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 "2.0.44-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_82598AT2),
80 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
82 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
84 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
107 /* required last entry */
110 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
112 #ifdef CONFIG_IXGBE_DCA
113 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
115 static struct notifier_block dca_notifier = {
116 .notifier_call = ixgbe_notify_dca,
122 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
123 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
124 MODULE_LICENSE("GPL");
125 MODULE_VERSION(DRV_VERSION);
127 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
129 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
133 /* Let firmware take over control of h/w */
134 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
135 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
136 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
139 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
143 /* Let firmware know the driver has taken over */
144 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
145 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
146 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
150 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
151 * @adapter: pointer to adapter struct
152 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
153 * @queue: queue to map the corresponding interrupt to
154 * @msix_vector: the vector to map to the corresponding queue
157 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
158 u8 queue, u8 msix_vector)
161 struct ixgbe_hw *hw = &adapter->hw;
162 switch (hw->mac.type) {
163 case ixgbe_mac_82598EB:
164 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
167 index = (((direction * 64) + queue) >> 2) & 0x1F;
168 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
169 ivar &= ~(0xFF << (8 * (queue & 0x3)));
170 ivar |= (msix_vector << (8 * (queue & 0x3)));
171 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
173 case ixgbe_mac_82599EB:
174 if (direction == -1) {
176 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
177 index = ((queue & 1) * 8);
178 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
179 ivar &= ~(0xFF << index);
180 ivar |= (msix_vector << index);
181 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
184 /* tx or rx causes */
185 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
186 index = ((16 * (queue & 1)) + (8 * direction));
187 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
188 ivar &= ~(0xFF << index);
189 ivar |= (msix_vector << index);
190 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
198 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
203 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
204 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
205 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
207 mask = (qmask & 0xFFFFFFFF);
208 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
209 mask = (qmask >> 32);
210 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
214 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
215 struct ixgbe_tx_buffer
218 tx_buffer_info->dma = 0;
219 if (tx_buffer_info->skb) {
220 skb_dma_unmap(&adapter->pdev->dev, tx_buffer_info->skb,
222 dev_kfree_skb_any(tx_buffer_info->skb);
223 tx_buffer_info->skb = NULL;
225 tx_buffer_info->time_stamp = 0;
226 /* tx_buffer_info must be completely set up in the transmit path */
230 * ixgbe_tx_is_paused - check if the tx ring is paused
231 * @adapter: the ixgbe adapter
232 * @tx_ring: the corresponding tx_ring
234 * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
235 * corresponding TC of this tx_ring when checking TFCS.
237 * Returns : true if paused
239 static inline bool ixgbe_tx_is_paused(struct ixgbe_adapter *adapter,
240 struct ixgbe_ring *tx_ring)
243 u32 txoff = IXGBE_TFCS_TXOFF;
245 #ifdef CONFIG_IXGBE_DCB
246 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
247 int reg_idx = tx_ring->reg_idx;
248 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
250 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
252 txoff = IXGBE_TFCS_TXOFF0;
253 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
255 txoff = IXGBE_TFCS_TXOFF;
259 if (tc == 2) /* TC2, TC3 */
260 tc += (reg_idx - 64) >> 4;
261 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
262 tc += 1 + ((reg_idx - 96) >> 3);
263 } else if (dcb_i == 4) {
267 tc += (reg_idx - 64) >> 5;
268 if (tc == 2) /* TC2, TC3 */
269 tc += (reg_idx - 96) >> 4;
276 return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
279 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
280 struct ixgbe_ring *tx_ring,
283 struct ixgbe_hw *hw = &adapter->hw;
285 /* Detect a transmit hang in hardware, this serializes the
286 * check with the clearing of time_stamp and movement of eop */
287 adapter->detect_tx_hung = false;
288 if (tx_ring->tx_buffer_info[eop].time_stamp &&
289 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
290 !ixgbe_tx_is_paused(adapter, tx_ring)) {
291 /* detected Tx unit hang */
292 union ixgbe_adv_tx_desc *tx_desc;
293 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
294 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
296 " TDH, TDT <%x>, <%x>\n"
297 " next_to_use <%x>\n"
298 " next_to_clean <%x>\n"
299 "tx_buffer_info[next_to_clean]\n"
300 " time_stamp <%lx>\n"
302 tx_ring->queue_index,
303 IXGBE_READ_REG(hw, tx_ring->head),
304 IXGBE_READ_REG(hw, tx_ring->tail),
305 tx_ring->next_to_use, eop,
306 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
313 #define IXGBE_MAX_TXD_PWR 14
314 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
316 /* Tx Descriptors needed, worst case */
317 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
318 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
319 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
320 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
322 static void ixgbe_tx_timeout(struct net_device *netdev);
325 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
326 * @q_vector: structure containing interrupt and ring information
327 * @tx_ring: tx ring to clean
329 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
330 struct ixgbe_ring *tx_ring)
332 struct ixgbe_adapter *adapter = q_vector->adapter;
333 struct net_device *netdev = adapter->netdev;
334 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
335 struct ixgbe_tx_buffer *tx_buffer_info;
336 unsigned int i, eop, count = 0;
337 unsigned int total_bytes = 0, total_packets = 0;
339 i = tx_ring->next_to_clean;
340 eop = tx_ring->tx_buffer_info[i].next_to_watch;
341 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
343 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
344 (count < tx_ring->work_limit)) {
345 bool cleaned = false;
346 for ( ; !cleaned; count++) {
348 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
349 tx_buffer_info = &tx_ring->tx_buffer_info[i];
350 cleaned = (i == eop);
351 skb = tx_buffer_info->skb;
353 if (cleaned && skb) {
354 unsigned int segs, bytecount;
355 unsigned int hlen = skb_headlen(skb);
357 /* gso_segs is currently only valid for tcp */
358 segs = skb_shinfo(skb)->gso_segs ?: 1;
360 /* adjust for FCoE Sequence Offload */
361 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
362 && (skb->protocol == htons(ETH_P_FCOE)) &&
364 hlen = skb_transport_offset(skb) +
365 sizeof(struct fc_frame_header) +
366 sizeof(struct fcoe_crc_eof);
367 segs = DIV_ROUND_UP(skb->len - hlen,
368 skb_shinfo(skb)->gso_size);
370 #endif /* IXGBE_FCOE */
371 /* multiply data chunks by size of headers */
372 bytecount = ((segs - 1) * hlen) + skb->len;
373 total_packets += segs;
374 total_bytes += bytecount;
377 ixgbe_unmap_and_free_tx_resource(adapter,
380 tx_desc->wb.status = 0;
383 if (i == tx_ring->count)
387 eop = tx_ring->tx_buffer_info[i].next_to_watch;
388 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
391 tx_ring->next_to_clean = i;
393 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
394 if (unlikely(count && netif_carrier_ok(netdev) &&
395 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
396 /* Make sure that anybody stopping the queue after this
397 * sees the new next_to_clean.
400 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
401 !test_bit(__IXGBE_DOWN, &adapter->state)) {
402 netif_wake_subqueue(netdev, tx_ring->queue_index);
403 ++adapter->restart_queue;
407 if (adapter->detect_tx_hung) {
408 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
409 /* schedule immediate reset if we believe we hung */
411 "tx hang %d detected, resetting adapter\n",
412 adapter->tx_timeout_count + 1);
413 ixgbe_tx_timeout(adapter->netdev);
417 /* re-arm the interrupt */
418 if (count >= tx_ring->work_limit)
419 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
421 tx_ring->total_bytes += total_bytes;
422 tx_ring->total_packets += total_packets;
423 tx_ring->stats.packets += total_packets;
424 tx_ring->stats.bytes += total_bytes;
425 adapter->net_stats.tx_bytes += total_bytes;
426 adapter->net_stats.tx_packets += total_packets;
427 return (count < tx_ring->work_limit);
430 #ifdef CONFIG_IXGBE_DCA
431 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
432 struct ixgbe_ring *rx_ring)
436 int q = rx_ring - adapter->rx_ring;
438 if (rx_ring->cpu != cpu) {
439 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
440 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
441 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
442 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
443 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
444 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
445 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
446 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
448 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
449 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
450 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
451 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
452 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
453 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
459 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
460 struct ixgbe_ring *tx_ring)
464 int q = tx_ring - adapter->tx_ring;
466 if (tx_ring->cpu != cpu) {
467 txctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q));
468 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
469 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
470 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
471 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
472 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
473 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
474 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
476 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
477 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q), txctrl);
483 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
487 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
490 /* always use CB2 mode, difference is masked in the CB driver */
491 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
493 for (i = 0; i < adapter->num_tx_queues; i++) {
494 adapter->tx_ring[i].cpu = -1;
495 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
497 for (i = 0; i < adapter->num_rx_queues; i++) {
498 adapter->rx_ring[i].cpu = -1;
499 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
503 static int __ixgbe_notify_dca(struct device *dev, void *data)
505 struct net_device *netdev = dev_get_drvdata(dev);
506 struct ixgbe_adapter *adapter = netdev_priv(netdev);
507 unsigned long event = *(unsigned long *)data;
510 case DCA_PROVIDER_ADD:
511 /* if we're already enabled, don't do it again */
512 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
514 if (dca_add_requester(dev) == 0) {
515 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
516 ixgbe_setup_dca(adapter);
519 /* Fall Through since DCA is disabled. */
520 case DCA_PROVIDER_REMOVE:
521 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
522 dca_remove_requester(dev);
523 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
524 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
532 #endif /* CONFIG_IXGBE_DCA */
534 * ixgbe_receive_skb - Send a completed packet up the stack
535 * @adapter: board private structure
536 * @skb: packet to send up
537 * @status: hardware indication of status of receive
538 * @rx_ring: rx descriptor ring (for a specific queue) to setup
539 * @rx_desc: rx descriptor
541 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
542 struct sk_buff *skb, u8 status,
543 struct ixgbe_ring *ring,
544 union ixgbe_adv_rx_desc *rx_desc)
546 struct ixgbe_adapter *adapter = q_vector->adapter;
547 struct napi_struct *napi = &q_vector->napi;
548 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
549 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
551 skb_record_rx_queue(skb, ring->queue_index);
552 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
553 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
554 vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
556 napi_gro_receive(napi, skb);
558 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
559 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
566 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
567 * @adapter: address of board private structure
568 * @status_err: hardware indication of status of receive
569 * @skb: skb currently being received and modified
571 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
572 union ixgbe_adv_rx_desc *rx_desc,
575 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
577 skb->ip_summed = CHECKSUM_NONE;
579 /* Rx csum disabled */
580 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
583 /* if IP and error */
584 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
585 (status_err & IXGBE_RXDADV_ERR_IPE)) {
586 adapter->hw_csum_rx_error++;
590 if (!(status_err & IXGBE_RXD_STAT_L4CS))
593 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
594 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
597 * 82599 errata, UDP frames with a 0 checksum can be marked as
600 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
601 (adapter->hw.mac.type == ixgbe_mac_82599EB))
604 adapter->hw_csum_rx_error++;
608 /* It must be a TCP or UDP packet with a valid checksum */
609 skb->ip_summed = CHECKSUM_UNNECESSARY;
610 adapter->hw_csum_rx_good++;
613 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
614 struct ixgbe_ring *rx_ring, u32 val)
617 * Force memory writes to complete before letting h/w
618 * know there are new descriptors to fetch. (Only
619 * applicable for weak-ordered memory model archs,
623 IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
627 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
628 * @adapter: address of board private structure
630 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
631 struct ixgbe_ring *rx_ring,
634 struct pci_dev *pdev = adapter->pdev;
635 union ixgbe_adv_rx_desc *rx_desc;
636 struct ixgbe_rx_buffer *bi;
639 i = rx_ring->next_to_use;
640 bi = &rx_ring->rx_buffer_info[i];
642 while (cleaned_count--) {
643 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
646 (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
648 bi->page = alloc_page(GFP_ATOMIC);
650 adapter->alloc_rx_page_failed++;
655 /* use a half page if we're re-using */
656 bi->page_offset ^= (PAGE_SIZE / 2);
659 bi->page_dma = pci_map_page(pdev, bi->page,
667 skb = netdev_alloc_skb(adapter->netdev,
668 (rx_ring->rx_buf_len +
672 adapter->alloc_rx_buff_failed++;
677 * Make buffer alignment 2 beyond a 16 byte boundary
678 * this will result in a 16 byte aligned IP header after
679 * the 14 byte MAC header is removed
681 skb_reserve(skb, NET_IP_ALIGN);
684 bi->dma = pci_map_single(pdev, skb->data,
688 /* Refresh the desc even if buffer_addrs didn't change because
689 * each write-back erases this info. */
690 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
691 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
692 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
694 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
698 if (i == rx_ring->count)
700 bi = &rx_ring->rx_buffer_info[i];
704 if (rx_ring->next_to_use != i) {
705 rx_ring->next_to_use = i;
707 i = (rx_ring->count - 1);
709 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
713 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
715 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
718 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
720 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
723 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
725 return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
726 IXGBE_RXDADV_RSCCNT_MASK) >>
727 IXGBE_RXDADV_RSCCNT_SHIFT;
731 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
732 * @skb: pointer to the last skb in the rsc queue
734 * This function changes a queue full of hw rsc buffers into a completed
735 * packet. It uses the ->prev pointers to find the first packet and then
736 * turns it into the frag list owner.
738 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
740 unsigned int frag_list_size = 0;
743 struct sk_buff *prev = skb->prev;
744 frag_list_size += skb->len;
749 skb_shinfo(skb)->frag_list = skb->next;
751 skb->len += frag_list_size;
752 skb->data_len += frag_list_size;
753 skb->truesize += frag_list_size;
757 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
758 struct ixgbe_ring *rx_ring,
759 int *work_done, int work_to_do)
761 struct ixgbe_adapter *adapter = q_vector->adapter;
762 struct pci_dev *pdev = adapter->pdev;
763 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
764 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
766 unsigned int i, rsc_count = 0;
769 bool cleaned = false;
770 int cleaned_count = 0;
771 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
774 #endif /* IXGBE_FCOE */
776 i = rx_ring->next_to_clean;
777 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
778 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
779 rx_buffer_info = &rx_ring->rx_buffer_info[i];
781 while (staterr & IXGBE_RXD_STAT_DD) {
783 if (*work_done >= work_to_do)
787 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
788 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
789 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
790 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
791 if (hdr_info & IXGBE_RXDADV_SPH)
792 adapter->rx_hdr_split++;
793 if (len > IXGBE_RX_HDR_SIZE)
794 len = IXGBE_RX_HDR_SIZE;
795 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
797 len = le16_to_cpu(rx_desc->wb.upper.length);
801 skb = rx_buffer_info->skb;
802 prefetch(skb->data - NET_IP_ALIGN);
803 rx_buffer_info->skb = NULL;
805 if (rx_buffer_info->dma) {
806 pci_unmap_single(pdev, rx_buffer_info->dma,
809 rx_buffer_info->dma = 0;
814 pci_unmap_page(pdev, rx_buffer_info->page_dma,
815 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
816 rx_buffer_info->page_dma = 0;
817 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
818 rx_buffer_info->page,
819 rx_buffer_info->page_offset,
822 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
823 (page_count(rx_buffer_info->page) != 1))
824 rx_buffer_info->page = NULL;
826 get_page(rx_buffer_info->page);
828 skb->len += upper_len;
829 skb->data_len += upper_len;
830 skb->truesize += upper_len;
834 if (i == rx_ring->count)
837 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
841 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
842 rsc_count = ixgbe_get_rsc_count(rx_desc);
845 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
846 IXGBE_RXDADV_NEXTP_SHIFT;
847 next_buffer = &rx_ring->rx_buffer_info[nextp];
848 rx_ring->rsc_count += (rsc_count - 1);
850 next_buffer = &rx_ring->rx_buffer_info[i];
853 if (staterr & IXGBE_RXD_STAT_EOP) {
855 skb = ixgbe_transform_rsc_queue(skb);
856 rx_ring->stats.packets++;
857 rx_ring->stats.bytes += skb->len;
859 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
860 rx_buffer_info->skb = next_buffer->skb;
861 rx_buffer_info->dma = next_buffer->dma;
862 next_buffer->skb = skb;
863 next_buffer->dma = 0;
865 skb->next = next_buffer->skb;
866 skb->next->prev = skb;
868 adapter->non_eop_descs++;
872 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
873 dev_kfree_skb_irq(skb);
877 ixgbe_rx_checksum(adapter, rx_desc, skb);
879 /* probably a little skewed due to removing CRC */
880 total_rx_bytes += skb->len;
883 skb->protocol = eth_type_trans(skb, adapter->netdev);
885 /* if ddp, not passing to ULD unless for FCP_RSP or error */
886 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
887 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
891 #endif /* IXGBE_FCOE */
892 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
895 rx_desc->wb.upper.status_error = 0;
897 /* return some buffers to hardware, one at a time is too slow */
898 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
899 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
903 /* use prefetched values */
905 rx_buffer_info = &rx_ring->rx_buffer_info[i];
907 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
910 rx_ring->next_to_clean = i;
911 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
914 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
917 /* include DDPed FCoE data */
921 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
922 sizeof(struct fc_frame_header) -
923 sizeof(struct fcoe_crc_eof);
926 total_rx_bytes += ddp_bytes;
927 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
929 #endif /* IXGBE_FCOE */
931 rx_ring->total_packets += total_rx_packets;
932 rx_ring->total_bytes += total_rx_bytes;
933 adapter->net_stats.rx_bytes += total_rx_bytes;
934 adapter->net_stats.rx_packets += total_rx_packets;
939 static int ixgbe_clean_rxonly(struct napi_struct *, int);
941 * ixgbe_configure_msix - Configure MSI-X hardware
942 * @adapter: board private structure
944 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
947 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
949 struct ixgbe_q_vector *q_vector;
950 int i, j, q_vectors, v_idx, r_idx;
953 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
956 * Populate the IVAR table and set the ITR values to the
957 * corresponding register.
