ixgbe: fix unmap length bug
[linux-2.6] / drivers / net / ixgbe / ixgbe_main.c
1 /*******************************************************************************
2
3   Intel 10 Gigabit PCI Express Linux driver
4   Copyright(c) 1999 - 2009 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
34 #include <linux/in.h>
35 #include <linux/ip.h>
36 #include <linux/tcp.h>
37 #include <linux/ipv6.h>
38 #include <net/checksum.h>
39 #include <net/ip6_checksum.h>
40 #include <linux/ethtool.h>
41 #include <linux/if_vlan.h>
42 #include <scsi/fc/fc_fcoe.h>
43
44 #include "ixgbe.h"
45 #include "ixgbe_common.h"
46
47 char ixgbe_driver_name[] = "ixgbe";
48 static const char ixgbe_driver_string[] =
49                               "Intel(R) 10 Gigabit PCI Express Network Driver";
50
51 #define DRV_VERSION "2.0.34-k2"
52 const char ixgbe_driver_version[] = DRV_VERSION;
53 static char ixgbe_copyright[] = "Copyright (c) 1999-2009 Intel Corporation.";
54
55 static const struct ixgbe_info *ixgbe_info_tbl[] = {
56         [board_82598] = &ixgbe_82598_info,
57         [board_82599] = &ixgbe_82599_info,
58 };
59
60 /* ixgbe_pci_tbl - PCI Device ID Table
61  *
62  * Wildcard entries (PCI_ANY_ID) should come last
63  * Last entry must be all 0s
64  *
65  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
66  *   Class, Class Mask, private data (not used) }
67  */
68 static struct pci_device_id ixgbe_pci_tbl[] = {
69         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
70          board_82598 },
71         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
72          board_82598 },
73         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
74          board_82598 },
75         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
76          board_82598 },
77         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
78          board_82598 },
79         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
80          board_82598 },
81         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
82          board_82598 },
83         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
84          board_82598 },
85         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
86          board_82598 },
87         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
88          board_82598 },
89         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
90          board_82598 },
91         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
92          board_82599 },
93         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
94          board_82599 },
95         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
96          board_82599 },
97
98         /* required last entry */
99         {0, }
100 };
101 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
102
103 #ifdef CONFIG_IXGBE_DCA
104 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
105                             void *p);
106 static struct notifier_block dca_notifier = {
107         .notifier_call = ixgbe_notify_dca,
108         .next          = NULL,
109         .priority      = 0
110 };
111 #endif
112
113 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
114 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
115 MODULE_LICENSE("GPL");
116 MODULE_VERSION(DRV_VERSION);
117
118 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
119
120 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
121 {
122         u32 ctrl_ext;
123
124         /* Let firmware take over control of h/w */
125         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
126         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
127                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
128 }
129
130 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
131 {
132         u32 ctrl_ext;
133
134         /* Let firmware know the driver has taken over */
135         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
136         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
137                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
138 }
139
140 /*
141  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
142  * @adapter: pointer to adapter struct
143  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
144  * @queue: queue to map the corresponding interrupt to
145  * @msix_vector: the vector to map to the corresponding queue
146  *
147  */
148 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
149                            u8 queue, u8 msix_vector)
150 {
151         u32 ivar, index;
152         struct ixgbe_hw *hw = &adapter->hw;
153         switch (hw->mac.type) {
154         case ixgbe_mac_82598EB:
155                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
156                 if (direction == -1)
157                         direction = 0;
158                 index = (((direction * 64) + queue) >> 2) & 0x1F;
159                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
160                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
161                 ivar |= (msix_vector << (8 * (queue & 0x3)));
162                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
163                 break;
164         case ixgbe_mac_82599EB:
165                 if (direction == -1) {
166                         /* other causes */
167                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
168                         index = ((queue & 1) * 8);
169                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
170                         ivar &= ~(0xFF << index);
171                         ivar |= (msix_vector << index);
172                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
173                         break;
174                 } else {
175                         /* tx or rx causes */
176                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
177                         index = ((16 * (queue & 1)) + (8 * direction));
178                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
179                         ivar &= ~(0xFF << index);
180                         ivar |= (msix_vector << index);
181                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
182                         break;
183                 }
184         default:
185                 break;
186         }
187 }
188
189 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
190                                           u64 qmask)
191 {
192         u32 mask;
193
194         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
195                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
196                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
197         } else {
198                 mask = (qmask & 0xFFFFFFFF);
199                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
200                 mask = (qmask >> 32);
201                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
202         }
203 }
204
205 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
206                                              struct ixgbe_tx_buffer
207                                              *tx_buffer_info)
208 {
209         tx_buffer_info->dma = 0;
210         if (tx_buffer_info->skb) {
211                 skb_dma_unmap(&adapter->pdev->dev, tx_buffer_info->skb,
212                               DMA_TO_DEVICE);
213                 dev_kfree_skb_any(tx_buffer_info->skb);
214                 tx_buffer_info->skb = NULL;
215         }
216         tx_buffer_info->time_stamp = 0;
217         /* tx_buffer_info must be completely set up in the transmit path */
218 }
219
220 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
221                                        struct ixgbe_ring *tx_ring,
222                                        unsigned int eop)
223 {
224         struct ixgbe_hw *hw = &adapter->hw;
225
226         /* Detect a transmit hang in hardware, this serializes the
227          * check with the clearing of time_stamp and movement of eop */
228         adapter->detect_tx_hung = false;
229         if (tx_ring->tx_buffer_info[eop].time_stamp &&
230             time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
231             !(IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & IXGBE_TFCS_TXOFF)) {
232                 /* detected Tx unit hang */
233                 union ixgbe_adv_tx_desc *tx_desc;
234                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
235                 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
236                         "  Tx Queue             <%d>\n"
237                         "  TDH, TDT             <%x>, <%x>\n"
238                         "  next_to_use          <%x>\n"
239                         "  next_to_clean        <%x>\n"
240                         "tx_buffer_info[next_to_clean]\n"
241                         "  time_stamp           <%lx>\n"
242                         "  jiffies              <%lx>\n",
243                         tx_ring->queue_index,
244                         IXGBE_READ_REG(hw, tx_ring->head),
245                         IXGBE_READ_REG(hw, tx_ring->tail),
246                         tx_ring->next_to_use, eop,
247                         tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
248                 return true;
249         }
250
251         return false;
252 }
253
254 #define IXGBE_MAX_TXD_PWR       14
255 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
256
257 /* Tx Descriptors needed, worst case */
258 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
259                          (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
260 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
261         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
262
263 static void ixgbe_tx_timeout(struct net_device *netdev);
264
265 /**
266  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
267  * @q_vector: structure containing interrupt and ring information
268  * @tx_ring: tx ring to clean
269  **/
270 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
271                                struct ixgbe_ring *tx_ring)
272 {
273         struct ixgbe_adapter *adapter = q_vector->adapter;
274         struct net_device *netdev = adapter->netdev;
275         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
276         struct ixgbe_tx_buffer *tx_buffer_info;
277         unsigned int i, eop, count = 0;
278         unsigned int total_bytes = 0, total_packets = 0;
279
280         i = tx_ring->next_to_clean;
281         eop = tx_ring->tx_buffer_info[i].next_to_watch;
282         eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
283
284         while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
285                (count < tx_ring->work_limit)) {
286                 bool cleaned = false;
287                 for ( ; !cleaned; count++) {
288                         struct sk_buff *skb;
289                         tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
290                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
291                         cleaned = (i == eop);
292                         skb = tx_buffer_info->skb;
293
294                         if (cleaned && skb) {
295                                 unsigned int segs, bytecount;
296                                 unsigned int hlen = skb_headlen(skb);
297
298                                 /* gso_segs is currently only valid for tcp */
299                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
300 #ifdef IXGBE_FCOE
301                                 /* adjust for FCoE Sequence Offload */
302                                 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
303                                     && (skb->protocol == htons(ETH_P_FCOE)) &&
304                                     skb_is_gso(skb)) {
305                                         hlen = skb_transport_offset(skb) +
306                                                 sizeof(struct fc_frame_header) +
307                                                 sizeof(struct fcoe_crc_eof);
308                                         segs = DIV_ROUND_UP(skb->len - hlen,
309                                                 skb_shinfo(skb)->gso_size);
310                                 }
311 #endif /* IXGBE_FCOE */
312                                 /* multiply data chunks by size of headers */
313                                 bytecount = ((segs - 1) * hlen) + skb->len;
314                                 total_packets += segs;
315                                 total_bytes += bytecount;
316                         }
317
318                         ixgbe_unmap_and_free_tx_resource(adapter,
319                                                          tx_buffer_info);
320
321                         tx_desc->wb.status = 0;
322
323                         i++;
324                         if (i == tx_ring->count)
325                                 i = 0;
326                 }
327
328                 eop = tx_ring->tx_buffer_info[i].next_to_watch;
329                 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
330         }
331
332         tx_ring->next_to_clean = i;
333
334 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
335         if (unlikely(count && netif_carrier_ok(netdev) &&
336                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
337                 /* Make sure that anybody stopping the queue after this
338                  * sees the new next_to_clean.
339                  */
340                 smp_mb();
341                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
342                     !test_bit(__IXGBE_DOWN, &adapter->state)) {
343                         netif_wake_subqueue(netdev, tx_ring->queue_index);
344                         ++adapter->restart_queue;
345                 }
346         }
347
348         if (adapter->detect_tx_hung) {
349                 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
350                         /* schedule immediate reset if we believe we hung */
351                         DPRINTK(PROBE, INFO,
352                                 "tx hang %d detected, resetting adapter\n",
353                                 adapter->tx_timeout_count + 1);
354                         ixgbe_tx_timeout(adapter->netdev);
355                 }
356         }
357
358         /* re-arm the interrupt */
359         if (count >= tx_ring->work_limit)
360                 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
361
362         tx_ring->total_bytes += total_bytes;
363         tx_ring->total_packets += total_packets;
364         tx_ring->stats.packets += total_packets;
365         tx_ring->stats.bytes += total_bytes;
366         adapter->net_stats.tx_bytes += total_bytes;
367         adapter->net_stats.tx_packets += total_packets;
368         return (count < tx_ring->work_limit);
369 }
370
371 #ifdef CONFIG_IXGBE_DCA
372 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
373                                 struct ixgbe_ring *rx_ring)
374 {
375         u32 rxctrl;
376         int cpu = get_cpu();
377         int q = rx_ring - adapter->rx_ring;
378
379         if (rx_ring->cpu != cpu) {
380                 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
381                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
382                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
383                         rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
384                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
385                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
386                         rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
387                                    IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
388                 }
389                 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
390                 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
391                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
392                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
393                             IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
394                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
395                 rx_ring->cpu = cpu;
396         }
397         put_cpu();
398 }
399
400 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
401                                 struct ixgbe_ring *tx_ring)
402 {
403         u32 txctrl;
404         int cpu = get_cpu();
405         int q = tx_ring - adapter->tx_ring;
406
407         if (tx_ring->cpu != cpu) {
408                 txctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q));
409                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
410                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
411                         txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
412                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
413                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
414                         txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
415                                    IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
416                 }
417                 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
418                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q), txctrl);
419                 tx_ring->cpu = cpu;
420         }
421         put_cpu();
422 }
423
424 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
425 {
426         int i;
427
428         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
429                 return;
430
431         /* always use CB2 mode, difference is masked in the CB driver */
432         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
433
434         for (i = 0; i < adapter->num_tx_queues; i++) {
435                 adapter->tx_ring[i].cpu = -1;
436                 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
437         }
438         for (i = 0; i < adapter->num_rx_queues; i++) {
439                 adapter->rx_ring[i].cpu = -1;
440                 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
441         }
442 }
443
444 static int __ixgbe_notify_dca(struct device *dev, void *data)
445 {
446         struct net_device *netdev = dev_get_drvdata(dev);
447         struct ixgbe_adapter *adapter = netdev_priv(netdev);
448         unsigned long event = *(unsigned long *)data;
449
450         switch (event) {
451         case DCA_PROVIDER_ADD:
452                 /* if we're already enabled, don't do it again */
453                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
454                         break;
455                 if (dca_add_requester(dev) == 0) {
456                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
457                         ixgbe_setup_dca(adapter);
458                         break;
459                 }
460                 /* Fall Through since DCA is disabled. */
461         case DCA_PROVIDER_REMOVE:
462                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
463                         dca_remove_requester(dev);
464                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
465                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
466                 }
467                 break;
468         }
469
470         return 0;
471 }
472
473 #endif /* CONFIG_IXGBE_DCA */
474 /**
475  * ixgbe_receive_skb - Send a completed packet up the stack
476  * @adapter: board private structure
477  * @skb: packet to send up
478  * @status: hardware indication of status of receive
479  * @rx_ring: rx descriptor ring (for a specific queue) to setup
480  * @rx_desc: rx descriptor
481  **/
482 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
483                               struct sk_buff *skb, u8 status,
484                               struct ixgbe_ring *ring,
485                               union ixgbe_adv_rx_desc *rx_desc)
486 {
487         struct ixgbe_adapter *adapter = q_vector->adapter;
488         struct napi_struct *napi = &q_vector->napi;
489         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
490         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
491
492         skb_record_rx_queue(skb, ring->queue_index);
493         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
494                 if (adapter->vlgrp && is_vlan && (tag != 0))
495                         vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
496                 else
497                         napi_gro_receive(napi, skb);
498         } else {
499                 if (adapter->vlgrp && is_vlan && (tag != 0))
500                         vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
501                 else
502                         netif_rx(skb);
503         }
504 }
505
506 /**
507  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
508  * @adapter: address of board private structure
509  * @status_err: hardware indication of status of receive
510  * @skb: skb currently being received and modified
511  **/
512 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
513                                      u32 status_err, struct sk_buff *skb)
514 {
515         skb->ip_summed = CHECKSUM_NONE;
516
517         /* Rx csum disabled */
518         if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
519                 return;
520
521         /* if IP and error */
522         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
523             (status_err & IXGBE_RXDADV_ERR_IPE)) {
524                 adapter->hw_csum_rx_error++;
525                 return;
526         }
527
528         if (!(status_err & IXGBE_RXD_STAT_L4CS))
529                 return;
530
531         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
532                 adapter->hw_csum_rx_error++;
533                 return;
534         }
535
536         /* It must be a TCP or UDP packet with a valid checksum */
537         skb->ip_summed = CHECKSUM_UNNECESSARY;
538         adapter->hw_csum_rx_good++;
539 }
540
541 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
542                                          struct ixgbe_ring *rx_ring, u32 val)
543 {
544         /*
545          * Force memory writes to complete before letting h/w
546          * know there are new descriptors to fetch.  (Only
547          * applicable for weak-ordered memory model archs,
548          * such as IA-64).
549          */
550         wmb();
551         IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
552 }
553
554 /**
555  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
556  * @adapter: address of board private structure
557  **/
558 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
559                                    struct ixgbe_ring *rx_ring,
560                                    int cleaned_count)
561 {
562         struct pci_dev *pdev = adapter->pdev;
563         union ixgbe_adv_rx_desc *rx_desc;
564         struct ixgbe_rx_buffer *bi;
565         unsigned int i;
566
567         i = rx_ring->next_to_use;
568         bi = &rx_ring->rx_buffer_info[i];
569
570         while (cleaned_count--) {
571                 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
572
573                 if (!bi->page_dma &&
574                     (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
575                         if (!bi->page) {
576                                 bi->page = alloc_page(GFP_ATOMIC);
577                                 if (!bi->page) {
578                                         adapter->alloc_rx_page_failed++;
579                                         goto no_buffers;
580                                 }
581                                 bi->page_offset = 0;
582                         } else {
583                                 /* use a half page if we're re-using */
584                                 bi->page_offset ^= (PAGE_SIZE / 2);
585                         }
586
587                         bi->page_dma = pci_map_page(pdev, bi->page,
588                                                     bi->page_offset,
589                                                     (PAGE_SIZE / 2),
590                                                     PCI_DMA_FROMDEVICE);
591                 }
592
593                 if (!bi->skb) {
594                         struct sk_buff *skb;
595                         skb = netdev_alloc_skb(adapter->netdev,
596                                                (rx_ring->rx_buf_len +
597                                                 NET_IP_ALIGN));
598
599                         if (!skb) {
600                                 adapter->alloc_rx_buff_failed++;
601                                 goto no_buffers;
602                         }
603
604                         /*
605                          * Make buffer alignment 2 beyond a 16 byte boundary
606                          * this will result in a 16 byte aligned IP header after
607                          * the 14 byte MAC header is removed
608                          */
609                         skb_reserve(skb, NET_IP_ALIGN);
610
611                         bi->skb = skb;
612                         bi->dma = pci_map_single(pdev, skb->data,
613                                                  rx_ring->rx_buf_len,
614                                                  PCI_DMA_FROMDEVICE);
615                 }
616                 /* Refresh the desc even if buffer_addrs didn't change because
617                  * each write-back erases this info. */
618                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
619                         rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
620                         rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
621                 } else {
622                         rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
623                 }
624
625                 i++;
626                 if (i == rx_ring->count)
627                         i = 0;
628                 bi = &rx_ring->rx_buffer_info[i];
629         }
630
631 no_buffers:
632         if (rx_ring->next_to_use != i) {
633                 rx_ring->next_to_use = i;
634                 if (i-- == 0)
635                         i = (rx_ring->count - 1);
636
637                 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
638         }
639 }
640
641 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
642 {
643         return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
644 }
645
646 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
647 {
648         return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
649 }
650
651 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
652 {
653         return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
654                 IXGBE_RXDADV_RSCCNT_MASK) >>
655                 IXGBE_RXDADV_RSCCNT_SHIFT;
656 }
657
658 /**
659  * ixgbe_transform_rsc_queue - change rsc queue into a full packet
660  * @skb: pointer to the last skb in the rsc queue
661  *
662  * This function changes a queue full of hw rsc buffers into a completed
663  * packet.  It uses the ->prev pointers to find the first packet and then
664  * turns it into the frag list owner.
665  **/
666 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
667 {
668         unsigned int frag_list_size = 0;
669
670         while (skb->prev) {
671                 struct sk_buff *prev = skb->prev;
672                 frag_list_size += skb->len;
673                 skb->prev = NULL;
674                 skb = prev;
675         }
676
677         skb_shinfo(skb)->frag_list = skb->next;
678         skb->next = NULL;
679         skb->len += frag_list_size;
680         skb->data_len += frag_list_size;
681         skb->truesize += frag_list_size;
682         return skb;
683 }
684
685 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
686                                struct ixgbe_ring *rx_ring,
687                                int *work_done, int work_to_do)
688 {
689         struct ixgbe_adapter *adapter = q_vector->adapter;
690         struct pci_dev *pdev = adapter->pdev;
691         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
692         struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
693         struct sk_buff *skb;
694         unsigned int i, rsc_count = 0;
695         u32 len, staterr;
696         u16 hdr_info;
697         bool cleaned = false;
698         int cleaned_count = 0;
699         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
700 #ifdef IXGBE_FCOE
701         int ddp_bytes = 0;
702 #endif /* IXGBE_FCOE */
703
704         i = rx_ring->next_to_clean;
705         rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
706         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
707         rx_buffer_info = &rx_ring->rx_buffer_info[i];
708
709         while (staterr & IXGBE_RXD_STAT_DD) {
710                 u32 upper_len = 0;
711                 if (*work_done >= work_to_do)
712                         break;
713                 (*work_done)++;
714
715                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
716                         hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
717                         len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
718                                IXGBE_RXDADV_HDRBUFLEN_SHIFT;
719                         if (hdr_info & IXGBE_RXDADV_SPH)
720                                 adapter->rx_hdr_split++;
721                         if (len > IXGBE_RX_HDR_SIZE)
722                                 len = IXGBE_RX_HDR_SIZE;
723                         upper_len = le16_to_cpu(rx_desc->wb.upper.length);
724                 } else {
725                         len = le16_to_cpu(rx_desc->wb.upper.length);
726                 }
727
728                 cleaned = true;
729                 skb = rx_buffer_info->skb;
730                 prefetch(skb->data - NET_IP_ALIGN);
731                 rx_buffer_info->skb = NULL;
732
733                 if (rx_buffer_info->dma) {
734                         pci_unmap_single(pdev, rx_buffer_info->dma,
735                                          rx_ring->rx_buf_len,
736                                          PCI_DMA_FROMDEVICE);
737                         rx_buffer_info->dma = 0;
738                         skb_put(skb, len);
739                 }
740
741                 if (upper_len) {
742                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
743                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
744                         rx_buffer_info->page_dma = 0;
745                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
746                                            rx_buffer_info->page,
747                                            rx_buffer_info->page_offset,
748                                            upper_len);
749
750                         if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
751                             (page_count(rx_buffer_info->page) != 1))
752                                 rx_buffer_info->page = NULL;
753                         else
754                                 get_page(rx_buffer_info->page);
755
756                         skb->len += upper_len;
757                         skb->data_len += upper_len;
758                         skb->truesize += upper_len;
759                 }
760
761                 i++;
762                 if (i == rx_ring->count)
763                         i = 0;
764
765                 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
766                 prefetch(next_rxd);
767                 cleaned_count++;
768
769                 if (adapter->flags & IXGBE_FLAG2_RSC_CAPABLE)
770                         rsc_count = ixgbe_get_rsc_count(rx_desc);
771
772                 if (rsc_count) {
773                         u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
774                                      IXGBE_RXDADV_NEXTP_SHIFT;
775                         next_buffer = &rx_ring->rx_buffer_info[nextp];
776                         rx_ring->rsc_count += (rsc_count - 1);
777                 } else {
778                         next_buffer = &rx_ring->rx_buffer_info[i];
779                 }
780
781                 if (staterr & IXGBE_RXD_STAT_EOP) {
782                         if (skb->prev)
783                                 skb = ixgbe_transform_rsc_queue(skb);
784                         rx_ring->stats.packets++;
785                         rx_ring->stats.bytes += skb->len;
786                 } else {
787                         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
788                                 rx_buffer_info->skb = next_buffer->skb;
789                                 rx_buffer_info->dma = next_buffer->dma;
790                                 next_buffer->skb = skb;
791                                 next_buffer->dma = 0;
792                         } else {
793                                 skb->next = next_buffer->skb;
794                                 skb->next->prev = skb;
795                         }
796                         adapter->non_eop_descs++;
797                         goto next_desc;
798                 }
799
800                 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
801                         dev_kfree_skb_irq(skb);
802                         goto next_desc;
803                 }
804
805                 ixgbe_rx_checksum(adapter, staterr, skb);
806
807                 /* probably a little skewed due to removing CRC */
808                 total_rx_bytes += skb->len;
809                 total_rx_packets++;
810
811                 skb->protocol = eth_type_trans(skb, adapter->netdev);
812 #ifdef IXGBE_FCOE
813                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
814                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
815                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
816                         if (!ddp_bytes)
817                                 goto next_desc;
818                 }
819 #endif /* IXGBE_FCOE */
820                 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
821
822 next_desc:
823                 rx_desc->wb.upper.status_error = 0;
824
825                 /* return some buffers to hardware, one at a time is too slow */
826                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
827                         ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
828                         cleaned_count = 0;
829                 }
830
831                 /* use prefetched values */
832                 rx_desc = next_rxd;
833                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
834
835                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
836         }
837
838         rx_ring->next_to_clean = i;
839         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
840
841         if (cleaned_count)
842                 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
843
844 #ifdef IXGBE_FCOE
845         /* include DDPed FCoE data */
846         if (ddp_bytes > 0) {
847                 unsigned int mss;
848
849                 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
850                         sizeof(struct fc_frame_header) -
851                         sizeof(struct fcoe_crc_eof);
852                 if (mss > 512)
853                         mss &= ~511;
854                 total_rx_bytes += ddp_bytes;
855                 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
856         }
857 #endif /* IXGBE_FCOE */
858
859         rx_ring->total_packets += total_rx_packets;
860         rx_ring->total_bytes += total_rx_bytes;
861         adapter->net_stats.rx_bytes += total_rx_bytes;
862         adapter->net_stats.rx_packets += total_rx_packets;
863
864         return cleaned;
865 }
866
867 static int ixgbe_clean_rxonly(struct napi_struct *, int);
868 /**
869  * ixgbe_configure_msix - Configure MSI-X hardware
870  * @adapter: board private structure
871  *
872  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
873  * interrupts.
