ixgbevf: fold ixgbevf_pull_tail into ixgbevf_add_rx_frag
[linux-2.6/btrfs-unstable.git] / drivers / net / ethernet / intel / ixgbevf / ixgbevf_main.c
blobacfa05154436cc75dac90887c813b4053419d1d9
1 /*******************************************************************************
3 Intel 82599 Virtual Function driver
4 Copyright(c) 1999 - 2015 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
13 more details.
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, see <http://www.gnu.org/licenses/>.
18 The full GNU General Public License is included in this distribution in
19 the file called "COPYING".
21 Contact Information:
22 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 *******************************************************************************/
27 /******************************************************************************
28 Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
29 ******************************************************************************/
31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33 #include <linux/types.h>
34 #include <linux/bitops.h>
35 #include <linux/module.h>
36 #include <linux/pci.h>
37 #include <linux/netdevice.h>
38 #include <linux/vmalloc.h>
39 #include <linux/string.h>
40 #include <linux/in.h>
41 #include <linux/ip.h>
42 #include <linux/tcp.h>
43 #include <linux/sctp.h>
44 #include <linux/ipv6.h>
45 #include <linux/slab.h>
46 #include <net/checksum.h>
47 #include <net/ip6_checksum.h>
48 #include <linux/ethtool.h>
49 #include <linux/if.h>
50 #include <linux/if_vlan.h>
51 #include <linux/prefetch.h>
53 #include "ixgbevf.h"
55 const char ixgbevf_driver_name[] = "ixgbevf";
56 static const char ixgbevf_driver_string[] =
57 "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
59 #define DRV_VERSION "2.12.1-k"
60 const char ixgbevf_driver_version[] = DRV_VERSION;
61 static char ixgbevf_copyright[] =
62 "Copyright (c) 2009 - 2012 Intel Corporation.";
64 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
65 [board_82599_vf] = &ixgbevf_82599_vf_info,
66 [board_X540_vf] = &ixgbevf_X540_vf_info,
67 [board_X550_vf] = &ixgbevf_X550_vf_info,
68 [board_X550EM_x_vf] = &ixgbevf_X550EM_x_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 const struct pci_device_id 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 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
83 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
84 /* required last entry */
85 {0, }
87 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
89 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
90 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
91 MODULE_LICENSE("GPL");
92 MODULE_VERSION(DRV_VERSION);
94 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
95 static int debug = -1;
96 module_param(debug, int, 0);
97 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
99 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
101 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
102 !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
103 !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
104 schedule_work(&adapter->service_task);
107 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
109 BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
111 /* flush memory to make sure state is correct before next watchdog */
112 smp_mb__before_atomic();
113 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
116 /* forward decls */
117 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
118 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
119 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
121 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
123 struct ixgbevf_adapter *adapter = hw->back;
125 if (!hw->hw_addr)
126 return;
127 hw->hw_addr = NULL;
128 dev_err(&adapter->pdev->dev, "Adapter removed\n");
129 if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
130 ixgbevf_service_event_schedule(adapter);
133 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
135 u32 value;
137 /* The following check not only optimizes a bit by not
138 * performing a read on the status register when the
139 * register just read was a status register read that
140 * returned IXGBE_FAILED_READ_REG. It also blocks any
141 * potential recursion.
143 if (reg == IXGBE_VFSTATUS) {
144 ixgbevf_remove_adapter(hw);
145 return;
147 value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
148 if (value == IXGBE_FAILED_READ_REG)
149 ixgbevf_remove_adapter(hw);
152 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
154 u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
155 u32 value;
157 if (IXGBE_REMOVED(reg_addr))
158 return IXGBE_FAILED_READ_REG;
159 value = readl(reg_addr + reg);
160 if (unlikely(value == IXGBE_FAILED_READ_REG))
161 ixgbevf_check_remove(hw, reg);
162 return value;
166 * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
167 * @adapter: pointer to adapter struct
168 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
169 * @queue: queue to map the corresponding interrupt to
170 * @msix_vector: the vector to map to the corresponding queue
172 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
173 u8 queue, u8 msix_vector)
175 u32 ivar, index;
176 struct ixgbe_hw *hw = &adapter->hw;
178 if (direction == -1) {
179 /* other causes */
180 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
181 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
182 ivar &= ~0xFF;
183 ivar |= msix_vector;
184 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
185 } else {
186 /* Tx or Rx causes */
187 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
188 index = ((16 * (queue & 1)) + (8 * direction));
189 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
190 ivar &= ~(0xFF << index);
191 ivar |= (msix_vector << index);
192 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
196 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
197 struct ixgbevf_tx_buffer *tx_buffer)
199 if (tx_buffer->skb) {
200 dev_kfree_skb_any(tx_buffer->skb);
201 if (dma_unmap_len(tx_buffer, len))
202 dma_unmap_single(tx_ring->dev,
203 dma_unmap_addr(tx_buffer, dma),
204 dma_unmap_len(tx_buffer, len),
205 DMA_TO_DEVICE);
206 } else if (dma_unmap_len(tx_buffer, len)) {
207 dma_unmap_page(tx_ring->dev,
208 dma_unmap_addr(tx_buffer, dma),
209 dma_unmap_len(tx_buffer, len),
210 DMA_TO_DEVICE);
212 tx_buffer->next_to_watch = NULL;
213 tx_buffer->skb = NULL;
214 dma_unmap_len_set(tx_buffer, len, 0);
215 /* tx_buffer must be completely set up in the transmit path */
218 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
220 return ring->stats.packets;
223 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
225 struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
226 struct ixgbe_hw *hw = &adapter->hw;
228 u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
229 u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
231 if (head != tail)
232 return (head < tail) ?
233 tail - head : (tail + ring->count - head);
235 return 0;
238 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
240 u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
241 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
242 u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
244 clear_check_for_tx_hang(tx_ring);
246 /* Check for a hung queue, but be thorough. This verifies
247 * that a transmit has been completed since the previous
248 * check AND there is at least one packet pending. The
249 * ARMED bit is set to indicate a potential hang.
251 if ((tx_done_old == tx_done) && tx_pending) {
252 /* make sure it is true for two checks in a row */
253 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
254 &tx_ring->state);
256 /* reset the countdown */
257 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
259 /* update completed stats and continue */
260 tx_ring->tx_stats.tx_done_old = tx_done;
262 return false;
265 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
267 /* Do the reset outside of interrupt context */
268 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
269 adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
270 ixgbevf_service_event_schedule(adapter);
275 * ixgbevf_tx_timeout - Respond to a Tx Hang
276 * @netdev: network interface device structure
278 static void ixgbevf_tx_timeout(struct net_device *netdev)
280 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
282 ixgbevf_tx_timeout_reset(adapter);
286 * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
287 * @q_vector: board private structure
288 * @tx_ring: tx ring to clean
290 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
291 struct ixgbevf_ring *tx_ring)
293 struct ixgbevf_adapter *adapter = q_vector->adapter;
294 struct ixgbevf_tx_buffer *tx_buffer;
295 union ixgbe_adv_tx_desc *tx_desc;
296 unsigned int total_bytes = 0, total_packets = 0;
297 unsigned int budget = tx_ring->count / 2;
298 unsigned int i = tx_ring->next_to_clean;
300 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
301 return true;
303 tx_buffer = &tx_ring->tx_buffer_info[i];
304 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
305 i -= tx_ring->count;
307 do {
308 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
310 /* if next_to_watch is not set then there is no work pending */
311 if (!eop_desc)
312 break;
314 /* prevent any other reads prior to eop_desc */
315 read_barrier_depends();
317 /* if DD is not set pending work has not been completed */
318 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
319 break;
321 /* clear next_to_watch to prevent false hangs */
322 tx_buffer->next_to_watch = NULL;
324 /* update the statistics for this packet */
325 total_bytes += tx_buffer->bytecount;
326 total_packets += tx_buffer->gso_segs;
328 /* free the skb */
329 dev_kfree_skb_any(tx_buffer->skb);
331 /* unmap skb header data */
332 dma_unmap_single(tx_ring->dev,
333 dma_unmap_addr(tx_buffer, dma),
334 dma_unmap_len(tx_buffer, len),
335 DMA_TO_DEVICE);
337 /* clear tx_buffer data */
338 tx_buffer->skb = NULL;
339 dma_unmap_len_set(tx_buffer, len, 0);
341 /* unmap remaining buffers */
342 while (tx_desc != eop_desc) {
343 tx_buffer++;
344 tx_desc++;
345 i++;
346 if (unlikely(!i)) {
347 i -= tx_ring->count;
348 tx_buffer = tx_ring->tx_buffer_info;
349 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
352 /* unmap any remaining paged data */
353 if (dma_unmap_len(tx_buffer, len)) {
354 dma_unmap_page(tx_ring->dev,
355 dma_unmap_addr(tx_buffer, dma),
356 dma_unmap_len(tx_buffer, len),
357 DMA_TO_DEVICE);
358 dma_unmap_len_set(tx_buffer, len, 0);
362 /* move us one more past the eop_desc for start of next pkt */
363 tx_buffer++;
364 tx_desc++;
365 i++;
366 if (unlikely(!i)) {
367 i -= tx_ring->count;
368 tx_buffer = tx_ring->tx_buffer_info;
369 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
372 /* issue prefetch for next Tx descriptor */
373 prefetch(tx_desc);
375 /* update budget accounting */
376 budget--;
377 } while (likely(budget));
379 i += tx_ring->count;
380 tx_ring->next_to_clean = i;
381 u64_stats_update_begin(&tx_ring->syncp);
382 tx_ring->stats.bytes += total_bytes;
383 tx_ring->stats.packets += total_packets;
384 u64_stats_update_end(&tx_ring->syncp);
385 q_vector->tx.total_bytes += total_bytes;
386 q_vector->tx.total_packets += total_packets;
388 if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
389 struct ixgbe_hw *hw = &adapter->hw;
390 union ixgbe_adv_tx_desc *eop_desc;
392 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
394 pr_err("Detected Tx Unit Hang\n"
395 " Tx Queue <%d>\n"
396 " TDH, TDT <%x>, <%x>\n"
397 " next_to_use <%x>\n"
398 " next_to_clean <%x>\n"
399 "tx_buffer_info[next_to_clean]\n"
400 " next_to_watch <%p>\n"
401 " eop_desc->wb.status <%x>\n"
402 " time_stamp <%lx>\n"
403 " jiffies <%lx>\n",
404 tx_ring->queue_index,
405 IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
406 IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
407 tx_ring->next_to_use, i,
408 eop_desc, (eop_desc ? eop_desc->wb.status : 0),
409 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
411 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
413 /* schedule immediate reset if we believe we hung */
414 ixgbevf_tx_timeout_reset(adapter);
416 return true;
419 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
420 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
421 (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
422 /* Make sure that anybody stopping the queue after this
423 * sees the new next_to_clean.
425 smp_mb();
427 if (__netif_subqueue_stopped(tx_ring->netdev,
428 tx_ring->queue_index) &&
429 !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
430 netif_wake_subqueue(tx_ring->netdev,
431 tx_ring->queue_index);
432 ++tx_ring->tx_stats.restart_queue;
436 return !!budget;
440 * ixgbevf_rx_skb - Helper function to determine proper Rx method
441 * @q_vector: structure containing interrupt and ring information
442 * @skb: packet to send up
444 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
445 struct sk_buff *skb)
447 #ifdef CONFIG_NET_RX_BUSY_POLL
448 skb_mark_napi_id(skb, &q_vector->napi);
450 if (ixgbevf_qv_busy_polling(q_vector)) {
451 netif_receive_skb(skb);
452 /* exit early if we busy polled */
453 return;
455 #endif /* CONFIG_NET_RX_BUSY_POLL */
457 napi_gro_receive(&q_vector->napi, skb);
461 * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
462 * @ring: structure containig ring specific data
463 * @rx_desc: current Rx descriptor being processed
464 * @skb: skb currently being received and modified
466 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
467 union ixgbe_adv_rx_desc *rx_desc,
468 struct sk_buff *skb)
470 skb_checksum_none_assert(skb);
472 /* Rx csum disabled */
473 if (!(ring->netdev->features & NETIF_F_RXCSUM))
474 return;
476 /* if IP and error */
477 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
478 ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
479 ring->rx_stats.csum_err++;
480 return;
483 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
484 return;
486 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
487 ring->rx_stats.csum_err++;
488 return;
491 /* It must be a TCP or UDP packet with a valid checksum */
492 skb->ip_summed = CHECKSUM_UNNECESSARY;
496 * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
497 * @rx_ring: rx descriptor ring packet is being transacted on
498 * @rx_desc: pointer to the EOP Rx descriptor
499 * @skb: pointer to current skb being populated
501 * This function checks the ring, descriptor, and packet information in
502 * order to populate the checksum, VLAN, protocol, and other fields within
503 * the skb.
505 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
506 union ixgbe_adv_rx_desc *rx_desc,
507 struct sk_buff *skb)
509 ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
511 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
512 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
513 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
515 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
516 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
519 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
523 * ixgbevf_is_non_eop - process handling of non-EOP buffers
524 * @rx_ring: Rx ring being processed
525 * @rx_desc: Rx descriptor for current buffer
526 * @skb: current socket buffer containing buffer in progress
528 * This function updates next to clean. If the buffer is an EOP buffer
529 * this function exits returning false, otherwise it will place the
530 * sk_buff in the next buffer to be chained and return true indicating
531 * that this is in fact a non-EOP buffer.
