Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[deliverable/linux.git] / drivers / net / ixgbe / ixgbe_main.c
1 /*******************************************************************************
2
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2009 Intel Corporation.
5
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
9
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 more details.
14
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
21
22 Contact Information:
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
34 #include <linux/in.h>
35 #include <linux/ip.h>
36 #include <linux/tcp.h>
37 #include <linux/pkt_sched.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/ethtool.h>
42 #include <linux/if_vlan.h>
43 #include <scsi/fc/fc_fcoe.h>
44
45 #include "ixgbe.h"
46 #include "ixgbe_common.h"
47
48 char ixgbe_driver_name[] = "ixgbe";
49 static const char ixgbe_driver_string[] =
50 "Intel(R) 10 Gigabit PCI Express Network Driver";
51
52 #define DRV_VERSION "2.0.44-k2"
53 const char ixgbe_driver_version[] = DRV_VERSION;
54 static char ixgbe_copyright[] = "Copyright (c) 1999-2009 Intel Corporation.";
55
56 static const struct ixgbe_info *ixgbe_info_tbl[] = {
57 [board_82598] = &ixgbe_82598_info,
58 [board_82599] = &ixgbe_82599_info,
59 };
60
61 /* ixgbe_pci_tbl - PCI Device ID Table
62 *
63 * Wildcard entries (PCI_ANY_ID) should come last
64 * Last entry must be all 0s
65 *
66 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
67 * Class, Class Mask, private data (not used) }
68 */
69 static struct pci_device_id ixgbe_pci_tbl[] = {
70 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
71 board_82598 },
72 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
73 board_82598 },
74 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
75 board_82598 },
76 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
77 board_82598 },
78 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
79 board_82598 },
80 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
81 board_82598 },
82 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
83 board_82598 },
84 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
85 board_82598 },
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
87 board_82598 },
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
89 board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
91 board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
93 board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
95 board_82599 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
97 board_82599 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
99 board_82599 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
101 board_82599 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
103 board_82599 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
105 board_82599 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
107 board_82599 },
108
109 /* required last entry */
110 {0, }
111 };
112 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
113
114 #ifdef CONFIG_IXGBE_DCA
115 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
116 void *p);
117 static struct notifier_block dca_notifier = {
118 .notifier_call = ixgbe_notify_dca,
119 .next = NULL,
120 .priority = 0
121 };
122 #endif
123
124 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
125 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
126 MODULE_LICENSE("GPL");
127 MODULE_VERSION(DRV_VERSION);
128
129 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
130
131 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
132 {
133 u32 ctrl_ext;
134
135 /* Let firmware take over control of h/w */
136 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
137 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
138 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
139 }
140
141 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
142 {
143 u32 ctrl_ext;
144
145 /* Let firmware know the driver has taken over */
146 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
147 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
148 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
149 }
150
151 /*
152 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
153 * @adapter: pointer to adapter struct
154 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
155 * @queue: queue to map the corresponding interrupt to
156 * @msix_vector: the vector to map to the corresponding queue
157 *
158 */
159 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
160 u8 queue, u8 msix_vector)
161 {
162 u32 ivar, index;
163 struct ixgbe_hw *hw = &adapter->hw;
164 switch (hw->mac.type) {
165 case ixgbe_mac_82598EB:
166 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
167 if (direction == -1)
168 direction = 0;
169 index = (((direction * 64) + queue) >> 2) & 0x1F;
170 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
171 ivar &= ~(0xFF << (8 * (queue & 0x3)));
172 ivar |= (msix_vector << (8 * (queue & 0x3)));
173 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
174 break;
175 case ixgbe_mac_82599EB:
176 if (direction == -1) {
177 /* other causes */
178 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
179 index = ((queue & 1) * 8);
180 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
181 ivar &= ~(0xFF << index);
182 ivar |= (msix_vector << index);
183 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
184 break;
185 } else {
186 /* tx or rx causes */
187 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
188 index = ((16 * (queue & 1)) + (8 * direction));
189 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
190 ivar &= ~(0xFF << index);
191 ivar |= (msix_vector << index);
192 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
193 break;
194 }
195 default:
196 break;
197 }
198 }
199
200 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
201 u64 qmask)
202 {
203 u32 mask;
204
205 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
206 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
207 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
208 } else {
209 mask = (qmask & 0xFFFFFFFF);
210 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
211 mask = (qmask >> 32);
212 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
213 }
214 }
215
216 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
217 struct ixgbe_tx_buffer
218 *tx_buffer_info)
219 {
220 tx_buffer_info->dma = 0;
221 if (tx_buffer_info->skb) {
222 skb_dma_unmap(&adapter->pdev->dev, tx_buffer_info->skb,
223 DMA_TO_DEVICE);
224 dev_kfree_skb_any(tx_buffer_info->skb);
225 tx_buffer_info->skb = NULL;
226 }
227 tx_buffer_info->time_stamp = 0;
228 /* tx_buffer_info must be completely set up in the transmit path */
229 }
230
231 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
232 struct ixgbe_ring *tx_ring,
233 unsigned int eop)
234 {
235 struct ixgbe_hw *hw = &adapter->hw;
236
237 /* Detect a transmit hang in hardware, this serializes the
238 * check with the clearing of time_stamp and movement of eop */
239 adapter->detect_tx_hung = false;
240 if (tx_ring->tx_buffer_info[eop].time_stamp &&
241 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
242 !(IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & IXGBE_TFCS_TXOFF)) {
243 /* detected Tx unit hang */
244 union ixgbe_adv_tx_desc *tx_desc;
245 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
246 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
247 " Tx Queue <%d>\n"
248 " TDH, TDT <%x>, <%x>\n"
249 " next_to_use <%x>\n"
250 " next_to_clean <%x>\n"
251 "tx_buffer_info[next_to_clean]\n"
252 " time_stamp <%lx>\n"
253 " jiffies <%lx>\n",
254 tx_ring->queue_index,
255 IXGBE_READ_REG(hw, tx_ring->head),
256 IXGBE_READ_REG(hw, tx_ring->tail),
257 tx_ring->next_to_use, eop,
258 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
259 return true;
260 }
261
262 return false;
263 }
264
265 #define IXGBE_MAX_TXD_PWR 14
266 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
267
268 /* Tx Descriptors needed, worst case */
269 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
270 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
271 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
272 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
273
274 static void ixgbe_tx_timeout(struct net_device *netdev);
275
276 /**
277 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
278 * @q_vector: structure containing interrupt and ring information
279 * @tx_ring: tx ring to clean
280 **/
281 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
282 struct ixgbe_ring *tx_ring)
283 {
284 struct ixgbe_adapter *adapter = q_vector->adapter;
285 struct net_device *netdev = adapter->netdev;
286 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
287 struct ixgbe_tx_buffer *tx_buffer_info;
288 unsigned int i, eop, count = 0;
289 unsigned int total_bytes = 0, total_packets = 0;
290
291 i = tx_ring->next_to_clean;
292 eop = tx_ring->tx_buffer_info[i].next_to_watch;
293 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
294
295 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
296 (count < tx_ring->work_limit)) {
297 bool cleaned = false;
298 for ( ; !cleaned; count++) {
299 struct sk_buff *skb;
300 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
301 tx_buffer_info = &tx_ring->tx_buffer_info[i];
302 cleaned = (i == eop);
303 skb = tx_buffer_info->skb;
304
305 if (cleaned && skb) {
306 unsigned int segs, bytecount;
307 unsigned int hlen = skb_headlen(skb);
308
309 /* gso_segs is currently only valid for tcp */
310 segs = skb_shinfo(skb)->gso_segs ?: 1;
311 #ifdef IXGBE_FCOE
312 /* adjust for FCoE Sequence Offload */
313 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
314 && (skb->protocol == htons(ETH_P_FCOE)) &&
315 skb_is_gso(skb)) {
316 hlen = skb_transport_offset(skb) +
317 sizeof(struct fc_frame_header) +
318 sizeof(struct fcoe_crc_eof);
319 segs = DIV_ROUND_UP(skb->len - hlen,
320 skb_shinfo(skb)->gso_size);
321 }
322 #endif /* IXGBE_FCOE */
323 /* multiply data chunks by size of headers */
324 bytecount = ((segs - 1) * hlen) + skb->len;
325 total_packets += segs;
326 total_bytes += bytecount;
327 }
328
329 ixgbe_unmap_and_free_tx_resource(adapter,
330 tx_buffer_info);
331
332 tx_desc->wb.status = 0;
333
334 i++;
335 if (i == tx_ring->count)
336 i = 0;
337 }
338
339 eop = tx_ring->tx_buffer_info[i].next_to_watch;
340 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
341 }
342
343 tx_ring->next_to_clean = i;
344
345 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
346 if (unlikely(count && netif_carrier_ok(netdev) &&
347 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
348 /* Make sure that anybody stopping the queue after this
349 * sees the new next_to_clean.
350 */
351 smp_mb();
352 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
353 !test_bit(__IXGBE_DOWN, &adapter->state)) {
354 netif_wake_subqueue(netdev, tx_ring->queue_index);
355 ++adapter->restart_queue;
356 }
357 }
358
359 if (adapter->detect_tx_hung) {
360 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
361 /* schedule immediate reset if we believe we hung */
362 DPRINTK(PROBE, INFO,
363 "tx hang %d detected, resetting adapter\n",
364 adapter->tx_timeout_count + 1);
365 ixgbe_tx_timeout(adapter->netdev);
366 }
367 }
368
369 /* re-arm the interrupt */
370 if (count >= tx_ring->work_limit)
371 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
372
373 tx_ring->total_bytes += total_bytes;
374 tx_ring->total_packets += total_packets;
375 tx_ring->stats.packets += total_packets;
376 tx_ring->stats.bytes += total_bytes;
377 netdev->stats.tx_bytes += total_bytes;
378 netdev->stats.tx_packets += total_packets;
379 return (count < tx_ring->work_limit);
380 }
381
382 #ifdef CONFIG_IXGBE_DCA
383 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
384 struct ixgbe_ring *rx_ring)
385 {
386 u32 rxctrl;
387 int cpu = get_cpu();
388 int q = rx_ring - adapter->rx_ring;
389
390 if (rx_ring->cpu != cpu) {
391 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
392 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
393 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
394 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
395 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
396 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
397 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
398 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
399 }
400 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
401 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
402 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
403 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
404 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
405 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
406 rx_ring->cpu = cpu;
407 }
408 put_cpu();
409 }
410
411 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
412 struct ixgbe_ring *tx_ring)
413 {
414 u32 txctrl;
415 int cpu = get_cpu();
416 int q = tx_ring - adapter->tx_ring;
417
418 if (tx_ring->cpu != cpu) {
419 txctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q));
420 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
421 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
422 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
423 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
424 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
425 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
426 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
427 }
428 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
429 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q), txctrl);
430 tx_ring->cpu = cpu;
431 }
432 put_cpu();
433 }
434
435 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
436 {
437 int i;
438
439 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
440 return;
441
442 /* always use CB2 mode, difference is masked in the CB driver */
443 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
444
445 for (i = 0; i < adapter->num_tx_queues; i++) {
446 adapter->tx_ring[i].cpu = -1;
447 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
448 }
449 for (i = 0; i < adapter->num_rx_queues; i++) {
450 adapter->rx_ring[i].cpu = -1;
451 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
452 }
453 }
454
455 static int __ixgbe_notify_dca(struct device *dev, void *data)
456 {
457 struct net_device *netdev = dev_get_drvdata(dev);
458 struct ixgbe_adapter *adapter = netdev_priv(netdev);
459 unsigned long event = *(unsigned long *)data;
460
461 switch (event) {
462 case DCA_PROVIDER_ADD:
463 /* if we're already enabled, don't do it again */
464 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
465 break;
466 if (dca_add_requester(dev) == 0) {
467 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
468 ixgbe_setup_dca(adapter);
469 break;
470 }
471 /* Fall Through since DCA is disabled. */
472 case DCA_PROVIDER_REMOVE:
473 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
474 dca_remove_requester(dev);
475 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
476 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
477 }
478 break;
479 }
480
481 return 0;
482 }
483
484 #endif /* CONFIG_IXGBE_DCA */
485 /**
486 * ixgbe_receive_skb - Send a completed packet up the stack
487 * @adapter: board private structure
488 * @skb: packet to send up
489 * @status: hardware indication of status of receive
490 * @rx_ring: rx descriptor ring (for a specific queue) to setup
491 * @rx_desc: rx descriptor
492 **/
493 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
494 struct sk_buff *skb, u8 status,
495 struct ixgbe_ring *ring,
496 union ixgbe_adv_rx_desc *rx_desc)
497 {
498 struct ixgbe_adapter *adapter = q_vector->adapter;
499 struct napi_struct *napi = &q_vector->napi;
500 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
501 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
502
503 skb_record_rx_queue(skb, ring->queue_index);
504 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
505 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
506 vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
507 else
508 napi_gro_receive(napi, skb);
509 } else {
510 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
511 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
512 else
513 netif_rx(skb);
514 }
515 }
516
517 /**
518 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
519 * @adapter: address of board private structure
520 * @status_err: hardware indication of status of receive
521 * @skb: skb currently being received and modified
522 **/
523 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
524 union ixgbe_adv_rx_desc *rx_desc,
525 struct sk_buff *skb)
526 {
527 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
528
529 skb->ip_summed = CHECKSUM_NONE;
530
531 /* Rx csum disabled */
532 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
533 return;
534
535 /* if IP and error */
536 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
537 (status_err & IXGBE_RXDADV_ERR_IPE)) {
538 adapter->hw_csum_rx_error++;
539 return;
540 }
541
542 if (!(status_err & IXGBE_RXD_STAT_L4CS))
543 return;
544
545 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
546 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
547
548 /*
549 * 82599 errata, UDP frames with a 0 checksum can be marked as
550 * checksum errors.
551 */
552 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
553 (adapter->hw.mac.type == ixgbe_mac_82599EB))
554 return;
555
556 adapter->hw_csum_rx_error++;
557 return;
558 }
559
560 /* It must be a TCP or UDP packet with a valid checksum */
561 skb->ip_summed = CHECKSUM_UNNECESSARY;
562 adapter->hw_csum_rx_good++;
563 }
564
565 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
566 struct ixgbe_ring *rx_ring, u32 val)
567 {
568 /*
569 * Force memory writes to complete before letting h/w
570 * know there are new descriptors to fetch. (Only
571 * applicable for weak-ordered memory model archs,
572 * such as IA-64).
573 */
574 wmb();
575 IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
576 }
577
578 /**
579 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
580 * @adapter: address of board private structure
581 **/
582 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
583 struct ixgbe_ring *rx_ring,
584 int cleaned_count)
585 {
586 struct pci_dev *pdev = adapter->pdev;
587 union ixgbe_adv_rx_desc *rx_desc;
588 struct ixgbe_rx_buffer *bi;
589 unsigned int i;
590
591 i = rx_ring->next_to_use;
592 bi = &rx_ring->rx_buffer_info[i];
593
594 while (cleaned_count--) {
595 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
596
597 if (!bi->page_dma &&
598 (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
599 if (!bi->page) {
600 bi->page = alloc_page(GFP_ATOMIC);
601 if (!bi->page) {
602 adapter->alloc_rx_page_failed++;
603 goto no_buffers;
604 }
605 bi->page_offset = 0;
606 } else {
607 /* use a half page if we're re-using */
608 bi->page_offset ^= (PAGE_SIZE / 2);
609 }
610
611 bi->page_dma = pci_map_page(pdev, bi->page,
612 bi->page_offset,
613 (PAGE_SIZE / 2),
614 PCI_DMA_FROMDEVICE);
615 }
616
617 if (!bi->skb) {
618 struct sk_buff *skb;
619 skb = netdev_alloc_skb_ip_align(adapter->netdev,
620 rx_ring->rx_buf_len);
621
622 if (!skb) {
623 adapter->alloc_rx_buff_failed++;
624 goto no_buffers;
625 }
626
627 bi->skb = skb;
628 bi->dma = pci_map_single(pdev, skb->data,
629 rx_ring->rx_buf_len,
630 PCI_DMA_FROMDEVICE);
631 }
632 /* Refresh the desc even if buffer_addrs didn't change because
633 * each write-back erases this info. */
634 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
635 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
636 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
637 } else {
638 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
639 }
640
641 i++;
642 if (i == rx_ring->count)
643 i = 0;
644 bi = &rx_ring->rx_buffer_info[i];
645 }
646
647 no_buffers:
648 if (rx_ring->next_to_use != i) {
649 rx_ring->next_to_use = i;
650 if (i-- == 0)
651 i = (rx_ring->count - 1);
652
653 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
654 }
655 }
656
657 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
658 {
659 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
660 }
661
662 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
663 {
664 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
665 }
666
667 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
668 {
669 return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
670 IXGBE_RXDADV_RSCCNT_MASK) >>
671 IXGBE_RXDADV_RSCCNT_SHIFT;
672 }
673
674 /**
675 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
676 * @skb: pointer to the last skb in the rsc queue
677 *
678 * This function changes a queue full of hw rsc buffers into a completed
679 * packet. It uses the ->prev pointers to find the first packet and then
680 * turns it into the frag list owner.
681 **/
682 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
683 {
684 unsigned int frag_list_size = 0;
685
686 while (skb->prev) {
687 struct sk_buff *prev = skb->prev;
688 frag_list_size += skb->len;
689 skb->prev = NULL;
690 skb = prev;
691 }
692
693 skb_shinfo(skb)->frag_list = skb->next;
694 skb->next = NULL;
695 skb->len += frag_list_size;
696 skb->data_len += frag_list_size;
697 skb->truesize += frag_list_size;
698 return skb;
699 }
700
701 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
702 struct ixgbe_ring *rx_ring,
703 int *work_done, int work_to_do)
704 {
705 struct ixgbe_adapter *adapter = q_vector->adapter;
706 struct net_device *netdev = adapter->netdev;
707 struct pci_dev *pdev = adapter->pdev;
708 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
709 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
710 struct sk_buff *skb;
711 unsigned int i, rsc_count = 0;
712 u32 len, staterr;
713 u16 hdr_info;
714 bool cleaned = false;
715 int cleaned_count = 0;
716 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
717 #ifdef IXGBE_FCOE
718 int ddp_bytes = 0;
719 #endif /* IXGBE_FCOE */
720
721 i = rx_ring->next_to_clean;
722 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
723 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
724 rx_buffer_info = &rx_ring->rx_buffer_info[i];
725
726 while (staterr & IXGBE_RXD_STAT_DD) {
727 u32 upper_len = 0;
728 if (*work_done >= work_to_do)
729 break;
730 (*work_done)++;
731
732 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
733 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
734 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
735 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
736 if (hdr_info & IXGBE_RXDADV_SPH)
737 adapter->rx_hdr_split++;
738 if (len > IXGBE_RX_HDR_SIZE)
739 len = IXGBE_RX_HDR_SIZE;
740 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
741 } else {
742 len = le16_to_cpu(rx_desc->wb.upper.length);
743 }
744
745 cleaned = true;
746 skb = rx_buffer_info->skb;
747 prefetch(skb->data - NET_IP_ALIGN);
748 rx_buffer_info->skb = NULL;
749
750 if (rx_buffer_info->dma) {
751 pci_unmap_single(pdev, rx_buffer_info->dma,
752 rx_ring->rx_buf_len,
753 PCI_DMA_FROMDEVICE);
754 rx_buffer_info->dma = 0;
755 skb_put(skb, len);
756 }
757
758 if (upper_len) {
759 pci_unmap_page(pdev, rx_buffer_info->page_dma,
760 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
761 rx_buffer_info->page_dma = 0;
762 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
763 rx_buffer_info->page,
764 rx_buffer_info->page_offset,
765 upper_len);
766
767 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
768 (page_count(rx_buffer_info->page) != 1))
769 rx_buffer_info->page = NULL;
770 else
771 get_page(rx_buffer_info->page);
772
773 skb->len += upper_len;
774 skb->data_len += upper_len;
775 skb->truesize += upper_len;
776 }
777
778 i++;
779 if (i == rx_ring->count)
780 i = 0;
781
782 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
783 prefetch(next_rxd);
784 cleaned_count++;
785
786 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
787 rsc_count = ixgbe_get_rsc_count(rx_desc);
788
789 if (rsc_count) {
790 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
791 IXGBE_RXDADV_NEXTP_SHIFT;
792 next_buffer = &rx_ring->rx_buffer_info[nextp];
793 rx_ring->rsc_count += (rsc_count - 1);
794 } else {
795 next_buffer = &rx_ring->rx_buffer_info[i];
796 }
797
798 if (staterr & IXGBE_RXD_STAT_EOP) {
799 if (skb->prev)
800 skb = ixgbe_transform_rsc_queue(skb);
801 rx_ring->stats.packets++;
802 rx_ring->stats.bytes += skb->len;
803 } else {
804 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
805 rx_buffer_info->skb = next_buffer->skb;
806 rx_buffer_info->dma = next_buffer->dma;
807 next_buffer->skb = skb;
808 next_buffer->dma = 0;
809 } else {
810 skb->next = next_buffer->skb;
811 skb->next->prev = skb;
812 }
813 adapter->non_eop_descs++;
814 goto next_desc;
815 }
816
817 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
818 dev_kfree_skb_irq(skb);
819 goto next_desc;
820 }
821
822 ixgbe_rx_checksum(adapter, rx_desc, skb);
823
824 /* probably a little skewed due to removing CRC */
825 total_rx_bytes += skb->len;
826 total_rx_packets++;
827
828 skb->protocol = eth_type_trans(skb, adapter->netdev);
829 #ifdef IXGBE_FCOE
830 /* if ddp, not passing to ULD unless for FCP_RSP or error */
831 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
832 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
833 if (!ddp_bytes)
834 goto next_desc;
835 }
836 #endif /* IXGBE_FCOE */
837 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
838
839 next_desc:
840 rx_desc->wb.upper.status_error = 0;
841
842 /* return some buffers to hardware, one at a time is too slow */
843 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
844 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
845 cleaned_count = 0;
846 }
847
848 /* use prefetched values */
849 rx_desc = next_rxd;
850 rx_buffer_info = &rx_ring->rx_buffer_info[i];
851
852 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
853 }
854
855 rx_ring->next_to_clean = i;
856 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
857
858 if (cleaned_count)
859 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
860
861 #ifdef IXGBE_FCOE
862 /* include DDPed FCoE data */
863 if (ddp_bytes > 0) {
864 unsigned int mss;
865
866 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
867 sizeof(struct fc_frame_header) -
868 sizeof(struct fcoe_crc_eof);
869 if (mss > 512)
870 mss &= ~511;
871 total_rx_bytes += ddp_bytes;
872 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
873 }
874 #endif /* IXGBE_FCOE */
875
876 rx_ring->total_packets += total_rx_packets;
877 rx_ring->total_bytes += total_rx_bytes;
878 netdev->stats.rx_bytes += total_rx_bytes;
879 netdev->stats.rx_packets += total_rx_packets;
880
881 return cleaned;
882 }
883
884 static int ixgbe_clean_rxonly(struct napi_struct *, int);
885 /**
886 * ixgbe_configure_msix - Configure MSI-X hardware
887 * @adapter: board private structure
888 *
889 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
890 * interrupts.