874  **/
875 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
876 {
877         struct ixgbe_q_vector *q_vector;
878         int i, j, q_vectors, v_idx, r_idx;
879         u32 mask;
880
881         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
882
883         /*
884          * Populate the IVAR table and set the ITR values to the
885          * corresponding register.
886          */
887         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
888                 q_vector = adapter->q_vector[v_idx];
889                 /* XXX for_each_bit(...) */
890                 r_idx = find_first_bit(q_vector->rxr_idx,
891                                        adapter->num_rx_queues);
892
893                 for (i = 0; i < q_vector->rxr_count; i++) {
894                         j = adapter->rx_ring[r_idx].reg_idx;
895                         ixgbe_set_ivar(adapter, 0, j, v_idx);
896                         r_idx = find_next_bit(q_vector->rxr_idx,
897                                               adapter->num_rx_queues,
898                                               r_idx + 1);
899                 }
900                 r_idx = find_first_bit(q_vector->txr_idx,
901                                        adapter->num_tx_queues);
902
903                 for (i = 0; i < q_vector->txr_count; i++) {
904                         j = adapter->tx_ring[r_idx].reg_idx;
905                         ixgbe_set_ivar(adapter, 1, j, v_idx);
906                         r_idx = find_next_bit(q_vector->txr_idx,
907                                               adapter->num_tx_queues,
908                                               r_idx + 1);
909                 }
910
911                 /* if this is a tx only vector halve the interrupt rate */
912                 if (q_vector->txr_count && !q_vector->rxr_count)
913                         q_vector->eitr = (adapter->eitr_param >> 1);
914                 else if (q_vector->rxr_count)
915                         /* rx only */
916                         q_vector->eitr = adapter->eitr_param;
917
918                 ixgbe_write_eitr(q_vector);
919         }
920
921         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
922                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
923                                v_idx);
924         else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
925                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
926         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
927
928         /* set up to autoclear timer, and the vectors */
929         mask = IXGBE_EIMS_ENABLE_MASK;
930         mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
931         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
932 }
933
934 enum latency_range {
935         lowest_latency = 0,
936         low_latency = 1,
937         bulk_latency = 2,
938         latency_invalid = 255
939 };
940
941 /**
942  * ixgbe_update_itr - update the dynamic ITR value based on statistics
943  * @adapter: pointer to adapter
944  * @eitr: eitr setting (ints per sec) to give last timeslice
945  * @itr_setting: current throttle rate in ints/second
946  * @packets: the number of packets during this measurement interval
947  * @bytes: the number of bytes during this measurement interval
948  *
949  *      Stores a new ITR value based on packets and byte
950  *      counts during the last interrupt.  The advantage of per interrupt
951  *      computation is faster updates and more accurate ITR for the current
952  *      traffic pattern.  Constants in this function were computed
953  *      based on theoretical maximum wire speed and thresholds were set based
954  *      on testing data as well as attempting to minimize response time
955  *      while increasing bulk throughput.
956  *      this functionality is controlled by the InterruptThrottleRate module
957  *      parameter (see ixgbe_param.c)
958  **/
959 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
960                            u32 eitr, u8 itr_setting,
961                            int packets, int bytes)
962 {
963         unsigned int retval = itr_setting;
964         u32 timepassed_us;
965         u64 bytes_perint;
966
967         if (packets == 0)
968                 goto update_itr_done;
969
970
971         /* simple throttlerate management
972          *    0-20MB/s lowest (100000 ints/s)
973          *   20-100MB/s low   (20000 ints/s)
974          *  100-1249MB/s bulk (8000 ints/s)
975          */
976         /* what was last interrupt timeslice? */
977         timepassed_us = 1000000/eitr;
978         bytes_perint = bytes / timepassed_us; /* bytes/usec */
979
980         switch (itr_setting) {
981         case lowest_latency:
982                 if (bytes_perint > adapter->eitr_low)
983                         retval = low_latency;
984                 break;
985         case low_latency:
986                 if (bytes_perint > adapter->eitr_high)
987                         retval = bulk_latency;
988                 else if (bytes_perint <= adapter->eitr_low)
989                         retval = lowest_latency;
990                 break;
991         case bulk_latency:
992                 if (bytes_perint <= adapter->eitr_high)
993                         retval = low_latency;
994                 break;
995         }
996
997 update_itr_done:
998         return retval;
999 }
1000
1001 /**
1002  * ixgbe_write_eitr - write EITR register in hardware specific way
1003  * @q_vector: structure containing interrupt and ring information
1004  *
1005  * This function is made to be called by ethtool and by the driver
1006  * when it needs to update EITR registers at runtime.  Hardware
1007  * specific quirks/differences are taken care of here.
1008  */
1009 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1010 {
1011         struct ixgbe_adapter *adapter = q_vector->adapter;
1012         struct ixgbe_hw *hw = &adapter->hw;
1013         int v_idx = q_vector->v_idx;
1014         u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1015
1016         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1017                 /* must write high and low 16 bits to reset counter */
1018                 itr_reg |= (itr_reg << 16);
1019         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1020                 /*
1021                  * set the WDIS bit to not clear the timer bits and cause an
1022                  * immediate assertion of the interrupt
1023                  */
1024                 itr_reg |= IXGBE_EITR_CNT_WDIS;
1025         }
1026         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1027 }
1028
1029 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1030 {
1031         struct ixgbe_adapter *adapter = q_vector->adapter;
1032         u32 new_itr;
1033         u8 current_itr, ret_itr;
1034         int i, r_idx;
1035         struct ixgbe_ring *rx_ring, *tx_ring;
1036
1037         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1038         for (i = 0; i < q_vector->txr_count; i++) {
1039                 tx_ring = &(adapter->tx_ring[r_idx]);
1040                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1041                                            q_vector->tx_itr,
1042                                            tx_ring->total_packets,
1043                                            tx_ring->total_bytes);
1044                 /* if the result for this queue would decrease interrupt
1045                  * rate for this vector then use that result */
1046                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1047                                     q_vector->tx_itr - 1 : ret_itr);
1048                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1049                                       r_idx + 1);
1050         }
1051
1052         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1053         for (i = 0; i < q_vector->rxr_count; i++) {
1054                 rx_ring = &(adapter->rx_ring[r_idx]);
1055                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1056                                            q_vector->rx_itr,
1057                                            rx_ring->total_packets,
1058                                            rx_ring->total_bytes);
1059                 /* if the result for this queue would decrease interrupt
1060                  * rate for this vector then use that result */
1061                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1062                                     q_vector->rx_itr - 1 : ret_itr);
1063                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1064                                       r_idx + 1);
1065         }
1066
1067         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1068
1069         switch (current_itr) {
1070         /* counts and packets in update_itr are dependent on these numbers */
1071         case lowest_latency:
1072                 new_itr = 100000;
1073                 break;
1074         case low_latency:
1075                 new_itr = 20000; /* aka hwitr = ~200 */
1076                 break;
1077         case bulk_latency:
1078         default:
1079                 new_itr = 8000;
1080                 break;
1081         }
1082
1083         if (new_itr != q_vector->eitr) {
1084                 /* do an exponential smoothing */
1085                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1086
1087                 /* save the algorithm value here, not the smoothed one */
1088                 q_vector->eitr = new_itr;
1089
1090                 ixgbe_write_eitr(q_vector);
1091         }
1092
1093         return;
1094 }
1095
1096 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1097 {
1098         struct ixgbe_hw *hw = &adapter->hw;
1099
1100         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1101             (eicr & IXGBE_EICR_GPI_SDP1)) {
1102                 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
1103                 /* write to clear the interrupt */
1104                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1105         }
1106 }
1107
1108 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1109 {
1110         struct ixgbe_hw *hw = &adapter->hw;
1111
1112         if (eicr & IXGBE_EICR_GPI_SDP1) {
1113                 /* Clear the interrupt */
1114                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1115                 schedule_work(&adapter->multispeed_fiber_task);
1116         } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1117                 /* Clear the interrupt */
1118                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1119                 schedule_work(&adapter->sfp_config_module_task);
1120         } else {
1121                 /* Interrupt isn't for us... */
1122                 return;
1123         }
1124 }
1125
1126 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1127 {
1128         struct ixgbe_hw *hw = &adapter->hw;
1129
1130         adapter->lsc_int++;
1131         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1132         adapter->link_check_timeout = jiffies;
1133         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1134                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1135                 schedule_work(&adapter->watchdog_task);
1136         }
1137 }
1138
1139 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1140 {
1141         struct net_device *netdev = data;
1142         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1143         struct ixgbe_hw *hw = &adapter->hw;
1144         u32 eicr;
1145
1146         /*
1147          * Workaround for Silicon errata.  Use clear-by-write instead
1148          * of clear-by-read.  Reading with EICS will return the
1149          * interrupt causes without clearing, which later be done
1150          * with the write to EICR.
1151          */
1152         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1153         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1154
1155         if (eicr & IXGBE_EICR_LSC)
1156                 ixgbe_check_lsc(adapter);
1157
1158         if (hw->mac.type == ixgbe_mac_82598EB)
1159                 ixgbe_check_fan_failure(adapter, eicr);
1160
1161         if (hw->mac.type == ixgbe_mac_82599EB) {
1162                 ixgbe_check_sfp_event(adapter, eicr);
1163
1164                 /* Handle Flow Director Full threshold interrupt */
1165                 if (eicr & IXGBE_EICR_FLOW_DIR) {
1166                         int i;
1167                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1168                         /* Disable transmits before FDIR Re-initialization */
1169                         netif_tx_stop_all_queues(netdev);
1170                         for (i = 0; i < adapter->num_tx_queues; i++) {
1171                                 struct ixgbe_ring *tx_ring =
1172                                                            &adapter->tx_ring[i];
1173                                 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1174                                                        &tx_ring->reinit_state))
1175                                         schedule_work(&adapter->fdir_reinit_task);
1176                         }
1177                 }
1178         }
1179         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1180                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1181
1182         return IRQ_HANDLED;
1183 }
1184
1185 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1186                                            u64 qmask)
1187 {
1188         u32 mask;
1189
1190         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1191                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1192                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1193         } else {
1194                 mask = (qmask & 0xFFFFFFFF);
1195                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1196                 mask = (qmask >> 32);
1197                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1198         }
1199         /* skip the flush */
1200 }
1201
1202 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1203                                             u64 qmask)
1204 {
1205         u32 mask;
1206
1207         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1208                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1209                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1210         } else {
1211                 mask = (qmask & 0xFFFFFFFF);
1212                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1213                 mask = (qmask >> 32);
1214                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1215         }
1216         /* skip the flush */
1217 }
1218
1219 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1220 {
1221         struct ixgbe_q_vector *q_vector = data;
1222         struct ixgbe_adapter  *adapter = q_vector->adapter;
1223         struct ixgbe_ring     *tx_ring;
1224         int i, r_idx;
1225
1226         if (!q_vector->txr_count)
1227                 return IRQ_HANDLED;
1228
1229         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1230         for (i = 0; i < q_vector->txr_count; i++) {
1231                 tx_ring = &(adapter->tx_ring[r_idx]);
1232                 tx_ring->total_bytes = 0;
1233                 tx_ring->total_packets = 0;
1234                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1235                                       r_idx + 1);
1236         }
1237
1238         /* disable interrupts on this vector only */
1239         ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1240         napi_schedule(&q_vector->napi);
1241
1242         return IRQ_HANDLED;
1243 }
1244
1245 /**
1246  * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1247  * @irq: unused
1248  * @data: pointer to our q_vector struct for this interrupt vector
1249  **/
1250 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1251 {
1252         struct ixgbe_q_vector *q_vector = data;
1253         struct ixgbe_adapter  *adapter = q_vector->adapter;
1254         struct ixgbe_ring  *rx_ring;
1255         int r_idx;
1256         int i;
1257
1258         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1259         for (i = 0;  i < q_vector->rxr_count; i++) {
1260                 rx_ring = &(adapter->rx_ring[r_idx]);
1261                 rx_ring->total_bytes = 0;
1262                 rx_ring->total_packets = 0;
1263                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1264                                       r_idx + 1);
1265         }
1266
1267         if (!q_vector->rxr_count)
1268                 return IRQ_HANDLED;
1269
1270         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1271         rx_ring = &(adapter->rx_ring[r_idx]);
1272         /* disable interrupts on this vector only */
1273         ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1274         napi_schedule(&q_vector->napi);
1275
1276         return IRQ_HANDLED;
1277 }
1278
1279 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1280 {
1281         struct ixgbe_q_vector *q_vector = data;
1282         struct ixgbe_adapter  *adapter = q_vector->adapter;
1283         struct ixgbe_ring  *ring;
1284         int r_idx;
1285         int i;
1286
1287         if (!q_vector->txr_count && !q_vector->rxr_count)
1288                 return IRQ_HANDLED;
1289
1290         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1291         for (i = 0; i < q_vector->txr_count; i++) {
1292                 ring = &(adapter->tx_ring[r_idx]);
1293                 ring->total_bytes = 0;
1294                 ring->total_packets = 0;
1295                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1296                                       r_idx + 1);
1297         }
1298
1299         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1300         for (i = 0; i < q_vector->rxr_count; i++) {
1301                 ring = &(adapter->rx_ring[r_idx]);
1302                 ring->total_bytes = 0;
1303                 ring->total_packets = 0;
1304                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1305                                       r_idx + 1);
1306         }
1307
1308         /* disable interrupts on this vector only */
1309         ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1310         napi_schedule(&q_vector->napi);
1311
1312         return IRQ_HANDLED;
1313 }
1314
1315 /**
1316  * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1317  * @napi: napi struct with our devices info in it
1318  * @budget: amount of work driver is allowed to do this pass, in packets
1319  *
1320  * This function is optimized for cleaning one queue only on a single
1321  * q_vector!!!
1322  **/
1323 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1324 {
1325         struct ixgbe_q_vector *q_vector =
1326                                container_of(napi, struct ixgbe_q_vector, napi);
1327         struct ixgbe_adapter *adapter = q_vector->adapter;
1328         struct ixgbe_ring *rx_ring = NULL;
1329         int work_done = 0;
1330         long r_idx;
1331
1332         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1333         rx_ring = &(adapter->rx_ring[r_idx]);
1334 #ifdef CONFIG_IXGBE_DCA
1335         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1336                 ixgbe_update_rx_dca(adapter, rx_ring);
1337 #endif
1338
1339         ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1340
1341         /* If all Rx work done, exit the polling mode */
1342         if (work_done < budget) {
1343                 napi_complete(napi);
1344                 if (adapter->itr_setting & 1)
1345                         ixgbe_set_itr_msix(q_vector);
1346                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1347                         ixgbe_irq_enable_queues(adapter,
1348                                                 ((u64)1 << q_vector->v_idx));
1349         }
1350
1351         return work_done;
1352 }
1353
1354 /**
1355  * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1356  * @napi: napi struct with our devices info in it
1357  * @budget: amount of work driver is allowed to do this pass, in packets
1358  *
1359  * This function will clean more than one rx queue associated with a
1360  * q_vector.
1361  **/
1362 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1363 {
1364         struct ixgbe_q_vector *q_vector =
1365                                container_of(napi, struct ixgbe_q_vector, napi);
1366         struct ixgbe_adapter *adapter = q_vector->adapter;
1367         struct ixgbe_ring *ring = NULL;
1368         int work_done = 0, i;
1369         long r_idx;
1370         bool tx_clean_complete = true;
1371
1372         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1373         for (i = 0; i < q_vector->txr_count; i++) {
1374                 ring = &(adapter->tx_ring[r_idx]);
1375 #ifdef CONFIG_IXGBE_DCA
1376                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1377                         ixgbe_update_tx_dca(adapter, ring);
1378 #endif
1379                 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1380                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1381                                       r_idx + 1);
1382         }
1383
1384         /* attempt to distribute budget to each queue fairly, but don't allow
1385          * the budget to go below 1 because we'll exit polling */
1386         budget /= (q_vector->rxr_count ?: 1);
1387         budget = max(budget, 1);
1388         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1389         for (i = 0; i < q_vector->rxr_count; i++) {
1390                 ring = &(adapter->rx_ring[r_idx]);
1391 #ifdef CONFIG_IXGBE_DCA
1392                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1393                         ixgbe_update_rx_dca(adapter, ring);
1394 #endif
1395                 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1396                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1397                                       r_idx + 1);
1398         }
1399
1400         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1401         ring = &(adapter->rx_ring[r_idx]);
1402         /* If all Rx work done, exit the polling mode */
1403         if (work_done < budget) {
1404                 napi_complete(napi);
1405                 if (adapter->itr_setting & 1)
1406                         ixgbe_set_itr_msix(q_vector);
1407                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1408                         ixgbe_irq_enable_queues(adapter,
1409                                                 ((u64)1 << q_vector->v_idx));
1410                 return 0;
1411         }
1412
1413         return work_done;
1414 }
1415
1416 /**
1417  * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1418  * @napi: napi struct with our devices info in it
1419  * @budget: amount of work driver is allowed to do this pass, in packets
1420  *
1421  * This function is optimized for cleaning one queue only on a single
1422  * q_vector!!!
1423  **/
1424 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1425 {
1426         struct ixgbe_q_vector *q_vector =
1427                                container_of(napi, struct ixgbe_q_vector, napi);
1428         struct ixgbe_adapter *adapter = q_vector->adapter;
1429         struct ixgbe_ring *tx_ring = NULL;
1430         int work_done = 0;
1431         long r_idx;
1432
1433         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1434         tx_ring = &(adapter->tx_ring[r_idx]);
1435 #ifdef CONFIG_IXGBE_DCA
1436         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1437                 ixgbe_update_tx_dca(adapter, tx_ring);
1438 #endif
1439
1440         if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
1441                 work_done = budget;
1442
1443         /* If all Rx work done, exit the polling mode */
1444         if (work_done < budget) {
1445                 napi_complete(napi);
1446                 if (adapter->itr_setting & 1)
1447                         ixgbe_set_itr_msix(q_vector);
1448                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1449                         ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
1450         }
1451
1452         return work_done;
1453 }
1454
1455 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1456                                      int r_idx)
1457 {
1458         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1459
1460         set_bit(r_idx, q_vector->rxr_idx);
1461         q_vector->rxr_count++;
1462 }
1463
1464 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1465                                      int t_idx)
1466 {
1467         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1468
1469         set_bit(t_idx, q_vector->txr_idx);
1470         q_vector->txr_count++;
1471 }
1472
1473 /**
1474  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1475  * @adapter: board private structure to initialize
1476  * @vectors: allotted vector count for descriptor rings
1477  *
1478  * This function maps descriptor rings to the queue-specific vectors
1479  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1480  * one vector per ring/queue, but on a constrained vector budget, we
1481  * group the rings as "efficiently" as possible.  You would add new
1482  * mapping configurations in here.
1483  **/
1484 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1485                                       int vectors)
1486 {
1487         int v_start = 0;
1488         int rxr_idx = 0, txr_idx = 0;
1489         int rxr_remaining = adapter->num_rx_queues;
1490         int txr_remaining = adapter->num_tx_queues;
1491         int i, j;
1492         int rqpv, tqpv;
1493         int err = 0;
1494
1495         /* No mapping required if MSI-X is disabled. */
1496         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1497                 goto out;
1498
1499         /*
1500          * The ideal configuration...
1501          * We have enough vectors to map one per queue.
1502          */
1503         if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1504                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1505                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1506
1507                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1508                         map_vector_to_txq(adapter, v_start, txr_idx);
1509
1510                 goto out;
1511         }
1512
1513         /*
1514          * If we don't have enough vectors for a 1-to-1
1515          * mapping, we'll have to group them so there are
1516          * multiple queues per vector.
1517          */
1518         /* Re-adjusting *qpv takes care of the remainder. */
1519         for (i = v_start; i < vectors; i++) {
1520                 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1521                 for (j = 0; j < rqpv; j++) {
1522                         map_vector_to_rxq(adapter, i, rxr_idx);
1523                         rxr_idx++;
1524                         rxr_remaining--;
1525                 }
1526         }
1527         for (i = v_start; i < vectors; i++) {
1528                 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1529                 for (j = 0; j < tqpv; j++) {
1530                         map_vector_to_txq(adapter, i, txr_idx);
1531                         txr_idx++;
1532                         txr_remaining--;
1533                 }
1534         }
1535
1536 out:
1537         return err;
1538 }
1539
1540 /**
1541  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1542  * @adapter: board private structure
1543  *
1544  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1545  * interrupts from the kernel.
