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>
37 const char ixgbevf_driver_name[] = "ixgbevf";
38 static const char ixgbevf_driver_string[] =
39 "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
41 static char ixgbevf_copyright[] =
42 "Copyright (c) 2009 - 2018 Intel Corporation.";
44 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
45 [board_82599_vf] = &ixgbevf_82599_vf_info,
46 [board_82599_vf_hv] = &ixgbevf_82599_vf_hv_info,
47 [board_X540_vf] = &ixgbevf_X540_vf_info,
48 [board_X540_vf_hv] = &ixgbevf_X540_vf_hv_info,
49 [board_X550_vf] = &ixgbevf_X550_vf_info,
50 [board_X550_vf_hv] = &ixgbevf_X550_vf_hv_info,
51 [board_X550EM_x_vf] = &ixgbevf_X550EM_x_vf_info,
52 [board_X550EM_x_vf_hv] = &ixgbevf_X550EM_x_vf_hv_info,
53 [board_x550em_a_vf] = &ixgbevf_x550em_a_vf_info,
56 /* ixgbevf_pci_tbl - PCI Device ID Table
58 * Wildcard entries (PCI_ANY_ID) should come last
59 * Last entry must be all 0s
61 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
62 * Class, Class Mask, private data (not used) }
64 static const struct pci_device_id ixgbevf_pci_tbl[] = {
65 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
66 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
67 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
68 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
69 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
70 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
71 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
72 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
73 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
74 /* required last entry */
77 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
79 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
80 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
81 MODULE_LICENSE("GPL v2");
83 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
84 static int debug = -1;
85 module_param(debug, int, 0);
86 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
88 static struct workqueue_struct *ixgbevf_wq;
90 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
92 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
93 !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
94 !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
95 queue_work(ixgbevf_wq, &adapter->service_task);
98 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
100 BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
102 /* flush memory to make sure state is correct before next watchdog */
103 smp_mb__before_atomic();
104 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
108 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
109 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
110 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
111 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer);
112 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
113 struct ixgbevf_rx_buffer *old_buff);
115 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
117 struct ixgbevf_adapter *adapter = hw->back;
122 dev_err(&adapter->pdev->dev, "Adapter removed\n");
123 if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
124 ixgbevf_service_event_schedule(adapter);
127 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
131 /* The following check not only optimizes a bit by not
132 * performing a read on the status register when the
133 * register just read was a status register read that
134 * returned IXGBE_FAILED_READ_REG. It also blocks any
135 * potential recursion.
137 if (reg == IXGBE_VFSTATUS) {
138 ixgbevf_remove_adapter(hw);
141 value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
142 if (value == IXGBE_FAILED_READ_REG)
143 ixgbevf_remove_adapter(hw);
146 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
148 u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
151 if (IXGBE_REMOVED(reg_addr))
152 return IXGBE_FAILED_READ_REG;
153 value = readl(reg_addr + reg);
154 if (unlikely(value == IXGBE_FAILED_READ_REG))
155 ixgbevf_check_remove(hw, reg);
160 * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
161 * @adapter: pointer to adapter struct
162 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
163 * @queue: queue to map the corresponding interrupt to
164 * @msix_vector: the vector to map to the corresponding queue
166 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
167 u8 queue, u8 msix_vector)
170 struct ixgbe_hw *hw = &adapter->hw;
172 if (direction == -1) {
174 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
175 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
178 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
180 /* Tx or Rx causes */
181 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
182 index = ((16 * (queue & 1)) + (8 * direction));
183 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
184 ivar &= ~(0xFF << index);
185 ivar |= (msix_vector << index);
186 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
190 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
192 return ring->stats.packets;
195 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
197 struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
198 struct ixgbe_hw *hw = &adapter->hw;
200 u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
201 u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
204 return (head < tail) ?
205 tail - head : (tail + ring->count - head);
210 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
212 u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
213 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
214 u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
216 clear_check_for_tx_hang(tx_ring);
218 /* Check for a hung queue, but be thorough. This verifies
219 * that a transmit has been completed since the previous
220 * check AND there is at least one packet pending. The
221 * ARMED bit is set to indicate a potential hang.
223 if ((tx_done_old == tx_done) && tx_pending) {
224 /* make sure it is true for two checks in a row */
225 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
228 /* reset the countdown */
229 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
231 /* update completed stats and continue */
232 tx_ring->tx_stats.tx_done_old = tx_done;
237 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
239 /* Do the reset outside of interrupt context */
240 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
241 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
242 ixgbevf_service_event_schedule(adapter);
247 * ixgbevf_tx_timeout - Respond to a Tx Hang
248 * @netdev: network interface device structure
249 * @txqueue: transmit queue hanging (unused)
251 static void ixgbevf_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
253 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
255 ixgbevf_tx_timeout_reset(adapter);
259 * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
260 * @q_vector: board private structure
261 * @tx_ring: tx ring to clean
262 * @napi_budget: Used to determine if we are in netpoll
264 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
265 struct ixgbevf_ring *tx_ring, int napi_budget)
267 struct ixgbevf_adapter *adapter = q_vector->adapter;
268 struct ixgbevf_tx_buffer *tx_buffer;
269 union ixgbe_adv_tx_desc *tx_desc;
270 unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
271 unsigned int budget = tx_ring->count / 2;
272 unsigned int i = tx_ring->next_to_clean;
274 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
277 tx_buffer = &tx_ring->tx_buffer_info[i];
278 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
282 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
284 /* if next_to_watch is not set then there is no work pending */
288 /* prevent any other reads prior to eop_desc */
291 /* if DD is not set pending work has not been completed */
292 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
295 /* clear next_to_watch to prevent false hangs */
296 tx_buffer->next_to_watch = NULL;
298 /* update the statistics for this packet */
299 total_bytes += tx_buffer->bytecount;
300 total_packets += tx_buffer->gso_segs;
301 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
305 if (ring_is_xdp(tx_ring))
306 page_frag_free(tx_buffer->data);
308 napi_consume_skb(tx_buffer->skb, napi_budget);
310 /* unmap skb header data */
311 dma_unmap_single(tx_ring->dev,
312 dma_unmap_addr(tx_buffer, dma),
313 dma_unmap_len(tx_buffer, len),
316 /* clear tx_buffer data */
317 dma_unmap_len_set(tx_buffer, len, 0);
319 /* unmap remaining buffers */
320 while (tx_desc != eop_desc) {
326 tx_buffer = tx_ring->tx_buffer_info;
327 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
330 /* unmap any remaining paged data */
331 if (dma_unmap_len(tx_buffer, len)) {
332 dma_unmap_page(tx_ring->dev,
333 dma_unmap_addr(tx_buffer, dma),
334 dma_unmap_len(tx_buffer, len),
336 dma_unmap_len_set(tx_buffer, len, 0);
340 /* move us one more past the eop_desc for start of next pkt */
346 tx_buffer = tx_ring->tx_buffer_info;
347 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
350 /* issue prefetch for next Tx descriptor */
353 /* update budget accounting */
355 } while (likely(budget));
358 tx_ring->next_to_clean = i;
359 u64_stats_update_begin(&tx_ring->syncp);
360 tx_ring->stats.bytes += total_bytes;
361 tx_ring->stats.packets += total_packets;
362 u64_stats_update_end(&tx_ring->syncp);
363 q_vector->tx.total_bytes += total_bytes;
364 q_vector->tx.total_packets += total_packets;
365 adapter->tx_ipsec += total_ipsec;
367 if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
368 struct ixgbe_hw *hw = &adapter->hw;
369 union ixgbe_adv_tx_desc *eop_desc;
371 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
373 pr_err("Detected Tx Unit Hang%s\n"
375 " TDH, TDT <%x>, <%x>\n"
376 " next_to_use <%x>\n"
377 " next_to_clean <%x>\n"
378 "tx_buffer_info[next_to_clean]\n"
379 " next_to_watch <%p>\n"
380 " eop_desc->wb.status <%x>\n"
381 " time_stamp <%lx>\n"
383 ring_is_xdp(tx_ring) ? " XDP" : "",
384 tx_ring->queue_index,
385 IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
386 IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
387 tx_ring->next_to_use, i,
388 eop_desc, (eop_desc ? eop_desc->wb.status : 0),
389 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
391 if (!ring_is_xdp(tx_ring))
392 netif_stop_subqueue(tx_ring->netdev,
393 tx_ring->queue_index);
395 /* schedule immediate reset if we believe we hung */
396 ixgbevf_tx_timeout_reset(adapter);
401 if (ring_is_xdp(tx_ring))
404 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
405 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
406 (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
407 /* Make sure that anybody stopping the queue after this
408 * sees the new next_to_clean.
412 if (__netif_subqueue_stopped(tx_ring->netdev,
413 tx_ring->queue_index) &&
414 !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
415 netif_wake_subqueue(tx_ring->netdev,
416 tx_ring->queue_index);
417 ++tx_ring->tx_stats.restart_queue;
425 * ixgbevf_rx_skb - Helper function to determine proper Rx method
426 * @q_vector: structure containing interrupt and ring information
427 * @skb: packet to send up
429 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
432 napi_gro_receive(&q_vector->napi, skb);
435 #define IXGBE_RSS_L4_TYPES_MASK \
436 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
437 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
438 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
439 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
441 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
442 union ixgbe_adv_rx_desc *rx_desc,
447 if (!(ring->netdev->features & NETIF_F_RXHASH))
450 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
451 IXGBE_RXDADV_RSSTYPE_MASK;
456 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
457 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
458 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
462 * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
463 * @ring: structure containig ring specific data
464 * @rx_desc: current Rx descriptor being processed
465 * @skb: skb currently being received and modified
467 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
468 union ixgbe_adv_rx_desc *rx_desc,
471 skb_checksum_none_assert(skb);
473 /* Rx csum disabled */
474 if (!(ring->netdev->features & NETIF_F_RXCSUM))
477 /* if IP and error */
478 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
479 ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
480 ring->rx_stats.csum_err++;
484 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
487 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
488 ring->rx_stats.csum_err++;
492 /* It must be a TCP or UDP packet with a valid checksum */
493 skb->ip_summed = CHECKSUM_UNNECESSARY;
497 * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
498 * @rx_ring: rx descriptor ring packet is being transacted on
499 * @rx_desc: pointer to the EOP Rx descriptor
500 * @skb: pointer to current skb being populated
502 * This function checks the ring, descriptor, and packet information in
503 * order to populate the checksum, VLAN, protocol, and other fields within
506 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
507 union ixgbe_adv_rx_desc *rx_desc,
510 ixgbevf_rx_hash(rx_ring, rx_desc, skb);
511 ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
513 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
514 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
515 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
517 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
518 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
521 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
522 ixgbevf_ipsec_rx(rx_ring, rx_desc, skb);
524 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
528 struct ixgbevf_rx_buffer *ixgbevf_get_rx_buffer(struct ixgbevf_ring *rx_ring,
529 const unsigned int size)
531 struct ixgbevf_rx_buffer *rx_buffer;
533 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
534 prefetchw(rx_buffer->page);
536 /* we are reusing so sync this buffer for CPU use */
537 dma_sync_single_range_for_cpu(rx_ring->dev,
539 rx_buffer->page_offset,
543 rx_buffer->pagecnt_bias--;
548 static void ixgbevf_put_rx_buffer(struct ixgbevf_ring *rx_ring,
549 struct ixgbevf_rx_buffer *rx_buffer,
552 if (ixgbevf_can_reuse_rx_page(rx_buffer)) {
553 /* hand second half of page back to the ring */
554 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
557 /* We are not reusing the buffer so unmap it and free
558 * any references we are holding to it
560 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
561 ixgbevf_rx_pg_size(rx_ring),
563 IXGBEVF_RX_DMA_ATTR);
564 __page_frag_cache_drain(rx_buffer->page,
565 rx_buffer->pagecnt_bias);
568 /* clear contents of rx_buffer */
569 rx_buffer->page = NULL;
573 * ixgbevf_is_non_eop - process handling of non-EOP buffers
574 * @rx_ring: Rx ring being processed
575 * @rx_desc: Rx descriptor for current buffer
577 * This function updates next to clean. If the buffer is an EOP buffer
578 * this function exits returning false, otherwise it will place the
579 * sk_buff in the next buffer to be chained and return true indicating
580 * that this is in fact a non-EOP buffer.
582 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
583 union ixgbe_adv_rx_desc *rx_desc)
585 u32 ntc = rx_ring->next_to_clean + 1;
587 /* fetch, update, and store next to clean */
588 ntc = (ntc < rx_ring->count) ? ntc : 0;
589 rx_ring->next_to_clean = ntc;
591 prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
593 if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
599 static inline unsigned int ixgbevf_rx_offset(struct ixgbevf_ring *rx_ring)
601 return ring_uses_build_skb(rx_ring) ? IXGBEVF_SKB_PAD : 0;
604 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
605 struct ixgbevf_rx_buffer *bi)
607 struct page *page = bi->page;
610 /* since we are recycling buffers we should seldom need to alloc */
614 /* alloc new page for storage */
615 page = dev_alloc_pages(ixgbevf_rx_pg_order(rx_ring));
616 if (unlikely(!page)) {
617 rx_ring->rx_stats.alloc_rx_page_failed++;
621 /* map page for use */
622 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
623 ixgbevf_rx_pg_size(rx_ring),
624 DMA_FROM_DEVICE, IXGBEVF_RX_DMA_ATTR);
626 /* if mapping failed free memory back to system since
627 * there isn't much point in holding memory we can't use
629 if (dma_mapping_error(rx_ring->dev, dma)) {
630 __free_pages(page, ixgbevf_rx_pg_order(rx_ring));
632 rx_ring->rx_stats.alloc_rx_page_failed++;
638 bi->page_offset = ixgbevf_rx_offset(rx_ring);
639 bi->pagecnt_bias = 1;
640 rx_ring->rx_stats.alloc_rx_page++;
646 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
647 * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
648 * @cleaned_count: number of buffers to replace
650 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
653 union ixgbe_adv_rx_desc *rx_desc;
654 struct ixgbevf_rx_buffer *bi;
655 unsigned int i = rx_ring->next_to_use;
657 /* nothing to do or no valid netdev defined */
658 if (!cleaned_count || !rx_ring->netdev)
661 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
662 bi = &rx_ring->rx_buffer_info[i];
666 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
669 /* sync the buffer for use by the device */
670 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
672 ixgbevf_rx_bufsz(rx_ring),
675 /* Refresh the desc even if pkt_addr didn't change
676 * because each write-back erases this info.