959 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
960 q_vector = adapter->q_vector[v_idx];
961 /* XXX for_each_bit(...) */
962 r_idx = find_first_bit(q_vector->rxr_idx,
963 adapter->num_rx_queues);
965 for (i = 0; i < q_vector->rxr_count; i++) {
966 j = adapter->rx_ring[r_idx].reg_idx;
967 ixgbe_set_ivar(adapter, 0, j, v_idx);
968 r_idx = find_next_bit(q_vector->rxr_idx,
969 adapter->num_rx_queues,
972 r_idx = find_first_bit(q_vector->txr_idx,
973 adapter->num_tx_queues);
975 for (i = 0; i < q_vector->txr_count; i++) {
976 j = adapter->tx_ring[r_idx].reg_idx;
977 ixgbe_set_ivar(adapter, 1, j, v_idx);
978 r_idx = find_next_bit(q_vector->txr_idx,
979 adapter->num_tx_queues,
983 if (q_vector->txr_count && !q_vector->rxr_count)
985 q_vector->eitr = adapter->tx_eitr_param;
986 else if (q_vector->rxr_count)
988 q_vector->eitr = adapter->rx_eitr_param;
990 ixgbe_write_eitr(q_vector);
993 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
994 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
996 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
997 ixgbe_set_ivar(adapter, -1, 1, v_idx);
998 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1000 /* set up to autoclear timer, and the vectors */
1001 mask = IXGBE_EIMS_ENABLE_MASK;
1002 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1003 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1006 enum latency_range {
1010 latency_invalid = 255
1014 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1015 * @adapter: pointer to adapter
1016 * @eitr: eitr setting (ints per sec) to give last timeslice
1017 * @itr_setting: current throttle rate in ints/second
1018 * @packets: the number of packets during this measurement interval
1019 * @bytes: the number of bytes during this measurement interval
1021 * Stores a new ITR value based on packets and byte
1022 * counts during the last interrupt. The advantage of per interrupt
1023 * computation is faster updates and more accurate ITR for the current
1024 * traffic pattern. Constants in this function were computed
1025 * based on theoretical maximum wire speed and thresholds were set based
1026 * on testing data as well as attempting to minimize response time
1027 * while increasing bulk throughput.
1028 * this functionality is controlled by the InterruptThrottleRate module
1029 * parameter (see ixgbe_param.c)
1031 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1032 u32 eitr, u8 itr_setting,
1033 int packets, int bytes)
1035 unsigned int retval = itr_setting;
1040 goto update_itr_done;
1043 /* simple throttlerate management
1044 * 0-20MB/s lowest (100000 ints/s)
1045 * 20-100MB/s low (20000 ints/s)
1046 * 100-1249MB/s bulk (8000 ints/s)
1048 /* what was last interrupt timeslice? */
1049 timepassed_us = 1000000/eitr;
1050 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1052 switch (itr_setting) {
1053 case lowest_latency:
1054 if (bytes_perint > adapter->eitr_low)
1055 retval = low_latency;
1058 if (bytes_perint > adapter->eitr_high)
1059 retval = bulk_latency;
1060 else if (bytes_perint <= adapter->eitr_low)
1061 retval = lowest_latency;
1064 if (bytes_perint <= adapter->eitr_high)
1065 retval = low_latency;
1074 * ixgbe_write_eitr - write EITR register in hardware specific way
1075 * @q_vector: structure containing interrupt and ring information
1077 * This function is made to be called by ethtool and by the driver
1078 * when it needs to update EITR registers at runtime. Hardware
1079 * specific quirks/differences are taken care of here.
1081 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1083 struct ixgbe_adapter *adapter = q_vector->adapter;
1084 struct ixgbe_hw *hw = &adapter->hw;
1085 int v_idx = q_vector->v_idx;
1086 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1088 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1089 /* must write high and low 16 bits to reset counter */
1090 itr_reg |= (itr_reg << 16);
1091 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1093 * set the WDIS bit to not clear the timer bits and cause an
1094 * immediate assertion of the interrupt
1096 itr_reg |= IXGBE_EITR_CNT_WDIS;
1098 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1101 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1103 struct ixgbe_adapter *adapter = q_vector->adapter;
1105 u8 current_itr, ret_itr;
1107 struct ixgbe_ring *rx_ring, *tx_ring;
1109 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1110 for (i = 0; i < q_vector->txr_count; i++) {
1111 tx_ring = &(adapter->tx_ring[r_idx]);
1112 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1114 tx_ring->total_packets,
1115 tx_ring->total_bytes);
1116 /* if the result for this queue would decrease interrupt
1117 * rate for this vector then use that result */
1118 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1119 q_vector->tx_itr - 1 : ret_itr);
1120 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1124 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1125 for (i = 0; i < q_vector->rxr_count; i++) {
1126 rx_ring = &(adapter->rx_ring[r_idx]);
1127 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1129 rx_ring->total_packets,
1130 rx_ring->total_bytes);
1131 /* if the result for this queue would decrease interrupt
1132 * rate for this vector then use that result */
1133 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1134 q_vector->rx_itr - 1 : ret_itr);
1135 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1139 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1141 switch (current_itr) {
1142 /* counts and packets in update_itr are dependent on these numbers */
1143 case lowest_latency:
1147 new_itr = 20000; /* aka hwitr = ~200 */
1155 if (new_itr != q_vector->eitr) {
1156 /* do an exponential smoothing */
1157 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1159 /* save the algorithm value here, not the smoothed one */
1160 q_vector->eitr = new_itr;
1162 ixgbe_write_eitr(q_vector);
1168 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1170 struct ixgbe_hw *hw = &adapter->hw;
1172 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1173 (eicr & IXGBE_EICR_GPI_SDP1)) {
1174 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
1175 /* write to clear the interrupt */
1176 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1180 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1182 struct ixgbe_hw *hw = &adapter->hw;
1184 if (eicr & IXGBE_EICR_GPI_SDP1) {
1185 /* Clear the interrupt */
1186 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1187 schedule_work(&adapter->multispeed_fiber_task);
1188 } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1189 /* Clear the interrupt */
1190 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1191 schedule_work(&adapter->sfp_config_module_task);
1193 /* Interrupt isn't for us... */
1198 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1200 struct ixgbe_hw *hw = &adapter->hw;
1203 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1204 adapter->link_check_timeout = jiffies;
1205 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1206 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1207 schedule_work(&adapter->watchdog_task);
1211 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1213 struct net_device *netdev = data;
1214 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1215 struct ixgbe_hw *hw = &adapter->hw;
1219 * Workaround for Silicon errata. Use clear-by-write instead
1220 * of clear-by-read. Reading with EICS will return the
1221 * interrupt causes without clearing, which later be done
1222 * with the write to EICR.
1224 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1225 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1227 if (eicr & IXGBE_EICR_LSC)
1228 ixgbe_check_lsc(adapter);
1230 if (hw->mac.type == ixgbe_mac_82598EB)
1231 ixgbe_check_fan_failure(adapter, eicr);
1233 if (hw->mac.type == ixgbe_mac_82599EB) {
1234 ixgbe_check_sfp_event(adapter, eicr);
1236 /* Handle Flow Director Full threshold interrupt */
1237 if (eicr & IXGBE_EICR_FLOW_DIR) {
1239 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1240 /* Disable transmits before FDIR Re-initialization */
1241 netif_tx_stop_all_queues(netdev);
1242 for (i = 0; i < adapter->num_tx_queues; i++) {
1243 struct ixgbe_ring *tx_ring =
1244 &adapter->tx_ring[i];
1245 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1246 &tx_ring->reinit_state))
1247 schedule_work(&adapter->fdir_reinit_task);
1251 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1252 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1257 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1262 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1263 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1264 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1266 mask = (qmask & 0xFFFFFFFF);
1267 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1268 mask = (qmask >> 32);
1269 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1271 /* skip the flush */
1274 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1279 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1280 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1281 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1283 mask = (qmask & 0xFFFFFFFF);
1284 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1285 mask = (qmask >> 32);
1286 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1288 /* skip the flush */
1291 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1293 struct ixgbe_q_vector *q_vector = data;
1294 struct ixgbe_adapter *adapter = q_vector->adapter;
1295 struct ixgbe_ring *tx_ring;
1298 if (!q_vector->txr_count)
1301 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1302 for (i = 0; i < q_vector->txr_count; i++) {
1303 tx_ring = &(adapter->tx_ring[r_idx]);
1304 tx_ring->total_bytes = 0;
1305 tx_ring->total_packets = 0;
1306 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1310 /* disable interrupts on this vector only */
1311 ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1312 napi_schedule(&q_vector->napi);
1318 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1320 * @data: pointer to our q_vector struct for this interrupt vector
1322 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1324 struct ixgbe_q_vector *q_vector = data;
1325 struct ixgbe_adapter *adapter = q_vector->adapter;
1326 struct ixgbe_ring *rx_ring;
1330 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1331 for (i = 0; i < q_vector->rxr_count; i++) {
1332 rx_ring = &(adapter->rx_ring[r_idx]);
1333 rx_ring->total_bytes = 0;
1334 rx_ring->total_packets = 0;
1335 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1339 if (!q_vector->rxr_count)
1342 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1343 rx_ring = &(adapter->rx_ring[r_idx]);
1344 /* disable interrupts on this vector only */
1345 ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1346 napi_schedule(&q_vector->napi);
1351 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1353 struct ixgbe_q_vector *q_vector = data;
1354 struct ixgbe_adapter *adapter = q_vector->adapter;
1355 struct ixgbe_ring *ring;
1359 if (!q_vector->txr_count && !q_vector->rxr_count)
1362 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1363 for (i = 0; i < q_vector->txr_count; i++) {
1364 ring = &(adapter->tx_ring[r_idx]);
1365 ring->total_bytes = 0;
1366 ring->total_packets = 0;
1367 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1371 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1372 for (i = 0; i < q_vector->rxr_count; i++) {
1373 ring = &(adapter->rx_ring[r_idx]);
1374 ring->total_bytes = 0;
1375 ring->total_packets = 0;
1376 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1380 /* disable interrupts on this vector only */
1381 ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1382 napi_schedule(&q_vector->napi);
1388 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1389 * @napi: napi struct with our devices info in it
1390 * @budget: amount of work driver is allowed to do this pass, in packets
1392 * This function is optimized for cleaning one queue only on a single
1395 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1397 struct ixgbe_q_vector *q_vector =
1398 container_of(napi, struct ixgbe_q_vector, napi);
1399 struct ixgbe_adapter *adapter = q_vector->adapter;
1400 struct ixgbe_ring *rx_ring = NULL;
1404 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1405 rx_ring = &(adapter->rx_ring[r_idx]);
1406 #ifdef CONFIG_IXGBE_DCA
1407 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1408 ixgbe_update_rx_dca(adapter, rx_ring);
1411 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1413 /* If all Rx work done, exit the polling mode */
1414 if (work_done < budget) {
1415 napi_complete(napi);
1416 if (adapter->rx_itr_setting & 1)
1417 ixgbe_set_itr_msix(q_vector);
1418 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1419 ixgbe_irq_enable_queues(adapter,
1420 ((u64)1 << q_vector->v_idx));
1427 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1428 * @napi: napi struct with our devices info in it
1429 * @budget: amount of work driver is allowed to do this pass, in packets
1431 * This function will clean more than one rx queue associated with a
1434 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1436 struct ixgbe_q_vector *q_vector =
1437 container_of(napi, struct ixgbe_q_vector, napi);
1438 struct ixgbe_adapter *adapter = q_vector->adapter;
1439 struct ixgbe_ring *ring = NULL;
1440 int work_done = 0, i;
1442 bool tx_clean_complete = true;
1444 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1445 for (i = 0; i < q_vector->txr_count; i++) {
1446 ring = &(adapter->tx_ring[r_idx]);
1447 #ifdef CONFIG_IXGBE_DCA
1448 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1449 ixgbe_update_tx_dca(adapter, ring);
1451 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1452 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1456 /* attempt to distribute budget to each queue fairly, but don't allow
1457 * the budget to go below 1 because we'll exit polling */
1458 budget /= (q_vector->rxr_count ?: 1);
1459 budget = max(budget, 1);
1460 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1461 for (i = 0; i < q_vector->rxr_count; i++) {
1462 ring = &(adapter->rx_ring[r_idx]);
1463 #ifdef CONFIG_IXGBE_DCA
1464 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1465 ixgbe_update_rx_dca(adapter, ring);
1467 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1468 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1472 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1473 ring = &(adapter->rx_ring[r_idx]);
1474 /* If all Rx work done, exit the polling mode */
1475 if (work_done < budget) {
1476 napi_complete(napi);
1477 if (adapter->rx_itr_setting & 1)
1478 ixgbe_set_itr_msix(q_vector);
1479 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1480 ixgbe_irq_enable_queues(adapter,
1481 ((u64)1 << q_vector->v_idx));
1489 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1490 * @napi: napi struct with our devices info in it
1491 * @budget: amount of work driver is allowed to do this pass, in packets
1493 * This function is optimized for cleaning one queue only on a single
1496 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1498 struct ixgbe_q_vector *q_vector =
1499 container_of(napi, struct ixgbe_q_vector, napi);
1500 struct ixgbe_adapter *adapter = q_vector->adapter;
1501 struct ixgbe_ring *tx_ring = NULL;
1505 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1506 tx_ring = &(adapter->tx_ring[r_idx]);
1507 #ifdef CONFIG_IXGBE_DCA
1508 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1509 ixgbe_update_tx_dca(adapter, tx_ring);
1512 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
1515 /* If all Tx work done, exit the polling mode */
1516 if (work_done < budget) {
1517 napi_complete(napi);
1518 if (adapter->tx_itr_setting & 1)
1519 ixgbe_set_itr_msix(q_vector);
1520 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1521 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
1527 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1530 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1532 set_bit(r_idx, q_vector->rxr_idx);
1533 q_vector->rxr_count++;
1536 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1539 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1541 set_bit(t_idx, q_vector->txr_idx);
1542 q_vector->txr_count++;
1546 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1547 * @adapter: board private structure to initialize
1548 * @vectors: allotted vector count for descriptor rings
1550 * This function maps descriptor rings to the queue-specific vectors
1551 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1552 * one vector per ring/queue, but on a constrained vector budget, we
1553 * group the rings as "efficiently" as possible. You would add new
1554 * mapping configurations in here.
1556 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1560 int rxr_idx = 0, txr_idx = 0;
1561 int rxr_remaining = adapter->num_rx_queues;
1562 int txr_remaining = adapter->num_tx_queues;
1567 /* No mapping required if MSI-X is disabled. */
1568 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1572 * The ideal configuration...
1573 * We have enough vectors to map one per queue.
1575 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1576 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1577 map_vector_to_rxq(adapter, v_start, rxr_idx);
1579 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1580 map_vector_to_txq(adapter, v_start, txr_idx);
1586 * If we don't have enough vectors for a 1-to-1
1587 * mapping, we'll have to group them so there are
1588 * multiple queues per vector.
1590 /* Re-adjusting *qpv takes care of the remainder. */
1591 for (i = v_start; i < vectors; i++) {
1592 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1593 for (j = 0; j < rqpv; j++) {
1594 map_vector_to_rxq(adapter, i, rxr_idx);
1599 for (i = v_start; i < vectors; i++) {
1600 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1601 for (j = 0; j < tqpv; j++) {
1602 map_vector_to_txq(adapter, i, txr_idx);
1613 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1614 * @adapter: board private structure
1616 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1617 * interrupts from the kernel.
