1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2008 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 "1.3.30-k2"
51 const char ixgbe_driver_version[] = DRV_VERSION;
52 static char ixgbe_copyright[] = "Copyright (c) 1999-2007 Intel Corporation.";
54 static const struct ixgbe_info *ixgbe_info_tbl[] = {
55 [board_82598] = &ixgbe_82598_info,
58 /* ixgbe_pci_tbl - PCI Device ID Table
60 * Wildcard entries (PCI_ANY_ID) should come last
61 * Last entry must be all 0s
63 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
64 * Class, Class Mask, private data (not used) }
66 static struct pci_device_id ixgbe_pci_tbl[] = {
67 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
69 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
71 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
73 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
75 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
78 /* required last entry */
81 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
83 #ifdef CONFIG_IXGBE_DCA
84 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
86 static struct notifier_block dca_notifier = {
87 .notifier_call = ixgbe_notify_dca,
93 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
94 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
95 MODULE_LICENSE("GPL");
96 MODULE_VERSION(DRV_VERSION);
98 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
100 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
104 /* Let firmware take over control of h/w */
105 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
106 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
107 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
110 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
114 /* Let firmware know the driver has taken over */
115 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
116 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
117 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
120 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, u16 int_alloc_entry,
125 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
126 index = (int_alloc_entry >> 2) & 0x1F;
127 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR(index));
128 ivar &= ~(0xFF << (8 * (int_alloc_entry & 0x3)));
129 ivar |= (msix_vector << (8 * (int_alloc_entry & 0x3)));
130 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR(index), ivar);
133 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
134 struct ixgbe_tx_buffer
137 if (tx_buffer_info->dma) {
138 pci_unmap_page(adapter->pdev, tx_buffer_info->dma,
139 tx_buffer_info->length, PCI_DMA_TODEVICE);
140 tx_buffer_info->dma = 0;
142 if (tx_buffer_info->skb) {
143 dev_kfree_skb_any(tx_buffer_info->skb);
144 tx_buffer_info->skb = NULL;
146 /* tx_buffer_info must be completely set up in the transmit path */
149 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
150 struct ixgbe_ring *tx_ring,
153 struct ixgbe_hw *hw = &adapter->hw;
156 /* Detect a transmit hang in hardware, this serializes the
157 * check with the clearing of time_stamp and movement of eop */
158 head = IXGBE_READ_REG(hw, tx_ring->head);
159 tail = IXGBE_READ_REG(hw, tx_ring->tail);
160 adapter->detect_tx_hung = false;
161 if ((head != tail) &&
162 tx_ring->tx_buffer_info[eop].time_stamp &&
163 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
164 !(IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & IXGBE_TFCS_TXOFF)) {
165 /* detected Tx unit hang */
166 union ixgbe_adv_tx_desc *tx_desc;
167 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
168 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
170 " TDH, TDT <%x>, <%x>\n"
171 " next_to_use <%x>\n"
172 " next_to_clean <%x>\n"
173 "tx_buffer_info[next_to_clean]\n"
174 " time_stamp <%lx>\n"
176 tx_ring->queue_index,
178 tx_ring->next_to_use, eop,
179 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
186 #define IXGBE_MAX_TXD_PWR 14
187 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
189 /* Tx Descriptors needed, worst case */
190 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
191 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
192 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
193 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
195 #define GET_TX_HEAD_FROM_RING(ring) (\
197 ((union ixgbe_adv_tx_desc *)(ring)->desc + (ring)->count))
198 static void ixgbe_tx_timeout(struct net_device *netdev);
201 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
202 * @adapter: board private structure
203 * @tx_ring: tx ring to clean
205 static bool ixgbe_clean_tx_irq(struct ixgbe_adapter *adapter,
206 struct ixgbe_ring *tx_ring)
208 union ixgbe_adv_tx_desc *tx_desc;
209 struct ixgbe_tx_buffer *tx_buffer_info;
210 struct net_device *netdev = adapter->netdev;
214 unsigned int count = 0;
215 unsigned int total_bytes = 0, total_packets = 0;
218 head = GET_TX_HEAD_FROM_RING(tx_ring);
219 head = le32_to_cpu(head);
220 i = tx_ring->next_to_clean;
223 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
224 tx_buffer_info = &tx_ring->tx_buffer_info[i];
225 skb = tx_buffer_info->skb;
228 unsigned int segs, bytecount;
230 /* gso_segs is currently only valid for tcp */
231 segs = skb_shinfo(skb)->gso_segs ?: 1;
232 /* multiply data chunks by size of headers */
233 bytecount = ((segs - 1) * skb_headlen(skb)) +
235 total_packets += segs;
236 total_bytes += bytecount;
239 ixgbe_unmap_and_free_tx_resource(adapter,
243 if (i == tx_ring->count)
247 if (count == tx_ring->count)
252 head = GET_TX_HEAD_FROM_RING(tx_ring);
253 head = le32_to_cpu(head);
259 tx_ring->next_to_clean = i;
261 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
262 if (unlikely(count && netif_carrier_ok(netdev) &&
263 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
264 /* Make sure that anybody stopping the queue after this
265 * sees the new next_to_clean.
268 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
269 !test_bit(__IXGBE_DOWN, &adapter->state)) {
270 netif_wake_subqueue(netdev, tx_ring->queue_index);
271 ++adapter->restart_queue;
275 if (adapter->detect_tx_hung) {
276 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
277 /* schedule immediate reset if we believe we hung */
279 "tx hang %d detected, resetting adapter\n",
280 adapter->tx_timeout_count + 1);
281 ixgbe_tx_timeout(adapter->netdev);
285 /* re-arm the interrupt */
286 if ((total_packets >= tx_ring->work_limit) ||
287 (count == tx_ring->count))
288 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, tx_ring->v_idx);
290 tx_ring->total_bytes += total_bytes;
291 tx_ring->total_packets += total_packets;
292 tx_ring->stats.bytes += total_bytes;
293 tx_ring->stats.packets += total_packets;
294 adapter->net_stats.tx_bytes += total_bytes;
295 adapter->net_stats.tx_packets += total_packets;
296 return (total_packets ? true : false);
299 #ifdef CONFIG_IXGBE_DCA
300 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
301 struct ixgbe_ring *rx_ring)
305 int q = rx_ring - adapter->rx_ring;
307 if (rx_ring->cpu != cpu) {
308 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
309 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
310 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
311 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
312 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
313 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
319 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
320 struct ixgbe_ring *tx_ring)
324 int q = tx_ring - adapter->tx_ring;
326 if (tx_ring->cpu != cpu) {
327 txctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q));
328 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
329 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
330 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
331 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q), txctrl);
337 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
341 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
344 for (i = 0; i < adapter->num_tx_queues; i++) {
345 adapter->tx_ring[i].cpu = -1;
346 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
348 for (i = 0; i < adapter->num_rx_queues; i++) {
349 adapter->rx_ring[i].cpu = -1;
350 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
354 static int __ixgbe_notify_dca(struct device *dev, void *data)
356 struct net_device *netdev = dev_get_drvdata(dev);
357 struct ixgbe_adapter *adapter = netdev_priv(netdev);
358 unsigned long event = *(unsigned long *)data;
361 case DCA_PROVIDER_ADD:
362 /* if we're already enabled, don't do it again */
363 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
365 /* Always use CB2 mode, difference is masked
366 * in the CB driver. */
367 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
368 if (dca_add_requester(dev) == 0) {
369 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
370 ixgbe_setup_dca(adapter);
373 /* Fall Through since DCA is disabled. */
374 case DCA_PROVIDER_REMOVE:
375 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
376 dca_remove_requester(dev);
377 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
378 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
386 #endif /* CONFIG_IXGBE_DCA */
388 * ixgbe_receive_skb - Send a completed packet up the stack
389 * @adapter: board private structure
390 * @skb: packet to send up
391 * @status: hardware indication of status of receive
392 * @rx_ring: rx descriptor ring (for a specific queue) to setup
393 * @rx_desc: rx descriptor
395 static void ixgbe_receive_skb(struct ixgbe_adapter *adapter,
396 struct sk_buff *skb, u8 status,
397 struct ixgbe_ring *ring,
398 union ixgbe_adv_rx_desc *rx_desc)
400 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
401 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
403 if (adapter->netdev->features & NETIF_F_LRO &&
404 skb->ip_summed == CHECKSUM_UNNECESSARY) {
405 if (adapter->vlgrp && is_vlan)
406 lro_vlan_hwaccel_receive_skb(&ring->lro_mgr, skb,
410 lro_receive_skb(&ring->lro_mgr, skb, rx_desc);
411 ring->lro_used = true;
413 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
414 if (adapter->vlgrp && is_vlan)
415 vlan_hwaccel_receive_skb(skb, adapter->vlgrp, tag);
417 netif_receive_skb(skb);
419 if (adapter->vlgrp && is_vlan)
420 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
428 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
429 * @adapter: address of board private structure
430 * @status_err: hardware indication of status of receive
431 * @skb: skb currently being received and modified
433 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
434 u32 status_err, struct sk_buff *skb)
436 skb->ip_summed = CHECKSUM_NONE;
438 /* Rx csum disabled */
439 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
442 /* if IP and error */
443 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
444 (status_err & IXGBE_RXDADV_ERR_IPE)) {
445 adapter->hw_csum_rx_error++;
449 if (!(status_err & IXGBE_RXD_STAT_L4CS))
452 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
453 adapter->hw_csum_rx_error++;
457 /* It must be a TCP or UDP packet with a valid checksum */
458 skb->ip_summed = CHECKSUM_UNNECESSARY;
459 adapter->hw_csum_rx_good++;
463 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
464 * @adapter: address of board private structure
466 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
467 struct ixgbe_ring *rx_ring,
470 struct pci_dev *pdev = adapter->pdev;
471 union ixgbe_adv_rx_desc *rx_desc;
472 struct ixgbe_rx_buffer *bi;
474 unsigned int bufsz = rx_ring->rx_buf_len + NET_IP_ALIGN;
476 i = rx_ring->next_to_use;
477 bi = &rx_ring->rx_buffer_info[i];
479 while (cleaned_count--) {
480 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
483 (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
485 bi->page = alloc_page(GFP_ATOMIC);
487 adapter->alloc_rx_page_failed++;
492 /* use a half page if we're re-using */
493 bi->page_offset ^= (PAGE_SIZE / 2);
496 bi->page_dma = pci_map_page(pdev, bi->page,
503 struct sk_buff *skb = netdev_alloc_skb(adapter->netdev,
507 adapter->alloc_rx_buff_failed++;
512 * Make buffer alignment 2 beyond a 16 byte boundary
513 * this will result in a 16 byte aligned IP header after
514 * the 14 byte MAC header is removed
516 skb_reserve(skb, NET_IP_ALIGN);
519 bi->dma = pci_map_single(pdev, skb->data, bufsz,
522 /* Refresh the desc even if buffer_addrs didn't change because
523 * each write-back erases this info. */
524 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
525 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
526 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
528 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
532 if (i == rx_ring->count)
534 bi = &rx_ring->rx_buffer_info[i];
538 if (rx_ring->next_to_use != i) {
539 rx_ring->next_to_use = i;
541 i = (rx_ring->count - 1);
544 * Force memory writes to complete before letting h/w
545 * know there are new descriptors to fetch. (Only
546 * applicable for weak-ordered memory model archs,
550 writel(i, adapter->hw.hw_addr + rx_ring->tail);
554 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
556 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
559 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
561 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
564 static bool ixgbe_clean_rx_irq(struct ixgbe_adapter *adapter,
565 struct ixgbe_ring *rx_ring,
566 int *work_done, int work_to_do)
568 struct pci_dev *pdev = adapter->pdev;
569 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
570 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
575 bool cleaned = false;
576 int cleaned_count = 0;
577 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
579 i = rx_ring->next_to_clean;
580 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
581 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
582 rx_buffer_info = &rx_ring->rx_buffer_info[i];
584 while (staterr & IXGBE_RXD_STAT_DD) {
586 if (*work_done >= work_to_do)
590 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
591 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
592 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
593 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
594 if (hdr_info & IXGBE_RXDADV_SPH)
595 adapter->rx_hdr_split++;
596 if (len > IXGBE_RX_HDR_SIZE)
597 len = IXGBE_RX_HDR_SIZE;
598 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
600 len = le16_to_cpu(rx_desc->wb.upper.length);
604 skb = rx_buffer_info->skb;
605 prefetch(skb->data - NET_IP_ALIGN);
606 rx_buffer_info->skb = NULL;
608 if (len && !skb_shinfo(skb)->nr_frags) {
609 pci_unmap_single(pdev, rx_buffer_info->dma,
610 rx_ring->rx_buf_len + NET_IP_ALIGN,
616 pci_unmap_page(pdev, rx_buffer_info->page_dma,
617 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
618 rx_buffer_info->page_dma = 0;
619 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
620 rx_buffer_info->page,
621 rx_buffer_info->page_offset,
624 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
625 (page_count(rx_buffer_info->page) != 1))
626 rx_buffer_info->page = NULL;
628 get_page(rx_buffer_info->page);
630 skb->len += upper_len;
631 skb->data_len += upper_len;
632 skb->truesize += upper_len;
636 if (i == rx_ring->count)
638 next_buffer = &rx_ring->rx_buffer_info[i];
640 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
644 if (staterr & IXGBE_RXD_STAT_EOP) {
645 rx_ring->stats.packets++;
646 rx_ring->stats.bytes += skb->len;
648 rx_buffer_info->skb = next_buffer->skb;
649 rx_buffer_info->dma = next_buffer->dma;
650 next_buffer->skb = skb;
651 next_buffer->dma = 0;
652 adapter->non_eop_descs++;
656 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
657 dev_kfree_skb_irq(skb);
661 ixgbe_rx_checksum(adapter, staterr, skb);
663 /* probably a little skewed due to removing CRC */
664 total_rx_bytes += skb->len;
667 skb->protocol = eth_type_trans(skb, adapter->netdev);
668 ixgbe_receive_skb(adapter, skb, staterr, rx_ring, rx_desc);
669 adapter->netdev->last_rx = jiffies;
672 rx_desc->wb.upper.status_error = 0;
674 /* return some buffers to hardware, one at a time is too slow */
675 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
676 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
680 /* use prefetched values */
682 rx_buffer_info = next_buffer;
684 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
687 if (rx_ring->lro_used) {
688 lro_flush_all(&rx_ring->lro_mgr);
689 rx_ring->lro_used = false;
692 rx_ring->next_to_clean = i;
693 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
696 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
698 rx_ring->total_packets += total_rx_packets;
699 rx_ring->total_bytes += total_rx_bytes;
700 adapter->net_stats.rx_bytes += total_rx_bytes;
701 adapter->net_stats.rx_packets += total_rx_packets;
706 static int ixgbe_clean_rxonly(struct napi_struct *, int);
708 * ixgbe_configure_msix - Configure MSI-X hardware
709 * @adapter: board private structure
711 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
714 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
716 struct ixgbe_q_vector *q_vector;
717 int i, j, q_vectors, v_idx, r_idx;
720 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
722 /* Populate the IVAR table and set the ITR values to the
723 * corresponding register.
