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 "2.11.3-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, htons(ETH_P_8021Q), tag);
296 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
297 napi_gro_receive(&q_vector->napi, skb);
303 * ixgbevf_rx_skb - Helper function to determine proper Rx method
304 * @q_vector: structure containing interrupt and ring information
305 * @skb: packet to send up
306 * @status: hardware indication of status of receive
307 * @rx_desc: rx descriptor
309 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
310 struct sk_buff *skb, u8 status,
311 union ixgbe_adv_rx_desc *rx_desc)
313 ixgbevf_receive_skb(q_vector, skb, status, rx_desc);
317 * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
318 * @ring: pointer to Rx descriptor ring structure
319 * @status_err: hardware indication of status of receive
320 * @skb: skb currently being received and modified
322 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
323 u32 status_err, struct sk_buff *skb)
325 skb_checksum_none_assert(skb);
327 /* Rx csum disabled */
328 if (!(ring->netdev->features & NETIF_F_RXCSUM))
331 /* if IP and error */
332 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
333 (status_err & IXGBE_RXDADV_ERR_IPE)) {
334 ring->hw_csum_rx_error++;
338 if (!(status_err & IXGBE_RXD_STAT_L4CS))
341 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
342 ring->hw_csum_rx_error++;
346 /* It must be a TCP or UDP packet with a valid checksum */
347 skb->ip_summed = CHECKSUM_UNNECESSARY;
348 ring->hw_csum_rx_good++;
352 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
353 * @adapter: address of board private structure
355 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_adapter *adapter,
356 struct ixgbevf_ring *rx_ring,
359 struct pci_dev *pdev = adapter->pdev;
360 union ixgbe_adv_rx_desc *rx_desc;
361 struct ixgbevf_rx_buffer *bi;
362 unsigned int i = rx_ring->next_to_use;
364 bi = &rx_ring->rx_buffer_info[i];
366 while (cleaned_count--) {
367 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
372 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
373 rx_ring->rx_buf_len);
375 adapter->alloc_rx_buff_failed++;
380 bi->dma = dma_map_single(&pdev->dev, skb->data,
383 if (dma_mapping_error(&pdev->dev, bi->dma)) {
386 dev_err(&pdev->dev, "RX DMA map failed\n");
390 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
393 if (i == rx_ring->count)
395 bi = &rx_ring->rx_buffer_info[i];
399 if (rx_ring->next_to_use != i) {
400 rx_ring->next_to_use = i;
401 ixgbevf_release_rx_desc(&adapter->hw, rx_ring, i);
405 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
408 struct ixgbe_hw *hw = &adapter->hw;
410 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
413 static bool ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
414 struct ixgbevf_ring *rx_ring,
417 struct ixgbevf_adapter *adapter = q_vector->adapter;
418 struct pci_dev *pdev = adapter->pdev;
419 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
420 struct ixgbevf_rx_buffer *rx_buffer_info, *next_buffer;
424 int cleaned_count = 0;
425 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
427 i = rx_ring->next_to_clean;
428 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
429 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
430 rx_buffer_info = &rx_ring->rx_buffer_info[i];
432 while (staterr & IXGBE_RXD_STAT_DD) {
437 rmb(); /* read descriptor and rx_buffer_info after status DD */
438 len = le16_to_cpu(rx_desc->wb.upper.length);
439 skb = rx_buffer_info->skb;
440 prefetch(skb->data - NET_IP_ALIGN);
441 rx_buffer_info->skb = NULL;
443 if (rx_buffer_info->dma) {
444 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
447 rx_buffer_info->dma = 0;
452 if (i == rx_ring->count)
455 next_rxd = IXGBEVF_RX_DESC(rx_ring, i);
459 next_buffer = &rx_ring->rx_buffer_info[i];
461 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
462 skb->next = next_buffer->skb;
463 IXGBE_CB(skb->next)->prev = skb;
464 adapter->non_eop_descs++;
468 /* we should not be chaining buffers, if we did drop the skb */
469 if (IXGBE_CB(skb)->prev) {
471 struct sk_buff *this = skb;
472 skb = IXGBE_CB(skb)->prev;
478 /* ERR_MASK will only have valid bits if EOP set */
479 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
480 dev_kfree_skb_irq(skb);
484 ixgbevf_rx_checksum(rx_ring, staterr, skb);
486 /* probably a little skewed due to removing CRC */
487 total_rx_bytes += skb->len;
491 * Work around issue of some types of VM to VM loop back
492 * packets not getting split correctly
494 if (staterr & IXGBE_RXD_STAT_LB) {
495 u32 header_fixup_len = skb_headlen(skb);
496 if (header_fixup_len < 14)
497 skb_push(skb, header_fixup_len);
499 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
501 /* Workaround hardware that can't do proper VEPA multicast
504 if ((skb->pkt_type & (PACKET_BROADCAST | PACKET_MULTICAST)) &&
505 ether_addr_equal(adapter->netdev->dev_addr,
506 eth_hdr(skb)->h_source)) {
507 dev_kfree_skb_irq(skb);
511 ixgbevf_rx_skb(q_vector, skb, staterr, rx_desc);
514 rx_desc->wb.upper.status_error = 0;
516 /* return some buffers to hardware, one at a time is too slow */
517 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
518 ixgbevf_alloc_rx_buffers(adapter, rx_ring,
523 /* use prefetched values */
525 rx_buffer_info = &rx_ring->rx_buffer_info[i];
527 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
530 rx_ring->next_to_clean = i;
531 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
534 ixgbevf_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
536 u64_stats_update_begin(&rx_ring->syncp);
537 rx_ring->total_packets += total_rx_packets;
538 rx_ring->total_bytes += total_rx_bytes;
539 u64_stats_update_end(&rx_ring->syncp);
540 q_vector->rx.total_packets += total_rx_packets;
541 q_vector->rx.total_bytes += total_rx_bytes;
547 * ixgbevf_poll - NAPI polling calback
548 * @napi: napi struct with our devices info in it
549 * @budget: amount of work driver is allowed to do this pass, in packets
551 * This function will clean more than one or more rings associated with a
554 static int ixgbevf_poll(struct napi_struct *napi, int budget)
556 struct ixgbevf_q_vector *q_vector =
557 container_of(napi, struct ixgbevf_q_vector, napi);
558 struct ixgbevf_adapter *adapter = q_vector->adapter;
559 struct ixgbevf_ring *ring;
561 bool clean_complete = true;
563 ixgbevf_for_each_ring(ring, q_vector->tx)
564 clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
566 /* attempt to distribute budget to each queue fairly, but don't allow
567 * the budget to go below 1 because we'll exit polling */
568 if (q_vector->rx.count > 1)
569 per_ring_budget = max(budget/q_vector->rx.count, 1);
571 per_ring_budget = budget;
573 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
574 ixgbevf_for_each_ring(ring, q_vector->rx)
575 clean_complete &= ixgbevf_clean_rx_irq(q_vector, ring,
577 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
579 /* If all work not completed, return budget and keep polling */
582 /* all work done, exit the polling mode */
584 if (adapter->rx_itr_setting & 1)
585 ixgbevf_set_itr(q_vector);
586 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
587 ixgbevf_irq_enable_queues(adapter,
588 1 << q_vector->v_idx);
594 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
595 * @q_vector: structure containing interrupt and ring information
597 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
599 struct ixgbevf_adapter *adapter = q_vector->adapter;
600 struct ixgbe_hw *hw = &adapter->hw;
601 int v_idx = q_vector->v_idx;
602 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
605 * set the WDIS bit to not clear the timer bits and cause an
606 * immediate assertion of the interrupt
608 itr_reg |= IXGBE_EITR_CNT_WDIS;
610 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
614 * ixgbevf_configure_msix - Configure MSI-X hardware
615 * @adapter: board private structure
617 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
620 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
622 struct ixgbevf_q_vector *q_vector;
623 int q_vectors, v_idx;
625 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
626 adapter->eims_enable_mask = 0;
629 * Populate the IVAR table and set the ITR values to the
630 * corresponding register.
632 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
633 struct ixgbevf_ring *ring;
634 q_vector = adapter->q_vector[v_idx];
636 ixgbevf_for_each_ring(ring, q_vector->rx)
637 ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
639 ixgbevf_for_each_ring(ring, q_vector->tx)
640 ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
642 if (q_vector->tx.ring && !q_vector->rx.ring) {
644 if (adapter->tx_itr_setting == 1)
645 q_vector->itr = IXGBE_10K_ITR;
647 q_vector->itr = adapter->tx_itr_setting;
649 /* rx or rx/tx vector */
650 if (adapter->rx_itr_setting == 1)
651 q_vector->itr = IXGBE_20K_ITR;
653 q_vector->itr = adapter->rx_itr_setting;
656 /* add q_vector eims value to global eims_enable_mask */
657 adapter->eims_enable_mask |= 1 << v_idx;
659 ixgbevf_write_eitr(q_vector);
662 ixgbevf_set_ivar(adapter, -1, 1, v_idx);
663 /* setup eims_other and add value to global eims_enable_mask */
664 adapter->eims_other = 1 << v_idx;
665 adapter->eims_enable_mask |= adapter->eims_other;
672 latency_invalid = 255
676 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
677 * @q_vector: structure containing interrupt and ring information
678 * @ring_container: structure containing ring performance data
680 * Stores a new ITR value based on packets and byte
681 * counts during the last interrupt. The advantage of per interrupt
682 * computation is faster updates and more accurate ITR for the current
683 * traffic pattern. Constants in this function were computed
684 * based on theoretical maximum wire speed and thresholds were set based
685 * on testing data as well as attempting to minimize response time
686 * while increasing bulk throughput.
