1 /*******************************************************************************
3 Intel 82599 Virtual Function driver
4 Copyright(c) 1999 - 2014 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
29 /******************************************************************************
30 Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
31 ******************************************************************************/
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35 #include <linux/types.h>
36 #include <linux/bitops.h>
37 #include <linux/module.h>
38 #include <linux/pci.h>
39 #include <linux/netdevice.h>
40 #include <linux/vmalloc.h>
41 #include <linux/string.h>
44 #include <linux/tcp.h>
45 #include <linux/sctp.h>
46 #include <linux/ipv6.h>
47 #include <linux/slab.h>
48 #include <net/checksum.h>
49 #include <net/ip6_checksum.h>
50 #include <linux/ethtool.h>
52 #include <linux/if_vlan.h>
53 #include <linux/prefetch.h>
57 const char ixgbevf_driver_name[] = "ixgbevf";
58 static const char ixgbevf_driver_string[] =
59 "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
61 #define DRV_VERSION "2.12.1-k"
62 const char ixgbevf_driver_version[] = DRV_VERSION;
63 static char ixgbevf_copyright[] =
64 "Copyright (c) 2009 - 2012 Intel Corporation.";
66 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
67 [board_82599_vf] = &ixgbevf_82599_vf_info,
68 [board_X540_vf] = &ixgbevf_X540_vf_info,
71 /* ixgbevf_pci_tbl - PCI Device ID Table
73 * Wildcard entries (PCI_ANY_ID) should come last
74 * Last entry must be all 0s
76 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
77 * Class, Class Mask, private data (not used) }
79 static const struct pci_device_id ixgbevf_pci_tbl[] = {
80 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
81 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
82 /* required last entry */
85 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
87 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
88 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
89 MODULE_LICENSE("GPL");
90 MODULE_VERSION(DRV_VERSION);
92 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
93 static int debug = -1;
94 module_param(debug, int, 0);
95 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
98 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
99 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
100 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
102 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
104 struct ixgbevf_adapter *adapter = hw->back;
109 dev_err(&adapter->pdev->dev, "Adapter removed\n");
110 if (test_bit(__IXGBEVF_WORK_INIT, &adapter->state))
111 schedule_work(&adapter->watchdog_task);
114 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
118 /* The following check not only optimizes a bit by not
119 * performing a read on the status register when the
120 * register just read was a status register read that
121 * returned IXGBE_FAILED_READ_REG. It also blocks any
122 * potential recursion.
124 if (reg == IXGBE_VFSTATUS) {
125 ixgbevf_remove_adapter(hw);
128 value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
129 if (value == IXGBE_FAILED_READ_REG)
130 ixgbevf_remove_adapter(hw);
133 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
135 u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
138 if (IXGBE_REMOVED(reg_addr))
139 return IXGBE_FAILED_READ_REG;
140 value = readl(reg_addr + reg);
141 if (unlikely(value == IXGBE_FAILED_READ_REG))
142 ixgbevf_check_remove(hw, reg);
147 * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
148 * @adapter: pointer to adapter struct
149 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
150 * @queue: queue to map the corresponding interrupt to
151 * @msix_vector: the vector to map to the corresponding queue
153 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
154 u8 queue, u8 msix_vector)
157 struct ixgbe_hw *hw = &adapter->hw;
158 if (direction == -1) {
160 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
161 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
