1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
4 /******************************************************************************
5 Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
6 ******************************************************************************/
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/types.h>
11 #include <linux/bitops.h>
12 #include <linux/module.h>
13 #include <linux/pci.h>
14 #include <linux/netdevice.h>
15 #include <linux/vmalloc.h>
16 #include <linux/string.h>
19 #include <linux/tcp.h>
20 #include <linux/sctp.h>
21 #include <linux/ipv6.h>
22 #include <linux/slab.h>
23 #include <net/checksum.h>
24 #include <net/ip6_checksum.h>
25 #include <linux/ethtool.h>
27 #include <linux/if_vlan.h>
28 #include <linux/prefetch.h>
30 #include <linux/bpf.h>
31 #include <linux/bpf_trace.h>
32 #include <linux/atomic.h>
36 const char ixgbevf_driver_name[] = "ixgbevf";
37 static const char ixgbevf_driver_string[] =
38 "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
40 #define DRV_VERSION "4.1.0-k"
41 const char ixgbevf_driver_version[] = DRV_VERSION;
42 static char ixgbevf_copyright[] =
43 "Copyright (c) 2009 - 2015 Intel Corporation.";
45 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
46 [board_82599_vf] = &ixgbevf_82599_vf_info,
47 [board_82599_vf_hv] = &ixgbevf_82599_vf_hv_info,
48 [board_X540_vf] = &ixgbevf_X540_vf_info,
49 [board_X540_vf_hv] = &ixgbevf_X540_vf_hv_info,
50 [board_X550_vf] = &ixgbevf_X550_vf_info,
51 [board_X550_vf_hv] = &ixgbevf_X550_vf_hv_info,
52 [board_X550EM_x_vf] = &ixgbevf_X550EM_x_vf_info,
53 [board_X550EM_x_vf_hv] = &ixgbevf_X550EM_x_vf_hv_info,
54 [board_x550em_a_vf] = &ixgbevf_x550em_a_vf_info,
57 /* ixgbevf_pci_tbl - PCI Device ID Table
59 * Wildcard entries (PCI_ANY_ID) should come last
60 * Last entry must be all 0s
62 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
63 * Class, Class Mask, private data (not used) }
65 static const struct pci_device_id ixgbevf_pci_tbl[] = {
66 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
67 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
68 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
69 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
70 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
71 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
72 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
73 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
74 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
75 /* required last entry */
78 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
80 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
81 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
82 MODULE_LICENSE("GPL");
83 MODULE_VERSION(DRV_VERSION);
85 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
86 static int debug = -1;
87 module_param(debug, int, 0);
88 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
90 static struct workqueue_struct *ixgbevf_wq;
92 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
94 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
95 !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
96 !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
97 queue_work(ixgbevf_wq, &adapter->service_task);
100 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
102 BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
104 /* flush memory to make sure state is correct before next watchdog */
105 smp_mb__before_atomic();
106 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
110 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
111 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
112 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
113 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer);
114 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
115 struct ixgbevf_rx_buffer *old_buff);
117 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
119 struct ixgbevf_adapter *adapter = hw->back;
124 dev_err(&adapter->pdev->dev, "Adapter removed\n");
125 if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
126 ixgbevf_service_event_schedule(adapter);
129 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
133 /* The following check not only optimizes a bit by not
134 * performing a read on the status register when the
135 * register just read was a status register read that
136 * returned IXGBE_FAILED_READ_REG. It also blocks any
137 * potential recursion.
139 if (reg == IXGBE_VFSTATUS) {
140 ixgbevf_remove_adapter(hw);
143 value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
144 if (value == IXGBE_FAILED_READ_REG)
145 ixgbevf_remove_adapter(hw);
148 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
150 u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
153 if (IXGBE_REMOVED(reg_addr))
154 return IXGBE_FAILED_READ_REG;
155 value = readl(reg_addr + reg);
156 if (unlikely(value == IXGBE_FAILED_READ_REG))
157 ixgbevf_check_remove(hw, reg);
162 * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
163 * @adapter: pointer to adapter struct
164 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
165 * @queue: queue to map the corresponding interrupt to
166 * @msix_vector: the vector to map to the corresponding queue
168 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
169 u8 queue, u8 msix_vector)
172 struct ixgbe_hw *hw = &adapter->hw;
174 if (direction == -1) {
176 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
177 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
180 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
182 /* Tx or Rx causes */
183 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
184 index = ((16 * (queue & 1)) + (8 * direction));
185 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
186 ivar &= ~(0xFF << index);
187 ivar |= (msix_vector << index);
188 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
192 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
194 return ring->stats.packets;
197 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
199 struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
200 struct ixgbe_hw *hw = &adapter->hw;
202 u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
203 u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
206 return (head < tail) ?
207 tail - head : (tail + ring->count - head);
212 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
214 u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
215 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
216 u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
218 clear_check_for_tx_hang(tx_ring);
220 /* Check for a hung queue, but be thorough. This verifies
221 * that a transmit has been completed since the previous
222 * check AND there is at least one packet pending. The
223 * ARMED bit is set to indicate a potential hang.
225 if ((tx_done_old == tx_done) && tx_pending) {
226 /* make sure it is true for two checks in a row */
227 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
230 /* reset the countdown */
231 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
233 /* update completed stats and continue */
234 tx_ring->tx_stats.tx_done_old = tx_done;
239 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
241 /* Do the reset outside of interrupt context */
242 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
243 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
244 ixgbevf_service_event_schedule(adapter);
249 * ixgbevf_tx_timeout - Respond to a Tx Hang
250 * @netdev: network interface device structure
252 static void ixgbevf_tx_timeout(struct net_device *netdev)
254 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
256 ixgbevf_tx_timeout_reset(adapter);
260 * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
261 * @q_vector: board private structure
262 * @tx_ring: tx ring to clean
263 * @napi_budget: Used to determine if we are in netpoll
265 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
266 struct ixgbevf_ring *tx_ring, int napi_budget)
268 struct ixgbevf_adapter *adapter = q_vector->adapter;
269 struct ixgbevf_tx_buffer *tx_buffer;
270 union ixgbe_adv_tx_desc *tx_desc;
271 unsigned int total_bytes = 0, total_packets = 0;
272 unsigned int budget = tx_ring->count / 2;
273 unsigned int i = tx_ring->next_to_clean;
275 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
278 tx_buffer = &tx_ring->tx_buffer_info[i];
279 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
283 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
285 /* if next_to_watch is not set then there is no work pending */
289 /* prevent any other reads prior to eop_desc */
292 /* if DD is not set pending work has not been completed */
293 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
296 /* clear next_to_watch to prevent false hangs */
297 tx_buffer->next_to_watch = NULL;
299 /* update the statistics for this packet */
300 total_bytes += tx_buffer->bytecount;
301 total_packets += tx_buffer->gso_segs;
304 if (ring_is_xdp(tx_ring))
305 page_frag_free(tx_buffer->data);
307 napi_consume_skb(tx_buffer->skb, napi_budget);
309 /* unmap skb header data */
310 dma_unmap_single(tx_ring->dev,
311 dma_unmap_addr(tx_buffer, dma),
312 dma_unmap_len(tx_buffer, len),
315 /* clear tx_buffer data */
316 dma_unmap_len_set(tx_buffer, len, 0);
318 /* unmap remaining buffers */
319 while (tx_desc != eop_desc) {
325 tx_buffer = tx_ring->tx_buffer_info;
326 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
329 /* unmap any remaining paged data */
330 if (dma_unmap_len(tx_buffer, len)) {
331 dma_unmap_page(tx_ring->dev,
332 dma_unmap_addr(tx_buffer, dma),
333 dma_unmap_len(tx_buffer, len),
335 dma_unmap_len_set(tx_buffer, len, 0);
339 /* move us one more past the eop_desc for start of next pkt */
345 tx_buffer = tx_ring->tx_buffer_info;
346 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
349 /* issue prefetch for next Tx descriptor */
352 /* update budget accounting */
354 } while (likely(budget));
357 tx_ring->next_to_clean = i;
358 u64_stats_update_begin(&tx_ring->syncp);
359 tx_ring->stats.bytes += total_bytes;
360 tx_ring->stats.packets += total_packets;
361 u64_stats_update_end(&tx_ring->syncp);
362 q_vector->tx.total_bytes += total_bytes;
363 q_vector->tx.total_packets += total_packets;
365 if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
366 struct ixgbe_hw *hw = &adapter->hw;
367 union ixgbe_adv_tx_desc *eop_desc;
369 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
371 pr_err("Detected Tx Unit Hang%s\n"
373 " TDH, TDT <%x>, <%x>\n"
374 " next_to_use <%x>\n"
375 " next_to_clean <%x>\n"
376 "tx_buffer_info[next_to_clean]\n"
377 " next_to_watch <%p>\n"
378 " eop_desc->wb.status <%x>\n"
379 " time_stamp <%lx>\n"
381 ring_is_xdp(tx_ring) ? " XDP" : "",
382 tx_ring->queue_index,
383 IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
384 IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
385 tx_ring->next_to_use, i,
386 eop_desc, (eop_desc ? eop_desc->wb.status : 0),
387 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
389 if (!ring_is_xdp(tx_ring))
390 netif_stop_subqueue(tx_ring->netdev,
391 tx_ring->queue_index);
393 /* schedule immediate reset if we believe we hung */
394 ixgbevf_tx_timeout_reset(adapter);
399 if (ring_is_xdp(tx_ring))
402 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
403 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
404 (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
405 /* Make sure that anybody stopping the queue after this
406 * sees the new next_to_clean.
410 if (__netif_subqueue_stopped(tx_ring->netdev,
411 tx_ring->queue_index) &&
412 !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
413 netif_wake_subqueue(tx_ring->netdev,
414 tx_ring->queue_index);
415 ++tx_ring->tx_stats.restart_queue;
423 * ixgbevf_rx_skb - Helper function to determine proper Rx method
424 * @q_vector: structure containing interrupt and ring information
425 * @skb: packet to send up
427 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
430 napi_gro_receive(&q_vector->napi, skb);
433 #define IXGBE_RSS_L4_TYPES_MASK \
434 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
435 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
436 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
437 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
439 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
440 union ixgbe_adv_rx_desc *rx_desc,
445 if (!(ring->netdev->features & NETIF_F_RXHASH))
448 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
449 IXGBE_RXDADV_RSSTYPE_MASK;
454 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
455 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
456 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
460 * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
461 * @ring: structure containig ring specific data
462 * @rx_desc: current Rx descriptor being processed
463 * @skb: skb currently being received and modified
465 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
466 union ixgbe_adv_rx_desc *rx_desc,
469 skb_checksum_none_assert(skb);
471 /* Rx csum disabled */
472 if (!(ring->netdev->features & NETIF_F_RXCSUM))
475 /* if IP and error */
476 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
477 ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
478 ring->rx_stats.csum_err++;
482 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
485 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
486 ring->rx_stats.csum_err++;
490 /* It must be a TCP or UDP packet with a valid checksum */
491 skb->ip_summed = CHECKSUM_UNNECESSARY;
495 * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
496 * @rx_ring: rx descriptor ring packet is being transacted on
497 * @rx_desc: pointer to the EOP Rx descriptor
498 * @skb: pointer to current skb being populated
500 * This function checks the ring, descriptor, and packet information in
501 * order to populate the checksum, VLAN, protocol, and other fields within
504 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
505 union ixgbe_adv_rx_desc *rx_desc,
508 ixgbevf_rx_hash(rx_ring, rx_desc, 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 struct ixgbevf_rx_buffer *ixgbevf_get_rx_buffer(struct ixgbevf_ring *rx_ring,
524 const unsigned int size)
526 struct ixgbevf_rx_buffer *rx_buffer;
528 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
529 prefetchw(rx_buffer->page);
531 /* we are reusing so sync this buffer for CPU use */
532 dma_sync_single_range_for_cpu(rx_ring->dev,
534 rx_buffer->page_offset,
538 rx_buffer->pagecnt_bias--;
543 static void ixgbevf_put_rx_buffer(struct ixgbevf_ring *rx_ring,
544 struct ixgbevf_rx_buffer *rx_buffer,
547 if (ixgbevf_can_reuse_rx_page(rx_buffer)) {
548 /* hand second half of page back to the ring */
549 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
552 /* We are not reusing the buffer so unmap it and free
553 * any references we are holding to it
555 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
556 ixgbevf_rx_pg_size(rx_ring),
558 IXGBEVF_RX_DMA_ATTR);
559 __page_frag_cache_drain(rx_buffer->page,
560 rx_buffer->pagecnt_bias);
563 /* clear contents of rx_buffer */
564 rx_buffer->page = NULL;
568 * ixgbevf_is_non_eop - process handling of non-EOP buffers
569 * @rx_ring: Rx ring being processed
570 * @rx_desc: Rx descriptor for current buffer
572 * This function updates next to clean. If the buffer is an EOP buffer
573 * this function exits returning false, otherwise it will place the
574 * sk_buff in the next buffer to be chained and return true indicating
575 * that this is in fact a non-EOP buffer.
577 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
578 union ixgbe_adv_rx_desc *rx_desc)
580 u32 ntc = rx_ring->next_to_clean + 1;
582 /* fetch, update, and store next to clean */
583 ntc = (ntc < rx_ring->count) ? ntc : 0;
584 rx_ring->next_to_clean = ntc;
586 prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
588 if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
594 static inline unsigned int ixgbevf_rx_offset(struct ixgbevf_ring *rx_ring)
596 return ring_uses_build_skb(rx_ring) ? IXGBEVF_SKB_PAD : 0;
599 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
600 struct ixgbevf_rx_buffer *bi)
602 struct page *page = bi->page;
605 /* since we are recycling buffers we should seldom need to alloc */
609 /* alloc new page for storage */
610 page = dev_alloc_pages(ixgbevf_rx_pg_order(rx_ring));
611 if (unlikely(!page)) {
612 rx_ring->rx_stats.alloc_rx_page_failed++;
616 /* map page for use */
617 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
618 ixgbevf_rx_pg_size(rx_ring),
619 DMA_FROM_DEVICE, IXGBEVF_RX_DMA_ATTR);
621 /* if mapping failed free memory back to system since
622 * there isn't much point in holding memory we can't use
624 if (dma_mapping_error(rx_ring->dev, dma)) {
625 __free_pages(page, ixgbevf_rx_pg_order(rx_ring));
627 rx_ring->rx_stats.alloc_rx_page_failed++;
633 bi->page_offset = ixgbevf_rx_offset(rx_ring);
634 bi->pagecnt_bias = 1;
635 rx_ring->rx_stats.alloc_rx_page++;
641 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
642 * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
643 * @cleaned_count: number of buffers to replace
645 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
648 union ixgbe_adv_rx_desc *rx_desc;
649 struct ixgbevf_rx_buffer *bi;
650 unsigned int i = rx_ring->next_to_use;
652 /* nothing to do or no valid netdev defined */
653 if (!cleaned_count || !rx_ring->netdev)
656 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
657 bi = &rx_ring->rx_buffer_info[i];
661 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
664 /* sync the buffer for use by the device */
665 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
667 ixgbevf_rx_bufsz(rx_ring),
670 /* Refresh the desc even if pkt_addr didn't change
671 * because each write-back erases this info.
