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;
1058 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
1063 act = bpf_prog_run_xdp(xdp_prog, xdp);
1068 xdp_ring = adapter->xdp_ring[rx_ring->queue_index];
1069 result = ixgbevf_xmit_xdp_ring(xdp_ring, xdp);
1070 if (result == IXGBEVF_XDP_CONSUMED)
1074 bpf_warn_invalid_xdp_action(act);
1078 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
1079 fallthrough; /* handle aborts by dropping packet */
1081 result = IXGBEVF_XDP_CONSUMED;
1086 return ERR_PTR(-result);
1089 static unsigned int ixgbevf_rx_frame_truesize(struct ixgbevf_ring *rx_ring,
1092 unsigned int truesize;
1094 #if (PAGE_SIZE < 8192)
1095 truesize = ixgbevf_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
1097 truesize = ring_uses_build_skb(rx_ring) ?
1098 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) +
1099 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
1100 SKB_DATA_ALIGN(size);
1105 static void ixgbevf_rx_buffer_flip(struct ixgbevf_ring *rx_ring,
1106 struct ixgbevf_rx_buffer *rx_buffer,
1109 unsigned int truesize = ixgbevf_rx_frame_truesize(rx_ring, size);
1111 #if (PAGE_SIZE < 8192)
1112 rx_buffer->page_offset ^= truesize;
1114 rx_buffer->page_offset += truesize;
1118 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
1119 struct ixgbevf_ring *rx_ring,
1122 unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
1123 struct ixgbevf_adapter *adapter = q_vector->adapter;
1124 u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
1125 struct sk_buff *skb = rx_ring->skb;
1126 bool xdp_xmit = false;
1127 struct xdp_buff xdp;
1129 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
1130 #if (PAGE_SIZE < 8192)
1131 frame_sz = ixgbevf_rx_frame_truesize(rx_ring, 0);
1133 xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
1135 while (likely(total_rx_packets < budget)) {
1136 struct ixgbevf_rx_buffer *rx_buffer;
1137 union ixgbe_adv_rx_desc *rx_desc;
1140 /* return some buffers to hardware, one at a time is too slow */
1141 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
1142 ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
1146 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
1147 size = le16_to_cpu(rx_desc->wb.upper.length);
1151 /* This memory barrier is needed to keep us from reading
1152 * any other fields out of the rx_desc until we know the
1153 * RXD_STAT_DD bit is set
1157 rx_buffer = ixgbevf_get_rx_buffer(rx_ring, size);
1159 /* retrieve a buffer from the ring */
1161 unsigned int offset = ixgbevf_rx_offset(rx_ring);
1162 unsigned char *hard_start;
1164 hard_start = page_address(rx_buffer->page) +
1165 rx_buffer->page_offset - offset;
1166 xdp_prepare_buff(&xdp, hard_start, offset, size, true);
1167 #if (PAGE_SIZE > 4096)
1168 /* At larger PAGE_SIZE, frame_sz depend on len size */
1169 xdp.frame_sz = ixgbevf_rx_frame_truesize(rx_ring, size);
1171 skb = ixgbevf_run_xdp(adapter, rx_ring, &xdp);
1175 if (PTR_ERR(skb) == -IXGBEVF_XDP_TX) {
1177 ixgbevf_rx_buffer_flip(rx_ring, rx_buffer,
1180 rx_buffer->pagecnt_bias++;
1183 total_rx_bytes += size;
1185 ixgbevf_add_rx_frag(rx_ring, rx_buffer, skb, size);
1186 } else if (ring_uses_build_skb(rx_ring)) {
1187 skb = ixgbevf_build_skb(rx_ring, rx_buffer,
1190 skb = ixgbevf_construct_skb(rx_ring, rx_buffer,
1194 /* exit if we failed to retrieve a buffer */
1196 rx_ring->rx_stats.alloc_rx_buff_failed++;
1197 rx_buffer->pagecnt_bias++;
1201 ixgbevf_put_rx_buffer(rx_ring, rx_buffer, skb);
1204 /* fetch next buffer in frame if non-eop */
1205 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
1208 /* verify the packet layout is correct */
1209 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
1214 /* probably a little skewed due to removing CRC */
1215 total_rx_bytes += skb->len;
1217 /* Workaround hardware that can't do proper VEPA multicast
1220 if ((skb->pkt_type == PACKET_BROADCAST ||
1221 skb->pkt_type == PACKET_MULTICAST) &&
1222 ether_addr_equal(rx_ring->netdev->dev_addr,
1223 eth_hdr(skb)->h_source)) {
1224 dev_kfree_skb_irq(skb);
1228 /* populate checksum, VLAN, and protocol */
1229 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
1231 ixgbevf_rx_skb(q_vector, skb);
1233 /* reset skb pointer */
1236 /* update budget accounting */
1240 /* place incomplete frames back on ring for completion */
1244 struct ixgbevf_ring *xdp_ring =
1245 adapter->xdp_ring[rx_ring->queue_index];
1247 /* Force memory writes to complete before letting h/w
1248 * know there are new descriptors to fetch.
1251 ixgbevf_write_tail(xdp_ring, xdp_ring->next_to_use);
1254 u64_stats_update_begin(&rx_ring->syncp);
1255 rx_ring->stats.packets += total_rx_packets;
1256 rx_ring->stats.bytes += total_rx_bytes;
1257 u64_stats_update_end(&rx_ring->syncp);
1258 q_vector->rx.total_packets += total_rx_packets;
1259 q_vector->rx.total_bytes += total_rx_bytes;
1261 return total_rx_packets;
1265 * ixgbevf_poll - NAPI polling calback
1266 * @napi: napi struct with our devices info in it
1267 * @budget: amount of work driver is allowed to do this pass, in packets
1269 * This function will clean more than one or more rings associated with a
1272 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1274 struct ixgbevf_q_vector *q_vector =
1275 container_of(napi, struct ixgbevf_q_vector, napi);
1276 struct ixgbevf_adapter *adapter = q_vector->adapter;
1277 struct ixgbevf_ring *ring;
1278 int per_ring_budget, work_done = 0;
1279 bool clean_complete = true;
1281 ixgbevf_for_each_ring(ring, q_vector->tx) {
1282 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1283 clean_complete = false;
1289 /* attempt to distribute budget to each queue fairly, but don't allow
1290 * the budget to go below 1 because we'll exit polling
1292 if (q_vector->rx.count > 1)
1293 per_ring_budget = max(budget/q_vector->rx.count, 1);
1295 per_ring_budget = budget;
1297 ixgbevf_for_each_ring(ring, q_vector->rx) {
1298 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1300 work_done += cleaned;
1301 if (cleaned >= per_ring_budget)
1302 clean_complete = false;
1305 /* If all work not completed, return budget and keep polling */
1306 if (!clean_complete)
1309 /* Exit the polling mode, but don't re-enable interrupts if stack might
1310 * poll us due to busy-polling
1312 if (likely(napi_complete_done(napi, work_done))) {
1313 if (adapter->rx_itr_setting == 1)
1314 ixgbevf_set_itr(q_vector);
1315 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1316 !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1317 ixgbevf_irq_enable_queues(adapter,
1318 BIT(q_vector->v_idx));
1321 return min(work_done, budget - 1);
1325 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1326 * @q_vector: structure containing interrupt and ring information
1328 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1330 struct ixgbevf_adapter *adapter = q_vector->adapter;
1331 struct ixgbe_hw *hw = &adapter->hw;
1332 int v_idx = q_vector->v_idx;
1333 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1335 /* set the WDIS bit to not clear the timer bits and cause an
1336 * immediate assertion of the interrupt
1338 itr_reg |= IXGBE_EITR_CNT_WDIS;
1340 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1344 * ixgbevf_configure_msix - Configure MSI-X hardware
1345 * @adapter: board private structure
1347 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1350 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1352 struct ixgbevf_q_vector *q_vector;
1353 int q_vectors, v_idx;
1355 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1356 adapter->eims_enable_mask = 0;
1358 /* Populate the IVAR table and set the ITR values to the
1359 * corresponding register.
1361 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1362 struct ixgbevf_ring *ring;
1364 q_vector = adapter->q_vector[v_idx];
1366 ixgbevf_for_each_ring(ring, q_vector->rx)
1367 ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1369 ixgbevf_for_each_ring(ring, q_vector->tx)
1370 ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1372 if (q_vector->tx.ring && !q_vector->rx.ring) {
1373 /* Tx only vector */
1374 if (adapter->tx_itr_setting == 1)
1375 q_vector->itr = IXGBE_12K_ITR;
1377 q_vector->itr = adapter->tx_itr_setting;
1379 /* Rx or Rx/Tx vector */
1380 if (adapter->rx_itr_setting == 1)
1381 q_vector->itr = IXGBE_20K_ITR;
1383 q_vector->itr = adapter->rx_itr_setting;
1386 /* add q_vector eims value to global eims_enable_mask */
1387 adapter->eims_enable_mask |= BIT(v_idx);
1389 ixgbevf_write_eitr(q_vector);
1392 ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1393 /* setup eims_other and add value to global eims_enable_mask */
1394 adapter->eims_other = BIT(v_idx);
1395 adapter->eims_enable_mask |= adapter->eims_other;
1398 enum latency_range {
1402 latency_invalid = 255
1406 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1407 * @q_vector: structure containing interrupt and ring information
1408 * @ring_container: structure containing ring performance data
1410 * Stores a new ITR value based on packets and byte
1411 * counts during the last interrupt. The advantage of per interrupt
1412 * computation is faster updates and more accurate ITR for the current
1413 * traffic pattern. Constants in this function were computed
1414 * based on theoretical maximum wire speed and thresholds were set based
1415 * on testing data as well as attempting to minimize response time
1416 * while increasing bulk throughput.
1418 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1419 struct ixgbevf_ring_container *ring_container)
1421 int bytes = ring_container->total_bytes;
1422 int packets = ring_container->total_packets;
1425 u8 itr_setting = ring_container->itr;
1430 /* simple throttle rate management
1431 * 0-20MB/s lowest (100000 ints/s)
1432 * 20-100MB/s low (20000 ints/s)
1433 * 100-1249MB/s bulk (12000 ints/s)
1435 /* what was last interrupt timeslice? */
1436 timepassed_us = q_vector->itr >> 2;
1437 if (timepassed_us == 0)
1440 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1442 switch (itr_setting) {
1443 case lowest_latency:
1444 if (bytes_perint > 10)
1445 itr_setting = low_latency;
1448 if (bytes_perint > 20)
1449 itr_setting = bulk_latency;
1450 else if (bytes_perint <= 10)
1451 itr_setting = lowest_latency;
1454 if (bytes_perint <= 20)
1455 itr_setting = low_latency;
1459 /* clear work counters since we have the values we need */
1460 ring_container->total_bytes = 0;
1461 ring_container->total_packets = 0;
1463 /* write updated itr to ring container */
1464 ring_container->itr = itr_setting;
1467 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1469 u32 new_itr = q_vector->itr;
1472 ixgbevf_update_itr(q_vector, &q_vector->tx);
1473 ixgbevf_update_itr(q_vector, &q_vector->rx);
1475 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1477 switch (current_itr) {
1478 /* counts and packets in update_itr are dependent on these numbers */
1479 case lowest_latency:
1480 new_itr = IXGBE_100K_ITR;
1483 new_itr = IXGBE_20K_ITR;
1486 new_itr = IXGBE_12K_ITR;
1492 if (new_itr != q_vector->itr) {
1493 /* do an exponential smoothing */
1494 new_itr = (10 * new_itr * q_vector->itr) /
1495 ((9 * new_itr) + q_vector->itr);
1497 /* save the algorithm value here */
1498 q_vector->itr = new_itr;
1500 ixgbevf_write_eitr(q_vector);
1504 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1506 struct ixgbevf_adapter *adapter = data;
1507 struct ixgbe_hw *hw = &adapter->hw;
1509 hw->mac.get_link_status = 1;
1511 ixgbevf_service_event_schedule(adapter);
1513 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1519 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1521 * @data: pointer to our q_vector struct for this interrupt vector
1523 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1525 struct ixgbevf_q_vector *q_vector = data;
1527 /* EIAM disabled interrupts (on this vector) for us */
1528 if (q_vector->rx.ring || q_vector->tx.ring)
1529 napi_schedule_irqoff(&q_vector->napi);
1535 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1536 * @adapter: board private structure
1538 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1539 * interrupts from the kernel.
