]> bbs.cooldavid.org Git - net-next-2.6.git/blame_incremental - drivers/net/ixgbe/ixgbe_main.c
git://bbs.cooldavid.org / net-next-2.6.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
ixgbe: Disable DROP_EN for Rx queues
[net-next-2.6.git] / drivers / net / ixgbe / ixgbe_main.c
... / ...
CommitLineData
1/*******************************************************************************
2
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2009 Intel Corporation.
5
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
9
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 more details.
14
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
21
22 Contact Information:
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26*******************************************************************************/
27
28#include <linux/types.h>
29#include <linux/module.h>
30#include <linux/pci.h>
31#include <linux/netdevice.h>
32#include <linux/vmalloc.h>
33#include <linux/string.h>
34#include <linux/in.h>
35#include <linux/ip.h>
36#include <linux/tcp.h>
37#include <linux/ipv6.h>
38#include <net/checksum.h>
39#include <net/ip6_checksum.h>
40#include <linux/ethtool.h>
41#include <linux/if_vlan.h>
42
43#include "ixgbe.h"
44#include "ixgbe_common.h"
45
46char ixgbe_driver_name[] = "ixgbe";
47static const char ixgbe_driver_string[] =
48 "Intel(R) 10 Gigabit PCI Express Network Driver";
49
50#define DRV_VERSION "2.0.8-k2"
51const char ixgbe_driver_version[] = DRV_VERSION;
52static char ixgbe_copyright[] = "Copyright (c) 1999-2009 Intel Corporation.";
53
54static const struct ixgbe_info *ixgbe_info_tbl[] = {
55 [board_82598] = &ixgbe_82598_info,
56 [board_82599] = &ixgbe_82599_info,
57};
58
59/* ixgbe_pci_tbl - PCI Device ID Table
60 *
61 * Wildcard entries (PCI_ANY_ID) should come last
62 * Last entry must be all 0s
63 *
64 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
65 * Class, Class Mask, private data (not used) }
66 */
67static struct pci_device_id ixgbe_pci_tbl[] = {
68 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
69 board_82598 },
70 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
71 board_82598 },
72 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
73 board_82598 },
74 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
75 board_82598 },
76 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
77 board_82598 },
78 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
79 board_82598 },
80 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
81 board_82598 },
82 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
83 board_82598 },
84 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
85 board_82598 },
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
87 board_82598 },
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
89 board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
91 board_82599 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
93 board_82599 },
94
95 /* required last entry */
96 {0, }
97};
98MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
99
100#ifdef CONFIG_IXGBE_DCA
101static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
102 void *p);
103static struct notifier_block dca_notifier = {
104 .notifier_call = ixgbe_notify_dca,
105 .next = NULL,
106 .priority = 0
107};
108#endif
109
110MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
111MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
112MODULE_LICENSE("GPL");
113MODULE_VERSION(DRV_VERSION);
114
115#define DEFAULT_DEBUG_LEVEL_SHIFT 3
116
117static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
118{
119 u32 ctrl_ext;
120
121 /* Let firmware take over control of h/w */
122 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
123 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
124 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
125}
126
127static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
128{
129 u32 ctrl_ext;
130
131 /* Let firmware know the driver has taken over */
132 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
133 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
134 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
135}
136
137/*
138 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
139 * @adapter: pointer to adapter struct
140 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
141 * @queue: queue to map the corresponding interrupt to
142 * @msix_vector: the vector to map to the corresponding queue
143 *
144 */
145static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
146 u8 queue, u8 msix_vector)
147{
148 u32 ivar, index;
149 struct ixgbe_hw *hw = &adapter->hw;
150 switch (hw->mac.type) {
151 case ixgbe_mac_82598EB:
152 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
153 if (direction == -1)
154 direction = 0;
155 index = (((direction * 64) + queue) >> 2) & 0x1F;
156 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
157 ivar &= ~(0xFF << (8 * (queue & 0x3)));
158 ivar |= (msix_vector << (8 * (queue & 0x3)));
159 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
160 break;
161 case ixgbe_mac_82599EB:
162 if (direction == -1) {
163 /* other causes */
164 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
165 index = ((queue & 1) * 8);
166 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
167 ivar &= ~(0xFF << index);
168 ivar |= (msix_vector << index);
169 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
170 break;
171 } else {
172 /* tx or rx causes */
173 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
174 index = ((16 * (queue & 1)) + (8 * direction));
175 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
176 ivar &= ~(0xFF << index);
177 ivar |= (msix_vector << index);
178 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
179 break;
180 }
181 default:
182 break;
183 }
184}
185
186static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
187 struct ixgbe_tx_buffer
188 *tx_buffer_info)
189{
190 if (tx_buffer_info->dma) {
191 pci_unmap_page(adapter->pdev, tx_buffer_info->dma,
192 tx_buffer_info->length, PCI_DMA_TODEVICE);
193 tx_buffer_info->dma = 0;
194 }
195 if (tx_buffer_info->skb) {
196 dev_kfree_skb_any(tx_buffer_info->skb);
197 tx_buffer_info->skb = NULL;
198 }
199 /* tx_buffer_info must be completely set up in the transmit path */
200}
201
202static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
203 struct ixgbe_ring *tx_ring,
204 unsigned int eop)
205{
206 struct ixgbe_hw *hw = &adapter->hw;
207 u32 head, tail;
208
209 /* Detect a transmit hang in hardware, this serializes the
210 * check with the clearing of time_stamp and movement of eop */
211 head = IXGBE_READ_REG(hw, tx_ring->head);
212 tail = IXGBE_READ_REG(hw, tx_ring->tail);
213 adapter->detect_tx_hung = false;
214 if ((head != tail) &&
215 tx_ring->tx_buffer_info[eop].time_stamp &&
216 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
217 !(IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & IXGBE_TFCS_TXOFF)) {
218 /* detected Tx unit hang */
219 union ixgbe_adv_tx_desc *tx_desc;
220 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
221 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
222 " Tx Queue <%d>\n"
223 " TDH, TDT <%x>, <%x>\n"
224 " next_to_use <%x>\n"
225 " next_to_clean <%x>\n"
226 "tx_buffer_info[next_to_clean]\n"
227 " time_stamp <%lx>\n"
228 " jiffies <%lx>\n",
229 tx_ring->queue_index,
230 head, tail,
231 tx_ring->next_to_use, eop,
232 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
233 return true;
234 }
235
236 return false;
237}
238
239#define IXGBE_MAX_TXD_PWR 14
240#define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
241
242/* Tx Descriptors needed, worst case */
243#define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
244 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
245#define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
246 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
247
248static void ixgbe_tx_timeout(struct net_device *netdev);
249
250/**
251 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
252 * @adapter: board private structure
253 * @tx_ring: tx ring to clean
254 **/
255static bool ixgbe_clean_tx_irq(struct ixgbe_adapter *adapter,
256 struct ixgbe_ring *tx_ring)
257{
258 struct net_device *netdev = adapter->netdev;
259 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
260 struct ixgbe_tx_buffer *tx_buffer_info;
261 unsigned int i, eop, count = 0;
262 unsigned int total_bytes = 0, total_packets = 0;
263
264 i = tx_ring->next_to_clean;
265 eop = tx_ring->tx_buffer_info[i].next_to_watch;
266 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
267
268 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
269 (count < tx_ring->count)) {
270 bool cleaned = false;
271 for ( ; !cleaned; count++) {
272 struct sk_buff *skb;
273 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
274 tx_buffer_info = &tx_ring->tx_buffer_info[i];
275 cleaned = (i == eop);
276 skb = tx_buffer_info->skb;
277
278 if (cleaned && skb) {
279 unsigned int segs, bytecount;
280
281 /* gso_segs is currently only valid for tcp */
282 segs = skb_shinfo(skb)->gso_segs ?: 1;
283 /* multiply data chunks by size of headers */
284 bytecount = ((segs - 1) * skb_headlen(skb)) +
285 skb->len;
286 total_packets += segs;
287 total_bytes += bytecount;
288 }
289
290 ixgbe_unmap_and_free_tx_resource(adapter,
291 tx_buffer_info);
292
293 tx_desc->wb.status = 0;
294
295 i++;
296 if (i == tx_ring->count)
297 i = 0;
298 }
299
300 eop = tx_ring->tx_buffer_info[i].next_to_watch;
301 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
302 }
303
304 tx_ring->next_to_clean = i;
305
306#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
307 if (unlikely(count && netif_carrier_ok(netdev) &&
308 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
309 /* Make sure that anybody stopping the queue after this
310 * sees the new next_to_clean.
311 */
312 smp_mb();
313 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
314 !test_bit(__IXGBE_DOWN, &adapter->state)) {
315 netif_wake_subqueue(netdev, tx_ring->queue_index);
316 ++adapter->restart_queue;
317 }
318 }
319
320 if (adapter->detect_tx_hung) {
321 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
322 /* schedule immediate reset if we believe we hung */
323 DPRINTK(PROBE, INFO,
324 "tx hang %d detected, resetting adapter\n",
325 adapter->tx_timeout_count + 1);
326 ixgbe_tx_timeout(adapter->netdev);
327 }
328 }
329
330 /* re-arm the interrupt */
331 if ((total_packets >= tx_ring->work_limit) ||
332 (count == tx_ring->count))
333 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, tx_ring->v_idx);
334
335 tx_ring->total_bytes += total_bytes;
336 tx_ring->total_packets += total_packets;
337 tx_ring->stats.packets += total_packets;
338 tx_ring->stats.bytes += total_bytes;
339 adapter->net_stats.tx_bytes += total_bytes;
340 adapter->net_stats.tx_packets += total_packets;
341 return (total_packets ? true : false);
342}
343
344#ifdef CONFIG_IXGBE_DCA
345static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
346 struct ixgbe_ring *rx_ring)
347{
348 u32 rxctrl;
349 int cpu = get_cpu();
350 int q = rx_ring - adapter->rx_ring;
351
352 if (rx_ring->cpu != cpu) {
353 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
354 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
355 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
356 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
357 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
358 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
359 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
360 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
361 }
362 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
363 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
364 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
365 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
366 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
367 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
368 rx_ring->cpu = cpu;
369 }
370 put_cpu();
371}
372
373static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
374 struct ixgbe_ring *tx_ring)
375{
376 u32 txctrl;
377 int cpu = get_cpu();
378 int q = tx_ring - adapter->tx_ring;
379
380 if (tx_ring->cpu != cpu) {
381 txctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q));
382 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
383 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
384 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
385 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
386 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
387 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
388 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
389 }
390 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
391 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q), txctrl);
392 tx_ring->cpu = cpu;
393 }
394 put_cpu();
395}
396
397static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
398{
399 int i;
400
401 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
402 return;
403
404 for (i = 0; i < adapter->num_tx_queues; i++) {
405 adapter->tx_ring[i].cpu = -1;
406 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
407 }
408 for (i = 0; i < adapter->num_rx_queues; i++) {
409 adapter->rx_ring[i].cpu = -1;
410 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
411 }
412}
413
414static int __ixgbe_notify_dca(struct device *dev, void *data)
415{
416 struct net_device *netdev = dev_get_drvdata(dev);
417 struct ixgbe_adapter *adapter = netdev_priv(netdev);
418 unsigned long event = *(unsigned long *)data;
419
420 switch (event) {
421 case DCA_PROVIDER_ADD:
422 /* if we're already enabled, don't do it again */
423 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
424 break;
425 /* Always use CB2 mode, difference is masked
426 * in the CB driver. */
427 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
428 if (dca_add_requester(dev) == 0) {
429 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
430 ixgbe_setup_dca(adapter);
431 break;
432 }
433 /* Fall Through since DCA is disabled. */
434 case DCA_PROVIDER_REMOVE:
435 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
436 dca_remove_requester(dev);
437 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
438 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
439 }
440 break;
441 }
442
443 return 0;
444}
445
446#endif /* CONFIG_IXGBE_DCA */
447/**
448 * ixgbe_receive_skb - Send a completed packet up the stack
449 * @adapter: board private structure
450 * @skb: packet to send up
451 * @status: hardware indication of status of receive
452 * @rx_ring: rx descriptor ring (for a specific queue) to setup
453 * @rx_desc: rx descriptor
454 **/
455static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
456 struct sk_buff *skb, u8 status,
457 union ixgbe_adv_rx_desc *rx_desc)
458{
459 struct ixgbe_adapter *adapter = q_vector->adapter;
460 struct napi_struct *napi = &q_vector->napi;
461 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
462 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
463
464 skb_record_rx_queue(skb, q_vector - &adapter->q_vector[0]);
465 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
466 if (adapter->vlgrp && is_vlan && (tag != 0))
467 vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
468 else
469 napi_gro_receive(napi, skb);
470 } else {
471 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
472 if (adapter->vlgrp && is_vlan && (tag != 0))
473 vlan_hwaccel_receive_skb(skb, adapter->vlgrp, tag);
474 else
475 netif_receive_skb(skb);
476 } else {
477 if (adapter->vlgrp && is_vlan && (tag != 0))
478 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
479 else
480 netif_rx(skb);
481 }
482 }
483}
484
485/**
486 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
487 * @adapter: address of board private structure
488 * @status_err: hardware indication of status of receive
489 * @skb: skb currently being received and modified
490 **/
491static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
492 u32 status_err, struct sk_buff *skb)
493{
494 skb->ip_summed = CHECKSUM_NONE;
495
496 /* Rx csum disabled */
497 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
498 return;
499
500 /* if IP and error */
501 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
502 (status_err & IXGBE_RXDADV_ERR_IPE)) {
503 adapter->hw_csum_rx_error++;
504 return;
505 }
506
507 if (!(status_err & IXGBE_RXD_STAT_L4CS))
508 return;
509
510 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
511 adapter->hw_csum_rx_error++;
512 return;
513 }
514
515 /* It must be a TCP or UDP packet with a valid checksum */
516 skb->ip_summed = CHECKSUM_UNNECESSARY;
517 adapter->hw_csum_rx_good++;
518}
519
520static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
521 struct ixgbe_ring *rx_ring, u32 val)
522{
523 /*
524 * Force memory writes to complete before letting h/w
525 * know there are new descriptors to fetch. (Only
526 * applicable for weak-ordered memory model archs,
527 * such as IA-64).
528 */
529 wmb();
530 IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
531}
532
533/**
534 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
535 * @adapter: address of board private structure
536 **/
537static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
538 struct ixgbe_ring *rx_ring,
539 int cleaned_count)
540{
541 struct pci_dev *pdev = adapter->pdev;
542 union ixgbe_adv_rx_desc *rx_desc;
543 struct ixgbe_rx_buffer *bi;
544 unsigned int i;
545 unsigned int bufsz = rx_ring->rx_buf_len + NET_IP_ALIGN;
546
547 i = rx_ring->next_to_use;
548 bi = &rx_ring->rx_buffer_info[i];
549
550 while (cleaned_count--) {
551 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
552
553 if (!bi->page_dma &&
554 (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
555 if (!bi->page) {
556 bi->page = alloc_page(GFP_ATOMIC);
557 if (!bi->page) {
558 adapter->alloc_rx_page_failed++;
559 goto no_buffers;
560 }
561 bi->page_offset = 0;
562 } else {
563 /* use a half page if we're re-using */
564 bi->page_offset ^= (PAGE_SIZE / 2);
565 }
566
567 bi->page_dma = pci_map_page(pdev, bi->page,
568 bi->page_offset,
569 (PAGE_SIZE / 2),
570 PCI_DMA_FROMDEVICE);
571 }
572
573 if (!bi->skb) {
574 struct sk_buff *skb;
575 skb = netdev_alloc_skb(adapter->netdev, bufsz);
576
577 if (!skb) {
578 adapter->alloc_rx_buff_failed++;
579 goto no_buffers;
580 }
581
582 /*
583 * Make buffer alignment 2 beyond a 16 byte boundary
584 * this will result in a 16 byte aligned IP header after
585 * the 14 byte MAC header is removed
586 */
587 skb_reserve(skb, NET_IP_ALIGN);
588
589 bi->skb = skb;
590 bi->dma = pci_map_single(pdev, skb->data, bufsz,
591 PCI_DMA_FROMDEVICE);
592 }
593 /* Refresh the desc even if buffer_addrs didn't change because
594 * each write-back erases this info. */
595 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
596 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
597 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
598 } else {
599 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
600 }
601
602 i++;
603 if (i == rx_ring->count)
604 i = 0;
605 bi = &rx_ring->rx_buffer_info[i];
606 }
607
608no_buffers:
609 if (rx_ring->next_to_use != i) {
610 rx_ring->next_to_use = i;
611 if (i-- == 0)
612 i = (rx_ring->count - 1);
613
614 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
615 }
616}
617
618static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
619{
620 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
621}
622
623static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
624{
625 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
626}
627
628static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
629 struct ixgbe_ring *rx_ring,
630 int *work_done, int work_to_do)
631{
632 struct ixgbe_adapter *adapter = q_vector->adapter;
633 struct pci_dev *pdev = adapter->pdev;
634 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
635 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
636 struct sk_buff *skb;
637 unsigned int i;
638 u32 len, staterr;
639 u16 hdr_info;
640 bool cleaned = false;
641 int cleaned_count = 0;
642 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
643
644 i = rx_ring->next_to_clean;
645 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
646 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
647 rx_buffer_info = &rx_ring->rx_buffer_info[i];
648
649 while (staterr & IXGBE_RXD_STAT_DD) {
650 u32 upper_len = 0;
651 if (*work_done >= work_to_do)
652 break;
653 (*work_done)++;
654
655 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
656 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
657 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
658 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
659 if (hdr_info & IXGBE_RXDADV_SPH)
660 adapter->rx_hdr_split++;
661 if (len > IXGBE_RX_HDR_SIZE)
662 len = IXGBE_RX_HDR_SIZE;
663 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
664 } else {
665 len = le16_to_cpu(rx_desc->wb.upper.length);
666 }
667
668 cleaned = true;
669 skb = rx_buffer_info->skb;
670 prefetch(skb->data - NET_IP_ALIGN);
671 rx_buffer_info->skb = NULL;
672
673 if (len && !skb_shinfo(skb)->nr_frags) {
674 pci_unmap_single(pdev, rx_buffer_info->dma,
675 rx_ring->rx_buf_len,
676 PCI_DMA_FROMDEVICE);
677 skb_put(skb, len);
678 }
679
680 if (upper_len) {
681 pci_unmap_page(pdev, rx_buffer_info->page_dma,
682 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
683 rx_buffer_info->page_dma = 0;
684 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
685 rx_buffer_info->page,
686 rx_buffer_info->page_offset,
687 upper_len);
688
689 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
690 (page_count(rx_buffer_info->page) != 1))
691 rx_buffer_info->page = NULL;
692 else
693 get_page(rx_buffer_info->page);
694
695 skb->len += upper_len;
696 skb->data_len += upper_len;
697 skb->truesize += upper_len;
698 }
699
700 i++;
701 if (i == rx_ring->count)
702 i = 0;
703 next_buffer = &rx_ring->rx_buffer_info[i];
704
705 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
706 prefetch(next_rxd);
707
708 cleaned_count++;
709 if (staterr & IXGBE_RXD_STAT_EOP) {
710 rx_ring->stats.packets++;
711 rx_ring->stats.bytes += skb->len;
712 } else {
713 rx_buffer_info->skb = next_buffer->skb;
714 rx_buffer_info->dma = next_buffer->dma;
715 next_buffer->skb = skb;
716 next_buffer->dma = 0;
717 adapter->non_eop_descs++;
718 goto next_desc;
719 }
720
721 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
722 dev_kfree_skb_irq(skb);
723 goto next_desc;
724 }
725
726 ixgbe_rx_checksum(adapter, staterr, skb);
727
728 /* probably a little skewed due to removing CRC */
729 total_rx_bytes += skb->len;
730 total_rx_packets++;
731
732 skb->protocol = eth_type_trans(skb, adapter->netdev);
733 ixgbe_receive_skb(q_vector, skb, staterr, rx_desc);
734
735next_desc:
736 rx_desc->wb.upper.status_error = 0;
737
738 /* return some buffers to hardware, one at a time is too slow */
739 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
740 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
741 cleaned_count = 0;
742 }
743
744 /* use prefetched values */
745 rx_desc = next_rxd;
746 rx_buffer_info = next_buffer;
747
748 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
749 }
750
751 rx_ring->next_to_clean = i;
752 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
753
754 if (cleaned_count)
755 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
756
757 rx_ring->total_packets += total_rx_packets;
758 rx_ring->total_bytes += total_rx_bytes;
759 adapter->net_stats.rx_bytes += total_rx_bytes;
760 adapter->net_stats.rx_packets += total_rx_packets;
761
762 return cleaned;
763}
764
765static int ixgbe_clean_rxonly(struct napi_struct *, int);
766/**
767 * ixgbe_configure_msix - Configure MSI-X hardware
768 * @adapter: board private structure
769 *
770 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
771 * interrupts.
