e1000e: fix close interrupt race
[safe/jmp/linux-2.6] / drivers / net / e1000e / netdev.c
index 2ffd752..b4ae046 100644 (file)
 #include <linux/cpu.h>
 #include <linux/smp.h>
 #include <linux/pm_qos_params.h>
+#include <linux/aer.h>
 
 #include "e1000.h"
 
-#define DRV_VERSION "0.3.3.3-k6"
+#define DRV_VERSION "0.3.3.4-k2"
 char e1000e_driver_name[] = "e1000e";
 const char e1000e_driver_version[] = DRV_VERSION;
 
@@ -56,6 +57,7 @@ static const struct e1000_info *e1000_info_tbl[] = {
        [board_82572]           = &e1000_82572_info,
        [board_82573]           = &e1000_82573_info,
        [board_82574]           = &e1000_82574_info,
+       [board_82583]           = &e1000_82583_info,
        [board_80003es2lan]     = &e1000_es2_info,
        [board_ich8lan]         = &e1000_ich8_info,
        [board_ich9lan]         = &e1000_ich9_info,
@@ -565,15 +567,14 @@ next_desc:
 static void e1000_put_txbuf(struct e1000_adapter *adapter,
                             struct e1000_buffer *buffer_info)
 {
-       if (buffer_info->dma) {
-               pci_unmap_page(adapter->pdev, buffer_info->dma,
-                              buffer_info->length, PCI_DMA_TODEVICE);
-               buffer_info->dma = 0;
-       }
+       buffer_info->dma = 0;
        if (buffer_info->skb) {
+               skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
+                             DMA_TO_DEVICE);
                dev_kfree_skb_any(buffer_info->skb);
                buffer_info->skb = NULL;
        }
+       buffer_info->time_stamp = 0;
 }
 
 static void e1000_print_tx_hang(struct e1000_adapter *adapter)
@@ -620,15 +621,16 @@ static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
        struct e1000_buffer *buffer_info;
        unsigned int i, eop;
        unsigned int count = 0;
-       bool cleaned = 0;
+       bool cleaned;
        unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 
        i = tx_ring->next_to_clean;
        eop = tx_ring->buffer_info[i].next_to_watch;
        eop_desc = E1000_TX_DESC(*tx_ring, eop);
 
-       while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
-               for (cleaned = 0; !cleaned; ) {
+       while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
+              (count < tx_ring->count)) {
+               for (cleaned = 0; !cleaned; count++) {
                        tx_desc = E1000_TX_DESC(*tx_ring, i);
                        buffer_info = &tx_ring->buffer_info[i];
                        cleaned = (i == eop);
@@ -654,10 +656,6 @@ static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 
                eop = tx_ring->buffer_info[i].next_to_watch;
                eop_desc = E1000_TX_DESC(*tx_ring, eop);
-#define E1000_TX_WEIGHT 64
-               /* weight of a sort for tx, to avoid endless transmit cleanup */
-               if (count++ == E1000_TX_WEIGHT)
-                       break;
        }
 
        tx_ring->next_to_clean = i;
@@ -678,13 +676,11 @@ static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
        }
 
        if (adapter->detect_tx_hung) {
-               /*
-                * Detect a transmit hang in hardware, this serializes the
-                * check with the clearing of time_stamp and movement of i
-                */
+               /* Detect a transmit hang in hardware, this serializes the
+                * check with the clearing of time_stamp and movement of i */
                adapter->detect_tx_hung = 0;
-               if (tx_ring->buffer_info[eop].dma &&
-                   time_after(jiffies, tx_ring->buffer_info[eop].time_stamp
+               if (tx_ring->buffer_info[i].time_stamp &&
+                   time_after(jiffies, tx_ring->buffer_info[i].time_stamp
                               + (adapter->tx_timeout_factor * HZ))
                    && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
                        e1000_print_tx_hang(adapter);
@@ -695,7 +691,7 @@ static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
        adapter->total_tx_packets += total_tx_packets;
        adapter->net_stats.tx_bytes += total_tx_bytes;
        adapter->net_stats.tx_packets += total_tx_packets;
-       return cleaned;
+       return (count < tx_ring->count);
 }
 
 /**
@@ -1152,7 +1148,7 @@ static irqreturn_t e1000_intr_msi(int irq, void *data)
         * read ICR disables interrupts using IAM
         */
 
-       if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
+       if (icr & E1000_ICR_LSC) {
                hw->mac.get_link_status = 1;
                /*
                 * ICH8 workaround-- Call gig speed drop workaround on cable
@@ -1218,7 +1214,7 @@ static irqreturn_t e1000_intr(int irq, void *data)
         * IMC write
         */
 
