d75c46ff31f65041859bfe98ade9d766057071e1
[linux-2.6.git] / 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 "2.0.62-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->reg_idx;
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->reg_idx;
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 struct ixgbe_rsc_cb {
822         dma_addr_t dma;
823 };
824
825 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
826
827 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
828                                struct ixgbe_ring *rx_ring,
829                                int *work_done, int work_to_do)
830 {
831         struct ixgbe_adapter *adapter = q_vector->adapter;
832         struct net_device *netdev = adapter->netdev;
833         struct pci_dev *pdev = adapter->pdev;
834         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
835         struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
836         struct sk_buff *skb;
837         unsigned int i, rsc_count = 0;
838         u32 len, staterr;
839         u16 hdr_info;
840         bool cleaned = false;
841         int cleaned_count = 0;
842         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
843 #ifdef IXGBE_FCOE
844         int ddp_bytes = 0;
845 #endif /* IXGBE_FCOE */
846
847         i = rx_ring->next_to_clean;
848         rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
849         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
850         rx_buffer_info = &rx_ring->rx_buffer_info[i];
851
852         while (staterr & IXGBE_RXD_STAT_DD) {
853                 u32 upper_len = 0;
854                 if (*work_done >= work_to_do)
855                         break;
856                 (*work_done)++;
857
858                 rmb(); /* read descriptor and rx_buffer_info after status DD */
859                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
860                         hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
861                         len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
862                                IXGBE_RXDADV_HDRBUFLEN_SHIFT;
863                         if (len > IXGBE_RX_HDR_SIZE)
864                                 len = IXGBE_RX_HDR_SIZE;
865                         upper_len = le16_to_cpu(rx_desc->wb.upper.length);
866                 } else {
867                         len = le16_to_cpu(rx_desc->wb.upper.length);
868                 }
869
870                 cleaned = true;
871                 skb = rx_buffer_info->skb;
872                 prefetch(skb->data);
873                 rx_buffer_info->skb = NULL;
874
875                 if (rx_buffer_info->dma) {
876                         if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
877                             (!(staterr & IXGBE_RXD_STAT_EOP)) &&
878                                  (!(skb->prev)))
879                                 /*
880                                  * When HWRSC is enabled, delay unmapping
881                                  * of the first packet. It carries the
882                                  * header information, HW may still
883                                  * access the header after the writeback.
884                                  * Only unmap it when EOP is reached
885                                  */
886                                 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
887                         else
888                                 pci_unmap_single(pdev, rx_buffer_info->dma,
889                                                  rx_ring->rx_buf_len,
890                                                  PCI_DMA_FROMDEVICE);
891                         rx_buffer_info->dma = 0;
892                         skb_put(skb, len);
893                 }
894
895                 if (upper_len) {
896                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
897                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
898                         rx_buffer_info->page_dma = 0;
899                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
900                                            rx_buffer_info->page,
901                                            rx_buffer_info->page_offset,
902                                            upper_len);
903
904                         if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
905                             (page_count(rx_buffer_info->page) != 1))
906                                 rx_buffer_info->page = NULL;
907                         else
908                                 get_page(rx_buffer_info->page);
909
910                         skb->len += upper_len;
911                         skb->data_len += upper_len;
912                         skb->truesize += upper_len;
913                 }
914
915                 i++;
916                 if (i == rx_ring->count)
917                         i = 0;
918
919                 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
920                 prefetch(next_rxd);
921                 cleaned_count++;
922
923                 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
924                         rsc_count = ixgbe_get_rsc_count(rx_desc);
925
926                 if (rsc_count) {
927                         u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
928                                      IXGBE_RXDADV_NEXTP_SHIFT;
929                         next_buffer = &rx_ring->rx_buffer_info[nextp];
930                 } else {
931                         next_buffer = &rx_ring->rx_buffer_info[i];
932                 }
933
934                 if (staterr & IXGBE_RXD_STAT_EOP) {
935                         if (skb->prev)
936                                 skb = ixgbe_transform_rsc_queue(skb, &(rx_ring->rsc_count));
937                         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
938                                 if (IXGBE_RSC_CB(skb)->dma) {
939                                         pci_unmap_single(pdev, IXGBE_RSC_CB(skb)->dma,
940                                                          rx_ring->rx_buf_len,
941                                                          PCI_DMA_FROMDEVICE);
942                                         IXGBE_RSC_CB(skb)->dma = 0;
943                                 }
944                                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)
945                                         rx_ring->rsc_count += skb_shinfo(skb)->nr_frags;
946                                 else
947                                         rx_ring->rsc_count++;
948                                 rx_ring->rsc_flush++;
949                         }
950                         rx_ring->stats.packets++;
951                         rx_ring->stats.bytes += skb->len;
952                 } else {
953                         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
954                                 rx_buffer_info->skb = next_buffer->skb;
955                                 rx_buffer_info->dma = next_buffer->dma;
956                                 next_buffer->skb = skb;
957                                 next_buffer->dma = 0;
958                         } else {
959                                 skb->next = next_buffer->skb;
960                                 skb->next->prev = skb;
961                         }
962                         rx_ring->non_eop_descs++;
963                         goto next_desc;
964                 }
965
966                 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
967                         dev_kfree_skb_irq(skb);
968                         goto next_desc;
969                 }
970
971                 ixgbe_rx_checksum(adapter, rx_desc, skb);
972
973                 /* probably a little skewed due to removing CRC */
974                 total_rx_bytes += skb->len;
975                 total_rx_packets++;
976
977                 skb->protocol = eth_type_trans(skb, adapter->netdev);
978 #ifdef IXGBE_FCOE
979                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
980                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
981                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
982                         if (!ddp_bytes)
983                                 goto next_desc;
984                 }
985 #endif /* IXGBE_FCOE */
986                 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
987
988 next_desc:
989                 rx_desc->wb.upper.status_error = 0;
990
991                 /* return some buffers to hardware, one at a time is too slow */
992                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
993                         ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
994                         cleaned_count = 0;
995                 }
996
997                 /* use prefetched values */
998                 rx_desc = next_rxd;
999                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1000
1001                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1002         }
1003
1004         rx_ring->next_to_clean = i;
1005         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1006
1007         if (cleaned_count)
1008                 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1009
1010 #ifdef IXGBE_FCOE
1011         /* include DDPed FCoE data */
1012         if (ddp_bytes > 0) {
1013                 unsigned int mss;
1014
1015                 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
1016                         sizeof(struct fc_frame_header) -
1017                         sizeof(struct fcoe_crc_eof);
1018                 if (mss > 512)
1019                         mss &= ~511;
1020                 total_rx_bytes += ddp_bytes;
1021                 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1022         }
1023 #endif /* IXGBE_FCOE */
1024
1025         rx_ring->total_packets += total_rx_packets;
1026         rx_ring->total_bytes += total_rx_bytes;
1027         netdev->stats.rx_bytes += total_rx_bytes;
1028         netdev->stats.rx_packets += total_rx_packets;
1029
1030         return cleaned;
1031 }
1032
1033 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1034 /**
1035  * ixgbe_configure_msix - Configure MSI-X hardware
1036  * @adapter: board private structure
1037  *
1038  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1039  * interrupts.
1040  **/
1041 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1042 {
1043         struct ixgbe_q_vector *q_vector;
1044         int i, j, q_vectors, v_idx, r_idx;
1045         u32 mask;
1046
1047         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1048
1049         /*
1050          * Populate the IVAR table and set the ITR values to the
1051          * corresponding register.
1052          */
1053         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1054                 q_vector = adapter->q_vector[v_idx];
1055                 /* XXX for_each_set_bit(...) */
1056                 r_idx = find_first_bit(q_vector->rxr_idx,
1057                                        adapter->num_rx_queues);
1058
1059                 for (i = 0; i < q_vector->rxr_count; i++) {
1060                         j = adapter->rx_ring[r_idx]->reg_idx;
1061                         ixgbe_set_ivar(adapter, 0, j, v_idx);
1062                         r_idx = find_next_bit(q_vector->rxr_idx,
1063                                               adapter->num_rx_queues,
1064                                               r_idx + 1);
1065                 }
1066                 r_idx = find_first_bit(q_vector->txr_idx,
1067                                        adapter->num_tx_queues);
1068
1069                 for (i = 0; i < q_vector->txr_count; i++) {
1070                         j = adapter->tx_ring[r_idx]->reg_idx;
1071                         ixgbe_set_ivar(adapter, 1, j, v_idx);
1072                         r_idx = find_next_bit(q_vector->txr_idx,
1073                                               adapter->num_tx_queues,
1074                                               r_idx + 1);
1075                 }
1076
1077                 if (q_vector->txr_count && !q_vector->rxr_count)
1078                         /* tx only */
1079                         q_vector->eitr = adapter->tx_eitr_param;
1080                 else if (q_vector->rxr_count)
1081                         /* rx or mixed */
1082                         q_vector->eitr = adapter->rx_eitr_param;
1083
1084                 ixgbe_write_eitr(q_vector);
1085         }
1086
1087         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1088                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1089                                v_idx);
1090         else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1091                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1092         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1093
1094         /* set up to autoclear timer, and the vectors */
1095         mask = IXGBE_EIMS_ENABLE_MASK;
1096         if (adapter->num_vfs)
1097                 mask &= ~(IXGBE_EIMS_OTHER |
1098                           IXGBE_EIMS_MAILBOX |
1099                           IXGBE_EIMS_LSC);
1100         else
1101                 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1102         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1103 }
1104
1105 enum latency_range {
1106         lowest_latency = 0,
1107         low_latency = 1,
1108         bulk_latency = 2,
1109         latency_invalid = 255
1110 };
1111
1112 /**
1113  * ixgbe_update_itr - update the dynamic ITR value based on statistics
1114  * @adapter: pointer to adapter
1115  * @eitr: eitr setting (ints per sec) to give last timeslice
1116  * @itr_setting: current throttle rate in ints/second
1117  * @packets: the number of packets during this measurement interval
1118  * @bytes: the number of bytes during this measurement interval
1119  *
1120  *      Stores a new ITR value based on packets and byte
1121  *      counts during the last interrupt.  The advantage of per interrupt
1122  *      computation is faster updates and more accurate ITR for the current
1123  *      traffic pattern.  Constants in this function were computed
1124  *      based on theoretical maximum wire speed and thresholds were set based
1125  *      on testing data as well as attempting to minimize response time
1126  *      while increasing bulk throughput.
1127  *      this functionality is controlled by the InterruptThrottleRate module
1128  *      parameter (see ixgbe_param.c)
1129  **/
1130 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1131                            u32 eitr, u8 itr_setting,
1132                            int packets, int bytes)
1133 {
1134         unsigned int retval = itr_setting;
1135         u32 timepassed_us;
1136         u64 bytes_perint;
1137
1138         if (packets == 0)
1139                 goto update_itr_done;
1140
1141
1142         /* simple throttlerate management
1143          *    0-20MB/s lowest (100000 ints/s)
1144          *   20-100MB/s low   (20000 ints/s)
1145          *  100-1249MB/s bulk (8000 ints/s)
1146          */
1147         /* what was last interrupt timeslice? */
1148         timepassed_us = 1000000/eitr;
1149         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1150
1151         switch (itr_setting) {
1152         case lowest_latency:
1153                 if (bytes_perint > adapter->eitr_low)
1154                         retval = low_latency;
1155                 break;
1156         case low_latency:
1157                 if (bytes_perint > adapter->eitr_high)
1158                         retval = bulk_latency;
1159                 else if (bytes_perint <= adapter->eitr_low)
1160                         retval = lowest_latency;
1161                 break;
1162         case bulk_latency:
1163                 if (bytes_perint <= adapter->eitr_high)
1164                         retval = low_latency;
1165                 break;
1166         }
1167
1168 update_itr_done:
1169         return retval;
1170 }
1171
1172 /**
1173  * ixgbe_write_eitr - write EITR register in hardware specific way
1174  * @q_vector: structure containing interrupt and ring information
1175  *
1176  * This function is made to be called by ethtool and by the driver
1177  * when it needs to update EITR registers at runtime.  Hardware
1178  * specific quirks/differences are taken care of here.
1179  */
1180 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1181 {
1182         struct ixgbe_adapter *adapter = q_vector->adapter;
1183         struct ixgbe_hw *hw = &adapter->hw;
1184         int v_idx = q_vector->v_idx;
1185         u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1186
1187         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1188                 /* must write high and low 16 bits to reset counter */
1189                 itr_reg |= (itr_reg << 16);
1190         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1191                 /*
1192                  * set the WDIS bit to not clear the timer bits and cause an
1193                  * immediate assertion of the interrupt
1194                  */
1195                 itr_reg |= IXGBE_EITR_CNT_WDIS;
1196         }
1197         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1198 }
1199
1200 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1201 {
1202         struct ixgbe_adapter *adapter = q_vector->adapter;
1203         u32 new_itr;
1204         u8 current_itr, ret_itr;
1205         int i, r_idx;
1206         struct ixgbe_ring *rx_ring, *tx_ring;
1207
1208         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1209         for (i = 0; i < q_vector->txr_count; i++) {
1210                 tx_ring = adapter->tx_ring[r_idx];
1211                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1212                                            q_vector->tx_itr,
1213                                            tx_ring->total_packets,
1214                                            tx_ring->total_bytes);
1215                 /* if the result for this queue would decrease interrupt
1216                  * rate for this vector then use that result */
1217                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1218                                     q_vector->tx_itr - 1 : ret_itr);
1219                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1220                                       r_idx + 1);
1221         }
1222
1223         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1224         for (i = 0; i < q_vector->rxr_count; i++) {
1225                 rx_ring = adapter->rx_ring[r_idx];
1226                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1227                                            q_vector->rx_itr,
1228                                            rx_ring->total_packets,
1229                                            rx_ring->total_bytes);
1230                 /* if the result for this queue would decrease interrupt
1231                  * rate for this vector then use that result */
1232                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1233                                     q_vector->rx_itr - 1 : ret_itr);
1234                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1235                                       r_idx + 1);
1236         }
1237
1238         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1239
1240         switch (current_itr) {
1241         /* counts and packets in update_itr are dependent on these numbers */
1242         case lowest_latency:
1243                 new_itr = 100000;
1244                 break;
1245         case low_latency:
1246                 new_itr = 20000; /* aka hwitr = ~200 */
1247                 break;
1248         case bulk_latency:
1249         default:
1250                 new_itr = 8000;
1251                 break;
1252         }
1253
1254         if (new_itr != q_vector->eitr) {
1255                 /* do an exponential smoothing */
1256                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1257
1258                 /* save the algorithm value here, not the smoothed one */
1259                 q_vector->eitr = new_itr;
1260
1261                 ixgbe_write_eitr(q_vector);
1262         }
1263
1264         return;
1265 }
1266
1267 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1268 {
1269         struct ixgbe_hw *hw = &adapter->hw;
1270
1271         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1272             (eicr & IXGBE_EICR_GPI_SDP1)) {
1273                 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
1274                 /* write to clear the interrupt */
1275                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1276         }
1277 }
1278
1279 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1280 {
1281         struct ixgbe_hw *hw = &adapter->hw;
1282
1283         if (eicr & IXGBE_EICR_GPI_SDP1) {
1284                 /* Clear the interrupt */
1285                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1286                 schedule_work(&adapter->multispeed_fiber_task);
1287         } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1288                 /* Clear the interrupt */
1289                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1290                 schedule_work(&adapter->sfp_config_module_task);
1291         } else {
1292                 /* Interrupt isn't for us... */
1293                 return;
1294         }
1295 }
1296
1297 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1298 {
1299         struct ixgbe_hw *hw = &adapter->hw;
1300
1301         adapter->lsc_int++;
1302         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1303         adapter->link_check_timeout = jiffies;
1304         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1305                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1306                 IXGBE_WRITE_FLUSH(hw);
1307                 schedule_work(&adapter->watchdog_task);
1308         }
1309 }
1310
1311 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1312 {
1313         struct net_device *netdev = data;
1314         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1315         struct ixgbe_hw *hw = &adapter->hw;
1316         u32 eicr;
1317
1318         /*
1319          * Workaround for Silicon errata.  Use clear-by-write instead
1320          * of clear-by-read.  Reading with EICS will return the
1321          * interrupt causes without clearing, which later be done
1322          * with the write to EICR.
1323          */
1324         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1325         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1326
1327         if (eicr & IXGBE_EICR_LSC)
1328                 ixgbe_check_lsc(adapter);
1329
1330         if (eicr & IXGBE_EICR_MAILBOX)
1331                 ixgbe_msg_task(adapter);
1332
1333         if (hw->mac.type == ixgbe_mac_82598EB)
1334                 ixgbe_check_fan_failure(adapter, eicr);
1335
1336         if (hw->mac.type == ixgbe_mac_82599EB) {
1337                 ixgbe_check_sfp_event(adapter, eicr);
1338
1339                 /* Handle Flow Director Full threshold interrupt */
1340                 if (eicr & IXGBE_EICR_FLOW_DIR) {
1341                         int i;
1342                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1343                         /* Disable transmits before FDIR Re-initialization */
1344                         netif_tx_stop_all_queues(netdev);
1345                         for (i = 0; i < adapter->num_tx_queues; i++) {
1346                                 struct ixgbe_ring *tx_ring =
1347                                                             adapter->tx_ring[i];
1348                                 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1349                                                        &tx_ring->reinit_state))
1350                                         schedule_work(&adapter->fdir_reinit_task);
1351                         }
1352                 }
1353         }
1354         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1355                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1356
1357         return IRQ_HANDLED;
1358 }
1359
1360 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1361                                            u64 qmask)
1362 {
1363         u32 mask;
1364
1365         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1366                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1367                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1368         } else {
1369                 mask = (qmask & 0xFFFFFFFF);
1370                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1371                 mask = (qmask >> 32);
1372                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1373         }
1374         /* skip the flush */
1375 }
1376
1377 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1378                                             u64 qmask)
1379 {
1380         u32 mask;
1381
1382         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1383                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1384                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1385         } else {
1386                 mask = (qmask & 0xFFFFFFFF);
1387                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1388                 mask = (qmask >> 32);
1389                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1390         }
1391         /* skip the flush */
1392 }
1393
1394 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1395 {
1396         struct ixgbe_q_vector *q_vector = data;
1397         struct ixgbe_adapter  *adapter = q_vector->adapter;
1398         struct ixgbe_ring     *tx_ring;
1399         int i, r_idx;
1400
1401         if (!q_vector->txr_count)
1402                 return IRQ_HANDLED;
1403
1404         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1405         for (i = 0; i < q_vector->txr_count; i++) {
1406                 tx_ring = adapter->tx_ring[r_idx];
1407                 tx_ring->total_bytes = 0;
1408                 tx_ring->total_packets = 0;
1409                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1410                                       r_idx + 1);
1411         }
1412
1413         /* EIAM disabled interrupts (on this vector) for us */
1414         napi_schedule(&q_vector->napi);
1415
1416         return IRQ_HANDLED;
1417 }
1418
1419 /**
1420  * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1421  * @irq: unused
1422  * @data: pointer to our q_vector struct for this interrupt vector
1423  **/
1424 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1425 {
1426         struct ixgbe_q_vector *q_vector = data;
1427         struct ixgbe_adapter  *adapter = q_vector->adapter;
1428         struct ixgbe_ring  *rx_ring;
1429         int r_idx;
1430         int i;
1431
1432         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1433         for (i = 0;  i < q_vector->rxr_count; i++) {
1434                 rx_ring = adapter->rx_ring[r_idx];
1435                 rx_ring->total_bytes = 0;
1436                 rx_ring->total_packets = 0;
1437                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1438                                       r_idx + 1);
1439         }
1440
1441         if (!q_vector->rxr_count)
1442                 return IRQ_HANDLED;
1443
1444         /* disable interrupts on this vector only */
1445         /* EIAM disabled interrupts (on this vector) for us */
1446         napi_schedule(&q_vector->napi);
1447
1448         return IRQ_HANDLED;
1449 }
1450
1451 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1452 {
1453         struct ixgbe_q_vector *q_vector = data;
1454         struct ixgbe_adapter  *adapter = q_vector->adapter;
1455         struct ixgbe_ring  *ring;
1456         int r_idx;
1457         int i;
1458
1459         if (!q_vector->txr_count && !q_vector->rxr_count)
1460                 return IRQ_HANDLED;
1461
1462         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1463         for (i = 0; i < q_vector->txr_count; i++) {
1464                 ring = adapter->tx_ring[r_idx];
1465                 ring->total_bytes = 0;
1466                 ring->total_packets = 0;
1467                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1468                                       r_idx + 1);
1469         }
1470
1471         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1472         for (i = 0; i < q_vector->rxr_count; i++) {
1473                 ring = adapter->rx_ring[r_idx];
1474                 ring->total_bytes = 0;
1475                 ring->total_packets = 0;
1476                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1477                                       r_idx + 1);
1478         }
1479
1480         /* EIAM disabled interrupts (on this vector) for us */
1481         napi_schedule(&q_vector->napi);
1482
1483         return IRQ_HANDLED;
1484 }
1485
1486 /**
1487  * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1488  * @napi: napi struct with our devices info in it
1489  * @budget: amount of work driver is allowed to do this pass, in packets
1490  *
1491  * This function is optimized for cleaning one queue only on a single
1492  * q_vector!!!