164 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
166 /* tx or rx causes */
167 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
168 index = ((16 * (queue & 1)) + (8 * direction));
169 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
170 ivar &= ~(0xFF << index);
171 ivar |= (msix_vector << index);
172 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
176 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
177 struct ixgbevf_tx_buffer *tx_buffer)
179 if (tx_buffer->skb) {
180 dev_kfree_skb_any(tx_buffer->skb);
181 if (dma_unmap_len(tx_buffer, len))
182 dma_unmap_single(tx_ring->dev,
183 dma_unmap_addr(tx_buffer, dma),
184 dma_unmap_len(tx_buffer, len),
186 } else if (dma_unmap_len(tx_buffer, len)) {
187 dma_unmap_page(tx_ring->dev,
188 dma_unmap_addr(tx_buffer, dma),
189 dma_unmap_len(tx_buffer, len),
192 tx_buffer->next_to_watch = NULL;
193 tx_buffer->skb = NULL;
194 dma_unmap_len_set(tx_buffer, len, 0);
195 /* tx_buffer must be completely set up in the transmit path */
198 #define IXGBE_MAX_TXD_PWR 14
199 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
201 /* Tx Descriptors needed, worst case */
202 #define TXD_USE_COUNT(S) DIV_ROUND_UP((S), IXGBE_MAX_DATA_PER_TXD)
203 #define DESC_NEEDED (MAX_SKB_FRAGS + 4)
205 static void ixgbevf_tx_timeout(struct net_device *netdev);
208 * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
209 * @q_vector: board private structure
210 * @tx_ring: tx ring to clean
212 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
213 struct ixgbevf_ring *tx_ring)
215 struct ixgbevf_adapter *adapter = q_vector->adapter;
216 struct ixgbevf_tx_buffer *tx_buffer;
217 union ixgbe_adv_tx_desc *tx_desc;
218 unsigned int total_bytes = 0, total_packets = 0;
219 unsigned int budget = tx_ring->count / 2;
220 unsigned int i = tx_ring->next_to_clean;
222 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
225 tx_buffer = &tx_ring->tx_buffer_info[i];
226 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
230 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
232 /* if next_to_watch is not set then there is no work pending */
236 /* prevent any other reads prior to eop_desc */
237 read_barrier_depends();
239 /* if DD is not set pending work has not been completed */
240 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
243 /* clear next_to_watch to prevent false hangs */
244 tx_buffer->next_to_watch = NULL;
246 /* update the statistics for this packet */
247 total_bytes += tx_buffer->bytecount;
248 total_packets += tx_buffer->gso_segs;
251 dev_kfree_skb_any(tx_buffer->skb);
253 /* unmap skb header data */
254 dma_unmap_single(tx_ring->dev,
255 dma_unmap_addr(tx_buffer, dma),
256 dma_unmap_len(tx_buffer, len),
259 /* clear tx_buffer data */
260 tx_buffer->skb = NULL;
261 dma_unmap_len_set(tx_buffer, len, 0);
263 /* unmap remaining buffers */
264 while (tx_desc != eop_desc) {
270 tx_buffer = tx_ring->tx_buffer_info;
271 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
274 /* unmap any remaining paged data */
275 if (dma_unmap_len(tx_buffer, len)) {
276 dma_unmap_page(tx_ring->dev,
277 dma_unmap_addr(tx_buffer, dma),
278 dma_unmap_len(tx_buffer, len),
280 dma_unmap_len_set(tx_buffer, len, 0);
284 /* move us one more past the eop_desc for start of next pkt */
290 tx_buffer = tx_ring->tx_buffer_info;
291 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
294 /* issue prefetch for next Tx descriptor */
297 /* update budget accounting */
299 } while (likely(budget));
302 tx_ring->next_to_clean = i;
303 u64_stats_update_begin(&tx_ring->syncp);
304 tx_ring->stats.bytes += total_bytes;
305 tx_ring->stats.packets += total_packets;
306 u64_stats_update_end(&tx_ring->syncp);
307 q_vector->tx.total_bytes += total_bytes;
308 q_vector->tx.total_packets += total_packets;
310 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
311 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
312 (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
313 /* Make sure that anybody stopping the queue after this
314 * sees the new next_to_clean.
