1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2010 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 <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <scsi/fc/fc_fcoe.h>
47 #include "ixgbe_common.h"
48 #include "ixgbe_dcb_82599.h"
49 #include "ixgbe_sriov.h"
51 char ixgbe_driver_name[] = "ixgbe";
52 static const char ixgbe_driver_string[] =
53 "Intel(R) 10 Gigabit PCI Express Network Driver";
55 #define DRV_VERSION "2.0.62-k2"
56 const char ixgbe_driver_version[] = DRV_VERSION;
57 static char ixgbe_copyright[] = "Copyright (c) 1999-2010 Intel Corporation.";
59 static const struct ixgbe_info *ixgbe_info_tbl[] = {
60 [board_82598] = &ixgbe_82598_info,
61 [board_82599] = &ixgbe_82599_info,
64 /* ixgbe_pci_tbl - PCI Device ID Table
66 * Wildcard entries (PCI_ANY_ID) should come last
67 * Last entry must be all 0s
69 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
70 * Class, Class Mask, private data (not used) }
72 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
73 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
75 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
77 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
79 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
81 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
83 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
85 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
114 /* required last entry */
117 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
119 #ifdef CONFIG_IXGBE_DCA
120 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
122 static struct notifier_block dca_notifier = {
123 .notifier_call = ixgbe_notify_dca,
129 #ifdef CONFIG_PCI_IOV
130 static unsigned int max_vfs;
131 module_param(max_vfs, uint, 0);
132 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
133 "per physical function");
134 #endif /* CONFIG_PCI_IOV */
136 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
137 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
138 MODULE_LICENSE("GPL");
139 MODULE_VERSION(DRV_VERSION);
141 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
143 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
145 struct ixgbe_hw *hw = &adapter->hw;
150 #ifdef CONFIG_PCI_IOV
151 /* disable iov and allow time for transactions to clear */
152 pci_disable_sriov(adapter->pdev);
155 /* turn off device IOV mode */
156 gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
157 gcr &= ~(IXGBE_GCR_EXT_SRIOV);
158 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
159 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
160 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
161 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
163 /* set default pool back to 0 */
164 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
165 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
166 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
168 /* take a breather then clean up driver data */
171 kfree(adapter->vfinfo);
172 adapter->vfinfo = NULL;
174 adapter->num_vfs = 0;
175 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
178 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
182 /* Let firmware take over control of h/w */
183 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
184 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
185 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
188 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
192 /* Let firmware know the driver has taken over */
193 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
194 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
195 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
199 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
200 * @adapter: pointer to adapter struct
201 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
202 * @queue: queue to map the corresponding interrupt to
203 * @msix_vector: the vector to map to the corresponding queue
206 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
207 u8 queue, u8 msix_vector)
210 struct ixgbe_hw *hw = &adapter->hw;
211 switch (hw->mac.type) {
212 case ixgbe_mac_82598EB:
213 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
216 index = (((direction * 64) + queue) >> 2) & 0x1F;
217 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
218 ivar &= ~(0xFF << (8 * (queue & 0x3)));
219 ivar |= (msix_vector << (8 * (queue & 0x3)));
220 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
222 case ixgbe_mac_82599EB:
223 if (direction == -1) {
225 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
226 index = ((queue & 1) * 8);
227 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
228 ivar &= ~(0xFF << index);
229 ivar |= (msix_vector << index);
230 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
233 /* tx or rx causes */
234 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
235 index = ((16 * (queue & 1)) + (8 * direction));
236 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
237 ivar &= ~(0xFF << index);
238 ivar |= (msix_vector << index);
239 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
247 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
252 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
253 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
254 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
256 mask = (qmask & 0xFFFFFFFF);
257 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
258 mask = (qmask >> 32);
259 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
263 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
264 struct ixgbe_tx_buffer
267 if (tx_buffer_info->dma) {
268 if (tx_buffer_info->mapped_as_page)
269 pci_unmap_page(adapter->pdev,
271 tx_buffer_info->length,
274 pci_unmap_single(adapter->pdev,
276 tx_buffer_info->length,
278 tx_buffer_info->dma = 0;
280 if (tx_buffer_info->skb) {
281 dev_kfree_skb_any(tx_buffer_info->skb);
282 tx_buffer_info->skb = NULL;
284 tx_buffer_info->time_stamp = 0;
285 /* tx_buffer_info must be completely set up in the transmit path */
289 * ixgbe_tx_is_paused - check if the tx ring is paused
290 * @adapter: the ixgbe adapter
291 * @tx_ring: the corresponding tx_ring
293 * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
294 * corresponding TC of this tx_ring when checking TFCS.
296 * Returns : true if paused
298 static inline bool ixgbe_tx_is_paused(struct ixgbe_adapter *adapter,
299 struct ixgbe_ring *tx_ring)
301 u32 txoff = IXGBE_TFCS_TXOFF;
303 #ifdef CONFIG_IXGBE_DCB
304 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
306 int reg_idx = tx_ring->reg_idx;
307 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
309 switch (adapter->hw.mac.type) {
310 case ixgbe_mac_82598EB:
312 txoff = IXGBE_TFCS_TXOFF0;
314 case ixgbe_mac_82599EB:
316 txoff = IXGBE_TFCS_TXOFF;
320 if (tc == 2) /* TC2, TC3 */
321 tc += (reg_idx - 64) >> 4;
322 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
323 tc += 1 + ((reg_idx - 96) >> 3);
324 } else if (dcb_i == 4) {
328 tc += (reg_idx - 64) >> 5;
329 if (tc == 2) /* TC2, TC3 */
330 tc += (reg_idx - 96) >> 4;
340 return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
343 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
344 struct ixgbe_ring *tx_ring,
347 struct ixgbe_hw *hw = &adapter->hw;
349 /* Detect a transmit hang in hardware, this serializes the
350 * check with the clearing of time_stamp and movement of eop */
351 adapter->detect_tx_hung = false;
352 if (tx_ring->tx_buffer_info[eop].time_stamp &&
353 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
354 !ixgbe_tx_is_paused(adapter, tx_ring)) {
355 /* detected Tx unit hang */
356 union ixgbe_adv_tx_desc *tx_desc;
357 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
358 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
360 " TDH, TDT <%x>, <%x>\n"
361 " next_to_use <%x>\n"
362 " next_to_clean <%x>\n"
363 "tx_buffer_info[next_to_clean]\n"
364 " time_stamp <%lx>\n"
366 tx_ring->queue_index,
367 IXGBE_READ_REG(hw, tx_ring->head),
368 IXGBE_READ_REG(hw, tx_ring->tail),
369 tx_ring->next_to_use, eop,
370 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
377 #define IXGBE_MAX_TXD_PWR 14
378 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
380 /* Tx Descriptors needed, worst case */
381 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
382 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
383 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
384 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
386 static void ixgbe_tx_timeout(struct net_device *netdev);
389 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
390 * @q_vector: structure containing interrupt and ring information
391 * @tx_ring: tx ring to clean
393 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
394 struct ixgbe_ring *tx_ring)
396 struct ixgbe_adapter *adapter = q_vector->adapter;
397 struct net_device *netdev = adapter->netdev;
398 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
399 struct ixgbe_tx_buffer *tx_buffer_info;
400 unsigned int i, eop, count = 0;
401 unsigned int total_bytes = 0, total_packets = 0;
403 i = tx_ring->next_to_clean;
404 eop = tx_ring->tx_buffer_info[i].next_to_watch;
405 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
407 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
408 (count < tx_ring->work_limit)) {
409 bool cleaned = false;
410 for ( ; !cleaned; count++) {
412 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
413 tx_buffer_info = &tx_ring->tx_buffer_info[i];
414 cleaned = (i == eop);
415 skb = tx_buffer_info->skb;
417 if (cleaned && skb) {
418 unsigned int segs, bytecount;
419 unsigned int hlen = skb_headlen(skb);
421 /* gso_segs is currently only valid for tcp */
422 segs = skb_shinfo(skb)->gso_segs ?: 1;
424 /* adjust for FCoE Sequence Offload */
425 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
426 && (skb->protocol == htons(ETH_P_FCOE)) &&
428 hlen = skb_transport_offset(skb) +
429 sizeof(struct fc_frame_header) +
430 sizeof(struct fcoe_crc_eof);
431 segs = DIV_ROUND_UP(skb->len - hlen,
432 skb_shinfo(skb)->gso_size);
434 #endif /* IXGBE_FCOE */
435 /* multiply data chunks by size of headers */
436 bytecount = ((segs - 1) * hlen) + skb->len;
437 total_packets += segs;
438 total_bytes += bytecount;
441 ixgbe_unmap_and_free_tx_resource(adapter,
444 tx_desc->wb.status = 0;
447 if (i == tx_ring->count)
451 eop = tx_ring->tx_buffer_info[i].next_to_watch;
452 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
455 tx_ring->next_to_clean = i;
457 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
458 if (unlikely(count && netif_carrier_ok(netdev) &&
459 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
460 /* Make sure that anybody stopping the queue after this
461 * sees the new next_to_clean.
464 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
465 !test_bit(__IXGBE_DOWN, &adapter->state)) {
466 netif_wake_subqueue(netdev, tx_ring->queue_index);
467 ++tx_ring->restart_queue;
471 if (adapter->detect_tx_hung) {
472 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
473 /* schedule immediate reset if we believe we hung */
475 "tx hang %d detected, resetting adapter\n",
476 adapter->tx_timeout_count + 1);
477 ixgbe_tx_timeout(adapter->netdev);
481 /* re-arm the interrupt */
482 if (count >= tx_ring->work_limit)
483 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
485 tx_ring->total_bytes += total_bytes;
486 tx_ring->total_packets += total_packets;
487 tx_ring->stats.packets += total_packets;
488 tx_ring->stats.bytes += total_bytes;
489 return (count < tx_ring->work_limit);
492 #ifdef CONFIG_IXGBE_DCA
493 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
494 struct ixgbe_ring *rx_ring)
498 int q = rx_ring->reg_idx;
500 if (rx_ring->cpu != cpu) {
501 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
502 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
503 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
504 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
505 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
506 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
507 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
508 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
510 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
511 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
512 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
513 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
514 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
515 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
521 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
522 struct ixgbe_ring *tx_ring)
526 int q = tx_ring->reg_idx;
527 struct ixgbe_hw *hw = &adapter->hw;
529 if (tx_ring->cpu != cpu) {
530 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
531 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(q));
532 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
533 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
534 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
535 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(q), txctrl);
536 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
537 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(q));
538 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
539 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
540 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
541 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
542 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(q), txctrl);
549 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
553 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
556 /* always use CB2 mode, difference is masked in the CB driver */
557 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
559 for (i = 0; i < adapter->num_tx_queues; i++) {
560 adapter->tx_ring[i]->cpu = -1;
561 ixgbe_update_tx_dca(adapter, adapter->tx_ring[i]);
563 for (i = 0; i < adapter->num_rx_queues; i++) {
564 adapter->rx_ring[i]->cpu = -1;
565 ixgbe_update_rx_dca(adapter, adapter->rx_ring[i]);
569 static int __ixgbe_notify_dca(struct device *dev, void *data)
571 struct net_device *netdev = dev_get_drvdata(dev);
572 struct ixgbe_adapter *adapter = netdev_priv(netdev);
573 unsigned long event = *(unsigned long *)data;
576 case DCA_PROVIDER_ADD:
577 /* if we're already enabled, don't do it again */
578 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
580 if (dca_add_requester(dev) == 0) {
581 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
582 ixgbe_setup_dca(adapter);
585 /* Fall Through since DCA is disabled. */
586 case DCA_PROVIDER_REMOVE:
587 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
588 dca_remove_requester(dev);
589 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
590 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
598 #endif /* CONFIG_IXGBE_DCA */
600 * ixgbe_receive_skb - Send a completed packet up the stack
601 * @adapter: board private structure
602 * @skb: packet to send up
603 * @status: hardware indication of status of receive
604 * @rx_ring: rx descriptor ring (for a specific queue) to setup
605 * @rx_desc: rx descriptor
607 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
608 struct sk_buff *skb, u8 status,
609 struct ixgbe_ring *ring,
610 union ixgbe_adv_rx_desc *rx_desc)
612 struct ixgbe_adapter *adapter = q_vector->adapter;
613 struct napi_struct *napi = &q_vector->napi;
614 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
615 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
617 skb_record_rx_queue(skb, ring->queue_index);
618 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
619 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
620 vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
622 napi_gro_receive(napi, skb);
624 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
625 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
632 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
633 * @adapter: address of board private structure
634 * @status_err: hardware indication of status of receive
635 * @skb: skb currently being received and modified
637 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
638 union ixgbe_adv_rx_desc *rx_desc,
641 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
643 skb->ip_summed = CHECKSUM_NONE;
645 /* Rx csum disabled */
646 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
649 /* if IP and error */
650 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
651 (status_err & IXGBE_RXDADV_ERR_IPE)) {
652 adapter->hw_csum_rx_error++;
656 if (!(status_err & IXGBE_RXD_STAT_L4CS))
659 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
660 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
663 * 82599 errata, UDP frames with a 0 checksum can be marked as
666 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
667 (adapter->hw.mac.type == ixgbe_mac_82599EB))
670 adapter->hw_csum_rx_error++;
674 /* It must be a TCP or UDP packet with a valid checksum */
675 skb->ip_summed = CHECKSUM_UNNECESSARY;
678 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
679 struct ixgbe_ring *rx_ring, u32 val)
682 * Force memory writes to complete before letting h/w
683 * know there are new descriptors to fetch. (Only
684 * applicable for weak-ordered memory model archs,
688 IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
692 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
693 * @adapter: address of board private structure
695 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
696 struct ixgbe_ring *rx_ring,
699 struct pci_dev *pdev = adapter->pdev;
700 union ixgbe_adv_rx_desc *rx_desc;
701 struct ixgbe_rx_buffer *bi;
704 i = rx_ring->next_to_use;
705 bi = &rx_ring->rx_buffer_info[i];
707 while (cleaned_count--) {
708 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
711 (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
713 bi->page = alloc_page(GFP_ATOMIC);
715 adapter->alloc_rx_page_failed++;
720 /* use a half page if we're re-using */
721 bi->page_offset ^= (PAGE_SIZE / 2);
724 bi->page_dma = pci_map_page(pdev, bi->page,
732 /* netdev_alloc_skb reserves 32 bytes up front!! */
733 uint bufsz = rx_ring->rx_buf_len + SMP_CACHE_BYTES;
734 skb = netdev_alloc_skb(adapter->netdev, bufsz);
737 adapter->alloc_rx_buff_failed++;
741 /* advance the data pointer to the next cache line */
742 skb_reserve(skb, (PTR_ALIGN(skb->data, SMP_CACHE_BYTES)
746 bi->dma = pci_map_single(pdev, skb->data,
750 /* Refresh the desc even if buffer_addrs didn't change because
751 * each write-back erases this info. */
752 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
753 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
754 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
756 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
760 if (i == rx_ring->count)
762 bi = &rx_ring->rx_buffer_info[i];
766 if (rx_ring->next_to_use != i) {
767 rx_ring->next_to_use = i;
769 i = (rx_ring->count - 1);
771 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
775 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
777 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
780 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
782 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
785 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
787 return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
788 IXGBE_RXDADV_RSCCNT_MASK) >>
789 IXGBE_RXDADV_RSCCNT_SHIFT;
793 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
794 * @skb: pointer to the last skb in the rsc queue
795 * @count: pointer to number of packets coalesced in this context
797 * This function changes a queue full of hw rsc buffers into a completed
798 * packet. It uses the ->prev pointers to find the first packet and then
799 * turns it into the frag list owner.
801 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
804 unsigned int frag_list_size = 0;
807 struct sk_buff *prev = skb->prev;
808 frag_list_size += skb->len;
814 skb_shinfo(skb)->frag_list = skb->next;
816 skb->len += frag_list_size;
817 skb->data_len += frag_list_size;
818 skb->truesize += frag_list_size;
822 struct ixgbe_rsc_cb {
826 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
828 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
829 struct ixgbe_ring *rx_ring,
830 int *work_done, int work_to_do)
832 struct ixgbe_adapter *adapter = q_vector->adapter;
833 struct net_device *netdev = adapter->netdev;
834 struct pci_dev *pdev = adapter->pdev;
835 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
836 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
838 unsigned int i, rsc_count = 0;
841 bool cleaned = false;
842 int cleaned_count = 0;
843 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
846 #endif /* IXGBE_FCOE */
848 i = rx_ring->next_to_clean;
849 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
850 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
851 rx_buffer_info = &rx_ring->rx_buffer_info[i];
853 while (staterr & IXGBE_RXD_STAT_DD) {
855 if (*work_done >= work_to_do)
859 rmb(); /* read descriptor and rx_buffer_info after status DD */
860 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
861 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
862 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
863 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
864 if (len > IXGBE_RX_HDR_SIZE)
865 len = IXGBE_RX_HDR_SIZE;
866 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
868 len = le16_to_cpu(rx_desc->wb.upper.length);
872 skb = rx_buffer_info->skb;
874 rx_buffer_info->skb = NULL;
876 if (rx_buffer_info->dma) {
877 if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
878 (!(staterr & IXGBE_RXD_STAT_EOP)) &&
881 * When HWRSC is enabled, delay unmapping
882 * of the first packet. It carries the
883 * header information, HW may still
884 * access the header after the writeback.
