Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / drivers / net / ixgbe / ixgbe_main.c
1 /*******************************************************************************
2
3   Intel 10 Gigabit PCI Express Linux driver
4   Copyright(c) 1999 - 2009 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
34 #include <linux/in.h>
35 #include <linux/ip.h>
36 #include <linux/tcp.h>
37 #include <linux/ipv6.h>
38 #include <net/checksum.h>
39 #include <net/ip6_checksum.h>
40 #include <linux/ethtool.h>
41 #include <linux/if_vlan.h>
42
43 #include "ixgbe.h"
44 #include "ixgbe_common.h"
45
46 char ixgbe_driver_name[] = "ixgbe";
47 static const char ixgbe_driver_string[] =
48                               "Intel(R) 10 Gigabit PCI Express Network Driver";
49
50 #define DRV_VERSION "2.0.8-k2"
51 const char ixgbe_driver_version[] = DRV_VERSION;
52 static char ixgbe_copyright[] = "Copyright (c) 1999-2009 Intel Corporation.";
53
54 static const struct ixgbe_info *ixgbe_info_tbl[] = {
55         [board_82598] = &ixgbe_82598_info,
56         [board_82599] = &ixgbe_82599_info,
57 };
58
59 /* ixgbe_pci_tbl - PCI Device ID Table
60  *
61  * Wildcard entries (PCI_ANY_ID) should come last
62  * Last entry must be all 0s
63  *
64  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
65  *   Class, Class Mask, private data (not used) }
66  */
67 static struct pci_device_id ixgbe_pci_tbl[] = {
68         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
69          board_82598 },
70         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
71          board_82598 },
72         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
73          board_82598 },
74         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
75          board_82598 },
76         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
77          board_82598 },
78         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
79          board_82598 },
80         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
81          board_82598 },
82         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
83          board_82598 },
84         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
85          board_82598 },
86         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
87          board_82598 },
88         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
89          board_82598 },
90         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
91          board_82599 },
92         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
93          board_82599 },
94
95         /* required last entry */
96         {0, }
97 };
98 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
99
100 #ifdef CONFIG_IXGBE_DCA
101 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
102                             void *p);
103 static struct notifier_block dca_notifier = {
104         .notifier_call = ixgbe_notify_dca,
105         .next          = NULL,
106         .priority      = 0
107 };
108 #endif
109
110 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
111 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
112 MODULE_LICENSE("GPL");
113 MODULE_VERSION(DRV_VERSION);
114
115 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
116
117 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
118 {
119         u32 ctrl_ext;
120
121         /* Let firmware take over control of h/w */
122         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
123         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
124                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
125 }
126
127 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
128 {
129         u32 ctrl_ext;
130
131         /* Let firmware know the driver has taken over */
132         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
133         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
134                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
135 }
136
137 /*
138  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
139  * @adapter: pointer to adapter struct
140  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
141  * @queue: queue to map the corresponding interrupt to
142  * @msix_vector: the vector to map to the corresponding queue
143  *
144  */
145 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
146                            u8 queue, u8 msix_vector)
147 {
148         u32 ivar, index;
149         struct ixgbe_hw *hw = &adapter->hw;
150         switch (hw->mac.type) {
151         case ixgbe_mac_82598EB:
152                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
153                 if (direction == -1)
154                         direction = 0;
155                 index = (((direction * 64) + queue) >> 2) & 0x1F;
156                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
157                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
158                 ivar |= (msix_vector << (8 * (queue & 0x3)));
159                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
160                 break;
161         case ixgbe_mac_82599EB:
162                 if (direction == -1) {
163                         /* other causes */
164                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
165                         index = ((queue & 1) * 8);
166                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
167                         ivar &= ~(0xFF << index);
168                         ivar |= (msix_vector << index);
169                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
170                         break;
171                 } else {
172                         /* tx or rx causes */
173                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
174                         index = ((16 * (queue & 1)) + (8 * direction));
175                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
176                         ivar &= ~(0xFF << index);
177                         ivar |= (msix_vector << index);
178                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
179                         break;
180                 }
181         default:
182                 break;
183         }
184 }
185
186 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
187                                              struct ixgbe_tx_buffer
188                                              *tx_buffer_info)
189 {
190         if (tx_buffer_info->dma) {
191                 pci_unmap_page(adapter->pdev, tx_buffer_info->dma,
192                                tx_buffer_info->length, PCI_DMA_TODEVICE);
193                 tx_buffer_info->dma = 0;
194         }
195         if (tx_buffer_info->skb) {
196                 dev_kfree_skb_any(tx_buffer_info->skb);
197                 tx_buffer_info->skb = NULL;
198         }
199         /* tx_buffer_info must be completely set up in the transmit path */
200 }
201
202 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
203                                        struct ixgbe_ring *tx_ring,
204                                        unsigned int eop)
205 {
206         struct ixgbe_hw *hw = &adapter->hw;
207         u32 head, tail;
208
209         /* Detect a transmit hang in hardware, this serializes the
210          * check with the clearing of time_stamp and movement of eop */
211         head = IXGBE_READ_REG(hw, tx_ring->head);
212         tail = IXGBE_READ_REG(hw, tx_ring->tail);
213         adapter->detect_tx_hung = false;
214         if ((head != tail) &&
215             tx_ring->tx_buffer_info[eop].time_stamp &&
216             time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
217             !(IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & IXGBE_TFCS_TXOFF)) {
218                 /* detected Tx unit hang */
219                 union ixgbe_adv_tx_desc *tx_desc;
220                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
221                 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
222                         "  Tx Queue             <%d>\n"
223                         "  TDH, TDT             <%x>, <%x>\n"
224                         "  next_to_use          <%x>\n"
225                         "  next_to_clean        <%x>\n"
226                         "tx_buffer_info[next_to_clean]\n"
227                         "  time_stamp           <%lx>\n"
228                         "  jiffies              <%lx>\n",
229                         tx_ring->queue_index,
230                         head, tail,
231                         tx_ring->next_to_use, eop,
232                         tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
233                 return true;
234         }
235
236         return false;
237 }
238
239 #define IXGBE_MAX_TXD_PWR       14
240 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
241
242 /* Tx Descriptors needed, worst case */
243 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
244                          (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
245 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
246         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
247
248 static void ixgbe_tx_timeout(struct net_device *netdev);
249
250 /**
251  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
252  * @adapter: board private structure
253  * @tx_ring: tx ring to clean
254  *
255  * returns true if transmit work is done
256  **/
257 static bool ixgbe_clean_tx_irq(struct ixgbe_adapter *adapter,
258                                struct ixgbe_ring *tx_ring)
259 {
260         struct net_device *netdev = adapter->netdev;
261         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
262         struct ixgbe_tx_buffer *tx_buffer_info;
263         unsigned int i, eop, count = 0;
264         unsigned int total_bytes = 0, total_packets = 0;
265
266         i = tx_ring->next_to_clean;
267         eop = tx_ring->tx_buffer_info[i].next_to_watch;
268         eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
269
270         while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
271                (count < tx_ring->work_limit)) {
272                 bool cleaned = false;
273                 for ( ; !cleaned; count++) {
274                         struct sk_buff *skb;
275                         tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
276                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
277                         cleaned = (i == eop);
278                         skb = tx_buffer_info->skb;
279
280                         if (cleaned && skb) {
281                                 unsigned int segs, bytecount;
282
283                                 /* gso_segs is currently only valid for tcp */
284                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
285                                 /* multiply data chunks by size of headers */
286                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
287                                             skb->len;
288                                 total_packets += segs;
289                                 total_bytes += bytecount;
290                         }
291
292                         ixgbe_unmap_and_free_tx_resource(adapter,
293                                                          tx_buffer_info);
294
295                         tx_desc->wb.status = 0;
296
297                         i++;
298                         if (i == tx_ring->count)
299                                 i = 0;
300                 }
301
302                 eop = tx_ring->tx_buffer_info[i].next_to_watch;
303                 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
304         }
305
306         tx_ring->next_to_clean = i;
307
308 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
309         if (unlikely(count && netif_carrier_ok(netdev) &&
310                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
311                 /* Make sure that anybody stopping the queue after this
312                  * sees the new next_to_clean.
313                  */
314                 smp_mb();
315                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
316                     !test_bit(__IXGBE_DOWN, &adapter->state)) {
317                         netif_wake_subqueue(netdev, tx_ring->queue_index);
318                         ++adapter->restart_queue;
319                 }
320         }
321
322         if (adapter->detect_tx_hung) {
323                 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
324                         /* schedule immediate reset if we believe we hung */
325                         DPRINTK(PROBE, INFO,
326                                 "tx hang %d detected, resetting adapter\n",
327                                 adapter->tx_timeout_count + 1);
328                         ixgbe_tx_timeout(adapter->netdev);
329                 }
330         }
331
332         /* re-arm the interrupt */
333         if (count >= tx_ring->work_limit)
334                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, tx_ring->v_idx);
335
336         tx_ring->total_bytes += total_bytes;
337         tx_ring->total_packets += total_packets;
338         tx_ring->stats.packets += total_packets;
339         tx_ring->stats.bytes += total_bytes;
340         adapter->net_stats.tx_bytes += total_bytes;
341         adapter->net_stats.tx_packets += total_packets;
342         return (count < tx_ring->work_limit);
343 }
344
345 #ifdef CONFIG_IXGBE_DCA
346 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
347                                 struct ixgbe_ring *rx_ring)
348 {
349         u32 rxctrl;
350         int cpu = get_cpu();
351         int q = rx_ring - adapter->rx_ring;
352
353         if (rx_ring->cpu != cpu) {
354                 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
355                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
356                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
357                         rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
358                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
359                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
360                         rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
361                                    IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
362                 }
363                 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
364                 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
365                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
366                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
367                             IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
368                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
369                 rx_ring->cpu = cpu;
370         }
371         put_cpu();
372 }
373
374 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
375                                 struct ixgbe_ring *tx_ring)
376 {
377         u32 txctrl;
378         int cpu = get_cpu();
379         int q = tx_ring - adapter->tx_ring;
380
381         if (tx_ring->cpu != cpu) {
382                 txctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q));
383                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
384                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
385                         txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
386                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
387                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
388                         txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
389                                    IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
390                 }
391                 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
392                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q), txctrl);
393                 tx_ring->cpu = cpu;
394         }
395         put_cpu();
396 }
397
398 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
399 {
400         int i;
401
402         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
403                 return;
404
405         for (i = 0; i < adapter->num_tx_queues; i++) {
406                 adapter->tx_ring[i].cpu = -1;
407                 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
408         }
409         for (i = 0; i < adapter->num_rx_queues; i++) {
410                 adapter->rx_ring[i].cpu = -1;
411                 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
412         }
413 }
414
415 static int __ixgbe_notify_dca(struct device *dev, void *data)
416 {
417         struct net_device *netdev = dev_get_drvdata(dev);
418         struct ixgbe_adapter *adapter = netdev_priv(netdev);
419         unsigned long event = *(unsigned long *)data;
420
421         switch (event) {
422         case DCA_PROVIDER_ADD:
423                 /* if we're already enabled, don't do it again */
424                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
425                         break;
426                 /* Always use CB2 mode, difference is masked
427                  * in the CB driver. */
428                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
429                 if (dca_add_requester(dev) == 0) {
430                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
431                         ixgbe_setup_dca(adapter);
432                         break;
433                 }
434                 /* Fall Through since DCA is disabled. */
435         case DCA_PROVIDER_REMOVE:
436                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
437                         dca_remove_requester(dev);
438                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
439                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
440                 }
441                 break;
442         }
443
444         return 0;
445 }
446
447 #endif /* CONFIG_IXGBE_DCA */
448 /**
449  * ixgbe_receive_skb - Send a completed packet up the stack
450  * @adapter: board private structure
451  * @skb: packet to send up
452  * @status: hardware indication of status of receive
453  * @rx_ring: rx descriptor ring (for a specific queue) to setup
454  * @rx_desc: rx descriptor
455  **/
456 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
457                               struct sk_buff *skb, u8 status,
458                               union ixgbe_adv_rx_desc *rx_desc)
459 {
460         struct ixgbe_adapter *adapter = q_vector->adapter;
461         struct napi_struct *napi = &q_vector->napi;
462         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
463         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
464
465         skb_record_rx_queue(skb, q_vector - &adapter->q_vector[0]);
466         if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
467                 if (adapter->vlgrp && is_vlan && (tag != 0))
468                         vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
469                 else
470                         napi_gro_receive(napi, skb);
471         } else {
472                 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
473                         if (adapter->vlgrp && is_vlan && (tag != 0))
474                                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp, tag);
475                         else
476                                 netif_receive_skb(skb);
477                 } else {
478                         if (adapter->vlgrp && is_vlan && (tag != 0))
479                                 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
480                         else
481                                 netif_rx(skb);
482                 }
483         }
484 }
485
486 /**
487  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
488  * @adapter: address of board private structure
489  * @status_err: hardware indication of status of receive
490  * @skb: skb currently being received and modified
491  **/
492 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
493                                      u32 status_err, struct sk_buff *skb)
494 {
495         skb->ip_summed = CHECKSUM_NONE;
496
497         /* Rx csum disabled */
498         if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
499                 return;
500
501         /* if IP and error */
502         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
503             (status_err & IXGBE_RXDADV_ERR_IPE)) {
504                 adapter->hw_csum_rx_error++;
505                 return;
506         }
507
508         if (!(status_err & IXGBE_RXD_STAT_L4CS))
509                 return;
510
511         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
512                 adapter->hw_csum_rx_error++;
513                 return;
514         }
515
516         /* It must be a TCP or UDP packet with a valid checksum */
517         skb->ip_summed = CHECKSUM_UNNECESSARY;
518         adapter->hw_csum_rx_good++;
519 }
520
521 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
522                                          struct ixgbe_ring *rx_ring, u32 val)
523 {
524         /*
525          * Force memory writes to complete before letting h/w
526          * know there are new descriptors to fetch.  (Only
527          * applicable for weak-ordered memory model archs,
528          * such as IA-64).
529          */
530         wmb();
531         IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
532 }
533
534 /**
535  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
536  * @adapter: address of board private structure
537  **/
538 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
539                                    struct ixgbe_ring *rx_ring,
540                                    int cleaned_count)
541 {
542         struct pci_dev *pdev = adapter->pdev;
543         union ixgbe_adv_rx_desc *rx_desc;
544         struct ixgbe_rx_buffer *bi;
545         unsigned int i;
546         unsigned int bufsz = rx_ring->rx_buf_len + NET_IP_ALIGN;
547
548         i = rx_ring->next_to_use;
549         bi = &rx_ring->rx_buffer_info[i];
550
551         while (cleaned_count--) {
552                 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
553
554                 if (!bi->page_dma &&
555                     (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
556                         if (!bi->page) {
557                                 bi->page = alloc_page(GFP_ATOMIC);
558                                 if (!bi->page) {
559                                         adapter->alloc_rx_page_failed++;
560                                         goto no_buffers;
561                                 }
562                                 bi->page_offset = 0;
563                         } else {
564                                 /* use a half page if we're re-using */
565                                 bi->page_offset ^= (PAGE_SIZE / 2);
566                         }
567
568                         bi->page_dma = pci_map_page(pdev, bi->page,
569                                                     bi->page_offset,
570                                                     (PAGE_SIZE / 2),
571                                                     PCI_DMA_FROMDEVICE);
572                 }
573
574                 if (!bi->skb) {
575                         struct sk_buff *skb;
576                         skb = netdev_alloc_skb(adapter->netdev, bufsz);
577
578                         if (!skb) {
579                                 adapter->alloc_rx_buff_failed++;
580                                 goto no_buffers;
581                         }
582
583                         /*
584                          * Make buffer alignment 2 beyond a 16 byte boundary
585                          * this will result in a 16 byte aligned IP header after
586                          * the 14 byte MAC header is removed
587                          */
588                         skb_reserve(skb, NET_IP_ALIGN);
589
590                         bi->skb = skb;
591                         bi->dma = pci_map_single(pdev, skb->data, bufsz,
592                                                  PCI_DMA_FROMDEVICE);
593                 }
594                 /* Refresh the desc even if buffer_addrs didn't change because
595                  * each write-back erases this info. */
596                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
597                         rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
598                         rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
599                 } else {
600                         rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
601                 }
602
603                 i++;
604                 if (i == rx_ring->count)
605                         i = 0;
606                 bi = &rx_ring->rx_buffer_info[i];
607         }
608
609 no_buffers:
610         if (rx_ring->next_to_use != i) {
611                 rx_ring->next_to_use = i;
612                 if (i-- == 0)
613                         i = (rx_ring->count - 1);
614
615                 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
616         }
617 }
618
619 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
620 {
621         return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
622 }
623
624 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
625 {
626         return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
627 }
628
629 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
630                                struct ixgbe_ring *rx_ring,
631                                int *work_done, int work_to_do)
632 {
633         struct ixgbe_adapter *adapter = q_vector->adapter;
634         struct pci_dev *pdev = adapter->pdev;
635         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
636         struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
637         struct sk_buff *skb;
638         unsigned int i;
639         u32 len, staterr;
640         u16 hdr_info;
641         bool cleaned = false;
642         int cleaned_count = 0;
643         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
644
645         i = rx_ring->next_to_clean;
646         rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
647         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
648         rx_buffer_info = &rx_ring->rx_buffer_info[i];
649
650         while (staterr & IXGBE_RXD_STAT_DD) {
651                 u32 upper_len = 0;
652                 if (*work_done >= work_to_do)
653                         break;
654                 (*work_done)++;
655
656                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
657                         hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
658                         len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
659                                IXGBE_RXDADV_HDRBUFLEN_SHIFT;
660                         if (hdr_info & IXGBE_RXDADV_SPH)
661                                 adapter->rx_hdr_split++;
662                         if (len > IXGBE_RX_HDR_SIZE)
663                                 len = IXGBE_RX_HDR_SIZE;
664                         upper_len = le16_to_cpu(rx_desc->wb.upper.length);
665                 } else {
666                         len = le16_to_cpu(rx_desc->wb.upper.length);
667                 }
668
669                 cleaned = true;
670                 skb = rx_buffer_info->skb;
671                 prefetch(skb->data - NET_IP_ALIGN);
672                 rx_buffer_info->skb = NULL;
673
674                 if (len && !skb_shinfo(skb)->nr_frags) {
675                         pci_unmap_single(pdev, rx_buffer_info->dma,
676                                          rx_ring->rx_buf_len,
677                                          PCI_DMA_FROMDEVICE);
678                         skb_put(skb, len);
679                 }
680
681                 if (upper_len) {
682                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
683                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
684                         rx_buffer_info->page_dma = 0;
685                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
686                                            rx_buffer_info->page,
687                                            rx_buffer_info->page_offset,
688                                            upper_len);
689
690                         if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
691                             (page_count(rx_buffer_info->page) != 1))
692                                 rx_buffer_info->page = NULL;
693                         else
694                                 get_page(rx_buffer_info->page);
695
696                         skb->len += upper_len;
697                         skb->data_len += upper_len;
698                         skb->truesize += upper_len;
699                 }
700
701                 i++;
702                 if (i == rx_ring->count)
703                         i = 0;
704                 next_buffer = &rx_ring->rx_buffer_info[i];
705
706                 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
707                 prefetch(next_rxd);
708
709                 cleaned_count++;
710                 if (staterr & IXGBE_RXD_STAT_EOP) {
711                         rx_ring->stats.packets++;
712                         rx_ring->stats.bytes += skb->len;
713                 } else {
714                         rx_buffer_info->skb = next_buffer->skb;
715                         rx_buffer_info->dma = next_buffer->dma;
716                         next_buffer->skb = skb;
717                         next_buffer->dma = 0;
718                         adapter->non_eop_descs++;
719                         goto next_desc;
720                 }
721
722                 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
723                         dev_kfree_skb_irq(skb);
724                         goto next_desc;
725                 }
726
727                 ixgbe_rx_checksum(adapter, staterr, skb);
728
729                 /* probably a little skewed due to removing CRC */
730                 total_rx_bytes += skb->len;
731                 total_rx_packets++;
732
733                 skb->protocol = eth_type_trans(skb, adapter->netdev);
734                 ixgbe_receive_skb(q_vector, skb, staterr, rx_desc);
735
736 next_desc:
737                 rx_desc->wb.upper.status_error = 0;
738
739                 /* return some buffers to hardware, one at a time is too slow */
740                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
741                         ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
742                         cleaned_count = 0;
743                 }
744
745                 /* use prefetched values */
746                 rx_desc = next_rxd;
747                 rx_buffer_info = next_buffer;
748
749                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
750         }
751
752         rx_ring->next_to_clean = i;
753         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
754
755         if (cleaned_count)
756                 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
757
758         rx_ring->total_packets += total_rx_packets;
759         rx_ring->total_bytes += total_rx_bytes;
760         adapter->net_stats.rx_bytes += total_rx_bytes;
761         adapter->net_stats.rx_packets += total_rx_packets;
762
763         return cleaned;
764 }
765
766 static int ixgbe_clean_rxonly(struct napi_struct *, int);
767 /**
768  * ixgbe_configure_msix - Configure MSI-X hardware
769  * @adapter: board private structure
770  *
771  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
772  * interrupts.