318 if (__netif_subqueue_stopped(tx_ring->netdev,
319 tx_ring->queue_index) &&
320 !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
321 netif_wake_subqueue(tx_ring->netdev,
322 tx_ring->queue_index);
323 ++tx_ring->tx_stats.restart_queue;
331 * ixgbevf_rx_skb - Helper function to determine proper Rx method
332 * @q_vector: structure containing interrupt and ring information
333 * @skb: packet to send up
335 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
338 #ifdef CONFIG_NET_RX_BUSY_POLL
339 skb_mark_napi_id(skb, &q_vector->napi);
341 if (ixgbevf_qv_busy_polling(q_vector)) {
342 netif_receive_skb(skb);
343 /* exit early if we busy polled */
346 #endif /* CONFIG_NET_RX_BUSY_POLL */
348 napi_gro_receive(&q_vector->napi, skb);
351 /* ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
352 * @ring: structure containig ring specific data
353 * @rx_desc: current Rx descriptor being processed
354 * @skb: skb currently being received and modified
356 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
357 union ixgbe_adv_rx_desc *rx_desc,
360 skb_checksum_none_assert(skb);
362 /* Rx csum disabled */
363 if (!(ring->netdev->features & NETIF_F_RXCSUM))
366 /* if IP and error */
367 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
368 ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
369 ring->rx_stats.csum_err++;
373 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
376 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
377 ring->rx_stats.csum_err++;
381 /* It must be a TCP or UDP packet with a valid checksum */
382 skb->ip_summed = CHECKSUM_UNNECESSARY;
385 /* ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
386 * @rx_ring: rx descriptor ring packet is being transacted on
387 * @rx_desc: pointer to the EOP Rx descriptor
388 * @skb: pointer to current skb being populated
390 * This function checks the ring, descriptor, and packet information in
391 * order to populate the checksum, VLAN, protocol, and other fields within
394 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
395 union ixgbe_adv_rx_desc *rx_desc,
398 ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
400 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
401 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
402 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
404 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
405 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
408 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
412 * ixgbevf_is_non_eop - process handling of non-EOP buffers
413 * @rx_ring: Rx ring being processed
414 * @rx_desc: Rx descriptor for current buffer
415 * @skb: current socket buffer containing buffer in progress
417 * This function updates next to clean. If the buffer is an EOP buffer
418 * this function exits returning false, otherwise it will place the
419 * sk_buff in the next buffer to be chained and return true indicating
420 * that this is in fact a non-EOP buffer.
422 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
423 union ixgbe_adv_rx_desc *rx_desc)
425 u32 ntc = rx_ring->next_to_clean + 1;
427 /* fetch, update, and store next to clean */
428 ntc = (ntc < rx_ring->count) ? ntc : 0;
429 rx_ring->next_to_clean = ntc;
431 prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
433 if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
439 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
440 struct ixgbevf_rx_buffer *bi)
442 struct page *page = bi->page;
443 dma_addr_t dma = bi->dma;
445 /* since we are recycling buffers we should seldom need to alloc */
449 /* alloc new page for storage */
450 page = dev_alloc_page();
451 if (unlikely(!page)) {
452 rx_ring->rx_stats.alloc_rx_page_failed++;
456 /* map page for use */
457 dma = dma_map_page(rx_ring->dev, page, 0,
458 PAGE_SIZE, DMA_FROM_DEVICE);
460 /* if mapping failed free memory back to system since
461 * there isn't much point in holding memory we can't use
463 if (dma_mapping_error(rx_ring->dev, dma)) {
466 rx_ring->rx_stats.alloc_rx_buff_failed++;
478 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
479 * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
480 * @cleaned_count: number of buffers to replace
482 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
485 union ixgbe_adv_rx_desc *rx_desc;
486 struct ixgbevf_rx_buffer *bi;
487 unsigned int i = rx_ring->next_to_use;
489 /* nothing to do or no valid netdev defined */
490 if (!cleaned_count || !rx_ring->netdev)
493 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
494 bi = &rx_ring->rx_buffer_info[i];
498 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
501 /* Refresh the desc even if pkt_addr didn't change
502 * because each write-back erases this info.
504 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
510 rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
511 bi = rx_ring->rx_buffer_info;
515 /* clear the hdr_addr for the next_to_use descriptor */
516 rx_desc->read.hdr_addr = 0;
519 } while (cleaned_count);
523 if (rx_ring->next_to_use != i) {
524 /* record the next descriptor to use */
525 rx_ring->next_to_use = i;
527 /* update next to alloc since we have filled the ring */
528 rx_ring->next_to_alloc = i;
530 /* Force memory writes to complete before letting h/w
531 * know there are new descriptors to fetch. (Only
532 * applicable for weak-ordered memory model archs,
536 ixgbevf_write_tail(rx_ring, i);
540 /* ixgbevf_pull_tail - ixgbevf specific version of skb_pull_tail
541 * @rx_ring: rx descriptor ring packet is being transacted on
542 * @skb: pointer to current skb being adjusted
544 * This function is an ixgbevf specific version of __pskb_pull_tail. The
545 * main difference between this version and the original function is that
546 * this function can make several assumptions about the state of things
547 * that allow for significant optimizations versus the standard function.
548 * As a result we can do things like drop a frag and maintain an accurate
549 * truesize for the skb.