688 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
689 struct ixgbevf_ring_container *ring_container)
691 int bytes = ring_container->total_bytes;
692 int packets = ring_container->total_packets;
695 u8 itr_setting = ring_container->itr;
700 /* simple throttlerate management
701 * 0-20MB/s lowest (100000 ints/s)
702 * 20-100MB/s low (20000 ints/s)
703 * 100-1249MB/s bulk (8000 ints/s)
705 /* what was last interrupt timeslice? */
706 timepassed_us = q_vector->itr >> 2;
707 bytes_perint = bytes / timepassed_us; /* bytes/usec */
709 switch (itr_setting) {
711 if (bytes_perint > 10)
712 itr_setting = low_latency;
715 if (bytes_perint > 20)
716 itr_setting = bulk_latency;
717 else if (bytes_perint <= 10)
718 itr_setting = lowest_latency;
721 if (bytes_perint <= 20)
722 itr_setting = low_latency;
726 /* clear work counters since we have the values we need */
727 ring_container->total_bytes = 0;
728 ring_container->total_packets = 0;
730 /* write updated itr to ring container */
731 ring_container->itr = itr_setting;
734 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
736 u32 new_itr = q_vector->itr;
739 ixgbevf_update_itr(q_vector, &q_vector->tx);
740 ixgbevf_update_itr(q_vector, &q_vector->rx);
742 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
744 switch (current_itr) {
745 /* counts and packets in update_itr are dependent on these numbers */
747 new_itr = IXGBE_100K_ITR;
750 new_itr = IXGBE_20K_ITR;
754 new_itr = IXGBE_8K_ITR;
758 if (new_itr != q_vector->itr) {
759 /* do an exponential smoothing */
760 new_itr = (10 * new_itr * q_vector->itr) /
761 ((9 * new_itr) + q_vector->itr);
763 /* save the algorithm value here */
764 q_vector->itr = new_itr;
766 ixgbevf_write_eitr(q_vector);
770 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
772 struct ixgbevf_adapter *adapter = data;
773 struct ixgbe_hw *hw = &adapter->hw;
775 hw->mac.get_link_status = 1;
777 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
778 mod_timer(&adapter->watchdog_timer, jiffies);
780 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
786 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
788 * @data: pointer to our q_vector struct for this interrupt vector
790 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
792 struct ixgbevf_q_vector *q_vector = data;
794 /* EIAM disabled interrupts (on this vector) for us */
795 if (q_vector->rx.ring || q_vector->tx.ring)
796 napi_schedule(&q_vector->napi);
801 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
804 struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
806 a->rx_ring[r_idx].next = q_vector->rx.ring;
807 q_vector->rx.ring = &a->rx_ring[r_idx];
808 q_vector->rx.count++;
811 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
814 struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
816 a->tx_ring[t_idx].next = q_vector->tx.ring;
817 q_vector->tx.ring = &a->tx_ring[t_idx];
818 q_vector->tx.count++;
822 * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
823 * @adapter: board private structure to initialize
825 * This function maps descriptor rings to the queue-specific vectors
826 * we were allotted through the MSI-X enabling code. Ideally, we'd have
827 * one vector per ring/queue, but on a constrained vector budget, we
828 * group the rings as "efficiently" as possible. You would add new
829 * mapping configurations in here.
831 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
835 int rxr_idx = 0, txr_idx = 0;
836 int rxr_remaining = adapter->num_rx_queues;
837 int txr_remaining = adapter->num_tx_queues;
842 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
845 * The ideal configuration...
846 * We have enough vectors to map one per queue.
848 if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
849 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
850 map_vector_to_rxq(adapter, v_start, rxr_idx);
852 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
853 map_vector_to_txq(adapter, v_start, txr_idx);
858 * If we don't have enough vectors for a 1-to-1
859 * mapping, we'll have to group them so there are
860 * multiple queues per vector.
862 /* Re-adjusting *qpv takes care of the remainder. */
863 for (i = v_start; i < q_vectors; i++) {
864 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
865 for (j = 0; j < rqpv; j++) {
866 map_vector_to_rxq(adapter, i, rxr_idx);
871 for (i = v_start; i < q_vectors; i++) {
872 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
873 for (j = 0; j < tqpv; j++) {
874 map_vector_to_txq(adapter, i, txr_idx);
885 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
886 * @adapter: board private structure
888 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
889 * interrupts from the kernel.
891 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
893 struct net_device *netdev = adapter->netdev;
894 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
898 for (vector = 0; vector < q_vectors; vector++) {
899 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
900 struct msix_entry *entry = &adapter->msix_entries[vector];
902 if (q_vector->tx.ring && q_vector->rx.ring) {
903 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
904 "%s-%s-%d", netdev->name, "TxRx", ri++);
906 } else if (q_vector->rx.ring) {
907 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
908 "%s-%s-%d", netdev->name, "rx", ri++);
909 } else if (q_vector->tx.ring) {
910 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
911 "%s-%s-%d", netdev->name, "tx", ti++);
913 /* skip this unused q_vector */
916 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
917 q_vector->name, q_vector);
920 "request_irq failed for MSIX interrupt "
922 goto free_queue_irqs;
926 err = request_irq(adapter->msix_entries[vector].vector,
927 &ixgbevf_msix_other, 0, netdev->name, adapter);
930 "request_irq for msix_other failed: %d\n", err);
931 goto free_queue_irqs;
939 free_irq(adapter->msix_entries[vector].vector,
940 adapter->q_vector[vector]);
942 /* This failure is non-recoverable - it indicates the system is
943 * out of MSIX vector resources and the VF driver cannot run
944 * without them. Set the number of msix vectors to zero
945 * indicating that not enough can be allocated. The error
946 * will be returned to the user indicating device open failed.
947 * Any further attempts to force the driver to open will also
948 * fail. The only way to recover is to unload the driver and
949 * reload it again. If the system has recovered some MSIX
950 * vectors then it may succeed.
952 adapter->num_msix_vectors = 0;
956 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
958 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
960 for (i = 0; i < q_vectors; i++) {
961 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
962 q_vector->rx.ring = NULL;
963 q_vector->tx.ring = NULL;
964 q_vector->rx.count = 0;
965 q_vector->tx.count = 0;
970 * ixgbevf_request_irq - initialize interrupts
971 * @adapter: board private structure
973 * Attempts to configure interrupts using the best available
974 * capabilities of the hardware and kernel.
976 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
980 err = ixgbevf_request_msix_irqs(adapter);
984 "request_irq failed, Error %d\n", err);
989 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
993 q_vectors = adapter->num_msix_vectors;
996 free_irq(adapter->msix_entries[i].vector, adapter);
999 for (; i >= 0; i--) {
1000 /* free only the irqs that were actually requested */
1001 if (!adapter->q_vector[i]->rx.ring &&
1002 !adapter->q_vector[i]->tx.ring)
1005 free_irq(adapter->msix_entries[i].vector,
1006 adapter->q_vector[i]);
1009 ixgbevf_reset_q_vectors(adapter);
1013 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1014 * @adapter: board private structure
1016 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1018 struct ixgbe_hw *hw = &adapter->hw;
1021 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1022 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1023 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1025 IXGBE_WRITE_FLUSH(hw);
1027 for (i = 0; i < adapter->num_msix_vectors; i++)
1028 synchronize_irq(adapter->msix_entries[i].vector);
1032 * ixgbevf_irq_enable - Enable default interrupt generation settings
1033 * @adapter: board private structure
1035 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1037 struct ixgbe_hw *hw = &adapter->hw;
1039 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1040 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1041 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1045 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1046 * @adapter: board private structure
1048 * Configure the Tx unit of the MAC after a reset.
1050 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1053 struct ixgbe_hw *hw = &adapter->hw;
1054 u32 i, j, tdlen, txctrl;
1056 /* Setup the HW Tx Head and Tail descriptor pointers */
1057 for (i = 0; i < adapter->num_tx_queues; i++) {
1058 struct ixgbevf_ring *ring = &adapter->tx_ring[i];
1061 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1062 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1063 (tdba & DMA_BIT_MASK(32)));
1064 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1065 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j), tdlen);
1066 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(j), 0);
1067 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(j), 0);
1068 adapter->tx_ring[i].head = IXGBE_VFTDH(j);
1069 adapter->tx_ring[i].tail = IXGBE_VFTDT(j);
1070 /* Disable Tx Head Writeback RO bit, since this hoses
1071 * bookkeeping if things aren't delivered in order.
1073 txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1074 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1075 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1079 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1081 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1083 struct ixgbevf_ring *rx_ring;
1084 struct ixgbe_hw *hw = &adapter->hw;
1087 rx_ring = &adapter->rx_ring[index];
1089 srrctl = IXGBE_SRRCTL_DROP_EN;
1091 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1093 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
1094 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1096 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1099 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1101 struct ixgbe_hw *hw = &adapter->hw;
1103 /* PSRTYPE must be initialized in 82599 */
1104 u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1105 IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1106 IXGBE_PSRTYPE_L2HDR;
1108 if (adapter->num_rx_queues > 1)
1111 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1114 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter)
1116 struct ixgbe_hw *hw = &adapter->hw;
1117 struct net_device *netdev = adapter->netdev;
1118 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1122 /* notify the PF of our intent to use this size of frame */
1123 ixgbevf_rlpml_set_vf(hw, max_frame);
1125 /* PF will allow an extra 4 bytes past for vlan tagged frames */
1126 max_frame += VLAN_HLEN;
1129 * Allocate buffer sizes that fit well into 32K and
1130 * take into account max frame size of 9.5K
1132 if ((hw->mac.type == ixgbe_mac_X540_vf) &&
1133 (max_frame <= MAXIMUM_ETHERNET_VLAN_SIZE))
1134 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1135 else if (max_frame <= IXGBEVF_RXBUFFER_2K)
1136 rx_buf_len = IXGBEVF_RXBUFFER_2K;
1137 else if (max_frame <= IXGBEVF_RXBUFFER_4K)
1138 rx_buf_len = IXGBEVF_RXBUFFER_4K;
1139 else if (max_frame <= IXGBEVF_RXBUFFER_8K)
1140 rx_buf_len = IXGBEVF_RXBUFFER_8K;
1142 rx_buf_len = IXGBEVF_RXBUFFER_10K;
1144 for (i = 0; i < adapter->num_rx_queues; i++)
1145 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1149 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1150 * @adapter: board private structure
1152 * Configure the Rx unit of the MAC after a reset.