773  **/
774 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
775 {
776         struct ixgbe_q_vector *q_vector;
777         int i, j, q_vectors, v_idx, r_idx;
778         u32 mask;
779
780         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
781
782         /*
783          * Populate the IVAR table and set the ITR values to the
784          * corresponding register.
785          */
786         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
787                 q_vector = &adapter->q_vector[v_idx];
788                 /* XXX for_each_bit(...) */
789                 r_idx = find_first_bit(q_vector->rxr_idx,
790                                        adapter->num_rx_queues);
791
792                 for (i = 0; i < q_vector->rxr_count; i++) {
793                         j = adapter->rx_ring[r_idx].reg_idx;
794                         ixgbe_set_ivar(adapter, 0, j, v_idx);
795                         r_idx = find_next_bit(q_vector->rxr_idx,
796                                               adapter->num_rx_queues,
797                                               r_idx + 1);
798                 }
799                 r_idx = find_first_bit(q_vector->txr_idx,
800                                        adapter->num_tx_queues);
801
802                 for (i = 0; i < q_vector->txr_count; i++) {
803                         j = adapter->tx_ring[r_idx].reg_idx;
804                         ixgbe_set_ivar(adapter, 1, j, v_idx);
805                         r_idx = find_next_bit(q_vector->txr_idx,
806                                               adapter->num_tx_queues,
807                                               r_idx + 1);
808                 }
809
810                 /* if this is a tx only vector halve the interrupt rate */
811                 if (q_vector->txr_count && !q_vector->rxr_count)
812                         q_vector->eitr = (adapter->eitr_param >> 1);
813                 else if (q_vector->rxr_count)
814                         /* rx only */
815                         q_vector->eitr = adapter->eitr_param;
816
817                 /*
818                  * since this is initial set up don't need to call
819                  * ixgbe_write_eitr helper
820                  */
821                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx),
822                                 EITR_INTS_PER_SEC_TO_REG(q_vector->eitr));
823         }
824
825         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
826                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
827                                v_idx);
828         else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
829                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
830         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
831
832         /* set up to autoclear timer, and the vectors */
833         mask = IXGBE_EIMS_ENABLE_MASK;
834         mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
835         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
836 }
837
838 enum latency_range {
839         lowest_latency = 0,
840         low_latency = 1,
841         bulk_latency = 2,
842         latency_invalid = 255
843 };
844
845 /**
846  * ixgbe_update_itr - update the dynamic ITR value based on statistics
847  * @adapter: pointer to adapter
848  * @eitr: eitr setting (ints per sec) to give last timeslice
849  * @itr_setting: current throttle rate in ints/second
850  * @packets: the number of packets during this measurement interval
851  * @bytes: the number of bytes during this measurement interval
852  *
853  *      Stores a new ITR value based on packets and byte
854  *      counts during the last interrupt.  The advantage of per interrupt
855  *      computation is faster updates and more accurate ITR for the current
856  *      traffic pattern.  Constants in this function were computed
857  *      based on theoretical maximum wire speed and thresholds were set based
858  *      on testing data as well as attempting to minimize response time
859  *      while increasing bulk throughput.
860  *      this functionality is controlled by the InterruptThrottleRate module
861  *      parameter (see ixgbe_param.c)
862  **/
863 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
864                            u32 eitr, u8 itr_setting,
865                            int packets, int bytes)
866 {
867         unsigned int retval = itr_setting;
868         u32 timepassed_us;
869         u64 bytes_perint;
870
871         if (packets == 0)
872                 goto update_itr_done;
873
874
875         /* simple throttlerate management
876          *    0-20MB/s lowest (100000 ints/s)
877          *   20-100MB/s low   (20000 ints/s)
878          *  100-1249MB/s bulk (8000 ints/s)
879          */
880         /* what was last interrupt timeslice? */
881         timepassed_us = 1000000/eitr;
882         bytes_perint = bytes / timepassed_us; /* bytes/usec */
883
884         switch (itr_setting) {
885         case lowest_latency:
886                 if (bytes_perint > adapter->eitr_low)
887                         retval = low_latency;
888                 break;
889         case low_latency:
890                 if (bytes_perint > adapter->eitr_high)
891                         retval = bulk_latency;
892                 else if (bytes_perint <= adapter->eitr_low)
893                         retval = lowest_latency;
894                 break;
895         case bulk_latency:
896                 if (bytes_perint <= adapter->eitr_high)
897                         retval = low_latency;
898                 break;
899         }
900
901 update_itr_done:
902         return retval;
903 }
904
905 /**
906  * ixgbe_write_eitr - write EITR register in hardware specific way
907  * @adapter: pointer to adapter struct
908  * @v_idx: vector index into q_vector array
909  * @itr_reg: new value to be written in *register* format, not ints/s
910  *
911  * This function is made to be called by ethtool and by the driver
912  * when it needs to update EITR registers at runtime.  Hardware
913  * specific quirks/differences are taken care of here.
914  */
915 void ixgbe_write_eitr(struct ixgbe_adapter *adapter, int v_idx, u32 itr_reg)
916 {
917         struct ixgbe_hw *hw = &adapter->hw;
918         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
919                 /* must write high and low 16 bits to reset counter */
920                 itr_reg |= (itr_reg << 16);
921         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
922                 /*
923                  * set the WDIS bit to not clear the timer bits and cause an
924                  * immediate assertion of the interrupt
925                  */
926                 itr_reg |= IXGBE_EITR_CNT_WDIS;
927         }
928         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
929 }
930
931 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
932 {
933         struct ixgbe_adapter *adapter = q_vector->adapter;
934         u32 new_itr;
935         u8 current_itr, ret_itr;
936         int i, r_idx, v_idx = ((void *)q_vector - (void *)(adapter->q_vector)) /
937                                sizeof(struct ixgbe_q_vector);
938         struct ixgbe_ring *rx_ring, *tx_ring;
939
940         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
941         for (i = 0; i < q_vector->txr_count; i++) {
942                 tx_ring = &(adapter->tx_ring[r_idx]);
943                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
944                                            q_vector->tx_itr,
945                                            tx_ring->total_packets,
946                                            tx_ring->total_bytes);
947                 /* if the result for this queue would decrease interrupt
948                  * rate for this vector then use that result */
949                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
950                                     q_vector->tx_itr - 1 : ret_itr);
951                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
952                                       r_idx + 1);
953         }
954
955         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
956         for (i = 0; i < q_vector->rxr_count; i++) {
957                 rx_ring = &(adapter->rx_ring[r_idx]);
958                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
959                                            q_vector->rx_itr,
960                                            rx_ring->total_packets,
961                                            rx_ring->total_bytes);
962                 /* if the result for this queue would decrease interrupt
963                  * rate for this vector then use that result */
964                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
965                                     q_vector->rx_itr - 1 : ret_itr);
966                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
967                                       r_idx + 1);
968         }
969
970         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
971
972         switch (current_itr) {
973         /* counts and packets in update_itr are dependent on these numbers */
974         case lowest_latency:
975                 new_itr = 100000;
976                 break;
977         case low_latency:
978                 new_itr = 20000; /* aka hwitr = ~200 */
979                 break;
980         case bulk_latency:
981         default:
982                 new_itr = 8000;
983                 break;
984         }
985
986         if (new_itr != q_vector->eitr) {
987                 u32 itr_reg;
988
989                 /* save the algorithm value here, not the smoothed one */
990                 q_vector->eitr = new_itr;
991                 /* do an exponential smoothing */
992                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
993                 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
994                 ixgbe_write_eitr(adapter, v_idx, itr_reg);
995         }
996
997         return;
998 }
999
1000 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1001 {
1002         struct ixgbe_hw *hw = &adapter->hw;
1003
1004         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1005             (eicr & IXGBE_EICR_GPI_SDP1)) {
1006                 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
1007                 /* write to clear the interrupt */
1008                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1009         }
1010 }
1011
1012 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1013 {
1014         struct ixgbe_hw *hw = &adapter->hw;
1015
1016         if (eicr & IXGBE_EICR_GPI_SDP1) {
1017                 /* Clear the interrupt */
1018                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1019                 schedule_work(&adapter->multispeed_fiber_task);
1020         } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1021                 /* Clear the interrupt */
1022                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1023                 schedule_work(&adapter->sfp_config_module_task);
1024         } else {
1025                 /* Interrupt isn't for us... */
1026                 return;
1027         }
1028 }
1029
1030 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1031 {
1032         struct ixgbe_hw *hw = &adapter->hw;
1033
1034         adapter->lsc_int++;
1035         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1036         adapter->link_check_timeout = jiffies;
1037         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1038                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1039                 schedule_work(&adapter->watchdog_task);
1040         }
1041 }
1042
1043 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1044 {
1045         struct net_device *netdev = data;
1046         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1047         struct ixgbe_hw *hw = &adapter->hw;
1048         u32 eicr;
1049
1050         /*
1051          * Workaround for Silicon errata.  Use clear-by-write instead
1052          * of clear-by-read.  Reading with EICS will return the
1053          * interrupt causes without clearing, which later be done
1054          * with the write to EICR.
1055          */
1056         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1057         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1058
1059         if (eicr & IXGBE_EICR_LSC)
1060                 ixgbe_check_lsc(adapter);
1061
1062         if (hw->mac.type == ixgbe_mac_82598EB)
1063                 ixgbe_check_fan_failure(adapter, eicr);
1064
1065         if (hw->mac.type == ixgbe_mac_82599EB)
1066                 ixgbe_check_sfp_event(adapter, eicr);
1067         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1068                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1069
1070         return IRQ_HANDLED;
1071 }
1072
1073 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1074 {
1075         struct ixgbe_q_vector *q_vector = data;
1076         struct ixgbe_adapter  *adapter = q_vector->adapter;
1077         struct ixgbe_ring     *tx_ring;
1078         int i, r_idx;
1079
1080         if (!q_vector->txr_count)
1081                 return IRQ_HANDLED;
1082
1083         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1084         for (i = 0; i < q_vector->txr_count; i++) {
1085                 tx_ring = &(adapter->tx_ring[r_idx]);
1086 #ifdef CONFIG_IXGBE_DCA
1087                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1088                         ixgbe_update_tx_dca(adapter, tx_ring);
1089 #endif
1090                 tx_ring->total_bytes = 0;
1091                 tx_ring->total_packets = 0;
1092                 ixgbe_clean_tx_irq(adapter, tx_ring);
1093                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1094                                       r_idx + 1);
1095         }
1096
1097         return IRQ_HANDLED;
1098 }
1099
1100 /**
1101  * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1102  * @irq: unused
1103  * @data: pointer to our q_vector struct for this interrupt vector
1104  **/
1105 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1106 {
1107         struct ixgbe_q_vector *q_vector = data;
1108         struct ixgbe_adapter  *adapter = q_vector->adapter;
1109         struct ixgbe_ring  *rx_ring;
1110         int r_idx;
1111         int i;
1112
1113         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1114         for (i = 0;  i < q_vector->rxr_count; i++) {
1115                 rx_ring = &(adapter->rx_ring[r_idx]);
1116                 rx_ring->total_bytes = 0;
1117                 rx_ring->total_packets = 0;
1118                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1119                                       r_idx + 1);
1120         }
1121
1122         if (!q_vector->rxr_count)
1123                 return IRQ_HANDLED;
1124
1125         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1126         rx_ring = &(adapter->rx_ring[r_idx]);
1127         /* disable interrupts on this vector only */
1128         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, rx_ring->v_idx);
1129         napi_schedule(&q_vector->napi);
1130
1131         return IRQ_HANDLED;
1132 }
1133
1134 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1135 {
1136         ixgbe_msix_clean_rx(irq, data);
1137         ixgbe_msix_clean_tx(irq, data);
1138
1139         return IRQ_HANDLED;
1140 }
1141
1142 /**
1143  * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1144  * @napi: napi struct with our devices info in it
1145  * @budget: amount of work driver is allowed to do this pass, in packets
1146  *
1147  * This function is optimized for cleaning one queue only on a single
1148  * q_vector!!!
1149  **/
1150 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1151 {
1152         struct ixgbe_q_vector *q_vector =
1153                                container_of(napi, struct ixgbe_q_vector, napi);
1154         struct ixgbe_adapter *adapter = q_vector->adapter;
1155         struct ixgbe_ring *rx_ring = NULL;
1156         int work_done = 0;
1157         long r_idx;
1158
1159         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1160         rx_ring = &(adapter->rx_ring[r_idx]);
1161 #ifdef CONFIG_IXGBE_DCA
1162         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1163                 ixgbe_update_rx_dca(adapter, rx_ring);
1164 #endif
1165
1166         ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1167
1168         /* If all Rx work done, exit the polling mode */
1169         if (work_done < budget) {
1170                 napi_complete(napi);
1171                 if (adapter->itr_setting & 1)
1172                         ixgbe_set_itr_msix(q_vector);
1173                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1174                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, rx_ring->v_idx);
1175         }
1176
1177         return work_done;
1178 }
1179
1180 /**
1181  * ixgbe_clean_rxonly_many - msix (aka one shot) rx clean routine
1182  * @napi: napi struct with our devices info in it
1183  * @budget: amount of work driver is allowed to do this pass, in packets
1184  *
1185  * This function will clean more than one rx queue associated with a
1186  * q_vector.
1187  **/
1188 static int ixgbe_clean_rxonly_many(struct napi_struct *napi, int budget)
1189 {
1190         struct ixgbe_q_vector *q_vector =
1191                                container_of(napi, struct ixgbe_q_vector, napi);
1192         struct ixgbe_adapter *adapter = q_vector->adapter;
1193         struct ixgbe_ring *rx_ring = NULL;
1194         int work_done = 0, i;
1195         long r_idx;
1196         u16 enable_mask = 0;
1197
1198         /* attempt to distribute budget to each queue fairly, but don't allow
1199          * the budget to go below 1 because we'll exit polling */
1200         budget /= (q_vector->rxr_count ?: 1);
1201         budget = max(budget, 1);
1202         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1203         for (i = 0; i < q_vector->rxr_count; i++) {
1204                 rx_ring = &(adapter->rx_ring[r_idx]);
1205 #ifdef CONFIG_IXGBE_DCA
1206                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1207                         ixgbe_update_rx_dca(adapter, rx_ring);
1208 #endif
1209                 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1210                 enable_mask |= rx_ring->v_idx;
1211                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1212                                       r_idx + 1);
1213         }
1214
1215         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1216         rx_ring = &(adapter->rx_ring[r_idx]);
1217         /* If all Rx work done, exit the polling mode */
1218         if (work_done < budget) {
1219                 napi_complete(napi);
1220                 if (adapter->itr_setting & 1)
1221                         ixgbe_set_itr_msix(q_vector);
1222                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1223                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, enable_mask);
1224                 return 0;
1225         }
1226
1227         return work_done;
1228 }
1229 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1230                                      int r_idx)
1231 {
1232         a->q_vector[v_idx].adapter = a;
1233         set_bit(r_idx, a->q_vector[v_idx].rxr_idx);
1234         a->q_vector[v_idx].rxr_count++;
1235         a->rx_ring[r_idx].v_idx = 1 << v_idx;
1236 }
1237
1238 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1239                                      int r_idx)
1240 {
1241         a->q_vector[v_idx].adapter = a;
1242         set_bit(r_idx, a->q_vector[v_idx].txr_idx);
1243         a->q_vector[v_idx].txr_count++;
1244         a->tx_ring[r_idx].v_idx = 1 << v_idx;
1245 }
1246
1247 /**
1248  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1249  * @adapter: board private structure to initialize
1250  * @vectors: allotted vector count for descriptor rings
1251  *
1252  * This function maps descriptor rings to the queue-specific vectors
1253  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1254  * one vector per ring/queue, but on a constrained vector budget, we
1255  * group the rings as "efficiently" as possible.  You would add new
1256  * mapping configurations in here.
1257  **/
1258 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1259                                       int vectors)
1260 {
1261         int v_start = 0;
1262         int rxr_idx = 0, txr_idx = 0;
1263         int rxr_remaining = adapter->num_rx_queues;
1264         int txr_remaining = adapter->num_tx_queues;
1265         int i, j;
1266         int rqpv, tqpv;
1267         int err = 0;
1268
1269         /* No mapping required if MSI-X is disabled. */
1270         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1271                 goto out;
1272
1273         /*
1274          * The ideal configuration...
1275          * We have enough vectors to map one per queue.
1276          */
1277         if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1278                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1279                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1280
1281                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1282                         map_vector_to_txq(adapter, v_start, txr_idx);
1283
1284                 goto out;
1285         }
1286
1287         /*
1288          * If we don't have enough vectors for a 1-to-1
1289          * mapping, we'll have to group them so there are
1290          * multiple queues per vector.
