1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2009 Intel Corporation.
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.
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
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.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
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>
36 #include <linux/tcp.h>
37 #include <linux/pkt_sched.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/ethtool.h>
42 #include <linux/if_vlan.h>
43 #include <scsi/fc/fc_fcoe.h>
46 #include "ixgbe_common.h"
48 char ixgbe_driver_name[] = "ixgbe";
49 static const char ixgbe_driver_string[] =
50 "Intel(R) 10 Gigabit PCI Express Network Driver";
52 #define DRV_VERSION "2.0.34-k2"
53 const char ixgbe_driver_version[] = DRV_VERSION;
54 static char ixgbe_copyright[] = "Copyright (c) 1999-2009 Intel Corporation.";
56 static const struct ixgbe_info *ixgbe_info_tbl[] = {
57 [board_82598] = &ixgbe_82598_info,
58 [board_82599] = &ixgbe_82599_info,
61 /* ixgbe_pci_tbl - PCI Device ID Table
63 * Wildcard entries (PCI_ANY_ID) should come last
64 * Last entry must be all 0s
66 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
67 * Class, Class Mask, private data (not used) }
69 static struct pci_device_id ixgbe_pci_tbl[] = {
70 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
72 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
74 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
76 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
78 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
80 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
82 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
84 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
99 /* required last entry */
102 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
104 #ifdef CONFIG_IXGBE_DCA
105 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
107 static struct notifier_block dca_notifier = {
108 .notifier_call = ixgbe_notify_dca,
114 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
115 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
116 MODULE_LICENSE("GPL");
117 MODULE_VERSION(DRV_VERSION);
119 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
121 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
125 /* Let firmware take over control of h/w */
126 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
127 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
128 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
131 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
135 /* Let firmware know the driver has taken over */
136 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
137 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
138 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
142 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
143 * @adapter: pointer to adapter struct
144 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
145 * @queue: queue to map the corresponding interrupt to
146 * @msix_vector: the vector to map to the corresponding queue
149 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
150 u8 queue, u8 msix_vector)
153 struct ixgbe_hw *hw = &adapter->hw;
154 switch (hw->mac.type) {
155 case ixgbe_mac_82598EB:
156 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
159 index = (((direction * 64) + queue) >> 2) & 0x1F;
160 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
161 ivar &= ~(0xFF << (8 * (queue & 0x3)));
162 ivar |= (msix_vector << (8 * (queue & 0x3)));
163 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
165 case ixgbe_mac_82599EB:
166 if (direction == -1) {
168 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
169 index = ((queue & 1) * 8);
170 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
171 ivar &= ~(0xFF << index);
172 ivar |= (msix_vector << index);
173 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
176 /* tx or rx causes */
177 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
178 index = ((16 * (queue & 1)) + (8 * direction));
179 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
180 ivar &= ~(0xFF << index);
181 ivar |= (msix_vector << index);
182 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
190 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
195 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
196 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
197 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
199 mask = (qmask & 0xFFFFFFFF);
200 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
201 mask = (qmask >> 32);
202 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
206 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
207 struct ixgbe_tx_buffer
210 tx_buffer_info->dma = 0;
211 if (tx_buffer_info->skb) {
212 skb_dma_unmap(&adapter->pdev->dev, tx_buffer_info->skb,
214 dev_kfree_skb_any(tx_buffer_info->skb);
215 tx_buffer_info->skb = NULL;
217 tx_buffer_info->time_stamp = 0;
218 /* tx_buffer_info must be completely set up in the transmit path */
221 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
222 struct ixgbe_ring *tx_ring,
225 struct ixgbe_hw *hw = &adapter->hw;
227 /* Detect a transmit hang in hardware, this serializes the
228 * check with the clearing of time_stamp and movement of eop */
229 adapter->detect_tx_hung = false;
230 if (tx_ring->tx_buffer_info[eop].time_stamp &&
231 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
232 !(IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & IXGBE_TFCS_TXOFF)) {
233 /* detected Tx unit hang */
234 union ixgbe_adv_tx_desc *tx_desc;
235 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
236 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
238 " TDH, TDT <%x>, <%x>\n"
239 " next_to_use <%x>\n"
240 " next_to_clean <%x>\n"
241 "tx_buffer_info[next_to_clean]\n"
242 " time_stamp <%lx>\n"
244 tx_ring->queue_index,
245 IXGBE_READ_REG(hw, tx_ring->head),
246 IXGBE_READ_REG(hw, tx_ring->tail),
247 tx_ring->next_to_use, eop,
248 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
255 #define IXGBE_MAX_TXD_PWR 14
256 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
258 /* Tx Descriptors needed, worst case */
259 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
260 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
261 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
262 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
264 static void ixgbe_tx_timeout(struct net_device *netdev);
267 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
268 * @q_vector: structure containing interrupt and ring information
269 * @tx_ring: tx ring to clean
271 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
272 struct ixgbe_ring *tx_ring)
274 struct ixgbe_adapter *adapter = q_vector->adapter;
275 struct net_device *netdev = adapter->netdev;
276 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
277 struct ixgbe_tx_buffer *tx_buffer_info;
278 unsigned int i, eop, count = 0;
279 unsigned int total_bytes = 0, total_packets = 0;
281 i = tx_ring->next_to_clean;
282 eop = tx_ring->tx_buffer_info[i].next_to_watch;
283 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
285 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
286 (count < tx_ring->work_limit)) {
287 bool cleaned = false;
288 for ( ; !cleaned; count++) {
290 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
291 tx_buffer_info = &tx_ring->tx_buffer_info[i];
292 cleaned = (i == eop);
293 skb = tx_buffer_info->skb;
295 if (cleaned && skb) {
296 unsigned int segs, bytecount;
297 unsigned int hlen = skb_headlen(skb);
299 /* gso_segs is currently only valid for tcp */
300 segs = skb_shinfo(skb)->gso_segs ?: 1;
302 /* adjust for FCoE Sequence Offload */
303 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
304 && (skb->protocol == htons(ETH_P_FCOE)) &&
306 hlen = skb_transport_offset(skb) +
307 sizeof(struct fc_frame_header) +
308 sizeof(struct fcoe_crc_eof);
309 segs = DIV_ROUND_UP(skb->len - hlen,
310 skb_shinfo(skb)->gso_size);
312 #endif /* IXGBE_FCOE */
313 /* multiply data chunks by size of headers */
314 bytecount = ((segs - 1) * hlen) + skb->len;
315 total_packets += segs;
316 total_bytes += bytecount;
319 ixgbe_unmap_and_free_tx_resource(adapter,
322 tx_desc->wb.status = 0;
325 if (i == tx_ring->count)
329 eop = tx_ring->tx_buffer_info[i].next_to_watch;
330 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
333 tx_ring->next_to_clean = i;
335 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
336 if (unlikely(count && netif_carrier_ok(netdev) &&
337 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
338 /* Make sure that anybody stopping the queue after this
339 * sees the new next_to_clean.
342 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
343 !test_bit(__IXGBE_DOWN, &adapter->state)) {
344 netif_wake_subqueue(netdev, tx_ring->queue_index);
345 ++adapter->restart_queue;
349 if (adapter->detect_tx_hung) {
350 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
351 /* schedule immediate reset if we believe we hung */
353 "tx hang %d detected, resetting adapter\n",
354 adapter->tx_timeout_count + 1);
355 ixgbe_tx_timeout(adapter->netdev);
359 /* re-arm the interrupt */
360 if (count >= tx_ring->work_limit)
361 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
363 tx_ring->total_bytes += total_bytes;
364 tx_ring->total_packets += total_packets;
365 tx_ring->stats.packets += total_packets;
366 tx_ring->stats.bytes += total_bytes;
367 adapter->net_stats.tx_bytes += total_bytes;
368 adapter->net_stats.tx_packets += total_packets;
369 return (count < tx_ring->work_limit);
372 #ifdef CONFIG_IXGBE_DCA
373 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
374 struct ixgbe_ring *rx_ring)
378 int q = rx_ring - adapter->rx_ring;
380 if (rx_ring->cpu != cpu) {
381 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
382 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
383 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
384 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
385 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
386 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
387 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
388 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
390 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
391 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
392 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
393 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
394 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
395 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
401 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
402 struct ixgbe_ring *tx_ring)
406 int q = tx_ring - adapter->tx_ring;
408 if (tx_ring->cpu != cpu) {
409 txctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q));
410 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
411 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
412 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
413 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
414 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
415 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
416 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
418 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
419 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q), txctrl);
425 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
429 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
432 /* always use CB2 mode, difference is masked in the CB driver */
433 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
435 for (i = 0; i < adapter->num_tx_queues; i++) {
436 adapter->tx_ring[i].cpu = -1;
437 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
439 for (i = 0; i < adapter->num_rx_queues; i++) {
440 adapter->rx_ring[i].cpu = -1;
441 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
445 static int __ixgbe_notify_dca(struct device *dev, void *data)
447 struct net_device *netdev = dev_get_drvdata(dev);
448 struct ixgbe_adapter *adapter = netdev_priv(netdev);
449 unsigned long event = *(unsigned long *)data;
452 case DCA_PROVIDER_ADD:
453 /* if we're already enabled, don't do it again */
454 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
456 if (dca_add_requester(dev) == 0) {
457 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
458 ixgbe_setup_dca(adapter);
461 /* Fall Through since DCA is disabled. */
462 case DCA_PROVIDER_REMOVE:
463 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
464 dca_remove_requester(dev);
465 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
466 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
474 #endif /* CONFIG_IXGBE_DCA */
476 * ixgbe_receive_skb - Send a completed packet up the stack
477 * @adapter: board private structure
478 * @skb: packet to send up
479 * @status: hardware indication of status of receive
480 * @rx_ring: rx descriptor ring (for a specific queue) to setup
481 * @rx_desc: rx descriptor
483 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
484 struct sk_buff *skb, u8 status,
485 struct ixgbe_ring *ring,
486 union ixgbe_adv_rx_desc *rx_desc)
488 struct ixgbe_adapter *adapter = q_vector->adapter;
489 struct napi_struct *napi = &q_vector->napi;
490 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
491 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
493 skb_record_rx_queue(skb, ring->queue_index);
494 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
495 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
496 vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
498 napi_gro_receive(napi, skb);
500 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
501 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
508 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
509 * @adapter: address of board private structure
510 * @status_err: hardware indication of status of receive
511 * @skb: skb currently being received and modified
513 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
514 union ixgbe_adv_rx_desc *rx_desc,
517 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
519 skb->ip_summed = CHECKSUM_NONE;
521 /* Rx csum disabled */
522 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
525 /* if IP and error */
526 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
527 (status_err & IXGBE_RXDADV_ERR_IPE)) {
528 adapter->hw_csum_rx_error++;
532 if (!(status_err & IXGBE_RXD_STAT_L4CS))
535 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
536 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
539 * 82599 errata, UDP frames with a 0 checksum can be marked as
542 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
543 (adapter->hw.mac.type == ixgbe_mac_82599EB))
546 adapter->hw_csum_rx_error++;
550 /* It must be a TCP or UDP packet with a valid checksum */
551 skb->ip_summed = CHECKSUM_UNNECESSARY;
552 adapter->hw_csum_rx_good++;
555 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
556 struct ixgbe_ring *rx_ring, u32 val)
559 * Force memory writes to complete before letting h/w
560 * know there are new descriptors to fetch. (Only
561 * applicable for weak-ordered memory model archs,
565 IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
569 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
570 * @adapter: address of board private structure
572 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
573 struct ixgbe_ring *rx_ring,
576 struct pci_dev *pdev = adapter->pdev;
577 union ixgbe_adv_rx_desc *rx_desc;
578 struct ixgbe_rx_buffer *bi;
581 i = rx_ring->next_to_use;
582 bi = &rx_ring->rx_buffer_info[i];
584 while (cleaned_count--) {
585 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
588 (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
590 bi->page = alloc_page(GFP_ATOMIC);
592 adapter->alloc_rx_page_failed++;
597 /* use a half page if we're re-using */
598 bi->page_offset ^= (PAGE_SIZE / 2);
601 bi->page_dma = pci_map_page(pdev, bi->page,
609 skb = netdev_alloc_skb(adapter->netdev,
610 (rx_ring->rx_buf_len +
614 adapter->alloc_rx_buff_failed++;
619 * Make buffer alignment 2 beyond a 16 byte boundary
620 * this will result in a 16 byte aligned IP header after
621 * the 14 byte MAC header is removed
623 skb_reserve(skb, NET_IP_ALIGN);
626 bi->dma = pci_map_single(pdev, skb->data,
630 /* Refresh the desc even if buffer_addrs didn't change because
631 * each write-back erases this info. */
632 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
633 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
634 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
636 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
640 if (i == rx_ring->count)
642 bi = &rx_ring->rx_buffer_info[i];
646 if (rx_ring->next_to_use != i) {
647 rx_ring->next_to_use = i;
649 i = (rx_ring->count - 1);
651 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
655 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
657 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
660 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
662 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
665 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
667 return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
668 IXGBE_RXDADV_RSCCNT_MASK) >>
669 IXGBE_RXDADV_RSCCNT_SHIFT;
673 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
674 * @skb: pointer to the last skb in the rsc queue
676 * This function changes a queue full of hw rsc buffers into a completed
677 * packet. It uses the ->prev pointers to find the first packet and then
678 * turns it into the frag list owner.
680 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
682 unsigned int frag_list_size = 0;
685 struct sk_buff *prev = skb->prev;
686 frag_list_size += skb->len;
691 skb_shinfo(skb)->frag_list = skb->next;
693 skb->len += frag_list_size;
694 skb->data_len += frag_list_size;
695 skb->truesize += frag_list_size;
699 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
700 struct ixgbe_ring *rx_ring,
701 int *work_done, int work_to_do)
703 struct ixgbe_adapter *adapter = q_vector->adapter;
704 struct pci_dev *pdev = adapter->pdev;
705 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
706 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
708 unsigned int i, rsc_count = 0;
711 bool cleaned = false;
712 int cleaned_count = 0;
713 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
716 #endif /* IXGBE_FCOE */
718 i = rx_ring->next_to_clean;
719 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
720 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
721 rx_buffer_info = &rx_ring->rx_buffer_info[i];
723 while (staterr & IXGBE_RXD_STAT_DD) {
725 if (*work_done >= work_to_do)
729 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
730 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
731 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
732 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
733 if (hdr_info & IXGBE_RXDADV_SPH)
734 adapter->rx_hdr_split++;
735 if (len > IXGBE_RX_HDR_SIZE)
736 len = IXGBE_RX_HDR_SIZE;
737 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
739 len = le16_to_cpu(rx_desc->wb.upper.length);
743 skb = rx_buffer_info->skb;
744 prefetch(skb->data - NET_IP_ALIGN);
745 rx_buffer_info->skb = NULL;
747 if (rx_buffer_info->dma) {
748 pci_unmap_single(pdev, rx_buffer_info->dma,
751 rx_buffer_info->dma = 0;
756 pci_unmap_page(pdev, rx_buffer_info->page_dma,
757 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
758 rx_buffer_info->page_dma = 0;
759 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
760 rx_buffer_info->page,
761 rx_buffer_info->page_offset,
764 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
765 (page_count(rx_buffer_info->page) != 1))
766 rx_buffer_info->page = NULL;
768 get_page(rx_buffer_info->page);
770 skb->len += upper_len;
771 skb->data_len += upper_len;
772 skb->truesize += upper_len;
776 if (i == rx_ring->count)
779 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
783 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
784 rsc_count = ixgbe_get_rsc_count(rx_desc);
787 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
788 IXGBE_RXDADV_NEXTP_SHIFT;
789 next_buffer = &rx_ring->rx_buffer_info[nextp];
790 rx_ring->rsc_count += (rsc_count - 1);
792 next_buffer = &rx_ring->rx_buffer_info[i];
795 if (staterr & IXGBE_RXD_STAT_EOP) {
797 skb = ixgbe_transform_rsc_queue(skb);
798 rx_ring->stats.packets++;
799 rx_ring->stats.bytes += skb->len;
801 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
802 rx_buffer_info->skb = next_buffer->skb;
803 rx_buffer_info->dma = next_buffer->dma;
804 next_buffer->skb = skb;
805 next_buffer->dma = 0;
807 skb->next = next_buffer->skb;
808 skb->next->prev = skb;
810 adapter->non_eop_descs++;
814 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
815 dev_kfree_skb_irq(skb);
819 ixgbe_rx_checksum(adapter, rx_desc, skb);
821 /* probably a little skewed due to removing CRC */
822 total_rx_bytes += skb->len;
825 skb->protocol = eth_type_trans(skb, adapter->netdev);
827 /* if ddp, not passing to ULD unless for FCP_RSP or error */
828 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
829 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
833 #endif /* IXGBE_FCOE */
834 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
837 rx_desc->wb.upper.status_error = 0;
839 /* return some buffers to hardware, one at a time is too slow */
840 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
841 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
845 /* use prefetched values */
847 rx_buffer_info = &rx_ring->rx_buffer_info[i];
849 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
852 rx_ring->next_to_clean = i;
853 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
856 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
859 /* include DDPed FCoE data */
863 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
864 sizeof(struct fc_frame_header) -
865 sizeof(struct fcoe_crc_eof);
868 total_rx_bytes += ddp_bytes;
869 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
871 #endif /* IXGBE_FCOE */
873 rx_ring->total_packets += total_rx_packets;
874 rx_ring->total_bytes += total_rx_bytes;
875 adapter->net_stats.rx_bytes += total_rx_bytes;
876 adapter->net_stats.rx_packets += total_rx_packets;
881 static int ixgbe_clean_rxonly(struct napi_struct *, int);
883 * ixgbe_configure_msix - Configure MSI-X hardware
884 * @adapter: board private structure
886 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
889 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
891 struct ixgbe_q_vector *q_vector;
892 int i, j, q_vectors, v_idx, r_idx;
895 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
898 * Populate the IVAR table and set the ITR values to the
899 * corresponding register.
