1 /*******************************************************************************
3 Intel 82599 Virtual Function driver
4 Copyright(c) 1999 - 2010 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
29 /******************************************************************************
30 Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
31 ******************************************************************************/
32 #include <linux/types.h>
33 #include <linux/module.h>
34 #include <linux/pci.h>
35 #include <linux/netdevice.h>
36 #include <linux/vmalloc.h>
37 #include <linux/string.h>
40 #include <linux/tcp.h>
41 #include <linux/ipv6.h>
42 #include <linux/slab.h>
43 #include <net/checksum.h>
44 #include <net/ip6_checksum.h>
45 #include <linux/ethtool.h>
46 #include <linux/if_vlan.h>
47 #include <linux/prefetch.h>
51 char ixgbevf_driver_name[] = "ixgbevf";
52 static const char ixgbevf_driver_string[] =
53 "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
55 #define DRV_VERSION "2.0.0-k2"
56 const char ixgbevf_driver_version[] = DRV_VERSION;
57 static char ixgbevf_copyright[] =
58 "Copyright (c) 2009 - 2010 Intel Corporation.";
60 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
61 [board_82599_vf] = &ixgbevf_82599_vf_info,
62 [board_X540_vf] = &ixgbevf_X540_vf_info,
65 /* ixgbevf_pci_tbl - PCI Device ID Table
67 * Wildcard entries (PCI_ANY_ID) should come last
68 * Last entry must be all 0s
70 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
71 * Class, Class Mask, private data (not used) }
73 static struct pci_device_id ixgbevf_pci_tbl[] = {
74 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF),
76 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF),
79 /* required last entry */
82 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
85 MODULE_DESCRIPTION("Intel(R) 82599 Virtual Function Driver");
86 MODULE_LICENSE("GPL");
87 MODULE_VERSION(DRV_VERSION);
89 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
92 static void ixgbevf_set_itr_msix(struct ixgbevf_q_vector *q_vector);
93 static void ixgbevf_write_eitr(struct ixgbevf_adapter *adapter, int v_idx,
96 static inline void ixgbevf_release_rx_desc(struct ixgbe_hw *hw,
97 struct ixgbevf_ring *rx_ring,
101 * Force memory writes to complete before letting h/w
102 * know there are new descriptors to fetch. (Only
103 * applicable for weak-ordered memory model archs,
107 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rx_ring->reg_idx), val);
111 * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
112 * @adapter: pointer to adapter struct
113 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
114 * @queue: queue to map the corresponding interrupt to
115 * @msix_vector: the vector to map to the corresponding queue
118 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
119 u8 queue, u8 msix_vector)
122 struct ixgbe_hw *hw = &adapter->hw;
123 if (direction == -1) {
125 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
126 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
129 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
131 /* tx or rx causes */
132 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
133 index = ((16 * (queue & 1)) + (8 * direction));
134 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
135 ivar &= ~(0xFF << index);
136 ivar |= (msix_vector << index);
137 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
141 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_adapter *adapter,
142 struct ixgbevf_tx_buffer
145 if (tx_buffer_info->dma) {
146 if (tx_buffer_info->mapped_as_page)
147 dma_unmap_page(&adapter->pdev->dev,
149 tx_buffer_info->length,
152 dma_unmap_single(&adapter->pdev->dev,
154 tx_buffer_info->length,
156 tx_buffer_info->dma = 0;
158 if (tx_buffer_info->skb) {
159 dev_kfree_skb_any(tx_buffer_info->skb);
160 tx_buffer_info->skb = NULL;
162 tx_buffer_info->time_stamp = 0;
163 /* tx_buffer_info must be completely set up in the transmit path */
166 #define IXGBE_MAX_TXD_PWR 14
167 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
169 /* Tx Descriptors needed, worst case */
170 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
171 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
173 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
174 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
176 #define DESC_NEEDED TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD)
179 static void ixgbevf_tx_timeout(struct net_device *netdev);
182 * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
183 * @adapter: board private structure
184 * @tx_ring: tx ring to clean
186 static bool ixgbevf_clean_tx_irq(struct ixgbevf_adapter *adapter,
187 struct ixgbevf_ring *tx_ring)
189 struct net_device *netdev = adapter->netdev;
190 struct ixgbe_hw *hw = &adapter->hw;
191 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
192 struct ixgbevf_tx_buffer *tx_buffer_info;
193 unsigned int i, eop, count = 0;
194 unsigned int total_bytes = 0, total_packets = 0;
196 i = tx_ring->next_to_clean;
197 eop = tx_ring->tx_buffer_info[i].next_to_watch;
198 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
200 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
201 (count < tx_ring->work_limit)) {
202 bool cleaned = false;
203 rmb(); /* read buffer_info after eop_desc */
204 for ( ; !cleaned; count++) {
206 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
207 tx_buffer_info = &tx_ring->tx_buffer_info[i];
208 cleaned = (i == eop);
209 skb = tx_buffer_info->skb;
211 if (cleaned && skb) {
212 unsigned int segs, bytecount;
214 /* gso_segs is currently only valid for tcp */
215 segs = skb_shinfo(skb)->gso_segs ?: 1;
216 /* multiply data chunks by size of headers */
217 bytecount = ((segs - 1) * skb_headlen(skb)) +
219 total_packets += segs;
220 total_bytes += bytecount;
223 ixgbevf_unmap_and_free_tx_resource(adapter,
226 tx_desc->wb.status = 0;
229 if (i == tx_ring->count)
233 eop = tx_ring->tx_buffer_info[i].next_to_watch;
234 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
237 tx_ring->next_to_clean = i;
239 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
240 if (unlikely(count && netif_carrier_ok(netdev) &&
241 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
242 /* Make sure that anybody stopping the queue after this
243 * sees the new next_to_clean.
247 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
248 !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
249 netif_wake_subqueue(netdev, tx_ring->queue_index);
250 ++adapter->restart_queue;
253 if (netif_queue_stopped(netdev) &&
254 !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
255 netif_wake_queue(netdev);
256 ++adapter->restart_queue;
261 /* re-arm the interrupt */
262 if ((count >= tx_ring->work_limit) &&
263 (!test_bit(__IXGBEVF_DOWN, &adapter->state))) {
264 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, tx_ring->v_idx);
267 tx_ring->total_bytes += total_bytes;
268 tx_ring->total_packets += total_packets;
270 netdev->stats.tx_bytes += total_bytes;
271 netdev->stats.tx_packets += total_packets;
273 return count < tx_ring->work_limit;
277 * ixgbevf_receive_skb - Send a completed packet up the stack
278 * @q_vector: structure containing interrupt and ring information
279 * @skb: packet to send up
280 * @status: hardware indication of status of receive
281 * @rx_ring: rx descriptor ring (for a specific queue) to setup
282 * @rx_desc: rx descriptor
284 static void ixgbevf_receive_skb(struct ixgbevf_q_vector *q_vector,
285 struct sk_buff *skb, u8 status,
286 struct ixgbevf_ring *ring,
287 union ixgbe_adv_rx_desc *rx_desc)
289 struct ixgbevf_adapter *adapter = q_vector->adapter;
290 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
291 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
293 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
294 if (adapter->vlgrp && is_vlan)
295 vlan_gro_receive(&q_vector->napi,
299 napi_gro_receive(&q_vector->napi, skb);
301 if (adapter->vlgrp && is_vlan)
302 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
309 * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
310 * @adapter: address of board private structure
311 * @status_err: hardware indication of status of receive
312 * @skb: skb currently being received and modified
314 static inline void ixgbevf_rx_checksum(struct ixgbevf_adapter *adapter,
315 u32 status_err, struct sk_buff *skb)
317 skb_checksum_none_assert(skb);
319 /* Rx csum disabled */
320 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
323 /* if IP and error */
324 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
325 (status_err & IXGBE_RXDADV_ERR_IPE)) {
326 adapter->hw_csum_rx_error++;
330 if (!(status_err & IXGBE_RXD_STAT_L4CS))
333 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
334 adapter->hw_csum_rx_error++;
338 /* It must be a TCP or UDP packet with a valid checksum */
339 skb->ip_summed = CHECKSUM_UNNECESSARY;
340 adapter->hw_csum_rx_good++;
344 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
345 * @adapter: address of board private structure
347 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_adapter *adapter,
348 struct ixgbevf_ring *rx_ring,
351 struct pci_dev *pdev = adapter->pdev;
352 union ixgbe_adv_rx_desc *rx_desc;
353 struct ixgbevf_rx_buffer *bi;
356 unsigned int bufsz = rx_ring->rx_buf_len + NET_IP_ALIGN;
358 i = rx_ring->next_to_use;
359 bi = &rx_ring->rx_buffer_info[i];
361 while (cleaned_count--) {
362 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
365 (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
367 bi->page = netdev_alloc_page(adapter->netdev);
369 adapter->alloc_rx_page_failed++;
374 /* use a half page if we're re-using */
375 bi->page_offset ^= (PAGE_SIZE / 2);
378 bi->page_dma = dma_map_page(&pdev->dev, bi->page,
386 skb = netdev_alloc_skb(adapter->netdev,
390 adapter->alloc_rx_buff_failed++;
395 * Make buffer alignment 2 beyond a 16 byte boundary
396 * this will result in a 16 byte aligned IP header after
397 * the 14 byte MAC header is removed
399 skb_reserve(skb, NET_IP_ALIGN);
404 bi->dma = dma_map_single(&pdev->dev, skb->data,
408 /* Refresh the desc even if buffer_addrs didn't change because
409 * each write-back erases this info. */
410 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
411 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
412 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
414 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
418 if (i == rx_ring->count)
420 bi = &rx_ring->rx_buffer_info[i];
424 if (rx_ring->next_to_use != i) {
425 rx_ring->next_to_use = i;
427 i = (rx_ring->count - 1);
