projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'fix' of git://git.kernel.org/pub/scm/linux/kernel/git/ycmiao/pxa-linux-2.6
[deliverable/linux.git] / drivers / net / ixgbe / ixgbe_main.c
1 /*******************************************************************************
2
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 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 /* enable the optics */
2986 if (hw->phy.multispeed_fiber)
2987 hw->mac.ops.enable_tx_laser(hw);
2988
2989 clear_bit(__IXGBE_DOWN, &adapter->state);
2990 ixgbe_napi_enable_all(adapter);
2991
2992 /* clear any pending interrupts, may auto mask */
2993 IXGBE_READ_REG(hw, IXGBE_EICR);
2994
2995 ixgbe_irq_enable(adapter);
2996
2997 /*
2998 * If this adapter has a fan, check to see if we had a failure
2999 * before we enabled the interrupt.
3000 */
3001 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3002 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3003 if (esdp & IXGBE_ESDP_SDP1)
3004 DPRINTK(DRV, CRIT,
3005 "Fan has stopped, replace the adapter\n");
3006 }
3007
3008 /*
3009 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3010 * arrived before interrupts were enabled but after probe. Such
3011 * devices wouldn't have their type identified yet. We need to
3012 * kick off the SFP+ module setup first, then try to bring up link.
3013 * If we're not hot-pluggable SFP+, we just need to configure link
3014 * and bring it up.
3015 */
3016 if (hw->phy.type == ixgbe_phy_unknown) {
3017 err = hw->phy.ops.identify(hw);
3018 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3019 /*
3020 * Take the device down and schedule the sfp tasklet
3021 * which will unregister_netdev and log it.
3022 */
3023 ixgbe_down(adapter);
3024 schedule_work(&adapter->sfp_config_module_task);
3025 return err;
3026 }
3027 }
3028
3029 if (ixgbe_is_sfp(hw)) {
3030 ixgbe_sfp_link_config(adapter);
3031 } else {
3032 err = ixgbe_non_sfp_link_config(hw);
3033 if (err)
3034 DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
3035 }
3036
3037 for (i = 0; i < adapter->num_tx_queues; i++)
3038 set_bit(__IXGBE_FDIR_INIT_DONE,
3039 &(adapter->tx_ring[i]->reinit_state));
3040
3041 /* enable transmits */
3042 netif_tx_start_all_queues(netdev);
3043
3044 /* bring the link up in the watchdog, this could race with our first
3045 * link up interrupt but shouldn't be a problem */
3046 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3047 adapter->link_check_timeout = jiffies;
3048 mod_timer(&adapter->watchdog_timer, jiffies);
3049
3050 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3051 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3052 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3053 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3054
3055 return 0;
3056 }
3057
3058 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3059 {
3060 WARN_ON(in_interrupt());
3061 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3062 msleep(1);
3063 ixgbe_down(adapter);
3064 /*
3065 * If SR-IOV enabled then wait a bit before bringing the adapter
3066 * back up to give the VFs time to respond to the reset. The
3067 * two second wait is based upon the watchdog timer cycle in
3068 * the VF driver.
3069 */
3070 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3071 msleep(2000);
3072 ixgbe_up(adapter);
3073 clear_bit(__IXGBE_RESETTING, &adapter->state);
3074 }
3075
3076 int ixgbe_up(struct ixgbe_adapter *adapter)
3077 {
3078 /* hardware has been reset, we need to reload some things */
3079 ixgbe_configure(adapter);
3080
3081 return ixgbe_up_complete(adapter);
3082 }
3083
3084 void ixgbe_reset(struct ixgbe_adapter *adapter)
3085 {
3086 struct ixgbe_hw *hw = &adapter->hw;
3087 int err;
3088
3089 err = hw->mac.ops.init_hw(hw);
3090 switch (err) {
3091 case 0:
3092 case IXGBE_ERR_SFP_NOT_PRESENT:
3093 break;
3094 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3095 dev_err(&adapter->pdev->dev, "master disable timed out\n");
3096 break;
3097 case IXGBE_ERR_EEPROM_VERSION:
3098 /* We are running on a pre-production device, log a warning */
3099 dev_warn(&adapter->pdev->dev, "This device is a pre-production "
3100 "adapter/LOM. Please be aware there may be issues "
3101 "associated with your hardware. If you are "
3102 "experiencing problems please contact your Intel or "
3103 "hardware representative who provided you with this "
3104 "hardware.\n");
3105 break;
3106 default:
3107 dev_err(&adapter->pdev->dev, "Hardware Error: %d\n", err);
3108 }
3109
3110 /* reprogram the RAR[0] in case user changed it. */
3111 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3112 IXGBE_RAH_AV);
3113 }
3114
3115 /**
3116 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3117 * @adapter: board private structure
3118 * @rx_ring: ring to free buffers from
3119 **/
3120 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
3121 struct ixgbe_ring *rx_ring)
3122 {
3123 struct pci_dev *pdev = adapter->pdev;
3124 unsigned long size;
3125 unsigned int i;
3126
3127 /* Free all the Rx ring sk_buffs */
3128
3129 for (i = 0; i < rx_ring->count; i++) {
3130 struct ixgbe_rx_buffer *rx_buffer_info;
3131
3132 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3133 if (rx_buffer_info->dma) {
3134 pci_unmap_single(pdev, rx_buffer_info->dma,
3135 rx_ring->rx_buf_len,
3136 PCI_DMA_FROMDEVICE);
3137 rx_buffer_info->dma = 0;
3138 }
3139 if (rx_buffer_info->skb) {
3140 struct sk_buff *skb = rx_buffer_info->skb;
3141 rx_buffer_info->skb = NULL;
3142 do {
3143 struct sk_buff *this = skb;
3144 if (IXGBE_RSC_CB(this)->dma) {
3145 pci_unmap_single(pdev, IXGBE_RSC_CB(this)->dma,
3146 rx_ring->rx_buf_len,
3147 PCI_DMA_FROMDEVICE);
3148 IXGBE_RSC_CB(this)->dma = 0;
3149 }
3150 skb = skb->prev;
3151 dev_kfree_skb(this);
3152 } while (skb);
3153 }
3154 if (!rx_buffer_info->page)
3155 continue;
3156 if (rx_buffer_info->page_dma) {
3157 pci_unmap_page(pdev, rx_buffer_info->page_dma,
3158 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
3159 rx_buffer_info->page_dma = 0;
3160 }
3161 put_page(rx_buffer_info->page);
3162 rx_buffer_info->page = NULL;
3163 rx_buffer_info->page_offset = 0;
3164 }
3165
3166 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3167 memset(rx_ring->rx_buffer_info, 0, size);
3168
3169 /* Zero out the descriptor ring */
3170 memset(rx_ring->desc, 0, rx_ring->size);
3171
3172 rx_ring->next_to_clean = 0;
3173 rx_ring->next_to_use = 0;
3174
3175 if (rx_ring->head)
3176 writel(0, adapter->hw.hw_addr + rx_ring->head);
3177 if (rx_ring->tail)
3178 writel(0, adapter->hw.hw_addr + rx_ring->tail);
3179 }
3180
3181 /**
3182 * ixgbe_clean_tx_ring - Free Tx Buffers
3183 * @adapter: board private structure
3184 * @tx_ring: ring to be cleaned
3185 **/
3186 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
3187 struct ixgbe_ring *tx_ring)
3188 {
3189 struct ixgbe_tx_buffer *tx_buffer_info;
3190 unsigned long size;
3191 unsigned int i;
3192
3193 /* Free all the Tx ring sk_buffs */
3194
3195 for (i = 0; i < tx_ring->count; i++) {
3196 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3197 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3198 }
3199
3200 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3201 memset(tx_ring->tx_buffer_info, 0, size);
3202
3203 /* Zero out the descriptor ring */
3204 memset(tx_ring->desc, 0, tx_ring->size);
3205
3206 tx_ring->next_to_use = 0;
3207 tx_ring->next_to_clean = 0;
3208
3209 if (tx_ring->head)
3210 writel(0, adapter->hw.hw_addr + tx_ring->head);
3211 if (tx_ring->tail)
3212 writel(0, adapter->hw.hw_addr + tx_ring->tail);
3213 }
3214
3215 /**
3216 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3217 * @adapter: board private structure
3218 **/
3219 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3220 {
3221 int i;
3222
3223 for (i = 0; i < adapter->num_rx_queues; i++)
3224 ixgbe_clean_rx_ring(adapter, adapter->rx_ring[i]);
3225 }
3226
3227 /**
3228 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3229 * @adapter: board private structure
3230 **/
3231 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3232 {
3233 int i;
3234
3235 for (i = 0; i < adapter->num_tx_queues; i++)
3236 ixgbe_clean_tx_ring(adapter, adapter->tx_ring[i]);
3237 }
3238
3239 void ixgbe_down(struct ixgbe_adapter *adapter)
3240 {
3241 struct net_device *netdev = adapter->netdev;
3242 struct ixgbe_hw *hw = &adapter->hw;
3243 u32 rxctrl;
3244 u32 txdctl;
3245 int i, j;
3246
3247 /* signal that we are down to the interrupt handler */
3248 set_bit(__IXGBE_DOWN, &adapter->state);
3249
3250 /* power down the optics */
3251 if (hw->phy.multispeed_fiber)
3252 hw->mac.ops.disable_tx_laser(hw);
3253
3254 /* disable receive for all VFs and wait one second */
3255 if (adapter->num_vfs) {
3256 /* ping all the active vfs to let them know we are going down */
3257 ixgbe_ping_all_vfs(adapter);
3258
3259 /* Disable all VFTE/VFRE TX/RX */
3260 ixgbe_disable_tx_rx(adapter);
3261
3262 /* Mark all the VFs as inactive */
3263 for (i = 0 ; i < adapter->num_vfs; i++)
3264 adapter->vfinfo[i].clear_to_send = 0;
3265 }
3266
3267 /* disable receives */
3268 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3269 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3270
3271 netif_tx_disable(netdev);
3272
3273 IXGBE_WRITE_FLUSH(hw);
3274 msleep(10);
3275
3276 netif_tx_stop_all_queues(netdev);
3277
3278 ixgbe_irq_disable(adapter);
3279
3280 ixgbe_napi_disable_all(adapter);
3281
3282 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3283 del_timer_sync(&adapter->sfp_timer);
3284 del_timer_sync(&adapter->watchdog_timer);
3285 cancel_work_sync(&adapter->watchdog_task);
3286
3287 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3288 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3289 cancel_work_sync(&adapter->fdir_reinit_task);
3290
3291 /* disable transmits in the hardware now that interrupts are off */
3292 for (i = 0; i < adapter->num_tx_queues; i++) {
3293 j = adapter->tx_ring[i]->reg_idx;
3294 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3295 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3296 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3297 }
3298 /* Disable the Tx DMA engine on 82599 */
3299 if (hw->mac.type == ixgbe_mac_82599EB)
3300 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3301 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3302 ~IXGBE_DMATXCTL_TE));
3303
3304 netif_carrier_off(netdev);
3305
3306 /* clear n-tuple filters that are cached */
3307 ethtool_ntuple_flush(netdev);
3308
3309 if (!pci_channel_offline(adapter->pdev))
3310 ixgbe_reset(adapter);
3311 ixgbe_clean_all_tx_rings(adapter);
3312 ixgbe_clean_all_rx_rings(adapter);
3313
3314 #ifdef CONFIG_IXGBE_DCA
3315 /* since we reset the hardware DCA settings were cleared */
3316 ixgbe_setup_dca(adapter);
3317 #endif
3318 }
3319
3320 /**
3321 * ixgbe_poll - NAPI Rx polling callback
3322 * @napi: structure for representing this polling device
3323 * @budget: how many packets driver is allowed to clean
3324 *
3325 * This function is used for legacy and MSI, NAPI mode
3326 **/
3327 static int ixgbe_poll(struct napi_struct *napi, int budget)
3328 {
3329 struct ixgbe_q_vector *q_vector =
3330 container_of(napi, struct ixgbe_q_vector, napi);
3331 struct ixgbe_adapter *adapter = q_vector->adapter;
3332 int tx_clean_complete, work_done = 0;
3333
3334 #ifdef CONFIG_IXGBE_DCA
3335 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3336 ixgbe_update_tx_dca(adapter, adapter->tx_ring[0]);
3337 ixgbe_update_rx_dca(adapter, adapter->rx_ring[0]);
3338 }
3339 #endif
3340
3341 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3342 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3343
3344 if (!tx_clean_complete)
3345 work_done = budget;
3346
3347 /* If budget not fully consumed, exit the polling mode */
3348 if (work_done < budget) {
3349 napi_complete(napi);
3350 if (adapter->rx_itr_setting & 1)
3351 ixgbe_set_itr(adapter);
3352 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3353 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3354 }
3355 return work_done;
3356 }
3357
3358 /**
3359 * ixgbe_tx_timeout - Respond to a Tx Hang
3360 * @netdev: network interface device structure
3361 **/
3362 static void ixgbe_tx_timeout(struct net_device *netdev)
3363 {
3364 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3365
3366 /* Do the reset outside of interrupt context */
3367 schedule_work(&adapter->reset_task);
3368 }
3369
3370 static void ixgbe_reset_task(struct work_struct *work)
3371 {
3372 struct ixgbe_adapter *adapter;
3373 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3374
3375 /* If we're already down or resetting, just bail */
3376 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3377 test_bit(__IXGBE_RESETTING, &adapter->state))
3378 return;
3379
3380 adapter->tx_timeout_count++;
3381
3382 ixgbe_reinit_locked(adapter);
3383 }
3384
3385 #ifdef CONFIG_IXGBE_DCB
3386 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3387 {
3388 bool ret = false;
3389 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3390
3391 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3392 return ret;
3393
3394 f->mask = 0x7 << 3;
3395 adapter->num_rx_queues = f->indices;
3396 adapter->num_tx_queues = f->indices;
3397 ret = true;
3398
3399 return ret;
3400 }
3401 #endif
3402
3403 /**
3404 * ixgbe_set_rss_queues: Allocate queues for RSS
3405 * @adapter: board private structure to initialize
3406 *
3407 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3408 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3409 *
3410 **/
3411 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3412 {
3413 bool ret = false;
3414 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3415
3416 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3417 f->mask = 0xF;
3418 adapter->num_rx_queues = f->indices;
3419 adapter->num_tx_queues = f->indices;
3420 ret = true;
3421 } else {
3422 ret = false;
3423 }
3424
3425 return ret;
3426 }
3427
3428 /**
3429 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3430 * @adapter: board private structure to initialize
3431 *
3432 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3433 * to the original CPU that initiated the Tx session. This runs in addition
3434 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3435 * Rx load across CPUs using RSS.
