1 /*******************************************************************************
3 Intel 82599 Virtual Function driver
4 Copyright(c) 1999 - 2012 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
29 /******************************************************************************
30 Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
31 ******************************************************************************/
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35 #include <linux/types.h>
36 #include <linux/bitops.h>
37 #include <linux/module.h>
38 #include <linux/pci.h>
39 #include <linux/netdevice.h>
40 #include <linux/vmalloc.h>
41 #include <linux/string.h>
44 #include <linux/tcp.h>
45 #include <linux/sctp.h>
46 #include <linux/ipv6.h>
47 #include <linux/slab.h>
48 #include <net/checksum.h>
49 #include <net/ip6_checksum.h>
50 #include <linux/ethtool.h>
52 #include <linux/if_vlan.h>
53 #include <linux/prefetch.h>
57 const char ixgbevf_driver_name[] = "ixgbevf";
58 static const char ixgbevf_driver_string[] =
59 "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
61 #define DRV_VERSION "2.7.12-k"
62 const char ixgbevf_driver_version[] = DRV_VERSION;
63 static char ixgbevf_copyright[] =
64 "Copyright (c) 2009 - 2012 Intel Corporation.";
66 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
67 [board_82599_vf] = &ixgbevf_82599_vf_info,
68 [board_X540_vf] = &ixgbevf_X540_vf_info,
71 /* ixgbevf_pci_tbl - PCI Device ID Table
73 * Wildcard entries (PCI_ANY_ID) should come last
74 * Last entry must be all 0s
76 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
77 * Class, Class Mask, private data (not used) }
79 static DEFINE_PCI_DEVICE_TABLE(ixgbevf_pci_tbl) = {
80 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
81 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
82 /* required last entry */
85 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
87 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
88 MODULE_DESCRIPTION("Intel(R) 82599 Virtual Function Driver");
89 MODULE_LICENSE("GPL");
90 MODULE_VERSION(DRV_VERSION);
92 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
93 static int debug = -1;
94 module_param(debug, int, 0);
95 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
98 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
99 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
101 static inline void ixgbevf_release_rx_desc(struct ixgbe_hw *hw,
102 struct ixgbevf_ring *rx_ring,
106 * Force memory writes to complete before letting h/w
107 * know there are new descriptors to fetch. (Only
108 * applicable for weak-ordered memory model archs,
112 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rx_ring->reg_idx), val);
116 * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
117 * @adapter: pointer to adapter struct
118 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
119 * @queue: queue to map the corresponding interrupt to
120 * @msix_vector: the vector to map to the corresponding queue
122 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
123 u8 queue, u8 msix_vector)
126 struct ixgbe_hw *hw = &adapter->hw;
127 if (direction == -1) {
129 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
130 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
133 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
135 /* tx or rx causes */
136 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
137 index = ((16 * (queue & 1)) + (8 * direction));
138 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
139 ivar &= ~(0xFF << index);
140 ivar |= (msix_vector << index);
141 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
145 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
146 struct ixgbevf_tx_buffer
149 if (tx_buffer_info->dma) {
150 if (tx_buffer_info->mapped_as_page)
151 dma_unmap_page(tx_ring->dev,
153 tx_buffer_info->length,
156 dma_unmap_single(tx_ring->dev,
158 tx_buffer_info->length,
160 tx_buffer_info->dma = 0;
162 if (tx_buffer_info->skb) {
163 dev_kfree_skb_any(tx_buffer_info->skb);
164 tx_buffer_info->skb = NULL;
166 tx_buffer_info->time_stamp = 0;
167 /* tx_buffer_info must be completely set up in the transmit path */
170 #define IXGBE_MAX_TXD_PWR 14
171 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
173 /* Tx Descriptors needed, worst case */
174 #define TXD_USE_COUNT(S) DIV_ROUND_UP((S), IXGBE_MAX_DATA_PER_TXD)
175 #define DESC_NEEDED (MAX_SKB_FRAGS + 4)
177 static void ixgbevf_tx_timeout(struct net_device *netdev);
180 * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
181 * @q_vector: board private structure
182 * @tx_ring: tx ring to clean
184 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
185 struct ixgbevf_ring *tx_ring)
187 struct ixgbevf_adapter *adapter = q_vector->adapter;
188 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
189 struct ixgbevf_tx_buffer *tx_buffer_info;
190 unsigned int i, count = 0;
191 unsigned int total_bytes = 0, total_packets = 0;
193 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
196 i = tx_ring->next_to_clean;
197 tx_buffer_info = &tx_ring->tx_buffer_info[i];
198 eop_desc = tx_buffer_info->next_to_watch;
201 bool cleaned = false;
203 /* if next_to_watch is not set then there is no work pending */
207 /* prevent any other reads prior to eop_desc */
208 read_barrier_depends();
210 /* if DD is not set pending work has not been completed */
211 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
214 /* clear next_to_watch to prevent false hangs */
215 tx_buffer_info->next_to_watch = NULL;
217 for ( ; !cleaned; count++) {
219 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
220 cleaned = (tx_desc == eop_desc);
221 skb = tx_buffer_info->skb;
223 if (cleaned && skb) {
224 unsigned int segs, bytecount;
226 /* gso_segs is currently only valid for tcp */
227 segs = skb_shinfo(skb)->gso_segs ?: 1;
228 /* multiply data chunks by size of headers */
229 bytecount = ((segs - 1) * skb_headlen(skb)) +
231 total_packets += segs;
232 total_bytes += bytecount;
235 ixgbevf_unmap_and_free_tx_resource(tx_ring,
238 tx_desc->wb.status = 0;
241 if (i == tx_ring->count)
244 tx_buffer_info = &tx_ring->tx_buffer_info[i];
247 eop_desc = tx_buffer_info->next_to_watch;
248 } while (count < tx_ring->count);
250 tx_ring->next_to_clean = i;
252 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
253 if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
254 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
255 /* Make sure that anybody stopping the queue after this
256 * sees the new next_to_clean.
259 if (__netif_subqueue_stopped(tx_ring->netdev,
260 tx_ring->queue_index) &&
261 !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
262 netif_wake_subqueue(tx_ring->netdev,
263 tx_ring->queue_index);
264 ++adapter->restart_queue;
268 u64_stats_update_begin(&tx_ring->syncp);
269 tx_ring->total_bytes += total_bytes;
270 tx_ring->total_packets += total_packets;
271 u64_stats_update_end(&tx_ring->syncp);
272 q_vector->tx.total_bytes += total_bytes;
273 q_vector->tx.total_packets += total_packets;
275 return count < tx_ring->count;
279 * ixgbevf_receive_skb - Send a completed packet up the stack
280 * @q_vector: structure containing interrupt and ring information
281 * @skb: packet to send up
282 * @status: hardware indication of status of receive
283 * @rx_desc: rx descriptor
285 static void ixgbevf_receive_skb(struct ixgbevf_q_vector *q_vector,
286 struct sk_buff *skb, u8 status,
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 (is_vlan && test_bit(tag & VLAN_VID_MASK, adapter->active_vlans))
294 __vlan_hwaccel_put_tag(skb, tag);
296 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
297 napi_gro_receive(&q_vector->napi, skb);
303 * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
304 * @ring: pointer to Rx descriptor ring structure
305 * @status_err: hardware indication of status of receive
306 * @skb: skb currently being received and modified
308 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
309 u32 status_err, struct sk_buff *skb)
311 skb_checksum_none_assert(skb);
313 /* Rx csum disabled */
314 if (!(ring->netdev->features & NETIF_F_RXCSUM))
317 /* if IP and error */
318 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
319 (status_err & IXGBE_RXDADV_ERR_IPE)) {
320 ring->hw_csum_rx_error++;
324 if (!(status_err & IXGBE_RXD_STAT_L4CS))
327 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
328 ring->hw_csum_rx_error++;
332 /* It must be a TCP or UDP packet with a valid checksum */
333 skb->ip_summed = CHECKSUM_UNNECESSARY;
334 ring->hw_csum_rx_good++;
338 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
339 * @adapter: address of board private structure
341 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_adapter *adapter,
342 struct ixgbevf_ring *rx_ring,
345 struct pci_dev *pdev = adapter->pdev;
346 union ixgbe_adv_rx_desc *rx_desc;
347 struct ixgbevf_rx_buffer *bi;
348 unsigned int i = rx_ring->next_to_use;
350 bi = &rx_ring->rx_buffer_info[i];
352 while (cleaned_count--) {
353 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
358 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
359 rx_ring->rx_buf_len);
361 adapter->alloc_rx_buff_failed++;
366 bi->dma = dma_map_single(&pdev->dev, skb->data,
369 if (dma_mapping_error(&pdev->dev, bi->dma)) {
372 dev_err(&pdev->dev, "RX DMA map failed\n");
376 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
379 if (i == rx_ring->count)
381 bi = &rx_ring->rx_buffer_info[i];
385 if (rx_ring->next_to_use != i) {
386 rx_ring->next_to_use = i;
387 ixgbevf_release_rx_desc(&adapter->hw, rx_ring, i);
391 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
394 struct ixgbe_hw *hw = &adapter->hw;
396 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
399 static bool ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
400 struct ixgbevf_ring *rx_ring,
403 struct ixgbevf_adapter *adapter = q_vector->adapter;
404 struct pci_dev *pdev = adapter->pdev;
405 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
406 struct ixgbevf_rx_buffer *rx_buffer_info, *next_buffer;
410 int cleaned_count = 0;
411 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
413 i = rx_ring->next_to_clean;
414 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
415 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
416 rx_buffer_info = &rx_ring->rx_buffer_info[i];
418 while (staterr & IXGBE_RXD_STAT_DD) {
423 rmb(); /* read descriptor and rx_buffer_info after status DD */
424 len = le16_to_cpu(rx_desc->wb.upper.length);
425 skb = rx_buffer_info->skb;
426 prefetch(skb->data - NET_IP_ALIGN);
427 rx_buffer_info->skb = NULL;
429 if (rx_buffer_info->dma) {
430 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
433 rx_buffer_info->dma = 0;
438 if (i == rx_ring->count)
441 next_rxd = IXGBEVF_RX_DESC(rx_ring, i);
445 next_buffer = &rx_ring->rx_buffer_info[i];
447 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
448 skb->next = next_buffer->skb;
449 IXGBE_CB(skb->next)->prev = skb;
450 adapter->non_eop_descs++;
454 /* we should not be chaining buffers, if we did drop the skb */
455 if (IXGBE_CB(skb)->prev) {
457 struct sk_buff *this = skb;
458 skb = IXGBE_CB(skb)->prev;
464 /* ERR_MASK will only have valid bits if EOP set */
465 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
466 dev_kfree_skb_irq(skb);
470 ixgbevf_rx_checksum(rx_ring, staterr, skb);
472 /* probably a little skewed due to removing CRC */
473 total_rx_bytes += skb->len;
477 * Work around issue of some types of VM to VM loop back
478 * packets not getting split correctly
480 if (staterr & IXGBE_RXD_STAT_LB) {
481 u32 header_fixup_len = skb_headlen(skb);
482 if (header_fixup_len < 14)
483 skb_push(skb, header_fixup_len);
485 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
487 /* Workaround hardware that can't do proper VEPA multicast
490 if ((skb->pkt_type & (PACKET_BROADCAST | PACKET_MULTICAST)) &&
491 !(compare_ether_addr(adapter->netdev->dev_addr,