678 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
684 rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
685 bi = rx_ring->rx_buffer_info;
689 /* clear the length for the next_to_use descriptor */
690 rx_desc->wb.upper.length = 0;
693 } while (cleaned_count);
697 if (rx_ring->next_to_use != i) {
698 /* record the next descriptor to use */
699 rx_ring->next_to_use = i;
701 /* update next to alloc since we have filled the ring */
702 rx_ring->next_to_alloc = i;
704 /* Force memory writes to complete before letting h/w
705 * know there are new descriptors to fetch. (Only
706 * applicable for weak-ordered memory model archs,
710 ixgbevf_write_tail(rx_ring, i);
715 * ixgbevf_cleanup_headers - Correct corrupted or empty headers
716 * @rx_ring: rx descriptor ring packet is being transacted on
717 * @rx_desc: pointer to the EOP Rx descriptor
718 * @skb: pointer to current skb being fixed
720 * Check for corrupted packet headers caused by senders on the local L2
721 * embedded NIC switch not setting up their Tx Descriptors right. These
722 * should be very rare.
724 * Also address the case where we are pulling data in on pages only
725 * and as such no data is present in the skb header.
727 * In addition if skb is not at least 60 bytes we need to pad it so that
728 * it is large enough to qualify as a valid Ethernet frame.
730 * Returns true if an error was encountered and skb was freed.
732 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
733 union ixgbe_adv_rx_desc *rx_desc,
736 /* XDP packets use error pointer so abort at this point */
740 /* verify that the packet does not have any known errors */
741 if (unlikely(ixgbevf_test_staterr(rx_desc,
742 IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
743 struct net_device *netdev = rx_ring->netdev;
745 if (!(netdev->features & NETIF_F_RXALL)) {
746 dev_kfree_skb_any(skb);
751 /* if eth_skb_pad returns an error the skb was freed */
752 if (eth_skb_pad(skb))
759 * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
760 * @rx_ring: rx descriptor ring to store buffers on
761 * @old_buff: donor buffer to have page reused
763 * Synchronizes page for reuse by the adapter
765 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
766 struct ixgbevf_rx_buffer *old_buff)
768 struct ixgbevf_rx_buffer *new_buff;
769 u16 nta = rx_ring->next_to_alloc;
771 new_buff = &rx_ring->rx_buffer_info[nta];
773 /* update, and store next to alloc */
775 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
777 /* transfer page from old buffer to new buffer */
778 new_buff->page = old_buff->page;
779 new_buff->dma = old_buff->dma;
780 new_buff->page_offset = old_buff->page_offset;
781 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
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 and pfmemalloc pages */
790 if (!dev_page_is_reusable(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 */
865 net_prefetch(xdp->data);
867 /* Note, we get here by enabling legacy-rx via:
869 * ethtool --set-priv-flags <dev> legacy-rx on
871 * In this mode, we currently get 0 extra XDP headroom as
872 * opposed to having legacy-rx off, where we process XDP
873 * packets going to stack via ixgbevf_build_skb().
875 * For ixgbevf_construct_skb() mode it means that the
876 * xdp->data_meta will always point to xdp->data, since
877 * the helper cannot expand the head. Should this ever
878 * changed in future for legacy-rx mode on, then lets also
879 * add xdp->data_meta handling here.
882 /* allocate a skb to store the frags */
883 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBEVF_RX_HDR_SIZE);
887 /* Determine available headroom for copy */
889 if (headlen > IXGBEVF_RX_HDR_SIZE)
890 headlen = eth_get_headlen(skb->dev, xdp->data,
891 IXGBEVF_RX_HDR_SIZE);
893 /* align pull length to size of long to optimize memcpy performance */
894 memcpy(__skb_put(skb, headlen), xdp->data,
895 ALIGN(headlen, sizeof(long)));
897 /* update all of the pointers */
900 skb_add_rx_frag(skb, 0, rx_buffer->page,
901 (xdp->data + headlen) -
902 page_address(rx_buffer->page),
904 #if (PAGE_SIZE < 8192)
905 rx_buffer->page_offset ^= truesize;
907 rx_buffer->page_offset += truesize;
910 rx_buffer->pagecnt_bias++;
916 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
919 struct ixgbe_hw *hw = &adapter->hw;
921 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
924 static struct sk_buff *ixgbevf_build_skb(struct ixgbevf_ring *rx_ring,
925 struct ixgbevf_rx_buffer *rx_buffer,
926 struct xdp_buff *xdp,
927 union ixgbe_adv_rx_desc *rx_desc)
929 unsigned int metasize = xdp->data - xdp->data_meta;
930 #if (PAGE_SIZE < 8192)
931 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
933 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
934 SKB_DATA_ALIGN(xdp->data_end -
935 xdp->data_hard_start);
939 /* Prefetch first cache line of first page. If xdp->data_meta
940 * is unused, this points to xdp->data, otherwise, we likely
941 * have a consumer accessing first few bytes of meta data,
942 * and then actual data.
944 net_prefetch(xdp->data_meta);
946 /* build an skb around the page buffer */
947 skb = build_skb(xdp->data_hard_start, truesize);
951 /* update pointers within the skb to store the data */
952 skb_reserve(skb, xdp->data - xdp->data_hard_start);
953 __skb_put(skb, xdp->data_end - xdp->data);
955 skb_metadata_set(skb, metasize);
957 /* update buffer offset */
958 #if (PAGE_SIZE < 8192)
959 rx_buffer->page_offset ^= truesize;
961 rx_buffer->page_offset += truesize;
967 #define IXGBEVF_XDP_PASS 0
968 #define IXGBEVF_XDP_CONSUMED 1
969 #define IXGBEVF_XDP_TX 2
971 static int ixgbevf_xmit_xdp_ring(struct ixgbevf_ring *ring,
972 struct xdp_buff *xdp)
974 struct ixgbevf_tx_buffer *tx_buffer;
975 union ixgbe_adv_tx_desc *tx_desc;
980 len = xdp->data_end - xdp->data;
982 if (unlikely(!ixgbevf_desc_unused(ring)))
983 return IXGBEVF_XDP_CONSUMED;
985 dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE);
986 if (dma_mapping_error(ring->dev, dma))
987 return IXGBEVF_XDP_CONSUMED;
989 /* record the location of the first descriptor for this packet */
990 i = ring->next_to_use;
991 tx_buffer = &ring->tx_buffer_info[i];
993 dma_unmap_len_set(tx_buffer, len, len);
994 dma_unmap_addr_set(tx_buffer, dma, dma);
995 tx_buffer->data = xdp->data;
996 tx_buffer->bytecount = len;
997 tx_buffer->gso_segs = 1;
998 tx_buffer->protocol = 0;
1000 /* Populate minimal context descriptor that will provide for the
1001 * fact that we are expected to process Ethernet frames.
1003 if (!test_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state)) {
1004 struct ixgbe_adv_tx_context_desc *context_desc;
1006 set_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1008 context_desc = IXGBEVF_TX_CTXTDESC(ring, 0);
1009 context_desc->vlan_macip_lens =
1010 cpu_to_le32(ETH_HLEN << IXGBE_ADVTXD_MACLEN_SHIFT);
1011 context_desc->fceof_saidx = 0;
1012 context_desc->type_tucmd_mlhl =
1013 cpu_to_le32(IXGBE_TXD_CMD_DEXT |
1014 IXGBE_ADVTXD_DTYP_CTXT);
1015 context_desc->mss_l4len_idx = 0;
1020 /* put descriptor type bits */
1021 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
1022 IXGBE_ADVTXD_DCMD_DEXT |
1023 IXGBE_ADVTXD_DCMD_IFCS;
1024 cmd_type |= len | IXGBE_TXD_CMD;
1026 tx_desc = IXGBEVF_TX_DESC(ring, i);
1027 tx_desc->read.buffer_addr = cpu_to_le64(dma);
1029 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1030 tx_desc->read.olinfo_status =
1031 cpu_to_le32((len << IXGBE_ADVTXD_PAYLEN_SHIFT) |
1034 /* Avoid any potential race with cleanup */
1037 /* set next_to_watch value indicating a packet is present */
1039 if (i == ring->count)
1042 tx_buffer->next_to_watch = tx_desc;
1043 ring->next_to_use = i;
1045 return IXGBEVF_XDP_TX;
1048 static struct sk_buff *ixgbevf_run_xdp(struct ixgbevf_adapter *adapter,
1049 struct ixgbevf_ring *rx_ring,
1050 struct xdp_buff *xdp)
1052 int result = IXGBEVF_XDP_PASS;
1053 struct ixgbevf_ring *xdp_ring;
1054 struct bpf_prog *xdp_prog;
1057 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
1062 act = bpf_prog_run_xdp(xdp_prog, xdp);
1067 xdp_ring = adapter->xdp_ring[rx_ring->queue_index];
1068 result = ixgbevf_xmit_xdp_ring(xdp_ring, xdp);
1069 if (result == IXGBEVF_XDP_CONSUMED)
1073 bpf_warn_invalid_xdp_action(act);
1077 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
1078 fallthrough; /* handle aborts by dropping packet */
1080 result = IXGBEVF_XDP_CONSUMED;
1084 return ERR_PTR(-result);
1087 static unsigned int ixgbevf_rx_frame_truesize(struct ixgbevf_ring *rx_ring,
1090 unsigned int truesize;
1092 #if (PAGE_SIZE < 8192)
1093 truesize = ixgbevf_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
1095 truesize = ring_uses_build_skb(rx_ring) ?
1096 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) +
1097 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
1098 SKB_DATA_ALIGN(size);
1103 static void ixgbevf_rx_buffer_flip(struct ixgbevf_ring *rx_ring,
1104 struct ixgbevf_rx_buffer *rx_buffer,
1107 unsigned int truesize = ixgbevf_rx_frame_truesize(rx_ring, size);
1109 #if (PAGE_SIZE < 8192)
1110 rx_buffer->page_offset ^= truesize;
1112 rx_buffer->page_offset += truesize;
1116 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
1117 struct ixgbevf_ring *rx_ring,
1120 unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
1121 struct ixgbevf_adapter *adapter = q_vector->adapter;
1122 u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
1123 struct sk_buff *skb = rx_ring->skb;
1124 bool xdp_xmit = false;
1125 struct xdp_buff xdp;
1127 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
1128 #if (PAGE_SIZE < 8192)
1129 frame_sz = ixgbevf_rx_frame_truesize(rx_ring, 0);
1131 xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
1133 while (likely(total_rx_packets < budget)) {
1134 struct ixgbevf_rx_buffer *rx_buffer;
1135 union ixgbe_adv_rx_desc *rx_desc;
1138 /* return some buffers to hardware, one at a time is too slow */
1139 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
1140 ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
1144 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
1145 size = le16_to_cpu(rx_desc->wb.upper.length);
1149 /* This memory barrier is needed to keep us from reading
1150 * any other fields out of the rx_desc until we know the
1151 * RXD_STAT_DD bit is set
1155 rx_buffer = ixgbevf_get_rx_buffer(rx_ring, size);
1157 /* retrieve a buffer from the ring */
1159 unsigned int offset = ixgbevf_rx_offset(rx_ring);
1160 unsigned char *hard_start;
1162 hard_start = page_address(rx_buffer->page) +
1163 rx_buffer->page_offset - offset;
1164 xdp_prepare_buff(&xdp, hard_start, offset, size, true);
1165 #if (PAGE_SIZE > 4096)
1166 /* At larger PAGE_SIZE, frame_sz depend on len size */
1167 xdp.frame_sz = ixgbevf_rx_frame_truesize(rx_ring, size);
1169 skb = ixgbevf_run_xdp(adapter, rx_ring, &xdp);
1173 if (PTR_ERR(skb) == -IXGBEVF_XDP_TX) {
1175 ixgbevf_rx_buffer_flip(rx_ring, rx_buffer,
1178 rx_buffer->pagecnt_bias++;
1181 total_rx_bytes += size;
1183 ixgbevf_add_rx_frag(rx_ring, rx_buffer, skb, size);
1184 } else if (ring_uses_build_skb(rx_ring)) {
1185 skb = ixgbevf_build_skb(rx_ring, rx_buffer,
1188 skb = ixgbevf_construct_skb(rx_ring, rx_buffer,
1192 /* exit if we failed to retrieve a buffer */
1194 rx_ring->rx_stats.alloc_rx_buff_failed++;
1195 rx_buffer->pagecnt_bias++;
1199 ixgbevf_put_rx_buffer(rx_ring, rx_buffer, skb);
1202 /* fetch next buffer in frame if non-eop */
1203 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
1206 /* verify the packet layout is correct */
1207 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
1212 /* probably a little skewed due to removing CRC */
1213 total_rx_bytes += skb->len;
1215 /* Workaround hardware that can't do proper VEPA multicast
1218 if ((skb->pkt_type == PACKET_BROADCAST ||
1219 skb->pkt_type == PACKET_MULTICAST) &&
1220 ether_addr_equal(rx_ring->netdev->dev_addr,
1221 eth_hdr(skb)->h_source)) {
1222 dev_kfree_skb_irq(skb);
1226 /* populate checksum, VLAN, and protocol */
1227 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
1229 ixgbevf_rx_skb(q_vector, skb);
1231 /* reset skb pointer */
1234 /* update budget accounting */
1238 /* place incomplete frames back on ring for completion */
1242 struct ixgbevf_ring *xdp_ring =
1243 adapter->xdp_ring[rx_ring->queue_index];
1245 /* Force memory writes to complete before letting h/w
1246 * know there are new descriptors to fetch.