533 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
534 union ixgbe_adv_rx_desc *rx_desc)
536 u32 ntc = rx_ring->next_to_clean + 1;
538 /* fetch, update, and store next to clean */
539 ntc = (ntc < rx_ring->count) ? ntc : 0;
540 rx_ring->next_to_clean = ntc;
542 prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
544 if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
545 return false;
547 return true;
550 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
551 struct ixgbevf_rx_buffer *bi)
553 struct page *page = bi->page;
554 dma_addr_t dma = bi->dma;
556 /* since we are recycling buffers we should seldom need to alloc */
557 if (likely(page))
558 return true;
560 /* alloc new page for storage */
561 page = dev_alloc_page();
562 if (unlikely(!page)) {
563 rx_ring->rx_stats.alloc_rx_page_failed++;
564 return false;
567 /* map page for use */
568 dma = dma_map_page(rx_ring->dev, page, 0,
569 PAGE_SIZE, DMA_FROM_DEVICE);
571 /* if mapping failed free memory back to system since
572 * there isn't much point in holding memory we can't use
574 if (dma_mapping_error(rx_ring->dev, dma)) {
575 __free_page(page);
577 rx_ring->rx_stats.alloc_rx_buff_failed++;
578 return false;
581 bi->dma = dma;
582 bi->page = page;
583 bi->page_offset = 0;
585 return true;
589 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
590 * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
591 * @cleaned_count: number of buffers to replace
593 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
594 u16 cleaned_count)
596 union ixgbe_adv_rx_desc *rx_desc;
597 struct ixgbevf_rx_buffer *bi;
598 unsigned int i = rx_ring->next_to_use;
600 /* nothing to do or no valid netdev defined */
601 if (!cleaned_count || !rx_ring->netdev)
602 return;
604 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
605 bi = &rx_ring->rx_buffer_info[i];
606 i -= rx_ring->count;
608 do {
609 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
610 break;
612 /* Refresh the desc even if pkt_addr didn't change
613 * because each write-back erases this info.
615 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
617 rx_desc++;
618 bi++;
619 i++;
620 if (unlikely(!i)) {
621 rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
622 bi = rx_ring->rx_buffer_info;
623 i -= rx_ring->count;
626 /* clear the hdr_addr for the next_to_use descriptor */
627 rx_desc->read.hdr_addr = 0;
629 cleaned_count--;
630 } while (cleaned_count);
632 i += rx_ring->count;
634 if (rx_ring->next_to_use != i) {
635 /* record the next descriptor to use */
636 rx_ring->next_to_use = i;
638 /* update next to alloc since we have filled the ring */
639 rx_ring->next_to_alloc = i;
641 /* Force memory writes to complete before letting h/w
642 * know there are new descriptors to fetch. (Only
643 * applicable for weak-ordered memory model archs,
644 * such as IA-64).
646 wmb();
647 ixgbevf_write_tail(rx_ring, i);
652 * ixgbevf_cleanup_headers - Correct corrupted or empty headers
653 * @rx_ring: rx descriptor ring packet is being transacted on
654 * @rx_desc: pointer to the EOP Rx descriptor
655 * @skb: pointer to current skb being fixed
657 * Check for corrupted packet headers caused by senders on the local L2
658 * embedded NIC switch not setting up their Tx Descriptors right. These
659 * should be very rare.
661 * Also address the case where we are pulling data in on pages only
662 * and as such no data is present in the skb header.
664 * In addition if skb is not at least 60 bytes we need to pad it so that
665 * it is large enough to qualify as a valid Ethernet frame.
667 * Returns true if an error was encountered and skb was freed.
669 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
670 union ixgbe_adv_rx_desc *rx_desc,
671 struct sk_buff *skb)
673 /* verify that the packet does not have any known errors */
674 if (unlikely(ixgbevf_test_staterr(rx_desc,
675 IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
676 struct net_device *netdev = rx_ring->netdev;
678 if (!(netdev->features & NETIF_F_RXALL)) {
679 dev_kfree_skb_any(skb);
680 return true;
684 /* if eth_skb_pad returns an error the skb was freed */
685 if (eth_skb_pad(skb))
686 return true;
688 return false;
692 * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
693 * @rx_ring: rx descriptor ring to store buffers on
694 * @old_buff: donor buffer to have page reused
696 * Synchronizes page for reuse by the adapter
698 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
699 struct ixgbevf_rx_buffer *old_buff)
701 struct ixgbevf_rx_buffer *new_buff;
702 u16 nta = rx_ring->next_to_alloc;
704 new_buff = &rx_ring->rx_buffer_info[nta];
706 /* update, and store next to alloc */
707 nta++;
708 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
710 /* transfer page from old buffer to new buffer */
711 new_buff->page = old_buff->page;
712 new_buff->dma = old_buff->dma;
713 new_buff->page_offset = old_buff->page_offset;
715 /* sync the buffer for use by the device */
716 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
717 new_buff->page_offset,
718 IXGBEVF_RX_BUFSZ,
719 DMA_FROM_DEVICE);
722 static inline bool ixgbevf_page_is_reserved(struct page *page)
724 return (page_to_nid(page) != numa_mem_id()) || page->pfmemalloc;
728 * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
729 * @rx_ring: rx descriptor ring to transact packets on
730 * @rx_buffer: buffer containing page to add
731 * @rx_desc: descriptor containing length of buffer written by hardware
732 * @skb: sk_buff to place the data into
734 * This function will add the data contained in rx_buffer->page to the skb.
735 * This is done either through a direct copy if the data in the buffer is
736 * less than the skb header size, otherwise it will just attach the page as
737 * a frag to the skb.
739 * The function will then update the page offset if necessary and return
740 * true if the buffer can be reused by the adapter.
742 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
743 struct ixgbevf_rx_buffer *rx_buffer,
744 union ixgbe_adv_rx_desc *rx_desc,
745 struct sk_buff *skb)
747 struct page *page = rx_buffer->page;
748 unsigned char *va = page_address(page) + rx_buffer->page_offset;
749 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
750 #if (PAGE_SIZE < 8192)
751 unsigned int truesize = IXGBEVF_RX_BUFSZ;
752 #else
753 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
754 #endif
755 unsigned int pull_len;
757 if (unlikely(skb_is_nonlinear(skb)))
758 goto add_tail_frag;
760 if (likely(size <= IXGBEVF_RX_HDR_SIZE)) {
761 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
763 /* page is not reserved, we can reuse buffer as is */
764 if (likely(!ixgbevf_page_is_reserved(page)))
765 return true;
767 /* this page cannot be reused so discard it */
768 put_page(page);
769 return false;
772 /* we need the header to contain the greater of either ETH_HLEN or
773 * 60 bytes if the skb->len is less than 60 for skb_pad.
775 pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
777 /* align pull length to size of long to optimize memcpy performance */
778 memcpy(__skb_put(skb, pull_len), va, ALIGN(pull_len, sizeof(long)));
780 /* update all of the pointers */
781 va += pull_len;
782 size -= pull_len;
784 add_tail_frag:
785 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
786 (unsigned long)va & ~PAGE_MASK, size, truesize);
788 /* avoid re-using remote pages */
789 if (unlikely(ixgbevf_page_is_reserved(page)))
790 return false;
792 #if (PAGE_SIZE < 8192)
793 /* if we are only owner of page we can reuse it */
794 if (unlikely(page_count(page) != 1))
795 return false;
797 /* flip page offset to other buffer */
798 rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
800 #else
801 /* move offset up to the next cache line */
802 rx_buffer->page_offset += truesize;
804 if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
805 return false;
807 #endif
808 /* Even if we own the page, we are not allowed to use atomic_set()
809 * This would break get_page_unless_zero() users.
811 atomic_inc(&page->_count);
813 return true;
816 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
817 union ixgbe_adv_rx_desc *rx_desc,
818 struct sk_buff *skb)
820 struct ixgbevf_rx_buffer *rx_buffer;
821 struct page *page;
823 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
824 page = rx_buffer->page;
825 prefetchw(page);
827 if (likely(!skb)) {
828 void *page_addr = page_address(page) +
829 rx_buffer->page_offset;
831 /* prefetch first cache line of first page */
832 prefetch(page_addr);
833 #if L1_CACHE_BYTES < 128
834 prefetch(page_addr + L1_CACHE_BYTES);
835 #endif
837 /* allocate a skb to store the frags */
838 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
839 IXGBEVF_RX_HDR_SIZE);
840 if (unlikely(!skb)) {
841 rx_ring->rx_stats.alloc_rx_buff_failed++;
842 return NULL;
845 /* we will be copying header into skb->data in
846 * pskb_may_pull so it is in our interest to prefetch
847 * it now to avoid a possible cache miss
849 prefetchw(skb->data);
852 /* we are reusing so sync this buffer for CPU use */
853 dma_sync_single_range_for_cpu(rx_ring->dev,
854 rx_buffer->dma,
855 rx_buffer->page_offset,
856 IXGBEVF_RX_BUFSZ,
857 DMA_FROM_DEVICE);
859 /* pull page into skb */
860 if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
861 /* hand second half of page back to the ring */
862 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
863 } else {
864 /* we are not reusing the buffer so unmap it */
865 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
866 PAGE_SIZE, DMA_FROM_DEVICE);
869 /* clear contents of buffer_info */
870 rx_buffer->dma = 0;
871 rx_buffer->page = NULL;
873 return skb;
876 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
877 u32 qmask)
879 struct ixgbe_hw *hw = &adapter->hw;
881 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
884 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
885 struct ixgbevf_ring *rx_ring,
886 int budget)
888 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
889 u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
890 struct sk_buff *skb = rx_ring->skb;
892 while (likely(total_rx_packets < budget)) {
893 union ixgbe_adv_rx_desc *rx_desc;
895 /* return some buffers to hardware, one at a time is too slow */
896 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
897 ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
898 cleaned_count = 0;
901 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
903 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
904 break;
906 /* This memory barrier is needed to keep us from reading
907 * any other fields out of the rx_desc until we know the
908 * RXD_STAT_DD bit is set
910 rmb();
912 /* retrieve a buffer from the ring */
913 skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
915 /* exit if we failed to retrieve a buffer */
916 if (!skb)
917 break;
919 cleaned_count++;
921 /* fetch next buffer in frame if non-eop */
922 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
923 continue;
925 /* verify the packet layout is correct */
926 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
927 skb = NULL;
928 continue;
931 /* probably a little skewed due to removing CRC */
932 total_rx_bytes += skb->len;
934 /* Workaround hardware that can't do proper VEPA multicast
935 * source pruning.
937 if ((skb->pkt_type == PACKET_BROADCAST ||
938 skb->pkt_type == PACKET_MULTICAST) &&
939 ether_addr_equal(rx_ring->netdev->dev_addr,
940 eth_hdr(skb)->h_source)) {
941 dev_kfree_skb_irq(skb);
942 continue;
945 /* populate checksum, VLAN, and protocol */
946 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
948 ixgbevf_rx_skb(q_vector, skb);
950 /* reset skb pointer */
951 skb = NULL;
953 /* update budget accounting */
954 total_rx_packets++;
957 /* place incomplete frames back on ring for completion */
958 rx_ring->skb = skb;
960 u64_stats_update_begin(&rx_ring->syncp);
961 rx_ring->stats.packets += total_rx_packets;
962 rx_ring->stats.bytes += total_rx_bytes;
963 u64_stats_update_end(&rx_ring->syncp);
964 q_vector->rx.total_packets += total_rx_packets;
965 q_vector->rx.total_bytes += total_rx_bytes;
967 return total_rx_packets;
971 * ixgbevf_poll - NAPI polling calback
972 * @napi: napi struct with our devices info in it
973 * @budget: amount of work driver is allowed to do this pass, in packets
975 * This function will clean more than one or more rings associated with a
976 * q_vector.
978 static int ixgbevf_poll(struct napi_struct *napi, int budget)
980 struct ixgbevf_q_vector *q_vector =
981 container_of(napi, struct ixgbevf_q_vector, napi);
982 struct ixgbevf_adapter *adapter = q_vector->adapter;
983 struct ixgbevf_ring *ring;
984 int per_ring_budget;
985 bool clean_complete = true;
987 ixgbevf_for_each_ring(ring, q_vector->tx)
988 clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
990 #ifdef CONFIG_NET_RX_BUSY_POLL
991 if (!ixgbevf_qv_lock_napi(q_vector))
992 return budget;
993 #endif
995 /* attempt to distribute budget to each queue fairly, but don't allow
996 * the budget to go below 1 because we'll exit polling
998 if (q_vector->rx.count > 1)
999 per_ring_budget = max(budget/q_vector->rx.count, 1);
1000 else
1001 per_ring_budget = budget;
1003 ixgbevf_for_each_ring(ring, q_vector->rx)
1004 clean_complete &= (ixgbevf_clean_rx_irq(q_vector, ring,
1005 per_ring_budget)
1006 < per_ring_budget);
1008 #ifdef CONFIG_NET_RX_BUSY_POLL
1009 ixgbevf_qv_unlock_napi(q_vector);
1010 #endif
1012 /* If all work not completed, return budget and keep polling */
1013 if (!clean_complete)
1014 return budget;
1015 /* all work done, exit the polling mode */
1016 napi_complete(napi);
1017 if (adapter->rx_itr_setting & 1)
1018 ixgbevf_set_itr(q_vector);
1019 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1020 !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1021 ixgbevf_irq_enable_queues(adapter,
1022 1 << q_vector->v_idx);
1024 return 0;
1028 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1029 * @q_vector: structure containing interrupt and ring information
1031 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1033 struct ixgbevf_adapter *adapter = q_vector->adapter;
1034 struct ixgbe_hw *hw = &adapter->hw;
1035 int v_idx = q_vector->v_idx;
1036 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1038 /* set the WDIS bit to not clear the timer bits and cause an
1039 * immediate assertion of the interrupt
1041 itr_reg |= IXGBE_EITR_CNT_WDIS;
1043 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1046 #ifdef CONFIG_NET_RX_BUSY_POLL
1047 /* must be called with local_bh_disable()d */
1048 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
1050 struct ixgbevf_q_vector *q_vector =
1051 container_of(napi, struct ixgbevf_q_vector, napi);
1052 struct ixgbevf_adapter *adapter = q_vector->adapter;
1053 struct ixgbevf_ring *ring;
1054 int found = 0;
1056 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
1057 return LL_FLUSH_FAILED;
1059 if (!ixgbevf_qv_lock_poll(q_vector))
1060 return LL_FLUSH_BUSY;
1062 ixgbevf_for_each_ring(ring, q_vector->rx) {
1063 found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
1064 #ifdef BP_EXTENDED_STATS
1065 if (found)
1066 ring->stats.cleaned += found;
1067 else
1068 ring->stats.misses++;
1069 #endif
1070 if (found)
1071 break;
1074 ixgbevf_qv_unlock_poll(q_vector);
1076 return found;
1078 #endif /* CONFIG_NET_RX_BUSY_POLL */
1081 * ixgbevf_configure_msix - Configure MSI-X hardware
1082 * @adapter: board private structure
1084 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1085 * interrupts.