492 eth_hdr(skb)->h_source))) {
493 dev_kfree_skb_irq(skb);
497 ixgbevf_receive_skb(q_vector, skb, staterr, rx_desc);
500 rx_desc->wb.upper.status_error = 0;
502 /* return some buffers to hardware, one at a time is too slow */
503 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
504 ixgbevf_alloc_rx_buffers(adapter, rx_ring,
509 /* use prefetched values */
511 rx_buffer_info = &rx_ring->rx_buffer_info[i];
513 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
516 rx_ring->next_to_clean = i;
517 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
520 ixgbevf_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
522 u64_stats_update_begin(&rx_ring->syncp);
523 rx_ring->total_packets += total_rx_packets;
524 rx_ring->total_bytes += total_rx_bytes;
525 u64_stats_update_end(&rx_ring->syncp);
526 q_vector->rx.total_packets += total_rx_packets;
527 q_vector->rx.total_bytes += total_rx_bytes;
533 * ixgbevf_poll - NAPI polling calback
534 * @napi: napi struct with our devices info in it
535 * @budget: amount of work driver is allowed to do this pass, in packets
537 * This function will clean more than one or more rings associated with a
540 static int ixgbevf_poll(struct napi_struct *napi, int budget)
542 struct ixgbevf_q_vector *q_vector =
543 container_of(napi, struct ixgbevf_q_vector, napi);
544 struct ixgbevf_adapter *adapter = q_vector->adapter;
545 struct ixgbevf_ring *ring;
547 bool clean_complete = true;
549 ixgbevf_for_each_ring(ring, q_vector->tx)
550 clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
552 /* attempt to distribute budget to each queue fairly, but don't allow
553 * the budget to go below 1 because we'll exit polling */
554 if (q_vector->rx.count > 1)
555 per_ring_budget = max(budget/q_vector->rx.count, 1);
557 per_ring_budget = budget;
559 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
560 ixgbevf_for_each_ring(ring, q_vector->rx)
561 clean_complete &= ixgbevf_clean_rx_irq(q_vector, ring,
563 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
565 /* If all work not completed, return budget and keep polling */
568 /* all work done, exit the polling mode */
570 if (adapter->rx_itr_setting & 1)
571 ixgbevf_set_itr(q_vector);
572 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
573 ixgbevf_irq_enable_queues(adapter,
574 1 << q_vector->v_idx);
580 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
581 * @q_vector: structure containing interrupt and ring information
583 static void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
585 struct ixgbevf_adapter *adapter = q_vector->adapter;
586 struct ixgbe_hw *hw = &adapter->hw;
587 int v_idx = q_vector->v_idx;
588 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
591 * set the WDIS bit to not clear the timer bits and cause an
592 * immediate assertion of the interrupt
594 itr_reg |= IXGBE_EITR_CNT_WDIS;
596 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
600 * ixgbevf_configure_msix - Configure MSI-X hardware
601 * @adapter: board private structure
603 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
606 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
608 struct ixgbevf_q_vector *q_vector;
609 int q_vectors, v_idx;
611 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
612 adapter->eims_enable_mask = 0;
615 * Populate the IVAR table and set the ITR values to the
616 * corresponding register.
618 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
619 struct ixgbevf_ring *ring;
620 q_vector = adapter->q_vector[v_idx];
622 ixgbevf_for_each_ring(ring, q_vector->rx)
623 ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
625 ixgbevf_for_each_ring(ring, q_vector->tx)
626 ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
628 if (q_vector->tx.ring && !q_vector->rx.ring) {
630 if (adapter->tx_itr_setting == 1)
631 q_vector->itr = IXGBE_10K_ITR;
633 q_vector->itr = adapter->tx_itr_setting;
635 /* rx or rx/tx vector */
636 if (adapter->rx_itr_setting == 1)
637 q_vector->itr = IXGBE_20K_ITR;
639 q_vector->itr = adapter->rx_itr_setting;
642 /* add q_vector eims value to global eims_enable_mask */
643 adapter->eims_enable_mask |= 1 << v_idx;
645 ixgbevf_write_eitr(q_vector);
648 ixgbevf_set_ivar(adapter, -1, 1, v_idx);
649 /* setup eims_other and add value to global eims_enable_mask */
650 adapter->eims_other = 1 << v_idx;
651 adapter->eims_enable_mask |= adapter->eims_other;
658 latency_invalid = 255
662 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
663 * @q_vector: structure containing interrupt and ring information
664 * @ring_container: structure containing ring performance data
666 * Stores a new ITR value based on packets and byte
667 * counts during the last interrupt. The advantage of per interrupt
668 * computation is faster updates and more accurate ITR for the current
669 * traffic pattern. Constants in this function were computed
670 * based on theoretical maximum wire speed and thresholds were set based
671 * on testing data as well as attempting to minimize response time
672 * while increasing bulk throughput.
674 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
675 struct ixgbevf_ring_container *ring_container)
677 int bytes = ring_container->total_bytes;
678 int packets = ring_container->total_packets;
681 u8 itr_setting = ring_container->itr;
686 /* simple throttlerate management
687 * 0-20MB/s lowest (100000 ints/s)
688 * 20-100MB/s low (20000 ints/s)
689 * 100-1249MB/s bulk (8000 ints/s)
691 /* what was last interrupt timeslice? */
692 timepassed_us = q_vector->itr >> 2;
693 bytes_perint = bytes / timepassed_us; /* bytes/usec */
695 switch (itr_setting) {
697 if (bytes_perint > 10)
698 itr_setting = low_latency;
701 if (bytes_perint > 20)
702 itr_setting = bulk_latency;
703 else if (bytes_perint <= 10)
704 itr_setting = lowest_latency;
707 if (bytes_perint <= 20)
708 itr_setting = low_latency;
712 /* clear work counters since we have the values we need */
713 ring_container->total_bytes = 0;
714 ring_container->total_packets = 0;
716 /* write updated itr to ring container */
717 ring_container->itr = itr_setting;
720 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
722 u32 new_itr = q_vector->itr;
725 ixgbevf_update_itr(q_vector, &q_vector->tx);
726 ixgbevf_update_itr(q_vector, &q_vector->rx);
728 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
730 switch (current_itr) {
731 /* counts and packets in update_itr are dependent on these numbers */
733 new_itr = IXGBE_100K_ITR;
736 new_itr = IXGBE_20K_ITR;
740 new_itr = IXGBE_8K_ITR;
744 if (new_itr != q_vector->itr) {
745 /* do an exponential smoothing */
746 new_itr = (10 * new_itr * q_vector->itr) /
747 ((9 * new_itr) + q_vector->itr);
749 /* save the algorithm value here */
750 q_vector->itr = new_itr;
752 ixgbevf_write_eitr(q_vector);
756 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
758 struct ixgbevf_adapter *adapter = data;
759 struct pci_dev *pdev = adapter->pdev;
760 struct ixgbe_hw *hw = &adapter->hw;
762 bool got_ack = false;
764 hw->mac.get_link_status = 1;
765 if (!hw->mbx.ops.check_for_ack(hw))
768 if (!hw->mbx.ops.check_for_msg(hw)) {
769 hw->mbx.ops.read(hw, &msg, 1);
771 if ((msg & IXGBE_MBVFICR_VFREQ_MASK) == IXGBE_PF_CONTROL_MSG) {
772 mod_timer(&adapter->watchdog_timer,
773 round_jiffies(jiffies + 1));
774 adapter->link_up = false;
777 if (msg & IXGBE_VT_MSGTYPE_NACK)
779 "Last Request of type %2.2x to PF Nacked\n",
781 hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFSTS;
784 /* checking for the ack clears the PFACK bit. Place
785 * it back in the v2p_mailbox cache so that anyone
786 * polling for an ack will not miss it
789 hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFACK;
791 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
797 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
799 * @data: pointer to our q_vector struct for this interrupt vector
801 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
803 struct ixgbevf_q_vector *q_vector = data;
805 /* EIAM disabled interrupts (on this vector) for us */
806 if (q_vector->rx.ring || q_vector->tx.ring)
807 napi_schedule(&q_vector->napi);
812 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
815 struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
817 a->rx_ring[r_idx].next = q_vector->rx.ring;
818 q_vector->rx.ring = &a->rx_ring[r_idx];
819 q_vector->rx.count++;
822 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
825 struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
827 a->tx_ring[t_idx].next = q_vector->tx.ring;
828 q_vector->tx.ring = &a->tx_ring[t_idx];
829 q_vector->tx.count++;
833 * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
834 * @adapter: board private structure to initialize
836 * This function maps descriptor rings to the queue-specific vectors
837 * we were allotted through the MSI-X enabling code. Ideally, we'd have
838 * one vector per ring/queue, but on a constrained vector budget, we
839 * group the rings as "efficiently" as possible. You would add new
840 * mapping configurations in here.
842 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
846 int rxr_idx = 0, txr_idx = 0;
847 int rxr_remaining = adapter->num_rx_queues;
848 int txr_remaining = adapter->num_tx_queues;
853 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
856 * The ideal configuration...
857 * We have enough vectors to map one per queue.
859 if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
860 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
861 map_vector_to_rxq(adapter, v_start, rxr_idx);
863 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
864 map_vector_to_txq(adapter, v_start, txr_idx);
869 * If we don't have enough vectors for a 1-to-1
870 * mapping, we'll have to group them so there are
871 * multiple queues per vector.
873 /* Re-adjusting *qpv takes care of the remainder. */
874 for (i = v_start; i < q_vectors; i++) {
875 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
876 for (j = 0; j < rqpv; j++) {
877 map_vector_to_rxq(adapter, i, rxr_idx);
882 for (i = v_start; i < q_vectors; i++) {
883 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
884 for (j = 0; j < tqpv; j++) {
885 map_vector_to_txq(adapter, i, txr_idx);
896 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
897 * @adapter: board private structure
899 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
900 * interrupts from the kernel.
902 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
904 struct net_device *netdev = adapter->netdev;
905 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
909 for (vector = 0; vector < q_vectors; vector++) {
910 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
911 struct msix_entry *entry = &adapter->msix_entries[vector];
913 if (q_vector->tx.ring && q_vector->rx.ring) {
914 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
915 "%s-%s-%d", netdev->name, "TxRx", ri++);
917 } else if (q_vector->rx.ring) {
918 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
919 "%s-%s-%d", netdev->name, "rx", ri++);
920 } else if (q_vector->tx.ring) {
921 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
922 "%s-%s-%d", netdev->name, "tx", ti++);
924 /* skip this unused q_vector */
927 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
928 q_vector->name, q_vector);
931 "request_irq failed for MSIX interrupt "
933 goto free_queue_irqs;
937 err = request_irq(adapter->msix_entries[vector].vector,
938 &ixgbevf_msix_other, 0, netdev->name, adapter);
941 "request_irq for msix_other failed: %d\n", err);
942 goto free_queue_irqs;
950 free_irq(adapter->msix_entries[vector].vector,
951 adapter->q_vector[vector]);
953 /* This failure is non-recoverable - it indicates the system is
954 * out of MSIX vector resources and the VF driver cannot run
955 * without them. Set the number of msix vectors to zero
956 * indicating that not enough can be allocated. The error
957 * will be returned to the user indicating device open failed.