901 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
902 q_vector = adapter->q_vector[v_idx];
903 /* XXX for_each_bit(...) */
904 r_idx = find_first_bit(q_vector->rxr_idx,
905 adapter->num_rx_queues);
907 for (i = 0; i < q_vector->rxr_count; i++) {
908 j = adapter->rx_ring[r_idx].reg_idx;
909 ixgbe_set_ivar(adapter, 0, j, v_idx);
910 r_idx = find_next_bit(q_vector->rxr_idx,
911 adapter->num_rx_queues,
914 r_idx = find_first_bit(q_vector->txr_idx,
915 adapter->num_tx_queues);
917 for (i = 0; i < q_vector->txr_count; i++) {
918 j = adapter->tx_ring[r_idx].reg_idx;
919 ixgbe_set_ivar(adapter, 1, j, v_idx);
920 r_idx = find_next_bit(q_vector->txr_idx,
921 adapter->num_tx_queues,
925 /* if this is a tx only vector halve the interrupt rate */
926 if (q_vector->txr_count && !q_vector->rxr_count)
927 q_vector->eitr = (adapter->eitr_param >> 1);
928 else if (q_vector->rxr_count)
930 q_vector->eitr = adapter->eitr_param;
932 ixgbe_write_eitr(q_vector);
935 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
936 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
938 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
939 ixgbe_set_ivar(adapter, -1, 1, v_idx);
940 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
942 /* set up to autoclear timer, and the vectors */
943 mask = IXGBE_EIMS_ENABLE_MASK;
944 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
945 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
952 latency_invalid = 255
956 * ixgbe_update_itr - update the dynamic ITR value based on statistics
957 * @adapter: pointer to adapter
958 * @eitr: eitr setting (ints per sec) to give last timeslice
959 * @itr_setting: current throttle rate in ints/second
960 * @packets: the number of packets during this measurement interval
961 * @bytes: the number of bytes during this measurement interval
963 * Stores a new ITR value based on packets and byte
964 * counts during the last interrupt. The advantage of per interrupt
965 * computation is faster updates and more accurate ITR for the current
966 * traffic pattern. Constants in this function were computed
967 * based on theoretical maximum wire speed and thresholds were set based
968 * on testing data as well as attempting to minimize response time
969 * while increasing bulk throughput.
970 * this functionality is controlled by the InterruptThrottleRate module
971 * parameter (see ixgbe_param.c)
973 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
974 u32 eitr, u8 itr_setting,
975 int packets, int bytes)
977 unsigned int retval = itr_setting;
982 goto update_itr_done;
985 /* simple throttlerate management
986 * 0-20MB/s lowest (100000 ints/s)
987 * 20-100MB/s low (20000 ints/s)
988 * 100-1249MB/s bulk (8000 ints/s)
990 /* what was last interrupt timeslice? */
991 timepassed_us = 1000000/eitr;
992 bytes_perint = bytes / timepassed_us; /* bytes/usec */
994 switch (itr_setting) {
996 if (bytes_perint > adapter->eitr_low)
997 retval = low_latency;
1000 if (bytes_perint > adapter->eitr_high)
1001 retval = bulk_latency;
1002 else if (bytes_perint <= adapter->eitr_low)
1003 retval = lowest_latency;
1006 if (bytes_perint <= adapter->eitr_high)
1007 retval = low_latency;
1016 * ixgbe_write_eitr - write EITR register in hardware specific way
1017 * @q_vector: structure containing interrupt and ring information
1019 * This function is made to be called by ethtool and by the driver
1020 * when it needs to update EITR registers at runtime. Hardware
1021 * specific quirks/differences are taken care of here.
1023 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1025 struct ixgbe_adapter *adapter = q_vector->adapter;
1026 struct ixgbe_hw *hw = &adapter->hw;
1027 int v_idx = q_vector->v_idx;
1028 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1030 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1031 /* must write high and low 16 bits to reset counter */
1032 itr_reg |= (itr_reg << 16);
1033 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1035 * set the WDIS bit to not clear the timer bits and cause an
1036 * immediate assertion of the interrupt
1038 itr_reg |= IXGBE_EITR_CNT_WDIS;
1040 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1043 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1045 struct ixgbe_adapter *adapter = q_vector->adapter;
1047 u8 current_itr, ret_itr;
1049 struct ixgbe_ring *rx_ring, *tx_ring;
1051 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1052 for (i = 0; i < q_vector->txr_count; i++) {
1053 tx_ring = &(adapter->tx_ring[r_idx]);
1054 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1056 tx_ring->total_packets,
1057 tx_ring->total_bytes);
1058 /* if the result for this queue would decrease interrupt
1059 * rate for this vector then use that result */
1060 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1061 q_vector->tx_itr - 1 : ret_itr);
1062 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1066 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1067 for (i = 0; i < q_vector->rxr_count; i++) {
1068 rx_ring = &(adapter->rx_ring[r_idx]);
1069 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1071 rx_ring->total_packets,
1072 rx_ring->total_bytes);
1073 /* if the result for this queue would decrease interrupt
1074 * rate for this vector then use that result */
1075 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1076 q_vector->rx_itr - 1 : ret_itr);
1077 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1081 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1083 switch (current_itr) {
1084 /* counts and packets in update_itr are dependent on these numbers */
1085 case lowest_latency:
1089 new_itr = 20000; /* aka hwitr = ~200 */
1097 if (new_itr != q_vector->eitr) {
1098 /* do an exponential smoothing */
1099 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1101 /* save the algorithm value here, not the smoothed one */
1102 q_vector->eitr = new_itr;
1104 ixgbe_write_eitr(q_vector);
1110 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1112 struct ixgbe_hw *hw = &adapter->hw;
1114 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1115 (eicr & IXGBE_EICR_GPI_SDP1)) {
1116 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
1117 /* write to clear the interrupt */
1118 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1122 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1124 struct ixgbe_hw *hw = &adapter->hw;
1126 if (eicr & IXGBE_EICR_GPI_SDP1) {
1127 /* Clear the interrupt */
1128 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1129 schedule_work(&adapter->multispeed_fiber_task);
1130 } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1131 /* Clear the interrupt */
1132 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1133 schedule_work(&adapter->sfp_config_module_task);
1135 /* Interrupt isn't for us... */
1140 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1142 struct ixgbe_hw *hw = &adapter->hw;
1145 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1146 adapter->link_check_timeout = jiffies;
1147 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1148 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1149 schedule_work(&adapter->watchdog_task);
1153 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1155 struct net_device *netdev = data;
1156 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1157 struct ixgbe_hw *hw = &adapter->hw;
1161 * Workaround for Silicon errata. Use clear-by-write instead
1162 * of clear-by-read. Reading with EICS will return the
1163 * interrupt causes without clearing, which later be done
1164 * with the write to EICR.
1166 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1167 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1169 if (eicr & IXGBE_EICR_LSC)
1170 ixgbe_check_lsc(adapter);
1172 if (hw->mac.type == ixgbe_mac_82598EB)
1173 ixgbe_check_fan_failure(adapter, eicr);
1175 if (hw->mac.type == ixgbe_mac_82599EB) {
1176 ixgbe_check_sfp_event(adapter, eicr);
1178 /* Handle Flow Director Full threshold interrupt */
1179 if (eicr & IXGBE_EICR_FLOW_DIR) {
1181 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1182 /* Disable transmits before FDIR Re-initialization */
1183 netif_tx_stop_all_queues(netdev);
1184 for (i = 0; i < adapter->num_tx_queues; i++) {
1185 struct ixgbe_ring *tx_ring =
1186 &adapter->tx_ring[i];
1187 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1188 &tx_ring->reinit_state))
1189 schedule_work(&adapter->fdir_reinit_task);
1193 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1194 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1199 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1204 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1205 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1206 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1208 mask = (qmask & 0xFFFFFFFF);
1209 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1210 mask = (qmask >> 32);
1211 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1213 /* skip the flush */
1216 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1221 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1222 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1223 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1225 mask = (qmask & 0xFFFFFFFF);
1226 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1227 mask = (qmask >> 32);
1228 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1230 /* skip the flush */
1233 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1235 struct ixgbe_q_vector *q_vector = data;
1236 struct ixgbe_adapter *adapter = q_vector->adapter;
1237 struct ixgbe_ring *tx_ring;
1240 if (!q_vector->txr_count)
1243 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1244 for (i = 0; i < q_vector->txr_count; i++) {
1245 tx_ring = &(adapter->tx_ring[r_idx]);
1246 tx_ring->total_bytes = 0;
1247 tx_ring->total_packets = 0;
1248 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1252 /* disable interrupts on this vector only */
1253 ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1254 napi_schedule(&q_vector->napi);
1260 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1262 * @data: pointer to our q_vector struct for this interrupt vector
1264 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1266 struct ixgbe_q_vector *q_vector = data;
1267 struct ixgbe_adapter *adapter = q_vector->adapter;
1268 struct ixgbe_ring *rx_ring;
1272 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1273 for (i = 0; i < q_vector->rxr_count; i++) {
1274 rx_ring = &(adapter->rx_ring[r_idx]);
1275 rx_ring->total_bytes = 0;
1276 rx_ring->total_packets = 0;
1277 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1281 if (!q_vector->rxr_count)
1284 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1285 rx_ring = &(adapter->rx_ring[r_idx]);
1286 /* disable interrupts on this vector only */
1287 ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1288 napi_schedule(&q_vector->napi);
1293 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1295 struct ixgbe_q_vector *q_vector = data;
1296 struct ixgbe_adapter *adapter = q_vector->adapter;
1297 struct ixgbe_ring *ring;
1301 if (!q_vector->txr_count && !q_vector->rxr_count)
1304 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1305 for (i = 0; i < q_vector->txr_count; i++) {
1306 ring = &(adapter->tx_ring[r_idx]);
1307 ring->total_bytes = 0;
1308 ring->total_packets = 0;
1309 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1313 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1314 for (i = 0; i < q_vector->rxr_count; i++) {
1315 ring = &(adapter->rx_ring[r_idx]);
1316 ring->total_bytes = 0;
1317 ring->total_packets = 0;
1318 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1322 /* disable interrupts on this vector only */
1323 ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1324 napi_schedule(&q_vector->napi);
1330 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1331 * @napi: napi struct with our devices info in it
1332 * @budget: amount of work driver is allowed to do this pass, in packets
1334 * This function is optimized for cleaning one queue only on a single
1337 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1339 struct ixgbe_q_vector *q_vector =
1340 container_of(napi, struct ixgbe_q_vector, napi);
1341 struct ixgbe_adapter *adapter = q_vector->adapter;
1342 struct ixgbe_ring *rx_ring = NULL;
1346 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1347 rx_ring = &(adapter->rx_ring[r_idx]);
1348 #ifdef CONFIG_IXGBE_DCA
1349 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1350 ixgbe_update_rx_dca(adapter, rx_ring);
1353 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1355 /* If all Rx work done, exit the polling mode */
1356 if (work_done < budget) {
1357 napi_complete(napi);
1358 if (adapter->itr_setting & 1)
1359 ixgbe_set_itr_msix(q_vector);
1360 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1361 ixgbe_irq_enable_queues(adapter,
1362 ((u64)1 << q_vector->v_idx));
1369 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1370 * @napi: napi struct with our devices info in it
1371 * @budget: amount of work driver is allowed to do this pass, in packets
1373 * This function will clean more than one rx queue associated with a
1376 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1378 struct ixgbe_q_vector *q_vector =
1379 container_of(napi, struct ixgbe_q_vector, napi);
1380 struct ixgbe_adapter *adapter = q_vector->adapter;
1381 struct ixgbe_ring *ring = NULL;
1382 int work_done = 0, i;
1384 bool tx_clean_complete = true;
1386 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1387 for (i = 0; i < q_vector->txr_count; i++) {
1388 ring = &(adapter->tx_ring[r_idx]);
1389 #ifdef CONFIG_IXGBE_DCA
1390 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1391 ixgbe_update_tx_dca(adapter, ring);
1393 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1394 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1398 /* attempt to distribute budget to each queue fairly, but don't allow
1399 * the budget to go below 1 because we'll exit polling */
1400 budget /= (q_vector->rxr_count ?: 1);
1401 budget = max(budget, 1);
1402 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1403 for (i = 0; i < q_vector->rxr_count; i++) {
1404 ring = &(adapter->rx_ring[r_idx]);
1405 #ifdef CONFIG_IXGBE_DCA
1406 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1407 ixgbe_update_rx_dca(adapter, ring);
1409 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1410 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1414 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1415 ring = &(adapter->rx_ring[r_idx]);
1416 /* If all Rx work done, exit the polling mode */
1417 if (work_done < budget) {
1418 napi_complete(napi);
1419 if (adapter->itr_setting & 1)
1420 ixgbe_set_itr_msix(q_vector);
1421 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1422 ixgbe_irq_enable_queues(adapter,
1423 ((u64)1 << q_vector->v_idx));
1431 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1432 * @napi: napi struct with our devices info in it
1433 * @budget: amount of work driver is allowed to do this pass, in packets
1435 * This function is optimized for cleaning one queue only on a single
1438 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1440 struct ixgbe_q_vector *q_vector =
1441 container_of(napi, struct ixgbe_q_vector, napi);
1442 struct ixgbe_adapter *adapter = q_vector->adapter;
1443 struct ixgbe_ring *tx_ring = NULL;
1447 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1448 tx_ring = &(adapter->tx_ring[r_idx]);
1449 #ifdef CONFIG_IXGBE_DCA
1450 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1451 ixgbe_update_tx_dca(adapter, tx_ring);
1454 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
1457 /* If all Rx work done, exit the polling mode */
1458 if (work_done < budget) {
1459 napi_complete(napi);
1460 if (adapter->itr_setting & 1)
1461 ixgbe_set_itr_msix(q_vector);
1462 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1463 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
1469 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1472 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1474 set_bit(r_idx, q_vector->rxr_idx);
1475 q_vector->rxr_count++;
1478 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1481 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1483 set_bit(t_idx, q_vector->txr_idx);
1484 q_vector->txr_count++;
1488 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1489 * @adapter: board private structure to initialize
1490 * @vectors: allotted vector count for descriptor rings
1492 * This function maps descriptor rings to the queue-specific vectors
1493 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1494 * one vector per ring/queue, but on a constrained vector budget, we
1495 * group the rings as "efficiently" as possible. You would add new
1496 * mapping configurations in here.
1498 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1502 int rxr_idx = 0, txr_idx = 0;
1503 int rxr_remaining = adapter->num_rx_queues;
1504 int txr_remaining = adapter->num_tx_queues;
1509 /* No mapping required if MSI-X is disabled. */
1510 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1514 * The ideal configuration...
1515 * We have enough vectors to map one per queue.
1517 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1518 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1519 map_vector_to_rxq(adapter, v_start, rxr_idx);
1521 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1522 map_vector_to_txq(adapter, v_start, txr_idx);
1528 * If we don't have enough vectors for a 1-to-1
1529 * mapping, we'll have to group them so there are
1530 * multiple queues per vector.
1532 /* Re-adjusting *qpv takes care of the remainder. */
1533 for (i = v_start; i < vectors; i++) {
1534 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1535 for (j = 0; j < rqpv; j++) {
1536 map_vector_to_rxq(adapter, i, rxr_idx);
1541 for (i = v_start; i < vectors; i++) {
1542 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1543 for (j = 0; j < tqpv; j++) {
1544 map_vector_to_txq(adapter, i, txr_idx);
1555 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1556 * @adapter: board private structure
1558 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1559 * interrupts from the kernel.