1249 ixgbevf_write_tail(xdp_ring, xdp_ring->next_to_use);
1252 u64_stats_update_begin(&rx_ring->syncp);
1253 rx_ring->stats.packets += total_rx_packets;
1254 rx_ring->stats.bytes += total_rx_bytes;
1255 u64_stats_update_end(&rx_ring->syncp);
1256 q_vector->rx.total_packets += total_rx_packets;
1257 q_vector->rx.total_bytes += total_rx_bytes;
1259 return total_rx_packets;
1263 * ixgbevf_poll - NAPI polling calback
1264 * @napi: napi struct with our devices info in it
1265 * @budget: amount of work driver is allowed to do this pass, in packets
1267 * This function will clean more than one or more rings associated with a
1270 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1272 struct ixgbevf_q_vector *q_vector =
1273 container_of(napi, struct ixgbevf_q_vector, napi);
1274 struct ixgbevf_adapter *adapter = q_vector->adapter;
1275 struct ixgbevf_ring *ring;
1276 int per_ring_budget, work_done = 0;
1277 bool clean_complete = true;
1279 ixgbevf_for_each_ring(ring, q_vector->tx) {
1280 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1281 clean_complete = false;
1287 /* attempt to distribute budget to each queue fairly, but don't allow
1288 * the budget to go below 1 because we'll exit polling
1290 if (q_vector->rx.count > 1)
1291 per_ring_budget = max(budget/q_vector->rx.count, 1);
1293 per_ring_budget = budget;
1295 ixgbevf_for_each_ring(ring, q_vector->rx) {
1296 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1298 work_done += cleaned;
1299 if (cleaned >= per_ring_budget)
1300 clean_complete = false;
1303 /* If all work not completed, return budget and keep polling */
1304 if (!clean_complete)
1307 /* Exit the polling mode, but don't re-enable interrupts if stack might
1308 * poll us due to busy-polling
1310 if (likely(napi_complete_done(napi, work_done))) {
1311 if (adapter->rx_itr_setting == 1)
1312 ixgbevf_set_itr(q_vector);
1313 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1314 !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1315 ixgbevf_irq_enable_queues(adapter,
1316 BIT(q_vector->v_idx));
1319 return min(work_done, budget - 1);
1323 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1324 * @q_vector: structure containing interrupt and ring information
1326 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1328 struct ixgbevf_adapter *adapter = q_vector->adapter;
1329 struct ixgbe_hw *hw = &adapter->hw;
1330 int v_idx = q_vector->v_idx;
1331 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1333 /* set the WDIS bit to not clear the timer bits and cause an
1334 * immediate assertion of the interrupt
1336 itr_reg |= IXGBE_EITR_CNT_WDIS;
1338 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1342 * ixgbevf_configure_msix - Configure MSI-X hardware
1343 * @adapter: board private structure
1345 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1348 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1350 struct ixgbevf_q_vector *q_vector;
1351 int q_vectors, v_idx;
1353 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1354 adapter->eims_enable_mask = 0;
1356 /* Populate the IVAR table and set the ITR values to the
1357 * corresponding register.
1359 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1360 struct ixgbevf_ring *ring;
1362 q_vector = adapter->q_vector[v_idx];
1364 ixgbevf_for_each_ring(ring, q_vector->rx)
1365 ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1367 ixgbevf_for_each_ring(ring, q_vector->tx)
1368 ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1370 if (q_vector->tx.ring && !q_vector->rx.ring) {
1371 /* Tx only vector */
1372 if (adapter->tx_itr_setting == 1)
1373 q_vector->itr = IXGBE_12K_ITR;
1375 q_vector->itr = adapter->tx_itr_setting;
1377 /* Rx or Rx/Tx vector */
1378 if (adapter->rx_itr_setting == 1)
1379 q_vector->itr = IXGBE_20K_ITR;
1381 q_vector->itr = adapter->rx_itr_setting;
1384 /* add q_vector eims value to global eims_enable_mask */
1385 adapter->eims_enable_mask |= BIT(v_idx);
1387 ixgbevf_write_eitr(q_vector);
1390 ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1391 /* setup eims_other and add value to global eims_enable_mask */
1392 adapter->eims_other = BIT(v_idx);
1393 adapter->eims_enable_mask |= adapter->eims_other;
1396 enum latency_range {
1400 latency_invalid = 255
1404 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1405 * @q_vector: structure containing interrupt and ring information
1406 * @ring_container: structure containing ring performance data
1408 * Stores a new ITR value based on packets and byte
1409 * counts during the last interrupt. The advantage of per interrupt
1410 * computation is faster updates and more accurate ITR for the current
1411 * traffic pattern. Constants in this function were computed
1412 * based on theoretical maximum wire speed and thresholds were set based
1413 * on testing data as well as attempting to minimize response time
1414 * while increasing bulk throughput.
1416 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1417 struct ixgbevf_ring_container *ring_container)
1419 int bytes = ring_container->total_bytes;
1420 int packets = ring_container->total_packets;
1423 u8 itr_setting = ring_container->itr;
1428 /* simple throttle rate management
1429 * 0-20MB/s lowest (100000 ints/s)
1430 * 20-100MB/s low (20000 ints/s)
1431 * 100-1249MB/s bulk (12000 ints/s)
1433 /* what was last interrupt timeslice? */
1434 timepassed_us = q_vector->itr >> 2;
1435 if (timepassed_us == 0)
1438 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1440 switch (itr_setting) {
1441 case lowest_latency:
1442 if (bytes_perint > 10)
1443 itr_setting = low_latency;
1446 if (bytes_perint > 20)
1447 itr_setting = bulk_latency;
1448 else if (bytes_perint <= 10)
1449 itr_setting = lowest_latency;
1452 if (bytes_perint <= 20)
1453 itr_setting = low_latency;
1457 /* clear work counters since we have the values we need */
1458 ring_container->total_bytes = 0;
1459 ring_container->total_packets = 0;
1461 /* write updated itr to ring container */
1462 ring_container->itr = itr_setting;
1465 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1467 u32 new_itr = q_vector->itr;
1470 ixgbevf_update_itr(q_vector, &q_vector->tx);
1471 ixgbevf_update_itr(q_vector, &q_vector->rx);
1473 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1475 switch (current_itr) {
1476 /* counts and packets in update_itr are dependent on these numbers */
1477 case lowest_latency:
1478 new_itr = IXGBE_100K_ITR;
1481 new_itr = IXGBE_20K_ITR;
1484 new_itr = IXGBE_12K_ITR;
1490 if (new_itr != q_vector->itr) {
1491 /* do an exponential smoothing */
1492 new_itr = (10 * new_itr * q_vector->itr) /
1493 ((9 * new_itr) + q_vector->itr);
1495 /* save the algorithm value here */
1496 q_vector->itr = new_itr;
1498 ixgbevf_write_eitr(q_vector);
1502 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1504 struct ixgbevf_adapter *adapter = data;
1505 struct ixgbe_hw *hw = &adapter->hw;
1507 hw->mac.get_link_status = 1;
1509 ixgbevf_service_event_schedule(adapter);
1511 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1517 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1519 * @data: pointer to our q_vector struct for this interrupt vector
1521 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1523 struct ixgbevf_q_vector *q_vector = data;
1525 /* EIAM disabled interrupts (on this vector) for us */
1526 if (q_vector->rx.ring || q_vector->tx.ring)
1527 napi_schedule_irqoff(&q_vector->napi);
1533 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1534 * @adapter: board private structure
1536 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1537 * interrupts from the kernel.
1539 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1541 struct net_device *netdev = adapter->netdev;
1542 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1543 unsigned int ri = 0, ti = 0;
1546 for (vector = 0; vector < q_vectors; vector++) {
1547 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1548 struct msix_entry *entry = &adapter->msix_entries[vector];
1550 if (q_vector->tx.ring && q_vector->rx.ring) {
1551 snprintf(q_vector->name, sizeof(q_vector->name),
1552 "%s-TxRx-%u", netdev->name, ri++);
1554 } else if (q_vector->rx.ring) {
1555 snprintf(q_vector->name, sizeof(q_vector->name),
1556 "%s-rx-%u", netdev->name, ri++);
1557 } else if (q_vector->tx.ring) {
1558 snprintf(q_vector->name, sizeof(q_vector->name),
1559 "%s-tx-%u", netdev->name, ti++);
1561 /* skip this unused q_vector */
1564 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1565 q_vector->name, q_vector);
1567 hw_dbg(&adapter->hw,
1568 "request_irq failed for MSIX interrupt Error: %d\n",
1570 goto free_queue_irqs;
1574 err = request_irq(adapter->msix_entries[vector].vector,
1575 &ixgbevf_msix_other, 0, netdev->name, adapter);
1577 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1579 goto free_queue_irqs;
1587 free_irq(adapter->msix_entries[vector].vector,
1588 adapter->q_vector[vector]);
1590 /* This failure is non-recoverable - it indicates the system is
1591 * out of MSIX vector resources and the VF driver cannot run
1592 * without them. Set the number of msix vectors to zero
1593 * indicating that not enough can be allocated. The error
1594 * will be returned to the user indicating device open failed.
1595 * Any further attempts to force the driver to open will also
1596 * fail. The only way to recover is to unload the driver and
1597 * reload it again. If the system has recovered some MSIX
1598 * vectors then it may succeed.
1600 adapter->num_msix_vectors = 0;
1605 * ixgbevf_request_irq - initialize interrupts
1606 * @adapter: board private structure
1608 * Attempts to configure interrupts using the best available
1609 * capabilities of the hardware and kernel.
1611 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1613 int err = ixgbevf_request_msix_irqs(adapter);
1616 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1621 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1625 if (!adapter->msix_entries)
1628 q_vectors = adapter->num_msix_vectors;
1631 free_irq(adapter->msix_entries[i].vector, adapter);
1634 for (; i >= 0; i--) {
1635 /* free only the irqs that were actually requested */
1636 if (!adapter->q_vector[i]->rx.ring &&
1637 !adapter->q_vector[i]->tx.ring)
1640 free_irq(adapter->msix_entries[i].vector,
1641 adapter->q_vector[i]);
1646 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1647 * @adapter: board private structure
1649 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1651 struct ixgbe_hw *hw = &adapter->hw;
1654 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1655 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1656 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1658 IXGBE_WRITE_FLUSH(hw);
1660 for (i = 0; i < adapter->num_msix_vectors; i++)
1661 synchronize_irq(adapter->msix_entries[i].vector);
1665 * ixgbevf_irq_enable - Enable default interrupt generation settings
1666 * @adapter: board private structure
1668 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1670 struct ixgbe_hw *hw = &adapter->hw;
1672 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1673 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1674 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1678 * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1679 * @adapter: board private structure
1680 * @ring: structure containing ring specific data
1682 * Configure the Tx descriptor ring after a reset.
1684 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1685 struct ixgbevf_ring *ring)
1687 struct ixgbe_hw *hw = &adapter->hw;
1688 u64 tdba = ring->dma;
1690 u32 txdctl = IXGBE_TXDCTL_ENABLE;
1691 u8 reg_idx = ring->reg_idx;
1693 /* disable queue to avoid issues while updating state */
1694 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1695 IXGBE_WRITE_FLUSH(hw);
1697 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1698 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1699 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1700 ring->count * sizeof(union ixgbe_adv_tx_desc));
1702 /* disable head writeback */
1703 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1704 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1706 /* enable relaxed ordering */
1707 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1708 (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1709 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1711 /* reset head and tail pointers */
1712 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1713 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1714 ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1716 /* reset ntu and ntc to place SW in sync with hardwdare */
1717 ring->next_to_clean = 0;
1718 ring->next_to_use = 0;
1720 /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1721 * to or less than the number of on chip descriptors, which is
1724 txdctl |= (8 << 16); /* WTHRESH = 8 */
1726 /* Setting PTHRESH to 32 both improves performance */
1727 txdctl |= (1u << 8) | /* HTHRESH = 1 */
1728 32; /* PTHRESH = 32 */
1730 /* reinitialize tx_buffer_info */
1731 memset(ring->tx_buffer_info, 0,
1732 sizeof(struct ixgbevf_tx_buffer) * ring->count);
1734 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1735 clear_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1737 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1739 /* poll to verify queue is enabled */
1741 usleep_range(1000, 2000);
1742 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1743 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1745 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1749 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1750 * @adapter: board private structure
1752 * Configure the Tx unit of the MAC after a reset.
1754 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1758 /* Setup the HW Tx Head and Tail descriptor pointers */
1759 for (i = 0; i < adapter->num_tx_queues; i++)
1760 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1761 for (i = 0; i < adapter->num_xdp_queues; i++)
1762 ixgbevf_configure_tx_ring(adapter, adapter->xdp_ring[i]);
1765 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1767 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter,
1768 struct ixgbevf_ring *ring, int index)
1770 struct ixgbe_hw *hw = &adapter->hw;
1773 srrctl = IXGBE_SRRCTL_DROP_EN;
1775 srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1776 if (ring_uses_large_buffer(ring))
1777 srrctl |= IXGBEVF_RXBUFFER_3072 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1779 srrctl |= IXGBEVF_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1780 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1782 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1785 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1787 struct ixgbe_hw *hw = &adapter->hw;
1789 /* PSRTYPE must be initialized in 82599 */
1790 u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1791 IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1792 IXGBE_PSRTYPE_L2HDR;
1794 if (adapter->num_rx_queues > 1)
1797 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1800 #define IXGBEVF_MAX_RX_DESC_POLL 10
1801 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1802 struct ixgbevf_ring *ring)
1804 struct ixgbe_hw *hw = &adapter->hw;
1805 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1807 u8 reg_idx = ring->reg_idx;
1809 if (IXGBE_REMOVED(hw->hw_addr))
1811 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1812 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1814 /* write value back with RXDCTL.ENABLE bit cleared */
1815 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1817 /* the hardware may take up to 100us to really disable the Rx queue */
1820 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1821 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1824 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1828 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1829 struct ixgbevf_ring *ring)
1831 struct ixgbe_hw *hw = &adapter->hw;
1832 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1834 u8 reg_idx = ring->reg_idx;
1836 if (IXGBE_REMOVED(hw->hw_addr))
1839 usleep_range(1000, 2000);
1840 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1841 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1844 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1849 * ixgbevf_init_rss_key - Initialize adapter RSS key
1850 * @adapter: device handle
1852 * Allocates and initializes the RSS key if it is not allocated.
