?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ixgbe: fix slow load times on 82598 nics
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / net / ixgbe / ixgbe_main.c
blob18d4353afa65352492add64d94ee34b70c00bc83
1 /*******************************************************************************
2
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2008 Intel Corporation.
5
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
9
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 more details.
14
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
21
22 Contact Information:
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
34 #include <linux/in.h>
35 #include <linux/ip.h>
36 #include <linux/tcp.h>
37 #include <linux/ipv6.h>
38 #include <net/checksum.h>
39 #include <net/ip6_checksum.h>
40 #include <linux/ethtool.h>
41 #include <linux/if_vlan.h>
42
43 #include "ixgbe.h"
44 #include "ixgbe_common.h"
45
46 char ixgbe_driver_name[] = "ixgbe";
47 static const char ixgbe_driver_string[] =
48 "Intel(R) 10 Gigabit PCI Express Network Driver";
49
50 #define DRV_VERSION "1.3.30-k2"
51 const char ixgbe_driver_version[] = DRV_VERSION;
52 static char ixgbe_copyright[] = "Copyright (c) 1999-2007 Intel Corporation.";
53
54 static const struct ixgbe_info *ixgbe_info_tbl[] = {
55 [board_82598] = &ixgbe_82598_info,
56 };
57
58 /* ixgbe_pci_tbl - PCI Device ID Table
59 *
60 * Wildcard entries (PCI_ANY_ID) should come last
61 * Last entry must be all 0s
62 *
63 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
64 * Class, Class Mask, private data (not used) }
65 */
66 static struct pci_device_id ixgbe_pci_tbl[] = {
67 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
68 board_82598 },
69 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
70 board_82598 },
71 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
72 board_82598 },
73 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
74 board_82598 },
75 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
76 board_82598 },
77 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
78 board_82598 },
79 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
80 board_82598 },
81 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
82 board_82598 },
83 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
84 board_82598 },
85
86 /* required last entry */
87 {0, }
88 };
89 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
90
91 #ifdef CONFIG_IXGBE_DCA
92 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
93 void *p);
94 static struct notifier_block dca_notifier = {
95 .notifier_call = ixgbe_notify_dca,
96 .next = NULL,
97 .priority = 0
98 };
99 #endif
100
101 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
102 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
103 MODULE_LICENSE("GPL");
104 MODULE_VERSION(DRV_VERSION);
105
106 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
107
108 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
109 {
110 u32 ctrl_ext;
111
112 /* Let firmware take over control of h/w */
113 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
114 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
115 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
116 }
117
118 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
119 {
120 u32 ctrl_ext;
121
122 /* Let firmware know the driver has taken over */
123 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
124 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
125 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
126 }
127
128 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, u16 int_alloc_entry,
129 u8 msix_vector)
130 {
131 u32 ivar, index;
132
133 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
134 index = (int_alloc_entry >> 2) & 0x1F;
135 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR(index));
136 ivar &= ~(0xFF << (8 * (int_alloc_entry & 0x3)));
137 ivar |= (msix_vector << (8 * (int_alloc_entry & 0x3)));
138 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR(index), ivar);
139 }
140
141 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
142 struct ixgbe_tx_buffer
143 *tx_buffer_info)
144 {
145 if (tx_buffer_info->dma) {
146 pci_unmap_page(adapter->pdev, tx_buffer_info->dma,
147 tx_buffer_info->length, PCI_DMA_TODEVICE);
148 tx_buffer_info->dma = 0;
149 }
150 if (tx_buffer_info->skb) {
151 dev_kfree_skb_any(tx_buffer_info->skb);
152 tx_buffer_info->skb = NULL;
153 }
154 /* tx_buffer_info must be completely set up in the transmit path */
155 }
156
157 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
158 struct ixgbe_ring *tx_ring,
159 unsigned int eop)
160 {
161 struct ixgbe_hw *hw = &adapter->hw;
162 u32 head, tail;
163
164 /* Detect a transmit hang in hardware, this serializes the
165 * check with the clearing of time_stamp and movement of eop */
166 head = IXGBE_READ_REG(hw, tx_ring->head);
167 tail = IXGBE_READ_REG(hw, tx_ring->tail);
168 adapter->detect_tx_hung = false;
169 if ((head != tail) &&
170 tx_ring->tx_buffer_info[eop].time_stamp &&
171 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
172 !(IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & IXGBE_TFCS_TXOFF)) {
173 /* detected Tx unit hang */
174 union ixgbe_adv_tx_desc *tx_desc;
175 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
176 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
177 " Tx Queue <%d>\n"
178 " TDH, TDT <%x>, <%x>\n"
179 " next_to_use <%x>\n"
180 " next_to_clean <%x>\n"
181 "tx_buffer_info[next_to_clean]\n"
182 " time_stamp <%lx>\n"
183 " jiffies <%lx>\n",
184 tx_ring->queue_index,
185 head, tail,
186 tx_ring->next_to_use, eop,
187 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
188 return true;
189 }
190
191 return false;
192 }
193
194 #define IXGBE_MAX_TXD_PWR 14
195 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
196
197 /* Tx Descriptors needed, worst case */
198 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
199 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
200 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
201 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
202
203 #define GET_TX_HEAD_FROM_RING(ring) (\
204 *(volatile u32 *) \
205 ((union ixgbe_adv_tx_desc *)(ring)->desc + (ring)->count))
206 static void ixgbe_tx_timeout(struct net_device *netdev);
207
208 /**
209 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
210 * @adapter: board private structure
211 * @tx_ring: tx ring to clean
212 **/
213 static bool ixgbe_clean_tx_irq(struct ixgbe_adapter *adapter,
214 struct ixgbe_ring *tx_ring)
215 {
216 union ixgbe_adv_tx_desc *tx_desc;
217 struct ixgbe_tx_buffer *tx_buffer_info;
218 struct net_device *netdev = adapter->netdev;
219 struct sk_buff *skb;
220 unsigned int i;
221 u32 head, oldhead;
222 unsigned int count = 0;
223 unsigned int total_bytes = 0, total_packets = 0;
224
225 rmb();
226 head = GET_TX_HEAD_FROM_RING(tx_ring);
227 head = le32_to_cpu(head);
228 i = tx_ring->next_to_clean;
229 while (1) {
230 while (i != head) {
231 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
232 tx_buffer_info = &tx_ring->tx_buffer_info[i];
233 skb = tx_buffer_info->skb;
234
235 if (skb) {
236 unsigned int segs, bytecount;
237
238 /* gso_segs is currently only valid for tcp */
239 segs = skb_shinfo(skb)->gso_segs ?: 1;
240 /* multiply data chunks by size of headers */
241 bytecount = ((segs - 1) * skb_headlen(skb)) +
242 skb->len;
243 total_packets += segs;
244 total_bytes += bytecount;
245 }
246
247 ixgbe_unmap_and_free_tx_resource(adapter,
248 tx_buffer_info);
249
250 i++;
251 if (i == tx_ring->count)
252 i = 0;
253
254 count++;
255 if (count == tx_ring->count)
256 goto done_cleaning;
257 }
258 oldhead = head;
259 rmb();
260 head = GET_TX_HEAD_FROM_RING(tx_ring);
261 head = le32_to_cpu(head);
262 if (head == oldhead)
263 goto done_cleaning;
264 } /* while (1) */
265
266 done_cleaning:
267 tx_ring->next_to_clean = i;
268
269 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
270 if (unlikely(count && netif_carrier_ok(netdev) &&
271 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
272 /* Make sure that anybody stopping the queue after this
273 * sees the new next_to_clean.
274 */
275 smp_mb();
276 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
277 !test_bit(__IXGBE_DOWN, &adapter->state)) {
278 netif_wake_subqueue(netdev, tx_ring->queue_index);
279 ++adapter->restart_queue;
280 }
281 }
282
283 if (adapter->detect_tx_hung) {
284 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
285 /* schedule immediate reset if we believe we hung */
286 DPRINTK(PROBE, INFO,
287 "tx hang %d detected, resetting adapter\n",
288 adapter->tx_timeout_count + 1);
289 ixgbe_tx_timeout(adapter->netdev);
290 }
291 }
292
293 /* re-arm the interrupt */
294 if ((total_packets >= tx_ring->work_limit) ||
295 (count == tx_ring->count))
296 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, tx_ring->v_idx);
297
298 tx_ring->total_bytes += total_bytes;
299 tx_ring->total_packets += total_packets;
300 tx_ring->stats.bytes += total_bytes;
301 tx_ring->stats.packets += total_packets;
302 adapter->net_stats.tx_bytes += total_bytes;
303 adapter->net_stats.tx_packets += total_packets;
304 return (total_packets ? true : false);
305 }
306
307 #ifdef CONFIG_IXGBE_DCA
308 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
309 struct ixgbe_ring *rx_ring)
310 {
311 u32 rxctrl;
312 int cpu = get_cpu();
313 int q = rx_ring - adapter->rx_ring;
314
315 if (rx_ring->cpu != cpu) {
316 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
317 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
318 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
319 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
320 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
321 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
322 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
323 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
324 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
325 rx_ring->cpu = cpu;
326 }
327 put_cpu();
328 }
329
330 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
331 struct ixgbe_ring *tx_ring)
332 {
333 u32 txctrl;
334 int cpu = get_cpu();
335 int q = tx_ring - adapter->tx_ring;
336
337 if (tx_ring->cpu != cpu) {
338 txctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q));
339 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
340 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
341 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
342 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q), txctrl);
343 tx_ring->cpu = cpu;
344 }
345 put_cpu();
346 }
347
348 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
349 {
350 int i;
351
352 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
353 return;
354
355 for (i = 0; i < adapter->num_tx_queues; i++) {
356 adapter->tx_ring[i].cpu = -1;
357 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
358 }
359 for (i = 0; i < adapter->num_rx_queues; i++) {
360 adapter->rx_ring[i].cpu = -1;
361 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
362 }
363 }
364
365 static int __ixgbe_notify_dca(struct device *dev, void *data)
366 {
367 struct net_device *netdev = dev_get_drvdata(dev);
368 struct ixgbe_adapter *adapter = netdev_priv(netdev);
369 unsigned long event = *(unsigned long *)data;
370
371 switch (event) {
372 case DCA_PROVIDER_ADD:
373 /* if we're already enabled, don't do it again */
374 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
375 break;
376 /* Always use CB2 mode, difference is masked
377 * in the CB driver. */
378 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
379 if (dca_add_requester(dev) == 0) {
380 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
381 ixgbe_setup_dca(adapter);
382 break;
383 }
384 /* Fall Through since DCA is disabled. */
385 case DCA_PROVIDER_REMOVE:
386 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
387 dca_remove_requester(dev);
388 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
389 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
390 }
391 break;
392 }
393
394 return 0;
395 }
396
397 #endif /* CONFIG_IXGBE_DCA */
398 /**
399 * ixgbe_receive_skb - Send a completed packet up the stack
400 * @adapter: board private structure
401 * @skb: packet to send up
402 * @status: hardware indication of status of receive
403 * @rx_ring: rx descriptor ring (for a specific queue) to setup
404 * @rx_desc: rx descriptor
405 **/
406 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
407 struct sk_buff *skb, u8 status,
408 union ixgbe_adv_rx_desc *rx_desc)
409 {
410 struct ixgbe_adapter *adapter = q_vector->adapter;
411 struct napi_struct *napi = &q_vector->napi;
412 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
413 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
414
415 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
416 if (adapter->vlgrp && is_vlan && (tag != 0))
417 vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
418 else
419 napi_gro_receive(napi, skb);
420 } else {
421 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
422 if (adapter->vlgrp && is_vlan && (tag != 0))
423 vlan_hwaccel_receive_skb(skb, adapter->vlgrp, tag);
424 else
425 netif_receive_skb(skb);
426 } else {
427 if (adapter->vlgrp && is_vlan && (tag != 0))
428 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
429 else
430 netif_rx(skb);
431 }
432 }
433 }
434
435 /**
436 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
437 * @adapter: address of board private structure
438 * @status_err: hardware indication of status of receive
439 * @skb: skb currently being received and modified
440 **/
441 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
442 u32 status_err, struct sk_buff *skb)
443 {
444 skb->ip_summed = CHECKSUM_NONE;
445
446 /* Rx csum disabled */
447 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
448 return;
449
450 /* if IP and error */
451 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
452 (status_err & IXGBE_RXDADV_ERR_IPE)) {
453 adapter->hw_csum_rx_error++;
454 return;
455 }
456
457 if (!(status_err & IXGBE_RXD_STAT_L4CS))
458 return;
459
460 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
461 adapter->hw_csum_rx_error++;
462 return;
463 }
464
465 /* It must be a TCP or UDP packet with a valid checksum */
466 skb->ip_summed = CHECKSUM_UNNECESSARY;
467 adapter->hw_csum_rx_good++;
468 }
469
470 /**
471 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
472 * @adapter: address of board private structure
473 **/
474 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
475 struct ixgbe_ring *rx_ring,
476 int cleaned_count)
477 {
478 struct pci_dev *pdev = adapter->pdev;
479 union ixgbe_adv_rx_desc *rx_desc;
480 struct ixgbe_rx_buffer *bi;
481 unsigned int i;
482
483 i = rx_ring->next_to_use;
484 bi = &rx_ring->rx_buffer_info[i];
485
486 while (cleaned_count--) {
487 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
488
489 if (!bi->page_dma &&
490 (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
491 if (!bi->page) {
492 bi->page = alloc_page(GFP_ATOMIC);
493 if (!bi->page) {
494 adapter->alloc_rx_page_failed++;
495 goto no_buffers;
496 }
497 bi->page_offset = 0;
498 } else {
499 /* use a half page if we're re-using */
500 bi->page_offset ^= (PAGE_SIZE / 2);
501 }
502
503 bi->page_dma = pci_map_page(pdev, bi->page,
504 bi->page_offset,
505 (PAGE_SIZE / 2),
506 PCI_DMA_FROMDEVICE);
507 }
508
509 if (!bi->skb) {
510 struct sk_buff *skb;
511 skb = netdev_alloc_skb(adapter->netdev,
512 (rx_ring->rx_buf_len +
513 NET_IP_ALIGN));
514
515 if (!skb) {
516 adapter->alloc_rx_buff_failed++;
517 goto no_buffers;
518 }
519
520 /*
521 * Make buffer alignment 2 beyond a 16 byte boundary
522 * this will result in a 16 byte aligned IP header after
523 * the 14 byte MAC header is removed
524 */
525 skb_reserve(skb, NET_IP_ALIGN);
526
527 bi->skb = skb;
528 bi->dma = pci_map_single(pdev, skb->data,
529 rx_ring->rx_buf_len,
530 PCI_DMA_FROMDEVICE);
531 }
532 /* Refresh the desc even if buffer_addrs didn't change because
533 * each write-back erases this info. */
534 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
535 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
536 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
537 } else {
538 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
539 }
540
541 i++;
542 if (i == rx_ring->count)
543 i = 0;
544 bi = &rx_ring->rx_buffer_info[i];
545 }
546
547 no_buffers:
548 if (rx_ring->next_to_use != i) {
549 rx_ring->next_to_use = i;
550 if (i-- == 0)
551 i = (rx_ring->count - 1);
552
553 /*
554 * Force memory writes to complete before letting h/w
555 * know there are new descriptors to fetch. (Only
556 * applicable for weak-ordered memory model archs,
557 * such as IA-64).