1493  **/
1494 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1495 {
1496         struct ixgbe_q_vector *q_vector =
1497                                container_of(napi, struct ixgbe_q_vector, napi);
1498         struct ixgbe_adapter *adapter = q_vector->adapter;
1499         struct ixgbe_ring *rx_ring = NULL;
1500         int work_done = 0;
1501         long r_idx;
1502
1503         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1504         rx_ring = adapter->rx_ring[r_idx];
1505 #ifdef CONFIG_IXGBE_DCA
1506         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1507                 ixgbe_update_rx_dca(adapter, rx_ring);
1508 #endif
1509
1510         ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1511
1512         /* If all Rx work done, exit the polling mode */
1513         if (work_done < budget) {
1514                 napi_complete(napi);
1515                 if (adapter->rx_itr_setting & 1)
1516                         ixgbe_set_itr_msix(q_vector);
1517                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1518                         ixgbe_irq_enable_queues(adapter,
1519                                                 ((u64)1 << q_vector->v_idx));
1520         }
1521
1522         return work_done;
1523 }
1524
1525 /**
1526  * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1527  * @napi: napi struct with our devices info in it
1528  * @budget: amount of work driver is allowed to do this pass, in packets
1529  *
1530  * This function will clean more than one rx queue associated with a
1531  * q_vector.
1532  **/
1533 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1534 {
1535         struct ixgbe_q_vector *q_vector =
1536                                container_of(napi, struct ixgbe_q_vector, napi);
1537         struct ixgbe_adapter *adapter = q_vector->adapter;
1538         struct ixgbe_ring *ring = NULL;
1539         int work_done = 0, i;
1540         long r_idx;
1541         bool tx_clean_complete = true;
1542
1543         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1544         for (i = 0; i < q_vector->txr_count; i++) {
1545                 ring = adapter->tx_ring[r_idx];
1546 #ifdef CONFIG_IXGBE_DCA
1547                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1548                         ixgbe_update_tx_dca(adapter, ring);
1549 #endif
1550                 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1551                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1552                                       r_idx + 1);
1553         }
1554
1555         /* attempt to distribute budget to each queue fairly, but don't allow
1556          * the budget to go below 1 because we'll exit polling */
1557         budget /= (q_vector->rxr_count ?: 1);
1558         budget = max(budget, 1);
1559         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1560         for (i = 0; i < q_vector->rxr_count; i++) {
1561                 ring = adapter->rx_ring[r_idx];
1562 #ifdef CONFIG_IXGBE_DCA
1563                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1564                         ixgbe_update_rx_dca(adapter, ring);
1565 #endif
1566                 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1567                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1568                                       r_idx + 1);
1569         }
1570
1571         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1572         ring = adapter->rx_ring[r_idx];
1573         /* If all Rx work done, exit the polling mode */
1574         if (work_done < budget) {
1575                 napi_complete(napi);
1576                 if (adapter->rx_itr_setting & 1)
1577                         ixgbe_set_itr_msix(q_vector);
1578                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1579                         ixgbe_irq_enable_queues(adapter,
1580                                                 ((u64)1 << q_vector->v_idx));
1581                 return 0;
1582         }
1583
1584         return work_done;
1585 }
1586
1587 /**
1588  * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1589  * @napi: napi struct with our devices info in it
1590  * @budget: amount of work driver is allowed to do this pass, in packets
1591  *
1592  * This function is optimized for cleaning one queue only on a single
1593  * q_vector!!!
1594  **/
1595 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1596 {
1597         struct ixgbe_q_vector *q_vector =
1598                                container_of(napi, struct ixgbe_q_vector, napi);
1599         struct ixgbe_adapter *adapter = q_vector->adapter;
1600         struct ixgbe_ring *tx_ring = NULL;
1601         int work_done = 0;
1602         long r_idx;
1603
1604         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1605         tx_ring = adapter->tx_ring[r_idx];
1606 #ifdef CONFIG_IXGBE_DCA
1607         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1608                 ixgbe_update_tx_dca(adapter, tx_ring);
1609 #endif
1610
1611         if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
1612                 work_done = budget;
1613
1614         /* If all Tx work done, exit the polling mode */
1615         if (work_done < budget) {
1616                 napi_complete(napi);
1617                 if (adapter->tx_itr_setting & 1)
1618                         ixgbe_set_itr_msix(q_vector);
1619                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1620                         ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
1621         }
1622
1623         return work_done;
1624 }
1625
1626 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1627                                      int r_idx)
1628 {
1629         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1630
1631         set_bit(r_idx, q_vector->rxr_idx);
1632         q_vector->rxr_count++;
1633 }
1634
1635 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1636                                      int t_idx)
1637 {
1638         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1639
1640         set_bit(t_idx, q_vector->txr_idx);
1641         q_vector->txr_count++;
1642 }
1643
1644 /**
1645  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1646  * @adapter: board private structure to initialize
1647  * @vectors: allotted vector count for descriptor rings
1648  *
1649  * This function maps descriptor rings to the queue-specific vectors
1650  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1651  * one vector per ring/queue, but on a constrained vector budget, we
1652  * group the rings as "efficiently" as possible.  You would add new
1653  * mapping configurations in here.
1654  **/
1655 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1656                                       int vectors)
1657 {
1658         int v_start = 0;
1659         int rxr_idx = 0, txr_idx = 0;
1660         int rxr_remaining = adapter->num_rx_queues;
1661         int txr_remaining = adapter->num_tx_queues;
1662         int i, j;
1663         int rqpv, tqpv;
1664         int err = 0;
1665
1666         /* No mapping required if MSI-X is disabled. */
1667         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1668                 goto out;
1669
1670         /*
1671          * The ideal configuration...
1672          * We have enough vectors to map one per queue.
1673          */
1674         if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1675                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1676                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1677
1678                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1679                         map_vector_to_txq(adapter, v_start, txr_idx);
1680
1681                 goto out;
1682         }
1683
1684         /*
1685          * If we don't have enough vectors for a 1-to-1
1686          * mapping, we'll have to group them so there are
1687          * multiple queues per vector.
1688          */
1689         /* Re-adjusting *qpv takes care of the remainder. */
1690         for (i = v_start; i < vectors; i++) {
1691                 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1692                 for (j = 0; j < rqpv; j++) {
1693                         map_vector_to_rxq(adapter, i, rxr_idx);
1694                         rxr_idx++;
1695                         rxr_remaining--;
1696                 }
1697         }
1698         for (i = v_start; i < vectors; i++) {
1699                 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1700                 for (j = 0; j < tqpv; j++) {
1701                         map_vector_to_txq(adapter, i, txr_idx);
1702                         txr_idx++;
1703                         txr_remaining--;
1704                 }
1705         }
1706
1707 out:
1708         return err;
1709 }
1710
1711 /**
1712  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1713  * @adapter: board private structure
1714  *
1715  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1716  * interrupts from the kernel.
1717  **/
1718 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1719 {
1720         struct net_device *netdev = adapter->netdev;
1721         irqreturn_t (*handler)(int, void *);
1722         int i, vector, q_vectors, err;
1723         int ri=0, ti=0;
1724
1725         /* Decrement for Other and TCP Timer vectors */
1726         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1727
1728         /* Map the Tx/Rx rings to the vectors we were allotted. */
1729         err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1730         if (err)
1731                 goto out;
1732
1733 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1734                          (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1735                          &ixgbe_msix_clean_many)
1736         for (vector = 0; vector < q_vectors; vector++) {
1737                 handler = SET_HANDLER(adapter->q_vector[vector]);
1738
1739                 if(handler == &ixgbe_msix_clean_rx) {
1740                         sprintf(adapter->name[vector], "%s-%s-%d",
1741                                 netdev->name, "rx", ri++);
1742                 }
1743                 else if(handler == &ixgbe_msix_clean_tx) {
1744                         sprintf(adapter->name[vector], "%s-%s-%d",
1745                                 netdev->name, "tx", ti++);
1746                 }
1747                 else
1748                         sprintf(adapter->name[vector], "%s-%s-%d",
1749                                 netdev->name, "TxRx", vector);
1750
1751                 err = request_irq(adapter->msix_entries[vector].vector,
1752                                   handler, 0, adapter->name[vector],
1753                                   adapter->q_vector[vector]);
1754                 if (err) {
1755                         DPRINTK(PROBE, ERR,
1756                                 "request_irq failed for MSIX interrupt "
1757                                 "Error: %d\n", err);
1758                         goto free_queue_irqs;
1759                 }
1760         }
1761
1762         sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1763         err = request_irq(adapter->msix_entries[vector].vector,
1764                           ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1765         if (err) {
1766                 DPRINTK(PROBE, ERR,
1767                         "request_irq for msix_lsc failed: %d\n", err);
1768                 goto free_queue_irqs;
1769         }
1770
1771         return 0;
1772
1773 free_queue_irqs:
1774         for (i = vector - 1; i >= 0; i--)
1775                 free_irq(adapter->msix_entries[--vector].vector,
1776                          adapter->q_vector[i]);
1777         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1778         pci_disable_msix(adapter->pdev);
1779         kfree(adapter->msix_entries);
1780         adapter->msix_entries = NULL;
1781 out:
1782         return err;
1783 }
1784
1785 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1786 {
1787         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1788         u8 current_itr;
1789         u32 new_itr = q_vector->eitr;
1790         struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
1791         struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
1792
1793         q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1794                                             q_vector->tx_itr,
1795                                             tx_ring->total_packets,
1796                                             tx_ring->total_bytes);
1797         q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1798                                             q_vector->rx_itr,
1799                                             rx_ring->total_packets,
1800                                             rx_ring->total_bytes);
1801
1802         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1803
1804         switch (current_itr) {
1805         /* counts and packets in update_itr are dependent on these numbers */
1806         case lowest_latency:
1807                 new_itr = 100000;
1808                 break;
1809         case low_latency:
1810                 new_itr = 20000; /* aka hwitr = ~200 */
1811                 break;
1812         case bulk_latency:
1813                 new_itr = 8000;
1814                 break;
1815         default:
1816                 break;
1817         }
1818
1819         if (new_itr != q_vector->eitr) {
1820                 /* do an exponential smoothing */
1821                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1822
1823                 /* save the algorithm value here, not the smoothed one */
1824                 q_vector->eitr = new_itr;
1825
1826                 ixgbe_write_eitr(q_vector);
1827         }
1828
1829         return;
1830 }
1831
1832 /**
1833  * ixgbe_irq_enable - Enable default interrupt generation settings
1834  * @adapter: board private structure
1835  **/
1836 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1837 {
1838         u32 mask;
1839
1840         mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1841         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1842                 mask |= IXGBE_EIMS_GPI_SDP1;
1843         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1844                 mask |= IXGBE_EIMS_ECC;
1845                 mask |= IXGBE_EIMS_GPI_SDP1;
1846                 mask |= IXGBE_EIMS_GPI_SDP2;
1847                 if (adapter->num_vfs)
1848                         mask |= IXGBE_EIMS_MAILBOX;
1849         }
1850         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
1851             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
1852                 mask |= IXGBE_EIMS_FLOW_DIR;
1853
1854         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1855         ixgbe_irq_enable_queues(adapter, ~0);
1856         IXGBE_WRITE_FLUSH(&adapter->hw);
1857
1858         if (adapter->num_vfs > 32) {
1859                 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1860                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1861         }
1862 }
1863
1864 /**
1865  * ixgbe_intr - legacy mode Interrupt Handler
1866  * @irq: interrupt number
1867  * @data: pointer to a network interface device structure
1868  **/
1869 static irqreturn_t ixgbe_intr(int irq, void *data)
1870 {
1871         struct net_device *netdev = data;
1872         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1873         struct ixgbe_hw *hw = &adapter->hw;
1874         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1875         u32 eicr;
1876
1877         /*
1878          * Workaround for silicon errata.  Mask the interrupts
1879          * before the read of EICR.
1880          */
1881         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1882
1883         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1884          * therefore no explict interrupt disable is necessary */
1885         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1886         if (!eicr) {
1887                 /* shared interrupt alert!
1888                  * make sure interrupts are enabled because the read will
1889                  * have disabled interrupts due to EIAM */
1890                 ixgbe_irq_enable(adapter);
1891                 return IRQ_NONE;        /* Not our interrupt */
1892         }
1893
1894         if (eicr & IXGBE_EICR_LSC)
1895                 ixgbe_check_lsc(adapter);
1896
1897         if (hw->mac.type == ixgbe_mac_82599EB)
1898                 ixgbe_check_sfp_event(adapter, eicr);
1899
1900         ixgbe_check_fan_failure(adapter, eicr);
1901
1902         if (napi_schedule_prep(&(q_vector->napi))) {
1903                 adapter->tx_ring[0]->total_packets = 0;
1904                 adapter->tx_ring[0]->total_bytes = 0;
1905                 adapter->rx_ring[0]->total_packets = 0;
1906                 adapter->rx_ring[0]->total_bytes = 0;
1907                 /* would disable interrupts here but EIAM disabled it */
1908                 __napi_schedule(&(q_vector->napi));
1909         }
1910
1911         return IRQ_HANDLED;
1912 }
1913
1914 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1915 {
1916         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1917
1918         for (i = 0; i < q_vectors; i++) {
1919                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
1920                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1921                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1922                 q_vector->rxr_count = 0;
1923                 q_vector->txr_count = 0;
1924         }
1925 }
1926
1927 /**
1928  * ixgbe_request_irq - initialize interrupts
1929  * @adapter: board private structure
1930  *
1931  * Attempts to configure interrupts using the best available
1932  * capabilities of the hardware and kernel.
1933  **/
1934 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1935 {
1936         struct net_device *netdev = adapter->netdev;
1937         int err;
1938
1939         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1940                 err = ixgbe_request_msix_irqs(adapter);
1941         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1942                 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
1943                                   netdev->name, netdev);
1944         } else {
1945                 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
1946                                   netdev->name, netdev);
1947         }
1948
1949         if (err)
1950                 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1951
1952         return err;
1953 }
1954
1955 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1956 {
1957         struct net_device *netdev = adapter->netdev;
1958
1959         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1960                 int i, q_vectors;
1961
1962                 q_vectors = adapter->num_msix_vectors;
1963
1964                 i = q_vectors - 1;
1965                 free_irq(adapter->msix_entries[i].vector, netdev);
1966
1967                 i--;
1968                 for (; i >= 0; i--) {
1969                         free_irq(adapter->msix_entries[i].vector,
1970                                  adapter->q_vector[i]);
1971                 }
1972
1973                 ixgbe_reset_q_vectors(adapter);
1974         } else {
1975                 free_irq(adapter->pdev->irq, netdev);
1976         }
1977 }
1978
1979 /**
1980  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1981  * @adapter: board private structure
1982  **/
1983 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1984 {
1985         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1986                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1987         } else {
1988                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
1989                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
1990                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
1991                 if (adapter->num_vfs > 32)
1992                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
1993         }
1994         IXGBE_WRITE_FLUSH(&adapter->hw);
1995         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1996                 int i;
1997                 for (i = 0; i < adapter->num_msix_vectors; i++)
1998                         synchronize_irq(adapter->msix_entries[i].vector);
1999         } else {
2000                 synchronize_irq(adapter->pdev->irq);
2001         }
2002 }
2003
2004 /**
2005  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2006  *
2007  **/
2008 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2009 {
2010         struct ixgbe_hw *hw = &adapter->hw;
2011
2012         IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2013                         EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2014
2015         ixgbe_set_ivar(adapter, 0, 0, 0);
2016         ixgbe_set_ivar(adapter, 1, 0, 0);
2017
2018         map_vector_to_rxq(adapter, 0, 0);
2019         map_vector_to_txq(adapter, 0, 0);
2020
2021         DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
2022 }
2023
2024 /**
2025  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2026  * @adapter: board private structure
2027  *
2028  * Configure the Tx unit of the MAC after a reset.
2029  **/
2030 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2031 {
2032         u64 tdba;
2033         struct ixgbe_hw *hw = &adapter->hw;
2034         u32 i, j, tdlen, txctrl;
2035
2036         /* Setup the HW Tx Head and Tail descriptor pointers */
2037         for (i = 0; i < adapter->num_tx_queues; i++) {
2038                 struct ixgbe_ring *ring = adapter->tx_ring[i];
2039                 j = ring->reg_idx;
2040                 tdba = ring->dma;
2041                 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
2042                 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
2043                                 (tdba & DMA_BIT_MASK(32)));
2044                 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
2045                 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
2046                 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
2047                 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
2048                 adapter->tx_ring[i]->head = IXGBE_TDH(j);
2049                 adapter->tx_ring[i]->tail = IXGBE_TDT(j);
2050                 /*
2051                  * Disable Tx Head Writeback RO bit, since this hoses
2052                  * bookkeeping if things aren't delivered in order.