1291          */
1292         /* Re-adjusting *qpv takes care of the remainder. */
1293         for (i = v_start; i < vectors; i++) {
1294                 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1295                 for (j = 0; j < rqpv; j++) {
1296                         map_vector_to_rxq(adapter, i, rxr_idx);
1297                         rxr_idx++;
1298                         rxr_remaining--;
1299                 }
1300         }
1301         for (i = v_start; i < vectors; i++) {
1302                 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1303                 for (j = 0; j < tqpv; j++) {
1304                         map_vector_to_txq(adapter, i, txr_idx);
1305                         txr_idx++;
1306                         txr_remaining--;
1307                 }
1308         }
1309
1310 out:
1311         return err;
1312 }
1313
1314 /**
1315  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1316  * @adapter: board private structure
1317  *
1318  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1319  * interrupts from the kernel.
1320  **/
1321 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1322 {
1323         struct net_device *netdev = adapter->netdev;
1324         irqreturn_t (*handler)(int, void *);
1325         int i, vector, q_vectors, err;
1326         int ri=0, ti=0;
1327
1328         /* Decrement for Other and TCP Timer vectors */
1329         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1330
1331         /* Map the Tx/Rx rings to the vectors we were allotted. */
1332         err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1333         if (err)
1334                 goto out;
1335
1336 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1337                          (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1338                          &ixgbe_msix_clean_many)
1339         for (vector = 0; vector < q_vectors; vector++) {
1340                 handler = SET_HANDLER(&adapter->q_vector[vector]);
1341
1342                 if(handler == &ixgbe_msix_clean_rx) {
1343                         sprintf(adapter->name[vector], "%s-%s-%d",
1344                                 netdev->name, "rx", ri++);
1345                 }
1346                 else if(handler == &ixgbe_msix_clean_tx) {
1347                         sprintf(adapter->name[vector], "%s-%s-%d",
1348                                 netdev->name, "tx", ti++);
1349                 }
1350                 else
1351                         sprintf(adapter->name[vector], "%s-%s-%d",
1352                                 netdev->name, "TxRx", vector);
1353
1354                 err = request_irq(adapter->msix_entries[vector].vector,
1355                                   handler, 0, adapter->name[vector],
1356                                   &(adapter->q_vector[vector]));
1357                 if (err) {
1358                         DPRINTK(PROBE, ERR,
1359                                 "request_irq failed for MSIX interrupt "
1360                                 "Error: %d\n", err);
1361                         goto free_queue_irqs;
1362                 }
1363         }
1364
1365         sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1366         err = request_irq(adapter->msix_entries[vector].vector,
1367                           &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1368         if (err) {
1369                 DPRINTK(PROBE, ERR,
1370                         "request_irq for msix_lsc failed: %d\n", err);
1371                 goto free_queue_irqs;
1372         }
1373
1374         return 0;
1375
1376 free_queue_irqs:
1377         for (i = vector - 1; i >= 0; i--)
1378                 free_irq(adapter->msix_entries[--vector].vector,
1379                          &(adapter->q_vector[i]));
1380         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1381         pci_disable_msix(adapter->pdev);
1382         kfree(adapter->msix_entries);
1383         adapter->msix_entries = NULL;
1384 out:
1385         return err;
1386 }
1387
1388 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1389 {
1390         struct ixgbe_q_vector *q_vector = adapter->q_vector;
1391         u8 current_itr;
1392         u32 new_itr = q_vector->eitr;
1393         struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1394         struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1395
1396         q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1397                                             q_vector->tx_itr,
1398                                             tx_ring->total_packets,
1399                                             tx_ring->total_bytes);
1400         q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1401                                             q_vector->rx_itr,
1402                                             rx_ring->total_packets,
1403                                             rx_ring->total_bytes);
1404
1405         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1406
1407         switch (current_itr) {
1408         /* counts and packets in update_itr are dependent on these numbers */
1409         case lowest_latency:
1410                 new_itr = 100000;
1411                 break;
1412         case low_latency:
1413                 new_itr = 20000; /* aka hwitr = ~200 */
1414                 break;
1415         case bulk_latency:
1416                 new_itr = 8000;
1417                 break;
1418         default:
1419                 break;
1420         }
1421
1422         if (new_itr != q_vector->eitr) {
1423                 u32 itr_reg;
1424
1425                 /* save the algorithm value here, not the smoothed one */
1426                 q_vector->eitr = new_itr;
1427                 /* do an exponential smoothing */
1428                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1429                 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
1430                 ixgbe_write_eitr(adapter, 0, itr_reg);
1431         }
1432
1433         return;
1434 }
1435
1436 /**
1437  * ixgbe_irq_enable - Enable default interrupt generation settings
1438  * @adapter: board private structure
1439  **/
1440 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1441 {
1442         u32 mask;
1443         mask = IXGBE_EIMS_ENABLE_MASK;
1444         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1445                 mask |= IXGBE_EIMS_GPI_SDP1;
1446         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1447                 mask |= IXGBE_EIMS_ECC;
1448                 mask |= IXGBE_EIMS_GPI_SDP1;
1449                 mask |= IXGBE_EIMS_GPI_SDP2;
1450         }
1451
1452         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1453         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1454                 /* enable the rest of the queue vectors */
1455                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1),
1456                                 (IXGBE_EIMS_RTX_QUEUE << 16));
1457                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(2),
1458                                 ((IXGBE_EIMS_RTX_QUEUE << 16) |
1459                                   IXGBE_EIMS_RTX_QUEUE));
1460         }
1461         IXGBE_WRITE_FLUSH(&adapter->hw);
1462 }
1463
1464 /**
1465  * ixgbe_intr - legacy mode Interrupt Handler
1466  * @irq: interrupt number
1467  * @data: pointer to a network interface device structure
1468  **/
1469 static irqreturn_t ixgbe_intr(int irq, void *data)
1470 {
1471         struct net_device *netdev = data;
1472         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1473         struct ixgbe_hw *hw = &adapter->hw;
1474         u32 eicr;
1475
1476         /*
1477          * Workaround for silicon errata.  Mask the interrupts
1478          * before the read of EICR.
1479          */
1480         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1481
1482         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1483          * therefore no explict interrupt disable is necessary */
1484         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1485         if (!eicr) {
1486                 /* shared interrupt alert!
1487                  * make sure interrupts are enabled because the read will
1488                  * have disabled interrupts due to EIAM */
1489                 ixgbe_irq_enable(adapter);
1490                 return IRQ_NONE;        /* Not our interrupt */
1491         }
1492
1493         if (eicr & IXGBE_EICR_LSC)
1494                 ixgbe_check_lsc(adapter);
1495
1496         if (hw->mac.type == ixgbe_mac_82599EB)
1497                 ixgbe_check_sfp_event(adapter, eicr);
1498
1499         ixgbe_check_fan_failure(adapter, eicr);
1500
1501         if (napi_schedule_prep(&adapter->q_vector[0].napi)) {
1502                 adapter->tx_ring[0].total_packets = 0;
1503                 adapter->tx_ring[0].total_bytes = 0;
1504                 adapter->rx_ring[0].total_packets = 0;
1505                 adapter->rx_ring[0].total_bytes = 0;
1506                 /* would disable interrupts here but EIAM disabled it */
1507                 __napi_schedule(&adapter->q_vector[0].napi);
1508         }
1509
1510         return IRQ_HANDLED;
1511 }
1512
1513 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1514 {
1515         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1516
1517         for (i = 0; i < q_vectors; i++) {
1518                 struct ixgbe_q_vector *q_vector = &adapter->q_vector[i];
1519                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1520                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1521                 q_vector->rxr_count = 0;
1522                 q_vector->txr_count = 0;
1523         }
1524 }
1525
1526 /**
1527  * ixgbe_request_irq - initialize interrupts
1528  * @adapter: board private structure
1529  *
1530  * Attempts to configure interrupts using the best available
1531  * capabilities of the hardware and kernel.
1532  **/
1533 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1534 {
1535         struct net_device *netdev = adapter->netdev;
1536         int err;
1537
1538         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1539                 err = ixgbe_request_msix_irqs(adapter);
1540         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1541                 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1542                                   netdev->name, netdev);
1543         } else {
1544                 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1545                                   netdev->name, netdev);
1546         }
1547
1548         if (err)
1549                 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1550
1551         return err;
1552 }
1553
1554 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1555 {
1556         struct net_device *netdev = adapter->netdev;
1557
1558         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1559                 int i, q_vectors;
1560
1561                 q_vectors = adapter->num_msix_vectors;
1562
1563                 i = q_vectors - 1;
1564                 free_irq(adapter->msix_entries[i].vector, netdev);
1565
1566                 i--;
1567                 for (; i >= 0; i--) {
1568                         free_irq(adapter->msix_entries[i].vector,
1569                                  &(adapter->q_vector[i]));
1570                 }
1571
1572                 ixgbe_reset_q_vectors(adapter);
1573         } else {
1574                 free_irq(adapter->pdev->irq, netdev);
1575         }
1576 }
1577
1578 /**
1579  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1580  * @adapter: board private structure
1581  **/
1582 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1583 {
1584         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1585         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1586                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
1587                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(2), ~0);
1588         }
1589         IXGBE_WRITE_FLUSH(&adapter->hw);
1590         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1591                 int i;
1592                 for (i = 0; i < adapter->num_msix_vectors; i++)
1593                         synchronize_irq(adapter->msix_entries[i].vector);
1594         } else {
1595                 synchronize_irq(adapter->pdev->irq);
1596         }
1597 }
1598
1599 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter)
1600 {
1601         u32 mask = IXGBE_EIMS_RTX_QUEUE;
1602         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1603         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1604                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask << 16);
1605                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(2),
1606                                 (mask << 16 | mask));
1607         }
1608         /* skip the flush */
1609 }
1610
1611 /**
1612  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1613  *
1614  **/
1615 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1616 {
1617         struct ixgbe_hw *hw = &adapter->hw;
1618
1619         IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1620                         EITR_INTS_PER_SEC_TO_REG(adapter->eitr_param));
1621
1622         ixgbe_set_ivar(adapter, 0, 0, 0);
1623         ixgbe_set_ivar(adapter, 1, 0, 0);
1624
1625         map_vector_to_rxq(adapter, 0, 0);
1626         map_vector_to_txq(adapter, 0, 0);
1627
1628         DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1629 }
1630
1631 /**
1632  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1633  * @adapter: board private structure
1634  *
1635  * Configure the Tx unit of the MAC after a reset.
1636  **/
1637 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1638 {
1639         u64 tdba;
1640         struct ixgbe_hw *hw = &adapter->hw;
1641         u32 i, j, tdlen, txctrl;
1642
1643         /* Setup the HW Tx Head and Tail descriptor pointers */
1644         for (i = 0; i < adapter->num_tx_queues; i++) {
1645                 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1646                 j = ring->reg_idx;
1647                 tdba = ring->dma;
1648                 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1649                 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1650                                 (tdba & DMA_32BIT_MASK));
1651                 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1652                 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1653                 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1654                 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1655                 adapter->tx_ring[i].head = IXGBE_TDH(j);
1656                 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1657                 /* Disable Tx Head Writeback RO bit, since this hoses
1658                  * bookkeeping if things aren't delivered in order.
1659                  */
1660                 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1661                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1662                 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1663         }
1664         if (hw->mac.type == ixgbe_mac_82599EB) {
1665                 /* We enable 8 traffic classes, DCB only */
1666                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
1667                         IXGBE_WRITE_REG(hw, IXGBE_MTQC, (IXGBE_MTQC_RT_ENA |
1668                                         IXGBE_MTQC_8TC_8TQ));
1669         }
1670 }
1671
1672 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1673
1674 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter, int index)
1675 {
1676         struct ixgbe_ring *rx_ring;
1677         u32 srrctl;
1678         int queue0 = 0;
1679         unsigned long mask;
1680
1681         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1682                 queue0 = index;
1683         } else {
1684                 mask = (unsigned long) adapter->ring_feature[RING_F_RSS].mask;
1685                 queue0 = index & mask;
1686                 index = index & mask;
1687         }
1688
1689         rx_ring = &adapter->rx_ring[queue0];
1690
1691         srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
1692
1693         srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1694         srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1695
1696         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1697                 u16 bufsz = IXGBE_RXBUFFER_2048;
1698                 /* grow the amount we can receive on large page machines */
1699                 if (bufsz < (PAGE_SIZE / 2))
1700                         bufsz = (PAGE_SIZE / 2);
1701                 /* cap the bufsz at our largest descriptor size */
1702                 bufsz = min((u16)IXGBE_MAX_RXBUFFER, bufsz);
1703
1704                 srrctl |= bufsz >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1705                 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1706                 srrctl |= ((IXGBE_RX_HDR_SIZE <<
1707                             IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1708                            IXGBE_SRRCTL_BSIZEHDR_MASK);
1709         } else {
1710                 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1711
1712                 if (rx_ring->rx_buf_len == MAXIMUM_ETHERNET_VLAN_SIZE)
1713                         srrctl |= IXGBE_RXBUFFER_2048 >>
1714                                   IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1715                 else
1716                         srrctl |= rx_ring->rx_buf_len >>
1717                                   IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1718         }
1719
1720         IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
1721 }
1722
1723 /**
1724  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
1725  * @adapter: board private structure
1726  *
1727  * Configure the Rx unit of the MAC after a reset.
1728  **/
1729 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
1730 {
1731         u64 rdba;
1732         struct ixgbe_hw *hw = &adapter->hw;
1733         struct net_device *netdev = adapter->netdev;
1734         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1735         int i, j;
1736         u32 rdlen, rxctrl, rxcsum;
1737         static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
1738                           0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
1739                           0x6A3E67EA, 0x14364D17, 0x3BED200D};
1740         u32 fctrl, hlreg0;
1741         u32 reta = 0, mrqc = 0;
1742         u32 rdrxctl;
1743         int rx_buf_len;
1744
1745         /* Decide whether to use packet split mode or not */
1746         adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
1747
1748         /* Set the RX buffer length according to the mode */
1749         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1750                 rx_buf_len = IXGBE_RX_HDR_SIZE;
1751                 if (hw->mac.type == ixgbe_mac_82599EB) {
1752                         /* PSRTYPE must be initialized in 82599 */
1753                         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
1754                                       IXGBE_PSRTYPE_UDPHDR |
1755                                       IXGBE_PSRTYPE_IPV4HDR |
1756                                       IXGBE_PSRTYPE_IPV6HDR;
1757                         IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), psrtype);
1758                 }
1759         } else {
1760                 if (netdev->mtu <= ETH_DATA_LEN)
1761                         rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1762                 else
1763                         rx_buf_len = ALIGN(max_frame, 1024);
1764         }
1765
1766         fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
1767         fctrl |= IXGBE_FCTRL_BAM;
1768         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
1769         fctrl |= IXGBE_FCTRL_PMCF;
1770         IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
1771
1772         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
1773         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1774                 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
1775         else
1776                 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
1777         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
1778
1779         rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1780         /* disable receives while setting up the descriptors */
1781         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
1782         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
1783
1784         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1785          * the Base and Length of the Rx Descriptor Ring */
1786         for (i = 0; i < adapter->num_rx_queues; i++) {
1787                 rdba = adapter->rx_ring[i].dma;
1788                 j = adapter->rx_ring[i].reg_idx;
1789                 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_32BIT_MASK));
1790                 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
1791                 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
1792                 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
1793                 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
1794                 adapter->rx_ring[i].head = IXGBE_RDH(j);
1795                 adapter->rx_ring[i].tail = IXGBE_RDT(j);
1796                 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1797
1798                 ixgbe_configure_srrctl(adapter, j);
1799         }
1800
1801         if (hw->mac.type == ixgbe_mac_82598EB) {
1802                 /*
1803                  * For VMDq support of different descriptor types or
1804                  * buffer sizes through the use of multiple SRRCTL
1805                  * registers, RDRXCTL.MVMEN must be set to 1
1806                  *
1807                  * also, the manual doesn't mention it clearly but DCA hints
1808                  * will only use queue 0's tags unless this bit is set.  Side
1809                  * effects of setting this bit are only that SRRCTL must be
1810                  * fully programmed [0..15]
1811                  */
1812                 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
1813                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
1814                 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
1815         }
1816
1817         /* Program MRQC for the distribution of queues */
1818         if (hw->mac.type == ixgbe_mac_82599EB) {
1819                 int mask = adapter->flags & (
1820                                 IXGBE_FLAG_RSS_ENABLED
1821                                 | IXGBE_FLAG_DCB_ENABLED
1822                                 );
1823
1824                 switch (mask) {
1825                 case (IXGBE_FLAG_RSS_ENABLED):
1826                         mrqc = IXGBE_MRQC_RSSEN;
1827                         break;
1828                 case (IXGBE_FLAG_DCB_ENABLED):
1829                         mrqc = IXGBE_MRQC_RT8TCEN;
1830                         break;
1831                 default:
1832                         break;
1833                 }
1834         }
1835         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
1836                 /* Fill out redirection table */
1837                 for (i = 0, j = 0; i < 128; i++, j++) {
1838                         if (j == adapter->ring_feature[RING_F_RSS].indices)
1839                                 j = 0;
1840                         /* reta = 4-byte sliding window of
1841                          * 0x00..(indices-1)(indices-1)00..etc. */
1842                         reta = (reta << 8) | (j * 0x11);
1843                         if ((i & 3) == 3)
1844                                 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
1845                 }
1846
1847                 /* Fill out hash function seeds */
1848                 for (i = 0; i < 10; i++)
1849                         IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
1850
1851                 if (hw->mac.type == ixgbe_mac_82598EB)
1852                         mrqc |= IXGBE_MRQC_RSSEN;
1853                     /* Perform hash on these packet types */
1854                 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
1855                       | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
1856                       | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
1857                       | IXGBE_MRQC_RSS_FIELD_IPV6
1858                       | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
1859                       | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
1860         }
1861         IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
1862
1863         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
1864
1865         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
1866             adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
1867                 /* Disable indicating checksum in descriptor, enables
1868                  * RSS hash */
1869                 rxcsum |= IXGBE_RXCSUM_PCSD;
1870         }
1871         if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
1872                 /* Enable IPv4 payload checksum for UDP fragments
1873                  * if PCSD is not set */
1874                 rxcsum |= IXGBE_RXCSUM_IPPCSE;
1875         }
1876
1877         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
1878
1879         if (hw->mac.type == ixgbe_mac_82599EB) {
1880                 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
1881                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
1882                 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
1883         }
1884 }
1885
1886 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1887 {
1888         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1889         struct ixgbe_hw *hw = &adapter->hw;
1890
1891         /* add VID to filter table */
1892         hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
1893 }
1894
1895 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1896 {
1897         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1898         struct ixgbe_hw *hw = &adapter->hw;
1899
1900         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1901                 ixgbe_irq_disable(adapter);
1902
1903         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1904
1905         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1906                 ixgbe_irq_enable(adapter);
1907
1908         /* remove VID from filter table */
1909         hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
1910 }
1911
1912 static void ixgbe_vlan_rx_register(struct net_device *netdev,
1913                                    struct vlan_group *grp)
1914 {
1915         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1916         u32 ctrl;
1917         int i, j;
1918
1919         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1920                 ixgbe_irq_disable(adapter);
1921         adapter->vlgrp = grp;
1922
1923         /*
1924          * For a DCB driver, always enable VLAN tag stripping so we can
1925          * still receive traffic from a DCB-enabled host even if we're
1926          * not in DCB mode.