1087 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1089 struct ixgbevf_q_vector *q_vector;
1090 int q_vectors, v_idx;
1092 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1093 adapter->eims_enable_mask = 0;
1095 /* Populate the IVAR table and set the ITR values to the
1096 * corresponding register.
1098 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1099 struct ixgbevf_ring *ring;
1101 q_vector = adapter->q_vector[v_idx];
1103 ixgbevf_for_each_ring(ring, q_vector->rx)
1104 ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1106 ixgbevf_for_each_ring(ring, q_vector->tx)
1107 ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1109 if (q_vector->tx.ring && !q_vector->rx.ring) {
1110 /* Tx only vector */
1111 if (adapter->tx_itr_setting == 1)
1112 q_vector->itr = IXGBE_10K_ITR;
1113 else
1114 q_vector->itr = adapter->tx_itr_setting;
1115 } else {
1116 /* Rx or Rx/Tx vector */
1117 if (adapter->rx_itr_setting == 1)
1118 q_vector->itr = IXGBE_20K_ITR;
1119 else
1120 q_vector->itr = adapter->rx_itr_setting;
1123 /* add q_vector eims value to global eims_enable_mask */
1124 adapter->eims_enable_mask |= 1 << v_idx;
1126 ixgbevf_write_eitr(q_vector);
1129 ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1130 /* setup eims_other and add value to global eims_enable_mask */
1131 adapter->eims_other = 1 << v_idx;
1132 adapter->eims_enable_mask |= adapter->eims_other;
1135 enum latency_range {
1136 lowest_latency = 0,
1137 low_latency = 1,
1138 bulk_latency = 2,
1139 latency_invalid = 255
1143 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1144 * @q_vector: structure containing interrupt and ring information
1145 * @ring_container: structure containing ring performance data
1147 * Stores a new ITR value based on packets and byte
1148 * counts during the last interrupt. The advantage of per interrupt
1149 * computation is faster updates and more accurate ITR for the current
1150 * traffic pattern. Constants in this function were computed
1151 * based on theoretical maximum wire speed and thresholds were set based
1152 * on testing data as well as attempting to minimize response time
1153 * while increasing bulk throughput.
1155 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1156 struct ixgbevf_ring_container *ring_container)
1158 int bytes = ring_container->total_bytes;
1159 int packets = ring_container->total_packets;
1160 u32 timepassed_us;
1161 u64 bytes_perint;
1162 u8 itr_setting = ring_container->itr;
1164 if (packets == 0)
1165 return;
1167 /* simple throttle rate management
1168 * 0-20MB/s lowest (100000 ints/s)
1169 * 20-100MB/s low (20000 ints/s)
1170 * 100-1249MB/s bulk (8000 ints/s)
1172 /* what was last interrupt timeslice? */
1173 timepassed_us = q_vector->itr >> 2;
1174 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1176 switch (itr_setting) {
1177 case lowest_latency:
1178 if (bytes_perint > 10)
1179 itr_setting = low_latency;
1180 break;
1181 case low_latency:
1182 if (bytes_perint > 20)
1183 itr_setting = bulk_latency;
1184 else if (bytes_perint <= 10)
1185 itr_setting = lowest_latency;
1186 break;
1187 case bulk_latency:
1188 if (bytes_perint <= 20)
1189 itr_setting = low_latency;
1190 break;
1193 /* clear work counters since we have the values we need */
1194 ring_container->total_bytes = 0;
1195 ring_container->total_packets = 0;
1197 /* write updated itr to ring container */
1198 ring_container->itr = itr_setting;
1201 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1203 u32 new_itr = q_vector->itr;
1204 u8 current_itr;
1206 ixgbevf_update_itr(q_vector, &q_vector->tx);
1207 ixgbevf_update_itr(q_vector, &q_vector->rx);
1209 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1211 switch (current_itr) {
1212 /* counts and packets in update_itr are dependent on these numbers */
1213 case lowest_latency:
1214 new_itr = IXGBE_100K_ITR;
1215 break;
1216 case low_latency:
1217 new_itr = IXGBE_20K_ITR;
1218 break;
1219 case bulk_latency:
1220 default:
1221 new_itr = IXGBE_8K_ITR;
1222 break;
1225 if (new_itr != q_vector->itr) {
1226 /* do an exponential smoothing */
1227 new_itr = (10 * new_itr * q_vector->itr) /
1228 ((9 * new_itr) + q_vector->itr);
1230 /* save the algorithm value here */
1231 q_vector->itr = new_itr;
1233 ixgbevf_write_eitr(q_vector);
1237 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1239 struct ixgbevf_adapter *adapter = data;
1240 struct ixgbe_hw *hw = &adapter->hw;
1242 hw->mac.get_link_status = 1;
1244 ixgbevf_service_event_schedule(adapter);
1246 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1248 return IRQ_HANDLED;
1252 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1253 * @irq: unused
1254 * @data: pointer to our q_vector struct for this interrupt vector
1256 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1258 struct ixgbevf_q_vector *q_vector = data;
1260 /* EIAM disabled interrupts (on this vector) for us */
1261 if (q_vector->rx.ring || q_vector->tx.ring)
1262 napi_schedule(&q_vector->napi);
1264 return IRQ_HANDLED;
1267 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1268 int r_idx)
1270 struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1272 a->rx_ring[r_idx]->next = q_vector->rx.ring;
1273 q_vector->rx.ring = a->rx_ring[r_idx];
1274 q_vector->rx.count++;
1277 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1278 int t_idx)
1280 struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1282 a->tx_ring[t_idx]->next = q_vector->tx.ring;
1283 q_vector->tx.ring = a->tx_ring[t_idx];
1284 q_vector->tx.count++;
1288 * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1289 * @adapter: board private structure to initialize
1291 * This function maps descriptor rings to the queue-specific vectors
1292 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1293 * one vector per ring/queue, but on a constrained vector budget, we
1294 * group the rings as "efficiently" as possible. You would add new
1295 * mapping configurations in here.
1297 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1299 int q_vectors;
1300 int v_start = 0;
1301 int rxr_idx = 0, txr_idx = 0;
1302 int rxr_remaining = adapter->num_rx_queues;
1303 int txr_remaining = adapter->num_tx_queues;
1304 int i, j;
1305 int rqpv, tqpv;
1306 int err = 0;
1308 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1310 /* The ideal configuration...
1311 * We have enough vectors to map one per queue.
1313 if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1314 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1315 map_vector_to_rxq(adapter, v_start, rxr_idx);
1317 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1318 map_vector_to_txq(adapter, v_start, txr_idx);
1319 goto out;
1322 /* If we don't have enough vectors for a 1-to-1
1323 * mapping, we'll have to group them so there are
1324 * multiple queues per vector.
1326 /* Re-adjusting *qpv takes care of the remainder. */
1327 for (i = v_start; i < q_vectors; i++) {
1328 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1329 for (j = 0; j < rqpv; j++) {
1330 map_vector_to_rxq(adapter, i, rxr_idx);
1331 rxr_idx++;
1332 rxr_remaining--;
1335 for (i = v_start; i < q_vectors; i++) {
1336 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1337 for (j = 0; j < tqpv; j++) {
1338 map_vector_to_txq(adapter, i, txr_idx);
1339 txr_idx++;
1340 txr_remaining--;
1344 out:
1345 return err;
1349 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1350 * @adapter: board private structure
1352 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1353 * interrupts from the kernel.
1355 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1357 struct net_device *netdev = adapter->netdev;
1358 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1359 int vector, err;
1360 int ri = 0, ti = 0;
1362 for (vector = 0; vector < q_vectors; vector++) {
1363 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1364 struct msix_entry *entry = &adapter->msix_entries[vector];
1366 if (q_vector->tx.ring && q_vector->rx.ring) {
1367 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1368 "%s-%s-%d", netdev->name, "TxRx", ri++);
1369 ti++;
1370 } else if (q_vector->rx.ring) {
1371 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1372 "%s-%s-%d", netdev->name, "rx", ri++);
1373 } else if (q_vector->tx.ring) {
1374 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1375 "%s-%s-%d", netdev->name, "tx", ti++);
1376 } else {
1377 /* skip this unused q_vector */
1378 continue;
1380 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1381 q_vector->name, q_vector);
1382 if (err) {
1383 hw_dbg(&adapter->hw,
1384 "request_irq failed for MSIX interrupt Error: %d\n",
1385 err);
1386 goto free_queue_irqs;
1390 err = request_irq(adapter->msix_entries[vector].vector,
1391 &ixgbevf_msix_other, 0, netdev->name, adapter);
1392 if (err) {
1393 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1394 err);
1395 goto free_queue_irqs;
1398 return 0;
1400 free_queue_irqs:
1401 while (vector) {
1402 vector--;
1403 free_irq(adapter->msix_entries[vector].vector,
1404 adapter->q_vector[vector]);
1406 /* This failure is non-recoverable - it indicates the system is
1407 * out of MSIX vector resources and the VF driver cannot run
1408 * without them. Set the number of msix vectors to zero
1409 * indicating that not enough can be allocated. The error
1410 * will be returned to the user indicating device open failed.
1411 * Any further attempts to force the driver to open will also
1412 * fail. The only way to recover is to unload the driver and
1413 * reload it again. If the system has recovered some MSIX
1414 * vectors then it may succeed.
1416 adapter->num_msix_vectors = 0;
1417 return err;
1420 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1422 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1424 for (i = 0; i < q_vectors; i++) {
1425 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1427 q_vector->rx.ring = NULL;
1428 q_vector->tx.ring = NULL;
1429 q_vector->rx.count = 0;
1430 q_vector->tx.count = 0;
1435 * ixgbevf_request_irq - initialize interrupts
1436 * @adapter: board private structure
1438 * Attempts to configure interrupts using the best available
1439 * capabilities of the hardware and kernel.
1441 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1443 int err = 0;
1445 err = ixgbevf_request_msix_irqs(adapter);
1447 if (err)
1448 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1450 return err;
1453 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1455 int i, q_vectors;
1457 q_vectors = adapter->num_msix_vectors;
1458 i = q_vectors - 1;
1460 free_irq(adapter->msix_entries[i].vector, adapter);
1461 i--;
1463 for (; i >= 0; i--) {
1464 /* free only the irqs that were actually requested */
1465 if (!adapter->q_vector[i]->rx.ring &&
1466 !adapter->q_vector[i]->tx.ring)
1467 continue;
1469 free_irq(adapter->msix_entries[i].vector,
1470 adapter->q_vector[i]);
1473 ixgbevf_reset_q_vectors(adapter);
1477 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1478 * @adapter: board private structure
1480 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1482 struct ixgbe_hw *hw = &adapter->hw;
1483 int i;
1485 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1486 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1487 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1489 IXGBE_WRITE_FLUSH(hw);
1491 for (i = 0; i < adapter->num_msix_vectors; i++)
1492 synchronize_irq(adapter->msix_entries[i].vector);
1496 * ixgbevf_irq_enable - Enable default interrupt generation settings
1497 * @adapter: board private structure
1499 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1501 struct ixgbe_hw *hw = &adapter->hw;
1503 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1504 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1505 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1509 * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1510 * @adapter: board private structure
1511 * @ring: structure containing ring specific data
1513 * Configure the Tx descriptor ring after a reset.
1515 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1516 struct ixgbevf_ring *ring)
1518 struct ixgbe_hw *hw = &adapter->hw;
1519 u64 tdba = ring->dma;
1520 int wait_loop = 10;
1521 u32 txdctl = IXGBE_TXDCTL_ENABLE;
1522 u8 reg_idx = ring->reg_idx;
1524 /* disable queue to avoid issues while updating state */
1525 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1526 IXGBE_WRITE_FLUSH(hw);
1528 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1529 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1530 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1531 ring->count * sizeof(union ixgbe_adv_tx_desc));
1533 /* disable head writeback */
1534 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1535 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1537 /* enable relaxed ordering */
1538 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1539 (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1540 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1542 /* reset head and tail pointers */
1543 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1544 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1545 ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1547 /* reset ntu and ntc to place SW in sync with hardwdare */
1548 ring->next_to_clean = 0;
1549 ring->next_to_use = 0;
1551 /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1552 * to or less than the number of on chip descriptors, which is
1553 * currently 40.
1555 txdctl |= (8 << 16); /* WTHRESH = 8 */
1557 /* Setting PTHRESH to 32 both improves performance */
1558 txdctl |= (1 << 8) | /* HTHRESH = 1 */
1559 32; /* PTHRESH = 32 */
1561 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1563 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1565 /* poll to verify queue is enabled */
1566 do {
1567 usleep_range(1000, 2000);
1568 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1569 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1570 if (!wait_loop)
1571 pr_err("Could not enable Tx Queue %d\n", reg_idx);
1575 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1576 * @adapter: board private structure
1578 * Configure the Tx unit of the MAC after a reset.
