Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[safe/jmp/linux-2.6] / drivers / net / ixgbe / ixgbe_main.c
1 /*******************************************************************************
2
3   Intel 10 Gigabit PCI Express Linux driver
4   Copyright(c) 1999 - 2010 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 #include "ixgbe_dcb_82599.h"
48 #include "ixgbe_sriov.h"
49
50 char ixgbe_driver_name[] = "ixgbe";
51 static const char ixgbe_driver_string[] =
52                               "Intel(R) 10 Gigabit PCI Express Network Driver";
53
54 #define DRV_VERSION "2.0.44-k2"
55 const char ixgbe_driver_version[] = DRV_VERSION;
56 static char ixgbe_copyright[] = "Copyright (c) 1999-2010 Intel Corporation.";
57
58 static const struct ixgbe_info *ixgbe_info_tbl[] = {
59         [board_82598] = &ixgbe_82598_info,
60         [board_82599] = &ixgbe_82599_info,
61 };
62
63 /* ixgbe_pci_tbl - PCI Device ID Table
64  *
65  * Wildcard entries (PCI_ANY_ID) should come last
66  * Last entry must be all 0s
67  *
68  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
69  *   Class, Class Mask, private data (not used) }
70  */
71 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
72         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
73          board_82598 },
74         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
75          board_82598 },
76         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
77          board_82598 },
78         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
79          board_82598 },
80         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
81          board_82598 },
82         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
83          board_82598 },
84         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
85          board_82598 },
86         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
87          board_82598 },
88         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
89          board_82598 },
90         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
91          board_82598 },
92         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
93          board_82598 },
94         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
95          board_82598 },
96         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
97          board_82599 },
98         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
99          board_82599 },
100         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
101          board_82599 },
102         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
103          board_82599 },
104         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
105          board_82599 },
106         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
107          board_82599 },
108         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
109          board_82599 },
110         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
111          board_82599 },
112
113         /* required last entry */
114         {0, }
115 };
116 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
117
118 #ifdef CONFIG_IXGBE_DCA
119 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
120                             void *p);
121 static struct notifier_block dca_notifier = {
122         .notifier_call = ixgbe_notify_dca,
123         .next          = NULL,
124         .priority      = 0
125 };
126 #endif
127
128 #ifdef CONFIG_PCI_IOV
129 static unsigned int max_vfs;
130 module_param(max_vfs, uint, 0);
131 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
132                  "per physical function");
133 #endif /* CONFIG_PCI_IOV */
134
135 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
136 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
137 MODULE_LICENSE("GPL");
138 MODULE_VERSION(DRV_VERSION);
139
140 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
141
142 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
143 {
144         struct ixgbe_hw *hw = &adapter->hw;
145         u32 gcr;
146         u32 gpie;
147         u32 vmdctl;
148
149 #ifdef CONFIG_PCI_IOV
150         /* disable iov and allow time for transactions to clear */
151         pci_disable_sriov(adapter->pdev);
152 #endif
153
154         /* turn off device IOV mode */
155         gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
156         gcr &= ~(IXGBE_GCR_EXT_SRIOV);
157         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
158         gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
159         gpie &= ~IXGBE_GPIE_VTMODE_MASK;
160         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
161
162         /* set default pool back to 0 */
163         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
164         vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
165         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
166
167         /* take a breather then clean up driver data */
168         msleep(100);
169         if (adapter->vfinfo)
170                 kfree(adapter->vfinfo);
171         adapter->vfinfo = NULL;
172
173         adapter->num_vfs = 0;
174         adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
175 }
176
177 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
178 {
179         u32 ctrl_ext;
180
181         /* Let firmware take over control of h/w */
182         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
183         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
184                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
185 }
186
187 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
188 {
189         u32 ctrl_ext;
190
191         /* Let firmware know the driver has taken over */
192         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
193         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
194                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
195 }
196
197 /*
198  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
199  * @adapter: pointer to adapter struct
200  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
201  * @queue: queue to map the corresponding interrupt to
202  * @msix_vector: the vector to map to the corresponding queue
203  *
204  */
205 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
206                            u8 queue, u8 msix_vector)
207 {
208         u32 ivar, index;
209         struct ixgbe_hw *hw = &adapter->hw;
210         switch (hw->mac.type) {
211         case ixgbe_mac_82598EB:
212                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
213                 if (direction == -1)
214                         direction = 0;
215                 index = (((direction * 64) + queue) >> 2) & 0x1F;
216                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
217                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
218                 ivar |= (msix_vector << (8 * (queue & 0x3)));
219                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
220                 break;
221         case ixgbe_mac_82599EB:
222                 if (direction == -1) {
223                         /* other causes */
224                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
225                         index = ((queue & 1) * 8);
226                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
227                         ivar &= ~(0xFF << index);
228                         ivar |= (msix_vector << index);
229                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
230                         break;
231                 } else {
232                         /* tx or rx causes */
233                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
234                         index = ((16 * (queue & 1)) + (8 * direction));
235                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
236                         ivar &= ~(0xFF << index);
237                         ivar |= (msix_vector << index);
238                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
239                         break;
240                 }
241         default:
242                 break;
243         }
244 }
245
246 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
247                                           u64 qmask)
248 {
249         u32 mask;
250
251         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
252                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
253                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
254         } else {
255                 mask = (qmask & 0xFFFFFFFF);
256                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
257                 mask = (qmask >> 32);
258                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
259         }
260 }
261
262 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
263                                              struct ixgbe_tx_buffer
264                                              *tx_buffer_info)
265 {
266         if (tx_buffer_info->dma) {
267                 if (tx_buffer_info->mapped_as_page)
268                         pci_unmap_page(adapter->pdev,
269                                        tx_buffer_info->dma,
270                                        tx_buffer_info->length,
271                                        PCI_DMA_TODEVICE);
272                 else
273                         pci_unmap_single(adapter->pdev,
274                                          tx_buffer_info->dma,
275                                          tx_buffer_info->length,
276                                          PCI_DMA_TODEVICE);
277                 tx_buffer_info->dma = 0;
278         }
279         if (tx_buffer_info->skb) {
280                 dev_kfree_skb_any(tx_buffer_info->skb);
281                 tx_buffer_info->skb = NULL;
282         }
283         tx_buffer_info->time_stamp = 0;
284         /* tx_buffer_info must be completely set up in the transmit path */
285 }
286
287 /**
288  * ixgbe_tx_is_paused - check if the tx ring is paused
289  * @adapter: the ixgbe adapter
290  * @tx_ring: the corresponding tx_ring
291  *
292  * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
293  * corresponding TC of this tx_ring when checking TFCS.
294  *
295  * Returns : true if paused
296  */
297 static inline bool ixgbe_tx_is_paused(struct ixgbe_adapter *adapter,
298                                       struct ixgbe_ring *tx_ring)
299 {
300         u32 txoff = IXGBE_TFCS_TXOFF;
301
302 #ifdef CONFIG_IXGBE_DCB
303         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
304                 int tc;
305                 int reg_idx = tx_ring->reg_idx;
306                 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
307
308                 switch (adapter->hw.mac.type) {
309                 case ixgbe_mac_82598EB:
310                         tc = reg_idx >> 2;
311                         txoff = IXGBE_TFCS_TXOFF0;
312                         break;
313                 case ixgbe_mac_82599EB:
314                         tc = 0;
315                         txoff = IXGBE_TFCS_TXOFF;
316                         if (dcb_i == 8) {
317                                 /* TC0, TC1 */
318                                 tc = reg_idx >> 5;
319                                 if (tc == 2) /* TC2, TC3 */
320                                         tc += (reg_idx - 64) >> 4;
321                                 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
322                                         tc += 1 + ((reg_idx - 96) >> 3);
323                         } else if (dcb_i == 4) {
324                                 /* TC0, TC1 */
325                                 tc = reg_idx >> 6;
326                                 if (tc == 1) {
327                                         tc += (reg_idx - 64) >> 5;
328                                         if (tc == 2) /* TC2, TC3 */
329                                                 tc += (reg_idx - 96) >> 4;
330                                 }
331                         }
332                         break;
333                 default:
334                         tc = 0;
335                 }
336                 txoff <<= tc;
337         }
338 #endif
339         return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
340 }
341
342 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
343                                        struct ixgbe_ring *tx_ring,
344                                        unsigned int eop)
345 {
346         struct ixgbe_hw *hw = &adapter->hw;
347
348         /* Detect a transmit hang in hardware, this serializes the
349          * check with the clearing of time_stamp and movement of eop */
350         adapter->detect_tx_hung = false;
351         if (tx_ring->tx_buffer_info[eop].time_stamp &&
352             time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
353             !ixgbe_tx_is_paused(adapter, tx_ring)) {
354                 /* detected Tx unit hang */
355                 union ixgbe_adv_tx_desc *tx_desc;
356                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
357                 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
358                         "  Tx Queue             <%d>\n"
359                         "  TDH, TDT             <%x>, <%x>\n"
360                         "  next_to_use          <%x>\n"
361                         "  next_to_clean        <%x>\n"
362                         "tx_buffer_info[next_to_clean]\n"
363                         "  time_stamp           <%lx>\n"
364                         "  jiffies              <%lx>\n",
365                         tx_ring->queue_index,
366                         IXGBE_READ_REG(hw, tx_ring->head),
367                         IXGBE_READ_REG(hw, tx_ring->tail),
368                         tx_ring->next_to_use, eop,
369                         tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
370                 return true;
371         }
372
373         return false;
374 }
375
376 #define IXGBE_MAX_TXD_PWR       14
377 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
378
379 /* Tx Descriptors needed, worst case */
380 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
381                          (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
382 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
383         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
384
385 static void ixgbe_tx_timeout(struct net_device *netdev);
386
387 /**
388  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
389  * @q_vector: structure containing interrupt and ring information
390  * @tx_ring: tx ring to clean
391  **/
392 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
393                                struct ixgbe_ring *tx_ring)
394 {
395         struct ixgbe_adapter *adapter = q_vector->adapter;
396         struct net_device *netdev = adapter->netdev;
397         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
398         struct ixgbe_tx_buffer *tx_buffer_info;
399         unsigned int i, eop, count = 0;
400         unsigned int total_bytes = 0, total_packets = 0;
401
402         i = tx_ring->next_to_clean;
403         eop = tx_ring->tx_buffer_info[i].next_to_watch;
404         eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
405
406         while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
407                (count < tx_ring->work_limit)) {
408                 bool cleaned = false;
409                 for ( ; !cleaned; count++) {
410                         struct sk_buff *skb;
411                         tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
412                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
413                         cleaned = (i == eop);
414                         skb = tx_buffer_info->skb;
415
416                         if (cleaned && skb) {
417                                 unsigned int segs, bytecount;
418                                 unsigned int hlen = skb_headlen(skb);
419
420                                 /* gso_segs is currently only valid for tcp */
421                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
422 #ifdef IXGBE_FCOE
423                                 /* adjust for FCoE Sequence Offload */
424                                 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
425                                     && (skb->protocol == htons(ETH_P_FCOE)) &&
426                                     skb_is_gso(skb)) {
427                                         hlen = skb_transport_offset(skb) +
428                                                 sizeof(struct fc_frame_header) +
429                                                 sizeof(struct fcoe_crc_eof);
430                                         segs = DIV_ROUND_UP(skb->len - hlen,
431                                                 skb_shinfo(skb)->gso_size);
432                                 }
433 #endif /* IXGBE_FCOE */
434                                 /* multiply data chunks by size of headers */
435                                 bytecount = ((segs - 1) * hlen) + skb->len;
436                                 total_packets += segs;
437                                 total_bytes += bytecount;
438                         }
439
440                         ixgbe_unmap_and_free_tx_resource(adapter,
441                                                          tx_buffer_info);
442
443                         tx_desc->wb.status = 0;
444
445                         i++;
446                         if (i == tx_ring->count)
447                                 i = 0;
448                 }
449
450                 eop = tx_ring->tx_buffer_info[i].next_to_watch;
451                 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
452         }
453
454         tx_ring->next_to_clean = i;
455
456 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
457         if (unlikely(count && netif_carrier_ok(netdev) &&
458                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
459                 /* Make sure that anybody stopping the queue after this
460                  * sees the new next_to_clean.
461                  */
462                 smp_mb();
463                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
464                     !test_bit(__IXGBE_DOWN, &adapter->state)) {
465                         netif_wake_subqueue(netdev, tx_ring->queue_index);
466                         ++tx_ring->restart_queue;
467                 }
468         }
469
470         if (adapter->detect_tx_hung) {
471                 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
472                         /* schedule immediate reset if we believe we hung */
473                         DPRINTK(PROBE, INFO,
474                                 "tx hang %d detected, resetting adapter\n",
475                                 adapter->tx_timeout_count + 1);
476                         ixgbe_tx_timeout(adapter->netdev);
477                 }
478         }
479
480         /* re-arm the interrupt */
481         if (count >= tx_ring->work_limit)
482                 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
483
484         tx_ring->total_bytes += total_bytes;
485         tx_ring->total_packets += total_packets;
486         tx_ring->stats.packets += total_packets;
487         tx_ring->stats.bytes += total_bytes;
488         return (count < tx_ring->work_limit);
489 }
490
491 #ifdef CONFIG_IXGBE_DCA
492 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
493                                 struct ixgbe_ring *rx_ring)
494 {
495         u32 rxctrl;
496         int cpu = get_cpu();
497         int q = rx_ring - adapter->rx_ring;
498
499         if (rx_ring->cpu != cpu) {
500                 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
501                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
502                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
503                         rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
504                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
505                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
506                         rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
507                                    IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
508                 }
509                 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
510                 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
511                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
512                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
513                             IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
514                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
515                 rx_ring->cpu = cpu;
516         }
517         put_cpu();
518 }
519
520 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
521                                 struct ixgbe_ring *tx_ring)
522 {
523         u32 txctrl;
524         int cpu = get_cpu();
525         int q = tx_ring - adapter->tx_ring;
526         struct ixgbe_hw *hw = &adapter->hw;
527
528         if (tx_ring->cpu != cpu) {
529                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
530                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(q));
531                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
532                         txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
533                         txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
534                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(q), txctrl);
535                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
536                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(q));
537                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
538                         txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
539                                   IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
540                         txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
541                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(q), txctrl);
542                 }
543                 tx_ring->cpu = cpu;
544         }
545         put_cpu();
546 }
547
548 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
549 {
550         int i;
551
552         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
553                 return;
554
555         /* always use CB2 mode, difference is masked in the CB driver */
556         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
557
558         for (i = 0; i < adapter->num_tx_queues; i++) {
559                 adapter->tx_ring[i].cpu = -1;
560                 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
561         }
562         for (i = 0; i < adapter->num_rx_queues; i++) {
563                 adapter->rx_ring[i].cpu = -1;
564                 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
565         }
566 }
567
568 static int __ixgbe_notify_dca(struct device *dev, void *data)
569 {
570         struct net_device *netdev = dev_get_drvdata(dev);
571         struct ixgbe_adapter *adapter = netdev_priv(netdev);
572         unsigned long event = *(unsigned long *)data;
573
574         switch (event) {
575         case DCA_PROVIDER_ADD:
576                 /* if we're already enabled, don't do it again */
577                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
578                         break;
579                 if (dca_add_requester(dev) == 0) {
580                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
581                         ixgbe_setup_dca(adapter);
582                         break;
583                 }
584                 /* Fall Through since DCA is disabled. */
585         case DCA_PROVIDER_REMOVE:
586                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
587                         dca_remove_requester(dev);
588                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
589                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
590                 }
591                 break;
592         }
593
594         return 0;
595 }
596
597 #endif /* CONFIG_IXGBE_DCA */
598 /**
599  * ixgbe_receive_skb - Send a completed packet up the stack
600  * @adapter: board private structure
601  * @skb: packet to send up
602  * @status: hardware indication of status of receive
603  * @rx_ring: rx descriptor ring (for a specific queue) to setup
604  * @rx_desc: rx descriptor
605  **/
606 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
607                               struct sk_buff *skb, u8 status,
608                               struct ixgbe_ring *ring,
609                               union ixgbe_adv_rx_desc *rx_desc)
610 {
611         struct ixgbe_adapter *adapter = q_vector->adapter;
612         struct napi_struct *napi = &q_vector->napi;
613         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
614         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
615
616         skb_record_rx_queue(skb, ring->queue_index);
617         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
618                 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
619                         vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
620                 else
621                         napi_gro_receive(napi, skb);
622         } else {
623                 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
624                         vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
625                 else
626                         netif_rx(skb);
627         }
628 }
629
630 /**
631  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
632  * @adapter: address of board private structure
633  * @status_err: hardware indication of status of receive
634  * @skb: skb currently being received and modified
635  **/
636 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
637                                      union ixgbe_adv_rx_desc *rx_desc,
638                                      struct sk_buff *skb)
639 {
640         u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
641
642         skb->ip_summed = CHECKSUM_NONE;
643
644         /* Rx csum disabled */
645         if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
646                 return;
647
648         /* if IP and error */
649         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
650             (status_err & IXGBE_RXDADV_ERR_IPE)) {
651                 adapter->hw_csum_rx_error++;
652                 return;
653         }
654
655         if (!(status_err & IXGBE_RXD_STAT_L4CS))
656                 return;
657
658         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
659                 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
660
661                 /*
662                  * 82599 errata, UDP frames with a 0 checksum can be marked as
663                  * checksum errors.
664                  */
665                 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
666                     (adapter->hw.mac.type == ixgbe_mac_82599EB))
667                         return;
668
669                 adapter->hw_csum_rx_error++;
670                 return;
671         }
672
673         /* It must be a TCP or UDP packet with a valid checksum */
674         skb->ip_summed = CHECKSUM_UNNECESSARY;
675 }
676
677 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
678                                          struct ixgbe_ring *rx_ring, u32 val)
679 {
680         /*
681          * Force memory writes to complete before letting h/w
682          * know there are new descriptors to fetch.  (Only
683          * applicable for weak-ordered memory model archs,
684          * such as IA-64).
685          */
686         wmb();
687         IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
688 }
689
690 /**
691  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
692  * @adapter: address of board private structure
693  **/
694 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
695                                    struct ixgbe_ring *rx_ring,
696                                    int cleaned_count)
697 {
698         struct pci_dev *pdev = adapter->pdev;
699         union ixgbe_adv_rx_desc *rx_desc;
700         struct ixgbe_rx_buffer *bi;
701         unsigned int i;
702
703         i = rx_ring->next_to_use;
704         bi = &rx_ring->rx_buffer_info[i];
705
706         while (cleaned_count--) {
707                 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
708
709                 if (!bi->page_dma &&
710                     (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
711                         if (!bi->page) {
712                                 bi->page = alloc_page(GFP_ATOMIC);
713                                 if (!bi->page) {
714                                         adapter->alloc_rx_page_failed++;
715                                         goto no_buffers;
716                                 }
717                                 bi->page_offset = 0;
718                         } else {
719                                 /* use a half page if we're re-using */
720                                 bi->page_offset ^= (PAGE_SIZE / 2);
721                         }
722
723                         bi->page_dma = pci_map_page(pdev, bi->page,
724                                                     bi->page_offset,
725                                                     (PAGE_SIZE / 2),
726                                                     PCI_DMA_FROMDEVICE);
727                 }
728
729                 if (!bi->skb) {
730                         struct sk_buff *skb;
731                         /* netdev_alloc_skb reserves 32 bytes up front!! */
732                         uint bufsz = rx_ring->rx_buf_len + SMP_CACHE_BYTES;
733                         skb = netdev_alloc_skb(adapter->netdev, bufsz);
734
735                         if (!skb) {
736                                 adapter->alloc_rx_buff_failed++;
737                                 goto no_buffers;
738                         }
739
740                         /* advance the data pointer to the next cache line */
741                         skb_reserve(skb, (PTR_ALIGN(skb->data, SMP_CACHE_BYTES)
742                                           - skb->data));
743
744                         bi->skb = skb;
745                         bi->dma = pci_map_single(pdev, skb->data,
746                                                  rx_ring->rx_buf_len,
747                                                  PCI_DMA_FROMDEVICE);
748                 }
749                 /* Refresh the desc even if buffer_addrs didn't change because
750                  * each write-back erases this info. */
751                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
752                         rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
753                         rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
754                 } else {
755                         rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
756                 }
757
758                 i++;
759                 if (i == rx_ring->count)
760                         i = 0;
761                 bi = &rx_ring->rx_buffer_info[i];
762         }
763
764 no_buffers:
765         if (rx_ring->next_to_use != i) {
766                 rx_ring->next_to_use = i;
767                 if (i-- == 0)
768                         i = (rx_ring->count - 1);
769
770                 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
771         }
772 }
773
774 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
775 {
776         return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
777 }
778
779 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
780 {
781         return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
782 }
783
784 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
785 {
786         return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
787                 IXGBE_RXDADV_RSCCNT_MASK) >>
788                 IXGBE_RXDADV_RSCCNT_SHIFT;
789 }
790
791 /**
792  * ixgbe_transform_rsc_queue - change rsc queue into a full packet
793  * @skb: pointer to the last skb in the rsc queue
794  * @count: pointer to number of packets coalesced in this context
795  *
796  * This function changes a queue full of hw rsc buffers into a completed
797  * packet.  It uses the ->prev pointers to find the first packet and then
798  * turns it into the frag list owner.