1154 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1157 struct ixgbe_hw *hw = &adapter->hw;
1161 ixgbevf_setup_psrtype(adapter);
1163 /* set_rx_buffer_len must be called before ring initialization */
1164 ixgbevf_set_rx_buffer_len(adapter);
1166 rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1167 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1168 * the Base and Length of the Rx Descriptor Ring */
1169 for (i = 0; i < adapter->num_rx_queues; i++) {
1170 rdba = adapter->rx_ring[i].dma;
1171 j = adapter->rx_ring[i].reg_idx;
1172 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1173 (rdba & DMA_BIT_MASK(32)));
1174 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1175 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j), rdlen);
1176 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(j), 0);
1177 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(j), 0);
1178 adapter->rx_ring[i].head = IXGBE_VFRDH(j);
1179 adapter->rx_ring[i].tail = IXGBE_VFRDT(j);
1181 ixgbevf_configure_srrctl(adapter, j);
1185 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1186 __be16 proto, u16 vid)
1188 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1189 struct ixgbe_hw *hw = &adapter->hw;
1192 spin_lock_bh(&adapter->mbx_lock);
1194 /* add VID to filter table */
1195 err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1197 spin_unlock_bh(&adapter->mbx_lock);
1199 /* translate error return types so error makes sense */
1200 if (err == IXGBE_ERR_MBX)
1203 if (err == IXGBE_ERR_INVALID_ARGUMENT)
1206 set_bit(vid, adapter->active_vlans);
1211 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1212 __be16 proto, u16 vid)
1214 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1215 struct ixgbe_hw *hw = &adapter->hw;
1216 int err = -EOPNOTSUPP;
1218 spin_lock_bh(&adapter->mbx_lock);
1220 /* remove VID from filter table */
1221 err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1223 spin_unlock_bh(&adapter->mbx_lock);
1225 clear_bit(vid, adapter->active_vlans);
1230 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1234 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1235 ixgbevf_vlan_rx_add_vid(adapter->netdev,
1236 htons(ETH_P_8021Q), vid);
1239 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1241 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1242 struct ixgbe_hw *hw = &adapter->hw;
1245 if ((netdev_uc_count(netdev)) > 10) {
1246 pr_err("Too many unicast filters - No Space\n");
1250 if (!netdev_uc_empty(netdev)) {
1251 struct netdev_hw_addr *ha;
1252 netdev_for_each_uc_addr(ha, netdev) {
1253 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1258 * If the list is empty then send message to PF driver to
1259 * clear all macvlans on this VF.
1261 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1268 * ixgbevf_set_rx_mode - Multicast and unicast set
1269 * @netdev: network interface device structure
1271 * The set_rx_method entry point is called whenever the multicast address
1272 * list, unicast address list or the network interface flags are updated.
1273 * This routine is responsible for configuring the hardware for proper
1274 * multicast mode and configuring requested unicast filters.
1276 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1278 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1279 struct ixgbe_hw *hw = &adapter->hw;
1281 spin_lock_bh(&adapter->mbx_lock);
1283 /* reprogram multicast list */
1284 hw->mac.ops.update_mc_addr_list(hw, netdev);
1286 ixgbevf_write_uc_addr_list(netdev);
1288 spin_unlock_bh(&adapter->mbx_lock);
1291 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1294 struct ixgbevf_q_vector *q_vector;
1295 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1297 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1298 q_vector = adapter->q_vector[q_idx];
1299 napi_enable(&q_vector->napi);
1303 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1306 struct ixgbevf_q_vector *q_vector;
1307 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1309 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1310 q_vector = adapter->q_vector[q_idx];
1311 napi_disable(&q_vector->napi);
1315 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1317 struct net_device *netdev = adapter->netdev;
1320 ixgbevf_set_rx_mode(netdev);
1322 ixgbevf_restore_vlan(adapter);
1324 ixgbevf_configure_tx(adapter);
1325 ixgbevf_configure_rx(adapter);
1326 for (i = 0; i < adapter->num_rx_queues; i++) {
1327 struct ixgbevf_ring *ring = &adapter->rx_ring[i];
1328 ixgbevf_alloc_rx_buffers(adapter, ring,
1329 IXGBE_DESC_UNUSED(ring));
1333 #define IXGBEVF_MAX_RX_DESC_POLL 10
1334 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1337 struct ixgbe_hw *hw = &adapter->hw;
1338 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1340 int j = adapter->rx_ring[rxr].reg_idx;
1343 usleep_range(1000, 2000);
1344 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1345 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1348 hw_dbg(hw, "RXDCTL.ENABLE queue %d not set while polling\n",
1351 ixgbevf_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
1352 (adapter->rx_ring[rxr].count - 1));
1355 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1356 struct ixgbevf_ring *ring)
1358 struct ixgbe_hw *hw = &adapter->hw;
1359 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1361 u8 reg_idx = ring->reg_idx;
1363 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1364 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1366 /* write value back with RXDCTL.ENABLE bit cleared */
1367 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1369 /* the hardware may take up to 100us to really disable the rx queue */
1372 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1373 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1376 hw_dbg(hw, "RXDCTL.ENABLE queue %d not cleared while polling\n",
1380 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1382 /* Only save pre-reset stats if there are some */
1383 if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1384 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1385 adapter->stats.base_vfgprc;
1386 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1387 adapter->stats.base_vfgptc;
1388 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1389 adapter->stats.base_vfgorc;
1390 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1391 adapter->stats.base_vfgotc;
1392 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1393 adapter->stats.base_vfmprc;
1397 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1399 struct ixgbe_hw *hw = &adapter->hw;
1401 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1402 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1403 adapter->stats.last_vfgorc |=
1404 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1405 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1406 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1407 adapter->stats.last_vfgotc |=
1408 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1409 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1411 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1412 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1413 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1414 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1415 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
1418 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
1420 struct ixgbe_hw *hw = &adapter->hw;
1421 int api[] = { ixgbe_mbox_api_11,
1423 ixgbe_mbox_api_unknown };
1424 int err = 0, idx = 0;
1426 spin_lock_bh(&adapter->mbx_lock);
1428 while (api[idx] != ixgbe_mbox_api_unknown) {
1429 err = ixgbevf_negotiate_api_version(hw, api[idx]);
1435 spin_unlock_bh(&adapter->mbx_lock);
1438 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
1440 struct net_device *netdev = adapter->netdev;
1441 struct ixgbe_hw *hw = &adapter->hw;
1443 int num_rx_rings = adapter->num_rx_queues;
1446 for (i = 0; i < adapter->num_tx_queues; i++) {
1447 j = adapter->tx_ring[i].reg_idx;
1448 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1449 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
1450 txdctl |= (8 << 16);
1451 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1454 for (i = 0; i < adapter->num_tx_queues; i++) {
1455 j = adapter->tx_ring[i].reg_idx;
1456 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1457 txdctl |= IXGBE_TXDCTL_ENABLE;
1458 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1461 for (i = 0; i < num_rx_rings; i++) {
1462 j = adapter->rx_ring[i].reg_idx;
1463 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1464 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1465 if (hw->mac.type == ixgbe_mac_X540_vf) {
1466 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
1467 rxdctl |= ((netdev->mtu + ETH_HLEN + ETH_FCS_LEN) |
1468 IXGBE_RXDCTL_RLPML_EN);
1470 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1471 ixgbevf_rx_desc_queue_enable(adapter, i);
1474 ixgbevf_configure_msix(adapter);
1476 spin_lock_bh(&adapter->mbx_lock);
1478 if (is_valid_ether_addr(hw->mac.addr))
1479 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
1481 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
1483 spin_unlock_bh(&adapter->mbx_lock);
1485 clear_bit(__IXGBEVF_DOWN, &adapter->state);
1486 ixgbevf_napi_enable_all(adapter);
1488 /* enable transmits */
1489 netif_tx_start_all_queues(netdev);
1491 ixgbevf_save_reset_stats(adapter);
1492 ixgbevf_init_last_counter_stats(adapter);
1494 hw->mac.get_link_status = 1;
1495 mod_timer(&adapter->watchdog_timer, jiffies);
1498 static int ixgbevf_reset_queues(struct ixgbevf_adapter *adapter)
1500 struct ixgbe_hw *hw = &adapter->hw;
1501 struct ixgbevf_ring *rx_ring;
1502 unsigned int def_q = 0;
1503 unsigned int num_tcs = 0;
1504 unsigned int num_rx_queues = 1;
1507 spin_lock_bh(&adapter->mbx_lock);
1509 /* fetch queue configuration from the PF */
1510 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1512 spin_unlock_bh(&adapter->mbx_lock);
1518 /* update default Tx ring register index */
1519 adapter->tx_ring[0].reg_idx = def_q;
1521 /* we need as many queues as traffic classes */
1522 num_rx_queues = num_tcs;
1525 /* nothing to do if we have the correct number of queues */
1526 if (adapter->num_rx_queues == num_rx_queues)
1529 /* allocate new rings */
1530 rx_ring = kcalloc(num_rx_queues,
1531 sizeof(struct ixgbevf_ring), GFP_KERNEL);
1535 /* setup ring fields */
1536 for (i = 0; i < num_rx_queues; i++) {
1537 rx_ring[i].count = adapter->rx_ring_count;
1538 rx_ring[i].queue_index = i;
1539 rx_ring[i].reg_idx = i;
1540 rx_ring[i].dev = &adapter->pdev->dev;
1541 rx_ring[i].netdev = adapter->netdev;
1543 /* allocate resources on the ring */
1544 err = ixgbevf_setup_rx_resources(adapter, &rx_ring[i]);
1548 ixgbevf_free_rx_resources(adapter, &rx_ring[i]);
1555 /* free the existing rings and queues */
1556 ixgbevf_free_all_rx_resources(adapter);