772 **/
773static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
774{
775 struct ixgbe_q_vector *q_vector;
776 int i, j, q_vectors, v_idx, r_idx;
777 u32 mask;
778
779 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
780
781 /* Populate the IVAR table and set the ITR values to the
782 * corresponding register.
783 */
784 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
785 q_vector = &adapter->q_vector[v_idx];
786 /* XXX for_each_bit(...) */
787 r_idx = find_first_bit(q_vector->rxr_idx,
788 adapter->num_rx_queues);
789
790 for (i = 0; i < q_vector->rxr_count; i++) {
791 j = adapter->rx_ring[r_idx].reg_idx;
792 ixgbe_set_ivar(adapter, 0, j, v_idx);
793 r_idx = find_next_bit(q_vector->rxr_idx,
794 adapter->num_rx_queues,
795 r_idx + 1);
796 }
797 r_idx = find_first_bit(q_vector->txr_idx,
798 adapter->num_tx_queues);
799
800 for (i = 0; i < q_vector->txr_count; i++) {
801 j = adapter->tx_ring[r_idx].reg_idx;
802 ixgbe_set_ivar(adapter, 1, j, v_idx);
803 r_idx = find_next_bit(q_vector->txr_idx,
804 adapter->num_tx_queues,
805 r_idx + 1);
806 }
807
808 /* if this is a tx only vector halve the interrupt rate */
809 if (q_vector->txr_count && !q_vector->rxr_count)
810 q_vector->eitr = (adapter->eitr_param >> 1);
811 else
812 /* rx only */
813 q_vector->eitr = adapter->eitr_param;
814
815 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx),
816 EITR_INTS_PER_SEC_TO_REG(q_vector->eitr));
817 }
818
819 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
820 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
821 v_idx);
822 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
823 ixgbe_set_ivar(adapter, -1, 1, v_idx);
824 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
825
826 /* set up to autoclear timer, and the vectors */
827 mask = IXGBE_EIMS_ENABLE_MASK;
828 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
829 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
830}
831
832enum latency_range {
833 lowest_latency = 0,
834 low_latency = 1,
835 bulk_latency = 2,
836 latency_invalid = 255
837};
838
839/**
840 * ixgbe_update_itr - update the dynamic ITR value based on statistics
841 * @adapter: pointer to adapter
842 * @eitr: eitr setting (ints per sec) to give last timeslice
843 * @itr_setting: current throttle rate in ints/second
844 * @packets: the number of packets during this measurement interval
845 * @bytes: the number of bytes during this measurement interval
846 *
847 * Stores a new ITR value based on packets and byte
848 * counts during the last interrupt. The advantage of per interrupt
849 * computation is faster updates and more accurate ITR for the current
850 * traffic pattern. Constants in this function were computed
851 * based on theoretical maximum wire speed and thresholds were set based
852 * on testing data as well as attempting to minimize response time
853 * while increasing bulk throughput.
854 * this functionality is controlled by the InterruptThrottleRate module
855 * parameter (see ixgbe_param.c)
856 **/
857static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
858 u32 eitr, u8 itr_setting,
859 int packets, int bytes)
860{
861 unsigned int retval = itr_setting;
862 u32 timepassed_us;
863 u64 bytes_perint;
864
865 if (packets == 0)
866 goto update_itr_done;
867
868
869 /* simple throttlerate management
870 * 0-20MB/s lowest (100000 ints/s)
871 * 20-100MB/s low (20000 ints/s)
872 * 100-1249MB/s bulk (8000 ints/s)
873 */
874 /* what was last interrupt timeslice? */
875 timepassed_us = 1000000/eitr;
876 bytes_perint = bytes / timepassed_us; /* bytes/usec */
877
878 switch (itr_setting) {
879 case lowest_latency:
880 if (bytes_perint > adapter->eitr_low)
881 retval = low_latency;
882 break;
883 case low_latency:
884 if (bytes_perint > adapter->eitr_high)
885 retval = bulk_latency;
886 else if (bytes_perint <= adapter->eitr_low)
887 retval = lowest_latency;
888 break;
889 case bulk_latency:
890 if (bytes_perint <= adapter->eitr_high)
891 retval = low_latency;
892 break;
893 }
894
895update_itr_done:
896 return retval;
897}
898
899static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
900{
901 struct ixgbe_adapter *adapter = q_vector->adapter;
902 struct ixgbe_hw *hw = &adapter->hw;
903 u32 new_itr;
904 u8 current_itr, ret_itr;
905 int i, r_idx, v_idx = ((void *)q_vector - (void *)(adapter->q_vector)) /
906 sizeof(struct ixgbe_q_vector);
907 struct ixgbe_ring *rx_ring, *tx_ring;
908
909 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
910 for (i = 0; i < q_vector->txr_count; i++) {
911 tx_ring = &(adapter->tx_ring[r_idx]);
912 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
913 q_vector->tx_itr,
914 tx_ring->total_packets,
915 tx_ring->total_bytes);
916 /* if the result for this queue would decrease interrupt
917 * rate for this vector then use that result */
918 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
919 q_vector->tx_itr - 1 : ret_itr);
920 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
921 r_idx + 1);
922 }
923
924 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
925 for (i = 0; i < q_vector->rxr_count; i++) {
926 rx_ring = &(adapter->rx_ring[r_idx]);
927 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
928 q_vector->rx_itr,
929 rx_ring->total_packets,
930 rx_ring->total_bytes);
931 /* if the result for this queue would decrease interrupt
932 * rate for this vector then use that result */
933 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
934 q_vector->rx_itr - 1 : ret_itr);
935 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
936 r_idx + 1);
937 }
938
939 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
940
941 switch (current_itr) {
942 /* counts and packets in update_itr are dependent on these numbers */
943 case lowest_latency:
944 new_itr = 100000;
945 break;
946 case low_latency:
947 new_itr = 20000; /* aka hwitr = ~200 */
948 break;
949 case bulk_latency:
950 default:
951 new_itr = 8000;
952 break;
953 }
954
955 if (new_itr != q_vector->eitr) {
956 u32 itr_reg;
957 /* do an exponential smoothing */
958 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
959 q_vector->eitr = new_itr;
960 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
961 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
962 /* Resolution is 2 usec on 82599, so halve the rate */
963 itr_reg >>= 1;
964 /* must write high and low 16 bits to reset counter */
965 DPRINTK(TX_ERR, DEBUG, "writing eitr(%d): %08X\n", v_idx,
966 itr_reg);
967 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg | (itr_reg)<<16);
968 }
969
970 return;
971}
972
973static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
974{
975 struct ixgbe_hw *hw = &adapter->hw;
976
977 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
978 (eicr & IXGBE_EICR_GPI_SDP1)) {
979 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
980 /* write to clear the interrupt */
981 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
982 }
983}
984
985static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
986{
987 struct ixgbe_hw *hw = &adapter->hw;
988
989 if (eicr & IXGBE_EICR_GPI_SDP1) {
990 /* Clear the interrupt */
991 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
992 schedule_work(&adapter->multispeed_fiber_task);
993 } else if (eicr & IXGBE_EICR_GPI_SDP2) {
994 /* Clear the interrupt */
995 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
996 schedule_work(&adapter->sfp_config_module_task);
997 } else {
998 /* Interrupt isn't for us... */
999 return;
1000 }
1001}
1002
1003static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1004{
1005 struct ixgbe_hw *hw = &adapter->hw;
1006
1007 adapter->lsc_int++;
1008 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1009 adapter->link_check_timeout = jiffies;
1010 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1011 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1012 schedule_work(&adapter->watchdog_task);
1013 }
1014}
1015
1016static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1017{
1018 struct net_device *netdev = data;
1019 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1020 struct ixgbe_hw *hw = &adapter->hw;
1021 u32 eicr;
1022
1023 /*
1024 * Workaround for Silicon errata. Use clear-by-write instead
1025 * of clear-by-read. Reading with EICS will return the
1026 * interrupt causes without clearing, which later be done
1027 * with the write to EICR.
1028 */
1029 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1030 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1031
1032 if (eicr & IXGBE_EICR_LSC)
1033 ixgbe_check_lsc(adapter);
1034
1035 if (hw->mac.type == ixgbe_mac_82598EB)
1036 ixgbe_check_fan_failure(adapter, eicr);
1037
1038 if (hw->mac.type == ixgbe_mac_82599EB)
1039 ixgbe_check_sfp_event(adapter, eicr);
1040 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1041 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1042
1043 return IRQ_HANDLED;
1044}
1045
1046static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1047{
1048 struct ixgbe_q_vector *q_vector = data;
1049 struct ixgbe_adapter *adapter = q_vector->adapter;
1050 struct ixgbe_ring *tx_ring;
1051 int i, r_idx;
1052
1053 if (!q_vector->txr_count)
1054 return IRQ_HANDLED;
1055
1056 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1057 for (i = 0; i < q_vector->txr_count; i++) {
1058 tx_ring = &(adapter->tx_ring[r_idx]);
1059#ifdef CONFIG_IXGBE_DCA
1060 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1061 ixgbe_update_tx_dca(adapter, tx_ring);
1062#endif
1063 tx_ring->total_bytes = 0;
1064 tx_ring->total_packets = 0;
1065 ixgbe_clean_tx_irq(adapter, tx_ring);
1066 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1067 r_idx + 1);
1068 }
1069
1070 return IRQ_HANDLED;
1071}
1072
1073/**
1074 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1075 * @irq: unused
1076 * @data: pointer to our q_vector struct for this interrupt vector
1077 **/
1078static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1079{
1080 struct ixgbe_q_vector *q_vector = data;
1081 struct ixgbe_adapter *adapter = q_vector->adapter;
1082 struct ixgbe_ring *rx_ring;
1083 int r_idx;
1084 int i;
1085
1086 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1087 for (i = 0; i < q_vector->rxr_count; i++) {
1088 rx_ring = &(adapter->rx_ring[r_idx]);
1089 rx_ring->total_bytes = 0;
1090 rx_ring->total_packets = 0;
1091 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1092 r_idx + 1);
1093 }
1094
1095 if (!q_vector->rxr_count)
1096 return IRQ_HANDLED;
1097
1098 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1099 rx_ring = &(adapter->rx_ring[r_idx]);
1100 /* disable interrupts on this vector only */
1101 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, rx_ring->v_idx);
1102 napi_schedule(&q_vector->napi);
1103
1104 return IRQ_HANDLED;
1105}
1106
1107static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1108{
1109 ixgbe_msix_clean_rx(irq, data);
1110 ixgbe_msix_clean_tx(irq, data);
1111
1112 return IRQ_HANDLED;
1113}
1114
1115/**
1116 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1117 * @napi: napi struct with our devices info in it
1118 * @budget: amount of work driver is allowed to do this pass, in packets
1119 *
1120 * This function is optimized for cleaning one queue only on a single
1121 * q_vector!!!
1122 **/
1123static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1124{
1125 struct ixgbe_q_vector *q_vector =
1126 container_of(napi, struct ixgbe_q_vector, napi);
1127 struct ixgbe_adapter *adapter = q_vector->adapter;
1128 struct ixgbe_ring *rx_ring = NULL;
1129 int work_done = 0;
1130 long r_idx;
1131
1132 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1133 rx_ring = &(adapter->rx_ring[r_idx]);
1134#ifdef CONFIG_IXGBE_DCA
1135 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1136 ixgbe_update_rx_dca(adapter, rx_ring);
1137#endif
1138
1139 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1140
1141 /* If all Rx work done, exit the polling mode */
1142 if (work_done < budget) {
1143 napi_complete(napi);
1144 if (adapter->itr_setting & 3)
1145 ixgbe_set_itr_msix(q_vector);
1146 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1147 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, rx_ring->v_idx);
1148 }
1149
1150 return work_done;
1151}
1152
1153/**
1154 * ixgbe_clean_rxonly_many - msix (aka one shot) rx clean routine
1155 * @napi: napi struct with our devices info in it
1156 * @budget: amount of work driver is allowed to do this pass, in packets
1157 *
1158 * This function will clean more than one rx queue associated with a
1159 * q_vector.
1160 **/
1161static int ixgbe_clean_rxonly_many(struct napi_struct *napi, int budget)
1162{
1163 struct ixgbe_q_vector *q_vector =
1164 container_of(napi, struct ixgbe_q_vector, napi);
1165 struct ixgbe_adapter *adapter = q_vector->adapter;
1166 struct ixgbe_ring *rx_ring = NULL;
1167 int work_done = 0, i;
1168 long r_idx;
1169 u16 enable_mask = 0;
1170
1171 /* attempt to distribute budget to each queue fairly, but don't allow
1172 * the budget to go below 1 because we'll exit polling */
1173 budget /= (q_vector->rxr_count ?: 1);
1174 budget = max(budget, 1);
1175 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1176 for (i = 0; i < q_vector->rxr_count; i++) {
1177 rx_ring = &(adapter->rx_ring[r_idx]);
1178#ifdef CONFIG_IXGBE_DCA
1179 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1180 ixgbe_update_rx_dca(adapter, rx_ring);
1181#endif
1182 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1183 enable_mask |= rx_ring->v_idx;
1184 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1185 r_idx + 1);
1186 }
1187
1188 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1189 rx_ring = &(adapter->rx_ring[r_idx]);
1190 /* If all Rx work done, exit the polling mode */
1191 if (work_done < budget) {
1192 napi_complete(napi);
1193 if (adapter->itr_setting & 3)
1194 ixgbe_set_itr_msix(q_vector);
1195 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1196 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, enable_mask);
1197 return 0;
1198 }
1199
1200 return work_done;
1201}
1202static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1203 int r_idx)
1204{
1205 a->q_vector[v_idx].adapter = a;
1206 set_bit(r_idx, a->q_vector[v_idx].rxr_idx);
1207 a->q_vector[v_idx].rxr_count++;
1208 a->rx_ring[r_idx].v_idx = 1 << v_idx;
1209}
1210
1211static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1212 int r_idx)
1213{
1214 a->q_vector[v_idx].adapter = a;
1215 set_bit(r_idx, a->q_vector[v_idx].txr_idx);
1216 a->q_vector[v_idx].txr_count++;
1217 a->tx_ring[r_idx].v_idx = 1 << v_idx;
1218}
1219
1220/**
1221 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1222 * @adapter: board private structure to initialize
1223 * @vectors: allotted vector count for descriptor rings
1224 *
1225 * This function maps descriptor rings to the queue-specific vectors
1226 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1227 * one vector per ring/queue, but on a constrained vector budget, we
1228 * group the rings as "efficiently" as possible. You would add new
1229 * mapping configurations in here.
1230 **/
1231static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1232 int vectors)
1233{
1234 int v_start = 0;
1235 int rxr_idx = 0, txr_idx = 0;
1236 int rxr_remaining = adapter->num_rx_queues;
1237 int txr_remaining = adapter->num_tx_queues;
1238 int i, j;
1239 int rqpv, tqpv;
1240 int err = 0;
1241
1242 /* No mapping required if MSI-X is disabled. */
1243 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1244 goto out;
1245
1246 /*
1247 * The ideal configuration...
1248 * We have enough vectors to map one per queue.
1249 */
1250 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1251 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1252 map_vector_to_rxq(adapter, v_start, rxr_idx);
1253
1254 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1255 map_vector_to_txq(adapter, v_start, txr_idx);
1256
1257 goto out;
1258 }
1259
1260 /*
1261 * If we don't have enough vectors for a 1-to-1
1262 * mapping, we'll have to group them so there are
1263 * multiple queues per vector.
