projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit...
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / net / ixgbe / ixgbe_main.c
1 /*******************************************************************************
2
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2010 Intel Corporation.
5
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
9
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 more details.
14
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
21
22 Contact Information:
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
34 #include <linux/in.h>
35 #include <linux/ip.h>
36 #include <linux/tcp.h>
37 #include <linux/pkt_sched.h>
38 #include <linux/ipv6.h>
39 #include <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <scsi/fc/fc_fcoe.h>
45
46 #include "ixgbe.h"
47 #include "ixgbe_common.h"
48 #include "ixgbe_dcb_82599.h"
49 #include "ixgbe_sriov.h"
50
51 char ixgbe_driver_name[] = "ixgbe";
52 static const char ixgbe_driver_string[] =
53 "Intel(R) 10 Gigabit PCI Express Network Driver";
54
55 #define DRV_VERSION "2.0.62-k2"
56 const char ixgbe_driver_version[] = DRV_VERSION;
57 static char ixgbe_copyright[] = "Copyright (c) 1999-2010 Intel Corporation.";
58
59 static const struct ixgbe_info *ixgbe_info_tbl[] = {
60 [board_82598] = &ixgbe_82598_info,
61 [board_82599] = &ixgbe_82599_info,
62 };
63
64 /* ixgbe_pci_tbl - PCI Device ID Table
65 *
66 * Wildcard entries (PCI_ANY_ID) should come last
67 * Last entry must be all 0s
68 *
69 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
70 * Class, Class Mask, private data (not used) }
71 */
72 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
73 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
74 board_82598 },
75 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
76 board_82598 },
77 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
78 board_82598 },
79 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
80 board_82598 },
81 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
82 board_82598 },
83 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
84 board_82598 },
85 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
86 board_82598 },
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
88 board_82598 },
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
90 board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
92 board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
94 board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
96 board_82598 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
98 board_82599 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
100 board_82599 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
102 board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
104 board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
106 board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
108 board_82599 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
110 board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
112 board_82599 },
113
114 /* required last entry */
115 {0, }
116 };
117 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
118
119 #ifdef CONFIG_IXGBE_DCA
120 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
121 void *p);
122 static struct notifier_block dca_notifier = {
123 .notifier_call = ixgbe_notify_dca,
124 .next = NULL,
125 .priority = 0
126 };
127 #endif
128
129 #ifdef CONFIG_PCI_IOV
130 static unsigned int max_vfs;
131 module_param(max_vfs, uint, 0);
132 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
133 "per physical function");
134 #endif /* CONFIG_PCI_IOV */
135
136 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
137 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
138 MODULE_LICENSE("GPL");
139 MODULE_VERSION(DRV_VERSION);
140
141 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
142
143 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
144 {
145 struct ixgbe_hw *hw = &adapter->hw;
146 u32 gcr;
147 u32 gpie;
148 u32 vmdctl;
149
150 #ifdef CONFIG_PCI_IOV
151 /* disable iov and allow time for transactions to clear */
152 pci_disable_sriov(adapter->pdev);
153 #endif
154
155 /* turn off device IOV mode */
156 gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
157 gcr &= ~(IXGBE_GCR_EXT_SRIOV);
158 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
159 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
160 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
161 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
162
163 /* set default pool back to 0 */
164 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
165 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
166 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
167
168 /* take a breather then clean up driver data */
169 msleep(100);
170 if (adapter->vfinfo)
171 kfree(adapter->vfinfo);
172 adapter->vfinfo = NULL;
173
174 adapter->num_vfs = 0;
175 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
176 }
177
178 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
179 {
180 u32 ctrl_ext;
181
182 /* Let firmware take over control of h/w */
183 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
184 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
185 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
186 }
187
188 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
189 {
190 u32 ctrl_ext;
191
192 /* Let firmware know the driver has taken over */
193 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
194 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
195 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
196 }
197
198 /*
199 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
200 * @adapter: pointer to adapter struct
201 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
202 * @queue: queue to map the corresponding interrupt to
203 * @msix_vector: the vector to map to the corresponding queue
204 *
205 */
206 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
207 u8 queue, u8 msix_vector)
208 {
209 u32 ivar, index;
210 struct ixgbe_hw *hw = &adapter->hw;
211 switch (hw->mac.type) {
212 case ixgbe_mac_82598EB:
213 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
214 if (direction == -1)
215 direction = 0;
216 index = (((direction * 64) + queue) >> 2) & 0x1F;
217 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
218 ivar &= ~(0xFF << (8 * (queue & 0x3)));
219 ivar |= (msix_vector << (8 * (queue & 0x3)));
220 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
221 break;
222 case ixgbe_mac_82599EB:
223 if (direction == -1) {
224 /* other causes */
225 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
226 index = ((queue & 1) * 8);
227 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
228 ivar &= ~(0xFF << index);
229 ivar |= (msix_vector << index);
230 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
231 break;
232 } else {
233 /* tx or rx causes */
234 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
235 index = ((16 * (queue & 1)) + (8 * direction));
236 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
237 ivar &= ~(0xFF << index);
238 ivar |= (msix_vector << index);
239 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
240 break;
241 }
242 default:
243 break;
244 }
245 }
246
247 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
248 u64 qmask)
249 {
250 u32 mask;
251
252 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
253 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
254 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
255 } else {
256 mask = (qmask & 0xFFFFFFFF);
257 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
258 mask = (qmask >> 32);
259 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
260 }
261 }
262
263 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
264 struct ixgbe_tx_buffer
265 *tx_buffer_info)
266 {
267 if (tx_buffer_info->dma) {
268 if (tx_buffer_info->mapped_as_page)
269 pci_unmap_page(adapter->pdev,
270 tx_buffer_info->dma,
271 tx_buffer_info->length,
272 PCI_DMA_TODEVICE);
273 else
274 pci_unmap_single(adapter->pdev,
275 tx_buffer_info->dma,
276 tx_buffer_info->length,
277 PCI_DMA_TODEVICE);
278 tx_buffer_info->dma = 0;
279 }
280 if (tx_buffer_info->skb) {
281 dev_kfree_skb_any(tx_buffer_info->skb);
282 tx_buffer_info->skb = NULL;
283 }
284 tx_buffer_info->time_stamp = 0;
285 /* tx_buffer_info must be completely set up in the transmit path */
286 }
287
288 /**
289 * ixgbe_tx_is_paused - check if the tx ring is paused
290 * @adapter: the ixgbe adapter
291 * @tx_ring: the corresponding tx_ring
292 *
293 * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
294 * corresponding TC of this tx_ring when checking TFCS.
295 *
296 * Returns : true if paused
297 */
298 static inline bool ixgbe_tx_is_paused(struct ixgbe_adapter *adapter,
299 struct ixgbe_ring *tx_ring)
300 {
301 u32 txoff = IXGBE_TFCS_TXOFF;
302
303 #ifdef CONFIG_IXGBE_DCB
304 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
305 int tc;
306 int reg_idx = tx_ring->reg_idx;
307 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
308
309 switch (adapter->hw.mac.type) {
310 case ixgbe_mac_82598EB:
311 tc = reg_idx >> 2;
312 txoff = IXGBE_TFCS_TXOFF0;
313 break;
314 case ixgbe_mac_82599EB:
315 tc = 0;
316 txoff = IXGBE_TFCS_TXOFF;
317 if (dcb_i == 8) {
318 /* TC0, TC1 */
319 tc = reg_idx >> 5;
320 if (tc == 2) /* TC2, TC3 */
321 tc += (reg_idx - 64) >> 4;
322 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
323 tc += 1 + ((reg_idx - 96) >> 3);
324 } else if (dcb_i == 4) {
325 /* TC0, TC1 */
326 tc = reg_idx >> 6;
327 if (tc == 1) {
328 tc += (reg_idx - 64) >> 5;
329 if (tc == 2) /* TC2, TC3 */
330 tc += (reg_idx - 96) >> 4;
331 }
332 }
333 break;
334 default:
335 tc = 0;
336 }
337 txoff <<= tc;
338 }
339 #endif
340 return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
341 }
342
343 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
344 struct ixgbe_ring *tx_ring,
345 unsigned int eop)
346 {
347 struct ixgbe_hw *hw = &adapter->hw;
348
349 /* Detect a transmit hang in hardware, this serializes the
350 * check with the clearing of time_stamp and movement of eop */
351 adapter->detect_tx_hung = false;
352 if (tx_ring->tx_buffer_info[eop].time_stamp &&
353 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
354 !ixgbe_tx_is_paused(adapter, tx_ring)) {
355 /* detected Tx unit hang */
356 union ixgbe_adv_tx_desc *tx_desc;
357 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
358 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
359 " Tx Queue <%d>\n"
360 " TDH, TDT <%x>, <%x>\n"
361 " next_to_use <%x>\n"
362 " next_to_clean <%x>\n"
363 "tx_buffer_info[next_to_clean]\n"
364 " time_stamp <%lx>\n"
365 " jiffies <%lx>\n",
366 tx_ring->queue_index,
367 IXGBE_READ_REG(hw, tx_ring->head),
368 IXGBE_READ_REG(hw, tx_ring->tail),
369 tx_ring->next_to_use, eop,
370 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
371 return true;
372 }
373
374 return false;
375 }
376
377 #define IXGBE_MAX_TXD_PWR 14
378 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
379
380 /* Tx Descriptors needed, worst case */
381 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
382 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
383 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
384 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
385
386 static void ixgbe_tx_timeout(struct net_device *netdev);
387
388 /**
389 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
390 * @q_vector: structure containing interrupt and ring information
391 * @tx_ring: tx ring to clean
392 **/
393 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
394 struct ixgbe_ring *tx_ring)
395 {
396 struct ixgbe_adapter *adapter = q_vector->adapter;
397 struct net_device *netdev = adapter->netdev;
398 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
399 struct ixgbe_tx_buffer *tx_buffer_info;
400 unsigned int i, eop, count = 0;
401 unsigned int total_bytes = 0, total_packets = 0;
402
403 i = tx_ring->next_to_clean;
404 eop = tx_ring->tx_buffer_info[i].next_to_watch;
405 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
406
407 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
408 (count < tx_ring->work_limit)) {
409 bool cleaned = false;
410 for ( ; !cleaned; count++) {
411 struct sk_buff *skb;
412 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
413 tx_buffer_info = &tx_ring->tx_buffer_info[i];
414 cleaned = (i == eop);
415 skb = tx_buffer_info->skb;
416
417 if (cleaned && skb) {
418 unsigned int segs, bytecount;
419 unsigned int hlen = skb_headlen(skb);
420
421 /* gso_segs is currently only valid for tcp */
422 segs = skb_shinfo(skb)->gso_segs ?: 1;
423 #ifdef IXGBE_FCOE
424 /* adjust for FCoE Sequence Offload */
425 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
426 && (skb->protocol == htons(ETH_P_FCOE)) &&
427 skb_is_gso(skb)) {
428 hlen = skb_transport_offset(skb) +
429 sizeof(struct fc_frame_header) +
430 sizeof(struct fcoe_crc_eof);
431 segs = DIV_ROUND_UP(skb->len - hlen,
432 skb_shinfo(skb)->gso_size);
433 }
434 #endif /* IXGBE_FCOE */
435 /* multiply data chunks by size of headers */
436 bytecount = ((segs - 1) * hlen) + skb->len;
437 total_packets += segs;
438 total_bytes += bytecount;
439 }
440
441 ixgbe_unmap_and_free_tx_resource(adapter,
442 tx_buffer_info);
443
444 tx_desc->wb.status = 0;
445
446 i++;
447 if (i == tx_ring->count)
448 i = 0;
449 }
450
451 eop = tx_ring->tx_buffer_info[i].next_to_watch;
452 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
453 }
454
455 tx_ring->next_to_clean = i;
456
457 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
458 if (unlikely(count && netif_carrier_ok(netdev) &&
459 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
460 /* Make sure that anybody stopping the queue after this
461 * sees the new next_to_clean.
462 */
463 smp_mb();
464 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
465 !test_bit(__IXGBE_DOWN, &adapter->state)) {
466 netif_wake_subqueue(netdev, tx_ring->queue_index);
467 ++tx_ring->restart_queue;
468 }
469 }
470
471 if (adapter->detect_tx_hung) {
472 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
473 /* schedule immediate reset if we believe we hung */
474 DPRINTK(PROBE, INFO,
475 "tx hang %d detected, resetting adapter\n",
476 adapter->tx_timeout_count + 1);
477 ixgbe_tx_timeout(adapter->netdev);
478 }
479 }
480
481 /* re-arm the interrupt */
482 if (count >= tx_ring->work_limit)
483 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
484
485 tx_ring->total_bytes += total_bytes;
486 tx_ring->total_packets += total_packets;
487 tx_ring->stats.packets += total_packets;
488 tx_ring->stats.bytes += total_bytes;
489 return (count < tx_ring->work_limit);
490 }
491
492 #ifdef CONFIG_IXGBE_DCA
493 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
494 struct ixgbe_ring *rx_ring)
495 {
496 u32 rxctrl;
497 int cpu = get_cpu();
498 int q = rx_ring->reg_idx;
499
500 if (rx_ring->cpu != cpu) {
501 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
502 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
503 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
504 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
505 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
506 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
507 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
508 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
509 }
510 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
511 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
512 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
513 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
514 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
515 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
516 rx_ring->cpu = cpu;
517 }
518 put_cpu();
519 }
520
521 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
522 struct ixgbe_ring *tx_ring)
523 {
524 u32 txctrl;
525 int cpu = get_cpu();
526 int q = tx_ring->reg_idx;
527 struct ixgbe_hw *hw = &adapter->hw;
528
529 if (tx_ring->cpu != cpu) {
530 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
531 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(q));
532 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
533 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
534 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
535 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(q), txctrl);
536 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
537 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(q));
538 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
539 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
540 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
541 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
542 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(q), txctrl);
543 }
544 tx_ring->cpu = cpu;
545 }
546 put_cpu();
547 }
548
549 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
550 {
551 int i;
552
553 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
554 return;
555
556 /* always use CB2 mode, difference is masked in the CB driver */
557 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
558
559 for (i = 0; i < adapter->num_tx_queues; i++) {
560 adapter->tx_ring[i]->cpu = -1;
561 ixgbe_update_tx_dca(adapter, adapter->tx_ring[i]);
562 }
563 for (i = 0; i < adapter->num_rx_queues; i++) {
564 adapter->rx_ring[i]->cpu = -1;
565 ixgbe_update_rx_dca(adapter, adapter->rx_ring[i]);
566 }
567 }
568
569 static int __ixgbe_notify_dca(struct device *dev, void *data)
570 {
571 struct net_device *netdev = dev_get_drvdata(dev);
572 struct ixgbe_adapter *adapter = netdev_priv(netdev);
573 unsigned long event = *(unsigned long *)data;
574
575 switch (event) {
576 case DCA_PROVIDER_ADD:
577 /* if we're already enabled, don't do it again */
578 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
579 break;
580 if (dca_add_requester(dev) == 0) {
581 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
582 ixgbe_setup_dca(adapter);
583 break;
584 }
585 /* Fall Through since DCA is disabled. */
586 case DCA_PROVIDER_REMOVE:
587 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
588 dca_remove_requester(dev);
589 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
590 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
591 }
592 break;
593 }
594
595 return 0;
596 }
597
598 #endif /* CONFIG_IXGBE_DCA */
599 /**
600 * ixgbe_receive_skb - Send a completed packet up the stack
601 * @adapter: board private structure
602 * @skb: packet to send up
603 * @status: hardware indication of status of receive
604 * @rx_ring: rx descriptor ring (for a specific queue) to setup
605 * @rx_desc: rx descriptor
606 **/
607 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
608 struct sk_buff *skb, u8 status,
609 struct ixgbe_ring *ring,
610 union ixgbe_adv_rx_desc *rx_desc)
611 {
612 struct ixgbe_adapter *adapter = q_vector->adapter;
613 struct napi_struct *napi = &q_vector->napi;
614 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
615 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
616
617 skb_record_rx_queue(skb, ring->queue_index);
618 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
619 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
620 vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
621 else
622 napi_gro_receive(napi, skb);
623 } else {
624 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
625 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
626 else
627 netif_rx(skb);
628 }
629 }
630
631 /**
632 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
633 * @adapter: address of board private structure
634 * @status_err: hardware indication of status of receive
635 * @skb: skb currently being received and modified
636 **/
637 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
638 union ixgbe_adv_rx_desc *rx_desc,
639 struct sk_buff *skb)
640 {
641 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
642
643 skb->ip_summed = CHECKSUM_NONE;
644
645 /* Rx csum disabled */
646 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
647 return;
648
649 /* if IP and error */
650 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
651 (status_err & IXGBE_RXDADV_ERR_IPE)) {
652 adapter->hw_csum_rx_error++;
653 return;
654 }
655
656 if (!(status_err & IXGBE_RXD_STAT_L4CS))
657 return;
658
659 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
660 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
661
662 /*
663 * 82599 errata, UDP frames with a 0 checksum can be marked as
664 * checksum errors.
665 */
666 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
667 (adapter->hw.mac.type == ixgbe_mac_82599EB))
668 return;
669
670 adapter->hw_csum_rx_error++;
671 return;
672 }
673
674 /* It must be a TCP or UDP packet with a valid checksum */
675 skb->ip_summed = CHECKSUM_UNNECESSARY;
676 }
677
678 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
679 struct ixgbe_ring *rx_ring, u32 val)
680 {
681 /*
682 * Force memory writes to complete before letting h/w
683 * know there are new descriptors to fetch. (Only
684 * applicable for weak-ordered memory model archs,
685 * such as IA-64).
686 */
687 wmb();
688 IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
689 }
690
691 /**
692 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
693 * @adapter: address of board private structure
694 **/
695 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
696 struct ixgbe_ring *rx_ring,
697 int cleaned_count)
698 {
699 struct pci_dev *pdev = adapter->pdev;
700 union ixgbe_adv_rx_desc *rx_desc;
701 struct ixgbe_rx_buffer *bi;
702 unsigned int i;
703
704 i = rx_ring->next_to_use;
705 bi = &rx_ring->rx_buffer_info[i];
706
707 while (cleaned_count--) {
708 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
709
710 if (!bi->page_dma &&
711 (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
712 if (!bi->page) {
713 bi->page = alloc_page(GFP_ATOMIC);
714 if (!bi->page) {
715 adapter->alloc_rx_page_failed++;
716 goto no_buffers;
717 }
718 bi->page_offset = 0;
719 } else {
720 /* use a half page if we're re-using */
721 bi->page_offset ^= (PAGE_SIZE / 2);
722 }
723
724 bi->page_dma = pci_map_page(pdev, bi->page,
725 bi->page_offset,
726 (PAGE_SIZE / 2),
727 PCI_DMA_FROMDEVICE);
728 }
729
730 if (!bi->skb) {
731 struct sk_buff *skb;
732 /* netdev_alloc_skb reserves 32 bytes up front!! */
733 uint bufsz = rx_ring->rx_buf_len + SMP_CACHE_BYTES;
734 skb = netdev_alloc_skb(adapter->netdev, bufsz);
735
736 if (!skb) {
737 adapter->alloc_rx_buff_failed++;
738 goto no_buffers;
739 }
740
741 /* advance the data pointer to the next cache line */
742 skb_reserve(skb, (PTR_ALIGN(skb->data, SMP_CACHE_BYTES)
743 - skb->data));
744
745 bi->skb = skb;
746 bi->dma = pci_map_single(pdev, skb->data,
747 rx_ring->rx_buf_len,
748 PCI_DMA_FROMDEVICE);
749 }
750 /* Refresh the desc even if buffer_addrs didn't change because
751 * each write-back erases this info. */
752 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
753 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
754 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
755 } else {
756 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
757 }
758
759 i++;
760 if (i == rx_ring->count)
761 i = 0;
762 bi = &rx_ring->rx_buffer_info[i];
763 }
764
765 no_buffers:
766 if (rx_ring->next_to_use != i) {
767 rx_ring->next_to_use = i;
768 if (i-- == 0)
769 i = (rx_ring->count - 1);
770
771 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
772 }
773 }
774
775 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
776 {
777 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
778 }
779
780 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
781 {
782 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
783 }
784
785 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
786 {
787 return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
788 IXGBE_RXDADV_RSCCNT_MASK) >>
789 IXGBE_RXDADV_RSCCNT_SHIFT;
790 }
791
792 /**
793 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
794 * @skb: pointer to the last skb in the rsc queue
795 * @count: pointer to number of packets coalesced in this context
796 *
797 * This function changes a queue full of hw rsc buffers into a completed
798 * packet. It uses the ->prev pointers to find the first packet and then
799 * turns it into the frag list owner.
800 **/
801 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
802 u64 *count)
803 {
804 unsigned int frag_list_size = 0;
805
806 while (skb->prev) {
807 struct sk_buff *prev = skb->prev;
808 frag_list_size += skb->len;
809 skb->prev = NULL;
810 skb = prev;
811 *count += 1;
812 }
813
814 skb_shinfo(skb)->frag_list = skb->next;
815 skb->next = NULL;
816 skb->len += frag_list_size;
817 skb->data_len += frag_list_size;
818 skb->truesize += frag_list_size;
819 return skb;
820 }
821
822 struct ixgbe_rsc_cb {
823 dma_addr_t dma;
824 };
825
826 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
827
828 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
829 struct ixgbe_ring *rx_ring,
830 int *work_done, int work_to_do)
831 {
832 struct ixgbe_adapter *adapter = q_vector->adapter;
833 struct net_device *netdev = adapter->netdev;
834 struct pci_dev *pdev = adapter->pdev;
835 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
836 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
837 struct sk_buff *skb;
838 unsigned int i, rsc_count = 0;
839 u32 len, staterr;
840 u16 hdr_info;
841 bool cleaned = false;
842 int cleaned_count = 0;
843 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
844 #ifdef IXGBE_FCOE
845 int ddp_bytes = 0;
846 #endif /* IXGBE_FCOE */
847
848 i = rx_ring->next_to_clean;
849 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
850 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
851 rx_buffer_info = &rx_ring->rx_buffer_info[i];
852
853 while (staterr & IXGBE_RXD_STAT_DD) {
854 u32 upper_len = 0;
855 if (*work_done >= work_to_do)
856 break;
857 (*work_done)++;
858
859 rmb(); /* read descriptor and rx_buffer_info after status DD */
860 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
861 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
862 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
863 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
864 if (len > IXGBE_RX_HDR_SIZE)
865 len = IXGBE_RX_HDR_SIZE;
866 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
867 } else {
868 len = le16_to_cpu(rx_desc->wb.upper.length);
869 }
870
871 cleaned = true;
872 skb = rx_buffer_info->skb;
873 prefetch(skb->data);
874 rx_buffer_info->skb = NULL;
875
876 if (rx_buffer_info->dma) {
877 if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
878 (!(staterr & IXGBE_RXD_STAT_EOP)) &&
879 (!(skb->prev)))
880 /*
881 * When HWRSC is enabled, delay unmapping
882 * of the first packet. It carries the
883 * header information, HW may still
884 * access the header after the writeback.
