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 #define DRV_VERSION "4.1.0-k"
42 const char ixgbevf_driver_version[] = DRV_VERSION;
43 static char ixgbevf_copyright[] =
44 "Copyright (c) 2009 - 2018 Intel Corporation.";
46 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
47 [board_82599_vf] = &ixgbevf_82599_vf_info,
48 [board_82599_vf_hv] = &ixgbevf_82599_vf_hv_info,
49 [board_X540_vf] = &ixgbevf_X540_vf_info,
50 [board_X540_vf_hv] = &ixgbevf_X540_vf_hv_info,
51 [board_X550_vf] = &ixgbevf_X550_vf_info,
52 [board_X550_vf_hv] = &ixgbevf_X550_vf_hv_info,
53 [board_X550EM_x_vf] = &ixgbevf_X550EM_x_vf_info,
54 [board_X550EM_x_vf_hv] = &ixgbevf_X550EM_x_vf_hv_info,
55 [board_x550em_a_vf] = &ixgbevf_x550em_a_vf_info,
58 /* ixgbevf_pci_tbl - PCI Device ID Table
60 * Wildcard entries (PCI_ANY_ID) should come last
61 * Last entry must be all 0s
63 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
64 * Class, Class Mask, private data (not used) }
66 static const struct pci_device_id ixgbevf_pci_tbl[] = {
67 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
68 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
69 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
70 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
71 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
72 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
73 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
74 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
75 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
76 /* required last entry */
79 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
81 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
82 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
83 MODULE_LICENSE("GPL v2");
84 MODULE_VERSION(DRV_VERSION);
86 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
87 static int debug = -1;
88 module_param(debug, int, 0);
89 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
91 static struct workqueue_struct *ixgbevf_wq;
93 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
95 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
96 !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
97 !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
98 queue_work(ixgbevf_wq, &adapter->service_task);
101 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
103 BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
105 /* flush memory to make sure state is correct before next watchdog */
106 smp_mb__before_atomic();
107 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
111 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
112 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
113 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
114 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer);
115 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
116 struct ixgbevf_rx_buffer *old_buff);
118 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
120 struct ixgbevf_adapter *adapter = hw->back;
125 dev_err(&adapter->pdev->dev, "Adapter removed\n");
126 if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
127 ixgbevf_service_event_schedule(adapter);
130 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
134 /* The following check not only optimizes a bit by not
135 * performing a read on the status register when the
136 * register just read was a status register read that
137 * returned IXGBE_FAILED_READ_REG. It also blocks any
138 * potential recursion.
140 if (reg == IXGBE_VFSTATUS) {
141 ixgbevf_remove_adapter(hw);
144 value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
145 if (value == IXGBE_FAILED_READ_REG)
146 ixgbevf_remove_adapter(hw);
149 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
151 u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
154 if (IXGBE_REMOVED(reg_addr))
155 return IXGBE_FAILED_READ_REG;
156 value = readl(reg_addr + reg);
157 if (unlikely(value == IXGBE_FAILED_READ_REG))
158 ixgbevf_check_remove(hw, reg);
163 * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
164 * @adapter: pointer to adapter struct
165 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
166 * @queue: queue to map the corresponding interrupt to
167 * @msix_vector: the vector to map to the corresponding queue
169 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
170 u8 queue, u8 msix_vector)
173 struct ixgbe_hw *hw = &adapter->hw;
175 if (direction == -1) {
177 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
178 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
181 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
183 /* Tx or Rx causes */
184 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
185 index = ((16 * (queue & 1)) + (8 * direction));
186 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
187 ivar &= ~(0xFF << index);
188 ivar |= (msix_vector << index);
189 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
193 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
195 return ring->stats.packets;
198 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
200 struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
201 struct ixgbe_hw *hw = &adapter->hw;
203 u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
204 u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
207 return (head < tail) ?
208 tail - head : (tail + ring->count - head);
213 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
215 u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
216 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
217 u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
219 clear_check_for_tx_hang(tx_ring);
221 /* Check for a hung queue, but be thorough. This verifies
222 * that a transmit has been completed since the previous
223 * check AND there is at least one packet pending. The
224 * ARMED bit is set to indicate a potential hang.
226 if ((tx_done_old == tx_done) && tx_pending) {
227 /* make sure it is true for two checks in a row */
228 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
231 /* reset the countdown */
232 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
234 /* update completed stats and continue */
235 tx_ring->tx_stats.tx_done_old = tx_done;
240 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
242 /* Do the reset outside of interrupt context */
243 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
244 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
245 ixgbevf_service_event_schedule(adapter);
250 * ixgbevf_tx_timeout - Respond to a Tx Hang
251 * @netdev: network interface device structure
253 static void ixgbevf_tx_timeout(struct net_device *netdev)
255 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
257 ixgbevf_tx_timeout_reset(adapter);
261 * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
262 * @q_vector: board private structure
263 * @tx_ring: tx ring to clean
264 * @napi_budget: Used to determine if we are in netpoll
266 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
267 struct ixgbevf_ring *tx_ring, int napi_budget)
269 struct ixgbevf_adapter *adapter = q_vector->adapter;
270 struct ixgbevf_tx_buffer *tx_buffer;
271 union ixgbe_adv_tx_desc *tx_desc;
272 unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
273 unsigned int budget = tx_ring->count / 2;
274 unsigned int i = tx_ring->next_to_clean;
276 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
279 tx_buffer = &tx_ring->tx_buffer_info[i];
280 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
284 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
286 /* if next_to_watch is not set then there is no work pending */
290 /* prevent any other reads prior to eop_desc */
293 /* if DD is not set pending work has not been completed */
294 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
297 /* clear next_to_watch to prevent false hangs */
298 tx_buffer->next_to_watch = NULL;
300 /* update the statistics for this packet */
301 total_bytes += tx_buffer->bytecount;
302 total_packets += tx_buffer->gso_segs;
303 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
307 if (ring_is_xdp(tx_ring))
308 page_frag_free(tx_buffer->data);
310 napi_consume_skb(tx_buffer->skb, napi_budget);
312 /* unmap skb header data */
313 dma_unmap_single(tx_ring->dev,
314 dma_unmap_addr(tx_buffer, dma),
315 dma_unmap_len(tx_buffer, len),
318 /* clear tx_buffer data */
319 dma_unmap_len_set(tx_buffer, len, 0);
321 /* unmap remaining buffers */
322 while (tx_desc != eop_desc) {
328 tx_buffer = tx_ring->tx_buffer_info;
329 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
332 /* unmap any remaining paged data */
333 if (dma_unmap_len(tx_buffer, len)) {
334 dma_unmap_page(tx_ring->dev,
335 dma_unmap_addr(tx_buffer, dma),
336 dma_unmap_len(tx_buffer, len),
338 dma_unmap_len_set(tx_buffer, len, 0);
342 /* move us one more past the eop_desc for start of next pkt */
348 tx_buffer = tx_ring->tx_buffer_info;
349 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
352 /* issue prefetch for next Tx descriptor */
355 /* update budget accounting */
357 } while (likely(budget));
360 tx_ring->next_to_clean = i;
361 u64_stats_update_begin(&tx_ring->syncp);
362 tx_ring->stats.bytes += total_bytes;
363 tx_ring->stats.packets += total_packets;
364 u64_stats_update_end(&tx_ring->syncp);
365 q_vector->tx.total_bytes += total_bytes;
366 q_vector->tx.total_packets += total_packets;
367 adapter->tx_ipsec += total_ipsec;
369 if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
370 struct ixgbe_hw *hw = &adapter->hw;
371 union ixgbe_adv_tx_desc *eop_desc;
373 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
375 pr_err("Detected Tx Unit Hang%s\n"
377 " TDH, TDT <%x>, <%x>\n"
378 " next_to_use <%x>\n"
379 " next_to_clean <%x>\n"
380 "tx_buffer_info[next_to_clean]\n"
381 " next_to_watch <%p>\n"
382 " eop_desc->wb.status <%x>\n"
383 " time_stamp <%lx>\n"
385 ring_is_xdp(tx_ring) ? " XDP" : "",
386 tx_ring->queue_index,
387 IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
388 IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
389 tx_ring->next_to_use, i,
390 eop_desc, (eop_desc ? eop_desc->wb.status : 0),
391 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
393 if (!ring_is_xdp(tx_ring))
394 netif_stop_subqueue(tx_ring->netdev,
395 tx_ring->queue_index);
397 /* schedule immediate reset if we believe we hung */
398 ixgbevf_tx_timeout_reset(adapter);
403 if (ring_is_xdp(tx_ring))
406 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
407 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
408 (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
409 /* Make sure that anybody stopping the queue after this
410 * sees the new next_to_clean.
414 if (__netif_subqueue_stopped(tx_ring->netdev,
415 tx_ring->queue_index) &&
416 !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
417 netif_wake_subqueue(tx_ring->netdev,
418 tx_ring->queue_index);
419 ++tx_ring->tx_stats.restart_queue;
427 * ixgbevf_rx_skb - Helper function to determine proper Rx method
428 * @q_vector: structure containing interrupt and ring information
429 * @skb: packet to send up
431 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
434 napi_gro_receive(&q_vector->napi, skb);
437 #define IXGBE_RSS_L4_TYPES_MASK \
438 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
439 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
440 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
441 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
443 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
444 union ixgbe_adv_rx_desc *rx_desc,
449 if (!(ring->netdev->features & NETIF_F_RXHASH))
452 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
453 IXGBE_RXDADV_RSSTYPE_MASK;
458 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
459 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
460 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
464 * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
465 * @ring: structure containig ring specific data
466 * @rx_desc: current Rx descriptor being processed
467 * @skb: skb currently being received and modified
469 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
470 union ixgbe_adv_rx_desc *rx_desc,
473 skb_checksum_none_assert(skb);
475 /* Rx csum disabled */
476 if (!(ring->netdev->features & NETIF_F_RXCSUM))
479 /* if IP and error */
480 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
481 ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
482 ring->rx_stats.csum_err++;
486 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
489 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
490 ring->rx_stats.csum_err++;
494 /* It must be a TCP or UDP packet with a valid checksum */
495 skb->ip_summed = CHECKSUM_UNNECESSARY;
499 * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
500 * @rx_ring: rx descriptor ring packet is being transacted on
501 * @rx_desc: pointer to the EOP Rx descriptor
502 * @skb: pointer to current skb being populated
504 * This function checks the ring, descriptor, and packet information in
505 * order to populate the checksum, VLAN, protocol, and other fields within
508 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
509 union ixgbe_adv_rx_desc *rx_desc,
512 ixgbevf_rx_hash(rx_ring, rx_desc, skb);
513 ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
515 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
516 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
517 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
519 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
520 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
523 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
524 ixgbevf_ipsec_rx(rx_ring, rx_desc, skb);
526 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
530 struct ixgbevf_rx_buffer *ixgbevf_get_rx_buffer(struct ixgbevf_ring *rx_ring,
531 const unsigned int size)
533 struct ixgbevf_rx_buffer *rx_buffer;
535 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
536 prefetchw(rx_buffer->page);
538 /* we are reusing so sync this buffer for CPU use */
539 dma_sync_single_range_for_cpu(rx_ring->dev,
541 rx_buffer->page_offset,
545 rx_buffer->pagecnt_bias--;
550 static void ixgbevf_put_rx_buffer(struct ixgbevf_ring *rx_ring,
551 struct ixgbevf_rx_buffer *rx_buffer,
554 if (ixgbevf_can_reuse_rx_page(rx_buffer)) {
555 /* hand second half of page back to the ring */
556 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
559 /* We are not reusing the buffer so unmap it and free
560 * any references we are holding to it
562 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
563 ixgbevf_rx_pg_size(rx_ring),
565 IXGBEVF_RX_DMA_ATTR);
566 __page_frag_cache_drain(rx_buffer->page,
567 rx_buffer->pagecnt_bias);
570 /* clear contents of rx_buffer */
571 rx_buffer->page = NULL;
575 * ixgbevf_is_non_eop - process handling of non-EOP buffers
576 * @rx_ring: Rx ring being processed
577 * @rx_desc: Rx descriptor for current buffer
579 * This function updates next to clean. If the buffer is an EOP buffer
580 * this function exits returning false, otherwise it will place the
581 * sk_buff in the next buffer to be chained and return true indicating
582 * that this is in fact a non-EOP buffer.
584 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
585 union ixgbe_adv_rx_desc *rx_desc)
587 u32 ntc = rx_ring->next_to_clean + 1;
589 /* fetch, update, and store next to clean */
590 ntc = (ntc < rx_ring->count) ? ntc : 0;
591 rx_ring->next_to_clean = ntc;
593 prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
595 if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
601 static inline unsigned int ixgbevf_rx_offset(struct ixgbevf_ring *rx_ring)
603 return ring_uses_build_skb(rx_ring) ? IXGBEVF_SKB_PAD : 0;
606 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
607 struct ixgbevf_rx_buffer *bi)
609 struct page *page = bi->page;
612 /* since we are recycling buffers we should seldom need to alloc */
616 /* alloc new page for storage */
617 page = dev_alloc_pages(ixgbevf_rx_pg_order(rx_ring));
618 if (unlikely(!page)) {
619 rx_ring->rx_stats.alloc_rx_page_failed++;
623 /* map page for use */
624 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
625 ixgbevf_rx_pg_size(rx_ring),
626 DMA_FROM_DEVICE, IXGBEVF_RX_DMA_ATTR);
628 /* if mapping failed free memory back to system since
629 * there isn't much point in holding memory we can't use
631 if (dma_mapping_error(rx_ring->dev, dma)) {
632 __free_pages(page, ixgbevf_rx_pg_order(rx_ring));
634 rx_ring->rx_stats.alloc_rx_page_failed++;
640 bi->page_offset = ixgbevf_rx_offset(rx_ring);
641 bi->pagecnt_bias = 1;
642 rx_ring->rx_stats.alloc_rx_page++;
648 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
649 * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
650 * @cleaned_count: number of buffers to replace
652 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
655 union ixgbe_adv_rx_desc *rx_desc;
656 struct ixgbevf_rx_buffer *bi;
657 unsigned int i = rx_ring->next_to_use;
659 /* nothing to do or no valid netdev defined */
660 if (!cleaned_count || !rx_ring->netdev)
663 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
664 bi = &rx_ring->rx_buffer_info[i];
668 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
671 /* sync the buffer for use by the device */
672 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
674 ixgbevf_rx_bufsz(rx_ring),
677 /* Refresh the desc even if pkt_addr didn't change
678 * because each write-back erases this info.