799  **/
800 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
801                                                         u64 *count)
802 {
803         unsigned int frag_list_size = 0;
804
805         while (skb->prev) {
806                 struct sk_buff *prev = skb->prev;
807                 frag_list_size += skb->len;
808                 skb->prev = NULL;
809                 skb = prev;
810                 *count += 1;
811         }
812
813         skb_shinfo(skb)->frag_list = skb->next;
814         skb->next = NULL;
815         skb->len += frag_list_size;
816         skb->data_len += frag_list_size;
817         skb->truesize += frag_list_size;
818         return skb;
819 }
820
821 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
822                                struct ixgbe_ring *rx_ring,
823                                int *work_done, int work_to_do)
824 {
825         struct ixgbe_adapter *adapter = q_vector->adapter;
826         struct net_device *netdev = adapter->netdev;
827         struct pci_dev *pdev = adapter->pdev;
828         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
829         struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
830         struct sk_buff *skb;
831         unsigned int i, rsc_count = 0;
832         u32 len, staterr;
833         u16 hdr_info;
834         bool cleaned = false;
835         int cleaned_count = 0;
836         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
837 #ifdef IXGBE_FCOE
838         int ddp_bytes = 0;
839 #endif /* IXGBE_FCOE */
840
841         i = rx_ring->next_to_clean;
842         rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
843         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
844         rx_buffer_info = &rx_ring->rx_buffer_info[i];
845
846         while (staterr & IXGBE_RXD_STAT_DD) {
847                 u32 upper_len = 0;
848                 if (*work_done >= work_to_do)
849                         break;
850                 (*work_done)++;
851
852                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
853                         hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
854                         len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
855                                IXGBE_RXDADV_HDRBUFLEN_SHIFT;
856                         if (len > IXGBE_RX_HDR_SIZE)
857                                 len = IXGBE_RX_HDR_SIZE;
858                         upper_len = le16_to_cpu(rx_desc->wb.upper.length);
859                 } else {
860                         len = le16_to_cpu(rx_desc->wb.upper.length);
861                 }
862
863                 cleaned = true;
864                 skb = rx_buffer_info->skb;
865                 prefetch(skb->data);
866                 rx_buffer_info->skb = NULL;
867
868                 if (rx_buffer_info->dma) {
869                         pci_unmap_single(pdev, rx_buffer_info->dma,
870                                          rx_ring->rx_buf_len,
871                                          PCI_DMA_FROMDEVICE);
872                         rx_buffer_info->dma = 0;
873                         skb_put(skb, len);
874                 }
875
876                 if (upper_len) {
877                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
878                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
879                         rx_buffer_info->page_dma = 0;
880                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
881                                            rx_buffer_info->page,
882                                            rx_buffer_info->page_offset,
883                                            upper_len);
884
885                         if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
886                             (page_count(rx_buffer_info->page) != 1))
887                                 rx_buffer_info->page = NULL;
888                         else
889                                 get_page(rx_buffer_info->page);
890
891                         skb->len += upper_len;
892                         skb->data_len += upper_len;
893                         skb->truesize += upper_len;
894                 }
895
896                 i++;
897                 if (i == rx_ring->count)
898                         i = 0;
899
900                 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
901                 prefetch(next_rxd);
902                 cleaned_count++;
903
904                 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
905                         rsc_count = ixgbe_get_rsc_count(rx_desc);
906
907                 if (rsc_count) {
908                         u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
909                                      IXGBE_RXDADV_NEXTP_SHIFT;
910                         next_buffer = &rx_ring->rx_buffer_info[nextp];
911                 } else {
912                         next_buffer = &rx_ring->rx_buffer_info[i];
913                 }
914
915                 if (staterr & IXGBE_RXD_STAT_EOP) {
916                         if (skb->prev)
917                                 skb = ixgbe_transform_rsc_queue(skb, &(rx_ring->rsc_count));
918                         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
919                                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)
920                                         rx_ring->rsc_count += skb_shinfo(skb)->nr_frags;
921                                 else
922                                         rx_ring->rsc_count++;
923                                 rx_ring->rsc_flush++;
924                         }
925                         rx_ring->stats.packets++;
926                         rx_ring->stats.bytes += skb->len;
927                 } else {
928                         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
929                                 rx_buffer_info->skb = next_buffer->skb;
930                                 rx_buffer_info->dma = next_buffer->dma;
931                                 next_buffer->skb = skb;
932                                 next_buffer->dma = 0;
933                         } else {
934                                 skb->next = next_buffer->skb;
935                                 skb->next->prev = skb;
936                         }
937                         rx_ring->non_eop_descs++;
938                         goto next_desc;
939                 }
940
941                 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
942                         dev_kfree_skb_irq(skb);
943                         goto next_desc;
944                 }
945
946                 ixgbe_rx_checksum(adapter, rx_desc, skb);
947
948                 /* probably a little skewed due to removing CRC */
949                 total_rx_bytes += skb->len;
950                 total_rx_packets++;
951
952                 skb->protocol = eth_type_trans(skb, adapter->netdev);
953 #ifdef IXGBE_FCOE
954                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
955                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
956                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
957                         if (!ddp_bytes)
958                                 goto next_desc;
959                 }
960 #endif /* IXGBE_FCOE */
961                 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
962
963 next_desc:
964                 rx_desc->wb.upper.status_error = 0;
965
966                 /* return some buffers to hardware, one at a time is too slow */
967                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
968                         ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
969                         cleaned_count = 0;
970                 }
971
972                 /* use prefetched values */
973                 rx_desc = next_rxd;
974                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
975
976                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
977         }
978
979         rx_ring->next_to_clean = i;
980         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
981
982         if (cleaned_count)
983                 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
984
985 #ifdef IXGBE_FCOE
986         /* include DDPed FCoE data */
987         if (ddp_bytes > 0) {
988                 unsigned int mss;
989
990                 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
991                         sizeof(struct fc_frame_header) -
992                         sizeof(struct fcoe_crc_eof);
993                 if (mss > 512)
994                         mss &= ~511;
995                 total_rx_bytes += ddp_bytes;
996                 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
997         }
998 #endif /* IXGBE_FCOE */
999
1000         rx_ring->total_packets += total_rx_packets;
1001         rx_ring->total_bytes += total_rx_bytes;
1002         netdev->stats.rx_bytes += total_rx_bytes;
1003         netdev->stats.rx_packets += total_rx_packets;
1004
1005         return cleaned;
1006 }
1007
1008 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1009 /**
1010  * ixgbe_configure_msix - Configure MSI-X hardware
1011  * @adapter: board private structure
1012  *
1013  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1014  * interrupts.
1015  **/
1016 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1017 {
1018         struct ixgbe_q_vector *q_vector;
1019         int i, j, q_vectors, v_idx, r_idx;
1020         u32 mask;
1021
1022         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1023
1024         /*
1025          * Populate the IVAR table and set the ITR values to the
1026          * corresponding register.
1027          */
1028         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1029                 q_vector = adapter->q_vector[v_idx];
1030                 /* XXX for_each_bit(...) */
1031                 r_idx = find_first_bit(q_vector->rxr_idx,
1032                                        adapter->num_rx_queues);
1033
1034                 for (i = 0; i < q_vector->rxr_count; i++) {
1035                         j = adapter->rx_ring[r_idx].reg_idx;
1036                         ixgbe_set_ivar(adapter, 0, j, v_idx);
1037                         r_idx = find_next_bit(q_vector->rxr_idx,
1038                                               adapter->num_rx_queues,
1039                                               r_idx + 1);
1040                 }
1041                 r_idx = find_first_bit(q_vector->txr_idx,
1042                                        adapter->num_tx_queues);
1043
1044                 for (i = 0; i < q_vector->txr_count; i++) {
1045                         j = adapter->tx_ring[r_idx].reg_idx;
1046                         ixgbe_set_ivar(adapter, 1, j, v_idx);
1047                         r_idx = find_next_bit(q_vector->txr_idx,
1048                                               adapter->num_tx_queues,
1049                                               r_idx + 1);
1050                 }
1051
1052                 if (q_vector->txr_count && !q_vector->rxr_count)
1053                         /* tx only */
1054                         q_vector->eitr = adapter->tx_eitr_param;
1055                 else if (q_vector->rxr_count)
1056                         /* rx or mixed */
1057                         q_vector->eitr = adapter->rx_eitr_param;
1058
1059                 ixgbe_write_eitr(q_vector);
1060         }
1061
1062         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1063                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1064                                v_idx);
1065         else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1066                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1067         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1068
1069         /* set up to autoclear timer, and the vectors */
1070         mask = IXGBE_EIMS_ENABLE_MASK;
1071         if (adapter->num_vfs)
1072                 mask &= ~(IXGBE_EIMS_OTHER |
1073                           IXGBE_EIMS_MAILBOX |
1074                           IXGBE_EIMS_LSC);
1075         else
1076                 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1077         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1078 }
1079
1080 enum latency_range {
1081         lowest_latency = 0,
1082         low_latency = 1,
1083         bulk_latency = 2,
1084         latency_invalid = 255
1085 };
1086
1087 /**
1088  * ixgbe_update_itr - update the dynamic ITR value based on statistics
1089  * @adapter: pointer to adapter
1090  * @eitr: eitr setting (ints per sec) to give last timeslice
1091  * @itr_setting: current throttle rate in ints/second
1092  * @packets: the number of packets during this measurement interval
1093  * @bytes: the number of bytes during this measurement interval
1094  *
1095  *      Stores a new ITR value based on packets and byte
1096  *      counts during the last interrupt.  The advantage of per interrupt
1097  *      computation is faster updates and more accurate ITR for the current
1098  *      traffic pattern.  Constants in this function were computed
1099  *      based on theoretical maximum wire speed and thresholds were set based
1100  *      on testing data as well as attempting to minimize response time
1101  *      while increasing bulk throughput.
1102  *      this functionality is controlled by the InterruptThrottleRate module
1103  *      parameter (see ixgbe_param.c)
1104  **/
1105 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1106                            u32 eitr, u8 itr_setting,
1107                            int packets, int bytes)
1108 {
1109         unsigned int retval = itr_setting;
1110         u32 timepassed_us;
1111         u64 bytes_perint;
1112
1113         if (packets == 0)
1114                 goto update_itr_done;
1115
1116
1117         /* simple throttlerate management
1118          *    0-20MB/s lowest (100000 ints/s)
1119          *   20-100MB/s low   (20000 ints/s)
1120          *  100-1249MB/s bulk (8000 ints/s)
1121          */
1122         /* what was last interrupt timeslice? */
1123         timepassed_us = 1000000/eitr;
1124         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1125
1126         switch (itr_setting) {
1127         case lowest_latency:
1128                 if (bytes_perint > adapter->eitr_low)
1129                         retval = low_latency;
1130                 break;
1131         case low_latency:
1132                 if (bytes_perint > adapter->eitr_high)
1133                         retval = bulk_latency;
1134                 else if (bytes_perint <= adapter->eitr_low)
1135                         retval = lowest_latency;
1136                 break;
1137         case bulk_latency:
1138                 if (bytes_perint <= adapter->eitr_high)
1139                         retval = low_latency;
1140                 break;
1141         }
1142
1143 update_itr_done:
1144         return retval;
1145 }
1146
1147 /**
1148  * ixgbe_write_eitr - write EITR register in hardware specific way
1149  * @q_vector: structure containing interrupt and ring information
1150  *
1151  * This function is made to be called by ethtool and by the driver
1152  * when it needs to update EITR registers at runtime.  Hardware
1153  * specific quirks/differences are taken care of here.
1154  */
1155 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1156 {
1157         struct ixgbe_adapter *adapter = q_vector->adapter;
1158         struct ixgbe_hw *hw = &adapter->hw;
1159         int v_idx = q_vector->v_idx;
1160         u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1161
1162         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1163                 /* must write high and low 16 bits to reset counter */
1164                 itr_reg |= (itr_reg << 16);
1165         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1166                 /*
1167                  * set the WDIS bit to not clear the timer bits and cause an
1168                  * immediate assertion of the interrupt
1169                  */
1170                 itr_reg |= IXGBE_EITR_CNT_WDIS;
1171         }
1172         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1173 }
1174
1175 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1176 {
1177         struct ixgbe_adapter *adapter = q_vector->adapter;
1178         u32 new_itr;
1179         u8 current_itr, ret_itr;
1180         int i, r_idx;
1181         struct ixgbe_ring *rx_ring, *tx_ring;
1182
1183         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1184         for (i = 0; i < q_vector->txr_count; i++) {
1185                 tx_ring = &(adapter->tx_ring[r_idx]);
1186                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1187                                            q_vector->tx_itr,
1188                                            tx_ring->total_packets,
1189                                            tx_ring->total_bytes);
1190                 /* if the result for this queue would decrease interrupt
1191                  * rate for this vector then use that result */
1192                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1193                                     q_vector->tx_itr - 1 : ret_itr);
1194                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1195                                       r_idx + 1);
1196         }
1197
1198         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1199         for (i = 0; i < q_vector->rxr_count; i++) {
1200                 rx_ring = &(adapter->rx_ring[r_idx]);
1201                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1202                                            q_vector->rx_itr,
1203                                            rx_ring->total_packets,
1204                                            rx_ring->total_bytes);
1205                 /* if the result for this queue would decrease interrupt
1206                  * rate for this vector then use that result */
1207                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1208                                     q_vector->rx_itr - 1 : ret_itr);
1209                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1210                                       r_idx + 1);
1211         }
1212
1213         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1214
1215         switch (current_itr) {
1216         /* counts and packets in update_itr are dependent on these numbers */
1217         case lowest_latency:
1218                 new_itr = 100000;
1219                 break;
1220         case low_latency:
1221                 new_itr = 20000; /* aka hwitr = ~200 */
1222                 break;
1223         case bulk_latency:
1224         default:
1225                 new_itr = 8000;
1226                 break;
1227         }
1228
1229         if (new_itr != q_vector->eitr) {
1230                 /* do an exponential smoothing */
1231                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1232
1233                 /* save the algorithm value here, not the smoothed one */
1234                 q_vector->eitr = new_itr;
1235
1236                 ixgbe_write_eitr(q_vector);
1237         }
1238
1239         return;
1240 }
1241
1242 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1243 {
1244         struct ixgbe_hw *hw = &adapter->hw;
1245
1246         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1247             (eicr & IXGBE_EICR_GPI_SDP1)) {
1248                 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
1249                 /* write to clear the interrupt */
1250                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1251         }
1252 }
1253
1254 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1255 {
1256         struct ixgbe_hw *hw = &adapter->hw;
1257
1258         if (eicr & IXGBE_EICR_GPI_SDP1) {
1259                 /* Clear the interrupt */
1260                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1261                 schedule_work(&adapter->multispeed_fiber_task);
1262         } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1263                 /* Clear the interrupt */
1264                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1265                 schedule_work(&adapter->sfp_config_module_task);
1266         } else {
1267                 /* Interrupt isn't for us... */
1268                 return;
1269         }
1270 }
1271
1272 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1273 {
1274         struct ixgbe_hw *hw = &adapter->hw;
1275
1276         adapter->lsc_int++;
1277         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1278         adapter->link_check_timeout = jiffies;
1279         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1280                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1281                 IXGBE_WRITE_FLUSH(hw);
1282                 schedule_work(&adapter->watchdog_task);
1283         }
1284 }
1285
1286 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1287 {
1288         struct net_device *netdev = data;
1289         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1290         struct ixgbe_hw *hw = &adapter->hw;
1291         u32 eicr;
1292
1293         /*
1294          * Workaround for Silicon errata.  Use clear-by-write instead
1295          * of clear-by-read.  Reading with EICS will return the
1296          * interrupt causes without clearing, which later be done
1297          * with the write to EICR.
1298          */
1299         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1300         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1301
1302         if (eicr & IXGBE_EICR_LSC)
1303                 ixgbe_check_lsc(adapter);
1304
1305         if (eicr & IXGBE_EICR_MAILBOX)
1306                 ixgbe_msg_task(adapter);
1307
1308         if (hw->mac.type == ixgbe_mac_82598EB)
1309                 ixgbe_check_fan_failure(adapter, eicr);
1310
1311         if (hw->mac.type == ixgbe_mac_82599EB) {
1312                 ixgbe_check_sfp_event(adapter, eicr);
1313
1314                 /* Handle Flow Director Full threshold interrupt */
1315                 if (eicr & IXGBE_EICR_FLOW_DIR) {
1316                         int i;
1317                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1318                         /* Disable transmits before FDIR Re-initialization */
1319                         netif_tx_stop_all_queues(netdev);
1320                         for (i = 0; i < adapter->num_tx_queues; i++) {
1321                                 struct ixgbe_ring *tx_ring =
1322                                                            &adapter->tx_ring[i];
1323                                 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1324                                                        &tx_ring->reinit_state))
1325                                         schedule_work(&adapter->fdir_reinit_task);
1326                         }
1327                 }
1328         }
1329         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1330                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1331
1332         return IRQ_HANDLED;
1333 }
1334
1335 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1336                                            u64 qmask)
1337 {
1338         u32 mask;
1339
1340         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1341                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1342                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1343         } else {
1344                 mask = (qmask & 0xFFFFFFFF);
1345                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1346                 mask = (qmask >> 32);
1347                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1348         }
1349         /* skip the flush */
1350 }
1351
1352 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1353                                             u64 qmask)
1354 {
1355         u32 mask;
1356
1357         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1358                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1359                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1360         } else {
1361                 mask = (qmask & 0xFFFFFFFF);
1362                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1363                 mask = (qmask >> 32);
1364                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1365         }
1366         /* skip the flush */
1367 }
1368
1369 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1370 {
1371         struct ixgbe_q_vector *q_vector = data;
1372         struct ixgbe_adapter  *adapter = q_vector->adapter;
1373         struct ixgbe_ring     *tx_ring;
1374         int i, r_idx;
1375
1376         if (!q_vector->txr_count)
1377                 return IRQ_HANDLED;
1378
1379         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1380         for (i = 0; i < q_vector->txr_count; i++) {
1381                 tx_ring = &(adapter->tx_ring[r_idx]);
1382                 tx_ring->total_bytes = 0;
1383                 tx_ring->total_packets = 0;
1384                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1385                                       r_idx + 1);
1386         }
1387
1388         /* EIAM disabled interrupts (on this vector) for us */
1389         napi_schedule(&q_vector->napi);
1390
1391         return IRQ_HANDLED;
1392 }
1393
1394 /**
1395  * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1396  * @irq: unused
1397  * @data: pointer to our q_vector struct for this interrupt vector
1398  **/
1399 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1400 {
1401         struct ixgbe_q_vector *q_vector = data;
1402         struct ixgbe_adapter  *adapter = q_vector->adapter;
1403         struct ixgbe_ring  *rx_ring;
1404         int r_idx;
1405         int i;
1406
1407         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1408         for (i = 0;  i < q_vector->rxr_count; i++) {
1409                 rx_ring = &(adapter->rx_ring[r_idx]);
1410                 rx_ring->total_bytes = 0;
1411                 rx_ring->total_packets = 0;
1412                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1413                                       r_idx + 1);
1414         }
1415
1416         if (!q_vector->rxr_count)
1417                 return IRQ_HANDLED;
1418
1419         /* disable interrupts on this vector only */
1420         /* EIAM disabled interrupts (on this vector) for us */
1421         napi_schedule(&q_vector->napi);
1422
1423         return IRQ_HANDLED;
1424 }
1425
1426 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1427 {
1428         struct ixgbe_q_vector *q_vector = data;
1429         struct ixgbe_adapter  *adapter = q_vector->adapter;
1430         struct ixgbe_ring  *ring;
1431         int r_idx;
1432         int i;
1433
1434         if (!q_vector->txr_count && !q_vector->rxr_count)
1435                 return IRQ_HANDLED;
1436
1437         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1438         for (i = 0; i < q_vector->txr_count; i++) {
1439                 ring = &(adapter->tx_ring[r_idx]);
1440                 ring->total_bytes = 0;
1441                 ring->total_packets = 0;
1442                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1443                                       r_idx + 1);
1444         }
1445
1446         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1447         for (i = 0; i < q_vector->rxr_count; i++) {
1448                 ring = &(adapter->rx_ring[r_idx]);
1449                 ring->total_bytes = 0;
1450                 ring->total_packets = 0;
1451                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1452                                       r_idx + 1);
1453         }
1454
1455         /* EIAM disabled interrupts (on this vector) for us */
1456         napi_schedule(&q_vector->napi);
1457
1458         return IRQ_HANDLED;
1459 }
1460
1461 /**
1462  * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1463  * @napi: napi struct with our devices info in it
1464  * @budget: amount of work driver is allowed to do this pass, in packets
1465  *
1466  * This function is optimized for cleaning one queue only on a single
1467  * q_vector!!!
1468  **/
1469 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1470 {
1471         struct ixgbe_q_vector *q_vector =
1472                                container_of(napi, struct ixgbe_q_vector, napi);
1473         struct ixgbe_adapter *adapter = q_vector->adapter;
1474         struct ixgbe_ring *rx_ring = NULL;
1475         int work_done = 0;
1476         long r_idx;
1477
1478         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1479         rx_ring = &(adapter->rx_ring[r_idx]);
1480 #ifdef CONFIG_IXGBE_DCA
1481         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1482                 ixgbe_update_rx_dca(adapter, rx_ring);
1483 #endif
1484
1485         ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1486
1487         /* If all Rx work done, exit the polling mode */
1488         if (work_done < budget) {
1489                 napi_complete(napi);
1490                 if (adapter->rx_itr_setting & 1)
1491                         ixgbe_set_itr_msix(q_vector);
1492                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1493                         ixgbe_irq_enable_queues(adapter,
1494                                                 ((u64)1 << q_vector->v_idx));
1495         }
1496
1497         return work_done;
1498 }
1499
1500 /**
1501  * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1502  * @napi: napi struct with our devices info in it
1503  * @budget: amount of work driver is allowed to do this pass, in packets
1504  *
1505  * This function will clean more than one rx queue associated with a
1506  * q_vector.
1507  **/
1508 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1509 {
1510         struct ixgbe_q_vector *q_vector =
1511                                container_of(napi, struct ixgbe_q_vector, napi);
1512         struct ixgbe_adapter *adapter = q_vector->adapter;
1513         struct ixgbe_ring *ring = NULL;
1514         int work_done = 0, i;
1515         long r_idx;
1516         bool tx_clean_complete = true;
1517
1518         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1519         for (i = 0; i < q_vector->txr_count; i++) {
1520                 ring = &(adapter->tx_ring[r_idx]);
1521 #ifdef CONFIG_IXGBE_DCA
1522                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1523                         ixgbe_update_tx_dca(adapter, ring);
1524 #endif
1525                 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1526                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1527                                       r_idx + 1);
1528         }
1529
1530         /* attempt to distribute budget to each queue fairly, but don't allow
1531          * the budget to go below 1 because we'll exit polling */
1532         budget /= (q_vector->rxr_count ?: 1);
1533         budget = max(budget, 1);
1534         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1535         for (i = 0; i < q_vector->rxr_count; i++) {
1536                 ring = &(adapter->rx_ring[r_idx]);
1537 #ifdef CONFIG_IXGBE_DCA
1538                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1539                         ixgbe_update_rx_dca(adapter, ring);
1540 #endif
1541                 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1542                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1543                                       r_idx + 1);
1544         }
1545
1546         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1547         ring = &(adapter->rx_ring[r_idx]);
1548         /* If all Rx work done, exit the polling mode */
1549         if (work_done < budget) {
1550                 napi_complete(napi);
1551                 if (adapter->rx_itr_setting & 1)
1552                         ixgbe_set_itr_msix(q_vector);
1553                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1554                         ixgbe_irq_enable_queues(adapter,
1555                                                 ((u64)1 << q_vector->v_idx));
1556                 return 0;
1557         }
1558
1559         return work_done;
1560 }
1561
1562 /**
1563  * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1564  * @napi: napi struct with our devices info in it
1565  * @budget: amount of work driver is allowed to do this pass, in packets
1566  *
1567  * This function is optimized for cleaning one queue only on a single
1568  * q_vector!!!
1569  **/
1570 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1571 {
1572         struct ixgbe_q_vector *q_vector =
1573                                container_of(napi, struct ixgbe_q_vector, napi);
1574         struct ixgbe_adapter *adapter = q_vector->adapter;
1575         struct ixgbe_ring *tx_ring = NULL;
1576         int work_done = 0;
1577         long r_idx;
1578
1579         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1580         tx_ring = &(adapter->tx_ring[r_idx]);
1581 #ifdef CONFIG_IXGBE_DCA
1582         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1583                 ixgbe_update_tx_dca(adapter, tx_ring);
1584 #endif
1585
1586         if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
1587                 work_done = budget;
1588
1589         /* If all Tx work done, exit the polling mode */
1590         if (work_done < budget) {
1591                 napi_complete(napi);
1592                 if (adapter->tx_itr_setting & 1)
1593                         ixgbe_set_itr_msix(q_vector);
1594                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1595                         ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
1596         }
1597
1598         return work_done;
1599 }
1600
1601 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1602                                      int r_idx)
1603 {
1604         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1605
1606         set_bit(r_idx, q_vector->rxr_idx);
1607         q_vector->rxr_count++;
1608 }
1609
1610 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1611                                      int t_idx)
1612 {
1613         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1614
1615         set_bit(t_idx, q_vector->txr_idx);
1616         q_vector->txr_count++;
1617 }
1618
1619 /**
1620  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1621  * @adapter: board private structure to initialize
1622  * @vectors: allotted vector count for descriptor rings
1623  *
1624  * This function maps descriptor rings to the queue-specific vectors
1625  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1626  * one vector per ring/queue, but on a constrained vector budget, we
1627  * group the rings as "efficiently" as possible.  You would add new
1628  * mapping configurations in here.
1629  **/
1630 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1631                                       int vectors)
1632 {
1633         int v_start = 0;
1634         int rxr_idx = 0, txr_idx = 0;
1635         int rxr_remaining = adapter->num_rx_queues;
1636         int txr_remaining = adapter->num_tx_queues;
1637         int i, j;
1638         int rqpv, tqpv;
1639         int err = 0;
1640
1641         /* No mapping required if MSI-X is disabled. */
1642         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1643                 goto out;
1644
1645         /*
1646          * The ideal configuration...
1647          * We have enough vectors to map one per queue.
1648          */
1649         if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1650                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1651                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1652
1653                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1654                         map_vector_to_txq(adapter, v_start, txr_idx);
1655
1656                 goto out;
1657         }
1658
1659         /*
1660          * If we don't have enough vectors for a 1-to-1
1661          * mapping, we'll have to group them so there are
1662          * multiple queues per vector.
1663          */
1664         /* Re-adjusting *qpv takes care of the remainder. */
1665         for (i = v_start; i < vectors; i++) {
1666                 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1667                 for (j = 0; j < rqpv; j++) {
1668                         map_vector_to_rxq(adapter, i, rxr_idx);
1669                         rxr_idx++;
1670                         rxr_remaining--;
1671                 }
1672         }
1673         for (i = v_start; i < vectors; i++) {
1674                 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1675                 for (j = 0; j < tqpv; j++) {
1676                         map_vector_to_txq(adapter, i, txr_idx);
1677                         txr_idx++;
1678                         txr_remaining--;
1679                 }
1680         }
1681
1682 out:
1683         return err;
1684 }
1685
1686 /**
1687  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1688  * @adapter: board private structure
1689  *
1690  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1691  * interrupts from the kernel.