1854 static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter)
1858 if (!adapter->rss_key) {
1859 rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL);
1860 if (unlikely(!rss_key))
1863 netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE);
1864 adapter->rss_key = rss_key;
1870 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1872 struct ixgbe_hw *hw = &adapter->hw;
1873 u32 vfmrqc = 0, vfreta = 0;
1874 u16 rss_i = adapter->num_rx_queues;
1877 /* Fill out hash function seeds */
1878 for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1879 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i));
1881 for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1885 adapter->rss_indir_tbl[i] = j;
1887 vfreta |= j << (i & 0x3) * 8;
1889 IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1894 /* Perform hash on these packet types */
1895 vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1896 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1897 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1898 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1900 vfmrqc |= IXGBE_VFMRQC_RSSEN;
1902 IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1905 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1906 struct ixgbevf_ring *ring)
1908 struct ixgbe_hw *hw = &adapter->hw;
1909 union ixgbe_adv_rx_desc *rx_desc;
1910 u64 rdba = ring->dma;
1912 u8 reg_idx = ring->reg_idx;
1914 /* disable queue to avoid issues while updating state */
1915 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1916 ixgbevf_disable_rx_queue(adapter, ring);
1918 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1919 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1920 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1921 ring->count * sizeof(union ixgbe_adv_rx_desc));
1923 #ifndef CONFIG_SPARC
1924 /* enable relaxed ordering */
1925 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1926 IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1928 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1929 IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1930 IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1933 /* reset head and tail pointers */
1934 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1935 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1936 ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1938 /* initialize rx_buffer_info */
1939 memset(ring->rx_buffer_info, 0,
1940 sizeof(struct ixgbevf_rx_buffer) * ring->count);
1942 /* initialize Rx descriptor 0 */
1943 rx_desc = IXGBEVF_RX_DESC(ring, 0);
1944 rx_desc->wb.upper.length = 0;
1946 /* reset ntu and ntc to place SW in sync with hardwdare */
1947 ring->next_to_clean = 0;
1948 ring->next_to_use = 0;
1949 ring->next_to_alloc = 0;
1951 ixgbevf_configure_srrctl(adapter, ring, reg_idx);
1953 /* RXDCTL.RLPML does not work on 82599 */
1954 if (adapter->hw.mac.type != ixgbe_mac_82599_vf) {
1955 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
1956 IXGBE_RXDCTL_RLPML_EN);
1958 #if (PAGE_SIZE < 8192)
1959 /* Limit the maximum frame size so we don't overrun the skb */
1960 if (ring_uses_build_skb(ring) &&
1961 !ring_uses_large_buffer(ring))
1962 rxdctl |= IXGBEVF_MAX_FRAME_BUILD_SKB |
1963 IXGBE_RXDCTL_RLPML_EN;
1967 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1968 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1970 ixgbevf_rx_desc_queue_enable(adapter, ring);
1971 ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1974 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter,
1975 struct ixgbevf_ring *rx_ring)
1977 struct net_device *netdev = adapter->netdev;
1978 unsigned int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1980 /* set build_skb and buffer size flags */
1981 clear_ring_build_skb_enabled(rx_ring);
1982 clear_ring_uses_large_buffer(rx_ring);
1984 if (adapter->flags & IXGBEVF_FLAGS_LEGACY_RX)
1987 set_ring_build_skb_enabled(rx_ring);
1989 if (PAGE_SIZE < 8192) {
1990 if (max_frame <= IXGBEVF_MAX_FRAME_BUILD_SKB)
1993 set_ring_uses_large_buffer(rx_ring);
1998 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1999 * @adapter: board private structure
2001 * Configure the Rx unit of the MAC after a reset.
2003 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
2005 struct ixgbe_hw *hw = &adapter->hw;
2006 struct net_device *netdev = adapter->netdev;
2009 ixgbevf_setup_psrtype(adapter);
2010 if (hw->mac.type >= ixgbe_mac_X550_vf)
2011 ixgbevf_setup_vfmrqc(adapter);
2013 spin_lock_bh(&adapter->mbx_lock);
2014 /* notify the PF of our intent to use this size of frame */
2015 ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
2016 spin_unlock_bh(&adapter->mbx_lock);
2018 dev_err(&adapter->pdev->dev,
2019 "Failed to set MTU at %d\n", netdev->mtu);
2021 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2022 * the Base and Length of the Rx Descriptor Ring
2024 for (i = 0; i < adapter->num_rx_queues; i++) {
2025 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
2027 ixgbevf_set_rx_buffer_len(adapter, rx_ring);
2028 ixgbevf_configure_rx_ring(adapter, rx_ring);
2032 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
2033 __be16 proto, u16 vid)
2035 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2036 struct ixgbe_hw *hw = &adapter->hw;
2039 spin_lock_bh(&adapter->mbx_lock);
2041 /* add VID to filter table */
2042 err = hw->mac.ops.set_vfta(hw, vid, 0, true);
2044 spin_unlock_bh(&adapter->mbx_lock);
2046 /* translate error return types so error makes sense */
2047 if (err == IXGBE_ERR_MBX)
2050 if (err == IXGBE_ERR_INVALID_ARGUMENT)
2053 set_bit(vid, adapter->active_vlans);
2058 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
2059 __be16 proto, u16 vid)
2061 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2062 struct ixgbe_hw *hw = &adapter->hw;
2065 spin_lock_bh(&adapter->mbx_lock);
2067 /* remove VID from filter table */
2068 err = hw->mac.ops.set_vfta(hw, vid, 0, false);
2070 spin_unlock_bh(&adapter->mbx_lock);
2072 clear_bit(vid, adapter->active_vlans);
2077 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
2081 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2082 ixgbevf_vlan_rx_add_vid(adapter->netdev,
2083 htons(ETH_P_8021Q), vid);
2086 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
2088 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2089 struct ixgbe_hw *hw = &adapter->hw;
2092 if (!netdev_uc_empty(netdev)) {
2093 struct netdev_hw_addr *ha;
2095 netdev_for_each_uc_addr(ha, netdev) {
2096 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
2100 /* If the list is empty then send message to PF driver to
2101 * clear all MAC VLANs on this VF.
2103 hw->mac.ops.set_uc_addr(hw, 0, NULL);
2110 * ixgbevf_set_rx_mode - Multicast and unicast set
2111 * @netdev: network interface device structure
2113 * The set_rx_method entry point is called whenever the multicast address
2114 * list, unicast address list or the network interface flags are updated.
2115 * This routine is responsible for configuring the hardware for proper
2116 * multicast mode and configuring requested unicast filters.
2118 static void ixgbevf_set_rx_mode(struct net_device *netdev)
2120 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2121 struct ixgbe_hw *hw = &adapter->hw;
2122 unsigned int flags = netdev->flags;
2125 /* request the most inclusive mode we need */
2126 if (flags & IFF_PROMISC)
2127 xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
2128 else if (flags & IFF_ALLMULTI)
2129 xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
2130 else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
2131 xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
2133 xcast_mode = IXGBEVF_XCAST_MODE_NONE;
2135 spin_lock_bh(&adapter->mbx_lock);
2137 hw->mac.ops.update_xcast_mode(hw, xcast_mode);
2139 /* reprogram multicast list */
2140 hw->mac.ops.update_mc_addr_list(hw, netdev);
2142 ixgbevf_write_uc_addr_list(netdev);
2144 spin_unlock_bh(&adapter->mbx_lock);
2147 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
2150 struct ixgbevf_q_vector *q_vector;
2151 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2153 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2154 q_vector = adapter->q_vector[q_idx];
2155 napi_enable(&q_vector->napi);
2159 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
2162 struct ixgbevf_q_vector *q_vector;
2163 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2165 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2166 q_vector = adapter->q_vector[q_idx];
2167 napi_disable(&q_vector->napi);
2171 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
2173 struct ixgbe_hw *hw = &adapter->hw;
2174 unsigned int def_q = 0;
2175 unsigned int num_tcs = 0;
2176 unsigned int num_rx_queues = adapter->num_rx_queues;
2177 unsigned int num_tx_queues = adapter->num_tx_queues;
2180 spin_lock_bh(&adapter->mbx_lock);
2182 /* fetch queue configuration from the PF */
2183 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2185 spin_unlock_bh(&adapter->mbx_lock);
2191 /* we need only one Tx queue */
2194 /* update default Tx ring register index */
2195 adapter->tx_ring[0]->reg_idx = def_q;
2197 /* we need as many queues as traffic classes */
2198 num_rx_queues = num_tcs;
2201 /* if we have a bad config abort request queue reset */
2202 if ((adapter->num_rx_queues != num_rx_queues) ||
2203 (adapter->num_tx_queues != num_tx_queues)) {
2204 /* force mailbox timeout to prevent further messages */
2205 hw->mbx.timeout = 0;
2207 /* wait for watchdog to come around and bail us out */
2208 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2214 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2216 ixgbevf_configure_dcb(adapter);
2218 ixgbevf_set_rx_mode(adapter->netdev);
2220 ixgbevf_restore_vlan(adapter);
2221 ixgbevf_ipsec_restore(adapter);
2223 ixgbevf_configure_tx(adapter);
2224 ixgbevf_configure_rx(adapter);
2227 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2229 /* Only save pre-reset stats if there are some */
2230 if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2231 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2232 adapter->stats.base_vfgprc;
2233 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2234 adapter->stats.base_vfgptc;
2235 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2236 adapter->stats.base_vfgorc;
2237 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2238 adapter->stats.base_vfgotc;
2239 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2240 adapter->stats.base_vfmprc;
2244 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2246 struct ixgbe_hw *hw = &adapter->hw;
2248 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2249 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2250 adapter->stats.last_vfgorc |=
2251 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2252 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2253 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2254 adapter->stats.last_vfgotc |=
2255 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2256 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2258 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2259 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2260 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2261 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2262 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2265 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2267 struct ixgbe_hw *hw = &adapter->hw;
2268 static const int api[] = {
2274 ixgbe_mbox_api_unknown
2278 spin_lock_bh(&adapter->mbx_lock);
2280 while (api[idx] != ixgbe_mbox_api_unknown) {
2281 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2287 spin_unlock_bh(&adapter->mbx_lock);
2290 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2292 struct net_device *netdev = adapter->netdev;
2293 struct ixgbe_hw *hw = &adapter->hw;
2295 ixgbevf_configure_msix(adapter);
2297 spin_lock_bh(&adapter->mbx_lock);
2299 if (is_valid_ether_addr(hw->mac.addr))
2300 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2302 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2304 spin_unlock_bh(&adapter->mbx_lock);
2306 smp_mb__before_atomic();
2307 clear_bit(__IXGBEVF_DOWN, &adapter->state);
2308 ixgbevf_napi_enable_all(adapter);
2310 /* clear any pending interrupts, may auto mask */
2311 IXGBE_READ_REG(hw, IXGBE_VTEICR);
2312 ixgbevf_irq_enable(adapter);
2314 /* enable transmits */
2315 netif_tx_start_all_queues(netdev);
2317 ixgbevf_save_reset_stats(adapter);
2318 ixgbevf_init_last_counter_stats(adapter);
2320 hw->mac.get_link_status = 1;
2321 mod_timer(&adapter->service_timer, jiffies);
2324 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2326 ixgbevf_configure(adapter);
2328 ixgbevf_up_complete(adapter);
2332 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2333 * @rx_ring: ring to free buffers from
2335 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2337 u16 i = rx_ring->next_to_clean;
2339 /* Free Rx ring sk_buff */
2341 dev_kfree_skb(rx_ring->skb);
2342 rx_ring->skb = NULL;
2345 /* Free all the Rx ring pages */
2346 while (i != rx_ring->next_to_alloc) {
2347 struct ixgbevf_rx_buffer *rx_buffer;
2349 rx_buffer = &rx_ring->rx_buffer_info[i];
2351 /* Invalidate cache lines that may have been written to by
2352 * device so that we avoid corrupting memory.
2354 dma_sync_single_range_for_cpu(rx_ring->dev,
2356 rx_buffer->page_offset,
2357 ixgbevf_rx_bufsz(rx_ring),
2360 /* free resources associated with mapping */
2361 dma_unmap_page_attrs(rx_ring->dev,
2363 ixgbevf_rx_pg_size(rx_ring),
2365 IXGBEVF_RX_DMA_ATTR);
2367 __page_frag_cache_drain(rx_buffer->page,
2368 rx_buffer->pagecnt_bias);
2371 if (i == rx_ring->count)
2375 rx_ring->next_to_alloc = 0;
2376 rx_ring->next_to_clean = 0;
2377 rx_ring->next_to_use = 0;
2381 * ixgbevf_clean_tx_ring - Free Tx Buffers
2382 * @tx_ring: ring to be cleaned
2384 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2386 u16 i = tx_ring->next_to_clean;
2387 struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
2389 while (i != tx_ring->next_to_use) {
2390 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
2392 /* Free all the Tx ring sk_buffs */
2393 if (ring_is_xdp(tx_ring))
2394 page_frag_free(tx_buffer->data);
2396 dev_kfree_skb_any(tx_buffer->skb);
2398 /* unmap skb header data */
2399 dma_unmap_single(tx_ring->dev,
2400 dma_unmap_addr(tx_buffer, dma),
2401 dma_unmap_len(tx_buffer, len),
2404 /* check for eop_desc to determine the end of the packet */
2405 eop_desc = tx_buffer->next_to_watch;
2406 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2408 /* unmap remaining buffers */
2409 while (tx_desc != eop_desc) {
2413 if (unlikely(i == tx_ring->count)) {
2415 tx_buffer = tx_ring->tx_buffer_info;
2416 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
2419 /* unmap any remaining paged data */
2420 if (dma_unmap_len(tx_buffer, len))
2421 dma_unmap_page(tx_ring->dev,
2422 dma_unmap_addr(tx_buffer, dma),
2423 dma_unmap_len(tx_buffer, len),
2427 /* move us one more past the eop_desc for start of next pkt */
2430 if (unlikely(i == tx_ring->count)) {
2432 tx_buffer = tx_ring->tx_buffer_info;
2436 /* reset next_to_use and next_to_clean */
2437 tx_ring->next_to_use = 0;
2438 tx_ring->next_to_clean = 0;
2443 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2444 * @adapter: board private structure
2446 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2450 for (i = 0; i < adapter->num_rx_queues; i++)
2451 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2455 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2456 * @adapter: board private structure
2458 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2462 for (i = 0; i < adapter->num_tx_queues; i++)
2463 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2464 for (i = 0; i < adapter->num_xdp_queues; i++)
2465 ixgbevf_clean_tx_ring(adapter->xdp_ring[i]);
2468 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2470 struct net_device *netdev = adapter->netdev;
2471 struct ixgbe_hw *hw = &adapter->hw;
2474 /* signal that we are down to the interrupt handler */
2475 if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2476 return; /* do nothing if already down */
2478 /* disable all enabled Rx queues */
2479 for (i = 0; i < adapter->num_rx_queues; i++)
2480 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2482 usleep_range(10000, 20000);
2484 netif_tx_stop_all_queues(netdev);
2486 /* call carrier off first to avoid false dev_watchdog timeouts */
2487 netif_carrier_off(netdev);
2488 netif_tx_disable(netdev);
2490 ixgbevf_irq_disable(adapter);
2492 ixgbevf_napi_disable_all(adapter);
2494 del_timer_sync(&adapter->service_timer);
2496 /* disable transmits in the hardware now that interrupts are off */
2497 for (i = 0; i < adapter->num_tx_queues; i++) {
2498 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2500 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2501 IXGBE_TXDCTL_SWFLSH);
2504 for (i = 0; i < adapter->num_xdp_queues; i++) {
2505 u8 reg_idx = adapter->xdp_ring[i]->reg_idx;
2507 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2508 IXGBE_TXDCTL_SWFLSH);
2511 if (!pci_channel_offline(adapter->pdev))
2512 ixgbevf_reset(adapter);
2514 ixgbevf_clean_all_tx_rings(adapter);
2515 ixgbevf_clean_all_rx_rings(adapter);
2518 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2520 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2523 ixgbevf_down(adapter);
2524 pci_set_master(adapter->pdev);
2525 ixgbevf_up(adapter);
2527 clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2530 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2532 struct ixgbe_hw *hw = &adapter->hw;
2533 struct net_device *netdev = adapter->netdev;
2535 if (hw->mac.ops.reset_hw(hw)) {
2536 hw_dbg(hw, "PF still resetting\n");
2538 hw->mac.ops.init_hw(hw);
2539 ixgbevf_negotiate_api(adapter);
2542 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2543 ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2544 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2547 adapter->last_reset = jiffies;
2550 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2553 int vector_threshold;
2555 /* We'll want at least 2 (vector_threshold):
2556 * 1) TxQ[0] + RxQ[0] handler
2557 * 2) Other (Link Status Change, etc.)