3436 *
3437 **/
3438 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3439 {
3440 bool ret = false;
3441 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3442
3443 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3444 f_fdir->mask = 0;
3445
3446 /* Flow Director must have RSS enabled */
3447 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3448 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3449 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3450 adapter->num_tx_queues = f_fdir->indices;
3451 adapter->num_rx_queues = f_fdir->indices;
3452 ret = true;
3453 } else {
3454 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3455 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3456 }
3457 return ret;
3458 }
3459
3460 #ifdef IXGBE_FCOE
3461 /**
3462 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3463 * @adapter: board private structure to initialize
3464 *
3465 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3466 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3467 * rx queues out of the max number of rx queues, instead, it is used as the
3468 * index of the first rx queue used by FCoE.
3469 *
3470 **/
3471 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3472 {
3473 bool ret = false;
3474 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3475
3476 f->indices = min((int)num_online_cpus(), f->indices);
3477 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3478 adapter->num_rx_queues = 1;
3479 adapter->num_tx_queues = 1;
3480 #ifdef CONFIG_IXGBE_DCB
3481 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3482 DPRINTK(PROBE, INFO, "FCoE enabled with DCB \n");
3483 ixgbe_set_dcb_queues(adapter);
3484 }
3485 #endif
3486 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3487 DPRINTK(PROBE, INFO, "FCoE enabled with RSS \n");
3488 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3489 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3490 ixgbe_set_fdir_queues(adapter);
3491 else
3492 ixgbe_set_rss_queues(adapter);
3493 }
3494 /* adding FCoE rx rings to the end */
3495 f->mask = adapter->num_rx_queues;
3496 adapter->num_rx_queues += f->indices;
3497 adapter->num_tx_queues += f->indices;
3498
3499 ret = true;
3500 }
3501
3502 return ret;
3503 }
3504
3505 #endif /* IXGBE_FCOE */
3506 /**
3507 * ixgbe_set_sriov_queues: Allocate queues for IOV use
3508 * @adapter: board private structure to initialize
3509 *
3510 * IOV doesn't actually use anything, so just NAK the
3511 * request for now and let the other queue routines
3512 * figure out what to do.
3513 */
3514 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
3515 {
3516 return false;
3517 }
3518
3519 /*
3520 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3521 * @adapter: board private structure to initialize
3522 *
3523 * This is the top level queue allocation routine. The order here is very
3524 * important, starting with the "most" number of features turned on at once,
3525 * and ending with the smallest set of features. This way large combinations
3526 * can be allocated if they're turned on, and smaller combinations are the
3527 * fallthrough conditions.
3528 *
3529 **/
3530 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
3531 {
3532 /* Start with base case */
3533 adapter->num_rx_queues = 1;
3534 adapter->num_tx_queues = 1;
3535 adapter->num_rx_pools = adapter->num_rx_queues;
3536 adapter->num_rx_queues_per_pool = 1;
3537
3538 if (ixgbe_set_sriov_queues(adapter))
3539 return;
3540
3541 #ifdef IXGBE_FCOE
3542 if (ixgbe_set_fcoe_queues(adapter))
3543 goto done;
3544
3545 #endif /* IXGBE_FCOE */
3546 #ifdef CONFIG_IXGBE_DCB
3547 if (ixgbe_set_dcb_queues(adapter))
3548 goto done;
3549
3550 #endif
3551 if (ixgbe_set_fdir_queues(adapter))
3552 goto done;
3553
3554 if (ixgbe_set_rss_queues(adapter))
3555 goto done;
3556
3557 /* fallback to base case */
3558 adapter->num_rx_queues = 1;
3559 adapter->num_tx_queues = 1;
3560
3561 done:
3562 /* Notify the stack of the (possibly) reduced Tx Queue count. */
3563 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
3564 }
3565
3566 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
3567 int vectors)
3568 {
3569 int err, vector_threshold;
3570
3571 /* We'll want at least 3 (vector_threshold):
3572 * 1) TxQ[0] Cleanup
3573 * 2) RxQ[0] Cleanup
3574 * 3) Other (Link Status Change, etc.)
3575 * 4) TCP Timer (optional)
3576 */
3577 vector_threshold = MIN_MSIX_COUNT;
3578
3579 /* The more we get, the more we will assign to Tx/Rx Cleanup
3580 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
3581 * Right now, we simply care about how many we'll get; we'll
3582 * set them up later while requesting irq's.
3583 */
3584 while (vectors >= vector_threshold) {
3585 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
3586 vectors);
3587 if (!err) /* Success in acquiring all requested vectors. */
3588 break;
3589 else if (err < 0)
3590 vectors = 0; /* Nasty failure, quit now */
3591 else /* err == number of vectors we should try again with */
3592 vectors = err;
3593 }
3594
3595 if (vectors < vector_threshold) {
3596 /* Can't allocate enough MSI-X interrupts? Oh well.
3597 * This just means we'll go with either a single MSI
3598 * vector or fall back to legacy interrupts.
3599 */
3600 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
3601 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3602 kfree(adapter->msix_entries);
3603 adapter->msix_entries = NULL;
3604 } else {
3605 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
3606 /*
3607 * Adjust for only the vectors we'll use, which is minimum
3608 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
3609 * vectors we were allocated.
3610 */
3611 adapter->num_msix_vectors = min(vectors,
3612 adapter->max_msix_q_vectors + NON_Q_VECTORS);
3613 }
3614 }
3615
3616 /**
3617 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
3618 * @adapter: board private structure to initialize
3619 *
3620 * Cache the descriptor ring offsets for RSS to the assigned rings.
3621 *
3622 **/
3623 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
3624 {
3625 int i;
3626 bool ret = false;
3627
3628 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3629 for (i = 0; i < adapter->num_rx_queues; i++)
3630 adapter->rx_ring[i]->reg_idx = i;
3631 for (i = 0; i < adapter->num_tx_queues; i++)
3632 adapter->tx_ring[i]->reg_idx = i;
3633 ret = true;
3634 } else {
3635 ret = false;
3636 }
3637
3638 return ret;
3639 }
3640
3641 #ifdef CONFIG_IXGBE_DCB
3642 /**
3643 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
3644 * @adapter: board private structure to initialize
3645 *
3646 * Cache the descriptor ring offsets for DCB to the assigned rings.
3647 *
3648 **/
3649 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
3650 {
3651 int i;
3652 bool ret = false;
3653 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
3654
3655 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3656 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3657 /* the number of queues is assumed to be symmetric */
3658 for (i = 0; i < dcb_i; i++) {
3659 adapter->rx_ring[i]->reg_idx = i << 3;
3660 adapter->tx_ring[i]->reg_idx = i << 2;
3661 }
3662 ret = true;
3663 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
3664 if (dcb_i == 8) {
3665 /*
3666 * Tx TC0 starts at: descriptor queue 0
3667 * Tx TC1 starts at: descriptor queue 32
3668 * Tx TC2 starts at: descriptor queue 64
3669 * Tx TC3 starts at: descriptor queue 80
3670 * Tx TC4 starts at: descriptor queue 96
3671 * Tx TC5 starts at: descriptor queue 104
3672 * Tx TC6 starts at: descriptor queue 112
3673 * Tx TC7 starts at: descriptor queue 120
3674 *
3675 * Rx TC0-TC7 are offset by 16 queues each
3676 */
3677 for (i = 0; i < 3; i++) {
3678 adapter->tx_ring[i]->reg_idx = i << 5;
3679 adapter->rx_ring[i]->reg_idx = i << 4;
3680 }
3681 for ( ; i < 5; i++) {
3682 adapter->tx_ring[i]->reg_idx =
3683 ((i + 2) << 4);
3684 adapter->rx_ring[i]->reg_idx = i << 4;
3685 }
3686 for ( ; i < dcb_i; i++) {
3687 adapter->tx_ring[i]->reg_idx =
3688 ((i + 8) << 3);
3689 adapter->rx_ring[i]->reg_idx = i << 4;
3690 }
3691
3692 ret = true;
3693 } else if (dcb_i == 4) {
3694 /*
3695 * Tx TC0 starts at: descriptor queue 0
3696 * Tx TC1 starts at: descriptor queue 64
3697 * Tx TC2 starts at: descriptor queue 96
3698 * Tx TC3 starts at: descriptor queue 112
3699 *
3700 * Rx TC0-TC3 are offset by 32 queues each
3701 */
3702 adapter->tx_ring[0]->reg_idx = 0;
3703 adapter->tx_ring[1]->reg_idx = 64;
3704 adapter->tx_ring[2]->reg_idx = 96;
3705 adapter->tx_ring[3]->reg_idx = 112;
3706 for (i = 0 ; i < dcb_i; i++)
3707 adapter->rx_ring[i]->reg_idx = i << 5;
3708
3709 ret = true;
3710 } else {
3711 ret = false;
3712 }
3713 } else {
3714 ret = false;
3715 }
3716 } else {
3717 ret = false;
3718 }
3719
3720 return ret;
3721 }
3722 #endif
3723
3724 /**
3725 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
3726 * @adapter: board private structure to initialize
3727 *
3728 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
3729 *
3730 **/
3731 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
3732 {
3733 int i;
3734 bool ret = false;
3735
3736 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3737 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3738 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
3739 for (i = 0; i < adapter->num_rx_queues; i++)
3740 adapter->rx_ring[i]->reg_idx = i;
3741 for (i = 0; i < adapter->num_tx_queues; i++)
3742 adapter->tx_ring[i]->reg_idx = i;
3743 ret = true;
3744 }
3745
3746 return ret;
3747 }
3748
3749 #ifdef IXGBE_FCOE
3750 /**
3751 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
3752 * @adapter: board private structure to initialize
3753 *
3754 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
3755 *
3756 */
3757 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
3758 {
3759 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
3760 bool ret = false;
3761 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3762
3763 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3764 #ifdef CONFIG_IXGBE_DCB
3765 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3766 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
3767
3768 ixgbe_cache_ring_dcb(adapter);
3769 /* find out queues in TC for FCoE */
3770 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
3771 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
3772 /*
3773 * In 82599, the number of Tx queues for each traffic
3774 * class for both 8-TC and 4-TC modes are:
3775 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
3776 * 8 TCs: 32 32 16 16 8 8 8 8
3777 * 4 TCs: 64 64 32 32
3778 * We have max 8 queues for FCoE, where 8 the is
3779 * FCoE redirection table size. If TC for FCoE is
3780 * less than or equal to TC3, we have enough queues
3781 * to add max of 8 queues for FCoE, so we start FCoE
3782 * tx descriptor from the next one, i.e., reg_idx + 1.
3783 * If TC for FCoE is above TC3, implying 8 TC mode,
3784 * and we need 8 for FCoE, we have to take all queues
3785 * in that traffic class for FCoE.
3786 */
3787 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
3788 fcoe_tx_i--;
3789 }
3790 #endif /* CONFIG_IXGBE_DCB */
3791 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3792 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3793 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3794 ixgbe_cache_ring_fdir(adapter);
3795 else
3796 ixgbe_cache_ring_rss(adapter);
3797
3798 fcoe_rx_i = f->mask;
3799 fcoe_tx_i = f->mask;
3800 }
3801 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
3802 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
3803 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
3804 }
3805 ret = true;
3806 }
3807 return ret;
3808 }
3809
3810 #endif /* IXGBE_FCOE */
3811 /**
3812 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
3813 * @adapter: board private structure to initialize
3814 *
3815 * SR-IOV doesn't use any descriptor rings but changes the default if
3816 * no other mapping is used.
3817 *
3818 */
3819 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
3820 {
3821 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
3822 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
3823 if (adapter->num_vfs)
3824 return true;
3825 else
3826 return false;
3827 }
3828
3829 /**
3830 * ixgbe_cache_ring_register - Descriptor ring to register mapping
3831 * @adapter: board private structure to initialize
3832 *
3833 * Once we know the feature-set enabled for the device, we'll cache
3834 * the register offset the descriptor ring is assigned to.
3835 *
3836 * Note, the order the various feature calls is important. It must start with
3837 * the "most" features enabled at the same time, then trickle down to the
3838 * least amount of features turned on at once.
3839 **/
3840 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
3841 {
3842 /* start with default case */
3843 adapter->rx_ring[0]->reg_idx = 0;
3844 adapter->tx_ring[0]->reg_idx = 0;
3845
3846 if (ixgbe_cache_ring_sriov(adapter))
3847 return;
3848
3849 #ifdef IXGBE_FCOE
3850 if (ixgbe_cache_ring_fcoe(adapter))
3851 return;
3852
3853 #endif /* IXGBE_FCOE */
3854 #ifdef CONFIG_IXGBE_DCB
3855 if (ixgbe_cache_ring_dcb(adapter))
3856 return;
3857
3858 #endif
3859 if (ixgbe_cache_ring_fdir(adapter))
3860 return;
3861
3862 if (ixgbe_cache_ring_rss(adapter))
3863 return;
3864 }
3865
3866 /**
3867 * ixgbe_alloc_queues - Allocate memory for all rings
3868 * @adapter: board private structure to initialize
3869 *
3870 * We allocate one ring per queue at run-time since we don't know the
3871 * number of queues at compile-time. The polling_netdev array is
3872 * intended for Multiqueue, but should work fine with a single queue.
