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_82598AT),
73 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
75 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
77 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
80 /* required last entry */
83 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
85 #ifdef CONFIG_IXGBE_DCA
86 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
88 static struct notifier_block dca_notifier = {
89 .notifier_call = ixgbe_notify_dca,
95 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
96 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
97 MODULE_LICENSE("GPL");
98 MODULE_VERSION(DRV_VERSION);
100 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
102 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
106 /* Let firmware take over control of h/w */
107 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
108 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
109 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
112 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
116 /* Let firmware know the driver has taken over */
117 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
118 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
119 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
122 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, u16 int_alloc_entry,
127 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
128 index = (int_alloc_entry >> 2) & 0x1F;
129 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR(index));
130 ivar &= ~(0xFF << (8 * (int_alloc_entry & 0x3)));
131 ivar |= (msix_vector << (8 * (int_alloc_entry & 0x3)));
132 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR(index), ivar);
135 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
136 struct ixgbe_tx_buffer
139 if (tx_buffer_info->dma) {
140 pci_unmap_page(adapter->pdev, tx_buffer_info->dma,
141 tx_buffer_info->length, PCI_DMA_TODEVICE);
142 tx_buffer_info->dma = 0;
144 if (tx_buffer_info->skb) {
145 dev_kfree_skb_any(tx_buffer_info->skb);
146 tx_buffer_info->skb = NULL;
148 /* tx_buffer_info must be completely set up in the transmit path */
151 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
152 struct ixgbe_ring *tx_ring,
155 struct ixgbe_hw *hw = &adapter->hw;
158 /* Detect a transmit hang in hardware, this serializes the
159 * check with the clearing of time_stamp and movement of eop */
160 head = IXGBE_READ_REG(hw, tx_ring->head);
161 tail = IXGBE_READ_REG(hw, tx_ring->tail);
162 adapter->detect_tx_hung = false;
163 if ((head != tail) &&
164 tx_ring->tx_buffer_info[eop].time_stamp &&
165 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
166 !(IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & IXGBE_TFCS_TXOFF)) {
167 /* detected Tx unit hang */
168 union ixgbe_adv_tx_desc *tx_desc;
169 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
170 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
172 " TDH, TDT <%x>, <%x>\n"
173 " next_to_use <%x>\n"
174 " next_to_clean <%x>\n"
175 "tx_buffer_info[next_to_clean]\n"
176 " time_stamp <%lx>\n"
178 tx_ring->queue_index,
180 tx_ring->next_to_use, eop,
181 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
188 #define IXGBE_MAX_TXD_PWR 14
189 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
191 /* Tx Descriptors needed, worst case */
192 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
193 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
194 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
195 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
197 #define GET_TX_HEAD_FROM_RING(ring) (\
199 ((union ixgbe_adv_tx_desc *)(ring)->desc + (ring)->count))
200 static void ixgbe_tx_timeout(struct net_device *netdev);
203 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
204 * @adapter: board private structure
205 * @tx_ring: tx ring to clean
207 static bool ixgbe_clean_tx_irq(struct ixgbe_adapter *adapter,
208 struct ixgbe_ring *tx_ring)
210 union ixgbe_adv_tx_desc *tx_desc;
211 struct ixgbe_tx_buffer *tx_buffer_info;
212 struct net_device *netdev = adapter->netdev;
216 unsigned int count = 0;
217 unsigned int total_bytes = 0, total_packets = 0;
220 head = GET_TX_HEAD_FROM_RING(tx_ring);
221 head = le32_to_cpu(head);
222 i = tx_ring->next_to_clean;
225 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
226 tx_buffer_info = &tx_ring->tx_buffer_info[i];
227 skb = tx_buffer_info->skb;
230 unsigned int segs, bytecount;
232 /* gso_segs is currently only valid for tcp */
233 segs = skb_shinfo(skb)->gso_segs ?: 1;
234 /* multiply data chunks by size of headers */
235 bytecount = ((segs - 1) * skb_headlen(skb)) +
237 total_packets += segs;
238 total_bytes += bytecount;
241 ixgbe_unmap_and_free_tx_resource(adapter,
245 if (i == tx_ring->count)
249 if (count == tx_ring->count)
254 head = GET_TX_HEAD_FROM_RING(tx_ring);
255 head = le32_to_cpu(head);
261 tx_ring->next_to_clean = i;
263 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
264 if (unlikely(count && netif_carrier_ok(netdev) &&
265 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
266 /* Make sure that anybody stopping the queue after this
267 * sees the new next_to_clean.
270 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
271 !test_bit(__IXGBE_DOWN, &adapter->state)) {
272 netif_wake_subqueue(netdev, tx_ring->queue_index);
273 ++adapter->restart_queue;
277 if (adapter->detect_tx_hung) {
278 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
279 /* schedule immediate reset if we believe we hung */
281 "tx hang %d detected, resetting adapter\n",
282 adapter->tx_timeout_count + 1);
283 ixgbe_tx_timeout(adapter->netdev);
287 /* re-arm the interrupt */
288 if ((total_packets >= tx_ring->work_limit) ||
289 (count == tx_ring->count))
290 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, tx_ring->v_idx);
292 tx_ring->total_bytes += total_bytes;
293 tx_ring->total_packets += total_packets;
294 tx_ring->stats.bytes += total_bytes;
295 tx_ring->stats.packets += total_packets;
296 adapter->net_stats.tx_bytes += total_bytes;
297 adapter->net_stats.tx_packets += total_packets;
298 return (total_packets ? true : false);
301 #ifdef CONFIG_IXGBE_DCA
302 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
303 struct ixgbe_ring *rx_ring)
307 int q = rx_ring - adapter->rx_ring;
309 if (rx_ring->cpu != cpu) {
310 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
311 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
312 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
313 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
314 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
315 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
321 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
322 struct ixgbe_ring *tx_ring)
326 int q = tx_ring - adapter->tx_ring;
328 if (tx_ring->cpu != cpu) {
329 txctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q));
330 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
331 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
332 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
333 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q), txctrl);
339 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
343 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
346 for (i = 0; i < adapter->num_tx_queues; i++) {
347 adapter->tx_ring[i].cpu = -1;
348 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
350 for (i = 0; i < adapter->num_rx_queues; i++) {
351 adapter->rx_ring[i].cpu = -1;
352 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
356 static int __ixgbe_notify_dca(struct device *dev, void *data)
358 struct net_device *netdev = dev_get_drvdata(dev);
359 struct ixgbe_adapter *adapter = netdev_priv(netdev);
360 unsigned long event = *(unsigned long *)data;
363 case DCA_PROVIDER_ADD:
364 /* if we're already enabled, don't do it again */
365 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
367 /* Always use CB2 mode, difference is masked
368 * in the CB driver. */
369 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
370 if (dca_add_requester(dev) == 0) {
371 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
372 ixgbe_setup_dca(adapter);
375 /* Fall Through since DCA is disabled. */
376 case DCA_PROVIDER_REMOVE:
377 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
378 dca_remove_requester(dev);
379 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
380 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
388 #endif /* CONFIG_IXGBE_DCA */
390 * ixgbe_receive_skb - Send a completed packet up the stack
391 * @adapter: board private structure
392 * @skb: packet to send up
393 * @status: hardware indication of status of receive
394 * @rx_ring: rx descriptor ring (for a specific queue) to setup
395 * @rx_desc: rx descriptor
397 static void ixgbe_receive_skb(struct ixgbe_adapter *adapter,
398 struct sk_buff *skb, u8 status,
399 struct ixgbe_ring *ring,
400 union ixgbe_adv_rx_desc *rx_desc)
402 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
403 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
405 if (adapter->netdev->features & NETIF_F_LRO &&
406 skb->ip_summed == CHECKSUM_UNNECESSARY) {
407 if (adapter->vlgrp && is_vlan)
408 lro_vlan_hwaccel_receive_skb(&ring->lro_mgr, skb,
412 lro_receive_skb(&ring->lro_mgr, skb, rx_desc);
413 ring->lro_used = true;
415 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
416 if (adapter->vlgrp && is_vlan)
417 vlan_hwaccel_receive_skb(skb, adapter->vlgrp, tag);
419 netif_receive_skb(skb);
421 if (adapter->vlgrp && is_vlan)
422 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
430 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
431 * @adapter: address of board private structure
432 * @status_err: hardware indication of status of receive
433 * @skb: skb currently being received and modified
435 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
436 u32 status_err, struct sk_buff *skb)
438 skb->ip_summed = CHECKSUM_NONE;
440 /* Rx csum disabled */
441 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
444 /* if IP and error */
445 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
446 (status_err & IXGBE_RXDADV_ERR_IPE)) {
447 adapter->hw_csum_rx_error++;
451 if (!(status_err & IXGBE_RXD_STAT_L4CS))
454 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
455 adapter->hw_csum_rx_error++;
459 /* It must be a TCP or UDP packet with a valid checksum */
460 skb->ip_summed = CHECKSUM_UNNECESSARY;
461 adapter->hw_csum_rx_good++;
465 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
466 * @adapter: address of board private structure
468 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
469 struct ixgbe_ring *rx_ring,
472 struct pci_dev *pdev = adapter->pdev;
473 union ixgbe_adv_rx_desc *rx_desc;
474 struct ixgbe_rx_buffer *bi;
476 unsigned int bufsz = rx_ring->rx_buf_len + NET_IP_ALIGN;
478 i = rx_ring->next_to_use;
479 bi = &rx_ring->rx_buffer_info[i];
481 while (cleaned_count--) {
482 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
485 (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
487 bi->page = alloc_page(GFP_ATOMIC);
489 adapter->alloc_rx_page_failed++;
494 /* use a half page if we're re-using */
495 bi->page_offset ^= (PAGE_SIZE / 2);
498 bi->page_dma = pci_map_page(pdev, bi->page,
505 struct sk_buff *skb = netdev_alloc_skb(adapter->netdev,
509 adapter->alloc_rx_buff_failed++;
514 * Make buffer alignment 2 beyond a 16 byte boundary
515 * this will result in a 16 byte aligned IP header after
516 * the 14 byte MAC header is removed
518 skb_reserve(skb, NET_IP_ALIGN);
521 bi->dma = pci_map_single(pdev, skb->data, bufsz,
524 /* Refresh the desc even if buffer_addrs didn't change because
525 * each write-back erases this info. */
526 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
527 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
528 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
530 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
534 if (i == rx_ring->count)
536 bi = &rx_ring->rx_buffer_info[i];
540 if (rx_ring->next_to_use != i) {
541 rx_ring->next_to_use = i;
543 i = (rx_ring->count - 1);
546 * Force memory writes to complete before letting h/w
547 * know there are new descriptors to fetch. (Only
548 * applicable for weak-ordered memory model archs,
552 writel(i, adapter->hw.hw_addr + rx_ring->tail);
556 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
558 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
561 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
563 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
566 static bool ixgbe_clean_rx_irq(struct ixgbe_adapter *adapter,
567 struct ixgbe_ring *rx_ring,
568 int *work_done, int work_to_do)
570 struct pci_dev *pdev = adapter->pdev;
571 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
572 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
577 bool cleaned = false;
578 int cleaned_count = 0;
579 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
581 i = rx_ring->next_to_clean;
582 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
583 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
584 rx_buffer_info = &rx_ring->rx_buffer_info[i];
586 while (staterr & IXGBE_RXD_STAT_DD) {
588 if (*work_done >= work_to_do)
592 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
593 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
594 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
595 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
596 if (hdr_info & IXGBE_RXDADV_SPH)
597 adapter->rx_hdr_split++;
598 if (len > IXGBE_RX_HDR_SIZE)
599 len = IXGBE_RX_HDR_SIZE;
600 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
602 len = le16_to_cpu(rx_desc->wb.upper.length);
606 skb = rx_buffer_info->skb;
607 prefetch(skb->data - NET_IP_ALIGN);
608 rx_buffer_info->skb = NULL;
610 if (len && !skb_shinfo(skb)->nr_frags) {
611 pci_unmap_single(pdev, rx_buffer_info->dma,
612 rx_ring->rx_buf_len + NET_IP_ALIGN,
618 pci_unmap_page(pdev, rx_buffer_info->page_dma,
619 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
620 rx_buffer_info->page_dma = 0;
621 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
622 rx_buffer_info->page,
623 rx_buffer_info->page_offset,
626 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
627 (page_count(rx_buffer_info->page) != 1))
628 rx_buffer_info->page = NULL;
630 get_page(rx_buffer_info->page);
632 skb->len += upper_len;
633 skb->data_len += upper_len;
634 skb->truesize += upper_len;
638 if (i == rx_ring->count)
640 next_buffer = &rx_ring->rx_buffer_info[i];
642 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
646 if (staterr & IXGBE_RXD_STAT_EOP) {
647 rx_ring->stats.packets++;
648 rx_ring->stats.bytes += skb->len;
650 rx_buffer_info->skb = next_buffer->skb;
651 rx_buffer_info->dma = next_buffer->dma;
652 next_buffer->skb = skb;
653 next_buffer->dma = 0;
654 adapter->non_eop_descs++;
658 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
659 dev_kfree_skb_irq(skb);
663 ixgbe_rx_checksum(adapter, staterr, skb);
665 /* probably a little skewed due to removing CRC */
666 total_rx_bytes += skb->len;
669 skb->protocol = eth_type_trans(skb, adapter->netdev);
670 ixgbe_receive_skb(adapter, skb, staterr, rx_ring, rx_desc);
673 rx_desc->wb.upper.status_error = 0;
675 /* return some buffers to hardware, one at a time is too slow */
676 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
677 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
681 /* use prefetched values */
683 rx_buffer_info = next_buffer;
685 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
688 if (rx_ring->lro_used) {
689 lro_flush_all(&rx_ring->lro_mgr);
690 rx_ring->lro_used = false;
693 rx_ring->next_to_clean = i;
694 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
697 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
699 rx_ring->total_packets += total_rx_packets;
700 rx_ring->total_bytes += total_rx_bytes;
701 adapter->net_stats.rx_bytes += total_rx_bytes;
702 adapter->net_stats.rx_packets += total_rx_packets;
707 static int ixgbe_clean_rxonly(struct napi_struct *, int);
709 * ixgbe_configure_msix - Configure MSI-X hardware
710 * @adapter: board private structure
712 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
715 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
717 struct ixgbe_q_vector *q_vector;
718 int i, j, q_vectors, v_idx, r_idx;
721 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
723 /* Populate the IVAR table and set the ITR values to the
724 * corresponding register.