2559 vector_threshold = MIN_MSIX_COUNT;
2561 /* The more we get, the more we will assign to Tx/Rx Cleanup
2562 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2563 * Right now, we simply care about how many we'll get; we'll
2564 * set them up later while requesting irq's.
2566 vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2567 vector_threshold, vectors);
2570 dev_err(&adapter->pdev->dev,
2571 "Unable to allocate MSI-X interrupts\n");
2572 kfree(adapter->msix_entries);
2573 adapter->msix_entries = NULL;
2577 /* Adjust for only the vectors we'll use, which is minimum
2578 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2579 * vectors we were allocated.
2581 adapter->num_msix_vectors = vectors;
2587 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2588 * @adapter: board private structure to initialize
2590 * This is the top level queue allocation routine. The order here is very
2591 * important, starting with the "most" number of features turned on at once,
2592 * and ending with the smallest set of features. This way large combinations
2593 * can be allocated if they're turned on, and smaller combinations are the
2594 * fall through conditions.
2597 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2599 struct ixgbe_hw *hw = &adapter->hw;
2600 unsigned int def_q = 0;
2601 unsigned int num_tcs = 0;
2604 /* Start with base case */
2605 adapter->num_rx_queues = 1;
2606 adapter->num_tx_queues = 1;
2607 adapter->num_xdp_queues = 0;
2609 spin_lock_bh(&adapter->mbx_lock);
2611 /* fetch queue configuration from the PF */
2612 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2614 spin_unlock_bh(&adapter->mbx_lock);
2619 /* we need as many queues as traffic classes */
2621 adapter->num_rx_queues = num_tcs;
2623 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2625 switch (hw->api_version) {
2626 case ixgbe_mbox_api_11:
2627 case ixgbe_mbox_api_12:
2628 case ixgbe_mbox_api_13:
2629 case ixgbe_mbox_api_14:
2630 if (adapter->xdp_prog &&
2631 hw->mac.max_tx_queues == rss)
2632 rss = rss > 3 ? 2 : 1;
2634 adapter->num_rx_queues = rss;
2635 adapter->num_tx_queues = rss;
2636 adapter->num_xdp_queues = adapter->xdp_prog ? rss : 0;
2645 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2646 * @adapter: board private structure to initialize
2648 * Attempt to configure the interrupts using the best available
2649 * capabilities of the hardware and the kernel.
2651 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2653 int vector, v_budget;
2655 /* It's easy to be greedy for MSI-X vectors, but it really
2656 * doesn't do us much good if we have a lot more vectors
2657 * than CPU's. So let's be conservative and only ask for
2658 * (roughly) the same number of vectors as there are CPU's.
2659 * The default is to use pairs of vectors.
2661 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2662 v_budget = min_t(int, v_budget, num_online_cpus());
2663 v_budget += NON_Q_VECTORS;
2665 adapter->msix_entries = kcalloc(v_budget,
2666 sizeof(struct msix_entry), GFP_KERNEL);
2667 if (!adapter->msix_entries)
2670 for (vector = 0; vector < v_budget; vector++)
2671 adapter->msix_entries[vector].entry = vector;
2673 /* A failure in MSI-X entry allocation isn't fatal, but the VF driver
2674 * does not support any other modes, so we will simply fail here. Note
2675 * that we clean up the msix_entries pointer else-where.
2677 return ixgbevf_acquire_msix_vectors(adapter, v_budget);
2680 static void ixgbevf_add_ring(struct ixgbevf_ring *ring,
2681 struct ixgbevf_ring_container *head)
2683 ring->next = head->ring;
2689 * ixgbevf_alloc_q_vector - Allocate memory for a single interrupt vector
2690 * @adapter: board private structure to initialize
2691 * @v_idx: index of vector in adapter struct
2692 * @txr_count: number of Tx rings for q vector
2693 * @txr_idx: index of first Tx ring to assign
2694 * @xdp_count: total number of XDP rings to allocate
2695 * @xdp_idx: index of first XDP ring to allocate
2696 * @rxr_count: number of Rx rings for q vector
2697 * @rxr_idx: index of first Rx ring to assign
2699 * We allocate one q_vector. If allocation fails we return -ENOMEM.
2701 static int ixgbevf_alloc_q_vector(struct ixgbevf_adapter *adapter, int v_idx,
2702 int txr_count, int txr_idx,
2703 int xdp_count, int xdp_idx,
2704 int rxr_count, int rxr_idx)
2706 struct ixgbevf_q_vector *q_vector;
2707 int reg_idx = txr_idx + xdp_idx;
2708 struct ixgbevf_ring *ring;
2709 int ring_count, size;
2711 ring_count = txr_count + xdp_count + rxr_count;
2712 size = sizeof(*q_vector) + (sizeof(*ring) * ring_count);
2714 /* allocate q_vector and rings */
2715 q_vector = kzalloc(size, GFP_KERNEL);
2719 /* initialize NAPI */
2720 netif_napi_add(adapter->netdev, &q_vector->napi, ixgbevf_poll, 64);
2722 /* tie q_vector and adapter together */
2723 adapter->q_vector[v_idx] = q_vector;
2724 q_vector->adapter = adapter;
2725 q_vector->v_idx = v_idx;
2727 /* initialize pointer to rings */
2728 ring = q_vector->ring;
2731 /* assign generic ring traits */
2732 ring->dev = &adapter->pdev->dev;
2733 ring->netdev = adapter->netdev;
2735 /* configure backlink on ring */
2736 ring->q_vector = q_vector;
2738 /* update q_vector Tx values */
2739 ixgbevf_add_ring(ring, &q_vector->tx);
2741 /* apply Tx specific ring traits */
2742 ring->count = adapter->tx_ring_count;
2743 ring->queue_index = txr_idx;
2744 ring->reg_idx = reg_idx;
2746 /* assign ring to adapter */
2747 adapter->tx_ring[txr_idx] = ring;
2749 /* update count and index */
2754 /* push pointer to next ring */
2759 /* assign generic ring traits */
2760 ring->dev = &adapter->pdev->dev;
2761 ring->netdev = adapter->netdev;
2763 /* configure backlink on ring */
2764 ring->q_vector = q_vector;
2766 /* update q_vector Tx values */
2767 ixgbevf_add_ring(ring, &q_vector->tx);
2769 /* apply Tx specific ring traits */
2770 ring->count = adapter->tx_ring_count;
2771 ring->queue_index = xdp_idx;
2772 ring->reg_idx = reg_idx;
2775 /* assign ring to adapter */
2776 adapter->xdp_ring[xdp_idx] = ring;
2778 /* update count and index */
2783 /* push pointer to next ring */
2788 /* assign generic ring traits */
2789 ring->dev = &adapter->pdev->dev;
2790 ring->netdev = adapter->netdev;
2792 /* configure backlink on ring */
2793 ring->q_vector = q_vector;
2795 /* update q_vector Rx values */
2796 ixgbevf_add_ring(ring, &q_vector->rx);
2798 /* apply Rx specific ring traits */
2799 ring->count = adapter->rx_ring_count;
2800 ring->queue_index = rxr_idx;
2801 ring->reg_idx = rxr_idx;
2803 /* assign ring to adapter */
2804 adapter->rx_ring[rxr_idx] = ring;
2806 /* update count and index */
2810 /* push pointer to next ring */
2818 * ixgbevf_free_q_vector - Free memory allocated for specific interrupt vector
2819 * @adapter: board private structure to initialize
2820 * @v_idx: index of vector in adapter struct
2822 * This function frees the memory allocated to the q_vector. In addition if
2823 * NAPI is enabled it will delete any references to the NAPI struct prior
2824 * to freeing the q_vector.
2826 static void ixgbevf_free_q_vector(struct ixgbevf_adapter *adapter, int v_idx)
2828 struct ixgbevf_q_vector *q_vector = adapter->q_vector[v_idx];
2829 struct ixgbevf_ring *ring;
2831 ixgbevf_for_each_ring(ring, q_vector->tx) {
2832 if (ring_is_xdp(ring))
2833 adapter->xdp_ring[ring->queue_index] = NULL;
2835 adapter->tx_ring[ring->queue_index] = NULL;
2838 ixgbevf_for_each_ring(ring, q_vector->rx)
2839 adapter->rx_ring[ring->queue_index] = NULL;
2841 adapter->q_vector[v_idx] = NULL;
2842 netif_napi_del(&q_vector->napi);
2844 /* ixgbevf_get_stats() might access the rings on this vector,
2845 * we must wait a grace period before freeing it.
2847 kfree_rcu(q_vector, rcu);
2851 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2852 * @adapter: board private structure to initialize
2854 * We allocate one q_vector per queue interrupt. If allocation fails we
2857 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2859 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2860 int rxr_remaining = adapter->num_rx_queues;
2861 int txr_remaining = adapter->num_tx_queues;
2862 int xdp_remaining = adapter->num_xdp_queues;
2863 int rxr_idx = 0, txr_idx = 0, xdp_idx = 0, v_idx = 0;
2866 if (q_vectors >= (rxr_remaining + txr_remaining + xdp_remaining)) {
2867 for (; rxr_remaining; v_idx++, q_vectors--) {
2868 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2870 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2871 0, 0, 0, 0, rqpv, rxr_idx);
2875 /* update counts and index */
2876 rxr_remaining -= rqpv;
2881 for (; q_vectors; v_idx++, q_vectors--) {
2882 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2883 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors);
2884 int xqpv = DIV_ROUND_UP(xdp_remaining, q_vectors);
2886 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2894 /* update counts and index */
2895 rxr_remaining -= rqpv;
2897 txr_remaining -= tqpv;
2899 xdp_remaining -= xqpv;
2908 ixgbevf_free_q_vector(adapter, v_idx);
2915 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2916 * @adapter: board private structure to initialize
2918 * This function frees the memory allocated to the q_vectors. In addition if
2919 * NAPI is enabled it will delete any references to the NAPI struct prior
2920 * to freeing the q_vector.
2922 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2924 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2928 ixgbevf_free_q_vector(adapter, q_vectors);
2933 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2934 * @adapter: board private structure
2937 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2939 if (!adapter->msix_entries)
2942 pci_disable_msix(adapter->pdev);
2943 kfree(adapter->msix_entries);
2944 adapter->msix_entries = NULL;
2948 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2949 * @adapter: board private structure to initialize
2952 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2956 /* Number of supported queues */
2957 ixgbevf_set_num_queues(adapter);
2959 err = ixgbevf_set_interrupt_capability(adapter);
2961 hw_dbg(&adapter->hw,
2962 "Unable to setup interrupt capabilities\n");
2963 goto err_set_interrupt;
2966 err = ixgbevf_alloc_q_vectors(adapter);
2968 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2969 goto err_alloc_q_vectors;
2972 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u XDP Queue count %u\n",
2973 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
2974 adapter->num_rx_queues, adapter->num_tx_queues,
2975 adapter->num_xdp_queues);
2977 set_bit(__IXGBEVF_DOWN, &adapter->state);
2980 err_alloc_q_vectors:
2981 ixgbevf_reset_interrupt_capability(adapter);
2987 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2988 * @adapter: board private structure to clear interrupt scheme on
2990 * We go through and clear interrupt specific resources and reset the structure
2991 * to pre-load conditions
2993 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2995 adapter->num_tx_queues = 0;
2996 adapter->num_xdp_queues = 0;
2997 adapter->num_rx_queues = 0;
2999 ixgbevf_free_q_vectors(adapter);
3000 ixgbevf_reset_interrupt_capability(adapter);
3004 * ixgbevf_sw_init - Initialize general software structures
3005 * @adapter: board private structure to initialize
3007 * ixgbevf_sw_init initializes the Adapter private data structure.
3008 * Fields are initialized based on PCI device information and
3009 * OS network device settings (MTU size).