429 ixgbevf_release_rx_desc(&adapter->hw, rx_ring, i);
433 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
437 struct ixgbe_hw *hw = &adapter->hw;
439 mask = (qmask & 0xFFFFFFFF);
440 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, mask);
443 static inline u16 ixgbevf_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
445 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
448 static inline u16 ixgbevf_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
450 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
453 static bool ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
454 struct ixgbevf_ring *rx_ring,
455 int *work_done, int work_to_do)
457 struct ixgbevf_adapter *adapter = q_vector->adapter;
458 struct pci_dev *pdev = adapter->pdev;
459 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
460 struct ixgbevf_rx_buffer *rx_buffer_info, *next_buffer;
465 bool cleaned = false;
466 int cleaned_count = 0;
467 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
469 i = rx_ring->next_to_clean;
470 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
471 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
472 rx_buffer_info = &rx_ring->rx_buffer_info[i];
474 while (staterr & IXGBE_RXD_STAT_DD) {
476 if (*work_done >= work_to_do)
480 rmb(); /* read descriptor and rx_buffer_info after status DD */
481 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
482 hdr_info = le16_to_cpu(ixgbevf_get_hdr_info(rx_desc));
483 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
484 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
485 if (hdr_info & IXGBE_RXDADV_SPH)
486 adapter->rx_hdr_split++;
487 if (len > IXGBEVF_RX_HDR_SIZE)
488 len = IXGBEVF_RX_HDR_SIZE;
489 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
491 len = le16_to_cpu(rx_desc->wb.upper.length);
494 skb = rx_buffer_info->skb;
495 prefetch(skb->data - NET_IP_ALIGN);
496 rx_buffer_info->skb = NULL;
498 if (rx_buffer_info->dma) {
499 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
502 rx_buffer_info->dma = 0;
507 dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
508 PAGE_SIZE / 2, DMA_FROM_DEVICE);
509 rx_buffer_info->page_dma = 0;
510 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
511 rx_buffer_info->page,
512 rx_buffer_info->page_offset,
515 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
516 (page_count(rx_buffer_info->page) != 1))
517 rx_buffer_info->page = NULL;
519 get_page(rx_buffer_info->page);
521 skb->len += upper_len;
522 skb->data_len += upper_len;
523 skb->truesize += upper_len;
527 if (i == rx_ring->count)
530 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
534 next_buffer = &rx_ring->rx_buffer_info[i];
536 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
537 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
538 rx_buffer_info->skb = next_buffer->skb;
539 rx_buffer_info->dma = next_buffer->dma;
540 next_buffer->skb = skb;
541 next_buffer->dma = 0;
543 skb->next = next_buffer->skb;
544 skb->next->prev = skb;
546 adapter->non_eop_descs++;
550 /* ERR_MASK will only have valid bits if EOP set */
551 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
552 dev_kfree_skb_irq(skb);
556 ixgbevf_rx_checksum(adapter, staterr, skb);
558 /* probably a little skewed due to removing CRC */
559 total_rx_bytes += skb->len;
563 * Work around issue of some types of VM to VM loop back
564 * packets not getting split correctly
566 if (staterr & IXGBE_RXD_STAT_LB) {
567 u32 header_fixup_len = skb_headlen(skb);
568 if (header_fixup_len < 14)
569 skb_push(skb, header_fixup_len);
571 skb->protocol = eth_type_trans(skb, adapter->netdev);
573 ixgbevf_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
576 rx_desc->wb.upper.status_error = 0;
578 /* return some buffers to hardware, one at a time is too slow */
579 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
580 ixgbevf_alloc_rx_buffers(adapter, rx_ring,
585 /* use prefetched values */
587 rx_buffer_info = &rx_ring->rx_buffer_info[i];
589 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
592 rx_ring->next_to_clean = i;
593 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
596 ixgbevf_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
598 rx_ring->total_packets += total_rx_packets;
599 rx_ring->total_bytes += total_rx_bytes;
600 adapter->netdev->stats.rx_bytes += total_rx_bytes;
601 adapter->netdev->stats.rx_packets += total_rx_packets;
607 * ixgbevf_clean_rxonly - msix (aka one shot) rx clean routine
608 * @napi: napi struct with our devices info in it
609 * @budget: amount of work driver is allowed to do this pass, in packets
611 * This function is optimized for cleaning one queue only on a single
614 static int ixgbevf_clean_rxonly(struct napi_struct *napi, int budget)
616 struct ixgbevf_q_vector *q_vector =
617 container_of(napi, struct ixgbevf_q_vector, napi);
618 struct ixgbevf_adapter *adapter = q_vector->adapter;
619 struct ixgbevf_ring *rx_ring = NULL;
623 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
624 rx_ring = &(adapter->rx_ring[r_idx]);
626 ixgbevf_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
628 /* If all Rx work done, exit the polling mode */
629 if (work_done < budget) {
631 if (adapter->itr_setting & 1)
632 ixgbevf_set_itr_msix(q_vector);
633 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
634 ixgbevf_irq_enable_queues(adapter, rx_ring->v_idx);
641 * ixgbevf_clean_rxonly_many - msix (aka one shot) rx clean routine
642 * @napi: napi struct with our devices info in it
643 * @budget: amount of work driver is allowed to do this pass, in packets
645 * This function will clean more than one rx queue associated with a
648 static int ixgbevf_clean_rxonly_many(struct napi_struct *napi, int budget)
650 struct ixgbevf_q_vector *q_vector =
651 container_of(napi, struct ixgbevf_q_vector, napi);
652 struct ixgbevf_adapter *adapter = q_vector->adapter;
653 struct ixgbevf_ring *rx_ring = NULL;
654 int work_done = 0, i;
658 /* attempt to distribute budget to each queue fairly, but don't allow
659 * the budget to go below 1 because we'll exit polling */
660 budget /= (q_vector->rxr_count ?: 1);
661 budget = max(budget, 1);
662 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
663 for (i = 0; i < q_vector->rxr_count; i++) {
664 rx_ring = &(adapter->rx_ring[r_idx]);
665 ixgbevf_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
666 enable_mask |= rx_ring->v_idx;
667 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
671 #ifndef HAVE_NETDEV_NAPI_LIST
672 if (!netif_running(adapter->netdev))
676 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
677 rx_ring = &(adapter->rx_ring[r_idx]);
679 /* If all Rx work done, exit the polling mode */
680 if (work_done < budget) {
682 if (adapter->itr_setting & 1)
683 ixgbevf_set_itr_msix(q_vector);
684 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
685 ixgbevf_irq_enable_queues(adapter, enable_mask);
693 * ixgbevf_configure_msix - Configure MSI-X hardware
694 * @adapter: board private structure
696 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
699 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
701 struct ixgbevf_q_vector *q_vector;
702 struct ixgbe_hw *hw = &adapter->hw;
703 int i, j, q_vectors, v_idx, r_idx;
706 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
709 * Populate the IVAR table and set the ITR values to the
710 * corresponding register.
712 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
713 q_vector = adapter->q_vector[v_idx];
714 /* XXX for_each_set_bit(...) */
715 r_idx = find_first_bit(q_vector->rxr_idx,
716 adapter->num_rx_queues);
718 for (i = 0; i < q_vector->rxr_count; i++) {
719 j = adapter->rx_ring[r_idx].reg_idx;
720 ixgbevf_set_ivar(adapter, 0, j, v_idx);
721 r_idx = find_next_bit(q_vector->rxr_idx,
722 adapter->num_rx_queues,
725 r_idx = find_first_bit(q_vector->txr_idx,
726 adapter->num_tx_queues);
728 for (i = 0; i < q_vector->txr_count; i++) {
729 j = adapter->tx_ring[r_idx].reg_idx;
730 ixgbevf_set_ivar(adapter, 1, j, v_idx);
731 r_idx = find_next_bit(q_vector->txr_idx,
732 adapter->num_tx_queues,
736 /* if this is a tx only vector halve the interrupt rate */
737 if (q_vector->txr_count && !q_vector->rxr_count)
738 q_vector->eitr = (adapter->eitr_param >> 1);
739 else if (q_vector->rxr_count)
741 q_vector->eitr = adapter->eitr_param;
743 ixgbevf_write_eitr(adapter, v_idx, q_vector->eitr);
746 ixgbevf_set_ivar(adapter, -1, 1, v_idx);
748 /* set up to autoclear timer, and the vectors */
749 mask = IXGBE_EIMS_ENABLE_MASK;
750 mask &= ~IXGBE_EIMS_OTHER;
751 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, mask);
758 latency_invalid = 255
762 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
763 * @adapter: pointer to adapter
764 * @eitr: eitr setting (ints per sec) to give last timeslice
765 * @itr_setting: current throttle rate in ints/second
766 * @packets: the number of packets during this measurement interval
767 * @bytes: the number of bytes during this measurement interval
769 * Stores a new ITR value based on packets and byte
770 * counts during the last interrupt. The advantage of per interrupt
771 * computation is faster updates and more accurate ITR for the current
772 * traffic pattern. Constants in this function were computed
773 * based on theoretical maximum wire speed and thresholds were set based
774 * on testing data as well as attempting to minimize response time
775 * while increasing bulk throughput.
777 static u8 ixgbevf_update_itr(struct ixgbevf_adapter *adapter,
778 u32 eitr, u8 itr_setting,
779 int packets, int bytes)
781 unsigned int retval = itr_setting;
786 goto update_itr_done;
789 /* simple throttlerate management
790 * 0-20MB/s lowest (100000 ints/s)
791 * 20-100MB/s low (20000 ints/s)
792 * 100-1249MB/s bulk (8000 ints/s)
794 /* what was last interrupt timeslice? */
795 timepassed_us = 1000000/eitr;
796 bytes_perint = bytes / timepassed_us; /* bytes/usec */
798 switch (itr_setting) {
800 if (bytes_perint > adapter->eitr_low)
801 retval = low_latency;
804 if (bytes_perint > adapter->eitr_high)
805 retval = bulk_latency;
806 else if (bytes_perint <= adapter->eitr_low)
807 retval = lowest_latency;
810 if (bytes_perint <= adapter->eitr_high)
811 retval = low_latency;
820 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
821 * @adapter: pointer to adapter struct
822 * @v_idx: vector index into q_vector array
823 * @itr_reg: new value to be written in *register* format, not ints/s
825 * This function is made to be called by ethtool and by the driver
826 * when it needs to update VTEITR registers at runtime. Hardware
827 * specific quirks/differences are taken care of here.