2053                  */
2054                 switch (hw->mac.type) {
2055                 case ixgbe_mac_82598EB:
2056                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
2057                         break;
2058                 case ixgbe_mac_82599EB:
2059                 default:
2060                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(j));
2061                         break;
2062                 }
2063                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
2064                 switch (hw->mac.type) {
2065                 case ixgbe_mac_82598EB:
2066                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
2067                         break;
2068                 case ixgbe_mac_82599EB:
2069                 default:
2070                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(j), txctrl);
2071                         break;
2072                 }
2073         }
2074
2075         if (hw->mac.type == ixgbe_mac_82599EB) {
2076                 u32 rttdcs;
2077                 u32 mask;
2078
2079                 /* disable the arbiter while setting MTQC */
2080                 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2081                 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2082                 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2083
2084                 /* set transmit pool layout */
2085                 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2086                 switch (adapter->flags & mask) {
2087
2088                 case (IXGBE_FLAG_SRIOV_ENABLED):
2089                         IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2090                                         (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2091                         break;
2092
2093                 case (IXGBE_FLAG_DCB_ENABLED):
2094                         /* We enable 8 traffic classes, DCB only */
2095                         IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2096                                       (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2097                         break;
2098
2099                 default:
2100                         IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2101                         break;
2102                 }
2103
2104                 /* re-eable the arbiter */
2105                 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2106                 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2107         }
2108 }
2109
2110 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2111
2112 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2113                                    struct ixgbe_ring *rx_ring)
2114 {
2115         u32 srrctl;
2116         int index;
2117         struct ixgbe_ring_feature *feature = adapter->ring_feature;
2118
2119         index = rx_ring->reg_idx;
2120         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2121                 unsigned long mask;
2122                 mask = (unsigned long) feature[RING_F_RSS].mask;
2123                 index = index & mask;
2124         }
2125         srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2126
2127         srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2128         srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2129
2130         srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2131                   IXGBE_SRRCTL_BSIZEHDR_MASK;
2132
2133         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2134 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2135                 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2136 #else
2137                 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2138 #endif
2139                 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2140         } else {
2141                 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2142                           IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2143                 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2144         }
2145
2146         IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2147 }
2148
2149 static u32 ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2150 {
2151         u32 mrqc = 0;
2152         int mask;
2153
2154         if (!(adapter->hw.mac.type == ixgbe_mac_82599EB))
2155                 return mrqc;
2156
2157         mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2158 #ifdef CONFIG_IXGBE_DCB
2159                                  | IXGBE_FLAG_DCB_ENABLED
2160 #endif
2161                                  | IXGBE_FLAG_SRIOV_ENABLED
2162                                 );
2163
2164         switch (mask) {
2165         case (IXGBE_FLAG_RSS_ENABLED):
2166                 mrqc = IXGBE_MRQC_RSSEN;
2167                 break;
2168         case (IXGBE_FLAG_SRIOV_ENABLED):
2169                 mrqc = IXGBE_MRQC_VMDQEN;
2170                 break;
2171 #ifdef CONFIG_IXGBE_DCB
2172         case (IXGBE_FLAG_DCB_ENABLED):
2173                 mrqc = IXGBE_MRQC_RT8TCEN;
2174                 break;
2175 #endif /* CONFIG_IXGBE_DCB */
2176         default:
2177                 break;
2178         }
2179
2180         return mrqc;
2181 }
2182
2183 /**
2184  * ixgbe_configure_rscctl - enable RSC for the indicated ring
2185  * @adapter:    address of board private structure
2186  * @index:      index of ring to set
2187  **/
2188 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter, int index)
2189 {
2190         struct ixgbe_ring *rx_ring;
2191         struct ixgbe_hw *hw = &adapter->hw;
2192         int j;
2193         u32 rscctrl;
2194         int rx_buf_len;
2195
2196         rx_ring = adapter->rx_ring[index];
2197         j = rx_ring->reg_idx;
2198         rx_buf_len = rx_ring->rx_buf_len;
2199         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2200         rscctrl |= IXGBE_RSCCTL_RSCEN;
2201         /*
2202          * we must limit the number of descriptors so that the
2203          * total size of max desc * buf_len is not greater
2204          * than 65535
2205          */
2206         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2207 #if (MAX_SKB_FRAGS > 16)
2208                 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2209 #elif (MAX_SKB_FRAGS > 8)
2210                 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2211 #elif (MAX_SKB_FRAGS > 4)
2212                 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2213 #else
2214                 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2215 #endif
2216         } else {
2217                 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2218                         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2219                 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2220                         rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2221                 else
2222                         rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2223         }
2224         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2225 }
2226
2227 /**
2228  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2229  * @adapter: board private structure
2230  *
2231  * Configure the Rx unit of the MAC after a reset.
2232  **/
2233 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2234 {
2235         u64 rdba;
2236         struct ixgbe_hw *hw = &adapter->hw;
2237         struct ixgbe_ring *rx_ring;
2238         struct net_device *netdev = adapter->netdev;
2239         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2240         int i, j;
2241         u32 rdlen, rxctrl, rxcsum;
2242         static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2243                           0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2244                           0x6A3E67EA, 0x14364D17, 0x3BED200D};
2245         u32 fctrl, hlreg0;
2246         u32 reta = 0, mrqc = 0;
2247         u32 rdrxctl;
2248         int rx_buf_len;
2249
2250         /* Decide whether to use packet split mode or not */
2251         /* Do not use packet split if we're in SR-IOV Mode */
2252         if (!adapter->num_vfs)
2253                 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2254
2255         /* Set the RX buffer length according to the mode */
2256         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2257                 rx_buf_len = IXGBE_RX_HDR_SIZE;
2258                 if (hw->mac.type == ixgbe_mac_82599EB) {
2259                         /* PSRTYPE must be initialized in 82599 */
2260                         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2261                                       IXGBE_PSRTYPE_UDPHDR |
2262                                       IXGBE_PSRTYPE_IPV4HDR |
2263                                       IXGBE_PSRTYPE_IPV6HDR |
2264                                       IXGBE_PSRTYPE_L2HDR;
2265                         IXGBE_WRITE_REG(hw,
2266                                         IXGBE_PSRTYPE(adapter->num_vfs),
2267                                         psrtype);
2268                 }
2269         } else {
2270                 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2271                     (netdev->mtu <= ETH_DATA_LEN))
2272                         rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2273                 else
2274                         rx_buf_len = ALIGN(max_frame, 1024);
2275         }
2276
2277         fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
2278         fctrl |= IXGBE_FCTRL_BAM;
2279         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
2280         fctrl |= IXGBE_FCTRL_PMCF;
2281         IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
2282
2283         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2284         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2285                 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2286         else
2287                 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2288 #ifdef IXGBE_FCOE
2289         if (netdev->features & NETIF_F_FCOE_MTU)
2290                 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2291 #endif
2292         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2293
2294         rdlen = adapter->rx_ring[0]->count * sizeof(union ixgbe_adv_rx_desc);
2295         /* disable receives while setting up the descriptors */
2296         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2297         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2298
2299         /*
2300          * Setup the HW Rx Head and Tail Descriptor Pointers and
2301          * the Base and Length of the Rx Descriptor Ring
2302          */
2303         for (i = 0; i < adapter->num_rx_queues; i++) {
2304                 rx_ring = adapter->rx_ring[i];
2305                 rdba = rx_ring->dma;
2306                 j = rx_ring->reg_idx;
2307                 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_BIT_MASK(32)));
2308                 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
2309                 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
2310                 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
2311                 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
2312                 rx_ring->head = IXGBE_RDH(j);
2313                 rx_ring->tail = IXGBE_RDT(j);
2314                 rx_ring->rx_buf_len = rx_buf_len;
2315
2316                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2317                         rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2318                 else
2319                         rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2320
2321 #ifdef IXGBE_FCOE
2322                 if (netdev->features & NETIF_F_FCOE_MTU) {
2323                         struct ixgbe_ring_feature *f;
2324                         f = &adapter->ring_feature[RING_F_FCOE];
2325                         if ((i >= f->mask) && (i < f->mask + f->indices)) {
2326                                 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2327                                 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2328                                         rx_ring->rx_buf_len =
2329                                                 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2330                         }
2331                 }
2332
2333 #endif /* IXGBE_FCOE */
2334                 ixgbe_configure_srrctl(adapter, rx_ring);
2335         }
2336
2337         if (hw->mac.type == ixgbe_mac_82598EB) {
2338                 /*
2339                  * For VMDq support of different descriptor types or
2340                  * buffer sizes through the use of multiple SRRCTL
2341                  * registers, RDRXCTL.MVMEN must be set to 1
2342                  *
2343                  * also, the manual doesn't mention it clearly but DCA hints
2344                  * will only use queue 0's tags unless this bit is set.  Side
2345                  * effects of setting this bit are only that SRRCTL must be
2346                  * fully programmed [0..15]
2347                  */
2348                 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2349                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2350                 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2351         }
2352
2353         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2354                 u32 vt_reg_bits;
2355                 u32 reg_offset, vf_shift;
2356                 u32 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2357                 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN
2358                         | IXGBE_VT_CTL_REPLEN;
2359                 vt_reg_bits |= (adapter->num_vfs <<
2360                                 IXGBE_VT_CTL_POOL_SHIFT);
2361                 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2362                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, 0);
2363
2364                 vf_shift = adapter->num_vfs % 32;
2365                 reg_offset = adapter->num_vfs / 32;
2366                 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), 0);
2367                 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), 0);
2368                 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), 0);
2369                 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), 0);
2370                 /* Enable only the PF's pool for Tx/Rx */
2371                 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2372                 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2373                 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2374                 ixgbe_set_vmolr(hw, adapter->num_vfs);
2375         }
2376
2377         /* Program MRQC for the distribution of queues */
2378         mrqc = ixgbe_setup_mrqc(adapter);
2379
2380         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2381                 /* Fill out redirection table */
2382                 for (i = 0, j = 0; i < 128; i++, j++) {
2383                         if (j == adapter->ring_feature[RING_F_RSS].indices)
2384                                 j = 0;
2385                         /* reta = 4-byte sliding window of
2386                          * 0x00..(indices-1)(indices-1)00..etc. */
2387                         reta = (reta << 8) | (j * 0x11);
2388                         if ((i & 3) == 3)
2389                                 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2390                 }
2391
2392                 /* Fill out hash function seeds */
2393                 for (i = 0; i < 10; i++)
2394                         IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2395
2396                 if (hw->mac.type == ixgbe_mac_82598EB)
2397                         mrqc |= IXGBE_MRQC_RSSEN;
2398                     /* Perform hash on these packet types */
2399                 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2400                       | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2401                       | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
2402                       | IXGBE_MRQC_RSS_FIELD_IPV6
2403                       | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
2404                       | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2405         }
2406         IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2407
2408         if (adapter->num_vfs) {
2409                 u32 reg;
2410
2411                 /* Map PF MAC address in RAR Entry 0 to first pool
2412                  * following VFs */
2413                 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2414
2415                 /* Set up VF register offsets for selected VT Mode, i.e.
2416                  * 64 VFs for SR-IOV */
2417                 reg = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2418                 reg |= IXGBE_GCR_EXT_SRIOV;
2419                 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, reg);
2420         }
2421
2422         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2423
2424         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
2425             adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
2426                 /* Disable indicating checksum in descriptor, enables
2427                  * RSS hash */
2428                 rxcsum |= IXGBE_RXCSUM_PCSD;
2429         }
2430         if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
2431                 /* Enable IPv4 payload checksum for UDP fragments
2432                  * if PCSD is not set */
2433                 rxcsum |= IXGBE_RXCSUM_IPPCSE;
2434         }
2435
2436         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2437
2438         if (hw->mac.type == ixgbe_mac_82599EB) {
2439                 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2440                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2441                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2442                 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2443         }
2444
2445         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
2446                 /* Enable 82599 HW-RSC */
2447                 for (i = 0; i < adapter->num_rx_queues; i++)
2448                         ixgbe_configure_rscctl(adapter, i);
2449
2450                 /* Disable RSC for ACK packets */
2451                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2452                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2453         }
2454 }
2455
2456 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2457 {
2458         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2459         struct ixgbe_hw *hw = &adapter->hw;
2460         int pool_ndx = adapter->num_vfs;
2461
2462         /* add VID to filter table */
2463         hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
2464 }
2465
2466 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2467 {
2468         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2469         struct ixgbe_hw *hw = &adapter->hw;
2470         int pool_ndx = adapter->num_vfs;
2471
2472         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2473                 ixgbe_irq_disable(adapter);
2474
2475         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2476
2477         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2478                 ixgbe_irq_enable(adapter);
2479
2480         /* remove VID from filter table */
2481         hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
2482 }
2483
2484 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2485                                    struct vlan_group *grp)
2486 {
2487         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2488         u32 ctrl;
2489         int i, j;
2490
2491         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2492                 ixgbe_irq_disable(adapter);
2493         adapter->vlgrp = grp;
2494
2495         /*
2496          * For a DCB driver, always enable VLAN tag stripping so we can
2497          * still receive traffic from a DCB-enabled host even if we're
2498          * not in DCB mode.
2499          */
2500         ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2501
2502         /* Disable CFI check */
2503         ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2504
2505         /* enable VLAN tag stripping */
2506         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2507                 ctrl |= IXGBE_VLNCTRL_VME;
2508         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2509                 for (i = 0; i < adapter->num_rx_queues; i++) {
2510                         u32 ctrl;
2511                         j = adapter->rx_ring[i]->reg_idx;
2512                         ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_RXDCTL(j));
2513                         ctrl |= IXGBE_RXDCTL_VME;
2514                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_RXDCTL(j), ctrl);
2515                 }
2516         }
2517
2518         IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2519
2520         ixgbe_vlan_rx_add_vid(netdev, 0);
2521
2522         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2523                 ixgbe_irq_enable(adapter);
2524 }
2525
2526 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2527 {
2528         ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2529
2530         if (adapter->vlgrp) {
2531                 u16 vid;
2532                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2533                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2534                                 continue;
2535                         ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2536                 }
2537         }
2538 }
2539
2540 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
2541 {
2542         struct dev_mc_list *mc_ptr;
2543         u8 *addr = *mc_addr_ptr;
2544         *vmdq = 0;
2545
2546         mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
2547         if (mc_ptr->next)
2548                 *mc_addr_ptr = mc_ptr->next->dmi_addr;
2549         else
2550                 *mc_addr_ptr = NULL;
2551
2552         return addr;
2553 }
2554
2555 /**
2556  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
2557  * @netdev: network interface device structure
2558  *
2559  * The set_rx_method entry point is called whenever the unicast/multicast
2560  * address list or the network interface flags are updated.  This routine is
2561  * responsible for configuring the hardware for proper unicast, multicast and
2562  * promiscuous mode.
2563  **/
2564 void ixgbe_set_rx_mode(struct net_device *netdev)
2565 {
2566         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2567         struct ixgbe_hw *hw = &adapter->hw;
2568         u32 fctrl, vlnctrl;
2569         u8 *addr_list = NULL;
2570         int addr_count = 0;
2571
2572         /* Check for Promiscuous and All Multicast modes */
2573
2574         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2575         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2576
2577         if (netdev->flags & IFF_PROMISC) {
2578                 hw->addr_ctrl.user_set_promisc = 1;
2579                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2580                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2581         } else {
2582                 if (netdev->flags & IFF_ALLMULTI) {
2583                         fctrl |= IXGBE_FCTRL_MPE;
2584                         fctrl &= ~IXGBE_FCTRL_UPE;
2585                 } else {
2586                         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2587                 }
2588                 vlnctrl |= IXGBE_VLNCTRL_VFE;
2589                 hw->addr_ctrl.user_set_promisc = 0;
2590         }
2591
2592         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
2593         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2594
2595         /* reprogram secondary unicast list */
2596         hw->mac.ops.update_uc_addr_list(hw, netdev);
2597
2598         /* reprogram multicast list */
2599         addr_count = netdev_mc_count(netdev);
2600         if (addr_count)
2601                 addr_list = netdev->mc_list->dmi_addr;
2602         hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
2603                                         ixgbe_addr_list_itr);
2604         if (adapter->num_vfs)
2605                 ixgbe_restore_vf_multicasts(adapter);
2606 }
2607
2608 static void ixgbe_napi_enable_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                 struct napi_struct *napi;
2620                 q_vector = adapter->q_vector[q_idx];
2621                 napi = &q_vector->napi;
2622                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2623                         if (!q_vector->rxr_count || !q_vector->txr_count) {
2624                                 if (q_vector->txr_count == 1)
2625                                         napi->poll = &ixgbe_clean_txonly;
2626                                 else if (q_vector->rxr_count == 1)
2627                                         napi->poll = &ixgbe_clean_rxonly;
2628                         }
2629                 }
2630
2631                 napi_enable(napi);
2632         }
2633 }
2634
2635 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2636 {
2637         int q_idx;
2638         struct ixgbe_q_vector *q_vector;
2639         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2640
2641         /* legacy and MSI only use one vector */
2642         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2643                 q_vectors = 1;
2644
2645         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2646                 q_vector = adapter->q_vector[q_idx];
2647                 napi_disable(&q_vector->napi);
2648         }
2649 }
2650
2651 #ifdef CONFIG_IXGBE_DCB
2652 /*
2653  * ixgbe_configure_dcb - Configure DCB hardware
2654  * @adapter: ixgbe adapter struct
2655  *
2656  * This is called by the driver on open to configure the DCB hardware.
2657  * This is also called by the gennetlink interface when reconfiguring
2658  * the DCB state.
2659  */
2660 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2661 {
2662         struct ixgbe_hw *hw = &adapter->hw;
2663         u32 txdctl, vlnctrl;
2664         int i, j;
2665
2666         ixgbe_dcb_check_config(&adapter->dcb_cfg);
2667         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2668         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2669
2670         /* reconfigure the hardware */
2671         ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
2672
2673         for (i = 0; i < adapter->num_tx_queues; i++) {
2674                 j = adapter->tx_ring[i]->reg_idx;
2675                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2676                 /* PThresh workaround for Tx hang with DFP enabled. */
2677                 txdctl |= 32;
2678                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2679         }
2680         /* Enable VLAN tag insert/strip */
2681         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2682         if (hw->mac.type == ixgbe_mac_82598EB) {
2683                 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2684                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2685                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2686         } else if (hw->mac.type == ixgbe_mac_82599EB) {
2687                 vlnctrl |= IXGBE_VLNCTRL_VFE;
2688                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2689                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2690                 for (i = 0; i < adapter->num_rx_queues; i++) {
2691                         j = adapter->rx_ring[i]->reg_idx;
2692                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2693                         vlnctrl |= IXGBE_RXDCTL_VME;
2694                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2695                 }
2696         }
2697         hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2698 }
2699
2700 #endif
2701 static void ixgbe_configure(struct ixgbe_adapter *adapter)
2702 {
2703         struct net_device *netdev = adapter->netdev;
2704         struct ixgbe_hw *hw = &adapter->hw;
2705         int i;
2706
2707         ixgbe_set_rx_mode(netdev);
2708
2709         ixgbe_restore_vlan(adapter);
2710 #ifdef CONFIG_IXGBE_DCB
2711         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2712                 if (hw->mac.type == ixgbe_mac_82598EB)
2713                         netif_set_gso_max_size(netdev, 32768);
2714                 else
2715                         netif_set_gso_max_size(netdev, 65536);
2716                 ixgbe_configure_dcb(adapter);
2717         } else {
2718                 netif_set_gso_max_size(netdev, 65536);
2719         }
2720 #else
2721         netif_set_gso_max_size(netdev, 65536);
2722 #endif
2723
2724 #ifdef IXGBE_FCOE
2725         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2726                 ixgbe_configure_fcoe(adapter);
2727
2728 #endif /* IXGBE_FCOE */
2729         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2730                 for (i = 0; i < adapter->num_tx_queues; i++)
2731                         adapter->tx_ring[i]->atr_sample_rate =
2732                                                        adapter->atr_sample_rate;
2733                 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
2734         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
2735                 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
2736         }
2737
2738         ixgbe_configure_tx(adapter);
2739         ixgbe_configure_rx(adapter);
2740         for (i = 0; i < adapter->num_rx_queues; i++)
2741                 ixgbe_alloc_rx_buffers(adapter, adapter->rx_ring[i],
2742                                        (adapter->rx_ring[i]->count - 1));
2743 }
2744
2745 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2746 {
2747         switch (hw->phy.type) {
2748         case ixgbe_phy_sfp_avago:
2749         case ixgbe_phy_sfp_ftl:
2750         case ixgbe_phy_sfp_intel:
2751         case ixgbe_phy_sfp_unknown:
2752         case ixgbe_phy_tw_tyco:
2753         case ixgbe_phy_tw_unknown:
2754                 return true;
2755         default:
2756                 return false;
2757         }
2758 }
2759
2760 /**
2761  * ixgbe_sfp_link_config - set up SFP+ link
2762  * @adapter: pointer to private adapter struct
2763  **/
2764 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2765 {
2766         struct ixgbe_hw *hw = &adapter->hw;
2767
2768                 if (hw->phy.multispeed_fiber) {
2769                         /*
2770                          * In multispeed fiber setups, the device may not have
2771                          * had a physical connection when the driver loaded.