1541 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1543 struct net_device *netdev = adapter->netdev;
1544 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1545 unsigned int ri = 0, ti = 0;
1548 for (vector = 0; vector < q_vectors; vector++) {
1549 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1550 struct msix_entry *entry = &adapter->msix_entries[vector];
1552 if (q_vector->tx.ring && q_vector->rx.ring) {
1553 snprintf(q_vector->name, sizeof(q_vector->name),
1554 "%s-TxRx-%u", netdev->name, ri++);
1556 } else if (q_vector->rx.ring) {
1557 snprintf(q_vector->name, sizeof(q_vector->name),
1558 "%s-rx-%u", netdev->name, ri++);
1559 } else if (q_vector->tx.ring) {
1560 snprintf(q_vector->name, sizeof(q_vector->name),
1561 "%s-tx-%u", netdev->name, ti++);
1563 /* skip this unused q_vector */
1566 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1567 q_vector->name, q_vector);
1569 hw_dbg(&adapter->hw,
1570 "request_irq failed for MSIX interrupt Error: %d\n",
1572 goto free_queue_irqs;
1576 err = request_irq(adapter->msix_entries[vector].vector,
1577 &ixgbevf_msix_other, 0, netdev->name, adapter);
1579 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1581 goto free_queue_irqs;
1589 free_irq(adapter->msix_entries[vector].vector,
1590 adapter->q_vector[vector]);
1592 /* This failure is non-recoverable - it indicates the system is
1593 * out of MSIX vector resources and the VF driver cannot run
1594 * without them. Set the number of msix vectors to zero
1595 * indicating that not enough can be allocated. The error
1596 * will be returned to the user indicating device open failed.
1597 * Any further attempts to force the driver to open will also
1598 * fail. The only way to recover is to unload the driver and
1599 * reload it again. If the system has recovered some MSIX
1600 * vectors then it may succeed.
1602 adapter->num_msix_vectors = 0;
1607 * ixgbevf_request_irq - initialize interrupts
1608 * @adapter: board private structure
1610 * Attempts to configure interrupts using the best available
1611 * capabilities of the hardware and kernel.
1613 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1615 int err = ixgbevf_request_msix_irqs(adapter);
1618 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1623 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1627 if (!adapter->msix_entries)
1630 q_vectors = adapter->num_msix_vectors;
1633 free_irq(adapter->msix_entries[i].vector, adapter);
1636 for (; i >= 0; i--) {
1637 /* free only the irqs that were actually requested */
1638 if (!adapter->q_vector[i]->rx.ring &&
1639 !adapter->q_vector[i]->tx.ring)
1642 free_irq(adapter->msix_entries[i].vector,
1643 adapter->q_vector[i]);
1648 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1649 * @adapter: board private structure
1651 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1653 struct ixgbe_hw *hw = &adapter->hw;
1656 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1657 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1658 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1660 IXGBE_WRITE_FLUSH(hw);
1662 for (i = 0; i < adapter->num_msix_vectors; i++)
1663 synchronize_irq(adapter->msix_entries[i].vector);
1667 * ixgbevf_irq_enable - Enable default interrupt generation settings
1668 * @adapter: board private structure
1670 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1672 struct ixgbe_hw *hw = &adapter->hw;
1674 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1675 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1676 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1680 * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1681 * @adapter: board private structure
1682 * @ring: structure containing ring specific data
1684 * Configure the Tx descriptor ring after a reset.
1686 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1687 struct ixgbevf_ring *ring)
1689 struct ixgbe_hw *hw = &adapter->hw;
1690 u64 tdba = ring->dma;
1692 u32 txdctl = IXGBE_TXDCTL_ENABLE;
1693 u8 reg_idx = ring->reg_idx;
1695 /* disable queue to avoid issues while updating state */
1696 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1697 IXGBE_WRITE_FLUSH(hw);
1699 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1700 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1701 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1702 ring->count * sizeof(union ixgbe_adv_tx_desc));
1704 /* disable head writeback */
1705 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1706 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1708 /* enable relaxed ordering */
1709 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1710 (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1711 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1713 /* reset head and tail pointers */
1714 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1715 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1716 ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1718 /* reset ntu and ntc to place SW in sync with hardwdare */
1719 ring->next_to_clean = 0;
1720 ring->next_to_use = 0;
1722 /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1723 * to or less than the number of on chip descriptors, which is
1726 txdctl |= (8 << 16); /* WTHRESH = 8 */
1728 /* Setting PTHRESH to 32 both improves performance */
1729 txdctl |= (1u << 8) | /* HTHRESH = 1 */
1730 32; /* PTHRESH = 32 */
1732 /* reinitialize tx_buffer_info */
1733 memset(ring->tx_buffer_info, 0,
1734 sizeof(struct ixgbevf_tx_buffer) * ring->count);
1736 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1737 clear_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1739 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1741 /* poll to verify queue is enabled */
1743 usleep_range(1000, 2000);
1744 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1745 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1747 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1751 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1752 * @adapter: board private structure
1754 * Configure the Tx unit of the MAC after a reset.
1756 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1760 /* Setup the HW Tx Head and Tail descriptor pointers */
1761 for (i = 0; i < adapter->num_tx_queues; i++)
1762 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1763 for (i = 0; i < adapter->num_xdp_queues; i++)
1764 ixgbevf_configure_tx_ring(adapter, adapter->xdp_ring[i]);
1767 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1769 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter,
1770 struct ixgbevf_ring *ring, int index)
1772 struct ixgbe_hw *hw = &adapter->hw;
1775 srrctl = IXGBE_SRRCTL_DROP_EN;
1777 srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1778 if (ring_uses_large_buffer(ring))
1779 srrctl |= IXGBEVF_RXBUFFER_3072 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1781 srrctl |= IXGBEVF_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1782 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1784 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1787 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1789 struct ixgbe_hw *hw = &adapter->hw;
1791 /* PSRTYPE must be initialized in 82599 */
1792 u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1793 IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1794 IXGBE_PSRTYPE_L2HDR;
1796 if (adapter->num_rx_queues > 1)
1799 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1802 #define IXGBEVF_MAX_RX_DESC_POLL 10
1803 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1804 struct ixgbevf_ring *ring)
1806 struct ixgbe_hw *hw = &adapter->hw;
1807 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1809 u8 reg_idx = ring->reg_idx;
1811 if (IXGBE_REMOVED(hw->hw_addr))
1813 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1814 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1816 /* write value back with RXDCTL.ENABLE bit cleared */
1817 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1819 /* the hardware may take up to 100us to really disable the Rx queue */
1822 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1823 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1826 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1830 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1831 struct ixgbevf_ring *ring)
1833 struct ixgbe_hw *hw = &adapter->hw;
1834 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1836 u8 reg_idx = ring->reg_idx;
1838 if (IXGBE_REMOVED(hw->hw_addr))
1841 usleep_range(1000, 2000);
1842 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1843 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1846 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1851 * ixgbevf_init_rss_key - Initialize adapter RSS key
1852 * @adapter: device handle
1854 * Allocates and initializes the RSS key if it is not allocated.
1856 static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter)
1860 if (!adapter->rss_key) {
1861 rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL);
1862 if (unlikely(!rss_key))
1865 netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE);
1866 adapter->rss_key = rss_key;
1872 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1874 struct ixgbe_hw *hw = &adapter->hw;
1875 u32 vfmrqc = 0, vfreta = 0;
1876 u16 rss_i = adapter->num_rx_queues;
1879 /* Fill out hash function seeds */
1880 for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1881 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i));
1883 for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1887 adapter->rss_indir_tbl[i] = j;
1889 vfreta |= j << (i & 0x3) * 8;
1891 IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1896 /* Perform hash on these packet types */
1897 vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1898 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1899 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1900 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1902 vfmrqc |= IXGBE_VFMRQC_RSSEN;
1904 IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1907 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1908 struct ixgbevf_ring *ring)
1910 struct ixgbe_hw *hw = &adapter->hw;
1911 union ixgbe_adv_rx_desc *rx_desc;
1912 u64 rdba = ring->dma;
1914 u8 reg_idx = ring->reg_idx;
1916 /* disable queue to avoid issues while updating state */
1917 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1918 ixgbevf_disable_rx_queue(adapter, ring);
1920 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1921 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1922 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1923 ring->count * sizeof(union ixgbe_adv_rx_desc));
1925 #ifndef CONFIG_SPARC
1926 /* enable relaxed ordering */
1927 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1928 IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1930 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1931 IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1932 IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1935 /* reset head and tail pointers */
1936 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1937 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1938 ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1940 /* initialize rx_buffer_info */
1941 memset(ring->rx_buffer_info, 0,
1942 sizeof(struct ixgbevf_rx_buffer) * ring->count);
1944 /* initialize Rx descriptor 0 */
1945 rx_desc = IXGBEVF_RX_DESC(ring, 0);
1946 rx_desc->wb.upper.length = 0;
1948 /* reset ntu and ntc to place SW in sync with hardwdare */
1949 ring->next_to_clean = 0;
1950 ring->next_to_use = 0;
1951 ring->next_to_alloc = 0;
1953 ixgbevf_configure_srrctl(adapter, ring, reg_idx);
1955 /* RXDCTL.RLPML does not work on 82599 */
1956 if (adapter->hw.mac.type != ixgbe_mac_82599_vf) {
1957 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
1958 IXGBE_RXDCTL_RLPML_EN);
1960 #if (PAGE_SIZE < 8192)
1961 /* Limit the maximum frame size so we don't overrun the skb */
1962 if (ring_uses_build_skb(ring) &&
1963 !ring_uses_large_buffer(ring))
1964 rxdctl |= IXGBEVF_MAX_FRAME_BUILD_SKB |
1965 IXGBE_RXDCTL_RLPML_EN;
1969 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1970 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1972 ixgbevf_rx_desc_queue_enable(adapter, ring);
1973 ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1976 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter,
1977 struct ixgbevf_ring *rx_ring)
1979 struct net_device *netdev = adapter->netdev;
1980 unsigned int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1982 /* set build_skb and buffer size flags */
1983 clear_ring_build_skb_enabled(rx_ring);
1984 clear_ring_uses_large_buffer(rx_ring);
1986 if (adapter->flags & IXGBEVF_FLAGS_LEGACY_RX)
1989 set_ring_build_skb_enabled(rx_ring);
1991 if (PAGE_SIZE < 8192) {
1992 if (max_frame <= IXGBEVF_MAX_FRAME_BUILD_SKB)
1995 set_ring_uses_large_buffer(rx_ring);
2000 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
2001 * @adapter: board private structure
2003 * Configure the Rx unit of the MAC after a reset.
