ixgbe: fix for 82599 errata marking UDP checksum errors
[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/pkt_sched.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/ethtool.h>
42 #include <linux/if_vlan.h>
43 #include <scsi/fc/fc_fcoe.h>
44
45 #include "ixgbe.h"
46 #include "ixgbe_common.h"
47
48 char ixgbe_driver_name[] = "ixgbe";
49 static const char ixgbe_driver_string[] =
50                               "Intel(R) 10 Gigabit PCI Express Network Driver";
51
52 #define DRV_VERSION "2.0.34-k2"
53 const char ixgbe_driver_version[] = DRV_VERSION;
54 static char ixgbe_copyright[] = "Copyright (c) 1999-2009 Intel Corporation.";
55
56 static const struct ixgbe_info *ixgbe_info_tbl[] = {
57         [board_82598] = &ixgbe_82598_info,
58         [board_82599] = &ixgbe_82599_info,
59 };
60
61 /* ixgbe_pci_tbl - PCI Device ID Table
62  *
63  * Wildcard entries (PCI_ANY_ID) should come last
64  * Last entry must be all 0s
65  *
66  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
67  *   Class, Class Mask, private data (not used) }
68  */
69 static struct pci_device_id ixgbe_pci_tbl[] = {
70         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
71          board_82598 },
72         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
73          board_82598 },
74         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
75          board_82598 },
76         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
77          board_82598 },
78         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
79          board_82598 },
80         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
81          board_82598 },
82         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
83          board_82598 },
84         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
85          board_82598 },
86         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
87          board_82598 },
88         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
89          board_82598 },
90         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
91          board_82598 },
92         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
93          board_82599 },
94         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
95          board_82599 },
96         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
97          board_82599 },
98
99         /* required last entry */
100         {0, }
101 };
102 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
103
104 #ifdef CONFIG_IXGBE_DCA
105 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
106                             void *p);
107 static struct notifier_block dca_notifier = {
108         .notifier_call = ixgbe_notify_dca,
109         .next          = NULL,
110         .priority      = 0
111 };
112 #endif
113
114 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
115 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
116 MODULE_LICENSE("GPL");
117 MODULE_VERSION(DRV_VERSION);
118
119 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
120
121 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
122 {
123         u32 ctrl_ext;
124
125         /* Let firmware take over control of h/w */
126         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
127         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
128                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
129 }
130
131 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
132 {
133         u32 ctrl_ext;
134
135         /* Let firmware know the driver has taken over */
136         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
137         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
138                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
139 }
140
141 /*
142  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
143  * @adapter: pointer to adapter struct
144  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
145  * @queue: queue to map the corresponding interrupt to
146  * @msix_vector: the vector to map to the corresponding queue
147  *
148  */
149 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
150                            u8 queue, u8 msix_vector)
151 {
152         u32 ivar, index;
153         struct ixgbe_hw *hw = &adapter->hw;
154         switch (hw->mac.type) {
155         case ixgbe_mac_82598EB:
156                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
157                 if (direction == -1)
158                         direction = 0;
159                 index = (((direction * 64) + queue) >> 2) & 0x1F;
160                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
161                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
162                 ivar |= (msix_vector << (8 * (queue & 0x3)));
163                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
164                 break;
165         case ixgbe_mac_82599EB:
166                 if (direction == -1) {
167                         /* other causes */
168                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
169                         index = ((queue & 1) * 8);
170                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
171                         ivar &= ~(0xFF << index);
172                         ivar |= (msix_vector << index);
173                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
174                         break;
175                 } else {
176                         /* tx or rx causes */
177                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
178                         index = ((16 * (queue & 1)) + (8 * direction));
179                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
180                         ivar &= ~(0xFF << index);
181                         ivar |= (msix_vector << index);
182                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
183                         break;
184                 }
185         default:
186                 break;
187         }
188 }
189
190 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
191                                           u64 qmask)
192 {
193         u32 mask;
194
195         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
196                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
197                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
198         } else {
199                 mask = (qmask & 0xFFFFFFFF);
200                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
201                 mask = (qmask >> 32);
202                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
203         }
204 }
205
206 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
207                                              struct ixgbe_tx_buffer
208                                              *tx_buffer_info)
209 {
210         tx_buffer_info->dma = 0;
211         if (tx_buffer_info->skb) {
212                 skb_dma_unmap(&adapter->pdev->dev, tx_buffer_info->skb,
213                               DMA_TO_DEVICE);
214                 dev_kfree_skb_any(tx_buffer_info->skb);
215                 tx_buffer_info->skb = NULL;
216         }
217         tx_buffer_info->time_stamp = 0;
218         /* tx_buffer_info must be completely set up in the transmit path */
219 }
220
221 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
222                                        struct ixgbe_ring *tx_ring,
223                                        unsigned int eop)
224 {
225         struct ixgbe_hw *hw = &adapter->hw;
226
227         /* Detect a transmit hang in hardware, this serializes the
228          * check with the clearing of time_stamp and movement of eop */
229         adapter->detect_tx_hung = false;
230         if (tx_ring->tx_buffer_info[eop].time_stamp &&
231             time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
232             !(IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & IXGBE_TFCS_TXOFF)) {
233                 /* detected Tx unit hang */
234                 union ixgbe_adv_tx_desc *tx_desc;
235                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
236                 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
237                         "  Tx Queue             <%d>\n"
238                         "  TDH, TDT             <%x>, <%x>\n"
239                         "  next_to_use          <%x>\n"
240                         "  next_to_clean        <%x>\n"
241                         "tx_buffer_info[next_to_clean]\n"
242                         "  time_stamp           <%lx>\n"
243                         "  jiffies              <%lx>\n",
244                         tx_ring->queue_index,
245                         IXGBE_READ_REG(hw, tx_ring->head),
246                         IXGBE_READ_REG(hw, tx_ring->tail),
247                         tx_ring->next_to_use, eop,
248                         tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
249                 return true;
250         }
251
252         return false;
253 }
254
255 #define IXGBE_MAX_TXD_PWR       14
256 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
257
258 /* Tx Descriptors needed, worst case */
259 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
260                          (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
261 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
262         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
263
264 static void ixgbe_tx_timeout(struct net_device *netdev);
265
266 /**
267  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
268  * @q_vector: structure containing interrupt and ring information
269  * @tx_ring: tx ring to clean
270  **/
271 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
272                                struct ixgbe_ring *tx_ring)
273 {
274         struct ixgbe_adapter *adapter = q_vector->adapter;
275         struct net_device *netdev = adapter->netdev;
276         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
277         struct ixgbe_tx_buffer *tx_buffer_info;
278         unsigned int i, eop, count = 0;
279         unsigned int total_bytes = 0, total_packets = 0;
280
281         i = tx_ring->next_to_clean;
282         eop = tx_ring->tx_buffer_info[i].next_to_watch;
283         eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
284
285         while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
286                (count < tx_ring->work_limit)) {
287                 bool cleaned = false;
288                 for ( ; !cleaned; count++) {
289                         struct sk_buff *skb;
290                         tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
291                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
292                         cleaned = (i == eop);
293                         skb = tx_buffer_info->skb;
294
295                         if (cleaned && skb) {
296                                 unsigned int segs, bytecount;
297                                 unsigned int hlen = skb_headlen(skb);
298
299                                 /* gso_segs is currently only valid for tcp */
300                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
301 #ifdef IXGBE_FCOE
302                                 /* adjust for FCoE Sequence Offload */
303                                 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
304                                     && (skb->protocol == htons(ETH_P_FCOE)) &&
305                                     skb_is_gso(skb)) {
306                                         hlen = skb_transport_offset(skb) +
307                                                 sizeof(struct fc_frame_header) +
308                                                 sizeof(struct fcoe_crc_eof);
309                                         segs = DIV_ROUND_UP(skb->len - hlen,
310                                                 skb_shinfo(skb)->gso_size);
311                                 }
312 #endif /* IXGBE_FCOE */
313                                 /* multiply data chunks by size of headers */
314                                 bytecount = ((segs - 1) * hlen) + skb->len;
315                                 total_packets += segs;
316                                 total_bytes += bytecount;
317                         }
318
319                         ixgbe_unmap_and_free_tx_resource(adapter,
320                                                          tx_buffer_info);
321
322                         tx_desc->wb.status = 0;
323
324                         i++;
325                         if (i == tx_ring->count)
326                                 i = 0;
327                 }
328
329                 eop = tx_ring->tx_buffer_info[i].next_to_watch;
330                 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
331         }
332
333         tx_ring->next_to_clean = i;
334
335 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
336         if (unlikely(count && netif_carrier_ok(netdev) &&
337                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
338                 /* Make sure that anybody stopping the queue after this
339                  * sees the new next_to_clean.
340                  */
341                 smp_mb();
342                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
343                     !test_bit(__IXGBE_DOWN, &adapter->state)) {
344                         netif_wake_subqueue(netdev, tx_ring->queue_index);
345                         ++adapter->restart_queue;
346                 }
347         }
348
349         if (adapter->detect_tx_hung) {
350                 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
351                         /* schedule immediate reset if we believe we hung */
352                         DPRINTK(PROBE, INFO,
353                                 "tx hang %d detected, resetting adapter\n",
354                                 adapter->tx_timeout_count + 1);
355                         ixgbe_tx_timeout(adapter->netdev);
356                 }
357         }
358
359         /* re-arm the interrupt */
360         if (count >= tx_ring->work_limit)
361                 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
362
363         tx_ring->total_bytes += total_bytes;
364         tx_ring->total_packets += total_packets;
365         tx_ring->stats.packets += total_packets;
366         tx_ring->stats.bytes += total_bytes;
367         adapter->net_stats.tx_bytes += total_bytes;
368         adapter->net_stats.tx_packets += total_packets;
369         return (count < tx_ring->work_limit);
370 }
371
372 #ifdef CONFIG_IXGBE_DCA
373 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
374                                 struct ixgbe_ring *rx_ring)
375 {
376         u32 rxctrl;
377         int cpu = get_cpu();
378         int q = rx_ring - adapter->rx_ring;
379
380         if (rx_ring->cpu != cpu) {
381                 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
382                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
383                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
384                         rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
385                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
386                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
387                         rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
388                                    IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
389                 }
390                 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
391                 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
392                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
393                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
394                             IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
395                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
396                 rx_ring->cpu = cpu;
397         }
398         put_cpu();
399 }
400
401 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
402                                 struct ixgbe_ring *tx_ring)
403 {
404         u32 txctrl;
405         int cpu = get_cpu();
406         int q = tx_ring - adapter->tx_ring;
407
408         if (tx_ring->cpu != cpu) {
409                 txctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q));
410                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
411                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
412                         txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
413                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
414                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
415                         txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
416                                    IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
417                 }
418                 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
419                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q), txctrl);
420                 tx_ring->cpu = cpu;
421         }
422         put_cpu();
423 }
424
425 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
426 {
427         int i;
428
429         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
430                 return;
431
432         /* always use CB2 mode, difference is masked in the CB driver */
433         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
434
435         for (i = 0; i < adapter->num_tx_queues; i++) {
436                 adapter->tx_ring[i].cpu = -1;
437                 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
438         }
439         for (i = 0; i < adapter->num_rx_queues; i++) {
440                 adapter->rx_ring[i].cpu = -1;
441                 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
442         }
443 }
444
445 static int __ixgbe_notify_dca(struct device *dev, void *data)
446 {
447         struct net_device *netdev = dev_get_drvdata(dev);
448         struct ixgbe_adapter *adapter = netdev_priv(netdev);
449         unsigned long event = *(unsigned long *)data;
450
451         switch (event) {
452         case DCA_PROVIDER_ADD:
453                 /* if we're already enabled, don't do it again */
454                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
455                         break;
456                 if (dca_add_requester(dev) == 0) {
457                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
458                         ixgbe_setup_dca(adapter);
459                         break;
460                 }
461                 /* Fall Through since DCA is disabled. */
462         case DCA_PROVIDER_REMOVE:
463                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
464                         dca_remove_requester(dev);
465                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
466                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
467                 }
468                 break;
469         }
470
471         return 0;
472 }
473
474 #endif /* CONFIG_IXGBE_DCA */
475 /**
476  * ixgbe_receive_skb - Send a completed packet up the stack
477  * @adapter: board private structure
478  * @skb: packet to send up
479  * @status: hardware indication of status of receive
480  * @rx_ring: rx descriptor ring (for a specific queue) to setup
481  * @rx_desc: rx descriptor
482  **/
483 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
484                               struct sk_buff *skb, u8 status,
485                               struct ixgbe_ring *ring,
486                               union ixgbe_adv_rx_desc *rx_desc)
487 {
488         struct ixgbe_adapter *adapter = q_vector->adapter;
489         struct napi_struct *napi = &q_vector->napi;
490         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
491         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
492
493         skb_record_rx_queue(skb, ring->queue_index);
494         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
495                 if (adapter->vlgrp && is_vlan && (tag != 0))
496                         vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
497                 else
498                         napi_gro_receive(napi, skb);
499         } else {
500                 if (adapter->vlgrp && is_vlan && (tag != 0))
501                         vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
502                 else
503                         netif_rx(skb);
504         }
505 }
506
507 /**
508  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
509  * @adapter: address of board private structure
510  * @status_err: hardware indication of status of receive
511  * @skb: skb currently being received and modified
512  **/
513 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
514                                      union ixgbe_adv_rx_desc *rx_desc,
515                                      struct sk_buff *skb)
516 {
517         u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
518
519         skb->ip_summed = CHECKSUM_NONE;
520
521         /* Rx csum disabled */
522         if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
523                 return;
524
525         /* if IP and error */
526         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
527             (status_err & IXGBE_RXDADV_ERR_IPE)) {
528                 adapter->hw_csum_rx_error++;
529                 return;
530         }
531
532         if (!(status_err & IXGBE_RXD_STAT_L4CS))
533                 return;
534
535         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
536                 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
537
538                 /*
539                  * 82599 errata, UDP frames with a 0 checksum can be marked as
540                  * checksum errors.
541                  */
542                 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
543                     (adapter->hw.mac.type == ixgbe_mac_82599EB))
544                         return;
545
546                 adapter->hw_csum_rx_error++;
547                 return;
548         }
549
550         /* It must be a TCP or UDP packet with a valid checksum */
551         skb->ip_summed = CHECKSUM_UNNECESSARY;
552         adapter->hw_csum_rx_good++;
553 }
554
555 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
556                                          struct ixgbe_ring *rx_ring, u32 val)
557 {
558         /*
559          * Force memory writes to complete before letting h/w
560          * know there are new descriptors to fetch.  (Only
561          * applicable for weak-ordered memory model archs,
562          * such as IA-64).
563          */
564         wmb();
565         IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
566 }
567
568 /**
569  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
570  * @adapter: address of board private structure
571  **/
572 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
573                                    struct ixgbe_ring *rx_ring,
574                                    int cleaned_count)
575 {
576         struct pci_dev *pdev = adapter->pdev;
577         union ixgbe_adv_rx_desc *rx_desc;
578         struct ixgbe_rx_buffer *bi;
579         unsigned int i;
580
581         i = rx_ring->next_to_use;
582         bi = &rx_ring->rx_buffer_info[i];
583
584         while (cleaned_count--) {
585                 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
586
587                 if (!bi->page_dma &&
588                     (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
589                         if (!bi->page) {
590                                 bi->page = alloc_page(GFP_ATOMIC);
591                                 if (!bi->page) {
592                                         adapter->alloc_rx_page_failed++;
593                                         goto no_buffers;
594                                 }
595                                 bi->page_offset = 0;
596                         } else {
597                                 /* use a half page if we're re-using */
598                                 bi->page_offset ^= (PAGE_SIZE / 2);
599                         }
600
601                         bi->page_dma = pci_map_page(pdev, bi->page,
602                                                     bi->page_offset,
603                                                     (PAGE_SIZE / 2),
604                                                     PCI_DMA_FROMDEVICE);
605                 }
606
607                 if (!bi->skb) {
608                         struct sk_buff *skb;
609                         skb = netdev_alloc_skb(adapter->netdev,
610                                                (rx_ring->rx_buf_len +
611                                                 NET_IP_ALIGN));
612
613                         if (!skb) {
614                                 adapter->alloc_rx_buff_failed++;
615                                 goto no_buffers;
616                         }
617
618                         /*
619                          * Make buffer alignment 2 beyond a 16 byte boundary
620                          * this will result in a 16 byte aligned IP header after
621                          * the 14 byte MAC header is removed
622                          */
623                         skb_reserve(skb, NET_IP_ALIGN);
624
625                         bi->skb = skb;
626                         bi->dma = pci_map_single(pdev, skb->data,
627                                                  rx_ring->rx_buf_len,
628                                                  PCI_DMA_FROMDEVICE);
629                 }
630                 /* Refresh the desc even if buffer_addrs didn't change because
631                  * each write-back erases this info. */
632                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
633                         rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
634                         rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
635                 } else {
636                         rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
637                 }
638
639                 i++;
640                 if (i == rx_ring->count)
641                         i = 0;
642                 bi = &rx_ring->rx_buffer_info[i];
643         }
644
645 no_buffers:
646         if (rx_ring->next_to_use != i) {
647                 rx_ring->next_to_use = i;
648                 if (i-- == 0)
649                         i = (rx_ring->count - 1);
650
651                 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
652         }
653 }
654
655 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
656 {
657         return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
658 }
659
660 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
661 {
662         return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
663 }
664
665 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
666 {
667         return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
668                 IXGBE_RXDADV_RSCCNT_MASK) >>
669                 IXGBE_RXDADV_RSCCNT_SHIFT;
670 }
671
672 /**
673  * ixgbe_transform_rsc_queue - change rsc queue into a full packet
674  * @skb: pointer to the last skb in the rsc queue
675  *
676  * This function changes a queue full of hw rsc buffers into a completed
677  * packet.  It uses the ->prev pointers to find the first packet and then
678  * turns it into the frag list owner.
679  **/
680 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
681 {
682         unsigned int frag_list_size = 0;
683
684         while (skb->prev) {
685                 struct sk_buff *prev = skb->prev;
686                 frag_list_size += skb->len;
687                 skb->prev = NULL;
688                 skb = prev;
689         }
690
691         skb_shinfo(skb)->frag_list = skb->next;
692         skb->next = NULL;
693         skb->len += frag_list_size;
694         skb->data_len += frag_list_size;
695         skb->truesize += frag_list_size;
696         return skb;
697 }
698
699 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
700                                struct ixgbe_ring *rx_ring,
701                                int *work_done, int work_to_do)
702 {
703         struct ixgbe_adapter *adapter = q_vector->adapter;
704         struct pci_dev *pdev = adapter->pdev;
705         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
706         struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
707         struct sk_buff *skb;
708         unsigned int i, rsc_count = 0;
709         u32 len, staterr;
710         u16 hdr_info;
711         bool cleaned = false;
712         int cleaned_count = 0;
713         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
714 #ifdef IXGBE_FCOE
715         int ddp_bytes = 0;
716 #endif /* IXGBE_FCOE */
717
718         i = rx_ring->next_to_clean;
719         rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
720         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
721         rx_buffer_info = &rx_ring->rx_buffer_info[i];
722
723         while (staterr & IXGBE_RXD_STAT_DD) {
724                 u32 upper_len = 0;
725                 if (*work_done >= work_to_do)
726                         break;
727                 (*work_done)++;
728
729                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
730                         hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
731                         len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
732                                IXGBE_RXDADV_HDRBUFLEN_SHIFT;
733                         if (hdr_info & IXGBE_RXDADV_SPH)
734                                 adapter->rx_hdr_split++;
735                         if (len > IXGBE_RX_HDR_SIZE)
736                                 len = IXGBE_RX_HDR_SIZE;
737                         upper_len = le16_to_cpu(rx_desc->wb.upper.length);
738                 } else {
739                         len = le16_to_cpu(rx_desc->wb.upper.length);
740                 }
741
742                 cleaned = true;
743                 skb = rx_buffer_info->skb;
744                 prefetch(skb->data - NET_IP_ALIGN);
745                 rx_buffer_info->skb = NULL;
746
747                 if (rx_buffer_info->dma) {
748                         pci_unmap_single(pdev, rx_buffer_info->dma,
749                                          rx_ring->rx_buf_len,
750                                          PCI_DMA_FROMDEVICE);
751                         rx_buffer_info->dma = 0;
752                         skb_put(skb, len);
753                 }
754
755                 if (upper_len) {
756                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
757                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
758                         rx_buffer_info->page_dma = 0;
759                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
760                                            rx_buffer_info->page,
761                                            rx_buffer_info->page_offset,
762                                            upper_len);
763
764                         if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
765                             (page_count(rx_buffer_info->page) != 1))
766                                 rx_buffer_info->page = NULL;
767                         else
768                                 get_page(rx_buffer_info->page);
769
770                         skb->len += upper_len;
771                         skb->data_len += upper_len;
772                         skb->truesize += upper_len;
773                 }
774
775                 i++;
776                 if (i == rx_ring->count)
777                         i = 0;
778
779                 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
780                 prefetch(next_rxd);
781                 cleaned_count++;
782
783                 if (adapter->flags & IXGBE_FLAG2_RSC_CAPABLE)
784                         rsc_count = ixgbe_get_rsc_count(rx_desc);
785
786                 if (rsc_count) {
787                         u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
788                                      IXGBE_RXDADV_NEXTP_SHIFT;
789                         next_buffer = &rx_ring->rx_buffer_info[nextp];
790                         rx_ring->rsc_count += (rsc_count - 1);
791                 } else {
792                         next_buffer = &rx_ring->rx_buffer_info[i];
793                 }
794
795                 if (staterr & IXGBE_RXD_STAT_EOP) {
796                         if (skb->prev)
797                                 skb = ixgbe_transform_rsc_queue(skb);
798                         rx_ring->stats.packets++;
799                         rx_ring->stats.bytes += skb->len;
800                 } else {
801                         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
802                                 rx_buffer_info->skb = next_buffer->skb;
803                                 rx_buffer_info->dma = next_buffer->dma;
804                                 next_buffer->skb = skb;
805                                 next_buffer->dma = 0;
806                         } else {
807                                 skb->next = next_buffer->skb;
808                                 skb->next->prev = skb;
809                         }
810                         adapter->non_eop_descs++;
811                         goto next_desc;
812                 }
813
814                 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
815                         dev_kfree_skb_irq(skb);
816                         goto next_desc;
817                 }
818
819                 ixgbe_rx_checksum(adapter, rx_desc, skb);
820
821                 /* probably a little skewed due to removing CRC */
822                 total_rx_bytes += skb->len;
823                 total_rx_packets++;
824
825                 skb->protocol = eth_type_trans(skb, adapter->netdev);
826 #ifdef IXGBE_FCOE
827                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
828                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
829                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
830                         if (!ddp_bytes)
831                                 goto next_desc;
832                 }
833 #endif /* IXGBE_FCOE */
834                 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
835
836 next_desc:
837                 rx_desc->wb.upper.status_error = 0;
838
839                 /* return some buffers to hardware, one at a time is too slow */
840                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
841                         ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
842                         cleaned_count = 0;
843                 }
844
845                 /* use prefetched values */
846                 rx_desc = next_rxd;
847                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
848
849                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
850         }
851
852         rx_ring->next_to_clean = i;
853         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
854
855         if (cleaned_count)
856                 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
857
858 #ifdef IXGBE_FCOE
859         /* include DDPed FCoE data */
860         if (ddp_bytes > 0) {
861                 unsigned int mss;
862
863                 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
864                         sizeof(struct fc_frame_header) -
865                         sizeof(struct fcoe_crc_eof);
866                 if (mss > 512)
867                         mss &= ~511;
868                 total_rx_bytes += ddp_bytes;
869                 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
870         }
871 #endif /* IXGBE_FCOE */
872
873         rx_ring->total_packets += total_rx_packets;
874         rx_ring->total_bytes += total_rx_bytes;
875         adapter->net_stats.rx_bytes += total_rx_bytes;
876         adapter->net_stats.rx_packets += total_rx_packets;
877
878         return cleaned;
879 }
880
881 static int ixgbe_clean_rxonly(struct napi_struct *, int);
882 /**
883  * ixgbe_configure_msix - Configure MSI-X hardware
884  * @adapter: board private structure
885  *
886  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
887  * interrupts.