885 * Only unmap it when EOP is reached
886 */
887 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
888 else
889 pci_unmap_single(pdev, rx_buffer_info->dma,
890 rx_ring->rx_buf_len,
891 PCI_DMA_FROMDEVICE);
892 rx_buffer_info->dma = 0;
893 skb_put(skb, len);
894 }
895
896 if (upper_len) {
897 pci_unmap_page(pdev, rx_buffer_info->page_dma,
898 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
899 rx_buffer_info->page_dma = 0;
900 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
901 rx_buffer_info->page,
902 rx_buffer_info->page_offset,
903 upper_len);
904
905 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
906 (page_count(rx_buffer_info->page) != 1))
907 rx_buffer_info->page = NULL;
908 else
909 get_page(rx_buffer_info->page);
910
911 skb->len += upper_len;
912 skb->data_len += upper_len;
913 skb->truesize += upper_len;
914 }
915
916 i++;
917 if (i == rx_ring->count)
918 i = 0;
919
920 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
921 prefetch(next_rxd);
922 cleaned_count++;
923
924 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
925 rsc_count = ixgbe_get_rsc_count(rx_desc);
926
927 if (rsc_count) {
928 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
929 IXGBE_RXDADV_NEXTP_SHIFT;
930 next_buffer = &rx_ring->rx_buffer_info[nextp];
931 } else {
932 next_buffer = &rx_ring->rx_buffer_info[i];
933 }
934
935 if (staterr & IXGBE_RXD_STAT_EOP) {
936 if (skb->prev)
937 skb = ixgbe_transform_rsc_queue(skb, &(rx_ring->rsc_count));
938 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
939 if (IXGBE_RSC_CB(skb)->dma) {
940 pci_unmap_single(pdev, IXGBE_RSC_CB(skb)->dma,
941 rx_ring->rx_buf_len,
942 PCI_DMA_FROMDEVICE);
943 IXGBE_RSC_CB(skb)->dma = 0;
944 }
945 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)
946 rx_ring->rsc_count += skb_shinfo(skb)->nr_frags;
947 else
948 rx_ring->rsc_count++;
949 rx_ring->rsc_flush++;
950 }
951 rx_ring->stats.packets++;
952 rx_ring->stats.bytes += skb->len;
953 } else {
954 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
955 rx_buffer_info->skb = next_buffer->skb;
956 rx_buffer_info->dma = next_buffer->dma;
957 next_buffer->skb = skb;
958 next_buffer->dma = 0;
959 } else {
960 skb->next = next_buffer->skb;
961 skb->next->prev = skb;
962 }
963 rx_ring->non_eop_descs++;
964 goto next_desc;
965 }
966
967 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
968 dev_kfree_skb_irq(skb);
969 goto next_desc;
970 }
971
972 ixgbe_rx_checksum(adapter, rx_desc, skb);
973
974 /* probably a little skewed due to removing CRC */
975 total_rx_bytes += skb->len;
976 total_rx_packets++;
977
978 skb->protocol = eth_type_trans(skb, adapter->netdev);
979 #ifdef IXGBE_FCOE
980 /* if ddp, not passing to ULD unless for FCP_RSP or error */
981 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
982 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
983 if (!ddp_bytes)
984 goto next_desc;
985 }
986 #endif /* IXGBE_FCOE */
987 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
988
989 next_desc:
990 rx_desc->wb.upper.status_error = 0;
991
992 /* return some buffers to hardware, one at a time is too slow */
993 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
994 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
995 cleaned_count = 0;
996 }
997
998 /* use prefetched values */
999 rx_desc = next_rxd;
1000 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1001
1002 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1003 }
1004
1005 rx_ring->next_to_clean = i;
1006 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1007
1008 if (cleaned_count)
1009 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1010
1011 #ifdef IXGBE_FCOE
1012 /* include DDPed FCoE data */
1013 if (ddp_bytes > 0) {
1014 unsigned int mss;
1015
1016 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
1017 sizeof(struct fc_frame_header) -
1018 sizeof(struct fcoe_crc_eof);
1019 if (mss > 512)
1020 mss &= ~511;
1021 total_rx_bytes += ddp_bytes;
1022 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1023 }
1024 #endif /* IXGBE_FCOE */
1025
1026 rx_ring->total_packets += total_rx_packets;
1027 rx_ring->total_bytes += total_rx_bytes;
1028 netdev->stats.rx_bytes += total_rx_bytes;
1029 netdev->stats.rx_packets += total_rx_packets;
1030
1031 return cleaned;
1032 }
1033
1034 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1035 /**
1036 * ixgbe_configure_msix - Configure MSI-X hardware
1037 * @adapter: board private structure
1038 *
1039 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1040 * interrupts.
1041 **/
1042 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1043 {
1044 struct ixgbe_q_vector *q_vector;
1045 int i, j, q_vectors, v_idx, r_idx;
1046 u32 mask;
1047
1048 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1049
1050 /*
1051 * Populate the IVAR table and set the ITR values to the
1052 * corresponding register.
1053 */
1054 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1055 q_vector = adapter->q_vector[v_idx];
1056 /* XXX for_each_set_bit(...) */
1057 r_idx = find_first_bit(q_vector->rxr_idx,
1058 adapter->num_rx_queues);
1059
1060 for (i = 0; i < q_vector->rxr_count; i++) {
1061 j = adapter->rx_ring[r_idx]->reg_idx;
1062 ixgbe_set_ivar(adapter, 0, j, v_idx);
1063 r_idx = find_next_bit(q_vector->rxr_idx,
1064 adapter->num_rx_queues,
1065 r_idx + 1);
1066 }
1067 r_idx = find_first_bit(q_vector->txr_idx,
1068 adapter->num_tx_queues);
1069
1070 for (i = 0; i < q_vector->txr_count; i++) {
1071 j = adapter->tx_ring[r_idx]->reg_idx;
1072 ixgbe_set_ivar(adapter, 1, j, v_idx);
1073 r_idx = find_next_bit(q_vector->txr_idx,
1074 adapter->num_tx_queues,
1075 r_idx + 1);
1076 }
1077
1078 if (q_vector->txr_count && !q_vector->rxr_count)
1079 /* tx only */
1080 q_vector->eitr = adapter->tx_eitr_param;
1081 else if (q_vector->rxr_count)
1082 /* rx or mixed */
1083 q_vector->eitr = adapter->rx_eitr_param;
1084
1085 ixgbe_write_eitr(q_vector);
1086 }
1087
1088 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1089 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1090 v_idx);
1091 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1092 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1093 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1094
1095 /* set up to autoclear timer, and the vectors */
1096 mask = IXGBE_EIMS_ENABLE_MASK;
1097 if (adapter->num_vfs)
1098 mask &= ~(IXGBE_EIMS_OTHER |
1099 IXGBE_EIMS_MAILBOX |
1100 IXGBE_EIMS_LSC);
1101 else
1102 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1103 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1104 }
1105
1106 enum latency_range {
1107 lowest_latency = 0,
1108 low_latency = 1,
1109 bulk_latency = 2,
1110 latency_invalid = 255
1111 };
1112
1113 /**
1114 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1115 * @adapter: pointer to adapter
1116 * @eitr: eitr setting (ints per sec) to give last timeslice
1117 * @itr_setting: current throttle rate in ints/second
1118 * @packets: the number of packets during this measurement interval
1119 * @bytes: the number of bytes during this measurement interval
1120 *
1121 * Stores a new ITR value based on packets and byte
1122 * counts during the last interrupt. The advantage of per interrupt
1123 * computation is faster updates and more accurate ITR for the current
1124 * traffic pattern. Constants in this function were computed
1125 * based on theoretical maximum wire speed and thresholds were set based
1126 * on testing data as well as attempting to minimize response time
1127 * while increasing bulk throughput.
1128 * this functionality is controlled by the InterruptThrottleRate module
1129 * parameter (see ixgbe_param.c)
1130 **/
1131 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1132 u32 eitr, u8 itr_setting,
1133 int packets, int bytes)
1134 {
1135 unsigned int retval = itr_setting;
1136 u32 timepassed_us;
1137 u64 bytes_perint;
1138
1139 if (packets == 0)
1140 goto update_itr_done;
1141
1142
1143 /* simple throttlerate management
1144 * 0-20MB/s lowest (100000 ints/s)
1145 * 20-100MB/s low (20000 ints/s)
1146 * 100-1249MB/s bulk (8000 ints/s)
1147 */
1148 /* what was last interrupt timeslice? */
1149 timepassed_us = 1000000/eitr;
1150 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1151
1152 switch (itr_setting) {
1153 case lowest_latency:
1154 if (bytes_perint > adapter->eitr_low)
1155 retval = low_latency;
1156 break;
1157 case low_latency:
1158 if (bytes_perint > adapter->eitr_high)
1159 retval = bulk_latency;
1160 else if (bytes_perint <= adapter->eitr_low)
1161 retval = lowest_latency;
1162 break;
1163 case bulk_latency:
1164 if (bytes_perint <= adapter->eitr_high)
1165 retval = low_latency;
1166 break;
1167 }
1168
1169 update_itr_done:
1170 return retval;
1171 }
1172
1173 /**
1174 * ixgbe_write_eitr - write EITR register in hardware specific way
1175 * @q_vector: structure containing interrupt and ring information
1176 *
1177 * This function is made to be called by ethtool and by the driver
1178 * when it needs to update EITR registers at runtime. Hardware
1179 * specific quirks/differences are taken care of here.
1180 */
1181 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1182 {
1183 struct ixgbe_adapter *adapter = q_vector->adapter;
1184 struct ixgbe_hw *hw = &adapter->hw;
1185 int v_idx = q_vector->v_idx;
1186 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1187
1188 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1189 /* must write high and low 16 bits to reset counter */
1190 itr_reg |= (itr_reg << 16);
1191 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1192 /*
1193 * set the WDIS bit to not clear the timer bits and cause an
1194 * immediate assertion of the interrupt
1195 */
1196 itr_reg |= IXGBE_EITR_CNT_WDIS;
1197 }
1198 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1199 }
1200
1201 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1202 {
1203 struct ixgbe_adapter *adapter = q_vector->adapter;
1204 u32 new_itr;
1205 u8 current_itr, ret_itr;
1206 int i, r_idx;
1207 struct ixgbe_ring *rx_ring, *tx_ring;
1208
1209 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1210 for (i = 0; i < q_vector->txr_count; i++) {
1211 tx_ring = adapter->tx_ring[r_idx];
1212 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1213 q_vector->tx_itr,
1214 tx_ring->total_packets,
1215 tx_ring->total_bytes);
1216 /* if the result for this queue would decrease interrupt
1217 * rate for this vector then use that result */
1218 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1219 q_vector->tx_itr - 1 : ret_itr);
1220 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1221 r_idx + 1);
1222 }
1223
1224 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1225 for (i = 0; i < q_vector->rxr_count; i++) {
1226 rx_ring = adapter->rx_ring[r_idx];
1227 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1228 q_vector->rx_itr,
1229 rx_ring->total_packets,
1230 rx_ring->total_bytes);
1231 /* if the result for this queue would decrease interrupt
1232 * rate for this vector then use that result */
1233 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1234 q_vector->rx_itr - 1 : ret_itr);
1235 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1236 r_idx + 1);
1237 }
1238
1239 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1240
1241 switch (current_itr) {
1242 /* counts and packets in update_itr are dependent on these numbers */
1243 case lowest_latency:
1244 new_itr = 100000;
1245 break;
1246 case low_latency:
1247 new_itr = 20000; /* aka hwitr = ~200 */
1248 break;
1249 case bulk_latency:
1250 default:
1251 new_itr = 8000;
1252 break;
1253 }
1254
1255 if (new_itr != q_vector->eitr) {
1256 /* do an exponential smoothing */
1257 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1258
1259 /* save the algorithm value here, not the smoothed one */
1260 q_vector->eitr = new_itr;
1261
1262 ixgbe_write_eitr(q_vector);
1263 }
1264
1265 return;
1266 }
1267
1268 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1269 {
1270 struct ixgbe_hw *hw = &adapter->hw;
1271
1272 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1273 (eicr & IXGBE_EICR_GPI_SDP1)) {
1274 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
1275 /* write to clear the interrupt */
1276 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1277 }
1278 }
1279
1280 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1281 {
1282 struct ixgbe_hw *hw = &adapter->hw;
1283
1284 if (eicr & IXGBE_EICR_GPI_SDP1) {
1285 /* Clear the interrupt */
1286 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1287 schedule_work(&adapter->multispeed_fiber_task);
1288 } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1289 /* Clear the interrupt */
1290 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1291 schedule_work(&adapter->sfp_config_module_task);
1292 } else {
1293 /* Interrupt isn't for us... */
1294 return;
1295 }
1296 }
1297
1298 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1299 {
1300 struct ixgbe_hw *hw = &adapter->hw;
1301
1302 adapter->lsc_int++;
1303 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1304 adapter->link_check_timeout = jiffies;
1305 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1306 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1307 IXGBE_WRITE_FLUSH(hw);
1308 schedule_work(&adapter->watchdog_task);
1309 }
1310 }
1311
1312 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1313 {
1314 struct net_device *netdev = data;
1315 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1316 struct ixgbe_hw *hw = &adapter->hw;
1317 u32 eicr;
1318
1319 /*
1320 * Workaround for Silicon errata. Use clear-by-write instead
1321 * of clear-by-read. Reading with EICS will return the
1322 * interrupt causes without clearing, which later be done
1323 * with the write to EICR.
1324 */
1325 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1326 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1327
1328 if (eicr & IXGBE_EICR_LSC)
1329 ixgbe_check_lsc(adapter);
1330
1331 if (eicr & IXGBE_EICR_MAILBOX)
1332 ixgbe_msg_task(adapter);
1333
1334 if (hw->mac.type == ixgbe_mac_82598EB)
1335 ixgbe_check_fan_failure(adapter, eicr);
1336
1337 if (hw->mac.type == ixgbe_mac_82599EB) {
1338 ixgbe_check_sfp_event(adapter, eicr);
1339
1340 /* Handle Flow Director Full threshold interrupt */
1341 if (eicr & IXGBE_EICR_FLOW_DIR) {
1342 int i;
1343 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1344 /* Disable transmits before FDIR Re-initialization */
1345 netif_tx_stop_all_queues(netdev);
1346 for (i = 0; i < adapter->num_tx_queues; i++) {
1347 struct ixgbe_ring *tx_ring =
1348 adapter->tx_ring[i];
1349 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1350 &tx_ring->reinit_state))
1351 schedule_work(&adapter->fdir_reinit_task);
1352 }
1353 }
1354 }
1355 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1356 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1357
1358 return IRQ_HANDLED;
1359 }
1360
1361 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1362 u64 qmask)
1363 {
1364 u32 mask;
1365
1366 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1367 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1368 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1369 } else {
1370 mask = (qmask & 0xFFFFFFFF);
1371 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1372 mask = (qmask >> 32);
1373 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1374 }
1375 /* skip the flush */
1376 }
1377
1378 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1379 u64 qmask)
1380 {
1381 u32 mask;
1382
1383 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1384 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1385 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1386 } else {
1387 mask = (qmask & 0xFFFFFFFF);
1388 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1389 mask = (qmask >> 32);
1390 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1391 }
1392 /* skip the flush */
1393 }
1394
1395 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1396 {
1397 struct ixgbe_q_vector *q_vector = data;
1398 struct ixgbe_adapter *adapter = q_vector->adapter;
1399 struct ixgbe_ring *tx_ring;
1400 int i, r_idx;
1401
1402 if (!q_vector->txr_count)
1403 return IRQ_HANDLED;
1404
1405 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1406 for (i = 0; i < q_vector->txr_count; i++) {
1407 tx_ring = adapter->tx_ring[r_idx];
1408 tx_ring->total_bytes = 0;
1409 tx_ring->total_packets = 0;
1410 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1411 r_idx + 1);
1412 }
1413
1414 /* EIAM disabled interrupts (on this vector) for us */
1415 napi_schedule(&q_vector->napi);
1416
1417 return IRQ_HANDLED;
1418 }
1419
1420 /**
1421 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1422 * @irq: unused
1423 * @data: pointer to our q_vector struct for this interrupt vector
1424 **/
1425 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1426 {
1427 struct ixgbe_q_vector *q_vector = data;
1428 struct ixgbe_adapter *adapter = q_vector->adapter;
1429 struct ixgbe_ring *rx_ring;
1430 int r_idx;
1431 int i;
1432
1433 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1434 for (i = 0; i < q_vector->rxr_count; i++) {
1435 rx_ring = adapter->rx_ring[r_idx];
1436 rx_ring->total_bytes = 0;
1437 rx_ring->total_packets = 0;
1438 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1439 r_idx + 1);
1440 }
1441
1442 if (!q_vector->rxr_count)
1443 return IRQ_HANDLED;
1444
1445 /* disable interrupts on this vector only */
1446 /* EIAM disabled interrupts (on this vector) for us */
1447 napi_schedule(&q_vector->napi);
1448
1449 return IRQ_HANDLED;
1450 }
1451
1452 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1453 {
1454 struct ixgbe_q_vector *q_vector = data;
1455 struct ixgbe_adapter *adapter = q_vector->adapter;
1456 struct ixgbe_ring *ring;
1457 int r_idx;
1458 int i;
1459
1460 if (!q_vector->txr_count && !q_vector->rxr_count)
1461 return IRQ_HANDLED;
1462
1463 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1464 for (i = 0; i < q_vector->txr_count; i++) {
1465 ring = adapter->tx_ring[r_idx];
1466 ring->total_bytes = 0;
1467 ring->total_packets = 0;
1468 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1469 r_idx + 1);
1470 }
1471
1472 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1473 for (i = 0; i < q_vector->rxr_count; i++) {
1474 ring = adapter->rx_ring[r_idx];
1475 ring->total_bytes = 0;
1476 ring->total_packets = 0;
1477 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1478 r_idx + 1);
1479 }
1480
1481 /* EIAM disabled interrupts (on this vector) for us */
1482 napi_schedule(&q_vector->napi);
1483
1484 return IRQ_HANDLED;
1485 }
1486
1487 /**
1488 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1489 * @napi: napi struct with our devices info in it
1490 * @budget: amount of work driver is allowed to do this pass, in packets
1491 *
1492 * This function is optimized for cleaning one queue only on a single
1493 * q_vector!!!
1494 **/
1495 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1496 {
1497 struct ixgbe_q_vector *q_vector =
1498 container_of(napi, struct ixgbe_q_vector, napi);
1499 struct ixgbe_adapter *adapter = q_vector->adapter;
1500 struct ixgbe_ring *rx_ring = NULL;
1501 int work_done = 0;
1502 long r_idx;
1503
1504 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1505 rx_ring = adapter->rx_ring[r_idx];
1506 #ifdef CONFIG_IXGBE_DCA
1507 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1508 ixgbe_update_rx_dca(adapter, rx_ring);
1509 #endif
1510
1511 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1512
1513 /* If all Rx work done, exit the polling mode */
1514 if (work_done < budget) {
1515 napi_complete(napi);
1516 if (adapter->rx_itr_setting & 1)
1517 ixgbe_set_itr_msix(q_vector);
1518 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1519 ixgbe_irq_enable_queues(adapter,
1520 ((u64)1 << q_vector->v_idx));
1521 }
1522
1523 return work_done;
1524 }
1525
1526 /**
1527 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1528 * @napi: napi struct with our devices info in it
1529 * @budget: amount of work driver is allowed to do this pass, in packets
1530 *
1531 * This function will clean more than one rx queue associated with a
1532 * q_vector.
1533 **/
1534 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1535 {
1536 struct ixgbe_q_vector *q_vector =
1537 container_of(napi, struct ixgbe_q_vector, napi);
1538 struct ixgbe_adapter *adapter = q_vector->adapter;
1539 struct ixgbe_ring *ring = NULL;
1540 int work_done = 0, i;
1541 long r_idx;
1542 bool tx_clean_complete = true;
1543
1544 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1545 for (i = 0; i < q_vector->txr_count; i++) {
1546 ring = adapter->tx_ring[r_idx];
1547 #ifdef CONFIG_IXGBE_DCA
1548 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1549 ixgbe_update_tx_dca(adapter, ring);
1550 #endif
1551 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1552 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1553 r_idx + 1);
1554 }
1555
1556 /* attempt to distribute budget to each queue fairly, but don't allow
1557 * the budget to go below 1 because we'll exit polling */
1558 budget /= (q_vector->rxr_count ?: 1);
1559 budget = max(budget, 1);
1560 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1561 for (i = 0; i < q_vector->rxr_count; i++) {
1562 ring = adapter->rx_ring[r_idx];
1563 #ifdef CONFIG_IXGBE_DCA
1564 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1565 ixgbe_update_rx_dca(adapter, ring);
1566 #endif
1567 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1568 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1569 r_idx + 1);
1570 }
1571
1572 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1573 ring = adapter->rx_ring[r_idx];
1574 /* If all Rx work done, exit the polling mode */
1575 if (work_done < budget) {
1576 napi_complete(napi);
1577 if (adapter->rx_itr_setting & 1)
1578 ixgbe_set_itr_msix(q_vector);
1579 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1580 ixgbe_irq_enable_queues(adapter,
1581 ((u64)1 << q_vector->v_idx));
1582 return 0;
1583 }
1584
1585 return work_done;
1586 }
1587
1588 /**
1589 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1590 * @napi: napi struct with our devices info in it
1591 * @budget: amount of work driver is allowed to do this pass, in packets
1592 *
1593 * This function is optimized for cleaning one queue only on a single
1594 * q_vector!!!
1595 **/
1596 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1597 {
1598 struct ixgbe_q_vector *q_vector =
1599 container_of(napi, struct ixgbe_q_vector, napi);
1600 struct ixgbe_adapter *adapter = q_vector->adapter;
1601 struct ixgbe_ring *tx_ring = NULL;
1602 int work_done = 0;
1603 long r_idx;
1604
1605 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1606 tx_ring = adapter->tx_ring[r_idx];
1607 #ifdef CONFIG_IXGBE_DCA
1608 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1609 ixgbe_update_tx_dca(adapter, tx_ring);
1610 #endif
1611
1612 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
1613 work_done = budget;
1614
1615 /* If all Tx work done, exit the polling mode */
1616 if (work_done < budget) {
1617 napi_complete(napi);
1618 if (adapter->tx_itr_setting & 1)
1619 ixgbe_set_itr_msix(q_vector);
1620 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1621 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
1622 }
1623
1624 return work_done;
1625 }
1626
1627 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1628 int r_idx)
1629 {
1630 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1631
1632 set_bit(r_idx, q_vector->rxr_idx);
1633 q_vector->rxr_count++;
1634 }
1635
1636 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1637 int t_idx)
1638 {
1639 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1640
1641 set_bit(t_idx, q_vector->txr_idx);
1642 q_vector->txr_count++;
1643 }
1644
1645 /**
1646 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1647 * @adapter: board private structure to initialize
1648 * @vectors: allotted vector count for descriptor rings
1649 *
1650 * This function maps descriptor rings to the queue-specific vectors
1651 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1652 * one vector per ring/queue, but on a constrained vector budget, we
1653 * group the rings as "efficiently" as possible. You would add new
1654 * mapping configurations in here.
1655 **/
1656 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1657 int vectors)
1658 {
1659 int v_start = 0;
1660 int rxr_idx = 0, txr_idx = 0;
1661 int rxr_remaining = adapter->num_rx_queues;
1662 int txr_remaining = adapter->num_tx_queues;
1663 int i, j;
1664 int rqpv, tqpv;
1665 int err = 0;
1666
1667 /* No mapping required if MSI-X is disabled. */
1668 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1669 goto out;
1670
1671 /*
1672 * The ideal configuration...
1673 * We have enough vectors to map one per queue.
1674 */
1675 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1676 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1677 map_vector_to_rxq(adapter, v_start, rxr_idx);
1678
1679 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1680 map_vector_to_txq(adapter, v_start, txr_idx);
1681
1682 goto out;
1683 }
1684
1685 /*
1686 * If we don't have enough vectors for a 1-to-1
1687 * mapping, we'll have to group them so there are
1688 * multiple queues per vector.
1689 */
1690 /* Re-adjusting *qpv takes care of the remainder. */
1691 for (i = v_start; i < vectors; i++) {
1692 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1693 for (j = 0; j < rqpv; j++) {
1694 map_vector_to_rxq(adapter, i, rxr_idx);
1695 rxr_idx++;
1696 rxr_remaining--;
1697 }
1698 }
1699 for (i = v_start; i < vectors; i++) {
1700 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1701 for (j = 0; j < tqpv; j++) {
1702 map_vector_to_txq(adapter, i, txr_idx);
1703 txr_idx++;
1704 txr_remaining--;
1705 }
1706 }
1707
1708 out:
1709 return err;
1710 }
1711
1712 /**
1713 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1714 * @adapter: board private structure
1715 *
1716 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1717 * interrupts from the kernel.
