diff options
Diffstat (limited to 'drivers/net/myri10ge/myri10ge.c')
-rw-r--r-- | drivers/net/myri10ge/myri10ge.c | 185 |
1 files changed, 134 insertions, 51 deletions
diff --git a/drivers/net/myri10ge/myri10ge.c b/drivers/net/myri10ge/myri10ge.c index d6524db321af..6dce901c7f45 100644 --- a/drivers/net/myri10ge/myri10ge.c +++ b/drivers/net/myri10ge/myri10ge.c @@ -75,7 +75,7 @@ #include "myri10ge_mcp.h" #include "myri10ge_mcp_gen_header.h" -#define MYRI10GE_VERSION_STR "1.4.3-1.358" +#define MYRI10GE_VERSION_STR "1.4.3-1.369" MODULE_DESCRIPTION("Myricom 10G driver (10GbE)"); MODULE_AUTHOR("Maintainer: help@myri.com"); @@ -102,6 +102,8 @@ MODULE_LICENSE("Dual BSD/GPL"); #define MYRI10GE_ALLOC_SIZE ((1 << MYRI10GE_ALLOC_ORDER) * PAGE_SIZE) #define MYRI10GE_MAX_FRAGS_PER_FRAME (MYRI10GE_MAX_ETHER_MTU/MYRI10GE_ALLOC_SIZE + 1) +#define MYRI10GE_MAX_SLICES 32 + struct myri10ge_rx_buffer_state { struct page *page; int page_offset; @@ -138,6 +140,8 @@ struct myri10ge_rx_buf { struct myri10ge_tx_buf { struct mcp_kreq_ether_send __iomem *lanai; /* lanai ptr for sendq */ + __be32 __iomem *send_go; /* "go" doorbell ptr */ + __be32 __iomem *send_stop; /* "stop" doorbell ptr */ struct mcp_kreq_ether_send *req_list; /* host shadow of sendq */ char *req_bytes; struct myri10ge_tx_buffer_state *info; @@ -149,6 +153,7 @@ struct myri10ge_tx_buf { int done ____cacheline_aligned; /* transmit slots completed */ int pkt_done; /* packets completed */ int wake_queue; + int queue_active; }; struct myri10ge_rx_done { @@ -183,7 +188,7 @@ struct myri10ge_slice_state { dma_addr_t fw_stats_bus; int watchdog_tx_done; int watchdog_tx_req; -#ifdef CONFIG_DCA +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) int cached_dca_tag; int cpu; __be32 __iomem *dca_tag; @@ -215,7 +220,7 @@ struct myri10ge_priv { int msi_enabled; int msix_enabled; struct msix_entry *msix_vectors; -#ifdef CONFIG_DCA +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) int dca_enabled; #endif u32 link_state; @@ -418,6 +423,12 @@ myri10ge_send_cmd(struct myri10ge_priv *mgp, u32 cmd, return -ENOSYS; } else if (result == MXGEFW_CMD_ERROR_UNALIGNED) { return -E2BIG; + } else if (result == MXGEFW_CMD_ERROR_RANGE && + cmd == MXGEFW_CMD_ENABLE_RSS_QUEUES && + (data-> + data1 & MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES) != + 0) { + return -ERANGE; } else { dev_err(&mgp->pdev->dev, "command %d failed, result = %d\n", @@ -891,7 +902,7 @@ static int myri10ge_reset(struct myri10ge_priv *mgp) struct myri10ge_slice_state *ss; int i, status; size_t bytes; -#ifdef CONFIG_DCA +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) unsigned long dca_tag_off; #endif @@ -947,9 +958,24 @@ static int myri10ge_reset(struct myri10ge_priv *mgp) */ cmd.data0 = mgp->num_slices; - cmd.data1 = 1; /* use MSI-X */ + cmd.data1 = MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE; + if (mgp->dev->real_num_tx_queues > 1) + cmd.data1 |= MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES; status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ENABLE_RSS_QUEUES, &cmd, 0); + + /* Firmware older than 1.4.32 only supports multiple + * RX queues, so if we get an error, first retry using a + * single TX queue before giving up */ + if (status != 0 && mgp->dev->real_num_tx_queues > 1) { + mgp->dev->real_num_tx_queues = 1; + cmd.data0 = mgp->num_slices; + cmd.data1 = MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE; + status = myri10ge_send_cmd(mgp, + MXGEFW_CMD_ENABLE_RSS_QUEUES, + &cmd, 0); + } + if (status != 0) { dev_err(&mgp->pdev->dev, "failed to set number of slices\n"); @@ -986,7 +1012,7 @@ static int myri10ge_reset(struct myri10ge_priv *mgp) } put_be32(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr); -#ifdef CONFIG_DCA +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_DCA_OFFSET, &cmd, 0); dca_tag_off = cmd.data0; for (i = 0; i < mgp->num_slices; i++) { @@ -1025,7 +1051,7 @@ static int myri10ge_reset(struct myri10ge_priv *mgp) return status; } -#ifdef CONFIG_DCA +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) static void myri10ge_write_dca(struct myri10ge_slice_state *ss, int cpu, int tag) { @@ -1060,8 +1086,9 @@ static void myri10ge_setup_dca(struct myri10ge_priv *mgp) } err = dca_add_requester(&pdev->dev); if (err) { - dev_err(&pdev->dev, - "dca_add_requester() failed, err=%d\n", err); + if (err != -ENODEV) + dev_err(&pdev->dev, + "dca_add_requester() failed, err=%d\n", err); return; } mgp->dca_enabled = 1; @@ -1316,6 +1343,7 @@ myri10ge_tx_done(struct myri10ge_slice_state *ss, int mcp_index) { struct pci_dev *pdev = ss->mgp->pdev; struct myri10ge_tx_buf *tx = &ss->tx; + struct netdev_queue *dev_queue; struct sk_buff *skb; int idx, len; @@ -1349,11 +1377,31 @@ myri10ge_tx_done(struct myri10ge_slice_state *ss, int mcp_index) PCI_DMA_TODEVICE); } } + + dev_queue = netdev_get_tx_queue(ss->dev, ss - ss->mgp->ss); + /* + * Make a minimal effort to prevent the NIC from polling an + * idle tx queue. If we can't get the lock we leave the queue + * active. In this case, either a thread was about to start + * using the queue anyway, or we lost a race and the NIC will + * waste some of its resources polling an inactive queue for a + * while. + */ + + if ((ss->mgp->dev->real_num_tx_queues > 1) && + __netif_tx_trylock(dev_queue)) { + if (tx->req == tx->done) { + tx->queue_active = 0; + put_be32(htonl(1), tx->send_stop); + } + __netif_tx_unlock(dev_queue); + } + /* start the queue if we've stopped it */ - if (netif_queue_stopped(ss->dev) + if (netif_tx_queue_stopped(dev_queue) && tx->req - tx->done < (tx->mask >> 1)) { tx->wake_queue++; - netif_wake_queue(ss->dev); + netif_tx_wake_queue(dev_queue); } } @@ -1457,7 +1505,7 @@ static int myri10ge_poll(struct napi_struct *napi, int budget) struct net_device *netdev = ss->mgp->dev; int work_done; -#ifdef CONFIG_DCA +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) if (ss->mgp->dca_enabled) myri10ge_update_dca(ss); #endif @@ -1481,9 +1529,9 @@ static irqreturn_t myri10ge_intr(int irq, void *arg) u32 send_done_count; int i; - /* an interrupt on a non-zero slice is implicitly valid - * since MSI-X irqs are not shared */ - if (ss != mgp->ss) { + /* an interrupt on a non-zero receive-only slice is implicitly + * valid since MSI-X irqs are not shared */ + if ((mgp->dev->real_num_tx_queues == 1) && (ss != mgp->ss)) { netif_rx_schedule(ss->dev, &ss->napi); return (IRQ_HANDLED); } @@ -1525,7 +1573,9 @@ static irqreturn_t myri10ge_intr(int irq, void *arg) barrier(); } - myri10ge_check_statblock(mgp); + /* Only slice 0 updates stats */ + if (ss == mgp->ss) + myri10ge_check_statblock(mgp); put_be32(htonl(3), ss->irq_claim + 1); return (IRQ_HANDLED); @@ -1686,8 +1736,8 @@ static const char myri10ge_gstrings_main_stats[][ETH_GSTRING_LEN] = { "tx_boundary", "WC", "irq", "MSI", "MSIX", "read_dma_bw_MBs", "write_dma_bw_MBs", "read_write_dma_bw_MBs", "serial_number", "watchdog_resets", -#ifdef CONFIG_DCA - "dca_capable", "dca_enabled", +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) + "dca_capable_firmware", "dca_device_present", #endif "link_changes", "link_up", "dropped_link_overflow", "dropped_link_error_or_filtered", @@ -1765,7 +1815,7 @@ myri10ge_get_ethtool_stats(struct net_device *netdev, data[i++] = (unsigned int)mgp->read_write_dma; data[i++] = (unsigned int)mgp->serial_number; data[i++] = (unsigned int)mgp->watchdog_resets; -#ifdef CONFIG_DCA +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) data[i++] = (unsigned int)(mgp->ss[0].dca_tag != NULL); data[i++] = (unsigned int)(mgp->dca_enabled); #endif @@ -1883,6 +1933,7 @@ static int myri10ge_allocate_rings(struct myri10ge_slice_state *ss) /* ensure req_list entries are aligned to 8 bytes */ ss->tx.req_list = (struct mcp_kreq_ether_send *) ALIGN((unsigned long)ss->tx.req_bytes, 8); + ss->tx.queue_active = 0; bytes = rx_ring_entries * sizeof(*ss->rx_small.shadow); ss->rx_small.shadow = kzalloc(bytes, GFP_KERNEL); @@ -2200,11 +2251,14 @@ static int myri10ge_get_txrx(struct myri10ge_priv *mgp, int slice) int status; ss = &mgp->ss[slice]; - cmd.data0 = 0; /* single slice for now */ - status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_OFFSET, &cmd, 0); - ss->tx.lanai = (struct mcp_kreq_ether_send __iomem *) - (mgp->sram + cmd.data0); - + status = 0; + if (slice == 0 || (mgp->dev->real_num_tx_queues > 1)) { + cmd.data0 = slice; + status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_OFFSET, + &cmd, 0); + ss->tx.lanai = (struct mcp_kreq_ether_send __iomem *) + (mgp->sram + cmd.data0); + } cmd.data0 = slice; status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SMALL_RX_OFFSET, &cmd, 0); @@ -2216,6 +2270,10 @@ static int myri10ge_get_txrx(struct myri10ge_priv *mgp, int slice) ss->rx_big.lanai = (struct mcp_kreq_ether_recv __iomem *) (mgp->sram + cmd.data0); + ss->tx.send_go = (__iomem __be32 *) + (mgp->sram + MXGEFW_ETH_SEND_GO + 64 * slice); + ss->tx.send_stop = (__iomem __be32 *) + (mgp->sram + MXGEFW_ETH_SEND_STOP + 64 * slice); return status; } @@ -2229,7 +2287,7 @@ static int myri10ge_set_stats(struct myri10ge_priv *mgp, int slice) ss = &mgp->ss[slice]; cmd.data0 = MYRI10GE_LOWPART_TO_U32(ss->fw_stats_bus); cmd.data1 = MYRI10GE_HIGHPART_TO_U32(ss->fw_stats_bus); - cmd.data2 = sizeof(struct mcp_irq_data); + cmd.data2 = sizeof(struct mcp_irq_data) | (slice << 16); status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_STATS_DMA_V2, &cmd, 0); if (status == -ENOSYS) { dma_addr_t bus = ss->fw_stats_bus; @@ -2270,7 +2328,9 @@ static int myri10ge_open(struct net_device *dev) if (mgp->num_slices > 1) { cmd.data0 = mgp->num_slices; - cmd.data1 = 1; /* use MSI-X */ + cmd.data1 = MXGEFW_SLICE_INTR_MODE_ONE_PER_SLICE; + if (mgp->dev->real_num_tx_queues > 1) + cmd.data1 |= MXGEFW_SLICE_ENABLE_MULTIPLE_TX_QUEUES; status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ENABLE_RSS_QUEUES, &cmd, 0); if (status != 0) { @@ -2291,6 +2351,7 @@ static int