888  **/
889 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
890 {
891         struct ixgbe_q_vector *q_vector;
892         int i, j, q_vectors, v_idx, r_idx;
893         u32 mask;
894
895         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
896
897         /*
898          * Populate the IVAR table and set the ITR values to the
899          * corresponding register.
900          */
901         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
902                 q_vector = adapter->q_vector[v_idx];
903                 /* XXX for_each_bit(...) */
904                 r_idx = find_first_bit(q_vector->rxr_idx,
905                                        adapter->num_rx_queues);
906
907                 for (i = 0; i < q_vector->rxr_count; i++) {
908                         j = adapter->rx_ring[r_idx].reg_idx;
909                         ixgbe_set_ivar(adapter, 0, j, v_idx);
910                         r_idx = find_next_bit(q_vector->rxr_idx,
911                                               adapter->num_rx_queues,
912                                               r_idx + 1);
913                 }
914                 r_idx = find_first_bit(q_vector->txr_idx,
915                                        adapter->num_tx_queues);
916
917                 for (i = 0; i < q_vector->txr_count; i++) {
918                         j = adapter->tx_ring[r_idx].reg_idx;
919                         ixgbe_set_ivar(adapter, 1, j, v_idx);
920                         r_idx = find_next_bit(q_vector->txr_idx,
921                                               adapter->num_tx_queues,
922                                               r_idx + 1);
923                 }
924
925                 /* if this is a tx only vector halve the interrupt rate */
926                 if (q_vector->txr_count && !q_vector->rxr_count)
927                         q_vector->eitr = (adapter->eitr_param >> 1);
928                 else if (q_vector->rxr_count)
929                         /* rx only */
930                         q_vector->eitr = adapter->eitr_param;
931
932                 ixgbe_write_eitr(q_vector);
933         }
934
935         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
936                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
937                                v_idx);
938         else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
939                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
940         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
941
942         /* set up to autoclear timer, and the vectors */
943         mask = IXGBE_EIMS_ENABLE_MASK;
944         mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
945         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
946 }
947
948 enum latency_range {
949         lowest_latency = 0,
950         low_latency = 1,
951         bulk_latency = 2,
952         latency_invalid = 255
953 };
954
955 /**
956  * ixgbe_update_itr - update the dynamic ITR value based on statistics
957  * @adapter: pointer to adapter
958  * @eitr: eitr setting (ints per sec) to give last timeslice
959  * @itr_setting: current throttle rate in ints/second
960  * @packets: the number of packets during this measurement interval
961  * @bytes: the number of bytes during this measurement interval
962  *
963  *      Stores a new ITR value based on packets and byte
964  *      counts during the last interrupt.  The advantage of per interrupt
965  *      computation is faster updates and more accurate ITR for the current
966  *      traffic pattern.  Constants in this function were computed
967  *      based on theoretical maximum wire speed and thresholds were set based
968  *      on testing data as well as attempting to minimize response time
969  *      while increasing bulk throughput.
970  *      this functionality is controlled by the InterruptThrottleRate module
971  *      parameter (see ixgbe_param.c)
972  **/
973 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
974                            u32 eitr, u8 itr_setting,
975                            int packets, int bytes)
976 {
977         unsigned int retval = itr_setting;
978         u32 timepassed_us;
979         u64 bytes_perint;
980
981         if (packets == 0)
982                 goto update_itr_done;
983
984
985         /* simple throttlerate management
986          *    0-20MB/s lowest (100000 ints/s)
987          *   20-100MB/s low   (20000 ints/s)
988          *  100-1249MB/s bulk (8000 ints/s)
989          */
990         /* what was last interrupt timeslice? */
991         timepassed_us = 1000000/eitr;
992         bytes_perint = bytes / timepassed_us; /* bytes/usec */
993
994         switch (itr_setting) {
995         case lowest_latency:
996                 if (bytes_perint > adapter->eitr_low)
997                         retval = low_latency;
998                 break;
999         case low_latency:
1000                 if (bytes_perint > adapter->eitr_high)
1001                         retval = bulk_latency;
1002                 else if (bytes_perint <= adapter->eitr_low)
1003                         retval = lowest_latency;
1004                 break;
1005         case bulk_latency:
1006                 if (bytes_perint <= adapter->eitr_high)
1007                         retval = low_latency;
1008                 break;
1009         }
1010
1011 update_itr_done:
1012         return retval;
1013 }
1014
1015 /**
1016  * ixgbe_write_eitr - write EITR register in hardware specific way
1017  * @q_vector: structure containing interrupt and ring information
1018  *
1019  * This function is made to be called by ethtool and by the driver
1020  * when it needs to update EITR registers at runtime.  Hardware
1021  * specific quirks/differences are taken care of here.
1022  */
1023 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1024 {
1025         struct ixgbe_adapter *adapter = q_vector->adapter;
1026         struct ixgbe_hw *hw = &adapter->hw;
1027         int v_idx = q_vector->v_idx;
1028         u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1029
1030         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1031                 /* must write high and low 16 bits to reset counter */
1032                 itr_reg |= (itr_reg << 16);
1033         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1034                 /*
1035                  * set the WDIS bit to not clear the timer bits and cause an
1036                  * immediate assertion of the interrupt
1037                  */
1038                 itr_reg |= IXGBE_EITR_CNT_WDIS;
1039         }
1040         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1041 }
1042
1043 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1044 {
1045         struct ixgbe_adapter *adapter = q_vector->adapter;
1046         u32 new_itr;
1047         u8 current_itr, ret_itr;
1048         int i, r_idx;
1049         struct ixgbe_ring *rx_ring, *tx_ring;
1050
1051         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1052         for (i = 0; i < q_vector->txr_count; i++) {
1053                 tx_ring = &(adapter->tx_ring[r_idx]);
1054                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1055                                            q_vector->tx_itr,
1056                                            tx_ring->total_packets,
1057                                            tx_ring->total_bytes);
1058                 /* if the result for this queue would decrease interrupt
1059                  * rate for this vector then use that result */
1060                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1061                                     q_vector->tx_itr - 1 : ret_itr);
1062                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1063                                       r_idx + 1);
1064         }
1065
1066         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1067         for (i = 0; i < q_vector->rxr_count; i++) {
1068                 rx_ring = &(adapter->rx_ring[r_idx]);
1069                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1070                                            q_vector->rx_itr,
1071                                            rx_ring->total_packets,
1072                                            rx_ring->total_bytes);
1073                 /* if the result for this queue would decrease interrupt
1074                  * rate for this vector then use that result */
1075                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1076                                     q_vector->rx_itr - 1 : ret_itr);
1077                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1078                                       r_idx + 1);
1079         }
1080
1081         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1082
1083         switch (current_itr) {
1084         /* counts and packets in update_itr are dependent on these numbers */
1085         case lowest_latency:
1086                 new_itr = 100000;
1087                 break;
1088         case low_latency:
1089                 new_itr = 20000; /* aka hwitr = ~200 */
1090                 break;
1091         case bulk_latency:
1092         default:
1093                 new_itr = 8000;
1094                 break;
1095         }
1096
1097         if (new_itr != q_vector->eitr) {
1098                 /* do an exponential smoothing */
1099                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1100
1101                 /* save the algorithm value here, not the smoothed one */
1102                 q_vector->eitr = new_itr;
1103
1104                 ixgbe_write_eitr(q_vector);
1105         }
1106
1107         return;
1108 }
1109
1110 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1111 {
1112         struct ixgbe_hw *hw = &adapter->hw;
1113
1114         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1115             (eicr & IXGBE_EICR_GPI_SDP1)) {
1116                 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
1117                 /* write to clear the interrupt */
1118                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1119         }
1120 }
1121
1122 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1123 {
1124         struct ixgbe_hw *hw = &adapter->hw;
1125
1126         if (eicr & IXGBE_EICR_GPI_SDP1) {
1127                 /* Clear the interrupt */
1128                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1129                 schedule_work(&adapter->multispeed_fiber_task);
1130         } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1131                 /* Clear the interrupt */
1132                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1133                 schedule_work(&adapter->sfp_config_module_task);
1134         } else {
1135                 /* Interrupt isn't for us... */
1136                 return;
1137         }
1138 }
1139
1140 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1141 {
1142         struct ixgbe_hw *hw = &adapter->hw;
1143
1144         adapter->lsc_int++;
1145         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1146         adapter->link_check_timeout = jiffies;
1147         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1148                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1149                 schedule_work(&adapter->watchdog_task);
1150         }
1151 }
1152
1153 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1154 {
1155         struct net_device *netdev = data;
1156         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1157         struct ixgbe_hw *hw = &adapter->hw;
1158         u32 eicr;
1159
1160         /*
1161          * Workaround for Silicon errata.  Use clear-by-write instead
1162          * of clear-by-read.  Reading with EICS will return the
1163          * interrupt causes without clearing, which later be done
1164          * with the write to EICR.
1165          */
1166         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1167         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1168
1169         if (eicr & IXGBE_EICR_LSC)
1170                 ixgbe_check_lsc(adapter);
1171
1172         if (hw->mac.type == ixgbe_mac_82598EB)
1173                 ixgbe_check_fan_failure(adapter, eicr);
1174
1175         if (hw->mac.type == ixgbe_mac_82599EB) {
1176                 ixgbe_check_sfp_event(adapter, eicr);
1177
1178                 /* Handle Flow Director Full threshold interrupt */
1179                 if (eicr & IXGBE_EICR_FLOW_DIR) {
1180                         int i;
1181                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1182                         /* Disable transmits before FDIR Re-initialization */
1183                         netif_tx_stop_all_queues(netdev);
1184                         for (i = 0; i < adapter->num_tx_queues; i++) {
1185                                 struct ixgbe_ring *tx_ring =
1186                                                            &adapter->tx_ring[i];
1187                                 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1188                                                        &tx_ring->reinit_state))
1189                                         schedule_work(&adapter->fdir_reinit_task);
1190                         }
1191                 }
1192         }
1193         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1194                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1195
1196         return IRQ_HANDLED;
1197 }
1198
1199 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1200                                            u64 qmask)
1201 {
1202         u32 mask;
1203
1204         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1205                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1206                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1207         } else {
1208                 mask = (qmask & 0xFFFFFFFF);
1209                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1210                 mask = (qmask >> 32);
1211                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1212         }
1213         /* skip the flush */
1214 }
1215
1216 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1217                                             u64 qmask)
1218 {
1219         u32 mask;
1220
1221         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1222                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1223                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1224         } else {
1225                 mask = (qmask & 0xFFFFFFFF);
1226                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1227                 mask = (qmask >> 32);
1228                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1229         }
1230         /* skip the flush */
1231 }
1232
1233 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1234 {
1235         struct ixgbe_q_vector *q_vector = data;
1236         struct ixgbe_adapter  *adapter = q_vector->adapter;
1237         struct ixgbe_ring     *tx_ring;
1238         int i, r_idx;
1239
1240         if (!q_vector->txr_count)
1241                 return IRQ_HANDLED;
1242
1243         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1244         for (i = 0; i < q_vector->txr_count; i++) {
1245                 tx_ring = &(adapter->tx_ring[r_idx]);
1246                 tx_ring->total_bytes = 0;
1247                 tx_ring->total_packets = 0;
1248                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1249                                       r_idx + 1);
1250         }
1251
1252         /* disable interrupts on this vector only */
1253         ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1254         napi_schedule(&q_vector->napi);
1255
1256         return IRQ_HANDLED;
1257 }
1258
1259 /**
1260  * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1261  * @irq: unused
1262  * @data: pointer to our q_vector struct for this interrupt vector
1263  **/
1264 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1265 {
1266         struct ixgbe_q_vector *q_vector = data;
1267         struct ixgbe_adapter  *adapter = q_vector->adapter;
1268         struct ixgbe_ring  *rx_ring;
1269         int r_idx;
1270         int i;
1271
1272         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1273         for (i = 0;  i < q_vector->rxr_count; i++) {
1274                 rx_ring = &(adapter->rx_ring[r_idx]);
1275                 rx_ring->total_bytes = 0;
1276                 rx_ring->total_packets = 0;
1277                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1278                                       r_idx + 1);
1279         }
1280
1281         if (!q_vector->rxr_count)
1282                 return IRQ_HANDLED;
1283
1284         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1285         rx_ring = &(adapter->rx_ring[r_idx]);
1286         /* disable interrupts on this vector only */
1287         ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1288         napi_schedule(&q_vector->napi);
1289
1290         return IRQ_HANDLED;
1291 }
1292
1293 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1294 {
1295         struct ixgbe_q_vector *q_vector = data;
1296         struct ixgbe_adapter  *adapter = q_vector->adapter;
1297         struct ixgbe_ring  *ring;
1298         int r_idx;
1299         int i;
1300
1301         if (!q_vector->txr_count && !q_vector->rxr_count)
1302                 return IRQ_HANDLED;
1303
1304         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1305         for (i = 0; i < q_vector->txr_count; i++) {
1306                 ring = &(adapter->tx_ring[r_idx]);
1307                 ring->total_bytes = 0;
1308                 ring->total_packets = 0;
1309                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1310                                       r_idx + 1);
1311         }
1312
1313         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1314         for (i = 0; i < q_vector->rxr_count; i++) {
1315                 ring = &(adapter->rx_ring[r_idx]);
1316                 ring->total_bytes = 0;
1317                 ring->total_packets = 0;
1318                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1319                                       r_idx + 1);
1320         }
1321
1322         /* disable interrupts on this vector only */
1323         ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1324         napi_schedule(&q_vector->napi);
1325
1326         return IRQ_HANDLED;
1327 }
1328
1329 /**
1330  * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1331  * @napi: napi struct with our devices info in it
1332  * @budget: amount of work driver is allowed to do this pass, in packets
1333  *
1334  * This function is optimized for cleaning one queue only on a single
1335  * q_vector!!!
1336  **/
1337 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1338 {
1339         struct ixgbe_q_vector *q_vector =
1340                                container_of(napi, struct ixgbe_q_vector, napi);
1341         struct ixgbe_adapter *adapter = q_vector->adapter;
1342         struct ixgbe_ring *rx_ring = NULL;
1343         int work_done = 0;
1344         long r_idx;
1345
1346         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1347         rx_ring = &(adapter->rx_ring[r_idx]);
1348 #ifdef CONFIG_IXGBE_DCA
1349         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1350                 ixgbe_update_rx_dca(adapter, rx_ring);
1351 #endif
1352
1353         ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1354
1355         /* If all Rx work done, exit the polling mode */
1356         if (work_done < budget) {
1357                 napi_complete(napi);
1358                 if (adapter->itr_setting & 1)
1359                         ixgbe_set_itr_msix(q_vector);
1360                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1361                         ixgbe_irq_enable_queues(adapter,
1362                                                 ((u64)1 << q_vector->v_idx));
1363         }
1364
1365         return work_done;
1366 }
1367
1368 /**
1369  * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1370  * @napi: napi struct with our devices info in it
1371  * @budget: amount of work driver is allowed to do this pass, in packets
1372  *
1373  * This function will clean more than one rx queue associated with a
1374  * q_vector.
1375  **/
1376 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1377 {
1378         struct ixgbe_q_vector *q_vector =
1379                                container_of(napi, struct ixgbe_q_vector, napi);
1380         struct ixgbe_adapter *adapter = q_vector->adapter;
1381         struct ixgbe_ring *ring = NULL;
1382         int work_done = 0, i;
1383         long r_idx;
1384         bool tx_clean_complete = true;
1385
1386         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1387         for (i = 0; i < q_vector->txr_count; i++) {
1388                 ring = &(adapter->tx_ring[r_idx]);
1389 #ifdef CONFIG_IXGBE_DCA
1390                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1391                         ixgbe_update_tx_dca(adapter, ring);
1392 #endif
1393                 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1394                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1395                                       r_idx + 1);
1396         }
1397
1398         /* attempt to distribute budget to each queue fairly, but don't allow
1399          * the budget to go below 1 because we'll exit polling */
1400         budget /= (q_vector->rxr_count ?: 1);
1401         budget = max(budget, 1);
1402         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1403         for (i = 0; i < q_vector->rxr_count; i++) {
1404                 ring = &(adapter->rx_ring[r_idx]);
1405 #ifdef CONFIG_IXGBE_DCA
1406                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1407                         ixgbe_update_rx_dca(adapter, ring);
1408 #endif
1409                 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1410                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1411                                       r_idx + 1);
1412         }
1413
1414         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1415         ring = &(adapter->rx_ring[r_idx]);
1416         /* If all Rx work done, exit the polling mode */
1417         if (work_done < budget) {
1418                 napi_complete(napi);
1419                 if (adapter->itr_setting & 1)
1420                         ixgbe_set_itr_msix(q_vector);
1421                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1422                         ixgbe_irq_enable_queues(adapter,
1423                                                 ((u64)1 << q_vector->v_idx));
1424                 return 0;
1425         }
1426
1427         return work_done;
1428 }
1429
1430 /**
1431  * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1432  * @napi: napi struct with our devices info in it
1433  * @budget: amount of work driver is allowed to do this pass, in packets
1434  *
1435  * This function is optimized for cleaning one queue only on a single
1436  * q_vector!!!
1437  **/
1438 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1439 {
1440         struct ixgbe_q_vector *q_vector =
1441                                container_of(napi, struct ixgbe_q_vector, napi);
1442         struct ixgbe_adapter *adapter = q_vector->adapter;
1443         struct ixgbe_ring *tx_ring = NULL;
1444         int work_done = 0;
1445         long r_idx;
1446
1447         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1448         tx_ring = &(adapter->tx_ring[r_idx]);
1449 #ifdef CONFIG_IXGBE_DCA
1450         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1451                 ixgbe_update_tx_dca(adapter, tx_ring);
1452 #endif
1453
1454         if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
1455                 work_done = budget;
1456
1457         /* If all Rx work done, exit the polling mode */
1458         if (work_done < budget) {
1459                 napi_complete(napi);
1460                 if (adapter->itr_setting & 1)
1461                         ixgbe_set_itr_msix(q_vector);
1462                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1463                         ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
1464         }
1465
1466         return work_done;
1467 }
1468
1469 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1470                                      int r_idx)
1471 {
1472         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1473
1474         set_bit(r_idx, q_vector->rxr_idx);
1475         q_vector->rxr_count++;
1476 }
1477
1478 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1479                                      int t_idx)
1480 {
1481         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1482
1483         set_bit(t_idx, q_vector->txr_idx);
1484         q_vector->txr_count++;
1485 }
1486
1487 /**
1488  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1489  * @adapter: board private structure to initialize
1490  * @vectors: allotted vector count for descriptor rings
1491  *
1492  * This function maps descriptor rings to the queue-specific vectors
1493  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1494  * one vector per ring/queue, but on a constrained vector budget, we
1495  * group the rings as "efficiently" as possible.  You would add new
1496  * mapping configurations in here.
1497  **/
1498 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1499                                       int vectors)
1500 {
1501         int v_start = 0;
1502         int rxr_idx = 0, txr_idx = 0;
1503         int rxr_remaining = adapter->num_rx_queues;
1504         int txr_remaining = adapter->num_tx_queues;
1505         int i, j;
1506         int rqpv, tqpv;
1507         int err = 0;
1508
1509         /* No mapping required if MSI-X is disabled. */
1510         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1511                 goto out;
1512
1513         /*
1514          * The ideal configuration...
1515          * We have enough vectors to map one per queue.
1516          */
1517         if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1518                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1519                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1520
1521                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1522                         map_vector_to_txq(adapter, v_start, txr_idx);
1523
1524                 goto out;
1525         }
1526
1527         /*
1528          * If we don't have enough vectors for a 1-to-1
1529          * mapping, we'll have to group them so there are
1530          * multiple queues per vector.
1531          */
1532         /* Re-adjusting *qpv takes care of the remainder. */
1533         for (i = v_start; i < vectors; i++) {
1534                 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1535                 for (j = 0; j < rqpv; j++) {
1536                         map_vector_to_rxq(adapter, i, rxr_idx);
1537                         rxr_idx++;
1538                         rxr_remaining--;
1539                 }
1540         }
1541         for (i = v_start; i < vectors; i++) {
1542                 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1543                 for (j = 0; j < tqpv; j++) {
1544                         map_vector_to_txq(adapter, i, txr_idx);
1545                         txr_idx++;
1546                         txr_remaining--;
1547                 }
1548         }
1549
1550 out:
1551         return err;
1552 }
1553
1554 /**
1555  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1556  * @adapter: board private structure
1557  *
1558  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1559  * interrupts from the kernel.