726 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
727 q_vector = &adapter->q_vector[v_idx];
728 /* XXX for_each_bit(...) */
729 r_idx = find_first_bit(q_vector->rxr_idx,
730 adapter->num_rx_queues);
732 for (i = 0; i < q_vector->rxr_count; i++) {
733 j = adapter->rx_ring[r_idx].reg_idx;
734 ixgbe_set_ivar(adapter, IXGBE_IVAR_RX_QUEUE(j), v_idx);
735 r_idx = find_next_bit(q_vector->rxr_idx,
736 adapter->num_rx_queues,
739 r_idx = find_first_bit(q_vector->txr_idx,
740 adapter->num_tx_queues);
742 for (i = 0; i < q_vector->txr_count; i++) {
743 j = adapter->tx_ring[r_idx].reg_idx;
744 ixgbe_set_ivar(adapter, IXGBE_IVAR_TX_QUEUE(j), v_idx);
745 r_idx = find_next_bit(q_vector->txr_idx,
746 adapter->num_tx_queues,
750 /* if this is a tx only vector halve the interrupt rate */
751 if (q_vector->txr_count && !q_vector->rxr_count)
752 q_vector->eitr = (adapter->eitr_param >> 1);
755 q_vector->eitr = adapter->eitr_param;
757 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx),
758 EITR_INTS_PER_SEC_TO_REG(q_vector->eitr));
761 ixgbe_set_ivar(adapter, IXGBE_IVAR_OTHER_CAUSES_INDEX, v_idx);
762 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
764 /* set up to autoclear timer, and the vectors */
765 mask = IXGBE_EIMS_ENABLE_MASK;
766 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
767 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
774 latency_invalid = 255
778 * ixgbe_update_itr - update the dynamic ITR value based on statistics
779 * @adapter: pointer to adapter
780 * @eitr: eitr setting (ints per sec) to give last timeslice
781 * @itr_setting: current throttle rate in ints/second
782 * @packets: the number of packets during this measurement interval
783 * @bytes: the number of bytes during this measurement interval
785 * Stores a new ITR value based on packets and byte
786 * counts during the last interrupt. The advantage of per interrupt
787 * computation is faster updates and more accurate ITR for the current
788 * traffic pattern. Constants in this function were computed
789 * based on theoretical maximum wire speed and thresholds were set based
790 * on testing data as well as attempting to minimize response time
791 * while increasing bulk throughput.
792 * this functionality is controlled by the InterruptThrottleRate module
793 * parameter (see ixgbe_param.c)
795 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
796 u32 eitr, u8 itr_setting,
797 int packets, int bytes)
799 unsigned int retval = itr_setting;
804 goto update_itr_done;
807 /* simple throttlerate management
808 * 0-20MB/s lowest (100000 ints/s)
809 * 20-100MB/s low (20000 ints/s)
810 * 100-1249MB/s bulk (8000 ints/s)
812 /* what was last interrupt timeslice? */
813 timepassed_us = 1000000/eitr;
814 bytes_perint = bytes / timepassed_us; /* bytes/usec */
816 switch (itr_setting) {
818 if (bytes_perint > adapter->eitr_low)
819 retval = low_latency;
822 if (bytes_perint > adapter->eitr_high)
823 retval = bulk_latency;
824 else if (bytes_perint <= adapter->eitr_low)
825 retval = lowest_latency;
828 if (bytes_perint <= adapter->eitr_high)
829 retval = low_latency;
837 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
839 struct ixgbe_adapter *adapter = q_vector->adapter;
840 struct ixgbe_hw *hw = &adapter->hw;
842 u8 current_itr, ret_itr;
843 int i, r_idx, v_idx = ((void *)q_vector - (void *)(adapter->q_vector)) /
844 sizeof(struct ixgbe_q_vector);
845 struct ixgbe_ring *rx_ring, *tx_ring;
847 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
848 for (i = 0; i < q_vector->txr_count; i++) {
849 tx_ring = &(adapter->tx_ring[r_idx]);
850 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
852 tx_ring->total_packets,
853 tx_ring->total_bytes);
854 /* if the result for this queue would decrease interrupt
855 * rate for this vector then use that result */
856 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
857 q_vector->tx_itr - 1 : ret_itr);
858 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
862 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
863 for (i = 0; i < q_vector->rxr_count; i++) {
864 rx_ring = &(adapter->rx_ring[r_idx]);
865 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
867 rx_ring->total_packets,
868 rx_ring->total_bytes);
869 /* if the result for this queue would decrease interrupt
870 * rate for this vector then use that result */
871 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
872 q_vector->rx_itr - 1 : ret_itr);
873 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
877 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
879 switch (current_itr) {
880 /* counts and packets in update_itr are dependent on these numbers */
885 new_itr = 20000; /* aka hwitr = ~200 */
893 if (new_itr != q_vector->eitr) {
895 /* do an exponential smoothing */
896 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
897 q_vector->eitr = new_itr;
898 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
899 /* must write high and low 16 bits to reset counter */
900 DPRINTK(TX_ERR, DEBUG, "writing eitr(%d): %08X\n", v_idx,
902 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg | (itr_reg)<<16);
908 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
910 struct ixgbe_hw *hw = &adapter->hw;
912 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
913 (eicr & IXGBE_EICR_GPI_SDP1)) {
914 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
915 /* write to clear the interrupt */
916 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
920 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
922 struct ixgbe_hw *hw = &adapter->hw;
925 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
926 adapter->link_check_timeout = jiffies;
927 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
928 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
929 schedule_work(&adapter->watchdog_task);
933 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
935 struct net_device *netdev = data;
936 struct ixgbe_adapter *adapter = netdev_priv(netdev);
937 struct ixgbe_hw *hw = &adapter->hw;
938 u32 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
940 if (eicr & IXGBE_EICR_LSC)
941 ixgbe_check_lsc(adapter);
943 ixgbe_check_fan_failure(adapter, eicr);
945 if (!test_bit(__IXGBE_DOWN, &adapter->state))
946 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
951 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
953 struct ixgbe_q_vector *q_vector = data;
954 struct ixgbe_adapter *adapter = q_vector->adapter;
955 struct ixgbe_ring *tx_ring;
958 if (!q_vector->txr_count)
961 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
962 for (i = 0; i < q_vector->txr_count; i++) {
963 tx_ring = &(adapter->tx_ring[r_idx]);
964 #ifdef CONFIG_IXGBE_DCA
965 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
966 ixgbe_update_tx_dca(adapter, tx_ring);
968 tx_ring->total_bytes = 0;
969 tx_ring->total_packets = 0;
970 ixgbe_clean_tx_irq(adapter, tx_ring);
971 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
979 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
981 * @data: pointer to our q_vector struct for this interrupt vector
983 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
985 struct ixgbe_q_vector *q_vector = data;
986 struct ixgbe_adapter *adapter = q_vector->adapter;
987 struct ixgbe_ring *rx_ring;
991 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
992 for (i = 0; i < q_vector->rxr_count; i++) {
993 rx_ring = &(adapter->rx_ring[r_idx]);
994 rx_ring->total_bytes = 0;
995 rx_ring->total_packets = 0;
996 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1000 if (!q_vector->rxr_count)
1003 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1004 rx_ring = &(adapter->rx_ring[r_idx]);
1005 /* disable interrupts on this vector only */
1006 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, rx_ring->v_idx);
1007 netif_rx_schedule(adapter->netdev, &q_vector->napi);
1012 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1014 ixgbe_msix_clean_rx(irq, data);
1015 ixgbe_msix_clean_tx(irq, data);
1021 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1022 * @napi: napi struct with our devices info in it
1023 * @budget: amount of work driver is allowed to do this pass, in packets
1025 * This function is optimized for cleaning one queue only on a single
1028 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1030 struct ixgbe_q_vector *q_vector =
1031 container_of(napi, struct ixgbe_q_vector, napi);
1032 struct ixgbe_adapter *adapter = q_vector->adapter;
1033 struct ixgbe_ring *rx_ring = NULL;
1037 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1038 rx_ring = &(adapter->rx_ring[r_idx]);
1039 #ifdef CONFIG_IXGBE_DCA
1040 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1041 ixgbe_update_rx_dca(adapter, rx_ring);
1044 ixgbe_clean_rx_irq(adapter, rx_ring, &work_done, budget);
1046 /* If all Rx work done, exit the polling mode */
1047 if (work_done < budget) {
1048 netif_rx_complete(adapter->netdev, napi);
1049 if (adapter->itr_setting & 3)
1050 ixgbe_set_itr_msix(q_vector);
1051 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1052 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, rx_ring->v_idx);
1059 * ixgbe_clean_rxonly_many - msix (aka one shot) rx clean routine
1060 * @napi: napi struct with our devices info in it
1061 * @budget: amount of work driver is allowed to do this pass, in packets
1063 * This function will clean more than one rx queue associated with a
1066 static int ixgbe_clean_rxonly_many(struct napi_struct *napi, int budget)
1068 struct ixgbe_q_vector *q_vector =
1069 container_of(napi, struct ixgbe_q_vector, napi);
1070 struct ixgbe_adapter *adapter = q_vector->adapter;
1071 struct ixgbe_ring *rx_ring = NULL;
1072 int work_done = 0, i;
1074 u16 enable_mask = 0;
1076 /* attempt to distribute budget to each queue fairly, but don't allow
1077 * the budget to go below 1 because we'll exit polling */
1078 budget /= (q_vector->rxr_count ?: 1);
1079 budget = max(budget, 1);
1080 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1081 for (i = 0; i < q_vector->rxr_count; i++) {
1082 rx_ring = &(adapter->rx_ring[r_idx]);
1083 #ifdef CONFIG_IXGBE_DCA
1084 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1085 ixgbe_update_rx_dca(adapter, rx_ring);
1087 ixgbe_clean_rx_irq(adapter, rx_ring, &work_done, budget);
1088 enable_mask |= rx_ring->v_idx;
1089 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1093 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1094 rx_ring = &(adapter->rx_ring[r_idx]);
1095 /* If all Rx work done, exit the polling mode */
1096 if (work_done < budget) {
1097 netif_rx_complete(adapter->netdev, napi);
1098 if (adapter->itr_setting & 3)
1099 ixgbe_set_itr_msix(q_vector);
1100 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1101 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, enable_mask);
1107 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1110 a->q_vector[v_idx].adapter = a;
1111 set_bit(r_idx, a->q_vector[v_idx].rxr_idx);
1112 a->q_vector[v_idx].rxr_count++;
1113 a->rx_ring[r_idx].v_idx = 1 << v_idx;
1116 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1119 a->q_vector[v_idx].adapter = a;
1120 set_bit(r_idx, a->q_vector[v_idx].txr_idx);
1121 a->q_vector[v_idx].txr_count++;
1122 a->tx_ring[r_idx].v_idx = 1 << v_idx;
1126 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1127 * @adapter: board private structure to initialize
1128 * @vectors: allotted vector count for descriptor rings
1130 * This function maps descriptor rings to the queue-specific vectors
1131 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1132 * one vector per ring/queue, but on a constrained vector budget, we
1133 * group the rings as "efficiently" as possible. You would add new
1134 * mapping configurations in here.