myri10ge_open(struct net_device *dev) printk(KERN_ERR "myri10ge: %s: failed to setup rss tables\n", dev->name); + goto abort_with_nothing; } /* just enable an identity mapping */ @@ -2361,7 +2422,11 @@ static int myri10ge_open(struct net_device *dev) status = myri10ge_allocate_rings(ss); if (status != 0) goto abort_with_rings; - if (slice == 0) + + /* only firmware which supports multiple TX queues + * supports setting up the tx stats on non-zero + * slices */ + if (slice == 0 || mgp->dev->real_num_tx_queues > 1) status = myri10ge_set_stats(mgp, slice); if (status) { printk(KERN_ERR @@ -2427,7 +2492,8 @@ static int myri10ge_open(struct net_device *dev) mgp->running = MYRI10GE_ETH_RUNNING; mgp->watchdog_timer.expires = jiffies + myri10ge_watchdog_timeout * HZ; add_timer(&mgp->watchdog_timer); - netif_wake_queue(dev); + netif_tx_wake_all_queues(dev); + return 0; abort_with_rings: @@ -2460,7 +2526,8 @@ static int myri10ge_close(struct net_device *dev) napi_disable(&mgp->ss[i].napi); } netif_carrier_off(dev); - netif_stop_queue(dev); + + netif_tx_stop_all_queues(dev); old_down_cnt = mgp->down_cnt; mb(); status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_DOWN, &cmd, 0); @@ -2565,18 +2632,21 @@ static int myri10ge_xmit(struct sk_buff *skb, struct net_device *dev) struct mcp_kreq_ether_send *req; struct myri10ge_tx_buf *tx; struct skb_frag_struct *frag; + struct netdev_queue *netdev_queue; dma_addr_t bus; u32 low; __be32 high_swapped; unsigned int len; int idx, last_idx, avail, frag_cnt, frag_idx, count, mss, max_segments; - u16 pseudo_hdr_offset, cksum_offset; + u16 pseudo_hdr_offset, cksum_offset, queue; int cum_len, seglen, boundary, rdma_count; u8 flags, odd_flag; - /* always transmit through slot 0 */ - ss = mgp->ss; + queue = skb_get_queue_mapping(skb); + ss = &mgp->ss[queue]; + netdev_queue = netdev_get_tx_queue(mgp->dev, queue); tx = &ss->tx; + again: req = tx->req_list; avail = tx->mask - 1 - (tx->req - tx->done); @@ -2592,7 +2662,7 @@ again: if ((unlikely(avail < max_segments))) { /* we are out of transmit resources */ tx->stop_queue++; - netif_stop_queue(dev); + netif_tx_stop_queue(netdev_queue); return 1; } @@ -2785,10 +2855,16 @@ again: idx = ((count - 1) + tx->req) & tx->mask; tx->info[idx].last = 1; myri10ge_submit_req(tx, tx->req_list, count); + /* if using multiple tx queues, make sure NIC polls the + * current slice */ + if ((mgp->dev->real_num_tx_queues > 1) && tx->queue_active == 0) { + tx->queue_active = 1; + put_be32(htonl(1), tx->send_go); + } tx->pkt_start++; if ((avail - count) < MXGEFW_MAX_SEND_DESC) { tx->stop_queue++; - netif_stop_queue(dev); + netif_tx_stop_queue(netdev_queue); } dev->trans_start = jiffies; return 0; @@ -3366,20 +3442,21 @@ static void myri10ge_watchdog(struct work_struct *work) for (i = 0; i < mgp->num_slices; i++) { tx = &mgp->ss[i].tx; printk(KERN_INFO - "myri10ge: %s: (%d): %d %d %d %d %d\n", - mgp->dev->name, i, tx->req, tx->done, - tx->pkt_start, tx->pkt_done, + "myri10ge: %s: (%d): %d %d %d %d %d %d\n", + mgp->dev->name, i, tx->queue_active, tx->req, + tx->done, tx->pkt_start, tx->pkt_done, (int)ntohl(mgp->ss[i].fw_stats-> send_done_count)); msleep(2000); printk(KERN_INFO - "myri10ge: %s: (%d): %d %d %d %d %d\n", - mgp->dev->name, i, tx->req, tx->done, - tx->pkt_start, tx->pkt_done, + "myri10ge: %s: (%d): %d %d %d %d %d %d\n", + mgp->dev->name, i, tx->queue_active, tx->req, + tx->done, tx->pkt_start, tx->pkt_done, (int)ntohl(mgp->ss[i].fw_stats-> send_done_count)); } } + rtnl_lock(); myri10ge_close(mgp->dev); status = myri10ge_load_firmware(mgp, 1); @@ -3434,10 +3511,14 @@ static void myri10ge_watchdog_timer(unsigned long arg) /* nic seems like it might be stuck.. */ if (rx_pause_cnt != mgp->watchdog_pause) { if (net_ratelimit()) - printk(KERN_WARNING "myri10ge %s:" + printk(KERN_WARNING + "myri10ge %s slice %d:" "TX paused, check link partner\n", - mgp->dev->name); + mgp->dev->name, i); } else { + printk(KERN_WARNING + "myri10ge %s slice %d stuck:", + mgp->dev->name, i); reset_needed = 1; } } @@ -3652,7 +3733,7 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent) int status = -ENXIO; int dac_enabled; - netdev = alloc_etherdev(sizeof(*mgp)); + netdev = alloc_etherdev_mq(sizeof(*mgp), MYRI10GE_MAX_SLICES); if (netdev == NULL) { dev_err(dev, "Could not allocate ethernet device\n"); return -ENOMEM; @@ -3757,13 +3838,13 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent) dev_err(&pdev->dev, "failed to alloc slice state\n"); goto abort_with_firmware; } - + netdev->real_num_tx_queues = mgp->num_slices; status = myri10ge_reset(mgp); if (status != 0) { dev_err(&pdev->dev, "failed reset\n"); goto abort_with_slices; } -#ifdef CONFIG_DCA +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) myri10ge_setup_dca(mgp); #endif pci_set_drvdata(pdev, mgp); @@ -3781,6 +3862,7 @@ static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent) netdev->set_multicast_list = myri10ge_set_multicast_list; netdev->set_mac_address = myri10ge_set_mac_address; netdev->features = mgp->features; + if (dac_enabled) netdev->features |= NETIF_F_HIGHDMA; @@ -3866,7 +3948,7 @@ static void myri10ge_remove(struct pci_dev *pdev) netdev = mgp->dev; unregister_netdev(netdev); -#ifdef CONFIG_DCA +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) myri10ge_teardown_dca(mgp); #endif myri10ge_dummy_rdma(mgp, 0); @@ -3911,7 +3993,7 @@ static struct pci_driver myri10ge_driver = { #endif }; -#ifdef CONFIG_DCA +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) static int myri10ge_notify_dca(struct notifier_block *nb, unsigned long event, void *p) { @@ -3936,16 +4018,17 @@ static __init int myri10ge_init_module(void) printk(KERN_INFO "%s: Version %s\n", myri10ge_driver.name, MYRI10GE_VERSION_STR); - if (myri10ge_rss_hash > MXGEFW_RSS_HASH_TYPE_SRC_PORT || - myri10ge_rss_hash < MXGEFW_RSS_HASH_TYPE_IPV4) { + if (myri10ge_rss_hash > MXGEFW_RSS_HASH_TYPE_MAX) { printk(KERN_ERR "%s: Illegal rssh hash type %d, defaulting to source port\n", myri10ge_driver.name, myri10ge_rss_hash); myri10ge_rss_hash = MXGEFW_RSS_HASH_TYPE_SRC_PORT; } -#ifdef CONFIG_DCA +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) dca_register_notify(&myri10ge_dca_notifier); #endif + if (myri10ge_max_slices > MYRI10GE_MAX_SLICES) + myri10ge_max_slices = MYRI10GE_MAX_SLICES; return pci_register_driver(&myri10ge_driver); } @@ -3954,7 +4037,7 @@ module_init(myri10ge_init_module); static __exit void myri10ge_cleanup_module(void) { -#ifdef CONFIG_DCA +#if (defined CONFIG_DCA) || (defined CONFIG_DCA_MODULE) dca_unregister_notify(&myri10ge_dca_notifier); #endif pci_unregister_driver(&myri10ge_driver); |