1718 **/
1719 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1720 {
1721 struct net_device *netdev = adapter->netdev;
1722 irqreturn_t (*handler)(int, void *);
1723 int i, vector, q_vectors, err;
1724 int ri=0, ti=0;
1725
1726 /* Decrement for Other and TCP Timer vectors */
1727 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1728
1729 /* Map the Tx/Rx rings to the vectors we were allotted. */
1730 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1731 if (err)
1732 goto out;
1733
1734 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1735 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1736 &ixgbe_msix_clean_many)
1737 for (vector = 0; vector < q_vectors; vector++) {
1738 handler = SET_HANDLER(adapter->q_vector[vector]);
1739
1740 if(handler == &ixgbe_msix_clean_rx) {
1741 sprintf(adapter->name[vector], "%s-%s-%d",
1742 netdev->name, "rx", ri++);
1743 }
1744 else if(handler == &ixgbe_msix_clean_tx) {
1745 sprintf(adapter->name[vector], "%s-%s-%d",
1746 netdev->name, "tx", ti++);
1747 }
1748 else
1749 sprintf(adapter->name[vector], "%s-%s-%d",
1750 netdev->name, "TxRx", vector);
1751
1752 err = request_irq(adapter->msix_entries[vector].vector,
1753 handler, 0, adapter->name[vector],
1754 adapter->q_vector[vector]);
1755 if (err) {
1756 DPRINTK(PROBE, ERR,
1757 "request_irq failed for MSIX interrupt "
1758 "Error: %d\n", err);
1759 goto free_queue_irqs;
1760 }
1761 }
1762
1763 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1764 err = request_irq(adapter->msix_entries[vector].vector,
1765 ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1766 if (err) {
1767 DPRINTK(PROBE, ERR,
1768 "request_irq for msix_lsc failed: %d\n", err);
1769 goto free_queue_irqs;
1770 }
1771
1772 return 0;
1773
1774 free_queue_irqs:
1775 for (i = vector - 1; i >= 0; i--)
1776 free_irq(adapter->msix_entries[--vector].vector,
1777 adapter->q_vector[i]);
1778 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1779 pci_disable_msix(adapter->pdev);
1780 kfree(adapter->msix_entries);
1781 adapter->msix_entries = NULL;
1782 out:
1783 return err;
1784 }
1785
1786 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1787 {
1788 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1789 u8 current_itr;
1790 u32 new_itr = q_vector->eitr;
1791 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
1792 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
1793
1794 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1795 q_vector->tx_itr,
1796 tx_ring->total_packets,
1797 tx_ring->total_bytes);
1798 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1799 q_vector->rx_itr,
1800 rx_ring->total_packets,
1801 rx_ring->total_bytes);
1802
1803 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1804
1805 switch (current_itr) {
1806 /* counts and packets in update_itr are dependent on these numbers */
1807 case lowest_latency:
1808 new_itr = 100000;
1809 break;
1810 case low_latency:
1811 new_itr = 20000; /* aka hwitr = ~200 */
1812 break;
1813 case bulk_latency:
1814 new_itr = 8000;
1815 break;
1816 default:
1817 break;
1818 }
1819
1820 if (new_itr != q_vector->eitr) {
1821 /* do an exponential smoothing */
1822 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1823
1824 /* save the algorithm value here, not the smoothed one */
1825 q_vector->eitr = new_itr;
1826
1827 ixgbe_write_eitr(q_vector);
1828 }
1829
1830 return;
1831 }
1832
1833 /**
1834 * ixgbe_irq_enable - Enable default interrupt generation settings
1835 * @adapter: board private structure
1836 **/
1837 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1838 {
1839 u32 mask;
1840
1841 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1842 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1843 mask |= IXGBE_EIMS_GPI_SDP1;
1844 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1845 mask |= IXGBE_EIMS_ECC;
1846 mask |= IXGBE_EIMS_GPI_SDP1;
1847 mask |= IXGBE_EIMS_GPI_SDP2;
1848 if (adapter->num_vfs)
1849 mask |= IXGBE_EIMS_MAILBOX;
1850 }
1851 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
1852 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
1853 mask |= IXGBE_EIMS_FLOW_DIR;
1854
1855 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1856 ixgbe_irq_enable_queues(adapter, ~0);
1857 IXGBE_WRITE_FLUSH(&adapter->hw);
1858
1859 if (adapter->num_vfs > 32) {
1860 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1861 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1862 }
1863 }
1864
1865 /**
1866 * ixgbe_intr - legacy mode Interrupt Handler
1867 * @irq: interrupt number
1868 * @data: pointer to a network interface device structure
1869 **/
1870 static irqreturn_t ixgbe_intr(int irq, void *data)
1871 {
1872 struct net_device *netdev = data;
1873 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1874 struct ixgbe_hw *hw = &adapter->hw;
1875 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1876 u32 eicr;
1877
1878 /*
1879 * Workaround for silicon errata. Mask the interrupts
1880 * before the read of EICR.
1881 */
1882 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1883
1884 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1885 * therefore no explict interrupt disable is necessary */
1886 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1887 if (!eicr) {
1888 /* shared interrupt alert!
1889 * make sure interrupts are enabled because the read will
1890 * have disabled interrupts due to EIAM */
1891 ixgbe_irq_enable(adapter);
1892 return IRQ_NONE; /* Not our interrupt */
1893 }
1894
1895 if (eicr & IXGBE_EICR_LSC)
1896 ixgbe_check_lsc(adapter);
1897
1898 if (hw->mac.type == ixgbe_mac_82599EB)
1899 ixgbe_check_sfp_event(adapter, eicr);
1900
1901 ixgbe_check_fan_failure(adapter, eicr);
1902
1903 if (napi_schedule_prep(&(q_vector->napi))) {
1904 adapter->tx_ring[0]->total_packets = 0;
1905 adapter->tx_ring[0]->total_bytes = 0;
1906 adapter->rx_ring[0]->total_packets = 0;
1907 adapter->rx_ring[0]->total_bytes = 0;
1908 /* would disable interrupts here but EIAM disabled it */
1909 __napi_schedule(&(q_vector->napi));
1910 }
1911
1912 return IRQ_HANDLED;
1913 }
1914
1915 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1916 {
1917 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1918
1919 for (i = 0; i < q_vectors; i++) {
1920 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
1921 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1922 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1923 q_vector->rxr_count = 0;
1924 q_vector->txr_count = 0;
1925 }
1926 }
1927
1928 /**
1929 * ixgbe_request_irq - initialize interrupts
1930 * @adapter: board private structure
1931 *
1932 * Attempts to configure interrupts using the best available
1933 * capabilities of the hardware and kernel.
1934 **/
1935 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1936 {
1937 struct net_device *netdev = adapter->netdev;
1938 int err;
1939
1940 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1941 err = ixgbe_request_msix_irqs(adapter);
1942 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1943 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
1944 netdev->name, netdev);
1945 } else {
1946 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
1947 netdev->name, netdev);
1948 }
1949
1950 if (err)
1951 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1952
1953 return err;
1954 }
1955
1956 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1957 {
1958 struct net_device *netdev = adapter->netdev;
1959
1960 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1961 int i, q_vectors;
1962
1963 q_vectors = adapter->num_msix_vectors;
1964
1965 i = q_vectors - 1;
1966 free_irq(adapter->msix_entries[i].vector, netdev);
1967
1968 i--;
1969 for (; i >= 0; i--) {
1970 free_irq(adapter->msix_entries[i].vector,
1971 adapter->q_vector[i]);
1972 }
1973
1974 ixgbe_reset_q_vectors(adapter);
1975 } else {
1976 free_irq(adapter->pdev->irq, netdev);
1977 }
1978 }
1979
1980 /**
1981 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1982 * @adapter: board private structure
1983 **/
1984 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1985 {
1986 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1987 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1988 } else {
1989 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
1990 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
1991 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
1992 if (adapter->num_vfs > 32)
1993 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
1994 }
1995 IXGBE_WRITE_FLUSH(&adapter->hw);
1996 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1997 int i;
1998 for (i = 0; i < adapter->num_msix_vectors; i++)
1999 synchronize_irq(adapter->msix_entries[i].vector);
2000 } else {
2001 synchronize_irq(adapter->pdev->irq);
2002 }
2003 }
2004
2005 /**
2006 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2007 *
2008 **/
2009 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2010 {
2011 struct ixgbe_hw *hw = &adapter->hw;
2012
2013 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2014 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2015
2016 ixgbe_set_ivar(adapter, 0, 0, 0);
2017 ixgbe_set_ivar(adapter, 1, 0, 0);
2018
2019 map_vector_to_rxq(adapter, 0, 0);
2020 map_vector_to_txq(adapter, 0, 0);
2021
2022 DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
2023 }
2024
2025 /**
2026 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2027 * @adapter: board private structure
2028 *
2029 * Configure the Tx unit of the MAC after a reset.
2030 **/
2031 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2032 {
2033 u64 tdba;
2034 struct ixgbe_hw *hw = &adapter->hw;
2035 u32 i, j, tdlen, txctrl;
2036
2037 /* Setup the HW Tx Head and Tail descriptor pointers */
2038 for (i = 0; i < adapter->num_tx_queues; i++) {
2039 struct ixgbe_ring *ring = adapter->tx_ring[i];
2040 j = ring->reg_idx;
2041 tdba = ring->dma;
2042 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
2043 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
2044 (tdba & DMA_BIT_MASK(32)));
2045 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
2046 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
2047 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
2048 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
2049 adapter->tx_ring[i]->head = IXGBE_TDH(j);
2050 adapter->tx_ring[i]->tail = IXGBE_TDT(j);
2051 /*
2052 * Disable Tx Head Writeback RO bit, since this hoses
2053 * bookkeeping if things aren't delivered in order.
2054 */
2055 switch (hw->mac.type) {
2056 case ixgbe_mac_82598EB:
2057 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
2058 break;
2059 case ixgbe_mac_82599EB:
2060 default:
2061 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(j));
2062 break;
2063 }
2064 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
2065 switch (hw->mac.type) {
2066 case ixgbe_mac_82598EB:
2067 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
2068 break;
2069 case ixgbe_mac_82599EB:
2070 default:
2071 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(j), txctrl);
2072 break;
2073 }
2074 }
2075
2076 if (hw->mac.type == ixgbe_mac_82599EB) {
2077 u32 rttdcs;
2078 u32 mask;
2079
2080 /* disable the arbiter while setting MTQC */
2081 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2082 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2083 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2084
2085 /* set transmit pool layout */
2086 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2087 switch (adapter->flags & mask) {
2088
2089 case (IXGBE_FLAG_SRIOV_ENABLED):
2090 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2091 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2092 break;
2093
2094 case (IXGBE_FLAG_DCB_ENABLED):
2095 /* We enable 8 traffic classes, DCB only */
2096 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2097 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2098 break;
2099
2100 default:
2101 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2102 break;
2103 }
2104
2105 /* re-eable the arbiter */
2106 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2107 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2108 }
2109 }
2110
2111 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2112
2113 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2114 struct ixgbe_ring *rx_ring)
2115 {
2116 u32 srrctl;
2117 int index;
2118 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2119
2120 index = rx_ring->reg_idx;
2121 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2122 unsigned long mask;
2123 mask = (unsigned long) feature[RING_F_RSS].mask;
2124 index = index & mask;
2125 }
2126 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2127
2128 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2129 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2130
2131 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2132 IXGBE_SRRCTL_BSIZEHDR_MASK;
2133
2134 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2135 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2136 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2137 #else
2138 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2139 #endif
2140 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2141 } else {
2142 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2143 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2144 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2145 }
2146
2147 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2148 }
2149
2150 static u32 ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2151 {
2152 u32 mrqc = 0;
2153 int mask;
2154
2155 if (!(adapter->hw.mac.type == ixgbe_mac_82599EB))
2156 return mrqc;
2157
2158 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2159 #ifdef CONFIG_IXGBE_DCB
2160 | IXGBE_FLAG_DCB_ENABLED
2161 #endif
2162 | IXGBE_FLAG_SRIOV_ENABLED
2163 );
2164
2165 switch (mask) {
2166 case (IXGBE_FLAG_RSS_ENABLED):
2167 mrqc = IXGBE_MRQC_RSSEN;
2168 break;
2169 case (IXGBE_FLAG_SRIOV_ENABLED):
2170 mrqc = IXGBE_MRQC_VMDQEN;
2171 break;
2172 #ifdef CONFIG_IXGBE_DCB
2173 case (IXGBE_FLAG_DCB_ENABLED):
2174 mrqc = IXGBE_MRQC_RT8TCEN;
2175 break;
2176 #endif /* CONFIG_IXGBE_DCB */
2177 default:
2178 break;
2179 }
2180
2181 return mrqc;
2182 }
2183
2184 /**
2185 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2186 * @adapter: address of board private structure
2187 * @index: index of ring to set
2188 **/
2189 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter, int index)
2190 {
2191 struct ixgbe_ring *rx_ring;
2192 struct ixgbe_hw *hw = &adapter->hw;
2193 int j;
2194 u32 rscctrl;
2195 int rx_buf_len;
2196
2197 rx_ring = adapter->rx_ring[index];
2198 j = rx_ring->reg_idx;
2199 rx_buf_len = rx_ring->rx_buf_len;
2200 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2201 rscctrl |= IXGBE_RSCCTL_RSCEN;
2202 /*
2203 * we must limit the number of descriptors so that the
2204 * total size of max desc * buf_len is not greater
2205 * than 65535
2206 */
2207 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2208 #if (MAX_SKB_FRAGS > 16)
2209 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2210 #elif (MAX_SKB_FRAGS > 8)
2211 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2212 #elif (MAX_SKB_FRAGS > 4)
2213 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2214 #else
2215 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2216 #endif
2217 } else {
2218 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2219 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2220 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2221 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2222 else
2223 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2224 }
2225 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2226 }
2227
2228 /**
2229 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2230 * @adapter: board private structure
2231 *
2232 * Configure the Rx unit of the MAC after a reset.
2233 **/
2234 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2235 {
2236 u64 rdba;
2237 struct ixgbe_hw *hw = &adapter->hw;
2238 struct ixgbe_ring *rx_ring;
2239 struct net_device *netdev = adapter->netdev;
2240 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2241 int i, j;
2242 u32 rdlen, rxctrl, rxcsum;
2243 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2244 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2245 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2246 u32 fctrl, hlreg0;
2247 u32 reta = 0, mrqc = 0;
2248 u32 rdrxctl;
2249 int rx_buf_len;
2250
2251 /* Decide whether to use packet split mode or not */
2252 /* Do not use packet split if we're in SR-IOV Mode */
2253 if (!adapter->num_vfs)
2254 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2255
2256 /* Set the RX buffer length according to the mode */
2257 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2258 rx_buf_len = IXGBE_RX_HDR_SIZE;
2259 if (hw->mac.type == ixgbe_mac_82599EB) {
2260 /* PSRTYPE must be initialized in 82599 */
2261 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2262 IXGBE_PSRTYPE_UDPHDR |
2263 IXGBE_PSRTYPE_IPV4HDR |
2264 IXGBE_PSRTYPE_IPV6HDR |
2265 IXGBE_PSRTYPE_L2HDR;
2266 IXGBE_WRITE_REG(hw,
2267 IXGBE_PSRTYPE(adapter->num_vfs),
2268 psrtype);
2269 }
2270 } else {
2271 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2272 (netdev->mtu <= ETH_DATA_LEN))
2273 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2274 else
2275 rx_buf_len = ALIGN(max_frame, 1024);
2276 }
2277
2278 fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
2279 fctrl |= IXGBE_FCTRL_BAM;
2280 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
2281 fctrl |= IXGBE_FCTRL_PMCF;
2282 IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
2283
2284 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2285 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2286 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2287 else
2288 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2289 #ifdef IXGBE_FCOE
2290 if (netdev->features & NETIF_F_FCOE_MTU)
2291 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2292 #endif
2293 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2294
2295 rdlen = adapter->rx_ring[0]->count * sizeof(union ixgbe_adv_rx_desc);
2296 /* disable receives while setting up the descriptors */
2297 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2298 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2299
2300 /*
2301 * Setup the HW Rx Head and Tail Descriptor Pointers and
2302 * the Base and Length of the Rx Descriptor Ring
2303 */
2304 for (i = 0; i < adapter->num_rx_queues; i++) {
2305 rx_ring = adapter->rx_ring[i];
2306 rdba = rx_ring->dma;
2307 j = rx_ring->reg_idx;
2308 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_BIT_MASK(32)));
2309 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
2310 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
2311 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
2312 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
2313 rx_ring->head = IXGBE_RDH(j);
2314 rx_ring->tail = IXGBE_RDT(j);
2315 rx_ring->rx_buf_len = rx_buf_len;
2316
2317 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2318 rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2319 else
2320 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2321
2322 #ifdef IXGBE_FCOE
2323 if (netdev->features & NETIF_F_FCOE_MTU) {
2324 struct ixgbe_ring_feature *f;
2325 f = &adapter->ring_feature[RING_F_FCOE];
2326 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2327 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2328 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2329 rx_ring->rx_buf_len =
2330 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2331 }
2332 }
2333
2334 #endif /* IXGBE_FCOE */
2335 ixgbe_configure_srrctl(adapter, rx_ring);
2336 }
2337
2338 if (hw->mac.type == ixgbe_mac_82598EB) {
2339 /*
2340 * For VMDq support of different descriptor types or
2341 * buffer sizes through the use of multiple SRRCTL
2342 * registers, RDRXCTL.MVMEN must be set to 1
2343 *
2344 * also, the manual doesn't mention it clearly but DCA hints
2345 * will only use queue 0's tags unless this bit is set. Side
2346 * effects of setting this bit are only that SRRCTL must be
2347 * fully programmed [0..15]
2348 */
2349 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2350 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2351 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2352 }
2353
2354 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2355 u32 vt_reg_bits;
2356 u32 reg_offset, vf_shift;
2357 u32 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2358 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN
2359 | IXGBE_VT_CTL_REPLEN;
2360 vt_reg_bits |= (adapter->num_vfs <<
2361 IXGBE_VT_CTL_POOL_SHIFT);
2362 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2363 IXGBE_WRITE_REG(hw, IXGBE_MRQC, 0);
2364
2365 vf_shift = adapter->num_vfs % 32;
2366 reg_offset = adapter->num_vfs / 32;
2367 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), 0);
2368 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), 0);
2369 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), 0);
2370 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), 0);
2371 /* Enable only the PF's pool for Tx/Rx */
2372 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2373 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2374 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2375 ixgbe_set_vmolr(hw, adapter->num_vfs);
2376 }
2377
2378 /* Program MRQC for the distribution of queues */
2379 mrqc = ixgbe_setup_mrqc(adapter);
2380
2381 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2382 /* Fill out redirection table */
2383 for (i = 0, j = 0; i < 128; i++, j++) {
2384 if (j == adapter->ring_feature[RING_F_RSS].indices)
2385 j = 0;
2386 /* reta = 4-byte sliding window of
2387 * 0x00..(indices-1)(indices-1)00..etc. */
2388 reta = (reta << 8) | (j * 0x11);
2389 if ((i & 3) == 3)
2390 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2391 }
2392
2393 /* Fill out hash function seeds */
2394 for (i = 0; i < 10; i++)
2395 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2396
2397 if (hw->mac.type == ixgbe_mac_82598EB)
2398 mrqc |= IXGBE_MRQC_RSSEN;
2399 /* Perform hash on these packet types */
2400 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2401 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2402 | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
2403 | IXGBE_MRQC_RSS_FIELD_IPV6
2404 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
2405 | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2406 }
2407 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2408
2409 if (adapter->num_vfs) {
2410 u32 reg;
2411
2412 /* Map PF MAC address in RAR Entry 0 to first pool
2413 * following VFs */
2414 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2415
2416 /* Set up VF register offsets for selected VT Mode, i.e.
2417 * 64 VFs for SR-IOV */
2418 reg = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2419 reg |= IXGBE_GCR_EXT_SRIOV;
2420 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, reg);
2421 }
2422
2423 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2424
2425 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
2426 adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
2427 /* Disable indicating checksum in descriptor, enables
2428 * RSS hash */
2429 rxcsum |= IXGBE_RXCSUM_PCSD;
2430 }
2431 if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
2432 /* Enable IPv4 payload checksum for UDP fragments
2433 * if PCSD is not set */
2434 rxcsum |= IXGBE_RXCSUM_IPPCSE;
2435 }
2436
2437 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2438
2439 if (hw->mac.type == ixgbe_mac_82599EB) {
2440 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2441 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2442 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2443 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2444 }
2445
2446 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
2447 /* Enable 82599 HW-RSC */
2448 for (i = 0; i < adapter->num_rx_queues; i++)
2449 ixgbe_configure_rscctl(adapter, i);
2450
2451 /* Disable RSC for ACK packets */
2452 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2453 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2454 }
2455 }
2456
2457 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2458 {
2459 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2460 struct ixgbe_hw *hw = &adapter->hw;
2461 int pool_ndx = adapter->num_vfs;
2462
2463 /* add VID to filter table */
2464 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
2465 }
2466
2467 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2468 {
2469 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2470 struct ixgbe_hw *hw = &adapter->hw;
2471 int pool_ndx = adapter->num_vfs;
2472
2473 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2474 ixgbe_irq_disable(adapter);
2475
2476 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2477
2478 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2479 ixgbe_irq_enable(adapter);
2480
2481 /* remove VID from filter table */
2482 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
2483 }
2484
2485 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2486 struct vlan_group *grp)
2487 {
2488 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2489 u32 ctrl;
2490 int i, j;
2491
2492 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2493 ixgbe_irq_disable(adapter);
2494 adapter->vlgrp = grp;
2495
2496 /*
2497 * For a DCB driver, always enable VLAN tag stripping so we can
2498 * still receive traffic from a DCB-enabled host even if we're
2499 * not in DCB mode.
2500 */
2501 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2502
2503 /* Disable CFI check */
2504 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2505
2506 /* enable VLAN tag stripping */
2507 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2508 ctrl |= IXGBE_VLNCTRL_VME;
2509 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2510 for (i = 0; i < adapter->num_rx_queues; i++) {
2511 u32 ctrl;
2512 j = adapter->rx_ring[i]->reg_idx;
2513 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_RXDCTL(j));
2514 ctrl |= IXGBE_RXDCTL_VME;
2515 IXGBE_WRITE_REG(&adapter->hw, IXGBE_RXDCTL(j), ctrl);
2516 }
2517 }
2518
2519 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2520
2521 ixgbe_vlan_rx_add_vid(netdev, 0);
2522
2523 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2524 ixgbe_irq_enable(adapter);
2525 }
2526
2527 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2528 {
2529 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2530
2531 if (adapter->vlgrp) {
2532 u16 vid;
2533 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2534 if (!vlan_group_get_device(adapter->vlgrp, vid))
2535 continue;
2536 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2537 }
2538 }
2539 }
2540
2541 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
2542 {
2543 struct dev_mc_list *mc_ptr;
2544 u8 *addr = *mc_addr_ptr;
2545 *vmdq = 0;
2546
2547 mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
2548 if (mc_ptr->next)
2549 *mc_addr_ptr = mc_ptr->next->dmi_addr;
2550 else
2551 *mc_addr_ptr = NULL;
2552
2553 return addr;
2554 }
2555
2556 /**
2557 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
2558 * @netdev: network interface device structure
2559 *
2560 * The set_rx_method entry point is called whenever the unicast/multicast
2561 * address list or the network interface flags are updated. This routine is
2562 * responsible for configuring the hardware for proper unicast, multicast and
2563 * promiscuous mode.