829 static void ixgbevf_write_eitr(struct ixgbevf_adapter *adapter, int v_idx,
832 struct ixgbe_hw *hw = &adapter->hw;
834 itr_reg = EITR_INTS_PER_SEC_TO_REG(itr_reg);
837 * set the WDIS bit to not clear the timer bits and cause an
838 * immediate assertion of the interrupt
840 itr_reg |= IXGBE_EITR_CNT_WDIS;
842 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
845 static void ixgbevf_set_itr_msix(struct ixgbevf_q_vector *q_vector)
847 struct ixgbevf_adapter *adapter = q_vector->adapter;
849 u8 current_itr, ret_itr;
850 int i, r_idx, v_idx = q_vector->v_idx;
851 struct ixgbevf_ring *rx_ring, *tx_ring;
853 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
854 for (i = 0; i < q_vector->txr_count; i++) {
855 tx_ring = &(adapter->tx_ring[r_idx]);
856 ret_itr = ixgbevf_update_itr(adapter, q_vector->eitr,
858 tx_ring->total_packets,
859 tx_ring->total_bytes);
860 /* if the result for this queue would decrease interrupt
861 * rate for this vector then use that result */
862 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
863 q_vector->tx_itr - 1 : ret_itr);
864 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
868 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
869 for (i = 0; i < q_vector->rxr_count; i++) {
870 rx_ring = &(adapter->rx_ring[r_idx]);
871 ret_itr = ixgbevf_update_itr(adapter, q_vector->eitr,
873 rx_ring->total_packets,
874 rx_ring->total_bytes);
875 /* if the result for this queue would decrease interrupt
876 * rate for this vector then use that result */
877 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
878 q_vector->rx_itr - 1 : ret_itr);
879 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
883 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
885 switch (current_itr) {
886 /* counts and packets in update_itr are dependent on these numbers */
891 new_itr = 20000; /* aka hwitr = ~200 */
899 if (new_itr != q_vector->eitr) {
902 /* save the algorithm value here, not the smoothed one */
903 q_vector->eitr = new_itr;
904 /* do an exponential smoothing */
905 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
906 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
907 ixgbevf_write_eitr(adapter, v_idx, itr_reg);
911 static irqreturn_t ixgbevf_msix_mbx(int irq, void *data)
913 struct net_device *netdev = data;
914 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
915 struct ixgbe_hw *hw = &adapter->hw;
919 eicr = IXGBE_READ_REG(hw, IXGBE_VTEICS);
920 IXGBE_WRITE_REG(hw, IXGBE_VTEICR, eicr);
922 if (!hw->mbx.ops.check_for_ack(hw)) {
924 * checking for the ack clears the PFACK bit. Place
925 * it back in the v2p_mailbox cache so that anyone
926 * polling for an ack will not miss it. Also
927 * avoid the read below because the code to read
928 * the mailbox will also clear the ack bit. This was
929 * causing lost acks. Just cache the bit and exit
932 hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFACK;
936 /* Not an ack interrupt, go ahead and read the message */
937 hw->mbx.ops.read(hw, &msg, 1);
939 if ((msg & IXGBE_MBVFICR_VFREQ_MASK) == IXGBE_PF_CONTROL_MSG)
940 mod_timer(&adapter->watchdog_timer,
941 round_jiffies(jiffies + 1));
947 static irqreturn_t ixgbevf_msix_clean_tx(int irq, void *data)
949 struct ixgbevf_q_vector *q_vector = data;
950 struct ixgbevf_adapter *adapter = q_vector->adapter;
951 struct ixgbevf_ring *tx_ring;
954 if (!q_vector->txr_count)
957 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
958 for (i = 0; i < q_vector->txr_count; i++) {
959 tx_ring = &(adapter->tx_ring[r_idx]);
960 tx_ring->total_bytes = 0;
961 tx_ring->total_packets = 0;
962 ixgbevf_clean_tx_irq(adapter, tx_ring);
963 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
967 if (adapter->itr_setting & 1)
968 ixgbevf_set_itr_msix(q_vector);
974 * ixgbevf_msix_clean_rx - single unshared vector rx clean (all queues)
976 * @data: pointer to our q_vector struct for this interrupt vector
978 static irqreturn_t ixgbevf_msix_clean_rx(int irq, void *data)
980 struct ixgbevf_q_vector *q_vector = data;
981 struct ixgbevf_adapter *adapter = q_vector->adapter;
982 struct ixgbe_hw *hw = &adapter->hw;
983 struct ixgbevf_ring *rx_ring;
987 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
988 for (i = 0; i < q_vector->rxr_count; i++) {
989 rx_ring = &(adapter->rx_ring[r_idx]);
990 rx_ring->total_bytes = 0;
991 rx_ring->total_packets = 0;
992 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
996 if (!q_vector->rxr_count)
999 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1000 rx_ring = &(adapter->rx_ring[r_idx]);
1001 /* disable interrupts on this vector only */
1002 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, rx_ring->v_idx);
1003 napi_schedule(&q_vector->napi);
1009 static irqreturn_t ixgbevf_msix_clean_many(int irq, void *data)
1011 ixgbevf_msix_clean_rx(irq, data);
1012 ixgbevf_msix_clean_tx(irq, data);
1017 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1020 struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1022 set_bit(r_idx, q_vector->rxr_idx);
1023 q_vector->rxr_count++;
1024 a->rx_ring[r_idx].v_idx = 1 << v_idx;
1027 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1030 struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1032 set_bit(t_idx, q_vector->txr_idx);
1033 q_vector->txr_count++;
1034 a->tx_ring[t_idx].v_idx = 1 << v_idx;
1038 * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1039 * @adapter: board private structure to initialize
1041 * This function maps descriptor rings to the queue-specific vectors
1042 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1043 * one vector per ring/queue, but on a constrained vector budget, we
1044 * group the rings as "efficiently" as possible. You would add new
1045 * mapping configurations in here.
1047 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1051 int rxr_idx = 0, txr_idx = 0;
1052 int rxr_remaining = adapter->num_rx_queues;
1053 int txr_remaining = adapter->num_tx_queues;
1058 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1061 * The ideal configuration...
1062 * We have enough vectors to map one per queue.
1064 if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1065 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1066 map_vector_to_rxq(adapter, v_start, rxr_idx);
1068 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1069 map_vector_to_txq(adapter, v_start, txr_idx);
1074 * If we don't have enough vectors for a 1-to-1
1075 * mapping, we'll have to group them so there are
1076 * multiple queues per vector.
1078 /* Re-adjusting *qpv takes care of the remainder. */
1079 for (i = v_start; i < q_vectors; i++) {
1080 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1081 for (j = 0; j < rqpv; j++) {
1082 map_vector_to_rxq(adapter, i, rxr_idx);
1087 for (i = v_start; i < q_vectors; i++) {
1088 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1089 for (j = 0; j < tqpv; j++) {
1090 map_vector_to_txq(adapter, i, txr_idx);
1101 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1102 * @adapter: board private structure
1104 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1105 * interrupts from the kernel.
1107 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1109 struct net_device *netdev = adapter->netdev;
1110 irqreturn_t (*handler)(int, void *);
1111 int i, vector, q_vectors, err;
1114 /* Decrement for Other and TCP Timer vectors */
1115 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1117 #define SET_HANDLER(_v) (((_v)->rxr_count && (_v)->txr_count) \
1118 ? &ixgbevf_msix_clean_many : \
1119 (_v)->rxr_count ? &ixgbevf_msix_clean_rx : \
1120 (_v)->txr_count ? &ixgbevf_msix_clean_tx : \
1122 for (vector = 0; vector < q_vectors; vector++) {
1123 handler = SET_HANDLER(adapter->q_vector[vector]);
1125 if (handler == &ixgbevf_msix_clean_rx) {
1126 sprintf(adapter->name[vector], "%s-%s-%d",
1127 netdev->name, "rx", ri++);
1128 } else if (handler == &ixgbevf_msix_clean_tx) {
1129 sprintf(adapter->name[vector], "%s-%s-%d",
1130 netdev->name, "tx", ti++);
1131 } else if (handler == &ixgbevf_msix_clean_many) {
1132 sprintf(adapter->name[vector], "%s-%s-%d",
1133 netdev->name, "TxRx", vector);
1135 /* skip this unused q_vector */
1138 err = request_irq(adapter->msix_entries[vector].vector,
1139 handler, 0, adapter->name[vector],
1140 adapter->q_vector[vector]);
1142 hw_dbg(&adapter->hw,
1143 "request_irq failed for MSIX interrupt "
1144 "Error: %d\n", err);
1145 goto free_queue_irqs;
1149 sprintf(adapter->name[vector], "%s:mbx", netdev->name);
1150 err = request_irq(adapter->msix_entries[vector].vector,
1151 &ixgbevf_msix_mbx, 0, adapter->name[vector], netdev);
1153 hw_dbg(&adapter->hw,
1154 "request_irq for msix_mbx failed: %d\n", err);
1155 goto free_queue_irqs;
1161 for (i = vector - 1; i >= 0; i--)
1162 free_irq(adapter->msix_entries[--vector].vector,
1163 &(adapter->q_vector[i]));
1164 pci_disable_msix(adapter->pdev);
1165 kfree(adapter->msix_entries);
1166 adapter->msix_entries = NULL;
1170 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1172 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1174 for (i = 0; i < q_vectors; i++) {
1175 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1176 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1177 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1178 q_vector->rxr_count = 0;
1179 q_vector->txr_count = 0;
1180 q_vector->eitr = adapter->eitr_param;
1185 * ixgbevf_request_irq - initialize interrupts
1186 * @adapter: board private structure
1188 * Attempts to configure interrupts using the best available
1189 * capabilities of the hardware and kernel.
1191 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1195 err = ixgbevf_request_msix_irqs(adapter);
1198 hw_dbg(&adapter->hw,
1199 "request_irq failed, Error %d\n", err);
1204 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1206 struct net_device *netdev = adapter->netdev;
1209 q_vectors = adapter->num_msix_vectors;
1213 free_irq(adapter->msix_entries[i].vector, netdev);
1216 for (; i >= 0; i--) {
1217 free_irq(adapter->msix_entries[i].vector,
1218 adapter->q_vector[i]);
1221 ixgbevf_reset_q_vectors(adapter);
1225 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1226 * @adapter: board private structure
1228 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1231 struct ixgbe_hw *hw = &adapter->hw;
1233 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1235 IXGBE_WRITE_FLUSH(hw);
1237 for (i = 0; i < adapter->num_msix_vectors; i++)
1238 synchronize_irq(adapter->msix_entries[i].vector);
1242 * ixgbevf_irq_enable - Enable default interrupt generation settings
1243 * @adapter: board private structure
1245 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter,
1246 bool queues, bool flush)
1248 struct ixgbe_hw *hw = &adapter->hw;
1252 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1255 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, mask);
1258 ixgbevf_irq_enable_queues(adapter, qmask);
1261 IXGBE_WRITE_FLUSH(hw);
1265 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1266 * @adapter: board private structure
1268 * Configure the Tx unit of the MAC after a reset.
1270 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1273 struct ixgbe_hw *hw = &adapter->hw;
1274 u32 i, j, tdlen, txctrl;
1276 /* Setup the HW Tx Head and Tail descriptor pointers */
1277 for (i = 0; i < adapter->num_tx_queues; i++) {
1278 struct ixgbevf_ring *ring = &adapter->tx_ring[i];
1281 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1282 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1283 (tdba & DMA_BIT_MASK(32)));
1284 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1285 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j), tdlen);
1286 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(j), 0);
1287 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(j), 0);
1288 adapter->tx_ring[i].head = IXGBE_VFTDH(j);
1289 adapter->tx_ring[i].tail = IXGBE_VFTDT(j);
1290 /* Disable Tx Head Writeback RO bit, since this hoses
1291 * bookkeeping if things aren't delivered in order.
1293 txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1294 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1295 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1299 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1301 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1303 struct ixgbevf_ring *rx_ring;
1304 struct ixgbe_hw *hw = &adapter->hw;
1307 rx_ring = &adapter->rx_ring[index];
1309 srrctl = IXGBE_SRRCTL_DROP_EN;
1311 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1312 u16 bufsz = IXGBEVF_RXBUFFER_2048;
1313 /* grow the amount we can receive on large page machines */
1314 if (bufsz < (PAGE_SIZE / 2))
1315 bufsz = (PAGE_SIZE / 2);
1316 /* cap the bufsz at our largest descriptor size */
1317 bufsz = min((u16)IXGBEVF_MAX_RXBUFFER, bufsz);
1319 srrctl |= bufsz >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1320 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1321 srrctl |= ((IXGBEVF_RX_HDR_SIZE <<
1322 IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1323 IXGBE_SRRCTL_BSIZEHDR_MASK);
1325 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1327 if (rx_ring->rx_buf_len == MAXIMUM_ETHERNET_VLAN_SIZE)
1328 srrctl |= IXGBEVF_RXBUFFER_2048 >>
1329 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1331 srrctl |= rx_ring->rx_buf_len >>
1332 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1334 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1338 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1339 * @adapter: board private structure
1341 * Configure the Rx unit of the MAC after a reset.