680 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
686 rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
687 bi = rx_ring->rx_buffer_info;
691 /* clear the length for the next_to_use descriptor */
692 rx_desc->wb.upper.length = 0;
695 } while (cleaned_count);
699 if (rx_ring->next_to_use != i) {
700 /* record the next descriptor to use */
701 rx_ring->next_to_use = i;
703 /* update next to alloc since we have filled the ring */
704 rx_ring->next_to_alloc = i;
706 /* Force memory writes to complete before letting h/w
707 * know there are new descriptors to fetch. (Only
708 * applicable for weak-ordered memory model archs,
712 ixgbevf_write_tail(rx_ring, i);
717 * ixgbevf_cleanup_headers - Correct corrupted or empty headers
718 * @rx_ring: rx descriptor ring packet is being transacted on
719 * @rx_desc: pointer to the EOP Rx descriptor
720 * @skb: pointer to current skb being fixed
722 * Check for corrupted packet headers caused by senders on the local L2
723 * embedded NIC switch not setting up their Tx Descriptors right. These
724 * should be very rare.
726 * Also address the case where we are pulling data in on pages only
727 * and as such no data is present in the skb header.
729 * In addition if skb is not at least 60 bytes we need to pad it so that
730 * it is large enough to qualify as a valid Ethernet frame.
732 * Returns true if an error was encountered and skb was freed.
734 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
735 union ixgbe_adv_rx_desc *rx_desc,
738 /* XDP packets use error pointer so abort at this point */
742 /* verify that the packet does not have any known errors */
743 if (unlikely(ixgbevf_test_staterr(rx_desc,
744 IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
745 struct net_device *netdev = rx_ring->netdev;
747 if (!(netdev->features & NETIF_F_RXALL)) {
748 dev_kfree_skb_any(skb);
753 /* if eth_skb_pad returns an error the skb was freed */
754 if (eth_skb_pad(skb))
761 * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
762 * @rx_ring: rx descriptor ring to store buffers on
763 * @old_buff: donor buffer to have page reused
765 * Synchronizes page for reuse by the adapter
767 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
768 struct ixgbevf_rx_buffer *old_buff)
770 struct ixgbevf_rx_buffer *new_buff;
771 u16 nta = rx_ring->next_to_alloc;
773 new_buff = &rx_ring->rx_buffer_info[nta];
775 /* update, and store next to alloc */
777 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
779 /* transfer page from old buffer to new buffer */
780 new_buff->page = old_buff->page;
781 new_buff->dma = old_buff->dma;
782 new_buff->page_offset = old_buff->page_offset;
783 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
786 static inline bool ixgbevf_page_is_reserved(struct page *page)
788 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
791 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer)
793 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
794 struct page *page = rx_buffer->page;
796 /* avoid re-using remote pages */
797 if (unlikely(ixgbevf_page_is_reserved(page)))
800 #if (PAGE_SIZE < 8192)
801 /* if we are only owner of page we can reuse it */
802 if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
805 #define IXGBEVF_LAST_OFFSET \
806 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBEVF_RXBUFFER_2048)
808 if (rx_buffer->page_offset > IXGBEVF_LAST_OFFSET)
813 /* If we have drained the page fragment pool we need to update
814 * the pagecnt_bias and page count so that we fully restock the
815 * number of references the driver holds.
817 if (unlikely(!pagecnt_bias)) {
818 page_ref_add(page, USHRT_MAX);
819 rx_buffer->pagecnt_bias = USHRT_MAX;
826 * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
827 * @rx_ring: rx descriptor ring to transact packets on
828 * @rx_buffer: buffer containing page to add
829 * @skb: sk_buff to place the data into
830 * @size: size of buffer to be added
832 * This function will add the data contained in rx_buffer->page to the skb.
834 static void ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
835 struct ixgbevf_rx_buffer *rx_buffer,
839 #if (PAGE_SIZE < 8192)
840 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
842 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
843 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
844 SKB_DATA_ALIGN(size);
846 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
847 rx_buffer->page_offset, size, truesize);
848 #if (PAGE_SIZE < 8192)
849 rx_buffer->page_offset ^= truesize;
851 rx_buffer->page_offset += truesize;
856 struct sk_buff *ixgbevf_construct_skb(struct ixgbevf_ring *rx_ring,
857 struct ixgbevf_rx_buffer *rx_buffer,
858 struct xdp_buff *xdp,
859 union ixgbe_adv_rx_desc *rx_desc)
861 unsigned int size = xdp->data_end - xdp->data;
862 #if (PAGE_SIZE < 8192)
863 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
865 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
866 xdp->data_hard_start);
868 unsigned int headlen;
871 /* prefetch first cache line of first page */
873 #if L1_CACHE_BYTES < 128
874 prefetch(xdp->data + L1_CACHE_BYTES);
876 /* Note, we get here by enabling legacy-rx via:
878 * ethtool --set-priv-flags <dev> legacy-rx on
880 * In this mode, we currently get 0 extra XDP headroom as
881 * opposed to having legacy-rx off, where we process XDP
882 * packets going to stack via ixgbevf_build_skb().
884 * For ixgbevf_construct_skb() mode it means that the
885 * xdp->data_meta will always point to xdp->data, since
886 * the helper cannot expand the head. Should this ever
887 * changed in future for legacy-rx mode on, then lets also
888 * add xdp->data_meta handling here.
891 /* allocate a skb to store the frags */
892 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBEVF_RX_HDR_SIZE);
896 /* Determine available headroom for copy */
898 if (headlen > IXGBEVF_RX_HDR_SIZE)
899 headlen = eth_get_headlen(skb->dev, xdp->data,
900 IXGBEVF_RX_HDR_SIZE);
902 /* align pull length to size of long to optimize memcpy performance */
903 memcpy(__skb_put(skb, headlen), xdp->data,
904 ALIGN(headlen, sizeof(long)));
906 /* update all of the pointers */
909 skb_add_rx_frag(skb, 0, rx_buffer->page,
910 (xdp->data + headlen) -
911 page_address(rx_buffer->page),
913 #if (PAGE_SIZE < 8192)
914 rx_buffer->page_offset ^= truesize;
916 rx_buffer->page_offset += truesize;
919 rx_buffer->pagecnt_bias++;
925 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
928 struct ixgbe_hw *hw = &adapter->hw;
930 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
933 static struct sk_buff *ixgbevf_build_skb(struct ixgbevf_ring *rx_ring,
934 struct ixgbevf_rx_buffer *rx_buffer,
935 struct xdp_buff *xdp,
936 union ixgbe_adv_rx_desc *rx_desc)
938 unsigned int metasize = xdp->data - xdp->data_meta;
939 #if (PAGE_SIZE < 8192)
940 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
942 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
943 SKB_DATA_ALIGN(xdp->data_end -
944 xdp->data_hard_start);
948 /* Prefetch first cache line of first page. If xdp->data_meta
949 * is unused, this points to xdp->data, otherwise, we likely
950 * have a consumer accessing first few bytes of meta data,
951 * and then actual data.
953 prefetch(xdp->data_meta);
954 #if L1_CACHE_BYTES < 128
955 prefetch(xdp->data_meta + L1_CACHE_BYTES);
958 /* build an skb around the page buffer */
959 skb = build_skb(xdp->data_hard_start, truesize);
963 /* update pointers within the skb to store the data */
964 skb_reserve(skb, xdp->data - xdp->data_hard_start);
965 __skb_put(skb, xdp->data_end - xdp->data);
967 skb_metadata_set(skb, metasize);
969 /* update buffer offset */
970 #if (PAGE_SIZE < 8192)
971 rx_buffer->page_offset ^= truesize;
973 rx_buffer->page_offset += truesize;
979 #define IXGBEVF_XDP_PASS 0
980 #define IXGBEVF_XDP_CONSUMED 1
981 #define IXGBEVF_XDP_TX 2
983 static int ixgbevf_xmit_xdp_ring(struct ixgbevf_ring *ring,
984 struct xdp_buff *xdp)
986 struct ixgbevf_tx_buffer *tx_buffer;
987 union ixgbe_adv_tx_desc *tx_desc;
992 len = xdp->data_end - xdp->data;
994 if (unlikely(!ixgbevf_desc_unused(ring)))
995 return IXGBEVF_XDP_CONSUMED;
997 dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE);
998 if (dma_mapping_error(ring->dev, dma))
999 return IXGBEVF_XDP_CONSUMED;
1001 /* record the location of the first descriptor for this packet */
1002 i = ring->next_to_use;
1003 tx_buffer = &ring->tx_buffer_info[i];
1005 dma_unmap_len_set(tx_buffer, len, len);
1006 dma_unmap_addr_set(tx_buffer, dma, dma);
1007 tx_buffer->data = xdp->data;
1008 tx_buffer->bytecount = len;
1009 tx_buffer->gso_segs = 1;
1010 tx_buffer->protocol = 0;
1012 /* Populate minimal context descriptor that will provide for the
1013 * fact that we are expected to process Ethernet frames.
1015 if (!test_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state)) {
1016 struct ixgbe_adv_tx_context_desc *context_desc;
1018 set_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1020 context_desc = IXGBEVF_TX_CTXTDESC(ring, 0);
1021 context_desc->vlan_macip_lens =
1022 cpu_to_le32(ETH_HLEN << IXGBE_ADVTXD_MACLEN_SHIFT);
1023 context_desc->fceof_saidx = 0;
1024 context_desc->type_tucmd_mlhl =
1025 cpu_to_le32(IXGBE_TXD_CMD_DEXT |
1026 IXGBE_ADVTXD_DTYP_CTXT);
1027 context_desc->mss_l4len_idx = 0;
1032 /* put descriptor type bits */
1033 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
1034 IXGBE_ADVTXD_DCMD_DEXT |
1035 IXGBE_ADVTXD_DCMD_IFCS;
1036 cmd_type |= len | IXGBE_TXD_CMD;
1038 tx_desc = IXGBEVF_TX_DESC(ring, i);
1039 tx_desc->read.buffer_addr = cpu_to_le64(dma);
1041 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1042 tx_desc->read.olinfo_status =
1043 cpu_to_le32((len << IXGBE_ADVTXD_PAYLEN_SHIFT) |
1046 /* Avoid any potential race with cleanup */
1049 /* set next_to_watch value indicating a packet is present */
1051 if (i == ring->count)
1054 tx_buffer->next_to_watch = tx_desc;
1055 ring->next_to_use = i;
1057 return IXGBEVF_XDP_TX;
1060 static struct sk_buff *ixgbevf_run_xdp(struct ixgbevf_adapter *adapter,
1061 struct ixgbevf_ring *rx_ring,
1062 struct xdp_buff *xdp)
1064 int result = IXGBEVF_XDP_PASS;
1065 struct ixgbevf_ring *xdp_ring;
1066 struct bpf_prog *xdp_prog;
1070 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
1075 act = bpf_prog_run_xdp(xdp_prog, xdp);
1080 xdp_ring = adapter->xdp_ring[rx_ring->queue_index];
1081 result = ixgbevf_xmit_xdp_ring(xdp_ring, xdp);
1084 bpf_warn_invalid_xdp_action(act);
1087 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
1088 /* fallthrough -- handle aborts by dropping packet */
1090 result = IXGBEVF_XDP_CONSUMED;
1095 return ERR_PTR(-result);
1098 static void ixgbevf_rx_buffer_flip(struct ixgbevf_ring *rx_ring,
1099 struct ixgbevf_rx_buffer *rx_buffer,
1102 #if (PAGE_SIZE < 8192)
1103 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
1105 rx_buffer->page_offset ^= truesize;
1107 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
1108 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
1109 SKB_DATA_ALIGN(size);
1111 rx_buffer->page_offset += truesize;
1115 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
1116 struct ixgbevf_ring *rx_ring,
1119 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1120 struct ixgbevf_adapter *adapter = q_vector->adapter;
1121 u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
1122 struct sk_buff *skb = rx_ring->skb;
1123 bool xdp_xmit = false;
1124 struct xdp_buff xdp;
1126 xdp.rxq = &rx_ring->xdp_rxq;
1128 while (likely(total_rx_packets < budget)) {
1129 struct ixgbevf_rx_buffer *rx_buffer;
1130 union ixgbe_adv_rx_desc *rx_desc;
1133 /* return some buffers to hardware, one at a time is too slow */
1134 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
1135 ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
1139 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
1140 size = le16_to_cpu(rx_desc->wb.upper.length);
1144 /* This memory barrier is needed to keep us from reading
1145 * any other fields out of the rx_desc until we know the
1146 * RXD_STAT_DD bit is set
1150 rx_buffer = ixgbevf_get_rx_buffer(rx_ring, size);
1152 /* retrieve a buffer from the ring */
1154 xdp.data = page_address(rx_buffer->page) +
1155 rx_buffer->page_offset;
1156 xdp.data_meta = xdp.data;
1157 xdp.data_hard_start = xdp.data -
1158 ixgbevf_rx_offset(rx_ring);
1159 xdp.data_end = xdp.data + size;
1161 skb = ixgbevf_run_xdp(adapter, rx_ring, &xdp);
1165 if (PTR_ERR(skb) == -IXGBEVF_XDP_TX) {
1167 ixgbevf_rx_buffer_flip(rx_ring, rx_buffer,
1170 rx_buffer->pagecnt_bias++;
1173 total_rx_bytes += size;
1175 ixgbevf_add_rx_frag(rx_ring, rx_buffer, skb, size);
1176 } else if (ring_uses_build_skb(rx_ring)) {
1177 skb = ixgbevf_build_skb(rx_ring, rx_buffer,
1180 skb = ixgbevf_construct_skb(rx_ring, rx_buffer,
1184 /* exit if we failed to retrieve a buffer */
1186 rx_ring->rx_stats.alloc_rx_buff_failed++;
1187 rx_buffer->pagecnt_bias++;
1191 ixgbevf_put_rx_buffer(rx_ring, rx_buffer, skb);
1194 /* fetch next buffer in frame if non-eop */
1195 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
1198 /* verify the packet layout is correct */
1199 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
1204 /* probably a little skewed due to removing CRC */
1205 total_rx_bytes += skb->len;
1207 /* Workaround hardware that can't do proper VEPA multicast
1210 if ((skb->pkt_type == PACKET_BROADCAST ||
1211 skb->pkt_type == PACKET_MULTICAST) &&
1212 ether_addr_equal(rx_ring->netdev->dev_addr,
1213 eth_hdr(skb)->h_source)) {
1214 dev_kfree_skb_irq(skb);
1218 /* populate checksum, VLAN, and protocol */
1219 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
1221 ixgbevf_rx_skb(q_vector, skb);
1223 /* reset skb pointer */
1226 /* update budget accounting */
1230 /* place incomplete frames back on ring for completion */
1234 struct ixgbevf_ring *xdp_ring =
1235 adapter->xdp_ring[rx_ring->queue_index];
1237 /* Force memory writes to complete before letting h/w
1238 * know there are new descriptors to fetch.