1264 */
1265 /* Re-adjusting *qpv takes care of the remainder. */
1266 for (i = v_start; i < vectors; i++) {
1267 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1268 for (j = 0; j < rqpv; j++) {
1269 map_vector_to_rxq(adapter, i, rxr_idx);
1270 rxr_idx++;
1271 rxr_remaining--;
1272 }
1273 }
1274 for (i = v_start; i < vectors; i++) {
1275 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1276 for (j = 0; j < tqpv; j++) {
1277 map_vector_to_txq(adapter, i, txr_idx);
1278 txr_idx++;
1279 txr_remaining--;
1280 }
1281 }
1282
1283out:
1284 return err;
1285}
1286
1287/**
1288 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1289 * @adapter: board private structure
1290 *
1291 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1292 * interrupts from the kernel.
1293 **/
1294static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1295{
1296 struct net_device *netdev = adapter->netdev;
1297 irqreturn_t (*handler)(int, void *);
1298 int i, vector, q_vectors, err;
1299 int ri=0, ti=0;
1300
1301 /* Decrement for Other and TCP Timer vectors */
1302 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1303
1304 /* Map the Tx/Rx rings to the vectors we were allotted. */
1305 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1306 if (err)
1307 goto out;
1308
1309#define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1310 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1311 &ixgbe_msix_clean_many)
1312 for (vector = 0; vector < q_vectors; vector++) {
1313 handler = SET_HANDLER(&adapter->q_vector[vector]);
1314
1315 if(handler == &ixgbe_msix_clean_rx) {
1316 sprintf(adapter->name[vector], "%s-%s-%d",
1317 netdev->name, "rx", ri++);
1318 }
1319 else if(handler == &ixgbe_msix_clean_tx) {
1320 sprintf(adapter->name[vector], "%s-%s-%d",
1321 netdev->name, "tx", ti++);
1322 }
1323 else
1324 sprintf(adapter->name[vector], "%s-%s-%d",
1325 netdev->name, "TxRx", vector);
1326
1327 err = request_irq(adapter->msix_entries[vector].vector,
1328 handler, 0, adapter->name[vector],
1329 &(adapter->q_vector[vector]));
1330 if (err) {
1331 DPRINTK(PROBE, ERR,
1332 "request_irq failed for MSIX interrupt "
1333 "Error: %d\n", err);
1334 goto free_queue_irqs;
1335 }
1336 }
1337
1338 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1339 err = request_irq(adapter->msix_entries[vector].vector,
1340 &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1341 if (err) {
1342 DPRINTK(PROBE, ERR,
1343 "request_irq for msix_lsc failed: %d\n", err);
1344 goto free_queue_irqs;
1345 }
1346
1347 return 0;
1348
1349free_queue_irqs:
1350 for (i = vector - 1; i >= 0; i--)
1351 free_irq(adapter->msix_entries[--vector].vector,
1352 &(adapter->q_vector[i]));
1353 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1354 pci_disable_msix(adapter->pdev);
1355 kfree(adapter->msix_entries);
1356 adapter->msix_entries = NULL;
1357out:
1358 return err;
1359}
1360
1361static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1362{
1363 struct ixgbe_hw *hw = &adapter->hw;
1364 struct ixgbe_q_vector *q_vector = adapter->q_vector;
1365 u8 current_itr;
1366 u32 new_itr = q_vector->eitr;
1367 struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1368 struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1369
1370 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1371 q_vector->tx_itr,
1372 tx_ring->total_packets,
1373 tx_ring->total_bytes);
1374 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1375 q_vector->rx_itr,
1376 rx_ring->total_packets,
1377 rx_ring->total_bytes);
1378
1379 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1380
1381 switch (current_itr) {
1382 /* counts and packets in update_itr are dependent on these numbers */
1383 case lowest_latency:
1384 new_itr = 100000;
1385 break;
1386 case low_latency:
1387 new_itr = 20000; /* aka hwitr = ~200 */
1388 break;
1389 case bulk_latency:
1390 new_itr = 8000;
1391 break;
1392 default:
1393 break;
1394 }
1395
1396 if (new_itr != q_vector->eitr) {
1397 u32 itr_reg;
1398 /* do an exponential smoothing */
1399 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1400 q_vector->eitr = new_itr;
1401 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
1402 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1403 /* Resolution is 2 usec on 82599, so halve the rate */
1404 itr_reg >>= 1;
1405 /* must write high and low 16 bits to reset counter */
1406 IXGBE_WRITE_REG(hw, IXGBE_EITR(0), itr_reg | (itr_reg)<<16);
1407 }
1408
1409 return;
1410}
1411
1412/**
1413 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1414 * @adapter: board private structure
1415 **/
1416static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1417{
1418 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1419 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1420 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
1421 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(2), ~0);
1422 }
1423 IXGBE_WRITE_FLUSH(&adapter->hw);
1424 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1425 int i;
1426 for (i = 0; i < adapter->num_msix_vectors; i++)
1427 synchronize_irq(adapter->msix_entries[i].vector);
1428 } else {
1429 synchronize_irq(adapter->pdev->irq);
1430 }
1431}
1432
1433/**
1434 * ixgbe_irq_enable - Enable default interrupt generation settings
1435 * @adapter: board private structure
1436 **/
1437static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1438{
1439 u32 mask;
1440 mask = IXGBE_EIMS_ENABLE_MASK;
1441 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1442 mask |= IXGBE_EIMS_GPI_SDP1;
1443 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1444 mask |= IXGBE_EIMS_GPI_SDP1;
1445 mask |= IXGBE_EIMS_GPI_SDP2;
1446 }
1447
1448 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1449 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1450 /* enable the rest of the queue vectors */
1451 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1),
1452 (IXGBE_EIMS_RTX_QUEUE << 16));
1453 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(2),
1454 ((IXGBE_EIMS_RTX_QUEUE << 16) |
1455 IXGBE_EIMS_RTX_QUEUE));
1456 }
1457 IXGBE_WRITE_FLUSH(&adapter->hw);
1458}
1459
1460/**
1461 * ixgbe_intr - legacy mode Interrupt Handler
1462 * @irq: interrupt number
1463 * @data: pointer to a network interface device structure
1464 **/
1465static irqreturn_t ixgbe_intr(int irq, void *data)
1466{
1467 struct net_device *netdev = data;
1468 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1469 struct ixgbe_hw *hw = &adapter->hw;
1470 u32 eicr;
1471
1472 /*
1473 * Workaround for silicon errata. Mask the interrupts
1474 * before the read of EICR.
1475 */
1476 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1477
1478 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1479 * therefore no explict interrupt disable is necessary */
1480 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1481 if (!eicr) {
1482 /* shared interrupt alert!
1483 * make sure interrupts are enabled because the read will
1484 * have disabled interrupts due to EIAM */
1485 ixgbe_irq_enable(adapter);
1486 return IRQ_NONE; /* Not our interrupt */
1487 }
1488
1489 if (eicr & IXGBE_EICR_LSC)
1490 ixgbe_check_lsc(adapter);
1491
1492 if (hw->mac.type == ixgbe_mac_82599EB)
1493 ixgbe_check_sfp_event(adapter, eicr);
1494
1495 ixgbe_check_fan_failure(adapter, eicr);
1496
1497 if (napi_schedule_prep(&adapter->q_vector[0].napi)) {
1498 adapter->tx_ring[0].total_packets = 0;
1499 adapter->tx_ring[0].total_bytes = 0;
1500 adapter->rx_ring[0].total_packets = 0;
1501 adapter->rx_ring[0].total_bytes = 0;
1502 /* would disable interrupts here but EIAM disabled it */
1503 __napi_schedule(&adapter->q_vector[0].napi);
1504 }
1505
1506 return IRQ_HANDLED;
1507}
1508
1509static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1510{
1511 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1512
1513 for (i = 0; i < q_vectors; i++) {
1514 struct ixgbe_q_vector *q_vector = &adapter->q_vector[i];
1515 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1516 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1517 q_vector->rxr_count = 0;
1518 q_vector->txr_count = 0;
1519 }
1520}
1521
1522/**
1523 * ixgbe_request_irq - initialize interrupts
1524 * @adapter: board private structure
1525 *
1526 * Attempts to configure interrupts using the best available
1527 * capabilities of the hardware and kernel.
1528 **/
1529static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1530{
1531 struct net_device *netdev = adapter->netdev;
1532 int err;
1533
1534 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1535 err = ixgbe_request_msix_irqs(adapter);
1536 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1537 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1538 netdev->name, netdev);
1539 } else {
1540 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1541 netdev->name, netdev);
1542 }
1543
1544 if (err)
1545 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1546
1547 return err;
1548}
1549
1550static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1551{
1552 struct net_device *netdev = adapter->netdev;
1553
1554 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1555 int i, q_vectors;
1556
1557 q_vectors = adapter->num_msix_vectors;
1558
1559 i = q_vectors - 1;
1560 free_irq(adapter->msix_entries[i].vector, netdev);
1561
1562 i--;
1563 for (; i >= 0; i--) {
1564 free_irq(adapter->msix_entries[i].vector,
1565 &(adapter->q_vector[i]));
1566 }
1567
1568 ixgbe_reset_q_vectors(adapter);
1569 } else {
1570 free_irq(adapter->pdev->irq, netdev);
1571 }
1572}
1573
1574/**
1575 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1576 *
1577 **/
1578static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1579{
1580 struct ixgbe_hw *hw = &adapter->hw;
1581
1582 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1583 EITR_INTS_PER_SEC_TO_REG(adapter->eitr_param));
1584
1585 ixgbe_set_ivar(adapter, 0, 0, 0);
1586 ixgbe_set_ivar(adapter, 1, 0, 0);
1587
1588 map_vector_to_rxq(adapter, 0, 0);
1589 map_vector_to_txq(adapter, 0, 0);
1590
1591 DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1592}
1593
1594/**
1595 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1596 * @adapter: board private structure
1597 *
1598 * Configure the Tx unit of the MAC after a reset.
1599 **/
1600static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1601{
1602 u64 tdba;
1603 struct ixgbe_hw *hw = &adapter->hw;
1604 u32 i, j, tdlen, txctrl;
1605
1606 /* Setup the HW Tx Head and Tail descriptor pointers */
1607 for (i = 0; i < adapter->num_tx_queues; i++) {
1608 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1609 j = ring->reg_idx;
1610 tdba = ring->dma;
1611 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1612 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1613 (tdba & DMA_32BIT_MASK));
1614 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1615 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1616 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1617 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1618 adapter->tx_ring[i].head = IXGBE_TDH(j);
1619 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1620 /* Disable Tx Head Writeback RO bit, since this hoses
1621 * bookkeeping if things aren't delivered in order.
1622 */
1623 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1624 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1625 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1626 }
1627 if (hw->mac.type == ixgbe_mac_82599EB) {
1628 /* We enable 8 traffic classes, DCB only */
1629 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
1630 IXGBE_WRITE_REG(hw, IXGBE_MTQC, (IXGBE_MTQC_RT_ENA |
1631 IXGBE_MTQC_8TC_8TQ));
1632 }
1633}
1634
1635#define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1636
1637static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter, int index)
1638{
1639 struct ixgbe_ring *rx_ring;
1640 u32 srrctl;
1641 int queue0 = 0;
1642 unsigned long mask;
1643
1644 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1645 queue0 = index;
1646 } else {
1647 mask = (unsigned long) adapter->ring_feature[RING_F_RSS].mask;
1648 queue0 = index & mask;
1649 index = index & mask;
1650 }
1651
1652 rx_ring = &adapter->rx_ring[queue0];
1653
1654 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
1655
1656 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1657 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1658
1659 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1660 u16 bufsz = IXGBE_RXBUFFER_2048;
1661 /* grow the amount we can receive on large page machines */
1662 if (bufsz < (PAGE_SIZE / 2))
1663 bufsz = (PAGE_SIZE / 2);
1664 /* cap the bufsz at our largest descriptor size */
1665 bufsz = min((u16)IXGBE_MAX_RXBUFFER, bufsz);
1666
1667 srrctl |= bufsz >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1668 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1669 srrctl |= ((IXGBE_RX_HDR_SIZE <<
1670 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1671 IXGBE_SRRCTL_BSIZEHDR_MASK);
1672 } else {
1673 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1674
1675 if (rx_ring->rx_buf_len == MAXIMUM_ETHERNET_VLAN_SIZE)
1676 srrctl |= IXGBE_RXBUFFER_2048 >>
1677 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1678 else
1679 srrctl |= rx_ring->rx_buf_len >>
1680 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1681 }
1682
1683 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
1684}
1685
1686/**
1687 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
1688 * @adapter: board private structure
1689 *
1690 * Configure the Rx unit of the MAC after a reset.
1691 **/
1692static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
1693{
1694 u64 rdba;
1695 struct ixgbe_hw *hw = &adapter->hw;
1696 struct net_device *netdev = adapter->netdev;
1697 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1698 int i, j;
1699 u32 rdlen, rxctrl, rxcsum;
1700 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
1701 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
1702 0x6A3E67EA, 0x14364D17, 0x3BED200D};
1703 u32 fctrl, hlreg0;
1704 u32 reta = 0, mrqc;
1705 u32 rdrxctl;
1706 int rx_buf_len;
1707
1708 /* Decide whether to use packet split mode or not */
1709 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
1710
1711 /* Set the RX buffer length according to the mode */
1712 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1713 rx_buf_len = IXGBE_RX_HDR_SIZE;
1714 if (hw->mac.type == ixgbe_mac_82599EB) {
1715 /* PSRTYPE must be initialized in 82599 */
1716 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
1717 IXGBE_PSRTYPE_UDPHDR |
1718 IXGBE_PSRTYPE_IPV4HDR |
1719 IXGBE_PSRTYPE_IPV6HDR;
1720 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), psrtype);
1721 }
1722 } else {
1723 if (netdev->mtu <= ETH_DATA_LEN)
1724 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1725 else
1726 rx_buf_len = ALIGN(max_frame, 1024);
1727 }
1728
1729 fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
1730 fctrl |= IXGBE_FCTRL_BAM;
1731 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
1732 fctrl |= IXGBE_FCTRL_PMCF;
1733 IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
1734
1735 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
1736 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1737 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
1738 else
1739 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
1740 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
1741
1742 rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1743 /* disable receives while setting up the descriptors */
1744 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
1745 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
1746
1747 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1748 * the Base and Length of the Rx Descriptor Ring */
1749 for (i = 0; i < adapter->num_rx_queues; i++) {
1750 rdba = adapter->rx_ring[i].dma;
1751 j = adapter->rx_ring[i].reg_idx;
1752 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_32BIT_MASK));
1753 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
1754 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
1755 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
1756 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
1757 adapter->rx_ring[i].head = IXGBE_RDH(j);
1758 adapter->rx_ring[i].tail = IXGBE_RDT(j);
1759 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1760
1761 ixgbe_configure_srrctl(adapter, j);
1762 }
1763
1764 if (hw->mac.type == ixgbe_mac_82598EB) {
1765 /*
1766 * For VMDq support of different descriptor types or
1767 * buffer sizes through the use of multiple SRRCTL
1768 * registers, RDRXCTL.MVMEN must be set to 1
1769 *
1770 * also, the manual doesn't mention it clearly but DCA hints
1771 * will only use queue 0's tags unless this bit is set. Side
1772 * effects of setting this bit are only that SRRCTL must be
1773 * fully programmed [0..15]
1774 */
1775 if (adapter->flags &
1776 (IXGBE_FLAG_RSS_ENABLED | IXGBE_FLAG_VMDQ_ENABLED)) {
1777 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
1778 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
1779 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
1780 }
1781 }
1782
1783 /* Program MRQC for the distribution of queues */
1784 if (hw->mac.type == ixgbe_mac_82599EB) {
1785 int mask = adapter->flags & (
1786 IXGBE_FLAG_RSS_ENABLED
1787 | IXGBE_FLAG_DCB_ENABLED
1788 );
1789
1790 switch (mask) {
1791 case (IXGBE_FLAG_RSS_ENABLED):
1792 mrqc = IXGBE_MRQC_RSSEN;
1793 break;
1794 case (IXGBE_FLAG_DCB_ENABLED):
1795 mrqc = IXGBE_MRQC_RT8TCEN;
1796 break;
1797 default:
1798 break;
1799 }
1800 }
1801 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
1802 /* Fill out redirection table */
1803 for (i = 0, j = 0; i < 128; i++, j++) {
1804 if (j == adapter->ring_feature[RING_F_RSS].indices)
1805 j = 0;
1806 /* reta = 4-byte sliding window of
1807 * 0x00..(indices-1)(indices-1)00..etc. */
1808 reta = (reta << 8) | (j * 0x11);
1809 if ((i & 3) == 3)
1810 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
1811 }
1812
1813 /* Fill out hash function seeds */
1814 for (i = 0; i < 10; i++)
1815 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
1816
1817 mrqc = IXGBE_MRQC_RSSEN
1818 /* Perform hash on these packet types */
1819 | IXGBE_MRQC_RSS_FIELD_IPV4
1820 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
1821 | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
1822 | IXGBE_MRQC_RSS_FIELD_IPV6
1823 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
1824 | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
1825 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
1826 }
1827
1828 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
1829
1830 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
1831 adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
1832 /* Disable indicating checksum in descriptor, enables
1833 * RSS hash */
1834 rxcsum |= IXGBE_RXCSUM_PCSD;
1835 }
1836 if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
1837 /* Enable IPv4 payload checksum for UDP fragments
1838 * if PCSD is not set */
1839 rxcsum |= IXGBE_RXCSUM_IPPCSE;
1840 }
1841
1842 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
1843
1844 if (hw->mac.type == ixgbe_mac_82599EB) {
1845 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
1846 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
1847 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
1848 }
1849}
1850
1851static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1852{
1853 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1854 struct ixgbe_hw *hw = &adapter->hw;
1855
1856 /* add VID to filter table */
1857 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
1858}
1859
1860static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1861{
1862 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1863 struct ixgbe_hw *hw = &adapter->hw;
1864
1865 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1866 ixgbe_irq_disable(adapter);
1867
1868 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1869
1870 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1871 ixgbe_irq_enable(adapter);
1872
1873 /* remove VID from filter table */
1874 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
1875}
1876
1877static void ixgbe_vlan_rx_register(struct net_device *netdev,
1878 struct vlan_group *grp)
1879{
1880 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1881 u32 ctrl;
1882 int i, j;
1883
1884 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1885 ixgbe_irq_disable(adapter);
1886 adapter->vlgrp = grp;
1887
1888 /*
1889 * For a DCB driver, always enable VLAN tag stripping so we can
1890 * still receive traffic from a DCB-enabled host even if we're
1891 * not in DCB mode.