2772                          * If that's the case, the initial link configuration
2773                          * couldn't get the MAC into 10G or 1G mode, so we'll
2774                          * never have a link status change interrupt fire.
2775                          * We need to try and force an autonegotiation
2776                          * session, then bring up link.
2777                          */
2778                         hw->mac.ops.setup_sfp(hw);
2779                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2780                                 schedule_work(&adapter->multispeed_fiber_task);
2781                 } else {
2782                         /*
2783                          * Direct Attach Cu and non-multispeed fiber modules
2784                          * still need to be configured properly prior to
2785                          * attempting link.
2786                          */
2787                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2788                                 schedule_work(&adapter->sfp_config_module_task);
2789                 }
2790 }
2791
2792 /**
2793  * ixgbe_non_sfp_link_config - set up non-SFP+ link
2794  * @hw: pointer to private hardware struct
2795  *
2796  * Returns 0 on success, negative on failure
2797  **/
2798 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2799 {
2800         u32 autoneg;
2801         bool negotiation, link_up = false;
2802         u32 ret = IXGBE_ERR_LINK_SETUP;
2803
2804         if (hw->mac.ops.check_link)
2805                 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2806
2807         if (ret)
2808                 goto link_cfg_out;
2809
2810         if (hw->mac.ops.get_link_capabilities)
2811                 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
2812         if (ret)
2813                 goto link_cfg_out;
2814
2815         if (hw->mac.ops.setup_link)
2816                 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
2817 link_cfg_out:
2818         return ret;
2819 }
2820
2821 #define IXGBE_MAX_RX_DESC_POLL 10
2822 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2823                                               int rxr)
2824 {
2825         int j = adapter->rx_ring[rxr]->reg_idx;
2826         int k;
2827
2828         for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2829                 if (IXGBE_READ_REG(&adapter->hw,
2830                                    IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2831                         break;
2832                 else
2833                         msleep(1);
2834         }
2835         if (k >= IXGBE_MAX_RX_DESC_POLL) {
2836                 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2837                         "not set within the polling period\n", rxr);
2838         }
2839         ixgbe_release_rx_desc(&adapter->hw, adapter->rx_ring[rxr],
2840                               (adapter->rx_ring[rxr]->count - 1));
2841 }
2842
2843 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2844 {
2845         struct net_device *netdev = adapter->netdev;
2846         struct ixgbe_hw *hw = &adapter->hw;
2847         int i, j = 0;
2848         int num_rx_rings = adapter->num_rx_queues;
2849         int err;
2850         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2851         u32 txdctl, rxdctl, mhadd;
2852         u32 dmatxctl;
2853         u32 gpie;
2854         u32 ctrl_ext;
2855
2856         ixgbe_get_hw_control(adapter);
2857
2858         if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2859             (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2860                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2861                         gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2862                                 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2863                 } else {
2864                         /* MSI only */
2865                         gpie = 0;
2866                 }
2867                 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2868                         gpie &= ~IXGBE_GPIE_VTMODE_MASK;
2869                         gpie |= IXGBE_GPIE_VTMODE_64;
2870                 }
2871                 /* XXX: to interrupt immediately for EICS writes, enable this */
2872                 /* gpie |= IXGBE_GPIE_EIMEN; */
2873                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2874         }
2875
2876         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2877                 /*
2878                  * use EIAM to auto-mask when MSI-X interrupt is asserted
2879                  * this saves a register write for every interrupt
2880                  */
2881                 switch (hw->mac.type) {
2882                 case ixgbe_mac_82598EB:
2883                         IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2884                         break;
2885                 default:
2886                 case ixgbe_mac_82599EB:
2887                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
2888                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
2889                         break;
2890                 }
2891         } else {
2892                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2893                  * specifically only auto mask tx and rx interrupts */
2894                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2895         }
2896
2897         /* Enable fan failure interrupt if media type is copper */
2898         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2899                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2900                 gpie |= IXGBE_SDP1_GPIEN;
2901                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2902         }
2903
2904         if (hw->mac.type == ixgbe_mac_82599EB) {
2905                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2906                 gpie |= IXGBE_SDP1_GPIEN;
2907                 gpie |= IXGBE_SDP2_GPIEN;
2908                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2909         }
2910
2911 #ifdef IXGBE_FCOE
2912         /* adjust max frame to be able to do baby jumbo for FCoE */
2913         if ((netdev->features & NETIF_F_FCOE_MTU) &&
2914             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2915                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2916
2917 #endif /* IXGBE_FCOE */
2918         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2919         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2920                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2921                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2922
2923                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2924         }
2925
2926         for (i = 0; i < adapter->num_tx_queues; i++) {
2927                 j = adapter->tx_ring[i]->reg_idx;
2928                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2929                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2930                 txdctl |= (8 << 16);
2931                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2932         }
2933
2934         if (hw->mac.type == ixgbe_mac_82599EB) {
2935                 /* DMATXCTL.EN must be set after all Tx queue config is done */
2936                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2937                 dmatxctl |= IXGBE_DMATXCTL_TE;
2938                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2939         }
2940         for (i = 0; i < adapter->num_tx_queues; i++) {
2941                 j = adapter->tx_ring[i]->reg_idx;
2942                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2943                 txdctl |= IXGBE_TXDCTL_ENABLE;
2944                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2945                 if (hw->mac.type == ixgbe_mac_82599EB) {
2946                         int wait_loop = 10;
2947                         /* poll for Tx Enable ready */
2948                         do {
2949                                 msleep(1);
2950                                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2951                         } while (--wait_loop &&
2952                                  !(txdctl & IXGBE_TXDCTL_ENABLE));
2953                         if (!wait_loop)
2954                                 DPRINTK(DRV, ERR, "Could not enable "
2955                                         "Tx Queue %d\n", j);
2956                 }
2957         }
2958
2959         for (i = 0; i < num_rx_rings; i++) {
2960                 j = adapter->rx_ring[i]->reg_idx;
2961                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2962                 /* enable PTHRESH=32 descriptors (half the internal cache)
2963                  * and HTHRESH=0 descriptors (to minimize latency on fetch),
2964                  * this also removes a pesky rx_no_buffer_count increment */
2965                 rxdctl |= 0x0020;
2966                 rxdctl |= IXGBE_RXDCTL_ENABLE;
2967                 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2968                 if (hw->mac.type == ixgbe_mac_82599EB)
2969                         ixgbe_rx_desc_queue_enable(adapter, i);
2970         }
2971         /* enable all receives */
2972         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2973         if (hw->mac.type == ixgbe_mac_82598EB)
2974                 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2975         else
2976                 rxdctl |= IXGBE_RXCTRL_RXEN;
2977         hw->mac.ops.enable_rx_dma(hw, rxdctl);
2978
2979         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2980                 ixgbe_configure_msix(adapter);
2981         else
2982                 ixgbe_configure_msi_and_legacy(adapter);
2983
2984         clear_bit(__IXGBE_DOWN, &adapter->state);
2985         ixgbe_napi_enable_all(adapter);
2986
2987         /* clear any pending interrupts, may auto mask */
2988         IXGBE_READ_REG(hw, IXGBE_EICR);
2989
2990         ixgbe_irq_enable(adapter);
2991
2992         /*
2993          * If this adapter has a fan, check to see if we had a failure
2994          * before we enabled the interrupt.
2995          */
2996         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2997                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
2998                 if (esdp & IXGBE_ESDP_SDP1)
2999                         DPRINTK(DRV, CRIT,
3000                                 "Fan has stopped, replace the adapter\n");
3001         }
3002
3003         /*
3004          * For hot-pluggable SFP+ devices, a new SFP+ module may have
3005          * arrived before interrupts were enabled but after probe.  Such
3006          * devices wouldn't have their type identified yet. We need to
3007          * kick off the SFP+ module setup first, then try to bring up link.
3008          * If we're not hot-pluggable SFP+, we just need to configure link
3009          * and bring it up.
3010          */
3011         if (hw->phy.type == ixgbe_phy_unknown) {
3012                 err = hw->phy.ops.identify(hw);
3013                 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3014                         /*
3015                          * Take the device down and schedule the sfp tasklet
3016                          * which will unregister_netdev and log it.
3017                          */
3018                         ixgbe_down(adapter);
3019                         schedule_work(&adapter->sfp_config_module_task);
3020                         return err;
3021                 }
3022         }
3023
3024         if (ixgbe_is_sfp(hw)) {
3025                 ixgbe_sfp_link_config(adapter);
3026         } else {
3027                 err = ixgbe_non_sfp_link_config(hw);
3028                 if (err)
3029                         DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
3030         }
3031
3032         for (i = 0; i < adapter->num_tx_queues; i++)
3033                 set_bit(__IXGBE_FDIR_INIT_DONE,
3034                         &(adapter->tx_ring[i]->reinit_state));
3035
3036         /* enable transmits */
3037         netif_tx_start_all_queues(netdev);
3038
3039         /* bring the link up in the watchdog, this could race with our first
3040          * link up interrupt but shouldn't be a problem */
3041         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3042         adapter->link_check_timeout = jiffies;
3043         mod_timer(&adapter->watchdog_timer, jiffies);
3044
3045         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3046         ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3047         ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3048         IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3049
3050         return 0;
3051 }
3052
3053 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3054 {
3055         WARN_ON(in_interrupt());
3056         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3057                 msleep(1);
3058         ixgbe_down(adapter);
3059         ixgbe_up(adapter);
3060         clear_bit(__IXGBE_RESETTING, &adapter->state);
3061 }
3062
3063 int ixgbe_up(struct ixgbe_adapter *adapter)
3064 {
3065         /* hardware has been reset, we need to reload some things */
3066         ixgbe_configure(adapter);
3067
3068         return ixgbe_up_complete(adapter);
3069 }
3070
3071 void ixgbe_reset(struct ixgbe_adapter *adapter)
3072 {
3073         struct ixgbe_hw *hw = &adapter->hw;
3074         int err;
3075
3076         err = hw->mac.ops.init_hw(hw);
3077         switch (err) {
3078         case 0:
3079         case IXGBE_ERR_SFP_NOT_PRESENT:
3080                 break;
3081         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3082                 dev_err(&adapter->pdev->dev, "master disable timed out\n");
3083                 break;
3084         case IXGBE_ERR_EEPROM_VERSION:
3085                 /* We are running on a pre-production device, log a warning */
3086                 dev_warn(&adapter->pdev->dev, "This device is a pre-production "
3087                          "adapter/LOM.  Please be aware there may be issues "
3088                          "associated with your hardware.  If you are "
3089                          "experiencing problems please contact your Intel or "
3090                          "hardware representative who provided you with this "
3091                          "hardware.\n");
3092                 break;
3093         default:
3094                 dev_err(&adapter->pdev->dev, "Hardware Error: %d\n", err);
3095         }
3096
3097         /* reprogram the RAR[0] in case user changed it. */
3098         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3099                             IXGBE_RAH_AV);
3100 }
3101
3102 /**
3103  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3104  * @adapter: board private structure
3105  * @rx_ring: ring to free buffers from
3106  **/
3107 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
3108                                 struct ixgbe_ring *rx_ring)
3109 {
3110         struct pci_dev *pdev = adapter->pdev;
3111         unsigned long size;
3112         unsigned int i;
3113
3114         /* Free all the Rx ring sk_buffs */
3115
3116         for (i = 0; i < rx_ring->count; i++) {
3117                 struct ixgbe_rx_buffer *rx_buffer_info;
3118
3119                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3120                 if (rx_buffer_info->dma) {
3121                         pci_unmap_single(pdev, rx_buffer_info->dma,
3122                                          rx_ring->rx_buf_len,
3123                                          PCI_DMA_FROMDEVICE);
3124                         rx_buffer_info->dma = 0;
3125                 }
3126                 if (rx_buffer_info->skb) {
3127                         struct sk_buff *skb = rx_buffer_info->skb;
3128                         rx_buffer_info->skb = NULL;
3129                         do {
3130                                 struct sk_buff *this = skb;
3131                                 if (IXGBE_RSC_CB(this)->dma) {
3132                                         pci_unmap_single(pdev, IXGBE_RSC_CB(this)->dma,
3133                                                          rx_ring->rx_buf_len,
3134                                                          PCI_DMA_FROMDEVICE);
3135                                         IXGBE_RSC_CB(this)->dma = 0;
3136                                 }
3137                                 skb = skb->prev;
3138                                 dev_kfree_skb(this);
3139                         } while (skb);
3140                 }
3141                 if (!rx_buffer_info->page)
3142                         continue;
3143                 if (rx_buffer_info->page_dma) {
3144                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
3145                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
3146                         rx_buffer_info->page_dma = 0;
3147                 }
3148                 put_page(rx_buffer_info->page);
3149                 rx_buffer_info->page = NULL;
3150                 rx_buffer_info->page_offset = 0;
3151         }
3152
3153         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3154         memset(rx_ring->rx_buffer_info, 0, size);
3155
3156         /* Zero out the descriptor ring */
3157         memset(rx_ring->desc, 0, rx_ring->size);
3158
3159         rx_ring->next_to_clean = 0;
3160         rx_ring->next_to_use = 0;
3161
3162         if (rx_ring->head)
3163                 writel(0, adapter->hw.hw_addr + rx_ring->head);
3164         if (rx_ring->tail)
3165                 writel(0, adapter->hw.hw_addr + rx_ring->tail);
3166 }
3167
3168 /**
3169  * ixgbe_clean_tx_ring - Free Tx Buffers
3170  * @adapter: board private structure
3171  * @tx_ring: ring to be cleaned
3172  **/
3173 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
3174                                 struct ixgbe_ring *tx_ring)
3175 {
3176         struct ixgbe_tx_buffer *tx_buffer_info;
3177         unsigned long size;
3178         unsigned int i;
3179
3180         /* Free all the Tx ring sk_buffs */
3181
3182         for (i = 0; i < tx_ring->count; i++) {
3183                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3184                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3185         }
3186
3187         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3188         memset(tx_ring->tx_buffer_info, 0, size);
3189
3190         /* Zero out the descriptor ring */
3191         memset(tx_ring->desc, 0, tx_ring->size);
3192
3193         tx_ring->next_to_use = 0;
3194         tx_ring->next_to_clean = 0;
3195
3196         if (tx_ring->head)
3197                 writel(0, adapter->hw.hw_addr + tx_ring->head);
3198         if (tx_ring->tail)
3199                 writel(0, adapter->hw.hw_addr + tx_ring->tail);
3200 }
3201
3202 /**
3203  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3204  * @adapter: board private structure
3205  **/
3206 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3207 {
3208         int i;
3209
3210         for (i = 0; i < adapter->num_rx_queues; i++)
3211                 ixgbe_clean_rx_ring(adapter, adapter->rx_ring[i]);
3212 }
3213
3214 /**
3215  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3216  * @adapter: board private structure
3217  **/
3218 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3219 {
3220         int i;
3221
3222         for (i = 0; i < adapter->num_tx_queues; i++)
3223                 ixgbe_clean_tx_ring(adapter, adapter->tx_ring[i]);
3224 }
3225
3226 void ixgbe_down(struct ixgbe_adapter *adapter)
3227 {
3228         struct net_device *netdev = adapter->netdev;
3229         struct ixgbe_hw *hw = &adapter->hw;
3230         u32 rxctrl;
3231         u32 txdctl;
3232         int i, j;
3233
3234         /* signal that we are down to the interrupt handler */
3235         set_bit(__IXGBE_DOWN, &adapter->state);
3236
3237         /* disable receive for all VFs and wait one second */
3238         if (adapter->num_vfs) {
3239                 for (i = 0 ; i < adapter->num_vfs; i++)
3240                         adapter->vfinfo[i].clear_to_send = 0;
3241
3242                 /* ping all the active vfs to let them know we are going down */
3243                 ixgbe_ping_all_vfs(adapter);
3244                 /* Disable all VFTE/VFRE TX/RX */
3245                 ixgbe_disable_tx_rx(adapter);
3246         }
3247
3248         /* disable receives */
3249         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3250         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3251
3252         netif_tx_disable(netdev);
3253
3254         IXGBE_WRITE_FLUSH(hw);
3255         msleep(10);
3256
3257         netif_tx_stop_all_queues(netdev);
3258
3259         ixgbe_irq_disable(adapter);
3260
3261         ixgbe_napi_disable_all(adapter);
3262
3263         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3264         del_timer_sync(&adapter->sfp_timer);
3265         del_timer_sync(&adapter->watchdog_timer);
3266         cancel_work_sync(&adapter->watchdog_task);
3267
3268         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3269             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3270                 cancel_work_sync(&adapter->fdir_reinit_task);
3271
3272         /* disable transmits in the hardware now that interrupts are off */
3273         for (i = 0; i < adapter->num_tx_queues; i++) {
3274                 j = adapter->tx_ring[i]->reg_idx;
3275                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3276                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3277                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3278         }
3279         /* Disable the Tx DMA engine on 82599 */
3280         if (hw->mac.type == ixgbe_mac_82599EB)
3281                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3282                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3283                                  ~IXGBE_DMATXCTL_TE));
3284
3285         netif_carrier_off(netdev);
3286
3287         /* clear n-tuple filters that are cached */
3288         ethtool_ntuple_flush(netdev);
3289
3290         if (!pci_channel_offline(adapter->pdev))
3291                 ixgbe_reset(adapter);
3292         ixgbe_clean_all_tx_rings(adapter);
3293         ixgbe_clean_all_rx_rings(adapter);
3294
3295 #ifdef CONFIG_IXGBE_DCA
3296         /* since we reset the hardware DCA settings were cleared */
3297         ixgbe_setup_dca(adapter);
3298 #endif
3299 }
3300
3301 /**
3302  * ixgbe_poll - NAPI Rx polling callback
3303  * @napi: structure for representing this polling device
3304  * @budget: how many packets driver is allowed to clean
3305  *
3306  * This function is used for legacy and MSI, NAPI mode
3307  **/
3308 static int ixgbe_poll(struct napi_struct *napi, int budget)
3309 {
3310         struct ixgbe_q_vector *q_vector =
3311                                 container_of(napi, struct ixgbe_q_vector, napi);
3312         struct ixgbe_adapter *adapter = q_vector->adapter;
3313         int tx_clean_complete, work_done = 0;
3314
3315 #ifdef CONFIG_IXGBE_DCA
3316         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3317                 ixgbe_update_tx_dca(adapter, adapter->tx_ring[0]);
3318                 ixgbe_update_rx_dca(adapter, adapter->rx_ring[0]);
3319         }
3320 #endif
3321
3322         tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3323         ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3324
3325         if (!tx_clean_complete)
3326                 work_done = budget;
3327
3328         /* If budget not fully consumed, exit the polling mode */
3329         if (work_done < budget) {
3330                 napi_complete(napi);
3331                 if (adapter->rx_itr_setting & 1)
3332                         ixgbe_set_itr(adapter);
3333                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3334                         ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3335         }
3336         return work_done;
3337 }
3338
3339 /**
3340  * ixgbe_tx_timeout - Respond to a Tx Hang
3341  * @netdev: network interface device structure
3342  **/
3343 static void ixgbe_tx_timeout(struct net_device *netdev)
3344 {
3345         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3346
3347         /* Do the reset outside of interrupt context */
3348         schedule_work(&adapter->reset_task);
3349 }
3350
3351 static void ixgbe_reset_task(struct work_struct *work)
3352 {
3353         struct ixgbe_adapter *adapter;
3354         adapter = container_of(work, struct ixgbe_adapter, reset_task);
3355
3356         /* If we're already down or resetting, just bail */
3357         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3358             test_bit(__IXGBE_RESETTING, &adapter->state))
3359                 return;
3360
3361         adapter->tx_timeout_count++;
3362
3363         ixgbe_reinit_locked(adapter);
3364 }
3365
3366 #ifdef CONFIG_IXGBE_DCB
3367 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3368 {
3369         bool ret = false;
3370         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3371
3372         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3373                 return ret;
3374
3375         f->mask = 0x7 << 3;
3376         adapter->num_rx_queues = f->indices;
3377         adapter->num_tx_queues = f->indices;
3378         ret = true;
3379
3380         return ret;
3381 }
3382 #endif
3383
3384 /**
3385  * ixgbe_set_rss_queues: Allocate queues for RSS
3386  * @adapter: board private structure to initialize
3387  *
3388  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
3389  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3390  *
3391  **/
3392 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3393 {
3394         bool ret = false;
3395         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3396
3397         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3398                 f->mask = 0xF;
3399                 adapter->num_rx_queues = f->indices;
3400                 adapter->num_tx_queues = f->indices;
3401                 ret = true;
3402         } else {
3403                 ret = false;
3404         }
3405
3406         return ret;
3407 }
3408
3409 /**
3410  * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3411  * @adapter: board private structure to initialize
3412  *
3413  * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3414  * to the original CPU that initiated the Tx session.  This runs in addition
3415  * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3416  * Rx load across CPUs using RSS.