3873 **/
3874 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
3875 {
3876 int i;
3877 int orig_node = adapter->node;
3878
3879 for (i = 0; i < adapter->num_tx_queues; i++) {
3880 struct ixgbe_ring *ring = adapter->tx_ring[i];
3881 if (orig_node == -1) {
3882 int cur_node = next_online_node(adapter->node);
3883 if (cur_node == MAX_NUMNODES)
3884 cur_node = first_online_node;
3885 adapter->node = cur_node;
3886 }
3887 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
3888 adapter->node);
3889 if (!ring)
3890 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
3891 if (!ring)
3892 goto err_tx_ring_allocation;
3893 ring->count = adapter->tx_ring_count;
3894 ring->queue_index = i;
3895 ring->numa_node = adapter->node;
3896
3897 adapter->tx_ring[i] = ring;
3898 }
3899
3900 /* Restore the adapter's original node */
3901 adapter->node = orig_node;
3902
3903 for (i = 0; i < adapter->num_rx_queues; i++) {
3904 struct ixgbe_ring *ring = adapter->rx_ring[i];
3905 if (orig_node == -1) {
3906 int cur_node = next_online_node(adapter->node);
3907 if (cur_node == MAX_NUMNODES)
3908 cur_node = first_online_node;
3909 adapter->node = cur_node;
3910 }
3911 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
3912 adapter->node);
3913 if (!ring)
3914 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
3915 if (!ring)
3916 goto err_rx_ring_allocation;
3917 ring->count = adapter->rx_ring_count;
3918 ring->queue_index = i;
3919 ring->numa_node = adapter->node;
3920
3921 adapter->rx_ring[i] = ring;
3922 }
3923
3924 /* Restore the adapter's original node */
3925 adapter->node = orig_node;
3926
3927 ixgbe_cache_ring_register(adapter);
3928
3929 return 0;
3930
3931 err_rx_ring_allocation:
3932 for (i = 0; i < adapter->num_tx_queues; i++)
3933 kfree(adapter->tx_ring[i]);
3934 err_tx_ring_allocation:
3935 return -ENOMEM;
3936 }
3937
3938 /**
3939 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
3940 * @adapter: board private structure to initialize
3941 *
3942 * Attempt to configure the interrupts using the best available
3943 * capabilities of the hardware and the kernel.
3944 **/
3945 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
3946 {
3947 struct ixgbe_hw *hw = &adapter->hw;
3948 int err = 0;
3949 int vector, v_budget;
3950
3951 /*
3952 * It's easy to be greedy for MSI-X vectors, but it really
3953 * doesn't do us much good if we have a lot more vectors
3954 * than CPU's. So let's be conservative and only ask for
3955 * (roughly) the same number of vectors as there are CPU's.
3956 */
3957 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
3958 (int)num_online_cpus()) + NON_Q_VECTORS;
3959
3960 /*
3961 * At the same time, hardware can only support a maximum of
3962 * hw.mac->max_msix_vectors vectors. With features
3963 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
3964 * descriptor queues supported by our device. Thus, we cap it off in
3965 * those rare cases where the cpu count also exceeds our vector limit.
3966 */
3967 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
3968
3969 /* A failure in MSI-X entry allocation isn't fatal, but it does
3970 * mean we disable MSI-X capabilities of the adapter. */
3971 adapter->msix_entries = kcalloc(v_budget,
3972 sizeof(struct msix_entry), GFP_KERNEL);
3973 if (adapter->msix_entries) {
3974 for (vector = 0; vector < v_budget; vector++)
3975 adapter->msix_entries[vector].entry = vector;
3976
3977 ixgbe_acquire_msix_vectors(adapter, v_budget);
3978
3979 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3980 goto out;
3981 }
3982
3983 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
3984 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
3985 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3986 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3987 adapter->atr_sample_rate = 0;
3988 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3989 ixgbe_disable_sriov(adapter);
3990
3991 ixgbe_set_num_queues(adapter);
3992
3993 err = pci_enable_msi(adapter->pdev);
3994 if (!err) {
3995 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
3996 } else {
3997 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
3998 "falling back to legacy. Error: %d\n", err);
3999 /* reset err */
4000 err = 0;
4001 }
4002
4003 out:
4004 return err;
4005 }
4006
4007 /**
4008 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4009 * @adapter: board private structure to initialize
4010 *
4011 * We allocate one q_vector per queue interrupt. If allocation fails we
4012 * return -ENOMEM.
4013 **/
4014 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4015 {
4016 int q_idx, num_q_vectors;
4017 struct ixgbe_q_vector *q_vector;
4018 int napi_vectors;
4019 int (*poll)(struct napi_struct *, int);
4020
4021 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4022 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4023 napi_vectors = adapter->num_rx_queues;
4024 poll = &ixgbe_clean_rxtx_many;
4025 } else {
4026 num_q_vectors = 1;
4027 napi_vectors = 1;
4028 poll = &ixgbe_poll;
4029 }
4030
4031 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4032 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4033 GFP_KERNEL, adapter->node);
4034 if (!q_vector)
4035 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4036 GFP_KERNEL);
4037 if (!q_vector)
4038 goto err_out;
4039 q_vector->adapter = adapter;
4040 if (q_vector->txr_count && !q_vector->rxr_count)
4041 q_vector->eitr = adapter->tx_eitr_param;
4042 else
4043 q_vector->eitr = adapter->rx_eitr_param;
4044 q_vector->v_idx = q_idx;
4045 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4046 adapter->q_vector[q_idx] = q_vector;
4047 }
4048
4049 return 0;
4050
4051 err_out:
4052 while (q_idx) {
4053 q_idx--;
4054 q_vector = adapter->q_vector[q_idx];
4055 netif_napi_del(&q_vector->napi);
4056 kfree(q_vector);
4057 adapter->q_vector[q_idx] = NULL;
4058 }
4059 return -ENOMEM;
4060 }
4061
4062 /**
4063 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4064 * @adapter: board private structure to initialize
4065 *
4066 * This function frees the memory allocated to the q_vectors. In addition if
4067 * NAPI is enabled it will delete any references to the NAPI struct prior
4068 * to freeing the q_vector.
4069 **/
4070 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4071 {
4072 int q_idx, num_q_vectors;
4073
4074 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4075 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4076 else
4077 num_q_vectors = 1;
4078
4079 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4080 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4081 adapter->q_vector[q_idx] = NULL;
4082 netif_napi_del(&q_vector->napi);
4083 kfree(q_vector);
4084 }
4085 }
4086
4087 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4088 {
4089 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4090 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4091 pci_disable_msix(adapter->pdev);
4092 kfree(adapter->msix_entries);
4093 adapter->msix_entries = NULL;
4094 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4095 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4096 pci_disable_msi(adapter->pdev);
4097 }
4098 return;
4099 }
4100
4101 /**
4102 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4103 * @adapter: board private structure to initialize
4104 *
4105 * We determine which interrupt scheme to use based on...
4106 * - Kernel support (MSI, MSI-X)
4107 * - which can be user-defined (via MODULE_PARAM)
4108 * - Hardware queue count (num_*_queues)
4109 * - defined by miscellaneous hardware support/features (RSS, etc.)
4110 **/
4111 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4112 {
4113 int err;
4114
4115 /* Number of supported queues */
4116 ixgbe_set_num_queues(adapter);
4117
4118 err = ixgbe_set_interrupt_capability(adapter);
4119 if (err) {
4120 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
4121 goto err_set_interrupt;
4122 }
4123
4124 err = ixgbe_alloc_q_vectors(adapter);
4125 if (err) {
4126 DPRINTK(PROBE, ERR, "Unable to allocate memory for queue "
4127 "vectors\n");
4128 goto err_alloc_q_vectors;
4129 }
4130
4131 err = ixgbe_alloc_queues(adapter);
4132 if (err) {
4133 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
4134 goto err_alloc_queues;
4135 }
4136
4137 DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
4138 "Tx Queue count = %u\n",
4139 (adapter->num_rx_queues > 1) ? "Enabled" :
4140 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
4141
4142 set_bit(__IXGBE_DOWN, &adapter->state);
4143
4144 return 0;
4145
4146 err_alloc_queues:
4147 ixgbe_free_q_vectors(adapter);
4148 err_alloc_q_vectors:
4149 ixgbe_reset_interrupt_capability(adapter);
4150 err_set_interrupt:
4151 return err;
4152 }
4153
4154 /**
4155 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4156 * @adapter: board private structure to clear interrupt scheme on
4157 *
4158 * We go through and clear interrupt specific resources and reset the structure
4159 * to pre-load conditions
4160 **/
4161 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4162 {
4163 int i;
4164
4165 for (i = 0; i < adapter->num_tx_queues; i++) {
4166 kfree(adapter->tx_ring[i]);
4167 adapter->tx_ring[i] = NULL;
4168 }
4169 for (i = 0; i < adapter->num_rx_queues; i++) {
4170 kfree(adapter->rx_ring[i]);
4171 adapter->rx_ring[i] = NULL;
4172 }
4173
4174 ixgbe_free_q_vectors(adapter);
4175 ixgbe_reset_interrupt_capability(adapter);
4176 }
4177
4178 /**
4179 * ixgbe_sfp_timer - worker thread to find a missing module
4180 * @data: pointer to our adapter struct
4181 **/
4182 static void ixgbe_sfp_timer(unsigned long data)
4183 {
4184 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4185
4186 /*
4187 * Do the sfp_timer outside of interrupt context due to the
4188 * delays that sfp+ detection requires
4189 */
4190 schedule_work(&adapter->sfp_task);
4191 }
4192
4193 /**
4194 * ixgbe_sfp_task - worker thread to find a missing module
4195 * @work: pointer to work_struct containing our data
4196 **/
4197 static void ixgbe_sfp_task(struct work_struct *work)
4198 {
4199 struct ixgbe_adapter *adapter = container_of(work,
4200 struct ixgbe_adapter,
4201 sfp_task);
4202 struct ixgbe_hw *hw = &adapter->hw;
4203
4204 if ((hw->phy.type == ixgbe_phy_nl) &&
4205 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4206 s32 ret = hw->phy.ops.identify_sfp(hw);
4207 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4208 goto reschedule;
4209 ret = hw->phy.ops.reset(hw);
4210 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4211 dev_err(&adapter->pdev->dev, "failed to initialize "
4212 "because an unsupported SFP+ module type "
4213 "was detected.\n"
4214 "Reload the driver after installing a "
4215 "supported module.\n");
4216 unregister_netdev(adapter->netdev);
4217 } else {
4218 DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
4219 hw->phy.sfp_type);
4220 }
4221 /* don't need this routine any more */
4222 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4223 }
4224 return;
4225 reschedule:
4226 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4227 mod_timer(&adapter->sfp_timer,
4228 round_jiffies(jiffies + (2 * HZ)));
4229 }
4230
4231 /**
4232 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4233 * @adapter: board private structure to initialize
4234 *
4235 * ixgbe_sw_init initializes the Adapter private data structure.
4236 * Fields are initialized based on PCI device information and
4237 * OS network device settings (MTU size).