1546  **/
1547 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1548 {
1549         struct net_device *netdev = adapter->netdev;
1550         irqreturn_t (*handler)(int, void *);
1551         int i, vector, q_vectors, err;
1552         int ri=0, ti=0;
1553
1554         /* Decrement for Other and TCP Timer vectors */
1555         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1556
1557         /* Map the Tx/Rx rings to the vectors we were allotted. */
1558         err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1559         if (err)
1560                 goto out;
1561
1562 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1563                          (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1564                          &ixgbe_msix_clean_many)
1565         for (vector = 0; vector < q_vectors; vector++) {
1566                 handler = SET_HANDLER(adapter->q_vector[vector]);
1567
1568                 if(handler == &ixgbe_msix_clean_rx) {
1569                         sprintf(adapter->name[vector], "%s-%s-%d",
1570                                 netdev->name, "rx", ri++);
1571                 }
1572                 else if(handler == &ixgbe_msix_clean_tx) {
1573                         sprintf(adapter->name[vector], "%s-%s-%d",
1574                                 netdev->name, "tx", ti++);
1575                 }
1576                 else
1577                         sprintf(adapter->name[vector], "%s-%s-%d",
1578                                 netdev->name, "TxRx", vector);
1579
1580                 err = request_irq(adapter->msix_entries[vector].vector,
1581                                   handler, 0, adapter->name[vector],
1582                                   adapter->q_vector[vector]);
1583                 if (err) {
1584                         DPRINTK(PROBE, ERR,
1585                                 "request_irq failed for MSIX interrupt "
1586                                 "Error: %d\n", err);
1587                         goto free_queue_irqs;
1588                 }
1589         }
1590
1591         sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1592         err = request_irq(adapter->msix_entries[vector].vector,
1593                           &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1594         if (err) {
1595                 DPRINTK(PROBE, ERR,
1596                         "request_irq for msix_lsc failed: %d\n", err);
1597                 goto free_queue_irqs;
1598         }
1599
1600         return 0;
1601
1602 free_queue_irqs:
1603         for (i = vector - 1; i >= 0; i--)
1604                 free_irq(adapter->msix_entries[--vector].vector,
1605                          adapter->q_vector[i]);
1606         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1607         pci_disable_msix(adapter->pdev);
1608         kfree(adapter->msix_entries);
1609         adapter->msix_entries = NULL;
1610 out:
1611         return err;
1612 }
1613
1614 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1615 {
1616         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1617         u8 current_itr;
1618         u32 new_itr = q_vector->eitr;
1619         struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1620         struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1621
1622         q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1623                                             q_vector->tx_itr,
1624                                             tx_ring->total_packets,
1625                                             tx_ring->total_bytes);
1626         q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1627                                             q_vector->rx_itr,
1628                                             rx_ring->total_packets,
1629                                             rx_ring->total_bytes);
1630
1631         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1632
1633         switch (current_itr) {
1634         /* counts and packets in update_itr are dependent on these numbers */
1635         case lowest_latency:
1636                 new_itr = 100000;
1637                 break;
1638         case low_latency:
1639                 new_itr = 20000; /* aka hwitr = ~200 */
1640                 break;
1641         case bulk_latency:
1642                 new_itr = 8000;
1643                 break;
1644         default:
1645                 break;
1646         }
1647
1648         if (new_itr != q_vector->eitr) {
1649                 /* do an exponential smoothing */
1650                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1651
1652                 /* save the algorithm value here, not the smoothed one */
1653                 q_vector->eitr = new_itr;
1654
1655                 ixgbe_write_eitr(q_vector);
1656         }
1657
1658         return;
1659 }
1660
1661 /**
1662  * ixgbe_irq_enable - Enable default interrupt generation settings
1663  * @adapter: board private structure
1664  **/
1665 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1666 {
1667         u32 mask;
1668
1669         mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1670         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1671                 mask |= IXGBE_EIMS_GPI_SDP1;
1672         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1673                 mask |= IXGBE_EIMS_ECC;
1674                 mask |= IXGBE_EIMS_GPI_SDP1;
1675                 mask |= IXGBE_EIMS_GPI_SDP2;
1676         }
1677         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
1678             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
1679                 mask |= IXGBE_EIMS_FLOW_DIR;
1680
1681         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1682         ixgbe_irq_enable_queues(adapter, ~0);
1683         IXGBE_WRITE_FLUSH(&adapter->hw);
1684 }
1685
1686 /**
1687  * ixgbe_intr - legacy mode Interrupt Handler
1688  * @irq: interrupt number
1689  * @data: pointer to a network interface device structure
1690  **/
1691 static irqreturn_t ixgbe_intr(int irq, void *data)
1692 {
1693         struct net_device *netdev = data;
1694         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1695         struct ixgbe_hw *hw = &adapter->hw;
1696         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1697         u32 eicr;
1698
1699         /*
1700          * Workaround for silicon errata.  Mask the interrupts
1701          * before the read of EICR.
1702          */
1703         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1704
1705         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1706          * therefore no explict interrupt disable is necessary */
1707         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1708         if (!eicr) {
1709                 /* shared interrupt alert!
1710                  * make sure interrupts are enabled because the read will
1711                  * have disabled interrupts due to EIAM */
1712                 ixgbe_irq_enable(adapter);
1713                 return IRQ_NONE;        /* Not our interrupt */
1714         }
1715
1716         if (eicr & IXGBE_EICR_LSC)
1717                 ixgbe_check_lsc(adapter);
1718
1719         if (hw->mac.type == ixgbe_mac_82599EB)
1720                 ixgbe_check_sfp_event(adapter, eicr);
1721
1722         ixgbe_check_fan_failure(adapter, eicr);
1723
1724         if (napi_schedule_prep(&(q_vector->napi))) {
1725                 adapter->tx_ring[0].total_packets = 0;
1726                 adapter->tx_ring[0].total_bytes = 0;
1727                 adapter->rx_ring[0].total_packets = 0;
1728                 adapter->rx_ring[0].total_bytes = 0;
1729                 /* would disable interrupts here but EIAM disabled it */
1730                 __napi_schedule(&(q_vector->napi));
1731         }
1732
1733         return IRQ_HANDLED;
1734 }
1735
1736 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1737 {
1738         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1739
1740         for (i = 0; i < q_vectors; i++) {
1741                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
1742                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1743                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1744                 q_vector->rxr_count = 0;
1745                 q_vector->txr_count = 0;
1746         }
1747 }
1748
1749 /**
1750  * ixgbe_request_irq - initialize interrupts
1751  * @adapter: board private structure
1752  *
1753  * Attempts to configure interrupts using the best available
1754  * capabilities of the hardware and kernel.
1755  **/
1756 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1757 {
1758         struct net_device *netdev = adapter->netdev;
1759         int err;
1760
1761         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1762                 err = ixgbe_request_msix_irqs(adapter);
1763         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1764                 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1765                                   netdev->name, netdev);
1766         } else {
1767                 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1768                                   netdev->name, netdev);
1769         }
1770
1771         if (err)
1772                 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1773
1774         return err;
1775 }
1776
1777 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1778 {
1779         struct net_device *netdev = adapter->netdev;
1780
1781         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1782                 int i, q_vectors;
1783
1784                 q_vectors = adapter->num_msix_vectors;
1785
1786                 i = q_vectors - 1;
1787                 free_irq(adapter->msix_entries[i].vector, netdev);
1788
1789                 i--;
1790                 for (; i >= 0; i--) {
1791                         free_irq(adapter->msix_entries[i].vector,
1792                                  adapter->q_vector[i]);
1793                 }
1794
1795                 ixgbe_reset_q_vectors(adapter);
1796         } else {
1797                 free_irq(adapter->pdev->irq, netdev);
1798         }
1799 }
1800
1801 /**
1802  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1803  * @adapter: board private structure
1804  **/
1805 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1806 {
1807         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1808                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1809         } else {
1810                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
1811                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
1812                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
1813         }
1814         IXGBE_WRITE_FLUSH(&adapter->hw);
1815         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1816                 int i;
1817                 for (i = 0; i < adapter->num_msix_vectors; i++)
1818                         synchronize_irq(adapter->msix_entries[i].vector);
1819         } else {
1820                 synchronize_irq(adapter->pdev->irq);
1821         }
1822 }
1823
1824 /**
1825  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1826  *
1827  **/
1828 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1829 {
1830         struct ixgbe_hw *hw = &adapter->hw;
1831
1832         IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1833                         EITR_INTS_PER_SEC_TO_REG(adapter->eitr_param));
1834
1835         ixgbe_set_ivar(adapter, 0, 0, 0);
1836         ixgbe_set_ivar(adapter, 1, 0, 0);
1837
1838         map_vector_to_rxq(adapter, 0, 0);
1839         map_vector_to_txq(adapter, 0, 0);
1840
1841         DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1842 }
1843
1844 /**
1845  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1846  * @adapter: board private structure
1847  *
1848  * Configure the Tx unit of the MAC after a reset.
1849  **/
1850 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1851 {
1852         u64 tdba;
1853         struct ixgbe_hw *hw = &adapter->hw;
1854         u32 i, j, tdlen, txctrl;
1855
1856         /* Setup the HW Tx Head and Tail descriptor pointers */
1857         for (i = 0; i < adapter->num_tx_queues; i++) {
1858                 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1859                 j = ring->reg_idx;
1860                 tdba = ring->dma;
1861                 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1862                 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1863                                 (tdba & DMA_BIT_MASK(32)));
1864                 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1865                 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1866                 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1867                 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1868                 adapter->tx_ring[i].head = IXGBE_TDH(j);
1869                 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1870                 /* Disable Tx Head Writeback RO bit, since this hoses
1871                  * bookkeeping if things aren't delivered in order.
1872                  */
1873                 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1874                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1875                 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1876         }
1877         if (hw->mac.type == ixgbe_mac_82599EB) {
1878                 /* We enable 8 traffic classes, DCB only */
1879                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
1880                         IXGBE_WRITE_REG(hw, IXGBE_MTQC, (IXGBE_MTQC_RT_ENA |
1881                                         IXGBE_MTQC_8TC_8TQ));
1882         }
1883 }
1884
1885 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1886
1887 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter, int index)
1888 {
1889         struct ixgbe_ring *rx_ring;
1890         u32 srrctl;
1891         int queue0 = 0;
1892         unsigned long mask;
1893         struct ixgbe_ring_feature *feature = adapter->ring_feature;
1894
1895         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1896                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
1897                         int dcb_i = feature[RING_F_DCB].indices;
1898                         if (dcb_i == 8)
1899                                 queue0 = index >> 4;
1900                         else if (dcb_i == 4)
1901                                 queue0 = index >> 5;
1902                         else
1903                                 dev_err(&adapter->pdev->dev, "Invalid DCB "
1904                                         "configuration\n");
1905 #ifdef IXGBE_FCOE
1906                         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1907                                 struct ixgbe_ring_feature *f;
1908
1909                                 rx_ring = &adapter->rx_ring[queue0];
1910                                 f = &adapter->ring_feature[RING_F_FCOE];
1911                                 if ((queue0 == 0) && (index > rx_ring->reg_idx))
1912                                         queue0 = f->mask + index -
1913                                                  rx_ring->reg_idx - 1;
1914                         }
1915 #endif /* IXGBE_FCOE */
1916                 } else {
1917                         queue0 = index;
1918                 }
1919         } else {
1920                 mask = (unsigned long) feature[RING_F_RSS].mask;
1921                 queue0 = index & mask;
1922                 index = index & mask;
1923         }
1924
1925         rx_ring = &adapter->rx_ring[queue0];
1926
1927         srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
1928
1929         srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1930         srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1931
1932         srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1933                   IXGBE_SRRCTL_BSIZEHDR_MASK;
1934
1935         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1936 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
1937                 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1938 #else
1939                 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1940 #endif
1941                 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1942         } else {
1943                 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
1944                           IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1945                 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1946         }
1947
1948         IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
1949 }
1950
1951 static u32 ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
1952 {
1953         u32 mrqc = 0;
1954         int mask;
1955
1956         if (!(adapter->hw.mac.type == ixgbe_mac_82599EB))
1957                 return mrqc;
1958
1959         mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
1960 #ifdef CONFIG_IXGBE_DCB
1961                                  | IXGBE_FLAG_DCB_ENABLED
1962 #endif
1963                                 );
1964
1965         switch (mask) {
1966         case (IXGBE_FLAG_RSS_ENABLED):
1967                 mrqc = IXGBE_MRQC_RSSEN;
1968                 break;
1969 #ifdef CONFIG_IXGBE_DCB
1970         case (IXGBE_FLAG_DCB_ENABLED):
1971                 mrqc = IXGBE_MRQC_RT8TCEN;
1972                 break;
1973 #endif /* CONFIG_IXGBE_DCB */
1974         default:
1975                 break;
1976         }
1977
1978         return mrqc;
1979 }
1980
1981 /**
1982  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
1983  * @adapter: board private structure
1984  *
1985  * Configure the Rx unit of the MAC after a reset.
1986  **/
1987 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
1988 {
1989         u64 rdba;
1990         struct ixgbe_hw *hw = &adapter->hw;
1991         struct net_device *netdev = adapter->netdev;
1992         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1993         int i, j;
1994         u32 rdlen, rxctrl, rxcsum;
1995         static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
1996                           0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
1997                           0x6A3E67EA, 0x14364D17, 0x3BED200D};
1998         u32 fctrl, hlreg0;
1999         u32 reta = 0, mrqc = 0;
2000         u32 rdrxctl;
2001         u32 rscctrl;
2002         int rx_buf_len;
2003
2004         /* Decide whether to use packet split mode or not */
2005         adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2006
2007 #ifdef IXGBE_FCOE
2008         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2009                 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2010 #endif /* IXGBE_FCOE */
2011
2012         /* Set the RX buffer length according to the mode */
2013         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2014                 rx_buf_len = IXGBE_RX_HDR_SIZE;
2015                 if (hw->mac.type == ixgbe_mac_82599EB) {
2016                         /* PSRTYPE must be initialized in 82599 */
2017                         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2018                                       IXGBE_PSRTYPE_UDPHDR |
2019                                       IXGBE_PSRTYPE_IPV4HDR |
2020                                       IXGBE_PSRTYPE_IPV6HDR |
2021                                       IXGBE_PSRTYPE_L2HDR;
2022                         IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), psrtype);
2023                 }
2024         } else {
2025                 if (!(adapter->flags & IXGBE_FLAG2_RSC_ENABLED) &&
2026                     (netdev->mtu <= ETH_DATA_LEN))
2027                         rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2028                 else
2029                         rx_buf_len = ALIGN(max_frame, 1024);
2030         }
2031
2032         fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
2033         fctrl |= IXGBE_FCTRL_BAM;
2034         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
2035         fctrl |= IXGBE_FCTRL_PMCF;
2036         IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
2037
2038         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2039         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2040                 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2041         else
2042                 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2043 #ifdef IXGBE_FCOE
2044         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2045                 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2046 #endif
2047         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2048
2049         rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
2050         /* disable receives while setting up the descriptors */
2051         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2052         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2053
2054         /*
2055          * Setup the HW Rx Head and Tail Descriptor Pointers and
2056          * the Base and Length of the Rx Descriptor Ring
2057          */
2058         for (i = 0; i < adapter->num_rx_queues; i++) {
2059                 rdba = adapter->rx_ring[i].dma;
2060                 j = adapter->rx_ring[i].reg_idx;
2061                 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_BIT_MASK(32)));
2062                 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
2063                 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
2064                 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
2065                 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
2066                 adapter->rx_ring[i].head = IXGBE_RDH(j);
2067                 adapter->rx_ring[i].tail = IXGBE_RDT(j);
2068                 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
2069
2070 #ifdef IXGBE_FCOE
2071                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
2072                         struct ixgbe_ring_feature *f;
2073                         f = &adapter->ring_feature[RING_F_FCOE];
2074                         if ((rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
2075                             (i >= f->mask) && (i < f->mask + f->indices))
2076                                 adapter->rx_ring[i].rx_buf_len =
2077                                         IXGBE_FCOE_JUMBO_FRAME_SIZE;
2078                 }
2079
2080 #endif /* IXGBE_FCOE */
2081                 ixgbe_configure_srrctl(adapter, j);
2082         }
2083
2084         if (hw->mac.type == ixgbe_mac_82598EB) {
2085                 /*
2086                  * For VMDq support of different descriptor types or
2087                  * buffer sizes through the use of multiple SRRCTL
2088                  * registers, RDRXCTL.MVMEN must be set to 1
2089                  *
2090                  * also, the manual doesn't mention it clearly but DCA hints
2091                  * will only use queue 0's tags unless this bit is set.  Side
2092                  * effects of setting this bit are only that SRRCTL must be
2093                  * fully programmed [0..15]
2094                  */
2095                 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2096                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2097                 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2098         }
2099
2100         /* Program MRQC for the distribution of queues */
2101         mrqc = ixgbe_setup_mrqc(adapter);
2102
2103         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2104                 /* Fill out redirection table */
2105                 for (i = 0, j = 0; i < 128; i++, j++) {
2106                         if (j == adapter->ring_feature[RING_F_RSS].indices)
2107                                 j = 0;
2108                         /* reta = 4-byte sliding window of
2109                          * 0x00..(indices-1)(indices-1)00..etc. */
2110                         reta = (reta << 8) | (j * 0x11);
2111                         if ((i & 3) == 3)
2112                                 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2113                 }
2114
2115                 /* Fill out hash function seeds */
2116                 for (i = 0; i < 10; i++)
2117                         IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2118
2119                 if (hw->mac.type == ixgbe_mac_82598EB)
2120                         mrqc |= IXGBE_MRQC_RSSEN;
2121                     /* Perform hash on these packet types */
2122                 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2123                       | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2124                       | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
2125                       | IXGBE_MRQC_RSS_FIELD_IPV6
2126                       | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
2127                       | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2128         }
2129         IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2130
2131         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2132
2133         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
2134             adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
2135                 /* Disable indicating checksum in descriptor, enables
2136                  * RSS hash */
2137                 rxcsum |= IXGBE_RXCSUM_PCSD;
2138         }
2139         if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
2140                 /* Enable IPv4 payload checksum for UDP fragments
2141                  * if PCSD is not set */
2142                 rxcsum |= IXGBE_RXCSUM_IPPCSE;
2143         }
2144
2145         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2146
2147         if (hw->mac.type == ixgbe_mac_82599EB) {
2148                 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2149                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2150                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2151                 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2152         }
2153
2154         if (adapter->flags & IXGBE_FLAG2_RSC_ENABLED) {
2155                 /* Enable 82599 HW-RSC */
2156                 for (i = 0; i < adapter->num_rx_queues; i++) {
2157                         j = adapter->rx_ring[i].reg_idx;
2158                         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2159                         rscctrl |= IXGBE_RSCCTL_RSCEN;
2160                         /*
2161                          * we must limit the number of descriptors so that the
2162                          * total size of max desc * buf_len is not greater
2163                          * than 65535
2164                          */
2165                         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2166 #if (MAX_SKB_FRAGS > 16)
2167                                 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2168 #elif (MAX_SKB_FRAGS > 8)
2169                                 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2170 #elif (MAX_SKB_FRAGS > 4)
2171                                 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2172 #else
2173                                 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2174 #endif
2175                         } else {
2176                                 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2177                                         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2178                                 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2179                                         rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2180                                 else
2181                                         rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2182                         }
2183                         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2184                 }
2185                 /* Disable RSC for ACK packets */
2186                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2187                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2188         }
2189 }
2190
2191 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2192 {
2193         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2194         struct ixgbe_hw *hw = &adapter->hw;
2195
2196         /* add VID to filter table */
2197         hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
2198 }
2199
2200 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2201 {
2202         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2203         struct ixgbe_hw *hw = &adapter->hw;
2204
2205         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2206                 ixgbe_irq_disable(adapter);
2207
2208         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2209
2210         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2211                 ixgbe_irq_enable(adapter);
2212
2213         /* remove VID from filter table */
2214         hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
2215 }
2216
2217 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2218                                    struct vlan_group *grp)
2219 {
2220         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2221         u32 ctrl;
2222         int i, j;
2223
2224         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2225                 ixgbe_irq_disable(adapter);
2226         adapter->vlgrp = grp;
2227
2228         /*
2229          * For a DCB driver, always enable VLAN tag stripping so we can
2230          * still receive traffic from a DCB-enabled host even if we're
2231          * not in DCB mode.
2232          */
2233         ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2234         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2235                 ctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2236                 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2237                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2238         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2239                 ctrl |= IXGBE_VLNCTRL_VFE;
2240                 /* enable VLAN tag insert/strip */
2241                 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2242                 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2243                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2244                 for (i = 0; i < adapter->num_rx_queues; i++) {
2245                         j = adapter->rx_ring[i].reg_idx;
2246                         ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_RXDCTL(j));
2247                         ctrl |= IXGBE_RXDCTL_VME;
2248                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_RXDCTL(j), ctrl);
2249                 }
2250         }
2251         ixgbe_vlan_rx_add_vid(netdev, 0);
2252
2253         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2254                 ixgbe_irq_enable(adapter);
2255 }
2256
2257 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2258 {
2259         ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2260
2261         if (adapter->vlgrp) {
2262                 u16 vid;
2263                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2264                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2265                                 continue;
2266                         ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2267                 }
2268         }
2269 }
2270
2271 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
2272 {
2273         struct dev_mc_list *mc_ptr;
2274         u8 *addr = *mc_addr_ptr;
2275         *vmdq = 0;
2276
2277         mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
2278         if (mc_ptr->next)
2279                 *mc_addr_ptr = mc_ptr->next->dmi_addr;
2280         else
2281                 *mc_addr_ptr = NULL;
2282
2283         return addr;
2284 }
2285
2286 /**
2287  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
2288  * @netdev: network interface device structure
2289  *
2290  * The set_rx_method entry point is called whenever the unicast/multicast
2291  * address list or the network interface flags are updated.  This routine is
2292  * responsible for configuring the hardware for proper unicast, multicast and
2293  * promiscuous mode.