1557 adapter->num_rx_queues = 0;
1558 kfree(adapter->rx_ring);
1560 /* move new rings into position on the adapter struct */
1561 adapter->rx_ring = rx_ring;
1562 adapter->num_rx_queues = num_rx_queues;
1564 /* reset ring to vector mapping */
1565 ixgbevf_reset_q_vectors(adapter);
1566 ixgbevf_map_rings_to_vectors(adapter);
1571 void ixgbevf_up(struct ixgbevf_adapter *adapter)
1573 struct ixgbe_hw *hw = &adapter->hw;
1575 ixgbevf_reset_queues(adapter);
1577 ixgbevf_configure(adapter);
1579 ixgbevf_up_complete(adapter);
1581 /* clear any pending interrupts, may auto mask */
1582 IXGBE_READ_REG(hw, IXGBE_VTEICR);
1584 ixgbevf_irq_enable(adapter);
1588 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
1589 * @adapter: board private structure
1590 * @rx_ring: ring to free buffers from
1592 static void ixgbevf_clean_rx_ring(struct ixgbevf_adapter *adapter,
1593 struct ixgbevf_ring *rx_ring)
1595 struct pci_dev *pdev = adapter->pdev;
1599 if (!rx_ring->rx_buffer_info)
1602 /* Free all the Rx ring sk_buffs */
1603 for (i = 0; i < rx_ring->count; i++) {
1604 struct ixgbevf_rx_buffer *rx_buffer_info;
1606 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1607 if (rx_buffer_info->dma) {
1608 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
1609 rx_ring->rx_buf_len,
1611 rx_buffer_info->dma = 0;
1613 if (rx_buffer_info->skb) {
1614 struct sk_buff *skb = rx_buffer_info->skb;
1615 rx_buffer_info->skb = NULL;
1617 struct sk_buff *this = skb;
1618 skb = IXGBE_CB(skb)->prev;
1619 dev_kfree_skb(this);
1624 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
1625 memset(rx_ring->rx_buffer_info, 0, size);
1627 /* Zero out the descriptor ring */
1628 memset(rx_ring->desc, 0, rx_ring->size);
1630 rx_ring->next_to_clean = 0;
1631 rx_ring->next_to_use = 0;
1634 writel(0, adapter->hw.hw_addr + rx_ring->head);
1636 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1640 * ixgbevf_clean_tx_ring - Free Tx Buffers
1641 * @adapter: board private structure
1642 * @tx_ring: ring to be cleaned
1644 static void ixgbevf_clean_tx_ring(struct ixgbevf_adapter *adapter,
1645 struct ixgbevf_ring *tx_ring)
1647 struct ixgbevf_tx_buffer *tx_buffer_info;
1651 if (!tx_ring->tx_buffer_info)
1654 /* Free all the Tx ring sk_buffs */
1655 for (i = 0; i < tx_ring->count; i++) {
1656 tx_buffer_info = &tx_ring->tx_buffer_info[i];
1657 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
1660 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
1661 memset(tx_ring->tx_buffer_info, 0, size);
1663 memset(tx_ring->desc, 0, tx_ring->size);
1665 tx_ring->next_to_use = 0;
1666 tx_ring->next_to_clean = 0;
1669 writel(0, adapter->hw.hw_addr + tx_ring->head);
1671 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1675 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
1676 * @adapter: board private structure
1678 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
1682 for (i = 0; i < adapter->num_rx_queues; i++)
1683 ixgbevf_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1687 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
1688 * @adapter: board private structure
1690 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
1694 for (i = 0; i < adapter->num_tx_queues; i++)
1695 ixgbevf_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1698 void ixgbevf_down(struct ixgbevf_adapter *adapter)
1700 struct net_device *netdev = adapter->netdev;
1701 struct ixgbe_hw *hw = &adapter->hw;
1705 /* signal that we are down to the interrupt handler */
1706 set_bit(__IXGBEVF_DOWN, &adapter->state);
1708 /* disable all enabled rx queues */
1709 for (i = 0; i < adapter->num_rx_queues; i++)
1710 ixgbevf_disable_rx_queue(adapter, &adapter->rx_ring[i]);
1712 netif_tx_disable(netdev);
1716 netif_tx_stop_all_queues(netdev);
1718 ixgbevf_irq_disable(adapter);
1720 ixgbevf_napi_disable_all(adapter);
1722 del_timer_sync(&adapter->watchdog_timer);
1723 /* can't call flush scheduled work here because it can deadlock
1724 * if linkwatch_event tries to acquire the rtnl_lock which we are
1726 while (adapter->flags & IXGBE_FLAG_IN_WATCHDOG_TASK)
1729 /* disable transmits in the hardware now that interrupts are off */
1730 for (i = 0; i < adapter->num_tx_queues; i++) {
1731 j = adapter->tx_ring[i].reg_idx;
1732 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1733 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j),
1734 (txdctl & ~IXGBE_TXDCTL_ENABLE));
1737 netif_carrier_off(netdev);
1739 if (!pci_channel_offline(adapter->pdev))
1740 ixgbevf_reset(adapter);
1742 ixgbevf_clean_all_tx_rings(adapter);
1743 ixgbevf_clean_all_rx_rings(adapter);
1746 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
1748 WARN_ON(in_interrupt());
1750 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
1753 ixgbevf_down(adapter);
1754 ixgbevf_up(adapter);
1756 clear_bit(__IXGBEVF_RESETTING, &adapter->state);
1759 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
1761 struct ixgbe_hw *hw = &adapter->hw;
1762 struct net_device *netdev = adapter->netdev;
1764 if (hw->mac.ops.reset_hw(hw)) {
1765 hw_dbg(hw, "PF still resetting\n");
1767 hw->mac.ops.init_hw(hw);
1768 ixgbevf_negotiate_api(adapter);
1771 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1772 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1774 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1779 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
1783 int vector_threshold;
1785 /* We'll want at least 2 (vector_threshold):
1786 * 1) TxQ[0] + RxQ[0] handler
1787 * 2) Other (Link Status Change, etc.)
1789 vector_threshold = MIN_MSIX_COUNT;
1791 /* The more we get, the more we will assign to Tx/Rx Cleanup
1792 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
1793 * Right now, we simply care about how many we'll get; we'll
1794 * set them up later while requesting irq's.
1796 while (vectors >= vector_threshold) {
1797 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
1799 if (!err || err < 0) /* Success or a nasty failure. */
1801 else /* err == number of vectors we should try again with */
1805 if (vectors < vector_threshold)
1809 dev_err(&adapter->pdev->dev,
1810 "Unable to allocate MSI-X interrupts\n");
1811 kfree(adapter->msix_entries);
1812 adapter->msix_entries = NULL;
1815 * Adjust for only the vectors we'll use, which is minimum
1816 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
1817 * vectors we were allocated.
1819 adapter->num_msix_vectors = vectors;
1826 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
1827 * @adapter: board private structure to initialize
1829 * This is the top level queue allocation routine. The order here is very
1830 * important, starting with the "most" number of features turned on at once,
1831 * and ending with the smallest set of features. This way large combinations
1832 * can be allocated if they're turned on, and smaller combinations are the
1833 * fallthrough conditions.
1836 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
1838 /* Start with base case */
1839 adapter->num_rx_queues = 1;
1840 adapter->num_tx_queues = 1;
1844 * ixgbevf_alloc_queues - Allocate memory for all rings
1845 * @adapter: board private structure to initialize
1847 * We allocate one ring per queue at run-time since we don't know the
1848 * number of queues at compile-time. The polling_netdev array is
1849 * intended for Multiqueue, but should work fine with a single queue.
1851 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
1855 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1856 sizeof(struct ixgbevf_ring), GFP_KERNEL);
1857 if (!adapter->tx_ring)
1858 goto err_tx_ring_allocation;
1860 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1861 sizeof(struct ixgbevf_ring), GFP_KERNEL);
1862 if (!adapter->rx_ring)
1863 goto err_rx_ring_allocation;
1865 for (i = 0; i < adapter->num_tx_queues; i++) {
1866 adapter->tx_ring[i].count = adapter->tx_ring_count;
1867 adapter->tx_ring[i].queue_index = i;
1868 /* reg_idx may be remapped later by DCB config */
1869 adapter->tx_ring[i].reg_idx = i;
1870 adapter->tx_ring[i].dev = &adapter->pdev->dev;
1871 adapter->tx_ring[i].netdev = adapter->netdev;
1874 for (i = 0; i < adapter->num_rx_queues; i++) {
1875 adapter->rx_ring[i].count = adapter->rx_ring_count;
1876 adapter->rx_ring[i].queue_index = i;
1877 adapter->rx_ring[i].reg_idx = i;
1878 adapter->rx_ring[i].dev = &adapter->pdev->dev;
1879 adapter->rx_ring[i].netdev = adapter->netdev;
1884 err_rx_ring_allocation:
1885 kfree(adapter->tx_ring);
1886 err_tx_ring_allocation:
1891 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
1892 * @adapter: board private structure to initialize
1894 * Attempt to configure the interrupts using the best available
1895 * capabilities of the hardware and the kernel.
1897 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
1899 struct net_device *netdev = adapter->netdev;
1901 int vector, v_budget;
1904 * It's easy to be greedy for MSI-X vectors, but it really
1905 * doesn't do us much good if we have a lot more vectors
1906 * than CPU's. So let's be conservative and only ask for
1907 * (roughly) the same number of vectors as there are CPU's.
1908 * The default is to use pairs of vectors.
1910 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
1911 v_budget = min_t(int, v_budget, num_online_cpus());
1912 v_budget += NON_Q_VECTORS;
1914 /* A failure in MSI-X entry allocation isn't fatal, but it does
1915 * mean we disable MSI-X capabilities of the adapter. */
1916 adapter->msix_entries = kcalloc(v_budget,
1917 sizeof(struct msix_entry), GFP_KERNEL);
1918 if (!adapter->msix_entries) {
1923 for (vector = 0; vector < v_budget; vector++)
1924 adapter->msix_entries[vector].entry = vector;
1926 err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
1930 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
1934 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
1941 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
1942 * @adapter: board private structure to initialize
1944 * We allocate one q_vector per queue interrupt. If allocation fails we
1947 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
1949 int q_idx, num_q_vectors;
1950 struct ixgbevf_q_vector *q_vector;
1952 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1954 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1955 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
1958 q_vector->adapter = adapter;
1959 q_vector->v_idx = q_idx;
1960 netif_napi_add(adapter->netdev, &q_vector->napi,
1962 adapter->q_vector[q_idx] = q_vector;
1970 q_vector = adapter->q_vector[q_idx];
1971 netif_napi_del(&q_vector->napi);
1973 adapter->q_vector[q_idx] = NULL;
1979 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
1980 * @adapter: board private structure to initialize
1982 * This function frees the memory allocated to the q_vectors. In addition if
1983 * NAPI is enabled it will delete any references to the NAPI struct prior
1984 * to freeing the q_vector.