1241 ixgbevf_write_tail(xdp_ring, xdp_ring->next_to_use);
1244 u64_stats_update_begin(&rx_ring->syncp);
1245 rx_ring->stats.packets += total_rx_packets;
1246 rx_ring->stats.bytes += total_rx_bytes;
1247 u64_stats_update_end(&rx_ring->syncp);
1248 q_vector->rx.total_packets += total_rx_packets;
1249 q_vector->rx.total_bytes += total_rx_bytes;
1251 return total_rx_packets;
1255 * ixgbevf_poll - NAPI polling calback
1256 * @napi: napi struct with our devices info in it
1257 * @budget: amount of work driver is allowed to do this pass, in packets
1259 * This function will clean more than one or more rings associated with a
1262 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1264 struct ixgbevf_q_vector *q_vector =
1265 container_of(napi, struct ixgbevf_q_vector, napi);
1266 struct ixgbevf_adapter *adapter = q_vector->adapter;
1267 struct ixgbevf_ring *ring;
1268 int per_ring_budget, work_done = 0;
1269 bool clean_complete = true;
1271 ixgbevf_for_each_ring(ring, q_vector->tx) {
1272 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1273 clean_complete = false;
1279 /* attempt to distribute budget to each queue fairly, but don't allow
1280 * the budget to go below 1 because we'll exit polling
1282 if (q_vector->rx.count > 1)
1283 per_ring_budget = max(budget/q_vector->rx.count, 1);
1285 per_ring_budget = budget;
1287 ixgbevf_for_each_ring(ring, q_vector->rx) {
1288 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1290 work_done += cleaned;
1291 if (cleaned >= per_ring_budget)
1292 clean_complete = false;
1295 /* If all work not completed, return budget and keep polling */
1296 if (!clean_complete)
1299 /* Exit the polling mode, but don't re-enable interrupts if stack might
1300 * poll us due to busy-polling
1302 if (likely(napi_complete_done(napi, work_done))) {
1303 if (adapter->rx_itr_setting == 1)
1304 ixgbevf_set_itr(q_vector);
1305 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1306 !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1307 ixgbevf_irq_enable_queues(adapter,
1308 BIT(q_vector->v_idx));
1311 return min(work_done, budget - 1);
1315 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1316 * @q_vector: structure containing interrupt and ring information
1318 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1320 struct ixgbevf_adapter *adapter = q_vector->adapter;
1321 struct ixgbe_hw *hw = &adapter->hw;
1322 int v_idx = q_vector->v_idx;
1323 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1325 /* set the WDIS bit to not clear the timer bits and cause an
1326 * immediate assertion of the interrupt
1328 itr_reg |= IXGBE_EITR_CNT_WDIS;
1330 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1334 * ixgbevf_configure_msix - Configure MSI-X hardware
1335 * @adapter: board private structure
1337 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1340 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1342 struct ixgbevf_q_vector *q_vector;
1343 int q_vectors, v_idx;
1345 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1346 adapter->eims_enable_mask = 0;
1348 /* Populate the IVAR table and set the ITR values to the
1349 * corresponding register.
1351 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1352 struct ixgbevf_ring *ring;
1354 q_vector = adapter->q_vector[v_idx];
1356 ixgbevf_for_each_ring(ring, q_vector->rx)
1357 ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1359 ixgbevf_for_each_ring(ring, q_vector->tx)
1360 ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1362 if (q_vector->tx.ring && !q_vector->rx.ring) {
1363 /* Tx only vector */
1364 if (adapter->tx_itr_setting == 1)
1365 q_vector->itr = IXGBE_12K_ITR;
1367 q_vector->itr = adapter->tx_itr_setting;
1369 /* Rx or Rx/Tx vector */
1370 if (adapter->rx_itr_setting == 1)
1371 q_vector->itr = IXGBE_20K_ITR;
1373 q_vector->itr = adapter->rx_itr_setting;
1376 /* add q_vector eims value to global eims_enable_mask */
1377 adapter->eims_enable_mask |= BIT(v_idx);
1379 ixgbevf_write_eitr(q_vector);
1382 ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1383 /* setup eims_other and add value to global eims_enable_mask */
1384 adapter->eims_other = BIT(v_idx);
1385 adapter->eims_enable_mask |= adapter->eims_other;
1388 enum latency_range {
1392 latency_invalid = 255
1396 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1397 * @q_vector: structure containing interrupt and ring information
1398 * @ring_container: structure containing ring performance data
1400 * Stores a new ITR value based on packets and byte
1401 * counts during the last interrupt. The advantage of per interrupt
1402 * computation is faster updates and more accurate ITR for the current
1403 * traffic pattern. Constants in this function were computed
1404 * based on theoretical maximum wire speed and thresholds were set based
1405 * on testing data as well as attempting to minimize response time
1406 * while increasing bulk throughput.
1408 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1409 struct ixgbevf_ring_container *ring_container)
1411 int bytes = ring_container->total_bytes;
1412 int packets = ring_container->total_packets;
1415 u8 itr_setting = ring_container->itr;
1420 /* simple throttle rate management
1421 * 0-20MB/s lowest (100000 ints/s)
1422 * 20-100MB/s low (20000 ints/s)
1423 * 100-1249MB/s bulk (12000 ints/s)
1425 /* what was last interrupt timeslice? */
1426 timepassed_us = q_vector->itr >> 2;
1427 if (timepassed_us == 0)
1430 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1432 switch (itr_setting) {
1433 case lowest_latency:
1434 if (bytes_perint > 10)
1435 itr_setting = low_latency;
1438 if (bytes_perint > 20)
1439 itr_setting = bulk_latency;
1440 else if (bytes_perint <= 10)
1441 itr_setting = lowest_latency;
1444 if (bytes_perint <= 20)
1445 itr_setting = low_latency;
1449 /* clear work counters since we have the values we need */
1450 ring_container->total_bytes = 0;
1451 ring_container->total_packets = 0;
1453 /* write updated itr to ring container */
1454 ring_container->itr = itr_setting;
1457 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1459 u32 new_itr = q_vector->itr;
1462 ixgbevf_update_itr(q_vector, &q_vector->tx);
1463 ixgbevf_update_itr(q_vector, &q_vector->rx);
1465 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1467 switch (current_itr) {
1468 /* counts and packets in update_itr are dependent on these numbers */
1469 case lowest_latency:
1470 new_itr = IXGBE_100K_ITR;
1473 new_itr = IXGBE_20K_ITR;
1476 new_itr = IXGBE_12K_ITR;
1482 if (new_itr != q_vector->itr) {
1483 /* do an exponential smoothing */
1484 new_itr = (10 * new_itr * q_vector->itr) /
1485 ((9 * new_itr) + q_vector->itr);
1487 /* save the algorithm value here */
1488 q_vector->itr = new_itr;
1490 ixgbevf_write_eitr(q_vector);
1494 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1496 struct ixgbevf_adapter *adapter = data;
1497 struct ixgbe_hw *hw = &adapter->hw;
1499 hw->mac.get_link_status = 1;
1501 ixgbevf_service_event_schedule(adapter);
1503 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1509 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1511 * @data: pointer to our q_vector struct for this interrupt vector
1513 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1515 struct ixgbevf_q_vector *q_vector = data;
1517 /* EIAM disabled interrupts (on this vector) for us */
1518 if (q_vector->rx.ring || q_vector->tx.ring)
1519 napi_schedule_irqoff(&q_vector->napi);
1525 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1526 * @adapter: board private structure
1528 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1529 * interrupts from the kernel.
1531 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1533 struct net_device *netdev = adapter->netdev;
1534 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1535 unsigned int ri = 0, ti = 0;
1538 for (vector = 0; vector < q_vectors; vector++) {
1539 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1540 struct msix_entry *entry = &adapter->msix_entries[vector];
1542 if (q_vector->tx.ring && q_vector->rx.ring) {
1543 snprintf(q_vector->name, sizeof(q_vector->name),
1544 "%s-TxRx-%u", netdev->name, ri++);
1546 } else if (q_vector->rx.ring) {
1547 snprintf(q_vector->name, sizeof(q_vector->name),
1548 "%s-rx-%u", netdev->name, ri++);
1549 } else if (q_vector->tx.ring) {
1550 snprintf(q_vector->name, sizeof(q_vector->name),
1551 "%s-tx-%u", netdev->name, ti++);
1553 /* skip this unused q_vector */
1556 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1557 q_vector->name, q_vector);
1559 hw_dbg(&adapter->hw,
1560 "request_irq failed for MSIX interrupt Error: %d\n",
1562 goto free_queue_irqs;
1566 err = request_irq(adapter->msix_entries[vector].vector,
1567 &ixgbevf_msix_other, 0, netdev->name, adapter);
1569 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1571 goto free_queue_irqs;
1579 free_irq(adapter->msix_entries[vector].vector,
1580 adapter->q_vector[vector]);
1582 /* This failure is non-recoverable - it indicates the system is
1583 * out of MSIX vector resources and the VF driver cannot run
1584 * without them. Set the number of msix vectors to zero
1585 * indicating that not enough can be allocated. The error
1586 * will be returned to the user indicating device open failed.
1587 * Any further attempts to force the driver to open will also
1588 * fail. The only way to recover is to unload the driver and
1589 * reload it again. If the system has recovered some MSIX
1590 * vectors then it may succeed.
1592 adapter->num_msix_vectors = 0;
1597 * ixgbevf_request_irq - initialize interrupts
1598 * @adapter: board private structure
1600 * Attempts to configure interrupts using the best available
1601 * capabilities of the hardware and kernel.
1603 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1605 int err = ixgbevf_request_msix_irqs(adapter);
1608 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1613 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1617 if (!adapter->msix_entries)
1620 q_vectors = adapter->num_msix_vectors;
1623 free_irq(adapter->msix_entries[i].vector, adapter);
1626 for (; i >= 0; i--) {
1627 /* free only the irqs that were actually requested */
1628 if (!adapter->q_vector[i]->rx.ring &&
1629 !adapter->q_vector[i]->tx.ring)
1632 free_irq(adapter->msix_entries[i].vector,
1633 adapter->q_vector[i]);
1638 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1639 * @adapter: board private structure
1641 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1643 struct ixgbe_hw *hw = &adapter->hw;
1646 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1647 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1648 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1650 IXGBE_WRITE_FLUSH(hw);
1652 for (i = 0; i < adapter->num_msix_vectors; i++)
1653 synchronize_irq(adapter->msix_entries[i].vector);
1657 * ixgbevf_irq_enable - Enable default interrupt generation settings
1658 * @adapter: board private structure
1660 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1662 struct ixgbe_hw *hw = &adapter->hw;
1664 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1665 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1666 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1670 * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1671 * @adapter: board private structure
1672 * @ring: structure containing ring specific data
1674 * Configure the Tx descriptor ring after a reset.
1676 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1677 struct ixgbevf_ring *ring)
1679 struct ixgbe_hw *hw = &adapter->hw;
1680 u64 tdba = ring->dma;
1682 u32 txdctl = IXGBE_TXDCTL_ENABLE;
1683 u8 reg_idx = ring->reg_idx;
1685 /* disable queue to avoid issues while updating state */
1686 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1687 IXGBE_WRITE_FLUSH(hw);
1689 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1690 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1691 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1692 ring->count * sizeof(union ixgbe_adv_tx_desc));
1694 /* disable head writeback */
1695 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1696 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1698 /* enable relaxed ordering */
1699 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1700 (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1701 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1703 /* reset head and tail pointers */
1704 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1705 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1706 ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1708 /* reset ntu and ntc to place SW in sync with hardwdare */
1709 ring->next_to_clean = 0;
1710 ring->next_to_use = 0;
1712 /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1713 * to or less than the number of on chip descriptors, which is
1716 txdctl |= (8 << 16); /* WTHRESH = 8 */
1718 /* Setting PTHRESH to 32 both improves performance */
1719 txdctl |= (1u << 8) | /* HTHRESH = 1 */
1720 32; /* PTHRESH = 32 */
1722 /* reinitialize tx_buffer_info */
1723 memset(ring->tx_buffer_info, 0,
1724 sizeof(struct ixgbevf_tx_buffer) * ring->count);
1726 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1727 clear_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1729 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1731 /* poll to verify queue is enabled */
1733 usleep_range(1000, 2000);
1734 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1735 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1737 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1741 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1742 * @adapter: board private structure
1744 * Configure the Tx unit of the MAC after a reset.
1746 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1750 /* Setup the HW Tx Head and Tail descriptor pointers */
1751 for (i = 0; i < adapter->num_tx_queues; i++)
1752 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1753 for (i = 0; i < adapter->num_xdp_queues; i++)
1754 ixgbevf_configure_tx_ring(adapter, adapter->xdp_ring[i]);
1757 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1759 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter,
1760 struct ixgbevf_ring *ring, int index)
1762 struct ixgbe_hw *hw = &adapter->hw;
1765 srrctl = IXGBE_SRRCTL_DROP_EN;
1767 srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1768 if (ring_uses_large_buffer(ring))
1769 srrctl |= IXGBEVF_RXBUFFER_3072 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1771 srrctl |= IXGBEVF_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1772 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1774 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1777 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1779 struct ixgbe_hw *hw = &adapter->hw;
1781 /* PSRTYPE must be initialized in 82599 */
1782 u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1783 IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1784 IXGBE_PSRTYPE_L2HDR;
1786 if (adapter->num_rx_queues > 1)
1789 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1792 #define IXGBEVF_MAX_RX_DESC_POLL 10
1793 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1794 struct ixgbevf_ring *ring)
1796 struct ixgbe_hw *hw = &adapter->hw;
1797 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1799 u8 reg_idx = ring->reg_idx;
1801 if (IXGBE_REMOVED(hw->hw_addr))
1803 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1804 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1806 /* write value back with RXDCTL.ENABLE bit cleared */
1807 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1809 /* the hardware may take up to 100us to really disable the Rx queue */
1812 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1813 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1816 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1820 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1821 struct ixgbevf_ring *ring)
1823 struct ixgbe_hw *hw = &adapter->hw;
1824 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1826 u8 reg_idx = ring->reg_idx;
1828 if (IXGBE_REMOVED(hw->hw_addr))
1831 usleep_range(1000, 2000);
1832 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1833 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1836 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1841 * ixgbevf_init_rss_key - Initialize adapter RSS key
1842 * @adapter: device handle
1844 * Allocates and initializes the RSS key if it is not allocated.