673 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
679 rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
680 bi = rx_ring->rx_buffer_info;
684 /* clear the length for the next_to_use descriptor */
685 rx_desc->wb.upper.length = 0;
688 } while (cleaned_count);
692 if (rx_ring->next_to_use != i) {
693 /* record the next descriptor to use */
694 rx_ring->next_to_use = i;
696 /* update next to alloc since we have filled the ring */
697 rx_ring->next_to_alloc = i;
699 /* Force memory writes to complete before letting h/w
700 * know there are new descriptors to fetch. (Only
701 * applicable for weak-ordered memory model archs,
705 ixgbevf_write_tail(rx_ring, i);
710 * ixgbevf_cleanup_headers - Correct corrupted or empty headers
711 * @rx_ring: rx descriptor ring packet is being transacted on
712 * @rx_desc: pointer to the EOP Rx descriptor
713 * @skb: pointer to current skb being fixed
715 * Check for corrupted packet headers caused by senders on the local L2
716 * embedded NIC switch not setting up their Tx Descriptors right. These
717 * should be very rare.
719 * Also address the case where we are pulling data in on pages only
720 * and as such no data is present in the skb header.
722 * In addition if skb is not at least 60 bytes we need to pad it so that
723 * it is large enough to qualify as a valid Ethernet frame.
725 * Returns true if an error was encountered and skb was freed.
727 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
728 union ixgbe_adv_rx_desc *rx_desc,
731 /* XDP packets use error pointer so abort at this point */
735 /* verify that the packet does not have any known errors */
736 if (unlikely(ixgbevf_test_staterr(rx_desc,
737 IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
738 struct net_device *netdev = rx_ring->netdev;
740 if (!(netdev->features & NETIF_F_RXALL)) {
741 dev_kfree_skb_any(skb);
746 /* if eth_skb_pad returns an error the skb was freed */
747 if (eth_skb_pad(skb))
754 * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
755 * @rx_ring: rx descriptor ring to store buffers on
756 * @old_buff: donor buffer to have page reused
758 * Synchronizes page for reuse by the adapter
760 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
761 struct ixgbevf_rx_buffer *old_buff)
763 struct ixgbevf_rx_buffer *new_buff;
764 u16 nta = rx_ring->next_to_alloc;
766 new_buff = &rx_ring->rx_buffer_info[nta];
768 /* update, and store next to alloc */
770 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
772 /* transfer page from old buffer to new buffer */
773 new_buff->page = old_buff->page;
774 new_buff->dma = old_buff->dma;
775 new_buff->page_offset = old_buff->page_offset;
776 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
779 static inline bool ixgbevf_page_is_reserved(struct page *page)
781 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
784 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer)
786 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
787 struct page *page = rx_buffer->page;
789 /* avoid re-using remote pages */
790 if (unlikely(ixgbevf_page_is_reserved(page)))
793 #if (PAGE_SIZE < 8192)
794 /* if we are only owner of page we can reuse it */
795 if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
798 #define IXGBEVF_LAST_OFFSET \
799 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBEVF_RXBUFFER_2048)
801 if (rx_buffer->page_offset > IXGBEVF_LAST_OFFSET)
806 /* If we have drained the page fragment pool we need to update
807 * the pagecnt_bias and page count so that we fully restock the
808 * number of references the driver holds.
810 if (unlikely(!pagecnt_bias)) {
811 page_ref_add(page, USHRT_MAX);
812 rx_buffer->pagecnt_bias = USHRT_MAX;
819 * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
820 * @rx_ring: rx descriptor ring to transact packets on
821 * @rx_buffer: buffer containing page to add
822 * @skb: sk_buff to place the data into
823 * @size: size of buffer to be added
825 * This function will add the data contained in rx_buffer->page to the skb.
827 static void ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
828 struct ixgbevf_rx_buffer *rx_buffer,
832 #if (PAGE_SIZE < 8192)
833 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
835 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
836 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
837 SKB_DATA_ALIGN(size);
839 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
840 rx_buffer->page_offset, size, truesize);
841 #if (PAGE_SIZE < 8192)
842 rx_buffer->page_offset ^= truesize;
844 rx_buffer->page_offset += truesize;
849 struct sk_buff *ixgbevf_construct_skb(struct ixgbevf_ring *rx_ring,
850 struct ixgbevf_rx_buffer *rx_buffer,
851 struct xdp_buff *xdp,
852 union ixgbe_adv_rx_desc *rx_desc)
854 unsigned int size = xdp->data_end - xdp->data;
855 #if (PAGE_SIZE < 8192)
856 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
858 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
859 xdp->data_hard_start);
861 unsigned int headlen;
864 /* prefetch first cache line of first page */
866 #if L1_CACHE_BYTES < 128
867 prefetch(xdp->data + L1_CACHE_BYTES);
869 /* Note, we get here by enabling legacy-rx via:
871 * ethtool --set-priv-flags <dev> legacy-rx on
873 * In this mode, we currently get 0 extra XDP headroom as
874 * opposed to having legacy-rx off, where we process XDP
875 * packets going to stack via ixgbevf_build_skb().
877 * For ixgbevf_construct_skb() mode it means that the
878 * xdp->data_meta will always point to xdp->data, since
879 * the helper cannot expand the head. Should this ever
880 * changed in future for legacy-rx mode on, then lets also
881 * add xdp->data_meta handling here.
884 /* allocate a skb to store the frags */
885 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBEVF_RX_HDR_SIZE);
889 /* Determine available headroom for copy */
891 if (headlen > IXGBEVF_RX_HDR_SIZE)
892 headlen = eth_get_headlen(xdp->data, IXGBEVF_RX_HDR_SIZE);
894 /* align pull length to size of long to optimize memcpy performance */
895 memcpy(__skb_put(skb, headlen), xdp->data,
896 ALIGN(headlen, sizeof(long)));
898 /* update all of the pointers */
901 skb_add_rx_frag(skb, 0, rx_buffer->page,
902 (xdp->data + headlen) -
903 page_address(rx_buffer->page),
905 #if (PAGE_SIZE < 8192)
906 rx_buffer->page_offset ^= truesize;
908 rx_buffer->page_offset += truesize;
911 rx_buffer->pagecnt_bias++;
917 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
920 struct ixgbe_hw *hw = &adapter->hw;
922 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
925 static struct sk_buff *ixgbevf_build_skb(struct ixgbevf_ring *rx_ring,
926 struct ixgbevf_rx_buffer *rx_buffer,
927 struct xdp_buff *xdp,
928 union ixgbe_adv_rx_desc *rx_desc)
930 unsigned int metasize = xdp->data - xdp->data_meta;
931 #if (PAGE_SIZE < 8192)
932 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
934 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
935 SKB_DATA_ALIGN(xdp->data_end -
936 xdp->data_hard_start);
940 /* Prefetch first cache line of first page. If xdp->data_meta
941 * is unused, this points to xdp->data, otherwise, we likely
942 * have a consumer accessing first few bytes of meta data,
943 * and then actual data.
945 prefetch(xdp->data_meta);
946 #if L1_CACHE_BYTES < 128
947 prefetch(xdp->data_meta + L1_CACHE_BYTES);
950 /* build an skb around the page buffer */
951 skb = build_skb(xdp->data_hard_start, truesize);
955 /* update pointers within the skb to store the data */
956 skb_reserve(skb, xdp->data - xdp->data_hard_start);
957 __skb_put(skb, xdp->data_end - xdp->data);
959 skb_metadata_set(skb, metasize);
961 /* update buffer offset */
962 #if (PAGE_SIZE < 8192)
963 rx_buffer->page_offset ^= truesize;
965 rx_buffer->page_offset += truesize;
971 #define IXGBEVF_XDP_PASS 0
972 #define IXGBEVF_XDP_CONSUMED 1
973 #define IXGBEVF_XDP_TX 2
975 static int ixgbevf_xmit_xdp_ring(struct ixgbevf_ring *ring,
976 struct xdp_buff *xdp)
978 struct ixgbevf_tx_buffer *tx_buffer;
979 union ixgbe_adv_tx_desc *tx_desc;
984 len = xdp->data_end - xdp->data;
986 if (unlikely(!ixgbevf_desc_unused(ring)))
987 return IXGBEVF_XDP_CONSUMED;
989 dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE);
990 if (dma_mapping_error(ring->dev, dma))
991 return IXGBEVF_XDP_CONSUMED;
993 /* record the location of the first descriptor for this packet */
994 i = ring->next_to_use;
995 tx_buffer = &ring->tx_buffer_info[i];
997 dma_unmap_len_set(tx_buffer, len, len);
998 dma_unmap_addr_set(tx_buffer, dma, dma);
999 tx_buffer->data = xdp->data;
1000 tx_buffer->bytecount = len;
1001 tx_buffer->gso_segs = 1;
1002 tx_buffer->protocol = 0;
1004 /* Populate minimal context descriptor that will provide for the
1005 * fact that we are expected to process Ethernet frames.
1007 if (!test_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state)) {
1008 struct ixgbe_adv_tx_context_desc *context_desc;
1010 set_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1012 context_desc = IXGBEVF_TX_CTXTDESC(ring, 0);
1013 context_desc->vlan_macip_lens =
1014 cpu_to_le32(ETH_HLEN << IXGBE_ADVTXD_MACLEN_SHIFT);
1015 context_desc->seqnum_seed = 0;
1016 context_desc->type_tucmd_mlhl =
1017 cpu_to_le32(IXGBE_TXD_CMD_DEXT |
1018 IXGBE_ADVTXD_DTYP_CTXT);
1019 context_desc->mss_l4len_idx = 0;
1024 /* put descriptor type bits */
1025 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
1026 IXGBE_ADVTXD_DCMD_DEXT |
1027 IXGBE_ADVTXD_DCMD_IFCS;
1028 cmd_type |= len | IXGBE_TXD_CMD;
1030 tx_desc = IXGBEVF_TX_DESC(ring, i);
1031 tx_desc->read.buffer_addr = cpu_to_le64(dma);
1033 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1034 tx_desc->read.olinfo_status =
1035 cpu_to_le32((len << IXGBE_ADVTXD_PAYLEN_SHIFT) |
1038 /* Avoid any potential race with cleanup */
1041 /* set next_to_watch value indicating a packet is present */
1043 if (i == ring->count)
1046 tx_buffer->next_to_watch = tx_desc;
1047 ring->next_to_use = i;
1049 return IXGBEVF_XDP_TX;
1052 static struct sk_buff *ixgbevf_run_xdp(struct ixgbevf_adapter *adapter,
1053 struct ixgbevf_ring *rx_ring,
1054 struct xdp_buff *xdp)
1056 int result = IXGBEVF_XDP_PASS;
1057 struct ixgbevf_ring *xdp_ring;
1058 struct bpf_prog *xdp_prog;
1062 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
1067 act = bpf_prog_run_xdp(xdp_prog, xdp);
1072 xdp_ring = adapter->xdp_ring[rx_ring->queue_index];
1073 result = ixgbevf_xmit_xdp_ring(xdp_ring, xdp);
1076 bpf_warn_invalid_xdp_action(act);
1079 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
1080 /* fallthrough -- handle aborts by dropping packet */
1082 result = IXGBEVF_XDP_CONSUMED;
1087 return ERR_PTR(-result);
1090 static void ixgbevf_rx_buffer_flip(struct ixgbevf_ring *rx_ring,
1091 struct ixgbevf_rx_buffer *rx_buffer,
1094 #if (PAGE_SIZE < 8192)
1095 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
1097 rx_buffer->page_offset ^= truesize;
1099 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
1100 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
1101 SKB_DATA_ALIGN(size);
1103 rx_buffer->page_offset += truesize;
1107 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
1108 struct ixgbevf_ring *rx_ring,
1111 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1112 struct ixgbevf_adapter *adapter = q_vector->adapter;
1113 u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
1114 struct sk_buff *skb = rx_ring->skb;
1115 bool xdp_xmit = false;
1116 struct xdp_buff xdp;
1118 xdp.rxq = &rx_ring->xdp_rxq;
1120 while (likely(total_rx_packets < budget)) {
1121 struct ixgbevf_rx_buffer *rx_buffer;
1122 union ixgbe_adv_rx_desc *rx_desc;
1125 /* return some buffers to hardware, one at a time is too slow */
1126 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
1127 ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
1131 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
1132 size = le16_to_cpu(rx_desc->wb.upper.length);
1136 /* This memory barrier is needed to keep us from reading
1137 * any other fields out of the rx_desc until we know the
1138 * RXD_STAT_DD bit is set
1142 rx_buffer = ixgbevf_get_rx_buffer(rx_ring, size);
1144 /* retrieve a buffer from the ring */
1146 xdp.data = page_address(rx_buffer->page) +
1147 rx_buffer->page_offset;
1148 xdp.data_meta = xdp.data;
1149 xdp.data_hard_start = xdp.data -
1150 ixgbevf_rx_offset(rx_ring);
1151 xdp.data_end = xdp.data + size;
1153 skb = ixgbevf_run_xdp(adapter, rx_ring, &xdp);
1157 if (PTR_ERR(skb) == -IXGBEVF_XDP_TX) {
1159 ixgbevf_rx_buffer_flip(rx_ring, rx_buffer,
1162 rx_buffer->pagecnt_bias++;
1165 total_rx_bytes += size;
1167 ixgbevf_add_rx_frag(rx_ring, rx_buffer, skb, size);
1168 } else if (ring_uses_build_skb(rx_ring)) {
1169 skb = ixgbevf_build_skb(rx_ring, rx_buffer,
1172 skb = ixgbevf_construct_skb(rx_ring, rx_buffer,
1176 /* exit if we failed to retrieve a buffer */
1178 rx_ring->rx_stats.alloc_rx_buff_failed++;
1179 rx_buffer->pagecnt_bias++;
1183 ixgbevf_put_rx_buffer(rx_ring, rx_buffer, skb);
1186 /* fetch next buffer in frame if non-eop */
1187 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
1190 /* verify the packet layout is correct */
1191 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
1196 /* probably a little skewed due to removing CRC */
1197 total_rx_bytes += skb->len;
1199 /* Workaround hardware that can't do proper VEPA multicast
1202 if ((skb->pkt_type == PACKET_BROADCAST ||
1203 skb->pkt_type == PACKET_MULTICAST) &&
1204 ether_addr_equal(rx_ring->netdev->dev_addr,
1205 eth_hdr(skb)->h_source)) {
1206 dev_kfree_skb_irq(skb);
1210 /* populate checksum, VLAN, and protocol */
1211 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
1213 ixgbevf_rx_skb(q_vector, skb);
1215 /* reset skb pointer */
1218 /* update budget accounting */
1222 /* place incomplete frames back on ring for completion */
1226 struct ixgbevf_ring *xdp_ring =
1227 adapter->xdp_ring[rx_ring->queue_index];
1229 /* Force memory writes to complete before letting h/w
1230 * know there are new descriptors to fetch.