3011 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
3013 struct ixgbe_hw *hw = &adapter->hw;
3014 struct pci_dev *pdev = adapter->pdev;
3015 struct net_device *netdev = adapter->netdev;
3018 /* PCI config space info */
3019 hw->vendor_id = pdev->vendor;
3020 hw->device_id = pdev->device;
3021 hw->revision_id = pdev->revision;
3022 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3023 hw->subsystem_device_id = pdev->subsystem_device;
3025 hw->mbx.ops.init_params(hw);
3027 if (hw->mac.type >= ixgbe_mac_X550_vf) {
3028 err = ixgbevf_init_rss_key(adapter);
3033 /* assume legacy case in which PF would only give VF 2 queues */
3034 hw->mac.max_tx_queues = 2;
3035 hw->mac.max_rx_queues = 2;
3037 /* lock to protect mailbox accesses */
3038 spin_lock_init(&adapter->mbx_lock);
3040 err = hw->mac.ops.reset_hw(hw);
3042 dev_info(&pdev->dev,
3043 "PF still in reset state. Is the PF interface up?\n");
3045 err = hw->mac.ops.init_hw(hw);
3047 pr_err("init_shared_code failed: %d\n", err);
3050 ixgbevf_negotiate_api(adapter);
3051 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
3053 dev_info(&pdev->dev, "Error reading MAC address\n");
3054 else if (is_zero_ether_addr(adapter->hw.mac.addr))
3055 dev_info(&pdev->dev,
3056 "MAC address not assigned by administrator.\n");
3057 ether_addr_copy(netdev->dev_addr, hw->mac.addr);
3060 if (!is_valid_ether_addr(netdev->dev_addr)) {
3061 dev_info(&pdev->dev, "Assigning random MAC address\n");
3062 eth_hw_addr_random(netdev);
3063 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
3064 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
3067 /* Enable dynamic interrupt throttling rates */
3068 adapter->rx_itr_setting = 1;
3069 adapter->tx_itr_setting = 1;
3071 /* set default ring sizes */
3072 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
3073 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
3075 set_bit(__IXGBEVF_DOWN, &adapter->state);
3082 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
3084 u32 current_counter = IXGBE_READ_REG(hw, reg); \
3085 if (current_counter < last_counter) \
3086 counter += 0x100000000LL; \
3087 last_counter = current_counter; \
3088 counter &= 0xFFFFFFFF00000000LL; \
3089 counter |= current_counter; \
3092 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
3094 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
3095 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
3096 u64 current_counter = (current_counter_msb << 32) | \
3097 current_counter_lsb; \
3098 if (current_counter < last_counter) \
3099 counter += 0x1000000000LL; \
3100 last_counter = current_counter; \
3101 counter &= 0xFFFFFFF000000000LL; \
3102 counter |= current_counter; \
3105 * ixgbevf_update_stats - Update the board statistics counters.
3106 * @adapter: board private structure
3108 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
3110 struct ixgbe_hw *hw = &adapter->hw;
3111 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
3112 u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
3115 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3116 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3119 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
3120 adapter->stats.vfgprc);
3121 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
3122 adapter->stats.vfgptc);
3123 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
3124 adapter->stats.last_vfgorc,
3125 adapter->stats.vfgorc);
3126 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
3127 adapter->stats.last_vfgotc,
3128 adapter->stats.vfgotc);
3129 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
3130 adapter->stats.vfmprc);
3132 for (i = 0; i < adapter->num_rx_queues; i++) {
3133 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
3135 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
3136 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
3137 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
3138 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
3141 adapter->hw_csum_rx_error = hw_csum_rx_error;
3142 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
3143 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
3144 adapter->alloc_rx_page = alloc_rx_page;
3148 * ixgbevf_service_timer - Timer Call-back
3149 * @t: pointer to timer_list struct
3151 static void ixgbevf_service_timer(struct timer_list *t)
3153 struct ixgbevf_adapter *adapter = from_timer(adapter, t,
3156 /* Reset the timer */
3157 mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
3159 ixgbevf_service_event_schedule(adapter);
3162 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
3164 if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
3168 /* If we're already down or resetting, just bail */
3169 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3170 test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
3171 test_bit(__IXGBEVF_RESETTING, &adapter->state)) {
3176 adapter->tx_timeout_count++;
3178 ixgbevf_reinit_locked(adapter);
3183 * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
3184 * @adapter: pointer to the device adapter structure
3186 * This function serves two purposes. First it strobes the interrupt lines
3187 * in order to make certain interrupts are occurring. Secondly it sets the
3188 * bits needed to check for TX hangs. As a result we should immediately
3189 * determine if a hang has occurred.
3191 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
3193 struct ixgbe_hw *hw = &adapter->hw;
3197 /* If we're down or resetting, just bail */
3198 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3199 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3202 /* Force detection of hung controller */
3203 if (netif_carrier_ok(adapter->netdev)) {
3204 for (i = 0; i < adapter->num_tx_queues; i++)
3205 set_check_for_tx_hang(adapter->tx_ring[i]);
3206 for (i = 0; i < adapter->num_xdp_queues; i++)
3207 set_check_for_tx_hang(adapter->xdp_ring[i]);
3210 /* get one bit for every active Tx/Rx interrupt vector */
3211 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
3212 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
3214 if (qv->rx.ring || qv->tx.ring)
3218 /* Cause software interrupt to ensure rings are cleaned */
3219 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
3223 * ixgbevf_watchdog_update_link - update the link status
3224 * @adapter: pointer to the device adapter structure
3226 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
3228 struct ixgbe_hw *hw = &adapter->hw;
3229 u32 link_speed = adapter->link_speed;
3230 bool link_up = adapter->link_up;
3233 spin_lock_bh(&adapter->mbx_lock);
3235 err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3237 spin_unlock_bh(&adapter->mbx_lock);
3239 /* if check for link returns error we will need to reset */
3240 if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
3241 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
3245 adapter->link_up = link_up;
3246 adapter->link_speed = link_speed;
3250 * ixgbevf_watchdog_link_is_up - update netif_carrier status and
3251 * print link up message
3252 * @adapter: pointer to the device adapter structure
3254 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
3256 struct net_device *netdev = adapter->netdev;
3258 /* only continue if link was previously down */
3259 if (netif_carrier_ok(netdev))
3262 dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
3263 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
3265 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
3267 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
3271 netif_carrier_on(netdev);
3275 * ixgbevf_watchdog_link_is_down - update netif_carrier status and
3276 * print link down message
3277 * @adapter: pointer to the adapter structure
3279 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
3281 struct net_device *netdev = adapter->netdev;
3283 adapter->link_speed = 0;
3285 /* only continue if link was up previously */
3286 if (!netif_carrier_ok(netdev))
3289 dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
3291 netif_carrier_off(netdev);
3295 * ixgbevf_watchdog_subtask - worker thread to bring link up
3296 * @adapter: board private structure
3298 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
3300 /* if interface is down do nothing */
3301 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3302 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3305 ixgbevf_watchdog_update_link(adapter);
3307 if (adapter->link_up)
3308 ixgbevf_watchdog_link_is_up(adapter);
3310 ixgbevf_watchdog_link_is_down(adapter);
3312 ixgbevf_update_stats(adapter);
3316 * ixgbevf_service_task - manages and runs subtasks
3317 * @work: pointer to work_struct containing our data
3319 static void ixgbevf_service_task(struct work_struct *work)
3321 struct ixgbevf_adapter *adapter = container_of(work,
3322 struct ixgbevf_adapter,
3324 struct ixgbe_hw *hw = &adapter->hw;
3326 if (IXGBE_REMOVED(hw->hw_addr)) {
3327 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
3329 ixgbevf_down(adapter);
3335 ixgbevf_queue_reset_subtask(adapter);
3336 ixgbevf_reset_subtask(adapter);
3337 ixgbevf_watchdog_subtask(adapter);
3338 ixgbevf_check_hang_subtask(adapter);
3340 ixgbevf_service_event_complete(adapter);
3344 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
3345 * @tx_ring: Tx descriptor ring for a specific queue
3347 * Free all transmit software resources
3349 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
3351 ixgbevf_clean_tx_ring(tx_ring);
3353 vfree(tx_ring->tx_buffer_info);
3354 tx_ring->tx_buffer_info = NULL;
3356 /* if not set, then don't free */
3360 dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
3363 tx_ring->desc = NULL;
3367 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
3368 * @adapter: board private structure
3370 * Free all transmit software resources
3372 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
3376 for (i = 0; i < adapter->num_tx_queues; i++)
3377 if (adapter->tx_ring[i]->desc)
3378 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3379 for (i = 0; i < adapter->num_xdp_queues; i++)
3380 if (adapter->xdp_ring[i]->desc)
3381 ixgbevf_free_tx_resources(adapter->xdp_ring[i]);
3385 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
3386 * @tx_ring: Tx descriptor ring (for a specific queue) to setup
3388 * Return 0 on success, negative on failure
3390 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
3392 struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3395 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3396 tx_ring->tx_buffer_info = vmalloc(size);
3397 if (!tx_ring->tx_buffer_info)
3400 u64_stats_init(&tx_ring->syncp);
3402 /* round up to nearest 4K */
3403 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3404 tx_ring->size = ALIGN(tx_ring->size, 4096);
3406 tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3407 &tx_ring->dma, GFP_KERNEL);
3414 vfree(tx_ring->tx_buffer_info);
3415 tx_ring->tx_buffer_info = NULL;
3416 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3421 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3422 * @adapter: board private structure
3424 * If this function returns with an error, then it's possible one or
3425 * more of the rings is populated (while the rest are not). It is the
3426 * callers duty to clean those orphaned rings.
3428 * Return 0 on success, negative on failure
3430 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3432 int i, j = 0, err = 0;
3434 for (i = 0; i < adapter->num_tx_queues; i++) {
3435 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3438 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3442 for (j = 0; j < adapter->num_xdp_queues; j++) {
3443 err = ixgbevf_setup_tx_resources(adapter->xdp_ring[j]);
3446 hw_dbg(&adapter->hw, "Allocation for XDP Queue %u failed\n", j);
3452 /* rewind the index freeing the rings as we go */
3454 ixgbevf_free_tx_resources(adapter->xdp_ring[j]);
3456 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3462 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3463 * @adapter: board private structure
3464 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3466 * Returns 0 on success, negative on failure
3468 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
3469 struct ixgbevf_ring *rx_ring)
3473 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3474 rx_ring->rx_buffer_info = vmalloc(size);
3475 if (!rx_ring->rx_buffer_info)
3478 u64_stats_init(&rx_ring->syncp);
3480 /* Round up to nearest 4K */
3481 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3482 rx_ring->size = ALIGN(rx_ring->size, 4096);
3484 rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3485 &rx_ring->dma, GFP_KERNEL);
3490 /* XDP RX-queue info */
3491 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
3492 rx_ring->queue_index, 0) < 0)
3495 rx_ring->xdp_prog = adapter->xdp_prog;
3499 vfree(rx_ring->rx_buffer_info);
3500 rx_ring->rx_buffer_info = NULL;
3501 dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3506 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3507 * @adapter: board private structure
3509 * If this function returns with an error, then it's possible one or
3510 * more of the rings is populated (while the rest are not). It is the
3511 * callers duty to clean those orphaned rings.
3513 * Return 0 on success, negative on failure
3515 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3519 for (i = 0; i < adapter->num_rx_queues; i++) {
3520 err = ixgbevf_setup_rx_resources(adapter, adapter->rx_ring[i]);
3523 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3529 /* rewind the index freeing the rings as we go */
3531 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3536 * ixgbevf_free_rx_resources - Free Rx Resources
3537 * @rx_ring: ring to clean the resources from
3539 * Free all receive software resources
3541 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3543 ixgbevf_clean_rx_ring(rx_ring);
3545 rx_ring->xdp_prog = NULL;
3546 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
3547 vfree(rx_ring->rx_buffer_info);
3548 rx_ring->rx_buffer_info = NULL;
3550 dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3553 rx_ring->desc = NULL;
3557 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3558 * @adapter: board private structure
3560 * Free all receive software resources
3562 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3566 for (i = 0; i < adapter->num_rx_queues; i++)
3567 if (adapter->rx_ring[i]->desc)
3568 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3572 * ixgbevf_open - Called when a network interface is made active
3573 * @netdev: network interface device structure
3575 * Returns 0 on success, negative value on failure
3577 * The open entry point is called when a network interface is made
3578 * active by the system (IFF_UP). At this point all resources needed
3579 * for transmit and receive operations are allocated, the interrupt
3580 * handler is registered with the OS, the watchdog timer is started,
3581 * and the stack is notified that the interface is ready.
3583 int ixgbevf_open(struct net_device *netdev)
3585 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3586 struct ixgbe_hw *hw = &adapter->hw;
3589 /* A previous failure to open the device because of a lack of
3590 * available MSIX vector resources may have reset the number
3591 * of msix vectors variable to zero. The only way to recover
3592 * is to unload/reload the driver and hope that the system has
3593 * been able to recover some MSIX vector resources.
3595 if (!adapter->num_msix_vectors)
3598 if (hw->adapter_stopped) {
3599 ixgbevf_reset(adapter);
3600 /* if adapter is still stopped then PF isn't up and
3601 * the VF can't start.
3603 if (hw->adapter_stopped) {
3604 err = IXGBE_ERR_MBX;
3605 pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3606 goto err_setup_reset;
3610 /* disallow open during test */
3611 if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3614 netif_carrier_off(netdev);
3616 /* allocate transmit descriptors */
3617 err = ixgbevf_setup_all_tx_resources(adapter);
3621 /* allocate receive descriptors */
3622 err = ixgbevf_setup_all_rx_resources(adapter);
3626 ixgbevf_configure(adapter);
3628 err = ixgbevf_request_irq(adapter);
3632 /* Notify the stack of the actual queue counts. */
3633 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
3635 goto err_set_queues;
3637 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
3639 goto err_set_queues;
3641 ixgbevf_up_complete(adapter);
3646 ixgbevf_free_irq(adapter);
3648 ixgbevf_free_all_rx_resources(adapter);
3650 ixgbevf_free_all_tx_resources(adapter);
3652 ixgbevf_reset(adapter);
3659 * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3660 * @adapter: the private adapter struct
3662 * This function should contain the necessary work common to both suspending
3663 * and closing of the device.
3665 static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3667 ixgbevf_down(adapter);
3668 ixgbevf_free_irq(adapter);
3669 ixgbevf_free_all_tx_resources(adapter);
3670 ixgbevf_free_all_rx_resources(adapter);
3674 * ixgbevf_close - Disables a network interface
3675 * @netdev: network interface device structure
3677 * Returns 0, this is not allowed to fail
3679 * The close entry point is called when an interface is de-activated
3680 * by the OS. The hardware is still under the drivers control, but
3681 * needs to be disabled. A global MAC reset is issued to stop the
3682 * hardware, and all transmit and receive resources are freed.
3684 int ixgbevf_close(struct net_device *netdev)
3686 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3688 if (netif_device_present(netdev))
3689 ixgbevf_close_suspend(adapter);
3694 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3696 struct net_device *dev = adapter->netdev;
3698 if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3702 /* if interface is down do nothing */
3703 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3704 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3707 /* Hardware has to reinitialize queues and interrupts to
3708 * match packet buffer alignment. Unfortunately, the
3709 * hardware is not flexible enough to do this dynamically.