1692  **/
1693 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1694 {
1695         struct net_device *netdev = adapter->netdev;
1696         irqreturn_t (*handler)(int, void *);
1697         int i, vector, q_vectors, err;
1698         int ri=0, ti=0;
1699
1700         /* Decrement for Other and TCP Timer vectors */
1701         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1702
1703         /* Map the Tx/Rx rings to the vectors we were allotted. */
1704         err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1705         if (err)
1706                 goto out;
1707
1708 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1709                          (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1710                          &ixgbe_msix_clean_many)
1711         for (vector = 0; vector < q_vectors; vector++) {
1712                 handler = SET_HANDLER(adapter->q_vector[vector]);
1713
1714                 if(handler == &ixgbe_msix_clean_rx) {
1715                         sprintf(adapter->name[vector], "%s-%s-%d",
1716                                 netdev->name, "rx", ri++);
1717                 }
1718                 else if(handler == &ixgbe_msix_clean_tx) {
1719                         sprintf(adapter->name[vector], "%s-%s-%d",
1720                                 netdev->name, "tx", ti++);
1721                 }
1722                 else
1723                         sprintf(adapter->name[vector], "%s-%s-%d",
1724                                 netdev->name, "TxRx", vector);
1725
1726                 err = request_irq(adapter->msix_entries[vector].vector,
1727                                   handler, 0, adapter->name[vector],
1728                                   adapter->q_vector[vector]);
1729                 if (err) {
1730                         DPRINTK(PROBE, ERR,
1731                                 "request_irq failed for MSIX interrupt "
1732                                 "Error: %d\n", err);
1733                         goto free_queue_irqs;
1734                 }
1735         }
1736
1737         sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1738         err = request_irq(adapter->msix_entries[vector].vector,
1739                           ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1740         if (err) {
1741                 DPRINTK(PROBE, ERR,
1742                         "request_irq for msix_lsc failed: %d\n", err);
1743                 goto free_queue_irqs;
1744         }
1745
1746         return 0;
1747
1748 free_queue_irqs:
1749         for (i = vector - 1; i >= 0; i--)
1750                 free_irq(adapter->msix_entries[--vector].vector,
1751                          adapter->q_vector[i]);
1752         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1753         pci_disable_msix(adapter->pdev);
1754         kfree(adapter->msix_entries);
1755         adapter->msix_entries = NULL;
1756 out:
1757         return err;
1758 }
1759
1760 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1761 {
1762         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1763         u8 current_itr;
1764         u32 new_itr = q_vector->eitr;
1765         struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1766         struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1767
1768         q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1769                                             q_vector->tx_itr,
1770                                             tx_ring->total_packets,
1771                                             tx_ring->total_bytes);
1772         q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1773                                             q_vector->rx_itr,
1774                                             rx_ring->total_packets,
1775                                             rx_ring->total_bytes);
1776
1777         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1778
1779         switch (current_itr) {
1780         /* counts and packets in update_itr are dependent on these numbers */
1781         case lowest_latency:
1782                 new_itr = 100000;
1783                 break;
1784         case low_latency:
1785                 new_itr = 20000; /* aka hwitr = ~200 */
1786                 break;
1787         case bulk_latency:
1788                 new_itr = 8000;
1789                 break;
1790         default:
1791                 break;
1792         }
1793
1794         if (new_itr != q_vector->eitr) {
1795                 /* do an exponential smoothing */
1796                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1797
1798                 /* save the algorithm value here, not the smoothed one */
1799                 q_vector->eitr = new_itr;
1800
1801                 ixgbe_write_eitr(q_vector);
1802         }
1803
1804         return;
1805 }
1806
1807 /**
1808  * ixgbe_irq_enable - Enable default interrupt generation settings
1809  * @adapter: board private structure
1810  **/
1811 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1812 {
1813         u32 mask;
1814
1815         mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1816         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1817                 mask |= IXGBE_EIMS_GPI_SDP1;
1818         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1819                 mask |= IXGBE_EIMS_ECC;
1820                 mask |= IXGBE_EIMS_GPI_SDP1;
1821                 mask |= IXGBE_EIMS_GPI_SDP2;
1822                 if (adapter->num_vfs)
1823                         mask |= IXGBE_EIMS_MAILBOX;
1824         }
1825         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
1826             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
1827                 mask |= IXGBE_EIMS_FLOW_DIR;
1828
1829         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1830         ixgbe_irq_enable_queues(adapter, ~0);
1831         IXGBE_WRITE_FLUSH(&adapter->hw);
1832
1833         if (adapter->num_vfs > 32) {
1834                 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1835                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1836         }
1837 }
1838
1839 /**
1840  * ixgbe_intr - legacy mode Interrupt Handler
1841  * @irq: interrupt number
1842  * @data: pointer to a network interface device structure
1843  **/
1844 static irqreturn_t ixgbe_intr(int irq, void *data)
1845 {
1846         struct net_device *netdev = data;
1847         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1848         struct ixgbe_hw *hw = &adapter->hw;
1849         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1850         u32 eicr;
1851
1852         /*
1853          * Workaround for silicon errata.  Mask the interrupts
1854          * before the read of EICR.
1855          */
1856         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1857
1858         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1859          * therefore no explict interrupt disable is necessary */
1860         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1861         if (!eicr) {
1862                 /* shared interrupt alert!
1863                  * make sure interrupts are enabled because the read will
1864                  * have disabled interrupts due to EIAM */
1865                 ixgbe_irq_enable(adapter);
1866                 return IRQ_NONE;        /* Not our interrupt */
1867         }
1868
1869         if (eicr & IXGBE_EICR_LSC)
1870                 ixgbe_check_lsc(adapter);
1871
1872         if (hw->mac.type == ixgbe_mac_82599EB)
1873                 ixgbe_check_sfp_event(adapter, eicr);
1874
1875         ixgbe_check_fan_failure(adapter, eicr);
1876
1877         if (napi_schedule_prep(&(q_vector->napi))) {
1878                 adapter->tx_ring[0].total_packets = 0;
1879                 adapter->tx_ring[0].total_bytes = 0;
1880                 adapter->rx_ring[0].total_packets = 0;
1881                 adapter->rx_ring[0].total_bytes = 0;
1882                 /* would disable interrupts here but EIAM disabled it */
1883                 __napi_schedule(&(q_vector->napi));
1884         }
1885
1886         return IRQ_HANDLED;
1887 }
1888
1889 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1890 {
1891         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1892
1893         for (i = 0; i < q_vectors; i++) {
1894                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
1895                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1896                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1897                 q_vector->rxr_count = 0;
1898                 q_vector->txr_count = 0;
1899         }
1900 }
1901
1902 /**
1903  * ixgbe_request_irq - initialize interrupts
1904  * @adapter: board private structure
1905  *
1906  * Attempts to configure interrupts using the best available
1907  * capabilities of the hardware and kernel.
1908  **/
1909 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1910 {
1911         struct net_device *netdev = adapter->netdev;
1912         int err;
1913
1914         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1915                 err = ixgbe_request_msix_irqs(adapter);
1916         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1917                 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
1918                                   netdev->name, netdev);
1919         } else {
1920                 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
1921                                   netdev->name, netdev);
1922         }
1923
1924         if (err)
1925                 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1926
1927         return err;
1928 }
1929
1930 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1931 {
1932         struct net_device *netdev = adapter->netdev;
1933
1934         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1935                 int i, q_vectors;
1936
1937                 q_vectors = adapter->num_msix_vectors;
1938
1939                 i = q_vectors - 1;
1940                 free_irq(adapter->msix_entries[i].vector, netdev);
1941
1942                 i--;
1943                 for (; i >= 0; i--) {
1944                         free_irq(adapter->msix_entries[i].vector,
1945                                  adapter->q_vector[i]);
1946                 }
1947
1948                 ixgbe_reset_q_vectors(adapter);
1949         } else {
1950                 free_irq(adapter->pdev->irq, netdev);
1951         }
1952 }
1953
1954 /**
1955  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1956  * @adapter: board private structure
1957  **/
1958 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1959 {
1960         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1961                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1962         } else {
1963                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
1964                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
1965                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
1966                 if (adapter->num_vfs > 32)
1967                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
1968         }
1969         IXGBE_WRITE_FLUSH(&adapter->hw);
1970         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1971                 int i;
1972                 for (i = 0; i < adapter->num_msix_vectors; i++)
1973                         synchronize_irq(adapter->msix_entries[i].vector);
1974         } else {
1975                 synchronize_irq(adapter->pdev->irq);
1976         }
1977 }
1978
1979 /**
1980  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1981  *
1982  **/
1983 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1984 {
1985         struct ixgbe_hw *hw = &adapter->hw;
1986
1987         IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1988                         EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
1989
1990         ixgbe_set_ivar(adapter, 0, 0, 0);
1991         ixgbe_set_ivar(adapter, 1, 0, 0);
1992
1993         map_vector_to_rxq(adapter, 0, 0);
1994         map_vector_to_txq(adapter, 0, 0);
1995
1996         DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1997 }
1998
1999 /**
2000  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2001  * @adapter: board private structure
2002  *
2003  * Configure the Tx unit of the MAC after a reset.
2004  **/
2005 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2006 {
2007         u64 tdba;
2008         struct ixgbe_hw *hw = &adapter->hw;
2009         u32 i, j, tdlen, txctrl;
2010
2011         /* Setup the HW Tx Head and Tail descriptor pointers */
2012         for (i = 0; i < adapter->num_tx_queues; i++) {
2013                 struct ixgbe_ring *ring = &adapter->tx_ring[i];
2014                 j = ring->reg_idx;
2015                 tdba = ring->dma;
2016                 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
2017                 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
2018                                 (tdba & DMA_BIT_MASK(32)));
2019                 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
2020                 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
2021                 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
2022                 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
2023                 adapter->tx_ring[i].head = IXGBE_TDH(j);
2024                 adapter->tx_ring[i].tail = IXGBE_TDT(j);
2025                 /*
2026                  * Disable Tx Head Writeback RO bit, since this hoses
2027                  * bookkeeping if things aren't delivered in order.
2028                  */
2029                 switch (hw->mac.type) {
2030                 case ixgbe_mac_82598EB:
2031                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
2032                         break;
2033                 case ixgbe_mac_82599EB:
2034                 default:
2035                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(j));
2036                         break;
2037                 }
2038                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
2039                 switch (hw->mac.type) {
2040                 case ixgbe_mac_82598EB:
2041                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
2042                         break;
2043                 case ixgbe_mac_82599EB:
2044                 default:
2045                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(j), txctrl);
2046                         break;
2047                 }
2048         }
2049
2050         if (hw->mac.type == ixgbe_mac_82599EB) {
2051                 u32 rttdcs;
2052                 u32 mask;
2053
2054                 /* disable the arbiter while setting MTQC */
2055                 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2056                 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2057                 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2058
2059                 /* set transmit pool layout */
2060                 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2061                 switch (adapter->flags & mask) {
2062
2063                 case (IXGBE_FLAG_SRIOV_ENABLED):
2064                         IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2065                                         (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2066                         break;
2067
2068                 case (IXGBE_FLAG_DCB_ENABLED):
2069                         /* We enable 8 traffic classes, DCB only */
2070                         IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2071                                       (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2072                         break;
2073
2074                 default:
2075                         IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2076                         break;
2077                 }
2078
2079                 /* re-eable the arbiter */
2080                 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2081                 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2082         }
2083 }
2084
2085 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2086
2087 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2088                                    struct ixgbe_ring *rx_ring)
2089 {
2090         u32 srrctl;
2091         int index;
2092         struct ixgbe_ring_feature *feature = adapter->ring_feature;
2093
2094         index = rx_ring->reg_idx;
2095         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2096                 unsigned long mask;
2097                 mask = (unsigned long) feature[RING_F_RSS].mask;
2098                 index = index & mask;
2099         }
2100         srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2101
2102         srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2103         srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2104
2105         srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2106                   IXGBE_SRRCTL_BSIZEHDR_MASK;
2107
2108         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2109 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2110                 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2111 #else
2112                 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2113 #endif
2114                 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2115         } else {
2116                 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2117                           IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2118                 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2119         }
2120
2121         IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2122 }
2123
2124 static u32 ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2125 {
2126         u32 mrqc = 0;
2127         int mask;
2128
2129         if (!(adapter->hw.mac.type == ixgbe_mac_82599EB))
2130                 return mrqc;
2131
2132         mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2133 #ifdef CONFIG_IXGBE_DCB
2134                                  | IXGBE_FLAG_DCB_ENABLED
2135 #endif
2136                                  | IXGBE_FLAG_SRIOV_ENABLED
2137                                 );
2138
2139         switch (mask) {
2140         case (IXGBE_FLAG_RSS_ENABLED):
2141                 mrqc = IXGBE_MRQC_RSSEN;
2142                 break;
2143         case (IXGBE_FLAG_SRIOV_ENABLED):
2144                 mrqc = IXGBE_MRQC_VMDQEN;
2145                 break;
2146 #ifdef CONFIG_IXGBE_DCB
2147         case (IXGBE_FLAG_DCB_ENABLED):
2148                 mrqc = IXGBE_MRQC_RT8TCEN;
2149                 break;
2150 #endif /* CONFIG_IXGBE_DCB */
2151         default:
2152                 break;
2153         }
2154
2155         return mrqc;
2156 }
2157
2158 /**
2159  * ixgbe_configure_rscctl - enable RSC for the indicated ring
2160  * @adapter:    address of board private structure
2161  * @index:      index of ring to set
2162  **/
2163 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter, int index)
2164 {
2165         struct ixgbe_ring *rx_ring;
2166         struct ixgbe_hw *hw = &adapter->hw;
2167         int j;
2168         u32 rscctrl;
2169         int rx_buf_len;
2170
2171         rx_ring = &adapter->rx_ring[index];
2172         j = rx_ring->reg_idx;
2173         rx_buf_len = rx_ring->rx_buf_len;
2174         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2175         rscctrl |= IXGBE_RSCCTL_RSCEN;
2176         /*
2177          * we must limit the number of descriptors so that the
2178          * total size of max desc * buf_len is not greater
2179          * than 65535
2180          */
2181         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2182 #if (MAX_SKB_FRAGS > 16)
2183                 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2184 #elif (MAX_SKB_FRAGS > 8)
2185                 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2186 #elif (MAX_SKB_FRAGS > 4)
2187                 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2188 #else
2189                 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2190 #endif
2191         } else {
2192                 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2193                         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2194                 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2195                         rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2196                 else
2197                         rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2198         }
2199         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2200 }
2201
2202 /**
2203  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2204  * @adapter: board private structure
2205  *
2206  * Configure the Rx unit of the MAC after a reset.
2207  **/
2208 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2209 {
2210         u64 rdba;
2211         struct ixgbe_hw *hw = &adapter->hw;
2212         struct ixgbe_ring *rx_ring;
2213         struct net_device *netdev = adapter->netdev;
2214         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2215         int i, j;
2216         u32 rdlen, rxctrl, rxcsum;
2217         static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2218                           0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2219                           0x6A3E67EA, 0x14364D17, 0x3BED200D};
2220         u32 fctrl, hlreg0;
2221         u32 reta = 0, mrqc = 0;
2222         u32 rdrxctl;
2223         int rx_buf_len;
2224
2225         /* Decide whether to use packet split mode or not */
2226         /* Do not use packet split if we're in SR-IOV Mode */
2227         if (!adapter->num_vfs)
2228                 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2229
2230         /* Set the RX buffer length according to the mode */
2231         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2232                 rx_buf_len = IXGBE_RX_HDR_SIZE;
2233                 if (hw->mac.type == ixgbe_mac_82599EB) {
2234                         /* PSRTYPE must be initialized in 82599 */
2235                         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2236                                       IXGBE_PSRTYPE_UDPHDR |
2237                                       IXGBE_PSRTYPE_IPV4HDR |
2238                                       IXGBE_PSRTYPE_IPV6HDR |
2239                                       IXGBE_PSRTYPE_L2HDR;
2240                         IXGBE_WRITE_REG(hw,
2241                                         IXGBE_PSRTYPE(adapter->num_vfs),
2242                                         psrtype);
2243                 }
2244         } else {
2245                 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2246                     (netdev->mtu <= ETH_DATA_LEN))
2247                         rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2248                 else
2249                         rx_buf_len = ALIGN(max_frame, 1024);
2250         }
2251
2252         fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
2253         fctrl |= IXGBE_FCTRL_BAM;
2254         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
2255         fctrl |= IXGBE_FCTRL_PMCF;
2256         IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
2257
2258         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2259         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2260                 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2261         else
2262                 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2263 #ifdef IXGBE_FCOE
2264         if (netdev->features & NETIF_F_FCOE_MTU)
2265                 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2266 #endif
2267         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2268
2269         rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
2270         /* disable receives while setting up the descriptors */
2271         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2272         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2273
2274         /*
2275          * Setup the HW Rx Head and Tail Descriptor Pointers and
2276          * the Base and Length of the Rx Descriptor Ring
2277          */
2278         for (i = 0; i < adapter->num_rx_queues; i++) {
2279                 rx_ring = &adapter->rx_ring[i];
2280                 rdba = rx_ring->dma;
2281                 j = rx_ring->reg_idx;
2282                 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_BIT_MASK(32)));
2283                 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
2284                 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
2285                 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
2286                 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
2287                 rx_ring->head = IXGBE_RDH(j);
2288                 rx_ring->tail = IXGBE_RDT(j);
2289                 rx_ring->rx_buf_len = rx_buf_len;
2290
2291                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2292                         rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2293                 else
2294                         rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2295
2296 #ifdef IXGBE_FCOE
2297                 if (netdev->features & NETIF_F_FCOE_MTU) {
2298                         struct ixgbe_ring_feature *f;
2299                         f = &adapter->ring_feature[RING_F_FCOE];
2300                         if ((i >= f->mask) && (i < f->mask + f->indices)) {
2301                                 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2302                                 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2303                                         rx_ring->rx_buf_len =
2304                                                 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2305                         }
2306                 }
2307
2308 #endif /* IXGBE_FCOE */
2309                 ixgbe_configure_srrctl(adapter, rx_ring);
2310         }
2311
2312         if (hw->mac.type == ixgbe_mac_82598EB) {
2313                 /*
2314                  * For VMDq support of different descriptor types or
2315                  * buffer sizes through the use of multiple SRRCTL
2316                  * registers, RDRXCTL.MVMEN must be set to 1
2317                  *
2318                  * also, the manual doesn't mention it clearly but DCA hints
2319                  * will only use queue 0's tags unless this bit is set.  Side
2320                  * effects of setting this bit are only that SRRCTL must be
2321                  * fully programmed [0..15]
2322                  */
2323                 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2324                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2325                 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2326         }
2327
2328         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2329                 u32 vt_reg_bits;
2330                 u32 reg_offset, vf_shift;
2331                 u32 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2332                 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN
2333                         | IXGBE_VT_CTL_REPLEN;
2334                 vt_reg_bits |= (adapter->num_vfs <<
2335                                 IXGBE_VT_CTL_POOL_SHIFT);
2336                 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2337                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, 0);
2338
2339                 vf_shift = adapter->num_vfs % 32;
2340                 reg_offset = adapter->num_vfs / 32;
2341                 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), 0);
2342                 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), 0);
2343                 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), 0);
2344                 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), 0);
2345                 /* Enable only the PF's pool for Tx/Rx */
2346                 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2347                 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2348                 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2349                 ixgbe_set_vmolr(hw, adapter->num_vfs);
2350         }
2351
2352         /* Program MRQC for the distribution of queues */
2353         mrqc = ixgbe_setup_mrqc(adapter);
2354
2355         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2356                 /* Fill out redirection table */
2357                 for (i = 0, j = 0; i < 128; i++, j++) {
2358                         if (j == adapter->ring_feature[RING_F_RSS].indices)
2359                                 j = 0;
2360                         /* reta = 4-byte sliding window of
2361                          * 0x00..(indices-1)(indices-1)00..etc. */
2362                         reta = (reta << 8) | (j * 0x11);
2363                         if ((i & 3) == 3)
2364                                 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2365                 }
2366
2367                 /* Fill out hash function seeds */
2368                 for (i = 0; i < 10; i++)
2369                         IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2370
2371                 if (hw->mac.type == ixgbe_mac_82598EB)
2372                         mrqc |= IXGBE_MRQC_RSSEN;
2373                     /* Perform hash on these packet types */
2374                 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2375                       | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2376                       | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
2377                       | IXGBE_MRQC_RSS_FIELD_IPV6
2378                       | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
2379                       | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2380         }
2381         IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2382
2383         if (adapter->num_vfs) {
2384                 u32 reg;
2385
2386                 /* Map PF MAC address in RAR Entry 0 to first pool
2387                  * following VFs */
2388                 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2389
2390                 /* Set up VF register offsets for selected VT Mode, i.e.
2391                  * 64 VFs for SR-IOV */
2392                 reg = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2393                 reg |= IXGBE_GCR_EXT_SRIOV;
2394                 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, reg);
2395         }
2396
2397         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2398
2399         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
2400             adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
2401                 /* Disable indicating checksum in descriptor, enables
2402                  * RSS hash */
2403                 rxcsum |= IXGBE_RXCSUM_PCSD;
2404         }
2405         if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
2406                 /* Enable IPv4 payload checksum for UDP fragments
2407                  * if PCSD is not set */
2408                 rxcsum |= IXGBE_RXCSUM_IPPCSE;
2409         }
2410
2411         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2412
2413         if (hw->mac.type == ixgbe_mac_82599EB) {
2414                 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2415                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2416                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2417                 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2418         }
2419
2420         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
2421                 /* Enable 82599 HW-RSC */
2422                 for (i = 0; i < adapter->num_rx_queues; i++)
2423                         ixgbe_configure_rscctl(adapter, i);
2424
2425                 /* Disable RSC for ACK packets */
2426                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2427                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2428         }
2429 }
2430
2431 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2432 {
2433         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2434         struct ixgbe_hw *hw = &adapter->hw;
2435
2436         /* add VID to filter table */
2437         hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
2438 }
2439
2440 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2441 {
2442         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2443         struct ixgbe_hw *hw = &adapter->hw;
2444
2445         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2446                 ixgbe_irq_disable(adapter);
2447
2448         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2449
2450         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2451                 ixgbe_irq_enable(adapter);
2452
2453         /* remove VID from filter table */
2454         hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
2455 }
2456
2457 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2458                                    struct vlan_group *grp)
2459 {
2460         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2461         u32 ctrl;
2462         int i, j;
2463
2464         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2465                 ixgbe_irq_disable(adapter);
2466         adapter->vlgrp = grp;
2467
2468         /*
2469          * For a DCB driver, always enable VLAN tag stripping so we can
2470          * still receive traffic from a DCB-enabled host even if we're
2471          * not in DCB mode.
2472          */
2473         ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2474
2475         /* Disable CFI check */
2476         ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2477
2478         /* enable VLAN tag stripping */
2479         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2480                 ctrl |= IXGBE_VLNCTRL_VME;
2481         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2482                 for (i = 0; i < adapter->num_rx_queues; i++) {
2483                         u32 ctrl;
2484                         j = adapter->rx_ring[i].reg_idx;
2485                         ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_RXDCTL(j));
2486                         ctrl |= IXGBE_RXDCTL_VME;
2487                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_RXDCTL(j), ctrl);
2488                 }
2489         }
2490
2491         IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2492
2493         ixgbe_vlan_rx_add_vid(netdev, 0);
2494
2495         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2496                 ixgbe_irq_enable(adapter);
2497 }
2498
2499 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2500 {
2501         ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2502
2503         if (adapter->vlgrp) {
2504                 u16 vid;
2505                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2506                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2507                                 continue;
2508                         ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2509                 }
2510         }
2511 }
2512
2513 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
2514 {
2515         struct dev_mc_list *mc_ptr;
2516         u8 *addr = *mc_addr_ptr;
2517         *vmdq = 0;
2518
2519         mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
2520         if (mc_ptr->next)
2521                 *mc_addr_ptr = mc_ptr->next->dmi_addr;
2522         else
2523                 *mc_addr_ptr = NULL;
2524
2525         return addr;
2526 }
2527
2528 /**
2529  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
2530  * @netdev: network interface device structure
2531  *
2532  * The set_rx_method entry point is called whenever the unicast/multicast
2533  * address list or the network interface flags are updated.  This routine is
2534  * responsible for configuring the hardware for proper unicast, multicast and
2535  * promiscuous mode.