1927          */
1928         ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
1929         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1930                 ctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
1931                 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
1932                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
1933         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1934                 ctrl |= IXGBE_VLNCTRL_VFE;
1935                 /* enable VLAN tag insert/strip */
1936                 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
1937                 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
1938                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
1939                 for (i = 0; i < adapter->num_rx_queues; i++) {
1940                         j = adapter->rx_ring[i].reg_idx;
1941                         ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_RXDCTL(j));
1942                         ctrl |= IXGBE_RXDCTL_VME;
1943                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_RXDCTL(j), ctrl);
1944                 }
1945         }
1946         ixgbe_vlan_rx_add_vid(netdev, 0);
1947
1948         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1949                 ixgbe_irq_enable(adapter);
1950 }
1951
1952 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
1953 {
1954         ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1955
1956         if (adapter->vlgrp) {
1957                 u16 vid;
1958                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1959                         if (!vlan_group_get_device(adapter->vlgrp, vid))
1960                                 continue;
1961                         ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
1962                 }
1963         }
1964 }
1965
1966 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
1967 {
1968         struct dev_mc_list *mc_ptr;
1969         u8 *addr = *mc_addr_ptr;
1970         *vmdq = 0;
1971
1972         mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
1973         if (mc_ptr->next)
1974                 *mc_addr_ptr = mc_ptr->next->dmi_addr;
1975         else
1976                 *mc_addr_ptr = NULL;
1977
1978         return addr;
1979 }
1980
1981 /**
1982  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
1983  * @netdev: network interface device structure
1984  *
1985  * The set_rx_method entry point is called whenever the unicast/multicast
1986  * address list or the network interface flags are updated.  This routine is
1987  * responsible for configuring the hardware for proper unicast, multicast and
1988  * promiscuous mode.
1989  **/
1990 static void ixgbe_set_rx_mode(struct net_device *netdev)
1991 {
1992         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1993         struct ixgbe_hw *hw = &adapter->hw;
1994         u32 fctrl, vlnctrl;
1995         u8 *addr_list = NULL;
1996         int addr_count = 0;
1997
1998         /* Check for Promiscuous and All Multicast modes */
1999
2000         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2001         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2002
2003         if (netdev->flags & IFF_PROMISC) {
2004                 hw->addr_ctrl.user_set_promisc = 1;
2005                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2006                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2007         } else {
2008                 if (netdev->flags & IFF_ALLMULTI) {
2009                         fctrl |= IXGBE_FCTRL_MPE;
2010                         fctrl &= ~IXGBE_FCTRL_UPE;
2011                 } else {
2012                         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2013                 }
2014                 vlnctrl |= IXGBE_VLNCTRL_VFE;
2015                 hw->addr_ctrl.user_set_promisc = 0;
2016         }
2017
2018         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
2019         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2020
2021         /* reprogram secondary unicast list */
2022         addr_count = netdev->uc_count;
2023         if (addr_count)
2024                 addr_list = netdev->uc_list->dmi_addr;
2025         hw->mac.ops.update_uc_addr_list(hw, addr_list, addr_count,
2026                                           ixgbe_addr_list_itr);
2027
2028         /* reprogram multicast list */
2029         addr_count = netdev->mc_count;
2030         if (addr_count)
2031                 addr_list = netdev->mc_list->dmi_addr;
2032         hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
2033                                         ixgbe_addr_list_itr);
2034 }
2035
2036 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
2037 {
2038         int q_idx;
2039         struct ixgbe_q_vector *q_vector;
2040         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2041
2042         /* legacy and MSI only use one vector */
2043         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2044                 q_vectors = 1;
2045
2046         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2047                 struct napi_struct *napi;
2048                 q_vector = &adapter->q_vector[q_idx];
2049                 if (!q_vector->rxr_count)
2050                         continue;
2051                 napi = &q_vector->napi;
2052                 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) &&
2053                     (q_vector->rxr_count > 1))
2054                         napi->poll = &ixgbe_clean_rxonly_many;
2055
2056                 napi_enable(napi);
2057         }
2058 }
2059
2060 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2061 {
2062         int q_idx;
2063         struct ixgbe_q_vector *q_vector;
2064         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2065
2066         /* legacy and MSI only use one vector */
2067         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2068                 q_vectors = 1;
2069
2070         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2071                 q_vector = &adapter->q_vector[q_idx];
2072                 if (!q_vector->rxr_count)
2073                         continue;
2074                 napi_disable(&q_vector->napi);
2075         }
2076 }
2077
2078 #ifdef CONFIG_IXGBE_DCB
2079 /*
2080  * ixgbe_configure_dcb - Configure DCB hardware
2081  * @adapter: ixgbe adapter struct
2082  *
2083  * This is called by the driver on open to configure the DCB hardware.
2084  * This is also called by the gennetlink interface when reconfiguring
2085  * the DCB state.
2086  */
2087 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2088 {
2089         struct ixgbe_hw *hw = &adapter->hw;
2090         u32 txdctl, vlnctrl;
2091         int i, j;
2092
2093         ixgbe_dcb_check_config(&adapter->dcb_cfg);
2094         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2095         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2096
2097         /* reconfigure the hardware */
2098         ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
2099
2100         for (i = 0; i < adapter->num_tx_queues; i++) {
2101                 j = adapter->tx_ring[i].reg_idx;
2102                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2103                 /* PThresh workaround for Tx hang with DFP enabled. */
2104                 txdctl |= 32;
2105                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2106         }
2107         /* Enable VLAN tag insert/strip */
2108         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2109         if (hw->mac.type == ixgbe_mac_82598EB) {
2110                 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2111                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2112                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2113         } else if (hw->mac.type == ixgbe_mac_82599EB) {
2114                 vlnctrl |= IXGBE_VLNCTRL_VFE;
2115                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2116                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2117                 for (i = 0; i < adapter->num_rx_queues; i++) {
2118                         j = adapter->rx_ring[i].reg_idx;
2119                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2120                         vlnctrl |= IXGBE_RXDCTL_VME;
2121                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2122                 }
2123         }
2124         hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2125 }
2126
2127 #endif
2128 static void ixgbe_configure(struct ixgbe_adapter *adapter)
2129 {
2130         struct net_device *netdev = adapter->netdev;
2131         int i;
2132
2133         ixgbe_set_rx_mode(netdev);
2134
2135         ixgbe_restore_vlan(adapter);
2136 #ifdef CONFIG_IXGBE_DCB
2137         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2138                 netif_set_gso_max_size(netdev, 32768);
2139                 ixgbe_configure_dcb(adapter);
2140         } else {
2141                 netif_set_gso_max_size(netdev, 65536);
2142         }
2143 #else
2144         netif_set_gso_max_size(netdev, 65536);
2145 #endif
2146
2147         ixgbe_configure_tx(adapter);
2148         ixgbe_configure_rx(adapter);
2149         for (i = 0; i < adapter->num_rx_queues; i++)
2150                 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
2151                                        (adapter->rx_ring[i].count - 1));
2152 }
2153
2154 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2155 {
2156         switch (hw->phy.type) {
2157         case ixgbe_phy_sfp_avago:
2158         case ixgbe_phy_sfp_ftl:
2159         case ixgbe_phy_sfp_intel:
2160         case ixgbe_phy_sfp_unknown:
2161         case ixgbe_phy_tw_tyco:
2162         case ixgbe_phy_tw_unknown:
2163                 return true;
2164         default:
2165                 return false;
2166         }
2167 }
2168
2169 /**
2170  * ixgbe_sfp_link_config - set up SFP+ link
2171  * @adapter: pointer to private adapter struct
2172  **/
2173 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2174 {
2175         struct ixgbe_hw *hw = &adapter->hw;
2176
2177                 if (hw->phy.multispeed_fiber) {
2178                         /*
2179                          * In multispeed fiber setups, the device may not have
2180                          * had a physical connection when the driver loaded.
2181                          * If that's the case, the initial link configuration
2182                          * couldn't get the MAC into 10G or 1G mode, so we'll
2183                          * never have a link status change interrupt fire.
2184                          * We need to try and force an autonegotiation
2185                          * session, then bring up link.
2186                          */
2187                         hw->mac.ops.setup_sfp(hw);
2188                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2189                                 schedule_work(&adapter->multispeed_fiber_task);
2190                 } else {
2191                         /*
2192                          * Direct Attach Cu and non-multispeed fiber modules
2193                          * still need to be configured properly prior to
2194                          * attempting link.
2195                          */
2196                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2197                                 schedule_work(&adapter->sfp_config_module_task);
2198                 }
2199 }
2200
2201 /**
2202  * ixgbe_non_sfp_link_config - set up non-SFP+ link
2203  * @hw: pointer to private hardware struct
2204  *
2205  * Returns 0 on success, negative on failure
2206  **/
2207 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2208 {
2209         u32 autoneg;
2210         bool link_up = false;
2211         u32 ret = IXGBE_ERR_LINK_SETUP;
2212
2213         if (hw->mac.ops.check_link)
2214                 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2215
2216         if (ret)
2217                 goto link_cfg_out;
2218
2219         if (hw->mac.ops.get_link_capabilities)
2220                 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
2221                                                         &hw->mac.autoneg);
2222         if (ret)
2223                 goto link_cfg_out;
2224
2225         if (hw->mac.ops.setup_link_speed)
2226                 ret = hw->mac.ops.setup_link_speed(hw, autoneg, true, link_up);
2227 link_cfg_out:
2228         return ret;
2229 }
2230
2231 #define IXGBE_MAX_RX_DESC_POLL 10
2232 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2233                                               int rxr)
2234 {
2235         int j = adapter->rx_ring[rxr].reg_idx;
2236         int k;
2237
2238         for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2239                 if (IXGBE_READ_REG(&adapter->hw,
2240                                    IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2241                         break;
2242                 else
2243                         msleep(1);
2244         }
2245         if (k >= IXGBE_MAX_RX_DESC_POLL) {
2246                 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2247                         "not set within the polling period\n", rxr);
2248         }
2249         ixgbe_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
2250                               (adapter->rx_ring[rxr].count - 1));
2251 }
2252
2253 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2254 {
2255         struct net_device *netdev = adapter->netdev;
2256         struct ixgbe_hw *hw = &adapter->hw;
2257         int i, j = 0;
2258         int num_rx_rings = adapter->num_rx_queues;
2259         int err;
2260         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2261         u32 txdctl, rxdctl, mhadd;
2262         u32 dmatxctl;
2263         u32 gpie;
2264
2265         ixgbe_get_hw_control(adapter);
2266
2267         if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2268             (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2269                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2270                         gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2271                                 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2272                 } else {
2273                         /* MSI only */
2274                         gpie = 0;
2275                 }
2276                 /* XXX: to interrupt immediately for EICS writes, enable this */
2277                 /* gpie |= IXGBE_GPIE_EIMEN; */
2278                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2279         }
2280
2281         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2282                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2283                  * specifically only auto mask tx and rx interrupts */
2284                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2285         }
2286
2287         /* Enable fan failure interrupt if media type is copper */
2288         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2289                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2290                 gpie |= IXGBE_SDP1_GPIEN;
2291                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2292         }
2293
2294         if (hw->mac.type == ixgbe_mac_82599EB) {
2295                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2296                 gpie |= IXGBE_SDP1_GPIEN;
2297                 gpie |= IXGBE_SDP2_GPIEN;
2298                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2299         }
2300
2301         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2302         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2303                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2304                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2305
2306                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2307         }
2308
2309         for (i = 0; i < adapter->num_tx_queues; i++) {
2310                 j = adapter->tx_ring[i].reg_idx;
2311                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2312                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2313                 txdctl |= (8 << 16);
2314                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2315         }
2316
2317         if (hw->mac.type == ixgbe_mac_82599EB) {
2318                 /* DMATXCTL.EN must be set after all Tx queue config is done */
2319                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2320                 dmatxctl |= IXGBE_DMATXCTL_TE;
2321                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2322         }
2323         for (i = 0; i < adapter->num_tx_queues; i++) {
2324                 j = adapter->tx_ring[i].reg_idx;
2325                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2326                 txdctl |= IXGBE_TXDCTL_ENABLE;
2327                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2328         }
2329
2330         for (i = 0; i < num_rx_rings; i++) {
2331                 j = adapter->rx_ring[i].reg_idx;
2332                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2333                 /* enable PTHRESH=32 descriptors (half the internal cache)
2334                  * and HTHRESH=0 descriptors (to minimize latency on fetch),
2335                  * this also removes a pesky rx_no_buffer_count increment */
2336                 rxdctl |= 0x0020;
2337                 rxdctl |= IXGBE_RXDCTL_ENABLE;
2338                 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2339                 if (hw->mac.type == ixgbe_mac_82599EB)
2340                         ixgbe_rx_desc_queue_enable(adapter, i);
2341         }
2342         /* enable all receives */
2343         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2344         if (hw->mac.type == ixgbe_mac_82598EB)
2345                 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2346         else
2347                 rxdctl |= IXGBE_RXCTRL_RXEN;
2348         hw->mac.ops.enable_rx_dma(hw, rxdctl);
2349
2350         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2351                 ixgbe_configure_msix(adapter);
2352         else
2353                 ixgbe_configure_msi_and_legacy(adapter);
2354
2355         clear_bit(__IXGBE_DOWN, &adapter->state);
2356         ixgbe_napi_enable_all(adapter);
2357
2358         /* clear any pending interrupts, may auto mask */
2359         IXGBE_READ_REG(hw, IXGBE_EICR);
2360
2361         ixgbe_irq_enable(adapter);
2362
2363         /*
2364          * For hot-pluggable SFP+ devices, a new SFP+ module may have
2365          * arrived before interrupts were enabled.  We need to kick off
2366          * the SFP+ module setup first, then try to bring up link.
2367          * If we're not hot-pluggable SFP+, we just need to configure link
2368          * and bring it up.