2005 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
2007 struct ixgbe_hw *hw = &adapter->hw;
2008 struct net_device *netdev = adapter->netdev;
2011 ixgbevf_setup_psrtype(adapter);
2012 if (hw->mac.type >= ixgbe_mac_X550_vf)
2013 ixgbevf_setup_vfmrqc(adapter);
2015 spin_lock_bh(&adapter->mbx_lock);
2016 /* notify the PF of our intent to use this size of frame */
2017 ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
2018 spin_unlock_bh(&adapter->mbx_lock);
2020 dev_err(&adapter->pdev->dev,
2021 "Failed to set MTU at %d\n", netdev->mtu);
2023 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2024 * the Base and Length of the Rx Descriptor Ring
2026 for (i = 0; i < adapter->num_rx_queues; i++) {
2027 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
2029 ixgbevf_set_rx_buffer_len(adapter, rx_ring);
2030 ixgbevf_configure_rx_ring(adapter, rx_ring);
2034 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
2035 __be16 proto, u16 vid)
2037 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2038 struct ixgbe_hw *hw = &adapter->hw;
2041 spin_lock_bh(&adapter->mbx_lock);
2043 /* add VID to filter table */
2044 err = hw->mac.ops.set_vfta(hw, vid, 0, true);
2046 spin_unlock_bh(&adapter->mbx_lock);
2048 /* translate error return types so error makes sense */
2049 if (err == IXGBE_ERR_MBX)
2052 if (err == IXGBE_ERR_INVALID_ARGUMENT)
2055 set_bit(vid, adapter->active_vlans);
2060 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
2061 __be16 proto, u16 vid)
2063 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2064 struct ixgbe_hw *hw = &adapter->hw;
2067 spin_lock_bh(&adapter->mbx_lock);
2069 /* remove VID from filter table */
2070 err = hw->mac.ops.set_vfta(hw, vid, 0, false);
2072 spin_unlock_bh(&adapter->mbx_lock);
2074 clear_bit(vid, adapter->active_vlans);
2079 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
2083 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2084 ixgbevf_vlan_rx_add_vid(adapter->netdev,
2085 htons(ETH_P_8021Q), vid);
2088 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
2090 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2091 struct ixgbe_hw *hw = &adapter->hw;
2094 if (!netdev_uc_empty(netdev)) {
2095 struct netdev_hw_addr *ha;
2097 netdev_for_each_uc_addr(ha, netdev) {
2098 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
2102 /* If the list is empty then send message to PF driver to
2103 * clear all MAC VLANs on this VF.
2105 hw->mac.ops.set_uc_addr(hw, 0, NULL);
2112 * ixgbevf_set_rx_mode - Multicast and unicast set
2113 * @netdev: network interface device structure
2115 * The set_rx_method entry point is called whenever the multicast address
2116 * list, unicast address list or the network interface flags are updated.
2117 * This routine is responsible for configuring the hardware for proper
2118 * multicast mode and configuring requested unicast filters.
2120 static void ixgbevf_set_rx_mode(struct net_device *netdev)
2122 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2123 struct ixgbe_hw *hw = &adapter->hw;
2124 unsigned int flags = netdev->flags;
2127 /* request the most inclusive mode we need */
2128 if (flags & IFF_PROMISC)
2129 xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
2130 else if (flags & IFF_ALLMULTI)
2131 xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
2132 else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
2133 xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
2135 xcast_mode = IXGBEVF_XCAST_MODE_NONE;
2137 spin_lock_bh(&adapter->mbx_lock);
2139 hw->mac.ops.update_xcast_mode(hw, xcast_mode);
2141 /* reprogram multicast list */
2142 hw->mac.ops.update_mc_addr_list(hw, netdev);
2144 ixgbevf_write_uc_addr_list(netdev);
2146 spin_unlock_bh(&adapter->mbx_lock);
2149 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
2152 struct ixgbevf_q_vector *q_vector;
2153 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2155 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2156 q_vector = adapter->q_vector[q_idx];
2157 napi_enable(&q_vector->napi);
2161 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
2164 struct ixgbevf_q_vector *q_vector;
2165 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2167 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2168 q_vector = adapter->q_vector[q_idx];
2169 napi_disable(&q_vector->napi);
2173 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
2175 struct ixgbe_hw *hw = &adapter->hw;
2176 unsigned int def_q = 0;
2177 unsigned int num_tcs = 0;
2178 unsigned int num_rx_queues = adapter->num_rx_queues;
2179 unsigned int num_tx_queues = adapter->num_tx_queues;
2182 spin_lock_bh(&adapter->mbx_lock);
2184 /* fetch queue configuration from the PF */
2185 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2187 spin_unlock_bh(&adapter->mbx_lock);
2193 /* we need only one Tx queue */
2196 /* update default Tx ring register index */
2197 adapter->tx_ring[0]->reg_idx = def_q;
2199 /* we need as many queues as traffic classes */
2200 num_rx_queues = num_tcs;
2203 /* if we have a bad config abort request queue reset */
2204 if ((adapter->num_rx_queues != num_rx_queues) ||
2205 (adapter->num_tx_queues != num_tx_queues)) {
2206 /* force mailbox timeout to prevent further messages */
2207 hw->mbx.timeout = 0;
2209 /* wait for watchdog to come around and bail us out */
2210 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2216 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2218 ixgbevf_configure_dcb(adapter);
2220 ixgbevf_set_rx_mode(adapter->netdev);
2222 ixgbevf_restore_vlan(adapter);
2223 ixgbevf_ipsec_restore(adapter);
2225 ixgbevf_configure_tx(adapter);
2226 ixgbevf_configure_rx(adapter);
2229 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2231 /* Only save pre-reset stats if there are some */
2232 if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2233 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2234 adapter->stats.base_vfgprc;
2235 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2236 adapter->stats.base_vfgptc;
2237 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2238 adapter->stats.base_vfgorc;
2239 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2240 adapter->stats.base_vfgotc;
2241 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2242 adapter->stats.base_vfmprc;
2246 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2248 struct ixgbe_hw *hw = &adapter->hw;
2250 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2251 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2252 adapter->stats.last_vfgorc |=
2253 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2254 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2255 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2256 adapter->stats.last_vfgotc |=
2257 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2258 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2260 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2261 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2262 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2263 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2264 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2267 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2269 struct ixgbe_hw *hw = &adapter->hw;
2270 static const int api[] = {
2276 ixgbe_mbox_api_unknown
2280 spin_lock_bh(&adapter->mbx_lock);
2282 while (api[idx] != ixgbe_mbox_api_unknown) {
2283 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2289 spin_unlock_bh(&adapter->mbx_lock);
2292 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2294 struct net_device *netdev = adapter->netdev;
2295 struct ixgbe_hw *hw = &adapter->hw;
2297 ixgbevf_configure_msix(adapter);
2299 spin_lock_bh(&adapter->mbx_lock);
2301 if (is_valid_ether_addr(hw->mac.addr))
2302 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2304 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2306 spin_unlock_bh(&adapter->mbx_lock);
2308 smp_mb__before_atomic();
2309 clear_bit(__IXGBEVF_DOWN, &adapter->state);
2310 ixgbevf_napi_enable_all(adapter);
2312 /* clear any pending interrupts, may auto mask */
2313 IXGBE_READ_REG(hw, IXGBE_VTEICR);
2314 ixgbevf_irq_enable(adapter);
2316 /* enable transmits */
2317 netif_tx_start_all_queues(netdev);
2319 ixgbevf_save_reset_stats(adapter);
2320 ixgbevf_init_last_counter_stats(adapter);
2322 hw->mac.get_link_status = 1;
2323 mod_timer(&adapter->service_timer, jiffies);
2326 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2328 ixgbevf_configure(adapter);
2330 ixgbevf_up_complete(adapter);
2334 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2335 * @rx_ring: ring to free buffers from
2337 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2339 u16 i = rx_ring->next_to_clean;
2341 /* Free Rx ring sk_buff */
2343 dev_kfree_skb(rx_ring->skb);
2344 rx_ring->skb = NULL;
2347 /* Free all the Rx ring pages */
2348 while (i != rx_ring->next_to_alloc) {
2349 struct ixgbevf_rx_buffer *rx_buffer;
2351 rx_buffer = &rx_ring->rx_buffer_info[i];
2353 /* Invalidate cache lines that may have been written to by
2354 * device so that we avoid corrupting memory.
2356 dma_sync_single_range_for_cpu(rx_ring->dev,
2358 rx_buffer->page_offset,
2359 ixgbevf_rx_bufsz(rx_ring),
2362 /* free resources associated with mapping */
2363 dma_unmap_page_attrs(rx_ring->dev,
2365 ixgbevf_rx_pg_size(rx_ring),
2367 IXGBEVF_RX_DMA_ATTR);
2369 __page_frag_cache_drain(rx_buffer->page,
2370 rx_buffer->pagecnt_bias);
2373 if (i == rx_ring->count)
2377 rx_ring->next_to_alloc = 0;
2378 rx_ring->next_to_clean = 0;
2379 rx_ring->next_to_use = 0;
2383 * ixgbevf_clean_tx_ring - Free Tx Buffers
2384 * @tx_ring: ring to be cleaned
2386 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2388 u16 i = tx_ring->next_to_clean;
2389 struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
2391 while (i != tx_ring->next_to_use) {
2392 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
2394 /* Free all the Tx ring sk_buffs */
2395 if (ring_is_xdp(tx_ring))
2396 page_frag_free(tx_buffer->data);
2398 dev_kfree_skb_any(tx_buffer->skb);
2400 /* unmap skb header data */
2401 dma_unmap_single(tx_ring->dev,
2402 dma_unmap_addr(tx_buffer, dma),
2403 dma_unmap_len(tx_buffer, len),
2406 /* check for eop_desc to determine the end of the packet */
2407 eop_desc = tx_buffer->next_to_watch;
2408 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2410 /* unmap remaining buffers */
2411 while (tx_desc != eop_desc) {
2415 if (unlikely(i == tx_ring->count)) {
2417 tx_buffer = tx_ring->tx_buffer_info;
2418 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
2421 /* unmap any remaining paged data */
2422 if (dma_unmap_len(tx_buffer, len))
2423 dma_unmap_page(tx_ring->dev,
2424 dma_unmap_addr(tx_buffer, dma),
2425 dma_unmap_len(tx_buffer, len),
2429 /* move us one more past the eop_desc for start of next pkt */
2432 if (unlikely(i == tx_ring->count)) {
2434 tx_buffer = tx_ring->tx_buffer_info;
2438 /* reset next_to_use and next_to_clean */
2439 tx_ring->next_to_use = 0;
2440 tx_ring->next_to_clean = 0;
2445 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2446 * @adapter: board private structure
2448 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2452 for (i = 0; i < adapter->num_rx_queues; i++)
2453 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2457 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2458 * @adapter: board private structure
2460 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2464 for (i = 0; i < adapter->num_tx_queues; i++)
2465 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2466 for (i = 0; i < adapter->num_xdp_queues; i++)
2467 ixgbevf_clean_tx_ring(adapter->xdp_ring[i]);
2470 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2472 struct net_device *netdev = adapter->netdev;
2473 struct ixgbe_hw *hw = &adapter->hw;
2476 /* signal that we are down to the interrupt handler */
2477 if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2478 return; /* do nothing if already down */
2480 /* disable all enabled Rx queues */
2481 for (i = 0; i < adapter->num_rx_queues; i++)
2482 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2484 usleep_range(10000, 20000);
2486 netif_tx_stop_all_queues(netdev);
2488 /* call carrier off first to avoid false dev_watchdog timeouts */
2489 netif_carrier_off(netdev);
2490 netif_tx_disable(netdev);
2492 ixgbevf_irq_disable(adapter);
2494 ixgbevf_napi_disable_all(adapter);
2496 del_timer_sync(&adapter->service_timer);
2498 /* disable transmits in the hardware now that interrupts are off */
2499 for (i = 0; i < adapter->num_tx_queues; i++) {
2500 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2502 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2503 IXGBE_TXDCTL_SWFLSH);
2506 for (i = 0; i < adapter->num_xdp_queues; i++) {
2507 u8 reg_idx = adapter->xdp_ring[i]->reg_idx;
2509 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2510 IXGBE_TXDCTL_SWFLSH);
2513 if (!pci_channel_offline(adapter->pdev))
2514 ixgbevf_reset(adapter);
2516 ixgbevf_clean_all_tx_rings(adapter);
2517 ixgbevf_clean_all_rx_rings(adapter);
2520 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2522 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2525 ixgbevf_down(adapter);
2526 pci_set_master(adapter->pdev);
2527 ixgbevf_up(adapter);
2529 clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2532 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2534 struct ixgbe_hw *hw = &adapter->hw;
2535 struct net_device *netdev = adapter->netdev;
2537 if (hw->mac.ops.reset_hw(hw)) {
2538 hw_dbg(hw, "PF still resetting\n");
2540 hw->mac.ops.init_hw(hw);
2541 ixgbevf_negotiate_api(adapter);
2544 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2545 ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2546 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2549 adapter->last_reset = jiffies;
2552 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2555 int vector_threshold;
2557 /* We'll want at least 2 (vector_threshold):
2558 * 1) TxQ[0] + RxQ[0] handler
2559 * 2) Other (Link Status Change, etc.)