1233 ixgbevf_write_tail(xdp_ring, xdp_ring->next_to_use);
1236 u64_stats_update_begin(&rx_ring->syncp);
1237 rx_ring->stats.packets += total_rx_packets;
1238 rx_ring->stats.bytes += total_rx_bytes;
1239 u64_stats_update_end(&rx_ring->syncp);
1240 q_vector->rx.total_packets += total_rx_packets;
1241 q_vector->rx.total_bytes += total_rx_bytes;
1243 return total_rx_packets;
1247 * ixgbevf_poll - NAPI polling calback
1248 * @napi: napi struct with our devices info in it
1249 * @budget: amount of work driver is allowed to do this pass, in packets
1251 * This function will clean more than one or more rings associated with a
1254 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1256 struct ixgbevf_q_vector *q_vector =
1257 container_of(napi, struct ixgbevf_q_vector, napi);
1258 struct ixgbevf_adapter *adapter = q_vector->adapter;
1259 struct ixgbevf_ring *ring;
1260 int per_ring_budget, work_done = 0;
1261 bool clean_complete = true;
1263 ixgbevf_for_each_ring(ring, q_vector->tx) {
1264 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1265 clean_complete = false;
1271 /* attempt to distribute budget to each queue fairly, but don't allow
1272 * the budget to go below 1 because we'll exit polling
1274 if (q_vector->rx.count > 1)
1275 per_ring_budget = max(budget/q_vector->rx.count, 1);
1277 per_ring_budget = budget;
1279 ixgbevf_for_each_ring(ring, q_vector->rx) {
1280 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1282 work_done += cleaned;
1283 if (cleaned >= per_ring_budget)
1284 clean_complete = false;
1287 /* If all work not completed, return budget and keep polling */
1288 if (!clean_complete)
1290 /* all work done, exit the polling mode */
1291 napi_complete_done(napi, work_done);
1292 if (adapter->rx_itr_setting == 1)
1293 ixgbevf_set_itr(q_vector);
1294 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1295 !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1296 ixgbevf_irq_enable_queues(adapter,
1297 BIT(q_vector->v_idx));
1303 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1304 * @q_vector: structure containing interrupt and ring information
1306 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1308 struct ixgbevf_adapter *adapter = q_vector->adapter;
1309 struct ixgbe_hw *hw = &adapter->hw;
1310 int v_idx = q_vector->v_idx;
1311 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1313 /* set the WDIS bit to not clear the timer bits and cause an
1314 * immediate assertion of the interrupt
1316 itr_reg |= IXGBE_EITR_CNT_WDIS;
1318 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1322 * ixgbevf_configure_msix - Configure MSI-X hardware
1323 * @adapter: board private structure
1325 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1328 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1330 struct ixgbevf_q_vector *q_vector;
1331 int q_vectors, v_idx;
1333 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1334 adapter->eims_enable_mask = 0;
1336 /* Populate the IVAR table and set the ITR values to the
1337 * corresponding register.
1339 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1340 struct ixgbevf_ring *ring;
1342 q_vector = adapter->q_vector[v_idx];
1344 ixgbevf_for_each_ring(ring, q_vector->rx)
1345 ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1347 ixgbevf_for_each_ring(ring, q_vector->tx)
1348 ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1350 if (q_vector->tx.ring && !q_vector->rx.ring) {
1351 /* Tx only vector */
1352 if (adapter->tx_itr_setting == 1)
1353 q_vector->itr = IXGBE_12K_ITR;
1355 q_vector->itr = adapter->tx_itr_setting;
1357 /* Rx or Rx/Tx vector */
1358 if (adapter->rx_itr_setting == 1)
1359 q_vector->itr = IXGBE_20K_ITR;
1361 q_vector->itr = adapter->rx_itr_setting;
1364 /* add q_vector eims value to global eims_enable_mask */
1365 adapter->eims_enable_mask |= BIT(v_idx);
1367 ixgbevf_write_eitr(q_vector);
1370 ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1371 /* setup eims_other and add value to global eims_enable_mask */
1372 adapter->eims_other = BIT(v_idx);
1373 adapter->eims_enable_mask |= adapter->eims_other;
1376 enum latency_range {
1380 latency_invalid = 255
1384 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1385 * @q_vector: structure containing interrupt and ring information
1386 * @ring_container: structure containing ring performance data
1388 * Stores a new ITR value based on packets and byte
1389 * counts during the last interrupt. The advantage of per interrupt
1390 * computation is faster updates and more accurate ITR for the current
1391 * traffic pattern. Constants in this function were computed
1392 * based on theoretical maximum wire speed and thresholds were set based
1393 * on testing data as well as attempting to minimize response time
1394 * while increasing bulk throughput.
1396 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1397 struct ixgbevf_ring_container *ring_container)
1399 int bytes = ring_container->total_bytes;
1400 int packets = ring_container->total_packets;
1403 u8 itr_setting = ring_container->itr;
1408 /* simple throttle rate management
1409 * 0-20MB/s lowest (100000 ints/s)
1410 * 20-100MB/s low (20000 ints/s)
1411 * 100-1249MB/s bulk (12000 ints/s)
1413 /* what was last interrupt timeslice? */
1414 timepassed_us = q_vector->itr >> 2;
1415 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1417 switch (itr_setting) {
1418 case lowest_latency:
1419 if (bytes_perint > 10)
1420 itr_setting = low_latency;
1423 if (bytes_perint > 20)
1424 itr_setting = bulk_latency;
1425 else if (bytes_perint <= 10)
1426 itr_setting = lowest_latency;
1429 if (bytes_perint <= 20)
1430 itr_setting = low_latency;
1434 /* clear work counters since we have the values we need */
1435 ring_container->total_bytes = 0;
1436 ring_container->total_packets = 0;
1438 /* write updated itr to ring container */
1439 ring_container->itr = itr_setting;
1442 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1444 u32 new_itr = q_vector->itr;
1447 ixgbevf_update_itr(q_vector, &q_vector->tx);
1448 ixgbevf_update_itr(q_vector, &q_vector->rx);
1450 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1452 switch (current_itr) {
1453 /* counts and packets in update_itr are dependent on these numbers */
1454 case lowest_latency:
1455 new_itr = IXGBE_100K_ITR;
1458 new_itr = IXGBE_20K_ITR;
1461 new_itr = IXGBE_12K_ITR;
1467 if (new_itr != q_vector->itr) {
1468 /* do an exponential smoothing */
1469 new_itr = (10 * new_itr * q_vector->itr) /
1470 ((9 * new_itr) + q_vector->itr);
1472 /* save the algorithm value here */
1473 q_vector->itr = new_itr;
1475 ixgbevf_write_eitr(q_vector);
1479 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1481 struct ixgbevf_adapter *adapter = data;
1482 struct ixgbe_hw *hw = &adapter->hw;
1484 hw->mac.get_link_status = 1;
1486 ixgbevf_service_event_schedule(adapter);
1488 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1494 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1496 * @data: pointer to our q_vector struct for this interrupt vector
1498 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1500 struct ixgbevf_q_vector *q_vector = data;
1502 /* EIAM disabled interrupts (on this vector) for us */
1503 if (q_vector->rx.ring || q_vector->tx.ring)
1504 napi_schedule_irqoff(&q_vector->napi);
1510 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1511 * @adapter: board private structure
1513 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1514 * interrupts from the kernel.
1516 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1518 struct net_device *netdev = adapter->netdev;
1519 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1520 unsigned int ri = 0, ti = 0;
1523 for (vector = 0; vector < q_vectors; vector++) {
1524 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1525 struct msix_entry *entry = &adapter->msix_entries[vector];
1527 if (q_vector->tx.ring && q_vector->rx.ring) {
1528 snprintf(q_vector->name, sizeof(q_vector->name),
1529 "%s-TxRx-%u", netdev->name, ri++);
1531 } else if (q_vector->rx.ring) {
1532 snprintf(q_vector->name, sizeof(q_vector->name),
1533 "%s-rx-%u", netdev->name, ri++);
1534 } else if (q_vector->tx.ring) {
1535 snprintf(q_vector->name, sizeof(q_vector->name),
1536 "%s-tx-%u", netdev->name, ti++);
1538 /* skip this unused q_vector */
1541 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1542 q_vector->name, q_vector);
1544 hw_dbg(&adapter->hw,
1545 "request_irq failed for MSIX interrupt Error: %d\n",
1547 goto free_queue_irqs;
1551 err = request_irq(adapter->msix_entries[vector].vector,
1552 &ixgbevf_msix_other, 0, netdev->name, adapter);
1554 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1556 goto free_queue_irqs;
1564 free_irq(adapter->msix_entries[vector].vector,
1565 adapter->q_vector[vector]);
1567 /* This failure is non-recoverable - it indicates the system is
1568 * out of MSIX vector resources and the VF driver cannot run
1569 * without them. Set the number of msix vectors to zero
1570 * indicating that not enough can be allocated. The error
1571 * will be returned to the user indicating device open failed.
1572 * Any further attempts to force the driver to open will also
1573 * fail. The only way to recover is to unload the driver and
1574 * reload it again. If the system has recovered some MSIX
1575 * vectors then it may succeed.
1577 adapter->num_msix_vectors = 0;
1582 * ixgbevf_request_irq - initialize interrupts
1583 * @adapter: board private structure
1585 * Attempts to configure interrupts using the best available
1586 * capabilities of the hardware and kernel.
1588 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1590 int err = ixgbevf_request_msix_irqs(adapter);
1593 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1598 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1602 if (!adapter->msix_entries)
1605 q_vectors = adapter->num_msix_vectors;
1608 free_irq(adapter->msix_entries[i].vector, adapter);
1611 for (; i >= 0; i--) {
1612 /* free only the irqs that were actually requested */
1613 if (!adapter->q_vector[i]->rx.ring &&
1614 !adapter->q_vector[i]->tx.ring)
1617 free_irq(adapter->msix_entries[i].vector,
1618 adapter->q_vector[i]);
1623 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1624 * @adapter: board private structure
1626 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1628 struct ixgbe_hw *hw = &adapter->hw;
1631 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1632 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1633 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1635 IXGBE_WRITE_FLUSH(hw);
1637 for (i = 0; i < adapter->num_msix_vectors; i++)
1638 synchronize_irq(adapter->msix_entries[i].vector);
1642 * ixgbevf_irq_enable - Enable default interrupt generation settings
1643 * @adapter: board private structure
1645 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1647 struct ixgbe_hw *hw = &adapter->hw;
1649 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1650 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1651 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1655 * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1656 * @adapter: board private structure
1657 * @ring: structure containing ring specific data
1659 * Configure the Tx descriptor ring after a reset.
1661 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1662 struct ixgbevf_ring *ring)
1664 struct ixgbe_hw *hw = &adapter->hw;
1665 u64 tdba = ring->dma;
1667 u32 txdctl = IXGBE_TXDCTL_ENABLE;
1668 u8 reg_idx = ring->reg_idx;
1670 /* disable queue to avoid issues while updating state */
1671 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1672 IXGBE_WRITE_FLUSH(hw);
1674 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1675 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1676 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1677 ring->count * sizeof(union ixgbe_adv_tx_desc));
1679 /* disable head writeback */
1680 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1681 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1683 /* enable relaxed ordering */
1684 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1685 (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1686 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1688 /* reset head and tail pointers */
1689 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1690 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1691 ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1693 /* reset ntu and ntc to place SW in sync with hardwdare */
1694 ring->next_to_clean = 0;
1695 ring->next_to_use = 0;
1697 /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1698 * to or less than the number of on chip descriptors, which is
1701 txdctl |= (8 << 16); /* WTHRESH = 8 */
1703 /* Setting PTHRESH to 32 both improves performance */
1704 txdctl |= (1u << 8) | /* HTHRESH = 1 */
1705 32; /* PTHRESH = 32 */
1707 /* reinitialize tx_buffer_info */
1708 memset(ring->tx_buffer_info, 0,
1709 sizeof(struct ixgbevf_tx_buffer) * ring->count);
1711 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1712 clear_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1714 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1716 /* poll to verify queue is enabled */
1718 usleep_range(1000, 2000);
1719 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1720 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1722 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1726 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1727 * @adapter: board private structure
1729 * Configure the Tx unit of the MAC after a reset.
1731 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1735 /* Setup the HW Tx Head and Tail descriptor pointers */
1736 for (i = 0; i < adapter->num_tx_queues; i++)
1737 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1738 for (i = 0; i < adapter->num_xdp_queues; i++)
1739 ixgbevf_configure_tx_ring(adapter, adapter->xdp_ring[i]);
1742 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1744 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter,
1745 struct ixgbevf_ring *ring, int index)
1747 struct ixgbe_hw *hw = &adapter->hw;
1750 srrctl = IXGBE_SRRCTL_DROP_EN;
1752 srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1753 if (ring_uses_large_buffer(ring))
1754 srrctl |= IXGBEVF_RXBUFFER_3072 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1756 srrctl |= IXGBEVF_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1757 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1759 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1762 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1764 struct ixgbe_hw *hw = &adapter->hw;
1766 /* PSRTYPE must be initialized in 82599 */
1767 u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1768 IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1769 IXGBE_PSRTYPE_L2HDR;
1771 if (adapter->num_rx_queues > 1)
1774 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1777 #define IXGBEVF_MAX_RX_DESC_POLL 10
1778 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1779 struct ixgbevf_ring *ring)
1781 struct ixgbe_hw *hw = &adapter->hw;
1782 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1784 u8 reg_idx = ring->reg_idx;
1786 if (IXGBE_REMOVED(hw->hw_addr))
1788 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1789 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1791 /* write value back with RXDCTL.ENABLE bit cleared */
1792 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1794 /* the hardware may take up to 100us to really disable the Rx queue */
1797 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1798 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1801 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1805 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1806 struct ixgbevf_ring *ring)
1808 struct ixgbe_hw *hw = &adapter->hw;
1809 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1811 u8 reg_idx = ring->reg_idx;
1813 if (IXGBE_REMOVED(hw->hw_addr))
1816 usleep_range(1000, 2000);
1817 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1818 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1821 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1826 * ixgbevf_init_rss_key - Initialize adapter RSS key
1827 * @adapter: device handle
1829 * Allocates and initializes the RSS key if it is not allocated.