1892 */
1893 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
1894 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1895 ctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
1896 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
1897 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
1898 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1899 ctrl |= IXGBE_VLNCTRL_VFE;
1900 /* enable VLAN tag insert/strip */
1901 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
1902 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
1903 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
1904 for (i = 0; i < adapter->num_rx_queues; i++) {
1905 j = adapter->rx_ring[i].reg_idx;
1906 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_RXDCTL(j));
1907 ctrl |= IXGBE_RXDCTL_VME;
1908 IXGBE_WRITE_REG(&adapter->hw, IXGBE_RXDCTL(j), ctrl);
1909 }
1910 }
1911 ixgbe_vlan_rx_add_vid(netdev, 0);
1912
1913 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1914 ixgbe_irq_enable(adapter);
1915}
1916
1917static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
1918{
1919 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1920
1921 if (adapter->vlgrp) {
1922 u16 vid;
1923 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1924 if (!vlan_group_get_device(adapter->vlgrp, vid))
1925 continue;
1926 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
1927 }
1928 }
1929}
1930
1931static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
1932{
1933 struct dev_mc_list *mc_ptr;
1934 u8 *addr = *mc_addr_ptr;
1935 *vmdq = 0;
1936
1937 mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
1938 if (mc_ptr->next)
1939 *mc_addr_ptr = mc_ptr->next->dmi_addr;
1940 else
1941 *mc_addr_ptr = NULL;
1942
1943 return addr;
1944}
1945
1946/**
1947 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
1948 * @netdev: network interface device structure
1949 *
1950 * The set_rx_method entry point is called whenever the unicast/multicast
1951 * address list or the network interface flags are updated. This routine is
1952 * responsible for configuring the hardware for proper unicast, multicast and
1953 * promiscuous mode.
1954 **/
1955static void ixgbe_set_rx_mode(struct net_device *netdev)
1956{
1957 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1958 struct ixgbe_hw *hw = &adapter->hw;
1959 u32 fctrl, vlnctrl;
1960 u8 *addr_list = NULL;
1961 int addr_count = 0;
1962
1963 /* Check for Promiscuous and All Multicast modes */
1964
1965 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
1966 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
1967
1968 if (netdev->flags & IFF_PROMISC) {
1969 hw->addr_ctrl.user_set_promisc = 1;
1970 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
1971 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
1972 } else {
1973 if (netdev->flags & IFF_ALLMULTI) {
1974 fctrl |= IXGBE_FCTRL_MPE;
1975 fctrl &= ~IXGBE_FCTRL_UPE;
1976 } else {
1977 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
1978 }
1979 vlnctrl |= IXGBE_VLNCTRL_VFE;
1980 hw->addr_ctrl.user_set_promisc = 0;
1981 }
1982
1983 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
1984 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
1985
1986 /* reprogram secondary unicast list */
1987 addr_count = netdev->uc_count;
1988 if (addr_count)
1989 addr_list = netdev->uc_list->dmi_addr;
1990 hw->mac.ops.update_uc_addr_list(hw, addr_list, addr_count,
1991 ixgbe_addr_list_itr);
1992
1993 /* reprogram multicast list */
1994 addr_count = netdev->mc_count;
1995 if (addr_count)
1996 addr_list = netdev->mc_list->dmi_addr;
1997 hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
1998 ixgbe_addr_list_itr);
1999}
2000
2001static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
2002{
2003 int q_idx;
2004 struct ixgbe_q_vector *q_vector;
2005 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2006
2007 /* legacy and MSI only use one vector */
2008 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2009 q_vectors = 1;
2010
2011 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2012 struct napi_struct *napi;
2013 q_vector = &adapter->q_vector[q_idx];
2014 if (!q_vector->rxr_count)
2015 continue;
2016 napi = &q_vector->napi;
2017 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) &&
2018 (q_vector->rxr_count > 1))
2019 napi->poll = &ixgbe_clean_rxonly_many;
2020
2021 napi_enable(napi);
2022 }
2023}
2024
2025static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2026{
2027 int q_idx;
2028 struct ixgbe_q_vector *q_vector;
2029 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2030
2031 /* legacy and MSI only use one vector */
2032 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2033 q_vectors = 1;
2034
2035 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2036 q_vector = &adapter->q_vector[q_idx];
2037 if (!q_vector->rxr_count)
2038 continue;
2039 napi_disable(&q_vector->napi);
2040 }
2041}
2042
2043#ifdef CONFIG_IXGBE_DCB
2044/*
2045 * ixgbe_configure_dcb - Configure DCB hardware
2046 * @adapter: ixgbe adapter struct
2047 *
2048 * This is called by the driver on open to configure the DCB hardware.
2049 * This is also called by the gennetlink interface when reconfiguring
2050 * the DCB state.
2051 */
2052static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2053{
2054 struct ixgbe_hw *hw = &adapter->hw;
2055 u32 txdctl, vlnctrl;
2056 int i, j;
2057
2058 ixgbe_dcb_check_config(&adapter->dcb_cfg);
2059 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2060 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2061
2062 /* reconfigure the hardware */
2063 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
2064
2065 for (i = 0; i < adapter->num_tx_queues; i++) {
2066 j = adapter->tx_ring[i].reg_idx;
2067 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2068 /* PThresh workaround for Tx hang with DFP enabled. */
2069 txdctl |= 32;
2070 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2071 }
2072 /* Enable VLAN tag insert/strip */
2073 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2074 if (hw->mac.type == ixgbe_mac_82598EB) {
2075 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2076 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2077 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2078 } else if (hw->mac.type == ixgbe_mac_82599EB) {
2079 vlnctrl |= IXGBE_VLNCTRL_VFE;
2080 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2081 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2082 for (i = 0; i < adapter->num_rx_queues; i++) {
2083 j = adapter->rx_ring[i].reg_idx;
2084 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2085 vlnctrl |= IXGBE_RXDCTL_VME;
2086 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2087 }
2088 }
2089 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2090}
2091
2092#endif
2093static void ixgbe_configure(struct ixgbe_adapter *adapter)
2094{
2095 struct net_device *netdev = adapter->netdev;
2096 int i;
2097
2098 ixgbe_set_rx_mode(netdev);
2099
2100 ixgbe_restore_vlan(adapter);
2101#ifdef CONFIG_IXGBE_DCB
2102 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2103 netif_set_gso_max_size(netdev, 32768);
2104 ixgbe_configure_dcb(adapter);
2105 } else {
2106 netif_set_gso_max_size(netdev, 65536);
2107 }
2108#else
2109 netif_set_gso_max_size(netdev, 65536);
2110#endif
2111
2112 ixgbe_configure_tx(adapter);
2113 ixgbe_configure_rx(adapter);
2114 for (i = 0; i < adapter->num_rx_queues; i++)
2115 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
2116 (adapter->rx_ring[i].count - 1));
2117}
2118
2119static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2120{
2121 switch (hw->phy.type) {
2122 case ixgbe_phy_sfp_avago:
2123 case ixgbe_phy_sfp_ftl:
2124 case ixgbe_phy_sfp_intel:
2125 case ixgbe_phy_sfp_unknown:
2126 case ixgbe_phy_tw_tyco:
2127 case ixgbe_phy_tw_unknown:
2128 return true;
2129 default:
2130 return false;
2131 }
2132}
2133
2134/**
2135 * ixgbe_sfp_link_config - set up SFP+ link
2136 * @adapter: pointer to private adapter struct
2137 **/
2138static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2139{
2140 struct ixgbe_hw *hw = &adapter->hw;
2141
2142 if (hw->phy.multispeed_fiber) {
2143 /*
2144 * In multispeed fiber setups, the device may not have
2145 * had a physical connection when the driver loaded.
2146 * If that's the case, the initial link configuration
2147 * couldn't get the MAC into 10G or 1G mode, so we'll
2148 * never have a link status change interrupt fire.
2149 * We need to try and force an autonegotiation
2150 * session, then bring up link.
2151 */
2152 hw->mac.ops.setup_sfp(hw);
2153 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2154 schedule_work(&adapter->multispeed_fiber_task);
2155 } else {
2156 /*
2157 * Direct Attach Cu and non-multispeed fiber modules
2158 * still need to be configured properly prior to
2159 * attempting link.
2160 */
2161 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2162 schedule_work(&adapter->sfp_config_module_task);
2163 }
2164}
2165
2166/**
2167 * ixgbe_non_sfp_link_config - set up non-SFP+ link
2168 * @hw: pointer to private hardware struct
2169 *
2170 * Returns 0 on success, negative on failure
2171 **/
2172static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2173{
2174 u32 autoneg;
2175 bool link_up = false;
2176 u32 ret = IXGBE_ERR_LINK_SETUP;
2177
2178 if (hw->mac.ops.check_link)
2179 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2180
2181 if (ret)
2182 goto link_cfg_out;
2183
2184 if (hw->mac.ops.get_link_capabilities)
2185 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
2186 &hw->mac.autoneg);
2187 if (ret)
2188 goto link_cfg_out;
2189
2190 if (hw->mac.ops.setup_link_speed)
2191 ret = hw->mac.ops.setup_link_speed(hw, autoneg, true, link_up);
2192link_cfg_out:
2193 return ret;
2194}
2195
2196#define IXGBE_MAX_RX_DESC_POLL 10
2197static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2198 int rxr)
2199{
2200 int j = adapter->rx_ring[rxr].reg_idx;
2201 int k;
2202
2203 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2204 if (IXGBE_READ_REG(&adapter->hw,
2205 IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2206 break;
2207 else
2208 msleep(1);
2209 }
2210 if (k >= IXGBE_MAX_RX_DESC_POLL) {
2211 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2212 "not set within the polling period\n", rxr);
2213 }
2214 ixgbe_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
2215 (adapter->rx_ring[rxr].count - 1));
2216}
2217
2218static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2219{
2220 struct net_device *netdev = adapter->netdev;
2221 struct ixgbe_hw *hw = &adapter->hw;
2222 int i, j = 0;
2223 int num_rx_rings = adapter->num_rx_queues;
2224 int err;
2225 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2226 u32 txdctl, rxdctl, mhadd;
2227 u32 dmatxctl;
2228 u32 gpie;
2229
2230 ixgbe_get_hw_control(adapter);
2231
2232 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2233 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2234 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2235 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2236 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2237 } else {
2238 /* MSI only */
2239 gpie = 0;
2240 }
2241 /* XXX: to interrupt immediately for EICS writes, enable this */
2242 /* gpie |= IXGBE_GPIE_EIMEN; */
2243 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2244 }
2245
2246 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2247 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2248 * specifically only auto mask tx and rx interrupts */
2249 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2250 }
2251
2252 /* Enable fan failure interrupt if media type is copper */
2253 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2254 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2255 gpie |= IXGBE_SDP1_GPIEN;
2256 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2257 }
2258
2259 if (hw->mac.type == ixgbe_mac_82599EB) {
2260 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2261 gpie |= IXGBE_SDP1_GPIEN;
2262 gpie |= IXGBE_SDP2_GPIEN;
2263 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2264 }
2265
2266 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2267 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2268 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2269 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2270
2271 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2272 }
2273
2274 for (i = 0; i < adapter->num_tx_queues; i++) {
2275 j = adapter->tx_ring[i].reg_idx;
2276 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2277 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2278 txdctl |= (8 << 16);
2279 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2280 }
2281
2282 if (hw->mac.type == ixgbe_mac_82599EB) {
2283 /* DMATXCTL.EN must be set after all Tx queue config is done */
2284 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2285 dmatxctl |= IXGBE_DMATXCTL_TE;
2286 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2287 }
2288 for (i = 0; i < adapter->num_tx_queues; i++) {
2289 j = adapter->tx_ring[i].reg_idx;
2290 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2291 txdctl |= IXGBE_TXDCTL_ENABLE;
2292 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2293 }
2294
2295 for (i = 0; i < num_rx_rings; i++) {
2296 j = adapter->rx_ring[i].reg_idx;
2297 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2298 /* enable PTHRESH=32 descriptors (half the internal cache)
2299 * and HTHRESH=0 descriptors (to minimize latency on fetch),
2300 * this also removes a pesky rx_no_buffer_count increment */
2301 rxdctl |= 0x0020;
2302 rxdctl |= IXGBE_RXDCTL_ENABLE;
2303 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2304 if (hw->mac.type == ixgbe_mac_82599EB)
2305 ixgbe_rx_desc_queue_enable(adapter, i);
2306 }
2307 /* enable all receives */
2308 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2309 if (hw->mac.type == ixgbe_mac_82598EB)
2310 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2311 else
2312 rxdctl |= IXGBE_RXCTRL_RXEN;
2313 hw->mac.ops.enable_rx_dma(hw, rxdctl);
2314
2315 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2316 ixgbe_configure_msix(adapter);
2317 else
2318 ixgbe_configure_msi_and_legacy(adapter);
2319
2320 ixgbe_napi_add_all(adapter);
2321
2322 clear_bit(__IXGBE_DOWN, &adapter->state);
2323 ixgbe_napi_enable_all(adapter);
2324
2325 /* clear any pending interrupts, may auto mask */
2326 IXGBE_READ_REG(hw, IXGBE_EICR);
2327
2328 ixgbe_irq_enable(adapter);
2329
2330 /*
2331 * For hot-pluggable SFP+ devices, a new SFP+ module may have
2332 * arrived before interrupts were enabled. We need to kick off
2333 * the SFP+ module setup first, then try to bring up link.
2334 * If we're not hot-pluggable SFP+, we just need to configure link
2335 * and bring it up.