1136 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1140 int rxr_idx = 0, txr_idx = 0;
1141 int rxr_remaining = adapter->num_rx_queues;
1142 int txr_remaining = adapter->num_tx_queues;
1147 /* No mapping required if MSI-X is disabled. */
1148 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1152 * The ideal configuration...
1153 * We have enough vectors to map one per queue.
1155 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1156 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1157 map_vector_to_rxq(adapter, v_start, rxr_idx);
1159 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1160 map_vector_to_txq(adapter, v_start, txr_idx);
1166 * If we don't have enough vectors for a 1-to-1
1167 * mapping, we'll have to group them so there are
1168 * multiple queues per vector.
1170 /* Re-adjusting *qpv takes care of the remainder. */
1171 for (i = v_start; i < vectors; i++) {
1172 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1173 for (j = 0; j < rqpv; j++) {
1174 map_vector_to_rxq(adapter, i, rxr_idx);
1179 for (i = v_start; i < vectors; i++) {
1180 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1181 for (j = 0; j < tqpv; j++) {
1182 map_vector_to_txq(adapter, i, txr_idx);
1193 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1194 * @adapter: board private structure
1196 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1197 * interrupts from the kernel.
1199 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1201 struct net_device *netdev = adapter->netdev;
1202 irqreturn_t (*handler)(int, void *);
1203 int i, vector, q_vectors, err;
1205 /* Decrement for Other and TCP Timer vectors */
1206 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1208 /* Map the Tx/Rx rings to the vectors we were allotted. */
1209 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1213 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1214 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1215 &ixgbe_msix_clean_many)
1216 for (vector = 0; vector < q_vectors; vector++) {
1217 handler = SET_HANDLER(&adapter->q_vector[vector]);
1218 sprintf(adapter->name[vector], "%s:v%d-%s",
1219 netdev->name, vector,
1220 (handler == &ixgbe_msix_clean_rx) ? "Rx" :
1221 ((handler == &ixgbe_msix_clean_tx) ? "Tx" : "TxRx"));
1222 err = request_irq(adapter->msix_entries[vector].vector,
1223 handler, 0, adapter->name[vector],
1224 &(adapter->q_vector[vector]));
1227 "request_irq failed for MSIX interrupt "
1228 "Error: %d\n", err);
1229 goto free_queue_irqs;
1233 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1234 err = request_irq(adapter->msix_entries[vector].vector,
1235 &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1238 "request_irq for msix_lsc failed: %d\n", err);
1239 goto free_queue_irqs;
1245 for (i = vector - 1; i >= 0; i--)
1246 free_irq(adapter->msix_entries[--vector].vector,
1247 &(adapter->q_vector[i]));
1248 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1249 pci_disable_msix(adapter->pdev);
1250 kfree(adapter->msix_entries);
1251 adapter->msix_entries = NULL;
1256 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1258 struct ixgbe_hw *hw = &adapter->hw;
1259 struct ixgbe_q_vector *q_vector = adapter->q_vector;
1261 u32 new_itr = q_vector->eitr;
1262 struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1263 struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1265 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1267 tx_ring->total_packets,
1268 tx_ring->total_bytes);
1269 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1271 rx_ring->total_packets,
1272 rx_ring->total_bytes);
1274 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1276 switch (current_itr) {
1277 /* counts and packets in update_itr are dependent on these numbers */
1278 case lowest_latency:
1282 new_itr = 20000; /* aka hwitr = ~200 */
1291 if (new_itr != q_vector->eitr) {
1293 /* do an exponential smoothing */
1294 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1295 q_vector->eitr = new_itr;
1296 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
1297 /* must write high and low 16 bits to reset counter */
1298 IXGBE_WRITE_REG(hw, IXGBE_EITR(0), itr_reg | (itr_reg)<<16);
1305 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1306 * @adapter: board private structure
1308 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1310 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1311 IXGBE_WRITE_FLUSH(&adapter->hw);
1312 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1314 for (i = 0; i < adapter->num_msix_vectors; i++)
1315 synchronize_irq(adapter->msix_entries[i].vector);
1317 synchronize_irq(adapter->pdev->irq);
1322 * ixgbe_irq_enable - Enable default interrupt generation settings
1323 * @adapter: board private structure
1325 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1328 mask = IXGBE_EIMS_ENABLE_MASK;
1329 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1330 mask |= IXGBE_EIMS_GPI_SDP1;
1331 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1332 IXGBE_WRITE_FLUSH(&adapter->hw);
1336 * ixgbe_intr - legacy mode Interrupt Handler
1337 * @irq: interrupt number
1338 * @data: pointer to a network interface device structure
1339 * @pt_regs: CPU registers structure
1341 static irqreturn_t ixgbe_intr(int irq, void *data)
1343 struct net_device *netdev = data;
1344 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1345 struct ixgbe_hw *hw = &adapter->hw;
1348 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1349 * therefore no explict interrupt disable is necessary */
1350 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1352 /* shared interrupt alert!
1353 * make sure interrupts are enabled because the read will
1354 * have disabled interrupts due to EIAM */
1355 ixgbe_irq_enable(adapter);
1356 return IRQ_NONE; /* Not our interrupt */
1359 if (eicr & IXGBE_EICR_LSC)
1360 ixgbe_check_lsc(adapter);
1362 ixgbe_check_fan_failure(adapter, eicr);
1364 if (netif_rx_schedule_prep(netdev, &adapter->q_vector[0].napi)) {
1365 adapter->tx_ring[0].total_packets = 0;
1366 adapter->tx_ring[0].total_bytes = 0;
1367 adapter->rx_ring[0].total_packets = 0;
1368 adapter->rx_ring[0].total_bytes = 0;
1369 /* would disable interrupts here but EIAM disabled it */
1370 __netif_rx_schedule(netdev, &adapter->q_vector[0].napi);
1376 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1378 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1380 for (i = 0; i < q_vectors; i++) {
1381 struct ixgbe_q_vector *q_vector = &adapter->q_vector[i];
1382 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1383 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1384 q_vector->rxr_count = 0;
1385 q_vector->txr_count = 0;
1390 * ixgbe_request_irq - initialize interrupts
1391 * @adapter: board private structure
1393 * Attempts to configure interrupts using the best available
1394 * capabilities of the hardware and kernel.
1396 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1398 struct net_device *netdev = adapter->netdev;
1401 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1402 err = ixgbe_request_msix_irqs(adapter);
1403 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1404 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1405 netdev->name, netdev);
1407 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1408 netdev->name, netdev);
1412 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1417 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1419 struct net_device *netdev = adapter->netdev;
1421 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1424 q_vectors = adapter->num_msix_vectors;
1427 free_irq(adapter->msix_entries[i].vector, netdev);
1430 for (; i >= 0; i--) {
1431 free_irq(adapter->msix_entries[i].vector,
1432 &(adapter->q_vector[i]));
1435 ixgbe_reset_q_vectors(adapter);
1437 free_irq(adapter->pdev->irq, netdev);
1442 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1445 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1447 struct ixgbe_hw *hw = &adapter->hw;
1449 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1450 EITR_INTS_PER_SEC_TO_REG(adapter->eitr_param));
1452 ixgbe_set_ivar(adapter, IXGBE_IVAR_RX_QUEUE(0), 0);
1453 ixgbe_set_ivar(adapter, IXGBE_IVAR_TX_QUEUE(0), 0);
1455 map_vector_to_rxq(adapter, 0, 0);
1456 map_vector_to_txq(adapter, 0, 0);
1458 DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1462 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1463 * @adapter: board private structure
1465 * Configure the Tx unit of the MAC after a reset.
1467 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1470 struct ixgbe_hw *hw = &adapter->hw;
1471 u32 i, j, tdlen, txctrl;
1473 /* Setup the HW Tx Head and Tail descriptor pointers */
1474 for (i = 0; i < adapter->num_tx_queues; i++) {
1475 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1478 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1479 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1480 (tdba & DMA_32BIT_MASK));
1481 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1483 (ring->count * sizeof(union ixgbe_adv_tx_desc));
1484 tdwba |= IXGBE_TDWBAL_HEAD_WB_ENABLE;
1485 IXGBE_WRITE_REG(hw, IXGBE_TDWBAL(j), tdwba & DMA_32BIT_MASK);
1486 IXGBE_WRITE_REG(hw, IXGBE_TDWBAH(j), (tdwba >> 32));
1487 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1488 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1489 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1490 adapter->tx_ring[i].head = IXGBE_TDH(j);
1491 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1492 /* Disable Tx Head Writeback RO bit, since this hoses
1493 * bookkeeping if things aren't delivered in order.
1495 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1496 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1497 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1501 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1503 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter, int index)
1505 struct ixgbe_ring *rx_ring;
1510 /* program one srrctl register per VMDq index */
1511 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
1513 mask = (unsigned long) adapter->ring_feature[RING_F_RSS].mask;
1514 len = sizeof(adapter->ring_feature[RING_F_VMDQ].mask) * 8;
1515 shift = find_first_bit(&mask, len);
1516 queue0 = index & mask;
1517 index = (index & mask) >> shift;
1518 /* program one srrctl per RSS queue since RDRXCTL.MVMEN is enabled */
1520 mask = (unsigned long) adapter->ring_feature[RING_F_RSS].mask;
1521 queue0 = index & mask;
1522 index = index & mask;
1525 rx_ring = &adapter->rx_ring[queue0];
1527 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
1529 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1530 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1532 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1533 srrctl |= IXGBE_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1534 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1535 srrctl |= ((IXGBE_RX_HDR_SIZE <<
1536 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1537 IXGBE_SRRCTL_BSIZEHDR_MASK);
1539 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1541 if (rx_ring->rx_buf_len == MAXIMUM_ETHERNET_VLAN_SIZE)
1542 srrctl |= IXGBE_RXBUFFER_2048 >>
1543 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1545 srrctl |= rx_ring->rx_buf_len >>
1546 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1548 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
1552 * ixgbe_get_skb_hdr - helper function for LRO header processing
1553 * @skb: pointer to sk_buff to be added to LRO packet
1554 * @iphdr: pointer to ip header structure
1555 * @tcph: pointer to tcp header structure
1556 * @hdr_flags: pointer to header flags
1557 * @priv: private data
1559 static int ixgbe_get_skb_hdr(struct sk_buff *skb, void **iphdr, void **tcph,
1560 u64 *hdr_flags, void *priv)
1562 union ixgbe_adv_rx_desc *rx_desc = priv;
1564 /* Verify that this is a valid IPv4 TCP packet */
1565 if (!((ixgbe_get_pkt_info(rx_desc) & IXGBE_RXDADV_PKTTYPE_IPV4) &&
1566 (ixgbe_get_pkt_info(rx_desc) & IXGBE_RXDADV_PKTTYPE_TCP)))
1569 /* Set network headers */
1570 skb_reset_network_header(skb);
1571 skb_set_transport_header(skb, ip_hdrlen(skb));
1572 *iphdr = ip_hdr(skb);
1573 *tcph = tcp_hdr(skb);
1574 *hdr_flags = LRO_IPV4 | LRO_TCP;
1578 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1579 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1582 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
1583 * @adapter: board private structure
1585 * Configure the Rx unit of the MAC after a reset.