2294  **/
2295 static void ixgbe_set_rx_mode(struct net_device *netdev)
2296 {
2297         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2298         struct ixgbe_hw *hw = &adapter->hw;
2299         u32 fctrl, vlnctrl;
2300         u8 *addr_list = NULL;
2301         int addr_count = 0;
2302
2303         /* Check for Promiscuous and All Multicast modes */
2304
2305         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2306         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2307
2308         if (netdev->flags & IFF_PROMISC) {
2309                 hw->addr_ctrl.user_set_promisc = 1;
2310                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2311                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2312         } else {
2313                 if (netdev->flags & IFF_ALLMULTI) {
2314                         fctrl |= IXGBE_FCTRL_MPE;
2315                         fctrl &= ~IXGBE_FCTRL_UPE;
2316                 } else {
2317                         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2318                 }
2319                 vlnctrl |= IXGBE_VLNCTRL_VFE;
2320                 hw->addr_ctrl.user_set_promisc = 0;
2321         }
2322
2323         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
2324         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2325
2326         /* reprogram secondary unicast list */
2327         hw->mac.ops.update_uc_addr_list(hw, &netdev->uc.list);
2328
2329         /* reprogram multicast list */
2330         addr_count = netdev->mc_count;
2331         if (addr_count)
2332                 addr_list = netdev->mc_list->dmi_addr;
2333         hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
2334                                         ixgbe_addr_list_itr);
2335 }
2336
2337 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
2338 {
2339         int q_idx;
2340         struct ixgbe_q_vector *q_vector;
2341         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2342
2343         /* legacy and MSI only use one vector */
2344         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2345                 q_vectors = 1;
2346
2347         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2348                 struct napi_struct *napi;
2349                 q_vector = adapter->q_vector[q_idx];
2350                 napi = &q_vector->napi;
2351                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2352                         if (!q_vector->rxr_count || !q_vector->txr_count) {
2353                                 if (q_vector->txr_count == 1)
2354                                         napi->poll = &ixgbe_clean_txonly;
2355                                 else if (q_vector->rxr_count == 1)
2356                                         napi->poll = &ixgbe_clean_rxonly;
2357                         }
2358                 }
2359
2360                 napi_enable(napi);
2361         }
2362 }
2363
2364 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2365 {
2366         int q_idx;
2367         struct ixgbe_q_vector *q_vector;
2368         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2369
2370         /* legacy and MSI only use one vector */
2371         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2372                 q_vectors = 1;
2373
2374         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2375                 q_vector = adapter->q_vector[q_idx];
2376                 napi_disable(&q_vector->napi);
2377         }
2378 }
2379
2380 #ifdef CONFIG_IXGBE_DCB
2381 /*
2382  * ixgbe_configure_dcb - Configure DCB hardware
2383  * @adapter: ixgbe adapter struct
2384  *
2385  * This is called by the driver on open to configure the DCB hardware.
2386  * This is also called by the gennetlink interface when reconfiguring
2387  * the DCB state.
2388  */
2389 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2390 {
2391         struct ixgbe_hw *hw = &adapter->hw;
2392         u32 txdctl, vlnctrl;
2393         int i, j;
2394
2395         ixgbe_dcb_check_config(&adapter->dcb_cfg);
2396         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2397         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2398
2399         /* reconfigure the hardware */
2400         ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
2401
2402         for (i = 0; i < adapter->num_tx_queues; i++) {
2403                 j = adapter->tx_ring[i].reg_idx;
2404                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2405                 /* PThresh workaround for Tx hang with DFP enabled. */
2406                 txdctl |= 32;
2407                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2408         }
2409         /* Enable VLAN tag insert/strip */
2410         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2411         if (hw->mac.type == ixgbe_mac_82598EB) {
2412                 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2413                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2414                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2415         } else if (hw->mac.type == ixgbe_mac_82599EB) {
2416                 vlnctrl |= IXGBE_VLNCTRL_VFE;
2417                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2418                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2419                 for (i = 0; i < adapter->num_rx_queues; i++) {
2420                         j = adapter->rx_ring[i].reg_idx;
2421                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2422                         vlnctrl |= IXGBE_RXDCTL_VME;
2423                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2424                 }
2425         }
2426         hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2427 }
2428
2429 #endif
2430 static void ixgbe_configure(struct ixgbe_adapter *adapter)
2431 {
2432         struct net_device *netdev = adapter->netdev;
2433         struct ixgbe_hw *hw = &adapter->hw;
2434         int i;
2435
2436         ixgbe_set_rx_mode(netdev);
2437
2438         ixgbe_restore_vlan(adapter);
2439 #ifdef CONFIG_IXGBE_DCB
2440         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2441                 netif_set_gso_max_size(netdev, 32768);
2442                 ixgbe_configure_dcb(adapter);
2443         } else {
2444                 netif_set_gso_max_size(netdev, 65536);
2445         }
2446 #else
2447         netif_set_gso_max_size(netdev, 65536);
2448 #endif
2449
2450 #ifdef IXGBE_FCOE
2451         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2452                 ixgbe_configure_fcoe(adapter);
2453
2454 #endif /* IXGBE_FCOE */
2455         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2456                 for (i = 0; i < adapter->num_tx_queues; i++)
2457                         adapter->tx_ring[i].atr_sample_rate =
2458                                                        adapter->atr_sample_rate;
2459                 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
2460         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
2461                 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
2462         }
2463
2464         ixgbe_configure_tx(adapter);
2465         ixgbe_configure_rx(adapter);
2466         for (i = 0; i < adapter->num_rx_queues; i++)
2467                 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
2468                                        (adapter->rx_ring[i].count - 1));
2469 }
2470
2471 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2472 {
2473         switch (hw->phy.type) {
2474         case ixgbe_phy_sfp_avago:
2475         case ixgbe_phy_sfp_ftl:
2476         case ixgbe_phy_sfp_intel:
2477         case ixgbe_phy_sfp_unknown:
2478         case ixgbe_phy_tw_tyco:
2479         case ixgbe_phy_tw_unknown:
2480                 return true;
2481         default:
2482                 return false;
2483         }
2484 }
2485
2486 /**
2487  * ixgbe_sfp_link_config - set up SFP+ link
2488  * @adapter: pointer to private adapter struct
2489  **/
2490 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2491 {
2492         struct ixgbe_hw *hw = &adapter->hw;
2493
2494                 if (hw->phy.multispeed_fiber) {
2495                         /*
2496                          * In multispeed fiber setups, the device may not have
2497                          * had a physical connection when the driver loaded.
2498                          * If that's the case, the initial link configuration
2499                          * couldn't get the MAC into 10G or 1G mode, so we'll
2500                          * never have a link status change interrupt fire.
2501                          * We need to try and force an autonegotiation
2502                          * session, then bring up link.
2503                          */
2504                         hw->mac.ops.setup_sfp(hw);
2505                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2506                                 schedule_work(&adapter->multispeed_fiber_task);
2507                 } else {
2508                         /*
2509                          * Direct Attach Cu and non-multispeed fiber modules
2510                          * still need to be configured properly prior to
2511                          * attempting link.
2512                          */
2513                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2514                                 schedule_work(&adapter->sfp_config_module_task);
2515                 }
2516 }
2517
2518 /**
2519  * ixgbe_non_sfp_link_config - set up non-SFP+ link
2520  * @hw: pointer to private hardware struct
2521  *
2522  * Returns 0 on success, negative on failure
2523  **/
2524 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2525 {
2526         u32 autoneg;
2527         bool link_up = false;
2528         u32 ret = IXGBE_ERR_LINK_SETUP;
2529
2530         if (hw->mac.ops.check_link)
2531                 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2532
2533         if (ret)
2534                 goto link_cfg_out;
2535
2536         if (hw->mac.ops.get_link_capabilities)
2537                 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
2538                                                         &hw->mac.autoneg);
2539         if (ret)
2540                 goto link_cfg_out;
2541
2542         if (hw->mac.ops.setup_link_speed)
2543                 ret = hw->mac.ops.setup_link_speed(hw, autoneg, true, link_up);
2544 link_cfg_out:
2545         return ret;
2546 }
2547
2548 #define IXGBE_MAX_RX_DESC_POLL 10
2549 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2550                                               int rxr)
2551 {
2552         int j = adapter->rx_ring[rxr].reg_idx;
2553         int k;
2554
2555         for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2556                 if (IXGBE_READ_REG(&adapter->hw,
2557                                    IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2558                         break;
2559                 else
2560                         msleep(1);
2561         }
2562         if (k >= IXGBE_MAX_RX_DESC_POLL) {
2563                 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2564                         "not set within the polling period\n", rxr);
2565         }
2566         ixgbe_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
2567                               (adapter->rx_ring[rxr].count - 1));
2568 }
2569
2570 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2571 {
2572         struct net_device *netdev = adapter->netdev;
2573         struct ixgbe_hw *hw = &adapter->hw;
2574         int i, j = 0;
2575         int num_rx_rings = adapter->num_rx_queues;
2576         int err;
2577         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2578         u32 txdctl, rxdctl, mhadd;
2579         u32 dmatxctl;
2580         u32 gpie;
2581
2582         ixgbe_get_hw_control(adapter);
2583
2584         if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2585             (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2586                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2587                         gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2588                                 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2589                 } else {
2590                         /* MSI only */
2591                         gpie = 0;
2592                 }
2593                 /* XXX: to interrupt immediately for EICS writes, enable this */
2594                 /* gpie |= IXGBE_GPIE_EIMEN; */
2595                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2596         }
2597
2598         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2599                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2600                  * specifically only auto mask tx and rx interrupts */
2601                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2602         }
2603
2604         /* Enable fan failure interrupt if media type is copper */
2605         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2606                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2607                 gpie |= IXGBE_SDP1_GPIEN;
2608                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2609         }
2610
2611         if (hw->mac.type == ixgbe_mac_82599EB) {
2612                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2613                 gpie |= IXGBE_SDP1_GPIEN;
2614                 gpie |= IXGBE_SDP2_GPIEN;
2615                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2616         }
2617
2618 #ifdef IXGBE_FCOE
2619         /* adjust max frame to be able to do baby jumbo for FCoE */
2620         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2621             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2622                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2623
2624 #endif /* IXGBE_FCOE */
2625         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2626         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2627                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2628                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2629
2630                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2631         }
2632
2633         for (i = 0; i < adapter->num_tx_queues; i++) {
2634                 j = adapter->tx_ring[i].reg_idx;
2635                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2636                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2637                 txdctl |= (8 << 16);
2638                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2639         }
2640
2641         if (hw->mac.type == ixgbe_mac_82599EB) {
2642                 /* DMATXCTL.EN must be set after all Tx queue config is done */
2643                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2644                 dmatxctl |= IXGBE_DMATXCTL_TE;
2645                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2646         }
2647         for (i = 0; i < adapter->num_tx_queues; i++) {
2648                 j = adapter->tx_ring[i].reg_idx;
2649                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2650                 txdctl |= IXGBE_TXDCTL_ENABLE;
2651                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2652         }
2653
2654         for (i = 0; i < num_rx_rings; i++) {
2655                 j = adapter->rx_ring[i].reg_idx;
2656                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2657                 /* enable PTHRESH=32 descriptors (half the internal cache)
2658                  * and HTHRESH=0 descriptors (to minimize latency on fetch),
2659                  * this also removes a pesky rx_no_buffer_count increment */
2660                 rxdctl |= 0x0020;
2661                 rxdctl |= IXGBE_RXDCTL_ENABLE;
2662                 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2663                 if (hw->mac.type == ixgbe_mac_82599EB)
2664                         ixgbe_rx_desc_queue_enable(adapter, i);
2665         }
2666         /* enable all receives */
2667         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2668         if (hw->mac.type == ixgbe_mac_82598EB)
2669                 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2670         else
2671                 rxdctl |= IXGBE_RXCTRL_RXEN;
2672         hw->mac.ops.enable_rx_dma(hw, rxdctl);
2673
2674         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2675                 ixgbe_configure_msix(adapter);
2676         else
2677                 ixgbe_configure_msi_and_legacy(adapter);
2678
2679         clear_bit(__IXGBE_DOWN, &adapter->state);
2680         ixgbe_napi_enable_all(adapter);
2681
2682         /* clear any pending interrupts, may auto mask */
2683         IXGBE_READ_REG(hw, IXGBE_EICR);
2684
2685         ixgbe_irq_enable(adapter);
2686
2687         /*
2688          * If this adapter has a fan, check to see if we had a failure
2689          * before we enabled the interrupt.
2690          */
2691         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2692                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
2693                 if (esdp & IXGBE_ESDP_SDP1)
2694                         DPRINTK(DRV, CRIT,
2695                                 "Fan has stopped, replace the adapter\n");
2696         }
2697
2698         /*
2699          * For hot-pluggable SFP+ devices, a new SFP+ module may have
2700          * arrived before interrupts were enabled.  We need to kick off
2701          * the SFP+ module setup first, then try to bring up link.
2702          * If we're not hot-pluggable SFP+, we just need to configure link
2703          * and bring it up.
2704          */
2705         err = hw->phy.ops.identify(hw);
2706         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
2707                 dev_err(&adapter->pdev->dev, "failed to initialize because "
2708                         "an unsupported SFP+ module type was detected.\n"
2709                         "Reload the driver after installing a supported "
2710                         "module.\n");
2711                 ixgbe_down(adapter);
2712                 return err;
2713         }
2714
2715         if (ixgbe_is_sfp(hw)) {
2716                 ixgbe_sfp_link_config(adapter);
2717         } else {
2718                 err = ixgbe_non_sfp_link_config(hw);
2719                 if (err)
2720                         DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
2721         }
2722
2723         for (i = 0; i < adapter->num_tx_queues; i++)
2724                 set_bit(__IXGBE_FDIR_INIT_DONE,
2725                         &(adapter->tx_ring[i].reinit_state));
2726
2727         /* enable transmits */
2728         netif_tx_start_all_queues(netdev);
2729
2730         /* bring the link up in the watchdog, this could race with our first
2731          * link up interrupt but shouldn't be a problem */
2732         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2733         adapter->link_check_timeout = jiffies;
2734         mod_timer(&adapter->watchdog_timer, jiffies);
2735         return 0;
2736 }
2737
2738 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
2739 {
2740         WARN_ON(in_interrupt());
2741         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
2742                 msleep(1);
2743         ixgbe_down(adapter);
2744         ixgbe_up(adapter);
2745         clear_bit(__IXGBE_RESETTING, &adapter->state);
2746 }
2747
2748 int ixgbe_up(struct ixgbe_adapter *adapter)
2749 {
2750         /* hardware has been reset, we need to reload some things */
2751         ixgbe_configure(adapter);
2752
2753         return ixgbe_up_complete(adapter);
2754 }
2755
2756 void ixgbe_reset(struct ixgbe_adapter *adapter)
2757 {
2758         struct ixgbe_hw *hw = &adapter->hw;
2759         int err;
2760
2761         err = hw->mac.ops.init_hw(hw);
2762         switch (err) {
2763         case 0:
2764         case IXGBE_ERR_SFP_NOT_PRESENT:
2765                 break;
2766         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
2767                 dev_err(&adapter->pdev->dev, "master disable timed out\n");
2768                 break;
2769         case IXGBE_ERR_EEPROM_VERSION:
2770                 /* We are running on a pre-production device, log a warning */
2771                 dev_warn(&adapter->pdev->dev, "This device is a pre-production "
2772                          "adapter/LOM.  Please be aware there may be issues "
2773                          "associated with your hardware.  If you are "
2774                          "experiencing problems please contact your Intel or "
2775                          "hardware representative who provided you with this "
2776                          "hardware.\n");
2777                 break;
2778         default:
2779                 dev_err(&adapter->pdev->dev, "Hardware Error: %d\n", err);
2780         }
2781
2782         /* reprogram the RAR[0] in case user changed it. */
2783         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2784 }
2785
2786 /**
2787  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2788  * @adapter: board private structure
2789  * @rx_ring: ring to free buffers from
2790  **/
2791 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2792                                 struct ixgbe_ring *rx_ring)
2793 {
2794         struct pci_dev *pdev = adapter->pdev;
2795         unsigned long size;
2796         unsigned int i;
2797
2798         /* Free all the Rx ring sk_buffs */
2799
2800         for (i = 0; i < rx_ring->count; i++) {
2801                 struct ixgbe_rx_buffer *rx_buffer_info;
2802
2803                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2804                 if (rx_buffer_info->dma) {
2805                         pci_unmap_single(pdev, rx_buffer_info->dma,
2806                                          rx_ring->rx_buf_len,
2807                                          PCI_DMA_FROMDEVICE);
2808                         rx_buffer_info->dma = 0;
2809                 }
2810                 if (rx_buffer_info->skb) {
2811                         struct sk_buff *skb = rx_buffer_info->skb;
2812                         rx_buffer_info->skb = NULL;
2813                         do {
2814                                 struct sk_buff *this = skb;
2815                                 skb = skb->prev;
2816                                 dev_kfree_skb(this);
2817                         } while (skb);
2818                 }
2819                 if (!rx_buffer_info->page)
2820                         continue;
2821                 if (rx_buffer_info->page_dma) {
2822                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
2823                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
2824                         rx_buffer_info->page_dma = 0;
2825                 }
2826                 put_page(rx_buffer_info->page);
2827                 rx_buffer_info->page = NULL;
2828                 rx_buffer_info->page_offset = 0;
2829         }
2830
2831         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2832         memset(rx_ring->rx_buffer_info, 0, size);
2833
2834         /* Zero out the descriptor ring */
2835         memset(rx_ring->desc, 0, rx_ring->size);
2836
2837         rx_ring->next_to_clean = 0;
2838         rx_ring->next_to_use = 0;
2839
2840         if (rx_ring->head)
2841                 writel(0, adapter->hw.hw_addr + rx_ring->head);
2842         if (rx_ring->tail)
2843                 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2844 }
2845
2846 /**
2847  * ixgbe_clean_tx_ring - Free Tx Buffers
2848  * @adapter: board private structure
2849  * @tx_ring: ring to be cleaned
2850  **/
2851 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2852                                 struct ixgbe_ring *tx_ring)
2853 {
2854         struct ixgbe_tx_buffer *tx_buffer_info;
2855         unsigned long size;
2856         unsigned int i;
2857
2858         /* Free all the Tx ring sk_buffs */
2859
2860         for (i = 0; i < tx_ring->count; i++) {
2861                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2862                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2863         }
2864
2865         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2866         memset(tx_ring->tx_buffer_info, 0, size);
2867
2868         /* Zero out the descriptor ring */
2869         memset(tx_ring->desc, 0, tx_ring->size);
2870
2871         tx_ring->next_to_use = 0;
2872         tx_ring->next_to_clean = 0;
2873
2874         if (tx_ring->head)
2875                 writel(0, adapter->hw.hw_addr + tx_ring->head);
2876         if (tx_ring->tail)
2877                 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2878 }
2879
2880 /**
2881  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2882  * @adapter: board private structure
2883  **/
2884 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2885 {
2886         int i;
2887
2888         for (i = 0; i < adapter->num_rx_queues; i++)
2889                 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2890 }
2891
2892 /**
2893  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2894  * @adapter: board private structure
2895  **/
2896 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2897 {
2898         int i;
2899
2900         for (i = 0; i < adapter->num_tx_queues; i++)
2901                 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2902 }
2903
2904 void ixgbe_down(struct ixgbe_adapter *adapter)
2905 {
2906         struct net_device *netdev = adapter->netdev;
2907         struct ixgbe_hw *hw = &adapter->hw;
2908         u32 rxctrl;
2909         u32 txdctl;
2910         int i, j;
2911
2912         /* signal that we are down to the interrupt handler */
2913         set_bit(__IXGBE_DOWN, &adapter->state);
2914
2915         /* disable receives */
2916         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2917         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2918
2919         netif_tx_disable(netdev);
2920
2921         IXGBE_WRITE_FLUSH(hw);
2922         msleep(10);
2923
2924         netif_tx_stop_all_queues(netdev);
2925
2926         ixgbe_irq_disable(adapter);
2927
2928         ixgbe_napi_disable_all(adapter);
2929
2930         del_timer_sync(&adapter->watchdog_timer);
2931         cancel_work_sync(&adapter->watchdog_task);
2932
2933         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2934             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2935                 cancel_work_sync(&adapter->fdir_reinit_task);
2936
2937         /* disable transmits in the hardware now that interrupts are off */
2938         for (i = 0; i < adapter->num_tx_queues; i++) {
2939                 j = adapter->tx_ring[i].reg_idx;
2940                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2941                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
2942                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
2943         }
2944         /* Disable the Tx DMA engine on 82599 */
2945         if (hw->mac.type == ixgbe_mac_82599EB)
2946                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
2947                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
2948                                  ~IXGBE_DMATXCTL_TE));
2949
2950         netif_carrier_off(netdev);
2951
2952         if (!pci_channel_offline(adapter->pdev))
2953                 ixgbe_reset(adapter);
2954         ixgbe_clean_all_tx_rings(adapter);
2955         ixgbe_clean_all_rx_rings(adapter);
2956
2957 #ifdef CONFIG_IXGBE_DCA
2958         /* since we reset the hardware DCA settings were cleared */
2959         ixgbe_setup_dca(adapter);
2960 #endif
2961 }
2962
2963 /**
2964  * ixgbe_poll - NAPI Rx polling callback
2965  * @napi: structure for representing this polling device
2966  * @budget: how many packets driver is allowed to clean
2967  *
2968  * This function is used for legacy and MSI, NAPI mode
2969  **/
2970 static int ixgbe_poll(struct napi_struct *napi, int budget)
2971 {
2972         struct ixgbe_q_vector *q_vector =
2973                                 container_of(napi, struct ixgbe_q_vector, napi);
2974         struct ixgbe_adapter *adapter = q_vector->adapter;
2975         int tx_clean_complete, work_done = 0;
2976
2977 #ifdef CONFIG_IXGBE_DCA
2978         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2979                 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
2980                 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
2981         }
2982 #endif
2983
2984         tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring);
2985         ixgbe_clean_rx_irq(q_vector, adapter->rx_ring, &work_done, budget);
2986
2987         if (!tx_clean_complete)
2988                 work_done = budget;
2989
2990         /* If budget not fully consumed, exit the polling mode */
2991         if (work_done < budget) {
2992                 napi_complete(napi);
2993                 if (adapter->itr_setting & 1)
2994                         ixgbe_set_itr(adapter);
2995                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2996                         ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
2997         }
2998         return work_done;
2999 }
3000
3001 /**
3002  * ixgbe_tx_timeout - Respond to a Tx Hang
3003  * @netdev: network interface device structure
3004  **/
3005 static void ixgbe_tx_timeout(struct net_device *netdev)
3006 {
3007         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3008
3009         /* Do the reset outside of interrupt context */
3010         schedule_work(&adapter->reset_task);
3011 }
3012
3013 static void ixgbe_reset_task(struct work_struct *work)
3014 {
3015         struct ixgbe_adapter *adapter;
3016         adapter = container_of(work, struct ixgbe_adapter, reset_task);
3017
3018         /* If we're already down or resetting, just bail */
3019         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3020             test_bit(__IXGBE_RESETTING, &adapter->state))
3021                 return;
3022
3023         adapter->tx_timeout_count++;
3024
3025         ixgbe_reinit_locked(adapter);
3026 }
3027
3028 #ifdef CONFIG_IXGBE_DCB
3029 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3030 {
3031         bool ret = false;
3032         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3033
3034         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3035                 return ret;
3036
3037         f->mask = 0x7 << 3;
3038         adapter->num_rx_queues = f->indices;
3039         adapter->num_tx_queues = f->indices;
3040         ret = true;
3041
3042         return ret;
3043 }
3044 #endif
3045
3046 /**
3047  * ixgbe_set_rss_queues: Allocate queues for RSS
3048  * @adapter: board private structure to initialize
3049  *
3050  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
3051  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3052  *
3053  **/
3054 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3055 {
3056         bool ret = false;
3057         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3058
3059         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3060                 f->mask = 0xF;
3061                 adapter->num_rx_queues = f->indices;
3062                 adapter->num_tx_queues = f->indices;
3063                 ret = true;
3064         } else {
3065                 ret = false;
3066         }
3067
3068         return ret;
3069 }
3070
3071 /**
3072  * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3073  * @adapter: board private structure to initialize
3074  *
3075  * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3076  * to the original CPU that initiated the Tx session.  This runs in addition
3077  * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3078  * Rx load across CPUs using RSS.