2536  **/
2537 void ixgbe_set_rx_mode(struct net_device *netdev)
2538 {
2539         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2540         struct ixgbe_hw *hw = &adapter->hw;
2541         u32 fctrl, vlnctrl;
2542         u8 *addr_list = NULL;
2543         int addr_count = 0;
2544
2545         /* Check for Promiscuous and All Multicast modes */
2546
2547         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2548         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2549
2550         if (netdev->flags & IFF_PROMISC) {
2551                 hw->addr_ctrl.user_set_promisc = 1;
2552                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2553                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2554         } else {
2555                 if (netdev->flags & IFF_ALLMULTI) {
2556                         fctrl |= IXGBE_FCTRL_MPE;
2557                         fctrl &= ~IXGBE_FCTRL_UPE;
2558                 } else {
2559                         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2560                 }
2561                 vlnctrl |= IXGBE_VLNCTRL_VFE;
2562                 hw->addr_ctrl.user_set_promisc = 0;
2563         }
2564
2565         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
2566         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2567
2568         /* reprogram secondary unicast list */
2569         hw->mac.ops.update_uc_addr_list(hw, &netdev->uc.list);
2570
2571         /* reprogram multicast list */
2572         addr_count = netdev->mc_count;
2573         if (addr_count)
2574                 addr_list = netdev->mc_list->dmi_addr;
2575         hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
2576                                         ixgbe_addr_list_itr);
2577         if (adapter->num_vfs)
2578                 ixgbe_restore_vf_multicasts(adapter);
2579 }
2580
2581 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
2582 {
2583         int q_idx;
2584         struct ixgbe_q_vector *q_vector;
2585         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2586
2587         /* legacy and MSI only use one vector */
2588         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2589                 q_vectors = 1;
2590
2591         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2592                 struct napi_struct *napi;
2593                 q_vector = adapter->q_vector[q_idx];
2594                 napi = &q_vector->napi;
2595                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2596                         if (!q_vector->rxr_count || !q_vector->txr_count) {
2597                                 if (q_vector->txr_count == 1)
2598                                         napi->poll = &ixgbe_clean_txonly;
2599                                 else if (q_vector->rxr_count == 1)
2600                                         napi->poll = &ixgbe_clean_rxonly;
2601                         }
2602                 }
2603
2604                 napi_enable(napi);
2605         }
2606 }
2607
2608 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2609 {
2610         int q_idx;
2611         struct ixgbe_q_vector *q_vector;
2612         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2613
2614         /* legacy and MSI only use one vector */
2615         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2616                 q_vectors = 1;
2617
2618         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2619                 q_vector = adapter->q_vector[q_idx];
2620                 napi_disable(&q_vector->napi);
2621         }
2622 }
2623
2624 #ifdef CONFIG_IXGBE_DCB
2625 /*
2626  * ixgbe_configure_dcb - Configure DCB hardware
2627  * @adapter: ixgbe adapter struct
2628  *
2629  * This is called by the driver on open to configure the DCB hardware.
2630  * This is also called by the gennetlink interface when reconfiguring
2631  * the DCB state.
2632  */
2633 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2634 {
2635         struct ixgbe_hw *hw = &adapter->hw;
2636         u32 txdctl, vlnctrl;
2637         int i, j;
2638
2639         ixgbe_dcb_check_config(&adapter->dcb_cfg);
2640         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2641         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2642
2643         /* reconfigure the hardware */
2644         ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
2645
2646         for (i = 0; i < adapter->num_tx_queues; i++) {
2647                 j = adapter->tx_ring[i].reg_idx;
2648                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2649                 /* PThresh workaround for Tx hang with DFP enabled. */
2650                 txdctl |= 32;
2651                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2652         }
2653         /* Enable VLAN tag insert/strip */
2654         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2655         if (hw->mac.type == ixgbe_mac_82598EB) {
2656                 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2657                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2658                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2659         } else if (hw->mac.type == ixgbe_mac_82599EB) {
2660                 vlnctrl |= IXGBE_VLNCTRL_VFE;
2661                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2662                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2663                 for (i = 0; i < adapter->num_rx_queues; i++) {
2664                         j = adapter->rx_ring[i].reg_idx;
2665                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2666                         vlnctrl |= IXGBE_RXDCTL_VME;
2667                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2668                 }
2669         }
2670         hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2671 }
2672
2673 #endif
2674 static void ixgbe_configure(struct ixgbe_adapter *adapter)
2675 {
2676         struct net_device *netdev = adapter->netdev;
2677         struct ixgbe_hw *hw = &adapter->hw;
2678         int i;
2679
2680         ixgbe_set_rx_mode(netdev);
2681
2682         ixgbe_restore_vlan(adapter);
2683 #ifdef CONFIG_IXGBE_DCB
2684         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2685                 if (hw->mac.type == ixgbe_mac_82598EB)
2686                         netif_set_gso_max_size(netdev, 32768);
2687                 else
2688                         netif_set_gso_max_size(netdev, 65536);
2689                 ixgbe_configure_dcb(adapter);
2690         } else {
2691                 netif_set_gso_max_size(netdev, 65536);
2692         }
2693 #else
2694         netif_set_gso_max_size(netdev, 65536);
2695 #endif
2696
2697 #ifdef IXGBE_FCOE
2698         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2699                 ixgbe_configure_fcoe(adapter);
2700
2701 #endif /* IXGBE_FCOE */
2702         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2703                 for (i = 0; i < adapter->num_tx_queues; i++)
2704                         adapter->tx_ring[i].atr_sample_rate =
2705                                                        adapter->atr_sample_rate;
2706                 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
2707         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
2708                 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
2709         }
2710
2711         ixgbe_configure_tx(adapter);
2712         ixgbe_configure_rx(adapter);
2713         for (i = 0; i < adapter->num_rx_queues; i++)
2714                 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
2715                                        (adapter->rx_ring[i].count - 1));
2716 }
2717
2718 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2719 {
2720         switch (hw->phy.type) {
2721         case ixgbe_phy_sfp_avago:
2722         case ixgbe_phy_sfp_ftl:
2723         case ixgbe_phy_sfp_intel:
2724         case ixgbe_phy_sfp_unknown:
2725         case ixgbe_phy_tw_tyco:
2726         case ixgbe_phy_tw_unknown:
2727                 return true;
2728         default:
2729                 return false;
2730         }
2731 }
2732
2733 /**
2734  * ixgbe_sfp_link_config - set up SFP+ link
2735  * @adapter: pointer to private adapter struct
2736  **/
2737 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2738 {
2739         struct ixgbe_hw *hw = &adapter->hw;
2740
2741                 if (hw->phy.multispeed_fiber) {
2742                         /*
2743                          * In multispeed fiber setups, the device may not have
2744                          * had a physical connection when the driver loaded.
2745                          * If that's the case, the initial link configuration
2746                          * couldn't get the MAC into 10G or 1G mode, so we'll
2747                          * never have a link status change interrupt fire.
2748                          * We need to try and force an autonegotiation
2749                          * session, then bring up link.
2750                          */
2751                         hw->mac.ops.setup_sfp(hw);
2752                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2753                                 schedule_work(&adapter->multispeed_fiber_task);
2754                 } else {
2755                         /*
2756                          * Direct Attach Cu and non-multispeed fiber modules
2757                          * still need to be configured properly prior to
2758                          * attempting link.
2759                          */
2760                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2761                                 schedule_work(&adapter->sfp_config_module_task);
2762                 }
2763 }
2764
2765 /**
2766  * ixgbe_non_sfp_link_config - set up non-SFP+ link
2767  * @hw: pointer to private hardware struct
2768  *
2769  * Returns 0 on success, negative on failure
2770  **/
2771 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2772 {
2773         u32 autoneg;
2774         bool negotiation, link_up = false;
2775         u32 ret = IXGBE_ERR_LINK_SETUP;
2776
2777         if (hw->mac.ops.check_link)
2778                 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2779
2780         if (ret)
2781                 goto link_cfg_out;
2782
2783         if (hw->mac.ops.get_link_capabilities)
2784                 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
2785         if (ret)
2786                 goto link_cfg_out;
2787
2788         if (hw->mac.ops.setup_link)
2789                 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
2790 link_cfg_out:
2791         return ret;
2792 }
2793
2794 #define IXGBE_MAX_RX_DESC_POLL 10
2795 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2796                                               int rxr)
2797 {
2798         int j = adapter->rx_ring[rxr].reg_idx;
2799         int k;
2800
2801         for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2802                 if (IXGBE_READ_REG(&adapter->hw,
2803                                    IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2804                         break;
2805                 else
2806                         msleep(1);
2807         }
2808         if (k >= IXGBE_MAX_RX_DESC_POLL) {
2809                 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2810                         "not set within the polling period\n", rxr);
2811         }
2812         ixgbe_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
2813                               (adapter->rx_ring[rxr].count - 1));
2814 }
2815
2816 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2817 {
2818         struct net_device *netdev = adapter->netdev;
2819         struct ixgbe_hw *hw = &adapter->hw;
2820         int i, j = 0;
2821         int num_rx_rings = adapter->num_rx_queues;
2822         int err;
2823         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2824         u32 txdctl, rxdctl, mhadd;
2825         u32 dmatxctl;
2826         u32 gpie;
2827
2828         ixgbe_get_hw_control(adapter);
2829
2830         if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2831             (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2832                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2833                         gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2834                                 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2835                 } else {
2836                         /* MSI only */
2837                         gpie = 0;
2838                 }
2839                 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2840                         gpie &= ~IXGBE_GPIE_VTMODE_MASK;
2841                         gpie |= IXGBE_GPIE_VTMODE_64;
2842                 }
2843                 /* XXX: to interrupt immediately for EICS writes, enable this */
2844                 /* gpie |= IXGBE_GPIE_EIMEN; */
2845                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2846         }
2847
2848         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2849                 /*
2850                  * use EIAM to auto-mask when MSI-X interrupt is asserted
2851                  * this saves a register write for every interrupt
2852                  */
2853                 switch (hw->mac.type) {
2854                 case ixgbe_mac_82598EB:
2855                         IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2856                         break;
2857                 default:
2858                 case ixgbe_mac_82599EB:
2859                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
2860                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
2861                         break;
2862                 }
2863         } else {
2864                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2865                  * specifically only auto mask tx and rx interrupts */
2866                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2867         }
2868
2869         /* Enable fan failure interrupt if media type is copper */
2870         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2871                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2872                 gpie |= IXGBE_SDP1_GPIEN;
2873                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2874         }
2875
2876         if (hw->mac.type == ixgbe_mac_82599EB) {
2877                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2878                 gpie |= IXGBE_SDP1_GPIEN;
2879                 gpie |= IXGBE_SDP2_GPIEN;
2880                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2881         }
2882
2883 #ifdef IXGBE_FCOE
2884         /* adjust max frame to be able to do baby jumbo for FCoE */
2885         if ((netdev->features & NETIF_F_FCOE_MTU) &&
2886             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2887                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2888
2889 #endif /* IXGBE_FCOE */
2890         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2891         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2892                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2893                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2894
2895                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2896         }
2897
2898         for (i = 0; i < adapter->num_tx_queues; i++) {
2899                 j = adapter->tx_ring[i].reg_idx;
2900                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2901                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2902                 txdctl |= (8 << 16);
2903                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2904         }
2905
2906         if (hw->mac.type == ixgbe_mac_82599EB) {
2907                 /* DMATXCTL.EN must be set after all Tx queue config is done */
2908                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2909                 dmatxctl |= IXGBE_DMATXCTL_TE;
2910                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2911         }
2912         for (i = 0; i < adapter->num_tx_queues; i++) {
2913                 j = adapter->tx_ring[i].reg_idx;
2914                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2915                 txdctl |= IXGBE_TXDCTL_ENABLE;
2916                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2917                 if (hw->mac.type == ixgbe_mac_82599EB) {
2918                         int wait_loop = 10;
2919                         /* poll for Tx Enable ready */
2920                         do {
2921                                 msleep(1);
2922                                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2923                         } while (--wait_loop &&
2924                                  !(txdctl & IXGBE_TXDCTL_ENABLE));
2925                         if (!wait_loop)
2926                                 DPRINTK(DRV, ERR, "Could not enable "
2927                                         "Tx Queue %d\n", j);
2928                 }
2929         }
2930
2931         for (i = 0; i < num_rx_rings; i++) {
2932                 j = adapter->rx_ring[i].reg_idx;
2933                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2934                 /* enable PTHRESH=32 descriptors (half the internal cache)
2935                  * and HTHRESH=0 descriptors (to minimize latency on fetch),
2936                  * this also removes a pesky rx_no_buffer_count increment */
2937                 rxdctl |= 0x0020;
2938                 rxdctl |= IXGBE_RXDCTL_ENABLE;
2939                 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2940                 if (hw->mac.type == ixgbe_mac_82599EB)
2941                         ixgbe_rx_desc_queue_enable(adapter, i);
2942         }
2943         /* enable all receives */
2944         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2945         if (hw->mac.type == ixgbe_mac_82598EB)
2946                 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2947         else
2948                 rxdctl |= IXGBE_RXCTRL_RXEN;
2949         hw->mac.ops.enable_rx_dma(hw, rxdctl);
2950
2951         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2952                 ixgbe_configure_msix(adapter);
2953         else
2954                 ixgbe_configure_msi_and_legacy(adapter);
2955
2956         clear_bit(__IXGBE_DOWN, &adapter->state);
2957         ixgbe_napi_enable_all(adapter);
2958
2959         /* clear any pending interrupts, may auto mask */
2960         IXGBE_READ_REG(hw, IXGBE_EICR);
2961
2962         ixgbe_irq_enable(adapter);
2963
2964         /*
2965          * If this adapter has a fan, check to see if we had a failure
2966          * before we enabled the interrupt.
2967          */
2968         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2969                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
2970                 if (esdp & IXGBE_ESDP_SDP1)
2971                         DPRINTK(DRV, CRIT,
2972                                 "Fan has stopped, replace the adapter\n");
2973         }
2974
2975         /*
2976          * For hot-pluggable SFP+ devices, a new SFP+ module may have
2977          * arrived before interrupts were enabled but after probe.  Such
2978          * devices wouldn't have their type identified yet. We need to
2979          * kick off the SFP+ module setup first, then try to bring up link.
2980          * If we're not hot-pluggable SFP+, we just need to configure link
2981          * and bring it up.
2982          */
2983         if (hw->phy.type == ixgbe_phy_unknown) {
2984                 err = hw->phy.ops.identify(hw);
2985                 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
2986                         /*
2987                          * Take the device down and schedule the sfp tasklet
2988                          * which will unregister_netdev and log it.
2989                          */
2990                         ixgbe_down(adapter);
2991                         schedule_work(&adapter->sfp_config_module_task);
2992                         return err;
2993                 }
2994         }
2995
2996         if (ixgbe_is_sfp(hw)) {
2997                 ixgbe_sfp_link_config(adapter);
2998         } else {
2999                 err = ixgbe_non_sfp_link_config(hw);
3000                 if (err)
3001                         DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
3002         }
3003
3004         for (i = 0; i < adapter->num_tx_queues; i++)
3005                 set_bit(__IXGBE_FDIR_INIT_DONE,
3006                         &(adapter->tx_ring[i].reinit_state));
3007
3008         /* enable transmits */
3009         netif_tx_start_all_queues(netdev);
3010
3011         /* bring the link up in the watchdog, this could race with our first
3012          * link up interrupt but shouldn't be a problem */
3013         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3014         adapter->link_check_timeout = jiffies;
3015         mod_timer(&adapter->watchdog_timer, jiffies);
3016         return 0;
3017 }
3018
3019 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3020 {
3021         WARN_ON(in_interrupt());
3022         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3023                 msleep(1);
3024         ixgbe_down(adapter);
3025         ixgbe_up(adapter);
3026         clear_bit(__IXGBE_RESETTING, &adapter->state);
3027 }
3028
3029 int ixgbe_up(struct ixgbe_adapter *adapter)
3030 {
3031         /* hardware has been reset, we need to reload some things */
3032         ixgbe_configure(adapter);
3033
3034         return ixgbe_up_complete(adapter);
3035 }
3036
3037 void ixgbe_reset(struct ixgbe_adapter *adapter)
3038 {
3039         struct ixgbe_hw *hw = &adapter->hw;
3040         int err;
3041
3042         err = hw->mac.ops.init_hw(hw);
3043         switch (err) {
3044         case 0:
3045         case IXGBE_ERR_SFP_NOT_PRESENT:
3046                 break;
3047         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3048                 dev_err(&adapter->pdev->dev, "master disable timed out\n");
3049                 break;
3050         case IXGBE_ERR_EEPROM_VERSION:
3051                 /* We are running on a pre-production device, log a warning */
3052                 dev_warn(&adapter->pdev->dev, "This device is a pre-production "
3053                          "adapter/LOM.  Please be aware there may be issues "
3054                          "associated with your hardware.  If you are "
3055                          "experiencing problems please contact your Intel or "
3056                          "hardware representative who provided you with this "
3057                          "hardware.\n");
3058                 break;
3059         default:
3060                 dev_err(&adapter->pdev->dev, "Hardware Error: %d\n", err);
3061         }
3062
3063         /* reprogram the RAR[0] in case user changed it. */
3064         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3065                             IXGBE_RAH_AV);
3066 }
3067
3068 /**
3069  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3070  * @adapter: board private structure
3071  * @rx_ring: ring to free buffers from
3072  **/
3073 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
3074                                 struct ixgbe_ring *rx_ring)
3075 {
3076         struct pci_dev *pdev = adapter->pdev;
3077         unsigned long size;
3078         unsigned int i;
3079
3080         /* Free all the Rx ring sk_buffs */
3081
3082         for (i = 0; i < rx_ring->count; i++) {
3083                 struct ixgbe_rx_buffer *rx_buffer_info;
3084
3085                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3086                 if (rx_buffer_info->dma) {
3087                         pci_unmap_single(pdev, rx_buffer_info->dma,
3088                                          rx_ring->rx_buf_len,
3089                                          PCI_DMA_FROMDEVICE);
3090                         rx_buffer_info->dma = 0;
3091                 }
3092                 if (rx_buffer_info->skb) {
3093                         struct sk_buff *skb = rx_buffer_info->skb;
3094                         rx_buffer_info->skb = NULL;
3095                         do {
3096                                 struct sk_buff *this = skb;
3097                                 skb = skb->prev;
3098                                 dev_kfree_skb(this);
3099                         } while (skb);
3100                 }
3101                 if (!rx_buffer_info->page)
3102                         continue;
3103                 if (rx_buffer_info->page_dma) {
3104                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
3105                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
3106                         rx_buffer_info->page_dma = 0;
3107                 }
3108                 put_page(rx_buffer_info->page);
3109                 rx_buffer_info->page = NULL;
3110                 rx_buffer_info->page_offset = 0;
3111         }
3112
3113         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3114         memset(rx_ring->rx_buffer_info, 0, size);
3115
3116         /* Zero out the descriptor ring */
3117         memset(rx_ring->desc, 0, rx_ring->size);
3118
3119         rx_ring->next_to_clean = 0;
3120         rx_ring->next_to_use = 0;
3121
3122         if (rx_ring->head)
3123                 writel(0, adapter->hw.hw_addr + rx_ring->head);
3124         if (rx_ring->tail)
3125                 writel(0, adapter->hw.hw_addr + rx_ring->tail);
3126 }
3127
3128 /**
3129  * ixgbe_clean_tx_ring - Free Tx Buffers
3130  * @adapter: board private structure
3131  * @tx_ring: ring to be cleaned
3132  **/
3133 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
3134                                 struct ixgbe_ring *tx_ring)
3135 {
3136         struct ixgbe_tx_buffer *tx_buffer_info;
3137         unsigned long size;
3138         unsigned int i;
3139
3140         /* Free all the Tx ring sk_buffs */
3141
3142         for (i = 0; i < tx_ring->count; i++) {
3143                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3144                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3145         }
3146
3147         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3148         memset(tx_ring->tx_buffer_info, 0, size);
3149
3150         /* Zero out the descriptor ring */
3151         memset(tx_ring->desc, 0, tx_ring->size);
3152
3153         tx_ring->next_to_use = 0;
3154         tx_ring->next_to_clean = 0;
3155
3156         if (tx_ring->head)
3157                 writel(0, adapter->hw.hw_addr + tx_ring->head);
3158         if (tx_ring->tail)
3159                 writel(0, adapter->hw.hw_addr + tx_ring->tail);
3160 }
3161
3162 /**
3163  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3164  * @adapter: board private structure
3165  **/
3166 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3167 {
3168         int i;
3169
3170         for (i = 0; i < adapter->num_rx_queues; i++)
3171                 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
3172 }
3173
3174 /**
3175  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3176  * @adapter: board private structure
3177  **/
3178 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3179 {
3180         int i;
3181
3182         for (i = 0; i < adapter->num_tx_queues; i++)
3183                 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
3184 }
3185
3186 void ixgbe_down(struct ixgbe_adapter *adapter)
3187 {
3188         struct net_device *netdev = adapter->netdev;
3189         struct ixgbe_hw *hw = &adapter->hw;
3190         u32 rxctrl;
3191         u32 txdctl;
3192         int i, j;
3193
3194         /* signal that we are down to the interrupt handler */
3195         set_bit(__IXGBE_DOWN, &adapter->state);
3196
3197         /* disable receives */
3198         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3199         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3200
3201         netif_tx_disable(netdev);
3202
3203         IXGBE_WRITE_FLUSH(hw);
3204         msleep(10);
3205
3206         netif_tx_stop_all_queues(netdev);
3207
3208         ixgbe_irq_disable(adapter);
3209
3210         ixgbe_napi_disable_all(adapter);
3211
3212         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3213         del_timer_sync(&adapter->sfp_timer);
3214         del_timer_sync(&adapter->watchdog_timer);
3215         cancel_work_sync(&adapter->watchdog_task);
3216
3217         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3218             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3219                 cancel_work_sync(&adapter->fdir_reinit_task);
3220
3221         /* disable transmits in the hardware now that interrupts are off */
3222         for (i = 0; i < adapter->num_tx_queues; i++) {
3223                 j = adapter->tx_ring[i].reg_idx;
3224                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3225                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3226                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3227         }
3228         /* Disable the Tx DMA engine on 82599 */
3229         if (hw->mac.type == ixgbe_mac_82599EB)
3230                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3231                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3232                                  ~IXGBE_DMATXCTL_TE));
3233
3234         netif_carrier_off(netdev);
3235
3236         if (!pci_channel_offline(adapter->pdev))
3237                 ixgbe_reset(adapter);
3238         ixgbe_clean_all_tx_rings(adapter);
3239         ixgbe_clean_all_rx_rings(adapter);
3240
3241 #ifdef CONFIG_IXGBE_DCA
3242         /* since we reset the hardware DCA settings were cleared */
3243         ixgbe_setup_dca(adapter);
3244 #endif
3245 }
3246
3247 /**
3248  * ixgbe_poll - NAPI Rx polling callback
3249  * @napi: structure for representing this polling device
3250  * @budget: how many packets driver is allowed to clean
3251  *
3252  * This function is used for legacy and MSI, NAPI mode
3253  **/
3254 static int ixgbe_poll(struct napi_struct *napi, int budget)
3255 {
3256         struct ixgbe_q_vector *q_vector =
3257                                 container_of(napi, struct ixgbe_q_vector, napi);
3258         struct ixgbe_adapter *adapter = q_vector->adapter;
3259         int tx_clean_complete, work_done = 0;
3260
3261 #ifdef CONFIG_IXGBE_DCA
3262         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3263                 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
3264                 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
3265         }
3266 #endif
3267
3268         tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring);
3269         ixgbe_clean_rx_irq(q_vector, adapter->rx_ring, &work_done, budget);
3270
3271         if (!tx_clean_complete)
3272                 work_done = budget;
3273
3274         /* If budget not fully consumed, exit the polling mode */
3275         if (work_done < budget) {
3276                 napi_complete(napi);
3277                 if (adapter->rx_itr_setting & 1)
3278                         ixgbe_set_itr(adapter);
3279                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3280                         ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3281         }
3282         return work_done;
3283 }
3284
3285 /**
3286  * ixgbe_tx_timeout - Respond to a Tx Hang
3287  * @netdev: network interface device structure
3288  **/
3289 static void ixgbe_tx_timeout(struct net_device *netdev)
3290 {
3291         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3292
3293         /* Do the reset outside of interrupt context */
3294         schedule_work(&adapter->reset_task);
3295 }
3296
3297 static void ixgbe_reset_task(struct work_struct *work)
3298 {
3299         struct ixgbe_adapter *adapter;
3300         adapter = container_of(work, struct ixgbe_adapter, reset_task);
3301
3302         /* If we're already down or resetting, just bail */
3303         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3304             test_bit(__IXGBE_RESETTING, &adapter->state))
3305                 return;
3306
3307         adapter->tx_timeout_count++;
3308
3309         ixgbe_reinit_locked(adapter);
3310 }
3311
3312 #ifdef CONFIG_IXGBE_DCB
3313 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3314 {
3315         bool ret = false;
3316         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3317
3318         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3319                 return ret;
3320
3321         f->mask = 0x7 << 3;
3322         adapter->num_rx_queues = f->indices;
3323         adapter->num_tx_queues = f->indices;
3324         ret = true;
3325
3326         return ret;
3327 }
3328 #endif
3329
3330 /**
3331  * ixgbe_set_rss_queues: Allocate queues for RSS
3332  * @adapter: board private structure to initialize
3333  *
3334  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
3335  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3336  *
3337  **/
3338 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3339 {
3340         bool ret = false;
3341         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3342
3343         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3344                 f->mask = 0xF;
3345                 adapter->num_rx_queues = f->indices;
3346                 adapter->num_tx_queues = f->indices;
3347                 ret = true;
3348         } else {
3349                 ret = false;
3350         }
3351
3352         return ret;
3353 }
3354
3355 /**
3356  * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3357  * @adapter: board private structure to initialize
3358  *
3359  * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3360  * to the original CPU that initiated the Tx session.  This runs in addition
3361  * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3362  * Rx load across CPUs using RSS.