1619 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1621 struct net_device *netdev = adapter->netdev;
1622 irqreturn_t (*handler)(int, void *);
1623 int i, vector, q_vectors, err;
1626 /* Decrement for Other and TCP Timer vectors */
1627 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1629 /* Map the Tx/Rx rings to the vectors we were allotted. */
1630 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1634 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1635 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1636 &ixgbe_msix_clean_many)
1637 for (vector = 0; vector < q_vectors; vector++) {
1638 handler = SET_HANDLER(adapter->q_vector[vector]);
1640 if(handler == &ixgbe_msix_clean_rx) {
1641 sprintf(adapter->name[vector], "%s-%s-%d",
1642 netdev->name, "rx", ri++);
1644 else if(handler == &ixgbe_msix_clean_tx) {
1645 sprintf(adapter->name[vector], "%s-%s-%d",
1646 netdev->name, "tx", ti++);
1649 sprintf(adapter->name[vector], "%s-%s-%d",
1650 netdev->name, "TxRx", vector);
1652 err = request_irq(adapter->msix_entries[vector].vector,
1653 handler, 0, adapter->name[vector],
1654 adapter->q_vector[vector]);
1657 "request_irq failed for MSIX interrupt "
1658 "Error: %d\n", err);
1659 goto free_queue_irqs;
1663 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1664 err = request_irq(adapter->msix_entries[vector].vector,
1665 &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1668 "request_irq for msix_lsc failed: %d\n", err);
1669 goto free_queue_irqs;
1675 for (i = vector - 1; i >= 0; i--)
1676 free_irq(adapter->msix_entries[--vector].vector,
1677 adapter->q_vector[i]);
1678 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1679 pci_disable_msix(adapter->pdev);
1680 kfree(adapter->msix_entries);
1681 adapter->msix_entries = NULL;
1686 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1688 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1690 u32 new_itr = q_vector->eitr;
1691 struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1692 struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1694 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1696 tx_ring->total_packets,
1697 tx_ring->total_bytes);
1698 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1700 rx_ring->total_packets,
1701 rx_ring->total_bytes);
1703 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1705 switch (current_itr) {
1706 /* counts and packets in update_itr are dependent on these numbers */
1707 case lowest_latency:
1711 new_itr = 20000; /* aka hwitr = ~200 */
1720 if (new_itr != q_vector->eitr) {
1721 /* do an exponential smoothing */
1722 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1724 /* save the algorithm value here, not the smoothed one */
1725 q_vector->eitr = new_itr;
1727 ixgbe_write_eitr(q_vector);
1734 * ixgbe_irq_enable - Enable default interrupt generation settings
1735 * @adapter: board private structure
1737 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1741 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1742 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1743 mask |= IXGBE_EIMS_GPI_SDP1;
1744 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1745 mask |= IXGBE_EIMS_ECC;
1746 mask |= IXGBE_EIMS_GPI_SDP1;
1747 mask |= IXGBE_EIMS_GPI_SDP2;
1749 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
1750 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
1751 mask |= IXGBE_EIMS_FLOW_DIR;
1753 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1754 ixgbe_irq_enable_queues(adapter, ~0);
1755 IXGBE_WRITE_FLUSH(&adapter->hw);
1759 * ixgbe_intr - legacy mode Interrupt Handler
1760 * @irq: interrupt number
1761 * @data: pointer to a network interface device structure
1763 static irqreturn_t ixgbe_intr(int irq, void *data)
1765 struct net_device *netdev = data;
1766 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1767 struct ixgbe_hw *hw = &adapter->hw;
1768 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1772 * Workaround for silicon errata. Mask the interrupts
1773 * before the read of EICR.
1775 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1777 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1778 * therefore no explict interrupt disable is necessary */
1779 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1781 /* shared interrupt alert!
1782 * make sure interrupts are enabled because the read will
1783 * have disabled interrupts due to EIAM */
1784 ixgbe_irq_enable(adapter);
1785 return IRQ_NONE; /* Not our interrupt */
1788 if (eicr & IXGBE_EICR_LSC)
1789 ixgbe_check_lsc(adapter);
1791 if (hw->mac.type == ixgbe_mac_82599EB)
1792 ixgbe_check_sfp_event(adapter, eicr);
1794 ixgbe_check_fan_failure(adapter, eicr);
1796 if (napi_schedule_prep(&(q_vector->napi))) {
1797 adapter->tx_ring[0].total_packets = 0;
1798 adapter->tx_ring[0].total_bytes = 0;
1799 adapter->rx_ring[0].total_packets = 0;
1800 adapter->rx_ring[0].total_bytes = 0;
1801 /* would disable interrupts here but EIAM disabled it */
1802 __napi_schedule(&(q_vector->napi));
1808 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1810 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1812 for (i = 0; i < q_vectors; i++) {
1813 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
1814 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1815 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1816 q_vector->rxr_count = 0;
1817 q_vector->txr_count = 0;
1822 * ixgbe_request_irq - initialize interrupts
1823 * @adapter: board private structure
1825 * Attempts to configure interrupts using the best available
1826 * capabilities of the hardware and kernel.
1828 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1830 struct net_device *netdev = adapter->netdev;
1833 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1834 err = ixgbe_request_msix_irqs(adapter);
1835 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1836 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1837 netdev->name, netdev);
1839 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1840 netdev->name, netdev);
1844 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1849 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1851 struct net_device *netdev = adapter->netdev;
1853 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1856 q_vectors = adapter->num_msix_vectors;
1859 free_irq(adapter->msix_entries[i].vector, netdev);
1862 for (; i >= 0; i--) {
1863 free_irq(adapter->msix_entries[i].vector,
1864 adapter->q_vector[i]);
1867 ixgbe_reset_q_vectors(adapter);
1869 free_irq(adapter->pdev->irq, netdev);
1874 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1875 * @adapter: board private structure
1877 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1879 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1880 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1882 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
1883 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
1884 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
1886 IXGBE_WRITE_FLUSH(&adapter->hw);
1887 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1889 for (i = 0; i < adapter->num_msix_vectors; i++)
1890 synchronize_irq(adapter->msix_entries[i].vector);
1892 synchronize_irq(adapter->pdev->irq);
1897 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1900 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1902 struct ixgbe_hw *hw = &adapter->hw;
1904 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1905 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
1907 ixgbe_set_ivar(adapter, 0, 0, 0);
1908 ixgbe_set_ivar(adapter, 1, 0, 0);
1910 map_vector_to_rxq(adapter, 0, 0);
1911 map_vector_to_txq(adapter, 0, 0);
1913 DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1917 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1918 * @adapter: board private structure
1920 * Configure the Tx unit of the MAC after a reset.
1922 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1925 struct ixgbe_hw *hw = &adapter->hw;
1926 u32 i, j, tdlen, txctrl;
1928 /* Setup the HW Tx Head and Tail descriptor pointers */
1929 for (i = 0; i < adapter->num_tx_queues; i++) {
1930 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1933 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1934 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1935 (tdba & DMA_BIT_MASK(32)));
1936 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1937 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1938 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1939 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1940 adapter->tx_ring[i].head = IXGBE_TDH(j);
1941 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1943 * Disable Tx Head Writeback RO bit, since this hoses
1944 * bookkeeping if things aren't delivered in order.
1946 switch (hw->mac.type) {
1947 case ixgbe_mac_82598EB:
1948 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1950 case ixgbe_mac_82599EB:
1952 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(j));
1955 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1956 switch (hw->mac.type) {
1957 case ixgbe_mac_82598EB:
1958 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1960 case ixgbe_mac_82599EB:
1962 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(j), txctrl);
1966 if (hw->mac.type == ixgbe_mac_82599EB) {
1967 /* We enable 8 traffic classes, DCB only */
1968 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
1969 IXGBE_WRITE_REG(hw, IXGBE_MTQC, (IXGBE_MTQC_RT_ENA |
1970 IXGBE_MTQC_8TC_8TQ));
1974 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1976 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
1977 struct ixgbe_ring *rx_ring)
1981 struct ixgbe_ring_feature *feature = adapter->ring_feature;
1983 index = rx_ring->reg_idx;
1984 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1986 mask = (unsigned long) feature[RING_F_RSS].mask;
1987 index = index & mask;
1989 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
1991 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1992 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1994 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1995 IXGBE_SRRCTL_BSIZEHDR_MASK;
1997 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1998 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
1999 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2001 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2003 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2005 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2006 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2007 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2010 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2013 static u32 ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2018 if (!(adapter->hw.mac.type == ixgbe_mac_82599EB))
2021 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2022 #ifdef CONFIG_IXGBE_DCB
2023 | IXGBE_FLAG_DCB_ENABLED
2028 case (IXGBE_FLAG_RSS_ENABLED):
2029 mrqc = IXGBE_MRQC_RSSEN;
2031 #ifdef CONFIG_IXGBE_DCB
2032 case (IXGBE_FLAG_DCB_ENABLED):
2033 mrqc = IXGBE_MRQC_RT8TCEN;
2035 #endif /* CONFIG_IXGBE_DCB */
2044 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2045 * @adapter: address of board private structure
2046 * @index: index of ring to set
2047 * @rx_buf_len: rx buffer length
2049 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter, int index,
2052 struct ixgbe_ring *rx_ring;
2053 struct ixgbe_hw *hw = &adapter->hw;
2057 rx_ring = &adapter->rx_ring[index];
2058 j = rx_ring->reg_idx;
2059 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2060 rscctrl |= IXGBE_RSCCTL_RSCEN;
2062 * we must limit the number of descriptors so that the
2063 * total size of max desc * buf_len is not greater
2066 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2067 #if (MAX_SKB_FRAGS > 16)
2068 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2069 #elif (MAX_SKB_FRAGS > 8)
2070 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2071 #elif (MAX_SKB_FRAGS > 4)
2072 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2074 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2077 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2078 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2079 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2080 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2082 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2084 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2088 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2089 * @adapter: board private structure
2091 * Configure the Rx unit of the MAC after a reset.
2093 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2096 struct ixgbe_hw *hw = &adapter->hw;
2097 struct ixgbe_ring *rx_ring;
2098 struct net_device *netdev = adapter->netdev;
2099 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2101 u32 rdlen, rxctrl, rxcsum;
2102 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2103 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2104 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2106 u32 reta = 0, mrqc = 0;
2110 /* Decide whether to use packet split mode or not */
2111 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2113 /* Set the RX buffer length according to the mode */
2114 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2115 rx_buf_len = IXGBE_RX_HDR_SIZE;
2116 if (hw->mac.type == ixgbe_mac_82599EB) {
2117 /* PSRTYPE must be initialized in 82599 */
2118 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2119 IXGBE_PSRTYPE_UDPHDR |
2120 IXGBE_PSRTYPE_IPV4HDR |
2121 IXGBE_PSRTYPE_IPV6HDR |
2122 IXGBE_PSRTYPE_L2HDR;
2123 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), psrtype);
2126 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2127 (netdev->mtu <= ETH_DATA_LEN))
2128 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2130 rx_buf_len = ALIGN(max_frame, 1024);
2133 fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
2134 fctrl |= IXGBE_FCTRL_BAM;
2135 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
2136 fctrl |= IXGBE_FCTRL_PMCF;
2137 IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
2139 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2140 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2141 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2143 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2145 if (netdev->features & NETIF_F_FCOE_MTU)
2146 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2148 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2150 rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
2151 /* disable receives while setting up the descriptors */
2152 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2153 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2156 * Setup the HW Rx Head and Tail Descriptor Pointers and
2157 * the Base and Length of the Rx Descriptor Ring
2159 for (i = 0; i < adapter->num_rx_queues; i++) {
2160 rx_ring = &adapter->rx_ring[i];
2161 rdba = rx_ring->dma;
2162 j = rx_ring->reg_idx;
2163 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_BIT_MASK(32)));
2164 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
2165 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
2166 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
2167 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
2168 rx_ring->head = IXGBE_RDH(j);
2169 rx_ring->tail = IXGBE_RDT(j);
2170 rx_ring->rx_buf_len = rx_buf_len;
2172 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2173 rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2175 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2178 if (netdev->features & NETIF_F_FCOE_MTU) {
2179 struct ixgbe_ring_feature *f;
2180 f = &adapter->ring_feature[RING_F_FCOE];
2181 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2182 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2183 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2184 rx_ring->rx_buf_len =
2185 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2189 #endif /* IXGBE_FCOE */
2190 ixgbe_configure_srrctl(adapter, rx_ring);
2193 if (hw->mac.type == ixgbe_mac_82598EB) {
2195 * For VMDq support of different descriptor types or
2196 * buffer sizes through the use of multiple SRRCTL
2197 * registers, RDRXCTL.MVMEN must be set to 1
2199 * also, the manual doesn't mention it clearly but DCA hints
2200 * will only use queue 0's tags unless this bit is set. Side
2201 * effects of setting this bit are only that SRRCTL must be
2202 * fully programmed [0..15]
2204 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2205 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2206 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2209 /* Program MRQC for the distribution of queues */
2210 mrqc = ixgbe_setup_mrqc(adapter);
2212 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2213 /* Fill out redirection table */
2214 for (i = 0, j = 0; i < 128; i++, j++) {
2215 if (j == adapter->ring_feature[RING_F_RSS].indices)
2217 /* reta = 4-byte sliding window of
2218 * 0x00..(indices-1)(indices-1)00..etc. */
2219 reta = (reta << 8) | (j * 0x11);
2221 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2224 /* Fill out hash function seeds */
2225 for (i = 0; i < 10; i++)
2226 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2228 if (hw->mac.type == ixgbe_mac_82598EB)
2229 mrqc |= IXGBE_MRQC_RSSEN;
2230 /* Perform hash on these packet types */
2231 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2232 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2233 | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
2234 | IXGBE_MRQC_RSS_FIELD_IPV6
2235 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
2236 | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2238 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2240 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2242 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
2243 adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
2244 /* Disable indicating checksum in descriptor, enables
2246 rxcsum |= IXGBE_RXCSUM_PCSD;
2248 if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
2249 /* Enable IPv4 payload checksum for UDP fragments
2250 * if PCSD is not set */
2251 rxcsum |= IXGBE_RXCSUM_IPPCSE;
2254 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2256 if (hw->mac.type == ixgbe_mac_82599EB) {
2257 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2258 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2259 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2260 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2263 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
2264 /* Enable 82599 HW-RSC */
2265 for (i = 0; i < adapter->num_rx_queues; i++)
2266 ixgbe_configure_rscctl(adapter, i, rx_buf_len);
2268 /* Disable RSC for ACK packets */
2269 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2270 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2274 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2276 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2277 struct ixgbe_hw *hw = &adapter->hw;
2279 /* add VID to filter table */
2280 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
2283 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2285 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2286 struct ixgbe_hw *hw = &adapter->hw;
2288 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2289 ixgbe_irq_disable(adapter);
2291 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2293 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2294 ixgbe_irq_enable(adapter);
2296 /* remove VID from filter table */
2297 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
2300 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2301 struct vlan_group *grp)
2303 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2307 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2308 ixgbe_irq_disable(adapter);
2309 adapter->vlgrp = grp;
2312 * For a DCB driver, always enable VLAN tag stripping so we can
2313 * still receive traffic from a DCB-enabled host even if we're
2316 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2317 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2318 ctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2319 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2320 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2321 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2322 ctrl |= IXGBE_VLNCTRL_VFE;
2323 /* enable VLAN tag insert/strip */
2324 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2325 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2326 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2327 for (i = 0; i < adapter->num_rx_queues; i++) {
2328 j = adapter->rx_ring[i].reg_idx;
2329 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_RXDCTL(j));
2330 ctrl |= IXGBE_RXDCTL_VME;
2331 IXGBE_WRITE_REG(&adapter->hw, IXGBE_RXDCTL(j), ctrl);
2334 ixgbe_vlan_rx_add_vid(netdev, 0);
2336 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2337 ixgbe_irq_enable(adapter);
2340 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2342 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2344 if (adapter->vlgrp) {
2346 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2347 if (!vlan_group_get_device(adapter->vlgrp, vid))
2349 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2354 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
2356 struct dev_mc_list *mc_ptr;
2357 u8 *addr = *mc_addr_ptr;
2360 mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
2362 *mc_addr_ptr = mc_ptr->next->dmi_addr;
2364 *mc_addr_ptr = NULL;
2370 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
2371 * @netdev: network interface device structure
2373 * The set_rx_method entry point is called whenever the unicast/multicast
2374 * address list or the network interface flags are updated. This routine is
2375 * responsible for configuring the hardware for proper unicast, multicast and
2378 static void ixgbe_set_rx_mode(struct net_device *netdev)
2380 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2381 struct ixgbe_hw *hw = &adapter->hw;
2383 u8 *addr_list = NULL;
2386 /* Check for Promiscuous and All Multicast modes */
2388 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2389 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2391 if (netdev->flags & IFF_PROMISC) {
2392 hw->addr_ctrl.user_set_promisc = 1;
2393 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2394 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2396 if (netdev->flags & IFF_ALLMULTI) {
2397 fctrl |= IXGBE_FCTRL_MPE;
2398 fctrl &= ~IXGBE_FCTRL_UPE;
2400 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2402 vlnctrl |= IXGBE_VLNCTRL_VFE;
2403 hw->addr_ctrl.user_set_promisc = 0;
2406 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
2407 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2409 /* reprogram secondary unicast list */
2410 hw->mac.ops.update_uc_addr_list(hw, &netdev->uc.list);
2412 /* reprogram multicast list */
2413 addr_count = netdev->mc_count;
2415 addr_list = netdev->mc_list->dmi_addr;
2416 hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
2417 ixgbe_addr_list_itr);
2420 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
2423 struct ixgbe_q_vector *q_vector;
2424 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2426 /* legacy and MSI only use one vector */
2427 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2430 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2431 struct napi_struct *napi;
2432 q_vector = adapter->q_vector[q_idx];
2433 napi = &q_vector->napi;
2434 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2435 if (!q_vector->rxr_count || !q_vector->txr_count) {
2436 if (q_vector->txr_count == 1)
2437 napi->poll = &ixgbe_clean_txonly;
2438 else if (q_vector->rxr_count == 1)
2439 napi->poll = &ixgbe_clean_rxonly;
2447 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2450 struct ixgbe_q_vector *q_vector;
2451 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2453 /* legacy and MSI only use one vector */
2454 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2457 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2458 q_vector = adapter->q_vector[q_idx];
2459 napi_disable(&q_vector->napi);
2463 #ifdef CONFIG_IXGBE_DCB
2465 * ixgbe_configure_dcb - Configure DCB hardware
2466 * @adapter: ixgbe adapter struct
2468 * This is called by the driver on open to configure the DCB hardware.