2564 **/
2565 void ixgbe_set_rx_mode(struct net_device *netdev)
2566 {
2567 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2568 struct ixgbe_hw *hw = &adapter->hw;
2569 u32 fctrl, vlnctrl;
2570 u8 *addr_list = NULL;
2571 int addr_count = 0;
2572
2573 /* Check for Promiscuous and All Multicast modes */
2574
2575 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2576 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2577
2578 if (netdev->flags & IFF_PROMISC) {
2579 hw->addr_ctrl.user_set_promisc = 1;
2580 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2581 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2582 } else {
2583 if (netdev->flags & IFF_ALLMULTI) {
2584 fctrl |= IXGBE_FCTRL_MPE;
2585 fctrl &= ~IXGBE_FCTRL_UPE;
2586 } else {
2587 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2588 }
2589 vlnctrl |= IXGBE_VLNCTRL_VFE;
2590 hw->addr_ctrl.user_set_promisc = 0;
2591 }
2592
2593 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
2594 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2595
2596 /* reprogram secondary unicast list */
2597 hw->mac.ops.update_uc_addr_list(hw, netdev);
2598
2599 /* reprogram multicast list */
2600 addr_count = netdev_mc_count(netdev);
2601 if (addr_count)
2602 addr_list = netdev->mc_list->dmi_addr;
2603 hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
2604 ixgbe_addr_list_itr);
2605 if (adapter->num_vfs)
2606 ixgbe_restore_vf_multicasts(adapter);
2607 }
2608
2609 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
2610 {
2611 int q_idx;
2612 struct ixgbe_q_vector *q_vector;
2613 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2614
2615 /* legacy and MSI only use one vector */
2616 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2617 q_vectors = 1;
2618
2619 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2620 struct napi_struct *napi;
2621 q_vector = adapter->q_vector[q_idx];
2622 napi = &q_vector->napi;
2623 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2624 if (!q_vector->rxr_count || !q_vector->txr_count) {
2625 if (q_vector->txr_count == 1)
2626 napi->poll = &ixgbe_clean_txonly;
2627 else if (q_vector->rxr_count == 1)
2628 napi->poll = &ixgbe_clean_rxonly;
2629 }
2630 }
2631
2632 napi_enable(napi);
2633 }
2634 }
2635
2636 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2637 {
2638 int q_idx;
2639 struct ixgbe_q_vector *q_vector;
2640 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2641
2642 /* legacy and MSI only use one vector */
2643 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2644 q_vectors = 1;
2645
2646 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2647 q_vector = adapter->q_vector[q_idx];
2648 napi_disable(&q_vector->napi);
2649 }
2650 }
2651
2652 #ifdef CONFIG_IXGBE_DCB
2653 /*
2654 * ixgbe_configure_dcb - Configure DCB hardware
2655 * @adapter: ixgbe adapter struct
2656 *
2657 * This is called by the driver on open to configure the DCB hardware.
2658 * This is also called by the gennetlink interface when reconfiguring
2659 * the DCB state.
2660 */
2661 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2662 {
2663 struct ixgbe_hw *hw = &adapter->hw;
2664 u32 txdctl, vlnctrl;
2665 int i, j;
2666
2667 ixgbe_dcb_check_config(&adapter->dcb_cfg);
2668 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2669 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2670
2671 /* reconfigure the hardware */
2672 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
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 /* PThresh workaround for Tx hang with DFP enabled. */
2678 txdctl |= 32;
2679 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2680 }
2681 /* Enable VLAN tag insert/strip */
2682 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2683 if (hw->mac.type == ixgbe_mac_82598EB) {
2684 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2685 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2686 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2687 } else if (hw->mac.type == ixgbe_mac_82599EB) {
2688 vlnctrl |= IXGBE_VLNCTRL_VFE;
2689 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2690 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2691 for (i = 0; i < adapter->num_rx_queues; i++) {
2692 j = adapter->rx_ring[i]->reg_idx;
2693 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2694 vlnctrl |= IXGBE_RXDCTL_VME;
2695 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2696 }
2697 }
2698 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2699 }
2700
2701 #endif
2702 static void ixgbe_configure(struct ixgbe_adapter *adapter)
2703 {
2704 struct net_device *netdev = adapter->netdev;
2705 struct ixgbe_hw *hw = &adapter->hw;
2706 int i;
2707
2708 ixgbe_set_rx_mode(netdev);
2709
2710 ixgbe_restore_vlan(adapter);
2711 #ifdef CONFIG_IXGBE_DCB
2712 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2713 if (hw->mac.type == ixgbe_mac_82598EB)
2714 netif_set_gso_max_size(netdev, 32768);
2715 else
2716 netif_set_gso_max_size(netdev, 65536);
2717 ixgbe_configure_dcb(adapter);
2718 } else {
2719 netif_set_gso_max_size(netdev, 65536);
2720 }
2721 #else
2722 netif_set_gso_max_size(netdev, 65536);
2723 #endif
2724
2725 #ifdef IXGBE_FCOE
2726 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2727 ixgbe_configure_fcoe(adapter);
2728
2729 #endif /* IXGBE_FCOE */
2730 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2731 for (i = 0; i < adapter->num_tx_queues; i++)
2732 adapter->tx_ring[i]->atr_sample_rate =
2733 adapter->atr_sample_rate;
2734 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
2735 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
2736 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
2737 }
2738
2739 ixgbe_configure_tx(adapter);
2740 ixgbe_configure_rx(adapter);
2741 for (i = 0; i < adapter->num_rx_queues; i++)
2742 ixgbe_alloc_rx_buffers(adapter, adapter->rx_ring[i],
2743 (adapter->rx_ring[i]->count - 1));
2744 }
2745
2746 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2747 {
2748 switch (hw->phy.type) {
2749 case ixgbe_phy_sfp_avago:
2750 case ixgbe_phy_sfp_ftl:
2751 case ixgbe_phy_sfp_intel:
2752 case ixgbe_phy_sfp_unknown:
2753 case ixgbe_phy_tw_tyco:
2754 case ixgbe_phy_tw_unknown:
2755 return true;
2756 default:
2757 return false;
2758 }
2759 }
2760
2761 /**
2762 * ixgbe_sfp_link_config - set up SFP+ link
2763 * @adapter: pointer to private adapter struct
2764 **/
2765 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2766 {
2767 struct ixgbe_hw *hw = &adapter->hw;
2768
2769 if (hw->phy.multispeed_fiber) {
2770 /*
2771 * In multispeed fiber setups, the device may not have
2772 * had a physical connection when the driver loaded.
2773 * If that's the case, the initial link configuration
2774 * couldn't get the MAC into 10G or 1G mode, so we'll
2775 * never have a link status change interrupt fire.
2776 * We need to try and force an autonegotiation
2777 * session, then bring up link.
2778 */
2779 hw->mac.ops.setup_sfp(hw);
2780 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2781 schedule_work(&adapter->multispeed_fiber_task);
2782 } else {
2783 /*
2784 * Direct Attach Cu and non-multispeed fiber modules
2785 * still need to be configured properly prior to
2786 * attempting link.
2787 */
2788 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2789 schedule_work(&adapter->sfp_config_module_task);
2790 }
2791 }
2792
2793 /**
2794 * ixgbe_non_sfp_link_config - set up non-SFP+ link
2795 * @hw: pointer to private hardware struct
2796 *
2797 * Returns 0 on success, negative on failure
2798 **/
2799 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2800 {
2801 u32 autoneg;
2802 bool negotiation, link_up = false;
2803 u32 ret = IXGBE_ERR_LINK_SETUP;
2804
2805 if (hw->mac.ops.check_link)
2806 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2807
2808 if (ret)
2809 goto link_cfg_out;
2810
2811 if (hw->mac.ops.get_link_capabilities)
2812 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
2813 if (ret)
2814 goto link_cfg_out;
2815
2816 if (hw->mac.ops.setup_link)
2817 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
2818 link_cfg_out:
2819 return ret;
2820 }
2821
2822 #define IXGBE_MAX_RX_DESC_POLL 10
2823 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2824 int rxr)
2825 {
2826 int j = adapter->rx_ring[rxr]->reg_idx;
2827 int k;
2828
2829 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2830 if (IXGBE_READ_REG(&adapter->hw,
2831 IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2832 break;
2833 else
2834 msleep(1);
2835 }
2836 if (k >= IXGBE_MAX_RX_DESC_POLL) {
2837 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2838 "not set within the polling period\n", rxr);
2839 }
2840 ixgbe_release_rx_desc(&adapter->hw, adapter->rx_ring[rxr],
2841 (adapter->rx_ring[rxr]->count - 1));
2842 }
2843
2844 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2845 {
2846 struct net_device *netdev = adapter->netdev;
2847 struct ixgbe_hw *hw = &adapter->hw;
2848 int i, j = 0;
2849 int num_rx_rings = adapter->num_rx_queues;
2850 int err;
2851 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2852 u32 txdctl, rxdctl, mhadd;
2853 u32 dmatxctl;
2854 u32 gpie;
2855 u32 ctrl_ext;
2856
2857 ixgbe_get_hw_control(adapter);
2858
2859 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2860 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2861 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2862 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2863 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2864 } else {
2865 /* MSI only */
2866 gpie = 0;
2867 }
2868 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2869 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
2870 gpie |= IXGBE_GPIE_VTMODE_64;
2871 }
2872 /* XXX: to interrupt immediately for EICS writes, enable this */
2873 /* gpie |= IXGBE_GPIE_EIMEN; */
2874 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2875 }
2876
2877 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2878 /*
2879 * use EIAM to auto-mask when MSI-X interrupt is asserted
2880 * this saves a register write for every interrupt
2881 */
2882 switch (hw->mac.type) {
2883 case ixgbe_mac_82598EB:
2884 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2885 break;
2886 default:
2887 case ixgbe_mac_82599EB:
2888 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
2889 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
2890 break;
2891 }
2892 } else {
2893 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2894 * specifically only auto mask tx and rx interrupts */
2895 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2896 }
2897
2898 /* Enable fan failure interrupt if media type is copper */
2899 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2900 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2901 gpie |= IXGBE_SDP1_GPIEN;
2902 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2903 }
2904
2905 if (hw->mac.type == ixgbe_mac_82599EB) {
2906 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2907 gpie |= IXGBE_SDP1_GPIEN;
2908 gpie |= IXGBE_SDP2_GPIEN;
2909 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2910 }
2911
2912 #ifdef IXGBE_FCOE
2913 /* adjust max frame to be able to do baby jumbo for FCoE */
2914 if ((netdev->features & NETIF_F_FCOE_MTU) &&
2915 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2916 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2917
2918 #endif /* IXGBE_FCOE */
2919 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2920 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2921 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2922 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2923
2924 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2925 }
2926
2927 for (i = 0; i < adapter->num_tx_queues; i++) {
2928 j = adapter->tx_ring[i]->reg_idx;
2929 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2930 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2931 txdctl |= (8 << 16);
2932 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2933 }
2934
2935 if (hw->mac.type == ixgbe_mac_82599EB) {
2936 /* DMATXCTL.EN must be set after all Tx queue config is done */
2937 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2938 dmatxctl |= IXGBE_DMATXCTL_TE;
2939 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2940 }
2941 for (i = 0; i < adapter->num_tx_queues; i++) {
2942 j = adapter->tx_ring[i]->reg_idx;
2943 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2944 txdctl |= IXGBE_TXDCTL_ENABLE;
2945 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2946 if (hw->mac.type == ixgbe_mac_82599EB) {
2947 int wait_loop = 10;
2948 /* poll for Tx Enable ready */
2949 do {
2950 msleep(1);
2951 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2952 } while (--wait_loop &&
2953 !(txdctl & IXGBE_TXDCTL_ENABLE));
2954 if (!wait_loop)
2955 DPRINTK(DRV, ERR, "Could not enable "
2956 "Tx Queue %d\n", j);
2957 }
2958 }
2959
2960 for (i = 0; i < num_rx_rings; i++) {
2961 j = adapter->rx_ring[i]->reg_idx;
2962 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2963 /* enable PTHRESH=32 descriptors (half the internal cache)
2964 * and HTHRESH=0 descriptors (to minimize latency on fetch),
2965 * this also removes a pesky rx_no_buffer_count increment */
2966 rxdctl |= 0x0020;
2967 rxdctl |= IXGBE_RXDCTL_ENABLE;
2968 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2969 if (hw->mac.type == ixgbe_mac_82599EB)
2970 ixgbe_rx_desc_queue_enable(adapter, i);
2971 }
2972 /* enable all receives */
2973 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2974 if (hw->mac.type == ixgbe_mac_82598EB)
2975 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2976 else
2977 rxdctl |= IXGBE_RXCTRL_RXEN;
2978 hw->mac.ops.enable_rx_dma(hw, rxdctl);
2979
2980 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2981 ixgbe_configure_msix(adapter);
2982 else
2983 ixgbe_configure_msi_and_legacy(adapter);
2984
2985 clear_bit(__IXGBE_DOWN, &adapter->state);
2986 ixgbe_napi_enable_all(adapter);
2987
2988 /* clear any pending interrupts, may auto mask */
2989 IXGBE_READ_REG(hw, IXGBE_EICR);
2990
2991 ixgbe_irq_enable(adapter);
2992
2993 /*
2994 * If this adapter has a fan, check to see if we had a failure
2995 * before we enabled the interrupt.
2996 */
2997 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2998 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
2999 if (esdp & IXGBE_ESDP_SDP1)
3000 DPRINTK(DRV, CRIT,
3001 "Fan has stopped, replace the adapter\n");
3002 }
3003
3004 /*
3005 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3006 * arrived before interrupts were enabled but after probe. Such
3007 * devices wouldn't have their type identified yet. We need to
3008 * kick off the SFP+ module setup first, then try to bring up link.
3009 * If we're not hot-pluggable SFP+, we just need to configure link
3010 * and bring it up.
3011 */
3012 if (hw->phy.type == ixgbe_phy_unknown) {
3013 err = hw->phy.ops.identify(hw);
3014 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3015 /*
3016 * Take the device down and schedule the sfp tasklet
3017 * which will unregister_netdev and log it.
3018 */
3019 ixgbe_down(adapter);
3020 schedule_work(&adapter->sfp_config_module_task);
3021 return err;
3022 }
3023 }
3024
3025 if (ixgbe_is_sfp(hw)) {
3026 ixgbe_sfp_link_config(adapter);
3027 } else {
3028 err = ixgbe_non_sfp_link_config(hw);
3029 if (err)
3030 DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
3031 }
3032
3033 for (i = 0; i < adapter->num_tx_queues; i++)
3034 set_bit(__IXGBE_FDIR_INIT_DONE,
3035 &(adapter->tx_ring[i]->reinit_state));
3036
3037 /* enable transmits */
3038 netif_tx_start_all_queues(netdev);
3039
3040 /* bring the link up in the watchdog, this could race with our first
3041 * link up interrupt but shouldn't be a problem */
3042 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3043 adapter->link_check_timeout = jiffies;
3044 mod_timer(&adapter->watchdog_timer, jiffies);
3045
3046 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3047 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3048 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3049 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3050
3051 return 0;
3052 }
3053
3054 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3055 {
3056 WARN_ON(in_interrupt());
3057 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3058 msleep(1);
3059 ixgbe_down(adapter);
3060 /*
3061 * If SR-IOV enabled then wait a bit before bringing the adapter
3062 * back up to give the VFs time to respond to the reset. The
3063 * two second wait is based upon the watchdog timer cycle in
3064 * the VF driver.
3065 */
3066 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3067 msleep(2000);
3068 ixgbe_up(adapter);
3069 clear_bit(__IXGBE_RESETTING, &adapter->state);
3070 }
3071
3072 int ixgbe_up(struct ixgbe_adapter *adapter)
3073 {
3074 /* hardware has been reset, we need to reload some things */
3075 ixgbe_configure(adapter);
3076
3077 return ixgbe_up_complete(adapter);
3078 }
3079
3080 void ixgbe_reset(struct ixgbe_adapter *adapter)
3081 {
3082 struct ixgbe_hw *hw = &adapter->hw;
3083 int err;
3084
3085 err = hw->mac.ops.init_hw(hw);
3086 switch (err) {
3087 case 0:
3088 case IXGBE_ERR_SFP_NOT_PRESENT:
3089 break;
3090 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3091 dev_err(&adapter->pdev->dev, "master disable timed out\n");
3092 break;
3093 case IXGBE_ERR_EEPROM_VERSION:
3094 /* We are running on a pre-production device, log a warning */
3095 dev_warn(&adapter->pdev->dev, "This device is a pre-production "
3096 "adapter/LOM. Please be aware there may be issues "
3097 "associated with your hardware. If you are "
3098 "experiencing problems please contact your Intel or "
3099 "hardware representative who provided you with this "
3100 "hardware.\n");
3101 break;
3102 default:
3103 dev_err(&adapter->pdev->dev, "Hardware Error: %d\n", err);
3104 }
3105
3106 /* reprogram the RAR[0] in case user changed it. */
3107 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3108 IXGBE_RAH_AV);
3109 }
3110
3111 /**
3112 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3113 * @adapter: board private structure
3114 * @rx_ring: ring to free buffers from
3115 **/
3116 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
3117 struct ixgbe_ring *rx_ring)
3118 {
3119 struct pci_dev *pdev = adapter->pdev;
3120 unsigned long size;
3121 unsigned int i;
3122
3123 /* Free all the Rx ring sk_buffs */
3124
3125 for (i = 0; i < rx_ring->count; i++) {
3126 struct ixgbe_rx_buffer *rx_buffer_info;
3127
3128 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3129 if (rx_buffer_info->dma) {
3130 pci_unmap_single(pdev, rx_buffer_info->dma,
3131 rx_ring->rx_buf_len,
3132 PCI_DMA_FROMDEVICE);
3133 rx_buffer_info->dma = 0;
3134 }
3135 if (rx_buffer_info->skb) {
3136 struct sk_buff *skb = rx_buffer_info->skb;
3137 rx_buffer_info->skb = NULL;
3138 do {
3139 struct sk_buff *this = skb;
3140 if (IXGBE_RSC_CB(this)->dma) {
3141 pci_unmap_single(pdev, IXGBE_RSC_CB(this)->dma,
3142 rx_ring->rx_buf_len,
3143 PCI_DMA_FROMDEVICE);
3144 IXGBE_RSC_CB(this)->dma = 0;
3145 }
3146 skb = skb->prev;
3147 dev_kfree_skb(this);
3148 } while (skb);
3149 }
3150 if (!rx_buffer_info->page)
3151 continue;
3152 if (rx_buffer_info->page_dma) {
3153 pci_unmap_page(pdev, rx_buffer_info->page_dma,
3154 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
3155 rx_buffer_info->page_dma = 0;
3156 }
3157 put_page(rx_buffer_info->page);
3158 rx_buffer_info->page = NULL;
3159 rx_buffer_info->page_offset = 0;
3160 }
3161
3162 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3163 memset(rx_ring->rx_buffer_info, 0, size);
3164
3165 /* Zero out the descriptor ring */
3166 memset(rx_ring->desc, 0, rx_ring->size);
3167
3168 rx_ring->next_to_clean = 0;
3169 rx_ring->next_to_use = 0;
3170
3171 if (rx_ring->head)
3172 writel(0, adapter->hw.hw_addr + rx_ring->head);
3173 if (rx_ring->tail)
3174 writel(0, adapter->hw.hw_addr + rx_ring->tail);
3175 }
3176
3177 /**
3178 * ixgbe_clean_tx_ring - Free Tx Buffers
3179 * @adapter: board private structure
3180 * @tx_ring: ring to be cleaned
3181 **/
3182 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
3183 struct ixgbe_ring *tx_ring)
3184 {
3185 struct ixgbe_tx_buffer *tx_buffer_info;
3186 unsigned long size;
3187 unsigned int i;
3188
3189 /* Free all the Tx ring sk_buffs */
3190
3191 for (i = 0; i < tx_ring->count; i++) {
3192 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3193 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3194 }
3195
3196 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3197 memset(tx_ring->tx_buffer_info, 0, size);
3198
3199 /* Zero out the descriptor ring */
3200 memset(tx_ring->desc, 0, tx_ring->size);
3201
3202 tx_ring->next_to_use = 0;
3203 tx_ring->next_to_clean = 0;
3204
3205 if (tx_ring->head)
3206 writel(0, adapter->hw.hw_addr + tx_ring->head);
3207 if (tx_ring->tail)
3208 writel(0, adapter->hw.hw_addr + tx_ring->tail);
3209 }
3210
3211 /**
3212 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3213 * @adapter: board private structure
3214 **/
3215 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3216 {
3217 int i;
3218
3219 for (i = 0; i < adapter->num_rx_queues; i++)
3220 ixgbe_clean_rx_ring(adapter, adapter->rx_ring[i]);
3221 }
3222
3223 /**
3224 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3225 * @adapter: board private structure
3226 **/
3227 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3228 {
3229 int i;
3230
3231 for (i = 0; i < adapter->num_tx_queues; i++)
3232 ixgbe_clean_tx_ring(adapter, adapter->tx_ring[i]);
3233 }
3234
3235 void ixgbe_down(struct ixgbe_adapter *adapter)
3236 {
3237 struct net_device *netdev = adapter->netdev;
3238 struct ixgbe_hw *hw = &adapter->hw;
3239 u32 rxctrl;
3240 u32 txdctl;
3241 int i, j;
3242
3243 /* signal that we are down to the interrupt handler */
3244 set_bit(__IXGBE_DOWN, &adapter->state);
3245
3246 /* disable receive for all VFs and wait one second */
3247 if (adapter->num_vfs) {
3248 /* ping all the active vfs to let them know we are going down */
3249 ixgbe_ping_all_vfs(adapter);
3250
3251 /* Disable all VFTE/VFRE TX/RX */
3252 ixgbe_disable_tx_rx(adapter);
3253
3254 /* Mark all the VFs as inactive */
3255 for (i = 0 ; i < adapter->num_vfs; i++)
3256 adapter->vfinfo[i].clear_to_send = 0;
3257 }
3258
3259 /* disable receives */
3260 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3261 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3262
3263 netif_tx_disable(netdev);
3264
3265 IXGBE_WRITE_FLUSH(hw);
3266 msleep(10);
3267
3268 netif_tx_stop_all_queues(netdev);
3269
3270 ixgbe_irq_disable(adapter);
3271
3272 ixgbe_napi_disable_all(adapter);
3273
3274 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3275 del_timer_sync(&adapter->sfp_timer);
3276 del_timer_sync(&adapter->watchdog_timer);
3277 cancel_work_sync(&adapter->watchdog_task);
3278
3279 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3280 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3281 cancel_work_sync(&adapter->fdir_reinit_task);
3282
3283 /* disable transmits in the hardware now that interrupts are off */
3284 for (i = 0; i < adapter->num_tx_queues; i++) {
3285 j = adapter->tx_ring[i]->reg_idx;
3286 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3287 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3288 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3289 }
3290 /* Disable the Tx DMA engine on 82599 */
3291 if (hw->mac.type == ixgbe_mac_82599EB)
3292 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3293 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3294 ~IXGBE_DMATXCTL_TE));
3295
3296 netif_carrier_off(netdev);
3297
3298 /* clear n-tuple filters that are cached */
3299 ethtool_ntuple_flush(netdev);
3300
3301 if (!pci_channel_offline(adapter->pdev))
3302 ixgbe_reset(adapter);
3303 ixgbe_clean_all_tx_rings(adapter);
3304 ixgbe_clean_all_rx_rings(adapter);
3305
3306 #ifdef CONFIG_IXGBE_DCA
3307 /* since we reset the hardware DCA settings were cleared */
3308 ixgbe_setup_dca(adapter);
3309 #endif
3310 }
3311
3312 /**
3313 * ixgbe_poll - NAPI Rx polling callback
3314 * @napi: structure for representing this polling device
3315 * @budget: how many packets driver is allowed to clean
3316 *
3317 * This function is used for legacy and MSI, NAPI mode
3318 **/
3319 static int ixgbe_poll(struct napi_struct *napi, int budget)
3320 {
3321 struct ixgbe_q_vector *q_vector =
3322 container_of(napi, struct ixgbe_q_vector, napi);
3323 struct ixgbe_adapter *adapter = q_vector->adapter;
3324 int tx_clean_complete, work_done = 0;
3325
3326 #ifdef CONFIG_IXGBE_DCA
3327 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3328 ixgbe_update_tx_dca(adapter, adapter->tx_ring[0]);
3329 ixgbe_update_rx_dca(adapter, adapter->rx_ring[0]);
3330 }
3331 #endif
3332
3333 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3334 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3335
3336 if (!tx_clean_complete)
3337 work_done = budget;
3338
3339 /* If budget not fully consumed, exit the polling mode */
3340 if (work_done < budget) {
3341 napi_complete(napi);
3342 if (adapter->rx_itr_setting & 1)
3343 ixgbe_set_itr(adapter);
3344 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3345 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3346 }
3347 return work_done;
3348 }
3349
3350 /**
3351 * ixgbe_tx_timeout - Respond to a Tx Hang
3352 * @netdev: network interface device structure
3353 **/
3354 static void ixgbe_tx_timeout(struct net_device *netdev)
3355 {
3356 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3357
3358 /* Do the reset outside of interrupt context */
3359 schedule_work(&adapter->reset_task);
3360 }
3361
3362 static void ixgbe_reset_task(struct work_struct *work)
3363 {
3364 struct ixgbe_adapter *adapter;
3365 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3366
3367 /* If we're already down or resetting, just bail */
3368 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3369 test_bit(__IXGBE_RESETTING, &adapter->state))
3370 return;
3371
3372 adapter->tx_timeout_count++;
3373
3374 ixgbe_reinit_locked(adapter);
3375 }
3376
3377 #ifdef CONFIG_IXGBE_DCB
3378 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3379 {
3380 bool ret = false;
3381 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3382
3383 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3384 return ret;
3385
3386 f->mask = 0x7 << 3;
3387 adapter->num_rx_queues = f->indices;
3388 adapter->num_tx_queues = f->indices;
3389 ret = true;
3390
3391 return ret;
3392 }
3393 #endif
3394
3395 /**
3396 * ixgbe_set_rss_queues: Allocate queues for RSS
3397 * @adapter: board private structure to initialize
3398 *
3399 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3400 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3401 *
3402 **/
3403 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3404 {
3405 bool ret = false;
3406 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3407
3408 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3409 f->mask = 0xF;
3410 adapter->num_rx_queues = f->indices;
3411 adapter->num_tx_queues = f->indices;
3412 ret = true;
3413 } else {
3414 ret = false;
3415 }
3416
3417 return ret;
3418 }
3419
3420 /**
3421 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3422 * @adapter: board private structure to initialize
3423 *
3424 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3425 * to the original CPU that initiated the Tx session. This runs in addition
3426 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3427 * Rx load across CPUs using RSS.