3713 if (netif_running(dev))
3716 ixgbevf_clear_interrupt_scheme(adapter);
3717 ixgbevf_init_interrupt_scheme(adapter);
3719 if (netif_running(dev))
3725 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3726 u32 vlan_macip_lens, u32 fceof_saidx,
3727 u32 type_tucmd, u32 mss_l4len_idx)
3729 struct ixgbe_adv_tx_context_desc *context_desc;
3730 u16 i = tx_ring->next_to_use;
3732 context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3735 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3737 /* set bits to identify this as an advanced context descriptor */
3738 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3740 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3741 context_desc->fceof_saidx = cpu_to_le32(fceof_saidx);
3742 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
3743 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3746 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3747 struct ixgbevf_tx_buffer *first,
3749 struct ixgbevf_ipsec_tx_data *itd)
3751 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3752 struct sk_buff *skb = first->skb;
3762 u32 paylen, l4_offset;
3763 u32 fceof_saidx = 0;
3766 if (skb->ip_summed != CHECKSUM_PARTIAL)
3769 if (!skb_is_gso(skb))
3772 err = skb_cow_head(skb, 0);
3776 if (eth_p_mpls(first->protocol))
3777 ip.hdr = skb_inner_network_header(skb);
3779 ip.hdr = skb_network_header(skb);
3780 l4.hdr = skb_checksum_start(skb);
3782 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3783 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3785 /* initialize outer IP header fields */
3786 if (ip.v4->version == 4) {
3787 unsigned char *csum_start = skb_checksum_start(skb);
3788 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3789 int len = csum_start - trans_start;
3791 /* IP header will have to cancel out any data that
3792 * is not a part of the outer IP header, so set to
3793 * a reverse csum if needed, else init check to 0.
3795 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
3796 csum_fold(csum_partial(trans_start,
3798 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3801 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3802 IXGBE_TX_FLAGS_CSUM |
3803 IXGBE_TX_FLAGS_IPV4;
3805 ip.v6->payload_len = 0;
3806 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3807 IXGBE_TX_FLAGS_CSUM;
3810 /* determine offset of inner transport header */
3811 l4_offset = l4.hdr - skb->data;
3813 /* compute length of segmentation header */
3814 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3816 /* remove payload length from inner checksum */
3817 paylen = skb->len - l4_offset;
3818 csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
3820 /* update gso size and bytecount with header size */
3821 first->gso_segs = skb_shinfo(skb)->gso_segs;
3822 first->bytecount += (first->gso_segs - 1) * *hdr_len;
3824 /* mss_l4len_id: use 1 as index for TSO */
3825 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3826 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3827 mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3829 fceof_saidx |= itd->pfsa;
3830 type_tucmd |= itd->flags | itd->trailer_len;
3832 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3833 vlan_macip_lens = l4.hdr - ip.hdr;
3834 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3835 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3837 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
3843 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3844 struct ixgbevf_tx_buffer *first,
3845 struct ixgbevf_ipsec_tx_data *itd)
3847 struct sk_buff *skb = first->skb;
3848 u32 vlan_macip_lens = 0;
3849 u32 fceof_saidx = 0;
3852 if (skb->ip_summed != CHECKSUM_PARTIAL)
3855 switch (skb->csum_offset) {
3856 case offsetof(struct tcphdr, check):
3857 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3859 case offsetof(struct udphdr, check):
3861 case offsetof(struct sctphdr, checksum):
3862 /* validate that this is actually an SCTP request */
3863 if (skb_csum_is_sctp(skb)) {
3864 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3869 skb_checksum_help(skb);
3873 if (first->protocol == htons(ETH_P_IP))
3874 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3876 /* update TX checksum flag */
3877 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3878 vlan_macip_lens = skb_checksum_start_offset(skb) -
3879 skb_network_offset(skb);
3881 /* vlan_macip_lens: MACLEN, VLAN tag */
3882 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3883 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3885 fceof_saidx |= itd->pfsa;
3886 type_tucmd |= itd->flags | itd->trailer_len;
3888 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3889 fceof_saidx, type_tucmd, 0);
3892 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3894 /* set type for advanced descriptor with frame checksum insertion */
3895 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3896 IXGBE_ADVTXD_DCMD_IFCS |
3897 IXGBE_ADVTXD_DCMD_DEXT);
3899 /* set HW VLAN bit if VLAN is present */
3900 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3901 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3903 /* set segmentation enable bits for TSO/FSO */
3904 if (tx_flags & IXGBE_TX_FLAGS_TSO)
3905 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3910 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3911 u32 tx_flags, unsigned int paylen)
3913 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3915 /* enable L4 checksum for TSO and TX checksum offload */
3916 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3917 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3919 /* enble IPv4 checksum for TSO */
3920 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3921 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3924 if (tx_flags & IXGBE_TX_FLAGS_IPSEC)
3925 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IPSEC);
3927 /* use index 1 context for TSO/FSO/FCOE/IPSEC */
3928 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_IPSEC))
3929 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3931 /* Check Context must be set if Tx switch is enabled, which it
3932 * always is for case where virtual functions are running
3934 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3936 tx_desc->read.olinfo_status = olinfo_status;
3939 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3940 struct ixgbevf_tx_buffer *first,
3943 struct sk_buff *skb = first->skb;
3944 struct ixgbevf_tx_buffer *tx_buffer;
3945 union ixgbe_adv_tx_desc *tx_desc;
3948 unsigned int data_len, size;
3949 u32 tx_flags = first->tx_flags;
3950 __le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3951 u16 i = tx_ring->next_to_use;
3953 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3955 ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
3957 size = skb_headlen(skb);
3958 data_len = skb->data_len;
3960 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3964 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3965 if (dma_mapping_error(tx_ring->dev, dma))
3968 /* record length, and DMA address */
3969 dma_unmap_len_set(tx_buffer, len, size);
3970 dma_unmap_addr_set(tx_buffer, dma, dma);
3972 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3974 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3975 tx_desc->read.cmd_type_len =
3976 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3980 if (i == tx_ring->count) {
3981 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3984 tx_desc->read.olinfo_status = 0;
3986 dma += IXGBE_MAX_DATA_PER_TXD;
3987 size -= IXGBE_MAX_DATA_PER_TXD;
3989 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3992 if (likely(!data_len))
3995 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3999 if (i == tx_ring->count) {
4000 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
4003 tx_desc->read.olinfo_status = 0;
4005 size = skb_frag_size(frag);
4008 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
4011 tx_buffer = &tx_ring->tx_buffer_info[i];
4014 /* write last descriptor with RS and EOP bits */
4015 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
4016 tx_desc->read.cmd_type_len = cmd_type;
4018 /* set the timestamp */
4019 first->time_stamp = jiffies;
4021 skb_tx_timestamp(skb);
4023 /* Force memory writes to complete before letting h/w know there
4024 * are new descriptors to fetch. (Only applicable for weak-ordered
4025 * memory model archs, such as IA-64).
4027 * We also need this memory barrier (wmb) to make certain all of the
4028 * status bits have been updated before next_to_watch is written.
4032 /* set next_to_watch value indicating a packet is present */
4033 first->next_to_watch = tx_desc;
4036 if (i == tx_ring->count)
4039 tx_ring->next_to_use = i;
4041 /* notify HW of packet */
4042 ixgbevf_write_tail(tx_ring, i);
4046 dev_err(tx_ring->dev, "TX DMA map failed\n");
4047 tx_buffer = &tx_ring->tx_buffer_info[i];
4049 /* clear dma mappings for failed tx_buffer_info map */
4050 while (tx_buffer != first) {
4051 if (dma_unmap_len(tx_buffer, len))
4052 dma_unmap_page(tx_ring->dev,
4053 dma_unmap_addr(tx_buffer, dma),
4054 dma_unmap_len(tx_buffer, len),
4056 dma_unmap_len_set(tx_buffer, len, 0);
4059 i += tx_ring->count;
4060 tx_buffer = &tx_ring->tx_buffer_info[i];
4063 if (dma_unmap_len(tx_buffer, len))
4064 dma_unmap_single(tx_ring->dev,
4065 dma_unmap_addr(tx_buffer, dma),
4066 dma_unmap_len(tx_buffer, len),
4068 dma_unmap_len_set(tx_buffer, len, 0);
4070 dev_kfree_skb_any(tx_buffer->skb);
4071 tx_buffer->skb = NULL;
4073 tx_ring->next_to_use = i;
4076 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4078 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
4079 /* Herbert's original patch had:
4080 * smp_mb__after_netif_stop_queue();
4081 * but since that doesn't exist yet, just open code it.
4085 /* We need to check again in a case another CPU has just
4086 * made room available.
4088 if (likely(ixgbevf_desc_unused(tx_ring) < size))
4091 /* A reprieve! - use start_queue because it doesn't call schedule */
4092 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
4093 ++tx_ring->tx_stats.restart_queue;
4098 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4100 if (likely(ixgbevf_desc_unused(tx_ring) >= size))
4102 return __ixgbevf_maybe_stop_tx(tx_ring, size);
4105 static int ixgbevf_xmit_frame_ring(struct sk_buff *skb,
4106 struct ixgbevf_ring *tx_ring)
4108 struct ixgbevf_tx_buffer *first;
4111 u16 count = TXD_USE_COUNT(skb_headlen(skb));
4112 struct ixgbevf_ipsec_tx_data ipsec_tx = { 0 };
4113 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4117 u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
4119 if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
4120 dev_kfree_skb_any(skb);
4121 return NETDEV_TX_OK;
4124 /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
4125 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
4126 * + 2 desc gap to keep tail from touching head,
4127 * + 1 desc for context descriptor,
4128 * otherwise try next time
4130 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4131 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
4132 skb_frag_t *frag = &skb_shinfo(skb)->frags[f];
4134 count += TXD_USE_COUNT(skb_frag_size(frag));
4137 count += skb_shinfo(skb)->nr_frags;
4139 if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
4140 tx_ring->tx_stats.tx_busy++;
4141 return NETDEV_TX_BUSY;
4144 /* record the location of the first descriptor for this packet */
4145 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4147 first->bytecount = skb->len;
4148 first->gso_segs = 1;
4150 if (skb_vlan_tag_present(skb)) {
4151 tx_flags |= skb_vlan_tag_get(skb);
4152 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4153 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4156 /* record initial flags and protocol */
4157 first->tx_flags = tx_flags;
4158 first->protocol = vlan_get_protocol(skb);
4160 #ifdef CONFIG_IXGBEVF_IPSEC
4161 if (xfrm_offload(skb) && !ixgbevf_ipsec_tx(tx_ring, first, &ipsec_tx))
4164 tso = ixgbevf_tso(tx_ring, first, &hdr_len, &ipsec_tx);
4168 ixgbevf_tx_csum(tx_ring, first, &ipsec_tx);
4170 ixgbevf_tx_map(tx_ring, first, hdr_len);
4172 ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
4174 return NETDEV_TX_OK;
4177 dev_kfree_skb_any(first->skb);
4180 return NETDEV_TX_OK;
4183 static netdev_tx_t ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4185 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4186 struct ixgbevf_ring *tx_ring;
4188 if (skb->len <= 0) {
4189 dev_kfree_skb_any(skb);
4190 return NETDEV_TX_OK;
4193 /* The minimum packet size for olinfo paylen is 17 so pad the skb
4194 * in order to meet this minimum size requirement.
4196 if (skb->len < 17) {
4197 if (skb_padto(skb, 17))
4198 return NETDEV_TX_OK;
4202 tx_ring = adapter->tx_ring[skb->queue_mapping];
4203 return ixgbevf_xmit_frame_ring(skb, tx_ring);
4207 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
4208 * @netdev: network interface device structure
4209 * @p: pointer to an address structure
4211 * Returns 0 on success, negative on failure
4213 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
4215 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4216 struct ixgbe_hw *hw = &adapter->hw;
4217 struct sockaddr *addr = p;
4220 if (!is_valid_ether_addr(addr->sa_data))
4221 return -EADDRNOTAVAIL;
4223 spin_lock_bh(&adapter->mbx_lock);
4225 err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
4227 spin_unlock_bh(&adapter->mbx_lock);
4232 ether_addr_copy(hw->mac.addr, addr->sa_data);
4233 ether_addr_copy(hw->mac.perm_addr, addr->sa_data);
4234 ether_addr_copy(netdev->dev_addr, addr->sa_data);
4240 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
4241 * @netdev: network interface device structure
4242 * @new_mtu: new value for maximum frame size
4244 * Returns 0 on success, negative on failure
4246 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
4248 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4249 struct ixgbe_hw *hw = &adapter->hw;
4250 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4253 /* prevent MTU being changed to a size unsupported by XDP */
4254 if (adapter->xdp_prog) {
4255 dev_warn(&adapter->pdev->dev, "MTU cannot be changed while XDP program is loaded\n");
4259 spin_lock_bh(&adapter->mbx_lock);
4260 /* notify the PF of our intent to use this size of frame */
4261 ret = hw->mac.ops.set_rlpml(hw, max_frame);
4262 spin_unlock_bh(&adapter->mbx_lock);
4266 hw_dbg(hw, "changing MTU from %d to %d\n",
4267 netdev->mtu, new_mtu);
4269 /* must set new MTU before calling down or up */
4270 netdev->mtu = new_mtu;
4272 if (netif_running(netdev))
4273 ixgbevf_reinit_locked(adapter);
4278 static int __maybe_unused ixgbevf_suspend(struct device *dev_d)
4280 struct net_device *netdev = dev_get_drvdata(dev_d);
4281 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4284 netif_device_detach(netdev);
4286 if (netif_running(netdev))
4287 ixgbevf_close_suspend(adapter);
4289 ixgbevf_clear_interrupt_scheme(adapter);
4295 static int __maybe_unused ixgbevf_resume(struct device *dev_d)
4297 struct pci_dev *pdev = to_pci_dev(dev_d);
4298 struct net_device *netdev = pci_get_drvdata(pdev);
4299 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
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 static void ixgbevf_shutdown(struct pci_dev *pdev)
4324 ixgbevf_suspend(&pdev->dev);
4327 static void ixgbevf_get_tx_ring_stats(struct rtnl_link_stats64 *stats,
4328 const struct ixgbevf_ring *ring)
4335 start = u64_stats_fetch_begin_irq(&ring->syncp);
4336 bytes = ring->stats.bytes;
4337 packets = ring->stats.packets;
4338 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4339 stats->tx_bytes += bytes;
4340 stats->tx_packets += packets;
4344 static void ixgbevf_get_stats(struct net_device *netdev,
4345 struct rtnl_link_stats64 *stats)
4347 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4350 const struct ixgbevf_ring *ring;
4353 ixgbevf_update_stats(adapter);
4355 stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
4358 for (i = 0; i < adapter->num_rx_queues; i++) {
4359 ring = adapter->rx_ring[i];
4361 start = u64_stats_fetch_begin_irq(&ring->syncp);
4362 bytes = ring->stats.bytes;
4363 packets = ring->stats.packets;
4364 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4365 stats->rx_bytes += bytes;
4366 stats->rx_packets += packets;
4369 for (i = 0; i < adapter->num_tx_queues; i++) {
4370 ring = adapter->tx_ring[i];
4371 ixgbevf_get_tx_ring_stats(stats, ring);
4374 for (i = 0; i < adapter->num_xdp_queues; i++) {
4375 ring = adapter->xdp_ring[i];
4376 ixgbevf_get_tx_ring_stats(stats, ring);
4381 #define IXGBEVF_MAX_MAC_HDR_LEN 127
4382 #define IXGBEVF_MAX_NETWORK_HDR_LEN 511
4384 static netdev_features_t
4385 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
4386 netdev_features_t features)
4388 unsigned int network_hdr_len, mac_hdr_len;
4390 /* Make certain the headers can be described by a context descriptor */
4391 mac_hdr_len = skb_network_header(skb) - skb->data;
4392 if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
4393 return features & ~(NETIF_F_HW_CSUM |
4395 NETIF_F_HW_VLAN_CTAG_TX |
4399 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
4400 if (unlikely(network_hdr_len > IXGBEVF_MAX_NETWORK_HDR_LEN))
4401 return features & ~(NETIF_F_HW_CSUM |
4406 /* We can only support IPV4 TSO in tunnels if we can mangle the
4407 * inner IP ID field, so strip TSO if MANGLEID is not supported.