-       if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
+       if (icr & E1000_ICR_LSC) {
                hw->mac.get_link_status = 1;
                /*
                 * ICH8 workaround-- Call gig speed drop workaround on cable
@@ -1265,7 +1261,8 @@ static irqreturn_t e1000_msix_other(int irq, void *data)
        u32 icr = er32(ICR);
 
        if (!(icr & E1000_ICR_INT_ASSERTED)) {
-               ew32(IMS, E1000_IMS_OTHER);
+               if (!test_bit(__E1000_DOWN, &adapter->state))
+                       ew32(IMS, E1000_IMS_OTHER);
                return IRQ_NONE;
        }
 
@@ -1282,7 +1279,8 @@ static irqreturn_t e1000_msix_other(int irq, void *data)
        }
 
 no_link_interrupt:
-       ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
+       if (!test_bit(__E1000_DOWN, &adapter->state))
+               ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
 
        return IRQ_HANDLED;
 }
@@ -1698,7 +1696,6 @@ int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
 
        tx_ring->next_to_use = 0;
        tx_ring->next_to_clean = 0;
-       spin_lock_init(&adapter->tx_queue_lock);
 
        return 0;
 err:
@@ -2007,21 +2004,12 @@ static int e1000_clean(struct napi_struct *napi, int budget)
            !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
                goto clean_rx;
 
-       /*
-        * e1000_clean is called per-cpu.  This lock protects
-        * tx_ring from being cleaned by multiple cpus
-        * simultaneously.  A failure obtaining the lock means
-        * tx_ring is currently being cleaned anyway.
-        */
-       if (spin_trylock(&adapter->tx_queue_lock)) {
-               tx_cleaned = e1000_clean_tx_irq(adapter);
-               spin_unlock(&adapter->tx_queue_lock);
-       }
+       tx_cleaned = e1000_clean_tx_irq(adapter);
 
 clean_rx:
        adapter->clean_rx(adapter, &work_done, budget);
 
-       if (tx_cleaned)
+       if (!tx_cleaned)
                work_done = budget;
 
        /* If budget not fully consumed, exit the polling mode */
@@ -2029,10 +2017,12 @@ clean_rx:
                if (adapter->itr_setting & 3)
                        e1000_set_itr(adapter);
                napi_complete(napi);
-               if (adapter->msix_entries)
-                       ew32(IMS, adapter->rx_ring->ims_val);
-               else
-                       e1000_irq_enable(adapter);
+               if (!test_bit(__E1000_DOWN, &adapter->state)) {
+                       if (adapter->msix_entries)
+                               ew32(IMS, adapter->rx_ring->ims_val);
+                       else
+                               e1000_irq_enable(adapter);
+               }
        }
 
        return work_done;
@@ -2922,8 +2912,6 @@ static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
        if (e1000_alloc_queues(adapter))
                return -ENOMEM;
 
-       spin_lock_init(&adapter->tx_queue_lock);
-
        /* Explicitly disable IRQ since the NIC can be in any state. */
        e1000_irq_disable(adapter);
 
@@ -3320,7 +3308,7 @@ void e1000e_update_stats(struct e1000_adapter *adapter)
 
        adapter->stats.algnerrc += er32(ALGNERRC);
        adapter->stats.rxerrc += er32(RXERRC);
-       if (hw->mac.type != e1000_82574)
+       if ((hw->mac.type != e1000_82574) && (hw->mac.type != e1000_82583))
                adapter->stats.tncrs += er32(TNCRS);
        adapter->stats.cexterr += er32(CEXTERR);
        adapter->stats.tsctc += er32(TSCTC);
@@ -3777,23 +3765,30 @@ static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
        unsigned int i;
        u8 css;
        u32 cmd_len = E1000_TXD_CMD_DEXT;
+       __be16 protocol;
 
        if (skb->ip_summed != CHECKSUM_PARTIAL)
                return 0;
 
-       switch (skb->protocol) {
-       case __constant_htons(ETH_P_IP):
+       if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
+               protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
+       else
+               protocol = skb->protocol;
+
+       switch (protocol) {
+       case cpu_to_be16(ETH_P_IP):
                if (ip_hdr(skb)->protocol == IPPROTO_TCP)
                        cmd_len |= E1000_TXD_CMD_TCP;
                break;
-       case __constant_htons(ETH_P_IPV6):
+       case cpu_to_be16(ETH_P_IPV6):
                /* XXX not handling all IPV6 headers */
                if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
                        cmd_len |= E1000_TXD_CMD_TCP;
                break;
        default:
                if (unlikely(net_ratelimit()))
-                       e_warn("checksum_partial proto=%x!\n", skb->protocol);
+                       e_warn("checksum_partial proto=%x!\n",
+                              be16_to_cpu(protocol));
                break;
        }
 