2561 vector_threshold = MIN_MSIX_COUNT;
2563 /* The more we get, the more we will assign to Tx/Rx Cleanup
2564 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2565 * Right now, we simply care about how many we'll get; we'll
2566 * set them up later while requesting irq's.
2568 vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2569 vector_threshold, vectors);
2572 dev_err(&adapter->pdev->dev,
2573 "Unable to allocate MSI-X interrupts\n");
2574 kfree(adapter->msix_entries);
2575 adapter->msix_entries = NULL;
2579 /* Adjust for only the vectors we'll use, which is minimum
2580 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2581 * vectors we were allocated.
2583 adapter->num_msix_vectors = vectors;
2589 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2590 * @adapter: board private structure to initialize
2592 * This is the top level queue allocation routine. The order here is very
2593 * important, starting with the "most" number of features turned on at once,
2594 * and ending with the smallest set of features. This way large combinations
2595 * can be allocated if they're turned on, and smaller combinations are the
2596 * fall through conditions.
2599 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2601 struct ixgbe_hw *hw = &adapter->hw;
2602 unsigned int def_q = 0;
2603 unsigned int num_tcs = 0;
2606 /* Start with base case */
2607 adapter->num_rx_queues = 1;
2608 adapter->num_tx_queues = 1;
2609 adapter->num_xdp_queues = 0;
2611 spin_lock_bh(&adapter->mbx_lock);
2613 /* fetch queue configuration from the PF */
2614 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2616 spin_unlock_bh(&adapter->mbx_lock);
2621 /* we need as many queues as traffic classes */
2623 adapter->num_rx_queues = num_tcs;
2625 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2627 switch (hw->api_version) {
2628 case ixgbe_mbox_api_11:
2629 case ixgbe_mbox_api_12:
2630 case ixgbe_mbox_api_13:
2631 case ixgbe_mbox_api_14:
2632 if (adapter->xdp_prog &&
2633 hw->mac.max_tx_queues == rss)
2634 rss = rss > 3 ? 2 : 1;
2636 adapter->num_rx_queues = rss;
2637 adapter->num_tx_queues = rss;
2638 adapter->num_xdp_queues = adapter->xdp_prog ? rss : 0;
2647 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2648 * @adapter: board private structure to initialize
2650 * Attempt to configure the interrupts using the best available
2651 * capabilities of the hardware and the kernel.
2653 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2655 int vector, v_budget;
2657 /* It's easy to be greedy for MSI-X vectors, but it really
2658 * doesn't do us much good if we have a lot more vectors
2659 * than CPU's. So let's be conservative and only ask for
2660 * (roughly) the same number of vectors as there are CPU's.
2661 * The default is to use pairs of vectors.
2663 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2664 v_budget = min_t(int, v_budget, num_online_cpus());
2665 v_budget += NON_Q_VECTORS;
2667 adapter->msix_entries = kcalloc(v_budget,
2668 sizeof(struct msix_entry), GFP_KERNEL);
2669 if (!adapter->msix_entries)
2672 for (vector = 0; vector < v_budget; vector++)
2673 adapter->msix_entries[vector].entry = vector;
2675 /* A failure in MSI-X entry allocation isn't fatal, but the VF driver
2676 * does not support any other modes, so we will simply fail here. Note
2677 * that we clean up the msix_entries pointer else-where.
2679 return ixgbevf_acquire_msix_vectors(adapter, v_budget);
2682 static void ixgbevf_add_ring(struct ixgbevf_ring *ring,
2683 struct ixgbevf_ring_container *head)
2685 ring->next = head->ring;
2691 * ixgbevf_alloc_q_vector - Allocate memory for a single interrupt vector
2692 * @adapter: board private structure to initialize
2693 * @v_idx: index of vector in adapter struct
2694 * @txr_count: number of Tx rings for q vector
2695 * @txr_idx: index of first Tx ring to assign
2696 * @xdp_count: total number of XDP rings to allocate
2697 * @xdp_idx: index of first XDP ring to allocate
2698 * @rxr_count: number of Rx rings for q vector
2699 * @rxr_idx: index of first Rx ring to assign
2701 * We allocate one q_vector. If allocation fails we return -ENOMEM.
2703 static int ixgbevf_alloc_q_vector(struct ixgbevf_adapter *adapter, int v_idx,
2704 int txr_count, int txr_idx,
2705 int xdp_count, int xdp_idx,
2706 int rxr_count, int rxr_idx)
2708 struct ixgbevf_q_vector *q_vector;
2709 int reg_idx = txr_idx + xdp_idx;
2710 struct ixgbevf_ring *ring;
2711 int ring_count, size;
2713 ring_count = txr_count + xdp_count + rxr_count;
2714 size = sizeof(*q_vector) + (sizeof(*ring) * ring_count);
2716 /* allocate q_vector and rings */
2717 q_vector = kzalloc(size, GFP_KERNEL);
2721 /* initialize NAPI */
2722 netif_napi_add(adapter->netdev, &q_vector->napi, ixgbevf_poll, 64);
2724 /* tie q_vector and adapter together */
2725 adapter->q_vector[v_idx] = q_vector;
2726 q_vector->adapter = adapter;
2727 q_vector->v_idx = v_idx;
2729 /* initialize pointer to rings */
2730 ring = q_vector->ring;
2733 /* assign generic ring traits */
2734 ring->dev = &adapter->pdev->dev;
2735 ring->netdev = adapter->netdev;
2737 /* configure backlink on ring */
2738 ring->q_vector = q_vector;
2740 /* update q_vector Tx values */
2741 ixgbevf_add_ring(ring, &q_vector->tx);
2743 /* apply Tx specific ring traits */
2744 ring->count = adapter->tx_ring_count;
2745 ring->queue_index = txr_idx;
2746 ring->reg_idx = reg_idx;
2748 /* assign ring to adapter */
2749 adapter->tx_ring[txr_idx] = ring;
2751 /* update count and index */
2756 /* push pointer to next ring */
2761 /* assign generic ring traits */
2762 ring->dev = &adapter->pdev->dev;
2763 ring->netdev = adapter->netdev;
2765 /* configure backlink on ring */
2766 ring->q_vector = q_vector;
2768 /* update q_vector Tx values */
2769 ixgbevf_add_ring(ring, &q_vector->tx);
2771 /* apply Tx specific ring traits */
2772 ring->count = adapter->tx_ring_count;
2773 ring->queue_index = xdp_idx;
2774 ring->reg_idx = reg_idx;
2777 /* assign ring to adapter */
2778 adapter->xdp_ring[xdp_idx] = ring;
2780 /* update count and index */
2785 /* push pointer to next ring */
2790 /* assign generic ring traits */
2791 ring->dev = &adapter->pdev->dev;
2792 ring->netdev = adapter->netdev;
2794 /* configure backlink on ring */
2795 ring->q_vector = q_vector;
2797 /* update q_vector Rx values */
2798 ixgbevf_add_ring(ring, &q_vector->rx);
2800 /* apply Rx specific ring traits */
2801 ring->count = adapter->rx_ring_count;
2802 ring->queue_index = rxr_idx;
2803 ring->reg_idx = rxr_idx;
2805 /* assign ring to adapter */
2806 adapter->rx_ring[rxr_idx] = ring;
2808 /* update count and index */
2812 /* push pointer to next ring */
2820 * ixgbevf_free_q_vector - Free memory allocated for specific interrupt vector
2821 * @adapter: board private structure to initialize
2822 * @v_idx: index of vector in adapter struct
2824 * This function frees the memory allocated to the q_vector. In addition if
2825 * NAPI is enabled it will delete any references to the NAPI struct prior
2826 * to freeing the q_vector.
2828 static void ixgbevf_free_q_vector(struct ixgbevf_adapter *adapter, int v_idx)
2830 struct ixgbevf_q_vector *q_vector = adapter->q_vector[v_idx];
2831 struct ixgbevf_ring *ring;
2833 ixgbevf_for_each_ring(ring, q_vector->tx) {
2834 if (ring_is_xdp(ring))
2835 adapter->xdp_ring[ring->queue_index] = NULL;
2837 adapter->tx_ring[ring->queue_index] = NULL;
2840 ixgbevf_for_each_ring(ring, q_vector->rx)
2841 adapter->rx_ring[ring->queue_index] = NULL;
2843 adapter->q_vector[v_idx] = NULL;
2844 netif_napi_del(&q_vector->napi);
2846 /* ixgbevf_get_stats() might access the rings on this vector,
2847 * we must wait a grace period before freeing it.
2849 kfree_rcu(q_vector, rcu);
2853 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2854 * @adapter: board private structure to initialize
2856 * We allocate one q_vector per queue interrupt. If allocation fails we
2859 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2861 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2862 int rxr_remaining = adapter->num_rx_queues;
2863 int txr_remaining = adapter->num_tx_queues;
2864 int xdp_remaining = adapter->num_xdp_queues;
2865 int rxr_idx = 0, txr_idx = 0, xdp_idx = 0, v_idx = 0;
2868 if (q_vectors >= (rxr_remaining + txr_remaining + xdp_remaining)) {
2869 for (; rxr_remaining; v_idx++, q_vectors--) {
2870 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2872 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2873 0, 0, 0, 0, rqpv, rxr_idx);
2877 /* update counts and index */
2878 rxr_remaining -= rqpv;
2883 for (; q_vectors; v_idx++, q_vectors--) {
2884 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2885 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors);
2886 int xqpv = DIV_ROUND_UP(xdp_remaining, q_vectors);
2888 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2896 /* update counts and index */
2897 rxr_remaining -= rqpv;
2899 txr_remaining -= tqpv;
2901 xdp_remaining -= xqpv;
2910 ixgbevf_free_q_vector(adapter, v_idx);
2917 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2918 * @adapter: board private structure to initialize
2920 * This function frees the memory allocated to the q_vectors. In addition if
2921 * NAPI is enabled it will delete any references to the NAPI struct prior
2922 * to freeing the q_vector.