2469 * This is also called by the gennetlink interface when reconfiguring
2472 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2474 struct ixgbe_hw *hw = &adapter->hw;
2475 u32 txdctl, vlnctrl;
2478 ixgbe_dcb_check_config(&adapter->dcb_cfg);
2479 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2480 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2482 /* reconfigure the hardware */
2483 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
2485 for (i = 0; i < adapter->num_tx_queues; i++) {
2486 j = adapter->tx_ring[i].reg_idx;
2487 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2488 /* PThresh workaround for Tx hang with DFP enabled. */
2490 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2492 /* Enable VLAN tag insert/strip */
2493 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2494 if (hw->mac.type == ixgbe_mac_82598EB) {
2495 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2496 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2497 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2498 } else if (hw->mac.type == ixgbe_mac_82599EB) {
2499 vlnctrl |= IXGBE_VLNCTRL_VFE;
2500 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2501 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2502 for (i = 0; i < adapter->num_rx_queues; i++) {
2503 j = adapter->rx_ring[i].reg_idx;
2504 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2505 vlnctrl |= IXGBE_RXDCTL_VME;
2506 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2509 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2513 static void ixgbe_configure(struct ixgbe_adapter *adapter)
2515 struct net_device *netdev = adapter->netdev;
2516 struct ixgbe_hw *hw = &adapter->hw;
2519 ixgbe_set_rx_mode(netdev);
2521 ixgbe_restore_vlan(adapter);
2522 #ifdef CONFIG_IXGBE_DCB
2523 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2524 if (hw->mac.type == ixgbe_mac_82598EB)
2525 netif_set_gso_max_size(netdev, 32768);
2527 netif_set_gso_max_size(netdev, 65536);
2528 ixgbe_configure_dcb(adapter);
2530 netif_set_gso_max_size(netdev, 65536);
2533 netif_set_gso_max_size(netdev, 65536);
2537 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2538 ixgbe_configure_fcoe(adapter);
2540 #endif /* IXGBE_FCOE */
2541 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2542 for (i = 0; i < adapter->num_tx_queues; i++)
2543 adapter->tx_ring[i].atr_sample_rate =
2544 adapter->atr_sample_rate;
2545 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
2546 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
2547 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
2550 ixgbe_configure_tx(adapter);
2551 ixgbe_configure_rx(adapter);
2552 for (i = 0; i < adapter->num_rx_queues; i++)
2553 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
2554 (adapter->rx_ring[i].count - 1));
2557 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2559 switch (hw->phy.type) {
2560 case ixgbe_phy_sfp_avago:
2561 case ixgbe_phy_sfp_ftl:
2562 case ixgbe_phy_sfp_intel:
2563 case ixgbe_phy_sfp_unknown:
2564 case ixgbe_phy_tw_tyco:
2565 case ixgbe_phy_tw_unknown:
2573 * ixgbe_sfp_link_config - set up SFP+ link
2574 * @adapter: pointer to private adapter struct
2576 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2578 struct ixgbe_hw *hw = &adapter->hw;
2580 if (hw->phy.multispeed_fiber) {
2582 * In multispeed fiber setups, the device may not have
2583 * had a physical connection when the driver loaded.
2584 * If that's the case, the initial link configuration
2585 * couldn't get the MAC into 10G or 1G mode, so we'll
2586 * never have a link status change interrupt fire.
2587 * We need to try and force an autonegotiation
2588 * session, then bring up link.
2590 hw->mac.ops.setup_sfp(hw);
2591 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2592 schedule_work(&adapter->multispeed_fiber_task);
2595 * Direct Attach Cu and non-multispeed fiber modules
2596 * still need to be configured properly prior to
2599 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2600 schedule_work(&adapter->sfp_config_module_task);
2605 * ixgbe_non_sfp_link_config - set up non-SFP+ link
2606 * @hw: pointer to private hardware struct
2608 * Returns 0 on success, negative on failure
2610 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2613 bool negotiation, link_up = false;
2614 u32 ret = IXGBE_ERR_LINK_SETUP;
2616 if (hw->mac.ops.check_link)
2617 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2622 if (hw->mac.ops.get_link_capabilities)
2623 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
2627 if (hw->mac.ops.setup_link)
2628 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
2633 #define IXGBE_MAX_RX_DESC_POLL 10
2634 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2637 int j = adapter->rx_ring[rxr].reg_idx;
2640 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2641 if (IXGBE_READ_REG(&adapter->hw,
2642 IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2647 if (k >= IXGBE_MAX_RX_DESC_POLL) {
2648 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2649 "not set within the polling period\n", rxr);
2651 ixgbe_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
2652 (adapter->rx_ring[rxr].count - 1));
2655 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2657 struct net_device *netdev = adapter->netdev;
2658 struct ixgbe_hw *hw = &adapter->hw;
2660 int num_rx_rings = adapter->num_rx_queues;
2662 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2663 u32 txdctl, rxdctl, mhadd;
2667 ixgbe_get_hw_control(adapter);
2669 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2670 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2671 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2672 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2673 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2678 /* XXX: to interrupt immediately for EICS writes, enable this */
2679 /* gpie |= IXGBE_GPIE_EIMEN; */
2680 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2683 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2684 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2685 * specifically only auto mask tx and rx interrupts */
2686 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2689 /* Enable fan failure interrupt if media type is copper */
2690 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2691 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2692 gpie |= IXGBE_SDP1_GPIEN;
2693 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2696 if (hw->mac.type == ixgbe_mac_82599EB) {
2697 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2698 gpie |= IXGBE_SDP1_GPIEN;
2699 gpie |= IXGBE_SDP2_GPIEN;
2700 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2704 /* adjust max frame to be able to do baby jumbo for FCoE */
2705 if ((netdev->features & NETIF_F_FCOE_MTU) &&
2706 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2707 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2709 #endif /* IXGBE_FCOE */
2710 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2711 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2712 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2713 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2715 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2718 for (i = 0; i < adapter->num_tx_queues; i++) {
2719 j = adapter->tx_ring[i].reg_idx;
2720 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2721 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2722 txdctl |= (8 << 16);
2723 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2726 if (hw->mac.type == ixgbe_mac_82599EB) {
2727 /* DMATXCTL.EN must be set after all Tx queue config is done */
2728 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2729 dmatxctl |= IXGBE_DMATXCTL_TE;
2730 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2732 for (i = 0; i < adapter->num_tx_queues; i++) {
2733 j = adapter->tx_ring[i].reg_idx;
2734 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2735 txdctl |= IXGBE_TXDCTL_ENABLE;
2736 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2739 for (i = 0; i < num_rx_rings; i++) {
2740 j = adapter->rx_ring[i].reg_idx;
2741 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2742 /* enable PTHRESH=32 descriptors (half the internal cache)
2743 * and HTHRESH=0 descriptors (to minimize latency on fetch),
2744 * this also removes a pesky rx_no_buffer_count increment */
2746 rxdctl |= IXGBE_RXDCTL_ENABLE;
2747 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2748 if (hw->mac.type == ixgbe_mac_82599EB)
2749 ixgbe_rx_desc_queue_enable(adapter, i);
2751 /* enable all receives */
2752 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2753 if (hw->mac.type == ixgbe_mac_82598EB)
2754 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2756 rxdctl |= IXGBE_RXCTRL_RXEN;
2757 hw->mac.ops.enable_rx_dma(hw, rxdctl);
2759 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2760 ixgbe_configure_msix(adapter);
2762 ixgbe_configure_msi_and_legacy(adapter);
2764 clear_bit(__IXGBE_DOWN, &adapter->state);
2765 ixgbe_napi_enable_all(adapter);
2767 /* clear any pending interrupts, may auto mask */
2768 IXGBE_READ_REG(hw, IXGBE_EICR);
2770 ixgbe_irq_enable(adapter);
2773 * If this adapter has a fan, check to see if we had a failure
2774 * before we enabled the interrupt.
2776 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2777 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
2778 if (esdp & IXGBE_ESDP_SDP1)
2780 "Fan has stopped, replace the adapter\n");
2784 * For hot-pluggable SFP+ devices, a new SFP+ module may have
2785 * arrived before interrupts were enabled but after probe. Such
2786 * devices wouldn't have their type identified yet. We need to
2787 * kick off the SFP+ module setup first, then try to bring up link.
2788 * If we're not hot-pluggable SFP+, we just need to configure link
2791 if (hw->phy.type == ixgbe_phy_unknown) {
2792 err = hw->phy.ops.identify(hw);
2793 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
2795 * Take the device down and schedule the sfp tasklet
2796 * which will unregister_netdev and log it.
2798 ixgbe_down(adapter);
2799 schedule_work(&adapter->sfp_config_module_task);
2804 if (ixgbe_is_sfp(hw)) {
2805 ixgbe_sfp_link_config(adapter);
2807 err = ixgbe_non_sfp_link_config(hw);
2809 DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
2812 for (i = 0; i < adapter->num_tx_queues; i++)
2813 set_bit(__IXGBE_FDIR_INIT_DONE,
2814 &(adapter->tx_ring[i].reinit_state));
2816 /* enable transmits */
2817 netif_tx_start_all_queues(netdev);
2819 /* bring the link up in the watchdog, this could race with our first
2820 * link up interrupt but shouldn't be a problem */
2821 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2822 adapter->link_check_timeout = jiffies;
2823 mod_timer(&adapter->watchdog_timer, jiffies);
2827 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
2829 WARN_ON(in_interrupt());
2830 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
2832 ixgbe_down(adapter);
2834 clear_bit(__IXGBE_RESETTING, &adapter->state);
2837 int ixgbe_up(struct ixgbe_adapter *adapter)
2839 /* hardware has been reset, we need to reload some things */
2840 ixgbe_configure(adapter);
2842 return ixgbe_up_complete(adapter);
2845 void ixgbe_reset(struct ixgbe_adapter *adapter)
2847 struct ixgbe_hw *hw = &adapter->hw;
2850 err = hw->mac.ops.init_hw(hw);
2853 case IXGBE_ERR_SFP_NOT_PRESENT:
2855 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
2856 dev_err(&adapter->pdev->dev, "master disable timed out\n");
2858 case IXGBE_ERR_EEPROM_VERSION:
2859 /* We are running on a pre-production device, log a warning */
2860 dev_warn(&adapter->pdev->dev, "This device is a pre-production "
2861 "adapter/LOM. Please be aware there may be issues "
2862 "associated with your hardware. If you are "
2863 "experiencing problems please contact your Intel or "
2864 "hardware representative who provided you with this "
2868 dev_err(&adapter->pdev->dev, "Hardware Error: %d\n", err);
2871 /* reprogram the RAR[0] in case user changed it. */
2872 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2876 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2877 * @adapter: board private structure
2878 * @rx_ring: ring to free buffers from
2880 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2881 struct ixgbe_ring *rx_ring)
2883 struct pci_dev *pdev = adapter->pdev;
2887 /* Free all the Rx ring sk_buffs */
2889 for (i = 0; i < rx_ring->count; i++) {
2890 struct ixgbe_rx_buffer *rx_buffer_info;
2892 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2893 if (rx_buffer_info->dma) {
2894 pci_unmap_single(pdev, rx_buffer_info->dma,
2895 rx_ring->rx_buf_len,
2896 PCI_DMA_FROMDEVICE);
2897 rx_buffer_info->dma = 0;
2899 if (rx_buffer_info->skb) {
2900 struct sk_buff *skb = rx_buffer_info->skb;
2901 rx_buffer_info->skb = NULL;
2903 struct sk_buff *this = skb;
2905 dev_kfree_skb(this);
2908 if (!rx_buffer_info->page)
2910 if (rx_buffer_info->page_dma) {
2911 pci_unmap_page(pdev, rx_buffer_info->page_dma,
2912 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
2913 rx_buffer_info->page_dma = 0;
2915 put_page(rx_buffer_info->page);
2916 rx_buffer_info->page = NULL;
2917 rx_buffer_info->page_offset = 0;
2920 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2921 memset(rx_ring->rx_buffer_info, 0, size);
2923 /* Zero out the descriptor ring */
2924 memset(rx_ring->desc, 0, rx_ring->size);
2926 rx_ring->next_to_clean = 0;
2927 rx_ring->next_to_use = 0;
2930 writel(0, adapter->hw.hw_addr + rx_ring->head);
2932 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2936 * ixgbe_clean_tx_ring - Free Tx Buffers
2937 * @adapter: board private structure
2938 * @tx_ring: ring to be cleaned
2940 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2941 struct ixgbe_ring *tx_ring)
2943 struct ixgbe_tx_buffer *tx_buffer_info;
2947 /* Free all the Tx ring sk_buffs */
2949 for (i = 0; i < tx_ring->count; i++) {
2950 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2951 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2954 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2955 memset(tx_ring->tx_buffer_info, 0, size);
2957 /* Zero out the descriptor ring */
2958 memset(tx_ring->desc, 0, tx_ring->size);
2960 tx_ring->next_to_use = 0;
2961 tx_ring->next_to_clean = 0;
2964 writel(0, adapter->hw.hw_addr + tx_ring->head);
2966 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2970 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2971 * @adapter: board private structure
2973 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2977 for (i = 0; i < adapter->num_rx_queues; i++)
2978 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2982 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2983 * @adapter: board private structure
2985 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2989 for (i = 0; i < adapter->num_tx_queues; i++)
2990 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2993 void ixgbe_down(struct ixgbe_adapter *adapter)
2995 struct net_device *netdev = adapter->netdev;
2996 struct ixgbe_hw *hw = &adapter->hw;
3001 /* signal that we are down to the interrupt handler */
3002 set_bit(__IXGBE_DOWN, &adapter->state);
3004 /* disable receives */
3005 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3006 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3008 netif_tx_disable(netdev);
3010 IXGBE_WRITE_FLUSH(hw);
3013 netif_tx_stop_all_queues(netdev);
3015 ixgbe_irq_disable(adapter);
3017 ixgbe_napi_disable_all(adapter);
3019 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3020 del_timer_sync(&adapter->sfp_timer);
3021 del_timer_sync(&adapter->watchdog_timer);
3022 cancel_work_sync(&adapter->watchdog_task);
3024 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3025 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3026 cancel_work_sync(&adapter->fdir_reinit_task);
3028 /* disable transmits in the hardware now that interrupts are off */
3029 for (i = 0; i < adapter->num_tx_queues; i++) {
3030 j = adapter->tx_ring[i].reg_idx;
3031 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3032 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3033 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3035 /* Disable the Tx DMA engine on 82599 */
3036 if (hw->mac.type == ixgbe_mac_82599EB)
3037 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3038 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3039 ~IXGBE_DMATXCTL_TE));
3041 netif_carrier_off(netdev);
3043 if (!pci_channel_offline(adapter->pdev))
3044 ixgbe_reset(adapter);
3045 ixgbe_clean_all_tx_rings(adapter);
3046 ixgbe_clean_all_rx_rings(adapter);
3048 #ifdef CONFIG_IXGBE_DCA
3049 /* since we reset the hardware DCA settings were cleared */
3050 ixgbe_setup_dca(adapter);
3055 * ixgbe_poll - NAPI Rx polling callback
3056 * @napi: structure for representing this polling device
3057 * @budget: how many packets driver is allowed to clean
3059 * This function is used for legacy and MSI, NAPI mode
3061 static int ixgbe_poll(struct napi_struct *napi, int budget)
3063 struct ixgbe_q_vector *q_vector =
3064 container_of(napi, struct ixgbe_q_vector, napi);
3065 struct ixgbe_adapter *adapter = q_vector->adapter;
3066 int tx_clean_complete, work_done = 0;
3068 #ifdef CONFIG_IXGBE_DCA
3069 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3070 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
3071 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
3075 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring);
3076 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring, &work_done, budget);
3078 if (!tx_clean_complete)
3081 /* If budget not fully consumed, exit the polling mode */
3082 if (work_done < budget) {
3083 napi_complete(napi);
3084 if (adapter->rx_itr_setting & 1)
3085 ixgbe_set_itr(adapter);
3086 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3087 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3093 * ixgbe_tx_timeout - Respond to a Tx Hang
3094 * @netdev: network interface device structure
3096 static void ixgbe_tx_timeout(struct net_device *netdev)
3098 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3100 /* Do the reset outside of interrupt context */
3101 schedule_work(&adapter->reset_task);
3104 static void ixgbe_reset_task(struct work_struct *work)
3106 struct ixgbe_adapter *adapter;
3107 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3109 /* If we're already down or resetting, just bail */
3110 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3111 test_bit(__IXGBE_RESETTING, &adapter->state))
3114 adapter->tx_timeout_count++;
3116 ixgbe_reinit_locked(adapter);
3119 #ifdef CONFIG_IXGBE_DCB
3120 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3123 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3125 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3129 adapter->num_rx_queues = f->indices;
3130 adapter->num_tx_queues = f->indices;
3138 * ixgbe_set_rss_queues: Allocate queues for RSS
3139 * @adapter: board private structure to initialize
3141 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3142 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3145 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3148 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3150 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3152 adapter->num_rx_queues = f->indices;
3153 adapter->num_tx_queues = f->indices;
3163 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3164 * @adapter: board private structure to initialize
3166 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3167 * to the original CPU that initiated the Tx session. This runs in addition
3168 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3169 * Rx load across CPUs using RSS.