958 * Any further attempts to force the driver to open will also
959 * fail. The only way to recover is to unload the driver and
960 * reload it again. If the system has recovered some MSIX
961 * vectors then it may succeed.
963 adapter->num_msix_vectors = 0;
967 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
969 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
971 for (i = 0; i < q_vectors; i++) {
972 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
973 q_vector->rx.ring = NULL;
974 q_vector->tx.ring = NULL;
975 q_vector->rx.count = 0;
976 q_vector->tx.count = 0;
981 * ixgbevf_request_irq - initialize interrupts
982 * @adapter: board private structure
984 * Attempts to configure interrupts using the best available
985 * capabilities of the hardware and kernel.
987 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
991 err = ixgbevf_request_msix_irqs(adapter);
995 "request_irq failed, Error %d\n", err);
1000 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1004 q_vectors = adapter->num_msix_vectors;
1007 free_irq(adapter->msix_entries[i].vector, adapter);
1010 for (; i >= 0; i--) {
1011 /* free only the irqs that were actually requested */
1012 if (!adapter->q_vector[i]->rx.ring &&
1013 !adapter->q_vector[i]->tx.ring)
1016 free_irq(adapter->msix_entries[i].vector,
1017 adapter->q_vector[i]);
1020 ixgbevf_reset_q_vectors(adapter);
1024 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1025 * @adapter: board private structure
1027 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1029 struct ixgbe_hw *hw = &adapter->hw;
1032 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1033 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1034 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1036 IXGBE_WRITE_FLUSH(hw);
1038 for (i = 0; i < adapter->num_msix_vectors; i++)
1039 synchronize_irq(adapter->msix_entries[i].vector);
1043 * ixgbevf_irq_enable - Enable default interrupt generation settings
1044 * @adapter: board private structure
1046 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1048 struct ixgbe_hw *hw = &adapter->hw;
1050 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1051 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1052 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1056 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1057 * @adapter: board private structure
1059 * Configure the Tx unit of the MAC after a reset.
1061 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1064 struct ixgbe_hw *hw = &adapter->hw;
1065 u32 i, j, tdlen, txctrl;
1067 /* Setup the HW Tx Head and Tail descriptor pointers */
1068 for (i = 0; i < adapter->num_tx_queues; i++) {
1069 struct ixgbevf_ring *ring = &adapter->tx_ring[i];
1072 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1073 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1074 (tdba & DMA_BIT_MASK(32)));
1075 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1076 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j), tdlen);
1077 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(j), 0);
1078 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(j), 0);
1079 adapter->tx_ring[i].head = IXGBE_VFTDH(j);
1080 adapter->tx_ring[i].tail = IXGBE_VFTDT(j);
1081 /* Disable Tx Head Writeback RO bit, since this hoses
1082 * bookkeeping if things aren't delivered in order.
1084 txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1085 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1086 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1090 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1092 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1094 struct ixgbevf_ring *rx_ring;
1095 struct ixgbe_hw *hw = &adapter->hw;
1098 rx_ring = &adapter->rx_ring[index];
1100 srrctl = IXGBE_SRRCTL_DROP_EN;
1102 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1104 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
1105 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1107 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1110 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter)
1112 struct ixgbe_hw *hw = &adapter->hw;
1113 struct net_device *netdev = adapter->netdev;
1114 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1118 /* notify the PF of our intent to use this size of frame */
1119 ixgbevf_rlpml_set_vf(hw, max_frame);
1121 /* PF will allow an extra 4 bytes past for vlan tagged frames */
1122 max_frame += VLAN_HLEN;
1125 * Allocate buffer sizes that fit well into 32K and
1126 * take into account max frame size of 9.5K
1128 if ((hw->mac.type == ixgbe_mac_X540_vf) &&
1129 (max_frame <= MAXIMUM_ETHERNET_VLAN_SIZE))
1130 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1131 else if (max_frame <= IXGBEVF_RXBUFFER_2K)
1132 rx_buf_len = IXGBEVF_RXBUFFER_2K;
1133 else if (max_frame <= IXGBEVF_RXBUFFER_4K)
1134 rx_buf_len = IXGBEVF_RXBUFFER_4K;
1135 else if (max_frame <= IXGBEVF_RXBUFFER_8K)
1136 rx_buf_len = IXGBEVF_RXBUFFER_8K;
1138 rx_buf_len = IXGBEVF_RXBUFFER_10K;
1140 for (i = 0; i < adapter->num_rx_queues; i++)
1141 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1145 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1146 * @adapter: board private structure
1148 * Configure the Rx unit of the MAC after a reset.
1150 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1153 struct ixgbe_hw *hw = &adapter->hw;
1157 /* PSRTYPE must be initialized in 82599 */
1158 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, 0);
1160 /* set_rx_buffer_len must be called before ring initialization */
1161 ixgbevf_set_rx_buffer_len(adapter);
1163 rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1164 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1165 * the Base and Length of the Rx Descriptor Ring */
1166 for (i = 0; i < adapter->num_rx_queues; i++) {
1167 rdba = adapter->rx_ring[i].dma;
1168 j = adapter->rx_ring[i].reg_idx;
1169 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1170 (rdba & DMA_BIT_MASK(32)));
1171 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1172 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j), rdlen);
1173 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(j), 0);
1174 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(j), 0);
1175 adapter->rx_ring[i].head = IXGBE_VFRDH(j);
1176 adapter->rx_ring[i].tail = IXGBE_VFRDT(j);
1178 ixgbevf_configure_srrctl(adapter, j);
1182 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1184 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1185 struct ixgbe_hw *hw = &adapter->hw;
1188 spin_lock_bh(&adapter->mbx_lock);
1190 /* add VID to filter table */
1191 err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1193 spin_unlock_bh(&adapter->mbx_lock);
1195 /* translate error return types so error makes sense */
1196 if (err == IXGBE_ERR_MBX)
1199 if (err == IXGBE_ERR_INVALID_ARGUMENT)
1202 set_bit(vid, adapter->active_vlans);
1207 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1209 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1210 struct ixgbe_hw *hw = &adapter->hw;
1211 int err = -EOPNOTSUPP;
1213 spin_lock_bh(&adapter->mbx_lock);
1215 /* remove VID from filter table */
1216 err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1218 spin_unlock_bh(&adapter->mbx_lock);
1220 clear_bit(vid, adapter->active_vlans);
1225 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1229 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1230 ixgbevf_vlan_rx_add_vid(adapter->netdev, vid);
1233 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1235 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1236 struct ixgbe_hw *hw = &adapter->hw;
1239 if ((netdev_uc_count(netdev)) > 10) {
1240 pr_err("Too many unicast filters - No Space\n");
1244 if (!netdev_uc_empty(netdev)) {
1245 struct netdev_hw_addr *ha;
1246 netdev_for_each_uc_addr(ha, netdev) {
1247 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1252 * If the list is empty then send message to PF driver to
1253 * clear all macvlans on this VF.
1255 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1262 * ixgbevf_set_rx_mode - Multicast and unicast set
1263 * @netdev: network interface device structure
1265 * The set_rx_method entry point is called whenever the multicast address
1266 * list, unicast address list or the network interface flags are updated.
1267 * This routine is responsible for configuring the hardware for proper
1268 * multicast mode and configuring requested unicast filters.