1846 static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter)
1850 if (!adapter->rss_key) {
1851 rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL);
1852 if (unlikely(!rss_key))
1855 netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE);
1856 adapter->rss_key = rss_key;
1862 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1864 struct ixgbe_hw *hw = &adapter->hw;
1865 u32 vfmrqc = 0, vfreta = 0;
1866 u16 rss_i = adapter->num_rx_queues;
1869 /* Fill out hash function seeds */
1870 for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1871 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i));
1873 for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1877 adapter->rss_indir_tbl[i] = j;
1879 vfreta |= j << (i & 0x3) * 8;
1881 IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1886 /* Perform hash on these packet types */
1887 vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1888 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1889 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1890 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1892 vfmrqc |= IXGBE_VFMRQC_RSSEN;
1894 IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1897 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1898 struct ixgbevf_ring *ring)
1900 struct ixgbe_hw *hw = &adapter->hw;
1901 union ixgbe_adv_rx_desc *rx_desc;
1902 u64 rdba = ring->dma;
1904 u8 reg_idx = ring->reg_idx;
1906 /* disable queue to avoid issues while updating state */
1907 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1908 ixgbevf_disable_rx_queue(adapter, ring);
1910 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1911 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1912 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1913 ring->count * sizeof(union ixgbe_adv_rx_desc));
1915 #ifndef CONFIG_SPARC
1916 /* enable relaxed ordering */
1917 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1918 IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1920 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1921 IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1922 IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1925 /* reset head and tail pointers */
1926 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1927 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1928 ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1930 /* initialize rx_buffer_info */
1931 memset(ring->rx_buffer_info, 0,
1932 sizeof(struct ixgbevf_rx_buffer) * ring->count);
1934 /* initialize Rx descriptor 0 */
1935 rx_desc = IXGBEVF_RX_DESC(ring, 0);
1936 rx_desc->wb.upper.length = 0;
1938 /* reset ntu and ntc to place SW in sync with hardwdare */
1939 ring->next_to_clean = 0;
1940 ring->next_to_use = 0;
1941 ring->next_to_alloc = 0;
1943 ixgbevf_configure_srrctl(adapter, ring, reg_idx);
1945 /* RXDCTL.RLPML does not work on 82599 */
1946 if (adapter->hw.mac.type != ixgbe_mac_82599_vf) {
1947 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
1948 IXGBE_RXDCTL_RLPML_EN);
1950 #if (PAGE_SIZE < 8192)
1951 /* Limit the maximum frame size so we don't overrun the skb */
1952 if (ring_uses_build_skb(ring) &&
1953 !ring_uses_large_buffer(ring))
1954 rxdctl |= IXGBEVF_MAX_FRAME_BUILD_SKB |
1955 IXGBE_RXDCTL_RLPML_EN;
1959 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1960 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1962 ixgbevf_rx_desc_queue_enable(adapter, ring);
1963 ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1966 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter,
1967 struct ixgbevf_ring *rx_ring)
1969 struct net_device *netdev = adapter->netdev;
1970 unsigned int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1972 /* set build_skb and buffer size flags */
1973 clear_ring_build_skb_enabled(rx_ring);
1974 clear_ring_uses_large_buffer(rx_ring);
1976 if (adapter->flags & IXGBEVF_FLAGS_LEGACY_RX)
1979 set_ring_build_skb_enabled(rx_ring);
1981 if (PAGE_SIZE < 8192) {
1982 if (max_frame <= IXGBEVF_MAX_FRAME_BUILD_SKB)
1985 set_ring_uses_large_buffer(rx_ring);
1990 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1991 * @adapter: board private structure
1993 * Configure the Rx unit of the MAC after a reset.
1995 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1997 struct ixgbe_hw *hw = &adapter->hw;
1998 struct net_device *netdev = adapter->netdev;
2001 ixgbevf_setup_psrtype(adapter);
2002 if (hw->mac.type >= ixgbe_mac_X550_vf)
2003 ixgbevf_setup_vfmrqc(adapter);
2005 spin_lock_bh(&adapter->mbx_lock);
2006 /* notify the PF of our intent to use this size of frame */
2007 ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
2008 spin_unlock_bh(&adapter->mbx_lock);
2010 dev_err(&adapter->pdev->dev,
2011 "Failed to set MTU at %d\n", netdev->mtu);
2013 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2014 * the Base and Length of the Rx Descriptor Ring
2016 for (i = 0; i < adapter->num_rx_queues; i++) {
2017 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
2019 ixgbevf_set_rx_buffer_len(adapter, rx_ring);
2020 ixgbevf_configure_rx_ring(adapter, rx_ring);
2024 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
2025 __be16 proto, u16 vid)
2027 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2028 struct ixgbe_hw *hw = &adapter->hw;
2031 spin_lock_bh(&adapter->mbx_lock);
2033 /* add VID to filter table */
2034 err = hw->mac.ops.set_vfta(hw, vid, 0, true);
2036 spin_unlock_bh(&adapter->mbx_lock);
2038 /* translate error return types so error makes sense */
2039 if (err == IXGBE_ERR_MBX)
2042 if (err == IXGBE_ERR_INVALID_ARGUMENT)
2045 set_bit(vid, adapter->active_vlans);
2050 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
2051 __be16 proto, u16 vid)
2053 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2054 struct ixgbe_hw *hw = &adapter->hw;
2057 spin_lock_bh(&adapter->mbx_lock);
2059 /* remove VID from filter table */
2060 err = hw->mac.ops.set_vfta(hw, vid, 0, false);
2062 spin_unlock_bh(&adapter->mbx_lock);
2064 clear_bit(vid, adapter->active_vlans);
2069 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
2073 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2074 ixgbevf_vlan_rx_add_vid(adapter->netdev,
2075 htons(ETH_P_8021Q), vid);
2078 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
2080 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2081 struct ixgbe_hw *hw = &adapter->hw;
2084 if ((netdev_uc_count(netdev)) > 10) {
2085 pr_err("Too many unicast filters - No Space\n");
2089 if (!netdev_uc_empty(netdev)) {
2090 struct netdev_hw_addr *ha;
2092 netdev_for_each_uc_addr(ha, netdev) {
2093 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
2097 /* If the list is empty then send message to PF driver to
2098 * clear all MAC VLANs on this VF.
2100 hw->mac.ops.set_uc_addr(hw, 0, NULL);
2107 * ixgbevf_set_rx_mode - Multicast and unicast set
2108 * @netdev: network interface device structure
2110 * The set_rx_method entry point is called whenever the multicast address
2111 * list, unicast address list or the network interface flags are updated.
2112 * This routine is responsible for configuring the hardware for proper
2113 * multicast mode and configuring requested unicast filters.
2115 static void ixgbevf_set_rx_mode(struct net_device *netdev)
2117 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2118 struct ixgbe_hw *hw = &adapter->hw;
2119 unsigned int flags = netdev->flags;
2122 /* request the most inclusive mode we need */
2123 if (flags & IFF_PROMISC)
2124 xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
2125 else if (flags & IFF_ALLMULTI)
2126 xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
2127 else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
2128 xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
2130 xcast_mode = IXGBEVF_XCAST_MODE_NONE;
2132 spin_lock_bh(&adapter->mbx_lock);
2134 hw->mac.ops.update_xcast_mode(hw, xcast_mode);
2136 /* reprogram multicast list */
2137 hw->mac.ops.update_mc_addr_list(hw, netdev);
2139 ixgbevf_write_uc_addr_list(netdev);
2141 spin_unlock_bh(&adapter->mbx_lock);
2144 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
2147 struct ixgbevf_q_vector *q_vector;
2148 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2150 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2151 q_vector = adapter->q_vector[q_idx];
2152 napi_enable(&q_vector->napi);
2156 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
2159 struct ixgbevf_q_vector *q_vector;
2160 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2162 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2163 q_vector = adapter->q_vector[q_idx];
2164 napi_disable(&q_vector->napi);
2168 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
2170 struct ixgbe_hw *hw = &adapter->hw;
2171 unsigned int def_q = 0;
2172 unsigned int num_tcs = 0;
2173 unsigned int num_rx_queues = adapter->num_rx_queues;
2174 unsigned int num_tx_queues = adapter->num_tx_queues;
2177 spin_lock_bh(&adapter->mbx_lock);
2179 /* fetch queue configuration from the PF */
2180 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2182 spin_unlock_bh(&adapter->mbx_lock);
2188 /* we need only one Tx queue */
2191 /* update default Tx ring register index */
2192 adapter->tx_ring[0]->reg_idx = def_q;
2194 /* we need as many queues as traffic classes */
2195 num_rx_queues = num_tcs;
2198 /* if we have a bad config abort request queue reset */
2199 if ((adapter->num_rx_queues != num_rx_queues) ||
2200 (adapter->num_tx_queues != num_tx_queues)) {
2201 /* force mailbox timeout to prevent further messages */
2202 hw->mbx.timeout = 0;
2204 /* wait for watchdog to come around and bail us out */
2205 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2211 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2213 ixgbevf_configure_dcb(adapter);
2215 ixgbevf_set_rx_mode(adapter->netdev);
2217 ixgbevf_restore_vlan(adapter);
2218 ixgbevf_ipsec_restore(adapter);
2220 ixgbevf_configure_tx(adapter);
2221 ixgbevf_configure_rx(adapter);
2224 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2226 /* Only save pre-reset stats if there are some */
2227 if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2228 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2229 adapter->stats.base_vfgprc;
2230 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2231 adapter->stats.base_vfgptc;
2232 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2233 adapter->stats.base_vfgorc;
2234 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2235 adapter->stats.base_vfgotc;
2236 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2237 adapter->stats.base_vfmprc;
2241 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2243 struct ixgbe_hw *hw = &adapter->hw;
2245 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2246 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2247 adapter->stats.last_vfgorc |=
2248 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2249 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2250 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2251 adapter->stats.last_vfgotc |=
2252 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2253 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2255 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2256 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2257 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2258 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2259 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2262 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2264 struct ixgbe_hw *hw = &adapter->hw;
2265 int api[] = { ixgbe_mbox_api_14,
2270 ixgbe_mbox_api_unknown };
2273 spin_lock_bh(&adapter->mbx_lock);
2275 while (api[idx] != ixgbe_mbox_api_unknown) {
2276 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2282 spin_unlock_bh(&adapter->mbx_lock);
2285 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2287 struct net_device *netdev = adapter->netdev;
2288 struct ixgbe_hw *hw = &adapter->hw;
2290 ixgbevf_configure_msix(adapter);
2292 spin_lock_bh(&adapter->mbx_lock);
2294 if (is_valid_ether_addr(hw->mac.addr))
2295 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2297 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2299 spin_unlock_bh(&adapter->mbx_lock);
2301 smp_mb__before_atomic();
2302 clear_bit(__IXGBEVF_DOWN, &adapter->state);
2303 ixgbevf_napi_enable_all(adapter);
2305 /* clear any pending interrupts, may auto mask */
2306 IXGBE_READ_REG(hw, IXGBE_VTEICR);
2307 ixgbevf_irq_enable(adapter);
2309 /* enable transmits */
2310 netif_tx_start_all_queues(netdev);
2312 ixgbevf_save_reset_stats(adapter);
2313 ixgbevf_init_last_counter_stats(adapter);
2315 hw->mac.get_link_status = 1;
2316 mod_timer(&adapter->service_timer, jiffies);
2319 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2321 ixgbevf_configure(adapter);
2323 ixgbevf_up_complete(adapter);
2327 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2328 * @rx_ring: ring to free buffers from
2330 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2332 u16 i = rx_ring->next_to_clean;
2334 /* Free Rx ring sk_buff */
2336 dev_kfree_skb(rx_ring->skb);
2337 rx_ring->skb = NULL;
2340 /* Free all the Rx ring pages */
2341 while (i != rx_ring->next_to_alloc) {
2342 struct ixgbevf_rx_buffer *rx_buffer;
2344 rx_buffer = &rx_ring->rx_buffer_info[i];
2346 /* Invalidate cache lines that may have been written to by
2347 * device so that we avoid corrupting memory.
2349 dma_sync_single_range_for_cpu(rx_ring->dev,
2351 rx_buffer->page_offset,
2352 ixgbevf_rx_bufsz(rx_ring),
2355 /* free resources associated with mapping */
2356 dma_unmap_page_attrs(rx_ring->dev,
2358 ixgbevf_rx_pg_size(rx_ring),
2360 IXGBEVF_RX_DMA_ATTR);
2362 __page_frag_cache_drain(rx_buffer->page,
2363 rx_buffer->pagecnt_bias);
2366 if (i == rx_ring->count)
2370 rx_ring->next_to_alloc = 0;
2371 rx_ring->next_to_clean = 0;
2372 rx_ring->next_to_use = 0;
2376 * ixgbevf_clean_tx_ring - Free Tx Buffers
2377 * @tx_ring: ring to be cleaned
2379 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2381 u16 i = tx_ring->next_to_clean;
2382 struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
2384 while (i != tx_ring->next_to_use) {
2385 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
2387 /* Free all the Tx ring sk_buffs */
2388 if (ring_is_xdp(tx_ring))
2389 page_frag_free(tx_buffer->data);
2391 dev_kfree_skb_any(tx_buffer->skb);
2393 /* unmap skb header data */
2394 dma_unmap_single(tx_ring->dev,
2395 dma_unmap_addr(tx_buffer, dma),
2396 dma_unmap_len(tx_buffer, len),
2399 /* check for eop_desc to determine the end of the packet */
2400 eop_desc = tx_buffer->next_to_watch;
2401 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2403 /* unmap remaining buffers */
2404 while (tx_desc != eop_desc) {
2408 if (unlikely(i == tx_ring->count)) {
2410 tx_buffer = tx_ring->tx_buffer_info;
2411 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
2414 /* unmap any remaining paged data */
2415 if (dma_unmap_len(tx_buffer, len))
2416 dma_unmap_page(tx_ring->dev,
2417 dma_unmap_addr(tx_buffer, dma),
2418 dma_unmap_len(tx_buffer, len),
2422 /* move us one more past the eop_desc for start of next pkt */
2425 if (unlikely(i == tx_ring->count)) {
2427 tx_buffer = tx_ring->tx_buffer_info;
2431 /* reset next_to_use and next_to_clean */
2432 tx_ring->next_to_use = 0;
2433 tx_ring->next_to_clean = 0;
2438 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2439 * @adapter: board private structure
2441 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2445 for (i = 0; i < adapter->num_rx_queues; i++)
2446 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2450 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2451 * @adapter: board private structure
2453 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2457 for (i = 0; i < adapter->num_tx_queues; i++)
2458 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2459 for (i = 0; i < adapter->num_xdp_queues; i++)
2460 ixgbevf_clean_tx_ring(adapter->xdp_ring[i]);
2463 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2465 struct net_device *netdev = adapter->netdev;
2466 struct ixgbe_hw *hw = &adapter->hw;
2469 /* signal that we are down to the interrupt handler */
2470 if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2471 return; /* do nothing if already down */
2473 /* disable all enabled Rx queues */
2474 for (i = 0; i < adapter->num_rx_queues; i++)
2475 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2477 usleep_range(10000, 20000);
2479 netif_tx_stop_all_queues(netdev);
2481 /* call carrier off first to avoid false dev_watchdog timeouts */
2482 netif_carrier_off(netdev);
2483 netif_tx_disable(netdev);
2485 ixgbevf_irq_disable(adapter);
2487 ixgbevf_napi_disable_all(adapter);
2489 del_timer_sync(&adapter->service_timer);
2491 /* disable transmits in the hardware now that interrupts are off */
2492 for (i = 0; i < adapter->num_tx_queues; i++) {
2493 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2495 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2496 IXGBE_TXDCTL_SWFLSH);
2499 for (i = 0; i < adapter->num_xdp_queues; i++) {
2500 u8 reg_idx = adapter->xdp_ring[i]->reg_idx;
2502 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2503 IXGBE_TXDCTL_SWFLSH);
2506 if (!pci_channel_offline(adapter->pdev))
2507 ixgbevf_reset(adapter);
2509 ixgbevf_clean_all_tx_rings(adapter);
2510 ixgbevf_clean_all_rx_rings(adapter);
2513 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2515 WARN_ON(in_interrupt());
2517 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2520 ixgbevf_down(adapter);
2521 ixgbevf_up(adapter);
2523 clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2526 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2528 struct ixgbe_hw *hw = &adapter->hw;
2529 struct net_device *netdev = adapter->netdev;
2531 if (hw->mac.ops.reset_hw(hw)) {
2532 hw_dbg(hw, "PF still resetting\n");
2534 hw->mac.ops.init_hw(hw);
2535 ixgbevf_negotiate_api(adapter);
2538 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2539 ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2540 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2543 adapter->last_reset = jiffies;
2546 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2549 int vector_threshold;
2551 /* We'll want at least 2 (vector_threshold):
2552 * 1) TxQ[0] + RxQ[0] handler
2553 * 2) Other (Link Status Change, etc.)