3428 *
3429 **/
3430 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3431 {
3432 bool ret = false;
3433 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3434
3435 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3436 f_fdir->mask = 0;
3437
3438 /* Flow Director must have RSS enabled */
3439 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3440 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3441 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3442 adapter->num_tx_queues = f_fdir->indices;
3443 adapter->num_rx_queues = f_fdir->indices;
3444 ret = true;
3445 } else {
3446 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3447 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3448 }
3449 return ret;
3450 }
3451
3452 #ifdef IXGBE_FCOE
3453 /**
3454 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3455 * @adapter: board private structure to initialize
3456 *
3457 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3458 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3459 * rx queues out of the max number of rx queues, instead, it is used as the
3460 * index of the first rx queue used by FCoE.
3461 *
3462 **/
3463 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3464 {
3465 bool ret = false;
3466 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3467
3468 f->indices = min((int)num_online_cpus(), f->indices);
3469 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3470 adapter->num_rx_queues = 1;
3471 adapter->num_tx_queues = 1;
3472 #ifdef CONFIG_IXGBE_DCB
3473 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3474 DPRINTK(PROBE, INFO, "FCoE enabled with DCB \n");
3475 ixgbe_set_dcb_queues(adapter);
3476 }
3477 #endif
3478 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3479 DPRINTK(PROBE, INFO, "FCoE enabled with RSS \n");
3480 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3481 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3482 ixgbe_set_fdir_queues(adapter);
3483 else
3484 ixgbe_set_rss_queues(adapter);
3485 }
3486 /* adding FCoE rx rings to the end */
3487 f->mask = adapter->num_rx_queues;
3488 adapter->num_rx_queues += f->indices;
3489 adapter->num_tx_queues += f->indices;
3490
3491 ret = true;
3492 }
3493
3494 return ret;
3495 }
3496
3497 #endif /* IXGBE_FCOE */
3498 /**
3499 * ixgbe_set_sriov_queues: Allocate queues for IOV use
3500 * @adapter: board private structure to initialize
3501 *
3502 * IOV doesn't actually use anything, so just NAK the
3503 * request for now and let the other queue routines
3504 * figure out what to do.
3505 */
3506 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
3507 {
3508 return false;
3509 }
3510
3511 /*
3512 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3513 * @adapter: board private structure to initialize
3514 *
3515 * This is the top level queue allocation routine. The order here is very
3516 * important, starting with the "most" number of features turned on at once,
3517 * and ending with the smallest set of features. This way large combinations
3518 * can be allocated if they're turned on, and smaller combinations are the
3519 * fallthrough conditions.
3520 *
3521 **/
3522 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
3523 {
3524 /* Start with base case */
3525 adapter->num_rx_queues = 1;
3526 adapter->num_tx_queues = 1;
3527 adapter->num_rx_pools = adapter->num_rx_queues;
3528 adapter->num_rx_queues_per_pool = 1;
3529
3530 if (ixgbe_set_sriov_queues(adapter))
3531 return;
3532
3533 #ifdef IXGBE_FCOE
3534 if (ixgbe_set_fcoe_queues(adapter))
3535 goto done;
3536
3537 #endif /* IXGBE_FCOE */
3538 #ifdef CONFIG_IXGBE_DCB
3539 if (ixgbe_set_dcb_queues(adapter))
3540 goto done;
3541
3542 #endif
3543 if (ixgbe_set_fdir_queues(adapter))
3544 goto done;
3545
3546 if (ixgbe_set_rss_queues(adapter))
3547 goto done;
3548
3549 /* fallback to base case */
3550 adapter->num_rx_queues = 1;
3551 adapter->num_tx_queues = 1;
3552
3553 done:
3554 /* Notify the stack of the (possibly) reduced Tx Queue count. */
3555 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
3556 }
3557
3558 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
3559 int vectors)
3560 {
3561 int err, vector_threshold;
3562
3563 /* We'll want at least 3 (vector_threshold):
3564 * 1) TxQ[0] Cleanup
3565 * 2) RxQ[0] Cleanup
3566 * 3) Other (Link Status Change, etc.)
3567 * 4) TCP Timer (optional)
3568 */
3569 vector_threshold = MIN_MSIX_COUNT;
3570
3571 /* The more we get, the more we will assign to Tx/Rx Cleanup
3572 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
3573 * Right now, we simply care about how many we'll get; we'll
3574 * set them up later while requesting irq's.
3575 */
3576 while (vectors >= vector_threshold) {
3577 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
3578 vectors);
3579 if (!err) /* Success in acquiring all requested vectors. */
3580 break;
3581 else if (err < 0)
3582 vectors = 0; /* Nasty failure, quit now */
3583 else /* err == number of vectors we should try again with */
3584 vectors = err;
3585 }
3586
3587 if (vectors < vector_threshold) {
3588 /* Can't allocate enough MSI-X interrupts? Oh well.
3589 * This just means we'll go with either a single MSI
3590 * vector or fall back to legacy interrupts.
3591 */
3592 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
3593 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3594 kfree(adapter->msix_entries);
3595 adapter->msix_entries = NULL;
3596 } else {
3597 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
3598 /*
3599 * Adjust for only the vectors we'll use, which is minimum
3600 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
3601 * vectors we were allocated.
3602 */
3603 adapter->num_msix_vectors = min(vectors,
3604 adapter->max_msix_q_vectors + NON_Q_VECTORS);
3605 }
3606 }
3607
3608 /**
3609 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
3610 * @adapter: board private structure to initialize
3611 *
3612 * Cache the descriptor ring offsets for RSS to the assigned rings.
3613 *
3614 **/
3615 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
3616 {
3617 int i;
3618 bool ret = false;
3619
3620 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3621 for (i = 0; i < adapter->num_rx_queues; i++)
3622 adapter->rx_ring[i]->reg_idx = i;
3623 for (i = 0; i < adapter->num_tx_queues; i++)
3624 adapter->tx_ring[i]->reg_idx = i;
3625 ret = true;
3626 } else {
3627 ret = false;
3628 }
3629
3630 return ret;
3631 }
3632
3633 #ifdef CONFIG_IXGBE_DCB
3634 /**
3635 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
3636 * @adapter: board private structure to initialize
3637 *
3638 * Cache the descriptor ring offsets for DCB to the assigned rings.
3639 *
3640 **/
3641 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
3642 {
3643 int i;
3644 bool ret = false;
3645 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
3646
3647 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3648 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3649 /* the number of queues is assumed to be symmetric */
3650 for (i = 0; i < dcb_i; i++) {
3651 adapter->rx_ring[i]->reg_idx = i << 3;
3652 adapter->tx_ring[i]->reg_idx = i << 2;
3653 }
3654 ret = true;
3655 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
3656 if (dcb_i == 8) {
3657 /*
3658 * Tx TC0 starts at: descriptor queue 0
3659 * Tx TC1 starts at: descriptor queue 32
3660 * Tx TC2 starts at: descriptor queue 64
3661 * Tx TC3 starts at: descriptor queue 80
3662 * Tx TC4 starts at: descriptor queue 96
3663 * Tx TC5 starts at: descriptor queue 104
3664 * Tx TC6 starts at: descriptor queue 112
3665 * Tx TC7 starts at: descriptor queue 120
3666 *
3667 * Rx TC0-TC7 are offset by 16 queues each
3668 */
3669 for (i = 0; i < 3; i++) {
3670 adapter->tx_ring[i]->reg_idx = i << 5;
3671 adapter->rx_ring[i]->reg_idx = i << 4;
3672 }
3673 for ( ; i < 5; i++) {
3674 adapter->tx_ring[i]->reg_idx =
3675 ((i + 2) << 4);
3676 adapter->rx_ring[i]->reg_idx = i << 4;
3677 }
3678 for ( ; i < dcb_i; i++) {
3679 adapter->tx_ring[i]->reg_idx =
3680 ((i + 8) << 3);
3681 adapter->rx_ring[i]->reg_idx = i << 4;
3682 }
3683
3684 ret = true;
3685 } else if (dcb_i == 4) {
3686 /*
3687 * Tx TC0 starts at: descriptor queue 0
3688 * Tx TC1 starts at: descriptor queue 64
3689 * Tx TC2 starts at: descriptor queue 96
3690 * Tx TC3 starts at: descriptor queue 112
3691 *
3692 * Rx TC0-TC3 are offset by 32 queues each
3693 */
3694 adapter->tx_ring[0]->reg_idx = 0;
3695 adapter->tx_ring[1]->reg_idx = 64;
3696 adapter->tx_ring[2]->reg_idx = 96;
3697 adapter->tx_ring[3]->reg_idx = 112;
3698 for (i = 0 ; i < dcb_i; i++)
3699 adapter->rx_ring[i]->reg_idx = i << 5;
3700
3701 ret = true;
3702 } else {
3703 ret = false;
3704 }
3705 } else {
3706 ret = false;
3707 }
3708 } else {
3709 ret = false;
3710 }
3711
3712 return ret;
3713 }
3714 #endif
3715
3716 /**
3717 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
3718 * @adapter: board private structure to initialize
3719 *
3720 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
3721 *
3722 **/
3723 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
3724 {
3725 int i;
3726 bool ret = false;
3727
3728 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3729 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3730 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
3731 for (i = 0; i < adapter->num_rx_queues; i++)
3732 adapter->rx_ring[i]->reg_idx = i;
3733 for (i = 0; i < adapter->num_tx_queues; i++)
3734 adapter->tx_ring[i]->reg_idx = i;
3735 ret = true;
3736 }
3737
3738 return ret;
3739 }
3740
3741 #ifdef IXGBE_FCOE
3742 /**
3743 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
3744 * @adapter: board private structure to initialize
3745 *
3746 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
3747 *
3748 */
3749 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
3750 {
3751 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
3752 bool ret = false;
3753 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3754
3755 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3756 #ifdef CONFIG_IXGBE_DCB
3757 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3758 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
3759
3760 ixgbe_cache_ring_dcb(adapter);
3761 /* find out queues in TC for FCoE */
3762 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
3763 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
3764 /*
3765 * In 82599, the number of Tx queues for each traffic
3766 * class for both 8-TC and 4-TC modes are:
3767 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
3768 * 8 TCs: 32 32 16 16 8 8 8 8
3769 * 4 TCs: 64 64 32 32
3770 * We have max 8 queues for FCoE, where 8 the is
3771 * FCoE redirection table size. If TC for FCoE is
3772 * less than or equal to TC3, we have enough queues
3773 * to add max of 8 queues for FCoE, so we start FCoE
3774 * tx descriptor from the next one, i.e., reg_idx + 1.
3775 * If TC for FCoE is above TC3, implying 8 TC mode,
3776 * and we need 8 for FCoE, we have to take all queues
3777 * in that traffic class for FCoE.
3778 */
3779 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
3780 fcoe_tx_i--;
3781 }
3782 #endif /* CONFIG_IXGBE_DCB */
3783 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3784 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3785 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3786 ixgbe_cache_ring_fdir(adapter);
3787 else
3788 ixgbe_cache_ring_rss(adapter);
3789
3790 fcoe_rx_i = f->mask;
3791 fcoe_tx_i = f->mask;
3792 }
3793 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
3794 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
3795 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
3796 }
3797 ret = true;
3798 }
3799 return ret;
3800 }
3801
3802 #endif /* IXGBE_FCOE */
3803 /**
3804 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
3805 * @adapter: board private structure to initialize
3806 *
3807 * SR-IOV doesn't use any descriptor rings but changes the default if
3808 * no other mapping is used.
3809 *
3810 */
3811 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
3812 {
3813 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
3814 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
3815 if (adapter->num_vfs)
3816 return true;
3817 else
3818 return false;
3819 }
3820
3821 /**
3822 * ixgbe_cache_ring_register - Descriptor ring to register mapping
3823 * @adapter: board private structure to initialize
3824 *
3825 * Once we know the feature-set enabled for the device, we'll cache
3826 * the register offset the descriptor ring is assigned to.
3827 *
3828 * Note, the order the various feature calls is important. It must start with
3829 * the "most" features enabled at the same time, then trickle down to the
3830 * least amount of features turned on at once.
3831 **/
3832 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
3833 {
3834 /* start with default case */
3835 adapter->rx_ring[0]->reg_idx = 0;
3836 adapter->tx_ring[0]->reg_idx = 0;
3837
3838 if (ixgbe_cache_ring_sriov(adapter))
3839 return;
3840
3841 #ifdef IXGBE_FCOE
3842 if (ixgbe_cache_ring_fcoe(adapter))
3843 return;
3844
3845 #endif /* IXGBE_FCOE */
3846 #ifdef CONFIG_IXGBE_DCB
3847 if (ixgbe_cache_ring_dcb(adapter))
3848 return;
3849
3850 #endif
3851 if (ixgbe_cache_ring_fdir(adapter))
3852 return;
3853
3854 if (ixgbe_cache_ring_rss(adapter))
3855 return;
3856 }
3857
3858 /**
3859 * ixgbe_alloc_queues - Allocate memory for all rings
3860 * @adapter: board private structure to initialize
3861 *
3862 * We allocate one ring per queue at run-time since we don't know the
3863 * number of queues at compile-time. The polling_netdev array is
3864 * intended for Multiqueue, but should work fine with a single queue.
3865 **/
3866 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
3867 {
3868 int i;
3869 int orig_node = adapter->node;
3870
3871 for (i = 0; i < adapter->num_tx_queues; i++) {
3872 struct ixgbe_ring *ring = adapter->tx_ring[i];
3873 if (orig_node == -1) {
3874 int cur_node = next_online_node(adapter->node);
3875 if (cur_node == MAX_NUMNODES)
3876 cur_node = first_online_node;
3877 adapter->node = cur_node;
3878 }
3879 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
3880 adapter->node);
3881 if (!ring)
3882 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
3883 if (!ring)
3884 goto err_tx_ring_allocation;
3885 ring->count = adapter->tx_ring_count;
3886 ring->queue_index = i;
3887 ring->numa_node = adapter->node;
3888
3889 adapter->tx_ring[i] = ring;
3890 }
3891
3892 /* Restore the adapter's original node */
3893 adapter->node = orig_node;
3894
3895 for (i = 0; i < adapter->num_rx_queues; i++) {
3896 struct ixgbe_ring *ring = adapter->rx_ring[i];
3897 if (orig_node == -1) {
3898 int cur_node = next_online_node(adapter->node);
3899 if (cur_node == MAX_NUMNODES)
3900 cur_node = first_online_node;
3901 adapter->node = cur_node;
3902 }
3903 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
3904 adapter->node);
3905 if (!ring)
3906 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
3907 if (!ring)
3908 goto err_rx_ring_allocation;
3909 ring->count = adapter->rx_ring_count;
3910 ring->queue_index = i;
3911 ring->numa_node = adapter->node;
3912
3913 adapter->rx_ring[i] = ring;
3914 }
3915
3916 /* Restore the adapter's original node */
3917 adapter->node = orig_node;
3918
3919 ixgbe_cache_ring_register(adapter);
3920
3921 return 0;
3922
3923 err_rx_ring_allocation:
3924 for (i = 0; i < adapter->num_tx_queues; i++)
3925 kfree(adapter->tx_ring[i]);
3926 err_tx_ring_allocation:
3927 return -ENOMEM;
3928 }
3929
3930 /**
3931 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
3932 * @adapter: board private structure to initialize
3933 *
3934 * Attempt to configure the interrupts using the best available
3935 * capabilities of the hardware and the kernel.
3936 **/
3937 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
3938 {
3939 struct ixgbe_hw *hw = &adapter->hw;
3940 int err = 0;
3941 int vector, v_budget;
3942
3943 /*
3944 * It's easy to be greedy for MSI-X vectors, but it really
3945 * doesn't do us much good if we have a lot more vectors
3946 * than CPU's. So let's be conservative and only ask for
3947 * (roughly) the same number of vectors as there are CPU's.
3948 */
3949 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
3950 (int)num_online_cpus()) + NON_Q_VECTORS;
3951
3952 /*
3953 * At the same time, hardware can only support a maximum of
3954 * hw.mac->max_msix_vectors vectors. With features
3955 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
3956 * descriptor queues supported by our device. Thus, we cap it off in
3957 * those rare cases where the cpu count also exceeds our vector limit.
3958 */
3959 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
3960
3961 /* A failure in MSI-X entry allocation isn't fatal, but it does
3962 * mean we disable MSI-X capabilities of the adapter. */
3963 adapter->msix_entries = kcalloc(v_budget,
3964 sizeof(struct msix_entry), GFP_KERNEL);
3965 if (adapter->msix_entries) {
3966 for (vector = 0; vector < v_budget; vector++)
3967 adapter->msix_entries[vector].entry = vector;
3968
3969 ixgbe_acquire_msix_vectors(adapter, v_budget);
3970
3971 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3972 goto out;
3973 }
3974
3975 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
3976 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
3977 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3978 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3979 adapter->atr_sample_rate = 0;
3980 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3981 ixgbe_disable_sriov(adapter);
3982
3983 ixgbe_set_num_queues(adapter);
3984
3985 err = pci_enable_msi(adapter->pdev);
3986 if (!err) {
3987 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
3988 } else {
3989 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
3990 "falling back to legacy. Error: %d\n", err);
3991 /* reset err */
3992 err = 0;
3993 }
3994
3995 out:
3996 return err;
3997 }
3998
3999 /**
4000 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4001 * @adapter: board private structure to initialize
4002 *
4003 * We allocate one q_vector per queue interrupt. If allocation fails we
4004 * return -ENOMEM.
4005 **/
4006 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4007 {
4008 int q_idx, num_q_vectors;
4009 struct ixgbe_q_vector *q_vector;
4010 int napi_vectors;
4011 int (*poll)(struct napi_struct *, int);
4012
4013 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4014 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4015 napi_vectors = adapter->num_rx_queues;
4016 poll = &ixgbe_clean_rxtx_many;
4017 } else {
4018 num_q_vectors = 1;
4019 napi_vectors = 1;
4020 poll = &ixgbe_poll;
4021 }
4022
4023 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4024 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4025 GFP_KERNEL, adapter->node);
4026 if (!q_vector)
4027 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4028 GFP_KERNEL);
4029 if (!q_vector)
4030 goto err_out;
4031 q_vector->adapter = adapter;
4032 if (q_vector->txr_count && !q_vector->rxr_count)
4033 q_vector->eitr = adapter->tx_eitr_param;
4034 else
4035 q_vector->eitr = adapter->rx_eitr_param;
4036 q_vector->v_idx = q_idx;
4037 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4038 adapter->q_vector[q_idx] = q_vector;
4039 }
4040
4041 return 0;
4042
4043 err_out:
4044 while (q_idx) {
4045 q_idx--;
4046 q_vector = adapter->q_vector[q_idx];
4047 netif_napi_del(&q_vector->napi);
4048 kfree(q_vector);
4049 adapter->q_vector[q_idx] = NULL;
4050 }
4051 return -ENOMEM;
4052 }
4053
4054 /**
4055 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4056 * @adapter: board private structure to initialize
4057 *
4058 * This function frees the memory allocated to the q_vectors. In addition if
4059 * NAPI is enabled it will delete any references to the NAPI struct prior
4060 * to freeing the q_vector.
4061 **/
4062 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4063 {
4064 int q_idx, num_q_vectors;
4065
4066 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4067 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4068 else
4069 num_q_vectors = 1;
4070
4071 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4072 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4073 adapter->q_vector[q_idx] = NULL;
4074 netif_napi_del(&q_vector->napi);
4075 kfree(q_vector);
4076 }
4077 }
4078
4079 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4080 {
4081 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4082 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4083 pci_disable_msix(adapter->pdev);
4084 kfree(adapter->msix_entries);
4085 adapter->msix_entries = NULL;
4086 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4087 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4088 pci_disable_msi(adapter->pdev);
4089 }
4090 return;
4091 }
4092
4093 /**
4094 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4095 * @adapter: board private structure to initialize
4096 *
4097 * We determine which interrupt scheme to use based on...
4098 * - Kernel support (MSI, MSI-X)
4099 * - which can be user-defined (via MODULE_PARAM)
4100 * - Hardware queue count (num_*_queues)
4101 * - defined by miscellaneous hardware support/features (RSS, etc.)
4102 **/
4103 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4104 {
4105 int err;
4106
4107 /* Number of supported queues */
4108 ixgbe_set_num_queues(adapter);
4109
4110 err = ixgbe_set_interrupt_capability(adapter);
4111 if (err) {
4112 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
4113 goto err_set_interrupt;
4114 }
4115
4116 err = ixgbe_alloc_q_vectors(adapter);
4117 if (err) {
4118 DPRINTK(PROBE, ERR, "Unable to allocate memory for queue "
4119 "vectors\n");
4120 goto err_alloc_q_vectors;
4121 }
4122
4123 err = ixgbe_alloc_queues(adapter);
4124 if (err) {
4125 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
4126 goto err_alloc_queues;
4127 }
4128
4129 DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
4130 "Tx Queue count = %u\n",
4131 (adapter->num_rx_queues > 1) ? "Enabled" :
4132 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
4133
4134 set_bit(__IXGBE_DOWN, &adapter->state);
4135
4136 return 0;
4137
4138 err_alloc_queues:
4139 ixgbe_free_q_vectors(adapter);
4140 err_alloc_q_vectors:
4141 ixgbe_reset_interrupt_capability(adapter);
4142 err_set_interrupt:
4143 return err;
4144 }
4145
4146 /**
4147 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4148 * @adapter: board private structure to clear interrupt scheme on
4149 *
4150 * We go through and clear interrupt specific resources and reset the structure
4151 * to pre-load conditions
4152 **/
4153 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4154 {
4155 int i;
4156
4157 for (i = 0; i < adapter->num_tx_queues; i++) {
4158 kfree(adapter->tx_ring[i]);
4159 adapter->tx_ring[i] = NULL;
4160 }
4161 for (i = 0; i < adapter->num_rx_queues; i++) {
4162 kfree(adapter->rx_ring[i]);
4163 adapter->rx_ring[i] = NULL;
4164 }
4165
4166 ixgbe_free_q_vectors(adapter);
4167 ixgbe_reset_interrupt_capability(adapter);
4168 }
4169
4170 /**
4171 * ixgbe_sfp_timer - worker thread to find a missing module
4172 * @data: pointer to our adapter struct
4173 **/
4174 static void ixgbe_sfp_timer(unsigned long data)
4175 {
4176 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4177
4178 /*
4179 * Do the sfp_timer outside of interrupt context due to the
4180 * delays that sfp+ detection requires
4181 */
4182 schedule_work(&adapter->sfp_task);
4183 }
4184
4185 /**
4186 * ixgbe_sfp_task - worker thread to find a missing module
4187 * @work: pointer to work_struct containing our data
4188 **/
4189 static void ixgbe_sfp_task(struct work_struct *work)
4190 {
4191 struct ixgbe_adapter *adapter = container_of(work,
4192 struct ixgbe_adapter,
4193 sfp_task);
4194 struct ixgbe_hw *hw = &adapter->hw;
4195
4196 if ((hw->phy.type == ixgbe_phy_nl) &&
4197 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4198 s32 ret = hw->phy.ops.identify_sfp(hw);
4199 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4200 goto reschedule;
4201 ret = hw->phy.ops.reset(hw);
4202 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4203 dev_err(&adapter->pdev->dev, "failed to initialize "
4204 "because an unsupported SFP+ module type "
4205 "was detected.\n"
4206 "Reload the driver after installing a "
4207 "supported module.\n");
4208 unregister_netdev(adapter->netdev);
4209 } else {
4210 DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
4211 hw->phy.sfp_type);
4212 }
4213 /* don't need this routine any more */
4214 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4215 }
4216 return;
4217 reschedule:
4218 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4219 mod_timer(&adapter->sfp_timer,
4220 round_jiffies(jiffies + (2 * HZ)));
4221 }
4222
4223 /**
4224 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4225 * @adapter: board private structure to initialize
4226 *
4227 * ixgbe_sw_init initializes the Adapter private data structure.