1986 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
1988 int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1990 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1991 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
1993 adapter->q_vector[q_idx] = NULL;
1994 netif_napi_del(&q_vector->napi);
2000 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2001 * @adapter: board private structure
2004 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2006 pci_disable_msix(adapter->pdev);
2007 kfree(adapter->msix_entries);
2008 adapter->msix_entries = NULL;
2012 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2013 * @adapter: board private structure to initialize
2016 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2020 /* Number of supported queues */
2021 ixgbevf_set_num_queues(adapter);
2023 err = ixgbevf_set_interrupt_capability(adapter);
2025 hw_dbg(&adapter->hw,
2026 "Unable to setup interrupt capabilities\n");
2027 goto err_set_interrupt;
2030 err = ixgbevf_alloc_q_vectors(adapter);
2032 hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
2034 goto err_alloc_q_vectors;
2037 err = ixgbevf_alloc_queues(adapter);
2039 pr_err("Unable to allocate memory for queues\n");
2040 goto err_alloc_queues;
2043 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
2044 "Tx Queue count = %u\n",
2045 (adapter->num_rx_queues > 1) ? "Enabled" :
2046 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2048 set_bit(__IXGBEVF_DOWN, &adapter->state);
2052 ixgbevf_free_q_vectors(adapter);
2053 err_alloc_q_vectors:
2054 ixgbevf_reset_interrupt_capability(adapter);
2060 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2061 * @adapter: board private structure to clear interrupt scheme on
2063 * We go through and clear interrupt specific resources and reset the structure
2064 * to pre-load conditions
2066 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2068 adapter->num_tx_queues = 0;
2069 adapter->num_rx_queues = 0;
2071 ixgbevf_free_q_vectors(adapter);
2072 ixgbevf_reset_interrupt_capability(adapter);
2076 * ixgbevf_sw_init - Initialize general software structures
2077 * (struct ixgbevf_adapter)
2078 * @adapter: board private structure to initialize
2080 * ixgbevf_sw_init initializes the Adapter private data structure.
2081 * Fields are initialized based on PCI device information and
2082 * OS network device settings (MTU size).
2084 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2086 struct ixgbe_hw *hw = &adapter->hw;
2087 struct pci_dev *pdev = adapter->pdev;
2088 struct net_device *netdev = adapter->netdev;
2091 /* PCI config space info */
2093 hw->vendor_id = pdev->vendor;
2094 hw->device_id = pdev->device;
2095 hw->revision_id = pdev->revision;
2096 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2097 hw->subsystem_device_id = pdev->subsystem_device;
2099 hw->mbx.ops.init_params(hw);
2101 /* assume legacy case in which PF would only give VF 2 queues */
2102 hw->mac.max_tx_queues = 2;
2103 hw->mac.max_rx_queues = 2;
2105 /* lock to protect mailbox accesses */
2106 spin_lock_init(&adapter->mbx_lock);
2108 err = hw->mac.ops.reset_hw(hw);
2110 dev_info(&pdev->dev,
2111 "PF still in reset state. Is the PF interface up?\n");
2113 err = hw->mac.ops.init_hw(hw);
2115 pr_err("init_shared_code failed: %d\n", err);
2118 ixgbevf_negotiate_api(adapter);
2119 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2121 dev_info(&pdev->dev, "Error reading MAC address\n");
2122 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2123 dev_info(&pdev->dev,
2124 "MAC address not assigned by administrator.\n");
2125 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2128 if (!is_valid_ether_addr(netdev->dev_addr)) {
2129 dev_info(&pdev->dev, "Assigning random MAC address\n");
2130 eth_hw_addr_random(netdev);
2131 memcpy(hw->mac.addr, netdev->dev_addr, netdev->addr_len);
2134 /* Enable dynamic interrupt throttling rates */
2135 adapter->rx_itr_setting = 1;
2136 adapter->tx_itr_setting = 1;
2138 /* set default ring sizes */
2139 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2140 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2142 set_bit(__IXGBEVF_DOWN, &adapter->state);
2149 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
2151 u32 current_counter = IXGBE_READ_REG(hw, reg); \
2152 if (current_counter < last_counter) \
2153 counter += 0x100000000LL; \
2154 last_counter = current_counter; \
2155 counter &= 0xFFFFFFFF00000000LL; \
2156 counter |= current_counter; \
2159 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2161 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
2162 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
2163 u64 current_counter = (current_counter_msb << 32) | \
2164 current_counter_lsb; \
2165 if (current_counter < last_counter) \
2166 counter += 0x1000000000LL; \
2167 last_counter = current_counter; \
2168 counter &= 0xFFFFFFF000000000LL; \
2169 counter |= current_counter; \
2172 * ixgbevf_update_stats - Update the board statistics counters.
2173 * @adapter: board private structure
2175 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2177 struct ixgbe_hw *hw = &adapter->hw;
2180 if (!adapter->link_up)
2183 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2184 adapter->stats.vfgprc);
2185 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2186 adapter->stats.vfgptc);
2187 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2188 adapter->stats.last_vfgorc,
2189 adapter->stats.vfgorc);
2190 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2191 adapter->stats.last_vfgotc,
2192 adapter->stats.vfgotc);
2193 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2194 adapter->stats.vfmprc);
2196 for (i = 0; i < adapter->num_rx_queues; i++) {
2197 adapter->hw_csum_rx_error +=
2198 adapter->rx_ring[i].hw_csum_rx_error;
2199 adapter->hw_csum_rx_good +=
2200 adapter->rx_ring[i].hw_csum_rx_good;
2201 adapter->rx_ring[i].hw_csum_rx_error = 0;
2202 adapter->rx_ring[i].hw_csum_rx_good = 0;
2207 * ixgbevf_watchdog - Timer Call-back
2208 * @data: pointer to adapter cast into an unsigned long
2210 static void ixgbevf_watchdog(unsigned long data)
2212 struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2213 struct ixgbe_hw *hw = &adapter->hw;
2218 * Do the watchdog outside of interrupt context due to the lovely
2219 * delays that some of the newer hardware requires
2222 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
2223 goto watchdog_short_circuit;
2225 /* get one bit for every active tx/rx interrupt vector */
2226 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2227 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2228 if (qv->rx.ring || qv->tx.ring)
2232 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2234 watchdog_short_circuit:
2235 schedule_work(&adapter->watchdog_task);
2239 * ixgbevf_tx_timeout - Respond to a Tx Hang
2240 * @netdev: network interface device structure
2242 static void ixgbevf_tx_timeout(struct net_device *netdev)
2244 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2246 /* Do the reset outside of interrupt context */
2247 schedule_work(&adapter->reset_task);
2250 static void ixgbevf_reset_task(struct work_struct *work)
2252 struct ixgbevf_adapter *adapter;
2253 adapter = container_of(work, struct ixgbevf_adapter, reset_task);
2255 /* If we're already down or resetting, just bail */
2256 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2257 test_bit(__IXGBEVF_RESETTING, &adapter->state))
2260 adapter->tx_timeout_count++;
2262 ixgbevf_reinit_locked(adapter);
2266 * ixgbevf_watchdog_task - worker thread to bring link up
2267 * @work: pointer to work_struct containing our data
2269 static void ixgbevf_watchdog_task(struct work_struct *work)
2271 struct ixgbevf_adapter *adapter = container_of(work,
2272 struct ixgbevf_adapter,
2274 struct net_device *netdev = adapter->netdev;
2275 struct ixgbe_hw *hw = &adapter->hw;
2276 u32 link_speed = adapter->link_speed;
2277 bool link_up = adapter->link_up;
2280 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
2283 * Always check the link on the watchdog because we have
2286 spin_lock_bh(&adapter->mbx_lock);
2288 need_reset = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2290 spin_unlock_bh(&adapter->mbx_lock);
2293 adapter->link_up = link_up;
2294 adapter->link_speed = link_speed;
2295 netif_carrier_off(netdev);
2296 netif_tx_stop_all_queues(netdev);
2297 schedule_work(&adapter->reset_task);
2300 adapter->link_up = link_up;
2301 adapter->link_speed = link_speed;
2304 if (!netif_carrier_ok(netdev)) {
2305 char *link_speed_string;
2306 switch (link_speed) {
2307 case IXGBE_LINK_SPEED_10GB_FULL:
2308 link_speed_string = "10 Gbps";
2310 case IXGBE_LINK_SPEED_1GB_FULL:
2311 link_speed_string = "1 Gbps";
2313 case IXGBE_LINK_SPEED_100_FULL:
2314 link_speed_string = "100 Mbps";
2317 link_speed_string = "unknown speed";
2320 dev_info(&adapter->pdev->dev,
2321 "NIC Link is Up, %s\n", link_speed_string);
2322 netif_carrier_on(netdev);
2323 netif_tx_wake_all_queues(netdev);
2326 adapter->link_up = false;
2327 adapter->link_speed = 0;
2328 if (netif_carrier_ok(netdev)) {
2329 dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2330 netif_carrier_off(netdev);
2331 netif_tx_stop_all_queues(netdev);
2335 ixgbevf_update_stats(adapter);
2338 /* Reset the timer */
2339 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
2340 mod_timer(&adapter->watchdog_timer,
2341 round_jiffies(jiffies + (2 * HZ)));
2343 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
2347 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2348 * @adapter: board private structure
2349 * @tx_ring: Tx descriptor ring for a specific queue
2351 * Free all transmit software resources
2353 void ixgbevf_free_tx_resources(struct ixgbevf_adapter *adapter,
2354 struct ixgbevf_ring *tx_ring)
2356 struct pci_dev *pdev = adapter->pdev;
2358 ixgbevf_clean_tx_ring(adapter, tx_ring);
2360 vfree(tx_ring->tx_buffer_info);
2361 tx_ring->tx_buffer_info = NULL;
2363 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2366 tx_ring->desc = NULL;
2370 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2371 * @adapter: board private structure
2373 * Free all transmit software resources
2375 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2379 for (i = 0; i < adapter->num_tx_queues; i++)
2380 if (adapter->tx_ring[i].desc)
2381 ixgbevf_free_tx_resources(adapter,
2382 &adapter->tx_ring[i]);
2387 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2388 * @adapter: board private structure
2389 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2391 * Return 0 on success, negative on failure
2393 int ixgbevf_setup_tx_resources(struct ixgbevf_adapter *adapter,
2394 struct ixgbevf_ring *tx_ring)
2396 struct pci_dev *pdev = adapter->pdev;
2399 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2400 tx_ring->tx_buffer_info = vzalloc(size);
2401 if (!tx_ring->tx_buffer_info)
2404 /* round up to nearest 4K */
2405 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2406 tx_ring->size = ALIGN(tx_ring->size, 4096);
2408 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
2409 &tx_ring->dma, GFP_KERNEL);
2413 tx_ring->next_to_use = 0;
2414 tx_ring->next_to_clean = 0;
2418 vfree(tx_ring->tx_buffer_info);
2419 tx_ring->tx_buffer_info = NULL;
2420 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2421 "descriptor ring\n");
2426 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2427 * @adapter: board private structure
2429 * If this function returns with an error, then it's possible one or
2430 * more of the rings is populated (while the rest are not). It is the
2431 * callers duty to clean those orphaned rings.