1831 static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter)
1835 if (!adapter->rss_key) {
1836 rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL);
1837 if (unlikely(!rss_key))
1840 netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE);
1841 adapter->rss_key = rss_key;
1847 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1849 struct ixgbe_hw *hw = &adapter->hw;
1850 u32 vfmrqc = 0, vfreta = 0;
1851 u16 rss_i = adapter->num_rx_queues;
1854 /* Fill out hash function seeds */
1855 for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1856 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i));
1858 for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1862 adapter->rss_indir_tbl[i] = j;
1864 vfreta |= j << (i & 0x3) * 8;
1866 IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1871 /* Perform hash on these packet types */
1872 vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1873 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1874 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1875 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1877 vfmrqc |= IXGBE_VFMRQC_RSSEN;
1879 IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1882 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1883 struct ixgbevf_ring *ring)
1885 struct ixgbe_hw *hw = &adapter->hw;
1886 union ixgbe_adv_rx_desc *rx_desc;
1887 u64 rdba = ring->dma;
1889 u8 reg_idx = ring->reg_idx;
1891 /* disable queue to avoid issues while updating state */
1892 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1893 ixgbevf_disable_rx_queue(adapter, ring);
1895 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1896 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1897 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1898 ring->count * sizeof(union ixgbe_adv_rx_desc));
1900 #ifndef CONFIG_SPARC
1901 /* enable relaxed ordering */
1902 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1903 IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1905 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1906 IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1907 IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1910 /* reset head and tail pointers */
1911 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1912 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1913 ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1915 /* initialize rx_buffer_info */
1916 memset(ring->rx_buffer_info, 0,
1917 sizeof(struct ixgbevf_rx_buffer) * ring->count);
1919 /* initialize Rx descriptor 0 */
1920 rx_desc = IXGBEVF_RX_DESC(ring, 0);
1921 rx_desc->wb.upper.length = 0;
1923 /* reset ntu and ntc to place SW in sync with hardwdare */
1924 ring->next_to_clean = 0;
1925 ring->next_to_use = 0;
1926 ring->next_to_alloc = 0;
1928 ixgbevf_configure_srrctl(adapter, ring, reg_idx);
1930 /* RXDCTL.RLPML does not work on 82599 */
1931 if (adapter->hw.mac.type != ixgbe_mac_82599_vf) {
1932 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
1933 IXGBE_RXDCTL_RLPML_EN);
1935 #if (PAGE_SIZE < 8192)
1936 /* Limit the maximum frame size so we don't overrun the skb */
1937 if (ring_uses_build_skb(ring) &&
1938 !ring_uses_large_buffer(ring))
1939 rxdctl |= IXGBEVF_MAX_FRAME_BUILD_SKB |
1940 IXGBE_RXDCTL_RLPML_EN;
1944 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1945 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1947 ixgbevf_rx_desc_queue_enable(adapter, ring);
1948 ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1951 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter,
1952 struct ixgbevf_ring *rx_ring)
1954 struct net_device *netdev = adapter->netdev;
1955 unsigned int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1957 /* set build_skb and buffer size flags */
1958 clear_ring_build_skb_enabled(rx_ring);
1959 clear_ring_uses_large_buffer(rx_ring);
1961 if (adapter->flags & IXGBEVF_FLAGS_LEGACY_RX)
1964 set_ring_build_skb_enabled(rx_ring);
1966 if (PAGE_SIZE < 8192) {
1967 if (max_frame <= IXGBEVF_MAX_FRAME_BUILD_SKB)
1970 set_ring_uses_large_buffer(rx_ring);
1975 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1976 * @adapter: board private structure
1978 * Configure the Rx unit of the MAC after a reset.
1980 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1982 struct ixgbe_hw *hw = &adapter->hw;
1983 struct net_device *netdev = adapter->netdev;
1986 ixgbevf_setup_psrtype(adapter);
1987 if (hw->mac.type >= ixgbe_mac_X550_vf)
1988 ixgbevf_setup_vfmrqc(adapter);
1990 spin_lock_bh(&adapter->mbx_lock);
1991 /* notify the PF of our intent to use this size of frame */
1992 ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1993 spin_unlock_bh(&adapter->mbx_lock);
1995 dev_err(&adapter->pdev->dev,
1996 "Failed to set MTU at %d\n", netdev->mtu);
1998 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1999 * the Base and Length of the Rx Descriptor Ring
2001 for (i = 0; i < adapter->num_rx_queues; i++) {
2002 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
2004 ixgbevf_set_rx_buffer_len(adapter, rx_ring);
2005 ixgbevf_configure_rx_ring(adapter, rx_ring);
2009 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
2010 __be16 proto, u16 vid)
2012 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2013 struct ixgbe_hw *hw = &adapter->hw;
2016 spin_lock_bh(&adapter->mbx_lock);
2018 /* add VID to filter table */
2019 err = hw->mac.ops.set_vfta(hw, vid, 0, true);
2021 spin_unlock_bh(&adapter->mbx_lock);
2023 /* translate error return types so error makes sense */
2024 if (err == IXGBE_ERR_MBX)
2027 if (err == IXGBE_ERR_INVALID_ARGUMENT)
2030 set_bit(vid, adapter->active_vlans);
2035 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
2036 __be16 proto, u16 vid)
2038 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2039 struct ixgbe_hw *hw = &adapter->hw;
2042 spin_lock_bh(&adapter->mbx_lock);
2044 /* remove VID from filter table */
2045 err = hw->mac.ops.set_vfta(hw, vid, 0, false);
2047 spin_unlock_bh(&adapter->mbx_lock);
2049 clear_bit(vid, adapter->active_vlans);
2054 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
2058 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2059 ixgbevf_vlan_rx_add_vid(adapter->netdev,
2060 htons(ETH_P_8021Q), vid);
2063 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
2065 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2066 struct ixgbe_hw *hw = &adapter->hw;
2069 if ((netdev_uc_count(netdev)) > 10) {
2070 pr_err("Too many unicast filters - No Space\n");
2074 if (!netdev_uc_empty(netdev)) {
2075 struct netdev_hw_addr *ha;
2077 netdev_for_each_uc_addr(ha, netdev) {
2078 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
2082 /* If the list is empty then send message to PF driver to
2083 * clear all MAC VLANs on this VF.
2085 hw->mac.ops.set_uc_addr(hw, 0, NULL);
2092 * ixgbevf_set_rx_mode - Multicast and unicast set
2093 * @netdev: network interface device structure
2095 * The set_rx_method entry point is called whenever the multicast address
2096 * list, unicast address list or the network interface flags are updated.
2097 * This routine is responsible for configuring the hardware for proper
2098 * multicast mode and configuring requested unicast filters.
2100 static void ixgbevf_set_rx_mode(struct net_device *netdev)
2102 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2103 struct ixgbe_hw *hw = &adapter->hw;
2104 unsigned int flags = netdev->flags;
2107 /* request the most inclusive mode we need */
2108 if (flags & IFF_PROMISC)
2109 xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
2110 else if (flags & IFF_ALLMULTI)
2111 xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
2112 else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
2113 xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
2115 xcast_mode = IXGBEVF_XCAST_MODE_NONE;
2117 spin_lock_bh(&adapter->mbx_lock);
2119 hw->mac.ops.update_xcast_mode(hw, xcast_mode);
2121 /* reprogram multicast list */
2122 hw->mac.ops.update_mc_addr_list(hw, netdev);
2124 ixgbevf_write_uc_addr_list(netdev);
2126 spin_unlock_bh(&adapter->mbx_lock);
2129 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
2132 struct ixgbevf_q_vector *q_vector;
2133 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2135 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2136 q_vector = adapter->q_vector[q_idx];
2137 napi_enable(&q_vector->napi);
2141 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
2144 struct ixgbevf_q_vector *q_vector;
2145 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2147 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2148 q_vector = adapter->q_vector[q_idx];
2149 napi_disable(&q_vector->napi);
2153 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
2155 struct ixgbe_hw *hw = &adapter->hw;
2156 unsigned int def_q = 0;
2157 unsigned int num_tcs = 0;
2158 unsigned int num_rx_queues = adapter->num_rx_queues;
2159 unsigned int num_tx_queues = adapter->num_tx_queues;
2162 spin_lock_bh(&adapter->mbx_lock);
2164 /* fetch queue configuration from the PF */
2165 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2167 spin_unlock_bh(&adapter->mbx_lock);
2173 /* we need only one Tx queue */
2176 /* update default Tx ring register index */
2177 adapter->tx_ring[0]->reg_idx = def_q;
2179 /* we need as many queues as traffic classes */
2180 num_rx_queues = num_tcs;
2183 /* if we have a bad config abort request queue reset */
2184 if ((adapter->num_rx_queues != num_rx_queues) ||
2185 (adapter->num_tx_queues != num_tx_queues)) {
2186 /* force mailbox timeout to prevent further messages */
2187 hw->mbx.timeout = 0;
2189 /* wait for watchdog to come around and bail us out */
2190 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2196 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2198 ixgbevf_configure_dcb(adapter);
2200 ixgbevf_set_rx_mode(adapter->netdev);
2202 ixgbevf_restore_vlan(adapter);
2204 ixgbevf_configure_tx(adapter);
2205 ixgbevf_configure_rx(adapter);
2208 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2210 /* Only save pre-reset stats if there are some */
2211 if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2212 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2213 adapter->stats.base_vfgprc;
2214 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2215 adapter->stats.base_vfgptc;
2216 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2217 adapter->stats.base_vfgorc;
2218 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2219 adapter->stats.base_vfgotc;
2220 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2221 adapter->stats.base_vfmprc;
2225 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2227 struct ixgbe_hw *hw = &adapter->hw;
2229 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2230 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2231 adapter->stats.last_vfgorc |=
2232 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2233 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2234 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2235 adapter->stats.last_vfgotc |=
2236 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2237 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2239 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2240 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2241 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2242 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2243 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2246 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2248 struct ixgbe_hw *hw = &adapter->hw;
2249 int api[] = { ixgbe_mbox_api_13,
2253 ixgbe_mbox_api_unknown };
2256 spin_lock_bh(&adapter->mbx_lock);
2258 while (api[idx] != ixgbe_mbox_api_unknown) {
2259 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2265 spin_unlock_bh(&adapter->mbx_lock);
2268 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2270 struct net_device *netdev = adapter->netdev;
2271 struct ixgbe_hw *hw = &adapter->hw;
2273 ixgbevf_configure_msix(adapter);
2275 spin_lock_bh(&adapter->mbx_lock);
2277 if (is_valid_ether_addr(hw->mac.addr))
2278 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2280 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2282 spin_unlock_bh(&adapter->mbx_lock);
2284 smp_mb__before_atomic();
2285 clear_bit(__IXGBEVF_DOWN, &adapter->state);
2286 ixgbevf_napi_enable_all(adapter);
2288 /* clear any pending interrupts, may auto mask */
2289 IXGBE_READ_REG(hw, IXGBE_VTEICR);
2290 ixgbevf_irq_enable(adapter);
2292 /* enable transmits */
2293 netif_tx_start_all_queues(netdev);
2295 ixgbevf_save_reset_stats(adapter);
2296 ixgbevf_init_last_counter_stats(adapter);
2298 hw->mac.get_link_status = 1;
2299 mod_timer(&adapter->service_timer, jiffies);
2302 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2304 ixgbevf_configure(adapter);
2306 ixgbevf_up_complete(adapter);
2310 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2311 * @rx_ring: ring to free buffers from
2313 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2315 u16 i = rx_ring->next_to_clean;
2317 /* Free Rx ring sk_buff */
2319 dev_kfree_skb(rx_ring->skb);
2320 rx_ring->skb = NULL;
2323 /* Free all the Rx ring pages */
2324 while (i != rx_ring->next_to_alloc) {
2325 struct ixgbevf_rx_buffer *rx_buffer;
2327 rx_buffer = &rx_ring->rx_buffer_info[i];
2329 /* Invalidate cache lines that may have been written to by
2330 * device so that we avoid corrupting memory.