4228 * Fields are initialized based on PCI device information and
4229 * OS network device settings (MTU size).
4230 **/
4231 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4232 {
4233 struct ixgbe_hw *hw = &adapter->hw;
4234 struct pci_dev *pdev = adapter->pdev;
4235 struct net_device *dev = adapter->netdev;
4236 unsigned int rss;
4237 #ifdef CONFIG_IXGBE_DCB
4238 int j;
4239 struct tc_configuration *tc;
4240 #endif
4241
4242 /* PCI config space info */
4243
4244 hw->vendor_id = pdev->vendor;
4245 hw->device_id = pdev->device;
4246 hw->revision_id = pdev->revision;
4247 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4248 hw->subsystem_device_id = pdev->subsystem_device;
4249
4250 /* Set capability flags */
4251 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4252 adapter->ring_feature[RING_F_RSS].indices = rss;
4253 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4254 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4255 if (hw->mac.type == ixgbe_mac_82598EB) {
4256 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4257 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4258 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4259 } else if (hw->mac.type == ixgbe_mac_82599EB) {
4260 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4261 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4262 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4263 if (dev->features & NETIF_F_NTUPLE) {
4264 /* Flow Director perfect filter enabled */
4265 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4266 adapter->atr_sample_rate = 0;
4267 spin_lock_init(&adapter->fdir_perfect_lock);
4268 } else {
4269 /* Flow Director hash filters enabled */
4270 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4271 adapter->atr_sample_rate = 20;
4272 }
4273 adapter->ring_feature[RING_F_FDIR].indices =
4274 IXGBE_MAX_FDIR_INDICES;
4275 adapter->fdir_pballoc = 0;
4276 #ifdef IXGBE_FCOE
4277 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4278 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4279 adapter->ring_feature[RING_F_FCOE].indices = 0;
4280 #ifdef CONFIG_IXGBE_DCB
4281 /* Default traffic class to use for FCoE */
4282 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4283 #endif
4284 #endif /* IXGBE_FCOE */
4285 }
4286
4287 #ifdef CONFIG_IXGBE_DCB
4288 /* Configure DCB traffic classes */
4289 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4290 tc = &adapter->dcb_cfg.tc_config[j];
4291 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4292 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4293 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4294 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4295 tc->dcb_pfc = pfc_disabled;
4296 }
4297 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4298 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4299 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4300 adapter->dcb_cfg.pfc_mode_enable = false;
4301 adapter->dcb_cfg.round_robin_enable = false;
4302 adapter->dcb_set_bitmap = 0x00;
4303 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4304 adapter->ring_feature[RING_F_DCB].indices);
4305
4306 #endif
4307
4308 /* default flow control settings */
4309 hw->fc.requested_mode = ixgbe_fc_full;
4310 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4311 #ifdef CONFIG_DCB
4312 adapter->last_lfc_mode = hw->fc.current_mode;
4313 #endif
4314 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
4315 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
4316 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4317 hw->fc.send_xon = true;
4318 hw->fc.disable_fc_autoneg = false;
4319
4320 /* enable itr by default in dynamic mode */
4321 adapter->rx_itr_setting = 1;
4322 adapter->rx_eitr_param = 20000;
4323 adapter->tx_itr_setting = 1;
4324 adapter->tx_eitr_param = 10000;
4325
4326 /* set defaults for eitr in MegaBytes */
4327 adapter->eitr_low = 10;
4328 adapter->eitr_high = 20;
4329
4330 /* set default ring sizes */
4331 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4332 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4333
4334 /* initialize eeprom parameters */
4335 if (ixgbe_init_eeprom_params_generic(hw)) {
4336 dev_err(&pdev->dev, "EEPROM initialization failed\n");
4337 return -EIO;
4338 }
4339
4340 /* enable rx csum by default */
4341 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4342
4343 /* get assigned NUMA node */
4344 adapter->node = dev_to_node(&pdev->dev);
4345
4346 set_bit(__IXGBE_DOWN, &adapter->state);
4347
4348 return 0;
4349 }
4350
4351 /**
4352 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4353 * @adapter: board private structure
4354 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4355 *
4356 * Return 0 on success, negative on failure
4357 **/
4358 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4359 struct ixgbe_ring *tx_ring)
4360 {
4361 struct pci_dev *pdev = adapter->pdev;
4362 int size;
4363
4364 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4365 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4366 if (!tx_ring->tx_buffer_info)
4367 tx_ring->tx_buffer_info = vmalloc(size);
4368 if (!tx_ring->tx_buffer_info)
4369 goto err;
4370 memset(tx_ring->tx_buffer_info, 0, size);
4371
4372 /* round up to nearest 4K */
4373 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4374 tx_ring->size = ALIGN(tx_ring->size, 4096);
4375
4376 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
4377 &tx_ring->dma);
4378 if (!tx_ring->desc)
4379 goto err;
4380
4381 tx_ring->next_to_use = 0;
4382 tx_ring->next_to_clean = 0;
4383 tx_ring->work_limit = tx_ring->count;
4384 return 0;
4385
4386 err:
4387 vfree(tx_ring->tx_buffer_info);
4388 tx_ring->tx_buffer_info = NULL;
4389 DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
4390 "descriptor ring\n");
4391 return -ENOMEM;
4392 }
4393
4394 /**
4395 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4396 * @adapter: board private structure
4397 *
4398 * If this function returns with an error, then it's possible one or
4399 * more of the rings is populated (while the rest are not). It is the
4400 * callers duty to clean those orphaned rings.
4401 *
4402 * Return 0 on success, negative on failure
4403 **/
4404 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4405 {
4406 int i, err = 0;
4407
4408 for (i = 0; i < adapter->num_tx_queues; i++) {
4409 err = ixgbe_setup_tx_resources(adapter, adapter->tx_ring[i]);
4410 if (!err)
4411 continue;
4412 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
4413 break;
4414 }
4415
4416 return err;
4417 }
4418
4419 /**
4420 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4421 * @adapter: board private structure
4422 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4423 *
4424 * Returns 0 on success, negative on failure
4425 **/
4426 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4427 struct ixgbe_ring *rx_ring)
4428 {
4429 struct pci_dev *pdev = adapter->pdev;
4430 int size;
4431
4432 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4433 rx_ring->rx_buffer_info = vmalloc_node(size, adapter->node);
4434 if (!rx_ring->rx_buffer_info)
4435 rx_ring->rx_buffer_info = vmalloc(size);
4436 if (!rx_ring->rx_buffer_info) {
4437 DPRINTK(PROBE, ERR,
4438 "vmalloc allocation failed for the rx desc ring\n");
4439 goto alloc_failed;
4440 }
4441 memset(rx_ring->rx_buffer_info, 0, size);
4442
4443 /* Round up to nearest 4K */
4444 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4445 rx_ring->size = ALIGN(rx_ring->size, 4096);
4446
4447 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
4448
4449 if (!rx_ring->desc) {
4450 DPRINTK(PROBE, ERR,
4451 "Memory allocation failed for the rx desc ring\n");
4452 vfree(rx_ring->rx_buffer_info);
4453 goto alloc_failed;
4454 }
4455
4456 rx_ring->next_to_clean = 0;
4457 rx_ring->next_to_use = 0;
4458
4459 return 0;
4460
4461 alloc_failed:
4462 return -ENOMEM;
4463 }
4464
4465 /**
4466 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4467 * @adapter: board private structure
4468 *
4469 * If this function returns with an error, then it's possible one or
4470 * more of the rings is populated (while the rest are not). It is the
4471 * callers duty to clean those orphaned rings.
4472 *
4473 * Return 0 on success, negative on failure
4474 **/
4475
4476 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4477 {
4478 int i, err = 0;
4479
4480 for (i = 0; i < adapter->num_rx_queues; i++) {
4481 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
4482 if (!err)
4483 continue;
4484 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
4485 break;
4486 }
4487
4488 return err;
4489 }
4490
4491 /**
4492 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4493 * @adapter: board private structure
4494 * @tx_ring: Tx descriptor ring for a specific queue
4495 *
4496 * Free all transmit software resources
4497 **/
4498 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4499 struct ixgbe_ring *tx_ring)
4500 {
4501 struct pci_dev *pdev = adapter->pdev;
4502
4503 ixgbe_clean_tx_ring(adapter, tx_ring);
4504
4505 vfree(tx_ring->tx_buffer_info);
4506 tx_ring->tx_buffer_info = NULL;
4507
4508 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
4509
4510 tx_ring->desc = NULL;
4511 }
4512
4513 /**
4514 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4515 * @adapter: board private structure
4516 *
4517 * Free all transmit software resources
4518 **/
4519 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4520 {
4521 int i;
4522
4523 for (i = 0; i < adapter->num_tx_queues; i++)
4524 if (adapter->tx_ring[i]->desc)
4525 ixgbe_free_tx_resources(adapter, adapter->tx_ring[i]);
4526 }
4527
4528 /**
4529 * ixgbe_free_rx_resources - Free Rx Resources
4530 * @adapter: board private structure
4531 * @rx_ring: ring to clean the resources from
4532 *
4533 * Free all receive software resources
4534 **/
4535 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
4536 struct ixgbe_ring *rx_ring)
4537 {
4538 struct pci_dev *pdev = adapter->pdev;
4539
4540 ixgbe_clean_rx_ring(adapter, rx_ring);
4541
4542 vfree(rx_ring->rx_buffer_info);
4543 rx_ring->rx_buffer_info = NULL;
4544
4545 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
4546
4547 rx_ring->desc = NULL;
4548 }
4549
4550 /**
4551 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4552 * @adapter: board private structure
4553 *
4554 * Free all receive software resources
4555 **/
4556 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4557 {
4558 int i;
4559
4560 for (i = 0; i < adapter->num_rx_queues; i++)
4561 if (adapter->rx_ring[i]->desc)
4562 ixgbe_free_rx_resources(adapter, adapter->rx_ring[i]);
4563 }
4564
4565 /**
4566 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4567 * @netdev: network interface device structure
4568 * @new_mtu: new value for maximum frame size
4569 *
4570 * Returns 0 on success, negative on failure
4571 **/
4572 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4573 {
4574 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4575 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4576
4577 /* MTU < 68 is an error and causes problems on some kernels */
4578 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4579 return -EINVAL;
4580
4581 DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
4582 netdev->mtu, new_mtu);
4583 /* must set new MTU before calling down or up */
4584 netdev->mtu = new_mtu;
4585
4586 if (netif_running(netdev))
4587 ixgbe_reinit_locked(adapter);
4588
4589 return 0;
4590 }
4591
4592 /**
4593 * ixgbe_open - Called when a network interface is made active
4594 * @netdev: network interface device structure
4595 *
4596 * Returns 0 on success, negative value on failure
4597 *
4598 * The open entry point is called when a network interface is made
4599 * active by the system (IFF_UP). At this point all resources needed
4600 * for transmit and receive operations are allocated, the interrupt
4601 * handler is registered with the OS, the watchdog timer is started,
4602 * and the stack is notified that the interface is ready.
4603 **/
4604 static int ixgbe_open(struct net_device *netdev)
4605 {
4606 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4607 int err;
4608
4609 /* disallow open during test */
4610 if (test_bit(__IXGBE_TESTING, &adapter->state))
4611 return -EBUSY;
4612
4613 netif_carrier_off(netdev);
4614
4615 /* allocate transmit descriptors */
4616 err = ixgbe_setup_all_tx_resources(adapter);
4617 if (err)
4618 goto err_setup_tx;
4619
4620 /* allocate receive descriptors */
4621 err = ixgbe_setup_all_rx_resources(adapter);
4622 if (err)
4623 goto err_setup_rx;
4624
4625 ixgbe_configure(adapter);
4626
4627 err = ixgbe_request_irq(adapter);
4628 if (err)
4629 goto err_req_irq;
4630
4631 err = ixgbe_up_complete(adapter);
4632 if (err)
4633 goto err_up;
4634
4635 netif_tx_start_all_queues(netdev);
4636
4637 return 0;
4638
4639 err_up:
4640 ixgbe_release_hw_control(adapter);
4641 ixgbe_free_irq(adapter);
4642 err_req_irq:
4643 err_setup_rx:
4644 ixgbe_free_all_rx_resources(adapter);
4645 err_setup_tx:
4646 ixgbe_free_all_tx_resources(adapter);
4647 ixgbe_reset(adapter);
4648
4649 return err;
4650 }
4651
4652 /**
4653 * ixgbe_close - Disables a network interface
4654 * @netdev: network interface device structure
4655 *
4656 * Returns 0, this is not allowed to fail
4657 *
4658 * The close entry point is called when an interface is de-activated
4659 * by the OS. The hardware is still under the drivers control, but
4660 * needs to be disabled. A global MAC reset is issued to stop the
4661 * hardware, and all transmit and receive resources are freed.
4662 **/
4663 static int ixgbe_close(struct net_device *netdev)
4664 {
4665 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4666
4667 ixgbe_down(adapter);
4668 ixgbe_free_irq(adapter);
4669
4670 ixgbe_free_all_tx_resources(adapter);
4671 ixgbe_free_all_rx_resources(adapter);
4672
4673 ixgbe_release_hw_control(adapter);
4674
4675 return 0;
4676 }
4677
4678 #ifdef CONFIG_PM
4679 static int ixgbe_resume(struct pci_dev *pdev)
4680 {
4681 struct net_device *netdev = pci_get_drvdata(pdev);
4682 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4683 u32 err;
4684
4685 pci_set_power_state(pdev, PCI_D0);
4686 pci_restore_state(pdev);
4687 /*
4688 * pci_restore_state clears dev->state_saved so call
4689 * pci_save_state to restore it.
4690 */
4691 pci_save_state(pdev);
4692
4693 err = pci_enable_device_mem(pdev);
4694 if (err) {
4695 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
4696 "suspend\n");
4697 return err;
4698 }
4699 pci_set_master(pdev);
4700
4701 pci_wake_from_d3(pdev, false);
4702
4703 err = ixgbe_init_interrupt_scheme(adapter);
4704 if (err) {
4705 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
4706 "device\n");
4707 return err;
4708 }
4709
4710 ixgbe_reset(adapter);
4711
4712 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4713
4714 if (netif_running(netdev)) {
4715 err = ixgbe_open(adapter->netdev);
4716 if (err)
4717 return err;
4718 }
4719
4720 netif_device_attach(netdev);
4721
4722 return 0;
4723 }
4724 #endif /* CONFIG_PM */
4725
4726 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4727 {
4728 struct net_device *netdev = pci_get_drvdata(pdev);
4729 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4730 struct ixgbe_hw *hw = &adapter->hw;
4731 u32 ctrl, fctrl;
4732 u32 wufc = adapter->wol;
4733 #ifdef CONFIG_PM
4734 int retval = 0;
4735 #endif
4736
4737 netif_device_detach(netdev);
4738
4739 if (netif_running(netdev)) {
4740 ixgbe_down(adapter);
4741 ixgbe_free_irq(adapter);
4742 ixgbe_free_all_tx_resources(adapter);
4743 ixgbe_free_all_rx_resources(adapter);
4744 }
4745 ixgbe_clear_interrupt_scheme(adapter);
4746
4747 #ifdef CONFIG_PM
4748 retval = pci_save_state(pdev);
4749 if (retval)
4750 return retval;
4751
4752 #endif
4753 if (wufc) {
4754 ixgbe_set_rx_mode(netdev);
4755
4756 /* turn on all-multi mode if wake on multicast is enabled */
4757 if (wufc & IXGBE_WUFC_MC) {
4758 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4759 fctrl |= IXGBE_FCTRL_MPE;
4760 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4761 }
4762
4763 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4764 ctrl |= IXGBE_CTRL_GIO_DIS;
4765 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4766
4767 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4768 } else {
4769 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4770 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4771 }
4772
4773 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
4774 pci_wake_from_d3(pdev, true);
4775 else
4776 pci_wake_from_d3(pdev, false);
4777
4778 *enable_wake = !!wufc;
4779
4780 ixgbe_release_hw_control(adapter);
4781
4782 pci_disable_device(pdev);
4783
4784 return 0;
4785 }
4786
4787 #ifdef CONFIG_PM
4788 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4789 {
4790 int retval;
4791 bool wake;
4792
4793 retval = __ixgbe_shutdown(pdev, &wake);
4794 if (retval)
4795 return retval;
4796
4797 if (wake) {
4798 pci_prepare_to_sleep(pdev);
4799 } else {
4800 pci_wake_from_d3(pdev, false);
4801 pci_set_power_state(pdev, PCI_D3hot);
4802 }
4803
4804 return 0;
4805 }
4806 #endif /* CONFIG_PM */
4807
4808 static void ixgbe_shutdown(struct pci_dev *pdev)
4809 {
4810 bool wake;
4811
4812 __ixgbe_shutdown(pdev, &wake);
4813
4814 if (system_state == SYSTEM_POWER_OFF) {
4815 pci_wake_from_d3(pdev, wake);
4816 pci_set_power_state(pdev, PCI_D3hot);
4817 }
4818 }
4819
4820 /**
4821 * ixgbe_update_stats - Update the board statistics counters.
4822 * @adapter: board private structure
4823 **/
4824 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4825 {
4826 struct net_device *netdev = adapter->netdev;
4827 struct ixgbe_hw *hw = &adapter->hw;
4828 u64 total_mpc = 0;
4829 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4830 u64 non_eop_descs = 0, restart_queue = 0;
4831
4832 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
4833 u64 rsc_count = 0;
4834 u64 rsc_flush = 0;
4835 for (i = 0; i < 16; i++)
4836 adapter->hw_rx_no_dma_resources +=
4837 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4838 for (i = 0; i < adapter->num_rx_queues; i++) {
4839 rsc_count += adapter->rx_ring[i]->rsc_count;
4840 rsc_flush += adapter->rx_ring[i]->rsc_flush;
4841 }
4842 adapter->rsc_total_count = rsc_count;
4843 adapter->rsc_total_flush = rsc_flush;
4844 }
4845
4846 /* gather some stats to the adapter struct that are per queue */
4847 for (i = 0; i < adapter->num_tx_queues; i++)
4848 restart_queue += adapter->tx_ring[i]->restart_queue;
4849 adapter->restart_queue = restart_queue;
4850
4851 for (i = 0; i < adapter->num_rx_queues; i++)
4852 non_eop_descs += adapter->rx_ring[i]->non_eop_descs;
4853 adapter->non_eop_descs = non_eop_descs;
4854
4855 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
4856 for (i = 0; i < 8; i++) {
4857 /* for packet buffers not used, the register should read 0 */
4858 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
4859 missed_rx += mpc;
4860 adapter->stats.mpc[i] += mpc;
4861 total_mpc += adapter->stats.mpc[i];
4862 if (hw->mac.type == ixgbe_mac_82598EB)
4863 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
4864 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
4865 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
4866 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
4867 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
4868 if (hw->mac.type == ixgbe_mac_82599EB) {
4869 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4870 IXGBE_PXONRXCNT(i));
4871 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4872 IXGBE_PXOFFRXCNT(i));
4873 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4874 } else {
4875 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4876 IXGBE_PXONRXC(i));
4877 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4878 IXGBE_PXOFFRXC(i));
4879 }
4880 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
4881 IXGBE_PXONTXC(i));
4882 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
4883 IXGBE_PXOFFTXC(i));
4884 }
4885 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
4886 /* work around hardware counting issue */
4887 adapter->stats.gprc -= missed_rx;
4888
4889 /* 82598 hardware only has a 32 bit counter in the high register */
4890 if (hw->mac.type == ixgbe_mac_82599EB) {
4891 u64 tmp;
4892 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
4893 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
4894 adapter->stats.gorc += (tmp << 32);
4895 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
4896 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
4897 adapter->stats.gotc += (tmp << 32);
4898 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
4899 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
4900 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
4901 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
4902 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
4903 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
4904 #ifdef IXGBE_FCOE
4905 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
4906 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
4907 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
4908 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
4909 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
4910 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
4911 #endif /* IXGBE_FCOE */
4912 } else {
4913 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
4914 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
4915 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
4916 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
4917 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
4918 }
4919 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
4920 adapter->stats.bprc += bprc;
4921 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
4922 if (hw->mac.type == ixgbe_mac_82598EB)
4923 adapter->stats.mprc -= bprc;
4924 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
4925 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
4926 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
4927 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
4928 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
4929 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
4930 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
4931 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
4932 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
4933 adapter->stats.lxontxc += lxon;
4934 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
4935 adapter->stats.lxofftxc += lxoff;
4936 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4937 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
4938 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
4939 /*
4940 * 82598 errata - tx of flow control packets is included in tx counters
4941 */
4942 xon_off_tot = lxon + lxoff;
4943 adapter->stats.gptc -= xon_off_tot;
4944 adapter->stats.mptc -= xon_off_tot;
4945 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
4946 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4947 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
4948 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
4949 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
4950 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
4951 adapter->stats.ptc64 -= xon_off_tot;
4952 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
4953 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
4954 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
4955 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
4956 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
4957 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
4958
4959 /* Fill out the OS statistics structure */
4960 netdev->stats.multicast = adapter->stats.mprc;
4961
4962 /* Rx Errors */
4963 netdev->stats.rx_errors = adapter->stats.crcerrs +
4964 adapter->stats.rlec;
4965 netdev->stats.rx_dropped = 0;
4966 netdev->stats.rx_length_errors = adapter->stats.rlec;
4967 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4968 netdev->stats.rx_missed_errors = total_mpc;
4969 }
4970
4971 /**
4972 * ixgbe_watchdog - Timer Call-back
4973 * @data: pointer to adapter cast into an unsigned long
4974 **/
4975 static void ixgbe_watchdog(unsigned long data)
4976 {
4977 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4978 struct ixgbe_hw *hw = &adapter->hw;
4979 u64 eics = 0;
4980 int i;
4981
4982 /*
4983 * Do the watchdog outside of interrupt context due to the lovely
4984 * delays that some of the newer hardware requires
4985 */
4986
4987 if (test_bit(__IXGBE_DOWN, &adapter->state))
4988 goto watchdog_short_circuit;
4989
4990 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
4991 /*
4992 * for legacy and MSI interrupts don't set any bits
4993 * that are enabled for EIAM, because this operation
4994 * would set *both* EIMS and EICS for any bit in EIAM
4995 */
4996 IXGBE_WRITE_REG(hw, IXGBE_EICS,
4997 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
4998 goto watchdog_reschedule;
4999 }
5000
5001 /* get one bit for every active tx/rx interrupt vector */
5002 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5003 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5004 if (qv->rxr_count || qv->txr_count)
5005 eics |= ((u64)1 << i);
5006 }
5007
5008 /* Cause software interrupt to ensure rx rings are cleaned */
5009 ixgbe_irq_rearm_queues(adapter, eics);
5010
5011 watchdog_reschedule:
5012 /* Reset the timer */
5013 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5014
5015 watchdog_short_circuit:
5016 schedule_work(&adapter->watchdog_task);
5017 }
5018
5019 /**
5020 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5021 * @work: pointer to work_struct containing our data
5022 **/
5023 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5024 {
5025 struct ixgbe_adapter *adapter = container_of(work,
5026 struct ixgbe_adapter,
5027 multispeed_fiber_task);
5028 struct ixgbe_hw *hw = &adapter->hw;
5029 u32 autoneg;
5030 bool negotiation;
5031
5032 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5033 autoneg = hw->phy.autoneg_advertised;
5034 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5035 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5036 hw->mac.autotry_restart = false;
5037 if (hw->mac.ops.setup_link)
5038 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5039 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5040 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5041 }
5042
5043 /**
5044 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5045 * @work: pointer to work_struct containing our data
5046 **/
5047 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5048 {
5049 struct ixgbe_adapter *adapter = container_of(work,
5050 struct ixgbe_adapter,
5051 sfp_config_module_task);
5052 struct ixgbe_hw *hw = &adapter->hw;
5053 u32 err;
5054
5055 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5056
5057 /* Time for electrical oscillations to settle down */
5058 msleep(100);
5059 err = hw->phy.ops.identify_sfp(hw);
5060
5061 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5062 dev_err(&adapter->pdev->dev, "failed to initialize because "
5063 "an unsupported SFP+ module type was detected.\n"
5064 "Reload the driver after installing a supported "
5065 "module.\n");
5066 unregister_netdev(adapter->netdev);
5067 return;
5068 }
5069 hw->mac.ops.setup_sfp(hw);
5070
5071 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5072 /* This will also work for DA Twinax connections */
5073 schedule_work(&adapter->multispeed_fiber_task);
5074 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5075 }
5076
5077 /**
5078 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5079 * @work: pointer to work_struct containing our data
5080 **/
5081 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5082 {
5083 struct ixgbe_adapter *adapter = container_of(work,
5084 struct ixgbe_adapter,
5085 fdir_reinit_task);
5086 struct ixgbe_hw *hw = &adapter->hw;
5087 int i;
5088
5089 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5090 for (i = 0; i < adapter->num_tx_queues; i++)
5091 set_bit(__IXGBE_FDIR_INIT_DONE,
5092 &(adapter->tx_ring[i]->reinit_state));
5093 } else {
5094 DPRINTK(PROBE, ERR, "failed to finish FDIR re-initialization, "
5095 "ignored adding FDIR ATR filters \n");
5096 }
5097 /* Done FDIR Re-initialization, enable transmits */
5098 netif_tx_start_all_queues(adapter->netdev);
5099 }
5100
5101 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5102
5103 /**
5104 * ixgbe_watchdog_task - worker thread to bring link up
5105 * @work: pointer to work_struct containing our data
5106 **/
5107 static void ixgbe_watchdog_task(struct work_struct *work)
5108 {
5109 struct ixgbe_adapter *adapter = container_of(work,
5110 struct ixgbe_adapter,
5111 watchdog_task);
5112 struct net_device *netdev = adapter->netdev;
5113 struct ixgbe_hw *hw = &adapter->hw;
5114 u32 link_speed;
5115 bool link_up;
5116 int i;
5117 struct ixgbe_ring *tx_ring;
5118 int some_tx_pending = 0;
5119
5120 mutex_lock(&ixgbe_watchdog_lock);
5121
5122 link_up = adapter->link_up;
5123 link_speed = adapter->link_speed;
5124
5125 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5126 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5127 if (link_up) {
5128 #ifdef CONFIG_DCB
5129 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5130 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5131 hw->mac.ops.fc_enable(hw, i);
5132 } else {
5133 hw->mac.ops.fc_enable(hw, 0);
5134 }
5135 #else
5136 hw->mac.ops.fc_enable(hw, 0);
5137 #endif
5138 }
5139
5140 if (link_up ||
5141 time_after(jiffies, (adapter->link_check_timeout +
5142 IXGBE_TRY_LINK_TIMEOUT))) {
5143 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5144 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5145 }
5146 adapter->link_up = link_up;
5147 adapter->link_speed = link_speed;
5148 }
5149
5150 if (link_up) {
5151 if (!netif_carrier_ok(netdev)) {
5152 bool flow_rx, flow_tx;
5153
5154 if (hw->mac.type == ixgbe_mac_82599EB) {
5155 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5156 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5157 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5158 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5159 } else {
5160 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5161 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5162 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5163 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5164 }
5165
5166 printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
5167 "Flow Control: %s\n",
5168 netdev->name,
5169 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5170 "10 Gbps" :
5171 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5172 "1 Gbps" : "unknown speed")),
5173 ((flow_rx && flow_tx) ? "RX/TX" :
5174 (flow_rx ? "RX" :
5175 (flow_tx ? "TX" : "None"))));
5176
5177 netif_carrier_on(netdev);
5178 } else {
5179 /* Force detection of hung controller */
5180 adapter->detect_tx_hung = true;
5181 }
5182 } else {
5183 adapter->link_up = false;
5184 adapter->link_speed = 0;
5185 if (netif_carrier_ok(netdev)) {
5186 printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
5187 netdev->name);
5188 netif_carrier_off(netdev);
5189 }
5190 }
5191
5192 if (!netif_carrier_ok(netdev)) {
5193 for (i = 0; i < adapter->num_tx_queues; i++) {
5194 tx_ring = adapter->tx_ring[i];
5195 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5196 some_tx_pending = 1;
5197 break;
5198 }
5199 }
5200
5201 if (some_tx_pending) {
5202 /* We've lost link, so the controller stops DMA,
5203 * but we've got queued Tx work that's never going
5204 * to get done, so reset controller to flush Tx.