4238 **/
4239 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4240 {
4241 struct ixgbe_hw *hw = &adapter->hw;
4242 struct pci_dev *pdev = adapter->pdev;
4243 struct net_device *dev = adapter->netdev;
4244 unsigned int rss;
4245 #ifdef CONFIG_IXGBE_DCB
4246 int j;
4247 struct tc_configuration *tc;
4248 #endif
4249
4250 /* PCI config space info */
4251
4252 hw->vendor_id = pdev->vendor;
4253 hw->device_id = pdev->device;
4254 hw->revision_id = pdev->revision;
4255 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4256 hw->subsystem_device_id = pdev->subsystem_device;
4257
4258 /* Set capability flags */
4259 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4260 adapter->ring_feature[RING_F_RSS].indices = rss;
4261 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4262 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4263 if (hw->mac.type == ixgbe_mac_82598EB) {
4264 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4265 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4266 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4267 } else if (hw->mac.type == ixgbe_mac_82599EB) {
4268 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4269 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4270 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4271 if (dev->features & NETIF_F_NTUPLE) {
4272 /* Flow Director perfect filter enabled */
4273 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4274 adapter->atr_sample_rate = 0;
4275 spin_lock_init(&adapter->fdir_perfect_lock);
4276 } else {
4277 /* Flow Director hash filters enabled */
4278 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4279 adapter->atr_sample_rate = 20;
4280 }
4281 adapter->ring_feature[RING_F_FDIR].indices =
4282 IXGBE_MAX_FDIR_INDICES;
4283 adapter->fdir_pballoc = 0;
4284 #ifdef IXGBE_FCOE
4285 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4286 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4287 adapter->ring_feature[RING_F_FCOE].indices = 0;
4288 #ifdef CONFIG_IXGBE_DCB
4289 /* Default traffic class to use for FCoE */
4290 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4291 #endif
4292 #endif /* IXGBE_FCOE */
4293 }
4294
4295 #ifdef CONFIG_IXGBE_DCB
4296 /* Configure DCB traffic classes */
4297 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4298 tc = &adapter->dcb_cfg.tc_config[j];
4299 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4300 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4301 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4302 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4303 tc->dcb_pfc = pfc_disabled;
4304 }
4305 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4306 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4307 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4308 adapter->dcb_cfg.pfc_mode_enable = false;
4309 adapter->dcb_cfg.round_robin_enable = false;
4310 adapter->dcb_set_bitmap = 0x00;
4311 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4312 adapter->ring_feature[RING_F_DCB].indices);
4313
4314 #endif
4315
4316 /* default flow control settings */
4317 hw->fc.requested_mode = ixgbe_fc_full;
4318 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4319 #ifdef CONFIG_DCB
4320 adapter->last_lfc_mode = hw->fc.current_mode;
4321 #endif
4322 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
4323 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
4324 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4325 hw->fc.send_xon = true;
4326 hw->fc.disable_fc_autoneg = false;
4327
4328 /* enable itr by default in dynamic mode */
4329 adapter->rx_itr_setting = 1;
4330 adapter->rx_eitr_param = 20000;
4331 adapter->tx_itr_setting = 1;
4332 adapter->tx_eitr_param = 10000;
4333
4334 /* set defaults for eitr in MegaBytes */
4335 adapter->eitr_low = 10;
4336 adapter->eitr_high = 20;
4337
4338 /* set default ring sizes */
4339 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4340 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4341
4342 /* initialize eeprom parameters */
4343 if (ixgbe_init_eeprom_params_generic(hw)) {
4344 dev_err(&pdev->dev, "EEPROM initialization failed\n");
4345 return -EIO;
4346 }
4347
4348 /* enable rx csum by default */
4349 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4350
4351 /* get assigned NUMA node */
4352 adapter->node = dev_to_node(&pdev->dev);
4353
4354 set_bit(__IXGBE_DOWN, &adapter->state);
4355
4356 return 0;
4357 }
4358
4359 /**
4360 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4361 * @adapter: board private structure
4362 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4363 *
4364 * Return 0 on success, negative on failure
4365 **/
4366 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4367 struct ixgbe_ring *tx_ring)
4368 {
4369 struct pci_dev *pdev = adapter->pdev;
4370 int size;
4371
4372 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4373 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4374 if (!tx_ring->tx_buffer_info)
4375 tx_ring->tx_buffer_info = vmalloc(size);
4376 if (!tx_ring->tx_buffer_info)
4377 goto err;
4378 memset(tx_ring->tx_buffer_info, 0, size);
4379
4380 /* round up to nearest 4K */
4381 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4382 tx_ring->size = ALIGN(tx_ring->size, 4096);
4383
4384 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
4385 &tx_ring->dma);
4386 if (!tx_ring->desc)
4387 goto err;
4388
4389 tx_ring->next_to_use = 0;
4390 tx_ring->next_to_clean = 0;
4391 tx_ring->work_limit = tx_ring->count;
4392 return 0;
4393
4394 err:
4395 vfree(tx_ring->tx_buffer_info);
4396 tx_ring->tx_buffer_info = NULL;
4397 DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
4398 "descriptor ring\n");
4399 return -ENOMEM;
4400 }
4401
4402 /**
4403 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4404 * @adapter: board private structure
4405 *
4406 * If this function returns with an error, then it's possible one or
4407 * more of the rings is populated (while the rest are not). It is the
4408 * callers duty to clean those orphaned rings.
4409 *
4410 * Return 0 on success, negative on failure
4411 **/
4412 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4413 {
4414 int i, err = 0;
4415
4416 for (i = 0; i < adapter->num_tx_queues; i++) {
4417 err = ixgbe_setup_tx_resources(adapter, adapter->tx_ring[i]);
4418 if (!err)
4419 continue;
4420 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
4421 break;
4422 }
4423
4424 return err;
4425 }
4426
4427 /**
4428 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4429 * @adapter: board private structure
4430 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4431 *
4432 * Returns 0 on success, negative on failure
4433 **/
4434 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4435 struct ixgbe_ring *rx_ring)
4436 {
4437 struct pci_dev *pdev = adapter->pdev;
4438 int size;
4439
4440 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4441 rx_ring->rx_buffer_info = vmalloc_node(size, adapter->node);
4442 if (!rx_ring->rx_buffer_info)
4443 rx_ring->rx_buffer_info = vmalloc(size);
4444 if (!rx_ring->rx_buffer_info) {
4445 DPRINTK(PROBE, ERR,
4446 "vmalloc allocation failed for the rx desc ring\n");
4447 goto alloc_failed;
4448 }
4449 memset(rx_ring->rx_buffer_info, 0, size);
4450
4451 /* Round up to nearest 4K */
4452 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4453 rx_ring->size = ALIGN(rx_ring->size, 4096);
4454
4455 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
4456
4457 if (!rx_ring->desc) {
4458 DPRINTK(PROBE, ERR,
4459 "Memory allocation failed for the rx desc ring\n");
4460 vfree(rx_ring->rx_buffer_info);
4461 goto alloc_failed;
4462 }
4463
4464 rx_ring->next_to_clean = 0;
4465 rx_ring->next_to_use = 0;
4466
4467 return 0;
4468
4469 alloc_failed:
4470 return -ENOMEM;
4471 }
4472
4473 /**
4474 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4475 * @adapter: board private structure
4476 *
4477 * If this function returns with an error, then it's possible one or
4478 * more of the rings is populated (while the rest are not). It is the
4479 * callers duty to clean those orphaned rings.
4480 *
4481 * Return 0 on success, negative on failure
4482 **/
4483
4484 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4485 {
4486 int i, err = 0;
4487
4488 for (i = 0; i < adapter->num_rx_queues; i++) {
4489 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
4490 if (!err)
4491 continue;
4492 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
4493 break;
4494 }
4495
4496 return err;
4497 }
4498
4499 /**
4500 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4501 * @adapter: board private structure
4502 * @tx_ring: Tx descriptor ring for a specific queue
4503 *
4504 * Free all transmit software resources
4505 **/
4506 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4507 struct ixgbe_ring *tx_ring)
4508 {
4509 struct pci_dev *pdev = adapter->pdev;
4510
4511 ixgbe_clean_tx_ring(adapter, tx_ring);
4512
4513 vfree(tx_ring->tx_buffer_info);
4514 tx_ring->tx_buffer_info = NULL;
4515
4516 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
4517
4518 tx_ring->desc = NULL;
4519 }
4520
4521 /**
4522 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4523 * @adapter: board private structure
4524 *
4525 * Free all transmit software resources
4526 **/
4527 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4528 {
4529 int i;
4530
4531 for (i = 0; i < adapter->num_tx_queues; i++)
4532 if (adapter->tx_ring[i]->desc)
4533 ixgbe_free_tx_resources(adapter, adapter->tx_ring[i]);
4534 }
4535
4536 /**
4537 * ixgbe_free_rx_resources - Free Rx Resources
4538 * @adapter: board private structure
4539 * @rx_ring: ring to clean the resources from
4540 *
4541 * Free all receive software resources
4542 **/
4543 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
4544 struct ixgbe_ring *rx_ring)
4545 {
4546 struct pci_dev *pdev = adapter->pdev;
4547
4548 ixgbe_clean_rx_ring(adapter, rx_ring);
4549
4550 vfree(rx_ring->rx_buffer_info);
4551 rx_ring->rx_buffer_info = NULL;
4552
4553 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
4554
4555 rx_ring->desc = NULL;
4556 }
4557
4558 /**
4559 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4560 * @adapter: board private structure
4561 *
4562 * Free all receive software resources
4563 **/
4564 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4565 {
4566 int i;
4567
4568 for (i = 0; i < adapter->num_rx_queues; i++)
4569 if (adapter->rx_ring[i]->desc)
4570 ixgbe_free_rx_resources(adapter, adapter->rx_ring[i]);
4571 }
4572
4573 /**
4574 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4575 * @netdev: network interface device structure
4576 * @new_mtu: new value for maximum frame size
4577 *
4578 * Returns 0 on success, negative on failure
4579 **/
4580 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4581 {
4582 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4583 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4584
4585 /* MTU < 68 is an error and causes problems on some kernels */
4586 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4587 return -EINVAL;
4588
4589 DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
4590 netdev->mtu, new_mtu);
4591 /* must set new MTU before calling down or up */
4592 netdev->mtu = new_mtu;
4593
4594 if (netif_running(netdev))
4595 ixgbe_reinit_locked(adapter);
4596
4597 return 0;
4598 }
4599
4600 /**
4601 * ixgbe_open - Called when a network interface is made active
4602 * @netdev: network interface device structure
4603 *
4604 * Returns 0 on success, negative value on failure
4605 *
4606 * The open entry point is called when a network interface is made
4607 * active by the system (IFF_UP). At this point all resources needed
4608 * for transmit and receive operations are allocated, the interrupt
4609 * handler is registered with the OS, the watchdog timer is started,
4610 * and the stack is notified that the interface is ready.
4611 **/
4612 static int ixgbe_open(struct net_device *netdev)
4613 {
4614 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4615 int err;
4616
4617 /* disallow open during test */
4618 if (test_bit(__IXGBE_TESTING, &adapter->state))
4619 return -EBUSY;
4620
4621 netif_carrier_off(netdev);
4622
4623 /* allocate transmit descriptors */
4624 err = ixgbe_setup_all_tx_resources(adapter);
4625 if (err)
4626 goto err_setup_tx;
4627
4628 /* allocate receive descriptors */
4629 err = ixgbe_setup_all_rx_resources(adapter);
4630 if (err)
4631 goto err_setup_rx;
4632
4633 ixgbe_configure(adapter);
4634
4635 err = ixgbe_request_irq(adapter);
4636 if (err)
4637 goto err_req_irq;
4638
4639 err = ixgbe_up_complete(adapter);
4640 if (err)
4641 goto err_up;
4642
4643 netif_tx_start_all_queues(netdev);
4644
4645 return 0;
4646
4647 err_up:
4648 ixgbe_release_hw_control(adapter);
4649 ixgbe_free_irq(adapter);
4650 err_req_irq:
4651 err_setup_rx:
4652 ixgbe_free_all_rx_resources(adapter);
4653 err_setup_tx:
4654 ixgbe_free_all_tx_resources(adapter);
4655 ixgbe_reset(adapter);
4656
4657 return err;
4658 }
4659
4660 /**
4661 * ixgbe_close - Disables a network interface
4662 * @netdev: network interface device structure
4663 *
4664 * Returns 0, this is not allowed to fail
4665 *
4666 * The close entry point is called when an interface is de-activated
4667 * by the OS. The hardware is still under the drivers control, but
4668 * needs to be disabled. A global MAC reset is issued to stop the
4669 * hardware, and all transmit and receive resources are freed.
4670 **/
4671 static int ixgbe_close(struct net_device *netdev)
4672 {
4673 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4674
4675 ixgbe_down(adapter);
4676 ixgbe_free_irq(adapter);
4677
4678 ixgbe_free_all_tx_resources(adapter);
4679 ixgbe_free_all_rx_resources(adapter);
4680
4681 ixgbe_release_hw_control(adapter);
4682
4683 return 0;
4684 }
4685
4686 #ifdef CONFIG_PM
4687 static int ixgbe_resume(struct pci_dev *pdev)
4688 {
4689 struct net_device *netdev = pci_get_drvdata(pdev);
4690 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4691 u32 err;
4692
4693 pci_set_power_state(pdev, PCI_D0);
4694 pci_restore_state(pdev);
4695 /*
4696 * pci_restore_state clears dev->state_saved so call
4697 * pci_save_state to restore it.
4698 */
4699 pci_save_state(pdev);
4700
4701 err = pci_enable_device_mem(pdev);
4702 if (err) {
4703 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
4704 "suspend\n");
4705 return err;
4706 }
4707 pci_set_master(pdev);
4708
4709 pci_wake_from_d3(pdev, false);
4710
4711 err = ixgbe_init_interrupt_scheme(adapter);
4712 if (err) {
4713 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
4714 "device\n");
4715 return err;
4716 }
4717
4718 ixgbe_reset(adapter);
4719
4720 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4721
4722 if (netif_running(netdev)) {
4723 err = ixgbe_open(adapter->netdev);
4724 if (err)
4725 return err;
4726 }
4727
4728 netif_device_attach(netdev);
4729
4730 return 0;
4731 }
4732 #endif /* CONFIG_PM */
4733
4734 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4735 {
4736 struct net_device *netdev = pci_get_drvdata(pdev);
4737 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4738 struct ixgbe_hw *hw = &adapter->hw;
4739 u32 ctrl, fctrl;
4740 u32 wufc = adapter->wol;
4741 #ifdef CONFIG_PM
4742 int retval = 0;
4743 #endif
4744
4745 netif_device_detach(netdev);
4746
4747 if (netif_running(netdev)) {
4748 ixgbe_down(adapter);
4749 ixgbe_free_irq(adapter);
4750 ixgbe_free_all_tx_resources(adapter);
4751 ixgbe_free_all_rx_resources(adapter);
4752 }
4753 ixgbe_clear_interrupt_scheme(adapter);
4754
4755 #ifdef CONFIG_PM
4756 retval = pci_save_state(pdev);
4757 if (retval)
4758 return retval;
4759
4760 #endif
4761 if (wufc) {
4762 ixgbe_set_rx_mode(netdev);
4763
4764 /* turn on all-multi mode if wake on multicast is enabled */
4765 if (wufc & IXGBE_WUFC_MC) {
4766 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4767 fctrl |= IXGBE_FCTRL_MPE;
4768 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4769 }
4770
4771 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4772 ctrl |= IXGBE_CTRL_GIO_DIS;
4773 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4774
4775 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4776 } else {
4777 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4778 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4779 }
4780
4781 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
4782 pci_wake_from_d3(pdev, true);
4783 else
4784 pci_wake_from_d3(pdev, false);
4785
4786 *enable_wake = !!wufc;
4787
4788 ixgbe_release_hw_control(adapter);
4789
4790 pci_disable_device(pdev);
4791
4792 return 0;
4793 }
4794
4795 #ifdef CONFIG_PM
4796 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4797 {
4798 int retval;
4799 bool wake;
4800
4801 retval = __ixgbe_shutdown(pdev, &wake);
4802 if (retval)
4803 return retval;
4804
4805 if (wake) {
4806 pci_prepare_to_sleep(pdev);
4807 } else {
4808 pci_wake_from_d3(pdev, false);
4809 pci_set_power_state(pdev, PCI_D3hot);
4810 }
4811
4812 return 0;
4813 }
4814 #endif /* CONFIG_PM */
4815
4816 static void ixgbe_shutdown(struct pci_dev *pdev)
4817 {
4818 bool wake;
4819
4820 __ixgbe_shutdown(pdev, &wake);
4821
4822 if (system_state == SYSTEM_POWER_OFF) {
4823 pci_wake_from_d3(pdev, wake);
4824 pci_set_power_state(pdev, PCI_D3hot);
4825 }
4826 }
4827
4828 /**
4829 * ixgbe_update_stats - Update the board statistics counters.