1580 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1582 u32 i;
1584 /* Setup the HW Tx Head and Tail descriptor pointers */
1585 for (i = 0; i < adapter->num_tx_queues; i++)
1586 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1589 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1591 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1593 struct ixgbe_hw *hw = &adapter->hw;
1594 u32 srrctl;
1596 srrctl = IXGBE_SRRCTL_DROP_EN;
1598 srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1599 srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1600 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1602 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1605 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1607 struct ixgbe_hw *hw = &adapter->hw;
1609 /* PSRTYPE must be initialized in 82599 */
1610 u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1611 IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1612 IXGBE_PSRTYPE_L2HDR;
1614 if (adapter->num_rx_queues > 1)
1615 psrtype |= 1 << 29;
1617 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1620 #define IXGBEVF_MAX_RX_DESC_POLL 10
1621 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1622 struct ixgbevf_ring *ring)
1624 struct ixgbe_hw *hw = &adapter->hw;
1625 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1626 u32 rxdctl;
1627 u8 reg_idx = ring->reg_idx;
1629 if (IXGBE_REMOVED(hw->hw_addr))
1630 return;
1631 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1632 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1634 /* write value back with RXDCTL.ENABLE bit cleared */
1635 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1637 /* the hardware may take up to 100us to really disable the Rx queue */
1638 do {
1639 udelay(10);
1640 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1641 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1643 if (!wait_loop)
1644 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1645 reg_idx);
1648 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1649 struct ixgbevf_ring *ring)
1651 struct ixgbe_hw *hw = &adapter->hw;
1652 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1653 u32 rxdctl;
1654 u8 reg_idx = ring->reg_idx;
1656 if (IXGBE_REMOVED(hw->hw_addr))
1657 return;
1658 do {
1659 usleep_range(1000, 2000);
1660 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1661 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1663 if (!wait_loop)
1664 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1665 reg_idx);
1668 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1670 struct ixgbe_hw *hw = &adapter->hw;
1671 u32 vfmrqc = 0, vfreta = 0;
1672 u32 rss_key[10];
1673 u16 rss_i = adapter->num_rx_queues;
1674 int i, j;
1676 /* Fill out hash function seeds */
1677 netdev_rss_key_fill(rss_key, sizeof(rss_key));
1678 for (i = 0; i < 10; i++)
1679 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), rss_key[i]);
1681 /* Fill out redirection table */
1682 for (i = 0, j = 0; i < 64; i++, j++) {
1683 if (j == rss_i)
1684 j = 0;
1685 vfreta = (vfreta << 8) | (j * 0x1);
1686 if ((i & 3) == 3)
1687 IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1690 /* Perform hash on these packet types */
1691 vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1692 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1693 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1694 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1696 vfmrqc |= IXGBE_VFMRQC_RSSEN;
1698 IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1701 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1702 struct ixgbevf_ring *ring)
1704 struct ixgbe_hw *hw = &adapter->hw;
1705 u64 rdba = ring->dma;
1706 u32 rxdctl;
1707 u8 reg_idx = ring->reg_idx;
1709 /* disable queue to avoid issues while updating state */
1710 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1711 ixgbevf_disable_rx_queue(adapter, ring);
1713 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1714 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1715 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1716 ring->count * sizeof(union ixgbe_adv_rx_desc));
1718 /* enable relaxed ordering */
1719 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1720 IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1722 /* reset head and tail pointers */
1723 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1724 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1725 ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1727 /* reset ntu and ntc to place SW in sync with hardwdare */
1728 ring->next_to_clean = 0;
1729 ring->next_to_use = 0;
1730 ring->next_to_alloc = 0;
1732 ixgbevf_configure_srrctl(adapter, reg_idx);
1734 /* allow any size packet since we can handle overflow */
1735 rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1737 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1738 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1740 ixgbevf_rx_desc_queue_enable(adapter, ring);
1741 ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1745 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1746 * @adapter: board private structure
1748 * Configure the Rx unit of the MAC after a reset.
1750 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1752 int i;
1753 struct ixgbe_hw *hw = &adapter->hw;
1754 struct net_device *netdev = adapter->netdev;
1756 ixgbevf_setup_psrtype(adapter);
1757 if (hw->mac.type >= ixgbe_mac_X550_vf)
1758 ixgbevf_setup_vfmrqc(adapter);
1760 /* notify the PF of our intent to use this size of frame */
1761 ixgbevf_rlpml_set_vf(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1763 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1764 * the Base and Length of the Rx Descriptor Ring
1766 for (i = 0; i < adapter->num_rx_queues; i++)
1767 ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1770 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1771 __be16 proto, u16 vid)
1773 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1774 struct ixgbe_hw *hw = &adapter->hw;
1775 int err;
1777 spin_lock_bh(&adapter->mbx_lock);
1779 /* add VID to filter table */
1780 err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1782 spin_unlock_bh(&adapter->mbx_lock);
1784 /* translate error return types so error makes sense */
1785 if (err == IXGBE_ERR_MBX)
1786 return -EIO;
1788 if (err == IXGBE_ERR_INVALID_ARGUMENT)
1789 return -EACCES;
1791 set_bit(vid, adapter->active_vlans);
1793 return err;
1796 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1797 __be16 proto, u16 vid)
1799 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1800 struct ixgbe_hw *hw = &adapter->hw;
1801 int err = -EOPNOTSUPP;
1803 spin_lock_bh(&adapter->mbx_lock);
1805 /* remove VID from filter table */
1806 err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1808 spin_unlock_bh(&adapter->mbx_lock);
1810 clear_bit(vid, adapter->active_vlans);
1812 return err;
1815 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1817 u16 vid;
1819 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1820 ixgbevf_vlan_rx_add_vid(adapter->netdev,
1821 htons(ETH_P_8021Q), vid);
1824 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1826 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1827 struct ixgbe_hw *hw = &adapter->hw;
1828 int count = 0;
1830 if ((netdev_uc_count(netdev)) > 10) {
1831 pr_err("Too many unicast filters - No Space\n");
1832 return -ENOSPC;
1835 if (!netdev_uc_empty(netdev)) {
1836 struct netdev_hw_addr *ha;
1838 netdev_for_each_uc_addr(ha, netdev) {
1839 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1840 udelay(200);
1842 } else {
1843 /* If the list is empty then send message to PF driver to
1844 * clear all MAC VLANs on this VF.
1846 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1849 return count;
1853 * ixgbevf_set_rx_mode - Multicast and unicast set
1854 * @netdev: network interface device structure
1856 * The set_rx_method entry point is called whenever the multicast address
1857 * list, unicast address list or the network interface flags are updated.
1858 * This routine is responsible for configuring the hardware for proper
1859 * multicast mode and configuring requested unicast filters.
1861 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1863 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1864 struct ixgbe_hw *hw = &adapter->hw;
1866 spin_lock_bh(&adapter->mbx_lock);
1868 /* reprogram multicast list */
1869 hw->mac.ops.update_mc_addr_list(hw, netdev);
1871 ixgbevf_write_uc_addr_list(netdev);
1873 spin_unlock_bh(&adapter->mbx_lock);
1876 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1878 int q_idx;
1879 struct ixgbevf_q_vector *q_vector;
1880 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1882 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1883 q_vector = adapter->q_vector[q_idx];
1884 #ifdef CONFIG_NET_RX_BUSY_POLL
1885 ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1886 #endif
1887 napi_enable(&q_vector->napi);
1891 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1893 int q_idx;
1894 struct ixgbevf_q_vector *q_vector;
1895 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1897 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1898 q_vector = adapter->q_vector[q_idx];
1899 napi_disable(&q_vector->napi);
1900 #ifdef CONFIG_NET_RX_BUSY_POLL
1901 while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1902 pr_info("QV %d locked\n", q_idx);
1903 usleep_range(1000, 20000);
1905 #endif /* CONFIG_NET_RX_BUSY_POLL */
1909 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1911 struct ixgbe_hw *hw = &adapter->hw;
1912 unsigned int def_q = 0;
1913 unsigned int num_tcs = 0;
1914 unsigned int num_rx_queues = adapter->num_rx_queues;
1915 unsigned int num_tx_queues = adapter->num_tx_queues;
1916 int err;
1918 spin_lock_bh(&adapter->mbx_lock);
1920 /* fetch queue configuration from the PF */
1921 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1923 spin_unlock_bh(&adapter->mbx_lock);
1925 if (err)
1926 return err;
1928 if (num_tcs > 1) {
1929 /* we need only one Tx queue */
1930 num_tx_queues = 1;
1932 /* update default Tx ring register index */
1933 adapter->tx_ring[0]->reg_idx = def_q;
1935 /* we need as many queues as traffic classes */
1936 num_rx_queues = num_tcs;
1939 /* if we have a bad config abort request queue reset */
1940 if ((adapter->num_rx_queues != num_rx_queues) ||
1941 (adapter->num_tx_queues != num_tx_queues)) {
1942 /* force mailbox timeout to prevent further messages */
1943 hw->mbx.timeout = 0;
1945 /* wait for watchdog to come around and bail us out */
1946 adapter->flags |= IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
1949 return 0;
1952 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1954 ixgbevf_configure_dcb(adapter);
1956 ixgbevf_set_rx_mode(adapter->netdev);
1958 ixgbevf_restore_vlan(adapter);
1960 ixgbevf_configure_tx(adapter);
1961 ixgbevf_configure_rx(adapter);
1964 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1966 /* Only save pre-reset stats if there are some */
1967 if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1968 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1969 adapter->stats.base_vfgprc;
1970 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1971 adapter->stats.base_vfgptc;
1972 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1973 adapter->stats.base_vfgorc;
1974 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1975 adapter->stats.base_vfgotc;
1976 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1977 adapter->stats.base_vfmprc;
1981 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1983 struct ixgbe_hw *hw = &adapter->hw;
1985 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1986 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1987 adapter->stats.last_vfgorc |=
1988 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1989 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1990 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1991 adapter->stats.last_vfgotc |=
1992 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1993 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1995 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1996 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1997 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1998 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1999 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2002 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2004 struct ixgbe_hw *hw = &adapter->hw;
2005 int api[] = { ixgbe_mbox_api_12,
2006 ixgbe_mbox_api_11,
2007 ixgbe_mbox_api_10,
2008 ixgbe_mbox_api_unknown };
2009 int err = 0, idx = 0;
2011 spin_lock_bh(&adapter->mbx_lock);
2013 while (api[idx] != ixgbe_mbox_api_unknown) {
2014 err = ixgbevf_negotiate_api_version(hw, api[idx]);
2015 if (!err)
2016 break;
2017 idx++;
2020 spin_unlock_bh(&adapter->mbx_lock);
2023 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2025 struct net_device *netdev = adapter->netdev;
2026 struct ixgbe_hw *hw = &adapter->hw;
2028 ixgbevf_configure_msix(adapter);
2030 spin_lock_bh(&adapter->mbx_lock);
2032 if (is_valid_ether_addr(hw->mac.addr))
2033 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2034 else
2035 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2037 spin_unlock_bh(&adapter->mbx_lock);
2039 smp_mb__before_atomic();
2040 clear_bit(__IXGBEVF_DOWN, &adapter->state);
2041 ixgbevf_napi_enable_all(adapter);
2043 /* clear any pending interrupts, may auto mask */
2044 IXGBE_READ_REG(hw, IXGBE_VTEICR);
2045 ixgbevf_irq_enable(adapter);
2047 /* enable transmits */
2048 netif_tx_start_all_queues(netdev);
2050 ixgbevf_save_reset_stats(adapter);
2051 ixgbevf_init_last_counter_stats(adapter);
2053 hw->mac.get_link_status = 1;
2054 mod_timer(&adapter->service_timer, jiffies);
2057 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2059 ixgbevf_configure(adapter);
2061 ixgbevf_up_complete(adapter);
2065 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2066 * @rx_ring: ring to free buffers from
2068 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2070 struct device *dev = rx_ring->dev;
2071 unsigned long size;
2072 unsigned int i;
2074 /* Free Rx ring sk_buff */
2075 if (rx_ring->skb) {
2076 dev_kfree_skb(rx_ring->skb);
2077 rx_ring->skb = NULL;
2080 /* ring already cleared, nothing to do */
2081 if (!rx_ring->rx_buffer_info)
2082 return;
2084 /* Free all the Rx ring pages */
2085 for (i = 0; i < rx_ring->count; i++) {
2086 struct ixgbevf_rx_buffer *rx_buffer;
2088 rx_buffer = &rx_ring->rx_buffer_info[i];
2089 if (rx_buffer->dma)
2090 dma_unmap_page(dev, rx_buffer->dma,
2091 PAGE_SIZE, DMA_FROM_DEVICE);
2092 rx_buffer->dma = 0;
2093 if (rx_buffer->page)
2094 __free_page(rx_buffer->page);
2095 rx_buffer->page = NULL;
2098 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2099 memset(rx_ring->rx_buffer_info, 0, size);
2101 /* Zero out the descriptor ring */
2102 memset(rx_ring->desc, 0, rx_ring->size);
2106 * ixgbevf_clean_tx_ring - Free Tx Buffers
2107 * @tx_ring: ring to be cleaned
2109 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2111 struct ixgbevf_tx_buffer *tx_buffer_info;
2112 unsigned long size;
2113 unsigned int i;
2115 if (!tx_ring->tx_buffer_info)
2116 return;
2118 /* Free all the Tx ring sk_buffs */
2119 for (i = 0; i < tx_ring->count; i++) {
2120 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2121 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2124 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2125 memset(tx_ring->tx_buffer_info, 0, size);
2127 memset(tx_ring->desc, 0, tx_ring->size);
2131 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2132 * @adapter: board private structure
2134 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2136 int i;
2138 for (i = 0; i < adapter->num_rx_queues; i++)
2139 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2143 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2144 * @adapter: board private structure
2146 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2148 int i;
2150 for (i = 0; i < adapter->num_tx_queues; i++)
2151 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2154 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2156 struct net_device *netdev = adapter->netdev;
2157 struct ixgbe_hw *hw = &adapter->hw;
2158 int i;
2160 /* signal that we are down to the interrupt handler */
2161 if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2162 return; /* do nothing if already down */
2164 /* disable all enabled Rx queues */
2165 for (i = 0; i < adapter->num_rx_queues; i++)
2166 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2168 usleep_range(10000, 20000);
2170 netif_tx_stop_all_queues(netdev);
2172 /* call carrier off first to avoid false dev_watchdog timeouts */
2173 netif_carrier_off(netdev);
2174 netif_tx_disable(netdev);
2176 ixgbevf_irq_disable(adapter);
2178 ixgbevf_napi_disable_all(adapter);
2180 del_timer_sync(&adapter->service_timer);
2182 /* disable transmits in the hardware now that interrupts are off */
2183 for (i = 0; i < adapter->num_tx_queues; i++) {
2184 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2186 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2187 IXGBE_TXDCTL_SWFLSH);
2190 if (!pci_channel_offline(adapter->pdev))
2191 ixgbevf_reset(adapter);
2193 ixgbevf_clean_all_tx_rings(adapter);
2194 ixgbevf_clean_all_rx_rings(adapter);
2197 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2199 WARN_ON(in_interrupt());
2201 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2202 msleep(1);
2204 ixgbevf_down(adapter);
2205 ixgbevf_up(adapter);
2207 clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2210 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2212 struct ixgbe_hw *hw = &adapter->hw;
2213 struct net_device *netdev = adapter->netdev;
2215 if (hw->mac.ops.reset_hw(hw)) {
2216 hw_dbg(hw, "PF still resetting\n");
2217 } else {
2218 hw->mac.ops.init_hw(hw);
2219 ixgbevf_negotiate_api(adapter);
2222 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2223 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2224 netdev->addr_len);
2225 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
2226 netdev->addr_len);
2229 adapter->last_reset = jiffies;
2232 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2233 int vectors)
2235 int vector_threshold;
2237 /* We'll want at least 2 (vector_threshold):
2238 * 1) TxQ[0] + RxQ[0] handler
2239 * 2) Other (Link Status Change, etc.)