5205 * (Do the reset outside of interrupt context).
5206 */
5207 schedule_work(&adapter->reset_task);
5208 }
5209 }
5210
5211 ixgbe_update_stats(adapter);
5212 mutex_unlock(&ixgbe_watchdog_lock);
5213 }
5214
5215 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5216 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5217 u32 tx_flags, u8 *hdr_len)
5218 {
5219 struct ixgbe_adv_tx_context_desc *context_desc;
5220 unsigned int i;
5221 int err;
5222 struct ixgbe_tx_buffer *tx_buffer_info;
5223 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5224 u32 mss_l4len_idx, l4len;
5225
5226 if (skb_is_gso(skb)) {
5227 if (skb_header_cloned(skb)) {
5228 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5229 if (err)
5230 return err;
5231 }
5232 l4len = tcp_hdrlen(skb);
5233 *hdr_len += l4len;
5234
5235 if (skb->protocol == htons(ETH_P_IP)) {
5236 struct iphdr *iph = ip_hdr(skb);
5237 iph->tot_len = 0;
5238 iph->check = 0;
5239 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5240 iph->daddr, 0,
5241 IPPROTO_TCP,
5242 0);
5243 } else if (skb_is_gso_v6(skb)) {
5244 ipv6_hdr(skb)->payload_len = 0;
5245 tcp_hdr(skb)->check =
5246 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5247 &ipv6_hdr(skb)->daddr,
5248 0, IPPROTO_TCP, 0);
5249 }
5250
5251 i = tx_ring->next_to_use;
5252
5253 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5254 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5255
5256 /* VLAN MACLEN IPLEN */
5257 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5258 vlan_macip_lens |=
5259 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5260 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5261 IXGBE_ADVTXD_MACLEN_SHIFT);
5262 *hdr_len += skb_network_offset(skb);
5263 vlan_macip_lens |=
5264 (skb_transport_header(skb) - skb_network_header(skb));
5265 *hdr_len +=
5266 (skb_transport_header(skb) - skb_network_header(skb));
5267 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5268 context_desc->seqnum_seed = 0;
5269
5270 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5271 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5272 IXGBE_ADVTXD_DTYP_CTXT);
5273
5274 if (skb->protocol == htons(ETH_P_IP))
5275 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5276 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5277 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5278
5279 /* MSS L4LEN IDX */
5280 mss_l4len_idx =
5281 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5282 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5283 /* use index 1 for TSO */
5284 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5285 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5286
5287 tx_buffer_info->time_stamp = jiffies;
5288 tx_buffer_info->next_to_watch = i;
5289
5290 i++;
5291 if (i == tx_ring->count)
5292 i = 0;
5293 tx_ring->next_to_use = i;
5294
5295 return true;
5296 }
5297 return false;
5298 }
5299
5300 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5301 struct ixgbe_ring *tx_ring,
5302 struct sk_buff *skb, u32 tx_flags)
5303 {
5304 struct ixgbe_adv_tx_context_desc *context_desc;
5305 unsigned int i;
5306 struct ixgbe_tx_buffer *tx_buffer_info;
5307 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5308
5309 if (skb->ip_summed == CHECKSUM_PARTIAL ||
5310 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5311 i = tx_ring->next_to_use;
5312 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5313 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5314
5315 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5316 vlan_macip_lens |=
5317 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5318 vlan_macip_lens |= (skb_network_offset(skb) <<
5319 IXGBE_ADVTXD_MACLEN_SHIFT);
5320 if (skb->ip_summed == CHECKSUM_PARTIAL)
5321 vlan_macip_lens |= (skb_transport_header(skb) -
5322 skb_network_header(skb));
5323
5324 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5325 context_desc->seqnum_seed = 0;
5326
5327 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5328 IXGBE_ADVTXD_DTYP_CTXT);
5329
5330 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5331 __be16 protocol;
5332
5333 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
5334 const struct vlan_ethhdr *vhdr =
5335 (const struct vlan_ethhdr *)skb->data;
5336
5337 protocol = vhdr->h_vlan_encapsulated_proto;
5338 } else {
5339 protocol = skb->protocol;
5340 }
5341
5342 switch (protocol) {
5343 case cpu_to_be16(ETH_P_IP):
5344 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5345 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5346 type_tucmd_mlhl |=
5347 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5348 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
5349 type_tucmd_mlhl |=
5350 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5351 break;
5352 case cpu_to_be16(ETH_P_IPV6):
5353 /* XXX what about other V6 headers?? */
5354 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5355 type_tucmd_mlhl |=
5356 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5357 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
5358 type_tucmd_mlhl |=
5359 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5360 break;
5361 default:
5362 if (unlikely(net_ratelimit())) {
5363 DPRINTK(PROBE, WARNING,
5364 "partial checksum but proto=%x!\n",
5365 skb->protocol);
5366 }
5367 break;
5368 }
5369 }
5370
5371 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5372 /* use index zero for tx checksum offload */
5373 context_desc->mss_l4len_idx = 0;
5374
5375 tx_buffer_info->time_stamp = jiffies;
5376 tx_buffer_info->next_to_watch = i;
5377
5378 i++;
5379 if (i == tx_ring->count)
5380 i = 0;
5381 tx_ring->next_to_use = i;
5382
5383 return true;
5384 }
5385
5386 return false;
5387 }
5388
5389 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5390 struct ixgbe_ring *tx_ring,
5391 struct sk_buff *skb, u32 tx_flags,
5392 unsigned int first)
5393 {
5394 struct pci_dev *pdev = adapter->pdev;
5395 struct ixgbe_tx_buffer *tx_buffer_info;
5396 unsigned int len;
5397 unsigned int total = skb->len;
5398 unsigned int offset = 0, size, count = 0, i;
5399 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5400 unsigned int f;
5401
5402 i = tx_ring->next_to_use;
5403
5404 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5405 /* excluding fcoe_crc_eof for FCoE */
5406 total -= sizeof(struct fcoe_crc_eof);
5407
5408 len = min(skb_headlen(skb), total);
5409 while (len) {
5410 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5411 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5412
5413 tx_buffer_info->length = size;
5414 tx_buffer_info->mapped_as_page = false;
5415 tx_buffer_info->dma = pci_map_single(pdev,
5416 skb->data + offset,
5417 size, PCI_DMA_TODEVICE);
5418 if (pci_dma_mapping_error(pdev, tx_buffer_info->dma))
5419 goto dma_error;
5420 tx_buffer_info->time_stamp = jiffies;
5421 tx_buffer_info->next_to_watch = i;
5422
5423 len -= size;
5424 total -= size;
5425 offset += size;
5426 count++;
5427
5428 if (len) {
5429 i++;
5430 if (i == tx_ring->count)
5431 i = 0;
5432 }
5433 }
5434
5435 for (f = 0; f < nr_frags; f++) {
5436 struct skb_frag_struct *frag;
5437
5438 frag = &skb_shinfo(skb)->frags[f];
5439 len = min((unsigned int)frag->size, total);
5440 offset = frag->page_offset;
5441
5442 while (len) {
5443 i++;
5444 if (i == tx_ring->count)
5445 i = 0;
5446
5447 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5448 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5449
5450 tx_buffer_info->length = size;
5451 tx_buffer_info->dma = pci_map_page(adapter->pdev,
5452 frag->page,
5453 offset, size,
5454 PCI_DMA_TODEVICE);
5455 tx_buffer_info->mapped_as_page = true;
5456 if (pci_dma_mapping_error(pdev, tx_buffer_info->dma))
5457 goto dma_error;
5458 tx_buffer_info->time_stamp = jiffies;
5459 tx_buffer_info->next_to_watch = i;
5460
5461 len -= size;
5462 total -= size;
5463 offset += size;
5464 count++;
5465 }
5466 if (total == 0)
5467 break;
5468 }
5469
5470 tx_ring->tx_buffer_info[i].skb = skb;
5471 tx_ring->tx_buffer_info[first].next_to_watch = i;
5472
5473 return count;
5474
5475 dma_error:
5476 dev_err(&pdev->dev, "TX DMA map failed\n");
5477
5478 /* clear timestamp and dma mappings for failed tx_buffer_info map */
5479 tx_buffer_info->dma = 0;
5480 tx_buffer_info->time_stamp = 0;
5481 tx_buffer_info->next_to_watch = 0;
5482 if (count)
5483 count--;
5484
5485 /* clear timestamp and dma mappings for remaining portion of packet */
5486 while (count--) {
5487 if (i==0)
5488 i += tx_ring->count;
5489 i--;
5490 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5491 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
5492 }
5493
5494 return 0;
5495 }
5496
5497 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
5498 struct ixgbe_ring *tx_ring,
5499 int tx_flags, int count, u32 paylen, u8 hdr_len)
5500 {
5501 union ixgbe_adv_tx_desc *tx_desc = NULL;
5502 struct ixgbe_tx_buffer *tx_buffer_info;
5503 u32 olinfo_status = 0, cmd_type_len = 0;
5504 unsigned int i;
5505 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
5506
5507 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
5508
5509 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
5510
5511 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5512 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
5513
5514 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
5515 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5516
5517 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5518 IXGBE_ADVTXD_POPTS_SHIFT;
5519
5520 /* use index 1 context for tso */
5521 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5522 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5523 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
5524 IXGBE_ADVTXD_POPTS_SHIFT;
5525
5526 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5527 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5528 IXGBE_ADVTXD_POPTS_SHIFT;
5529
5530 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5531 olinfo_status |= IXGBE_ADVTXD_CC;
5532 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5533 if (tx_flags & IXGBE_TX_FLAGS_FSO)
5534 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5535 }
5536
5537 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
5538
5539 i = tx_ring->next_to_use;
5540 while (count--) {
5541 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5542 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
5543 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
5544 tx_desc->read.cmd_type_len =
5545 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
5546 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5547 i++;
5548 if (i == tx_ring->count)
5549 i = 0;
5550 }
5551
5552 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
5553
5554 /*
5555 * Force memory writes to complete before letting h/w
5556 * know there are new descriptors to fetch. (Only
5557 * applicable for weak-ordered memory model archs,
5558 * such as IA-64).
5559 */
5560 wmb();
5561
5562 tx_ring->next_to_use = i;
5563 writel(i, adapter->hw.hw_addr + tx_ring->tail);
5564 }
5565
5566 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5567 int queue, u32 tx_flags)
5568 {
5569 /* Right now, we support IPv4 only */
5570 struct ixgbe_atr_input atr_input;
5571 struct tcphdr *th;
5572 struct iphdr *iph = ip_hdr(skb);
5573 struct ethhdr *eth = (struct ethhdr *)skb->data;
5574 u16 vlan_id, src_port, dst_port, flex_bytes;
5575 u32 src_ipv4_addr, dst_ipv4_addr;
5576 u8 l4type = 0;
5577
5578 /* check if we're UDP or TCP */
5579 if (iph->protocol == IPPROTO_TCP) {
5580 th = tcp_hdr(skb);
5581 src_port = th->source;
5582 dst_port = th->dest;
5583 l4type |= IXGBE_ATR_L4TYPE_TCP;
5584 /* l4type IPv4 type is 0, no need to assign */
5585 } else {
5586 /* Unsupported L4 header, just bail here */
5587 return;
5588 }
5589
5590 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
5591
5592 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
5593 IXGBE_TX_FLAGS_VLAN_SHIFT;
5594 src_ipv4_addr = iph->saddr;
5595 dst_ipv4_addr = iph->daddr;
5596 flex_bytes = eth->h_proto;
5597
5598 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
5599 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
5600 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
5601 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
5602 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
5603 /* src and dst are inverted, think how the receiver sees them */
5604 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
5605 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
5606
5607 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
5608 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
5609 }
5610
5611 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
5612 struct ixgbe_ring *tx_ring, int size)
5613 {
5614 netif_stop_subqueue(netdev, tx_ring->queue_index);
5615 /* Herbert's original patch had:
5616 * smp_mb__after_netif_stop_queue();
5617 * but since that doesn't exist yet, just open code it. */
5618 smp_mb();
5619
5620 /* We need to check again in a case another CPU has just
5621 * made room available. */
5622 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
5623 return -EBUSY;
5624
5625 /* A reprieve! - use start_queue because it doesn't call schedule */
5626 netif_start_subqueue(netdev, tx_ring->queue_index);
5627 ++tx_ring->restart_queue;
5628 return 0;
5629 }
5630
5631 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
5632 struct ixgbe_ring *tx_ring, int size)
5633 {
5634 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
5635 return 0;
5636 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
5637 }
5638
5639 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
5640 {
5641 struct ixgbe_adapter *adapter = netdev_priv(dev);
5642 int txq = smp_processor_id();
5643
5644 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5645 while (unlikely(txq >= dev->real_num_tx_queues))
5646 txq -= dev->real_num_tx_queues;
5647 return txq;
5648 }
5649
5650 #ifdef IXGBE_FCOE
5651 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5652 ((skb->protocol == htons(ETH_P_FCOE)) ||
5653 (skb->protocol == htons(ETH_P_FIP)))) {
5654 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
5655 txq += adapter->ring_feature[RING_F_FCOE].mask;
5656 return txq;
5657 }
5658 #endif
5659 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5660 if (skb->priority == TC_PRIO_CONTROL)
5661 txq = adapter->ring_feature[RING_F_DCB].indices-1;
5662 else
5663 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
5664 >> 13;
5665 return txq;
5666 }
5667
5668 return skb_tx_hash(dev, skb);
5669 }
5670
5671 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
5672 struct net_device *netdev)
5673 {
5674 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5675 struct ixgbe_ring *tx_ring;
5676 struct netdev_queue *txq;
5677 unsigned int first;
5678 unsigned int tx_flags = 0;
5679 u8 hdr_len = 0;
5680 int tso;
5681 int count = 0;
5682 unsigned int f;
5683
5684 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
5685 tx_flags |= vlan_tx_tag_get(skb);
5686 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5687 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
5688 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
5689 }
5690 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5691 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5692 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5693 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
5694 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5695 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5696 }
5697
5698 tx_ring = adapter->tx_ring[skb->queue_mapping];
5699
5700 #ifdef IXGBE_FCOE
5701 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
5702 #ifdef CONFIG_IXGBE_DCB
5703 /* for FCoE with DCB, we force the priority to what
5704 * was specified by the switch */
5705 if ((skb->protocol == htons(ETH_P_FCOE)) ||
5706 (skb->protocol == htons(ETH_P_FIP))) {
5707 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
5708 << IXGBE_TX_FLAGS_VLAN_SHIFT);
5709 tx_flags |= ((adapter->fcoe.up << 13)
5710 << IXGBE_TX_FLAGS_VLAN_SHIFT);
5711 }
5712 #endif
5713 /* flag for FCoE offloads */
5714 if (skb->protocol == htons(ETH_P_FCOE))
5715 tx_flags |= IXGBE_TX_FLAGS_FCOE;
5716 }
5717 #endif
5718
5719 /* four things can cause us to need a context descriptor */
5720 if (skb_is_gso(skb) ||
5721 (skb->ip_summed == CHECKSUM_PARTIAL) ||
5722 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
5723 (tx_flags & IXGBE_TX_FLAGS_FCOE))
5724 count++;
5725
5726 count += TXD_USE_COUNT(skb_headlen(skb));
5727 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5728 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5729
5730 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
5731 adapter->tx_busy++;
5732 return NETDEV_TX_BUSY;
5733 }
5734
5735 first = tx_ring->next_to_use;
5736 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5737 #ifdef IXGBE_FCOE
5738 /* setup tx offload for FCoE */
5739 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5740 if (tso < 0) {
5741 dev_kfree_skb_any(skb);
5742 return NETDEV_TX_OK;
5743 }
5744 if (tso)
5745 tx_flags |= IXGBE_TX_FLAGS_FSO;
5746 #endif /* IXGBE_FCOE */
5747 } else {
5748 if (skb->protocol == htons(ETH_P_IP))
5749 tx_flags |= IXGBE_TX_FLAGS_IPV4;
5750 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5751 if (tso < 0) {
5752 dev_kfree_skb_any(skb);
5753 return NETDEV_TX_OK;
5754 }
5755
5756 if (tso)
5757 tx_flags |= IXGBE_TX_FLAGS_TSO;
5758 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
5759 (skb->ip_summed == CHECKSUM_PARTIAL))
5760 tx_flags |= IXGBE_TX_FLAGS_CSUM;
5761 }
5762
5763 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
5764 if (count) {
5765 /* add the ATR filter if ATR is on */
5766 if (tx_ring->atr_sample_rate) {
5767 ++tx_ring->atr_count;
5768 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
5769 test_bit(__IXGBE_FDIR_INIT_DONE,
5770 &tx_ring->reinit_state)) {
5771 ixgbe_atr(adapter, skb, tx_ring->queue_index,
5772 tx_flags);
5773 tx_ring->atr_count = 0;
5774 }
5775 }
5776 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
5777 txq->tx_bytes += skb->len;
5778 txq->tx_packets++;
5779 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
5780 hdr_len);
5781 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
5782
5783 } else {
5784 dev_kfree_skb_any(skb);
5785 tx_ring->tx_buffer_info[first].time_stamp = 0;
5786 tx_ring->next_to_use = first;
5787 }
5788
5789 return NETDEV_TX_OK;
5790 }
5791
5792 /**
5793 * ixgbe_set_mac - Change the Ethernet Address of the NIC
5794 * @netdev: network interface device structure
5795 * @p: pointer to an address structure
5796 *
5797 * Returns 0 on success, negative on failure
5798 **/
5799 static int ixgbe_set_mac(struct net_device *netdev, void *p)
5800 {
5801 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5802 struct ixgbe_hw *hw = &adapter->hw;
5803 struct sockaddr *addr = p;
5804
5805 if (!is_valid_ether_addr(addr->sa_data))
5806 return -EADDRNOTAVAIL;
5807
5808 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
5809 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5810
5811 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
5812 IXGBE_RAH_AV);
5813
5814 return 0;
5815 }
5816
5817 static int
5818 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
5819 {
5820 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5821 struct ixgbe_hw *hw = &adapter->hw;
5822 u16 value;
5823 int rc;
5824
5825 if (prtad != hw->phy.mdio.prtad)
5826 return -EINVAL;
5827 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
5828 if (!rc)
5829 rc = value;
5830 return rc;
5831 }
5832
5833 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
5834 u16 addr, u16 value)
5835 {
5836 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5837 struct ixgbe_hw *hw = &adapter->hw;
5838
5839 if (prtad != hw->phy.mdio.prtad)
5840 return -EINVAL;
5841 return hw->phy.ops.write_reg(hw, addr, devad, value);
5842 }
5843
5844 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
5845 {
5846 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5847
5848 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
5849 }
5850
5851 /**
5852 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
5853 * netdev->dev_addrs
5854 * @netdev: network interface device structure
5855 *
5856 * Returns non-zero on failure
5857 **/
5858 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
5859 {
5860 int err = 0;
5861 struct ixgbe_adapter *adapter = netdev_priv(dev);
5862 struct ixgbe_mac_info *mac = &adapter->hw.mac;
5863
5864 if (is_valid_ether_addr(mac->san_addr)) {
5865 rtnl_lock();
5866 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5867 rtnl_unlock();
5868 }
5869 return err;
5870 }
5871
5872 /**
5873 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
5874 * netdev->dev_addrs
5875 * @netdev: network interface device structure
5876 *
5877 * Returns non-zero on failure
5878 **/
5879 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
5880 {
5881 int err = 0;
5882 struct ixgbe_adapter *adapter = netdev_priv(dev);
5883 struct ixgbe_mac_info *mac = &adapter->hw.mac;
5884
5885 if (is_valid_ether_addr(mac->san_addr)) {
5886 rtnl_lock();
5887 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5888 rtnl_unlock();
5889 }
5890 return err;
5891 }
5892
5893 #ifdef CONFIG_NET_POLL_CONTROLLER
5894 /*
5895 * Polling 'interrupt' - used by things like netconsole to send skbs
5896 * without having to re-enable interrupts. It's not called while
5897 * the interrupt routine is executing.