891 **/
892 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
893 {
894 struct ixgbe_q_vector *q_vector;
895 int i, j, q_vectors, v_idx, r_idx;
896 u32 mask;
897
898 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
899
900 /*
901 * Populate the IVAR table and set the ITR values to the
902 * corresponding register.
903 */
904 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
905 q_vector = adapter->q_vector[v_idx];
906 /* XXX for_each_bit(...) */
907 r_idx = find_first_bit(q_vector->rxr_idx,
908 adapter->num_rx_queues);
909
910 for (i = 0; i < q_vector->rxr_count; i++) {
911 j = adapter->rx_ring[r_idx].reg_idx;
912 ixgbe_set_ivar(adapter, 0, j, v_idx);
913 r_idx = find_next_bit(q_vector->rxr_idx,
914 adapter->num_rx_queues,
915 r_idx + 1);
916 }
917 r_idx = find_first_bit(q_vector->txr_idx,
918 adapter->num_tx_queues);
919
920 for (i = 0; i < q_vector->txr_count; i++) {
921 j = adapter->tx_ring[r_idx].reg_idx;
922 ixgbe_set_ivar(adapter, 1, j, v_idx);
923 r_idx = find_next_bit(q_vector->txr_idx,
924 adapter->num_tx_queues,
925 r_idx + 1);
926 }
927
928 if (q_vector->txr_count && !q_vector->rxr_count)
929 /* tx only */
930 q_vector->eitr = adapter->tx_eitr_param;
931 else if (q_vector->rxr_count)
932 /* rx or mixed */
933 q_vector->eitr = adapter->rx_eitr_param;
934
935 ixgbe_write_eitr(q_vector);
936 }
937
938 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
939 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
940 v_idx);
941 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
942 ixgbe_set_ivar(adapter, -1, 1, v_idx);
943 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
944
945 /* set up to autoclear timer, and the vectors */
946 mask = IXGBE_EIMS_ENABLE_MASK;
947 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
948 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
949 }
950
951 enum latency_range {
952 lowest_latency = 0,
953 low_latency = 1,
954 bulk_latency = 2,
955 latency_invalid = 255
956 };
957
958 /**
959 * ixgbe_update_itr - update the dynamic ITR value based on statistics
960 * @adapter: pointer to adapter
961 * @eitr: eitr setting (ints per sec) to give last timeslice
962 * @itr_setting: current throttle rate in ints/second
963 * @packets: the number of packets during this measurement interval
964 * @bytes: the number of bytes during this measurement interval
965 *
966 * Stores a new ITR value based on packets and byte
967 * counts during the last interrupt. The advantage of per interrupt
968 * computation is faster updates and more accurate ITR for the current
969 * traffic pattern. Constants in this function were computed
970 * based on theoretical maximum wire speed and thresholds were set based
971 * on testing data as well as attempting to minimize response time
972 * while increasing bulk throughput.
973 * this functionality is controlled by the InterruptThrottleRate module
974 * parameter (see ixgbe_param.c)
975 **/
976 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
977 u32 eitr, u8 itr_setting,
978 int packets, int bytes)
979 {
980 unsigned int retval = itr_setting;
981 u32 timepassed_us;
982 u64 bytes_perint;
983
984 if (packets == 0)
985 goto update_itr_done;
986
987
988 /* simple throttlerate management
989 * 0-20MB/s lowest (100000 ints/s)
990 * 20-100MB/s low (20000 ints/s)
991 * 100-1249MB/s bulk (8000 ints/s)
992 */
993 /* what was last interrupt timeslice? */
994 timepassed_us = 1000000/eitr;
995 bytes_perint = bytes / timepassed_us; /* bytes/usec */
996
997 switch (itr_setting) {
998 case lowest_latency:
999 if (bytes_perint > adapter->eitr_low)
1000 retval = low_latency;
1001 break;
1002 case low_latency:
1003 if (bytes_perint > adapter->eitr_high)
1004 retval = bulk_latency;
1005 else if (bytes_perint <= adapter->eitr_low)
1006 retval = lowest_latency;
1007 break;
1008 case bulk_latency:
1009 if (bytes_perint <= adapter->eitr_high)
1010 retval = low_latency;
1011 break;
1012 }
1013
1014 update_itr_done:
1015 return retval;
1016 }
1017
1018 /**
1019 * ixgbe_write_eitr - write EITR register in hardware specific way
1020 * @q_vector: structure containing interrupt and ring information
1021 *
1022 * This function is made to be called by ethtool and by the driver
1023 * when it needs to update EITR registers at runtime. Hardware
1024 * specific quirks/differences are taken care of here.
1025 */
1026 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1027 {
1028 struct ixgbe_adapter *adapter = q_vector->adapter;
1029 struct ixgbe_hw *hw = &adapter->hw;
1030 int v_idx = q_vector->v_idx;
1031 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1032
1033 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1034 /* must write high and low 16 bits to reset counter */
1035 itr_reg |= (itr_reg << 16);
1036 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1037 /*
1038 * set the WDIS bit to not clear the timer bits and cause an
1039 * immediate assertion of the interrupt
1040 */
1041 itr_reg |= IXGBE_EITR_CNT_WDIS;
1042 }
1043 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1044 }
1045
1046 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1047 {
1048 struct ixgbe_adapter *adapter = q_vector->adapter;
1049 u32 new_itr;
1050 u8 current_itr, ret_itr;
1051 int i, r_idx;
1052 struct ixgbe_ring *rx_ring, *tx_ring;
1053
1054 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1055 for (i = 0; i < q_vector->txr_count; i++) {
1056 tx_ring = &(adapter->tx_ring[r_idx]);
1057 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1058 q_vector->tx_itr,
1059 tx_ring->total_packets,
1060 tx_ring->total_bytes);
1061 /* if the result for this queue would decrease interrupt
1062 * rate for this vector then use that result */
1063 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1064 q_vector->tx_itr - 1 : ret_itr);
1065 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1066 r_idx + 1);
1067 }
1068
1069 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1070 for (i = 0; i < q_vector->rxr_count; i++) {
1071 rx_ring = &(adapter->rx_ring[r_idx]);
1072 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1073 q_vector->rx_itr,
1074 rx_ring->total_packets,
1075 rx_ring->total_bytes);
1076 /* if the result for this queue would decrease interrupt
1077 * rate for this vector then use that result */
1078 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1079 q_vector->rx_itr - 1 : ret_itr);
1080 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1081 r_idx + 1);
1082 }
1083
1084 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1085
1086 switch (current_itr) {
1087 /* counts and packets in update_itr are dependent on these numbers */
1088 case lowest_latency:
1089 new_itr = 100000;
1090 break;
1091 case low_latency:
1092 new_itr = 20000; /* aka hwitr = ~200 */
1093 break;
1094 case bulk_latency:
1095 default:
1096 new_itr = 8000;
1097 break;
1098 }
1099
1100 if (new_itr != q_vector->eitr) {
1101 /* do an exponential smoothing */
1102 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1103
1104 /* save the algorithm value here, not the smoothed one */
1105 q_vector->eitr = new_itr;
1106
1107 ixgbe_write_eitr(q_vector);
1108 }
1109
1110 return;
1111 }
1112
1113 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1114 {
1115 struct ixgbe_hw *hw = &adapter->hw;
1116
1117 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1118 (eicr & IXGBE_EICR_GPI_SDP1)) {
1119 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
1120 /* write to clear the interrupt */
1121 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1122 }
1123 }
1124
1125 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1126 {
1127 struct ixgbe_hw *hw = &adapter->hw;
1128
1129 if (eicr & IXGBE_EICR_GPI_SDP1) {
1130 /* Clear the interrupt */
1131 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1132 schedule_work(&adapter->multispeed_fiber_task);
1133 } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1134 /* Clear the interrupt */
1135 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1136 schedule_work(&adapter->sfp_config_module_task);
1137 } else {
1138 /* Interrupt isn't for us... */
1139 return;
1140 }
1141 }
1142
1143 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1144 {
1145 struct ixgbe_hw *hw = &adapter->hw;
1146
1147 adapter->lsc_int++;
1148 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1149 adapter->link_check_timeout = jiffies;
1150 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1151 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1152 schedule_work(&adapter->watchdog_task);
1153 }
1154 }
1155
1156 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1157 {
1158 struct net_device *netdev = data;
1159 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1160 struct ixgbe_hw *hw = &adapter->hw;
1161 u32 eicr;
1162
1163 /*
1164 * Workaround for Silicon errata. Use clear-by-write instead
1165 * of clear-by-read. Reading with EICS will return the
1166 * interrupt causes without clearing, which later be done
1167 * with the write to EICR.
1168 */
1169 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1170 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1171
1172 if (eicr & IXGBE_EICR_LSC)
1173 ixgbe_check_lsc(adapter);
1174
1175 if (hw->mac.type == ixgbe_mac_82598EB)
1176 ixgbe_check_fan_failure(adapter, eicr);
1177
1178 if (hw->mac.type == ixgbe_mac_82599EB) {
1179 ixgbe_check_sfp_event(adapter, eicr);
1180
1181 /* Handle Flow Director Full threshold interrupt */
1182 if (eicr & IXGBE_EICR_FLOW_DIR) {
1183 int i;
1184 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1185 /* Disable transmits before FDIR Re-initialization */
1186 netif_tx_stop_all_queues(netdev);
1187 for (i = 0; i < adapter->num_tx_queues; i++) {
1188 struct ixgbe_ring *tx_ring =
1189 &adapter->tx_ring[i];
1190 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1191 &tx_ring->reinit_state))
1192 schedule_work(&adapter->fdir_reinit_task);
1193 }
1194 }
1195 }
1196 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1197 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1198
1199 return IRQ_HANDLED;
1200 }
1201
1202 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1203 u64 qmask)
1204 {
1205 u32 mask;
1206
1207 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1208 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1209 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1210 } else {
1211 mask = (qmask & 0xFFFFFFFF);
1212 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1213 mask = (qmask >> 32);
1214 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1215 }
1216 /* skip the flush */
1217 }
1218
1219 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1220 u64 qmask)
1221 {
1222 u32 mask;
1223
1224 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1225 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1226 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1227 } else {
1228 mask = (qmask & 0xFFFFFFFF);
1229 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1230 mask = (qmask >> 32);
1231 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1232 }
1233 /* skip the flush */
1234 }
1235
1236 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1237 {
1238 struct ixgbe_q_vector *q_vector = data;
1239 struct ixgbe_adapter *adapter = q_vector->adapter;
1240 struct ixgbe_ring *tx_ring;
1241 int i, r_idx;
1242
1243 if (!q_vector->txr_count)
1244 return IRQ_HANDLED;
1245
1246 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1247 for (i = 0; i < q_vector->txr_count; i++) {
1248 tx_ring = &(adapter->tx_ring[r_idx]);
1249 tx_ring->total_bytes = 0;
1250 tx_ring->total_packets = 0;
1251 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1252 r_idx + 1);
1253 }
1254
1255 /* disable interrupts on this vector only */
1256 ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1257 napi_schedule(&q_vector->napi);
1258
1259 return IRQ_HANDLED;
1260 }
1261
1262 /**
1263 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1264 * @irq: unused
1265 * @data: pointer to our q_vector struct for this interrupt vector
1266 **/
1267 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1268 {
1269 struct ixgbe_q_vector *q_vector = data;
1270 struct ixgbe_adapter *adapter = q_vector->adapter;
1271 struct ixgbe_ring *rx_ring;
1272 int r_idx;
1273 int i;
1274
1275 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1276 for (i = 0; i < q_vector->rxr_count; i++) {
1277 rx_ring = &(adapter->rx_ring[r_idx]);
1278 rx_ring->total_bytes = 0;
1279 rx_ring->total_packets = 0;
1280 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1281 r_idx + 1);
1282 }
1283
1284 if (!q_vector->rxr_count)
1285 return IRQ_HANDLED;
1286
1287 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1288 rx_ring = &(adapter->rx_ring[r_idx]);
1289 /* disable interrupts on this vector only */
1290 ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1291 napi_schedule(&q_vector->napi);
1292
1293 return IRQ_HANDLED;
1294 }
1295
1296 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1297 {
1298 struct ixgbe_q_vector *q_vector = data;
1299 struct ixgbe_adapter *adapter = q_vector->adapter;
1300 struct ixgbe_ring *ring;
1301 int r_idx;
1302 int i;
1303
1304 if (!q_vector->txr_count && !q_vector->rxr_count)
1305 return IRQ_HANDLED;
1306
1307 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1308 for (i = 0; i < q_vector->txr_count; i++) {
1309 ring = &(adapter->tx_ring[r_idx]);
1310 ring->total_bytes = 0;
1311 ring->total_packets = 0;
1312 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1313 r_idx + 1);
1314 }
1315
1316 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1317 for (i = 0; i < q_vector->rxr_count; i++) {
1318 ring = &(adapter->rx_ring[r_idx]);
1319 ring->total_bytes = 0;
1320 ring->total_packets = 0;
1321 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1322 r_idx + 1);
1323 }
1324
1325 /* disable interrupts on this vector only */
1326 ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1327 napi_schedule(&q_vector->napi);
1328
1329 return IRQ_HANDLED;
1330 }
1331
1332 /**
1333 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1334 * @napi: napi struct with our devices info in it
1335 * @budget: amount of work driver is allowed to do this pass, in packets
1336 *
1337 * This function is optimized for cleaning one queue only on a single
1338 * q_vector!!!
1339 **/
1340 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1341 {
1342 struct ixgbe_q_vector *q_vector =
1343 container_of(napi, struct ixgbe_q_vector, napi);
1344 struct ixgbe_adapter *adapter = q_vector->adapter;
1345 struct ixgbe_ring *rx_ring = NULL;
1346 int work_done = 0;
1347 long r_idx;
1348
1349 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1350 rx_ring = &(adapter->rx_ring[r_idx]);
1351 #ifdef CONFIG_IXGBE_DCA
1352 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1353 ixgbe_update_rx_dca(adapter, rx_ring);
1354 #endif
1355
1356 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1357
1358 /* If all Rx work done, exit the polling mode */
1359 if (work_done < budget) {
1360 napi_complete(napi);
1361 if (adapter->rx_itr_setting & 1)
1362 ixgbe_set_itr_msix(q_vector);
1363 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1364 ixgbe_irq_enable_queues(adapter,
1365 ((u64)1 << q_vector->v_idx));
1366 }
1367
1368 return work_done;
1369 }
1370
1371 /**
1372 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1373 * @napi: napi struct with our devices info in it
1374 * @budget: amount of work driver is allowed to do this pass, in packets
1375 *
1376 * This function will clean more than one rx queue associated with a
1377 * q_vector.
1378 **/
1379 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1380 {
1381 struct ixgbe_q_vector *q_vector =
1382 container_of(napi, struct ixgbe_q_vector, napi);
1383 struct ixgbe_adapter *adapter = q_vector->adapter;
1384 struct ixgbe_ring *ring = NULL;
1385 int work_done = 0, i;
1386 long r_idx;
1387 bool tx_clean_complete = true;
1388
1389 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1390 for (i = 0; i < q_vector->txr_count; i++) {
1391 ring = &(adapter->tx_ring[r_idx]);
1392 #ifdef CONFIG_IXGBE_DCA
1393 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1394 ixgbe_update_tx_dca(adapter, ring);
1395 #endif
1396 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1397 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1398 r_idx + 1);
1399 }
1400
1401 /* attempt to distribute budget to each queue fairly, but don't allow
1402 * the budget to go below 1 because we'll exit polling */
1403 budget /= (q_vector->rxr_count ?: 1);
1404 budget = max(budget, 1);
1405 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1406 for (i = 0; i < q_vector->rxr_count; i++) {
1407 ring = &(adapter->rx_ring[r_idx]);
1408 #ifdef CONFIG_IXGBE_DCA
1409 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1410 ixgbe_update_rx_dca(adapter, ring);
1411 #endif
1412 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1413 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1414 r_idx + 1);
1415 }
1416
1417 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1418 ring = &(adapter->rx_ring[r_idx]);
1419 /* If all Rx work done, exit the polling mode */
1420 if (work_done < budget) {
1421 napi_complete(napi);
1422 if (adapter->rx_itr_setting & 1)
1423 ixgbe_set_itr_msix(q_vector);
1424 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1425 ixgbe_irq_enable_queues(adapter,
1426 ((u64)1 << q_vector->v_idx));
1427 return 0;
1428 }
1429
1430 return work_done;
1431 }
1432
1433 /**
1434 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1435 * @napi: napi struct with our devices info in it
1436 * @budget: amount of work driver is allowed to do this pass, in packets
1437 *
1438 * This function is optimized for cleaning one queue only on a single
1439 * q_vector!!!
1440 **/
1441 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1442 {
1443 struct ixgbe_q_vector *q_vector =
1444 container_of(napi, struct ixgbe_q_vector, napi);
1445 struct ixgbe_adapter *adapter = q_vector->adapter;
1446 struct ixgbe_ring *tx_ring = NULL;
1447 int work_done = 0;
1448 long r_idx;
1449
1450 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1451 tx_ring = &(adapter->tx_ring[r_idx]);
1452 #ifdef CONFIG_IXGBE_DCA
1453 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1454 ixgbe_update_tx_dca(adapter, tx_ring);
1455 #endif
1456
1457 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
1458 work_done = budget;
1459
1460 /* If all Tx work done, exit the polling mode */
1461 if (work_done < budget) {
1462 napi_complete(napi);
1463 if (adapter->tx_itr_setting & 1)
1464 ixgbe_set_itr_msix(q_vector);
1465 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1466 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
1467 }
1468
1469 return work_done;
1470 }
1471
1472 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1473 int r_idx)
1474 {
1475 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1476
1477 set_bit(r_idx, q_vector->rxr_idx);
1478 q_vector->rxr_count++;
1479 }
1480
1481 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1482 int t_idx)
1483 {
1484 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1485
1486 set_bit(t_idx, q_vector->txr_idx);
1487 q_vector->txr_count++;
1488 }
1489
1490 /**
1491 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1492 * @adapter: board private structure to initialize
1493 * @vectors: allotted vector count for descriptor rings
1494 *
1495 * This function maps descriptor rings to the queue-specific vectors
1496 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1497 * one vector per ring/queue, but on a constrained vector budget, we
1498 * group the rings as "efficiently" as possible. You would add new
1499 * mapping configurations in here.
1500 **/
1501 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1502 int vectors)
1503 {
1504 int v_start = 0;
1505 int rxr_idx = 0, txr_idx = 0;
1506 int rxr_remaining = adapter->num_rx_queues;
1507 int txr_remaining = adapter->num_tx_queues;
1508 int i, j;
1509 int rqpv, tqpv;
1510 int err = 0;
1511
1512 /* No mapping required if MSI-X is disabled. */
1513 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1514 goto out;
1515
1516 /*
1517 * The ideal configuration...
1518 * We have enough vectors to map one per queue.
1519 */
1520 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1521 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1522 map_vector_to_rxq(adapter, v_start, rxr_idx);
1523
1524 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1525 map_vector_to_txq(adapter, v_start, txr_idx);
1526
1527 goto out;
1528 }
1529
1530 /*
1531 * If we don't have enough vectors for a 1-to-1
1532 * mapping, we'll have to group them so there are
1533 * multiple queues per vector.
1534 */
1535 /* Re-adjusting *qpv takes care of the remainder. */
1536 for (i = v_start; i < vectors; i++) {
1537 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1538 for (j = 0; j < rqpv; j++) {
1539 map_vector_to_rxq(adapter, i, rxr_idx);
1540 rxr_idx++;
1541 rxr_remaining--;
1542 }
1543 }
1544 for (i = v_start; i < vectors; i++) {
1545 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1546 for (j = 0; j < tqpv; j++) {
1547 map_vector_to_txq(adapter, i, txr_idx);
1548 txr_idx++;
1549 txr_remaining--;
1550 }
1551 }
1552
1553 out:
1554 return err;
1555 }
1556
1557 /**
1558 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1559 * @adapter: board private structure
1560 *
1561 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1562 * interrupts from the kernel.