1343 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1346 struct ixgbe_hw *hw = &adapter->hw;
1347 struct net_device *netdev = adapter->netdev;
1348 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1353 /* Decide whether to use packet split mode or not */
1354 if (netdev->mtu > ETH_DATA_LEN) {
1355 if (adapter->flags & IXGBE_FLAG_RX_PS_CAPABLE)
1356 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
1358 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
1360 if (adapter->flags & IXGBE_FLAG_RX_1BUF_CAPABLE)
1361 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
1363 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
1366 /* Set the RX buffer length according to the mode */
1367 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1368 /* PSRTYPE must be initialized in 82599 */
1369 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
1370 IXGBE_PSRTYPE_UDPHDR |
1371 IXGBE_PSRTYPE_IPV4HDR |
1372 IXGBE_PSRTYPE_IPV6HDR |
1373 IXGBE_PSRTYPE_L2HDR;
1374 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1375 rx_buf_len = IXGBEVF_RX_HDR_SIZE;
1377 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, 0);
1378 if (netdev->mtu <= ETH_DATA_LEN)
1379 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1381 rx_buf_len = ALIGN(max_frame, 1024);
1384 rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1385 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1386 * the Base and Length of the Rx Descriptor Ring */
1387 for (i = 0; i < adapter->num_rx_queues; i++) {
1388 rdba = adapter->rx_ring[i].dma;
1389 j = adapter->rx_ring[i].reg_idx;
1390 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1391 (rdba & DMA_BIT_MASK(32)));
1392 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1393 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j), rdlen);
1394 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(j), 0);
1395 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(j), 0);
1396 adapter->rx_ring[i].head = IXGBE_VFRDH(j);
1397 adapter->rx_ring[i].tail = IXGBE_VFRDT(j);
1398 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1400 ixgbevf_configure_srrctl(adapter, j);
1404 static void ixgbevf_vlan_rx_register(struct net_device *netdev,
1405 struct vlan_group *grp)
1407 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1408 struct ixgbe_hw *hw = &adapter->hw;
1412 adapter->vlgrp = grp;
1414 for (i = 0; i < adapter->num_rx_queues; i++) {
1415 j = adapter->rx_ring[i].reg_idx;
1416 ctrl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1417 ctrl |= IXGBE_RXDCTL_VME;
1418 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), ctrl);
1422 static void ixgbevf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1424 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1425 struct ixgbe_hw *hw = &adapter->hw;
1427 /* add VID to filter table */
1428 if (hw->mac.ops.set_vfta)
1429 hw->mac.ops.set_vfta(hw, vid, 0, true);
1432 static void ixgbevf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1434 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1435 struct ixgbe_hw *hw = &adapter->hw;
1437 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
1438 ixgbevf_irq_disable(adapter);
1440 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1442 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
1443 ixgbevf_irq_enable(adapter, true, true);
1445 /* remove VID from filter table */
1446 if (hw->mac.ops.set_vfta)
1447 hw->mac.ops.set_vfta(hw, vid, 0, false);
1450 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1452 ixgbevf_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1454 if (adapter->vlgrp) {
1456 for (vid = 0; vid < VLAN_N_VID; vid++) {
1457 if (!vlan_group_get_device(adapter->vlgrp, vid))
1459 ixgbevf_vlan_rx_add_vid(adapter->netdev, vid);
1464 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1466 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1467 struct ixgbe_hw *hw = &adapter->hw;
1470 if ((netdev_uc_count(netdev)) > 10) {
1471 printk(KERN_ERR "Too many unicast filters - No Space\n");
1475 if (!netdev_uc_empty(netdev)) {
1476 struct netdev_hw_addr *ha;
1477 netdev_for_each_uc_addr(ha, netdev) {
1478 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1483 * If the list is empty then send message to PF driver to
1484 * clear all macvlans on this VF.
1486 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1493 * ixgbevf_set_rx_mode - Multicast set
1494 * @netdev: network interface device structure
1496 * The set_rx_method entry point is called whenever the multicast address
1497 * list or the network interface flags are updated. This routine is
1498 * responsible for configuring the hardware for proper multicast mode.
1500 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1502 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1503 struct ixgbe_hw *hw = &adapter->hw;
1505 /* reprogram multicast list */
1506 if (hw->mac.ops.update_mc_addr_list)
1507 hw->mac.ops.update_mc_addr_list(hw, netdev);
1509 ixgbevf_write_uc_addr_list(netdev);
1512 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1515 struct ixgbevf_q_vector *q_vector;
1516 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1518 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1519 struct napi_struct *napi;
1520 q_vector = adapter->q_vector[q_idx];
1521 if (!q_vector->rxr_count)
1523 napi = &q_vector->napi;
1524 if (q_vector->rxr_count > 1)
1525 napi->poll = &ixgbevf_clean_rxonly_many;
1531 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1534 struct ixgbevf_q_vector *q_vector;
1535 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1537 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1538 q_vector = adapter->q_vector[q_idx];
1539 if (!q_vector->rxr_count)
1541 napi_disable(&q_vector->napi);
1545 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1547 struct net_device *netdev = adapter->netdev;
1550 ixgbevf_set_rx_mode(netdev);
1552 ixgbevf_restore_vlan(adapter);
1554 ixgbevf_configure_tx(adapter);
1555 ixgbevf_configure_rx(adapter);
1556 for (i = 0; i < adapter->num_rx_queues; i++) {
1557 struct ixgbevf_ring *ring = &adapter->rx_ring[i];
1558 ixgbevf_alloc_rx_buffers(adapter, ring, ring->count);
1559 ring->next_to_use = ring->count - 1;
1560 writel(ring->next_to_use, adapter->hw.hw_addr + ring->tail);
1564 #define IXGBE_MAX_RX_DESC_POLL 10
1565 static inline void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1568 struct ixgbe_hw *hw = &adapter->hw;
1569 int j = adapter->rx_ring[rxr].reg_idx;
1572 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
1573 if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
1578 if (k >= IXGBE_MAX_RX_DESC_POLL) {
1579 hw_dbg(hw, "RXDCTL.ENABLE on Rx queue %d "
1580 "not set within the polling period\n", rxr);
1583 ixgbevf_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
1584 (adapter->rx_ring[rxr].count - 1));
1587 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1589 /* Only save pre-reset stats if there are some */
1590 if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1591 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1592 adapter->stats.base_vfgprc;
1593 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1594 adapter->stats.base_vfgptc;
1595 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1596 adapter->stats.base_vfgorc;
1597 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1598 adapter->stats.base_vfgotc;
1599 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1600 adapter->stats.base_vfmprc;
1604 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1606 struct ixgbe_hw *hw = &adapter->hw;
1608 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1609 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1610 adapter->stats.last_vfgorc |=
1611 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1612 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1613 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1614 adapter->stats.last_vfgotc |=
1615 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1616 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1618 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1619 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1620 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1621 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1622 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
1625 static int ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
1627 struct net_device *netdev = adapter->netdev;
1628 struct ixgbe_hw *hw = &adapter->hw;
1630 int num_rx_rings = adapter->num_rx_queues;
1633 for (i = 0; i < adapter->num_tx_queues; i++) {
1634 j = adapter->tx_ring[i].reg_idx;
1635 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1636 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
1637 txdctl |= (8 << 16);
1638 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1641 for (i = 0; i < adapter->num_tx_queues; i++) {
1642 j = adapter->tx_ring[i].reg_idx;
1643 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1644 txdctl |= IXGBE_TXDCTL_ENABLE;
1645 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1648 for (i = 0; i < num_rx_rings; i++) {
1649 j = adapter->rx_ring[i].reg_idx;
1650 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1651 rxdctl |= IXGBE_RXDCTL_ENABLE;
1652 if (hw->mac.type == ixgbe_mac_X540_vf) {
1653 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
1654 rxdctl |= ((netdev->mtu + ETH_HLEN + ETH_FCS_LEN) |
1655 IXGBE_RXDCTL_RLPML_EN);
1657 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1658 ixgbevf_rx_desc_queue_enable(adapter, i);
1661 ixgbevf_configure_msix(adapter);
1663 if (hw->mac.ops.set_rar) {
1664 if (is_valid_ether_addr(hw->mac.addr))
1665 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
1667 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
1670 clear_bit(__IXGBEVF_DOWN, &adapter->state);
1671 ixgbevf_napi_enable_all(adapter);
1673 /* enable transmits */
1674 netif_tx_start_all_queues(netdev);
1676 ixgbevf_save_reset_stats(adapter);
1677 ixgbevf_init_last_counter_stats(adapter);
1679 /* bring the link up in the watchdog, this could race with our first
1680 * link up interrupt but shouldn't be a problem */
1681 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1682 adapter->link_check_timeout = jiffies;
1683 mod_timer(&adapter->watchdog_timer, jiffies);
1687 int ixgbevf_up(struct ixgbevf_adapter *adapter)
1690 struct ixgbe_hw *hw = &adapter->hw;
1692 ixgbevf_configure(adapter);
1694 err = ixgbevf_up_complete(adapter);
1696 /* clear any pending interrupts, may auto mask */
1697 IXGBE_READ_REG(hw, IXGBE_VTEICR);
1699 ixgbevf_irq_enable(adapter, true, true);
1705 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
1706 * @adapter: board private structure
1707 * @rx_ring: ring to free buffers from
1709 static void ixgbevf_clean_rx_ring(struct ixgbevf_adapter *adapter,
1710 struct ixgbevf_ring *rx_ring)
1712 struct pci_dev *pdev = adapter->pdev;
1716 if (!rx_ring->rx_buffer_info)
1719 /* Free all the Rx ring sk_buffs */
1720 for (i = 0; i < rx_ring->count; i++) {
1721 struct ixgbevf_rx_buffer *rx_buffer_info;
1723 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1724 if (rx_buffer_info->dma) {
1725 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
1726 rx_ring->rx_buf_len,
1728 rx_buffer_info->dma = 0;
1730 if (rx_buffer_info->skb) {
1731 struct sk_buff *skb = rx_buffer_info->skb;
1732 rx_buffer_info->skb = NULL;
1734 struct sk_buff *this = skb;
1736 dev_kfree_skb(this);
1739 if (!rx_buffer_info->page)
1741 dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
1742 PAGE_SIZE / 2, DMA_FROM_DEVICE);
1743 rx_buffer_info->page_dma = 0;
1744 put_page(rx_buffer_info->page);
1745 rx_buffer_info->page = NULL;
1746 rx_buffer_info->page_offset = 0;
1749 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
1750 memset(rx_ring->rx_buffer_info, 0, size);
1752 /* Zero out the descriptor ring */
1753 memset(rx_ring->desc, 0, rx_ring->size);
1755 rx_ring->next_to_clean = 0;
1756 rx_ring->next_to_use = 0;
1759 writel(0, adapter->hw.hw_addr + rx_ring->head);
1761 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1765 * ixgbevf_clean_tx_ring - Free Tx Buffers
1766 * @adapter: board private structure
1767 * @tx_ring: ring to be cleaned
1769 static void ixgbevf_clean_tx_ring(struct ixgbevf_adapter *adapter,
1770 struct ixgbevf_ring *tx_ring)
1772 struct ixgbevf_tx_buffer *tx_buffer_info;
1776 if (!tx_ring->tx_buffer_info)
1779 /* Free all the Tx ring sk_buffs */
1781 for (i = 0; i < tx_ring->count; i++) {
1782 tx_buffer_info = &tx_ring->tx_buffer_info[i];
1783 ixgbevf_unmap_and_free_tx_resource(adapter, tx_buffer_info);
1786 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
1787 memset(tx_ring->tx_buffer_info, 0, size);
1789 memset(tx_ring->desc, 0, tx_ring->size);
1791 tx_ring->next_to_use = 0;
1792 tx_ring->next_to_clean = 0;
1795 writel(0, adapter->hw.hw_addr + tx_ring->head);
1797 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1801 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
1802 * @adapter: board private structure
1804 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
1808 for (i = 0; i < adapter->num_rx_queues; i++)
1809 ixgbevf_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1813 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
1814 * @adapter: board private structure
1816 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
1820 for (i = 0; i < adapter->num_tx_queues; i++)
1821 ixgbevf_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1824 void ixgbevf_down(struct ixgbevf_adapter *adapter)
1826 struct net_device *netdev = adapter->netdev;
1827 struct ixgbe_hw *hw = &adapter->hw;
1831 /* signal that we are down to the interrupt handler */
1832 set_bit(__IXGBEVF_DOWN, &adapter->state);
1833 /* disable receives */
1835 netif_tx_disable(netdev);
1839 netif_tx_stop_all_queues(netdev);
1841 ixgbevf_irq_disable(adapter);
1843 ixgbevf_napi_disable_all(adapter);
1845 del_timer_sync(&adapter->watchdog_timer);
1846 /* can't call flush scheduled work here because it can deadlock
1847 * if linkwatch_event tries to acquire the rtnl_lock which we are
1849 while (adapter->flags & IXGBE_FLAG_IN_WATCHDOG_TASK)
1852 /* disable transmits in the hardware now that interrupts are off */
1853 for (i = 0; i < adapter->num_tx_queues; i++) {
1854 j = adapter->tx_ring[i].reg_idx;
1855 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1856 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j),
1857 (txdctl & ~IXGBE_TXDCTL_ENABLE));
1860 netif_carrier_off(netdev);
1862 if (!pci_channel_offline(adapter->pdev))
1863 ixgbevf_reset(adapter);
1865 ixgbevf_clean_all_tx_rings(adapter);
1866 ixgbevf_clean_all_rx_rings(adapter);
1869 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
1871 struct ixgbe_hw *hw = &adapter->hw;
1873 WARN_ON(in_interrupt());
1875 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
1879 * Check if PF is up before re-init. If not then skip until
1880 * later when the PF is up and ready to service requests from
1881 * the VF via mailbox. If the VF is up and running then the
1882 * watchdog task will continue to schedule reset tasks until
1883 * the PF is up and running.