1561 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1563 struct net_device *netdev = adapter->netdev;
1564 irqreturn_t (*handler)(int, void *);
1565 int i, vector, q_vectors, err;
1568 /* Decrement for Other and TCP Timer vectors */
1569 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1571 /* Map the Tx/Rx rings to the vectors we were allotted. */
1572 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1576 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1577 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1578 &ixgbe_msix_clean_many)
1579 for (vector = 0; vector < q_vectors; vector++) {
1580 handler = SET_HANDLER(adapter->q_vector[vector]);
1582 if(handler == &ixgbe_msix_clean_rx) {
1583 sprintf(adapter->name[vector], "%s-%s-%d",
1584 netdev->name, "rx", ri++);
1586 else if(handler == &ixgbe_msix_clean_tx) {
1587 sprintf(adapter->name[vector], "%s-%s-%d",
1588 netdev->name, "tx", ti++);
1591 sprintf(adapter->name[vector], "%s-%s-%d",
1592 netdev->name, "TxRx", vector);
1594 err = request_irq(adapter->msix_entries[vector].vector,
1595 handler, 0, adapter->name[vector],
1596 adapter->q_vector[vector]);
1599 "request_irq failed for MSIX interrupt "
1600 "Error: %d\n", err);
1601 goto free_queue_irqs;
1605 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1606 err = request_irq(adapter->msix_entries[vector].vector,
1607 &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1610 "request_irq for msix_lsc failed: %d\n", err);
1611 goto free_queue_irqs;
1617 for (i = vector - 1; i >= 0; i--)
1618 free_irq(adapter->msix_entries[--vector].vector,
1619 adapter->q_vector[i]);
1620 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1621 pci_disable_msix(adapter->pdev);
1622 kfree(adapter->msix_entries);
1623 adapter->msix_entries = NULL;
1628 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1630 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1632 u32 new_itr = q_vector->eitr;
1633 struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1634 struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1636 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1638 tx_ring->total_packets,
1639 tx_ring->total_bytes);
1640 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1642 rx_ring->total_packets,
1643 rx_ring->total_bytes);
1645 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1647 switch (current_itr) {
1648 /* counts and packets in update_itr are dependent on these numbers */
1649 case lowest_latency:
1653 new_itr = 20000; /* aka hwitr = ~200 */
1662 if (new_itr != q_vector->eitr) {
1663 /* do an exponential smoothing */
1664 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1666 /* save the algorithm value here, not the smoothed one */
1667 q_vector->eitr = new_itr;
1669 ixgbe_write_eitr(q_vector);
1676 * ixgbe_irq_enable - Enable default interrupt generation settings
1677 * @adapter: board private structure
1679 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1683 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1684 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1685 mask |= IXGBE_EIMS_GPI_SDP1;
1686 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1687 mask |= IXGBE_EIMS_ECC;
1688 mask |= IXGBE_EIMS_GPI_SDP1;
1689 mask |= IXGBE_EIMS_GPI_SDP2;
1691 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
1692 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
1693 mask |= IXGBE_EIMS_FLOW_DIR;
1695 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1696 ixgbe_irq_enable_queues(adapter, ~0);
1697 IXGBE_WRITE_FLUSH(&adapter->hw);
1701 * ixgbe_intr - legacy mode Interrupt Handler
1702 * @irq: interrupt number
1703 * @data: pointer to a network interface device structure
1705 static irqreturn_t ixgbe_intr(int irq, void *data)
1707 struct net_device *netdev = data;
1708 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1709 struct ixgbe_hw *hw = &adapter->hw;
1710 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1714 * Workaround for silicon errata. Mask the interrupts
1715 * before the read of EICR.
1717 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1719 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1720 * therefore no explict interrupt disable is necessary */
1721 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1723 /* shared interrupt alert!
1724 * make sure interrupts are enabled because the read will
1725 * have disabled interrupts due to EIAM */
1726 ixgbe_irq_enable(adapter);
1727 return IRQ_NONE; /* Not our interrupt */
1730 if (eicr & IXGBE_EICR_LSC)
1731 ixgbe_check_lsc(adapter);
1733 if (hw->mac.type == ixgbe_mac_82599EB)
1734 ixgbe_check_sfp_event(adapter, eicr);
1736 ixgbe_check_fan_failure(adapter, eicr);
1738 if (napi_schedule_prep(&(q_vector->napi))) {
1739 adapter->tx_ring[0].total_packets = 0;
1740 adapter->tx_ring[0].total_bytes = 0;
1741 adapter->rx_ring[0].total_packets = 0;
1742 adapter->rx_ring[0].total_bytes = 0;
1743 /* would disable interrupts here but EIAM disabled it */
1744 __napi_schedule(&(q_vector->napi));
1750 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1752 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1754 for (i = 0; i < q_vectors; i++) {
1755 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
1756 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1757 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1758 q_vector->rxr_count = 0;
1759 q_vector->txr_count = 0;
1764 * ixgbe_request_irq - initialize interrupts
1765 * @adapter: board private structure
1767 * Attempts to configure interrupts using the best available
1768 * capabilities of the hardware and kernel.
1770 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1772 struct net_device *netdev = adapter->netdev;
1775 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1776 err = ixgbe_request_msix_irqs(adapter);
1777 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1778 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1779 netdev->name, netdev);
1781 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1782 netdev->name, netdev);
1786 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1791 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1793 struct net_device *netdev = adapter->netdev;
1795 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1798 q_vectors = adapter->num_msix_vectors;
1801 free_irq(adapter->msix_entries[i].vector, netdev);
1804 for (; i >= 0; i--) {
1805 free_irq(adapter->msix_entries[i].vector,
1806 adapter->q_vector[i]);
1809 ixgbe_reset_q_vectors(adapter);
1811 free_irq(adapter->pdev->irq, netdev);
1816 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1817 * @adapter: board private structure
1819 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1821 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1822 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1824 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
1825 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
1826 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
1828 IXGBE_WRITE_FLUSH(&adapter->hw);
1829 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1831 for (i = 0; i < adapter->num_msix_vectors; i++)
1832 synchronize_irq(adapter->msix_entries[i].vector);
1834 synchronize_irq(adapter->pdev->irq);
1839 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1842 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1844 struct ixgbe_hw *hw = &adapter->hw;
1846 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1847 EITR_INTS_PER_SEC_TO_REG(adapter->eitr_param));
1849 ixgbe_set_ivar(adapter, 0, 0, 0);
1850 ixgbe_set_ivar(adapter, 1, 0, 0);
1852 map_vector_to_rxq(adapter, 0, 0);
1853 map_vector_to_txq(adapter, 0, 0);
1855 DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1859 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1860 * @adapter: board private structure
1862 * Configure the Tx unit of the MAC after a reset.
1864 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1867 struct ixgbe_hw *hw = &adapter->hw;
1868 u32 i, j, tdlen, txctrl;
1870 /* Setup the HW Tx Head and Tail descriptor pointers */
1871 for (i = 0; i < adapter->num_tx_queues; i++) {
1872 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1875 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1876 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1877 (tdba & DMA_BIT_MASK(32)));
1878 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1879 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1880 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1881 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1882 adapter->tx_ring[i].head = IXGBE_TDH(j);
1883 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1884 /* Disable Tx Head Writeback RO bit, since this hoses
1885 * bookkeeping if things aren't delivered in order.
1887 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1888 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1889 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1891 if (hw->mac.type == ixgbe_mac_82599EB) {
1892 /* We enable 8 traffic classes, DCB only */
1893 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
1894 IXGBE_WRITE_REG(hw, IXGBE_MTQC, (IXGBE_MTQC_RT_ENA |
1895 IXGBE_MTQC_8TC_8TQ));
1899 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1901 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
1902 struct ixgbe_ring *rx_ring)
1906 struct ixgbe_ring_feature *feature = adapter->ring_feature;
1908 index = rx_ring->reg_idx;
1909 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1911 mask = (unsigned long) feature[RING_F_RSS].mask;
1912 index = index & mask;
1914 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
1916 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1917 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1919 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1920 IXGBE_SRRCTL_BSIZEHDR_MASK;
1922 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1923 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
1924 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1926 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1928 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1930 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
1931 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1932 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1935 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
1938 static u32 ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
1943 if (!(adapter->hw.mac.type == ixgbe_mac_82599EB))
1946 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
1947 #ifdef CONFIG_IXGBE_DCB
1948 | IXGBE_FLAG_DCB_ENABLED
1953 case (IXGBE_FLAG_RSS_ENABLED):
1954 mrqc = IXGBE_MRQC_RSSEN;
1956 #ifdef CONFIG_IXGBE_DCB
1957 case (IXGBE_FLAG_DCB_ENABLED):
1958 mrqc = IXGBE_MRQC_RT8TCEN;
1960 #endif /* CONFIG_IXGBE_DCB */
1969 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
1970 * @adapter: board private structure
1972 * Configure the Rx unit of the MAC after a reset.
1974 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
1977 struct ixgbe_hw *hw = &adapter->hw;
1978 struct ixgbe_ring *rx_ring;
1979 struct net_device *netdev = adapter->netdev;
1980 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1982 u32 rdlen, rxctrl, rxcsum;
1983 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
1984 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
1985 0x6A3E67EA, 0x14364D17, 0x3BED200D};
1987 u32 reta = 0, mrqc = 0;
1992 /* Decide whether to use packet split mode or not */
1993 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
1995 /* Set the RX buffer length according to the mode */
1996 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1997 rx_buf_len = IXGBE_RX_HDR_SIZE;
1998 if (hw->mac.type == ixgbe_mac_82599EB) {
1999 /* PSRTYPE must be initialized in 82599 */
2000 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2001 IXGBE_PSRTYPE_UDPHDR |
2002 IXGBE_PSRTYPE_IPV4HDR |
2003 IXGBE_PSRTYPE_IPV6HDR |
2004 IXGBE_PSRTYPE_L2HDR;
2005 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), psrtype);
2008 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2009 (netdev->mtu <= ETH_DATA_LEN))
2010 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2012 rx_buf_len = ALIGN(max_frame, 1024);
2015 fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
2016 fctrl |= IXGBE_FCTRL_BAM;
2017 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
2018 fctrl |= IXGBE_FCTRL_PMCF;
2019 IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
2021 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2022 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2023 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2025 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2027 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2028 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2030 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2032 rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
2033 /* disable receives while setting up the descriptors */
2034 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2035 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2038 * Setup the HW Rx Head and Tail Descriptor Pointers and
2039 * the Base and Length of the Rx Descriptor Ring
2041 for (i = 0; i < adapter->num_rx_queues; i++) {
2042 rx_ring = &adapter->rx_ring[i];
2043 rdba = rx_ring->dma;
2044 j = rx_ring->reg_idx;
2045 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_BIT_MASK(32)));
2046 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
2047 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
2048 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
2049 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
2050 rx_ring->head = IXGBE_RDH(j);
2051 rx_ring->tail = IXGBE_RDT(j);
2052 rx_ring->rx_buf_len = rx_buf_len;
2054 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2055 rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2058 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
2059 struct ixgbe_ring_feature *f;
2060 f = &adapter->ring_feature[RING_F_FCOE];
2061 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2062 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2063 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2064 rx_ring->rx_buf_len =
2065 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2069 #endif /* IXGBE_FCOE */
2070 ixgbe_configure_srrctl(adapter, rx_ring);
2073 if (hw->mac.type == ixgbe_mac_82598EB) {
2075 * For VMDq support of different descriptor types or
2076 * buffer sizes through the use of multiple SRRCTL
2077 * registers, RDRXCTL.MVMEN must be set to 1
2079 * also, the manual doesn't mention it clearly but DCA hints
2080 * will only use queue 0's tags unless this bit is set. Side
2081 * effects of setting this bit are only that SRRCTL must be
2082 * fully programmed [0..15]
2084 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2085 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2086 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2089 /* Program MRQC for the distribution of queues */
2090 mrqc = ixgbe_setup_mrqc(adapter);
2092 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2093 /* Fill out redirection table */
2094 for (i = 0, j = 0; i < 128; i++, j++) {
2095 if (j == adapter->ring_feature[RING_F_RSS].indices)
2097 /* reta = 4-byte sliding window of
2098 * 0x00..(indices-1)(indices-1)00..etc. */
2099 reta = (reta << 8) | (j * 0x11);
2101 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2104 /* Fill out hash function seeds */
2105 for (i = 0; i < 10; i++)
2106 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2108 if (hw->mac.type == ixgbe_mac_82598EB)
2109 mrqc |= IXGBE_MRQC_RSSEN;
2110 /* Perform hash on these packet types */
2111 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2112 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2113 | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
2114 | IXGBE_MRQC_RSS_FIELD_IPV6
2115 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
2116 | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2118 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2120 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2122 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
2123 adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
2124 /* Disable indicating checksum in descriptor, enables
2126 rxcsum |= IXGBE_RXCSUM_PCSD;
2128 if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
2129 /* Enable IPv4 payload checksum for UDP fragments
2130 * if PCSD is not set */
2131 rxcsum |= IXGBE_RXCSUM_IPPCSE;
2134 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2136 if (hw->mac.type == ixgbe_mac_82599EB) {
2137 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2138 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2139 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2140 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2143 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
2144 /* Enable 82599 HW-RSC */
2145 for (i = 0; i < adapter->num_rx_queues; i++) {
2146 rx_ring = &adapter->rx_ring[i];
2147 j = rx_ring->reg_idx;
2148 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2149 rscctrl |= IXGBE_RSCCTL_RSCEN;
2151 * we must limit the number of descriptors so that the
2152 * total size of max desc * buf_len is not greater
2155 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2156 #if (MAX_SKB_FRAGS > 16)
2157 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2158 #elif (MAX_SKB_FRAGS > 8)
2159 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2160 #elif (MAX_SKB_FRAGS > 4)
2161 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2163 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2166 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2167 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2168 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2169 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2171 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2173 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2175 /* Disable RSC for ACK packets */
2176 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2177 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2181 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2183 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2184 struct ixgbe_hw *hw = &adapter->hw;
2186 /* add VID to filter table */
2187 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
2190 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2192 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2193 struct ixgbe_hw *hw = &adapter->hw;
2195 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2196 ixgbe_irq_disable(adapter);
2198 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2200 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2201 ixgbe_irq_enable(adapter);
2203 /* remove VID from filter table */
2204 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
2207 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2208 struct vlan_group *grp)
2210 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2214 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2215 ixgbe_irq_disable(adapter);
2216 adapter->vlgrp = grp;
2219 * For a DCB driver, always enable VLAN tag stripping so we can
2220 * still receive traffic from a DCB-enabled host even if we're
2223 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2224 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2225 ctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2226 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2227 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2228 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2229 ctrl |= IXGBE_VLNCTRL_VFE;
2230 /* enable VLAN tag insert/strip */
2231 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2232 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2233 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2234 for (i = 0; i < adapter->num_rx_queues; i++) {
2235 j = adapter->rx_ring[i].reg_idx;
2236 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_RXDCTL(j));
2237 ctrl |= IXGBE_RXDCTL_VME;
2238 IXGBE_WRITE_REG(&adapter->hw, IXGBE_RXDCTL(j), ctrl);
2241 ixgbe_vlan_rx_add_vid(netdev, 0);
2243 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2244 ixgbe_irq_enable(adapter);
2247 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2249 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2251 if (adapter->vlgrp) {
2253 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2254 if (!vlan_group_get_device(adapter->vlgrp, vid))
2256 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2261 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
2263 struct dev_mc_list *mc_ptr;
2264 u8 *addr = *mc_addr_ptr;
2267 mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
2269 *mc_addr_ptr = mc_ptr->next->dmi_addr;
2271 *mc_addr_ptr = NULL;
2277 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
2278 * @netdev: network interface device structure
2280 * The set_rx_method entry point is called whenever the unicast/multicast
2281 * address list or the network interface flags are updated. This routine is
2282 * responsible for configuring the hardware for proper unicast, multicast and
2285 static void ixgbe_set_rx_mode(struct net_device *netdev)
2287 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2288 struct ixgbe_hw *hw = &adapter->hw;
2290 u8 *addr_list = NULL;
2293 /* Check for Promiscuous and All Multicast modes */
2295 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2296 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2298 if (netdev->flags & IFF_PROMISC) {
2299 hw->addr_ctrl.user_set_promisc = 1;
2300 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2301 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2303 if (netdev->flags & IFF_ALLMULTI) {
2304 fctrl |= IXGBE_FCTRL_MPE;
2305 fctrl &= ~IXGBE_FCTRL_UPE;
2307 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2309 vlnctrl |= IXGBE_VLNCTRL_VFE;
2310 hw->addr_ctrl.user_set_promisc = 0;
2313 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
2314 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2316 /* reprogram secondary unicast list */
2317 hw->mac.ops.update_uc_addr_list(hw, &netdev->uc.list);
2319 /* reprogram multicast list */
2320 addr_count = netdev->mc_count;
2322 addr_list = netdev->mc_list->dmi_addr;
2323 hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
2324 ixgbe_addr_list_itr);
2327 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
2330 struct ixgbe_q_vector *q_vector;
2331 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2333 /* legacy and MSI only use one vector */
2334 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2337 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2338 struct napi_struct *napi;
2339 q_vector = adapter->q_vector[q_idx];
2340 napi = &q_vector->napi;
2341 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2342 if (!q_vector->rxr_count || !q_vector->txr_count) {
2343 if (q_vector->txr_count == 1)
2344 napi->poll = &ixgbe_clean_txonly;
2345 else if (q_vector->rxr_count == 1)
2346 napi->poll = &ixgbe_clean_rxonly;
2354 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2357 struct ixgbe_q_vector *q_vector;
2358 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2360 /* legacy and MSI only use one vector */
2361 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2364 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2365 q_vector = adapter->q_vector[q_idx];
2366 napi_disable(&q_vector->napi);
2370 #ifdef CONFIG_IXGBE_DCB
2372 * ixgbe_configure_dcb - Configure DCB hardware
2373 * @adapter: ixgbe adapter struct
2375 * This is called by the driver on open to configure the DCB hardware.