5898 */
5899 static void ixgbe_netpoll(struct net_device *netdev)
5900 {
5901 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5902 int i;
5903
5904 /* if interface is down do nothing */
5905 if (test_bit(__IXGBE_DOWN, &adapter->state))
5906 return;
5907
5908 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
5909 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5910 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5911 for (i = 0; i < num_q_vectors; i++) {
5912 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
5913 ixgbe_msix_clean_many(0, q_vector);
5914 }
5915 } else {
5916 ixgbe_intr(adapter->pdev->irq, netdev);
5917 }
5918 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
5919 }
5920 #endif
5921
5922 static const struct net_device_ops ixgbe_netdev_ops = {
5923 .ndo_open = ixgbe_open,
5924 .ndo_stop = ixgbe_close,
5925 .ndo_start_xmit = ixgbe_xmit_frame,
5926 .ndo_select_queue = ixgbe_select_queue,
5927 .ndo_set_rx_mode = ixgbe_set_rx_mode,
5928 .ndo_set_multicast_list = ixgbe_set_rx_mode,
5929 .ndo_validate_addr = eth_validate_addr,
5930 .ndo_set_mac_address = ixgbe_set_mac,
5931 .ndo_change_mtu = ixgbe_change_mtu,
5932 .ndo_tx_timeout = ixgbe_tx_timeout,
5933 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
5934 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
5935 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
5936 .ndo_do_ioctl = ixgbe_ioctl,
5937 #ifdef CONFIG_NET_POLL_CONTROLLER
5938 .ndo_poll_controller = ixgbe_netpoll,
5939 #endif
5940 #ifdef IXGBE_FCOE
5941 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
5942 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
5943 .ndo_fcoe_enable = ixgbe_fcoe_enable,
5944 .ndo_fcoe_disable = ixgbe_fcoe_disable,
5945 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
5946 #endif /* IXGBE_FCOE */
5947 };
5948
5949 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
5950 const struct ixgbe_info *ii)
5951 {
5952 #ifdef CONFIG_PCI_IOV
5953 struct ixgbe_hw *hw = &adapter->hw;
5954 int err;
5955
5956 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
5957 return;
5958
5959 /* The 82599 supports up to 64 VFs per physical function
5960 * but this implementation limits allocation to 63 so that
5961 * basic networking resources are still available to the
5962 * physical function
5963 */
5964 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
5965 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
5966 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
5967 if (err) {
5968 DPRINTK(PROBE, ERR,
5969 "Failed to enable PCI sriov: %d\n", err);
5970 goto err_novfs;
5971 }
5972 /* If call to enable VFs succeeded then allocate memory
5973 * for per VF control structures.
5974 */
5975 adapter->vfinfo =
5976 kcalloc(adapter->num_vfs,
5977 sizeof(struct vf_data_storage), GFP_KERNEL);
5978 if (adapter->vfinfo) {
5979 /* Now that we're sure SR-IOV is enabled
5980 * and memory allocated set up the mailbox parameters
5981 */
5982 ixgbe_init_mbx_params_pf(hw);
5983 memcpy(&hw->mbx.ops, ii->mbx_ops,
5984 sizeof(hw->mbx.ops));
5985
5986 /* Disable RSC when in SR-IOV mode */
5987 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
5988 IXGBE_FLAG2_RSC_ENABLED);
5989 return;
5990 }
5991
5992 /* Oh oh */
5993 DPRINTK(PROBE, ERR,
5994 "Unable to allocate memory for VF "
5995 "Data Storage - SRIOV disabled\n");
5996 pci_disable_sriov(adapter->pdev);
5997
5998 err_novfs:
5999 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6000 adapter->num_vfs = 0;
6001 #endif /* CONFIG_PCI_IOV */
6002 }
6003
6004 /**
6005 * ixgbe_probe - Device Initialization Routine
6006 * @pdev: PCI device information struct
6007 * @ent: entry in ixgbe_pci_tbl
6008 *
6009 * Returns 0 on success, negative on failure
6010 *
6011 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6012 * The OS initialization, configuring of the adapter private structure,
6013 * and a hardware reset occur.
6014 **/
6015 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6016 const struct pci_device_id *ent)
6017 {
6018 struct net_device *netdev;
6019 struct ixgbe_adapter *adapter = NULL;
6020 struct ixgbe_hw *hw;
6021 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6022 static int cards_found;
6023 int i, err, pci_using_dac;
6024 unsigned int indices = num_possible_cpus();
6025 #ifdef IXGBE_FCOE
6026 u16 device_caps;
6027 #endif
6028 u32 part_num, eec;
6029
6030 err = pci_enable_device_mem(pdev);
6031 if (err)
6032 return err;
6033
6034 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
6035 !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
6036 pci_using_dac = 1;
6037 } else {
6038 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
6039 if (err) {
6040 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
6041 if (err) {
6042 dev_err(&pdev->dev, "No usable DMA "
6043 "configuration, aborting\n");
6044 goto err_dma;
6045 }
6046 }
6047 pci_using_dac = 0;
6048 }
6049
6050 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6051 IORESOURCE_MEM), ixgbe_driver_name);
6052 if (err) {
6053 dev_err(&pdev->dev,
6054 "pci_request_selected_regions failed 0x%x\n", err);
6055 goto err_pci_reg;
6056 }
6057
6058 pci_enable_pcie_error_reporting(pdev);
6059
6060 pci_set_master(pdev);
6061 pci_save_state(pdev);
6062
6063 if (ii->mac == ixgbe_mac_82598EB)
6064 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6065 else
6066 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6067
6068 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6069 #ifdef IXGBE_FCOE
6070 indices += min_t(unsigned int, num_possible_cpus(),
6071 IXGBE_MAX_FCOE_INDICES);
6072 #endif
6073 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6074 if (!netdev) {
6075 err = -ENOMEM;
6076 goto err_alloc_etherdev;
6077 }
6078
6079 SET_NETDEV_DEV(netdev, &pdev->dev);
6080
6081 pci_set_drvdata(pdev, netdev);
6082 adapter = netdev_priv(netdev);
6083
6084 adapter->netdev = netdev;
6085 adapter->pdev = pdev;
6086 hw = &adapter->hw;
6087 hw->back = adapter;
6088 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6089
6090 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6091 pci_resource_len(pdev, 0));
6092 if (!hw->hw_addr) {
6093 err = -EIO;
6094 goto err_ioremap;
6095 }
6096
6097 for (i = 1; i <= 5; i++) {
6098 if (pci_resource_len(pdev, i) == 0)
6099 continue;
6100 }
6101
6102 netdev->netdev_ops = &ixgbe_netdev_ops;
6103 ixgbe_set_ethtool_ops(netdev);
6104 netdev->watchdog_timeo = 5 * HZ;
6105 strcpy(netdev->name, pci_name(pdev));
6106
6107 adapter->bd_number = cards_found;
6108
6109 /* Setup hw api */
6110 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6111 hw->mac.type = ii->mac;
6112
6113 /* EEPROM */
6114 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6115 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6116 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6117 if (!(eec & (1 << 8)))
6118 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6119
6120 /* PHY */
6121 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6122 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6123 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6124 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6125 hw->phy.mdio.mmds = 0;
6126 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6127 hw->phy.mdio.dev = netdev;
6128 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6129 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6130
6131 /* set up this timer and work struct before calling get_invariants
6132 * which might start the timer
6133 */
6134 init_timer(&adapter->sfp_timer);
6135 adapter->sfp_timer.function = &ixgbe_sfp_timer;
6136 adapter->sfp_timer.data = (unsigned long) adapter;
6137
6138 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6139
6140 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6141 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6142
6143 /* a new SFP+ module arrival, called from GPI SDP2 context */
6144 INIT_WORK(&adapter->sfp_config_module_task,
6145 ixgbe_sfp_config_module_task);
6146
6147 ii->get_invariants(hw);
6148
6149 /* setup the private structure */
6150 err = ixgbe_sw_init(adapter);
6151 if (err)
6152 goto err_sw_init;
6153
6154 /* Make it possible the adapter to be woken up via WOL */
6155 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6156 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6157
6158 /*
6159 * If there is a fan on this device and it has failed log the
6160 * failure.
6161 */
6162 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6163 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6164 if (esdp & IXGBE_ESDP_SDP1)
6165 DPRINTK(PROBE, CRIT,
6166 "Fan has stopped, replace the adapter\n");
6167 }
6168
6169 /* reset_hw fills in the perm_addr as well */
6170 err = hw->mac.ops.reset_hw(hw);
6171 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6172 hw->mac.type == ixgbe_mac_82598EB) {
6173 /*
6174 * Start a kernel thread to watch for a module to arrive.
6175 * Only do this for 82598, since 82599 will generate
6176 * interrupts on module arrival.
6177 */
6178 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6179 mod_timer(&adapter->sfp_timer,
6180 round_jiffies(jiffies + (2 * HZ)));
6181 err = 0;
6182 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6183 dev_err(&adapter->pdev->dev, "failed to initialize because "
6184 "an unsupported SFP+ module type was detected.\n"
6185 "Reload the driver after installing a supported "
6186 "module.\n");
6187 goto err_sw_init;
6188 } else if (err) {
6189 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
6190 goto err_sw_init;
6191 }
6192
6193 ixgbe_probe_vf(adapter, ii);
6194
6195 netdev->features = NETIF_F_SG |
6196 NETIF_F_IP_CSUM |
6197 NETIF_F_HW_VLAN_TX |
6198 NETIF_F_HW_VLAN_RX |
6199 NETIF_F_HW_VLAN_FILTER;
6200
6201 netdev->features |= NETIF_F_IPV6_CSUM;
6202 netdev->features |= NETIF_F_TSO;
6203 netdev->features |= NETIF_F_TSO6;
6204 netdev->features |= NETIF_F_GRO;
6205
6206 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6207 netdev->features |= NETIF_F_SCTP_CSUM;
6208
6209 netdev->vlan_features |= NETIF_F_TSO;
6210 netdev->vlan_features |= NETIF_F_TSO6;
6211 netdev->vlan_features |= NETIF_F_IP_CSUM;
6212 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6213 netdev->vlan_features |= NETIF_F_SG;
6214
6215 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6216 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6217 IXGBE_FLAG_DCB_ENABLED);
6218 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6219 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6220
6221 #ifdef CONFIG_IXGBE_DCB
6222 netdev->dcbnl_ops = &dcbnl_ops;
6223 #endif
6224
6225 #ifdef IXGBE_FCOE
6226 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6227 if (hw->mac.ops.get_device_caps) {
6228 hw->mac.ops.get_device_caps(hw, &device_caps);
6229 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6230 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6231 }
6232 }
6233 #endif /* IXGBE_FCOE */
6234 if (pci_using_dac)
6235 netdev->features |= NETIF_F_HIGHDMA;
6236
6237 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6238 netdev->features |= NETIF_F_LRO;
6239
6240 /* make sure the EEPROM is good */
6241 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6242 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
6243 err = -EIO;
6244 goto err_eeprom;
6245 }
6246
6247 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6248 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6249
6250 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6251 dev_err(&pdev->dev, "invalid MAC address\n");
6252 err = -EIO;
6253 goto err_eeprom;
6254 }
6255
6256 init_timer(&adapter->watchdog_timer);
6257 adapter->watchdog_timer.function = &ixgbe_watchdog;
6258 adapter->watchdog_timer.data = (unsigned long)adapter;
6259
6260 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6261 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6262
6263 err = ixgbe_init_interrupt_scheme(adapter);
6264 if (err)
6265 goto err_sw_init;
6266
6267 switch (pdev->device) {
6268 case IXGBE_DEV_ID_82599_KX4:
6269 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
6270 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
6271 break;
6272 default:
6273 adapter->wol = 0;
6274 break;
6275 }
6276 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6277
6278 /* pick up the PCI bus settings for reporting later */
6279 hw->mac.ops.get_bus_info(hw);
6280
6281 /* print bus type/speed/width info */
6282 dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
6283 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
6284 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
6285 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
6286 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
6287 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
6288 "Unknown"),
6289 netdev->dev_addr);
6290 ixgbe_read_pba_num_generic(hw, &part_num);
6291 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
6292 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
6293 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
6294 (part_num >> 8), (part_num & 0xff));
6295 else
6296 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
6297 hw->mac.type, hw->phy.type,
6298 (part_num >> 8), (part_num & 0xff));
6299
6300 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
6301 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
6302 "this card is not sufficient for optimal "
6303 "performance.\n");
6304 dev_warn(&pdev->dev, "For optimal performance a x8 "
6305 "PCI-Express slot is required.\n");
6306 }
6307
6308 /* save off EEPROM version number */
6309 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
6310
6311 /* reset the hardware with the new settings */
6312 err = hw->mac.ops.start_hw(hw);
6313
6314 if (err == IXGBE_ERR_EEPROM_VERSION) {
6315 /* We are running on a pre-production device, log a warning */
6316 dev_warn(&pdev->dev, "This device is a pre-production "
6317 "adapter/LOM. Please be aware there may be issues "
6318 "associated with your hardware. If you are "
6319 "experiencing problems please contact your Intel or "
6320 "hardware representative who provided you with this "
6321 "hardware.\n");
6322 }
6323 strcpy(netdev->name, "eth%d");
6324 err = register_netdev(netdev);
6325 if (err)
6326 goto err_register;
6327
6328 /* carrier off reporting is important to ethtool even BEFORE open */
6329 netif_carrier_off(netdev);
6330
6331 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6332 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6333 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
6334
6335 #ifdef CONFIG_IXGBE_DCA
6336 if (dca_add_requester(&pdev->dev) == 0) {
6337 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
6338 ixgbe_setup_dca(adapter);
6339 }
6340 #endif
6341 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6342 DPRINTK(PROBE, INFO, "IOV is enabled with %d VFs\n",
6343 adapter->num_vfs);
6344 for (i = 0; i < adapter->num_vfs; i++)
6345 ixgbe_vf_configuration(pdev, (i | 0x10000000));
6346 }
6347
6348 /* add san mac addr to netdev */
6349 ixgbe_add_sanmac_netdev(netdev);
6350
6351 dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
6352 cards_found++;
6353 return 0;
6354
6355 err_register:
6356 ixgbe_release_hw_control(adapter);
6357 ixgbe_clear_interrupt_scheme(adapter);
6358 err_sw_init:
6359 err_eeprom:
6360 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6361 ixgbe_disable_sriov(adapter);
6362 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6363 del_timer_sync(&adapter->sfp_timer);
6364 cancel_work_sync(&adapter->sfp_task);
6365 cancel_work_sync(&adapter->multispeed_fiber_task);
6366 cancel_work_sync(&adapter->sfp_config_module_task);
6367 iounmap(hw->hw_addr);
6368 err_ioremap:
6369 free_netdev(netdev);
6370 err_alloc_etherdev:
6371 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6372 IORESOURCE_MEM));
6373 err_pci_reg:
6374 err_dma:
6375 pci_disable_device(pdev);
6376 return err;
6377 }
6378
6379 /**
6380 * ixgbe_remove - Device Removal Routine
6381 * @pdev: PCI device information struct
6382 *
6383 * ixgbe_remove is called by the PCI subsystem to alert the driver
6384 * that it should release a PCI device. The could be caused by a
6385 * Hot-Plug event, or because the driver is going to be removed from
6386 * memory.
6387 **/
6388 static void __devexit ixgbe_remove(struct pci_dev *pdev)
6389 {
6390 struct net_device *netdev = pci_get_drvdata(pdev);
6391 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6392
6393 set_bit(__IXGBE_DOWN, &adapter->state);
6394 /* clear the module not found bit to make sure the worker won't
6395 * reschedule
6396 */
6397 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6398 del_timer_sync(&adapter->watchdog_timer);
6399
6400 del_timer_sync(&adapter->sfp_timer);
6401 cancel_work_sync(&adapter->watchdog_task);
6402 cancel_work_sync(&adapter->sfp_task);
6403 if (adapter->hw.phy.multispeed_fiber) {
6404 struct ixgbe_hw *hw = &adapter->hw;
6405 /*
6406 * Restart clause 37 autoneg, disable and re-enable
6407 * the tx laser, to clear & alert the link partner
6408 * that it needs to restart autotry
6409 */
6410 hw->mac.autotry_restart = true;
6411 hw->mac.ops.flap_tx_laser(hw);
6412 }
6413 cancel_work_sync(&adapter->multispeed_fiber_task);
6414 cancel_work_sync(&adapter->sfp_config_module_task);
6415 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6416 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6417 cancel_work_sync(&adapter->fdir_reinit_task);
6418 flush_scheduled_work();
6419
6420 #ifdef CONFIG_IXGBE_DCA
6421 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
6422 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
6423 dca_remove_requester(&pdev->dev);
6424 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
6425 }
6426
6427 #endif
6428 #ifdef IXGBE_FCOE
6429 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
6430 ixgbe_cleanup_fcoe(adapter);
6431
6432 #endif /* IXGBE_FCOE */
6433
6434 /* remove the added san mac */
6435 ixgbe_del_sanmac_netdev(netdev);
6436
6437 if (netdev->reg_state == NETREG_REGISTERED)
6438 unregister_netdev(netdev);
6439
6440 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6441 ixgbe_disable_sriov(adapter);
6442
6443 ixgbe_clear_interrupt_scheme(adapter);
6444
6445 ixgbe_release_hw_control(adapter);
6446
6447 iounmap(adapter->hw.hw_addr);
6448 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6449 IORESOURCE_MEM));
6450
6451 DPRINTK(PROBE, INFO, "complete\n");
6452
6453 free_netdev(netdev);
6454
6455 pci_disable_pcie_error_reporting(pdev);
6456
6457 pci_disable_device(pdev);
6458 }
6459
6460 /**
6461 * ixgbe_io_error_detected - called when PCI error is detected
6462 * @pdev: Pointer to PCI device
6463 * @state: The current pci connection state
6464 *
6465 * This function is called after a PCI bus error affecting
6466 * this device has been detected.
6467 */
6468 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
6469 pci_channel_state_t state)
6470 {
6471 struct net_device *netdev = pci_get_drvdata(pdev);
6472 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6473
6474 netif_device_detach(netdev);
6475
6476 if (state == pci_channel_io_perm_failure)
6477 return PCI_ERS_RESULT_DISCONNECT;
6478
6479 if (netif_running(netdev))
6480 ixgbe_down(adapter);
6481 pci_disable_device(pdev);
6482
6483 /* Request a slot reset. */
6484 return PCI_ERS_RESULT_NEED_RESET;
6485 }
6486
6487 /**
6488 * ixgbe_io_slot_reset - called after the pci bus has been reset.
6489 * @pdev: Pointer to PCI device
6490 *
6491 * Restart the card from scratch, as if from a cold-boot.
6492 */
6493 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
6494 {
6495 struct net_device *netdev = pci_get_drvdata(pdev);
6496 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6497 pci_ers_result_t result;
6498 int err;
6499
6500 if (pci_enable_device_mem(pdev)) {
6501 DPRINTK(PROBE, ERR,
6502 "Cannot re-enable PCI device after reset.\n");
6503 result = PCI_ERS_RESULT_DISCONNECT;
6504 } else {
6505 pci_set_master(pdev);
6506 pci_restore_state(pdev);
6507 pci_save_state(pdev);
6508
6509 pci_wake_from_d3(pdev, false);
6510
6511 ixgbe_reset(adapter);
6512 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6513 result = PCI_ERS_RESULT_RECOVERED;
6514 }
6515
6516 err = pci_cleanup_aer_uncorrect_error_status(pdev);
6517 if (err) {
6518 dev_err(&pdev->dev,
6519 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
6520 /* non-fatal, continue */
6521 }
6522
6523 return result;
6524 }
6525
6526 /**
6527 * ixgbe_io_resume - called when traffic can start flowing again.
6528 * @pdev: Pointer to PCI device
6529 *
6530 * This callback is called when the error recovery driver tells us that
6531 * its OK to resume normal operation.
6532 */
6533 static void ixgbe_io_resume(struct pci_dev *pdev)
6534 {
6535 struct net_device *netdev = pci_get_drvdata(pdev);
6536 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6537
6538 if (netif_running(netdev)) {
6539 if (ixgbe_up(adapter)) {
6540 DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
6541 return;
6542 }
6543 }
6544
6545 netif_device_attach(netdev);
6546 }
6547
6548 static struct pci_error_handlers ixgbe_err_handler = {
6549 .error_detected = ixgbe_io_error_detected,
6550 .slot_reset = ixgbe_io_slot_reset,
6551 .resume = ixgbe_io_resume,
6552 };
6553
6554 static struct pci_driver ixgbe_driver = {
6555 .name = ixgbe_driver_name,
6556 .id_table = ixgbe_pci_tbl,
6557 .probe = ixgbe_probe,
6558 .remove = __devexit_p(ixgbe_remove),
6559 #ifdef CONFIG_PM
6560 .suspend = ixgbe_suspend,
6561 .resume = ixgbe_resume,
6562 #endif
6563 .shutdown = ixgbe_shutdown,
6564 .err_handler = &ixgbe_err_handler
6565 };
6566
6567 /**
6568 * ixgbe_init_module - Driver Registration Routine
6569 *
6570 * ixgbe_init_module is the first routine called when the driver is
6571 * loaded. All it does is register with the PCI subsystem.
6572 **/
6573 static int __init ixgbe_init_module(void)
6574 {
6575 int ret;
6576 printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
6577 ixgbe_driver_string, ixgbe_driver_version);
6578
6579 printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
6580
6581 #ifdef CONFIG_IXGBE_DCA
6582 dca_register_notify(&dca_notifier);
6583 #endif
6584
6585 ret = pci_register_driver(&ixgbe_driver);
6586 return ret;
6587 }
6588
6589 module_init(ixgbe_init_module);
6590
6591 /**
6592 * ixgbe_exit_module - Driver Exit Cleanup Routine
6593 *
6594 * ixgbe_exit_module is called just before the driver is removed
6595 * from memory.
6596 **/
6597 static void __exit ixgbe_exit_module(void)
6598 {
6599 #ifdef CONFIG_IXGBE_DCA
6600 dca_unregister_notify(&dca_notifier);
6601 #endif
6602 pci_unregister_driver(&ixgbe_driver);
6603 }
6604
6605 #ifdef CONFIG_IXGBE_DCA
6606 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
6607 void *p)
6608 {
6609 int ret_val;
6610
6611 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
6612 __ixgbe_notify_dca);
6613
6614 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6615 }
6616
6617 #endif /* CONFIG_IXGBE_DCA */
6618 #ifdef DEBUG
6619 /**
6620 * ixgbe_get_hw_dev_name - return device name string
6621 * used by hardware layer to print debugging information
6622 **/
6623 char *ixgbe_get_hw_dev_name(struct ixgbe_hw *hw)
6624 {
6625 struct ixgbe_adapter *adapter = hw->back;
6626 return adapter->netdev->name;
6627 }
6628
6629 #endif
6630 module_exit(ixgbe_exit_module);
6631
6632 /* ixgbe_main.c */
kernel source for telekom puls (alcatel/mediatek)