2369          */
2370         err = hw->phy.ops.identify(hw);
2371         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
2372                 DPRINTK(PROBE, ERR, "PHY not supported on this NIC %d\n", err);
2373                 ixgbe_down(adapter);
2374                 return err;
2375         }
2376
2377         if (ixgbe_is_sfp(hw)) {
2378                 ixgbe_sfp_link_config(adapter);
2379         } else {
2380                 err = ixgbe_non_sfp_link_config(hw);
2381                 if (err)
2382                         DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
2383         }
2384
2385         /* enable transmits */
2386         netif_tx_start_all_queues(netdev);
2387
2388         /* bring the link up in the watchdog, this could race with our first
2389          * link up interrupt but shouldn't be a problem */
2390         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2391         adapter->link_check_timeout = jiffies;
2392         mod_timer(&adapter->watchdog_timer, jiffies);
2393         return 0;
2394 }
2395
2396 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
2397 {
2398         WARN_ON(in_interrupt());
2399         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
2400                 msleep(1);
2401         ixgbe_down(adapter);
2402         ixgbe_up(adapter);
2403         clear_bit(__IXGBE_RESETTING, &adapter->state);
2404 }
2405
2406 int ixgbe_up(struct ixgbe_adapter *adapter)
2407 {
2408         /* hardware has been reset, we need to reload some things */
2409         ixgbe_configure(adapter);
2410
2411         ixgbe_napi_add_all(adapter);
2412
2413         return ixgbe_up_complete(adapter);
2414 }
2415
2416 void ixgbe_reset(struct ixgbe_adapter *adapter)
2417 {
2418         struct ixgbe_hw *hw = &adapter->hw;
2419         if (hw->mac.ops.init_hw(hw))
2420                 dev_err(&adapter->pdev->dev, "Hardware Error\n");
2421
2422         /* reprogram the RAR[0] in case user changed it. */
2423         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2424
2425 }
2426
2427 /**
2428  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2429  * @adapter: board private structure
2430  * @rx_ring: ring to free buffers from
2431  **/
2432 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2433                                 struct ixgbe_ring *rx_ring)
2434 {
2435         struct pci_dev *pdev = adapter->pdev;
2436         unsigned long size;
2437         unsigned int i;
2438
2439         /* Free all the Rx ring sk_buffs */
2440
2441         for (i = 0; i < rx_ring->count; i++) {
2442                 struct ixgbe_rx_buffer *rx_buffer_info;
2443
2444                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2445                 if (rx_buffer_info->dma) {
2446                         pci_unmap_single(pdev, rx_buffer_info->dma,
2447                                          rx_ring->rx_buf_len,
2448                                          PCI_DMA_FROMDEVICE);
2449                         rx_buffer_info->dma = 0;
2450                 }
2451                 if (rx_buffer_info->skb) {
2452                         dev_kfree_skb(rx_buffer_info->skb);
2453                         rx_buffer_info->skb = NULL;
2454                 }
2455                 if (!rx_buffer_info->page)
2456                         continue;
2457                 pci_unmap_page(pdev, rx_buffer_info->page_dma, PAGE_SIZE / 2,
2458                                PCI_DMA_FROMDEVICE);
2459                 rx_buffer_info->page_dma = 0;
2460                 put_page(rx_buffer_info->page);
2461                 rx_buffer_info->page = NULL;
2462                 rx_buffer_info->page_offset = 0;
2463         }
2464
2465         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2466         memset(rx_ring->rx_buffer_info, 0, size);
2467
2468         /* Zero out the descriptor ring */
2469         memset(rx_ring->desc, 0, rx_ring->size);
2470
2471         rx_ring->next_to_clean = 0;
2472         rx_ring->next_to_use = 0;
2473
2474         if (rx_ring->head)
2475                 writel(0, adapter->hw.hw_addr + rx_ring->head);
2476         if (rx_ring->tail)
2477                 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2478 }
2479
2480 /**
2481  * ixgbe_clean_tx_ring - Free Tx Buffers
2482  * @adapter: board private structure
2483  * @tx_ring: ring to be cleaned
2484  **/
2485 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2486                                 struct ixgbe_ring *tx_ring)
2487 {
2488         struct ixgbe_tx_buffer *tx_buffer_info;
2489         unsigned long size;
2490         unsigned int i;
2491
2492         /* Free all the Tx ring sk_buffs */
2493
2494         for (i = 0; i < tx_ring->count; i++) {
2495                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2496                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2497         }
2498
2499         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2500         memset(tx_ring->tx_buffer_info, 0, size);
2501
2502         /* Zero out the descriptor ring */
2503         memset(tx_ring->desc, 0, tx_ring->size);
2504
2505         tx_ring->next_to_use = 0;
2506         tx_ring->next_to_clean = 0;
2507
2508         if (tx_ring->head)
2509                 writel(0, adapter->hw.hw_addr + tx_ring->head);
2510         if (tx_ring->tail)
2511                 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2512 }
2513
2514 /**
2515  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2516  * @adapter: board private structure
2517  **/
2518 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2519 {
2520         int i;
2521
2522         for (i = 0; i < adapter->num_rx_queues; i++)
2523                 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2524 }
2525
2526 /**
2527  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2528  * @adapter: board private structure
2529  **/
2530 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2531 {
2532         int i;
2533
2534         for (i = 0; i < adapter->num_tx_queues; i++)
2535                 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2536 }
2537
2538 void ixgbe_down(struct ixgbe_adapter *adapter)
2539 {
2540         struct net_device *netdev = adapter->netdev;
2541         struct ixgbe_hw *hw = &adapter->hw;
2542         u32 rxctrl;
2543         u32 txdctl;
2544         int i, j;
2545
2546         /* signal that we are down to the interrupt handler */
2547         set_bit(__IXGBE_DOWN, &adapter->state);
2548
2549         /* disable receives */
2550         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2551         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2552
2553         netif_tx_disable(netdev);
2554
2555         IXGBE_WRITE_FLUSH(hw);
2556         msleep(10);
2557
2558         netif_tx_stop_all_queues(netdev);
2559
2560         ixgbe_irq_disable(adapter);
2561
2562         ixgbe_napi_disable_all(adapter);
2563
2564         del_timer_sync(&adapter->watchdog_timer);
2565         cancel_work_sync(&adapter->watchdog_task);
2566
2567         /* disable transmits in the hardware now that interrupts are off */
2568         for (i = 0; i < adapter->num_tx_queues; i++) {
2569                 j = adapter->tx_ring[i].reg_idx;
2570                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2571                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
2572                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
2573         }
2574         /* Disable the Tx DMA engine on 82599 */
2575         if (hw->mac.type == ixgbe_mac_82599EB)
2576                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
2577                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
2578                                  ~IXGBE_DMATXCTL_TE));
2579
2580         netif_carrier_off(netdev);
2581
2582 #ifdef CONFIG_IXGBE_DCA
2583         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2584                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
2585                 dca_remove_requester(&adapter->pdev->dev);
2586         }
2587
2588 #endif
2589         if (!pci_channel_offline(adapter->pdev))
2590                 ixgbe_reset(adapter);
2591         ixgbe_clean_all_tx_rings(adapter);
2592         ixgbe_clean_all_rx_rings(adapter);
2593
2594 #ifdef CONFIG_IXGBE_DCA
2595         /* since we reset the hardware DCA settings were cleared */
2596         if (dca_add_requester(&adapter->pdev->dev) == 0) {
2597                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
2598                 /* always use CB2 mode, difference is masked
2599                  * in the CB driver */
2600                 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
2601                 ixgbe_setup_dca(adapter);
2602         }
2603 #endif
2604 }
2605
2606 /**
2607  * ixgbe_poll - NAPI Rx polling callback
2608  * @napi: structure for representing this polling device
2609  * @budget: how many packets driver is allowed to clean
2610  *
2611  * This function is used for legacy and MSI, NAPI mode
2612  **/
2613 static int ixgbe_poll(struct napi_struct *napi, int budget)
2614 {
2615         struct ixgbe_q_vector *q_vector =
2616                                 container_of(napi, struct ixgbe_q_vector, napi);
2617         struct ixgbe_adapter *adapter = q_vector->adapter;
2618         int tx_clean_complete, work_done = 0;
2619
2620 #ifdef CONFIG_IXGBE_DCA
2621         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2622                 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
2623                 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
2624         }
2625 #endif
2626
2627         tx_clean_complete = ixgbe_clean_tx_irq(adapter, adapter->tx_ring);
2628         ixgbe_clean_rx_irq(q_vector, adapter->rx_ring, &work_done, budget);
2629
2630         if (!tx_clean_complete)
2631                 work_done = budget;
2632
2633         /* If budget not fully consumed, exit the polling mode */
2634         if (work_done < budget) {
2635                 napi_complete(napi);
2636                 if (adapter->itr_setting & 1)
2637                         ixgbe_set_itr(adapter);
2638                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2639                         ixgbe_irq_enable_queues(adapter);
2640         }
2641         return work_done;
2642 }
2643
2644 /**
2645  * ixgbe_tx_timeout - Respond to a Tx Hang
2646  * @netdev: network interface device structure
2647  **/
2648 static void ixgbe_tx_timeout(struct net_device *netdev)
2649 {
2650         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2651
2652         /* Do the reset outside of interrupt context */
2653         schedule_work(&adapter->reset_task);
2654 }
2655
2656 static void ixgbe_reset_task(struct work_struct *work)
2657 {
2658         struct ixgbe_adapter *adapter;
2659         adapter = container_of(work, struct ixgbe_adapter, reset_task);
2660
2661         /* If we're already down or resetting, just bail */
2662         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
2663             test_bit(__IXGBE_RESETTING, &adapter->state))
2664                 return;
2665
2666         adapter->tx_timeout_count++;
2667
2668         ixgbe_reinit_locked(adapter);
2669 }
2670
2671 #ifdef CONFIG_IXGBE_DCB
2672 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
2673 {
2674         bool ret = false;
2675
2676         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2677                 adapter->ring_feature[RING_F_DCB].mask = 0x7 << 3;
2678                 adapter->num_rx_queues =
2679                                       adapter->ring_feature[RING_F_DCB].indices;
2680                 adapter->num_tx_queues =
2681                                       adapter->ring_feature[RING_F_DCB].indices;
2682                 ret = true;
2683         } else {
2684                 ret = false;
2685         }
2686
2687         return ret;
2688 }
2689 #endif
2690
2691 /**
2692  * ixgbe_set_rss_queues: Allocate queues for RSS
2693  * @adapter: board private structure to initialize
2694  *
2695  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
2696  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
2697  *
2698  **/
2699 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
2700 {
2701         bool ret = false;
2702
2703         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2704                 adapter->ring_feature[RING_F_RSS].mask = 0xF;
2705                 adapter->num_rx_queues =
2706                                       adapter->ring_feature[RING_F_RSS].indices;
2707                 adapter->num_tx_queues =
2708                                       adapter->ring_feature[RING_F_RSS].indices;
2709                 ret = true;
2710         } else {
2711                 ret = false;
2712         }
2713
2714         return ret;
2715 }
2716
2717 /*
2718  * ixgbe_set_num_queues: Allocate queues for device, feature dependant
2719  * @adapter: board private structure to initialize
2720  *
2721  * This is the top level queue allocation routine.  The order here is very
2722  * important, starting with the "most" number of features turned on at once,
2723  * and ending with the smallest set of features.  This way large combinations
2724  * can be allocated if they're turned on, and smaller combinations are the
2725  * fallthrough conditions.
2726  *
2727  **/
2728 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
2729 {
2730         /* Start with base case */
2731         adapter->num_rx_queues = 1;
2732         adapter->num_tx_queues = 1;
2733
2734 #ifdef CONFIG_IXGBE_DCB
2735         if (ixgbe_set_dcb_queues(adapter))
2736                 return;
2737
2738 #endif
2739         if (ixgbe_set_rss_queues(adapter))
2740                 return;
2741 }
2742
2743 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
2744                                        int vectors)
2745 {
2746         int err, vector_threshold;
2747
2748         /* We'll want at least 3 (vector_threshold):
2749          * 1) TxQ[0] Cleanup
2750          * 2) RxQ[0] Cleanup
2751          * 3) Other (Link Status Change, etc.)
2752          * 4) TCP Timer (optional)
2753          */
2754         vector_threshold = MIN_MSIX_COUNT;
2755
2756         /* The more we get, the more we will assign to Tx/Rx Cleanup
2757          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2758          * Right now, we simply care about how many we'll get; we'll
2759          * set them up later while requesting irq's.
2760          */
2761         while (vectors >= vector_threshold) {
2762                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
2763                                       vectors);
2764                 if (!err) /* Success in acquiring all requested vectors. */
2765                         break;
2766                 else if (err < 0)
2767                         vectors = 0; /* Nasty failure, quit now */
2768                 else /* err == number of vectors we should try again with */
2769                         vectors = err;
2770         }
2771
2772         if (vectors < vector_threshold) {
2773                 /* Can't allocate enough MSI-X interrupts?  Oh well.
2774                  * This just means we'll go with either a single MSI
2775                  * vector or fall back to legacy interrupts.
2776                  */
2777                 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
2778                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2779                 kfree(adapter->msix_entries);
2780                 adapter->msix_entries = NULL;
2781                 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
2782                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2783                 ixgbe_set_num_queues(adapter);
2784         } else {
2785                 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
2786                 /*
2787                  * Adjust for only the vectors we'll use, which is minimum
2788                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2789                  * vectors we were allocated.
2790                  */
2791                 adapter->num_msix_vectors = min(vectors,
2792                                    adapter->max_msix_q_vectors + NON_Q_VECTORS);
2793         }
2794 }
2795
2796 /**
2797  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
2798  * @adapter: board private structure to initialize
2799  *
2800  * Cache the descriptor ring offsets for RSS to the assigned rings.
2801  *
2802  **/
2803 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
2804 {
2805         int i;
2806         bool ret = false;
2807
2808         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2809                 for (i = 0; i < adapter->num_rx_queues; i++)
2810                         adapter->rx_ring[i].reg_idx = i;
2811                 for (i = 0; i < adapter->num_tx_queues; i++)
2812                         adapter->tx_ring[i].reg_idx = i;
2813                 ret = true;
2814         } else {
2815                 ret = false;
2816         }
2817
2818         return ret;
2819 }
2820
2821 #ifdef CONFIG_IXGBE_DCB
2822 /**
2823  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
2824  * @adapter: board private structure to initialize
2825  *
2826  * Cache the descriptor ring offsets for DCB to the assigned rings.
2827  *
2828  **/
2829 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
2830 {
2831         int i;
2832         bool ret = false;
2833         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
2834
2835         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2836                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2837                         /* the number of queues is assumed to be symmetric */
2838                         for (i = 0; i < dcb_i; i++) {
2839                                 adapter->rx_ring[i].reg_idx = i << 3;
2840                                 adapter->tx_ring[i].reg_idx = i << 2;
2841                         }
2842                         ret = true;
2843                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2844                         for (i = 0; i < dcb_i; i++) {
2845                                 adapter->rx_ring[i].reg_idx = i << 4;
2846                                 adapter->tx_ring[i].reg_idx = i << 4;
2847                         }
2848                         ret = true;
2849                 } else {
2850                         ret = false;
2851                 }
2852         } else {
2853                 ret = false;
2854         }
2855
2856         return ret;
2857 }
2858 #endif
2859
2860 /**
2861  * ixgbe_cache_ring_register - Descriptor ring to register mapping
2862  * @adapter: board private structure to initialize
2863  *
2864  * Once we know the feature-set enabled for the device, we'll cache
2865  * the register offset the descriptor ring is assigned to.
2866  *
2867  * Note, the order the various feature calls is important.  It must start with
2868  * the "most" features enabled at the same time, then trickle down to the
2869  * least amount of features turned on at once.
2870  **/
2871 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
2872 {
2873         /* start with default case */
2874         adapter->rx_ring[0].reg_idx = 0;
2875         adapter->tx_ring[0].reg_idx = 0;
2876
2877 #ifdef CONFIG_IXGBE_DCB
2878         if (ixgbe_cache_ring_dcb(adapter))
2879                 return;
2880
2881 #endif
2882         if (ixgbe_cache_ring_rss(adapter))
2883                 return;
2884 }
2885
2886 /**
2887  * ixgbe_alloc_queues - Allocate memory for all rings
2888  * @adapter: board private structure to initialize
2889  *
2890  * We allocate one ring per queue at run-time since we don't know the
2891  * number of queues at compile-time.  The polling_netdev array is
2892  * intended for Multiqueue, but should work fine with a single queue.
2893  **/
2894 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
2895 {
2896         int i;
2897
2898         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
2899                                    sizeof(struct ixgbe_ring), GFP_KERNEL);
2900         if (!adapter->tx_ring)
2901                 goto err_tx_ring_allocation;
2902
2903         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
2904                                    sizeof(struct ixgbe_ring), GFP_KERNEL);
2905         if (!adapter->rx_ring)
2906                 goto err_rx_ring_allocation;
2907
2908         for (i = 0; i < adapter->num_tx_queues; i++) {
2909                 adapter->tx_ring[i].count = adapter->tx_ring_count;
2910                 adapter->tx_ring[i].queue_index = i;
2911         }
2912
2913         for (i = 0; i < adapter->num_rx_queues; i++) {
2914                 adapter->rx_ring[i].count = adapter->rx_ring_count;
2915                 adapter->rx_ring[i].queue_index = i;
2916         }
2917
2918         ixgbe_cache_ring_register(adapter);
2919
2920         return 0;
2921
2922 err_rx_ring_allocation:
2923         kfree(adapter->tx_ring);
2924 err_tx_ring_allocation:
2925         return -ENOMEM;
2926 }
2927
2928 /**
2929  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
2930  * @adapter: board private structure to initialize
2931  *
2932  * Attempt to configure the interrupts using the best available
2933  * capabilities of the hardware and the kernel.
2934  **/
2935 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
2936 {
2937         int err = 0;
2938         int vector, v_budget;
2939
2940         /*
2941          * It's easy to be greedy for MSI-X vectors, but it really
2942          * doesn't do us much good if we have a lot more vectors
2943          * than CPU's.  So let's be conservative and only ask for
2944          * (roughly) twice the number of vectors as there are CPU's.
2945          */
2946         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
2947                        (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
2948
2949         /*
2950          * At the same time, hardware can only support a maximum of
2951          * MAX_MSIX_COUNT vectors.  With features such as RSS and VMDq,
2952          * we can easily reach upwards of 64 Rx descriptor queues and
2953          * 32 Tx queues.  Thus, we cap it off in those rare cases where
2954          * the cpu count also exceeds our vector limit.
2955          */
2956         v_budget = min(v_budget, MAX_MSIX_COUNT);
2957
2958         /* A failure in MSI-X entry allocation isn't fatal, but it does
2959          * mean we disable MSI-X capabilities of the adapter. */
2960         adapter->msix_entries = kcalloc(v_budget,
2961                                         sizeof(struct msix_entry), GFP_KERNEL);
2962         if (!adapter->msix_entries) {
2963                 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
2964                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2965                 ixgbe_set_num_queues(adapter);
2966                 kfree(adapter->tx_ring);
2967                 kfree(adapter->rx_ring);
2968                 err = ixgbe_alloc_queues(adapter);
2969                 if (err) {
2970                         DPRINTK(PROBE, ERR, "Unable to allocate memory "
2971                                 "for queues\n");
2972                         goto out;
2973                 }
2974
2975                 goto try_msi;
2976         }
2977
2978         for (vector = 0; vector < v_budget; vector++)
2979                 adapter->msix_entries[vector].entry = vector;
2980
2981         ixgbe_acquire_msix_vectors(adapter, v_budget);
2982
2983         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2984                 goto out;
2985
2986 try_msi:
2987         err = pci_enable_msi(adapter->pdev);
2988         if (!err) {
2989                 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
2990         } else {
2991                 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
2992                         "falling back to legacy.  Error: %d\n", err);
2993                 /* reset err */
2994                 err = 0;
2995         }
2996
2997 out:
2998         /* Notify the stack of the (possibly) reduced Tx Queue count. */
2999         adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
3000
3001         return err;
3002 }
3003
3004 void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
3005 {
3006         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3007                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3008                 pci_disable_msix(adapter->pdev);
3009                 kfree(adapter->msix_entries);
3010                 adapter->msix_entries = NULL;
3011         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
3012                 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
3013                 pci_disable_msi(adapter->pdev);
3014         }
3015         return;
3016 }
3017
3018 /**
3019  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
3020  * @adapter: board private structure to initialize
3021  *
3022  * We determine which interrupt scheme to use based on...
3023  * - Kernel support (MSI, MSI-X)
3024  *   - which can be user-defined (via MODULE_PARAM)
3025  * - Hardware queue count (num_*_queues)
3026  *   - defined by miscellaneous hardware support/features (RSS, etc.)
3027  **/
3028 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
3029 {
3030         int err;
3031
3032         /* Number of supported queues */
3033         ixgbe_set_num_queues(adapter);
3034
3035         err = ixgbe_alloc_queues(adapter);
3036         if (err) {
3037                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
3038                 goto err_alloc_queues;
3039         }
3040
3041         err = ixgbe_set_interrupt_capability(adapter);
3042         if (err) {
3043                 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
3044                 goto err_set_interrupt;
3045         }
3046
3047         DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
3048                 "Tx Queue count = %u\n",
3049                 (adapter->num_rx_queues > 1) ? "Enabled" :
3050                 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
3051
3052         set_bit(__IXGBE_DOWN, &adapter->state);
3053
3054         return 0;
3055
3056 err_set_interrupt:
3057         kfree(adapter->tx_ring);
3058         kfree(adapter->rx_ring);
3059 err_alloc_queues:
3060         return err;
3061 }
3062
3063 /**
3064  * ixgbe_sfp_timer - worker thread to find a missing module
3065  * @data: pointer to our adapter struct
3066  **/
3067 static void ixgbe_sfp_timer(unsigned long data)
3068 {
3069         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3070
3071         /*
3072          * Do the sfp_timer outside of interrupt context due to the
3073          * delays that sfp+ detection requires
3074          */
3075         schedule_work(&adapter->sfp_task);
3076 }
3077
3078 /**
3079  * ixgbe_sfp_task - worker thread to find a missing module
3080  * @work: pointer to work_struct containing our data
3081  **/
3082 static void ixgbe_sfp_task(struct work_struct *work)
3083 {
3084         struct ixgbe_adapter *adapter = container_of(work,
3085                                                      struct ixgbe_adapter,
3086                                                      sfp_task);
3087         struct ixgbe_hw *hw = &adapter->hw;
3088
3089         if ((hw->phy.type == ixgbe_phy_nl) &&
3090             (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
3091                 s32 ret = hw->phy.ops.identify_sfp(hw);
3092                 if (ret)
3093                         goto reschedule;
3094                 ret = hw->phy.ops.reset(hw);
3095                 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3096                         DPRINTK(PROBE, ERR, "failed to initialize because an "
3097                                 "unsupported SFP+ module type was detected.\n"
3098                                 "Reload the driver after installing a "
3099                                 "supported module.\n");
3100                         unregister_netdev(adapter->netdev);
3101                 } else {
3102                         DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
3103                                 hw->phy.sfp_type);
3104                 }
3105                 /* don't need this routine any more */
3106                 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3107         }
3108         return;
3109 reschedule:
3110         if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
3111                 mod_timer(&adapter->sfp_timer,
3112                           round_jiffies(jiffies + (2 * HZ)));
3113 }
3114
3115 /**
3116  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
3117  * @adapter: board private structure to initialize
3118  *
3119  * ixgbe_sw_init initializes the Adapter private data structure.
3120  * Fields are initialized based on PCI device information and
3121  * OS network device settings (MTU size).