2376 * This is also called by the gennetlink interface when reconfiguring
2379 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2381 struct ixgbe_hw *hw = &adapter->hw;
2382 u32 txdctl, vlnctrl;
2385 ixgbe_dcb_check_config(&adapter->dcb_cfg);
2386 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2387 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2389 /* reconfigure the hardware */
2390 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
2392 for (i = 0; i < adapter->num_tx_queues; i++) {
2393 j = adapter->tx_ring[i].reg_idx;
2394 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2395 /* PThresh workaround for Tx hang with DFP enabled. */
2397 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2399 /* Enable VLAN tag insert/strip */
2400 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2401 if (hw->mac.type == ixgbe_mac_82598EB) {
2402 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2403 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2404 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2405 } else if (hw->mac.type == ixgbe_mac_82599EB) {
2406 vlnctrl |= IXGBE_VLNCTRL_VFE;
2407 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2408 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2409 for (i = 0; i < adapter->num_rx_queues; i++) {
2410 j = adapter->rx_ring[i].reg_idx;
2411 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2412 vlnctrl |= IXGBE_RXDCTL_VME;
2413 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2416 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2420 static void ixgbe_configure(struct ixgbe_adapter *adapter)
2422 struct net_device *netdev = adapter->netdev;
2423 struct ixgbe_hw *hw = &adapter->hw;
2426 ixgbe_set_rx_mode(netdev);
2428 ixgbe_restore_vlan(adapter);
2429 #ifdef CONFIG_IXGBE_DCB
2430 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2431 netif_set_gso_max_size(netdev, 32768);
2432 ixgbe_configure_dcb(adapter);
2434 netif_set_gso_max_size(netdev, 65536);
2437 netif_set_gso_max_size(netdev, 65536);
2441 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2442 ixgbe_configure_fcoe(adapter);
2444 #endif /* IXGBE_FCOE */
2445 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2446 for (i = 0; i < adapter->num_tx_queues; i++)
2447 adapter->tx_ring[i].atr_sample_rate =
2448 adapter->atr_sample_rate;
2449 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
2450 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
2451 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
2454 ixgbe_configure_tx(adapter);
2455 ixgbe_configure_rx(adapter);
2456 for (i = 0; i < adapter->num_rx_queues; i++)
2457 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
2458 (adapter->rx_ring[i].count - 1));
2461 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2463 switch (hw->phy.type) {
2464 case ixgbe_phy_sfp_avago:
2465 case ixgbe_phy_sfp_ftl:
2466 case ixgbe_phy_sfp_intel:
2467 case ixgbe_phy_sfp_unknown:
2468 case ixgbe_phy_tw_tyco:
2469 case ixgbe_phy_tw_unknown:
2477 * ixgbe_sfp_link_config - set up SFP+ link
2478 * @adapter: pointer to private adapter struct
2480 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2482 struct ixgbe_hw *hw = &adapter->hw;
2484 if (hw->phy.multispeed_fiber) {
2486 * In multispeed fiber setups, the device may not have
2487 * had a physical connection when the driver loaded.
2488 * If that's the case, the initial link configuration
2489 * couldn't get the MAC into 10G or 1G mode, so we'll
2490 * never have a link status change interrupt fire.
2491 * We need to try and force an autonegotiation
2492 * session, then bring up link.
2494 hw->mac.ops.setup_sfp(hw);
2495 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2496 schedule_work(&adapter->multispeed_fiber_task);
2499 * Direct Attach Cu and non-multispeed fiber modules
2500 * still need to be configured properly prior to
2503 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2504 schedule_work(&adapter->sfp_config_module_task);
2509 * ixgbe_non_sfp_link_config - set up non-SFP+ link
2510 * @hw: pointer to private hardware struct
2512 * Returns 0 on success, negative on failure
2514 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2517 bool link_up = false;
2518 u32 ret = IXGBE_ERR_LINK_SETUP;
2520 if (hw->mac.ops.check_link)
2521 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2526 if (hw->mac.ops.get_link_capabilities)
2527 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
2532 if (hw->mac.ops.setup_link_speed)
2533 ret = hw->mac.ops.setup_link_speed(hw, autoneg, true, link_up);
2538 #define IXGBE_MAX_RX_DESC_POLL 10
2539 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2542 int j = adapter->rx_ring[rxr].reg_idx;
2545 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2546 if (IXGBE_READ_REG(&adapter->hw,
2547 IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2552 if (k >= IXGBE_MAX_RX_DESC_POLL) {
2553 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2554 "not set within the polling period\n", rxr);
2556 ixgbe_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
2557 (adapter->rx_ring[rxr].count - 1));
2560 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2562 struct net_device *netdev = adapter->netdev;
2563 struct ixgbe_hw *hw = &adapter->hw;
2565 int num_rx_rings = adapter->num_rx_queues;
2567 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2568 u32 txdctl, rxdctl, mhadd;
2572 ixgbe_get_hw_control(adapter);
2574 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2575 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2576 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2577 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2578 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2583 /* XXX: to interrupt immediately for EICS writes, enable this */
2584 /* gpie |= IXGBE_GPIE_EIMEN; */
2585 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2588 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2589 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2590 * specifically only auto mask tx and rx interrupts */
2591 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2594 /* Enable fan failure interrupt if media type is copper */
2595 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2596 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2597 gpie |= IXGBE_SDP1_GPIEN;
2598 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2601 if (hw->mac.type == ixgbe_mac_82599EB) {
2602 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2603 gpie |= IXGBE_SDP1_GPIEN;
2604 gpie |= IXGBE_SDP2_GPIEN;
2605 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2609 /* adjust max frame to be able to do baby jumbo for FCoE */
2610 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2611 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2612 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2614 #endif /* IXGBE_FCOE */
2615 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2616 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2617 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2618 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2620 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2623 for (i = 0; i < adapter->num_tx_queues; i++) {
2624 j = adapter->tx_ring[i].reg_idx;
2625 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2626 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2627 txdctl |= (8 << 16);
2628 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2631 if (hw->mac.type == ixgbe_mac_82599EB) {
2632 /* DMATXCTL.EN must be set after all Tx queue config is done */
2633 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2634 dmatxctl |= IXGBE_DMATXCTL_TE;
2635 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2637 for (i = 0; i < adapter->num_tx_queues; i++) {
2638 j = adapter->tx_ring[i].reg_idx;
2639 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2640 txdctl |= IXGBE_TXDCTL_ENABLE;
2641 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2644 for (i = 0; i < num_rx_rings; i++) {
2645 j = adapter->rx_ring[i].reg_idx;
2646 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2647 /* enable PTHRESH=32 descriptors (half the internal cache)
2648 * and HTHRESH=0 descriptors (to minimize latency on fetch),
2649 * this also removes a pesky rx_no_buffer_count increment */
2651 rxdctl |= IXGBE_RXDCTL_ENABLE;
2652 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2653 if (hw->mac.type == ixgbe_mac_82599EB)
2654 ixgbe_rx_desc_queue_enable(adapter, i);
2656 /* enable all receives */
2657 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2658 if (hw->mac.type == ixgbe_mac_82598EB)
2659 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2661 rxdctl |= IXGBE_RXCTRL_RXEN;
2662 hw->mac.ops.enable_rx_dma(hw, rxdctl);
2664 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2665 ixgbe_configure_msix(adapter);
2667 ixgbe_configure_msi_and_legacy(adapter);
2669 clear_bit(__IXGBE_DOWN, &adapter->state);
2670 ixgbe_napi_enable_all(adapter);
2672 /* clear any pending interrupts, may auto mask */
2673 IXGBE_READ_REG(hw, IXGBE_EICR);
2675 ixgbe_irq_enable(adapter);
2678 * If this adapter has a fan, check to see if we had a failure
2679 * before we enabled the interrupt.
2681 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2682 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
2683 if (esdp & IXGBE_ESDP_SDP1)
2685 "Fan has stopped, replace the adapter\n");
2689 * For hot-pluggable SFP+ devices, a new SFP+ module may have
2690 * arrived before interrupts were enabled but after probe. Such
2691 * devices wouldn't have their type identified yet. We need to
2692 * kick off the SFP+ module setup first, then try to bring up link.
2693 * If we're not hot-pluggable SFP+, we just need to configure link
2696 if (hw->phy.type == ixgbe_phy_unknown) {
2697 err = hw->phy.ops.identify(hw);
2698 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
2700 * Take the device down and schedule the sfp tasklet
2701 * which will unregister_netdev and log it.
2703 ixgbe_down(adapter);
2704 schedule_work(&adapter->sfp_config_module_task);
2709 if (ixgbe_is_sfp(hw)) {
2710 ixgbe_sfp_link_config(adapter);
2712 err = ixgbe_non_sfp_link_config(hw);
2714 DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
2717 for (i = 0; i < adapter->num_tx_queues; i++)
2718 set_bit(__IXGBE_FDIR_INIT_DONE,
2719 &(adapter->tx_ring[i].reinit_state));
2721 /* enable transmits */
2722 netif_tx_start_all_queues(netdev);
2724 /* bring the link up in the watchdog, this could race with our first
2725 * link up interrupt but shouldn't be a problem */
2726 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2727 adapter->link_check_timeout = jiffies;
2728 mod_timer(&adapter->watchdog_timer, jiffies);
2732 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
2734 WARN_ON(in_interrupt());
2735 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
2737 ixgbe_down(adapter);
2739 clear_bit(__IXGBE_RESETTING, &adapter->state);
2742 int ixgbe_up(struct ixgbe_adapter *adapter)
2744 /* hardware has been reset, we need to reload some things */
2745 ixgbe_configure(adapter);
2747 return ixgbe_up_complete(adapter);
2750 void ixgbe_reset(struct ixgbe_adapter *adapter)
2752 struct ixgbe_hw *hw = &adapter->hw;
2755 err = hw->mac.ops.init_hw(hw);
2758 case IXGBE_ERR_SFP_NOT_PRESENT:
2760 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
2761 dev_err(&adapter->pdev->dev, "master disable timed out\n");
2763 case IXGBE_ERR_EEPROM_VERSION:
2764 /* We are running on a pre-production device, log a warning */
2765 dev_warn(&adapter->pdev->dev, "This device is a pre-production "
2766 "adapter/LOM. Please be aware there may be issues "
2767 "associated with your hardware. If you are "
2768 "experiencing problems please contact your Intel or "
2769 "hardware representative who provided you with this "
2773 dev_err(&adapter->pdev->dev, "Hardware Error: %d\n", err);
2776 /* reprogram the RAR[0] in case user changed it. */
2777 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2781 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2782 * @adapter: board private structure
2783 * @rx_ring: ring to free buffers from
2785 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2786 struct ixgbe_ring *rx_ring)
2788 struct pci_dev *pdev = adapter->pdev;
2792 /* Free all the Rx ring sk_buffs */
2794 for (i = 0; i < rx_ring->count; i++) {
2795 struct ixgbe_rx_buffer *rx_buffer_info;
2797 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2798 if (rx_buffer_info->dma) {
2799 pci_unmap_single(pdev, rx_buffer_info->dma,
2800 rx_ring->rx_buf_len,
2801 PCI_DMA_FROMDEVICE);
2802 rx_buffer_info->dma = 0;
2804 if (rx_buffer_info->skb) {
2805 struct sk_buff *skb = rx_buffer_info->skb;
2806 rx_buffer_info->skb = NULL;
2808 struct sk_buff *this = skb;
2810 dev_kfree_skb(this);
2813 if (!rx_buffer_info->page)
2815 if (rx_buffer_info->page_dma) {
2816 pci_unmap_page(pdev, rx_buffer_info->page_dma,
2817 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
2818 rx_buffer_info->page_dma = 0;
2820 put_page(rx_buffer_info->page);
2821 rx_buffer_info->page = NULL;
2822 rx_buffer_info->page_offset = 0;
2825 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2826 memset(rx_ring->rx_buffer_info, 0, size);
2828 /* Zero out the descriptor ring */
2829 memset(rx_ring->desc, 0, rx_ring->size);
2831 rx_ring->next_to_clean = 0;
2832 rx_ring->next_to_use = 0;
2835 writel(0, adapter->hw.hw_addr + rx_ring->head);
2837 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2841 * ixgbe_clean_tx_ring - Free Tx Buffers
2842 * @adapter: board private structure
2843 * @tx_ring: ring to be cleaned
2845 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2846 struct ixgbe_ring *tx_ring)
2848 struct ixgbe_tx_buffer *tx_buffer_info;
2852 /* Free all the Tx ring sk_buffs */
2854 for (i = 0; i < tx_ring->count; i++) {
2855 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2856 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2859 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2860 memset(tx_ring->tx_buffer_info, 0, size);
2862 /* Zero out the descriptor ring */
2863 memset(tx_ring->desc, 0, tx_ring->size);
2865 tx_ring->next_to_use = 0;
2866 tx_ring->next_to_clean = 0;
2869 writel(0, adapter->hw.hw_addr + tx_ring->head);
2871 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2875 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2876 * @adapter: board private structure
2878 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2882 for (i = 0; i < adapter->num_rx_queues; i++)
2883 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2887 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2888 * @adapter: board private structure
2890 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2894 for (i = 0; i < adapter->num_tx_queues; i++)
2895 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2898 void ixgbe_down(struct ixgbe_adapter *adapter)
2900 struct net_device *netdev = adapter->netdev;
2901 struct ixgbe_hw *hw = &adapter->hw;
2906 /* signal that we are down to the interrupt handler */
2907 set_bit(__IXGBE_DOWN, &adapter->state);
2909 /* disable receives */
2910 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2911 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2913 netif_tx_disable(netdev);
2915 IXGBE_WRITE_FLUSH(hw);
2918 netif_tx_stop_all_queues(netdev);
2920 ixgbe_irq_disable(adapter);
2922 ixgbe_napi_disable_all(adapter);
2924 del_timer_sync(&adapter->watchdog_timer);
2925 cancel_work_sync(&adapter->watchdog_task);
2927 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2928 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2929 cancel_work_sync(&adapter->fdir_reinit_task);
2931 /* disable transmits in the hardware now that interrupts are off */
2932 for (i = 0; i < adapter->num_tx_queues; i++) {
2933 j = adapter->tx_ring[i].reg_idx;
2934 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2935 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
2936 (txdctl & ~IXGBE_TXDCTL_ENABLE));
2938 /* Disable the Tx DMA engine on 82599 */
2939 if (hw->mac.type == ixgbe_mac_82599EB)
2940 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
2941 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
2942 ~IXGBE_DMATXCTL_TE));
2944 netif_carrier_off(netdev);
2946 if (!pci_channel_offline(adapter->pdev))
2947 ixgbe_reset(adapter);
2948 ixgbe_clean_all_tx_rings(adapter);
2949 ixgbe_clean_all_rx_rings(adapter);
2951 #ifdef CONFIG_IXGBE_DCA
2952 /* since we reset the hardware DCA settings were cleared */
2953 ixgbe_setup_dca(adapter);
2958 * ixgbe_poll - NAPI Rx polling callback
2959 * @napi: structure for representing this polling device
2960 * @budget: how many packets driver is allowed to clean
2962 * This function is used for legacy and MSI, NAPI mode
2964 static int ixgbe_poll(struct napi_struct *napi, int budget)
2966 struct ixgbe_q_vector *q_vector =
2967 container_of(napi, struct ixgbe_q_vector, napi);
2968 struct ixgbe_adapter *adapter = q_vector->adapter;
2969 int tx_clean_complete, work_done = 0;
2971 #ifdef CONFIG_IXGBE_DCA
2972 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2973 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
2974 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
2978 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring);
2979 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring, &work_done, budget);
2981 if (!tx_clean_complete)
2984 /* If budget not fully consumed, exit the polling mode */
2985 if (work_done < budget) {
2986 napi_complete(napi);
2987 if (adapter->itr_setting & 1)
2988 ixgbe_set_itr(adapter);
2989 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2990 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
2996 * ixgbe_tx_timeout - Respond to a Tx Hang
2997 * @netdev: network interface device structure
2999 static void ixgbe_tx_timeout(struct net_device *netdev)
3001 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3003 /* Do the reset outside of interrupt context */
3004 schedule_work(&adapter->reset_task);
3007 static void ixgbe_reset_task(struct work_struct *work)
3009 struct ixgbe_adapter *adapter;
3010 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3012 /* If we're already down or resetting, just bail */
3013 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3014 test_bit(__IXGBE_RESETTING, &adapter->state))
3017 adapter->tx_timeout_count++;
3019 ixgbe_reinit_locked(adapter);
3022 #ifdef CONFIG_IXGBE_DCB
3023 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3026 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3028 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3032 adapter->num_rx_queues = f->indices;
3033 adapter->num_tx_queues = f->indices;
3041 * ixgbe_set_rss_queues: Allocate queues for RSS
3042 * @adapter: board private structure to initialize
3044 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3045 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3048 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3051 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3053 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3055 adapter->num_rx_queues = f->indices;
3056 adapter->num_tx_queues = f->indices;
3066 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3067 * @adapter: board private structure to initialize
3069 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3070 * to the original CPU that initiated the Tx session. This runs in addition
3071 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3072 * Rx load across CPUs using RSS.