2336 */
2337 err = hw->phy.ops.identify(hw);
2338 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
2339 DPRINTK(PROBE, ERR, "PHY not supported on this NIC %d\n", err);
2340 ixgbe_down(adapter);
2341 return err;
2342 }
2343
2344 if (ixgbe_is_sfp(hw)) {
2345 ixgbe_sfp_link_config(adapter);
2346 } else {
2347 err = ixgbe_non_sfp_link_config(hw);
2348 if (err)
2349 DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
2350 }
2351
2352 /* enable transmits */
2353 netif_tx_start_all_queues(netdev);
2354
2355 /* bring the link up in the watchdog, this could race with our first
2356 * link up interrupt but shouldn't be a problem */
2357 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2358 adapter->link_check_timeout = jiffies;
2359 mod_timer(&adapter->watchdog_timer, jiffies);
2360 return 0;
2361}
2362
2363void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
2364{
2365 WARN_ON(in_interrupt());
2366 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
2367 msleep(1);
2368 ixgbe_down(adapter);
2369 ixgbe_up(adapter);
2370 clear_bit(__IXGBE_RESETTING, &adapter->state);
2371}
2372
2373int ixgbe_up(struct ixgbe_adapter *adapter)
2374{
2375 /* hardware has been reset, we need to reload some things */
2376 ixgbe_configure(adapter);
2377
2378 return ixgbe_up_complete(adapter);
2379}
2380
2381void ixgbe_reset(struct ixgbe_adapter *adapter)
2382{
2383 struct ixgbe_hw *hw = &adapter->hw;
2384 if (hw->mac.ops.init_hw(hw))
2385 dev_err(&adapter->pdev->dev, "Hardware Error\n");
2386
2387 /* reprogram the RAR[0] in case user changed it. */
2388 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2389
2390}
2391
2392/**
2393 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2394 * @adapter: board private structure
2395 * @rx_ring: ring to free buffers from
2396 **/
2397static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2398 struct ixgbe_ring *rx_ring)
2399{
2400 struct pci_dev *pdev = adapter->pdev;
2401 unsigned long size;
2402 unsigned int i;
2403
2404 /* Free all the Rx ring sk_buffs */
2405
2406 for (i = 0; i < rx_ring->count; i++) {
2407 struct ixgbe_rx_buffer *rx_buffer_info;
2408
2409 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2410 if (rx_buffer_info->dma) {
2411 pci_unmap_single(pdev, rx_buffer_info->dma,
2412 rx_ring->rx_buf_len,
2413 PCI_DMA_FROMDEVICE);
2414 rx_buffer_info->dma = 0;
2415 }
2416 if (rx_buffer_info->skb) {
2417 dev_kfree_skb(rx_buffer_info->skb);
2418 rx_buffer_info->skb = NULL;
2419 }
2420 if (!rx_buffer_info->page)
2421 continue;
2422 pci_unmap_page(pdev, rx_buffer_info->page_dma, PAGE_SIZE / 2,
2423 PCI_DMA_FROMDEVICE);
2424 rx_buffer_info->page_dma = 0;
2425 put_page(rx_buffer_info->page);
2426 rx_buffer_info->page = NULL;
2427 rx_buffer_info->page_offset = 0;
2428 }
2429
2430 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2431 memset(rx_ring->rx_buffer_info, 0, size);
2432
2433 /* Zero out the descriptor ring */
2434 memset(rx_ring->desc, 0, rx_ring->size);
2435
2436 rx_ring->next_to_clean = 0;
2437 rx_ring->next_to_use = 0;
2438
2439 writel(0, adapter->hw.hw_addr + rx_ring->head);
2440 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2441}
2442
2443/**
2444 * ixgbe_clean_tx_ring - Free Tx Buffers
2445 * @adapter: board private structure
2446 * @tx_ring: ring to be cleaned
2447 **/
2448static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2449 struct ixgbe_ring *tx_ring)
2450{
2451 struct ixgbe_tx_buffer *tx_buffer_info;
2452 unsigned long size;
2453 unsigned int i;
2454
2455 /* Free all the Tx ring sk_buffs */
2456
2457 for (i = 0; i < tx_ring->count; i++) {
2458 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2459 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2460 }
2461
2462 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2463 memset(tx_ring->tx_buffer_info, 0, size);
2464
2465 /* Zero out the descriptor ring */
2466 memset(tx_ring->desc, 0, tx_ring->size);
2467
2468 tx_ring->next_to_use = 0;
2469 tx_ring->next_to_clean = 0;
2470
2471 writel(0, adapter->hw.hw_addr + tx_ring->head);
2472 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2473}
2474
2475/**
2476 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2477 * @adapter: board private structure
2478 **/
2479static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2480{
2481 int i;
2482
2483 for (i = 0; i < adapter->num_rx_queues; i++)
2484 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2485}
2486
2487/**
2488 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2489 * @adapter: board private structure
2490 **/
2491static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2492{
2493 int i;
2494
2495 for (i = 0; i < adapter->num_tx_queues; i++)
2496 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2497}
2498
2499void ixgbe_down(struct ixgbe_adapter *adapter)
2500{
2501 struct net_device *netdev = adapter->netdev;
2502 struct ixgbe_hw *hw = &adapter->hw;
2503 u32 rxctrl;
2504 u32 txdctl;
2505 int i, j;
2506
2507 /* signal that we are down to the interrupt handler */
2508 set_bit(__IXGBE_DOWN, &adapter->state);
2509
2510 /* disable receives */
2511 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2512 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2513
2514 netif_tx_disable(netdev);
2515
2516 IXGBE_WRITE_FLUSH(hw);
2517 msleep(10);
2518
2519 netif_tx_stop_all_queues(netdev);
2520
2521 ixgbe_irq_disable(adapter);
2522
2523 ixgbe_napi_disable_all(adapter);
2524
2525 del_timer_sync(&adapter->watchdog_timer);
2526 cancel_work_sync(&adapter->watchdog_task);
2527
2528 /* disable transmits in the hardware now that interrupts are off */
2529 for (i = 0; i < adapter->num_tx_queues; i++) {
2530 j = adapter->tx_ring[i].reg_idx;
2531 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2532 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
2533 (txdctl & ~IXGBE_TXDCTL_ENABLE));
2534 }
2535
2536 netif_carrier_off(netdev);
2537
2538#ifdef CONFIG_IXGBE_DCA
2539 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2540 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
2541 dca_remove_requester(&adapter->pdev->dev);
2542 }
2543
2544#endif
2545 if (!pci_channel_offline(adapter->pdev))
2546 ixgbe_reset(adapter);
2547 ixgbe_clean_all_tx_rings(adapter);
2548 ixgbe_clean_all_rx_rings(adapter);
2549
2550#ifdef CONFIG_IXGBE_DCA
2551 /* since we reset the hardware DCA settings were cleared */
2552 if (dca_add_requester(&adapter->pdev->dev) == 0) {
2553 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
2554 /* always use CB2 mode, difference is masked
2555 * in the CB driver */
2556 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
2557 ixgbe_setup_dca(adapter);
2558 }
2559#endif
2560}
2561
2562/**
2563 * ixgbe_poll - NAPI Rx polling callback
2564 * @napi: structure for representing this polling device
2565 * @budget: how many packets driver is allowed to clean
2566 *
2567 * This function is used for legacy and MSI, NAPI mode
2568 **/
2569static int ixgbe_poll(struct napi_struct *napi, int budget)
2570{
2571 struct ixgbe_q_vector *q_vector = container_of(napi,
2572 struct ixgbe_q_vector, napi);
2573 struct ixgbe_adapter *adapter = q_vector->adapter;
2574 int tx_cleaned, work_done = 0;
2575
2576#ifdef CONFIG_IXGBE_DCA
2577 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2578 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
2579 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
2580 }
2581#endif
2582
2583 tx_cleaned = ixgbe_clean_tx_irq(adapter, adapter->tx_ring);
2584 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring, &work_done, budget);
2585
2586 if (tx_cleaned)
2587 work_done = budget;
2588
2589 /* If budget not fully consumed, exit the polling mode */
2590 if (work_done < budget) {
2591 napi_complete(napi);
2592 if (adapter->itr_setting & 3)
2593 ixgbe_set_itr(adapter);
2594 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2595 ixgbe_irq_enable(adapter);
2596 }
2597 return work_done;
2598}
2599
2600/**
2601 * ixgbe_tx_timeout - Respond to a Tx Hang
2602 * @netdev: network interface device structure
2603 **/
2604static void ixgbe_tx_timeout(struct net_device *netdev)
2605{
2606 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2607
2608 /* Do the reset outside of interrupt context */
2609 schedule_work(&adapter->reset_task);
2610}
2611
2612static void ixgbe_reset_task(struct work_struct *work)
2613{
2614 struct ixgbe_adapter *adapter;
2615 adapter = container_of(work, struct ixgbe_adapter, reset_task);
2616
2617 /* If we're already down or resetting, just bail */
2618 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
2619 test_bit(__IXGBE_RESETTING, &adapter->state))
2620 return;
2621
2622 adapter->tx_timeout_count++;
2623
2624 ixgbe_reinit_locked(adapter);
2625}
2626
2627#ifdef CONFIG_IXGBE_DCB
2628static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
2629{
2630 bool ret = false;
2631
2632 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2633 adapter->ring_feature[RING_F_DCB].mask = 0x7 << 3;
2634 adapter->num_rx_queues =
2635 adapter->ring_feature[RING_F_DCB].indices;
2636 adapter->num_tx_queues =
2637 adapter->ring_feature[RING_F_DCB].indices;
2638 ret = true;
2639 } else {
2640 ret = false;
2641 }
2642
2643 return ret;
2644}
2645#endif
2646
2647static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
2648{
2649 bool ret = false;
2650
2651 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2652 adapter->ring_feature[RING_F_RSS].mask = 0xF;
2653 adapter->num_rx_queues =
2654 adapter->ring_feature[RING_F_RSS].indices;
2655 adapter->num_tx_queues =
2656 adapter->ring_feature[RING_F_RSS].indices;
2657 ret = true;
2658 } else {
2659 ret = false;
2660 }
2661
2662 return ret;
2663}
2664
2665static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
2666{
2667 /* Start with base case */
2668 adapter->num_rx_queues = 1;
2669 adapter->num_tx_queues = 1;
2670
2671#ifdef CONFIG_IXGBE_DCB
2672 if (ixgbe_set_dcb_queues(adapter))
2673 return;
2674
2675#endif
2676 if (ixgbe_set_rss_queues(adapter))
2677 return;
2678}
2679
2680static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
2681 int vectors)
2682{
2683 int err, vector_threshold;
2684
2685 /* We'll want at least 3 (vector_threshold):
2686 * 1) TxQ[0] Cleanup
2687 * 2) RxQ[0] Cleanup
2688 * 3) Other (Link Status Change, etc.)
2689 * 4) TCP Timer (optional)
2690 */
2691 vector_threshold = MIN_MSIX_COUNT;
2692
2693 /* The more we get, the more we will assign to Tx/Rx Cleanup
2694 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2695 * Right now, we simply care about how many we'll get; we'll
2696 * set them up later while requesting irq's.
2697 */
2698 while (vectors >= vector_threshold) {
2699 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
2700 vectors);
2701 if (!err) /* Success in acquiring all requested vectors. */
2702 break;
2703 else if (err < 0)
2704 vectors = 0; /* Nasty failure, quit now */
2705 else /* err == number of vectors we should try again with */
2706 vectors = err;
2707 }
2708
2709 if (vectors < vector_threshold) {
2710 /* Can't allocate enough MSI-X interrupts? Oh well.
2711 * This just means we'll go with either a single MSI
2712 * vector or fall back to legacy interrupts.
2713 */
2714 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
2715 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2716 kfree(adapter->msix_entries);
2717 adapter->msix_entries = NULL;
2718 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
2719 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2720 ixgbe_set_num_queues(adapter);
2721 } else {
2722 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
2723 /*
2724 * Adjust for only the vectors we'll use, which is minimum
2725 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2726 * vectors we were allocated.
2727 */
2728 adapter->num_msix_vectors = min(vectors,
2729 adapter->max_msix_q_vectors + NON_Q_VECTORS);
2730 }
2731}
2732
2733/**
2734 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
2735 * @adapter: board private structure to initialize
2736 *
2737 * Cache the descriptor ring offsets for RSS to the assigned rings.
2738 *
2739 **/
2740static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
2741{
2742 int i;
2743 bool ret = false;
2744
2745 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2746 for (i = 0; i < adapter->num_rx_queues; i++)
2747 adapter->rx_ring[i].reg_idx = i;
2748 for (i = 0; i < adapter->num_tx_queues; i++)
2749 adapter->tx_ring[i].reg_idx = i;
2750 ret = true;
2751 } else {
2752 ret = false;
2753 }
2754
2755 return ret;
2756}
2757
2758#ifdef CONFIG_IXGBE_DCB
2759/**
2760 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
2761 * @adapter: board private structure to initialize
2762 *
2763 * Cache the descriptor ring offsets for DCB to the assigned rings.
2764 *
2765 **/
2766static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
2767{
2768 int i;
2769 bool ret = false;
2770 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
2771
2772 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2773 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2774 /* the number of queues is assumed to be symmetric */
2775 for (i = 0; i < dcb_i; i++) {
2776 adapter->rx_ring[i].reg_idx = i << 3;
2777 adapter->tx_ring[i].reg_idx = i << 2;
2778 }
2779 ret = true;
2780 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2781 for (i = 0; i < dcb_i; i++) {
2782 adapter->rx_ring[i].reg_idx = i << 4;
2783 adapter->tx_ring[i].reg_idx = i << 4;
2784 }
2785 ret = true;
2786 } else {
2787 ret = false;
2788 }
2789 } else {
2790 ret = false;
2791 }
2792
2793 return ret;
2794}
2795#endif
2796
2797/**
2798 * ixgbe_cache_ring_register - Descriptor ring to register mapping
2799 * @adapter: board private structure to initialize
2800 *
2801 * Once we know the feature-set enabled for the device, we'll cache
2802 * the register offset the descriptor ring is assigned to.
2803 *
2804 * Note, the order the various feature calls is important. It must start with
2805 * the "most" features enabled at the same time, then trickle down to the
2806 * least amount of features turned on at once.
2807 **/
2808static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
2809{
2810 /* start with default case */
2811 adapter->rx_ring[0].reg_idx = 0;
2812 adapter->tx_ring[0].reg_idx = 0;
2813
2814#ifdef CONFIG_IXGBE_DCB
2815 if (ixgbe_cache_ring_dcb(adapter))
2816 return;
2817
2818#endif
2819 if (ixgbe_cache_ring_rss(adapter))
2820 return;
2821}
2822
2823/**
2824 * ixgbe_alloc_queues - Allocate memory for all rings
2825 * @adapter: board private structure to initialize
2826 *
2827 * We allocate one ring per queue at run-time since we don't know the
2828 * number of queues at compile-time.
2829 **/
2830static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
2831{
2832 int i;
2833
2834 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
2835 sizeof(struct ixgbe_ring), GFP_KERNEL);
2836 if (!adapter->tx_ring)
2837 goto err_tx_ring_allocation;
2838
2839 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
2840 sizeof(struct ixgbe_ring), GFP_KERNEL);
2841 if (!adapter->rx_ring)
2842 goto err_rx_ring_allocation;
2843
2844 for (i = 0; i < adapter->num_tx_queues; i++) {
2845 adapter->tx_ring[i].count = adapter->tx_ring_count;
2846 adapter->tx_ring[i].queue_index = i;
2847 }
2848
2849 for (i = 0; i < adapter->num_rx_queues; i++) {
2850 adapter->rx_ring[i].count = adapter->rx_ring_count;
2851 adapter->rx_ring[i].queue_index = i;
2852 }
2853
2854 ixgbe_cache_ring_register(adapter);
2855
2856 return 0;
2857
2858err_rx_ring_allocation:
2859 kfree(adapter->tx_ring);
2860err_tx_ring_allocation:
2861 return -ENOMEM;
2862}
2863
2864/**
2865 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
2866 * @adapter: board private structure to initialize
2867 *
2868 * Attempt to configure the interrupts using the best available
2869 * capabilities of the hardware and the kernel.
2870 **/
2871static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
2872{
2873 int err = 0;
2874 int vector, v_budget;
2875
2876 /*
2877 * It's easy to be greedy for MSI-X vectors, but it really
2878 * doesn't do us much good if we have a lot more vectors
2879 * than CPU's. So let's be conservative and only ask for
2880 * (roughly) twice the number of vectors as there are CPU's.
2881 */
2882 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
2883 (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
2884
2885 /*
2886 * At the same time, hardware can only support a maximum of
2887 * MAX_MSIX_COUNT vectors. With features such as RSS and VMDq,
2888 * we can easily reach upwards of 64 Rx descriptor queues and
2889 * 32 Tx queues. Thus, we cap it off in those rare cases where
2890 * the cpu count also exceeds our vector limit.
2891 */
2892 v_budget = min(v_budget, MAX_MSIX_COUNT);
2893
2894 /* A failure in MSI-X entry allocation isn't fatal, but it does
2895 * mean we disable MSI-X capabilities of the adapter. */
2896 adapter->msix_entries = kcalloc(v_budget,
2897 sizeof(struct msix_entry), GFP_KERNEL);
2898 if (!adapter->msix_entries) {
2899 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
2900 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2901 ixgbe_set_num_queues(adapter);
2902 kfree(adapter->tx_ring);
2903 kfree(adapter->rx_ring);
2904 err = ixgbe_alloc_queues(adapter);
2905 if (err) {
2906 DPRINTK(PROBE, ERR, "Unable to allocate memory "
2907 "for queues\n");
2908 goto out;
2909 }
2910
2911 goto try_msi;
2912 }
2913
2914 for (vector = 0; vector < v_budget; vector++)
2915 adapter->msix_entries[vector].entry = vector;
2916
2917 ixgbe_acquire_msix_vectors(adapter, v_budget);
2918
2919 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2920 goto out;
2921
2922try_msi:
2923 err = pci_enable_msi(adapter->pdev);
2924 if (!err) {
2925 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
2926 } else {
2927 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
2928 "falling back to legacy. Error: %d\n", err);
2929 /* reset err */
2930 err = 0;
2931 }
2932
2933out:
2934 /* Notify the stack of the (possibly) reduced Tx Queue count. */
2935 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
2936
2937 return err;
2938}
2939
2940void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
2941{
2942 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2943 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2944 pci_disable_msix(adapter->pdev);
2945 kfree(adapter->msix_entries);
2946 adapter->msix_entries = NULL;
2947 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2948 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
2949 pci_disable_msi(adapter->pdev);
2950 }
2951 return;
2952}
2953
2954/**
2955 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
2956 * @adapter: board private structure to initialize
2957 *
2958 * We determine which interrupt scheme to use based on...
2959 * - Kernel support (MSI, MSI-X)
2960 * - which can be user-defined (via MODULE_PARAM)
2961 * - Hardware queue count (num_*_queues)
2962 * - defined by miscellaneous hardware support/features (RSS, etc.)
2963 **/
2964int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
2965{
2966 int err;
2967
2968 /* Number of supported queues */
2969 ixgbe_set_num_queues(adapter);
2970
2971 err = ixgbe_alloc_queues(adapter);
2972 if (err) {
2973 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
2974 goto err_alloc_queues;
2975 }
2976
2977 err = ixgbe_set_interrupt_capability(adapter);
2978 if (err) {
2979 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
2980 goto err_set_interrupt;
2981 }
2982
2983 DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
2984 "Tx Queue count = %u\n",
2985 (adapter->num_rx_queues > 1) ? "Enabled" :
2986 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2987
2988 set_bit(__IXGBE_DOWN, &adapter->state);
2989
2990 return 0;
2991
2992err_set_interrupt:
2993 kfree(adapter->tx_ring);
2994 kfree(adapter->rx_ring);
2995err_alloc_queues:
2996 return err;
2997}
2998
2999/**
3000 * ixgbe_sfp_timer - worker thread to find a missing module
3001 * @data: pointer to our adapter struct
3002 **/
3003static void ixgbe_sfp_timer(unsigned long data)
3004{
3005 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3006
3007 /* Do the sfp_timer outside of interrupt context due to the
3008 * delays that sfp+ detection requires
3009 */
3010 schedule_work(&adapter->sfp_task);
3011}
3012
3013/**
3014 * ixgbe_sfp_task - worker thread to find a missing module
3015 * @work: pointer to work_struct containing our data
3016 **/
3017static void ixgbe_sfp_task(struct work_struct *work)
3018{
3019 struct ixgbe_adapter *adapter = container_of(work,
3020 struct ixgbe_adapter,
3021 sfp_task);
3022 struct ixgbe_hw *hw = &adapter->hw;
3023
3024 if ((hw->phy.type == ixgbe_phy_nl) &&
3025 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
3026 s32 ret = hw->phy.ops.identify_sfp(hw);
3027 if (ret)
3028 goto reschedule;
3029 ret = hw->phy.ops.reset(hw);
3030 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3031 DPRINTK(PROBE, ERR, "failed to initialize because an "
3032 "unsupported SFP+ module type was detected.\n"
3033 "Reload the driver after installing a "
3034 "supported module.\n");
3035 unregister_netdev(adapter->netdev);
3036 } else {
3037 DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
3038 hw->phy.sfp_type);
3039 }
3040 /* don't need this routine any more */
3041 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3042 }
3043 return;
3044reschedule:
3045 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
3046 mod_timer(&adapter->sfp_timer,
3047 round_jiffies(jiffies + (2 * HZ)));
3048}
3049
3050/**
3051 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
3052 * @adapter: board private structure to initialize
3053 *
3054 * ixgbe_sw_init initializes the Adapter private data structure.
3055 * Fields are initialized based on PCI device information and
3056 * OS network device settings (MTU size).