3122  **/
3123 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
3124 {
3125         struct ixgbe_hw *hw = &adapter->hw;
3126         struct pci_dev *pdev = adapter->pdev;
3127         unsigned int rss;
3128 #ifdef CONFIG_IXGBE_DCB
3129         int j;
3130         struct tc_configuration *tc;
3131 #endif
3132
3133         /* PCI config space info */
3134
3135         hw->vendor_id = pdev->vendor;
3136         hw->device_id = pdev->device;
3137         hw->revision_id = pdev->revision;
3138         hw->subsystem_vendor_id = pdev->subsystem_vendor;
3139         hw->subsystem_device_id = pdev->subsystem_device;
3140
3141         /* Set capability flags */
3142         rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
3143         adapter->ring_feature[RING_F_RSS].indices = rss;
3144         adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
3145         adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
3146         if (hw->mac.type == ixgbe_mac_82598EB)
3147                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
3148         else if (hw->mac.type == ixgbe_mac_82599EB)
3149                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
3150
3151 #ifdef CONFIG_IXGBE_DCB
3152         /* Configure DCB traffic classes */
3153         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
3154                 tc = &adapter->dcb_cfg.tc_config[j];
3155                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
3156                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
3157                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
3158                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
3159                 tc->dcb_pfc = pfc_disabled;
3160         }
3161         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
3162         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
3163         adapter->dcb_cfg.rx_pba_cfg = pba_equal;
3164         adapter->dcb_cfg.round_robin_enable = false;
3165         adapter->dcb_set_bitmap = 0x00;
3166         ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
3167                            adapter->ring_feature[RING_F_DCB].indices);
3168
3169 #endif
3170
3171         /* default flow control settings */
3172         hw->fc.requested_mode = ixgbe_fc_none;
3173         hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
3174         hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
3175         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
3176         hw->fc.send_xon = true;
3177
3178         /* enable itr by default in dynamic mode */
3179         adapter->itr_setting = 1;
3180         adapter->eitr_param = 20000;
3181
3182         /* set defaults for eitr in MegaBytes */
3183         adapter->eitr_low = 10;
3184         adapter->eitr_high = 20;
3185
3186         /* set default ring sizes */
3187         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
3188         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
3189
3190         /* initialize eeprom parameters */
3191         if (ixgbe_init_eeprom_params_generic(hw)) {
3192                 dev_err(&pdev->dev, "EEPROM initialization failed\n");
3193                 return -EIO;
3194         }
3195
3196         /* enable rx csum by default */
3197         adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
3198
3199         set_bit(__IXGBE_DOWN, &adapter->state);
3200
3201         return 0;
3202 }
3203
3204 /**
3205  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
3206  * @adapter: board private structure
3207  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
3208  *
3209  * Return 0 on success, negative on failure
3210  **/
3211 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
3212                              struct ixgbe_ring *tx_ring)
3213 {
3214         struct pci_dev *pdev = adapter->pdev;
3215         int size;
3216
3217         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3218         tx_ring->tx_buffer_info = vmalloc(size);
3219         if (!tx_ring->tx_buffer_info)
3220                 goto err;
3221         memset(tx_ring->tx_buffer_info, 0, size);
3222
3223         /* round up to nearest 4K */
3224         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3225         tx_ring->size = ALIGN(tx_ring->size, 4096);
3226
3227         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
3228                                              &tx_ring->dma);
3229         if (!tx_ring->desc)
3230                 goto err;
3231
3232         tx_ring->next_to_use = 0;
3233         tx_ring->next_to_clean = 0;
3234         tx_ring->work_limit = tx_ring->count;
3235         return 0;
3236
3237 err:
3238         vfree(tx_ring->tx_buffer_info);
3239         tx_ring->tx_buffer_info = NULL;
3240         DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
3241                             "descriptor ring\n");
3242         return -ENOMEM;
3243 }
3244
3245 /**
3246  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
3247  * @adapter: board private structure
3248  *
3249  * If this function returns with an error, then it's possible one or
3250  * more of the rings is populated (while the rest are not).  It is the
3251  * callers duty to clean those orphaned rings.
3252  *
3253  * Return 0 on success, negative on failure
3254  **/
3255 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
3256 {
3257         int i, err = 0;
3258
3259         for (i = 0; i < adapter->num_tx_queues; i++) {
3260                 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
3261                 if (!err)
3262                         continue;
3263                 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
3264                 break;
3265         }
3266
3267         return err;
3268 }
3269
3270 /**
3271  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
3272  * @adapter: board private structure
3273  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
3274  *
3275  * Returns 0 on success, negative on failure
3276  **/
3277 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
3278                              struct ixgbe_ring *rx_ring)
3279 {
3280         struct pci_dev *pdev = adapter->pdev;
3281         int size;
3282
3283         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3284         rx_ring->rx_buffer_info = vmalloc(size);
3285         if (!rx_ring->rx_buffer_info) {
3286                 DPRINTK(PROBE, ERR,
3287                         "vmalloc allocation failed for the rx desc ring\n");
3288                 goto alloc_failed;
3289         }
3290         memset(rx_ring->rx_buffer_info, 0, size);
3291
3292         /* Round up to nearest 4K */
3293         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3294         rx_ring->size = ALIGN(rx_ring->size, 4096);
3295
3296         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
3297
3298         if (!rx_ring->desc) {
3299                 DPRINTK(PROBE, ERR,
3300                         "Memory allocation failed for the rx desc ring\n");
3301                 vfree(rx_ring->rx_buffer_info);
3302                 goto alloc_failed;
3303         }
3304
3305         rx_ring->next_to_clean = 0;
3306         rx_ring->next_to_use = 0;
3307
3308         return 0;
3309
3310 alloc_failed:
3311         return -ENOMEM;
3312 }
3313
3314 /**
3315  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
3316  * @adapter: board private structure
3317  *
3318  * If this function returns with an error, then it's possible one or
3319  * more of the rings is populated (while the rest are not).  It is the
3320  * callers duty to clean those orphaned rings.
3321  *
3322  * Return 0 on success, negative on failure
3323  **/
3324
3325 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
3326 {
3327         int i, err = 0;
3328
3329         for (i = 0; i < adapter->num_rx_queues; i++) {
3330                 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
3331                 if (!err)
3332                         continue;
3333                 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
3334                 break;
3335         }
3336
3337         return err;
3338 }
3339
3340 /**
3341  * ixgbe_free_tx_resources - Free Tx Resources per Queue
3342  * @adapter: board private structure
3343  * @tx_ring: Tx descriptor ring for a specific queue
3344  *
3345  * Free all transmit software resources
3346  **/
3347 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
3348                              struct ixgbe_ring *tx_ring)
3349 {
3350         struct pci_dev *pdev = adapter->pdev;
3351
3352         ixgbe_clean_tx_ring(adapter, tx_ring);
3353
3354         vfree(tx_ring->tx_buffer_info);
3355         tx_ring->tx_buffer_info = NULL;
3356
3357         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
3358
3359         tx_ring->desc = NULL;
3360 }
3361
3362 /**
3363  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
3364  * @adapter: board private structure
3365  *
3366  * Free all transmit software resources
3367  **/
3368 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
3369 {
3370         int i;
3371
3372         for (i = 0; i < adapter->num_tx_queues; i++)
3373                 if (adapter->tx_ring[i].desc)
3374                         ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
3375 }
3376
3377 /**
3378  * ixgbe_free_rx_resources - Free Rx Resources
3379  * @adapter: board private structure
3380  * @rx_ring: ring to clean the resources from
3381  *
3382  * Free all receive software resources
3383  **/
3384 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
3385                              struct ixgbe_ring *rx_ring)
3386 {
3387         struct pci_dev *pdev = adapter->pdev;
3388
3389         ixgbe_clean_rx_ring(adapter, rx_ring);
3390
3391         vfree(rx_ring->rx_buffer_info);
3392         rx_ring->rx_buffer_info = NULL;
3393
3394         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
3395
3396         rx_ring->desc = NULL;
3397 }
3398
3399 /**
3400  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
3401  * @adapter: board private structure
3402  *
3403  * Free all receive software resources
3404  **/
3405 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
3406 {
3407         int i;
3408
3409         for (i = 0; i < adapter->num_rx_queues; i++)
3410                 if (adapter->rx_ring[i].desc)
3411                         ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
3412 }
3413
3414 /**
3415  * ixgbe_change_mtu - Change the Maximum Transfer Unit
3416  * @netdev: network interface device structure
3417  * @new_mtu: new value for maximum frame size
3418  *
3419  * Returns 0 on success, negative on failure
3420  **/
3421 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
3422 {
3423         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3424         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3425
3426         /* MTU < 68 is an error and causes problems on some kernels */
3427         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
3428                 return -EINVAL;
3429
3430         DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
3431                 netdev->mtu, new_mtu);
3432         /* must set new MTU before calling down or up */
3433         netdev->mtu = new_mtu;
3434
3435         if (netif_running(netdev))
3436                 ixgbe_reinit_locked(adapter);
3437
3438         return 0;
3439 }
3440
3441 /**
3442  * ixgbe_open - Called when a network interface is made active
3443  * @netdev: network interface device structure
3444  *
3445  * Returns 0 on success, negative value on failure
3446  *
3447  * The open entry point is called when a network interface is made
3448  * active by the system (IFF_UP).  At this point all resources needed
3449  * for transmit and receive operations are allocated, the interrupt
3450  * handler is registered with the OS, the watchdog timer is started,
3451  * and the stack is notified that the interface is ready.
3452  **/
3453 static int ixgbe_open(struct net_device *netdev)
3454 {
3455         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3456         int err;
3457
3458         /* disallow open during test */
3459         if (test_bit(__IXGBE_TESTING, &adapter->state))
3460                 return -EBUSY;
3461
3462         /* allocate transmit descriptors */
3463         err = ixgbe_setup_all_tx_resources(adapter);
3464         if (err)
3465                 goto err_setup_tx;
3466
3467         /* allocate receive descriptors */
3468         err = ixgbe_setup_all_rx_resources(adapter);
3469         if (err)
3470                 goto err_setup_rx;
3471
3472         ixgbe_configure(adapter);
3473
3474         ixgbe_napi_add_all(adapter);
3475
3476         err = ixgbe_request_irq(adapter);
3477         if (err)
3478                 goto err_req_irq;
3479
3480         err = ixgbe_up_complete(adapter);
3481         if (err)
3482                 goto err_up;
3483
3484         netif_tx_start_all_queues(netdev);
3485
3486         return 0;
3487
3488 err_up:
3489         ixgbe_release_hw_control(adapter);
3490         ixgbe_free_irq(adapter);
3491 err_req_irq:
3492         ixgbe_free_all_rx_resources(adapter);
3493 err_setup_rx:
3494         ixgbe_free_all_tx_resources(adapter);
3495 err_setup_tx:
3496         ixgbe_reset(adapter);
3497
3498         return err;
3499 }
3500
3501 /**
3502  * ixgbe_close - Disables a network interface
3503  * @netdev: network interface device structure
3504  *
3505  * Returns 0, this is not allowed to fail
3506  *
3507  * The close entry point is called when an interface is de-activated
3508  * by the OS.  The hardware is still under the drivers control, but
3509  * needs to be disabled.  A global MAC reset is issued to stop the
3510  * hardware, and all transmit and receive resources are freed.
3511  **/
3512 static int ixgbe_close(struct net_device *netdev)
3513 {
3514         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3515
3516         ixgbe_down(adapter);
3517         ixgbe_free_irq(adapter);
3518
3519         ixgbe_free_all_tx_resources(adapter);
3520         ixgbe_free_all_rx_resources(adapter);
3521
3522         ixgbe_release_hw_control(adapter);
3523
3524         return 0;
3525 }
3526
3527 /**
3528  * ixgbe_napi_add_all - prep napi structs for use
3529  * @adapter: private struct
3530  *
3531  * helper function to napi_add each possible q_vector->napi
3532  */
3533 void ixgbe_napi_add_all(struct ixgbe_adapter *adapter)
3534 {
3535         int q_idx, q_vectors;
3536         struct net_device *netdev = adapter->netdev;
3537         int (*poll)(struct napi_struct *, int);
3538
3539         /* check if we already have our netdev->napi_list populated */
3540         if (&netdev->napi_list != netdev->napi_list.next)
3541                 return;
3542
3543         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3544                 poll = &ixgbe_clean_rxonly;
3545                 /* Only enable as many vectors as we have rx queues. */
3546                 q_vectors = adapter->num_rx_queues;
3547         } else {
3548                 poll = &ixgbe_poll;
3549                 /* only one q_vector for legacy modes */
3550                 q_vectors = 1;
3551         }
3552
3553         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3554                 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
3555                 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
3556         }
3557 }
3558
3559 void ixgbe_napi_del_all(struct ixgbe_adapter *adapter)
3560 {
3561         int q_idx;
3562         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3563
3564         /* legacy and MSI only use one vector */
3565         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3566                 q_vectors = 1;
3567
3568         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3569                 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
3570                 if (!q_vector->rxr_count)
3571                         continue;
3572                 netif_napi_del(&q_vector->napi);
3573         }
3574 }
3575
3576 #ifdef CONFIG_PM
3577 static int ixgbe_resume(struct pci_dev *pdev)
3578 {
3579         struct net_device *netdev = pci_get_drvdata(pdev);
3580         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3581         u32 err;
3582
3583         pci_set_power_state(pdev, PCI_D0);
3584         pci_restore_state(pdev);
3585         err = pci_enable_device(pdev);
3586         if (err) {
3587                 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
3588                                 "suspend\n");
3589                 return err;
3590         }
3591         pci_set_master(pdev);
3592
3593         pci_enable_wake(pdev, PCI_D3hot, 0);
3594         pci_enable_wake(pdev, PCI_D3cold, 0);
3595
3596         err = ixgbe_init_interrupt_scheme(adapter);
3597         if (err) {
3598                 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
3599                                 "device\n");
3600                 return err;
3601         }
3602
3603         ixgbe_reset(adapter);
3604
3605         if (netif_running(netdev)) {
3606                 err = ixgbe_open(adapter->netdev);
3607                 if (err)
3608                         return err;
3609         }
3610
3611         netif_device_attach(netdev);
3612
3613         return 0;
3614 }
3615
3616 #endif /* CONFIG_PM */
3617 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
3618 {
3619         struct net_device *netdev = pci_get_drvdata(pdev);
3620         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3621         struct ixgbe_hw *hw = &adapter->hw;
3622         u32 ctrl, fctrl;
3623         u32 wufc = adapter->wol;
3624 #ifdef CONFIG_PM
3625         int retval = 0;
3626 #endif
3627
3628         netif_device_detach(netdev);
3629
3630         if (netif_running(netdev)) {
3631                 ixgbe_down(adapter);
3632                 ixgbe_free_irq(adapter);
3633                 ixgbe_free_all_tx_resources(adapter);
3634                 ixgbe_free_all_rx_resources(adapter);
3635         }
3636         ixgbe_reset_interrupt_capability(adapter);
3637         ixgbe_napi_del_all(adapter);
3638         INIT_LIST_HEAD(&netdev->napi_list);
3639         kfree(adapter->tx_ring);
3640         kfree(adapter->rx_ring);
3641
3642 #ifdef CONFIG_PM
3643         retval = pci_save_state(pdev);
3644         if (retval)
3645                 return retval;
3646
3647 #endif
3648         if (wufc) {
3649                 ixgbe_set_rx_mode(netdev);
3650
3651                 /* turn on all-multi mode if wake on multicast is enabled */
3652                 if (wufc & IXGBE_WUFC_MC) {
3653                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3654                         fctrl |= IXGBE_FCTRL_MPE;
3655                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3656                 }
3657
3658                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
3659                 ctrl |= IXGBE_CTRL_GIO_DIS;
3660                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
3661
3662                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
3663         } else {
3664                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
3665                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
3666         }
3667
3668         if (wufc && hw->mac.type == ixgbe_mac_82599EB) {
3669                 pci_enable_wake(pdev, PCI_D3hot, 1);
3670                 pci_enable_wake(pdev, PCI_D3cold, 1);
3671         } else {
3672                 pci_enable_wake(pdev, PCI_D3hot, 0);
3673                 pci_enable_wake(pdev, PCI_D3cold, 0);
3674         }
3675
3676         ixgbe_release_hw_control(adapter);
3677
3678         pci_disable_device(pdev);
3679
3680         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3681
3682         return 0;
3683 }
3684
3685 static void ixgbe_shutdown(struct pci_dev *pdev)
3686 {
3687         ixgbe_suspend(pdev, PMSG_SUSPEND);
3688 }
3689
3690 /**
3691  * ixgbe_update_stats - Update the board statistics counters.