4830 * @adapter: board private structure
4831 **/
4832 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4833 {
4834 struct net_device *netdev = adapter->netdev;
4835 struct ixgbe_hw *hw = &adapter->hw;
4836 u64 total_mpc = 0;
4837 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4838 u64 non_eop_descs = 0, restart_queue = 0;
4839
4840 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
4841 u64 rsc_count = 0;
4842 u64 rsc_flush = 0;
4843 for (i = 0; i < 16; i++)
4844 adapter->hw_rx_no_dma_resources +=
4845 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4846 for (i = 0; i < adapter->num_rx_queues; i++) {
4847 rsc_count += adapter->rx_ring[i]->rsc_count;
4848 rsc_flush += adapter->rx_ring[i]->rsc_flush;
4849 }
4850 adapter->rsc_total_count = rsc_count;
4851 adapter->rsc_total_flush = rsc_flush;
4852 }
4853
4854 /* gather some stats to the adapter struct that are per queue */
4855 for (i = 0; i < adapter->num_tx_queues; i++)
4856 restart_queue += adapter->tx_ring[i]->restart_queue;
4857 adapter->restart_queue = restart_queue;
4858
4859 for (i = 0; i < adapter->num_rx_queues; i++)
4860 non_eop_descs += adapter->rx_ring[i]->non_eop_descs;
4861 adapter->non_eop_descs = non_eop_descs;
4862
4863 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
4864 for (i = 0; i < 8; i++) {
4865 /* for packet buffers not used, the register should read 0 */
4866 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
4867 missed_rx += mpc;
4868 adapter->stats.mpc[i] += mpc;
4869 total_mpc += adapter->stats.mpc[i];
4870 if (hw->mac.type == ixgbe_mac_82598EB)
4871 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
4872 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
4873 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
4874 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
4875 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
4876 if (hw->mac.type == ixgbe_mac_82599EB) {
4877 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4878 IXGBE_PXONRXCNT(i));
4879 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4880 IXGBE_PXOFFRXCNT(i));
4881 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4882 } else {
4883 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4884 IXGBE_PXONRXC(i));
4885 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4886 IXGBE_PXOFFRXC(i));
4887 }
4888 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
4889 IXGBE_PXONTXC(i));
4890 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
4891 IXGBE_PXOFFTXC(i));
4892 }
4893 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
4894 /* work around hardware counting issue */
4895 adapter->stats.gprc -= missed_rx;
4896
4897 /* 82598 hardware only has a 32 bit counter in the high register */
4898 if (hw->mac.type == ixgbe_mac_82599EB) {
4899 u64 tmp;
4900 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
4901 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
4902 adapter->stats.gorc += (tmp << 32);
4903 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
4904 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
4905 adapter->stats.gotc += (tmp << 32);
4906 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
4907 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
4908 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
4909 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
4910 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
4911 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
4912 #ifdef IXGBE_FCOE
4913 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
4914 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
4915 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
4916 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
4917 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
4918 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
4919 #endif /* IXGBE_FCOE */
4920 } else {
4921 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
4922 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
4923 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
4924 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
4925 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
4926 }
4927 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
4928 adapter->stats.bprc += bprc;
4929 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
4930 if (hw->mac.type == ixgbe_mac_82598EB)
4931 adapter->stats.mprc -= bprc;
4932 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
4933 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
4934 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
4935 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
4936 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
4937 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
4938 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
4939 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
4940 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
4941 adapter->stats.lxontxc += lxon;
4942 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
4943 adapter->stats.lxofftxc += lxoff;
4944 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4945 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
4946 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
4947 /*
4948 * 82598 errata - tx of flow control packets is included in tx counters
4949 */
4950 xon_off_tot = lxon + lxoff;
4951 adapter->stats.gptc -= xon_off_tot;
4952 adapter->stats.mptc -= xon_off_tot;
4953 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
4954 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4955 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
4956 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
4957 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
4958 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
4959 adapter->stats.ptc64 -= xon_off_tot;
4960 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
4961 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
4962 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
4963 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
4964 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
4965 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
4966
4967 /* Fill out the OS statistics structure */
4968 netdev->stats.multicast = adapter->stats.mprc;
4969
4970 /* Rx Errors */
4971 netdev->stats.rx_errors = adapter->stats.crcerrs +
4972 adapter->stats.rlec;
4973 netdev->stats.rx_dropped = 0;
4974 netdev->stats.rx_length_errors = adapter->stats.rlec;
4975 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4976 netdev->stats.rx_missed_errors = total_mpc;
4977 }
4978
4979 /**
4980 * ixgbe_watchdog - Timer Call-back
4981 * @data: pointer to adapter cast into an unsigned long
4982 **/
4983 static void ixgbe_watchdog(unsigned long data)
4984 {
4985 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4986 struct ixgbe_hw *hw = &adapter->hw;
4987 u64 eics = 0;
4988 int i;
4989
4990 /*
4991 * Do the watchdog outside of interrupt context due to the lovely
4992 * delays that some of the newer hardware requires
4993 */
4994
4995 if (test_bit(__IXGBE_DOWN, &adapter->state))
4996 goto watchdog_short_circuit;
4997
4998 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
4999 /*
5000 * for legacy and MSI interrupts don't set any bits
5001 * that are enabled for EIAM, because this operation
5002 * would set *both* EIMS and EICS for any bit in EIAM
5003 */
5004 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5005 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5006 goto watchdog_reschedule;
5007 }
5008
5009 /* get one bit for every active tx/rx interrupt vector */
5010 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5011 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5012 if (qv->rxr_count || qv->txr_count)
5013 eics |= ((u64)1 << i);
5014 }
5015
5016 /* Cause software interrupt to ensure rx rings are cleaned */
5017 ixgbe_irq_rearm_queues(adapter, eics);
5018
5019 watchdog_reschedule:
5020 /* Reset the timer */
5021 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5022
5023 watchdog_short_circuit:
5024 schedule_work(&adapter->watchdog_task);
5025 }
5026
5027 /**
5028 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5029 * @work: pointer to work_struct containing our data
5030 **/
5031 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5032 {
5033 struct ixgbe_adapter *adapter = container_of(work,
5034 struct ixgbe_adapter,
5035 multispeed_fiber_task);
5036 struct ixgbe_hw *hw = &adapter->hw;
5037 u32 autoneg;
5038 bool negotiation;
5039
5040 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5041 autoneg = hw->phy.autoneg_advertised;
5042 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5043 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5044 hw->mac.autotry_restart = false;
5045 if (hw->mac.ops.setup_link)
5046 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5047 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5048 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5049 }
5050
5051 /**
5052 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5053 * @work: pointer to work_struct containing our data
5054 **/
5055 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5056 {
5057 struct ixgbe_adapter *adapter = container_of(work,
5058 struct ixgbe_adapter,
5059 sfp_config_module_task);
5060 struct ixgbe_hw *hw = &adapter->hw;
5061 u32 err;
5062
5063 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5064
5065 /* Time for electrical oscillations to settle down */
5066 msleep(100);
5067 err = hw->phy.ops.identify_sfp(hw);
5068
5069 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5070 dev_err(&adapter->pdev->dev, "failed to initialize because "
5071 "an unsupported SFP+ module type was detected.\n"
5072 "Reload the driver after installing a supported "
5073 "module.\n");
5074 unregister_netdev(adapter->netdev);
5075 return;
5076 }
5077 hw->mac.ops.setup_sfp(hw);
5078
5079 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5080 /* This will also work for DA Twinax connections */
5081 schedule_work(&adapter->multispeed_fiber_task);
5082 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5083 }
5084
5085 /**
5086 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5087 * @work: pointer to work_struct containing our data
5088 **/
5089 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5090 {
5091 struct ixgbe_adapter *adapter = container_of(work,
5092 struct ixgbe_adapter,
5093 fdir_reinit_task);
5094 struct ixgbe_hw *hw = &adapter->hw;
5095 int i;
5096
5097 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5098 for (i = 0; i < adapter->num_tx_queues; i++)
5099 set_bit(__IXGBE_FDIR_INIT_DONE,
5100 &(adapter->tx_ring[i]->reinit_state));
5101 } else {
5102 DPRINTK(PROBE, ERR, "failed to finish FDIR re-initialization, "
5103 "ignored adding FDIR ATR filters \n");
5104 }
5105 /* Done FDIR Re-initialization, enable transmits */
5106 netif_tx_start_all_queues(adapter->netdev);
5107 }
5108
5109 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5110
5111 /**
5112 * ixgbe_watchdog_task - worker thread to bring link up
5113 * @work: pointer to work_struct containing our data
5114 **/
5115 static void ixgbe_watchdog_task(struct work_struct *work)
5116 {
5117 struct ixgbe_adapter *adapter = container_of(work,
5118 struct ixgbe_adapter,
5119 watchdog_task);
5120 struct net_device *netdev = adapter->netdev;
5121 struct ixgbe_hw *hw = &adapter->hw;
5122 u32 link_speed;
5123 bool link_up;
5124 int i;
5125 struct ixgbe_ring *tx_ring;
5126 int some_tx_pending = 0;
5127
5128 mutex_lock(&ixgbe_watchdog_lock);
5129
5130 link_up = adapter->link_up;
5131 link_speed = adapter->link_speed;
5132
5133 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5134 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5135 if (link_up) {
5136 #ifdef CONFIG_DCB
5137 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5138 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5139 hw->mac.ops.fc_enable(hw, i);
5140 } else {
5141 hw->mac.ops.fc_enable(hw, 0);
5142 }
5143 #else
5144 hw->mac.ops.fc_enable(hw, 0);
5145 #endif
5146 }
5147
5148 if (link_up ||
5149 time_after(jiffies, (adapter->link_check_timeout +
5150 IXGBE_TRY_LINK_TIMEOUT))) {
5151 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5152 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5153 }
5154 adapter->link_up = link_up;
5155 adapter->link_speed = link_speed;
5156 }
5157
5158 if (link_up) {
5159 if (!netif_carrier_ok(netdev)) {
5160 bool flow_rx, flow_tx;
5161
5162 if (hw->mac.type == ixgbe_mac_82599EB) {
5163 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5164 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5165 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5166 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5167 } else {
5168 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5169 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5170 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5171 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5172 }
5173
5174 printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
5175 "Flow Control: %s\n",
5176 netdev->name,
5177 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5178 "10 Gbps" :
5179 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5180 "1 Gbps" : "unknown speed")),
5181 ((flow_rx && flow_tx) ? "RX/TX" :
5182 (flow_rx ? "RX" :
5183 (flow_tx ? "TX" : "None"))));
5184
5185 netif_carrier_on(netdev);
5186 } else {
5187 /* Force detection of hung controller */
5188 adapter->detect_tx_hung = true;
5189 }
5190 } else {
5191 adapter->link_up = false;
5192 adapter->link_speed = 0;
5193 if (netif_carrier_ok(netdev)) {
5194 printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
5195 netdev->name);
5196 netif_carrier_off(netdev);
5197 }
5198 }
5199
5200 if (!netif_carrier_ok(netdev)) {
5201 for (i = 0; i < adapter->num_tx_queues; i++) {
5202 tx_ring = adapter->tx_ring[i];
5203 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5204 some_tx_pending = 1;
5205 break;
5206 }
5207 }
5208
5209 if (some_tx_pending) {
5210 /* We've lost link, so the controller stops DMA,
5211 * but we've got queued Tx work that's never going
5212 * to get done, so reset controller to flush Tx.
5213 * (Do the reset outside of interrupt context).