2555 vector_threshold = MIN_MSIX_COUNT;
2557 /* The more we get, the more we will assign to Tx/Rx Cleanup
2558 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2559 * Right now, we simply care about how many we'll get; we'll
2560 * set them up later while requesting irq's.
2562 vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2563 vector_threshold, vectors);
2566 dev_err(&adapter->pdev->dev,
2567 "Unable to allocate MSI-X interrupts\n");
2568 kfree(adapter->msix_entries);
2569 adapter->msix_entries = NULL;
2573 /* Adjust for only the vectors we'll use, which is minimum
2574 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2575 * vectors we were allocated.
2577 adapter->num_msix_vectors = vectors;
2583 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2584 * @adapter: board private structure to initialize
2586 * This is the top level queue allocation routine. The order here is very
2587 * important, starting with the "most" number of features turned on at once,
2588 * and ending with the smallest set of features. This way large combinations
2589 * can be allocated if they're turned on, and smaller combinations are the
2590 * fallthrough conditions.
2593 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2595 struct ixgbe_hw *hw = &adapter->hw;
2596 unsigned int def_q = 0;
2597 unsigned int num_tcs = 0;
2600 /* Start with base case */
2601 adapter->num_rx_queues = 1;
2602 adapter->num_tx_queues = 1;
2603 adapter->num_xdp_queues = 0;
2605 spin_lock_bh(&adapter->mbx_lock);
2607 /* fetch queue configuration from the PF */
2608 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2610 spin_unlock_bh(&adapter->mbx_lock);
2615 /* we need as many queues as traffic classes */
2617 adapter->num_rx_queues = num_tcs;
2619 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2621 switch (hw->api_version) {
2622 case ixgbe_mbox_api_11:
2623 case ixgbe_mbox_api_12:
2624 case ixgbe_mbox_api_13:
2625 case ixgbe_mbox_api_14:
2626 if (adapter->xdp_prog &&
2627 hw->mac.max_tx_queues == rss)
2628 rss = rss > 3 ? 2 : 1;
2630 adapter->num_rx_queues = rss;
2631 adapter->num_tx_queues = rss;
2632 adapter->num_xdp_queues = adapter->xdp_prog ? rss : 0;
2640 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2641 * @adapter: board private structure to initialize
2643 * Attempt to configure the interrupts using the best available
2644 * capabilities of the hardware and the kernel.
2646 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2648 int vector, v_budget;
2650 /* It's easy to be greedy for MSI-X vectors, but it really
2651 * doesn't do us much good if we have a lot more vectors
2652 * than CPU's. So let's be conservative and only ask for
2653 * (roughly) the same number of vectors as there are CPU's.
2654 * The default is to use pairs of vectors.
2656 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2657 v_budget = min_t(int, v_budget, num_online_cpus());
2658 v_budget += NON_Q_VECTORS;
2660 adapter->msix_entries = kcalloc(v_budget,
2661 sizeof(struct msix_entry), GFP_KERNEL);
2662 if (!adapter->msix_entries)
2665 for (vector = 0; vector < v_budget; vector++)
2666 adapter->msix_entries[vector].entry = vector;
2668 /* A failure in MSI-X entry allocation isn't fatal, but the VF driver
2669 * does not support any other modes, so we will simply fail here. Note
2670 * that we clean up the msix_entries pointer else-where.
2672 return ixgbevf_acquire_msix_vectors(adapter, v_budget);
2675 static void ixgbevf_add_ring(struct ixgbevf_ring *ring,
2676 struct ixgbevf_ring_container *head)
2678 ring->next = head->ring;
2684 * ixgbevf_alloc_q_vector - Allocate memory for a single interrupt vector
2685 * @adapter: board private structure to initialize
2686 * @v_idx: index of vector in adapter struct
2687 * @txr_count: number of Tx rings for q vector
2688 * @txr_idx: index of first Tx ring to assign
2689 * @xdp_count: total number of XDP rings to allocate
2690 * @xdp_idx: index of first XDP ring to allocate
2691 * @rxr_count: number of Rx rings for q vector
2692 * @rxr_idx: index of first Rx ring to assign
2694 * We allocate one q_vector. If allocation fails we return -ENOMEM.
2696 static int ixgbevf_alloc_q_vector(struct ixgbevf_adapter *adapter, int v_idx,
2697 int txr_count, int txr_idx,
2698 int xdp_count, int xdp_idx,
2699 int rxr_count, int rxr_idx)
2701 struct ixgbevf_q_vector *q_vector;
2702 int reg_idx = txr_idx + xdp_idx;
2703 struct ixgbevf_ring *ring;
2704 int ring_count, size;
2706 ring_count = txr_count + xdp_count + rxr_count;
2707 size = sizeof(*q_vector) + (sizeof(*ring) * ring_count);
2709 /* allocate q_vector and rings */
2710 q_vector = kzalloc(size, GFP_KERNEL);
2714 /* initialize NAPI */
2715 netif_napi_add(adapter->netdev, &q_vector->napi, ixgbevf_poll, 64);
2717 /* tie q_vector and adapter together */
2718 adapter->q_vector[v_idx] = q_vector;
2719 q_vector->adapter = adapter;
2720 q_vector->v_idx = v_idx;
2722 /* initialize pointer to rings */
2723 ring = q_vector->ring;
2726 /* assign generic ring traits */
2727 ring->dev = &adapter->pdev->dev;
2728 ring->netdev = adapter->netdev;
2730 /* configure backlink on ring */
2731 ring->q_vector = q_vector;
2733 /* update q_vector Tx values */
2734 ixgbevf_add_ring(ring, &q_vector->tx);
2736 /* apply Tx specific ring traits */
2737 ring->count = adapter->tx_ring_count;
2738 ring->queue_index = txr_idx;
2739 ring->reg_idx = reg_idx;
2741 /* assign ring to adapter */
2742 adapter->tx_ring[txr_idx] = ring;
2744 /* update count and index */
2749 /* push pointer to next ring */
2754 /* assign generic ring traits */
2755 ring->dev = &adapter->pdev->dev;
2756 ring->netdev = adapter->netdev;
2758 /* configure backlink on ring */
2759 ring->q_vector = q_vector;
2761 /* update q_vector Tx values */
2762 ixgbevf_add_ring(ring, &q_vector->tx);
2764 /* apply Tx specific ring traits */
2765 ring->count = adapter->tx_ring_count;
2766 ring->queue_index = xdp_idx;
2767 ring->reg_idx = reg_idx;
2770 /* assign ring to adapter */
2771 adapter->xdp_ring[xdp_idx] = ring;
2773 /* update count and index */
2778 /* push pointer to next ring */
2783 /* assign generic ring traits */
2784 ring->dev = &adapter->pdev->dev;
2785 ring->netdev = adapter->netdev;
2787 /* configure backlink on ring */
2788 ring->q_vector = q_vector;
2790 /* update q_vector Rx values */
2791 ixgbevf_add_ring(ring, &q_vector->rx);
2793 /* apply Rx specific ring traits */
2794 ring->count = adapter->rx_ring_count;
2795 ring->queue_index = rxr_idx;
2796 ring->reg_idx = rxr_idx;
2798 /* assign ring to adapter */
2799 adapter->rx_ring[rxr_idx] = ring;
2801 /* update count and index */
2805 /* push pointer to next ring */
2813 * ixgbevf_free_q_vector - Free memory allocated for specific interrupt vector
2814 * @adapter: board private structure to initialize
2815 * @v_idx: index of vector in adapter struct
2817 * This function frees the memory allocated to the q_vector. In addition if
2818 * NAPI is enabled it will delete any references to the NAPI struct prior
2819 * to freeing the q_vector.
2821 static void ixgbevf_free_q_vector(struct ixgbevf_adapter *adapter, int v_idx)
2823 struct ixgbevf_q_vector *q_vector = adapter->q_vector[v_idx];
2824 struct ixgbevf_ring *ring;
2826 ixgbevf_for_each_ring(ring, q_vector->tx) {
2827 if (ring_is_xdp(ring))
2828 adapter->xdp_ring[ring->queue_index] = NULL;
2830 adapter->tx_ring[ring->queue_index] = NULL;
2833 ixgbevf_for_each_ring(ring, q_vector->rx)
2834 adapter->rx_ring[ring->queue_index] = NULL;
2836 adapter->q_vector[v_idx] = NULL;
2837 netif_napi_del(&q_vector->napi);
2839 /* ixgbevf_get_stats() might access the rings on this vector,
2840 * we must wait a grace period before freeing it.
2842 kfree_rcu(q_vector, rcu);
2846 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2847 * @adapter: board private structure to initialize
2849 * We allocate one q_vector per queue interrupt. If allocation fails we
2852 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2854 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2855 int rxr_remaining = adapter->num_rx_queues;
2856 int txr_remaining = adapter->num_tx_queues;
2857 int xdp_remaining = adapter->num_xdp_queues;
2858 int rxr_idx = 0, txr_idx = 0, xdp_idx = 0, v_idx = 0;
2861 if (q_vectors >= (rxr_remaining + txr_remaining + xdp_remaining)) {
2862 for (; rxr_remaining; v_idx++, q_vectors--) {
2863 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2865 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2866 0, 0, 0, 0, rqpv, rxr_idx);
2870 /* update counts and index */
2871 rxr_remaining -= rqpv;
2876 for (; q_vectors; v_idx++, q_vectors--) {
2877 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2878 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors);
2879 int xqpv = DIV_ROUND_UP(xdp_remaining, q_vectors);
2881 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2889 /* update counts and index */
2890 rxr_remaining -= rqpv;
2892 txr_remaining -= tqpv;
2894 xdp_remaining -= xqpv;
2903 ixgbevf_free_q_vector(adapter, v_idx);
2910 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2911 * @adapter: board private structure to initialize
2913 * This function frees the memory allocated to the q_vectors. In addition if
2914 * NAPI is enabled it will delete any references to the NAPI struct prior
2915 * to freeing the q_vector.
2917 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2919 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2923 ixgbevf_free_q_vector(adapter, q_vectors);
2928 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2929 * @adapter: board private structure
2932 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2934 if (!adapter->msix_entries)
2937 pci_disable_msix(adapter->pdev);
2938 kfree(adapter->msix_entries);
2939 adapter->msix_entries = NULL;
2943 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2944 * @adapter: board private structure to initialize
2947 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2951 /* Number of supported queues */
2952 ixgbevf_set_num_queues(adapter);
2954 err = ixgbevf_set_interrupt_capability(adapter);
2956 hw_dbg(&adapter->hw,
2957 "Unable to setup interrupt capabilities\n");
2958 goto err_set_interrupt;
2961 err = ixgbevf_alloc_q_vectors(adapter);
2963 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2964 goto err_alloc_q_vectors;
2967 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u XDP Queue count %u\n",
2968 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
2969 adapter->num_rx_queues, adapter->num_tx_queues,
2970 adapter->num_xdp_queues);
2972 set_bit(__IXGBEVF_DOWN, &adapter->state);
2975 err_alloc_q_vectors:
2976 ixgbevf_reset_interrupt_capability(adapter);
2982 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2983 * @adapter: board private structure to clear interrupt scheme on
2985 * We go through and clear interrupt specific resources and reset the structure
2986 * to pre-load conditions
2988 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2990 adapter->num_tx_queues = 0;
2991 adapter->num_xdp_queues = 0;
2992 adapter->num_rx_queues = 0;
2994 ixgbevf_free_q_vectors(adapter);
2995 ixgbevf_reset_interrupt_capability(adapter);
2999 * ixgbevf_sw_init - Initialize general software structures
3000 * @adapter: board private structure to initialize
3002 * ixgbevf_sw_init initializes the Adapter private data structure.
3003 * Fields are initialized based on PCI device information and
3004 * OS network device settings (MTU size).