2332 dma_sync_single_range_for_cpu(rx_ring->dev,
2334 rx_buffer->page_offset,
2335 ixgbevf_rx_bufsz(rx_ring),
2338 /* free resources associated with mapping */
2339 dma_unmap_page_attrs(rx_ring->dev,
2341 ixgbevf_rx_pg_size(rx_ring),
2343 IXGBEVF_RX_DMA_ATTR);
2345 __page_frag_cache_drain(rx_buffer->page,
2346 rx_buffer->pagecnt_bias);
2349 if (i == rx_ring->count)
2353 rx_ring->next_to_alloc = 0;
2354 rx_ring->next_to_clean = 0;
2355 rx_ring->next_to_use = 0;
2359 * ixgbevf_clean_tx_ring - Free Tx Buffers
2360 * @tx_ring: ring to be cleaned
2362 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2364 u16 i = tx_ring->next_to_clean;
2365 struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
2367 while (i != tx_ring->next_to_use) {
2368 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
2370 /* Free all the Tx ring sk_buffs */
2371 if (ring_is_xdp(tx_ring))
2372 page_frag_free(tx_buffer->data);
2374 dev_kfree_skb_any(tx_buffer->skb);
2376 /* unmap skb header data */
2377 dma_unmap_single(tx_ring->dev,
2378 dma_unmap_addr(tx_buffer, dma),
2379 dma_unmap_len(tx_buffer, len),
2382 /* check for eop_desc to determine the end of the packet */
2383 eop_desc = tx_buffer->next_to_watch;
2384 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2386 /* unmap remaining buffers */
2387 while (tx_desc != eop_desc) {
2391 if (unlikely(i == tx_ring->count)) {
2393 tx_buffer = tx_ring->tx_buffer_info;
2394 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
2397 /* unmap any remaining paged data */
2398 if (dma_unmap_len(tx_buffer, len))
2399 dma_unmap_page(tx_ring->dev,
2400 dma_unmap_addr(tx_buffer, dma),
2401 dma_unmap_len(tx_buffer, len),
2405 /* move us one more past the eop_desc for start of next pkt */
2408 if (unlikely(i == tx_ring->count)) {
2410 tx_buffer = tx_ring->tx_buffer_info;
2414 /* reset next_to_use and next_to_clean */
2415 tx_ring->next_to_use = 0;
2416 tx_ring->next_to_clean = 0;
2421 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2422 * @adapter: board private structure
2424 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2428 for (i = 0; i < adapter->num_rx_queues; i++)
2429 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2433 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2434 * @adapter: board private structure
2436 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2440 for (i = 0; i < adapter->num_tx_queues; i++)
2441 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2442 for (i = 0; i < adapter->num_xdp_queues; i++)
2443 ixgbevf_clean_tx_ring(adapter->xdp_ring[i]);
2446 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2448 struct net_device *netdev = adapter->netdev;
2449 struct ixgbe_hw *hw = &adapter->hw;
2452 /* signal that we are down to the interrupt handler */
2453 if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2454 return; /* do nothing if already down */
2456 /* disable all enabled Rx queues */
2457 for (i = 0; i < adapter->num_rx_queues; i++)
2458 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2460 usleep_range(10000, 20000);
2462 netif_tx_stop_all_queues(netdev);
2464 /* call carrier off first to avoid false dev_watchdog timeouts */
2465 netif_carrier_off(netdev);
2466 netif_tx_disable(netdev);
2468 ixgbevf_irq_disable(adapter);
2470 ixgbevf_napi_disable_all(adapter);
2472 del_timer_sync(&adapter->service_timer);
2474 /* disable transmits in the hardware now that interrupts are off */
2475 for (i = 0; i < adapter->num_tx_queues; i++) {
2476 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2478 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2479 IXGBE_TXDCTL_SWFLSH);
2482 for (i = 0; i < adapter->num_xdp_queues; i++) {
2483 u8 reg_idx = adapter->xdp_ring[i]->reg_idx;
2485 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2486 IXGBE_TXDCTL_SWFLSH);
2489 if (!pci_channel_offline(adapter->pdev))
2490 ixgbevf_reset(adapter);
2492 ixgbevf_clean_all_tx_rings(adapter);
2493 ixgbevf_clean_all_rx_rings(adapter);
2496 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2498 WARN_ON(in_interrupt());
2500 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2503 ixgbevf_down(adapter);
2504 ixgbevf_up(adapter);
2506 clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2509 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2511 struct ixgbe_hw *hw = &adapter->hw;
2512 struct net_device *netdev = adapter->netdev;
2514 if (hw->mac.ops.reset_hw(hw)) {
2515 hw_dbg(hw, "PF still resetting\n");
2517 hw->mac.ops.init_hw(hw);
2518 ixgbevf_negotiate_api(adapter);
2521 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2522 ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2523 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2526 adapter->last_reset = jiffies;
2529 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2532 int vector_threshold;
2534 /* We'll want at least 2 (vector_threshold):
2535 * 1) TxQ[0] + RxQ[0] handler
2536 * 2) Other (Link Status Change, etc.)
2538 vector_threshold = MIN_MSIX_COUNT;
2540 /* The more we get, the more we will assign to Tx/Rx Cleanup
2541 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2542 * Right now, we simply care about how many we'll get; we'll
2543 * set them up later while requesting irq's.
2545 vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2546 vector_threshold, vectors);
2549 dev_err(&adapter->pdev->dev,
2550 "Unable to allocate MSI-X interrupts\n");
2551 kfree(adapter->msix_entries);
2552 adapter->msix_entries = NULL;
2556 /* Adjust for only the vectors we'll use, which is minimum
2557 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2558 * vectors we were allocated.
2560 adapter->num_msix_vectors = vectors;
2566 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2567 * @adapter: board private structure to initialize
2569 * This is the top level queue allocation routine. The order here is very
2570 * important, starting with the "most" number of features turned on at once,
2571 * and ending with the smallest set of features. This way large combinations
2572 * can be allocated if they're turned on, and smaller combinations are the
2573 * fallthrough conditions.
2576 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2578 struct ixgbe_hw *hw = &adapter->hw;
2579 unsigned int def_q = 0;
2580 unsigned int num_tcs = 0;
2583 /* Start with base case */
2584 adapter->num_rx_queues = 1;
2585 adapter->num_tx_queues = 1;
2586 adapter->num_xdp_queues = 0;
2588 spin_lock_bh(&adapter->mbx_lock);
2590 /* fetch queue configuration from the PF */
2591 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2593 spin_unlock_bh(&adapter->mbx_lock);
2598 /* we need as many queues as traffic classes */
2600 adapter->num_rx_queues = num_tcs;
2602 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2604 switch (hw->api_version) {
2605 case ixgbe_mbox_api_11:
2606 case ixgbe_mbox_api_12:
2607 case ixgbe_mbox_api_13:
2608 if (adapter->xdp_prog &&
2609 hw->mac.max_tx_queues == rss)
2610 rss = rss > 3 ? 2 : 1;
2612 adapter->num_rx_queues = rss;
2613 adapter->num_tx_queues = rss;
2614 adapter->num_xdp_queues = adapter->xdp_prog ? rss : 0;
2622 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2623 * @adapter: board private structure to initialize
2625 * Attempt to configure the interrupts using the best available
2626 * capabilities of the hardware and the kernel.
2628 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2630 int vector, v_budget;
2632 /* It's easy to be greedy for MSI-X vectors, but it really
2633 * doesn't do us much good if we have a lot more vectors
2634 * than CPU's. So let's be conservative and only ask for
2635 * (roughly) the same number of vectors as there are CPU's.
2636 * The default is to use pairs of vectors.
2638 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2639 v_budget = min_t(int, v_budget, num_online_cpus());
2640 v_budget += NON_Q_VECTORS;
2642 adapter->msix_entries = kcalloc(v_budget,
2643 sizeof(struct msix_entry), GFP_KERNEL);
2644 if (!adapter->msix_entries)
2647 for (vector = 0; vector < v_budget; vector++)
2648 adapter->msix_entries[vector].entry = vector;
2650 /* A failure in MSI-X entry allocation isn't fatal, but the VF driver
2651 * does not support any other modes, so we will simply fail here. Note
2652 * that we clean up the msix_entries pointer else-where.
2654 return ixgbevf_acquire_msix_vectors(adapter, v_budget);
2657 static void ixgbevf_add_ring(struct ixgbevf_ring *ring,
2658 struct ixgbevf_ring_container *head)
2660 ring->next = head->ring;
2666 * ixgbevf_alloc_q_vector - Allocate memory for a single interrupt vector
2667 * @adapter: board private structure to initialize
2668 * @v_idx: index of vector in adapter struct
2669 * @txr_count: number of Tx rings for q vector
2670 * @txr_idx: index of first Tx ring to assign
2671 * @xdp_count: total number of XDP rings to allocate
2672 * @xdp_idx: index of first XDP ring to allocate
2673 * @rxr_count: number of Rx rings for q vector
2674 * @rxr_idx: index of first Rx ring to assign
2676 * We allocate one q_vector. If allocation fails we return -ENOMEM.
2678 static int ixgbevf_alloc_q_vector(struct ixgbevf_adapter *adapter, int v_idx,
2679 int txr_count, int txr_idx,
2680 int xdp_count, int xdp_idx,
2681 int rxr_count, int rxr_idx)
2683 struct ixgbevf_q_vector *q_vector;
2684 int reg_idx = txr_idx + xdp_idx;
2685 struct ixgbevf_ring *ring;
2686 int ring_count, size;
2688 ring_count = txr_count + xdp_count + rxr_count;
2689 size = sizeof(*q_vector) + (sizeof(*ring) * ring_count);
2691 /* allocate q_vector and rings */
2692 q_vector = kzalloc(size, GFP_KERNEL);
2696 /* initialize NAPI */
2697 netif_napi_add(adapter->netdev, &q_vector->napi, ixgbevf_poll, 64);
2699 /* tie q_vector and adapter together */
2700 adapter->q_vector[v_idx] = q_vector;
2701 q_vector->adapter = adapter;
2702 q_vector->v_idx = v_idx;
2704 /* initialize pointer to rings */
2705 ring = q_vector->ring;
2708 /* assign generic ring traits */
2709 ring->dev = &adapter->pdev->dev;
2710 ring->netdev = adapter->netdev;
2712 /* configure backlink on ring */
2713 ring->q_vector = q_vector;
2715 /* update q_vector Tx values */
2716 ixgbevf_add_ring(ring, &q_vector->tx);
2718 /* apply Tx specific ring traits */
2719 ring->count = adapter->tx_ring_count;
2720 ring->queue_index = txr_idx;
2721 ring->reg_idx = reg_idx;
2723 /* assign ring to adapter */
2724 adapter->tx_ring[txr_idx] = ring;
2726 /* update count and index */
2731 /* push pointer to next ring */
2736 /* assign generic ring traits */
2737 ring->dev = &adapter->pdev->dev;
2738 ring->netdev = adapter->netdev;
2740 /* configure backlink on ring */
2741 ring->q_vector = q_vector;
2743 /* update q_vector Tx values */
2744 ixgbevf_add_ring(ring, &q_vector->tx);
2746 /* apply Tx specific ring traits */
2747 ring->count = adapter->tx_ring_count;
2748 ring->queue_index = xdp_idx;
2749 ring->reg_idx = reg_idx;
2752 /* assign ring to adapter */
2753 adapter->xdp_ring[xdp_idx] = ring;
2755 /* update count and index */
2760 /* push pointer to next ring */
2765 /* assign generic ring traits */
2766 ring->dev = &adapter->pdev->dev;
2767 ring->netdev = adapter->netdev;
2769 /* configure backlink on ring */
2770 ring->q_vector = q_vector;
2772 /* update q_vector Rx values */
2773 ixgbevf_add_ring(ring, &q_vector->rx);
2775 /* apply Rx specific ring traits */
2776 ring->count = adapter->rx_ring_count;
2777 ring->queue_index = rxr_idx;
2778 ring->reg_idx = rxr_idx;
2780 /* assign ring to adapter */
2781 adapter->rx_ring[rxr_idx] = ring;
2783 /* update count and index */
2787 /* push pointer to next ring */
2795 * ixgbevf_free_q_vector - Free memory allocated for specific interrupt vector
2796 * @adapter: board private structure to initialize
2797 * @v_idx: index of vector in adapter struct
2799 * This function frees the memory allocated to the q_vector. In addition if
2800 * NAPI is enabled it will delete any references to the NAPI struct prior
2801 * to freeing the q_vector.
2803 static void ixgbevf_free_q_vector(struct ixgbevf_adapter *adapter, int v_idx)
2805 struct ixgbevf_q_vector *q_vector = adapter->q_vector[v_idx];
2806 struct ixgbevf_ring *ring;
2808 ixgbevf_for_each_ring(ring, q_vector->tx) {
2809 if (ring_is_xdp(ring))
2810 adapter->xdp_ring[ring->queue_index] = NULL;
2812 adapter->tx_ring[ring->queue_index] = NULL;
2815 ixgbevf_for_each_ring(ring, q_vector->rx)
2816 adapter->rx_ring[ring->queue_index] = NULL;
2818 adapter->q_vector[v_idx] = NULL;
2819 netif_napi_del(&q_vector->napi);
2821 /* ixgbevf_get_stats() might access the rings on this vector,
2822 * we must wait a grace period before freeing it.
2824 kfree_rcu(q_vector, rcu);
2828 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2829 * @adapter: board private structure to initialize
2831 * We allocate one q_vector per queue interrupt. If allocation fails we
2834 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2836 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2837 int rxr_remaining = adapter->num_rx_queues;
2838 int txr_remaining = adapter->num_tx_queues;
2839 int xdp_remaining = adapter->num_xdp_queues;
2840 int rxr_idx = 0, txr_idx = 0, xdp_idx = 0, v_idx = 0;
2843 if (q_vectors >= (rxr_remaining + txr_remaining + xdp_remaining)) {
2844 for (; rxr_remaining; v_idx++, q_vectors--) {
2845 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2847 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2848 0, 0, 0, 0, rqpv, rxr_idx);
2852 /* update counts and index */
2853 rxr_remaining -= rqpv;
2858 for (; q_vectors; v_idx++, q_vectors--) {
2859 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2860 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors);
2861 int xqpv = DIV_ROUND_UP(xdp_remaining, q_vectors);
2863 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2871 /* update counts and index */
2872 rxr_remaining -= rqpv;
2874 txr_remaining -= tqpv;
2876 xdp_remaining -= xqpv;
2885 ixgbevf_free_q_vector(adapter, v_idx);
2892 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2893 * @adapter: board private structure to initialize
2895 * This function frees the memory allocated to the q_vectors. In addition if
2896 * NAPI is enabled it will delete any references to the NAPI struct prior
2897 * to freeing the q_vector.
2899 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2901 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2905 ixgbevf_free_q_vector(adapter, q_vectors);
2910 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2911 * @adapter: board private structure
2914 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2916 if (!adapter->msix_entries)
2919 pci_disable_msix(adapter->pdev);
2920 kfree(adapter->msix_entries);
2921 adapter->msix_entries = NULL;
2925 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2926 * @adapter: board private structure to initialize
2929 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2933 /* Number of supported queues */
2934 ixgbevf_set_num_queues(adapter);
2936 err = ixgbevf_set_interrupt_capability(adapter);
2938 hw_dbg(&adapter->hw,
2939 "Unable to setup interrupt capabilities\n");
2940 goto err_set_interrupt;
2943 err = ixgbevf_alloc_q_vectors(adapter);
2945 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2946 goto err_alloc_q_vectors;
2949 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u XDP Queue count %u\n",
2950 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
2951 adapter->num_rx_queues, adapter->num_tx_queues,
2952 adapter->num_xdp_queues);
2954 set_bit(__IXGBEVF_DOWN, &adapter->state);
2957 err_alloc_q_vectors:
2958 ixgbevf_reset_interrupt_capability(adapter);
2964 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2965 * @adapter: board private structure to clear interrupt scheme on
2967 * We go through and clear interrupt specific resources and reset the structure
2968 * to pre-load conditions
2970 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2972 adapter->num_tx_queues = 0;
2973 adapter->num_xdp_queues = 0;
2974 adapter->num_rx_queues = 0;
2976 ixgbevf_free_q_vectors(adapter);
2977 ixgbevf_reset_interrupt_capability(adapter);
2981 * ixgbevf_sw_init - Initialize general software structures
2982 * @adapter: board private structure to initialize
2984 * ixgbevf_sw_init initializes the Adapter private data structure.
2985 * Fields are initialized based on PCI device information and
2986 * OS network device settings (MTU size).