725 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
726 q_vector = &adapter->q_vector[v_idx];
727 /* XXX for_each_bit(...) */
728 r_idx = find_first_bit(q_vector->rxr_idx,
729 adapter->num_rx_queues);
731 for (i = 0; i < q_vector->rxr_count; i++) {
732 j = adapter->rx_ring[r_idx].reg_idx;
733 ixgbe_set_ivar(adapter, IXGBE_IVAR_RX_QUEUE(j), v_idx);
734 r_idx = find_next_bit(q_vector->rxr_idx,
735 adapter->num_rx_queues,
738 r_idx = find_first_bit(q_vector->txr_idx,
739 adapter->num_tx_queues);
741 for (i = 0; i < q_vector->txr_count; i++) {
742 j = adapter->tx_ring[r_idx].reg_idx;
743 ixgbe_set_ivar(adapter, IXGBE_IVAR_TX_QUEUE(j), v_idx);
744 r_idx = find_next_bit(q_vector->txr_idx,
745 adapter->num_tx_queues,
749 /* if this is a tx only vector halve the interrupt rate */
750 if (q_vector->txr_count && !q_vector->rxr_count)
751 q_vector->eitr = (adapter->eitr_param >> 1);
754 q_vector->eitr = adapter->eitr_param;
756 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx),
757 EITR_INTS_PER_SEC_TO_REG(q_vector->eitr));
760 ixgbe_set_ivar(adapter, IXGBE_IVAR_OTHER_CAUSES_INDEX, v_idx);
761 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
763 /* set up to autoclear timer, and the vectors */
764 mask = IXGBE_EIMS_ENABLE_MASK;
765 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
766 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
773 latency_invalid = 255
777 * ixgbe_update_itr - update the dynamic ITR value based on statistics
778 * @adapter: pointer to adapter
779 * @eitr: eitr setting (ints per sec) to give last timeslice
780 * @itr_setting: current throttle rate in ints/second
781 * @packets: the number of packets during this measurement interval
782 * @bytes: the number of bytes during this measurement interval
784 * Stores a new ITR value based on packets and byte
785 * counts during the last interrupt. The advantage of per interrupt
786 * computation is faster updates and more accurate ITR for the current
787 * traffic pattern. Constants in this function were computed
788 * based on theoretical maximum wire speed and thresholds were set based
789 * on testing data as well as attempting to minimize response time
790 * while increasing bulk throughput.
791 * this functionality is controlled by the InterruptThrottleRate module
792 * parameter (see ixgbe_param.c)
794 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
795 u32 eitr, u8 itr_setting,
796 int packets, int bytes)
798 unsigned int retval = itr_setting;
803 goto update_itr_done;
806 /* simple throttlerate management
807 * 0-20MB/s lowest (100000 ints/s)
808 * 20-100MB/s low (20000 ints/s)
809 * 100-1249MB/s bulk (8000 ints/s)
811 /* what was last interrupt timeslice? */
812 timepassed_us = 1000000/eitr;
813 bytes_perint = bytes / timepassed_us; /* bytes/usec */
815 switch (itr_setting) {
817 if (bytes_perint > adapter->eitr_low)
818 retval = low_latency;
821 if (bytes_perint > adapter->eitr_high)
822 retval = bulk_latency;
823 else if (bytes_perint <= adapter->eitr_low)
824 retval = lowest_latency;
827 if (bytes_perint <= adapter->eitr_high)
828 retval = low_latency;
836 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
838 struct ixgbe_adapter *adapter = q_vector->adapter;
839 struct ixgbe_hw *hw = &adapter->hw;
841 u8 current_itr, ret_itr;
842 int i, r_idx, v_idx = ((void *)q_vector - (void *)(adapter->q_vector)) /
843 sizeof(struct ixgbe_q_vector);
844 struct ixgbe_ring *rx_ring, *tx_ring;
846 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
847 for (i = 0; i < q_vector->txr_count; i++) {
848 tx_ring = &(adapter->tx_ring[r_idx]);
849 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
851 tx_ring->total_packets,
852 tx_ring->total_bytes);
853 /* if the result for this queue would decrease interrupt
854 * rate for this vector then use that result */
855 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
856 q_vector->tx_itr - 1 : ret_itr);
857 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
861 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
862 for (i = 0; i < q_vector->rxr_count; i++) {
863 rx_ring = &(adapter->rx_ring[r_idx]);
864 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
866 rx_ring->total_packets,
867 rx_ring->total_bytes);
868 /* if the result for this queue would decrease interrupt
869 * rate for this vector then use that result */
870 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
871 q_vector->rx_itr - 1 : ret_itr);
872 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
876 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
878 switch (current_itr) {
879 /* counts and packets in update_itr are dependent on these numbers */
884 new_itr = 20000; /* aka hwitr = ~200 */
892 if (new_itr != q_vector->eitr) {
894 /* do an exponential smoothing */
895 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
896 q_vector->eitr = new_itr;
897 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
898 /* must write high and low 16 bits to reset counter */
899 DPRINTK(TX_ERR, DEBUG, "writing eitr(%d): %08X\n", v_idx,
901 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg | (itr_reg)<<16);
908 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
910 struct ixgbe_hw *hw = &adapter->hw;
913 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
914 adapter->link_check_timeout = jiffies;
915 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
916 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
917 schedule_work(&adapter->watchdog_task);
921 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
923 struct net_device *netdev = data;
924 struct ixgbe_adapter *adapter = netdev_priv(netdev);
925 struct ixgbe_hw *hw = &adapter->hw;
926 u32 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
928 if (eicr & IXGBE_EICR_LSC)
929 ixgbe_check_lsc(adapter);
931 if (!test_bit(__IXGBE_DOWN, &adapter->state))
932 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
937 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
939 struct ixgbe_q_vector *q_vector = data;
940 struct ixgbe_adapter *adapter = q_vector->adapter;
941 struct ixgbe_ring *tx_ring;
944 if (!q_vector->txr_count)
947 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
948 for (i = 0; i < q_vector->txr_count; i++) {
949 tx_ring = &(adapter->tx_ring[r_idx]);
950 #ifdef CONFIG_IXGBE_DCA
951 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
952 ixgbe_update_tx_dca(adapter, tx_ring);
954 tx_ring->total_bytes = 0;
955 tx_ring->total_packets = 0;
956 ixgbe_clean_tx_irq(adapter, tx_ring);
957 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
965 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
967 * @data: pointer to our q_vector struct for this interrupt vector
969 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
971 struct ixgbe_q_vector *q_vector = data;
972 struct ixgbe_adapter *adapter = q_vector->adapter;
973 struct ixgbe_ring *rx_ring;
977 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
978 for (i = 0; i < q_vector->rxr_count; i++) {
979 rx_ring = &(adapter->rx_ring[r_idx]);
980 rx_ring->total_bytes = 0;
981 rx_ring->total_packets = 0;
982 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
986 if (!q_vector->rxr_count)
989 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
990 rx_ring = &(adapter->rx_ring[r_idx]);
991 /* disable interrupts on this vector only */
992 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, rx_ring->v_idx);
993 netif_rx_schedule(adapter->netdev, &q_vector->napi);
998 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1000 ixgbe_msix_clean_rx(irq, data);
1001 ixgbe_msix_clean_tx(irq, data);
1007 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1008 * @napi: napi struct with our devices info in it
1009 * @budget: amount of work driver is allowed to do this pass, in packets
1011 * This function is optimized for cleaning one queue only on a single
1014 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1016 struct ixgbe_q_vector *q_vector =
1017 container_of(napi, struct ixgbe_q_vector, napi);
1018 struct ixgbe_adapter *adapter = q_vector->adapter;
1019 struct ixgbe_ring *rx_ring = NULL;
1023 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1024 rx_ring = &(adapter->rx_ring[r_idx]);
1025 #ifdef CONFIG_IXGBE_DCA
1026 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1027 ixgbe_update_rx_dca(adapter, rx_ring);
1030 ixgbe_clean_rx_irq(adapter, rx_ring, &work_done, budget);
1032 /* If all Rx work done, exit the polling mode */
1033 if (work_done < budget) {
1034 netif_rx_complete(adapter->netdev, napi);
1035 if (adapter->itr_setting & 3)
1036 ixgbe_set_itr_msix(q_vector);
1037 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1038 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, rx_ring->v_idx);
1045 * ixgbe_clean_rxonly_many - msix (aka one shot) rx clean routine
1046 * @napi: napi struct with our devices info in it
1047 * @budget: amount of work driver is allowed to do this pass, in packets
1049 * This function will clean more than one rx queue associated with a
1052 static int ixgbe_clean_rxonly_many(struct napi_struct *napi, int budget)
1054 struct ixgbe_q_vector *q_vector =
1055 container_of(napi, struct ixgbe_q_vector, napi);
1056 struct ixgbe_adapter *adapter = q_vector->adapter;
1057 struct ixgbe_ring *rx_ring = NULL;
1058 int work_done = 0, i;
1060 u16 enable_mask = 0;
1062 /* attempt to distribute budget to each queue fairly, but don't allow
1063 * the budget to go below 1 because we'll exit polling */
1064 budget /= (q_vector->rxr_count ?: 1);
1065 budget = max(budget, 1);
1066 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1067 for (i = 0; i < q_vector->rxr_count; i++) {
1068 rx_ring = &(adapter->rx_ring[r_idx]);
1069 #ifdef CONFIG_IXGBE_DCA
1070 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1071 ixgbe_update_rx_dca(adapter, rx_ring);
1073 ixgbe_clean_rx_irq(adapter, rx_ring, &work_done, budget);
1074 enable_mask |= rx_ring->v_idx;
1075 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1079 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1080 rx_ring = &(adapter->rx_ring[r_idx]);
1081 /* If all Rx work done, exit the polling mode */
1082 if (work_done < budget) {
1083 netif_rx_complete(adapter->netdev, napi);
1084 if (adapter->itr_setting & 3)
1085 ixgbe_set_itr_msix(q_vector);
1086 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1087 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, enable_mask);
1093 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1096 a->q_vector[v_idx].adapter = a;
1097 set_bit(r_idx, a->q_vector[v_idx].rxr_idx);
1098 a->q_vector[v_idx].rxr_count++;
1099 a->rx_ring[r_idx].v_idx = 1 << v_idx;
1102 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1105 a->q_vector[v_idx].adapter = a;
1106 set_bit(r_idx, a->q_vector[v_idx].txr_idx);
1107 a->q_vector[v_idx].txr_count++;
1108 a->tx_ring[r_idx].v_idx = 1 << v_idx;
1112 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1113 * @adapter: board private structure to initialize
1114 * @vectors: allotted vector count for descriptor rings
1116 * This function maps descriptor rings to the queue-specific vectors
1117 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1118 * one vector per ring/queue, but on a constrained vector budget, we
1119 * group the rings as "efficiently" as possible. You would add new
1120 * mapping configurations in here.