1270 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1272 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1273 struct ixgbe_hw *hw = &adapter->hw;
1275 spin_lock_bh(&adapter->mbx_lock);
1277 /* reprogram multicast list */
1278 hw->mac.ops.update_mc_addr_list(hw, netdev);
1280 ixgbevf_write_uc_addr_list(netdev);
1282 spin_unlock_bh(&adapter->mbx_lock);
1285 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1288 struct ixgbevf_q_vector *q_vector;
1289 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1291 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1292 q_vector = adapter->q_vector[q_idx];
1293 napi_enable(&q_vector->napi);
1297 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1300 struct ixgbevf_q_vector *q_vector;
1301 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1303 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1304 q_vector = adapter->q_vector[q_idx];
1305 napi_disable(&q_vector->napi);
1309 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1311 struct net_device *netdev = adapter->netdev;
1314 ixgbevf_set_rx_mode(netdev);
1316 ixgbevf_restore_vlan(adapter);
1318 ixgbevf_configure_tx(adapter);
1319 ixgbevf_configure_rx(adapter);
1320 for (i = 0; i < adapter->num_rx_queues; i++) {
1321 struct ixgbevf_ring *ring = &adapter->rx_ring[i];
1322 ixgbevf_alloc_rx_buffers(adapter, ring,
1323 IXGBE_DESC_UNUSED(ring));
1327 #define IXGBE_MAX_RX_DESC_POLL 10
1328 static inline void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1331 struct ixgbe_hw *hw = &adapter->hw;
1332 int j = adapter->rx_ring[rxr].reg_idx;
1335 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
1336 if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
1341 if (k >= IXGBE_MAX_RX_DESC_POLL) {
1342 hw_dbg(hw, "RXDCTL.ENABLE on Rx queue %d "
1343 "not set within the polling period\n", rxr);
1346 ixgbevf_release_rx_desc(hw, &adapter->rx_ring[rxr],
1347 adapter->rx_ring[rxr].count - 1);
1350 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1352 /* Only save pre-reset stats if there are some */
1353 if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1354 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1355 adapter->stats.base_vfgprc;
1356 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1357 adapter->stats.base_vfgptc;
1358 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1359 adapter->stats.base_vfgorc;
1360 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1361 adapter->stats.base_vfgotc;
1362 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1363 adapter->stats.base_vfmprc;
1367 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1369 struct ixgbe_hw *hw = &adapter->hw;
1371 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1372 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1373 adapter->stats.last_vfgorc |=
1374 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1375 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1376 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1377 adapter->stats.last_vfgotc |=
1378 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1379 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1381 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1382 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1383 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1384 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1385 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
1388 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
1390 struct ixgbe_hw *hw = &adapter->hw;
1391 int api[] = { ixgbe_mbox_api_11,
1393 ixgbe_mbox_api_unknown };
1394 int err = 0, idx = 0;
1396 spin_lock_bh(&adapter->mbx_lock);
1398 while (api[idx] != ixgbe_mbox_api_unknown) {
1399 err = ixgbevf_negotiate_api_version(hw, api[idx]);
1405 spin_unlock_bh(&adapter->mbx_lock);
1408 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
1410 struct net_device *netdev = adapter->netdev;
1411 struct ixgbe_hw *hw = &adapter->hw;
1413 int num_rx_rings = adapter->num_rx_queues;
1416 for (i = 0; i < adapter->num_tx_queues; i++) {
1417 j = adapter->tx_ring[i].reg_idx;
1418 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1419 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
1420 txdctl |= (8 << 16);
1421 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1424 for (i = 0; i < adapter->num_tx_queues; i++) {
1425 j = adapter->tx_ring[i].reg_idx;
1426 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1427 txdctl |= IXGBE_TXDCTL_ENABLE;
1428 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1431 for (i = 0; i < num_rx_rings; i++) {
1432 j = adapter->rx_ring[i].reg_idx;
1433 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1434 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1435 if (hw->mac.type == ixgbe_mac_X540_vf) {
1436 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
1437 rxdctl |= ((netdev->mtu + ETH_HLEN + ETH_FCS_LEN) |
1438 IXGBE_RXDCTL_RLPML_EN);
1440 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1441 ixgbevf_rx_desc_queue_enable(adapter, i);
1444 ixgbevf_configure_msix(adapter);
1446 spin_lock_bh(&adapter->mbx_lock);
1448 if (is_valid_ether_addr(hw->mac.addr))
1449 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
1451 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
1453 spin_unlock_bh(&adapter->mbx_lock);
1455 clear_bit(__IXGBEVF_DOWN, &adapter->state);
1456 ixgbevf_napi_enable_all(adapter);
1458 /* enable transmits */
1459 netif_tx_start_all_queues(netdev);
1461 ixgbevf_save_reset_stats(adapter);
1462 ixgbevf_init_last_counter_stats(adapter);
1464 hw->mac.get_link_status = 1;
1465 mod_timer(&adapter->watchdog_timer, jiffies);
1468 static int ixgbevf_reset_queues(struct ixgbevf_adapter *adapter)
1470 struct ixgbe_hw *hw = &adapter->hw;
1471 struct ixgbevf_ring *rx_ring;
1472 unsigned int def_q = 0;
1473 unsigned int num_tcs = 0;
1474 unsigned int num_rx_queues = 1;
1477 spin_lock_bh(&adapter->mbx_lock);
1479 /* fetch queue configuration from the PF */
1480 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1482 spin_unlock_bh(&adapter->mbx_lock);
1488 /* update default Tx ring register index */
1489 adapter->tx_ring[0].reg_idx = def_q;
1491 /* we need as many queues as traffic classes */
1492 num_rx_queues = num_tcs;
1495 /* nothing to do if we have the correct number of queues */
1496 if (adapter->num_rx_queues == num_rx_queues)
1499 /* allocate new rings */
1500 rx_ring = kcalloc(num_rx_queues,
1501 sizeof(struct ixgbevf_ring), GFP_KERNEL);
1505 /* setup ring fields */
1506 for (i = 0; i < num_rx_queues; i++) {
1507 rx_ring[i].count = adapter->rx_ring_count;
1508 rx_ring[i].queue_index = i;
1509 rx_ring[i].reg_idx = i;
1510 rx_ring[i].dev = &adapter->pdev->dev;
1511 rx_ring[i].netdev = adapter->netdev;
1513 /* allocate resources on the ring */
1514 err = ixgbevf_setup_rx_resources(adapter, &rx_ring[i]);
1518 ixgbevf_free_rx_resources(adapter, &rx_ring[i]);
1525 /* free the existing rings and queues */
1526 ixgbevf_free_all_rx_resources(adapter);
1527 adapter->num_rx_queues = 0;
1528 kfree(adapter->rx_ring);
1530 /* move new rings into position on the adapter struct */
1531 adapter->rx_ring = rx_ring;
1532 adapter->num_rx_queues = num_rx_queues;
1534 /* reset ring to vector mapping */
1535 ixgbevf_reset_q_vectors(adapter);
1536 ixgbevf_map_rings_to_vectors(adapter);
1541 void ixgbevf_up(struct ixgbevf_adapter *adapter)
1543 struct ixgbe_hw *hw = &adapter->hw;
1545 ixgbevf_negotiate_api(adapter);
1547 ixgbevf_reset_queues(adapter);
1549 ixgbevf_configure(adapter);
1551 ixgbevf_up_complete(adapter);
1553 /* clear any pending interrupts, may auto mask */
1554 IXGBE_READ_REG(hw, IXGBE_VTEICR);
1556 ixgbevf_irq_enable(adapter);
1560 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
1561 * @adapter: board private structure
1562 * @rx_ring: ring to free buffers from
1564 static void ixgbevf_clean_rx_ring(struct ixgbevf_adapter *adapter,
1565 struct ixgbevf_ring *rx_ring)
1567 struct pci_dev *pdev = adapter->pdev;
1571 if (!rx_ring->rx_buffer_info)
1574 /* Free all the Rx ring sk_buffs */
1575 for (i = 0; i < rx_ring->count; i++) {
1576 struct ixgbevf_rx_buffer *rx_buffer_info;
1578 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1579 if (rx_buffer_info->dma) {
1580 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
1581 rx_ring->rx_buf_len,
1583 rx_buffer_info->dma = 0;
1585 if (rx_buffer_info->skb) {
1586 struct sk_buff *skb = rx_buffer_info->skb;
1587 rx_buffer_info->skb = NULL;
1589 struct sk_buff *this = skb;
1590 skb = IXGBE_CB(skb)->prev;
1591 dev_kfree_skb(this);
1596 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
1597 memset(rx_ring->rx_buffer_info, 0, size);
1599 /* Zero out the descriptor ring */
1600 memset(rx_ring->desc, 0, rx_ring->size);
1602 rx_ring->next_to_clean = 0;
1603 rx_ring->next_to_use = 0;
1606 writel(0, adapter->hw.hw_addr + rx_ring->head);
1608 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1612 * ixgbevf_clean_tx_ring - Free Tx Buffers
1613 * @adapter: board private structure
1614 * @tx_ring: ring to be cleaned
1616 static void ixgbevf_clean_tx_ring(struct ixgbevf_adapter *adapter,
1617 struct ixgbevf_ring *tx_ring)
1619 struct ixgbevf_tx_buffer *tx_buffer_info;
1623 if (!tx_ring->tx_buffer_info)
1626 /* Free all the Tx ring sk_buffs */
1627 for (i = 0; i < tx_ring->count; i++) {
1628 tx_buffer_info = &tx_ring->tx_buffer_info[i];
1629 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
1632 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
1633 memset(tx_ring->tx_buffer_info, 0, size);
1635 memset(tx_ring->desc, 0, tx_ring->size);
1637 tx_ring->next_to_use = 0;
1638 tx_ring->next_to_clean = 0;
1641 writel(0, adapter->hw.hw_addr + tx_ring->head);
1643 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1647 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
1648 * @adapter: board private structure
1650 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
1654 for (i = 0; i < adapter->num_rx_queues; i++)
1655 ixgbevf_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1659 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
1660 * @adapter: board private structure
1662 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
1666 for (i = 0; i < adapter->num_tx_queues; i++)
1667 ixgbevf_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1670 void ixgbevf_down(struct ixgbevf_adapter *adapter)
1672 struct net_device *netdev = adapter->netdev;
1673 struct ixgbe_hw *hw = &adapter->hw;
1677 /* signal that we are down to the interrupt handler */
1678 set_bit(__IXGBEVF_DOWN, &adapter->state);
1679 /* disable receives */
1681 netif_tx_disable(netdev);
1685 netif_tx_stop_all_queues(netdev);
1687 ixgbevf_irq_disable(adapter);
1689 ixgbevf_napi_disable_all(adapter);
1691 del_timer_sync(&adapter->watchdog_timer);
1692 /* can't call flush scheduled work here because it can deadlock
1693 * if linkwatch_event tries to acquire the rtnl_lock which we are
1695 while (adapter->flags & IXGBE_FLAG_IN_WATCHDOG_TASK)
1698 /* disable transmits in the hardware now that interrupts are off */
1699 for (i = 0; i < adapter->num_tx_queues; i++) {
1700 j = adapter->tx_ring[i].reg_idx;
1701 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1702 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j),
1703 (txdctl & ~IXGBE_TXDCTL_ENABLE));
1706 netif_carrier_off(netdev);
1708 if (!pci_channel_offline(adapter->pdev))
1709 ixgbevf_reset(adapter);
1711 ixgbevf_clean_all_tx_rings(adapter);
1712 ixgbevf_clean_all_rx_rings(adapter);
1715 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
1717 WARN_ON(in_interrupt());
1719 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
1722 ixgbevf_down(adapter);
1723 ixgbevf_up(adapter);
1725 clear_bit(__IXGBEVF_RESETTING, &adapter->state);
1728 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
1730 struct ixgbe_hw *hw = &adapter->hw;
1731 struct net_device *netdev = adapter->netdev;
1733 if (hw->mac.ops.reset_hw(hw))
1734 hw_dbg(hw, "PF still resetting\n");
1736 hw->mac.ops.init_hw(hw);
1738 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1739 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1741 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1746 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
1750 int vector_threshold;
1752 /* We'll want at least 2 (vector_threshold):
1753 * 1) TxQ[0] + RxQ[0] handler
1754 * 2) Other (Link Status Change, etc.)
1756 vector_threshold = MIN_MSIX_COUNT;
1758 /* The more we get, the more we will assign to Tx/Rx Cleanup
1759 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
1760 * Right now, we simply care about how many we'll get; we'll
1761 * set them up later while requesting irq's.
1763 while (vectors >= vector_threshold) {
1764 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
1766 if (!err || err < 0) /* Success or a nasty failure. */
1768 else /* err == number of vectors we should try again with */
1772 if (vectors < vector_threshold)
1776 dev_err(&adapter->pdev->dev,
1777 "Unable to allocate MSI-X interrupts\n");
1778 kfree(adapter->msix_entries);
1779 adapter->msix_entries = NULL;
1782 * Adjust for only the vectors we'll use, which is minimum
1783 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
1784 * vectors we were allocated.
1786 adapter->num_msix_vectors = vectors;
1793 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
1794 * @adapter: board private structure to initialize
1796 * This is the top level queue allocation routine. The order here is very
1797 * important, starting with the "most" number of features turned on at once,
1798 * and ending with the smallest set of features. This way large combinations
1799 * can be allocated if they're turned on, and smaller combinations are the
1800 * fallthrough conditions.
1803 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
1805 /* Start with base case */
1806 adapter->num_rx_queues = 1;
1807 adapter->num_tx_queues = 1;
1811 * ixgbevf_alloc_queues - Allocate memory for all rings
1812 * @adapter: board private structure to initialize
1814 * We allocate one ring per queue at run-time since we don't know the
1815 * number of queues at compile-time. The polling_netdev array is
1816 * intended for Multiqueue, but should work fine with a single queue.