1563 **/
1564 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1565 {
1566 struct net_device *netdev = adapter->netdev;
1567 irqreturn_t (*handler)(int, void *);
1568 int i, vector, q_vectors, err;
1569 int ri=0, ti=0;
1570
1571 /* Decrement for Other and TCP Timer vectors */
1572 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1573
1574 /* Map the Tx/Rx rings to the vectors we were allotted. */
1575 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1576 if (err)
1577 goto out;
1578
1579 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1580 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1581 &ixgbe_msix_clean_many)
1582 for (vector = 0; vector < q_vectors; vector++) {
1583 handler = SET_HANDLER(adapter->q_vector[vector]);
1584
1585 if(handler == &ixgbe_msix_clean_rx) {
1586 sprintf(adapter->name[vector], "%s-%s-%d",
1587 netdev->name, "rx", ri++);
1588 }
1589 else if(handler == &ixgbe_msix_clean_tx) {
1590 sprintf(adapter->name[vector], "%s-%s-%d",
1591 netdev->name, "tx", ti++);
1592 }
1593 else
1594 sprintf(adapter->name[vector], "%s-%s-%d",
1595 netdev->name, "TxRx", vector);
1596
1597 err = request_irq(adapter->msix_entries[vector].vector,
1598 handler, 0, adapter->name[vector],
1599 adapter->q_vector[vector]);
1600 if (err) {
1601 DPRINTK(PROBE, ERR,
1602 "request_irq failed for MSIX interrupt "
1603 "Error: %d\n", err);
1604 goto free_queue_irqs;
1605 }
1606 }
1607
1608 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1609 err = request_irq(adapter->msix_entries[vector].vector,
1610 &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1611 if (err) {
1612 DPRINTK(PROBE, ERR,
1613 "request_irq for msix_lsc failed: %d\n", err);
1614 goto free_queue_irqs;
1615 }
1616
1617 return 0;
1618
1619 free_queue_irqs:
1620 for (i = vector - 1; i >= 0; i--)
1621 free_irq(adapter->msix_entries[--vector].vector,
1622 adapter->q_vector[i]);
1623 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1624 pci_disable_msix(adapter->pdev);
1625 kfree(adapter->msix_entries);
1626 adapter->msix_entries = NULL;
1627 out:
1628 return err;
1629 }
1630
1631 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1632 {
1633 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1634 u8 current_itr;
1635 u32 new_itr = q_vector->eitr;
1636 struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1637 struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1638
1639 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1640 q_vector->tx_itr,
1641 tx_ring->total_packets,
1642 tx_ring->total_bytes);
1643 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1644 q_vector->rx_itr,
1645 rx_ring->total_packets,
1646 rx_ring->total_bytes);
1647
1648 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1649
1650 switch (current_itr) {
1651 /* counts and packets in update_itr are dependent on these numbers */
1652 case lowest_latency:
1653 new_itr = 100000;
1654 break;
1655 case low_latency:
1656 new_itr = 20000; /* aka hwitr = ~200 */
1657 break;
1658 case bulk_latency:
1659 new_itr = 8000;
1660 break;
1661 default:
1662 break;
1663 }
1664
1665 if (new_itr != q_vector->eitr) {
1666 /* do an exponential smoothing */
1667 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1668
1669 /* save the algorithm value here, not the smoothed one */
1670 q_vector->eitr = new_itr;
1671
1672 ixgbe_write_eitr(q_vector);
1673 }
1674
1675 return;
1676 }
1677
1678 /**
1679 * ixgbe_irq_enable - Enable default interrupt generation settings
1680 * @adapter: board private structure
1681 **/
1682 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1683 {
1684 u32 mask;
1685
1686 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1687 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1688 mask |= IXGBE_EIMS_GPI_SDP1;
1689 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1690 mask |= IXGBE_EIMS_ECC;
1691 mask |= IXGBE_EIMS_GPI_SDP1;
1692 mask |= IXGBE_EIMS_GPI_SDP2;
1693 }
1694 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
1695 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
1696 mask |= IXGBE_EIMS_FLOW_DIR;
1697
1698 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1699 ixgbe_irq_enable_queues(adapter, ~0);
1700 IXGBE_WRITE_FLUSH(&adapter->hw);
1701 }
1702
1703 /**
1704 * ixgbe_intr - legacy mode Interrupt Handler
1705 * @irq: interrupt number
1706 * @data: pointer to a network interface device structure
1707 **/
1708 static irqreturn_t ixgbe_intr(int irq, void *data)
1709 {
1710 struct net_device *netdev = data;
1711 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1712 struct ixgbe_hw *hw = &adapter->hw;
1713 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1714 u32 eicr;
1715
1716 /*
1717 * Workaround for silicon errata. Mask the interrupts
1718 * before the read of EICR.
1719 */
1720 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1721
1722 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1723 * therefore no explict interrupt disable is necessary */
1724 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1725 if (!eicr) {
1726 /* shared interrupt alert!
1727 * make sure interrupts are enabled because the read will
1728 * have disabled interrupts due to EIAM */
1729 ixgbe_irq_enable(adapter);
1730 return IRQ_NONE; /* Not our interrupt */
1731 }
1732
1733 if (eicr & IXGBE_EICR_LSC)
1734 ixgbe_check_lsc(adapter);
1735
1736 if (hw->mac.type == ixgbe_mac_82599EB)
1737 ixgbe_check_sfp_event(adapter, eicr);
1738
1739 ixgbe_check_fan_failure(adapter, eicr);
1740
1741 if (napi_schedule_prep(&(q_vector->napi))) {
1742 adapter->tx_ring[0].total_packets = 0;
1743 adapter->tx_ring[0].total_bytes = 0;
1744 adapter->rx_ring[0].total_packets = 0;
1745 adapter->rx_ring[0].total_bytes = 0;
1746 /* would disable interrupts here but EIAM disabled it */
1747 __napi_schedule(&(q_vector->napi));
1748 }
1749
1750 return IRQ_HANDLED;
1751 }
1752
1753 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1754 {
1755 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1756
1757 for (i = 0; i < q_vectors; i++) {
1758 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
1759 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1760 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1761 q_vector->rxr_count = 0;
1762 q_vector->txr_count = 0;
1763 }
1764 }
1765
1766 /**
1767 * ixgbe_request_irq - initialize interrupts
1768 * @adapter: board private structure
1769 *
1770 * Attempts to configure interrupts using the best available
1771 * capabilities of the hardware and kernel.
1772 **/
1773 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1774 {
1775 struct net_device *netdev = adapter->netdev;
1776 int err;
1777
1778 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1779 err = ixgbe_request_msix_irqs(adapter);
1780 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1781 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1782 netdev->name, netdev);
1783 } else {
1784 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1785 netdev->name, netdev);
1786 }
1787
1788 if (err)
1789 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1790
1791 return err;
1792 }
1793
1794 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1795 {
1796 struct net_device *netdev = adapter->netdev;
1797
1798 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1799 int i, q_vectors;
1800
1801 q_vectors = adapter->num_msix_vectors;
1802
1803 i = q_vectors - 1;
1804 free_irq(adapter->msix_entries[i].vector, netdev);
1805
1806 i--;
1807 for (; i >= 0; i--) {
1808 free_irq(adapter->msix_entries[i].vector,
1809 adapter->q_vector[i]);
1810 }
1811
1812 ixgbe_reset_q_vectors(adapter);
1813 } else {
1814 free_irq(adapter->pdev->irq, netdev);
1815 }
1816 }
1817
1818 /**
1819 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1820 * @adapter: board private structure
1821 **/
1822 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1823 {
1824 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1825 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1826 } else {
1827 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
1828 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
1829 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
1830 }
1831 IXGBE_WRITE_FLUSH(&adapter->hw);
1832 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1833 int i;
1834 for (i = 0; i < adapter->num_msix_vectors; i++)
1835 synchronize_irq(adapter->msix_entries[i].vector);
1836 } else {
1837 synchronize_irq(adapter->pdev->irq);
1838 }
1839 }
1840
1841 /**
1842 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1843 *
1844 **/
1845 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1846 {
1847 struct ixgbe_hw *hw = &adapter->hw;
1848
1849 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1850 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
1851
1852 ixgbe_set_ivar(adapter, 0, 0, 0);
1853 ixgbe_set_ivar(adapter, 1, 0, 0);
1854
1855 map_vector_to_rxq(adapter, 0, 0);
1856 map_vector_to_txq(adapter, 0, 0);
1857
1858 DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1859 }
1860
1861 /**
1862 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1863 * @adapter: board private structure
1864 *
1865 * Configure the Tx unit of the MAC after a reset.
1866 **/
1867 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1868 {
1869 u64 tdba;
1870 struct ixgbe_hw *hw = &adapter->hw;
1871 u32 i, j, tdlen, txctrl;
1872
1873 /* Setup the HW Tx Head and Tail descriptor pointers */
1874 for (i = 0; i < adapter->num_tx_queues; i++) {
1875 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1876 j = ring->reg_idx;
1877 tdba = ring->dma;
1878 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1879 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1880 (tdba & DMA_BIT_MASK(32)));
1881 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1882 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1883 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1884 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1885 adapter->tx_ring[i].head = IXGBE_TDH(j);
1886 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1887 /*
1888 * Disable Tx Head Writeback RO bit, since this hoses
1889 * bookkeeping if things aren't delivered in order.
1890 */
1891 switch (hw->mac.type) {
1892 case ixgbe_mac_82598EB:
1893 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1894 break;
1895 case ixgbe_mac_82599EB:
1896 default:
1897 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(j));
1898 break;
1899 }
1900 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1901 switch (hw->mac.type) {
1902 case ixgbe_mac_82598EB:
1903 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1904 break;
1905 case ixgbe_mac_82599EB:
1906 default:
1907 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(j), txctrl);
1908 break;
1909 }
1910 }
1911 if (hw->mac.type == ixgbe_mac_82599EB) {
1912 /* We enable 8 traffic classes, DCB only */
1913 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
1914 IXGBE_WRITE_REG(hw, IXGBE_MTQC, (IXGBE_MTQC_RT_ENA |
1915 IXGBE_MTQC_8TC_8TQ));
1916 }
1917 }
1918
1919 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1920
1921 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
1922 struct ixgbe_ring *rx_ring)
1923 {
1924 u32 srrctl;
1925 int index;
1926 struct ixgbe_ring_feature *feature = adapter->ring_feature;
1927
1928 index = rx_ring->reg_idx;
1929 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1930 unsigned long mask;
1931 mask = (unsigned long) feature[RING_F_RSS].mask;
1932 index = index & mask;
1933 }
1934 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
1935
1936 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1937 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1938
1939 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1940 IXGBE_SRRCTL_BSIZEHDR_MASK;
1941
1942 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1943 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
1944 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1945 #else
1946 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1947 #endif
1948 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1949 } else {
1950 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
1951 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1952 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1953 }
1954
1955 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
1956 }
1957
1958 static u32 ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
1959 {
1960 u32 mrqc = 0;
1961 int mask;
1962
1963 if (!(adapter->hw.mac.type == ixgbe_mac_82599EB))
1964 return mrqc;
1965
1966 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
1967 #ifdef CONFIG_IXGBE_DCB
1968 | IXGBE_FLAG_DCB_ENABLED
1969 #endif
1970 );
1971
1972 switch (mask) {
1973 case (IXGBE_FLAG_RSS_ENABLED):
1974 mrqc = IXGBE_MRQC_RSSEN;
1975 break;
1976 #ifdef CONFIG_IXGBE_DCB
1977 case (IXGBE_FLAG_DCB_ENABLED):
1978 mrqc = IXGBE_MRQC_RT8TCEN;
1979 break;
1980 #endif /* CONFIG_IXGBE_DCB */
1981 default:
1982 break;
1983 }
1984
1985 return mrqc;
1986 }
1987
1988 /**
1989 * ixgbe_configure_rscctl - enable RSC for the indicated ring
1990 * @adapter: address of board private structure
1991 * @index: index of ring to set
1992 * @rx_buf_len: rx buffer length
1993 **/
1994 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter, int index,
1995 int rx_buf_len)
1996 {
1997 struct ixgbe_ring *rx_ring;
1998 struct ixgbe_hw *hw = &adapter->hw;
1999 int j;
2000 u32 rscctrl;
2001
2002 rx_ring = &adapter->rx_ring[index];
2003 j = rx_ring->reg_idx;
2004 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2005 rscctrl |= IXGBE_RSCCTL_RSCEN;
2006 /*
2007 * we must limit the number of descriptors so that the
2008 * total size of max desc * buf_len is not greater
2009 * than 65535
2010 */
2011 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2012 #if (MAX_SKB_FRAGS > 16)
2013 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2014 #elif (MAX_SKB_FRAGS > 8)
2015 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2016 #elif (MAX_SKB_FRAGS > 4)
2017 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2018 #else
2019 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2020 #endif
2021 } else {
2022 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2023 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2024 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2025 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2026 else
2027 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2028 }
2029 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2030 }
2031
2032 /**
2033 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2034 * @adapter: board private structure
2035 *
2036 * Configure the Rx unit of the MAC after a reset.
2037 **/
2038 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2039 {
2040 u64 rdba;
2041 struct ixgbe_hw *hw = &adapter->hw;
2042 struct ixgbe_ring *rx_ring;
2043 struct net_device *netdev = adapter->netdev;
2044 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2045 int i, j;
2046 u32 rdlen, rxctrl, rxcsum;
2047 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2048 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2049 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2050 u32 fctrl, hlreg0;
2051 u32 reta = 0, mrqc = 0;
2052 u32 rdrxctl;
2053 int rx_buf_len;
2054
2055 /* Decide whether to use packet split mode or not */
2056 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2057
2058 /* Set the RX buffer length according to the mode */
2059 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2060 rx_buf_len = IXGBE_RX_HDR_SIZE;
2061 if (hw->mac.type == ixgbe_mac_82599EB) {
2062 /* PSRTYPE must be initialized in 82599 */
2063 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2064 IXGBE_PSRTYPE_UDPHDR |
2065 IXGBE_PSRTYPE_IPV4HDR |
2066 IXGBE_PSRTYPE_IPV6HDR |
2067 IXGBE_PSRTYPE_L2HDR;
2068 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), psrtype);
2069 }
2070 } else {
2071 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2072 (netdev->mtu <= ETH_DATA_LEN))
2073 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2074 else
2075 rx_buf_len = ALIGN(max_frame, 1024);
2076 }
2077
2078 fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
2079 fctrl |= IXGBE_FCTRL_BAM;
2080 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
2081 fctrl |= IXGBE_FCTRL_PMCF;
2082 IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
2083
2084 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2085 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2086 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2087 else
2088 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2089 #ifdef IXGBE_FCOE
2090 if (netdev->features & NETIF_F_FCOE_MTU)
2091 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2092 #endif
2093 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2094
2095 rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
2096 /* disable receives while setting up the descriptors */
2097 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2098 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2099
2100 /*
2101 * Setup the HW Rx Head and Tail Descriptor Pointers and
2102 * the Base and Length of the Rx Descriptor Ring
2103 */
2104 for (i = 0; i < adapter->num_rx_queues; i++) {
2105 rx_ring = &adapter->rx_ring[i];
2106 rdba = rx_ring->dma;
2107 j = rx_ring->reg_idx;
2108 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_BIT_MASK(32)));
2109 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
2110 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
2111 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
2112 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
2113 rx_ring->head = IXGBE_RDH(j);
2114 rx_ring->tail = IXGBE_RDT(j);
2115 rx_ring->rx_buf_len = rx_buf_len;
2116
2117 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2118 rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2119 else
2120 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2121
2122 #ifdef IXGBE_FCOE
2123 if (netdev->features & NETIF_F_FCOE_MTU) {
2124 struct ixgbe_ring_feature *f;
2125 f = &adapter->ring_feature[RING_F_FCOE];
2126 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2127 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2128 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2129 rx_ring->rx_buf_len =
2130 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2131 }
2132 }
2133
2134 #endif /* IXGBE_FCOE */
2135 ixgbe_configure_srrctl(adapter, rx_ring);
2136 }
2137
2138 if (hw->mac.type == ixgbe_mac_82598EB) {
2139 /*
2140 * For VMDq support of different descriptor types or
2141 * buffer sizes through the use of multiple SRRCTL
2142 * registers, RDRXCTL.MVMEN must be set to 1
2143 *
2144 * also, the manual doesn't mention it clearly but DCA hints
2145 * will only use queue 0's tags unless this bit is set. Side
2146 * effects of setting this bit are only that SRRCTL must be
2147 * fully programmed [0..15]
2148 */
2149 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2150 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2151 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2152 }
2153
2154 /* Program MRQC for the distribution of queues */
2155 mrqc = ixgbe_setup_mrqc(adapter);
2156
2157 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2158 /* Fill out redirection table */
2159 for (i = 0, j = 0; i < 128; i++, j++) {
2160 if (j == adapter->ring_feature[RING_F_RSS].indices)
2161 j = 0;
2162 /* reta = 4-byte sliding window of
2163 * 0x00..(indices-1)(indices-1)00..etc. */
2164 reta = (reta << 8) | (j * 0x11);
2165 if ((i & 3) == 3)
2166 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2167 }
2168
2169 /* Fill out hash function seeds */
2170 for (i = 0; i < 10; i++)
2171 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2172
2173 if (hw->mac.type == ixgbe_mac_82598EB)
2174 mrqc |= IXGBE_MRQC_RSSEN;
2175 /* Perform hash on these packet types */
2176 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2177 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2178 | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
2179 | IXGBE_MRQC_RSS_FIELD_IPV6
2180 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
2181 | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2182 }
2183 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2184
2185 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2186
2187 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
2188 adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
2189 /* Disable indicating checksum in descriptor, enables
2190 * RSS hash */
2191 rxcsum |= IXGBE_RXCSUM_PCSD;
2192 }
2193 if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
2194 /* Enable IPv4 payload checksum for UDP fragments
2195 * if PCSD is not set */
2196 rxcsum |= IXGBE_RXCSUM_IPPCSE;
2197 }
2198
2199 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2200
2201 if (hw->mac.type == ixgbe_mac_82599EB) {
2202 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2203 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2204 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2205 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2206 }
2207
2208 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
2209 /* Enable 82599 HW-RSC */
2210 for (i = 0; i < adapter->num_rx_queues; i++)
2211 ixgbe_configure_rscctl(adapter, i, rx_buf_len);
2212
2213 /* Disable RSC for ACK packets */
2214 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2215 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2216 }
2217 }
2218
2219 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2220 {
2221 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2222 struct ixgbe_hw *hw = &adapter->hw;
2223
2224 /* add VID to filter table */
2225 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
2226 }
2227
2228 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2229 {
2230 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2231 struct ixgbe_hw *hw = &adapter->hw;
2232
2233 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2234 ixgbe_irq_disable(adapter);
2235
2236 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2237
2238 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2239 ixgbe_irq_enable(adapter);
2240
2241 /* remove VID from filter table */
2242 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
2243 }
2244
2245 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2246 struct vlan_group *grp)
2247 {
2248 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2249 u32 ctrl;
2250 int i, j;
2251
2252 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2253 ixgbe_irq_disable(adapter);
2254 adapter->vlgrp = grp;
2255
2256 /*
2257 * For a DCB driver, always enable VLAN tag stripping so we can
2258 * still receive traffic from a DCB-enabled host even if we're
2259 * not in DCB mode.
2260 */
2261 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2262 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2263 ctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2264 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2265 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2266 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2267 ctrl |= IXGBE_VLNCTRL_VFE;
2268 /* enable VLAN tag insert/strip */
2269 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2270 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2271 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2272 for (i = 0; i < adapter->num_rx_queues; i++) {
2273 j = adapter->rx_ring[i].reg_idx;
2274 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_RXDCTL(j));
2275 ctrl |= IXGBE_RXDCTL_VME;
2276 IXGBE_WRITE_REG(&adapter->hw, IXGBE_RXDCTL(j), ctrl);
2277 }
2278 }
2279 ixgbe_vlan_rx_add_vid(netdev, 0);
2280
2281 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2282 ixgbe_irq_enable(adapter);
2283 }
2284
2285 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2286 {
2287 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2288
2289 if (adapter->vlgrp) {
2290 u16 vid;
2291 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2292 if (!vlan_group_get_device(adapter->vlgrp, vid))
2293 continue;
2294 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2295 }
2296 }
2297 }
2298
2299 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
2300 {
2301 struct dev_mc_list *mc_ptr;
2302 u8 *addr = *mc_addr_ptr;
2303 *vmdq = 0;
2304
2305 mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
2306 if (mc_ptr->next)
2307 *mc_addr_ptr = mc_ptr->next->dmi_addr;
2308 else
2309 *mc_addr_ptr = NULL;
2310
2311 return addr;
2312 }
2313
2314 /**
2315 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
2316 * @netdev: network interface device structure
2317 *
2318 * The set_rx_method entry point is called whenever the unicast/multicast
2319 * address list or the network interface flags are updated. This routine is
2320 * responsible for configuring the hardware for proper unicast, multicast and
2321 * promiscuous mode.
2322 **/
2323 static void ixgbe_set_rx_mode(struct net_device *netdev)
2324 {
2325 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2326 struct ixgbe_hw *hw = &adapter->hw;
2327 u32 fctrl, vlnctrl;
2328 u8 *addr_list = NULL;
2329 int addr_count = 0;
2330
2331 /* Check for Promiscuous and All Multicast modes */
2332
2333 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2334 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2335
2336 if (netdev->flags & IFF_PROMISC) {
2337 hw->addr_ctrl.user_set_promisc = 1;
2338 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2339 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2340 } else {
2341 if (netdev->flags & IFF_ALLMULTI) {
2342 fctrl |= IXGBE_FCTRL_MPE;
2343 fctrl &= ~IXGBE_FCTRL_UPE;
2344 } else {
2345 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2346 }
2347 vlnctrl |= IXGBE_VLNCTRL_VFE;
2348 hw->addr_ctrl.user_set_promisc = 0;
2349 }
2350
2351 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
2352 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2353
2354 /* reprogram secondary unicast list */
2355 hw->mac.ops.update_uc_addr_list(hw, &netdev->uc.list);
2356
2357 /* reprogram multicast list */
2358 addr_count = netdev->mc_count;
2359 if (addr_count)
2360 addr_list = netdev->mc_list->dmi_addr;
2361 hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
2362 ixgbe_addr_list_itr);
2363 }
2364
2365 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
2366 {
2367 int q_idx;
2368 struct ixgbe_q_vector *q_vector;
2369 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2370
2371 /* legacy and MSI only use one vector */
2372 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2373 q_vectors = 1;
2374
2375 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2376 struct napi_struct *napi;
2377 q_vector = adapter->q_vector[q_idx];
2378 napi = &q_vector->napi;
2379 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2380 if (!q_vector->rxr_count || !q_vector->txr_count) {
2381 if (q_vector->txr_count == 1)
2382 napi->poll = &ixgbe_clean_txonly;
2383 else if (q_vector->rxr_count == 1)
2384 napi->poll = &ixgbe_clean_rxonly;
2385 }
2386 }
2387
2388 napi_enable(napi);
2389 }
2390 }
2391
2392 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2393 {
2394 int q_idx;
2395 struct ixgbe_q_vector *q_vector;
2396 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2397
2398 /* legacy and MSI only use one vector */
2399 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2400 q_vectors = 1;
2401
2402 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2403 q_vector = adapter->q_vector[q_idx];
2404 napi_disable(&q_vector->napi);
2405 }
2406 }
2407
2408 #ifdef CONFIG_IXGBE_DCB
2409 /*
2410 * ixgbe_configure_dcb - Configure DCB hardware
2411 * @adapter: ixgbe adapter struct
2412 *
2413 * This is called by the driver on open to configure the DCB hardware.