1122 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1126 int rxr_idx = 0, txr_idx = 0;
1127 int rxr_remaining = adapter->num_rx_queues;
1128 int txr_remaining = adapter->num_tx_queues;
1133 /* No mapping required if MSI-X is disabled. */
1134 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1138 * The ideal configuration...
1139 * We have enough vectors to map one per queue.
1141 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1142 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1143 map_vector_to_rxq(adapter, v_start, rxr_idx);
1145 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1146 map_vector_to_txq(adapter, v_start, txr_idx);
1152 * If we don't have enough vectors for a 1-to-1
1153 * mapping, we'll have to group them so there are
1154 * multiple queues per vector.
1156 /* Re-adjusting *qpv takes care of the remainder. */
1157 for (i = v_start; i < vectors; i++) {
1158 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1159 for (j = 0; j < rqpv; j++) {
1160 map_vector_to_rxq(adapter, i, rxr_idx);
1165 for (i = v_start; i < vectors; i++) {
1166 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1167 for (j = 0; j < tqpv; j++) {
1168 map_vector_to_txq(adapter, i, txr_idx);
1179 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1180 * @adapter: board private structure
1182 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1183 * interrupts from the kernel.
1185 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1187 struct net_device *netdev = adapter->netdev;
1188 irqreturn_t (*handler)(int, void *);
1189 int i, vector, q_vectors, err;
1191 /* Decrement for Other and TCP Timer vectors */
1192 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1194 /* Map the Tx/Rx rings to the vectors we were allotted. */
1195 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1199 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1200 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1201 &ixgbe_msix_clean_many)
1202 for (vector = 0; vector < q_vectors; vector++) {
1203 handler = SET_HANDLER(&adapter->q_vector[vector]);
1204 sprintf(adapter->name[vector], "%s:v%d-%s",
1205 netdev->name, vector,
1206 (handler == &ixgbe_msix_clean_rx) ? "Rx" :
1207 ((handler == &ixgbe_msix_clean_tx) ? "Tx" : "TxRx"));
1208 err = request_irq(adapter->msix_entries[vector].vector,
1209 handler, 0, adapter->name[vector],
1210 &(adapter->q_vector[vector]));
1213 "request_irq failed for MSIX interrupt "
1214 "Error: %d\n", err);
1215 goto free_queue_irqs;
1219 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1220 err = request_irq(adapter->msix_entries[vector].vector,
1221 &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1224 "request_irq for msix_lsc failed: %d\n", err);
1225 goto free_queue_irqs;
1231 for (i = vector - 1; i >= 0; i--)
1232 free_irq(adapter->msix_entries[--vector].vector,
1233 &(adapter->q_vector[i]));
1234 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1235 pci_disable_msix(adapter->pdev);
1236 kfree(adapter->msix_entries);
1237 adapter->msix_entries = NULL;
1242 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1244 struct ixgbe_hw *hw = &adapter->hw;
1245 struct ixgbe_q_vector *q_vector = adapter->q_vector;
1247 u32 new_itr = q_vector->eitr;
1248 struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1249 struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1251 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1253 tx_ring->total_packets,
1254 tx_ring->total_bytes);
1255 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1257 rx_ring->total_packets,
1258 rx_ring->total_bytes);
1260 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1262 switch (current_itr) {
1263 /* counts and packets in update_itr are dependent on these numbers */
1264 case lowest_latency:
1268 new_itr = 20000; /* aka hwitr = ~200 */
1277 if (new_itr != q_vector->eitr) {
1279 /* do an exponential smoothing */
1280 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1281 q_vector->eitr = new_itr;
1282 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
1283 /* must write high and low 16 bits to reset counter */
1284 IXGBE_WRITE_REG(hw, IXGBE_EITR(0), itr_reg | (itr_reg)<<16);
1291 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1292 * @adapter: board private structure
1294 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1296 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1297 IXGBE_WRITE_FLUSH(&adapter->hw);
1298 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1300 for (i = 0; i < adapter->num_msix_vectors; i++)
1301 synchronize_irq(adapter->msix_entries[i].vector);
1303 synchronize_irq(adapter->pdev->irq);
1308 * ixgbe_irq_enable - Enable default interrupt generation settings
1309 * @adapter: board private structure
1311 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1314 mask = IXGBE_EIMS_ENABLE_MASK;
1315 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1316 IXGBE_WRITE_FLUSH(&adapter->hw);
1320 * ixgbe_intr - legacy mode Interrupt Handler
1321 * @irq: interrupt number
1322 * @data: pointer to a network interface device structure
1323 * @pt_regs: CPU registers structure
1325 static irqreturn_t ixgbe_intr(int irq, void *data)
1327 struct net_device *netdev = data;
1328 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1329 struct ixgbe_hw *hw = &adapter->hw;
1332 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1333 * therefore no explict interrupt disable is necessary */
1334 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1336 /* shared interrupt alert!
1337 * make sure interrupts are enabled because the read will
1338 * have disabled interrupts due to EIAM */
1339 ixgbe_irq_enable(adapter);
1340 return IRQ_NONE; /* Not our interrupt */
1343 if (eicr & IXGBE_EICR_LSC)
1344 ixgbe_check_lsc(adapter);
1346 if (netif_rx_schedule_prep(netdev, &adapter->q_vector[0].napi)) {
1347 adapter->tx_ring[0].total_packets = 0;
1348 adapter->tx_ring[0].total_bytes = 0;
1349 adapter->rx_ring[0].total_packets = 0;
1350 adapter->rx_ring[0].total_bytes = 0;
1351 /* would disable interrupts here but EIAM disabled it */
1352 __netif_rx_schedule(netdev, &adapter->q_vector[0].napi);
1358 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1360 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1362 for (i = 0; i < q_vectors; i++) {
1363 struct ixgbe_q_vector *q_vector = &adapter->q_vector[i];
1364 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1365 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1366 q_vector->rxr_count = 0;
1367 q_vector->txr_count = 0;
1372 * ixgbe_request_irq - initialize interrupts
1373 * @adapter: board private structure
1375 * Attempts to configure interrupts using the best available
1376 * capabilities of the hardware and kernel.
1378 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1380 struct net_device *netdev = adapter->netdev;
1383 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1384 err = ixgbe_request_msix_irqs(adapter);
1385 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1386 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1387 netdev->name, netdev);
1389 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1390 netdev->name, netdev);
1394 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1399 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1401 struct net_device *netdev = adapter->netdev;
1403 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1406 q_vectors = adapter->num_msix_vectors;
1409 free_irq(adapter->msix_entries[i].vector, netdev);
1412 for (; i >= 0; i--) {
1413 free_irq(adapter->msix_entries[i].vector,
1414 &(adapter->q_vector[i]));
1417 ixgbe_reset_q_vectors(adapter);
1419 free_irq(adapter->pdev->irq, netdev);
1424 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1427 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1429 struct ixgbe_hw *hw = &adapter->hw;
1431 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1432 EITR_INTS_PER_SEC_TO_REG(adapter->eitr_param));
1434 ixgbe_set_ivar(adapter, IXGBE_IVAR_RX_QUEUE(0), 0);
1435 ixgbe_set_ivar(adapter, IXGBE_IVAR_TX_QUEUE(0), 0);
1437 map_vector_to_rxq(adapter, 0, 0);
1438 map_vector_to_txq(adapter, 0, 0);
1440 DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1444 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1445 * @adapter: board private structure
1447 * Configure the Tx unit of the MAC after a reset.
1449 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1452 struct ixgbe_hw *hw = &adapter->hw;
1453 u32 i, j, tdlen, txctrl;
1455 /* Setup the HW Tx Head and Tail descriptor pointers */
1456 for (i = 0; i < adapter->num_tx_queues; i++) {
1457 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1460 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1461 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1462 (tdba & DMA_32BIT_MASK));
1463 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1465 (ring->count * sizeof(union ixgbe_adv_tx_desc));
1466 tdwba |= IXGBE_TDWBAL_HEAD_WB_ENABLE;
1467 IXGBE_WRITE_REG(hw, IXGBE_TDWBAL(j), tdwba & DMA_32BIT_MASK);
1468 IXGBE_WRITE_REG(hw, IXGBE_TDWBAH(j), (tdwba >> 32));
1469 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1470 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1471 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1472 adapter->tx_ring[i].head = IXGBE_TDH(j);
1473 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1474 /* Disable Tx Head Writeback RO bit, since this hoses
1475 * bookkeeping if things aren't delivered in order.
1477 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1478 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1479 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1483 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1485 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter, int index)
1487 struct ixgbe_ring *rx_ring;
1492 /* program one srrctl register per VMDq index */
1493 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
1495 mask = (unsigned long) adapter->ring_feature[RING_F_RSS].mask;
1496 len = sizeof(adapter->ring_feature[RING_F_VMDQ].mask) * 8;
1497 shift = find_first_bit(&mask, len);
1498 queue0 = index & mask;
1499 index = (index & mask) >> shift;
1500 /* program one srrctl per RSS queue since RDRXCTL.MVMEN is enabled */
1502 mask = (unsigned long) adapter->ring_feature[RING_F_RSS].mask;
1503 queue0 = index & mask;
1504 index = index & mask;
1507 rx_ring = &adapter->rx_ring[queue0];
1509 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
1511 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1512 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1514 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1515 srrctl |= IXGBE_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1516 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1517 srrctl |= ((IXGBE_RX_HDR_SIZE <<
1518 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1519 IXGBE_SRRCTL_BSIZEHDR_MASK);
1521 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1523 if (rx_ring->rx_buf_len == MAXIMUM_ETHERNET_VLAN_SIZE)
1524 srrctl |= IXGBE_RXBUFFER_2048 >>
1525 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1527 srrctl |= rx_ring->rx_buf_len >>
1528 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1530 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
1534 * ixgbe_get_skb_hdr - helper function for LRO header processing
1535 * @skb: pointer to sk_buff to be added to LRO packet
1536 * @iphdr: pointer to ip header structure
1537 * @tcph: pointer to tcp header structure
1538 * @hdr_flags: pointer to header flags
1539 * @priv: private data
1541 static int ixgbe_get_skb_hdr(struct sk_buff *skb, void **iphdr, void **tcph,
1542 u64 *hdr_flags, void *priv)
1544 union ixgbe_adv_rx_desc *rx_desc = priv;
1546 /* Verify that this is a valid IPv4 TCP packet */
1547 if (!((ixgbe_get_pkt_info(rx_desc) & IXGBE_RXDADV_PKTTYPE_IPV4) &&
1548 (ixgbe_get_pkt_info(rx_desc) & IXGBE_RXDADV_PKTTYPE_TCP)))
1551 /* Set network headers */
1552 skb_reset_network_header(skb);
1553 skb_set_transport_header(skb, ip_hdrlen(skb));
1554 *iphdr = ip_hdr(skb);
1555 *tcph = tcp_hdr(skb);
1556 *hdr_flags = LRO_IPV4 | LRO_TCP;
1560 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1561 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1564 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
1565 * @adapter: board private structure
1567 * Configure the Rx unit of the MAC after a reset.