1587 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
1590 struct ixgbe_hw *hw = &adapter->hw;
1591 struct net_device *netdev = adapter->netdev;
1592 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1594 u32 rdlen, rxctrl, rxcsum;
1595 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
1596 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
1597 0x6A3E67EA, 0x14364D17, 0x3BED200D};
1604 /* Decide whether to use packet split mode or not */
1605 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
1607 /* Set the RX buffer length according to the mode */
1608 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1609 rx_buf_len = IXGBE_RX_HDR_SIZE;
1611 if (netdev->mtu <= ETH_DATA_LEN)
1612 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1614 rx_buf_len = ALIGN(max_frame, 1024);
1617 fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
1618 fctrl |= IXGBE_FCTRL_BAM;
1619 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
1620 IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
1622 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
1623 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1624 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
1626 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
1627 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
1629 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1631 rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1632 /* disable receives while setting up the descriptors */
1633 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
1634 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
1636 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1637 * the Base and Length of the Rx Descriptor Ring */
1638 for (i = 0; i < adapter->num_rx_queues; i++) {
1639 rdba = adapter->rx_ring[i].dma;
1640 j = adapter->rx_ring[i].reg_idx;
1641 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_32BIT_MASK));
1642 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
1643 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
1644 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
1645 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
1646 adapter->rx_ring[i].head = IXGBE_RDH(j);
1647 adapter->rx_ring[i].tail = IXGBE_RDT(j);
1648 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1649 /* Intitial LRO Settings */
1650 adapter->rx_ring[i].lro_mgr.max_aggr = IXGBE_MAX_LRO_AGGREGATE;
1651 adapter->rx_ring[i].lro_mgr.max_desc = IXGBE_MAX_LRO_DESCRIPTORS;
1652 adapter->rx_ring[i].lro_mgr.get_skb_header = ixgbe_get_skb_hdr;
1653 adapter->rx_ring[i].lro_mgr.features = LRO_F_EXTRACT_VLAN_ID;
1654 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1655 adapter->rx_ring[i].lro_mgr.features |= LRO_F_NAPI;
1656 adapter->rx_ring[i].lro_mgr.dev = adapter->netdev;
1657 adapter->rx_ring[i].lro_mgr.ip_summed = CHECKSUM_UNNECESSARY;
1658 adapter->rx_ring[i].lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1660 ixgbe_configure_srrctl(adapter, j);
1664 * For VMDq support of different descriptor types or
1665 * buffer sizes through the use of multiple SRRCTL
1666 * registers, RDRXCTL.MVMEN must be set to 1
1668 * also, the manual doesn't mention it clearly but DCA hints
1669 * will only use queue 0's tags unless this bit is set. Side
1670 * effects of setting this bit are only that SRRCTL must be
1671 * fully programmed [0..15]
1673 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
1674 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
1675 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
1678 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
1679 /* Fill out redirection table */
1680 for (i = 0, j = 0; i < 128; i++, j++) {
1681 if (j == adapter->ring_feature[RING_F_RSS].indices)
1683 /* reta = 4-byte sliding window of
1684 * 0x00..(indices-1)(indices-1)00..etc. */
1685 reta = (reta << 8) | (j * 0x11);
1687 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
1690 /* Fill out hash function seeds */
1691 for (i = 0; i < 10; i++)
1692 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
1694 mrqc = IXGBE_MRQC_RSSEN
1695 /* Perform hash on these packet types */
1696 | IXGBE_MRQC_RSS_FIELD_IPV4
1697 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
1698 | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
1699 | IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP
1700 | IXGBE_MRQC_RSS_FIELD_IPV6_EX
1701 | IXGBE_MRQC_RSS_FIELD_IPV6
1702 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
1703 | IXGBE_MRQC_RSS_FIELD_IPV6_UDP
1704 | IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
1705 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
1708 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
1710 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
1711 adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
1712 /* Disable indicating checksum in descriptor, enables
1714 rxcsum |= IXGBE_RXCSUM_PCSD;
1716 if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
1717 /* Enable IPv4 payload checksum for UDP fragments
1718 * if PCSD is not set */
1719 rxcsum |= IXGBE_RXCSUM_IPPCSE;
1722 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
1725 static void ixgbe_vlan_rx_register(struct net_device *netdev,
1726 struct vlan_group *grp)
1728 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1731 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1732 ixgbe_irq_disable(adapter);
1733 adapter->vlgrp = grp;
1736 /* enable VLAN tag insert/strip */
1737 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
1738 ctrl |= IXGBE_VLNCTRL_VME;
1739 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
1740 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
1743 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1744 ixgbe_irq_enable(adapter);
1747 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1749 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1750 struct ixgbe_hw *hw = &adapter->hw;
1752 /* add VID to filter table */
1753 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
1756 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1758 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1759 struct ixgbe_hw *hw = &adapter->hw;
1761 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1762 ixgbe_irq_disable(adapter);
1764 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1766 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1767 ixgbe_irq_enable(adapter);
1769 /* remove VID from filter table */
1770 hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
1773 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
1775 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1777 if (adapter->vlgrp) {
1779 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1780 if (!vlan_group_get_device(adapter->vlgrp, vid))
1782 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
1787 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
1789 struct dev_mc_list *mc_ptr;
1790 u8 *addr = *mc_addr_ptr;
1793 mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
1795 *mc_addr_ptr = mc_ptr->next->dmi_addr;
1797 *mc_addr_ptr = NULL;
1803 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
1804 * @netdev: network interface device structure
1806 * The set_rx_method entry point is called whenever the unicast/multicast
1807 * address list or the network interface flags are updated. This routine is
1808 * responsible for configuring the hardware for proper unicast, multicast and
1811 static void ixgbe_set_rx_mode(struct net_device *netdev)
1813 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1814 struct ixgbe_hw *hw = &adapter->hw;
1816 u8 *addr_list = NULL;
1819 /* Check for Promiscuous and All Multicast modes */
1821 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
1822 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
1824 if (netdev->flags & IFF_PROMISC) {
1825 hw->addr_ctrl.user_set_promisc = 1;
1826 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
1827 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
1829 if (netdev->flags & IFF_ALLMULTI) {
1830 fctrl |= IXGBE_FCTRL_MPE;
1831 fctrl &= ~IXGBE_FCTRL_UPE;
1833 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
1835 vlnctrl |= IXGBE_VLNCTRL_VFE;
1836 hw->addr_ctrl.user_set_promisc = 0;
1839 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
1840 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
1842 /* reprogram secondary unicast list */
1843 addr_count = netdev->uc_count;
1845 addr_list = netdev->uc_list->dmi_addr;
1846 hw->mac.ops.update_uc_addr_list(hw, addr_list, addr_count,
1847 ixgbe_addr_list_itr);
1849 /* reprogram multicast list */
1850 addr_count = netdev->mc_count;
1852 addr_list = netdev->mc_list->dmi_addr;
1853 hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
1854 ixgbe_addr_list_itr);
1857 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
1860 struct ixgbe_q_vector *q_vector;
1861 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1863 /* legacy and MSI only use one vector */
1864 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1867 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1868 struct napi_struct *napi;
1869 q_vector = &adapter->q_vector[q_idx];
1870 if (!q_vector->rxr_count)
1872 napi = &q_vector->napi;
1873 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) &&
1874 (q_vector->rxr_count > 1))
1875 napi->poll = &ixgbe_clean_rxonly_many;
1881 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
1884 struct ixgbe_q_vector *q_vector;
1885 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1887 /* legacy and MSI only use one vector */
1888 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1891 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1892 q_vector = &adapter->q_vector[q_idx];
1893 if (!q_vector->rxr_count)
1895 napi_disable(&q_vector->napi);
1899 static void ixgbe_configure(struct ixgbe_adapter *adapter)
1901 struct net_device *netdev = adapter->netdev;
1904 ixgbe_set_rx_mode(netdev);
1906 ixgbe_restore_vlan(adapter);
1908 ixgbe_configure_tx(adapter);
1909 ixgbe_configure_rx(adapter);
1910 for (i = 0; i < adapter->num_rx_queues; i++)
1911 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
1912 (adapter->rx_ring[i].count - 1));
1915 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
1917 struct net_device *netdev = adapter->netdev;
1918 struct ixgbe_hw *hw = &adapter->hw;
1920 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1921 u32 txdctl, rxdctl, mhadd;
1924 ixgbe_get_hw_control(adapter);
1926 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
1927 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
1928 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1929 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
1930 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
1935 /* XXX: to interrupt immediately for EICS writes, enable this */
1936 /* gpie |= IXGBE_GPIE_EIMEN; */
1937 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
1940 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
1941 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
1942 * specifically only auto mask tx and rx interrupts */
1943 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
1946 /* Enable fan failure interrupt if media type is copper */
1947 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
1948 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
1949 gpie |= IXGBE_SDP1_GPIEN;
1950 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
1953 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
1954 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
1955 mhadd &= ~IXGBE_MHADD_MFS_MASK;
1956 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
1958 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
1961 for (i = 0; i < adapter->num_tx_queues; i++) {
1962 j = adapter->tx_ring[i].reg_idx;
1963 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
1964 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
1965 txdctl |= (8 << 16);
1966 txdctl |= IXGBE_TXDCTL_ENABLE;
1967 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
1970 for (i = 0; i < adapter->num_rx_queues; i++) {
1971 j = adapter->rx_ring[i].reg_idx;
1972 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
1973 /* enable PTHRESH=32 descriptors (half the internal cache)
1974 * and HTHRESH=0 descriptors (to minimize latency on fetch),
1975 * this also removes a pesky rx_no_buffer_count increment */
1977 rxdctl |= IXGBE_RXDCTL_ENABLE;
1978 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
1980 /* enable all receives */
1981 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
1982 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
1983 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxdctl);
1985 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
1986 ixgbe_configure_msix(adapter);
1988 ixgbe_configure_msi_and_legacy(adapter);
1990 clear_bit(__IXGBE_DOWN, &adapter->state);
1991 ixgbe_napi_enable_all(adapter);
1993 /* clear any pending interrupts, may auto mask */
1994 IXGBE_READ_REG(hw, IXGBE_EICR);
1996 ixgbe_irq_enable(adapter);
1998 /* enable transmits */
1999 netif_tx_start_all_queues(netdev);
2001 /* bring the link up in the watchdog, this could race with our first
2002 * link up interrupt but shouldn't be a problem */
2003 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2004 adapter->link_check_timeout = jiffies;
2005 mod_timer(&adapter->watchdog_timer, jiffies);
2009 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
2011 WARN_ON(in_interrupt());
2012 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
2014 ixgbe_down(adapter);
2016 clear_bit(__IXGBE_RESETTING, &adapter->state);
2019 int ixgbe_up(struct ixgbe_adapter *adapter)
2021 /* hardware has been reset, we need to reload some things */
2022 ixgbe_configure(adapter);
2024 return ixgbe_up_complete(adapter);
2027 void ixgbe_reset(struct ixgbe_adapter *adapter)
2029 struct ixgbe_hw *hw = &adapter->hw;
2030 if (hw->mac.ops.init_hw(hw))
2031 dev_err(&adapter->pdev->dev, "Hardware Error\n");
2033 /* reprogram the RAR[0] in case user changed it. */