2414 * This is also called by the gennetlink interface when reconfiguring
2415 * the DCB state.
2416 */
2417 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2418 {
2419 struct ixgbe_hw *hw = &adapter->hw;
2420 u32 txdctl, vlnctrl;
2421 int i, j;
2422
2423 ixgbe_dcb_check_config(&adapter->dcb_cfg);
2424 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2425 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2426
2427 /* reconfigure the hardware */
2428 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
2429
2430 for (i = 0; i < adapter->num_tx_queues; i++) {
2431 j = adapter->tx_ring[i].reg_idx;
2432 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2433 /* PThresh workaround for Tx hang with DFP enabled. */
2434 txdctl |= 32;
2435 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2436 }
2437 /* Enable VLAN tag insert/strip */
2438 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2439 if (hw->mac.type == ixgbe_mac_82598EB) {
2440 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2441 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2442 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2443 } else if (hw->mac.type == ixgbe_mac_82599EB) {
2444 vlnctrl |= IXGBE_VLNCTRL_VFE;
2445 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2446 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2447 for (i = 0; i < adapter->num_rx_queues; i++) {
2448 j = adapter->rx_ring[i].reg_idx;
2449 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2450 vlnctrl |= IXGBE_RXDCTL_VME;
2451 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2452 }
2453 }
2454 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2455 }
2456
2457 #endif
2458 static void ixgbe_configure(struct ixgbe_adapter *adapter)
2459 {
2460 struct net_device *netdev = adapter->netdev;
2461 struct ixgbe_hw *hw = &adapter->hw;
2462 int i;
2463
2464 ixgbe_set_rx_mode(netdev);
2465
2466 ixgbe_restore_vlan(adapter);
2467 #ifdef CONFIG_IXGBE_DCB
2468 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2469 netif_set_gso_max_size(netdev, 32768);
2470 ixgbe_configure_dcb(adapter);
2471 } else {
2472 netif_set_gso_max_size(netdev, 65536);
2473 }
2474 #else
2475 netif_set_gso_max_size(netdev, 65536);
2476 #endif
2477
2478 #ifdef IXGBE_FCOE
2479 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2480 ixgbe_configure_fcoe(adapter);
2481
2482 #endif /* IXGBE_FCOE */
2483 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2484 for (i = 0; i < adapter->num_tx_queues; i++)
2485 adapter->tx_ring[i].atr_sample_rate =
2486 adapter->atr_sample_rate;
2487 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
2488 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
2489 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
2490 }
2491
2492 ixgbe_configure_tx(adapter);
2493 ixgbe_configure_rx(adapter);
2494 for (i = 0; i < adapter->num_rx_queues; i++)
2495 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
2496 (adapter->rx_ring[i].count - 1));
2497 }
2498
2499 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2500 {
2501 switch (hw->phy.type) {
2502 case ixgbe_phy_sfp_avago:
2503 case ixgbe_phy_sfp_ftl:
2504 case ixgbe_phy_sfp_intel:
2505 case ixgbe_phy_sfp_unknown:
2506 case ixgbe_phy_tw_tyco:
2507 case ixgbe_phy_tw_unknown:
2508 return true;
2509 default:
2510 return false;
2511 }
2512 }
2513
2514 /**
2515 * ixgbe_sfp_link_config - set up SFP+ link
2516 * @adapter: pointer to private adapter struct
2517 **/
2518 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2519 {
2520 struct ixgbe_hw *hw = &adapter->hw;
2521
2522 if (hw->phy.multispeed_fiber) {
2523 /*
2524 * In multispeed fiber setups, the device may not have
2525 * had a physical connection when the driver loaded.
2526 * If that's the case, the initial link configuration
2527 * couldn't get the MAC into 10G or 1G mode, so we'll
2528 * never have a link status change interrupt fire.
2529 * We need to try and force an autonegotiation
2530 * session, then bring up link.
2531 */
2532 hw->mac.ops.setup_sfp(hw);
2533 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2534 schedule_work(&adapter->multispeed_fiber_task);
2535 } else {
2536 /*
2537 * Direct Attach Cu and non-multispeed fiber modules
2538 * still need to be configured properly prior to
2539 * attempting link.
2540 */
2541 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2542 schedule_work(&adapter->sfp_config_module_task);
2543 }
2544 }
2545
2546 /**
2547 * ixgbe_non_sfp_link_config - set up non-SFP+ link
2548 * @hw: pointer to private hardware struct
2549 *
2550 * Returns 0 on success, negative on failure
2551 **/
2552 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2553 {
2554 u32 autoneg;
2555 bool negotiation, link_up = false;
2556 u32 ret = IXGBE_ERR_LINK_SETUP;
2557
2558 if (hw->mac.ops.check_link)
2559 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2560
2561 if (ret)
2562 goto link_cfg_out;
2563
2564 if (hw->mac.ops.get_link_capabilities)
2565 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
2566 if (ret)
2567 goto link_cfg_out;
2568
2569 if (hw->mac.ops.setup_link)
2570 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
2571 link_cfg_out:
2572 return ret;
2573 }
2574
2575 #define IXGBE_MAX_RX_DESC_POLL 10
2576 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2577 int rxr)
2578 {
2579 int j = adapter->rx_ring[rxr].reg_idx;
2580 int k;
2581
2582 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2583 if (IXGBE_READ_REG(&adapter->hw,
2584 IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2585 break;
2586 else
2587 msleep(1);
2588 }
2589 if (k >= IXGBE_MAX_RX_DESC_POLL) {
2590 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2591 "not set within the polling period\n", rxr);
2592 }
2593 ixgbe_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
2594 (adapter->rx_ring[rxr].count - 1));
2595 }
2596
2597 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2598 {
2599 struct net_device *netdev = adapter->netdev;
2600 struct ixgbe_hw *hw = &adapter->hw;
2601 int i, j = 0;
2602 int num_rx_rings = adapter->num_rx_queues;
2603 int err;
2604 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2605 u32 txdctl, rxdctl, mhadd;
2606 u32 dmatxctl;
2607 u32 gpie;
2608
2609 ixgbe_get_hw_control(adapter);
2610
2611 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2612 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2613 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2614 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2615 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2616 } else {
2617 /* MSI only */
2618 gpie = 0;
2619 }
2620 /* XXX: to interrupt immediately for EICS writes, enable this */
2621 /* gpie |= IXGBE_GPIE_EIMEN; */
2622 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2623 }
2624
2625 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2626 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2627 * specifically only auto mask tx and rx interrupts */
2628 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2629 }
2630
2631 /* Enable fan failure interrupt if media type is copper */
2632 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2633 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2634 gpie |= IXGBE_SDP1_GPIEN;
2635 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2636 }
2637
2638 if (hw->mac.type == ixgbe_mac_82599EB) {
2639 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2640 gpie |= IXGBE_SDP1_GPIEN;
2641 gpie |= IXGBE_SDP2_GPIEN;
2642 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2643 }
2644
2645 #ifdef IXGBE_FCOE
2646 /* adjust max frame to be able to do baby jumbo for FCoE */
2647 if ((netdev->features & NETIF_F_FCOE_MTU) &&
2648 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2649 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2650
2651 #endif /* IXGBE_FCOE */
2652 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2653 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2654 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2655 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2656
2657 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2658 }
2659
2660 for (i = 0; i < adapter->num_tx_queues; i++) {
2661 j = adapter->tx_ring[i].reg_idx;
2662 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2663 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2664 txdctl |= (8 << 16);
2665 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2666 }
2667
2668 if (hw->mac.type == ixgbe_mac_82599EB) {
2669 /* DMATXCTL.EN must be set after all Tx queue config is done */
2670 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2671 dmatxctl |= IXGBE_DMATXCTL_TE;
2672 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2673 }
2674 for (i = 0; i < adapter->num_tx_queues; i++) {
2675 j = adapter->tx_ring[i].reg_idx;
2676 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2677 txdctl |= IXGBE_TXDCTL_ENABLE;
2678 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2679 }
2680
2681 for (i = 0; i < num_rx_rings; i++) {
2682 j = adapter->rx_ring[i].reg_idx;
2683 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2684 /* enable PTHRESH=32 descriptors (half the internal cache)
2685 * and HTHRESH=0 descriptors (to minimize latency on fetch),
2686 * this also removes a pesky rx_no_buffer_count increment */
2687 rxdctl |= 0x0020;
2688 rxdctl |= IXGBE_RXDCTL_ENABLE;
2689 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2690 if (hw->mac.type == ixgbe_mac_82599EB)
2691 ixgbe_rx_desc_queue_enable(adapter, i);
2692 }
2693 /* enable all receives */
2694 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2695 if (hw->mac.type == ixgbe_mac_82598EB)
2696 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2697 else
2698 rxdctl |= IXGBE_RXCTRL_RXEN;
2699 hw->mac.ops.enable_rx_dma(hw, rxdctl);
2700
2701 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2702 ixgbe_configure_msix(adapter);
2703 else
2704 ixgbe_configure_msi_and_legacy(adapter);
2705
2706 clear_bit(__IXGBE_DOWN, &adapter->state);
2707 ixgbe_napi_enable_all(adapter);
2708
2709 /* clear any pending interrupts, may auto mask */
2710 IXGBE_READ_REG(hw, IXGBE_EICR);
2711
2712 ixgbe_irq_enable(adapter);
2713
2714 /*
2715 * If this adapter has a fan, check to see if we had a failure
2716 * before we enabled the interrupt.
2717 */
2718 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2719 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
2720 if (esdp & IXGBE_ESDP_SDP1)
2721 DPRINTK(DRV, CRIT,
2722 "Fan has stopped, replace the adapter\n");
2723 }
2724
2725 /*
2726 * For hot-pluggable SFP+ devices, a new SFP+ module may have
2727 * arrived before interrupts were enabled but after probe. Such
2728 * devices wouldn't have their type identified yet. We need to
2729 * kick off the SFP+ module setup first, then try to bring up link.
2730 * If we're not hot-pluggable SFP+, we just need to configure link
2731 * and bring it up.
2732 */
2733 if (hw->phy.type == ixgbe_phy_unknown) {
2734 err = hw->phy.ops.identify(hw);
2735 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
2736 /*
2737 * Take the device down and schedule the sfp tasklet
2738 * which will unregister_netdev and log it.
2739 */
2740 ixgbe_down(adapter);
2741 schedule_work(&adapter->sfp_config_module_task);
2742 return err;
2743 }
2744 }
2745
2746 if (ixgbe_is_sfp(hw)) {
2747 ixgbe_sfp_link_config(adapter);
2748 } else {
2749 err = ixgbe_non_sfp_link_config(hw);
2750 if (err)
2751 DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
2752 }
2753
2754 for (i = 0; i < adapter->num_tx_queues; i++)
2755 set_bit(__IXGBE_FDIR_INIT_DONE,
2756 &(adapter->tx_ring[i].reinit_state));
2757
2758 /* enable transmits */
2759 netif_tx_start_all_queues(netdev);
2760
2761 /* bring the link up in the watchdog, this could race with our first
2762 * link up interrupt but shouldn't be a problem */
2763 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2764 adapter->link_check_timeout = jiffies;
2765 mod_timer(&adapter->watchdog_timer, jiffies);
2766 return 0;
2767 }
2768
2769 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
2770 {
2771 WARN_ON(in_interrupt());
2772 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
2773 msleep(1);
2774 ixgbe_down(adapter);
2775 ixgbe_up(adapter);
2776 clear_bit(__IXGBE_RESETTING, &adapter->state);
2777 }
2778
2779 int ixgbe_up(struct ixgbe_adapter *adapter)
2780 {
2781 /* hardware has been reset, we need to reload some things */
2782 ixgbe_configure(adapter);
2783
2784 return ixgbe_up_complete(adapter);
2785 }
2786
2787 void ixgbe_reset(struct ixgbe_adapter *adapter)
2788 {
2789 struct ixgbe_hw *hw = &adapter->hw;
2790 int err;
2791
2792 err = hw->mac.ops.init_hw(hw);
2793 switch (err) {
2794 case 0:
2795 case IXGBE_ERR_SFP_NOT_PRESENT:
2796 break;
2797 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
2798 dev_err(&adapter->pdev->dev, "master disable timed out\n");
2799 break;
2800 case IXGBE_ERR_EEPROM_VERSION:
2801 /* We are running on a pre-production device, log a warning */
2802 dev_warn(&adapter->pdev->dev, "This device is a pre-production "
2803 "adapter/LOM. Please be aware there may be issues "
2804 "associated with your hardware. If you are "
2805 "experiencing problems please contact your Intel or "
2806 "hardware representative who provided you with this "
2807 "hardware.\n");
2808 break;
2809 default:
2810 dev_err(&adapter->pdev->dev, "Hardware Error: %d\n", err);
2811 }
2812
2813 /* reprogram the RAR[0] in case user changed it. */
2814 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2815 }
2816
2817 /**
2818 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2819 * @adapter: board private structure
2820 * @rx_ring: ring to free buffers from
2821 **/
2822 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2823 struct ixgbe_ring *rx_ring)
2824 {
2825 struct pci_dev *pdev = adapter->pdev;
2826 unsigned long size;
2827 unsigned int i;
2828
2829 /* Free all the Rx ring sk_buffs */
2830
2831 for (i = 0; i < rx_ring->count; i++) {
2832 struct ixgbe_rx_buffer *rx_buffer_info;
2833
2834 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2835 if (rx_buffer_info->dma) {
2836 pci_unmap_single(pdev, rx_buffer_info->dma,
2837 rx_ring->rx_buf_len,
2838 PCI_DMA_FROMDEVICE);
2839 rx_buffer_info->dma = 0;
2840 }
2841 if (rx_buffer_info->skb) {
2842 struct sk_buff *skb = rx_buffer_info->skb;
2843 rx_buffer_info->skb = NULL;
2844 do {
2845 struct sk_buff *this = skb;
2846 skb = skb->prev;
2847 dev_kfree_skb(this);
2848 } while (skb);
2849 }
2850 if (!rx_buffer_info->page)
2851 continue;
2852 if (rx_buffer_info->page_dma) {
2853 pci_unmap_page(pdev, rx_buffer_info->page_dma,
2854 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
2855 rx_buffer_info->page_dma = 0;
2856 }
2857 put_page(rx_buffer_info->page);
2858 rx_buffer_info->page = NULL;
2859 rx_buffer_info->page_offset = 0;
2860 }
2861
2862 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2863 memset(rx_ring->rx_buffer_info, 0, size);
2864
2865 /* Zero out the descriptor ring */
2866 memset(rx_ring->desc, 0, rx_ring->size);
2867
2868 rx_ring->next_to_clean = 0;
2869 rx_ring->next_to_use = 0;
2870
2871 if (rx_ring->head)
2872 writel(0, adapter->hw.hw_addr + rx_ring->head);
2873 if (rx_ring->tail)
2874 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2875 }
2876
2877 /**
2878 * ixgbe_clean_tx_ring - Free Tx Buffers
2879 * @adapter: board private structure
2880 * @tx_ring: ring to be cleaned
2881 **/
2882 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2883 struct ixgbe_ring *tx_ring)
2884 {
2885 struct ixgbe_tx_buffer *tx_buffer_info;
2886 unsigned long size;
2887 unsigned int i;
2888
2889 /* Free all the Tx ring sk_buffs */
2890
2891 for (i = 0; i < tx_ring->count; i++) {
2892 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2893 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2894 }
2895
2896 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2897 memset(tx_ring->tx_buffer_info, 0, size);
2898
2899 /* Zero out the descriptor ring */
2900 memset(tx_ring->desc, 0, tx_ring->size);
2901
2902 tx_ring->next_to_use = 0;
2903 tx_ring->next_to_clean = 0;
2904
2905 if (tx_ring->head)
2906 writel(0, adapter->hw.hw_addr + tx_ring->head);
2907 if (tx_ring->tail)
2908 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2909 }
2910
2911 /**
2912 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2913 * @adapter: board private structure
2914 **/
2915 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2916 {
2917 int i;
2918
2919 for (i = 0; i < adapter->num_rx_queues; i++)
2920 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2921 }
2922
2923 /**
2924 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2925 * @adapter: board private structure
2926 **/
2927 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2928 {
2929 int i;
2930
2931 for (i = 0; i < adapter->num_tx_queues; i++)
2932 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2933 }
2934
2935 void ixgbe_down(struct ixgbe_adapter *adapter)
2936 {
2937 struct net_device *netdev = adapter->netdev;
2938 struct ixgbe_hw *hw = &adapter->hw;
2939 u32 rxctrl;
2940 u32 txdctl;
2941 int i, j;
2942
2943 /* signal that we are down to the interrupt handler */
2944 set_bit(__IXGBE_DOWN, &adapter->state);
2945
2946 /* disable receives */
2947 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2948 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2949
2950 netif_tx_disable(netdev);
2951
2952 IXGBE_WRITE_FLUSH(hw);
2953 msleep(10);
2954
2955 netif_tx_stop_all_queues(netdev);
2956
2957 ixgbe_irq_disable(adapter);
2958
2959 ixgbe_napi_disable_all(adapter);
2960
2961 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
2962 del_timer_sync(&adapter->sfp_timer);
2963 del_timer_sync(&adapter->watchdog_timer);
2964 cancel_work_sync(&adapter->watchdog_task);
2965
2966 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2967 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2968 cancel_work_sync(&adapter->fdir_reinit_task);
2969
2970 /* disable transmits in the hardware now that interrupts are off */
2971 for (i = 0; i < adapter->num_tx_queues; i++) {
2972 j = adapter->tx_ring[i].reg_idx;
2973 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2974 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
2975 (txdctl & ~IXGBE_TXDCTL_ENABLE));
2976 }
2977 /* Disable the Tx DMA engine on 82599 */
2978 if (hw->mac.type == ixgbe_mac_82599EB)
2979 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
2980 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
2981 ~IXGBE_DMATXCTL_TE));
2982
2983 netif_carrier_off(netdev);
2984
2985 if (!pci_channel_offline(adapter->pdev))
2986 ixgbe_reset(adapter);
2987 ixgbe_clean_all_tx_rings(adapter);
2988 ixgbe_clean_all_rx_rings(adapter);
2989
2990 #ifdef CONFIG_IXGBE_DCA
2991 /* since we reset the hardware DCA settings were cleared */
2992 ixgbe_setup_dca(adapter);
2993 #endif
2994 }
2995
2996 /**
2997 * ixgbe_poll - NAPI Rx polling callback
2998 * @napi: structure for representing this polling device
2999 * @budget: how many packets driver is allowed to clean
3000 *
3001 * This function is used for legacy and MSI, NAPI mode
3002 **/
3003 static int ixgbe_poll(struct napi_struct *napi, int budget)
3004 {
3005 struct ixgbe_q_vector *q_vector =
3006 container_of(napi, struct ixgbe_q_vector, napi);
3007 struct ixgbe_adapter *adapter = q_vector->adapter;
3008 int tx_clean_complete, work_done = 0;
3009
3010 #ifdef CONFIG_IXGBE_DCA
3011 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3012 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
3013 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
3014 }
3015 #endif
3016
3017 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring);
3018 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring, &work_done, budget);
3019
3020 if (!tx_clean_complete)
3021 work_done = budget;
3022
3023 /* If budget not fully consumed, exit the polling mode */
3024 if (work_done < budget) {
3025 napi_complete(napi);
3026 if (adapter->rx_itr_setting & 1)
3027 ixgbe_set_itr(adapter);
3028 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3029 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3030 }
3031 return work_done;
3032 }
3033
3034 /**
3035 * ixgbe_tx_timeout - Respond to a Tx Hang
3036 * @netdev: network interface device structure
3037 **/
3038 static void ixgbe_tx_timeout(struct net_device *netdev)
3039 {
3040 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3041
3042 /* Do the reset outside of interrupt context */
3043 schedule_work(&adapter->reset_task);
3044 }
3045
3046 static void ixgbe_reset_task(struct work_struct *work)
3047 {
3048 struct ixgbe_adapter *adapter;
3049 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3050
3051 /* If we're already down or resetting, just bail */
3052 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3053 test_bit(__IXGBE_RESETTING, &adapter->state))
3054 return;
3055
3056 adapter->tx_timeout_count++;
3057
3058 ixgbe_reinit_locked(adapter);
3059 }
3060
3061 #ifdef CONFIG_IXGBE_DCB
3062 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3063 {
3064 bool ret = false;
3065 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3066
3067 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3068 return ret;
3069
3070 f->mask = 0x7 << 3;
3071 adapter->num_rx_queues = f->indices;
3072 adapter->num_tx_queues = f->indices;
3073 ret = true;
3074
3075 return ret;
3076 }
3077 #endif
3078
3079 /**
3080 * ixgbe_set_rss_queues: Allocate queues for RSS
3081 * @adapter: board private structure to initialize
3082 *
3083 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3084 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3085 *
3086 **/
3087 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3088 {
3089 bool ret = false;
3090 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3091
3092 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3093 f->mask = 0xF;
3094 adapter->num_rx_queues = f->indices;
3095 adapter->num_tx_queues = f->indices;
3096 ret = true;
3097 } else {
3098 ret = false;
3099 }
3100
3101 return ret;
3102 }
3103
3104 /**
3105 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3106 * @adapter: board private structure to initialize
3107 *
3108 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3109 * to the original CPU that initiated the Tx session. This runs in addition
3110 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3111 * Rx load across CPUs using RSS.