3417  *
3418  **/
3419 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3420 {
3421         bool ret = false;
3422         struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3423
3424         f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3425         f_fdir->mask = 0;
3426
3427         /* Flow Director must have RSS enabled */
3428         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3429             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3430              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3431                 adapter->num_tx_queues = f_fdir->indices;
3432                 adapter->num_rx_queues = f_fdir->indices;
3433                 ret = true;
3434         } else {
3435                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3436                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3437         }
3438         return ret;
3439 }
3440
3441 #ifdef IXGBE_FCOE
3442 /**
3443  * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3444  * @adapter: board private structure to initialize
3445  *
3446  * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3447  * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3448  * rx queues out of the max number of rx queues, instead, it is used as the
3449  * index of the first rx queue used by FCoE.
3450  *
3451  **/
3452 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3453 {
3454         bool ret = false;
3455         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3456
3457         f->indices = min((int)num_online_cpus(), f->indices);
3458         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3459                 adapter->num_rx_queues = 1;
3460                 adapter->num_tx_queues = 1;
3461 #ifdef CONFIG_IXGBE_DCB
3462                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3463                         DPRINTK(PROBE, INFO, "FCoE enabled with DCB \n");
3464                         ixgbe_set_dcb_queues(adapter);
3465                 }
3466 #endif
3467                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3468                         DPRINTK(PROBE, INFO, "FCoE enabled with RSS \n");
3469                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3470                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3471                                 ixgbe_set_fdir_queues(adapter);
3472                         else
3473                                 ixgbe_set_rss_queues(adapter);
3474                 }
3475                 /* adding FCoE rx rings to the end */
3476                 f->mask = adapter->num_rx_queues;
3477                 adapter->num_rx_queues += f->indices;
3478                 adapter->num_tx_queues += f->indices;
3479
3480                 ret = true;
3481         }
3482
3483         return ret;
3484 }
3485
3486 #endif /* IXGBE_FCOE */
3487 /**
3488  * ixgbe_set_sriov_queues: Allocate queues for IOV use
3489  * @adapter: board private structure to initialize
3490  *
3491  * IOV doesn't actually use anything, so just NAK the
3492  * request for now and let the other queue routines
3493  * figure out what to do.
3494  */
3495 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
3496 {
3497         return false;
3498 }
3499
3500 /*
3501  * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3502  * @adapter: board private structure to initialize
3503  *
3504  * This is the top level queue allocation routine.  The order here is very
3505  * important, starting with the "most" number of features turned on at once,
3506  * and ending with the smallest set of features.  This way large combinations
3507  * can be allocated if they're turned on, and smaller combinations are the
3508  * fallthrough conditions.
3509  *
3510  **/
3511 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
3512 {
3513         /* Start with base case */
3514         adapter->num_rx_queues = 1;
3515         adapter->num_tx_queues = 1;
3516         adapter->num_rx_pools = adapter->num_rx_queues;
3517         adapter->num_rx_queues_per_pool = 1;
3518
3519         if (ixgbe_set_sriov_queues(adapter))
3520                 return;
3521
3522 #ifdef IXGBE_FCOE
3523         if (ixgbe_set_fcoe_queues(adapter))
3524                 goto done;
3525
3526 #endif /* IXGBE_FCOE */
3527 #ifdef CONFIG_IXGBE_DCB
3528         if (ixgbe_set_dcb_queues(adapter))
3529                 goto done;
3530
3531 #endif
3532         if (ixgbe_set_fdir_queues(adapter))
3533                 goto done;
3534
3535         if (ixgbe_set_rss_queues(adapter))
3536                 goto done;
3537
3538         /* fallback to base case */
3539         adapter->num_rx_queues = 1;
3540         adapter->num_tx_queues = 1;
3541
3542 done:
3543         /* Notify the stack of the (possibly) reduced Tx Queue count. */
3544         adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
3545 }
3546
3547 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
3548                                        int vectors)
3549 {
3550         int err, vector_threshold;
3551
3552         /* We'll want at least 3 (vector_threshold):
3553          * 1) TxQ[0] Cleanup
3554          * 2) RxQ[0] Cleanup
3555          * 3) Other (Link Status Change, etc.)
3556          * 4) TCP Timer (optional)
3557          */
3558         vector_threshold = MIN_MSIX_COUNT;
3559
3560         /* The more we get, the more we will assign to Tx/Rx Cleanup
3561          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
3562          * Right now, we simply care about how many we'll get; we'll
3563          * set them up later while requesting irq's.
3564          */
3565         while (vectors >= vector_threshold) {
3566                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
3567                                       vectors);
3568                 if (!err) /* Success in acquiring all requested vectors. */
3569                         break;
3570                 else if (err < 0)
3571                         vectors = 0; /* Nasty failure, quit now */
3572                 else /* err == number of vectors we should try again with */
3573                         vectors = err;
3574         }
3575
3576         if (vectors < vector_threshold) {
3577                 /* Can't allocate enough MSI-X interrupts?  Oh well.
3578                  * This just means we'll go with either a single MSI
3579                  * vector or fall back to legacy interrupts.
3580                  */
3581                 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
3582                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3583                 kfree(adapter->msix_entries);
3584                 adapter->msix_entries = NULL;
3585         } else {
3586                 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
3587                 /*
3588                  * Adjust for only the vectors we'll use, which is minimum
3589                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
3590                  * vectors we were allocated.
3591                  */
3592                 adapter->num_msix_vectors = min(vectors,
3593                                    adapter->max_msix_q_vectors + NON_Q_VECTORS);
3594         }
3595 }
3596
3597 /**
3598  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
3599  * @adapter: board private structure to initialize
3600  *
3601  * Cache the descriptor ring offsets for RSS to the assigned rings.
3602  *
3603  **/
3604 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
3605 {
3606         int i;
3607         bool ret = false;
3608
3609         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3610                 for (i = 0; i < adapter->num_rx_queues; i++)
3611                         adapter->rx_ring[i]->reg_idx = i;
3612                 for (i = 0; i < adapter->num_tx_queues; i++)
3613                         adapter->tx_ring[i]->reg_idx = i;
3614                 ret = true;
3615         } else {
3616                 ret = false;
3617         }
3618
3619         return ret;
3620 }
3621
3622 #ifdef CONFIG_IXGBE_DCB
3623 /**
3624  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
3625  * @adapter: board private structure to initialize
3626  *
3627  * Cache the descriptor ring offsets for DCB to the assigned rings.
3628  *
3629  **/
3630 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
3631 {
3632         int i;
3633         bool ret = false;
3634         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
3635
3636         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3637                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3638                         /* the number of queues is assumed to be symmetric */
3639                         for (i = 0; i < dcb_i; i++) {
3640                                 adapter->rx_ring[i]->reg_idx = i << 3;
3641                                 adapter->tx_ring[i]->reg_idx = i << 2;
3642                         }
3643                         ret = true;
3644                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
3645                         if (dcb_i == 8) {
3646                                 /*
3647                                  * Tx TC0 starts at: descriptor queue 0
3648                                  * Tx TC1 starts at: descriptor queue 32
3649                                  * Tx TC2 starts at: descriptor queue 64
3650                                  * Tx TC3 starts at: descriptor queue 80
3651                                  * Tx TC4 starts at: descriptor queue 96
3652                                  * Tx TC5 starts at: descriptor queue 104
3653                                  * Tx TC6 starts at: descriptor queue 112
3654                                  * Tx TC7 starts at: descriptor queue 120
3655                                  *
3656                                  * Rx TC0-TC7 are offset by 16 queues each
3657                                  */
3658                                 for (i = 0; i < 3; i++) {
3659                                         adapter->tx_ring[i]->reg_idx = i << 5;
3660                                         adapter->rx_ring[i]->reg_idx = i << 4;
3661                                 }
3662                                 for ( ; i < 5; i++) {
3663                                         adapter->tx_ring[i]->reg_idx =
3664                                                                  ((i + 2) << 4);
3665                                         adapter->rx_ring[i]->reg_idx = i << 4;
3666                                 }
3667                                 for ( ; i < dcb_i; i++) {
3668                                         adapter->tx_ring[i]->reg_idx =
3669                                                                  ((i + 8) << 3);
3670                                         adapter->rx_ring[i]->reg_idx = i << 4;
3671                                 }
3672
3673                                 ret = true;
3674                         } else if (dcb_i == 4) {
3675                                 /*
3676                                  * Tx TC0 starts at: descriptor queue 0
3677                                  * Tx TC1 starts at: descriptor queue 64
3678                                  * Tx TC2 starts at: descriptor queue 96
3679                                  * Tx TC3 starts at: descriptor queue 112
3680                                  *
3681                                  * Rx TC0-TC3 are offset by 32 queues each
3682                                  */
3683                                 adapter->tx_ring[0]->reg_idx = 0;
3684                                 adapter->tx_ring[1]->reg_idx = 64;
3685                                 adapter->tx_ring[2]->reg_idx = 96;
3686                                 adapter->tx_ring[3]->reg_idx = 112;
3687                                 for (i = 0 ; i < dcb_i; i++)
3688                                         adapter->rx_ring[i]->reg_idx = i << 5;
3689
3690                                 ret = true;
3691                         } else {
3692                                 ret = false;
3693                         }
3694                 } else {
3695                         ret = false;
3696                 }
3697         } else {
3698                 ret = false;
3699         }
3700
3701         return ret;
3702 }
3703 #endif
3704
3705 /**
3706  * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
3707  * @adapter: board private structure to initialize
3708  *
3709  * Cache the descriptor ring offsets for Flow Director to the assigned rings.
3710  *
3711  **/
3712 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
3713 {
3714         int i;
3715         bool ret = false;
3716
3717         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3718             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3719              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
3720                 for (i = 0; i < adapter->num_rx_queues; i++)
3721                         adapter->rx_ring[i]->reg_idx = i;
3722                 for (i = 0; i < adapter->num_tx_queues; i++)
3723                         adapter->tx_ring[i]->reg_idx = i;
3724                 ret = true;
3725         }
3726
3727         return ret;
3728 }
3729
3730 #ifdef IXGBE_FCOE
3731 /**
3732  * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
3733  * @adapter: board private structure to initialize
3734  *
3735  * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
3736  *
3737  */
3738 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
3739 {
3740         int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
3741         bool ret = false;
3742         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3743
3744         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3745 #ifdef CONFIG_IXGBE_DCB
3746                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3747                         struct ixgbe_fcoe *fcoe = &adapter->fcoe;
3748
3749                         ixgbe_cache_ring_dcb(adapter);
3750                         /* find out queues in TC for FCoE */
3751                         fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
3752                         fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
3753                         /*
3754                          * In 82599, the number of Tx queues for each traffic
3755                          * class for both 8-TC and 4-TC modes are:
3756                          * TCs  : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
3757                          * 8 TCs:  32  32  16  16   8   8   8   8
3758                          * 4 TCs:  64  64  32  32
3759                          * We have max 8 queues for FCoE, where 8 the is
3760                          * FCoE redirection table size. If TC for FCoE is
3761                          * less than or equal to TC3, we have enough queues
3762                          * to add max of 8 queues for FCoE, so we start FCoE
3763                          * tx descriptor from the next one, i.e., reg_idx + 1.
3764                          * If TC for FCoE is above TC3, implying 8 TC mode,
3765                          * and we need 8 for FCoE, we have to take all queues
3766                          * in that traffic class for FCoE.
3767                          */
3768                         if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
3769                                 fcoe_tx_i--;
3770                 }
3771 #endif /* CONFIG_IXGBE_DCB */
3772                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3773                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3774                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3775                                 ixgbe_cache_ring_fdir(adapter);
3776                         else
3777                                 ixgbe_cache_ring_rss(adapter);
3778
3779                         fcoe_rx_i = f->mask;
3780                         fcoe_tx_i = f->mask;
3781                 }
3782                 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
3783                         adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
3784                         adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
3785                 }
3786                 ret = true;
3787         }
3788         return ret;
3789 }
3790
3791 #endif /* IXGBE_FCOE */
3792 /**
3793  * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
3794  * @adapter: board private structure to initialize
3795  *
3796  * SR-IOV doesn't use any descriptor rings but changes the default if
3797  * no other mapping is used.
3798  *
3799  */
3800 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
3801 {
3802         adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
3803         adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
3804         if (adapter->num_vfs)
3805                 return true;
3806         else
3807                 return false;
3808 }
3809
3810 /**
3811  * ixgbe_cache_ring_register - Descriptor ring to register mapping
3812  * @adapter: board private structure to initialize
3813  *
3814  * Once we know the feature-set enabled for the device, we'll cache
3815  * the register offset the descriptor ring is assigned to.
3816  *
3817  * Note, the order the various feature calls is important.  It must start with
3818  * the "most" features enabled at the same time, then trickle down to the
3819  * least amount of features turned on at once.
3820  **/
3821 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
3822 {
3823         /* start with default case */
3824         adapter->rx_ring[0]->reg_idx = 0;
3825         adapter->tx_ring[0]->reg_idx = 0;
3826
3827         if (ixgbe_cache_ring_sriov(adapter))
3828                 return;
3829
3830 #ifdef IXGBE_FCOE
3831         if (ixgbe_cache_ring_fcoe(adapter))
3832                 return;
3833
3834 #endif /* IXGBE_FCOE */
3835 #ifdef CONFIG_IXGBE_DCB
3836         if (ixgbe_cache_ring_dcb(adapter))
3837                 return;
3838
3839 #endif
3840         if (ixgbe_cache_ring_fdir(adapter))
3841                 return;
3842
3843         if (ixgbe_cache_ring_rss(adapter))
3844                 return;
3845 }
3846
3847 /**
3848  * ixgbe_alloc_queues - Allocate memory for all rings
3849  * @adapter: board private structure to initialize
3850  *
3851  * We allocate one ring per queue at run-time since we don't know the
3852  * number of queues at compile-time.  The polling_netdev array is
3853  * intended for Multiqueue, but should work fine with a single queue.
3854  **/
3855 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
3856 {
3857         int i;
3858         int orig_node = adapter->node;
3859
3860         for (i = 0; i < adapter->num_tx_queues; i++) {
3861                 struct ixgbe_ring *ring = adapter->tx_ring[i];
3862                 if (orig_node == -1) {
3863                         int cur_node = next_online_node(adapter->node);
3864                         if (cur_node == MAX_NUMNODES)
3865                                 cur_node = first_online_node;
3866                         adapter->node = cur_node;
3867                 }
3868                 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
3869                                     adapter->node);
3870                 if (!ring)
3871                         ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
3872                 if (!ring)
3873                         goto err_tx_ring_allocation;
3874                 ring->count = adapter->tx_ring_count;
3875                 ring->queue_index = i;
3876                 ring->numa_node = adapter->node;
3877
3878                 adapter->tx_ring[i] = ring;
3879         }
3880
3881         /* Restore the adapter's original node */
3882         adapter->node = orig_node;
3883
3884         for (i = 0; i < adapter->num_rx_queues; i++) {
3885                 struct ixgbe_ring *ring = adapter->rx_ring[i];
3886                 if (orig_node == -1) {
3887                         int cur_node = next_online_node(adapter->node);
3888                         if (cur_node == MAX_NUMNODES)
3889                                 cur_node = first_online_node;
3890                         adapter->node = cur_node;
3891                 }
3892                 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
3893                                     adapter->node);
3894                 if (!ring)
3895                         ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
3896                 if (!ring)
3897                         goto err_rx_ring_allocation;
3898                 ring->count = adapter->rx_ring_count;
3899                 ring->queue_index = i;
3900                 ring->numa_node = adapter->node;
3901
3902                 adapter->rx_ring[i] = ring;
3903         }
3904
3905         /* Restore the adapter's original node */
3906         adapter->node = orig_node;
3907
3908         ixgbe_cache_ring_register(adapter);
3909
3910         return 0;
3911
3912 err_rx_ring_allocation:
3913         for (i = 0; i < adapter->num_tx_queues; i++)
3914                 kfree(adapter->tx_ring[i]);
3915 err_tx_ring_allocation:
3916         return -ENOMEM;
3917 }
3918
3919 /**
3920  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
3921  * @adapter: board private structure to initialize
3922  *
3923  * Attempt to configure the interrupts using the best available
3924  * capabilities of the hardware and the kernel.