1818 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
1822 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1823 sizeof(struct ixgbevf_ring), GFP_KERNEL);
1824 if (!adapter->tx_ring)
1825 goto err_tx_ring_allocation;
1827 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1828 sizeof(struct ixgbevf_ring), GFP_KERNEL);
1829 if (!adapter->rx_ring)
1830 goto err_rx_ring_allocation;
1832 for (i = 0; i < adapter->num_tx_queues; i++) {
1833 adapter->tx_ring[i].count = adapter->tx_ring_count;
1834 adapter->tx_ring[i].queue_index = i;
1835 /* reg_idx may be remapped later by DCB config */
1836 adapter->tx_ring[i].reg_idx = i;
1837 adapter->tx_ring[i].dev = &adapter->pdev->dev;
1838 adapter->tx_ring[i].netdev = adapter->netdev;
1841 for (i = 0; i < adapter->num_rx_queues; i++) {
1842 adapter->rx_ring[i].count = adapter->rx_ring_count;
1843 adapter->rx_ring[i].queue_index = i;
1844 adapter->rx_ring[i].reg_idx = i;
1845 adapter->rx_ring[i].dev = &adapter->pdev->dev;
1846 adapter->rx_ring[i].netdev = adapter->netdev;
1851 err_rx_ring_allocation:
1852 kfree(adapter->tx_ring);
1853 err_tx_ring_allocation:
1858 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
1859 * @adapter: board private structure to initialize
1861 * Attempt to configure the interrupts using the best available
1862 * capabilities of the hardware and the kernel.
1864 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
1866 struct net_device *netdev = adapter->netdev;
1868 int vector, v_budget;
1871 * It's easy to be greedy for MSI-X vectors, but it really
1872 * doesn't do us much good if we have a lot more vectors
1873 * than CPU's. So let's be conservative and only ask for
1874 * (roughly) the same number of vectors as there are CPU's.
1875 * The default is to use pairs of vectors.
1877 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
1878 v_budget = min_t(int, v_budget, num_online_cpus());
1879 v_budget += NON_Q_VECTORS;
1881 /* A failure in MSI-X entry allocation isn't fatal, but it does
1882 * mean we disable MSI-X capabilities of the adapter. */
1883 adapter->msix_entries = kcalloc(v_budget,
1884 sizeof(struct msix_entry), GFP_KERNEL);
1885 if (!adapter->msix_entries) {
1890 for (vector = 0; vector < v_budget; vector++)
1891 adapter->msix_entries[vector].entry = vector;
1893 err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
1897 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
1901 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
1908 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
1909 * @adapter: board private structure to initialize
1911 * We allocate one q_vector per queue interrupt. If allocation fails we
1914 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
1916 int q_idx, num_q_vectors;
1917 struct ixgbevf_q_vector *q_vector;
1919 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1921 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1922 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
1925 q_vector->adapter = adapter;
1926 q_vector->v_idx = q_idx;
1927 netif_napi_add(adapter->netdev, &q_vector->napi,
1929 adapter->q_vector[q_idx] = q_vector;
1937 q_vector = adapter->q_vector[q_idx];
1938 netif_napi_del(&q_vector->napi);
1940 adapter->q_vector[q_idx] = NULL;
1946 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
1947 * @adapter: board private structure to initialize
1949 * This function frees the memory allocated to the q_vectors. In addition if
1950 * NAPI is enabled it will delete any references to the NAPI struct prior
1951 * to freeing the q_vector.
1953 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
1955 int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1957 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1958 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
1960 adapter->q_vector[q_idx] = NULL;
1961 netif_napi_del(&q_vector->napi);
1967 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
1968 * @adapter: board private structure
1971 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
1973 pci_disable_msix(adapter->pdev);
1974 kfree(adapter->msix_entries);
1975 adapter->msix_entries = NULL;
1979 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
1980 * @adapter: board private structure to initialize
1983 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
1987 /* Number of supported queues */
1988 ixgbevf_set_num_queues(adapter);
1990 err = ixgbevf_set_interrupt_capability(adapter);
1992 hw_dbg(&adapter->hw,
1993 "Unable to setup interrupt capabilities\n");
1994 goto err_set_interrupt;
1997 err = ixgbevf_alloc_q_vectors(adapter);
1999 hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
2001 goto err_alloc_q_vectors;
2004 err = ixgbevf_alloc_queues(adapter);
2006 pr_err("Unable to allocate memory for queues\n");
2007 goto err_alloc_queues;
2010 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
2011 "Tx Queue count = %u\n",
2012 (adapter->num_rx_queues > 1) ? "Enabled" :
2013 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2015 set_bit(__IXGBEVF_DOWN, &adapter->state);
2019 ixgbevf_free_q_vectors(adapter);
2020 err_alloc_q_vectors:
2021 ixgbevf_reset_interrupt_capability(adapter);
2027 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2028 * @adapter: board private structure to clear interrupt scheme on
2030 * We go through and clear interrupt specific resources and reset the structure
2031 * to pre-load conditions
2033 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2035 adapter->num_tx_queues = 0;
2036 adapter->num_rx_queues = 0;
2038 ixgbevf_free_q_vectors(adapter);
2039 ixgbevf_reset_interrupt_capability(adapter);
2043 * ixgbevf_sw_init - Initialize general software structures
2044 * (struct ixgbevf_adapter)
2045 * @adapter: board private structure to initialize
2047 * ixgbevf_sw_init initializes the Adapter private data structure.
2048 * Fields are initialized based on PCI device information and
2049 * OS network device settings (MTU size).
2051 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2053 struct ixgbe_hw *hw = &adapter->hw;
2054 struct pci_dev *pdev = adapter->pdev;
2055 struct net_device *netdev = adapter->netdev;
2058 /* PCI config space info */
2060 hw->vendor_id = pdev->vendor;
2061 hw->device_id = pdev->device;
2062 hw->revision_id = pdev->revision;
2063 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2064 hw->subsystem_device_id = pdev->subsystem_device;
2066 hw->mbx.ops.init_params(hw);
2068 /* assume legacy case in which PF would only give VF 2 queues */
2069 hw->mac.max_tx_queues = 2;
2070 hw->mac.max_rx_queues = 2;
2072 err = hw->mac.ops.reset_hw(hw);
2074 dev_info(&pdev->dev,
2075 "PF still in reset state. Is the PF interface up?\n");
2077 err = hw->mac.ops.init_hw(hw);
2079 pr_err("init_shared_code failed: %d\n", err);
2082 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2084 dev_info(&pdev->dev, "Error reading MAC address\n");
2085 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2086 dev_info(&pdev->dev,
2087 "MAC address not assigned by administrator.\n");
2088 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2091 if (!is_valid_ether_addr(netdev->dev_addr)) {
2092 dev_info(&pdev->dev, "Assigning random MAC address\n");
2093 eth_hw_addr_random(netdev);
2094 memcpy(hw->mac.addr, netdev->dev_addr, netdev->addr_len);
2097 /* lock to protect mailbox accesses */
2098 spin_lock_init(&adapter->mbx_lock);
2100 /* Enable dynamic interrupt throttling rates */
2101 adapter->rx_itr_setting = 1;
2102 adapter->tx_itr_setting = 1;
2104 /* set default ring sizes */
2105 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2106 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2108 set_bit(__IXGBEVF_DOWN, &adapter->state);
2115 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
2117 u32 current_counter = IXGBE_READ_REG(hw, reg); \
2118 if (current_counter < last_counter) \
2119 counter += 0x100000000LL; \
2120 last_counter = current_counter; \
2121 counter &= 0xFFFFFFFF00000000LL; \
2122 counter |= current_counter; \
2125 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2127 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
2128 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
2129 u64 current_counter = (current_counter_msb << 32) | \
2130 current_counter_lsb; \
2131 if (current_counter < last_counter) \
2132 counter += 0x1000000000LL; \
2133 last_counter = current_counter; \
2134 counter &= 0xFFFFFFF000000000LL; \
2135 counter |= current_counter; \
2138 * ixgbevf_update_stats - Update the board statistics counters.