3172 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3175 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3177 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3180 /* Flow Director must have RSS enabled */
3181 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3182 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3183 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3184 adapter->num_tx_queues = f_fdir->indices;
3185 adapter->num_rx_queues = f_fdir->indices;
3188 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3189 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3196 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3197 * @adapter: board private structure to initialize
3199 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3200 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3201 * rx queues out of the max number of rx queues, instead, it is used as the
3202 * index of the first rx queue used by FCoE.
3205 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3208 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3210 f->indices = min((int)num_online_cpus(), f->indices);
3211 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3212 adapter->num_rx_queues = 1;
3213 adapter->num_tx_queues = 1;
3214 #ifdef CONFIG_IXGBE_DCB
3215 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3216 DPRINTK(PROBE, INFO, "FCoE enabled with DCB \n");
3217 ixgbe_set_dcb_queues(adapter);
3220 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3221 DPRINTK(PROBE, INFO, "FCoE enabled with RSS \n");
3222 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3223 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3224 ixgbe_set_fdir_queues(adapter);
3226 ixgbe_set_rss_queues(adapter);
3228 /* adding FCoE rx rings to the end */
3229 f->mask = adapter->num_rx_queues;
3230 adapter->num_rx_queues += f->indices;
3231 adapter->num_tx_queues += f->indices;
3239 #endif /* IXGBE_FCOE */
3241 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3242 * @adapter: board private structure to initialize
3244 * This is the top level queue allocation routine. The order here is very
3245 * important, starting with the "most" number of features turned on at once,
3246 * and ending with the smallest set of features. This way large combinations
3247 * can be allocated if they're turned on, and smaller combinations are the
3248 * fallthrough conditions.
3251 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
3254 if (ixgbe_set_fcoe_queues(adapter))
3257 #endif /* IXGBE_FCOE */
3258 #ifdef CONFIG_IXGBE_DCB
3259 if (ixgbe_set_dcb_queues(adapter))
3263 if (ixgbe_set_fdir_queues(adapter))
3266 if (ixgbe_set_rss_queues(adapter))
3269 /* fallback to base case */
3270 adapter->num_rx_queues = 1;
3271 adapter->num_tx_queues = 1;
3274 /* Notify the stack of the (possibly) reduced Tx Queue count. */
3275 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
3278 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
3281 int err, vector_threshold;
3283 /* We'll want at least 3 (vector_threshold):
3286 * 3) Other (Link Status Change, etc.)
3287 * 4) TCP Timer (optional)
3289 vector_threshold = MIN_MSIX_COUNT;
3291 /* The more we get, the more we will assign to Tx/Rx Cleanup
3292 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
3293 * Right now, we simply care about how many we'll get; we'll
3294 * set them up later while requesting irq's.
3296 while (vectors >= vector_threshold) {
3297 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
3299 if (!err) /* Success in acquiring all requested vectors. */
3302 vectors = 0; /* Nasty failure, quit now */
3303 else /* err == number of vectors we should try again with */
3307 if (vectors < vector_threshold) {
3308 /* Can't allocate enough MSI-X interrupts? Oh well.
3309 * This just means we'll go with either a single MSI
3310 * vector or fall back to legacy interrupts.
3312 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
3313 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3314 kfree(adapter->msix_entries);
3315 adapter->msix_entries = NULL;
3317 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
3319 * Adjust for only the vectors we'll use, which is minimum
3320 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
3321 * vectors we were allocated.
3323 adapter->num_msix_vectors = min(vectors,
3324 adapter->max_msix_q_vectors + NON_Q_VECTORS);
3329 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
3330 * @adapter: board private structure to initialize
3332 * Cache the descriptor ring offsets for RSS to the assigned rings.
3335 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
3340 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3341 for (i = 0; i < adapter->num_rx_queues; i++)
3342 adapter->rx_ring[i].reg_idx = i;
3343 for (i = 0; i < adapter->num_tx_queues; i++)
3344 adapter->tx_ring[i].reg_idx = i;
3353 #ifdef CONFIG_IXGBE_DCB
3355 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
3356 * @adapter: board private structure to initialize
3358 * Cache the descriptor ring offsets for DCB to the assigned rings.
3361 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
3365 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
3367 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3368 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3369 /* the number of queues is assumed to be symmetric */
3370 for (i = 0; i < dcb_i; i++) {
3371 adapter->rx_ring[i].reg_idx = i << 3;
3372 adapter->tx_ring[i].reg_idx = i << 2;
3375 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
3378 * Tx TC0 starts at: descriptor queue 0
3379 * Tx TC1 starts at: descriptor queue 32
3380 * Tx TC2 starts at: descriptor queue 64
3381 * Tx TC3 starts at: descriptor queue 80
3382 * Tx TC4 starts at: descriptor queue 96
3383 * Tx TC5 starts at: descriptor queue 104
3384 * Tx TC6 starts at: descriptor queue 112
3385 * Tx TC7 starts at: descriptor queue 120
3387 * Rx TC0-TC7 are offset by 16 queues each
3389 for (i = 0; i < 3; i++) {
3390 adapter->tx_ring[i].reg_idx = i << 5;
3391 adapter->rx_ring[i].reg_idx = i << 4;
3393 for ( ; i < 5; i++) {
3394 adapter->tx_ring[i].reg_idx =
3396 adapter->rx_ring[i].reg_idx = i << 4;
3398 for ( ; i < dcb_i; i++) {
3399 adapter->tx_ring[i].reg_idx =
3401 adapter->rx_ring[i].reg_idx = i << 4;
3405 } else if (dcb_i == 4) {
3407 * Tx TC0 starts at: descriptor queue 0
3408 * Tx TC1 starts at: descriptor queue 64
3409 * Tx TC2 starts at: descriptor queue 96
3410 * Tx TC3 starts at: descriptor queue 112
3412 * Rx TC0-TC3 are offset by 32 queues each
3414 adapter->tx_ring[0].reg_idx = 0;
3415 adapter->tx_ring[1].reg_idx = 64;
3416 adapter->tx_ring[2].reg_idx = 96;
3417 adapter->tx_ring[3].reg_idx = 112;
3418 for (i = 0 ; i < dcb_i; i++)
3419 adapter->rx_ring[i].reg_idx = i << 5;
3437 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
3438 * @adapter: board private structure to initialize
3440 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
3443 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
3448 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3449 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3450 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
3451 for (i = 0; i < adapter->num_rx_queues; i++)
3452 adapter->rx_ring[i].reg_idx = i;
3453 for (i = 0; i < adapter->num_tx_queues; i++)
3454 adapter->tx_ring[i].reg_idx = i;
3463 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
3464 * @adapter: board private structure to initialize
3466 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
3469 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
3471 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
3473 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3475 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3476 #ifdef CONFIG_IXGBE_DCB
3477 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3478 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
3480 ixgbe_cache_ring_dcb(adapter);
3481 /* find out queues in TC for FCoE */
3482 fcoe_rx_i = adapter->rx_ring[fcoe->tc].reg_idx + 1;
3483 fcoe_tx_i = adapter->tx_ring[fcoe->tc].reg_idx + 1;
3485 * In 82599, the number of Tx queues for each traffic
3486 * class for both 8-TC and 4-TC modes are:
3487 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
3488 * 8 TCs: 32 32 16 16 8 8 8 8
3489 * 4 TCs: 64 64 32 32
3490 * We have max 8 queues for FCoE, where 8 the is
3491 * FCoE redirection table size. If TC for FCoE is
3492 * less than or equal to TC3, we have enough queues
3493 * to add max of 8 queues for FCoE, so we start FCoE
3494 * tx descriptor from the next one, i.e., reg_idx + 1.
3495 * If TC for FCoE is above TC3, implying 8 TC mode,
3496 * and we need 8 for FCoE, we have to take all queues
3497 * in that traffic class for FCoE.
3499 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
3502 #endif /* CONFIG_IXGBE_DCB */
3503 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3504 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3505 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3506 ixgbe_cache_ring_fdir(adapter);
3508 ixgbe_cache_ring_rss(adapter);
3510 fcoe_rx_i = f->mask;
3511 fcoe_tx_i = f->mask;
3513 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
3514 adapter->rx_ring[f->mask + i].reg_idx = fcoe_rx_i;
3515 adapter->tx_ring[f->mask + i].reg_idx = fcoe_tx_i;
3522 #endif /* IXGBE_FCOE */
3524 * ixgbe_cache_ring_register - Descriptor ring to register mapping
3525 * @adapter: board private structure to initialize
3527 * Once we know the feature-set enabled for the device, we'll cache
3528 * the register offset the descriptor ring is assigned to.
3530 * Note, the order the various feature calls is important. It must start with
3531 * the "most" features enabled at the same time, then trickle down to the
3532 * least amount of features turned on at once.
3534 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
3536 /* start with default case */
3537 adapter->rx_ring[0].reg_idx = 0;
3538 adapter->tx_ring[0].reg_idx = 0;
3541 if (ixgbe_cache_ring_fcoe(adapter))
3544 #endif /* IXGBE_FCOE */
3545 #ifdef CONFIG_IXGBE_DCB
3546 if (ixgbe_cache_ring_dcb(adapter))
3550 if (ixgbe_cache_ring_fdir(adapter))
3553 if (ixgbe_cache_ring_rss(adapter))
3558 * ixgbe_alloc_queues - Allocate memory for all rings
3559 * @adapter: board private structure to initialize
3561 * We allocate one ring per queue at run-time since we don't know the
3562 * number of queues at compile-time. The polling_netdev array is
3563 * intended for Multiqueue, but should work fine with a single queue.
3565 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
3569 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
3570 sizeof(struct ixgbe_ring), GFP_KERNEL);
3571 if (!adapter->tx_ring)
3572 goto err_tx_ring_allocation;
3574 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
3575 sizeof(struct ixgbe_ring), GFP_KERNEL);
3576 if (!adapter->rx_ring)
3577 goto err_rx_ring_allocation;
3579 for (i = 0; i < adapter->num_tx_queues; i++) {
3580 adapter->tx_ring[i].count = adapter->tx_ring_count;
3581 adapter->tx_ring[i].queue_index = i;
3584 for (i = 0; i < adapter->num_rx_queues; i++) {
3585 adapter->rx_ring[i].count = adapter->rx_ring_count;
3586 adapter->rx_ring[i].queue_index = i;
3589 ixgbe_cache_ring_register(adapter);
3593 err_rx_ring_allocation:
3594 kfree(adapter->tx_ring);
3595 err_tx_ring_allocation:
3600 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
3601 * @adapter: board private structure to initialize
3603 * Attempt to configure the interrupts using the best available
3604 * capabilities of the hardware and the kernel.
3606 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
3608 struct ixgbe_hw *hw = &adapter->hw;
3610 int vector, v_budget;
3613 * It's easy to be greedy for MSI-X vectors, but it really
3614 * doesn't do us much good if we have a lot more vectors
3615 * than CPU's. So let's be conservative and only ask for
3616 * (roughly) twice the number of vectors as there are CPU's.
3618 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
3619 (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
3622 * At the same time, hardware can only support a maximum of
3623 * hw.mac->max_msix_vectors vectors. With features
3624 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
3625 * descriptor queues supported by our device. Thus, we cap it off in
3626 * those rare cases where the cpu count also exceeds our vector limit.
3628 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
3630 /* A failure in MSI-X entry allocation isn't fatal, but it does
3631 * mean we disable MSI-X capabilities of the adapter. */
3632 adapter->msix_entries = kcalloc(v_budget,
3633 sizeof(struct msix_entry), GFP_KERNEL);
3634 if (adapter->msix_entries) {
3635 for (vector = 0; vector < v_budget; vector++)
3636 adapter->msix_entries[vector].entry = vector;
3638 ixgbe_acquire_msix_vectors(adapter, v_budget);
3640 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3644 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
3645 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
3646 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3647 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3648 adapter->atr_sample_rate = 0;
3649 ixgbe_set_num_queues(adapter);
3651 err = pci_enable_msi(adapter->pdev);
3653 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
3655 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
3656 "falling back to legacy. Error: %d\n", err);
3666 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
3667 * @adapter: board private structure to initialize
3669 * We allocate one q_vector per queue interrupt. If allocation fails we
3672 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
3674 int q_idx, num_q_vectors;
3675 struct ixgbe_q_vector *q_vector;
3677 int (*poll)(struct napi_struct *, int);
3679 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3680 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3681 napi_vectors = adapter->num_rx_queues;
3682 poll = &ixgbe_clean_rxtx_many;
3689 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3690 q_vector = kzalloc(sizeof(struct ixgbe_q_vector), GFP_KERNEL);
3693 q_vector->adapter = adapter;
3694 if (q_vector->txr_count && !q_vector->rxr_count)
3695 q_vector->eitr = adapter->tx_eitr_param;
3697 q_vector->eitr = adapter->rx_eitr_param;
3698 q_vector->v_idx = q_idx;
3699 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
3700 adapter->q_vector[q_idx] = q_vector;
3708 q_vector = adapter->q_vector[q_idx];
3709 netif_napi_del(&q_vector->napi);
3711 adapter->q_vector[q_idx] = NULL;
3717 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
3718 * @adapter: board private structure to initialize
3720 * This function frees the memory allocated to the q_vectors. In addition if
3721 * NAPI is enabled it will delete any references to the NAPI struct prior
3722 * to freeing the q_vector.
3724 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
3726 int q_idx, num_q_vectors;
3728 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3729 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3733 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3734 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
3735 adapter->q_vector[q_idx] = NULL;
3736 netif_napi_del(&q_vector->napi);
3741 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
3743 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3744 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3745 pci_disable_msix(adapter->pdev);
3746 kfree(adapter->msix_entries);
3747 adapter->msix_entries = NULL;
3748 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
3749 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
3750 pci_disable_msi(adapter->pdev);
3756 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
3757 * @adapter: board private structure to initialize
3759 * We determine which interrupt scheme to use based on...
3760 * - Kernel support (MSI, MSI-X)
3761 * - which can be user-defined (via MODULE_PARAM)
3762 * - Hardware queue count (num_*_queues)
3763 * - defined by miscellaneous hardware support/features (RSS, etc.)
3765 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
3769 /* Number of supported queues */
3770 ixgbe_set_num_queues(adapter);
3772 err = ixgbe_set_interrupt_capability(adapter);
3774 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
3775 goto err_set_interrupt;
3778 err = ixgbe_alloc_q_vectors(adapter);
3780 DPRINTK(PROBE, ERR, "Unable to allocate memory for queue "
3782 goto err_alloc_q_vectors;
3785 err = ixgbe_alloc_queues(adapter);
3787 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
3788 goto err_alloc_queues;
3791 DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
3792 "Tx Queue count = %u\n",
3793 (adapter->num_rx_queues > 1) ? "Enabled" :
3794 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
3796 set_bit(__IXGBE_DOWN, &adapter->state);
3801 ixgbe_free_q_vectors(adapter);
3802 err_alloc_q_vectors:
3803 ixgbe_reset_interrupt_capability(adapter);
3809 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
3810 * @adapter: board private structure to clear interrupt scheme on
3812 * We go through and clear interrupt specific resources and reset the structure
3813 * to pre-load conditions
3815 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
3817 kfree(adapter->tx_ring);
3818 kfree(adapter->rx_ring);
3819 adapter->tx_ring = NULL;
3820 adapter->rx_ring = NULL;
3822 ixgbe_free_q_vectors(adapter);
3823 ixgbe_reset_interrupt_capability(adapter);
3827 * ixgbe_sfp_timer - worker thread to find a missing module
3828 * @data: pointer to our adapter struct
3830 static void ixgbe_sfp_timer(unsigned long data)
3832 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3835 * Do the sfp_timer outside of interrupt context due to the
3836 * delays that sfp+ detection requires
3838 schedule_work(&adapter->sfp_task);
3842 * ixgbe_sfp_task - worker thread to find a missing module
3843 * @work: pointer to work_struct containing our data
3845 static void ixgbe_sfp_task(struct work_struct *work)
3847 struct ixgbe_adapter *adapter = container_of(work,
3848 struct ixgbe_adapter,
3850 struct ixgbe_hw *hw = &adapter->hw;
3852 if ((hw->phy.type == ixgbe_phy_nl) &&
3853 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
3854 s32 ret = hw->phy.ops.identify_sfp(hw);
3855 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
3857 ret = hw->phy.ops.reset(hw);
3858 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3859 dev_err(&adapter->pdev->dev, "failed to initialize "
3860 "because an unsupported SFP+ module type "
3862 "Reload the driver after installing a "
3863 "supported module.\n");
3864 unregister_netdev(adapter->netdev);
3866 DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
3869 /* don't need this routine any more */
3870 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3874 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
3875 mod_timer(&adapter->sfp_timer,
3876 round_jiffies(jiffies + (2 * HZ)));
3880 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
3881 * @adapter: board private structure to initialize
3883 * ixgbe_sw_init initializes the Adapter private data structure.