885 * Only unmap it when EOP is reached
887 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
889 pci_unmap_single(pdev, rx_buffer_info->dma,
892 rx_buffer_info->dma = 0;
897 pci_unmap_page(pdev, rx_buffer_info->page_dma,
898 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
899 rx_buffer_info->page_dma = 0;
900 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
901 rx_buffer_info->page,
902 rx_buffer_info->page_offset,
905 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
906 (page_count(rx_buffer_info->page) != 1))
907 rx_buffer_info->page = NULL;
909 get_page(rx_buffer_info->page);
911 skb->len += upper_len;
912 skb->data_len += upper_len;
913 skb->truesize += upper_len;
917 if (i == rx_ring->count)
920 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
924 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
925 rsc_count = ixgbe_get_rsc_count(rx_desc);
928 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
929 IXGBE_RXDADV_NEXTP_SHIFT;
930 next_buffer = &rx_ring->rx_buffer_info[nextp];
932 next_buffer = &rx_ring->rx_buffer_info[i];
935 if (staterr & IXGBE_RXD_STAT_EOP) {
937 skb = ixgbe_transform_rsc_queue(skb, &(rx_ring->rsc_count));
938 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
939 if (IXGBE_RSC_CB(skb)->dma) {
940 pci_unmap_single(pdev, IXGBE_RSC_CB(skb)->dma,
943 IXGBE_RSC_CB(skb)->dma = 0;
945 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)
946 rx_ring->rsc_count += skb_shinfo(skb)->nr_frags;
948 rx_ring->rsc_count++;
949 rx_ring->rsc_flush++;
951 rx_ring->stats.packets++;
952 rx_ring->stats.bytes += skb->len;
954 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
955 rx_buffer_info->skb = next_buffer->skb;
956 rx_buffer_info->dma = next_buffer->dma;
957 next_buffer->skb = skb;
958 next_buffer->dma = 0;
960 skb->next = next_buffer->skb;
961 skb->next->prev = skb;
963 rx_ring->non_eop_descs++;
967 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
968 dev_kfree_skb_irq(skb);
972 ixgbe_rx_checksum(adapter, rx_desc, skb);
974 /* probably a little skewed due to removing CRC */
975 total_rx_bytes += skb->len;
978 skb->protocol = eth_type_trans(skb, adapter->netdev);
980 /* if ddp, not passing to ULD unless for FCP_RSP or error */
981 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
982 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
986 #endif /* IXGBE_FCOE */
987 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
990 rx_desc->wb.upper.status_error = 0;
992 /* return some buffers to hardware, one at a time is too slow */
993 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
994 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
998 /* use prefetched values */
1000 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1002 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1005 rx_ring->next_to_clean = i;
1006 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1009 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1012 /* include DDPed FCoE data */
1013 if (ddp_bytes > 0) {
1016 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
1017 sizeof(struct fc_frame_header) -
1018 sizeof(struct fcoe_crc_eof);
1021 total_rx_bytes += ddp_bytes;
1022 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1024 #endif /* IXGBE_FCOE */
1026 rx_ring->total_packets += total_rx_packets;
1027 rx_ring->total_bytes += total_rx_bytes;
1028 netdev->stats.rx_bytes += total_rx_bytes;
1029 netdev->stats.rx_packets += total_rx_packets;
1034 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1036 * ixgbe_configure_msix - Configure MSI-X hardware
1037 * @adapter: board private structure
1039 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1042 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1044 struct ixgbe_q_vector *q_vector;
1045 int i, j, q_vectors, v_idx, r_idx;
1048 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1051 * Populate the IVAR table and set the ITR values to the
1052 * corresponding register.
1054 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1055 q_vector = adapter->q_vector[v_idx];
1056 /* XXX for_each_set_bit(...) */
1057 r_idx = find_first_bit(q_vector->rxr_idx,
1058 adapter->num_rx_queues);
1060 for (i = 0; i < q_vector->rxr_count; i++) {
1061 j = adapter->rx_ring[r_idx]->reg_idx;
1062 ixgbe_set_ivar(adapter, 0, j, v_idx);
1063 r_idx = find_next_bit(q_vector->rxr_idx,
1064 adapter->num_rx_queues,
1067 r_idx = find_first_bit(q_vector->txr_idx,
1068 adapter->num_tx_queues);
1070 for (i = 0; i < q_vector->txr_count; i++) {
1071 j = adapter->tx_ring[r_idx]->reg_idx;
1072 ixgbe_set_ivar(adapter, 1, j, v_idx);
1073 r_idx = find_next_bit(q_vector->txr_idx,
1074 adapter->num_tx_queues,
1078 if (q_vector->txr_count && !q_vector->rxr_count)
1080 q_vector->eitr = adapter->tx_eitr_param;
1081 else if (q_vector->rxr_count)
1083 q_vector->eitr = adapter->rx_eitr_param;
1085 ixgbe_write_eitr(q_vector);
1088 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1089 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1091 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1092 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1093 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1095 /* set up to autoclear timer, and the vectors */
1096 mask = IXGBE_EIMS_ENABLE_MASK;
1097 if (adapter->num_vfs)
1098 mask &= ~(IXGBE_EIMS_OTHER |
1099 IXGBE_EIMS_MAILBOX |
1102 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1103 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1106 enum latency_range {
1110 latency_invalid = 255
1114 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1115 * @adapter: pointer to adapter
1116 * @eitr: eitr setting (ints per sec) to give last timeslice
1117 * @itr_setting: current throttle rate in ints/second
1118 * @packets: the number of packets during this measurement interval
1119 * @bytes: the number of bytes during this measurement interval
1121 * Stores a new ITR value based on packets and byte
1122 * counts during the last interrupt. The advantage of per interrupt
1123 * computation is faster updates and more accurate ITR for the current
1124 * traffic pattern. Constants in this function were computed
1125 * based on theoretical maximum wire speed and thresholds were set based
1126 * on testing data as well as attempting to minimize response time
1127 * while increasing bulk throughput.
1128 * this functionality is controlled by the InterruptThrottleRate module
1129 * parameter (see ixgbe_param.c)
1131 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1132 u32 eitr, u8 itr_setting,
1133 int packets, int bytes)
1135 unsigned int retval = itr_setting;
1140 goto update_itr_done;
1143 /* simple throttlerate management
1144 * 0-20MB/s lowest (100000 ints/s)
1145 * 20-100MB/s low (20000 ints/s)
1146 * 100-1249MB/s bulk (8000 ints/s)
1148 /* what was last interrupt timeslice? */
1149 timepassed_us = 1000000/eitr;
1150 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1152 switch (itr_setting) {
1153 case lowest_latency:
1154 if (bytes_perint > adapter->eitr_low)
1155 retval = low_latency;
1158 if (bytes_perint > adapter->eitr_high)
1159 retval = bulk_latency;
1160 else if (bytes_perint <= adapter->eitr_low)
1161 retval = lowest_latency;
1164 if (bytes_perint <= adapter->eitr_high)
1165 retval = low_latency;
1174 * ixgbe_write_eitr - write EITR register in hardware specific way
1175 * @q_vector: structure containing interrupt and ring information
1177 * This function is made to be called by ethtool and by the driver
1178 * when it needs to update EITR registers at runtime. Hardware
1179 * specific quirks/differences are taken care of here.
1181 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1183 struct ixgbe_adapter *adapter = q_vector->adapter;
1184 struct ixgbe_hw *hw = &adapter->hw;
1185 int v_idx = q_vector->v_idx;
1186 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1188 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1189 /* must write high and low 16 bits to reset counter */
1190 itr_reg |= (itr_reg << 16);
1191 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1193 * 82599 can support a value of zero, so allow it for
1194 * max interrupt rate, but there is an errata where it can
1195 * not be zero with RSC
1198 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1202 * set the WDIS bit to not clear the timer bits and cause an
1203 * immediate assertion of the interrupt
1205 itr_reg |= IXGBE_EITR_CNT_WDIS;
1207 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1210 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1212 struct ixgbe_adapter *adapter = q_vector->adapter;
1214 u8 current_itr, ret_itr;
1216 struct ixgbe_ring *rx_ring, *tx_ring;
1218 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1219 for (i = 0; i < q_vector->txr_count; i++) {
1220 tx_ring = adapter->tx_ring[r_idx];
1221 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1223 tx_ring->total_packets,
1224 tx_ring->total_bytes);
1225 /* if the result for this queue would decrease interrupt
1226 * rate for this vector then use that result */
1227 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1228 q_vector->tx_itr - 1 : ret_itr);
1229 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1233 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1234 for (i = 0; i < q_vector->rxr_count; i++) {
1235 rx_ring = adapter->rx_ring[r_idx];
1236 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1238 rx_ring->total_packets,
1239 rx_ring->total_bytes);
1240 /* if the result for this queue would decrease interrupt
1241 * rate for this vector then use that result */
1242 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1243 q_vector->rx_itr - 1 : ret_itr);
1244 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1248 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1250 switch (current_itr) {
1251 /* counts and packets in update_itr are dependent on these numbers */
1252 case lowest_latency:
1256 new_itr = 20000; /* aka hwitr = ~200 */
1264 if (new_itr != q_vector->eitr) {
1265 /* do an exponential smoothing */
1266 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1268 /* save the algorithm value here, not the smoothed one */
1269 q_vector->eitr = new_itr;
1271 ixgbe_write_eitr(q_vector);
1277 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1279 struct ixgbe_hw *hw = &adapter->hw;
1281 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1282 (eicr & IXGBE_EICR_GPI_SDP1)) {
1283 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
1284 /* write to clear the interrupt */
1285 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1289 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1291 struct ixgbe_hw *hw = &adapter->hw;
1293 if (eicr & IXGBE_EICR_GPI_SDP1) {
1294 /* Clear the interrupt */
1295 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1296 schedule_work(&adapter->multispeed_fiber_task);
1297 } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1298 /* Clear the interrupt */
1299 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1300 schedule_work(&adapter->sfp_config_module_task);
1302 /* Interrupt isn't for us... */
1307 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1309 struct ixgbe_hw *hw = &adapter->hw;
1312 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1313 adapter->link_check_timeout = jiffies;
1314 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1315 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1316 IXGBE_WRITE_FLUSH(hw);
1317 schedule_work(&adapter->watchdog_task);
1321 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1323 struct net_device *netdev = data;
1324 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1325 struct ixgbe_hw *hw = &adapter->hw;
1329 * Workaround for Silicon errata. Use clear-by-write instead
1330 * of clear-by-read. Reading with EICS will return the
1331 * interrupt causes without clearing, which later be done
1332 * with the write to EICR.
1334 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1335 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1337 if (eicr & IXGBE_EICR_LSC)
1338 ixgbe_check_lsc(adapter);
1340 if (eicr & IXGBE_EICR_MAILBOX)
1341 ixgbe_msg_task(adapter);
1343 if (hw->mac.type == ixgbe_mac_82598EB)
1344 ixgbe_check_fan_failure(adapter, eicr);
1346 if (hw->mac.type == ixgbe_mac_82599EB) {
1347 ixgbe_check_sfp_event(adapter, eicr);
1349 /* Handle Flow Director Full threshold interrupt */
1350 if (eicr & IXGBE_EICR_FLOW_DIR) {
1352 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1353 /* Disable transmits before FDIR Re-initialization */
1354 netif_tx_stop_all_queues(netdev);
1355 for (i = 0; i < adapter->num_tx_queues; i++) {
1356 struct ixgbe_ring *tx_ring =
1357 adapter->tx_ring[i];
1358 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1359 &tx_ring->reinit_state))
1360 schedule_work(&adapter->fdir_reinit_task);
1364 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1365 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1370 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1375 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1376 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1377 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1379 mask = (qmask & 0xFFFFFFFF);
1380 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1381 mask = (qmask >> 32);
1382 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1384 /* skip the flush */
1387 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1392 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1393 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1394 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1396 mask = (qmask & 0xFFFFFFFF);
1397 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1398 mask = (qmask >> 32);
1399 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1401 /* skip the flush */
1404 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1406 struct ixgbe_q_vector *q_vector = data;
1407 struct ixgbe_adapter *adapter = q_vector->adapter;
1408 struct ixgbe_ring *tx_ring;
1411 if (!q_vector->txr_count)
1414 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1415 for (i = 0; i < q_vector->txr_count; i++) {
1416 tx_ring = adapter->tx_ring[r_idx];
1417 tx_ring->total_bytes = 0;
1418 tx_ring->total_packets = 0;
1419 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1423 /* EIAM disabled interrupts (on this vector) for us */
1424 napi_schedule(&q_vector->napi);
1430 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1432 * @data: pointer to our q_vector struct for this interrupt vector
1434 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1436 struct ixgbe_q_vector *q_vector = data;
1437 struct ixgbe_adapter *adapter = q_vector->adapter;
1438 struct ixgbe_ring *rx_ring;
1442 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1443 for (i = 0; i < q_vector->rxr_count; i++) {
1444 rx_ring = adapter->rx_ring[r_idx];
1445 rx_ring->total_bytes = 0;
1446 rx_ring->total_packets = 0;
1447 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1451 if (!q_vector->rxr_count)
1454 /* disable interrupts on this vector only */
1455 /* EIAM disabled interrupts (on this vector) for us */
1456 napi_schedule(&q_vector->napi);
1461 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1463 struct ixgbe_q_vector *q_vector = data;
1464 struct ixgbe_adapter *adapter = q_vector->adapter;
1465 struct ixgbe_ring *ring;
1469 if (!q_vector->txr_count && !q_vector->rxr_count)
1472 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1473 for (i = 0; i < q_vector->txr_count; i++) {
1474 ring = adapter->tx_ring[r_idx];
1475 ring->total_bytes = 0;
1476 ring->total_packets = 0;
1477 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1481 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1482 for (i = 0; i < q_vector->rxr_count; i++) {
1483 ring = adapter->rx_ring[r_idx];
1484 ring->total_bytes = 0;
1485 ring->total_packets = 0;
1486 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1490 /* EIAM disabled interrupts (on this vector) for us */
1491 napi_schedule(&q_vector->napi);
1497 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1498 * @napi: napi struct with our devices info in it
1499 * @budget: amount of work driver is allowed to do this pass, in packets
1501 * This function is optimized for cleaning one queue only on a single
1504 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1506 struct ixgbe_q_vector *q_vector =
1507 container_of(napi, struct ixgbe_q_vector, napi);
1508 struct ixgbe_adapter *adapter = q_vector->adapter;
1509 struct ixgbe_ring *rx_ring = NULL;
1513 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1514 rx_ring = adapter->rx_ring[r_idx];
1515 #ifdef CONFIG_IXGBE_DCA
1516 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1517 ixgbe_update_rx_dca(adapter, rx_ring);
1520 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1522 /* If all Rx work done, exit the polling mode */
1523 if (work_done < budget) {
1524 napi_complete(napi);
1525 if (adapter->rx_itr_setting & 1)
1526 ixgbe_set_itr_msix(q_vector);
1527 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1528 ixgbe_irq_enable_queues(adapter,
1529 ((u64)1 << q_vector->v_idx));
1536 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1537 * @napi: napi struct with our devices info in it
1538 * @budget: amount of work driver is allowed to do this pass, in packets
1540 * This function will clean more than one rx queue associated with a
1543 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1545 struct ixgbe_q_vector *q_vector =
1546 container_of(napi, struct ixgbe_q_vector, napi);
1547 struct ixgbe_adapter *adapter = q_vector->adapter;
1548 struct ixgbe_ring *ring = NULL;
1549 int work_done = 0, i;
1551 bool tx_clean_complete = true;
1553 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1554 for (i = 0; i < q_vector->txr_count; i++) {
1555 ring = adapter->tx_ring[r_idx];
1556 #ifdef CONFIG_IXGBE_DCA
1557 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1558 ixgbe_update_tx_dca(adapter, ring);
1560 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1561 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1565 /* attempt to distribute budget to each queue fairly, but don't allow
1566 * the budget to go below 1 because we'll exit polling */
1567 budget /= (q_vector->rxr_count ?: 1);
1568 budget = max(budget, 1);
1569 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1570 for (i = 0; i < q_vector->rxr_count; i++) {
1571 ring = adapter->rx_ring[r_idx];
1572 #ifdef CONFIG_IXGBE_DCA
1573 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1574 ixgbe_update_rx_dca(adapter, ring);
1576 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1577 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1581 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1582 ring = adapter->rx_ring[r_idx];
1583 /* If all Rx work done, exit the polling mode */
1584 if (work_done < budget) {
1585 napi_complete(napi);
1586 if (adapter->rx_itr_setting & 1)
1587 ixgbe_set_itr_msix(q_vector);
1588 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1589 ixgbe_irq_enable_queues(adapter,
1590 ((u64)1 << q_vector->v_idx));
1598 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1599 * @napi: napi struct with our devices info in it
1600 * @budget: amount of work driver is allowed to do this pass, in packets
1602 * This function is optimized for cleaning one queue only on a single
1605 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1607 struct ixgbe_q_vector *q_vector =
1608 container_of(napi, struct ixgbe_q_vector, napi);
1609 struct ixgbe_adapter *adapter = q_vector->adapter;
1610 struct ixgbe_ring *tx_ring = NULL;
1614 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1615 tx_ring = adapter->tx_ring[r_idx];
1616 #ifdef CONFIG_IXGBE_DCA
1617 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1618 ixgbe_update_tx_dca(adapter, tx_ring);
1621 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
1624 /* If all Tx work done, exit the polling mode */
1625 if (work_done < budget) {
1626 napi_complete(napi);
1627 if (adapter->tx_itr_setting & 1)
1628 ixgbe_set_itr_msix(q_vector);
1629 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1630 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
1636 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1639 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1641 set_bit(r_idx, q_vector->rxr_idx);
1642 q_vector->rxr_count++;
1645 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1648 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1650 set_bit(t_idx, q_vector->txr_idx);
1651 q_vector->txr_count++;
1655 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1656 * @adapter: board private structure to initialize
1657 * @vectors: allotted vector count for descriptor rings
1659 * This function maps descriptor rings to the queue-specific vectors
1660 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1661 * one vector per ring/queue, but on a constrained vector budget, we
1662 * group the rings as "efficiently" as possible. You would add new
1663 * mapping configurations in here.
1665 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1669 int rxr_idx = 0, txr_idx = 0;
1670 int rxr_remaining = adapter->num_rx_queues;
1671 int txr_remaining = adapter->num_tx_queues;
1676 /* No mapping required if MSI-X is disabled. */
1677 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1681 * The ideal configuration...
1682 * We have enough vectors to map one per queue.
1684 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1685 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1686 map_vector_to_rxq(adapter, v_start, rxr_idx);
1688 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1689 map_vector_to_txq(adapter, v_start, txr_idx);
1695 * If we don't have enough vectors for a 1-to-1
1696 * mapping, we'll have to group them so there are
1697 * multiple queues per vector.
1699 /* Re-adjusting *qpv takes care of the remainder. */
1700 for (i = v_start; i < vectors; i++) {
1701 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1702 for (j = 0; j < rqpv; j++) {
1703 map_vector_to_rxq(adapter, i, rxr_idx);
1708 for (i = v_start; i < vectors; i++) {
1709 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1710 for (j = 0; j < tqpv; j++) {
1711 map_vector_to_txq(adapter, i, txr_idx);
1722 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1723 * @adapter: board private structure
1725 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1726 * interrupts from the kernel.
1728 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1730 struct net_device *netdev = adapter->netdev;
1731 irqreturn_t (*handler)(int, void *);
1732 int i, vector, q_vectors, err;
1735 /* Decrement for Other and TCP Timer vectors */
1736 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1738 /* Map the Tx/Rx rings to the vectors we were allotted. */
1739 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1743 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1744 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1745 &ixgbe_msix_clean_many)
1746 for (vector = 0; vector < q_vectors; vector++) {
1747 handler = SET_HANDLER(adapter->q_vector[vector]);
1749 if(handler == &ixgbe_msix_clean_rx) {
1750 sprintf(adapter->name[vector], "%s-%s-%d",
1751 netdev->name, "rx", ri++);
1753 else if(handler == &ixgbe_msix_clean_tx) {
1754 sprintf(adapter->name[vector], "%s-%s-%d",
1755 netdev->name, "tx", ti++);
1758 sprintf(adapter->name[vector], "%s-%s-%d",
1759 netdev->name, "TxRx", vector);
1761 err = request_irq(adapter->msix_entries[vector].vector,
1762 handler, 0, adapter->name[vector],
1763 adapter->q_vector[vector]);
1766 "request_irq failed for MSIX interrupt "
1767 "Error: %d\n", err);
1768 goto free_queue_irqs;
1772 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1773 err = request_irq(adapter->msix_entries[vector].vector,
1774 ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1777 "request_irq for msix_lsc failed: %d\n", err);
1778 goto free_queue_irqs;
1784 for (i = vector - 1; i >= 0; i--)
1785 free_irq(adapter->msix_entries[--vector].vector,
1786 adapter->q_vector[i]);
1787 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1788 pci_disable_msix(adapter->pdev);
1789 kfree(adapter->msix_entries);
1790 adapter->msix_entries = NULL;
1795 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1797 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1799 u32 new_itr = q_vector->eitr;
1800 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
1801 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
1803 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1805 tx_ring->total_packets,
1806 tx_ring->total_bytes);
1807 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1809 rx_ring->total_packets,
1810 rx_ring->total_bytes);
1812 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1814 switch (current_itr) {
1815 /* counts and packets in update_itr are dependent on these numbers */
1816 case lowest_latency:
1820 new_itr = 20000; /* aka hwitr = ~200 */
1829 if (new_itr != q_vector->eitr) {
1830 /* do an exponential smoothing */
1831 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1833 /* save the algorithm value here, not the smoothed one */
1834 q_vector->eitr = new_itr;
1836 ixgbe_write_eitr(q_vector);
1843 * ixgbe_irq_enable - Enable default interrupt generation settings
1844 * @adapter: board private structure
1846 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1850 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1851 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1852 mask |= IXGBE_EIMS_GPI_SDP1;
1853 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1854 mask |= IXGBE_EIMS_ECC;
1855 mask |= IXGBE_EIMS_GPI_SDP1;
1856 mask |= IXGBE_EIMS_GPI_SDP2;
1857 if (adapter->num_vfs)
1858 mask |= IXGBE_EIMS_MAILBOX;
1860 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
1861 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
1862 mask |= IXGBE_EIMS_FLOW_DIR;
1864 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1865 ixgbe_irq_enable_queues(adapter, ~0);
1866 IXGBE_WRITE_FLUSH(&adapter->hw);
1868 if (adapter->num_vfs > 32) {
1869 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1870 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1875 * ixgbe_intr - legacy mode Interrupt Handler
1876 * @irq: interrupt number
1877 * @data: pointer to a network interface device structure
1879 static irqreturn_t ixgbe_intr(int irq, void *data)
1881 struct net_device *netdev = data;
1882 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1883 struct ixgbe_hw *hw = &adapter->hw;
1884 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1888 * Workaround for silicon errata. Mask the interrupts
1889 * before the read of EICR.
1891 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1893 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1894 * therefore no explict interrupt disable is necessary */
1895 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1897 /* shared interrupt alert!
1898 * make sure interrupts are enabled because the read will
1899 * have disabled interrupts due to EIAM */
1900 ixgbe_irq_enable(adapter);
1901 return IRQ_NONE; /* Not our interrupt */
1904 if (eicr & IXGBE_EICR_LSC)
1905 ixgbe_check_lsc(adapter);
1907 if (hw->mac.type == ixgbe_mac_82599EB)
1908 ixgbe_check_sfp_event(adapter, eicr);
1910 ixgbe_check_fan_failure(adapter, eicr);
1912 if (napi_schedule_prep(&(q_vector->napi))) {
1913 adapter->tx_ring[0]->total_packets = 0;
1914 adapter->tx_ring[0]->total_bytes = 0;
1915 adapter->rx_ring[0]->total_packets = 0;
1916 adapter->rx_ring[0]->total_bytes = 0;
1917 /* would disable interrupts here but EIAM disabled it */
1918 __napi_schedule(&(q_vector->napi));
1924 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1926 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1928 for (i = 0; i < q_vectors; i++) {
1929 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
1930 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1931 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1932 q_vector->rxr_count = 0;
1933 q_vector->txr_count = 0;
1938 * ixgbe_request_irq - initialize interrupts
1939 * @adapter: board private structure
1941 * Attempts to configure interrupts using the best available
1942 * capabilities of the hardware and kernel.
1944 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1946 struct net_device *netdev = adapter->netdev;
1949 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1950 err = ixgbe_request_msix_irqs(adapter);
1951 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1952 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
1953 netdev->name, netdev);
1955 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
1956 netdev->name, netdev);
1960 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1965 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1967 struct net_device *netdev = adapter->netdev;
1969 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1972 q_vectors = adapter->num_msix_vectors;
1975 free_irq(adapter->msix_entries[i].vector, netdev);
1978 for (; i >= 0; i--) {
1979 free_irq(adapter->msix_entries[i].vector,
1980 adapter->q_vector[i]);
1983 ixgbe_reset_q_vectors(adapter);
1985 free_irq(adapter->pdev->irq, netdev);
1990 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1991 * @adapter: board private structure
1993 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1995 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1996 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1998 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
1999 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2000 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2001 if (adapter->num_vfs > 32)
2002 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2004 IXGBE_WRITE_FLUSH(&adapter->hw);
2005 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2007 for (i = 0; i < adapter->num_msix_vectors; i++)
2008 synchronize_irq(adapter->msix_entries[i].vector);
2010 synchronize_irq(adapter->pdev->irq);
2015 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2018 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2020 struct ixgbe_hw *hw = &adapter->hw;
2022 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2023 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2025 ixgbe_set_ivar(adapter, 0, 0, 0);
2026 ixgbe_set_ivar(adapter, 1, 0, 0);
2028 map_vector_to_rxq(adapter, 0, 0);
2029 map_vector_to_txq(adapter, 0, 0);
2031 DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
2035 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2036 * @adapter: board private structure
2038 * Configure the Tx unit of the MAC after a reset.
2040 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2043 struct ixgbe_hw *hw = &adapter->hw;
2044 u32 i, j, tdlen, txctrl;
2046 /* Setup the HW Tx Head and Tail descriptor pointers */
2047 for (i = 0; i < adapter->num_tx_queues; i++) {
2048 struct ixgbe_ring *ring = adapter->tx_ring[i];
2051 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
2052 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
2053 (tdba & DMA_BIT_MASK(32)));
2054 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
2055 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
2056 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
2057 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
2058 adapter->tx_ring[i]->head = IXGBE_TDH(j);
2059 adapter->tx_ring[i]->tail = IXGBE_TDT(j);
2061 * Disable Tx Head Writeback RO bit, since this hoses
2062 * bookkeeping if things aren't delivered in order.
2064 switch (hw->mac.type) {
2065 case ixgbe_mac_82598EB:
2066 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
2068 case ixgbe_mac_82599EB:
2070 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(j));
2073 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
2074 switch (hw->mac.type) {
2075 case ixgbe_mac_82598EB:
2076 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
2078 case ixgbe_mac_82599EB:
2080 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(j), txctrl);
2085 if (hw->mac.type == ixgbe_mac_82599EB) {
2089 /* disable the arbiter while setting MTQC */
2090 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2091 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2092 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2094 /* set transmit pool layout */
2095 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2096 switch (adapter->flags & mask) {
2098 case (IXGBE_FLAG_SRIOV_ENABLED):
2099 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2100 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2103 case (IXGBE_FLAG_DCB_ENABLED):
2104 /* We enable 8 traffic classes, DCB only */
2105 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2106 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2110 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2114 /* re-eable the arbiter */
2115 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2116 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2120 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2122 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2123 struct ixgbe_ring *rx_ring)
2127 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2129 index = rx_ring->reg_idx;
2130 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2132 mask = (unsigned long) feature[RING_F_RSS].mask;
2133 index = index & mask;
2135 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2137 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2138 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2140 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2141 IXGBE_SRRCTL_BSIZEHDR_MASK;
2143 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2144 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2145 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2147 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2149 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2151 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2152 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2153 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2156 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2159 static u32 ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2164 if (!(adapter->hw.mac.type == ixgbe_mac_82599EB))
2167 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2168 #ifdef CONFIG_IXGBE_DCB
2169 | IXGBE_FLAG_DCB_ENABLED
2171 | IXGBE_FLAG_SRIOV_ENABLED
2175 case (IXGBE_FLAG_RSS_ENABLED):
2176 mrqc = IXGBE_MRQC_RSSEN;
2178 case (IXGBE_FLAG_SRIOV_ENABLED):
2179 mrqc = IXGBE_MRQC_VMDQEN;
2181 #ifdef CONFIG_IXGBE_DCB
2182 case (IXGBE_FLAG_DCB_ENABLED):
2183 mrqc = IXGBE_MRQC_RT8TCEN;
2185 #endif /* CONFIG_IXGBE_DCB */
2194 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2195 * @adapter: address of board private structure
2196 * @index: index of ring to set
2198 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter, int index)
2200 struct ixgbe_ring *rx_ring;
2201 struct ixgbe_hw *hw = &adapter->hw;
2206 rx_ring = adapter->rx_ring[index];
2207 j = rx_ring->reg_idx;
2208 rx_buf_len = rx_ring->rx_buf_len;
2209 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2210 rscctrl |= IXGBE_RSCCTL_RSCEN;
2212 * we must limit the number of descriptors so that the
2213 * total size of max desc * buf_len is not greater
2216 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2217 #if (MAX_SKB_FRAGS > 16)
2218 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2219 #elif (MAX_SKB_FRAGS > 8)
2220 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2221 #elif (MAX_SKB_FRAGS > 4)
2222 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2224 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2227 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2228 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2229 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2230 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2232 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2234 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2238 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2239 * @adapter: board private structure
2241 * Configure the Rx unit of the MAC after a reset.
2243 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2246 struct ixgbe_hw *hw = &adapter->hw;
2247 struct ixgbe_ring *rx_ring;
2248 struct net_device *netdev = adapter->netdev;
2249 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2251 u32 rdlen, rxctrl, rxcsum;
2252 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2253 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2254 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2256 u32 reta = 0, mrqc = 0;
2260 /* Decide whether to use packet split mode or not */
2261 /* Do not use packet split if we're in SR-IOV Mode */
2262 if (!adapter->num_vfs)
2263 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2265 /* Set the RX buffer length according to the mode */
2266 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2267 rx_buf_len = IXGBE_RX_HDR_SIZE;
2268 if (hw->mac.type == ixgbe_mac_82599EB) {
2269 /* PSRTYPE must be initialized in 82599 */
2270 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2271 IXGBE_PSRTYPE_UDPHDR |
2272 IXGBE_PSRTYPE_IPV4HDR |
2273 IXGBE_PSRTYPE_IPV6HDR |
2274 IXGBE_PSRTYPE_L2HDR;
2276 IXGBE_PSRTYPE(adapter->num_vfs),
2280 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2281 (netdev->mtu <= ETH_DATA_LEN))
2282 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2284 rx_buf_len = ALIGN(max_frame, 1024);
2287 fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
2288 fctrl |= IXGBE_FCTRL_BAM;
2289 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
2290 fctrl |= IXGBE_FCTRL_PMCF;
2291 IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
2293 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2294 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2295 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2297 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2299 if (netdev->features & NETIF_F_FCOE_MTU)
2300 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2302 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2304 rdlen = adapter->rx_ring[0]->count * sizeof(union ixgbe_adv_rx_desc);
2305 /* disable receives while setting up the descriptors */
2306 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2307 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2310 * Setup the HW Rx Head and Tail Descriptor Pointers and
2311 * the Base and Length of the Rx Descriptor Ring
2313 for (i = 0; i < adapter->num_rx_queues; i++) {
2314 rx_ring = adapter->rx_ring[i];
2315 rdba = rx_ring->dma;
2316 j = rx_ring->reg_idx;
2317 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_BIT_MASK(32)));
2318 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
2319 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
2320 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
2321 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
2322 rx_ring->head = IXGBE_RDH(j);
2323 rx_ring->tail = IXGBE_RDT(j);
2324 rx_ring->rx_buf_len = rx_buf_len;
2326 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2327 rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2329 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2332 if (netdev->features & NETIF_F_FCOE_MTU) {
2333 struct ixgbe_ring_feature *f;
2334 f = &adapter->ring_feature[RING_F_FCOE];
2335 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2336 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2337 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2338 rx_ring->rx_buf_len =
2339 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2343 #endif /* IXGBE_FCOE */
2344 ixgbe_configure_srrctl(adapter, rx_ring);
2347 if (hw->mac.type == ixgbe_mac_82598EB) {
2349 * For VMDq support of different descriptor types or
2350 * buffer sizes through the use of multiple SRRCTL
2351 * registers, RDRXCTL.MVMEN must be set to 1
2353 * also, the manual doesn't mention it clearly but DCA hints
2354 * will only use queue 0's tags unless this bit is set. Side
2355 * effects of setting this bit are only that SRRCTL must be
2356 * fully programmed [0..15]
2358 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2359 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2360 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2363 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2365 u32 reg_offset, vf_shift;
2366 u32 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2367 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN
2368 | IXGBE_VT_CTL_REPLEN;
2369 vt_reg_bits |= (adapter->num_vfs <<
2370 IXGBE_VT_CTL_POOL_SHIFT);
2371 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2372 IXGBE_WRITE_REG(hw, IXGBE_MRQC, 0);
2374 vf_shift = adapter->num_vfs % 32;
2375 reg_offset = adapter->num_vfs / 32;
2376 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), 0);
2377 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), 0);
2378 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), 0);
2379 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), 0);
2380 /* Enable only the PF's pool for Tx/Rx */
2381 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2382 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2383 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2384 ixgbe_set_vmolr(hw, adapter->num_vfs);
2387 /* Program MRQC for the distribution of queues */
2388 mrqc = ixgbe_setup_mrqc(adapter);
2390 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2391 /* Fill out redirection table */
2392 for (i = 0, j = 0; i < 128; i++, j++) {
2393 if (j == adapter->ring_feature[RING_F_RSS].indices)
2395 /* reta = 4-byte sliding window of
2396 * 0x00..(indices-1)(indices-1)00..etc. */
2397 reta = (reta << 8) | (j * 0x11);
2399 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2402 /* Fill out hash function seeds */
2403 for (i = 0; i < 10; i++)
2404 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2406 if (hw->mac.type == ixgbe_mac_82598EB)
2407 mrqc |= IXGBE_MRQC_RSSEN;
2408 /* Perform hash on these packet types */
2409 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2410 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2411 | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
2412 | IXGBE_MRQC_RSS_FIELD_IPV6
2413 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
2414 | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2416 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2418 if (adapter->num_vfs) {
2421 /* Map PF MAC address in RAR Entry 0 to first pool
2423 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2425 /* Set up VF register offsets for selected VT Mode, i.e.
2426 * 64 VFs for SR-IOV */
2427 reg = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2428 reg |= IXGBE_GCR_EXT_SRIOV;
2429 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, reg);
2432 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2434 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
2435 adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
2436 /* Disable indicating checksum in descriptor, enables
2438 rxcsum |= IXGBE_RXCSUM_PCSD;
2440 if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
2441 /* Enable IPv4 payload checksum for UDP fragments
2442 * if PCSD is not set */
2443 rxcsum |= IXGBE_RXCSUM_IPPCSE;
2446 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2448 if (hw->mac.type == ixgbe_mac_82599EB) {
2449 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2450 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2451 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2452 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2455 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
2456 /* Enable 82599 HW-RSC */
2457 for (i = 0; i < adapter->num_rx_queues; i++)
2458 ixgbe_configure_rscctl(adapter, i);
2460 /* Disable RSC for ACK packets */
2461 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2462 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2466 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2468 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2469 struct ixgbe_hw *hw = &adapter->hw;
2470 int pool_ndx = adapter->num_vfs;
2472 /* add VID to filter table */
2473 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
2476 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2478 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2479 struct ixgbe_hw *hw = &adapter->hw;
2480 int pool_ndx = adapter->num_vfs;
2482 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2483 ixgbe_irq_disable(adapter);
2485 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2487 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2488 ixgbe_irq_enable(adapter);
2490 /* remove VID from filter table */
2491 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
2495 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
2496 * @adapter: driver data
2498 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
2500 struct ixgbe_hw *hw = &adapter->hw;
2501 u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2504 switch (hw->mac.type) {
2505 case ixgbe_mac_82598EB:
2506 vlnctrl &= ~(IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE);
2507 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2508 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2510 case ixgbe_mac_82599EB:
2511 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2512 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2513 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2514 for (i = 0; i < adapter->num_rx_queues; i++) {
2515 j = adapter->rx_ring[i]->reg_idx;
2516 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2517 vlnctrl &= ~IXGBE_RXDCTL_VME;
2518 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2527 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
2528 * @adapter: driver data
2530 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
2532 struct ixgbe_hw *hw = &adapter->hw;
2533 u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2536 switch (hw->mac.type) {
2537 case ixgbe_mac_82598EB:
2538 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2539 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2540 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2542 case ixgbe_mac_82599EB:
2543 vlnctrl |= IXGBE_VLNCTRL_VFE;
2544 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2545 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2546 for (i = 0; i < adapter->num_rx_queues; i++) {
2547 j = adapter->rx_ring[i]->reg_idx;
2548 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2549 vlnctrl |= IXGBE_RXDCTL_VME;
2550 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2558 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2559 struct vlan_group *grp)
2561 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2563 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2564 ixgbe_irq_disable(adapter);
2565 adapter->vlgrp = grp;
2568 * For a DCB driver, always enable VLAN tag stripping so we can
2569 * still receive traffic from a DCB-enabled host even if we're
2572 ixgbe_vlan_filter_enable(adapter);
2574 ixgbe_vlan_rx_add_vid(netdev, 0);
2576 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2577 ixgbe_irq_enable(adapter);
2580 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2582 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2584 if (adapter->vlgrp) {
2586 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2587 if (!vlan_group_get_device(adapter->vlgrp, vid))
2589 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2595 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
2596 * @netdev: network interface device structure
2598 * The set_rx_method entry point is called whenever the unicast/multicast
2599 * address list or the network interface flags are updated. This routine is
2600 * responsible for configuring the hardware for proper unicast, multicast and
2603 void ixgbe_set_rx_mode(struct net_device *netdev)
2605 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2606 struct ixgbe_hw *hw = &adapter->hw;
2609 /* Check for Promiscuous and All Multicast modes */
2611 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2613 if (netdev->flags & IFF_PROMISC) {
2614 hw->addr_ctrl.user_set_promisc = 1;
2615 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2616 /* don't hardware filter vlans in promisc mode */
2617 ixgbe_vlan_filter_disable(adapter);
2619 if (netdev->flags & IFF_ALLMULTI) {
2620 fctrl |= IXGBE_FCTRL_MPE;
2621 fctrl &= ~IXGBE_FCTRL_UPE;
2623 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2625 ixgbe_vlan_filter_enable(adapter);
2626 hw->addr_ctrl.user_set_promisc = 0;
2629 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
2631 /* reprogram secondary unicast list */
2632 hw->mac.ops.update_uc_addr_list(hw, netdev);
2634 /* reprogram multicast list */
2635 hw->mac.ops.update_mc_addr_list(hw, netdev);
2637 if (adapter->num_vfs)
2638 ixgbe_restore_vf_multicasts(adapter);
2641 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
2644 struct ixgbe_q_vector *q_vector;
2645 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2647 /* legacy and MSI only use one vector */
2648 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2651 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2652 struct napi_struct *napi;
2653 q_vector = adapter->q_vector[q_idx];
2654 napi = &q_vector->napi;
2655 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2656 if (!q_vector->rxr_count || !q_vector->txr_count) {
2657 if (q_vector->txr_count == 1)
2658 napi->poll = &ixgbe_clean_txonly;
2659 else if (q_vector->rxr_count == 1)
2660 napi->poll = &ixgbe_clean_rxonly;
2668 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2671 struct ixgbe_q_vector *q_vector;
2672 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2674 /* legacy and MSI only use one vector */
2675 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2678 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2679 q_vector = adapter->q_vector[q_idx];
2680 napi_disable(&q_vector->napi);
2684 #ifdef CONFIG_IXGBE_DCB
2686 * ixgbe_configure_dcb - Configure DCB hardware
2687 * @adapter: ixgbe adapter struct
2689 * This is called by the driver on open to configure the DCB hardware.
2690 * This is also called by the gennetlink interface when reconfiguring
2693 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2695 struct ixgbe_hw *hw = &adapter->hw;
2699 ixgbe_dcb_check_config(&adapter->dcb_cfg);
2700 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2701 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2703 /* reconfigure the hardware */
2704 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
2706 for (i = 0; i < adapter->num_tx_queues; i++) {
2707 j = adapter->tx_ring[i]->reg_idx;
2708 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2709 /* PThresh workaround for Tx hang with DFP enabled. */
2711 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2713 /* Enable VLAN tag insert/strip */
2714 ixgbe_vlan_filter_enable(adapter);
2716 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2720 static void ixgbe_configure(struct ixgbe_adapter *adapter)
2722 struct net_device *netdev = adapter->netdev;
2723 struct ixgbe_hw *hw = &adapter->hw;
2726 ixgbe_set_rx_mode(netdev);
2728 ixgbe_restore_vlan(adapter);
2729 #ifdef CONFIG_IXGBE_DCB
2730 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2731 if (hw->mac.type == ixgbe_mac_82598EB)
2732 netif_set_gso_max_size(netdev, 32768);
2734 netif_set_gso_max_size(netdev, 65536);
2735 ixgbe_configure_dcb(adapter);
2737 netif_set_gso_max_size(netdev, 65536);
2740 netif_set_gso_max_size(netdev, 65536);
2744 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2745 ixgbe_configure_fcoe(adapter);
2747 #endif /* IXGBE_FCOE */
2748 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2749 for (i = 0; i < adapter->num_tx_queues; i++)
2750 adapter->tx_ring[i]->atr_sample_rate =
2751 adapter->atr_sample_rate;
2752 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
2753 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
2754 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
2757 ixgbe_configure_tx(adapter);
2758 ixgbe_configure_rx(adapter);
2759 for (i = 0; i < adapter->num_rx_queues; i++)
2760 ixgbe_alloc_rx_buffers(adapter, adapter->rx_ring[i],
2761 (adapter->rx_ring[i]->count - 1));
2764 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2766 switch (hw->phy.type) {
2767 case ixgbe_phy_sfp_avago:
2768 case ixgbe_phy_sfp_ftl:
2769 case ixgbe_phy_sfp_intel:
2770 case ixgbe_phy_sfp_unknown:
2771 case ixgbe_phy_tw_tyco:
2772 case ixgbe_phy_tw_unknown:
2780 * ixgbe_sfp_link_config - set up SFP+ link
2781 * @adapter: pointer to private adapter struct
2783 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2785 struct ixgbe_hw *hw = &adapter->hw;
2787 if (hw->phy.multispeed_fiber) {
2789 * In multispeed fiber setups, the device may not have
2790 * had a physical connection when the driver loaded.
2791 * If that's the case, the initial link configuration
2792 * couldn't get the MAC into 10G or 1G mode, so we'll
2793 * never have a link status change interrupt fire.
2794 * We need to try and force an autonegotiation
2795 * session, then bring up link.
2797 hw->mac.ops.setup_sfp(hw);
2798 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2799 schedule_work(&adapter->multispeed_fiber_task);
2802 * Direct Attach Cu and non-multispeed fiber modules
2803 * still need to be configured properly prior to
2806 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2807 schedule_work(&adapter->sfp_config_module_task);
2812 * ixgbe_non_sfp_link_config - set up non-SFP+ link
2813 * @hw: pointer to private hardware struct
2815 * Returns 0 on success, negative on failure
2817 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2820 bool negotiation, link_up = false;
2821 u32 ret = IXGBE_ERR_LINK_SETUP;
2823 if (hw->mac.ops.check_link)
2824 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2829 if (hw->mac.ops.get_link_capabilities)
2830 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
2834 if (hw->mac.ops.setup_link)
2835 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
2840 #define IXGBE_MAX_RX_DESC_POLL 10
2841 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2844 int j = adapter->rx_ring[rxr]->reg_idx;
2847 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2848 if (IXGBE_READ_REG(&adapter->hw,
2849 IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2854 if (k >= IXGBE_MAX_RX_DESC_POLL) {
2855 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2856 "not set within the polling period\n", rxr);
2858 ixgbe_release_rx_desc(&adapter->hw, adapter->rx_ring[rxr],
2859 (adapter->rx_ring[rxr]->count - 1));
2862 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2864 struct net_device *netdev = adapter->netdev;
2865 struct ixgbe_hw *hw = &adapter->hw;
2867 int num_rx_rings = adapter->num_rx_queues;
2869 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2870 u32 txdctl, rxdctl, mhadd;
2875 ixgbe_get_hw_control(adapter);
2877 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2878 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2879 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2880 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2881 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2886 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2887 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
2888 gpie |= IXGBE_GPIE_VTMODE_64;
2890 /* XXX: to interrupt immediately for EICS writes, enable this */
2891 /* gpie |= IXGBE_GPIE_EIMEN; */
2892 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2895 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2897 * use EIAM to auto-mask when MSI-X interrupt is asserted
2898 * this saves a register write for every interrupt
2900 switch (hw->mac.type) {
2901 case ixgbe_mac_82598EB:
2902 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2905 case ixgbe_mac_82599EB:
2906 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
2907 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
2911 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2912 * specifically only auto mask tx and rx interrupts */
2913 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2916 /* Enable fan failure interrupt if media type is copper */
2917 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2918 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2919 gpie |= IXGBE_SDP1_GPIEN;
2920 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2923 if (hw->mac.type == ixgbe_mac_82599EB) {
2924 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2925 gpie |= IXGBE_SDP1_GPIEN;
2926 gpie |= IXGBE_SDP2_GPIEN;
2927 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2931 /* adjust max frame to be able to do baby jumbo for FCoE */
2932 if ((netdev->features & NETIF_F_FCOE_MTU) &&
2933 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2934 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2936 #endif /* IXGBE_FCOE */
2937 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2938 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2939 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2940 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2942 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2945 for (i = 0; i < adapter->num_tx_queues; i++) {
2946 j = adapter->tx_ring[i]->reg_idx;
2947 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2948 if (adapter->rx_itr_setting == 0) {
2949 /* cannot set wthresh when itr==0 */
2950 txdctl &= ~0x007F0000;
2952 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2953 txdctl |= (8 << 16);
2955 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2958 if (hw->mac.type == ixgbe_mac_82599EB) {
2959 /* DMATXCTL.EN must be set after all Tx queue config is done */
2960 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2961 dmatxctl |= IXGBE_DMATXCTL_TE;
2962 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2964 for (i = 0; i < adapter->num_tx_queues; i++) {
2965 j = adapter->tx_ring[i]->reg_idx;
2966 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2967 txdctl |= IXGBE_TXDCTL_ENABLE;
2968 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2969 if (hw->mac.type == ixgbe_mac_82599EB) {
2971 /* poll for Tx Enable ready */
2974 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2975 } while (--wait_loop &&
2976 !(txdctl & IXGBE_TXDCTL_ENABLE));
2978 DPRINTK(DRV, ERR, "Could not enable "
2979 "Tx Queue %d\n", j);
2983 for (i = 0; i < num_rx_rings; i++) {
2984 j = adapter->rx_ring[i]->reg_idx;
2985 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2986 /* enable PTHRESH=32 descriptors (half the internal cache)
2987 * and HTHRESH=0 descriptors (to minimize latency on fetch),
2988 * this also removes a pesky rx_no_buffer_count increment */
2990 rxdctl |= IXGBE_RXDCTL_ENABLE;
2991 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2992 if (hw->mac.type == ixgbe_mac_82599EB)
2993 ixgbe_rx_desc_queue_enable(adapter, i);
2995 /* enable all receives */
2996 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2997 if (hw->mac.type == ixgbe_mac_82598EB)
2998 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
3000 rxdctl |= IXGBE_RXCTRL_RXEN;
3001 hw->mac.ops.enable_rx_dma(hw, rxdctl);
3003 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3004 ixgbe_configure_msix(adapter);
3006 ixgbe_configure_msi_and_legacy(adapter);
3008 /* enable the optics */
3009 if (hw->phy.multispeed_fiber)
3010 hw->mac.ops.enable_tx_laser(hw);
3012 clear_bit(__IXGBE_DOWN, &adapter->state);
3013 ixgbe_napi_enable_all(adapter);
3015 /* clear any pending interrupts, may auto mask */
3016 IXGBE_READ_REG(hw, IXGBE_EICR);
3018 ixgbe_irq_enable(adapter);
3021 * If this adapter has a fan, check to see if we had a failure
3022 * before we enabled the interrupt.
3024 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3025 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3026 if (esdp & IXGBE_ESDP_SDP1)
3028 "Fan has stopped, replace the adapter\n");
3032 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3033 * arrived before interrupts were enabled but after probe. Such
3034 * devices wouldn't have their type identified yet. We need to
3035 * kick off the SFP+ module setup first, then try to bring up link.
3036 * If we're not hot-pluggable SFP+, we just need to configure link
3039 if (hw->phy.type == ixgbe_phy_unknown) {
3040 err = hw->phy.ops.identify(hw);
3041 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3043 * Take the device down and schedule the sfp tasklet
3044 * which will unregister_netdev and log it.
3046 ixgbe_down(adapter);
3047 schedule_work(&adapter->sfp_config_module_task);
3052 if (ixgbe_is_sfp(hw)) {
3053 ixgbe_sfp_link_config(adapter);
3055 err = ixgbe_non_sfp_link_config(hw);
3057 DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
3060 for (i = 0; i < adapter->num_tx_queues; i++)
3061 set_bit(__IXGBE_FDIR_INIT_DONE,
3062 &(adapter->tx_ring[i]->reinit_state));
3064 /* enable transmits */
3065 netif_tx_start_all_queues(netdev);
3067 /* bring the link up in the watchdog, this could race with our first
3068 * link up interrupt but shouldn't be a problem */
3069 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3070 adapter->link_check_timeout = jiffies;
3071 mod_timer(&adapter->watchdog_timer, jiffies);
3073 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3074 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3075 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3076 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3081 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3083 WARN_ON(in_interrupt());
3084 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3086 ixgbe_down(adapter);
3088 * If SR-IOV enabled then wait a bit before bringing the adapter
3089 * back up to give the VFs time to respond to the reset. The
3090 * two second wait is based upon the watchdog timer cycle in
3093 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3096 clear_bit(__IXGBE_RESETTING, &adapter->state);
3099 int ixgbe_up(struct ixgbe_adapter *adapter)
3101 /* hardware has been reset, we need to reload some things */
3102 ixgbe_configure(adapter);
3104 return ixgbe_up_complete(adapter);
3107 void ixgbe_reset(struct ixgbe_adapter *adapter)
3109 struct ixgbe_hw *hw = &adapter->hw;
3112 err = hw->mac.ops.init_hw(hw);
3115 case IXGBE_ERR_SFP_NOT_PRESENT:
3117 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3118 dev_err(&adapter->pdev->dev, "master disable timed out\n");
3120 case IXGBE_ERR_EEPROM_VERSION:
3121 /* We are running on a pre-production device, log a warning */
3122 dev_warn(&adapter->pdev->dev, "This device is a pre-production "
3123 "adapter/LOM. Please be aware there may be issues "
3124 "associated with your hardware. If you are "
3125 "experiencing problems please contact your Intel or "
3126 "hardware representative who provided you with this "
3130 dev_err(&adapter->pdev->dev, "Hardware Error: %d\n", err);
3133 /* reprogram the RAR[0] in case user changed it. */
3134 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3139 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3140 * @adapter: board private structure
3141 * @rx_ring: ring to free buffers from
3143 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
3144 struct ixgbe_ring *rx_ring)
3146 struct pci_dev *pdev = adapter->pdev;
3150 /* Free all the Rx ring sk_buffs */
3152 for (i = 0; i < rx_ring->count; i++) {
3153 struct ixgbe_rx_buffer *rx_buffer_info;
3155 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3156 if (rx_buffer_info->dma) {
3157 pci_unmap_single(pdev, rx_buffer_info->dma,
3158 rx_ring->rx_buf_len,
3159 PCI_DMA_FROMDEVICE);
3160 rx_buffer_info->dma = 0;
3162 if (rx_buffer_info->skb) {
3163 struct sk_buff *skb = rx_buffer_info->skb;
3164 rx_buffer_info->skb = NULL;
3166 struct sk_buff *this = skb;
3167 if (IXGBE_RSC_CB(this)->dma) {
3168 pci_unmap_single(pdev, IXGBE_RSC_CB(this)->dma,
3169 rx_ring->rx_buf_len,
3170 PCI_DMA_FROMDEVICE);
3171 IXGBE_RSC_CB(this)->dma = 0;
3174 dev_kfree_skb(this);
3177 if (!rx_buffer_info->page)
3179 if (rx_buffer_info->page_dma) {
3180 pci_unmap_page(pdev, rx_buffer_info->page_dma,
3181 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
3182 rx_buffer_info->page_dma = 0;
3184 put_page(rx_buffer_info->page);
3185 rx_buffer_info->page = NULL;
3186 rx_buffer_info->page_offset = 0;
3189 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3190 memset(rx_ring->rx_buffer_info, 0, size);
3192 /* Zero out the descriptor ring */
3193 memset(rx_ring->desc, 0, rx_ring->size);
3195 rx_ring->next_to_clean = 0;
3196 rx_ring->next_to_use = 0;
3199 writel(0, adapter->hw.hw_addr + rx_ring->head);
3201 writel(0, adapter->hw.hw_addr + rx_ring->tail);
3205 * ixgbe_clean_tx_ring - Free Tx Buffers
3206 * @adapter: board private structure
3207 * @tx_ring: ring to be cleaned
3209 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
3210 struct ixgbe_ring *tx_ring)
3212 struct ixgbe_tx_buffer *tx_buffer_info;
3216 /* Free all the Tx ring sk_buffs */
3218 for (i = 0; i < tx_ring->count; i++) {
3219 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3220 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3223 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3224 memset(tx_ring->tx_buffer_info, 0, size);
3226 /* Zero out the descriptor ring */
3227 memset(tx_ring->desc, 0, tx_ring->size);
3229 tx_ring->next_to_use = 0;
3230 tx_ring->next_to_clean = 0;
3233 writel(0, adapter->hw.hw_addr + tx_ring->head);
3235 writel(0, adapter->hw.hw_addr + tx_ring->tail);
3239 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3240 * @adapter: board private structure
3242 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3246 for (i = 0; i < adapter->num_rx_queues; i++)
3247 ixgbe_clean_rx_ring(adapter, adapter->rx_ring[i]);
3251 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3252 * @adapter: board private structure
3254 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3258 for (i = 0; i < adapter->num_tx_queues; i++)
3259 ixgbe_clean_tx_ring(adapter, adapter->tx_ring[i]);
3262 void ixgbe_down(struct ixgbe_adapter *adapter)
3264 struct net_device *netdev = adapter->netdev;
3265 struct ixgbe_hw *hw = &adapter->hw;
3270 /* signal that we are down to the interrupt handler */
3271 set_bit(__IXGBE_DOWN, &adapter->state);
3273 /* power down the optics */
3274 if (hw->phy.multispeed_fiber)
3275 hw->mac.ops.disable_tx_laser(hw);
3277 /* disable receive for all VFs and wait one second */
3278 if (adapter->num_vfs) {
3279 /* ping all the active vfs to let them know we are going down */
3280 ixgbe_ping_all_vfs(adapter);
3282 /* Disable all VFTE/VFRE TX/RX */
3283 ixgbe_disable_tx_rx(adapter);
3285 /* Mark all the VFs as inactive */
3286 for (i = 0 ; i < adapter->num_vfs; i++)
3287 adapter->vfinfo[i].clear_to_send = 0;
3290 /* disable receives */
3291 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3292 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3294 IXGBE_WRITE_FLUSH(hw);
3297 netif_tx_stop_all_queues(netdev);
3299 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3300 del_timer_sync(&adapter->sfp_timer);
3301 del_timer_sync(&adapter->watchdog_timer);
3302 cancel_work_sync(&adapter->watchdog_task);
3304 netif_carrier_off(netdev);
3305 netif_tx_disable(netdev);
3307 ixgbe_irq_disable(adapter);
3309 ixgbe_napi_disable_all(adapter);
3311 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3312 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3313 cancel_work_sync(&adapter->fdir_reinit_task);
3315 /* disable transmits in the hardware now that interrupts are off */
3316 for (i = 0; i < adapter->num_tx_queues; i++) {
3317 j = adapter->tx_ring[i]->reg_idx;
3318 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3319 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3320 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3322 /* Disable the Tx DMA engine on 82599 */
3323 if (hw->mac.type == ixgbe_mac_82599EB)
3324 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3325 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3326 ~IXGBE_DMATXCTL_TE));
3328 /* clear n-tuple filters that are cached */
3329 ethtool_ntuple_flush(netdev);
3331 if (!pci_channel_offline(adapter->pdev))
3332 ixgbe_reset(adapter);
3333 ixgbe_clean_all_tx_rings(adapter);
3334 ixgbe_clean_all_rx_rings(adapter);
3336 #ifdef CONFIG_IXGBE_DCA
3337 /* since we reset the hardware DCA settings were cleared */
3338 ixgbe_setup_dca(adapter);
3343 * ixgbe_poll - NAPI Rx polling callback
3344 * @napi: structure for representing this polling device
3345 * @budget: how many packets driver is allowed to clean
3347 * This function is used for legacy and MSI, NAPI mode
3349 static int ixgbe_poll(struct napi_struct *napi, int budget)
3351 struct ixgbe_q_vector *q_vector =
3352 container_of(napi, struct ixgbe_q_vector, napi);
3353 struct ixgbe_adapter *adapter = q_vector->adapter;
3354 int tx_clean_complete, work_done = 0;
3356 #ifdef CONFIG_IXGBE_DCA
3357 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3358 ixgbe_update_tx_dca(adapter, adapter->tx_ring[0]);
3359 ixgbe_update_rx_dca(adapter, adapter->rx_ring[0]);
3363 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3364 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3366 if (!tx_clean_complete)
3369 /* If budget not fully consumed, exit the polling mode */
3370 if (work_done < budget) {
3371 napi_complete(napi);
3372 if (adapter->rx_itr_setting & 1)
3373 ixgbe_set_itr(adapter);
3374 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3375 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3381 * ixgbe_tx_timeout - Respond to a Tx Hang
3382 * @netdev: network interface device structure
3384 static void ixgbe_tx_timeout(struct net_device *netdev)
3386 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3388 /* Do the reset outside of interrupt context */
3389 schedule_work(&adapter->reset_task);
3392 static void ixgbe_reset_task(struct work_struct *work)
3394 struct ixgbe_adapter *adapter;
3395 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3397 /* If we're already down or resetting, just bail */
3398 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3399 test_bit(__IXGBE_RESETTING, &adapter->state))
3402 adapter->tx_timeout_count++;
3404 ixgbe_reinit_locked(adapter);
3407 #ifdef CONFIG_IXGBE_DCB
3408 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3411 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3413 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3417 adapter->num_rx_queues = f->indices;
3418 adapter->num_tx_queues = f->indices;
3426 * ixgbe_set_rss_queues: Allocate queues for RSS
3427 * @adapter: board private structure to initialize
3429 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3430 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3433 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3436 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3438 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3440 adapter->num_rx_queues = f->indices;
3441 adapter->num_tx_queues = f->indices;
3451 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3452 * @adapter: board private structure to initialize
3454 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3455 * to the original CPU that initiated the Tx session. This runs in addition
3456 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3457 * Rx load across CPUs using RSS.
3460 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3463 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3465 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3468 /* Flow Director must have RSS enabled */
3469 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3470 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3471 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3472 adapter->num_tx_queues = f_fdir->indices;
3473 adapter->num_rx_queues = f_fdir->indices;
3476 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3477 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3484 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3485 * @adapter: board private structure to initialize
3487 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3488 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3489 * rx queues out of the max number of rx queues, instead, it is used as the
3490 * index of the first rx queue used by FCoE.
3493 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3496 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3498 f->indices = min((int)num_online_cpus(), f->indices);
3499 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3500 adapter->num_rx_queues = 1;
3501 adapter->num_tx_queues = 1;
3502 #ifdef CONFIG_IXGBE_DCB
3503 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3504 DPRINTK(PROBE, INFO, "FCoE enabled with DCB\n");
3505 ixgbe_set_dcb_queues(adapter);
3508 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3509 DPRINTK(PROBE, INFO, "FCoE enabled with RSS\n");
3510 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3511 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3512 ixgbe_set_fdir_queues(adapter);
3514 ixgbe_set_rss_queues(adapter);
3516 /* adding FCoE rx rings to the end */
3517 f->mask = adapter->num_rx_queues;
3518 adapter->num_rx_queues += f->indices;
3519 adapter->num_tx_queues += f->indices;
3527 #endif /* IXGBE_FCOE */
3529 * ixgbe_set_sriov_queues: Allocate queues for IOV use
3530 * @adapter: board private structure to initialize
3532 * IOV doesn't actually use anything, so just NAK the
3533 * request for now and let the other queue routines
3534 * figure out what to do.
3536 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
3542 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3543 * @adapter: board private structure to initialize
3545 * This is the top level queue allocation routine. The order here is very
3546 * important, starting with the "most" number of features turned on at once,
3547 * and ending with the smallest set of features. This way large combinations
3548 * can be allocated if they're turned on, and smaller combinations are the
3549 * fallthrough conditions.
3552 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
3554 /* Start with base case */
3555 adapter->num_rx_queues = 1;
3556 adapter->num_tx_queues = 1;
3557 adapter->num_rx_pools = adapter->num_rx_queues;
3558 adapter->num_rx_queues_per_pool = 1;
3560 if (ixgbe_set_sriov_queues(adapter))
3564 if (ixgbe_set_fcoe_queues(adapter))
3567 #endif /* IXGBE_FCOE */
3568 #ifdef CONFIG_IXGBE_DCB
3569 if (ixgbe_set_dcb_queues(adapter))
3573 if (ixgbe_set_fdir_queues(adapter))
3576 if (ixgbe_set_rss_queues(adapter))
3579 /* fallback to base case */
3580 adapter->num_rx_queues = 1;
3581 adapter->num_tx_queues = 1;
3584 /* Notify the stack of the (possibly) reduced Tx Queue count. */
3585 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
3588 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
3591 int err, vector_threshold;
3593 /* We'll want at least 3 (vector_threshold):
3596 * 3) Other (Link Status Change, etc.)
3597 * 4) TCP Timer (optional)
3599 vector_threshold = MIN_MSIX_COUNT;
3601 /* The more we get, the more we will assign to Tx/Rx Cleanup
3602 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
3603 * Right now, we simply care about how many we'll get; we'll
3604 * set them up later while requesting irq's.
3606 while (vectors >= vector_threshold) {
3607 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
3609 if (!err) /* Success in acquiring all requested vectors. */
3612 vectors = 0; /* Nasty failure, quit now */
3613 else /* err == number of vectors we should try again with */
3617 if (vectors < vector_threshold) {
3618 /* Can't allocate enough MSI-X interrupts? Oh well.
3619 * This just means we'll go with either a single MSI
3620 * vector or fall back to legacy interrupts.
3622 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
3623 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3624 kfree(adapter->msix_entries);
3625 adapter->msix_entries = NULL;
3627 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
3629 * Adjust for only the vectors we'll use, which is minimum
3630 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
3631 * vectors we were allocated.
3633 adapter->num_msix_vectors = min(vectors,
3634 adapter->max_msix_q_vectors + NON_Q_VECTORS);
3639 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
3640 * @adapter: board private structure to initialize
3642 * Cache the descriptor ring offsets for RSS to the assigned rings.
3645 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
3650 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3651 for (i = 0; i < adapter->num_rx_queues; i++)
3652 adapter->rx_ring[i]->reg_idx = i;
3653 for (i = 0; i < adapter->num_tx_queues; i++)
3654 adapter->tx_ring[i]->reg_idx = i;
3663 #ifdef CONFIG_IXGBE_DCB
3665 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
3666 * @adapter: board private structure to initialize
3668 * Cache the descriptor ring offsets for DCB to the assigned rings.
3671 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
3675 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
3677 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3678 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3679 /* the number of queues is assumed to be symmetric */
3680 for (i = 0; i < dcb_i; i++) {
3681 adapter->rx_ring[i]->reg_idx = i << 3;
3682 adapter->tx_ring[i]->reg_idx = i << 2;
3685 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
3688 * Tx TC0 starts at: descriptor queue 0
3689 * Tx TC1 starts at: descriptor queue 32
3690 * Tx TC2 starts at: descriptor queue 64
3691 * Tx TC3 starts at: descriptor queue 80
3692 * Tx TC4 starts at: descriptor queue 96
3693 * Tx TC5 starts at: descriptor queue 104
3694 * Tx TC6 starts at: descriptor queue 112
3695 * Tx TC7 starts at: descriptor queue 120
3697 * Rx TC0-TC7 are offset by 16 queues each
3699 for (i = 0; i < 3; i++) {
3700 adapter->tx_ring[i]->reg_idx = i << 5;
3701 adapter->rx_ring[i]->reg_idx = i << 4;
3703 for ( ; i < 5; i++) {
3704 adapter->tx_ring[i]->reg_idx =
3706 adapter->rx_ring[i]->reg_idx = i << 4;
3708 for ( ; i < dcb_i; i++) {
3709 adapter->tx_ring[i]->reg_idx =
3711 adapter->rx_ring[i]->reg_idx = i << 4;
3715 } else if (dcb_i == 4) {
3717 * Tx TC0 starts at: descriptor queue 0
3718 * Tx TC1 starts at: descriptor queue 64
3719 * Tx TC2 starts at: descriptor queue 96
3720 * Tx TC3 starts at: descriptor queue 112
3722 * Rx TC0-TC3 are offset by 32 queues each
3724 adapter->tx_ring[0]->reg_idx = 0;
3725 adapter->tx_ring[1]->reg_idx = 64;
3726 adapter->tx_ring[2]->reg_idx = 96;
3727 adapter->tx_ring[3]->reg_idx = 112;
3728 for (i = 0 ; i < dcb_i; i++)
3729 adapter->rx_ring[i]->reg_idx = i << 5;
3747 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
3748 * @adapter: board private structure to initialize
3750 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
3753 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
3758 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3759 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3760 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
3761 for (i = 0; i < adapter->num_rx_queues; i++)
3762 adapter->rx_ring[i]->reg_idx = i;
3763 for (i = 0; i < adapter->num_tx_queues; i++)
3764 adapter->tx_ring[i]->reg_idx = i;
3773 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
3774 * @adapter: board private structure to initialize
3776 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
3779 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
3781 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
3783 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3785 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3786 #ifdef CONFIG_IXGBE_DCB
3787 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3788 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
3790 ixgbe_cache_ring_dcb(adapter);
3791 /* find out queues in TC for FCoE */
3792 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
3793 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
3795 * In 82599, the number of Tx queues for each traffic
3796 * class for both 8-TC and 4-TC modes are:
3797 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
3798 * 8 TCs: 32 32 16 16 8 8 8 8
3799 * 4 TCs: 64 64 32 32
3800 * We have max 8 queues for FCoE, where 8 the is
3801 * FCoE redirection table size. If TC for FCoE is
3802 * less than or equal to TC3, we have enough queues
3803 * to add max of 8 queues for FCoE, so we start FCoE
3804 * tx descriptor from the next one, i.e., reg_idx + 1.
3805 * If TC for FCoE is above TC3, implying 8 TC mode,
3806 * and we need 8 for FCoE, we have to take all queues
3807 * in that traffic class for FCoE.
3809 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
3812 #endif /* CONFIG_IXGBE_DCB */
3813 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3814 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3815 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3816 ixgbe_cache_ring_fdir(adapter);
3818 ixgbe_cache_ring_rss(adapter);
3820 fcoe_rx_i = f->mask;
3821 fcoe_tx_i = f->mask;
3823 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
3824 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
3825 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
3832 #endif /* IXGBE_FCOE */
3834 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
3835 * @adapter: board private structure to initialize
3837 * SR-IOV doesn't use any descriptor rings but changes the default if
3838 * no other mapping is used.
3841 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
3843 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
3844 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
3845 if (adapter->num_vfs)
3852 * ixgbe_cache_ring_register - Descriptor ring to register mapping
3853 * @adapter: board private structure to initialize
3855 * Once we know the feature-set enabled for the device, we'll cache
3856 * the register offset the descriptor ring is assigned to.
3858 * Note, the order the various feature calls is important. It must start with
3859 * the "most" features enabled at the same time, then trickle down to the
3860 * least amount of features turned on at once.
3862 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
3864 /* start with default case */
3865 adapter->rx_ring[0]->reg_idx = 0;
3866 adapter->tx_ring[0]->reg_idx = 0;
3868 if (ixgbe_cache_ring_sriov(adapter))
3872 if (ixgbe_cache_ring_fcoe(adapter))
3875 #endif /* IXGBE_FCOE */
3876 #ifdef CONFIG_IXGBE_DCB
3877 if (ixgbe_cache_ring_dcb(adapter))
3881 if (ixgbe_cache_ring_fdir(adapter))
3884 if (ixgbe_cache_ring_rss(adapter))
3889 * ixgbe_alloc_queues - Allocate memory for all rings
3890 * @adapter: board private structure to initialize
3892 * We allocate one ring per queue at run-time since we don't know the
3893 * number of queues at compile-time. The polling_netdev array is
3894 * intended for Multiqueue, but should work fine with a single queue.
3896 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
3899 int orig_node = adapter->node;
3901 for (i = 0; i < adapter->num_tx_queues; i++) {
3902 struct ixgbe_ring *ring = adapter->tx_ring[i];
3903 if (orig_node == -1) {
3904 int cur_node = next_online_node(adapter->node);
3905 if (cur_node == MAX_NUMNODES)
3906 cur_node = first_online_node;
3907 adapter->node = cur_node;
3909 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
3912 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
3914 goto err_tx_ring_allocation;
3915 ring->count = adapter->tx_ring_count;
3916 ring->queue_index = i;
3917 ring->numa_node = adapter->node;
3919 adapter->tx_ring[i] = ring;
3922 /* Restore the adapter's original node */
3923 adapter->node = orig_node;
3925 for (i = 0; i < adapter->num_rx_queues; i++) {
3926 struct ixgbe_ring *ring = adapter->rx_ring[i];
3927 if (orig_node == -1) {
3928 int cur_node = next_online_node(adapter->node);
3929 if (cur_node == MAX_NUMNODES)
3930 cur_node = first_online_node;
3931 adapter->node = cur_node;
3933 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
3936 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
3938 goto err_rx_ring_allocation;
3939 ring->count = adapter->rx_ring_count;
3940 ring->queue_index = i;
3941 ring->numa_node = adapter->node;
3943 adapter->rx_ring[i] = ring;
3946 /* Restore the adapter's original node */
3947 adapter->node = orig_node;
3949 ixgbe_cache_ring_register(adapter);
3953 err_rx_ring_allocation:
3954 for (i = 0; i < adapter->num_tx_queues; i++)
3955 kfree(adapter->tx_ring[i]);
3956 err_tx_ring_allocation:
3961 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
3962 * @adapter: board private structure to initialize
3964 * Attempt to configure the interrupts using the best available
3965 * capabilities of the hardware and the kernel.
3967 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
3969 struct ixgbe_hw *hw = &adapter->hw;
3971 int vector, v_budget;
3974 * It's easy to be greedy for MSI-X vectors, but it really
3975 * doesn't do us much good if we have a lot more vectors
3976 * than CPU's. So let's be conservative and only ask for
3977 * (roughly) the same number of vectors as there are CPU's.
3979 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
3980 (int)num_online_cpus()) + NON_Q_VECTORS;
3983 * At the same time, hardware can only support a maximum of
3984 * hw.mac->max_msix_vectors vectors. With features
3985 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
3986 * descriptor queues supported by our device. Thus, we cap it off in
3987 * those rare cases where the cpu count also exceeds our vector limit.
3989 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
3991 /* A failure in MSI-X entry allocation isn't fatal, but it does
3992 * mean we disable MSI-X capabilities of the adapter. */
3993 adapter->msix_entries = kcalloc(v_budget,
3994 sizeof(struct msix_entry), GFP_KERNEL);
3995 if (adapter->msix_entries) {
3996 for (vector = 0; vector < v_budget; vector++)
3997 adapter->msix_entries[vector].entry = vector;
3999 ixgbe_acquire_msix_vectors(adapter, v_budget);
4001 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4005 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4006 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4007 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4008 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4009 adapter->atr_sample_rate = 0;
4010 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4011 ixgbe_disable_sriov(adapter);
4013 ixgbe_set_num_queues(adapter);
4015 err = pci_enable_msi(adapter->pdev);
4017 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4019 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
4020 "falling back to legacy. Error: %d\n", err);
4030 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4031 * @adapter: board private structure to initialize
4033 * We allocate one q_vector per queue interrupt. If allocation fails we
4036 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4038 int q_idx, num_q_vectors;
4039 struct ixgbe_q_vector *q_vector;
4041 int (*poll)(struct napi_struct *, int);
4043 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4044 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4045 napi_vectors = adapter->num_rx_queues;
4046 poll = &ixgbe_clean_rxtx_many;
4053 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4054 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4055 GFP_KERNEL, adapter->node);
4057 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4061 q_vector->adapter = adapter;
4062 if (q_vector->txr_count && !q_vector->rxr_count)
4063 q_vector->eitr = adapter->tx_eitr_param;
4065 q_vector->eitr = adapter->rx_eitr_param;
4066 q_vector->v_idx = q_idx;
4067 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4068 adapter->q_vector[q_idx] = q_vector;
4076 q_vector = adapter->q_vector[q_idx];
4077 netif_napi_del(&q_vector->napi);
4079 adapter->q_vector[q_idx] = NULL;
4085 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4086 * @adapter: board private structure to initialize
4088 * This function frees the memory allocated to the q_vectors. In addition if
4089 * NAPI is enabled it will delete any references to the NAPI struct prior
4090 * to freeing the q_vector.
4092 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4094 int q_idx, num_q_vectors;
4096 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4097 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4101 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4102 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4103 adapter->q_vector[q_idx] = NULL;
4104 netif_napi_del(&q_vector->napi);
4109 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4111 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4112 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4113 pci_disable_msix(adapter->pdev);
4114 kfree(adapter->msix_entries);
4115 adapter->msix_entries = NULL;
4116 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4117 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4118 pci_disable_msi(adapter->pdev);
4124 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4125 * @adapter: board private structure to initialize
4127 * We determine which interrupt scheme to use based on...
4128 * - Kernel support (MSI, MSI-X)
4129 * - which can be user-defined (via MODULE_PARAM)
4130 * - Hardware queue count (num_*_queues)
4131 * - defined by miscellaneous hardware support/features (RSS, etc.)
4133 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4137 /* Number of supported queues */
4138 ixgbe_set_num_queues(adapter);
4140 err = ixgbe_set_interrupt_capability(adapter);
4142 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
4143 goto err_set_interrupt;
4146 err = ixgbe_alloc_q_vectors(adapter);
4148 DPRINTK(PROBE, ERR, "Unable to allocate memory for queue "
4150 goto err_alloc_q_vectors;
4153 err = ixgbe_alloc_queues(adapter);
4155 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
4156 goto err_alloc_queues;
4159 DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
4160 "Tx Queue count = %u\n",
4161 (adapter->num_rx_queues > 1) ? "Enabled" :
4162 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
4164 set_bit(__IXGBE_DOWN, &adapter->state);
4169 ixgbe_free_q_vectors(adapter);
4170 err_alloc_q_vectors:
4171 ixgbe_reset_interrupt_capability(adapter);
4177 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4178 * @adapter: board private structure to clear interrupt scheme on
4180 * We go through and clear interrupt specific resources and reset the structure
4181 * to pre-load conditions
4183 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4187 for (i = 0; i < adapter->num_tx_queues; i++) {
4188 kfree(adapter->tx_ring[i]);
4189 adapter->tx_ring[i] = NULL;
4191 for (i = 0; i < adapter->num_rx_queues; i++) {
4192 kfree(adapter->rx_ring[i]);
4193 adapter->rx_ring[i] = NULL;
4196 ixgbe_free_q_vectors(adapter);
4197 ixgbe_reset_interrupt_capability(adapter);
4201 * ixgbe_sfp_timer - worker thread to find a missing module
4202 * @data: pointer to our adapter struct
4204 static void ixgbe_sfp_timer(unsigned long data)
4206 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4209 * Do the sfp_timer outside of interrupt context due to the
4210 * delays that sfp+ detection requires
4212 schedule_work(&adapter->sfp_task);
4216 * ixgbe_sfp_task - worker thread to find a missing module
4217 * @work: pointer to work_struct containing our data
4219 static void ixgbe_sfp_task(struct work_struct *work)
4221 struct ixgbe_adapter *adapter = container_of(work,
4222 struct ixgbe_adapter,
4224 struct ixgbe_hw *hw = &adapter->hw;
4226 if ((hw->phy.type == ixgbe_phy_nl) &&
4227 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4228 s32 ret = hw->phy.ops.identify_sfp(hw);
4229 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4231 ret = hw->phy.ops.reset(hw);
4232 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4233 dev_err(&adapter->pdev->dev, "failed to initialize "
4234 "because an unsupported SFP+ module type "
4236 "Reload the driver after installing a "
4237 "supported module.\n");
4238 unregister_netdev(adapter->netdev);
4240 DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
4243 /* don't need this routine any more */
4244 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4248 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4249 mod_timer(&adapter->sfp_timer,
4250 round_jiffies(jiffies + (2 * HZ)));
4254 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4255 * @adapter: board private structure to initialize
4257 * ixgbe_sw_init initializes the Adapter private data structure.
4258 * Fields are initialized based on PCI device information and
4259 * OS network device settings (MTU size).
4261 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4263 struct ixgbe_hw *hw = &adapter->hw;
4264 struct pci_dev *pdev = adapter->pdev;
4265 struct net_device *dev = adapter->netdev;
4267 #ifdef CONFIG_IXGBE_DCB
4269 struct tc_configuration *tc;
4272 /* PCI config space info */
4274 hw->vendor_id = pdev->vendor;
4275 hw->device_id = pdev->device;
4276 hw->revision_id = pdev->revision;
4277 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4278 hw->subsystem_device_id = pdev->subsystem_device;
4280 /* Set capability flags */
4281 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4282 adapter->ring_feature[RING_F_RSS].indices = rss;
4283 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4284 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4285 if (hw->mac.type == ixgbe_mac_82598EB) {
4286 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4287 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4288 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4289 } else if (hw->mac.type == ixgbe_mac_82599EB) {
4290 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4291 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4292 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4293 if (dev->features & NETIF_F_NTUPLE) {
4294 /* Flow Director perfect filter enabled */
4295 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4296 adapter->atr_sample_rate = 0;
4297 spin_lock_init(&adapter->fdir_perfect_lock);
4299 /* Flow Director hash filters enabled */
4300 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4301 adapter->atr_sample_rate = 20;
4303 adapter->ring_feature[RING_F_FDIR].indices =
4304 IXGBE_MAX_FDIR_INDICES;
4305 adapter->fdir_pballoc = 0;
4307 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4308 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4309 adapter->ring_feature[RING_F_FCOE].indices = 0;
4310 #ifdef CONFIG_IXGBE_DCB
4311 /* Default traffic class to use for FCoE */
4312 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4314 #endif /* IXGBE_FCOE */
4317 #ifdef CONFIG_IXGBE_DCB
4318 /* Configure DCB traffic classes */
4319 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4320 tc = &adapter->dcb_cfg.tc_config[j];
4321 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4322 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4323 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4324 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4325 tc->dcb_pfc = pfc_disabled;
4327 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4328 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4329 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4330 adapter->dcb_cfg.pfc_mode_enable = false;
4331 adapter->dcb_cfg.round_robin_enable = false;
4332 adapter->dcb_set_bitmap = 0x00;
4333 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4334 adapter->ring_feature[RING_F_DCB].indices);
4338 /* default flow control settings */
4339 hw->fc.requested_mode = ixgbe_fc_full;
4340 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4342 adapter->last_lfc_mode = hw->fc.current_mode;
4344 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
4345 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
4346 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4347 hw->fc.send_xon = true;
4348 hw->fc.disable_fc_autoneg = false;
4350 /* enable itr by default in dynamic mode */
4351 adapter->rx_itr_setting = 1;
4352 adapter->rx_eitr_param = 20000;
4353 adapter->tx_itr_setting = 1;
4354 adapter->tx_eitr_param = 10000;
4356 /* set defaults for eitr in MegaBytes */
4357 adapter->eitr_low = 10;
4358 adapter->eitr_high = 20;
4360 /* set default ring sizes */
4361 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4362 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4364 /* initialize eeprom parameters */
4365 if (ixgbe_init_eeprom_params_generic(hw)) {
4366 dev_err(&pdev->dev, "EEPROM initialization failed\n");
4370 /* enable rx csum by default */
4371 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4373 /* get assigned NUMA node */
4374 adapter->node = dev_to_node(&pdev->dev);
4376 set_bit(__IXGBE_DOWN, &adapter->state);
4382 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4383 * @adapter: board private structure
4384 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4386 * Return 0 on success, negative on failure
4388 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4389 struct ixgbe_ring *tx_ring)
4391 struct pci_dev *pdev = adapter->pdev;
4394 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4395 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4396 if (!tx_ring->tx_buffer_info)
4397 tx_ring->tx_buffer_info = vmalloc(size);
4398 if (!tx_ring->tx_buffer_info)
4400 memset(tx_ring->tx_buffer_info, 0, size);
4402 /* round up to nearest 4K */
4403 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4404 tx_ring->size = ALIGN(tx_ring->size, 4096);
4406 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
4411 tx_ring->next_to_use = 0;
4412 tx_ring->next_to_clean = 0;
4413 tx_ring->work_limit = tx_ring->count;
4417 vfree(tx_ring->tx_buffer_info);
4418 tx_ring->tx_buffer_info = NULL;
4419 DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
4420 "descriptor ring\n");
4425 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4426 * @adapter: board private structure
4428 * If this function returns with an error, then it's possible one or
4429 * more of the rings is populated (while the rest are not). It is the
4430 * callers duty to clean those orphaned rings.
4432 * Return 0 on success, negative on failure
4434 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4438 for (i = 0; i < adapter->num_tx_queues; i++) {
4439 err = ixgbe_setup_tx_resources(adapter, adapter->tx_ring[i]);
4442 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
4450 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4451 * @adapter: board private structure
4452 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4454 * Returns 0 on success, negative on failure
4456 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4457 struct ixgbe_ring *rx_ring)
4459 struct pci_dev *pdev = adapter->pdev;
4462 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4463 rx_ring->rx_buffer_info = vmalloc_node(size, adapter->node);
4464 if (!rx_ring->rx_buffer_info)
4465 rx_ring->rx_buffer_info = vmalloc(size);
4466 if (!rx_ring->rx_buffer_info) {
4468 "vmalloc allocation failed for the rx desc ring\n");
4471 memset(rx_ring->rx_buffer_info, 0, size);
4473 /* Round up to nearest 4K */
4474 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4475 rx_ring->size = ALIGN(rx_ring->size, 4096);
4477 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
4479 if (!rx_ring->desc) {
4481 "Memory allocation failed for the rx desc ring\n");
4482 vfree(rx_ring->rx_buffer_info);
4486 rx_ring->next_to_clean = 0;
4487 rx_ring->next_to_use = 0;
4496 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4497 * @adapter: board private structure
4499 * If this function returns with an error, then it's possible one or
4500 * more of the rings is populated (while the rest are not). It is the
4501 * callers duty to clean those orphaned rings.
4503 * Return 0 on success, negative on failure
4506 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4510 for (i = 0; i < adapter->num_rx_queues; i++) {
4511 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
4514 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
4522 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4523 * @adapter: board private structure
4524 * @tx_ring: Tx descriptor ring for a specific queue
4526 * Free all transmit software resources
4528 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4529 struct ixgbe_ring *tx_ring)
4531 struct pci_dev *pdev = adapter->pdev;
4533 ixgbe_clean_tx_ring(adapter, tx_ring);
4535 vfree(tx_ring->tx_buffer_info);
4536 tx_ring->tx_buffer_info = NULL;
4538 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
4540 tx_ring->desc = NULL;
4544 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4545 * @adapter: board private structure
4547 * Free all transmit software resources
4549 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4553 for (i = 0; i < adapter->num_tx_queues; i++)
4554 if (adapter->tx_ring[i]->desc)
4555 ixgbe_free_tx_resources(adapter, adapter->tx_ring[i]);
4559 * ixgbe_free_rx_resources - Free Rx Resources
4560 * @adapter: board private structure
4561 * @rx_ring: ring to clean the resources from
4563 * Free all receive software resources
4565 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
4566 struct ixgbe_ring *rx_ring)
4568 struct pci_dev *pdev = adapter->pdev;
4570 ixgbe_clean_rx_ring(adapter, rx_ring);
4572 vfree(rx_ring->rx_buffer_info);
4573 rx_ring->rx_buffer_info = NULL;
4575 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
4577 rx_ring->desc = NULL;
4581 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4582 * @adapter: board private structure
4584 * Free all receive software resources
4586 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4590 for (i = 0; i < adapter->num_rx_queues; i++)
4591 if (adapter->rx_ring[i]->desc)
4592 ixgbe_free_rx_resources(adapter, adapter->rx_ring[i]);
4596 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4597 * @netdev: network interface device structure
4598 * @new_mtu: new value for maximum frame size
4600 * Returns 0 on success, negative on failure
4602 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4604 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4605 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4607 /* MTU < 68 is an error and causes problems on some kernels */
4608 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4611 DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
4612 netdev->mtu, new_mtu);
4613 /* must set new MTU before calling down or up */
4614 netdev->mtu = new_mtu;
4616 if (netif_running(netdev))
4617 ixgbe_reinit_locked(adapter);
4623 * ixgbe_open - Called when a network interface is made active
4624 * @netdev: network interface device structure
4626 * Returns 0 on success, negative value on failure
4628 * The open entry point is called when a network interface is made
4629 * active by the system (IFF_UP). At this point all resources needed
4630 * for transmit and receive operations are allocated, the interrupt
4631 * handler is registered with the OS, the watchdog timer is started,
4632 * and the stack is notified that the interface is ready.
4634 static int ixgbe_open(struct net_device *netdev)
4636 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4639 /* disallow open during test */
4640 if (test_bit(__IXGBE_TESTING, &adapter->state))
4643 netif_carrier_off(netdev);
4645 /* allocate transmit descriptors */
4646 err = ixgbe_setup_all_tx_resources(adapter);
4650 /* allocate receive descriptors */
4651 err = ixgbe_setup_all_rx_resources(adapter);
4655 ixgbe_configure(adapter);
4657 err = ixgbe_request_irq(adapter);
4661 err = ixgbe_up_complete(adapter);
4665 netif_tx_start_all_queues(netdev);
4670 ixgbe_release_hw_control(adapter);
4671 ixgbe_free_irq(adapter);
4674 ixgbe_free_all_rx_resources(adapter);
4676 ixgbe_free_all_tx_resources(adapter);
4677 ixgbe_reset(adapter);
4683 * ixgbe_close - Disables a network interface
4684 * @netdev: network interface device structure
4686 * Returns 0, this is not allowed to fail
4688 * The close entry point is called when an interface is de-activated
4689 * by the OS. The hardware is still under the drivers control, but
4690 * needs to be disabled. A global MAC reset is issued to stop the
4691 * hardware, and all transmit and receive resources are freed.
4693 static int ixgbe_close(struct net_device *netdev)
4695 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4697 ixgbe_down(adapter);
4698 ixgbe_free_irq(adapter);
4700 ixgbe_free_all_tx_resources(adapter);
4701 ixgbe_free_all_rx_resources(adapter);
4703 ixgbe_release_hw_control(adapter);
4709 static int ixgbe_resume(struct pci_dev *pdev)
4711 struct net_device *netdev = pci_get_drvdata(pdev);
4712 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4715 pci_set_power_state(pdev, PCI_D0);
4716 pci_restore_state(pdev);
4718 * pci_restore_state clears dev->state_saved so call
4719 * pci_save_state to restore it.
4721 pci_save_state(pdev);
4723 err = pci_enable_device_mem(pdev);
4725 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
4729 pci_set_master(pdev);
4731 pci_wake_from_d3(pdev, false);
4733 err = ixgbe_init_interrupt_scheme(adapter);
4735 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
4740 ixgbe_reset(adapter);
4742 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4744 if (netif_running(netdev)) {
4745 err = ixgbe_open(adapter->netdev);
4750 netif_device_attach(netdev);
4754 #endif /* CONFIG_PM */
4756 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4758 struct net_device *netdev = pci_get_drvdata(pdev);
4759 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4760 struct ixgbe_hw *hw = &adapter->hw;
4762 u32 wufc = adapter->wol;
4767 netif_device_detach(netdev);
4769 if (netif_running(netdev)) {
4770 ixgbe_down(adapter);
4771 ixgbe_free_irq(adapter);
4772 ixgbe_free_all_tx_resources(adapter);
4773 ixgbe_free_all_rx_resources(adapter);
4775 ixgbe_clear_interrupt_scheme(adapter);
4778 retval = pci_save_state(pdev);
4784 ixgbe_set_rx_mode(netdev);
4786 /* turn on all-multi mode if wake on multicast is enabled */
4787 if (wufc & IXGBE_WUFC_MC) {
4788 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4789 fctrl |= IXGBE_FCTRL_MPE;
4790 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4793 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4794 ctrl |= IXGBE_CTRL_GIO_DIS;
4795 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4797 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4799 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4800 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4803 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
4804 pci_wake_from_d3(pdev, true);
4806 pci_wake_from_d3(pdev, false);
4808 *enable_wake = !!wufc;
4810 ixgbe_release_hw_control(adapter);
4812 pci_disable_device(pdev);
4818 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4823 retval = __ixgbe_shutdown(pdev, &wake);
4828 pci_prepare_to_sleep(pdev);
4830 pci_wake_from_d3(pdev, false);
4831 pci_set_power_state(pdev, PCI_D3hot);
4836 #endif /* CONFIG_PM */
4838 static void ixgbe_shutdown(struct pci_dev *pdev)
4842 __ixgbe_shutdown(pdev, &wake);
4844 if (system_state == SYSTEM_POWER_OFF) {
4845 pci_wake_from_d3(pdev, wake);
4846 pci_set_power_state(pdev, PCI_D3hot);
4851 * ixgbe_update_stats - Update the board statistics counters.
4852 * @adapter: board private structure
4854 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4856 struct net_device *netdev = adapter->netdev;
4857 struct ixgbe_hw *hw = &adapter->hw;
4859 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4860 u64 non_eop_descs = 0, restart_queue = 0;
4862 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
4865 for (i = 0; i < 16; i++)
4866 adapter->hw_rx_no_dma_resources +=
4867 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4868 for (i = 0; i < adapter->num_rx_queues; i++) {
4869 rsc_count += adapter->rx_ring[i]->rsc_count;
4870 rsc_flush += adapter->rx_ring[i]->rsc_flush;
4872 adapter->rsc_total_count = rsc_count;
4873 adapter->rsc_total_flush = rsc_flush;
4876 /* gather some stats to the adapter struct that are per queue */
4877 for (i = 0; i < adapter->num_tx_queues; i++)
4878 restart_queue += adapter->tx_ring[i]->restart_queue;
4879 adapter->restart_queue = restart_queue;
4881 for (i = 0; i < adapter->num_rx_queues; i++)
4882 non_eop_descs += adapter->rx_ring[i]->non_eop_descs;
4883 adapter->non_eop_descs = non_eop_descs;
4885 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
4886 for (i = 0; i < 8; i++) {
4887 /* for packet buffers not used, the register should read 0 */
4888 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
4890 adapter->stats.mpc[i] += mpc;
4891 total_mpc += adapter->stats.mpc[i];
4892 if (hw->mac.type == ixgbe_mac_82598EB)
4893 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
4894 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
4895 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
4896 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
4897 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
4898 if (hw->mac.type == ixgbe_mac_82599EB) {
4899 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4900 IXGBE_PXONRXCNT(i));
4901 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4902 IXGBE_PXOFFRXCNT(i));
4903 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4905 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4907 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4910 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
4912 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
4915 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
4916 /* work around hardware counting issue */
4917 adapter->stats.gprc -= missed_rx;
4919 /* 82598 hardware only has a 32 bit counter in the high register */
4920 if (hw->mac.type == ixgbe_mac_82599EB) {
4922 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
4923 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
4924 adapter->stats.gorc += (tmp << 32);
4925 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
4926 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
4927 adapter->stats.gotc += (tmp << 32);
4928 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
4929 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
4930 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
4931 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
4932 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
4933 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
4935 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
4936 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
4937 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
4938 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
4939 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
4940 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
4941 #endif /* IXGBE_FCOE */
4943 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
4944 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
4945 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
4946 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
4947 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
4949 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
4950 adapter->stats.bprc += bprc;
4951 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
4952 if (hw->mac.type == ixgbe_mac_82598EB)
4953 adapter->stats.mprc -= bprc;
4954 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
4955 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
4956 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
4957 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
4958 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
4959 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
4960 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
4961 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
4962 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
4963 adapter->stats.lxontxc += lxon;
4964 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
4965 adapter->stats.lxofftxc += lxoff;
4966 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4967 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
4968 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
4970 * 82598 errata - tx of flow control packets is included in tx counters
4972 xon_off_tot = lxon + lxoff;
4973 adapter->stats.gptc -= xon_off_tot;
4974 adapter->stats.mptc -= xon_off_tot;
4975 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
4976 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4977 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
4978 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
4979 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
4980 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
4981 adapter->stats.ptc64 -= xon_off_tot;
4982 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
4983 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
4984 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
4985 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
4986 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
4987 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
4989 /* Fill out the OS statistics structure */
4990 netdev->stats.multicast = adapter->stats.mprc;
4993 netdev->stats.rx_errors = adapter->stats.crcerrs +
4994 adapter->stats.rlec;
4995 netdev->stats.rx_dropped = 0;
4996 netdev->stats.rx_length_errors = adapter->stats.rlec;
4997 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4998 netdev->stats.rx_missed_errors = total_mpc;
5002 * ixgbe_watchdog - Timer Call-back
5003 * @data: pointer to adapter cast into an unsigned long
5005 static void ixgbe_watchdog(unsigned long data)
5007 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5008 struct ixgbe_hw *hw = &adapter->hw;
5013 * Do the watchdog outside of interrupt context due to the lovely
5014 * delays that some of the newer hardware requires
5017 if (test_bit(__IXGBE_DOWN, &adapter->state))
5018 goto watchdog_short_circuit;
5020 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5022 * for legacy and MSI interrupts don't set any bits
5023 * that are enabled for EIAM, because this operation
5024 * would set *both* EIMS and EICS for any bit in EIAM
5026 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5027 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5028 goto watchdog_reschedule;
5031 /* get one bit for every active tx/rx interrupt vector */
5032 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5033 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5034 if (qv->rxr_count || qv->txr_count)
5035 eics |= ((u64)1 << i);
5038 /* Cause software interrupt to ensure rx rings are cleaned */
5039 ixgbe_irq_rearm_queues(adapter, eics);
5041 watchdog_reschedule:
5042 /* Reset the timer */
5043 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5045 watchdog_short_circuit:
5046 schedule_work(&adapter->watchdog_task);
5050 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5051 * @work: pointer to work_struct containing our data
5053 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5055 struct ixgbe_adapter *adapter = container_of(work,
5056 struct ixgbe_adapter,
5057 multispeed_fiber_task);
5058 struct ixgbe_hw *hw = &adapter->hw;
5062 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5063 autoneg = hw->phy.autoneg_advertised;
5064 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5065 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5066 hw->mac.autotry_restart = false;
5067 if (hw->mac.ops.setup_link)
5068 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5069 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5070 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5074 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5075 * @work: pointer to work_struct containing our data
5077 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5079 struct ixgbe_adapter *adapter = container_of(work,
5080 struct ixgbe_adapter,
5081 sfp_config_module_task);
5082 struct ixgbe_hw *hw = &adapter->hw;
5085 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5087 /* Time for electrical oscillations to settle down */
5089 err = hw->phy.ops.identify_sfp(hw);
5091 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5092 dev_err(&adapter->pdev->dev, "failed to initialize because "
5093 "an unsupported SFP+ module type was detected.\n"
5094 "Reload the driver after installing a supported "
5096 unregister_netdev(adapter->netdev);
5099 hw->mac.ops.setup_sfp(hw);
5101 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5102 /* This will also work for DA Twinax connections */
5103 schedule_work(&adapter->multispeed_fiber_task);
5104 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5108 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5109 * @work: pointer to work_struct containing our data
5111 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5113 struct ixgbe_adapter *adapter = container_of(work,
5114 struct ixgbe_adapter,
5116 struct ixgbe_hw *hw = &adapter->hw;
5119 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5120 for (i = 0; i < adapter->num_tx_queues; i++)
5121 set_bit(__IXGBE_FDIR_INIT_DONE,
5122 &(adapter->tx_ring[i]->reinit_state));
5124 DPRINTK(PROBE, ERR, "failed to finish FDIR re-initialization, "
5125 "ignored adding FDIR ATR filters\n");
5127 /* Done FDIR Re-initialization, enable transmits */
5128 netif_tx_start_all_queues(adapter->netdev);
5131 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5134 * ixgbe_watchdog_task - worker thread to bring link up
5135 * @work: pointer to work_struct containing our data
5137 static void ixgbe_watchdog_task(struct work_struct *work)
5139 struct ixgbe_adapter *adapter = container_of(work,
5140 struct ixgbe_adapter,
5142 struct net_device *netdev = adapter->netdev;
5143 struct ixgbe_hw *hw = &adapter->hw;
5147 struct ixgbe_ring *tx_ring;
5148 int some_tx_pending = 0;
5150 mutex_lock(&ixgbe_watchdog_lock);
5152 link_up = adapter->link_up;
5153 link_speed = adapter->link_speed;
5155 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5156 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5159 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5160 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5161 hw->mac.ops.fc_enable(hw, i);
5163 hw->mac.ops.fc_enable(hw, 0);
5166 hw->mac.ops.fc_enable(hw, 0);
5171 time_after(jiffies, (adapter->link_check_timeout +
5172 IXGBE_TRY_LINK_TIMEOUT))) {
5173 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5174 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5176 adapter->link_up = link_up;
5177 adapter->link_speed = link_speed;
5181 if (!netif_carrier_ok(netdev)) {
5182 bool flow_rx, flow_tx;
5184 if (hw->mac.type == ixgbe_mac_82599EB) {
5185 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5186 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5187 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5188 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5190 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5191 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5192 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5193 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5196 printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
5197 "Flow Control: %s\n",
5199 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5201 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5202 "1 Gbps" : "unknown speed")),
5203 ((flow_rx && flow_tx) ? "RX/TX" :
5205 (flow_tx ? "TX" : "None"))));
5207 netif_carrier_on(netdev);
5209 /* Force detection of hung controller */
5210 adapter->detect_tx_hung = true;
5213 adapter->link_up = false;
5214 adapter->link_speed = 0;
5215 if (netif_carrier_ok(netdev)) {
5216 printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
5218 netif_carrier_off(netdev);
5222 if (!netif_carrier_ok(netdev)) {
5223 for (i = 0; i < adapter->num_tx_queues; i++) {
5224 tx_ring = adapter->tx_ring[i];
5225 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5226 some_tx_pending = 1;
5231 if (some_tx_pending) {
5232 /* We've lost link, so the controller stops DMA,
5233 * but we've got queued Tx work that's never going
5234 * to get done, so reset controller to flush Tx.
5235 * (Do the reset outside of interrupt context).
5237 schedule_work(&adapter->reset_task);
5241 ixgbe_update_stats(adapter);
5242 mutex_unlock(&ixgbe_watchdog_lock);
5245 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5246 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5247 u32 tx_flags, u8 *hdr_len)
5249 struct ixgbe_adv_tx_context_desc *context_desc;
5252 struct ixgbe_tx_buffer *tx_buffer_info;
5253 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5254 u32 mss_l4len_idx, l4len;
5256 if (skb_is_gso(skb)) {
5257 if (skb_header_cloned(skb)) {
5258 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5262 l4len = tcp_hdrlen(skb);
5265 if (skb->protocol == htons(ETH_P_IP)) {
5266 struct iphdr *iph = ip_hdr(skb);
5269 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5273 } else if (skb_is_gso_v6(skb)) {
5274 ipv6_hdr(skb)->payload_len = 0;
5275 tcp_hdr(skb)->check =
5276 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5277 &ipv6_hdr(skb)->daddr,
5281 i = tx_ring->next_to_use;
5283 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5284 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5286 /* VLAN MACLEN IPLEN */
5287 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5289 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5290 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5291 IXGBE_ADVTXD_MACLEN_SHIFT);
5292 *hdr_len += skb_network_offset(skb);
5294 (skb_transport_header(skb) - skb_network_header(skb));
5296 (skb_transport_header(skb) - skb_network_header(skb));
5297 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5298 context_desc->seqnum_seed = 0;
5300 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5301 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5302 IXGBE_ADVTXD_DTYP_CTXT);
5304 if (skb->protocol == htons(ETH_P_IP))
5305 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5306 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5307 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5311 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5312 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5313 /* use index 1 for TSO */
5314 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5315 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5317 tx_buffer_info->time_stamp = jiffies;
5318 tx_buffer_info->next_to_watch = i;
5321 if (i == tx_ring->count)
5323 tx_ring->next_to_use = i;
5330 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5331 struct ixgbe_ring *tx_ring,
5332 struct sk_buff *skb, u32 tx_flags)
5334 struct ixgbe_adv_tx_context_desc *context_desc;
5336 struct ixgbe_tx_buffer *tx_buffer_info;
5337 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5339 if (skb->ip_summed == CHECKSUM_PARTIAL ||
5340 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5341 i = tx_ring->next_to_use;
5342 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5343 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5345 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5347 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5348 vlan_macip_lens |= (skb_network_offset(skb) <<
5349 IXGBE_ADVTXD_MACLEN_SHIFT);
5350 if (skb->ip_summed == CHECKSUM_PARTIAL)
5351 vlan_macip_lens |= (skb_transport_header(skb) -
5352 skb_network_header(skb));
5354 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5355 context_desc->seqnum_seed = 0;
5357 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5358 IXGBE_ADVTXD_DTYP_CTXT);
5360 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5363 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
5364 const struct vlan_ethhdr *vhdr =
5365 (const struct vlan_ethhdr *)skb->data;
5367 protocol = vhdr->h_vlan_encapsulated_proto;
5369 protocol = skb->protocol;
5373 case cpu_to_be16(ETH_P_IP):
5374 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5375 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5377 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5378 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
5380 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5382 case cpu_to_be16(ETH_P_IPV6):
5383 /* XXX what about other V6 headers?? */
5384 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5386 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5387 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
5389 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5392 if (unlikely(net_ratelimit())) {
5393 DPRINTK(PROBE, WARNING,
5394 "partial checksum but proto=%x!\n",
5401 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5402 /* use index zero for tx checksum offload */
5403 context_desc->mss_l4len_idx = 0;
5405 tx_buffer_info->time_stamp = jiffies;
5406 tx_buffer_info->next_to_watch = i;
5409 if (i == tx_ring->count)
5411 tx_ring->next_to_use = i;
5419 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5420 struct ixgbe_ring *tx_ring,
5421 struct sk_buff *skb, u32 tx_flags,
5424 struct pci_dev *pdev = adapter->pdev;
5425 struct ixgbe_tx_buffer *tx_buffer_info;
5427 unsigned int total = skb->len;
5428 unsigned int offset = 0, size, count = 0, i;
5429 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5432 i = tx_ring->next_to_use;
5434 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5435 /* excluding fcoe_crc_eof for FCoE */
5436 total -= sizeof(struct fcoe_crc_eof);
5438 len = min(skb_headlen(skb), total);
5440 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5441 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5443 tx_buffer_info->length = size;
5444 tx_buffer_info->mapped_as_page = false;
5445 tx_buffer_info->dma = pci_map_single(pdev,
5447 size, PCI_DMA_TODEVICE);
5448 if (pci_dma_mapping_error(pdev, tx_buffer_info->dma))
5450 tx_buffer_info->time_stamp = jiffies;
5451 tx_buffer_info->next_to_watch = i;
5460 if (i == tx_ring->count)
5465 for (f = 0; f < nr_frags; f++) {
5466 struct skb_frag_struct *frag;
5468 frag = &skb_shinfo(skb)->frags[f];
5469 len = min((unsigned int)frag->size, total);
5470 offset = frag->page_offset;
5474 if (i == tx_ring->count)
5477 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5478 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5480 tx_buffer_info->length = size;
5481 tx_buffer_info->dma = pci_map_page(adapter->pdev,
5485 tx_buffer_info->mapped_as_page = true;
5486 if (pci_dma_mapping_error(pdev, tx_buffer_info->dma))
5488 tx_buffer_info->time_stamp = jiffies;
5489 tx_buffer_info->next_to_watch = i;
5500 tx_ring->tx_buffer_info[i].skb = skb;
5501 tx_ring->tx_buffer_info[first].next_to_watch = i;
5506 dev_err(&pdev->dev, "TX DMA map failed\n");
5508 /* clear timestamp and dma mappings for failed tx_buffer_info map */
5509 tx_buffer_info->dma = 0;
5510 tx_buffer_info->time_stamp = 0;
5511 tx_buffer_info->next_to_watch = 0;
5515 /* clear timestamp and dma mappings for remaining portion of packet */
5518 i += tx_ring->count;
5520 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5521 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
5527 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
5528 struct ixgbe_ring *tx_ring,
5529 int tx_flags, int count, u32 paylen, u8 hdr_len)
5531 union ixgbe_adv_tx_desc *tx_desc = NULL;
5532 struct ixgbe_tx_buffer *tx_buffer_info;
5533 u32 olinfo_status = 0, cmd_type_len = 0;
5535 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
5537 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
5539 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
5541 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5542 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
5544 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
5545 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5547 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5548 IXGBE_ADVTXD_POPTS_SHIFT;
5550 /* use index 1 context for tso */
5551 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5552 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5553 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
5554 IXGBE_ADVTXD_POPTS_SHIFT;
5556 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5557 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5558 IXGBE_ADVTXD_POPTS_SHIFT;
5560 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5561 olinfo_status |= IXGBE_ADVTXD_CC;
5562 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5563 if (tx_flags & IXGBE_TX_FLAGS_FSO)
5564 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5567 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
5569 i = tx_ring->next_to_use;
5571 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5572 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
5573 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
5574 tx_desc->read.cmd_type_len =
5575 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
5576 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5578 if (i == tx_ring->count)
5582 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
5585 * Force memory writes to complete before letting h/w
5586 * know there are new descriptors to fetch. (Only
5587 * applicable for weak-ordered memory model archs,
5592 tx_ring->next_to_use = i;
5593 writel(i, adapter->hw.hw_addr + tx_ring->tail);
5596 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5597 int queue, u32 tx_flags)
5599 /* Right now, we support IPv4 only */
5600 struct ixgbe_atr_input atr_input;
5602 struct iphdr *iph = ip_hdr(skb);
5603 struct ethhdr *eth = (struct ethhdr *)skb->data;
5604 u16 vlan_id, src_port, dst_port, flex_bytes;
5605 u32 src_ipv4_addr, dst_ipv4_addr;
5608 /* check if we're UDP or TCP */
5609 if (iph->protocol == IPPROTO_TCP) {
5611 src_port = th->source;
5612 dst_port = th->dest;
5613 l4type |= IXGBE_ATR_L4TYPE_TCP;
5614 /* l4type IPv4 type is 0, no need to assign */
5616 /* Unsupported L4 header, just bail here */
5620 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
5622 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
5623 IXGBE_TX_FLAGS_VLAN_SHIFT;
5624 src_ipv4_addr = iph->saddr;
5625 dst_ipv4_addr = iph->daddr;
5626 flex_bytes = eth->h_proto;
5628 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
5629 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
5630 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
5631 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
5632 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
5633 /* src and dst are inverted, think how the receiver sees them */
5634 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
5635 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
5637 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
5638 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
5641 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
5642 struct ixgbe_ring *tx_ring, int size)
5644 netif_stop_subqueue(netdev, tx_ring->queue_index);
5645 /* Herbert's original patch had:
5646 * smp_mb__after_netif_stop_queue();
5647 * but since that doesn't exist yet, just open code it. */
5650 /* We need to check again in a case another CPU has just
5651 * made room available. */
5652 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
5655 /* A reprieve! - use start_queue because it doesn't call schedule */
5656 netif_start_subqueue(netdev, tx_ring->queue_index);
5657 ++tx_ring->restart_queue;
5661 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
5662 struct ixgbe_ring *tx_ring, int size)
5664 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
5666 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
5669 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
5671 struct ixgbe_adapter *adapter = netdev_priv(dev);
5672 int txq = smp_processor_id();
5674 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5675 while (unlikely(txq >= dev->real_num_tx_queues))
5676 txq -= dev->real_num_tx_queues;
5681 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5682 ((skb->protocol == htons(ETH_P_FCOE)) ||
5683 (skb->protocol == htons(ETH_P_FIP)))) {
5684 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
5685 txq += adapter->ring_feature[RING_F_FCOE].mask;
5689 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5690 if (skb->priority == TC_PRIO_CONTROL)
5691 txq = adapter->ring_feature[RING_F_DCB].indices-1;
5693 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
5698 return skb_tx_hash(dev, skb);
5701 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
5702 struct net_device *netdev)
5704 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5705 struct ixgbe_ring *tx_ring;
5706 struct netdev_queue *txq;
5708 unsigned int tx_flags = 0;
5714 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
5715 tx_flags |= vlan_tx_tag_get(skb);
5716 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5717 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
5718 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
5720 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5721 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5722 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5723 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
5724 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5725 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5728 tx_ring = adapter->tx_ring[skb->queue_mapping];
5731 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
5732 #ifdef CONFIG_IXGBE_DCB
5733 /* for FCoE with DCB, we force the priority to what
5734 * was specified by the switch */
5735 if ((skb->protocol == htons(ETH_P_FCOE)) ||
5736 (skb->protocol == htons(ETH_P_FIP))) {
5737 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
5738 << IXGBE_TX_FLAGS_VLAN_SHIFT);
5739 tx_flags |= ((adapter->fcoe.up << 13)
5740 << IXGBE_TX_FLAGS_VLAN_SHIFT);
5743 /* flag for FCoE offloads */
5744 if (skb->protocol == htons(ETH_P_FCOE))
5745 tx_flags |= IXGBE_TX_FLAGS_FCOE;
5749 /* four things can cause us to need a context descriptor */
5750 if (skb_is_gso(skb) ||
5751 (skb->ip_summed == CHECKSUM_PARTIAL) ||
5752 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
5753 (tx_flags & IXGBE_TX_FLAGS_FCOE))
5756 count += TXD_USE_COUNT(skb_headlen(skb));
5757 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5758 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5760 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
5762 return NETDEV_TX_BUSY;
5765 first = tx_ring->next_to_use;
5766 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5768 /* setup tx offload for FCoE */
5769 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5771 dev_kfree_skb_any(skb);
5772 return NETDEV_TX_OK;
5775 tx_flags |= IXGBE_TX_FLAGS_FSO;
5776 #endif /* IXGBE_FCOE */
5778 if (skb->protocol == htons(ETH_P_IP))
5779 tx_flags |= IXGBE_TX_FLAGS_IPV4;
5780 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5782 dev_kfree_skb_any(skb);
5783 return NETDEV_TX_OK;
5787 tx_flags |= IXGBE_TX_FLAGS_TSO;
5788 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
5789 (skb->ip_summed == CHECKSUM_PARTIAL))
5790 tx_flags |= IXGBE_TX_FLAGS_CSUM;
5793 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
5795 /* add the ATR filter if ATR is on */
5796 if (tx_ring->atr_sample_rate) {
5797 ++tx_ring->atr_count;
5798 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
5799 test_bit(__IXGBE_FDIR_INIT_DONE,
5800 &tx_ring->reinit_state)) {
5801 ixgbe_atr(adapter, skb, tx_ring->queue_index,
5803 tx_ring->atr_count = 0;
5806 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
5807 txq->tx_bytes += skb->len;
5809 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
5811 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
5814 dev_kfree_skb_any(skb);
5815 tx_ring->tx_buffer_info[first].time_stamp = 0;
5816 tx_ring->next_to_use = first;
5819 return NETDEV_TX_OK;
5823 * ixgbe_set_mac - Change the Ethernet Address of the NIC
5824 * @netdev: network interface device structure
5825 * @p: pointer to an address structure
5827 * Returns 0 on success, negative on failure
5829 static int ixgbe_set_mac(struct net_device *netdev, void *p)
5831 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5832 struct ixgbe_hw *hw = &adapter->hw;
5833 struct sockaddr *addr = p;
5835 if (!is_valid_ether_addr(addr->sa_data))
5836 return -EADDRNOTAVAIL;
5838 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
5839 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5841 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
5848 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
5850 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5851 struct ixgbe_hw *hw = &adapter->hw;
5855 if (prtad != hw->phy.mdio.prtad)
5857 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
5863 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
5864 u16 addr, u16 value)
5866 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5867 struct ixgbe_hw *hw = &adapter->hw;
5869 if (prtad != hw->phy.mdio.prtad)
5871 return hw->phy.ops.write_reg(hw, addr, devad, value);
5874 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
5876 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5878 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
5882 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
5884 * @netdev: network interface device structure
5886 * Returns non-zero on failure
5888 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
5891 struct ixgbe_adapter *adapter = netdev_priv(dev);
5892 struct ixgbe_mac_info *mac = &adapter->hw.mac;
5894 if (is_valid_ether_addr(mac->san_addr)) {
5896 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5903 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
5905 * @netdev: network interface device structure
5907 * Returns non-zero on failure
5909 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
5912 struct ixgbe_adapter *adapter = netdev_priv(dev);
5913 struct ixgbe_mac_info *mac = &adapter->hw.mac;
5915 if (is_valid_ether_addr(mac->san_addr)) {
5917 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5923 #ifdef CONFIG_NET_POLL_CONTROLLER
5925 * Polling 'interrupt' - used by things like netconsole to send skbs
5926 * without having to re-enable interrupts. It's not called while
5927 * the interrupt routine is executing.
5929 static void ixgbe_netpoll(struct net_device *netdev)
5931 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5934 /* if interface is down do nothing */
5935 if (test_bit(__IXGBE_DOWN, &adapter->state))
5938 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
5939 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5940 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5941 for (i = 0; i < num_q_vectors; i++) {
5942 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
5943 ixgbe_msix_clean_many(0, q_vector);
5946 ixgbe_intr(adapter->pdev->irq, netdev);
5948 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
5952 static const struct net_device_ops ixgbe_netdev_ops = {
5953 .ndo_open = ixgbe_open,
5954 .ndo_stop = ixgbe_close,
5955 .ndo_start_xmit = ixgbe_xmit_frame,
5956 .ndo_select_queue = ixgbe_select_queue,
5957 .ndo_set_rx_mode = ixgbe_set_rx_mode,
5958 .ndo_set_multicast_list = ixgbe_set_rx_mode,
5959 .ndo_validate_addr = eth_validate_addr,
5960 .ndo_set_mac_address = ixgbe_set_mac,
5961 .ndo_change_mtu = ixgbe_change_mtu,
5962 .ndo_tx_timeout = ixgbe_tx_timeout,
5963 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
5964 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
5965 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
5966 .ndo_do_ioctl = ixgbe_ioctl,
5967 #ifdef CONFIG_NET_POLL_CONTROLLER
5968 .ndo_poll_controller = ixgbe_netpoll,
5971 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
5972 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
5973 .ndo_fcoe_enable = ixgbe_fcoe_enable,
5974 .ndo_fcoe_disable = ixgbe_fcoe_disable,
5975 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
5976 #endif /* IXGBE_FCOE */
5979 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
5980 const struct ixgbe_info *ii)
5982 #ifdef CONFIG_PCI_IOV
5983 struct ixgbe_hw *hw = &adapter->hw;
5986 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
5989 /* The 82599 supports up to 64 VFs per physical function
5990 * but this implementation limits allocation to 63 so that
5991 * basic networking resources are still available to the
5994 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
5995 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
5996 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
5999 "Failed to enable PCI sriov: %d\n", err);
6002 /* If call to enable VFs succeeded then allocate memory
6003 * for per VF control structures.
6006 kcalloc(adapter->num_vfs,
6007 sizeof(struct vf_data_storage), GFP_KERNEL);
6008 if (adapter->vfinfo) {
6009 /* Now that we're sure SR-IOV is enabled
6010 * and memory allocated set up the mailbox parameters
6012 ixgbe_init_mbx_params_pf(hw);
6013 memcpy(&hw->mbx.ops, ii->mbx_ops,
6014 sizeof(hw->mbx.ops));
6016 /* Disable RSC when in SR-IOV mode */
6017 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6018 IXGBE_FLAG2_RSC_ENABLED);
6024 "Unable to allocate memory for VF "
6025 "Data Storage - SRIOV disabled\n");
6026 pci_disable_sriov(adapter->pdev);
6029 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6030 adapter->num_vfs = 0;
6031 #endif /* CONFIG_PCI_IOV */
6035 * ixgbe_probe - Device Initialization Routine
6036 * @pdev: PCI device information struct
6037 * @ent: entry in ixgbe_pci_tbl
6039 * Returns 0 on success, negative on failure
6041 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6042 * The OS initialization, configuring of the adapter private structure,
6043 * and a hardware reset occur.
6045 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6046 const struct pci_device_id *ent)
6048 struct net_device *netdev;
6049 struct ixgbe_adapter *adapter = NULL;
6050 struct ixgbe_hw *hw;
6051 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6052 static int cards_found;
6053 int i, err, pci_using_dac;
6054 unsigned int indices = num_possible_cpus();
6060 err = pci_enable_device_mem(pdev);
6064 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
6065 !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
6068 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
6070 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
6072 dev_err(&pdev->dev, "No usable DMA "
6073 "configuration, aborting\n");
6080 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6081 IORESOURCE_MEM), ixgbe_driver_name);
6084 "pci_request_selected_regions failed 0x%x\n", err);
6088 pci_enable_pcie_error_reporting(pdev);
6090 pci_set_master(pdev);
6091 pci_save_state(pdev);
6093 if (ii->mac == ixgbe_mac_82598EB)
6094 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6096 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6098 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6100 indices += min_t(unsigned int, num_possible_cpus(),
6101 IXGBE_MAX_FCOE_INDICES);
6103 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6106 goto err_alloc_etherdev;
6109 SET_NETDEV_DEV(netdev, &pdev->dev);
6111 pci_set_drvdata(pdev, netdev);
6112 adapter = netdev_priv(netdev);
6114 adapter->netdev = netdev;
6115 adapter->pdev = pdev;
6118 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6120 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6121 pci_resource_len(pdev, 0));
6127 for (i = 1; i <= 5; i++) {
6128 if (pci_resource_len(pdev, i) == 0)
6132 netdev->netdev_ops = &ixgbe_netdev_ops;
6133 ixgbe_set_ethtool_ops(netdev);
6134 netdev->watchdog_timeo = 5 * HZ;
6135 strcpy(netdev->name, pci_name(pdev));
6137 adapter->bd_number = cards_found;
6140 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6141 hw->mac.type = ii->mac;
6144 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6145 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6146 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6147 if (!(eec & (1 << 8)))
6148 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6151 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6152 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6153 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6154 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6155 hw->phy.mdio.mmds = 0;
6156 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6157 hw->phy.mdio.dev = netdev;
6158 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6159 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6161 /* set up this timer and work struct before calling get_invariants
6162 * which might start the timer
6164 init_timer(&adapter->sfp_timer);
6165 adapter->sfp_timer.function = &ixgbe_sfp_timer;
6166 adapter->sfp_timer.data = (unsigned long) adapter;
6168 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6170 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6171 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6173 /* a new SFP+ module arrival, called from GPI SDP2 context */
6174 INIT_WORK(&adapter->sfp_config_module_task,
6175 ixgbe_sfp_config_module_task);
6177 ii->get_invariants(hw);
6179 /* setup the private structure */
6180 err = ixgbe_sw_init(adapter);
6184 /* Make it possible the adapter to be woken up via WOL */
6185 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6186 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6189 * If there is a fan on this device and it has failed log the
6192 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6193 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6194 if (esdp & IXGBE_ESDP_SDP1)
6195 DPRINTK(PROBE, CRIT,
6196 "Fan has stopped, replace the adapter\n");
6199 /* reset_hw fills in the perm_addr as well */
6200 err = hw->mac.ops.reset_hw(hw);
6201 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6202 hw->mac.type == ixgbe_mac_82598EB) {
6204 * Start a kernel thread to watch for a module to arrive.
6205 * Only do this for 82598, since 82599 will generate
6206 * interrupts on module arrival.
6208 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6209 mod_timer(&adapter->sfp_timer,
6210 round_jiffies(jiffies + (2 * HZ)));
6212 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6213 dev_err(&adapter->pdev->dev, "failed to initialize because "
6214 "an unsupported SFP+ module type was detected.\n"
6215 "Reload the driver after installing a supported "
6219 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
6223 ixgbe_probe_vf(adapter, ii);
6225 netdev->features = NETIF_F_SG |
6227 NETIF_F_HW_VLAN_TX |
6228 NETIF_F_HW_VLAN_RX |
6229 NETIF_F_HW_VLAN_FILTER;
6231 netdev->features |= NETIF_F_IPV6_CSUM;
6232 netdev->features |= NETIF_F_TSO;
6233 netdev->features |= NETIF_F_TSO6;
6234 netdev->features |= NETIF_F_GRO;
6236 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6237 netdev->features |= NETIF_F_SCTP_CSUM;
6239 netdev->vlan_features |= NETIF_F_TSO;
6240 netdev->vlan_features |= NETIF_F_TSO6;
6241 netdev->vlan_features |= NETIF_F_IP_CSUM;
6242 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6243 netdev->vlan_features |= NETIF_F_SG;
6245 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6246 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6247 IXGBE_FLAG_DCB_ENABLED);
6248 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6249 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6251 #ifdef CONFIG_IXGBE_DCB
6252 netdev->dcbnl_ops = &dcbnl_ops;
6256 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6257 if (hw->mac.ops.get_device_caps) {
6258 hw->mac.ops.get_device_caps(hw, &device_caps);
6259 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6260 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6263 #endif /* IXGBE_FCOE */
6265 netdev->features |= NETIF_F_HIGHDMA;
6267 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6268 netdev->features |= NETIF_F_LRO;
6270 /* make sure the EEPROM is good */
6271 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6272 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
6277 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6278 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6280 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6281 dev_err(&pdev->dev, "invalid MAC address\n");
6286 /* power down the optics */
6287 if (hw->phy.multispeed_fiber)
6288 hw->mac.ops.disable_tx_laser(hw);
6290 init_timer(&adapter->watchdog_timer);
6291 adapter->watchdog_timer.function = &ixgbe_watchdog;
6292 adapter->watchdog_timer.data = (unsigned long)adapter;
6294 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6295 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6297 err = ixgbe_init_interrupt_scheme(adapter);
6301 switch (pdev->device) {
6302 case IXGBE_DEV_ID_82599_KX4:
6303 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
6304 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
6310 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6312 /* pick up the PCI bus settings for reporting later */
6313 hw->mac.ops.get_bus_info(hw);
6315 /* print bus type/speed/width info */
6316 dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
6317 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
6318 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
6319 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
6320 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
6321 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
6324 ixgbe_read_pba_num_generic(hw, &part_num);
6325 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
6326 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
6327 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
6328 (part_num >> 8), (part_num & 0xff));
6330 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
6331 hw->mac.type, hw->phy.type,
6332 (part_num >> 8), (part_num & 0xff));
6334 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
6335 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
6336 "this card is not sufficient for optimal "
6338 dev_warn(&pdev->dev, "For optimal performance a x8 "
6339 "PCI-Express slot is required.\n");
6342 /* save off EEPROM version number */
6343 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
6345 /* reset the hardware with the new settings */
6346 err = hw->mac.ops.start_hw(hw);
6348 if (err == IXGBE_ERR_EEPROM_VERSION) {
6349 /* We are running on a pre-production device, log a warning */
6350 dev_warn(&pdev->dev, "This device is a pre-production "
6351 "adapter/LOM. Please be aware there may be issues "
6352 "associated with your hardware. If you are "
6353 "experiencing problems please contact your Intel or "
6354 "hardware representative who provided you with this "
6357 strcpy(netdev->name, "eth%d");
6358 err = register_netdev(netdev);
6362 /* carrier off reporting is important to ethtool even BEFORE open */
6363 netif_carrier_off(netdev);
6365 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6366 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6367 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
6369 #ifdef CONFIG_IXGBE_DCA
6370 if (dca_add_requester(&pdev->dev) == 0) {
6371 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
6372 ixgbe_setup_dca(adapter);
6375 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6376 DPRINTK(PROBE, INFO, "IOV is enabled with %d VFs\n",
6378 for (i = 0; i < adapter->num_vfs; i++)
6379 ixgbe_vf_configuration(pdev, (i | 0x10000000));
6382 /* add san mac addr to netdev */
6383 ixgbe_add_sanmac_netdev(netdev);
6385 dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
6390 ixgbe_release_hw_control(adapter);
6391 ixgbe_clear_interrupt_scheme(adapter);
6394 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6395 ixgbe_disable_sriov(adapter);
6396 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6397 del_timer_sync(&adapter->sfp_timer);
6398 cancel_work_sync(&adapter->sfp_task);
6399 cancel_work_sync(&adapter->multispeed_fiber_task);
6400 cancel_work_sync(&adapter->sfp_config_module_task);
6401 iounmap(hw->hw_addr);
6403 free_netdev(netdev);
6405 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6409 pci_disable_device(pdev);
6414 * ixgbe_remove - Device Removal Routine
6415 * @pdev: PCI device information struct
6417 * ixgbe_remove is called by the PCI subsystem to alert the driver
6418 * that it should release a PCI device. The could be caused by a
6419 * Hot-Plug event, or because the driver is going to be removed from
6422 static void __devexit ixgbe_remove(struct pci_dev *pdev)
6424 struct net_device *netdev = pci_get_drvdata(pdev);
6425 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6427 set_bit(__IXGBE_DOWN, &adapter->state);
6428 /* clear the module not found bit to make sure the worker won't
6431 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6432 del_timer_sync(&adapter->watchdog_timer);
6434 del_timer_sync(&adapter->sfp_timer);
6435 cancel_work_sync(&adapter->watchdog_task);
6436 cancel_work_sync(&adapter->sfp_task);
6437 cancel_work_sync(&adapter->multispeed_fiber_task);
6438 cancel_work_sync(&adapter->sfp_config_module_task);
6439 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6440 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6441 cancel_work_sync(&adapter->fdir_reinit_task);
6442 flush_scheduled_work();
6444 #ifdef CONFIG_IXGBE_DCA
6445 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
6446 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
6447 dca_remove_requester(&pdev->dev);
6448 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
6453 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
6454 ixgbe_cleanup_fcoe(adapter);
6456 #endif /* IXGBE_FCOE */
6458 /* remove the added san mac */
6459 ixgbe_del_sanmac_netdev(netdev);
6461 if (netdev->reg_state == NETREG_REGISTERED)
6462 unregister_netdev(netdev);
6464 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6465 ixgbe_disable_sriov(adapter);
6467 ixgbe_clear_interrupt_scheme(adapter);
6469 ixgbe_release_hw_control(adapter);
6471 iounmap(adapter->hw.hw_addr);
6472 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6475 DPRINTK(PROBE, INFO, "complete\n");
6477 free_netdev(netdev);
6479 pci_disable_pcie_error_reporting(pdev);
6481 pci_disable_device(pdev);
6485 * ixgbe_io_error_detected - called when PCI error is detected
6486 * @pdev: Pointer to PCI device
6487 * @state: The current pci connection state
6489 * This function is called after a PCI bus error affecting
6490 * this device has been detected.
6492 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
6493 pci_channel_state_t state)
6495 struct net_device *netdev = pci_get_drvdata(pdev);
6496 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6498 netif_device_detach(netdev);
6500 if (state == pci_channel_io_perm_failure)
6501 return PCI_ERS_RESULT_DISCONNECT;
6503 if (netif_running(netdev))
6504 ixgbe_down(adapter);
6505 pci_disable_device(pdev);
6507 /* Request a slot reset. */
6508 return PCI_ERS_RESULT_NEED_RESET;
6512 * ixgbe_io_slot_reset - called after the pci bus has been reset.
6513 * @pdev: Pointer to PCI device
6515 * Restart the card from scratch, as if from a cold-boot.
6517 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
6519 struct net_device *netdev = pci_get_drvdata(pdev);
6520 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6521 pci_ers_result_t result;
6524 if (pci_enable_device_mem(pdev)) {
6526 "Cannot re-enable PCI device after reset.\n");
6527 result = PCI_ERS_RESULT_DISCONNECT;
6529 pci_set_master(pdev);
6530 pci_restore_state(pdev);
6531 pci_save_state(pdev);
6533 pci_wake_from_d3(pdev, false);
6535 ixgbe_reset(adapter);
6536 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6537 result = PCI_ERS_RESULT_RECOVERED;
6540 err = pci_cleanup_aer_uncorrect_error_status(pdev);
6543 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
6544 /* non-fatal, continue */
6551 * ixgbe_io_resume - called when traffic can start flowing again.
6552 * @pdev: Pointer to PCI device
6554 * This callback is called when the error recovery driver tells us that
6555 * its OK to resume normal operation.
6557 static void ixgbe_io_resume(struct pci_dev *pdev)
6559 struct net_device *netdev = pci_get_drvdata(pdev);
6560 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6562 if (netif_running(netdev)) {
6563 if (ixgbe_up(adapter)) {
6564 DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
6569 netif_device_attach(netdev);
6572 static struct pci_error_handlers ixgbe_err_handler = {
6573 .error_detected = ixgbe_io_error_detected,
6574 .slot_reset = ixgbe_io_slot_reset,
6575 .resume = ixgbe_io_resume,
6578 static struct pci_driver ixgbe_driver = {
6579 .name = ixgbe_driver_name,
6580 .id_table = ixgbe_pci_tbl,
6581 .probe = ixgbe_probe,
6582 .remove = __devexit_p(ixgbe_remove),
6584 .suspend = ixgbe_suspend,
6585 .resume = ixgbe_resume,
6587 .shutdown = ixgbe_shutdown,
6588 .err_handler = &ixgbe_err_handler
6592 * ixgbe_init_module - Driver Registration Routine
6594 * ixgbe_init_module is the first routine called when the driver is
6595 * loaded. All it does is register with the PCI subsystem.
6597 static int __init ixgbe_init_module(void)
6600 printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
6601 ixgbe_driver_string, ixgbe_driver_version);
6603 printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
6605 #ifdef CONFIG_IXGBE_DCA
6606 dca_register_notify(&dca_notifier);
6609 ret = pci_register_driver(&ixgbe_driver);
6613 module_init(ixgbe_init_module);
6616 * ixgbe_exit_module - Driver Exit Cleanup Routine
6618 * ixgbe_exit_module is called just before the driver is removed
6621 static void __exit ixgbe_exit_module(void)
6623 #ifdef CONFIG_IXGBE_DCA
6624 dca_unregister_notify(&dca_notifier);
6626 pci_unregister_driver(&ixgbe_driver);
6629 #ifdef CONFIG_IXGBE_DCA
6630 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
6635 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
6636 __ixgbe_notify_dca);
6638 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6641 #endif /* CONFIG_IXGBE_DCA */
6644 * ixgbe_get_hw_dev_name - return device name string
6645 * used by hardware layer to print debugging information
6647 char *ixgbe_get_hw_dev_name(struct ixgbe_hw *hw)
6649 struct ixgbe_adapter *adapter = hw->back;
6650 return adapter->netdev->name;
6654 module_exit(ixgbe_exit_module);