2988 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2990 struct ixgbe_hw *hw = &adapter->hw;
2991 struct pci_dev *pdev = adapter->pdev;
2992 struct net_device *netdev = adapter->netdev;
2995 /* PCI config space info */
2996 hw->vendor_id = pdev->vendor;
2997 hw->device_id = pdev->device;
2998 hw->revision_id = pdev->revision;
2999 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3000 hw->subsystem_device_id = pdev->subsystem_device;
3002 hw->mbx.ops.init_params(hw);
3004 if (hw->mac.type >= ixgbe_mac_X550_vf) {
3005 err = ixgbevf_init_rss_key(adapter);
3010 /* assume legacy case in which PF would only give VF 2 queues */
3011 hw->mac.max_tx_queues = 2;
3012 hw->mac.max_rx_queues = 2;
3014 /* lock to protect mailbox accesses */
3015 spin_lock_init(&adapter->mbx_lock);
3017 err = hw->mac.ops.reset_hw(hw);
3019 dev_info(&pdev->dev,
3020 "PF still in reset state. Is the PF interface up?\n");
3022 err = hw->mac.ops.init_hw(hw);
3024 pr_err("init_shared_code failed: %d\n", err);
3027 ixgbevf_negotiate_api(adapter);
3028 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
3030 dev_info(&pdev->dev, "Error reading MAC address\n");
3031 else if (is_zero_ether_addr(adapter->hw.mac.addr))
3032 dev_info(&pdev->dev,
3033 "MAC address not assigned by administrator.\n");
3034 ether_addr_copy(netdev->dev_addr, hw->mac.addr);
3037 if (!is_valid_ether_addr(netdev->dev_addr)) {
3038 dev_info(&pdev->dev, "Assigning random MAC address\n");
3039 eth_hw_addr_random(netdev);
3040 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
3041 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
3044 /* Enable dynamic interrupt throttling rates */
3045 adapter->rx_itr_setting = 1;
3046 adapter->tx_itr_setting = 1;
3048 /* set default ring sizes */
3049 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
3050 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
3052 set_bit(__IXGBEVF_DOWN, &adapter->state);
3059 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
3061 u32 current_counter = IXGBE_READ_REG(hw, reg); \
3062 if (current_counter < last_counter) \
3063 counter += 0x100000000LL; \
3064 last_counter = current_counter; \
3065 counter &= 0xFFFFFFFF00000000LL; \
3066 counter |= current_counter; \
3069 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
3071 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
3072 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
3073 u64 current_counter = (current_counter_msb << 32) | \
3074 current_counter_lsb; \
3075 if (current_counter < last_counter) \
3076 counter += 0x1000000000LL; \
3077 last_counter = current_counter; \
3078 counter &= 0xFFFFFFF000000000LL; \
3079 counter |= current_counter; \
3082 * ixgbevf_update_stats - Update the board statistics counters.
3083 * @adapter: board private structure
3085 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
3087 struct ixgbe_hw *hw = &adapter->hw;
3088 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
3089 u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
3092 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3093 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3096 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
3097 adapter->stats.vfgprc);
3098 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
3099 adapter->stats.vfgptc);
3100 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
3101 adapter->stats.last_vfgorc,
3102 adapter->stats.vfgorc);
3103 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
3104 adapter->stats.last_vfgotc,
3105 adapter->stats.vfgotc);
3106 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
3107 adapter->stats.vfmprc);
3109 for (i = 0; i < adapter->num_rx_queues; i++) {
3110 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
3112 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
3113 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
3114 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
3115 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
3118 adapter->hw_csum_rx_error = hw_csum_rx_error;
3119 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
3120 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
3121 adapter->alloc_rx_page = alloc_rx_page;
3125 * ixgbevf_service_timer - Timer Call-back
3126 * @t: pointer to timer_list struct
3128 static void ixgbevf_service_timer(struct timer_list *t)
3130 struct ixgbevf_adapter *adapter = from_timer(adapter, t,
3133 /* Reset the timer */
3134 mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
3136 ixgbevf_service_event_schedule(adapter);
3139 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
3141 if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
3145 /* If we're already down or resetting, just bail */
3146 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3147 test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
3148 test_bit(__IXGBEVF_RESETTING, &adapter->state)) {
3153 adapter->tx_timeout_count++;
3155 ixgbevf_reinit_locked(adapter);
3160 * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
3161 * @adapter: pointer to the device adapter structure
3163 * This function serves two purposes. First it strobes the interrupt lines
3164 * in order to make certain interrupts are occurring. Secondly it sets the
3165 * bits needed to check for TX hangs. As a result we should immediately
3166 * determine if a hang has occurred.
3168 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
3170 struct ixgbe_hw *hw = &adapter->hw;
3174 /* If we're down or resetting, just bail */
3175 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3176 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3179 /* Force detection of hung controller */
3180 if (netif_carrier_ok(adapter->netdev)) {
3181 for (i = 0; i < adapter->num_tx_queues; i++)
3182 set_check_for_tx_hang(adapter->tx_ring[i]);
3183 for (i = 0; i < adapter->num_xdp_queues; i++)
3184 set_check_for_tx_hang(adapter->xdp_ring[i]);
3187 /* get one bit for every active Tx/Rx interrupt vector */
3188 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
3189 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
3191 if (qv->rx.ring || qv->tx.ring)
3195 /* Cause software interrupt to ensure rings are cleaned */
3196 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
3200 * ixgbevf_watchdog_update_link - update the link status
3201 * @adapter: pointer to the device adapter structure
3203 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
3205 struct ixgbe_hw *hw = &adapter->hw;
3206 u32 link_speed = adapter->link_speed;
3207 bool link_up = adapter->link_up;
3210 spin_lock_bh(&adapter->mbx_lock);
3212 err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3214 spin_unlock_bh(&adapter->mbx_lock);
3216 /* if check for link returns error we will need to reset */
3217 if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
3218 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
3222 adapter->link_up = link_up;
3223 adapter->link_speed = link_speed;
3227 * ixgbevf_watchdog_link_is_up - update netif_carrier status and
3228 * print link up message
3229 * @adapter: pointer to the device adapter structure
3231 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
3233 struct net_device *netdev = adapter->netdev;
3235 /* only continue if link was previously down */
3236 if (netif_carrier_ok(netdev))
3239 dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
3240 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
3242 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
3244 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
3248 netif_carrier_on(netdev);
3252 * ixgbevf_watchdog_link_is_down - update netif_carrier status and
3253 * print link down message
3254 * @adapter: pointer to the adapter structure
3256 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
3258 struct net_device *netdev = adapter->netdev;
3260 adapter->link_speed = 0;
3262 /* only continue if link was up previously */
3263 if (!netif_carrier_ok(netdev))
3266 dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
3268 netif_carrier_off(netdev);
3272 * ixgbevf_watchdog_subtask - worker thread to bring link up
3273 * @adapter: board private structure
3275 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
3277 /* if interface is down do nothing */
3278 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3279 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3282 ixgbevf_watchdog_update_link(adapter);
3284 if (adapter->link_up)
3285 ixgbevf_watchdog_link_is_up(adapter);
3287 ixgbevf_watchdog_link_is_down(adapter);
3289 ixgbevf_update_stats(adapter);
3293 * ixgbevf_service_task - manages and runs subtasks
3294 * @work: pointer to work_struct containing our data
3296 static void ixgbevf_service_task(struct work_struct *work)
3298 struct ixgbevf_adapter *adapter = container_of(work,
3299 struct ixgbevf_adapter,
3301 struct ixgbe_hw *hw = &adapter->hw;
3303 if (IXGBE_REMOVED(hw->hw_addr)) {
3304 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
3306 ixgbevf_down(adapter);
3312 ixgbevf_queue_reset_subtask(adapter);
3313 ixgbevf_reset_subtask(adapter);
3314 ixgbevf_watchdog_subtask(adapter);
3315 ixgbevf_check_hang_subtask(adapter);
3317 ixgbevf_service_event_complete(adapter);
3321 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
3322 * @tx_ring: Tx descriptor ring for a specific queue
3324 * Free all transmit software resources
3326 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
3328 ixgbevf_clean_tx_ring(tx_ring);
3330 vfree(tx_ring->tx_buffer_info);
3331 tx_ring->tx_buffer_info = NULL;
3333 /* if not set, then don't free */
3337 dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
3340 tx_ring->desc = NULL;
3344 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
3345 * @adapter: board private structure
3347 * Free all transmit software resources
3349 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
3353 for (i = 0; i < adapter->num_tx_queues; i++)
3354 if (adapter->tx_ring[i]->desc)
3355 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3356 for (i = 0; i < adapter->num_xdp_queues; i++)
3357 if (adapter->xdp_ring[i]->desc)
3358 ixgbevf_free_tx_resources(adapter->xdp_ring[i]);
3362 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
3363 * @tx_ring: Tx descriptor ring (for a specific queue) to setup
3365 * Return 0 on success, negative on failure
3367 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
3369 struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3372 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3373 tx_ring->tx_buffer_info = vmalloc(size);
3374 if (!tx_ring->tx_buffer_info)
3377 u64_stats_init(&tx_ring->syncp);
3379 /* round up to nearest 4K */
3380 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3381 tx_ring->size = ALIGN(tx_ring->size, 4096);
3383 tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3384 &tx_ring->dma, GFP_KERNEL);
3391 vfree(tx_ring->tx_buffer_info);
3392 tx_ring->tx_buffer_info = NULL;
3393 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3398 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3399 * @adapter: board private structure
3401 * If this function returns with an error, then it's possible one or
3402 * more of the rings is populated (while the rest are not). It is the
3403 * callers duty to clean those orphaned rings.
3405 * Return 0 on success, negative on failure
3407 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3409 int i, j = 0, err = 0;
3411 for (i = 0; i < adapter->num_tx_queues; i++) {
3412 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3415 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3419 for (j = 0; j < adapter->num_xdp_queues; j++) {
3420 err = ixgbevf_setup_tx_resources(adapter->xdp_ring[j]);
3423 hw_dbg(&adapter->hw, "Allocation for XDP Queue %u failed\n", j);
3429 /* rewind the index freeing the rings as we go */
3431 ixgbevf_free_tx_resources(adapter->xdp_ring[j]);
3433 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3439 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3440 * @adapter: board private structure
3441 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3443 * Returns 0 on success, negative on failure
3445 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
3446 struct ixgbevf_ring *rx_ring)
3450 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3451 rx_ring->rx_buffer_info = vmalloc(size);
3452 if (!rx_ring->rx_buffer_info)
3455 u64_stats_init(&rx_ring->syncp);
3457 /* Round up to nearest 4K */
3458 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3459 rx_ring->size = ALIGN(rx_ring->size, 4096);
3461 rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3462 &rx_ring->dma, GFP_KERNEL);
3467 /* XDP RX-queue info */
3468 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
3469 rx_ring->queue_index) < 0)
3472 rx_ring->xdp_prog = adapter->xdp_prog;
3476 vfree(rx_ring->rx_buffer_info);
3477 rx_ring->rx_buffer_info = NULL;
3478 dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3483 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3484 * @adapter: board private structure
3486 * If this function returns with an error, then it's possible one or
3487 * more of the rings is populated (while the rest are not). It is the
3488 * callers duty to clean those orphaned rings.
3490 * Return 0 on success, negative on failure
3492 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3496 for (i = 0; i < adapter->num_rx_queues; i++) {
3497 err = ixgbevf_setup_rx_resources(adapter, adapter->rx_ring[i]);
3500 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3506 /* rewind the index freeing the rings as we go */
3508 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3513 * ixgbevf_free_rx_resources - Free Rx Resources
3514 * @rx_ring: ring to clean the resources from
3516 * Free all receive software resources
3518 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3520 ixgbevf_clean_rx_ring(rx_ring);
3522 rx_ring->xdp_prog = NULL;
3523 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
3524 vfree(rx_ring->rx_buffer_info);
3525 rx_ring->rx_buffer_info = NULL;
3527 dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3530 rx_ring->desc = NULL;
3534 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3535 * @adapter: board private structure
3537 * Free all receive software resources
3539 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3543 for (i = 0; i < adapter->num_rx_queues; i++)
3544 if (adapter->rx_ring[i]->desc)
3545 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3549 * ixgbevf_open - Called when a network interface is made active
3550 * @netdev: network interface device structure
3552 * Returns 0 on success, negative value on failure
3554 * The open entry point is called when a network interface is made
3555 * active by the system (IFF_UP). At this point all resources needed
3556 * for transmit and receive operations are allocated, the interrupt
3557 * handler is registered with the OS, the watchdog timer is started,
3558 * and the stack is notified that the interface is ready.
3560 int ixgbevf_open(struct net_device *netdev)
3562 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3563 struct ixgbe_hw *hw = &adapter->hw;
3566 /* A previous failure to open the device because of a lack of
3567 * available MSIX vector resources may have reset the number
3568 * of msix vectors variable to zero. The only way to recover
3569 * is to unload/reload the driver and hope that the system has
3570 * been able to recover some MSIX vector resources.
3572 if (!adapter->num_msix_vectors)
3575 if (hw->adapter_stopped) {
3576 ixgbevf_reset(adapter);
3577 /* if adapter is still stopped then PF isn't up and
3578 * the VF can't start.
3580 if (hw->adapter_stopped) {
3581 err = IXGBE_ERR_MBX;
3582 pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3583 goto err_setup_reset;
3587 /* disallow open during test */
3588 if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3591 netif_carrier_off(netdev);
3593 /* allocate transmit descriptors */
3594 err = ixgbevf_setup_all_tx_resources(adapter);
3598 /* allocate receive descriptors */
3599 err = ixgbevf_setup_all_rx_resources(adapter);
3603 ixgbevf_configure(adapter);
3605 err = ixgbevf_request_irq(adapter);
3609 /* Notify the stack of the actual queue counts. */
3610 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
3612 goto err_set_queues;
3614 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
3616 goto err_set_queues;
3618 ixgbevf_up_complete(adapter);
3623 ixgbevf_free_irq(adapter);
3625 ixgbevf_free_all_rx_resources(adapter);
3627 ixgbevf_free_all_tx_resources(adapter);
3629 ixgbevf_reset(adapter);
3636 * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3637 * @adapter: the private adapter struct
3639 * This function should contain the necessary work common to both suspending
3640 * and closing of the device.
3642 static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3644 ixgbevf_down(adapter);
3645 ixgbevf_free_irq(adapter);
3646 ixgbevf_free_all_tx_resources(adapter);
3647 ixgbevf_free_all_rx_resources(adapter);
3651 * ixgbevf_close - Disables a network interface
3652 * @netdev: network interface device structure
3654 * Returns 0, this is not allowed to fail
3656 * The close entry point is called when an interface is de-activated
3657 * by the OS. The hardware is still under the drivers control, but
3658 * needs to be disabled. A global MAC reset is issued to stop the
3659 * hardware, and all transmit and receive resources are freed.
3661 int ixgbevf_close(struct net_device *netdev)
3663 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3665 if (netif_device_present(netdev))
3666 ixgbevf_close_suspend(adapter);
3671 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3673 struct net_device *dev = adapter->netdev;
3675 if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3679 /* if interface is down do nothing */
3680 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3681 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3684 /* Hardware has to reinitialize queues and interrupts to
3685 * match packet buffer alignment. Unfortunately, the
3686 * hardware is not flexible enough to do this dynamically.