3079  *
3080  **/
3081 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3082 {
3083         bool ret = false;
3084         struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3085
3086         f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3087         f_fdir->mask = 0;
3088
3089         /* Flow Director must have RSS enabled */
3090         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3091             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3092              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3093                 adapter->num_tx_queues = f_fdir->indices;
3094                 adapter->num_rx_queues = f_fdir->indices;
3095                 ret = true;
3096         } else {
3097                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3098                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3099         }
3100         return ret;
3101 }
3102
3103 #ifdef IXGBE_FCOE
3104 /**
3105  * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3106  * @adapter: board private structure to initialize
3107  *
3108  * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3109  * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3110  * rx queues out of the max number of rx queues, instead, it is used as the
3111  * index of the first rx queue used by FCoE.
3112  *
3113  **/
3114 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3115 {
3116         bool ret = false;
3117         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3118
3119         f->indices = min((int)num_online_cpus(), f->indices);
3120         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3121 #ifdef CONFIG_IXGBE_DCB
3122                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3123                         DPRINTK(PROBE, INFO, "FCOE enabled with DCB \n");
3124                         ixgbe_set_dcb_queues(adapter);
3125                 }
3126 #endif
3127                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3128                         DPRINTK(PROBE, INFO, "FCOE enabled with RSS \n");
3129                         ixgbe_set_rss_queues(adapter);
3130                 }
3131                 /* adding FCoE rx rings to the end */
3132                 f->mask = adapter->num_rx_queues;
3133                 adapter->num_rx_queues += f->indices;
3134                 if (adapter->num_tx_queues == 0)
3135                         adapter->num_tx_queues = f->indices;
3136
3137                 ret = true;
3138         }
3139
3140         return ret;
3141 }
3142
3143 #endif /* IXGBE_FCOE */
3144 /*
3145  * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3146  * @adapter: board private structure to initialize
3147  *
3148  * This is the top level queue allocation routine.  The order here is very
3149  * important, starting with the "most" number of features turned on at once,
3150  * and ending with the smallest set of features.  This way large combinations
3151  * can be allocated if they're turned on, and smaller combinations are the
3152  * fallthrough conditions.
3153  *
3154  **/
3155 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
3156 {
3157 #ifdef IXGBE_FCOE
3158         if (ixgbe_set_fcoe_queues(adapter))
3159                 goto done;
3160
3161 #endif /* IXGBE_FCOE */
3162 #ifdef CONFIG_IXGBE_DCB
3163         if (ixgbe_set_dcb_queues(adapter))
3164                 goto done;
3165
3166 #endif
3167         if (ixgbe_set_fdir_queues(adapter))
3168                 goto done;
3169
3170         if (ixgbe_set_rss_queues(adapter))
3171                 goto done;
3172
3173         /* fallback to base case */
3174         adapter->num_rx_queues = 1;
3175         adapter->num_tx_queues = 1;
3176
3177 done:
3178         /* Notify the stack of the (possibly) reduced Tx Queue count. */
3179         adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
3180 }
3181
3182 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
3183                                        int vectors)
3184 {
3185         int err, vector_threshold;
3186
3187         /* We'll want at least 3 (vector_threshold):
3188          * 1) TxQ[0] Cleanup
3189          * 2) RxQ[0] Cleanup
3190          * 3) Other (Link Status Change, etc.)
3191          * 4) TCP Timer (optional)
3192          */
3193         vector_threshold = MIN_MSIX_COUNT;
3194
3195         /* The more we get, the more we will assign to Tx/Rx Cleanup
3196          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
3197          * Right now, we simply care about how many we'll get; we'll
3198          * set them up later while requesting irq's.
3199          */
3200         while (vectors >= vector_threshold) {
3201                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
3202                                       vectors);
3203                 if (!err) /* Success in acquiring all requested vectors. */
3204                         break;
3205                 else if (err < 0)
3206                         vectors = 0; /* Nasty failure, quit now */
3207                 else /* err == number of vectors we should try again with */
3208                         vectors = err;
3209         }
3210
3211         if (vectors < vector_threshold) {
3212                 /* Can't allocate enough MSI-X interrupts?  Oh well.
3213                  * This just means we'll go with either a single MSI
3214                  * vector or fall back to legacy interrupts.
3215                  */
3216                 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
3217                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3218                 kfree(adapter->msix_entries);
3219                 adapter->msix_entries = NULL;
3220         } else {
3221                 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
3222                 /*
3223                  * Adjust for only the vectors we'll use, which is minimum
3224                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
3225                  * vectors we were allocated.
3226                  */
3227                 adapter->num_msix_vectors = min(vectors,
3228                                    adapter->max_msix_q_vectors + NON_Q_VECTORS);
3229         }
3230 }
3231
3232 /**
3233  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
3234  * @adapter: board private structure to initialize
3235  *
3236  * Cache the descriptor ring offsets for RSS to the assigned rings.
3237  *
3238  **/
3239 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
3240 {
3241         int i;
3242         bool ret = false;
3243
3244         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3245                 for (i = 0; i < adapter->num_rx_queues; i++)
3246                         adapter->rx_ring[i].reg_idx = i;
3247                 for (i = 0; i < adapter->num_tx_queues; i++)
3248                         adapter->tx_ring[i].reg_idx = i;
3249                 ret = true;
3250         } else {
3251                 ret = false;
3252         }
3253
3254         return ret;
3255 }
3256
3257 #ifdef CONFIG_IXGBE_DCB
3258 /**
3259  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
3260  * @adapter: board private structure to initialize
3261  *
3262  * Cache the descriptor ring offsets for DCB to the assigned rings.
3263  *
3264  **/
3265 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
3266 {
3267         int i;
3268         bool ret = false;
3269         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
3270
3271         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3272                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3273                         /* the number of queues is assumed to be symmetric */
3274                         for (i = 0; i < dcb_i; i++) {
3275                                 adapter->rx_ring[i].reg_idx = i << 3;
3276                                 adapter->tx_ring[i].reg_idx = i << 2;
3277                         }
3278                         ret = true;
3279                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
3280                         if (dcb_i == 8) {
3281                                 /*
3282                                  * Tx TC0 starts at: descriptor queue 0
3283                                  * Tx TC1 starts at: descriptor queue 32
3284                                  * Tx TC2 starts at: descriptor queue 64
3285                                  * Tx TC3 starts at: descriptor queue 80
3286                                  * Tx TC4 starts at: descriptor queue 96
3287                                  * Tx TC5 starts at: descriptor queue 104
3288                                  * Tx TC6 starts at: descriptor queue 112
3289                                  * Tx TC7 starts at: descriptor queue 120
3290                                  *
3291                                  * Rx TC0-TC7 are offset by 16 queues each
3292                                  */
3293                                 for (i = 0; i < 3; i++) {
3294                                         adapter->tx_ring[i].reg_idx = i << 5;
3295                                         adapter->rx_ring[i].reg_idx = i << 4;
3296                                 }
3297                                 for ( ; i < 5; i++) {
3298                                         adapter->tx_ring[i].reg_idx =
3299                                                                  ((i + 2) << 4);
3300                                         adapter->rx_ring[i].reg_idx = i << 4;
3301                                 }
3302                                 for ( ; i < dcb_i; i++) {
3303                                         adapter->tx_ring[i].reg_idx =
3304                                                                  ((i + 8) << 3);
3305                                         adapter->rx_ring[i].reg_idx = i << 4;
3306                                 }
3307
3308                                 ret = true;
3309                         } else if (dcb_i == 4) {
3310                                 /*
3311                                  * Tx TC0 starts at: descriptor queue 0
3312                                  * Tx TC1 starts at: descriptor queue 64
3313                                  * Tx TC2 starts at: descriptor queue 96
3314                                  * Tx TC3 starts at: descriptor queue 112
3315                                  *
3316                                  * Rx TC0-TC3 are offset by 32 queues each
3317                                  */
3318                                 adapter->tx_ring[0].reg_idx = 0;
3319                                 adapter->tx_ring[1].reg_idx = 64;
3320                                 adapter->tx_ring[2].reg_idx = 96;
3321                                 adapter->tx_ring[3].reg_idx = 112;
3322                                 for (i = 0 ; i < dcb_i; i++)
3323                                         adapter->rx_ring[i].reg_idx = i << 5;
3324
3325                                 ret = true;
3326                         } else {
3327                                 ret = false;
3328                         }
3329                 } else {
3330                         ret = false;
3331                 }
3332         } else {
3333                 ret = false;
3334         }
3335
3336         return ret;
3337 }
3338 #endif
3339
3340 /**
3341  * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
3342  * @adapter: board private structure to initialize
3343  *
3344  * Cache the descriptor ring offsets for Flow Director to the assigned rings.
3345  *
3346  **/
3347 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
3348 {
3349         int i;
3350         bool ret = false;
3351
3352         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3353             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3354              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
3355                 for (i = 0; i < adapter->num_rx_queues; i++)
3356                         adapter->rx_ring[i].reg_idx = i;
3357                 for (i = 0; i < adapter->num_tx_queues; i++)
3358                         adapter->tx_ring[i].reg_idx = i;
3359                 ret = true;
3360         }
3361
3362         return ret;
3363 }
3364
3365 #ifdef IXGBE_FCOE
3366 /**
3367  * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
3368  * @adapter: board private structure to initialize
3369  *
3370  * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
3371  *
3372  */
3373 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
3374 {
3375         int i, fcoe_i = 0;
3376         bool ret = false;
3377         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3378
3379         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3380 #ifdef CONFIG_IXGBE_DCB
3381                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3382                         ixgbe_cache_ring_dcb(adapter);
3383                         fcoe_i = adapter->rx_ring[0].reg_idx + 1;
3384                 }
3385 #endif /* CONFIG_IXGBE_DCB */
3386                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3387                         ixgbe_cache_ring_rss(adapter);
3388                         fcoe_i = f->mask;
3389                 }
3390                 for (i = 0; i < f->indices; i++, fcoe_i++)
3391                         adapter->rx_ring[f->mask + i].reg_idx = fcoe_i;
3392                 ret = true;
3393         }
3394         return ret;
3395 }
3396
3397 #endif /* IXGBE_FCOE */
3398 /**
3399  * ixgbe_cache_ring_register - Descriptor ring to register mapping
3400  * @adapter: board private structure to initialize
3401  *
3402  * Once we know the feature-set enabled for the device, we'll cache
3403  * the register offset the descriptor ring is assigned to.
3404  *
3405  * Note, the order the various feature calls is important.  It must start with
3406  * the "most" features enabled at the same time, then trickle down to the
3407  * least amount of features turned on at once.
3408  **/
3409 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
3410 {
3411         /* start with default case */
3412         adapter->rx_ring[0].reg_idx = 0;
3413         adapter->tx_ring[0].reg_idx = 0;
3414
3415 #ifdef IXGBE_FCOE
3416         if (ixgbe_cache_ring_fcoe(adapter))
3417                 return;
3418
3419 #endif /* IXGBE_FCOE */
3420 #ifdef CONFIG_IXGBE_DCB
3421         if (ixgbe_cache_ring_dcb(adapter))
3422                 return;
3423
3424 #endif
3425         if (ixgbe_cache_ring_fdir(adapter))
3426                 return;
3427
3428         if (ixgbe_cache_ring_rss(adapter))
3429                 return;
3430 }
3431
3432 /**
3433  * ixgbe_alloc_queues - Allocate memory for all rings
3434  * @adapter: board private structure to initialize
3435  *
3436  * We allocate one ring per queue at run-time since we don't know the
3437  * number of queues at compile-time.  The polling_netdev array is
3438  * intended for Multiqueue, but should work fine with a single queue.
3439  **/
3440 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
3441 {
3442         int i;
3443
3444         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
3445                                    sizeof(struct ixgbe_ring), GFP_KERNEL);
3446         if (!adapter->tx_ring)
3447                 goto err_tx_ring_allocation;
3448
3449         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
3450                                    sizeof(struct ixgbe_ring), GFP_KERNEL);
3451         if (!adapter->rx_ring)
3452                 goto err_rx_ring_allocation;
3453
3454         for (i = 0; i < adapter->num_tx_queues; i++) {
3455                 adapter->tx_ring[i].count = adapter->tx_ring_count;
3456                 adapter->tx_ring[i].queue_index = i;
3457         }
3458
3459         for (i = 0; i < adapter->num_rx_queues; i++) {
3460                 adapter->rx_ring[i].count = adapter->rx_ring_count;
3461                 adapter->rx_ring[i].queue_index = i;
3462         }
3463
3464         ixgbe_cache_ring_register(adapter);
3465
3466         return 0;
3467
3468 err_rx_ring_allocation:
3469         kfree(adapter->tx_ring);
3470 err_tx_ring_allocation:
3471         return -ENOMEM;
3472 }
3473
3474 /**
3475  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
3476  * @adapter: board private structure to initialize
3477  *
3478  * Attempt to configure the interrupts using the best available
3479  * capabilities of the hardware and the kernel.
3480  **/
3481 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
3482 {
3483         struct ixgbe_hw *hw = &adapter->hw;
3484         int err = 0;
3485         int vector, v_budget;
3486
3487         /*
3488          * It's easy to be greedy for MSI-X vectors, but it really
3489          * doesn't do us much good if we have a lot more vectors
3490          * than CPU's.  So let's be conservative and only ask for
3491          * (roughly) twice the number of vectors as there are CPU's.
3492          */
3493         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
3494                        (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
3495
3496         /*
3497          * At the same time, hardware can only support a maximum of
3498          * hw.mac->max_msix_vectors vectors.  With features
3499          * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
3500          * descriptor queues supported by our device.  Thus, we cap it off in
3501          * those rare cases where the cpu count also exceeds our vector limit.
3502          */
3503         v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
3504
3505         /* A failure in MSI-X entry allocation isn't fatal, but it does
3506          * mean we disable MSI-X capabilities of the adapter. */
3507         adapter->msix_entries = kcalloc(v_budget,
3508                                         sizeof(struct msix_entry), GFP_KERNEL);
3509         if (adapter->msix_entries) {
3510                 for (vector = 0; vector < v_budget; vector++)
3511                         adapter->msix_entries[vector].entry = vector;
3512
3513                 ixgbe_acquire_msix_vectors(adapter, v_budget);
3514
3515                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3516                         goto out;
3517         }
3518
3519         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
3520         adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
3521         adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3522         adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3523         adapter->atr_sample_rate = 0;
3524         ixgbe_set_num_queues(adapter);
3525
3526         err = pci_enable_msi(adapter->pdev);
3527         if (!err) {
3528                 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
3529         } else {
3530                 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
3531                         "falling back to legacy.  Error: %d\n", err);
3532                 /* reset err */
3533                 err = 0;
3534         }
3535
3536 out:
3537         return err;
3538 }
3539
3540 /**
3541  * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
3542  * @adapter: board private structure to initialize
3543  *
3544  * We allocate one q_vector per queue interrupt.  If allocation fails we
3545  * return -ENOMEM.
3546  **/
3547 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
3548 {
3549         int q_idx, num_q_vectors;
3550         struct ixgbe_q_vector *q_vector;
3551         int napi_vectors;
3552         int (*poll)(struct napi_struct *, int);
3553
3554         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3555                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3556                 napi_vectors = adapter->num_rx_queues;
3557                 poll = &ixgbe_clean_rxtx_many;
3558         } else {
3559                 num_q_vectors = 1;
3560                 napi_vectors = 1;
3561                 poll = &ixgbe_poll;
3562         }
3563
3564         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3565                 q_vector = kzalloc(sizeof(struct ixgbe_q_vector), GFP_KERNEL);
3566                 if (!q_vector)
3567                         goto err_out;
3568                 q_vector->adapter = adapter;
3569                 q_vector->eitr = adapter->eitr_param;
3570                 q_vector->v_idx = q_idx;
3571                 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
3572                 adapter->q_vector[q_idx] = q_vector;
3573         }
3574
3575         return 0;
3576
3577 err_out:
3578         while (q_idx) {
3579                 q_idx--;
3580                 q_vector = adapter->q_vector[q_idx];
3581                 netif_napi_del(&q_vector->napi);
3582                 kfree(q_vector);
3583                 adapter->q_vector[q_idx] = NULL;
3584         }
3585         return -ENOMEM;
3586 }
3587
3588 /**
3589  * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
3590  * @adapter: board private structure to initialize
3591  *
3592  * This function frees the memory allocated to the q_vectors.  In addition if
3593  * NAPI is enabled it will delete any references to the NAPI struct prior
3594  * to freeing the q_vector.
3595  **/
3596 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
3597 {
3598         int q_idx, num_q_vectors;
3599
3600         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3601                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3602         else
3603                 num_q_vectors = 1;
3604
3605         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3606                 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
3607                 adapter->q_vector[q_idx] = NULL;
3608                 netif_napi_del(&q_vector->napi);
3609                 kfree(q_vector);
3610         }
3611 }
3612
3613 void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
3614 {
3615         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3616                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3617                 pci_disable_msix(adapter->pdev);
3618                 kfree(adapter->msix_entries);
3619                 adapter->msix_entries = NULL;
3620         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
3621                 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
3622                 pci_disable_msi(adapter->pdev);
3623         }
3624         return;
3625 }
3626
3627 /**
3628  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
3629  * @adapter: board private structure to initialize
3630  *
3631  * We determine which interrupt scheme to use based on...
3632  * - Kernel support (MSI, MSI-X)
3633  *   - which can be user-defined (via MODULE_PARAM)
3634  * - Hardware queue count (num_*_queues)
3635  *   - defined by miscellaneous hardware support/features (RSS, etc.)
3636  **/
3637 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
3638 {
3639         int err;
3640
3641         /* Number of supported queues */
3642         ixgbe_set_num_queues(adapter);
3643
3644         err = ixgbe_set_interrupt_capability(adapter);
3645         if (err) {
3646                 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
3647                 goto err_set_interrupt;
3648         }
3649
3650         err = ixgbe_alloc_q_vectors(adapter);
3651         if (err) {
3652                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queue "
3653                         "vectors\n");
3654                 goto err_alloc_q_vectors;
3655         }
3656
3657         err = ixgbe_alloc_queues(adapter);
3658         if (err) {
3659                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
3660                 goto err_alloc_queues;
3661         }
3662
3663         DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
3664                 "Tx Queue count = %u\n",
3665                 (adapter->num_rx_queues > 1) ? "Enabled" :
3666                 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
3667
3668         set_bit(__IXGBE_DOWN, &adapter->state);
3669
3670         return 0;
3671
3672 err_alloc_queues:
3673         ixgbe_free_q_vectors(adapter);
3674 err_alloc_q_vectors:
3675         ixgbe_reset_interrupt_capability(adapter);
3676 err_set_interrupt:
3677         return err;
3678 }
3679
3680 /**
3681  * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
3682  * @adapter: board private structure to clear interrupt scheme on
3683  *
3684  * We go through and clear interrupt specific resources and reset the structure
3685  * to pre-load conditions
3686  **/
3687 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
3688 {
3689         kfree(adapter->tx_ring);
3690         kfree(adapter->rx_ring);
3691         adapter->tx_ring = NULL;
3692         adapter->rx_ring = NULL;
3693
3694         ixgbe_free_q_vectors(adapter);
3695         ixgbe_reset_interrupt_capability(adapter);
3696 }
3697
3698 /**
3699  * ixgbe_sfp_timer - worker thread to find a missing module
3700  * @data: pointer to our adapter struct
3701  **/
3702 static void ixgbe_sfp_timer(unsigned long data)
3703 {
3704         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3705
3706         /*
3707          * Do the sfp_timer outside of interrupt context due to the
3708          * delays that sfp+ detection requires
3709          */
3710         schedule_work(&adapter->sfp_task);
3711 }
3712
3713 /**
3714  * ixgbe_sfp_task - worker thread to find a missing module
3715  * @work: pointer to work_struct containing our data
3716  **/
3717 static void ixgbe_sfp_task(struct work_struct *work)
3718 {
3719         struct ixgbe_adapter *adapter = container_of(work,
3720                                                      struct ixgbe_adapter,
3721                                                      sfp_task);
3722         struct ixgbe_hw *hw = &adapter->hw;
3723
3724         if ((hw->phy.type == ixgbe_phy_nl) &&
3725             (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
3726                 s32 ret = hw->phy.ops.identify_sfp(hw);
3727                 if (ret)
3728                         goto reschedule;
3729                 ret = hw->phy.ops.reset(hw);
3730                 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3731                         dev_err(&adapter->pdev->dev, "failed to initialize "
3732                                 "because an unsupported SFP+ module type "
3733                                 "was detected.\n"
3734                                 "Reload the driver after installing a "
3735                                 "supported module.\n");
3736                         unregister_netdev(adapter->netdev);
3737                 } else {
3738                         DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
3739                                 hw->phy.sfp_type);
3740                 }
3741                 /* don't need this routine any more */
3742                 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3743         }
3744         return;
3745 reschedule:
3746         if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
3747                 mod_timer(&adapter->sfp_timer,
3748                           round_jiffies(jiffies + (2 * HZ)));
3749 }
3750
3751 /**
3752  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
3753  * @adapter: board private structure to initialize
3754  *
3755  * ixgbe_sw_init initializes the Adapter private data structure.
3756  * Fields are initialized based on PCI device information and
3757  * OS network device settings (MTU size).