3925  **/
3926 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
3927 {
3928         struct ixgbe_hw *hw = &adapter->hw;
3929         int err = 0;
3930         int vector, v_budget;
3931
3932         /*
3933          * It's easy to be greedy for MSI-X vectors, but it really
3934          * doesn't do us much good if we have a lot more vectors
3935          * than CPU's.  So let's be conservative and only ask for
3936          * (roughly) the same number of vectors as there are CPU's.
3937          */
3938         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
3939                        (int)num_online_cpus()) + NON_Q_VECTORS;
3940
3941         /*
3942          * At the same time, hardware can only support a maximum of
3943          * hw.mac->max_msix_vectors vectors.  With features
3944          * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
3945          * descriptor queues supported by our device.  Thus, we cap it off in
3946          * those rare cases where the cpu count also exceeds our vector limit.
3947          */
3948         v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
3949
3950         /* A failure in MSI-X entry allocation isn't fatal, but it does
3951          * mean we disable MSI-X capabilities of the adapter. */
3952         adapter->msix_entries = kcalloc(v_budget,
3953                                         sizeof(struct msix_entry), GFP_KERNEL);
3954         if (adapter->msix_entries) {
3955                 for (vector = 0; vector < v_budget; vector++)
3956                         adapter->msix_entries[vector].entry = vector;
3957
3958                 ixgbe_acquire_msix_vectors(adapter, v_budget);
3959
3960                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3961                         goto out;
3962         }
3963
3964         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
3965         adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
3966         adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3967         adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3968         adapter->atr_sample_rate = 0;
3969         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3970                 ixgbe_disable_sriov(adapter);
3971
3972         ixgbe_set_num_queues(adapter);
3973
3974         err = pci_enable_msi(adapter->pdev);
3975         if (!err) {
3976                 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
3977         } else {
3978                 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
3979                         "falling back to legacy.  Error: %d\n", err);
3980                 /* reset err */
3981                 err = 0;
3982         }
3983
3984 out:
3985         return err;
3986 }
3987
3988 /**
3989  * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
3990  * @adapter: board private structure to initialize
3991  *
3992  * We allocate one q_vector per queue interrupt.  If allocation fails we
3993  * return -ENOMEM.
3994  **/
3995 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
3996 {
3997         int q_idx, num_q_vectors;
3998         struct ixgbe_q_vector *q_vector;
3999         int napi_vectors;
4000         int (*poll)(struct napi_struct *, int);
4001
4002         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4003                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4004                 napi_vectors = adapter->num_rx_queues;
4005                 poll = &ixgbe_clean_rxtx_many;
4006         } else {
4007                 num_q_vectors = 1;
4008                 napi_vectors = 1;
4009                 poll = &ixgbe_poll;
4010         }
4011
4012         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4013                 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4014                                         GFP_KERNEL, adapter->node);
4015                 if (!q_vector)
4016                         q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4017                                            GFP_KERNEL);
4018                 if (!q_vector)
4019                         goto err_out;
4020                 q_vector->adapter = adapter;
4021                 if (q_vector->txr_count && !q_vector->rxr_count)
4022                         q_vector->eitr = adapter->tx_eitr_param;
4023                 else
4024                         q_vector->eitr = adapter->rx_eitr_param;
4025                 q_vector->v_idx = q_idx;
4026                 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4027                 adapter->q_vector[q_idx] = q_vector;
4028         }
4029
4030         return 0;
4031
4032 err_out:
4033         while (q_idx) {
4034                 q_idx--;
4035                 q_vector = adapter->q_vector[q_idx];
4036                 netif_napi_del(&q_vector->napi);
4037                 kfree(q_vector);
4038                 adapter->q_vector[q_idx] = NULL;
4039         }
4040         return -ENOMEM;
4041 }
4042
4043 /**
4044  * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4045  * @adapter: board private structure to initialize
4046  *
4047  * This function frees the memory allocated to the q_vectors.  In addition if
4048  * NAPI is enabled it will delete any references to the NAPI struct prior
4049  * to freeing the q_vector.
4050  **/
4051 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4052 {
4053         int q_idx, num_q_vectors;
4054
4055         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4056                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4057         else
4058                 num_q_vectors = 1;
4059
4060         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4061                 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4062                 adapter->q_vector[q_idx] = NULL;
4063                 netif_napi_del(&q_vector->napi);
4064                 kfree(q_vector);
4065         }
4066 }
4067
4068 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4069 {
4070         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4071                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4072                 pci_disable_msix(adapter->pdev);
4073                 kfree(adapter->msix_entries);
4074                 adapter->msix_entries = NULL;
4075         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4076                 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4077                 pci_disable_msi(adapter->pdev);
4078         }
4079         return;
4080 }
4081
4082 /**
4083  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4084  * @adapter: board private structure to initialize
4085  *
4086  * We determine which interrupt scheme to use based on...
4087  * - Kernel support (MSI, MSI-X)
4088  *   - which can be user-defined (via MODULE_PARAM)
4089  * - Hardware queue count (num_*_queues)
4090  *   - defined by miscellaneous hardware support/features (RSS, etc.)
4091  **/
4092 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4093 {
4094         int err;
4095
4096         /* Number of supported queues */
4097         ixgbe_set_num_queues(adapter);
4098
4099         err = ixgbe_set_interrupt_capability(adapter);
4100         if (err) {
4101                 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
4102                 goto err_set_interrupt;
4103         }
4104
4105         err = ixgbe_alloc_q_vectors(adapter);
4106         if (err) {
4107                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queue "
4108                         "vectors\n");
4109                 goto err_alloc_q_vectors;
4110         }
4111
4112         err = ixgbe_alloc_queues(adapter);
4113         if (err) {
4114                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
4115                 goto err_alloc_queues;
4116         }
4117
4118         DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
4119                 "Tx Queue count = %u\n",
4120                 (adapter->num_rx_queues > 1) ? "Enabled" :
4121                 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
4122
4123         set_bit(__IXGBE_DOWN, &adapter->state);
4124
4125         return 0;
4126
4127 err_alloc_queues:
4128         ixgbe_free_q_vectors(adapter);
4129 err_alloc_q_vectors:
4130         ixgbe_reset_interrupt_capability(adapter);
4131 err_set_interrupt:
4132         return err;
4133 }
4134
4135 /**
4136  * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4137  * @adapter: board private structure to clear interrupt scheme on
4138  *
4139  * We go through and clear interrupt specific resources and reset the structure
4140  * to pre-load conditions
4141  **/
4142 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4143 {
4144         int i;
4145
4146         for (i = 0; i < adapter->num_tx_queues; i++) {
4147                 kfree(adapter->tx_ring[i]);
4148                 adapter->tx_ring[i] = NULL;
4149         }
4150         for (i = 0; i < adapter->num_rx_queues; i++) {
4151                 kfree(adapter->rx_ring[i]);
4152                 adapter->rx_ring[i] = NULL;
4153         }
4154
4155         ixgbe_free_q_vectors(adapter);
4156         ixgbe_reset_interrupt_capability(adapter);
4157 }
4158
4159 /**
4160  * ixgbe_sfp_timer - worker thread to find a missing module
4161  * @data: pointer to our adapter struct
4162  **/
4163 static void ixgbe_sfp_timer(unsigned long data)
4164 {
4165         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4166
4167         /*
4168          * Do the sfp_timer outside of interrupt context due to the
4169          * delays that sfp+ detection requires
4170          */
4171         schedule_work(&adapter->sfp_task);
4172 }
4173
4174 /**
4175  * ixgbe_sfp_task - worker thread to find a missing module
4176  * @work: pointer to work_struct containing our data
4177  **/
4178 static void ixgbe_sfp_task(struct work_struct *work)
4179 {
4180         struct ixgbe_adapter *adapter = container_of(work,
4181                                                      struct ixgbe_adapter,
4182                                                      sfp_task);
4183         struct ixgbe_hw *hw = &adapter->hw;
4184
4185         if ((hw->phy.type == ixgbe_phy_nl) &&
4186             (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4187                 s32 ret = hw->phy.ops.identify_sfp(hw);
4188                 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4189                         goto reschedule;
4190                 ret = hw->phy.ops.reset(hw);
4191                 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4192                         dev_err(&adapter->pdev->dev, "failed to initialize "
4193                                 "because an unsupported SFP+ module type "
4194                                 "was detected.\n"
4195                                 "Reload the driver after installing a "
4196                                 "supported module.\n");
4197                         unregister_netdev(adapter->netdev);
4198                 } else {
4199                         DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
4200                                 hw->phy.sfp_type);
4201                 }
4202                 /* don't need this routine any more */
4203                 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4204         }
4205         return;
4206 reschedule:
4207         if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4208                 mod_timer(&adapter->sfp_timer,
4209                           round_jiffies(jiffies + (2 * HZ)));
4210 }
4211
4212 /**
4213  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4214  * @adapter: board private structure to initialize
4215  *
4216  * ixgbe_sw_init initializes the Adapter private data structure.
4217  * Fields are initialized based on PCI device information and
4218  * OS network device settings (MTU size).
4219  **/
4220 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4221 {
4222         struct ixgbe_hw *hw = &adapter->hw;
4223         struct pci_dev *pdev = adapter->pdev;
4224         struct net_device *dev = adapter->netdev;
4225         unsigned int rss;
4226 #ifdef CONFIG_IXGBE_DCB
4227         int j;
4228         struct tc_configuration *tc;
4229 #endif
4230
4231         /* PCI config space info */
4232
4233         hw->vendor_id = pdev->vendor;
4234         hw->device_id = pdev->device;
4235         hw->revision_id = pdev->revision;
4236         hw->subsystem_vendor_id = pdev->subsystem_vendor;
4237         hw->subsystem_device_id = pdev->subsystem_device;
4238
4239         /* Set capability flags */
4240         rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4241         adapter->ring_feature[RING_F_RSS].indices = rss;
4242         adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4243         adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4244         if (hw->mac.type == ixgbe_mac_82598EB) {
4245                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4246                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4247                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4248         } else if (hw->mac.type == ixgbe_mac_82599EB) {
4249                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4250                 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4251                 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4252                 if (dev->features & NETIF_F_NTUPLE) {
4253                         /* Flow Director perfect filter enabled */
4254                         adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4255                         adapter->atr_sample_rate = 0;
4256                         spin_lock_init(&adapter->fdir_perfect_lock);
4257                 } else {
4258                         /* Flow Director hash filters enabled */
4259                         adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4260                         adapter->atr_sample_rate = 20;
4261                 }
4262                 adapter->ring_feature[RING_F_FDIR].indices =
4263                                                          IXGBE_MAX_FDIR_INDICES;
4264                 adapter->fdir_pballoc = 0;
4265 #ifdef IXGBE_FCOE
4266                 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4267                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4268                 adapter->ring_feature[RING_F_FCOE].indices = 0;
4269 #ifdef CONFIG_IXGBE_DCB
4270                 /* Default traffic class to use for FCoE */
4271                 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4272 #endif
4273 #endif /* IXGBE_FCOE */
4274         }
4275
4276 #ifdef CONFIG_IXGBE_DCB
4277         /* Configure DCB traffic classes */
4278         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4279                 tc = &adapter->dcb_cfg.tc_config[j];
4280                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4281                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4282                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4283                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4284                 tc->dcb_pfc = pfc_disabled;
4285         }
4286         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4287         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4288         adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4289         adapter->dcb_cfg.pfc_mode_enable = false;
4290         adapter->dcb_cfg.round_robin_enable = false;
4291         adapter->dcb_set_bitmap = 0x00;
4292         ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4293                            adapter->ring_feature[RING_F_DCB].indices);
4294
4295 #endif
4296
4297         /* default flow control settings */
4298         hw->fc.requested_mode = ixgbe_fc_full;
4299         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
4300 #ifdef CONFIG_DCB
4301         adapter->last_lfc_mode = hw->fc.current_mode;
4302 #endif
4303         hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
4304         hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
4305         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4306         hw->fc.send_xon = true;
4307         hw->fc.disable_fc_autoneg = false;
4308
4309         /* enable itr by default in dynamic mode */
4310         adapter->rx_itr_setting = 1;
4311         adapter->rx_eitr_param = 20000;
4312         adapter->tx_itr_setting = 1;
4313         adapter->tx_eitr_param = 10000;
4314
4315         /* set defaults for eitr in MegaBytes */
4316         adapter->eitr_low = 10;
4317         adapter->eitr_high = 20;
4318
4319         /* set default ring sizes */
4320         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4321         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4322
4323         /* initialize eeprom parameters */
4324         if (ixgbe_init_eeprom_params_generic(hw)) {
4325                 dev_err(&pdev->dev, "EEPROM initialization failed\n");
4326                 return -EIO;
4327         }
4328
4329         /* enable rx csum by default */
4330         adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4331
4332         /* get assigned NUMA node */
4333         adapter->node = dev_to_node(&pdev->dev);
4334
4335         set_bit(__IXGBE_DOWN, &adapter->state);
4336
4337         return 0;
4338 }
4339
4340 /**
4341  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4342  * @adapter: board private structure
4343  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
4344  *
4345  * Return 0 on success, negative on failure
4346  **/
4347 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4348                              struct ixgbe_ring *tx_ring)
4349 {
4350         struct pci_dev *pdev = adapter->pdev;
4351         int size;
4352
4353         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4354         tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4355         if (!tx_ring->tx_buffer_info)
4356                 tx_ring->tx_buffer_info = vmalloc(size);
4357         if (!tx_ring->tx_buffer_info)
4358                 goto err;
4359         memset(tx_ring->tx_buffer_info, 0, size);
4360
4361         /* round up to nearest 4K */
4362         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4363         tx_ring->size = ALIGN(tx_ring->size, 4096);
4364
4365         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
4366                                              &tx_ring->dma);
4367         if (!tx_ring->desc)
4368                 goto err;
4369
4370         tx_ring->next_to_use = 0;
4371         tx_ring->next_to_clean = 0;
4372         tx_ring->work_limit = tx_ring->count;
4373         return 0;
4374
4375 err:
4376         vfree(tx_ring->tx_buffer_info);
4377         tx_ring->tx_buffer_info = NULL;
4378         DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
4379                             "descriptor ring\n");
4380         return -ENOMEM;
4381 }
4382
4383 /**
4384  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4385  * @adapter: board private structure
4386  *
4387  * If this function returns with an error, then it's possible one or
4388  * more of the rings is populated (while the rest are not).  It is the
4389  * callers duty to clean those orphaned rings.
4390  *
4391  * Return 0 on success, negative on failure
4392  **/
4393 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4394 {
4395         int i, err = 0;
4396
4397         for (i = 0; i < adapter->num_tx_queues; i++) {
4398                 err = ixgbe_setup_tx_resources(adapter, adapter->tx_ring[i]);
4399                 if (!err)
4400                         continue;
4401                 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
4402                 break;
4403         }
4404
4405         return err;
4406 }
4407
4408 /**
4409  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4410  * @adapter: board private structure
4411  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
4412  *
4413  * Returns 0 on success, negative on failure
4414  **/
4415 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4416                              struct ixgbe_ring *rx_ring)
4417 {
4418         struct pci_dev *pdev = adapter->pdev;
4419         int size;
4420
4421         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4422         rx_ring->rx_buffer_info = vmalloc_node(size, adapter->node);
4423         if (!rx_ring->rx_buffer_info)
4424                 rx_ring->rx_buffer_info = vmalloc(size);
4425         if (!rx_ring->rx_buffer_info) {
4426                 DPRINTK(PROBE, ERR,
4427                         "vmalloc allocation failed for the rx desc ring\n");
4428                 goto alloc_failed;
4429         }
4430         memset(rx_ring->rx_buffer_info, 0, size);
4431
4432         /* Round up to nearest 4K */
4433         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4434         rx_ring->size = ALIGN(rx_ring->size, 4096);
4435
4436         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
4437
4438         if (!rx_ring->desc) {
4439                 DPRINTK(PROBE, ERR,
4440                         "Memory allocation failed for the rx desc ring\n");
4441                 vfree(rx_ring->rx_buffer_info);
4442                 goto alloc_failed;
4443         }
4444
4445         rx_ring->next_to_clean = 0;
4446         rx_ring->next_to_use = 0;
4447
4448         return 0;
4449
4450 alloc_failed:
4451         return -ENOMEM;
4452 }
4453
4454 /**
4455  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4456  * @adapter: board private structure
4457  *
4458  * If this function returns with an error, then it's possible one or
4459  * more of the rings is populated (while the rest are not).  It is the
4460  * callers duty to clean those orphaned rings.
4461  *
4462  * Return 0 on success, negative on failure
4463  **/
4464
4465 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4466 {
4467         int i, err = 0;
4468
4469         for (i = 0; i < adapter->num_rx_queues; i++) {
4470                 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
4471                 if (!err)
4472                         continue;
4473                 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
4474                 break;
4475         }
4476
4477         return err;
4478 }
4479
4480 /**
4481  * ixgbe_free_tx_resources - Free Tx Resources per Queue
4482  * @adapter: board private structure
4483  * @tx_ring: Tx descriptor ring for a specific queue
4484  *
4485  * Free all transmit software resources
4486  **/
4487 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4488                              struct ixgbe_ring *tx_ring)
4489 {
4490         struct pci_dev *pdev = adapter->pdev;
4491
4492         ixgbe_clean_tx_ring(adapter, tx_ring);
4493
4494         vfree(tx_ring->tx_buffer_info);
4495         tx_ring->tx_buffer_info = NULL;
4496
4497         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
4498
4499         tx_ring->desc = NULL;
4500 }
4501
4502 /**
4503  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4504  * @adapter: board private structure
4505  *
4506  * Free all transmit software resources
4507  **/
4508 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4509 {
4510         int i;
4511
4512         for (i = 0; i < adapter->num_tx_queues; i++)
4513                 if (adapter->tx_ring[i]->desc)
4514                         ixgbe_free_tx_resources(adapter, adapter->tx_ring[i]);
4515 }
4516
4517 /**
4518  * ixgbe_free_rx_resources - Free Rx Resources
4519  * @adapter: board private structure
4520  * @rx_ring: ring to clean the resources from
4521  *
4522  * Free all receive software resources
4523  **/
4524 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
4525                              struct ixgbe_ring *rx_ring)
4526 {
4527         struct pci_dev *pdev = adapter->pdev;
4528
4529         ixgbe_clean_rx_ring(adapter, rx_ring);
4530
4531         vfree(rx_ring->rx_buffer_info);
4532         rx_ring->rx_buffer_info = NULL;
4533
4534         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
4535
4536         rx_ring->desc = NULL;
4537 }
4538
4539 /**
4540  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4541  * @adapter: board private structure
4542  *
4543  * Free all receive software resources
4544  **/
4545 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4546 {
4547         int i;
4548
4549         for (i = 0; i < adapter->num_rx_queues; i++)
4550                 if (adapter->rx_ring[i]->desc)
4551                         ixgbe_free_rx_resources(adapter, adapter->rx_ring[i]);
4552 }
4553
4554 /**
4555  * ixgbe_change_mtu - Change the Maximum Transfer Unit
4556  * @netdev: network interface device structure
4557  * @new_mtu: new value for maximum frame size
4558  *
4559  * Returns 0 on success, negative on failure
4560  **/
4561 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4562 {
4563         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4564         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4565
4566         /* MTU < 68 is an error and causes problems on some kernels */
4567         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4568                 return -EINVAL;
4569
4570         DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
4571                 netdev->mtu, new_mtu);
4572         /* must set new MTU before calling down or up */
4573         netdev->mtu = new_mtu;
4574
4575         if (netif_running(netdev))
4576                 ixgbe_reinit_locked(adapter);
4577
4578         return 0;
4579 }
4580
4581 /**
4582  * ixgbe_open - Called when a network interface is made active
4583  * @netdev: network interface device structure
4584  *
4585  * Returns 0 on success, negative value on failure
4586  *
4587  * The open entry point is called when a network interface is made
4588  * active by the system (IFF_UP).  At this point all resources needed
4589  * for transmit and receive operations are allocated, the interrupt
4590  * handler is registered with the OS, the watchdog timer is started,
4591  * and the stack is notified that the interface is ready.