1569 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
1572 struct ixgbe_hw *hw = &adapter->hw;
1573 struct net_device *netdev = adapter->netdev;
1574 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1576 u32 rdlen, rxctrl, rxcsum;
1577 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
1578 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
1579 0x6A3E67EA, 0x14364D17, 0x3BED200D};
1586 /* Decide whether to use packet split mode or not */
1587 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
1589 /* Set the RX buffer length according to the mode */
1590 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1591 rx_buf_len = IXGBE_RX_HDR_SIZE;
1593 if (netdev->mtu <= ETH_DATA_LEN)
1594 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1596 rx_buf_len = ALIGN(max_frame, 1024);
1599 fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
1600 fctrl |= IXGBE_FCTRL_BAM;
1601 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
1602 IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
1604 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
1605 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1606 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
1608 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
1609 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
1611 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1613 rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1614 /* disable receives while setting up the descriptors */
1615 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
1616 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
1618 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1619 * the Base and Length of the Rx Descriptor Ring */
1620 for (i = 0; i < adapter->num_rx_queues; i++) {
1621 rdba = adapter->rx_ring[i].dma;
1622 j = adapter->rx_ring[i].reg_idx;
1623 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_32BIT_MASK));
1624 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
1625 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
1626 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
1627 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
1628 adapter->rx_ring[i].head = IXGBE_RDH(j);
1629 adapter->rx_ring[i].tail = IXGBE_RDT(j);
1630 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1631 /* Intitial LRO Settings */
1632 adapter->rx_ring[i].lro_mgr.max_aggr = IXGBE_MAX_LRO_AGGREGATE;
1633 adapter->rx_ring[i].lro_mgr.max_desc = IXGBE_MAX_LRO_DESCRIPTORS;
1634 adapter->rx_ring[i].lro_mgr.get_skb_header = ixgbe_get_skb_hdr;
1635 adapter->rx_ring[i].lro_mgr.features = LRO_F_EXTRACT_VLAN_ID;
1636 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1637 adapter->rx_ring[i].lro_mgr.features |= LRO_F_NAPI;
1638 adapter->rx_ring[i].lro_mgr.dev = adapter->netdev;
1639 adapter->rx_ring[i].lro_mgr.ip_summed = CHECKSUM_UNNECESSARY;
1640 adapter->rx_ring[i].lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1642 ixgbe_configure_srrctl(adapter, j);
1646 * For VMDq support of different descriptor types or
1647 * buffer sizes through the use of multiple SRRCTL
1648 * registers, RDRXCTL.MVMEN must be set to 1
1650 * also, the manual doesn't mention it clearly but DCA hints
1651 * will only use queue 0's tags unless this bit is set. Side
1652 * effects of setting this bit are only that SRRCTL must be
1653 * fully programmed [0..15]
1655 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
1656 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
1657 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
1660 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
1661 /* Fill out redirection table */
1662 for (i = 0, j = 0; i < 128; i++, j++) {
1663 if (j == adapter->ring_feature[RING_F_RSS].indices)
1665 /* reta = 4-byte sliding window of
1666 * 0x00..(indices-1)(indices-1)00..etc. */
1667 reta = (reta << 8) | (j * 0x11);
1669 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
1672 /* Fill out hash function seeds */
1673 for (i = 0; i < 10; i++)
1674 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
1676 mrqc = IXGBE_MRQC_RSSEN
1677 /* Perform hash on these packet types */
1678 | IXGBE_MRQC_RSS_FIELD_IPV4
1679 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
1680 | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
1681 | IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP
1682 | IXGBE_MRQC_RSS_FIELD_IPV6_EX
1683 | IXGBE_MRQC_RSS_FIELD_IPV6
1684 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
1685 | IXGBE_MRQC_RSS_FIELD_IPV6_UDP
1686 | IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
1687 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
1690 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
1692 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
1693 adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
1694 /* Disable indicating checksum in descriptor, enables
1696 rxcsum |= IXGBE_RXCSUM_PCSD;
1698 if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
1699 /* Enable IPv4 payload checksum for UDP fragments
1700 * if PCSD is not set */
1701 rxcsum |= IXGBE_RXCSUM_IPPCSE;
1704 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
1707 static void ixgbe_vlan_rx_register(struct net_device *netdev,
1708 struct vlan_group *grp)
1710 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1713 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1714 ixgbe_irq_disable(adapter);
1715 adapter->vlgrp = grp;
1718 /* enable VLAN tag insert/strip */
1719 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
1720 ctrl |= IXGBE_VLNCTRL_VME;
1721 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
1722 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
1725 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1726 ixgbe_irq_enable(adapter);
1729 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1731 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1732 struct ixgbe_hw *hw = &adapter->hw;
1734 /* add VID to filter table */
1735 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
1738 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1740 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1741 struct ixgbe_hw *hw = &adapter->hw;
1743 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1744 ixgbe_irq_disable(adapter);
1746 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1748 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1749 ixgbe_irq_enable(adapter);
1751 /* remove VID from filter table */
1752 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
1755 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
1757 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1759 if (adapter->vlgrp) {
1761 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1762 if (!vlan_group_get_device(adapter->vlgrp, vid))
1764 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
1769 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
1771 struct dev_mc_list *mc_ptr;
1772 u8 *addr = *mc_addr_ptr;
1775 mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
1777 *mc_addr_ptr = mc_ptr->next->dmi_addr;
1779 *mc_addr_ptr = NULL;
1785 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
1786 * @netdev: network interface device structure
1788 * The set_rx_method entry point is called whenever the unicast/multicast
1789 * address list or the network interface flags are updated. This routine is
1790 * responsible for configuring the hardware for proper unicast, multicast and
1793 static void ixgbe_set_rx_mode(struct net_device *netdev)
1795 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1796 struct ixgbe_hw *hw = &adapter->hw;
1798 u8 *addr_list = NULL;
1801 /* Check for Promiscuous and All Multicast modes */
1803 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
1804 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
1806 if (netdev->flags & IFF_PROMISC) {
1807 hw->addr_ctrl.user_set_promisc = 1;
1808 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
1809 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
1811 if (netdev->flags & IFF_ALLMULTI) {
1812 fctrl |= IXGBE_FCTRL_MPE;
1813 fctrl &= ~IXGBE_FCTRL_UPE;
1815 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
1817 vlnctrl |= IXGBE_VLNCTRL_VFE;
1818 hw->addr_ctrl.user_set_promisc = 0;
1821 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
1822 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
1824 /* reprogram secondary unicast list */
1825 addr_count = netdev->uc_count;
1827 addr_list = netdev->uc_list->dmi_addr;
1828 hw->mac.ops.update_uc_addr_list(hw, addr_list, addr_count,
1829 ixgbe_addr_list_itr);
1831 /* reprogram multicast list */
1832 addr_count = netdev->mc_count;
1834 addr_list = netdev->mc_list->dmi_addr;
1835 hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
1836 ixgbe_addr_list_itr);
1839 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
1842 struct ixgbe_q_vector *q_vector;
1843 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1845 /* legacy and MSI only use one vector */
1846 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1849 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1850 struct napi_struct *napi;
1851 q_vector = &adapter->q_vector[q_idx];
1852 if (!q_vector->rxr_count)
1854 napi = &q_vector->napi;
1855 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) &&
1856 (q_vector->rxr_count > 1))
1857 napi->poll = &ixgbe_clean_rxonly_many;
1863 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
1866 struct ixgbe_q_vector *q_vector;
1867 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1869 /* legacy and MSI only use one vector */
1870 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1873 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1874 q_vector = &adapter->q_vector[q_idx];
1875 if (!q_vector->rxr_count)
1877 napi_disable(&q_vector->napi);
1881 static void ixgbe_configure(struct ixgbe_adapter *adapter)
1883 struct net_device *netdev = adapter->netdev;
1886 ixgbe_set_rx_mode(netdev);
1888 ixgbe_restore_vlan(adapter);
1890 ixgbe_configure_tx(adapter);
1891 ixgbe_configure_rx(adapter);
1892 for (i = 0; i < adapter->num_rx_queues; i++)
1893 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
1894 (adapter->rx_ring[i].count - 1));
1897 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
1899 struct net_device *netdev = adapter->netdev;
1900 struct ixgbe_hw *hw = &adapter->hw;
1902 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1903 u32 txdctl, rxdctl, mhadd;
1906 ixgbe_get_hw_control(adapter);
1908 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
1909 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
1910 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1911 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
1912 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
1917 /* XXX: to interrupt immediately for EICS writes, enable this */
1918 /* gpie |= IXGBE_GPIE_EIMEN; */
1919 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
1922 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
1923 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
1924 * specifically only auto mask tx and rx interrupts */
1925 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
1928 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
1929 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
1930 mhadd &= ~IXGBE_MHADD_MFS_MASK;
1931 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
1933 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
1936 for (i = 0; i < adapter->num_tx_queues; i++) {
1937 j = adapter->tx_ring[i].reg_idx;
1938 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
1939 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
1940 txdctl |= (8 << 16);
1941 txdctl |= IXGBE_TXDCTL_ENABLE;
1942 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
1945 for (i = 0; i < adapter->num_rx_queues; i++) {
1946 j = adapter->rx_ring[i].reg_idx;
1947 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
1948 /* enable PTHRESH=32 descriptors (half the internal cache)
1949 * and HTHRESH=0 descriptors (to minimize latency on fetch),
1950 * this also removes a pesky rx_no_buffer_count increment */
1952 rxdctl |= IXGBE_RXDCTL_ENABLE;
1953 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
1955 /* enable all receives */
1956 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
1957 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
1958 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxdctl);
1960 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
1961 ixgbe_configure_msix(adapter);
1963 ixgbe_configure_msi_and_legacy(adapter);
1965 clear_bit(__IXGBE_DOWN, &adapter->state);
1966 ixgbe_napi_enable_all(adapter);
1968 /* clear any pending interrupts, may auto mask */
1969 IXGBE_READ_REG(hw, IXGBE_EICR);
1971 ixgbe_irq_enable(adapter);
1973 /* bring the link up in the watchdog, this could race with our first
1974 * link up interrupt but shouldn't be a problem */
1975 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1976 adapter->link_check_timeout = jiffies;
1977 mod_timer(&adapter->watchdog_timer, jiffies);
1981 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
1983 WARN_ON(in_interrupt());
1984 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
1986 ixgbe_down(adapter);
1988 clear_bit(__IXGBE_RESETTING, &adapter->state);
1991 int ixgbe_up(struct ixgbe_adapter *adapter)
1993 /* hardware has been reset, we need to reload some things */
1994 ixgbe_configure(adapter);
1996 return ixgbe_up_complete(adapter);
1999 void ixgbe_reset(struct ixgbe_adapter *adapter)
2001 struct ixgbe_hw *hw = &adapter->hw;
2002 if (hw->mac.ops.init_hw(hw))
2003 dev_err(&adapter->pdev->dev, "Hardware Error\n");
2005 /* reprogram the RAR[0] in case user changed it. */
2006 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2011 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2012 * @adapter: board private structure
2013 * @rx_ring: ring to free buffers from