2034 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2039 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2040 * @adapter: board private structure
2041 * @rx_ring: ring to free buffers from
2043 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2044 struct ixgbe_ring *rx_ring)
2046 struct pci_dev *pdev = adapter->pdev;
2050 /* Free all the Rx ring sk_buffs */
2052 for (i = 0; i < rx_ring->count; i++) {
2053 struct ixgbe_rx_buffer *rx_buffer_info;
2055 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2056 if (rx_buffer_info->dma) {
2057 pci_unmap_single(pdev, rx_buffer_info->dma,
2058 rx_ring->rx_buf_len,
2059 PCI_DMA_FROMDEVICE);
2060 rx_buffer_info->dma = 0;
2062 if (rx_buffer_info->skb) {
2063 dev_kfree_skb(rx_buffer_info->skb);
2064 rx_buffer_info->skb = NULL;
2066 if (!rx_buffer_info->page)
2068 pci_unmap_page(pdev, rx_buffer_info->page_dma, PAGE_SIZE / 2,
2069 PCI_DMA_FROMDEVICE);
2070 rx_buffer_info->page_dma = 0;
2071 put_page(rx_buffer_info->page);
2072 rx_buffer_info->page = NULL;
2073 rx_buffer_info->page_offset = 0;
2076 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2077 memset(rx_ring->rx_buffer_info, 0, size);
2079 /* Zero out the descriptor ring */
2080 memset(rx_ring->desc, 0, rx_ring->size);
2082 rx_ring->next_to_clean = 0;
2083 rx_ring->next_to_use = 0;
2085 writel(0, adapter->hw.hw_addr + rx_ring->head);
2086 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2090 * ixgbe_clean_tx_ring - Free Tx Buffers
2091 * @adapter: board private structure
2092 * @tx_ring: ring to be cleaned
2094 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2095 struct ixgbe_ring *tx_ring)
2097 struct ixgbe_tx_buffer *tx_buffer_info;
2101 /* Free all the Tx ring sk_buffs */
2103 for (i = 0; i < tx_ring->count; i++) {
2104 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2105 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2108 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2109 memset(tx_ring->tx_buffer_info, 0, size);
2111 /* Zero out the descriptor ring */
2112 memset(tx_ring->desc, 0, tx_ring->size);
2114 tx_ring->next_to_use = 0;
2115 tx_ring->next_to_clean = 0;
2117 writel(0, adapter->hw.hw_addr + tx_ring->head);
2118 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2122 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2123 * @adapter: board private structure
2125 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2129 for (i = 0; i < adapter->num_rx_queues; i++)
2130 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2134 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2135 * @adapter: board private structure
2137 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2141 for (i = 0; i < adapter->num_tx_queues; i++)
2142 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2145 void ixgbe_down(struct ixgbe_adapter *adapter)
2147 struct net_device *netdev = adapter->netdev;
2148 struct ixgbe_hw *hw = &adapter->hw;
2153 /* signal that we are down to the interrupt handler */
2154 set_bit(__IXGBE_DOWN, &adapter->state);
2156 /* disable receives */
2157 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2158 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2160 netif_tx_disable(netdev);
2162 IXGBE_WRITE_FLUSH(hw);
2165 netif_tx_stop_all_queues(netdev);
2167 ixgbe_irq_disable(adapter);
2169 ixgbe_napi_disable_all(adapter);
2171 del_timer_sync(&adapter->watchdog_timer);
2172 cancel_work_sync(&adapter->watchdog_task);
2174 /* disable transmits in the hardware now that interrupts are off */
2175 for (i = 0; i < adapter->num_tx_queues; i++) {
2176 j = adapter->tx_ring[i].reg_idx;
2177 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2178 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
2179 (txdctl & ~IXGBE_TXDCTL_ENABLE));
2182 netif_carrier_off(netdev);
2184 #ifdef CONFIG_IXGBE_DCA
2185 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2186 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
2187 dca_remove_requester(&adapter->pdev->dev);
2191 if (!pci_channel_offline(adapter->pdev))
2192 ixgbe_reset(adapter);
2193 ixgbe_clean_all_tx_rings(adapter);
2194 ixgbe_clean_all_rx_rings(adapter);
2196 #ifdef CONFIG_IXGBE_DCA
2197 /* since we reset the hardware DCA settings were cleared */
2198 if (dca_add_requester(&adapter->pdev->dev) == 0) {
2199 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
2200 /* always use CB2 mode, difference is masked
2201 * in the CB driver */
2202 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
2203 ixgbe_setup_dca(adapter);
2209 * ixgbe_poll - NAPI Rx polling callback
2210 * @napi: structure for representing this polling device
2211 * @budget: how many packets driver is allowed to clean
2213 * This function is used for legacy and MSI, NAPI mode
2215 static int ixgbe_poll(struct napi_struct *napi, int budget)
2217 struct ixgbe_q_vector *q_vector = container_of(napi,
2218 struct ixgbe_q_vector, napi);
2219 struct ixgbe_adapter *adapter = q_vector->adapter;
2220 int tx_cleaned, work_done = 0;
2222 #ifdef CONFIG_IXGBE_DCA
2223 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2224 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
2225 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
2229 tx_cleaned = ixgbe_clean_tx_irq(adapter, adapter->tx_ring);
2230 ixgbe_clean_rx_irq(adapter, adapter->rx_ring, &work_done, budget);
2235 /* If budget not fully consumed, exit the polling mode */
2236 if (work_done < budget) {
2237 netif_rx_complete(adapter->netdev, napi);
2238 if (adapter->itr_setting & 3)
2239 ixgbe_set_itr(adapter);
2240 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2241 ixgbe_irq_enable(adapter);
2247 * ixgbe_tx_timeout - Respond to a Tx Hang
2248 * @netdev: network interface device structure
2250 static void ixgbe_tx_timeout(struct net_device *netdev)
2252 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2254 /* Do the reset outside of interrupt context */
2255 schedule_work(&adapter->reset_task);
2258 static void ixgbe_reset_task(struct work_struct *work)
2260 struct ixgbe_adapter *adapter;
2261 adapter = container_of(work, struct ixgbe_adapter, reset_task);
2263 adapter->tx_timeout_count++;
2265 ixgbe_reinit_locked(adapter);
2268 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
2270 int nrq = 1, ntq = 1;
2271 int feature_mask = 0, rss_i, rss_m;
2273 /* Number of supported queues */
2274 switch (adapter->hw.mac.type) {
2275 case ixgbe_mac_82598EB:
2276 rss_i = adapter->ring_feature[RING_F_RSS].indices;
2278 feature_mask |= IXGBE_FLAG_RSS_ENABLED;
2280 switch (adapter->flags & feature_mask) {
2281 case (IXGBE_FLAG_RSS_ENABLED):
2295 adapter->ring_feature[RING_F_RSS].indices = rss_i;
2296 adapter->ring_feature[RING_F_RSS].mask = rss_m;
2304 adapter->num_rx_queues = nrq;
2305 adapter->num_tx_queues = ntq;
2308 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
2311 int err, vector_threshold;
2313 /* We'll want at least 3 (vector_threshold):
2316 * 3) Other (Link Status Change, etc.)
2317 * 4) TCP Timer (optional)
2319 vector_threshold = MIN_MSIX_COUNT;
2321 /* The more we get, the more we will assign to Tx/Rx Cleanup
2322 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2323 * Right now, we simply care about how many we'll get; we'll
2324 * set them up later while requesting irq's.
2326 while (vectors >= vector_threshold) {
2327 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
2329 if (!err) /* Success in acquiring all requested vectors. */
2332 vectors = 0; /* Nasty failure, quit now */
2333 else /* err == number of vectors we should try again with */
2337 if (vectors < vector_threshold) {
2338 /* Can't allocate enough MSI-X interrupts? Oh well.
2339 * This just means we'll go with either a single MSI
2340 * vector or fall back to legacy interrupts.
2342 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
2343 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2344 kfree(adapter->msix_entries);
2345 adapter->msix_entries = NULL;
2346 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2347 ixgbe_set_num_queues(adapter);
2349 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
2350 adapter->num_msix_vectors = vectors;
2355 * ixgbe_cache_ring_register - Descriptor ring to register mapping
2356 * @adapter: board private structure to initialize
2358 * Once we know the feature-set enabled for the device, we'll cache
2359 * the register offset the descriptor ring is assigned to.
2361 static void __devinit ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
2363 int feature_mask = 0, rss_i;
2364 int i, txr_idx, rxr_idx;
2366 /* Number of supported queues */
2367 switch (adapter->hw.mac.type) {
2368 case ixgbe_mac_82598EB:
2369 rss_i = adapter->ring_feature[RING_F_RSS].indices;
2372 feature_mask |= IXGBE_FLAG_RSS_ENABLED;
2373 switch (adapter->flags & feature_mask) {
2374 case (IXGBE_FLAG_RSS_ENABLED):
2375 for (i = 0; i < adapter->num_rx_queues; i++)
2376 adapter->rx_ring[i].reg_idx = i;
2377 for (i = 0; i < adapter->num_tx_queues; i++)
2378 adapter->tx_ring[i].reg_idx = i;
2391 * ixgbe_alloc_queues - Allocate memory for all rings
2392 * @adapter: board private structure to initialize
2394 * We allocate one ring per queue at run-time since we don't know the
2395 * number of queues at compile-time. The polling_netdev array is
2396 * intended for Multiqueue, but should work fine with a single queue.
2398 static int __devinit ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
2402 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
2403 sizeof(struct ixgbe_ring), GFP_KERNEL);
2404 if (!adapter->tx_ring)
2405 goto err_tx_ring_allocation;
2407 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
2408 sizeof(struct ixgbe_ring), GFP_KERNEL);
2409 if (!adapter->rx_ring)
2410 goto err_rx_ring_allocation;
2412 for (i = 0; i < adapter->num_tx_queues; i++) {
2413 adapter->tx_ring[i].count = adapter->tx_ring_count;
2414 adapter->tx_ring[i].queue_index = i;
2417 for (i = 0; i < adapter->num_rx_queues; i++) {
2418 adapter->rx_ring[i].count = adapter->rx_ring_count;
2419 adapter->rx_ring[i].queue_index = i;
2422 ixgbe_cache_ring_register(adapter);
2426 err_rx_ring_allocation:
2427 kfree(adapter->tx_ring);
2428 err_tx_ring_allocation:
2433 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
2434 * @adapter: board private structure to initialize
2436 * Attempt to configure the interrupts using the best available
2437 * capabilities of the hardware and the kernel.
2439 static int __devinit ixgbe_set_interrupt_capability(struct ixgbe_adapter
2443 int vector, v_budget;
2446 * It's easy to be greedy for MSI-X vectors, but it really
2447 * doesn't do us much good if we have a lot more vectors
2448 * than CPU's. So let's be conservative and only ask for
2449 * (roughly) twice the number of vectors as there are CPU's.
2451 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
2452 (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
2455 * At the same time, hardware can only support a maximum of
2456 * MAX_MSIX_COUNT vectors. With features such as RSS and VMDq,
2457 * we can easily reach upwards of 64 Rx descriptor queues and
2458 * 32 Tx queues. Thus, we cap it off in those rare cases where
2459 * the cpu count also exceeds our vector limit.
2461 v_budget = min(v_budget, MAX_MSIX_COUNT);
2463 /* A failure in MSI-X entry allocation isn't fatal, but it does
2464 * mean we disable MSI-X capabilities of the adapter. */
2465 adapter->msix_entries = kcalloc(v_budget,
2466 sizeof(struct msix_entry), GFP_KERNEL);
2467 if (!adapter->msix_entries) {
2468 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2469 ixgbe_set_num_queues(adapter);
2470 kfree(adapter->tx_ring);
2471 kfree(adapter->rx_ring);
2472 err = ixgbe_alloc_queues(adapter);
2474 DPRINTK(PROBE, ERR, "Unable to allocate memory "
2482 for (vector = 0; vector < v_budget; vector++)
2483 adapter->msix_entries[vector].entry = vector;
2485 ixgbe_acquire_msix_vectors(adapter, v_budget);
2487 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2491 err = pci_enable_msi(adapter->pdev);
2493 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
2495 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
2496 "falling back to legacy. Error: %d\n", err);
2502 /* Notify the stack of the (possibly) reduced Tx Queue count. */
2503 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
2508 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
2510 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2511 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2512 pci_disable_msix(adapter->pdev);
2513 kfree(adapter->msix_entries);
2514 adapter->msix_entries = NULL;
2515 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2516 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
2517 pci_disable_msi(adapter->pdev);
2523 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
2524 * @adapter: board private structure to initialize
2526 * We determine which interrupt scheme to use based on...
2527 * - Kernel support (MSI, MSI-X)
2528 * - which can be user-defined (via MODULE_PARAM)
2529 * - Hardware queue count (num_*_queues)
2530 * - defined by miscellaneous hardware support/features (RSS, etc.)
2532 static int __devinit ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
2536 /* Number of supported queues */
2537 ixgbe_set_num_queues(adapter);
2539 err = ixgbe_alloc_queues(adapter);
2541 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
2542 goto err_alloc_queues;
2545 err = ixgbe_set_interrupt_capability(adapter);
2547 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
2548 goto err_set_interrupt;
2551 DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
2552 "Tx Queue count = %u\n",
2553 (adapter->num_rx_queues > 1) ? "Enabled" :
2554 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2556 set_bit(__IXGBE_DOWN, &adapter->state);
2561 kfree(adapter->tx_ring);
2562 kfree(adapter->rx_ring);
2568 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
2569 * @adapter: board private structure to initialize
2571 * ixgbe_sw_init initializes the Adapter private data structure.
2572 * Fields are initialized based on PCI device information and
2573 * OS network device settings (MTU size).