1560  **/
1561 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1562 {
1563         struct net_device *netdev = adapter->netdev;
1564         irqreturn_t (*handler)(int, void *);
1565         int i, vector, q_vectors, err;
1566         int ri=0, ti=0;
1567
1568         /* Decrement for Other and TCP Timer vectors */
1569         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1570
1571         /* Map the Tx/Rx rings to the vectors we were allotted. */
1572         err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1573         if (err)
1574                 goto out;
1575
1576 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1577                          (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1578                          &ixgbe_msix_clean_many)
1579         for (vector = 0; vector < q_vectors; vector++) {
1580                 handler = SET_HANDLER(adapter->q_vector[vector]);
1581
1582                 if(handler == &ixgbe_msix_clean_rx) {
1583                         sprintf(adapter->name[vector], "%s-%s-%d",
1584                                 netdev->name, "rx", ri++);
1585                 }
1586                 else if(handler == &ixgbe_msix_clean_tx) {
1587                         sprintf(adapter->name[vector], "%s-%s-%d",
1588                                 netdev->name, "tx", ti++);
1589                 }
1590                 else
1591                         sprintf(adapter->name[vector], "%s-%s-%d",
1592                                 netdev->name, "TxRx", vector);
1593
1594                 err = request_irq(adapter->msix_entries[vector].vector,
1595                                   handler, 0, adapter->name[vector],
1596                                   adapter->q_vector[vector]);
1597                 if (err) {
1598                         DPRINTK(PROBE, ERR,
1599                                 "request_irq failed for MSIX interrupt "
1600                                 "Error: %d\n", err);
1601                         goto free_queue_irqs;
1602                 }
1603         }
1604
1605         sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1606         err = request_irq(adapter->msix_entries[vector].vector,
1607                           &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1608         if (err) {
1609                 DPRINTK(PROBE, ERR,
1610                         "request_irq for msix_lsc failed: %d\n", err);
1611                 goto free_queue_irqs;
1612         }
1613
1614         return 0;
1615
1616 free_queue_irqs:
1617         for (i = vector - 1; i >= 0; i--)
1618                 free_irq(adapter->msix_entries[--vector].vector,
1619                          adapter->q_vector[i]);
1620         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1621         pci_disable_msix(adapter->pdev);
1622         kfree(adapter->msix_entries);
1623         adapter->msix_entries = NULL;
1624 out:
1625         return err;
1626 }
1627
1628 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1629 {
1630         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1631         u8 current_itr;
1632         u32 new_itr = q_vector->eitr;
1633         struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1634         struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1635
1636         q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1637                                             q_vector->tx_itr,
1638                                             tx_ring->total_packets,
1639                                             tx_ring->total_bytes);
1640         q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1641                                             q_vector->rx_itr,
1642                                             rx_ring->total_packets,
1643                                             rx_ring->total_bytes);
1644
1645         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1646
1647         switch (current_itr) {
1648         /* counts and packets in update_itr are dependent on these numbers */
1649         case lowest_latency:
1650                 new_itr = 100000;
1651                 break;
1652         case low_latency:
1653                 new_itr = 20000; /* aka hwitr = ~200 */
1654                 break;
1655         case bulk_latency:
1656                 new_itr = 8000;
1657                 break;
1658         default:
1659                 break;
1660         }
1661
1662         if (new_itr != q_vector->eitr) {
1663                 /* do an exponential smoothing */
1664                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1665
1666                 /* save the algorithm value here, not the smoothed one */
1667                 q_vector->eitr = new_itr;
1668
1669                 ixgbe_write_eitr(q_vector);
1670         }
1671
1672         return;
1673 }
1674
1675 /**
1676  * ixgbe_irq_enable - Enable default interrupt generation settings
1677  * @adapter: board private structure
1678  **/
1679 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1680 {
1681         u32 mask;
1682
1683         mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1684         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1685                 mask |= IXGBE_EIMS_GPI_SDP1;
1686         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1687                 mask |= IXGBE_EIMS_ECC;
1688                 mask |= IXGBE_EIMS_GPI_SDP1;
1689                 mask |= IXGBE_EIMS_GPI_SDP2;
1690         }
1691         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
1692             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
1693                 mask |= IXGBE_EIMS_FLOW_DIR;
1694
1695         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1696         ixgbe_irq_enable_queues(adapter, ~0);
1697         IXGBE_WRITE_FLUSH(&adapter->hw);
1698 }
1699
1700 /**
1701  * ixgbe_intr - legacy mode Interrupt Handler
1702  * @irq: interrupt number
1703  * @data: pointer to a network interface device structure
1704  **/
1705 static irqreturn_t ixgbe_intr(int irq, void *data)
1706 {
1707         struct net_device *netdev = data;
1708         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1709         struct ixgbe_hw *hw = &adapter->hw;
1710         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1711         u32 eicr;
1712
1713         /*
1714          * Workaround for silicon errata.  Mask the interrupts
1715          * before the read of EICR.
1716          */
1717         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1718
1719         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1720          * therefore no explict interrupt disable is necessary */
1721         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1722         if (!eicr) {
1723                 /* shared interrupt alert!
1724                  * make sure interrupts are enabled because the read will
1725                  * have disabled interrupts due to EIAM */
1726                 ixgbe_irq_enable(adapter);
1727                 return IRQ_NONE;        /* Not our interrupt */
1728         }
1729
1730         if (eicr & IXGBE_EICR_LSC)
1731                 ixgbe_check_lsc(adapter);
1732
1733         if (hw->mac.type == ixgbe_mac_82599EB)
1734                 ixgbe_check_sfp_event(adapter, eicr);
1735
1736         ixgbe_check_fan_failure(adapter, eicr);
1737
1738         if (napi_schedule_prep(&(q_vector->napi))) {
1739                 adapter->tx_ring[0].total_packets = 0;
1740                 adapter->tx_ring[0].total_bytes = 0;
1741                 adapter->rx_ring[0].total_packets = 0;
1742                 adapter->rx_ring[0].total_bytes = 0;
1743                 /* would disable interrupts here but EIAM disabled it */
1744                 __napi_schedule(&(q_vector->napi));
1745         }
1746
1747         return IRQ_HANDLED;
1748 }
1749
1750 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1751 {
1752         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1753
1754         for (i = 0; i < q_vectors; i++) {
1755                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
1756                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1757                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1758                 q_vector->rxr_count = 0;
1759                 q_vector->txr_count = 0;
1760         }
1761 }
1762
1763 /**
1764  * ixgbe_request_irq - initialize interrupts
1765  * @adapter: board private structure
1766  *
1767  * Attempts to configure interrupts using the best available
1768  * capabilities of the hardware and kernel.
1769  **/
1770 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1771 {
1772         struct net_device *netdev = adapter->netdev;
1773         int err;
1774
1775         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1776                 err = ixgbe_request_msix_irqs(adapter);
1777         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1778                 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1779                                   netdev->name, netdev);
1780         } else {
1781                 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1782                                   netdev->name, netdev);
1783         }
1784
1785         if (err)
1786                 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1787
1788         return err;
1789 }
1790
1791 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1792 {
1793         struct net_device *netdev = adapter->netdev;
1794
1795         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1796                 int i, q_vectors;
1797
1798                 q_vectors = adapter->num_msix_vectors;
1799
1800                 i = q_vectors - 1;
1801                 free_irq(adapter->msix_entries[i].vector, netdev);
1802
1803                 i--;
1804                 for (; i >= 0; i--) {
1805                         free_irq(adapter->msix_entries[i].vector,
1806                                  adapter->q_vector[i]);
1807                 }
1808
1809                 ixgbe_reset_q_vectors(adapter);
1810         } else {
1811                 free_irq(adapter->pdev->irq, netdev);
1812         }
1813 }
1814
1815 /**
1816  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1817  * @adapter: board private structure
1818  **/
1819 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1820 {
1821         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1822                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1823         } else {
1824                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
1825                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
1826                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
1827         }
1828         IXGBE_WRITE_FLUSH(&adapter->hw);
1829         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1830                 int i;
1831                 for (i = 0; i < adapter->num_msix_vectors; i++)
1832                         synchronize_irq(adapter->msix_entries[i].vector);
1833         } else {
1834                 synchronize_irq(adapter->pdev->irq);
1835         }
1836 }
1837
1838 /**
1839  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1840  *
1841  **/
1842 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1843 {
1844         struct ixgbe_hw *hw = &adapter->hw;
1845
1846         IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1847                         EITR_INTS_PER_SEC_TO_REG(adapter->eitr_param));
1848
1849         ixgbe_set_ivar(adapter, 0, 0, 0);
1850         ixgbe_set_ivar(adapter, 1, 0, 0);
1851
1852         map_vector_to_rxq(adapter, 0, 0);
1853         map_vector_to_txq(adapter, 0, 0);
1854
1855         DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1856 }
1857
1858 /**
1859  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1860  * @adapter: board private structure
1861  *
1862  * Configure the Tx unit of the MAC after a reset.
1863  **/
1864 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1865 {
1866         u64 tdba;
1867         struct ixgbe_hw *hw = &adapter->hw;
1868         u32 i, j, tdlen, txctrl;
1869
1870         /* Setup the HW Tx Head and Tail descriptor pointers */
1871         for (i = 0; i < adapter->num_tx_queues; i++) {
1872                 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1873                 j = ring->reg_idx;
1874                 tdba = ring->dma;
1875                 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1876                 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1877                                 (tdba & DMA_BIT_MASK(32)));
1878                 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1879                 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1880                 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1881                 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1882                 adapter->tx_ring[i].head = IXGBE_TDH(j);
1883                 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1884                 /* Disable Tx Head Writeback RO bit, since this hoses
1885                  * bookkeeping if things aren't delivered in order.
1886                  */
1887                 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1888                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1889                 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1890         }
1891         if (hw->mac.type == ixgbe_mac_82599EB) {
1892                 /* We enable 8 traffic classes, DCB only */
1893                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
1894                         IXGBE_WRITE_REG(hw, IXGBE_MTQC, (IXGBE_MTQC_RT_ENA |
1895                                         IXGBE_MTQC_8TC_8TQ));
1896         }
1897 }
1898
1899 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1900
1901 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter, int index)
1902 {
1903         struct ixgbe_ring *rx_ring;
1904         u32 srrctl;
1905         int queue0 = 0;
1906         unsigned long mask;
1907         struct ixgbe_ring_feature *feature = adapter->ring_feature;
1908
1909         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1910                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
1911                         int dcb_i = feature[RING_F_DCB].indices;
1912                         if (dcb_i == 8)
1913                                 queue0 = index >> 4;
1914                         else if (dcb_i == 4)
1915                                 queue0 = index >> 5;
1916                         else
1917                                 dev_err(&adapter->pdev->dev, "Invalid DCB "
1918                                         "configuration\n");
1919 #ifdef IXGBE_FCOE
1920                         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1921                                 struct ixgbe_ring_feature *f;
1922
1923                                 rx_ring = &adapter->rx_ring[queue0];
1924                                 f = &adapter->ring_feature[RING_F_FCOE];
1925                                 if ((queue0 == 0) && (index > rx_ring->reg_idx))
1926                                         queue0 = f->mask + index -
1927                                                  rx_ring->reg_idx - 1;
1928                         }
1929 #endif /* IXGBE_FCOE */
1930                 } else {
1931                         queue0 = index;
1932                 }
1933         } else {
1934                 mask = (unsigned long) feature[RING_F_RSS].mask;
1935                 queue0 = index & mask;
1936                 index = index & mask;
1937         }
1938
1939         rx_ring = &adapter->rx_ring[queue0];
1940
1941         srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
1942
1943         srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1944         srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1945
1946         srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1947                   IXGBE_SRRCTL_BSIZEHDR_MASK;
1948
1949         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1950 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
1951                 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1952 #else
1953                 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1954 #endif
1955                 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1956         } else {
1957                 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
1958                           IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1959                 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1960         }
1961
1962         IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
1963 }
1964
1965 static u32 ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
1966 {
1967         u32 mrqc = 0;
1968         int mask;
1969
1970         if (!(adapter->hw.mac.type == ixgbe_mac_82599EB))
1971                 return mrqc;
1972
1973         mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
1974 #ifdef CONFIG_IXGBE_DCB
1975                                  | IXGBE_FLAG_DCB_ENABLED
1976 #endif
1977                                 );
1978
1979         switch (mask) {
1980         case (IXGBE_FLAG_RSS_ENABLED):
1981                 mrqc = IXGBE_MRQC_RSSEN;
1982                 break;
1983 #ifdef CONFIG_IXGBE_DCB
1984         case (IXGBE_FLAG_DCB_ENABLED):
1985                 mrqc = IXGBE_MRQC_RT8TCEN;
1986                 break;
1987 #endif /* CONFIG_IXGBE_DCB */
1988         default:
1989                 break;
1990         }
1991
1992         return mrqc;
1993 }
1994
1995 /**
1996  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
1997  * @adapter: board private structure
1998  *
1999  * Configure the Rx unit of the MAC after a reset.
2000  **/
2001 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2002 {
2003         u64 rdba;
2004         struct ixgbe_hw *hw = &adapter->hw;
2005         struct net_device *netdev = adapter->netdev;
2006         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2007         int i, j;
2008         u32 rdlen, rxctrl, rxcsum;
2009         static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2010                           0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2011                           0x6A3E67EA, 0x14364D17, 0x3BED200D};
2012         u32 fctrl, hlreg0;
2013         u32 reta = 0, mrqc = 0;
2014         u32 rdrxctl;
2015         u32 rscctrl;
2016         int rx_buf_len;
2017
2018         /* Decide whether to use packet split mode or not */
2019         adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2020
2021 #ifdef IXGBE_FCOE
2022         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2023                 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
2024 #endif /* IXGBE_FCOE */
2025
2026         /* Set the RX buffer length according to the mode */
2027         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2028                 rx_buf_len = IXGBE_RX_HDR_SIZE;
2029                 if (hw->mac.type == ixgbe_mac_82599EB) {
2030                         /* PSRTYPE must be initialized in 82599 */
2031                         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2032                                       IXGBE_PSRTYPE_UDPHDR |
2033                                       IXGBE_PSRTYPE_IPV4HDR |
2034                                       IXGBE_PSRTYPE_IPV6HDR |
2035                                       IXGBE_PSRTYPE_L2HDR;
2036                         IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), psrtype);
2037                 }
2038         } else {
2039                 if (!(adapter->flags & IXGBE_FLAG2_RSC_ENABLED) &&
2040                     (netdev->mtu <= ETH_DATA_LEN))
2041                         rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2042                 else
2043                         rx_buf_len = ALIGN(max_frame, 1024);
2044         }
2045
2046         fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
2047         fctrl |= IXGBE_FCTRL_BAM;
2048         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
2049         fctrl |= IXGBE_FCTRL_PMCF;
2050         IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
2051
2052         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2053         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2054                 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2055         else
2056                 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2057 #ifdef IXGBE_FCOE
2058         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2059                 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2060 #endif
2061         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2062
2063         rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
2064         /* disable receives while setting up the descriptors */
2065         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2066         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2067
2068         /*
2069          * Setup the HW Rx Head and Tail Descriptor Pointers and
2070          * the Base and Length of the Rx Descriptor Ring
2071          */
2072         for (i = 0; i < adapter->num_rx_queues; i++) {
2073                 rdba = adapter->rx_ring[i].dma;
2074                 j = adapter->rx_ring[i].reg_idx;
2075                 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_BIT_MASK(32)));
2076                 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
2077                 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
2078                 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
2079                 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
2080                 adapter->rx_ring[i].head = IXGBE_RDH(j);
2081                 adapter->rx_ring[i].tail = IXGBE_RDT(j);
2082                 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
2083
2084 #ifdef IXGBE_FCOE
2085                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
2086                         struct ixgbe_ring_feature *f;
2087                         f = &adapter->ring_feature[RING_F_FCOE];
2088                         if ((rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
2089                             (i >= f->mask) && (i < f->mask + f->indices))
2090                                 adapter->rx_ring[i].rx_buf_len =
2091                                         IXGBE_FCOE_JUMBO_FRAME_SIZE;
2092                 }
2093
2094 #endif /* IXGBE_FCOE */
2095                 ixgbe_configure_srrctl(adapter, j);
2096         }
2097
2098         if (hw->mac.type == ixgbe_mac_82598EB) {
2099                 /*
2100                  * For VMDq support of different descriptor types or
2101                  * buffer sizes through the use of multiple SRRCTL
2102                  * registers, RDRXCTL.MVMEN must be set to 1
2103                  *
2104                  * also, the manual doesn't mention it clearly but DCA hints
2105                  * will only use queue 0's tags unless this bit is set.  Side
2106                  * effects of setting this bit are only that SRRCTL must be
2107                  * fully programmed [0..15]
2108                  */
2109                 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2110                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2111                 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2112         }
2113
2114         /* Program MRQC for the distribution of queues */
2115         mrqc = ixgbe_setup_mrqc(adapter);
2116
2117         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2118                 /* Fill out redirection table */
2119                 for (i = 0, j = 0; i < 128; i++, j++) {
2120                         if (j == adapter->ring_feature[RING_F_RSS].indices)
2121                                 j = 0;
2122                         /* reta = 4-byte sliding window of
2123                          * 0x00..(indices-1)(indices-1)00..etc. */
2124                         reta = (reta << 8) | (j * 0x11);
2125                         if ((i & 3) == 3)
2126                                 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2127                 }
2128
2129                 /* Fill out hash function seeds */
2130                 for (i = 0; i < 10; i++)
2131                         IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2132
2133                 if (hw->mac.type == ixgbe_mac_82598EB)
2134                         mrqc |= IXGBE_MRQC_RSSEN;
2135                     /* Perform hash on these packet types */
2136                 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2137                       | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2138                       | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
2139                       | IXGBE_MRQC_RSS_FIELD_IPV6
2140                       | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
2141                       | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2142         }
2143         IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2144
2145         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2146
2147         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
2148             adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
2149                 /* Disable indicating checksum in descriptor, enables
2150                  * RSS hash */
2151                 rxcsum |= IXGBE_RXCSUM_PCSD;
2152         }
2153         if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
2154                 /* Enable IPv4 payload checksum for UDP fragments
2155                  * if PCSD is not set */
2156                 rxcsum |= IXGBE_RXCSUM_IPPCSE;
2157         }
2158
2159         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2160
2161         if (hw->mac.type == ixgbe_mac_82599EB) {
2162                 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2163                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2164                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2165                 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2166         }
2167
2168         if (adapter->flags & IXGBE_FLAG2_RSC_ENABLED) {
2169                 /* Enable 82599 HW-RSC */
2170                 for (i = 0; i < adapter->num_rx_queues; i++) {
2171                         j = adapter->rx_ring[i].reg_idx;
2172                         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2173                         rscctrl |= IXGBE_RSCCTL_RSCEN;
2174                         /*
2175                          * we must limit the number of descriptors so that the
2176                          * total size of max desc * buf_len is not greater
2177                          * than 65535
2178                          */
2179                         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2180 #if (MAX_SKB_FRAGS > 16)
2181                                 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2182 #elif (MAX_SKB_FRAGS > 8)
2183                                 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2184 #elif (MAX_SKB_FRAGS > 4)
2185                                 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2186 #else
2187                                 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2188 #endif
2189                         } else {
2190                                 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2191                                         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2192                                 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2193                                         rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2194                                 else
2195                                         rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2196                         }
2197                         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2198                 }
2199                 /* Disable RSC for ACK packets */
2200                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2201                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2202         }
2203 }
2204
2205 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2206 {
2207         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2208         struct ixgbe_hw *hw = &adapter->hw;
2209
2210         /* add VID to filter table */
2211         hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
2212 }
2213
2214 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2215 {
2216         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2217         struct ixgbe_hw *hw = &adapter->hw;
2218
2219         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2220                 ixgbe_irq_disable(adapter);
2221
2222         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2223
2224         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2225                 ixgbe_irq_enable(adapter);
2226
2227         /* remove VID from filter table */
2228         hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
2229 }
2230
2231 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2232                                    struct vlan_group *grp)
2233 {
2234         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2235         u32 ctrl;
2236         int i, j;
2237
2238         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2239                 ixgbe_irq_disable(adapter);
2240         adapter->vlgrp = grp;
2241
2242         /*
2243          * For a DCB driver, always enable VLAN tag stripping so we can
2244          * still receive traffic from a DCB-enabled host even if we're
2245          * not in DCB mode.
2246          */
2247         ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2248         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2249                 ctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2250                 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2251                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2252         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2253                 ctrl |= IXGBE_VLNCTRL_VFE;
2254                 /* enable VLAN tag insert/strip */
2255                 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2256                 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2257                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2258                 for (i = 0; i < adapter->num_rx_queues; i++) {
2259                         j = adapter->rx_ring[i].reg_idx;
2260                         ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_RXDCTL(j));
2261                         ctrl |= IXGBE_RXDCTL_VME;
2262                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_RXDCTL(j), ctrl);
2263                 }
2264         }
2265         ixgbe_vlan_rx_add_vid(netdev, 0);
2266
2267         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2268                 ixgbe_irq_enable(adapter);
2269 }
2270
2271 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2272 {
2273         ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2274
2275         if (adapter->vlgrp) {
2276                 u16 vid;
2277                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2278                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2279                                 continue;
2280                         ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2281                 }
2282         }
2283 }
2284
2285 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
2286 {
2287         struct dev_mc_list *mc_ptr;
2288         u8 *addr = *mc_addr_ptr;
2289         *vmdq = 0;
2290
2291         mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
2292         if (mc_ptr->next)
2293                 *mc_addr_ptr = mc_ptr->next->dmi_addr;
2294         else
2295                 *mc_addr_ptr = NULL;
2296
2297         return addr;
2298 }
2299
2300 /**
2301  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
2302  * @netdev: network interface device structure
2303  *
2304  * The set_rx_method entry point is called whenever the unicast/multicast
2305  * address list or the network interface flags are updated.  This routine is
2306  * responsible for configuring the hardware for proper unicast, multicast and
2307  * promiscuous mode.