3363  *
3364  **/
3365 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3366 {
3367         bool ret = false;
3368         struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3369
3370         f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3371         f_fdir->mask = 0;
3372
3373         /* Flow Director must have RSS enabled */
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                 adapter->num_tx_queues = f_fdir->indices;
3378                 adapter->num_rx_queues = f_fdir->indices;
3379                 ret = true;
3380         } else {
3381                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3382                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3383         }
3384         return ret;
3385 }
3386
3387 #ifdef IXGBE_FCOE
3388 /**
3389  * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3390  * @adapter: board private structure to initialize
3391  *
3392  * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3393  * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3394  * rx queues out of the max number of rx queues, instead, it is used as the
3395  * index of the first rx queue used by FCoE.
3396  *
3397  **/
3398 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3399 {
3400         bool ret = false;
3401         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3402
3403         f->indices = min((int)num_online_cpus(), f->indices);
3404         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3405                 adapter->num_rx_queues = 1;
3406                 adapter->num_tx_queues = 1;
3407 #ifdef CONFIG_IXGBE_DCB
3408                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3409                         DPRINTK(PROBE, INFO, "FCoE enabled with DCB \n");
3410                         ixgbe_set_dcb_queues(adapter);
3411                 }
3412 #endif
3413                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3414                         DPRINTK(PROBE, INFO, "FCoE enabled with RSS \n");
3415                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3416                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3417                                 ixgbe_set_fdir_queues(adapter);
3418                         else
3419                                 ixgbe_set_rss_queues(adapter);
3420                 }
3421                 /* adding FCoE rx rings to the end */
3422                 f->mask = adapter->num_rx_queues;
3423                 adapter->num_rx_queues += f->indices;
3424                 adapter->num_tx_queues += f->indices;
3425
3426                 ret = true;
3427         }
3428
3429         return ret;
3430 }
3431
3432 #endif /* IXGBE_FCOE */
3433 /**
3434  * ixgbe_set_sriov_queues: Allocate queues for IOV use
3435  * @adapter: board private structure to initialize
3436  *
3437  * IOV doesn't actually use anything, so just NAK the
3438  * request for now and let the other queue routines
3439  * figure out what to do.
3440  */
3441 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
3442 {
3443         return false;
3444 }
3445
3446 /*
3447  * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3448  * @adapter: board private structure to initialize
3449  *
3450  * This is the top level queue allocation routine.  The order here is very
3451  * important, starting with the "most" number of features turned on at once,
3452  * and ending with the smallest set of features.  This way large combinations
3453  * can be allocated if they're turned on, and smaller combinations are the
3454  * fallthrough conditions.
3455  *
3456  **/
3457 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
3458 {
3459         /* Start with base case */
3460         adapter->num_rx_queues = 1;
3461         adapter->num_tx_queues = 1;
3462         adapter->num_rx_pools = adapter->num_rx_queues;
3463         adapter->num_rx_queues_per_pool = 1;
3464
3465         if (ixgbe_set_sriov_queues(adapter))
3466                 return;
3467
3468 #ifdef IXGBE_FCOE
3469         if (ixgbe_set_fcoe_queues(adapter))
3470                 goto done;
3471
3472 #endif /* IXGBE_FCOE */
3473 #ifdef CONFIG_IXGBE_DCB
3474         if (ixgbe_set_dcb_queues(adapter))
3475                 goto done;
3476
3477 #endif
3478         if (ixgbe_set_fdir_queues(adapter))
3479                 goto done;
3480
3481         if (ixgbe_set_rss_queues(adapter))
3482                 goto done;
3483
3484         /* fallback to base case */
3485         adapter->num_rx_queues = 1;
3486         adapter->num_tx_queues = 1;
3487
3488 done:
3489         /* Notify the stack of the (possibly) reduced Tx Queue count. */
3490         adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
3491 }
3492
3493 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
3494                                        int vectors)
3495 {
3496         int err, vector_threshold;
3497
3498         /* We'll want at least 3 (vector_threshold):
3499          * 1) TxQ[0] Cleanup
3500          * 2) RxQ[0] Cleanup
3501          * 3) Other (Link Status Change, etc.)
3502          * 4) TCP Timer (optional)
3503          */
3504         vector_threshold = MIN_MSIX_COUNT;
3505
3506         /* The more we get, the more we will assign to Tx/Rx Cleanup
3507          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
3508          * Right now, we simply care about how many we'll get; we'll
3509          * set them up later while requesting irq's.
3510          */
3511         while (vectors >= vector_threshold) {
3512                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
3513                                       vectors);
3514                 if (!err) /* Success in acquiring all requested vectors. */
3515                         break;
3516                 else if (err < 0)
3517                         vectors = 0; /* Nasty failure, quit now */
3518                 else /* err == number of vectors we should try again with */
3519                         vectors = err;
3520         }
3521
3522         if (vectors < vector_threshold) {
3523                 /* Can't allocate enough MSI-X interrupts?  Oh well.
3524                  * This just means we'll go with either a single MSI
3525                  * vector or fall back to legacy interrupts.
3526                  */
3527                 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
3528                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3529                 kfree(adapter->msix_entries);
3530                 adapter->msix_entries = NULL;
3531         } else {
3532                 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
3533                 /*
3534                  * Adjust for only the vectors we'll use, which is minimum
3535                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
3536                  * vectors we were allocated.
3537                  */
3538                 adapter->num_msix_vectors = min(vectors,
3539                                    adapter->max_msix_q_vectors + NON_Q_VECTORS);
3540         }
3541 }
3542
3543 /**
3544  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
3545  * @adapter: board private structure to initialize
3546  *
3547  * Cache the descriptor ring offsets for RSS to the assigned rings.
3548  *
3549  **/
3550 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
3551 {
3552         int i;
3553         bool ret = false;
3554
3555         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3556                 for (i = 0; i < adapter->num_rx_queues; i++)
3557                         adapter->rx_ring[i].reg_idx = i;
3558                 for (i = 0; i < adapter->num_tx_queues; i++)
3559                         adapter->tx_ring[i].reg_idx = i;
3560                 ret = true;
3561         } else {
3562                 ret = false;
3563         }
3564
3565         return ret;
3566 }
3567
3568 #ifdef CONFIG_IXGBE_DCB
3569 /**
3570  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
3571  * @adapter: board private structure to initialize
3572  *
3573  * Cache the descriptor ring offsets for DCB to the assigned rings.
3574  *
3575  **/
3576 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
3577 {
3578         int i;
3579         bool ret = false;
3580         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
3581
3582         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3583                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3584                         /* the number of queues is assumed to be symmetric */
3585                         for (i = 0; i < dcb_i; i++) {
3586                                 adapter->rx_ring[i].reg_idx = i << 3;
3587                                 adapter->tx_ring[i].reg_idx = i << 2;
3588                         }
3589                         ret = true;
3590                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
3591                         if (dcb_i == 8) {
3592                                 /*
3593                                  * Tx TC0 starts at: descriptor queue 0
3594                                  * Tx TC1 starts at: descriptor queue 32
3595                                  * Tx TC2 starts at: descriptor queue 64
3596                                  * Tx TC3 starts at: descriptor queue 80
3597                                  * Tx TC4 starts at: descriptor queue 96
3598                                  * Tx TC5 starts at: descriptor queue 104
3599                                  * Tx TC6 starts at: descriptor queue 112
3600                                  * Tx TC7 starts at: descriptor queue 120
3601                                  *
3602                                  * Rx TC0-TC7 are offset by 16 queues each
3603                                  */
3604                                 for (i = 0; i < 3; i++) {
3605                                         adapter->tx_ring[i].reg_idx = i << 5;
3606                                         adapter->rx_ring[i].reg_idx = i << 4;
3607                                 }
3608                                 for ( ; i < 5; i++) {
3609                                         adapter->tx_ring[i].reg_idx =
3610                                                                  ((i + 2) << 4);
3611                                         adapter->rx_ring[i].reg_idx = i << 4;
3612                                 }
3613                                 for ( ; i < dcb_i; i++) {
3614                                         adapter->tx_ring[i].reg_idx =
3615                                                                  ((i + 8) << 3);
3616                                         adapter->rx_ring[i].reg_idx = i << 4;
3617                                 }
3618
3619                                 ret = true;
3620                         } else if (dcb_i == 4) {
3621                                 /*
3622                                  * Tx TC0 starts at: descriptor queue 0
3623                                  * Tx TC1 starts at: descriptor queue 64
3624                                  * Tx TC2 starts at: descriptor queue 96
3625                                  * Tx TC3 starts at: descriptor queue 112
3626                                  *
3627                                  * Rx TC0-TC3 are offset by 32 queues each
3628                                  */
3629                                 adapter->tx_ring[0].reg_idx = 0;
3630                                 adapter->tx_ring[1].reg_idx = 64;
3631                                 adapter->tx_ring[2].reg_idx = 96;
3632                                 adapter->tx_ring[3].reg_idx = 112;
3633                                 for (i = 0 ; i < dcb_i; i++)
3634                                         adapter->rx_ring[i].reg_idx = i << 5;
3635
3636                                 ret = true;
3637                         } else {
3638                                 ret = false;
3639                         }
3640                 } else {
3641                         ret = false;
3642                 }
3643         } else {
3644                 ret = false;
3645         }
3646
3647         return ret;
3648 }
3649 #endif
3650
3651 /**
3652  * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
3653  * @adapter: board private structure to initialize
3654  *
3655  * Cache the descriptor ring offsets for Flow Director to the assigned rings.
3656  *
3657  **/
3658 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
3659 {
3660         int i;
3661         bool ret = false;
3662
3663         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3664             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3665              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
3666                 for (i = 0; i < adapter->num_rx_queues; i++)
3667                         adapter->rx_ring[i].reg_idx = i;
3668                 for (i = 0; i < adapter->num_tx_queues; i++)
3669                         adapter->tx_ring[i].reg_idx = i;
3670                 ret = true;
3671         }
3672
3673         return ret;
3674 }
3675
3676 #ifdef IXGBE_FCOE
3677 /**
3678  * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
3679  * @adapter: board private structure to initialize
3680  *
3681  * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
3682  *
3683  */
3684 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
3685 {
3686         int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
3687         bool ret = false;
3688         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3689
3690         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3691 #ifdef CONFIG_IXGBE_DCB
3692                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3693                         struct ixgbe_fcoe *fcoe = &adapter->fcoe;
3694
3695                         ixgbe_cache_ring_dcb(adapter);
3696                         /* find out queues in TC for FCoE */
3697                         fcoe_rx_i = adapter->rx_ring[fcoe->tc].reg_idx + 1;
3698                         fcoe_tx_i = adapter->tx_ring[fcoe->tc].reg_idx + 1;
3699                         /*
3700                          * In 82599, the number of Tx queues for each traffic
3701                          * class for both 8-TC and 4-TC modes are:
3702                          * TCs  : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
3703                          * 8 TCs:  32  32  16  16   8   8   8   8
3704                          * 4 TCs:  64  64  32  32
3705                          * We have max 8 queues for FCoE, where 8 the is
3706                          * FCoE redirection table size. If TC for FCoE is
3707                          * less than or equal to TC3, we have enough queues
3708                          * to add max of 8 queues for FCoE, so we start FCoE
3709                          * tx descriptor from the next one, i.e., reg_idx + 1.
3710                          * If TC for FCoE is above TC3, implying 8 TC mode,
3711                          * and we need 8 for FCoE, we have to take all queues
3712                          * in that traffic class for FCoE.
3713                          */
3714                         if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
3715                                 fcoe_tx_i--;
3716                 }
3717 #endif /* CONFIG_IXGBE_DCB */
3718                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3719                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3720                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3721                                 ixgbe_cache_ring_fdir(adapter);
3722                         else
3723                                 ixgbe_cache_ring_rss(adapter);
3724
3725                         fcoe_rx_i = f->mask;
3726                         fcoe_tx_i = f->mask;
3727                 }
3728                 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
3729                         adapter->rx_ring[f->mask + i].reg_idx = fcoe_rx_i;
3730                         adapter->tx_ring[f->mask + i].reg_idx = fcoe_tx_i;
3731                 }
3732                 ret = true;
3733         }
3734         return ret;
3735 }
3736
3737 #endif /* IXGBE_FCOE */
3738 /**
3739  * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
3740  * @adapter: board private structure to initialize
3741  *
3742  * SR-IOV doesn't use any descriptor rings but changes the default if
3743  * no other mapping is used.
3744  *
3745  */
3746 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
3747 {
3748         adapter->rx_ring[0].reg_idx = adapter->num_vfs * 2;
3749         adapter->tx_ring[0].reg_idx = adapter->num_vfs * 2;
3750         if (adapter->num_vfs)
3751                 return true;
3752         else
3753                 return false;
3754 }
3755
3756 /**
3757  * ixgbe_cache_ring_register - Descriptor ring to register mapping
3758  * @adapter: board private structure to initialize
3759  *
3760  * Once we know the feature-set enabled for the device, we'll cache
3761  * the register offset the descriptor ring is assigned to.
3762  *
3763  * Note, the order the various feature calls is important.  It must start with
3764  * the "most" features enabled at the same time, then trickle down to the
3765  * least amount of features turned on at once.
3766  **/
3767 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
3768 {
3769         /* start with default case */
3770         adapter->rx_ring[0].reg_idx = 0;
3771         adapter->tx_ring[0].reg_idx = 0;
3772
3773         if (ixgbe_cache_ring_sriov(adapter))
3774                 return;
3775
3776 #ifdef IXGBE_FCOE
3777         if (ixgbe_cache_ring_fcoe(adapter))
3778                 return;
3779
3780 #endif /* IXGBE_FCOE */
3781 #ifdef CONFIG_IXGBE_DCB
3782         if (ixgbe_cache_ring_dcb(adapter))
3783                 return;
3784
3785 #endif
3786         if (ixgbe_cache_ring_fdir(adapter))
3787                 return;
3788
3789         if (ixgbe_cache_ring_rss(adapter))
3790                 return;
3791 }
3792
3793 /**
3794  * ixgbe_alloc_queues - Allocate memory for all rings
3795  * @adapter: board private structure to initialize
3796  *
3797  * We allocate one ring per queue at run-time since we don't know the
3798  * number of queues at compile-time.  The polling_netdev array is
3799  * intended for Multiqueue, but should work fine with a single queue.
3800  **/
3801 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
3802 {
3803         int i;
3804
3805         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
3806                                    sizeof(struct ixgbe_ring), GFP_KERNEL);
3807         if (!adapter->tx_ring)
3808                 goto err_tx_ring_allocation;
3809
3810         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
3811                                    sizeof(struct ixgbe_ring), GFP_KERNEL);
3812         if (!adapter->rx_ring)
3813                 goto err_rx_ring_allocation;
3814
3815         for (i = 0; i < adapter->num_tx_queues; i++) {
3816                 adapter->tx_ring[i].count = adapter->tx_ring_count;
3817                 adapter->tx_ring[i].queue_index = i;
3818         }
3819
3820         for (i = 0; i < adapter->num_rx_queues; i++) {
3821                 adapter->rx_ring[i].count = adapter->rx_ring_count;
3822                 adapter->rx_ring[i].queue_index = i;
3823         }
3824
3825         ixgbe_cache_ring_register(adapter);
3826
3827         return 0;
3828
3829 err_rx_ring_allocation:
3830         kfree(adapter->tx_ring);
3831 err_tx_ring_allocation:
3832         return -ENOMEM;
3833 }
3834
3835 /**
3836  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
3837  * @adapter: board private structure to initialize
3838  *
3839  * Attempt to configure the interrupts using the best available
3840  * capabilities of the hardware and the kernel.
3841  **/
3842 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
3843 {
3844         struct ixgbe_hw *hw = &adapter->hw;
3845         int err = 0;
3846         int vector, v_budget;
3847
3848         /*
3849          * It's easy to be greedy for MSI-X vectors, but it really
3850          * doesn't do us much good if we have a lot more vectors
3851          * than CPU's.  So let's be conservative and only ask for
3852          * (roughly) the same number of vectors as there are CPU's.
3853          */
3854         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
3855                        (int)num_online_cpus()) + NON_Q_VECTORS;
3856
3857         /*
3858          * At the same time, hardware can only support a maximum of
3859          * hw.mac->max_msix_vectors vectors.  With features
3860          * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
3861          * descriptor queues supported by our device.  Thus, we cap it off in
3862          * those rare cases where the cpu count also exceeds our vector limit.
3863          */
3864         v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
3865
3866         /* A failure in MSI-X entry allocation isn't fatal, but it does
3867          * mean we disable MSI-X capabilities of the adapter. */
3868         adapter->msix_entries = kcalloc(v_budget,
3869                                         sizeof(struct msix_entry), GFP_KERNEL);
3870         if (adapter->msix_entries) {
3871                 for (vector = 0; vector < v_budget; vector++)
3872                         adapter->msix_entries[vector].entry = vector;
3873
3874                 ixgbe_acquire_msix_vectors(adapter, v_budget);
3875
3876                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3877                         goto out;
3878         }
3879
3880         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
3881         adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
3882         adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3883         adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3884         adapter->atr_sample_rate = 0;
3885         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3886                 ixgbe_disable_sriov(adapter);
3887
3888         ixgbe_set_num_queues(adapter);
3889
3890         err = pci_enable_msi(adapter->pdev);
3891         if (!err) {
3892                 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
3893         } else {
3894                 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
3895                         "falling back to legacy.  Error: %d\n", err);
3896                 /* reset err */
3897                 err = 0;
3898         }
3899
3900 out:
3901         return err;
3902 }
3903
3904 /**
3905  * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
3906  * @adapter: board private structure to initialize
3907  *
3908  * We allocate one q_vector per queue interrupt.  If allocation fails we
3909  * return -ENOMEM.
3910  **/
3911 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
3912 {
3913         int q_idx, num_q_vectors;
3914         struct ixgbe_q_vector *q_vector;
3915         int napi_vectors;
3916         int (*poll)(struct napi_struct *, int);
3917
3918         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3919                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3920                 napi_vectors = adapter->num_rx_queues;
3921                 poll = &ixgbe_clean_rxtx_many;
3922         } else {
3923                 num_q_vectors = 1;
3924                 napi_vectors = 1;
3925                 poll = &ixgbe_poll;
3926         }
3927
3928         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3929                 q_vector = kzalloc(sizeof(struct ixgbe_q_vector), GFP_KERNEL);
3930                 if (!q_vector)
3931                         goto err_out;
3932                 q_vector->adapter = adapter;
3933                 if (q_vector->txr_count && !q_vector->rxr_count)
3934                         q_vector->eitr = adapter->tx_eitr_param;
3935                 else
3936                         q_vector->eitr = adapter->rx_eitr_param;
3937                 q_vector->v_idx = q_idx;
3938                 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
3939                 adapter->q_vector[q_idx] = q_vector;
3940         }
3941
3942         return 0;
3943
3944 err_out:
3945         while (q_idx) {
3946                 q_idx--;
3947                 q_vector = adapter->q_vector[q_idx];
3948                 netif_napi_del(&q_vector->napi);
3949                 kfree(q_vector);
3950                 adapter->q_vector[q_idx] = NULL;
3951         }
3952         return -ENOMEM;
3953 }
3954
3955 /**
3956  * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
3957  * @adapter: board private structure to initialize
3958  *
3959  * This function frees the memory allocated to the q_vectors.  In addition if
3960  * NAPI is enabled it will delete any references to the NAPI struct prior
3961  * to freeing the q_vector.
3962  **/
3963 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
3964 {
3965         int q_idx, num_q_vectors;
3966
3967         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3968                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3969         else
3970                 num_q_vectors = 1;
3971
3972         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3973                 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
3974                 adapter->q_vector[q_idx] = NULL;
3975                 netif_napi_del(&q_vector->napi);
3976                 kfree(q_vector);
3977         }
3978 }
3979
3980 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
3981 {
3982         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3983                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3984                 pci_disable_msix(adapter->pdev);
3985                 kfree(adapter->msix_entries);
3986                 adapter->msix_entries = NULL;
3987         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
3988                 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
3989                 pci_disable_msi(adapter->pdev);
3990         }
3991         return;
3992 }
3993
3994 /**
3995  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
3996  * @adapter: board private structure to initialize
3997  *
3998  * We determine which interrupt scheme to use based on...
3999  * - Kernel support (MSI, MSI-X)
4000  *   - which can be user-defined (via MODULE_PARAM)
4001  * - Hardware queue count (num_*_queues)
4002  *   - defined by miscellaneous hardware support/features (RSS, etc.)
4003  **/
4004 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4005 {
4006         int err;
4007
4008         /* Number of supported queues */
4009         ixgbe_set_num_queues(adapter);
4010
4011         err = ixgbe_set_interrupt_capability(adapter);
4012         if (err) {
4013                 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
4014                 goto err_set_interrupt;
4015         }
4016
4017         err = ixgbe_alloc_q_vectors(adapter);
4018         if (err) {
4019                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queue "
4020                         "vectors\n");
4021                 goto err_alloc_q_vectors;
4022         }
4023
4024         err = ixgbe_alloc_queues(adapter);
4025         if (err) {
4026                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
4027                 goto err_alloc_queues;
4028         }
4029
4030         DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
4031                 "Tx Queue count = %u\n",
4032                 (adapter->num_rx_queues > 1) ? "Enabled" :
4033                 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
4034
4035         set_bit(__IXGBE_DOWN, &adapter->state);
4036
4037         return 0;
4038
4039 err_alloc_queues:
4040         ixgbe_free_q_vectors(adapter);
4041 err_alloc_q_vectors:
4042         ixgbe_reset_interrupt_capability(adapter);
4043 err_set_interrupt:
4044         return err;
4045 }
4046
4047 /**
4048  * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4049  * @adapter: board private structure to clear interrupt scheme on
4050  *
4051  * We go through and clear interrupt specific resources and reset the structure
4052  * to pre-load conditions
4053  **/
4054 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4055 {
4056         kfree(adapter->tx_ring);
4057         kfree(adapter->rx_ring);
4058         adapter->tx_ring = NULL;
4059         adapter->rx_ring = NULL;
4060
4061         ixgbe_free_q_vectors(adapter);
4062         ixgbe_reset_interrupt_capability(adapter);
4063 }
4064
4065 /**
4066  * ixgbe_sfp_timer - worker thread to find a missing module
4067  * @data: pointer to our adapter struct
4068  **/
4069 static void ixgbe_sfp_timer(unsigned long data)
4070 {
4071         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4072
4073         /*
4074          * Do the sfp_timer outside of interrupt context due to the
4075          * delays that sfp+ detection requires
4076          */
4077         schedule_work(&adapter->sfp_task);
4078 }
4079
4080 /**
4081  * ixgbe_sfp_task - worker thread to find a missing module
4082  * @work: pointer to work_struct containing our data
4083  **/
4084 static void ixgbe_sfp_task(struct work_struct *work)
4085 {
4086         struct ixgbe_adapter *adapter = container_of(work,
4087                                                      struct ixgbe_adapter,
4088                                                      sfp_task);
4089         struct ixgbe_hw *hw = &adapter->hw;
4090
4091         if ((hw->phy.type == ixgbe_phy_nl) &&
4092             (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4093                 s32 ret = hw->phy.ops.identify_sfp(hw);
4094                 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4095                         goto reschedule;
4096                 ret = hw->phy.ops.reset(hw);
4097                 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4098                         dev_err(&adapter->pdev->dev, "failed to initialize "
4099                                 "because an unsupported SFP+ module type "
4100                                 "was detected.\n"
4101                                 "Reload the driver after installing a "
4102                                 "supported module.\n");
4103                         unregister_netdev(adapter->netdev);
4104                 } else {
4105                         DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
4106                                 hw->phy.sfp_type);
4107                 }
4108                 /* don't need this routine any more */
4109                 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4110         }
4111         return;
4112 reschedule:
4113         if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4114                 mod_timer(&adapter->sfp_timer,
4115                           round_jiffies(jiffies + (2 * HZ)));
4116 }
4117
4118 /**
4119  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4120  * @adapter: board private structure to initialize
4121  *
4122  * ixgbe_sw_init initializes the Adapter private data structure.