551 static void ixgbevf_pull_tail(struct ixgbevf_ring *rx_ring,
554 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
556 unsigned int pull_len;
558 /* it is valid to use page_address instead of kmap since we are
559 * working with pages allocated out of the lomem pool per
560 * alloc_page(GFP_ATOMIC)
562 va = skb_frag_address(frag);
564 /* we need the header to contain the greater of either ETH_HLEN or
565 * 60 bytes if the skb->len is less than 60 for skb_pad.
567 pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
569 /* align pull length to size of long to optimize memcpy performance */
570 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
572 /* update all of the pointers */
573 skb_frag_size_sub(frag, pull_len);
574 frag->page_offset += pull_len;
575 skb->data_len -= pull_len;
576 skb->tail += pull_len;
579 /* ixgbevf_cleanup_headers - Correct corrupted or empty headers
580 * @rx_ring: rx descriptor ring packet is being transacted on
581 * @rx_desc: pointer to the EOP Rx descriptor
582 * @skb: pointer to current skb being fixed
584 * Check for corrupted packet headers caused by senders on the local L2
585 * embedded NIC switch not setting up their Tx Descriptors right. These
586 * should be very rare.
588 * Also address the case where we are pulling data in on pages only
589 * and as such no data is present in the skb header.
591 * In addition if skb is not at least 60 bytes we need to pad it so that
592 * it is large enough to qualify as a valid Ethernet frame.
594 * Returns true if an error was encountered and skb was freed.
596 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
597 union ixgbe_adv_rx_desc *rx_desc,
600 /* verify that the packet does not have any known errors */
601 if (unlikely(ixgbevf_test_staterr(rx_desc,
602 IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
603 struct net_device *netdev = rx_ring->netdev;
605 if (!(netdev->features & NETIF_F_RXALL)) {
606 dev_kfree_skb_any(skb);
611 /* place header in linear portion of buffer */
612 if (skb_is_nonlinear(skb))
613 ixgbevf_pull_tail(rx_ring, skb);
615 /* if skb_pad returns an error the skb was freed */
616 if (unlikely(skb->len < 60)) {
617 int pad_len = 60 - skb->len;
619 if (skb_pad(skb, pad_len))
621 __skb_put(skb, pad_len);
627 /* ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
628 * @rx_ring: rx descriptor ring to store buffers on
629 * @old_buff: donor buffer to have page reused
631 * Synchronizes page for reuse by the adapter
633 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
634 struct ixgbevf_rx_buffer *old_buff)
636 struct ixgbevf_rx_buffer *new_buff;
637 u16 nta = rx_ring->next_to_alloc;
639 new_buff = &rx_ring->rx_buffer_info[nta];
641 /* update, and store next to alloc */
643 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
645 /* transfer page from old buffer to new buffer */
646 new_buff->page = old_buff->page;
647 new_buff->dma = old_buff->dma;
648 new_buff->page_offset = old_buff->page_offset;
650 /* sync the buffer for use by the device */
651 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
652 new_buff->page_offset,
657 static inline bool ixgbevf_page_is_reserved(struct page *page)
659 return (page_to_nid(page) != numa_mem_id()) || page->pfmemalloc;
662 /* ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
663 * @rx_ring: rx descriptor ring to transact packets on
664 * @rx_buffer: buffer containing page to add
665 * @rx_desc: descriptor containing length of buffer written by hardware
666 * @skb: sk_buff to place the data into
668 * This function will add the data contained in rx_buffer->page to the skb.
669 * This is done either through a direct copy if the data in the buffer is
670 * less than the skb header size, otherwise it will just attach the page as
673 * The function will then update the page offset if necessary and return
674 * true if the buffer can be reused by the adapter.