2308  **/
2309 static void ixgbe_set_rx_mode(struct net_device *netdev)
2310 {
2311         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2312         struct ixgbe_hw *hw = &adapter->hw;
2313         u32 fctrl, vlnctrl;
2314         u8 *addr_list = NULL;
2315         int addr_count = 0;
2316
2317         /* Check for Promiscuous and All Multicast modes */
2318
2319         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2320         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2321
2322         if (netdev->flags & IFF_PROMISC) {
2323                 hw->addr_ctrl.user_set_promisc = 1;
2324                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2325                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2326         } else {
2327                 if (netdev->flags & IFF_ALLMULTI) {
2328                         fctrl |= IXGBE_FCTRL_MPE;
2329                         fctrl &= ~IXGBE_FCTRL_UPE;
2330                 } else {
2331                         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2332                 }
2333                 vlnctrl |= IXGBE_VLNCTRL_VFE;
2334                 hw->addr_ctrl.user_set_promisc = 0;
2335         }
2336
2337         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
2338         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2339
2340         /* reprogram secondary unicast list */
2341         hw->mac.ops.update_uc_addr_list(hw, &netdev->uc.list);
2342
2343         /* reprogram multicast list */
2344         addr_count = netdev->mc_count;
2345         if (addr_count)
2346                 addr_list = netdev->mc_list->dmi_addr;
2347         hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
2348                                         ixgbe_addr_list_itr);
2349 }
2350
2351 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
2352 {
2353         int q_idx;
2354         struct ixgbe_q_vector *q_vector;
2355         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2356
2357         /* legacy and MSI only use one vector */
2358         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2359                 q_vectors = 1;
2360
2361         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2362                 struct napi_struct *napi;
2363                 q_vector = adapter->q_vector[q_idx];
2364                 napi = &q_vector->napi;
2365                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2366                         if (!q_vector->rxr_count || !q_vector->txr_count) {
2367                                 if (q_vector->txr_count == 1)
2368                                         napi->poll = &ixgbe_clean_txonly;
2369                                 else if (q_vector->rxr_count == 1)
2370                                         napi->poll = &ixgbe_clean_rxonly;
2371                         }
2372                 }
2373
2374                 napi_enable(napi);
2375         }
2376 }
2377
2378 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2379 {
2380         int q_idx;
2381         struct ixgbe_q_vector *q_vector;
2382         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2383
2384         /* legacy and MSI only use one vector */
2385         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2386                 q_vectors = 1;
2387
2388         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2389                 q_vector = adapter->q_vector[q_idx];
2390                 napi_disable(&q_vector->napi);
2391         }
2392 }
2393
2394 #ifdef CONFIG_IXGBE_DCB
2395 /*
2396  * ixgbe_configure_dcb - Configure DCB hardware
2397  * @adapter: ixgbe adapter struct
2398  *
2399  * This is called by the driver on open to configure the DCB hardware.
2400  * This is also called by the gennetlink interface when reconfiguring
2401  * the DCB state.
2402  */
2403 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2404 {
2405         struct ixgbe_hw *hw = &adapter->hw;
2406         u32 txdctl, vlnctrl;
2407         int i, j;
2408
2409         ixgbe_dcb_check_config(&adapter->dcb_cfg);
2410         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2411         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2412
2413         /* reconfigure the hardware */
2414         ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
2415
2416         for (i = 0; i < adapter->num_tx_queues; i++) {
2417                 j = adapter->tx_ring[i].reg_idx;
2418                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2419                 /* PThresh workaround for Tx hang with DFP enabled. */
2420                 txdctl |= 32;
2421                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2422         }
2423         /* Enable VLAN tag insert/strip */
2424         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2425         if (hw->mac.type == ixgbe_mac_82598EB) {
2426                 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2427                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2428                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2429         } else if (hw->mac.type == ixgbe_mac_82599EB) {
2430                 vlnctrl |= IXGBE_VLNCTRL_VFE;
2431                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2432                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2433                 for (i = 0; i < adapter->num_rx_queues; i++) {
2434                         j = adapter->rx_ring[i].reg_idx;
2435                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2436                         vlnctrl |= IXGBE_RXDCTL_VME;
2437                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2438                 }
2439         }
2440         hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2441 }
2442
2443 #endif
2444 static void ixgbe_configure(struct ixgbe_adapter *adapter)
2445 {
2446         struct net_device *netdev = adapter->netdev;
2447         struct ixgbe_hw *hw = &adapter->hw;
2448         int i;
2449
2450         ixgbe_set_rx_mode(netdev);
2451
2452         ixgbe_restore_vlan(adapter);
2453 #ifdef CONFIG_IXGBE_DCB
2454         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2455                 netif_set_gso_max_size(netdev, 32768);
2456                 ixgbe_configure_dcb(adapter);
2457         } else {
2458                 netif_set_gso_max_size(netdev, 65536);
2459         }
2460 #else
2461         netif_set_gso_max_size(netdev, 65536);
2462 #endif
2463
2464 #ifdef IXGBE_FCOE
2465         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2466                 ixgbe_configure_fcoe(adapter);
2467
2468 #endif /* IXGBE_FCOE */
2469         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2470                 for (i = 0; i < adapter->num_tx_queues; i++)
2471                         adapter->tx_ring[i].atr_sample_rate =
2472                                                        adapter->atr_sample_rate;
2473                 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
2474         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
2475                 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
2476         }
2477
2478         ixgbe_configure_tx(adapter);
2479         ixgbe_configure_rx(adapter);
2480         for (i = 0; i < adapter->num_rx_queues; i++)
2481                 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
2482                                        (adapter->rx_ring[i].count - 1));
2483 }
2484
2485 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2486 {
2487         switch (hw->phy.type) {
2488         case ixgbe_phy_sfp_avago:
2489         case ixgbe_phy_sfp_ftl:
2490         case ixgbe_phy_sfp_intel:
2491         case ixgbe_phy_sfp_unknown:
2492         case ixgbe_phy_tw_tyco:
2493         case ixgbe_phy_tw_unknown:
2494                 return true;
2495         default:
2496                 return false;
2497         }
2498 }
2499
2500 /**
2501  * ixgbe_sfp_link_config - set up SFP+ link
2502  * @adapter: pointer to private adapter struct
2503  **/
2504 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2505 {
2506         struct ixgbe_hw *hw = &adapter->hw;
2507
2508                 if (hw->phy.multispeed_fiber) {
2509                         /*
2510                          * In multispeed fiber setups, the device may not have
2511                          * had a physical connection when the driver loaded.
2512                          * If that's the case, the initial link configuration
2513                          * couldn't get the MAC into 10G or 1G mode, so we'll
2514                          * never have a link status change interrupt fire.
2515                          * We need to try and force an autonegotiation
2516                          * session, then bring up link.
2517                          */
2518                         hw->mac.ops.setup_sfp(hw);
2519                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2520                                 schedule_work(&adapter->multispeed_fiber_task);
2521                 } else {
2522                         /*
2523                          * Direct Attach Cu and non-multispeed fiber modules
2524                          * still need to be configured properly prior to
2525                          * attempting link.
2526                          */
2527                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2528                                 schedule_work(&adapter->sfp_config_module_task);
2529                 }
2530 }
2531
2532 /**
2533  * ixgbe_non_sfp_link_config - set up non-SFP+ link
2534  * @hw: pointer to private hardware struct
2535  *
2536  * Returns 0 on success, negative on failure
2537  **/
2538 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2539 {
2540         u32 autoneg;
2541         bool link_up = false;
2542         u32 ret = IXGBE_ERR_LINK_SETUP;
2543
2544         if (hw->mac.ops.check_link)
2545                 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2546
2547         if (ret)
2548                 goto link_cfg_out;
2549
2550         if (hw->mac.ops.get_link_capabilities)
2551                 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
2552                                                         &hw->mac.autoneg);
2553         if (ret)
2554                 goto link_cfg_out;
2555
2556         if (hw->mac.ops.setup_link_speed)
2557                 ret = hw->mac.ops.setup_link_speed(hw, autoneg, true, link_up);
2558 link_cfg_out:
2559         return ret;
2560 }
2561
2562 #define IXGBE_MAX_RX_DESC_POLL 10
2563 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2564                                               int rxr)
2565 {
2566         int j = adapter->rx_ring[rxr].reg_idx;
2567         int k;
2568
2569         for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2570                 if (IXGBE_READ_REG(&adapter->hw,
2571                                    IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2572                         break;
2573                 else
2574                         msleep(1);
2575         }
2576         if (k >= IXGBE_MAX_RX_DESC_POLL) {
2577                 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2578                         "not set within the polling period\n", rxr);
2579         }
2580         ixgbe_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
2581                               (adapter->rx_ring[rxr].count - 1));
2582 }
2583
2584 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2585 {
2586         struct net_device *netdev = adapter->netdev;
2587         struct ixgbe_hw *hw = &adapter->hw;
2588         int i, j = 0;
2589         int num_rx_rings = adapter->num_rx_queues;
2590         int err;
2591         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2592         u32 txdctl, rxdctl, mhadd;
2593         u32 dmatxctl;
2594         u32 gpie;
2595
2596         ixgbe_get_hw_control(adapter);
2597
2598         if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2599             (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2600                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2601                         gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2602                                 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2603                 } else {
2604                         /* MSI only */
2605                         gpie = 0;
2606                 }
2607                 /* XXX: to interrupt immediately for EICS writes, enable this */
2608                 /* gpie |= IXGBE_GPIE_EIMEN; */
2609                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2610         }
2611
2612         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2613                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2614                  * specifically only auto mask tx and rx interrupts */
2615                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2616         }
2617
2618         /* Enable fan failure interrupt if media type is copper */
2619         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2620                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2621                 gpie |= IXGBE_SDP1_GPIEN;
2622                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2623         }
2624
2625         if (hw->mac.type == ixgbe_mac_82599EB) {
2626                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2627                 gpie |= IXGBE_SDP1_GPIEN;
2628                 gpie |= IXGBE_SDP2_GPIEN;
2629                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2630         }
2631
2632 #ifdef IXGBE_FCOE
2633         /* adjust max frame to be able to do baby jumbo for FCoE */
2634         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2635             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2636                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2637
2638 #endif /* IXGBE_FCOE */
2639         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2640         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2641                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2642                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2643
2644                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2645         }
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                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2651                 txdctl |= (8 << 16);
2652                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2653         }
2654
2655         if (hw->mac.type == ixgbe_mac_82599EB) {
2656                 /* DMATXCTL.EN must be set after all Tx queue config is done */
2657                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2658                 dmatxctl |= IXGBE_DMATXCTL_TE;
2659                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2660         }
2661         for (i = 0; i < adapter->num_tx_queues; i++) {
2662                 j = adapter->tx_ring[i].reg_idx;
2663                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2664                 txdctl |= IXGBE_TXDCTL_ENABLE;
2665                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2666         }
2667
2668         for (i = 0; i < num_rx_rings; i++) {
2669                 j = adapter->rx_ring[i].reg_idx;
2670                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2671                 /* enable PTHRESH=32 descriptors (half the internal cache)
2672                  * and HTHRESH=0 descriptors (to minimize latency on fetch),
2673                  * this also removes a pesky rx_no_buffer_count increment */
2674                 rxdctl |= 0x0020;
2675                 rxdctl |= IXGBE_RXDCTL_ENABLE;
2676                 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2677                 if (hw->mac.type == ixgbe_mac_82599EB)
2678                         ixgbe_rx_desc_queue_enable(adapter, i);
2679         }
2680         /* enable all receives */
2681         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2682         if (hw->mac.type == ixgbe_mac_82598EB)
2683                 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2684         else
2685                 rxdctl |= IXGBE_RXCTRL_RXEN;
2686         hw->mac.ops.enable_rx_dma(hw, rxdctl);
2687
2688         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2689                 ixgbe_configure_msix(adapter);
2690         else
2691                 ixgbe_configure_msi_and_legacy(adapter);
2692
2693         clear_bit(__IXGBE_DOWN, &adapter->state);
2694         ixgbe_napi_enable_all(adapter);
2695
2696         /* clear any pending interrupts, may auto mask */
2697         IXGBE_READ_REG(hw, IXGBE_EICR);
2698
2699         ixgbe_irq_enable(adapter);
2700
2701         /*
2702          * If this adapter has a fan, check to see if we had a failure
2703          * before we enabled the interrupt.
2704          */
2705         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2706                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
2707                 if (esdp & IXGBE_ESDP_SDP1)
2708                         DPRINTK(DRV, CRIT,
2709                                 "Fan has stopped, replace the adapter\n");
2710         }
2711
2712         /*
2713          * For hot-pluggable SFP+ devices, a new SFP+ module may have
2714          * arrived before interrupts were enabled but after probe.  Such
2715          * devices wouldn't have their type identified yet. We need to
2716          * kick off the SFP+ module setup first, then try to bring up link.
2717          * If we're not hot-pluggable SFP+, we just need to configure link
2718          * and bring it up.
2719          */
2720         if (hw->phy.type == ixgbe_phy_unknown) {
2721                 err = hw->phy.ops.identify(hw);
2722                 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
2723                         /*
2724                          * Take the device down and schedule the sfp tasklet
2725                          * which will unregister_netdev and log it.
2726                          */
2727                         ixgbe_down(adapter);
2728                         schedule_work(&adapter->sfp_config_module_task);
2729                         return err;
2730                 }
2731         }
2732
2733         if (ixgbe_is_sfp(hw)) {
2734                 ixgbe_sfp_link_config(adapter);
2735         } else {
2736                 err = ixgbe_non_sfp_link_config(hw);
2737                 if (err)
2738                         DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
2739         }
2740
2741         for (i = 0; i < adapter->num_tx_queues; i++)
2742                 set_bit(__IXGBE_FDIR_INIT_DONE,
2743                         &(adapter->tx_ring[i].reinit_state));
2744
2745         /* enable transmits */
2746         netif_tx_start_all_queues(netdev);
2747
2748         /* bring the link up in the watchdog, this could race with our first
2749          * link up interrupt but shouldn't be a problem */
2750         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2751         adapter->link_check_timeout = jiffies;
2752         mod_timer(&adapter->watchdog_timer, jiffies);
2753         return 0;
2754 }
2755
2756 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
2757 {
2758         WARN_ON(in_interrupt());
2759         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
2760                 msleep(1);
2761         ixgbe_down(adapter);
2762         ixgbe_up(adapter);
2763         clear_bit(__IXGBE_RESETTING, &adapter->state);
2764 }
2765
2766 int ixgbe_up(struct ixgbe_adapter *adapter)
2767 {
2768         /* hardware has been reset, we need to reload some things */
2769         ixgbe_configure(adapter);
2770
2771         return ixgbe_up_complete(adapter);
2772 }
2773
2774 void ixgbe_reset(struct ixgbe_adapter *adapter)
2775 {
2776         struct ixgbe_hw *hw = &adapter->hw;
2777         int err;
2778
2779         err = hw->mac.ops.init_hw(hw);
2780         switch (err) {
2781         case 0:
2782         case IXGBE_ERR_SFP_NOT_PRESENT:
2783                 break;
2784         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
2785                 dev_err(&adapter->pdev->dev, "master disable timed out\n");
2786                 break;
2787         case IXGBE_ERR_EEPROM_VERSION:
2788                 /* We are running on a pre-production device, log a warning */
2789                 dev_warn(&adapter->pdev->dev, "This device is a pre-production "
2790                          "adapter/LOM.  Please be aware there may be issues "
2791                          "associated with your hardware.  If you are "
2792                          "experiencing problems please contact your Intel or "
2793                          "hardware representative who provided you with this "
2794                          "hardware.\n");
2795                 break;
2796         default:
2797                 dev_err(&adapter->pdev->dev, "Hardware Error: %d\n", err);
2798         }
2799
2800         /* reprogram the RAR[0] in case user changed it. */
2801         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2802 }
2803
2804 /**
2805  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2806  * @adapter: board private structure
2807  * @rx_ring: ring to free buffers from
2808  **/
2809 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2810                                 struct ixgbe_ring *rx_ring)
2811 {
2812         struct pci_dev *pdev = adapter->pdev;
2813         unsigned long size;
2814         unsigned int i;
2815
2816         /* Free all the Rx ring sk_buffs */
2817
2818         for (i = 0; i < rx_ring->count; i++) {
2819                 struct ixgbe_rx_buffer *rx_buffer_info;
2820
2821                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2822                 if (rx_buffer_info->dma) {
2823                         pci_unmap_single(pdev, rx_buffer_info->dma,
2824                                          rx_ring->rx_buf_len,
2825                                          PCI_DMA_FROMDEVICE);
2826                         rx_buffer_info->dma = 0;
2827                 }
2828                 if (rx_buffer_info->skb) {
2829                         struct sk_buff *skb = rx_buffer_info->skb;
2830                         rx_buffer_info->skb = NULL;
2831                         do {
2832                                 struct sk_buff *this = skb;
2833                                 skb = skb->prev;
2834                                 dev_kfree_skb(this);
2835                         } while (skb);
2836                 }
2837                 if (!rx_buffer_info->page)
2838                         continue;
2839                 if (rx_buffer_info->page_dma) {
2840                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
2841                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
2842                         rx_buffer_info->page_dma = 0;
2843                 }
2844                 put_page(rx_buffer_info->page);
2845                 rx_buffer_info->page = NULL;
2846                 rx_buffer_info->page_offset = 0;
2847         }
2848
2849         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2850         memset(rx_ring->rx_buffer_info, 0, size);
2851
2852         /* Zero out the descriptor ring */
2853         memset(rx_ring->desc, 0, rx_ring->size);
2854
2855         rx_ring->next_to_clean = 0;
2856         rx_ring->next_to_use = 0;
2857
2858         if (rx_ring->head)
2859                 writel(0, adapter->hw.hw_addr + rx_ring->head);
2860         if (rx_ring->tail)
2861                 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2862 }
2863
2864 /**
2865  * ixgbe_clean_tx_ring - Free Tx Buffers
2866  * @adapter: board private structure
2867  * @tx_ring: ring to be cleaned
2868  **/
2869 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2870                                 struct ixgbe_ring *tx_ring)
2871 {
2872         struct ixgbe_tx_buffer *tx_buffer_info;
2873         unsigned long size;
2874         unsigned int i;
2875
2876         /* Free all the Tx ring sk_buffs */
2877
2878         for (i = 0; i < tx_ring->count; i++) {
2879                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2880                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2881         }
2882
2883         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2884         memset(tx_ring->tx_buffer_info, 0, size);
2885
2886         /* Zero out the descriptor ring */
2887         memset(tx_ring->desc, 0, tx_ring->size);
2888
2889         tx_ring->next_to_use = 0;
2890         tx_ring->next_to_clean = 0;
2891
2892         if (tx_ring->head)
2893                 writel(0, adapter->hw.hw_addr + tx_ring->head);
2894         if (tx_ring->tail)
2895                 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2896 }
2897
2898 /**
2899  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2900  * @adapter: board private structure
2901  **/
2902 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2903 {
2904         int i;
2905
2906         for (i = 0; i < adapter->num_rx_queues; i++)
2907                 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2908 }
2909
2910 /**
2911  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2912  * @adapter: board private structure
2913  **/
2914 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2915 {
2916         int i;
2917
2918         for (i = 0; i < adapter->num_tx_queues; i++)
2919                 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2920 }
2921
2922 void ixgbe_down(struct ixgbe_adapter *adapter)
2923 {
2924         struct net_device *netdev = adapter->netdev;
2925         struct ixgbe_hw *hw = &adapter->hw;
2926         u32 rxctrl;
2927         u32 txdctl;
2928         int i, j;
2929
2930         /* signal that we are down to the interrupt handler */
2931         set_bit(__IXGBE_DOWN, &adapter->state);
2932
2933         /* disable receives */
2934         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2935         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2936
2937         netif_tx_disable(netdev);
2938
2939         IXGBE_WRITE_FLUSH(hw);
2940         msleep(10);
2941
2942         netif_tx_stop_all_queues(netdev);
2943
2944         ixgbe_irq_disable(adapter);
2945
2946         ixgbe_napi_disable_all(adapter);
2947
2948         del_timer_sync(&adapter->watchdog_timer);
2949         cancel_work_sync(&adapter->watchdog_task);
2950
2951         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2952             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2953                 cancel_work_sync(&adapter->fdir_reinit_task);
2954
2955         /* disable transmits in the hardware now that interrupts are off */
2956         for (i = 0; i < adapter->num_tx_queues; i++) {
2957                 j = adapter->tx_ring[i].reg_idx;
2958                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2959                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
2960                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
2961         }
2962         /* Disable the Tx DMA engine on 82599 */
2963         if (hw->mac.type == ixgbe_mac_82599EB)
2964                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
2965                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
2966                                  ~IXGBE_DMATXCTL_TE));
2967
2968         netif_carrier_off(netdev);
2969
2970         if (!pci_channel_offline(adapter->pdev))
2971                 ixgbe_reset(adapter);
2972         ixgbe_clean_all_tx_rings(adapter);
2973         ixgbe_clean_all_rx_rings(adapter);
2974
2975 #ifdef CONFIG_IXGBE_DCA
2976         /* since we reset the hardware DCA settings were cleared */
2977         ixgbe_setup_dca(adapter);
2978 #endif
2979 }
2980
2981 /**
2982  * ixgbe_poll - NAPI Rx polling callback
2983  * @napi: structure for representing this polling device
2984  * @budget: how many packets driver is allowed to clean
2985  *
2986  * This function is used for legacy and MSI, NAPI mode
2987  **/
2988 static int ixgbe_poll(struct napi_struct *napi, int budget)
2989 {
2990         struct ixgbe_q_vector *q_vector =
2991                                 container_of(napi, struct ixgbe_q_vector, napi);
2992         struct ixgbe_adapter *adapter = q_vector->adapter;
2993         int tx_clean_complete, work_done = 0;
2994
2995 #ifdef CONFIG_IXGBE_DCA
2996         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2997                 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
2998                 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
2999         }
3000 #endif
3001
3002         tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring);
3003         ixgbe_clean_rx_irq(q_vector, adapter->rx_ring, &work_done, budget);
3004
3005         if (!tx_clean_complete)
3006                 work_done = budget;
3007
3008         /* If budget not fully consumed, exit the polling mode */
3009         if (work_done < budget) {
3010                 napi_complete(napi);
3011                 if (adapter->itr_setting & 1)
3012                         ixgbe_set_itr(adapter);
3013                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3014                         ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3015         }
3016         return work_done;
3017 }
3018
3019 /**
3020  * ixgbe_tx_timeout - Respond to a Tx Hang
3021  * @netdev: network interface device structure
3022  **/
3023 static void ixgbe_tx_timeout(struct net_device *netdev)
3024 {
3025         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3026
3027         /* Do the reset outside of interrupt context */
3028         schedule_work(&adapter->reset_task);
3029 }
3030
3031 static void ixgbe_reset_task(struct work_struct *work)
3032 {
3033         struct ixgbe_adapter *adapter;
3034         adapter = container_of(work, struct ixgbe_adapter, reset_task);
3035
3036         /* If we're already down or resetting, just bail */
3037         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3038             test_bit(__IXGBE_RESETTING, &adapter->state))
3039                 return;
3040
3041         adapter->tx_timeout_count++;
3042
3043         ixgbe_reinit_locked(adapter);
3044 }
3045
3046 #ifdef CONFIG_IXGBE_DCB
3047 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3048 {
3049         bool ret = false;
3050         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3051
3052         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3053                 return ret;
3054
3055         f->mask = 0x7 << 3;
3056         adapter->num_rx_queues = f->indices;
3057         adapter->num_tx_queues = f->indices;
3058         ret = true;
3059
3060         return ret;
3061 }
3062 #endif
3063
3064 /**
3065  * ixgbe_set_rss_queues: Allocate queues for RSS
3066  * @adapter: board private structure to initialize
3067  *
3068  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
3069  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3070  *
3071  **/
3072 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3073 {
3074         bool ret = false;
3075         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3076
3077         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3078                 f->mask = 0xF;
3079                 adapter->num_rx_queues = f->indices;
3080                 adapter->num_tx_queues = f->indices;
3081                 ret = true;
3082         } else {
3083                 ret = false;
3084         }
3085
3086         return ret;
3087 }
3088
3089 /**
3090  * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3091  * @adapter: board private structure to initialize
3092  *
3093  * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3094  * to the original CPU that initiated the Tx session.  This runs in addition
3095  * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3096  * Rx load across CPUs using RSS.