3006 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
3008 struct ixgbe_hw *hw = &adapter->hw;
3009 struct pci_dev *pdev = adapter->pdev;
3010 struct net_device *netdev = adapter->netdev;
3013 /* PCI config space info */
3014 hw->vendor_id = pdev->vendor;
3015 hw->device_id = pdev->device;
3016 hw->revision_id = pdev->revision;
3017 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3018 hw->subsystem_device_id = pdev->subsystem_device;
3020 hw->mbx.ops.init_params(hw);
3022 if (hw->mac.type >= ixgbe_mac_X550_vf) {
3023 err = ixgbevf_init_rss_key(adapter);
3028 /* assume legacy case in which PF would only give VF 2 queues */
3029 hw->mac.max_tx_queues = 2;
3030 hw->mac.max_rx_queues = 2;
3032 /* lock to protect mailbox accesses */
3033 spin_lock_init(&adapter->mbx_lock);
3035 err = hw->mac.ops.reset_hw(hw);
3037 dev_info(&pdev->dev,
3038 "PF still in reset state. Is the PF interface up?\n");
3040 err = hw->mac.ops.init_hw(hw);
3042 pr_err("init_shared_code failed: %d\n", err);
3045 ixgbevf_negotiate_api(adapter);
3046 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
3048 dev_info(&pdev->dev, "Error reading MAC address\n");
3049 else if (is_zero_ether_addr(adapter->hw.mac.addr))
3050 dev_info(&pdev->dev,
3051 "MAC address not assigned by administrator.\n");
3052 ether_addr_copy(netdev->dev_addr, hw->mac.addr);
3055 if (!is_valid_ether_addr(netdev->dev_addr)) {
3056 dev_info(&pdev->dev, "Assigning random MAC address\n");
3057 eth_hw_addr_random(netdev);
3058 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
3059 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
3062 /* Enable dynamic interrupt throttling rates */
3063 adapter->rx_itr_setting = 1;
3064 adapter->tx_itr_setting = 1;
3066 /* set default ring sizes */
3067 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
3068 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
3070 set_bit(__IXGBEVF_DOWN, &adapter->state);
3077 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
3079 u32 current_counter = IXGBE_READ_REG(hw, reg); \
3080 if (current_counter < last_counter) \
3081 counter += 0x100000000LL; \
3082 last_counter = current_counter; \
3083 counter &= 0xFFFFFFFF00000000LL; \
3084 counter |= current_counter; \
3087 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
3089 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
3090 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
3091 u64 current_counter = (current_counter_msb << 32) | \
3092 current_counter_lsb; \
3093 if (current_counter < last_counter) \
3094 counter += 0x1000000000LL; \
3095 last_counter = current_counter; \
3096 counter &= 0xFFFFFFF000000000LL; \
3097 counter |= current_counter; \
3100 * ixgbevf_update_stats - Update the board statistics counters.
3101 * @adapter: board private structure
3103 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
3105 struct ixgbe_hw *hw = &adapter->hw;
3106 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
3107 u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
3110 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3111 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3114 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
3115 adapter->stats.vfgprc);
3116 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
3117 adapter->stats.vfgptc);
3118 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
3119 adapter->stats.last_vfgorc,
3120 adapter->stats.vfgorc);
3121 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
3122 adapter->stats.last_vfgotc,
3123 adapter->stats.vfgotc);
3124 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
3125 adapter->stats.vfmprc);
3127 for (i = 0; i < adapter->num_rx_queues; i++) {
3128 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
3130 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
3131 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
3132 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
3133 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
3136 adapter->hw_csum_rx_error = hw_csum_rx_error;
3137 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
3138 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
3139 adapter->alloc_rx_page = alloc_rx_page;
3143 * ixgbevf_service_timer - Timer Call-back
3144 * @t: pointer to timer_list struct
3146 static void ixgbevf_service_timer(struct timer_list *t)
3148 struct ixgbevf_adapter *adapter = from_timer(adapter, t,
3151 /* Reset the timer */
3152 mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
3154 ixgbevf_service_event_schedule(adapter);
3157 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
3159 if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
3163 /* If we're already down or resetting, just bail */
3164 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3165 test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
3166 test_bit(__IXGBEVF_RESETTING, &adapter->state)) {
3171 adapter->tx_timeout_count++;
3173 ixgbevf_reinit_locked(adapter);
3178 * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
3179 * @adapter: pointer to the device adapter structure
3181 * This function serves two purposes. First it strobes the interrupt lines
3182 * in order to make certain interrupts are occurring. Secondly it sets the
3183 * bits needed to check for TX hangs. As a result we should immediately
3184 * determine if a hang has occurred.
3186 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
3188 struct ixgbe_hw *hw = &adapter->hw;
3192 /* If we're down or resetting, just bail */
3193 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3194 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3197 /* Force detection of hung controller */
3198 if (netif_carrier_ok(adapter->netdev)) {
3199 for (i = 0; i < adapter->num_tx_queues; i++)
3200 set_check_for_tx_hang(adapter->tx_ring[i]);
3201 for (i = 0; i < adapter->num_xdp_queues; i++)
3202 set_check_for_tx_hang(adapter->xdp_ring[i]);
3205 /* get one bit for every active Tx/Rx interrupt vector */
3206 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
3207 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
3209 if (qv->rx.ring || qv->tx.ring)
3213 /* Cause software interrupt to ensure rings are cleaned */
3214 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
3218 * ixgbevf_watchdog_update_link - update the link status
3219 * @adapter: pointer to the device adapter structure
3221 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
3223 struct ixgbe_hw *hw = &adapter->hw;
3224 u32 link_speed = adapter->link_speed;
3225 bool link_up = adapter->link_up;
3228 spin_lock_bh(&adapter->mbx_lock);
3230 err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3232 spin_unlock_bh(&adapter->mbx_lock);
3234 /* if check for link returns error we will need to reset */
3235 if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
3236 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
3240 adapter->link_up = link_up;
3241 adapter->link_speed = link_speed;
3245 * ixgbevf_watchdog_link_is_up - update netif_carrier status and
3246 * print link up message
3247 * @adapter: pointer to the device adapter structure
3249 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
3251 struct net_device *netdev = adapter->netdev;
3253 /* only continue if link was previously down */
3254 if (netif_carrier_ok(netdev))
3257 dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
3258 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
3260 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
3262 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
3266 netif_carrier_on(netdev);
3270 * ixgbevf_watchdog_link_is_down - update netif_carrier status and
3271 * print link down message
3272 * @adapter: pointer to the adapter structure
3274 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
3276 struct net_device *netdev = adapter->netdev;
3278 adapter->link_speed = 0;
3280 /* only continue if link was up previously */
3281 if (!netif_carrier_ok(netdev))
3284 dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
3286 netif_carrier_off(netdev);
3290 * ixgbevf_watchdog_subtask - worker thread to bring link up
3291 * @adapter: board private structure
3293 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
3295 /* if interface is down do nothing */
3296 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3297 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3300 ixgbevf_watchdog_update_link(adapter);
3302 if (adapter->link_up)
3303 ixgbevf_watchdog_link_is_up(adapter);
3305 ixgbevf_watchdog_link_is_down(adapter);
3307 ixgbevf_update_stats(adapter);
3311 * ixgbevf_service_task - manages and runs subtasks
3312 * @work: pointer to work_struct containing our data
3314 static void ixgbevf_service_task(struct work_struct *work)
3316 struct ixgbevf_adapter *adapter = container_of(work,
3317 struct ixgbevf_adapter,
3319 struct ixgbe_hw *hw = &adapter->hw;
3321 if (IXGBE_REMOVED(hw->hw_addr)) {
3322 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
3324 ixgbevf_down(adapter);
3330 ixgbevf_queue_reset_subtask(adapter);
3331 ixgbevf_reset_subtask(adapter);
3332 ixgbevf_watchdog_subtask(adapter);
3333 ixgbevf_check_hang_subtask(adapter);
3335 ixgbevf_service_event_complete(adapter);
3339 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
3340 * @tx_ring: Tx descriptor ring for a specific queue
3342 * Free all transmit software resources
3344 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
3346 ixgbevf_clean_tx_ring(tx_ring);
3348 vfree(tx_ring->tx_buffer_info);
3349 tx_ring->tx_buffer_info = NULL;
3351 /* if not set, then don't free */
3355 dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
3358 tx_ring->desc = NULL;
3362 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
3363 * @adapter: board private structure
3365 * Free all transmit software resources
3367 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
3371 for (i = 0; i < adapter->num_tx_queues; i++)
3372 if (adapter->tx_ring[i]->desc)
3373 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3374 for (i = 0; i < adapter->num_xdp_queues; i++)
3375 if (adapter->xdp_ring[i]->desc)
3376 ixgbevf_free_tx_resources(adapter->xdp_ring[i]);
3380 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
3381 * @tx_ring: Tx descriptor ring (for a specific queue) to setup
3383 * Return 0 on success, negative on failure
3385 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
3387 struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3390 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3391 tx_ring->tx_buffer_info = vmalloc(size);
3392 if (!tx_ring->tx_buffer_info)
3395 u64_stats_init(&tx_ring->syncp);
3397 /* round up to nearest 4K */
3398 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3399 tx_ring->size = ALIGN(tx_ring->size, 4096);
3401 tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3402 &tx_ring->dma, GFP_KERNEL);
3409 vfree(tx_ring->tx_buffer_info);
3410 tx_ring->tx_buffer_info = NULL;
3411 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3416 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3417 * @adapter: board private structure
3419 * If this function returns with an error, then it's possible one or
3420 * more of the rings is populated (while the rest are not). It is the
3421 * callers duty to clean those orphaned rings.
3423 * Return 0 on success, negative on failure
3425 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3427 int i, j = 0, err = 0;
3429 for (i = 0; i < adapter->num_tx_queues; i++) {
3430 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3433 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3437 for (j = 0; j < adapter->num_xdp_queues; j++) {
3438 err = ixgbevf_setup_tx_resources(adapter->xdp_ring[j]);
3441 hw_dbg(&adapter->hw, "Allocation for XDP Queue %u failed\n", j);
3447 /* rewind the index freeing the rings as we go */
3449 ixgbevf_free_tx_resources(adapter->xdp_ring[j]);
3451 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3457 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3458 * @adapter: board private structure
3459 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3461 * Returns 0 on success, negative on failure
3463 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
3464 struct ixgbevf_ring *rx_ring)
3468 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3469 rx_ring->rx_buffer_info = vmalloc(size);
3470 if (!rx_ring->rx_buffer_info)
3473 u64_stats_init(&rx_ring->syncp);
3475 /* Round up to nearest 4K */
3476 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3477 rx_ring->size = ALIGN(rx_ring->size, 4096);
3479 rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3480 &rx_ring->dma, GFP_KERNEL);
3485 /* XDP RX-queue info */
3486 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
3487 rx_ring->queue_index) < 0)
3490 rx_ring->xdp_prog = adapter->xdp_prog;
3494 vfree(rx_ring->rx_buffer_info);
3495 rx_ring->rx_buffer_info = NULL;
3496 dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3501 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3502 * @adapter: board private structure
3504 * If this function returns with an error, then it's possible one or
3505 * more of the rings is populated (while the rest are not). It is the
3506 * callers duty to clean those orphaned rings.
3508 * Return 0 on success, negative on failure
3510 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3514 for (i = 0; i < adapter->num_rx_queues; i++) {
3515 err = ixgbevf_setup_rx_resources(adapter, adapter->rx_ring[i]);
3518 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3524 /* rewind the index freeing the rings as we go */
3526 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3531 * ixgbevf_free_rx_resources - Free Rx Resources
3532 * @rx_ring: ring to clean the resources from
3534 * Free all receive software resources
3536 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3538 ixgbevf_clean_rx_ring(rx_ring);
3540 rx_ring->xdp_prog = NULL;
3541 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
3542 vfree(rx_ring->rx_buffer_info);
3543 rx_ring->rx_buffer_info = NULL;
3545 dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3548 rx_ring->desc = NULL;
3552 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3553 * @adapter: board private structure
3555 * Free all receive software resources
3557 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3561 for (i = 0; i < adapter->num_rx_queues; i++)
3562 if (adapter->rx_ring[i]->desc)
3563 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3567 * ixgbevf_open - Called when a network interface is made active
3568 * @netdev: network interface device structure
3570 * Returns 0 on success, negative value on failure
3572 * The open entry point is called when a network interface is made
3573 * active by the system (IFF_UP). At this point all resources needed
3574 * for transmit and receive operations are allocated, the interrupt
3575 * handler is registered with the OS, the watchdog timer is started,
3576 * and the stack is notified that the interface is ready.
3578 int ixgbevf_open(struct net_device *netdev)
3580 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3581 struct ixgbe_hw *hw = &adapter->hw;
3584 /* A previous failure to open the device because of a lack of
3585 * available MSIX vector resources may have reset the number
3586 * of msix vectors variable to zero. The only way to recover
3587 * is to unload/reload the driver and hope that the system has
3588 * been able to recover some MSIX vector resources.
3590 if (!adapter->num_msix_vectors)
3593 if (hw->adapter_stopped) {
3594 ixgbevf_reset(adapter);
3595 /* if adapter is still stopped then PF isn't up and
3596 * the VF can't start.
3598 if (hw->adapter_stopped) {
3599 err = IXGBE_ERR_MBX;
3600 pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3601 goto err_setup_reset;
3605 /* disallow open during test */
3606 if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3609 netif_carrier_off(netdev);
3611 /* allocate transmit descriptors */
3612 err = ixgbevf_setup_all_tx_resources(adapter);
3616 /* allocate receive descriptors */
3617 err = ixgbevf_setup_all_rx_resources(adapter);
3621 ixgbevf_configure(adapter);
3623 err = ixgbevf_request_irq(adapter);
3627 /* Notify the stack of the actual queue counts. */
3628 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
3630 goto err_set_queues;
3632 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
3634 goto err_set_queues;
3636 ixgbevf_up_complete(adapter);
3641 ixgbevf_free_irq(adapter);
3643 ixgbevf_free_all_rx_resources(adapter);
3645 ixgbevf_free_all_tx_resources(adapter);
3647 ixgbevf_reset(adapter);
3654 * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3655 * @adapter: the private adapter struct
3657 * This function should contain the necessary work common to both suspending
3658 * and closing of the device.
3660 static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3662 ixgbevf_down(adapter);
3663 ixgbevf_free_irq(adapter);
3664 ixgbevf_free_all_tx_resources(adapter);
3665 ixgbevf_free_all_rx_resources(adapter);
3669 * ixgbevf_close - Disables a network interface
3670 * @netdev: network interface device structure
3672 * Returns 0, this is not allowed to fail
3674 * The close entry point is called when an interface is de-activated
3675 * by the OS. The hardware is still under the drivers control, but
3676 * needs to be disabled. A global MAC reset is issued to stop the
3677 * hardware, and all transmit and receive resources are freed.
3679 int ixgbevf_close(struct net_device *netdev)
3681 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3683 if (netif_device_present(netdev))
3684 ixgbevf_close_suspend(adapter);
3689 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3691 struct net_device *dev = adapter->netdev;
3693 if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3697 /* if interface is down do nothing */
3698 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3699 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3702 /* Hardware has to reinitialize queues and interrupts to
3703 * match packet buffer alignment. Unfortunately, the
3704 * hardware is not flexible enough to do this dynamically.