2015 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2016 struct ixgbe_ring *rx_ring)
2018 struct pci_dev *pdev = adapter->pdev;
2022 /* Free all the Rx ring sk_buffs */
2024 for (i = 0; i < rx_ring->count; i++) {
2025 struct ixgbe_rx_buffer *rx_buffer_info;
2027 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2028 if (rx_buffer_info->dma) {
2029 pci_unmap_single(pdev, rx_buffer_info->dma,
2030 rx_ring->rx_buf_len,
2031 PCI_DMA_FROMDEVICE);
2032 rx_buffer_info->dma = 0;
2034 if (rx_buffer_info->skb) {
2035 dev_kfree_skb(rx_buffer_info->skb);
2036 rx_buffer_info->skb = NULL;
2038 if (!rx_buffer_info->page)
2040 pci_unmap_page(pdev, rx_buffer_info->page_dma, PAGE_SIZE / 2,
2041 PCI_DMA_FROMDEVICE);
2042 rx_buffer_info->page_dma = 0;
2043 put_page(rx_buffer_info->page);
2044 rx_buffer_info->page = NULL;
2045 rx_buffer_info->page_offset = 0;
2048 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2049 memset(rx_ring->rx_buffer_info, 0, size);
2051 /* Zero out the descriptor ring */
2052 memset(rx_ring->desc, 0, rx_ring->size);
2054 rx_ring->next_to_clean = 0;
2055 rx_ring->next_to_use = 0;
2057 writel(0, adapter->hw.hw_addr + rx_ring->head);
2058 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2062 * ixgbe_clean_tx_ring - Free Tx Buffers
2063 * @adapter: board private structure
2064 * @tx_ring: ring to be cleaned
2066 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2067 struct ixgbe_ring *tx_ring)
2069 struct ixgbe_tx_buffer *tx_buffer_info;
2073 /* Free all the Tx ring sk_buffs */
2075 for (i = 0; i < tx_ring->count; i++) {
2076 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2077 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2080 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2081 memset(tx_ring->tx_buffer_info, 0, size);
2083 /* Zero out the descriptor ring */
2084 memset(tx_ring->desc, 0, tx_ring->size);
2086 tx_ring->next_to_use = 0;
2087 tx_ring->next_to_clean = 0;
2089 writel(0, adapter->hw.hw_addr + tx_ring->head);
2090 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2094 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2095 * @adapter: board private structure
2097 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2101 for (i = 0; i < adapter->num_rx_queues; i++)
2102 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2106 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2107 * @adapter: board private structure
2109 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2113 for (i = 0; i < adapter->num_tx_queues; i++)
2114 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2117 void ixgbe_down(struct ixgbe_adapter *adapter)
2119 struct net_device *netdev = adapter->netdev;
2120 struct ixgbe_hw *hw = &adapter->hw;
2125 /* signal that we are down to the interrupt handler */
2126 set_bit(__IXGBE_DOWN, &adapter->state);
2128 /* disable receives */
2129 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2130 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2132 netif_tx_disable(netdev);
2134 IXGBE_WRITE_FLUSH(hw);
2137 netif_tx_stop_all_queues(netdev);
2139 ixgbe_irq_disable(adapter);
2141 ixgbe_napi_disable_all(adapter);
2143 del_timer_sync(&adapter->watchdog_timer);
2144 cancel_work_sync(&adapter->watchdog_task);
2146 /* disable transmits in the hardware now that interrupts are off */
2147 for (i = 0; i < adapter->num_tx_queues; i++) {
2148 j = adapter->tx_ring[i].reg_idx;
2149 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2150 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
2151 (txdctl & ~IXGBE_TXDCTL_ENABLE));
2154 netif_carrier_off(netdev);
2156 #ifdef CONFIG_IXGBE_DCA
2157 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2158 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
2159 dca_remove_requester(&adapter->pdev->dev);
2163 if (!pci_channel_offline(adapter->pdev))
2164 ixgbe_reset(adapter);
2165 ixgbe_clean_all_tx_rings(adapter);
2166 ixgbe_clean_all_rx_rings(adapter);
2168 #ifdef CONFIG_IXGBE_DCA
2169 /* since we reset the hardware DCA settings were cleared */
2170 if (dca_add_requester(&adapter->pdev->dev) == 0) {
2171 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
2172 /* always use CB2 mode, difference is masked
2173 * in the CB driver */
2174 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
2175 ixgbe_setup_dca(adapter);
2181 * ixgbe_poll - NAPI Rx polling callback
2182 * @napi: structure for representing this polling device
2183 * @budget: how many packets driver is allowed to clean
2185 * This function is used for legacy and MSI, NAPI mode
2187 static int ixgbe_poll(struct napi_struct *napi, int budget)
2189 struct ixgbe_q_vector *q_vector = container_of(napi,
2190 struct ixgbe_q_vector, napi);
2191 struct ixgbe_adapter *adapter = q_vector->adapter;
2192 int tx_cleaned, work_done = 0;
2194 #ifdef CONFIG_IXGBE_DCA
2195 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2196 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
2197 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
2201 tx_cleaned = ixgbe_clean_tx_irq(adapter, adapter->tx_ring);
2202 ixgbe_clean_rx_irq(adapter, adapter->rx_ring, &work_done, budget);
2207 /* If budget not fully consumed, exit the polling mode */
2208 if (work_done < budget) {
2209 netif_rx_complete(adapter->netdev, napi);
2210 if (adapter->itr_setting & 3)
2211 ixgbe_set_itr(adapter);
2212 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2213 ixgbe_irq_enable(adapter);
2219 * ixgbe_tx_timeout - Respond to a Tx Hang
2220 * @netdev: network interface device structure
2222 static void ixgbe_tx_timeout(struct net_device *netdev)
2224 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2226 /* Do the reset outside of interrupt context */
2227 schedule_work(&adapter->reset_task);
2230 static void ixgbe_reset_task(struct work_struct *work)
2232 struct ixgbe_adapter *adapter;
2233 adapter = container_of(work, struct ixgbe_adapter, reset_task);
2235 adapter->tx_timeout_count++;
2237 ixgbe_reinit_locked(adapter);
2240 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
2242 int nrq = 1, ntq = 1;
2243 int feature_mask = 0, rss_i, rss_m;
2245 /* Number of supported queues */
2246 switch (adapter->hw.mac.type) {
2247 case ixgbe_mac_82598EB:
2248 rss_i = adapter->ring_feature[RING_F_RSS].indices;
2250 feature_mask |= IXGBE_FLAG_RSS_ENABLED;
2252 switch (adapter->flags & feature_mask) {
2253 case (IXGBE_FLAG_RSS_ENABLED):
2267 adapter->ring_feature[RING_F_RSS].indices = rss_i;
2268 adapter->ring_feature[RING_F_RSS].mask = rss_m;
2276 adapter->num_rx_queues = nrq;
2277 adapter->num_tx_queues = ntq;
2280 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
2283 int err, vector_threshold;
2285 /* We'll want at least 3 (vector_threshold):
2288 * 3) Other (Link Status Change, etc.)
2289 * 4) TCP Timer (optional)
2291 vector_threshold = MIN_MSIX_COUNT;
2293 /* The more we get, the more we will assign to Tx/Rx Cleanup
2294 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2295 * Right now, we simply care about how many we'll get; we'll
2296 * set them up later while requesting irq's.
2298 while (vectors >= vector_threshold) {
2299 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
2301 if (!err) /* Success in acquiring all requested vectors. */
2304 vectors = 0; /* Nasty failure, quit now */
2305 else /* err == number of vectors we should try again with */
2309 if (vectors < vector_threshold) {
2310 /* Can't allocate enough MSI-X interrupts? Oh well.
2311 * This just means we'll go with either a single MSI
2312 * vector or fall back to legacy interrupts.
2314 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
2315 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2316 kfree(adapter->msix_entries);
2317 adapter->msix_entries = NULL;
2318 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2319 ixgbe_set_num_queues(adapter);
2321 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
2322 adapter->num_msix_vectors = vectors;
2327 * ixgbe_cache_ring_register - Descriptor ring to register mapping
2328 * @adapter: board private structure to initialize
2330 * Once we know the feature-set enabled for the device, we'll cache
2331 * the register offset the descriptor ring is assigned to.
2333 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
2335 int feature_mask = 0, rss_i;
2336 int i, txr_idx, rxr_idx;
2338 /* Number of supported queues */
2339 switch (adapter->hw.mac.type) {
2340 case ixgbe_mac_82598EB:
2341 rss_i = adapter->ring_feature[RING_F_RSS].indices;
2344 feature_mask |= IXGBE_FLAG_RSS_ENABLED;
2345 switch (adapter->flags & feature_mask) {
2346 case (IXGBE_FLAG_RSS_ENABLED):
2347 for (i = 0; i < adapter->num_rx_queues; i++)
2348 adapter->rx_ring[i].reg_idx = i;
2349 for (i = 0; i < adapter->num_tx_queues; i++)
2350 adapter->tx_ring[i].reg_idx = i;
2363 * ixgbe_alloc_queues - Allocate memory for all rings
2364 * @adapter: board private structure to initialize
2366 * We allocate one ring per queue at run-time since we don't know the
2367 * number of queues at compile-time. The polling_netdev array is
2368 * intended for Multiqueue, but should work fine with a single queue.
2370 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
2374 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
2375 sizeof(struct ixgbe_ring), GFP_KERNEL);
2376 if (!adapter->tx_ring)
2377 goto err_tx_ring_allocation;
2379 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
2380 sizeof(struct ixgbe_ring), GFP_KERNEL);
2381 if (!adapter->rx_ring)
2382 goto err_rx_ring_allocation;
2384 for (i = 0; i < adapter->num_tx_queues; i++) {
2385 adapter->tx_ring[i].count = adapter->tx_ring_count;
2386 adapter->tx_ring[i].queue_index = i;
2389 for (i = 0; i < adapter->num_rx_queues; i++) {
2390 adapter->rx_ring[i].count = adapter->rx_ring_count;
2391 adapter->rx_ring[i].queue_index = i;
2394 ixgbe_cache_ring_register(adapter);
2398 err_rx_ring_allocation:
2399 kfree(adapter->tx_ring);
2400 err_tx_ring_allocation:
2405 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
2406 * @adapter: board private structure to initialize
2408 * Attempt to configure the interrupts using the best available
2409 * capabilities of the hardware and the kernel.
2411 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
2414 int vector, v_budget;
2417 * It's easy to be greedy for MSI-X vectors, but it really
2418 * doesn't do us much good if we have a lot more vectors
2419 * than CPU's. So let's be conservative and only ask for
2420 * (roughly) twice the number of vectors as there are CPU's.
2422 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
2423 (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
2426 * At the same time, hardware can only support a maximum of
2427 * MAX_MSIX_COUNT vectors. With features such as RSS and VMDq,
2428 * we can easily reach upwards of 64 Rx descriptor queues and
2429 * 32 Tx queues. Thus, we cap it off in those rare cases where
2430 * the cpu count also exceeds our vector limit.
2432 v_budget = min(v_budget, MAX_MSIX_COUNT);
2434 /* A failure in MSI-X entry allocation isn't fatal, but it does
2435 * mean we disable MSI-X capabilities of the adapter. */
2436 adapter->msix_entries = kcalloc(v_budget,
2437 sizeof(struct msix_entry), GFP_KERNEL);
2438 if (!adapter->msix_entries) {
2439 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2440 ixgbe_set_num_queues(adapter);
2441 kfree(adapter->tx_ring);
2442 kfree(adapter->rx_ring);
2443 err = ixgbe_alloc_queues(adapter);
2445 DPRINTK(PROBE, ERR, "Unable to allocate memory "
2453 for (vector = 0; vector < v_budget; vector++)
2454 adapter->msix_entries[vector].entry = vector;
2456 ixgbe_acquire_msix_vectors(adapter, v_budget);
2458 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2462 err = pci_enable_msi(adapter->pdev);
2464 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
2466 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
2467 "falling back to legacy. Error: %d\n", err);
2473 /* Notify the stack of the (possibly) reduced Tx Queue count. */
2474 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
2479 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
2481 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2482 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2483 pci_disable_msix(adapter->pdev);
2484 kfree(adapter->msix_entries);
2485 adapter->msix_entries = NULL;
2486 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2487 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
2488 pci_disable_msi(adapter->pdev);
2494 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
2495 * @adapter: board private structure to initialize
2497 * We determine which interrupt scheme to use based on...
2498 * - Kernel support (MSI, MSI-X)
2499 * - which can be user-defined (via MODULE_PARAM)
2500 * - Hardware queue count (num_*_queues)
2501 * - defined by miscellaneous hardware support/features (RSS, etc.)
2503 static int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
2507 /* Number of supported queues */
2508 ixgbe_set_num_queues(adapter);
2510 err = ixgbe_alloc_queues(adapter);
2512 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
2513 goto err_alloc_queues;
2516 err = ixgbe_set_interrupt_capability(adapter);
2518 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
2519 goto err_set_interrupt;
2522 DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
2523 "Tx Queue count = %u\n",
2524 (adapter->num_rx_queues > 1) ? "Enabled" :
2525 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2527 set_bit(__IXGBE_DOWN, &adapter->state);
2532 kfree(adapter->tx_ring);
2533 kfree(adapter->rx_ring);
2539 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
2540 * @adapter: board private structure to initialize
2542 * ixgbe_sw_init initializes the Adapter private data structure.
2543 * Fields are initialized based on PCI device information and
2544 * OS network device settings (MTU size).