4592  **/
4593 static int ixgbe_open(struct net_device *netdev)
4594 {
4595         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4596         int err;
4597
4598         /* disallow open during test */
4599         if (test_bit(__IXGBE_TESTING, &adapter->state))
4600                 return -EBUSY;
4601
4602         netif_carrier_off(netdev);
4603
4604         /* allocate transmit descriptors */
4605         err = ixgbe_setup_all_tx_resources(adapter);
4606         if (err)
4607                 goto err_setup_tx;
4608
4609         /* allocate receive descriptors */
4610         err = ixgbe_setup_all_rx_resources(adapter);
4611         if (err)
4612                 goto err_setup_rx;
4613
4614         ixgbe_configure(adapter);
4615
4616         err = ixgbe_request_irq(adapter);
4617         if (err)
4618                 goto err_req_irq;
4619
4620         err = ixgbe_up_complete(adapter);
4621         if (err)
4622                 goto err_up;
4623
4624         netif_tx_start_all_queues(netdev);
4625
4626         return 0;
4627
4628 err_up:
4629         ixgbe_release_hw_control(adapter);
4630         ixgbe_free_irq(adapter);
4631 err_req_irq:
4632 err_setup_rx:
4633         ixgbe_free_all_rx_resources(adapter);
4634 err_setup_tx:
4635         ixgbe_free_all_tx_resources(adapter);
4636         ixgbe_reset(adapter);
4637
4638         return err;
4639 }
4640
4641 /**
4642  * ixgbe_close - Disables a network interface
4643  * @netdev: network interface device structure
4644  *
4645  * Returns 0, this is not allowed to fail
4646  *
4647  * The close entry point is called when an interface is de-activated
4648  * by the OS.  The hardware is still under the drivers control, but
4649  * needs to be disabled.  A global MAC reset is issued to stop the
4650  * hardware, and all transmit and receive resources are freed.
4651  **/
4652 static int ixgbe_close(struct net_device *netdev)
4653 {
4654         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4655
4656         ixgbe_down(adapter);
4657         ixgbe_free_irq(adapter);
4658
4659         ixgbe_free_all_tx_resources(adapter);
4660         ixgbe_free_all_rx_resources(adapter);
4661
4662         ixgbe_release_hw_control(adapter);
4663
4664         return 0;
4665 }
4666
4667 #ifdef CONFIG_PM
4668 static int ixgbe_resume(struct pci_dev *pdev)
4669 {
4670         struct net_device *netdev = pci_get_drvdata(pdev);
4671         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4672         u32 err;
4673
4674         pci_set_power_state(pdev, PCI_D0);
4675         pci_restore_state(pdev);
4676         /*
4677          * pci_restore_state clears dev->state_saved so call
4678          * pci_save_state to restore it.
4679          */
4680         pci_save_state(pdev);
4681
4682         err = pci_enable_device_mem(pdev);
4683         if (err) {
4684                 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
4685                                 "suspend\n");
4686                 return err;
4687         }
4688         pci_set_master(pdev);
4689
4690         pci_wake_from_d3(pdev, false);
4691
4692         err = ixgbe_init_interrupt_scheme(adapter);
4693         if (err) {
4694                 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
4695                                 "device\n");
4696                 return err;
4697         }
4698
4699         ixgbe_reset(adapter);
4700
4701         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4702
4703         if (netif_running(netdev)) {
4704                 err = ixgbe_open(adapter->netdev);
4705                 if (err)
4706                         return err;
4707         }
4708
4709         netif_device_attach(netdev);
4710
4711         return 0;
4712 }
4713 #endif /* CONFIG_PM */
4714
4715 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4716 {
4717         struct net_device *netdev = pci_get_drvdata(pdev);
4718         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4719         struct ixgbe_hw *hw = &adapter->hw;
4720         u32 ctrl, fctrl;
4721         u32 wufc = adapter->wol;
4722 #ifdef CONFIG_PM
4723         int retval = 0;
4724 #endif
4725
4726         netif_device_detach(netdev);
4727
4728         if (netif_running(netdev)) {
4729                 ixgbe_down(adapter);
4730                 ixgbe_free_irq(adapter);
4731                 ixgbe_free_all_tx_resources(adapter);
4732                 ixgbe_free_all_rx_resources(adapter);
4733         }
4734         ixgbe_clear_interrupt_scheme(adapter);
4735
4736 #ifdef CONFIG_PM
4737         retval = pci_save_state(pdev);
4738         if (retval)
4739                 return retval;
4740
4741 #endif
4742         if (wufc) {
4743                 ixgbe_set_rx_mode(netdev);
4744
4745                 /* turn on all-multi mode if wake on multicast is enabled */
4746                 if (wufc & IXGBE_WUFC_MC) {
4747                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4748                         fctrl |= IXGBE_FCTRL_MPE;
4749                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4750                 }
4751
4752                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4753                 ctrl |= IXGBE_CTRL_GIO_DIS;
4754                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4755
4756                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4757         } else {
4758                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4759                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4760         }
4761
4762         if (wufc && hw->mac.type == ixgbe_mac_82599EB)
4763                 pci_wake_from_d3(pdev, true);
4764         else
4765                 pci_wake_from_d3(pdev, false);
4766
4767         *enable_wake = !!wufc;
4768
4769         ixgbe_release_hw_control(adapter);
4770
4771         pci_disable_device(pdev);
4772
4773         return 0;
4774 }
4775
4776 #ifdef CONFIG_PM
4777 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4778 {
4779         int retval;
4780         bool wake;
4781
4782         retval = __ixgbe_shutdown(pdev, &wake);
4783         if (retval)
4784                 return retval;
4785
4786         if (wake) {
4787                 pci_prepare_to_sleep(pdev);
4788         } else {
4789                 pci_wake_from_d3(pdev, false);
4790                 pci_set_power_state(pdev, PCI_D3hot);
4791         }
4792
4793         return 0;
4794 }
4795 #endif /* CONFIG_PM */
4796
4797 static void ixgbe_shutdown(struct pci_dev *pdev)
4798 {
4799         bool wake;
4800
4801         __ixgbe_shutdown(pdev, &wake);
4802
4803         if (system_state == SYSTEM_POWER_OFF) {
4804                 pci_wake_from_d3(pdev, wake);
4805                 pci_set_power_state(pdev, PCI_D3hot);
4806         }
4807 }
4808
4809 /**
4810  * ixgbe_update_stats - Update the board statistics counters.
4811  * @adapter: board private structure
4812  **/
4813 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4814 {
4815         struct net_device *netdev = adapter->netdev;
4816         struct ixgbe_hw *hw = &adapter->hw;
4817         u64 total_mpc = 0;
4818         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4819         u64 non_eop_descs = 0, restart_queue = 0;
4820
4821         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
4822                 u64 rsc_count = 0;
4823                 u64 rsc_flush = 0;
4824                 for (i = 0; i < 16; i++)
4825                         adapter->hw_rx_no_dma_resources +=
4826                                              IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4827                 for (i = 0; i < adapter->num_rx_queues; i++) {
4828                         rsc_count += adapter->rx_ring[i]->rsc_count;
4829                         rsc_flush += adapter->rx_ring[i]->rsc_flush;
4830                 }
4831                 adapter->rsc_total_count = rsc_count;
4832                 adapter->rsc_total_flush = rsc_flush;
4833         }
4834
4835         /* gather some stats to the adapter struct that are per queue */
4836         for (i = 0; i < adapter->num_tx_queues; i++)
4837                 restart_queue += adapter->tx_ring[i]->restart_queue;
4838         adapter->restart_queue = restart_queue;
4839
4840         for (i = 0; i < adapter->num_rx_queues; i++)
4841                 non_eop_descs += adapter->rx_ring[i]->non_eop_descs;
4842         adapter->non_eop_descs = non_eop_descs;
4843
4844         adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
4845         for (i = 0; i < 8; i++) {
4846                 /* for packet buffers not used, the register should read 0 */
4847                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
4848                 missed_rx += mpc;
4849                 adapter->stats.mpc[i] += mpc;
4850                 total_mpc += adapter->stats.mpc[i];
4851                 if (hw->mac.type == ixgbe_mac_82598EB)
4852                         adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
4853                 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
4854                 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
4855                 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
4856                 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
4857                 if (hw->mac.type == ixgbe_mac_82599EB) {
4858                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4859                                                             IXGBE_PXONRXCNT(i));
4860                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4861                                                            IXGBE_PXOFFRXCNT(i));
4862                         adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4863                 } else {
4864                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4865                                                               IXGBE_PXONRXC(i));
4866                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4867                                                              IXGBE_PXOFFRXC(i));
4868                 }
4869                 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
4870                                                             IXGBE_PXONTXC(i));
4871                 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
4872                                                              IXGBE_PXOFFTXC(i));
4873         }
4874         adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
4875         /* work around hardware counting issue */
4876         adapter->stats.gprc -= missed_rx;
4877
4878         /* 82598 hardware only has a 32 bit counter in the high register */
4879         if (hw->mac.type == ixgbe_mac_82599EB) {
4880                 u64 tmp;
4881                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
4882                 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
4883                 adapter->stats.gorc += (tmp << 32);
4884                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
4885                 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
4886                 adapter->stats.gotc += (tmp << 32);
4887                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
4888                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
4889                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
4890                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
4891                 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
4892                 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
4893 #ifdef IXGBE_FCOE
4894                 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
4895                 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
4896                 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
4897                 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
4898                 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
4899                 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
4900 #endif /* IXGBE_FCOE */
4901         } else {
4902                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
4903                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
4904                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
4905                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
4906                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
4907         }
4908         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
4909         adapter->stats.bprc += bprc;
4910         adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
4911         if (hw->mac.type == ixgbe_mac_82598EB)
4912                 adapter->stats.mprc -= bprc;
4913         adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
4914         adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
4915         adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
4916         adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
4917         adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
4918         adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
4919         adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
4920         adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
4921         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
4922         adapter->stats.lxontxc += lxon;
4923         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
4924         adapter->stats.lxofftxc += lxoff;
4925         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4926         adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
4927         adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
4928         /*
4929          * 82598 errata - tx of flow control packets is included in tx counters
4930          */
4931         xon_off_tot = lxon + lxoff;
4932         adapter->stats.gptc -= xon_off_tot;
4933         adapter->stats.mptc -= xon_off_tot;
4934         adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
4935         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4936         adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
4937         adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
4938         adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
4939         adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
4940         adapter->stats.ptc64 -= xon_off_tot;
4941         adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
4942         adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
4943         adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
4944         adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
4945         adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
4946         adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
4947
4948         /* Fill out the OS statistics structure */
4949         netdev->stats.multicast = adapter->stats.mprc;
4950
4951         /* Rx Errors */
4952         netdev->stats.rx_errors = adapter->stats.crcerrs +
4953                                        adapter->stats.rlec;
4954         netdev->stats.rx_dropped = 0;
4955         netdev->stats.rx_length_errors = adapter->stats.rlec;
4956         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4957         netdev->stats.rx_missed_errors = total_mpc;
4958 }
4959
4960 /**
4961  * ixgbe_watchdog - Timer Call-back
4962  * @data: pointer to adapter cast into an unsigned long
4963  **/
4964 static void ixgbe_watchdog(unsigned long data)
4965 {
4966         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4967         struct ixgbe_hw *hw = &adapter->hw;
4968         u64 eics = 0;
4969         int i;
4970
4971         /*
4972          *  Do the watchdog outside of interrupt context due to the lovely
4973          * delays that some of the newer hardware requires
4974          */
4975
4976         if (test_bit(__IXGBE_DOWN, &adapter->state))
4977                 goto watchdog_short_circuit;
4978
4979         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
4980                 /*
4981                  * for legacy and MSI interrupts don't set any bits
4982                  * that are enabled for EIAM, because this operation
4983                  * would set *both* EIMS and EICS for any bit in EIAM
4984                  */
4985                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
4986                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
4987                 goto watchdog_reschedule;
4988         }
4989
4990         /* get one bit for every active tx/rx interrupt vector */
4991         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
4992                 struct ixgbe_q_vector *qv = adapter->q_vector[i];
4993                 if (qv->rxr_count || qv->txr_count)
4994                         eics |= ((u64)1 << i);
4995         }
4996
4997         /* Cause software interrupt to ensure rx rings are cleaned */
4998         ixgbe_irq_rearm_queues(adapter, eics);
4999
5000 watchdog_reschedule:
5001         /* Reset the timer */
5002         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5003
5004 watchdog_short_circuit:
5005         schedule_work(&adapter->watchdog_task);
5006 }
5007
5008 /**
5009  * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5010  * @work: pointer to work_struct containing our data
5011  **/
5012 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5013 {
5014         struct ixgbe_adapter *adapter = container_of(work,
5015                                                      struct ixgbe_adapter,
5016                                                      multispeed_fiber_task);
5017         struct ixgbe_hw *hw = &adapter->hw;
5018         u32 autoneg;
5019         bool negotiation;
5020
5021         adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5022         autoneg = hw->phy.autoneg_advertised;
5023         if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5024                 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5025         hw->mac.autotry_restart = false;
5026         if (hw->mac.ops.setup_link)
5027                 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5028         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5029         adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5030 }
5031
5032 /**
5033  * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5034  * @work: pointer to work_struct containing our data
5035  **/
5036 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5037 {
5038         struct ixgbe_adapter *adapter = container_of(work,
5039                                                      struct ixgbe_adapter,
5040                                                      sfp_config_module_task);
5041         struct ixgbe_hw *hw = &adapter->hw;
5042         u32 err;
5043
5044         adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5045
5046         /* Time for electrical oscillations to settle down */
5047         msleep(100);
5048         err = hw->phy.ops.identify_sfp(hw);
5049
5050         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5051                 dev_err(&adapter->pdev->dev, "failed to initialize because "
5052                         "an unsupported SFP+ module type was detected.\n"
5053                         "Reload the driver after installing a supported "
5054                         "module.\n");
5055                 unregister_netdev(adapter->netdev);
5056                 return;
5057         }
5058         hw->mac.ops.setup_sfp(hw);
5059
5060         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5061                 /* This will also work for DA Twinax connections */
5062                 schedule_work(&adapter->multispeed_fiber_task);
5063         adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5064 }
5065
5066 /**
5067  * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5068  * @work: pointer to work_struct containing our data
5069  **/
5070 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5071 {
5072         struct ixgbe_adapter *adapter = container_of(work,
5073                                                      struct ixgbe_adapter,
5074                                                      fdir_reinit_task);
5075         struct ixgbe_hw *hw = &adapter->hw;
5076         int i;
5077
5078         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5079                 for (i = 0; i < adapter->num_tx_queues; i++)
5080                         set_bit(__IXGBE_FDIR_INIT_DONE,
5081                                 &(adapter->tx_ring[i]->reinit_state));
5082         } else {
5083                 DPRINTK(PROBE, ERR, "failed to finish FDIR re-initialization, "
5084                         "ignored adding FDIR ATR filters \n");
5085         }
5086         /* Done FDIR Re-initialization, enable transmits */
5087         netif_tx_start_all_queues(adapter->netdev);
5088 }
5089
5090 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5091
5092 /**
5093  * ixgbe_watchdog_task - worker thread to bring link up
5094  * @work: pointer to work_struct containing our data
5095  **/
5096 static void ixgbe_watchdog_task(struct work_struct *work)
5097 {
5098         struct ixgbe_adapter *adapter = container_of(work,
5099                                                      struct ixgbe_adapter,
5100                                                      watchdog_task);
5101         struct net_device *netdev = adapter->netdev;
5102         struct ixgbe_hw *hw = &adapter->hw;
5103         u32 link_speed;
5104         bool link_up;
5105         int i;
5106         struct ixgbe_ring *tx_ring;
5107         int some_tx_pending = 0;
5108
5109         mutex_lock(&ixgbe_watchdog_lock);
5110
5111         link_up = adapter->link_up;
5112         link_speed = adapter->link_speed;
5113
5114         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5115                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5116                 if (link_up) {
5117 #ifdef CONFIG_DCB
5118                         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5119                                 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5120                                         hw->mac.ops.fc_enable(hw, i);
5121                         } else {
5122                                 hw->mac.ops.fc_enable(hw, 0);
5123                         }
5124 #else
5125                         hw->mac.ops.fc_enable(hw, 0);
5126 #endif
5127                 }
5128
5129                 if (link_up ||
5130                     time_after(jiffies, (adapter->link_check_timeout +
5131                                          IXGBE_TRY_LINK_TIMEOUT))) {
5132                         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5133                         IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5134                 }
5135                 adapter->link_up = link_up;
5136                 adapter->link_speed = link_speed;
5137         }
5138
5139         if (link_up) {
5140                 if (!netif_carrier_ok(netdev)) {
5141                         bool flow_rx, flow_tx;
5142
5143                         if (hw->mac.type == ixgbe_mac_82599EB) {
5144                                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5145                                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5146                                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5147                                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5148                         } else {
5149                                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5150                                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5151                                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5152                                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5153                         }
5154
5155                         printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
5156                                "Flow Control: %s\n",
5157                                netdev->name,
5158                                (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5159                                 "10 Gbps" :
5160                                 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5161                                  "1 Gbps" : "unknown speed")),
5162                                ((flow_rx && flow_tx) ? "RX/TX" :
5163                                 (flow_rx ? "RX" :
5164                                 (flow_tx ? "TX" : "None"))));
5165
5166                         netif_carrier_on(netdev);
5167                 } else {
5168                         /* Force detection of hung controller */
5169                         adapter->detect_tx_hung = true;
5170                 }
5171         } else {
5172                 adapter->link_up = false;
5173                 adapter->link_speed = 0;
5174                 if (netif_carrier_ok(netdev)) {
5175                         printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
5176                                netdev->name);
5177                         netif_carrier_off(netdev);
5178                 }
5179         }
5180
5181         if (!netif_carrier_ok(netdev)) {
5182                 for (i = 0; i < adapter->num_tx_queues; i++) {
5183                         tx_ring = adapter->tx_ring[i];
5184                         if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5185                                 some_tx_pending = 1;
5186                                 break;
5187                         }
5188                 }
5189
5190                 if (some_tx_pending) {
5191                         /* We've lost link, so the controller stops DMA,
5192                          * but we've got queued Tx work that's never going
5193                          * to get done, so reset controller to flush Tx.