@@ -3832,42 +3827,40 @@ static int e1000_tx_map(struct e1000_adapter *adapter,
 {
        struct e1000_ring *tx_ring = adapter->tx_ring;
        struct e1000_buffer *buffer_info;
-       unsigned int len = skb->len - skb->data_len;
-       unsigned int offset = 0, size, count = 0, i;
+       unsigned int len = skb_headlen(skb);
+       unsigned int offset, size, count = 0, i;
        unsigned int f;
+       dma_addr_t *map;
 
        i = tx_ring->next_to_use;
 
+       if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
+               dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
+               adapter->tx_dma_failed++;
+               return 0;
+       }
+
+       map = skb_shinfo(skb)->dma_maps;
+       offset = 0;
+
        while (len) {
                buffer_info = &tx_ring->buffer_info[i];
                size = min(len, max_per_txd);
 
-               /* Workaround for premature desc write-backs
-                * in TSO mode.  Append 4-byte sentinel desc */
-               if (mss && !nr_frags && size == len && size > 8)
-                       size -= 4;
-
                buffer_info->length = size;
-               /* set time_stamp *before* dma to help avoid a possible race */
                buffer_info->time_stamp = jiffies;
-               buffer_info->dma =
-                       pci_map_single(adapter->pdev,
-                               skb->data + offset,
-                               size,
-                               PCI_DMA_TODEVICE);
-               if (pci_dma_mapping_error(adapter->pdev, buffer_info->dma)) {
-                       dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
-                       adapter->tx_dma_failed++;
-                       return -1;
-               }
                buffer_info->next_to_watch = i;
+               buffer_info->dma = map[0] + offset;
+               count++;
 
                len -= size;
                offset += size;
-               count++;
-               i++;
-               if (i == tx_ring->count)
-                       i = 0;
+
+               if (len) {
+                       i++;
+                       if (i == tx_ring->count)
+                               i = 0;
+               }
        }
 
        for (f = 0; f < nr_frags; f++) {
@@ -3875,49 +3868,27 @@ static int e1000_tx_map(struct e1000_adapter *adapter,
 
                frag = &skb_shinfo(skb)->frags[f];
                len = frag->size;
-               offset = frag->page_offset;
+               offset = 0;
 
                while (len) {
+                       i++;
+                       if (i == tx_ring->count)
+                               i = 0;
+
                        buffer_info = &tx_ring->buffer_info[i];
                        size = min(len, max_per_txd);
-                       /* Workaround for premature desc write-backs
-                        * in TSO mode.  Append 4-byte sentinel desc */
-                       if (mss && f == (nr_frags-1) && size == len && size > 8)
-                               size -= 4;
 
                        buffer_info->length = size;
                        buffer_info->time_stamp = jiffies;
-                       buffer_info->dma =
-                               pci_map_page(adapter->pdev,
-                                       frag->page,
-                                       offset,
-                                       size,
-                                       PCI_DMA_TODEVICE);
-                       if (pci_dma_mapping_error(adapter->pdev,
-                                                 buffer_info->dma)) {
-                               dev_err(&adapter->pdev->dev,
-                                       "TX DMA page map failed\n");
-                               adapter->tx_dma_failed++;
-                               return -1;
-                       }
-
                        buffer_info->next_to_watch = i;
+                       buffer_info->dma = map[f + 1] + offset;
 
                        len -= size;
                        offset += size;
                        count++;
-
-                       i++;
-                       if (i == tx_ring->count)
-                               i = 0;
                }
        }
 
-       if (i == 0)
-               i = tx_ring->count - 1;
-       else
-               i--;
-
        tx_ring->buffer_info[i].skb = skb;
        tx_ring->buffer_info[first].next_to_watch = i;
 
@@ -4069,7 +4040,6 @@ static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
        unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
        unsigned int tx_flags = 0;
        unsigned int len = skb->len - skb->data_len;
-       unsigned long irq_flags;
        unsigned int nr_frags;
        unsigned int mss;
        int count = 0;
@@ -4138,18 +4108,12 @@ static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
        if (adapter->hw.mac.tx_pkt_filtering)
                e1000_transfer_dhcp_info(adapter, skb);
 
-       if (!spin_trylock_irqsave(&adapter->tx_queue_lock, irq_flags))
-               /* Collision - tell upper layer to requeue */
-               return NETDEV_TX_LOCKED;
-
        /*
         * need: count + 2 desc gap to keep tail from touching
         * head, otherwise try next time
         */
-       if (e1000_maybe_stop_tx(netdev, count + 2)) {
-               spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
+       if (e1000_maybe_stop_tx(netdev, count + 2))
                return NETDEV_TX_BUSY;
-       }
 
        if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
                tx_flags |= E1000_TX_FLAGS_VLAN;
@@ -4161,7 +4125,6 @@ static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
        tso = e1000_tso(adapter, skb);
        if (tso < 0) {
                dev_kfree_skb_any(skb);
-               spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
                return NETDEV_TX_OK;
        }
 