2433 * Return 0 on success, negative on failure
2435 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2439 for (i = 0; i < adapter->num_tx_queues; i++) {
2440 err = ixgbevf_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2443 hw_dbg(&adapter->hw,
2444 "Allocation for Tx Queue %u failed\n", i);
2452 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2453 * @adapter: board private structure
2454 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2456 * Returns 0 on success, negative on failure
2458 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
2459 struct ixgbevf_ring *rx_ring)
2461 struct pci_dev *pdev = adapter->pdev;
2464 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2465 rx_ring->rx_buffer_info = vzalloc(size);
2466 if (!rx_ring->rx_buffer_info)
2469 /* Round up to nearest 4K */
2470 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2471 rx_ring->size = ALIGN(rx_ring->size, 4096);
2473 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
2474 &rx_ring->dma, GFP_KERNEL);
2476 if (!rx_ring->desc) {
2477 vfree(rx_ring->rx_buffer_info);
2478 rx_ring->rx_buffer_info = NULL;
2482 rx_ring->next_to_clean = 0;
2483 rx_ring->next_to_use = 0;
2491 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
2492 * @adapter: board private structure
2494 * If this function returns with an error, then it's possible one or
2495 * more of the rings is populated (while the rest are not). It is the
2496 * callers duty to clean those orphaned rings.
2498 * Return 0 on success, negative on failure
2500 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
2504 for (i = 0; i < adapter->num_rx_queues; i++) {
2505 err = ixgbevf_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2508 hw_dbg(&adapter->hw,
2509 "Allocation for Rx Queue %u failed\n", i);
2516 * ixgbevf_free_rx_resources - Free Rx Resources
2517 * @adapter: board private structure
2518 * @rx_ring: ring to clean the resources from
2520 * Free all receive software resources
2522 void ixgbevf_free_rx_resources(struct ixgbevf_adapter *adapter,
2523 struct ixgbevf_ring *rx_ring)
2525 struct pci_dev *pdev = adapter->pdev;
2527 ixgbevf_clean_rx_ring(adapter, rx_ring);
2529 vfree(rx_ring->rx_buffer_info);
2530 rx_ring->rx_buffer_info = NULL;
2532 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2535 rx_ring->desc = NULL;
2539 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
2540 * @adapter: board private structure
2542 * Free all receive software resources
2544 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
2548 for (i = 0; i < adapter->num_rx_queues; i++)
2549 if (adapter->rx_ring[i].desc)
2550 ixgbevf_free_rx_resources(adapter,
2551 &adapter->rx_ring[i]);
2554 static int ixgbevf_setup_queues(struct ixgbevf_adapter *adapter)
2556 struct ixgbe_hw *hw = &adapter->hw;
2557 struct ixgbevf_ring *rx_ring;
2558 unsigned int def_q = 0;
2559 unsigned int num_tcs = 0;
2560 unsigned int num_rx_queues = 1;
2563 spin_lock_bh(&adapter->mbx_lock);
2565 /* fetch queue configuration from the PF */
2566 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2568 spin_unlock_bh(&adapter->mbx_lock);
2574 /* update default Tx ring register index */
2575 adapter->tx_ring[0].reg_idx = def_q;
2577 /* we need as many queues as traffic classes */
2578 num_rx_queues = num_tcs;
2581 /* nothing to do if we have the correct number of queues */
2582 if (adapter->num_rx_queues == num_rx_queues)
2585 /* allocate new rings */
2586 rx_ring = kcalloc(num_rx_queues,
2587 sizeof(struct ixgbevf_ring), GFP_KERNEL);
2591 /* setup ring fields */
2592 for (i = 0; i < num_rx_queues; i++) {
2593 rx_ring[i].count = adapter->rx_ring_count;
2594 rx_ring[i].queue_index = i;
2595 rx_ring[i].reg_idx = i;
2596 rx_ring[i].dev = &adapter->pdev->dev;
2597 rx_ring[i].netdev = adapter->netdev;
2600 /* free the existing ring and queues */
2601 adapter->num_rx_queues = 0;
2602 kfree(adapter->rx_ring);
2604 /* move new rings into position on the adapter struct */
2605 adapter->rx_ring = rx_ring;
2606 adapter->num_rx_queues = num_rx_queues;
2612 * ixgbevf_open - Called when a network interface is made active
2613 * @netdev: network interface device structure
2615 * Returns 0 on success, negative value on failure
2617 * The open entry point is called when a network interface is made
2618 * active by the system (IFF_UP). At this point all resources needed
2619 * for transmit and receive operations are allocated, the interrupt
2620 * handler is registered with the OS, the watchdog timer is started,
2621 * and the stack is notified that the interface is ready.
2623 static int ixgbevf_open(struct net_device *netdev)
2625 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2626 struct ixgbe_hw *hw = &adapter->hw;
2629 /* A previous failure to open the device because of a lack of
2630 * available MSIX vector resources may have reset the number
2631 * of msix vectors variable to zero. The only way to recover
2632 * is to unload/reload the driver and hope that the system has
2633 * been able to recover some MSIX vector resources.
2635 if (!adapter->num_msix_vectors)
2638 /* disallow open during test */
2639 if (test_bit(__IXGBEVF_TESTING, &adapter->state))
2642 if (hw->adapter_stopped) {
2643 ixgbevf_reset(adapter);
2644 /* if adapter is still stopped then PF isn't up and
2645 * the vf can't start. */
2646 if (hw->adapter_stopped) {
2647 err = IXGBE_ERR_MBX;
2648 pr_err("Unable to start - perhaps the PF Driver isn't "
2650 goto err_setup_reset;
2654 /* setup queue reg_idx and Rx queue count */
2655 err = ixgbevf_setup_queues(adapter);
2657 goto err_setup_queues;
2659 /* allocate transmit descriptors */
2660 err = ixgbevf_setup_all_tx_resources(adapter);
2664 /* allocate receive descriptors */
2665 err = ixgbevf_setup_all_rx_resources(adapter);
2669 ixgbevf_configure(adapter);
2672 * Map the Tx/Rx rings to the vectors we were allotted.
2673 * if request_irq will be called in this function map_rings
2674 * must be called *before* up_complete
2676 ixgbevf_map_rings_to_vectors(adapter);
2678 ixgbevf_up_complete(adapter);
2680 /* clear any pending interrupts, may auto mask */
2681 IXGBE_READ_REG(hw, IXGBE_VTEICR);
2682 err = ixgbevf_request_irq(adapter);
2686 ixgbevf_irq_enable(adapter);
2691 ixgbevf_down(adapter);
2693 ixgbevf_free_all_rx_resources(adapter);
2695 ixgbevf_free_all_tx_resources(adapter);
2697 ixgbevf_reset(adapter);
2705 * ixgbevf_close - Disables a network interface
2706 * @netdev: network interface device structure
2708 * Returns 0, this is not allowed to fail
2710 * The close entry point is called when an interface is de-activated
2711 * by the OS. The hardware is still under the drivers control, but
2712 * needs to be disabled. A global MAC reset is issued to stop the
2713 * hardware, and all transmit and receive resources are freed.