3097  *
3098  **/
3099 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3100 {
3101         bool ret = false;
3102         struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3103
3104         f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3105         f_fdir->mask = 0;
3106
3107         /* Flow Director must have RSS enabled */
3108         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3109             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3110              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3111                 adapter->num_tx_queues = f_fdir->indices;
3112                 adapter->num_rx_queues = f_fdir->indices;
3113                 ret = true;
3114         } else {
3115                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3116                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3117         }
3118         return ret;
3119 }
3120
3121 #ifdef IXGBE_FCOE
3122 /**
3123  * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3124  * @adapter: board private structure to initialize
3125  *
3126  * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3127  * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3128  * rx queues out of the max number of rx queues, instead, it is used as the
3129  * index of the first rx queue used by FCoE.
3130  *
3131  **/
3132 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3133 {
3134         bool ret = false;
3135         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3136
3137         f->indices = min((int)num_online_cpus(), f->indices);
3138         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3139 #ifdef CONFIG_IXGBE_DCB
3140                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3141                         DPRINTK(PROBE, INFO, "FCOE enabled with DCB \n");
3142                         ixgbe_set_dcb_queues(adapter);
3143                 }
3144 #endif
3145                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3146                         DPRINTK(PROBE, INFO, "FCOE enabled with RSS \n");
3147                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3148                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3149                                 ixgbe_set_fdir_queues(adapter);
3150                         else
3151                                 ixgbe_set_rss_queues(adapter);
3152                 }
3153                 /* adding FCoE rx rings to the end */
3154                 f->mask = adapter->num_rx_queues;
3155                 adapter->num_rx_queues += f->indices;
3156                 if (adapter->num_tx_queues == 0)
3157                         adapter->num_tx_queues = f->indices;
3158
3159                 ret = true;
3160         }
3161
3162         return ret;
3163 }
3164
3165 #endif /* IXGBE_FCOE */
3166 /*
3167  * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3168  * @adapter: board private structure to initialize
3169  *
3170  * This is the top level queue allocation routine.  The order here is very
3171  * important, starting with the "most" number of features turned on at once,
3172  * and ending with the smallest set of features.  This way large combinations
3173  * can be allocated if they're turned on, and smaller combinations are the
3174  * fallthrough conditions.
3175  *
3176  **/
3177 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
3178 {
3179 #ifdef IXGBE_FCOE
3180         if (ixgbe_set_fcoe_queues(adapter))
3181                 goto done;
3182
3183 #endif /* IXGBE_FCOE */
3184 #ifdef CONFIG_IXGBE_DCB
3185         if (ixgbe_set_dcb_queues(adapter))
3186                 goto done;
3187
3188 #endif
3189         if (ixgbe_set_fdir_queues(adapter))
3190                 goto done;
3191
3192         if (ixgbe_set_rss_queues(adapter))
3193                 goto done;
3194
3195         /* fallback to base case */
3196         adapter->num_rx_queues = 1;
3197         adapter->num_tx_queues = 1;
3198
3199 done:
3200         /* Notify the stack of the (possibly) reduced Tx Queue count. */
3201         adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
3202 }
3203
3204 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
3205                                        int vectors)
3206 {
3207         int err, vector_threshold;
3208
3209         /* We'll want at least 3 (vector_threshold):
3210          * 1) TxQ[0] Cleanup
3211          * 2) RxQ[0] Cleanup
3212          * 3) Other (Link Status Change, etc.)
3213          * 4) TCP Timer (optional)
3214          */
3215         vector_threshold = MIN_MSIX_COUNT;
3216
3217         /* The more we get, the more we will assign to Tx/Rx Cleanup
3218          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
3219          * Right now, we simply care about how many we'll get; we'll
3220          * set them up later while requesting irq's.
3221          */
3222         while (vectors >= vector_threshold) {
3223                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
3224                                       vectors);
3225                 if (!err) /* Success in acquiring all requested vectors. */
3226                         break;
3227                 else if (err < 0)
3228                         vectors = 0; /* Nasty failure, quit now */
3229                 else /* err == number of vectors we should try again with */
3230                         vectors = err;
3231         }
3232
3233         if (vectors < vector_threshold) {
3234                 /* Can't allocate enough MSI-X interrupts?  Oh well.
3235                  * This just means we'll go with either a single MSI
3236                  * vector or fall back to legacy interrupts.
3237                  */
3238                 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
3239                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3240                 kfree(adapter->msix_entries);
3241                 adapter->msix_entries = NULL;
3242         } else {
3243                 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
3244                 /*
3245                  * Adjust for only the vectors we'll use, which is minimum
3246                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
3247                  * vectors we were allocated.
3248                  */
3249                 adapter->num_msix_vectors = min(vectors,
3250                                    adapter->max_msix_q_vectors + NON_Q_VECTORS);
3251         }
3252 }
3253
3254 /**
3255  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
3256  * @adapter: board private structure to initialize
3257  *
3258  * Cache the descriptor ring offsets for RSS to the assigned rings.
3259  *
3260  **/
3261 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
3262 {
3263         int i;
3264         bool ret = false;
3265
3266         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3267                 for (i = 0; i < adapter->num_rx_queues; i++)
3268                         adapter->rx_ring[i].reg_idx = i;
3269                 for (i = 0; i < adapter->num_tx_queues; i++)
3270                         adapter->tx_ring[i].reg_idx = i;
3271                 ret = true;
3272         } else {
3273                 ret = false;
3274         }
3275
3276         return ret;
3277 }
3278
3279 #ifdef CONFIG_IXGBE_DCB
3280 /**
3281  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
3282  * @adapter: board private structure to initialize
3283  *
3284  * Cache the descriptor ring offsets for DCB to the assigned rings.
3285  *
3286  **/
3287 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
3288 {
3289         int i;
3290         bool ret = false;
3291         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
3292
3293         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3294                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3295                         /* the number of queues is assumed to be symmetric */
3296                         for (i = 0; i < dcb_i; i++) {
3297                                 adapter->rx_ring[i].reg_idx = i << 3;
3298                                 adapter->tx_ring[i].reg_idx = i << 2;
3299                         }
3300                         ret = true;
3301                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
3302                         if (dcb_i == 8) {
3303                                 /*
3304                                  * Tx TC0 starts at: descriptor queue 0
3305                                  * Tx TC1 starts at: descriptor queue 32
3306                                  * Tx TC2 starts at: descriptor queue 64
3307                                  * Tx TC3 starts at: descriptor queue 80
3308                                  * Tx TC4 starts at: descriptor queue 96
3309                                  * Tx TC5 starts at: descriptor queue 104
3310                                  * Tx TC6 starts at: descriptor queue 112
3311                                  * Tx TC7 starts at: descriptor queue 120
3312                                  *
3313                                  * Rx TC0-TC7 are offset by 16 queues each
3314                                  */
3315                                 for (i = 0; i < 3; i++) {
3316                                         adapter->tx_ring[i].reg_idx = i << 5;
3317                                         adapter->rx_ring[i].reg_idx = i << 4;
3318                                 }
3319                                 for ( ; i < 5; i++) {
3320                                         adapter->tx_ring[i].reg_idx =
3321                                                                  ((i + 2) << 4);
3322                                         adapter->rx_ring[i].reg_idx = i << 4;
3323                                 }
3324                                 for ( ; i < dcb_i; i++) {
3325                                         adapter->tx_ring[i].reg_idx =
3326                                                                  ((i + 8) << 3);
3327                                         adapter->rx_ring[i].reg_idx = i << 4;
3328                                 }
3329
3330                                 ret = true;
3331                         } else if (dcb_i == 4) {
3332                                 /*
3333                                  * Tx TC0 starts at: descriptor queue 0
3334                                  * Tx TC1 starts at: descriptor queue 64
3335                                  * Tx TC2 starts at: descriptor queue 96
3336                                  * Tx TC3 starts at: descriptor queue 112
3337                                  *
3338                                  * Rx TC0-TC3 are offset by 32 queues each
3339                                  */
3340                                 adapter->tx_ring[0].reg_idx = 0;
3341                                 adapter->tx_ring[1].reg_idx = 64;
3342                                 adapter->tx_ring[2].reg_idx = 96;
3343                                 adapter->tx_ring[3].reg_idx = 112;
3344                                 for (i = 0 ; i < dcb_i; i++)
3345                                         adapter->rx_ring[i].reg_idx = i << 5;
3346
3347                                 ret = true;
3348                         } else {
3349                                 ret = false;
3350                         }
3351                 } else {
3352                         ret = false;
3353                 }
3354         } else {
3355                 ret = false;
3356         }
3357
3358         return ret;
3359 }
3360 #endif
3361
3362 /**
3363  * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
3364  * @adapter: board private structure to initialize
3365  *
3366  * Cache the descriptor ring offsets for Flow Director to the assigned rings.
3367  *
3368  **/
3369 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
3370 {
3371         int i;
3372         bool ret = false;
3373
3374         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3375             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3376              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
3377                 for (i = 0; i < adapter->num_rx_queues; i++)
3378                         adapter->rx_ring[i].reg_idx = i;
3379                 for (i = 0; i < adapter->num_tx_queues; i++)
3380                         adapter->tx_ring[i].reg_idx = i;
3381                 ret = true;
3382         }
3383
3384         return ret;
3385 }
3386
3387 #ifdef IXGBE_FCOE
3388 /**
3389  * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
3390  * @adapter: board private structure to initialize
3391  *
3392  * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
3393  *
3394  */
3395 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
3396 {
3397         int i, fcoe_i = 0;
3398         bool ret = false;
3399         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3400
3401         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3402 #ifdef CONFIG_IXGBE_DCB
3403                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3404                         ixgbe_cache_ring_dcb(adapter);
3405                         fcoe_i = adapter->rx_ring[0].reg_idx + 1;
3406                 }
3407 #endif /* CONFIG_IXGBE_DCB */
3408                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3409                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3410                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3411                                 ixgbe_cache_ring_fdir(adapter);
3412                         else
3413                                 ixgbe_cache_ring_rss(adapter);
3414
3415                         fcoe_i = f->mask;
3416                 }
3417                 for (i = 0; i < f->indices; i++, fcoe_i++)
3418                         adapter->rx_ring[f->mask + i].reg_idx = fcoe_i;
3419                 ret = true;
3420         }
3421         return ret;
3422 }
3423
3424 #endif /* IXGBE_FCOE */
3425 /**
3426  * ixgbe_cache_ring_register - Descriptor ring to register mapping
3427  * @adapter: board private structure to initialize
3428  *
3429  * Once we know the feature-set enabled for the device, we'll cache
3430  * the register offset the descriptor ring is assigned to.
3431  *
3432  * Note, the order the various feature calls is important.  It must start with
3433  * the "most" features enabled at the same time, then trickle down to the
3434  * least amount of features turned on at once.
3435  **/
3436 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
3437 {
3438         /* start with default case */
3439         adapter->rx_ring[0].reg_idx = 0;
3440         adapter->tx_ring[0].reg_idx = 0;
3441
3442 #ifdef IXGBE_FCOE
3443         if (ixgbe_cache_ring_fcoe(adapter))
3444                 return;
3445
3446 #endif /* IXGBE_FCOE */
3447 #ifdef CONFIG_IXGBE_DCB
3448         if (ixgbe_cache_ring_dcb(adapter))
3449                 return;
3450
3451 #endif
3452         if (ixgbe_cache_ring_fdir(adapter))
3453                 return;
3454
3455         if (ixgbe_cache_ring_rss(adapter))
3456                 return;
3457 }
3458
3459 /**
3460  * ixgbe_alloc_queues - Allocate memory for all rings
3461  * @adapter: board private structure to initialize
3462  *
3463  * We allocate one ring per queue at run-time since we don't know the
3464  * number of queues at compile-time.  The polling_netdev array is
3465  * intended for Multiqueue, but should work fine with a single queue.
3466  **/
3467 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
3468 {
3469         int i;
3470
3471         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
3472                                    sizeof(struct ixgbe_ring), GFP_KERNEL);
3473         if (!adapter->tx_ring)
3474                 goto err_tx_ring_allocation;
3475
3476         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
3477                                    sizeof(struct ixgbe_ring), GFP_KERNEL);
3478         if (!adapter->rx_ring)
3479                 goto err_rx_ring_allocation;
3480
3481         for (i = 0; i < adapter->num_tx_queues; i++) {
3482                 adapter->tx_ring[i].count = adapter->tx_ring_count;
3483                 adapter->tx_ring[i].queue_index = i;
3484         }
3485
3486         for (i = 0; i < adapter->num_rx_queues; i++) {
3487                 adapter->rx_ring[i].count = adapter->rx_ring_count;
3488                 adapter->rx_ring[i].queue_index = i;
3489         }
3490
3491         ixgbe_cache_ring_register(adapter);
3492
3493         return 0;
3494
3495 err_rx_ring_allocation:
3496         kfree(adapter->tx_ring);
3497 err_tx_ring_allocation:
3498         return -ENOMEM;
3499 }
3500
3501 /**
3502  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
3503  * @adapter: board private structure to initialize
3504  *
3505  * Attempt to configure the interrupts using the best available
3506  * capabilities of the hardware and the kernel.
3507  **/
3508 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
3509 {
3510         struct ixgbe_hw *hw = &adapter->hw;
3511         int err = 0;
3512         int vector, v_budget;
3513
3514         /*
3515          * It's easy to be greedy for MSI-X vectors, but it really
3516          * doesn't do us much good if we have a lot more vectors
3517          * than CPU's.  So let's be conservative and only ask for
3518          * (roughly) twice the number of vectors as there are CPU's.
3519          */
3520         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
3521                        (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
3522
3523         /*
3524          * At the same time, hardware can only support a maximum of
3525          * hw.mac->max_msix_vectors vectors.  With features
3526          * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
3527          * descriptor queues supported by our device.  Thus, we cap it off in
3528          * those rare cases where the cpu count also exceeds our vector limit.
3529          */
3530         v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
3531
3532         /* A failure in MSI-X entry allocation isn't fatal, but it does
3533          * mean we disable MSI-X capabilities of the adapter. */
3534         adapter->msix_entries = kcalloc(v_budget,
3535                                         sizeof(struct msix_entry), GFP_KERNEL);
3536         if (adapter->msix_entries) {
3537                 for (vector = 0; vector < v_budget; vector++)
3538                         adapter->msix_entries[vector].entry = vector;
3539
3540                 ixgbe_acquire_msix_vectors(adapter, v_budget);
3541
3542                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3543                         goto out;
3544         }
3545
3546         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
3547         adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
3548         adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3549         adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3550         adapter->atr_sample_rate = 0;
3551         ixgbe_set_num_queues(adapter);
3552
3553         err = pci_enable_msi(adapter->pdev);
3554         if (!err) {
3555                 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
3556         } else {
3557                 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
3558                         "falling back to legacy.  Error: %d\n", err);
3559                 /* reset err */
3560                 err = 0;
3561         }
3562
3563 out:
3564         return err;
3565 }
3566
3567 /**
3568  * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
3569  * @adapter: board private structure to initialize
3570  *
3571  * We allocate one q_vector per queue interrupt.  If allocation fails we
3572  * return -ENOMEM.
3573  **/
3574 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
3575 {
3576         int q_idx, num_q_vectors;
3577         struct ixgbe_q_vector *q_vector;
3578         int napi_vectors;
3579         int (*poll)(struct napi_struct *, int);
3580
3581         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3582                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3583                 napi_vectors = adapter->num_rx_queues;
3584                 poll = &ixgbe_clean_rxtx_many;
3585         } else {
3586                 num_q_vectors = 1;
3587                 napi_vectors = 1;
3588                 poll = &ixgbe_poll;
3589         }
3590
3591         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3592                 q_vector = kzalloc(sizeof(struct ixgbe_q_vector), GFP_KERNEL);
3593                 if (!q_vector)
3594                         goto err_out;
3595                 q_vector->adapter = adapter;
3596                 q_vector->eitr = adapter->eitr_param;
3597                 q_vector->v_idx = q_idx;
3598                 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
3599                 adapter->q_vector[q_idx] = q_vector;
3600         }
3601
3602         return 0;
3603
3604 err_out:
3605         while (q_idx) {
3606                 q_idx--;
3607                 q_vector = adapter->q_vector[q_idx];
3608                 netif_napi_del(&q_vector->napi);
3609                 kfree(q_vector);
3610                 adapter->q_vector[q_idx] = NULL;
3611         }
3612         return -ENOMEM;
3613 }
3614
3615 /**
3616  * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
3617  * @adapter: board private structure to initialize
3618  *
3619  * This function frees the memory allocated to the q_vectors.  In addition if
3620  * NAPI is enabled it will delete any references to the NAPI struct prior
3621  * to freeing the q_vector.
3622  **/
3623 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
3624 {
3625         int q_idx, num_q_vectors;
3626
3627         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3628                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3629         else
3630                 num_q_vectors = 1;
3631
3632         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3633                 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
3634                 adapter->q_vector[q_idx] = NULL;
3635                 netif_napi_del(&q_vector->napi);
3636                 kfree(q_vector);
3637         }
3638 }
3639
3640 void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
3641 {
3642         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3643                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3644                 pci_disable_msix(adapter->pdev);
3645                 kfree(adapter->msix_entries);
3646                 adapter->msix_entries = NULL;
3647         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
3648                 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
3649                 pci_disable_msi(adapter->pdev);
3650         }
3651         return;
3652 }
3653
3654 /**
3655  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
3656  * @adapter: board private structure to initialize
3657  *
3658  * We determine which interrupt scheme to use based on...
3659  * - Kernel support (MSI, MSI-X)
3660  *   - which can be user-defined (via MODULE_PARAM)
3661  * - Hardware queue count (num_*_queues)
3662  *   - defined by miscellaneous hardware support/features (RSS, etc.)
3663  **/
3664 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
3665 {
3666         int err;
3667
3668         /* Number of supported queues */
3669         ixgbe_set_num_queues(adapter);
3670
3671         err = ixgbe_set_interrupt_capability(adapter);
3672         if (err) {
3673                 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
3674                 goto err_set_interrupt;
3675         }
3676
3677         err = ixgbe_alloc_q_vectors(adapter);
3678         if (err) {
3679                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queue "
3680                         "vectors\n");
3681                 goto err_alloc_q_vectors;
3682         }
3683
3684         err = ixgbe_alloc_queues(adapter);
3685         if (err) {
3686                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
3687                 goto err_alloc_queues;
3688         }
3689
3690         DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
3691                 "Tx Queue count = %u\n",
3692                 (adapter->num_rx_queues > 1) ? "Enabled" :
3693                 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
3694
3695         set_bit(__IXGBE_DOWN, &adapter->state);
3696
3697         return 0;
3698
3699 err_alloc_queues:
3700         ixgbe_free_q_vectors(adapter);
3701 err_alloc_q_vectors:
3702         ixgbe_reset_interrupt_capability(adapter);
3703 err_set_interrupt:
3704         return err;
3705 }
3706
3707 /**
3708  * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
3709  * @adapter: board private structure to clear interrupt scheme on
3710  *
3711  * We go through and clear interrupt specific resources and reset the structure
3712  * to pre-load conditions
3713  **/
3714 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
3715 {
3716         kfree(adapter->tx_ring);
3717         kfree(adapter->rx_ring);
3718         adapter->tx_ring = NULL;
3719         adapter->rx_ring = NULL;
3720
3721         ixgbe_free_q_vectors(adapter);
3722         ixgbe_reset_interrupt_capability(adapter);
3723 }
3724
3725 /**
3726  * ixgbe_sfp_timer - worker thread to find a missing module
3727  * @data: pointer to our adapter struct
3728  **/
3729 static void ixgbe_sfp_timer(unsigned long data)
3730 {
3731         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3732
3733         /*
3734          * Do the sfp_timer outside of interrupt context due to the
3735          * delays that sfp+ detection requires
3736          */
3737         schedule_work(&adapter->sfp_task);
3738 }
3739
3740 /**
3741  * ixgbe_sfp_task - worker thread to find a missing module
3742  * @work: pointer to work_struct containing our data
3743  **/
3744 static void ixgbe_sfp_task(struct work_struct *work)
3745 {
3746         struct ixgbe_adapter *adapter = container_of(work,
3747                                                      struct ixgbe_adapter,
3748                                                      sfp_task);
3749         struct ixgbe_hw *hw = &adapter->hw;
3750
3751         if ((hw->phy.type == ixgbe_phy_nl) &&
3752             (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
3753                 s32 ret = hw->phy.ops.identify_sfp(hw);
3754                 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
3755                         goto reschedule;
3756                 ret = hw->phy.ops.reset(hw);
3757                 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3758                         dev_err(&adapter->pdev->dev, "failed to initialize "
3759                                 "because an unsupported SFP+ module type "
3760                                 "was detected.\n"
3761                                 "Reload the driver after installing a "
3762                                 "supported module.\n");
3763                         unregister_netdev(adapter->netdev);
3764                 } else {
3765                         DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
3766                                 hw->phy.sfp_type);
3767                 }
3768                 /* don't need this routine any more */
3769                 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3770         }
3771         return;
3772 reschedule:
3773         if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
3774                 mod_timer(&adapter->sfp_timer,
3775                           round_jiffies(jiffies + (2 * HZ)));
3776 }
3777
3778 /**
3779  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
3780  * @adapter: board private structure to initialize
3781  *
3782  * ixgbe_sw_init initializes the Adapter private data structure.
3783  * Fields are initialized based on PCI device information and
3784  * OS network device settings (MTU size).