3884 * Fields are initialized based on PCI device information and
3885 * OS network device settings (MTU size).
3887 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
3889 struct ixgbe_hw *hw = &adapter->hw;
3890 struct pci_dev *pdev = adapter->pdev;
3892 #ifdef CONFIG_IXGBE_DCB
3894 struct tc_configuration *tc;
3897 /* PCI config space info */
3899 hw->vendor_id = pdev->vendor;
3900 hw->device_id = pdev->device;
3901 hw->revision_id = pdev->revision;
3902 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3903 hw->subsystem_device_id = pdev->subsystem_device;
3905 /* Set capability flags */
3906 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
3907 adapter->ring_feature[RING_F_RSS].indices = rss;
3908 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
3909 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
3910 if (hw->mac.type == ixgbe_mac_82598EB) {
3911 if (hw->device_id == IXGBE_DEV_ID_82598AT)
3912 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
3913 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
3914 } else if (hw->mac.type == ixgbe_mac_82599EB) {
3915 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
3916 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
3917 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
3918 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
3919 adapter->ring_feature[RING_F_FDIR].indices =
3920 IXGBE_MAX_FDIR_INDICES;
3921 adapter->atr_sample_rate = 20;
3922 adapter->fdir_pballoc = 0;
3924 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
3925 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
3926 adapter->ring_feature[RING_F_FCOE].indices = 0;
3927 /* Default traffic class to use for FCoE */
3928 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
3929 #endif /* IXGBE_FCOE */
3932 #ifdef CONFIG_IXGBE_DCB
3933 /* Configure DCB traffic classes */
3934 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
3935 tc = &adapter->dcb_cfg.tc_config[j];
3936 tc->path[DCB_TX_CONFIG].bwg_id = 0;
3937 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
3938 tc->path[DCB_RX_CONFIG].bwg_id = 0;
3939 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
3940 tc->dcb_pfc = pfc_disabled;
3942 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
3943 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
3944 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
3945 adapter->dcb_cfg.pfc_mode_enable = false;
3946 adapter->dcb_cfg.round_robin_enable = false;
3947 adapter->dcb_set_bitmap = 0x00;
3948 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
3949 adapter->ring_feature[RING_F_DCB].indices);
3953 /* default flow control settings */
3954 hw->fc.requested_mode = ixgbe_fc_full;
3955 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
3957 adapter->last_lfc_mode = hw->fc.current_mode;
3959 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
3960 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
3961 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
3962 hw->fc.send_xon = true;
3963 hw->fc.disable_fc_autoneg = false;
3965 /* enable itr by default in dynamic mode */
3966 adapter->rx_itr_setting = 1;
3967 adapter->rx_eitr_param = 20000;
3968 adapter->tx_itr_setting = 1;
3969 adapter->tx_eitr_param = 10000;
3971 /* set defaults for eitr in MegaBytes */
3972 adapter->eitr_low = 10;
3973 adapter->eitr_high = 20;
3975 /* set default ring sizes */
3976 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
3977 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
3979 /* initialize eeprom parameters */
3980 if (ixgbe_init_eeprom_params_generic(hw)) {
3981 dev_err(&pdev->dev, "EEPROM initialization failed\n");
3985 /* enable rx csum by default */
3986 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
3988 set_bit(__IXGBE_DOWN, &adapter->state);
3994 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
3995 * @adapter: board private structure
3996 * @tx_ring: tx descriptor ring (for a specific queue) to setup
3998 * Return 0 on success, negative on failure
4000 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4001 struct ixgbe_ring *tx_ring)
4003 struct pci_dev *pdev = adapter->pdev;
4006 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4007 tx_ring->tx_buffer_info = vmalloc(size);
4008 if (!tx_ring->tx_buffer_info)
4010 memset(tx_ring->tx_buffer_info, 0, size);
4012 /* round up to nearest 4K */
4013 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4014 tx_ring->size = ALIGN(tx_ring->size, 4096);
4016 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
4021 tx_ring->next_to_use = 0;
4022 tx_ring->next_to_clean = 0;
4023 tx_ring->work_limit = tx_ring->count;
4027 vfree(tx_ring->tx_buffer_info);
4028 tx_ring->tx_buffer_info = NULL;
4029 DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
4030 "descriptor ring\n");
4035 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4036 * @adapter: board private structure
4038 * If this function returns with an error, then it's possible one or
4039 * more of the rings is populated (while the rest are not). It is the
4040 * callers duty to clean those orphaned rings.
4042 * Return 0 on success, negative on failure
4044 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4048 for (i = 0; i < adapter->num_tx_queues; i++) {
4049 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
4052 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
4060 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4061 * @adapter: board private structure
4062 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4064 * Returns 0 on success, negative on failure
4066 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4067 struct ixgbe_ring *rx_ring)
4069 struct pci_dev *pdev = adapter->pdev;
4072 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4073 rx_ring->rx_buffer_info = vmalloc(size);
4074 if (!rx_ring->rx_buffer_info) {
4076 "vmalloc allocation failed for the rx desc ring\n");
4079 memset(rx_ring->rx_buffer_info, 0, size);
4081 /* Round up to nearest 4K */
4082 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4083 rx_ring->size = ALIGN(rx_ring->size, 4096);
4085 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
4087 if (!rx_ring->desc) {
4089 "Memory allocation failed for the rx desc ring\n");
4090 vfree(rx_ring->rx_buffer_info);
4094 rx_ring->next_to_clean = 0;
4095 rx_ring->next_to_use = 0;
4104 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4105 * @adapter: board private structure
4107 * If this function returns with an error, then it's possible one or
4108 * more of the rings is populated (while the rest are not). It is the
4109 * callers duty to clean those orphaned rings.
4111 * Return 0 on success, negative on failure
4114 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4118 for (i = 0; i < adapter->num_rx_queues; i++) {
4119 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
4122 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
4130 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4131 * @adapter: board private structure
4132 * @tx_ring: Tx descriptor ring for a specific queue
4134 * Free all transmit software resources
4136 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4137 struct ixgbe_ring *tx_ring)
4139 struct pci_dev *pdev = adapter->pdev;
4141 ixgbe_clean_tx_ring(adapter, tx_ring);
4143 vfree(tx_ring->tx_buffer_info);
4144 tx_ring->tx_buffer_info = NULL;
4146 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
4148 tx_ring->desc = NULL;
4152 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4153 * @adapter: board private structure
4155 * Free all transmit software resources
4157 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4161 for (i = 0; i < adapter->num_tx_queues; i++)
4162 if (adapter->tx_ring[i].desc)
4163 ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
4167 * ixgbe_free_rx_resources - Free Rx Resources
4168 * @adapter: board private structure
4169 * @rx_ring: ring to clean the resources from
4171 * Free all receive software resources
4173 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
4174 struct ixgbe_ring *rx_ring)
4176 struct pci_dev *pdev = adapter->pdev;
4178 ixgbe_clean_rx_ring(adapter, rx_ring);
4180 vfree(rx_ring->rx_buffer_info);
4181 rx_ring->rx_buffer_info = NULL;
4183 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
4185 rx_ring->desc = NULL;
4189 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4190 * @adapter: board private structure
4192 * Free all receive software resources
4194 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4198 for (i = 0; i < adapter->num_rx_queues; i++)
4199 if (adapter->rx_ring[i].desc)
4200 ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
4204 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4205 * @netdev: network interface device structure
4206 * @new_mtu: new value for maximum frame size
4208 * Returns 0 on success, negative on failure
4210 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4212 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4213 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4215 /* MTU < 68 is an error and causes problems on some kernels */
4216 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4219 DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
4220 netdev->mtu, new_mtu);
4221 /* must set new MTU before calling down or up */
4222 netdev->mtu = new_mtu;
4224 if (netif_running(netdev))
4225 ixgbe_reinit_locked(adapter);
4231 * ixgbe_open - Called when a network interface is made active
4232 * @netdev: network interface device structure
4234 * Returns 0 on success, negative value on failure
4236 * The open entry point is called when a network interface is made
4237 * active by the system (IFF_UP). At this point all resources needed
4238 * for transmit and receive operations are allocated, the interrupt
4239 * handler is registered with the OS, the watchdog timer is started,
4240 * and the stack is notified that the interface is ready.
4242 static int ixgbe_open(struct net_device *netdev)
4244 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4247 /* disallow open during test */
4248 if (test_bit(__IXGBE_TESTING, &adapter->state))
4251 netif_carrier_off(netdev);
4253 /* allocate transmit descriptors */
4254 err = ixgbe_setup_all_tx_resources(adapter);
4258 /* allocate receive descriptors */
4259 err = ixgbe_setup_all_rx_resources(adapter);
4263 ixgbe_configure(adapter);
4265 err = ixgbe_request_irq(adapter);
4269 err = ixgbe_up_complete(adapter);
4273 netif_tx_start_all_queues(netdev);
4278 ixgbe_release_hw_control(adapter);
4279 ixgbe_free_irq(adapter);
4282 ixgbe_free_all_rx_resources(adapter);
4284 ixgbe_free_all_tx_resources(adapter);
4285 ixgbe_reset(adapter);
4291 * ixgbe_close - Disables a network interface
4292 * @netdev: network interface device structure
4294 * Returns 0, this is not allowed to fail
4296 * The close entry point is called when an interface is de-activated
4297 * by the OS. The hardware is still under the drivers control, but
4298 * needs to be disabled. A global MAC reset is issued to stop the
4299 * hardware, and all transmit and receive resources are freed.
4301 static int ixgbe_close(struct net_device *netdev)
4303 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4305 ixgbe_down(adapter);
4306 ixgbe_free_irq(adapter);
4308 ixgbe_free_all_tx_resources(adapter);
4309 ixgbe_free_all_rx_resources(adapter);
4311 ixgbe_release_hw_control(adapter);
4317 static int ixgbe_resume(struct pci_dev *pdev)
4319 struct net_device *netdev = pci_get_drvdata(pdev);
4320 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4323 pci_set_power_state(pdev, PCI_D0);
4324 pci_restore_state(pdev);
4326 err = pci_enable_device_mem(pdev);
4328 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
4332 pci_set_master(pdev);
4334 pci_wake_from_d3(pdev, false);
4336 err = ixgbe_init_interrupt_scheme(adapter);
4338 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
4343 ixgbe_reset(adapter);
4345 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4347 if (netif_running(netdev)) {
4348 err = ixgbe_open(adapter->netdev);
4353 netif_device_attach(netdev);
4357 #endif /* CONFIG_PM */
4359 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4361 struct net_device *netdev = pci_get_drvdata(pdev);
4362 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4363 struct ixgbe_hw *hw = &adapter->hw;
4365 u32 wufc = adapter->wol;
4370 netif_device_detach(netdev);
4372 if (netif_running(netdev)) {
4373 ixgbe_down(adapter);
4374 ixgbe_free_irq(adapter);
4375 ixgbe_free_all_tx_resources(adapter);
4376 ixgbe_free_all_rx_resources(adapter);
4378 ixgbe_clear_interrupt_scheme(adapter);
4381 retval = pci_save_state(pdev);
4387 ixgbe_set_rx_mode(netdev);
4389 /* turn on all-multi mode if wake on multicast is enabled */
4390 if (wufc & IXGBE_WUFC_MC) {
4391 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4392 fctrl |= IXGBE_FCTRL_MPE;
4393 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4396 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4397 ctrl |= IXGBE_CTRL_GIO_DIS;
4398 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4400 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4402 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4403 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4406 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
4407 pci_wake_from_d3(pdev, true);
4409 pci_wake_from_d3(pdev, false);
4411 *enable_wake = !!wufc;
4413 ixgbe_release_hw_control(adapter);
4415 pci_disable_device(pdev);
4421 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4426 retval = __ixgbe_shutdown(pdev, &wake);
4431 pci_prepare_to_sleep(pdev);
4433 pci_wake_from_d3(pdev, false);
4434 pci_set_power_state(pdev, PCI_D3hot);
4439 #endif /* CONFIG_PM */
4441 static void ixgbe_shutdown(struct pci_dev *pdev)
4445 __ixgbe_shutdown(pdev, &wake);
4447 if (system_state == SYSTEM_POWER_OFF) {
4448 pci_wake_from_d3(pdev, wake);
4449 pci_set_power_state(pdev, PCI_D3hot);
4454 * ixgbe_update_stats - Update the board statistics counters.