3057 **/
3058static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
3059{
3060 struct ixgbe_hw *hw = &adapter->hw;
3061 struct pci_dev *pdev = adapter->pdev;
3062 unsigned int rss;
3063#ifdef CONFIG_IXGBE_DCB
3064 int j;
3065 struct tc_configuration *tc;
3066#endif
3067
3068 /* PCI config space info */
3069
3070 hw->vendor_id = pdev->vendor;
3071 hw->device_id = pdev->device;
3072 hw->revision_id = pdev->revision;
3073 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3074 hw->subsystem_device_id = pdev->subsystem_device;
3075
3076 /* Set capability flags */
3077 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
3078 adapter->ring_feature[RING_F_RSS].indices = rss;
3079 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
3080 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
3081 if (hw->mac.type == ixgbe_mac_82598EB)
3082 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
3083 else if (hw->mac.type == ixgbe_mac_82599EB)
3084 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
3085
3086#ifdef CONFIG_IXGBE_DCB
3087 /* Configure DCB traffic classes */
3088 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
3089 tc = &adapter->dcb_cfg.tc_config[j];
3090 tc->path[DCB_TX_CONFIG].bwg_id = 0;
3091 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
3092 tc->path[DCB_RX_CONFIG].bwg_id = 0;
3093 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
3094 tc->dcb_pfc = pfc_disabled;
3095 }
3096 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
3097 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
3098 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
3099 adapter->dcb_cfg.round_robin_enable = false;
3100 adapter->dcb_set_bitmap = 0x00;
3101 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
3102 adapter->ring_feature[RING_F_DCB].indices);
3103
3104#endif
3105
3106 /* default flow control settings */
3107 hw->fc.requested_mode = ixgbe_fc_none;
3108 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
3109 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
3110 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
3111 hw->fc.send_xon = true;
3112
3113 /* enable itr by default in dynamic mode */
3114 adapter->itr_setting = 1;
3115 adapter->eitr_param = 20000;
3116
3117 /* set defaults for eitr in MegaBytes */
3118 adapter->eitr_low = 10;
3119 adapter->eitr_high = 20;
3120
3121 /* set default ring sizes */
3122 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
3123 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
3124
3125 /* initialize eeprom parameters */
3126 if (ixgbe_init_eeprom_params_generic(hw)) {
3127 dev_err(&pdev->dev, "EEPROM initialization failed\n");
3128 return -EIO;
3129 }
3130
3131 /* enable rx csum by default */
3132 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
3133
3134 set_bit(__IXGBE_DOWN, &adapter->state);
3135
3136 return 0;
3137}
3138
3139/**
3140 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
3141 * @adapter: board private structure
3142 * @tx_ring: tx descriptor ring (for a specific queue) to setup
3143 *
3144 * Return 0 on success, negative on failure
3145 **/
3146int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
3147 struct ixgbe_ring *tx_ring)
3148{
3149 struct pci_dev *pdev = adapter->pdev;
3150 int size;
3151
3152 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3153 tx_ring->tx_buffer_info = vmalloc(size);
3154 if (!tx_ring->tx_buffer_info)
3155 goto err;
3156 memset(tx_ring->tx_buffer_info, 0, size);
3157
3158 /* round up to nearest 4K */
3159 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3160 tx_ring->size = ALIGN(tx_ring->size, 4096);
3161
3162 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
3163 &tx_ring->dma);
3164 if (!tx_ring->desc)
3165 goto err;
3166
3167 tx_ring->next_to_use = 0;
3168 tx_ring->next_to_clean = 0;
3169 tx_ring->work_limit = tx_ring->count;
3170 return 0;
3171
3172err:
3173 vfree(tx_ring->tx_buffer_info);
3174 tx_ring->tx_buffer_info = NULL;
3175 DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
3176 "descriptor ring\n");
3177 return -ENOMEM;
3178}
3179
3180/**
3181 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
3182 * @adapter: board private structure
3183 *
3184 * If this function returns with an error, then it's possible one or
3185 * more of the rings is populated (while the rest are not). It is the
3186 * callers duty to clean those orphaned rings.
3187 *
3188 * Return 0 on success, negative on failure
3189 **/
3190static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
3191{
3192 int i, err = 0;
3193
3194 for (i = 0; i < adapter->num_tx_queues; i++) {
3195 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
3196 if (!err)
3197 continue;
3198 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
3199 break;
3200 }
3201
3202 return err;
3203}
3204
3205/**
3206 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
3207 * @adapter: board private structure
3208 * @rx_ring: rx descriptor ring (for a specific queue) to setup
3209 *
3210 * Returns 0 on success, negative on failure
3211 **/
3212int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
3213 struct ixgbe_ring *rx_ring)
3214{
3215 struct pci_dev *pdev = adapter->pdev;
3216 int size;
3217
3218 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3219 rx_ring->rx_buffer_info = vmalloc(size);
3220 if (!rx_ring->rx_buffer_info) {
3221 DPRINTK(PROBE, ERR,
3222 "vmalloc allocation failed for the rx desc ring\n");
3223 goto alloc_failed;
3224 }
3225 memset(rx_ring->rx_buffer_info, 0, size);
3226
3227 /* Round up to nearest 4K */
3228 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3229 rx_ring->size = ALIGN(rx_ring->size, 4096);
3230
3231 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
3232
3233 if (!rx_ring->desc) {
3234 DPRINTK(PROBE, ERR,
3235 "Memory allocation failed for the rx desc ring\n");
3236 vfree(rx_ring->rx_buffer_info);
3237 goto alloc_failed;
3238 }
3239
3240 rx_ring->next_to_clean = 0;
3241 rx_ring->next_to_use = 0;
3242
3243 return 0;
3244
3245alloc_failed:
3246 return -ENOMEM;
3247}
3248
3249/**
3250 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
3251 * @adapter: board private structure
3252 *
3253 * If this function returns with an error, then it's possible one or
3254 * more of the rings is populated (while the rest are not). It is the
3255 * callers duty to clean those orphaned rings.
3256 *
3257 * Return 0 on success, negative on failure
3258 **/
3259
3260static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
3261{
3262 int i, err = 0;
3263
3264 for (i = 0; i < adapter->num_rx_queues; i++) {
3265 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
3266 if (!err)
3267 continue;
3268 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
3269 break;
3270 }
3271
3272 return err;
3273}
3274
3275/**
3276 * ixgbe_free_tx_resources - Free Tx Resources per Queue
3277 * @adapter: board private structure
3278 * @tx_ring: Tx descriptor ring for a specific queue
3279 *
3280 * Free all transmit software resources
3281 **/
3282void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
3283 struct ixgbe_ring *tx_ring)
3284{
3285 struct pci_dev *pdev = adapter->pdev;
3286
3287 ixgbe_clean_tx_ring(adapter, tx_ring);
3288
3289 vfree(tx_ring->tx_buffer_info);
3290 tx_ring->tx_buffer_info = NULL;
3291
3292 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
3293
3294 tx_ring->desc = NULL;
3295}
3296
3297/**
3298 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
3299 * @adapter: board private structure
3300 *
3301 * Free all transmit software resources
3302 **/
3303static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
3304{
3305 int i;
3306
3307 for (i = 0; i < adapter->num_tx_queues; i++)
3308 ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
3309}
3310
3311/**
3312 * ixgbe_free_rx_resources - Free Rx Resources
3313 * @adapter: board private structure
3314 * @rx_ring: ring to clean the resources from
3315 *
3316 * Free all receive software resources
3317 **/
3318void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
3319 struct ixgbe_ring *rx_ring)
3320{
3321 struct pci_dev *pdev = adapter->pdev;
3322
3323 ixgbe_clean_rx_ring(adapter, rx_ring);
3324
3325 vfree(rx_ring->rx_buffer_info);
3326 rx_ring->rx_buffer_info = NULL;
3327
3328 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
3329
3330 rx_ring->desc = NULL;
3331}
3332
3333/**
3334 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
3335 * @adapter: board private structure
3336 *
3337 * Free all receive software resources
3338 **/
3339static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
3340{
3341 int i;
3342
3343 for (i = 0; i < adapter->num_rx_queues; i++)
3344 ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
3345}
3346
3347/**
3348 * ixgbe_change_mtu - Change the Maximum Transfer Unit
3349 * @netdev: network interface device structure
3350 * @new_mtu: new value for maximum frame size
3351 *
3352 * Returns 0 on success, negative on failure
3353 **/
3354static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
3355{
3356 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3357 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3358
3359 /* MTU < 68 is an error and causes problems on some kernels */
3360 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
3361 return -EINVAL;
3362
3363 DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
3364 netdev->mtu, new_mtu);
3365 /* must set new MTU before calling down or up */
3366 netdev->mtu = new_mtu;
3367
3368 if (netif_running(netdev))
3369 ixgbe_reinit_locked(adapter);
3370
3371 return 0;
3372}
3373
3374/**
3375 * ixgbe_open - Called when a network interface is made active
3376 * @netdev: network interface device structure
3377 *
3378 * Returns 0 on success, negative value on failure
3379 *
3380 * The open entry point is called when a network interface is made
3381 * active by the system (IFF_UP). At this point all resources needed
3382 * for transmit and receive operations are allocated, the interrupt
3383 * handler is registered with the OS, the watchdog timer is started,
3384 * and the stack is notified that the interface is ready.
3385 **/
3386static int ixgbe_open(struct net_device *netdev)
3387{
3388 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3389 int err;
3390
3391 /* disallow open during test */
3392 if (test_bit(__IXGBE_TESTING, &adapter->state))
3393 return -EBUSY;
3394
3395 /* allocate transmit descriptors */
3396 err = ixgbe_setup_all_tx_resources(adapter);
3397 if (err)
3398 goto err_setup_tx;
3399
3400 /* allocate receive descriptors */
3401 err = ixgbe_setup_all_rx_resources(adapter);
3402 if (err)
3403 goto err_setup_rx;
3404
3405 ixgbe_configure(adapter);
3406
3407 err = ixgbe_request_irq(adapter);
3408 if (err)
3409 goto err_req_irq;
3410
3411 err = ixgbe_up_complete(adapter);
3412 if (err)
3413 goto err_up;
3414
3415 netif_tx_start_all_queues(netdev);
3416
3417 return 0;
3418
3419err_up:
3420 ixgbe_release_hw_control(adapter);
3421 ixgbe_free_irq(adapter);
3422err_req_irq:
3423 ixgbe_free_all_rx_resources(adapter);
3424err_setup_rx:
3425 ixgbe_free_all_tx_resources(adapter);
3426err_setup_tx:
3427 ixgbe_reset(adapter);
3428
3429 return err;
3430}
3431
3432/**
3433 * ixgbe_close - Disables a network interface
3434 * @netdev: network interface device structure
3435 *
3436 * Returns 0, this is not allowed to fail
3437 *
3438 * The close entry point is called when an interface is de-activated
3439 * by the OS. The hardware is still under the drivers control, but
3440 * needs to be disabled. A global MAC reset is issued to stop the
3441 * hardware, and all transmit and receive resources are freed.
3442 **/
3443static int ixgbe_close(struct net_device *netdev)
3444{
3445 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3446
3447 ixgbe_down(adapter);
3448 ixgbe_free_irq(adapter);
3449
3450 ixgbe_free_all_tx_resources(adapter);
3451 ixgbe_free_all_rx_resources(adapter);
3452
3453 ixgbe_release_hw_control(adapter);
3454
3455 return 0;
3456}
3457
3458/**
3459 * ixgbe_napi_add_all - prep napi structs for use
3460 * @adapter: private struct
3461 * helper function to napi_add each possible q_vector->napi
3462 */
3463void ixgbe_napi_add_all(struct ixgbe_adapter *adapter)
3464{
3465 int q_idx, q_vectors;
3466 struct net_device *netdev = adapter->netdev;
3467 int (*poll)(struct napi_struct *, int);
3468
3469 /* check if we already have our netdev->napi_list populated */
3470 if (&netdev->napi_list != netdev->napi_list.next)
3471 return;
3472
3473 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3474 poll = &ixgbe_clean_rxonly;
3475 /* Only enable as many vectors as we have rx queues. */
3476 q_vectors = adapter->num_rx_queues;
3477 } else {
3478 poll = &ixgbe_poll;
3479 /* only one q_vector for legacy modes */
3480 q_vectors = 1;
3481 }
3482
3483 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3484 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
3485 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
3486 }
3487}
3488
3489void ixgbe_napi_del_all(struct ixgbe_adapter *adapter)
3490{
3491 int q_idx;
3492 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3493
3494 /* legacy and MSI only use one vector */
3495 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3496 q_vectors = 1;
3497
3498 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3499 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
3500 if (!q_vector->rxr_count)
3501 continue;
3502 netif_napi_del(&q_vector->napi);
3503 }
3504}
3505
3506#ifdef CONFIG_PM
3507static int ixgbe_resume(struct pci_dev *pdev)
3508{
3509 struct net_device *netdev = pci_get_drvdata(pdev);
3510 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3511 u32 err;
3512
3513 pci_set_power_state(pdev, PCI_D0);
3514 pci_restore_state(pdev);
3515 err = pci_enable_device(pdev);
3516 if (err) {
3517 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
3518 "suspend\n");
3519 return err;
3520 }
3521 pci_set_master(pdev);
3522
3523 pci_enable_wake(pdev, PCI_D3hot, 0);
3524 pci_enable_wake(pdev, PCI_D3cold, 0);
3525
3526 err = ixgbe_init_interrupt_scheme(adapter);
3527 if (err) {
3528 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
3529 "device\n");
3530 return err;
3531 }
3532
3533 ixgbe_napi_add_all(adapter);
3534 ixgbe_reset(adapter);
3535
3536 if (netif_running(netdev)) {
3537 err = ixgbe_open(adapter->netdev);
3538 if (err)
3539 return err;
3540 }
3541
3542 netif_device_attach(netdev);
3543
3544 return 0;
3545}
3546
3547#endif /* CONFIG_PM */
3548static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
3549{
3550 struct net_device *netdev = pci_get_drvdata(pdev);
3551 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3552 struct ixgbe_hw *hw = &adapter->hw;
3553 u32 ctrl, fctrl;
3554 u32 wufc = adapter->wol;
3555#ifdef CONFIG_PM
3556 int retval = 0;
3557#endif
3558
3559 netif_device_detach(netdev);
3560
3561 if (netif_running(netdev)) {
3562 ixgbe_down(adapter);
3563 ixgbe_free_irq(adapter);
3564 ixgbe_free_all_tx_resources(adapter);
3565 ixgbe_free_all_rx_resources(adapter);
3566 }
3567 ixgbe_reset_interrupt_capability(adapter);
3568 ixgbe_napi_del_all(adapter);
3569 INIT_LIST_HEAD(&netdev->napi_list);
3570 kfree(adapter->tx_ring);
3571 kfree(adapter->rx_ring);
3572
3573#ifdef CONFIG_PM
3574 retval = pci_save_state(pdev);
3575 if (retval)
3576 return retval;
3577#endif
3578 if (wufc) {
3579 ixgbe_set_rx_mode(netdev);
3580
3581 /* turn on all-multi mode if wake on multicast is enabled */
3582 if (wufc & IXGBE_WUFC_MC) {
3583 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3584 fctrl |= IXGBE_FCTRL_MPE;
3585 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3586 }
3587
3588 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
3589 ctrl |= IXGBE_CTRL_GIO_DIS;
3590 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
3591
3592 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
3593 } else {
3594 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
3595 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
3596 }
3597
3598 if (wufc && hw->mac.type == ixgbe_mac_82599EB) {
3599 pci_enable_wake(pdev, PCI_D3hot, 1);
3600 pci_enable_wake(pdev, PCI_D3cold, 1);
3601 } else {
3602 pci_enable_wake(pdev, PCI_D3hot, 0);
3603 pci_enable_wake(pdev, PCI_D3cold, 0);
3604 }
3605
3606 ixgbe_release_hw_control(adapter);
3607
3608 pci_disable_device(pdev);
3609
3610 pci_set_power_state(pdev, pci_choose_state(pdev, state));
3611
3612 return 0;
3613}
3614
3615static void ixgbe_shutdown(struct pci_dev *pdev)
3616{
3617 ixgbe_suspend(pdev, PMSG_SUSPEND);
3618}
3619
3620/**
3621 * ixgbe_update_stats - Update the board statistics counters.
3622 * @adapter: board private structure
3623 **/
3624void ixgbe_update_stats(struct ixgbe_adapter *adapter)
3625{
3626 struct ixgbe_hw *hw = &adapter->hw;
3627 u64 total_mpc = 0;
3628 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
3629
3630 if (hw->mac.type == ixgbe_mac_82599EB) {
3631 for (i = 0; i < 16; i++)
3632 adapter->hw_rx_no_dma_resources +=
3633 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
3634 }
3635
3636 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
3637 for (i = 0; i < 8; i++) {
3638 /* for packet buffers not used, the register should read 0 */
3639 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
3640 missed_rx += mpc;
3641 adapter->stats.mpc[i] += mpc;
3642 total_mpc += adapter->stats.mpc[i];
3643 if (hw->mac.type == ixgbe_mac_82598EB)
3644 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
3645 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
3646 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
3647 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
3648 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
3649 if (hw->mac.type == ixgbe_mac_82599EB) {
3650 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
3651 IXGBE_PXONRXCNT(i));
3652 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
3653 IXGBE_PXOFFRXCNT(i));
3654 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
3655 } else {
3656 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
3657 IXGBE_PXONRXC(i));
3658 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
3659 IXGBE_PXOFFRXC(i));
3660 }
3661 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
3662 IXGBE_PXONTXC(i));
3663 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
3664 IXGBE_PXOFFTXC(i));
3665 }
3666 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
3667 /* work around hardware counting issue */
3668 adapter->stats.gprc -= missed_rx;
3669
3670 /* 82598 hardware only has a 32 bit counter in the high register */
3671 if (hw->mac.type == ixgbe_mac_82599EB) {
3672 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
3673 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
3674 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
3675 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
3676 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
3677 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
3678 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
3679 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
3680 } else {
3681 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
3682 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
3683 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
3684 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
3685 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
3686 }
3687 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
3688 adapter->stats.bprc += bprc;
3689 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
3690 if (hw->mac.type == ixgbe_mac_82598EB)
3691 adapter->stats.mprc -= bprc;
3692 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
3693 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
3694 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
3695 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
3696 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
3697 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
3698 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
3699 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
3700 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
3701 adapter->stats.lxontxc += lxon;
3702 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
3703 adapter->stats.lxofftxc += lxoff;
3704 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3705 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
3706 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
3707 /*
3708 * 82598 errata - tx of flow control packets is included in tx counters
3709 */
3710 xon_off_tot = lxon + lxoff;
3711 adapter->stats.gptc -= xon_off_tot;
3712 adapter->stats.mptc -= xon_off_tot;
3713 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
3714 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3715 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
3716 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
3717 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
3718 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
3719 adapter->stats.ptc64 -= xon_off_tot;
3720 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
3721 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
3722 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
3723 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
3724 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
3725 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
3726
3727 /* Fill out the OS statistics structure */
3728 adapter->net_stats.multicast = adapter->stats.mprc;
3729
3730 /* Rx Errors */
3731 adapter->net_stats.rx_errors = adapter->stats.crcerrs +
3732 adapter->stats.rlec;
3733 adapter->net_stats.rx_dropped = 0;
3734 adapter->net_stats.rx_length_errors = adapter->stats.rlec;
3735 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3736 adapter->net_stats.rx_missed_errors = total_mpc;
3737}
3738
3739/**
3740 * ixgbe_watchdog - Timer Call-back
3741 * @data: pointer to adapter cast into an unsigned long
3742 **/
3743static void ixgbe_watchdog(unsigned long data)
3744{
3745 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3746 struct ixgbe_hw *hw = &adapter->hw;
3747
3748 /* Do the watchdog outside of interrupt context due to the lovely
3749 * delays that some of the newer hardware requires */
3750 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
3751 /* Cause software interrupt to ensure rx rings are cleaned */
3752 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3753 u32 eics =
3754 (1 << (adapter->num_msix_vectors - NON_Q_VECTORS)) - 1;
3755 IXGBE_WRITE_REG(hw, IXGBE_EICS, eics);
3756 } else {
3757 /* For legacy and MSI interrupts don't set any bits that
3758 * are enabled for EIAM, because this operation would
3759 * set *both* EIMS and EICS for any bit in EIAM */
3760 IXGBE_WRITE_REG(hw, IXGBE_EICS,
3761 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
3762 }
3763 /* Reset the timer */
3764 mod_timer(&adapter->watchdog_timer,
3765 round_jiffies(jiffies + 2 * HZ));
3766 }
3767
3768 schedule_work(&adapter->watchdog_task);
3769}
3770
3771/**
3772 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
3773 * @work: pointer to work_struct containing our data
3774 **/
3775static void ixgbe_multispeed_fiber_task(struct work_struct *work)
3776{
3777 struct ixgbe_adapter *adapter = container_of(work,
3778 struct ixgbe_adapter,
3779 multispeed_fiber_task);
3780 struct ixgbe_hw *hw = &adapter->hw;
3781 u32 autoneg;
3782
3783 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
3784 if (hw->mac.ops.get_link_capabilities)
3785 hw->mac.ops.get_link_capabilities(hw, &autoneg,
3786 &hw->mac.autoneg);
3787 if (hw->mac.ops.setup_link_speed)
3788 hw->mac.ops.setup_link_speed(hw, autoneg, true, true);
3789 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3790 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
3791}
3792
3793/**
3794 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
3795 * @work: pointer to work_struct containing our data
3796 **/
3797static void ixgbe_sfp_config_module_task(struct work_struct *work)
3798{
3799 struct ixgbe_adapter *adapter = container_of(work,
3800 struct ixgbe_adapter,
3801 sfp_config_module_task);
3802 struct ixgbe_hw *hw = &adapter->hw;
3803 u32 err;
3804
3805 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
3806 err = hw->phy.ops.identify_sfp(hw);
3807 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3808 DPRINTK(PROBE, ERR, "PHY not supported on this NIC %d\n", err);
3809 ixgbe_down(adapter);
3810 return;
3811 }
3812 hw->mac.ops.setup_sfp(hw);
3813
3814 if (!adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK)
3815 /* This will also work for DA Twinax connections */
3816 schedule_work(&adapter->multispeed_fiber_task);
3817 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
3818}
3819
3820/**
3821 * ixgbe_watchdog_task - worker thread to bring link up
3822 * @work: pointer to work_struct containing our data
3823 **/
3824static void ixgbe_watchdog_task(struct work_struct *work)
3825{
3826 struct ixgbe_adapter *adapter = container_of(work,
3827 struct ixgbe_adapter,
3828 watchdog_task);
3829 struct net_device *netdev = adapter->netdev;
3830 struct ixgbe_hw *hw = &adapter->hw;
3831 u32 link_speed = adapter->link_speed;
3832 bool link_up = adapter->link_up;
3833
3834 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
3835
3836 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
3837 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3838 if (link_up ||
3839 time_after(jiffies, (adapter->link_check_timeout +
3840 IXGBE_TRY_LINK_TIMEOUT))) {
3841 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
3842 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
3843 }
3844 adapter->link_up = link_up;
3845 adapter->link_speed = link_speed;
3846 }
3847
3848 if (link_up) {
3849 if (!netif_carrier_ok(netdev)) {
3850 bool flow_rx, flow_tx;
3851
3852 if (hw->mac.type == ixgbe_mac_82599EB) {
3853 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
3854 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
3855 flow_rx = (mflcn & IXGBE_MFLCN_RFCE);
3856 flow_tx = (fccfg & IXGBE_FCCFG_TFCE_802_3X);
3857 } else {
3858 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3859 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
3860 flow_rx = (frctl & IXGBE_FCTRL_RFCE);
3861 flow_tx = (rmcs & IXGBE_RMCS_TFCE_802_3X);
3862 }
3863
3864 printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
3865 "Flow Control: %s\n",
3866 netdev->name,
3867 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
3868 "10 Gbps" :
3869 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
3870 "1 Gbps" : "unknown speed")),
3871 ((flow_rx && flow_tx) ? "RX/TX" :
3872 (flow_rx ? "RX" :
3873 (flow_tx ? "TX" : "None"))));
3874
3875 netif_carrier_on(netdev);
3876 } else {
3877 /* Force detection of hung controller */
3878 adapter->detect_tx_hung = true;
3879 }
3880 } else {
3881 adapter->link_up = false;
3882 adapter->link_speed = 0;
3883 if (netif_carrier_ok(netdev)) {
3884 printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
3885 netdev->name);
3886 netif_carrier_off(netdev);
3887 }
3888 }
3889
3890 ixgbe_update_stats(adapter);
3891 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
3892}
3893
3894static int ixgbe_tso(struct ixgbe_adapter *adapter,
3895 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
3896 u32 tx_flags, u8 *hdr_len)
3897{
3898 struct ixgbe_adv_tx_context_desc *context_desc;
3899 unsigned int i;
3900 int err;
3901 struct ixgbe_tx_buffer *tx_buffer_info;
3902 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
3903 u32 mss_l4len_idx, l4len;
3904
3905 if (skb_is_gso(skb)) {
3906 if (skb_header_cloned(skb)) {
3907 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3908 if (err)
3909 return err;
3910 }
3911 l4len = tcp_hdrlen(skb);
3912 *hdr_len += l4len;
3913
3914 if (skb->protocol == htons(ETH_P_IP)) {
3915 struct iphdr *iph = ip_hdr(skb);
3916 iph->tot_len = 0;
3917 iph->check = 0;
3918 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3919 iph->daddr, 0,
3920 IPPROTO_TCP,
3921 0);
3922 adapter->hw_tso_ctxt++;
3923 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3924 ipv6_hdr(skb)->payload_len = 0;
3925 tcp_hdr(skb)->check =
3926 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3927 &ipv6_hdr(skb)->daddr,
3928 0, IPPROTO_TCP, 0);
3929 adapter->hw_tso6_ctxt++;
3930 }
3931
3932 i = tx_ring->next_to_use;
3933
3934 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3935 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
3936
3937 /* VLAN MACLEN IPLEN */
3938 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3939 vlan_macip_lens |=
3940 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
3941 vlan_macip_lens |= ((skb_network_offset(skb)) <<
3942 IXGBE_ADVTXD_MACLEN_SHIFT);
3943 *hdr_len += skb_network_offset(skb);
3944 vlan_macip_lens |=
3945 (skb_transport_header(skb) - skb_network_header(skb));
3946 *hdr_len +=
3947 (skb_transport_header(skb) - skb_network_header(skb));
3948 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3949 context_desc->seqnum_seed = 0;
3950
3951 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3952 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
3953 IXGBE_ADVTXD_DTYP_CTXT);
3954
3955 if (skb->protocol == htons(ETH_P_IP))
3956 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
3957 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3958 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
3959
3960 /* MSS L4LEN IDX */
3961 mss_l4len_idx =
3962 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
3963 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
3964 /* use index 1 for TSO */
3965 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3966 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3967
3968 tx_buffer_info->time_stamp = jiffies;
3969 tx_buffer_info->next_to_watch = i;
3970
3971 i++;
3972 if (i == tx_ring->count)
3973 i = 0;
3974 tx_ring->next_to_use = i;
3975
3976 return true;
3977 }
3978 return false;
3979}
3980
3981static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
3982 struct ixgbe_ring *tx_ring,
3983 struct sk_buff *skb, u32 tx_flags)
3984{
3985 struct ixgbe_adv_tx_context_desc *context_desc;
3986 unsigned int i;
3987 struct ixgbe_tx_buffer *tx_buffer_info;
3988 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
3989
3990 if (skb->ip_summed == CHECKSUM_PARTIAL ||
3991 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
3992 i = tx_ring->next_to_use;
3993 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3994 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
3995
3996 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3997 vlan_macip_lens |=
3998 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
3999 vlan_macip_lens |= (skb_network_offset(skb) <<
4000 IXGBE_ADVTXD_MACLEN_SHIFT);
4001 if (skb->ip_summed == CHECKSUM_PARTIAL)
4002 vlan_macip_lens |= (skb_transport_header(skb) -
4003 skb_network_header(skb));
4004
4005 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4006 context_desc->seqnum_seed = 0;
4007
4008 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
4009 IXGBE_ADVTXD_DTYP_CTXT);
4010
4011 if (skb->ip_summed == CHECKSUM_PARTIAL) {
4012 switch (skb->protocol) {
4013 case cpu_to_be16(ETH_P_IP):
4014 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4015 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4016 type_tucmd_mlhl |=
4017 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4018 break;
4019 case cpu_to_be16(ETH_P_IPV6):
4020 /* XXX what about other V6 headers?? */
4021 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4022 type_tucmd_mlhl |=
4023 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4024 break;
4025 default:
4026 if (unlikely(net_ratelimit())) {
4027 DPRINTK(PROBE, WARNING,
4028 "partial checksum but proto=%x!\n",
4029 skb->protocol);
4030 }
4031 break;
4032 }
4033 }
4034
4035 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4036 /* use index zero for tx checksum offload */
4037 context_desc->mss_l4len_idx = 0;
4038
4039 tx_buffer_info->time_stamp = jiffies;
4040 tx_buffer_info->next_to_watch = i;
4041
4042 adapter->hw_csum_tx_good++;
4043 i++;
4044 if (i == tx_ring->count)
4045 i = 0;
4046 tx_ring->next_to_use = i;
4047
4048 return true;
4049 }
4050
4051 return false;
4052}
4053
4054static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
4055 struct ixgbe_ring *tx_ring,
4056 struct sk_buff *skb, unsigned int first)
4057{
4058 struct ixgbe_tx_buffer *tx_buffer_info;
4059 unsigned int len = skb->len;
4060 unsigned int offset = 0, size, count = 0, i;
4061 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
4062 unsigned int f;
4063
4064 len -= skb->data_len;
4065
4066 i = tx_ring->next_to_use;
4067
4068 while (len) {
4069 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4070 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4071
4072 tx_buffer_info->length = size;
4073 tx_buffer_info->dma = pci_map_single(adapter->pdev,
4074 skb->data + offset,
4075 size, PCI_DMA_TODEVICE);
4076 tx_buffer_info->time_stamp = jiffies;
4077 tx_buffer_info->next_to_watch = i;
4078
4079 len -= size;
4080 offset += size;
4081 count++;
4082 i++;
4083 if (i == tx_ring->count)
4084 i = 0;
4085 }
4086
4087 for (f = 0; f < nr_frags; f++) {
4088 struct skb_frag_struct *frag;
4089
4090 frag = &skb_shinfo(skb)->frags[f];
4091 len = frag->size;
4092 offset = frag->page_offset;
4093
4094 while (len) {
4095 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4096 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4097
4098 tx_buffer_info->length = size;
4099 tx_buffer_info->dma = pci_map_page(adapter->pdev,
4100 frag->page,
4101 offset,
4102 size,
4103 PCI_DMA_TODEVICE);
4104 tx_buffer_info->time_stamp = jiffies;
4105 tx_buffer_info->next_to_watch = i;
4106
4107 len -= size;
4108 offset += size;
4109 count++;
4110 i++;
4111 if (i == tx_ring->count)
4112 i = 0;
4113 }
4114 }
4115 if (i == 0)
4116 i = tx_ring->count - 1;
4117 else
4118 i = i - 1;
4119 tx_ring->tx_buffer_info[i].skb = skb;
4120 tx_ring->tx_buffer_info[first].next_to_watch = i;
4121
4122 return count;
4123}
4124
4125static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
4126 struct ixgbe_ring *tx_ring,
4127 int tx_flags, int count, u32 paylen, u8 hdr_len)
4128{
4129 union ixgbe_adv_tx_desc *tx_desc = NULL;
4130 struct ixgbe_tx_buffer *tx_buffer_info;
4131 u32 olinfo_status = 0, cmd_type_len = 0;
4132 unsigned int i;
4133 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
4134
4135 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
4136
4137 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
4138
4139 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4140 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
4141
4142 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
4143 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
4144
4145 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
4146 IXGBE_ADVTXD_POPTS_SHIFT;
4147
4148 /* use index 1 context for tso */
4149 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4150 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
4151 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
4152 IXGBE_ADVTXD_POPTS_SHIFT;
4153
4154 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
4155 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
4156 IXGBE_ADVTXD_POPTS_SHIFT;
4157
4158 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
4159
4160 i = tx_ring->next_to_use;
4161 while (count--) {
4162 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4163 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
4164 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
4165 tx_desc->read.cmd_type_len =
4166 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
4167 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4168 i++;
4169 if (i == tx_ring->count)
4170 i = 0;
4171 }
4172
4173 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
4174
4175 /*
4176 * Force memory writes to complete before letting h/w
4177 * know there are new descriptors to fetch. (Only
4178 * applicable for weak-ordered memory model archs,
4179 * such as IA-64).
4180 */
4181 wmb();
4182
4183 tx_ring->next_to_use = i;
4184 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4185}
4186
4187static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
4188 struct ixgbe_ring *tx_ring, int size)
4189{
4190 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4191
4192 netif_stop_subqueue(netdev, tx_ring->queue_index);
4193 /* Herbert's original patch had:
4194 * smp_mb__after_netif_stop_queue();
4195 * but since that doesn't exist yet, just open code it. */
4196 smp_mb();
4197
4198 /* We need to check again in a case another CPU has just
4199 * made room available. */
4200 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
4201 return -EBUSY;
4202
4203 /* A reprieve! - use start_queue because it doesn't call schedule */
4204 netif_start_subqueue(netdev, tx_ring->queue_index);
4205 ++adapter->restart_queue;
4206 return 0;
4207}
4208
4209static int ixgbe_maybe_stop_tx(struct net_device *netdev,
4210 struct ixgbe_ring *tx_ring, int size)
4211{
4212 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
4213 return 0;
4214 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
4215}
4216
4217static int ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4218{
4219 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4220 struct ixgbe_ring *tx_ring;
4221 unsigned int first;
4222 unsigned int tx_flags = 0;
4223 u8 hdr_len = 0;
4224 int r_idx = 0, tso;
4225 int count = 0;
4226 unsigned int f;
4227
4228 r_idx = (adapter->num_tx_queues - 1) & skb->queue_mapping;
4229 tx_ring = &adapter->tx_ring[r_idx];
4230
4231 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4232 tx_flags |= vlan_tx_tag_get(skb);
4233 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4234 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
4235 tx_flags |= (skb->queue_mapping << 13);
4236 }
4237 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4238 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4239 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4240 tx_flags |= (skb->queue_mapping << 13);
4241 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4242 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4243 }
4244 /* three things can cause us to need a context descriptor */
4245 if (skb_is_gso(skb) ||
4246 (skb->ip_summed == CHECKSUM_PARTIAL) ||
4247 (tx_flags & IXGBE_TX_FLAGS_VLAN))
4248 count++;
4249
4250 count += TXD_USE_COUNT(skb_headlen(skb));
4251 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4252 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4253
4254 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
4255 adapter->tx_busy++;
4256 return NETDEV_TX_BUSY;
4257 }
4258
4259 if (skb->protocol == htons(ETH_P_IP))
4260 tx_flags |= IXGBE_TX_FLAGS_IPV4;
4261 first = tx_ring->next_to_use;
4262 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
4263 if (tso < 0) {
4264 dev_kfree_skb_any(skb);
4265 return NETDEV_TX_OK;
4266 }
4267
4268 if (tso)
4269 tx_flags |= IXGBE_TX_FLAGS_TSO;
4270 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
4271 (skb->ip_summed == CHECKSUM_PARTIAL))
4272 tx_flags |= IXGBE_TX_FLAGS_CSUM;
4273
4274 ixgbe_tx_queue(adapter, tx_ring, tx_flags,
4275 ixgbe_tx_map(adapter, tx_ring, skb, first),
4276 skb->len, hdr_len);
4277
4278 netdev->trans_start = jiffies;
4279
4280 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
4281
4282 return NETDEV_TX_OK;
4283}
4284
4285/**
4286 * ixgbe_get_stats - Get System Network Statistics
4287 * @netdev: network interface device structure
4288 *
4289 * Returns the address of the device statistics structure.
4290 * The statistics are actually updated from the timer callback.