558 */
559 wmb();
560 writel(i, adapter->hw.hw_addr + rx_ring->tail);
561 }
562 }
563
564 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
565 {
566 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
567 }
568
569 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
570 {
571 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
572 }
573
574 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
575 struct ixgbe_ring *rx_ring,
576 int *work_done, int work_to_do)
577 {
578 struct ixgbe_adapter *adapter = q_vector->adapter;
579 struct pci_dev *pdev = adapter->pdev;
580 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
581 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
582 struct sk_buff *skb;
583 unsigned int i;
584 u32 len, staterr;
585 u16 hdr_info;
586 bool cleaned = false;
587 int cleaned_count = 0;
588 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
589
590 i = rx_ring->next_to_clean;
591 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
592 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
593 rx_buffer_info = &rx_ring->rx_buffer_info[i];
594
595 while (staterr & IXGBE_RXD_STAT_DD) {
596 u32 upper_len = 0;
597 if (*work_done >= work_to_do)
598 break;
599 (*work_done)++;
600
601 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
602 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
603 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
604 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
605 if (hdr_info & IXGBE_RXDADV_SPH)
606 adapter->rx_hdr_split++;
607 if (len > IXGBE_RX_HDR_SIZE)
608 len = IXGBE_RX_HDR_SIZE;
609 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
610 } else {
611 len = le16_to_cpu(rx_desc->wb.upper.length);
612 }
613
614 cleaned = true;
615 skb = rx_buffer_info->skb;
616 prefetch(skb->data - NET_IP_ALIGN);
617 rx_buffer_info->skb = NULL;
618
619 if (len && !skb_shinfo(skb)->nr_frags) {
620 pci_unmap_single(pdev, rx_buffer_info->dma,
621 rx_ring->rx_buf_len,
622 PCI_DMA_FROMDEVICE);
623 skb_put(skb, len);
624 }
625
626 if (upper_len) {
627 pci_unmap_page(pdev, rx_buffer_info->page_dma,
628 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
629 rx_buffer_info->page_dma = 0;
630 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
631 rx_buffer_info->page,
632 rx_buffer_info->page_offset,
633 upper_len);
634
635 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
636 (page_count(rx_buffer_info->page) != 1))
637 rx_buffer_info->page = NULL;
638 else
639 get_page(rx_buffer_info->page);
640
641 skb->len += upper_len;
642 skb->data_len += upper_len;
643 skb->truesize += upper_len;
644 }
645
646 i++;
647 if (i == rx_ring->count)
648 i = 0;
649 next_buffer = &rx_ring->rx_buffer_info[i];
650
651 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
652 prefetch(next_rxd);
653
654 cleaned_count++;
655 if (staterr & IXGBE_RXD_STAT_EOP) {
656 rx_ring->stats.packets++;
657 rx_ring->stats.bytes += skb->len;
658 } else {
659 rx_buffer_info->skb = next_buffer->skb;
660 rx_buffer_info->dma = next_buffer->dma;
661 next_buffer->skb = skb;
662 next_buffer->dma = 0;
663 adapter->non_eop_descs++;
664 goto next_desc;
665 }
666
667 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
668 dev_kfree_skb_irq(skb);
669 goto next_desc;
670 }
671
672 ixgbe_rx_checksum(adapter, staterr, skb);
673
674 /* probably a little skewed due to removing CRC */
675 total_rx_bytes += skb->len;
676 total_rx_packets++;
677
678 skb->protocol = eth_type_trans(skb, adapter->netdev);
679 ixgbe_receive_skb(q_vector, skb, staterr, rx_desc);
680
681 next_desc:
682 rx_desc->wb.upper.status_error = 0;
683
684 /* return some buffers to hardware, one at a time is too slow */
685 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
686 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
687 cleaned_count = 0;
688 }
689
690 /* use prefetched values */
691 rx_desc = next_rxd;
692 rx_buffer_info = next_buffer;
693
694 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
695 }
696
697 rx_ring->next_to_clean = i;
698 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
699
700 if (cleaned_count)
701 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
702
703 rx_ring->total_packets += total_rx_packets;
704 rx_ring->total_bytes += total_rx_bytes;
705 adapter->net_stats.rx_bytes += total_rx_bytes;
706 adapter->net_stats.rx_packets += total_rx_packets;
707
708 return cleaned;
709 }
710
711 static int ixgbe_clean_rxonly(struct napi_struct *, int);
712 /**
713 * ixgbe_configure_msix - Configure MSI-X hardware
714 * @adapter: board private structure
715 *
716 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
717 * interrupts.
718 **/
719 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
720 {
721 struct ixgbe_q_vector *q_vector;
722 int i, j, q_vectors, v_idx, r_idx;
723 u32 mask;
724
725 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
726
727 /* Populate the IVAR table and set the ITR values to the
728 * corresponding register.
729 */
730 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
731 q_vector = &adapter->q_vector[v_idx];
732 /* XXX for_each_bit(...) */
733 r_idx = find_first_bit(q_vector->rxr_idx,
734 adapter->num_rx_queues);
735
736 for (i = 0; i < q_vector->rxr_count; i++) {
737 j = adapter->rx_ring[r_idx].reg_idx;
738 ixgbe_set_ivar(adapter, IXGBE_IVAR_RX_QUEUE(j), v_idx);
739 r_idx = find_next_bit(q_vector->rxr_idx,
740 adapter->num_rx_queues,
741 r_idx + 1);
742 }
743 r_idx = find_first_bit(q_vector->txr_idx,
744 adapter->num_tx_queues);
745
746 for (i = 0; i < q_vector->txr_count; i++) {
747 j = adapter->tx_ring[r_idx].reg_idx;
748 ixgbe_set_ivar(adapter, IXGBE_IVAR_TX_QUEUE(j), v_idx);
749 r_idx = find_next_bit(q_vector->txr_idx,
750 adapter->num_tx_queues,
751 r_idx + 1);
752 }
753
754 /* if this is a tx only vector halve the interrupt rate */
755 if (q_vector->txr_count && !q_vector->rxr_count)
756 q_vector->eitr = (adapter->eitr_param >> 1);
757 else
758 /* rx only */
759 q_vector->eitr = adapter->eitr_param;
760
761 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx),
762 EITR_INTS_PER_SEC_TO_REG(q_vector->eitr));
763 }
764
765 ixgbe_set_ivar(adapter, IXGBE_IVAR_OTHER_CAUSES_INDEX, v_idx);
766 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
767
768 /* set up to autoclear timer, and the vectors */
769 mask = IXGBE_EIMS_ENABLE_MASK;
770 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
771 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
772 }
773
774 enum latency_range {
775 lowest_latency = 0,
776 low_latency = 1,
777 bulk_latency = 2,
778 latency_invalid = 255
779 };
780
781 /**
782 * ixgbe_update_itr - update the dynamic ITR value based on statistics
783 * @adapter: pointer to adapter
784 * @eitr: eitr setting (ints per sec) to give last timeslice
785 * @itr_setting: current throttle rate in ints/second
786 * @packets: the number of packets during this measurement interval
787 * @bytes: the number of bytes during this measurement interval
788 *
789 * Stores a new ITR value based on packets and byte
790 * counts during the last interrupt. The advantage of per interrupt
791 * computation is faster updates and more accurate ITR for the current
792 * traffic pattern. Constants in this function were computed
793 * based on theoretical maximum wire speed and thresholds were set based
794 * on testing data as well as attempting to minimize response time
795 * while increasing bulk throughput.
796 * this functionality is controlled by the InterruptThrottleRate module
797 * parameter (see ixgbe_param.c)
798 **/
799 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
800 u32 eitr, u8 itr_setting,
801 int packets, int bytes)
802 {
803 unsigned int retval = itr_setting;
804 u32 timepassed_us;
805 u64 bytes_perint;
806
807 if (packets == 0)
808 goto update_itr_done;
809
810
811 /* simple throttlerate management
812 * 0-20MB/s lowest (100000 ints/s)
813 * 20-100MB/s low (20000 ints/s)
814 * 100-1249MB/s bulk (8000 ints/s)
815 */
816 /* what was last interrupt timeslice? */
817 timepassed_us = 1000000/eitr;
818 bytes_perint = bytes / timepassed_us; /* bytes/usec */
819
820 switch (itr_setting) {
821 case lowest_latency:
822 if (bytes_perint > adapter->eitr_low)
823 retval = low_latency;
824 break;
825 case low_latency:
826 if (bytes_perint > adapter->eitr_high)
827 retval = bulk_latency;
828 else if (bytes_perint <= adapter->eitr_low)
829 retval = lowest_latency;
830 break;
831 case bulk_latency:
832 if (bytes_perint <= adapter->eitr_high)
833 retval = low_latency;
834 break;
835 }
836
837 update_itr_done:
838 return retval;
839 }
840
841 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
842 {
843 struct ixgbe_adapter *adapter = q_vector->adapter;
844 struct ixgbe_hw *hw = &adapter->hw;
845 u32 new_itr;
846 u8 current_itr, ret_itr;
847 int i, r_idx, v_idx = ((void *)q_vector - (void *)(adapter->q_vector)) /
848 sizeof(struct ixgbe_q_vector);
849 struct ixgbe_ring *rx_ring, *tx_ring;
850
851 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
852 for (i = 0; i < q_vector->txr_count; i++) {
853 tx_ring = &(adapter->tx_ring[r_idx]);
854 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
855 q_vector->tx_itr,
856 tx_ring->total_packets,
857 tx_ring->total_bytes);
858 /* if the result for this queue would decrease interrupt
859 * rate for this vector then use that result */
860 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
861 q_vector->tx_itr - 1 : ret_itr);
862 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
863 r_idx + 1);
864 }
865
866 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
867 for (i = 0; i < q_vector->rxr_count; i++) {
868 rx_ring = &(adapter->rx_ring[r_idx]);
869 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
870 q_vector->rx_itr,
871 rx_ring->total_packets,
872 rx_ring->total_bytes);
873 /* if the result for this queue would decrease interrupt
874 * rate for this vector then use that result */
875 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
876 q_vector->rx_itr - 1 : ret_itr);
877 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
878 r_idx + 1);
879 }
880
881 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
882
883 switch (current_itr) {
884 /* counts and packets in update_itr are dependent on these numbers */
885 case lowest_latency:
886 new_itr = 100000;
887 break;
888 case low_latency:
889 new_itr = 20000; /* aka hwitr = ~200 */
890 break;
891 case bulk_latency:
892 default:
893 new_itr = 8000;
894 break;
895 }
896
897 if (new_itr != q_vector->eitr) {
898 u32 itr_reg;
899 /* do an exponential smoothing */
900 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
901 q_vector->eitr = new_itr;
902 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
903 /* must write high and low 16 bits to reset counter */
904 DPRINTK(TX_ERR, DEBUG, "writing eitr(%d): %08X\n", v_idx,
905 itr_reg);
906 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg | (itr_reg)<<16);
907 }
908
909 return;
910 }
911
912 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
913 {
914 struct ixgbe_hw *hw = &adapter->hw;
915
916 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
917 (eicr & IXGBE_EICR_GPI_SDP1)) {
918 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
919 /* write to clear the interrupt */
920 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
921 }
922 }
923
924 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
925 {
926 struct ixgbe_hw *hw = &adapter->hw;
927
928 adapter->lsc_int++;
929 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
930 adapter->link_check_timeout = jiffies;
931 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
932 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
933 schedule_work(&adapter->watchdog_task);
934 }
935 }
936
937 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
938 {
939 struct net_device *netdev = data;
940 struct ixgbe_adapter *adapter = netdev_priv(netdev);
941 struct ixgbe_hw *hw = &adapter->hw;
942 u32 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
943
944 if (eicr & IXGBE_EICR_LSC)
945 ixgbe_check_lsc(adapter);
946
947 ixgbe_check_fan_failure(adapter, eicr);
948
949 if (!test_bit(__IXGBE_DOWN, &adapter->state))
950 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
951
952 return IRQ_HANDLED;
953 }
954
955 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
956 {
957 struct ixgbe_q_vector *q_vector = data;
958 struct ixgbe_adapter *adapter = q_vector->adapter;
959 struct ixgbe_ring *tx_ring;
960 int i, r_idx;
961
962 if (!q_vector->txr_count)
963 return IRQ_HANDLED;
964
965 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
966 for (i = 0; i < q_vector->txr_count; i++) {
967 tx_ring = &(adapter->tx_ring[r_idx]);
968 #ifdef CONFIG_IXGBE_DCA
969 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
970 ixgbe_update_tx_dca(adapter, tx_ring);
971 #endif
972 tx_ring->total_bytes = 0;
973 tx_ring->total_packets = 0;
974 ixgbe_clean_tx_irq(adapter, tx_ring);
975 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
976 r_idx + 1);
977 }
978
979 return IRQ_HANDLED;
980 }
981
982 /**
983 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
984 * @irq: unused
985 * @data: pointer to our q_vector struct for this interrupt vector
986 **/
987 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
988 {
989 struct ixgbe_q_vector *q_vector = data;
990 struct ixgbe_adapter *adapter = q_vector->adapter;
991 struct ixgbe_ring *rx_ring;
992 int r_idx;
993 int i;
994
995 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
996 for (i = 0; i < q_vector->rxr_count; i++) {
997 rx_ring = &(adapter->rx_ring[r_idx]);
998 rx_ring->total_bytes = 0;
999 rx_ring->total_packets = 0;
1000 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1001 r_idx + 1);
1002 }
1003
1004 if (!q_vector->rxr_count)
1005 return IRQ_HANDLED;
1006
1007 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1008 rx_ring = &(adapter->rx_ring[r_idx]);
1009 /* disable interrupts on this vector only */
1010 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, rx_ring->v_idx);
1011 napi_schedule(&q_vector->napi);
1012
1013 return IRQ_HANDLED;
1014 }
1015
1016 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1017 {
1018 ixgbe_msix_clean_rx(irq, data);
1019 ixgbe_msix_clean_tx(irq, data);
1020
1021 return IRQ_HANDLED;
1022 }
1023
1024 /**
1025 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1026 * @napi: napi struct with our devices info in it
1027 * @budget: amount of work driver is allowed to do this pass, in packets
1028 *
1029 * This function is optimized for cleaning one queue only on a single
1030 * q_vector!!!
1031 **/
1032 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1033 {
1034 struct ixgbe_q_vector *q_vector =
1035 container_of(napi, struct ixgbe_q_vector, napi);
1036 struct ixgbe_adapter *adapter = q_vector->adapter;
1037 struct ixgbe_ring *rx_ring = NULL;
1038 int work_done = 0;
1039 long r_idx;
1040
1041 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1042 rx_ring = &(adapter->rx_ring[r_idx]);
1043 #ifdef CONFIG_IXGBE_DCA
1044 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1045 ixgbe_update_rx_dca(adapter, rx_ring);
1046 #endif
1047
1048 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1049
1050 /* If all Rx work done, exit the polling mode */
1051 if (work_done < budget) {
1052 napi_complete(napi);
1053 if (adapter->itr_setting & 3)
1054 ixgbe_set_itr_msix(q_vector);
1055 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1056 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, rx_ring->v_idx);
1057 }
1058
1059 return work_done;
1060 }
1061
1062 /**
1063 * ixgbe_clean_rxonly_many - msix (aka one shot) rx clean routine
1064 * @napi: napi struct with our devices info in it
1065 * @budget: amount of work driver is allowed to do this pass, in packets
1066 *
1067 * This function will clean more than one rx queue associated with a
1068 * q_vector.
1069 **/
1070 static int ixgbe_clean_rxonly_many(struct napi_struct *napi, int budget)
1071 {
1072 struct ixgbe_q_vector *q_vector =
1073 container_of(napi, struct ixgbe_q_vector, napi);
1074 struct ixgbe_adapter *adapter = q_vector->adapter;
1075 struct ixgbe_ring *rx_ring = NULL;
1076 int work_done = 0, i;
1077 long r_idx;
1078 u16 enable_mask = 0;
1079
1080 /* attempt to distribute budget to each queue fairly, but don't allow
1081 * the budget to go below 1 because we'll exit polling */
1082 budget /= (q_vector->rxr_count ?: 1);
1083 budget = max(budget, 1);
1084 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1085 for (i = 0; i < q_vector->rxr_count; i++) {
1086 rx_ring = &(adapter->rx_ring[r_idx]);
1087 #ifdef CONFIG_IXGBE_DCA
1088 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1089 ixgbe_update_rx_dca(adapter, rx_ring);
1090 #endif
1091 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1092 enable_mask |= rx_ring->v_idx;
1093 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1094 r_idx + 1);
1095 }
1096
1097 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1098 rx_ring = &(adapter->rx_ring[r_idx]);
1099 /* If all Rx work done, exit the polling mode */
1100 if (work_done < budget) {
1101 napi_complete(napi);
1102 if (adapter->itr_setting & 3)
1103 ixgbe_set_itr_msix(q_vector);
1104 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1105 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, enable_mask);
1106 return 0;
1107 }
1108
1109 return work_done;
1110 }
1111 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1112 int r_idx)
1113 {
1114 a->q_vector[v_idx].adapter = a;
1115 set_bit(r_idx, a->q_vector[v_idx].rxr_idx);
1116 a->q_vector[v_idx].rxr_count++;
1117 a->rx_ring[r_idx].v_idx = 1 << v_idx;
1118 }
1119
1120 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1121 int r_idx)
1122 {
1123 a->q_vector[v_idx].adapter = a;
1124 set_bit(r_idx, a->q_vector[v_idx].txr_idx);
1125 a->q_vector[v_idx].txr_count++;
1126 a->tx_ring[r_idx].v_idx = 1 << v_idx;
1127 }
1128
1129 /**
1130 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1131 * @adapter: board private structure to initialize
1132 * @vectors: allotted vector count for descriptor rings
1133 *
1134 * This function maps descriptor rings to the queue-specific vectors
1135 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1136 * one vector per ring/queue, but on a constrained vector budget, we
1137 * group the rings as "efficiently" as possible. You would add new
1138 * mapping configurations in here.