2139 * @adapter: board private structure
2141 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2143 struct ixgbe_hw *hw = &adapter->hw;
2146 if (!adapter->link_up)
2149 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2150 adapter->stats.vfgprc);
2151 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2152 adapter->stats.vfgptc);
2153 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2154 adapter->stats.last_vfgorc,
2155 adapter->stats.vfgorc);
2156 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2157 adapter->stats.last_vfgotc,
2158 adapter->stats.vfgotc);
2159 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2160 adapter->stats.vfmprc);
2162 for (i = 0; i < adapter->num_rx_queues; i++) {
2163 adapter->hw_csum_rx_error +=
2164 adapter->rx_ring[i].hw_csum_rx_error;
2165 adapter->hw_csum_rx_good +=
2166 adapter->rx_ring[i].hw_csum_rx_good;
2167 adapter->rx_ring[i].hw_csum_rx_error = 0;
2168 adapter->rx_ring[i].hw_csum_rx_good = 0;
2173 * ixgbevf_watchdog - Timer Call-back
2174 * @data: pointer to adapter cast into an unsigned long
2176 static void ixgbevf_watchdog(unsigned long data)
2178 struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2179 struct ixgbe_hw *hw = &adapter->hw;
2184 * Do the watchdog outside of interrupt context due to the lovely
2185 * delays that some of the newer hardware requires
2188 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
2189 goto watchdog_short_circuit;
2191 /* get one bit for every active tx/rx interrupt vector */
2192 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2193 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2194 if (qv->rx.ring || qv->tx.ring)
2198 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2200 watchdog_short_circuit:
2201 schedule_work(&adapter->watchdog_task);
2205 * ixgbevf_tx_timeout - Respond to a Tx Hang
2206 * @netdev: network interface device structure
2208 static void ixgbevf_tx_timeout(struct net_device *netdev)
2210 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2212 /* Do the reset outside of interrupt context */
2213 schedule_work(&adapter->reset_task);
2216 static void ixgbevf_reset_task(struct work_struct *work)
2218 struct ixgbevf_adapter *adapter;
2219 adapter = container_of(work, struct ixgbevf_adapter, reset_task);
2221 /* If we're already down or resetting, just bail */
2222 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2223 test_bit(__IXGBEVF_RESETTING, &adapter->state))
2226 adapter->tx_timeout_count++;
2228 ixgbevf_reinit_locked(adapter);
2232 * ixgbevf_watchdog_task - worker thread to bring link up
2233 * @work: pointer to work_struct containing our data
2235 static void ixgbevf_watchdog_task(struct work_struct *work)
2237 struct ixgbevf_adapter *adapter = container_of(work,
2238 struct ixgbevf_adapter,
2240 struct net_device *netdev = adapter->netdev;
2241 struct ixgbe_hw *hw = &adapter->hw;
2242 u32 link_speed = adapter->link_speed;
2243 bool link_up = adapter->link_up;
2246 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
2249 * Always check the link on the watchdog because we have
2252 spin_lock_bh(&adapter->mbx_lock);
2254 need_reset = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2256 spin_unlock_bh(&adapter->mbx_lock);
2259 adapter->link_up = link_up;
2260 adapter->link_speed = link_speed;
2261 netif_carrier_off(netdev);
2262 netif_tx_stop_all_queues(netdev);
2263 schedule_work(&adapter->reset_task);
2266 adapter->link_up = link_up;
2267 adapter->link_speed = link_speed;
2270 if (!netif_carrier_ok(netdev)) {
2271 char *link_speed_string;
2272 switch (link_speed) {
2273 case IXGBE_LINK_SPEED_10GB_FULL:
2274 link_speed_string = "10 Gbps";
2276 case IXGBE_LINK_SPEED_1GB_FULL:
2277 link_speed_string = "1 Gbps";
2279 case IXGBE_LINK_SPEED_100_FULL:
2280 link_speed_string = "100 Mbps";
2283 link_speed_string = "unknown speed";
2286 dev_info(&adapter->pdev->dev,
2287 "NIC Link is Up, %s\n", link_speed_string);
2288 netif_carrier_on(netdev);
2289 netif_tx_wake_all_queues(netdev);
2292 adapter->link_up = false;
2293 adapter->link_speed = 0;
2294 if (netif_carrier_ok(netdev)) {
2295 dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2296 netif_carrier_off(netdev);
2297 netif_tx_stop_all_queues(netdev);
2301 ixgbevf_update_stats(adapter);
2304 /* Reset the timer */
2305 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
2306 mod_timer(&adapter->watchdog_timer,
2307 round_jiffies(jiffies + (2 * HZ)));
2309 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
2313 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2314 * @adapter: board private structure
2315 * @tx_ring: Tx descriptor ring for a specific queue
2317 * Free all transmit software resources
2319 void ixgbevf_free_tx_resources(struct ixgbevf_adapter *adapter,
2320 struct ixgbevf_ring *tx_ring)
2322 struct pci_dev *pdev = adapter->pdev;
2324 ixgbevf_clean_tx_ring(adapter, tx_ring);
2326 vfree(tx_ring->tx_buffer_info);
2327 tx_ring->tx_buffer_info = NULL;
2329 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2332 tx_ring->desc = NULL;
2336 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2337 * @adapter: board private structure
2339 * Free all transmit software resources
2341 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2345 for (i = 0; i < adapter->num_tx_queues; i++)
2346 if (adapter->tx_ring[i].desc)
2347 ixgbevf_free_tx_resources(adapter,
2348 &adapter->tx_ring[i]);
2353 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2354 * @adapter: board private structure
2355 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2357 * Return 0 on success, negative on failure
2359 int ixgbevf_setup_tx_resources(struct ixgbevf_adapter *adapter,
2360 struct ixgbevf_ring *tx_ring)
2362 struct pci_dev *pdev = adapter->pdev;
2365 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2366 tx_ring->tx_buffer_info = vzalloc(size);
2367 if (!tx_ring->tx_buffer_info)
2370 /* round up to nearest 4K */
2371 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2372 tx_ring->size = ALIGN(tx_ring->size, 4096);
2374 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
2375 &tx_ring->dma, GFP_KERNEL);
2379 tx_ring->next_to_use = 0;
2380 tx_ring->next_to_clean = 0;
2384 vfree(tx_ring->tx_buffer_info);
2385 tx_ring->tx_buffer_info = NULL;
2386 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2387 "descriptor ring\n");
2392 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2393 * @adapter: board private structure
2395 * If this function returns with an error, then it's possible one or
2396 * more of the rings is populated (while the rest are not). It is the
2397 * callers duty to clean those orphaned rings.
2399 * Return 0 on success, negative on failure
2401 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2405 for (i = 0; i < adapter->num_tx_queues; i++) {
2406 err = ixgbevf_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2409 hw_dbg(&adapter->hw,
2410 "Allocation for Tx Queue %u failed\n", i);
2418 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2419 * @adapter: board private structure
2420 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2422 * Returns 0 on success, negative on failure
2424 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
2425 struct ixgbevf_ring *rx_ring)
2427 struct pci_dev *pdev = adapter->pdev;
2430 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2431 rx_ring->rx_buffer_info = vzalloc(size);
2432 if (!rx_ring->rx_buffer_info)
2435 /* Round up to nearest 4K */
2436 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2437 rx_ring->size = ALIGN(rx_ring->size, 4096);
2439 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
2440 &rx_ring->dma, GFP_KERNEL);
2442 if (!rx_ring->desc) {
2443 vfree(rx_ring->rx_buffer_info);
2444 rx_ring->rx_buffer_info = NULL;
2448 rx_ring->next_to_clean = 0;
2449 rx_ring->next_to_use = 0;
2457 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
2458 * @adapter: board private structure
2460 * If this function returns with an error, then it's possible one or
2461 * more of the rings is populated (while the rest are not). It is the
2462 * callers duty to clean those orphaned rings.
2464 * Return 0 on success, negative on failure
2466 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
2470 for (i = 0; i < adapter->num_rx_queues; i++) {
2471 err = ixgbevf_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2474 hw_dbg(&adapter->hw,
2475 "Allocation for Rx Queue %u failed\n", i);
2482 * ixgbevf_free_rx_resources - Free Rx Resources
2483 * @adapter: board private structure
2484 * @rx_ring: ring to clean the resources from
2486 * Free all receive software resources
2488 void ixgbevf_free_rx_resources(struct ixgbevf_adapter *adapter,
2489 struct ixgbevf_ring *rx_ring)
2491 struct pci_dev *pdev = adapter->pdev;
2493 ixgbevf_clean_rx_ring(adapter, rx_ring);
2495 vfree(rx_ring->rx_buffer_info);
2496 rx_ring->rx_buffer_info = NULL;
2498 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2501 rx_ring->desc = NULL;
2505 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
2506 * @adapter: board private structure
2508 * Free all receive software resources
2510 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
2514 for (i = 0; i < adapter->num_rx_queues; i++)
2515 if (adapter->rx_ring[i].desc)
2516 ixgbevf_free_rx_resources(adapter,
2517 &adapter->rx_ring[i]);
2520 static int ixgbevf_setup_queues(struct ixgbevf_adapter *adapter)
2522 struct ixgbe_hw *hw = &adapter->hw;
2523 struct ixgbevf_ring *rx_ring;
2524 unsigned int def_q = 0;
2525 unsigned int num_tcs = 0;
2526 unsigned int num_rx_queues = 1;
2529 spin_lock_bh(&adapter->mbx_lock);
2531 /* fetch queue configuration from the PF */
2532 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2534 spin_unlock_bh(&adapter->mbx_lock);
2540 /* update default Tx ring register index */
2541 adapter->tx_ring[0].reg_idx = def_q;
2543 /* we need as many queues as traffic classes */
2544 num_rx_queues = num_tcs;
2547 /* nothing to do if we have the correct number of queues */
2548 if (adapter->num_rx_queues == num_rx_queues)
2551 /* allocate new rings */
2552 rx_ring = kcalloc(num_rx_queues,
2553 sizeof(struct ixgbevf_ring), GFP_KERNEL);
2557 /* setup ring fields */
2558 for (i = 0; i < num_rx_queues; i++) {
2559 rx_ring[i].count = adapter->rx_ring_count;
2560 rx_ring[i].queue_index = i;
2561 rx_ring[i].reg_idx = i;
2562 rx_ring[i].dev = &adapter->pdev->dev;
2563 rx_ring[i].netdev = adapter->netdev;
2566 /* free the existing ring and queues */
2567 adapter->num_rx_queues = 0;
2568 kfree(adapter->rx_ring);
2570 /* move new rings into position on the adapter struct */
2571 adapter->rx_ring = rx_ring;
2572 adapter->num_rx_queues = num_rx_queues;
2578 * ixgbevf_open - Called when a network interface is made active
2579 * @netdev: network interface device structure
2581 * Returns 0 on success, negative value on failure
2583 * The open entry point is called when a network interface is made
2584 * active by the system (IFF_UP). At this point all resources needed
2585 * for transmit and receive operations are allocated, the interrupt
2586 * handler is registered with the OS, the watchdog timer is started,
2587 * and the stack is notified that the interface is ready.
2589 static int ixgbevf_open(struct net_device *netdev)
2591 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2592 struct ixgbe_hw *hw = &adapter->hw;
2595 /* A previous failure to open the device because of a lack of
2596 * available MSIX vector resources may have reset the number
2597 * of msix vectors variable to zero. The only way to recover
2598 * is to unload/reload the driver and hope that the system has
2599 * been able to recover some MSIX vector resources.
2601 if (!adapter->num_msix_vectors)
2604 /* disallow open during test */
2605 if (test_bit(__IXGBEVF_TESTING, &adapter->state))
2608 if (hw->adapter_stopped) {
2609 ixgbevf_reset(adapter);
2610 /* if adapter is still stopped then PF isn't up and
2611 * the vf can't start. */
2612 if (hw->adapter_stopped) {
2613 err = IXGBE_ERR_MBX;
2614 pr_err("Unable to start - perhaps the PF Driver isn't "
2616 goto err_setup_reset;
2620 ixgbevf_negotiate_api(adapter);
2622 /* setup queue reg_idx and Rx queue count */
2623 err = ixgbevf_setup_queues(adapter);
2625 goto err_setup_queues;
2627 /* allocate transmit descriptors */
2628 err = ixgbevf_setup_all_tx_resources(adapter);
2632 /* allocate receive descriptors */
2633 err = ixgbevf_setup_all_rx_resources(adapter);
2637 ixgbevf_configure(adapter);
2640 * Map the Tx/Rx rings to the vectors we were allotted.
2641 * if request_irq will be called in this function map_rings
2642 * must be called *before* up_complete
2644 ixgbevf_map_rings_to_vectors(adapter);
2646 ixgbevf_up_complete(adapter);
2648 /* clear any pending interrupts, may auto mask */
2649 IXGBE_READ_REG(hw, IXGBE_VTEICR);
2650 err = ixgbevf_request_irq(adapter);
2654 ixgbevf_irq_enable(adapter);
2659 ixgbevf_down(adapter);
2661 ixgbevf_free_all_rx_resources(adapter);
2663 ixgbevf_free_all_tx_resources(adapter);
2665 ixgbevf_reset(adapter);
2673 * ixgbevf_close - Disables a network interface
2674 * @netdev: network interface device structure
2676 * Returns 0, this is not allowed to fail
2678 * The close entry point is called when an interface is de-activated
2679 * by the OS. The hardware is still under the drivers control, but
2680 * needs to be disabled. A global MAC reset is issued to stop the
2681 * hardware, and all transmit and receive resources are freed.