2924 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2926 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2930 ixgbevf_free_q_vector(adapter, q_vectors);
2935 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2936 * @adapter: board private structure
2939 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2941 if (!adapter->msix_entries)
2944 pci_disable_msix(adapter->pdev);
2945 kfree(adapter->msix_entries);
2946 adapter->msix_entries = NULL;
2950 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2951 * @adapter: board private structure to initialize
2954 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2958 /* Number of supported queues */
2959 ixgbevf_set_num_queues(adapter);
2961 err = ixgbevf_set_interrupt_capability(adapter);
2963 hw_dbg(&adapter->hw,
2964 "Unable to setup interrupt capabilities\n");
2965 goto err_set_interrupt;
2968 err = ixgbevf_alloc_q_vectors(adapter);
2970 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2971 goto err_alloc_q_vectors;
2974 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u XDP Queue count %u\n",
2975 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
2976 adapter->num_rx_queues, adapter->num_tx_queues,
2977 adapter->num_xdp_queues);
2979 set_bit(__IXGBEVF_DOWN, &adapter->state);
2982 err_alloc_q_vectors:
2983 ixgbevf_reset_interrupt_capability(adapter);
2989 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2990 * @adapter: board private structure to clear interrupt scheme on
2992 * We go through and clear interrupt specific resources and reset the structure
2993 * to pre-load conditions
2995 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2997 adapter->num_tx_queues = 0;
2998 adapter->num_xdp_queues = 0;
2999 adapter->num_rx_queues = 0;
3001 ixgbevf_free_q_vectors(adapter);
3002 ixgbevf_reset_interrupt_capability(adapter);
3006 * ixgbevf_sw_init - Initialize general software structures
3007 * @adapter: board private structure to initialize
3009 * ixgbevf_sw_init initializes the Adapter private data structure.
3010 * Fields are initialized based on PCI device information and
3011 * OS network device settings (MTU size).
3013 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
3015 struct ixgbe_hw *hw = &adapter->hw;
3016 struct pci_dev *pdev = adapter->pdev;
3017 struct net_device *netdev = adapter->netdev;
3020 /* PCI config space info */
3021 hw->vendor_id = pdev->vendor;
3022 hw->device_id = pdev->device;
3023 hw->revision_id = pdev->revision;
3024 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3025 hw->subsystem_device_id = pdev->subsystem_device;
3027 hw->mbx.ops.init_params(hw);
3029 if (hw->mac.type >= ixgbe_mac_X550_vf) {
3030 err = ixgbevf_init_rss_key(adapter);
3035 /* assume legacy case in which PF would only give VF 2 queues */
3036 hw->mac.max_tx_queues = 2;
3037 hw->mac.max_rx_queues = 2;
3039 /* lock to protect mailbox accesses */
3040 spin_lock_init(&adapter->mbx_lock);
3042 err = hw->mac.ops.reset_hw(hw);
3044 dev_info(&pdev->dev,
3045 "PF still in reset state. Is the PF interface up?\n");
3047 err = hw->mac.ops.init_hw(hw);
3049 pr_err("init_shared_code failed: %d\n", err);
3052 ixgbevf_negotiate_api(adapter);
3053 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
3055 dev_info(&pdev->dev, "Error reading MAC address\n");
3056 else if (is_zero_ether_addr(adapter->hw.mac.addr))
3057 dev_info(&pdev->dev,
3058 "MAC address not assigned by administrator.\n");
3059 ether_addr_copy(netdev->dev_addr, hw->mac.addr);
3062 if (!is_valid_ether_addr(netdev->dev_addr)) {
3063 dev_info(&pdev->dev, "Assigning random MAC address\n");
3064 eth_hw_addr_random(netdev);
3065 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
3066 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
3069 /* Enable dynamic interrupt throttling rates */
3070 adapter->rx_itr_setting = 1;
3071 adapter->tx_itr_setting = 1;
3073 /* set default ring sizes */
3074 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
3075 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
3077 set_bit(__IXGBEVF_DOWN, &adapter->state);
3084 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
3086 u32 current_counter = IXGBE_READ_REG(hw, reg); \
3087 if (current_counter < last_counter) \
3088 counter += 0x100000000LL; \
3089 last_counter = current_counter; \
3090 counter &= 0xFFFFFFFF00000000LL; \
3091 counter |= current_counter; \
3094 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
3096 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
3097 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
3098 u64 current_counter = (current_counter_msb << 32) | \
3099 current_counter_lsb; \
3100 if (current_counter < last_counter) \
3101 counter += 0x1000000000LL; \
3102 last_counter = current_counter; \
3103 counter &= 0xFFFFFFF000000000LL; \
3104 counter |= current_counter; \
3107 * ixgbevf_update_stats - Update the board statistics counters.
3108 * @adapter: board private structure
3110 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
3112 struct ixgbe_hw *hw = &adapter->hw;
3113 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
3114 u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
3117 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3118 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3121 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
3122 adapter->stats.vfgprc);
3123 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
3124 adapter->stats.vfgptc);
3125 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
3126 adapter->stats.last_vfgorc,
3127 adapter->stats.vfgorc);
3128 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
3129 adapter->stats.last_vfgotc,
3130 adapter->stats.vfgotc);
3131 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
3132 adapter->stats.vfmprc);
3134 for (i = 0; i < adapter->num_rx_queues; i++) {
3135 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
3137 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
3138 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
3139 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
3140 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
3143 adapter->hw_csum_rx_error = hw_csum_rx_error;
3144 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
3145 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
3146 adapter->alloc_rx_page = alloc_rx_page;
3150 * ixgbevf_service_timer - Timer Call-back
3151 * @t: pointer to timer_list struct
3153 static void ixgbevf_service_timer(struct timer_list *t)
3155 struct ixgbevf_adapter *adapter = from_timer(adapter, t,
3158 /* Reset the timer */
3159 mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
3161 ixgbevf_service_event_schedule(adapter);
3164 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
3166 if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
3170 /* If we're already down or resetting, just bail */
3171 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3172 test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
3173 test_bit(__IXGBEVF_RESETTING, &adapter->state)) {
3178 adapter->tx_timeout_count++;
3180 ixgbevf_reinit_locked(adapter);
3185 * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
3186 * @adapter: pointer to the device adapter structure
3188 * This function serves two purposes. First it strobes the interrupt lines
3189 * in order to make certain interrupts are occurring. Secondly it sets the
3190 * bits needed to check for TX hangs. As a result we should immediately
3191 * determine if a hang has occurred.
3193 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
3195 struct ixgbe_hw *hw = &adapter->hw;
3199 /* If we're down or resetting, just bail */
3200 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3201 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3204 /* Force detection of hung controller */
3205 if (netif_carrier_ok(adapter->netdev)) {
3206 for (i = 0; i < adapter->num_tx_queues; i++)
3207 set_check_for_tx_hang(adapter->tx_ring[i]);
3208 for (i = 0; i < adapter->num_xdp_queues; i++)
3209 set_check_for_tx_hang(adapter->xdp_ring[i]);
3212 /* get one bit for every active Tx/Rx interrupt vector */
3213 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
3214 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
3216 if (qv->rx.ring || qv->tx.ring)
3220 /* Cause software interrupt to ensure rings are cleaned */
3221 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
3225 * ixgbevf_watchdog_update_link - update the link status
3226 * @adapter: pointer to the device adapter structure
3228 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
3230 struct ixgbe_hw *hw = &adapter->hw;
3231 u32 link_speed = adapter->link_speed;
3232 bool link_up = adapter->link_up;
3235 spin_lock_bh(&adapter->mbx_lock);
3237 err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3239 spin_unlock_bh(&adapter->mbx_lock);
3241 /* if check for link returns error we will need to reset */
3242 if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
3243 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
3247 adapter->link_up = link_up;
3248 adapter->link_speed = link_speed;
3252 * ixgbevf_watchdog_link_is_up - update netif_carrier status and
3253 * print link up message
3254 * @adapter: pointer to the device adapter structure
3256 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
3258 struct net_device *netdev = adapter->netdev;
3260 /* only continue if link was previously down */
3261 if (netif_carrier_ok(netdev))
3264 dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
3265 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
3267 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
3269 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
3273 netif_carrier_on(netdev);
3277 * ixgbevf_watchdog_link_is_down - update netif_carrier status and
3278 * print link down message
3279 * @adapter: pointer to the adapter structure
3281 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
3283 struct net_device *netdev = adapter->netdev;
3285 adapter->link_speed = 0;
3287 /* only continue if link was up previously */
3288 if (!netif_carrier_ok(netdev))
3291 dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
3293 netif_carrier_off(netdev);
3297 * ixgbevf_watchdog_subtask - worker thread to bring link up
3298 * @adapter: board private structure
3300 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
3302 /* if interface is down do nothing */
3303 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3304 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3307 ixgbevf_watchdog_update_link(adapter);
3309 if (adapter->link_up)
3310 ixgbevf_watchdog_link_is_up(adapter);
3312 ixgbevf_watchdog_link_is_down(adapter);
3314 ixgbevf_update_stats(adapter);
3318 * ixgbevf_service_task - manages and runs subtasks
3319 * @work: pointer to work_struct containing our data
3321 static void ixgbevf_service_task(struct work_struct *work)
3323 struct ixgbevf_adapter *adapter = container_of(work,
3324 struct ixgbevf_adapter,
3326 struct ixgbe_hw *hw = &adapter->hw;
3328 if (IXGBE_REMOVED(hw->hw_addr)) {
3329 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
3331 ixgbevf_down(adapter);
3337 ixgbevf_queue_reset_subtask(adapter);
3338 ixgbevf_reset_subtask(adapter);
3339 ixgbevf_watchdog_subtask(adapter);
3340 ixgbevf_check_hang_subtask(adapter);
3342 ixgbevf_service_event_complete(adapter);
3346 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
3347 * @tx_ring: Tx descriptor ring for a specific queue
3349 * Free all transmit software resources
3351 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
3353 ixgbevf_clean_tx_ring(tx_ring);
3355 vfree(tx_ring->tx_buffer_info);
3356 tx_ring->tx_buffer_info = NULL;
3358 /* if not set, then don't free */
3362 dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
3365 tx_ring->desc = NULL;
3369 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
3370 * @adapter: board private structure
3372 * Free all transmit software resources
3374 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
3378 for (i = 0; i < adapter->num_tx_queues; i++)
3379 if (adapter->tx_ring[i]->desc)
3380 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3381 for (i = 0; i < adapter->num_xdp_queues; i++)
3382 if (adapter->xdp_ring[i]->desc)
3383 ixgbevf_free_tx_resources(adapter->xdp_ring[i]);
3387 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
3388 * @tx_ring: Tx descriptor ring (for a specific queue) to setup
3390 * Return 0 on success, negative on failure
3392 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
3394 struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3397 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3398 tx_ring->tx_buffer_info = vmalloc(size);
3399 if (!tx_ring->tx_buffer_info)
3402 u64_stats_init(&tx_ring->syncp);
3404 /* round up to nearest 4K */
3405 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3406 tx_ring->size = ALIGN(tx_ring->size, 4096);
3408 tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3409 &tx_ring->dma, GFP_KERNEL);
3416 vfree(tx_ring->tx_buffer_info);
3417 tx_ring->tx_buffer_info = NULL;
3418 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3423 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3424 * @adapter: board private structure
3426 * If this function returns with an error, then it's possible one or
3427 * more of the rings is populated (while the rest are not). It is the
3428 * callers duty to clean those orphaned rings.