4123  * Fields are initialized based on PCI device information and
4124  * OS network device settings (MTU size).
4125  **/
4126 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4127 {
4128         struct ixgbe_hw *hw = &adapter->hw;
4129         struct pci_dev *pdev = adapter->pdev;
4130         unsigned int rss;
4131 #ifdef CONFIG_IXGBE_DCB
4132         int j;
4133         struct tc_configuration *tc;
4134 #endif
4135
4136         /* PCI config space info */
4137
4138         hw->vendor_id = pdev->vendor;
4139         hw->device_id = pdev->device;
4140         hw->revision_id = pdev->revision;
4141         hw->subsystem_vendor_id = pdev->subsystem_vendor;
4142         hw->subsystem_device_id = pdev->subsystem_device;
4143
4144         /* Set capability flags */
4145         rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4146         adapter->ring_feature[RING_F_RSS].indices = rss;
4147         adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4148         adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4149         if (hw->mac.type == ixgbe_mac_82598EB) {
4150                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4151                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4152                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4153         } else if (hw->mac.type == ixgbe_mac_82599EB) {
4154                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4155                 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4156                 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4157                 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4158                 adapter->ring_feature[RING_F_FDIR].indices =
4159                                                          IXGBE_MAX_FDIR_INDICES;
4160                 adapter->atr_sample_rate = 20;
4161                 adapter->fdir_pballoc = 0;
4162 #ifdef IXGBE_FCOE
4163                 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4164                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4165                 adapter->ring_feature[RING_F_FCOE].indices = 0;
4166 #ifdef CONFIG_IXGBE_DCB
4167                 /* Default traffic class to use for FCoE */
4168                 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4169 #endif
4170 #endif /* IXGBE_FCOE */
4171         }
4172
4173 #ifdef CONFIG_IXGBE_DCB
4174         /* Configure DCB traffic classes */
4175         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4176                 tc = &adapter->dcb_cfg.tc_config[j];
4177                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4178                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4179                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4180                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4181                 tc->dcb_pfc = pfc_disabled;
4182         }
4183         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4184         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4185         adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4186         adapter->dcb_cfg.pfc_mode_enable = false;
4187         adapter->dcb_cfg.round_robin_enable = false;
4188         adapter->dcb_set_bitmap = 0x00;
4189         ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4190                            adapter->ring_feature[RING_F_DCB].indices);
4191
4192 #endif
4193
4194         /* default flow control settings */
4195         hw->fc.requested_mode = ixgbe_fc_full;
4196         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
4197 #ifdef CONFIG_DCB
4198         adapter->last_lfc_mode = hw->fc.current_mode;
4199 #endif
4200         hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
4201         hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
4202         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4203         hw->fc.send_xon = true;
4204         hw->fc.disable_fc_autoneg = false;
4205
4206         /* enable itr by default in dynamic mode */
4207         adapter->rx_itr_setting = 1;
4208         adapter->rx_eitr_param = 20000;
4209         adapter->tx_itr_setting = 1;
4210         adapter->tx_eitr_param = 10000;
4211
4212         /* set defaults for eitr in MegaBytes */
4213         adapter->eitr_low = 10;
4214         adapter->eitr_high = 20;
4215
4216         /* set default ring sizes */
4217         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4218         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4219
4220         /* initialize eeprom parameters */
4221         if (ixgbe_init_eeprom_params_generic(hw)) {
4222                 dev_err(&pdev->dev, "EEPROM initialization failed\n");
4223                 return -EIO;
4224         }
4225
4226         /* enable rx csum by default */
4227         adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4228
4229         set_bit(__IXGBE_DOWN, &adapter->state);
4230
4231         return 0;
4232 }
4233
4234 /**
4235  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4236  * @adapter: board private structure
4237  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
4238  *
4239  * Return 0 on success, negative on failure
4240  **/
4241 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4242                              struct ixgbe_ring *tx_ring)
4243 {
4244         struct pci_dev *pdev = adapter->pdev;
4245         int size;
4246
4247         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4248         tx_ring->tx_buffer_info = vmalloc(size);
4249         if (!tx_ring->tx_buffer_info)
4250                 goto err;
4251         memset(tx_ring->tx_buffer_info, 0, size);
4252
4253         /* round up to nearest 4K */
4254         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4255         tx_ring->size = ALIGN(tx_ring->size, 4096);
4256
4257         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
4258                                              &tx_ring->dma);
4259         if (!tx_ring->desc)
4260                 goto err;
4261
4262         tx_ring->next_to_use = 0;
4263         tx_ring->next_to_clean = 0;
4264         tx_ring->work_limit = tx_ring->count;
4265         return 0;
4266
4267 err:
4268         vfree(tx_ring->tx_buffer_info);
4269         tx_ring->tx_buffer_info = NULL;
4270         DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
4271                             "descriptor ring\n");
4272         return -ENOMEM;
4273 }
4274
4275 /**
4276  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4277  * @adapter: board private structure
4278  *
4279  * If this function returns with an error, then it's possible one or
4280  * more of the rings is populated (while the rest are not).  It is the
4281  * callers duty to clean those orphaned rings.
4282  *
4283  * Return 0 on success, negative on failure
4284  **/
4285 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4286 {
4287         int i, err = 0;
4288
4289         for (i = 0; i < adapter->num_tx_queues; i++) {
4290                 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
4291                 if (!err)
4292                         continue;
4293                 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
4294                 break;
4295         }
4296
4297         return err;
4298 }
4299
4300 /**
4301  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4302  * @adapter: board private structure
4303  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
4304  *
4305  * Returns 0 on success, negative on failure
4306  **/
4307 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4308                              struct ixgbe_ring *rx_ring)
4309 {
4310         struct pci_dev *pdev = adapter->pdev;
4311         int size;
4312
4313         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4314         rx_ring->rx_buffer_info = vmalloc(size);
4315         if (!rx_ring->rx_buffer_info) {
4316                 DPRINTK(PROBE, ERR,
4317                         "vmalloc allocation failed for the rx desc ring\n");
4318                 goto alloc_failed;
4319         }
4320         memset(rx_ring->rx_buffer_info, 0, size);
4321
4322         /* Round up to nearest 4K */
4323         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4324         rx_ring->size = ALIGN(rx_ring->size, 4096);
4325
4326         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
4327
4328         if (!rx_ring->desc) {
4329                 DPRINTK(PROBE, ERR,
4330                         "Memory allocation failed for the rx desc ring\n");
4331                 vfree(rx_ring->rx_buffer_info);
4332                 goto alloc_failed;
4333         }
4334
4335         rx_ring->next_to_clean = 0;
4336         rx_ring->next_to_use = 0;
4337
4338         return 0;
4339
4340 alloc_failed:
4341         return -ENOMEM;
4342 }
4343
4344 /**
4345  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4346  * @adapter: board private structure
4347  *
4348  * If this function returns with an error, then it's possible one or
4349  * more of the rings is populated (while the rest are not).  It is the
4350  * callers duty to clean those orphaned rings.
4351  *
4352  * Return 0 on success, negative on failure
4353  **/
4354
4355 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4356 {
4357         int i, err = 0;
4358
4359         for (i = 0; i < adapter->num_rx_queues; i++) {
4360                 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
4361                 if (!err)
4362                         continue;
4363                 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
4364                 break;
4365         }
4366
4367         return err;
4368 }
4369
4370 /**
4371  * ixgbe_free_tx_resources - Free Tx Resources per Queue
4372  * @adapter: board private structure
4373  * @tx_ring: Tx descriptor ring for a specific queue
4374  *
4375  * Free all transmit software resources
4376  **/
4377 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4378                              struct ixgbe_ring *tx_ring)
4379 {
4380         struct pci_dev *pdev = adapter->pdev;
4381
4382         ixgbe_clean_tx_ring(adapter, tx_ring);
4383
4384         vfree(tx_ring->tx_buffer_info);
4385         tx_ring->tx_buffer_info = NULL;
4386
4387         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
4388
4389         tx_ring->desc = NULL;
4390 }
4391
4392 /**
4393  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4394  * @adapter: board private structure
4395  *
4396  * Free all transmit software resources
4397  **/
4398 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4399 {
4400         int i;
4401
4402         for (i = 0; i < adapter->num_tx_queues; i++)
4403                 if (adapter->tx_ring[i].desc)
4404                         ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
4405 }
4406
4407 /**
4408  * ixgbe_free_rx_resources - Free Rx Resources
4409  * @adapter: board private structure
4410  * @rx_ring: ring to clean the resources from
4411  *
4412  * Free all receive software resources
4413  **/
4414 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
4415                              struct ixgbe_ring *rx_ring)
4416 {
4417         struct pci_dev *pdev = adapter->pdev;
4418
4419         ixgbe_clean_rx_ring(adapter, rx_ring);
4420
4421         vfree(rx_ring->rx_buffer_info);
4422         rx_ring->rx_buffer_info = NULL;
4423
4424         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
4425
4426         rx_ring->desc = NULL;
4427 }
4428
4429 /**
4430  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4431  * @adapter: board private structure
4432  *
4433  * Free all receive software resources
4434  **/
4435 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4436 {
4437         int i;
4438
4439         for (i = 0; i < adapter->num_rx_queues; i++)
4440                 if (adapter->rx_ring[i].desc)
4441                         ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
4442 }
4443
4444 /**
4445  * ixgbe_change_mtu - Change the Maximum Transfer Unit
4446  * @netdev: network interface device structure
4447  * @new_mtu: new value for maximum frame size
4448  *
4449  * Returns 0 on success, negative on failure
4450  **/
4451 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4452 {
4453         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4454         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4455
4456         /* MTU < 68 is an error and causes problems on some kernels */
4457         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4458                 return -EINVAL;
4459
4460         DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
4461                 netdev->mtu, new_mtu);
4462         /* must set new MTU before calling down or up */
4463         netdev->mtu = new_mtu;
4464
4465         if (netif_running(netdev))
4466                 ixgbe_reinit_locked(adapter);
4467
4468         return 0;
4469 }
4470
4471 /**
4472  * ixgbe_open - Called when a network interface is made active
4473  * @netdev: network interface device structure
4474  *
4475  * Returns 0 on success, negative value on failure
4476  *
4477  * The open entry point is called when a network interface is made
4478  * active by the system (IFF_UP).  At this point all resources needed
4479  * for transmit and receive operations are allocated, the interrupt
4480  * handler is registered with the OS, the watchdog timer is started,
4481  * and the stack is notified that the interface is ready.
4482  **/
4483 static int ixgbe_open(struct net_device *netdev)
4484 {
4485         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4486         int err;
4487
4488         /* disallow open during test */
4489         if (test_bit(__IXGBE_TESTING, &adapter->state))
4490                 return -EBUSY;
4491
4492         netif_carrier_off(netdev);
4493
4494         /* allocate transmit descriptors */
4495         err = ixgbe_setup_all_tx_resources(adapter);
4496         if (err)
4497                 goto err_setup_tx;
4498
4499         /* allocate receive descriptors */
4500         err = ixgbe_setup_all_rx_resources(adapter);
4501         if (err)
4502                 goto err_setup_rx;
4503
4504         ixgbe_configure(adapter);
4505
4506         err = ixgbe_request_irq(adapter);
4507         if (err)
4508                 goto err_req_irq;
4509
4510         err = ixgbe_up_complete(adapter);
4511         if (err)
4512                 goto err_up;
4513
4514         netif_tx_start_all_queues(netdev);
4515
4516         return 0;
4517
4518 err_up:
4519         ixgbe_release_hw_control(adapter);
4520         ixgbe_free_irq(adapter);
4521 err_req_irq:
4522 err_setup_rx:
4523         ixgbe_free_all_rx_resources(adapter);
4524 err_setup_tx:
4525         ixgbe_free_all_tx_resources(adapter);
4526         ixgbe_reset(adapter);
4527
4528         return err;
4529 }
4530
4531 /**
4532  * ixgbe_close - Disables a network interface
4533  * @netdev: network interface device structure
4534  *
4535  * Returns 0, this is not allowed to fail
4536  *
4537  * The close entry point is called when an interface is de-activated
4538  * by the OS.  The hardware is still under the drivers control, but
4539  * needs to be disabled.  A global MAC reset is issued to stop the
4540  * hardware, and all transmit and receive resources are freed.
4541  **/
4542 static int ixgbe_close(struct net_device *netdev)
4543 {
4544         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4545
4546         ixgbe_down(adapter);
4547         ixgbe_free_irq(adapter);
4548
4549         ixgbe_free_all_tx_resources(adapter);
4550         ixgbe_free_all_rx_resources(adapter);
4551
4552         ixgbe_release_hw_control(adapter);
4553
4554         return 0;
4555 }
4556
4557 #ifdef CONFIG_PM
4558 static int ixgbe_resume(struct pci_dev *pdev)
4559 {
4560         struct net_device *netdev = pci_get_drvdata(pdev);
4561         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4562         u32 err;
4563
4564         pci_set_power_state(pdev, PCI_D0);
4565         pci_restore_state(pdev);
4566         /*
4567          * pci_restore_state clears dev->state_saved so call
4568          * pci_save_state to restore it.
4569          */
4570         pci_save_state(pdev);
4571
4572         err = pci_enable_device_mem(pdev);
4573         if (err) {
4574                 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
4575                                 "suspend\n");
4576                 return err;
4577         }
4578         pci_set_master(pdev);
4579
4580         pci_wake_from_d3(pdev, false);
4581
4582         err = ixgbe_init_interrupt_scheme(adapter);
4583         if (err) {
4584                 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
4585                                 "device\n");
4586                 return err;
4587         }
4588
4589         ixgbe_reset(adapter);
4590
4591         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4592
4593         if (netif_running(netdev)) {
4594                 err = ixgbe_open(adapter->netdev);
4595                 if (err)
4596                         return err;
4597         }
4598
4599         netif_device_attach(netdev);
4600
4601         return 0;
4602 }
4603 #endif /* CONFIG_PM */
4604
4605 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4606 {
4607         struct net_device *netdev = pci_get_drvdata(pdev);
4608         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4609         struct ixgbe_hw *hw = &adapter->hw;
4610         u32 ctrl, fctrl;
4611         u32 wufc = adapter->wol;
4612 #ifdef CONFIG_PM
4613         int retval = 0;
4614 #endif
4615
4616         netif_device_detach(netdev);
4617
4618         if (netif_running(netdev)) {
4619                 ixgbe_down(adapter);
4620                 ixgbe_free_irq(adapter);
4621                 ixgbe_free_all_tx_resources(adapter);
4622                 ixgbe_free_all_rx_resources(adapter);
4623         }
4624         ixgbe_clear_interrupt_scheme(adapter);
4625
4626 #ifdef CONFIG_PM
4627         retval = pci_save_state(pdev);
4628         if (retval)
4629                 return retval;
4630
4631 #endif
4632         if (wufc) {
4633                 ixgbe_set_rx_mode(netdev);
4634
4635                 /* turn on all-multi mode if wake on multicast is enabled */
4636                 if (wufc & IXGBE_WUFC_MC) {
4637                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4638                         fctrl |= IXGBE_FCTRL_MPE;
4639                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4640                 }
4641
4642                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4643                 ctrl |= IXGBE_CTRL_GIO_DIS;
4644                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4645
4646                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4647         } else {
4648                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4649                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4650         }
4651
4652         if (wufc && hw->mac.type == ixgbe_mac_82599EB)
4653                 pci_wake_from_d3(pdev, true);
4654         else
4655                 pci_wake_from_d3(pdev, false);
4656
4657         *enable_wake = !!wufc;
4658
4659         ixgbe_release_hw_control(adapter);
4660
4661         pci_disable_device(pdev);
4662
4663         return 0;
4664 }
4665
4666 #ifdef CONFIG_PM
4667 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4668 {
4669         int retval;
4670         bool wake;
4671
4672         retval = __ixgbe_shutdown(pdev, &wake);
4673         if (retval)
4674                 return retval;
4675
4676         if (wake) {
4677                 pci_prepare_to_sleep(pdev);
4678         } else {
4679                 pci_wake_from_d3(pdev, false);
4680                 pci_set_power_state(pdev, PCI_D3hot);
4681         }
4682
4683         return 0;
4684 }
4685 #endif /* CONFIG_PM */
4686
4687 static void ixgbe_shutdown(struct pci_dev *pdev)
4688 {
4689         bool wake;
4690
4691         __ixgbe_shutdown(pdev, &wake);
4692
4693         if (system_state == SYSTEM_POWER_OFF) {
4694                 pci_wake_from_d3(pdev, wake);
4695                 pci_set_power_state(pdev, PCI_D3hot);
4696         }
4697 }
4698
4699 /**
4700  * ixgbe_update_stats - Update the board statistics counters.
4701  * @adapter: board private structure
4702  **/
4703 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4704 {
4705         struct net_device *netdev = adapter->netdev;
4706         struct ixgbe_hw *hw = &adapter->hw;
4707         u64 total_mpc = 0;
4708         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4709         u64 non_eop_descs = 0, restart_queue = 0;
4710
4711         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
4712                 u64 rsc_count = 0;
4713                 u64 rsc_flush = 0;
4714                 for (i = 0; i < 16; i++)
4715                         adapter->hw_rx_no_dma_resources +=
4716                                              IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4717                 for (i = 0; i < adapter->num_rx_queues; i++) {
4718                         rsc_count += adapter->rx_ring[i].rsc_count;
4719                         rsc_flush += adapter->rx_ring[i].rsc_flush;
4720                 }
4721                 adapter->rsc_total_count = rsc_count;
4722                 adapter->rsc_total_flush = rsc_flush;
4723         }
4724
4725         /* gather some stats to the adapter struct that are per queue */
4726         for (i = 0; i < adapter->num_tx_queues; i++)
4727                 restart_queue += adapter->tx_ring[i].restart_queue;
4728         adapter->restart_queue = restart_queue;
4729
4730         for (i = 0; i < adapter->num_rx_queues; i++)
4731                 non_eop_descs += adapter->rx_ring[i].non_eop_descs;
4732         adapter->non_eop_descs = non_eop_descs;
4733
4734         adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
4735         for (i = 0; i < 8; i++) {
4736                 /* for packet buffers not used, the register should read 0 */
4737                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
4738                 missed_rx += mpc;
4739                 adapter->stats.mpc[i] += mpc;
4740                 total_mpc += adapter->stats.mpc[i];
4741                 if (hw->mac.type == ixgbe_mac_82598EB)
4742                         adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
4743                 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
4744                 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
4745                 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
4746                 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
4747                 if (hw->mac.type == ixgbe_mac_82599EB) {
4748                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4749                                                             IXGBE_PXONRXCNT(i));
4750                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4751                                                            IXGBE_PXOFFRXCNT(i));
4752                         adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4753                 } else {
4754                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4755                                                               IXGBE_PXONRXC(i));
4756                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4757                                                              IXGBE_PXOFFRXC(i));
4758                 }
4759                 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
4760                                                             IXGBE_PXONTXC(i));
4761                 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
4762                                                              IXGBE_PXOFFTXC(i));
4763         }
4764         adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
4765         /* work around hardware counting issue */
4766         adapter->stats.gprc -= missed_rx;
4767
4768         /* 82598 hardware only has a 32 bit counter in the high register */
4769         if (hw->mac.type == ixgbe_mac_82599EB) {
4770                 u64 tmp;
4771                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
4772                 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
4773                 adapter->stats.gorc += (tmp << 32);
4774                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
4775                 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
4776                 adapter->stats.gotc += (tmp << 32);
4777                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
4778                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
4779                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
4780                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
4781                 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
4782                 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
4783 #ifdef IXGBE_FCOE
4784                 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
4785                 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
4786                 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
4787                 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
4788                 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
4789                 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
4790 #endif /* IXGBE_FCOE */
4791         } else {
4792                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
4793                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
4794                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
4795                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
4796                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
4797         }
4798         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
4799         adapter->stats.bprc += bprc;
4800         adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
4801         if (hw->mac.type == ixgbe_mac_82598EB)
4802                 adapter->stats.mprc -= bprc;
4803         adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
4804         adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
4805         adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
4806         adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
4807         adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
4808         adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
4809         adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
4810         adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
4811         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
4812         adapter->stats.lxontxc += lxon;
4813         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
4814         adapter->stats.lxofftxc += lxoff;
4815         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4816         adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
4817         adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
4818         /*
4819          * 82598 errata - tx of flow control packets is included in tx counters
4820          */
4821         xon_off_tot = lxon + lxoff;
4822         adapter->stats.gptc -= xon_off_tot;
4823         adapter->stats.mptc -= xon_off_tot;
4824         adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
4825         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4826         adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
4827         adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
4828         adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
4829         adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
4830         adapter->stats.ptc64 -= xon_off_tot;
4831         adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
4832         adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
4833         adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
4834         adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
4835         adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
4836         adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
4837
4838         /* Fill out the OS statistics structure */
4839         netdev->stats.multicast = adapter->stats.mprc;
4840
4841         /* Rx Errors */
4842         netdev->stats.rx_errors = adapter->stats.crcerrs +
4843                                        adapter->stats.rlec;
4844         netdev->stats.rx_dropped = 0;
4845         netdev->stats.rx_length_errors = adapter->stats.rlec;
4846         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4847         netdev->stats.rx_missed_errors = total_mpc;
4848 }
4849
4850 /**
4851  * ixgbe_watchdog - Timer Call-back
4852  * @data: pointer to adapter cast into an unsigned long
4853  **/
4854 static void ixgbe_watchdog(unsigned long data)
4855 {
4856         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4857         struct ixgbe_hw *hw = &adapter->hw;
4858         u64 eics = 0;
4859         int i;
4860
4861         /*
4862          *  Do the watchdog outside of interrupt context due to the lovely
4863          * delays that some of the newer hardware requires
4864          */
4865
4866         if (test_bit(__IXGBE_DOWN, &adapter->state))
4867                 goto watchdog_short_circuit;
4868
4869         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
4870                 /*
4871                  * for legacy and MSI interrupts don't set any bits
4872                  * that are enabled for EIAM, because this operation
4873                  * would set *both* EIMS and EICS for any bit in EIAM
4874                  */
4875                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
4876                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
4877                 goto watchdog_reschedule;
4878         }
4879
4880         /* get one bit for every active tx/rx interrupt vector */
4881         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
4882                 struct ixgbe_q_vector *qv = adapter->q_vector[i];
4883                 if (qv->rxr_count || qv->txr_count)
4884                         eics |= ((u64)1 << i);
4885         }
4886
4887         /* Cause software interrupt to ensure rx rings are cleaned */
4888         ixgbe_irq_rearm_queues(adapter, eics);
4889
4890 watchdog_reschedule:
4891         /* Reset the timer */
4892         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
4893
4894 watchdog_short_circuit:
4895         schedule_work(&adapter->watchdog_task);
4896 }
4897
4898 /**
4899  * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
4900  * @work: pointer to work_struct containing our data
4901  **/
4902 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
4903 {
4904         struct ixgbe_adapter *adapter = container_of(work,
4905                                                      struct ixgbe_adapter,
4906                                                      multispeed_fiber_task);
4907         struct ixgbe_hw *hw = &adapter->hw;
4908         u32 autoneg;
4909         bool negotiation;
4910
4911         adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
4912         autoneg = hw->phy.autoneg_advertised;
4913         if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
4914                 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
4915         if (hw->mac.ops.setup_link)
4916                 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
4917         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4918         adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
4919 }
4920
4921 /**
4922  * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
4923  * @work: pointer to work_struct containing our data
4924  **/
4925 static void ixgbe_sfp_config_module_task(struct work_struct *work)
4926 {
4927         struct ixgbe_adapter *adapter = container_of(work,
4928                                                      struct ixgbe_adapter,
4929                                                      sfp_config_module_task);
4930         struct ixgbe_hw *hw = &adapter->hw;
4931         u32 err;
4932
4933         adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
4934
4935         /* Time for electrical oscillations to settle down */
4936         msleep(100);
4937         err = hw->phy.ops.identify_sfp(hw);
4938
4939         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4940                 dev_err(&adapter->pdev->dev, "failed to initialize because "
4941                         "an unsupported SFP+ module type was detected.\n"
4942                         "Reload the driver after installing a supported "
4943                         "module.\n");
4944                 unregister_netdev(adapter->netdev);
4945                 return;
4946         }
4947         hw->mac.ops.setup_sfp(hw);
4948
4949         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
4950                 /* This will also work for DA Twinax connections */
4951                 schedule_work(&adapter->multispeed_fiber_task);
4952         adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
4953 }
4954
4955 /**
4956  * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
4957  * @work: pointer to work_struct containing our data
4958  **/
4959 static void ixgbe_fdir_reinit_task(struct work_struct *work)
4960 {
4961         struct ixgbe_adapter *adapter = container_of(work,
4962                                                      struct ixgbe_adapter,
4963                                                      fdir_reinit_task);
4964         struct ixgbe_hw *hw = &adapter->hw;
4965         int i;
4966
4967         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
4968                 for (i = 0; i < adapter->num_tx_queues; i++)
4969                         set_bit(__IXGBE_FDIR_INIT_DONE,
4970                                 &(adapter->tx_ring[i].reinit_state));
4971         } else {
4972                 DPRINTK(PROBE, ERR, "failed to finish FDIR re-initialization, "
4973                         "ignored adding FDIR ATR filters \n");
4974         }
4975         /* Done FDIR Re-initialization, enable transmits */
4976         netif_tx_start_all_queues(adapter->netdev);
4977 }
4978
4979 /**
4980  * ixgbe_watchdog_task - worker thread to bring link up
4981  * @work: pointer to work_struct containing our data
4982  **/
4983 static void ixgbe_watchdog_task(struct work_struct *work)
4984 {
4985         struct ixgbe_adapter *adapter = container_of(work,
4986                                                      struct ixgbe_adapter,
4987                                                      watchdog_task);
4988         struct net_device *netdev = adapter->netdev;
4989         struct ixgbe_hw *hw = &adapter->hw;
4990         u32 link_speed = adapter->link_speed;
4991         bool link_up = adapter->link_up;
4992         int i;
4993         struct ixgbe_ring *tx_ring;
4994         int some_tx_pending = 0;
4995
4996         adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
4997
4998         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
4999                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5000                 if (link_up) {
5001 #ifdef CONFIG_DCB
5002                         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5003                                 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5004                                         hw->mac.ops.fc_enable(hw, i);
5005                         } else {
5006                                 hw->mac.ops.fc_enable(hw, 0);
5007                         }
5008 #else
5009                         hw->mac.ops.fc_enable(hw, 0);
5010 #endif
5011                 }
5012
5013                 if (link_up ||
5014                     time_after(jiffies, (adapter->link_check_timeout +
5015                                          IXGBE_TRY_LINK_TIMEOUT))) {
5016                         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5017                         IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5018                 }
5019                 adapter->link_up = link_up;
5020                 adapter->link_speed = link_speed;
5021         }
5022
5023         if (link_up) {
5024                 if (!netif_carrier_ok(netdev)) {
5025                         bool flow_rx, flow_tx;
5026
5027                         if (hw->mac.type == ixgbe_mac_82599EB) {
5028                                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5029                                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5030                                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5031                                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5032                         } else {
5033                                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5034                                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5035                                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5036                                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5037                         }
5038
5039                         printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
5040                                "Flow Control: %s\n",
5041                                netdev->name,
5042                                (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5043                                 "10 Gbps" :
5044                                 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5045                                  "1 Gbps" : "unknown speed")),
5046                                ((flow_rx && flow_tx) ? "RX/TX" :
5047                                 (flow_rx ? "RX" :
5048                                 (flow_tx ? "TX" : "None"))));
5049
5050                         netif_carrier_on(netdev);
5051                 } else {
5052                         /* Force detection of hung controller */
5053                         adapter->detect_tx_hung = true;
5054                 }
5055         } else {
5056                 adapter->link_up = false;
5057                 adapter->link_speed = 0;
5058                 if (netif_carrier_ok(netdev)) {
5059                         printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
5060                                netdev->name);
5061                         netif_carrier_off(netdev);
5062                 }
5063         }
5064
5065         if (!netif_carrier_ok(netdev)) {
5066                 for (i = 0; i < adapter->num_tx_queues; i++) {
5067                         tx_ring = &adapter->tx_ring[i];
5068                         if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5069                                 some_tx_pending = 1;
5070                                 break;
5071                         }
5072                 }
5073
5074                 if (some_tx_pending) {
5075                         /* We've lost link, so the controller stops DMA,
5076                          * but we've got queued Tx work that's never going
5077                          * to get done, so reset controller to flush Tx.