4455 * @adapter: board private structure
4457 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4459 struct ixgbe_hw *hw = &adapter->hw;
4461 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4463 if (hw->mac.type == ixgbe_mac_82599EB) {
4465 for (i = 0; i < 16; i++)
4466 adapter->hw_rx_no_dma_resources +=
4467 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4468 for (i = 0; i < adapter->num_rx_queues; i++)
4469 rsc_count += adapter->rx_ring[i].rsc_count;
4470 adapter->rsc_count = rsc_count;
4473 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
4474 for (i = 0; i < 8; i++) {
4475 /* for packet buffers not used, the register should read 0 */
4476 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
4478 adapter->stats.mpc[i] += mpc;
4479 total_mpc += adapter->stats.mpc[i];
4480 if (hw->mac.type == ixgbe_mac_82598EB)
4481 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
4482 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
4483 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
4484 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
4485 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
4486 if (hw->mac.type == ixgbe_mac_82599EB) {
4487 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4488 IXGBE_PXONRXCNT(i));
4489 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4490 IXGBE_PXOFFRXCNT(i));
4491 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4493 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4495 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4498 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
4500 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
4503 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
4504 /* work around hardware counting issue */
4505 adapter->stats.gprc -= missed_rx;
4507 /* 82598 hardware only has a 32 bit counter in the high register */
4508 if (hw->mac.type == ixgbe_mac_82599EB) {
4510 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
4511 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
4512 adapter->stats.gorc += (tmp << 32);
4513 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
4514 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
4515 adapter->stats.gotc += (tmp << 32);
4516 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
4517 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
4518 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
4519 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
4520 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
4521 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
4523 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
4524 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
4525 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
4526 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
4527 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
4528 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
4529 #endif /* IXGBE_FCOE */
4531 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
4532 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
4533 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
4534 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
4535 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
4537 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
4538 adapter->stats.bprc += bprc;
4539 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
4540 if (hw->mac.type == ixgbe_mac_82598EB)
4541 adapter->stats.mprc -= bprc;
4542 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
4543 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
4544 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
4545 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
4546 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
4547 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
4548 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
4549 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
4550 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
4551 adapter->stats.lxontxc += lxon;
4552 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
4553 adapter->stats.lxofftxc += lxoff;
4554 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4555 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
4556 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
4558 * 82598 errata - tx of flow control packets is included in tx counters
4560 xon_off_tot = lxon + lxoff;
4561 adapter->stats.gptc -= xon_off_tot;
4562 adapter->stats.mptc -= xon_off_tot;
4563 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
4564 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4565 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
4566 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
4567 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
4568 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
4569 adapter->stats.ptc64 -= xon_off_tot;
4570 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
4571 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
4572 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
4573 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
4574 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
4575 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
4577 /* Fill out the OS statistics structure */
4578 adapter->net_stats.multicast = adapter->stats.mprc;
4581 adapter->net_stats.rx_errors = adapter->stats.crcerrs +
4582 adapter->stats.rlec;
4583 adapter->net_stats.rx_dropped = 0;
4584 adapter->net_stats.rx_length_errors = adapter->stats.rlec;
4585 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
4586 adapter->net_stats.rx_missed_errors = total_mpc;
4590 * ixgbe_watchdog - Timer Call-back
4591 * @data: pointer to adapter cast into an unsigned long
4593 static void ixgbe_watchdog(unsigned long data)
4595 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4596 struct ixgbe_hw *hw = &adapter->hw;
4601 * Do the watchdog outside of interrupt context due to the lovely
4602 * delays that some of the newer hardware requires
4605 if (test_bit(__IXGBE_DOWN, &adapter->state))
4606 goto watchdog_short_circuit;
4608 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
4610 * for legacy and MSI interrupts don't set any bits
4611 * that are enabled for EIAM, because this operation
4612 * would set *both* EIMS and EICS for any bit in EIAM
4614 IXGBE_WRITE_REG(hw, IXGBE_EICS,
4615 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
4616 goto watchdog_reschedule;
4619 /* get one bit for every active tx/rx interrupt vector */
4620 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
4621 struct ixgbe_q_vector *qv = adapter->q_vector[i];
4622 if (qv->rxr_count || qv->txr_count)
4623 eics |= ((u64)1 << i);
4626 /* Cause software interrupt to ensure rx rings are cleaned */
4627 ixgbe_irq_rearm_queues(adapter, eics);
4629 watchdog_reschedule:
4630 /* Reset the timer */
4631 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
4633 watchdog_short_circuit:
4634 schedule_work(&adapter->watchdog_task);
4638 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
4639 * @work: pointer to work_struct containing our data
4641 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
4643 struct ixgbe_adapter *adapter = container_of(work,
4644 struct ixgbe_adapter,
4645 multispeed_fiber_task);
4646 struct ixgbe_hw *hw = &adapter->hw;
4650 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
4651 autoneg = hw->phy.autoneg_advertised;
4652 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
4653 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
4654 if (hw->mac.ops.setup_link)
4655 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
4656 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4657 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
4661 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
4662 * @work: pointer to work_struct containing our data
4664 static void ixgbe_sfp_config_module_task(struct work_struct *work)
4666 struct ixgbe_adapter *adapter = container_of(work,
4667 struct ixgbe_adapter,
4668 sfp_config_module_task);
4669 struct ixgbe_hw *hw = &adapter->hw;
4672 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
4674 /* Time for electrical oscillations to settle down */
4676 err = hw->phy.ops.identify_sfp(hw);
4678 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4679 dev_err(&adapter->pdev->dev, "failed to initialize because "
4680 "an unsupported SFP+ module type was detected.\n"
4681 "Reload the driver after installing a supported "
4683 unregister_netdev(adapter->netdev);
4686 hw->mac.ops.setup_sfp(hw);
4688 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
4689 /* This will also work for DA Twinax connections */
4690 schedule_work(&adapter->multispeed_fiber_task);
4691 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
4695 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
4696 * @work: pointer to work_struct containing our data
4698 static void ixgbe_fdir_reinit_task(struct work_struct *work)
4700 struct ixgbe_adapter *adapter = container_of(work,
4701 struct ixgbe_adapter,
4703 struct ixgbe_hw *hw = &adapter->hw;
4706 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
4707 for (i = 0; i < adapter->num_tx_queues; i++)
4708 set_bit(__IXGBE_FDIR_INIT_DONE,
4709 &(adapter->tx_ring[i].reinit_state));
4711 DPRINTK(PROBE, ERR, "failed to finish FDIR re-initialization, "
4712 "ignored adding FDIR ATR filters \n");
4714 /* Done FDIR Re-initialization, enable transmits */
4715 netif_tx_start_all_queues(adapter->netdev);
4719 * ixgbe_watchdog_task - worker thread to bring link up
4720 * @work: pointer to work_struct containing our data
4722 static void ixgbe_watchdog_task(struct work_struct *work)
4724 struct ixgbe_adapter *adapter = container_of(work,
4725 struct ixgbe_adapter,
4727 struct net_device *netdev = adapter->netdev;
4728 struct ixgbe_hw *hw = &adapter->hw;
4729 u32 link_speed = adapter->link_speed;
4730 bool link_up = adapter->link_up;
4732 struct ixgbe_ring *tx_ring;
4733 int some_tx_pending = 0;
4735 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
4737 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
4738 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
4741 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4742 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
4743 hw->mac.ops.fc_enable(hw, i);
4745 hw->mac.ops.fc_enable(hw, 0);
4748 hw->mac.ops.fc_enable(hw, 0);
4753 time_after(jiffies, (adapter->link_check_timeout +
4754 IXGBE_TRY_LINK_TIMEOUT))) {
4755 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4756 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
4758 adapter->link_up = link_up;
4759 adapter->link_speed = link_speed;
4763 if (!netif_carrier_ok(netdev)) {
4764 bool flow_rx, flow_tx;
4766 if (hw->mac.type == ixgbe_mac_82599EB) {
4767 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
4768 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
4769 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
4770 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
4772 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4773 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
4774 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
4775 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
4778 printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
4779 "Flow Control: %s\n",
4781 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
4783 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
4784 "1 Gbps" : "unknown speed")),
4785 ((flow_rx && flow_tx) ? "RX/TX" :
4787 (flow_tx ? "TX" : "None"))));
4789 netif_carrier_on(netdev);
4791 /* Force detection of hung controller */
4792 adapter->detect_tx_hung = true;
4795 adapter->link_up = false;
4796 adapter->link_speed = 0;
4797 if (netif_carrier_ok(netdev)) {
4798 printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
4800 netif_carrier_off(netdev);
4804 if (!netif_carrier_ok(netdev)) {
4805 for (i = 0; i < adapter->num_tx_queues; i++) {
4806 tx_ring = &adapter->tx_ring[i];
4807 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
4808 some_tx_pending = 1;
4813 if (some_tx_pending) {
4814 /* We've lost link, so the controller stops DMA,
4815 * but we've got queued Tx work that's never going
4816 * to get done, so reset controller to flush Tx.
4817 * (Do the reset outside of interrupt context).
4819 schedule_work(&adapter->reset_task);
4823 ixgbe_update_stats(adapter);
4824 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
4827 static int ixgbe_tso(struct ixgbe_adapter *adapter,
4828 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
4829 u32 tx_flags, u8 *hdr_len)
4831 struct ixgbe_adv_tx_context_desc *context_desc;
4834 struct ixgbe_tx_buffer *tx_buffer_info;
4835 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
4836 u32 mss_l4len_idx, l4len;
4838 if (skb_is_gso(skb)) {
4839 if (skb_header_cloned(skb)) {
4840 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4844 l4len = tcp_hdrlen(skb);
4847 if (skb->protocol == htons(ETH_P_IP)) {
4848 struct iphdr *iph = ip_hdr(skb);
4851 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4855 adapter->hw_tso_ctxt++;
4856 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
4857 ipv6_hdr(skb)->payload_len = 0;
4858 tcp_hdr(skb)->check =
4859 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4860 &ipv6_hdr(skb)->daddr,
4862 adapter->hw_tso6_ctxt++;
4865 i = tx_ring->next_to_use;
4867 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4868 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4870 /* VLAN MACLEN IPLEN */
4871 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4873 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4874 vlan_macip_lens |= ((skb_network_offset(skb)) <<
4875 IXGBE_ADVTXD_MACLEN_SHIFT);
4876 *hdr_len += skb_network_offset(skb);
4878 (skb_transport_header(skb) - skb_network_header(skb));
4880 (skb_transport_header(skb) - skb_network_header(skb));
4881 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4882 context_desc->seqnum_seed = 0;
4884 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4885 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
4886 IXGBE_ADVTXD_DTYP_CTXT);
4888 if (skb->protocol == htons(ETH_P_IP))
4889 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4890 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
4891 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4895 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
4896 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
4897 /* use index 1 for TSO */
4898 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4899 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4901 tx_buffer_info->time_stamp = jiffies;
4902 tx_buffer_info->next_to_watch = i;
4905 if (i == tx_ring->count)
4907 tx_ring->next_to_use = i;
4914 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
4915 struct ixgbe_ring *tx_ring,
4916 struct sk_buff *skb, u32 tx_flags)
4918 struct ixgbe_adv_tx_context_desc *context_desc;
4920 struct ixgbe_tx_buffer *tx_buffer_info;
4921 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
4923 if (skb->ip_summed == CHECKSUM_PARTIAL ||
4924 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
4925 i = tx_ring->next_to_use;
4926 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4927 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4929 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4931 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4932 vlan_macip_lens |= (skb_network_offset(skb) <<
4933 IXGBE_ADVTXD_MACLEN_SHIFT);
4934 if (skb->ip_summed == CHECKSUM_PARTIAL)
4935 vlan_macip_lens |= (skb_transport_header(skb) -
4936 skb_network_header(skb));
4938 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4939 context_desc->seqnum_seed = 0;
4941 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
4942 IXGBE_ADVTXD_DTYP_CTXT);
4944 if (skb->ip_summed == CHECKSUM_PARTIAL) {
4945 switch (skb->protocol) {
4946 case cpu_to_be16(ETH_P_IP):
4947 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4948 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4950 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4951 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
4953 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
4955 case cpu_to_be16(ETH_P_IPV6):
4956 /* XXX what about other V6 headers?? */
4957 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4959 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4960 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
4962 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
4965 if (unlikely(net_ratelimit())) {
4966 DPRINTK(PROBE, WARNING,
4967 "partial checksum but proto=%x!\n",
4974 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4975 /* use index zero for tx checksum offload */
4976 context_desc->mss_l4len_idx = 0;
4978 tx_buffer_info->time_stamp = jiffies;
4979 tx_buffer_info->next_to_watch = i;
4981 adapter->hw_csum_tx_good++;
4983 if (i == tx_ring->count)
4985 tx_ring->next_to_use = i;
4993 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
4994 struct ixgbe_ring *tx_ring,
4995 struct sk_buff *skb, u32 tx_flags,
4998 struct ixgbe_tx_buffer *tx_buffer_info;
5000 unsigned int total = skb->len;
5001 unsigned int offset = 0, size, count = 0, i;
5002 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5006 i = tx_ring->next_to_use;
5008 if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
5009 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
5013 map = skb_shinfo(skb)->dma_maps;
5015 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5016 /* excluding fcoe_crc_eof for FCoE */
5017 total -= sizeof(struct fcoe_crc_eof);
5019 len = min(skb_headlen(skb), total);
5021 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5022 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5024 tx_buffer_info->length = size;
5025 tx_buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
5026 tx_buffer_info->time_stamp = jiffies;
5027 tx_buffer_info->next_to_watch = i;
5036 if (i == tx_ring->count)
5041 for (f = 0; f < nr_frags; f++) {
5042 struct skb_frag_struct *frag;
5044 frag = &skb_shinfo(skb)->frags[f];
5045 len = min((unsigned int)frag->size, total);
5050 if (i == tx_ring->count)
5053 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5054 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5056 tx_buffer_info->length = size;
5057 tx_buffer_info->dma = map[f] + offset;
5058 tx_buffer_info->time_stamp = jiffies;
5059 tx_buffer_info->next_to_watch = i;
5070 tx_ring->tx_buffer_info[i].skb = skb;
5071 tx_ring->tx_buffer_info[first].next_to_watch = i;
5076 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
5077 struct ixgbe_ring *tx_ring,
5078 int tx_flags, int count, u32 paylen, u8 hdr_len)
5080 union ixgbe_adv_tx_desc *tx_desc = NULL;
5081 struct ixgbe_tx_buffer *tx_buffer_info;
5082 u32 olinfo_status = 0, cmd_type_len = 0;
5084 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
5086 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
5088 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
5090 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5091 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
5093 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
5094 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5096 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5097 IXGBE_ADVTXD_POPTS_SHIFT;
5099 /* use index 1 context for tso */
5100 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5101 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5102 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
5103 IXGBE_ADVTXD_POPTS_SHIFT;
5105 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5106 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5107 IXGBE_ADVTXD_POPTS_SHIFT;
5109 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5110 olinfo_status |= IXGBE_ADVTXD_CC;
5111 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5112 if (tx_flags & IXGBE_TX_FLAGS_FSO)
5113 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5116 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
5118 i = tx_ring->next_to_use;
5120 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5121 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
5122 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
5123 tx_desc->read.cmd_type_len =
5124 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
5125 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5127 if (i == tx_ring->count)
5131 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
5134 * Force memory writes to complete before letting h/w
5135 * know there are new descriptors to fetch. (Only
5136 * applicable for weak-ordered memory model archs,
5141 tx_ring->next_to_use = i;
5142 writel(i, adapter->hw.hw_addr + tx_ring->tail);
5145 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5146 int queue, u32 tx_flags)
5148 /* Right now, we support IPv4 only */
5149 struct ixgbe_atr_input atr_input;
5151 struct iphdr *iph = ip_hdr(skb);
5152 struct ethhdr *eth = (struct ethhdr *)skb->data;
5153 u16 vlan_id, src_port, dst_port, flex_bytes;
5154 u32 src_ipv4_addr, dst_ipv4_addr;
5157 /* check if we're UDP or TCP */
5158 if (iph->protocol == IPPROTO_TCP) {
5160 src_port = th->source;
5161 dst_port = th->dest;
5162 l4type |= IXGBE_ATR_L4TYPE_TCP;
5163 /* l4type IPv4 type is 0, no need to assign */
5165 /* Unsupported L4 header, just bail here */
5169 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
5171 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
5172 IXGBE_TX_FLAGS_VLAN_SHIFT;
5173 src_ipv4_addr = iph->saddr;
5174 dst_ipv4_addr = iph->daddr;
5175 flex_bytes = eth->h_proto;
5177 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
5178 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
5179 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
5180 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
5181 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
5182 /* src and dst are inverted, think how the receiver sees them */
5183 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
5184 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
5186 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
5187 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
5190 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
5191 struct ixgbe_ring *tx_ring, int size)
5193 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5195 netif_stop_subqueue(netdev, tx_ring->queue_index);
5196 /* Herbert's original patch had:
5197 * smp_mb__after_netif_stop_queue();
5198 * but since that doesn't exist yet, just open code it. */
5201 /* We need to check again in a case another CPU has just
5202 * made room available. */
5203 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
5206 /* A reprieve! - use start_queue because it doesn't call schedule */
5207 netif_start_subqueue(netdev, tx_ring->queue_index);
5208 ++adapter->restart_queue;
5212 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
5213 struct ixgbe_ring *tx_ring, int size)
5215 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
5217 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
5220 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
5222 struct ixgbe_adapter *adapter = netdev_priv(dev);
5224 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
5225 return smp_processor_id();
5227 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5228 return (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK) >> 13;
5230 return skb_tx_hash(dev, skb);
5233 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
5234 struct net_device *netdev)
5236 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5237 struct ixgbe_ring *tx_ring;
5239 unsigned int tx_flags = 0;
5245 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
5246 tx_flags |= vlan_tx_tag_get(skb);
5247 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5248 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
5249 tx_flags |= (skb->queue_mapping << 13);
5251 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5252 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5253 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5254 if (skb->priority != TC_PRIO_CONTROL) {
5255 tx_flags |= (skb->queue_mapping << 13);
5256 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5257 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5259 skb->queue_mapping =
5260 adapter->ring_feature[RING_F_DCB].indices-1;
5264 r_idx = skb->queue_mapping;
5265 tx_ring = &adapter->tx_ring[r_idx];
5267 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5268 (skb->protocol == htons(ETH_P_FCOE))) {
5269 tx_flags |= IXGBE_TX_FLAGS_FCOE;
5271 r_idx = smp_processor_id();
5272 r_idx &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
5273 r_idx += adapter->ring_feature[RING_F_FCOE].mask;
5274 tx_ring = &adapter->tx_ring[r_idx];
5277 /* four things can cause us to need a context descriptor */
5278 if (skb_is_gso(skb) ||
5279 (skb->ip_summed == CHECKSUM_PARTIAL) ||
5280 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
5281 (tx_flags & IXGBE_TX_FLAGS_FCOE))
5284 count += TXD_USE_COUNT(skb_headlen(skb));
5285 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5286 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5288 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
5290 return NETDEV_TX_BUSY;
5293 first = tx_ring->next_to_use;
5294 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5296 /* setup tx offload for FCoE */
5297 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5299 dev_kfree_skb_any(skb);
5300 return NETDEV_TX_OK;
5303 tx_flags |= IXGBE_TX_FLAGS_FSO;
5304 #endif /* IXGBE_FCOE */
5306 if (skb->protocol == htons(ETH_P_IP))
5307 tx_flags |= IXGBE_TX_FLAGS_IPV4;
5308 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5310 dev_kfree_skb_any(skb);
5311 return NETDEV_TX_OK;
5315 tx_flags |= IXGBE_TX_FLAGS_TSO;
5316 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
5317 (skb->ip_summed == CHECKSUM_PARTIAL))
5318 tx_flags |= IXGBE_TX_FLAGS_CSUM;
5321 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
5323 /* add the ATR filter if ATR is on */
5324 if (tx_ring->atr_sample_rate) {
5325 ++tx_ring->atr_count;
5326 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
5327 test_bit(__IXGBE_FDIR_INIT_DONE,
5328 &tx_ring->reinit_state)) {
5329 ixgbe_atr(adapter, skb, tx_ring->queue_index,
5331 tx_ring->atr_count = 0;
5334 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
5336 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
5339 dev_kfree_skb_any(skb);
5340 tx_ring->tx_buffer_info[first].time_stamp = 0;
5341 tx_ring->next_to_use = first;
5344 return NETDEV_TX_OK;
5348 * ixgbe_get_stats - Get System Network Statistics
5349 * @netdev: network interface device structure
5351 * Returns the address of the device statistics structure.
5352 * The statistics are actually updated from the timer callback.