1885 if (!hw->mac.ops.reset_hw(hw)) {
1886 ixgbevf_down(adapter);
1887 ixgbevf_up(adapter);
1890 clear_bit(__IXGBEVF_RESETTING, &adapter->state);
1893 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
1895 struct ixgbe_hw *hw = &adapter->hw;
1896 struct net_device *netdev = adapter->netdev;
1898 if (hw->mac.ops.reset_hw(hw))
1899 hw_dbg(hw, "PF still resetting\n");
1901 hw->mac.ops.init_hw(hw);
1903 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1904 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1906 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1911 static void ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
1914 int err, vector_threshold;
1916 /* We'll want at least 3 (vector_threshold):
1919 * 3) Other (Link Status Change, etc.)
1921 vector_threshold = MIN_MSIX_COUNT;
1923 /* The more we get, the more we will assign to Tx/Rx Cleanup
1924 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
1925 * Right now, we simply care about how many we'll get; we'll
1926 * set them up later while requesting irq's.
1928 while (vectors >= vector_threshold) {
1929 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
1931 if (!err) /* Success in acquiring all requested vectors. */
1934 vectors = 0; /* Nasty failure, quit now */
1935 else /* err == number of vectors we should try again with */
1939 if (vectors < vector_threshold) {
1940 /* Can't allocate enough MSI-X interrupts? Oh well.
1941 * This just means we'll go with either a single MSI
1942 * vector or fall back to legacy interrupts.
1944 hw_dbg(&adapter->hw,
1945 "Unable to allocate MSI-X interrupts\n");
1946 kfree(adapter->msix_entries);
1947 adapter->msix_entries = NULL;
1950 * Adjust for only the vectors we'll use, which is minimum
1951 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
1952 * vectors we were allocated.
1954 adapter->num_msix_vectors = vectors;
1959 * ixgbevf_set_num_queues: Allocate queues for device, feature dependent
1960 * @adapter: board private structure to initialize
1962 * This is the top level queue allocation routine. The order here is very
1963 * important, starting with the "most" number of features turned on at once,
1964 * and ending with the smallest set of features. This way large combinations
1965 * can be allocated if they're turned on, and smaller combinations are the
1966 * fallthrough conditions.
1969 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
1971 /* Start with base case */
1972 adapter->num_rx_queues = 1;
1973 adapter->num_tx_queues = 1;
1974 adapter->num_rx_pools = adapter->num_rx_queues;
1975 adapter->num_rx_queues_per_pool = 1;
1979 * ixgbevf_alloc_queues - Allocate memory for all rings
1980 * @adapter: board private structure to initialize
1982 * We allocate one ring per queue at run-time since we don't know the
1983 * number of queues at compile-time. The polling_netdev array is
1984 * intended for Multiqueue, but should work fine with a single queue.
1986 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
1990 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1991 sizeof(struct ixgbevf_ring), GFP_KERNEL);
1992 if (!adapter->tx_ring)
1993 goto err_tx_ring_allocation;
1995 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1996 sizeof(struct ixgbevf_ring), GFP_KERNEL);
1997 if (!adapter->rx_ring)
1998 goto err_rx_ring_allocation;
2000 for (i = 0; i < adapter->num_tx_queues; i++) {
2001 adapter->tx_ring[i].count = adapter->tx_ring_count;
2002 adapter->tx_ring[i].queue_index = i;
2003 adapter->tx_ring[i].reg_idx = i;
2006 for (i = 0; i < adapter->num_rx_queues; i++) {
2007 adapter->rx_ring[i].count = adapter->rx_ring_count;
2008 adapter->rx_ring[i].queue_index = i;
2009 adapter->rx_ring[i].reg_idx = i;
2014 err_rx_ring_allocation:
2015 kfree(adapter->tx_ring);
2016 err_tx_ring_allocation:
2021 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2022 * @adapter: board private structure to initialize
2024 * Attempt to configure the interrupts using the best available
2025 * capabilities of the hardware and the kernel.
2027 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2030 int vector, v_budget;
2033 * It's easy to be greedy for MSI-X vectors, but it really
2034 * doesn't do us much good if we have a lot more vectors
2035 * than CPU's. So let's be conservative and only ask for
2036 * (roughly) twice the number of vectors as there are CPU's.
2038 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
2039 (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
2041 /* A failure in MSI-X entry allocation isn't fatal, but it does
2042 * mean we disable MSI-X capabilities of the adapter. */
2043 adapter->msix_entries = kcalloc(v_budget,
2044 sizeof(struct msix_entry), GFP_KERNEL);
2045 if (!adapter->msix_entries) {
2050 for (vector = 0; vector < v_budget; vector++)
2051 adapter->msix_entries[vector].entry = vector;
2053 ixgbevf_acquire_msix_vectors(adapter, v_budget);
2060 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2061 * @adapter: board private structure to initialize
2063 * We allocate one q_vector per queue interrupt. If allocation fails we
2066 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2068 int q_idx, num_q_vectors;
2069 struct ixgbevf_q_vector *q_vector;
2071 int (*poll)(struct napi_struct *, int);
2073 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2074 napi_vectors = adapter->num_rx_queues;
2075 poll = &ixgbevf_clean_rxonly;
2077 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2078 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2081 q_vector->adapter = adapter;
2082 q_vector->v_idx = q_idx;
2083 q_vector->eitr = adapter->eitr_param;
2084 if (q_idx < napi_vectors)
2085 netif_napi_add(adapter->netdev, &q_vector->napi,
2087 adapter->q_vector[q_idx] = q_vector;
2095 q_vector = adapter->q_vector[q_idx];
2096 netif_napi_del(&q_vector->napi);
2098 adapter->q_vector[q_idx] = NULL;
2104 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2105 * @adapter: board private structure to initialize
2107 * This function frees the memory allocated to the q_vectors. In addition if
2108 * NAPI is enabled it will delete any references to the NAPI struct prior
2109 * to freeing the q_vector.
2111 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2113 int q_idx, num_q_vectors;
2116 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2117 napi_vectors = adapter->num_rx_queues;
2119 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2120 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2122 adapter->q_vector[q_idx] = NULL;
2123 if (q_idx < napi_vectors)
2124 netif_napi_del(&q_vector->napi);
2130 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2131 * @adapter: board private structure
2134 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2136 pci_disable_msix(adapter->pdev);
2137 kfree(adapter->msix_entries);
2138 adapter->msix_entries = NULL;
2142 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2143 * @adapter: board private structure to initialize
2146 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2150 /* Number of supported queues */
2151 ixgbevf_set_num_queues(adapter);
2153 err = ixgbevf_set_interrupt_capability(adapter);
2155 hw_dbg(&adapter->hw,
2156 "Unable to setup interrupt capabilities\n");
2157 goto err_set_interrupt;
2160 err = ixgbevf_alloc_q_vectors(adapter);
2162 hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
2164 goto err_alloc_q_vectors;
2167 err = ixgbevf_alloc_queues(adapter);
2169 printk(KERN_ERR "Unable to allocate memory for queues\n");
2170 goto err_alloc_queues;
2173 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
2174 "Tx Queue count = %u\n",
2175 (adapter->num_rx_queues > 1) ? "Enabled" :
2176 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2178 set_bit(__IXGBEVF_DOWN, &adapter->state);
2182 ixgbevf_free_q_vectors(adapter);
2183 err_alloc_q_vectors:
2184 ixgbevf_reset_interrupt_capability(adapter);
2190 * ixgbevf_sw_init - Initialize general software structures
2191 * (struct ixgbevf_adapter)
2192 * @adapter: board private structure to initialize
2194 * ixgbevf_sw_init initializes the Adapter private data structure.
2195 * Fields are initialized based on PCI device information and
2196 * OS network device settings (MTU size).