2575 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
2577 struct ixgbe_hw *hw = &adapter->hw;
2578 struct pci_dev *pdev = adapter->pdev;
2581 /* PCI config space info */
2583 hw->vendor_id = pdev->vendor;
2584 hw->device_id = pdev->device;
2585 hw->revision_id = pdev->revision;
2586 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2587 hw->subsystem_device_id = pdev->subsystem_device;
2589 /* Set capability flags */
2590 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
2591 adapter->ring_feature[RING_F_RSS].indices = rss;
2592 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
2593 if (hw->mac.ops.get_media_type &&
2594 (hw->mac.ops.get_media_type(hw) == ixgbe_media_type_copper))
2595 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
2597 /* default flow control settings */
2598 hw->fc.original_type = ixgbe_fc_none;
2599 hw->fc.type = ixgbe_fc_none;
2600 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
2601 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
2602 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
2603 hw->fc.send_xon = true;
2605 /* select 10G link by default */
2606 hw->mac.link_mode_select = IXGBE_AUTOC_LMS_10G_LINK_NO_AN;
2608 /* enable itr by default in dynamic mode */
2609 adapter->itr_setting = 1;
2610 adapter->eitr_param = 20000;
2612 /* set defaults for eitr in MegaBytes */
2613 adapter->eitr_low = 10;
2614 adapter->eitr_high = 20;
2616 /* set default ring sizes */
2617 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
2618 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
2620 /* initialize eeprom parameters */
2621 if (ixgbe_init_eeprom_params_generic(hw)) {
2622 dev_err(&pdev->dev, "EEPROM initialization failed\n");
2626 /* enable rx csum by default */
2627 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
2629 set_bit(__IXGBE_DOWN, &adapter->state);
2635 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
2636 * @adapter: board private structure
2637 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2639 * Return 0 on success, negative on failure
2641 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
2642 struct ixgbe_ring *tx_ring)
2644 struct pci_dev *pdev = adapter->pdev;
2647 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2648 tx_ring->tx_buffer_info = vmalloc(size);
2649 if (!tx_ring->tx_buffer_info)
2651 memset(tx_ring->tx_buffer_info, 0, size);
2653 /* round up to nearest 4K */
2654 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc) +
2656 tx_ring->size = ALIGN(tx_ring->size, 4096);
2658 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
2663 tx_ring->next_to_use = 0;
2664 tx_ring->next_to_clean = 0;
2665 tx_ring->work_limit = tx_ring->count;
2669 vfree(tx_ring->tx_buffer_info);
2670 tx_ring->tx_buffer_info = NULL;
2671 DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
2672 "descriptor ring\n");
2677 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
2678 * @adapter: board private structure
2680 * If this function returns with an error, then it's possible one or
2681 * more of the rings is populated (while the rest are not). It is the
2682 * callers duty to clean those orphaned rings.
2684 * Return 0 on success, negative on failure
2686 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
2690 for (i = 0; i < adapter->num_tx_queues; i++) {
2691 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2694 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
2702 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
2703 * @adapter: board private structure
2704 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2706 * Returns 0 on success, negative on failure
2708 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
2709 struct ixgbe_ring *rx_ring)
2711 struct pci_dev *pdev = adapter->pdev;
2714 size = sizeof(struct net_lro_desc) * IXGBE_MAX_LRO_DESCRIPTORS;
2715 rx_ring->lro_mgr.lro_arr = vmalloc(size);
2716 if (!rx_ring->lro_mgr.lro_arr)
2718 memset(rx_ring->lro_mgr.lro_arr, 0, size);
2720 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2721 rx_ring->rx_buffer_info = vmalloc(size);
2722 if (!rx_ring->rx_buffer_info) {
2724 "vmalloc allocation failed for the rx desc ring\n");
2727 memset(rx_ring->rx_buffer_info, 0, size);
2729 /* Round up to nearest 4K */
2730 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2731 rx_ring->size = ALIGN(rx_ring->size, 4096);
2733 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
2735 if (!rx_ring->desc) {
2737 "Memory allocation failed for the rx desc ring\n");
2738 vfree(rx_ring->rx_buffer_info);
2742 rx_ring->next_to_clean = 0;
2743 rx_ring->next_to_use = 0;
2748 vfree(rx_ring->lro_mgr.lro_arr);
2749 rx_ring->lro_mgr.lro_arr = NULL;
2754 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
2755 * @adapter: board private structure
2757 * If this function returns with an error, then it's possible one or
2758 * more of the rings is populated (while the rest are not). It is the
2759 * callers duty to clean those orphaned rings.
2761 * Return 0 on success, negative on failure
2764 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
2768 for (i = 0; i < adapter->num_rx_queues; i++) {
2769 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2772 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
2780 * ixgbe_free_tx_resources - Free Tx Resources per Queue
2781 * @adapter: board private structure
2782 * @tx_ring: Tx descriptor ring for a specific queue
2784 * Free all transmit software resources
2786 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
2787 struct ixgbe_ring *tx_ring)
2789 struct pci_dev *pdev = adapter->pdev;
2791 ixgbe_clean_tx_ring(adapter, tx_ring);
2793 vfree(tx_ring->tx_buffer_info);
2794 tx_ring->tx_buffer_info = NULL;
2796 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2798 tx_ring->desc = NULL;
2802 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
2803 * @adapter: board private structure
2805 * Free all transmit software resources
2807 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
2811 for (i = 0; i < adapter->num_tx_queues; i++)
2812 ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
2816 * ixgbe_free_rx_resources - Free Rx Resources
2817 * @adapter: board private structure
2818 * @rx_ring: ring to clean the resources from
2820 * Free all receive software resources
2822 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
2823 struct ixgbe_ring *rx_ring)
2825 struct pci_dev *pdev = adapter->pdev;
2827 vfree(rx_ring->lro_mgr.lro_arr);
2828 rx_ring->lro_mgr.lro_arr = NULL;
2830 ixgbe_clean_rx_ring(adapter, rx_ring);
2832 vfree(rx_ring->rx_buffer_info);
2833 rx_ring->rx_buffer_info = NULL;
2835 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2837 rx_ring->desc = NULL;
2841 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
2842 * @adapter: board private structure
2844 * Free all receive software resources
2846 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
2850 for (i = 0; i < adapter->num_rx_queues; i++)
2851 ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
2855 * ixgbe_change_mtu - Change the Maximum Transfer Unit
2856 * @netdev: network interface device structure
2857 * @new_mtu: new value for maximum frame size
2859 * Returns 0 on success, negative on failure
2861 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
2863 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2864 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2866 /* MTU < 68 is an error and causes problems on some kernels */
2867 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
2870 DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
2871 netdev->mtu, new_mtu);
2872 /* must set new MTU before calling down or up */
2873 netdev->mtu = new_mtu;
2875 if (netif_running(netdev))
2876 ixgbe_reinit_locked(adapter);
2882 * ixgbe_open - Called when a network interface is made active
2883 * @netdev: network interface device structure
2885 * Returns 0 on success, negative value on failure
2887 * The open entry point is called when a network interface is made
2888 * active by the system (IFF_UP). At this point all resources needed
2889 * for transmit and receive operations are allocated, the interrupt
2890 * handler is registered with the OS, the watchdog timer is started,
2891 * and the stack is notified that the interface is ready.
2893 static int ixgbe_open(struct net_device *netdev)
2895 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2898 /* disallow open during test */
2899 if (test_bit(__IXGBE_TESTING, &adapter->state))
2902 /* allocate transmit descriptors */
2903 err = ixgbe_setup_all_tx_resources(adapter);
2907 /* allocate receive descriptors */
2908 err = ixgbe_setup_all_rx_resources(adapter);
2912 ixgbe_configure(adapter);
2914 err = ixgbe_request_irq(adapter);
2918 err = ixgbe_up_complete(adapter);
2922 netif_tx_start_all_queues(netdev);
2927 ixgbe_release_hw_control(adapter);
2928 ixgbe_free_irq(adapter);
2930 ixgbe_free_all_rx_resources(adapter);
2932 ixgbe_free_all_tx_resources(adapter);
2934 ixgbe_reset(adapter);
2940 * ixgbe_close - Disables a network interface
2941 * @netdev: network interface device structure
2943 * Returns 0, this is not allowed to fail
2945 * The close entry point is called when an interface is de-activated
2946 * by the OS. The hardware is still under the drivers control, but
2947 * needs to be disabled. A global MAC reset is issued to stop the
2948 * hardware, and all transmit and receive resources are freed.
2950 static int ixgbe_close(struct net_device *netdev)
2952 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2954 ixgbe_down(adapter);
2955 ixgbe_free_irq(adapter);
2957 ixgbe_free_all_tx_resources(adapter);
2958 ixgbe_free_all_rx_resources(adapter);
2960 ixgbe_release_hw_control(adapter);
2966 * ixgbe_napi_add_all - prep napi structs for use
2967 * @adapter: private struct
2968 * helper function to napi_add each possible q_vector->napi
2970 static void ixgbe_napi_add_all(struct ixgbe_adapter *adapter)
2972 int q_idx, q_vectors;
2973 int (*poll)(struct napi_struct *, int);
2975 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2976 poll = &ixgbe_clean_rxonly;
2977 /* Only enable as many vectors as we have rx queues. */
2978 q_vectors = adapter->num_rx_queues;
2981 /* only one q_vector for legacy modes */
2985 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2986 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
2987 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
2991 static void ixgbe_napi_del_all(struct ixgbe_adapter *adapter)
2994 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2996 /* legacy and MSI only use one vector */
2997 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3000 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3001 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
3002 if (!q_vector->rxr_count)
3004 netif_napi_del(&q_vector->napi);
3009 static int ixgbe_resume(struct pci_dev *pdev)
3011 struct net_device *netdev = pci_get_drvdata(pdev);
3012 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3015 pci_set_power_state(pdev, PCI_D0);
3016 pci_restore_state(pdev);
3017 err = pci_enable_device(pdev);
3019 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
3023 pci_set_master(pdev);
3025 pci_enable_wake(pdev, PCI_D3hot, 0);
3026 pci_enable_wake(pdev, PCI_D3cold, 0);
3028 err = ixgbe_init_interrupt_scheme(adapter);
3030 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
3035 ixgbe_napi_add_all(adapter);
3036 ixgbe_reset(adapter);
3038 if (netif_running(netdev)) {
3039 err = ixgbe_open(adapter->netdev);
3044 netif_device_attach(netdev);
3049 #endif /* CONFIG_PM */
3050 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
3052 struct net_device *netdev = pci_get_drvdata(pdev);
3053 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3058 netif_device_detach(netdev);
3060 if (netif_running(netdev)) {
3061 ixgbe_down(adapter);
3062 ixgbe_free_irq(adapter);
3063 ixgbe_free_all_tx_resources(adapter);
3064 ixgbe_free_all_rx_resources(adapter);
3066 ixgbe_reset_interrupt_capability(adapter);
3067 ixgbe_napi_del_all(adapter);
3068 kfree(adapter->tx_ring);
3069 kfree(adapter->rx_ring);
3072 retval = pci_save_state(pdev);
3077 pci_enable_wake(pdev, PCI_D3hot, 0);
3078 pci_enable_wake(pdev, PCI_D3cold, 0);
3080 ixgbe_release_hw_control(adapter);
3082 pci_disable_device(pdev);
3084 pci_set_power_state(pdev, pci_choose_state(pdev, state));
3089 static void ixgbe_shutdown(struct pci_dev *pdev)
3091 ixgbe_suspend(pdev, PMSG_SUSPEND);
3095 * ixgbe_update_stats - Update the board statistics counters.