5354 static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
5356 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5358 /* only return the current stats */
5359 return &adapter->net_stats;
5363 * ixgbe_set_mac - Change the Ethernet Address of the NIC
5364 * @netdev: network interface device structure
5365 * @p: pointer to an address structure
5367 * Returns 0 on success, negative on failure
5369 static int ixgbe_set_mac(struct net_device *netdev, void *p)
5371 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5372 struct ixgbe_hw *hw = &adapter->hw;
5373 struct sockaddr *addr = p;
5375 if (!is_valid_ether_addr(addr->sa_data))
5376 return -EADDRNOTAVAIL;
5378 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
5379 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5381 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
5387 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
5389 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5390 struct ixgbe_hw *hw = &adapter->hw;
5394 if (prtad != hw->phy.mdio.prtad)
5396 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
5402 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
5403 u16 addr, u16 value)
5405 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5406 struct ixgbe_hw *hw = &adapter->hw;
5408 if (prtad != hw->phy.mdio.prtad)
5410 return hw->phy.ops.write_reg(hw, addr, devad, value);
5413 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
5415 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5417 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
5421 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
5423 * @netdev: network interface device structure
5425 * Returns non-zero on failure
5427 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
5430 struct ixgbe_adapter *adapter = netdev_priv(dev);
5431 struct ixgbe_mac_info *mac = &adapter->hw.mac;
5433 if (is_valid_ether_addr(mac->san_addr)) {
5435 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5442 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
5444 * @netdev: network interface device structure
5446 * Returns non-zero on failure
5448 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
5451 struct ixgbe_adapter *adapter = netdev_priv(dev);
5452 struct ixgbe_mac_info *mac = &adapter->hw.mac;
5454 if (is_valid_ether_addr(mac->san_addr)) {
5456 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5462 #ifdef CONFIG_NET_POLL_CONTROLLER
5464 * Polling 'interrupt' - used by things like netconsole to send skbs
5465 * without having to re-enable interrupts. It's not called while
5466 * the interrupt routine is executing.
5468 static void ixgbe_netpoll(struct net_device *netdev)
5470 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5473 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
5474 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5475 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5476 for (i = 0; i < num_q_vectors; i++) {
5477 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
5478 ixgbe_msix_clean_many(0, q_vector);
5481 ixgbe_intr(adapter->pdev->irq, netdev);
5483 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
5487 static const struct net_device_ops ixgbe_netdev_ops = {
5488 .ndo_open = ixgbe_open,
5489 .ndo_stop = ixgbe_close,
5490 .ndo_start_xmit = ixgbe_xmit_frame,
5491 .ndo_select_queue = ixgbe_select_queue,
5492 .ndo_get_stats = ixgbe_get_stats,
5493 .ndo_set_rx_mode = ixgbe_set_rx_mode,
5494 .ndo_set_multicast_list = ixgbe_set_rx_mode,
5495 .ndo_validate_addr = eth_validate_addr,
5496 .ndo_set_mac_address = ixgbe_set_mac,
5497 .ndo_change_mtu = ixgbe_change_mtu,
5498 .ndo_tx_timeout = ixgbe_tx_timeout,
5499 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
5500 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
5501 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
5502 .ndo_do_ioctl = ixgbe_ioctl,
5503 #ifdef CONFIG_NET_POLL_CONTROLLER
5504 .ndo_poll_controller = ixgbe_netpoll,
5507 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
5508 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
5509 .ndo_fcoe_enable = ixgbe_fcoe_enable,
5510 .ndo_fcoe_disable = ixgbe_fcoe_disable,
5511 #endif /* IXGBE_FCOE */
5515 * ixgbe_probe - Device Initialization Routine
5516 * @pdev: PCI device information struct
5517 * @ent: entry in ixgbe_pci_tbl
5519 * Returns 0 on success, negative on failure
5521 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
5522 * The OS initialization, configuring of the adapter private structure,
5523 * and a hardware reset occur.
5525 static int __devinit ixgbe_probe(struct pci_dev *pdev,
5526 const struct pci_device_id *ent)
5528 struct net_device *netdev;
5529 struct ixgbe_adapter *adapter = NULL;
5530 struct ixgbe_hw *hw;
5531 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
5532 static int cards_found;
5533 int i, err, pci_using_dac;
5539 err = pci_enable_device_mem(pdev);
5543 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
5544 !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
5547 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5549 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
5551 dev_err(&pdev->dev, "No usable DMA "
5552 "configuration, aborting\n");
5559 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
5560 IORESOURCE_MEM), ixgbe_driver_name);
5563 "pci_request_selected_regions failed 0x%x\n", err);
5567 pci_enable_pcie_error_reporting(pdev);
5569 pci_set_master(pdev);
5570 pci_save_state(pdev);
5572 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
5575 goto err_alloc_etherdev;
5578 SET_NETDEV_DEV(netdev, &pdev->dev);
5580 pci_set_drvdata(pdev, netdev);
5581 adapter = netdev_priv(netdev);
5583 adapter->netdev = netdev;
5584 adapter->pdev = pdev;
5587 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
5589 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
5590 pci_resource_len(pdev, 0));
5596 for (i = 1; i <= 5; i++) {
5597 if (pci_resource_len(pdev, i) == 0)
5601 netdev->netdev_ops = &ixgbe_netdev_ops;
5602 ixgbe_set_ethtool_ops(netdev);
5603 netdev->watchdog_timeo = 5 * HZ;
5604 strcpy(netdev->name, pci_name(pdev));
5606 adapter->bd_number = cards_found;
5609 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
5610 hw->mac.type = ii->mac;
5613 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
5614 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
5615 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
5616 if (!(eec & (1 << 8)))
5617 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
5620 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
5621 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
5622 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
5623 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
5624 hw->phy.mdio.mmds = 0;
5625 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
5626 hw->phy.mdio.dev = netdev;
5627 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
5628 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
5630 /* set up this timer and work struct before calling get_invariants
5631 * which might start the timer
5633 init_timer(&adapter->sfp_timer);
5634 adapter->sfp_timer.function = &ixgbe_sfp_timer;
5635 adapter->sfp_timer.data = (unsigned long) adapter;
5637 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
5639 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
5640 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
5642 /* a new SFP+ module arrival, called from GPI SDP2 context */
5643 INIT_WORK(&adapter->sfp_config_module_task,
5644 ixgbe_sfp_config_module_task);
5646 ii->get_invariants(hw);
5648 /* setup the private structure */
5649 err = ixgbe_sw_init(adapter);
5654 * If there is a fan on this device and it has failed log the
5657 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5658 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5659 if (esdp & IXGBE_ESDP_SDP1)
5660 DPRINTK(PROBE, CRIT,
5661 "Fan has stopped, replace the adapter\n");
5664 /* reset_hw fills in the perm_addr as well */
5665 err = hw->mac.ops.reset_hw(hw);
5666 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
5667 hw->mac.type == ixgbe_mac_82598EB) {
5669 * Start a kernel thread to watch for a module to arrive.
5670 * Only do this for 82598, since 82599 will generate
5671 * interrupts on module arrival.
5673 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5674 mod_timer(&adapter->sfp_timer,
5675 round_jiffies(jiffies + (2 * HZ)));
5677 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5678 dev_err(&adapter->pdev->dev, "failed to initialize because "
5679 "an unsupported SFP+ module type was detected.\n"
5680 "Reload the driver after installing a supported "
5684 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
5688 netdev->features = NETIF_F_SG |
5690 NETIF_F_HW_VLAN_TX |
5691 NETIF_F_HW_VLAN_RX |
5692 NETIF_F_HW_VLAN_FILTER;
5694 netdev->features |= NETIF_F_IPV6_CSUM;
5695 netdev->features |= NETIF_F_TSO;
5696 netdev->features |= NETIF_F_TSO6;
5697 netdev->features |= NETIF_F_GRO;
5699 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
5700 netdev->features |= NETIF_F_SCTP_CSUM;
5702 netdev->vlan_features |= NETIF_F_TSO;
5703 netdev->vlan_features |= NETIF_F_TSO6;
5704 netdev->vlan_features |= NETIF_F_IP_CSUM;
5705 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
5706 netdev->vlan_features |= NETIF_F_SG;
5708 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5709 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
5711 #ifdef CONFIG_IXGBE_DCB
5712 netdev->dcbnl_ops = &dcbnl_ops;
5716 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
5717 if (hw->mac.ops.get_device_caps) {
5718 hw->mac.ops.get_device_caps(hw, &device_caps);
5719 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
5720 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5723 #endif /* IXGBE_FCOE */
5725 netdev->features |= NETIF_F_HIGHDMA;
5727 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
5728 netdev->features |= NETIF_F_LRO;
5730 /* make sure the EEPROM is good */
5731 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
5732 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
5737 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
5738 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
5740 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
5741 dev_err(&pdev->dev, "invalid MAC address\n");
5746 init_timer(&adapter->watchdog_timer);
5747 adapter->watchdog_timer.function = &ixgbe_watchdog;
5748 adapter->watchdog_timer.data = (unsigned long)adapter;
5750 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
5751 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
5753 err = ixgbe_init_interrupt_scheme(adapter);
5757 switch (pdev->device) {
5758 case IXGBE_DEV_ID_82599_KX4:
5759 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
5760 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
5761 /* Enable ACPI wakeup in GRC */
5762 IXGBE_WRITE_REG(hw, IXGBE_GRC,
5763 (IXGBE_READ_REG(hw, IXGBE_GRC) & ~IXGBE_GRC_APME));
5769 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5771 /* pick up the PCI bus settings for reporting later */
5772 hw->mac.ops.get_bus_info(hw);
5774 /* print bus type/speed/width info */
5775 dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
5776 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
5777 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
5778 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
5779 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
5780 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
5783 ixgbe_read_pba_num_generic(hw, &part_num);
5784 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
5785 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
5786 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
5787 (part_num >> 8), (part_num & 0xff));
5789 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5790 hw->mac.type, hw->phy.type,
5791 (part_num >> 8), (part_num & 0xff));
5793 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
5794 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
5795 "this card is not sufficient for optimal "
5797 dev_warn(&pdev->dev, "For optimal performance a x8 "
5798 "PCI-Express slot is required.\n");
5801 /* save off EEPROM version number */
5802 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
5804 /* reset the hardware with the new settings */
5805 err = hw->mac.ops.start_hw(hw);
5807 if (err == IXGBE_ERR_EEPROM_VERSION) {
5808 /* We are running on a pre-production device, log a warning */
5809 dev_warn(&pdev->dev, "This device is a pre-production "
5810 "adapter/LOM. Please be aware there may be issues "
5811 "associated with your hardware. If you are "
5812 "experiencing problems please contact your Intel or "
5813 "hardware representative who provided you with this "
5816 strcpy(netdev->name, "eth%d");
5817 err = register_netdev(netdev);
5821 /* carrier off reporting is important to ethtool even BEFORE open */
5822 netif_carrier_off(netdev);
5824 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5825 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5826 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
5828 #ifdef CONFIG_IXGBE_DCA
5829 if (dca_add_requester(&pdev->dev) == 0) {
5830 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
5831 ixgbe_setup_dca(adapter);
5834 /* add san mac addr to netdev */
5835 ixgbe_add_sanmac_netdev(netdev);
5837 dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
5842 ixgbe_release_hw_control(adapter);
5843 ixgbe_clear_interrupt_scheme(adapter);
5846 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5847 del_timer_sync(&adapter->sfp_timer);
5848 cancel_work_sync(&adapter->sfp_task);
5849 cancel_work_sync(&adapter->multispeed_fiber_task);
5850 cancel_work_sync(&adapter->sfp_config_module_task);
5851 iounmap(hw->hw_addr);
5853 free_netdev(netdev);
5855 pci_release_selected_regions(pdev, pci_select_bars(pdev,
5859 pci_disable_device(pdev);
5864 * ixgbe_remove - Device Removal Routine
5865 * @pdev: PCI device information struct
5867 * ixgbe_remove is called by the PCI subsystem to alert the driver
5868 * that it should release a PCI device. The could be caused by a
5869 * Hot-Plug event, or because the driver is going to be removed from
5872 static void __devexit ixgbe_remove(struct pci_dev *pdev)
5874 struct net_device *netdev = pci_get_drvdata(pdev);
5875 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5877 set_bit(__IXGBE_DOWN, &adapter->state);
5878 /* clear the module not found bit to make sure the worker won't
5881 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5882 del_timer_sync(&adapter->watchdog_timer);
5884 del_timer_sync(&adapter->sfp_timer);
5885 cancel_work_sync(&adapter->watchdog_task);
5886 cancel_work_sync(&adapter->sfp_task);
5887 cancel_work_sync(&adapter->multispeed_fiber_task);
5888 cancel_work_sync(&adapter->sfp_config_module_task);
5889 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5890 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5891 cancel_work_sync(&adapter->fdir_reinit_task);
5892 flush_scheduled_work();
5894 #ifdef CONFIG_IXGBE_DCA
5895 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
5896 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
5897 dca_remove_requester(&pdev->dev);
5898 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
5903 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
5904 ixgbe_cleanup_fcoe(adapter);
5906 #endif /* IXGBE_FCOE */
5908 /* remove the added san mac */
5909 ixgbe_del_sanmac_netdev(netdev);
5911 if (netdev->reg_state == NETREG_REGISTERED)
5912 unregister_netdev(netdev);
5914 ixgbe_clear_interrupt_scheme(adapter);
5916 ixgbe_release_hw_control(adapter);
5918 iounmap(adapter->hw.hw_addr);
5919 pci_release_selected_regions(pdev, pci_select_bars(pdev,
5922 DPRINTK(PROBE, INFO, "complete\n");
5924 free_netdev(netdev);
5926 pci_disable_pcie_error_reporting(pdev);
5928 pci_disable_device(pdev);
5932 * ixgbe_io_error_detected - called when PCI error is detected
5933 * @pdev: Pointer to PCI device
5934 * @state: The current pci connection state
5936 * This function is called after a PCI bus error affecting
5937 * this device has been detected.
5939 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
5940 pci_channel_state_t state)
5942 struct net_device *netdev = pci_get_drvdata(pdev);
5943 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5945 netif_device_detach(netdev);
5947 if (state == pci_channel_io_perm_failure)
5948 return PCI_ERS_RESULT_DISCONNECT;
5950 if (netif_running(netdev))
5951 ixgbe_down(adapter);
5952 pci_disable_device(pdev);
5954 /* Request a slot reset. */
5955 return PCI_ERS_RESULT_NEED_RESET;
5959 * ixgbe_io_slot_reset - called after the pci bus has been reset.
5960 * @pdev: Pointer to PCI device
5962 * Restart the card from scratch, as if from a cold-boot.
5964 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
5966 struct net_device *netdev = pci_get_drvdata(pdev);
5967 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5968 pci_ers_result_t result;
5971 if (pci_enable_device_mem(pdev)) {
5973 "Cannot re-enable PCI device after reset.\n");
5974 result = PCI_ERS_RESULT_DISCONNECT;
5976 pci_set_master(pdev);
5977 pci_restore_state(pdev);
5979 pci_wake_from_d3(pdev, false);
5981 ixgbe_reset(adapter);
5982 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5983 result = PCI_ERS_RESULT_RECOVERED;
5986 err = pci_cleanup_aer_uncorrect_error_status(pdev);
5989 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
5990 /* non-fatal, continue */
5997 * ixgbe_io_resume - called when traffic can start flowing again.
5998 * @pdev: Pointer to PCI device
6000 * This callback is called when the error recovery driver tells us that
6001 * its OK to resume normal operation.
6003 static void ixgbe_io_resume(struct pci_dev *pdev)
6005 struct net_device *netdev = pci_get_drvdata(pdev);
6006 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6008 if (netif_running(netdev)) {
6009 if (ixgbe_up(adapter)) {
6010 DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
6015 netif_device_attach(netdev);
6018 static struct pci_error_handlers ixgbe_err_handler = {
6019 .error_detected = ixgbe_io_error_detected,
6020 .slot_reset = ixgbe_io_slot_reset,
6021 .resume = ixgbe_io_resume,
6024 static struct pci_driver ixgbe_driver = {
6025 .name = ixgbe_driver_name,
6026 .id_table = ixgbe_pci_tbl,
6027 .probe = ixgbe_probe,
6028 .remove = __devexit_p(ixgbe_remove),
6030 .suspend = ixgbe_suspend,
6031 .resume = ixgbe_resume,
6033 .shutdown = ixgbe_shutdown,
6034 .err_handler = &ixgbe_err_handler
6038 * ixgbe_init_module - Driver Registration Routine
6040 * ixgbe_init_module is the first routine called when the driver is
6041 * loaded. All it does is register with the PCI subsystem.
6043 static int __init ixgbe_init_module(void)
6046 printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
6047 ixgbe_driver_string, ixgbe_driver_version);
6049 printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
6051 #ifdef CONFIG_IXGBE_DCA
6052 dca_register_notify(&dca_notifier);
6055 ret = pci_register_driver(&ixgbe_driver);
6059 module_init(ixgbe_init_module);
6062 * ixgbe_exit_module - Driver Exit Cleanup Routine
6064 * ixgbe_exit_module is called just before the driver is removed
6067 static void __exit ixgbe_exit_module(void)
6069 #ifdef CONFIG_IXGBE_DCA
6070 dca_unregister_notify(&dca_notifier);
6072 pci_unregister_driver(&ixgbe_driver);
6075 #ifdef CONFIG_IXGBE_DCA
6076 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
6081 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
6082 __ixgbe_notify_dca);
6084 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6087 #endif /* CONFIG_IXGBE_DCA */
6090 * ixgbe_get_hw_dev_name - return device name string
6091 * used by hardware layer to print debugging information
6093 char *ixgbe_get_hw_dev_name(struct ixgbe_hw *hw)
6095 struct ixgbe_adapter *adapter = hw->back;
6096 return adapter->netdev->name;
6100 module_exit(ixgbe_exit_module);