2241 vector_threshold = MIN_MSIX_COUNT;
2243 /* The more we get, the more we will assign to Tx/Rx Cleanup
2244 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2245 * Right now, we simply care about how many we'll get; we'll
2246 * set them up later while requesting irq's.
2248 vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2249 vector_threshold, vectors);
2251 if (vectors < 0) {
2252 dev_err(&adapter->pdev->dev,
2253 "Unable to allocate MSI-X interrupts\n");
2254 kfree(adapter->msix_entries);
2255 adapter->msix_entries = NULL;
2256 return vectors;
2259 /* Adjust for only the vectors we'll use, which is minimum
2260 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2261 * vectors we were allocated.
2263 adapter->num_msix_vectors = vectors;
2265 return 0;
2269 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2270 * @adapter: board private structure to initialize
2272 * This is the top level queue allocation routine. The order here is very
2273 * important, starting with the "most" number of features turned on at once,
2274 * and ending with the smallest set of features. This way large combinations
2275 * can be allocated if they're turned on, and smaller combinations are the
2276 * fallthrough conditions.
2279 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2281 struct ixgbe_hw *hw = &adapter->hw;
2282 unsigned int def_q = 0;
2283 unsigned int num_tcs = 0;
2284 int err;
2286 /* Start with base case */
2287 adapter->num_rx_queues = 1;
2288 adapter->num_tx_queues = 1;
2290 spin_lock_bh(&adapter->mbx_lock);
2292 /* fetch queue configuration from the PF */
2293 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2295 spin_unlock_bh(&adapter->mbx_lock);
2297 if (err)
2298 return;
2300 /* we need as many queues as traffic classes */
2301 if (num_tcs > 1) {
2302 adapter->num_rx_queues = num_tcs;
2303 } else {
2304 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2306 switch (hw->api_version) {
2307 case ixgbe_mbox_api_11:
2308 case ixgbe_mbox_api_12:
2309 adapter->num_rx_queues = rss;
2310 adapter->num_tx_queues = rss;
2311 default:
2312 break;
2318 * ixgbevf_alloc_queues - Allocate memory for all rings
2319 * @adapter: board private structure to initialize
2321 * We allocate one ring per queue at run-time since we don't know the
2322 * number of queues at compile-time. The polling_netdev array is
2323 * intended for Multiqueue, but should work fine with a single queue.
2325 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2327 struct ixgbevf_ring *ring;
2328 int rx = 0, tx = 0;
2330 for (; tx < adapter->num_tx_queues; tx++) {
2331 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2332 if (!ring)
2333 goto err_allocation;
2335 ring->dev = &adapter->pdev->dev;
2336 ring->netdev = adapter->netdev;
2337 ring->count = adapter->tx_ring_count;
2338 ring->queue_index = tx;
2339 ring->reg_idx = tx;
2341 adapter->tx_ring[tx] = ring;
2344 for (; rx < adapter->num_rx_queues; rx++) {
2345 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2346 if (!ring)
2347 goto err_allocation;
2349 ring->dev = &adapter->pdev->dev;
2350 ring->netdev = adapter->netdev;
2352 ring->count = adapter->rx_ring_count;
2353 ring->queue_index = rx;
2354 ring->reg_idx = rx;
2356 adapter->rx_ring[rx] = ring;
2359 return 0;
2361 err_allocation:
2362 while (tx) {
2363 kfree(adapter->tx_ring[--tx]);
2364 adapter->tx_ring[tx] = NULL;
2367 while (rx) {
2368 kfree(adapter->rx_ring[--rx]);
2369 adapter->rx_ring[rx] = NULL;
2371 return -ENOMEM;
2375 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2376 * @adapter: board private structure to initialize
2378 * Attempt to configure the interrupts using the best available
2379 * capabilities of the hardware and the kernel.
2381 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2383 struct net_device *netdev = adapter->netdev;
2384 int err = 0;
2385 int vector, v_budget;
2387 /* It's easy to be greedy for MSI-X vectors, but it really
2388 * doesn't do us much good if we have a lot more vectors
2389 * than CPU's. So let's be conservative and only ask for
2390 * (roughly) the same number of vectors as there are CPU's.
2391 * The default is to use pairs of vectors.
2393 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2394 v_budget = min_t(int, v_budget, num_online_cpus());
2395 v_budget += NON_Q_VECTORS;
2397 /* A failure in MSI-X entry allocation isn't fatal, but it does
2398 * mean we disable MSI-X capabilities of the adapter.
2400 adapter->msix_entries = kcalloc(v_budget,
2401 sizeof(struct msix_entry), GFP_KERNEL);
2402 if (!adapter->msix_entries) {
2403 err = -ENOMEM;
2404 goto out;
2407 for (vector = 0; vector < v_budget; vector++)
2408 adapter->msix_entries[vector].entry = vector;
2410 err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2411 if (err)
2412 goto out;
2414 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2415 if (err)
2416 goto out;
2418 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2420 out:
2421 return err;
2425 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2426 * @adapter: board private structure to initialize
2428 * We allocate one q_vector per queue interrupt. If allocation fails we
2429 * return -ENOMEM.
2431 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2433 int q_idx, num_q_vectors;
2434 struct ixgbevf_q_vector *q_vector;
2436 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2438 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2439 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2440 if (!q_vector)
2441 goto err_out;
2442 q_vector->adapter = adapter;
2443 q_vector->v_idx = q_idx;
2444 netif_napi_add(adapter->netdev, &q_vector->napi,
2445 ixgbevf_poll, 64);
2446 #ifdef CONFIG_NET_RX_BUSY_POLL
2447 napi_hash_add(&q_vector->napi);
2448 #endif
2449 adapter->q_vector[q_idx] = q_vector;
2452 return 0;
2454 err_out:
2455 while (q_idx) {
2456 q_idx--;
2457 q_vector = adapter->q_vector[q_idx];
2458 #ifdef CONFIG_NET_RX_BUSY_POLL
2459 napi_hash_del(&q_vector->napi);
2460 #endif
2461 netif_napi_del(&q_vector->napi);
2462 kfree(q_vector);
2463 adapter->q_vector[q_idx] = NULL;
2465 return -ENOMEM;
2469 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2470 * @adapter: board private structure to initialize
2472 * This function frees the memory allocated to the q_vectors. In addition if
2473 * NAPI is enabled it will delete any references to the NAPI struct prior
2474 * to freeing the q_vector.
2476 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2478 int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2480 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2481 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2483 adapter->q_vector[q_idx] = NULL;
2484 #ifdef CONFIG_NET_RX_BUSY_POLL
2485 napi_hash_del(&q_vector->napi);
2486 #endif
2487 netif_napi_del(&q_vector->napi);
2488 kfree(q_vector);
2493 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2494 * @adapter: board private structure
2497 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2499 pci_disable_msix(adapter->pdev);
2500 kfree(adapter->msix_entries);
2501 adapter->msix_entries = NULL;
2505 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2506 * @adapter: board private structure to initialize
2509 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2511 int err;
2513 /* Number of supported queues */
2514 ixgbevf_set_num_queues(adapter);
2516 err = ixgbevf_set_interrupt_capability(adapter);
2517 if (err) {
2518 hw_dbg(&adapter->hw,
2519 "Unable to setup interrupt capabilities\n");
2520 goto err_set_interrupt;
2523 err = ixgbevf_alloc_q_vectors(adapter);
2524 if (err) {
2525 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2526 goto err_alloc_q_vectors;
2529 err = ixgbevf_alloc_queues(adapter);
2530 if (err) {
2531 pr_err("Unable to allocate memory for queues\n");
2532 goto err_alloc_queues;
2535 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
2536 (adapter->num_rx_queues > 1) ? "Enabled" :
2537 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2539 set_bit(__IXGBEVF_DOWN, &adapter->state);
2541 return 0;
2542 err_alloc_queues:
2543 ixgbevf_free_q_vectors(adapter);
2544 err_alloc_q_vectors:
2545 ixgbevf_reset_interrupt_capability(adapter);
2546 err_set_interrupt:
2547 return err;
2551 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2552 * @adapter: board private structure to clear interrupt scheme on
2554 * We go through and clear interrupt specific resources and reset the structure
2555 * to pre-load conditions
2557 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2559 int i;
2561 for (i = 0; i < adapter->num_tx_queues; i++) {
2562 kfree(adapter->tx_ring[i]);
2563 adapter->tx_ring[i] = NULL;
2565 for (i = 0; i < adapter->num_rx_queues; i++) {
2566 kfree(adapter->rx_ring[i]);
2567 adapter->rx_ring[i] = NULL;
2570 adapter->num_tx_queues = 0;
2571 adapter->num_rx_queues = 0;
2573 ixgbevf_free_q_vectors(adapter);
2574 ixgbevf_reset_interrupt_capability(adapter);
2578 * ixgbevf_sw_init - Initialize general software structures
2579 * @adapter: board private structure to initialize
2581 * ixgbevf_sw_init initializes the Adapter private data structure.
2582 * Fields are initialized based on PCI device information and
2583 * OS network device settings (MTU size).
2585 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2587 struct ixgbe_hw *hw = &adapter->hw;
2588 struct pci_dev *pdev = adapter->pdev;
2589 struct net_device *netdev = adapter->netdev;
2590 int err;
2592 /* PCI config space info */
2593 hw->vendor_id = pdev->vendor;
2594 hw->device_id = pdev->device;
2595 hw->revision_id = pdev->revision;
2596 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2597 hw->subsystem_device_id = pdev->subsystem_device;
2599 hw->mbx.ops.init_params(hw);
2601 /* assume legacy case in which PF would only give VF 2 queues */
2602 hw->mac.max_tx_queues = 2;
2603 hw->mac.max_rx_queues = 2;
2605 /* lock to protect mailbox accesses */
2606 spin_lock_init(&adapter->mbx_lock);
2608 err = hw->mac.ops.reset_hw(hw);
2609 if (err) {
2610 dev_info(&pdev->dev,
2611 "PF still in reset state. Is the PF interface up?\n");
2612 } else {
2613 err = hw->mac.ops.init_hw(hw);
2614 if (err) {
2615 pr_err("init_shared_code failed: %d\n", err);
2616 goto out;
2618 ixgbevf_negotiate_api(adapter);
2619 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2620 if (err)
2621 dev_info(&pdev->dev, "Error reading MAC address\n");
2622 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2623 dev_info(&pdev->dev,
2624 "MAC address not assigned by administrator.\n");
2625 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2628 if (!is_valid_ether_addr(netdev->dev_addr)) {
2629 dev_info(&pdev->dev, "Assigning random MAC address\n");
2630 eth_hw_addr_random(netdev);
2631 memcpy(hw->mac.addr, netdev->dev_addr, netdev->addr_len);
2634 /* Enable dynamic interrupt throttling rates */
2635 adapter->rx_itr_setting = 1;
2636 adapter->tx_itr_setting = 1;
2638 /* set default ring sizes */
2639 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2640 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2642 set_bit(__IXGBEVF_DOWN, &adapter->state);
2643 return 0;
2645 out:
2646 return err;
2649 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
2651 u32 current_counter = IXGBE_READ_REG(hw, reg); \
2652 if (current_counter < last_counter) \
2653 counter += 0x100000000LL; \
2654 last_counter = current_counter; \
2655 counter &= 0xFFFFFFFF00000000LL; \
2656 counter |= current_counter; \
2659 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2661 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
2662 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
2663 u64 current_counter = (current_counter_msb << 32) | \
2664 current_counter_lsb; \
2665 if (current_counter < last_counter) \
2666 counter += 0x1000000000LL; \
2667 last_counter = current_counter; \
2668 counter &= 0xFFFFFFF000000000LL; \
2669 counter |= current_counter; \
2672 * ixgbevf_update_stats - Update the board statistics counters.