2546 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
2548 struct ixgbe_hw *hw = &adapter->hw;
2549 struct pci_dev *pdev = adapter->pdev;
2552 /* PCI config space info */
2554 hw->vendor_id = pdev->vendor;
2555 hw->device_id = pdev->device;
2556 hw->revision_id = pdev->revision;
2557 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2558 hw->subsystem_device_id = pdev->subsystem_device;
2560 /* Set capability flags */
2561 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
2562 adapter->ring_feature[RING_F_RSS].indices = rss;
2563 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
2565 /* default flow control settings */
2566 hw->fc.original_type = ixgbe_fc_none;
2567 hw->fc.type = ixgbe_fc_none;
2568 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
2569 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
2570 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
2571 hw->fc.send_xon = true;
2573 /* select 10G link by default */
2574 hw->mac.link_mode_select = IXGBE_AUTOC_LMS_10G_LINK_NO_AN;
2576 /* enable itr by default in dynamic mode */
2577 adapter->itr_setting = 1;
2578 adapter->eitr_param = 20000;
2580 /* set defaults for eitr in MegaBytes */
2581 adapter->eitr_low = 10;
2582 adapter->eitr_high = 20;
2584 /* set default ring sizes */
2585 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
2586 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
2588 /* initialize eeprom parameters */
2589 if (ixgbe_init_eeprom_params_generic(hw)) {
2590 dev_err(&pdev->dev, "EEPROM initialization failed\n");
2594 /* enable rx csum by default */
2595 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
2597 set_bit(__IXGBE_DOWN, &adapter->state);
2603 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
2604 * @adapter: board private structure
2605 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2607 * Return 0 on success, negative on failure
2609 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
2610 struct ixgbe_ring *tx_ring)
2612 struct pci_dev *pdev = adapter->pdev;
2615 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2616 tx_ring->tx_buffer_info = vmalloc(size);
2617 if (!tx_ring->tx_buffer_info)
2619 memset(tx_ring->tx_buffer_info, 0, size);
2621 /* round up to nearest 4K */
2622 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc) +
2624 tx_ring->size = ALIGN(tx_ring->size, 4096);
2626 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
2631 tx_ring->next_to_use = 0;
2632 tx_ring->next_to_clean = 0;
2633 tx_ring->work_limit = tx_ring->count;
2637 vfree(tx_ring->tx_buffer_info);
2638 tx_ring->tx_buffer_info = NULL;
2639 DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
2640 "descriptor ring\n");
2645 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
2646 * @adapter: board private structure
2648 * If this function returns with an error, then it's possible one or
2649 * more of the rings is populated (while the rest are not). It is the
2650 * callers duty to clean those orphaned rings.
2652 * Return 0 on success, negative on failure
2654 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
2658 for (i = 0; i < adapter->num_tx_queues; i++) {
2659 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2662 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
2670 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
2671 * @adapter: board private structure
2672 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2674 * Returns 0 on success, negative on failure
2676 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
2677 struct ixgbe_ring *rx_ring)
2679 struct pci_dev *pdev = adapter->pdev;
2682 size = sizeof(struct net_lro_desc) * IXGBE_MAX_LRO_DESCRIPTORS;
2683 rx_ring->lro_mgr.lro_arr = vmalloc(size);
2684 if (!rx_ring->lro_mgr.lro_arr)
2686 memset(rx_ring->lro_mgr.lro_arr, 0, size);
2688 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2689 rx_ring->rx_buffer_info = vmalloc(size);
2690 if (!rx_ring->rx_buffer_info) {
2692 "vmalloc allocation failed for the rx desc ring\n");
2695 memset(rx_ring->rx_buffer_info, 0, size);
2697 /* Round up to nearest 4K */
2698 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2699 rx_ring->size = ALIGN(rx_ring->size, 4096);
2701 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
2703 if (!rx_ring->desc) {
2705 "Memory allocation failed for the rx desc ring\n");
2706 vfree(rx_ring->rx_buffer_info);
2710 rx_ring->next_to_clean = 0;
2711 rx_ring->next_to_use = 0;
2716 vfree(rx_ring->lro_mgr.lro_arr);
2717 rx_ring->lro_mgr.lro_arr = NULL;
2722 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
2723 * @adapter: board private structure
2725 * If this function returns with an error, then it's possible one or
2726 * more of the rings is populated (while the rest are not). It is the
2727 * callers duty to clean those orphaned rings.
2729 * Return 0 on success, negative on failure
2732 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
2736 for (i = 0; i < adapter->num_rx_queues; i++) {
2737 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2740 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
2748 * ixgbe_free_tx_resources - Free Tx Resources per Queue
2749 * @adapter: board private structure
2750 * @tx_ring: Tx descriptor ring for a specific queue
2752 * Free all transmit software resources
2754 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
2755 struct ixgbe_ring *tx_ring)
2757 struct pci_dev *pdev = adapter->pdev;
2759 ixgbe_clean_tx_ring(adapter, tx_ring);
2761 vfree(tx_ring->tx_buffer_info);
2762 tx_ring->tx_buffer_info = NULL;
2764 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2766 tx_ring->desc = NULL;
2770 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
2771 * @adapter: board private structure
2773 * Free all transmit software resources
2775 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
2779 for (i = 0; i < adapter->num_tx_queues; i++)
2780 ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
2784 * ixgbe_free_rx_resources - Free Rx Resources
2785 * @adapter: board private structure
2786 * @rx_ring: ring to clean the resources from
2788 * Free all receive software resources
2790 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
2791 struct ixgbe_ring *rx_ring)
2793 struct pci_dev *pdev = adapter->pdev;
2795 vfree(rx_ring->lro_mgr.lro_arr);
2796 rx_ring->lro_mgr.lro_arr = NULL;
2798 ixgbe_clean_rx_ring(adapter, rx_ring);
2800 vfree(rx_ring->rx_buffer_info);
2801 rx_ring->rx_buffer_info = NULL;
2803 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2805 rx_ring->desc = NULL;
2809 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
2810 * @adapter: board private structure
2812 * Free all receive software resources
2814 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
2818 for (i = 0; i < adapter->num_rx_queues; i++)
2819 ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
2823 * ixgbe_change_mtu - Change the Maximum Transfer Unit
2824 * @netdev: network interface device structure
2825 * @new_mtu: new value for maximum frame size
2827 * Returns 0 on success, negative on failure
2829 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
2831 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2832 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2834 /* MTU < 68 is an error and causes problems on some kernels */
2835 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
2838 DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
2839 netdev->mtu, new_mtu);
2840 /* must set new MTU before calling down or up */
2841 netdev->mtu = new_mtu;
2843 if (netif_running(netdev))
2844 ixgbe_reinit_locked(adapter);
2850 * ixgbe_open - Called when a network interface is made active
2851 * @netdev: network interface device structure
2853 * Returns 0 on success, negative value on failure
2855 * The open entry point is called when a network interface is made
2856 * active by the system (IFF_UP). At this point all resources needed
2857 * for transmit and receive operations are allocated, the interrupt
2858 * handler is registered with the OS, the watchdog timer is started,
2859 * and the stack is notified that the interface is ready.
2861 static int ixgbe_open(struct net_device *netdev)
2863 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2866 /* disallow open during test */
2867 if (test_bit(__IXGBE_TESTING, &adapter->state))
2870 /* allocate transmit descriptors */
2871 err = ixgbe_setup_all_tx_resources(adapter);
2875 /* allocate receive descriptors */
2876 err = ixgbe_setup_all_rx_resources(adapter);
2880 ixgbe_configure(adapter);
2882 err = ixgbe_request_irq(adapter);
2886 err = ixgbe_up_complete(adapter);
2890 netif_tx_start_all_queues(netdev);
2895 ixgbe_release_hw_control(adapter);
2896 ixgbe_free_irq(adapter);
2898 ixgbe_free_all_rx_resources(adapter);
2900 ixgbe_free_all_tx_resources(adapter);
2902 ixgbe_reset(adapter);
2908 * ixgbe_close - Disables a network interface
2909 * @netdev: network interface device structure
2911 * Returns 0, this is not allowed to fail
2913 * The close entry point is called when an interface is de-activated
2914 * by the OS. The hardware is still under the drivers control, but
2915 * needs to be disabled. A global MAC reset is issued to stop the
2916 * hardware, and all transmit and receive resources are freed.
2918 static int ixgbe_close(struct net_device *netdev)
2920 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2922 ixgbe_down(adapter);
2923 ixgbe_free_irq(adapter);
2925 ixgbe_free_all_tx_resources(adapter);
2926 ixgbe_free_all_rx_resources(adapter);
2928 ixgbe_release_hw_control(adapter);
2934 * ixgbe_napi_add_all - prep napi structs for use
2935 * @adapter: private struct
2936 * helper function to napi_add each possible q_vector->napi
2938 static void ixgbe_napi_add_all(struct ixgbe_adapter *adapter)
2940 int q_idx, q_vectors;
2941 int (*poll)(struct napi_struct *, int);
2943 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2944 poll = &ixgbe_clean_rxonly;
2945 /* Only enable as many vectors as we have rx queues. */
2946 q_vectors = adapter->num_rx_queues;
2949 /* only one q_vector for legacy modes */
2953 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2954 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
2955 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
2959 static void ixgbe_napi_del_all(struct ixgbe_adapter *adapter)
2962 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2964 /* legacy and MSI only use one vector */
2965 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2968 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2969 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
2970 if (!q_vector->rxr_count)
2972 netif_napi_del(&q_vector->napi);
2977 static int ixgbe_resume(struct pci_dev *pdev)
2979 struct net_device *netdev = pci_get_drvdata(pdev);
2980 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2983 pci_set_power_state(pdev, PCI_D0);
2984 pci_restore_state(pdev);
2985 err = pci_enable_device(pdev);
2987 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
2991 pci_set_master(pdev);
2993 pci_enable_wake(pdev, PCI_D3hot, 0);
2994 pci_enable_wake(pdev, PCI_D3cold, 0);
2996 err = ixgbe_init_interrupt_scheme(adapter);
2998 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
3003 ixgbe_napi_add_all(adapter);
3004 ixgbe_reset(adapter);
3006 if (netif_running(netdev)) {
3007 err = ixgbe_open(adapter->netdev);
3012 netif_device_attach(netdev);
3017 #endif /* CONFIG_PM */
3018 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
3020 struct net_device *netdev = pci_get_drvdata(pdev);
3021 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3026 netif_device_detach(netdev);
3028 if (netif_running(netdev)) {
3029 ixgbe_down(adapter);
3030 ixgbe_free_irq(adapter);
3031 ixgbe_free_all_tx_resources(adapter);
3032 ixgbe_free_all_rx_resources(adapter);
3034 ixgbe_reset_interrupt_capability(adapter);
3035 ixgbe_napi_del_all(adapter);
3036 kfree(adapter->tx_ring);
3037 kfree(adapter->rx_ring);
3040 retval = pci_save_state(pdev);
3045 pci_enable_wake(pdev, PCI_D3hot, 0);
3046 pci_enable_wake(pdev, PCI_D3cold, 0);
3048 ixgbe_release_hw_control(adapter);
3050 pci_disable_device(pdev);
3052 pci_set_power_state(pdev, pci_choose_state(pdev, state));
3057 static void ixgbe_shutdown(struct pci_dev *pdev)
3059 ixgbe_suspend(pdev, PMSG_SUSPEND);
3063 * ixgbe_update_stats - Update the board statistics counters.