2715 static int ixgbevf_close(struct net_device *netdev)
2717 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2719 ixgbevf_down(adapter);
2720 ixgbevf_free_irq(adapter);
2722 ixgbevf_free_all_tx_resources(adapter);
2723 ixgbevf_free_all_rx_resources(adapter);
2728 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
2729 u32 vlan_macip_lens, u32 type_tucmd,
2732 struct ixgbe_adv_tx_context_desc *context_desc;
2733 u16 i = tx_ring->next_to_use;
2735 context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
2738 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2740 /* set bits to identify this as an advanced context descriptor */
2741 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
2743 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
2744 context_desc->seqnum_seed = 0;
2745 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
2746 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2749 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
2750 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2752 u32 vlan_macip_lens, type_tucmd;
2753 u32 mss_l4len_idx, l4len;
2755 if (!skb_is_gso(skb))
2758 if (skb_header_cloned(skb)) {
2759 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2764 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2765 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
2767 if (skb->protocol == htons(ETH_P_IP)) {
2768 struct iphdr *iph = ip_hdr(skb);
2771 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2775 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2776 } else if (skb_is_gso_v6(skb)) {
2777 ipv6_hdr(skb)->payload_len = 0;
2778 tcp_hdr(skb)->check =
2779 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2780 &ipv6_hdr(skb)->daddr,
2784 /* compute header lengths */
2785 l4len = tcp_hdrlen(skb);
2787 *hdr_len = skb_transport_offset(skb) + l4len;
2789 /* mss_l4len_id: use 1 as index for TSO */
2790 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
2791 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
2792 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
2794 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
2795 vlan_macip_lens = skb_network_header_len(skb);
2796 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2797 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2799 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2800 type_tucmd, mss_l4len_idx);
2805 static bool ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
2806 struct sk_buff *skb, u32 tx_flags)
2808 u32 vlan_macip_lens = 0;
2809 u32 mss_l4len_idx = 0;
2812 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2814 switch (skb->protocol) {
2815 case __constant_htons(ETH_P_IP):
2816 vlan_macip_lens |= skb_network_header_len(skb);
2817 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2818 l4_hdr = ip_hdr(skb)->protocol;
2820 case __constant_htons(ETH_P_IPV6):
2821 vlan_macip_lens |= skb_network_header_len(skb);
2822 l4_hdr = ipv6_hdr(skb)->nexthdr;
2825 if (unlikely(net_ratelimit())) {
2826 dev_warn(tx_ring->dev,
2827 "partial checksum but proto=%x!\n",
2835 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
2836 mss_l4len_idx = tcp_hdrlen(skb) <<
2837 IXGBE_ADVTXD_L4LEN_SHIFT;
2840 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
2841 mss_l4len_idx = sizeof(struct sctphdr) <<
2842 IXGBE_ADVTXD_L4LEN_SHIFT;
2845 mss_l4len_idx = sizeof(struct udphdr) <<
2846 IXGBE_ADVTXD_L4LEN_SHIFT;
2849 if (unlikely(net_ratelimit())) {
2850 dev_warn(tx_ring->dev,
2851 "partial checksum but l4 proto=%x!\n",
2858 /* vlan_macip_lens: MACLEN, VLAN tag */
2859 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2860 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2862 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2863 type_tucmd, mss_l4len_idx);
2865 return (skb->ip_summed == CHECKSUM_PARTIAL);
2868 static int ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
2869 struct sk_buff *skb, u32 tx_flags)
2871 struct ixgbevf_tx_buffer *tx_buffer_info;
2873 unsigned int total = skb->len;
2874 unsigned int offset = 0, size;
2876 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
2880 i = tx_ring->next_to_use;
2882 len = min(skb_headlen(skb), total);
2884 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2885 size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2887 tx_buffer_info->length = size;
2888 tx_buffer_info->mapped_as_page = false;
2889 tx_buffer_info->dma = dma_map_single(tx_ring->dev,
2891 size, DMA_TO_DEVICE);
2892 if (dma_mapping_error(tx_ring->dev, tx_buffer_info->dma))
2900 if (i == tx_ring->count)
2904 for (f = 0; f < nr_frags; f++) {
2905 const struct skb_frag_struct *frag;
2907 frag = &skb_shinfo(skb)->frags[f];
2908 len = min((unsigned int)skb_frag_size(frag), total);
2912 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2913 size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2915 tx_buffer_info->length = size;
2916 tx_buffer_info->dma =
2917 skb_frag_dma_map(tx_ring->dev, frag,
2918 offset, size, DMA_TO_DEVICE);
2919 if (dma_mapping_error(tx_ring->dev,
2920 tx_buffer_info->dma))
2922 tx_buffer_info->mapped_as_page = true;
2929 if (i == tx_ring->count)
2937 i = tx_ring->count - 1;
2940 tx_ring->tx_buffer_info[i].skb = skb;
2945 dev_err(tx_ring->dev, "TX DMA map failed\n");
2947 /* clear timestamp and dma mappings for failed tx_buffer_info map */
2948 tx_buffer_info->dma = 0;
2951 /* clear timestamp and dma mappings for remaining portion of packet */
2952 while (count >= 0) {
2956 i += tx_ring->count;
2957 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2958 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2964 static void ixgbevf_tx_queue(struct ixgbevf_ring *tx_ring, int tx_flags,
2965 int count, unsigned int first, u32 paylen,
2968 union ixgbe_adv_tx_desc *tx_desc = NULL;
2969 struct ixgbevf_tx_buffer *tx_buffer_info;
2970 u32 olinfo_status = 0, cmd_type_len = 0;
2973 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
2975 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
2977 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
2979 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2980 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
2982 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
2983 olinfo_status |= IXGBE_ADVTXD_POPTS_TXSM;
2985 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
2986 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
2988 /* use index 1 context for tso */
2989 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
2990 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
2991 olinfo_status |= IXGBE_ADVTXD_POPTS_IXSM;
2995 * Check Context must be set if Tx switch is enabled, which it
2996 * always is for case where virtual functions are running
2998 olinfo_status |= IXGBE_ADVTXD_CC;
3000 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
3002 i = tx_ring->next_to_use;
3004 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3005 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3006 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
3007 tx_desc->read.cmd_type_len =
3008 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
3009 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3011 if (i == tx_ring->count)
3015 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
3017 tx_ring->tx_buffer_info[first].time_stamp = jiffies;
3019 /* Force memory writes to complete before letting h/w
3020 * know there are new descriptors to fetch. (Only
3021 * applicable for weak-ordered memory model archs,
3026 tx_ring->tx_buffer_info[first].next_to_watch = tx_desc;
3027 tx_ring->next_to_use = i;
3030 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3032 struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3034 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3035 /* Herbert's original patch had:
3036 * smp_mb__after_netif_stop_queue();
3037 * but since that doesn't exist yet, just open code it. */
3040 /* We need to check again in a case another CPU has just
3041 * made room available. */
3042 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
3045 /* A reprieve! - use start_queue because it doesn't call schedule */
3046 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3047 ++adapter->restart_queue;
3051 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3053 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
3055 return __ixgbevf_maybe_stop_tx(tx_ring, size);
3058 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3060 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3061 struct ixgbevf_ring *tx_ring;
3063 unsigned int tx_flags = 0;
3066 u16 count = TXD_USE_COUNT(skb_headlen(skb));
3067 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3070 u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3071 if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3073 return NETDEV_TX_OK;
3076 tx_ring = &adapter->tx_ring[r_idx];
3079 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3080 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3081 * + 2 desc gap to keep tail from touching head,
3082 * + 1 desc for context descriptor,
3083 * otherwise try next time
3085 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3086 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3087 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3089 count += skb_shinfo(skb)->nr_frags;
3091 if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3093 return NETDEV_TX_BUSY;
3096 if (vlan_tx_tag_present(skb)) {
3097 tx_flags |= vlan_tx_tag_get(skb);
3098 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3099 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3102 first = tx_ring->next_to_use;
3104 if (skb->protocol == htons(ETH_P_IP))
3105 tx_flags |= IXGBE_TX_FLAGS_IPV4;
3106 tso = ixgbevf_tso(tx_ring, skb, tx_flags, &hdr_len);
3108 dev_kfree_skb_any(skb);
3109 return NETDEV_TX_OK;
3113 tx_flags |= IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_CSUM;
3114 else if (ixgbevf_tx_csum(tx_ring, skb, tx_flags))
3115 tx_flags |= IXGBE_TX_FLAGS_CSUM;
3117 ixgbevf_tx_queue(tx_ring, tx_flags,
3118 ixgbevf_tx_map(tx_ring, skb, tx_flags),
3119 first, skb->len, hdr_len);
3121 writel(tx_ring->next_to_use, adapter->hw.hw_addr + tx_ring->tail);
3123 ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3125 return NETDEV_TX_OK;
3129 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3130 * @netdev: network interface device structure
3131 * @p: pointer to an address structure
3133 * Returns 0 on success, negative on failure
3135 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3137 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3138 struct ixgbe_hw *hw = &adapter->hw;
3139 struct sockaddr *addr = p;
3141 if (!is_valid_ether_addr(addr->sa_data))
3142 return -EADDRNOTAVAIL;
3144 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3145 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3147 spin_lock_bh(&adapter->mbx_lock);
3149 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3151 spin_unlock_bh(&adapter->mbx_lock);
3157 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3158 * @netdev: network interface device structure
3159 * @new_mtu: new value for maximum frame size
3161 * Returns 0 on success, negative on failure
3163 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3165 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3166 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3167 int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3169 switch (adapter->hw.api_version) {
3170 case ixgbe_mbox_api_11:
3171 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3174 if (adapter->hw.mac.type == ixgbe_mac_X540_vf)
3175 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3179 /* MTU < 68 is an error and causes problems on some kernels */
3180 if ((new_mtu < 68) || (max_frame > max_possible_frame))
3183 hw_dbg(&adapter->hw, "changing MTU from %d to %d\n",
3184 netdev->mtu, new_mtu);
3185 /* must set new MTU before calling down or up */
3186 netdev->mtu = new_mtu;
3188 if (netif_running(netdev))
3189 ixgbevf_reinit_locked(adapter);
3194 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3196 struct net_device *netdev = pci_get_drvdata(pdev);
3197 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3202 netif_device_detach(netdev);
3204 if (netif_running(netdev)) {
3206 ixgbevf_down(adapter);
3207 ixgbevf_free_irq(adapter);
3208 ixgbevf_free_all_tx_resources(adapter);
3209 ixgbevf_free_all_rx_resources(adapter);
3213 ixgbevf_clear_interrupt_scheme(adapter);
3216 retval = pci_save_state(pdev);
3221 pci_disable_device(pdev);
3227 static int ixgbevf_resume(struct pci_dev *pdev)
3229 struct ixgbevf_adapter *adapter = pci_get_drvdata(pdev);
3230 struct net_device *netdev = adapter->netdev;
3233 pci_set_power_state(pdev, PCI_D0);
3234 pci_restore_state(pdev);
3236 * pci_restore_state clears dev->state_saved so call
3237 * pci_save_state to restore it.