3112 *
3113 **/
3114 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3115 {
3116 bool ret = false;
3117 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3118
3119 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3120 f_fdir->mask = 0;
3121
3122 /* Flow Director must have RSS enabled */
3123 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3124 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3125 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3126 adapter->num_tx_queues = f_fdir->indices;
3127 adapter->num_rx_queues = f_fdir->indices;
3128 ret = true;
3129 } else {
3130 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3131 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3132 }
3133 return ret;
3134 }
3135
3136 #ifdef IXGBE_FCOE
3137 /**
3138 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3139 * @adapter: board private structure to initialize
3140 *
3141 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3142 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3143 * rx queues out of the max number of rx queues, instead, it is used as the
3144 * index of the first rx queue used by FCoE.
3145 *
3146 **/
3147 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3148 {
3149 bool ret = false;
3150 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3151
3152 f->indices = min((int)num_online_cpus(), f->indices);
3153 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3154 adapter->num_rx_queues = 1;
3155 adapter->num_tx_queues = 1;
3156 #ifdef CONFIG_IXGBE_DCB
3157 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3158 DPRINTK(PROBE, INFO, "FCoE enabled with DCB \n");
3159 ixgbe_set_dcb_queues(adapter);
3160 }
3161 #endif
3162 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3163 DPRINTK(PROBE, INFO, "FCoE enabled with RSS \n");
3164 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3165 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3166 ixgbe_set_fdir_queues(adapter);
3167 else
3168 ixgbe_set_rss_queues(adapter);
3169 }
3170 /* adding FCoE rx rings to the end */
3171 f->mask = adapter->num_rx_queues;
3172 adapter->num_rx_queues += f->indices;
3173 adapter->num_tx_queues += f->indices;
3174
3175 ret = true;
3176 }
3177
3178 return ret;
3179 }
3180
3181 #endif /* IXGBE_FCOE */
3182 /*
3183 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3184 * @adapter: board private structure to initialize
3185 *
3186 * This is the top level queue allocation routine. The order here is very
3187 * important, starting with the "most" number of features turned on at once,
3188 * and ending with the smallest set of features. This way large combinations
3189 * can be allocated if they're turned on, and smaller combinations are the
3190 * fallthrough conditions.
3191 *
3192 **/
3193 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
3194 {
3195 #ifdef IXGBE_FCOE
3196 if (ixgbe_set_fcoe_queues(adapter))
3197 goto done;
3198
3199 #endif /* IXGBE_FCOE */
3200 #ifdef CONFIG_IXGBE_DCB
3201 if (ixgbe_set_dcb_queues(adapter))
3202 goto done;
3203
3204 #endif
3205 if (ixgbe_set_fdir_queues(adapter))
3206 goto done;
3207
3208 if (ixgbe_set_rss_queues(adapter))
3209 goto done;
3210
3211 /* fallback to base case */
3212 adapter->num_rx_queues = 1;
3213 adapter->num_tx_queues = 1;
3214
3215 done:
3216 /* Notify the stack of the (possibly) reduced Tx Queue count. */
3217 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
3218 }
3219
3220 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
3221 int vectors)
3222 {
3223 int err, vector_threshold;
3224
3225 /* We'll want at least 3 (vector_threshold):
3226 * 1) TxQ[0] Cleanup
3227 * 2) RxQ[0] Cleanup
3228 * 3) Other (Link Status Change, etc.)
3229 * 4) TCP Timer (optional)
3230 */
3231 vector_threshold = MIN_MSIX_COUNT;
3232
3233 /* The more we get, the more we will assign to Tx/Rx Cleanup
3234 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
3235 * Right now, we simply care about how many we'll get; we'll
3236 * set them up later while requesting irq's.
3237 */
3238 while (vectors >= vector_threshold) {
3239 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
3240 vectors);
3241 if (!err) /* Success in acquiring all requested vectors. */
3242 break;
3243 else if (err < 0)
3244 vectors = 0; /* Nasty failure, quit now */
3245 else /* err == number of vectors we should try again with */
3246 vectors = err;
3247 }
3248
3249 if (vectors < vector_threshold) {
3250 /* Can't allocate enough MSI-X interrupts? Oh well.
3251 * This just means we'll go with either a single MSI
3252 * vector or fall back to legacy interrupts.
3253 */
3254 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
3255 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3256 kfree(adapter->msix_entries);
3257 adapter->msix_entries = NULL;
3258 } else {
3259 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
3260 /*
3261 * Adjust for only the vectors we'll use, which is minimum
3262 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
3263 * vectors we were allocated.
3264 */
3265 adapter->num_msix_vectors = min(vectors,
3266 adapter->max_msix_q_vectors + NON_Q_VECTORS);
3267 }
3268 }
3269
3270 /**
3271 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
3272 * @adapter: board private structure to initialize
3273 *
3274 * Cache the descriptor ring offsets for RSS to the assigned rings.
3275 *
3276 **/
3277 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
3278 {
3279 int i;
3280 bool ret = false;
3281
3282 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3283 for (i = 0; i < adapter->num_rx_queues; i++)
3284 adapter->rx_ring[i].reg_idx = i;
3285 for (i = 0; i < adapter->num_tx_queues; i++)
3286 adapter->tx_ring[i].reg_idx = i;
3287 ret = true;
3288 } else {
3289 ret = false;
3290 }
3291
3292 return ret;
3293 }
3294
3295 #ifdef CONFIG_IXGBE_DCB
3296 /**
3297 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
3298 * @adapter: board private structure to initialize
3299 *
3300 * Cache the descriptor ring offsets for DCB to the assigned rings.
3301 *
3302 **/
3303 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
3304 {
3305 int i;
3306 bool ret = false;
3307 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
3308
3309 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3310 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3311 /* the number of queues is assumed to be symmetric */
3312 for (i = 0; i < dcb_i; i++) {
3313 adapter->rx_ring[i].reg_idx = i << 3;
3314 adapter->tx_ring[i].reg_idx = i << 2;
3315 }
3316 ret = true;
3317 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
3318 if (dcb_i == 8) {
3319 /*
3320 * Tx TC0 starts at: descriptor queue 0
3321 * Tx TC1 starts at: descriptor queue 32
3322 * Tx TC2 starts at: descriptor queue 64
3323 * Tx TC3 starts at: descriptor queue 80
3324 * Tx TC4 starts at: descriptor queue 96
3325 * Tx TC5 starts at: descriptor queue 104
3326 * Tx TC6 starts at: descriptor queue 112
3327 * Tx TC7 starts at: descriptor queue 120
3328 *
3329 * Rx TC0-TC7 are offset by 16 queues each
3330 */
3331 for (i = 0; i < 3; i++) {
3332 adapter->tx_ring[i].reg_idx = i << 5;
3333 adapter->rx_ring[i].reg_idx = i << 4;
3334 }
3335 for ( ; i < 5; i++) {
3336 adapter->tx_ring[i].reg_idx =
3337 ((i + 2) << 4);
3338 adapter->rx_ring[i].reg_idx = i << 4;
3339 }
3340 for ( ; i < dcb_i; i++) {
3341 adapter->tx_ring[i].reg_idx =
3342 ((i + 8) << 3);
3343 adapter->rx_ring[i].reg_idx = i << 4;
3344 }
3345
3346 ret = true;
3347 } else if (dcb_i == 4) {
3348 /*
3349 * Tx TC0 starts at: descriptor queue 0
3350 * Tx TC1 starts at: descriptor queue 64
3351 * Tx TC2 starts at: descriptor queue 96
3352 * Tx TC3 starts at: descriptor queue 112
3353 *
3354 * Rx TC0-TC3 are offset by 32 queues each
3355 */
3356 adapter->tx_ring[0].reg_idx = 0;
3357 adapter->tx_ring[1].reg_idx = 64;
3358 adapter->tx_ring[2].reg_idx = 96;
3359 adapter->tx_ring[3].reg_idx = 112;
3360 for (i = 0 ; i < dcb_i; i++)
3361 adapter->rx_ring[i].reg_idx = i << 5;
3362
3363 ret = true;
3364 } else {
3365 ret = false;
3366 }
3367 } else {
3368 ret = false;
3369 }
3370 } else {
3371 ret = false;
3372 }
3373
3374 return ret;
3375 }
3376 #endif
3377
3378 /**
3379 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
3380 * @adapter: board private structure to initialize
3381 *
3382 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
3383 *
3384 **/
3385 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
3386 {
3387 int i;
3388 bool ret = false;
3389
3390 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3391 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3392 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
3393 for (i = 0; i < adapter->num_rx_queues; i++)
3394 adapter->rx_ring[i].reg_idx = i;
3395 for (i = 0; i < adapter->num_tx_queues; i++)
3396 adapter->tx_ring[i].reg_idx = i;
3397 ret = true;
3398 }
3399
3400 return ret;
3401 }
3402
3403 #ifdef IXGBE_FCOE
3404 /**
3405 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
3406 * @adapter: board private structure to initialize
3407 *
3408 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
3409 *
3410 */
3411 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
3412 {
3413 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
3414 bool ret = false;
3415 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3416
3417 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3418 #ifdef CONFIG_IXGBE_DCB
3419 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3420 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
3421
3422 ixgbe_cache_ring_dcb(adapter);
3423 /* find out queues in TC for FCoE */
3424 fcoe_rx_i = adapter->rx_ring[fcoe->tc].reg_idx + 1;
3425 fcoe_tx_i = adapter->tx_ring[fcoe->tc].reg_idx + 1;
3426 /*
3427 * In 82599, the number of Tx queues for each traffic
3428 * class for both 8-TC and 4-TC modes are:
3429 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
3430 * 8 TCs: 32 32 16 16 8 8 8 8
3431 * 4 TCs: 64 64 32 32
3432 * We have max 8 queues for FCoE, where 8 the is
3433 * FCoE redirection table size. If TC for FCoE is
3434 * less than or equal to TC3, we have enough queues
3435 * to add max of 8 queues for FCoE, so we start FCoE
3436 * tx descriptor from the next one, i.e., reg_idx + 1.
3437 * If TC for FCoE is above TC3, implying 8 TC mode,
3438 * and we need 8 for FCoE, we have to take all queues
3439 * in that traffic class for FCoE.
3440 */
3441 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
3442 fcoe_tx_i--;
3443 }
3444 #endif /* CONFIG_IXGBE_DCB */
3445 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3446 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3447 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3448 ixgbe_cache_ring_fdir(adapter);
3449 else
3450 ixgbe_cache_ring_rss(adapter);
3451
3452 fcoe_rx_i = f->mask;
3453 fcoe_tx_i = f->mask;
3454 }
3455 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
3456 adapter->rx_ring[f->mask + i].reg_idx = fcoe_rx_i;
3457 adapter->tx_ring[f->mask + i].reg_idx = fcoe_tx_i;
3458 }
3459 ret = true;
3460 }
3461 return ret;
3462 }
3463
3464 #endif /* IXGBE_FCOE */
3465 /**
3466 * ixgbe_cache_ring_register - Descriptor ring to register mapping
3467 * @adapter: board private structure to initialize
3468 *
3469 * Once we know the feature-set enabled for the device, we'll cache
3470 * the register offset the descriptor ring is assigned to.
3471 *
3472 * Note, the order the various feature calls is important. It must start with
3473 * the "most" features enabled at the same time, then trickle down to the
3474 * least amount of features turned on at once.
3475 **/
3476 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
3477 {
3478 /* start with default case */
3479 adapter->rx_ring[0].reg_idx = 0;
3480 adapter->tx_ring[0].reg_idx = 0;
3481
3482 #ifdef IXGBE_FCOE
3483 if (ixgbe_cache_ring_fcoe(adapter))
3484 return;
3485
3486 #endif /* IXGBE_FCOE */
3487 #ifdef CONFIG_IXGBE_DCB
3488 if (ixgbe_cache_ring_dcb(adapter))
3489 return;
3490
3491 #endif
3492 if (ixgbe_cache_ring_fdir(adapter))
3493 return;
3494
3495 if (ixgbe_cache_ring_rss(adapter))
3496 return;
3497 }
3498
3499 /**
3500 * ixgbe_alloc_queues - Allocate memory for all rings
3501 * @adapter: board private structure to initialize
3502 *
3503 * We allocate one ring per queue at run-time since we don't know the
3504 * number of queues at compile-time. The polling_netdev array is
3505 * intended for Multiqueue, but should work fine with a single queue.
3506 **/
3507 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
3508 {
3509 int i;
3510
3511 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
3512 sizeof(struct ixgbe_ring), GFP_KERNEL);
3513 if (!adapter->tx_ring)
3514 goto err_tx_ring_allocation;
3515
3516 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
3517 sizeof(struct ixgbe_ring), GFP_KERNEL);
3518 if (!adapter->rx_ring)
3519 goto err_rx_ring_allocation;
3520
3521 for (i = 0; i < adapter->num_tx_queues; i++) {
3522 adapter->tx_ring[i].count = adapter->tx_ring_count;
3523 adapter->tx_ring[i].queue_index = i;
3524 }
3525
3526 for (i = 0; i < adapter->num_rx_queues; i++) {
3527 adapter->rx_ring[i].count = adapter->rx_ring_count;
3528 adapter->rx_ring[i].queue_index = i;
3529 }
3530
3531 ixgbe_cache_ring_register(adapter);
3532
3533 return 0;
3534
3535 err_rx_ring_allocation:
3536 kfree(adapter->tx_ring);
3537 err_tx_ring_allocation:
3538 return -ENOMEM;
3539 }
3540
3541 /**
3542 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
3543 * @adapter: board private structure to initialize
3544 *
3545 * Attempt to configure the interrupts using the best available
3546 * capabilities of the hardware and the kernel.
3547 **/
3548 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
3549 {
3550 struct ixgbe_hw *hw = &adapter->hw;
3551 int err = 0;
3552 int vector, v_budget;
3553
3554 /*
3555 * It's easy to be greedy for MSI-X vectors, but it really
3556 * doesn't do us much good if we have a lot more vectors
3557 * than CPU's. So let's be conservative and only ask for
3558 * (roughly) twice the number of vectors as there are CPU's.
3559 */
3560 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
3561 (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
3562
3563 /*
3564 * At the same time, hardware can only support a maximum of
3565 * hw.mac->max_msix_vectors vectors. With features
3566 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
3567 * descriptor queues supported by our device. Thus, we cap it off in
3568 * those rare cases where the cpu count also exceeds our vector limit.
3569 */
3570 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
3571
3572 /* A failure in MSI-X entry allocation isn't fatal, but it does
3573 * mean we disable MSI-X capabilities of the adapter. */
3574 adapter->msix_entries = kcalloc(v_budget,
3575 sizeof(struct msix_entry), GFP_KERNEL);
3576 if (adapter->msix_entries) {
3577 for (vector = 0; vector < v_budget; vector++)
3578 adapter->msix_entries[vector].entry = vector;
3579
3580 ixgbe_acquire_msix_vectors(adapter, v_budget);
3581
3582 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3583 goto out;
3584 }
3585
3586 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
3587 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
3588 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3589 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3590 adapter->atr_sample_rate = 0;
3591 ixgbe_set_num_queues(adapter);
3592
3593 err = pci_enable_msi(adapter->pdev);
3594 if (!err) {
3595 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
3596 } else {
3597 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
3598 "falling back to legacy. Error: %d\n", err);
3599 /* reset err */
3600 err = 0;
3601 }
3602
3603 out:
3604 return err;
3605 }
3606
3607 /**
3608 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
3609 * @adapter: board private structure to initialize
3610 *
3611 * We allocate one q_vector per queue interrupt. If allocation fails we
3612 * return -ENOMEM.
3613 **/
3614 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
3615 {
3616 int q_idx, num_q_vectors;
3617 struct ixgbe_q_vector *q_vector;
3618 int napi_vectors;
3619 int (*poll)(struct napi_struct *, int);
3620
3621 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3622 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3623 napi_vectors = adapter->num_rx_queues;
3624 poll = &ixgbe_clean_rxtx_many;
3625 } else {
3626 num_q_vectors = 1;
3627 napi_vectors = 1;
3628 poll = &ixgbe_poll;
3629 }
3630
3631 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3632 q_vector = kzalloc(sizeof(struct ixgbe_q_vector), GFP_KERNEL);
3633 if (!q_vector)
3634 goto err_out;
3635 q_vector->adapter = adapter;
3636 if (q_vector->txr_count && !q_vector->rxr_count)
3637 q_vector->eitr = adapter->tx_eitr_param;
3638 else
3639 q_vector->eitr = adapter->rx_eitr_param;
3640 q_vector->v_idx = q_idx;
3641 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
3642 adapter->q_vector[q_idx] = q_vector;
3643 }
3644
3645 return 0;
3646
3647 err_out:
3648 while (q_idx) {
3649 q_idx--;
3650 q_vector = adapter->q_vector[q_idx];
3651 netif_napi_del(&q_vector->napi);
3652 kfree(q_vector);
3653 adapter->q_vector[q_idx] = NULL;
3654 }
3655 return -ENOMEM;
3656 }
3657
3658 /**
3659 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
3660 * @adapter: board private structure to initialize
3661 *
3662 * This function frees the memory allocated to the q_vectors. In addition if
3663 * NAPI is enabled it will delete any references to the NAPI struct prior
3664 * to freeing the q_vector.
3665 **/
3666 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
3667 {
3668 int q_idx, num_q_vectors;
3669
3670 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3671 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3672 else
3673 num_q_vectors = 1;
3674
3675 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3676 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
3677 adapter->q_vector[q_idx] = NULL;
3678 netif_napi_del(&q_vector->napi);
3679 kfree(q_vector);
3680 }
3681 }
3682
3683 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
3684 {
3685 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3686 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3687 pci_disable_msix(adapter->pdev);
3688 kfree(adapter->msix_entries);
3689 adapter->msix_entries = NULL;
3690 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
3691 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
3692 pci_disable_msi(adapter->pdev);
3693 }
3694 return;
3695 }
3696
3697 /**
3698 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
3699 * @adapter: board private structure to initialize
3700 *
3701 * We determine which interrupt scheme to use based on...
3702 * - Kernel support (MSI, MSI-X)
3703 * - which can be user-defined (via MODULE_PARAM)
3704 * - Hardware queue count (num_*_queues)
3705 * - defined by miscellaneous hardware support/features (RSS, etc.)
3706 **/
3707 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
3708 {
3709 int err;
3710
3711 /* Number of supported queues */
3712 ixgbe_set_num_queues(adapter);
3713
3714 err = ixgbe_set_interrupt_capability(adapter);
3715 if (err) {
3716 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
3717 goto err_set_interrupt;
3718 }
3719
3720 err = ixgbe_alloc_q_vectors(adapter);
3721 if (err) {
3722 DPRINTK(PROBE, ERR, "Unable to allocate memory for queue "
3723 "vectors\n");
3724 goto err_alloc_q_vectors;
3725 }
3726
3727 err = ixgbe_alloc_queues(adapter);
3728 if (err) {
3729 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
3730 goto err_alloc_queues;
3731 }
3732
3733 DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
3734 "Tx Queue count = %u\n",
3735 (adapter->num_rx_queues > 1) ? "Enabled" :
3736 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
3737
3738 set_bit(__IXGBE_DOWN, &adapter->state);
3739
3740 return 0;
3741
3742 err_alloc_queues:
3743 ixgbe_free_q_vectors(adapter);
3744 err_alloc_q_vectors:
3745 ixgbe_reset_interrupt_capability(adapter);
3746 err_set_interrupt:
3747 return err;
3748 }
3749
3750 /**
3751 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
3752 * @adapter: board private structure to clear interrupt scheme on
3753 *
3754 * We go through and clear interrupt specific resources and reset the structure
3755 * to pre-load conditions
3756 **/
3757 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
3758 {
3759 kfree(adapter->tx_ring);
3760 kfree(adapter->rx_ring);
3761 adapter->tx_ring = NULL;
3762 adapter->rx_ring = NULL;
3763
3764 ixgbe_free_q_vectors(adapter);
3765 ixgbe_reset_interrupt_capability(adapter);
3766 }
3767
3768 /**
3769 * ixgbe_sfp_timer - worker thread to find a missing module
3770 * @data: pointer to our adapter struct
3771 **/
3772 static void ixgbe_sfp_timer(unsigned long data)
3773 {
3774 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3775
3776 /*
3777 * Do the sfp_timer outside of interrupt context due to the
3778 * delays that sfp+ detection requires
3779 */
3780 schedule_work(&adapter->sfp_task);
3781 }
3782
3783 /**
3784 * ixgbe_sfp_task - worker thread to find a missing module
3785 * @work: pointer to work_struct containing our data
3786 **/
3787 static void ixgbe_sfp_task(struct work_struct *work)
3788 {
3789 struct ixgbe_adapter *adapter = container_of(work,
3790 struct ixgbe_adapter,
3791 sfp_task);
3792 struct ixgbe_hw *hw = &adapter->hw;
3793
3794 if ((hw->phy.type == ixgbe_phy_nl) &&
3795 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
3796 s32 ret = hw->phy.ops.identify_sfp(hw);
3797 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
3798 goto reschedule;
3799 ret = hw->phy.ops.reset(hw);
3800 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3801 dev_err(&adapter->pdev->dev, "failed to initialize "
3802 "because an unsupported SFP+ module type "
3803 "was detected.\n"
3804 "Reload the driver after installing a "
3805 "supported module.\n");
3806 unregister_netdev(adapter->netdev);
3807 } else {
3808 DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
3809 hw->phy.sfp_type);
3810 }
3811 /* don't need this routine any more */
3812 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3813 }
3814 return;
3815 reschedule:
3816 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
3817 mod_timer(&adapter->sfp_timer,
3818 round_jiffies(jiffies + (2 * HZ)));
3819 }
3820
3821 /**
3822 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
3823 * @adapter: board private structure to initialize
3824 *
3825 * ixgbe_sw_init initializes the Adapter private data structure.