2198 static int __devinit ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2200 struct ixgbe_hw *hw = &adapter->hw;
2201 struct pci_dev *pdev = adapter->pdev;
2204 /* PCI config space info */
2206 hw->vendor_id = pdev->vendor;
2207 hw->device_id = pdev->device;
2208 hw->revision_id = pdev->revision;
2209 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2210 hw->subsystem_device_id = pdev->subsystem_device;
2212 hw->mbx.ops.init_params(hw);
2213 hw->mac.max_tx_queues = MAX_TX_QUEUES;
2214 hw->mac.max_rx_queues = MAX_RX_QUEUES;
2215 err = hw->mac.ops.reset_hw(hw);
2217 dev_info(&pdev->dev,
2218 "PF still in reset state, assigning new address\n");
2219 dev_hw_addr_random(adapter->netdev, hw->mac.addr);
2221 err = hw->mac.ops.init_hw(hw);
2223 printk(KERN_ERR "init_shared_code failed: %d\n", err);
2228 /* Enable dynamic interrupt throttling rates */
2229 adapter->eitr_param = 20000;
2230 adapter->itr_setting = 1;
2232 /* set defaults for eitr in MegaBytes */
2233 adapter->eitr_low = 10;
2234 adapter->eitr_high = 20;
2236 /* set default ring sizes */
2237 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2238 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2240 /* enable rx csum by default */
2241 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
2243 set_bit(__IXGBEVF_DOWN, &adapter->state);
2249 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
2251 u32 current_counter = IXGBE_READ_REG(hw, reg); \
2252 if (current_counter < last_counter) \
2253 counter += 0x100000000LL; \
2254 last_counter = current_counter; \
2255 counter &= 0xFFFFFFFF00000000LL; \
2256 counter |= current_counter; \
2259 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2261 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
2262 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
2263 u64 current_counter = (current_counter_msb << 32) | \
2264 current_counter_lsb; \
2265 if (current_counter < last_counter) \
2266 counter += 0x1000000000LL; \
2267 last_counter = current_counter; \
2268 counter &= 0xFFFFFFF000000000LL; \
2269 counter |= current_counter; \
2272 * ixgbevf_update_stats - Update the board statistics counters.
2273 * @adapter: board private structure
2275 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2277 struct ixgbe_hw *hw = &adapter->hw;
2279 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2280 adapter->stats.vfgprc);
2281 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2282 adapter->stats.vfgptc);
2283 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2284 adapter->stats.last_vfgorc,
2285 adapter->stats.vfgorc);
2286 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2287 adapter->stats.last_vfgotc,
2288 adapter->stats.vfgotc);
2289 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2290 adapter->stats.vfmprc);
2292 /* Fill out the OS statistics structure */
2293 adapter->netdev->stats.multicast = adapter->stats.vfmprc -
2294 adapter->stats.base_vfmprc;
2298 * ixgbevf_watchdog - Timer Call-back
2299 * @data: pointer to adapter cast into an unsigned long
2301 static void ixgbevf_watchdog(unsigned long data)
2303 struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2304 struct ixgbe_hw *hw = &adapter->hw;
2309 * Do the watchdog outside of interrupt context due to the lovely
2310 * delays that some of the newer hardware requires
2313 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
2314 goto watchdog_short_circuit;
2316 /* get one bit for every active tx/rx interrupt vector */
2317 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2318 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2319 if (qv->rxr_count || qv->txr_count)
2323 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, (u32)eics);
2325 watchdog_short_circuit:
2326 schedule_work(&adapter->watchdog_task);
2330 * ixgbevf_tx_timeout - Respond to a Tx Hang
2331 * @netdev: network interface device structure
2333 static void ixgbevf_tx_timeout(struct net_device *netdev)
2335 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2337 /* Do the reset outside of interrupt context */
2338 schedule_work(&adapter->reset_task);
2341 static void ixgbevf_reset_task(struct work_struct *work)
2343 struct ixgbevf_adapter *adapter;
2344 adapter = container_of(work, struct ixgbevf_adapter, reset_task);
2346 /* If we're already down or resetting, just bail */
2347 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2348 test_bit(__IXGBEVF_RESETTING, &adapter->state))
2351 adapter->tx_timeout_count++;
2353 ixgbevf_reinit_locked(adapter);
2357 * ixgbevf_watchdog_task - worker thread to bring link up
2358 * @work: pointer to work_struct containing our data
2360 static void ixgbevf_watchdog_task(struct work_struct *work)
2362 struct ixgbevf_adapter *adapter = container_of(work,
2363 struct ixgbevf_adapter,
2365 struct net_device *netdev = adapter->netdev;
2366 struct ixgbe_hw *hw = &adapter->hw;
2367 u32 link_speed = adapter->link_speed;
2368 bool link_up = adapter->link_up;
2370 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
2373 * Always check the link on the watchdog because we have
2376 if (hw->mac.ops.check_link) {
2377 if ((hw->mac.ops.check_link(hw, &link_speed,
2378 &link_up, false)) != 0) {
2379 adapter->link_up = link_up;
2380 adapter->link_speed = link_speed;
2381 netif_carrier_off(netdev);
2382 netif_tx_stop_all_queues(netdev);
2383 schedule_work(&adapter->reset_task);
2387 /* always assume link is up, if no check link
2389 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
2392 adapter->link_up = link_up;
2393 adapter->link_speed = link_speed;
2396 if (!netif_carrier_ok(netdev)) {
2397 hw_dbg(&adapter->hw, "NIC Link is Up, %u Gbps\n",
2398 (link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2400 netif_carrier_on(netdev);
2401 netif_tx_wake_all_queues(netdev);
2404 adapter->link_up = false;
2405 adapter->link_speed = 0;
2406 if (netif_carrier_ok(netdev)) {
2407 hw_dbg(&adapter->hw, "NIC Link is Down\n");
2408 netif_carrier_off(netdev);
2409 netif_tx_stop_all_queues(netdev);
2413 ixgbevf_update_stats(adapter);
2416 /* Reset the timer */
2417 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
2418 mod_timer(&adapter->watchdog_timer,
2419 round_jiffies(jiffies + (2 * HZ)));
2421 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
2425 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2426 * @adapter: board private structure
2427 * @tx_ring: Tx descriptor ring for a specific queue
2429 * Free all transmit software resources
2431 void ixgbevf_free_tx_resources(struct ixgbevf_adapter *adapter,
2432 struct ixgbevf_ring *tx_ring)
2434 struct pci_dev *pdev = adapter->pdev;
2436 ixgbevf_clean_tx_ring(adapter, tx_ring);
2438 vfree(tx_ring->tx_buffer_info);
2439 tx_ring->tx_buffer_info = NULL;
2441 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2444 tx_ring->desc = NULL;
2448 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2449 * @adapter: board private structure
2451 * Free all transmit software resources
2453 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2457 for (i = 0; i < adapter->num_tx_queues; i++)
2458 if (adapter->tx_ring[i].desc)
2459 ixgbevf_free_tx_resources(adapter,
2460 &adapter->tx_ring[i]);
2465 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2466 * @adapter: board private structure
2467 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2469 * Return 0 on success, negative on failure
2471 int ixgbevf_setup_tx_resources(struct ixgbevf_adapter *adapter,
2472 struct ixgbevf_ring *tx_ring)
2474 struct pci_dev *pdev = adapter->pdev;
2477 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2478 tx_ring->tx_buffer_info = vzalloc(size);
2479 if (!tx_ring->tx_buffer_info)
2482 /* round up to nearest 4K */
2483 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2484 tx_ring->size = ALIGN(tx_ring->size, 4096);
2486 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
2487 &tx_ring->dma, GFP_KERNEL);
2491 tx_ring->next_to_use = 0;
2492 tx_ring->next_to_clean = 0;
2493 tx_ring->work_limit = tx_ring->count;
2497 vfree(tx_ring->tx_buffer_info);
2498 tx_ring->tx_buffer_info = NULL;
2499 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2500 "descriptor ring\n");
2505 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2506 * @adapter: board private structure
2508 * If this function returns with an error, then it's possible one or
2509 * more of the rings is populated (while the rest are not). It is the
2510 * callers duty to clean those orphaned rings.
2512 * Return 0 on success, negative on failure
2514 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2518 for (i = 0; i < adapter->num_tx_queues; i++) {
2519 err = ixgbevf_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2522 hw_dbg(&adapter->hw,
2523 "Allocation for Tx Queue %u failed\n", i);
2531 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2532 * @adapter: board private structure
2533 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2535 * Returns 0 on success, negative on failure
2537 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
2538 struct ixgbevf_ring *rx_ring)
2540 struct pci_dev *pdev = adapter->pdev;
2543 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2544 rx_ring->rx_buffer_info = vzalloc(size);
2545 if (!rx_ring->rx_buffer_info) {
2546 hw_dbg(&adapter->hw,
2547 "Unable to vmalloc buffer memory for "
2548 "the receive descriptor ring\n");
2552 /* Round up to nearest 4K */
2553 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2554 rx_ring->size = ALIGN(rx_ring->size, 4096);
2556 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
2557 &rx_ring->dma, GFP_KERNEL);
2559 if (!rx_ring->desc) {
2560 hw_dbg(&adapter->hw,
2561 "Unable to allocate memory for "
2562 "the receive descriptor ring\n");
2563 vfree(rx_ring->rx_buffer_info);
2564 rx_ring->rx_buffer_info = NULL;
2568 rx_ring->next_to_clean = 0;
2569 rx_ring->next_to_use = 0;
2577 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
2578 * @adapter: board private structure
2580 * If this function returns with an error, then it's possible one or
2581 * more of the rings is populated (while the rest are not). It is the
2582 * callers duty to clean those orphaned rings.
2584 * Return 0 on success, negative on failure
2586 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
2590 for (i = 0; i < adapter->num_rx_queues; i++) {
2591 err = ixgbevf_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2594 hw_dbg(&adapter->hw,
2595 "Allocation for Rx Queue %u failed\n", i);
2602 * ixgbevf_free_rx_resources - Free Rx Resources
2603 * @adapter: board private structure
2604 * @rx_ring: ring to clean the resources from
2606 * Free all receive software resources
2608 void ixgbevf_free_rx_resources(struct ixgbevf_adapter *adapter,
2609 struct ixgbevf_ring *rx_ring)
2611 struct pci_dev *pdev = adapter->pdev;
2613 ixgbevf_clean_rx_ring(adapter, rx_ring);
2615 vfree(rx_ring->rx_buffer_info);
2616 rx_ring->rx_buffer_info = NULL;
2618 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2621 rx_ring->desc = NULL;
2625 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
2626 * @adapter: board private structure
2628 * Free all receive software resources
2630 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
2634 for (i = 0; i < adapter->num_rx_queues; i++)
2635 if (adapter->rx_ring[i].desc)
2636 ixgbevf_free_rx_resources(adapter,
2637 &adapter->rx_ring[i]);
2641 * ixgbevf_open - Called when a network interface is made active
2642 * @netdev: network interface device structure
2644 * Returns 0 on success, negative value on failure
2646 * The open entry point is called when a network interface is made
2647 * active by the system (IFF_UP). At this point all resources needed
2648 * for transmit and receive operations are allocated, the interrupt
2649 * handler is registered with the OS, the watchdog timer is started,
2650 * and the stack is notified that the interface is ready.
2652 static int ixgbevf_open(struct net_device *netdev)
2654 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2655 struct ixgbe_hw *hw = &adapter->hw;
2658 /* disallow open during test */
2659 if (test_bit(__IXGBEVF_TESTING, &adapter->state))
2662 if (hw->adapter_stopped) {
2663 ixgbevf_reset(adapter);
2664 /* if adapter is still stopped then PF isn't up and
2665 * the vf can't start. */
2666 if (hw->adapter_stopped) {
2667 err = IXGBE_ERR_MBX;
2668 printk(KERN_ERR "Unable to start - perhaps the PF"
2669 " Driver isn't up yet\n");
2670 goto err_setup_reset;
2674 /* allocate transmit descriptors */
2675 err = ixgbevf_setup_all_tx_resources(adapter);
2679 /* allocate receive descriptors */
2680 err = ixgbevf_setup_all_rx_resources(adapter);
2684 ixgbevf_configure(adapter);
2687 * Map the Tx/Rx rings to the vectors we were allotted.