3692  * @adapter: board private structure
3693  **/
3694 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
3695 {
3696         struct ixgbe_hw *hw = &adapter->hw;
3697         u64 total_mpc = 0;
3698         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
3699
3700         if (hw->mac.type == ixgbe_mac_82599EB) {
3701                 for (i = 0; i < 16; i++)
3702                         adapter->hw_rx_no_dma_resources +=
3703                                              IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
3704         }
3705
3706         adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
3707         for (i = 0; i < 8; i++) {
3708                 /* for packet buffers not used, the register should read 0 */
3709                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
3710                 missed_rx += mpc;
3711                 adapter->stats.mpc[i] += mpc;
3712                 total_mpc += adapter->stats.mpc[i];
3713                 if (hw->mac.type == ixgbe_mac_82598EB)
3714                         adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
3715                 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
3716                 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
3717                 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
3718                 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
3719                 if (hw->mac.type == ixgbe_mac_82599EB) {
3720                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
3721                                                             IXGBE_PXONRXCNT(i));
3722                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
3723                                                            IXGBE_PXOFFRXCNT(i));
3724                         adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
3725                 } else {
3726                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
3727                                                               IXGBE_PXONRXC(i));
3728                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
3729                                                              IXGBE_PXOFFRXC(i));
3730                 }
3731                 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
3732                                                             IXGBE_PXONTXC(i));
3733                 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
3734                                                              IXGBE_PXOFFTXC(i));
3735         }
3736         adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
3737         /* work around hardware counting issue */
3738         adapter->stats.gprc -= missed_rx;
3739
3740         /* 82598 hardware only has a 32 bit counter in the high register */
3741         if (hw->mac.type == ixgbe_mac_82599EB) {
3742                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
3743                 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
3744                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
3745                 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
3746                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
3747                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
3748                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
3749                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
3750         } else {
3751                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
3752                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
3753                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
3754                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
3755                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
3756         }
3757         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
3758         adapter->stats.bprc += bprc;
3759         adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
3760         if (hw->mac.type == ixgbe_mac_82598EB)
3761                 adapter->stats.mprc -= bprc;
3762         adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
3763         adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
3764         adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
3765         adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
3766         adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
3767         adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
3768         adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
3769         adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
3770         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
3771         adapter->stats.lxontxc += lxon;
3772         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
3773         adapter->stats.lxofftxc += lxoff;
3774         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3775         adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
3776         adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
3777         /*
3778          * 82598 errata - tx of flow control packets is included in tx counters
3779          */
3780         xon_off_tot = lxon + lxoff;
3781         adapter->stats.gptc -= xon_off_tot;
3782         adapter->stats.mptc -= xon_off_tot;
3783         adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
3784         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3785         adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
3786         adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
3787         adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
3788         adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
3789         adapter->stats.ptc64 -= xon_off_tot;
3790         adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
3791         adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
3792         adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
3793         adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
3794         adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
3795         adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
3796
3797         /* Fill out the OS statistics structure */
3798         adapter->net_stats.multicast = adapter->stats.mprc;
3799
3800         /* Rx Errors */
3801         adapter->net_stats.rx_errors = adapter->stats.crcerrs +
3802                                        adapter->stats.rlec;
3803         adapter->net_stats.rx_dropped = 0;
3804         adapter->net_stats.rx_length_errors = adapter->stats.rlec;
3805         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3806         adapter->net_stats.rx_missed_errors = total_mpc;
3807 }
3808
3809 /**
3810  * ixgbe_watchdog - Timer Call-back
3811  * @data: pointer to adapter cast into an unsigned long
3812  **/
3813 static void ixgbe_watchdog(unsigned long data)
3814 {
3815         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3816         struct ixgbe_hw *hw = &adapter->hw;
3817
3818         /* Do the watchdog outside of interrupt context due to the lovely
3819          * delays that some of the newer hardware requires */
3820         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
3821                 u64 eics = 0;
3822                 int i;
3823
3824                 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++)
3825                         eics |= (1 << i);
3826
3827                 /* Cause software interrupt to ensure rx rings are cleaned */
3828                 switch (hw->mac.type) {
3829                 case ixgbe_mac_82598EB:
3830                         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3831                                 IXGBE_WRITE_REG(hw, IXGBE_EICS, (u32)eics);
3832                         } else {
3833                                 /*
3834                                  * for legacy and MSI interrupts don't set any
3835                                  * bits that are enabled for EIAM, because this
3836                                  * operation would set *both* EIMS and EICS for
3837                                  * any bit in EIAM
3838                                  */
3839                                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
3840                                      (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
3841                         }
3842                         break;
3843                 case ixgbe_mac_82599EB:
3844                         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3845                                 /*
3846                                  * EICS(0..15) first 0-15 q vectors
3847                                  * EICS[1] (16..31) q vectors 16-31
3848                                  * EICS[2] (0..31) q vectors 32-63
3849                                  */
3850                                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
3851                                                 (u32)(eics & 0xFFFF));
3852                                 IXGBE_WRITE_REG(hw, IXGBE_EICS_EX(1),
3853                                                 (u32)(eics & 0xFFFF0000));
3854                                 IXGBE_WRITE_REG(hw, IXGBE_EICS_EX(2),
3855                                                 (u32)(eics >> 32));
3856                         } else {
3857                                 /*
3858                                  * for legacy and MSI interrupts don't set any
3859                                  * bits that are enabled for EIAM, because this
3860                                  * operation would set *both* EIMS and EICS for
3861                                  * any bit in EIAM
3862                                  */
3863                                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
3864                                      (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
3865                         }
3866                         break;
3867                 default:
3868                         break;
3869                 }
3870                 /* Reset the timer */
3871                 mod_timer(&adapter->watchdog_timer,
3872                           round_jiffies(jiffies + 2 * HZ));
3873         }
3874
3875         schedule_work(&adapter->watchdog_task);
3876 }
3877
3878 /**
3879  * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
3880  * @work: pointer to work_struct containing our data
3881  **/
3882 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
3883 {
3884         struct ixgbe_adapter *adapter = container_of(work,
3885                                                      struct ixgbe_adapter,
3886                                                      multispeed_fiber_task);
3887         struct ixgbe_hw *hw = &adapter->hw;
3888         u32 autoneg;
3889
3890         adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
3891         if (hw->mac.ops.get_link_capabilities)
3892                 hw->mac.ops.get_link_capabilities(hw, &autoneg,
3893                                                   &hw->mac.autoneg);
3894         if (hw->mac.ops.setup_link_speed)
3895                 hw->mac.ops.setup_link_speed(hw, autoneg, true, true);
3896         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3897         adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
3898 }
3899
3900 /**
3901  * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
3902  * @work: pointer to work_struct containing our data
3903  **/
3904 static void ixgbe_sfp_config_module_task(struct work_struct *work)
3905 {
3906         struct ixgbe_adapter *adapter = container_of(work,
3907                                                      struct ixgbe_adapter,
3908                                                      sfp_config_module_task);
3909         struct ixgbe_hw *hw = &adapter->hw;
3910         u32 err;
3911
3912         adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
3913         err = hw->phy.ops.identify_sfp(hw);
3914         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3915                 DPRINTK(PROBE, ERR, "PHY not supported on this NIC %d\n", err);
3916                 ixgbe_down(adapter);
3917                 return;
3918         }
3919         hw->mac.ops.setup_sfp(hw);
3920
3921         if (!adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK)
3922                 /* This will also work for DA Twinax connections */
3923                 schedule_work(&adapter->multispeed_fiber_task);
3924         adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
3925 }
3926
3927 /**
3928  * ixgbe_watchdog_task - worker thread to bring link up
3929  * @work: pointer to work_struct containing our data
3930  **/
3931 static void ixgbe_watchdog_task(struct work_struct *work)
3932 {
3933         struct ixgbe_adapter *adapter = container_of(work,
3934                                                      struct ixgbe_adapter,
3935                                                      watchdog_task);
3936         struct net_device *netdev = adapter->netdev;
3937         struct ixgbe_hw *hw = &adapter->hw;
3938         u32 link_speed = adapter->link_speed;
3939         bool link_up = adapter->link_up;
3940
3941         adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
3942
3943         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
3944                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3945                 if (link_up ||
3946                     time_after(jiffies, (adapter->link_check_timeout +
3947                                          IXGBE_TRY_LINK_TIMEOUT))) {
3948                         IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
3949                         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
3950                 }
3951                 adapter->link_up = link_up;
3952                 adapter->link_speed = link_speed;
3953         }
3954
3955         if (link_up) {
3956                 if (!netif_carrier_ok(netdev)) {
3957                         bool flow_rx, flow_tx;
3958
3959                         if (hw->mac.type == ixgbe_mac_82599EB) {
3960                                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
3961                                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
3962                                 flow_rx = (mflcn & IXGBE_MFLCN_RFCE);
3963                                 flow_tx = (fccfg & IXGBE_FCCFG_TFCE_802_3X);
3964                         } else {
3965                                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3966                                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
3967                                 flow_rx = (frctl & IXGBE_FCTRL_RFCE);
3968                                 flow_tx = (rmcs & IXGBE_RMCS_TFCE_802_3X);
3969                         }
3970
3971                         printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
3972                                "Flow Control: %s\n",
3973                                netdev->name,
3974                                (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
3975                                 "10 Gbps" :
3976                                 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
3977                                  "1 Gbps" : "unknown speed")),
3978                                ((flow_rx && flow_tx) ? "RX/TX" :
3979                                 (flow_rx ? "RX" :
3980                                 (flow_tx ? "TX" : "None"))));
3981
3982                         netif_carrier_on(netdev);
3983                 } else {
3984                         /* Force detection of hung controller */
3985                         adapter->detect_tx_hung = true;
3986                 }
3987         } else {
3988                 adapter->link_up = false;
3989                 adapter->link_speed = 0;
3990                 if (netif_carrier_ok(netdev)) {
3991                         printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
3992                                netdev->name);
3993                         netif_carrier_off(netdev);
3994                 }
3995         }
3996
3997         ixgbe_update_stats(adapter);
3998         adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
3999 }
4000
4001 static int ixgbe_tso(struct ixgbe_adapter *adapter,
4002                      struct ixgbe_ring *tx_ring, struct sk_buff *skb,
4003                      u32 tx_flags, u8 *hdr_len)
4004 {
4005         struct ixgbe_adv_tx_context_desc *context_desc;
4006         unsigned int i;
4007         int err;
4008         struct ixgbe_tx_buffer *tx_buffer_info;
4009         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
4010         u32 mss_l4len_idx, l4len;
4011
4012         if (skb_is_gso(skb)) {
4013                 if (skb_header_cloned(skb)) {
4014                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4015                         if (err)
4016                                 return err;
4017                 }
4018                 l4len = tcp_hdrlen(skb);
4019                 *hdr_len += l4len;
4020
4021                 if (skb->protocol == htons(ETH_P_IP)) {
4022                         struct iphdr *iph = ip_hdr(skb);
4023                         iph->tot_len = 0;
4024                         iph->check = 0;
4025                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4026                                                                  iph->daddr, 0,
4027                                                                  IPPROTO_TCP,
4028                                                                  0);
4029                         adapter->hw_tso_ctxt++;
4030                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
4031                         ipv6_hdr(skb)->payload_len = 0;
4032                         tcp_hdr(skb)->check =
4033                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4034                                              &ipv6_hdr(skb)->daddr,
4035                                              0, IPPROTO_TCP, 0);
4036                         adapter->hw_tso6_ctxt++;
4037                 }
4038
4039                 i = tx_ring->next_to_use;
4040
4041                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4042                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4043
4044                 /* VLAN MACLEN IPLEN */
4045                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4046                         vlan_macip_lens |=
4047                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4048                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
4049                                     IXGBE_ADVTXD_MACLEN_SHIFT);
4050                 *hdr_len += skb_network_offset(skb);
4051                 vlan_macip_lens |=
4052                     (skb_transport_header(skb) - skb_network_header(skb));
4053                 *hdr_len +=
4054                     (skb_transport_header(skb) - skb_network_header(skb));
4055                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4056                 context_desc->seqnum_seed = 0;
4057
4058                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4059                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
4060                                    IXGBE_ADVTXD_DTYP_CTXT);
4061
4062                 if (skb->protocol == htons(ETH_P_IP))
4063                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4064                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
4065                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4066
4067                 /* MSS L4LEN IDX */
4068                 mss_l4len_idx =
4069                     (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
4070                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
4071                 /* use index 1 for TSO */
4072                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4073                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4074
4075                 tx_buffer_info->time_stamp = jiffies;
4076                 tx_buffer_info->next_to_watch = i;
4077
4078                 i++;
4079                 if (i == tx_ring->count)
4080                         i = 0;
4081                 tx_ring->next_to_use = i;
4082
4083                 return true;
4084         }
4085         return false;
4086 }
4087
4088 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
4089                           struct ixgbe_ring *tx_ring,
4090                           struct sk_buff *skb, u32 tx_flags)
4091 {
4092         struct ixgbe_adv_tx_context_desc *context_desc;
4093         unsigned int i;
4094         struct ixgbe_tx_buffer *tx_buffer_info;
4095         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
4096
4097         if (skb->ip_summed == CHECKSUM_PARTIAL ||
4098             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
4099                 i = tx_ring->next_to_use;
4100                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4101                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4102
4103                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4104                         vlan_macip_lens |=
4105                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4106                 vlan_macip_lens |= (skb_network_offset(skb) <<
4107                                     IXGBE_ADVTXD_MACLEN_SHIFT);
4108                 if (skb->ip_summed == CHECKSUM_PARTIAL)
4109                         vlan_macip_lens |= (skb_transport_header(skb) -
4110                                             skb_network_header(skb));
4111
4112                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4113                 context_desc->seqnum_seed = 0;
4114
4115                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
4116                                     IXGBE_ADVTXD_DTYP_CTXT);
4117
4118                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
4119                         switch (skb->protocol) {
4120                         case cpu_to_be16(ETH_P_IP):
4121                                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4122                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4123                                         type_tucmd_mlhl |=
4124                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4125                                 break;
4126                         case cpu_to_be16(ETH_P_IPV6):
4127                                 /* XXX what about other V6 headers?? */
4128                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4129                                         type_tucmd_mlhl |=
4130                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4131                                 break;
4132                         default:
4133                                 if (unlikely(net_ratelimit())) {
4134                                         DPRINTK(PROBE, WARNING,
4135                                          "partial checksum but proto=%x!\n",
4136                                          skb->protocol);
4137                                 }
4138                                 break;
4139                         }
4140                 }
4141
4142                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4143                 /* use index zero for tx checksum offload */
4144                 context_desc->mss_l4len_idx = 0;
4145
4146                 tx_buffer_info->time_stamp = jiffies;
4147                 tx_buffer_info->next_to_watch = i;
4148
4149                 adapter->hw_csum_tx_good++;
4150                 i++;
4151                 if (i == tx_ring->count)
4152                         i = 0;
4153                 tx_ring->next_to_use = i;
4154
4155                 return true;
4156         }
4157
4158         return false;
4159 }
4160
4161 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
4162                         struct ixgbe_ring *tx_ring,
4163                         struct sk_buff *skb, unsigned int first)
4164 {
4165         struct ixgbe_tx_buffer *tx_buffer_info;
4166         unsigned int len = skb->len;
4167         unsigned int offset = 0, size, count = 0, i;
4168         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
4169         unsigned int f;
4170
4171         len -= skb->data_len;
4172
4173         i = tx_ring->next_to_use;
4174
4175         while (len) {
4176                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4177                 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4178
4179                 tx_buffer_info->length = size;
4180                 tx_buffer_info->dma = pci_map_single(adapter->pdev,
4181                                                      skb->data + offset,
4182                                                      size, PCI_DMA_TODEVICE);
4183                 tx_buffer_info->time_stamp = jiffies;
4184                 tx_buffer_info->next_to_watch = i;
4185
4186                 len -= size;
4187                 offset += size;
4188                 count++;
4189                 i++;
4190                 if (i == tx_ring->count)
4191                         i = 0;
4192         }
4193
4194         for (f = 0; f < nr_frags; f++) {
4195                 struct skb_frag_struct *frag;
4196
4197                 frag = &skb_shinfo(skb)->frags[f];
4198                 len = frag->size;
4199                 offset = frag->page_offset;
4200
4201                 while (len) {
4202                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
4203                         size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4204
4205                         tx_buffer_info->length = size;
4206                         tx_buffer_info->dma = pci_map_page(adapter->pdev,
4207                                                            frag->page,
4208                                                            offset,
4209                                                            size,
4210                                                            PCI_DMA_TODEVICE);
4211                         tx_buffer_info->time_stamp = jiffies;
4212                         tx_buffer_info->next_to_watch = i;
4213
4214                         len -= size;
4215                         offset += size;
4216                         count++;
4217                         i++;
4218                         if (i == tx_ring->count)
4219                                 i = 0;
4220                 }
4221         }
4222         if (i == 0)
4223                 i = tx_ring->count - 1;
4224         else
4225                 i = i - 1;
4226         tx_ring->tx_buffer_info[i].skb = skb;
4227         tx_ring->tx_buffer_info[first].next_to_watch = i;
4228
4229         return count;
4230 }
4231
4232 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
4233                            struct ixgbe_ring *tx_ring,
4234                            int tx_flags, int count, u32 paylen, u8 hdr_len)
4235 {
4236         union ixgbe_adv_tx_desc *tx_desc = NULL;
4237         struct ixgbe_tx_buffer *tx_buffer_info;
4238         u32 olinfo_status = 0, cmd_type_len = 0;
4239         unsigned int i;
4240         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
4241
4242         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
4243
4244         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
4245
4246         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4247                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
4248
4249         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
4250                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
4251
4252                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
4253                                  IXGBE_ADVTXD_POPTS_SHIFT;
4254
4255                 /* use index 1 context for tso */
4256                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4257                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
4258                         olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
4259                                          IXGBE_ADVTXD_POPTS_SHIFT;
4260
4261         } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
4262                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
4263                                  IXGBE_ADVTXD_POPTS_SHIFT;
4264
4265         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
4266
4267         i = tx_ring->next_to_use;
4268         while (count--) {
4269                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4270                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
4271                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
4272                 tx_desc->read.cmd_type_len =
4273                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
4274                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4275                 i++;
4276                 if (i == tx_ring->count)
4277                         i = 0;
4278         }
4279
4280         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
4281
4282         /*
4283          * Force memory writes to complete before letting h/w
4284          * know there are new descriptors to fetch.  (Only
4285          * applicable for weak-ordered memory model archs,
4286          * such as IA-64).
4287          */
4288         wmb();
4289
4290         tx_ring->next_to_use = i;
4291         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4292 }
4293
4294 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
4295                                  struct ixgbe_ring *tx_ring, int size)
4296 {
4297         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4298
4299         netif_stop_subqueue(netdev, tx_ring->queue_index);
4300         /* Herbert's original patch had:
4301          *  smp_mb__after_netif_stop_queue();
4302          * but since that doesn't exist yet, just open code it. */
4303         smp_mb();
4304
4305         /* We need to check again in a case another CPU has just
4306          * made room available. */
4307         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
4308                 return -EBUSY;
4309
4310         /* A reprieve! - use start_queue because it doesn't call schedule */
4311         netif_start_subqueue(netdev, tx_ring->queue_index);
4312         ++adapter->restart_queue;
4313         return 0;
4314 }
4315
4316 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
4317                               struct ixgbe_ring *tx_ring, int size)
4318 {
4319         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
4320                 return 0;
4321         return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
4322 }
4323
4324 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
4325 {
4326         struct ixgbe_adapter *adapter = netdev_priv(dev);
4327
4328         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
4329                 return 0;  /* All traffic should default to class 0 */
4330
4331         return skb_tx_hash(dev, skb);
4332 }
4333
4334 static int ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4335 {
4336         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4337         struct ixgbe_ring *tx_ring;
4338         unsigned int first;
4339         unsigned int tx_flags = 0;
4340         u8 hdr_len = 0;
4341         int r_idx = 0, tso;
4342         int count = 0;
4343         unsigned int f;
4344
4345         r_idx = (adapter->num_tx_queues - 1) & skb->queue_mapping;
4346         tx_ring = &adapter->tx_ring[r_idx];
4347
4348         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4349                 tx_flags |= vlan_tx_tag_get(skb);
4350                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4351                         tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
4352                         tx_flags |= (skb->queue_mapping << 13);
4353                 }
4354                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4355                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4356         } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4357                 tx_flags |= (skb->queue_mapping << 13);
4358                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4359                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4360         }
4361         /* three things can cause us to need a context descriptor */
4362         if (skb_is_gso(skb) ||
4363             (skb->ip_summed == CHECKSUM_PARTIAL) ||
4364             (tx_flags & IXGBE_TX_FLAGS_VLAN))
4365                 count++;
4366
4367         count += TXD_USE_COUNT(skb_headlen(skb));
4368         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4369                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4370
4371         if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
4372                 adapter->tx_busy++;
4373                 return NETDEV_TX_BUSY;
4374         }
4375
4376         if (skb->protocol == htons(ETH_P_IP))
4377                 tx_flags |= IXGBE_TX_FLAGS_IPV4;
4378         first = tx_ring->next_to_use;
4379         tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
4380         if (tso < 0) {
4381                 dev_kfree_skb_any(skb);
4382                 return NETDEV_TX_OK;
4383         }
4384
4385         if (tso)
4386                 tx_flags |= IXGBE_TX_FLAGS_TSO;
4387         else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
4388                  (skb->ip_summed == CHECKSUM_PARTIAL))
4389                 tx_flags |= IXGBE_TX_FLAGS_CSUM;
4390
4391         ixgbe_tx_queue(adapter, tx_ring, tx_flags,
4392                        ixgbe_tx_map(adapter, tx_ring, skb, first),
4393                        skb->len, hdr_len);
4394
4395         netdev->trans_start = jiffies;
4396
4397         ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
4398
4399         return NETDEV_TX_OK;
4400 }
4401
4402 /**
4403  * ixgbe_get_stats - Get System Network Statistics
4404  * @netdev: network interface device structure
4405  *
4406  * Returns the address of the device statistics structure.