676 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
677 struct ixgbevf_rx_buffer *rx_buffer,
678 union ixgbe_adv_rx_desc *rx_desc,
681 struct page *page = rx_buffer->page;
682 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
683 #if (PAGE_SIZE < 8192)
684 unsigned int truesize = IXGBEVF_RX_BUFSZ;
686 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
689 if ((size <= IXGBEVF_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
690 unsigned char *va = page_address(page) + rx_buffer->page_offset;
692 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
694 /* page is not reserved, we can reuse buffer as is */
695 if (likely(!ixgbevf_page_is_reserved(page)))
698 /* this page cannot be reused so discard it */
703 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
704 rx_buffer->page_offset, size, truesize);
706 /* avoid re-using remote pages */
707 if (unlikely(ixgbevf_page_is_reserved(page)))
710 #if (PAGE_SIZE < 8192)
711 /* if we are only owner of page we can reuse it */
712 if (unlikely(page_count(page) != 1))
715 /* flip page offset to other buffer */
716 rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
719 /* move offset up to the next cache line */
720 rx_buffer->page_offset += truesize;
722 if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
726 /* Even if we own the page, we are not allowed to use atomic_set()
727 * This would break get_page_unless_zero() users.
729 atomic_inc(&page->_count);
734 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
735 union ixgbe_adv_rx_desc *rx_desc,
738 struct ixgbevf_rx_buffer *rx_buffer;
741 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
742 page = rx_buffer->page;
746 void *page_addr = page_address(page) +
747 rx_buffer->page_offset;
749 /* prefetch first cache line of first page */
751 #if L1_CACHE_BYTES < 128
752 prefetch(page_addr + L1_CACHE_BYTES);
755 /* allocate a skb to store the frags */
756 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
757 IXGBEVF_RX_HDR_SIZE);
758 if (unlikely(!skb)) {
759 rx_ring->rx_stats.alloc_rx_buff_failed++;
763 /* we will be copying header into skb->data in
764 * pskb_may_pull so it is in our interest to prefetch
765 * it now to avoid a possible cache miss
767 prefetchw(skb->data);
770 /* we are reusing so sync this buffer for CPU use */
771 dma_sync_single_range_for_cpu(rx_ring->dev,
773 rx_buffer->page_offset,
777 /* pull page into skb */
778 if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
779 /* hand second half of page back to the ring */
780 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
782 /* we are not reusing the buffer so unmap it */
783 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
784 PAGE_SIZE, DMA_FROM_DEVICE);
787 /* clear contents of buffer_info */
789 rx_buffer->page = NULL;
794 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
797 struct ixgbe_hw *hw = &adapter->hw;
799 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
802 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
803 struct ixgbevf_ring *rx_ring,
806 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
807 u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
808 struct sk_buff *skb = rx_ring->skb;
810 while (likely(total_rx_packets < budget)) {
811 union ixgbe_adv_rx_desc *rx_desc;
813 /* return some buffers to hardware, one at a time is too slow */
814 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
815 ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
819 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
821 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
824 /* This memory barrier is needed to keep us from reading
825 * any other fields out of the rx_desc until we know the
826 * RXD_STAT_DD bit is set
830 /* retrieve a buffer from the ring */
831 skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
833 /* exit if we failed to retrieve a buffer */
839 /* fetch next buffer in frame if non-eop */
840 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
843 /* verify the packet layout is correct */
844 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
849 /* probably a little skewed due to removing CRC */
850 total_rx_bytes += skb->len;
852 /* Workaround hardware that can't do proper VEPA multicast
855 if ((skb->pkt_type == PACKET_BROADCAST ||
856 skb->pkt_type == PACKET_MULTICAST) &&
857 ether_addr_equal(rx_ring->netdev->dev_addr,
858 eth_hdr(skb)->h_source)) {
859 dev_kfree_skb_irq(skb);
863 /* populate checksum, VLAN, and protocol */
864 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
866 ixgbevf_rx_skb(q_vector, skb);
868 /* reset skb pointer */
871 /* update budget accounting */
875 /* place incomplete frames back on ring for completion */
878 u64_stats_update_begin(&rx_ring->syncp);
879 rx_ring->stats.packets += total_rx_packets;
880 rx_ring->stats.bytes += total_rx_bytes;
881 u64_stats_update_end(&rx_ring->syncp);
882 q_vector->rx.total_packets += total_rx_packets;
883 q_vector->rx.total_bytes += total_rx_bytes;
885 return total_rx_packets;
889 * ixgbevf_poll - NAPI polling calback