1139 **/
1140 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1141 int vectors)
1142 {
1143 int v_start = 0;
1144 int rxr_idx = 0, txr_idx = 0;
1145 int rxr_remaining = adapter->num_rx_queues;
1146 int txr_remaining = adapter->num_tx_queues;
1147 int i, j;
1148 int rqpv, tqpv;
1149 int err = 0;
1150
1151 /* No mapping required if MSI-X is disabled. */
1152 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1153 goto out;
1154
1155 /*
1156 * The ideal configuration...
1157 * We have enough vectors to map one per queue.
1158 */
1159 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1160 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1161 map_vector_to_rxq(adapter, v_start, rxr_idx);
1162
1163 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1164 map_vector_to_txq(adapter, v_start, txr_idx);
1165
1166 goto out;
1167 }
1168
1169 /*
1170 * If we don't have enough vectors for a 1-to-1
1171 * mapping, we'll have to group them so there are
1172 * multiple queues per vector.
1173 */
1174 /* Re-adjusting *qpv takes care of the remainder. */
1175 for (i = v_start; i < vectors; i++) {
1176 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1177 for (j = 0; j < rqpv; j++) {
1178 map_vector_to_rxq(adapter, i, rxr_idx);
1179 rxr_idx++;
1180 rxr_remaining--;
1181 }
1182 }
1183 for (i = v_start; i < vectors; i++) {
1184 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1185 for (j = 0; j < tqpv; j++) {
1186 map_vector_to_txq(adapter, i, txr_idx);
1187 txr_idx++;
1188 txr_remaining--;
1189 }
1190 }
1191
1192 out:
1193 return err;
1194 }
1195
1196 /**
1197 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1198 * @adapter: board private structure
1199 *
1200 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1201 * interrupts from the kernel.
1202 **/
1203 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1204 {
1205 struct net_device *netdev = adapter->netdev;
1206 irqreturn_t (*handler)(int, void *);
1207 int i, vector, q_vectors, err;
1208 int ri=0, ti=0;
1209
1210 /* Decrement for Other and TCP Timer vectors */
1211 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1212
1213 /* Map the Tx/Rx rings to the vectors we were allotted. */
1214 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1215 if (err)
1216 goto out;
1217
1218 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1219 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1220 &ixgbe_msix_clean_many)
1221 for (vector = 0; vector < q_vectors; vector++) {
1222 handler = SET_HANDLER(&adapter->q_vector[vector]);
1223
1224 if(handler == &ixgbe_msix_clean_rx) {
1225 sprintf(adapter->name[vector], "%s-%s-%d",
1226 netdev->name, "rx", ri++);
1227 }
1228 else if(handler == &ixgbe_msix_clean_tx) {
1229 sprintf(adapter->name[vector], "%s-%s-%d",
1230 netdev->name, "tx", ti++);
1231 }
1232 else
1233 sprintf(adapter->name[vector], "%s-%s-%d",
1234 netdev->name, "TxRx", vector);
1235
1236 err = request_irq(adapter->msix_entries[vector].vector,
1237 handler, 0, adapter->name[vector],
1238 &(adapter->q_vector[vector]));
1239 if (err) {
1240 DPRINTK(PROBE, ERR,
1241 "request_irq failed for MSIX interrupt "
1242 "Error: %d\n", err);
1243 goto free_queue_irqs;
1244 }
1245 }
1246
1247 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1248 err = request_irq(adapter->msix_entries[vector].vector,
1249 &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1250 if (err) {
1251 DPRINTK(PROBE, ERR,
1252 "request_irq for msix_lsc failed: %d\n", err);
1253 goto free_queue_irqs;
1254 }
1255
1256 return 0;
1257
1258 free_queue_irqs:
1259 for (i = vector - 1; i >= 0; i--)
1260 free_irq(adapter->msix_entries[--vector].vector,
1261 &(adapter->q_vector[i]));
1262 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1263 pci_disable_msix(adapter->pdev);
1264 kfree(adapter->msix_entries);
1265 adapter->msix_entries = NULL;
1266 out:
1267 return err;
1268 }
1269
1270 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1271 {
1272 struct ixgbe_hw *hw = &adapter->hw;
1273 struct ixgbe_q_vector *q_vector = adapter->q_vector;
1274 u8 current_itr;
1275 u32 new_itr = q_vector->eitr;
1276 struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1277 struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1278
1279 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1280 q_vector->tx_itr,
1281 tx_ring->total_packets,
1282 tx_ring->total_bytes);
1283 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1284 q_vector->rx_itr,
1285 rx_ring->total_packets,
1286 rx_ring->total_bytes);
1287
1288 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1289
1290 switch (current_itr) {
1291 /* counts and packets in update_itr are dependent on these numbers */
1292 case lowest_latency:
1293 new_itr = 100000;
1294 break;
1295 case low_latency:
1296 new_itr = 20000; /* aka hwitr = ~200 */
1297 break;
1298 case bulk_latency:
1299 new_itr = 8000;
1300 break;
1301 default:
1302 break;
1303 }
1304
1305 if (new_itr != q_vector->eitr) {
1306 u32 itr_reg;
1307 /* do an exponential smoothing */
1308 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1309 q_vector->eitr = new_itr;
1310 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
1311 /* must write high and low 16 bits to reset counter */
1312 IXGBE_WRITE_REG(hw, IXGBE_EITR(0), itr_reg | (itr_reg)<<16);
1313 }
1314
1315 return;
1316 }
1317
1318 /**
1319 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1320 * @adapter: board private structure
1321 **/
1322 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1323 {
1324 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1325 IXGBE_WRITE_FLUSH(&adapter->hw);
1326 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1327 int i;
1328 for (i = 0; i < adapter->num_msix_vectors; i++)
1329 synchronize_irq(adapter->msix_entries[i].vector);
1330 } else {
1331 synchronize_irq(adapter->pdev->irq);
1332 }
1333 }
1334
1335 /**
1336 * ixgbe_irq_enable - Enable default interrupt generation settings
1337 * @adapter: board private structure
1338 **/
1339 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1340 {
1341 u32 mask;
1342 mask = IXGBE_EIMS_ENABLE_MASK;
1343 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1344 mask |= IXGBE_EIMS_GPI_SDP1;
1345 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1346 IXGBE_WRITE_FLUSH(&adapter->hw);
1347 }
1348
1349 /**
1350 * ixgbe_intr - legacy mode Interrupt Handler
1351 * @irq: interrupt number
1352 * @data: pointer to a network interface device structure
1353 **/
1354 static irqreturn_t ixgbe_intr(int irq, void *data)
1355 {
1356 struct net_device *netdev = data;
1357 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1358 struct ixgbe_hw *hw = &adapter->hw;
1359 u32 eicr;
1360
1361 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1362 * therefore no explict interrupt disable is necessary */
1363 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1364 if (!eicr) {
1365 /* shared interrupt alert!
1366 * make sure interrupts are enabled because the read will
1367 * have disabled interrupts due to EIAM */
1368 ixgbe_irq_enable(adapter);
1369 return IRQ_NONE; /* Not our interrupt */
1370 }
1371
1372 if (eicr & IXGBE_EICR_LSC)
1373 ixgbe_check_lsc(adapter);
1374
1375 ixgbe_check_fan_failure(adapter, eicr);
1376
1377 if (napi_schedule_prep(&adapter->q_vector[0].napi)) {
1378 adapter->tx_ring[0].total_packets = 0;
1379 adapter->tx_ring[0].total_bytes = 0;
1380 adapter->rx_ring[0].total_packets = 0;
1381 adapter->rx_ring[0].total_bytes = 0;
1382 /* would disable interrupts here but EIAM disabled it */
1383 __napi_schedule(&adapter->q_vector[0].napi);
1384 }
1385
1386 return IRQ_HANDLED;
1387 }
1388
1389 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1390 {
1391 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1392
1393 for (i = 0; i < q_vectors; i++) {
1394 struct ixgbe_q_vector *q_vector = &adapter->q_vector[i];
1395 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1396 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1397 q_vector->rxr_count = 0;
1398 q_vector->txr_count = 0;
1399 }
1400 }
1401
1402 /**
1403 * ixgbe_request_irq - initialize interrupts
1404 * @adapter: board private structure
1405 *
1406 * Attempts to configure interrupts using the best available
1407 * capabilities of the hardware and kernel.
1408 **/
1409 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1410 {
1411 struct net_device *netdev = adapter->netdev;
1412 int err;
1413
1414 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1415 err = ixgbe_request_msix_irqs(adapter);
1416 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1417 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1418 netdev->name, netdev);
1419 } else {
1420 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1421 netdev->name, netdev);
1422 }
1423
1424 if (err)
1425 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1426
1427 return err;
1428 }
1429
1430 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1431 {
1432 struct net_device *netdev = adapter->netdev;
1433
1434 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1435 int i, q_vectors;
1436
1437 q_vectors = adapter->num_msix_vectors;
1438
1439 i = q_vectors - 1;
1440 free_irq(adapter->msix_entries[i].vector, netdev);
1441
1442 i--;
1443 for (; i >= 0; i--) {
1444 free_irq(adapter->msix_entries[i].vector,
1445 &(adapter->q_vector[i]));
1446 }
1447
1448 ixgbe_reset_q_vectors(adapter);
1449 } else {
1450 free_irq(adapter->pdev->irq, netdev);
1451 }
1452 }
1453
1454 /**
1455 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1456 *
1457 **/
1458 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1459 {
1460 struct ixgbe_hw *hw = &adapter->hw;
1461
1462 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1463 EITR_INTS_PER_SEC_TO_REG(adapter->eitr_param));
1464
1465 ixgbe_set_ivar(adapter, IXGBE_IVAR_RX_QUEUE(0), 0);
1466 ixgbe_set_ivar(adapter, IXGBE_IVAR_TX_QUEUE(0), 0);
1467
1468 map_vector_to_rxq(adapter, 0, 0);
1469 map_vector_to_txq(adapter, 0, 0);
1470
1471 DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1472 }
1473
1474 /**
1475 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1476 * @adapter: board private structure
1477 *
1478 * Configure the Tx unit of the MAC after a reset.
1479 **/
1480 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1481 {
1482 u64 tdba, tdwba;
1483 struct ixgbe_hw *hw = &adapter->hw;
1484 u32 i, j, tdlen, txctrl;
1485
1486 /* Setup the HW Tx Head and Tail descriptor pointers */
1487 for (i = 0; i < adapter->num_tx_queues; i++) {
1488 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1489 j = ring->reg_idx;
1490 tdba = ring->dma;
1491 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1492 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1493 (tdba & DMA_32BIT_MASK));
1494 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1495 tdwba = ring->dma +
1496 (ring->count * sizeof(union ixgbe_adv_tx_desc));
1497 tdwba |= IXGBE_TDWBAL_HEAD_WB_ENABLE;
1498 IXGBE_WRITE_REG(hw, IXGBE_TDWBAL(j), tdwba & DMA_32BIT_MASK);
1499 IXGBE_WRITE_REG(hw, IXGBE_TDWBAH(j), (tdwba >> 32));
1500 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1501 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1502 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1503 adapter->tx_ring[i].head = IXGBE_TDH(j);
1504 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1505 /* Disable Tx Head Writeback RO bit, since this hoses
1506 * bookkeeping if things aren't delivered in order.
1507 */
1508 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1509 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1510 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1511 }
1512 }
1513
1514 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1515
1516 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter, int index)
1517 {
1518 struct ixgbe_ring *rx_ring;
1519 u32 srrctl;
1520 int queue0;
1521 unsigned long mask;
1522
1523 /* program one srrctl register per VMDq index */
1524 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
1525 long shift, len;
1526 mask = (unsigned long) adapter->ring_feature[RING_F_RSS].mask;
1527 len = sizeof(adapter->ring_feature[RING_F_VMDQ].mask) * 8;
1528 shift = find_first_bit(&mask, len);
1529 queue0 = index & mask;
1530 index = (index & mask) >> shift;
1531 /* program one srrctl per RSS queue since RDRXCTL.MVMEN is enabled */
1532 } else {
1533 mask = (unsigned long) adapter->ring_feature[RING_F_RSS].mask;
1534 queue0 = index & mask;
1535 index = index & mask;
1536 }
1537
1538 rx_ring = &adapter->rx_ring[queue0];
1539
1540 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
1541
1542 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1543 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1544
1545 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1546 srrctl |= IXGBE_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1547 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1548 srrctl |= ((IXGBE_RX_HDR_SIZE <<
1549 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1550 IXGBE_SRRCTL_BSIZEHDR_MASK);
1551 } else {
1552 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1553
1554 if (rx_ring->rx_buf_len == MAXIMUM_ETHERNET_VLAN_SIZE)
1555 srrctl |= IXGBE_RXBUFFER_2048 >>
1556 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1557 else
1558 srrctl |= rx_ring->rx_buf_len >>
1559 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1560 }
1561 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
1562 }
1563
1564 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1565 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1566
1567 /**
1568 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
1569 * @adapter: board private structure
1570 *
1571 * Configure the Rx unit of the MAC after a reset.
1572 **/
1573 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
1574 {
1575 u64 rdba;
1576 struct ixgbe_hw *hw = &adapter->hw;
1577 struct net_device *netdev = adapter->netdev;
1578 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1579 int i, j;
1580 u32 rdlen, rxctrl, rxcsum;
1581 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
1582 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
1583 0x6A3E67EA, 0x14364D17, 0x3BED200D};
1584 u32 fctrl, hlreg0;
1585 u32 pages;
1586 u32 reta = 0, mrqc;
1587 u32 rdrxctl;
1588 int rx_buf_len;
1589
1590 /* Decide whether to use packet split mode or not */
1591 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
1592
1593 /* Set the RX buffer length according to the mode */
1594 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1595 rx_buf_len = IXGBE_RX_HDR_SIZE;
1596 } else {
1597 if (netdev->mtu <= ETH_DATA_LEN)
1598 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1599 else
1600 rx_buf_len = ALIGN(max_frame, 1024);
1601 }
1602
1603 fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
1604 fctrl |= IXGBE_FCTRL_BAM;
1605 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
1606 IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
1607
1608 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
1609 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1610 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
1611 else
1612 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
1613 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
1614
1615 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1616
1617 rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1618 /* disable receives while setting up the descriptors */
1619 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
1620 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
1621
1622 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1623 * the Base and Length of the Rx Descriptor Ring */
1624 for (i = 0; i < adapter->num_rx_queues; i++) {
1625 rdba = adapter->rx_ring[i].dma;
1626 j = adapter->rx_ring[i].reg_idx;
1627 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_32BIT_MASK));
1628 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
1629 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
1630 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
1631 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
1632 adapter->rx_ring[i].head = IXGBE_RDH(j);
1633 adapter->rx_ring[i].tail = IXGBE_RDT(j);
1634 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1635
1636 ixgbe_configure_srrctl(adapter, j);
1637 }
1638
1639 /*
1640 * For VMDq support of different descriptor types or
1641 * buffer sizes through the use of multiple SRRCTL
1642 * registers, RDRXCTL.MVMEN must be set to 1
1643 *
1644 * also, the manual doesn't mention it clearly but DCA hints
1645 * will only use queue 0's tags unless this bit is set. Side
1646 * effects of setting this bit are only that SRRCTL must be
1647 * fully programmed [0..15]
1648 */
1649 if (adapter->flags &
1650 (IXGBE_FLAG_RSS_ENABLED | IXGBE_FLAG_VMDQ_ENABLED)) {
1651 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
1652 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
1653 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
1654 }
1655
1656 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
1657 /* Fill out redirection table */
1658 for (i = 0, j = 0; i < 128; i++, j++) {
1659 if (j == adapter->ring_feature[RING_F_RSS].indices)
1660 j = 0;
1661 /* reta = 4-byte sliding window of
1662 * 0x00..(indices-1)(indices-1)00..etc. */
1663 reta = (reta << 8) | (j * 0x11);
1664 if ((i & 3) == 3)
1665 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
1666 }
1667
1668 /* Fill out hash function seeds */
1669 for (i = 0; i < 10; i++)
1670 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
1671
1672 mrqc = IXGBE_MRQC_RSSEN
1673 /* Perform hash on these packet types */
1674 | IXGBE_MRQC_RSS_FIELD_IPV4
1675 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
1676 | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
1677 | IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP
1678 | IXGBE_MRQC_RSS_FIELD_IPV6_EX
1679 | IXGBE_MRQC_RSS_FIELD_IPV6
1680 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
1681 | IXGBE_MRQC_RSS_FIELD_IPV6_UDP
1682 | IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
1683 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
1684 }
1685
1686 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
1687
1688 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
1689 adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
1690 /* Disable indicating checksum in descriptor, enables
1691 * RSS hash */
1692 rxcsum |= IXGBE_RXCSUM_PCSD;
1693 }
1694 if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
1695 /* Enable IPv4 payload checksum for UDP fragments
1696 * if PCSD is not set */
1697 rxcsum |= IXGBE_RXCSUM_IPPCSE;
1698 }
1699
1700 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
1701 }
1702
1703 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1704 {
1705 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1706 struct ixgbe_hw *hw = &adapter->hw;
1707
1708 /* add VID to filter table */
1709 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
1710 }
1711
1712 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1713 {
1714 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1715 struct ixgbe_hw *hw = &adapter->hw;
1716
1717 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1718 ixgbe_irq_disable(adapter);
1719
1720 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1721
1722 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1723 ixgbe_irq_enable(adapter);
1724
1725 /* remove VID from filter table */
1726 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
1727 }
1728
1729 static void ixgbe_vlan_rx_register(struct net_device *netdev,
1730 struct vlan_group *grp)
1731 {
1732 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1733 u32 ctrl;
1734
1735 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1736 ixgbe_irq_disable(adapter);
1737 adapter->vlgrp = grp;
1738
1739 /*
1740 * For a DCB driver, always enable VLAN tag stripping so we can
1741 * still receive traffic from a DCB-enabled host even if we're
1742 * not in DCB mode.