3430 * Return 0 on success, negative on failure
3432 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3434 int i, j = 0, err = 0;
3436 for (i = 0; i < adapter->num_tx_queues; i++) {
3437 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3440 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3444 for (j = 0; j < adapter->num_xdp_queues; j++) {
3445 err = ixgbevf_setup_tx_resources(adapter->xdp_ring[j]);
3448 hw_dbg(&adapter->hw, "Allocation for XDP Queue %u failed\n", j);
3454 /* rewind the index freeing the rings as we go */
3456 ixgbevf_free_tx_resources(adapter->xdp_ring[j]);
3458 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3464 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3465 * @adapter: board private structure
3466 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3468 * Returns 0 on success, negative on failure
3470 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
3471 struct ixgbevf_ring *rx_ring)
3475 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3476 rx_ring->rx_buffer_info = vmalloc(size);
3477 if (!rx_ring->rx_buffer_info)
3480 u64_stats_init(&rx_ring->syncp);
3482 /* Round up to nearest 4K */
3483 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3484 rx_ring->size = ALIGN(rx_ring->size, 4096);
3486 rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3487 &rx_ring->dma, GFP_KERNEL);
3492 /* XDP RX-queue info */
3493 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
3494 rx_ring->queue_index, 0) < 0)
3497 rx_ring->xdp_prog = adapter->xdp_prog;
3501 vfree(rx_ring->rx_buffer_info);
3502 rx_ring->rx_buffer_info = NULL;
3503 dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3508 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3509 * @adapter: board private structure
3511 * If this function returns with an error, then it's possible one or
3512 * more of the rings is populated (while the rest are not). It is the
3513 * callers duty to clean those orphaned rings.
3515 * Return 0 on success, negative on failure
3517 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3521 for (i = 0; i < adapter->num_rx_queues; i++) {
3522 err = ixgbevf_setup_rx_resources(adapter, adapter->rx_ring[i]);
3525 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3531 /* rewind the index freeing the rings as we go */
3533 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3538 * ixgbevf_free_rx_resources - Free Rx Resources
3539 * @rx_ring: ring to clean the resources from
3541 * Free all receive software resources
3543 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3545 ixgbevf_clean_rx_ring(rx_ring);
3547 rx_ring->xdp_prog = NULL;
3548 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
3549 vfree(rx_ring->rx_buffer_info);
3550 rx_ring->rx_buffer_info = NULL;
3552 dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3555 rx_ring->desc = NULL;
3559 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3560 * @adapter: board private structure
3562 * Free all receive software resources
3564 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3568 for (i = 0; i < adapter->num_rx_queues; i++)
3569 if (adapter->rx_ring[i]->desc)
3570 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3574 * ixgbevf_open - Called when a network interface is made active
3575 * @netdev: network interface device structure
3577 * Returns 0 on success, negative value on failure
3579 * The open entry point is called when a network interface is made
3580 * active by the system (IFF_UP). At this point all resources needed
3581 * for transmit and receive operations are allocated, the interrupt
3582 * handler is registered with the OS, the watchdog timer is started,
3583 * and the stack is notified that the interface is ready.
3585 int ixgbevf_open(struct net_device *netdev)
3587 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3588 struct ixgbe_hw *hw = &adapter->hw;
3591 /* A previous failure to open the device because of a lack of
3592 * available MSIX vector resources may have reset the number
3593 * of msix vectors variable to zero. The only way to recover
3594 * is to unload/reload the driver and hope that the system has
3595 * been able to recover some MSIX vector resources.
3597 if (!adapter->num_msix_vectors)
3600 if (hw->adapter_stopped) {
3601 ixgbevf_reset(adapter);
3602 /* if adapter is still stopped then PF isn't up and
3603 * the VF can't start.
3605 if (hw->adapter_stopped) {
3606 err = IXGBE_ERR_MBX;
3607 pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3608 goto err_setup_reset;
3612 /* disallow open during test */
3613 if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3616 netif_carrier_off(netdev);
3618 /* allocate transmit descriptors */
3619 err = ixgbevf_setup_all_tx_resources(adapter);
3623 /* allocate receive descriptors */
3624 err = ixgbevf_setup_all_rx_resources(adapter);
3628 ixgbevf_configure(adapter);
3630 err = ixgbevf_request_irq(adapter);
3634 /* Notify the stack of the actual queue counts. */
3635 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
3637 goto err_set_queues;
3639 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
3641 goto err_set_queues;
3643 ixgbevf_up_complete(adapter);
3648 ixgbevf_free_irq(adapter);
3650 ixgbevf_free_all_rx_resources(adapter);
3652 ixgbevf_free_all_tx_resources(adapter);
3654 ixgbevf_reset(adapter);
3661 * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3662 * @adapter: the private adapter struct
3664 * This function should contain the necessary work common to both suspending
3665 * and closing of the device.
3667 static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3669 ixgbevf_down(adapter);
3670 ixgbevf_free_irq(adapter);
3671 ixgbevf_free_all_tx_resources(adapter);
3672 ixgbevf_free_all_rx_resources(adapter);
3676 * ixgbevf_close - Disables a network interface
3677 * @netdev: network interface device structure
3679 * Returns 0, this is not allowed to fail
3681 * The close entry point is called when an interface is de-activated
3682 * by the OS. The hardware is still under the drivers control, but
3683 * needs to be disabled. A global MAC reset is issued to stop the
3684 * hardware, and all transmit and receive resources are freed.
3686 int ixgbevf_close(struct net_device *netdev)
3688 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3690 if (netif_device_present(netdev))
3691 ixgbevf_close_suspend(adapter);
3696 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3698 struct net_device *dev = adapter->netdev;
3700 if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3704 /* if interface is down do nothing */
3705 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3706 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3709 /* Hardware has to reinitialize queues and interrupts to
3710 * match packet buffer alignment. Unfortunately, the
3711 * hardware is not flexible enough to do this dynamically.
3715 if (netif_running(dev))
3718 ixgbevf_clear_interrupt_scheme(adapter);
3719 ixgbevf_init_interrupt_scheme(adapter);
3721 if (netif_running(dev))
3727 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3728 u32 vlan_macip_lens, u32 fceof_saidx,
3729 u32 type_tucmd, u32 mss_l4len_idx)
3731 struct ixgbe_adv_tx_context_desc *context_desc;
3732 u16 i = tx_ring->next_to_use;
3734 context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3737 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3739 /* set bits to identify this as an advanced context descriptor */
3740 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3742 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3743 context_desc->fceof_saidx = cpu_to_le32(fceof_saidx);
3744 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
3745 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3748 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3749 struct ixgbevf_tx_buffer *first,
3751 struct ixgbevf_ipsec_tx_data *itd)
3753 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3754 struct sk_buff *skb = first->skb;
3764 u32 paylen, l4_offset;
3765 u32 fceof_saidx = 0;
3768 if (skb->ip_summed != CHECKSUM_PARTIAL)
3771 if (!skb_is_gso(skb))
3774 err = skb_cow_head(skb, 0);
3778 if (eth_p_mpls(first->protocol))
3779 ip.hdr = skb_inner_network_header(skb);
3781 ip.hdr = skb_network_header(skb);
3782 l4.hdr = skb_checksum_start(skb);
3784 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3785 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3787 /* initialize outer IP header fields */
3788 if (ip.v4->version == 4) {
3789 unsigned char *csum_start = skb_checksum_start(skb);
3790 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3791 int len = csum_start - trans_start;
3793 /* IP header will have to cancel out any data that
3794 * is not a part of the outer IP header, so set to
3795 * a reverse csum if needed, else init check to 0.
3797 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
3798 csum_fold(csum_partial(trans_start,
3800 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3803 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3804 IXGBE_TX_FLAGS_CSUM |
3805 IXGBE_TX_FLAGS_IPV4;
3807 ip.v6->payload_len = 0;
3808 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3809 IXGBE_TX_FLAGS_CSUM;
3812 /* determine offset of inner transport header */
3813 l4_offset = l4.hdr - skb->data;
3815 /* compute length of segmentation header */
3816 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3818 /* remove payload length from inner checksum */
3819 paylen = skb->len - l4_offset;
3820 csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
3822 /* update gso size and bytecount with header size */
3823 first->gso_segs = skb_shinfo(skb)->gso_segs;
3824 first->bytecount += (first->gso_segs - 1) * *hdr_len;
3826 /* mss_l4len_id: use 1 as index for TSO */
3827 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3828 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3829 mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3831 fceof_saidx |= itd->pfsa;
3832 type_tucmd |= itd->flags | itd->trailer_len;
3834 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3835 vlan_macip_lens = l4.hdr - ip.hdr;
3836 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3837 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3839 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
3845 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3846 struct ixgbevf_tx_buffer *first,
3847 struct ixgbevf_ipsec_tx_data *itd)
3849 struct sk_buff *skb = first->skb;
3850 u32 vlan_macip_lens = 0;
3851 u32 fceof_saidx = 0;
3854 if (skb->ip_summed != CHECKSUM_PARTIAL)
3857 switch (skb->csum_offset) {
3858 case offsetof(struct tcphdr, check):
3859 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3861 case offsetof(struct udphdr, check):
3863 case offsetof(struct sctphdr, checksum):
3864 /* validate that this is actually an SCTP request */
3865 if (skb_csum_is_sctp(skb)) {
3866 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3871 skb_checksum_help(skb);
3875 if (first->protocol == htons(ETH_P_IP))
3876 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3878 /* update TX checksum flag */
3879 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3880 vlan_macip_lens = skb_checksum_start_offset(skb) -
3881 skb_network_offset(skb);
3883 /* vlan_macip_lens: MACLEN, VLAN tag */
3884 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3885 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3887 fceof_saidx |= itd->pfsa;
3888 type_tucmd |= itd->flags | itd->trailer_len;
3890 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3891 fceof_saidx, type_tucmd, 0);
3894 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3896 /* set type for advanced descriptor with frame checksum insertion */
3897 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3898 IXGBE_ADVTXD_DCMD_IFCS |
3899 IXGBE_ADVTXD_DCMD_DEXT);
3901 /* set HW VLAN bit if VLAN is present */
3902 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3903 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3905 /* set segmentation enable bits for TSO/FSO */
3906 if (tx_flags & IXGBE_TX_FLAGS_TSO)
3907 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3912 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3913 u32 tx_flags, unsigned int paylen)
3915 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3917 /* enable L4 checksum for TSO and TX checksum offload */
3918 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3919 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3921 /* enble IPv4 checksum for TSO */
3922 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3923 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3926 if (tx_flags & IXGBE_TX_FLAGS_IPSEC)
3927 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IPSEC);
3929 /* use index 1 context for TSO/FSO/FCOE/IPSEC */
3930 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_IPSEC))
3931 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3933 /* Check Context must be set if Tx switch is enabled, which it
3934 * always is for case where virtual functions are running
3936 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3938 tx_desc->read.olinfo_status = olinfo_status;
3941 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3942 struct ixgbevf_tx_buffer *first,
3945 struct sk_buff *skb = first->skb;
3946 struct ixgbevf_tx_buffer *tx_buffer;
3947 union ixgbe_adv_tx_desc *tx_desc;
3950 unsigned int data_len, size;
3951 u32 tx_flags = first->tx_flags;
3952 __le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3953 u16 i = tx_ring->next_to_use;
3955 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3957 ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
3959 size = skb_headlen(skb);
3960 data_len = skb->data_len;
3962 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3966 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3967 if (dma_mapping_error(tx_ring->dev, dma))
3970 /* record length, and DMA address */
3971 dma_unmap_len_set(tx_buffer, len, size);
3972 dma_unmap_addr_set(tx_buffer, dma, dma);
3974 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3976 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3977 tx_desc->read.cmd_type_len =
3978 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3982 if (i == tx_ring->count) {
3983 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3986 tx_desc->read.olinfo_status = 0;
3988 dma += IXGBE_MAX_DATA_PER_TXD;
3989 size -= IXGBE_MAX_DATA_PER_TXD;
3991 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3994 if (likely(!data_len))
3997 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
4001 if (i == tx_ring->count) {
4002 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
4005 tx_desc->read.olinfo_status = 0;
4007 size = skb_frag_size(frag);
4010 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
4013 tx_buffer = &tx_ring->tx_buffer_info[i];
4016 /* write last descriptor with RS and EOP bits */
4017 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
4018 tx_desc->read.cmd_type_len = cmd_type;
4020 /* set the timestamp */
4021 first->time_stamp = jiffies;
4023 skb_tx_timestamp(skb);
4025 /* Force memory writes to complete before letting h/w know there
4026 * are new descriptors to fetch. (Only applicable for weak-ordered
4027 * memory model archs, such as IA-64).