3064 * @adapter: board private structure
3066 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
3068 struct ixgbe_hw *hw = &adapter->hw;
3070 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
3072 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
3073 for (i = 0; i < 8; i++) {
3074 /* for packet buffers not used, the register should read 0 */
3075 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
3077 adapter->stats.mpc[i] += mpc;
3078 total_mpc += adapter->stats.mpc[i];
3079 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
3081 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
3082 /* work around hardware counting issue */
3083 adapter->stats.gprc -= missed_rx;
3085 /* 82598 hardware only has a 32 bit counter in the high register */
3086 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
3087 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
3088 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
3089 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
3090 adapter->stats.bprc += bprc;
3091 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
3092 adapter->stats.mprc -= bprc;
3093 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
3094 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
3095 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
3096 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
3097 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
3098 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
3099 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
3100 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
3101 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
3102 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
3103 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
3104 adapter->stats.lxontxc += lxon;
3105 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
3106 adapter->stats.lxofftxc += lxoff;
3107 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3108 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
3109 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
3111 * 82598 errata - tx of flow control packets is included in tx counters
3113 xon_off_tot = lxon + lxoff;
3114 adapter->stats.gptc -= xon_off_tot;
3115 adapter->stats.mptc -= xon_off_tot;
3116 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
3117 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3118 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
3119 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
3120 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
3121 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
3122 adapter->stats.ptc64 -= xon_off_tot;
3123 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
3124 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
3125 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
3126 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
3127 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
3128 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
3130 /* Fill out the OS statistics structure */
3131 adapter->net_stats.multicast = adapter->stats.mprc;
3134 adapter->net_stats.rx_errors = adapter->stats.crcerrs +
3135 adapter->stats.rlec;
3136 adapter->net_stats.rx_dropped = 0;
3137 adapter->net_stats.rx_length_errors = adapter->stats.rlec;
3138 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3139 adapter->net_stats.rx_missed_errors = total_mpc;
3143 * ixgbe_watchdog - Timer Call-back
3144 * @data: pointer to adapter cast into an unsigned long
3146 static void ixgbe_watchdog(unsigned long data)
3148 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3149 struct ixgbe_hw *hw = &adapter->hw;
3151 /* Do the watchdog outside of interrupt context due to the lovely
3152 * delays that some of the newer hardware requires */
3153 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
3154 /* Cause software interrupt to ensure rx rings are cleaned */
3155 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3157 (1 << (adapter->num_msix_vectors - NON_Q_VECTORS)) - 1;
3158 IXGBE_WRITE_REG(hw, IXGBE_EICS, eics);
3160 /* For legacy and MSI interrupts don't set any bits that
3161 * are enabled for EIAM, because this operation would
3162 * set *both* EIMS and EICS for any bit in EIAM */
3163 IXGBE_WRITE_REG(hw, IXGBE_EICS,
3164 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
3166 /* Reset the timer */
3167 mod_timer(&adapter->watchdog_timer,
3168 round_jiffies(jiffies + 2 * HZ));
3171 schedule_work(&adapter->watchdog_task);
3175 * ixgbe_watchdog_task - worker thread to bring link up
3176 * @work: pointer to work_struct containing our data
3178 static void ixgbe_watchdog_task(struct work_struct *work)
3180 struct ixgbe_adapter *adapter = container_of(work,
3181 struct ixgbe_adapter,
3183 struct net_device *netdev = adapter->netdev;
3184 struct ixgbe_hw *hw = &adapter->hw;
3185 u32 link_speed = adapter->link_speed;
3186 bool link_up = adapter->link_up;
3188 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
3190 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
3191 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3193 time_after(jiffies, (adapter->link_check_timeout +
3194 IXGBE_TRY_LINK_TIMEOUT))) {
3195 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
3196 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
3198 adapter->link_up = link_up;
3199 adapter->link_speed = link_speed;
3203 if (!netif_carrier_ok(netdev)) {
3204 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3205 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
3206 #define FLOW_RX (frctl & IXGBE_FCTRL_RFCE)
3207 #define FLOW_TX (rmcs & IXGBE_RMCS_TFCE_802_3X)
3208 DPRINTK(LINK, INFO, "NIC Link is Up %s, "
3209 "Flow Control: %s\n",
3210 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
3212 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
3213 "1 Gbps" : "unknown speed")),
3214 ((FLOW_RX && FLOW_TX) ? "RX/TX" :
3216 (FLOW_TX ? "TX" : "None"))));
3218 netif_carrier_on(netdev);
3219 netif_tx_wake_all_queues(netdev);
3221 /* Force detection of hung controller */
3222 adapter->detect_tx_hung = true;
3225 adapter->link_up = false;
3226 adapter->link_speed = 0;
3227 if (netif_carrier_ok(netdev)) {
3228 DPRINTK(LINK, INFO, "NIC Link is Down\n");
3229 netif_carrier_off(netdev);
3230 netif_tx_stop_all_queues(netdev);
3234 ixgbe_update_stats(adapter);
3235 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
3238 static int ixgbe_tso(struct ixgbe_adapter *adapter,
3239 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
3240 u32 tx_flags, u8 *hdr_len)
3242 struct ixgbe_adv_tx_context_desc *context_desc;
3245 struct ixgbe_tx_buffer *tx_buffer_info;
3246 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
3247 u32 mss_l4len_idx, l4len;
3249 if (skb_is_gso(skb)) {
3250 if (skb_header_cloned(skb)) {
3251 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3255 l4len = tcp_hdrlen(skb);
3258 if (skb->protocol == htons(ETH_P_IP)) {
3259 struct iphdr *iph = ip_hdr(skb);
3262 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3266 adapter->hw_tso_ctxt++;
3267 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3268 ipv6_hdr(skb)->payload_len = 0;
3269 tcp_hdr(skb)->check =
3270 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3271 &ipv6_hdr(skb)->daddr,
3273 adapter->hw_tso6_ctxt++;
3276 i = tx_ring->next_to_use;
3278 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3279 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
3281 /* VLAN MACLEN IPLEN */
3282 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3284 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
3285 vlan_macip_lens |= ((skb_network_offset(skb)) <<
3286 IXGBE_ADVTXD_MACLEN_SHIFT);
3287 *hdr_len += skb_network_offset(skb);
3289 (skb_transport_header(skb) - skb_network_header(skb));
3291 (skb_transport_header(skb) - skb_network_header(skb));
3292 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3293 context_desc->seqnum_seed = 0;
3295 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3296 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
3297 IXGBE_ADVTXD_DTYP_CTXT);
3299 if (skb->protocol == htons(ETH_P_IP))
3300 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
3301 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3302 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
3306 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
3307 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
3308 /* use index 1 for TSO */
3309 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3310 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3312 tx_buffer_info->time_stamp = jiffies;
3313 tx_buffer_info->next_to_watch = i;
3316 if (i == tx_ring->count)
3318 tx_ring->next_to_use = i;
3325 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
3326 struct ixgbe_ring *tx_ring,
3327 struct sk_buff *skb, u32 tx_flags)
3329 struct ixgbe_adv_tx_context_desc *context_desc;
3331 struct ixgbe_tx_buffer *tx_buffer_info;
3332 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
3334 if (skb->ip_summed == CHECKSUM_PARTIAL ||
3335 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
3336 i = tx_ring->next_to_use;
3337 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3338 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
3340 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3342 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
3343 vlan_macip_lens |= (skb_network_offset(skb) <<
3344 IXGBE_ADVTXD_MACLEN_SHIFT);
3345 if (skb->ip_summed == CHECKSUM_PARTIAL)
3346 vlan_macip_lens |= (skb_transport_header(skb) -
3347 skb_network_header(skb));
3349 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3350 context_desc->seqnum_seed = 0;
3352 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
3353 IXGBE_ADVTXD_DTYP_CTXT);
3355 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3356 switch (skb->protocol) {
3357 case __constant_htons(ETH_P_IP):
3358 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
3359 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3361 IXGBE_ADVTXD_TUCMD_L4T_TCP;
3363 case __constant_htons(ETH_P_IPV6):
3364 /* XXX what about other V6 headers?? */
3365 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3367 IXGBE_ADVTXD_TUCMD_L4T_TCP;
3370 if (unlikely(net_ratelimit())) {
3371 DPRINTK(PROBE, WARNING,
3372 "partial checksum but proto=%x!\n",
3379 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
3380 /* use index zero for tx checksum offload */
3381 context_desc->mss_l4len_idx = 0;
3383 tx_buffer_info->time_stamp = jiffies;
3384 tx_buffer_info->next_to_watch = i;
3386 adapter->hw_csum_tx_good++;
3388 if (i == tx_ring->count)
3390 tx_ring->next_to_use = i;
3398 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
3399 struct ixgbe_ring *tx_ring,
3400 struct sk_buff *skb, unsigned int first)
3402 struct ixgbe_tx_buffer *tx_buffer_info;
3403 unsigned int len = skb->len;
3404 unsigned int offset = 0, size, count = 0, i;
3405 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
3408 len -= skb->data_len;
3410 i = tx_ring->next_to_use;
3413 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3414 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
3416 tx_buffer_info->length = size;
3417 tx_buffer_info->dma = pci_map_single(adapter->pdev,
3419 size, PCI_DMA_TODEVICE);
3420 tx_buffer_info->time_stamp = jiffies;
3421 tx_buffer_info->next_to_watch = i;
3427 if (i == tx_ring->count)
3431 for (f = 0; f < nr_frags; f++) {
3432 struct skb_frag_struct *frag;
3434 frag = &skb_shinfo(skb)->frags[f];
3436 offset = frag->page_offset;
3439 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3440 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
3442 tx_buffer_info->length = size;
3443 tx_buffer_info->dma = pci_map_page(adapter->pdev,
3448 tx_buffer_info->time_stamp = jiffies;
3449 tx_buffer_info->next_to_watch = i;
3455 if (i == tx_ring->count)
3460 i = tx_ring->count - 1;
3463 tx_ring->tx_buffer_info[i].skb = skb;
3464 tx_ring->tx_buffer_info[first].next_to_watch = i;
3469 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
3470 struct ixgbe_ring *tx_ring,
3471 int tx_flags, int count, u32 paylen, u8 hdr_len)
3473 union ixgbe_adv_tx_desc *tx_desc = NULL;
3474 struct ixgbe_tx_buffer *tx_buffer_info;
3475 u32 olinfo_status = 0, cmd_type_len = 0;
3477 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
3479 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
3481 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
3483 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3484 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
3486 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
3487 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
3489 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
3490 IXGBE_ADVTXD_POPTS_SHIFT;
3492 /* use index 1 context for tso */
3493 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3494 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3495 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
3496 IXGBE_ADVTXD_POPTS_SHIFT;
3498 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3499 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
3500 IXGBE_ADVTXD_POPTS_SHIFT;
3502 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
3504 i = tx_ring->next_to_use;
3506 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3507 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
3508 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
3509 tx_desc->read.cmd_type_len =
3510 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
3511 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3513 if (i == tx_ring->count)
3517 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
3520 * Force memory writes to complete before letting h/w
3521 * know there are new descriptors to fetch. (Only
3522 * applicable for weak-ordered memory model archs,
3527 tx_ring->next_to_use = i;
3528 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3531 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
3532 struct ixgbe_ring *tx_ring, int size)
3534 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3536 netif_stop_subqueue(netdev, tx_ring->queue_index);
3537 /* Herbert's original patch had:
3538 * smp_mb__after_netif_stop_queue();
3539 * but since that doesn't exist yet, just open code it. */
3542 /* We need to check again in a case another CPU has just
3543 * made room available. */
3544 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
3547 /* A reprieve! - use start_queue because it doesn't call schedule */
3548 netif_start_subqueue(netdev, tx_ring->queue_index);
3549 ++adapter->restart_queue;
3553 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
3554 struct ixgbe_ring *tx_ring, int size)
3556 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
3558 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
3561 static int ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3563 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3564 struct ixgbe_ring *tx_ring;
3566 unsigned int tx_flags = 0;
3572 r_idx = (adapter->num_tx_queues - 1) & skb->queue_mapping;
3573 tx_ring = &adapter->tx_ring[r_idx];
3575 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3576 tx_flags |= vlan_tx_tag_get(skb);
3577 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3578 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3580 /* three things can cause us to need a context descriptor */
3581 if (skb_is_gso(skb) ||
3582 (skb->ip_summed == CHECKSUM_PARTIAL) ||
3583 (tx_flags & IXGBE_TX_FLAGS_VLAN))
3586 count += TXD_USE_COUNT(skb_headlen(skb));
3587 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3588 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3590 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
3592 return NETDEV_TX_BUSY;
3595 if (skb->protocol == htons(ETH_P_IP))
3596 tx_flags |= IXGBE_TX_FLAGS_IPV4;
3597 first = tx_ring->next_to_use;
3598 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
3600 dev_kfree_skb_any(skb);
3601 return NETDEV_TX_OK;
3605 tx_flags |= IXGBE_TX_FLAGS_TSO;
3606 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
3607 (skb->ip_summed == CHECKSUM_PARTIAL))
3608 tx_flags |= IXGBE_TX_FLAGS_CSUM;
3610 ixgbe_tx_queue(adapter, tx_ring, tx_flags,
3611 ixgbe_tx_map(adapter, tx_ring, skb, first),
3614 netdev->trans_start = jiffies;
3616 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
3618 return NETDEV_TX_OK;
3622 * ixgbe_get_stats - Get System Network Statistics
3623 * @netdev: network interface device structure
3625 * Returns the address of the device statistics structure.
3626 * The statistics are actually updated from the timer callback.
3628 static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
3630 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3632 /* only return the current stats */
3633 return &adapter->net_stats;
3637 * ixgbe_set_mac - Change the Ethernet Address of the NIC
3638 * @netdev: network interface device structure
3639 * @p: pointer to an address structure
3641 * Returns 0 on success, negative on failure
3643 static int ixgbe_set_mac(struct net_device *netdev, void *p)
3645 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3646 struct ixgbe_hw *hw = &adapter->hw;
3647 struct sockaddr *addr = p;
3649 if (!is_valid_ether_addr(addr->sa_data))
3650 return -EADDRNOTAVAIL;
3652 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3653 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3655 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
3660 #ifdef CONFIG_NET_POLL_CONTROLLER
3662 * Polling 'interrupt' - used by things like netconsole to send skbs
3663 * without having to re-enable interrupts. It's not called while
3664 * the interrupt routine is executing.
3666 static void ixgbe_netpoll(struct net_device *netdev)
3668 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3670 disable_irq(adapter->pdev->irq);
3671 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
3672 ixgbe_intr(adapter->pdev->irq, netdev);
3673 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
3674 enable_irq(adapter->pdev->irq);
3679 * ixgbe_link_config - set up initial link with default speed and duplex
3680 * @hw: pointer to private hardware struct
3682 * Returns 0 on success, negative on failure
3684 static int ixgbe_link_config(struct ixgbe_hw *hw)
3686 u32 autoneg = IXGBE_LINK_SPEED_10GB_FULL;
3688 /* must always autoneg for both 1G and 10G link */
3689 hw->mac.autoneg = true;
3691 return hw->mac.ops.setup_link_speed(hw, autoneg, true, true);
3695 * ixgbe_probe - Device Initialization Routine
3696 * @pdev: PCI device information struct
3697 * @ent: entry in ixgbe_pci_tbl
3699 * Returns 0 on success, negative on failure
3701 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
3702 * The OS initialization, configuring of the adapter private structure,
3703 * and a hardware reset occur.