3708 if (netif_running(dev))
3711 ixgbevf_clear_interrupt_scheme(adapter);
3712 ixgbevf_init_interrupt_scheme(adapter);
3714 if (netif_running(dev))
3720 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3721 u32 vlan_macip_lens, u32 fceof_saidx,
3722 u32 type_tucmd, u32 mss_l4len_idx)
3724 struct ixgbe_adv_tx_context_desc *context_desc;
3725 u16 i = tx_ring->next_to_use;
3727 context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3730 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3732 /* set bits to identify this as an advanced context descriptor */
3733 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3735 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3736 context_desc->fceof_saidx = cpu_to_le32(fceof_saidx);
3737 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
3738 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3741 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3742 struct ixgbevf_tx_buffer *first,
3744 struct ixgbevf_ipsec_tx_data *itd)
3746 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3747 struct sk_buff *skb = first->skb;
3757 u32 paylen, l4_offset;
3758 u32 fceof_saidx = 0;
3761 if (skb->ip_summed != CHECKSUM_PARTIAL)
3764 if (!skb_is_gso(skb))
3767 err = skb_cow_head(skb, 0);
3771 if (eth_p_mpls(first->protocol))
3772 ip.hdr = skb_inner_network_header(skb);
3774 ip.hdr = skb_network_header(skb);
3775 l4.hdr = skb_checksum_start(skb);
3777 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3778 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3780 /* initialize outer IP header fields */
3781 if (ip.v4->version == 4) {
3782 unsigned char *csum_start = skb_checksum_start(skb);
3783 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3784 int len = csum_start - trans_start;
3786 /* IP header will have to cancel out any data that
3787 * is not a part of the outer IP header, so set to
3788 * a reverse csum if needed, else init check to 0.
3790 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
3791 csum_fold(csum_partial(trans_start,
3793 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3796 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3797 IXGBE_TX_FLAGS_CSUM |
3798 IXGBE_TX_FLAGS_IPV4;
3800 ip.v6->payload_len = 0;
3801 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3802 IXGBE_TX_FLAGS_CSUM;
3805 /* determine offset of inner transport header */
3806 l4_offset = l4.hdr - skb->data;
3808 /* compute length of segmentation header */
3809 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3811 /* remove payload length from inner checksum */
3812 paylen = skb->len - l4_offset;
3813 csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
3815 /* update gso size and bytecount with header size */
3816 first->gso_segs = skb_shinfo(skb)->gso_segs;
3817 first->bytecount += (first->gso_segs - 1) * *hdr_len;
3819 /* mss_l4len_id: use 1 as index for TSO */
3820 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3821 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3822 mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3824 fceof_saidx |= itd->pfsa;
3825 type_tucmd |= itd->flags | itd->trailer_len;
3827 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3828 vlan_macip_lens = l4.hdr - ip.hdr;
3829 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3830 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3832 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
3838 static inline bool ixgbevf_ipv6_csum_is_sctp(struct sk_buff *skb)
3840 unsigned int offset = 0;
3842 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
3844 return offset == skb_checksum_start_offset(skb);
3847 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3848 struct ixgbevf_tx_buffer *first,
3849 struct ixgbevf_ipsec_tx_data *itd)
3851 struct sk_buff *skb = first->skb;
3852 u32 vlan_macip_lens = 0;
3853 u32 fceof_saidx = 0;
3856 if (skb->ip_summed != CHECKSUM_PARTIAL)
3859 switch (skb->csum_offset) {
3860 case offsetof(struct tcphdr, check):
3861 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3863 case offsetof(struct udphdr, check):
3865 case offsetof(struct sctphdr, checksum):
3866 /* validate that this is actually an SCTP request */
3867 if (((first->protocol == htons(ETH_P_IP)) &&
3868 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
3869 ((first->protocol == htons(ETH_P_IPV6)) &&
3870 ixgbevf_ipv6_csum_is_sctp(skb))) {
3871 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3876 skb_checksum_help(skb);
3880 if (first->protocol == htons(ETH_P_IP))
3881 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3883 /* update TX checksum flag */
3884 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3885 vlan_macip_lens = skb_checksum_start_offset(skb) -
3886 skb_network_offset(skb);
3888 /* vlan_macip_lens: MACLEN, VLAN tag */
3889 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3890 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3892 fceof_saidx |= itd->pfsa;
3893 type_tucmd |= itd->flags | itd->trailer_len;
3895 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3896 fceof_saidx, type_tucmd, 0);
3899 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3901 /* set type for advanced descriptor with frame checksum insertion */
3902 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3903 IXGBE_ADVTXD_DCMD_IFCS |
3904 IXGBE_ADVTXD_DCMD_DEXT);
3906 /* set HW VLAN bit if VLAN is present */
3907 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3908 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3910 /* set segmentation enable bits for TSO/FSO */
3911 if (tx_flags & IXGBE_TX_FLAGS_TSO)
3912 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3917 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3918 u32 tx_flags, unsigned int paylen)
3920 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3922 /* enable L4 checksum for TSO and TX checksum offload */
3923 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3924 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3926 /* enble IPv4 checksum for TSO */
3927 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3928 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3931 if (tx_flags & IXGBE_TX_FLAGS_IPSEC)
3932 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IPSEC);
3934 /* use index 1 context for TSO/FSO/FCOE/IPSEC */
3935 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_IPSEC))
3936 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3938 /* Check Context must be set if Tx switch is enabled, which it
3939 * always is for case where virtual functions are running
3941 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3943 tx_desc->read.olinfo_status = olinfo_status;
3946 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3947 struct ixgbevf_tx_buffer *first,
3950 struct sk_buff *skb = first->skb;
3951 struct ixgbevf_tx_buffer *tx_buffer;
3952 union ixgbe_adv_tx_desc *tx_desc;
3953 struct skb_frag_struct *frag;
3955 unsigned int data_len, size;
3956 u32 tx_flags = first->tx_flags;
3957 __le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3958 u16 i = tx_ring->next_to_use;
3960 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3962 ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
3964 size = skb_headlen(skb);
3965 data_len = skb->data_len;
3967 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3971 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3972 if (dma_mapping_error(tx_ring->dev, dma))
3975 /* record length, and DMA address */
3976 dma_unmap_len_set(tx_buffer, len, size);
3977 dma_unmap_addr_set(tx_buffer, dma, dma);
3979 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3981 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3982 tx_desc->read.cmd_type_len =
3983 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3987 if (i == tx_ring->count) {
3988 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3991 tx_desc->read.olinfo_status = 0;
3993 dma += IXGBE_MAX_DATA_PER_TXD;
3994 size -= IXGBE_MAX_DATA_PER_TXD;
3996 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3999 if (likely(!data_len))
4002 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
4006 if (i == tx_ring->count) {
4007 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
4010 tx_desc->read.olinfo_status = 0;
4012 size = skb_frag_size(frag);
4015 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
4018 tx_buffer = &tx_ring->tx_buffer_info[i];
4021 /* write last descriptor with RS and EOP bits */
4022 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
4023 tx_desc->read.cmd_type_len = cmd_type;
4025 /* set the timestamp */
4026 first->time_stamp = jiffies;
4028 skb_tx_timestamp(skb);
4030 /* Force memory writes to complete before letting h/w know there
4031 * are new descriptors to fetch. (Only applicable for weak-ordered
4032 * memory model archs, such as IA-64).
4034 * We also need this memory barrier (wmb) to make certain all of the
4035 * status bits have been updated before next_to_watch is written.
4039 /* set next_to_watch value indicating a packet is present */
4040 first->next_to_watch = tx_desc;
4043 if (i == tx_ring->count)
4046 tx_ring->next_to_use = i;
4048 /* notify HW of packet */
4049 ixgbevf_write_tail(tx_ring, i);
4053 dev_err(tx_ring->dev, "TX DMA map failed\n");
4054 tx_buffer = &tx_ring->tx_buffer_info[i];
4056 /* clear dma mappings for failed tx_buffer_info map */
4057 while (tx_buffer != first) {
4058 if (dma_unmap_len(tx_buffer, len))
4059 dma_unmap_page(tx_ring->dev,
4060 dma_unmap_addr(tx_buffer, dma),
4061 dma_unmap_len(tx_buffer, len),
4063 dma_unmap_len_set(tx_buffer, len, 0);
4066 i += tx_ring->count;
4067 tx_buffer = &tx_ring->tx_buffer_info[i];
4070 if (dma_unmap_len(tx_buffer, len))
4071 dma_unmap_single(tx_ring->dev,
4072 dma_unmap_addr(tx_buffer, dma),
4073 dma_unmap_len(tx_buffer, len),
4075 dma_unmap_len_set(tx_buffer, len, 0);
4077 dev_kfree_skb_any(tx_buffer->skb);
4078 tx_buffer->skb = NULL;
4080 tx_ring->next_to_use = i;
4083 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4085 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
4086 /* Herbert's original patch had:
4087 * smp_mb__after_netif_stop_queue();
4088 * but since that doesn't exist yet, just open code it.
4092 /* We need to check again in a case another CPU has just
4093 * made room available.
4095 if (likely(ixgbevf_desc_unused(tx_ring) < size))
4098 /* A reprieve! - use start_queue because it doesn't call schedule */
4099 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
4100 ++tx_ring->tx_stats.restart_queue;
4105 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4107 if (likely(ixgbevf_desc_unused(tx_ring) >= size))
4109 return __ixgbevf_maybe_stop_tx(tx_ring, size);
4112 static int ixgbevf_xmit_frame_ring(struct sk_buff *skb,
4113 struct ixgbevf_ring *tx_ring)
4115 struct ixgbevf_tx_buffer *first;
4118 u16 count = TXD_USE_COUNT(skb_headlen(skb));
4119 struct ixgbevf_ipsec_tx_data ipsec_tx = { 0 };
4120 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4124 u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
4126 if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
4127 dev_kfree_skb_any(skb);
4128 return NETDEV_TX_OK;
4131 /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
4132 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
4133 * + 2 desc gap to keep tail from touching head,
4134 * + 1 desc for context descriptor,
4135 * otherwise try next time
4137 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4138 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4139 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4141 count += skb_shinfo(skb)->nr_frags;
4143 if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
4144 tx_ring->tx_stats.tx_busy++;
4145 return NETDEV_TX_BUSY;
4148 /* record the location of the first descriptor for this packet */
4149 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4151 first->bytecount = skb->len;
4152 first->gso_segs = 1;
4154 if (skb_vlan_tag_present(skb)) {
4155 tx_flags |= skb_vlan_tag_get(skb);
4156 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4157 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4160 /* record initial flags and protocol */
4161 first->tx_flags = tx_flags;
4162 first->protocol = vlan_get_protocol(skb);
4164 #ifdef CONFIG_IXGBEVF_IPSEC
4165 if (xfrm_offload(skb) && !ixgbevf_ipsec_tx(tx_ring, first, &ipsec_tx))
4168 tso = ixgbevf_tso(tx_ring, first, &hdr_len, &ipsec_tx);
4172 ixgbevf_tx_csum(tx_ring, first, &ipsec_tx);
4174 ixgbevf_tx_map(tx_ring, first, hdr_len);
4176 ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
4178 return NETDEV_TX_OK;
4181 dev_kfree_skb_any(first->skb);
4184 return NETDEV_TX_OK;
4187 static netdev_tx_t ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4189 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4190 struct ixgbevf_ring *tx_ring;
4192 if (skb->len <= 0) {
4193 dev_kfree_skb_any(skb);
4194 return NETDEV_TX_OK;
4197 /* The minimum packet size for olinfo paylen is 17 so pad the skb
4198 * in order to meet this minimum size requirement.
4200 if (skb->len < 17) {
4201 if (skb_padto(skb, 17))
4202 return NETDEV_TX_OK;
4206 tx_ring = adapter->tx_ring[skb->queue_mapping];
4207 return ixgbevf_xmit_frame_ring(skb, tx_ring);
4211 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
4212 * @netdev: network interface device structure
4213 * @p: pointer to an address structure
4215 * Returns 0 on success, negative on failure
4217 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
4219 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4220 struct ixgbe_hw *hw = &adapter->hw;
4221 struct sockaddr *addr = p;
4224 if (!is_valid_ether_addr(addr->sa_data))
4225 return -EADDRNOTAVAIL;
4227 spin_lock_bh(&adapter->mbx_lock);
4229 err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
4231 spin_unlock_bh(&adapter->mbx_lock);
4236 ether_addr_copy(hw->mac.addr, addr->sa_data);
4237 ether_addr_copy(hw->mac.perm_addr, addr->sa_data);
4238 ether_addr_copy(netdev->dev_addr, addr->sa_data);
4244 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
4245 * @netdev: network interface device structure
4246 * @new_mtu: new value for maximum frame size
4248 * Returns 0 on success, negative on failure
4250 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
4252 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4253 struct ixgbe_hw *hw = &adapter->hw;
4254 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4257 /* prevent MTU being changed to a size unsupported by XDP */
4258 if (adapter->xdp_prog) {
4259 dev_warn(&adapter->pdev->dev, "MTU cannot be changed while XDP program is loaded\n");
4263 spin_lock_bh(&adapter->mbx_lock);
4264 /* notify the PF of our intent to use this size of frame */
4265 ret = hw->mac.ops.set_rlpml(hw, max_frame);
4266 spin_unlock_bh(&adapter->mbx_lock);
4270 hw_dbg(hw, "changing MTU from %d to %d\n",
4271 netdev->mtu, new_mtu);
4273 /* must set new MTU before calling down or up */
4274 netdev->mtu = new_mtu;
4276 if (netif_running(netdev))
4277 ixgbevf_reinit_locked(adapter);
4282 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
4284 struct net_device *netdev = pci_get_drvdata(pdev);
4285 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4291 netif_device_detach(netdev);
4293 if (netif_running(netdev))
4294 ixgbevf_close_suspend(adapter);
4296 ixgbevf_clear_interrupt_scheme(adapter);
4300 retval = pci_save_state(pdev);
4305 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4306 pci_disable_device(pdev);
4312 static int ixgbevf_resume(struct pci_dev *pdev)
4314 struct net_device *netdev = pci_get_drvdata(pdev);
4315 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4318 pci_restore_state(pdev);
4319 /* pci_restore_state clears dev->state_saved so call
4320 * pci_save_state to restore it.