3758  **/
3759 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
3760 {
3761         struct ixgbe_hw *hw = &adapter->hw;
3762         struct pci_dev *pdev = adapter->pdev;
3763         unsigned int rss;
3764 #ifdef CONFIG_IXGBE_DCB
3765         int j;
3766         struct tc_configuration *tc;
3767 #endif
3768
3769         /* PCI config space info */
3770
3771         hw->vendor_id = pdev->vendor;
3772         hw->device_id = pdev->device;
3773         hw->revision_id = pdev->revision;
3774         hw->subsystem_vendor_id = pdev->subsystem_vendor;
3775         hw->subsystem_device_id = pdev->subsystem_device;
3776
3777         /* Set capability flags */
3778         rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
3779         adapter->ring_feature[RING_F_RSS].indices = rss;
3780         adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
3781         adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
3782         if (hw->mac.type == ixgbe_mac_82598EB) {
3783                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
3784                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
3785                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
3786         } else if (hw->mac.type == ixgbe_mac_82599EB) {
3787                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
3788                 adapter->flags |= IXGBE_FLAG2_RSC_CAPABLE;
3789                 adapter->flags |= IXGBE_FLAG2_RSC_ENABLED;
3790                 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
3791                 adapter->ring_feature[RING_F_FDIR].indices =
3792                                                          IXGBE_MAX_FDIR_INDICES;
3793                 adapter->atr_sample_rate = 20;
3794                 adapter->fdir_pballoc = 0;
3795 #ifdef IXGBE_FCOE
3796                 adapter->flags |= IXGBE_FLAG_FCOE_ENABLED;
3797                 adapter->ring_feature[RING_F_FCOE].indices = IXGBE_FCRETA_SIZE;
3798 #endif /* IXGBE_FCOE */
3799         }
3800
3801 #ifdef CONFIG_IXGBE_DCB
3802         /* Configure DCB traffic classes */
3803         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
3804                 tc = &adapter->dcb_cfg.tc_config[j];
3805                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
3806                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
3807                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
3808                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
3809                 tc->dcb_pfc = pfc_disabled;
3810         }
3811         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
3812         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
3813         adapter->dcb_cfg.rx_pba_cfg = pba_equal;
3814         adapter->dcb_cfg.pfc_mode_enable = false;
3815         adapter->dcb_cfg.round_robin_enable = false;
3816         adapter->dcb_set_bitmap = 0x00;
3817         ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
3818                            adapter->ring_feature[RING_F_DCB].indices);
3819
3820 #endif
3821
3822         /* default flow control settings */
3823         hw->fc.requested_mode = ixgbe_fc_full;
3824         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
3825 #ifdef CONFIG_DCB
3826         adapter->last_lfc_mode = hw->fc.current_mode;
3827 #endif
3828         hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
3829         hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
3830         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
3831         hw->fc.send_xon = true;
3832         hw->fc.disable_fc_autoneg = false;
3833
3834         /* enable itr by default in dynamic mode */
3835         adapter->itr_setting = 1;
3836         adapter->eitr_param = 20000;
3837
3838         /* set defaults for eitr in MegaBytes */
3839         adapter->eitr_low = 10;
3840         adapter->eitr_high = 20;
3841
3842         /* set default ring sizes */
3843         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
3844         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
3845
3846         /* initialize eeprom parameters */
3847         if (ixgbe_init_eeprom_params_generic(hw)) {
3848                 dev_err(&pdev->dev, "EEPROM initialization failed\n");
3849                 return -EIO;
3850         }
3851
3852         /* enable rx csum by default */
3853         adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
3854
3855         set_bit(__IXGBE_DOWN, &adapter->state);
3856
3857         return 0;
3858 }
3859
3860 /**
3861  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
3862  * @adapter: board private structure
3863  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
3864  *
3865  * Return 0 on success, negative on failure
3866  **/
3867 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
3868                              struct ixgbe_ring *tx_ring)
3869 {
3870         struct pci_dev *pdev = adapter->pdev;
3871         int size;
3872
3873         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3874         tx_ring->tx_buffer_info = vmalloc(size);
3875         if (!tx_ring->tx_buffer_info)
3876                 goto err;
3877         memset(tx_ring->tx_buffer_info, 0, size);
3878
3879         /* round up to nearest 4K */
3880         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3881         tx_ring->size = ALIGN(tx_ring->size, 4096);
3882
3883         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
3884                                              &tx_ring->dma);
3885         if (!tx_ring->desc)
3886                 goto err;
3887
3888         tx_ring->next_to_use = 0;
3889         tx_ring->next_to_clean = 0;
3890         tx_ring->work_limit = tx_ring->count;
3891         return 0;
3892
3893 err:
3894         vfree(tx_ring->tx_buffer_info);
3895         tx_ring->tx_buffer_info = NULL;
3896         DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
3897                             "descriptor ring\n");
3898         return -ENOMEM;
3899 }
3900
3901 /**
3902  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
3903  * @adapter: board private structure
3904  *
3905  * If this function returns with an error, then it's possible one or
3906  * more of the rings is populated (while the rest are not).  It is the
3907  * callers duty to clean those orphaned rings.
3908  *
3909  * Return 0 on success, negative on failure
3910  **/
3911 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
3912 {
3913         int i, err = 0;
3914
3915         for (i = 0; i < adapter->num_tx_queues; i++) {
3916                 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
3917                 if (!err)
3918                         continue;
3919                 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
3920                 break;
3921         }
3922
3923         return err;
3924 }
3925
3926 /**
3927  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
3928  * @adapter: board private structure
3929  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
3930  *
3931  * Returns 0 on success, negative on failure
3932  **/
3933 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
3934                              struct ixgbe_ring *rx_ring)
3935 {
3936         struct pci_dev *pdev = adapter->pdev;
3937         int size;
3938
3939         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3940         rx_ring->rx_buffer_info = vmalloc(size);
3941         if (!rx_ring->rx_buffer_info) {
3942                 DPRINTK(PROBE, ERR,
3943                         "vmalloc allocation failed for the rx desc ring\n");
3944                 goto alloc_failed;
3945         }
3946         memset(rx_ring->rx_buffer_info, 0, size);
3947
3948         /* Round up to nearest 4K */
3949         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3950         rx_ring->size = ALIGN(rx_ring->size, 4096);
3951
3952         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
3953
3954         if (!rx_ring->desc) {
3955                 DPRINTK(PROBE, ERR,
3956                         "Memory allocation failed for the rx desc ring\n");
3957                 vfree(rx_ring->rx_buffer_info);
3958                 goto alloc_failed;
3959         }
3960
3961         rx_ring->next_to_clean = 0;
3962         rx_ring->next_to_use = 0;
3963
3964         return 0;
3965
3966 alloc_failed:
3967         return -ENOMEM;
3968 }
3969
3970 /**
3971  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
3972  * @adapter: board private structure
3973  *
3974  * If this function returns with an error, then it's possible one or
3975  * more of the rings is populated (while the rest are not).  It is the
3976  * callers duty to clean those orphaned rings.
3977  *
3978  * Return 0 on success, negative on failure
3979  **/
3980
3981 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
3982 {
3983         int i, err = 0;
3984
3985         for (i = 0; i < adapter->num_rx_queues; i++) {
3986                 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
3987                 if (!err)
3988                         continue;
3989                 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
3990                 break;
3991         }
3992
3993         return err;
3994 }
3995
3996 /**
3997  * ixgbe_free_tx_resources - Free Tx Resources per Queue
3998  * @adapter: board private structure
3999  * @tx_ring: Tx descriptor ring for a specific queue
4000  *
4001  * Free all transmit software resources
4002  **/
4003 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4004                              struct ixgbe_ring *tx_ring)
4005 {
4006         struct pci_dev *pdev = adapter->pdev;
4007
4008         ixgbe_clean_tx_ring(adapter, tx_ring);
4009
4010         vfree(tx_ring->tx_buffer_info);
4011         tx_ring->tx_buffer_info = NULL;
4012
4013         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
4014
4015         tx_ring->desc = NULL;
4016 }
4017
4018 /**
4019  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4020  * @adapter: board private structure
4021  *
4022  * Free all transmit software resources
4023  **/
4024 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4025 {
4026         int i;
4027
4028         for (i = 0; i < adapter->num_tx_queues; i++)
4029                 if (adapter->tx_ring[i].desc)
4030                         ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
4031 }
4032
4033 /**
4034  * ixgbe_free_rx_resources - Free Rx Resources
4035  * @adapter: board private structure
4036  * @rx_ring: ring to clean the resources from
4037  *
4038  * Free all receive software resources
4039  **/
4040 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
4041                              struct ixgbe_ring *rx_ring)
4042 {
4043         struct pci_dev *pdev = adapter->pdev;
4044
4045         ixgbe_clean_rx_ring(adapter, rx_ring);
4046
4047         vfree(rx_ring->rx_buffer_info);
4048         rx_ring->rx_buffer_info = NULL;
4049
4050         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
4051
4052         rx_ring->desc = NULL;
4053 }
4054
4055 /**
4056  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4057  * @adapter: board private structure
4058  *
4059  * Free all receive software resources
4060  **/
4061 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4062 {
4063         int i;
4064
4065         for (i = 0; i < adapter->num_rx_queues; i++)
4066                 if (adapter->rx_ring[i].desc)
4067                         ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
4068 }
4069
4070 /**
4071  * ixgbe_change_mtu - Change the Maximum Transfer Unit
4072  * @netdev: network interface device structure
4073  * @new_mtu: new value for maximum frame size
4074  *
4075  * Returns 0 on success, negative on failure
4076  **/
4077 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4078 {
4079         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4080         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4081
4082         /* MTU < 68 is an error and causes problems on some kernels */
4083         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4084                 return -EINVAL;
4085
4086         DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
4087                 netdev->mtu, new_mtu);
4088         /* must set new MTU before calling down or up */
4089         netdev->mtu = new_mtu;
4090
4091         if (netif_running(netdev))
4092                 ixgbe_reinit_locked(adapter);
4093
4094         return 0;
4095 }
4096
4097 /**
4098  * ixgbe_open - Called when a network interface is made active
4099  * @netdev: network interface device structure
4100  *
4101  * Returns 0 on success, negative value on failure
4102  *
4103  * The open entry point is called when a network interface is made
4104  * active by the system (IFF_UP).  At this point all resources needed
4105  * for transmit and receive operations are allocated, the interrupt
4106  * handler is registered with the OS, the watchdog timer is started,
4107  * and the stack is notified that the interface is ready.
4108  **/
4109 static int ixgbe_open(struct net_device *netdev)
4110 {
4111         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4112         int err;
4113
4114         /* disallow open during test */
4115         if (test_bit(__IXGBE_TESTING, &adapter->state))
4116                 return -EBUSY;
4117
4118         netif_carrier_off(netdev);
4119
4120         /* allocate transmit descriptors */
4121         err = ixgbe_setup_all_tx_resources(adapter);
4122         if (err)
4123                 goto err_setup_tx;
4124
4125         /* allocate receive descriptors */
4126         err = ixgbe_setup_all_rx_resources(adapter);
4127         if (err)
4128                 goto err_setup_rx;
4129
4130         ixgbe_configure(adapter);
4131
4132         err = ixgbe_request_irq(adapter);
4133         if (err)
4134                 goto err_req_irq;
4135
4136         err = ixgbe_up_complete(adapter);
4137         if (err)
4138                 goto err_up;
4139
4140         netif_tx_start_all_queues(netdev);
4141
4142         return 0;
4143
4144 err_up:
4145         ixgbe_release_hw_control(adapter);
4146         ixgbe_free_irq(adapter);
4147 err_req_irq:
4148 err_setup_rx:
4149         ixgbe_free_all_rx_resources(adapter);
4150 err_setup_tx:
4151         ixgbe_free_all_tx_resources(adapter);
4152         ixgbe_reset(adapter);
4153
4154         return err;
4155 }
4156
4157 /**
4158  * ixgbe_close - Disables a network interface
4159  * @netdev: network interface device structure
4160  *
4161  * Returns 0, this is not allowed to fail
4162  *
4163  * The close entry point is called when an interface is de-activated
4164  * by the OS.  The hardware is still under the drivers control, but
4165  * needs to be disabled.  A global MAC reset is issued to stop the
4166  * hardware, and all transmit and receive resources are freed.
4167  **/
4168 static int ixgbe_close(struct net_device *netdev)
4169 {
4170         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4171
4172         ixgbe_down(adapter);
4173         ixgbe_free_irq(adapter);
4174
4175         ixgbe_free_all_tx_resources(adapter);
4176         ixgbe_free_all_rx_resources(adapter);
4177
4178         ixgbe_release_hw_control(adapter);
4179
4180         return 0;
4181 }
4182
4183 #ifdef CONFIG_PM
4184 static int ixgbe_resume(struct pci_dev *pdev)
4185 {
4186         struct net_device *netdev = pci_get_drvdata(pdev);
4187         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4188         u32 err;
4189
4190         pci_set_power_state(pdev, PCI_D0);
4191         pci_restore_state(pdev);
4192
4193         err = pci_enable_device_mem(pdev);
4194         if (err) {
4195                 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
4196                                 "suspend\n");
4197                 return err;
4198         }
4199         pci_set_master(pdev);
4200
4201         pci_wake_from_d3(pdev, false);
4202
4203         err = ixgbe_init_interrupt_scheme(adapter);
4204         if (err) {
4205                 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
4206                                 "device\n");
4207                 return err;
4208         }
4209
4210         ixgbe_reset(adapter);
4211
4212         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4213
4214         if (netif_running(netdev)) {
4215                 err = ixgbe_open(adapter->netdev);
4216                 if (err)
4217                         return err;
4218         }
4219
4220         netif_device_attach(netdev);
4221
4222         return 0;
4223 }
4224 #endif /* CONFIG_PM */
4225
4226 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4227 {
4228         struct net_device *netdev = pci_get_drvdata(pdev);
4229         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4230         struct ixgbe_hw *hw = &adapter->hw;
4231         u32 ctrl, fctrl;
4232         u32 wufc = adapter->wol;
4233 #ifdef CONFIG_PM
4234         int retval = 0;
4235 #endif
4236
4237         netif_device_detach(netdev);
4238
4239         if (netif_running(netdev)) {
4240                 ixgbe_down(adapter);
4241                 ixgbe_free_irq(adapter);
4242                 ixgbe_free_all_tx_resources(adapter);
4243                 ixgbe_free_all_rx_resources(adapter);
4244         }
4245         ixgbe_clear_interrupt_scheme(adapter);
4246
4247 #ifdef CONFIG_PM
4248         retval = pci_save_state(pdev);
4249         if (retval)
4250                 return retval;
4251
4252 #endif
4253         if (wufc) {
4254                 ixgbe_set_rx_mode(netdev);
4255
4256                 /* turn on all-multi mode if wake on multicast is enabled */
4257                 if (wufc & IXGBE_WUFC_MC) {
4258                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4259                         fctrl |= IXGBE_FCTRL_MPE;
4260                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4261                 }
4262
4263                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4264                 ctrl |= IXGBE_CTRL_GIO_DIS;
4265                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4266
4267                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4268         } else {
4269                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4270                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4271         }
4272
4273         if (wufc && hw->mac.type == ixgbe_mac_82599EB)
4274                 pci_wake_from_d3(pdev, true);
4275         else
4276                 pci_wake_from_d3(pdev, false);
4277
4278         *enable_wake = !!wufc;
4279
4280         ixgbe_release_hw_control(adapter);
4281
4282         pci_disable_device(pdev);
4283
4284         return 0;
4285 }
4286
4287 #ifdef CONFIG_PM
4288 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4289 {
4290         int retval;
4291         bool wake;
4292
4293         retval = __ixgbe_shutdown(pdev, &wake);
4294         if (retval)
4295                 return retval;
4296
4297         if (wake) {
4298                 pci_prepare_to_sleep(pdev);
4299         } else {
4300                 pci_wake_from_d3(pdev, false);
4301                 pci_set_power_state(pdev, PCI_D3hot);
4302         }
4303
4304         return 0;
4305 }
4306 #endif /* CONFIG_PM */
4307
4308 static void ixgbe_shutdown(struct pci_dev *pdev)
4309 {
4310         bool wake;
4311
4312         __ixgbe_shutdown(pdev, &wake);
4313
4314         if (system_state == SYSTEM_POWER_OFF) {
4315                 pci_wake_from_d3(pdev, wake);
4316                 pci_set_power_state(pdev, PCI_D3hot);
4317         }
4318 }
4319
4320 /**
4321  * ixgbe_update_stats - Update the board statistics counters.
4322  * @adapter: board private structure
4323  **/
4324 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4325 {
4326         struct ixgbe_hw *hw = &adapter->hw;
4327         u64 total_mpc = 0;
4328         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4329
4330         if (hw->mac.type == ixgbe_mac_82599EB) {
4331                 u64 rsc_count = 0;
4332                 for (i = 0; i < 16; i++)
4333                         adapter->hw_rx_no_dma_resources +=
4334                                              IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4335                 for (i = 0; i < adapter->num_rx_queues; i++)
4336                         rsc_count += adapter->rx_ring[i].rsc_count;
4337                 adapter->rsc_count = rsc_count;
4338         }
4339
4340         adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
4341         for (i = 0; i < 8; i++) {
4342                 /* for packet buffers not used, the register should read 0 */
4343                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
4344                 missed_rx += mpc;
4345                 adapter->stats.mpc[i] += mpc;
4346                 total_mpc += adapter->stats.mpc[i];
4347                 if (hw->mac.type == ixgbe_mac_82598EB)
4348                         adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
4349                 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
4350                 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
4351                 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
4352                 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
4353                 if (hw->mac.type == ixgbe_mac_82599EB) {
4354                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4355                                                             IXGBE_PXONRXCNT(i));
4356                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4357                                                            IXGBE_PXOFFRXCNT(i));
4358                         adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4359                 } else {
4360                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4361                                                               IXGBE_PXONRXC(i));
4362                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4363                                                              IXGBE_PXOFFRXC(i));
4364                 }
4365                 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
4366                                                             IXGBE_PXONTXC(i));
4367                 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
4368                                                              IXGBE_PXOFFTXC(i));
4369         }
4370         adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
4371         /* work around hardware counting issue */
4372         adapter->stats.gprc -= missed_rx;
4373
4374         /* 82598 hardware only has a 32 bit counter in the high register */
4375         if (hw->mac.type == ixgbe_mac_82599EB) {
4376                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
4377                 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
4378                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
4379                 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
4380                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
4381                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
4382                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
4383                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
4384                 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
4385                 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
4386 #ifdef IXGBE_FCOE
4387                 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
4388                 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
4389                 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
4390                 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
4391                 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
4392                 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
4393 #endif /* IXGBE_FCOE */
4394         } else {
4395                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
4396                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
4397                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
4398                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
4399                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
4400         }
4401         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
4402         adapter->stats.bprc += bprc;
4403         adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
4404         if (hw->mac.type == ixgbe_mac_82598EB)
4405                 adapter->stats.mprc -= bprc;
4406         adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
4407         adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
4408         adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
4409         adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
4410         adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
4411         adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
4412         adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
4413         adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
4414         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
4415         adapter->stats.lxontxc += lxon;
4416         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
4417         adapter->stats.lxofftxc += lxoff;
4418         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4419         adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
4420         adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
4421         /*
4422          * 82598 errata - tx of flow control packets is included in tx counters
4423          */
4424         xon_off_tot = lxon + lxoff;
4425         adapter->stats.gptc -= xon_off_tot;
4426         adapter->stats.mptc -= xon_off_tot;
4427         adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
4428         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4429         adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
4430         adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
4431         adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
4432         adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
4433         adapter->stats.ptc64 -= xon_off_tot;
4434         adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
4435         adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
4436         adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
4437         adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
4438         adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
4439         adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
4440
4441         /* Fill out the OS statistics structure */
4442         adapter->net_stats.multicast = adapter->stats.mprc;
4443
4444         /* Rx Errors */
4445         adapter->net_stats.rx_errors = adapter->stats.crcerrs +
4446                                        adapter->stats.rlec;
4447         adapter->net_stats.rx_dropped = 0;
4448         adapter->net_stats.rx_length_errors = adapter->stats.rlec;
4449         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
4450         adapter->net_stats.rx_missed_errors = total_mpc;
4451 }
4452
4453 /**
4454  * ixgbe_watchdog - Timer Call-back
4455  * @data: pointer to adapter cast into an unsigned long
4456  **/
4457 static void ixgbe_watchdog(unsigned long data)
4458 {
4459         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4460         struct ixgbe_hw *hw = &adapter->hw;
4461         u64 eics = 0;
4462         int i;
4463
4464         /*
4465          *  Do the watchdog outside of interrupt context due to the lovely
4466          * delays that some of the newer hardware requires
4467          */
4468
4469         if (test_bit(__IXGBE_DOWN, &adapter->state))
4470                 goto watchdog_short_circuit;
4471
4472         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
4473                 /*
4474                  * for legacy and MSI interrupts don't set any bits
4475                  * that are enabled for EIAM, because this operation
4476                  * would set *both* EIMS and EICS for any bit in EIAM
4477                  */
4478                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
4479                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
4480                 goto watchdog_reschedule;
4481         }
4482
4483         /* get one bit for every active tx/rx interrupt vector */
4484         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
4485                 struct ixgbe_q_vector *qv = adapter->q_vector[i];
4486                 if (qv->rxr_count || qv->txr_count)
4487                         eics |= ((u64)1 << i);
4488         }
4489
4490         /* Cause software interrupt to ensure rx rings are cleaned */
4491         ixgbe_irq_rearm_queues(adapter, eics);
4492
4493 watchdog_reschedule:
4494         /* Reset the timer */
4495         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
4496
4497 watchdog_short_circuit:
4498         schedule_work(&adapter->watchdog_task);
4499 }
4500
4501 /**
4502  * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
4503  * @work: pointer to work_struct containing our data
4504  **/
4505 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
4506 {
4507         struct ixgbe_adapter *adapter = container_of(work,
4508                                                      struct ixgbe_adapter,
4509                                                      multispeed_fiber_task);
4510         struct ixgbe_hw *hw = &adapter->hw;
4511         u32 autoneg;
4512
4513         adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
4514         autoneg = hw->phy.autoneg_advertised;
4515         if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
4516                 hw->mac.ops.get_link_capabilities(hw, &autoneg,
4517                                                   &hw->mac.autoneg);
4518         if (hw->mac.ops.setup_link_speed)
4519                 hw->mac.ops.setup_link_speed(hw, autoneg, true, true);
4520         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4521         adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
4522 }
4523
4524 /**
4525  * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
4526  * @work: pointer to work_struct containing our data
4527  **/
4528 static void ixgbe_sfp_config_module_task(struct work_struct *work)
4529 {
4530         struct ixgbe_adapter *adapter = container_of(work,
4531                                                      struct ixgbe_adapter,
4532                                                      sfp_config_module_task);
4533         struct ixgbe_hw *hw = &adapter->hw;
4534         u32 err;
4535
4536         adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
4537         err = hw->phy.ops.identify_sfp(hw);
4538         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4539                 dev_err(&adapter->pdev->dev, "failed to initialize because "
4540                         "an unsupported SFP+ module type was detected.\n"
4541                         "Reload the driver after installing a supported "
4542                         "module.\n");
4543                 ixgbe_down(adapter);
4544                 return;
4545         }
4546         hw->mac.ops.setup_sfp(hw);
4547
4548         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
4549                 /* This will also work for DA Twinax connections */
4550                 schedule_work(&adapter->multispeed_fiber_task);
4551         adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
4552 }
4553
4554 /**
4555  * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
4556  * @work: pointer to work_struct containing our data
4557  **/
4558 static void ixgbe_fdir_reinit_task(struct work_struct *work)
4559 {
4560         struct ixgbe_adapter *adapter = container_of(work,
4561                                                      struct ixgbe_adapter,
4562                                                      fdir_reinit_task);
4563         struct ixgbe_hw *hw = &adapter->hw;
4564         int i;
4565
4566         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
4567                 for (i = 0; i < adapter->num_tx_queues; i++)
4568                         set_bit(__IXGBE_FDIR_INIT_DONE,
4569                                 &(adapter->tx_ring[i].reinit_state));
4570         } else {
4571                 DPRINTK(PROBE, ERR, "failed to finish FDIR re-initialization, "
4572                         "ignored adding FDIR ATR filters \n");
4573         }
4574         /* Done FDIR Re-initialization, enable transmits */
4575         netif_tx_start_all_queues(adapter->netdev);
4576 }
4577
4578 /**
4579  * ixgbe_watchdog_task - worker thread to bring link up
4580  * @work: pointer to work_struct containing our data
4581  **/
4582 static void ixgbe_watchdog_task(struct work_struct *work)
4583 {
4584         struct ixgbe_adapter *adapter = container_of(work,
4585                                                      struct ixgbe_adapter,
4586                                                      watchdog_task);
4587         struct net_device *netdev = adapter->netdev;
4588         struct ixgbe_hw *hw = &adapter->hw;
4589         u32 link_speed = adapter->link_speed;
4590         bool link_up = adapter->link_up;
4591         int i;
4592         struct ixgbe_ring *tx_ring;
4593         int some_tx_pending = 0;
4594
4595         adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
4596
4597         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
4598                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
4599                 if (link_up) {
4600 #ifdef CONFIG_DCB
4601                         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4602                                 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
4603                                         hw->mac.ops.fc_enable(hw, i);
4604                         } else {
4605                                 hw->mac.ops.fc_enable(hw, 0);
4606                         }
4607 #else
4608                         hw->mac.ops.fc_enable(hw, 0);
4609 #endif
4610                 }
4611
4612                 if (link_up ||
4613                     time_after(jiffies, (adapter->link_check_timeout +
4614                                          IXGBE_TRY_LINK_TIMEOUT))) {
4615                         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4616                         IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
4617                 }
4618                 adapter->link_up = link_up;
4619                 adapter->link_speed = link_speed;
4620         }
4621
4622         if (link_up) {
4623                 if (!netif_carrier_ok(netdev)) {
4624                         bool flow_rx, flow_tx;
4625
4626                         if (hw->mac.type == ixgbe_mac_82599EB) {
4627                                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
4628                                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
4629                                 flow_rx = (mflcn & IXGBE_MFLCN_RFCE);
4630                                 flow_tx = (fccfg & IXGBE_FCCFG_TFCE_802_3X);
4631                         } else {
4632                                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4633                                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
4634                                 flow_rx = (frctl & IXGBE_FCTRL_RFCE);
4635                                 flow_tx = (rmcs & IXGBE_RMCS_TFCE_802_3X);
4636                         }
4637
4638                         printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
4639                                "Flow Control: %s\n",
4640                                netdev->name,
4641                                (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
4642                                 "10 Gbps" :
4643                                 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
4644                                  "1 Gbps" : "unknown speed")),
4645                                ((flow_rx && flow_tx) ? "RX/TX" :
4646                                 (flow_rx ? "RX" :
4647                                 (flow_tx ? "TX" : "None"))));
4648
4649                         netif_carrier_on(netdev);
4650                 } else {
4651                         /* Force detection of hung controller */
4652                         adapter->detect_tx_hung = true;
4653                 }
4654         } else {
4655                 adapter->link_up = false;
4656                 adapter->link_speed = 0;
4657                 if (netif_carrier_ok(netdev)) {
4658                         printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
4659                                netdev->name);
4660                         netif_carrier_off(netdev);
4661                 }
4662         }
4663
4664         if (!netif_carrier_ok(netdev)) {
4665                 for (i = 0; i < adapter->num_tx_queues; i++) {
4666                         tx_ring = &adapter->tx_ring[i];
4667                         if (tx_ring->next_to_use != tx_ring->next_to_clean) {
4668                                 some_tx_pending = 1;
4669                                 break;
4670                         }
4671                 }
4672
4673                 if (some_tx_pending) {
4674                         /* We've lost link, so the controller stops DMA,
4675                          * but we've got queued Tx work that's never going
4676                          * to get done, so reset controller to flush Tx.