3690 if (netif_running(dev))
3693 ixgbevf_clear_interrupt_scheme(adapter);
3694 ixgbevf_init_interrupt_scheme(adapter);
3696 if (netif_running(dev))
3702 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3703 u32 vlan_macip_lens, u32 type_tucmd,
3706 struct ixgbe_adv_tx_context_desc *context_desc;
3707 u16 i = tx_ring->next_to_use;
3709 context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3712 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3714 /* set bits to identify this as an advanced context descriptor */
3715 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3717 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3718 context_desc->seqnum_seed = 0;
3719 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
3720 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3723 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3724 struct ixgbevf_tx_buffer *first,
3727 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3728 struct sk_buff *skb = first->skb;
3738 u32 paylen, l4_offset;
3741 if (skb->ip_summed != CHECKSUM_PARTIAL)
3744 if (!skb_is_gso(skb))
3747 err = skb_cow_head(skb, 0);
3751 if (eth_p_mpls(first->protocol))
3752 ip.hdr = skb_inner_network_header(skb);
3754 ip.hdr = skb_network_header(skb);
3755 l4.hdr = skb_checksum_start(skb);
3757 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3758 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3760 /* initialize outer IP header fields */
3761 if (ip.v4->version == 4) {
3762 unsigned char *csum_start = skb_checksum_start(skb);
3763 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3765 /* IP header will have to cancel out any data that
3766 * is not a part of the outer IP header
3768 ip.v4->check = csum_fold(csum_partial(trans_start,
3769 csum_start - trans_start,
3771 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3774 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3775 IXGBE_TX_FLAGS_CSUM |
3776 IXGBE_TX_FLAGS_IPV4;
3778 ip.v6->payload_len = 0;
3779 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3780 IXGBE_TX_FLAGS_CSUM;
3783 /* determine offset of inner transport header */
3784 l4_offset = l4.hdr - skb->data;
3786 /* compute length of segmentation header */
3787 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3789 /* remove payload length from inner checksum */
3790 paylen = skb->len - l4_offset;
3791 csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
3793 /* update gso size and bytecount with header size */
3794 first->gso_segs = skb_shinfo(skb)->gso_segs;
3795 first->bytecount += (first->gso_segs - 1) * *hdr_len;
3797 /* mss_l4len_id: use 1 as index for TSO */
3798 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3799 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3800 mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3802 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3803 vlan_macip_lens = l4.hdr - ip.hdr;
3804 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3805 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3807 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3808 type_tucmd, mss_l4len_idx);
3813 static inline bool ixgbevf_ipv6_csum_is_sctp(struct sk_buff *skb)
3815 unsigned int offset = 0;
3817 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
3819 return offset == skb_checksum_start_offset(skb);
3822 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3823 struct ixgbevf_tx_buffer *first)
3825 struct sk_buff *skb = first->skb;
3826 u32 vlan_macip_lens = 0;
3829 if (skb->ip_summed != CHECKSUM_PARTIAL)
3832 switch (skb->csum_offset) {
3833 case offsetof(struct tcphdr, check):
3834 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3836 case offsetof(struct udphdr, check):
3838 case offsetof(struct sctphdr, checksum):
3839 /* validate that this is actually an SCTP request */
3840 if (((first->protocol == htons(ETH_P_IP)) &&
3841 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
3842 ((first->protocol == htons(ETH_P_IPV6)) &&
3843 ixgbevf_ipv6_csum_is_sctp(skb))) {
3844 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3849 skb_checksum_help(skb);
3852 /* update TX checksum flag */
3853 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3854 vlan_macip_lens = skb_checksum_start_offset(skb) -
3855 skb_network_offset(skb);
3857 /* vlan_macip_lens: MACLEN, VLAN tag */
3858 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3859 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3861 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
3864 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3866 /* set type for advanced descriptor with frame checksum insertion */
3867 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3868 IXGBE_ADVTXD_DCMD_IFCS |
3869 IXGBE_ADVTXD_DCMD_DEXT);
3871 /* set HW VLAN bit if VLAN is present */
3872 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3873 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3875 /* set segmentation enable bits for TSO/FSO */
3876 if (tx_flags & IXGBE_TX_FLAGS_TSO)
3877 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3882 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3883 u32 tx_flags, unsigned int paylen)
3885 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3887 /* enable L4 checksum for TSO and TX checksum offload */
3888 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3889 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3891 /* enble IPv4 checksum for TSO */
3892 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3893 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3895 /* use index 1 context for TSO/FSO/FCOE */
3896 if (tx_flags & IXGBE_TX_FLAGS_TSO)
3897 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3899 /* Check Context must be set if Tx switch is enabled, which it
3900 * always is for case where virtual functions are running
3902 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3904 tx_desc->read.olinfo_status = olinfo_status;
3907 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3908 struct ixgbevf_tx_buffer *first,
3911 struct sk_buff *skb = first->skb;
3912 struct ixgbevf_tx_buffer *tx_buffer;
3913 union ixgbe_adv_tx_desc *tx_desc;
3914 struct skb_frag_struct *frag;
3916 unsigned int data_len, size;
3917 u32 tx_flags = first->tx_flags;
3918 __le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3919 u16 i = tx_ring->next_to_use;
3921 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3923 ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
3925 size = skb_headlen(skb);
3926 data_len = skb->data_len;
3928 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3932 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3933 if (dma_mapping_error(tx_ring->dev, dma))
3936 /* record length, and DMA address */
3937 dma_unmap_len_set(tx_buffer, len, size);
3938 dma_unmap_addr_set(tx_buffer, dma, dma);
3940 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3942 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3943 tx_desc->read.cmd_type_len =
3944 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3948 if (i == tx_ring->count) {
3949 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3952 tx_desc->read.olinfo_status = 0;
3954 dma += IXGBE_MAX_DATA_PER_TXD;
3955 size -= IXGBE_MAX_DATA_PER_TXD;
3957 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3960 if (likely(!data_len))
3963 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3967 if (i == tx_ring->count) {
3968 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3971 tx_desc->read.olinfo_status = 0;
3973 size = skb_frag_size(frag);
3976 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3979 tx_buffer = &tx_ring->tx_buffer_info[i];
3982 /* write last descriptor with RS and EOP bits */
3983 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3984 tx_desc->read.cmd_type_len = cmd_type;
3986 /* set the timestamp */
3987 first->time_stamp = jiffies;
3989 /* Force memory writes to complete before letting h/w know there
3990 * are new descriptors to fetch. (Only applicable for weak-ordered
3991 * memory model archs, such as IA-64).
3993 * We also need this memory barrier (wmb) to make certain all of the
3994 * status bits have been updated before next_to_watch is written.
3998 /* set next_to_watch value indicating a packet is present */
3999 first->next_to_watch = tx_desc;
4002 if (i == tx_ring->count)
4005 tx_ring->next_to_use = i;
4007 /* notify HW of packet */
4008 ixgbevf_write_tail(tx_ring, i);
4012 dev_err(tx_ring->dev, "TX DMA map failed\n");
4013 tx_buffer = &tx_ring->tx_buffer_info[i];
4015 /* clear dma mappings for failed tx_buffer_info map */
4016 while (tx_buffer != first) {
4017 if (dma_unmap_len(tx_buffer, len))
4018 dma_unmap_page(tx_ring->dev,
4019 dma_unmap_addr(tx_buffer, dma),
4020 dma_unmap_len(tx_buffer, len),
4022 dma_unmap_len_set(tx_buffer, len, 0);
4025 i += tx_ring->count;
4026 tx_buffer = &tx_ring->tx_buffer_info[i];
4029 if (dma_unmap_len(tx_buffer, len))
4030 dma_unmap_single(tx_ring->dev,
4031 dma_unmap_addr(tx_buffer, dma),
4032 dma_unmap_len(tx_buffer, len),
4034 dma_unmap_len_set(tx_buffer, len, 0);
4036 dev_kfree_skb_any(tx_buffer->skb);
4037 tx_buffer->skb = NULL;
4039 tx_ring->next_to_use = i;
4042 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4044 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
4045 /* Herbert's original patch had:
4046 * smp_mb__after_netif_stop_queue();
4047 * but since that doesn't exist yet, just open code it.
4051 /* We need to check again in a case another CPU has just
4052 * made room available.
4054 if (likely(ixgbevf_desc_unused(tx_ring) < size))
4057 /* A reprieve! - use start_queue because it doesn't call schedule */
4058 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
4059 ++tx_ring->tx_stats.restart_queue;
4064 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4066 if (likely(ixgbevf_desc_unused(tx_ring) >= size))
4068 return __ixgbevf_maybe_stop_tx(tx_ring, size);
4071 static int ixgbevf_xmit_frame_ring(struct sk_buff *skb,
4072 struct ixgbevf_ring *tx_ring)
4074 struct ixgbevf_tx_buffer *first;
4077 u16 count = TXD_USE_COUNT(skb_headlen(skb));
4078 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4082 u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
4084 if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
4085 dev_kfree_skb_any(skb);
4086 return NETDEV_TX_OK;
4089 /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
4090 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
4091 * + 2 desc gap to keep tail from touching head,
4092 * + 1 desc for context descriptor,
4093 * otherwise try next time
4095 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4096 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4097 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4099 count += skb_shinfo(skb)->nr_frags;
4101 if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
4102 tx_ring->tx_stats.tx_busy++;
4103 return NETDEV_TX_BUSY;
4106 /* record the location of the first descriptor for this packet */
4107 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4109 first->bytecount = skb->len;
4110 first->gso_segs = 1;
4112 if (skb_vlan_tag_present(skb)) {
4113 tx_flags |= skb_vlan_tag_get(skb);
4114 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4115 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4118 /* record initial flags and protocol */
4119 first->tx_flags = tx_flags;
4120 first->protocol = vlan_get_protocol(skb);
4122 tso = ixgbevf_tso(tx_ring, first, &hdr_len);
4126 ixgbevf_tx_csum(tx_ring, first);
4128 ixgbevf_tx_map(tx_ring, first, hdr_len);
4130 ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
4132 return NETDEV_TX_OK;
4135 dev_kfree_skb_any(first->skb);
4138 return NETDEV_TX_OK;
4141 static netdev_tx_t ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4143 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4144 struct ixgbevf_ring *tx_ring;
4146 if (skb->len <= 0) {
4147 dev_kfree_skb_any(skb);
4148 return NETDEV_TX_OK;
4151 /* The minimum packet size for olinfo paylen is 17 so pad the skb
4152 * in order to meet this minimum size requirement.
4154 if (skb->len < 17) {
4155 if (skb_padto(skb, 17))
4156 return NETDEV_TX_OK;
4160 tx_ring = adapter->tx_ring[skb->queue_mapping];
4161 return ixgbevf_xmit_frame_ring(skb, tx_ring);
4165 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
4166 * @netdev: network interface device structure
4167 * @p: pointer to an address structure
4169 * Returns 0 on success, negative on failure
4171 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
4173 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4174 struct ixgbe_hw *hw = &adapter->hw;
4175 struct sockaddr *addr = p;
4178 if (!is_valid_ether_addr(addr->sa_data))
4179 return -EADDRNOTAVAIL;
4181 spin_lock_bh(&adapter->mbx_lock);
4183 err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
4185 spin_unlock_bh(&adapter->mbx_lock);
4190 ether_addr_copy(hw->mac.addr, addr->sa_data);
4191 ether_addr_copy(hw->mac.perm_addr, addr->sa_data);
4192 ether_addr_copy(netdev->dev_addr, addr->sa_data);
4198 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
4199 * @netdev: network interface device structure
4200 * @new_mtu: new value for maximum frame size
4202 * Returns 0 on success, negative on failure
4204 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
4206 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4207 struct ixgbe_hw *hw = &adapter->hw;
4208 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4211 /* prevent MTU being changed to a size unsupported by XDP */
4212 if (adapter->xdp_prog) {
4213 dev_warn(&adapter->pdev->dev, "MTU cannot be changed while XDP program is loaded\n");
4217 spin_lock_bh(&adapter->mbx_lock);
4218 /* notify the PF of our intent to use this size of frame */
4219 ret = hw->mac.ops.set_rlpml(hw, max_frame);
4220 spin_unlock_bh(&adapter->mbx_lock);
4224 hw_dbg(hw, "changing MTU from %d to %d\n",
4225 netdev->mtu, new_mtu);
4227 /* must set new MTU before calling down or up */
4228 netdev->mtu = new_mtu;
4230 if (netif_running(netdev))
4231 ixgbevf_reinit_locked(adapter);
4236 #ifdef CONFIG_NET_POLL_CONTROLLER
4237 /* Polling 'interrupt' - used by things like netconsole to send skbs
4238 * without having to re-enable interrupts. It's not called while
4239 * the interrupt routine is executing.
4241 static void ixgbevf_netpoll(struct net_device *netdev)
4243 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4246 /* if interface is down do nothing */
4247 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
4249 for (i = 0; i < adapter->num_rx_queues; i++)
4250 ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
4252 #endif /* CONFIG_NET_POLL_CONTROLLER */
4254 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
4256 struct net_device *netdev = pci_get_drvdata(pdev);
4257 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4263 netif_device_detach(netdev);
4265 if (netif_running(netdev))
4266 ixgbevf_close_suspend(adapter);
4268 ixgbevf_clear_interrupt_scheme(adapter);
4272 retval = pci_save_state(pdev);
4277 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4278 pci_disable_device(pdev);
4284 static int ixgbevf_resume(struct pci_dev *pdev)
4286 struct net_device *netdev = pci_get_drvdata(pdev);
4287 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4290 pci_restore_state(pdev);
4291 /* pci_restore_state clears dev->state_saved so call
4292 * pci_save_state to restore it.