3826 * Fields are initialized based on PCI device information and
3827 * OS network device settings (MTU size).
3828 **/
3829 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
3830 {
3831 struct ixgbe_hw *hw = &adapter->hw;
3832 struct pci_dev *pdev = adapter->pdev;
3833 unsigned int rss;
3834 #ifdef CONFIG_IXGBE_DCB
3835 int j;
3836 struct tc_configuration *tc;
3837 #endif
3838
3839 /* PCI config space info */
3840
3841 hw->vendor_id = pdev->vendor;
3842 hw->device_id = pdev->device;
3843 hw->revision_id = pdev->revision;
3844 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3845 hw->subsystem_device_id = pdev->subsystem_device;
3846
3847 /* Set capability flags */
3848 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
3849 adapter->ring_feature[RING_F_RSS].indices = rss;
3850 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
3851 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
3852 if (hw->mac.type == ixgbe_mac_82598EB) {
3853 if (hw->device_id == IXGBE_DEV_ID_82598AT)
3854 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
3855 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
3856 } else if (hw->mac.type == ixgbe_mac_82599EB) {
3857 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
3858 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
3859 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
3860 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
3861 adapter->ring_feature[RING_F_FDIR].indices =
3862 IXGBE_MAX_FDIR_INDICES;
3863 adapter->atr_sample_rate = 20;
3864 adapter->fdir_pballoc = 0;
3865 #ifdef IXGBE_FCOE
3866 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
3867 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
3868 adapter->ring_feature[RING_F_FCOE].indices = 0;
3869 /* Default traffic class to use for FCoE */
3870 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
3871 #endif /* IXGBE_FCOE */
3872 }
3873
3874 #ifdef CONFIG_IXGBE_DCB
3875 /* Configure DCB traffic classes */
3876 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
3877 tc = &adapter->dcb_cfg.tc_config[j];
3878 tc->path[DCB_TX_CONFIG].bwg_id = 0;
3879 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
3880 tc->path[DCB_RX_CONFIG].bwg_id = 0;
3881 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
3882 tc->dcb_pfc = pfc_disabled;
3883 }
3884 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
3885 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
3886 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
3887 adapter->dcb_cfg.pfc_mode_enable = false;
3888 adapter->dcb_cfg.round_robin_enable = false;
3889 adapter->dcb_set_bitmap = 0x00;
3890 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
3891 adapter->ring_feature[RING_F_DCB].indices);
3892
3893 #endif
3894
3895 /* default flow control settings */
3896 hw->fc.requested_mode = ixgbe_fc_full;
3897 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
3898 #ifdef CONFIG_DCB
3899 adapter->last_lfc_mode = hw->fc.current_mode;
3900 #endif
3901 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
3902 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
3903 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
3904 hw->fc.send_xon = true;
3905 hw->fc.disable_fc_autoneg = false;
3906
3907 /* enable itr by default in dynamic mode */
3908 adapter->rx_itr_setting = 1;
3909 adapter->rx_eitr_param = 20000;
3910 adapter->tx_itr_setting = 1;
3911 adapter->tx_eitr_param = 10000;
3912
3913 /* set defaults for eitr in MegaBytes */
3914 adapter->eitr_low = 10;
3915 adapter->eitr_high = 20;
3916
3917 /* set default ring sizes */
3918 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
3919 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
3920
3921 /* initialize eeprom parameters */
3922 if (ixgbe_init_eeprom_params_generic(hw)) {
3923 dev_err(&pdev->dev, "EEPROM initialization failed\n");
3924 return -EIO;
3925 }
3926
3927 /* enable rx csum by default */
3928 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
3929
3930 set_bit(__IXGBE_DOWN, &adapter->state);
3931
3932 return 0;
3933 }
3934
3935 /**
3936 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
3937 * @adapter: board private structure
3938 * @tx_ring: tx descriptor ring (for a specific queue) to setup
3939 *
3940 * Return 0 on success, negative on failure
3941 **/
3942 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
3943 struct ixgbe_ring *tx_ring)
3944 {
3945 struct pci_dev *pdev = adapter->pdev;
3946 int size;
3947
3948 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3949 tx_ring->tx_buffer_info = vmalloc(size);
3950 if (!tx_ring->tx_buffer_info)
3951 goto err;
3952 memset(tx_ring->tx_buffer_info, 0, size);
3953
3954 /* round up to nearest 4K */
3955 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3956 tx_ring->size = ALIGN(tx_ring->size, 4096);
3957
3958 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
3959 &tx_ring->dma);
3960 if (!tx_ring->desc)
3961 goto err;
3962
3963 tx_ring->next_to_use = 0;
3964 tx_ring->next_to_clean = 0;
3965 tx_ring->work_limit = tx_ring->count;
3966 return 0;
3967
3968 err:
3969 vfree(tx_ring->tx_buffer_info);
3970 tx_ring->tx_buffer_info = NULL;
3971 DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
3972 "descriptor ring\n");
3973 return -ENOMEM;
3974 }
3975
3976 /**
3977 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
3978 * @adapter: board private structure
3979 *
3980 * If this function returns with an error, then it's possible one or
3981 * more of the rings is populated (while the rest are not). It is the
3982 * callers duty to clean those orphaned rings.
3983 *
3984 * Return 0 on success, negative on failure
3985 **/
3986 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
3987 {
3988 int i, err = 0;
3989
3990 for (i = 0; i < adapter->num_tx_queues; i++) {
3991 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
3992 if (!err)
3993 continue;
3994 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
3995 break;
3996 }
3997
3998 return err;
3999 }
4000
4001 /**
4002 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4003 * @adapter: board private structure
4004 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4005 *
4006 * Returns 0 on success, negative on failure
4007 **/
4008 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4009 struct ixgbe_ring *rx_ring)
4010 {
4011 struct pci_dev *pdev = adapter->pdev;
4012 int size;
4013
4014 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4015 rx_ring->rx_buffer_info = vmalloc(size);
4016 if (!rx_ring->rx_buffer_info) {
4017 DPRINTK(PROBE, ERR,
4018 "vmalloc allocation failed for the rx desc ring\n");
4019 goto alloc_failed;
4020 }
4021 memset(rx_ring->rx_buffer_info, 0, size);
4022
4023 /* Round up to nearest 4K */
4024 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4025 rx_ring->size = ALIGN(rx_ring->size, 4096);
4026
4027 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
4028
4029 if (!rx_ring->desc) {
4030 DPRINTK(PROBE, ERR,
4031 "Memory allocation failed for the rx desc ring\n");
4032 vfree(rx_ring->rx_buffer_info);
4033 goto alloc_failed;
4034 }
4035
4036 rx_ring->next_to_clean = 0;
4037 rx_ring->next_to_use = 0;
4038
4039 return 0;
4040
4041 alloc_failed:
4042 return -ENOMEM;
4043 }
4044
4045 /**
4046 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4047 * @adapter: board private structure
4048 *
4049 * If this function returns with an error, then it's possible one or
4050 * more of the rings is populated (while the rest are not). It is the
4051 * callers duty to clean those orphaned rings.
4052 *
4053 * Return 0 on success, negative on failure
4054 **/
4055
4056 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4057 {
4058 int i, err = 0;
4059
4060 for (i = 0; i < adapter->num_rx_queues; i++) {
4061 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
4062 if (!err)
4063 continue;
4064 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
4065 break;
4066 }
4067
4068 return err;
4069 }
4070
4071 /**
4072 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4073 * @adapter: board private structure
4074 * @tx_ring: Tx descriptor ring for a specific queue
4075 *
4076 * Free all transmit software resources
4077 **/
4078 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4079 struct ixgbe_ring *tx_ring)
4080 {
4081 struct pci_dev *pdev = adapter->pdev;
4082
4083 ixgbe_clean_tx_ring(adapter, tx_ring);
4084
4085 vfree(tx_ring->tx_buffer_info);
4086 tx_ring->tx_buffer_info = NULL;
4087
4088 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
4089
4090 tx_ring->desc = NULL;
4091 }
4092
4093 /**
4094 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4095 * @adapter: board private structure
4096 *
4097 * Free all transmit software resources
4098 **/
4099 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4100 {
4101 int i;
4102
4103 for (i = 0; i < adapter->num_tx_queues; i++)
4104 if (adapter->tx_ring[i].desc)
4105 ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
4106 }
4107
4108 /**
4109 * ixgbe_free_rx_resources - Free Rx Resources
4110 * @adapter: board private structure
4111 * @rx_ring: ring to clean the resources from
4112 *
4113 * Free all receive software resources
4114 **/
4115 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
4116 struct ixgbe_ring *rx_ring)
4117 {
4118 struct pci_dev *pdev = adapter->pdev;
4119
4120 ixgbe_clean_rx_ring(adapter, rx_ring);
4121
4122 vfree(rx_ring->rx_buffer_info);
4123 rx_ring->rx_buffer_info = NULL;
4124
4125 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
4126
4127 rx_ring->desc = NULL;
4128 }
4129
4130 /**
4131 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4132 * @adapter: board private structure
4133 *
4134 * Free all receive software resources
4135 **/
4136 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4137 {
4138 int i;
4139
4140 for (i = 0; i < adapter->num_rx_queues; i++)
4141 if (adapter->rx_ring[i].desc)
4142 ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
4143 }
4144
4145 /**
4146 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4147 * @netdev: network interface device structure
4148 * @new_mtu: new value for maximum frame size
4149 *
4150 * Returns 0 on success, negative on failure
4151 **/
4152 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4153 {
4154 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4155 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4156
4157 /* MTU < 68 is an error and causes problems on some kernels */
4158 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4159 return -EINVAL;
4160
4161 DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
4162 netdev->mtu, new_mtu);
4163 /* must set new MTU before calling down or up */
4164 netdev->mtu = new_mtu;
4165
4166 if (netif_running(netdev))
4167 ixgbe_reinit_locked(adapter);
4168
4169 return 0;
4170 }
4171
4172 /**
4173 * ixgbe_open - Called when a network interface is made active
4174 * @netdev: network interface device structure
4175 *
4176 * Returns 0 on success, negative value on failure
4177 *
4178 * The open entry point is called when a network interface is made
4179 * active by the system (IFF_UP). At this point all resources needed
4180 * for transmit and receive operations are allocated, the interrupt
4181 * handler is registered with the OS, the watchdog timer is started,
4182 * and the stack is notified that the interface is ready.
4183 **/
4184 static int ixgbe_open(struct net_device *netdev)
4185 {
4186 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4187 int err;
4188
4189 /* disallow open during test */
4190 if (test_bit(__IXGBE_TESTING, &adapter->state))
4191 return -EBUSY;
4192
4193 netif_carrier_off(netdev);
4194
4195 /* allocate transmit descriptors */
4196 err = ixgbe_setup_all_tx_resources(adapter);
4197 if (err)
4198 goto err_setup_tx;
4199
4200 /* allocate receive descriptors */
4201 err = ixgbe_setup_all_rx_resources(adapter);
4202 if (err)
4203 goto err_setup_rx;
4204
4205 ixgbe_configure(adapter);
4206
4207 err = ixgbe_request_irq(adapter);
4208 if (err)
4209 goto err_req_irq;
4210
4211 err = ixgbe_up_complete(adapter);
4212 if (err)
4213 goto err_up;
4214
4215 netif_tx_start_all_queues(netdev);
4216
4217 return 0;
4218
4219 err_up:
4220 ixgbe_release_hw_control(adapter);
4221 ixgbe_free_irq(adapter);
4222 err_req_irq:
4223 err_setup_rx:
4224 ixgbe_free_all_rx_resources(adapter);
4225 err_setup_tx:
4226 ixgbe_free_all_tx_resources(adapter);
4227 ixgbe_reset(adapter);
4228
4229 return err;
4230 }
4231
4232 /**
4233 * ixgbe_close - Disables a network interface
4234 * @netdev: network interface device structure
4235 *
4236 * Returns 0, this is not allowed to fail
4237 *
4238 * The close entry point is called when an interface is de-activated
4239 * by the OS. The hardware is still under the drivers control, but
4240 * needs to be disabled. A global MAC reset is issued to stop the
4241 * hardware, and all transmit and receive resources are freed.
4242 **/
4243 static int ixgbe_close(struct net_device *netdev)
4244 {
4245 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4246
4247 ixgbe_down(adapter);
4248 ixgbe_free_irq(adapter);
4249
4250 ixgbe_free_all_tx_resources(adapter);
4251 ixgbe_free_all_rx_resources(adapter);
4252
4253 ixgbe_release_hw_control(adapter);
4254
4255 return 0;
4256 }
4257
4258 #ifdef CONFIG_PM
4259 static int ixgbe_resume(struct pci_dev *pdev)
4260 {
4261 struct net_device *netdev = pci_get_drvdata(pdev);
4262 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4263 u32 err;
4264
4265 pci_set_power_state(pdev, PCI_D0);
4266 pci_restore_state(pdev);
4267
4268 err = pci_enable_device_mem(pdev);
4269 if (err) {
4270 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
4271 "suspend\n");
4272 return err;
4273 }
4274 pci_set_master(pdev);
4275
4276 pci_wake_from_d3(pdev, false);
4277
4278 err = ixgbe_init_interrupt_scheme(adapter);
4279 if (err) {
4280 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
4281 "device\n");
4282 return err;
4283 }
4284
4285 ixgbe_reset(adapter);
4286
4287 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4288
4289 if (netif_running(netdev)) {
4290 err = ixgbe_open(adapter->netdev);
4291 if (err)
4292 return err;
4293 }
4294
4295 netif_device_attach(netdev);
4296
4297 return 0;
4298 }
4299 #endif /* CONFIG_PM */
4300
4301 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4302 {
4303 struct net_device *netdev = pci_get_drvdata(pdev);
4304 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4305 struct ixgbe_hw *hw = &adapter->hw;
4306 u32 ctrl, fctrl;
4307 u32 wufc = adapter->wol;
4308 #ifdef CONFIG_PM
4309 int retval = 0;
4310 #endif
4311
4312 netif_device_detach(netdev);
4313
4314 if (netif_running(netdev)) {
4315 ixgbe_down(adapter);
4316 ixgbe_free_irq(adapter);
4317 ixgbe_free_all_tx_resources(adapter);
4318 ixgbe_free_all_rx_resources(adapter);
4319 }
4320 ixgbe_clear_interrupt_scheme(adapter);
4321
4322 #ifdef CONFIG_PM
4323 retval = pci_save_state(pdev);
4324 if (retval)
4325 return retval;
4326
4327 #endif
4328 if (wufc) {
4329 ixgbe_set_rx_mode(netdev);
4330
4331 /* turn on all-multi mode if wake on multicast is enabled */
4332 if (wufc & IXGBE_WUFC_MC) {
4333 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4334 fctrl |= IXGBE_FCTRL_MPE;
4335 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4336 }
4337
4338 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4339 ctrl |= IXGBE_CTRL_GIO_DIS;
4340 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4341
4342 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4343 } else {
4344 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4345 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4346 }
4347
4348 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
4349 pci_wake_from_d3(pdev, true);
4350 else
4351 pci_wake_from_d3(pdev, false);
4352
4353 *enable_wake = !!wufc;
4354
4355 ixgbe_release_hw_control(adapter);
4356
4357 pci_disable_device(pdev);
4358
4359 return 0;
4360 }
4361
4362 #ifdef CONFIG_PM
4363 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4364 {
4365 int retval;
4366 bool wake;
4367
4368 retval = __ixgbe_shutdown(pdev, &wake);
4369 if (retval)
4370 return retval;
4371
4372 if (wake) {
4373 pci_prepare_to_sleep(pdev);
4374 } else {
4375 pci_wake_from_d3(pdev, false);
4376 pci_set_power_state(pdev, PCI_D3hot);
4377 }
4378
4379 return 0;
4380 }
4381 #endif /* CONFIG_PM */
4382
4383 static void ixgbe_shutdown(struct pci_dev *pdev)
4384 {
4385 bool wake;
4386
4387 __ixgbe_shutdown(pdev, &wake);
4388
4389 if (system_state == SYSTEM_POWER_OFF) {
4390 pci_wake_from_d3(pdev, wake);
4391 pci_set_power_state(pdev, PCI_D3hot);
4392 }
4393 }
4394
4395 /**
4396 * ixgbe_update_stats - Update the board statistics counters.
4397 * @adapter: board private structure
4398 **/
4399 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4400 {
4401 struct net_device *netdev = adapter->netdev;
4402 struct ixgbe_hw *hw = &adapter->hw;
4403 u64 total_mpc = 0;
4404 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4405
4406 if (hw->mac.type == ixgbe_mac_82599EB) {
4407 u64 rsc_count = 0;
4408 for (i = 0; i < 16; i++)
4409 adapter->hw_rx_no_dma_resources +=
4410 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4411 for (i = 0; i < adapter->num_rx_queues; i++)
4412 rsc_count += adapter->rx_ring[i].rsc_count;
4413 adapter->rsc_count = rsc_count;
4414 }
4415
4416 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
4417 for (i = 0; i < 8; i++) {
4418 /* for packet buffers not used, the register should read 0 */
4419 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
4420 missed_rx += mpc;
4421 adapter->stats.mpc[i] += mpc;
4422 total_mpc += adapter->stats.mpc[i];
4423 if (hw->mac.type == ixgbe_mac_82598EB)
4424 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
4425 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
4426 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
4427 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
4428 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
4429 if (hw->mac.type == ixgbe_mac_82599EB) {
4430 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4431 IXGBE_PXONRXCNT(i));
4432 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4433 IXGBE_PXOFFRXCNT(i));
4434 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4435 } else {
4436 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4437 IXGBE_PXONRXC(i));
4438 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4439 IXGBE_PXOFFRXC(i));
4440 }
4441 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
4442 IXGBE_PXONTXC(i));
4443 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
4444 IXGBE_PXOFFTXC(i));
4445 }
4446 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
4447 /* work around hardware counting issue */
4448 adapter->stats.gprc -= missed_rx;
4449
4450 /* 82598 hardware only has a 32 bit counter in the high register */
4451 if (hw->mac.type == ixgbe_mac_82599EB) {
4452 u64 tmp;
4453 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
4454 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
4455 adapter->stats.gorc += (tmp << 32);
4456 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
4457 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
4458 adapter->stats.gotc += (tmp << 32);
4459 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
4460 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
4461 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
4462 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
4463 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
4464 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
4465 #ifdef IXGBE_FCOE
4466 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
4467 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
4468 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
4469 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
4470 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
4471 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
4472 #endif /* IXGBE_FCOE */
4473 } else {
4474 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
4475 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
4476 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
4477 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
4478 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
4479 }
4480 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
4481 adapter->stats.bprc += bprc;
4482 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
4483 if (hw->mac.type == ixgbe_mac_82598EB)
4484 adapter->stats.mprc -= bprc;
4485 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
4486 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
4487 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
4488 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
4489 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
4490 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
4491 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
4492 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
4493 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
4494 adapter->stats.lxontxc += lxon;
4495 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
4496 adapter->stats.lxofftxc += lxoff;
4497 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4498 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
4499 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
4500 /*
4501 * 82598 errata - tx of flow control packets is included in tx counters
4502 */
4503 xon_off_tot = lxon + lxoff;
4504 adapter->stats.gptc -= xon_off_tot;
4505 adapter->stats.mptc -= xon_off_tot;
4506 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
4507 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4508 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
4509 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
4510 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
4511 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
4512 adapter->stats.ptc64 -= xon_off_tot;
4513 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
4514 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
4515 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
4516 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
4517 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
4518 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
4519
4520 /* Fill out the OS statistics structure */
4521 netdev->stats.multicast = adapter->stats.mprc;
4522
4523 /* Rx Errors */
4524 netdev->stats.rx_errors = adapter->stats.crcerrs +
4525 adapter->stats.rlec;
4526 netdev->stats.rx_dropped = 0;
4527 netdev->stats.rx_length_errors = adapter->stats.rlec;
4528 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4529 netdev->stats.rx_missed_errors = total_mpc;
4530 }
4531
4532 /**
4533 * ixgbe_watchdog - Timer Call-back
4534 * @data: pointer to adapter cast into an unsigned long
4535 **/
4536 static void ixgbe_watchdog(unsigned long data)
4537 {
4538 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4539 struct ixgbe_hw *hw = &adapter->hw;
4540 u64 eics = 0;
4541 int i;
4542
4543 /*
4544 * Do the watchdog outside of interrupt context due to the lovely
4545 * delays that some of the newer hardware requires
4546 */
4547
4548 if (test_bit(__IXGBE_DOWN, &adapter->state))
4549 goto watchdog_short_circuit;
4550
4551 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
4552 /*
4553 * for legacy and MSI interrupts don't set any bits
4554 * that are enabled for EIAM, because this operation
4555 * would set *both* EIMS and EICS for any bit in EIAM
4556 */
4557 IXGBE_WRITE_REG(hw, IXGBE_EICS,
4558 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
4559 goto watchdog_reschedule;
4560 }
4561
4562 /* get one bit for every active tx/rx interrupt vector */
4563 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
4564 struct ixgbe_q_vector *qv = adapter->q_vector[i];
4565 if (qv->rxr_count || qv->txr_count)
4566 eics |= ((u64)1 << i);
4567 }
4568
4569 /* Cause software interrupt to ensure rx rings are cleaned */
4570 ixgbe_irq_rearm_queues(adapter, eics);
4571
4572 watchdog_reschedule:
4573 /* Reset the timer */
4574 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
4575
4576 watchdog_short_circuit:
4577 schedule_work(&adapter->watchdog_task);
4578 }
4579
4580 /**
4581 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
4582 * @work: pointer to work_struct containing our data
4583 **/
4584 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
4585 {
4586 struct ixgbe_adapter *adapter = container_of(work,
4587 struct ixgbe_adapter,
4588 multispeed_fiber_task);
4589 struct ixgbe_hw *hw = &adapter->hw;
4590 u32 autoneg;
4591 bool negotiation;
4592
4593 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
4594 autoneg = hw->phy.autoneg_advertised;
4595 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
4596 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
4597 if (hw->mac.ops.setup_link)
4598 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
4599 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4600 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
4601 }
4602
4603 /**
4604 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
4605 * @work: pointer to work_struct containing our data
4606 **/
4607 static void ixgbe_sfp_config_module_task(struct work_struct *work)
4608 {
4609 struct ixgbe_adapter *adapter = container_of(work,
4610 struct ixgbe_adapter,
4611 sfp_config_module_task);
4612 struct ixgbe_hw *hw = &adapter->hw;
4613 u32 err;
4614
4615 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
4616
4617 /* Time for electrical oscillations to settle down */
4618 msleep(100);
4619 err = hw->phy.ops.identify_sfp(hw);
4620
4621 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4622 dev_err(&adapter->pdev->dev, "failed to initialize because "
4623 "an unsupported SFP+ module type was detected.\n"
4624 "Reload the driver after installing a supported "
4625 "module.\n");
4626 unregister_netdev(adapter->netdev);
4627 return;
4628 }
4629 hw->mac.ops.setup_sfp(hw);
4630
4631 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
4632 /* This will also work for DA Twinax connections */
4633 schedule_work(&adapter->multispeed_fiber_task);
4634 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
4635 }
4636
4637 /**
4638 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
4639 * @work: pointer to work_struct containing our data
4640 **/
4641 static void ixgbe_fdir_reinit_task(struct work_struct *work)
4642 {
4643 struct ixgbe_adapter *adapter = container_of(work,
4644 struct ixgbe_adapter,
4645 fdir_reinit_task);
4646 struct ixgbe_hw *hw = &adapter->hw;
4647 int i;
4648
4649 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
4650 for (i = 0; i < adapter->num_tx_queues; i++)
4651 set_bit(__IXGBE_FDIR_INIT_DONE,
4652 &(adapter->tx_ring[i].reinit_state));
4653 } else {
4654 DPRINTK(PROBE, ERR, "failed to finish FDIR re-initialization, "
4655 "ignored adding FDIR ATR filters \n");
4656 }
4657 /* Done FDIR Re-initialization, enable transmits */
4658 netif_tx_start_all_queues(adapter->netdev);
4659 }
4660
4661 /**
4662 * ixgbe_watchdog_task - worker thread to bring link up
4663 * @work: pointer to work_struct containing our data
4664 **/
4665 static void ixgbe_watchdog_task(struct work_struct *work)
4666 {
4667 struct ixgbe_adapter *adapter = container_of(work,
4668 struct ixgbe_adapter,
4669 watchdog_task);
4670 struct net_device *netdev = adapter->netdev;
4671 struct ixgbe_hw *hw = &adapter->hw;
4672 u32 link_speed = adapter->link_speed;
4673 bool link_up = adapter->link_up;
4674 int i;
4675 struct ixgbe_ring *tx_ring;
4676 int some_tx_pending = 0;
4677
4678 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
4679
4680 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
4681 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
4682 if (link_up) {
4683 #ifdef CONFIG_DCB
4684 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4685 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
4686 hw->mac.ops.fc_enable(hw, i);
4687 } else {
4688 hw->mac.ops.fc_enable(hw, 0);
4689 }
4690 #else
4691 hw->mac.ops.fc_enable(hw, 0);
4692 #endif
4693 }
4694
4695 if (link_up ||
4696 time_after(jiffies, (adapter->link_check_timeout +
4697 IXGBE_TRY_LINK_TIMEOUT))) {
4698 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4699 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
4700 }
4701 adapter->link_up = link_up;
4702 adapter->link_speed = link_speed;
4703 }
4704
4705 if (link_up) {
4706 if (!netif_carrier_ok(netdev)) {
4707 bool flow_rx, flow_tx;
4708
4709 if (hw->mac.type == ixgbe_mac_82599EB) {
4710 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
4711 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
4712 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
4713 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
4714 } else {
4715 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4716 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
4717 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
4718 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
4719 }
4720
4721 printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
4722 "Flow Control: %s\n",
4723 netdev->name,
4724 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
4725 "10 Gbps" :
4726 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
4727 "1 Gbps" : "unknown speed")),
4728 ((flow_rx && flow_tx) ? "RX/TX" :
4729 (flow_rx ? "RX" :
4730 (flow_tx ? "TX" : "None"))));
4731
4732 netif_carrier_on(netdev);
4733 } else {
4734 /* Force detection of hung controller */
4735 adapter->detect_tx_hung = true;
4736 }
4737 } else {
4738 adapter->link_up = false;
4739 adapter->link_speed = 0;
4740 if (netif_carrier_ok(netdev)) {
4741 printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
4742 netdev->name);
4743 netif_carrier_off(netdev);
4744 }
4745 }
4746
4747 if (!netif_carrier_ok(netdev)) {
4748 for (i = 0; i < adapter->num_tx_queues; i++) {
4749 tx_ring = &adapter->tx_ring[i];
4750 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
4751 some_tx_pending = 1;
4752 break;
4753 }
4754 }
4755
4756 if (some_tx_pending) {
4757 /* We've lost link, so the controller stops DMA,
4758 * but we've got queued Tx work that's never going
4759 * to get done, so reset controller to flush Tx.