4677                          * (Do the reset outside of interrupt context).
4678                          */
4679                          schedule_work(&adapter->reset_task);
4680                 }
4681         }
4682
4683         ixgbe_update_stats(adapter);
4684         adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
4685 }
4686
4687 static int ixgbe_tso(struct ixgbe_adapter *adapter,
4688                      struct ixgbe_ring *tx_ring, struct sk_buff *skb,
4689                      u32 tx_flags, u8 *hdr_len)
4690 {
4691         struct ixgbe_adv_tx_context_desc *context_desc;
4692         unsigned int i;
4693         int err;
4694         struct ixgbe_tx_buffer *tx_buffer_info;
4695         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
4696         u32 mss_l4len_idx, l4len;
4697
4698         if (skb_is_gso(skb)) {
4699                 if (skb_header_cloned(skb)) {
4700                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4701                         if (err)
4702                                 return err;
4703                 }
4704                 l4len = tcp_hdrlen(skb);
4705                 *hdr_len += l4len;
4706
4707                 if (skb->protocol == htons(ETH_P_IP)) {
4708                         struct iphdr *iph = ip_hdr(skb);
4709                         iph->tot_len = 0;
4710                         iph->check = 0;
4711                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4712                                                                  iph->daddr, 0,
4713                                                                  IPPROTO_TCP,
4714                                                                  0);
4715                         adapter->hw_tso_ctxt++;
4716                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
4717                         ipv6_hdr(skb)->payload_len = 0;
4718                         tcp_hdr(skb)->check =
4719                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4720                                              &ipv6_hdr(skb)->daddr,
4721                                              0, IPPROTO_TCP, 0);
4722                         adapter->hw_tso6_ctxt++;
4723                 }
4724
4725                 i = tx_ring->next_to_use;
4726
4727                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4728                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4729
4730                 /* VLAN MACLEN IPLEN */
4731                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4732                         vlan_macip_lens |=
4733                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4734                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
4735                                     IXGBE_ADVTXD_MACLEN_SHIFT);
4736                 *hdr_len += skb_network_offset(skb);
4737                 vlan_macip_lens |=
4738                     (skb_transport_header(skb) - skb_network_header(skb));
4739                 *hdr_len +=
4740                     (skb_transport_header(skb) - skb_network_header(skb));
4741                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4742                 context_desc->seqnum_seed = 0;
4743
4744                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4745                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
4746                                    IXGBE_ADVTXD_DTYP_CTXT);
4747
4748                 if (skb->protocol == htons(ETH_P_IP))
4749                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4750                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
4751                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4752
4753                 /* MSS L4LEN IDX */
4754                 mss_l4len_idx =
4755                     (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
4756                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
4757                 /* use index 1 for TSO */
4758                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4759                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4760
4761                 tx_buffer_info->time_stamp = jiffies;
4762                 tx_buffer_info->next_to_watch = i;
4763
4764                 i++;
4765                 if (i == tx_ring->count)
4766                         i = 0;
4767                 tx_ring->next_to_use = i;
4768
4769                 return true;
4770         }
4771         return false;
4772 }
4773
4774 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
4775                           struct ixgbe_ring *tx_ring,
4776                           struct sk_buff *skb, u32 tx_flags)
4777 {
4778         struct ixgbe_adv_tx_context_desc *context_desc;
4779         unsigned int i;
4780         struct ixgbe_tx_buffer *tx_buffer_info;
4781         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
4782
4783         if (skb->ip_summed == CHECKSUM_PARTIAL ||
4784             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
4785                 i = tx_ring->next_to_use;
4786                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4787                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4788
4789                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4790                         vlan_macip_lens |=
4791                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4792                 vlan_macip_lens |= (skb_network_offset(skb) <<
4793                                     IXGBE_ADVTXD_MACLEN_SHIFT);
4794                 if (skb->ip_summed == CHECKSUM_PARTIAL)
4795                         vlan_macip_lens |= (skb_transport_header(skb) -
4796                                             skb_network_header(skb));
4797
4798                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4799                 context_desc->seqnum_seed = 0;
4800
4801                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
4802                                     IXGBE_ADVTXD_DTYP_CTXT);
4803
4804                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
4805                         switch (skb->protocol) {
4806                         case cpu_to_be16(ETH_P_IP):
4807                                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4808                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4809                                         type_tucmd_mlhl |=
4810                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4811                                 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
4812                                         type_tucmd_mlhl |=
4813                                                 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
4814                                 break;
4815                         case cpu_to_be16(ETH_P_IPV6):
4816                                 /* XXX what about other V6 headers?? */
4817                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4818                                         type_tucmd_mlhl |=
4819                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4820                                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
4821                                         type_tucmd_mlhl |=
4822                                                 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
4823                                 break;
4824                         default:
4825                                 if (unlikely(net_ratelimit())) {
4826                                         DPRINTK(PROBE, WARNING,
4827                                          "partial checksum but proto=%x!\n",
4828                                          skb->protocol);
4829                                 }
4830                                 break;
4831                         }
4832                 }
4833
4834                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4835                 /* use index zero for tx checksum offload */
4836                 context_desc->mss_l4len_idx = 0;
4837
4838                 tx_buffer_info->time_stamp = jiffies;
4839                 tx_buffer_info->next_to_watch = i;
4840
4841                 adapter->hw_csum_tx_good++;
4842                 i++;
4843                 if (i == tx_ring->count)
4844                         i = 0;
4845                 tx_ring->next_to_use = i;
4846
4847                 return true;
4848         }
4849
4850         return false;
4851 }
4852
4853 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
4854                         struct ixgbe_ring *tx_ring,
4855                         struct sk_buff *skb, u32 tx_flags,
4856                         unsigned int first)
4857 {
4858         struct ixgbe_tx_buffer *tx_buffer_info;
4859         unsigned int len;
4860         unsigned int total = skb->len;
4861         unsigned int offset = 0, size, count = 0, i;
4862         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
4863         unsigned int f;
4864         dma_addr_t *map;
4865
4866         i = tx_ring->next_to_use;
4867
4868         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
4869                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
4870                 return 0;
4871         }
4872
4873         map = skb_shinfo(skb)->dma_maps;
4874
4875         if (tx_flags & IXGBE_TX_FLAGS_FCOE)
4876                 /* excluding fcoe_crc_eof for FCoE */
4877                 total -= sizeof(struct fcoe_crc_eof);
4878
4879         len = min(skb_headlen(skb), total);
4880         while (len) {
4881                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4882                 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4883
4884                 tx_buffer_info->length = size;
4885                 tx_buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
4886                 tx_buffer_info->time_stamp = jiffies;
4887                 tx_buffer_info->next_to_watch = i;
4888
4889                 len -= size;
4890                 total -= size;
4891                 offset += size;
4892                 count++;
4893
4894                 if (len) {
4895                         i++;
4896                         if (i == tx_ring->count)
4897                                 i = 0;
4898                 }
4899         }
4900
4901         for (f = 0; f < nr_frags; f++) {
4902                 struct skb_frag_struct *frag;
4903
4904                 frag = &skb_shinfo(skb)->frags[f];
4905                 len = min((unsigned int)frag->size, total);
4906                 offset = 0;
4907
4908                 while (len) {
4909                         i++;
4910                         if (i == tx_ring->count)
4911                                 i = 0;
4912
4913                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
4914                         size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4915
4916                         tx_buffer_info->length = size;
4917                         tx_buffer_info->dma = map[f] + offset;
4918                         tx_buffer_info->time_stamp = jiffies;
4919                         tx_buffer_info->next_to_watch = i;
4920
4921                         len -= size;
4922                         total -= size;
4923                         offset += size;
4924                         count++;
4925                 }
4926                 if (total == 0)
4927                         break;
4928         }
4929
4930         tx_ring->tx_buffer_info[i].skb = skb;
4931         tx_ring->tx_buffer_info[first].next_to_watch = i;
4932
4933         return count;
4934 }
4935
4936 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
4937                            struct ixgbe_ring *tx_ring,
4938                            int tx_flags, int count, u32 paylen, u8 hdr_len)
4939 {
4940         union ixgbe_adv_tx_desc *tx_desc = NULL;
4941         struct ixgbe_tx_buffer *tx_buffer_info;
4942         u32 olinfo_status = 0, cmd_type_len = 0;
4943         unsigned int i;
4944         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
4945
4946         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
4947
4948         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
4949
4950         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4951                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
4952
4953         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
4954                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
4955
4956                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
4957                                  IXGBE_ADVTXD_POPTS_SHIFT;
4958
4959                 /* use index 1 context for tso */
4960                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4961                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
4962                         olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
4963                                          IXGBE_ADVTXD_POPTS_SHIFT;
4964
4965         } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
4966                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
4967                                  IXGBE_ADVTXD_POPTS_SHIFT;
4968
4969         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
4970                 olinfo_status |= IXGBE_ADVTXD_CC;
4971                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4972                 if (tx_flags & IXGBE_TX_FLAGS_FSO)
4973                         cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
4974         }
4975
4976         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
4977
4978         i = tx_ring->next_to_use;
4979         while (count--) {
4980                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4981                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
4982                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
4983                 tx_desc->read.cmd_type_len =
4984                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
4985                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4986                 i++;
4987                 if (i == tx_ring->count)
4988                         i = 0;
4989         }
4990
4991         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
4992
4993         /*
4994          * Force memory writes to complete before letting h/w
4995          * know there are new descriptors to fetch.  (Only
4996          * applicable for weak-ordered memory model archs,
4997          * such as IA-64).
4998          */
4999         wmb();
5000
5001         tx_ring->next_to_use = i;
5002         writel(i, adapter->hw.hw_addr + tx_ring->tail);
5003 }
5004
5005 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5006                       int queue, u32 tx_flags)
5007 {
5008         /* Right now, we support IPv4 only */
5009         struct ixgbe_atr_input atr_input;
5010         struct tcphdr *th;
5011         struct udphdr *uh;
5012         struct iphdr *iph = ip_hdr(skb);
5013         struct ethhdr *eth = (struct ethhdr *)skb->data;
5014         u16 vlan_id, src_port, dst_port, flex_bytes;
5015         u32 src_ipv4_addr, dst_ipv4_addr;
5016         u8 l4type = 0;
5017
5018         /* check if we're UDP or TCP */
5019         if (iph->protocol == IPPROTO_TCP) {
5020                 th = tcp_hdr(skb);
5021                 src_port = th->source;
5022                 dst_port = th->dest;
5023                 l4type |= IXGBE_ATR_L4TYPE_TCP;
5024                 /* l4type IPv4 type is 0, no need to assign */
5025         } else if(iph->protocol == IPPROTO_UDP) {
5026                 uh = udp_hdr(skb);
5027                 src_port = uh->source;
5028                 dst_port = uh->dest;
5029                 l4type |= IXGBE_ATR_L4TYPE_UDP;
5030                 /* l4type IPv4 type is 0, no need to assign */
5031         } else {
5032                 /* Unsupported L4 header, just bail here */
5033                 return;
5034         }
5035
5036         memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
5037
5038         vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
5039                    IXGBE_TX_FLAGS_VLAN_SHIFT;
5040         src_ipv4_addr = iph->saddr;
5041         dst_ipv4_addr = iph->daddr;
5042         flex_bytes = eth->h_proto;
5043
5044         ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
5045         ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
5046         ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
5047         ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
5048         ixgbe_atr_set_l4type_82599(&atr_input, l4type);
5049         /* src and dst are inverted, think how the receiver sees them */
5050         ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
5051         ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
5052
5053         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
5054         ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
5055 }
5056
5057 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
5058                                  struct ixgbe_ring *tx_ring, int size)
5059 {
5060         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5061
5062         netif_stop_subqueue(netdev, tx_ring->queue_index);
5063         /* Herbert's original patch had:
5064          *  smp_mb__after_netif_stop_queue();
5065          * but since that doesn't exist yet, just open code it. */
5066         smp_mb();
5067
5068         /* We need to check again in a case another CPU has just
5069          * made room available. */
5070         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
5071                 return -EBUSY;
5072
5073         /* A reprieve! - use start_queue because it doesn't call schedule */
5074         netif_start_subqueue(netdev, tx_ring->queue_index);
5075         ++adapter->restart_queue;
5076         return 0;
5077 }
5078
5079 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
5080                               struct ixgbe_ring *tx_ring, int size)
5081 {
5082         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
5083                 return 0;
5084         return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
5085 }
5086
5087 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
5088 {
5089         struct ixgbe_adapter *adapter = netdev_priv(dev);
5090
5091         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
5092                 return smp_processor_id();
5093
5094         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5095                 return 0;  /* All traffic should default to class 0 */
5096
5097         return skb_tx_hash(dev, skb);
5098 }
5099
5100 static int ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
5101 {
5102         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5103         struct ixgbe_ring *tx_ring;
5104         unsigned int first;
5105         unsigned int tx_flags = 0;
5106         u8 hdr_len = 0;
5107         int r_idx = 0, tso;
5108         int count = 0;
5109         unsigned int f;
5110
5111         r_idx = skb->queue_mapping;
5112         tx_ring = &adapter->tx_ring[r_idx];
5113
5114         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
5115                 tx_flags |= vlan_tx_tag_get(skb);
5116                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5117                         tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
5118                         tx_flags |= (skb->queue_mapping << 13);
5119                 }
5120                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5121                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5122         } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5123                 tx_flags |= (skb->queue_mapping << 13);
5124                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5125                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5126         }
5127
5128         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5129             (skb->protocol == htons(ETH_P_FCOE)))
5130                 tx_flags |= IXGBE_TX_FLAGS_FCOE;
5131
5132         /* four things can cause us to need a context descriptor */
5133         if (skb_is_gso(skb) ||
5134             (skb->ip_summed == CHECKSUM_PARTIAL) ||
5135             (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
5136             (tx_flags & IXGBE_TX_FLAGS_FCOE))
5137                 count++;
5138
5139         count += TXD_USE_COUNT(skb_headlen(skb));
5140         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5141                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5142
5143         if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
5144                 adapter->tx_busy++;
5145                 return NETDEV_TX_BUSY;
5146         }
5147
5148         first = tx_ring->next_to_use;
5149         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5150 #ifdef IXGBE_FCOE
5151                 /* setup tx offload for FCoE */
5152                 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5153                 if (tso < 0) {
5154                         dev_kfree_skb_any(skb);
5155                         return NETDEV_TX_OK;
5156                 }
5157                 if (tso)
5158                         tx_flags |= IXGBE_TX_FLAGS_FSO;
5159 #endif /* IXGBE_FCOE */
5160         } else {
5161                 if (skb->protocol == htons(ETH_P_IP))
5162                         tx_flags |= IXGBE_TX_FLAGS_IPV4;
5163                 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5164                 if (tso < 0) {
5165                         dev_kfree_skb_any(skb);
5166                         return NETDEV_TX_OK;
5167                 }
5168
5169                 if (tso)
5170                         tx_flags |= IXGBE_TX_FLAGS_TSO;
5171                 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
5172                          (skb->ip_summed == CHECKSUM_PARTIAL))
5173                         tx_flags |= IXGBE_TX_FLAGS_CSUM;
5174         }
5175
5176         count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
5177         if (count) {
5178                 /* add the ATR filter if ATR is on */
5179                 if (tx_ring->atr_sample_rate) {
5180                         ++tx_ring->atr_count;
5181                         if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
5182                              test_bit(__IXGBE_FDIR_INIT_DONE,
5183                                       &tx_ring->reinit_state)) {
5184                                 ixgbe_atr(adapter, skb, tx_ring->queue_index,
5185                                           tx_flags);
5186                                 tx_ring->atr_count = 0;
5187                         }
5188                 }
5189                 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
5190                                hdr_len);
5191                 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
5192
5193         } else {
5194                 dev_kfree_skb_any(skb);
5195                 tx_ring->tx_buffer_info[first].time_stamp = 0;
5196                 tx_ring->next_to_use = first;
5197         }
5198
5199         return NETDEV_TX_OK;
5200 }
5201
5202 /**
5203  * ixgbe_get_stats - Get System Network Statistics
5204  * @netdev: network interface device structure
5205  *
5206  * Returns the address of the device statistics structure.
5207  * The statistics are actually updated from the timer callback.
5208  **/
5209 static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
5210 {
5211         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5212
5213         /* only return the current stats */
5214         return &adapter->net_stats;
5215 }
5216
5217 /**
5218  * ixgbe_set_mac - Change the Ethernet Address of the NIC
5219  * @netdev: network interface device structure
5220  * @p: pointer to an address structure
5221  *
5222  * Returns 0 on success, negative on failure
5223  **/
5224 static int ixgbe_set_mac(struct net_device *netdev, void *p)
5225 {
5226         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5227         struct ixgbe_hw *hw = &adapter->hw;
5228         struct sockaddr *addr = p;
5229
5230         if (!is_valid_ether_addr(addr->sa_data))
5231                 return -EADDRNOTAVAIL;
5232
5233         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
5234         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5235
5236         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
5237
5238         return 0;
5239 }
5240
5241 static int
5242 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
5243 {
5244         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5245         struct ixgbe_hw *hw = &adapter->hw;
5246         u16 value;
5247         int rc;
5248
5249         if (prtad != hw->phy.mdio.prtad)
5250                 return -EINVAL;
5251         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
5252         if (!rc)
5253                 rc = value;
5254         return rc;
5255 }
5256
5257 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
5258                             u16 addr, u16 value)
5259 {
5260         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5261         struct ixgbe_hw *hw = &adapter->hw;
5262
5263         if (prtad != hw->phy.mdio.prtad)
5264                 return -EINVAL;
5265         return hw->phy.ops.write_reg(hw, addr, devad, value);
5266 }
5267
5268 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
5269 {
5270         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5271
5272         return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
5273 }
5274
5275 /**
5276  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
5277  * netdev->dev_addrs
5278  * @netdev: network interface device structure
5279  *
5280  * Returns non-zero on failure
5281  **/
5282 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
5283 {
5284         int err = 0;
5285         struct ixgbe_adapter *adapter = netdev_priv(dev);
5286         struct ixgbe_mac_info *mac = &adapter->hw.mac;
5287
5288         if (is_valid_ether_addr(mac->san_addr)) {
5289                 rtnl_lock();
5290                 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5291                 rtnl_unlock();
5292         }
5293         return err;
5294 }
5295
5296 /**
5297  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
5298  * netdev->dev_addrs
5299  * @netdev: network interface device structure
5300  *
5301  * Returns non-zero on failure
5302  **/
5303 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
5304 {
5305         int err = 0;
5306         struct ixgbe_adapter *adapter = netdev_priv(dev);
5307         struct ixgbe_mac_info *mac = &adapter->hw.mac;
5308
5309         if (is_valid_ether_addr(mac->san_addr)) {
5310                 rtnl_lock();
5311                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5312                 rtnl_unlock();
5313         }
5314         return err;
5315 }
5316
5317 #ifdef CONFIG_NET_POLL_CONTROLLER
5318 /*
5319  * Polling 'interrupt' - used by things like netconsole to send skbs
5320  * without having to re-enable interrupts. It's not called while
5321  * the interrupt routine is executing.