3239 pci_save_state(pdev);
3241 err = pci_enable_device_mem(pdev);
3243 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3246 pci_set_master(pdev);
3248 ixgbevf_reset(adapter);
3251 err = ixgbevf_init_interrupt_scheme(adapter);
3254 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3258 if (netif_running(netdev)) {
3259 err = ixgbevf_open(netdev);
3264 netif_device_attach(netdev);
3269 #endif /* CONFIG_PM */
3270 static void ixgbevf_shutdown(struct pci_dev *pdev)
3272 ixgbevf_suspend(pdev, PMSG_SUSPEND);
3275 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3276 struct rtnl_link_stats64 *stats)
3278 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3281 const struct ixgbevf_ring *ring;
3284 ixgbevf_update_stats(adapter);
3286 stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3288 for (i = 0; i < adapter->num_rx_queues; i++) {
3289 ring = &adapter->rx_ring[i];
3291 start = u64_stats_fetch_begin_bh(&ring->syncp);
3292 bytes = ring->total_bytes;
3293 packets = ring->total_packets;
3294 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3295 stats->rx_bytes += bytes;
3296 stats->rx_packets += packets;
3299 for (i = 0; i < adapter->num_tx_queues; i++) {
3300 ring = &adapter->tx_ring[i];
3302 start = u64_stats_fetch_begin_bh(&ring->syncp);
3303 bytes = ring->total_bytes;
3304 packets = ring->total_packets;
3305 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3306 stats->tx_bytes += bytes;
3307 stats->tx_packets += packets;
3313 static const struct net_device_ops ixgbevf_netdev_ops = {
3314 .ndo_open = ixgbevf_open,
3315 .ndo_stop = ixgbevf_close,
3316 .ndo_start_xmit = ixgbevf_xmit_frame,
3317 .ndo_set_rx_mode = ixgbevf_set_rx_mode,
3318 .ndo_get_stats64 = ixgbevf_get_stats,
3319 .ndo_validate_addr = eth_validate_addr,
3320 .ndo_set_mac_address = ixgbevf_set_mac,
3321 .ndo_change_mtu = ixgbevf_change_mtu,
3322 .ndo_tx_timeout = ixgbevf_tx_timeout,
3323 .ndo_vlan_rx_add_vid = ixgbevf_vlan_rx_add_vid,
3324 .ndo_vlan_rx_kill_vid = ixgbevf_vlan_rx_kill_vid,
3327 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3329 dev->netdev_ops = &ixgbevf_netdev_ops;
3330 ixgbevf_set_ethtool_ops(dev);
3331 dev->watchdog_timeo = 5 * HZ;
3335 * ixgbevf_probe - Device Initialization Routine
3336 * @pdev: PCI device information struct
3337 * @ent: entry in ixgbevf_pci_tbl
3339 * Returns 0 on success, negative on failure
3341 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3342 * The OS initialization, configuring of the adapter private structure,
3343 * and a hardware reset occur.
3345 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3347 struct net_device *netdev;
3348 struct ixgbevf_adapter *adapter = NULL;
3349 struct ixgbe_hw *hw = NULL;
3350 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3351 static int cards_found;
3352 int err, pci_using_dac;
3354 err = pci_enable_device(pdev);
3358 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
3359 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
3362 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
3364 err = dma_set_coherent_mask(&pdev->dev,
3367 dev_err(&pdev->dev, "No usable DMA "
3368 "configuration, aborting\n");
3375 err = pci_request_regions(pdev, ixgbevf_driver_name);
3377 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3381 pci_set_master(pdev);
3383 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3387 goto err_alloc_etherdev;
3390 SET_NETDEV_DEV(netdev, &pdev->dev);
3392 pci_set_drvdata(pdev, netdev);
3393 adapter = netdev_priv(netdev);
3395 adapter->netdev = netdev;
3396 adapter->pdev = pdev;
3399 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3402 * call save state here in standalone driver because it relies on
3403 * adapter struct to exist, and needs to call netdev_priv
3405 pci_save_state(pdev);
3407 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3408 pci_resource_len(pdev, 0));
3414 ixgbevf_assign_netdev_ops(netdev);
3416 adapter->bd_number = cards_found;
3419 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3420 hw->mac.type = ii->mac;
3422 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3423 sizeof(struct ixgbe_mbx_operations));
3425 /* setup the private structure */
3426 err = ixgbevf_sw_init(adapter);
3430 /* The HW MAC address was set and/or determined in sw_init */
3431 if (!is_valid_ether_addr(netdev->dev_addr)) {
3432 pr_err("invalid MAC address\n");
3437 netdev->hw_features = NETIF_F_SG |
3444 netdev->features = netdev->hw_features |
3445 NETIF_F_HW_VLAN_CTAG_TX |
3446 NETIF_F_HW_VLAN_CTAG_RX |
3447 NETIF_F_HW_VLAN_CTAG_FILTER;
3449 netdev->vlan_features |= NETIF_F_TSO;
3450 netdev->vlan_features |= NETIF_F_TSO6;
3451 netdev->vlan_features |= NETIF_F_IP_CSUM;
3452 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3453 netdev->vlan_features |= NETIF_F_SG;
3456 netdev->features |= NETIF_F_HIGHDMA;
3458 netdev->priv_flags |= IFF_UNICAST_FLT;
3460 init_timer(&adapter->watchdog_timer);
3461 adapter->watchdog_timer.function = ixgbevf_watchdog;
3462 adapter->watchdog_timer.data = (unsigned long)adapter;
3464 INIT_WORK(&adapter->reset_task, ixgbevf_reset_task);
3465 INIT_WORK(&adapter->watchdog_task, ixgbevf_watchdog_task);
3467 err = ixgbevf_init_interrupt_scheme(adapter);
3471 strcpy(netdev->name, "eth%d");
3473 err = register_netdev(netdev);
3477 netif_carrier_off(netdev);
3479 ixgbevf_init_last_counter_stats(adapter);
3481 /* print the MAC address */
3482 hw_dbg(hw, "%pM\n", netdev->dev_addr);
3484 hw_dbg(hw, "MAC: %d\n", hw->mac.type);
3486 hw_dbg(hw, "Intel(R) 82599 Virtual Function\n");
3491 ixgbevf_clear_interrupt_scheme(adapter);
3493 ixgbevf_reset_interrupt_capability(adapter);
3494 iounmap(hw->hw_addr);
3496 free_netdev(netdev);
3498 pci_release_regions(pdev);
3501 pci_disable_device(pdev);
3506 * ixgbevf_remove - Device Removal Routine
3507 * @pdev: PCI device information struct
3509 * ixgbevf_remove is called by the PCI subsystem to alert the driver
3510 * that it should release a PCI device. The could be caused by a
3511 * Hot-Plug event, or because the driver is going to be removed from
3514 static void ixgbevf_remove(struct pci_dev *pdev)
3516 struct net_device *netdev = pci_get_drvdata(pdev);
3517 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3519 set_bit(__IXGBEVF_DOWN, &adapter->state);
3521 del_timer_sync(&adapter->watchdog_timer);
3523 cancel_work_sync(&adapter->reset_task);
3524 cancel_work_sync(&adapter->watchdog_task);
3526 if (netdev->reg_state == NETREG_REGISTERED)
3527 unregister_netdev(netdev);
3529 ixgbevf_clear_interrupt_scheme(adapter);
3530 ixgbevf_reset_interrupt_capability(adapter);
3532 iounmap(adapter->hw.hw_addr);
3533 pci_release_regions(pdev);
3535 hw_dbg(&adapter->hw, "Remove complete\n");
3537 kfree(adapter->tx_ring);
3538 kfree(adapter->rx_ring);
3540 free_netdev(netdev);
3542 pci_disable_device(pdev);
3546 * ixgbevf_io_error_detected - called when PCI error is detected
3547 * @pdev: Pointer to PCI device
3548 * @state: The current pci connection state
3550 * This function is called after a PCI bus error affecting
3551 * this device has been detected.
3553 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
3554 pci_channel_state_t state)
3556 struct net_device *netdev = pci_get_drvdata(pdev);
3557 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3559 netif_device_detach(netdev);
3561 if (state == pci_channel_io_perm_failure)
3562 return PCI_ERS_RESULT_DISCONNECT;
3564 if (netif_running(netdev))
3565 ixgbevf_down(adapter);
3567 pci_disable_device(pdev);
3569 /* Request a slot slot reset. */
3570 return PCI_ERS_RESULT_NEED_RESET;
3574 * ixgbevf_io_slot_reset - called after the pci bus has been reset.
3575 * @pdev: Pointer to PCI device
3577 * Restart the card from scratch, as if from a cold-boot. Implementation
3578 * resembles the first-half of the ixgbevf_resume routine.
3580 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
3582 struct net_device *netdev = pci_get_drvdata(pdev);
3583 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3585 if (pci_enable_device_mem(pdev)) {
3587 "Cannot re-enable PCI device after reset.\n");
3588 return PCI_ERS_RESULT_DISCONNECT;
3591 pci_set_master(pdev);
3593 ixgbevf_reset(adapter);
3595 return PCI_ERS_RESULT_RECOVERED;
3599 * ixgbevf_io_resume - called when traffic can start flowing again.
3600 * @pdev: Pointer to PCI device
3602 * This callback is called when the error recovery driver tells us that
3603 * its OK to resume normal operation. Implementation resembles the
3604 * second-half of the ixgbevf_resume routine.
3606 static void ixgbevf_io_resume(struct pci_dev *pdev)
3608 struct net_device *netdev = pci_get_drvdata(pdev);
3609 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3611 if (netif_running(netdev))
3612 ixgbevf_up(adapter);
3614 netif_device_attach(netdev);
3617 /* PCI Error Recovery (ERS) */
3618 static const struct pci_error_handlers ixgbevf_err_handler = {
3619 .error_detected = ixgbevf_io_error_detected,
3620 .slot_reset = ixgbevf_io_slot_reset,
3621 .resume = ixgbevf_io_resume,
3624 static struct pci_driver ixgbevf_driver = {
3625 .name = ixgbevf_driver_name,
3626 .id_table = ixgbevf_pci_tbl,
3627 .probe = ixgbevf_probe,
3628 .remove = ixgbevf_remove,
3630 /* Power Management Hooks */
3631 .suspend = ixgbevf_suspend,
3632 .resume = ixgbevf_resume,
3634 .shutdown = ixgbevf_shutdown,
3635 .err_handler = &ixgbevf_err_handler
3639 * ixgbevf_init_module - Driver Registration Routine
3641 * ixgbevf_init_module is the first routine called when the driver is
3642 * loaded. All it does is register with the PCI subsystem.
3644 static int __init ixgbevf_init_module(void)
3647 pr_info("%s - version %s\n", ixgbevf_driver_string,
3648 ixgbevf_driver_version);
3650 pr_info("%s\n", ixgbevf_copyright);
3652 ret = pci_register_driver(&ixgbevf_driver);
3656 module_init(ixgbevf_init_module);
3659 * ixgbevf_exit_module - Driver Exit Cleanup Routine
3661 * ixgbevf_exit_module is called just before the driver is removed
3664 static void __exit ixgbevf_exit_module(void)
3666 pci_unregister_driver(&ixgbevf_driver);
3671 * ixgbevf_get_hw_dev_name - return device name string
3672 * used by hardware layer to print debugging information
3674 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
3676 struct ixgbevf_adapter *adapter = hw->back;
3677 return adapter->netdev->name;
3681 module_exit(ixgbevf_exit_module);
3683 /* ixgbevf_main.c */