890 * @napi: napi struct with our devices info in it
891 * @budget: amount of work driver is allowed to do this pass, in packets
893 * This function will clean more than one or more rings associated with a
896 static int ixgbevf_poll(struct napi_struct *napi, int budget)
898 struct ixgbevf_q_vector *q_vector =
899 container_of(napi, struct ixgbevf_q_vector, napi);
900 struct ixgbevf_adapter *adapter = q_vector->adapter;
901 struct ixgbevf_ring *ring;
903 bool clean_complete = true;
905 ixgbevf_for_each_ring(ring, q_vector->tx)
906 clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
908 #ifdef CONFIG_NET_RX_BUSY_POLL
909 if (!ixgbevf_qv_lock_napi(q_vector))
913 /* attempt to distribute budget to each queue fairly, but don't allow
914 * the budget to go below 1 because we'll exit polling */
915 if (q_vector->rx.count > 1)
916 per_ring_budget = max(budget/q_vector->rx.count, 1);
918 per_ring_budget = budget;
920 ixgbevf_for_each_ring(ring, q_vector->rx)
921 clean_complete &= (ixgbevf_clean_rx_irq(q_vector, ring,
925 #ifdef CONFIG_NET_RX_BUSY_POLL
926 ixgbevf_qv_unlock_napi(q_vector);
929 /* If all work not completed, return budget and keep polling */
932 /* all work done, exit the polling mode */
934 if (adapter->rx_itr_setting & 1)
935 ixgbevf_set_itr(q_vector);
936 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
937 !test_bit(__IXGBEVF_REMOVING, &adapter->state))
938 ixgbevf_irq_enable_queues(adapter,
939 1 << q_vector->v_idx);
945 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
946 * @q_vector: structure containing interrupt and ring information
948 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
950 struct ixgbevf_adapter *adapter = q_vector->adapter;
951 struct ixgbe_hw *hw = &adapter->hw;
952 int v_idx = q_vector->v_idx;
953 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
956 * set the WDIS bit to not clear the timer bits and cause an
957 * immediate assertion of the interrupt
959 itr_reg |= IXGBE_EITR_CNT_WDIS;
961 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
964 #ifdef CONFIG_NET_RX_BUSY_POLL
965 /* must be called with local_bh_disable()d */
966 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
968 struct ixgbevf_q_vector *q_vector =
969 container_of(napi, struct ixgbevf_q_vector, napi);
970 struct ixgbevf_adapter *adapter = q_vector->adapter;
971 struct ixgbevf_ring *ring;
974 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
975 return LL_FLUSH_FAILED;
977 if (!ixgbevf_qv_lock_poll(q_vector))
978 return LL_FLUSH_BUSY;
980 ixgbevf_for_each_ring(ring, q_vector->rx) {
981 found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
982 #ifdef BP_EXTENDED_STATS
984 ring->stats.cleaned += found;
986 ring->stats.misses++;
992 ixgbevf_qv_unlock_poll(q_vector);
996 #endif /* CONFIG_NET_RX_BUSY_POLL */
999 * ixgbevf_configure_msix - Configure MSI-X hardware
1000 * @adapter: board private structure
1002 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1005 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1007 struct ixgbevf_q_vector *q_vector;
1008 int q_vectors, v_idx;
1010 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1011 adapter->eims_enable_mask = 0;
1014 * Populate the IVAR table and set the ITR values to the
1015 * corresponding register.
1017 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1018 struct ixgbevf_ring *ring;
1019 q_vector = adapter->q_vector[v_idx];
1021 ixgbevf_for_each_ring(ring, q_vector->rx)
1022 ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1024 ixgbevf_for_each_ring(ring, q_vector->tx)
1025 ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1027 if (q_vector->tx.ring && !q_vector->rx.ring) {
1028 /* tx only vector */
1029 if (adapter->tx_itr_setting == 1)
1030 q_vector->itr = IXGBE_10K_ITR;
1032 q_vector->itr = adapter->tx_itr_setting;
1034 /* rx or rx/tx vector */
1035 if (adapter->rx_itr_setting == 1)
1036 q_vector->itr = IXGBE_20K_ITR;
1038 q_vector->itr = adapter->rx_itr_setting;
1041 /* add q_vector eims value to global eims_enable_mask */
1042 adapter->eims_enable_mask |= 1 << v_idx;
1044 ixgbevf_write_eitr(q_vector);
1047 ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1048 /* setup eims_other and add value to global eims_enable_mask */
1049 adapter->eims_other = 1 << v_idx;
1050 adapter->eims_enable_mask |= adapter->eims_other;
1053 enum latency_range {
1057 latency_invalid = 255
1061 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1062 * @q_vector: structure containing interrupt and ring information
1063 * @ring_container: structure containing ring performance data
1065 * Stores a new ITR value based on packets and byte
1066 * counts during the last interrupt. The advantage of per interrupt
1067 * computation is faster updates and more accurate ITR for the current
1068 * traffic pattern. Constants in this function were computed
1069 * based on theoretical maximum wire speed and thresholds were set based
1070 * on testing data as well as attempting to minimize response time
1071 * while increasing bulk throughput.