2683 static int ixgbevf_close(struct net_device *netdev)
2685 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2687 ixgbevf_down(adapter);
2688 ixgbevf_free_irq(adapter);
2690 ixgbevf_free_all_tx_resources(adapter);
2691 ixgbevf_free_all_rx_resources(adapter);
2696 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
2697 u32 vlan_macip_lens, u32 type_tucmd,
2700 struct ixgbe_adv_tx_context_desc *context_desc;
2701 u16 i = tx_ring->next_to_use;
2703 context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
2706 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2708 /* set bits to identify this as an advanced context descriptor */
2709 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
2711 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
2712 context_desc->seqnum_seed = 0;
2713 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
2714 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2717 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
2718 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2720 u32 vlan_macip_lens, type_tucmd;
2721 u32 mss_l4len_idx, l4len;
2723 if (!skb_is_gso(skb))
2726 if (skb_header_cloned(skb)) {
2727 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2732 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2733 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
2735 if (skb->protocol == htons(ETH_P_IP)) {
2736 struct iphdr *iph = ip_hdr(skb);
2739 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2743 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2744 } else if (skb_is_gso_v6(skb)) {
2745 ipv6_hdr(skb)->payload_len = 0;
2746 tcp_hdr(skb)->check =
2747 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2748 &ipv6_hdr(skb)->daddr,
2752 /* compute header lengths */
2753 l4len = tcp_hdrlen(skb);
2755 *hdr_len = skb_transport_offset(skb) + l4len;
2757 /* mss_l4len_id: use 1 as index for TSO */
2758 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
2759 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
2760 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
2762 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
2763 vlan_macip_lens = skb_network_header_len(skb);
2764 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2765 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2767 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2768 type_tucmd, mss_l4len_idx);
2773 static bool ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
2774 struct sk_buff *skb, u32 tx_flags)
2776 u32 vlan_macip_lens = 0;
2777 u32 mss_l4len_idx = 0;
2780 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2782 switch (skb->protocol) {
2783 case __constant_htons(ETH_P_IP):
2784 vlan_macip_lens |= skb_network_header_len(skb);
2785 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2786 l4_hdr = ip_hdr(skb)->protocol;
2788 case __constant_htons(ETH_P_IPV6):
2789 vlan_macip_lens |= skb_network_header_len(skb);
2790 l4_hdr = ipv6_hdr(skb)->nexthdr;
2793 if (unlikely(net_ratelimit())) {
2794 dev_warn(tx_ring->dev,
2795 "partial checksum but proto=%x!\n",
2803 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
2804 mss_l4len_idx = tcp_hdrlen(skb) <<
2805 IXGBE_ADVTXD_L4LEN_SHIFT;
2808 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
2809 mss_l4len_idx = sizeof(struct sctphdr) <<
2810 IXGBE_ADVTXD_L4LEN_SHIFT;
2813 mss_l4len_idx = sizeof(struct udphdr) <<
2814 IXGBE_ADVTXD_L4LEN_SHIFT;
2817 if (unlikely(net_ratelimit())) {
2818 dev_warn(tx_ring->dev,
2819 "partial checksum but l4 proto=%x!\n",
2826 /* vlan_macip_lens: MACLEN, VLAN tag */
2827 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2828 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2830 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2831 type_tucmd, mss_l4len_idx);
2833 return (skb->ip_summed == CHECKSUM_PARTIAL);
2836 static int ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
2837 struct sk_buff *skb, u32 tx_flags)
2839 struct ixgbevf_tx_buffer *tx_buffer_info;
2841 unsigned int total = skb->len;
2842 unsigned int offset = 0, size;
2844 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
2848 i = tx_ring->next_to_use;
2850 len = min(skb_headlen(skb), total);
2852 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2853 size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2855 tx_buffer_info->length = size;
2856 tx_buffer_info->mapped_as_page = false;
2857 tx_buffer_info->dma = dma_map_single(tx_ring->dev,
2859 size, DMA_TO_DEVICE);
2860 if (dma_mapping_error(tx_ring->dev, tx_buffer_info->dma))
2868 if (i == tx_ring->count)
2872 for (f = 0; f < nr_frags; f++) {
2873 const struct skb_frag_struct *frag;
2875 frag = &skb_shinfo(skb)->frags[f];
2876 len = min((unsigned int)skb_frag_size(frag), total);
2880 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2881 size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2883 tx_buffer_info->length = size;
2884 tx_buffer_info->dma =
2885 skb_frag_dma_map(tx_ring->dev, frag,
2886 offset, size, DMA_TO_DEVICE);
2887 if (dma_mapping_error(tx_ring->dev,
2888 tx_buffer_info->dma))
2890 tx_buffer_info->mapped_as_page = true;
2897 if (i == tx_ring->count)
2905 i = tx_ring->count - 1;
2908 tx_ring->tx_buffer_info[i].skb = skb;
2913 dev_err(tx_ring->dev, "TX DMA map failed\n");
2915 /* clear timestamp and dma mappings for failed tx_buffer_info map */
2916 tx_buffer_info->dma = 0;
2919 /* clear timestamp and dma mappings for remaining portion of packet */
2920 while (count >= 0) {
2924 i += tx_ring->count;
2925 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2926 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2932 static void ixgbevf_tx_queue(struct ixgbevf_ring *tx_ring, int tx_flags,
2933 int count, unsigned int first, u32 paylen,
2936 union ixgbe_adv_tx_desc *tx_desc = NULL;
2937 struct ixgbevf_tx_buffer *tx_buffer_info;
2938 u32 olinfo_status = 0, cmd_type_len = 0;
2941 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
2943 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
2945 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
2947 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2948 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
2950 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
2951 olinfo_status |= IXGBE_ADVTXD_POPTS_TXSM;
2953 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
2954 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
2956 /* use index 1 context for tso */
2957 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
2958 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
2959 olinfo_status |= IXGBE_ADVTXD_POPTS_IXSM;
2963 * Check Context must be set if Tx switch is enabled, which it
2964 * always is for case where virtual functions are running
2966 olinfo_status |= IXGBE_ADVTXD_CC;
2968 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
2970 i = tx_ring->next_to_use;
2972 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2973 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2974 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
2975 tx_desc->read.cmd_type_len =
2976 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
2977 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2979 if (i == tx_ring->count)
2983 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
2985 tx_ring->tx_buffer_info[first].time_stamp = jiffies;
2987 /* Force memory writes to complete before letting h/w
2988 * know there are new descriptors to fetch. (Only
2989 * applicable for weak-ordered memory model archs,
2994 tx_ring->tx_buffer_info[first].next_to_watch = tx_desc;
2995 tx_ring->next_to_use = i;
2998 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3000 struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3002 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3003 /* Herbert's original patch had:
3004 * smp_mb__after_netif_stop_queue();
3005 * but since that doesn't exist yet, just open code it. */
3008 /* We need to check again in a case another CPU has just
3009 * made room available. */
3010 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
3013 /* A reprieve! - use start_queue because it doesn't call schedule */
3014 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3015 ++adapter->restart_queue;
3019 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3021 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
3023 return __ixgbevf_maybe_stop_tx(tx_ring, size);
3026 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3028 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3029 struct ixgbevf_ring *tx_ring;
3031 unsigned int tx_flags = 0;
3034 u16 count = TXD_USE_COUNT(skb_headlen(skb));
3035 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3038 u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3039 if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3041 return NETDEV_TX_OK;
3044 tx_ring = &adapter->tx_ring[r_idx];
3047 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3048 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3049 * + 2 desc gap to keep tail from touching head,
3050 * + 1 desc for context descriptor,
3051 * otherwise try next time
3053 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3054 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3055 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3057 count += skb_shinfo(skb)->nr_frags;
3059 if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3061 return NETDEV_TX_BUSY;
3064 if (vlan_tx_tag_present(skb)) {
3065 tx_flags |= vlan_tx_tag_get(skb);
3066 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3067 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3070 first = tx_ring->next_to_use;
3072 if (skb->protocol == htons(ETH_P_IP))
3073 tx_flags |= IXGBE_TX_FLAGS_IPV4;
3074 tso = ixgbevf_tso(tx_ring, skb, tx_flags, &hdr_len);
3076 dev_kfree_skb_any(skb);
3077 return NETDEV_TX_OK;
3081 tx_flags |= IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_CSUM;
3082 else if (ixgbevf_tx_csum(tx_ring, skb, tx_flags))
3083 tx_flags |= IXGBE_TX_FLAGS_CSUM;
3085 ixgbevf_tx_queue(tx_ring, tx_flags,
3086 ixgbevf_tx_map(tx_ring, skb, tx_flags),
3087 first, skb->len, hdr_len);
3089 writel(tx_ring->next_to_use, adapter->hw.hw_addr + tx_ring->tail);
3091 ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3093 return NETDEV_TX_OK;
3097 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3098 * @netdev: network interface device structure
3099 * @p: pointer to an address structure
3101 * Returns 0 on success, negative on failure
3103 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3105 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3106 struct ixgbe_hw *hw = &adapter->hw;
3107 struct sockaddr *addr = p;
3109 if (!is_valid_ether_addr(addr->sa_data))
3110 return -EADDRNOTAVAIL;
3112 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3113 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3115 spin_lock_bh(&adapter->mbx_lock);
3117 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3119 spin_unlock_bh(&adapter->mbx_lock);
3125 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3126 * @netdev: network interface device structure
3127 * @new_mtu: new value for maximum frame size
3129 * Returns 0 on success, negative on failure
3131 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3133 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3134 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3135 int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3137 switch (adapter->hw.api_version) {
3138 case ixgbe_mbox_api_11:
3139 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3142 if (adapter->hw.mac.type == ixgbe_mac_X540_vf)
3143 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3147 /* MTU < 68 is an error and causes problems on some kernels */
3148 if ((new_mtu < 68) || (max_frame > max_possible_frame))
3151 hw_dbg(&adapter->hw, "changing MTU from %d to %d\n",
3152 netdev->mtu, new_mtu);
3153 /* must set new MTU before calling down or up */
3154 netdev->mtu = new_mtu;
3156 if (netif_running(netdev))
3157 ixgbevf_reinit_locked(adapter);
3162 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3164 struct net_device *netdev = pci_get_drvdata(pdev);
3165 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3170 netif_device_detach(netdev);
3172 if (netif_running(netdev)) {
3174 ixgbevf_down(adapter);
3175 ixgbevf_free_irq(adapter);
3176 ixgbevf_free_all_tx_resources(adapter);
3177 ixgbevf_free_all_rx_resources(adapter);
3181 ixgbevf_clear_interrupt_scheme(adapter);
3184 retval = pci_save_state(pdev);
3189 pci_disable_device(pdev);
3195 static int ixgbevf_resume(struct pci_dev *pdev)
3197 struct ixgbevf_adapter *adapter = pci_get_drvdata(pdev);
3198 struct net_device *netdev = adapter->netdev;
3201 pci_set_power_state(pdev, PCI_D0);
3202 pci_restore_state(pdev);
3204 * pci_restore_state clears dev->state_saved so call
3205 * pci_save_state to restore it.