4029 * We also need this memory barrier (wmb) to make certain all of the
4030 * status bits have been updated before next_to_watch is written.
4034 /* set next_to_watch value indicating a packet is present */
4035 first->next_to_watch = tx_desc;
4038 if (i == tx_ring->count)
4041 tx_ring->next_to_use = i;
4043 /* notify HW of packet */
4044 ixgbevf_write_tail(tx_ring, i);
4048 dev_err(tx_ring->dev, "TX DMA map failed\n");
4049 tx_buffer = &tx_ring->tx_buffer_info[i];
4051 /* clear dma mappings for failed tx_buffer_info map */
4052 while (tx_buffer != first) {
4053 if (dma_unmap_len(tx_buffer, len))
4054 dma_unmap_page(tx_ring->dev,
4055 dma_unmap_addr(tx_buffer, dma),
4056 dma_unmap_len(tx_buffer, len),
4058 dma_unmap_len_set(tx_buffer, len, 0);
4061 i += tx_ring->count;
4062 tx_buffer = &tx_ring->tx_buffer_info[i];
4065 if (dma_unmap_len(tx_buffer, len))
4066 dma_unmap_single(tx_ring->dev,
4067 dma_unmap_addr(tx_buffer, dma),
4068 dma_unmap_len(tx_buffer, len),
4070 dma_unmap_len_set(tx_buffer, len, 0);
4072 dev_kfree_skb_any(tx_buffer->skb);
4073 tx_buffer->skb = NULL;
4075 tx_ring->next_to_use = i;
4078 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4080 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
4081 /* Herbert's original patch had:
4082 * smp_mb__after_netif_stop_queue();
4083 * but since that doesn't exist yet, just open code it.
4087 /* We need to check again in a case another CPU has just
4088 * made room available.
4090 if (likely(ixgbevf_desc_unused(tx_ring) < size))
4093 /* A reprieve! - use start_queue because it doesn't call schedule */
4094 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
4095 ++tx_ring->tx_stats.restart_queue;
4100 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4102 if (likely(ixgbevf_desc_unused(tx_ring) >= size))
4104 return __ixgbevf_maybe_stop_tx(tx_ring, size);
4107 static int ixgbevf_xmit_frame_ring(struct sk_buff *skb,
4108 struct ixgbevf_ring *tx_ring)
4110 struct ixgbevf_tx_buffer *first;
4113 u16 count = TXD_USE_COUNT(skb_headlen(skb));
4114 struct ixgbevf_ipsec_tx_data ipsec_tx = { 0 };
4115 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4119 u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
4121 if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
4122 dev_kfree_skb_any(skb);
4123 return NETDEV_TX_OK;
4126 /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
4127 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
4128 * + 2 desc gap to keep tail from touching head,
4129 * + 1 desc for context descriptor,
4130 * otherwise try next time
4132 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4133 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
4134 skb_frag_t *frag = &skb_shinfo(skb)->frags[f];
4136 count += TXD_USE_COUNT(skb_frag_size(frag));
4139 count += skb_shinfo(skb)->nr_frags;
4141 if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
4142 tx_ring->tx_stats.tx_busy++;
4143 return NETDEV_TX_BUSY;
4146 /* record the location of the first descriptor for this packet */
4147 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4149 first->bytecount = skb->len;
4150 first->gso_segs = 1;
4152 if (skb_vlan_tag_present(skb)) {
4153 tx_flags |= skb_vlan_tag_get(skb);
4154 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4155 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4158 /* record initial flags and protocol */
4159 first->tx_flags = tx_flags;
4160 first->protocol = vlan_get_protocol(skb);
4162 #ifdef CONFIG_IXGBEVF_IPSEC
4163 if (xfrm_offload(skb) && !ixgbevf_ipsec_tx(tx_ring, first, &ipsec_tx))
4166 tso = ixgbevf_tso(tx_ring, first, &hdr_len, &ipsec_tx);
4170 ixgbevf_tx_csum(tx_ring, first, &ipsec_tx);
4172 ixgbevf_tx_map(tx_ring, first, hdr_len);
4174 ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
4176 return NETDEV_TX_OK;
4179 dev_kfree_skb_any(first->skb);
4182 return NETDEV_TX_OK;
4185 static netdev_tx_t ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4187 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4188 struct ixgbevf_ring *tx_ring;
4190 if (skb->len <= 0) {
4191 dev_kfree_skb_any(skb);
4192 return NETDEV_TX_OK;
4195 /* The minimum packet size for olinfo paylen is 17 so pad the skb
4196 * in order to meet this minimum size requirement.
4198 if (skb->len < 17) {
4199 if (skb_padto(skb, 17))
4200 return NETDEV_TX_OK;
4204 tx_ring = adapter->tx_ring[skb->queue_mapping];
4205 return ixgbevf_xmit_frame_ring(skb, tx_ring);
4209 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
4210 * @netdev: network interface device structure
4211 * @p: pointer to an address structure
4213 * Returns 0 on success, negative on failure
4215 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
4217 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4218 struct ixgbe_hw *hw = &adapter->hw;
4219 struct sockaddr *addr = p;
4222 if (!is_valid_ether_addr(addr->sa_data))
4223 return -EADDRNOTAVAIL;
4225 spin_lock_bh(&adapter->mbx_lock);
4227 err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
4229 spin_unlock_bh(&adapter->mbx_lock);
4234 ether_addr_copy(hw->mac.addr, addr->sa_data);
4235 ether_addr_copy(hw->mac.perm_addr, addr->sa_data);
4236 ether_addr_copy(netdev->dev_addr, addr->sa_data);
4242 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
4243 * @netdev: network interface device structure
4244 * @new_mtu: new value for maximum frame size
4246 * Returns 0 on success, negative on failure
4248 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
4250 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4251 struct ixgbe_hw *hw = &adapter->hw;
4252 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4255 /* prevent MTU being changed to a size unsupported by XDP */
4256 if (adapter->xdp_prog) {
4257 dev_warn(&adapter->pdev->dev, "MTU cannot be changed while XDP program is loaded\n");
4261 spin_lock_bh(&adapter->mbx_lock);
4262 /* notify the PF of our intent to use this size of frame */
4263 ret = hw->mac.ops.set_rlpml(hw, max_frame);
4264 spin_unlock_bh(&adapter->mbx_lock);
4268 hw_dbg(hw, "changing MTU from %d to %d\n",
4269 netdev->mtu, new_mtu);
4271 /* must set new MTU before calling down or up */
4272 netdev->mtu = new_mtu;
4274 if (netif_running(netdev))
4275 ixgbevf_reinit_locked(adapter);
4280 static int __maybe_unused ixgbevf_suspend(struct device *dev_d)
4282 struct net_device *netdev = dev_get_drvdata(dev_d);
4283 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4286 netif_device_detach(netdev);
4288 if (netif_running(netdev))
4289 ixgbevf_close_suspend(adapter);
4291 ixgbevf_clear_interrupt_scheme(adapter);
4297 static int __maybe_unused ixgbevf_resume(struct device *dev_d)
4299 struct pci_dev *pdev = to_pci_dev(dev_d);
4300 struct net_device *netdev = pci_get_drvdata(pdev);
4301 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4304 adapter->hw.hw_addr = adapter->io_addr;
4305 smp_mb__before_atomic();
4306 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4307 pci_set_master(pdev);
4309 ixgbevf_reset(adapter);
4312 err = ixgbevf_init_interrupt_scheme(adapter);
4313 if (!err && netif_running(netdev))
4314 err = ixgbevf_open(netdev);
4319 netif_device_attach(netdev);
4324 static void ixgbevf_shutdown(struct pci_dev *pdev)
4326 ixgbevf_suspend(&pdev->dev);
4329 static void ixgbevf_get_tx_ring_stats(struct rtnl_link_stats64 *stats,
4330 const struct ixgbevf_ring *ring)
4337 start = u64_stats_fetch_begin_irq(&ring->syncp);
4338 bytes = ring->stats.bytes;
4339 packets = ring->stats.packets;
4340 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4341 stats->tx_bytes += bytes;
4342 stats->tx_packets += packets;
4346 static void ixgbevf_get_stats(struct net_device *netdev,
4347 struct rtnl_link_stats64 *stats)
4349 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4352 const struct ixgbevf_ring *ring;
4355 ixgbevf_update_stats(adapter);
4357 stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
4360 for (i = 0; i < adapter->num_rx_queues; i++) {
4361 ring = adapter->rx_ring[i];
4363 start = u64_stats_fetch_begin_irq(&ring->syncp);
4364 bytes = ring->stats.bytes;
4365 packets = ring->stats.packets;
4366 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4367 stats->rx_bytes += bytes;
4368 stats->rx_packets += packets;
4371 for (i = 0; i < adapter->num_tx_queues; i++) {
4372 ring = adapter->tx_ring[i];
4373 ixgbevf_get_tx_ring_stats(stats, ring);
4376 for (i = 0; i < adapter->num_xdp_queues; i++) {
4377 ring = adapter->xdp_ring[i];
4378 ixgbevf_get_tx_ring_stats(stats, ring);
4383 #define IXGBEVF_MAX_MAC_HDR_LEN 127
4384 #define IXGBEVF_MAX_NETWORK_HDR_LEN 511
4386 static netdev_features_t
4387 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
4388 netdev_features_t features)
4390 unsigned int network_hdr_len, mac_hdr_len;
4392 /* Make certain the headers can be described by a context descriptor */
4393 mac_hdr_len = skb_network_header(skb) - skb->data;
4394 if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
4395 return features & ~(NETIF_F_HW_CSUM |
4397 NETIF_F_HW_VLAN_CTAG_TX |
4401 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
4402 if (unlikely(network_hdr_len > IXGBEVF_MAX_NETWORK_HDR_LEN))
4403 return features & ~(NETIF_F_HW_CSUM |
4408 /* We can only support IPV4 TSO in tunnels if we can mangle the
4409 * inner IP ID field, so strip TSO if MANGLEID is not supported.
4411 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
4412 features &= ~NETIF_F_TSO;
4417 static int ixgbevf_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
4419 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4420 struct ixgbevf_adapter *adapter = netdev_priv(dev);
4421 struct bpf_prog *old_prog;
4423 /* verify ixgbevf ring attributes are sufficient for XDP */
4424 for (i = 0; i < adapter->num_rx_queues; i++) {
4425 struct ixgbevf_ring *ring = adapter->rx_ring[i];
4427 if (frame_size > ixgbevf_rx_bufsz(ring))
4431 old_prog = xchg(&adapter->xdp_prog, prog);
4433 /* If transitioning XDP modes reconfigure rings */
4434 if (!!prog != !!old_prog) {
4435 /* Hardware has to reinitialize queues and interrupts to
4436 * match packet buffer alignment. Unfortunately, the
4437 * hardware is not flexible enough to do this dynamically.