2688 * if request_irq will be called in this function map_rings
2689 * must be called *before* up_complete
2691 ixgbevf_map_rings_to_vectors(adapter);
2693 err = ixgbevf_up_complete(adapter);
2697 /* clear any pending interrupts, may auto mask */
2698 IXGBE_READ_REG(hw, IXGBE_VTEICR);
2699 err = ixgbevf_request_irq(adapter);
2703 ixgbevf_irq_enable(adapter, true, true);
2708 ixgbevf_down(adapter);
2710 ixgbevf_free_irq(adapter);
2712 ixgbevf_free_all_rx_resources(adapter);
2714 ixgbevf_free_all_tx_resources(adapter);
2715 ixgbevf_reset(adapter);
2723 * ixgbevf_close - Disables a network interface
2724 * @netdev: network interface device structure
2726 * Returns 0, this is not allowed to fail
2728 * The close entry point is called when an interface is de-activated
2729 * by the OS. The hardware is still under the drivers control, but
2730 * needs to be disabled. A global MAC reset is issued to stop the
2731 * hardware, and all transmit and receive resources are freed.
2733 static int ixgbevf_close(struct net_device *netdev)
2735 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2737 ixgbevf_down(adapter);
2738 ixgbevf_free_irq(adapter);
2740 ixgbevf_free_all_tx_resources(adapter);
2741 ixgbevf_free_all_rx_resources(adapter);
2746 static int ixgbevf_tso(struct ixgbevf_adapter *adapter,
2747 struct ixgbevf_ring *tx_ring,
2748 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2750 struct ixgbe_adv_tx_context_desc *context_desc;
2753 struct ixgbevf_tx_buffer *tx_buffer_info;
2754 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
2755 u32 mss_l4len_idx, l4len;
2757 if (skb_is_gso(skb)) {
2758 if (skb_header_cloned(skb)) {
2759 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2763 l4len = tcp_hdrlen(skb);
2766 if (skb->protocol == htons(ETH_P_IP)) {
2767 struct iphdr *iph = ip_hdr(skb);
2770 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2774 adapter->hw_tso_ctxt++;
2775 } else if (skb_is_gso_v6(skb)) {
2776 ipv6_hdr(skb)->payload_len = 0;
2777 tcp_hdr(skb)->check =
2778 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2779 &ipv6_hdr(skb)->daddr,
2781 adapter->hw_tso6_ctxt++;
2784 i = tx_ring->next_to_use;
2786 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2787 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
2789 /* VLAN MACLEN IPLEN */
2790 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2792 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
2793 vlan_macip_lens |= ((skb_network_offset(skb)) <<
2794 IXGBE_ADVTXD_MACLEN_SHIFT);
2795 *hdr_len += skb_network_offset(skb);
2797 (skb_transport_header(skb) - skb_network_header(skb));
2799 (skb_transport_header(skb) - skb_network_header(skb));
2800 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
2801 context_desc->seqnum_seed = 0;
2803 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2804 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
2805 IXGBE_ADVTXD_DTYP_CTXT);
2807 if (skb->protocol == htons(ETH_P_IP))
2808 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
2809 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
2810 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
2814 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
2815 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
2816 /* use index 1 for TSO */
2817 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
2818 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2820 tx_buffer_info->time_stamp = jiffies;
2821 tx_buffer_info->next_to_watch = i;
2824 if (i == tx_ring->count)
2826 tx_ring->next_to_use = i;
2834 static bool ixgbevf_tx_csum(struct ixgbevf_adapter *adapter,
2835 struct ixgbevf_ring *tx_ring,
2836 struct sk_buff *skb, u32 tx_flags)
2838 struct ixgbe_adv_tx_context_desc *context_desc;
2840 struct ixgbevf_tx_buffer *tx_buffer_info;
2841 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
2843 if (skb->ip_summed == CHECKSUM_PARTIAL ||
2844 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
2845 i = tx_ring->next_to_use;
2846 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2847 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
2849 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2850 vlan_macip_lens |= (tx_flags &
2851 IXGBE_TX_FLAGS_VLAN_MASK);
2852 vlan_macip_lens |= (skb_network_offset(skb) <<
2853 IXGBE_ADVTXD_MACLEN_SHIFT);
2854 if (skb->ip_summed == CHECKSUM_PARTIAL)
2855 vlan_macip_lens |= (skb_transport_header(skb) -
2856 skb_network_header(skb));
2858 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
2859 context_desc->seqnum_seed = 0;
2861 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
2862 IXGBE_ADVTXD_DTYP_CTXT);
2864 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2865 switch (skb->protocol) {
2866 case __constant_htons(ETH_P_IP):
2867 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
2868 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2870 IXGBE_ADVTXD_TUCMD_L4T_TCP;
2872 case __constant_htons(ETH_P_IPV6):
2873 /* XXX what about other V6 headers?? */
2874 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2876 IXGBE_ADVTXD_TUCMD_L4T_TCP;
2879 if (unlikely(net_ratelimit())) {
2881 "partial checksum but "
2889 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
2890 /* use index zero for tx checksum offload */
2891 context_desc->mss_l4len_idx = 0;
2893 tx_buffer_info->time_stamp = jiffies;
2894 tx_buffer_info->next_to_watch = i;
2896 adapter->hw_csum_tx_good++;
2898 if (i == tx_ring->count)
2900 tx_ring->next_to_use = i;
2908 static int ixgbevf_tx_map(struct ixgbevf_adapter *adapter,
2909 struct ixgbevf_ring *tx_ring,
2910 struct sk_buff *skb, u32 tx_flags,
2913 struct pci_dev *pdev = adapter->pdev;
2914 struct ixgbevf_tx_buffer *tx_buffer_info;
2916 unsigned int total = skb->len;
2917 unsigned int offset = 0, size;
2919 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
2923 i = tx_ring->next_to_use;
2925 len = min(skb_headlen(skb), total);
2927 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2928 size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2930 tx_buffer_info->length = size;
2931 tx_buffer_info->mapped_as_page = false;
2932 tx_buffer_info->dma = dma_map_single(&adapter->pdev->dev,
2934 size, DMA_TO_DEVICE);
2935 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
2937 tx_buffer_info->time_stamp = jiffies;
2938 tx_buffer_info->next_to_watch = i;
2945 if (i == tx_ring->count)
2949 for (f = 0; f < nr_frags; f++) {
2950 struct skb_frag_struct *frag;
2952 frag = &skb_shinfo(skb)->frags[f];
2953 len = min((unsigned int)frag->size, total);
2954 offset = frag->page_offset;
2957 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2958 size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2960 tx_buffer_info->length = size;
2961 tx_buffer_info->dma = dma_map_page(&adapter->pdev->dev,
2966 tx_buffer_info->mapped_as_page = true;
2967 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
2969 tx_buffer_info->time_stamp = jiffies;
2970 tx_buffer_info->next_to_watch = i;
2977 if (i == tx_ring->count)
2985 i = tx_ring->count - 1;
2988 tx_ring->tx_buffer_info[i].skb = skb;
2989 tx_ring->tx_buffer_info[first].next_to_watch = i;
2994 dev_err(&pdev->dev, "TX DMA map failed\n");
2996 /* clear timestamp and dma mappings for failed tx_buffer_info map */
2997 tx_buffer_info->dma = 0;
2998 tx_buffer_info->time_stamp = 0;
2999 tx_buffer_info->next_to_watch = 0;
3002 /* clear timestamp and dma mappings for remaining portion of packet */
3003 while (count >= 0) {
3007 i += tx_ring->count;
3008 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3009 ixgbevf_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3015 static void ixgbevf_tx_queue(struct ixgbevf_adapter *adapter,
3016 struct ixgbevf_ring *tx_ring, int tx_flags,
3017 int count, u32 paylen, u8 hdr_len)
3019 union ixgbe_adv_tx_desc *tx_desc = NULL;
3020 struct ixgbevf_tx_buffer *tx_buffer_info;
3021 u32 olinfo_status = 0, cmd_type_len = 0;
3024 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
3026 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
3028 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
3030 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3031 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
3033 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
3034 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
3036 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
3037 IXGBE_ADVTXD_POPTS_SHIFT;
3039 /* use index 1 context for tso */
3040 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3041 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3042 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
3043 IXGBE_ADVTXD_POPTS_SHIFT;
3045 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3046 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
3047 IXGBE_ADVTXD_POPTS_SHIFT;
3049 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
3051 i = tx_ring->next_to_use;
3053 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3054 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
3055 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
3056 tx_desc->read.cmd_type_len =
3057 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
3058 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3060 if (i == tx_ring->count)
3064 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
3067 * Force memory writes to complete before letting h/w
3068 * know there are new descriptors to fetch. (Only
3069 * applicable for weak-ordered memory model archs,
3074 tx_ring->next_to_use = i;
3075 writel(i, adapter->hw.hw_addr + tx_ring->tail);
3078 static int __ixgbevf_maybe_stop_tx(struct net_device *netdev,
3079 struct ixgbevf_ring *tx_ring, int size)
3081 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3083 netif_stop_subqueue(netdev, tx_ring->queue_index);
3084 /* Herbert's original patch had:
3085 * smp_mb__after_netif_stop_queue();
3086 * but since that doesn't exist yet, just open code it. */
3089 /* We need to check again in a case another CPU has just
3090 * made room available. */
3091 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
3094 /* A reprieve! - use start_queue because it doesn't call schedule */
3095 netif_start_subqueue(netdev, tx_ring->queue_index);
3096 ++adapter->restart_queue;
3100 static int ixgbevf_maybe_stop_tx(struct net_device *netdev,
3101 struct ixgbevf_ring *tx_ring, int size)
3103 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
3105 return __ixgbevf_maybe_stop_tx(netdev, tx_ring, size);
3108 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3110 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3111 struct ixgbevf_ring *tx_ring;
3113 unsigned int tx_flags = 0;
3120 tx_ring = &adapter->tx_ring[r_idx];
3122 if (vlan_tx_tag_present(skb)) {
3123 tx_flags |= vlan_tx_tag_get(skb);