4322 pci_save_state(pdev);
4324 err = pci_enable_device_mem(pdev);
4326 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
4330 adapter->hw.hw_addr = adapter->io_addr;
4331 smp_mb__before_atomic();
4332 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4333 pci_set_master(pdev);
4335 ixgbevf_reset(adapter);
4338 err = ixgbevf_init_interrupt_scheme(adapter);
4339 if (!err && netif_running(netdev))
4340 err = ixgbevf_open(netdev);
4345 netif_device_attach(netdev);
4350 #endif /* CONFIG_PM */
4351 static void ixgbevf_shutdown(struct pci_dev *pdev)
4353 ixgbevf_suspend(pdev, PMSG_SUSPEND);
4356 static void ixgbevf_get_tx_ring_stats(struct rtnl_link_stats64 *stats,
4357 const struct ixgbevf_ring *ring)
4364 start = u64_stats_fetch_begin_irq(&ring->syncp);
4365 bytes = ring->stats.bytes;
4366 packets = ring->stats.packets;
4367 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4368 stats->tx_bytes += bytes;
4369 stats->tx_packets += packets;
4373 static void ixgbevf_get_stats(struct net_device *netdev,
4374 struct rtnl_link_stats64 *stats)
4376 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4379 const struct ixgbevf_ring *ring;
4382 ixgbevf_update_stats(adapter);
4384 stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
4387 for (i = 0; i < adapter->num_rx_queues; i++) {
4388 ring = adapter->rx_ring[i];
4390 start = u64_stats_fetch_begin_irq(&ring->syncp);
4391 bytes = ring->stats.bytes;
4392 packets = ring->stats.packets;
4393 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4394 stats->rx_bytes += bytes;
4395 stats->rx_packets += packets;
4398 for (i = 0; i < adapter->num_tx_queues; i++) {
4399 ring = adapter->tx_ring[i];
4400 ixgbevf_get_tx_ring_stats(stats, ring);
4403 for (i = 0; i < adapter->num_xdp_queues; i++) {
4404 ring = adapter->xdp_ring[i];
4405 ixgbevf_get_tx_ring_stats(stats, ring);
4410 #define IXGBEVF_MAX_MAC_HDR_LEN 127
4411 #define IXGBEVF_MAX_NETWORK_HDR_LEN 511
4413 static netdev_features_t
4414 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
4415 netdev_features_t features)
4417 unsigned int network_hdr_len, mac_hdr_len;
4419 /* Make certain the headers can be described by a context descriptor */
4420 mac_hdr_len = skb_network_header(skb) - skb->data;
4421 if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
4422 return features & ~(NETIF_F_HW_CSUM |
4424 NETIF_F_HW_VLAN_CTAG_TX |
4428 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
4429 if (unlikely(network_hdr_len > IXGBEVF_MAX_NETWORK_HDR_LEN))
4430 return features & ~(NETIF_F_HW_CSUM |
4435 /* We can only support IPV4 TSO in tunnels if we can mangle the
4436 * inner IP ID field, so strip TSO if MANGLEID is not supported.
4438 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
4439 features &= ~NETIF_F_TSO;
4444 static int ixgbevf_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
4446 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4447 struct ixgbevf_adapter *adapter = netdev_priv(dev);
4448 struct bpf_prog *old_prog;
4450 /* verify ixgbevf ring attributes are sufficient for XDP */
4451 for (i = 0; i < adapter->num_rx_queues; i++) {
4452 struct ixgbevf_ring *ring = adapter->rx_ring[i];
4454 if (frame_size > ixgbevf_rx_bufsz(ring))
4458 old_prog = xchg(&adapter->xdp_prog, prog);
4460 /* If transitioning XDP modes reconfigure rings */
4461 if (!!prog != !!old_prog) {
4462 /* Hardware has to reinitialize queues and interrupts to
4463 * match packet buffer alignment. Unfortunately, the
4464 * hardware is not flexible enough to do this dynamically.
4466 if (netif_running(dev))
4469 ixgbevf_clear_interrupt_scheme(adapter);
4470 ixgbevf_init_interrupt_scheme(adapter);
4472 if (netif_running(dev))
4475 for (i = 0; i < adapter->num_rx_queues; i++)
4476 xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
4480 bpf_prog_put(old_prog);
4485 static int ixgbevf_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4487 struct ixgbevf_adapter *adapter = netdev_priv(dev);
4489 switch (xdp->command) {
4490 case XDP_SETUP_PROG:
4491 return ixgbevf_xdp_setup(dev, xdp->prog);
4492 case XDP_QUERY_PROG:
4493 xdp->prog_id = adapter->xdp_prog ?
4494 adapter->xdp_prog->aux->id : 0;
4501 static const struct net_device_ops ixgbevf_netdev_ops = {
4502 .ndo_open = ixgbevf_open,
4503 .ndo_stop = ixgbevf_close,
4504 .ndo_start_xmit = ixgbevf_xmit_frame,
4505 .ndo_set_rx_mode = ixgbevf_set_rx_mode,
4506 .ndo_get_stats64 = ixgbevf_get_stats,
4507 .ndo_validate_addr = eth_validate_addr,
4508 .ndo_set_mac_address = ixgbevf_set_mac,
4509 .ndo_change_mtu = ixgbevf_change_mtu,
4510 .ndo_tx_timeout = ixgbevf_tx_timeout,
4511 .ndo_vlan_rx_add_vid = ixgbevf_vlan_rx_add_vid,
4512 .ndo_vlan_rx_kill_vid = ixgbevf_vlan_rx_kill_vid,
4513 .ndo_features_check = ixgbevf_features_check,
4514 .ndo_bpf = ixgbevf_xdp,
4517 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
4519 dev->netdev_ops = &ixgbevf_netdev_ops;
4520 ixgbevf_set_ethtool_ops(dev);
4521 dev->watchdog_timeo = 5 * HZ;
4525 * ixgbevf_probe - Device Initialization Routine
4526 * @pdev: PCI device information struct
4527 * @ent: entry in ixgbevf_pci_tbl
4529 * Returns 0 on success, negative on failure
4531 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
4532 * The OS initialization, configuring of the adapter private structure,
4533 * and a hardware reset occur.
4535 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4537 struct net_device *netdev;
4538 struct ixgbevf_adapter *adapter = NULL;
4539 struct ixgbe_hw *hw = NULL;
4540 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
4541 int err, pci_using_dac;
4542 bool disable_dev = false;
4544 err = pci_enable_device(pdev);
4548 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
4551 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
4553 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4559 err = pci_request_regions(pdev, ixgbevf_driver_name);
4561 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4565 pci_set_master(pdev);
4567 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4571 goto err_alloc_etherdev;
4574 SET_NETDEV_DEV(netdev, &pdev->dev);
4576 adapter = netdev_priv(netdev);
4578 adapter->netdev = netdev;
4579 adapter->pdev = pdev;
4582 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4584 /* call save state here in standalone driver because it relies on
4585 * adapter struct to exist, and needs to call netdev_priv
4587 pci_save_state(pdev);
4589 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4590 pci_resource_len(pdev, 0));
4591 adapter->io_addr = hw->hw_addr;
4597 ixgbevf_assign_netdev_ops(netdev);
4600 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4601 hw->mac.type = ii->mac;
4603 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4604 sizeof(struct ixgbe_mbx_operations));
4606 /* setup the private structure */
4607 err = ixgbevf_sw_init(adapter);
4611 /* The HW MAC address was set and/or determined in sw_init */
4612 if (!is_valid_ether_addr(netdev->dev_addr)) {
4613 pr_err("invalid MAC address\n");
4618 netdev->hw_features = NETIF_F_SG |
4625 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4626 NETIF_F_GSO_GRE_CSUM | \
4627 NETIF_F_GSO_IPXIP4 | \
4628 NETIF_F_GSO_IPXIP6 | \
4629 NETIF_F_GSO_UDP_TUNNEL | \
4630 NETIF_F_GSO_UDP_TUNNEL_CSUM)
4632 netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4633 netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4634 IXGBEVF_GSO_PARTIAL_FEATURES;
4636 netdev->features = netdev->hw_features;
4639 netdev->features |= NETIF_F_HIGHDMA;
4641 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4642 netdev->mpls_features |= NETIF_F_SG |
4646 netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES;
4647 netdev->hw_enc_features |= netdev->vlan_features;
4649 /* set this bit last since it cannot be part of vlan_features */
4650 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4651 NETIF_F_HW_VLAN_CTAG_RX |
4652 NETIF_F_HW_VLAN_CTAG_TX;
4654 netdev->priv_flags |= IFF_UNICAST_FLT;
4656 /* MTU range: 68 - 1504 or 9710 */
4657 netdev->min_mtu = ETH_MIN_MTU;
4658 switch (adapter->hw.api_version) {
4659 case ixgbe_mbox_api_11:
4660 case ixgbe_mbox_api_12:
4661 case ixgbe_mbox_api_13:
4662 case ixgbe_mbox_api_14:
4663 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4664 (ETH_HLEN + ETH_FCS_LEN);
4667 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4668 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4669 (ETH_HLEN + ETH_FCS_LEN);
4671 netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4675 if (IXGBE_REMOVED(hw->hw_addr)) {
4680 timer_setup(&adapter->service_timer, ixgbevf_service_timer, 0);
4682 INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4683 set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4684 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4686 err = ixgbevf_init_interrupt_scheme(adapter);
4690 strcpy(netdev->name, "eth%d");
4692 err = register_netdev(netdev);
4696 pci_set_drvdata(pdev, netdev);
4697 netif_carrier_off(netdev);
4698 ixgbevf_init_ipsec_offload(adapter);
4700 ixgbevf_init_last_counter_stats(adapter);
4702 /* print the VF info */
4703 dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4704 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4706 switch (hw->mac.type) {
4707 case ixgbe_mac_X550_vf:
4708 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4710 case ixgbe_mac_X540_vf:
4711 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4713 case ixgbe_mac_82599_vf:
4715 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4722 ixgbevf_clear_interrupt_scheme(adapter);
4724 ixgbevf_reset_interrupt_capability(adapter);
4725 iounmap(adapter->io_addr);
4726 kfree(adapter->rss_key);
4728 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4729 free_netdev(netdev);
4731 pci_release_regions(pdev);
4734 if (!adapter || disable_dev)
4735 pci_disable_device(pdev);
4740 * ixgbevf_remove - Device Removal Routine
4741 * @pdev: PCI device information struct
4743 * ixgbevf_remove is called by the PCI subsystem to alert the driver
4744 * that it should release a PCI device. The could be caused by a
4745 * Hot-Plug event, or because the driver is going to be removed from
4748 static void ixgbevf_remove(struct pci_dev *pdev)
4750 struct net_device *netdev = pci_get_drvdata(pdev);
4751 struct ixgbevf_adapter *adapter;
4757 adapter = netdev_priv(netdev);
4759 set_bit(__IXGBEVF_REMOVING, &adapter->state);
4760 cancel_work_sync(&adapter->service_task);
4762 if (netdev->reg_state == NETREG_REGISTERED)
4763 unregister_netdev(netdev);
4765 ixgbevf_stop_ipsec_offload(adapter);
4766 ixgbevf_clear_interrupt_scheme(adapter);
4767 ixgbevf_reset_interrupt_capability(adapter);
4769 iounmap(adapter->io_addr);
4770 pci_release_regions(pdev);
4772 hw_dbg(&adapter->hw, "Remove complete\n");
4774 kfree(adapter->rss_key);
4775 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4776 free_netdev(netdev);
4779 pci_disable_device(pdev);
4783 * ixgbevf_io_error_detected - called when PCI error is detected
4784 * @pdev: Pointer to PCI device
4785 * @state: The current pci connection state
4787 * This function is called after a PCI bus error affecting
4788 * this device has been detected.
4790 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4791 pci_channel_state_t state)
4793 struct net_device *netdev = pci_get_drvdata(pdev);
4794 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4796 if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4797 return PCI_ERS_RESULT_DISCONNECT;
4800 netif_device_detach(netdev);
4802 if (netif_running(netdev))
4803 ixgbevf_close_suspend(adapter);
4805 if (state == pci_channel_io_perm_failure) {
4807 return PCI_ERS_RESULT_DISCONNECT;
4810 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4811 pci_disable_device(pdev);
4814 /* Request a slot slot reset. */
4815 return PCI_ERS_RESULT_NEED_RESET;
4819 * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4820 * @pdev: Pointer to PCI device
4822 * Restart the card from scratch, as if from a cold-boot. Implementation
4823 * resembles the first-half of the ixgbevf_resume routine.
4825 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4827 struct net_device *netdev = pci_get_drvdata(pdev);
4828 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4830 if (pci_enable_device_mem(pdev)) {
4832 "Cannot re-enable PCI device after reset.\n");
4833 return PCI_ERS_RESULT_DISCONNECT;
4836 adapter->hw.hw_addr = adapter->io_addr;
4837 smp_mb__before_atomic();
4838 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4839 pci_set_master(pdev);
4841 ixgbevf_reset(adapter);
4843 return PCI_ERS_RESULT_RECOVERED;
4847 * ixgbevf_io_resume - called when traffic can start flowing again.
4848 * @pdev: Pointer to PCI device
4850 * This callback is called when the error recovery driver tells us that
4851 * its OK to resume normal operation. Implementation resembles the
4852 * second-half of the ixgbevf_resume routine.
4854 static void ixgbevf_io_resume(struct pci_dev *pdev)
4856 struct net_device *netdev = pci_get_drvdata(pdev);
4859 if (netif_running(netdev))
4860 ixgbevf_open(netdev);
4862 netif_device_attach(netdev);
4866 /* PCI Error Recovery (ERS) */
4867 static const struct pci_error_handlers ixgbevf_err_handler = {
4868 .error_detected = ixgbevf_io_error_detected,
4869 .slot_reset = ixgbevf_io_slot_reset,
4870 .resume = ixgbevf_io_resume,
4873 static struct pci_driver ixgbevf_driver = {
4874 .name = ixgbevf_driver_name,
4875 .id_table = ixgbevf_pci_tbl,
4876 .probe = ixgbevf_probe,
4877 .remove = ixgbevf_remove,
4879 /* Power Management Hooks */
4880 .suspend = ixgbevf_suspend,
4881 .resume = ixgbevf_resume,
4883 .shutdown = ixgbevf_shutdown,
4884 .err_handler = &ixgbevf_err_handler
4888 * ixgbevf_init_module - Driver Registration Routine
4890 * ixgbevf_init_module is the first routine called when the driver is
4891 * loaded. All it does is register with the PCI subsystem.
4893 static int __init ixgbevf_init_module(void)
4895 pr_info("%s - version %s\n", ixgbevf_driver_string,
4896 ixgbevf_driver_version);
4898 pr_info("%s\n", ixgbevf_copyright);
4899 ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4901 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4905 return pci_register_driver(&ixgbevf_driver);
4908 module_init(ixgbevf_init_module);
4911 * ixgbevf_exit_module - Driver Exit Cleanup Routine
4913 * ixgbevf_exit_module is called just before the driver is removed
4916 static void __exit ixgbevf_exit_module(void)
4918 pci_unregister_driver(&ixgbevf_driver);
4920 destroy_workqueue(ixgbevf_wq);
4927 * ixgbevf_get_hw_dev_name - return device name string
4928 * used by hardware layer to print debugging information
4929 * @hw: pointer to private hardware struct
4931 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4933 struct ixgbevf_adapter *adapter = hw->back;
4935 return adapter->netdev->name;
4939 module_exit(ixgbevf_exit_module);
4941 /* ixgbevf_main.c */