5194                          * (Do the reset outside of interrupt context).
5195                          */
5196                          schedule_work(&adapter->reset_task);
5197                 }
5198         }
5199
5200         ixgbe_update_stats(adapter);
5201         mutex_unlock(&ixgbe_watchdog_lock);
5202 }
5203
5204 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5205                      struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5206                      u32 tx_flags, u8 *hdr_len)
5207 {
5208         struct ixgbe_adv_tx_context_desc *context_desc;
5209         unsigned int i;
5210         int err;
5211         struct ixgbe_tx_buffer *tx_buffer_info;
5212         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5213         u32 mss_l4len_idx, l4len;
5214
5215         if (skb_is_gso(skb)) {
5216                 if (skb_header_cloned(skb)) {
5217                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5218                         if (err)
5219                                 return err;
5220                 }
5221                 l4len = tcp_hdrlen(skb);
5222                 *hdr_len += l4len;
5223
5224                 if (skb->protocol == htons(ETH_P_IP)) {
5225                         struct iphdr *iph = ip_hdr(skb);
5226                         iph->tot_len = 0;
5227                         iph->check = 0;
5228                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5229                                                                  iph->daddr, 0,
5230                                                                  IPPROTO_TCP,
5231                                                                  0);
5232                 } else if (skb_is_gso_v6(skb)) {
5233                         ipv6_hdr(skb)->payload_len = 0;
5234                         tcp_hdr(skb)->check =
5235                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5236                                              &ipv6_hdr(skb)->daddr,
5237                                              0, IPPROTO_TCP, 0);
5238                 }
5239
5240                 i = tx_ring->next_to_use;
5241
5242                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5243                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5244
5245                 /* VLAN MACLEN IPLEN */
5246                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5247                         vlan_macip_lens |=
5248                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5249                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5250                                     IXGBE_ADVTXD_MACLEN_SHIFT);
5251                 *hdr_len += skb_network_offset(skb);
5252                 vlan_macip_lens |=
5253                     (skb_transport_header(skb) - skb_network_header(skb));
5254                 *hdr_len +=
5255                     (skb_transport_header(skb) - skb_network_header(skb));
5256                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5257                 context_desc->seqnum_seed = 0;
5258
5259                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5260                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5261                                    IXGBE_ADVTXD_DTYP_CTXT);
5262
5263                 if (skb->protocol == htons(ETH_P_IP))
5264                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5265                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5266                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5267
5268                 /* MSS L4LEN IDX */
5269                 mss_l4len_idx =
5270                     (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5271                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5272                 /* use index 1 for TSO */
5273                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5274                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5275
5276                 tx_buffer_info->time_stamp = jiffies;
5277                 tx_buffer_info->next_to_watch = i;
5278
5279                 i++;
5280                 if (i == tx_ring->count)
5281                         i = 0;
5282                 tx_ring->next_to_use = i;
5283
5284                 return true;
5285         }
5286         return false;
5287 }
5288
5289 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5290                           struct ixgbe_ring *tx_ring,
5291                           struct sk_buff *skb, u32 tx_flags)
5292 {
5293         struct ixgbe_adv_tx_context_desc *context_desc;
5294         unsigned int i;
5295         struct ixgbe_tx_buffer *tx_buffer_info;
5296         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5297
5298         if (skb->ip_summed == CHECKSUM_PARTIAL ||
5299             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5300                 i = tx_ring->next_to_use;
5301                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5302                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5303
5304                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5305                         vlan_macip_lens |=
5306                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5307                 vlan_macip_lens |= (skb_network_offset(skb) <<
5308                                     IXGBE_ADVTXD_MACLEN_SHIFT);
5309                 if (skb->ip_summed == CHECKSUM_PARTIAL)
5310                         vlan_macip_lens |= (skb_transport_header(skb) -
5311                                             skb_network_header(skb));
5312
5313                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5314                 context_desc->seqnum_seed = 0;
5315
5316                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5317                                     IXGBE_ADVTXD_DTYP_CTXT);
5318
5319                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5320                         __be16 protocol;
5321
5322                         if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
5323                                 const struct vlan_ethhdr *vhdr =
5324                                         (const struct vlan_ethhdr *)skb->data;
5325
5326                                 protocol = vhdr->h_vlan_encapsulated_proto;
5327                         } else {
5328                                 protocol = skb->protocol;
5329                         }
5330
5331                         switch (protocol) {
5332                         case cpu_to_be16(ETH_P_IP):
5333                                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5334                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5335                                         type_tucmd_mlhl |=
5336                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5337                                 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
5338                                         type_tucmd_mlhl |=
5339                                                 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5340                                 break;
5341                         case cpu_to_be16(ETH_P_IPV6):
5342                                 /* XXX what about other V6 headers?? */
5343                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5344                                         type_tucmd_mlhl |=
5345                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5346                                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
5347                                         type_tucmd_mlhl |=
5348                                                 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5349                                 break;
5350                         default:
5351                                 if (unlikely(net_ratelimit())) {
5352                                         DPRINTK(PROBE, WARNING,
5353                                          "partial checksum but proto=%x!\n",
5354                                          skb->protocol);
5355                                 }
5356                                 break;
5357                         }
5358                 }
5359
5360                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5361                 /* use index zero for tx checksum offload */
5362                 context_desc->mss_l4len_idx = 0;
5363
5364                 tx_buffer_info->time_stamp = jiffies;
5365                 tx_buffer_info->next_to_watch = i;
5366
5367                 i++;
5368                 if (i == tx_ring->count)
5369                         i = 0;
5370                 tx_ring->next_to_use = i;
5371
5372                 return true;
5373         }
5374
5375         return false;
5376 }
5377
5378 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5379                         struct ixgbe_ring *tx_ring,
5380                         struct sk_buff *skb, u32 tx_flags,
5381                         unsigned int first)
5382 {
5383         struct pci_dev *pdev = adapter->pdev;
5384         struct ixgbe_tx_buffer *tx_buffer_info;
5385         unsigned int len;
5386         unsigned int total = skb->len;
5387         unsigned int offset = 0, size, count = 0, i;
5388         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5389         unsigned int f;
5390
5391         i = tx_ring->next_to_use;
5392
5393         if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5394                 /* excluding fcoe_crc_eof for FCoE */
5395                 total -= sizeof(struct fcoe_crc_eof);
5396
5397         len = min(skb_headlen(skb), total);
5398         while (len) {
5399                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5400                 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5401
5402                 tx_buffer_info->length = size;
5403                 tx_buffer_info->mapped_as_page = false;
5404                 tx_buffer_info->dma = pci_map_single(pdev,
5405                                                      skb->data + offset,
5406                                                      size, PCI_DMA_TODEVICE);
5407                 if (pci_dma_mapping_error(pdev, tx_buffer_info->dma))
5408                         goto dma_error;
5409                 tx_buffer_info->time_stamp = jiffies;
5410                 tx_buffer_info->next_to_watch = i;
5411
5412                 len -= size;
5413                 total -= size;
5414                 offset += size;
5415                 count++;
5416
5417                 if (len) {
5418                         i++;
5419                         if (i == tx_ring->count)
5420                                 i = 0;
5421                 }
5422         }
5423
5424         for (f = 0; f < nr_frags; f++) {
5425                 struct skb_frag_struct *frag;
5426
5427                 frag = &skb_shinfo(skb)->frags[f];
5428                 len = min((unsigned int)frag->size, total);
5429                 offset = frag->page_offset;
5430
5431                 while (len) {
5432                         i++;
5433                         if (i == tx_ring->count)
5434                                 i = 0;
5435
5436                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
5437                         size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5438
5439                         tx_buffer_info->length = size;
5440                         tx_buffer_info->dma = pci_map_page(adapter->pdev,
5441                                                            frag->page,
5442                                                            offset, size,
5443                                                            PCI_DMA_TODEVICE);
5444                         tx_buffer_info->mapped_as_page = true;
5445                         if (pci_dma_mapping_error(pdev, tx_buffer_info->dma))
5446                                 goto dma_error;
5447                         tx_buffer_info->time_stamp = jiffies;
5448                         tx_buffer_info->next_to_watch = i;
5449
5450                         len -= size;
5451                         total -= size;
5452                         offset += size;
5453                         count++;
5454                 }
5455                 if (total == 0)
5456                         break;
5457         }
5458
5459         tx_ring->tx_buffer_info[i].skb = skb;
5460         tx_ring->tx_buffer_info[first].next_to_watch = i;
5461
5462         return count;
5463
5464 dma_error:
5465         dev_err(&pdev->dev, "TX DMA map failed\n");
5466
5467         /* clear timestamp and dma mappings for failed tx_buffer_info map */
5468         tx_buffer_info->dma = 0;
5469         tx_buffer_info->time_stamp = 0;
5470         tx_buffer_info->next_to_watch = 0;
5471         if (count)
5472                 count--;
5473
5474         /* clear timestamp and dma mappings for remaining portion of packet */
5475         while (count--) {
5476                 if (i==0)
5477                         i += tx_ring->count;
5478                 i--;
5479                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5480                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
5481         }
5482
5483         return 0;
5484 }
5485
5486 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
5487                            struct ixgbe_ring *tx_ring,
5488                            int tx_flags, int count, u32 paylen, u8 hdr_len)
5489 {
5490         union ixgbe_adv_tx_desc *tx_desc = NULL;
5491         struct ixgbe_tx_buffer *tx_buffer_info;
5492         u32 olinfo_status = 0, cmd_type_len = 0;
5493         unsigned int i;
5494         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
5495
5496         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
5497
5498         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
5499
5500         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5501                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
5502
5503         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
5504                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5505
5506                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5507                                  IXGBE_ADVTXD_POPTS_SHIFT;
5508
5509                 /* use index 1 context for tso */
5510                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5511                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5512                         olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
5513                                          IXGBE_ADVTXD_POPTS_SHIFT;
5514
5515         } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5516                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5517                                  IXGBE_ADVTXD_POPTS_SHIFT;
5518
5519         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5520                 olinfo_status |= IXGBE_ADVTXD_CC;
5521                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5522                 if (tx_flags & IXGBE_TX_FLAGS_FSO)
5523                         cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5524         }
5525
5526         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
5527
5528         i = tx_ring->next_to_use;
5529         while (count--) {
5530                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5531                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
5532                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
5533                 tx_desc->read.cmd_type_len =
5534                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
5535                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5536                 i++;
5537                 if (i == tx_ring->count)
5538                         i = 0;
5539         }
5540
5541         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
5542
5543         /*
5544          * Force memory writes to complete before letting h/w
5545          * know there are new descriptors to fetch.  (Only
5546          * applicable for weak-ordered memory model archs,
5547          * such as IA-64).
5548          */
5549         wmb();
5550
5551         tx_ring->next_to_use = i;
5552         writel(i, adapter->hw.hw_addr + tx_ring->tail);
5553 }
5554
5555 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5556                       int queue, u32 tx_flags)
5557 {
5558         /* Right now, we support IPv4 only */
5559         struct ixgbe_atr_input atr_input;
5560         struct tcphdr *th;
5561         struct iphdr *iph = ip_hdr(skb);
5562         struct ethhdr *eth = (struct ethhdr *)skb->data;
5563         u16 vlan_id, src_port, dst_port, flex_bytes;
5564         u32 src_ipv4_addr, dst_ipv4_addr;
5565         u8 l4type = 0;
5566
5567         /* check if we're UDP or TCP */
5568         if (iph->protocol == IPPROTO_TCP) {
5569                 th = tcp_hdr(skb);
5570                 src_port = th->source;
5571                 dst_port = th->dest;
5572                 l4type |= IXGBE_ATR_L4TYPE_TCP;
5573                 /* l4type IPv4 type is 0, no need to assign */
5574         } else {
5575                 /* Unsupported L4 header, just bail here */
5576                 return;
5577         }
5578
5579         memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
5580
5581         vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
5582                    IXGBE_TX_FLAGS_VLAN_SHIFT;
5583         src_ipv4_addr = iph->saddr;
5584         dst_ipv4_addr = iph->daddr;
5585         flex_bytes = eth->h_proto;
5586
5587         ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
5588         ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
5589         ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
5590         ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
5591         ixgbe_atr_set_l4type_82599(&atr_input, l4type);
5592         /* src and dst are inverted, think how the receiver sees them */
5593         ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
5594         ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
5595
5596         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
5597         ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
5598 }
5599
5600 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
5601                                  struct ixgbe_ring *tx_ring, int size)
5602 {
5603         netif_stop_subqueue(netdev, tx_ring->queue_index);
5604         /* Herbert's original patch had:
5605          *  smp_mb__after_netif_stop_queue();
5606          * but since that doesn't exist yet, just open code it. */
5607         smp_mb();
5608
5609         /* We need to check again in a case another CPU has just
5610          * made room available. */
5611         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
5612                 return -EBUSY;
5613
5614         /* A reprieve! - use start_queue because it doesn't call schedule */
5615         netif_start_subqueue(netdev, tx_ring->queue_index);
5616         ++tx_ring->restart_queue;
5617         return 0;
5618 }
5619
5620 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
5621                               struct ixgbe_ring *tx_ring, int size)
5622 {
5623         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
5624                 return 0;
5625         return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
5626 }
5627
5628 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
5629 {
5630         struct ixgbe_adapter *adapter = netdev_priv(dev);
5631         int txq = smp_processor_id();
5632
5633         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5634                 while (unlikely(txq >= dev->real_num_tx_queues))
5635                         txq -= dev->real_num_tx_queues;
5636                 return txq;
5637         }
5638
5639 #ifdef IXGBE_FCOE
5640         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5641             (skb->protocol == htons(ETH_P_FCOE))) {
5642                 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
5643                 txq += adapter->ring_feature[RING_F_FCOE].mask;
5644                 return txq;
5645         }
5646 #endif
5647         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5648                 if (skb->priority == TC_PRIO_CONTROL)
5649                         txq = adapter->ring_feature[RING_F_DCB].indices-1;
5650                 else
5651                         txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
5652                                >> 13;
5653                 return txq;
5654         }
5655
5656         return skb_tx_hash(dev, skb);
5657 }
5658
5659 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
5660                                     struct net_device *netdev)
5661 {
5662         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5663         struct ixgbe_ring *tx_ring;
5664         struct netdev_queue *txq;
5665         unsigned int first;
5666         unsigned int tx_flags = 0;
5667         u8 hdr_len = 0;
5668         int tso;
5669         int count = 0;
5670         unsigned int f;
5671
5672         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
5673                 tx_flags |= vlan_tx_tag_get(skb);
5674                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5675                         tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
5676                         tx_flags |= ((skb->queue_mapping & 0x7) << 13);
5677                 }
5678                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5679                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5680         } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5681                 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
5682                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5683                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5684         }
5685
5686         tx_ring = adapter->tx_ring[skb->queue_mapping];
5687
5688         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5689             (skb->protocol == htons(ETH_P_FCOE))) {
5690                 tx_flags |= IXGBE_TX_FLAGS_FCOE;
5691 #ifdef IXGBE_FCOE
5692 #ifdef CONFIG_IXGBE_DCB
5693                 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
5694                               << IXGBE_TX_FLAGS_VLAN_SHIFT);
5695                 tx_flags |= ((adapter->fcoe.up << 13)
5696                               << IXGBE_TX_FLAGS_VLAN_SHIFT);
5697 #endif
5698 #endif
5699         }
5700         /* four things can cause us to need a context descriptor */
5701         if (skb_is_gso(skb) ||
5702             (skb->ip_summed == CHECKSUM_PARTIAL) ||
5703             (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
5704             (tx_flags & IXGBE_TX_FLAGS_FCOE))
5705                 count++;
5706
5707         count += TXD_USE_COUNT(skb_headlen(skb));
5708         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5709                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5710
5711         if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
5712                 adapter->tx_busy++;
5713                 return NETDEV_TX_BUSY;
5714         }
5715
5716         first = tx_ring->next_to_use;
5717         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5718 #ifdef IXGBE_FCOE
5719                 /* setup tx offload for FCoE */
5720                 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5721                 if (tso < 0) {
5722                         dev_kfree_skb_any(skb);
5723                         return NETDEV_TX_OK;
5724                 }
5725                 if (tso)
5726                         tx_flags |= IXGBE_TX_FLAGS_FSO;
5727 #endif /* IXGBE_FCOE */
5728         } else {
5729                 if (skb->protocol == htons(ETH_P_IP))
5730                         tx_flags |= IXGBE_TX_FLAGS_IPV4;
5731                 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5732                 if (tso < 0) {
5733                         dev_kfree_skb_any(skb);
5734                         return NETDEV_TX_OK;
5735                 }
5736
5737                 if (tso)
5738                         tx_flags |= IXGBE_TX_FLAGS_TSO;
5739                 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
5740                          (skb->ip_summed == CHECKSUM_PARTIAL))
5741                         tx_flags |= IXGBE_TX_FLAGS_CSUM;
5742         }
5743
5744         count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
5745         if (count) {
5746                 /* add the ATR filter if ATR is on */
5747                 if (tx_ring->atr_sample_rate) {
5748                         ++tx_ring->atr_count;
5749                         if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
5750                              test_bit(__IXGBE_FDIR_INIT_DONE,
5751                                       &tx_ring->reinit_state)) {
5752                                 ixgbe_atr(adapter, skb, tx_ring->queue_index,
5753                                           tx_flags);
5754                                 tx_ring->atr_count = 0;
5755                         }
5756                 }
5757                 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
5758                 txq->tx_bytes += skb->len;
5759                 txq->tx_packets++;
5760                 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
5761                                hdr_len);
5762                 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
5763
5764         } else {
5765                 dev_kfree_skb_any(skb);
5766                 tx_ring->tx_buffer_info[first].time_stamp = 0;
5767                 tx_ring->next_to_use = first;
5768         }
5769
5770         return NETDEV_TX_OK;
5771 }
5772
5773 /**
5774  * ixgbe_set_mac - Change the Ethernet Address of the NIC
5775  * @netdev: network interface device structure
5776  * @p: pointer to an address structure
5777  *
5778  * Returns 0 on success, negative on failure
5779  **/
5780 static int ixgbe_set_mac(struct net_device *netdev, void *p)
5781 {
5782         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5783         struct ixgbe_hw *hw = &adapter->hw;
5784         struct sockaddr *addr = p;
5785
5786         if (!is_valid_ether_addr(addr->sa_data))
5787                 return -EADDRNOTAVAIL;
5788
5789         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
5790         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5791
5792         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
5793                             IXGBE_RAH_AV);
5794
5795         return 0;
5796 }
5797
5798 static int
5799 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
5800 {
5801         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5802         struct ixgbe_hw *hw = &adapter->hw;
5803         u16 value;
5804         int rc;
5805
5806         if (prtad != hw->phy.mdio.prtad)
5807                 return -EINVAL;
5808         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
5809         if (!rc)
5810                 rc = value;
5811         return rc;
5812 }
5813
<