1073 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1074 struct ixgbevf_ring_container *ring_container)
1076 int bytes = ring_container->total_bytes;
1077 int packets = ring_container->total_packets;
1080 u8 itr_setting = ring_container->itr;
1085 /* simple throttlerate management
1086 * 0-20MB/s lowest (100000 ints/s)
1087 * 20-100MB/s low (20000 ints/s)
1088 * 100-1249MB/s bulk (8000 ints/s)
1090 /* what was last interrupt timeslice? */
1091 timepassed_us = q_vector->itr >> 2;
1092 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1094 switch (itr_setting) {
1095 case lowest_latency:
1096 if (bytes_perint > 10)
1097 itr_setting = low_latency;
1100 if (bytes_perint > 20)
1101 itr_setting = bulk_latency;
1102 else if (bytes_perint <= 10)
1103 itr_setting = lowest_latency;
1106 if (bytes_perint <= 20)
1107 itr_setting = low_latency;
1111 /* clear work counters since we have the values we need */
1112 ring_container->total_bytes = 0;
1113 ring_container->total_packets = 0;
1115 /* write updated itr to ring container */
1116 ring_container->itr = itr_setting;
1119 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1121 u32 new_itr = q_vector->itr;
1124 ixgbevf_update_itr(q_vector, &q_vector->tx);
1125 ixgbevf_update_itr(q_vector, &q_vector->rx);
1127 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1129 switch (current_itr) {
1130 /* counts and packets in update_itr are dependent on these numbers */
1131 case lowest_latency:
1132 new_itr = IXGBE_100K_ITR;
1135 new_itr = IXGBE_20K_ITR;
1139 new_itr = IXGBE_8K_ITR;
1143 if (new_itr != q_vector->itr) {
1144 /* do an exponential smoothing */
1145 new_itr = (10 * new_itr * q_vector->itr) /
1146 ((9 * new_itr) + q_vector->itr);
1148 /* save the algorithm value here */
1149 q_vector->itr = new_itr;
1151 ixgbevf_write_eitr(q_vector);
1155 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1157 struct ixgbevf_adapter *adapter = data;
1158 struct ixgbe_hw *hw = &adapter->hw;
1160 hw->mac.get_link_status = 1;
1162 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1163 !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1164 mod_timer(&adapter->watchdog_timer, jiffies);
1166 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1172 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1174 * @data: pointer to our q_vector struct for this interrupt vector
1176 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1178 struct ixgbevf_q_vector *q_vector = data;
1180 /* EIAM disabled interrupts (on this vector) for us */
1181 if (q_vector->rx.ring || q_vector->tx.ring)
1182 napi_schedule(&q_vector->napi);
1187 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1190 struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1192 a->rx_ring[r_idx]->next = q_vector->rx.ring;
1193 q_vector->rx.ring = a->rx_ring[r_idx];
1194 q_vector->rx.count++;
1197 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1200 struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1202 a->tx_ring[t_idx]->next = q_vector->tx.ring;
1203 q_vector->tx.ring = a->tx_ring[t_idx];
1204 q_vector->tx.count++;
1208 * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1209 * @adapter: board private structure to initialize
1211 * This function maps descriptor rings to the queue-specific vectors
1212 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1213 * one vector per ring/queue, but on a constrained vector budget, we
1214 * group the rings as "efficiently" as possible. You would add new
1215 * mapping configurations in here.
1217 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1221 int rxr_idx = 0, txr_idx = 0;
1222 int rxr_remaining = adapter->num_rx_queues;
1223 int txr_remaining = adapter->num_tx_queues;
1228 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1231 * The ideal configuration...
1232 * We have enough vectors to map one per queue.