4760 * (Do the reset outside of interrupt context).
4761 */
4762 schedule_work(&adapter->reset_task);
4763 }
4764 }
4765
4766 ixgbe_update_stats(adapter);
4767 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
4768 }
4769
4770 static int ixgbe_tso(struct ixgbe_adapter *adapter,
4771 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
4772 u32 tx_flags, u8 *hdr_len)
4773 {
4774 struct ixgbe_adv_tx_context_desc *context_desc;
4775 unsigned int i;
4776 int err;
4777 struct ixgbe_tx_buffer *tx_buffer_info;
4778 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
4779 u32 mss_l4len_idx, l4len;
4780
4781 if (skb_is_gso(skb)) {
4782 if (skb_header_cloned(skb)) {
4783 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4784 if (err)
4785 return err;
4786 }
4787 l4len = tcp_hdrlen(skb);
4788 *hdr_len += l4len;
4789
4790 if (skb->protocol == htons(ETH_P_IP)) {
4791 struct iphdr *iph = ip_hdr(skb);
4792 iph->tot_len = 0;
4793 iph->check = 0;
4794 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4795 iph->daddr, 0,
4796 IPPROTO_TCP,
4797 0);
4798 adapter->hw_tso_ctxt++;
4799 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
4800 ipv6_hdr(skb)->payload_len = 0;
4801 tcp_hdr(skb)->check =
4802 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4803 &ipv6_hdr(skb)->daddr,
4804 0, IPPROTO_TCP, 0);
4805 adapter->hw_tso6_ctxt++;
4806 }
4807
4808 i = tx_ring->next_to_use;
4809
4810 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4811 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4812
4813 /* VLAN MACLEN IPLEN */
4814 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4815 vlan_macip_lens |=
4816 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4817 vlan_macip_lens |= ((skb_network_offset(skb)) <<
4818 IXGBE_ADVTXD_MACLEN_SHIFT);
4819 *hdr_len += skb_network_offset(skb);
4820 vlan_macip_lens |=
4821 (skb_transport_header(skb) - skb_network_header(skb));
4822 *hdr_len +=
4823 (skb_transport_header(skb) - skb_network_header(skb));
4824 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4825 context_desc->seqnum_seed = 0;
4826
4827 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4828 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
4829 IXGBE_ADVTXD_DTYP_CTXT);
4830
4831 if (skb->protocol == htons(ETH_P_IP))
4832 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4833 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
4834 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4835
4836 /* MSS L4LEN IDX */
4837 mss_l4len_idx =
4838 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
4839 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
4840 /* use index 1 for TSO */
4841 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4842 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4843
4844 tx_buffer_info->time_stamp = jiffies;
4845 tx_buffer_info->next_to_watch = i;
4846
4847 i++;
4848 if (i == tx_ring->count)
4849 i = 0;
4850 tx_ring->next_to_use = i;
4851
4852 return true;
4853 }
4854 return false;
4855 }
4856
4857 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
4858 struct ixgbe_ring *tx_ring,
4859 struct sk_buff *skb, u32 tx_flags)
4860 {
4861 struct ixgbe_adv_tx_context_desc *context_desc;
4862 unsigned int i;
4863 struct ixgbe_tx_buffer *tx_buffer_info;
4864 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
4865
4866 if (skb->ip_summed == CHECKSUM_PARTIAL ||
4867 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
4868 i = tx_ring->next_to_use;
4869 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4870 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4871
4872 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4873 vlan_macip_lens |=
4874 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4875 vlan_macip_lens |= (skb_network_offset(skb) <<
4876 IXGBE_ADVTXD_MACLEN_SHIFT);
4877 if (skb->ip_summed == CHECKSUM_PARTIAL)
4878 vlan_macip_lens |= (skb_transport_header(skb) -
4879 skb_network_header(skb));
4880
4881 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4882 context_desc->seqnum_seed = 0;
4883
4884 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
4885 IXGBE_ADVTXD_DTYP_CTXT);
4886
4887 if (skb->ip_summed == CHECKSUM_PARTIAL) {
4888 switch (skb->protocol) {
4889 case cpu_to_be16(ETH_P_IP):
4890 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4891 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4892 type_tucmd_mlhl |=
4893 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4894 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
4895 type_tucmd_mlhl |=
4896 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
4897 break;
4898 case cpu_to_be16(ETH_P_IPV6):
4899 /* XXX what about other V6 headers?? */
4900 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4901 type_tucmd_mlhl |=
4902 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4903 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
4904 type_tucmd_mlhl |=
4905 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
4906 break;
4907 default:
4908 if (unlikely(net_ratelimit())) {
4909 DPRINTK(PROBE, WARNING,
4910 "partial checksum but proto=%x!\n",
4911 skb->protocol);
4912 }
4913 break;
4914 }
4915 }
4916
4917 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4918 /* use index zero for tx checksum offload */
4919 context_desc->mss_l4len_idx = 0;
4920
4921 tx_buffer_info->time_stamp = jiffies;
4922 tx_buffer_info->next_to_watch = i;
4923
4924 adapter->hw_csum_tx_good++;
4925 i++;
4926 if (i == tx_ring->count)
4927 i = 0;
4928 tx_ring->next_to_use = i;
4929
4930 return true;
4931 }
4932
4933 return false;
4934 }
4935
4936 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
4937 struct ixgbe_ring *tx_ring,
4938 struct sk_buff *skb, u32 tx_flags,
4939 unsigned int first)
4940 {
4941 struct ixgbe_tx_buffer *tx_buffer_info;
4942 unsigned int len;
4943 unsigned int total = skb->len;
4944 unsigned int offset = 0, size, count = 0, i;
4945 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
4946 unsigned int f;
4947 dma_addr_t *map;
4948
4949 i = tx_ring->next_to_use;
4950
4951 if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
4952 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
4953 return 0;
4954 }
4955
4956 map = skb_shinfo(skb)->dma_maps;
4957
4958 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
4959 /* excluding fcoe_crc_eof for FCoE */
4960 total -= sizeof(struct fcoe_crc_eof);
4961
4962 len = min(skb_headlen(skb), total);
4963 while (len) {
4964 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4965 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4966
4967 tx_buffer_info->length = size;
4968 tx_buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
4969 tx_buffer_info->time_stamp = jiffies;
4970 tx_buffer_info->next_to_watch = i;
4971
4972 len -= size;
4973 total -= size;
4974 offset += size;
4975 count++;
4976
4977 if (len) {
4978 i++;
4979 if (i == tx_ring->count)
4980 i = 0;
4981 }
4982 }
4983
4984 for (f = 0; f < nr_frags; f++) {
4985 struct skb_frag_struct *frag;
4986
4987 frag = &skb_shinfo(skb)->frags[f];
4988 len = min((unsigned int)frag->size, total);
4989 offset = 0;
4990
4991 while (len) {
4992 i++;
4993 if (i == tx_ring->count)
4994 i = 0;
4995
4996 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4997 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4998
4999 tx_buffer_info->length = size;
5000 tx_buffer_info->dma = map[f] + offset;
5001 tx_buffer_info->time_stamp = jiffies;
5002 tx_buffer_info->next_to_watch = i;
5003
5004 len -= size;
5005 total -= size;
5006 offset += size;
5007 count++;
5008 }
5009 if (total == 0)
5010 break;
5011 }
5012
5013 tx_ring->tx_buffer_info[i].skb = skb;
5014 tx_ring->tx_buffer_info[first].next_to_watch = i;
5015
5016 return count;
5017 }
5018
5019 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
5020 struct ixgbe_ring *tx_ring,
5021 int tx_flags, int count, u32 paylen, u8 hdr_len)
5022 {
5023 union ixgbe_adv_tx_desc *tx_desc = NULL;
5024 struct ixgbe_tx_buffer *tx_buffer_info;
5025 u32 olinfo_status = 0, cmd_type_len = 0;
5026 unsigned int i;
5027 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
5028
5029 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
5030
5031 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
5032
5033 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5034 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
5035
5036 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
5037 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5038
5039 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5040 IXGBE_ADVTXD_POPTS_SHIFT;
5041
5042 /* use index 1 context for tso */
5043 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5044 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5045 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
5046 IXGBE_ADVTXD_POPTS_SHIFT;
5047
5048 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5049 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5050 IXGBE_ADVTXD_POPTS_SHIFT;
5051
5052 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5053 olinfo_status |= IXGBE_ADVTXD_CC;
5054 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5055 if (tx_flags & IXGBE_TX_FLAGS_FSO)
5056 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5057 }
5058
5059 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
5060
5061 i = tx_ring->next_to_use;
5062 while (count--) {
5063 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5064 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
5065 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
5066 tx_desc->read.cmd_type_len =
5067 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
5068 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5069 i++;
5070 if (i == tx_ring->count)
5071 i = 0;
5072 }
5073
5074 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
5075
5076 /*
5077 * Force memory writes to complete before letting h/w
5078 * know there are new descriptors to fetch. (Only
5079 * applicable for weak-ordered memory model archs,
5080 * such as IA-64).
5081 */
5082 wmb();
5083
5084 tx_ring->next_to_use = i;
5085 writel(i, adapter->hw.hw_addr + tx_ring->tail);
5086 }
5087
5088 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5089 int queue, u32 tx_flags)
5090 {
5091 /* Right now, we support IPv4 only */
5092 struct ixgbe_atr_input atr_input;
5093 struct tcphdr *th;
5094 struct iphdr *iph = ip_hdr(skb);
5095 struct ethhdr *eth = (struct ethhdr *)skb->data;
5096 u16 vlan_id, src_port, dst_port, flex_bytes;
5097 u32 src_ipv4_addr, dst_ipv4_addr;
5098 u8 l4type = 0;
5099
5100 /* check if we're UDP or TCP */
5101 if (iph->protocol == IPPROTO_TCP) {
5102 th = tcp_hdr(skb);
5103 src_port = th->source;
5104 dst_port = th->dest;
5105 l4type |= IXGBE_ATR_L4TYPE_TCP;
5106 /* l4type IPv4 type is 0, no need to assign */
5107 } else {
5108 /* Unsupported L4 header, just bail here */
5109 return;
5110 }
5111
5112 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
5113
5114 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
5115 IXGBE_TX_FLAGS_VLAN_SHIFT;
5116 src_ipv4_addr = iph->saddr;
5117 dst_ipv4_addr = iph->daddr;
5118 flex_bytes = eth->h_proto;
5119
5120 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
5121 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
5122 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
5123 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
5124 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
5125 /* src and dst are inverted, think how the receiver sees them */
5126 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
5127 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
5128
5129 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
5130 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
5131 }
5132
5133 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
5134 struct ixgbe_ring *tx_ring, int size)
5135 {
5136 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5137
5138 netif_stop_subqueue(netdev, tx_ring->queue_index);
5139 /* Herbert's original patch had:
5140 * smp_mb__after_netif_stop_queue();
5141 * but since that doesn't exist yet, just open code it. */
5142 smp_mb();
5143
5144 /* We need to check again in a case another CPU has just
5145 * made room available. */
5146 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
5147 return -EBUSY;
5148
5149 /* A reprieve! - use start_queue because it doesn't call schedule */
5150 netif_start_subqueue(netdev, tx_ring->queue_index);
5151 ++adapter->restart_queue;
5152 return 0;
5153 }
5154
5155 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
5156 struct ixgbe_ring *tx_ring, int size)
5157 {
5158 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
5159 return 0;
5160 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
5161 }
5162
5163 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
5164 {
5165 struct ixgbe_adapter *adapter = netdev_priv(dev);
5166
5167 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
5168 return smp_processor_id();
5169
5170 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5171 return (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK) >> 13;
5172
5173 return skb_tx_hash(dev, skb);
5174 }
5175
5176 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
5177 struct net_device *netdev)
5178 {
5179 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5180 struct ixgbe_ring *tx_ring;
5181 unsigned int first;
5182 unsigned int tx_flags = 0;
5183 u8 hdr_len = 0;
5184 int r_idx = 0, tso;
5185 int count = 0;
5186 unsigned int f;
5187
5188 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
5189 tx_flags |= vlan_tx_tag_get(skb);
5190 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5191 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
5192 tx_flags |= (skb->queue_mapping << 13);
5193 }
5194 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5195 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5196 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5197 if (skb->priority != TC_PRIO_CONTROL) {
5198 tx_flags |= (skb->queue_mapping << 13);
5199 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5200 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5201 } else {
5202 skb->queue_mapping =
5203 adapter->ring_feature[RING_F_DCB].indices-1;
5204 }
5205 }
5206
5207 r_idx = skb->queue_mapping;
5208 tx_ring = &adapter->tx_ring[r_idx];
5209
5210 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5211 (skb->protocol == htons(ETH_P_FCOE))) {
5212 tx_flags |= IXGBE_TX_FLAGS_FCOE;
5213 #ifdef IXGBE_FCOE
5214 r_idx = smp_processor_id();
5215 r_idx &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
5216 r_idx += adapter->ring_feature[RING_F_FCOE].mask;
5217 tx_ring = &adapter->tx_ring[r_idx];
5218 #endif
5219 }
5220 /* four things can cause us to need a context descriptor */
5221 if (skb_is_gso(skb) ||
5222 (skb->ip_summed == CHECKSUM_PARTIAL) ||
5223 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
5224 (tx_flags & IXGBE_TX_FLAGS_FCOE))
5225 count++;
5226
5227 count += TXD_USE_COUNT(skb_headlen(skb));
5228 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5229 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5230
5231 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
5232 adapter->tx_busy++;
5233 return NETDEV_TX_BUSY;
5234 }
5235
5236 first = tx_ring->next_to_use;
5237 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5238 #ifdef IXGBE_FCOE
5239 /* setup tx offload for FCoE */
5240 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5241 if (tso < 0) {
5242 dev_kfree_skb_any(skb);
5243 return NETDEV_TX_OK;
5244 }
5245 if (tso)
5246 tx_flags |= IXGBE_TX_FLAGS_FSO;
5247 #endif /* IXGBE_FCOE */
5248 } else {
5249 if (skb->protocol == htons(ETH_P_IP))
5250 tx_flags |= IXGBE_TX_FLAGS_IPV4;
5251 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5252 if (tso < 0) {
5253 dev_kfree_skb_any(skb);
5254 return NETDEV_TX_OK;
5255 }
5256
5257 if (tso)
5258 tx_flags |= IXGBE_TX_FLAGS_TSO;
5259 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
5260 (skb->ip_summed == CHECKSUM_PARTIAL))
5261 tx_flags |= IXGBE_TX_FLAGS_CSUM;
5262 }
5263
5264 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
5265 if (count) {
5266 /* add the ATR filter if ATR is on */
5267 if (tx_ring->atr_sample_rate) {
5268 ++tx_ring->atr_count;
5269 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
5270 test_bit(__IXGBE_FDIR_INIT_DONE,
5271 &tx_ring->reinit_state)) {
5272 ixgbe_atr(adapter, skb, tx_ring->queue_index,
5273 tx_flags);
5274 tx_ring->atr_count = 0;
5275 }
5276 }
5277 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
5278 hdr_len);
5279 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
5280
5281 } else {
5282 dev_kfree_skb_any(skb);
5283 tx_ring->tx_buffer_info[first].time_stamp = 0;
5284 tx_ring->next_to_use = first;
5285 }
5286
5287 return NETDEV_TX_OK;
5288 }
5289
5290 /**
5291 * ixgbe_get_stats - Get System Network Statistics
5292 * @netdev: network interface device structure
5293 *
5294 * Returns the address of the device statistics structure.
5295 * The statistics are actually updated from the timer callback.