4439 if (netif_running(dev))
4442 ixgbevf_clear_interrupt_scheme(adapter);
4443 ixgbevf_init_interrupt_scheme(adapter);
4445 if (netif_running(dev))
4448 for (i = 0; i < adapter->num_rx_queues; i++)
4449 xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
4453 bpf_prog_put(old_prog);
4458 static int ixgbevf_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4460 switch (xdp->command) {
4461 case XDP_SETUP_PROG:
4462 return ixgbevf_xdp_setup(dev, xdp->prog);
4468 static const struct net_device_ops ixgbevf_netdev_ops = {
4469 .ndo_open = ixgbevf_open,
4470 .ndo_stop = ixgbevf_close,
4471 .ndo_start_xmit = ixgbevf_xmit_frame,
4472 .ndo_set_rx_mode = ixgbevf_set_rx_mode,
4473 .ndo_get_stats64 = ixgbevf_get_stats,
4474 .ndo_validate_addr = eth_validate_addr,
4475 .ndo_set_mac_address = ixgbevf_set_mac,
4476 .ndo_change_mtu = ixgbevf_change_mtu,
4477 .ndo_tx_timeout = ixgbevf_tx_timeout,
4478 .ndo_vlan_rx_add_vid = ixgbevf_vlan_rx_add_vid,
4479 .ndo_vlan_rx_kill_vid = ixgbevf_vlan_rx_kill_vid,
4480 .ndo_features_check = ixgbevf_features_check,
4481 .ndo_bpf = ixgbevf_xdp,
4484 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
4486 dev->netdev_ops = &ixgbevf_netdev_ops;
4487 ixgbevf_set_ethtool_ops(dev);
4488 dev->watchdog_timeo = 5 * HZ;
4492 * ixgbevf_probe - Device Initialization Routine
4493 * @pdev: PCI device information struct
4494 * @ent: entry in ixgbevf_pci_tbl
4496 * Returns 0 on success, negative on failure
4498 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
4499 * The OS initialization, configuring of the adapter private structure,
4500 * and a hardware reset occur.
4502 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4504 struct net_device *netdev;
4505 struct ixgbevf_adapter *adapter = NULL;
4506 struct ixgbe_hw *hw = NULL;
4507 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
4508 int err, pci_using_dac;
4509 bool disable_dev = false;
4511 err = pci_enable_device(pdev);
4515 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
4518 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
4520 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4526 err = pci_request_regions(pdev, ixgbevf_driver_name);
4528 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4532 pci_set_master(pdev);
4534 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4538 goto err_alloc_etherdev;
4541 SET_NETDEV_DEV(netdev, &pdev->dev);
4543 adapter = netdev_priv(netdev);
4545 adapter->netdev = netdev;
4546 adapter->pdev = pdev;
4549 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4551 /* call save state here in standalone driver because it relies on
4552 * adapter struct to exist, and needs to call netdev_priv
4554 pci_save_state(pdev);
4556 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4557 pci_resource_len(pdev, 0));
4558 adapter->io_addr = hw->hw_addr;
4564 ixgbevf_assign_netdev_ops(netdev);
4567 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4568 hw->mac.type = ii->mac;
4570 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4571 sizeof(struct ixgbe_mbx_operations));
4573 /* setup the private structure */
4574 err = ixgbevf_sw_init(adapter);
4578 /* The HW MAC address was set and/or determined in sw_init */
4579 if (!is_valid_ether_addr(netdev->dev_addr)) {
4580 pr_err("invalid MAC address\n");
4585 netdev->hw_features = NETIF_F_SG |
4592 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4593 NETIF_F_GSO_GRE_CSUM | \
4594 NETIF_F_GSO_IPXIP4 | \
4595 NETIF_F_GSO_IPXIP6 | \
4596 NETIF_F_GSO_UDP_TUNNEL | \
4597 NETIF_F_GSO_UDP_TUNNEL_CSUM)
4599 netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4600 netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4601 IXGBEVF_GSO_PARTIAL_FEATURES;
4603 netdev->features = netdev->hw_features;
4606 netdev->features |= NETIF_F_HIGHDMA;
4608 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4609 netdev->mpls_features |= NETIF_F_SG |
4613 netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES;
4614 netdev->hw_enc_features |= netdev->vlan_features;
4616 /* set this bit last since it cannot be part of vlan_features */
4617 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4618 NETIF_F_HW_VLAN_CTAG_RX |
4619 NETIF_F_HW_VLAN_CTAG_TX;
4621 netdev->priv_flags |= IFF_UNICAST_FLT;
4623 /* MTU range: 68 - 1504 or 9710 */
4624 netdev->min_mtu = ETH_MIN_MTU;
4625 switch (adapter->hw.api_version) {
4626 case ixgbe_mbox_api_11:
4627 case ixgbe_mbox_api_12:
4628 case ixgbe_mbox_api_13:
4629 case ixgbe_mbox_api_14:
4630 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4631 (ETH_HLEN + ETH_FCS_LEN);
4634 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4635 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4636 (ETH_HLEN + ETH_FCS_LEN);
4638 netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4642 if (IXGBE_REMOVED(hw->hw_addr)) {
4647 timer_setup(&adapter->service_timer, ixgbevf_service_timer, 0);
4649 INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4650 set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4651 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4653 err = ixgbevf_init_interrupt_scheme(adapter);
4657 strcpy(netdev->name, "eth%d");
4659 err = register_netdev(netdev);
4663 pci_set_drvdata(pdev, netdev);
4664 netif_carrier_off(netdev);
4665 ixgbevf_init_ipsec_offload(adapter);
4667 ixgbevf_init_last_counter_stats(adapter);
4669 /* print the VF info */
4670 dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4671 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4673 switch (hw->mac.type) {
4674 case ixgbe_mac_X550_vf:
4675 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4677 case ixgbe_mac_X540_vf:
4678 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4680 case ixgbe_mac_82599_vf:
4682 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4689 ixgbevf_clear_interrupt_scheme(adapter);
4691 ixgbevf_reset_interrupt_capability(adapter);
4692 iounmap(adapter->io_addr);
4693 kfree(adapter->rss_key);
4695 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4696 free_netdev(netdev);
4698 pci_release_regions(pdev);
4701 if (!adapter || disable_dev)
4702 pci_disable_device(pdev);
4707 * ixgbevf_remove - Device Removal Routine
4708 * @pdev: PCI device information struct
4710 * ixgbevf_remove is called by the PCI subsystem to alert the driver
4711 * that it should release a PCI device. The could be caused by a
4712 * Hot-Plug event, or because the driver is going to be removed from
4715 static void ixgbevf_remove(struct pci_dev *pdev)
4717 struct net_device *netdev = pci_get_drvdata(pdev);
4718 struct ixgbevf_adapter *adapter;
4724 adapter = netdev_priv(netdev);
4726 set_bit(__IXGBEVF_REMOVING, &adapter->state);
4727 cancel_work_sync(&adapter->service_task);
4729 if (netdev->reg_state == NETREG_REGISTERED)
4730 unregister_netdev(netdev);
4732 ixgbevf_stop_ipsec_offload(adapter);
4733 ixgbevf_clear_interrupt_scheme(adapter);
4734 ixgbevf_reset_interrupt_capability(adapter);
4736 iounmap(adapter->io_addr);
4737 pci_release_regions(pdev);
4739 hw_dbg(&adapter->hw, "Remove complete\n");
4741 kfree(adapter->rss_key);
4742 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4743 free_netdev(netdev);
4746 pci_disable_device(pdev);
4750 * ixgbevf_io_error_detected - called when PCI error is detected
4751 * @pdev: Pointer to PCI device
4752 * @state: The current pci connection state
4754 * This function is called after a PCI bus error affecting
4755 * this device has been detected.
4757 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4758 pci_channel_state_t state)
4760 struct net_device *netdev = pci_get_drvdata(pdev);
4761 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4763 if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4764 return PCI_ERS_RESULT_DISCONNECT;
4767 netif_device_detach(netdev);
4769 if (netif_running(netdev))
4770 ixgbevf_close_suspend(adapter);
4772 if (state == pci_channel_io_perm_failure) {
4774 return PCI_ERS_RESULT_DISCONNECT;
4777 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4778 pci_disable_device(pdev);
4781 /* Request a slot slot reset. */
4782 return PCI_ERS_RESULT_NEED_RESET;
4786 * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4787 * @pdev: Pointer to PCI device
4789 * Restart the card from scratch, as if from a cold-boot. Implementation
4790 * resembles the first-half of the ixgbevf_resume routine.
4792 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4794 struct net_device *netdev = pci_get_drvdata(pdev);
4795 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4797 if (pci_enable_device_mem(pdev)) {
4799 "Cannot re-enable PCI device after reset.\n");
4800 return PCI_ERS_RESULT_DISCONNECT;
4803 adapter->hw.hw_addr = adapter->io_addr;
4804 smp_mb__before_atomic();
4805 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4806 pci_set_master(pdev);
4808 ixgbevf_reset(adapter);
4810 return PCI_ERS_RESULT_RECOVERED;
4814 * ixgbevf_io_resume - called when traffic can start flowing again.
4815 * @pdev: Pointer to PCI device
4817 * This callback is called when the error recovery driver tells us that
4818 * its OK to resume normal operation. Implementation resembles the
4819 * second-half of the ixgbevf_resume routine.
4821 static void ixgbevf_io_resume(struct pci_dev *pdev)
4823 struct net_device *netdev = pci_get_drvdata(pdev);
4826 if (netif_running(netdev))
4827 ixgbevf_open(netdev);
4829 netif_device_attach(netdev);
4833 /* PCI Error Recovery (ERS) */
4834 static const struct pci_error_handlers ixgbevf_err_handler = {
4835 .error_detected = ixgbevf_io_error_detected,
4836 .slot_reset = ixgbevf_io_slot_reset,
4837 .resume = ixgbevf_io_resume,
4840 static SIMPLE_DEV_PM_OPS(ixgbevf_pm_ops, ixgbevf_suspend, ixgbevf_resume);
4842 static struct pci_driver ixgbevf_driver = {
4843 .name = ixgbevf_driver_name,
4844 .id_table = ixgbevf_pci_tbl,
4845 .probe = ixgbevf_probe,
4846 .remove = ixgbevf_remove,
4848 /* Power Management Hooks */
4849 .driver.pm = &ixgbevf_pm_ops,
4851 .shutdown = ixgbevf_shutdown,
4852 .err_handler = &ixgbevf_err_handler
4856 * ixgbevf_init_module - Driver Registration Routine
4858 * ixgbevf_init_module is the first routine called when the driver is
4859 * loaded. All it does is register with the PCI subsystem.
4861 static int __init ixgbevf_init_module(void)
4863 pr_info("%s\n", ixgbevf_driver_string);
4864 pr_info("%s\n", ixgbevf_copyright);
4865 ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4867 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4871 return pci_register_driver(&ixgbevf_driver);
4874 module_init(ixgbevf_init_module);
4877 * ixgbevf_exit_module - Driver Exit Cleanup Routine
4879 * ixgbevf_exit_module is called just before the driver is removed
4882 static void __exit ixgbevf_exit_module(void)
4884 pci_unregister_driver(&ixgbevf_driver);
4886 destroy_workqueue(ixgbevf_wq);
4893 * ixgbevf_get_hw_dev_name - return device name string
4894 * used by hardware layer to print debugging information
4895 * @hw: pointer to private hardware struct
4897 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4899 struct ixgbevf_adapter *adapter = hw->back;
4901 return adapter->netdev->name;
4905 module_exit(ixgbevf_exit_module);
4907 /* ixgbevf_main.c */