3075 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3078 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3080 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3083 /* Flow Director must have RSS enabled */
3084 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3085 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3086 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3087 adapter->num_tx_queues = f_fdir->indices;
3088 adapter->num_rx_queues = f_fdir->indices;
3091 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3092 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3099 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3100 * @adapter: board private structure to initialize
3102 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3103 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3104 * rx queues out of the max number of rx queues, instead, it is used as the
3105 * index of the first rx queue used by FCoE.
3108 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3111 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3113 f->indices = min((int)num_online_cpus(), f->indices);
3114 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3115 #ifdef CONFIG_IXGBE_DCB
3116 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3117 DPRINTK(PROBE, INFO, "FCOE enabled with DCB \n");
3118 ixgbe_set_dcb_queues(adapter);
3121 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3122 DPRINTK(PROBE, INFO, "FCOE enabled with RSS \n");
3123 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3124 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3125 ixgbe_set_fdir_queues(adapter);
3127 ixgbe_set_rss_queues(adapter);
3129 /* adding FCoE rx rings to the end */
3130 f->mask = adapter->num_rx_queues;
3131 adapter->num_rx_queues += f->indices;
3132 if (adapter->num_tx_queues == 0)
3133 adapter->num_tx_queues = f->indices;
3141 #endif /* IXGBE_FCOE */
3143 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3144 * @adapter: board private structure to initialize
3146 * This is the top level queue allocation routine. The order here is very
3147 * important, starting with the "most" number of features turned on at once,
3148 * and ending with the smallest set of features. This way large combinations
3149 * can be allocated if they're turned on, and smaller combinations are the
3150 * fallthrough conditions.
3153 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
3156 if (ixgbe_set_fcoe_queues(adapter))
3159 #endif /* IXGBE_FCOE */
3160 #ifdef CONFIG_IXGBE_DCB
3161 if (ixgbe_set_dcb_queues(adapter))
3165 if (ixgbe_set_fdir_queues(adapter))
3168 if (ixgbe_set_rss_queues(adapter))
3171 /* fallback to base case */
3172 adapter->num_rx_queues = 1;
3173 adapter->num_tx_queues = 1;
3176 /* Notify the stack of the (possibly) reduced Tx Queue count. */
3177 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
3180 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
3183 int err, vector_threshold;
3185 /* We'll want at least 3 (vector_threshold):
3188 * 3) Other (Link Status Change, etc.)
3189 * 4) TCP Timer (optional)
3191 vector_threshold = MIN_MSIX_COUNT;
3193 /* The more we get, the more we will assign to Tx/Rx Cleanup
3194 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
3195 * Right now, we simply care about how many we'll get; we'll
3196 * set them up later while requesting irq's.
3198 while (vectors >= vector_threshold) {
3199 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
3201 if (!err) /* Success in acquiring all requested vectors. */
3204 vectors = 0; /* Nasty failure, quit now */
3205 else /* err == number of vectors we should try again with */
3209 if (vectors < vector_threshold) {
3210 /* Can't allocate enough MSI-X interrupts? Oh well.
3211 * This just means we'll go with either a single MSI
3212 * vector or fall back to legacy interrupts.
3214 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
3215 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3216 kfree(adapter->msix_entries);
3217 adapter->msix_entries = NULL;
3219 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
3221 * Adjust for only the vectors we'll use, which is minimum
3222 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
3223 * vectors we were allocated.
3225 adapter->num_msix_vectors = min(vectors,
3226 adapter->max_msix_q_vectors + NON_Q_VECTORS);
3231 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
3232 * @adapter: board private structure to initialize
3234 * Cache the descriptor ring offsets for RSS to the assigned rings.
3237 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
3242 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3243 for (i = 0; i < adapter->num_rx_queues; i++)
3244 adapter->rx_ring[i].reg_idx = i;
3245 for (i = 0; i < adapter->num_tx_queues; i++)
3246 adapter->tx_ring[i].reg_idx = i;
3255 #ifdef CONFIG_IXGBE_DCB
3257 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
3258 * @adapter: board private structure to initialize
3260 * Cache the descriptor ring offsets for DCB to the assigned rings.
3263 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
3267 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
3269 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3270 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3271 /* the number of queues is assumed to be symmetric */
3272 for (i = 0; i < dcb_i; i++) {
3273 adapter->rx_ring[i].reg_idx = i << 3;
3274 adapter->tx_ring[i].reg_idx = i << 2;
3277 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
3280 * Tx TC0 starts at: descriptor queue 0
3281 * Tx TC1 starts at: descriptor queue 32
3282 * Tx TC2 starts at: descriptor queue 64
3283 * Tx TC3 starts at: descriptor queue 80
3284 * Tx TC4 starts at: descriptor queue 96
3285 * Tx TC5 starts at: descriptor queue 104
3286 * Tx TC6 starts at: descriptor queue 112
3287 * Tx TC7 starts at: descriptor queue 120
3289 * Rx TC0-TC7 are offset by 16 queues each
3291 for (i = 0; i < 3; i++) {
3292 adapter->tx_ring[i].reg_idx = i << 5;
3293 adapter->rx_ring[i].reg_idx = i << 4;
3295 for ( ; i < 5; i++) {
3296 adapter->tx_ring[i].reg_idx =
3298 adapter->rx_ring[i].reg_idx = i << 4;
3300 for ( ; i < dcb_i; i++) {
3301 adapter->tx_ring[i].reg_idx =
3303 adapter->rx_ring[i].reg_idx = i << 4;
3307 } else if (dcb_i == 4) {
3309 * Tx TC0 starts at: descriptor queue 0
3310 * Tx TC1 starts at: descriptor queue 64
3311 * Tx TC2 starts at: descriptor queue 96
3312 * Tx TC3 starts at: descriptor queue 112
3314 * Rx TC0-TC3 are offset by 32 queues each
3316 adapter->tx_ring[0].reg_idx = 0;
3317 adapter->tx_ring[1].reg_idx = 64;
3318 adapter->tx_ring[2].reg_idx = 96;
3319 adapter->tx_ring[3].reg_idx = 112;
3320 for (i = 0 ; i < dcb_i; i++)
3321 adapter->rx_ring[i].reg_idx = i << 5;
3339 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
3340 * @adapter: board private structure to initialize
3342 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
3345 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
3350 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3351 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3352 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
3353 for (i = 0; i < adapter->num_rx_queues; i++)
3354 adapter->rx_ring[i].reg_idx = i;
3355 for (i = 0; i < adapter->num_tx_queues; i++)
3356 adapter->tx_ring[i].reg_idx = i;
3365 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
3366 * @adapter: board private structure to initialize
3368 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
3371 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
3375 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3377 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3378 #ifdef CONFIG_IXGBE_DCB
3379 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3380 ixgbe_cache_ring_dcb(adapter);
3381 fcoe_i = adapter->rx_ring[0].reg_idx + 1;
3383 #endif /* CONFIG_IXGBE_DCB */
3384 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3385 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3386 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3387 ixgbe_cache_ring_fdir(adapter);
3389 ixgbe_cache_ring_rss(adapter);
3393 for (i = 0; i < f->indices; i++, fcoe_i++)
3394 adapter->rx_ring[f->mask + i].reg_idx = fcoe_i;
3400 #endif /* IXGBE_FCOE */
3402 * ixgbe_cache_ring_register - Descriptor ring to register mapping
3403 * @adapter: board private structure to initialize
3405 * Once we know the feature-set enabled for the device, we'll cache
3406 * the register offset the descriptor ring is assigned to.
3408 * Note, the order the various feature calls is important. It must start with
3409 * the "most" features enabled at the same time, then trickle down to the
3410 * least amount of features turned on at once.
3412 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
3414 /* start with default case */
3415 adapter->rx_ring[0].reg_idx = 0;
3416 adapter->tx_ring[0].reg_idx = 0;
3419 if (ixgbe_cache_ring_fcoe(adapter))
3422 #endif /* IXGBE_FCOE */
3423 #ifdef CONFIG_IXGBE_DCB
3424 if (ixgbe_cache_ring_dcb(adapter))
3428 if (ixgbe_cache_ring_fdir(adapter))
3431 if (ixgbe_cache_ring_rss(adapter))
3436 * ixgbe_alloc_queues - Allocate memory for all rings
3437 * @adapter: board private structure to initialize
3439 * We allocate one ring per queue at run-time since we don't know the
3440 * number of queues at compile-time. The polling_netdev array is
3441 * intended for Multiqueue, but should work fine with a single queue.
3443 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
3447 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
3448 sizeof(struct ixgbe_ring), GFP_KERNEL);
3449 if (!adapter->tx_ring)
3450 goto err_tx_ring_allocation;
3452 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
3453 sizeof(struct ixgbe_ring), GFP_KERNEL);
3454 if (!adapter->rx_ring)
3455 goto err_rx_ring_allocation;
3457 for (i = 0; i < adapter->num_tx_queues; i++) {
3458 adapter->tx_ring[i].count = adapter->tx_ring_count;
3459 adapter->tx_ring[i].queue_index = i;
3462 for (i = 0; i < adapter->num_rx_queues; i++) {
3463 adapter->rx_ring[i].count = adapter->rx_ring_count;
3464 adapter->rx_ring[i].queue_index = i;
3467 ixgbe_cache_ring_register(adapter);
3471 err_rx_ring_allocation:
3472 kfree(adapter->tx_ring);
3473 err_tx_ring_allocation:
3478 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
3479 * @adapter: board private structure to initialize
3481 * Attempt to configure the interrupts using the best available
3482 * capabilities of the hardware and the kernel.
3484 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
3486 struct ixgbe_hw *hw = &adapter->hw;
3488 int vector, v_budget;
3491 * It's easy to be greedy for MSI-X vectors, but it really
3492 * doesn't do us much good if we have a lot more vectors
3493 * than CPU's. So let's be conservative and only ask for
3494 * (roughly) twice the number of vectors as there are CPU's.
3496 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
3497 (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
3500 * At the same time, hardware can only support a maximum of
3501 * hw.mac->max_msix_vectors vectors. With features
3502 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
3503 * descriptor queues supported by our device. Thus, we cap it off in
3504 * those rare cases where the cpu count also exceeds our vector limit.
3506 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
3508 /* A failure in MSI-X entry allocation isn't fatal, but it does
3509 * mean we disable MSI-X capabilities of the adapter. */
3510 adapter->msix_entries = kcalloc(v_budget,
3511 sizeof(struct msix_entry), GFP_KERNEL);
3512 if (adapter->msix_entries) {
3513 for (vector = 0; vector < v_budget; vector++)
3514 adapter->msix_entries[vector].entry = vector;
3516 ixgbe_acquire_msix_vectors(adapter, v_budget);
3518 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3522 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
3523 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
3524 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3525 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3526 adapter->atr_sample_rate = 0;
3527 ixgbe_set_num_queues(adapter);
3529 err = pci_enable_msi(adapter->pdev);
3531 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
3533 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
3534 "falling back to legacy. Error: %d\n", err);
3544 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
3545 * @adapter: board private structure to initialize
3547 * We allocate one q_vector per queue interrupt. If allocation fails we
3550 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
3552 int q_idx, num_q_vectors;
3553 struct ixgbe_q_vector *q_vector;
3555 int (*poll)(struct napi_struct *, int);
3557 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3558 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3559 napi_vectors = adapter->num_rx_queues;
3560 poll = &ixgbe_clean_rxtx_many;
3567 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3568 q_vector = kzalloc(sizeof(struct ixgbe_q_vector), GFP_KERNEL);
3571 q_vector->adapter = adapter;
3572 q_vector->eitr = adapter->eitr_param;
3573 q_vector->v_idx = q_idx;
3574 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
3575 adapter->q_vector[q_idx] = q_vector;
3583 q_vector = adapter->q_vector[q_idx];
3584 netif_napi_del(&q_vector->napi);
3586 adapter->q_vector[q_idx] = NULL;
3592 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
3593 * @adapter: board private structure to initialize
3595 * This function frees the memory allocated to the q_vectors. In addition if
3596 * NAPI is enabled it will delete any references to the NAPI struct prior
3597 * to freeing the q_vector.
3599 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
3601 int q_idx, num_q_vectors;
3603 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3604 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3608 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3609 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
3610 adapter->q_vector[q_idx] = NULL;
3611 netif_napi_del(&q_vector->napi);
3616 void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
3618 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3619 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3620 pci_disable_msix(adapter->pdev);
3621 kfree(adapter->msix_entries);
3622 adapter->msix_entries = NULL;
3623 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
3624 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
3625 pci_disable_msi(adapter->pdev);
3631 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
3632 * @adapter: board private structure to initialize
3634 * We determine which interrupt scheme to use based on...
3635 * - Kernel support (MSI, MSI-X)
3636 * - which can be user-defined (via MODULE_PARAM)
3637 * - Hardware queue count (num_*_queues)
3638 * - defined by miscellaneous hardware support/features (RSS, etc.)
3640 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
3644 /* Number of supported queues */
3645 ixgbe_set_num_queues(adapter);
3647 err = ixgbe_set_interrupt_capability(adapter);
3649 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
3650 goto err_set_interrupt;
3653 err = ixgbe_alloc_q_vectors(adapter);
3655 DPRINTK(PROBE, ERR, "Unable to allocate memory for queue "
3657 goto err_alloc_q_vectors;
3660 err = ixgbe_alloc_queues(adapter);
3662 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
3663 goto err_alloc_queues;
3666 DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
3667 "Tx Queue count = %u\n",
3668 (adapter->num_rx_queues > 1) ? "Enabled" :
3669 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
3671 set_bit(__IXGBE_DOWN, &adapter->state);
3676 ixgbe_free_q_vectors(adapter);
3677 err_alloc_q_vectors:
3678 ixgbe_reset_interrupt_capability(adapter);
3684 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
3685 * @adapter: board private structure to clear interrupt scheme on
3687 * We go through and clear interrupt specific resources and reset the structure
3688 * to pre-load conditions
3690 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
3692 kfree(adapter->tx_ring);
3693 kfree(adapter->rx_ring);
3694 adapter->tx_ring = NULL;
3695 adapter->rx_ring = NULL;
3697 ixgbe_free_q_vectors(adapter);
3698 ixgbe_reset_interrupt_capability(adapter);
3702 * ixgbe_sfp_timer - worker thread to find a missing module
3703 * @data: pointer to our adapter struct
3705 static void ixgbe_sfp_timer(unsigned long data)
3707 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3710 * Do the sfp_timer outside of interrupt context due to the
3711 * delays that sfp+ detection requires
3713 schedule_work(&adapter->sfp_task);
3717 * ixgbe_sfp_task - worker thread to find a missing module
3718 * @work: pointer to work_struct containing our data
3720 static void ixgbe_sfp_task(struct work_struct *work)
3722 struct ixgbe_adapter *adapter = container_of(work,
3723 struct ixgbe_adapter,
3725 struct ixgbe_hw *hw = &adapter->hw;
3727 if ((hw->phy.type == ixgbe_phy_nl) &&
3728 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
3729 s32 ret = hw->phy.ops.identify_sfp(hw);
3730 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
3732 ret = hw->phy.ops.reset(hw);
3733 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3734 dev_err(&adapter->pdev->dev, "failed to initialize "
3735 "because an unsupported SFP+ module type "
3737 "Reload the driver after installing a "
3738 "supported module.\n");
3739 unregister_netdev(adapter->netdev);
3741 DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
3744 /* don't need this routine any more */
3745 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3749 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
3750 mod_timer(&adapter->sfp_timer,
3751 round_jiffies(jiffies + (2 * HZ)));
3755 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
3756 * @adapter: board private structure to initialize
3758 * ixgbe_sw_init initializes the Adapter private data structure.
3759 * Fields are initialized based on PCI device information and
3760 * OS network device settings (MTU size).