3207 pci_save_state(pdev);
3209 err = pci_enable_device_mem(pdev);
3211 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3214 pci_set_master(pdev);
3217 err = ixgbevf_init_interrupt_scheme(adapter);
3220 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3224 ixgbevf_reset(adapter);
3226 if (netif_running(netdev)) {
3227 err = ixgbevf_open(netdev);
3232 netif_device_attach(netdev);
3237 #endif /* CONFIG_PM */
3238 static void ixgbevf_shutdown(struct pci_dev *pdev)
3240 ixgbevf_suspend(pdev, PMSG_SUSPEND);
3243 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3244 struct rtnl_link_stats64 *stats)
3246 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3249 const struct ixgbevf_ring *ring;
3252 ixgbevf_update_stats(adapter);
3254 stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3256 for (i = 0; i < adapter->num_rx_queues; i++) {
3257 ring = &adapter->rx_ring[i];
3259 start = u64_stats_fetch_begin_bh(&ring->syncp);
3260 bytes = ring->total_bytes;
3261 packets = ring->total_packets;
3262 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3263 stats->rx_bytes += bytes;
3264 stats->rx_packets += packets;
3267 for (i = 0; i < adapter->num_tx_queues; i++) {
3268 ring = &adapter->tx_ring[i];
3270 start = u64_stats_fetch_begin_bh(&ring->syncp);
3271 bytes = ring->total_bytes;
3272 packets = ring->total_packets;
3273 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3274 stats->tx_bytes += bytes;
3275 stats->tx_packets += packets;
3281 static const struct net_device_ops ixgbevf_netdev_ops = {
3282 .ndo_open = ixgbevf_open,
3283 .ndo_stop = ixgbevf_close,
3284 .ndo_start_xmit = ixgbevf_xmit_frame,
3285 .ndo_set_rx_mode = ixgbevf_set_rx_mode,
3286 .ndo_get_stats64 = ixgbevf_get_stats,
3287 .ndo_validate_addr = eth_validate_addr,
3288 .ndo_set_mac_address = ixgbevf_set_mac,
3289 .ndo_change_mtu = ixgbevf_change_mtu,
3290 .ndo_tx_timeout = ixgbevf_tx_timeout,
3291 .ndo_vlan_rx_add_vid = ixgbevf_vlan_rx_add_vid,
3292 .ndo_vlan_rx_kill_vid = ixgbevf_vlan_rx_kill_vid,
3295 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3297 dev->netdev_ops = &ixgbevf_netdev_ops;
3298 ixgbevf_set_ethtool_ops(dev);
3299 dev->watchdog_timeo = 5 * HZ;
3303 * ixgbevf_probe - Device Initialization Routine
3304 * @pdev: PCI device information struct
3305 * @ent: entry in ixgbevf_pci_tbl
3307 * Returns 0 on success, negative on failure
3309 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3310 * The OS initialization, configuring of the adapter private structure,
3311 * and a hardware reset occur.
3313 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3315 struct net_device *netdev;
3316 struct ixgbevf_adapter *adapter = NULL;
3317 struct ixgbe_hw *hw = NULL;
3318 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3319 static int cards_found;
3320 int err, pci_using_dac;
3322 err = pci_enable_device(pdev);
3326 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
3327 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
3330 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
3332 err = dma_set_coherent_mask(&pdev->dev,
3335 dev_err(&pdev->dev, "No usable DMA "
3336 "configuration, aborting\n");
3343 err = pci_request_regions(pdev, ixgbevf_driver_name);
3345 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3349 pci_set_master(pdev);
3351 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3355 goto err_alloc_etherdev;
3358 SET_NETDEV_DEV(netdev, &pdev->dev);
3360 pci_set_drvdata(pdev, netdev);
3361 adapter = netdev_priv(netdev);
3363 adapter->netdev = netdev;
3364 adapter->pdev = pdev;
3367 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3370 * call save state here in standalone driver because it relies on
3371 * adapter struct to exist, and needs to call netdev_priv
3373 pci_save_state(pdev);
3375 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3376 pci_resource_len(pdev, 0));
3382 ixgbevf_assign_netdev_ops(netdev);
3384 adapter->bd_number = cards_found;
3387 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3388 hw->mac.type = ii->mac;
3390 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3391 sizeof(struct ixgbe_mbx_operations));
3393 /* setup the private structure */
3394 err = ixgbevf_sw_init(adapter);
3398 /* The HW MAC address was set and/or determined in sw_init */
3399 if (!is_valid_ether_addr(netdev->dev_addr)) {
3400 pr_err("invalid MAC address\n");
3405 netdev->hw_features = NETIF_F_SG |
3412 netdev->features = netdev->hw_features |
3413 NETIF_F_HW_VLAN_TX |
3414 NETIF_F_HW_VLAN_RX |
3415 NETIF_F_HW_VLAN_FILTER;
3417 netdev->vlan_features |= NETIF_F_TSO;
3418 netdev->vlan_features |= NETIF_F_TSO6;
3419 netdev->vlan_features |= NETIF_F_IP_CSUM;
3420 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3421 netdev->vlan_features |= NETIF_F_SG;
3424 netdev->features |= NETIF_F_HIGHDMA;
3426 netdev->priv_flags |= IFF_UNICAST_FLT;
3428 init_timer(&adapter->watchdog_timer);
3429 adapter->watchdog_timer.function = ixgbevf_watchdog;
3430 adapter->watchdog_timer.data = (unsigned long)adapter;
3432 INIT_WORK(&adapter->reset_task, ixgbevf_reset_task);
3433 INIT_WORK(&adapter->watchdog_task, ixgbevf_watchdog_task);
3435 err = ixgbevf_init_interrupt_scheme(adapter);
3439 strcpy(netdev->name, "eth%d");
3441 err = register_netdev(netdev);
3445 netif_carrier_off(netdev);
3447 ixgbevf_init_last_counter_stats(adapter);
3449 /* print the MAC address */
3450 hw_dbg(hw, "%pM\n", netdev->dev_addr);
3452 hw_dbg(hw, "MAC: %d\n", hw->mac.type);
3454 hw_dbg(hw, "Intel(R) 82599 Virtual Function\n");
3459 ixgbevf_clear_interrupt_scheme(adapter);
3461 ixgbevf_reset_interrupt_capability(adapter);
3462 iounmap(hw->hw_addr);
3464 free_netdev(netdev);
3466 pci_release_regions(pdev);
3469 pci_disable_device(pdev);
3474 * ixgbevf_remove - Device Removal Routine
3475 * @pdev: PCI device information struct
3477 * ixgbevf_remove is called by the PCI subsystem to alert the driver
3478 * that it should release a PCI device. The could be caused by a
3479 * Hot-Plug event, or because the driver is going to be removed from
3482 static void ixgbevf_remove(struct pci_dev *pdev)
3484 struct net_device *netdev = pci_get_drvdata(pdev);
3485 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3487 set_bit(__IXGBEVF_DOWN, &adapter->state);
3489 del_timer_sync(&adapter->watchdog_timer);
3491 cancel_work_sync(&adapter->reset_task);
3492 cancel_work_sync(&adapter->watchdog_task);
3494 if (netdev->reg_state == NETREG_REGISTERED)
3495 unregister_netdev(netdev);
3497 ixgbevf_clear_interrupt_scheme(adapter);
3498 ixgbevf_reset_interrupt_capability(adapter);
3500 iounmap(adapter->hw.hw_addr);
3501 pci_release_regions(pdev);
3503 hw_dbg(&adapter->hw, "Remove complete\n");
3505 kfree(adapter->tx_ring);
3506 kfree(adapter->rx_ring);
3508 free_netdev(netdev);
3510 pci_disable_device(pdev);
3514 * ixgbevf_io_error_detected - called when PCI error is detected
3515 * @pdev: Pointer to PCI device
3516 * @state: The current pci connection state
3518 * This function is called after a PCI bus error affecting
3519 * this device has been detected.
3521 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
3522 pci_channel_state_t state)
3524 struct net_device *netdev = pci_get_drvdata(pdev);
3525 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3527 netif_device_detach(netdev);
3529 if (state == pci_channel_io_perm_failure)
3530 return PCI_ERS_RESULT_DISCONNECT;
3532 if (netif_running(netdev))
3533 ixgbevf_down(adapter);
3535 pci_disable_device(pdev);
3537 /* Request a slot slot reset. */
3538 return PCI_ERS_RESULT_NEED_RESET;
3542 * ixgbevf_io_slot_reset - called after the pci bus has been reset.
3543 * @pdev: Pointer to PCI device
3545 * Restart the card from scratch, as if from a cold-boot. Implementation
3546 * resembles the first-half of the ixgbevf_resume routine.
3548 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
3550 struct net_device *netdev = pci_get_drvdata(pdev);
3551 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3553 if (pci_enable_device_mem(pdev)) {
3555 "Cannot re-enable PCI device after reset.\n");
3556 return PCI_ERS_RESULT_DISCONNECT;
3559 pci_set_master(pdev);
3561 ixgbevf_reset(adapter);
3563 return PCI_ERS_RESULT_RECOVERED;
3567 * ixgbevf_io_resume - called when traffic can start flowing again.
3568 * @pdev: Pointer to PCI device
3570 * This callback is called when the error recovery driver tells us that
3571 * its OK to resume normal operation. Implementation resembles the
3572 * second-half of the ixgbevf_resume routine.
3574 static void ixgbevf_io_resume(struct pci_dev *pdev)
3576 struct net_device *netdev = pci_get_drvdata(pdev);
3577 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3579 if (netif_running(netdev))
3580 ixgbevf_up(adapter);
3582 netif_device_attach(netdev);
3585 /* PCI Error Recovery (ERS) */
3586 static const struct pci_error_handlers ixgbevf_err_handler = {
3587 .error_detected = ixgbevf_io_error_detected,
3588 .slot_reset = ixgbevf_io_slot_reset,
3589 .resume = ixgbevf_io_resume,
3592 static struct pci_driver ixgbevf_driver = {
3593 .name = ixgbevf_driver_name,
3594 .id_table = ixgbevf_pci_tbl,
3595 .probe = ixgbevf_probe,
3596 .remove = ixgbevf_remove,
3598 /* Power Management Hooks */
3599 .suspend = ixgbevf_suspend,
3600 .resume = ixgbevf_resume,
3602 .shutdown = ixgbevf_shutdown,
3603 .err_handler = &ixgbevf_err_handler
3607 * ixgbevf_init_module - Driver Registration Routine
3609 * ixgbevf_init_module is the first routine called when the driver is
3610 * loaded. All it does is register with the PCI subsystem.
3612 static int __init ixgbevf_init_module(void)
3615 pr_info("%s - version %s\n", ixgbevf_driver_string,
3616 ixgbevf_driver_version);
3618 pr_info("%s\n", ixgbevf_copyright);
3620 ret = pci_register_driver(&ixgbevf_driver);
3624 module_init(ixgbevf_init_module);
3627 * ixgbevf_exit_module - Driver Exit Cleanup Routine
3629 * ixgbevf_exit_module is called just before the driver is removed
3632 static void __exit ixgbevf_exit_module(void)
3634 pci_unregister_driver(&ixgbevf_driver);
3639 * ixgbevf_get_hw_dev_name - return device name string
3640 * used by hardware layer to print debugging information
3642 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
3644 struct ixgbevf_adapter *adapter = hw->back;
3645 return adapter->netdev->name;
3649 module_exit(ixgbevf_exit_module);
3651 /* ixgbevf_main.c */