3785  **/
3786 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
3787 {
3788         struct ixgbe_hw *hw = &adapter->hw;
3789         struct pci_dev *pdev = adapter->pdev;
3790         unsigned int rss;
3791 #ifdef CONFIG_IXGBE_DCB
3792         int j;
3793         struct tc_configuration *tc;
3794 #endif
3795
3796         /* PCI config space info */
3797
3798         hw->vendor_id = pdev->vendor;
3799         hw->device_id = pdev->device;
3800         hw->revision_id = pdev->revision;
3801         hw->subsystem_vendor_id = pdev->subsystem_vendor;
3802         hw->subsystem_device_id = pdev->subsystem_device;
3803
3804         /* Set capability flags */
3805         rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
3806         adapter->ring_feature[RING_F_RSS].indices = rss;
3807         adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
3808         adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
3809         if (hw->mac.type == ixgbe_mac_82598EB) {
3810                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
3811                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
3812                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
3813         } else if (hw->mac.type == ixgbe_mac_82599EB) {
3814                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
3815                 adapter->flags |= IXGBE_FLAG2_RSC_CAPABLE;
3816                 adapter->flags |= IXGBE_FLAG2_RSC_ENABLED;
3817                 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
3818                 adapter->ring_feature[RING_F_FDIR].indices =
3819                                                          IXGBE_MAX_FDIR_INDICES;
3820                 adapter->atr_sample_rate = 20;
3821                 adapter->fdir_pballoc = 0;
3822 #ifdef IXGBE_FCOE
3823                 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
3824                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
3825                 adapter->ring_feature[RING_F_FCOE].indices = 0;
3826 #endif /* IXGBE_FCOE */
3827         }
3828
3829 #ifdef CONFIG_IXGBE_DCB
3830         /* Configure DCB traffic classes */
3831         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
3832                 tc = &adapter->dcb_cfg.tc_config[j];
3833                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
3834                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
3835                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
3836                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
3837                 tc->dcb_pfc = pfc_disabled;
3838         }
3839         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
3840         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
3841         adapter->dcb_cfg.rx_pba_cfg = pba_equal;
3842         adapter->dcb_cfg.pfc_mode_enable = false;
3843         adapter->dcb_cfg.round_robin_enable = false;
3844         adapter->dcb_set_bitmap = 0x00;
3845         ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
3846                            adapter->ring_feature[RING_F_DCB].indices);
3847
3848 #endif
3849
3850         /* default flow control settings */
3851         hw->fc.requested_mode = ixgbe_fc_full;
3852         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
3853 #ifdef CONFIG_DCB
3854         adapter->last_lfc_mode = hw->fc.current_mode;
3855 #endif
3856         hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
3857         hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
3858         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
3859         hw->fc.send_xon = true;
3860         hw->fc.disable_fc_autoneg = false;
3861
3862         /* enable itr by default in dynamic mode */
3863         adapter->itr_setting = 1;
3864         adapter->eitr_param = 20000;
3865
3866         /* set defaults for eitr in MegaBytes */
3867         adapter->eitr_low = 10;
3868         adapter->eitr_high = 20;
3869
3870         /* set default ring sizes */
3871         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
3872         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
3873
3874         /* initialize eeprom parameters */
3875         if (ixgbe_init_eeprom_params_generic(hw)) {
3876                 dev_err(&pdev->dev, "EEPROM initialization failed\n");
3877                 return -EIO;
3878         }
3879
3880         /* enable rx csum by default */
3881         adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
3882
3883         set_bit(__IXGBE_DOWN, &adapter->state);
3884
3885         return 0;
3886 }
3887
3888 /**
3889  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
3890  * @adapter: board private structure
3891  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
3892  *
3893  * Return 0 on success, negative on failure
3894  **/
3895 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
3896                              struct ixgbe_ring *tx_ring)
3897 {
3898         struct pci_dev *pdev = adapter->pdev;
3899         int size;
3900
3901         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3902         tx_ring->tx_buffer_info = vmalloc(size);
3903         if (!tx_ring->tx_buffer_info)
3904                 goto err;
3905         memset(tx_ring->tx_buffer_info, 0, size);
3906
3907         /* round up to nearest 4K */
3908         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3909         tx_ring->size = ALIGN(tx_ring->size, 4096);
3910
3911         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
3912                                              &tx_ring->dma);
3913         if (!tx_ring->desc)
3914                 goto err;
3915
3916         tx_ring->next_to_use = 0;
3917         tx_ring->next_to_clean = 0;
3918         tx_ring->work_limit = tx_ring->count;
3919         return 0;
3920
3921 err:
3922         vfree(tx_ring->tx_buffer_info);
3923         tx_ring->tx_buffer_info = NULL;
3924         DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
3925                             "descriptor ring\n");
3926         return -ENOMEM;
3927 }
3928
3929 /**
3930  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
3931  * @adapter: board private structure
3932  *
3933  * If this function returns with an error, then it's possible one or
3934  * more of the rings is populated (while the rest are not).  It is the
3935  * callers duty to clean those orphaned rings.
3936  *
3937  * Return 0 on success, negative on failure
3938  **/
3939 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
3940 {
3941         int i, err = 0;
3942
3943         for (i = 0; i < adapter->num_tx_queues; i++) {
3944                 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
3945                 if (!err)
3946                         continue;
3947                 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
3948                 break;
3949         }
3950
3951         return err;
3952 }
3953
3954 /**
3955  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
3956  * @adapter: board private structure
3957  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
3958  *
3959  * Returns 0 on success, negative on failure
3960  **/
3961 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
3962                              struct ixgbe_ring *rx_ring)
3963 {
3964         struct pci_dev *pdev = adapter->pdev;
3965         int size;
3966
3967         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3968         rx_ring->rx_buffer_info = vmalloc(size);
3969         if (!rx_ring->rx_buffer_info) {
3970                 DPRINTK(PROBE, ERR,
3971                         "vmalloc allocation failed for the rx desc ring\n");
3972                 goto alloc_failed;
3973         }
3974         memset(rx_ring->rx_buffer_info, 0, size);
3975
3976         /* Round up to nearest 4K */
3977         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3978         rx_ring->size = ALIGN(rx_ring->size, 4096);
3979
3980         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
3981
3982         if (!rx_ring->desc) {
3983                 DPRINTK(PROBE, ERR,
3984                         "Memory allocation failed for the rx desc ring\n");
3985                 vfree(rx_ring->rx_buffer_info);
3986                 goto alloc_failed;
3987         }
3988
3989         rx_ring->next_to_clean = 0;
3990         rx_ring->next_to_use = 0;
3991
3992         return 0;
3993
3994 alloc_failed:
3995         return -ENOMEM;
3996 }
3997
3998 /**
3999  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4000  * @adapter: board private structure
4001  *
4002  * If this function returns with an error, then it's possible one or
4003  * more of the rings is populated (while the rest are not).  It is the
4004  * callers duty to clean those orphaned rings.
4005  *
4006  * Return 0 on success, negative on failure
4007  **/
4008
4009 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4010 {
4011         int i, err = 0;
4012
4013         for (i = 0; i < adapter->num_rx_queues; i++) {
4014                 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
4015                 if (!err)
4016                         continue;
4017                 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
4018                 break;
4019         }
4020
4021         return err;
4022 }
4023
4024 /**
4025  * ixgbe_free_tx_resources - Free Tx Resources per Queue
4026  * @adapter: board private structure
4027  * @tx_ring: Tx descriptor ring for a specific queue
4028  *
4029  * Free all transmit software resources
4030  **/
4031 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4032                              struct ixgbe_ring *tx_ring)
4033 {
4034         struct pci_dev *pdev = adapter->pdev;
4035
4036         ixgbe_clean_tx_ring(adapter, tx_ring);
4037
4038         vfree(tx_ring->tx_buffer_info);
4039         tx_ring->tx_buffer_info = NULL;
4040
4041         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
4042
4043         tx_ring->desc = NULL;
4044 }
4045
4046 /**
4047  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4048  * @adapter: board private structure
4049  *
4050  * Free all transmit software resources
4051  **/
4052 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4053 {
4054         int i;
4055
4056         for (i = 0; i < adapter->num_tx_queues; i++)
4057                 if (adapter->tx_ring[i].desc)
4058                         ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
4059 }
4060
4061 /**
4062  * ixgbe_free_rx_resources - Free Rx Resources
4063  * @adapter: board private structure
4064  * @rx_ring: ring to clean the resources from
4065  *
4066  * Free all receive software resources
4067  **/
4068 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
4069                              struct ixgbe_ring *rx_ring)
4070 {
4071         struct pci_dev *pdev = adapter->pdev;
4072
4073         ixgbe_clean_rx_ring(adapter, rx_ring);
4074
4075         vfree(rx_ring->rx_buffer_info);
4076         rx_ring->rx_buffer_info = NULL;
4077
4078         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
4079
4080         rx_ring->desc = NULL;
4081 }
4082
4083 /**
4084  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4085  * @adapter: board private structure
4086  *
4087  * Free all receive software resources
4088  **/
4089 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4090 {
4091         int i;
4092
4093         for (i = 0; i < adapter->num_rx_queues; i++)
4094                 if (adapter->rx_ring[i].desc)
4095                         ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
4096 }
4097
4098 /**
4099  * ixgbe_change_mtu - Change the Maximum Transfer Unit
4100  * @netdev: network interface device structure
4101  * @new_mtu: new value for maximum frame size
4102  *
4103  * Returns 0 on success, negative on failure
4104  **/
4105 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4106 {
4107         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4108         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4109
4110         /* MTU < 68 is an error and causes problems on some kernels */
4111         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4112                 return -EINVAL;
4113
4114         DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
4115                 netdev->mtu, new_mtu);
4116         /* must set new MTU before calling down or up */
4117         netdev->mtu = new_mtu;
4118
4119         if (netif_running(netdev))
4120                 ixgbe_reinit_locked(adapter);
4121
4122         return 0;
4123 }
4124
4125 /**
4126  * ixgbe_open - Called when a network interface is made active
4127  * @netdev: network interface device structure
4128  *
4129  * Returns 0 on success, negative value on failure
4130  *
4131  * The open entry point is called when a network interface is made
4132  * active by the system (IFF_UP).  At this point all resources needed
4133  * for transmit and receive operations are allocated, the interrupt
4134  * handler is registered with the OS, the watchdog timer is started,
4135  * and the stack is notified that the interface is ready.
4136  **/
4137 static int ixgbe_open(struct net_device *netdev)
4138 {
4139         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4140         int err;
4141
4142         /* disallow open during test */
4143         if (test_bit(__IXGBE_TESTING, &adapter->state))
4144                 return -EBUSY;
4145
4146         netif_carrier_off(netdev);
4147
4148         /* allocate transmit descriptors */
4149         err = ixgbe_setup_all_tx_resources(adapter);
4150         if (err)
4151                 goto err_setup_tx;
4152
4153         /* allocate receive descriptors */
4154         err = ixgbe_setup_all_rx_resources(adapter);
4155         if (err)
4156                 goto err_setup_rx;
4157
4158         ixgbe_configure(adapter);
4159
4160         err = ixgbe_request_irq(adapter);
4161         if (err)
4162                 goto err_req_irq;
4163
4164         err = ixgbe_up_complete(adapter);
4165         if (err)
4166                 goto err_up;
4167
4168         netif_tx_start_all_queues(netdev);
4169
4170         return 0;
4171
4172 err_up:
4173         ixgbe_release_hw_control(adapter);
4174         ixgbe_free_irq(adapter);
4175 err_req_irq:
4176 err_setup_rx:
4177         ixgbe_free_all_rx_resources(adapter);
4178 err_setup_tx:
4179         ixgbe_free_all_tx_resources(adapter);
4180         ixgbe_reset(adapter);
4181
4182         return err;
4183 }
4184
4185 /**
4186  * ixgbe_close - Disables a network interface
4187  * @netdev: network interface device structure
4188  *
4189  * Returns 0, this is not allowed to fail
4190  *
4191  * The close entry point is called when an interface is de-activated
4192  * by the OS.  The hardware is still under the drivers control, but
4193  * needs to be disabled.  A global MAC reset is issued to stop the
4194  * hardware, and all transmit and receive resources are freed.
4195  **/
4196 static int ixgbe_close(struct net_device *netdev)
4197 {
4198         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4199
4200         ixgbe_down(adapter);
4201         ixgbe_free_irq(adapter);
4202
4203         ixgbe_free_all_tx_resources(adapter);
4204         ixgbe_free_all_rx_resources(adapter);
4205
4206         ixgbe_release_hw_control(adapter);
4207
4208         return 0;
4209 }
4210
4211 #ifdef CONFIG_PM
4212 static int ixgbe_resume(struct pci_dev *pdev)
4213 {
4214         struct net_device *netdev = pci_get_drvdata(pdev);
4215         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4216         u32 err;
4217
4218         pci_set_power_state(pdev, PCI_D0);
4219         pci_restore_state(pdev);
4220
4221         err = pci_enable_device_mem(pdev);
4222         if (err) {
4223                 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
4224                                 "suspend\n");
4225                 return err;
4226         }
4227         pci_set_master(pdev);
4228
4229         pci_wake_from_d3(pdev, false);
4230
4231         err = ixgbe_init_interrupt_scheme(adapter);
4232         if (err) {
4233                 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
4234                                 "device\n");
4235                 return err;
4236         }
4237
4238         ixgbe_reset(adapter);
4239
4240         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4241
4242         if (netif_running(netdev)) {
4243                 err = ixgbe_open(adapter->netdev);
4244                 if (err)
4245                         return err;
4246         }
4247
4248         netif_device_attach(netdev);
4249
4250         return 0;
4251 }
4252 #endif /* CONFIG_PM */
4253
4254 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4255 {
4256         struct net_device *netdev = pci_get_drvdata(pdev);
4257         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4258         struct ixgbe_hw *hw = &adapter->hw;
4259         u32 ctrl, fctrl;
4260         u32 wufc = adapter->wol;
4261 #ifdef CONFIG_PM
4262         int retval = 0;
4263 #endif
4264
4265         netif_device_detach(netdev);
4266
4267         if (netif_running(netdev)) {
4268                 ixgbe_down(adapter);
4269                 ixgbe_free_irq(adapter);
4270                 ixgbe_free_all_tx_resources(adapter);
4271                 ixgbe_free_all_rx_resources(adapter);
4272         }
4273         ixgbe_clear_interrupt_scheme(adapter);
4274
4275 #ifdef CONFIG_PM
4276         retval = pci_save_state(pdev);
4277         if (retval)
4278                 return retval;
4279
4280 #endif
4281         if (wufc) {
4282                 ixgbe_set_rx_mode(netdev);
4283
4284                 /* turn on all-multi mode if wake on multicast is enabled */
4285                 if (wufc & IXGBE_WUFC_MC) {
4286                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4287                         fctrl |= IXGBE_FCTRL_MPE;
4288                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4289                 }
4290
4291                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4292                 ctrl |= IXGBE_CTRL_GIO_DIS;
4293                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4294
4295                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4296         } else {
4297                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4298                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4299         }
4300
4301         if (wufc && hw->mac.type == ixgbe_mac_82599EB)
4302                 pci_wake_from_d3(pdev, true);
4303         else
4304                 pci_wake_from_d3(pdev, false);
4305
4306         *enable_wake = !!wufc;
4307
4308         ixgbe_release_hw_control(adapter);
4309
4310         pci_disable_device(pdev);
4311
4312         return 0;
4313 }
4314
4315 #ifdef CONFIG_PM
4316 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4317 {
4318         int retval;
4319         bool wake;
4320
4321         retval = __ixgbe_shutdown(pdev, &wake);
4322         if (retval)
4323                 return retval;
4324
4325         if (wake) {
4326                 pci_prepare_to_sleep(pdev);
4327         } else {
4328                 pci_wake_from_d3(pdev, false);
4329                 pci_set_power_state(pdev, PCI_D3hot);
4330         }
4331
4332         return 0;
4333 }
4334 #endif /* CONFIG_PM */
4335
4336 static void ixgbe_shutdown(struct pci_dev *pdev)
4337 {
4338         bool wake;
4339
4340         __ixgbe_shutdown(pdev, &wake);
4341
4342         if (system_state == SYSTEM_POWER_OFF) {
4343                 pci_wake_from_d3(pdev, wake);
4344                 pci_set_power_state(pdev, PCI_D3hot);
4345         }
4346 }
4347
4348 /**
4349  * ixgbe_update_stats - Update the board statistics counters.
4350  * @adapter: board private structure
4351  **/
4352 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4353 {
4354         struct ixgbe_hw *hw = &adapter->hw;
4355         u64 total_mpc = 0;
4356         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4357
4358         if (hw->mac.type == ixgbe_mac_82599EB) {
4359                 u64 rsc_count = 0;
4360                 for (i = 0; i < 16; i++)
4361                         adapter->hw_rx_no_dma_resources +=
4362                                              IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4363                 for (i = 0; i < adapter->num_rx_queues; i++)
4364                         rsc_count += adapter->rx_ring[i].rsc_count;
4365                 adapter->rsc_count = rsc_count;
4366         }
4367
4368         adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
4369         for (i = 0; i < 8; i++) {
4370                 /* for packet buffers not used, the register should read 0 */
4371                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
4372                 missed_rx += mpc;
4373                 adapter->stats.mpc[i] += mpc;
4374                 total_mpc += adapter->stats.mpc[i];
4375                 if (hw->mac.type == ixgbe_mac_82598EB)
4376                         adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
4377                 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
4378                 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
4379                 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
4380                 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
4381                 if (hw->mac.type == ixgbe_mac_82599EB) {
4382                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4383                                                             IXGBE_PXONRXCNT(i));
4384                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4385                                                            IXGBE_PXOFFRXCNT(i));
4386                         adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4387                 } else {
4388                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4389                                                               IXGBE_PXONRXC(i));
4390                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4391                                                              IXGBE_PXOFFRXC(i));
4392                 }
4393                 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
4394                                                             IXGBE_PXONTXC(i));
4395                 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
4396                                                              IXGBE_PXOFFTXC(i));
4397         }
4398         adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
4399         /* work around hardware counting issue */
4400         adapter->stats.gprc -= missed_rx;
4401
4402         /* 82598 hardware only has a 32 bit counter in the high register */
4403         if (hw->mac.type == ixgbe_mac_82599EB) {
4404                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
4405                 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
4406                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
4407                 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
4408                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
4409                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
4410                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
4411                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
4412                 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
4413                 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
4414 #ifdef IXGBE_FCOE
4415                 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
4416                 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
4417                 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
4418                 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
4419                 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
4420                 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
4421 #endif /* IXGBE_FCOE */
4422         } else {
4423                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
4424                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
4425                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
4426                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
4427                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
4428         }
4429         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
4430         adapter->stats.bprc += bprc;
4431         adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
4432         if (hw->mac.type == ixgbe_mac_82598EB)
4433                 adapter->stats.mprc -= bprc;
4434         adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
4435         adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
4436         adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
4437         adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
4438         adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
4439         adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
4440         adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
4441         adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
4442         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
4443         adapter->stats.lxontxc += lxon;
4444         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
4445         adapter->stats.lxofftxc += lxoff;
4446         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4447         adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
4448         adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
4449         /*
4450          * 82598 errata - tx of flow control packets is included in tx counters
4451          */
4452         xon_off_tot = lxon + lxoff;
4453         adapter->stats.gptc -= xon_off_tot;
4454         adapter->stats.mptc -= xon_off_tot;
4455         adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
4456         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4457         adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
4458         adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
4459         adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
4460         adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
4461         adapter->stats.ptc64 -= xon_off_tot;
4462         adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
4463         adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
4464         adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
4465         adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
4466         adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
4467         adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
4468
4469         /* Fill out the OS statistics structure */
4470         adapter->net_stats.multicast = adapter->stats.mprc;
4471
4472         /* Rx Errors */
4473         adapter->net_stats.rx_errors = adapter->stats.crcerrs +
4474                                        adapter->stats.rlec;
4475         adapter->net_stats.rx_dropped = 0;
4476         adapter->net_stats.rx_length_errors = adapter->stats.rlec;
4477         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
4478         adapter->net_stats.rx_missed_errors = total_mpc;
4479 }
4480
4481 /**
4482  * ixgbe_watchdog - Timer Call-back
4483  * @data: pointer to adapter cast into an unsigned long
4484  **/
4485 static void ixgbe_watchdog(unsigned long data)
4486 {
4487         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4488         struct ixgbe_hw *hw = &adapter->hw;
4489         u64 eics = 0;
4490         int i;
4491
4492         /*
4493          *  Do the watchdog outside of interrupt context due to the lovely
4494          * delays that some of the newer hardware requires
4495          */
4496
4497         if (test_bit(__IXGBE_DOWN, &adapter->state))
4498                 goto watchdog_short_circuit;
4499
4500         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
4501                 /*
4502                  * for legacy and MSI interrupts don't set any bits
4503                  * that are enabled for EIAM, because this operation
4504                  * would set *both* EIMS and EICS for any bit in EIAM
4505                  */
4506                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
4507                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
4508                 goto watchdog_reschedule;
4509         }
4510
4511         /* get one bit for every active tx/rx interrupt vector */
4512         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
4513                 struct ixgbe_q_vector *qv = adapter->q_vector[i];
4514                 if (qv->rxr_count || qv->txr_count)
4515                         eics |= ((u64)1 << i);
4516         }
4517
4518         /* Cause software interrupt to ensure rx rings are cleaned */
4519         ixgbe_irq_rearm_queues(adapter, eics);
4520
4521 watchdog_reschedule:
4522         /* Reset the timer */
4523         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
4524
4525 watchdog_short_circuit:
4526         schedule_work(&adapter->watchdog_task);
4527 }
4528
4529 /**
4530  * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
4531  * @work: pointer to work_struct containing our data
4532  **/
4533 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
4534 {
4535         struct ixgbe_adapter *adapter = container_of(work,
4536                                                      struct ixgbe_adapter,
4537                                                      multispeed_fiber_task);
4538         struct ixgbe_hw *hw = &adapter->hw;
4539         u32 autoneg;
4540
4541         adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
4542         autoneg = hw->phy.autoneg_advertised;
4543         if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
4544                 hw->mac.ops.get_link_capabilities(hw, &autoneg,
4545                                                   &hw->mac.autoneg);
4546         if (hw->mac.ops.setup_link_speed)
4547                 hw->mac.ops.setup_link_speed(hw, autoneg, true, true);
4548         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4549         adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
4550 }
4551
4552 /**
4553  * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
4554  * @work: pointer to work_struct containing our data
4555  **/
4556 static void ixgbe_sfp_config_module_task(struct work_struct *work)
4557 {
4558         struct ixgbe_adapter *adapter = container_of(work,
4559                                                      struct ixgbe_adapter,
4560                                                      sfp_config_module_task);
4561         struct ixgbe_hw *hw = &adapter->hw;
4562         u32 err;
4563
4564         adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
4565
4566         /* Time for electrical oscillations to settle down */
4567         msleep(100);
4568         err = hw->phy.ops.identify_sfp(hw);
4569
4570         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4571                 dev_err(&adapter->pdev->dev, "failed to initialize because "
4572                         "an unsupported SFP+ module type was detected.\n"
4573                         "Reload the driver after installing a supported "
4574                         "module.\n");
4575                 unregister_netdev(adapter->netdev);
4576                 return;
4577         }
4578         hw->mac.ops.setup_sfp(hw);
4579
4580         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
4581                 /* This will also work for DA Twinax connections */
4582                 schedule_work(&adapter->multispeed_fiber_task);
4583         adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
4584 }
4585
4586 /**
4587  * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
4588  * @work: pointer to work_struct containing our data
4589  **/
4590 static void ixgbe_fdir_reinit_task(struct work_struct *work)
4591 {
4592         struct ixgbe_adapter *adapter = container_of(work,
4593                                                      struct ixgbe_adapter,
4594                                                      fdir_reinit_task);
4595         struct ixgbe_hw *hw = &adapter->hw;
4596         int i;
4597
4598         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
4599                 for (i = 0; i < adapter->num_tx_queues; i++)
4600                         set_bit(__IXGBE_FDIR_INIT_DONE,
4601                                 &(adapter->tx_ring[i].reinit_state));
4602         } else {
4603                 DPRINTK(PROBE, ERR, "failed to finish FDIR re-initialization, "
4604                         "ignored adding FDIR ATR filters \n");
4605         }
4606         /* Done FDIR Re-initialization, enable transmits */
4607         netif_tx_start_all_queues(adapter->netdev);
4608 }
4609
4610 /**
4611  * ixgbe_watchdog_task - worker thread to bring link up
4612  * @work: pointer to work_struct containing our data
4613  **/
4614 static void ixgbe_watchdog_task(struct work_struct *work)
4615 {
4616         struct ixgbe_adapter *adapter = container_of(work,
4617                                                      struct ixgbe_adapter,
4618                                                      watchdog_task);
4619         struct net_device *netdev = adapter->netdev;
4620         struct ixgbe_hw *hw = &adapter->hw;
4621         u32 link_speed = adapter->link_speed;
4622         bool link_up = adapter->link_up;
4623         int i;
4624         struct ixgbe_ring *tx_ring;
4625         int some_tx_pending = 0;
4626
4627         adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
4628
4629         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
4630                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
4631                 if (link_up) {
4632 #ifdef CONFIG_DCB
4633                         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4634                                 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
4635                                         hw->mac.ops.fc_enable(hw, i);
4636                         } else {
4637                                 hw->mac.ops.fc_enable(hw, 0);
4638                         }
4639 #else
4640                         hw->mac.ops.fc_enable(hw, 0);
4641 #endif
4642                 }
4643
4644                 if (link_up ||
4645                     time_after(jiffies, (adapter->link_check_timeout +
4646                                          IXGBE_TRY_LINK_TIMEOUT))) {
4647                         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4648                         IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
4649                 }
4650                 adapter->link_up = link_up;
4651                 adapter->link_speed = link_speed;
4652         }
4653
4654         if (link_up) {
4655                 if (!netif_carrier_ok(netdev)) {
4656                         bool flow_rx, flow_tx;
4657
4658                         if (hw->mac.type == ixgbe_mac_82599EB) {
4659                                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
4660                                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
4661                                 flow_rx = (mflcn & IXGBE_MFLCN_RFCE);
4662                                 flow_tx = (fccfg & IXGBE_FCCFG_TFCE_802_3X);
4663                         } else {
4664                                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4665                                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
4666                                 flow_rx = (frctl & IXGBE_FCTRL_RFCE);
4667                                 flow_tx = (rmcs & IXGBE_RMCS_TFCE_802_3X);
4668                         }
4669
4670                         printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
4671                                "Flow Control: %s\n",
4672                                netdev->name,
4673                                (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
4674                                 "10 Gbps" :
4675                                 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
4676                                  "1 Gbps" : "unknown speed")),
4677                                ((flow_rx && flow_tx) ? "RX/TX" :
4678                                 (flow_rx ? "RX" :
4679                                 (flow_tx ? "TX" : "None"))));
4680
4681                         netif_carrier_on(netdev);
4682                 } else {
4683                         /* Force detection of hung controller */
4684                         adapter->detect_tx_hung = true;
4685                 }
4686         } else {
4687                 adapter->link_up = false;
4688                 adapter->link_speed = 0;
4689                 if (netif_carrier_ok(netdev)) {
4690                         printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
4691                                netdev->name);
4692                         netif_carrier_off(netdev);
4693                 }
4694         }
4695
4696         if (!netif_carrier_ok(netdev)) {
4697                 for (i = 0; i < adapter->num_tx_queues; i++) {
4698                         tx_ring = &adapter->tx_ring[i];
4699                         if (tx_ring->next_to_use != tx_ring->next_to_clean) {
4700                                 some_tx_pending = 1;
4701                                 break;
4702                         }
4703                 }
4704
4705                 if (some_tx_pending) {
4706                         /* We've lost link, so the controller stops DMA,
4707                          * but we've got queued Tx work that's never going
4708                          * to get done, so reset controller to flush Tx.