3762 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
3764 struct ixgbe_hw *hw = &adapter->hw;
3765 struct pci_dev *pdev = adapter->pdev;
3767 #ifdef CONFIG_IXGBE_DCB
3769 struct tc_configuration *tc;
3772 /* PCI config space info */
3774 hw->vendor_id = pdev->vendor;
3775 hw->device_id = pdev->device;
3776 hw->revision_id = pdev->revision;
3777 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3778 hw->subsystem_device_id = pdev->subsystem_device;
3780 /* Set capability flags */
3781 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
3782 adapter->ring_feature[RING_F_RSS].indices = rss;
3783 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
3784 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
3785 if (hw->mac.type == ixgbe_mac_82598EB) {
3786 if (hw->device_id == IXGBE_DEV_ID_82598AT)
3787 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
3788 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
3789 } else if (hw->mac.type == ixgbe_mac_82599EB) {
3790 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
3791 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
3792 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
3793 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
3794 adapter->ring_feature[RING_F_FDIR].indices =
3795 IXGBE_MAX_FDIR_INDICES;
3796 adapter->atr_sample_rate = 20;
3797 adapter->fdir_pballoc = 0;
3799 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
3800 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
3801 adapter->ring_feature[RING_F_FCOE].indices = 0;
3802 #endif /* IXGBE_FCOE */
3805 #ifdef CONFIG_IXGBE_DCB
3806 /* Configure DCB traffic classes */
3807 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
3808 tc = &adapter->dcb_cfg.tc_config[j];
3809 tc->path[DCB_TX_CONFIG].bwg_id = 0;
3810 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
3811 tc->path[DCB_RX_CONFIG].bwg_id = 0;
3812 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
3813 tc->dcb_pfc = pfc_disabled;
3815 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
3816 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
3817 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
3818 adapter->dcb_cfg.pfc_mode_enable = false;
3819 adapter->dcb_cfg.round_robin_enable = false;
3820 adapter->dcb_set_bitmap = 0x00;
3821 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
3822 adapter->ring_feature[RING_F_DCB].indices);
3826 /* default flow control settings */
3827 hw->fc.requested_mode = ixgbe_fc_full;
3828 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
3830 adapter->last_lfc_mode = hw->fc.current_mode;
3832 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
3833 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
3834 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
3835 hw->fc.send_xon = true;
3836 hw->fc.disable_fc_autoneg = false;
3838 /* enable itr by default in dynamic mode */
3839 adapter->itr_setting = 1;
3840 adapter->eitr_param = 20000;
3842 /* set defaults for eitr in MegaBytes */
3843 adapter->eitr_low = 10;
3844 adapter->eitr_high = 20;
3846 /* set default ring sizes */
3847 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
3848 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
3850 /* initialize eeprom parameters */
3851 if (ixgbe_init_eeprom_params_generic(hw)) {
3852 dev_err(&pdev->dev, "EEPROM initialization failed\n");
3856 /* enable rx csum by default */
3857 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
3859 set_bit(__IXGBE_DOWN, &adapter->state);
3865 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
3866 * @adapter: board private structure
3867 * @tx_ring: tx descriptor ring (for a specific queue) to setup
3869 * Return 0 on success, negative on failure
3871 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
3872 struct ixgbe_ring *tx_ring)
3874 struct pci_dev *pdev = adapter->pdev;
3877 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3878 tx_ring->tx_buffer_info = vmalloc(size);
3879 if (!tx_ring->tx_buffer_info)
3881 memset(tx_ring->tx_buffer_info, 0, size);
3883 /* round up to nearest 4K */
3884 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3885 tx_ring->size = ALIGN(tx_ring->size, 4096);
3887 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
3892 tx_ring->next_to_use = 0;
3893 tx_ring->next_to_clean = 0;
3894 tx_ring->work_limit = tx_ring->count;
3898 vfree(tx_ring->tx_buffer_info);
3899 tx_ring->tx_buffer_info = NULL;
3900 DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
3901 "descriptor ring\n");
3906 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
3907 * @adapter: board private structure
3909 * If this function returns with an error, then it's possible one or
3910 * more of the rings is populated (while the rest are not). It is the
3911 * callers duty to clean those orphaned rings.
3913 * Return 0 on success, negative on failure
3915 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
3919 for (i = 0; i < adapter->num_tx_queues; i++) {
3920 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
3923 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
3931 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
3932 * @adapter: board private structure
3933 * @rx_ring: rx descriptor ring (for a specific queue) to setup
3935 * Returns 0 on success, negative on failure
3937 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
3938 struct ixgbe_ring *rx_ring)
3940 struct pci_dev *pdev = adapter->pdev;
3943 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3944 rx_ring->rx_buffer_info = vmalloc(size);
3945 if (!rx_ring->rx_buffer_info) {
3947 "vmalloc allocation failed for the rx desc ring\n");
3950 memset(rx_ring->rx_buffer_info, 0, size);
3952 /* Round up to nearest 4K */
3953 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3954 rx_ring->size = ALIGN(rx_ring->size, 4096);
3956 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
3958 if (!rx_ring->desc) {
3960 "Memory allocation failed for the rx desc ring\n");
3961 vfree(rx_ring->rx_buffer_info);
3965 rx_ring->next_to_clean = 0;
3966 rx_ring->next_to_use = 0;
3975 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
3976 * @adapter: board private structure
3978 * If this function returns with an error, then it's possible one or
3979 * more of the rings is populated (while the rest are not). It is the
3980 * callers duty to clean those orphaned rings.
3982 * Return 0 on success, negative on failure
3985 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
3989 for (i = 0; i < adapter->num_rx_queues; i++) {
3990 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
3993 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
4001 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4002 * @adapter: board private structure
4003 * @tx_ring: Tx descriptor ring for a specific queue
4005 * Free all transmit software resources
4007 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4008 struct ixgbe_ring *tx_ring)
4010 struct pci_dev *pdev = adapter->pdev;
4012 ixgbe_clean_tx_ring(adapter, tx_ring);
4014 vfree(tx_ring->tx_buffer_info);
4015 tx_ring->tx_buffer_info = NULL;
4017 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
4019 tx_ring->desc = NULL;
4023 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4024 * @adapter: board private structure
4026 * Free all transmit software resources
4028 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4032 for (i = 0; i < adapter->num_tx_queues; i++)
4033 if (adapter->tx_ring[i].desc)
4034 ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
4038 * ixgbe_free_rx_resources - Free Rx Resources
4039 * @adapter: board private structure
4040 * @rx_ring: ring to clean the resources from
4042 * Free all receive software resources
4044 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
4045 struct ixgbe_ring *rx_ring)
4047 struct pci_dev *pdev = adapter->pdev;
4049 ixgbe_clean_rx_ring(adapter, rx_ring);
4051 vfree(rx_ring->rx_buffer_info);
4052 rx_ring->rx_buffer_info = NULL;
4054 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
4056 rx_ring->desc = NULL;
4060 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4061 * @adapter: board private structure
4063 * Free all receive software resources
4065 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4069 for (i = 0; i < adapter->num_rx_queues; i++)
4070 if (adapter->rx_ring[i].desc)
4071 ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
4075 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4076 * @netdev: network interface device structure
4077 * @new_mtu: new value for maximum frame size
4079 * Returns 0 on success, negative on failure
4081 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4083 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4084 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4086 /* MTU < 68 is an error and causes problems on some kernels */
4087 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4090 DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
4091 netdev->mtu, new_mtu);
4092 /* must set new MTU before calling down or up */
4093 netdev->mtu = new_mtu;
4095 if (netif_running(netdev))
4096 ixgbe_reinit_locked(adapter);
4102 * ixgbe_open - Called when a network interface is made active
4103 * @netdev: network interface device structure
4105 * Returns 0 on success, negative value on failure
4107 * The open entry point is called when a network interface is made
4108 * active by the system (IFF_UP). At this point all resources needed
4109 * for transmit and receive operations are allocated, the interrupt
4110 * handler is registered with the OS, the watchdog timer is started,
4111 * and the stack is notified that the interface is ready.
4113 static int ixgbe_open(struct net_device *netdev)
4115 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4118 /* disallow open during test */
4119 if (test_bit(__IXGBE_TESTING, &adapter->state))
4122 netif_carrier_off(netdev);
4124 /* allocate transmit descriptors */
4125 err = ixgbe_setup_all_tx_resources(adapter);
4129 /* allocate receive descriptors */
4130 err = ixgbe_setup_all_rx_resources(adapter);
4134 ixgbe_configure(adapter);
4136 err = ixgbe_request_irq(adapter);
4140 err = ixgbe_up_complete(adapter);
4144 netif_tx_start_all_queues(netdev);
4149 ixgbe_release_hw_control(adapter);
4150 ixgbe_free_irq(adapter);
4153 ixgbe_free_all_rx_resources(adapter);
4155 ixgbe_free_all_tx_resources(adapter);
4156 ixgbe_reset(adapter);
4162 * ixgbe_close - Disables a network interface
4163 * @netdev: network interface device structure
4165 * Returns 0, this is not allowed to fail
4167 * The close entry point is called when an interface is de-activated
4168 * by the OS. The hardware is still under the drivers control, but
4169 * needs to be disabled. A global MAC reset is issued to stop the
4170 * hardware, and all transmit and receive resources are freed.
4172 static int ixgbe_close(struct net_device *netdev)
4174 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4176 ixgbe_down(adapter);
4177 ixgbe_free_irq(adapter);
4179 ixgbe_free_all_tx_resources(adapter);
4180 ixgbe_free_all_rx_resources(adapter);
4182 ixgbe_release_hw_control(adapter);
4188 static int ixgbe_resume(struct pci_dev *pdev)
4190 struct net_device *netdev = pci_get_drvdata(pdev);
4191 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4194 pci_set_power_state(pdev, PCI_D0);
4195 pci_restore_state(pdev);
4197 err = pci_enable_device_mem(pdev);
4199 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
4203 pci_set_master(pdev);
4205 pci_wake_from_d3(pdev, false);
4207 err = ixgbe_init_interrupt_scheme(adapter);
4209 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
4214 ixgbe_reset(adapter);
4216 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4218 if (netif_running(netdev)) {
4219 err = ixgbe_open(adapter->netdev);
4224 netif_device_attach(netdev);
4228 #endif /* CONFIG_PM */
4230 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4232 struct net_device *netdev = pci_get_drvdata(pdev);
4233 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4234 struct ixgbe_hw *hw = &adapter->hw;
4236 u32 wufc = adapter->wol;
4241 netif_device_detach(netdev);
4243 if (netif_running(netdev)) {
4244 ixgbe_down(adapter);
4245 ixgbe_free_irq(adapter);
4246 ixgbe_free_all_tx_resources(adapter);
4247 ixgbe_free_all_rx_resources(adapter);
4249 ixgbe_clear_interrupt_scheme(adapter);
4252 retval = pci_save_state(pdev);
4258 ixgbe_set_rx_mode(netdev);
4260 /* turn on all-multi mode if wake on multicast is enabled */
4261 if (wufc & IXGBE_WUFC_MC) {
4262 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4263 fctrl |= IXGBE_FCTRL_MPE;
4264 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4267 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4268 ctrl |= IXGBE_CTRL_GIO_DIS;
4269 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4271 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4273 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4274 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4277 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
4278 pci_wake_from_d3(pdev, true);
4280 pci_wake_from_d3(pdev, false);
4282 *enable_wake = !!wufc;
4284 ixgbe_release_hw_control(adapter);
4286 pci_disable_device(pdev);
4292 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4297 retval = __ixgbe_shutdown(pdev, &wake);
4302 pci_prepare_to_sleep(pdev);
4304 pci_wake_from_d3(pdev, false);
4305 pci_set_power_state(pdev, PCI_D3hot);
4310 #endif /* CONFIG_PM */
4312 static void ixgbe_shutdown(struct pci_dev *pdev)
4316 __ixgbe_shutdown(pdev, &wake);
4318 if (system_state == SYSTEM_POWER_OFF) {
4319 pci_wake_from_d3(pdev, wake);
4320 pci_set_power_state(pdev, PCI_D3hot);
4325 * ixgbe_update_stats - Update the board statistics counters.
4326 * @adapter: board private structure
4328 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4330 struct ixgbe_hw *hw = &adapter->hw;
4332 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4334 if (hw->mac.type == ixgbe_mac_82599EB) {
4336 for (i = 0; i < 16; i++)
4337 adapter->hw_rx_no_dma_resources +=
4338 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4339 for (i = 0; i < adapter->num_rx_queues; i++)
4340 rsc_count += adapter->rx_ring[i].rsc_count;
4341 adapter->rsc_count = rsc_count;
4344 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
4345 for (i = 0; i < 8; i++) {
4346 /* for packet buffers not used, the register should read 0 */
4347 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
4349 adapter->stats.mpc[i] += mpc;
4350 total_mpc += adapter->stats.mpc[i];
4351 if (hw->mac.type == ixgbe_mac_82598EB)
4352 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
4353 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
4354 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
4355 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
4356 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
4357 if (hw->mac.type == ixgbe_mac_82599EB) {
4358 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4359 IXGBE_PXONRXCNT(i));
4360 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4361 IXGBE_PXOFFRXCNT(i));
4362 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4364 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4366 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4369 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
4371 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
4374 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
4375 /* work around hardware counting issue */
4376 adapter->stats.gprc -= missed_rx;
4378 /* 82598 hardware only has a 32 bit counter in the high register */
4379 if (hw->mac.type == ixgbe_mac_82599EB) {
4380 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
4381 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
4382 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
4383 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
4384 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
4385 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
4386 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
4387 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
4388 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
4389 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
4391 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
4392 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
4393 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
4394 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
4395 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
4396 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
4397 #endif /* IXGBE_FCOE */
4399 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
4400 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
4401 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
4402 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
4403 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
4405 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
4406 adapter->stats.bprc += bprc;
4407 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
4408 if (hw->mac.type == ixgbe_mac_82598EB)
4409 adapter->stats.mprc -= bprc;
4410 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
4411 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
4412 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
4413 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
4414 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
4415 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
4416 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
4417 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
4418 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
4419 adapter->stats.lxontxc += lxon;
4420 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
4421 adapter->stats.lxofftxc += lxoff;
4422 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4423 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
4424 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
4426 * 82598 errata - tx of flow control packets is included in tx counters
4428 xon_off_tot = lxon + lxoff;
4429 adapter->stats.gptc -= xon_off_tot;
4430 adapter->stats.mptc -= xon_off_tot;
4431 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
4432 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4433 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
4434 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
4435 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
4436 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
4437 adapter->stats.ptc64 -= xon_off_tot;
4438 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
4439 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
4440 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
4441 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
4442 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
4443 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
4445 /* Fill out the OS statistics structure */
4446 adapter->net_stats.multicast = adapter->stats.mprc;
4449 adapter->net_stats.rx_errors = adapter->stats.crcerrs +
4450 adapter->stats.rlec;
4451 adapter->net_stats.rx_dropped = 0;
4452 adapter->net_stats.rx_length_errors = adapter->stats.rlec;
4453 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
4454 adapter->net_stats.rx_missed_errors = total_mpc;
4458 * ixgbe_watchdog - Timer Call-back
4459 * @data: pointer to adapter cast into an unsigned long
4461 static void ixgbe_watchdog(unsigned long data)
4463 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4464 struct ixgbe_hw *hw = &adapter->hw;
4469 * Do the watchdog outside of interrupt context due to the lovely
4470 * delays that some of the newer hardware requires
4473 if (test_bit(__IXGBE_DOWN, &adapter->state))
4474 goto watchdog_short_circuit;
4476 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
4478 * for legacy and MSI interrupts don't set any bits
4479 * that are enabled for EIAM, because this operation
4480 * would set *both* EIMS and EICS for any bit in EIAM
4482 IXGBE_WRITE_REG(hw, IXGBE_EICS,
4483 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
4484 goto watchdog_reschedule;
4487 /* get one bit for every active tx/rx interrupt vector */
4488 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
4489 struct ixgbe_q_vector *qv = adapter->q_vector[i];
4490 if (qv->rxr_count || qv->txr_count)
4491 eics |= ((u64)1 << i);
4494 /* Cause software interrupt to ensure rx rings are cleaned */
4495 ixgbe_irq_rearm_queues(adapter, eics);
4497 watchdog_reschedule:
4498 /* Reset the timer */
4499 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
4501 watchdog_short_circuit:
4502 schedule_work(&adapter->watchdog_task);
4506 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
4507 * @work: pointer to work_struct containing our data
4509 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
4511 struct ixgbe_adapter *adapter = container_of(work,
4512 struct ixgbe_adapter,
4513 multispeed_fiber_task);
4514 struct ixgbe_hw *hw = &adapter->hw;
4517 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
4518 autoneg = hw->phy.autoneg_advertised;
4519 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
4520 hw->mac.ops.get_link_capabilities(hw, &autoneg,
4522 if (hw->mac.ops.setup_link_speed)
4523 hw->mac.ops.setup_link_speed(hw, autoneg, true, true);
4524 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4525 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
4529 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
4530 * @work: pointer to work_struct containing our data
4532 static void ixgbe_sfp_config_module_task(struct work_struct *work)
4534 struct ixgbe_adapter *adapter = container_of(work,
4535 struct ixgbe_adapter,
4536 sfp_config_module_task);
4537 struct ixgbe_hw *hw = &adapter->hw;
4540 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
4542 /* Time for electrical oscillations to settle down */
4544 err = hw->phy.ops.identify_sfp(hw);
4546 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4547 dev_err(&adapter->pdev->dev, "failed to initialize because "
4548 "an unsupported SFP+ module type was detected.\n"
4549 "Reload the driver after installing a supported "
4551 unregister_netdev(adapter->netdev);
4554 hw->mac.ops.setup_sfp(hw);
4556 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
4557 /* This will also work for DA Twinax connections */
4558 schedule_work(&adapter->multispeed_fiber_task);
4559 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
4563 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
4564 * @work: pointer to work_struct containing our data
4566 static void ixgbe_fdir_reinit_task(struct work_struct *work)
4568 struct ixgbe_adapter *adapter = container_of(work,
4569 struct ixgbe_adapter,
4571 struct ixgbe_hw *hw = &adapter->hw;
4574 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
4575 for (i = 0; i < adapter->num_tx_queues; i++)
4576 set_bit(__IXGBE_FDIR_INIT_DONE,
4577 &(adapter->tx_ring[i].reinit_state));
4579 DPRINTK(PROBE, ERR, "failed to finish FDIR re-initialization, "
4580 "ignored adding FDIR ATR filters \n");
4582 /* Done FDIR Re-initialization, enable transmits */
4583 netif_tx_start_all_queues(adapter->netdev);
4587 * ixgbe_watchdog_task - worker thread to bring link up
4588 * @work: pointer to work_struct containing our data
4590 static void ixgbe_watchdog_task(struct work_struct *work)
4592 struct ixgbe_adapter *adapter = container_of(work,
4593 struct ixgbe_adapter,
4595 struct net_device *netdev = adapter->netdev;
4596 struct ixgbe_hw *hw = &adapter->hw;
4597 u32 link_speed = adapter->link_speed;
4598 bool link_up = adapter->link_up;
4600 struct ixgbe_ring *tx_ring;
4601 int some_tx_pending = 0;
4603 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
4605 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
4606 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
4609 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4610 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
4611 hw->mac.ops.fc_enable(hw, i);
4613 hw->mac.ops.fc_enable(hw, 0);
4616 hw->mac.ops.fc_enable(hw, 0);
4621 time_after(jiffies, (adapter->link_check_timeout +
4622 IXGBE_TRY_LINK_TIMEOUT))) {
4623 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4624 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
4626 adapter->link_up = link_up;
4627 adapter->link_speed = link_speed;
4631 if (!netif_carrier_ok(netdev)) {
4632 bool flow_rx, flow_tx;
4634 if (hw->mac.type == ixgbe_mac_82599EB) {
4635 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
4636 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
4637 flow_rx = (mflcn & IXGBE_MFLCN_RFCE);
4638 flow_tx = (fccfg & IXGBE_FCCFG_TFCE_802_3X);
4640 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4641 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
4642 flow_rx = (frctl & IXGBE_FCTRL_RFCE);
4643 flow_tx = (rmcs & IXGBE_RMCS_TFCE_802_3X);
4646 printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
4647 "Flow Control: %s\n",
4649 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
4651 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
4652 "1 Gbps" : "unknown speed")),
4653 ((flow_rx && flow_tx) ? "RX/TX" :
4655 (flow_tx ? "TX" : "None"))));
4657 netif_carrier_on(netdev);
4659 /* Force detection of hung controller */
4660 adapter->detect_tx_hung = true;
4663 adapter->link_up = false;
4664 adapter->link_speed = 0;
4665 if (netif_carrier_ok(netdev)) {
4666 printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
4668 netif_carrier_off(netdev);
4672 if (!netif_carrier_ok(netdev)) {
4673 for (i = 0; i < adapter->num_tx_queues; i++) {
4674 tx_ring = &adapter->tx_ring[i];
4675 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
4676 some_tx_pending = 1;
4681 if (some_tx_pending) {
4682 /* We've lost link, so the controller stops DMA,
4683 * but we've got queued Tx work that's never going
4684 * to get done, so reset controller to flush Tx.