2673 * @adapter: board private structure
2675 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2677 struct ixgbe_hw *hw = &adapter->hw;
2678 int i;
2680 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2681 test_bit(__IXGBEVF_RESETTING, &adapter->state))
2682 return;
2684 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2685 adapter->stats.vfgprc);
2686 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2687 adapter->stats.vfgptc);
2688 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2689 adapter->stats.last_vfgorc,
2690 adapter->stats.vfgorc);
2691 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2692 adapter->stats.last_vfgotc,
2693 adapter->stats.vfgotc);
2694 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2695 adapter->stats.vfmprc);
2697 for (i = 0; i < adapter->num_rx_queues; i++) {
2698 adapter->hw_csum_rx_error +=
2699 adapter->rx_ring[i]->hw_csum_rx_error;
2700 adapter->rx_ring[i]->hw_csum_rx_error = 0;
2705 * ixgbevf_service_timer - Timer Call-back
2706 * @data: pointer to adapter cast into an unsigned long
2708 static void ixgbevf_service_timer(unsigned long data)
2710 struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2712 /* Reset the timer */
2713 mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
2715 ixgbevf_service_event_schedule(adapter);
2718 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
2720 if (!(adapter->flags & IXGBEVF_FLAG_RESET_REQUESTED))
2721 return;
2723 adapter->flags &= ~IXGBEVF_FLAG_RESET_REQUESTED;
2725 /* If we're already down or resetting, just bail */
2726 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2727 test_bit(__IXGBEVF_RESETTING, &adapter->state))
2728 return;
2730 adapter->tx_timeout_count++;
2732 ixgbevf_reinit_locked(adapter);
2736 * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2737 * @adapter: pointer to the device adapter structure
2739 * This function serves two purposes. First it strobes the interrupt lines
2740 * in order to make certain interrupts are occurring. Secondly it sets the
2741 * bits needed to check for TX hangs. As a result we should immediately
2742 * determine if a hang has occurred.
2744 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
2746 struct ixgbe_hw *hw = &adapter->hw;
2747 u32 eics = 0;
2748 int i;
2750 /* If we're down or resetting, just bail */
2751 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2752 test_bit(__IXGBEVF_RESETTING, &adapter->state))
2753 return;
2755 /* Force detection of hung controller */
2756 if (netif_carrier_ok(adapter->netdev)) {
2757 for (i = 0; i < adapter->num_tx_queues; i++)
2758 set_check_for_tx_hang(adapter->tx_ring[i]);
2761 /* get one bit for every active Tx/Rx interrupt vector */
2762 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2763 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2765 if (qv->rx.ring || qv->tx.ring)
2766 eics |= 1 << i;
2769 /* Cause software interrupt to ensure rings are cleaned */
2770 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2774 * ixgbevf_watchdog_update_link - update the link status
2775 * @adapter: pointer to the device adapter structure
2777 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
2779 struct ixgbe_hw *hw = &adapter->hw;
2780 u32 link_speed = adapter->link_speed;
2781 bool link_up = adapter->link_up;
2782 s32 err;
2784 spin_lock_bh(&adapter->mbx_lock);
2786 err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2788 spin_unlock_bh(&adapter->mbx_lock);
2790 /* if check for link returns error we will need to reset */
2791 if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
2792 adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
2793 link_up = false;
2796 adapter->link_up = link_up;
2797 adapter->link_speed = link_speed;
2801 * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2802 * print link up message
2803 * @adapter: pointer to the device adapter structure
2805 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
2807 struct net_device *netdev = adapter->netdev;
2809 /* only continue if link was previously down */
2810 if (netif_carrier_ok(netdev))
2811 return;
2813 dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
2814 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2815 "10 Gbps" :
2816 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
2817 "1 Gbps" :
2818 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
2819 "100 Mbps" :
2820 "unknown speed");
2822 netif_carrier_on(netdev);
2826 * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2827 * print link down message
2828 * @adapter: pointer to the adapter structure
2830 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
2832 struct net_device *netdev = adapter->netdev;
2834 adapter->link_speed = 0;
2836 /* only continue if link was up previously */
2837 if (!netif_carrier_ok(netdev))
2838 return;
2840 dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2842 netif_carrier_off(netdev);
2846 * ixgbevf_watchdog_subtask - worker thread to bring link up
2847 * @work: pointer to work_struct containing our data
2849 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
2851 /* if interface is down do nothing */
2852 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2853 test_bit(__IXGBEVF_RESETTING, &adapter->state))
2854 return;
2856 ixgbevf_watchdog_update_link(adapter);
2858 if (adapter->link_up)
2859 ixgbevf_watchdog_link_is_up(adapter);
2860 else
2861 ixgbevf_watchdog_link_is_down(adapter);
2863 ixgbevf_update_stats(adapter);
2867 * ixgbevf_service_task - manages and runs subtasks
2868 * @work: pointer to work_struct containing our data
2870 static void ixgbevf_service_task(struct work_struct *work)
2872 struct ixgbevf_adapter *adapter = container_of(work,
2873 struct ixgbevf_adapter,
2874 service_task);
2875 struct ixgbe_hw *hw = &adapter->hw;
2877 if (IXGBE_REMOVED(hw->hw_addr)) {
2878 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2879 rtnl_lock();
2880 ixgbevf_down(adapter);
2881 rtnl_unlock();
2883 return;
2886 ixgbevf_queue_reset_subtask(adapter);
2887 ixgbevf_reset_subtask(adapter);
2888 ixgbevf_watchdog_subtask(adapter);
2889 ixgbevf_check_hang_subtask(adapter);
2891 ixgbevf_service_event_complete(adapter);
2895 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2896 * @tx_ring: Tx descriptor ring for a specific queue
2898 * Free all transmit software resources
2900 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2902 ixgbevf_clean_tx_ring(tx_ring);
2904 vfree(tx_ring->tx_buffer_info);
2905 tx_ring->tx_buffer_info = NULL;
2907 /* if not set, then don't free */
2908 if (!tx_ring->desc)
2909 return;
2911 dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2912 tx_ring->dma);
2914 tx_ring->desc = NULL;
2918 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2919 * @adapter: board private structure
2921 * Free all transmit software resources
2923 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2925 int i;
2927 for (i = 0; i < adapter->num_tx_queues; i++)
2928 if (adapter->tx_ring[i]->desc)
2929 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2933 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2934 * @tx_ring: Tx descriptor ring (for a specific queue) to setup
2936 * Return 0 on success, negative on failure
2938 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2940 int size;
2942 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2943 tx_ring->tx_buffer_info = vzalloc(size);
2944 if (!tx_ring->tx_buffer_info)
2945 goto err;
2947 /* round up to nearest 4K */
2948 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2949 tx_ring->size = ALIGN(tx_ring->size, 4096);
2951 tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
2952 &tx_ring->dma, GFP_KERNEL);
2953 if (!tx_ring->desc)
2954 goto err;
2956 return 0;
2958 err:
2959 vfree(tx_ring->tx_buffer_info);
2960 tx_ring->tx_buffer_info = NULL;
2961 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
2962 return -ENOMEM;
2966 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2967 * @adapter: board private structure
2969 * If this function returns with an error, then it's possible one or
2970 * more of the rings is populated (while the rest are not). It is the
2971 * callers duty to clean those orphaned rings.
2973 * Return 0 on success, negative on failure
2975 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2977 int i, err = 0;
2979 for (i = 0; i < adapter->num_tx_queues; i++) {
2980 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
2981 if (!err)
2982 continue;
2983 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
2984 break;
2987 return err;
2991 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2992 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
2994 * Returns 0 on success, negative on failure
2996 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
2998 int size;
3000 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3001 rx_ring->rx_buffer_info = vzalloc(size);
3002 if (!rx_ring->rx_buffer_info)
3003 goto err;
3005 /* Round up to nearest 4K */
3006 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3007 rx_ring->size = ALIGN(rx_ring->size, 4096);
3009 rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3010 &rx_ring->dma, GFP_KERNEL);
3012 if (!rx_ring->desc)
3013 goto err;
3015 return 0;
3016 err:
3017 vfree(rx_ring->rx_buffer_info);
3018 rx_ring->rx_buffer_info = NULL;
3019 dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3020 return -ENOMEM;
3024 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3025 * @adapter: board private structure
3027 * If this function returns with an error, then it's possible one or
3028 * more of the rings is populated (while the rest are not). It is the
3029 * callers duty to clean those orphaned rings.
3031 * Return 0 on success, negative on failure
3033 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3035 int i, err = 0;
3037 for (i = 0; i < adapter->num_rx_queues; i++) {
3038 err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
3039 if (!err)
3040 continue;
3041 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3042 break;
3044 return err;
3048 * ixgbevf_free_rx_resources - Free Rx Resources
3049 * @rx_ring: ring to clean the resources from
3051 * Free all receive software resources
3053 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3055 ixgbevf_clean_rx_ring(rx_ring);
3057 vfree(rx_ring->rx_buffer_info);
3058 rx_ring->rx_buffer_info = NULL;
3060 dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3061 rx_ring->dma);
3063 rx_ring->desc = NULL;
3067 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3068 * @adapter: board private structure
3070 * Free all receive software resources
3072 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3074 int i;
3076 for (i = 0; i < adapter->num_rx_queues; i++)
3077 if (adapter->rx_ring[i]->desc)
3078 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3082 * ixgbevf_open - Called when a network interface is made active
3083 * @netdev: network interface device structure
3085 * Returns 0 on success, negative value on failure
3087 * The open entry point is called when a network interface is made
3088 * active by the system (IFF_UP). At this point all resources needed
3089 * for transmit and receive operations are allocated, the interrupt
3090 * handler is registered with the OS, the watchdog timer is started,
3091 * and the stack is notified that the interface is ready.
3093 static int ixgbevf_open(struct net_device *netdev)
3095 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3096 struct ixgbe_hw *hw = &adapter->hw;
3097 int err;
3099 /* A previous failure to open the device because of a lack of
3100 * available MSIX vector resources may have reset the number
3101 * of msix vectors variable to zero. The only way to recover
3102 * is to unload/reload the driver and hope that the system has
3103 * been able to recover some MSIX vector resources.
3105 if (!adapter->num_msix_vectors)
3106 return -ENOMEM;
3108 if (hw->adapter_stopped) {
3109 ixgbevf_reset(adapter);
3110 /* if adapter is still stopped then PF isn't up and
3111 * the VF can't start.
3113 if (hw->adapter_stopped) {
3114 err = IXGBE_ERR_MBX;
3115 pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3116 goto err_setup_reset;
3120 /* disallow open during test */
3121 if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3122 return -EBUSY;
3124 netif_carrier_off(netdev);
3126 /* allocate transmit descriptors */
3127 err = ixgbevf_setup_all_tx_resources(adapter);
3128 if (err)
3129 goto err_setup_tx;
3131 /* allocate receive descriptors */
3132 err = ixgbevf_setup_all_rx_resources(adapter);
3133 if (err)
3134 goto err_setup_rx;
3136 ixgbevf_configure(adapter);
3138 /* Map the Tx/Rx rings to the vectors we were allotted.
3139 * if request_irq will be called in this function map_rings
3140 * must be called *before* up_complete
3142 ixgbevf_map_rings_to_vectors(adapter);
3144 err = ixgbevf_request_irq(adapter);
3145 if (err)
3146 goto err_req_irq;
3148 ixgbevf_up_complete(adapter);
3150 return 0;
3152 err_req_irq:
3153 ixgbevf_down(adapter);
3154 err_setup_rx:
3155 ixgbevf_free_all_rx_resources(adapter);
3156 err_setup_tx:
3157 ixgbevf_free_all_tx_resources(adapter);
3158 ixgbevf_reset(adapter);
3160 err_setup_reset:
3162 return err;
3166 * ixgbevf_close - Disables a network interface
3167 * @netdev: network interface device structure
3169 * Returns 0, this is not allowed to fail
3171 * The close entry point is called when an interface is de-activated
3172 * by the OS. The hardware is still under the drivers control, but
3173 * needs to be disabled. A global MAC reset is issued to stop the
3174 * hardware, and all transmit and receive resources are freed.
3176 static int ixgbevf_close(struct net_device *netdev)
3178 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3180 ixgbevf_down(adapter);
3181 ixgbevf_free_irq(adapter);
3183 ixgbevf_free_all_tx_resources(adapter);
3184 ixgbevf_free_all_rx_resources(adapter);
3186 return 0;
3189 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3191 struct net_device *dev = adapter->netdev;
3193 if (!(adapter->flags & IXGBEVF_FLAG_QUEUE_RESET_REQUESTED))
3194 return;
3196 adapter->flags &= ~IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
3198 /* if interface is down do nothing */
3199 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3200 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3201 return;
3203 /* Hardware has to reinitialize queues and interrupts to
3204 * match packet buffer alignment. Unfortunately, the
3205 * hardware is not flexible enough to do this dynamically.