3705 static int __devinit ixgbe_probe(struct pci_dev *pdev,
3706 const struct pci_device_id *ent)
3708 struct net_device *netdev;
3709 struct ixgbe_adapter *adapter = NULL;
3710 struct ixgbe_hw *hw;
3711 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
3712 static int cards_found;
3713 int i, err, pci_using_dac;
3714 u16 link_status, link_speed, link_width;
3717 err = pci_enable_device(pdev);
3721 if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK) &&
3722 !pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK)) {
3725 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
3727 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
3729 dev_err(&pdev->dev, "No usable DMA "
3730 "configuration, aborting\n");
3737 err = pci_request_regions(pdev, ixgbe_driver_name);
3739 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3743 pci_set_master(pdev);
3744 pci_save_state(pdev);
3746 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
3749 goto err_alloc_etherdev;
3752 SET_NETDEV_DEV(netdev, &pdev->dev);
3754 pci_set_drvdata(pdev, netdev);
3755 adapter = netdev_priv(netdev);
3757 adapter->netdev = netdev;
3758 adapter->pdev = pdev;
3761 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
3763 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3764 pci_resource_len(pdev, 0));
3770 for (i = 1; i <= 5; i++) {
3771 if (pci_resource_len(pdev, i) == 0)
3775 netdev->open = &ixgbe_open;
3776 netdev->stop = &ixgbe_close;
3777 netdev->hard_start_xmit = &ixgbe_xmit_frame;
3778 netdev->get_stats = &ixgbe_get_stats;
3779 netdev->set_rx_mode = &ixgbe_set_rx_mode;
3780 netdev->set_multicast_list = &ixgbe_set_rx_mode;
3781 netdev->set_mac_address = &ixgbe_set_mac;
3782 netdev->change_mtu = &ixgbe_change_mtu;
3783 ixgbe_set_ethtool_ops(netdev);
3784 netdev->tx_timeout = &ixgbe_tx_timeout;
3785 netdev->watchdog_timeo = 5 * HZ;
3786 netdev->vlan_rx_register = ixgbe_vlan_rx_register;
3787 netdev->vlan_rx_add_vid = ixgbe_vlan_rx_add_vid;
3788 netdev->vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid;
3789 #ifdef CONFIG_NET_POLL_CONTROLLER
3790 netdev->poll_controller = ixgbe_netpoll;
3792 strcpy(netdev->name, pci_name(pdev));
3794 adapter->bd_number = cards_found;
3797 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3798 hw->mac.type = ii->mac;
3801 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
3802 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
3803 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
3804 if (!(eec & (1 << 8)))
3805 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
3808 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
3809 /* phy->sfp_type = ixgbe_sfp_type_unknown; */
3811 err = ii->get_invariants(hw);
3815 /* setup the private structure */
3816 err = ixgbe_sw_init(adapter);
3820 /* reset_hw fills in the perm_addr as well */
3821 err = hw->mac.ops.reset_hw(hw);
3823 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
3827 netdev->features = NETIF_F_SG |
3829 NETIF_F_HW_VLAN_TX |
3830 NETIF_F_HW_VLAN_RX |
3831 NETIF_F_HW_VLAN_FILTER;
3833 netdev->features |= NETIF_F_IPV6_CSUM;
3834 netdev->features |= NETIF_F_TSO;
3835 netdev->features |= NETIF_F_TSO6;
3836 netdev->features |= NETIF_F_LRO;
3838 netdev->vlan_features |= NETIF_F_TSO;
3839 netdev->vlan_features |= NETIF_F_TSO6;
3840 netdev->vlan_features |= NETIF_F_IP_CSUM;
3841 netdev->vlan_features |= NETIF_F_SG;
3844 netdev->features |= NETIF_F_HIGHDMA;
3846 /* make sure the EEPROM is good */
3847 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
3848 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
3853 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
3854 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
3856 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
3857 dev_err(&pdev->dev, "invalid MAC address\n");
3862 init_timer(&adapter->watchdog_timer);
3863 adapter->watchdog_timer.function = &ixgbe_watchdog;
3864 adapter->watchdog_timer.data = (unsigned long)adapter;
3866 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
3867 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
3869 err = ixgbe_init_interrupt_scheme(adapter);
3873 /* print bus type/speed/width info */
3874 pci_read_config_word(pdev, IXGBE_PCI_LINK_STATUS, &link_status);
3875 link_speed = link_status & IXGBE_PCI_LINK_SPEED;
3876 link_width = link_status & IXGBE_PCI_LINK_WIDTH;
3877 dev_info(&pdev->dev, "(PCI Express:%s:%s) "
3878 "%02x:%02x:%02x:%02x:%02x:%02x\n",
3879 ((link_speed == IXGBE_PCI_LINK_SPEED_5000) ? "5.0Gb/s" :
3880 (link_speed == IXGBE_PCI_LINK_SPEED_2500) ? "2.5Gb/s" :
3882 ((link_width == IXGBE_PCI_LINK_WIDTH_8) ? "Width x8" :
3883 (link_width == IXGBE_PCI_LINK_WIDTH_4) ? "Width x4" :
3884 (link_width == IXGBE_PCI_LINK_WIDTH_2) ? "Width x2" :
3885 (link_width == IXGBE_PCI_LINK_WIDTH_1) ? "Width x1" :
3887 netdev->dev_addr[0], netdev->dev_addr[1], netdev->dev_addr[2],
3888 netdev->dev_addr[3], netdev->dev_addr[4], netdev->dev_addr[5]);
3889 ixgbe_read_pba_num_generic(hw, &part_num);
3890 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
3891 hw->mac.type, hw->phy.type,
3892 (part_num >> 8), (part_num & 0xff));
3894 if (link_width <= IXGBE_PCI_LINK_WIDTH_4) {
3895 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
3896 "this card is not sufficient for optimal "
3898 dev_warn(&pdev->dev, "For optimal performance a x8 "
3899 "PCI-Express slot is required.\n");
3902 /* reset the hardware with the new settings */
3903 hw->mac.ops.start_hw(hw);
3905 /* link_config depends on start_hw being called at least once */
3906 err = ixgbe_link_config(hw);
3908 dev_err(&pdev->dev, "setup_link_speed FAILED %d\n", err);
3912 netif_carrier_off(netdev);
3913 netif_tx_stop_all_queues(netdev);
3915 ixgbe_napi_add_all(adapter);
3917 strcpy(netdev->name, "eth%d");
3918 err = register_netdev(netdev);
3922 #ifdef CONFIG_IXGBE_DCA
3923 if (dca_add_requester(&pdev->dev) == 0) {
3924 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
3925 /* always use CB2 mode, difference is masked
3926 * in the CB driver */
3927 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
3928 ixgbe_setup_dca(adapter);
3932 dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
3937 ixgbe_release_hw_control(adapter);
3940 ixgbe_reset_interrupt_capability(adapter);
3942 iounmap(hw->hw_addr);
3944 free_netdev(netdev);
3946 pci_release_regions(pdev);
3949 pci_disable_device(pdev);
3954 * ixgbe_remove - Device Removal Routine
3955 * @pdev: PCI device information struct
3957 * ixgbe_remove is called by the PCI subsystem to alert the driver
3958 * that it should release a PCI device. The could be caused by a
3959 * Hot-Plug event, or because the driver is going to be removed from
3962 static void __devexit ixgbe_remove(struct pci_dev *pdev)
3964 struct net_device *netdev = pci_get_drvdata(pdev);
3965 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3967 set_bit(__IXGBE_DOWN, &adapter->state);
3968 del_timer_sync(&adapter->watchdog_timer);
3970 flush_scheduled_work();
3972 #ifdef CONFIG_IXGBE_DCA
3973 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3974 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
3975 dca_remove_requester(&pdev->dev);
3976 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
3980 unregister_netdev(netdev);
3982 ixgbe_reset_interrupt_capability(adapter);
3984 ixgbe_release_hw_control(adapter);
3986 iounmap(adapter->hw.hw_addr);
3987 pci_release_regions(pdev);
3989 DPRINTK(PROBE, INFO, "complete\n");
3990 ixgbe_napi_del_all(adapter);
3991 kfree(adapter->tx_ring);
3992 kfree(adapter->rx_ring);
3994 free_netdev(netdev);
3996 pci_disable_device(pdev);
4000 * ixgbe_io_error_detected - called when PCI error is detected
4001 * @pdev: Pointer to PCI device
4002 * @state: The current pci connection state
4004 * This function is called after a PCI bus error affecting
4005 * this device has been detected.
4007 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
4008 pci_channel_state_t state)
4010 struct net_device *netdev = pci_get_drvdata(pdev);
4011 struct ixgbe_adapter *adapter = netdev->priv;
4013 netif_device_detach(netdev);
4015 if (netif_running(netdev))
4016 ixgbe_down(adapter);
4017 pci_disable_device(pdev);
4019 /* Request a slot reset. */
4020 return PCI_ERS_RESULT_NEED_RESET;
4024 * ixgbe_io_slot_reset - called after the pci bus has been reset.
4025 * @pdev: Pointer to PCI device
4027 * Restart the card from scratch, as if from a cold-boot.
4029 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
4031 struct net_device *netdev = pci_get_drvdata(pdev);
4032 struct ixgbe_adapter *adapter = netdev->priv;
4034 if (pci_enable_device(pdev)) {
4036 "Cannot re-enable PCI device after reset.\n");
4037 return PCI_ERS_RESULT_DISCONNECT;
4039 pci_set_master(pdev);
4040 pci_restore_state(pdev);
4042 pci_enable_wake(pdev, PCI_D3hot, 0);
4043 pci_enable_wake(pdev, PCI_D3cold, 0);
4045 ixgbe_reset(adapter);
4047 return PCI_ERS_RESULT_RECOVERED;
4051 * ixgbe_io_resume - called when traffic can start flowing again.
4052 * @pdev: Pointer to PCI device
4054 * This callback is called when the error recovery driver tells us that
4055 * its OK to resume normal operation.
4057 static void ixgbe_io_resume(struct pci_dev *pdev)
4059 struct net_device *netdev = pci_get_drvdata(pdev);
4060 struct ixgbe_adapter *adapter = netdev->priv;
4062 if (netif_running(netdev)) {
4063 if (ixgbe_up(adapter)) {
4064 DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
4069 netif_device_attach(netdev);
4072 static struct pci_error_handlers ixgbe_err_handler = {
4073 .error_detected = ixgbe_io_error_detected,
4074 .slot_reset = ixgbe_io_slot_reset,
4075 .resume = ixgbe_io_resume,
4078 static struct pci_driver ixgbe_driver = {
4079 .name = ixgbe_driver_name,
4080 .id_table = ixgbe_pci_tbl,
4081 .probe = ixgbe_probe,
4082 .remove = __devexit_p(ixgbe_remove),
4084 .suspend = ixgbe_suspend,
4085 .resume = ixgbe_resume,
4087 .shutdown = ixgbe_shutdown,
4088 .err_handler = &ixgbe_err_handler
4092 * ixgbe_init_module - Driver Registration Routine
4094 * ixgbe_init_module is the first routine called when the driver is
4095 * loaded. All it does is register with the PCI subsystem.
4097 static int __init ixgbe_init_module(void)
4100 printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
4101 ixgbe_driver_string, ixgbe_driver_version);
4103 printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
4105 #ifdef CONFIG_IXGBE_DCA
4106 dca_register_notify(&dca_notifier);
4109 ret = pci_register_driver(&ixgbe_driver);
4113 module_init(ixgbe_init_module);
4116 * ixgbe_exit_module - Driver Exit Cleanup Routine
4118 * ixgbe_exit_module is called just before the driver is removed
4121 static void __exit ixgbe_exit_module(void)
4123 #ifdef CONFIG_IXGBE_DCA
4124 dca_unregister_notify(&dca_notifier);
4126 pci_unregister_driver(&ixgbe_driver);
4129 #ifdef CONFIG_IXGBE_DCA
4130 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
4135 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
4136 __ixgbe_notify_dca);
4138 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4140 #endif /* CONFIG_IXGBE_DCA */
4142 module_exit(ixgbe_exit_module);