1743 */
1744 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
1745 ctrl |= IXGBE_VLNCTRL_VME;
1746 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
1747 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
1748 ixgbe_vlan_rx_add_vid(netdev, 0);
1749
1750 if (grp) {
1751 /* enable VLAN tag insert/strip */
1752 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
1753 ctrl |= IXGBE_VLNCTRL_VME;
1754 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
1755 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
1756 }
1757
1758 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1759 ixgbe_irq_enable(adapter);
1760 }
1761
1762 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
1763 {
1764 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1765
1766 if (adapter->vlgrp) {
1767 u16 vid;
1768 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1769 if (!vlan_group_get_device(adapter->vlgrp, vid))
1770 continue;
1771 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
1772 }
1773 }
1774 }
1775
1776 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
1777 {
1778 struct dev_mc_list *mc_ptr;
1779 u8 *addr = *mc_addr_ptr;
1780 *vmdq = 0;
1781
1782 mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
1783 if (mc_ptr->next)
1784 *mc_addr_ptr = mc_ptr->next->dmi_addr;
1785 else
1786 *mc_addr_ptr = NULL;
1787
1788 return addr;
1789 }
1790
1791 /**
1792 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
1793 * @netdev: network interface device structure
1794 *
1795 * The set_rx_method entry point is called whenever the unicast/multicast
1796 * address list or the network interface flags are updated. This routine is
1797 * responsible for configuring the hardware for proper unicast, multicast and
1798 * promiscuous mode.
1799 **/
1800 static void ixgbe_set_rx_mode(struct net_device *netdev)
1801 {
1802 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1803 struct ixgbe_hw *hw = &adapter->hw;
1804 u32 fctrl, vlnctrl;
1805 u8 *addr_list = NULL;
1806 int addr_count = 0;
1807
1808 /* Check for Promiscuous and All Multicast modes */
1809
1810 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
1811 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
1812
1813 if (netdev->flags & IFF_PROMISC) {
1814 hw->addr_ctrl.user_set_promisc = 1;
1815 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
1816 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
1817 } else {
1818 if (netdev->flags & IFF_ALLMULTI) {
1819 fctrl |= IXGBE_FCTRL_MPE;
1820 fctrl &= ~IXGBE_FCTRL_UPE;
1821 } else {
1822 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
1823 }
1824 vlnctrl |= IXGBE_VLNCTRL_VFE;
1825 hw->addr_ctrl.user_set_promisc = 0;
1826 }
1827
1828 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
1829 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
1830
1831 /* reprogram secondary unicast list */
1832 addr_count = netdev->uc_count;
1833 if (addr_count)
1834 addr_list = netdev->uc_list->dmi_addr;
1835 hw->mac.ops.update_uc_addr_list(hw, addr_list, addr_count,
1836 ixgbe_addr_list_itr);
1837
1838 /* reprogram multicast list */
1839 addr_count = netdev->mc_count;
1840 if (addr_count)
1841 addr_list = netdev->mc_list->dmi_addr;
1842 hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
1843 ixgbe_addr_list_itr);
1844 }
1845
1846 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
1847 {
1848 int q_idx;
1849 struct ixgbe_q_vector *q_vector;
1850 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1851
1852 /* legacy and MSI only use one vector */
1853 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1854 q_vectors = 1;
1855
1856 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1857 struct napi_struct *napi;
1858 q_vector = &adapter->q_vector[q_idx];
1859 if (!q_vector->rxr_count)
1860 continue;
1861 napi = &q_vector->napi;
1862 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) &&
1863 (q_vector->rxr_count > 1))
1864 napi->poll = &ixgbe_clean_rxonly_many;
1865
1866 napi_enable(napi);
1867 }
1868 }
1869
1870 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
1871 {
1872 int q_idx;
1873 struct ixgbe_q_vector *q_vector;
1874 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1875
1876 /* legacy and MSI only use one vector */
1877 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1878 q_vectors = 1;
1879
1880 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1881 q_vector = &adapter->q_vector[q_idx];
1882 if (!q_vector->rxr_count)
1883 continue;
1884 napi_disable(&q_vector->napi);
1885 }
1886 }
1887
1888 #ifdef CONFIG_IXGBE_DCB
1889 /*
1890 * ixgbe_configure_dcb - Configure DCB hardware
1891 * @adapter: ixgbe adapter struct
1892 *
1893 * This is called by the driver on open to configure the DCB hardware.
1894 * This is also called by the gennetlink interface when reconfiguring
1895 * the DCB state.
1896 */
1897 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
1898 {
1899 struct ixgbe_hw *hw = &adapter->hw;
1900 u32 txdctl, vlnctrl;
1901 int i, j;
1902
1903 ixgbe_dcb_check_config(&adapter->dcb_cfg);
1904 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
1905 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
1906
1907 /* reconfigure the hardware */
1908 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
1909
1910 for (i = 0; i < adapter->num_tx_queues; i++) {
1911 j = adapter->tx_ring[i].reg_idx;
1912 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
1913 /* PThresh workaround for Tx hang with DFP enabled. */
1914 txdctl |= 32;
1915 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
1916 }
1917 /* Enable VLAN tag insert/strip */
1918 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
1919 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
1920 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
1921 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
1922 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
1923 }
1924
1925 #endif
1926 static void ixgbe_configure(struct ixgbe_adapter *adapter)
1927 {
1928 struct net_device *netdev = adapter->netdev;
1929 int i;
1930
1931 ixgbe_set_rx_mode(netdev);
1932
1933 ixgbe_restore_vlan(adapter);
1934 #ifdef CONFIG_IXGBE_DCB
1935 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
1936 netif_set_gso_max_size(netdev, 32768);
1937 ixgbe_configure_dcb(adapter);
1938 } else {
1939 netif_set_gso_max_size(netdev, 65536);
1940 }
1941 #else
1942 netif_set_gso_max_size(netdev, 65536);
1943 #endif
1944
1945 ixgbe_configure_tx(adapter);
1946 ixgbe_configure_rx(adapter);
1947 for (i = 0; i < adapter->num_rx_queues; i++)
1948 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
1949 (adapter->rx_ring[i].count - 1));
1950 }
1951
1952 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
1953 {
1954 struct net_device *netdev = adapter->netdev;
1955 struct ixgbe_hw *hw = &adapter->hw;
1956 int i, j = 0;
1957 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1958 u32 txdctl, rxdctl, mhadd;
1959 u32 gpie;
1960
1961 ixgbe_get_hw_control(adapter);
1962
1963 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
1964 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
1965 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1966 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
1967 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
1968 } else {
1969 /* MSI only */
1970 gpie = 0;
1971 }
1972 /* XXX: to interrupt immediately for EICS writes, enable this */
1973 /* gpie |= IXGBE_GPIE_EIMEN; */
1974 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
1975 }
1976
1977 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
1978 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
1979 * specifically only auto mask tx and rx interrupts */
1980 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
1981 }
1982
1983 /* Enable fan failure interrupt if media type is copper */
1984 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
1985 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
1986 gpie |= IXGBE_SDP1_GPIEN;
1987 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
1988 }
1989
1990 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
1991 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
1992 mhadd &= ~IXGBE_MHADD_MFS_MASK;
1993 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
1994
1995 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
1996 }
1997
1998 for (i = 0; i < adapter->num_tx_queues; i++) {
1999 j = adapter->tx_ring[i].reg_idx;
2000 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2001 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2002 txdctl |= (8 << 16);
2003 txdctl |= IXGBE_TXDCTL_ENABLE;
2004 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2005 }
2006
2007 for (i = 0; i < adapter->num_rx_queues; i++) {
2008 j = adapter->rx_ring[i].reg_idx;
2009 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2010 /* enable PTHRESH=32 descriptors (half the internal cache)
2011 * and HTHRESH=0 descriptors (to minimize latency on fetch),
2012 * this also removes a pesky rx_no_buffer_count increment */
2013 rxdctl |= 0x0020;
2014 rxdctl |= IXGBE_RXDCTL_ENABLE;
2015 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2016 }
2017 /* enable all receives */
2018 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2019 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2020 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxdctl);
2021
2022 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2023 ixgbe_configure_msix(adapter);
2024 else
2025 ixgbe_configure_msi_and_legacy(adapter);
2026
2027 ixgbe_napi_add_all(adapter);
2028
2029 clear_bit(__IXGBE_DOWN, &adapter->state);
2030 ixgbe_napi_enable_all(adapter);
2031
2032 /* clear any pending interrupts, may auto mask */
2033 IXGBE_READ_REG(hw, IXGBE_EICR);
2034
2035 ixgbe_irq_enable(adapter);
2036
2037 /* enable transmits */
2038 netif_tx_start_all_queues(netdev);
2039
2040 /* bring the link up in the watchdog, this could race with our first
2041 * link up interrupt but shouldn't be a problem */
2042 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2043 adapter->link_check_timeout = jiffies;
2044 mod_timer(&adapter->watchdog_timer, jiffies);
2045 return 0;
2046 }
2047
2048 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
2049 {
2050 WARN_ON(in_interrupt());
2051 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
2052 msleep(1);
2053 ixgbe_down(adapter);
2054 ixgbe_up(adapter);
2055 clear_bit(__IXGBE_RESETTING, &adapter->state);
2056 }
2057
2058 int ixgbe_up(struct ixgbe_adapter *adapter)
2059 {
2060 /* hardware has been reset, we need to reload some things */
2061 ixgbe_configure(adapter);
2062
2063 return ixgbe_up_complete(adapter);
2064 }
2065
2066 void ixgbe_reset(struct ixgbe_adapter *adapter)
2067 {
2068 struct ixgbe_hw *hw = &adapter->hw;
2069 if (hw->mac.ops.init_hw(hw))
2070 dev_err(&adapter->pdev->dev, "Hardware Error\n");
2071
2072 /* reprogram the RAR[0] in case user changed it. */
2073 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2074
2075 }
2076
2077 /**
2078 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2079 * @adapter: board private structure
2080 * @rx_ring: ring to free buffers from
2081 **/
2082 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2083 struct ixgbe_ring *rx_ring)
2084 {
2085 struct pci_dev *pdev = adapter->pdev;
2086 unsigned long size;
2087 unsigned int i;
2088
2089 /* Free all the Rx ring sk_buffs */
2090
2091 for (i = 0; i < rx_ring->count; i++) {
2092 struct ixgbe_rx_buffer *rx_buffer_info;
2093
2094 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2095 if (rx_buffer_info->dma) {
2096 pci_unmap_single(pdev, rx_buffer_info->dma,
2097 rx_ring->rx_buf_len,
2098 PCI_DMA_FROMDEVICE);
2099 rx_buffer_info->dma = 0;
2100 }
2101 if (rx_buffer_info->skb) {
2102 dev_kfree_skb(rx_buffer_info->skb);
2103 rx_buffer_info->skb = NULL;
2104 }
2105 if (!rx_buffer_info->page)
2106 continue;
2107 pci_unmap_page(pdev, rx_buffer_info->page_dma, PAGE_SIZE / 2,
2108 PCI_DMA_FROMDEVICE);
2109 rx_buffer_info->page_dma = 0;
2110 put_page(rx_buffer_info->page);
2111 rx_buffer_info->page = NULL;
2112 rx_buffer_info->page_offset = 0;
2113 }
2114
2115 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2116 memset(rx_ring->rx_buffer_info, 0, size);
2117
2118 /* Zero out the descriptor ring */
2119 memset(rx_ring->desc, 0, rx_ring->size);
2120
2121 rx_ring->next_to_clean = 0;
2122 rx_ring->next_to_use = 0;
2123
2124 writel(0, adapter->hw.hw_addr + rx_ring->head);
2125 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2126 }
2127
2128 /**
2129 * ixgbe_clean_tx_ring - Free Tx Buffers
2130 * @adapter: board private structure
2131 * @tx_ring: ring to be cleaned
2132 **/
2133 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2134 struct ixgbe_ring *tx_ring)
2135 {
2136 struct ixgbe_tx_buffer *tx_buffer_info;
2137 unsigned long size;
2138 unsigned int i;
2139
2140 /* Free all the Tx ring sk_buffs */
2141
2142 for (i = 0; i < tx_ring->count; i++) {
2143 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2144 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2145 }
2146
2147 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2148 memset(tx_ring->tx_buffer_info, 0, size);
2149
2150 /* Zero out the descriptor ring */
2151 memset(tx_ring->desc, 0, tx_ring->size);
2152
2153 tx_ring->next_to_use = 0;
2154 tx_ring->next_to_clean = 0;
2155
2156 writel(0, adapter->hw.hw_addr + tx_ring->head);
2157 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2158 }
2159
2160 /**
2161 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2162 * @adapter: board private structure
2163 **/
2164 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2165 {
2166 int i;
2167
2168 for (i = 0; i < adapter->num_rx_queues; i++)
2169 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2170 }
2171
2172 /**
2173 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2174 * @adapter: board private structure
2175 **/
2176 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2177 {
2178 int i;
2179
2180 for (i = 0; i < adapter->num_tx_queues; i++)
2181 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2182 }
2183
2184 void ixgbe_down(struct ixgbe_adapter *adapter)
2185 {
2186 struct net_device *netdev = adapter->netdev;
2187 struct ixgbe_hw *hw = &adapter->hw;
2188 u32 rxctrl;
2189 u32 txdctl;
2190 int i, j;
2191
2192 /* signal that we are down to the interrupt handler */
2193 set_bit(__IXGBE_DOWN, &adapter->state);
2194
2195 /* disable receives */
2196 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2197 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2198
2199 netif_tx_disable(netdev);
2200
2201 IXGBE_WRITE_FLUSH(hw);
2202 msleep(10);
2203
2204 netif_tx_stop_all_queues(netdev);
2205
2206 ixgbe_irq_disable(adapter);
2207
2208 ixgbe_napi_disable_all(adapter);
2209
2210 del_timer_sync(&adapter->watchdog_timer);
2211 cancel_work_sync(&adapter->watchdog_task);
2212
2213 /* disable transmits in the hardware now that interrupts are off */
2214 for (i = 0; i < adapter->num_tx_queues; i++) {
2215 j = adapter->tx_ring[i].reg_idx;
2216 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2217 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
2218 (txdctl & ~IXGBE_TXDCTL_ENABLE));
2219 }
2220
2221 netif_carrier_off(netdev);
2222
2223 #ifdef CONFIG_IXGBE_DCA
2224 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2225 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
2226 dca_remove_requester(&adapter->pdev->dev);
2227 }
2228
2229 #endif
2230 if (!pci_channel_offline(adapter->pdev))
2231 ixgbe_reset(adapter);
2232 ixgbe_clean_all_tx_rings(adapter);
2233 ixgbe_clean_all_rx_rings(adapter);
2234
2235 #ifdef CONFIG_IXGBE_DCA
2236 /* since we reset the hardware DCA settings were cleared */
2237 if (dca_add_requester(&adapter->pdev->dev) == 0) {
2238 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
2239 /* always use CB2 mode, difference is masked
2240 * in the CB driver */
2241 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
2242 ixgbe_setup_dca(adapter);
2243 }
2244 #endif
2245 }
2246
2247 /**
2248 * ixgbe_poll - NAPI Rx polling callback
2249 * @napi: structure for representing this polling device
2250 * @budget: how many packets driver is allowed to clean
2251 *
2252 * This function is used for legacy and MSI, NAPI mode
2253 **/
2254 static int ixgbe_poll(struct napi_struct *napi, int budget)
2255 {
2256 struct ixgbe_q_vector *q_vector = container_of(napi,
2257 struct ixgbe_q_vector, napi);
2258 struct ixgbe_adapter *adapter = q_vector->adapter;
2259 int tx_cleaned, work_done = 0;
2260
2261 #ifdef CONFIG_IXGBE_DCA
2262 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2263 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
2264 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
2265 }
2266 #endif
2267
2268 tx_cleaned = ixgbe_clean_tx_irq(adapter, adapter->tx_ring);
2269 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring, &work_done, budget);
2270
2271 if (tx_cleaned)
2272 work_done = budget;
2273
2274 /* If budget not fully consumed, exit the polling mode */
2275 if (work_done < budget) {
2276 napi_complete(napi);
2277 if (adapter->itr_setting & 3)
2278 ixgbe_set_itr(adapter);
2279 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2280 ixgbe_irq_enable(adapter);
2281 }
2282 return work_done;
2283 }
2284
2285 /**
2286 * ixgbe_tx_timeout - Respond to a Tx Hang
2287 * @netdev: network interface device structure
2288 **/
2289 static void ixgbe_tx_timeout(struct net_device *netdev)
2290 {
2291 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2292
2293 /* Do the reset outside of interrupt context */
2294 schedule_work(&adapter->reset_task);
2295 }
2296
2297 static void ixgbe_reset_task(struct work_struct *work)
2298 {
2299 struct ixgbe_adapter *adapter;
2300 adapter = container_of(work, struct ixgbe_adapter, reset_task);
2301
2302 /* If we're already down or resetting, just bail */
2303 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
2304 test_bit(__IXGBE_RESETTING, &adapter->state))
2305 return;
2306
2307 adapter->tx_timeout_count++;
2308
2309 ixgbe_reinit_locked(adapter);
2310 }
2311
2312 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
2313 {
2314 int nrq = 1, ntq = 1;
2315 int feature_mask = 0, rss_i, rss_m;
2316 int dcb_i, dcb_m;
2317
2318 /* Number of supported queues */
2319 switch (adapter->hw.mac.type) {
2320 case ixgbe_mac_82598EB:
2321 dcb_i = adapter->ring_feature[RING_F_DCB].indices;
2322 dcb_m = 0;
2323 rss_i = adapter->ring_feature[RING_F_RSS].indices;
2324 rss_m = 0;
2325 feature_mask |= IXGBE_FLAG_RSS_ENABLED;
2326 feature_mask |= IXGBE_FLAG_DCB_ENABLED;
2327
2328 switch (adapter->flags & feature_mask) {
2329 case (IXGBE_FLAG_RSS_ENABLED | IXGBE_FLAG_DCB_ENABLED):
2330 dcb_m = 0x7 << 3;
2331 rss_i = min(8, rss_i);
2332 rss_m = 0x7;
2333 nrq = dcb_i * rss_i;
2334 ntq = min(MAX_TX_QUEUES, dcb_i * rss_i);
2335 break;
2336 case (IXGBE_FLAG_DCB_ENABLED):
2337 dcb_m = 0x7 << 3;
2338 nrq = dcb_i;
2339 ntq = dcb_i;
2340 break;
2341 case (IXGBE_FLAG_RSS_ENABLED):
2342 rss_m = 0xF;
2343 nrq = rss_i;
2344 ntq = rss_i;
2345 break;
2346 case 0:
2347 default:
2348 dcb_i = 0;
2349 dcb_m = 0;
2350 rss_i = 0;
2351 rss_m = 0;
2352 nrq = 1;
2353 ntq = 1;
2354 break;
2355 }
2356
2357 /* Sanity check, we should never have zero queues */
2358 nrq = (nrq ?:1);
2359 ntq = (ntq ?:1);
2360
2361 adapter->ring_feature[RING_F_DCB].indices = dcb_i;
2362 adapter->ring_feature[RING_F_DCB].mask = dcb_m;
2363 adapter->ring_feature[RING_F_RSS].indices = rss_i;
2364 adapter->ring_feature[RING_F_RSS].mask = rss_m;
2365 break;
2366 default:
2367 nrq = 1;
2368 ntq = 1;
2369 break;
2370 }
2371
2372 adapter->num_rx_queues = nrq;
2373 adapter->num_tx_queues = ntq;
2374 }
2375
2376 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
2377 int vectors)
2378 {
2379 int err, vector_threshold;
2380
2381 /* We'll want at least 3 (vector_threshold):
2382 * 1) TxQ[0] Cleanup
2383 * 2) RxQ[0] Cleanup
2384 * 3) Other (Link Status Change, etc.)