5214 */
5215 schedule_work(&adapter->reset_task);
5216 }
5217 }
5218
5219 ixgbe_update_stats(adapter);
5220 mutex_unlock(&ixgbe_watchdog_lock);
5221 }
5222
5223 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5224 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5225 u32 tx_flags, u8 *hdr_len)
5226 {
5227 struct ixgbe_adv_tx_context_desc *context_desc;
5228 unsigned int i;
5229 int err;
5230 struct ixgbe_tx_buffer *tx_buffer_info;
5231 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5232 u32 mss_l4len_idx, l4len;
5233
5234 if (skb_is_gso(skb)) {
5235 if (skb_header_cloned(skb)) {
5236 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5237 if (err)
5238 return err;
5239 }
5240 l4len = tcp_hdrlen(skb);
5241 *hdr_len += l4len;
5242
5243 if (skb->protocol == htons(ETH_P_IP)) {
5244 struct iphdr *iph = ip_hdr(skb);
5245 iph->tot_len = 0;
5246 iph->check = 0;
5247 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5248 iph->daddr, 0,
5249 IPPROTO_TCP,
5250 0);
5251 } else if (skb_is_gso_v6(skb)) {
5252 ipv6_hdr(skb)->payload_len = 0;
5253 tcp_hdr(skb)->check =
5254 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5255 &ipv6_hdr(skb)->daddr,
5256 0, IPPROTO_TCP, 0);
5257 }
5258
5259 i = tx_ring->next_to_use;
5260
5261 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5262 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5263
5264 /* VLAN MACLEN IPLEN */
5265 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5266 vlan_macip_lens |=
5267 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5268 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5269 IXGBE_ADVTXD_MACLEN_SHIFT);
5270 *hdr_len += skb_network_offset(skb);
5271 vlan_macip_lens |=
5272 (skb_transport_header(skb) - skb_network_header(skb));
5273 *hdr_len +=
5274 (skb_transport_header(skb) - skb_network_header(skb));
5275 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5276 context_desc->seqnum_seed = 0;
5277
5278 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5279 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5280 IXGBE_ADVTXD_DTYP_CTXT);
5281
5282 if (skb->protocol == htons(ETH_P_IP))
5283 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5284 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5285 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5286
5287 /* MSS L4LEN IDX */
5288 mss_l4len_idx =
5289 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5290 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5291 /* use index 1 for TSO */
5292 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5293 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5294
5295 tx_buffer_info->time_stamp = jiffies;
5296 tx_buffer_info->next_to_watch = i;
5297
5298 i++;
5299 if (i == tx_ring->count)
5300 i = 0;
5301 tx_ring->next_to_use = i;
5302
5303 return true;
5304 }
5305 return false;
5306 }
5307
5308 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5309 struct ixgbe_ring *tx_ring,
5310 struct sk_buff *skb, u32 tx_flags)
5311 {
5312 struct ixgbe_adv_tx_context_desc *context_desc;
5313 unsigned int i;
5314 struct ixgbe_tx_buffer *tx_buffer_info;
5315 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5316
5317 if (skb->ip_summed == CHECKSUM_PARTIAL ||
5318 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5319 i = tx_ring->next_to_use;
5320 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5321 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5322
5323 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5324 vlan_macip_lens |=
5325 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5326 vlan_macip_lens |= (skb_network_offset(skb) <<
5327 IXGBE_ADVTXD_MACLEN_SHIFT);
5328 if (skb->ip_summed == CHECKSUM_PARTIAL)
5329 vlan_macip_lens |= (skb_transport_header(skb) -
5330 skb_network_header(skb));
5331
5332 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5333 context_desc->seqnum_seed = 0;
5334
5335 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5336 IXGBE_ADVTXD_DTYP_CTXT);
5337
5338 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5339 __be16 protocol;
5340
5341 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
5342 const struct vlan_ethhdr *vhdr =
5343 (const struct vlan_ethhdr *)skb->data;
5344
5345 protocol = vhdr->h_vlan_encapsulated_proto;
5346 } else {
5347 protocol = skb->protocol;
5348 }
5349
5350 switch (protocol) {
5351 case cpu_to_be16(ETH_P_IP):
5352 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5353 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5354 type_tucmd_mlhl |=
5355 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5356 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
5357 type_tucmd_mlhl |=
5358 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5359 break;
5360 case cpu_to_be16(ETH_P_IPV6):
5361 /* XXX what about other V6 headers?? */
5362 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5363 type_tucmd_mlhl |=
5364 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5365 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
5366 type_tucmd_mlhl |=
5367 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5368 break;
5369 default:
5370 if (unlikely(net_ratelimit())) {
5371 DPRINTK(PROBE, WARNING,
5372 "partial checksum but proto=%x!\n",
5373 skb->protocol);
5374 }
5375 break;
5376 }
5377 }
5378
5379 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5380 /* use index zero for tx checksum offload */
5381 context_desc->mss_l4len_idx = 0;
5382
5383 tx_buffer_info->time_stamp = jiffies;
5384 tx_buffer_info->next_to_watch = i;
5385
5386 i++;
5387 if (i == tx_ring->count)
5388 i = 0;
5389 tx_ring->next_to_use = i;
5390
5391 return true;
5392 }
5393
5394 return false;
5395 }
5396
5397 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5398 struct ixgbe_ring *tx_ring,
5399 struct sk_buff *skb, u32 tx_flags,
5400 unsigned int first)
5401 {
5402 struct pci_dev *pdev = adapter->pdev;
5403 struct ixgbe_tx_buffer *tx_buffer_info;
5404 unsigned int len;
5405 unsigned int total = skb->len;
5406 unsigned int offset = 0, size, count = 0, i;
5407 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5408 unsigned int f;
5409
5410 i = tx_ring->next_to_use;
5411
5412 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5413 /* excluding fcoe_crc_eof for FCoE */
5414 total -= sizeof(struct fcoe_crc_eof);
5415
5416 len = min(skb_headlen(skb), total);
5417 while (len) {
5418 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5419 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5420
5421 tx_buffer_info->length = size;
5422 tx_buffer_info->mapped_as_page = false;
5423 tx_buffer_info->dma = pci_map_single(pdev,
5424 skb->data + offset,
5425 size, PCI_DMA_TODEVICE);
5426 if (pci_dma_mapping_error(pdev, tx_buffer_info->dma))
5427 goto dma_error;
5428 tx_buffer_info->time_stamp = jiffies;
5429 tx_buffer_info->next_to_watch = i;
5430
5431 len -= size;
5432 total -= size;
5433 offset += size;
5434 count++;
5435
5436 if (len) {
5437 i++;
5438 if (i == tx_ring->count)
5439 i = 0;
5440 }
5441 }
5442
5443 for (f = 0; f < nr_frags; f++) {
5444 struct skb_frag_struct *frag;
5445
5446 frag = &skb_shinfo(skb)->frags[f];
5447 len = min((unsigned int)frag->size, total);
5448 offset = frag->page_offset;
5449
5450 while (len) {
5451 i++;
5452 if (i == tx_ring->count)
5453 i = 0;
5454
5455 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5456 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5457
5458 tx_buffer_info->length = size;
5459 tx_buffer_info->dma = pci_map_page(adapter->pdev,
5460 frag->page,
5461 offset, size,
5462 PCI_DMA_TODEVICE);
5463 tx_buffer_info->mapped_as_page = true;
5464 if (pci_dma_mapping_error(pdev, tx_buffer_info->dma))
5465 goto dma_error;
5466 tx_buffer_info->time_stamp = jiffies;
5467 tx_buffer_info->next_to_watch = i;
5468
5469 len -= size;
5470 total -= size;
5471 offset += size;
5472 count++;
5473 }
5474 if (total == 0)
5475 break;
5476 }
5477
5478 tx_ring->tx_buffer_info[i].skb = skb;
5479 tx_ring->tx_buffer_info[first].next_to_watch = i;
5480
5481 return count;
5482
5483 dma_error:
5484 dev_err(&pdev->dev, "TX DMA map failed\n");
5485
5486 /* clear timestamp and dma mappings for failed tx_buffer_info map */
5487 tx_buffer_info->dma = 0;
5488 tx_buffer_info->time_stamp = 0;
5489 tx_buffer_info->next_to_watch = 0;
5490 if (count)
5491 count--;
5492
5493 /* clear timestamp and dma mappings for remaining portion of packet */
5494 while (count--) {
5495 if (i==0)
5496 i += tx_ring->count;
5497 i--;
5498 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5499 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
5500 }
5501
5502 return 0;
5503 }
5504
5505 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
5506 struct ixgbe_ring *tx_ring,
5507 int tx_flags, int count, u32 paylen, u8 hdr_len)
5508 {
5509 union ixgbe_adv_tx_desc *tx_desc = NULL;
5510 struct ixgbe_tx_buffer *tx_buffer_info;
5511 u32 olinfo_status = 0, cmd_type_len = 0;
5512 unsigned int i;
5513 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
5514
5515 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
5516
5517 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
5518
5519 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5520 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
5521
5522 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
5523 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5524
5525 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5526 IXGBE_ADVTXD_POPTS_SHIFT;
5527
5528 /* use index 1 context for tso */
5529 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5530 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5531 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
5532 IXGBE_ADVTXD_POPTS_SHIFT;
5533
5534 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5535 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5536 IXGBE_ADVTXD_POPTS_SHIFT;
5537
5538 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5539 olinfo_status |= IXGBE_ADVTXD_CC;
5540 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5541 if (tx_flags & IXGBE_TX_FLAGS_FSO)
5542 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5543 }
5544
5545 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
5546
5547 i = tx_ring->next_to_use;
5548 while (count--) {
5549 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5550 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
5551 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
5552 tx_desc->read.cmd_type_len =
5553 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
5554 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5555 i++;
5556 if (i == tx_ring->count)
5557 i = 0;
5558 }
5559
5560 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
5561
5562 /*
5563 * Force memory writes to complete before letting h/w
5564 * know there are new descriptors to fetch. (Only
5565 * applicable for weak-ordered memory model archs,
5566 * such as IA-64).
5567 */
5568 wmb();
5569
5570 tx_ring->next_to_use = i;
5571 writel(i, adapter->hw.hw_addr + tx_ring->tail);
5572 }
5573
5574 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5575 int queue, u32 tx_flags)
5576 {
5577 /* Right now, we support IPv4 only */
5578 struct ixgbe_atr_input atr_input;
5579 struct tcphdr *th;
5580 struct iphdr *iph = ip_hdr(skb);
5581 struct ethhdr *eth = (struct ethhdr *)skb->data;
5582 u16 vlan_id, src_port, dst_port, flex_bytes;
5583 u32 src_ipv4_addr, dst_ipv4_addr;
5584 u8 l4type = 0;
5585
5586 /* check if we're UDP or TCP */
5587 if (iph->protocol == IPPROTO_TCP) {
5588 th = tcp_hdr(skb);
5589 src_port = th->source;
5590 dst_port = th->dest;
5591 l4type |= IXGBE_ATR_L4TYPE_TCP;
5592 /* l4type IPv4 type is 0, no need to assign */
5593 } else {
5594 /* Unsupported L4 header, just bail here */
5595 return;
5596 }
5597
5598 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
5599
5600 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
5601 IXGBE_TX_FLAGS_VLAN_SHIFT;
5602 src_ipv4_addr = iph->saddr;
5603 dst_ipv4_addr = iph->daddr;
5604 flex_bytes = eth->h_proto;
5605
5606 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
5607 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
5608 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
5609 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
5610 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
5611 /* src and dst are inverted, think how the receiver sees them */
5612 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
5613 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
5614
5615 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
5616 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
5617 }
5618
5619 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
5620 struct ixgbe_ring *tx_ring, int size)
5621 {
5622 netif_stop_subqueue(netdev, tx_ring->queue_index);
5623 /* Herbert's original patch had:
5624 * smp_mb__after_netif_stop_queue();
5625 * but since that doesn't exist yet, just open code it. */
5626 smp_mb();
5627
5628 /* We need to check again in a case another CPU has just
5629 * made room available. */
5630 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
5631 return -EBUSY;
5632
5633 /* A reprieve! - use start_queue because it doesn't call schedule */
5634 netif_start_subqueue(netdev, tx_ring->queue_index);
5635 ++tx_ring->restart_queue;
5636 return 0;
5637 }
5638
5639 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
5640 struct ixgbe_ring *tx_ring, int size)
5641 {
5642 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
5643 return 0;
5644 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
5645 }
5646
5647 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
5648 {
5649 struct ixgbe_adapter *adapter = netdev_priv(dev);
5650 int txq = smp_processor_id();
5651
5652 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5653 while (unlikely(txq >= dev->real_num_tx_queues))
5654 txq -= dev->real_num_tx_queues;
5655 return txq;
5656 }
5657
5658 #ifdef IXGBE_FCOE
5659 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5660 ((skb->protocol == htons(ETH_P_FCOE)) ||
5661 (skb->protocol == htons(ETH_P_FIP)))) {
5662 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
5663 txq += adapter->ring_feature[RING_F_FCOE].mask;
5664 return txq;
5665 }
5666 #endif
5667 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5668 if (skb->priority == TC_PRIO_CONTROL)
5669 txq = adapter->ring_feature[RING_F_DCB].indices-1;
5670 else
5671 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
5672 >> 13;
5673 return txq;
5674 }
5675
5676 return skb_tx_hash(dev, skb);
5677 }
5678
5679 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
5680 struct net_device *netdev)
5681 {
5682 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5683 struct ixgbe_ring *tx_ring;
5684 struct netdev_queue *txq;
5685 unsigned int first;
5686 unsigned int tx_flags = 0;
5687 u8 hdr_len = 0;
5688 int tso;
5689 int count = 0;
5690 unsigned int f;
5691
5692 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
5693 tx_flags |= vlan_tx_tag_get(skb);
5694 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5695 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
5696 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
5697 }