3096 * @adapter: board private structure
3098 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
3100 struct ixgbe_hw *hw = &adapter->hw;
3102 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
3104 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
3105 for (i = 0; i < 8; i++) {
3106 /* for packet buffers not used, the register should read 0 */
3107 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
3109 adapter->stats.mpc[i] += mpc;
3110 total_mpc += adapter->stats.mpc[i];
3111 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
3113 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
3114 /* work around hardware counting issue */
3115 adapter->stats.gprc -= missed_rx;
3117 /* 82598 hardware only has a 32 bit counter in the high register */
3118 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
3119 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
3120 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
3121 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
3122 adapter->stats.bprc += bprc;
3123 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
3124 adapter->stats.mprc -= bprc;
3125 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
3126 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
3127 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
3128 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
3129 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
3130 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
3131 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
3132 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
3133 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
3134 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
3135 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
3136 adapter->stats.lxontxc += lxon;
3137 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
3138 adapter->stats.lxofftxc += lxoff;
3139 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3140 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
3141 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
3143 * 82598 errata - tx of flow control packets is included in tx counters
3145 xon_off_tot = lxon + lxoff;
3146 adapter->stats.gptc -= xon_off_tot;
3147 adapter->stats.mptc -= xon_off_tot;
3148 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
3149 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3150 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
3151 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
3152 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
3153 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
3154 adapter->stats.ptc64 -= xon_off_tot;
3155 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
3156 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
3157 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
3158 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
3159 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
3160 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
3162 /* Fill out the OS statistics structure */
3163 adapter->net_stats.multicast = adapter->stats.mprc;
3166 adapter->net_stats.rx_errors = adapter->stats.crcerrs +
3167 adapter->stats.rlec;
3168 adapter->net_stats.rx_dropped = 0;
3169 adapter->net_stats.rx_length_errors = adapter->stats.rlec;
3170 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3171 adapter->net_stats.rx_missed_errors = total_mpc;
3175 * ixgbe_watchdog - Timer Call-back
3176 * @data: pointer to adapter cast into an unsigned long
3178 static void ixgbe_watchdog(unsigned long data)
3180 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3181 struct ixgbe_hw *hw = &adapter->hw;
3183 /* Do the watchdog outside of interrupt context due to the lovely
3184 * delays that some of the newer hardware requires */
3185 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
3186 /* Cause software interrupt to ensure rx rings are cleaned */
3187 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3189 (1 << (adapter->num_msix_vectors - NON_Q_VECTORS)) - 1;
3190 IXGBE_WRITE_REG(hw, IXGBE_EICS, eics);
3192 /* For legacy and MSI interrupts don't set any bits that
3193 * are enabled for EIAM, because this operation would
3194 * set *both* EIMS and EICS for any bit in EIAM */
3195 IXGBE_WRITE_REG(hw, IXGBE_EICS,
3196 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
3198 /* Reset the timer */
3199 mod_timer(&adapter->watchdog_timer,
3200 round_jiffies(jiffies + 2 * HZ));
3203 schedule_work(&adapter->watchdog_task);
3207 * ixgbe_watchdog_task - worker thread to bring link up
3208 * @work: pointer to work_struct containing our data
3210 static void ixgbe_watchdog_task(struct work_struct *work)
3212 struct ixgbe_adapter *adapter = container_of(work,
3213 struct ixgbe_adapter,
3215 struct net_device *netdev = adapter->netdev;
3216 struct ixgbe_hw *hw = &adapter->hw;
3217 u32 link_speed = adapter->link_speed;
3218 bool link_up = adapter->link_up;
3220 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
3222 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
3223 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3225 time_after(jiffies, (adapter->link_check_timeout +
3226 IXGBE_TRY_LINK_TIMEOUT))) {
3227 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
3228 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
3230 adapter->link_up = link_up;
3231 adapter->link_speed = link_speed;
3235 if (!netif_carrier_ok(netdev)) {
3236 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3237 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
3238 #define FLOW_RX (frctl & IXGBE_FCTRL_RFCE)
3239 #define FLOW_TX (rmcs & IXGBE_RMCS_TFCE_802_3X)
3240 DPRINTK(LINK, INFO, "NIC Link is Up %s, "
3241 "Flow Control: %s\n",
3242 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
3244 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
3245 "1 Gbps" : "unknown speed")),
3246 ((FLOW_RX && FLOW_TX) ? "RX/TX" :
3248 (FLOW_TX ? "TX" : "None"))));
3250 netif_carrier_on(netdev);
3252 /* Force detection of hung controller */
3253 adapter->detect_tx_hung = true;
3256 adapter->link_up = false;
3257 adapter->link_speed = 0;
3258 if (netif_carrier_ok(netdev)) {
3259 DPRINTK(LINK, INFO, "NIC Link is Down\n");
3260 netif_carrier_off(netdev);
3264 ixgbe_update_stats(adapter);
3265 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
3268 static int ixgbe_tso(struct ixgbe_adapter *adapter,
3269 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
3270 u32 tx_flags, u8 *hdr_len)
3272 struct ixgbe_adv_tx_context_desc *context_desc;
3275 struct ixgbe_tx_buffer *tx_buffer_info;
3276 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
3277 u32 mss_l4len_idx, l4len;
3279 if (skb_is_gso(skb)) {
3280 if (skb_header_cloned(skb)) {
3281 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3285 l4len = tcp_hdrlen(skb);
3288 if (skb->protocol == htons(ETH_P_IP)) {
3289 struct iphdr *iph = ip_hdr(skb);
3292 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3296 adapter->hw_tso_ctxt++;
3297 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3298 ipv6_hdr(skb)->payload_len = 0;
3299 tcp_hdr(skb)->check =
3300 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3301 &ipv6_hdr(skb)->daddr,
3303 adapter->hw_tso6_ctxt++;
3306 i = tx_ring->next_to_use;
3308 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3309 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
3311 /* VLAN MACLEN IPLEN */
3312 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3314 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
3315 vlan_macip_lens |= ((skb_network_offset(skb)) <<
3316 IXGBE_ADVTXD_MACLEN_SHIFT);
3317 *hdr_len += skb_network_offset(skb);
3319 (skb_transport_header(skb) - skb_network_header(skb));
3321 (skb_transport_header(skb) - skb_network_header(skb));
3322 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3323 context_desc->seqnum_seed = 0;
3325 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3326 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
3327 IXGBE_ADVTXD_DTYP_CTXT);
3329 if (skb->protocol == htons(ETH_P_IP))
3330 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
3331 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3332 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
3336 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
3337 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
3338 /* use index 1 for TSO */
3339 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3340 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3342 tx_buffer_info->time_stamp = jiffies;
3343 tx_buffer_info->next_to_watch = i;
3346 if (i == tx_ring->count)
3348 tx_ring->next_to_use = i;
3355 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
3356 struct ixgbe_ring *tx_ring,
3357 struct sk_buff *skb, u32 tx_flags)
3359 struct ixgbe_adv_tx_context_desc *context_desc;
3361 struct ixgbe_tx_buffer *tx_buffer_info;
3362 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
3364 if (skb->ip_summed == CHECKSUM_PARTIAL ||
3365 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
3366 i = tx_ring->next_to_use;
3367 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3368 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
3370 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3372 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
3373 vlan_macip_lens |= (skb_network_offset(skb) <<
3374 IXGBE_ADVTXD_MACLEN_SHIFT);
3375 if (skb->ip_summed == CHECKSUM_PARTIAL)
3376 vlan_macip_lens |= (skb_transport_header(skb) -
3377 skb_network_header(skb));
3379 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3380 context_desc->seqnum_seed = 0;
3382 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
3383 IXGBE_ADVTXD_DTYP_CTXT);
3385 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3386 switch (skb->protocol) {
3387 case __constant_htons(ETH_P_IP):
3388 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
3389 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3391 IXGBE_ADVTXD_TUCMD_L4T_TCP;
3393 case __constant_htons(ETH_P_IPV6):
3394 /* XXX what about other V6 headers?? */
3395 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3397 IXGBE_ADVTXD_TUCMD_L4T_TCP;
3400 if (unlikely(net_ratelimit())) {
3401 DPRINTK(PROBE, WARNING,
3402 "partial checksum but proto=%x!\n",
3409 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
3410 /* use index zero for tx checksum offload */
3411 context_desc->mss_l4len_idx = 0;
3413 tx_buffer_info->time_stamp = jiffies;
3414 tx_buffer_info->next_to_watch = i;
3416 adapter->hw_csum_tx_good++;
3418 if (i == tx_ring->count)
3420 tx_ring->next_to_use = i;
3428 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
3429 struct ixgbe_ring *tx_ring,
3430 struct sk_buff *skb, unsigned int first)
3432 struct ixgbe_tx_buffer *tx_buffer_info;
3433 unsigned int len = skb->len;
3434 unsigned int offset = 0, size, count = 0, i;
3435 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
3438 len -= skb->data_len;
3440 i = tx_ring->next_to_use;
3443 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3444 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
3446 tx_buffer_info->length = size;
3447 tx_buffer_info->dma = pci_map_single(adapter->pdev,
3449 size, PCI_DMA_TODEVICE);
3450 tx_buffer_info->time_stamp = jiffies;
3451 tx_buffer_info->next_to_watch = i;
3457 if (i == tx_ring->count)
3461 for (f = 0; f < nr_frags; f++) {
3462 struct skb_frag_struct *frag;
3464 frag = &skb_shinfo(skb)->frags[f];
3466 offset = frag->page_offset;
3469 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3470 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
3472 tx_buffer_info->length = size;
3473 tx_buffer_info->dma = pci_map_page(adapter->pdev,
3478 tx_buffer_info->time_stamp = jiffies;
3479 tx_buffer_info->next_to_watch = i;
3485 if (i == tx_ring->count)
3490 i = tx_ring->count - 1;
3493 tx_ring->tx_buffer_info[i].skb = skb;
3494 tx_ring->tx_buffer_info[first].next_to_watch = i;
3499 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
3500 struct ixgbe_ring *tx_ring,
3501 int tx_flags, int count, u32 paylen, u8 hdr_len)
3503 union ixgbe_adv_tx_desc *tx_desc = NULL;
3504 struct ixgbe_tx_buffer *tx_buffer_info;
3505 u32 olinfo_status = 0, cmd_type_len = 0;
3507 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
3509 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
3511 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
3513 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3514 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
3516 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
3517 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
3519 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
3520 IXGBE_ADVTXD_POPTS_SHIFT;
3522 /* use index 1 context for tso */
3523 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3524 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3525 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
3526 IXGBE_ADVTXD_POPTS_SHIFT;
3528 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3529 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
3530 IXGBE_ADVTXD_POPTS_SHIFT;
3532 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
3534 i = tx_ring->next_to_use;
3536 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3537 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
3538 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
3539 tx_desc->read.cmd_type_len =
3540 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
3541 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3543 if (i == tx_ring->count)
3547 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
3550 * Force memory writes to complete before letting h/w
3551 * know there are new descriptors to fetch. (Only
3552 * applicable for weak-ordered memory model archs,
3557 tx_ring->next_to_use = i;
3558 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3561 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
3562 struct ixgbe_ring *tx_ring, int size)
3564 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3566 netif_stop_subqueue(netdev, tx_ring->queue_index);
3567 /* Herbert's original patch had:
3568 * smp_mb__after_netif_stop_queue();
3569 * but since that doesn't exist yet, just open code it. */
3572 /* We need to check again in a case another CPU has just
3573 * made room available. */
3574 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
3577 /* A reprieve! - use start_queue because it doesn't call schedule */
3578 netif_start_subqueue(netdev, tx_ring->queue_index);
3579 ++adapter->restart_queue;
3583 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
3584 struct ixgbe_ring *tx_ring, int size)
3586 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
3588 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
3591 static int ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3593 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3594 struct ixgbe_ring *tx_ring;
3596 unsigned int tx_flags = 0;
3602 r_idx = (adapter->num_tx_queues - 1) & skb->queue_mapping;
3603 tx_ring = &adapter->tx_ring[r_idx];
3605 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3606 tx_flags |= vlan_tx_tag_get(skb);
3607 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3608 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3610 /* three things can cause us to need a context descriptor */
3611 if (skb_is_gso(skb) ||
3612 (skb->ip_summed == CHECKSUM_PARTIAL) ||
3613 (tx_flags & IXGBE_TX_FLAGS_VLAN))
3616 count += TXD_USE_COUNT(skb_headlen(skb));
3617 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3618 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3620 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
3622 return NETDEV_TX_BUSY;
3625 if (skb->protocol == htons(ETH_P_IP))
3626 tx_flags |= IXGBE_TX_FLAGS_IPV4;
3627 first = tx_ring->next_to_use;
3628 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
3630 dev_kfree_skb_any(skb);
3631 return NETDEV_TX_OK;
3635 tx_flags |= IXGBE_TX_FLAGS_TSO;
3636 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
3637 (skb->ip_summed == CHECKSUM_PARTIAL))
3638 tx_flags |= IXGBE_TX_FLAGS_CSUM;
3640 ixgbe_tx_queue(adapter, tx_ring, tx_flags,
3641 ixgbe_tx_map(adapter, tx_ring, skb, first),
3644 netdev->trans_start = jiffies;
3646 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
3648 return NETDEV_TX_OK;
3652 * ixgbe_get_stats - Get System Network Statistics
3653 * @netdev: network interface device structure
3655 * Returns the address of the device statistics structure.
3656 * The statistics are actually updated from the timer callback.
3658 static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
3660 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3662 /* only return the current stats */
3663 return &adapter->net_stats;
3667 * ixgbe_set_mac - Change the Ethernet Address of the NIC
3668 * @netdev: network interface device structure
3669 * @p: pointer to an address structure
3671 * Returns 0 on success, negative on failure
3673 static int ixgbe_set_mac(struct net_device *netdev, void *p)
3675 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3676 struct ixgbe_hw *hw = &adapter->hw;
3677 struct sockaddr *addr = p;
3679 if (!is_valid_ether_addr(addr->sa_data))
3680 return -EADDRNOTAVAIL;
3682 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3683 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3685 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
3690 #ifdef CONFIG_NET_POLL_CONTROLLER
3692 * Polling 'interrupt' - used by things like netconsole to send skbs
3693 * without having to re-enable interrupts. It's not called while
3694 * the interrupt routine is executing.