2385 * 4) TCP Timer (optional)
2386 */
2387 vector_threshold = MIN_MSIX_COUNT;
2388
2389 /* The more we get, the more we will assign to Tx/Rx Cleanup
2390 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2391 * Right now, we simply care about how many we'll get; we'll
2392 * set them up later while requesting irq's.
2393 */
2394 while (vectors >= vector_threshold) {
2395 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
2396 vectors);
2397 if (!err) /* Success in acquiring all requested vectors. */
2398 break;
2399 else if (err < 0)
2400 vectors = 0; /* Nasty failure, quit now */
2401 else /* err == number of vectors we should try again with */
2402 vectors = err;
2403 }
2404
2405 if (vectors < vector_threshold) {
2406 /* Can't allocate enough MSI-X interrupts? Oh well.
2407 * This just means we'll go with either a single MSI
2408 * vector or fall back to legacy interrupts.
2409 */
2410 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
2411 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2412 kfree(adapter->msix_entries);
2413 adapter->msix_entries = NULL;
2414 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
2415 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2416 ixgbe_set_num_queues(adapter);
2417 } else {
2418 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
2419 adapter->num_msix_vectors = vectors;
2420 }
2421 }
2422
2423 /**
2424 * ixgbe_cache_ring_register - Descriptor ring to register mapping
2425 * @adapter: board private structure to initialize
2426 *
2427 * Once we know the feature-set enabled for the device, we'll cache
2428 * the register offset the descriptor ring is assigned to.
2429 **/
2430 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
2431 {
2432 int feature_mask = 0, rss_i;
2433 int i, txr_idx, rxr_idx;
2434 int dcb_i;
2435
2436 /* Number of supported queues */
2437 switch (adapter->hw.mac.type) {
2438 case ixgbe_mac_82598EB:
2439 dcb_i = adapter->ring_feature[RING_F_DCB].indices;
2440 rss_i = adapter->ring_feature[RING_F_RSS].indices;
2441 txr_idx = 0;
2442 rxr_idx = 0;
2443 feature_mask |= IXGBE_FLAG_DCB_ENABLED;
2444 feature_mask |= IXGBE_FLAG_RSS_ENABLED;
2445 switch (adapter->flags & feature_mask) {
2446 case (IXGBE_FLAG_RSS_ENABLED | IXGBE_FLAG_DCB_ENABLED):
2447 for (i = 0; i < dcb_i; i++) {
2448 int j;
2449 /* Rx first */
2450 for (j = 0; j < adapter->num_rx_queues; j++) {
2451 adapter->rx_ring[rxr_idx].reg_idx =
2452 i << 3 | j;
2453 rxr_idx++;
2454 }
2455 /* Tx now */
2456 for (j = 0; j < adapter->num_tx_queues; j++) {
2457 adapter->tx_ring[txr_idx].reg_idx =
2458 i << 2 | (j >> 1);
2459 if (j & 1)
2460 txr_idx++;
2461 }
2462 }
2463 case (IXGBE_FLAG_DCB_ENABLED):
2464 /* the number of queues is assumed to be symmetric */
2465 for (i = 0; i < dcb_i; i++) {
2466 adapter->rx_ring[i].reg_idx = i << 3;
2467 adapter->tx_ring[i].reg_idx = i << 2;
2468 }
2469 break;
2470 case (IXGBE_FLAG_RSS_ENABLED):
2471 for (i = 0; i < adapter->num_rx_queues; i++)
2472 adapter->rx_ring[i].reg_idx = i;
2473 for (i = 0; i < adapter->num_tx_queues; i++)
2474 adapter->tx_ring[i].reg_idx = i;
2475 break;
2476 case 0:
2477 default:
2478 break;
2479 }
2480 break;
2481 default:
2482 break;
2483 }
2484 }
2485
2486 /**
2487 * ixgbe_alloc_queues - Allocate memory for all rings
2488 * @adapter: board private structure to initialize
2489 *
2490 * We allocate one ring per queue at run-time since we don't know the
2491 * number of queues at compile-time.
2492 **/
2493 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
2494 {
2495 int i;
2496
2497 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
2498 sizeof(struct ixgbe_ring), GFP_KERNEL);
2499 if (!adapter->tx_ring)
2500 goto err_tx_ring_allocation;
2501
2502 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
2503 sizeof(struct ixgbe_ring), GFP_KERNEL);
2504 if (!adapter->rx_ring)
2505 goto err_rx_ring_allocation;
2506
2507 for (i = 0; i < adapter->num_tx_queues; i++) {
2508 adapter->tx_ring[i].count = adapter->tx_ring_count;
2509 adapter->tx_ring[i].queue_index = i;
2510 }
2511
2512 for (i = 0; i < adapter->num_rx_queues; i++) {
2513 adapter->rx_ring[i].count = adapter->rx_ring_count;
2514 adapter->rx_ring[i].queue_index = i;
2515 }
2516
2517 ixgbe_cache_ring_register(adapter);
2518
2519 return 0;
2520
2521 err_rx_ring_allocation:
2522 kfree(adapter->tx_ring);
2523 err_tx_ring_allocation:
2524 return -ENOMEM;
2525 }
2526
2527 /**
2528 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
2529 * @adapter: board private structure to initialize
2530 *
2531 * Attempt to configure the interrupts using the best available
2532 * capabilities of the hardware and the kernel.
2533 **/
2534 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
2535 {
2536 int err = 0;
2537 int vector, v_budget;
2538
2539 /*
2540 * It's easy to be greedy for MSI-X vectors, but it really
2541 * doesn't do us much good if we have a lot more vectors
2542 * than CPU's. So let's be conservative and only ask for
2543 * (roughly) twice the number of vectors as there are CPU's.
2544 */
2545 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
2546 (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
2547
2548 /*
2549 * At the same time, hardware can only support a maximum of
2550 * MAX_MSIX_COUNT vectors. With features such as RSS and VMDq,
2551 * we can easily reach upwards of 64 Rx descriptor queues and
2552 * 32 Tx queues. Thus, we cap it off in those rare cases where
2553 * the cpu count also exceeds our vector limit.
2554 */
2555 v_budget = min(v_budget, MAX_MSIX_COUNT);
2556
2557 /* A failure in MSI-X entry allocation isn't fatal, but it does
2558 * mean we disable MSI-X capabilities of the adapter. */
2559 adapter->msix_entries = kcalloc(v_budget,
2560 sizeof(struct msix_entry), GFP_KERNEL);
2561 if (!adapter->msix_entries) {
2562 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
2563 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2564 ixgbe_set_num_queues(adapter);
2565 kfree(adapter->tx_ring);
2566 kfree(adapter->rx_ring);
2567 err = ixgbe_alloc_queues(adapter);
2568 if (err) {
2569 DPRINTK(PROBE, ERR, "Unable to allocate memory "
2570 "for queues\n");
2571 goto out;
2572 }
2573
2574 goto try_msi;
2575 }
2576
2577 for (vector = 0; vector < v_budget; vector++)
2578 adapter->msix_entries[vector].entry = vector;
2579
2580 ixgbe_acquire_msix_vectors(adapter, v_budget);
2581
2582 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2583 goto out;
2584
2585 try_msi:
2586 err = pci_enable_msi(adapter->pdev);
2587 if (!err) {
2588 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
2589 } else {
2590 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
2591 "falling back to legacy. Error: %d\n", err);
2592 /* reset err */
2593 err = 0;
2594 }
2595
2596 out:
2597 /* Notify the stack of the (possibly) reduced Tx Queue count. */
2598 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
2599
2600 return err;
2601 }
2602
2603 void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
2604 {
2605 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2606 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2607 pci_disable_msix(adapter->pdev);
2608 kfree(adapter->msix_entries);
2609 adapter->msix_entries = NULL;
2610 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2611 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
2612 pci_disable_msi(adapter->pdev);
2613 }
2614 return;
2615 }
2616
2617 /**
2618 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
2619 * @adapter: board private structure to initialize
2620 *
2621 * We determine which interrupt scheme to use based on...
2622 * - Kernel support (MSI, MSI-X)
2623 * - which can be user-defined (via MODULE_PARAM)
2624 * - Hardware queue count (num_*_queues)
2625 * - defined by miscellaneous hardware support/features (RSS, etc.)
2626 **/
2627 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
2628 {
2629 int err;
2630
2631 /* Number of supported queues */
2632 ixgbe_set_num_queues(adapter);
2633
2634 err = ixgbe_alloc_queues(adapter);
2635 if (err) {
2636 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
2637 goto err_alloc_queues;
2638 }
2639
2640 err = ixgbe_set_interrupt_capability(adapter);
2641 if (err) {
2642 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
2643 goto err_set_interrupt;
2644 }
2645
2646 DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
2647 "Tx Queue count = %u\n",
2648 (adapter->num_rx_queues > 1) ? "Enabled" :
2649 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2650
2651 set_bit(__IXGBE_DOWN, &adapter->state);
2652
2653 return 0;
2654
2655 err_set_interrupt:
2656 kfree(adapter->tx_ring);
2657 kfree(adapter->rx_ring);
2658 err_alloc_queues:
2659 return err;
2660 }
2661
2662 /**
2663 * ixgbe_sfp_timer - worker thread to find a missing module
2664 * @data: pointer to our adapter struct
2665 **/
2666 static void ixgbe_sfp_timer(unsigned long data)
2667 {
2668 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
2669
2670 /* Do the sfp_timer outside of interrupt context due to the
2671 * delays that sfp+ detection requires
2672 */
2673 schedule_work(&adapter->sfp_task);
2674 }
2675
2676 /**
2677 * ixgbe_sfp_task - worker thread to find a missing module
2678 * @work: pointer to work_struct containing our data
2679 **/
2680 static void ixgbe_sfp_task(struct work_struct *work)
2681 {
2682 struct ixgbe_adapter *adapter = container_of(work,
2683 struct ixgbe_adapter,
2684 sfp_task);
2685 struct ixgbe_hw *hw = &adapter->hw;
2686
2687 if ((hw->phy.type == ixgbe_phy_nl) &&
2688 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
2689 s32 ret = hw->phy.ops.identify_sfp(hw);
2690 if (ret)
2691 goto reschedule;
2692 ret = hw->phy.ops.reset(hw);
2693 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
2694 DPRINTK(PROBE, ERR, "failed to initialize because an "
2695 "unsupported SFP+ module type was detected.\n"
2696 "Reload the driver after installing a "
2697 "supported module.\n");
2698 unregister_netdev(adapter->netdev);
2699 } else {
2700 DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
2701 hw->phy.sfp_type);
2702 }
2703 /* don't need this routine any more */
2704 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
2705 }
2706 return;
2707 reschedule:
2708 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
2709 mod_timer(&adapter->sfp_timer,
2710 round_jiffies(jiffies + (2 * HZ)));
2711 }
2712
2713 /**
2714 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
2715 * @adapter: board private structure to initialize
2716 *
2717 * ixgbe_sw_init initializes the Adapter private data structure.
2718 * Fields are initialized based on PCI device information and
2719 * OS network device settings (MTU size).