3124 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3125 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3128 /* four things can cause us to need a context descriptor */
3129 if (skb_is_gso(skb) ||
3130 (skb->ip_summed == CHECKSUM_PARTIAL) ||
3131 (tx_flags & IXGBE_TX_FLAGS_VLAN))
3134 count += TXD_USE_COUNT(skb_headlen(skb));
3135 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3136 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3138 if (ixgbevf_maybe_stop_tx(netdev, tx_ring, count)) {
3140 return NETDEV_TX_BUSY;
3143 first = tx_ring->next_to_use;
3145 if (skb->protocol == htons(ETH_P_IP))
3146 tx_flags |= IXGBE_TX_FLAGS_IPV4;
3147 tso = ixgbevf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
3149 dev_kfree_skb_any(skb);
3150 return NETDEV_TX_OK;
3154 tx_flags |= IXGBE_TX_FLAGS_TSO;
3155 else if (ixgbevf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
3156 (skb->ip_summed == CHECKSUM_PARTIAL))
3157 tx_flags |= IXGBE_TX_FLAGS_CSUM;
3159 ixgbevf_tx_queue(adapter, tx_ring, tx_flags,
3160 ixgbevf_tx_map(adapter, tx_ring, skb, tx_flags, first),
3163 ixgbevf_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
3165 return NETDEV_TX_OK;
3169 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3170 * @netdev: network interface device structure
3171 * @p: pointer to an address structure
3173 * Returns 0 on success, negative on failure
3175 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3177 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3178 struct ixgbe_hw *hw = &adapter->hw;
3179 struct sockaddr *addr = p;
3181 if (!is_valid_ether_addr(addr->sa_data))
3182 return -EADDRNOTAVAIL;
3184 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3185 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3187 if (hw->mac.ops.set_rar)
3188 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3194 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3195 * @netdev: network interface device structure
3196 * @new_mtu: new value for maximum frame size
3198 * Returns 0 on success, negative on failure
3200 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3202 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3203 struct ixgbe_hw *hw = &adapter->hw;
3204 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3205 int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3208 if (adapter->hw.mac.type == ixgbe_mac_X540_vf)
3209 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3211 /* MTU < 68 is an error and causes problems on some kernels */
3212 if ((new_mtu < 68) || (max_frame > max_possible_frame))
3215 hw_dbg(&adapter->hw, "changing MTU from %d to %d\n",
3216 netdev->mtu, new_mtu);
3217 /* must set new MTU before calling down or up */
3218 netdev->mtu = new_mtu;
3220 msg[0] = IXGBE_VF_SET_LPE;
3222 hw->mbx.ops.write_posted(hw, msg, 2);
3224 if (netif_running(netdev))
3225 ixgbevf_reinit_locked(adapter);
3230 static void ixgbevf_shutdown(struct pci_dev *pdev)
3232 struct net_device *netdev = pci_get_drvdata(pdev);
3233 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3235 netif_device_detach(netdev);
3237 if (netif_running(netdev)) {
3238 ixgbevf_down(adapter);
3239 ixgbevf_free_irq(adapter);
3240 ixgbevf_free_all_tx_resources(adapter);
3241 ixgbevf_free_all_rx_resources(adapter);
3245 pci_save_state(pdev);
3248 pci_disable_device(pdev);
3251 static const struct net_device_ops ixgbe_netdev_ops = {
3252 .ndo_open = &ixgbevf_open,
3253 .ndo_stop = &ixgbevf_close,
3254 .ndo_start_xmit = &ixgbevf_xmit_frame,
3255 .ndo_set_rx_mode = &ixgbevf_set_rx_mode,
3256 .ndo_set_multicast_list = &ixgbevf_set_rx_mode,
3257 .ndo_validate_addr = eth_validate_addr,
3258 .ndo_set_mac_address = &ixgbevf_set_mac,
3259 .ndo_change_mtu = &ixgbevf_change_mtu,
3260 .ndo_tx_timeout = &ixgbevf_tx_timeout,
3261 .ndo_vlan_rx_register = &ixgbevf_vlan_rx_register,
3262 .ndo_vlan_rx_add_vid = &ixgbevf_vlan_rx_add_vid,
3263 .ndo_vlan_rx_kill_vid = &ixgbevf_vlan_rx_kill_vid,
3266 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3268 dev->netdev_ops = &ixgbe_netdev_ops;
3269 ixgbevf_set_ethtool_ops(dev);
3270 dev->watchdog_timeo = 5 * HZ;
3274 * ixgbevf_probe - Device Initialization Routine
3275 * @pdev: PCI device information struct
3276 * @ent: entry in ixgbevf_pci_tbl
3278 * Returns 0 on success, negative on failure
3280 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3281 * The OS initialization, configuring of the adapter private structure,
3282 * and a hardware reset occur.
3284 static int __devinit ixgbevf_probe(struct pci_dev *pdev,
3285 const struct pci_device_id *ent)
3287 struct net_device *netdev;
3288 struct ixgbevf_adapter *adapter = NULL;
3289 struct ixgbe_hw *hw = NULL;
3290 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3291 static int cards_found;
3292 int err, pci_using_dac;
3294 err = pci_enable_device(pdev);
3298 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
3299 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
3302 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
3304 err = dma_set_coherent_mask(&pdev->dev,
3307 dev_err(&pdev->dev, "No usable DMA "
3308 "configuration, aborting\n");
3315 err = pci_request_regions(pdev, ixgbevf_driver_name);
3317 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3321 pci_set_master(pdev);
3324 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3327 netdev = alloc_etherdev(sizeof(struct ixgbevf_adapter));
3331 goto err_alloc_etherdev;
3334 SET_NETDEV_DEV(netdev, &pdev->dev);
3336 pci_set_drvdata(pdev, netdev);
3337 adapter = netdev_priv(netdev);
3339 adapter->netdev = netdev;
3340 adapter->pdev = pdev;
3343 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
3346 * call save state here in standalone driver because it relies on
3347 * adapter struct to exist, and needs to call netdev_priv
3349 pci_save_state(pdev);
3351 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3352 pci_resource_len(pdev, 0));
3358 ixgbevf_assign_netdev_ops(netdev);
3360 adapter->bd_number = cards_found;
3363 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3364 hw->mac.type = ii->mac;
3366 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3367 sizeof(struct ixgbe_mac_operations));
3369 adapter->flags &= ~IXGBE_FLAG_RX_PS_CAPABLE;
3370 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3371 adapter->flags |= IXGBE_FLAG_RX_1BUF_CAPABLE;
3373 /* setup the private structure */
3374 err = ixgbevf_sw_init(adapter);
3376 netdev->features = NETIF_F_SG |
3378 NETIF_F_HW_VLAN_TX |
3379 NETIF_F_HW_VLAN_RX |
3380 NETIF_F_HW_VLAN_FILTER;
3382 netdev->features |= NETIF_F_IPV6_CSUM;
3383 netdev->features |= NETIF_F_TSO;
3384 netdev->features |= NETIF_F_TSO6;
3385 netdev->features |= NETIF_F_GRO;
3386 netdev->vlan_features |= NETIF_F_TSO;
3387 netdev->vlan_features |= NETIF_F_TSO6;
3388 netdev->vlan_features |= NETIF_F_IP_CSUM;
3389 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3390 netdev->vlan_features |= NETIF_F_SG;
3393 netdev->features |= NETIF_F_HIGHDMA;
3395 /* The HW MAC address was set and/or determined in sw_init */
3396 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
3397 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
3399 if (!is_valid_ether_addr(netdev->dev_addr)) {
3400 printk(KERN_ERR "invalid MAC address\n");
3405 init_timer(&adapter->watchdog_timer);
3406 adapter->watchdog_timer.function = ixgbevf_watchdog;
3407 adapter->watchdog_timer.data = (unsigned long)adapter;
3409 INIT_WORK(&adapter->reset_task, ixgbevf_reset_task);
3410 INIT_WORK(&adapter->watchdog_task, ixgbevf_watchdog_task);
3412 err = ixgbevf_init_interrupt_scheme(adapter);
3416 /* pick up the PCI bus settings for reporting later */
3417 if (hw->mac.ops.get_bus_info)
3418 hw->mac.ops.get_bus_info(hw);
3420 strcpy(netdev->name, "eth%d");
3422 err = register_netdev(netdev);
3426 adapter->netdev_registered = true;
3428 netif_carrier_off(netdev);
3430 ixgbevf_init_last_counter_stats(adapter);
3432 /* print the MAC address */
3433 hw_dbg(hw, "%2.2x:%2.2x:%2.2x:%2.2x:%2.2x:%2.2x\n",
3434 netdev->dev_addr[0],
3435 netdev->dev_addr[1],
3436 netdev->dev_addr[2],
3437 netdev->dev_addr[3],
3438 netdev->dev_addr[4],
3439 netdev->dev_addr[5]);
3441 hw_dbg(hw, "MAC: %d\n", hw->mac.type);
3443 hw_dbg(hw, "LRO is disabled\n");
3445 hw_dbg(hw, "Intel(R) 82599 Virtual Function\n");
3451 ixgbevf_reset_interrupt_capability(adapter);
3452 iounmap(hw->hw_addr);
3454 free_netdev(netdev);
3456 pci_release_regions(pdev);
3459 pci_disable_device(pdev);
3464 * ixgbevf_remove - Device Removal Routine
3465 * @pdev: PCI device information struct
3467 * ixgbevf_remove is called by the PCI subsystem to alert the driver
3468 * that it should release a PCI device. The could be caused by a
3469 * Hot-Plug event, or because the driver is going to be removed from
3472 static void __devexit ixgbevf_remove(struct pci_dev *pdev)
3474 struct net_device *netdev = pci_get_drvdata(pdev);
3475 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3477 set_bit(__IXGBEVF_DOWN, &adapter->state);
3479 del_timer_sync(&adapter->watchdog_timer);
3481 cancel_work_sync(&adapter->reset_task);
3482 cancel_work_sync(&adapter->watchdog_task);
3484 if (adapter->netdev_registered) {
3485 unregister_netdev(netdev);
3486 adapter->netdev_registered = false;
3489 ixgbevf_reset_interrupt_capability(adapter);
3491 iounmap(adapter->hw.hw_addr);
3492 pci_release_regions(pdev);
3494 hw_dbg(&adapter->hw, "Remove complete\n");
3496 kfree(adapter->tx_ring);
3497 kfree(adapter->rx_ring);
3499 free_netdev(netdev);
3501 pci_disable_device(pdev);
3504 static struct pci_driver ixgbevf_driver = {
3505 .name = ixgbevf_driver_name,
3506 .id_table = ixgbevf_pci_tbl,
3507 .probe = ixgbevf_probe,
3508 .remove = __devexit_p(ixgbevf_remove),
3509 .shutdown = ixgbevf_shutdown,
3513 * ixgbevf_init_module - Driver Registration Routine
3515 * ixgbevf_init_module is the first routine called when the driver is
3516 * loaded. All it does is register with the PCI subsystem.
3518 static int __init ixgbevf_init_module(void)
3521 printk(KERN_INFO "ixgbevf: %s - version %s\n", ixgbevf_driver_string,
3522 ixgbevf_driver_version);
3524 printk(KERN_INFO "%s\n", ixgbevf_copyright);
3526 ret = pci_register_driver(&ixgbevf_driver);
3530 module_init(ixgbevf_init_module);
3533 * ixgbevf_exit_module - Driver Exit Cleanup Routine
3535 * ixgbevf_exit_module is called just before the driver is removed
3538 static void __exit ixgbevf_exit_module(void)
3540 pci_unregister_driver(&ixgbevf_driver);
3545 * ixgbevf_get_hw_dev_name - return device name string
3546 * used by hardware layer to print debugging information
3548 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
3550 struct ixgbevf_adapter *adapter = hw->back;
3551 return adapter->netdev->name;
3555 module_exit(ixgbevf_exit_module);
3557 /* ixgbevf_main.c */