4407  * The statistics are actually updated from the timer callback.
4408  **/
4409 static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
4410 {
4411         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4412
4413         /* only return the current stats */
4414         return &adapter->net_stats;
4415 }
4416
4417 /**
4418  * ixgbe_set_mac - Change the Ethernet Address of the NIC
4419  * @netdev: network interface device structure
4420  * @p: pointer to an address structure
4421  *
4422  * Returns 0 on success, negative on failure
4423  **/
4424 static int ixgbe_set_mac(struct net_device *netdev, void *p)
4425 {
4426         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4427         struct ixgbe_hw *hw = &adapter->hw;
4428         struct sockaddr *addr = p;
4429
4430         if (!is_valid_ether_addr(addr->sa_data))
4431                 return -EADDRNOTAVAIL;
4432
4433         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4434         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
4435
4436         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
4437
4438         return 0;
4439 }
4440
4441 #ifdef CONFIG_NET_POLL_CONTROLLER
4442 /*
4443  * Polling 'interrupt' - used by things like netconsole to send skbs
4444  * without having to re-enable interrupts. It's not called while
4445  * the interrupt routine is executing.
4446  */
4447 static void ixgbe_netpoll(struct net_device *netdev)
4448 {
4449         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4450
4451         disable_irq(adapter->pdev->irq);
4452         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
4453         ixgbe_intr(adapter->pdev->irq, netdev);
4454         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
4455         enable_irq(adapter->pdev->irq);
4456 }
4457 #endif
4458
4459 static const struct net_device_ops ixgbe_netdev_ops = {
4460         .ndo_open               = ixgbe_open,
4461         .ndo_stop               = ixgbe_close,
4462         .ndo_start_xmit         = ixgbe_xmit_frame,
4463         .ndo_select_queue       = ixgbe_select_queue,
4464         .ndo_get_stats          = ixgbe_get_stats,
4465         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
4466         .ndo_set_multicast_list = ixgbe_set_rx_mode,
4467         .ndo_validate_addr      = eth_validate_addr,
4468         .ndo_set_mac_address    = ixgbe_set_mac,
4469         .ndo_change_mtu         = ixgbe_change_mtu,
4470         .ndo_tx_timeout         = ixgbe_tx_timeout,
4471         .ndo_vlan_rx_register   = ixgbe_vlan_rx_register,
4472         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
4473         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
4474 #ifdef CONFIG_NET_POLL_CONTROLLER
4475         .ndo_poll_controller    = ixgbe_netpoll,
4476 #endif
4477 };
4478
4479 /**
4480  * ixgbe_probe - Device Initialization Routine
4481  * @pdev: PCI device information struct
4482  * @ent: entry in ixgbe_pci_tbl
4483  *
4484  * Returns 0 on success, negative on failure
4485  *
4486  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
4487  * The OS initialization, configuring of the adapter private structure,
4488  * and a hardware reset occur.
4489  **/
4490 static int __devinit ixgbe_probe(struct pci_dev *pdev,
4491                                  const struct pci_device_id *ent)
4492 {
4493         struct net_device *netdev;
4494         struct ixgbe_adapter *adapter = NULL;
4495         struct ixgbe_hw *hw;
4496         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
4497         static int cards_found;
4498         int i, err, pci_using_dac;
4499         u16 pm_value = 0;
4500         u32 part_num, eec;
4501
4502         err = pci_enable_device(pdev);
4503         if (err)
4504                 return err;
4505
4506         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK) &&
4507             !pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK)) {
4508                 pci_using_dac = 1;
4509         } else {
4510                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4511                 if (err) {
4512                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
4513                         if (err) {
4514                                 dev_err(&pdev->dev, "No usable DMA "
4515                                         "configuration, aborting\n");
4516                                 goto err_dma;
4517                         }
4518                 }
4519                 pci_using_dac = 0;
4520         }
4521
4522         err = pci_request_regions(pdev, ixgbe_driver_name);
4523         if (err) {
4524                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4525                 goto err_pci_reg;
4526         }
4527
4528         err = pci_enable_pcie_error_reporting(pdev);
4529         if (err) {
4530                 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
4531                                     "0x%x\n", err);
4532                 /* non-fatal, continue */
4533         }
4534
4535         pci_set_master(pdev);
4536         pci_save_state(pdev);
4537
4538         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
4539         if (!netdev) {
4540                 err = -ENOMEM;
4541                 goto err_alloc_etherdev;
4542         }
4543
4544         SET_NETDEV_DEV(netdev, &pdev->dev);
4545
4546         pci_set_drvdata(pdev, netdev);
4547         adapter = netdev_priv(netdev);
4548
4549         adapter->netdev = netdev;
4550         adapter->pdev = pdev;
4551         hw = &adapter->hw;
4552         hw->back = adapter;
4553         adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
4554
4555         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4556                               pci_resource_len(pdev, 0));
4557         if (!hw->hw_addr) {
4558                 err = -EIO;
4559                 goto err_ioremap;
4560         }
4561
4562         for (i = 1; i <= 5; i++) {
4563                 if (pci_resource_len(pdev, i) == 0)
4564                         continue;
4565         }
4566
4567         netdev->netdev_ops = &ixgbe_netdev_ops;
4568         ixgbe_set_ethtool_ops(netdev);
4569         netdev->watchdog_timeo = 5 * HZ;
4570         strcpy(netdev->name, pci_name(pdev));
4571
4572         adapter->bd_number = cards_found;
4573
4574         /* Setup hw api */
4575         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4576         hw->mac.type  = ii->mac;
4577
4578         /* EEPROM */
4579         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
4580         eec = IXGBE_READ_REG(hw, IXGBE_EEC);
4581         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
4582         if (!(eec & (1 << 8)))
4583                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
4584
4585         /* PHY */
4586         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
4587         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
4588
4589         /* set up this timer and work struct before calling get_invariants
4590          * which might start the timer
4591          */
4592         init_timer(&adapter->sfp_timer);
4593         adapter->sfp_timer.function = &ixgbe_sfp_timer;
4594         adapter->sfp_timer.data = (unsigned long) adapter;
4595
4596         INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
4597
4598         /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
4599         INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
4600
4601         /* a new SFP+ module arrival, called from GPI SDP2 context */
4602         INIT_WORK(&adapter->sfp_config_module_task,
4603                   ixgbe_sfp_config_module_task);
4604
4605         err = ii->get_invariants(hw);
4606         if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
4607                 /* start a kernel thread to watch for a module to arrive */
4608                 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4609                 mod_timer(&adapter->sfp_timer,
4610                           round_jiffies(jiffies + (2 * HZ)));
4611                 err = 0;
4612         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4613                 DPRINTK(PROBE, ERR, "failed to load because an "
4614                         "unsupported SFP+ module type was detected.\n");
4615                 goto err_hw_init;
4616         } else if (err) {
4617                 goto err_hw_init;
4618         }
4619
4620         /* setup the private structure */
4621         err = ixgbe_sw_init(adapter);
4622         if (err)
4623                 goto err_sw_init;
4624
4625         /* reset_hw fills in the perm_addr as well */
4626         err = hw->mac.ops.reset_hw(hw);
4627         if (err) {
4628                 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
4629                 goto err_sw_init;
4630         }
4631
4632         netdev->features = NETIF_F_SG |
4633                            NETIF_F_IP_CSUM |
4634                            NETIF_F_HW_VLAN_TX |
4635                            NETIF_F_HW_VLAN_RX |
4636                            NETIF_F_HW_VLAN_FILTER;
4637
4638         netdev->features |= NETIF_F_IPV6_CSUM;
4639         netdev->features |= NETIF_F_TSO;
4640         netdev->features |= NETIF_F_TSO6;
4641         netdev->features |= NETIF_F_GRO;
4642
4643         netdev->vlan_features |= NETIF_F_TSO;
4644         netdev->vlan_features |= NETIF_F_TSO6;
4645         netdev->vlan_features |= NETIF_F_IP_CSUM;
4646         netdev->vlan_features |= NETIF_F_SG;
4647
4648         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
4649                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4650
4651 #ifdef CONFIG_IXGBE_DCB
4652         netdev->dcbnl_ops = &dcbnl_ops;
4653 #endif
4654
4655         if (pci_using_dac)
4656                 netdev->features |= NETIF_F_HIGHDMA;
4657
4658         /* make sure the EEPROM is good */
4659         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
4660                 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
4661                 err = -EIO;
4662                 goto err_eeprom;
4663         }
4664
4665         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
4666         memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
4667
4668         if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
4669                 dev_err(&pdev->dev, "invalid MAC address\n");
4670                 err = -EIO;
4671                 goto err_eeprom;
4672         }
4673
4674         init_timer(&adapter->watchdog_timer);
4675         adapter->watchdog_timer.function = &ixgbe_watchdog;
4676         adapter->watchdog_timer.data = (unsigned long)adapter;
4677
4678         INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
4679         INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
4680
4681         err = ixgbe_init_interrupt_scheme(adapter);
4682         if (err)
4683                 goto err_sw_init;
4684
4685         switch (pdev->device) {
4686         case IXGBE_DEV_ID_82599_KX4:
4687 #define IXGBE_PCIE_PMCSR 0x44
4688                 adapter->wol = IXGBE_WUFC_MAG;
4689                 pci_read_config_word(pdev, IXGBE_PCIE_PMCSR, &pm_value);
4690                 pci_write_config_word(pdev, IXGBE_PCIE_PMCSR,
4691                                       (pm_value | (1 << 8)));
4692                 break;
4693         default:
4694                 adapter->wol = 0;
4695                 break;
4696         }
4697         device_init_wakeup(&adapter->pdev->dev, true);
4698         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
4699
4700         /* print bus type/speed/width info */
4701         dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
4702                 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
4703                  (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
4704                 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
4705                  (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
4706                  (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
4707                  "Unknown"),
4708                 netdev->dev_addr);
4709         ixgbe_read_pba_num_generic(hw, &part_num);
4710         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
4711                 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
4712                          hw->mac.type, hw->phy.type, hw->phy.sfp_type,
4713                          (part_num >> 8), (part_num & 0xff));
4714         else
4715                 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4716                          hw->mac.type, hw->phy.type,
4717                          (part_num >> 8), (part_num & 0xff));
4718
4719         if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
4720                 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
4721                          "this card is not sufficient for optimal "
4722                          "performance.\n");
4723                 dev_warn(&pdev->dev, "For optimal performance a x8 "
4724                          "PCI-Express slot is required.\n");
4725         }
4726
4727         /* save off EEPROM version number */
4728         hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
4729
4730         /* reset the hardware with the new settings */
4731         hw->mac.ops.start_hw(hw);
4732
4733         netif_carrier_off(netdev);
4734
4735         strcpy(netdev->name, "eth%d");
4736         err = register_netdev(netdev);
4737         if (err)
4738                 goto err_register;
4739
4740 #ifdef CONFIG_IXGBE_DCA
4741         if (dca_add_requester(&pdev->dev) == 0) {
4742                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
4743                 /* always use CB2 mode, difference is masked
4744                  * in the CB driver */
4745                 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
4746                 ixgbe_setup_dca(adapter);
4747         }
4748 #endif
4749
4750         dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
4751         cards_found++;
4752         return 0;
4753
4754 err_register:
4755         ixgbe_release_hw_control(adapter);
4756 err_hw_init:
4757 err_sw_init:
4758         ixgbe_reset_interrupt_capability(adapter);
4759 err_eeprom:
4760         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4761         del_timer_sync(&adapter->sfp_timer);
4762         cancel_work_sync(&adapter->sfp_task);
4763         cancel_work_sync(&adapter->multispeed_fiber_task);
4764         cancel_work_sync(&adapter->sfp_config_module_task);
4765         iounmap(hw->hw_addr);
4766 err_ioremap:
4767         free_netdev(netdev);
4768 err_alloc_etherdev:
4769         pci_release_regions(pdev);
4770 err_pci_reg:
4771 err_dma:
4772         pci_disable_device(pdev);
4773         return err;
4774 }
4775
4776 /**
4777  * ixgbe_remove - Device Removal Routine
4778  * @pdev: PCI device information struct
4779  *
4780  * ixgbe_remove is called by the PCI subsystem to alert the driver
4781  * that it should release a PCI device.  The could be caused by a
4782  * Hot-Plug event, or because the driver is going to be removed from
4783  * memory.
4784  **/
4785 static void __devexit ixgbe_remove(struct pci_dev *pdev)
4786 {
4787         struct net_device *netdev = pci_get_drvdata(pdev);
4788         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4789         int err;
4790
4791         set_bit(__IXGBE_DOWN, &adapter->state);
4792         /* clear the module not found bit to make sure the worker won't
4793          * reschedule
4794          */
4795         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4796         del_timer_sync(&adapter->watchdog_timer);
4797
4798         del_timer_sync(&adapter->sfp_timer);
4799         cancel_work_sync(&adapter->watchdog_task);
4800         cancel_work_sync(&adapter->sfp_task);
4801         cancel_work_sync(&adapter->multispeed_fiber_task);
4802         cancel_work_sync(&adapter->sfp_config_module_task);
4803         flush_scheduled_work();
4804
4805 #ifdef CONFIG_IXGBE_DCA
4806         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
4807                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
4808                 dca_remove_requester(&pdev->dev);
4809                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
4810         }
4811
4812 #endif
4813         if (netdev->reg_state == NETREG_REGISTERED)
4814                 unregister_netdev(netdev);
4815
4816         ixgbe_reset_interrupt_capability(adapter);
4817
4818         ixgbe_release_hw_control(adapter);
4819
4820         iounmap(adapter->hw.hw_addr);
4821         pci_release_regions(pdev);
4822
4823         DPRINTK(PROBE, INFO, "complete\n");
4824         kfree(adapter->tx_ring);
4825         kfree(adapter->rx_ring);
4826
4827         free_netdev(netdev);
4828
4829         err = pci_disable_pcie_error_reporting(pdev);
4830         if (err)
4831                 dev_err(&pdev->dev,
4832                         "pci_disable_pcie_error_reporting failed 0x%x\n", err);
4833
4834         pci_disable_device(pdev);
4835 }
4836
4837 /**
4838  * ixgbe_io_error_detected - called when PCI error is detected
4839  * @pdev: Pointer to PCI device
4840  * @state: The current pci connection state
4841  *
4842  * This function is called after a PCI bus error affecting
4843  * this device has been detected.
4844  */
4845 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
4846                                                 pci_channel_state_t state)
4847 {
4848         struct net_device *netdev = pci_get_drvdata(pdev);
4849         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4850
4851         netif_device_detach(netdev);
4852
4853         if (netif_running(netdev))
4854                 ixgbe_down(adapter);
4855         pci_disable_device(pdev);
4856
4857         /* Request a slot reset. */
4858         return PCI_ERS_RESULT_NEED_RESET;
4859 }
4860
4861 /**
4862  * ixgbe_io_slot_reset - called after the pci bus has been reset.
4863  * @pdev: Pointer to PCI device
4864  *
4865  * Restart the card from scratch, as if from a cold-boot.
4866  */
4867 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
4868 {
4869         struct net_device *netdev = pci_get_drvdata(pdev);
4870         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4871         pci_ers_result_t result;
4872         int err;
4873
4874         if (pci_enable_device(pdev)) {
4875                 DPRINTK(PROBE, ERR,
4876                         "Cannot re-enable PCI device after reset.\n");
4877                 result = PCI_ERS_RESULT_DISCONNECT;
4878         } else {
4879                 pci_set_master(pdev);
4880                 pci_restore_state(pdev);
4881
4882                 pci_enable_wake(pdev, PCI_D3hot, 0);
4883                 pci_enable_wake(pdev, PCI_D3cold, 0);
4884
4885                 ixgbe_reset(adapter);
4886                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4887                 result = PCI_ERS_RESULT_RECOVERED;
4888         }
4889
4890         err = pci_cleanup_aer_uncorrect_error_status(pdev);
4891         if (err) {
4892                 dev_err(&pdev->dev,
4893                   "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
4894                 /* non-fatal, continue */
4895         }
4896
4897         return result;
4898 }
4899
4900 /**
4901  * ixgbe_io_resume - called when traffic can start flowing again.
4902  * @pdev: Pointer to PCI device
4903  *
4904  * This callback is called when the error recovery driver tells us that
4905  * its OK to resume normal operation.
4906  */
4907 static void ixgbe_io_resume(struct pci_dev *pdev)
4908 {
4909         struct net_device *netdev = pci_get_drvdata(pdev);
4910         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4911
4912         if (netif_running(netdev)) {
4913                 if (ixgbe_up(adapter)) {
4914                         DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
4915                         return;
4916                 }
4917         }
4918
4919         netif_device_attach(netdev);
4920 }
4921
4922 static struct pci_error_handlers ixgbe_err_handler = {
4923         .error_detected = ixgbe_io_error_detected,
4924         .slot_reset = ixgbe_io_slot_reset,
4925         .resume = ixgbe_io_resume,
4926 };
4927
4928 static struct pci_driver ixgbe_driver = {
4929         .name     = ixgbe_driver_name,
4930         .id_table = ixgbe_pci_tbl,
4931         .probe    = ixgbe_probe,
4932         .remove   = __devexit_p(ixgbe_remove),
4933 #ifdef CONFIG_PM
4934         .suspend  = ixgbe_suspend,
4935         .resume   = ixgbe_resume,
4936 #endif
4937         .shutdown = ixgbe_shutdown,
4938         .err_handler = &ixgbe_err_handler
4939 };
4940
4941 /**
4942  * ixgbe_init_module - Driver Registration Routine
4943  *
4944  * ixgbe_init_module is the first routine called when the driver is
4945  * loaded. All it does is register with the PCI subsystem.
4946  **/
4947 static int __init ixgbe_init_module(void)
4948 {
4949         int ret;
4950         printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
4951                ixgbe_driver_string, ixgbe_driver_version);
4952
4953         printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
4954
4955 #ifdef CONFIG_IXGBE_DCA
4956         dca_register_notify(&dca_notifier);
4957 #endif
4958
4959         ret = pci_register_driver(&ixgbe_driver);
4960         return ret;
4961 }
4962
4963 module_init(ixgbe_init_module);
4964
4965 /**
4966  * ixgbe_exit_module - Driver Exit Cleanup Routine
4967  *
4968  * ixgbe_exit_module is called just before the driver is removed
4969  * from memory.
4970  **/
4971 static void __exit ixgbe_exit_module(void)
4972 {
4973 #ifdef CONFIG_IXGBE_DCA
4974         dca_unregister_notify(&dca_notifier);
4975 #endif
4976         pci_unregister_driver(&ixgbe_driver);
4977 }
4978
4979 #ifdef CONFIG_IXGBE_DCA
4980 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
4981                             void *p)
4982 {
4983         int ret_val;
4984
4985         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
4986                                          __ixgbe_notify_dca);
4987
4988         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4989 }
4990 #endif /* CONFIG_IXGBE_DCA */
4991
4992 module_exit(ixgbe_exit_module);
4993
4994 /* ixgbe_main.c */