3207 if (netif_running(dev))
3208 ixgbevf_close(dev);
3210 ixgbevf_clear_interrupt_scheme(adapter);
3211 ixgbevf_init_interrupt_scheme(adapter);
3213 if (netif_running(dev))
3214 ixgbevf_open(dev);
3217 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3218 u32 vlan_macip_lens, u32 type_tucmd,
3219 u32 mss_l4len_idx)
3221 struct ixgbe_adv_tx_context_desc *context_desc;
3222 u16 i = tx_ring->next_to_use;
3224 context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3226 i++;
3227 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3229 /* set bits to identify this as an advanced context descriptor */
3230 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3232 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3233 context_desc->seqnum_seed = 0;
3234 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
3235 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3238 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3239 struct ixgbevf_tx_buffer *first,
3240 u8 *hdr_len)
3242 struct sk_buff *skb = first->skb;
3243 u32 vlan_macip_lens, type_tucmd;
3244 u32 mss_l4len_idx, l4len;
3245 int err;
3247 if (skb->ip_summed != CHECKSUM_PARTIAL)
3248 return 0;
3250 if (!skb_is_gso(skb))
3251 return 0;
3253 err = skb_cow_head(skb, 0);
3254 if (err < 0)
3255 return err;
3257 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3258 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3260 if (first->protocol == htons(ETH_P_IP)) {
3261 struct iphdr *iph = ip_hdr(skb);
3263 iph->tot_len = 0;
3264 iph->check = 0;
3265 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3266 iph->daddr, 0,
3267 IPPROTO_TCP,
3269 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3270 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3271 IXGBE_TX_FLAGS_CSUM |
3272 IXGBE_TX_FLAGS_IPV4;
3273 } else if (skb_is_gso_v6(skb)) {
3274 ipv6_hdr(skb)->payload_len = 0;
3275 tcp_hdr(skb)->check =
3276 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3277 &ipv6_hdr(skb)->daddr,
3278 0, IPPROTO_TCP, 0);
3279 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3280 IXGBE_TX_FLAGS_CSUM;
3283 /* compute header lengths */
3284 l4len = tcp_hdrlen(skb);
3285 *hdr_len += l4len;
3286 *hdr_len = skb_transport_offset(skb) + l4len;
3288 /* update GSO size and bytecount with header size */
3289 first->gso_segs = skb_shinfo(skb)->gso_segs;
3290 first->bytecount += (first->gso_segs - 1) * *hdr_len;
3292 /* mss_l4len_id: use 1 as index for TSO */
3293 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
3294 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3295 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
3297 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3298 vlan_macip_lens = skb_network_header_len(skb);
3299 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3300 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3302 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3303 type_tucmd, mss_l4len_idx);
3305 return 1;
3308 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3309 struct ixgbevf_tx_buffer *first)
3311 struct sk_buff *skb = first->skb;
3312 u32 vlan_macip_lens = 0;
3313 u32 mss_l4len_idx = 0;
3314 u32 type_tucmd = 0;
3316 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3317 u8 l4_hdr = 0;
3319 switch (first->protocol) {
3320 case htons(ETH_P_IP):
3321 vlan_macip_lens |= skb_network_header_len(skb);
3322 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3323 l4_hdr = ip_hdr(skb)->protocol;
3324 break;
3325 case htons(ETH_P_IPV6):
3326 vlan_macip_lens |= skb_network_header_len(skb);
3327 l4_hdr = ipv6_hdr(skb)->nexthdr;
3328 break;
3329 default:
3330 if (unlikely(net_ratelimit())) {
3331 dev_warn(tx_ring->dev,
3332 "partial checksum but proto=%x!\n",
3333 first->protocol);
3335 break;
3338 switch (l4_hdr) {
3339 case IPPROTO_TCP:
3340 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3341 mss_l4len_idx = tcp_hdrlen(skb) <<
3342 IXGBE_ADVTXD_L4LEN_SHIFT;
3343 break;
3344 case IPPROTO_SCTP:
3345 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3346 mss_l4len_idx = sizeof(struct sctphdr) <<
3347 IXGBE_ADVTXD_L4LEN_SHIFT;
3348 break;
3349 case IPPROTO_UDP:
3350 mss_l4len_idx = sizeof(struct udphdr) <<
3351 IXGBE_ADVTXD_L4LEN_SHIFT;
3352 break;
3353 default:
3354 if (unlikely(net_ratelimit())) {
3355 dev_warn(tx_ring->dev,
3356 "partial checksum but l4 proto=%x!\n",
3357 l4_hdr);
3359 break;
3362 /* update TX checksum flag */
3363 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3366 /* vlan_macip_lens: MACLEN, VLAN tag */
3367 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3368 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3370 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3371 type_tucmd, mss_l4len_idx);
3374 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3376 /* set type for advanced descriptor with frame checksum insertion */
3377 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3378 IXGBE_ADVTXD_DCMD_IFCS |
3379 IXGBE_ADVTXD_DCMD_DEXT);
3381 /* set HW VLAN bit if VLAN is present */
3382 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3383 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3385 /* set segmentation enable bits for TSO/FSO */
3386 if (tx_flags & IXGBE_TX_FLAGS_TSO)
3387 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3389 return cmd_type;
3392 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3393 u32 tx_flags, unsigned int paylen)
3395 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3397 /* enable L4 checksum for TSO and TX checksum offload */
3398 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3399 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3401 /* enble IPv4 checksum for TSO */
3402 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3403 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3405 /* use index 1 context for TSO/FSO/FCOE */
3406 if (tx_flags & IXGBE_TX_FLAGS_TSO)
3407 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
3409 /* Check Context must be set if Tx switch is enabled, which it
3410 * always is for case where virtual functions are running
3412 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3414 tx_desc->read.olinfo_status = olinfo_status;
3417 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3418 struct ixgbevf_tx_buffer *first,
3419 const u8 hdr_len)
3421 dma_addr_t dma;
3422 struct sk_buff *skb = first->skb;
3423 struct ixgbevf_tx_buffer *tx_buffer;
3424 union ixgbe_adv_tx_desc *tx_desc;
3425 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3426 unsigned int data_len = skb->data_len;
3427 unsigned int size = skb_headlen(skb);
3428 unsigned int paylen = skb->len - hdr_len;
3429 u32 tx_flags = first->tx_flags;
3430 __le32 cmd_type;
3431 u16 i = tx_ring->next_to_use;
3433 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3435 ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3436 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3438 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3439 if (dma_mapping_error(tx_ring->dev, dma))
3440 goto dma_error;
3442 /* record length, and DMA address */
3443 dma_unmap_len_set(first, len, size);
3444 dma_unmap_addr_set(first, dma, dma);
3446 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3448 for (;;) {
3449 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3450 tx_desc->read.cmd_type_len =
3451 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3453 i++;
3454 tx_desc++;
3455 if (i == tx_ring->count) {
3456 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3457 i = 0;
3460 dma += IXGBE_MAX_DATA_PER_TXD;
3461 size -= IXGBE_MAX_DATA_PER_TXD;
3463 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3464 tx_desc->read.olinfo_status = 0;
3467 if (likely(!data_len))
3468 break;
3470 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3472 i++;
3473 tx_desc++;
3474 if (i == tx_ring->count) {
3475 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3476 i = 0;
3479 size = skb_frag_size(frag);
3480 data_len -= size;
3482 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3483 DMA_TO_DEVICE);
3484 if (dma_mapping_error(tx_ring->dev, dma))
3485 goto dma_error;
3487 tx_buffer = &tx_ring->tx_buffer_info[i];
3488 dma_unmap_len_set(tx_buffer, len, size);
3489 dma_unmap_addr_set(tx_buffer, dma, dma);
3491 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3492 tx_desc->read.olinfo_status = 0;
3494 frag++;
3497 /* write last descriptor with RS and EOP bits */
3498 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3499 tx_desc->read.cmd_type_len = cmd_type;
3501 /* set the timestamp */
3502 first->time_stamp = jiffies;
3504 /* Force memory writes to complete before letting h/w know there
3505 * are new descriptors to fetch. (Only applicable for weak-ordered
3506 * memory model archs, such as IA-64).
3508 * We also need this memory barrier (wmb) to make certain all of the
3509 * status bits have been updated before next_to_watch is written.
3511 wmb();
3513 /* set next_to_watch value indicating a packet is present */
3514 first->next_to_watch = tx_desc;
3516 i++;
3517 if (i == tx_ring->count)
3518 i = 0;
3520 tx_ring->next_to_use = i;
3522 /* notify HW of packet */
3523 ixgbevf_write_tail(tx_ring, i);
3525 return;
3526 dma_error:
3527 dev_err(tx_ring->dev, "TX DMA map failed\n");
3529 /* clear dma mappings for failed tx_buffer_info map */
3530 for (;;) {
3531 tx_buffer = &tx_ring->tx_buffer_info[i];
3532 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3533 if (tx_buffer == first)
3534 break;
3535 if (i == 0)
3536 i = tx_ring->count;
3537 i--;
3540 tx_ring->next_to_use = i;
3543 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3545 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3546 /* Herbert's original patch had:
3547 * smp_mb__after_netif_stop_queue();
3548 * but since that doesn't exist yet, just open code it.
3550 smp_mb();
3552 /* We need to check again in a case another CPU has just
3553 * made room available.
3555 if (likely(ixgbevf_desc_unused(tx_ring) < size))
3556 return -EBUSY;
3558 /* A reprieve! - use start_queue because it doesn't call schedule */
3559 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3560 ++tx_ring->tx_stats.restart_queue;
3562 return 0;
3565 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3567 if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3568 return 0;
3569 return __ixgbevf_maybe_stop_tx(tx_ring, size);
3572 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3574 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3575 struct ixgbevf_tx_buffer *first;
3576 struct ixgbevf_ring *tx_ring;
3577 int tso;
3578 u32 tx_flags = 0;
3579 u16 count = TXD_USE_COUNT(skb_headlen(skb));
3580 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3581 unsigned short f;
3582 #endif
3583 u8 hdr_len = 0;
3584 u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3586 if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3587 dev_kfree_skb_any(skb);
3588 return NETDEV_TX_OK;
3591 tx_ring = adapter->tx_ring[skb->queue_mapping];
3593 /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3594 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3595 * + 2 desc gap to keep tail from touching head,
3596 * + 1 desc for context descriptor,
3597 * otherwise try next time
3599 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3600 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3601 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3602 #else
3603 count += skb_shinfo(skb)->nr_frags;
3604 #endif
3605 if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3606 tx_ring->tx_stats.tx_busy++;
3607 return NETDEV_TX_BUSY;
3610 /* record the location of the first descriptor for this packet */
3611 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3612 first->skb = skb;
3613 first->bytecount = skb->len;
3614 first->gso_segs = 1;
3616 if (skb_vlan_tag_present(skb)) {
3617 tx_flags |= skb_vlan_tag_get(skb);
3618 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3619 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3622 /* record initial flags and protocol */
3623 first->tx_flags = tx_flags;
3624 first->protocol = vlan_get_protocol(skb);
3626 tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3627 if (tso < 0)
3628 goto out_drop;
3629 else if (!tso)
3630 ixgbevf_tx_csum(tx_ring, first);
3632 ixgbevf_tx_map(tx_ring, first, hdr_len);
3634 ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3636 return NETDEV_TX_OK;
3638 out_drop:
3639 dev_kfree_skb_any(first->skb);
3640 first->skb = NULL;
3642 return NETDEV_TX_OK;
3646 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3647 * @netdev: network interface device structure
3648 * @p: pointer to an address structure
3650 * Returns 0 on success, negative on failure
3652 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3654 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3655 struct ixgbe_hw *hw = &adapter->hw;
3656 struct sockaddr *addr = p;
3658 if (!is_valid_ether_addr(addr->sa_data))
3659 return -EADDRNOTAVAIL;
3661 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3662 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3664 spin_lock_bh(&adapter->mbx_lock);
3666 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3668 spin_unlock_bh(&adapter->mbx_lock);
3670 return 0;
3674 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3675 * @netdev: network interface device structure
3676 * @new_mtu: new value for maximum frame size
3678 * Returns 0 on success, negative on failure
3680 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3682 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3683 struct ixgbe_hw *hw = &adapter->hw;
3684 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3685 int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3687 switch (adapter->hw.api_version) {
3688 case ixgbe_mbox_api_11:
3689 case ixgbe_mbox_api_12:
3690 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3691 break;
3692 default:
3693 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
3694 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3695 break;
3698 /* MTU < 68 is an error and causes problems on some kernels */
3699 if ((new_mtu < 68) || (max_frame > max_possible_frame))
3700 return -EINVAL;
3702 hw_dbg(hw, "changing MTU from %d to %d\n",
3703 netdev->mtu, new_mtu);
3704 /* must set new MTU before calling down or up */
3705 netdev->mtu = new_mtu;
3707 /* notify the PF of our intent to use this size of frame */
3708 ixgbevf_rlpml_set_vf(hw, max_frame);
3710 return 0;
3713 #ifdef CONFIG_NET_POLL_CONTROLLER
3714 /* Polling 'interrupt' - used by things like netconsole to send skbs
3715 * without having to re-enable interrupts. It's not called while
3716 * the interrupt routine is executing.
3718 static void ixgbevf_netpoll(struct net_device *netdev)
3720 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3721 int i;
3723 /* if interface is down do nothing */
3724 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
3725 return;
3726 for (i = 0; i < adapter->num_rx_queues; i++)
3727 ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
3729 #endif /* CONFIG_NET_POLL_CONTROLLER */
3731 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3733 struct net_device *netdev = pci_get_drvdata(pdev);
3734 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3735 #ifdef CONFIG_PM
3736 int retval = 0;
3737 #endif
3739 netif_device_detach(netdev);
3741 if (netif_running(netdev)) {
3742 rtnl_lock();
3743 ixgbevf_down(adapter);
3744 ixgbevf_free_irq(adapter);
3745 ixgbevf_free_all_tx_resources(adapter);
3746 ixgbevf_free_all_rx_resources(adapter);
3747 rtnl_unlock();
3750 ixgbevf_clear_interrupt_scheme(adapter);
3752 #ifdef CONFIG_PM
3753 retval = pci_save_state(pdev);
3754 if (retval)
3755 return retval;
3757 #endif
3758 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3759 pci_disable_device(pdev);
3761 return 0;
3764 #ifdef CONFIG_PM
3765 static int ixgbevf_resume(struct pci_dev *pdev)
3767 struct net_device *netdev = pci_get_drvdata(pdev);
3768 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3769 u32 err;
3771 pci_restore_state(pdev);
3772 /* pci_restore_state clears dev->state_saved so call
3773 * pci_save_state to restore it.