4709                          * (Do the reset outside of interrupt context).
4710                          */
4711                          schedule_work(&adapter->reset_task);
4712                 }
4713         }
4714
4715         ixgbe_update_stats(adapter);
4716         adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
4717 }
4718
4719 static int ixgbe_tso(struct ixgbe_adapter *adapter,
4720                      struct ixgbe_ring *tx_ring, struct sk_buff *skb,
4721                      u32 tx_flags, u8 *hdr_len)
4722 {
4723         struct ixgbe_adv_tx_context_desc *context_desc;
4724         unsigned int i;
4725         int err;
4726         struct ixgbe_tx_buffer *tx_buffer_info;
4727         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
4728         u32 mss_l4len_idx, l4len;
4729
4730         if (skb_is_gso(skb)) {
4731                 if (skb_header_cloned(skb)) {
4732                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4733                         if (err)
4734                                 return err;
4735                 }
4736                 l4len = tcp_hdrlen(skb);
4737                 *hdr_len += l4len;
4738
4739                 if (skb->protocol == htons(ETH_P_IP)) {
4740                         struct iphdr *iph = ip_hdr(skb);
4741                         iph->tot_len = 0;
4742                         iph->check = 0;
4743                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4744                                                                  iph->daddr, 0,
4745                                                                  IPPROTO_TCP,
4746                                                                  0);
4747                         adapter->hw_tso_ctxt++;
4748                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
4749                         ipv6_hdr(skb)->payload_len = 0;
4750                         tcp_hdr(skb)->check =
4751                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4752                                              &ipv6_hdr(skb)->daddr,
4753                                              0, IPPROTO_TCP, 0);
4754                         adapter->hw_tso6_ctxt++;
4755                 }
4756
4757                 i = tx_ring->next_to_use;
4758
4759                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4760                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4761
4762                 /* VLAN MACLEN IPLEN */
4763                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4764                         vlan_macip_lens |=
4765                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4766                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
4767                                     IXGBE_ADVTXD_MACLEN_SHIFT);
4768                 *hdr_len += skb_network_offset(skb);
4769                 vlan_macip_lens |=
4770                     (skb_transport_header(skb) - skb_network_header(skb));
4771                 *hdr_len +=
4772                     (skb_transport_header(skb) - skb_network_header(skb));
4773                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4774                 context_desc->seqnum_seed = 0;
4775
4776                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4777                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
4778                                    IXGBE_ADVTXD_DTYP_CTXT);
4779
4780                 if (skb->protocol == htons(ETH_P_IP))
4781                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4782                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
4783                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4784
4785                 /* MSS L4LEN IDX */
4786                 mss_l4len_idx =
4787                     (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
4788                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
4789                 /* use index 1 for TSO */
4790                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4791                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4792
4793                 tx_buffer_info->time_stamp = jiffies;
4794                 tx_buffer_info->next_to_watch = i;
4795
4796                 i++;
4797                 if (i == tx_ring->count)
4798                         i = 0;
4799                 tx_ring->next_to_use = i;
4800
4801                 return true;
4802         }
4803         return false;
4804 }
4805
4806 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
4807                           struct ixgbe_ring *tx_ring,
4808                           struct sk_buff *skb, u32 tx_flags)
4809 {
4810         struct ixgbe_adv_tx_context_desc *context_desc;
4811         unsigned int i;
4812         struct ixgbe_tx_buffer *tx_buffer_info;
4813         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
4814
4815         if (skb->ip_summed == CHECKSUM_PARTIAL ||
4816             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
4817                 i = tx_ring->next_to_use;
4818                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4819                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4820
4821                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4822                         vlan_macip_lens |=
4823                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4824                 vlan_macip_lens |= (skb_network_offset(skb) <<
4825                                     IXGBE_ADVTXD_MACLEN_SHIFT);
4826                 if (skb->ip_summed == CHECKSUM_PARTIAL)
4827                         vlan_macip_lens |= (skb_transport_header(skb) -
4828                                             skb_network_header(skb));
4829
4830                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4831                 context_desc->seqnum_seed = 0;
4832
4833                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
4834                                     IXGBE_ADVTXD_DTYP_CTXT);
4835
4836                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
4837                         switch (skb->protocol) {
4838                         case cpu_to_be16(ETH_P_IP):
4839                                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4840                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4841                                         type_tucmd_mlhl |=
4842                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4843                                 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
4844                                         type_tucmd_mlhl |=
4845                                                 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
4846                                 break;
4847                         case cpu_to_be16(ETH_P_IPV6):
4848                                 /* XXX what about other V6 headers?? */
4849                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4850                                         type_tucmd_mlhl |=
4851                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4852                                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
4853                                         type_tucmd_mlhl |=
4854                                                 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
4855                                 break;
4856                         default:
4857                                 if (unlikely(net_ratelimit())) {
4858                                         DPRINTK(PROBE, WARNING,
4859                                          "partial checksum but proto=%x!\n",
4860                                          skb->protocol);
4861                                 }
4862                                 break;
4863                         }
4864                 }
4865
4866                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4867                 /* use index zero for tx checksum offload */
4868                 context_desc->mss_l4len_idx = 0;
4869
4870                 tx_buffer_info->time_stamp = jiffies;
4871                 tx_buffer_info->next_to_watch = i;
4872
4873                 adapter->hw_csum_tx_good++;
4874                 i++;
4875                 if (i == tx_ring->count)
4876                         i = 0;
4877                 tx_ring->next_to_use = i;
4878
4879                 return true;
4880         }
4881
4882         return false;
4883 }
4884
4885 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
4886                         struct ixgbe_ring *tx_ring,
4887                         struct sk_buff *skb, u32 tx_flags,
4888                         unsigned int first)
4889 {
4890         struct ixgbe_tx_buffer *tx_buffer_info;
4891         unsigned int len;
4892         unsigned int total = skb->len;
4893         unsigned int offset = 0, size, count = 0, i;
4894         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
4895         unsigned int f;
4896         dma_addr_t *map;
4897
4898         i = tx_ring->next_to_use;
4899
4900         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
4901                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
4902                 return 0;
4903         }
4904
4905         map = skb_shinfo(skb)->dma_maps;
4906
4907         if (tx_flags & IXGBE_TX_FLAGS_FCOE)
4908                 /* excluding fcoe_crc_eof for FCoE */
4909                 total -= sizeof(struct fcoe_crc_eof);
4910
4911         len = min(skb_headlen(skb), total);
4912         while (len) {
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 = skb_shinfo(skb)->dma_head + 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 (len) {
4927                         i++;
4928                         if (i == tx_ring->count)
4929                                 i = 0;
4930                 }
4931         }
4932
4933         for (f = 0; f < nr_frags; f++) {
4934                 struct skb_frag_struct *frag;
4935
4936                 frag = &skb_shinfo(skb)->frags[f];
4937                 len = min((unsigned int)frag->size, total);
4938                 offset = 0;
4939
4940                 while (len) {
4941                         i++;
4942                         if (i == tx_ring->count)
4943                                 i = 0;
4944
4945                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
4946                         size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4947
4948                         tx_buffer_info->length = size;
4949                         tx_buffer_info->dma = map[f] + offset;
4950                         tx_buffer_info->time_stamp = jiffies;
4951                         tx_buffer_info->next_to_watch = i;
4952
4953                         len -= size;
4954                         total -= size;
4955                         offset += size;
4956                         count++;
4957                 }
4958                 if (total == 0)
4959                         break;
4960         }
4961
4962         tx_ring->tx_buffer_info[i].skb = skb;
4963         tx_ring->tx_buffer_info[first].next_to_watch = i;
4964
4965         return count;
4966 }
4967
4968 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
4969                            struct ixgbe_ring *tx_ring,
4970                            int tx_flags, int count, u32 paylen, u8 hdr_len)
4971 {
4972         union ixgbe_adv_tx_desc *tx_desc = NULL;
4973         struct ixgbe_tx_buffer *tx_buffer_info;
4974         u32 olinfo_status = 0, cmd_type_len = 0;
4975         unsigned int i;
4976         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
4977
4978         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
4979
4980         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
4981
4982         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4983                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
4984
4985         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
4986                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
4987
4988                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
4989                                  IXGBE_ADVTXD_POPTS_SHIFT;
4990
4991                 /* use index 1 context for tso */
4992                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4993                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
4994                         olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
4995                                          IXGBE_ADVTXD_POPTS_SHIFT;
4996
4997         } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
4998                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
4999                                  IXGBE_ADVTXD_POPTS_SHIFT;
5000
5001         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5002                 olinfo_status |= IXGBE_ADVTXD_CC;
5003                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5004                 if (tx_flags & IXGBE_TX_FLAGS_FSO)
5005                         cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5006         }
5007
5008         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
5009
5010         i = tx_ring->next_to_use;
5011         while (count--) {
5012                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5013                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
5014                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
5015                 tx_desc->read.cmd_type_len =
5016                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
5017                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5018                 i++;
5019                 if (i == tx_ring->count)
5020                         i = 0;
5021         }
5022
5023         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
5024
5025         /*
5026          * Force memory writes to complete before letting h/w
5027          * know there are new descriptors to fetch.  (Only
5028          * applicable for weak-ordered memory model archs,
5029          * such as IA-64).
5030          */
5031         wmb();
5032
5033         tx_ring->next_to_use = i;
5034         writel(i, adapter->hw.hw_addr + tx_ring->tail);
5035 }
5036
5037 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5038                       int queue, u32 tx_flags)
5039 {
5040         /* Right now, we support IPv4 only */
5041         struct ixgbe_atr_input atr_input;
5042         struct tcphdr *th;
5043         struct udphdr *uh;
5044         struct iphdr *iph = ip_hdr(skb);
5045         struct ethhdr *eth = (struct ethhdr *)skb->data;
5046         u16 vlan_id, src_port, dst_port, flex_bytes;
5047         u32 src_ipv4_addr, dst_ipv4_addr;
5048         u8 l4type = 0;
5049
5050         /* check if we're UDP or TCP */
5051         if (iph->protocol == IPPROTO_TCP) {
5052                 th = tcp_hdr(skb);
5053                 src_port = th->source;
5054                 dst_port = th->dest;
5055                 l4type |= IXGBE_ATR_L4TYPE_TCP;
5056                 /* l4type IPv4 type is 0, no need to assign */
5057         } else if(iph->protocol == IPPROTO_UDP) {
5058                 uh = udp_hdr(skb);
5059                 src_port = uh->source;
5060                 dst_port = uh->dest;
5061                 l4type |= IXGBE_ATR_L4TYPE_UDP;
5062                 /* l4type IPv4 type is 0, no need to assign */
5063         } else {
5064                 /* Unsupported L4 header, just bail here */
5065                 return;
5066         }
5067
5068         memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
5069
5070         vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
5071                    IXGBE_TX_FLAGS_VLAN_SHIFT;
5072         src_ipv4_addr = iph->saddr;
5073         dst_ipv4_addr = iph->daddr;
5074         flex_bytes = eth->h_proto;
5075
5076         ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
5077         ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
5078         ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
5079         ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
5080         ixgbe_atr_set_l4type_82599(&atr_input, l4type);
5081         /* src and dst are inverted, think how the receiver sees them */
5082         ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
5083         ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
5084
5085         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
5086         ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
5087 }
5088
5089 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
5090                                  struct ixgbe_ring *tx_ring, int size)
5091 {
5092         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5093
5094         netif_stop_subqueue(netdev, tx_ring->queue_index);
5095         /* Herbert's original patch had:
5096          *  smp_mb__after_netif_stop_queue();
5097          * but since that doesn't exist yet, just open code it. */
5098         smp_mb();
5099
5100         /* We need to check again in a case another CPU has just
5101          * made room available. */
5102         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
5103                 return -EBUSY;
5104
5105         /* A reprieve! - use start_queue because it doesn't call schedule */
5106         netif_start_subqueue(netdev, tx_ring->queue_index);
5107         ++adapter->restart_queue;
5108         return 0;
5109 }
5110
5111 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
5112                               struct ixgbe_ring *tx_ring, int size)
5113 {
5114         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
5115                 return 0;
5116         return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
5117 }
5118
5119 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
5120 {
5121         struct ixgbe_adapter *adapter = netdev_priv(dev);
5122
5123         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
5124                 return smp_processor_id();
5125
5126         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5127                 return 0;  /* All traffic should default to class 0 */
5128
5129         return skb_tx_hash(dev, skb);
5130 }
5131
5132 static int ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
5133 {
5134         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5135         struct ixgbe_ring *tx_ring;
5136         unsigned int first;
5137         unsigned int tx_flags = 0;
5138         u8 hdr_len = 0;
5139         int r_idx = 0, tso;
5140         int count = 0;
5141         unsigned int f;
5142
5143         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
5144                 tx_flags |= vlan_tx_tag_get(skb);
5145                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5146                         tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
5147                         tx_flags |= (skb->queue_mapping << 13);
5148                 }
5149                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5150                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5151         } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5152                 if (skb->priority != TC_PRIO_CONTROL) {
5153                         tx_flags |= (skb->queue_mapping << 13);
5154                         tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5155                         tx_flags |= IXGBE_TX_FLAGS_VLAN;
5156                 } else {
5157                         skb->queue_mapping =
5158                                 adapter->ring_feature[RING_F_DCB].indices-1;
5159                 }
5160         }
5161
5162         r_idx = skb->queue_mapping;
5163         tx_ring = &adapter->tx_ring[r_idx];
5164
5165         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5166             (skb->protocol == htons(ETH_P_FCOE)))
5167                 tx_flags |= IXGBE_TX_FLAGS_FCOE;
5168
5169         /* four things can cause us to need a context descriptor */
5170         if (skb_is_gso(skb) ||
5171             (skb->ip_summed == CHECKSUM_PARTIAL) ||
5172             (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
5173             (tx_flags & IXGBE_TX_FLAGS_FCOE))
5174                 count++;
5175
5176         count += TXD_USE_COUNT(skb_headlen(skb));
5177         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5178                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5179
5180         if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
5181                 adapter->tx_busy++;
5182                 return NETDEV_TX_BUSY;
5183         }
5184
5185         first = tx_ring->next_to_use;
5186         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5187 #ifdef IXGBE_FCOE
5188                 /* setup tx offload for FCoE */
5189                 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5190                 if (tso < 0) {
5191                         dev_kfree_skb_any(skb);
5192                         return NETDEV_TX_OK;
5193                 }
5194                 if (tso)
5195                         tx_flags |= IXGBE_TX_FLAGS_FSO;
5196 #endif /* IXGBE_FCOE */
5197         } else {
5198                 if (skb->protocol == htons(ETH_P_IP))
5199                         tx_flags |= IXGBE_TX_FLAGS_IPV4;
5200                 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5201                 if (tso < 0) {
5202                         dev_kfree_skb_any(skb);
5203                         return NETDEV_TX_OK;
5204                 }
5205
5206                 if (tso)
5207                         tx_flags |= IXGBE_TX_FLAGS_TSO;
5208                 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
5209                          (skb->ip_summed == CHECKSUM_PARTIAL))
5210                         tx_flags |= IXGBE_TX_FLAGS_CSUM;
5211         }
5212
5213         count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
5214         if (count) {
5215                 /* add the ATR filter if ATR is on */
5216                 if (tx_ring->atr_sample_rate) {
5217                         ++tx_ring->atr_count;
5218                         if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
5219                              test_bit(__IXGBE_FDIR_INIT_DONE,
5220                                       &tx_ring->reinit_state)) {
5221                                 ixgbe_atr(adapter, skb, tx_ring->queue_index,
5222                                           tx_flags);
5223                                 tx_ring->atr_count = 0;
5224                         }
5225                 }
5226                 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
5227                                hdr_len);
5228                 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
5229
5230         } else {
5231                 dev_kfree_skb_any(skb);
5232                 tx_ring->tx_buffer_info[first].time_stamp = 0;
5233                 tx_ring->next_to_use = first;
5234         }
5235
5236         return NETDEV_TX_OK;
5237 }
5238
5239 /**
5240  * ixgbe_get_stats - Get System Network Statistics
5241  * @netdev: network interface device structure
5242  *
5243  * Returns the address of the device statistics structure.
5244  * The statistics are actually updated from the timer callback.