2720 **/
2721 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
2722 {
2723 struct ixgbe_hw *hw = &adapter->hw;
2724 struct pci_dev *pdev = adapter->pdev;
2725 unsigned int rss;
2726 #ifdef CONFIG_IXGBE_DCB
2727 int j;
2728 struct tc_configuration *tc;
2729 #endif
2730
2731 /* PCI config space info */
2732
2733 hw->vendor_id = pdev->vendor;
2734 hw->device_id = pdev->device;
2735 hw->revision_id = pdev->revision;
2736 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2737 hw->subsystem_device_id = pdev->subsystem_device;
2738
2739 /* Set capability flags */
2740 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
2741 adapter->ring_feature[RING_F_RSS].indices = rss;
2742 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
2743 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
2744
2745 #ifdef CONFIG_IXGBE_DCB
2746 /* Configure DCB traffic classes */
2747 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
2748 tc = &adapter->dcb_cfg.tc_config[j];
2749 tc->path[DCB_TX_CONFIG].bwg_id = 0;
2750 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
2751 tc->path[DCB_RX_CONFIG].bwg_id = 0;
2752 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
2753 tc->dcb_pfc = pfc_disabled;
2754 }
2755 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
2756 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
2757 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
2758 adapter->dcb_cfg.round_robin_enable = false;
2759 adapter->dcb_set_bitmap = 0x00;
2760 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
2761 adapter->ring_feature[RING_F_DCB].indices);
2762
2763 #endif
2764 if (hw->mac.ops.get_media_type &&
2765 (hw->mac.ops.get_media_type(hw) == ixgbe_media_type_copper))
2766 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
2767
2768 /* default flow control settings */
2769 hw->fc.original_type = ixgbe_fc_none;
2770 hw->fc.type = ixgbe_fc_none;
2771 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
2772 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
2773 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
2774 hw->fc.send_xon = true;
2775
2776 /* select 10G link by default */
2777 hw->mac.link_mode_select = IXGBE_AUTOC_LMS_10G_LINK_NO_AN;
2778
2779 /* enable itr by default in dynamic mode */
2780 adapter->itr_setting = 1;
2781 adapter->eitr_param = 20000;
2782
2783 /* set defaults for eitr in MegaBytes */
2784 adapter->eitr_low = 10;
2785 adapter->eitr_high = 20;
2786
2787 /* set default ring sizes */
2788 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
2789 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
2790
2791 /* initialize eeprom parameters */
2792 if (ixgbe_init_eeprom_params_generic(hw)) {
2793 dev_err(&pdev->dev, "EEPROM initialization failed\n");
2794 return -EIO;
2795 }
2796
2797 /* enable rx csum by default */
2798 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
2799
2800 set_bit(__IXGBE_DOWN, &adapter->state);
2801
2802 return 0;
2803 }
2804
2805 /**
2806 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
2807 * @adapter: board private structure
2808 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2809 *
2810 * Return 0 on success, negative on failure
2811 **/
2812 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
2813 struct ixgbe_ring *tx_ring)
2814 {
2815 struct pci_dev *pdev = adapter->pdev;
2816 int size;
2817
2818 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2819 tx_ring->tx_buffer_info = vmalloc(size);
2820 if (!tx_ring->tx_buffer_info)
2821 goto err;
2822 memset(tx_ring->tx_buffer_info, 0, size);
2823
2824 /* round up to nearest 4K */
2825 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc) +
2826 sizeof(u32);
2827 tx_ring->size = ALIGN(tx_ring->size, 4096);
2828
2829 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
2830 &tx_ring->dma);
2831 if (!tx_ring->desc)
2832 goto err;
2833
2834 tx_ring->next_to_use = 0;
2835 tx_ring->next_to_clean = 0;
2836 tx_ring->work_limit = tx_ring->count;
2837 return 0;
2838
2839 err:
2840 vfree(tx_ring->tx_buffer_info);
2841 tx_ring->tx_buffer_info = NULL;
2842 DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
2843 "descriptor ring\n");
2844 return -ENOMEM;
2845 }
2846
2847 /**
2848 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
2849 * @adapter: board private structure
2850 *
2851 * If this function returns with an error, then it's possible one or
2852 * more of the rings is populated (while the rest are not). It is the
2853 * callers duty to clean those orphaned rings.
2854 *
2855 * Return 0 on success, negative on failure
2856 **/
2857 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
2858 {
2859 int i, err = 0;
2860
2861 for (i = 0; i < adapter->num_tx_queues; i++) {
2862 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2863 if (!err)
2864 continue;
2865 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
2866 break;
2867 }
2868
2869 return err;
2870 }
2871
2872 /**
2873 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
2874 * @adapter: board private structure
2875 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2876 *
2877 * Returns 0 on success, negative on failure
2878 **/
2879 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
2880 struct ixgbe_ring *rx_ring)
2881 {
2882 struct pci_dev *pdev = adapter->pdev;
2883 int size;
2884
2885 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2886 rx_ring->rx_buffer_info = vmalloc(size);
2887 if (!rx_ring->rx_buffer_info) {
2888 DPRINTK(PROBE, ERR,
2889 "vmalloc allocation failed for the rx desc ring\n");
2890 goto alloc_failed;
2891 }
2892 memset(rx_ring->rx_buffer_info, 0, size);
2893
2894 /* Round up to nearest 4K */
2895 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2896 rx_ring->size = ALIGN(rx_ring->size, 4096);
2897
2898 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
2899
2900 if (!rx_ring->desc) {
2901 DPRINTK(PROBE, ERR,
2902 "Memory allocation failed for the rx desc ring\n");
2903 vfree(rx_ring->rx_buffer_info);
2904 goto alloc_failed;
2905 }
2906
2907 rx_ring->next_to_clean = 0;
2908 rx_ring->next_to_use = 0;
2909
2910 return 0;
2911
2912 alloc_failed:
2913 return -ENOMEM;
2914 }
2915
2916 /**
2917 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
2918 * @adapter: board private structure
2919 *
2920 * If this function returns with an error, then it's possible one or
2921 * more of the rings is populated (while the rest are not). It is the
2922 * callers duty to clean those orphaned rings.
2923 *
2924 * Return 0 on success, negative on failure
2925 **/
2926
2927 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
2928 {
2929 int i, err = 0;
2930
2931 for (i = 0; i < adapter->num_rx_queues; i++) {
2932 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2933 if (!err)
2934 continue;
2935 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
2936 break;
2937 }
2938
2939 return err;
2940 }
2941
2942 /**
2943 * ixgbe_free_tx_resources - Free Tx Resources per Queue
2944 * @adapter: board private structure
2945 * @tx_ring: Tx descriptor ring for a specific queue
2946 *
2947 * Free all transmit software resources
2948 **/
2949 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
2950 struct ixgbe_ring *tx_ring)
2951 {
2952 struct pci_dev *pdev = adapter->pdev;
2953
2954 ixgbe_clean_tx_ring(adapter, tx_ring);
2955
2956 vfree(tx_ring->tx_buffer_info);
2957 tx_ring->tx_buffer_info = NULL;
2958
2959 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2960
2961 tx_ring->desc = NULL;
2962 }
2963
2964 /**
2965 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
2966 * @adapter: board private structure
2967 *
2968 * Free all transmit software resources
2969 **/
2970 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
2971 {
2972 int i;
2973
2974 for (i = 0; i < adapter->num_tx_queues; i++)
2975 ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
2976 }
2977
2978 /**
2979 * ixgbe_free_rx_resources - Free Rx Resources
2980 * @adapter: board private structure
2981 * @rx_ring: ring to clean the resources from
2982 *
2983 * Free all receive software resources
2984 **/
2985 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
2986 struct ixgbe_ring *rx_ring)
2987 {
2988 struct pci_dev *pdev = adapter->pdev;
2989
2990 ixgbe_clean_rx_ring(adapter, rx_ring);
2991
2992 vfree(rx_ring->rx_buffer_info);
2993 rx_ring->rx_buffer_info = NULL;
2994
2995 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2996
2997 rx_ring->desc = NULL;
2998 }
2999
3000 /**
3001 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
3002 * @adapter: board private structure
3003 *
3004 * Free all receive software resources
3005 **/
3006 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
3007 {
3008 int i;
3009
3010 for (i = 0; i < adapter->num_rx_queues; i++)
3011 ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
3012 }
3013
3014 /**
3015 * ixgbe_change_mtu - Change the Maximum Transfer Unit
3016 * @netdev: network interface device structure
3017 * @new_mtu: new value for maximum frame size
3018 *
3019 * Returns 0 on success, negative on failure
3020 **/
3021 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
3022 {
3023 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3024 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3025
3026 /* MTU < 68 is an error and causes problems on some kernels */
3027 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
3028 return -EINVAL;
3029
3030 DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
3031 netdev->mtu, new_mtu);
3032 /* must set new MTU before calling down or up */
3033 netdev->mtu = new_mtu;
3034
3035 if (netif_running(netdev))
3036 ixgbe_reinit_locked(adapter);
3037
3038 return 0;
3039 }
3040
3041 /**
3042 * ixgbe_open - Called when a network interface is made active
3043 * @netdev: network interface device structure
3044 *
3045 * Returns 0 on success, negative value on failure
3046 *
3047 * The open entry point is called when a network interface is made
3048 * active by the system (IFF_UP). At this point all resources needed
3049 * for transmit and receive operations are allocated, the interrupt
3050 * handler is registered with the OS, the watchdog timer is started,
3051 * and the stack is notified that the interface is ready.
3052 **/
3053 static int ixgbe_open(struct net_device *netdev)
3054 {
3055 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3056 int err;
3057
3058 /* disallow open during test */
3059 if (test_bit(__IXGBE_TESTING, &adapter->state))
3060 return -EBUSY;
3061
3062 /* allocate transmit descriptors */
3063 err = ixgbe_setup_all_tx_resources(adapter);
3064 if (err)
3065 goto err_setup_tx;
3066
3067 /* allocate receive descriptors */
3068 err = ixgbe_setup_all_rx_resources(adapter);
3069 if (err)
3070 goto err_setup_rx;
3071
3072 ixgbe_configure(adapter);
3073
3074 err = ixgbe_request_irq(adapter);
3075 if (err)
3076 goto err_req_irq;
3077
3078 err = ixgbe_up_complete(adapter);
3079 if (err)
3080 goto err_up;
3081
3082 netif_tx_start_all_queues(netdev);
3083
3084 return 0;
3085
3086 err_up:
3087 ixgbe_release_hw_control(adapter);
3088 ixgbe_free_irq(adapter);
3089 err_req_irq:
3090 ixgbe_free_all_rx_resources(adapter);
3091 err_setup_rx:
3092 ixgbe_free_all_tx_resources(adapter);
3093 err_setup_tx:
3094 ixgbe_reset(adapter);
3095
3096 return err;
3097 }
3098
3099 /**
3100 * ixgbe_close - Disables a network interface
3101 * @netdev: network interface device structure
3102 *
3103 * Returns 0, this is not allowed to fail
3104 *
3105 * The close entry point is called when an interface is de-activated
3106 * by the OS. The hardware is still under the drivers control, but
3107 * needs to be disabled. A global MAC reset is issued to stop the
3108 * hardware, and all transmit and receive resources are freed.
3109 **/
3110 static int ixgbe_close(struct net_device *netdev)
3111 {
3112 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3113
3114 ixgbe_down(adapter);
3115 ixgbe_free_irq(adapter);
3116
3117 ixgbe_free_all_tx_resources(adapter);
3118 ixgbe_free_all_rx_resources(adapter);
3119
3120 ixgbe_release_hw_control(adapter);
3121
3122 return 0;
3123 }
3124
3125 /**
3126 * ixgbe_napi_add_all - prep napi structs for use
3127 * @adapter: private struct
3128 * helper function to napi_add each possible q_vector->napi
3129 */
3130 void ixgbe_napi_add_all(struct ixgbe_adapter *adapter)
3131 {
3132 int q_idx, q_vectors;
3133 struct net_device *netdev = adapter->netdev;
3134 int (*poll)(struct napi_struct *, int);
3135
3136 /* check if we already have our netdev->napi_list populated */
3137 if (&netdev->napi_list != netdev->napi_list.next)
3138 return;
3139
3140 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3141 poll = &ixgbe_clean_rxonly;
3142 /* Only enable as many vectors as we have rx queues. */
3143 q_vectors = adapter->num_rx_queues;
3144 } else {
3145 poll = &ixgbe_poll;
3146 /* only one q_vector for legacy modes */
3147 q_vectors = 1;
3148 }
3149
3150 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3151 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
3152 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
3153 }
3154 }
3155
3156 void ixgbe_napi_del_all(struct ixgbe_adapter *adapter)
3157 {
3158 int q_idx;
3159 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3160
3161 /* legacy and MSI only use one vector */
3162 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3163 q_vectors = 1;
3164
3165 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3166 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
3167 if (!q_vector->rxr_count)
3168 continue;
3169 netif_napi_del(&q_vector->napi);
3170 }
3171 }
3172
3173 #ifdef CONFIG_PM
3174 static int ixgbe_resume(struct pci_dev *pdev)
3175 {
3176 struct net_device *netdev = pci_get_drvdata(pdev);
3177 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3178 u32 err;
3179
3180 pci_set_power_state(pdev, PCI_D0);
3181 pci_restore_state(pdev);
3182 err = pci_enable_device(pdev);
3183 if (err) {
3184 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
3185 "suspend\n");
3186 return err;
3187 }
3188 pci_set_master(pdev);
3189
3190 pci_enable_wake(pdev, PCI_D3hot, 0);
3191 pci_enable_wake(pdev, PCI_D3cold, 0);
3192
3193 err = ixgbe_init_interrupt_scheme(adapter);
3194 if (err) {
3195 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
3196 "device\n");
3197 return err;
3198 }
3199
3200 ixgbe_napi_add_all(adapter);
3201 ixgbe_reset(adapter);
3202
3203 if (netif_running(netdev)) {
3204 err = ixgbe_open(adapter->netdev);
3205 if (err)
3206 return err;
3207 }
3208
3209 netif_device_attach(netdev);
3210
3211 return 0;
3212 }
3213
3214 #endif /* CONFIG_PM */
3215 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
3216 {
3217 struct net_device *netdev = pci_get_drvdata(pdev);
3218 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3219 #ifdef CONFIG_PM
3220 int retval = 0;
3221 #endif
3222
3223 netif_device_detach(netdev);
3224
3225 if (netif_running(netdev)) {
3226 ixgbe_down(adapter);
3227 ixgbe_free_irq(adapter);
3228 ixgbe_free_all_tx_resources(adapter);
3229 ixgbe_free_all_rx_resources(adapter);
3230 }
3231 ixgbe_reset_interrupt_capability(adapter);
3232 ixgbe_napi_del_all(adapter);
3233 INIT_LIST_HEAD(&netdev->napi_list);
3234 kfree(adapter->tx_ring);
3235 kfree(adapter->rx_ring);
3236
3237 #ifdef CONFIG_PM
3238 retval = pci_save_state(pdev);
3239 if (retval)
3240 return retval;
3241 #endif
3242
3243 pci_enable_wake(pdev, PCI_D3hot, 0);
3244 pci_enable_wake(pdev, PCI_D3cold, 0);
3245
3246 ixgbe_release_hw_control(adapter);
3247
3248 pci_disable_device(pdev);
3249
3250 pci_set_power_state(pdev, pci_choose_state(pdev, state));
3251
3252 return 0;
3253 }
3254
3255 static void ixgbe_shutdown(struct pci_dev *pdev)
3256 {
3257 ixgbe_suspend(pdev, PMSG_SUSPEND);
3258 }
3259
3260 /**
3261 * ixgbe_update_stats - Update the board statistics counters.