5322  */
5323 static void ixgbe_netpoll(struct net_device *netdev)
5324 {
5325         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5326
5327         disable_irq(adapter->pdev->irq);
5328         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
5329         ixgbe_intr(adapter->pdev->irq, netdev);
5330         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
5331         enable_irq(adapter->pdev->irq);
5332 }
5333 #endif
5334
5335 static const struct net_device_ops ixgbe_netdev_ops = {
5336         .ndo_open               = ixgbe_open,
5337         .ndo_stop               = ixgbe_close,
5338         .ndo_start_xmit         = ixgbe_xmit_frame,
5339         .ndo_select_queue       = ixgbe_select_queue,
5340         .ndo_get_stats          = ixgbe_get_stats,
5341         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
5342         .ndo_set_multicast_list = ixgbe_set_rx_mode,
5343         .ndo_validate_addr      = eth_validate_addr,
5344         .ndo_set_mac_address    = ixgbe_set_mac,
5345         .ndo_change_mtu         = ixgbe_change_mtu,
5346         .ndo_tx_timeout         = ixgbe_tx_timeout,
5347         .ndo_vlan_rx_register   = ixgbe_vlan_rx_register,
5348         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
5349         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
5350         .ndo_do_ioctl           = ixgbe_ioctl,
5351 #ifdef CONFIG_NET_POLL_CONTROLLER
5352         .ndo_poll_controller    = ixgbe_netpoll,
5353 #endif
5354 #ifdef IXGBE_FCOE
5355         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
5356         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
5357 #endif /* IXGBE_FCOE */
5358 };
5359
5360 /**
5361  * ixgbe_probe - Device Initialization Routine
5362  * @pdev: PCI device information struct
5363  * @ent: entry in ixgbe_pci_tbl
5364  *
5365  * Returns 0 on success, negative on failure
5366  *
5367  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
5368  * The OS initialization, configuring of the adapter private structure,
5369  * and a hardware reset occur.
5370  **/
5371 static int __devinit ixgbe_probe(struct pci_dev *pdev,
5372                                  const struct pci_device_id *ent)
5373 {
5374         struct net_device *netdev;
5375         struct ixgbe_adapter *adapter = NULL;
5376         struct ixgbe_hw *hw;
5377         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
5378         static int cards_found;
5379         int i, err, pci_using_dac;
5380 #ifdef IXGBE_FCOE
5381         u16 device_caps;
5382 #endif
5383         u32 part_num, eec;
5384
5385         err = pci_enable_device_mem(pdev);
5386         if (err)
5387                 return err;
5388
5389         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
5390             !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
5391                 pci_using_dac = 1;
5392         } else {
5393                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5394                 if (err) {
5395                         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
5396                         if (err) {
5397                                 dev_err(&pdev->dev, "No usable DMA "
5398                                         "configuration, aborting\n");
5399                                 goto err_dma;
5400                         }
5401                 }
5402                 pci_using_dac = 0;
5403         }
5404
5405         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
5406                                            IORESOURCE_MEM), ixgbe_driver_name);
5407         if (err) {
5408                 dev_err(&pdev->dev,
5409                         "pci_request_selected_regions failed 0x%x\n", err);
5410                 goto err_pci_reg;
5411         }
5412
5413         err = pci_enable_pcie_error_reporting(pdev);
5414         if (err) {
5415                 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
5416                                     "0x%x\n", err);
5417                 /* non-fatal, continue */
5418         }
5419
5420         pci_set_master(pdev);
5421         pci_save_state(pdev);
5422
5423         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
5424         if (!netdev) {
5425                 err = -ENOMEM;
5426                 goto err_alloc_etherdev;
5427         }
5428
5429         SET_NETDEV_DEV(netdev, &pdev->dev);
5430
5431         pci_set_drvdata(pdev, netdev);
5432         adapter = netdev_priv(netdev);
5433
5434         adapter->netdev = netdev;
5435         adapter->pdev = pdev;
5436         hw = &adapter->hw;
5437         hw->back = adapter;
5438         adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
5439
5440         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
5441                               pci_resource_len(pdev, 0));
5442         if (!hw->hw_addr) {
5443                 err = -EIO;
5444                 goto err_ioremap;
5445         }
5446
5447         for (i = 1; i <= 5; i++) {
5448                 if (pci_resource_len(pdev, i) == 0)
5449                         continue;
5450         }
5451
5452         netdev->netdev_ops = &ixgbe_netdev_ops;
5453         ixgbe_set_ethtool_ops(netdev);
5454         netdev->watchdog_timeo = 5 * HZ;
5455         strcpy(netdev->name, pci_name(pdev));
5456
5457         adapter->bd_number = cards_found;
5458
5459         /* Setup hw api */
5460         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
5461         hw->mac.type  = ii->mac;
5462
5463         /* EEPROM */
5464         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
5465         eec = IXGBE_READ_REG(hw, IXGBE_EEC);
5466         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
5467         if (!(eec & (1 << 8)))
5468                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
5469
5470         /* PHY */
5471         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
5472         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
5473         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
5474         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
5475         hw->phy.mdio.mmds = 0;
5476         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
5477         hw->phy.mdio.dev = netdev;
5478         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
5479         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
5480
5481         /* set up this timer and work struct before calling get_invariants
5482          * which might start the timer
5483          */
5484         init_timer(&adapter->sfp_timer);
5485         adapter->sfp_timer.function = &ixgbe_sfp_timer;
5486         adapter->sfp_timer.data = (unsigned long) adapter;
5487
5488         INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
5489
5490         /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
5491         INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
5492
5493         /* a new SFP+ module arrival, called from GPI SDP2 context */
5494         INIT_WORK(&adapter->sfp_config_module_task,
5495                   ixgbe_sfp_config_module_task);
5496
5497         ii->get_invariants(hw);
5498
5499         /* setup the private structure */
5500         err = ixgbe_sw_init(adapter);
5501         if (err)
5502                 goto err_sw_init;
5503
5504         /*
5505          * If there is a fan on this device and it has failed log the
5506          * failure.
5507          */
5508         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5509                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5510                 if (esdp & IXGBE_ESDP_SDP1)
5511                         DPRINTK(PROBE, CRIT,
5512                                 "Fan has stopped, replace the adapter\n");
5513         }
5514
5515         /* reset_hw fills in the perm_addr as well */
5516         err = hw->mac.ops.reset_hw(hw);
5517         if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
5518             hw->mac.type == ixgbe_mac_82598EB) {
5519                 /*
5520                  * Start a kernel thread to watch for a module to arrive.
5521                  * Only do this for 82598, since 82599 will generate
5522                  * interrupts on module arrival.
5523                  */
5524                 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5525                 mod_timer(&adapter->sfp_timer,
5526                           round_jiffies(jiffies + (2 * HZ)));
5527                 err = 0;
5528         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5529                 dev_err(&adapter->pdev->dev, "failed to initialize because "
5530                         "an unsupported SFP+ module type was detected.\n"
5531                         "Reload the driver after installing a supported "
5532                         "module.\n");
5533                 goto err_sw_init;
5534         } else if (err) {
5535                 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
5536                 goto err_sw_init;
5537         }
5538
5539         netdev->features = NETIF_F_SG |
5540                            NETIF_F_IP_CSUM |
5541                            NETIF_F_HW_VLAN_TX |
5542                            NETIF_F_HW_VLAN_RX |
5543                            NETIF_F_HW_VLAN_FILTER;
5544
5545         netdev->features |= NETIF_F_IPV6_CSUM;
5546         netdev->features |= NETIF_F_TSO;
5547         netdev->features |= NETIF_F_TSO6;
5548         netdev->features |= NETIF_F_GRO;
5549
5550         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
5551                 netdev->features |= NETIF_F_SCTP_CSUM;
5552
5553         netdev->vlan_features |= NETIF_F_TSO;
5554         netdev->vlan_features |= NETIF_F_TSO6;
5555         netdev->vlan_features |= NETIF_F_IP_CSUM;
5556         netdev->vlan_features |= NETIF_F_SG;
5557
5558         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5559                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
5560
5561 #ifdef CONFIG_IXGBE_DCB
5562         netdev->dcbnl_ops = &dcbnl_ops;
5563 #endif
5564
5565 #ifdef IXGBE_FCOE
5566         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
5567                 if (hw->mac.ops.get_device_caps) {
5568                         hw->mac.ops.get_device_caps(hw, &device_caps);
5569                         if (!(device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)) {
5570                                 netdev->features |= NETIF_F_FCOE_CRC;
5571                                 netdev->features |= NETIF_F_FSO;
5572                                 netdev->fcoe_ddp_xid = IXGBE_FCOE_DDP_MAX - 1;
5573                                 DPRINTK(DRV, INFO, "FCoE enabled, "
5574                                         "disabling Flow Director\n");
5575                                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
5576                                 adapter->flags &=
5577                                         ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
5578                                 adapter->atr_sample_rate = 0;
5579                         } else {
5580                                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5581                         }
5582                 }
5583         }
5584 #endif /* IXGBE_FCOE */
5585         if (pci_using_dac)
5586                 netdev->features |= NETIF_F_HIGHDMA;
5587
5588         if (adapter->flags & IXGBE_FLAG2_RSC_ENABLED)
5589                 netdev->features |= NETIF_F_LRO;
5590
5591         /* make sure the EEPROM is good */
5592         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
5593                 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
5594                 err = -EIO;
5595                 goto err_eeprom;
5596         }
5597
5598         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
5599         memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
5600
5601         if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
5602                 dev_err(&pdev->dev, "invalid MAC address\n");
5603                 err = -EIO;
5604                 goto err_eeprom;
5605         }
5606
5607         init_timer(&adapter->watchdog_timer);
5608         adapter->watchdog_timer.function = &ixgbe_watchdog;
5609         adapter->watchdog_timer.data = (unsigned long)adapter;
5610
5611         INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
5612         INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
5613
5614         err = ixgbe_init_interrupt_scheme(adapter);
5615         if (err)
5616                 goto err_sw_init;
5617
5618         switch (pdev->device) {
5619         case IXGBE_DEV_ID_82599_KX4:
5620                 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
5621                                 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
5622                 /* Enable ACPI wakeup in GRC */
5623                 IXGBE_WRITE_REG(hw, IXGBE_GRC,
5624                              (IXGBE_READ_REG(hw, IXGBE_GRC) & ~IXGBE_GRC_APME));
5625                 break;
5626         default:
5627                 adapter->wol = 0;
5628                 break;
5629         }
5630         device_init_wakeup(&adapter->pdev->dev, true);
5631         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5632
5633         /* pick up the PCI bus settings for reporting later */
5634         hw->mac.ops.get_bus_info(hw);
5635
5636         /* print bus type/speed/width info */
5637         dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
5638                 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
5639                  (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
5640                 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
5641                  (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
5642                  (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
5643                  "Unknown"),
5644                 netdev->dev_addr);
5645         ixgbe_read_pba_num_generic(hw, &part_num);
5646         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
5647                 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
5648                          hw->mac.type, hw->phy.type, hw->phy.sfp_type,
5649                          (part_num >> 8), (part_num & 0xff));
5650         else
5651                 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5652                          hw->mac.type, hw->phy.type,
5653                          (part_num >> 8), (part_num & 0xff));
5654
5655         if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
5656                 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
5657                          "this card is not sufficient for optimal "
5658                          "performance.\n");
5659                 dev_warn(&pdev->dev, "For optimal performance a x8 "
5660                          "PCI-Express slot is required.\n");
5661         }
5662
5663         /* save off EEPROM version number */
5664         hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
5665
5666         /* reset the hardware with the new settings */
5667         err = hw->mac.ops.start_hw(hw);
5668
5669         if (err == IXGBE_ERR_EEPROM_VERSION) {
5670                 /* We are running on a pre-production device, log a warning */
5671                 dev_warn(&pdev->dev, "This device is a pre-production "
5672                          "adapter/LOM.  Please be aware there may be issues "
5673                          "associated with your hardware.  If you are "
5674                          "experiencing problems please contact your Intel or "
5675                          "hardware representative who provided you with this "
5676                          "hardware.\n");
5677         }
5678         strcpy(netdev->name, "eth%d");
5679         err = register_netdev(netdev);
5680         if (err)
5681                 goto err_register;
5682
5683         /* carrier off reporting is important to ethtool even BEFORE open */
5684         netif_carrier_off(netdev);
5685
5686         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5687             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5688                 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
5689
5690 #ifdef CONFIG_IXGBE_DCA
5691         if (dca_add_requester(&pdev->dev) == 0) {
5692                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
5693                 ixgbe_setup_dca(adapter);
5694         }
5695 #endif
5696         /* add san mac addr to netdev */
5697         ixgbe_add_sanmac_netdev(netdev);
5698
5699         dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
5700         cards_found++;
5701         return 0;
5702
5703 err_register:
5704         ixgbe_release_hw_control(adapter);
5705         ixgbe_clear_interrupt_scheme(adapter);
5706 err_sw_init:
5707 err_eeprom:
5708         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5709         del_timer_sync(&adapter->sfp_timer);
5710         cancel_work_sync(&adapter->sfp_task);
5711         cancel_work_sync(&adapter->multispeed_fiber_task);
5712         cancel_work_sync(&adapter->sfp_config_module_task);
5713         iounmap(hw->hw_addr);
5714 err_ioremap:
5715         free_netdev(netdev);
5716 err_alloc_etherdev:
5717         pci_release_selected_regions(pdev, pci_select_bars(pdev,
5718                                      IORESOURCE_MEM));
5719 err_pci_reg:
5720 err_dma:
5721         pci_disable_device(pdev);
5722         return err;
5723 }
5724
5725 /**
5726  * ixgbe_remove - Device Removal Routine
5727  * @pdev: PCI device information struct
5728  *
5729  * ixgbe_remove is called by the PCI subsystem to alert the driver
5730  * that it should release a PCI device.  The could be caused by a
5731  * Hot-Plug event, or because the driver is going to be removed from
5732  * memory.
5733  **/
5734 static void __devexit ixgbe_remove(struct pci_dev *pdev)
5735 {
5736         struct net_device *netdev = pci_get_drvdata(pdev);
5737         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5738         int err;
5739
5740         set_bit(__IXGBE_DOWN, &adapter->state);
5741         /* clear the module not found bit to make sure the worker won't
5742          * reschedule
5743          */
5744         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5745         del_timer_sync(&adapter->watchdog_timer);
5746
5747         del_timer_sync(&adapter->sfp_timer);
5748         cancel_work_sync(&adapter->watchdog_task);
5749         cancel_work_sync(&adapter->sfp_task);
5750         cancel_work_sync(&adapter->multispeed_fiber_task);
5751         cancel_work_sync(&adapter->sfp_config_module_task);
5752         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5753             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5754                 cancel_work_sync(&adapter->fdir_reinit_task);
5755         flush_scheduled_work();
5756
5757 #ifdef CONFIG_IXGBE_DCA
5758         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
5759                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
5760                 dca_remove_requester(&pdev->dev);
5761                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
5762         }
5763
5764 #endif
5765 #ifdef IXGBE_FCOE
5766         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
5767                 ixgbe_cleanup_fcoe(adapter);
5768
5769 #endif /* IXGBE_FCOE */
5770
5771         /* remove the added san mac */
5772         ixgbe_del_sanmac_netdev(netdev);
5773
5774         if (netdev->reg_state == NETREG_REGISTERED)
5775                 unregister_netdev(netdev);
5776
5777         ixgbe_clear_interrupt_scheme(adapter);
5778
5779         ixgbe_release_hw_control(adapter);
5780
5781         iounmap(adapter->hw.hw_addr);
5782         pci_release_selected_regions(pdev, pci_select_bars(pdev,
5783                                      IORESOURCE_MEM));
5784
5785         DPRINTK(PROBE, INFO, "complete\n");
5786
5787         free_netdev(netdev);
5788
5789         err = pci_disable_pcie_error_reporting(pdev);
5790         if (err)
5791                 dev_err(&pdev->dev,
5792                         "pci_disable_pcie_error_reporting failed 0x%x\n", err);
5793
5794         pci_disable_device(pdev);
5795 }
5796
5797 /**
5798  * ixgbe_io_error_detected - called when PCI error is detected
5799  * @pdev: Pointer to PCI device
5800  * @state: The current pci connection state
5801  *
5802  * This function is called after a PCI bus error affecting
5803  * this device has been detected.
5804  */
5805 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
5806                                                 pci_channel_state_t state)
5807 {
5808         struct net_device *netdev = pci_get_drvdata(pdev);
5809         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5810
5811         netif_device_detach(netdev);
5812
5813         if (state == pci_channel_io_perm_failure)
5814                 return PCI_ERS_RESULT_DISCONNECT;
5815
5816         if (netif_running(netdev))
5817                 ixgbe_down(adapter);
5818         pci_disable_device(pdev);
5819
5820         /* Request a slot reset. */
5821         return PCI_ERS_RESULT_NEED_RESET;
5822 }
5823
5824 /**
5825  * ixgbe_io_slot_reset - called after the pci bus has been reset.
5826  * @pdev: Pointer to PCI device
5827  *
5828  * Restart the card from scratch, as if from a cold-boot.
5829  */
5830 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
5831 {
5832         struct net_device *netdev = pci_get_drvdata(pdev);
5833         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5834         pci_ers_result_t result;
5835         int err;
5836
5837         if (pci_enable_device_mem(pdev)) {
5838                 DPRINTK(PROBE, ERR,
5839                         "Cannot re-enable PCI device after reset.\n");
5840                 result = PCI_ERS_RESULT_DISCONNECT;
5841         } else {
5842                 pci_set_master(pdev);
5843                 pci_restore_state(pdev);
5844
5845                 pci_wake_from_d3(pdev, false);
5846
5847                 ixgbe_reset(adapter);
5848                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5849                 result = PCI_ERS_RESULT_RECOVERED;
5850         }
5851
5852         err = pci_cleanup_aer_uncorrect_error_status(pdev);
5853         if (err) {
5854                 dev_err(&pdev->dev,
5855                   "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
5856                 /* non-fatal, continue */
5857         }
5858
5859         return result;
5860 }
5861
5862 /**
5863  * ixgbe_io_resume - called when traffic can start flowing again.
5864  * @pdev: Pointer to PCI device
5865  *
5866  * This callback is called when the error recovery driver tells us that
5867  * its OK to resume normal operation.
5868  */
5869 static void ixgbe_io_resume(struct pci_dev *pdev)
5870 {
5871         struct net_device *netdev = pci_get_drvdata(pdev);
5872         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5873
5874         if (netif_running(netdev)) {
5875                 if (ixgbe_up(adapter)) {
5876                         DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
5877                         return;
5878                 }
5879         }
5880
5881         netif_device_attach(netdev);
5882 }
5883
5884 static struct pci_error_handlers ixgbe_err_handler = {
5885         .error_detected = ixgbe_io_error_detected,
5886         .slot_reset = ixgbe_io_slot_reset,
5887         .resume = ixgbe_io_resume,
5888 };
5889
5890 static struct pci_driver ixgbe_driver = {
5891         .name     = ixgbe_driver_name,
5892         .id_table = ixgbe_pci_tbl,
5893         .probe    = ixgbe_probe,
5894         .remove   = __devexit_p(ixgbe_remove),
5895 #ifdef CONFIG_PM
5896         .suspend  = ixgbe_suspend,
5897         .resume   = ixgbe_resume,
5898 #endif
5899         .shutdown = ixgbe_shutdown,
5900         .err_handler = &ixgbe_err_handler
5901 };
5902
5903 /**
5904  * ixgbe_init_module - Driver Registration Routine
5905  *
5906  * ixgbe_init_module is the first routine called when the driver is
5907  * loaded. All it does is register with the PCI subsystem.
5908  **/
5909 static int __init ixgbe_init_module(void)
5910 {
5911         int ret;
5912         printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
5913                ixgbe_driver_string, ixgbe_driver_version);
5914
5915         printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
5916
5917 #ifdef CONFIG_IXGBE_DCA
5918         dca_register_notify(&dca_notifier);
5919 #endif
5920
5921         ret = pci_register_driver(&ixgbe_driver);
5922         return ret;
5923 }
5924
5925 module_init(ixgbe_init_module);
5926
5927 /**
5928  * ixgbe_exit_module - Driver Exit Cleanup Routine
5929  *
5930  * ixgbe_exit_module is called just before the driver is removed
5931  * from memory.
5932  **/
5933 static void __exit ixgbe_exit_module(void)
5934 {
5935 #ifdef CONFIG_IXGBE_DCA
5936         dca_unregister_notify(&dca_notifier);
5937 #endif
5938         pci_unregister_driver(&ixgbe_driver);
5939 }
5940
5941 #ifdef CONFIG_IXGBE_DCA
5942 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
5943                             void *p)
5944 {
5945         int ret_val;
5946
5947         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
5948                                          __ixgbe_notify_dca);
5949
5950         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5951 }
5952
5953 #endif /* CONFIG_IXGBE_DCA */
5954 #ifdef DEBUG
5955 /**
5956  * ixgbe_get_hw_dev_name - return device name string
5957  * used by hardware layer to print debugging information
5958  **/
5959 char *ixgbe_get_hw_dev_name(struct ixgbe_hw *hw)
5960 {
5961         struct ixgbe_adapter *adapter = hw->back;
5962         return adapter->netdev->name;
5963 }
5964
5965 #endif
5966 module_exit(ixgbe_exit_module);
5967
5968 /* ixgbe_main.c */