5078                          * (Do the reset outside of interrupt context).
5079                          */
5080                          schedule_work(&adapter->reset_task);
5081                 }
5082         }
5083
5084         ixgbe_update_stats(adapter);
5085         adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
5086 }
5087
5088 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5089                      struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5090                      u32 tx_flags, u8 *hdr_len)
5091 {
5092         struct ixgbe_adv_tx_context_desc *context_desc;
5093         unsigned int i;
5094         int err;
5095         struct ixgbe_tx_buffer *tx_buffer_info;
5096         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5097         u32 mss_l4len_idx, l4len;
5098
5099         if (skb_is_gso(skb)) {
5100                 if (skb_header_cloned(skb)) {
5101                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5102                         if (err)
5103                                 return err;
5104                 }
5105                 l4len = tcp_hdrlen(skb);
5106                 *hdr_len += l4len;
5107
5108                 if (skb->protocol == htons(ETH_P_IP)) {
5109                         struct iphdr *iph = ip_hdr(skb);
5110                         iph->tot_len = 0;
5111                         iph->check = 0;
5112                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5113                                                                  iph->daddr, 0,
5114                                                                  IPPROTO_TCP,
5115                                                                  0);
5116                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
5117                         ipv6_hdr(skb)->payload_len = 0;
5118                         tcp_hdr(skb)->check =
5119                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5120                                              &ipv6_hdr(skb)->daddr,
5121                                              0, IPPROTO_TCP, 0);
5122                 }
5123
5124                 i = tx_ring->next_to_use;
5125
5126                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5127                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5128
5129                 /* VLAN MACLEN IPLEN */
5130                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5131                         vlan_macip_lens |=
5132                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5133                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5134                                     IXGBE_ADVTXD_MACLEN_SHIFT);
5135                 *hdr_len += skb_network_offset(skb);
5136                 vlan_macip_lens |=
5137                     (skb_transport_header(skb) - skb_network_header(skb));
5138                 *hdr_len +=
5139                     (skb_transport_header(skb) - skb_network_header(skb));
5140                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5141                 context_desc->seqnum_seed = 0;
5142
5143                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5144                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5145                                    IXGBE_ADVTXD_DTYP_CTXT);
5146
5147                 if (skb->protocol == htons(ETH_P_IP))
5148                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5149                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5150                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5151
5152                 /* MSS L4LEN IDX */
5153                 mss_l4len_idx =
5154                     (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5155                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5156                 /* use index 1 for TSO */
5157                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5158                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5159
5160                 tx_buffer_info->time_stamp = jiffies;
5161                 tx_buffer_info->next_to_watch = i;
5162
5163                 i++;
5164                 if (i == tx_ring->count)
5165                         i = 0;
5166                 tx_ring->next_to_use = i;
5167
5168                 return true;
5169         }
5170         return false;
5171 }
5172
5173 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5174                           struct ixgbe_ring *tx_ring,
5175                           struct sk_buff *skb, u32 tx_flags)
5176 {
5177         struct ixgbe_adv_tx_context_desc *context_desc;
5178         unsigned int i;
5179         struct ixgbe_tx_buffer *tx_buffer_info;
5180         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5181
5182         if (skb->ip_summed == CHECKSUM_PARTIAL ||
5183             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5184                 i = tx_ring->next_to_use;
5185                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5186                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5187
5188                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5189                         vlan_macip_lens |=
5190                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5191                 vlan_macip_lens |= (skb_network_offset(skb) <<
5192                                     IXGBE_ADVTXD_MACLEN_SHIFT);
5193                 if (skb->ip_summed == CHECKSUM_PARTIAL)
5194                         vlan_macip_lens |= (skb_transport_header(skb) -
5195                                             skb_network_header(skb));
5196
5197                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5198                 context_desc->seqnum_seed = 0;
5199
5200                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5201                                     IXGBE_ADVTXD_DTYP_CTXT);
5202
5203                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5204                         __be16 protocol;
5205
5206                         if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
5207                                 const struct vlan_ethhdr *vhdr =
5208                                         (const struct vlan_ethhdr *)skb->data;
5209
5210                                 protocol = vhdr->h_vlan_encapsulated_proto;
5211                         } else {
5212                                 protocol = skb->protocol;
5213                         }
5214
5215                         switch (protocol) {
5216                         case cpu_to_be16(ETH_P_IP):
5217                                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5218                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5219                                         type_tucmd_mlhl |=
5220                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5221                                 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
5222                                         type_tucmd_mlhl |=
5223                                                 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5224                                 break;
5225                         case cpu_to_be16(ETH_P_IPV6):
5226                                 /* XXX what about other V6 headers?? */
5227                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5228                                         type_tucmd_mlhl |=
5229                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5230                                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
5231                                         type_tucmd_mlhl |=
5232                                                 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5233                                 break;
5234                         default:
5235                                 if (unlikely(net_ratelimit())) {
5236                                         DPRINTK(PROBE, WARNING,
5237                                          "partial checksum but proto=%x!\n",
5238                                          skb->protocol);
5239                                 }
5240                                 break;
5241                         }
5242                 }
5243
5244                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5245                 /* use index zero for tx checksum offload */
5246                 context_desc->mss_l4len_idx = 0;
5247
5248                 tx_buffer_info->time_stamp = jiffies;
5249                 tx_buffer_info->next_to_watch = i;
5250
5251                 i++;
5252                 if (i == tx_ring->count)
5253                         i = 0;
5254                 tx_ring->next_to_use = i;
5255
5256                 return true;
5257         }
5258
5259         return false;
5260 }
5261
5262 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5263                         struct ixgbe_ring *tx_ring,
5264                         struct sk_buff *skb, u32 tx_flags,
5265                         unsigned int first)
5266 {
5267         struct pci_dev *pdev = adapter->pdev;
5268         struct ixgbe_tx_buffer *tx_buffer_info;
5269         unsigned int len;
5270         unsigned int total = skb->len;
5271         unsigned int offset = 0, size, count = 0, i;
5272         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5273         unsigned int f;
5274
5275         i = tx_ring->next_to_use;
5276
5277         if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5278                 /* excluding fcoe_crc_eof for FCoE */
5279                 total -= sizeof(struct fcoe_crc_eof);
5280
5281         len = min(skb_headlen(skb), total);
5282         while (len) {
5283                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5284                 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5285
5286                 tx_buffer_info->length = size;
5287                 tx_buffer_info->mapped_as_page = false;
5288                 tx_buffer_info->dma = pci_map_single(pdev,
5289                                                      skb->data + offset,
5290                                                      size, PCI_DMA_TODEVICE);
5291                 if (pci_dma_mapping_error(pdev, tx_buffer_info->dma))
5292                         goto dma_error;
5293                 tx_buffer_info->time_stamp = jiffies;
5294                 tx_buffer_info->next_to_watch = i;
5295
5296                 len -= size;
5297                 total -= size;
5298                 offset += size;
5299                 count++;
5300
5301                 if (len) {
5302                         i++;
5303                         if (i == tx_ring->count)
5304                                 i = 0;
5305                 }
5306         }
5307
5308         for (f = 0; f < nr_frags; f++) {
5309                 struct skb_frag_struct *frag;
5310
5311                 frag = &skb_shinfo(skb)->frags[f];
5312                 len = min((unsigned int)frag->size, total);
5313                 offset = frag->page_offset;
5314
5315                 while (len) {
5316                         i++;
5317                         if (i == tx_ring->count)
5318                                 i = 0;
5319
5320                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
5321                         size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5322
5323                         tx_buffer_info->length = size;
5324                         tx_buffer_info->dma = pci_map_page(adapter->pdev,
5325                                                            frag->page,
5326                                                            offset, size,
5327                                                            PCI_DMA_TODEVICE);
5328                         tx_buffer_info->mapped_as_page = true;
5329                         if (pci_dma_mapping_error(pdev, tx_buffer_info->dma))
5330                                 goto dma_error;
5331                         tx_buffer_info->time_stamp = jiffies;
5332                         tx_buffer_info->next_to_watch = i;
5333
5334                         len -= size;
5335                         total -= size;
5336                         offset += size;
5337                         count++;
5338                 }
5339                 if (total == 0)
5340                         break;
5341         }
5342
5343         tx_ring->tx_buffer_info[i].skb = skb;
5344         tx_ring->tx_buffer_info[first].next_to_watch = i;
5345
5346         return count;
5347
5348 dma_error:
5349         dev_err(&pdev->dev, "TX DMA map failed\n");
5350
5351         /* clear timestamp and dma mappings for failed tx_buffer_info map */
5352         tx_buffer_info->dma = 0;
5353         tx_buffer_info->time_stamp = 0;
5354         tx_buffer_info->next_to_watch = 0;
5355         if (count)
5356                 count--;
5357
5358         /* clear timestamp and dma mappings for remaining portion of packet */
5359         while (count--) {
5360                 if (i==0)
5361                         i += tx_ring->count;
5362                 i--;
5363                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5364                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
5365         }
5366
5367         return count;
5368 }
5369
5370 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
5371                            struct ixgbe_ring *tx_ring,
5372                            int tx_flags, int count, u32 paylen, u8 hdr_len)
5373 {
5374         union ixgbe_adv_tx_desc *tx_desc = NULL;
5375         struct ixgbe_tx_buffer *tx_buffer_info;
5376         u32 olinfo_status = 0, cmd_type_len = 0;
5377         unsigned int i;
5378         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
5379
5380         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
5381
5382         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
5383
5384         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5385                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
5386
5387         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
5388                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5389
5390                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5391                                  IXGBE_ADVTXD_POPTS_SHIFT;
5392
5393                 /* use index 1 context for tso */
5394                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5395                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5396                         olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
5397                                          IXGBE_ADVTXD_POPTS_SHIFT;
5398
5399         } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5400                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5401                                  IXGBE_ADVTXD_POPTS_SHIFT;
5402
5403         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5404                 olinfo_status |= IXGBE_ADVTXD_CC;
5405                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5406                 if (tx_flags & IXGBE_TX_FLAGS_FSO)
5407                         cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5408         }
5409
5410         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
5411
5412         i = tx_ring->next_to_use;
5413         while (count--) {
5414                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5415                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
5416                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
5417                 tx_desc->read.cmd_type_len =
5418                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
5419                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5420                 i++;
5421                 if (i == tx_ring->count)
5422                         i = 0;
5423         }
5424
5425         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
5426
5427         /*
5428          * Force memory writes to complete before letting h/w
5429          * know there are new descriptors to fetch.  (Only
5430          * applicable for weak-ordered memory model archs,
5431          * such as IA-64).
5432          */
5433         wmb();
5434
5435         tx_ring->next_to_use = i;
5436         writel(i, adapter->hw.hw_addr + tx_ring->tail);
5437 }
5438
5439 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5440                       int queue, u32 tx_flags)
5441 {
5442         /* Right now, we support IPv4 only */
5443         struct ixgbe_atr_input atr_input;
5444         struct tcphdr *th;
5445         struct iphdr *iph = ip_hdr(skb);
5446         struct ethhdr *eth = (struct ethhdr *)skb->data;
5447         u16 vlan_id, src_port, dst_port, flex_bytes;
5448         u32 src_ipv4_addr, dst_ipv4_addr;
5449         u8 l4type = 0;
5450
5451         /* check if we're UDP or TCP */
5452         if (iph->protocol == IPPROTO_TCP) {
5453                 th = tcp_hdr(skb);
5454                 src_port = th->source;
5455                 dst_port = th->dest;
5456                 l4type |= IXGBE_ATR_L4TYPE_TCP;
5457                 /* l4type IPv4 type is 0, no need to assign */
5458         } else {
5459                 /* Unsupported L4 header, just bail here */
5460                 return;
5461         }
5462
5463         memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
5464
5465         vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
5466                    IXGBE_TX_FLAGS_VLAN_SHIFT;
5467         src_ipv4_addr = iph->saddr;
5468         dst_ipv4_addr = iph->daddr;
5469         flex_bytes = eth->h_proto;
5470
5471         ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
5472         ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
5473         ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
5474         ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
5475         ixgbe_atr_set_l4type_82599(&atr_input, l4type);
5476         /* src and dst are inverted, think how the receiver sees them */
5477         ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
5478         ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
5479
5480         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
5481         ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
5482 }
5483
5484 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
5485                                  struct ixgbe_ring *tx_ring, int size)
5486 {
5487         netif_stop_subqueue(netdev, tx_ring->queue_index);
5488         /* Herbert's original patch had:
5489          *  smp_mb__after_netif_stop_queue();
5490          * but since that doesn't exist yet, just open code it. */
5491         smp_mb();
5492
5493         /* We need to check again in a case another CPU has just
5494          * made room available. */
5495         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
5496                 return -EBUSY;
5497
5498         /* A reprieve! - use start_queue because it doesn't call schedule */
5499         netif_start_subqueue(netdev, tx_ring->queue_index);
5500         ++tx_ring->restart_queue;
5501         return 0;
5502 }
5503
5504 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
5505                               struct ixgbe_ring *tx_ring, int size)
5506 {
5507         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
5508                 return 0;
5509         return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
5510 }
5511
5512 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
5513 {
5514         struct ixgbe_adapter *adapter = netdev_priv(dev);
5515         int txq = smp_processor_id();
5516
5517         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
5518                 return txq;
5519
5520 #ifdef IXGBE_FCOE
5521         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5522             (skb->protocol == htons(ETH_P_FCOE))) {
5523                 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
5524                 txq += adapter->ring_feature[RING_F_FCOE].mask;
5525                 return txq;
5526         }
5527 #endif
5528         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5529                 return (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK) >> 13;
5530
5531         return skb_tx_hash(dev, skb);
5532 }
5533
5534 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
5535                                     struct net_device *netdev)
5536 {
5537         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5538         struct ixgbe_ring *tx_ring;
5539         struct netdev_queue *txq;
5540         unsigned int first;
5541         unsigned int tx_flags = 0;
5542         u8 hdr_len = 0;
5543         int tso;
5544         int count = 0;
5545         unsigned int f;
5546
5547         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
5548                 tx_flags |= vlan_tx_tag_get(skb);
5549                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5550                         tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
5551                         tx_flags |= ((skb->queue_mapping & 0x7) << 13);
5552                 }
5553                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5554                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5555         } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5556                 if (skb->priority != TC_PRIO_CONTROL) {
5557                         tx_flags |= ((skb->queue_mapping & 0x7) << 13);
5558                         tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5559                         tx_flags |= IXGBE_TX_FLAGS_VLAN;
5560                 } else {
5561                         skb->queue_mapping =
5562                                 adapter->ring_feature[RING_F_DCB].indices-1;
5563                 }
5564         }
5565
5566         tx_ring = &adapter->tx_ring[skb->queue_mapping];
5567
5568         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5569             (skb->protocol == htons(ETH_P_FCOE))) {
5570                 tx_flags |= IXGBE_TX_FLAGS_FCOE;
5571 #ifdef IXGBE_FCOE
5572 #ifdef CONFIG_IXGBE_DCB
5573                 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
5574                               << IXGBE_TX_FLAGS_VLAN_SHIFT);
5575                 tx_flags |= ((adapter->fcoe.up << 13)
5576                               << IXGBE_TX_FLAGS_VLAN_SHIFT);
5577 #endif
5578 #endif
5579         }
5580         /* four things can cause us to need a context descriptor */
5581         if (skb_is_gso(skb) ||
5582             (skb->ip_summed == CHECKSUM_PARTIAL) ||
5583             (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
5584             (tx_flags & IXGBE_TX_FLAGS_FCOE))
5585                 count++;
5586
5587         count += TXD_USE_COUNT(skb_headlen(skb));
5588         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5589                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5590
5591         if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
5592                 adapter->tx_busy++;
5593                 return NETDEV_TX_BUSY;
5594         }
5595
5596         first = tx_ring->next_to_use;
5597         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5598 #ifdef IXGBE_FCOE
5599                 /* setup tx offload for FCoE */
5600                 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5601                 if (tso < 0) {
5602                         dev_kfree_skb_any(skb);
5603                         return NETDEV_TX_OK;
5604                 }
5605                 if (tso)
5606                         tx_flags |= IXGBE_TX_FLAGS_FSO;
5607 #endif /* IXGBE_FCOE */
5608         } else {
5609                 if (skb->protocol == htons(ETH_P_IP))
5610                         tx_flags |= IXGBE_TX_FLAGS_IPV4;
5611                 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5612                 if (tso < 0) {
5613                         dev_kfree_skb_any(skb);
5614                         return NETDEV_TX_OK;
5615                 }
5616
5617                 if (tso)
5618                         tx_flags |= IXGBE_TX_FLAGS_TSO;
5619                 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
5620                          (skb->ip_summed == CHECKSUM_PARTIAL))
5621                         tx_flags |= IXGBE_TX_FLAGS_CSUM;
5622         }
5623
5624         count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
5625         if (count) {
5626                 /* add the ATR filter if ATR is on */
5627                 if (tx_ring->atr_sample_rate) {
5628                         ++tx_ring->atr_count;
5629                         if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
5630                              test_bit(__IXGBE_FDIR_INIT_DONE,
5631                                       &tx_ring->reinit_state)) {
5632                                 ixgbe_atr(adapter, skb, tx_ring->queue_index,
5633                                           tx_flags);
5634                                 tx_ring->atr_count = 0;
5635                         }
5636                 }
5637                 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
5638                 txq->tx_bytes += skb->len;
5639                 txq->tx_packets++;
5640                 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
5641                                hdr_len);
5642                 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
5643
5644         } else {
5645                 dev_kfree_skb_any(skb);
5646                 tx_ring->tx_buffer_info[first].time_stamp = 0;
5647                 tx_ring->next_to_use = first;
5648         }
5649
5650         return NETDEV_TX_OK;
5651 }
5652
5653 /**
5654  * ixgbe_set_mac - Change the Ethernet Address of the NIC
5655  * @netdev: network interface device structure
5656  * @p: pointer to an address structure
5657  *
5658  * Returns 0 on success, negative on failure
5659  **/
5660 static int ixgbe_set_mac(struct net_device *netdev, void *p)
5661 {
5662         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5663         struct ixgbe_hw *hw = &adapter->hw;
5664         struct sockaddr *addr = p;
5665
5666         if (!is_valid_ether_addr(addr->sa_data))
5667                 return -EADDRNOTAVAIL;
5668
5669         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
5670         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5671
5672         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
5673                             IXGBE_RAH_AV);
5674
5675         return 0;
5676 }
5677
5678 static int
5679 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
5680 {
5681         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5682         struct ixgbe_hw *hw = &adapter->hw;
5683         u16 value;
5684         int rc;
5685
5686         if (prtad != hw->phy.mdio.prtad)
5687                 return -EINVAL;
5688         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
5689         if (!rc)
5690                 rc = value;
5691         return rc;
5692 }
5693
5694 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
5695                             u16 addr, u16 value)
5696 {
5697         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5698         struct ixgbe_hw *hw = &adapter->hw;
5699
5700         if (prtad != hw->phy.mdio.prtad)
5701                 return -EINVAL;
5702         return hw->phy.ops.write_reg(hw, addr, devad, value);
5703 }
5704
5705 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
5706 {
5707         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5708
5709         return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
5710 }
5711
5712 /**
5713  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
5714  * netdev->dev_addrs
5715  * @netdev: network interface device structure
5716  *
5717  * Returns non-zero on failure
5718  **/
5719 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
5720 {
5721         int err = 0;
5722         struct ixgbe_adapter *adapter = netdev_priv(dev);
5723         struct ixgbe_mac_info *mac = &adapter->hw.mac;
5724
5725         if (is_valid_ether_addr(mac->san_addr)) {
5726                 rtnl_lock();
5727                 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5728                 rtnl_unlock();
5729         }
5730         return err;
5731 }
5732
5733 /**
5734  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
5735  * netdev->dev_addrs
5736  * @netdev: network interface device structure
5737  *
5738  * Returns non-zero on failure
5739  **/
5740 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
5741 {
5742         int err = 0;
5743         struct ixgbe_adapter *adapter = netdev_priv(dev);
5744         struct ixgbe_mac_info *mac = &adapter->hw.mac;
5745
5746         if (is_valid_ether_addr(mac->san_addr)) {
5747                 rtnl_lock();
5748                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5749                 rtnl_unlock();
5750         }
5751         return err;
5752 }
5753
5754 #ifdef CONFIG_NET_POLL_CONTROLLER
5755 /*
5756  * Polling 'interrupt' - used by things like netconsole to send skbs
5757  * without having to re-enable interrupts. It's not called while
5758  * the interrupt routine is executing.