4291 **/
4292static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
4293{
4294 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4295
4296 /* only return the current stats */
4297 return &adapter->net_stats;
4298}
4299
4300/**
4301 * ixgbe_set_mac - Change the Ethernet Address of the NIC
4302 * @netdev: network interface device structure
4303 * @p: pointer to an address structure
4304 *
4305 * Returns 0 on success, negative on failure
4306 **/
4307static int ixgbe_set_mac(struct net_device *netdev, void *p)
4308{
4309 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4310 struct ixgbe_hw *hw = &adapter->hw;
4311 struct sockaddr *addr = p;
4312
4313 if (!is_valid_ether_addr(addr->sa_data))
4314 return -EADDRNOTAVAIL;
4315
4316 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4317 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
4318
4319 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
4320
4321 return 0;
4322}
4323
4324#ifdef CONFIG_NET_POLL_CONTROLLER
4325/*
4326 * Polling 'interrupt' - used by things like netconsole to send skbs
4327 * without having to re-enable interrupts. It's not called while
4328 * the interrupt routine is executing.
4329 */
4330static void ixgbe_netpoll(struct net_device *netdev)
4331{
4332 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4333
4334 disable_irq(adapter->pdev->irq);
4335 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
4336 ixgbe_intr(adapter->pdev->irq, netdev);
4337 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
4338 enable_irq(adapter->pdev->irq);
4339}
4340#endif
4341
4342static const struct net_device_ops ixgbe_netdev_ops = {
4343 .ndo_open = ixgbe_open,
4344 .ndo_stop = ixgbe_close,
4345 .ndo_start_xmit = ixgbe_xmit_frame,
4346 .ndo_get_stats = ixgbe_get_stats,
4347 .ndo_set_multicast_list = ixgbe_set_rx_mode,
4348 .ndo_validate_addr = eth_validate_addr,
4349 .ndo_set_mac_address = ixgbe_set_mac,
4350 .ndo_change_mtu = ixgbe_change_mtu,
4351 .ndo_tx_timeout = ixgbe_tx_timeout,
4352 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
4353 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
4354 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
4355#ifdef CONFIG_NET_POLL_CONTROLLER
4356 .ndo_poll_controller = ixgbe_netpoll,
4357#endif
4358};
4359
4360/**
4361 * ixgbe_probe - Device Initialization Routine
4362 * @pdev: PCI device information struct
4363 * @ent: entry in ixgbe_pci_tbl
4364 *
4365 * Returns 0 on success, negative on failure
4366 *
4367 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
4368 * The OS initialization, configuring of the adapter private structure,
4369 * and a hardware reset occur.
4370 **/
4371static int __devinit ixgbe_probe(struct pci_dev *pdev,
4372 const struct pci_device_id *ent)
4373{
4374 struct net_device *netdev;
4375 struct ixgbe_adapter *adapter = NULL;
4376 struct ixgbe_hw *hw;
4377 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
4378 static int cards_found;
4379 int i, err, pci_using_dac;
4380 u16 pm_value = 0;
4381 u32 part_num, eec;
4382
4383 err = pci_enable_device(pdev);
4384 if (err)
4385 return err;
4386
4387 if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK) &&
4388 !pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK)) {
4389 pci_using_dac = 1;
4390 } else {
4391 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4392 if (err) {
4393 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
4394 if (err) {
4395 dev_err(&pdev->dev, "No usable DMA "
4396 "configuration, aborting\n");
4397 goto err_dma;
4398 }
4399 }
4400 pci_using_dac = 0;
4401 }
4402
4403 err = pci_request_regions(pdev, ixgbe_driver_name);
4404 if (err) {
4405 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4406 goto err_pci_reg;
4407 }
4408
4409 err = pci_enable_pcie_error_reporting(pdev);
4410 if (err) {
4411 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
4412 "0x%x\n", err);
4413 /* non-fatal, continue */
4414 }
4415
4416 pci_set_master(pdev);
4417 pci_save_state(pdev);
4418
4419 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
4420 if (!netdev) {
4421 err = -ENOMEM;
4422 goto err_alloc_etherdev;
4423 }
4424
4425 SET_NETDEV_DEV(netdev, &pdev->dev);
4426
4427 pci_set_drvdata(pdev, netdev);
4428 adapter = netdev_priv(netdev);
4429
4430 adapter->netdev = netdev;
4431 adapter->pdev = pdev;
4432 hw = &adapter->hw;
4433 hw->back = adapter;
4434 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
4435
4436 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4437 pci_resource_len(pdev, 0));
4438 if (!hw->hw_addr) {
4439 err = -EIO;
4440 goto err_ioremap;
4441 }
4442
4443 for (i = 1; i <= 5; i++) {
4444 if (pci_resource_len(pdev, i) == 0)
4445 continue;
4446 }
4447
4448 netdev->netdev_ops = &ixgbe_netdev_ops;
4449 ixgbe_set_ethtool_ops(netdev);
4450 netdev->watchdog_timeo = 5 * HZ;
4451 strcpy(netdev->name, pci_name(pdev));
4452
4453 adapter->bd_number = cards_found;
4454
4455 /* Setup hw api */
4456 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4457 hw->mac.type = ii->mac;
4458
4459 /* EEPROM */
4460 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
4461 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
4462 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
4463 if (!(eec & (1 << 8)))
4464 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
4465
4466 /* PHY */
4467 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
4468 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
4469
4470 /* set up this timer and work struct before calling get_invariants
4471 * which might start the timer
4472 */
4473 init_timer(&adapter->sfp_timer);
4474 adapter->sfp_timer.function = &ixgbe_sfp_timer;
4475 adapter->sfp_timer.data = (unsigned long) adapter;
4476
4477 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
4478
4479 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
4480 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
4481
4482 /* a new SFP+ module arrival, called from GPI SDP2 context */
4483 INIT_WORK(&adapter->sfp_config_module_task,
4484 ixgbe_sfp_config_module_task);
4485
4486 err = ii->get_invariants(hw);
4487 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
4488 /* start a kernel thread to watch for a module to arrive */
4489 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4490 mod_timer(&adapter->sfp_timer,
4491 round_jiffies(jiffies + (2 * HZ)));
4492 err = 0;
4493 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4494 DPRINTK(PROBE, ERR, "failed to load because an "
4495 "unsupported SFP+ module type was detected.\n");
4496 goto err_hw_init;
4497 } else if (err) {
4498 goto err_hw_init;
4499 }
4500
4501 /* setup the private structure */
4502 err = ixgbe_sw_init(adapter);
4503 if (err)
4504 goto err_sw_init;
4505
4506 /* reset_hw fills in the perm_addr as well */
4507 err = hw->mac.ops.reset_hw(hw);
4508 if (err) {
4509 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
4510 goto err_sw_init;
4511 }
4512
4513 netdev->features = NETIF_F_SG |
4514 NETIF_F_IP_CSUM |
4515 NETIF_F_HW_VLAN_TX |
4516 NETIF_F_HW_VLAN_RX |
4517 NETIF_F_HW_VLAN_FILTER;
4518
4519 netdev->features |= NETIF_F_IPV6_CSUM;
4520 netdev->features |= NETIF_F_TSO;
4521 netdev->features |= NETIF_F_TSO6;
4522 netdev->features |= NETIF_F_GRO;
4523
4524 netdev->vlan_features |= NETIF_F_TSO;
4525 netdev->vlan_features |= NETIF_F_TSO6;
4526 netdev->vlan_features |= NETIF_F_IP_CSUM;
4527 netdev->vlan_features |= NETIF_F_SG;
4528
4529 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
4530 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4531
4532#ifdef CONFIG_IXGBE_DCB
4533 netdev->dcbnl_ops = &dcbnl_ops;
4534#endif
4535
4536 if (pci_using_dac)
4537 netdev->features |= NETIF_F_HIGHDMA;
4538
4539 /* make sure the EEPROM is good */
4540 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
4541 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
4542 err = -EIO;
4543 goto err_eeprom;
4544 }
4545
4546 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
4547 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
4548
4549 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
4550 dev_err(&pdev->dev, "invalid MAC address\n");
4551 err = -EIO;
4552 goto err_eeprom;
4553 }
4554
4555 init_timer(&adapter->watchdog_timer);
4556 adapter->watchdog_timer.function = &ixgbe_watchdog;
4557 adapter->watchdog_timer.data = (unsigned long)adapter;
4558
4559 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
4560 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
4561
4562 err = ixgbe_init_interrupt_scheme(adapter);
4563 if (err)
4564 goto err_sw_init;
4565
4566 switch (pdev->device) {
4567 case IXGBE_DEV_ID_82599_KX4:
4568#define IXGBE_PCIE_PMCSR 0x44
4569 adapter->wol = IXGBE_WUFC_MAG;
4570 pci_read_config_word(pdev, IXGBE_PCIE_PMCSR, &pm_value);
4571 pci_write_config_word(pdev, IXGBE_PCIE_PMCSR,
4572 (pm_value | (1 << 8)));
4573 break;
4574 default:
4575 adapter->wol = 0;
4576 break;
4577 }
4578 device_init_wakeup(&adapter->pdev->dev, true);
4579 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
4580
4581 /* print bus type/speed/width info */
4582 dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
4583 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
4584 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
4585 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
4586 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
4587 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
4588 "Unknown"),
4589 netdev->dev_addr);
4590 ixgbe_read_pba_num_generic(hw, &part_num);
4591 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
4592 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
4593 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
4594 (part_num >> 8), (part_num & 0xff));
4595 else
4596 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4597 hw->mac.type, hw->phy.type,
4598 (part_num >> 8), (part_num & 0xff));
4599
4600 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
4601 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
4602 "this card is not sufficient for optimal "
4603 "performance.\n");
4604 dev_warn(&pdev->dev, "For optimal performance a x8 "
4605 "PCI-Express slot is required.\n");
4606 }
4607
4608 /* save off EEPROM version number */
4609 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
4610
4611 /* reset the hardware with the new settings */
4612 hw->mac.ops.start_hw(hw);
4613
4614 netif_carrier_off(netdev);
4615
4616 strcpy(netdev->name, "eth%d");
4617 err = register_netdev(netdev);
4618 if (err)
4619 goto err_register;
4620
4621#ifdef CONFIG_IXGBE_DCA
4622 if (dca_add_requester(&pdev->dev) == 0) {
4623 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
4624 /* always use CB2 mode, difference is masked
4625 * in the CB driver */
4626 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
4627 ixgbe_setup_dca(adapter);
4628 }
4629#endif
4630
4631 dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
4632 cards_found++;
4633 return 0;
4634
4635err_register:
4636 ixgbe_release_hw_control(adapter);
4637err_hw_init:
4638err_sw_init:
4639 ixgbe_reset_interrupt_capability(adapter);
4640err_eeprom:
4641 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4642 del_timer_sync(&adapter->sfp_timer);
4643 cancel_work_sync(&adapter->sfp_task);
4644 cancel_work_sync(&adapter->multispeed_fiber_task);
4645 cancel_work_sync(&adapter->sfp_config_module_task);
4646 iounmap(hw->hw_addr);
4647err_ioremap:
4648 free_netdev(netdev);
4649err_alloc_etherdev:
4650 pci_release_regions(pdev);
4651err_pci_reg:
4652err_dma:
4653 pci_disable_device(pdev);
4654 return err;
4655}
4656
4657/**
4658 * ixgbe_remove - Device Removal Routine
4659 * @pdev: PCI device information struct
4660 *
4661 * ixgbe_remove is called by the PCI subsystem to alert the driver
4662 * that it should release a PCI device. The could be caused by a
4663 * Hot-Plug event, or because the driver is going to be removed from
4664 * memory.
4665 **/
4666static void __devexit ixgbe_remove(struct pci_dev *pdev)
4667{
4668 struct net_device *netdev = pci_get_drvdata(pdev);
4669 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4670 int err;
4671
4672 set_bit(__IXGBE_DOWN, &adapter->state);
4673 /* clear the module not found bit to make sure the worker won't
4674 * reschedule
4675 */
4676 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4677 del_timer_sync(&adapter->watchdog_timer);
4678
4679 del_timer_sync(&adapter->sfp_timer);
4680 cancel_work_sync(&adapter->watchdog_task);
4681 cancel_work_sync(&adapter->sfp_task);
4682 cancel_work_sync(&adapter->multispeed_fiber_task);
4683 cancel_work_sync(&adapter->sfp_config_module_task);
4684 flush_scheduled_work();
4685
4686#ifdef CONFIG_IXGBE_DCA
4687 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
4688 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
4689 dca_remove_requester(&pdev->dev);
4690 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
4691 }
4692
4693#endif
4694 if (netdev->reg_state == NETREG_REGISTERED)
4695 unregister_netdev(netdev);
4696
4697 ixgbe_reset_interrupt_capability(adapter);
4698
4699 ixgbe_release_hw_control(adapter);
4700
4701 iounmap(adapter->hw.hw_addr);
4702 pci_release_regions(pdev);
4703
4704 DPRINTK(PROBE, INFO, "complete\n");
4705 kfree(adapter->tx_ring);
4706 kfree(adapter->rx_ring);
4707
4708 free_netdev(netdev);
4709
4710 err = pci_disable_pcie_error_reporting(pdev);
4711 if (err)
4712 dev_err(&pdev->dev,
4713 "pci_disable_pcie_error_reporting failed 0x%x\n", err);
4714
4715 pci_disable_device(pdev);
4716}
4717
4718/**
4719 * ixgbe_io_error_detected - called when PCI error is detected
4720 * @pdev: Pointer to PCI device
4721 * @state: The current pci connection state
4722 *
4723 * This function is called after a PCI bus error affecting
4724 * this device has been detected.
4725 */
4726static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
4727 pci_channel_state_t state)
4728{
4729 struct net_device *netdev = pci_get_drvdata(pdev);
4730 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4731
4732 netif_device_detach(netdev);
4733
4734 if (netif_running(netdev))
4735 ixgbe_down(adapter);
4736 pci_disable_device(pdev);
4737
4738 /* Request a slot reset. */
4739 return PCI_ERS_RESULT_NEED_RESET;
4740}
4741
4742/**
4743 * ixgbe_io_slot_reset - called after the pci bus has been reset.
4744 * @pdev: Pointer to PCI device
4745 *
4746 * Restart the card from scratch, as if from a cold-boot.
4747 */
4748static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
4749{
4750 struct net_device *netdev = pci_get_drvdata(pdev);
4751 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4752 pci_ers_result_t result;
4753 int err;
4754
4755 if (pci_enable_device(pdev)) {
4756 DPRINTK(PROBE, ERR,
4757 "Cannot re-enable PCI device after reset.\n");
4758 result = PCI_ERS_RESULT_DISCONNECT;
4759 } else {
4760 pci_set_master(pdev);
4761 pci_restore_state(pdev);
4762
4763 pci_enable_wake(pdev, PCI_D3hot, 0);
4764 pci_enable_wake(pdev, PCI_D3cold, 0);
4765
4766 ixgbe_reset(adapter);
4767
4768 result = PCI_ERS_RESULT_RECOVERED;
4769 }
4770
4771 err = pci_cleanup_aer_uncorrect_error_status(pdev);
4772 if (err) {
4773 dev_err(&pdev->dev,
4774 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
4775 /* non-fatal, continue */
4776 }
4777
4778 return result;
4779}
4780
4781/**
4782 * ixgbe_io_resume - called when traffic can start flowing again.
4783 * @pdev: Pointer to PCI device
4784 *
4785 * This callback is called when the error recovery driver tells us that
4786 * its OK to resume normal operation.
4787 */
4788static void ixgbe_io_resume(struct pci_dev *pdev)
4789{
4790 struct net_device *netdev = pci_get_drvdata(pdev);
4791 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4792
4793 if (netif_running(netdev)) {
4794 if (ixgbe_up(adapter)) {
4795 DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
4796 return;
4797 }
4798 }
4799
4800 netif_device_attach(netdev);
4801}
4802
4803static struct pci_error_handlers ixgbe_err_handler = {
4804 .error_detected = ixgbe_io_error_detected,
4805 .slot_reset = ixgbe_io_slot_reset,
4806 .resume = ixgbe_io_resume,
4807};
4808
4809static struct pci_driver ixgbe_driver = {
4810 .name = ixgbe_driver_name,
4811 .id_table = ixgbe_pci_tbl,
4812 .probe = ixgbe_probe,
4813 .remove = __devexit_p(ixgbe_remove),
4814#ifdef CONFIG_PM
4815 .suspend = ixgbe_suspend,
4816 .resume = ixgbe_resume,
4817#endif
4818 .shutdown = ixgbe_shutdown,
4819 .err_handler = &ixgbe_err_handler
4820};
4821
4822/**
4823 * ixgbe_init_module - Driver Registration Routine
4824 *
4825 * ixgbe_init_module is the first routine called when the driver is
4826 * loaded. All it does is register with the PCI subsystem.
4827 **/
4828static int __init ixgbe_init_module(void)
4829{
4830 int ret;
4831 printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
4832 ixgbe_driver_string, ixgbe_driver_version);
4833
4834 printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
4835
4836#ifdef CONFIG_IXGBE_DCA
4837 dca_register_notify(&dca_notifier);
4838#endif
4839
4840 ret = pci_register_driver(&ixgbe_driver);
4841 return ret;
4842}
4843
4844module_init(ixgbe_init_module);
4845
4846/**
4847 * ixgbe_exit_module - Driver Exit Cleanup Routine
4848 *
4849 * ixgbe_exit_module is called just before the driver is removed
4850 * from memory.
4851 **/
4852static void __exit ixgbe_exit_module(void)
4853{
4854#ifdef CONFIG_IXGBE_DCA
4855 dca_unregister_notify(&dca_notifier);
4856#endif
4857 pci_unregister_driver(&ixgbe_driver);
4858}
4859
4860#ifdef CONFIG_IXGBE_DCA
4861static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
4862 void *p)
4863{
4864 int ret_val;
4865
4866 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
4867 __ixgbe_notify_dca);
4868
4869 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4870}
4871#endif /* CONFIG_IXGBE_DCA */
4872
4873module_exit(ixgbe_exit_module);
4874
4875/* ixgbe_main.c */
Clone of davem's net-next-2.6 tree