4685 * (Do the reset outside of interrupt context).
4687 schedule_work(&adapter->reset_task);
4691 ixgbe_update_stats(adapter);
4692 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
4695 static int ixgbe_tso(struct ixgbe_adapter *adapter,
4696 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
4697 u32 tx_flags, u8 *hdr_len)
4699 struct ixgbe_adv_tx_context_desc *context_desc;
4702 struct ixgbe_tx_buffer *tx_buffer_info;
4703 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
4704 u32 mss_l4len_idx, l4len;
4706 if (skb_is_gso(skb)) {
4707 if (skb_header_cloned(skb)) {
4708 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4712 l4len = tcp_hdrlen(skb);
4715 if (skb->protocol == htons(ETH_P_IP)) {
4716 struct iphdr *iph = ip_hdr(skb);
4719 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4723 adapter->hw_tso_ctxt++;
4724 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
4725 ipv6_hdr(skb)->payload_len = 0;
4726 tcp_hdr(skb)->check =
4727 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4728 &ipv6_hdr(skb)->daddr,
4730 adapter->hw_tso6_ctxt++;
4733 i = tx_ring->next_to_use;
4735 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4736 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4738 /* VLAN MACLEN IPLEN */
4739 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4741 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4742 vlan_macip_lens |= ((skb_network_offset(skb)) <<
4743 IXGBE_ADVTXD_MACLEN_SHIFT);
4744 *hdr_len += skb_network_offset(skb);
4746 (skb_transport_header(skb) - skb_network_header(skb));
4748 (skb_transport_header(skb) - skb_network_header(skb));
4749 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4750 context_desc->seqnum_seed = 0;
4752 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4753 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
4754 IXGBE_ADVTXD_DTYP_CTXT);
4756 if (skb->protocol == htons(ETH_P_IP))
4757 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4758 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
4759 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4763 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
4764 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
4765 /* use index 1 for TSO */
4766 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4767 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4769 tx_buffer_info->time_stamp = jiffies;
4770 tx_buffer_info->next_to_watch = i;
4773 if (i == tx_ring->count)
4775 tx_ring->next_to_use = i;
4782 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
4783 struct ixgbe_ring *tx_ring,
4784 struct sk_buff *skb, u32 tx_flags)
4786 struct ixgbe_adv_tx_context_desc *context_desc;
4788 struct ixgbe_tx_buffer *tx_buffer_info;
4789 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
4791 if (skb->ip_summed == CHECKSUM_PARTIAL ||
4792 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
4793 i = tx_ring->next_to_use;
4794 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4795 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4797 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4799 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4800 vlan_macip_lens |= (skb_network_offset(skb) <<
4801 IXGBE_ADVTXD_MACLEN_SHIFT);
4802 if (skb->ip_summed == CHECKSUM_PARTIAL)
4803 vlan_macip_lens |= (skb_transport_header(skb) -
4804 skb_network_header(skb));
4806 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4807 context_desc->seqnum_seed = 0;
4809 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
4810 IXGBE_ADVTXD_DTYP_CTXT);
4812 if (skb->ip_summed == CHECKSUM_PARTIAL) {
4813 switch (skb->protocol) {
4814 case cpu_to_be16(ETH_P_IP):
4815 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4816 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4818 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4819 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
4821 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
4823 case cpu_to_be16(ETH_P_IPV6):
4824 /* XXX what about other V6 headers?? */
4825 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4827 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4828 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
4830 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
4833 if (unlikely(net_ratelimit())) {
4834 DPRINTK(PROBE, WARNING,
4835 "partial checksum but proto=%x!\n",
4842 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4843 /* use index zero for tx checksum offload */
4844 context_desc->mss_l4len_idx = 0;
4846 tx_buffer_info->time_stamp = jiffies;
4847 tx_buffer_info->next_to_watch = i;
4849 adapter->hw_csum_tx_good++;
4851 if (i == tx_ring->count)
4853 tx_ring->next_to_use = i;
4861 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
4862 struct ixgbe_ring *tx_ring,
4863 struct sk_buff *skb, u32 tx_flags,
4866 struct ixgbe_tx_buffer *tx_buffer_info;
4868 unsigned int total = skb->len;
4869 unsigned int offset = 0, size, count = 0, i;
4870 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
4874 i = tx_ring->next_to_use;
4876 if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
4877 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
4881 map = skb_shinfo(skb)->dma_maps;
4883 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
4884 /* excluding fcoe_crc_eof for FCoE */
4885 total -= sizeof(struct fcoe_crc_eof);
4887 len = min(skb_headlen(skb), total);
4889 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4890 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4892 tx_buffer_info->length = size;
4893 tx_buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
4894 tx_buffer_info->time_stamp = jiffies;
4895 tx_buffer_info->next_to_watch = i;
4904 if (i == tx_ring->count)
4909 for (f = 0; f < nr_frags; f++) {
4910 struct skb_frag_struct *frag;
4912 frag = &skb_shinfo(skb)->frags[f];
4913 len = min((unsigned int)frag->size, total);
4918 if (i == tx_ring->count)
4921 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4922 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4924 tx_buffer_info->length = size;
4925 tx_buffer_info->dma = map[f] + offset;
4926 tx_buffer_info->time_stamp = jiffies;
4927 tx_buffer_info->next_to_watch = i;
4938 tx_ring->tx_buffer_info[i].skb = skb;
4939 tx_ring->tx_buffer_info[first].next_to_watch = i;
4944 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
4945 struct ixgbe_ring *tx_ring,
4946 int tx_flags, int count, u32 paylen, u8 hdr_len)
4948 union ixgbe_adv_tx_desc *tx_desc = NULL;
4949 struct ixgbe_tx_buffer *tx_buffer_info;
4950 u32 olinfo_status = 0, cmd_type_len = 0;
4952 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
4954 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
4956 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
4958 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4959 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
4961 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
4962 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
4964 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
4965 IXGBE_ADVTXD_POPTS_SHIFT;
4967 /* use index 1 context for tso */
4968 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4969 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
4970 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
4971 IXGBE_ADVTXD_POPTS_SHIFT;
4973 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
4974 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
4975 IXGBE_ADVTXD_POPTS_SHIFT;
4977 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
4978 olinfo_status |= IXGBE_ADVTXD_CC;
4979 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4980 if (tx_flags & IXGBE_TX_FLAGS_FSO)
4981 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
4984 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
4986 i = tx_ring->next_to_use;
4988 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4989 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
4990 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
4991 tx_desc->read.cmd_type_len =
4992 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
4993 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4995 if (i == tx_ring->count)
4999 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
5002 * Force memory writes to complete before letting h/w
5003 * know there are new descriptors to fetch. (Only
5004 * applicable for weak-ordered memory model archs,
5009 tx_ring->next_to_use = i;
5010 writel(i, adapter->hw.hw_addr + tx_ring->tail);
5013 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5014 int queue, u32 tx_flags)
5016 /* Right now, we support IPv4 only */
5017 struct ixgbe_atr_input atr_input;
5020 struct iphdr *iph = ip_hdr(skb);
5021 struct ethhdr *eth = (struct ethhdr *)skb->data;
5022 u16 vlan_id, src_port, dst_port, flex_bytes;
5023 u32 src_ipv4_addr, dst_ipv4_addr;
5026 /* check if we're UDP or TCP */
5027 if (iph->protocol == IPPROTO_TCP) {
5029 src_port = th->source;
5030 dst_port = th->dest;
5031 l4type |= IXGBE_ATR_L4TYPE_TCP;
5032 /* l4type IPv4 type is 0, no need to assign */
5033 } else if(iph->protocol == IPPROTO_UDP) {
5035 src_port = uh->source;
5036 dst_port = uh->dest;
5037 l4type |= IXGBE_ATR_L4TYPE_UDP;
5038 /* l4type IPv4 type is 0, no need to assign */
5040 /* Unsupported L4 header, just bail here */
5044 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
5046 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
5047 IXGBE_TX_FLAGS_VLAN_SHIFT;
5048 src_ipv4_addr = iph->saddr;
5049 dst_ipv4_addr = iph->daddr;
5050 flex_bytes = eth->h_proto;
5052 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
5053 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
5054 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
5055 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
5056 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
5057 /* src and dst are inverted, think how the receiver sees them */
5058 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
5059 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
5061 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
5062 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
5065 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
5066 struct ixgbe_ring *tx_ring, int size)
5068 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5070 netif_stop_subqueue(netdev, tx_ring->queue_index);
5071 /* Herbert's original patch had:
5072 * smp_mb__after_netif_stop_queue();
5073 * but since that doesn't exist yet, just open code it. */
5076 /* We need to check again in a case another CPU has just
5077 * made room available. */
5078 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
5081 /* A reprieve! - use start_queue because it doesn't call schedule */
5082 netif_start_subqueue(netdev, tx_ring->queue_index);
5083 ++adapter->restart_queue;
5087 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
5088 struct ixgbe_ring *tx_ring, int size)
5090 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
5092 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
5095 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
5097 struct ixgbe_adapter *adapter = netdev_priv(dev);
5099 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
5100 return smp_processor_id();
5102 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5103 return 0; /* All traffic should default to class 0 */
5105 return skb_tx_hash(dev, skb);
5108 static int ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
5110 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5111 struct ixgbe_ring *tx_ring;
5113 unsigned int tx_flags = 0;
5119 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
5120 tx_flags |= vlan_tx_tag_get(skb);
5121 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5122 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
5123 tx_flags |= (skb->queue_mapping << 13);
5125 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5126 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5127 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5128 if (skb->priority != TC_PRIO_CONTROL) {
5129 tx_flags |= (skb->queue_mapping << 13);
5130 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5131 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5133 skb->queue_mapping =
5134 adapter->ring_feature[RING_F_DCB].indices-1;
5138 r_idx = skb->queue_mapping;
5139 tx_ring = &adapter->tx_ring[r_idx];
5141 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5142 (skb->protocol == htons(ETH_P_FCOE)))
5143 tx_flags |= IXGBE_TX_FLAGS_FCOE;
5145 /* four things can cause us to need a context descriptor */
5146 if (skb_is_gso(skb) ||
5147 (skb->ip_summed == CHECKSUM_PARTIAL) ||
5148 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
5149 (tx_flags & IXGBE_TX_FLAGS_FCOE))
5152 count += TXD_USE_COUNT(skb_headlen(skb));
5153 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5154 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5156 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
5158 return NETDEV_TX_BUSY;
5161 first = tx_ring->next_to_use;
5162 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5164 /* setup tx offload for FCoE */
5165 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5167 dev_kfree_skb_any(skb);
5168 return NETDEV_TX_OK;
5171 tx_flags |= IXGBE_TX_FLAGS_FSO;
5172 #endif /* IXGBE_FCOE */
5174 if (skb->protocol == htons(ETH_P_IP))
5175 tx_flags |= IXGBE_TX_FLAGS_IPV4;
5176 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5178 dev_kfree_skb_any(skb);
5179 return NETDEV_TX_OK;
5183 tx_flags |= IXGBE_TX_FLAGS_TSO;
5184 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
5185 (skb->ip_summed == CHECKSUM_PARTIAL))
5186 tx_flags |= IXGBE_TX_FLAGS_CSUM;
5189 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
5191 /* add the ATR filter if ATR is on */
5192 if (tx_ring->atr_sample_rate) {
5193 ++tx_ring->atr_count;
5194 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
5195 test_bit(__IXGBE_FDIR_INIT_DONE,
5196 &tx_ring->reinit_state)) {
5197 ixgbe_atr(adapter, skb, tx_ring->queue_index,
5199 tx_ring->atr_count = 0;
5202 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
5204 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
5207 dev_kfree_skb_any(skb);
5208 tx_ring->tx_buffer_info[first].time_stamp = 0;
5209 tx_ring->next_to_use = first;
5212 return NETDEV_TX_OK;
5216 * ixgbe_get_stats - Get System Network Statistics
5217 * @netdev: network interface device structure
5219 * Returns the address of the device statistics structure.
5220 * The statistics are actually updated from the timer callback.