5245  **/
5246 static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
5247 {
5248         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5249
5250         /* only return the current stats */
5251         return &adapter->net_stats;
5252 }
5253
5254 /**
5255  * ixgbe_set_mac - Change the Ethernet Address of the NIC
5256  * @netdev: network interface device structure
5257  * @p: pointer to an address structure
5258  *
5259  * Returns 0 on success, negative on failure
5260  **/
5261 static int ixgbe_set_mac(struct net_device *netdev, void *p)
5262 {
5263         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5264         struct ixgbe_hw *hw = &adapter->hw;
5265         struct sockaddr *addr = p;
5266
5267         if (!is_valid_ether_addr(addr->sa_data))
5268                 return -EADDRNOTAVAIL;
5269
5270         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
5271         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5272
5273         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
5274
5275         return 0;
5276 }
5277
5278 static int
5279 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
5280 {
5281         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5282         struct ixgbe_hw *hw = &adapter->hw;
5283         u16 value;
5284         int rc;
5285
5286         if (prtad != hw->phy.mdio.prtad)
5287                 return -EINVAL;
5288         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
5289         if (!rc)
5290                 rc = value;
5291         return rc;
5292 }
5293
5294 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
5295                             u16 addr, u16 value)
5296 {
5297         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5298         struct ixgbe_hw *hw = &adapter->hw;
5299
5300         if (prtad != hw->phy.mdio.prtad)
5301                 return -EINVAL;
5302         return hw->phy.ops.write_reg(hw, addr, devad, value);
5303 }
5304
5305 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
5306 {
5307         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5308
5309         return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
5310 }
5311
5312 /**
5313  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
5314  * netdev->dev_addrs
5315  * @netdev: network interface device structure
5316  *
5317  * Returns non-zero on failure
5318  **/
5319 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
5320 {
5321         int err = 0;
5322         struct ixgbe_adapter *adapter = netdev_priv(dev);
5323         struct ixgbe_mac_info *mac = &adapter->hw.mac;
5324
5325         if (is_valid_ether_addr(mac->san_addr)) {
5326                 rtnl_lock();
5327                 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5328                 rtnl_unlock();
5329         }
5330         return err;
5331 }
5332
5333 /**
5334  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
5335  * netdev->dev_addrs
5336  * @netdev: network interface device structure
5337  *
5338  * Returns non-zero on failure
5339  **/
5340 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
5341 {
5342         int err = 0;
5343         struct ixgbe_adapter *adapter = netdev_priv(dev);
5344         struct ixgbe_mac_info *mac = &adapter->hw.mac;
5345
5346         if (is_valid_ether_addr(mac->san_addr)) {
5347                 rtnl_lock();
5348                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5349                 rtnl_unlock();
5350         }
5351         return err;
5352 }
5353
5354 #ifdef CONFIG_NET_POLL_CONTROLLER
5355 /*
5356  * Polling 'interrupt' - used by things like netconsole to send skbs
5357  * without having to re-enable interrupts. It's not called while
5358  * the interrupt routine is executing.
5359  */
5360 static void ixgbe_netpoll(struct net_device *netdev)
5361 {
5362         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5363
5364         disable_irq(adapter->pdev->irq);
5365         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
5366         ixgbe_intr(adapter->pdev->irq, netdev);
5367         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
5368         enable_irq(adapter->pdev->irq);
5369 }
5370 #endif
5371
5372 static const struct net_device_ops ixgbe_netdev_ops = {
5373         .ndo_open               = ixgbe_open,
5374         .ndo_stop               = ixgbe_close,
5375         .ndo_start_xmit         = ixgbe_xmit_frame,
5376         .ndo_select_queue       = ixgbe_select_queue,
5377         .ndo_get_stats          = ixgbe_get_stats,
5378         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
5379         .ndo_set_multicast_list = ixgbe_set_rx_mode,
5380         .ndo_validate_addr      = eth_validate_addr,
5381         .ndo_set_mac_address    = ixgbe_set_mac,
5382         .ndo_change_mtu         = ixgbe_change_mtu,
5383         .ndo_tx_timeout         = ixgbe_tx_timeout,
5384         .ndo_vlan_rx_register   = ixgbe_vlan_rx_register,
5385         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
5386         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
5387         .ndo_do_ioctl           = ixgbe_ioctl,
5388 #ifdef CONFIG_NET_POLL_CONTROLLER
5389         .ndo_poll_controller    = ixgbe_netpoll,
5390 #endif
5391 #ifdef IXGBE_FCOE
5392         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
5393         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
5394 #endif /* IXGBE_FCOE */
5395 };
5396
5397 /**
5398  * ixgbe_probe - Device Initialization Routine
5399  * @pdev: PCI device information struct
5400  * @ent: entry in ixgbe_pci_tbl
5401  *
5402  * Returns 0 on success, negative on failure
5403  *
5404  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
5405  * The OS initialization, configuring of the adapter private structure,
5406  * and a hardware reset occur.
5407  **/
5408 static int __devinit ixgbe_probe(struct pci_dev *pdev,
5409                                  const struct pci_device_id *ent)
5410 {
5411         struct net_device *netdev;
5412         struct ixgbe_adapter *adapter = NULL;
5413         struct ixgbe_hw *hw;
5414         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
5415         static int cards_found;
5416         int i, err, pci_using_dac;
5417 #ifdef IXGBE_FCOE
5418         u16 device_caps;
5419 #endif
5420         u32 part_num, eec;
5421
5422         err = pci_enable_device_mem(pdev);
5423         if (err)
5424                 return err;
5425
5426         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
5427             !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
5428                 pci_using_dac = 1;
5429         } else {
5430                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5431                 if (err) {
5432                         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
5433                         if (err) {
5434                                 dev_err(&pdev->dev, "No usable DMA "
5435                                         "configuration, aborting\n");
5436                                 goto err_dma;
5437                         }
5438                 }
5439                 pci_using_dac = 0;
5440         }
5441
5442         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
5443                                            IORESOURCE_MEM), ixgbe_driver_name);
5444         if (err) {
5445                 dev_err(&pdev->dev,
5446                         "pci_request_selected_regions failed 0x%x\n", err);
5447                 goto err_pci_reg;
5448         }
5449
5450         err = pci_enable_pcie_error_reporting(pdev);
5451         if (err) {
5452                 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
5453                                     "0x%x\n", err);
5454                 /* non-fatal, continue */
5455         }
5456
5457         pci_set_master(pdev);
5458         pci_save_state(pdev);
5459
5460         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
5461         if (!netdev) {
5462                 err = -ENOMEM;
5463                 goto err_alloc_etherdev;
5464         }
5465
5466         SET_NETDEV_DEV(netdev, &pdev->dev);
5467
5468         pci_set_drvdata(pdev, netdev);
5469         adapter = netdev_priv(netdev);
5470
5471         adapter->netdev = netdev;
5472         adapter->pdev = pdev;
5473         hw = &adapter->hw;
5474         hw->back = adapter;
5475         adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
5476
5477         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
5478                               pci_resource_len(pdev, 0));
5479         if (!hw->hw_addr) {
5480                 err = -EIO;
5481                 goto err_ioremap;
5482         }
5483
5484         for (i = 1; i <= 5; i++) {
5485                 if (pci_resource_len(pdev, i) == 0)
5486                         continue;
5487         }
5488
5489         netdev->netdev_ops = &ixgbe_netdev_ops;
5490         ixgbe_set_ethtool_ops(netdev);
5491         netdev->watchdog_timeo = 5 * HZ;
5492         strcpy(netdev->name, pci_name(pdev));
5493
5494         adapter->bd_number = cards_found;
5495
5496         /* Setup hw api */
5497         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
5498         hw->mac.type  = ii->mac;
5499
5500         /* EEPROM */
5501         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
5502         eec = IXGBE_READ_REG(hw, IXGBE_EEC);
5503         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
5504         if (!(eec & (1 << 8)))
5505                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
5506
5507         /* PHY */
5508         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
5509         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
5510         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
5511         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
5512         hw->phy.mdio.mmds = 0;
5513         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
5514         hw->phy.mdio.dev = netdev;
5515         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
5516         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
5517
5518         /* set up this timer and work struct before calling get_invariants
5519          * which might start the timer
5520          */
5521         init_timer(&adapter->sfp_timer);
5522         adapter->sfp_timer.function = &ixgbe_sfp_timer;
5523         adapter->sfp_timer.data = (unsigned long) adapter;
5524
5525         INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
5526
5527         /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
5528         INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
5529
5530         /* a new SFP+ module arrival, called from GPI SDP2 context */
5531         INIT_WORK(&adapter->sfp_config_module_task,
5532                   ixgbe_sfp_config_module_task);
5533
5534         ii->get_invariants(hw);
5535
5536         /* setup the private structure */
5537         err = ixgbe_sw_init(adapter);
5538         if (err)
5539                 goto err_sw_init;
5540
5541         /*
5542          * If there is a fan on this device and it has failed log the
5543          * failure.
5544          */
5545         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5546                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5547                 if (esdp & IXGBE_ESDP_SDP1)
5548                         DPRINTK(PROBE, CRIT,
5549                                 "Fan has stopped, replace the adapter\n");
5550         }
5551
5552         /* reset_hw fills in the perm_addr as well */
5553         err = hw->mac.ops.reset_hw(hw);
5554         if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
5555             hw->mac.type == ixgbe_mac_82598EB) {
5556                 /*
5557                  * Start a kernel thread to watch for a module to arrive.
5558                  * Only do this for 82598, since 82599 will generate
5559                  * interrupts on module arrival.
5560                  */
5561                 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5562                 mod_timer(&adapter->sfp_timer,
5563                           round_jiffies(jiffies + (2 * HZ)));
5564                 err = 0;
5565         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5566                 dev_err(&adapter->pdev->dev, "failed to initialize because "
5567                         "an unsupported SFP+ module type was detected.\n"
5568                         "Reload the driver after installing a supported "
5569                         "module.\n");
5570                 goto err_sw_init;
5571         } else if (err) {
5572                 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
5573                 goto err_sw_init;
5574         }
5575
5576         netdev->features = NETIF_F_SG |
5577                            NETIF_F_IP_CSUM |
5578                            NETIF_F_HW_VLAN_TX |
5579                            NETIF_F_HW_VLAN_RX |
5580                            NETIF_F_HW_VLAN_FILTER;
5581
5582         netdev->features |= NETIF_F_IPV6_CSUM;
5583         netdev->features |= NETIF_F_TSO;
5584         netdev->features |= NETIF_F_TSO6;
5585         netdev->features |= NETIF_F_GRO;
5586
5587         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
5588                 netdev->features |= NETIF_F_SCTP_CSUM;
5589
5590         netdev->vlan_features |= NETIF_F_TSO;
5591         netdev->vlan_features |= NETIF_F_TSO6;
5592         netdev->vlan_features |= NETIF_F_IP_CSUM;
5593         netdev->vlan_features |= NETIF_F_SG;
5594
5595         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5596                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
5597
5598 #ifdef CONFIG_IXGBE_DCB
5599         netdev->dcbnl_ops = &dcbnl_ops;
5600 #endif
5601
5602 #ifdef IXGBE_FCOE
5603         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
5604                 if (hw->mac.ops.get_device_caps) {
5605                         hw->mac.ops.get_device_caps(hw, &device_caps);
5606                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
5607                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5608                 }
5609         }
5610 #endif /* IXGBE_FCOE */
5611         if (pci_using_dac)
5612                 netdev->features |= NETIF_F_HIGHDMA;
5613
5614         if (adapter->flags & IXGBE_FLAG2_RSC_ENABLED)
5615                 netdev->features |= NETIF_F_LRO;
5616
5617         /* make sure the EEPROM is good */
5618         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
5619                 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
5620                 err = -EIO;
5621                 goto err_eeprom;
5622         }
5623
5624         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
5625         memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
5626
5627         if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
5628                 dev_err(&pdev->dev, "invalid MAC address\n");
5629                 err = -EIO;
5630                 goto err_eeprom;
5631         }
5632
5633         init_timer(&adapter->watchdog_timer);
5634         adapter->watchdog_timer.function = &ixgbe_watchdog;
5635         adapter->watchdog_timer.data = (unsigned long)adapter;
5636
5637         INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
5638         INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
5639
5640         err = ixgbe_init_interrupt_scheme(adapter);
5641         if (err)
5642                 goto err_sw_init;
5643
5644         switch (pdev->device) {
5645         case IXGBE_DEV_ID_82599_KX4:
5646                 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
5647                                 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
5648                 /* Enable ACPI wakeup in GRC */
5649                 IXGBE_WRITE_REG(hw, IXGBE_GRC,
5650                              (IXGBE_READ_REG(hw, IXGBE_GRC) & ~IXGBE_GRC_APME));
5651                 break;
5652         default:
5653                 adapter->wol = 0;
5654                 break;
5655         }
5656         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5657
5658         /* pick up the PCI bus settings for reporting later */
5659         hw->mac.ops.get_bus_info(hw);
5660
5661         /* print bus type/speed/width info */
5662         dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
5663                 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
5664                  (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
5665                 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
5666                  (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
5667                  (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
5668                  "Unknown"),
5669                 netdev->dev_addr);
5670         ixgbe_read_pba_num_generic(hw, &part_num);
5671         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
5672                 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
5673                          hw->mac.type, hw->phy.type, hw->phy.sfp_type,
5674                          (part_num >> 8), (part_num & 0xff));
5675         else
5676                 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5677                          hw->mac.type, hw->phy.type,
5678                          (part_num >> 8), (part_num & 0xff));
5679
5680         if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
5681                 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
5682                          "this card is not sufficient for optimal "
5683                          "performance.\n");
5684                 dev_warn(&pdev->dev, "For optimal performance a x8 "
5685                          "PCI-Express slot is required.\n");
5686         }
5687
5688         /* save off EEPROM version number */
5689         hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
5690
5691         /* reset the hardware with the new settings */
5692         err = hw->mac.ops.start_hw(hw);
5693
5694         if (err == IXGBE_ERR_EEPROM_VERSION) {
5695                 /* We are running on a pre-production device, log a warning */
5696                 dev_warn(&pdev->dev, "This device is a pre-production "
5697                          "adapter/LOM.  Please be aware there may be issues "
5698                          "associated with your hardware.  If you are "
5699                          "experiencing problems please contact your Intel or "
5700                          "hardware representative who provided you with this "
5701                          "hardware.\n");
5702         }
5703         strcpy(netdev->name, "eth%d");
5704         err = register_netdev(netdev);
5705         if (err)
5706                 goto err_register;
5707
5708         /* carrier off reporting is important to ethtool even BEFORE open */
5709         netif_carrier_off(netdev);
5710
5711         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5712             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5713                 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
5714
5715 #ifdef CONFIG_IXGBE_DCA
5716         if (dca_add_requester(&pdev->dev) == 0) {
5717                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
5718                 ixgbe_setup_dca(adapter);
5719         }
5720 #endif
5721         /* add san mac addr to netdev */
5722         ixgbe_add_sanmac_netdev(netdev);
5723
5724         dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
5725         cards_found++;
5726         return 0;
5727
5728 err_register:
5729         ixgbe_release_hw_control(adapter);
5730         ixgbe_clear_interrupt_scheme(adapter);
5731 err_sw_init:
5732 err_eeprom:
5733         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5734         del_timer_sync(&adapter->sfp_timer);
5735         cancel_work_sync(&adapter->sfp_task);
5736         cancel_work_sync(&adapter->multispeed_fiber_task);
5737         cancel_work_sync(&adapter->sfp_config_module_task);
5738         iounmap(hw->hw_addr);
5739 err_ioremap:
5740         free_netdev(netdev);
5741 err_alloc_etherdev:
5742         pci_release_selected_regions(pdev, pci_select_bars(pdev,
5743                                      IORESOURCE_MEM));
5744 err_pci_reg:
5745 err_dma:
5746         pci_disable_device(pdev);
5747         return err;
5748 }
5749
5750 /**
5751  * ixgbe_remove - Device Removal Routine
5752  * @pdev: PCI device information struct
5753  *
5754  * ixgbe_remove is called by the PCI subsystem to alert the driver
5755  * that it should release a PCI device.  The could be caused by a
5756  * Hot-Plug event, or because the driver is going to be removed from
5757  * memory.
5758  **/
5759 static void __devexit ixgbe_remove(struct pci_dev *pdev)
5760 {
5761         struct net_device *netdev = pci_get_drvdata(pdev);
5762         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5763         int err;
5764
5765         set_bit(__IXGBE_DOWN, &adapter->state);
5766         /* clear the module not found bit to make sure the worker won't
5767          * reschedule
5768          */
5769         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5770         del_timer_sync(&adapter->watchdog_timer);
5771
5772         del_timer_sync(&adapter->sfp_timer);
5773         cancel_work_sync(&adapter->watchdog_task);
5774         cancel_work_sync(&adapter->sfp_task);
5775         cancel_work_sync(&adapter->multispeed_fiber_task);
5776         cancel_work_sync(&adapter->sfp_config_module_task);
5777         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5778             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5779                 cancel_work_sync(&adapter->fdir_reinit_task);
5780         flush_scheduled_work();
5781
5782 #ifdef CONFIG_IXGBE_DCA
5783         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
5784                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
5785                 dca_remove_requester(&pdev->dev);
5786                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
5787         }
5788
5789 #endif
5790 #ifdef IXGBE_FCOE
5791         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
5792                 ixgbe_cleanup_fcoe(adapter);
5793
5794 #endif /* IXGBE_FCOE */
5795
5796         /* remove the added san mac */
5797         ixgbe_del_sanmac_netdev(netdev);
5798
5799         if (netdev->reg_state == NETREG_REGISTERED)
5800                 unregister_netdev(netdev);
5801
5802         ixgbe_clear_interrupt_scheme(adapter);
5803
5804         ixgbe_release_hw_control(adapter);
5805
5806         iounmap(adapter->hw.hw_addr);
5807         pci_release_selected_regions(pdev, pci_select_bars(pdev,
5808                                      IORESOURCE_MEM));
5809
5810         DPRINTK(PROBE, INFO, "complete\n");
5811
5812         free_netdev(netdev);
5813
5814         err = pci_disable_pcie_error_reporting(pdev);
5815         if (err)
5816                 dev_err(&pdev->dev,
5817                         "pci_disable_pcie_error_reporting failed 0x%x\n", err);
5818
5819         pci_disable_device(pdev);
5820 }
5821
5822 /**
5823  * ixgbe_io_error_detected - called when PCI error is detected
5824  * @pdev: Pointer to PCI device
5825  * @state: The current pci connection state
5826  *
5827  * This function is called after a PCI bus error affecting
5828  * this device has been detected.
5829  */
5830 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
5831                                                 pci_channel_state_t state)
5832 {
5833         struct net_device *netdev = pci_get_drvdata(pdev);
5834         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5835
5836         netif_device_detach(netdev);
5837
5838         if (state == pci_channel_io_perm_failure)
5839                 return PCI_ERS_RESULT_DISCONNECT;
5840
5841         if (netif_running(netdev))
5842                 ixgbe_down(adapter);
5843         pci_disable_device(pdev);
5844
5845         /* Request a slot reset. */
5846         return PCI_ERS_RESULT_NEED_RESET;
5847 }
5848
5849 /**
5850  * ixgbe_io_slot_reset - called after the pci bus has been reset.
5851  * @pdev: Pointer to PCI device
5852  *
5853  * Restart the card from scratch, as if from a cold-boot.
5854  */
5855 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
5856 {
5857         struct net_device *netdev = pci_get_drvdata(pdev);
5858         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5859         pci_ers_result_t result;
5860         int err;
5861
5862         if (pci_enable_device_mem(pdev)) {
5863                 DPRINTK(PROBE, ERR,
5864                         "Cannot re-enable PCI device after reset.\n");
5865                 result = PCI_ERS_RESULT_DISCONNECT;
5866         } else {
5867                 pci_set_master(pdev);
5868                 pci_restore_state(pdev);
5869
5870                 pci_wake_from_d3(pdev, false);
5871
5872                 ixgbe_reset(adapter);
5873                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5874                 result = PCI_ERS_RESULT_RECOVERED;
5875         }
5876
5877         err = pci_cleanup_aer_uncorrect_error_status(pdev);
5878         if (err) {
5879                 dev_err(&pdev->dev,
5880                   "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
5881                 /* non-fatal, continue */
5882         }
5883
5884         return result;
5885 }
5886
5887 /**
5888  * ixgbe_io_resume - called when traffic can start flowing again.
5889  * @pdev: Pointer to PCI device
5890  *
5891  * This callback is called when the error recovery driver tells us that
5892  * its OK to resume normal operation.
5893  */
5894 static void ixgbe_io_resume(struct pci_dev *pdev)
5895 {
5896         struct net_device *netdev = pci_get_drvdata(pdev);
5897         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5898
5899         if (netif_running(netdev)) {
5900                 if (ixgbe_up(adapter)) {
5901                         DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
5902                         return;
5903                 }
5904         }
5905
5906         netif_device_attach(netdev);
5907 }
5908
5909 static struct pci_error_handlers ixgbe_err_handler = {
5910         .error_detected = ixgbe_io_error_detected,
5911         .slot_reset = ixgbe_io_slot_reset,
5912         .resume = ixgbe_io_resume,
5913 };
5914
5915 static struct pci_driver ixgbe_driver = {
5916         .name     = ixgbe_driver_name,
5917         .id_table = ixgbe_pci_tbl,
5918         .probe    = ixgbe_probe,
5919         .remove   = __devexit_p(ixgbe_remove),
5920 #ifdef CONFIG_PM
5921         .suspend  = ixgbe_suspend,
5922         .resume   = ixgbe_resume,
5923 #endif
5924         .shutdown = ixgbe_shutdown,
5925         .err_handler = &ixgbe_err_handler
5926 };
5927
5928 /**
5929  * ixgbe_init_module - Driver Registration Routine
5930  *
5931  * ixgbe_init_module is the first routine called when the driver is
5932  * loaded. All it does is register with the PCI subsystem.
5933  **/
5934 static int __init ixgbe_init_module(void)
5935 {
5936         int ret;
5937         printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
5938                ixgbe_driver_string, ixgbe_driver_version);
5939
5940         printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
5941
5942 #ifdef CONFIG_IXGBE_DCA
5943         dca_register_notify(&dca_notifier);
5944 #endif
5945
5946         ret = pci_register_driver(&ixgbe_driver);
5947         return ret;
5948 }
5949
5950 module_init(ixgbe_init_module);
5951
5952 /**
5953  * ixgbe_exit_module - Driver Exit Cleanup Routine
5954  *
5955  * ixgbe_exit_module is called just before the driver is removed
5956  * from memory.
5957  **/
5958 static void __exit ixgbe_exit_module(void)
5959 {
5960 #ifdef CONFIG_IXGBE_DCA
5961         dca_unregister_notify(&dca_notifier);
5962 #endif
5963         pci_unregister_driver(&ixgbe_driver);
5964 }
5965
5966 #ifdef CONFIG_IXGBE_DCA
5967 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
5968                             void *p)
5969 {
5970         int ret_val;
5971
5972         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
5973                                          __ixgbe_notify_dca);
5974
5975         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5976 }
5977
5978 #endif /* CONFIG_IXGBE_DCA */
5979 #ifdef DEBUG
5980 /**
5981  * ixgbe_get_hw_dev_name - return device name string
5982  * used by hardware layer to print debugging information
5983  **/
5984 char *ixgbe_get_hw_dev_name(struct ixgbe_hw *hw)
5985 {
5986         struct ixgbe_adapter *adapter = hw->back;
5987         return adapter->netdev->name;
5988 }
5989
5990 #endif
5991 module_exit(ixgbe_exit_module);
5992
5993 /* ixgbe_main.c */