4409 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
4410 features &= ~NETIF_F_TSO;
4415 static int ixgbevf_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
4417 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4418 struct ixgbevf_adapter *adapter = netdev_priv(dev);
4419 struct bpf_prog *old_prog;
4421 /* verify ixgbevf ring attributes are sufficient for XDP */
4422 for (i = 0; i < adapter->num_rx_queues; i++) {
4423 struct ixgbevf_ring *ring = adapter->rx_ring[i];
4425 if (frame_size > ixgbevf_rx_bufsz(ring))
4429 old_prog = xchg(&adapter->xdp_prog, prog);
4431 /* If transitioning XDP modes reconfigure rings */
4432 if (!!prog != !!old_prog) {
4433 /* Hardware has to reinitialize queues and interrupts to
4434 * match packet buffer alignment. Unfortunately, the
4435 * hardware is not flexible enough to do this dynamically.
4437 if (netif_running(dev))
4440 ixgbevf_clear_interrupt_scheme(adapter);
4441 ixgbevf_init_interrupt_scheme(adapter);
4443 if (netif_running(dev))
4446 for (i = 0; i < adapter->num_rx_queues; i++)
4447 xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
4451 bpf_prog_put(old_prog);
4456 static int ixgbevf_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4458 switch (xdp->command) {
4459 case XDP_SETUP_PROG:
4460 return ixgbevf_xdp_setup(dev, xdp->prog);
4466 static const struct net_device_ops ixgbevf_netdev_ops = {
4467 .ndo_open = ixgbevf_open,
4468 .ndo_stop = ixgbevf_close,
4469 .ndo_start_xmit = ixgbevf_xmit_frame,
4470 .ndo_set_rx_mode = ixgbevf_set_rx_mode,
4471 .ndo_get_stats64 = ixgbevf_get_stats,
4472 .ndo_validate_addr = eth_validate_addr,
4473 .ndo_set_mac_address = ixgbevf_set_mac,
4474 .ndo_change_mtu = ixgbevf_change_mtu,
4475 .ndo_tx_timeout = ixgbevf_tx_timeout,
4476 .ndo_vlan_rx_add_vid = ixgbevf_vlan_rx_add_vid,
4477 .ndo_vlan_rx_kill_vid = ixgbevf_vlan_rx_kill_vid,
4478 .ndo_features_check = ixgbevf_features_check,
4479 .ndo_bpf = ixgbevf_xdp,
4482 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
4484 dev->netdev_ops = &ixgbevf_netdev_ops;
4485 ixgbevf_set_ethtool_ops(dev);
4486 dev->watchdog_timeo = 5 * HZ;
4490 * ixgbevf_probe - Device Initialization Routine
4491 * @pdev: PCI device information struct
4492 * @ent: entry in ixgbevf_pci_tbl
4494 * Returns 0 on success, negative on failure
4496 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
4497 * The OS initialization, configuring of the adapter private structure,
4498 * and a hardware reset occur.
4500 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4502 struct net_device *netdev;
4503 struct ixgbevf_adapter *adapter = NULL;
4504 struct ixgbe_hw *hw = NULL;
4505 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
4506 int err, pci_using_dac;
4507 bool disable_dev = false;
4509 err = pci_enable_device(pdev);
4513 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
4516 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
4518 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4524 err = pci_request_regions(pdev, ixgbevf_driver_name);
4526 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4530 pci_set_master(pdev);
4532 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4536 goto err_alloc_etherdev;
4539 SET_NETDEV_DEV(netdev, &pdev->dev);
4541 adapter = netdev_priv(netdev);
4543 adapter->netdev = netdev;
4544 adapter->pdev = pdev;
4547 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4549 /* call save state here in standalone driver because it relies on
4550 * adapter struct to exist, and needs to call netdev_priv
4552 pci_save_state(pdev);
4554 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4555 pci_resource_len(pdev, 0));
4556 adapter->io_addr = hw->hw_addr;
4562 ixgbevf_assign_netdev_ops(netdev);
4565 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4566 hw->mac.type = ii->mac;
4568 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4569 sizeof(struct ixgbe_mbx_operations));
4571 /* setup the private structure */
4572 err = ixgbevf_sw_init(adapter);
4576 /* The HW MAC address was set and/or determined in sw_init */
4577 if (!is_valid_ether_addr(netdev->dev_addr)) {
4578 pr_err("invalid MAC address\n");
4583 netdev->hw_features = NETIF_F_SG |
4590 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4591 NETIF_F_GSO_GRE_CSUM | \
4592 NETIF_F_GSO_IPXIP4 | \
4593 NETIF_F_GSO_IPXIP6 | \
4594 NETIF_F_GSO_UDP_TUNNEL | \
4595 NETIF_F_GSO_UDP_TUNNEL_CSUM)
4597 netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4598 netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4599 IXGBEVF_GSO_PARTIAL_FEATURES;
4601 netdev->features = netdev->hw_features;
4604 netdev->features |= NETIF_F_HIGHDMA;
4606 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4607 netdev->mpls_features |= NETIF_F_SG |
4611 netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES;
4612 netdev->hw_enc_features |= netdev->vlan_features;
4614 /* set this bit last since it cannot be part of vlan_features */
4615 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4616 NETIF_F_HW_VLAN_CTAG_RX |
4617 NETIF_F_HW_VLAN_CTAG_TX;
4619 netdev->priv_flags |= IFF_UNICAST_FLT;
4621 /* MTU range: 68 - 1504 or 9710 */
4622 netdev->min_mtu = ETH_MIN_MTU;
4623 switch (adapter->hw.api_version) {
4624 case ixgbe_mbox_api_11:
4625 case ixgbe_mbox_api_12:
4626 case ixgbe_mbox_api_13:
4627 case ixgbe_mbox_api_14:
4628 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4629 (ETH_HLEN + ETH_FCS_LEN);
4632 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4633 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4634 (ETH_HLEN + ETH_FCS_LEN);
4636 netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4640 if (IXGBE_REMOVED(hw->hw_addr)) {
4645 timer_setup(&adapter->service_timer, ixgbevf_service_timer, 0);
4647 INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4648 set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4649 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4651 err = ixgbevf_init_interrupt_scheme(adapter);
4655 strcpy(netdev->name, "eth%d");
4657 err = register_netdev(netdev);
4661 pci_set_drvdata(pdev, netdev);
4662 netif_carrier_off(netdev);
4663 ixgbevf_init_ipsec_offload(adapter);
4665 ixgbevf_init_last_counter_stats(adapter);
4667 /* print the VF info */
4668 dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4669 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4671 switch (hw->mac.type) {
4672 case ixgbe_mac_X550_vf:
4673 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4675 case ixgbe_mac_X540_vf:
4676 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4678 case ixgbe_mac_82599_vf:
4680 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4687 ixgbevf_clear_interrupt_scheme(adapter);
4689 ixgbevf_reset_interrupt_capability(adapter);
4690 iounmap(adapter->io_addr);
4691 kfree(adapter->rss_key);
4693 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4694 free_netdev(netdev);
4696 pci_release_regions(pdev);
4699 if (!adapter || disable_dev)
4700 pci_disable_device(pdev);
4705 * ixgbevf_remove - Device Removal Routine
4706 * @pdev: PCI device information struct
4708 * ixgbevf_remove is called by the PCI subsystem to alert the driver
4709 * that it should release a PCI device. The could be caused by a
4710 * Hot-Plug event, or because the driver is going to be removed from
4713 static void ixgbevf_remove(struct pci_dev *pdev)
4715 struct net_device *netdev = pci_get_drvdata(pdev);
4716 struct ixgbevf_adapter *adapter;
4722 adapter = netdev_priv(netdev);
4724 set_bit(__IXGBEVF_REMOVING, &adapter->state);
4725 cancel_work_sync(&adapter->service_task);
4727 if (netdev->reg_state == NETREG_REGISTERED)
4728 unregister_netdev(netdev);
4730 ixgbevf_stop_ipsec_offload(adapter);
4731 ixgbevf_clear_interrupt_scheme(adapter);
4732 ixgbevf_reset_interrupt_capability(adapter);
4734 iounmap(adapter->io_addr);
4735 pci_release_regions(pdev);
4737 hw_dbg(&adapter->hw, "Remove complete\n");
4739 kfree(adapter->rss_key);
4740 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4741 free_netdev(netdev);
4744 pci_disable_device(pdev);
4748 * ixgbevf_io_error_detected - called when PCI error is detected
4749 * @pdev: Pointer to PCI device
4750 * @state: The current pci connection state
4752 * This function is called after a PCI bus error affecting
4753 * this device has been detected.
4755 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4756 pci_channel_state_t state)
4758 struct net_device *netdev = pci_get_drvdata(pdev);
4759 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4761 if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4762 return PCI_ERS_RESULT_DISCONNECT;
4765 netif_device_detach(netdev);
4767 if (netif_running(netdev))
4768 ixgbevf_close_suspend(adapter);
4770 if (state == pci_channel_io_perm_failure) {
4772 return PCI_ERS_RESULT_DISCONNECT;
4775 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4776 pci_disable_device(pdev);
4779 /* Request a slot slot reset. */
4780 return PCI_ERS_RESULT_NEED_RESET;
4784 * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4785 * @pdev: Pointer to PCI device
4787 * Restart the card from scratch, as if from a cold-boot. Implementation
4788 * resembles the first-half of the ixgbevf_resume routine.
4790 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4792 struct net_device *netdev = pci_get_drvdata(pdev);
4793 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4795 if (pci_enable_device_mem(pdev)) {
4797 "Cannot re-enable PCI device after reset.\n");
4798 return PCI_ERS_RESULT_DISCONNECT;
4801 adapter->hw.hw_addr = adapter->io_addr;
4802 smp_mb__before_atomic();
4803 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4804 pci_set_master(pdev);
4806 ixgbevf_reset(adapter);
4808 return PCI_ERS_RESULT_RECOVERED;
4812 * ixgbevf_io_resume - called when traffic can start flowing again.
4813 * @pdev: Pointer to PCI device
4815 * This callback is called when the error recovery driver tells us that
4816 * its OK to resume normal operation. Implementation resembles the
4817 * second-half of the ixgbevf_resume routine.
4819 static void ixgbevf_io_resume(struct pci_dev *pdev)
4821 struct net_device *netdev = pci_get_drvdata(pdev);
4824 if (netif_running(netdev))
4825 ixgbevf_open(netdev);
4827 netif_device_attach(netdev);
4831 /* PCI Error Recovery (ERS) */
4832 static const struct pci_error_handlers ixgbevf_err_handler = {
4833 .error_detected = ixgbevf_io_error_detected,
4834 .slot_reset = ixgbevf_io_slot_reset,
4835 .resume = ixgbevf_io_resume,
4838 static SIMPLE_DEV_PM_OPS(ixgbevf_pm_ops, ixgbevf_suspend, ixgbevf_resume);
4840 static struct pci_driver ixgbevf_driver = {
4841 .name = ixgbevf_driver_name,
4842 .id_table = ixgbevf_pci_tbl,
4843 .probe = ixgbevf_probe,
4844 .remove = ixgbevf_remove,
4846 /* Power Management Hooks */
4847 .driver.pm = &ixgbevf_pm_ops,
4849 .shutdown = ixgbevf_shutdown,
4850 .err_handler = &ixgbevf_err_handler
4854 * ixgbevf_init_module - Driver Registration Routine
4856 * ixgbevf_init_module is the first routine called when the driver is
4857 * loaded. All it does is register with the PCI subsystem.
4859 static int __init ixgbevf_init_module(void)
4861 pr_info("%s\n", ixgbevf_driver_string);
4862 pr_info("%s\n", ixgbevf_copyright);
4863 ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4865 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4869 return pci_register_driver(&ixgbevf_driver);
4872 module_init(ixgbevf_init_module);
4875 * ixgbevf_exit_module - Driver Exit Cleanup Routine
4877 * ixgbevf_exit_module is called just before the driver is removed
4880 static void __exit ixgbevf_exit_module(void)
4882 pci_unregister_driver(&ixgbevf_driver);
4884 destroy_workqueue(ixgbevf_wq);
4891 * ixgbevf_get_hw_dev_name - return device name string
4892 * used by hardware layer to print debugging information
4893 * @hw: pointer to private hardware struct
4895 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4897 struct ixgbevf_adapter *adapter = hw->back;
4899 return adapter->netdev->name;
4903 module_exit(ixgbevf_exit_module);
4905 /* ixgbevf_main.c */