5222 static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
5224 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5226 /* only return the current stats */
5227 return &adapter->net_stats;
5231 * ixgbe_set_mac - Change the Ethernet Address of the NIC
5232 * @netdev: network interface device structure
5233 * @p: pointer to an address structure
5235 * Returns 0 on success, negative on failure
5237 static int ixgbe_set_mac(struct net_device *netdev, void *p)
5239 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5240 struct ixgbe_hw *hw = &adapter->hw;
5241 struct sockaddr *addr = p;
5243 if (!is_valid_ether_addr(addr->sa_data))
5244 return -EADDRNOTAVAIL;
5246 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
5247 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5249 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
5255 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
5257 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5258 struct ixgbe_hw *hw = &adapter->hw;
5262 if (prtad != hw->phy.mdio.prtad)
5264 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
5270 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
5271 u16 addr, u16 value)
5273 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5274 struct ixgbe_hw *hw = &adapter->hw;
5276 if (prtad != hw->phy.mdio.prtad)
5278 return hw->phy.ops.write_reg(hw, addr, devad, value);
5281 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
5283 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5285 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
5289 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
5291 * @netdev: network interface device structure
5293 * Returns non-zero on failure
5295 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
5298 struct ixgbe_adapter *adapter = netdev_priv(dev);
5299 struct ixgbe_mac_info *mac = &adapter->hw.mac;
5301 if (is_valid_ether_addr(mac->san_addr)) {
5303 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5310 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
5312 * @netdev: network interface device structure
5314 * Returns non-zero on failure
5316 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
5319 struct ixgbe_adapter *adapter = netdev_priv(dev);
5320 struct ixgbe_mac_info *mac = &adapter->hw.mac;
5322 if (is_valid_ether_addr(mac->san_addr)) {
5324 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5330 #ifdef CONFIG_NET_POLL_CONTROLLER
5332 * Polling 'interrupt' - used by things like netconsole to send skbs
5333 * without having to re-enable interrupts. It's not called while
5334 * the interrupt routine is executing.
5336 static void ixgbe_netpoll(struct net_device *netdev)
5338 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5341 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
5342 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5343 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5344 for (i = 0; i < num_q_vectors; i++) {
5345 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
5346 ixgbe_msix_clean_many(0, q_vector);
5349 ixgbe_intr(adapter->pdev->irq, netdev);
5351 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
5355 static const struct net_device_ops ixgbe_netdev_ops = {
5356 .ndo_open = ixgbe_open,
5357 .ndo_stop = ixgbe_close,
5358 .ndo_start_xmit = ixgbe_xmit_frame,
5359 .ndo_select_queue = ixgbe_select_queue,
5360 .ndo_get_stats = ixgbe_get_stats,
5361 .ndo_set_rx_mode = ixgbe_set_rx_mode,
5362 .ndo_set_multicast_list = ixgbe_set_rx_mode,
5363 .ndo_validate_addr = eth_validate_addr,
5364 .ndo_set_mac_address = ixgbe_set_mac,
5365 .ndo_change_mtu = ixgbe_change_mtu,
5366 .ndo_tx_timeout = ixgbe_tx_timeout,
5367 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
5368 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
5369 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
5370 .ndo_do_ioctl = ixgbe_ioctl,
5371 #ifdef CONFIG_NET_POLL_CONTROLLER
5372 .ndo_poll_controller = ixgbe_netpoll,
5375 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
5376 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
5377 #endif /* IXGBE_FCOE */
5381 * ixgbe_probe - Device Initialization Routine
5382 * @pdev: PCI device information struct
5383 * @ent: entry in ixgbe_pci_tbl
5385 * Returns 0 on success, negative on failure
5387 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
5388 * The OS initialization, configuring of the adapter private structure,
5389 * and a hardware reset occur.
5391 static int __devinit ixgbe_probe(struct pci_dev *pdev,
5392 const struct pci_device_id *ent)
5394 struct net_device *netdev;
5395 struct ixgbe_adapter *adapter = NULL;
5396 struct ixgbe_hw *hw;
5397 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
5398 static int cards_found;
5399 int i, err, pci_using_dac;
5405 err = pci_enable_device_mem(pdev);
5409 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
5410 !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
5413 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5415 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
5417 dev_err(&pdev->dev, "No usable DMA "
5418 "configuration, aborting\n");
5425 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
5426 IORESOURCE_MEM), ixgbe_driver_name);
5429 "pci_request_selected_regions failed 0x%x\n", err);
5433 err = pci_enable_pcie_error_reporting(pdev);
5435 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
5437 /* non-fatal, continue */
5440 pci_set_master(pdev);
5441 pci_save_state(pdev);
5443 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
5446 goto err_alloc_etherdev;
5449 SET_NETDEV_DEV(netdev, &pdev->dev);
5451 pci_set_drvdata(pdev, netdev);
5452 adapter = netdev_priv(netdev);
5454 adapter->netdev = netdev;
5455 adapter->pdev = pdev;
5458 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
5460 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
5461 pci_resource_len(pdev, 0));
5467 for (i = 1; i <= 5; i++) {
5468 if (pci_resource_len(pdev, i) == 0)
5472 netdev->netdev_ops = &ixgbe_netdev_ops;
5473 ixgbe_set_ethtool_ops(netdev);
5474 netdev->watchdog_timeo = 5 * HZ;
5475 strcpy(netdev->name, pci_name(pdev));
5477 adapter->bd_number = cards_found;
5480 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
5481 hw->mac.type = ii->mac;
5484 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
5485 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
5486 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
5487 if (!(eec & (1 << 8)))
5488 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
5491 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
5492 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
5493 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
5494 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
5495 hw->phy.mdio.mmds = 0;
5496 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
5497 hw->phy.mdio.dev = netdev;
5498 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
5499 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
5501 /* set up this timer and work struct before calling get_invariants
5502 * which might start the timer
5504 init_timer(&adapter->sfp_timer);
5505 adapter->sfp_timer.function = &ixgbe_sfp_timer;
5506 adapter->sfp_timer.data = (unsigned long) adapter;
5508 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
5510 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
5511 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
5513 /* a new SFP+ module arrival, called from GPI SDP2 context */
5514 INIT_WORK(&adapter->sfp_config_module_task,
5515 ixgbe_sfp_config_module_task);
5517 ii->get_invariants(hw);
5519 /* setup the private structure */
5520 err = ixgbe_sw_init(adapter);
5525 * If there is a fan on this device and it has failed log the
5528 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5529 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5530 if (esdp & IXGBE_ESDP_SDP1)
5531 DPRINTK(PROBE, CRIT,
5532 "Fan has stopped, replace the adapter\n");
5535 /* reset_hw fills in the perm_addr as well */
5536 err = hw->mac.ops.reset_hw(hw);
5537 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
5538 hw->mac.type == ixgbe_mac_82598EB) {
5540 * Start a kernel thread to watch for a module to arrive.
5541 * Only do this for 82598, since 82599 will generate
5542 * interrupts on module arrival.
5544 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5545 mod_timer(&adapter->sfp_timer,
5546 round_jiffies(jiffies + (2 * HZ)));
5548 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5549 dev_err(&adapter->pdev->dev, "failed to initialize because "
5550 "an unsupported SFP+ module type was detected.\n"
5551 "Reload the driver after installing a supported "
5555 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
5559 netdev->features = NETIF_F_SG |
5561 NETIF_F_HW_VLAN_TX |
5562 NETIF_F_HW_VLAN_RX |
5563 NETIF_F_HW_VLAN_FILTER;
5565 netdev->features |= NETIF_F_IPV6_CSUM;
5566 netdev->features |= NETIF_F_TSO;
5567 netdev->features |= NETIF_F_TSO6;
5568 netdev->features |= NETIF_F_GRO;
5570 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
5571 netdev->features |= NETIF_F_SCTP_CSUM;
5573 netdev->vlan_features |= NETIF_F_TSO;
5574 netdev->vlan_features |= NETIF_F_TSO6;
5575 netdev->vlan_features |= NETIF_F_IP_CSUM;
5576 netdev->vlan_features |= NETIF_F_SG;
5578 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5579 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
5581 #ifdef CONFIG_IXGBE_DCB
5582 netdev->dcbnl_ops = &dcbnl_ops;
5586 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
5587 if (hw->mac.ops.get_device_caps) {
5588 hw->mac.ops.get_device_caps(hw, &device_caps);
5589 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
5590 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5593 #endif /* IXGBE_FCOE */
5595 netdev->features |= NETIF_F_HIGHDMA;
5597 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
5598 netdev->features |= NETIF_F_LRO;
5600 /* make sure the EEPROM is good */
5601 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
5602 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
5607 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
5608 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
5610 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
5611 dev_err(&pdev->dev, "invalid MAC address\n");
5616 init_timer(&adapter->watchdog_timer);
5617 adapter->watchdog_timer.function = &ixgbe_watchdog;
5618 adapter->watchdog_timer.data = (unsigned long)adapter;
5620 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
5621 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
5623 err = ixgbe_init_interrupt_scheme(adapter);
5627 switch (pdev->device) {
5628 case IXGBE_DEV_ID_82599_KX4:
5629 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
5630 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
5631 /* Enable ACPI wakeup in GRC */
5632 IXGBE_WRITE_REG(hw, IXGBE_GRC,
5633 (IXGBE_READ_REG(hw, IXGBE_GRC) & ~IXGBE_GRC_APME));
5639 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5641 /* pick up the PCI bus settings for reporting later */
5642 hw->mac.ops.get_bus_info(hw);
5644 /* print bus type/speed/width info */
5645 dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
5646 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
5647 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
5648 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
5649 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
5650 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
5653 ixgbe_read_pba_num_generic(hw, &part_num);
5654 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
5655 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
5656 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
5657 (part_num >> 8), (part_num & 0xff));
5659 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5660 hw->mac.type, hw->phy.type,
5661 (part_num >> 8), (part_num & 0xff));
5663 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
5664 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
5665 "this card is not sufficient for optimal "
5667 dev_warn(&pdev->dev, "For optimal performance a x8 "
5668 "PCI-Express slot is required.\n");
5671 /* save off EEPROM version number */
5672 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
5674 /* reset the hardware with the new settings */
5675 err = hw->mac.ops.start_hw(hw);
5677 if (err == IXGBE_ERR_EEPROM_VERSION) {
5678 /* We are running on a pre-production device, log a warning */
5679 dev_warn(&pdev->dev, "This device is a pre-production "
5680 "adapter/LOM. Please be aware there may be issues "
5681 "associated with your hardware. If you are "
5682 "experiencing problems please contact your Intel or "
5683 "hardware representative who provided you with this "
5686 strcpy(netdev->name, "eth%d");
5687 err = register_netdev(netdev);
5691 /* carrier off reporting is important to ethtool even BEFORE open */
5692 netif_carrier_off(netdev);
5694 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5695 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5696 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
5698 #ifdef CONFIG_IXGBE_DCA
5699 if (dca_add_requester(&pdev->dev) == 0) {
5700 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
5701 ixgbe_setup_dca(adapter);
5704 /* add san mac addr to netdev */
5705 ixgbe_add_sanmac_netdev(netdev);
5707 dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
5712 ixgbe_release_hw_control(adapter);
5713 ixgbe_clear_interrupt_scheme(adapter);
5716 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5717 del_timer_sync(&adapter->sfp_timer);
5718 cancel_work_sync(&adapter->sfp_task);
5719 cancel_work_sync(&adapter->multispeed_fiber_task);
5720 cancel_work_sync(&adapter->sfp_config_module_task);
5721 iounmap(hw->hw_addr);
5723 free_netdev(netdev);
5725 pci_release_selected_regions(pdev, pci_select_bars(pdev,
5729 pci_disable_device(pdev);
5734 * ixgbe_remove - Device Removal Routine
5735 * @pdev: PCI device information struct
5737 * ixgbe_remove is called by the PCI subsystem to alert the driver
5738 * that it should release a PCI device. The could be caused by a
5739 * Hot-Plug event, or because the driver is going to be removed from
5742 static void __devexit ixgbe_remove(struct pci_dev *pdev)
5744 struct net_device *netdev = pci_get_drvdata(pdev);
5745 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5748 set_bit(__IXGBE_DOWN, &adapter->state);
5749 /* clear the module not found bit to make sure the worker won't
5752 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5753 del_timer_sync(&adapter->watchdog_timer);
5755 del_timer_sync(&adapter->sfp_timer);
5756 cancel_work_sync(&adapter->watchdog_task);
5757 cancel_work_sync(&adapter->sfp_task);
5758 cancel_work_sync(&adapter->multispeed_fiber_task);
5759 cancel_work_sync(&adapter->sfp_config_module_task);
5760 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5761 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5762 cancel_work_sync(&adapter->fdir_reinit_task);
5763 flush_scheduled_work();
5765 #ifdef CONFIG_IXGBE_DCA
5766 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
5767 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
5768 dca_remove_requester(&pdev->dev);
5769 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
5774 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
5775 ixgbe_cleanup_fcoe(adapter);
5777 #endif /* IXGBE_FCOE */
5779 /* remove the added san mac */
5780 ixgbe_del_sanmac_netdev(netdev);
5782 if (netdev->reg_state == NETREG_REGISTERED)
5783 unregister_netdev(netdev);
5785 ixgbe_clear_interrupt_scheme(adapter);
5787 ixgbe_release_hw_control(adapter);
5789 iounmap(adapter->hw.hw_addr);
5790 pci_release_selected_regions(pdev, pci_select_bars(pdev,
5793 DPRINTK(PROBE, INFO, "complete\n");
5795 free_netdev(netdev);
5797 err = pci_disable_pcie_error_reporting(pdev);
5800 "pci_disable_pcie_error_reporting failed 0x%x\n", err);
5802 pci_disable_device(pdev);
5806 * ixgbe_io_error_detected - called when PCI error is detected
5807 * @pdev: Pointer to PCI device
5808 * @state: The current pci connection state
5810 * This function is called after a PCI bus error affecting
5811 * this device has been detected.
5813 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
5814 pci_channel_state_t state)
5816 struct net_device *netdev = pci_get_drvdata(pdev);
5817 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5819 netif_device_detach(netdev);
5821 if (state == pci_channel_io_perm_failure)
5822 return PCI_ERS_RESULT_DISCONNECT;
5824 if (netif_running(netdev))
5825 ixgbe_down(adapter);
5826 pci_disable_device(pdev);
5828 /* Request a slot reset. */
5829 return PCI_ERS_RESULT_NEED_RESET;
5833 * ixgbe_io_slot_reset - called after the pci bus has been reset.
5834 * @pdev: Pointer to PCI device
5836 * Restart the card from scratch, as if from a cold-boot.
5838 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
5840 struct net_device *netdev = pci_get_drvdata(pdev);
5841 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5842 pci_ers_result_t result;
5845 if (pci_enable_device_mem(pdev)) {
5847 "Cannot re-enable PCI device after reset.\n");
5848 result = PCI_ERS_RESULT_DISCONNECT;
5850 pci_set_master(pdev);
5851 pci_restore_state(pdev);
5853 pci_wake_from_d3(pdev, false);
5855 ixgbe_reset(adapter);
5856 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5857 result = PCI_ERS_RESULT_RECOVERED;
5860 err = pci_cleanup_aer_uncorrect_error_status(pdev);
5863 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
5864 /* non-fatal, continue */
5871 * ixgbe_io_resume - called when traffic can start flowing again.
5872 * @pdev: Pointer to PCI device
5874 * This callback is called when the error recovery driver tells us that
5875 * its OK to resume normal operation.
5877 static void ixgbe_io_resume(struct pci_dev *pdev)
5879 struct net_device *netdev = pci_get_drvdata(pdev);
5880 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5882 if (netif_running(netdev)) {
5883 if (ixgbe_up(adapter)) {
5884 DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
5889 netif_device_attach(netdev);
5892 static struct pci_error_handlers ixgbe_err_handler = {
5893 .error_detected = ixgbe_io_error_detected,
5894 .slot_reset = ixgbe_io_slot_reset,
5895 .resume = ixgbe_io_resume,
5898 static struct pci_driver ixgbe_driver = {
5899 .name = ixgbe_driver_name,
5900 .id_table = ixgbe_pci_tbl,
5901 .probe = ixgbe_probe,
5902 .remove = __devexit_p(ixgbe_remove),
5904 .suspend = ixgbe_suspend,
5905 .resume = ixgbe_resume,
5907 .shutdown = ixgbe_shutdown,
5908 .err_handler = &ixgbe_err_handler
5912 * ixgbe_init_module - Driver Registration Routine
5914 * ixgbe_init_module is the first routine called when the driver is
5915 * loaded. All it does is register with the PCI subsystem.
5917 static int __init ixgbe_init_module(void)
5920 printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
5921 ixgbe_driver_string, ixgbe_driver_version);
5923 printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
5925 #ifdef CONFIG_IXGBE_DCA
5926 dca_register_notify(&dca_notifier);
5929 ret = pci_register_driver(&ixgbe_driver);
5933 module_init(ixgbe_init_module);
5936 * ixgbe_exit_module - Driver Exit Cleanup Routine
5938 * ixgbe_exit_module is called just before the driver is removed
5941 static void __exit ixgbe_exit_module(void)
5943 #ifdef CONFIG_IXGBE_DCA
5944 dca_unregister_notify(&dca_notifier);
5946 pci_unregister_driver(&ixgbe_driver);
5949 #ifdef CONFIG_IXGBE_DCA
5950 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
5955 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
5956 __ixgbe_notify_dca);
5958 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5961 #endif /* CONFIG_IXGBE_DCA */
5964 * ixgbe_get_hw_dev_name - return device name string
5965 * used by hardware layer to print debugging information
5967 char *ixgbe_get_hw_dev_name(struct ixgbe_hw *hw)
5969 struct ixgbe_adapter *adapter = hw->back;
5970 return adapter->netdev->name;
5974 module_exit(ixgbe_exit_module);