5698 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5699 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5700 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5701 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
5702 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5703 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5704 }
5705
5706 tx_ring = adapter->tx_ring[skb->queue_mapping];
5707
5708 #ifdef IXGBE_FCOE
5709 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
5710 #ifdef CONFIG_IXGBE_DCB
5711 /* for FCoE with DCB, we force the priority to what
5712 * was specified by the switch */
5713 if ((skb->protocol == htons(ETH_P_FCOE)) ||
5714 (skb->protocol == htons(ETH_P_FIP))) {
5715 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
5716 << IXGBE_TX_FLAGS_VLAN_SHIFT);
5717 tx_flags |= ((adapter->fcoe.up << 13)
5718 << IXGBE_TX_FLAGS_VLAN_SHIFT);
5719 }
5720 #endif
5721 /* flag for FCoE offloads */
5722 if (skb->protocol == htons(ETH_P_FCOE))
5723 tx_flags |= IXGBE_TX_FLAGS_FCOE;
5724 }
5725 #endif
5726
5727 /* four things can cause us to need a context descriptor */
5728 if (skb_is_gso(skb) ||
5729 (skb->ip_summed == CHECKSUM_PARTIAL) ||
5730 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
5731 (tx_flags & IXGBE_TX_FLAGS_FCOE))
5732 count++;
5733
5734 count += TXD_USE_COUNT(skb_headlen(skb));
5735 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5736 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5737
5738 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
5739 adapter->tx_busy++;
5740 return NETDEV_TX_BUSY;
5741 }
5742
5743 first = tx_ring->next_to_use;
5744 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5745 #ifdef IXGBE_FCOE
5746 /* setup tx offload for FCoE */
5747 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5748 if (tso < 0) {
5749 dev_kfree_skb_any(skb);
5750 return NETDEV_TX_OK;
5751 }
5752 if (tso)
5753 tx_flags |= IXGBE_TX_FLAGS_FSO;
5754 #endif /* IXGBE_FCOE */
5755 } else {
5756 if (skb->protocol == htons(ETH_P_IP))
5757 tx_flags |= IXGBE_TX_FLAGS_IPV4;
5758 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5759 if (tso < 0) {
5760 dev_kfree_skb_any(skb);
5761 return NETDEV_TX_OK;
5762 }
5763
5764 if (tso)
5765 tx_flags |= IXGBE_TX_FLAGS_TSO;
5766 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
5767 (skb->ip_summed == CHECKSUM_PARTIAL))
5768 tx_flags |= IXGBE_TX_FLAGS_CSUM;
5769 }
5770
5771 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
5772 if (count) {
5773 /* add the ATR filter if ATR is on */
5774 if (tx_ring->atr_sample_rate) {
5775 ++tx_ring->atr_count;
5776 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
5777 test_bit(__IXGBE_FDIR_INIT_DONE,
5778 &tx_ring->reinit_state)) {
5779 ixgbe_atr(adapter, skb, tx_ring->queue_index,
5780 tx_flags);
5781 tx_ring->atr_count = 0;
5782 }
5783 }
5784 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
5785 txq->tx_bytes += skb->len;
5786 txq->tx_packets++;
5787 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
5788 hdr_len);
5789 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
5790
5791 } else {
5792 dev_kfree_skb_any(skb);
5793 tx_ring->tx_buffer_info[first].time_stamp = 0;
5794 tx_ring->next_to_use = first;
5795 }
5796
5797 return NETDEV_TX_OK;
5798 }
5799
5800 /**
5801 * ixgbe_set_mac - Change the Ethernet Address of the NIC
5802 * @netdev: network interface device structure
5803 * @p: pointer to an address structure
5804 *
5805 * Returns 0 on success, negative on failure
5806 **/
5807 static int ixgbe_set_mac(struct net_device *netdev, void *p)
5808 {
5809 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5810 struct ixgbe_hw *hw = &adapter->hw;
5811 struct sockaddr *addr = p;
5812
5813 if (!is_valid_ether_addr(addr->sa_data))
5814 return -EADDRNOTAVAIL;
5815
5816 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
5817 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5818
5819 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
5820 IXGBE_RAH_AV);
5821
5822 return 0;
5823 }
5824
5825 static int
5826 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
5827 {
5828 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5829 struct ixgbe_hw *hw = &adapter->hw;
5830 u16 value;
5831 int rc;
5832
5833 if (prtad != hw->phy.mdio.prtad)
5834 return -EINVAL;
5835 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
5836 if (!rc)
5837 rc = value;
5838 return rc;
5839 }
5840
5841 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
5842 u16 addr, u16 value)
5843 {
5844 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5845 struct ixgbe_hw *hw = &adapter->hw;
5846
5847 if (prtad != hw->phy.mdio.prtad)
5848 return -EINVAL;
5849 return hw->phy.ops.write_reg(hw, addr, devad, value);
5850 }
5851
5852 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
5853 {
5854 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5855
5856 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
5857 }
5858
5859 /**
5860 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
5861 * netdev->dev_addrs
5862 * @netdev: network interface device structure
5863 *
5864 * Returns non-zero on failure
5865 **/
5866 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
5867 {
5868 int err = 0;
5869 struct ixgbe_adapter *adapter = netdev_priv(dev);
5870 struct ixgbe_mac_info *mac = &adapter->hw.mac;
5871
5872 if (is_valid_ether_addr(mac->san_addr)) {
5873 rtnl_lock();
5874 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5875 rtnl_unlock();
5876 }
5877 return err;
5878 }
5879
5880 /**
5881 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
5882 * netdev->dev_addrs
5883 * @netdev: network interface device structure
5884 *
5885 * Returns non-zero on failure
5886 **/
5887 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
5888 {
5889 int err = 0;
5890 struct ixgbe_adapter *adapter = netdev_priv(dev);
5891 struct ixgbe_mac_info *mac = &adapter->hw.mac;
5892
5893 if (is_valid_ether_addr(mac->san_addr)) {
5894 rtnl_lock();
5895 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5896 rtnl_unlock();
5897 }
5898 return err;
5899 }
5900
5901 #ifdef CONFIG_NET_POLL_CONTROLLER
5902 /*
5903 * Polling 'interrupt' - used by things like netconsole to send skbs
5904 * without having to re-enable interrupts. It's not called while
5905 * the interrupt routine is executing.
5906 */
5907 static void ixgbe_netpoll(struct net_device *netdev)
5908 {
5909 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5910 int i;
5911
5912 /* if interface is down do nothing */
5913 if (test_bit(__IXGBE_DOWN, &adapter->state))
5914 return;
5915
5916 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
5917 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5918 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5919 for (i = 0; i < num_q_vectors; i++) {
5920 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
5921 ixgbe_msix_clean_many(0, q_vector);
5922 }
5923 } else {
5924 ixgbe_intr(adapter->pdev->irq, netdev);
5925 }
5926 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
5927 }
5928 #endif
5929
5930 static const struct net_device_ops ixgbe_netdev_ops = {
5931 .ndo_open = ixgbe_open,
5932 .ndo_stop = ixgbe_close,
5933 .ndo_start_xmit = ixgbe_xmit_frame,
5934 .ndo_select_queue = ixgbe_select_queue,
5935 .ndo_set_rx_mode = ixgbe_set_rx_mode,
5936 .ndo_set_multicast_list = ixgbe_set_rx_mode,
5937 .ndo_validate_addr = eth_validate_addr,
5938 .ndo_set_mac_address = ixgbe_set_mac,
5939 .ndo_change_mtu = ixgbe_change_mtu,
5940 .ndo_tx_timeout = ixgbe_tx_timeout,
5941 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
5942 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
5943 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
5944 .ndo_do_ioctl = ixgbe_ioctl,
5945 #ifdef CONFIG_NET_POLL_CONTROLLER
5946 .ndo_poll_controller = ixgbe_netpoll,
5947 #endif
5948 #ifdef IXGBE_FCOE
5949 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
5950 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
5951 .ndo_fcoe_enable = ixgbe_fcoe_enable,
5952 .ndo_fcoe_disable = ixgbe_fcoe_disable,
5953 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
5954 #endif /* IXGBE_FCOE */
5955 };
5956
5957 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
5958 const struct ixgbe_info *ii)
5959 {
5960 #ifdef CONFIG_PCI_IOV
5961 struct ixgbe_hw *hw = &adapter->hw;
5962 int err;
5963
5964 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
5965 return;
5966
5967 /* The 82599 supports up to 64 VFs per physical function
5968 * but this implementation limits allocation to 63 so that
5969 * basic networking resources are still available to the
5970 * physical function
5971 */
5972 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
5973 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
5974 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
5975 if (err) {
5976 DPRINTK(PROBE, ERR,
5977 "Failed to enable PCI sriov: %d\n", err);
5978 goto err_novfs;
5979 }
5980 /* If call to enable VFs succeeded then allocate memory
5981 * for per VF control structures.
5982 */
5983 adapter->vfinfo =
5984 kcalloc(adapter->num_vfs,
5985 sizeof(struct vf_data_storage), GFP_KERNEL);
5986 if (adapter->vfinfo) {
5987 /* Now that we're sure SR-IOV is enabled
5988 * and memory allocated set up the mailbox parameters
5989 */
5990 ixgbe_init_mbx_params_pf(hw);
5991 memcpy(&hw->mbx.ops, ii->mbx_ops,
5992 sizeof(hw->mbx.ops));
5993
5994 /* Disable RSC when in SR-IOV mode */
5995 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
5996 IXGBE_FLAG2_RSC_ENABLED);
5997 return;
5998 }
5999
6000 /* Oh oh */
6001 DPRINTK(PROBE, ERR,
6002 "Unable to allocate memory for VF "
6003 "Data Storage - SRIOV disabled\n");
6004 pci_disable_sriov(adapter->pdev);
6005
6006 err_novfs:
6007 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6008 adapter->num_vfs = 0;
6009 #endif /* CONFIG_PCI_IOV */
6010 }
6011
6012 /**
6013 * ixgbe_probe - Device Initialization Routine
6014 * @pdev: PCI device information struct
6015 * @ent: entry in ixgbe_pci_tbl
6016 *
6017 * Returns 0 on success, negative on failure
6018 *
6019 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6020 * The OS initialization, configuring of the adapter private structure,
6021 * and a hardware reset occur.
6022 **/
6023 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6024 const struct pci_device_id *ent)
6025 {
6026 struct net_device *netdev;
6027 struct ixgbe_adapter *adapter = NULL;
6028 struct ixgbe_hw *hw;
6029 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6030 static int cards_found;
6031 int i, err, pci_using_dac;
6032 unsigned int indices = num_possible_cpus();
6033 #ifdef IXGBE_FCOE
6034 u16 device_caps;
6035 #endif
6036 u32 part_num, eec;
6037
6038 err = pci_enable_device_mem(pdev);
6039 if (err)
6040 return err;
6041
6042 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
6043 !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
6044 pci_using_dac = 1;
6045 } else {
6046 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
6047 if (err) {
6048 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
6049 if (err) {
6050 dev_err(&pdev->dev, "No usable DMA "
6051 "configuration, aborting\n");
6052 goto err_dma;
6053 }
6054 }
6055 pci_using_dac = 0;
6056 }
6057
6058 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6059 IORESOURCE_MEM), ixgbe_driver_name);
6060 if (err) {
6061 dev_err(&pdev->dev,
6062 "pci_request_selected_regions failed 0x%x\n", err);
6063 goto err_pci_reg;
6064 }
6065
6066 pci_enable_pcie_error_reporting(pdev);
6067
6068 pci_set_master(pdev);
6069 pci_save_state(pdev);
6070
6071 if (ii->mac == ixgbe_mac_82598EB)
6072 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6073 else
6074 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6075
6076 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6077 #ifdef IXGBE_FCOE
6078 indices += min_t(unsigned int, num_possible_cpus(),
6079 IXGBE_MAX_FCOE_INDICES);
6080 #endif
6081 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6082 if (!netdev) {
6083 err = -ENOMEM;
6084 goto err_alloc_etherdev;
6085 }
6086
6087 SET_NETDEV_DEV(netdev, &pdev->dev);
6088
6089 pci_set_drvdata(pdev, netdev);
6090 adapter = netdev_priv(netdev);
6091
6092 adapter->netdev = netdev;
6093 adapter->pdev = pdev;
6094 hw = &adapter->hw;
6095 hw->back = adapter;
6096 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6097
6098 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6099 pci_resource_len(pdev, 0));
6100 if (!hw->hw_addr) {
6101 err = -EIO;
6102 goto err_ioremap;
6103 }
6104
6105 for (i = 1; i <= 5; i++) {
6106 if (pci_resource_len(pdev, i) == 0)
6107 continue;
6108 }
6109
6110 netdev->netdev_ops = &ixgbe_netdev_ops;
6111 ixgbe_set_ethtool_ops(netdev);
6112 netdev->watchdog_timeo = 5 * HZ;
6113 strcpy(netdev->name, pci_name(pdev));
6114
6115 adapter->bd_number = cards_found;
6116
6117 /* Setup hw api */
6118 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6119 hw->mac.type = ii->mac;
6120
6121 /* EEPROM */
6122 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6123 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6124 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6125 if (!(eec & (1 << 8)))
6126 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6127
6128 /* PHY */
6129 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6130 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6131 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6132 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6133 hw->phy.mdio.mmds = 0;
6134 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6135 hw->phy.mdio.dev = netdev;
6136 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6137 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6138
6139 /* set up this timer and work struct before calling get_invariants
6140 * which might start the timer
6141 */
6142 init_timer(&adapter->sfp_timer);
6143 adapter->sfp_timer.function = &ixgbe_sfp_timer;
6144 adapter->sfp_timer.data = (unsigned long) adapter;
6145
6146 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6147
6148 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6149 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6150
6151 /* a new SFP+ module arrival, called from GPI SDP2 context */
6152 INIT_WORK(&adapter->sfp_config_module_task,
6153 ixgbe_sfp_config_module_task);
6154
6155 ii->get_invariants(hw);
6156
6157 /* setup the private structure */
6158 err = ixgbe_sw_init(adapter);
6159 if (err)
6160 goto err_sw_init;
6161
6162 /* Make it possible the adapter to be woken up via WOL */
6163 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6164 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6165
6166 /*
6167 * If there is a fan on this device and it has failed log the
6168 * failure.
6169 */
6170 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6171 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6172 if (esdp & IXGBE_ESDP_SDP1)
6173 DPRINTK(PROBE, CRIT,
6174 "Fan has stopped, replace the adapter\n");
6175 }
6176
6177 /* reset_hw fills in the perm_addr as well */
6178 err = hw->mac.ops.reset_hw(hw);
6179 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6180 hw->mac.type == ixgbe_mac_82598EB) {
6181 /*
6182 * Start a kernel thread to watch for a module to arrive.
6183 * Only do this for 82598, since 82599 will generate
6184 * interrupts on module arrival.