3262 * @adapter: board private structure
3263 **/
3264 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
3265 {
3266 struct ixgbe_hw *hw = &adapter->hw;
3267 u64 total_mpc = 0;
3268 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
3269
3270 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
3271 for (i = 0; i < 8; i++) {
3272 /* for packet buffers not used, the register should read 0 */
3273 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
3274 missed_rx += mpc;
3275 adapter->stats.mpc[i] += mpc;
3276 total_mpc += adapter->stats.mpc[i];
3277 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
3278 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
3279 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
3280 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
3281 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
3282 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
3283 IXGBE_PXONRXC(i));
3284 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
3285 IXGBE_PXONTXC(i));
3286 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
3287 IXGBE_PXOFFRXC(i));
3288 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
3289 IXGBE_PXOFFTXC(i));
3290 }
3291 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
3292 /* work around hardware counting issue */
3293 adapter->stats.gprc -= missed_rx;
3294
3295 /* 82598 hardware only has a 32 bit counter in the high register */
3296 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
3297 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
3298 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
3299 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
3300 adapter->stats.bprc += bprc;
3301 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
3302 adapter->stats.mprc -= bprc;
3303 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
3304 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
3305 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
3306 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
3307 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
3308 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
3309 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
3310 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
3311 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
3312 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
3313 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
3314 adapter->stats.lxontxc += lxon;
3315 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
3316 adapter->stats.lxofftxc += lxoff;
3317 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3318 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
3319 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
3320 /*
3321 * 82598 errata - tx of flow control packets is included in tx counters
3322 */
3323 xon_off_tot = lxon + lxoff;
3324 adapter->stats.gptc -= xon_off_tot;
3325 adapter->stats.mptc -= xon_off_tot;
3326 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
3327 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3328 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
3329 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
3330 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
3331 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
3332 adapter->stats.ptc64 -= xon_off_tot;
3333 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
3334 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
3335 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
3336 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
3337 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
3338 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
3339
3340 /* Fill out the OS statistics structure */
3341 adapter->net_stats.multicast = adapter->stats.mprc;
3342
3343 /* Rx Errors */
3344 adapter->net_stats.rx_errors = adapter->stats.crcerrs +
3345 adapter->stats.rlec;
3346 adapter->net_stats.rx_dropped = 0;
3347 adapter->net_stats.rx_length_errors = adapter->stats.rlec;
3348 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3349 adapter->net_stats.rx_missed_errors = total_mpc;
3350 }
3351
3352 /**
3353 * ixgbe_watchdog - Timer Call-back
3354 * @data: pointer to adapter cast into an unsigned long
3355 **/
3356 static void ixgbe_watchdog(unsigned long data)
3357 {
3358 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3359 struct ixgbe_hw *hw = &adapter->hw;
3360
3361 /* Do the watchdog outside of interrupt context due to the lovely
3362 * delays that some of the newer hardware requires */
3363 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
3364 /* Cause software interrupt to ensure rx rings are cleaned */
3365 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3366 u32 eics =
3367 (1 << (adapter->num_msix_vectors - NON_Q_VECTORS)) - 1;
3368 IXGBE_WRITE_REG(hw, IXGBE_EICS, eics);
3369 } else {
3370 /* For legacy and MSI interrupts don't set any bits that
3371 * are enabled for EIAM, because this operation would
3372 * set *both* EIMS and EICS for any bit in EIAM */
3373 IXGBE_WRITE_REG(hw, IXGBE_EICS,
3374 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
3375 }
3376 /* Reset the timer */
3377 mod_timer(&adapter->watchdog_timer,
3378 round_jiffies(jiffies + 2 * HZ));
3379 }
3380
3381 schedule_work(&adapter->watchdog_task);
3382 }
3383
3384 /**
3385 * ixgbe_watchdog_task - worker thread to bring link up
3386 * @work: pointer to work_struct containing our data
3387 **/
3388 static void ixgbe_watchdog_task(struct work_struct *work)
3389 {
3390 struct ixgbe_adapter *adapter = container_of(work,
3391 struct ixgbe_adapter,
3392 watchdog_task);
3393 struct net_device *netdev = adapter->netdev;
3394 struct ixgbe_hw *hw = &adapter->hw;
3395 u32 link_speed = adapter->link_speed;
3396 bool link_up = adapter->link_up;
3397
3398 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
3399
3400 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
3401 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3402 if (link_up ||
3403 time_after(jiffies, (adapter->link_check_timeout +
3404 IXGBE_TRY_LINK_TIMEOUT))) {
3405 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
3406 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
3407 }
3408 adapter->link_up = link_up;
3409 adapter->link_speed = link_speed;
3410 }
3411
3412 if (link_up) {
3413 if (!netif_carrier_ok(netdev)) {
3414 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3415 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
3416 #define FLOW_RX (frctl & IXGBE_FCTRL_RFCE)
3417 #define FLOW_TX (rmcs & IXGBE_RMCS_TFCE_802_3X)
3418 printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
3419 "Flow Control: %s\n",
3420 netdev->name,
3421 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
3422 "10 Gbps" :
3423 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
3424 "1 Gbps" : "unknown speed")),
3425 ((FLOW_RX && FLOW_TX) ? "RX/TX" :
3426 (FLOW_RX ? "RX" :
3427 (FLOW_TX ? "TX" : "None"))));
3428
3429 netif_carrier_on(netdev);
3430 } else {
3431 /* Force detection of hung controller */
3432 adapter->detect_tx_hung = true;
3433 }
3434 } else {
3435 adapter->link_up = false;
3436 adapter->link_speed = 0;
3437 if (netif_carrier_ok(netdev)) {
3438 printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
3439 netdev->name);
3440 netif_carrier_off(netdev);
3441 }
3442 }
3443
3444 ixgbe_update_stats(adapter);
3445 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
3446 }
3447
3448 static int ixgbe_tso(struct ixgbe_adapter *adapter,
3449 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
3450 u32 tx_flags, u8 *hdr_len)
3451 {
3452 struct ixgbe_adv_tx_context_desc *context_desc;
3453 unsigned int i;
3454 int err;
3455 struct ixgbe_tx_buffer *tx_buffer_info;
3456 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
3457 u32 mss_l4len_idx, l4len;
3458
3459 if (skb_is_gso(skb)) {
3460 if (skb_header_cloned(skb)) {
3461 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3462 if (err)
3463 return err;
3464 }
3465 l4len = tcp_hdrlen(skb);
3466 *hdr_len += l4len;
3467
3468 if (skb->protocol == htons(ETH_P_IP)) {
3469 struct iphdr *iph = ip_hdr(skb);
3470 iph->tot_len = 0;
3471 iph->check = 0;
3472 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3473 iph->daddr, 0,
3474 IPPROTO_TCP,
3475 0);
3476 adapter->hw_tso_ctxt++;
3477 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3478 ipv6_hdr(skb)->payload_len = 0;
3479 tcp_hdr(skb)->check =
3480 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3481 &ipv6_hdr(skb)->daddr,
3482 0, IPPROTO_TCP, 0);
3483 adapter->hw_tso6_ctxt++;
3484 }
3485
3486 i = tx_ring->next_to_use;
3487
3488 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3489 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
3490
3491 /* VLAN MACLEN IPLEN */
3492 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3493 vlan_macip_lens |=
3494 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
3495 vlan_macip_lens |= ((skb_network_offset(skb)) <<
3496 IXGBE_ADVTXD_MACLEN_SHIFT);
3497 *hdr_len += skb_network_offset(skb);
3498 vlan_macip_lens |=
3499 (skb_transport_header(skb) - skb_network_header(skb));
3500 *hdr_len +=
3501 (skb_transport_header(skb) - skb_network_header(skb));
3502 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3503 context_desc->seqnum_seed = 0;
3504
3505 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3506 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
3507 IXGBE_ADVTXD_DTYP_CTXT);
3508
3509 if (skb->protocol == htons(ETH_P_IP))
3510 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
3511 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3512 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
3513
3514 /* MSS L4LEN IDX */
3515 mss_l4len_idx =
3516 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
3517 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
3518 /* use index 1 for TSO */
3519 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3520 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3521
3522 tx_buffer_info->time_stamp = jiffies;
3523 tx_buffer_info->next_to_watch = i;
3524
3525 i++;
3526 if (i == tx_ring->count)
3527 i = 0;
3528 tx_ring->next_to_use = i;
3529
3530 return true;
3531 }
3532 return false;
3533 }
3534
3535 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
3536 struct ixgbe_ring *tx_ring,
3537 struct sk_buff *skb, u32 tx_flags)
3538 {
3539 struct ixgbe_adv_tx_context_desc *context_desc;
3540 unsigned int i;
3541 struct ixgbe_tx_buffer *tx_buffer_info;
3542 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
3543
3544 if (skb->ip_summed == CHECKSUM_PARTIAL ||
3545 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
3546 i = tx_ring->next_to_use;
3547 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3548 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
3549
3550 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3551 vlan_macip_lens |=
3552 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
3553 vlan_macip_lens |= (skb_network_offset(skb) <<
3554 IXGBE_ADVTXD_MACLEN_SHIFT);
3555 if (skb->ip_summed == CHECKSUM_PARTIAL)
3556 vlan_macip_lens |= (skb_transport_header(skb) -
3557 skb_network_header(skb));
3558
3559 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3560 context_desc->seqnum_seed = 0;
3561
3562 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
3563 IXGBE_ADVTXD_DTYP_CTXT);
3564
3565 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3566 switch (skb->protocol) {
3567 case __constant_htons(ETH_P_IP):
3568 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
3569 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3570 type_tucmd_mlhl |=
3571 IXGBE_ADVTXD_TUCMD_L4T_TCP;
3572 break;
3573 case __constant_htons(ETH_P_IPV6):
3574 /* XXX what about other V6 headers?? */
3575 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3576 type_tucmd_mlhl |=
3577 IXGBE_ADVTXD_TUCMD_L4T_TCP;
3578 break;
3579 default:
3580 if (unlikely(net_ratelimit())) {
3581 DPRINTK(PROBE, WARNING,
3582 "partial checksum but proto=%x!\n",
3583 skb->protocol);
3584 }
3585 break;
3586 }
3587 }
3588
3589 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
3590 /* use index zero for tx checksum offload */
3591 context_desc->mss_l4len_idx = 0;
3592
3593 tx_buffer_info->time_stamp = jiffies;
3594 tx_buffer_info->next_to_watch = i;
3595
3596 adapter->hw_csum_tx_good++;
3597 i++;
3598 if (i == tx_ring->count)
3599 i = 0;
3600 tx_ring->next_to_use = i;
3601
3602 return true;
3603 }
3604
3605 return false;
3606 }
3607
3608 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
3609 struct ixgbe_ring *tx_ring,
3610 struct sk_buff *skb, unsigned int first)
3611 {
3612 struct ixgbe_tx_buffer *tx_buffer_info;
3613 unsigned int len = skb->len;
3614 unsigned int offset = 0, size, count = 0, i;
3615 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
3616 unsigned int f;
3617
3618 len -= skb->data_len;
3619
3620 i = tx_ring->next_to_use;
3621
3622 while (len) {
3623 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3624 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
3625
3626 tx_buffer_info->length = size;
3627 tx_buffer_info->dma = pci_map_single(adapter->pdev,
3628 skb->data + offset,
3629 size, PCI_DMA_TODEVICE);
3630 tx_buffer_info->time_stamp = jiffies;
3631 tx_buffer_info->next_to_watch = i;
3632
3633 len -= size;
3634 offset += size;
3635 count++;
3636 i++;
3637 if (i == tx_ring->count)
3638 i = 0;
3639 }
3640
3641 for (f = 0; f < nr_frags; f++) {
3642 struct skb_frag_struct *frag;
3643
3644 frag = &skb_shinfo(skb)->frags[f];
3645 len = frag->size;
3646 offset = frag->page_offset;
3647
3648 while (len) {
3649 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3650 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
3651
3652 tx_buffer_info->length = size;
3653 tx_buffer_info->dma = pci_map_page(adapter->pdev,
3654 frag->page,
3655 offset,
3656 size,
3657 PCI_DMA_TODEVICE);
3658 tx_buffer_info->time_stamp = jiffies;
3659 tx_buffer_info->next_to_watch = i;
3660
3661 len -= size;
3662 offset += size;
3663 count++;
3664 i++;
3665 if (i == tx_ring->count)
3666 i = 0;
3667 }
3668 }
3669 if (i == 0)
3670 i = tx_ring->count - 1;
3671 else
3672 i = i - 1;
3673 tx_ring->tx_buffer_info[i].skb = skb;
3674 tx_ring->tx_buffer_info[first].next_to_watch = i;
3675
3676 return count;
3677 }
3678
3679 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
3680 struct ixgbe_ring *tx_ring,
3681 int tx_flags, int count, u32 paylen, u8 hdr_len)
3682 {
3683 union ixgbe_adv_tx_desc *tx_desc = NULL;
3684 struct ixgbe_tx_buffer *tx_buffer_info;
3685 u32 olinfo_status = 0, cmd_type_len = 0;
3686 unsigned int i;
3687 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
3688
3689 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
3690
3691 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
3692
3693 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3694 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
3695
3696 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
3697 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
3698
3699 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
3700 IXGBE_ADVTXD_POPTS_SHIFT;
3701
3702 /* use index 1 context for tso */
3703 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3704 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3705 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
3706 IXGBE_ADVTXD_POPTS_SHIFT;
3707
3708 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3709 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
3710 IXGBE_ADVTXD_POPTS_SHIFT;
3711
3712 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
3713
3714 i = tx_ring->next_to_use;
3715 while (count--) {
3716 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3717 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
3718 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
3719 tx_desc->read.cmd_type_len =
3720 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
3721 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3722 i++;
3723 if (i == tx_ring->count)
3724 i = 0;
3725 }
3726
3727 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
3728
3729 /*
3730 * Force memory writes to complete before letting h/w
3731 * know there are new descriptors to fetch. (Only
3732 * applicable for weak-ordered memory model archs,
3733 * such as IA-64).
3734 */
3735 wmb();
3736
3737 tx_ring->next_to_use = i;
3738 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3739 }
3740
3741 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
3742 struct ixgbe_ring *tx_ring, int size)
3743 {
3744 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3745
3746 netif_stop_subqueue(netdev, tx_ring->queue_index);
3747 /* Herbert's original patch had:
3748 * smp_mb__after_netif_stop_queue();
3749 * but since that doesn't exist yet, just open code it. */
3750 smp_mb();
3751
3752 /* We need to check again in a case another CPU has just
3753 * made room available. */
3754 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
3755 return -EBUSY;
3756
3757 /* A reprieve! - use start_queue because it doesn't call schedule */
3758 netif_start_subqueue(netdev, tx_ring->queue_index);
3759 ++adapter->restart_queue;
3760 return 0;
3761 }
3762
3763 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
3764 struct ixgbe_ring *tx_ring, int size)
3765 {
3766 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
3767 return 0;
3768 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
3769 }
3770
3771 static int ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3772 {
3773 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3774 struct ixgbe_ring *tx_ring;
3775 unsigned int first;
3776 unsigned int tx_flags = 0;
3777 u8 hdr_len = 0;
3778 int r_idx = 0, tso;
3779 int count = 0;
3780 unsigned int f;
3781
3782 r_idx = (adapter->num_tx_queues - 1) & skb->queue_mapping;
3783 tx_ring = &adapter->tx_ring[r_idx];
3784
3785 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3786 tx_flags |= vlan_tx_tag_get(skb);
3787 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3788 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
3789 tx_flags |= (skb->queue_mapping << 13);
3790 }
3791 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3792 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3793 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3794 tx_flags |= (skb->queue_mapping << 13);
3795 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3796 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3797 }
3798 /* three things can cause us to need a context descriptor */
3799 if (skb_is_gso(skb) ||
3800 (skb->ip_summed == CHECKSUM_PARTIAL) ||
3801 (tx_flags & IXGBE_TX_FLAGS_VLAN))
3802 count++;
3803
3804 count += TXD_USE_COUNT(skb_headlen(skb));
3805 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3806 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3807
3808 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
3809 adapter->tx_busy++;
3810 return NETDEV_TX_BUSY;
3811 }
3812
3813 if (skb->protocol == htons(ETH_P_IP))
3814 tx_flags |= IXGBE_TX_FLAGS_IPV4;
3815 first = tx_ring->next_to_use;
3816 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
3817 if (tso < 0) {
3818 dev_kfree_skb_any(skb);
3819 return NETDEV_TX_OK;
3820 }
3821
3822 if (tso)
3823 tx_flags |= IXGBE_TX_FLAGS_TSO;
3824 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
3825 (skb->ip_summed == CHECKSUM_PARTIAL))
3826 tx_flags |= IXGBE_TX_FLAGS_CSUM;
3827
3828 ixgbe_tx_queue(adapter, tx_ring, tx_flags,
3829 ixgbe_tx_map(adapter, tx_ring, skb, first),
3830 skb->len, hdr_len);
3831
3832 netdev->trans_start = jiffies;
3833
3834 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
3835
3836 return NETDEV_TX_OK;
3837 }
3838
3839 /**
3840 * ixgbe_get_stats - Get System Network Statistics
3841 * @netdev: network interface device structure
3842 *
3843 * Returns the address of the device statistics structure.
3844 * The statistics are actually updated from the timer callback.
3845 **/
3846 static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
3847 {
3848 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3849
3850 /* only return the current stats */
3851 return &adapter->net_stats;
3852 }
3853
3854 /**
3855 * ixgbe_set_mac - Change the Ethernet Address of the NIC
3856 * @netdev: network interface device structure
3857 * @p: pointer to an address structure
3858 *
3859 * Returns 0 on success, negative on failure
3860 **/
3861 static int ixgbe_set_mac(struct net_device *netdev, void *p)
3862 {
3863 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3864 struct ixgbe_hw *hw = &adapter->hw;
3865 struct sockaddr *addr = p;
3866
3867 if (!is_valid_ether_addr(addr->sa_data))
3868 return -EADDRNOTAVAIL;
3869
3870 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3871 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3872
3873 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
3874
3875 return 0;
3876 }
3877
3878 #ifdef CONFIG_NET_POLL_CONTROLLER
3879 /*
3880 * Polling 'interrupt' - used by things like netconsole to send skbs
3881 * without having to re-enable interrupts. It's not called while
3882 * the interrupt routine is executing.
3883 */
3884 static void ixgbe_netpoll(struct net_device *netdev)
3885 {
3886 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3887
3888 disable_irq(adapter->pdev->irq);
3889 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
3890 ixgbe_intr(adapter->pdev->irq, netdev);
3891 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
3892 enable_irq(adapter->pdev->irq);
3893 }
3894 #endif
3895
3896 /**
3897 * ixgbe_link_config - set up initial link with default speed and duplex
3898 * @hw: pointer to private hardware struct
3899 *
3900 * Returns 0 on success, negative on failure
3901 **/
3902 static int ixgbe_link_config(struct ixgbe_hw *hw)
3903 {
3904 u32 autoneg;
3905 bool link_up = false;
3906 u32 ret = IXGBE_ERR_LINK_SETUP;
3907
3908 if (hw->mac.ops.check_link)
3909 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3910
3911 if (ret || !link_up)
3912 goto link_cfg_out;
3913
3914 if (hw->mac.ops.get_link_capabilities)
3915 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3916 &hw->mac.autoneg);
3917 if (ret)
3918 goto link_cfg_out;
3919
3920 if (hw->mac.ops.setup_link_speed)
3921 ret = hw->mac.ops.setup_link_speed(hw, autoneg, true, true);
3922
3923 link_cfg_out:
3924 return ret;
3925 }
3926
3927 static const struct net_device_ops ixgbe_netdev_ops = {
3928 .ndo_open = ixgbe_open,
3929 .ndo_stop = ixgbe_close,
3930 .ndo_start_xmit = ixgbe_xmit_frame,
3931 .ndo_get_stats = ixgbe_get_stats,
3932 .ndo_set_multicast_list = ixgbe_set_rx_mode,
3933 .ndo_validate_addr = eth_validate_addr,
3934 .ndo_set_mac_address = ixgbe_set_mac,
3935 .ndo_change_mtu = ixgbe_change_mtu,
3936 .ndo_tx_timeout = ixgbe_tx_timeout,
3937 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
3938 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
3939 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
3940 #ifdef CONFIG_NET_POLL_CONTROLLER
3941 .ndo_poll_controller = ixgbe_netpoll,
3942 #endif
3943 };
3944
3945 /**
3946 * ixgbe_probe - Device Initialization Routine
3947 * @pdev: PCI device information struct
3948 * @ent: entry in ixgbe_pci_tbl
3949 *
3950 * Returns 0 on success, negative on failure
3951 *
3952 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
3953 * The OS initialization, configuring of the adapter private structure,
3954 * and a hardware reset occur.