5759  */
5760 static void ixgbe_netpoll(struct net_device *netdev)
5761 {
5762         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5763         int i;
5764
5765         /* if interface is down do nothing */
5766         if (test_bit(__IXGBE_DOWN, &adapter->state))
5767                 return;
5768
5769         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
5770         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5771                 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5772                 for (i = 0; i < num_q_vectors; i++) {
5773                         struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
5774                         ixgbe_msix_clean_many(0, q_vector);
5775                 }
5776         } else {
5777                 ixgbe_intr(adapter->pdev->irq, netdev);
5778         }
5779         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
5780 }
5781 #endif
5782
5783 static const struct net_device_ops ixgbe_netdev_ops = {
5784         .ndo_open               = ixgbe_open,
5785         .ndo_stop               = ixgbe_close,
5786         .ndo_start_xmit         = ixgbe_xmit_frame,
5787         .ndo_select_queue       = ixgbe_select_queue,
5788         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
5789         .ndo_set_multicast_list = ixgbe_set_rx_mode,
5790         .ndo_validate_addr      = eth_validate_addr,
5791         .ndo_set_mac_address    = ixgbe_set_mac,
5792         .ndo_change_mtu         = ixgbe_change_mtu,
5793         .ndo_tx_timeout         = ixgbe_tx_timeout,
5794         .ndo_vlan_rx_register   = ixgbe_vlan_rx_register,
5795         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
5796         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
5797         .ndo_do_ioctl           = ixgbe_ioctl,
5798 #ifdef CONFIG_NET_POLL_CONTROLLER
5799         .ndo_poll_controller    = ixgbe_netpoll,
5800 #endif
5801 #ifdef IXGBE_FCOE
5802         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
5803         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
5804         .ndo_fcoe_enable = ixgbe_fcoe_enable,
5805         .ndo_fcoe_disable = ixgbe_fcoe_disable,
5806         .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
5807 #endif /* IXGBE_FCOE */
5808 };
5809
5810 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
5811                            const struct ixgbe_info *ii)
5812 {
5813 #ifdef CONFIG_PCI_IOV
5814         struct ixgbe_hw *hw = &adapter->hw;
5815         int err;
5816
5817         if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
5818                 return;
5819
5820         /* The 82599 supports up to 64 VFs per physical function
5821          * but this implementation limits allocation to 63 so that
5822          * basic networking resources are still available to the
5823          * physical function
5824          */
5825         adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
5826         adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
5827         err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
5828         if (err) {
5829                 DPRINTK(PROBE, ERR,
5830                         "Failed to enable PCI sriov: %d\n", err);
5831                 goto err_novfs;
5832         }
5833         /* If call to enable VFs succeeded then allocate memory
5834          * for per VF control structures.
5835          */
5836         adapter->vfinfo =
5837                 kcalloc(adapter->num_vfs,
5838                         sizeof(struct vf_data_storage), GFP_KERNEL);
5839         if (adapter->vfinfo) {
5840                 /* Now that we're sure SR-IOV is enabled
5841                  * and memory allocated set up the mailbox parameters
5842                  */
5843                 ixgbe_init_mbx_params_pf(hw);
5844                 memcpy(&hw->mbx.ops, ii->mbx_ops,
5845                        sizeof(hw->mbx.ops));
5846
5847                 /* Disable RSC when in SR-IOV mode */
5848                 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
5849                                      IXGBE_FLAG2_RSC_ENABLED);
5850                 return;
5851         }
5852
5853         /* Oh oh */
5854         DPRINTK(PROBE, ERR,
5855                 "Unable to allocate memory for VF "
5856                 "Data Storage - SRIOV disabled\n");
5857         pci_disable_sriov(adapter->pdev);
5858
5859 err_novfs:
5860         adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
5861         adapter->num_vfs = 0;
5862 #endif /* CONFIG_PCI_IOV */
5863 }
5864
5865 /**
5866  * ixgbe_probe - Device Initialization Routine
5867  * @pdev: PCI device information struct
5868  * @ent: entry in ixgbe_pci_tbl
5869  *
5870  * Returns 0 on success, negative on failure
5871  *
5872  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
5873  * The OS initialization, configuring of the adapter private structure,
5874  * and a hardware reset occur.
5875  **/
5876 static int __devinit ixgbe_probe(struct pci_dev *pdev,
5877                                  const struct pci_device_id *ent)
5878 {
5879         struct net_device *netdev;
5880         struct ixgbe_adapter *adapter = NULL;
5881         struct ixgbe_hw *hw;
5882         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
5883         static int cards_found;
5884         int i, err, pci_using_dac;
5885 #ifdef IXGBE_FCOE
5886         u16 device_caps;
5887 #endif
5888         u32 part_num, eec;
5889
5890         err = pci_enable_device_mem(pdev);
5891         if (err)
5892                 return err;
5893
5894         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
5895             !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
5896                 pci_using_dac = 1;
5897         } else {
5898                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5899                 if (err) {
5900                         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
5901                         if (err) {
5902                                 dev_err(&pdev->dev, "No usable DMA "
5903                                         "configuration, aborting\n");
5904                                 goto err_dma;
5905                         }
5906                 }
5907                 pci_using_dac = 0;
5908         }
5909
5910         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
5911                                            IORESOURCE_MEM), ixgbe_driver_name);
5912         if (err) {
5913                 dev_err(&pdev->dev,
5914                         "pci_request_selected_regions failed 0x%x\n", err);
5915                 goto err_pci_reg;
5916         }
5917
5918         pci_enable_pcie_error_reporting(pdev);
5919
5920         pci_set_master(pdev);
5921         pci_save_state(pdev);
5922
5923         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
5924         if (!netdev) {
5925                 err = -ENOMEM;
5926                 goto err_alloc_etherdev;
5927         }
5928
5929         SET_NETDEV_DEV(netdev, &pdev->dev);
5930
5931         pci_set_drvdata(pdev, netdev);
5932         adapter = netdev_priv(netdev);
5933
5934         adapter->netdev = netdev;
5935         adapter->pdev = pdev;
5936         hw = &adapter->hw;
5937         hw->back = adapter;
5938         adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
5939
5940         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
5941                               pci_resource_len(pdev, 0));
5942         if (!hw->hw_addr) {
5943                 err = -EIO;
5944                 goto err_ioremap;
5945         }
5946
5947         for (i = 1; i <= 5; i++) {
5948                 if (pci_resource_len(pdev, i) == 0)
5949                         continue;
5950         }
5951
5952         netdev->netdev_ops = &ixgbe_netdev_ops;
5953         ixgbe_set_ethtool_ops(netdev);
5954         netdev->watchdog_timeo = 5 * HZ;
5955         strcpy(netdev->name, pci_name(pdev));
5956
5957         adapter->bd_number = cards_found;
5958
5959         /* Setup hw api */
5960         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
5961         hw->mac.type  = ii->mac;
5962
5963         /* EEPROM */
5964         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
5965         eec = IXGBE_READ_REG(hw, IXGBE_EEC);
5966         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
5967         if (!(eec & (1 << 8)))
5968                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
5969
5970         /* PHY */
5971         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
5972         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
5973         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
5974         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
5975         hw->phy.mdio.mmds = 0;
5976         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
5977         hw->phy.mdio.dev = netdev;
5978         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
5979         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
5980
5981         /* set up this timer and work struct before calling get_invariants
5982          * which might start the timer
5983          */
5984         init_timer(&adapter->sfp_timer);
5985         adapter->sfp_timer.function = &ixgbe_sfp_timer;
5986         adapter->sfp_timer.data = (unsigned long) adapter;
5987
5988         INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
5989
5990         /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
5991         INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
5992
5993         /* a new SFP+ module arrival, called from GPI SDP2 context */
5994         INIT_WORK(&adapter->sfp_config_module_task,
5995                   ixgbe_sfp_config_module_task);
5996
5997         ii->get_invariants(hw);
5998
5999         /* setup the private structure */
6000         err = ixgbe_sw_init(adapter);
6001         if (err)
6002                 goto err_sw_init;
6003
6004         /*
6005          * If there is a fan on this device and it has failed log the
6006          * failure.
6007          */
6008         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6009                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6010                 if (esdp & IXGBE_ESDP_SDP1)
6011                         DPRINTK(PROBE, CRIT,
6012                                 "Fan has stopped, replace the adapter\n");
6013         }
6014
6015         /* reset_hw fills in the perm_addr as well */
6016         err = hw->mac.ops.reset_hw(hw);
6017         if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6018             hw->mac.type == ixgbe_mac_82598EB) {
6019                 /*
6020                  * Start a kernel thread to watch for a module to arrive.
6021                  * Only do this for 82598, since 82599 will generate
6022                  * interrupts on module arrival.
6023                  */
6024                 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6025                 mod_timer(&adapter->sfp_timer,
6026                           round_jiffies(jiffies + (2 * HZ)));
6027                 err = 0;
6028         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6029                 dev_err(&adapter->pdev->dev, "failed to initialize because "
6030                         "an unsupported SFP+ module type was detected.\n"
6031                         "Reload the driver after installing a supported "
6032                         "module.\n");
6033                 goto err_sw_init;
6034         } else if (err) {
6035                 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
6036                 goto err_sw_init;
6037         }
6038
6039         ixgbe_probe_vf(adapter, ii);
6040
6041         netdev->features = NETIF_F_SG |
6042                            NETIF_F_IP_CSUM |
6043                            NETIF_F_HW_VLAN_TX |
6044                            NETIF_F_HW_VLAN_RX |
6045                            NETIF_F_HW_VLAN_FILTER;
6046
6047         netdev->features |= NETIF_F_IPV6_CSUM;
6048         netdev->features |= NETIF_F_TSO;
6049         netdev->features |= NETIF_F_TSO6;
6050         netdev->features |= NETIF_F_GRO;
6051
6052         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6053                 netdev->features |= NETIF_F_SCTP_CSUM;
6054
6055         netdev->vlan_features |= NETIF_F_TSO;
6056         netdev->vlan_features |= NETIF_F_TSO6;
6057         netdev->vlan_features |= NETIF_F_IP_CSUM;
6058         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6059         netdev->vlan_features |= NETIF_F_SG;
6060
6061         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6062                 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6063                                     IXGBE_FLAG_DCB_ENABLED);
6064         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6065                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6066
6067 #ifdef CONFIG_IXGBE_DCB
6068         netdev->dcbnl_ops = &dcbnl_ops;
6069 #endif
6070
6071 #ifdef IXGBE_FCOE
6072         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6073                 if (hw->mac.ops.get_device_caps) {
6074                         hw->mac.ops.get_device_caps(hw, &device_caps);
6075                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6076                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6077                 }
6078         }
6079 #endif /* IXGBE_FCOE */
6080         if (pci_using_dac)
6081                 netdev->features |= NETIF_F_HIGHDMA;
6082
6083         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6084                 netdev->features |= NETIF_F_LRO;
6085
6086         /* make sure the EEPROM is good */
6087         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6088                 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
6089                 err = -EIO;
6090                 goto err_eeprom;
6091         }
6092
6093         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6094         memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6095
6096         if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6097                 dev_err(&pdev->dev, "invalid MAC address\n");
6098                 err = -EIO;
6099                 goto err_eeprom;
6100         }
6101
6102         init_timer(&adapter->watchdog_timer);
6103         adapter->watchdog_timer.function = &ixgbe_watchdog;
6104         adapter->watchdog_timer.data = (unsigned long)adapter;
6105
6106         INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6107         INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6108
6109         err = ixgbe_init_interrupt_scheme(adapter);
6110         if (err)
6111                 goto err_sw_init;
6112
6113         switch (pdev->device) {
6114         case IXGBE_DEV_ID_82599_KX4:
6115                 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
6116                                 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
6117                 /* Enable ACPI wakeup in GRC */
6118                 IXGBE_WRITE_REG(hw, IXGBE_GRC,
6119                              (IXGBE_READ_REG(hw, IXGBE_GRC) & ~IXGBE_GRC_APME));
6120                 break;
6121         default:
6122                 adapter->wol = 0;
6123                 break;
6124         }
6125         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6126
6127         /* pick up the PCI bus settings for reporting later */
6128         hw->mac.ops.get_bus_info(hw);
6129
6130         /* print bus type/speed/width info */
6131         dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
6132                 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
6133                  (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
6134                 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
6135                  (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
6136                  (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
6137                  "Unknown"),
6138                 netdev->dev_addr);
6139         ixgbe_read_pba_num_generic(hw, &part_num);
6140         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
6141                 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
6142                          hw->mac.type, hw->phy.type, hw->phy.sfp_type,
6143                          (part_num >> 8), (part_num & 0xff));
6144         else
6145                 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
6146                          hw->mac.type, hw->phy.type,
6147                          (part_num >> 8), (part_num & 0xff));
6148
6149         if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
6150                 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
6151                          "this card is not sufficient for optimal "
6152                          "performance.\n");
6153                 dev_warn(&pdev->dev, "For optimal performance a x8 "
6154                          "PCI-Express slot is required.\n");
6155         }
6156
6157         /* save off EEPROM version number */
6158         hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
6159
6160         /* reset the hardware with the new settings */
6161         err = hw->mac.ops.start_hw(hw);
6162
6163         if (err == IXGBE_ERR_EEPROM_VERSION) {
6164                 /* We are running on a pre-production device, log a warning */
6165                 dev_warn(&pdev->dev, "This device is a pre-production "
6166                          "adapter/LOM.  Please be aware there may be issues "
6167                          "associated with your hardware.  If you are "
6168                          "experiencing problems please contact your Intel or "
6169                          "hardware representative who provided you with this "
6170                          "hardware.\n");
6171         }
6172         strcpy(netdev->name, "eth%d");
6173         err = register_netdev(netdev);
6174         if (err)
6175                 goto err_register;
6176
6177         /* carrier off reporting is important to ethtool even BEFORE open */
6178         netif_carrier_off(netdev);
6179
6180         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6181             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6182                 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
6183
6184 #ifdef CONFIG_IXGBE_DCA
6185         if (dca_add_requester(&pdev->dev) == 0) {
6186                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
6187                 ixgbe_setup_dca(adapter);
6188         }
6189 #endif
6190         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6191                 DPRINTK(PROBE, INFO, "IOV is enabled with %d VFs\n",
6192                         adapter->num_vfs);
6193                 for (i = 0; i < adapter->num_vfs; i++)
6194                         ixgbe_vf_configuration(pdev, (i | 0x10000000));
6195         }
6196
6197         /* add san mac addr to netdev */
6198         ixgbe_add_sanmac_netdev(netdev);
6199
6200         dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
6201         cards_found++;
6202         return 0;
6203
6204 err_register:
6205         ixgbe_release_hw_control(adapter);
6206         ixgbe_clear_interrupt_scheme(adapter);
6207 err_sw_init:
6208 err_eeprom:
6209         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6210                 ixgbe_disable_sriov(adapter);
6211         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6212         del_timer_sync(&adapter->sfp_timer);
6213         cancel_work_sync(&adapter->sfp_task);
6214         cancel_work_sync(&adapter->multispeed_fiber_task);
6215         cancel_work_sync(&adapter->sfp_config_module_task);
6216         iounmap(hw->hw_addr);
6217 err_ioremap:
6218         free_netdev(netdev);
6219 err_alloc_etherdev:
6220         pci_release_selected_regions(pdev, pci_select_bars(pdev,
6221                                      IORESOURCE_MEM));
6222 err_pci_reg:
6223 err_dma:
6224         pci_disable_device(pdev);
6225         return err;
6226 }
6227
6228 /**
6229  * ixgbe_remove - Device Removal Routine
6230  * @pdev: PCI device information struct
6231  *
6232  * ixgbe_remove is called by the PCI subsystem to alert the driver
6233  * that it should release a PCI device.  The could be caused by a
6234  * Hot-Plug event, or because the driver is going to be removed from
6235  * memory.
6236  **/
6237 static void __devexit ixgbe_remove(struct pci_dev *pdev)
6238 {
6239         struct net_device *netdev = pci_get_drvdata(pdev);
6240         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6241
6242         set_bit(__IXGBE_DOWN, &adapter->state);
6243         /* clear the module not found bit to make sure the worker won't
6244          * reschedule
6245          */
6246         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6247         del_timer_sync(&adapter->watchdog_timer);
6248
6249         del_timer_sync(&adapter->sfp_timer);
6250         cancel_work_sync(&adapter->watchdog_task);
6251         cancel_work_sync(&adapter->sfp_task);
6252         cancel_work_sync(&adapter->multispeed_fiber_task);
6253         cancel_work_sync(&adapter->sfp_config_module_task);
6254         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6255             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6256                 cancel_work_sync(&adapter->fdir_reinit_task);
6257         flush_scheduled_work();
6258
6259 #ifdef CONFIG_IXGBE_DCA
6260         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
6261                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
6262                 dca_remove_requester(&pdev->dev);
6263                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
6264         }
6265
6266 #endif
6267 #ifdef IXGBE_FCOE
6268         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
6269                 ixgbe_cleanup_fcoe(adapter);
6270
6271 #endif /* IXGBE_FCOE */
6272
6273         /* remove the added san mac */
6274         ixgbe_del_sanmac_netdev(netdev);
6275
6276         if (netdev->reg_state == NETREG_REGISTERED)
6277                 unregister_netdev(netdev);
6278
6279         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6280                 ixgbe_disable_sriov(adapter);
6281
6282         ixgbe_clear_interrupt_scheme(adapter);
6283
6284         ixgbe_release_hw_control(adapter);
6285
6286         iounmap(adapter->hw.hw_addr);
6287         pci_release_selected_regions(pdev, pci_select_bars(pdev,
6288                                      IORESOURCE_MEM));
6289
6290         DPRINTK(PROBE, INFO, "complete\n");
6291
6292         free_netdev(netdev);
6293
6294         pci_disable_pcie_error_reporting(pdev);
6295
6296         pci_disable_device(pdev);
6297 }
6298
6299 /**
6300  * ixgbe_io_error_detected - called when PCI error is detected
6301  * @pdev: Pointer to PCI device
6302  * @state: The current pci connection state
6303  *
6304  * This function is called after a PCI bus error affecting
6305  * this device has been detected.
6306  */
6307 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
6308                                                 pci_channel_state_t state)
6309 {
6310         struct net_device *netdev = pci_get_drvdata(pdev);
6311         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6312
6313         netif_device_detach(netdev);
6314
6315         if (state == pci_channel_io_perm_failure)
6316                 return PCI_ERS_RESULT_DISCONNECT;
6317
6318         if (netif_running(netdev))
6319                 ixgbe_down(adapter);
6320         pci_disable_device(pdev);
6321
6322         /* Request a slot reset. */
6323         return PCI_ERS_RESULT_NEED_RESET;
6324 }
6325
6326 /**
6327  * ixgbe_io_slot_reset - called after the pci bus has been reset.
6328  * @pdev: Pointer to PCI device
6329  *
6330  * Restart the card from scratch, as if from a cold-boot.
6331  */
6332 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
6333 {
6334         struct net_device *netdev = pci_get_drvdata(pdev);
6335         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6336         pci_ers_result_t result;
6337         int err;
6338
6339         if (pci_enable_device_mem(pdev)) {
6340                 DPRINTK(PROBE, ERR,
6341                         "Cannot re-enable PCI device after reset.\n");
6342                 result = PCI_ERS_RESULT_DISCONNECT;
6343         } else {
6344                 pci_set_master(pdev);
6345                 pci_restore_state(pdev);
6346                 pci_save_state(pdev);
6347
6348                 pci_wake_from_d3(pdev, false);
6349
6350                 ixgbe_reset(adapter);
6351                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6352                 result = PCI_ERS_RESULT_RECOVERED;
6353         }
6354
6355         err = pci_cleanup_aer_uncorrect_error_status(pdev);
6356         if (err) {
6357                 dev_err(&pdev->dev,
6358                   "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
6359                 /* non-fatal, continue */
6360         }
6361
6362         return result;
6363 }
6364
6365 /**
6366  * ixgbe_io_resume - called when traffic can start flowing again.
6367  * @pdev: Pointer to PCI device
6368  *
6369  * This callback is called when the error recovery driver tells us that
6370  * its OK to resume normal operation.
6371  */
6372 static void ixgbe_io_resume(struct pci_dev *pdev)
6373 {
6374         struct net_device *netdev = pci_get_drvdata(pdev);
6375         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6376
6377         if (netif_running(netdev)) {
6378                 if (ixgbe_up(adapter)) {
6379                         DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
6380                         return;
6381                 }
6382         }
6383
6384         netif_device_attach(netdev);
6385 }
6386
6387 static struct pci_error_handlers ixgbe_err_handler = {
6388         .error_detected = ixgbe_io_error_detected,
6389         .slot_reset = ixgbe_io_slot_reset,
6390         .resume = ixgbe_io_resume,
6391 };
6392
6393 static struct pci_driver ixgbe_driver = {
6394         .name     = ixgbe_driver_name,
6395         .id_table = ixgbe_pci_tbl,
6396         .probe    = ixgbe_probe,
6397         .remove   = __devexit_p(ixgbe_remove),
6398 #ifdef CONFIG_PM
6399         .suspend  = ixgbe_suspend,
6400         .resume   = ixgbe_resume,
6401 #endif
6402         .shutdown = ixgbe_shutdown,
6403         .err_handler = &ixgbe_err_handler
6404 };
6405
6406 /**
6407  * ixgbe_init_module - Driver Registration Routine
6408  *
6409  * ixgbe_init_module is the first routine called when the driver is
6410  * loaded. All it does is register with the PCI subsystem.
6411  **/
6412 static int __init ixgbe_init_module(void)
6413 {
6414         int ret;
6415         printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
6416                ixgbe_driver_string, ixgbe_driver_version);
6417
6418         printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
6419
6420 #ifdef CONFIG_IXGBE_DCA
6421         dca_register_notify(&dca_notifier);
6422 #endif
6423
6424         ret = pci_register_driver(&ixgbe_driver);
6425         return ret;
6426 }
6427
6428 module_init(ixgbe_init_module);
6429
6430 /**
6431  * ixgbe_exit_module - Driver Exit Cleanup Routine
6432  *
6433  * ixgbe_exit_module is called just before the driver is removed
6434  * from memory.
6435  **/
6436 static void __exit ixgbe_exit_module(void)
6437 {
6438 #ifdef CONFIG_IXGBE_DCA
6439         dca_unregister_notify(&dca_notifier);
6440 #endif
6441         pci_unregister_driver(&ixgbe_driver);
6442 }
6443
6444 #ifdef CONFIG_IXGBE_DCA
6445 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
6446                             void *p)
6447 {
6448         int ret_val;
6449
6450         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
6451                                          __ixgbe_notify_dca);
6452
6453         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6454 }
6455
6456 #endif /* CONFIG_IXGBE_DCA */
6457 #ifdef DEBUG
6458 /**
6459  * ixgbe_get_hw_dev_name - return device name string
6460  * used by hardware layer to print debugging information
6461  **/
6462 char *ixgbe_get_hw_dev_name(struct ixgbe_hw *hw)
6463 {
6464         struct ixgbe_adapter *adapter = hw->back;
6465         return adapter->netdev->name;
6466 }
6467
6468 #endif
6469 module_exit(ixgbe_exit_module);
6470
6471 /* ixgbe_main.c */