@@ -4178,22 +4141,20 @@ static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
        if (skb->protocol == htons(ETH_P_IP))
                tx_flags |= E1000_TX_FLAGS_IPV4;
 
+       /* if count is 0 then mapping error has occured */
        count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
-       if (count < 0) {
-               /* handle pci_map_single() error in e1000_tx_map */
+       if (count) {
+               e1000_tx_queue(adapter, tx_flags, count);
+               netdev->trans_start = jiffies;
+               /* Make sure there is space in the ring for the next send. */
+               e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
+
+       } else {
                dev_kfree_skb_any(skb);
-               spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
-               return NETDEV_TX_OK;
+               tx_ring->buffer_info[first].time_stamp = 0;
+               tx_ring->next_to_use = first;
        }
 
-       e1000_tx_queue(adapter, tx_flags, count);
-
-       netdev->trans_start = jiffies;
-
-       /* Make sure there is space in the ring for the next send. */
-       e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
-
-       spin_unlock_irqrestore(&adapter->tx_queue_lock, irq_flags);
        return NETDEV_TX_OK;
 }
 
@@ -4543,6 +4504,14 @@ static int e1000_resume(struct pci_dev *pdev)
                return err;
        }
 
+       /* AER (Advanced Error Reporting) hooks */
+       err = pci_enable_pcie_error_reporting(pdev);
+       if (err) {
+               dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
+                                   "0x%x\n", err);
+               /* non-fatal, continue */
+       }
+
        pci_set_master(pdev);
 
        pci_enable_wake(pdev, PCI_D3hot, 0);
@@ -4637,24 +4606,29 @@ static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
        struct e1000_adapter *adapter = netdev_priv(netdev);
        struct e1000_hw *hw = &adapter->hw;
        int err;
+       pci_ers_result_t result;
 
        e1000e_disable_l1aspm(pdev);
        err = pci_enable_device_mem(pdev);
        if (err) {
                dev_err(&pdev->dev,
                        "Cannot re-enable PCI device after reset.\n");
-               return PCI_ERS_RESULT_DISCONNECT;
-       }
-       pci_set_master(pdev);
-       pci_restore_state(pdev);
+               result = PCI_ERS_RESULT_DISCONNECT;
+       } else {
+               pci_set_master(pdev);
+               pci_restore_state(pdev);
 
-       pci_enable_wake(pdev, PCI_D3hot, 0);
-       pci_enable_wake(pdev, PCI_D3cold, 0);
+               pci_enable_wake(pdev, PCI_D3hot, 0);
+               pci_enable_wake(pdev, PCI_D3cold, 0);
 
-       e1000e_reset(adapter);
-       ew32(WUS, ~0);
+               e1000e_reset(adapter);
+               ew32(WUS, ~0);
+               result = PCI_ERS_RESULT_RECOVERED;
+       }
+
+       pci_cleanup_aer_uncorrect_error_status(pdev);
 
-       return PCI_ERS_RESULT_RECOVERED;
+       return result;
 }
 
 /**
@@ -4922,12 +4896,6 @@ static int __devinit e1000_probe(struct pci_dev *pdev,
        if (pci_using_dac)
                netdev->features |= NETIF_F_HIGHDMA;
 
-       /*
-        * We should not be using LLTX anymore, but we are still Tx faster with
-        * it.
-        */
-       netdev->features |= NETIF_F_LLTX;
-
        if (e1000e_enable_mng_pass_thru(&adapter->hw))
                adapter->flags |= FLAG_MNG_PT_ENABLED;
 
@@ -5091,6 +5059,7 @@ static void __devexit e1000_remove(struct pci_dev *pdev)
 {
        struct net_device *netdev = pci_get_drvdata(pdev);
        struct e1000_adapter *adapter = netdev_priv(netdev);
+       int err;
 
        /*
         * flush_scheduled work may reschedule our watchdog task, so
@@ -5125,6 +5094,12 @@ static void __devexit e1000_remove(struct pci_dev *pdev)
 
        free_netdev(netdev);
 
+       /* AER disable */
+       err = pci_disable_pcie_error_reporting(pdev);
+       if (err)
+               dev_err(&pdev->dev,
+                       "pci_disable_pcie_error_reporting failed 0x%x\n", err);
+
        pci_disable_device(pdev);
 }
 
@@ -5156,6 +5131,8 @@ static struct pci_device_id e1000_pci_tbl[] = {
        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
 
        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
+       { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
+       { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
 
        { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
          board_80003es2lan },