3955 **/
3956 static int __devinit ixgbe_probe(struct pci_dev *pdev,
3957 const struct pci_device_id *ent)
3958 {
3959 struct net_device *netdev;
3960 struct ixgbe_adapter *adapter = NULL;
3961 struct ixgbe_hw *hw;
3962 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
3963 static int cards_found;
3964 int i, err, pci_using_dac;
3965 u16 link_status, link_speed, link_width;
3966 u32 part_num, eec;
3967
3968 err = pci_enable_device(pdev);
3969 if (err)
3970 return err;
3971
3972 if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK) &&
3973 !pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK)) {
3974 pci_using_dac = 1;
3975 } else {
3976 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
3977 if (err) {
3978 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
3979 if (err) {
3980 dev_err(&pdev->dev, "No usable DMA "
3981 "configuration, aborting\n");
3982 goto err_dma;
3983 }
3984 }
3985 pci_using_dac = 0;
3986 }
3987
3988 err = pci_request_regions(pdev, ixgbe_driver_name);
3989 if (err) {
3990 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3991 goto err_pci_reg;
3992 }
3993
3994 err = pci_enable_pcie_error_reporting(pdev);
3995 if (err) {
3996 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
3997 "0x%x\n", err);
3998 /* non-fatal, continue */
3999 }
4000
4001 pci_set_master(pdev);
4002 pci_save_state(pdev);
4003
4004 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
4005 if (!netdev) {
4006 err = -ENOMEM;
4007 goto err_alloc_etherdev;
4008 }
4009
4010 SET_NETDEV_DEV(netdev, &pdev->dev);
4011
4012 pci_set_drvdata(pdev, netdev);
4013 adapter = netdev_priv(netdev);
4014
4015 adapter->netdev = netdev;
4016 adapter->pdev = pdev;
4017 hw = &adapter->hw;
4018 hw->back = adapter;
4019 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
4020
4021 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4022 pci_resource_len(pdev, 0));
4023 if (!hw->hw_addr) {
4024 err = -EIO;
4025 goto err_ioremap;
4026 }
4027
4028 for (i = 1; i <= 5; i++) {
4029 if (pci_resource_len(pdev, i) == 0)
4030 continue;
4031 }
4032
4033 netdev->netdev_ops = &ixgbe_netdev_ops;
4034 ixgbe_set_ethtool_ops(netdev);
4035 netdev->watchdog_timeo = 5 * HZ;
4036 strcpy(netdev->name, pci_name(pdev));
4037
4038 adapter->bd_number = cards_found;
4039
4040 /* Setup hw api */
4041 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4042 hw->mac.type = ii->mac;
4043
4044 /* EEPROM */
4045 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
4046 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
4047 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
4048 if (!(eec & (1 << 8)))
4049 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
4050
4051 /* PHY */
4052 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
4053 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
4054
4055 /* set up this timer and work struct before calling get_invariants
4056 * which might start the timer
4057 */
4058 init_timer(&adapter->sfp_timer);
4059 adapter->sfp_timer.function = &ixgbe_sfp_timer;
4060 adapter->sfp_timer.data = (unsigned long) adapter;
4061
4062 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
4063
4064 err = ii->get_invariants(hw);
4065 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
4066 /* start a kernel thread to watch for a module to arrive */
4067 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4068 mod_timer(&adapter->sfp_timer,
4069 round_jiffies(jiffies + (2 * HZ)));
4070 err = 0;
4071 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4072 DPRINTK(PROBE, ERR, "failed to load because an "
4073 "unsupported SFP+ module type was detected.\n");
4074 goto err_hw_init;
4075 } else if (err) {
4076 goto err_hw_init;
4077 }
4078
4079 /* setup the private structure */
4080 err = ixgbe_sw_init(adapter);
4081 if (err)
4082 goto err_sw_init;
4083
4084 /* reset_hw fills in the perm_addr as well */
4085 err = hw->mac.ops.reset_hw(hw);
4086 if (err) {
4087 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
4088 goto err_sw_init;
4089 }
4090
4091 netdev->features = NETIF_F_SG |
4092 NETIF_F_IP_CSUM |
4093 NETIF_F_HW_VLAN_TX |
4094 NETIF_F_HW_VLAN_RX |
4095 NETIF_F_HW_VLAN_FILTER;
4096
4097 netdev->features |= NETIF_F_IPV6_CSUM;
4098 netdev->features |= NETIF_F_TSO;
4099 netdev->features |= NETIF_F_TSO6;
4100 netdev->features |= NETIF_F_GRO;
4101
4102 netdev->vlan_features |= NETIF_F_TSO;
4103 netdev->vlan_features |= NETIF_F_TSO6;
4104 netdev->vlan_features |= NETIF_F_IP_CSUM;
4105 netdev->vlan_features |= NETIF_F_SG;
4106
4107 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
4108 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4109
4110 #ifdef CONFIG_IXGBE_DCB
4111 netdev->dcbnl_ops = &dcbnl_ops;
4112 #endif
4113
4114 if (pci_using_dac)
4115 netdev->features |= NETIF_F_HIGHDMA;
4116
4117 /* make sure the EEPROM is good */
4118 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
4119 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
4120 err = -EIO;
4121 goto err_eeprom;
4122 }
4123
4124 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
4125 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
4126
4127 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
4128 dev_err(&pdev->dev, "invalid MAC address\n");
4129 err = -EIO;
4130 goto err_eeprom;
4131 }
4132
4133 init_timer(&adapter->watchdog_timer);
4134 adapter->watchdog_timer.function = &ixgbe_watchdog;
4135 adapter->watchdog_timer.data = (unsigned long)adapter;
4136
4137 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
4138 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
4139
4140 err = ixgbe_init_interrupt_scheme(adapter);
4141 if (err)
4142 goto err_sw_init;
4143
4144 /* print bus type/speed/width info */
4145 pci_read_config_word(pdev, IXGBE_PCI_LINK_STATUS, &link_status);
4146 link_speed = link_status & IXGBE_PCI_LINK_SPEED;
4147 link_width = link_status & IXGBE_PCI_LINK_WIDTH;
4148 dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
4149 ((link_speed == IXGBE_PCI_LINK_SPEED_5000) ? "5.0Gb/s" :
4150 (link_speed == IXGBE_PCI_LINK_SPEED_2500) ? "2.5Gb/s" :
4151 "Unknown"),
4152 ((link_width == IXGBE_PCI_LINK_WIDTH_8) ? "Width x8" :
4153 (link_width == IXGBE_PCI_LINK_WIDTH_4) ? "Width x4" :
4154 (link_width == IXGBE_PCI_LINK_WIDTH_2) ? "Width x2" :
4155 (link_width == IXGBE_PCI_LINK_WIDTH_1) ? "Width x1" :
4156 "Unknown"),
4157 netdev->dev_addr);
4158 ixgbe_read_pba_num_generic(hw, &part_num);
4159 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4160 hw->mac.type, hw->phy.type,
4161 (part_num >> 8), (part_num & 0xff));
4162
4163 if (link_width <= IXGBE_PCI_LINK_WIDTH_4) {
4164 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
4165 "this card is not sufficient for optimal "
4166 "performance.\n");
4167 dev_warn(&pdev->dev, "For optimal performance a x8 "
4168 "PCI-Express slot is required.\n");
4169 }
4170
4171 /* reset the hardware with the new settings */
4172 hw->mac.ops.start_hw(hw);
4173
4174 /* link_config depends on start_hw being called at least once */
4175 err = ixgbe_link_config(hw);
4176 if (err) {
4177 dev_err(&pdev->dev, "setup_link_speed FAILED %d\n", err);
4178 goto err_register;
4179 }
4180
4181 netif_carrier_off(netdev);
4182
4183 strcpy(netdev->name, "eth%d");
4184 err = register_netdev(netdev);
4185 if (err)
4186 goto err_register;
4187
4188 #ifdef CONFIG_IXGBE_DCA
4189 if (dca_add_requester(&pdev->dev) == 0) {
4190 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
4191 /* always use CB2 mode, difference is masked
4192 * in the CB driver */
4193 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
4194 ixgbe_setup_dca(adapter);
4195 }
4196 #endif
4197
4198 dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
4199 cards_found++;
4200 return 0;
4201
4202 err_register:
4203 ixgbe_release_hw_control(adapter);
4204 err_hw_init:
4205 err_sw_init:
4206 ixgbe_reset_interrupt_capability(adapter);
4207 err_eeprom:
4208 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4209 del_timer_sync(&adapter->sfp_timer);
4210 cancel_work_sync(&adapter->sfp_task);
4211 iounmap(hw->hw_addr);
4212 err_ioremap:
4213 free_netdev(netdev);
4214 err_alloc_etherdev:
4215 pci_release_regions(pdev);
4216 err_pci_reg:
4217 err_dma:
4218 pci_disable_device(pdev);
4219 return err;
4220 }
4221
4222 /**
4223 * ixgbe_remove - Device Removal Routine
4224 * @pdev: PCI device information struct
4225 *
4226 * ixgbe_remove is called by the PCI subsystem to alert the driver
4227 * that it should release a PCI device. The could be caused by a
4228 * Hot-Plug event, or because the driver is going to be removed from
4229 * memory.
4230 **/
4231 static void __devexit ixgbe_remove(struct pci_dev *pdev)
4232 {
4233 struct net_device *netdev = pci_get_drvdata(pdev);
4234 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4235 int err;
4236
4237 set_bit(__IXGBE_DOWN, &adapter->state);
4238 /* clear the module not found bit to make sure the worker won't
4239 * reschedule
4240 */
4241 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4242 del_timer_sync(&adapter->watchdog_timer);
4243
4244 del_timer_sync(&adapter->sfp_timer);
4245 cancel_work_sync(&adapter->watchdog_task);
4246 cancel_work_sync(&adapter->sfp_task);
4247 flush_scheduled_work();
4248
4249 #ifdef CONFIG_IXGBE_DCA
4250 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
4251 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
4252 dca_remove_requester(&pdev->dev);
4253 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
4254 }
4255
4256 #endif
4257 if (netdev->reg_state == NETREG_REGISTERED)
4258 unregister_netdev(netdev);
4259
4260 ixgbe_reset_interrupt_capability(adapter);
4261
4262 ixgbe_release_hw_control(adapter);
4263
4264 iounmap(adapter->hw.hw_addr);
4265 pci_release_regions(pdev);
4266
4267 DPRINTK(PROBE, INFO, "complete\n");
4268 kfree(adapter->tx_ring);
4269 kfree(adapter->rx_ring);
4270
4271 free_netdev(netdev);
4272
4273 err = pci_disable_pcie_error_reporting(pdev);
4274 if (err)
4275 dev_err(&pdev->dev,
4276 "pci_disable_pcie_error_reporting failed 0x%x\n", err);
4277
4278 pci_disable_device(pdev);
4279 }
4280
4281 /**
4282 * ixgbe_io_error_detected - called when PCI error is detected
4283 * @pdev: Pointer to PCI device
4284 * @state: The current pci connection state
4285 *
4286 * This function is called after a PCI bus error affecting
4287 * this device has been detected.
4288 */
4289 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
4290 pci_channel_state_t state)
4291 {
4292 struct net_device *netdev = pci_get_drvdata(pdev);
4293 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4294
4295 netif_device_detach(netdev);
4296
4297 if (netif_running(netdev))
4298 ixgbe_down(adapter);
4299 pci_disable_device(pdev);
4300
4301 /* Request a slot reset. */
4302 return PCI_ERS_RESULT_NEED_RESET;
4303 }
4304
4305 /**
4306 * ixgbe_io_slot_reset - called after the pci bus has been reset.
4307 * @pdev: Pointer to PCI device
4308 *
4309 * Restart the card from scratch, as if from a cold-boot.
4310 */
4311 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
4312 {
4313 struct net_device *netdev = pci_get_drvdata(pdev);
4314 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4315 pci_ers_result_t result;
4316 int err;
4317
4318 if (pci_enable_device(pdev)) {
4319 DPRINTK(PROBE, ERR,
4320 "Cannot re-enable PCI device after reset.\n");
4321 result = PCI_ERS_RESULT_DISCONNECT;
4322 } else {
4323 pci_set_master(pdev);
4324 pci_restore_state(pdev);
4325
4326 pci_enable_wake(pdev, PCI_D3hot, 0);
4327 pci_enable_wake(pdev, PCI_D3cold, 0);
4328
4329 ixgbe_reset(adapter);
4330
4331 result = PCI_ERS_RESULT_RECOVERED;
4332 }
4333
4334 err = pci_cleanup_aer_uncorrect_error_status(pdev);
4335 if (err) {
4336 dev_err(&pdev->dev,
4337 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
4338 /* non-fatal, continue */
4339 }
4340
4341 return result;
4342 }
4343
4344 /**
4345 * ixgbe_io_resume - called when traffic can start flowing again.
4346 * @pdev: Pointer to PCI device
4347 *
4348 * This callback is called when the error recovery driver tells us that
4349 * its OK to resume normal operation.
4350 */
4351 static void ixgbe_io_resume(struct pci_dev *pdev)
4352 {
4353 struct net_device *netdev = pci_get_drvdata(pdev);
4354 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4355
4356 if (netif_running(netdev)) {
4357 if (ixgbe_up(adapter)) {
4358 DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
4359 return;
4360 }
4361 }
4362
4363 netif_device_attach(netdev);
4364 }
4365
4366 static struct pci_error_handlers ixgbe_err_handler = {
4367 .error_detected = ixgbe_io_error_detected,
4368 .slot_reset = ixgbe_io_slot_reset,
4369 .resume = ixgbe_io_resume,
4370 };
4371
4372 static struct pci_driver ixgbe_driver = {
4373 .name = ixgbe_driver_name,
4374 .id_table = ixgbe_pci_tbl,
4375 .probe = ixgbe_probe,
4376 .remove = __devexit_p(ixgbe_remove),
4377 #ifdef CONFIG_PM
4378 .suspend = ixgbe_suspend,
4379 .resume = ixgbe_resume,
4380 #endif
4381 .shutdown = ixgbe_shutdown,
4382 .err_handler = &ixgbe_err_handler
4383 };
4384
4385 /**
4386 * ixgbe_init_module - Driver Registration Routine
4387 *
4388 * ixgbe_init_module is the first routine called when the driver is
4389 * loaded. All it does is register with the PCI subsystem.
4390 **/
4391 static int __init ixgbe_init_module(void)
4392 {
4393 int ret;
4394 printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
4395 ixgbe_driver_string, ixgbe_driver_version);
4396
4397 printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
4398
4399 #ifdef CONFIG_IXGBE_DCA
4400 dca_register_notify(&dca_notifier);
4401 #endif
4402
4403 ret = pci_register_driver(&ixgbe_driver);
4404 return ret;
4405 }
4406
4407 module_init(ixgbe_init_module);
4408
4409 /**
4410 * ixgbe_exit_module - Driver Exit Cleanup Routine
4411 *
4412 * ixgbe_exit_module is called just before the driver is removed
4413 * from memory.
4414 **/
4415 static void __exit ixgbe_exit_module(void)
4416 {
4417 #ifdef CONFIG_IXGBE_DCA
4418 dca_unregister_notify(&dca_notifier);
4419 #endif
4420 pci_unregister_driver(&ixgbe_driver);
4421 }
4422
4423 #ifdef CONFIG_IXGBE_DCA
4424 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
4425 void *p)
4426 {
4427 int ret_val;
4428
4429 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
4430 __ixgbe_notify_dca);
4431
4432 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4433 }
4434 #endif /* CONFIG_IXGBE_DCA */
4435
4436 module_exit(ixgbe_exit_module);
4437
4438 /* ixgbe_main.c */
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree