1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2009 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
36 #include <linux/tcp.h>
37 #include <linux/ipv6.h>
38 #include <net/checksum.h>
39 #include <net/ip6_checksum.h>
40 #include <linux/ethtool.h>
41 #include <linux/if_vlan.h>
44 #include "ixgbe_common.h"
46 char ixgbe_driver_name[] = "ixgbe";
47 static const char ixgbe_driver_string[] =
48 "Intel(R) 10 Gigabit PCI Express Network Driver";
50 #define DRV_VERSION "2.0.8-k2"
51 const char ixgbe_driver_version[] = DRV_VERSION;
52 static char ixgbe_copyright[] = "Copyright (c) 1999-2009 Intel Corporation.";
54 static const struct ixgbe_info *ixgbe_info_tbl[] = {
55 [board_82598] = &ixgbe_82598_info,
56 [board_82599] = &ixgbe_82599_info,
59 /* ixgbe_pci_tbl - PCI Device ID Table
61 * Wildcard entries (PCI_ANY_ID) should come last
62 * Last entry must be all 0s
64 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
65 * Class, Class Mask, private data (not used) }
67 static struct pci_device_id ixgbe_pci_tbl[] = {
68 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
70 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
72 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
74 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
76 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
78 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
80 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
82 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
84 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
95 /* required last entry */
98 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
100 #ifdef CONFIG_IXGBE_DCA
101 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
103 static struct notifier_block dca_notifier = {
104 .notifier_call = ixgbe_notify_dca,
110 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
111 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
112 MODULE_LICENSE("GPL");
113 MODULE_VERSION(DRV_VERSION);
115 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
117 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
121 /* Let firmware take over control of h/w */
122 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
123 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
124 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
127 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
131 /* Let firmware know the driver has taken over */
132 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
133 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
134 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
138 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
139 * @adapter: pointer to adapter struct
140 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
141 * @queue: queue to map the corresponding interrupt to
142 * @msix_vector: the vector to map to the corresponding queue
145 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
146 u8 queue, u8 msix_vector)
149 struct ixgbe_hw *hw = &adapter->hw;
150 switch (hw->mac.type) {
151 case ixgbe_mac_82598EB:
152 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
155 index = (((direction * 64) + queue) >> 2) & 0x1F;
156 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
157 ivar &= ~(0xFF << (8 * (queue & 0x3)));
158 ivar |= (msix_vector << (8 * (queue & 0x3)));
159 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
161 case ixgbe_mac_82599EB:
162 if (direction == -1) {
164 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
165 index = ((queue & 1) * 8);
166 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
167 ivar &= ~(0xFF << index);
168 ivar |= (msix_vector << index);
169 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
172 /* tx or rx causes */
173 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
174 index = ((16 * (queue & 1)) + (8 * direction));
175 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
176 ivar &= ~(0xFF << index);
177 ivar |= (msix_vector << index);
178 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
186 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
187 struct ixgbe_tx_buffer
190 if (tx_buffer_info->dma) {
191 pci_unmap_page(adapter->pdev, tx_buffer_info->dma,
192 tx_buffer_info->length, PCI_DMA_TODEVICE);
193 tx_buffer_info->dma = 0;
195 if (tx_buffer_info->skb) {
196 dev_kfree_skb_any(tx_buffer_info->skb);
197 tx_buffer_info->skb = NULL;
199 /* tx_buffer_info must be completely set up in the transmit path */
202 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
203 struct ixgbe_ring *tx_ring,
206 struct ixgbe_hw *hw = &adapter->hw;
209 /* Detect a transmit hang in hardware, this serializes the
210 * check with the clearing of time_stamp and movement of eop */
211 head = IXGBE_READ_REG(hw, tx_ring->head);
212 tail = IXGBE_READ_REG(hw, tx_ring->tail);
213 adapter->detect_tx_hung = false;
214 if ((head != tail) &&
215 tx_ring->tx_buffer_info[eop].time_stamp &&
216 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
217 !(IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & IXGBE_TFCS_TXOFF)) {
218 /* detected Tx unit hang */
219 union ixgbe_adv_tx_desc *tx_desc;
220 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
221 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
223 " TDH, TDT <%x>, <%x>\n"
224 " next_to_use <%x>\n"
225 " next_to_clean <%x>\n"
226 "tx_buffer_info[next_to_clean]\n"
227 " time_stamp <%lx>\n"
229 tx_ring->queue_index,
231 tx_ring->next_to_use, eop,
232 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
239 #define IXGBE_MAX_TXD_PWR 14
240 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
242 /* Tx Descriptors needed, worst case */
243 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
244 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
245 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
246 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
248 static void ixgbe_tx_timeout(struct net_device *netdev);
251 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
252 * @adapter: board private structure
253 * @tx_ring: tx ring to clean
255 static bool ixgbe_clean_tx_irq(struct ixgbe_adapter *adapter,
256 struct ixgbe_ring *tx_ring)
258 struct net_device *netdev = adapter->netdev;
259 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
260 struct ixgbe_tx_buffer *tx_buffer_info;
261 unsigned int i, eop, count = 0;
262 unsigned int total_bytes = 0, total_packets = 0;
264 i = tx_ring->next_to_clean;
265 eop = tx_ring->tx_buffer_info[i].next_to_watch;
266 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
268 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
269 (count < tx_ring->count)) {
270 bool cleaned = false;
271 for ( ; !cleaned; count++) {
273 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
274 tx_buffer_info = &tx_ring->tx_buffer_info[i];
275 cleaned = (i == eop);
276 skb = tx_buffer_info->skb;
278 if (cleaned && skb) {
279 unsigned int segs, bytecount;
281 /* gso_segs is currently only valid for tcp */
282 segs = skb_shinfo(skb)->gso_segs ?: 1;
283 /* multiply data chunks by size of headers */
284 bytecount = ((segs - 1) * skb_headlen(skb)) +
286 total_packets += segs;
287 total_bytes += bytecount;
290 ixgbe_unmap_and_free_tx_resource(adapter,
293 tx_desc->wb.status = 0;
296 if (i == tx_ring->count)
300 eop = tx_ring->tx_buffer_info[i].next_to_watch;
301 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
304 tx_ring->next_to_clean = i;
306 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
307 if (unlikely(count && netif_carrier_ok(netdev) &&
308 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
309 /* Make sure that anybody stopping the queue after this
310 * sees the new next_to_clean.
313 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
314 !test_bit(__IXGBE_DOWN, &adapter->state)) {
315 netif_wake_subqueue(netdev, tx_ring->queue_index);
316 ++adapter->restart_queue;
320 if (adapter->detect_tx_hung) {
321 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
322 /* schedule immediate reset if we believe we hung */
324 "tx hang %d detected, resetting adapter\n",
325 adapter->tx_timeout_count + 1);
326 ixgbe_tx_timeout(adapter->netdev);
330 /* re-arm the interrupt */
331 if ((total_packets >= tx_ring->work_limit) ||
332 (count == tx_ring->count))
333 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, tx_ring->v_idx);
335 tx_ring->total_bytes += total_bytes;
336 tx_ring->total_packets += total_packets;
337 tx_ring->stats.packets += total_packets;
338 tx_ring->stats.bytes += total_bytes;
339 adapter->net_stats.tx_bytes += total_bytes;
340 adapter->net_stats.tx_packets += total_packets;
341 return (total_packets ? true : false);
344 #ifdef CONFIG_IXGBE_DCA
345 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
346 struct ixgbe_ring *rx_ring)
350 int q = rx_ring - adapter->rx_ring;
352 if (rx_ring->cpu != cpu) {
353 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
354 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
355 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
356 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
357 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
358 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
359 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
360 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
362 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
363 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
364 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
365 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
366 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
367 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
373 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
374 struct ixgbe_ring *tx_ring)
378 int q = tx_ring - adapter->tx_ring;
380 if (tx_ring->cpu != cpu) {
381 txctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q));
382 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
383 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
384 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
385 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
386 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
387 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
388 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
390 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
391 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q), txctrl);
397 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
401 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
404 for (i = 0; i < adapter->num_tx_queues; i++) {
405 adapter->tx_ring[i].cpu = -1;
406 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
408 for (i = 0; i < adapter->num_rx_queues; i++) {
409 adapter->rx_ring[i].cpu = -1;
410 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
414 static int __ixgbe_notify_dca(struct device *dev, void *data)
416 struct net_device *netdev = dev_get_drvdata(dev);
417 struct ixgbe_adapter *adapter = netdev_priv(netdev);
418 unsigned long event = *(unsigned long *)data;
421 case DCA_PROVIDER_ADD:
422 /* if we're already enabled, don't do it again */
423 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
425 /* Always use CB2 mode, difference is masked
426 * in the CB driver. */
427 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
428 if (dca_add_requester(dev) == 0) {
429 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
430 ixgbe_setup_dca(adapter);
433 /* Fall Through since DCA is disabled. */
434 case DCA_PROVIDER_REMOVE:
435 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
436 dca_remove_requester(dev);
437 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
438 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
446 #endif /* CONFIG_IXGBE_DCA */
448 * ixgbe_receive_skb - Send a completed packet up the stack
449 * @adapter: board private structure
450 * @skb: packet to send up
451 * @status: hardware indication of status of receive
452 * @rx_ring: rx descriptor ring (for a specific queue) to setup
453 * @rx_desc: rx descriptor
455 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
456 struct sk_buff *skb, u8 status,
457 union ixgbe_adv_rx_desc *rx_desc)
459 struct ixgbe_adapter *adapter = q_vector->adapter;
460 struct napi_struct *napi = &q_vector->napi;
461 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
462 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
464 skb_record_rx_queue(skb, q_vector - &adapter->q_vector[0]);
465 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
466 if (adapter->vlgrp && is_vlan && (tag != 0))
467 vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
469 napi_gro_receive(napi, skb);
471 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
472 if (adapter->vlgrp && is_vlan && (tag != 0))
473 vlan_hwaccel_receive_skb(skb, adapter->vlgrp, tag);
475 netif_receive_skb(skb);
477 if (adapter->vlgrp && is_vlan && (tag != 0))
478 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
486 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
487 * @adapter: address of board private structure
488 * @status_err: hardware indication of status of receive
489 * @skb: skb currently being received and modified
491 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
492 u32 status_err, struct sk_buff *skb)
494 skb->ip_summed = CHECKSUM_NONE;
496 /* Rx csum disabled */
497 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
500 /* if IP and error */
501 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
502 (status_err & IXGBE_RXDADV_ERR_IPE)) {
503 adapter->hw_csum_rx_error++;
507 if (!(status_err & IXGBE_RXD_STAT_L4CS))
510 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
511 adapter->hw_csum_rx_error++;
515 /* It must be a TCP or UDP packet with a valid checksum */
516 skb->ip_summed = CHECKSUM_UNNECESSARY;
517 adapter->hw_csum_rx_good++;
520 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
521 struct ixgbe_ring *rx_ring, u32 val)
524 * Force memory writes to complete before letting h/w
525 * know there are new descriptors to fetch. (Only
526 * applicable for weak-ordered memory model archs,
530 IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
534 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
535 * @adapter: address of board private structure
537 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
538 struct ixgbe_ring *rx_ring,
541 struct pci_dev *pdev = adapter->pdev;
542 union ixgbe_adv_rx_desc *rx_desc;
543 struct ixgbe_rx_buffer *bi;
545 unsigned int bufsz = rx_ring->rx_buf_len + NET_IP_ALIGN;
547 i = rx_ring->next_to_use;
548 bi = &rx_ring->rx_buffer_info[i];
550 while (cleaned_count--) {
551 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
554 (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
556 bi->page = alloc_page(GFP_ATOMIC);
558 adapter->alloc_rx_page_failed++;
563 /* use a half page if we're re-using */
564 bi->page_offset ^= (PAGE_SIZE / 2);
567 bi->page_dma = pci_map_page(pdev, bi->page,
575 skb = netdev_alloc_skb(adapter->netdev, bufsz);
578 adapter->alloc_rx_buff_failed++;
583 * Make buffer alignment 2 beyond a 16 byte boundary
584 * this will result in a 16 byte aligned IP header after
585 * the 14 byte MAC header is removed
587 skb_reserve(skb, NET_IP_ALIGN);
590 bi->dma = pci_map_single(pdev, skb->data, bufsz,
593 /* Refresh the desc even if buffer_addrs didn't change because
594 * each write-back erases this info. */
595 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
596 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
597 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
599 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
603 if (i == rx_ring->count)
605 bi = &rx_ring->rx_buffer_info[i];
609 if (rx_ring->next_to_use != i) {
610 rx_ring->next_to_use = i;
612 i = (rx_ring->count - 1);
614 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
618 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
620 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
623 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
625 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
628 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
629 struct ixgbe_ring *rx_ring,
630 int *work_done, int work_to_do)
632 struct ixgbe_adapter *adapter = q_vector->adapter;
633 struct pci_dev *pdev = adapter->pdev;
634 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
635 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
640 bool cleaned = false;
641 int cleaned_count = 0;
642 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
644 i = rx_ring->next_to_clean;
645 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
646 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
647 rx_buffer_info = &rx_ring->rx_buffer_info[i];
649 while (staterr & IXGBE_RXD_STAT_DD) {
651 if (*work_done >= work_to_do)
655 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
656 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
657 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
658 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
659 if (hdr_info & IXGBE_RXDADV_SPH)
660 adapter->rx_hdr_split++;
661 if (len > IXGBE_RX_HDR_SIZE)
662 len = IXGBE_RX_HDR_SIZE;
663 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
665 len = le16_to_cpu(rx_desc->wb.upper.length);
669 skb = rx_buffer_info->skb;
670 prefetch(skb->data - NET_IP_ALIGN);
671 rx_buffer_info->skb = NULL;
673 if (len && !skb_shinfo(skb)->nr_frags) {
674 pci_unmap_single(pdev, rx_buffer_info->dma,
681 pci_unmap_page(pdev, rx_buffer_info->page_dma,
682 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
683 rx_buffer_info->page_dma = 0;
684 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
685 rx_buffer_info->page,
686 rx_buffer_info->page_offset,
689 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
690 (page_count(rx_buffer_info->page) != 1))
691 rx_buffer_info->page = NULL;
693 get_page(rx_buffer_info->page);
695 skb->len += upper_len;
696 skb->data_len += upper_len;
697 skb->truesize += upper_len;
701 if (i == rx_ring->count)
703 next_buffer = &rx_ring->rx_buffer_info[i];
705 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
709 if (staterr & IXGBE_RXD_STAT_EOP) {
710 rx_ring->stats.packets++;
711 rx_ring->stats.bytes += skb->len;
713 rx_buffer_info->skb = next_buffer->skb;
714 rx_buffer_info->dma = next_buffer->dma;
715 next_buffer->skb = skb;
716 next_buffer->dma = 0;
717 adapter->non_eop_descs++;
721 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
722 dev_kfree_skb_irq(skb);
726 ixgbe_rx_checksum(adapter, staterr, skb);
728 /* probably a little skewed due to removing CRC */
729 total_rx_bytes += skb->len;
732 skb->protocol = eth_type_trans(skb, adapter->netdev);
733 ixgbe_receive_skb(q_vector, skb, staterr, rx_desc);
736 rx_desc->wb.upper.status_error = 0;
738 /* return some buffers to hardware, one at a time is too slow */
739 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
740 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
744 /* use prefetched values */
746 rx_buffer_info = next_buffer;
748 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
751 rx_ring->next_to_clean = i;
752 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
755 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
757 rx_ring->total_packets += total_rx_packets;
758 rx_ring->total_bytes += total_rx_bytes;
759 adapter->net_stats.rx_bytes += total_rx_bytes;
760 adapter->net_stats.rx_packets += total_rx_packets;
765 static int ixgbe_clean_rxonly(struct napi_struct *, int);
767 * ixgbe_configure_msix - Configure MSI-X hardware
768 * @adapter: board private structure
770 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
773 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
775 struct ixgbe_q_vector *q_vector;
776 int i, j, q_vectors, v_idx, r_idx;
779 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
781 /* Populate the IVAR table and set the ITR values to the
782 * corresponding register.
784 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
785 q_vector = &adapter->q_vector[v_idx];
786 /* XXX for_each_bit(...) */
787 r_idx = find_first_bit(q_vector->rxr_idx,
788 adapter->num_rx_queues);
790 for (i = 0; i < q_vector->rxr_count; i++) {
791 j = adapter->rx_ring[r_idx].reg_idx;
792 ixgbe_set_ivar(adapter, 0, j, v_idx);
793 r_idx = find_next_bit(q_vector->rxr_idx,
794 adapter->num_rx_queues,
797 r_idx = find_first_bit(q_vector->txr_idx,
798 adapter->num_tx_queues);
800 for (i = 0; i < q_vector->txr_count; i++) {
801 j = adapter->tx_ring[r_idx].reg_idx;
802 ixgbe_set_ivar(adapter, 1, j, v_idx);
803 r_idx = find_next_bit(q_vector->txr_idx,
804 adapter->num_tx_queues,
808 /* if this is a tx only vector halve the interrupt rate */
809 if (q_vector->txr_count && !q_vector->rxr_count)
810 q_vector->eitr = (adapter->eitr_param >> 1);
813 q_vector->eitr = adapter->eitr_param;
815 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx),
816 EITR_INTS_PER_SEC_TO_REG(q_vector->eitr));
819 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
820 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
822 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
823 ixgbe_set_ivar(adapter, -1, 1, v_idx);
824 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
826 /* set up to autoclear timer, and the vectors */
827 mask = IXGBE_EIMS_ENABLE_MASK;
828 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
829 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
836 latency_invalid = 255
840 * ixgbe_update_itr - update the dynamic ITR value based on statistics
841 * @adapter: pointer to adapter
842 * @eitr: eitr setting (ints per sec) to give last timeslice
843 * @itr_setting: current throttle rate in ints/second
844 * @packets: the number of packets during this measurement interval
845 * @bytes: the number of bytes during this measurement interval
847 * Stores a new ITR value based on packets and byte
848 * counts during the last interrupt. The advantage of per interrupt
849 * computation is faster updates and more accurate ITR for the current
850 * traffic pattern. Constants in this function were computed
851 * based on theoretical maximum wire speed and thresholds were set based
852 * on testing data as well as attempting to minimize response time
853 * while increasing bulk throughput.
854 * this functionality is controlled by the InterruptThrottleRate module
855 * parameter (see ixgbe_param.c)
857 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
858 u32 eitr, u8 itr_setting,
859 int packets, int bytes)
861 unsigned int retval = itr_setting;
866 goto update_itr_done;
869 /* simple throttlerate management
870 * 0-20MB/s lowest (100000 ints/s)
871 * 20-100MB/s low (20000 ints/s)
872 * 100-1249MB/s bulk (8000 ints/s)
874 /* what was last interrupt timeslice? */
875 timepassed_us = 1000000/eitr;
876 bytes_perint = bytes / timepassed_us; /* bytes/usec */
878 switch (itr_setting) {
880 if (bytes_perint > adapter->eitr_low)
881 retval = low_latency;
884 if (bytes_perint > adapter->eitr_high)
885 retval = bulk_latency;
886 else if (bytes_perint <= adapter->eitr_low)
887 retval = lowest_latency;
890 if (bytes_perint <= adapter->eitr_high)
891 retval = low_latency;
899 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
901 struct ixgbe_adapter *adapter = q_vector->adapter;
902 struct ixgbe_hw *hw = &adapter->hw;
904 u8 current_itr, ret_itr;
905 int i, r_idx, v_idx = ((void *)q_vector - (void *)(adapter->q_vector)) /
906 sizeof(struct ixgbe_q_vector);
907 struct ixgbe_ring *rx_ring, *tx_ring;
909 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
910 for (i = 0; i < q_vector->txr_count; i++) {
911 tx_ring = &(adapter->tx_ring[r_idx]);
912 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
914 tx_ring->total_packets,
915 tx_ring->total_bytes);
916 /* if the result for this queue would decrease interrupt
917 * rate for this vector then use that result */
918 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
919 q_vector->tx_itr - 1 : ret_itr);
920 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
924 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
925 for (i = 0; i < q_vector->rxr_count; i++) {
926 rx_ring = &(adapter->rx_ring[r_idx]);
927 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
929 rx_ring->total_packets,
930 rx_ring->total_bytes);
931 /* if the result for this queue would decrease interrupt
932 * rate for this vector then use that result */
933 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
934 q_vector->rx_itr - 1 : ret_itr);
935 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
939 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
941 switch (current_itr) {
942 /* counts and packets in update_itr are dependent on these numbers */
947 new_itr = 20000; /* aka hwitr = ~200 */
955 if (new_itr != q_vector->eitr) {
957 /* do an exponential smoothing */
958 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
959 q_vector->eitr = new_itr;
960 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
961 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
962 /* Resolution is 2 usec on 82599, so halve the rate */
964 /* must write high and low 16 bits to reset counter */
965 DPRINTK(TX_ERR, DEBUG, "writing eitr(%d): %08X\n", v_idx,
967 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg | (itr_reg)<<16);
973 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
975 struct ixgbe_hw *hw = &adapter->hw;
977 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
978 (eicr & IXGBE_EICR_GPI_SDP1)) {
979 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
980 /* write to clear the interrupt */
981 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
985 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
987 struct ixgbe_hw *hw = &adapter->hw;
989 if (eicr & IXGBE_EICR_GPI_SDP1) {
990 /* Clear the interrupt */
991 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
992 schedule_work(&adapter->multispeed_fiber_task);
993 } else if (eicr & IXGBE_EICR_GPI_SDP2) {
994 /* Clear the interrupt */
995 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
996 schedule_work(&adapter->sfp_config_module_task);
998 /* Interrupt isn't for us... */
1003 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1005 struct ixgbe_hw *hw = &adapter->hw;
1008 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1009 adapter->link_check_timeout = jiffies;
1010 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1011 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1012 schedule_work(&adapter->watchdog_task);
1016 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1018 struct net_device *netdev = data;
1019 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1020 struct ixgbe_hw *hw = &adapter->hw;
1024 * Workaround for Silicon errata. Use clear-by-write instead
1025 * of clear-by-read. Reading with EICS will return the
1026 * interrupt causes without clearing, which later be done
1027 * with the write to EICR.
1029 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1030 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1032 if (eicr & IXGBE_EICR_LSC)
1033 ixgbe_check_lsc(adapter);
1035 if (hw->mac.type == ixgbe_mac_82598EB)
1036 ixgbe_check_fan_failure(adapter, eicr);
1038 if (hw->mac.type == ixgbe_mac_82599EB)
1039 ixgbe_check_sfp_event(adapter, eicr);
1040 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1041 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1046 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1048 struct ixgbe_q_vector *q_vector = data;
1049 struct ixgbe_adapter *adapter = q_vector->adapter;
1050 struct ixgbe_ring *tx_ring;
1053 if (!q_vector->txr_count)
1056 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1057 for (i = 0; i < q_vector->txr_count; i++) {
1058 tx_ring = &(adapter->tx_ring[r_idx]);
1059 #ifdef CONFIG_IXGBE_DCA
1060 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1061 ixgbe_update_tx_dca(adapter, tx_ring);
1063 tx_ring->total_bytes = 0;
1064 tx_ring->total_packets = 0;
1065 ixgbe_clean_tx_irq(adapter, tx_ring);
1066 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1074 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1076 * @data: pointer to our q_vector struct for this interrupt vector
1078 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1080 struct ixgbe_q_vector *q_vector = data;
1081 struct ixgbe_adapter *adapter = q_vector->adapter;
1082 struct ixgbe_ring *rx_ring;
1086 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1087 for (i = 0; i < q_vector->rxr_count; i++) {
1088 rx_ring = &(adapter->rx_ring[r_idx]);
1089 rx_ring->total_bytes = 0;
1090 rx_ring->total_packets = 0;
1091 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1095 if (!q_vector->rxr_count)
1098 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1099 rx_ring = &(adapter->rx_ring[r_idx]);
1100 /* disable interrupts on this vector only */
1101 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, rx_ring->v_idx);
1102 napi_schedule(&q_vector->napi);
1107 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1109 ixgbe_msix_clean_rx(irq, data);
1110 ixgbe_msix_clean_tx(irq, data);
1116 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1117 * @napi: napi struct with our devices info in it
1118 * @budget: amount of work driver is allowed to do this pass, in packets
1120 * This function is optimized for cleaning one queue only on a single
1123 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1125 struct ixgbe_q_vector *q_vector =
1126 container_of(napi, struct ixgbe_q_vector, napi);
1127 struct ixgbe_adapter *adapter = q_vector->adapter;
1128 struct ixgbe_ring *rx_ring = NULL;
1132 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1133 rx_ring = &(adapter->rx_ring[r_idx]);
1134 #ifdef CONFIG_IXGBE_DCA
1135 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1136 ixgbe_update_rx_dca(adapter, rx_ring);
1139 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1141 /* If all Rx work done, exit the polling mode */
1142 if (work_done < budget) {
1143 napi_complete(napi);
1144 if (adapter->itr_setting & 3)
1145 ixgbe_set_itr_msix(q_vector);
1146 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1147 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, rx_ring->v_idx);
1154 * ixgbe_clean_rxonly_many - msix (aka one shot) rx clean routine
1155 * @napi: napi struct with our devices info in it
1156 * @budget: amount of work driver is allowed to do this pass, in packets
1158 * This function will clean more than one rx queue associated with a
1161 static int ixgbe_clean_rxonly_many(struct napi_struct *napi, int budget)
1163 struct ixgbe_q_vector *q_vector =
1164 container_of(napi, struct ixgbe_q_vector, napi);
1165 struct ixgbe_adapter *adapter = q_vector->adapter;
1166 struct ixgbe_ring *rx_ring = NULL;
1167 int work_done = 0, i;
1169 u16 enable_mask = 0;
1171 /* attempt to distribute budget to each queue fairly, but don't allow
1172 * the budget to go below 1 because we'll exit polling */
1173 budget /= (q_vector->rxr_count ?: 1);
1174 budget = max(budget, 1);
1175 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1176 for (i = 0; i < q_vector->rxr_count; i++) {
1177 rx_ring = &(adapter->rx_ring[r_idx]);
1178 #ifdef CONFIG_IXGBE_DCA
1179 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1180 ixgbe_update_rx_dca(adapter, rx_ring);
1182 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1183 enable_mask |= rx_ring->v_idx;
1184 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1188 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1189 rx_ring = &(adapter->rx_ring[r_idx]);
1190 /* If all Rx work done, exit the polling mode */
1191 if (work_done < budget) {
1192 napi_complete(napi);
1193 if (adapter->itr_setting & 3)
1194 ixgbe_set_itr_msix(q_vector);
1195 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1196 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, enable_mask);
1202 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1205 a->q_vector[v_idx].adapter = a;
1206 set_bit(r_idx, a->q_vector[v_idx].rxr_idx);
1207 a->q_vector[v_idx].rxr_count++;
1208 a->rx_ring[r_idx].v_idx = 1 << v_idx;
1211 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1214 a->q_vector[v_idx].adapter = a;
1215 set_bit(r_idx, a->q_vector[v_idx].txr_idx);
1216 a->q_vector[v_idx].txr_count++;
1217 a->tx_ring[r_idx].v_idx = 1 << v_idx;
1221 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1222 * @adapter: board private structure to initialize
1223 * @vectors: allotted vector count for descriptor rings
1225 * This function maps descriptor rings to the queue-specific vectors
1226 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1227 * one vector per ring/queue, but on a constrained vector budget, we
1228 * group the rings as "efficiently" as possible. You would add new
1229 * mapping configurations in here.
1231 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1235 int rxr_idx = 0, txr_idx = 0;
1236 int rxr_remaining = adapter->num_rx_queues;
1237 int txr_remaining = adapter->num_tx_queues;
1242 /* No mapping required if MSI-X is disabled. */
1243 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1247 * The ideal configuration...
1248 * We have enough vectors to map one per queue.
1250 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1251 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1252 map_vector_to_rxq(adapter, v_start, rxr_idx);
1254 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1255 map_vector_to_txq(adapter, v_start, txr_idx);
1261 * If we don't have enough vectors for a 1-to-1
1262 * mapping, we'll have to group them so there are
1263 * multiple queues per vector.
1265 /* Re-adjusting *qpv takes care of the remainder. */
1266 for (i = v_start; i < vectors; i++) {
1267 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1268 for (j = 0; j < rqpv; j++) {
1269 map_vector_to_rxq(adapter, i, rxr_idx);
1274 for (i = v_start; i < vectors; i++) {
1275 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1276 for (j = 0; j < tqpv; j++) {
1277 map_vector_to_txq(adapter, i, txr_idx);
1288 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1289 * @adapter: board private structure
1291 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1292 * interrupts from the kernel.
1294 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1296 struct net_device *netdev = adapter->netdev;
1297 irqreturn_t (*handler)(int, void *);
1298 int i, vector, q_vectors, err;
1301 /* Decrement for Other and TCP Timer vectors */
1302 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1304 /* Map the Tx/Rx rings to the vectors we were allotted. */
1305 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1309 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1310 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1311 &ixgbe_msix_clean_many)
1312 for (vector = 0; vector < q_vectors; vector++) {
1313 handler = SET_HANDLER(&adapter->q_vector[vector]);
1315 if(handler == &ixgbe_msix_clean_rx) {
1316 sprintf(adapter->name[vector], "%s-%s-%d",
1317 netdev->name, "rx", ri++);
1319 else if(handler == &ixgbe_msix_clean_tx) {
1320 sprintf(adapter->name[vector], "%s-%s-%d",
1321 netdev->name, "tx", ti++);
1324 sprintf(adapter->name[vector], "%s-%s-%d",
1325 netdev->name, "TxRx", vector);
1327 err = request_irq(adapter->msix_entries[vector].vector,
1328 handler, 0, adapter->name[vector],
1329 &(adapter->q_vector[vector]));
1332 "request_irq failed for MSIX interrupt "
1333 "Error: %d\n", err);
1334 goto free_queue_irqs;
1338 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1339 err = request_irq(adapter->msix_entries[vector].vector,
1340 &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1343 "request_irq for msix_lsc failed: %d\n", err);
1344 goto free_queue_irqs;
1350 for (i = vector - 1; i >= 0; i--)
1351 free_irq(adapter->msix_entries[--vector].vector,
1352 &(adapter->q_vector[i]));
1353 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1354 pci_disable_msix(adapter->pdev);
1355 kfree(adapter->msix_entries);
1356 adapter->msix_entries = NULL;
1361 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1363 struct ixgbe_hw *hw = &adapter->hw;
1364 struct ixgbe_q_vector *q_vector = adapter->q_vector;
1366 u32 new_itr = q_vector->eitr;
1367 struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1368 struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1370 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1372 tx_ring->total_packets,
1373 tx_ring->total_bytes);
1374 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1376 rx_ring->total_packets,
1377 rx_ring->total_bytes);
1379 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1381 switch (current_itr) {
1382 /* counts and packets in update_itr are dependent on these numbers */
1383 case lowest_latency:
1387 new_itr = 20000; /* aka hwitr = ~200 */
1396 if (new_itr != q_vector->eitr) {
1398 /* do an exponential smoothing */
1399 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1400 q_vector->eitr = new_itr;
1401 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
1402 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1403 /* Resolution is 2 usec on 82599, so halve the rate */
1405 /* must write high and low 16 bits to reset counter */
1406 IXGBE_WRITE_REG(hw, IXGBE_EITR(0), itr_reg | (itr_reg)<<16);
1413 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1414 * @adapter: board private structure
1416 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1418 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1419 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1420 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
1421 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(2), ~0);
1423 IXGBE_WRITE_FLUSH(&adapter->hw);
1424 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1426 for (i = 0; i < adapter->num_msix_vectors; i++)
1427 synchronize_irq(adapter->msix_entries[i].vector);
1429 synchronize_irq(adapter->pdev->irq);
1434 * ixgbe_irq_enable - Enable default interrupt generation settings
1435 * @adapter: board private structure
1437 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1440 mask = IXGBE_EIMS_ENABLE_MASK;
1441 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1442 mask |= IXGBE_EIMS_GPI_SDP1;
1443 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1444 mask |= IXGBE_EIMS_GPI_SDP1;
1445 mask |= IXGBE_EIMS_GPI_SDP2;
1448 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1449 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1450 /* enable the rest of the queue vectors */
1451 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1),
1452 (IXGBE_EIMS_RTX_QUEUE << 16));
1453 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(2),
1454 ((IXGBE_EIMS_RTX_QUEUE << 16) |
1455 IXGBE_EIMS_RTX_QUEUE));
1457 IXGBE_WRITE_FLUSH(&adapter->hw);
1461 * ixgbe_intr - legacy mode Interrupt Handler
1462 * @irq: interrupt number
1463 * @data: pointer to a network interface device structure
1465 static irqreturn_t ixgbe_intr(int irq, void *data)
1467 struct net_device *netdev = data;
1468 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1469 struct ixgbe_hw *hw = &adapter->hw;
1473 * Workaround for silicon errata. Mask the interrupts
1474 * before the read of EICR.
1476 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1478 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1479 * therefore no explict interrupt disable is necessary */
1480 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1482 /* shared interrupt alert!
1483 * make sure interrupts are enabled because the read will
1484 * have disabled interrupts due to EIAM */
1485 ixgbe_irq_enable(adapter);
1486 return IRQ_NONE; /* Not our interrupt */
1489 if (eicr & IXGBE_EICR_LSC)
1490 ixgbe_check_lsc(adapter);
1492 if (hw->mac.type == ixgbe_mac_82599EB)
1493 ixgbe_check_sfp_event(adapter, eicr);
1495 ixgbe_check_fan_failure(adapter, eicr);
1497 if (napi_schedule_prep(&adapter->q_vector[0].napi)) {
1498 adapter->tx_ring[0].total_packets = 0;
1499 adapter->tx_ring[0].total_bytes = 0;
1500 adapter->rx_ring[0].total_packets = 0;
1501 adapter->rx_ring[0].total_bytes = 0;
1502 /* would disable interrupts here but EIAM disabled it */
1503 __napi_schedule(&adapter->q_vector[0].napi);
1509 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1511 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1513 for (i = 0; i < q_vectors; i++) {
1514 struct ixgbe_q_vector *q_vector = &adapter->q_vector[i];
1515 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1516 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1517 q_vector->rxr_count = 0;
1518 q_vector->txr_count = 0;
1523 * ixgbe_request_irq - initialize interrupts
1524 * @adapter: board private structure
1526 * Attempts to configure interrupts using the best available
1527 * capabilities of the hardware and kernel.
1529 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1531 struct net_device *netdev = adapter->netdev;
1534 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1535 err = ixgbe_request_msix_irqs(adapter);
1536 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1537 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1538 netdev->name, netdev);
1540 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1541 netdev->name, netdev);
1545 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1550 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1552 struct net_device *netdev = adapter->netdev;
1554 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1557 q_vectors = adapter->num_msix_vectors;
1560 free_irq(adapter->msix_entries[i].vector, netdev);
1563 for (; i >= 0; i--) {
1564 free_irq(adapter->msix_entries[i].vector,
1565 &(adapter->q_vector[i]));
1568 ixgbe_reset_q_vectors(adapter);
1570 free_irq(adapter->pdev->irq, netdev);
1575 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1578 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1580 struct ixgbe_hw *hw = &adapter->hw;
1582 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1583 EITR_INTS_PER_SEC_TO_REG(adapter->eitr_param));
1585 ixgbe_set_ivar(adapter, 0, 0, 0);
1586 ixgbe_set_ivar(adapter, 1, 0, 0);
1588 map_vector_to_rxq(adapter, 0, 0);
1589 map_vector_to_txq(adapter, 0, 0);
1591 DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1595 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1596 * @adapter: board private structure
1598 * Configure the Tx unit of the MAC after a reset.
1600 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1603 struct ixgbe_hw *hw = &adapter->hw;
1604 u32 i, j, tdlen, txctrl;
1606 /* Setup the HW Tx Head and Tail descriptor pointers */
1607 for (i = 0; i < adapter->num_tx_queues; i++) {
1608 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1611 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1612 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1613 (tdba & DMA_32BIT_MASK));
1614 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1615 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1616 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1617 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1618 adapter->tx_ring[i].head = IXGBE_TDH(j);
1619 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1620 /* Disable Tx Head Writeback RO bit, since this hoses
1621 * bookkeeping if things aren't delivered in order.
1623 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1624 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1625 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1627 if (hw->mac.type == ixgbe_mac_82599EB) {
1628 /* We enable 8 traffic classes, DCB only */
1629 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
1630 IXGBE_WRITE_REG(hw, IXGBE_MTQC, (IXGBE_MTQC_RT_ENA |
1631 IXGBE_MTQC_8TC_8TQ));
1635 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1637 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter, int index)
1639 struct ixgbe_ring *rx_ring;
1644 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1647 mask = (unsigned long) adapter->ring_feature[RING_F_RSS].mask;
1648 queue0 = index & mask;
1649 index = index & mask;
1652 rx_ring = &adapter->rx_ring[queue0];
1654 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
1656 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1657 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1659 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1660 u16 bufsz = IXGBE_RXBUFFER_2048;
1661 /* grow the amount we can receive on large page machines */
1662 if (bufsz < (PAGE_SIZE / 2))
1663 bufsz = (PAGE_SIZE / 2);
1664 /* cap the bufsz at our largest descriptor size */
1665 bufsz = min((u16)IXGBE_MAX_RXBUFFER, bufsz);
1667 srrctl |= bufsz >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1668 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1669 srrctl |= ((IXGBE_RX_HDR_SIZE <<
1670 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1671 IXGBE_SRRCTL_BSIZEHDR_MASK);
1673 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1675 if (rx_ring->rx_buf_len == MAXIMUM_ETHERNET_VLAN_SIZE)
1676 srrctl |= IXGBE_RXBUFFER_2048 >>
1677 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1679 srrctl |= rx_ring->rx_buf_len >>
1680 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1682 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1683 srrctl |= IXGBE_SRRCTL_DROP_EN;
1685 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
1689 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
1690 * @adapter: board private structure
1692 * Configure the Rx unit of the MAC after a reset.
1694 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
1697 struct ixgbe_hw *hw = &adapter->hw;
1698 struct net_device *netdev = adapter->netdev;
1699 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1701 u32 rdlen, rxctrl, rxcsum;
1702 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
1703 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
1704 0x6A3E67EA, 0x14364D17, 0x3BED200D};
1710 /* Decide whether to use packet split mode or not */
1711 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
1713 /* Set the RX buffer length according to the mode */
1714 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1715 rx_buf_len = IXGBE_RX_HDR_SIZE;
1716 if (hw->mac.type == ixgbe_mac_82599EB) {
1717 /* PSRTYPE must be initialized in 82599 */
1718 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
1719 IXGBE_PSRTYPE_UDPHDR |
1720 IXGBE_PSRTYPE_IPV4HDR |
1721 IXGBE_PSRTYPE_IPV6HDR;
1722 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), psrtype);
1725 if (netdev->mtu <= ETH_DATA_LEN)
1726 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1728 rx_buf_len = ALIGN(max_frame, 1024);
1731 fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
1732 fctrl |= IXGBE_FCTRL_BAM;
1733 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
1734 fctrl |= IXGBE_FCTRL_PMCF;
1735 IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
1737 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
1738 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1739 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
1741 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
1742 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
1744 rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1745 /* disable receives while setting up the descriptors */
1746 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
1747 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
1749 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1750 * the Base and Length of the Rx Descriptor Ring */
1751 for (i = 0; i < adapter->num_rx_queues; i++) {
1752 rdba = adapter->rx_ring[i].dma;
1753 j = adapter->rx_ring[i].reg_idx;
1754 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_32BIT_MASK));
1755 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
1756 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
1757 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
1758 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
1759 adapter->rx_ring[i].head = IXGBE_RDH(j);
1760 adapter->rx_ring[i].tail = IXGBE_RDT(j);
1761 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1763 ixgbe_configure_srrctl(adapter, j);
1766 if (hw->mac.type == ixgbe_mac_82598EB) {
1768 * For VMDq support of different descriptor types or
1769 * buffer sizes through the use of multiple SRRCTL
1770 * registers, RDRXCTL.MVMEN must be set to 1
1772 * also, the manual doesn't mention it clearly but DCA hints
1773 * will only use queue 0's tags unless this bit is set. Side
1774 * effects of setting this bit are only that SRRCTL must be
1775 * fully programmed [0..15]
1777 if (adapter->flags &
1778 (IXGBE_FLAG_RSS_ENABLED | IXGBE_FLAG_VMDQ_ENABLED)) {
1779 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
1780 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
1781 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
1785 /* Program MRQC for the distribution of queues */
1786 if (hw->mac.type == ixgbe_mac_82599EB) {
1787 int mask = adapter->flags & (
1788 IXGBE_FLAG_RSS_ENABLED
1789 | IXGBE_FLAG_DCB_ENABLED
1793 case (IXGBE_FLAG_RSS_ENABLED):
1794 mrqc = IXGBE_MRQC_RSSEN;
1796 case (IXGBE_FLAG_DCB_ENABLED):
1797 mrqc = IXGBE_MRQC_RT8TCEN;
1803 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
1804 /* Fill out redirection table */
1805 for (i = 0, j = 0; i < 128; i++, j++) {
1806 if (j == adapter->ring_feature[RING_F_RSS].indices)
1808 /* reta = 4-byte sliding window of
1809 * 0x00..(indices-1)(indices-1)00..etc. */
1810 reta = (reta << 8) | (j * 0x11);
1812 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
1815 /* Fill out hash function seeds */
1816 for (i = 0; i < 10; i++)
1817 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
1819 mrqc = IXGBE_MRQC_RSSEN
1820 /* Perform hash on these packet types */
1821 | IXGBE_MRQC_RSS_FIELD_IPV4
1822 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
1823 | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
1824 | IXGBE_MRQC_RSS_FIELD_IPV6
1825 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
1826 | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
1827 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
1830 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
1832 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
1833 adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
1834 /* Disable indicating checksum in descriptor, enables
1836 rxcsum |= IXGBE_RXCSUM_PCSD;
1838 if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
1839 /* Enable IPv4 payload checksum for UDP fragments
1840 * if PCSD is not set */
1841 rxcsum |= IXGBE_RXCSUM_IPPCSE;
1844 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
1846 if (hw->mac.type == ixgbe_mac_82599EB) {
1847 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
1848 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
1849 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
1853 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1855 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1856 struct ixgbe_hw *hw = &adapter->hw;
1858 /* add VID to filter table */
1859 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
1862 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1864 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1865 struct ixgbe_hw *hw = &adapter->hw;
1867 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1868 ixgbe_irq_disable(adapter);
1870 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1872 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1873 ixgbe_irq_enable(adapter);
1875 /* remove VID from filter table */
1876 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
1879 static void ixgbe_vlan_rx_register(struct net_device *netdev,
1880 struct vlan_group *grp)
1882 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1886 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1887 ixgbe_irq_disable(adapter);
1888 adapter->vlgrp = grp;
1891 * For a DCB driver, always enable VLAN tag stripping so we can
1892 * still receive traffic from a DCB-enabled host even if we're
1895 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
1896 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1897 ctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
1898 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
1899 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
1900 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1901 ctrl |= IXGBE_VLNCTRL_VFE;
1902 /* enable VLAN tag insert/strip */
1903 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
1904 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
1905 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
1906 for (i = 0; i < adapter->num_rx_queues; i++) {
1907 j = adapter->rx_ring[i].reg_idx;
1908 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_RXDCTL(j));
1909 ctrl |= IXGBE_RXDCTL_VME;
1910 IXGBE_WRITE_REG(&adapter->hw, IXGBE_RXDCTL(j), ctrl);
1913 ixgbe_vlan_rx_add_vid(netdev, 0);
1915 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1916 ixgbe_irq_enable(adapter);
1919 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
1921 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1923 if (adapter->vlgrp) {
1925 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1926 if (!vlan_group_get_device(adapter->vlgrp, vid))
1928 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
1933 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
1935 struct dev_mc_list *mc_ptr;
1936 u8 *addr = *mc_addr_ptr;
1939 mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
1941 *mc_addr_ptr = mc_ptr->next->dmi_addr;
1943 *mc_addr_ptr = NULL;
1949 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
1950 * @netdev: network interface device structure
1952 * The set_rx_method entry point is called whenever the unicast/multicast
1953 * address list or the network interface flags are updated. This routine is
1954 * responsible for configuring the hardware for proper unicast, multicast and
1957 static void ixgbe_set_rx_mode(struct net_device *netdev)
1959 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1960 struct ixgbe_hw *hw = &adapter->hw;
1962 u8 *addr_list = NULL;
1965 /* Check for Promiscuous and All Multicast modes */
1967 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
1968 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
1970 if (netdev->flags & IFF_PROMISC) {
1971 hw->addr_ctrl.user_set_promisc = 1;
1972 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
1973 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
1975 if (netdev->flags & IFF_ALLMULTI) {
1976 fctrl |= IXGBE_FCTRL_MPE;
1977 fctrl &= ~IXGBE_FCTRL_UPE;
1979 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
1981 vlnctrl |= IXGBE_VLNCTRL_VFE;
1982 hw->addr_ctrl.user_set_promisc = 0;
1985 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
1986 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
1988 /* reprogram secondary unicast list */
1989 addr_count = netdev->uc_count;
1991 addr_list = netdev->uc_list->dmi_addr;
1992 hw->mac.ops.update_uc_addr_list(hw, addr_list, addr_count,
1993 ixgbe_addr_list_itr);
1995 /* reprogram multicast list */
1996 addr_count = netdev->mc_count;
1998 addr_list = netdev->mc_list->dmi_addr;
1999 hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
2000 ixgbe_addr_list_itr);
2003 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
2006 struct ixgbe_q_vector *q_vector;
2007 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2009 /* legacy and MSI only use one vector */
2010 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2013 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2014 struct napi_struct *napi;
2015 q_vector = &adapter->q_vector[q_idx];
2016 if (!q_vector->rxr_count)
2018 napi = &q_vector->napi;
2019 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) &&
2020 (q_vector->rxr_count > 1))
2021 napi->poll = &ixgbe_clean_rxonly_many;
2027 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2030 struct ixgbe_q_vector *q_vector;
2031 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2033 /* legacy and MSI only use one vector */
2034 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2037 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2038 q_vector = &adapter->q_vector[q_idx];
2039 if (!q_vector->rxr_count)
2041 napi_disable(&q_vector->napi);
2045 #ifdef CONFIG_IXGBE_DCB
2047 * ixgbe_configure_dcb - Configure DCB hardware
2048 * @adapter: ixgbe adapter struct
2050 * This is called by the driver on open to configure the DCB hardware.
2051 * This is also called by the gennetlink interface when reconfiguring
2054 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2056 struct ixgbe_hw *hw = &adapter->hw;
2057 u32 txdctl, vlnctrl;
2060 ixgbe_dcb_check_config(&adapter->dcb_cfg);
2061 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2062 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2064 /* reconfigure the hardware */
2065 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
2067 for (i = 0; i < adapter->num_tx_queues; i++) {
2068 j = adapter->tx_ring[i].reg_idx;
2069 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2070 /* PThresh workaround for Tx hang with DFP enabled. */
2072 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2074 /* Enable VLAN tag insert/strip */
2075 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2076 if (hw->mac.type == ixgbe_mac_82598EB) {
2077 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2078 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2079 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2080 } else if (hw->mac.type == ixgbe_mac_82599EB) {
2081 vlnctrl |= IXGBE_VLNCTRL_VFE;
2082 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2083 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2084 for (i = 0; i < adapter->num_rx_queues; i++) {
2085 j = adapter->rx_ring[i].reg_idx;
2086 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2087 vlnctrl |= IXGBE_RXDCTL_VME;
2088 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2091 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2095 static void ixgbe_configure(struct ixgbe_adapter *adapter)
2097 struct net_device *netdev = adapter->netdev;
2100 ixgbe_set_rx_mode(netdev);
2102 ixgbe_restore_vlan(adapter);
2103 #ifdef CONFIG_IXGBE_DCB
2104 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2105 netif_set_gso_max_size(netdev, 32768);
2106 ixgbe_configure_dcb(adapter);
2108 netif_set_gso_max_size(netdev, 65536);
2111 netif_set_gso_max_size(netdev, 65536);
2114 ixgbe_configure_tx(adapter);
2115 ixgbe_configure_rx(adapter);
2116 for (i = 0; i < adapter->num_rx_queues; i++)
2117 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
2118 (adapter->rx_ring[i].count - 1));
2121 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2123 switch (hw->phy.type) {
2124 case ixgbe_phy_sfp_avago:
2125 case ixgbe_phy_sfp_ftl:
2126 case ixgbe_phy_sfp_intel:
2127 case ixgbe_phy_sfp_unknown:
2128 case ixgbe_phy_tw_tyco:
2129 case ixgbe_phy_tw_unknown:
2137 * ixgbe_sfp_link_config - set up SFP+ link
2138 * @adapter: pointer to private adapter struct
2140 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2142 struct ixgbe_hw *hw = &adapter->hw;
2144 if (hw->phy.multispeed_fiber) {
2146 * In multispeed fiber setups, the device may not have
2147 * had a physical connection when the driver loaded.
2148 * If that's the case, the initial link configuration
2149 * couldn't get the MAC into 10G or 1G mode, so we'll
2150 * never have a link status change interrupt fire.
2151 * We need to try and force an autonegotiation
2152 * session, then bring up link.
2154 hw->mac.ops.setup_sfp(hw);
2155 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2156 schedule_work(&adapter->multispeed_fiber_task);
2159 * Direct Attach Cu and non-multispeed fiber modules
2160 * still need to be configured properly prior to
2163 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2164 schedule_work(&adapter->sfp_config_module_task);
2169 * ixgbe_non_sfp_link_config - set up non-SFP+ link
2170 * @hw: pointer to private hardware struct
2172 * Returns 0 on success, negative on failure
2174 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2177 bool link_up = false;
2178 u32 ret = IXGBE_ERR_LINK_SETUP;
2180 if (hw->mac.ops.check_link)
2181 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2186 if (hw->mac.ops.get_link_capabilities)
2187 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
2192 if (hw->mac.ops.setup_link_speed)
2193 ret = hw->mac.ops.setup_link_speed(hw, autoneg, true, link_up);
2198 #define IXGBE_MAX_RX_DESC_POLL 10
2199 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2202 int j = adapter->rx_ring[rxr].reg_idx;
2205 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2206 if (IXGBE_READ_REG(&adapter->hw,
2207 IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2212 if (k >= IXGBE_MAX_RX_DESC_POLL) {
2213 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2214 "not set within the polling period\n", rxr);
2216 ixgbe_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
2217 (adapter->rx_ring[rxr].count - 1));
2220 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2222 struct net_device *netdev = adapter->netdev;
2223 struct ixgbe_hw *hw = &adapter->hw;
2225 int num_rx_rings = adapter->num_rx_queues;
2227 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2228 u32 txdctl, rxdctl, mhadd;
2232 ixgbe_get_hw_control(adapter);
2234 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2235 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2236 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2237 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2238 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2243 /* XXX: to interrupt immediately for EICS writes, enable this */
2244 /* gpie |= IXGBE_GPIE_EIMEN; */
2245 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2248 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2249 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2250 * specifically only auto mask tx and rx interrupts */
2251 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2254 /* Enable fan failure interrupt if media type is copper */
2255 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2256 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2257 gpie |= IXGBE_SDP1_GPIEN;
2258 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2261 if (hw->mac.type == ixgbe_mac_82599EB) {
2262 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2263 gpie |= IXGBE_SDP1_GPIEN;
2264 gpie |= IXGBE_SDP2_GPIEN;
2265 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2268 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2269 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2270 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2271 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2273 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2276 for (i = 0; i < adapter->num_tx_queues; i++) {
2277 j = adapter->tx_ring[i].reg_idx;
2278 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2279 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2280 txdctl |= (8 << 16);
2281 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2284 if (hw->mac.type == ixgbe_mac_82599EB) {
2285 /* DMATXCTL.EN must be set after all Tx queue config is done */
2286 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2287 dmatxctl |= IXGBE_DMATXCTL_TE;
2288 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2290 for (i = 0; i < adapter->num_tx_queues; i++) {
2291 j = adapter->tx_ring[i].reg_idx;
2292 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2293 txdctl |= IXGBE_TXDCTL_ENABLE;
2294 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2297 for (i = 0; i < num_rx_rings; i++) {
2298 j = adapter->rx_ring[i].reg_idx;
2299 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2300 /* enable PTHRESH=32 descriptors (half the internal cache)
2301 * and HTHRESH=0 descriptors (to minimize latency on fetch),
2302 * this also removes a pesky rx_no_buffer_count increment */
2304 rxdctl |= IXGBE_RXDCTL_ENABLE;
2305 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2306 if (hw->mac.type == ixgbe_mac_82599EB)
2307 ixgbe_rx_desc_queue_enable(adapter, i);
2309 /* enable all receives */
2310 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2311 if (hw->mac.type == ixgbe_mac_82598EB)
2312 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2314 rxdctl |= IXGBE_RXCTRL_RXEN;
2315 hw->mac.ops.enable_rx_dma(hw, rxdctl);
2317 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2318 ixgbe_configure_msix(adapter);
2320 ixgbe_configure_msi_and_legacy(adapter);
2322 ixgbe_napi_add_all(adapter);
2324 clear_bit(__IXGBE_DOWN, &adapter->state);
2325 ixgbe_napi_enable_all(adapter);
2327 /* clear any pending interrupts, may auto mask */
2328 IXGBE_READ_REG(hw, IXGBE_EICR);
2330 ixgbe_irq_enable(adapter);
2333 * For hot-pluggable SFP+ devices, a new SFP+ module may have
2334 * arrived before interrupts were enabled. We need to kick off
2335 * the SFP+ module setup first, then try to bring up link.
2336 * If we're not hot-pluggable SFP+, we just need to configure link
2339 err = hw->phy.ops.identify(hw);
2340 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
2341 DPRINTK(PROBE, ERR, "PHY not supported on this NIC %d\n", err);
2342 ixgbe_down(adapter);
2346 if (ixgbe_is_sfp(hw)) {
2347 ixgbe_sfp_link_config(adapter);
2349 err = ixgbe_non_sfp_link_config(hw);
2351 DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
2354 /* enable transmits */
2355 netif_tx_start_all_queues(netdev);
2357 /* bring the link up in the watchdog, this could race with our first
2358 * link up interrupt but shouldn't be a problem */
2359 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2360 adapter->link_check_timeout = jiffies;
2361 mod_timer(&adapter->watchdog_timer, jiffies);
2365 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
2367 WARN_ON(in_interrupt());
2368 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
2370 ixgbe_down(adapter);
2372 clear_bit(__IXGBE_RESETTING, &adapter->state);
2375 int ixgbe_up(struct ixgbe_adapter *adapter)
2377 /* hardware has been reset, we need to reload some things */
2378 ixgbe_configure(adapter);
2380 return ixgbe_up_complete(adapter);
2383 void ixgbe_reset(struct ixgbe_adapter *adapter)
2385 struct ixgbe_hw *hw = &adapter->hw;
2386 if (hw->mac.ops.init_hw(hw))
2387 dev_err(&adapter->pdev->dev, "Hardware Error\n");
2389 /* reprogram the RAR[0] in case user changed it. */
2390 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2395 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2396 * @adapter: board private structure
2397 * @rx_ring: ring to free buffers from
2399 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2400 struct ixgbe_ring *rx_ring)
2402 struct pci_dev *pdev = adapter->pdev;
2406 /* Free all the Rx ring sk_buffs */
2408 for (i = 0; i < rx_ring->count; i++) {
2409 struct ixgbe_rx_buffer *rx_buffer_info;
2411 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2412 if (rx_buffer_info->dma) {
2413 pci_unmap_single(pdev, rx_buffer_info->dma,
2414 rx_ring->rx_buf_len,
2415 PCI_DMA_FROMDEVICE);
2416 rx_buffer_info->dma = 0;
2418 if (rx_buffer_info->skb) {
2419 dev_kfree_skb(rx_buffer_info->skb);
2420 rx_buffer_info->skb = NULL;
2422 if (!rx_buffer_info->page)
2424 pci_unmap_page(pdev, rx_buffer_info->page_dma, PAGE_SIZE / 2,
2425 PCI_DMA_FROMDEVICE);
2426 rx_buffer_info->page_dma = 0;
2427 put_page(rx_buffer_info->page);
2428 rx_buffer_info->page = NULL;
2429 rx_buffer_info->page_offset = 0;
2432 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2433 memset(rx_ring->rx_buffer_info, 0, size);
2435 /* Zero out the descriptor ring */
2436 memset(rx_ring->desc, 0, rx_ring->size);
2438 rx_ring->next_to_clean = 0;
2439 rx_ring->next_to_use = 0;
2441 writel(0, adapter->hw.hw_addr + rx_ring->head);
2442 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2446 * ixgbe_clean_tx_ring - Free Tx Buffers
2447 * @adapter: board private structure
2448 * @tx_ring: ring to be cleaned
2450 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2451 struct ixgbe_ring *tx_ring)
2453 struct ixgbe_tx_buffer *tx_buffer_info;
2457 /* Free all the Tx ring sk_buffs */
2459 for (i = 0; i < tx_ring->count; i++) {
2460 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2461 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2464 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2465 memset(tx_ring->tx_buffer_info, 0, size);
2467 /* Zero out the descriptor ring */
2468 memset(tx_ring->desc, 0, tx_ring->size);
2470 tx_ring->next_to_use = 0;
2471 tx_ring->next_to_clean = 0;
2473 writel(0, adapter->hw.hw_addr + tx_ring->head);
2474 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2478 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2479 * @adapter: board private structure
2481 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2485 for (i = 0; i < adapter->num_rx_queues; i++)
2486 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2490 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2491 * @adapter: board private structure
2493 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2497 for (i = 0; i < adapter->num_tx_queues; i++)
2498 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2501 void ixgbe_down(struct ixgbe_adapter *adapter)
2503 struct net_device *netdev = adapter->netdev;
2504 struct ixgbe_hw *hw = &adapter->hw;
2509 /* signal that we are down to the interrupt handler */
2510 set_bit(__IXGBE_DOWN, &adapter->state);
2512 /* disable receives */
2513 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2514 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2516 netif_tx_disable(netdev);
2518 IXGBE_WRITE_FLUSH(hw);
2521 netif_tx_stop_all_queues(netdev);
2523 ixgbe_irq_disable(adapter);
2525 ixgbe_napi_disable_all(adapter);
2527 del_timer_sync(&adapter->watchdog_timer);
2528 cancel_work_sync(&adapter->watchdog_task);
2530 /* disable transmits in the hardware now that interrupts are off */
2531 for (i = 0; i < adapter->num_tx_queues; i++) {
2532 j = adapter->tx_ring[i].reg_idx;
2533 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2534 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
2535 (txdctl & ~IXGBE_TXDCTL_ENABLE));
2538 netif_carrier_off(netdev);
2540 #ifdef CONFIG_IXGBE_DCA
2541 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2542 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
2543 dca_remove_requester(&adapter->pdev->dev);
2547 if (!pci_channel_offline(adapter->pdev))
2548 ixgbe_reset(adapter);
2549 ixgbe_clean_all_tx_rings(adapter);
2550 ixgbe_clean_all_rx_rings(adapter);
2552 #ifdef CONFIG_IXGBE_DCA
2553 /* since we reset the hardware DCA settings were cleared */
2554 if (dca_add_requester(&adapter->pdev->dev) == 0) {
2555 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
2556 /* always use CB2 mode, difference is masked
2557 * in the CB driver */
2558 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
2559 ixgbe_setup_dca(adapter);
2565 * ixgbe_poll - NAPI Rx polling callback
2566 * @napi: structure for representing this polling device
2567 * @budget: how many packets driver is allowed to clean
2569 * This function is used for legacy and MSI, NAPI mode
2571 static int ixgbe_poll(struct napi_struct *napi, int budget)
2573 struct ixgbe_q_vector *q_vector = container_of(napi,
2574 struct ixgbe_q_vector, napi);
2575 struct ixgbe_adapter *adapter = q_vector->adapter;
2576 int tx_cleaned, work_done = 0;
2578 #ifdef CONFIG_IXGBE_DCA
2579 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2580 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
2581 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
2585 tx_cleaned = ixgbe_clean_tx_irq(adapter, adapter->tx_ring);
2586 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring, &work_done, budget);
2591 /* If budget not fully consumed, exit the polling mode */
2592 if (work_done < budget) {
2593 napi_complete(napi);
2594 if (adapter->itr_setting & 3)
2595 ixgbe_set_itr(adapter);
2596 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2597 ixgbe_irq_enable(adapter);
2603 * ixgbe_tx_timeout - Respond to a Tx Hang
2604 * @netdev: network interface device structure
2606 static void ixgbe_tx_timeout(struct net_device *netdev)
2608 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2610 /* Do the reset outside of interrupt context */
2611 schedule_work(&adapter->reset_task);
2614 static void ixgbe_reset_task(struct work_struct *work)
2616 struct ixgbe_adapter *adapter;
2617 adapter = container_of(work, struct ixgbe_adapter, reset_task);
2619 /* If we're already down or resetting, just bail */
2620 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
2621 test_bit(__IXGBE_RESETTING, &adapter->state))
2624 adapter->tx_timeout_count++;
2626 ixgbe_reinit_locked(adapter);
2629 #ifdef CONFIG_IXGBE_DCB
2630 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
2634 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2635 adapter->ring_feature[RING_F_DCB].mask = 0x7 << 3;
2636 adapter->num_rx_queues =
2637 adapter->ring_feature[RING_F_DCB].indices;
2638 adapter->num_tx_queues =
2639 adapter->ring_feature[RING_F_DCB].indices;
2649 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
2653 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2654 adapter->ring_feature[RING_F_RSS].mask = 0xF;
2655 adapter->num_rx_queues =
2656 adapter->ring_feature[RING_F_RSS].indices;
2657 adapter->num_tx_queues =
2658 adapter->ring_feature[RING_F_RSS].indices;
2667 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
2669 /* Start with base case */
2670 adapter->num_rx_queues = 1;
2671 adapter->num_tx_queues = 1;
2673 #ifdef CONFIG_IXGBE_DCB
2674 if (ixgbe_set_dcb_queues(adapter))
2678 if (ixgbe_set_rss_queues(adapter))
2682 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
2685 int err, vector_threshold;
2687 /* We'll want at least 3 (vector_threshold):
2690 * 3) Other (Link Status Change, etc.)
2691 * 4) TCP Timer (optional)
2693 vector_threshold = MIN_MSIX_COUNT;
2695 /* The more we get, the more we will assign to Tx/Rx Cleanup
2696 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2697 * Right now, we simply care about how many we'll get; we'll
2698 * set them up later while requesting irq's.
2700 while (vectors >= vector_threshold) {
2701 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
2703 if (!err) /* Success in acquiring all requested vectors. */
2706 vectors = 0; /* Nasty failure, quit now */
2707 else /* err == number of vectors we should try again with */
2711 if (vectors < vector_threshold) {
2712 /* Can't allocate enough MSI-X interrupts? Oh well.
2713 * This just means we'll go with either a single MSI
2714 * vector or fall back to legacy interrupts.
2716 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
2717 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2718 kfree(adapter->msix_entries);
2719 adapter->msix_entries = NULL;
2720 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
2721 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2722 ixgbe_set_num_queues(adapter);
2724 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
2726 * Adjust for only the vectors we'll use, which is minimum
2727 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2728 * vectors we were allocated.
2730 adapter->num_msix_vectors = min(vectors,
2731 adapter->max_msix_q_vectors + NON_Q_VECTORS);
2736 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
2737 * @adapter: board private structure to initialize
2739 * Cache the descriptor ring offsets for RSS to the assigned rings.
2742 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
2747 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2748 for (i = 0; i < adapter->num_rx_queues; i++)
2749 adapter->rx_ring[i].reg_idx = i;
2750 for (i = 0; i < adapter->num_tx_queues; i++)
2751 adapter->tx_ring[i].reg_idx = i;
2760 #ifdef CONFIG_IXGBE_DCB
2762 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
2763 * @adapter: board private structure to initialize
2765 * Cache the descriptor ring offsets for DCB to the assigned rings.
2768 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
2772 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
2774 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2775 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2776 /* the number of queues is assumed to be symmetric */
2777 for (i = 0; i < dcb_i; i++) {
2778 adapter->rx_ring[i].reg_idx = i << 3;
2779 adapter->tx_ring[i].reg_idx = i << 2;
2782 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2783 for (i = 0; i < dcb_i; i++) {
2784 adapter->rx_ring[i].reg_idx = i << 4;
2785 adapter->tx_ring[i].reg_idx = i << 4;
2800 * ixgbe_cache_ring_register - Descriptor ring to register mapping
2801 * @adapter: board private structure to initialize
2803 * Once we know the feature-set enabled for the device, we'll cache
2804 * the register offset the descriptor ring is assigned to.
2806 * Note, the order the various feature calls is important. It must start with
2807 * the "most" features enabled at the same time, then trickle down to the
2808 * least amount of features turned on at once.
2810 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
2812 /* start with default case */
2813 adapter->rx_ring[0].reg_idx = 0;
2814 adapter->tx_ring[0].reg_idx = 0;
2816 #ifdef CONFIG_IXGBE_DCB
2817 if (ixgbe_cache_ring_dcb(adapter))
2821 if (ixgbe_cache_ring_rss(adapter))
2826 * ixgbe_alloc_queues - Allocate memory for all rings
2827 * @adapter: board private structure to initialize
2829 * We allocate one ring per queue at run-time since we don't know the
2830 * number of queues at compile-time.
2832 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
2836 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
2837 sizeof(struct ixgbe_ring), GFP_KERNEL);
2838 if (!adapter->tx_ring)
2839 goto err_tx_ring_allocation;
2841 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
2842 sizeof(struct ixgbe_ring), GFP_KERNEL);
2843 if (!adapter->rx_ring)
2844 goto err_rx_ring_allocation;
2846 for (i = 0; i < adapter->num_tx_queues; i++) {
2847 adapter->tx_ring[i].count = adapter->tx_ring_count;
2848 adapter->tx_ring[i].queue_index = i;
2851 for (i = 0; i < adapter->num_rx_queues; i++) {
2852 adapter->rx_ring[i].count = adapter->rx_ring_count;
2853 adapter->rx_ring[i].queue_index = i;
2856 ixgbe_cache_ring_register(adapter);
2860 err_rx_ring_allocation:
2861 kfree(adapter->tx_ring);
2862 err_tx_ring_allocation:
2867 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
2868 * @adapter: board private structure to initialize
2870 * Attempt to configure the interrupts using the best available
2871 * capabilities of the hardware and the kernel.
2873 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
2876 int vector, v_budget;
2879 * It's easy to be greedy for MSI-X vectors, but it really
2880 * doesn't do us much good if we have a lot more vectors
2881 * than CPU's. So let's be conservative and only ask for
2882 * (roughly) twice the number of vectors as there are CPU's.
2884 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
2885 (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
2888 * At the same time, hardware can only support a maximum of
2889 * MAX_MSIX_COUNT vectors. With features such as RSS and VMDq,
2890 * we can easily reach upwards of 64 Rx descriptor queues and
2891 * 32 Tx queues. Thus, we cap it off in those rare cases where
2892 * the cpu count also exceeds our vector limit.
2894 v_budget = min(v_budget, MAX_MSIX_COUNT);
2896 /* A failure in MSI-X entry allocation isn't fatal, but it does
2897 * mean we disable MSI-X capabilities of the adapter. */
2898 adapter->msix_entries = kcalloc(v_budget,
2899 sizeof(struct msix_entry), GFP_KERNEL);
2900 if (!adapter->msix_entries) {
2901 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
2902 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2903 ixgbe_set_num_queues(adapter);
2904 kfree(adapter->tx_ring);
2905 kfree(adapter->rx_ring);
2906 err = ixgbe_alloc_queues(adapter);
2908 DPRINTK(PROBE, ERR, "Unable to allocate memory "
2916 for (vector = 0; vector < v_budget; vector++)
2917 adapter->msix_entries[vector].entry = vector;
2919 ixgbe_acquire_msix_vectors(adapter, v_budget);
2921 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2925 err = pci_enable_msi(adapter->pdev);
2927 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
2929 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
2930 "falling back to legacy. Error: %d\n", err);
2936 /* Notify the stack of the (possibly) reduced Tx Queue count. */
2937 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
2942 void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
2944 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2945 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2946 pci_disable_msix(adapter->pdev);
2947 kfree(adapter->msix_entries);
2948 adapter->msix_entries = NULL;
2949 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2950 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
2951 pci_disable_msi(adapter->pdev);
2957 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
2958 * @adapter: board private structure to initialize
2960 * We determine which interrupt scheme to use based on...
2961 * - Kernel support (MSI, MSI-X)
2962 * - which can be user-defined (via MODULE_PARAM)
2963 * - Hardware queue count (num_*_queues)
2964 * - defined by miscellaneous hardware support/features (RSS, etc.)
2966 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
2970 /* Number of supported queues */
2971 ixgbe_set_num_queues(adapter);
2973 err = ixgbe_alloc_queues(adapter);
2975 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
2976 goto err_alloc_queues;
2979 err = ixgbe_set_interrupt_capability(adapter);
2981 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
2982 goto err_set_interrupt;
2985 DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
2986 "Tx Queue count = %u\n",
2987 (adapter->num_rx_queues > 1) ? "Enabled" :
2988 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2990 set_bit(__IXGBE_DOWN, &adapter->state);
2995 kfree(adapter->tx_ring);
2996 kfree(adapter->rx_ring);
3002 * ixgbe_sfp_timer - worker thread to find a missing module
3003 * @data: pointer to our adapter struct
3005 static void ixgbe_sfp_timer(unsigned long data)
3007 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3009 /* Do the sfp_timer outside of interrupt context due to the
3010 * delays that sfp+ detection requires
3012 schedule_work(&adapter->sfp_task);
3016 * ixgbe_sfp_task - worker thread to find a missing module
3017 * @work: pointer to work_struct containing our data
3019 static void ixgbe_sfp_task(struct work_struct *work)
3021 struct ixgbe_adapter *adapter = container_of(work,
3022 struct ixgbe_adapter,
3024 struct ixgbe_hw *hw = &adapter->hw;
3026 if ((hw->phy.type == ixgbe_phy_nl) &&
3027 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
3028 s32 ret = hw->phy.ops.identify_sfp(hw);
3031 ret = hw->phy.ops.reset(hw);
3032 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3033 DPRINTK(PROBE, ERR, "failed to initialize because an "
3034 "unsupported SFP+ module type was detected.\n"
3035 "Reload the driver after installing a "
3036 "supported module.\n");
3037 unregister_netdev(adapter->netdev);
3039 DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
3042 /* don't need this routine any more */
3043 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3047 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
3048 mod_timer(&adapter->sfp_timer,
3049 round_jiffies(jiffies + (2 * HZ)));
3053 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
3054 * @adapter: board private structure to initialize
3056 * ixgbe_sw_init initializes the Adapter private data structure.
3057 * Fields are initialized based on PCI device information and
3058 * OS network device settings (MTU size).
3060 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
3062 struct ixgbe_hw *hw = &adapter->hw;
3063 struct pci_dev *pdev = adapter->pdev;
3065 #ifdef CONFIG_IXGBE_DCB
3067 struct tc_configuration *tc;
3070 /* PCI config space info */
3072 hw->vendor_id = pdev->vendor;
3073 hw->device_id = pdev->device;
3074 hw->revision_id = pdev->revision;
3075 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3076 hw->subsystem_device_id = pdev->subsystem_device;
3078 /* Set capability flags */
3079 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
3080 adapter->ring_feature[RING_F_RSS].indices = rss;
3081 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
3082 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
3083 if (hw->mac.type == ixgbe_mac_82598EB)
3084 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
3085 else if (hw->mac.type == ixgbe_mac_82599EB)
3086 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
3088 #ifdef CONFIG_IXGBE_DCB
3089 /* Configure DCB traffic classes */
3090 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
3091 tc = &adapter->dcb_cfg.tc_config[j];
3092 tc->path[DCB_TX_CONFIG].bwg_id = 0;
3093 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
3094 tc->path[DCB_RX_CONFIG].bwg_id = 0;
3095 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
3096 tc->dcb_pfc = pfc_disabled;
3098 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
3099 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
3100 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
3101 adapter->dcb_cfg.round_robin_enable = false;
3102 adapter->dcb_set_bitmap = 0x00;
3103 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
3104 adapter->ring_feature[RING_F_DCB].indices);
3108 /* default flow control settings */
3109 hw->fc.requested_mode = ixgbe_fc_none;
3110 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
3111 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
3112 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
3113 hw->fc.send_xon = true;
3115 /* enable itr by default in dynamic mode */
3116 adapter->itr_setting = 1;
3117 adapter->eitr_param = 20000;
3119 /* set defaults for eitr in MegaBytes */
3120 adapter->eitr_low = 10;
3121 adapter->eitr_high = 20;
3123 /* set default ring sizes */
3124 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
3125 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
3127 /* initialize eeprom parameters */
3128 if (ixgbe_init_eeprom_params_generic(hw)) {
3129 dev_err(&pdev->dev, "EEPROM initialization failed\n");
3133 /* enable rx csum by default */
3134 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
3136 set_bit(__IXGBE_DOWN, &adapter->state);
3142 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
3143 * @adapter: board private structure
3144 * @tx_ring: tx descriptor ring (for a specific queue) to setup
3146 * Return 0 on success, negative on failure
3148 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
3149 struct ixgbe_ring *tx_ring)
3151 struct pci_dev *pdev = adapter->pdev;
3154 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3155 tx_ring->tx_buffer_info = vmalloc(size);
3156 if (!tx_ring->tx_buffer_info)
3158 memset(tx_ring->tx_buffer_info, 0, size);
3160 /* round up to nearest 4K */
3161 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3162 tx_ring->size = ALIGN(tx_ring->size, 4096);
3164 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
3169 tx_ring->next_to_use = 0;
3170 tx_ring->next_to_clean = 0;
3171 tx_ring->work_limit = tx_ring->count;
3175 vfree(tx_ring->tx_buffer_info);
3176 tx_ring->tx_buffer_info = NULL;
3177 DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
3178 "descriptor ring\n");
3183 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
3184 * @adapter: board private structure
3186 * If this function returns with an error, then it's possible one or
3187 * more of the rings is populated (while the rest are not). It is the
3188 * callers duty to clean those orphaned rings.
3190 * Return 0 on success, negative on failure
3192 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
3196 for (i = 0; i < adapter->num_tx_queues; i++) {
3197 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
3200 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
3208 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
3209 * @adapter: board private structure
3210 * @rx_ring: rx descriptor ring (for a specific queue) to setup
3212 * Returns 0 on success, negative on failure
3214 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
3215 struct ixgbe_ring *rx_ring)
3217 struct pci_dev *pdev = adapter->pdev;
3220 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3221 rx_ring->rx_buffer_info = vmalloc(size);
3222 if (!rx_ring->rx_buffer_info) {
3224 "vmalloc allocation failed for the rx desc ring\n");
3227 memset(rx_ring->rx_buffer_info, 0, size);
3229 /* Round up to nearest 4K */
3230 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3231 rx_ring->size = ALIGN(rx_ring->size, 4096);
3233 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
3235 if (!rx_ring->desc) {
3237 "Memory allocation failed for the rx desc ring\n");
3238 vfree(rx_ring->rx_buffer_info);
3242 rx_ring->next_to_clean = 0;
3243 rx_ring->next_to_use = 0;
3252 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
3253 * @adapter: board private structure
3255 * If this function returns with an error, then it's possible one or
3256 * more of the rings is populated (while the rest are not). It is the
3257 * callers duty to clean those orphaned rings.
3259 * Return 0 on success, negative on failure
3262 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
3266 for (i = 0; i < adapter->num_rx_queues; i++) {
3267 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
3270 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
3278 * ixgbe_free_tx_resources - Free Tx Resources per Queue
3279 * @adapter: board private structure
3280 * @tx_ring: Tx descriptor ring for a specific queue
3282 * Free all transmit software resources
3284 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
3285 struct ixgbe_ring *tx_ring)
3287 struct pci_dev *pdev = adapter->pdev;
3289 ixgbe_clean_tx_ring(adapter, tx_ring);
3291 vfree(tx_ring->tx_buffer_info);
3292 tx_ring->tx_buffer_info = NULL;
3294 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
3296 tx_ring->desc = NULL;
3300 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
3301 * @adapter: board private structure
3303 * Free all transmit software resources
3305 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
3309 for (i = 0; i < adapter->num_tx_queues; i++)
3310 ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
3314 * ixgbe_free_rx_resources - Free Rx Resources
3315 * @adapter: board private structure
3316 * @rx_ring: ring to clean the resources from
3318 * Free all receive software resources
3320 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
3321 struct ixgbe_ring *rx_ring)
3323 struct pci_dev *pdev = adapter->pdev;
3325 ixgbe_clean_rx_ring(adapter, rx_ring);
3327 vfree(rx_ring->rx_buffer_info);
3328 rx_ring->rx_buffer_info = NULL;
3330 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
3332 rx_ring->desc = NULL;
3336 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
3337 * @adapter: board private structure
3339 * Free all receive software resources
3341 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
3345 for (i = 0; i < adapter->num_rx_queues; i++)
3346 ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
3350 * ixgbe_change_mtu - Change the Maximum Transfer Unit
3351 * @netdev: network interface device structure
3352 * @new_mtu: new value for maximum frame size
3354 * Returns 0 on success, negative on failure
3356 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
3358 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3359 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3361 /* MTU < 68 is an error and causes problems on some kernels */
3362 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
3365 DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
3366 netdev->mtu, new_mtu);
3367 /* must set new MTU before calling down or up */
3368 netdev->mtu = new_mtu;
3370 if (netif_running(netdev))
3371 ixgbe_reinit_locked(adapter);
3377 * ixgbe_open - Called when a network interface is made active
3378 * @netdev: network interface device structure
3380 * Returns 0 on success, negative value on failure
3382 * The open entry point is called when a network interface is made
3383 * active by the system (IFF_UP). At this point all resources needed
3384 * for transmit and receive operations are allocated, the interrupt
3385 * handler is registered with the OS, the watchdog timer is started,
3386 * and the stack is notified that the interface is ready.
3388 static int ixgbe_open(struct net_device *netdev)
3390 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3393 /* disallow open during test */
3394 if (test_bit(__IXGBE_TESTING, &adapter->state))
3397 /* allocate transmit descriptors */
3398 err = ixgbe_setup_all_tx_resources(adapter);
3402 /* allocate receive descriptors */
3403 err = ixgbe_setup_all_rx_resources(adapter);
3407 ixgbe_configure(adapter);
3409 err = ixgbe_request_irq(adapter);
3413 err = ixgbe_up_complete(adapter);
3417 netif_tx_start_all_queues(netdev);
3422 ixgbe_release_hw_control(adapter);
3423 ixgbe_free_irq(adapter);
3425 ixgbe_free_all_rx_resources(adapter);
3427 ixgbe_free_all_tx_resources(adapter);
3429 ixgbe_reset(adapter);
3435 * ixgbe_close - Disables a network interface
3436 * @netdev: network interface device structure
3438 * Returns 0, this is not allowed to fail
3440 * The close entry point is called when an interface is de-activated
3441 * by the OS. The hardware is still under the drivers control, but
3442 * needs to be disabled. A global MAC reset is issued to stop the
3443 * hardware, and all transmit and receive resources are freed.
3445 static int ixgbe_close(struct net_device *netdev)
3447 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3449 ixgbe_down(adapter);
3450 ixgbe_free_irq(adapter);
3452 ixgbe_free_all_tx_resources(adapter);
3453 ixgbe_free_all_rx_resources(adapter);
3455 ixgbe_release_hw_control(adapter);
3461 * ixgbe_napi_add_all - prep napi structs for use
3462 * @adapter: private struct
3463 * helper function to napi_add each possible q_vector->napi
3465 void ixgbe_napi_add_all(struct ixgbe_adapter *adapter)
3467 int q_idx, q_vectors;
3468 struct net_device *netdev = adapter->netdev;
3469 int (*poll)(struct napi_struct *, int);
3471 /* check if we already have our netdev->napi_list populated */
3472 if (&netdev->napi_list != netdev->napi_list.next)
3475 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3476 poll = &ixgbe_clean_rxonly;
3477 /* Only enable as many vectors as we have rx queues. */
3478 q_vectors = adapter->num_rx_queues;
3481 /* only one q_vector for legacy modes */
3485 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3486 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
3487 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
3491 void ixgbe_napi_del_all(struct ixgbe_adapter *adapter)
3494 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3496 /* legacy and MSI only use one vector */
3497 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3500 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3501 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
3502 if (!q_vector->rxr_count)
3504 netif_napi_del(&q_vector->napi);
3509 static int ixgbe_resume(struct pci_dev *pdev)
3511 struct net_device *netdev = pci_get_drvdata(pdev);
3512 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3515 pci_set_power_state(pdev, PCI_D0);
3516 pci_restore_state(pdev);
3517 err = pci_enable_device(pdev);
3519 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
3523 pci_set_master(pdev);
3525 pci_enable_wake(pdev, PCI_D3hot, 0);
3526 pci_enable_wake(pdev, PCI_D3cold, 0);
3528 err = ixgbe_init_interrupt_scheme(adapter);
3530 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
3535 ixgbe_napi_add_all(adapter);
3536 ixgbe_reset(adapter);
3538 if (netif_running(netdev)) {
3539 err = ixgbe_open(adapter->netdev);
3544 netif_device_attach(netdev);
3549 #endif /* CONFIG_PM */
3550 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
3552 struct net_device *netdev = pci_get_drvdata(pdev);
3553 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3554 struct ixgbe_hw *hw = &adapter->hw;
3556 u32 wufc = adapter->wol;
3561 netif_device_detach(netdev);
3563 if (netif_running(netdev)) {
3564 ixgbe_down(adapter);
3565 ixgbe_free_irq(adapter);
3566 ixgbe_free_all_tx_resources(adapter);
3567 ixgbe_free_all_rx_resources(adapter);
3569 ixgbe_reset_interrupt_capability(adapter);
3570 ixgbe_napi_del_all(adapter);
3571 INIT_LIST_HEAD(&netdev->napi_list);
3572 kfree(adapter->tx_ring);
3573 kfree(adapter->rx_ring);
3576 retval = pci_save_state(pdev);
3581 ixgbe_set_rx_mode(netdev);
3583 /* turn on all-multi mode if wake on multicast is enabled */
3584 if (wufc & IXGBE_WUFC_MC) {
3585 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3586 fctrl |= IXGBE_FCTRL_MPE;
3587 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3590 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
3591 ctrl |= IXGBE_CTRL_GIO_DIS;
3592 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
3594 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
3596 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
3597 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
3600 if (wufc && hw->mac.type == ixgbe_mac_82599EB) {
3601 pci_enable_wake(pdev, PCI_D3hot, 1);
3602 pci_enable_wake(pdev, PCI_D3cold, 1);
3604 pci_enable_wake(pdev, PCI_D3hot, 0);
3605 pci_enable_wake(pdev, PCI_D3cold, 0);
3608 ixgbe_release_hw_control(adapter);
3610 pci_disable_device(pdev);
3612 pci_set_power_state(pdev, pci_choose_state(pdev, state));
3617 static void ixgbe_shutdown(struct pci_dev *pdev)
3619 ixgbe_suspend(pdev, PMSG_SUSPEND);
3623 * ixgbe_update_stats - Update the board statistics counters.
3624 * @adapter: board private structure
3626 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
3628 struct ixgbe_hw *hw = &adapter->hw;
3630 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
3632 if (hw->mac.type == ixgbe_mac_82599EB) {
3633 for (i = 0; i < 16; i++)
3634 adapter->hw_rx_no_dma_resources +=
3635 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
3638 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
3639 for (i = 0; i < 8; i++) {
3640 /* for packet buffers not used, the register should read 0 */
3641 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
3643 adapter->stats.mpc[i] += mpc;
3644 total_mpc += adapter->stats.mpc[i];
3645 if (hw->mac.type == ixgbe_mac_82598EB)
3646 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
3647 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
3648 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
3649 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
3650 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
3651 if (hw->mac.type == ixgbe_mac_82599EB) {
3652 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
3653 IXGBE_PXONRXCNT(i));
3654 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
3655 IXGBE_PXOFFRXCNT(i));
3656 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
3658 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
3660 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
3663 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
3665 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
3668 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
3669 /* work around hardware counting issue */
3670 adapter->stats.gprc -= missed_rx;
3672 /* 82598 hardware only has a 32 bit counter in the high register */
3673 if (hw->mac.type == ixgbe_mac_82599EB) {
3674 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
3675 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
3676 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
3677 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
3678 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
3679 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
3680 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
3681 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
3683 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
3684 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
3685 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
3686 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
3687 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
3689 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
3690 adapter->stats.bprc += bprc;
3691 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
3692 if (hw->mac.type == ixgbe_mac_82598EB)
3693 adapter->stats.mprc -= bprc;
3694 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
3695 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
3696 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
3697 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
3698 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
3699 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
3700 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
3701 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
3702 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
3703 adapter->stats.lxontxc += lxon;
3704 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
3705 adapter->stats.lxofftxc += lxoff;
3706 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3707 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
3708 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
3710 * 82598 errata - tx of flow control packets is included in tx counters
3712 xon_off_tot = lxon + lxoff;
3713 adapter->stats.gptc -= xon_off_tot;
3714 adapter->stats.mptc -= xon_off_tot;
3715 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
3716 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3717 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
3718 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
3719 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
3720 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
3721 adapter->stats.ptc64 -= xon_off_tot;
3722 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
3723 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
3724 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
3725 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
3726 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
3727 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
3729 /* Fill out the OS statistics structure */
3730 adapter->net_stats.multicast = adapter->stats.mprc;
3733 adapter->net_stats.rx_errors = adapter->stats.crcerrs +
3734 adapter->stats.rlec;
3735 adapter->net_stats.rx_dropped = 0;
3736 adapter->net_stats.rx_length_errors = adapter->stats.rlec;
3737 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3738 adapter->net_stats.rx_missed_errors = total_mpc;
3742 * ixgbe_watchdog - Timer Call-back
3743 * @data: pointer to adapter cast into an unsigned long
3745 static void ixgbe_watchdog(unsigned long data)
3747 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3748 struct ixgbe_hw *hw = &adapter->hw;
3750 /* Do the watchdog outside of interrupt context due to the lovely
3751 * delays that some of the newer hardware requires */
3752 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
3753 /* Cause software interrupt to ensure rx rings are cleaned */
3754 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3756 (1 << (adapter->num_msix_vectors - NON_Q_VECTORS)) - 1;
3757 IXGBE_WRITE_REG(hw, IXGBE_EICS, eics);
3759 /* For legacy and MSI interrupts don't set any bits that
3760 * are enabled for EIAM, because this operation would
3761 * set *both* EIMS and EICS for any bit in EIAM */
3762 IXGBE_WRITE_REG(hw, IXGBE_EICS,
3763 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
3765 /* Reset the timer */
3766 mod_timer(&adapter->watchdog_timer,
3767 round_jiffies(jiffies + 2 * HZ));
3770 schedule_work(&adapter->watchdog_task);
3774 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
3775 * @work: pointer to work_struct containing our data
3777 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
3779 struct ixgbe_adapter *adapter = container_of(work,
3780 struct ixgbe_adapter,
3781 multispeed_fiber_task);
3782 struct ixgbe_hw *hw = &adapter->hw;
3785 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
3786 if (hw->mac.ops.get_link_capabilities)
3787 hw->mac.ops.get_link_capabilities(hw, &autoneg,
3789 if (hw->mac.ops.setup_link_speed)
3790 hw->mac.ops.setup_link_speed(hw, autoneg, true, true);
3791 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3792 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
3796 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
3797 * @work: pointer to work_struct containing our data
3799 static void ixgbe_sfp_config_module_task(struct work_struct *work)
3801 struct ixgbe_adapter *adapter = container_of(work,
3802 struct ixgbe_adapter,
3803 sfp_config_module_task);
3804 struct ixgbe_hw *hw = &adapter->hw;
3807 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
3808 err = hw->phy.ops.identify_sfp(hw);
3809 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3810 DPRINTK(PROBE, ERR, "PHY not supported on this NIC %d\n", err);
3811 ixgbe_down(adapter);
3814 hw->mac.ops.setup_sfp(hw);
3816 if (!adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK)
3817 /* This will also work for DA Twinax connections */
3818 schedule_work(&adapter->multispeed_fiber_task);
3819 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
3823 * ixgbe_watchdog_task - worker thread to bring link up
3824 * @work: pointer to work_struct containing our data
3826 static void ixgbe_watchdog_task(struct work_struct *work)
3828 struct ixgbe_adapter *adapter = container_of(work,
3829 struct ixgbe_adapter,
3831 struct net_device *netdev = adapter->netdev;
3832 struct ixgbe_hw *hw = &adapter->hw;
3833 u32 link_speed = adapter->link_speed;
3834 bool link_up = adapter->link_up;
3836 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
3838 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
3839 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3841 time_after(jiffies, (adapter->link_check_timeout +
3842 IXGBE_TRY_LINK_TIMEOUT))) {
3843 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
3844 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
3846 adapter->link_up = link_up;
3847 adapter->link_speed = link_speed;
3851 if (!netif_carrier_ok(netdev)) {
3852 bool flow_rx, flow_tx;
3854 if (hw->mac.type == ixgbe_mac_82599EB) {
3855 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
3856 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
3857 flow_rx = (mflcn & IXGBE_MFLCN_RFCE);
3858 flow_tx = (fccfg & IXGBE_FCCFG_TFCE_802_3X);
3860 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3861 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
3862 flow_rx = (frctl & IXGBE_FCTRL_RFCE);
3863 flow_tx = (rmcs & IXGBE_RMCS_TFCE_802_3X);
3866 printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
3867 "Flow Control: %s\n",
3869 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
3871 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
3872 "1 Gbps" : "unknown speed")),
3873 ((flow_rx && flow_tx) ? "RX/TX" :
3875 (flow_tx ? "TX" : "None"))));
3877 netif_carrier_on(netdev);
3879 /* Force detection of hung controller */
3880 adapter->detect_tx_hung = true;
3883 adapter->link_up = false;
3884 adapter->link_speed = 0;
3885 if (netif_carrier_ok(netdev)) {
3886 printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
3888 netif_carrier_off(netdev);
3892 ixgbe_update_stats(adapter);
3893 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
3896 static int ixgbe_tso(struct ixgbe_adapter *adapter,
3897 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
3898 u32 tx_flags, u8 *hdr_len)
3900 struct ixgbe_adv_tx_context_desc *context_desc;
3903 struct ixgbe_tx_buffer *tx_buffer_info;
3904 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
3905 u32 mss_l4len_idx, l4len;
3907 if (skb_is_gso(skb)) {
3908 if (skb_header_cloned(skb)) {
3909 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3913 l4len = tcp_hdrlen(skb);
3916 if (skb->protocol == htons(ETH_P_IP)) {
3917 struct iphdr *iph = ip_hdr(skb);
3920 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3924 adapter->hw_tso_ctxt++;
3925 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3926 ipv6_hdr(skb)->payload_len = 0;
3927 tcp_hdr(skb)->check =
3928 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3929 &ipv6_hdr(skb)->daddr,
3931 adapter->hw_tso6_ctxt++;
3934 i = tx_ring->next_to_use;
3936 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3937 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
3939 /* VLAN MACLEN IPLEN */
3940 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3942 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
3943 vlan_macip_lens |= ((skb_network_offset(skb)) <<
3944 IXGBE_ADVTXD_MACLEN_SHIFT);
3945 *hdr_len += skb_network_offset(skb);
3947 (skb_transport_header(skb) - skb_network_header(skb));
3949 (skb_transport_header(skb) - skb_network_header(skb));
3950 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3951 context_desc->seqnum_seed = 0;
3953 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3954 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
3955 IXGBE_ADVTXD_DTYP_CTXT);
3957 if (skb->protocol == htons(ETH_P_IP))
3958 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
3959 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3960 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
3964 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
3965 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
3966 /* use index 1 for TSO */
3967 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3968 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3970 tx_buffer_info->time_stamp = jiffies;
3971 tx_buffer_info->next_to_watch = i;
3974 if (i == tx_ring->count)
3976 tx_ring->next_to_use = i;
3983 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
3984 struct ixgbe_ring *tx_ring,
3985 struct sk_buff *skb, u32 tx_flags)
3987 struct ixgbe_adv_tx_context_desc *context_desc;
3989 struct ixgbe_tx_buffer *tx_buffer_info;
3990 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
3992 if (skb->ip_summed == CHECKSUM_PARTIAL ||
3993 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
3994 i = tx_ring->next_to_use;
3995 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3996 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
3998 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4000 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4001 vlan_macip_lens |= (skb_network_offset(skb) <<
4002 IXGBE_ADVTXD_MACLEN_SHIFT);
4003 if (skb->ip_summed == CHECKSUM_PARTIAL)
4004 vlan_macip_lens |= (skb_transport_header(skb) -
4005 skb_network_header(skb));
4007 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4008 context_desc->seqnum_seed = 0;
4010 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
4011 IXGBE_ADVTXD_DTYP_CTXT);
4013 if (skb->ip_summed == CHECKSUM_PARTIAL) {
4014 switch (skb->protocol) {
4015 case cpu_to_be16(ETH_P_IP):
4016 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4017 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4019 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4021 case cpu_to_be16(ETH_P_IPV6):
4022 /* XXX what about other V6 headers?? */
4023 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4025 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4028 if (unlikely(net_ratelimit())) {
4029 DPRINTK(PROBE, WARNING,
4030 "partial checksum but proto=%x!\n",
4037 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4038 /* use index zero for tx checksum offload */
4039 context_desc->mss_l4len_idx = 0;
4041 tx_buffer_info->time_stamp = jiffies;
4042 tx_buffer_info->next_to_watch = i;
4044 adapter->hw_csum_tx_good++;
4046 if (i == tx_ring->count)
4048 tx_ring->next_to_use = i;
4056 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
4057 struct ixgbe_ring *tx_ring,
4058 struct sk_buff *skb, unsigned int first)
4060 struct ixgbe_tx_buffer *tx_buffer_info;
4061 unsigned int len = skb->len;
4062 unsigned int offset = 0, size, count = 0, i;
4063 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
4066 len -= skb->data_len;
4068 i = tx_ring->next_to_use;
4071 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4072 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4074 tx_buffer_info->length = size;
4075 tx_buffer_info->dma = pci_map_single(adapter->pdev,
4077 size, PCI_DMA_TODEVICE);
4078 tx_buffer_info->time_stamp = jiffies;
4079 tx_buffer_info->next_to_watch = i;
4085 if (i == tx_ring->count)
4089 for (f = 0; f < nr_frags; f++) {
4090 struct skb_frag_struct *frag;
4092 frag = &skb_shinfo(skb)->frags[f];
4094 offset = frag->page_offset;
4097 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4098 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
4100 tx_buffer_info->length = size;
4101 tx_buffer_info->dma = pci_map_page(adapter->pdev,
4106 tx_buffer_info->time_stamp = jiffies;
4107 tx_buffer_info->next_to_watch = i;
4113 if (i == tx_ring->count)
4118 i = tx_ring->count - 1;
4121 tx_ring->tx_buffer_info[i].skb = skb;
4122 tx_ring->tx_buffer_info[first].next_to_watch = i;
4127 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
4128 struct ixgbe_ring *tx_ring,
4129 int tx_flags, int count, u32 paylen, u8 hdr_len)
4131 union ixgbe_adv_tx_desc *tx_desc = NULL;
4132 struct ixgbe_tx_buffer *tx_buffer_info;
4133 u32 olinfo_status = 0, cmd_type_len = 0;
4135 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
4137 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
4139 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
4141 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4142 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
4144 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
4145 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
4147 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
4148 IXGBE_ADVTXD_POPTS_SHIFT;
4150 /* use index 1 context for tso */
4151 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4152 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
4153 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
4154 IXGBE_ADVTXD_POPTS_SHIFT;
4156 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
4157 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
4158 IXGBE_ADVTXD_POPTS_SHIFT;
4160 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
4162 i = tx_ring->next_to_use;
4164 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4165 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
4166 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
4167 tx_desc->read.cmd_type_len =
4168 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
4169 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4171 if (i == tx_ring->count)
4175 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
4178 * Force memory writes to complete before letting h/w
4179 * know there are new descriptors to fetch. (Only
4180 * applicable for weak-ordered memory model archs,
4185 tx_ring->next_to_use = i;
4186 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4189 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
4190 struct ixgbe_ring *tx_ring, int size)
4192 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4194 netif_stop_subqueue(netdev, tx_ring->queue_index);
4195 /* Herbert's original patch had:
4196 * smp_mb__after_netif_stop_queue();
4197 * but since that doesn't exist yet, just open code it. */
4200 /* We need to check again in a case another CPU has just
4201 * made room available. */
4202 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
4205 /* A reprieve! - use start_queue because it doesn't call schedule */
4206 netif_start_subqueue(netdev, tx_ring->queue_index);
4207 ++adapter->restart_queue;
4211 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
4212 struct ixgbe_ring *tx_ring, int size)
4214 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
4216 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
4219 static int ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4221 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4222 struct ixgbe_ring *tx_ring;
4224 unsigned int tx_flags = 0;
4230 r_idx = (adapter->num_tx_queues - 1) & skb->queue_mapping;
4231 tx_ring = &adapter->tx_ring[r_idx];
4233 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4234 tx_flags |= vlan_tx_tag_get(skb);
4235 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4236 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
4237 tx_flags |= (skb->queue_mapping << 13);
4239 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4240 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4241 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4242 tx_flags |= (skb->queue_mapping << 13);
4243 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4244 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4246 /* three things can cause us to need a context descriptor */
4247 if (skb_is_gso(skb) ||
4248 (skb->ip_summed == CHECKSUM_PARTIAL) ||
4249 (tx_flags & IXGBE_TX_FLAGS_VLAN))
4252 count += TXD_USE_COUNT(skb_headlen(skb));
4253 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4254 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4256 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
4258 return NETDEV_TX_BUSY;
4261 if (skb->protocol == htons(ETH_P_IP))
4262 tx_flags |= IXGBE_TX_FLAGS_IPV4;
4263 first = tx_ring->next_to_use;
4264 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
4266 dev_kfree_skb_any(skb);
4267 return NETDEV_TX_OK;
4271 tx_flags |= IXGBE_TX_FLAGS_TSO;
4272 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
4273 (skb->ip_summed == CHECKSUM_PARTIAL))
4274 tx_flags |= IXGBE_TX_FLAGS_CSUM;
4276 ixgbe_tx_queue(adapter, tx_ring, tx_flags,
4277 ixgbe_tx_map(adapter, tx_ring, skb, first),
4280 netdev->trans_start = jiffies;
4282 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
4284 return NETDEV_TX_OK;
4288 * ixgbe_get_stats - Get System Network Statistics
4289 * @netdev: network interface device structure
4291 * Returns the address of the device statistics structure.
4292 * The statistics are actually updated from the timer callback.
4294 static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
4296 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4298 /* only return the current stats */
4299 return &adapter->net_stats;
4303 * ixgbe_set_mac - Change the Ethernet Address of the NIC
4304 * @netdev: network interface device structure
4305 * @p: pointer to an address structure
4307 * Returns 0 on success, negative on failure
4309 static int ixgbe_set_mac(struct net_device *netdev, void *p)
4311 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4312 struct ixgbe_hw *hw = &adapter->hw;
4313 struct sockaddr *addr = p;
4315 if (!is_valid_ether_addr(addr->sa_data))
4316 return -EADDRNOTAVAIL;
4318 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4319 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
4321 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
4326 #ifdef CONFIG_NET_POLL_CONTROLLER
4328 * Polling 'interrupt' - used by things like netconsole to send skbs
4329 * without having to re-enable interrupts. It's not called while
4330 * the interrupt routine is executing.
4332 static void ixgbe_netpoll(struct net_device *netdev)
4334 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4336 disable_irq(adapter->pdev->irq);
4337 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
4338 ixgbe_intr(adapter->pdev->irq, netdev);
4339 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
4340 enable_irq(adapter->pdev->irq);
4344 static const struct net_device_ops ixgbe_netdev_ops = {
4345 .ndo_open = ixgbe_open,
4346 .ndo_stop = ixgbe_close,
4347 .ndo_start_xmit = ixgbe_xmit_frame,
4348 .ndo_get_stats = ixgbe_get_stats,
4349 .ndo_set_multicast_list = ixgbe_set_rx_mode,
4350 .ndo_validate_addr = eth_validate_addr,
4351 .ndo_set_mac_address = ixgbe_set_mac,
4352 .ndo_change_mtu = ixgbe_change_mtu,
4353 .ndo_tx_timeout = ixgbe_tx_timeout,
4354 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
4355 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
4356 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
4357 #ifdef CONFIG_NET_POLL_CONTROLLER
4358 .ndo_poll_controller = ixgbe_netpoll,
4363 * ixgbe_probe - Device Initialization Routine
4364 * @pdev: PCI device information struct
4365 * @ent: entry in ixgbe_pci_tbl
4367 * Returns 0 on success, negative on failure
4369 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
4370 * The OS initialization, configuring of the adapter private structure,
4371 * and a hardware reset occur.
4373 static int __devinit ixgbe_probe(struct pci_dev *pdev,
4374 const struct pci_device_id *ent)
4376 struct net_device *netdev;
4377 struct ixgbe_adapter *adapter = NULL;
4378 struct ixgbe_hw *hw;
4379 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
4380 static int cards_found;
4381 int i, err, pci_using_dac;
4385 err = pci_enable_device(pdev);
4389 if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK) &&
4390 !pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK)) {
4393 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
4395 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
4397 dev_err(&pdev->dev, "No usable DMA "
4398 "configuration, aborting\n");
4405 err = pci_request_regions(pdev, ixgbe_driver_name);
4407 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4411 err = pci_enable_pcie_error_reporting(pdev);
4413 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
4415 /* non-fatal, continue */
4418 pci_set_master(pdev);
4419 pci_save_state(pdev);
4421 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
4424 goto err_alloc_etherdev;
4427 SET_NETDEV_DEV(netdev, &pdev->dev);
4429 pci_set_drvdata(pdev, netdev);
4430 adapter = netdev_priv(netdev);
4432 adapter->netdev = netdev;
4433 adapter->pdev = pdev;
4436 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
4438 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4439 pci_resource_len(pdev, 0));
4445 for (i = 1; i <= 5; i++) {
4446 if (pci_resource_len(pdev, i) == 0)
4450 netdev->netdev_ops = &ixgbe_netdev_ops;
4451 ixgbe_set_ethtool_ops(netdev);
4452 netdev->watchdog_timeo = 5 * HZ;
4453 strcpy(netdev->name, pci_name(pdev));
4455 adapter->bd_number = cards_found;
4458 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4459 hw->mac.type = ii->mac;
4462 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
4463 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
4464 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
4465 if (!(eec & (1 << 8)))
4466 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
4469 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
4470 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
4472 /* set up this timer and work struct before calling get_invariants
4473 * which might start the timer
4475 init_timer(&adapter->sfp_timer);
4476 adapter->sfp_timer.function = &ixgbe_sfp_timer;
4477 adapter->sfp_timer.data = (unsigned long) adapter;
4479 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
4481 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
4482 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
4484 /* a new SFP+ module arrival, called from GPI SDP2 context */
4485 INIT_WORK(&adapter->sfp_config_module_task,
4486 ixgbe_sfp_config_module_task);
4488 err = ii->get_invariants(hw);
4489 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
4490 /* start a kernel thread to watch for a module to arrive */
4491 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4492 mod_timer(&adapter->sfp_timer,
4493 round_jiffies(jiffies + (2 * HZ)));
4495 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4496 DPRINTK(PROBE, ERR, "failed to load because an "
4497 "unsupported SFP+ module type was detected.\n");
4503 /* setup the private structure */
4504 err = ixgbe_sw_init(adapter);
4508 /* reset_hw fills in the perm_addr as well */
4509 err = hw->mac.ops.reset_hw(hw);
4511 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
4515 netdev->features = NETIF_F_SG |
4517 NETIF_F_HW_VLAN_TX |
4518 NETIF_F_HW_VLAN_RX |
4519 NETIF_F_HW_VLAN_FILTER;
4521 netdev->features |= NETIF_F_IPV6_CSUM;
4522 netdev->features |= NETIF_F_TSO;
4523 netdev->features |= NETIF_F_TSO6;
4524 netdev->features |= NETIF_F_GRO;
4526 netdev->vlan_features |= NETIF_F_TSO;
4527 netdev->vlan_features |= NETIF_F_TSO6;
4528 netdev->vlan_features |= NETIF_F_IP_CSUM;
4529 netdev->vlan_features |= NETIF_F_SG;
4531 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
4532 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4534 #ifdef CONFIG_IXGBE_DCB
4535 netdev->dcbnl_ops = &dcbnl_ops;
4539 netdev->features |= NETIF_F_HIGHDMA;
4541 /* make sure the EEPROM is good */
4542 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
4543 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
4548 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
4549 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
4551 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
4552 dev_err(&pdev->dev, "invalid MAC address\n");
4557 init_timer(&adapter->watchdog_timer);
4558 adapter->watchdog_timer.function = &ixgbe_watchdog;
4559 adapter->watchdog_timer.data = (unsigned long)adapter;
4561 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
4562 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
4564 err = ixgbe_init_interrupt_scheme(adapter);
4568 switch (pdev->device) {
4569 case IXGBE_DEV_ID_82599_KX4:
4570 #define IXGBE_PCIE_PMCSR 0x44
4571 adapter->wol = IXGBE_WUFC_MAG;
4572 pci_read_config_word(pdev, IXGBE_PCIE_PMCSR, &pm_value);
4573 pci_write_config_word(pdev, IXGBE_PCIE_PMCSR,
4574 (pm_value | (1 << 8)));
4580 device_init_wakeup(&adapter->pdev->dev, true);
4581 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
4583 /* print bus type/speed/width info */
4584 dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
4585 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
4586 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
4587 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
4588 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
4589 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
4592 ixgbe_read_pba_num_generic(hw, &part_num);
4593 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
4594 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
4595 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
4596 (part_num >> 8), (part_num & 0xff));
4598 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
4599 hw->mac.type, hw->phy.type,
4600 (part_num >> 8), (part_num & 0xff));
4602 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
4603 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
4604 "this card is not sufficient for optimal "
4606 dev_warn(&pdev->dev, "For optimal performance a x8 "
4607 "PCI-Express slot is required.\n");
4610 /* save off EEPROM version number */
4611 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
4613 /* reset the hardware with the new settings */
4614 hw->mac.ops.start_hw(hw);
4616 netif_carrier_off(netdev);
4618 strcpy(netdev->name, "eth%d");
4619 err = register_netdev(netdev);
4623 #ifdef CONFIG_IXGBE_DCA
4624 if (dca_add_requester(&pdev->dev) == 0) {
4625 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
4626 /* always use CB2 mode, difference is masked
4627 * in the CB driver */
4628 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
4629 ixgbe_setup_dca(adapter);
4633 dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
4638 ixgbe_release_hw_control(adapter);
4641 ixgbe_reset_interrupt_capability(adapter);
4643 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4644 del_timer_sync(&adapter->sfp_timer);
4645 cancel_work_sync(&adapter->sfp_task);
4646 cancel_work_sync(&adapter->multispeed_fiber_task);
4647 cancel_work_sync(&adapter->sfp_config_module_task);
4648 iounmap(hw->hw_addr);
4650 free_netdev(netdev);
4652 pci_release_regions(pdev);
4655 pci_disable_device(pdev);
4660 * ixgbe_remove - Device Removal Routine
4661 * @pdev: PCI device information struct
4663 * ixgbe_remove is called by the PCI subsystem to alert the driver
4664 * that it should release a PCI device. The could be caused by a
4665 * Hot-Plug event, or because the driver is going to be removed from
4668 static void __devexit ixgbe_remove(struct pci_dev *pdev)
4670 struct net_device *netdev = pci_get_drvdata(pdev);
4671 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4674 set_bit(__IXGBE_DOWN, &adapter->state);
4675 /* clear the module not found bit to make sure the worker won't
4678 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4679 del_timer_sync(&adapter->watchdog_timer);
4681 del_timer_sync(&adapter->sfp_timer);
4682 cancel_work_sync(&adapter->watchdog_task);
4683 cancel_work_sync(&adapter->sfp_task);
4684 cancel_work_sync(&adapter->multispeed_fiber_task);
4685 cancel_work_sync(&adapter->sfp_config_module_task);
4686 flush_scheduled_work();
4688 #ifdef CONFIG_IXGBE_DCA
4689 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
4690 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
4691 dca_remove_requester(&pdev->dev);
4692 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
4696 if (netdev->reg_state == NETREG_REGISTERED)
4697 unregister_netdev(netdev);
4699 ixgbe_reset_interrupt_capability(adapter);
4701 ixgbe_release_hw_control(adapter);
4703 iounmap(adapter->hw.hw_addr);
4704 pci_release_regions(pdev);
4706 DPRINTK(PROBE, INFO, "complete\n");
4707 kfree(adapter->tx_ring);
4708 kfree(adapter->rx_ring);
4710 free_netdev(netdev);
4712 err = pci_disable_pcie_error_reporting(pdev);
4715 "pci_disable_pcie_error_reporting failed 0x%x\n", err);
4717 pci_disable_device(pdev);
4721 * ixgbe_io_error_detected - called when PCI error is detected
4722 * @pdev: Pointer to PCI device
4723 * @state: The current pci connection state
4725 * This function is called after a PCI bus error affecting
4726 * this device has been detected.
4728 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
4729 pci_channel_state_t state)
4731 struct net_device *netdev = pci_get_drvdata(pdev);
4732 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4734 netif_device_detach(netdev);
4736 if (netif_running(netdev))
4737 ixgbe_down(adapter);
4738 pci_disable_device(pdev);
4740 /* Request a slot reset. */
4741 return PCI_ERS_RESULT_NEED_RESET;
4745 * ixgbe_io_slot_reset - called after the pci bus has been reset.
4746 * @pdev: Pointer to PCI device
4748 * Restart the card from scratch, as if from a cold-boot.
4750 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
4752 struct net_device *netdev = pci_get_drvdata(pdev);
4753 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4754 pci_ers_result_t result;
4757 if (pci_enable_device(pdev)) {
4759 "Cannot re-enable PCI device after reset.\n");
4760 result = PCI_ERS_RESULT_DISCONNECT;
4762 pci_set_master(pdev);
4763 pci_restore_state(pdev);
4765 pci_enable_wake(pdev, PCI_D3hot, 0);
4766 pci_enable_wake(pdev, PCI_D3cold, 0);
4768 ixgbe_reset(adapter);
4770 result = PCI_ERS_RESULT_RECOVERED;
4773 err = pci_cleanup_aer_uncorrect_error_status(pdev);
4776 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
4777 /* non-fatal, continue */
4784 * ixgbe_io_resume - called when traffic can start flowing again.
4785 * @pdev: Pointer to PCI device
4787 * This callback is called when the error recovery driver tells us that
4788 * its OK to resume normal operation.
4790 static void ixgbe_io_resume(struct pci_dev *pdev)
4792 struct net_device *netdev = pci_get_drvdata(pdev);
4793 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4795 if (netif_running(netdev)) {
4796 if (ixgbe_up(adapter)) {
4797 DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
4802 netif_device_attach(netdev);
4805 static struct pci_error_handlers ixgbe_err_handler = {
4806 .error_detected = ixgbe_io_error_detected,
4807 .slot_reset = ixgbe_io_slot_reset,
4808 .resume = ixgbe_io_resume,
4811 static struct pci_driver ixgbe_driver = {
4812 .name = ixgbe_driver_name,
4813 .id_table = ixgbe_pci_tbl,
4814 .probe = ixgbe_probe,
4815 .remove = __devexit_p(ixgbe_remove),
4817 .suspend = ixgbe_suspend,
4818 .resume = ixgbe_resume,
4820 .shutdown = ixgbe_shutdown,
4821 .err_handler = &ixgbe_err_handler
4825 * ixgbe_init_module - Driver Registration Routine
4827 * ixgbe_init_module is the first routine called when the driver is
4828 * loaded. All it does is register with the PCI subsystem.
4830 static int __init ixgbe_init_module(void)
4833 printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
4834 ixgbe_driver_string, ixgbe_driver_version);
4836 printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
4838 #ifdef CONFIG_IXGBE_DCA
4839 dca_register_notify(&dca_notifier);
4842 ret = pci_register_driver(&ixgbe_driver);
4846 module_init(ixgbe_init_module);
4849 * ixgbe_exit_module - Driver Exit Cleanup Routine
4851 * ixgbe_exit_module is called just before the driver is removed
4854 static void __exit ixgbe_exit_module(void)
4856 #ifdef CONFIG_IXGBE_DCA
4857 dca_unregister_notify(&dca_notifier);
4859 pci_unregister_driver(&ixgbe_driver);
4862 #ifdef CONFIG_IXGBE_DCA
4863 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
4868 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
4869 __ixgbe_notify_dca);
4871 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4873 #endif /* CONFIG_IXGBE_DCA */
4875 module_exit(ixgbe_exit_module);