6185 */
6186 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6187 mod_timer(&adapter->sfp_timer,
6188 round_jiffies(jiffies + (2 * HZ)));
6189 err = 0;
6190 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6191 dev_err(&adapter->pdev->dev, "failed to initialize because "
6192 "an unsupported SFP+ module type was detected.\n"
6193 "Reload the driver after installing a supported "
6194 "module.\n");
6195 goto err_sw_init;
6196 } else if (err) {
6197 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
6198 goto err_sw_init;
6199 }
6200
6201 ixgbe_probe_vf(adapter, ii);
6202
6203 netdev->features = NETIF_F_SG |
6204 NETIF_F_IP_CSUM |
6205 NETIF_F_HW_VLAN_TX |
6206 NETIF_F_HW_VLAN_RX |
6207 NETIF_F_HW_VLAN_FILTER;
6208
6209 netdev->features |= NETIF_F_IPV6_CSUM;
6210 netdev->features |= NETIF_F_TSO;
6211 netdev->features |= NETIF_F_TSO6;
6212 netdev->features |= NETIF_F_GRO;
6213
6214 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6215 netdev->features |= NETIF_F_SCTP_CSUM;
6216
6217 netdev->vlan_features |= NETIF_F_TSO;
6218 netdev->vlan_features |= NETIF_F_TSO6;
6219 netdev->vlan_features |= NETIF_F_IP_CSUM;
6220 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6221 netdev->vlan_features |= NETIF_F_SG;
6222
6223 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6224 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6225 IXGBE_FLAG_DCB_ENABLED);
6226 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6227 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6228
6229 #ifdef CONFIG_IXGBE_DCB
6230 netdev->dcbnl_ops = &dcbnl_ops;
6231 #endif
6232
6233 #ifdef IXGBE_FCOE
6234 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6235 if (hw->mac.ops.get_device_caps) {
6236 hw->mac.ops.get_device_caps(hw, &device_caps);
6237 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6238 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6239 }
6240 }
6241 #endif /* IXGBE_FCOE */
6242 if (pci_using_dac)
6243 netdev->features |= NETIF_F_HIGHDMA;
6244
6245 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6246 netdev->features |= NETIF_F_LRO;
6247
6248 /* make sure the EEPROM is good */
6249 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6250 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
6251 err = -EIO;
6252 goto err_eeprom;
6253 }
6254
6255 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6256 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6257
6258 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6259 dev_err(&pdev->dev, "invalid MAC address\n");
6260 err = -EIO;
6261 goto err_eeprom;
6262 }
6263
6264 /* power down the optics */
6265 if (hw->phy.multispeed_fiber)
6266 hw->mac.ops.disable_tx_laser(hw);
6267
6268 init_timer(&adapter->watchdog_timer);
6269 adapter->watchdog_timer.function = &ixgbe_watchdog;
6270 adapter->watchdog_timer.data = (unsigned long)adapter;
6271
6272 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6273 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6274
6275 err = ixgbe_init_interrupt_scheme(adapter);
6276 if (err)
6277 goto err_sw_init;
6278
6279 switch (pdev->device) {
6280 case IXGBE_DEV_ID_82599_KX4:
6281 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
6282 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
6283 break;
6284 default:
6285 adapter->wol = 0;
6286 break;
6287 }
6288 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6289
6290 /* pick up the PCI bus settings for reporting later */
6291 hw->mac.ops.get_bus_info(hw);
6292
6293 /* print bus type/speed/width info */
6294 dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
6295 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
6296 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
6297 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
6298 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
6299 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
6300 "Unknown"),
6301 netdev->dev_addr);
6302 ixgbe_read_pba_num_generic(hw, &part_num);
6303 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
6304 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
6305 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
6306 (part_num >> 8), (part_num & 0xff));
6307 else
6308 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
6309 hw->mac.type, hw->phy.type,
6310 (part_num >> 8), (part_num & 0xff));
6311
6312 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
6313 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
6314 "this card is not sufficient for optimal "
6315 "performance.\n");
6316 dev_warn(&pdev->dev, "For optimal performance a x8 "
6317 "PCI-Express slot is required.\n");
6318 }
6319
6320 /* save off EEPROM version number */
6321 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
6322
6323 /* reset the hardware with the new settings */
6324 err = hw->mac.ops.start_hw(hw);
6325
6326 if (err == IXGBE_ERR_EEPROM_VERSION) {
6327 /* We are running on a pre-production device, log a warning */
6328 dev_warn(&pdev->dev, "This device is a pre-production "
6329 "adapter/LOM. Please be aware there may be issues "
6330 "associated with your hardware. If you are "
6331 "experiencing problems please contact your Intel or "
6332 "hardware representative who provided you with this "
6333 "hardware.\n");
6334 }
6335 strcpy(netdev->name, "eth%d");
6336 err = register_netdev(netdev);
6337 if (err)
6338 goto err_register;
6339
6340 /* carrier off reporting is important to ethtool even BEFORE open */
6341 netif_carrier_off(netdev);
6342
6343 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6344 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6345 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
6346
6347 #ifdef CONFIG_IXGBE_DCA
6348 if (dca_add_requester(&pdev->dev) == 0) {
6349 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
6350 ixgbe_setup_dca(adapter);
6351 }
6352 #endif
6353 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6354 DPRINTK(PROBE, INFO, "IOV is enabled with %d VFs\n",
6355 adapter->num_vfs);
6356 for (i = 0; i < adapter->num_vfs; i++)
6357 ixgbe_vf_configuration(pdev, (i | 0x10000000));
6358 }
6359
6360 /* add san mac addr to netdev */
6361 ixgbe_add_sanmac_netdev(netdev);
6362
6363 dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
6364 cards_found++;
6365 return 0;
6366
6367 err_register:
6368 ixgbe_release_hw_control(adapter);
6369 ixgbe_clear_interrupt_scheme(adapter);
6370 err_sw_init:
6371 err_eeprom:
6372 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6373 ixgbe_disable_sriov(adapter);
6374 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6375 del_timer_sync(&adapter->sfp_timer);
6376 cancel_work_sync(&adapter->sfp_task);
6377 cancel_work_sync(&adapter->multispeed_fiber_task);
6378 cancel_work_sync(&adapter->sfp_config_module_task);
6379 iounmap(hw->hw_addr);
6380 err_ioremap:
6381 free_netdev(netdev);
6382 err_alloc_etherdev:
6383 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6384 IORESOURCE_MEM));
6385 err_pci_reg:
6386 err_dma:
6387 pci_disable_device(pdev);
6388 return err;
6389 }
6390
6391 /**
6392 * ixgbe_remove - Device Removal Routine
6393 * @pdev: PCI device information struct
6394 *
6395 * ixgbe_remove is called by the PCI subsystem to alert the driver
6396 * that it should release a PCI device. The could be caused by a
6397 * Hot-Plug event, or because the driver is going to be removed from
6398 * memory.
6399 **/
6400 static void __devexit ixgbe_remove(struct pci_dev *pdev)
6401 {
6402 struct net_device *netdev = pci_get_drvdata(pdev);
6403 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6404
6405 set_bit(__IXGBE_DOWN, &adapter->state);
6406 /* clear the module not found bit to make sure the worker won't
6407 * reschedule
6408 */
6409 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6410 del_timer_sync(&adapter->watchdog_timer);
6411
6412 del_timer_sync(&adapter->sfp_timer);
6413 cancel_work_sync(&adapter->watchdog_task);
6414 cancel_work_sync(&adapter->sfp_task);
6415 cancel_work_sync(&adapter->multispeed_fiber_task);
6416 cancel_work_sync(&adapter->sfp_config_module_task);
6417 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6418 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6419 cancel_work_sync(&adapter->fdir_reinit_task);
6420 flush_scheduled_work();
6421
6422 #ifdef CONFIG_IXGBE_DCA
6423 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
6424 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
6425 dca_remove_requester(&pdev->dev);
6426 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
6427 }
6428
6429 #endif
6430 #ifdef IXGBE_FCOE
6431 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
6432 ixgbe_cleanup_fcoe(adapter);
6433
6434 #endif /* IXGBE_FCOE */
6435
6436 /* remove the added san mac */
6437 ixgbe_del_sanmac_netdev(netdev);
6438
6439 if (netdev->reg_state == NETREG_REGISTERED)
6440 unregister_netdev(netdev);
6441
6442 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6443 ixgbe_disable_sriov(adapter);
6444
6445 ixgbe_clear_interrupt_scheme(adapter);
6446
6447 ixgbe_release_hw_control(adapter);
6448
6449 iounmap(adapter->hw.hw_addr);
6450 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6451 IORESOURCE_MEM));
6452
6453 DPRINTK(PROBE, INFO, "complete\n");
6454
6455 free_netdev(netdev);
6456
6457 pci_disable_pcie_error_reporting(pdev);
6458
6459 pci_disable_device(pdev);
6460 }
6461
6462 /**
6463 * ixgbe_io_error_detected - called when PCI error is detected
6464 * @pdev: Pointer to PCI device
6465 * @state: The current pci connection state
6466 *
6467 * This function is called after a PCI bus error affecting
6468 * this device has been detected.
6469 */
6470 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
6471 pci_channel_state_t state)
6472 {
6473 struct net_device *netdev = pci_get_drvdata(pdev);
6474 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6475
6476 netif_device_detach(netdev);
6477
6478 if (state == pci_channel_io_perm_failure)
6479 return PCI_ERS_RESULT_DISCONNECT;
6480
6481 if (netif_running(netdev))
6482 ixgbe_down(adapter);
6483 pci_disable_device(pdev);
6484
6485 /* Request a slot reset. */
6486 return PCI_ERS_RESULT_NEED_RESET;
6487 }
6488
6489 /**
6490 * ixgbe_io_slot_reset - called after the pci bus has been reset.
6491 * @pdev: Pointer to PCI device
6492 *
6493 * Restart the card from scratch, as if from a cold-boot.
6494 */
6495 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
6496 {
6497 struct net_device *netdev = pci_get_drvdata(pdev);
6498 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6499 pci_ers_result_t result;
6500 int err;
6501
6502 if (pci_enable_device_mem(pdev)) {
6503 DPRINTK(PROBE, ERR,
6504 "Cannot re-enable PCI device after reset.\n");
6505 result = PCI_ERS_RESULT_DISCONNECT;
6506 } else {
6507 pci_set_master(pdev);
6508 pci_restore_state(pdev);
6509 pci_save_state(pdev);
6510
6511 pci_wake_from_d3(pdev, false);
6512
6513 ixgbe_reset(adapter);
6514 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6515 result = PCI_ERS_RESULT_RECOVERED;
6516 }
6517
6518 err = pci_cleanup_aer_uncorrect_error_status(pdev);
6519 if (err) {
6520 dev_err(&pdev->dev,
6521 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
6522 /* non-fatal, continue */
6523 }
6524
6525 return result;
6526 }
6527
6528 /**
6529 * ixgbe_io_resume - called when traffic can start flowing again.
6530 * @pdev: Pointer to PCI device
6531 *
6532 * This callback is called when the error recovery driver tells us that
6533 * its OK to resume normal operation.
6534 */
6535 static void ixgbe_io_resume(struct pci_dev *pdev)
6536 {
6537 struct net_device *netdev = pci_get_drvdata(pdev);
6538 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6539
6540 if (netif_running(netdev)) {
6541 if (ixgbe_up(adapter)) {
6542 DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
6543 return;
6544 }
6545 }
6546
6547 netif_device_attach(netdev);
6548 }
6549
6550 static struct pci_error_handlers ixgbe_err_handler = {
6551 .error_detected = ixgbe_io_error_detected,
6552 .slot_reset = ixgbe_io_slot_reset,
6553 .resume = ixgbe_io_resume,
6554 };
6555
6556 static struct pci_driver ixgbe_driver = {
6557 .name = ixgbe_driver_name,
6558 .id_table = ixgbe_pci_tbl,
6559 .probe = ixgbe_probe,
6560 .remove = __devexit_p(ixgbe_remove),
6561 #ifdef CONFIG_PM
6562 .suspend = ixgbe_suspend,
6563 .resume = ixgbe_resume,
6564 #endif
6565 .shutdown = ixgbe_shutdown,
6566 .err_handler = &ixgbe_err_handler
6567 };
6568
6569 /**
6570 * ixgbe_init_module - Driver Registration Routine
6571 *
6572 * ixgbe_init_module is the first routine called when the driver is
6573 * loaded. All it does is register with the PCI subsystem.
6574 **/
6575 static int __init ixgbe_init_module(void)
6576 {
6577 int ret;
6578 printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
6579 ixgbe_driver_string, ixgbe_driver_version);
6580
6581 printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
6582
6583 #ifdef CONFIG_IXGBE_DCA
6584 dca_register_notify(&dca_notifier);
6585 #endif
6586
6587 ret = pci_register_driver(&ixgbe_driver);
6588 return ret;
6589 }
6590
6591 module_init(ixgbe_init_module);
6592
6593 /**
6594 * ixgbe_exit_module - Driver Exit Cleanup Routine
6595 *
6596 * ixgbe_exit_module is called just before the driver is removed
6597 * from memory.
6598 **/
6599 static void __exit ixgbe_exit_module(void)
6600 {
6601 #ifdef CONFIG_IXGBE_DCA
6602 dca_unregister_notify(&dca_notifier);
6603 #endif
6604 pci_unregister_driver(&ixgbe_driver);
6605 }
6606
6607 #ifdef CONFIG_IXGBE_DCA
6608 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
6609 void *p)
6610 {
6611 int ret_val;
6612
6613 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
6614 __ixgbe_notify_dca);
6615
6616 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6617 }
6618
6619 #endif /* CONFIG_IXGBE_DCA */
6620 #ifdef DEBUG
6621 /**
6622 * ixgbe_get_hw_dev_name - return device name string
6623 * used by hardware layer to print debugging information
6624 **/
6625 char *ixgbe_get_hw_dev_name(struct ixgbe_hw *hw)
6626 {
6627 struct ixgbe_adapter *adapter = hw->back;
6628 return adapter->netdev->name;
6629 }
6630
6631 #endif
6632 module_exit(ixgbe_exit_module);
6633
6634 /* ixgbe_main.c */
Linux kernel - Deliverables
This page took 0.163823 seconds and 6 git commands to generate.