3775 pci_save_state(pdev);
3777 err = pci_enable_device_mem(pdev);
3778 if (err) {
3779 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3780 return err;
3782 smp_mb__before_atomic();
3783 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3784 pci_set_master(pdev);
3786 ixgbevf_reset(adapter);
3788 rtnl_lock();
3789 err = ixgbevf_init_interrupt_scheme(adapter);
3790 rtnl_unlock();
3791 if (err) {
3792 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3793 return err;
3796 if (netif_running(netdev)) {
3797 err = ixgbevf_open(netdev);
3798 if (err)
3799 return err;
3802 netif_device_attach(netdev);
3804 return err;
3807 #endif /* CONFIG_PM */
3808 static void ixgbevf_shutdown(struct pci_dev *pdev)
3810 ixgbevf_suspend(pdev, PMSG_SUSPEND);
3813 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3814 struct rtnl_link_stats64 *stats)
3816 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3817 unsigned int start;
3818 u64 bytes, packets;
3819 const struct ixgbevf_ring *ring;
3820 int i;
3822 ixgbevf_update_stats(adapter);
3824 stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3826 for (i = 0; i < adapter->num_rx_queues; i++) {
3827 ring = adapter->rx_ring[i];
3828 do {
3829 start = u64_stats_fetch_begin_irq(&ring->syncp);
3830 bytes = ring->stats.bytes;
3831 packets = ring->stats.packets;
3832 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3833 stats->rx_bytes += bytes;
3834 stats->rx_packets += packets;
3837 for (i = 0; i < adapter->num_tx_queues; i++) {
3838 ring = adapter->tx_ring[i];
3839 do {
3840 start = u64_stats_fetch_begin_irq(&ring->syncp);
3841 bytes = ring->stats.bytes;
3842 packets = ring->stats.packets;
3843 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3844 stats->tx_bytes += bytes;
3845 stats->tx_packets += packets;
3848 return stats;
3851 static const struct net_device_ops ixgbevf_netdev_ops = {
3852 .ndo_open = ixgbevf_open,
3853 .ndo_stop = ixgbevf_close,
3854 .ndo_start_xmit = ixgbevf_xmit_frame,
3855 .ndo_set_rx_mode = ixgbevf_set_rx_mode,
3856 .ndo_get_stats64 = ixgbevf_get_stats,
3857 .ndo_validate_addr = eth_validate_addr,
3858 .ndo_set_mac_address = ixgbevf_set_mac,
3859 .ndo_change_mtu = ixgbevf_change_mtu,
3860 .ndo_tx_timeout = ixgbevf_tx_timeout,
3861 .ndo_vlan_rx_add_vid = ixgbevf_vlan_rx_add_vid,
3862 .ndo_vlan_rx_kill_vid = ixgbevf_vlan_rx_kill_vid,
3863 #ifdef CONFIG_NET_RX_BUSY_POLL
3864 .ndo_busy_poll = ixgbevf_busy_poll_recv,
3865 #endif
3866 #ifdef CONFIG_NET_POLL_CONTROLLER
3867 .ndo_poll_controller = ixgbevf_netpoll,
3868 #endif
3871 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3873 dev->netdev_ops = &ixgbevf_netdev_ops;
3874 ixgbevf_set_ethtool_ops(dev);
3875 dev->watchdog_timeo = 5 * HZ;
3879 * ixgbevf_probe - Device Initialization Routine
3880 * @pdev: PCI device information struct
3881 * @ent: entry in ixgbevf_pci_tbl
3883 * Returns 0 on success, negative on failure
3885 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3886 * The OS initialization, configuring of the adapter private structure,
3887 * and a hardware reset occur.
3889 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3891 struct net_device *netdev;
3892 struct ixgbevf_adapter *adapter = NULL;
3893 struct ixgbe_hw *hw = NULL;
3894 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3895 int err, pci_using_dac;
3896 bool disable_dev = false;
3898 err = pci_enable_device(pdev);
3899 if (err)
3900 return err;
3902 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3903 pci_using_dac = 1;
3904 } else {
3905 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3906 if (err) {
3907 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
3908 goto err_dma;
3910 pci_using_dac = 0;
3913 err = pci_request_regions(pdev, ixgbevf_driver_name);
3914 if (err) {
3915 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3916 goto err_pci_reg;
3919 pci_set_master(pdev);
3921 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3922 MAX_TX_QUEUES);
3923 if (!netdev) {
3924 err = -ENOMEM;
3925 goto err_alloc_etherdev;
3928 SET_NETDEV_DEV(netdev, &pdev->dev);
3930 adapter = netdev_priv(netdev);
3932 adapter->netdev = netdev;
3933 adapter->pdev = pdev;
3934 hw = &adapter->hw;
3935 hw->back = adapter;
3936 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3938 /* call save state here in standalone driver because it relies on
3939 * adapter struct to exist, and needs to call netdev_priv
3941 pci_save_state(pdev);
3943 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3944 pci_resource_len(pdev, 0));
3945 adapter->io_addr = hw->hw_addr;
3946 if (!hw->hw_addr) {
3947 err = -EIO;
3948 goto err_ioremap;
3951 ixgbevf_assign_netdev_ops(netdev);
3953 /* Setup HW API */
3954 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3955 hw->mac.type = ii->mac;
3957 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3958 sizeof(struct ixgbe_mbx_operations));
3960 /* setup the private structure */
3961 err = ixgbevf_sw_init(adapter);
3962 if (err)
3963 goto err_sw_init;
3965 /* The HW MAC address was set and/or determined in sw_init */
3966 if (!is_valid_ether_addr(netdev->dev_addr)) {
3967 pr_err("invalid MAC address\n");
3968 err = -EIO;
3969 goto err_sw_init;
3972 netdev->hw_features = NETIF_F_SG |
3973 NETIF_F_IP_CSUM |
3974 NETIF_F_IPV6_CSUM |
3975 NETIF_F_TSO |
3976 NETIF_F_TSO6 |
3977 NETIF_F_RXCSUM;
3979 netdev->features = netdev->hw_features |
3980 NETIF_F_HW_VLAN_CTAG_TX |
3981 NETIF_F_HW_VLAN_CTAG_RX |
3982 NETIF_F_HW_VLAN_CTAG_FILTER;
3984 netdev->vlan_features |= NETIF_F_TSO |
3985 NETIF_F_TSO6 |
3986 NETIF_F_IP_CSUM |
3987 NETIF_F_IPV6_CSUM |
3988 NETIF_F_SG;
3990 if (pci_using_dac)
3991 netdev->features |= NETIF_F_HIGHDMA;
3993 netdev->priv_flags |= IFF_UNICAST_FLT;
3995 if (IXGBE_REMOVED(hw->hw_addr)) {
3996 err = -EIO;
3997 goto err_sw_init;
4000 setup_timer(&adapter->service_timer, &ixgbevf_service_timer,
4001 (unsigned long)adapter);
4003 INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4004 set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4005 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4007 err = ixgbevf_init_interrupt_scheme(adapter);
4008 if (err)
4009 goto err_sw_init;
4011 strcpy(netdev->name, "eth%d");
4013 err = register_netdev(netdev);
4014 if (err)
4015 goto err_register;
4017 pci_set_drvdata(pdev, netdev);
4018 netif_carrier_off(netdev);
4020 ixgbevf_init_last_counter_stats(adapter);
4022 /* print the VF info */
4023 dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4024 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4026 switch (hw->mac.type) {
4027 case ixgbe_mac_X550_vf:
4028 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4029 break;
4030 case ixgbe_mac_X540_vf:
4031 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4032 break;
4033 case ixgbe_mac_82599_vf:
4034 default:
4035 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4036 break;
4039 return 0;
4041 err_register:
4042 ixgbevf_clear_interrupt_scheme(adapter);
4043 err_sw_init:
4044 ixgbevf_reset_interrupt_capability(adapter);
4045 iounmap(adapter->io_addr);
4046 err_ioremap:
4047 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4048 free_netdev(netdev);
4049 err_alloc_etherdev:
4050 pci_release_regions(pdev);
4051 err_pci_reg:
4052 err_dma:
4053 if (!adapter || disable_dev)
4054 pci_disable_device(pdev);
4055 return err;
4059 * ixgbevf_remove - Device Removal Routine
4060 * @pdev: PCI device information struct
4062 * ixgbevf_remove is called by the PCI subsystem to alert the driver
4063 * that it should release a PCI device. The could be caused by a
4064 * Hot-Plug event, or because the driver is going to be removed from
4065 * memory.
4067 static void ixgbevf_remove(struct pci_dev *pdev)
4069 struct net_device *netdev = pci_get_drvdata(pdev);
4070 struct ixgbevf_adapter *adapter;
4071 bool disable_dev;
4073 if (!netdev)
4074 return;
4076 adapter = netdev_priv(netdev);
4078 set_bit(__IXGBEVF_REMOVING, &adapter->state);
4079 cancel_work_sync(&adapter->service_task);
4081 if (netdev->reg_state == NETREG_REGISTERED)
4082 unregister_netdev(netdev);
4084 ixgbevf_clear_interrupt_scheme(adapter);
4085 ixgbevf_reset_interrupt_capability(adapter);
4087 iounmap(adapter->io_addr);
4088 pci_release_regions(pdev);
4090 hw_dbg(&adapter->hw, "Remove complete\n");
4092 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4093 free_netdev(netdev);
4095 if (disable_dev)
4096 pci_disable_device(pdev);
4100 * ixgbevf_io_error_detected - called when PCI error is detected
4101 * @pdev: Pointer to PCI device
4102 * @state: The current pci connection state
4104 * This function is called after a PCI bus error affecting
4105 * this device has been detected.
4107 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4108 pci_channel_state_t state)
4110 struct net_device *netdev = pci_get_drvdata(pdev);
4111 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4113 if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4114 return PCI_ERS_RESULT_DISCONNECT;
4116 rtnl_lock();
4117 netif_device_detach(netdev);
4119 if (state == pci_channel_io_perm_failure) {
4120 rtnl_unlock();
4121 return PCI_ERS_RESULT_DISCONNECT;
4124 if (netif_running(netdev))
4125 ixgbevf_down(adapter);
4127 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4128 pci_disable_device(pdev);
4129 rtnl_unlock();
4131 /* Request a slot slot reset. */
4132 return PCI_ERS_RESULT_NEED_RESET;
4136 * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4137 * @pdev: Pointer to PCI device
4139 * Restart the card from scratch, as if from a cold-boot. Implementation
4140 * resembles the first-half of the ixgbevf_resume routine.
4142 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4144 struct net_device *netdev = pci_get_drvdata(pdev);
4145 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4147 if (pci_enable_device_mem(pdev)) {
4148 dev_err(&pdev->dev,
4149 "Cannot re-enable PCI device after reset.\n");
4150 return PCI_ERS_RESULT_DISCONNECT;
4153 smp_mb__before_atomic();
4154 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4155 pci_set_master(pdev);
4157 ixgbevf_reset(adapter);
4159 return PCI_ERS_RESULT_RECOVERED;
4163 * ixgbevf_io_resume - called when traffic can start flowing again.
4164 * @pdev: Pointer to PCI device
4166 * This callback is called when the error recovery driver tells us that
4167 * its OK to resume normal operation. Implementation resembles the
4168 * second-half of the ixgbevf_resume routine.
4170 static void ixgbevf_io_resume(struct pci_dev *pdev)
4172 struct net_device *netdev = pci_get_drvdata(pdev);
4173 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4175 if (netif_running(netdev))
4176 ixgbevf_up(adapter);
4178 netif_device_attach(netdev);
4181 /* PCI Error Recovery (ERS) */
4182 static const struct pci_error_handlers ixgbevf_err_handler = {
4183 .error_detected = ixgbevf_io_error_detected,
4184 .slot_reset = ixgbevf_io_slot_reset,
4185 .resume = ixgbevf_io_resume,
4188 static struct pci_driver ixgbevf_driver = {
4189 .name = ixgbevf_driver_name,
4190 .id_table = ixgbevf_pci_tbl,
4191 .probe = ixgbevf_probe,
4192 .remove = ixgbevf_remove,
4193 #ifdef CONFIG_PM
4194 /* Power Management Hooks */
4195 .suspend = ixgbevf_suspend,
4196 .resume = ixgbevf_resume,
4197 #endif
4198 .shutdown = ixgbevf_shutdown,
4199 .err_handler = &ixgbevf_err_handler
4203 * ixgbevf_init_module - Driver Registration Routine
4205 * ixgbevf_init_module is the first routine called when the driver is
4206 * loaded. All it does is register with the PCI subsystem.
4208 static int __init ixgbevf_init_module(void)
4210 int ret;
4212 pr_info("%s - version %s\n", ixgbevf_driver_string,
4213 ixgbevf_driver_version);
4215 pr_info("%s\n", ixgbevf_copyright);
4217 ret = pci_register_driver(&ixgbevf_driver);
4218 return ret;
4221 module_init(ixgbevf_init_module);
4224 * ixgbevf_exit_module - Driver Exit Cleanup Routine
4226 * ixgbevf_exit_module is called just before the driver is removed
4227 * from memory.
4229 static void __exit ixgbevf_exit_module(void)
4231 pci_unregister_driver(&ixgbevf_driver);
4234 #ifdef DEBUG
4236 * ixgbevf_get_hw_dev_name - return device name string
4237 * used by hardware layer to print debugging information
4239 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4241 struct ixgbevf_adapter *adapter = hw->back;
4243 return adapter->netdev->name;
4246 #endif
4247 module_exit(ixgbevf_exit_module);
4249 /* ixgbevf_main.c */