5296 **/
5297 static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
5298 {
5299 /* only return the current stats */
5300 return &netdev->stats;
5301 }
5302
5303 /**
5304 * ixgbe_set_mac - Change the Ethernet Address of the NIC
5305 * @netdev: network interface device structure
5306 * @p: pointer to an address structure
5307 *
5308 * Returns 0 on success, negative on failure
5309 **/
5310 static int ixgbe_set_mac(struct net_device *netdev, void *p)
5311 {
5312 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5313 struct ixgbe_hw *hw = &adapter->hw;
5314 struct sockaddr *addr = p;
5315
5316 if (!is_valid_ether_addr(addr->sa_data))
5317 return -EADDRNOTAVAIL;
5318
5319 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
5320 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5321
5322 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
5323
5324 return 0;
5325 }
5326
5327 static int
5328 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
5329 {
5330 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5331 struct ixgbe_hw *hw = &adapter->hw;
5332 u16 value;
5333 int rc;
5334
5335 if (prtad != hw->phy.mdio.prtad)
5336 return -EINVAL;
5337 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
5338 if (!rc)
5339 rc = value;
5340 return rc;
5341 }
5342
5343 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
5344 u16 addr, u16 value)
5345 {
5346 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5347 struct ixgbe_hw *hw = &adapter->hw;
5348
5349 if (prtad != hw->phy.mdio.prtad)
5350 return -EINVAL;
5351 return hw->phy.ops.write_reg(hw, addr, devad, value);
5352 }
5353
5354 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
5355 {
5356 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5357
5358 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
5359 }
5360
5361 /**
5362 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
5363 * netdev->dev_addrs
5364 * @netdev: network interface device structure
5365 *
5366 * Returns non-zero on failure
5367 **/
5368 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
5369 {
5370 int err = 0;
5371 struct ixgbe_adapter *adapter = netdev_priv(dev);
5372 struct ixgbe_mac_info *mac = &adapter->hw.mac;
5373
5374 if (is_valid_ether_addr(mac->san_addr)) {
5375 rtnl_lock();
5376 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5377 rtnl_unlock();
5378 }
5379 return err;
5380 }
5381
5382 /**
5383 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
5384 * netdev->dev_addrs
5385 * @netdev: network interface device structure
5386 *
5387 * Returns non-zero on failure
5388 **/
5389 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
5390 {
5391 int err = 0;
5392 struct ixgbe_adapter *adapter = netdev_priv(dev);
5393 struct ixgbe_mac_info *mac = &adapter->hw.mac;
5394
5395 if (is_valid_ether_addr(mac->san_addr)) {
5396 rtnl_lock();
5397 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5398 rtnl_unlock();
5399 }
5400 return err;
5401 }
5402
5403 #ifdef CONFIG_NET_POLL_CONTROLLER
5404 /*
5405 * Polling 'interrupt' - used by things like netconsole to send skbs
5406 * without having to re-enable interrupts. It's not called while
5407 * the interrupt routine is executing.
5408 */
5409 static void ixgbe_netpoll(struct net_device *netdev)
5410 {
5411 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5412 int i;
5413
5414 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
5415 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5416 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5417 for (i = 0; i < num_q_vectors; i++) {
5418 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
5419 ixgbe_msix_clean_many(0, q_vector);
5420 }
5421 } else {
5422 ixgbe_intr(adapter->pdev->irq, netdev);
5423 }
5424 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
5425 }
5426 #endif
5427
5428 static const struct net_device_ops ixgbe_netdev_ops = {
5429 .ndo_open = ixgbe_open,
5430 .ndo_stop = ixgbe_close,
5431 .ndo_start_xmit = ixgbe_xmit_frame,
5432 .ndo_select_queue = ixgbe_select_queue,
5433 .ndo_get_stats = ixgbe_get_stats,
5434 .ndo_set_rx_mode = ixgbe_set_rx_mode,
5435 .ndo_set_multicast_list = ixgbe_set_rx_mode,
5436 .ndo_validate_addr = eth_validate_addr,
5437 .ndo_set_mac_address = ixgbe_set_mac,
5438 .ndo_change_mtu = ixgbe_change_mtu,
5439 .ndo_tx_timeout = ixgbe_tx_timeout,
5440 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
5441 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
5442 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
5443 .ndo_do_ioctl = ixgbe_ioctl,
5444 #ifdef CONFIG_NET_POLL_CONTROLLER
5445 .ndo_poll_controller = ixgbe_netpoll,
5446 #endif
5447 #ifdef IXGBE_FCOE
5448 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
5449 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
5450 .ndo_fcoe_enable = ixgbe_fcoe_enable,
5451 .ndo_fcoe_disable = ixgbe_fcoe_disable,
5452 #endif /* IXGBE_FCOE */
5453 };
5454
5455 /**
5456 * ixgbe_probe - Device Initialization Routine
5457 * @pdev: PCI device information struct
5458 * @ent: entry in ixgbe_pci_tbl
5459 *
5460 * Returns 0 on success, negative on failure
5461 *
5462 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
5463 * The OS initialization, configuring of the adapter private structure,
5464 * and a hardware reset occur.
5465 **/
5466 static int __devinit ixgbe_probe(struct pci_dev *pdev,
5467 const struct pci_device_id *ent)
5468 {
5469 struct net_device *netdev;
5470 struct ixgbe_adapter *adapter = NULL;
5471 struct ixgbe_hw *hw;
5472 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
5473 static int cards_found;
5474 int i, err, pci_using_dac;
5475 #ifdef IXGBE_FCOE
5476 u16 device_caps;
5477 #endif
5478 u32 part_num, eec;
5479
5480 err = pci_enable_device_mem(pdev);
5481 if (err)
5482 return err;
5483
5484 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
5485 !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
5486 pci_using_dac = 1;
5487 } else {
5488 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5489 if (err) {
5490 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
5491 if (err) {
5492 dev_err(&pdev->dev, "No usable DMA "
5493 "configuration, aborting\n");
5494 goto err_dma;
5495 }
5496 }
5497 pci_using_dac = 0;
5498 }
5499
5500 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
5501 IORESOURCE_MEM), ixgbe_driver_name);
5502 if (err) {
5503 dev_err(&pdev->dev,
5504 "pci_request_selected_regions failed 0x%x\n", err);
5505 goto err_pci_reg;
5506 }
5507
5508 pci_enable_pcie_error_reporting(pdev);
5509
5510 pci_set_master(pdev);
5511 pci_save_state(pdev);
5512
5513 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
5514 if (!netdev) {
5515 err = -ENOMEM;
5516 goto err_alloc_etherdev;
5517 }
5518
5519 SET_NETDEV_DEV(netdev, &pdev->dev);
5520
5521 pci_set_drvdata(pdev, netdev);
5522 adapter = netdev_priv(netdev);
5523
5524 adapter->netdev = netdev;
5525 adapter->pdev = pdev;
5526 hw = &adapter->hw;
5527 hw->back = adapter;
5528 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
5529
5530 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
5531 pci_resource_len(pdev, 0));
5532 if (!hw->hw_addr) {
5533 err = -EIO;
5534 goto err_ioremap;
5535 }
5536
5537 for (i = 1; i <= 5; i++) {
5538 if (pci_resource_len(pdev, i) == 0)
5539 continue;
5540 }
5541
5542 netdev->netdev_ops = &ixgbe_netdev_ops;
5543 ixgbe_set_ethtool_ops(netdev);
5544 netdev->watchdog_timeo = 5 * HZ;
5545 strcpy(netdev->name, pci_name(pdev));
5546
5547 adapter->bd_number = cards_found;
5548
5549 /* Setup hw api */
5550 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
5551 hw->mac.type = ii->mac;
5552
5553 /* EEPROM */
5554 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
5555 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
5556 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
5557 if (!(eec & (1 << 8)))
5558 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
5559
5560 /* PHY */
5561 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
5562 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
5563 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
5564 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
5565 hw->phy.mdio.mmds = 0;
5566 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
5567 hw->phy.mdio.dev = netdev;
5568 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
5569 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
5570
5571 /* set up this timer and work struct before calling get_invariants
5572 * which might start the timer
5573 */
5574 init_timer(&adapter->sfp_timer);
5575 adapter->sfp_timer.function = &ixgbe_sfp_timer;
5576 adapter->sfp_timer.data = (unsigned long) adapter;
5577
5578 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
5579
5580 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
5581 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
5582
5583 /* a new SFP+ module arrival, called from GPI SDP2 context */
5584 INIT_WORK(&adapter->sfp_config_module_task,
5585 ixgbe_sfp_config_module_task);
5586
5587 ii->get_invariants(hw);
5588
5589 /* setup the private structure */
5590 err = ixgbe_sw_init(adapter);
5591 if (err)
5592 goto err_sw_init;
5593
5594 /*
5595 * If there is a fan on this device and it has failed log the
5596 * failure.
5597 */
5598 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5599 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5600 if (esdp & IXGBE_ESDP_SDP1)
5601 DPRINTK(PROBE, CRIT,
5602 "Fan has stopped, replace the adapter\n");
5603 }
5604
5605 /* reset_hw fills in the perm_addr as well */
5606 err = hw->mac.ops.reset_hw(hw);
5607 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
5608 hw->mac.type == ixgbe_mac_82598EB) {
5609 /*
5610 * Start a kernel thread to watch for a module to arrive.
5611 * Only do this for 82598, since 82599 will generate
5612 * interrupts on module arrival.
5613 */
5614 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5615 mod_timer(&adapter->sfp_timer,
5616 round_jiffies(jiffies + (2 * HZ)));
5617 err = 0;
5618 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5619 dev_err(&adapter->pdev->dev, "failed to initialize because "
5620 "an unsupported SFP+ module type was detected.\n"
5621 "Reload the driver after installing a supported "
5622 "module.\n");
5623 goto err_sw_init;
5624 } else if (err) {
5625 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
5626 goto err_sw_init;
5627 }
5628
5629 netdev->features = NETIF_F_SG |
5630 NETIF_F_IP_CSUM |
5631 NETIF_F_HW_VLAN_TX |
5632 NETIF_F_HW_VLAN_RX |
5633 NETIF_F_HW_VLAN_FILTER;
5634
5635 netdev->features |= NETIF_F_IPV6_CSUM;
5636 netdev->features |= NETIF_F_TSO;
5637 netdev->features |= NETIF_F_TSO6;
5638 netdev->features |= NETIF_F_GRO;
5639
5640 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
5641 netdev->features |= NETIF_F_SCTP_CSUM;
5642
5643 netdev->vlan_features |= NETIF_F_TSO;
5644 netdev->vlan_features |= NETIF_F_TSO6;
5645 netdev->vlan_features |= NETIF_F_IP_CSUM;
5646 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
5647 netdev->vlan_features |= NETIF_F_SG;
5648
5649 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5650 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
5651
5652 #ifdef CONFIG_IXGBE_DCB
5653 netdev->dcbnl_ops = &dcbnl_ops;
5654 #endif
5655
5656 #ifdef IXGBE_FCOE
5657 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
5658 if (hw->mac.ops.get_device_caps) {
5659 hw->mac.ops.get_device_caps(hw, &device_caps);
5660 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
5661 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5662 }
5663 }
5664 #endif /* IXGBE_FCOE */
5665 if (pci_using_dac)
5666 netdev->features |= NETIF_F_HIGHDMA;
5667
5668 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
5669 netdev->features |= NETIF_F_LRO;
5670
5671 /* make sure the EEPROM is good */
5672 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
5673 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
5674 err = -EIO;
5675 goto err_eeprom;
5676 }
5677
5678 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
5679 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
5680
5681 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
5682 dev_err(&pdev->dev, "invalid MAC address\n");
5683 err = -EIO;
5684 goto err_eeprom;
5685 }
5686
5687 init_timer(&adapter->watchdog_timer);
5688 adapter->watchdog_timer.function = &ixgbe_watchdog;
5689 adapter->watchdog_timer.data = (unsigned long)adapter;
5690
5691 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
5692 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
5693
5694 err = ixgbe_init_interrupt_scheme(adapter);
5695 if (err)
5696 goto err_sw_init;
5697
5698 switch (pdev->device) {
5699 case IXGBE_DEV_ID_82599_KX4:
5700 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
5701 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
5702 /* Enable ACPI wakeup in GRC */
5703 IXGBE_WRITE_REG(hw, IXGBE_GRC,
5704 (IXGBE_READ_REG(hw, IXGBE_GRC) & ~IXGBE_GRC_APME));
5705 break;
5706 default:
5707 adapter->wol = 0;
5708 break;
5709 }
5710 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5711
5712 /* pick up the PCI bus settings for reporting later */
5713 hw->mac.ops.get_bus_info(hw);
5714
5715 /* print bus type/speed/width info */
5716 dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
5717 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
5718 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
5719 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
5720 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
5721 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
5722 "Unknown"),
5723 netdev->dev_addr);
5724 ixgbe_read_pba_num_generic(hw, &part_num);
5725 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
5726 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
5727 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
5728 (part_num >> 8), (part_num & 0xff));
5729 else
5730 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5731 hw->mac.type, hw->phy.type,
5732 (part_num >> 8), (part_num & 0xff));
5733
5734 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
5735 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
5736 "this card is not sufficient for optimal "
5737 "performance.\n");
5738 dev_warn(&pdev->dev, "For optimal performance a x8 "
5739 "PCI-Express slot is required.\n");
5740 }
5741
5742 /* save off EEPROM version number */
5743 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
5744
5745 /* reset the hardware with the new settings */
5746 err = hw->mac.ops.start_hw(hw);
5747
5748 if (err == IXGBE_ERR_EEPROM_VERSION) {
5749 /* We are running on a pre-production device, log a warning */
5750 dev_warn(&pdev->dev, "This device is a pre-production "
5751 "adapter/LOM. Please be aware there may be issues "
5752 "associated with your hardware. If you are "
5753 "experiencing problems please contact your Intel or "
5754 "hardware representative who provided you with this "
5755 "hardware.\n");
5756 }
5757 strcpy(netdev->name, "eth%d");
5758 err = register_netdev(netdev);
5759 if (err)
5760 goto err_register;
5761
5762 /* carrier off reporting is important to ethtool even BEFORE open */
5763 netif_carrier_off(netdev);
5764
5765 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5766 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5767 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
5768
5769 #ifdef CONFIG_IXGBE_DCA
5770 if (dca_add_requester(&pdev->dev) == 0) {
5771 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
5772 ixgbe_setup_dca(adapter);
5773 }
5774 #endif
5775 /* add san mac addr to netdev */
5776 ixgbe_add_sanmac_netdev(netdev);
5777
5778 dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
5779 cards_found++;
5780 return 0;
5781
5782 err_register:
5783 ixgbe_release_hw_control(adapter);
5784 ixgbe_clear_interrupt_scheme(adapter);
5785 err_sw_init:
5786 err_eeprom:
5787 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5788 del_timer_sync(&adapter->sfp_timer);
5789 cancel_work_sync(&adapter->sfp_task);
5790 cancel_work_sync(&adapter->multispeed_fiber_task);
5791 cancel_work_sync(&adapter->sfp_config_module_task);
5792 iounmap(hw->hw_addr);
5793 err_ioremap:
5794 free_netdev(netdev);
5795 err_alloc_etherdev:
5796 pci_release_selected_regions(pdev, pci_select_bars(pdev,
5797 IORESOURCE_MEM));
5798 err_pci_reg:
5799 err_dma:
5800 pci_disable_device(pdev);
5801 return err;
5802 }
5803
5804 /**
5805 * ixgbe_remove - Device Removal Routine
5806 * @pdev: PCI device information struct
5807 *
5808 * ixgbe_remove is called by the PCI subsystem to alert the driver
5809 * that it should release a PCI device. The could be caused by a
5810 * Hot-Plug event, or because the driver is going to be removed from
5811 * memory.
5812 **/
5813 static void __devexit ixgbe_remove(struct pci_dev *pdev)
5814 {
5815 struct net_device *netdev = pci_get_drvdata(pdev);
5816 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5817
5818 set_bit(__IXGBE_DOWN, &adapter->state);
5819 /* clear the module not found bit to make sure the worker won't
5820 * reschedule
5821 */
5822 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5823 del_timer_sync(&adapter->watchdog_timer);
5824
5825 del_timer_sync(&adapter->sfp_timer);
5826 cancel_work_sync(&adapter->watchdog_task);
5827 cancel_work_sync(&adapter->sfp_task);
5828 cancel_work_sync(&adapter->multispeed_fiber_task);
5829 cancel_work_sync(&adapter->sfp_config_module_task);
5830 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5831 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5832 cancel_work_sync(&adapter->fdir_reinit_task);
5833 flush_scheduled_work();
5834
5835 #ifdef CONFIG_IXGBE_DCA
5836 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
5837 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
5838 dca_remove_requester(&pdev->dev);
5839 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
5840 }
5841
5842 #endif
5843 #ifdef IXGBE_FCOE
5844 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
5845 ixgbe_cleanup_fcoe(adapter);
5846
5847 #endif /* IXGBE_FCOE */
5848
5849 /* remove the added san mac */
5850 ixgbe_del_sanmac_netdev(netdev);
5851
5852 if (netdev->reg_state == NETREG_REGISTERED)
5853 unregister_netdev(netdev);
5854
5855 ixgbe_clear_interrupt_scheme(adapter);
5856
5857 ixgbe_release_hw_control(adapter);
5858
5859 iounmap(adapter->hw.hw_addr);
5860 pci_release_selected_regions(pdev, pci_select_bars(pdev,
5861 IORESOURCE_MEM));
5862
5863 DPRINTK(PROBE, INFO, "complete\n");
5864
5865 free_netdev(netdev);
5866
5867 pci_disable_pcie_error_reporting(pdev);
5868
5869 pci_disable_device(pdev);
5870 }
5871
5872 /**
5873 * ixgbe_io_error_detected - called when PCI error is detected
5874 * @pdev: Pointer to PCI device
5875 * @state: The current pci connection state
5876 *
5877 * This function is called after a PCI bus error affecting
5878 * this device has been detected.
5879 */
5880 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
5881 pci_channel_state_t state)
5882 {
5883 struct net_device *netdev = pci_get_drvdata(pdev);
5884 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5885
5886 netif_device_detach(netdev);
5887
5888 if (state == pci_channel_io_perm_failure)
5889 return PCI_ERS_RESULT_DISCONNECT;
5890
5891 if (netif_running(netdev))
5892 ixgbe_down(adapter);
5893 pci_disable_device(pdev);
5894
5895 /* Request a slot reset. */
5896 return PCI_ERS_RESULT_NEED_RESET;
5897 }
5898
5899 /**
5900 * ixgbe_io_slot_reset - called after the pci bus has been reset.
5901 * @pdev: Pointer to PCI device
5902 *
5903 * Restart the card from scratch, as if from a cold-boot.
5904 */
5905 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
5906 {
5907 struct net_device *netdev = pci_get_drvdata(pdev);
5908 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5909 pci_ers_result_t result;
5910 int err;
5911
5912 if (pci_enable_device_mem(pdev)) {
5913 DPRINTK(PROBE, ERR,
5914 "Cannot re-enable PCI device after reset.\n");
5915 result = PCI_ERS_RESULT_DISCONNECT;
5916 } else {
5917 pci_set_master(pdev);
5918 pci_restore_state(pdev);
5919
5920 pci_wake_from_d3(pdev, false);
5921
5922 ixgbe_reset(adapter);
5923 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5924 result = PCI_ERS_RESULT_RECOVERED;
5925 }
5926
5927 err = pci_cleanup_aer_uncorrect_error_status(pdev);
5928 if (err) {
5929 dev_err(&pdev->dev,
5930 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
5931 /* non-fatal, continue */
5932 }
5933
5934 return result;
5935 }
5936
5937 /**
5938 * ixgbe_io_resume - called when traffic can start flowing again.
5939 * @pdev: Pointer to PCI device
5940 *
5941 * This callback is called when the error recovery driver tells us that
5942 * its OK to resume normal operation.
5943 */
5944 static void ixgbe_io_resume(struct pci_dev *pdev)
5945 {
5946 struct net_device *netdev = pci_get_drvdata(pdev);
5947 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5948
5949 if (netif_running(netdev)) {
5950 if (ixgbe_up(adapter)) {
5951 DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
5952 return;
5953 }
5954 }
5955
5956 netif_device_attach(netdev);
5957 }
5958
5959 static struct pci_error_handlers ixgbe_err_handler = {
5960 .error_detected = ixgbe_io_error_detected,
5961 .slot_reset = ixgbe_io_slot_reset,
5962 .resume = ixgbe_io_resume,
5963 };
5964
5965 static struct pci_driver ixgbe_driver = {
5966 .name = ixgbe_driver_name,
5967 .id_table = ixgbe_pci_tbl,
5968 .probe = ixgbe_probe,
5969 .remove = __devexit_p(ixgbe_remove),
5970 #ifdef CONFIG_PM
5971 .suspend = ixgbe_suspend,
5972 .resume = ixgbe_resume,
5973 #endif
5974 .shutdown = ixgbe_shutdown,
5975 .err_handler = &ixgbe_err_handler
5976 };
5977
5978 /**
5979 * ixgbe_init_module - Driver Registration Routine
5980 *
5981 * ixgbe_init_module is the first routine called when the driver is
5982 * loaded. All it does is register with the PCI subsystem.
5983 **/
5984 static int __init ixgbe_init_module(void)
5985 {
5986 int ret;
5987 printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
5988 ixgbe_driver_string, ixgbe_driver_version);
5989
5990 printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
5991
5992 #ifdef CONFIG_IXGBE_DCA
5993 dca_register_notify(&dca_notifier);
5994 #endif
5995
5996 ret = pci_register_driver(&ixgbe_driver);
5997 return ret;
5998 }
5999
6000 module_init(ixgbe_init_module);
6001
6002 /**
6003 * ixgbe_exit_module - Driver Exit Cleanup Routine
6004 *
6005 * ixgbe_exit_module is called just before the driver is removed
6006 * from memory.
6007 **/
6008 static void __exit ixgbe_exit_module(void)
6009 {
6010 #ifdef CONFIG_IXGBE_DCA
6011 dca_unregister_notify(&dca_notifier);
6012 #endif
6013 pci_unregister_driver(&ixgbe_driver);
6014 }
6015
6016 #ifdef CONFIG_IXGBE_DCA
6017 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
6018 void *p)
6019 {
6020 int ret_val;
6021
6022 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
6023 __ixgbe_notify_dca);
6024
6025 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6026 }
6027
6028 #endif /* CONFIG_IXGBE_DCA */
6029 #ifdef DEBUG
6030 /**
6031 * ixgbe_get_hw_dev_name - return device name string
6032 * used by hardware layer to print debugging information
6033 **/
6034 char *ixgbe_get_hw_dev_name(struct ixgbe_hw *hw)
6035 {
6036 struct ixgbe_adapter *adapter = hw->back;
6037 return adapter->netdev->name;
6038 }
6039
6040 #endif
6041 module_exit(ixgbe_exit_module);
6042
6043 /* ixgbe_main.c */
This page took 0.153289 seconds and 6 git commands to generate.