3696 static void ixgbe_netpoll(struct net_device *netdev)
3698 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3700 disable_irq(adapter->pdev->irq);
3701 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
3702 ixgbe_intr(adapter->pdev->irq, netdev);
3703 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
3704 enable_irq(adapter->pdev->irq);
3709 * ixgbe_link_config - set up initial link with default speed and duplex
3710 * @hw: pointer to private hardware struct
3712 * Returns 0 on success, negative on failure
3714 static int ixgbe_link_config(struct ixgbe_hw *hw)
3716 u32 autoneg = IXGBE_LINK_SPEED_10GB_FULL;
3718 /* must always autoneg for both 1G and 10G link */
3719 hw->mac.autoneg = true;
3721 if ((hw->mac.type == ixgbe_mac_82598EB) &&
3722 (hw->phy.media_type == ixgbe_media_type_copper))
3723 autoneg = IXGBE_LINK_SPEED_82598_AUTONEG;
3725 return hw->mac.ops.setup_link_speed(hw, autoneg, true, true);
3728 static const struct net_device_ops ixgbe_netdev_ops = {
3729 .ndo_open = ixgbe_open,
3730 .ndo_stop = ixgbe_close,
3731 .ndo_start_xmit = ixgbe_xmit_frame,
3732 .ndo_get_stats = ixgbe_get_stats,
3733 .ndo_set_multicast_list = ixgbe_set_rx_mode,
3734 .ndo_validate_addr = eth_validate_addr,
3735 .ndo_set_mac_address = ixgbe_set_mac,
3736 .ndo_change_mtu = ixgbe_change_mtu,
3737 .ndo_tx_timeout = ixgbe_tx_timeout,
3738 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
3739 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
3740 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
3741 #ifdef CONFIG_NET_POLL_CONTROLLER
3742 .ndo_poll_controller = ixgbe_netpoll,
3747 * ixgbe_probe - Device Initialization Routine
3748 * @pdev: PCI device information struct
3749 * @ent: entry in ixgbe_pci_tbl
3751 * Returns 0 on success, negative on failure
3753 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
3754 * The OS initialization, configuring of the adapter private structure,
3755 * and a hardware reset occur.
3757 static int __devinit ixgbe_probe(struct pci_dev *pdev,
3758 const struct pci_device_id *ent)
3760 struct net_device *netdev;
3761 struct ixgbe_adapter *adapter = NULL;
3762 struct ixgbe_hw *hw;
3763 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
3764 static int cards_found;
3765 int i, err, pci_using_dac;
3766 u16 link_status, link_speed, link_width;
3769 err = pci_enable_device(pdev);
3773 if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK) &&
3774 !pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK)) {
3777 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
3779 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
3781 dev_err(&pdev->dev, "No usable DMA "
3782 "configuration, aborting\n");
3789 err = pci_request_regions(pdev, ixgbe_driver_name);
3791 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3795 pci_set_master(pdev);
3796 pci_save_state(pdev);
3798 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
3801 goto err_alloc_etherdev;
3804 SET_NETDEV_DEV(netdev, &pdev->dev);
3806 pci_set_drvdata(pdev, netdev);
3807 adapter = netdev_priv(netdev);
3809 adapter->netdev = netdev;
3810 adapter->pdev = pdev;
3813 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
3815 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3816 pci_resource_len(pdev, 0));
3822 for (i = 1; i <= 5; i++) {
3823 if (pci_resource_len(pdev, i) == 0)
3827 netdev->netdev_ops = &ixgbe_netdev_ops;
3828 ixgbe_set_ethtool_ops(netdev);
3829 netdev->watchdog_timeo = 5 * HZ;
3830 strcpy(netdev->name, pci_name(pdev));
3832 adapter->bd_number = cards_found;
3835 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3836 hw->mac.type = ii->mac;
3839 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
3840 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
3841 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
3842 if (!(eec & (1 << 8)))
3843 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
3846 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
3847 /* phy->sfp_type = ixgbe_sfp_type_unknown; */
3849 err = ii->get_invariants(hw);
3853 /* setup the private structure */
3854 err = ixgbe_sw_init(adapter);
3858 /* reset_hw fills in the perm_addr as well */
3859 err = hw->mac.ops.reset_hw(hw);
3861 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
3865 netdev->features = NETIF_F_SG |
3867 NETIF_F_HW_VLAN_TX |
3868 NETIF_F_HW_VLAN_RX |
3869 NETIF_F_HW_VLAN_FILTER;
3871 netdev->features |= NETIF_F_IPV6_CSUM;
3872 netdev->features |= NETIF_F_TSO;
3873 netdev->features |= NETIF_F_TSO6;
3874 netdev->features |= NETIF_F_LRO;
3876 netdev->vlan_features |= NETIF_F_TSO;
3877 netdev->vlan_features |= NETIF_F_TSO6;
3878 netdev->vlan_features |= NETIF_F_IP_CSUM;
3879 netdev->vlan_features |= NETIF_F_SG;
3882 netdev->features |= NETIF_F_HIGHDMA;
3884 /* make sure the EEPROM is good */
3885 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
3886 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
3891 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
3892 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
3894 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
3895 dev_err(&pdev->dev, "invalid MAC address\n");
3900 init_timer(&adapter->watchdog_timer);
3901 adapter->watchdog_timer.function = &ixgbe_watchdog;
3902 adapter->watchdog_timer.data = (unsigned long)adapter;
3904 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
3905 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
3907 err = ixgbe_init_interrupt_scheme(adapter);
3911 /* print bus type/speed/width info */
3912 pci_read_config_word(pdev, IXGBE_PCI_LINK_STATUS, &link_status);
3913 link_speed = link_status & IXGBE_PCI_LINK_SPEED;
3914 link_width = link_status & IXGBE_PCI_LINK_WIDTH;
3915 dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
3916 ((link_speed == IXGBE_PCI_LINK_SPEED_5000) ? "5.0Gb/s" :
3917 (link_speed == IXGBE_PCI_LINK_SPEED_2500) ? "2.5Gb/s" :
3919 ((link_width == IXGBE_PCI_LINK_WIDTH_8) ? "Width x8" :
3920 (link_width == IXGBE_PCI_LINK_WIDTH_4) ? "Width x4" :
3921 (link_width == IXGBE_PCI_LINK_WIDTH_2) ? "Width x2" :
3922 (link_width == IXGBE_PCI_LINK_WIDTH_1) ? "Width x1" :
3925 ixgbe_read_pba_num_generic(hw, &part_num);
3926 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
3927 hw->mac.type, hw->phy.type,
3928 (part_num >> 8), (part_num & 0xff));
3930 if (link_width <= IXGBE_PCI_LINK_WIDTH_4) {
3931 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
3932 "this card is not sufficient for optimal "
3934 dev_warn(&pdev->dev, "For optimal performance a x8 "
3935 "PCI-Express slot is required.\n");
3938 /* reset the hardware with the new settings */
3939 hw->mac.ops.start_hw(hw);
3941 /* link_config depends on start_hw being called at least once */
3942 err = ixgbe_link_config(hw);
3944 dev_err(&pdev->dev, "setup_link_speed FAILED %d\n", err);
3948 netif_carrier_off(netdev);
3950 ixgbe_napi_add_all(adapter);
3952 strcpy(netdev->name, "eth%d");
3953 err = register_netdev(netdev);
3957 #ifdef CONFIG_IXGBE_DCA
3958 if (dca_add_requester(&pdev->dev) == 0) {
3959 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
3960 /* always use CB2 mode, difference is masked
3961 * in the CB driver */
3962 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
3963 ixgbe_setup_dca(adapter);
3967 dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
3972 ixgbe_release_hw_control(adapter);
3975 ixgbe_reset_interrupt_capability(adapter);
3977 iounmap(hw->hw_addr);
3979 free_netdev(netdev);
3981 pci_release_regions(pdev);
3984 pci_disable_device(pdev);
3989 * ixgbe_remove - Device Removal Routine
3990 * @pdev: PCI device information struct
3992 * ixgbe_remove is called by the PCI subsystem to alert the driver
3993 * that it should release a PCI device. The could be caused by a
3994 * Hot-Plug event, or because the driver is going to be removed from
3997 static void __devexit ixgbe_remove(struct pci_dev *pdev)
3999 struct net_device *netdev = pci_get_drvdata(pdev);
4000 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4002 set_bit(__IXGBE_DOWN, &adapter->state);
4003 del_timer_sync(&adapter->watchdog_timer);
4005 flush_scheduled_work();
4007 #ifdef CONFIG_IXGBE_DCA
4008 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
4009 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
4010 dca_remove_requester(&pdev->dev);
4011 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
4015 unregister_netdev(netdev);
4017 ixgbe_reset_interrupt_capability(adapter);
4019 ixgbe_release_hw_control(adapter);
4021 iounmap(adapter->hw.hw_addr);
4022 pci_release_regions(pdev);
4024 DPRINTK(PROBE, INFO, "complete\n");
4025 ixgbe_napi_del_all(adapter);
4026 kfree(adapter->tx_ring);
4027 kfree(adapter->rx_ring);
4029 free_netdev(netdev);
4031 pci_disable_device(pdev);
4035 * ixgbe_io_error_detected - called when PCI error is detected
4036 * @pdev: Pointer to PCI device
4037 * @state: The current pci connection state
4039 * This function is called after a PCI bus error affecting
4040 * this device has been detected.
4042 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
4043 pci_channel_state_t state)
4045 struct net_device *netdev = pci_get_drvdata(pdev);
4046 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4048 netif_device_detach(netdev);
4050 if (netif_running(netdev))
4051 ixgbe_down(adapter);
4052 pci_disable_device(pdev);
4054 /* Request a slot reset. */
4055 return PCI_ERS_RESULT_NEED_RESET;
4059 * ixgbe_io_slot_reset - called after the pci bus has been reset.
4060 * @pdev: Pointer to PCI device
4062 * Restart the card from scratch, as if from a cold-boot.
4064 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
4066 struct net_device *netdev = pci_get_drvdata(pdev);
4067 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4069 if (pci_enable_device(pdev)) {
4071 "Cannot re-enable PCI device after reset.\n");
4072 return PCI_ERS_RESULT_DISCONNECT;
4074 pci_set_master(pdev);
4075 pci_restore_state(pdev);
4077 pci_enable_wake(pdev, PCI_D3hot, 0);
4078 pci_enable_wake(pdev, PCI_D3cold, 0);
4080 ixgbe_reset(adapter);
4082 return PCI_ERS_RESULT_RECOVERED;
4086 * ixgbe_io_resume - called when traffic can start flowing again.
4087 * @pdev: Pointer to PCI device
4089 * This callback is called when the error recovery driver tells us that
4090 * its OK to resume normal operation.
4092 static void ixgbe_io_resume(struct pci_dev *pdev)
4094 struct net_device *netdev = pci_get_drvdata(pdev);
4095 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4097 if (netif_running(netdev)) {
4098 if (ixgbe_up(adapter)) {
4099 DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
4104 netif_device_attach(netdev);
4107 static struct pci_error_handlers ixgbe_err_handler = {
4108 .error_detected = ixgbe_io_error_detected,
4109 .slot_reset = ixgbe_io_slot_reset,
4110 .resume = ixgbe_io_resume,
4113 static struct pci_driver ixgbe_driver = {
4114 .name = ixgbe_driver_name,
4115 .id_table = ixgbe_pci_tbl,
4116 .probe = ixgbe_probe,
4117 .remove = __devexit_p(ixgbe_remove),
4119 .suspend = ixgbe_suspend,
4120 .resume = ixgbe_resume,
4122 .shutdown = ixgbe_shutdown,
4123 .err_handler = &ixgbe_err_handler
4127 * ixgbe_init_module - Driver Registration Routine
4129 * ixgbe_init_module is the first routine called when the driver is
4130 * loaded. All it does is register with the PCI subsystem.
4132 static int __init ixgbe_init_module(void)
4135 printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
4136 ixgbe_driver_string, ixgbe_driver_version);
4138 printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
4140 #ifdef CONFIG_IXGBE_DCA
4141 dca_register_notify(&dca_notifier);
4144 ret = pci_register_driver(&ixgbe_driver);
4148 module_init(ixgbe_init_module);
4151 * ixgbe_exit_module - Driver Exit Cleanup Routine
4153 * ixgbe_exit_module is called just before the driver is removed
4156 static void __exit ixgbe_exit_module(void)
4158 #ifdef CONFIG_IXGBE_DCA
4159 dca_unregister_notify(&dca_notifier);
4161 pci_unregister_driver(&ixgbe_driver);
4164 #ifdef CONFIG_IXGBE_DCA
4165 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
4170 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
4171 __ixgbe_notify_dca);
4173 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4175 #endif /* CONFIG_IXGBE_DCA */
4177 module_exit(ixgbe_exit_module);