4294 pci_save_state(pdev);
4296 err = pci_enable_device_mem(pdev);
4298 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
4302 adapter->hw.hw_addr = adapter->io_addr;
4303 smp_mb__before_atomic();
4304 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4305 pci_set_master(pdev);
4307 ixgbevf_reset(adapter);
4310 err = ixgbevf_init_interrupt_scheme(adapter);
4311 if (!err && netif_running(netdev))
4312 err = ixgbevf_open(netdev);
4317 netif_device_attach(netdev);
4322 #endif /* CONFIG_PM */
4323 static void ixgbevf_shutdown(struct pci_dev *pdev)
4325 ixgbevf_suspend(pdev, PMSG_SUSPEND);
4328 static void ixgbevf_get_tx_ring_stats(struct rtnl_link_stats64 *stats,
4329 const struct ixgbevf_ring *ring)
4336 start = u64_stats_fetch_begin_irq(&ring->syncp);
4337 bytes = ring->stats.bytes;
4338 packets = ring->stats.packets;
4339 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4340 stats->tx_bytes += bytes;
4341 stats->tx_packets += packets;
4345 static void ixgbevf_get_stats(struct net_device *netdev,
4346 struct rtnl_link_stats64 *stats)
4348 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4351 const struct ixgbevf_ring *ring;
4354 ixgbevf_update_stats(adapter);
4356 stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
4359 for (i = 0; i < adapter->num_rx_queues; i++) {
4360 ring = adapter->rx_ring[i];
4362 start = u64_stats_fetch_begin_irq(&ring->syncp);
4363 bytes = ring->stats.bytes;
4364 packets = ring->stats.packets;
4365 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4366 stats->rx_bytes += bytes;
4367 stats->rx_packets += packets;
4370 for (i = 0; i < adapter->num_tx_queues; i++) {
4371 ring = adapter->tx_ring[i];
4372 ixgbevf_get_tx_ring_stats(stats, ring);
4375 for (i = 0; i < adapter->num_xdp_queues; i++) {
4376 ring = adapter->xdp_ring[i];
4377 ixgbevf_get_tx_ring_stats(stats, ring);
4382 #define IXGBEVF_MAX_MAC_HDR_LEN 127
4383 #define IXGBEVF_MAX_NETWORK_HDR_LEN 511
4385 static netdev_features_t
4386 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
4387 netdev_features_t features)
4389 unsigned int network_hdr_len, mac_hdr_len;
4391 /* Make certain the headers can be described by a context descriptor */
4392 mac_hdr_len = skb_network_header(skb) - skb->data;
4393 if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
4394 return features & ~(NETIF_F_HW_CSUM |
4396 NETIF_F_HW_VLAN_CTAG_TX |
4400 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
4401 if (unlikely(network_hdr_len > IXGBEVF_MAX_NETWORK_HDR_LEN))
4402 return features & ~(NETIF_F_HW_CSUM |
4407 /* We can only support IPV4 TSO in tunnels if we can mangle the
4408 * inner IP ID field, so strip TSO if MANGLEID is not supported.
4410 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
4411 features &= ~NETIF_F_TSO;
4416 static int ixgbevf_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
4418 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4419 struct ixgbevf_adapter *adapter = netdev_priv(dev);
4420 struct bpf_prog *old_prog;
4422 /* verify ixgbevf ring attributes are sufficient for XDP */
4423 for (i = 0; i < adapter->num_rx_queues; i++) {
4424 struct ixgbevf_ring *ring = adapter->rx_ring[i];
4426 if (frame_size > ixgbevf_rx_bufsz(ring))
4430 old_prog = xchg(&adapter->xdp_prog, prog);
4432 /* If transitioning XDP modes reconfigure rings */
4433 if (!!prog != !!old_prog) {
4434 /* Hardware has to reinitialize queues and interrupts to
4435 * match packet buffer alignment. Unfortunately, the
4436 * hardware is not flexible enough to do this dynamically.
4438 if (netif_running(dev))
4441 ixgbevf_clear_interrupt_scheme(adapter);
4442 ixgbevf_init_interrupt_scheme(adapter);
4444 if (netif_running(dev))
4447 for (i = 0; i < adapter->num_rx_queues; i++)
4448 xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
4452 bpf_prog_put(old_prog);
4457 static int ixgbevf_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4459 struct ixgbevf_adapter *adapter = netdev_priv(dev);
4461 switch (xdp->command) {
4462 case XDP_SETUP_PROG:
4463 return ixgbevf_xdp_setup(dev, xdp->prog);
4464 case XDP_QUERY_PROG:
4465 xdp->prog_id = adapter->xdp_prog ?
4466 adapter->xdp_prog->aux->id : 0;
4473 static const struct net_device_ops ixgbevf_netdev_ops = {
4474 .ndo_open = ixgbevf_open,
4475 .ndo_stop = ixgbevf_close,
4476 .ndo_start_xmit = ixgbevf_xmit_frame,
4477 .ndo_set_rx_mode = ixgbevf_set_rx_mode,
4478 .ndo_get_stats64 = ixgbevf_get_stats,
4479 .ndo_validate_addr = eth_validate_addr,
4480 .ndo_set_mac_address = ixgbevf_set_mac,
4481 .ndo_change_mtu = ixgbevf_change_mtu,
4482 .ndo_tx_timeout = ixgbevf_tx_timeout,
4483 .ndo_vlan_rx_add_vid = ixgbevf_vlan_rx_add_vid,
4484 .ndo_vlan_rx_kill_vid = ixgbevf_vlan_rx_kill_vid,
4485 #ifdef CONFIG_NET_POLL_CONTROLLER
4486 .ndo_poll_controller = ixgbevf_netpoll,
4488 .ndo_features_check = ixgbevf_features_check,
4489 .ndo_bpf = ixgbevf_xdp,
4492 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
4494 dev->netdev_ops = &ixgbevf_netdev_ops;
4495 ixgbevf_set_ethtool_ops(dev);
4496 dev->watchdog_timeo = 5 * HZ;
4500 * ixgbevf_probe - Device Initialization Routine
4501 * @pdev: PCI device information struct
4502 * @ent: entry in ixgbevf_pci_tbl
4504 * Returns 0 on success, negative on failure
4506 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
4507 * The OS initialization, configuring of the adapter private structure,
4508 * and a hardware reset occur.
4510 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4512 struct net_device *netdev;
4513 struct ixgbevf_adapter *adapter = NULL;
4514 struct ixgbe_hw *hw = NULL;
4515 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
4516 int err, pci_using_dac;
4517 bool disable_dev = false;
4519 err = pci_enable_device(pdev);
4523 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
4526 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
4528 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4534 err = pci_request_regions(pdev, ixgbevf_driver_name);
4536 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4540 pci_set_master(pdev);
4542 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4546 goto err_alloc_etherdev;
4549 SET_NETDEV_DEV(netdev, &pdev->dev);
4551 adapter = netdev_priv(netdev);
4553 adapter->netdev = netdev;
4554 adapter->pdev = pdev;
4557 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4559 /* call save state here in standalone driver because it relies on
4560 * adapter struct to exist, and needs to call netdev_priv
4562 pci_save_state(pdev);
4564 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4565 pci_resource_len(pdev, 0));
4566 adapter->io_addr = hw->hw_addr;
4572 ixgbevf_assign_netdev_ops(netdev);
4575 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4576 hw->mac.type = ii->mac;
4578 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4579 sizeof(struct ixgbe_mbx_operations));
4581 /* setup the private structure */
4582 err = ixgbevf_sw_init(adapter);
4586 /* The HW MAC address was set and/or determined in sw_init */
4587 if (!is_valid_ether_addr(netdev->dev_addr)) {
4588 pr_err("invalid MAC address\n");
4593 netdev->hw_features = NETIF_F_SG |
4600 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4601 NETIF_F_GSO_GRE_CSUM | \
4602 NETIF_F_GSO_IPXIP4 | \
4603 NETIF_F_GSO_IPXIP6 | \
4604 NETIF_F_GSO_UDP_TUNNEL | \
4605 NETIF_F_GSO_UDP_TUNNEL_CSUM)
4607 netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4608 netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4609 IXGBEVF_GSO_PARTIAL_FEATURES;
4611 netdev->features = netdev->hw_features;
4614 netdev->features |= NETIF_F_HIGHDMA;
4616 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4617 netdev->mpls_features |= NETIF_F_SG |
4621 netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES;
4622 netdev->hw_enc_features |= netdev->vlan_features;
4624 /* set this bit last since it cannot be part of vlan_features */
4625 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4626 NETIF_F_HW_VLAN_CTAG_RX |
4627 NETIF_F_HW_VLAN_CTAG_TX;
4629 netdev->priv_flags |= IFF_UNICAST_FLT;
4631 /* MTU range: 68 - 1504 or 9710 */
4632 netdev->min_mtu = ETH_MIN_MTU;
4633 switch (adapter->hw.api_version) {
4634 case ixgbe_mbox_api_11:
4635 case ixgbe_mbox_api_12:
4636 case ixgbe_mbox_api_13:
4637 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4638 (ETH_HLEN + ETH_FCS_LEN);
4641 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4642 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4643 (ETH_HLEN + ETH_FCS_LEN);
4645 netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4649 if (IXGBE_REMOVED(hw->hw_addr)) {
4654 timer_setup(&adapter->service_timer, ixgbevf_service_timer, 0);
4656 INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4657 set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4658 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4660 err = ixgbevf_init_interrupt_scheme(adapter);
4664 strcpy(netdev->name, "eth%d");
4666 err = register_netdev(netdev);
4670 pci_set_drvdata(pdev, netdev);
4671 netif_carrier_off(netdev);
4673 ixgbevf_init_last_counter_stats(adapter);
4675 /* print the VF info */
4676 dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4677 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4679 switch (hw->mac.type) {
4680 case ixgbe_mac_X550_vf:
4681 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4683 case ixgbe_mac_X540_vf:
4684 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4686 case ixgbe_mac_82599_vf:
4688 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4695 ixgbevf_clear_interrupt_scheme(adapter);
4697 ixgbevf_reset_interrupt_capability(adapter);
4698 iounmap(adapter->io_addr);
4699 kfree(adapter->rss_key);
4701 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4702 free_netdev(netdev);
4704 pci_release_regions(pdev);
4707 if (!adapter || disable_dev)
4708 pci_disable_device(pdev);
4713 * ixgbevf_remove - Device Removal Routine
4714 * @pdev: PCI device information struct
4716 * ixgbevf_remove is called by the PCI subsystem to alert the driver
4717 * that it should release a PCI device. The could be caused by a
4718 * Hot-Plug event, or because the driver is going to be removed from
4721 static void ixgbevf_remove(struct pci_dev *pdev)
4723 struct net_device *netdev = pci_get_drvdata(pdev);
4724 struct ixgbevf_adapter *adapter;
4730 adapter = netdev_priv(netdev);
4732 set_bit(__IXGBEVF_REMOVING, &adapter->state);
4733 cancel_work_sync(&adapter->service_task);
4735 if (netdev->reg_state == NETREG_REGISTERED)
4736 unregister_netdev(netdev);
4738 ixgbevf_clear_interrupt_scheme(adapter);
4739 ixgbevf_reset_interrupt_capability(adapter);
4741 iounmap(adapter->io_addr);
4742 pci_release_regions(pdev);
4744 hw_dbg(&adapter->hw, "Remove complete\n");
4746 kfree(adapter->rss_key);
4747 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4748 free_netdev(netdev);
4751 pci_disable_device(pdev);
4755 * ixgbevf_io_error_detected - called when PCI error is detected
4756 * @pdev: Pointer to PCI device
4757 * @state: The current pci connection state
4759 * This function is called after a PCI bus error affecting
4760 * this device has been detected.
4762 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4763 pci_channel_state_t state)
4765 struct net_device *netdev = pci_get_drvdata(pdev);
4766 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4768 if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4769 return PCI_ERS_RESULT_DISCONNECT;
4772 netif_device_detach(netdev);
4774 if (netif_running(netdev))
4775 ixgbevf_close_suspend(adapter);
4777 if (state == pci_channel_io_perm_failure) {
4779 return PCI_ERS_RESULT_DISCONNECT;
4782 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4783 pci_disable_device(pdev);
4786 /* Request a slot slot reset. */
4787 return PCI_ERS_RESULT_NEED_RESET;
4791 * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4792 * @pdev: Pointer to PCI device
4794 * Restart the card from scratch, as if from a cold-boot. Implementation
4795 * resembles the first-half of the ixgbevf_resume routine.
4797 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4799 struct net_device *netdev = pci_get_drvdata(pdev);
4800 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4802 if (pci_enable_device_mem(pdev)) {
4804 "Cannot re-enable PCI device after reset.\n");
4805 return PCI_ERS_RESULT_DISCONNECT;
4808 adapter->hw.hw_addr = adapter->io_addr;
4809 smp_mb__before_atomic();
4810 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4811 pci_set_master(pdev);
4813 ixgbevf_reset(adapter);
4815 return PCI_ERS_RESULT_RECOVERED;
4819 * ixgbevf_io_resume - called when traffic can start flowing again.
4820 * @pdev: Pointer to PCI device
4822 * This callback is called when the error recovery driver tells us that
4823 * its OK to resume normal operation. Implementation resembles the
4824 * second-half of the ixgbevf_resume routine.
4826 static void ixgbevf_io_resume(struct pci_dev *pdev)
4828 struct net_device *netdev = pci_get_drvdata(pdev);
4831 if (netif_running(netdev))
4832 ixgbevf_open(netdev);
4834 netif_device_attach(netdev);
4838 /* PCI Error Recovery (ERS) */
4839 static const struct pci_error_handlers ixgbevf_err_handler = {
4840 .error_detected = ixgbevf_io_error_detected,
4841 .slot_reset = ixgbevf_io_slot_reset,
4842 .resume = ixgbevf_io_resume,
4845 static struct pci_driver ixgbevf_driver = {
4846 .name = ixgbevf_driver_name,
4847 .id_table = ixgbevf_pci_tbl,
4848 .probe = ixgbevf_probe,
4849 .remove = ixgbevf_remove,
4851 /* Power Management Hooks */
4852 .suspend = ixgbevf_suspend,
4853 .resume = ixgbevf_resume,
4855 .shutdown = ixgbevf_shutdown,
4856 .err_handler = &ixgbevf_err_handler
4860 * ixgbevf_init_module - Driver Registration Routine
4862 * ixgbevf_init_module is the first routine called when the driver is
4863 * loaded. All it does is register with the PCI subsystem.
4865 static int __init ixgbevf_init_module(void)
4867 pr_info("%s - version %s\n", ixgbevf_driver_string,
4868 ixgbevf_driver_version);
4870 pr_info("%s\n", ixgbevf_copyright);
4871 ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4873 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4877 return pci_register_driver(&ixgbevf_driver);
4880 module_init(ixgbevf_init_module);
4883 * ixgbevf_exit_module - Driver Exit Cleanup Routine
4885 * ixgbevf_exit_module is called just before the driver is removed
4888 static void __exit ixgbevf_exit_module(void)
4890 pci_unregister_driver(&ixgbevf_driver);
4892 destroy_workqueue(ixgbevf_wq);
4899 * ixgbevf_get_hw_dev_name - return device name string
4900 * used by hardware layer to print debugging information
4901 * @hw: pointer to private hardware struct
4903 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4905 struct ixgbevf_adapter *adapter = hw->back;
4907 return adapter->netdev->name;
4911 module_exit(ixgbevf_exit_module);
4913 /* ixgbevf_main.c */