1234 if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1235 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1236 map_vector_to_rxq(adapter, v_start, rxr_idx);
1238 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1239 map_vector_to_txq(adapter, v_start, txr_idx);
1244 * If we don't have enough vectors for a 1-to-1
1245 * mapping, we'll have to group them so there are
1246 * multiple queues per vector.
1248 /* Re-adjusting *qpv takes care of the remainder. */
1249 for (i = v_start; i < q_vectors; i++) {
1250 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1251 for (j = 0; j < rqpv; j++) {
1252 map_vector_to_rxq(adapter, i, rxr_idx);
1257 for (i = v_start; i < q_vectors; i++) {
1258 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1259 for (j = 0; j < tqpv; j++) {
1260 map_vector_to_txq(adapter, i, txr_idx);
1271 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1272 * @adapter: board private structure
1274 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1275 * interrupts from the kernel.
1277 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1279 struct net_device *netdev = adapter->netdev;
1280 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1284 for (vector = 0; vector < q_vectors; vector++) {
1285 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1286 struct msix_entry *entry = &adapter->msix_entries[vector];
1288 if (q_vector->tx.ring && q_vector->rx.ring) {
1289 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1290 "%s-%s-%d", netdev->name, "TxRx", ri++);
1292 } else if (q_vector->rx.ring) {
1293 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1294 "%s-%s-%d", netdev->name, "rx", ri++);
1295 } else if (q_vector->tx.ring) {
1296 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1297 "%s-%s-%d", netdev->name, "tx", ti++);
1299 /* skip this unused q_vector */
1302 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1303 q_vector->name, q_vector);
1305 hw_dbg(&adapter->hw,
1306 "request_irq failed for MSIX interrupt "
1307 "Error: %d\n", err);
1308 goto free_queue_irqs;
1312 err = request_irq(adapter->msix_entries[vector].vector,
1313 &ixgbevf_msix_other, 0, netdev->name, adapter);
1315 hw_dbg(&adapter->hw,
1316 "request_irq for msix_other failed: %d\n", err);
1317 goto free_queue_irqs;
1325 free_irq(adapter->msix_entries[vector].vector,
1326 adapter->q_vector[vector]);
1328 /* This failure is non-recoverable - it indicates the system is
1329 * out of MSIX vector resources and the VF driver cannot run
1330 * without them. Set the number of msix vectors to zero
1331 * indicating that not enough can be allocated. The error
1332 * will be returned to the user indicating device open failed.
1333 * Any further attempts to force the driver to open will also
1334 * fail. The only way to recover is to unload the driver and
1335 * reload it again. If the system has recovered some MSIX
1336 * vectors then it may succeed.
1338 adapter->num_msix_vectors = 0;
1342 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1344 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1346 for (i = 0; i < q_vectors; i++) {
1347 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1348 q_vector->rx.ring = NULL;
1349 q_vector->tx.ring = NULL;
1350 q_vector->rx.count = 0;
1351 q_vector->tx.count = 0;
1356 * ixgbevf_request_irq - initialize interrupts
1357 * @adapter: board private structure
1359 * Attempts to configure interrupts using the best available
1360 * capabilities of the hardware and kernel.
1362 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1366 err = ixgbevf_request_msix_irqs(adapter);
1369 hw_dbg(&adapter->hw,
1370 "request_irq failed, Error %d\n", err);
1375 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1379 q_vectors = adapter->num_msix_vectors;
1382 free_irq(adapter->msix_entries[i].vector, adapter);
1385 for (; i >= 0; i--) {
1386 /* free only the irqs that were actually requested */
1387 if (!adapter->q_vector[i]->rx.ring &&
1388 !adapter->q_vector[i]->tx.ring)
1391 free_irq(adapter->msix_entries[i].vector,
1392 adapter->q_vector[i]);
1395 ixgbevf_reset_q_vectors(adapter);
1399 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1400 * @adapter: board private structure
1402 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1404 struct ixgbe_hw *hw = &adapter->hw;
1407 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1408 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1409 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1411 IXGBE_WRITE_FLUSH(hw);
1413 for (i = 0; i < adapter->num_msix_vectors; i++)
1414 synchronize_irq(adapter->msix_entries[i].vector);
1418 * ixgbevf_irq_enable - Enable default interrupt generation settings
1419 * @adapter: board private structure
1421 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1423 struct ixgbe_hw *hw = &adapter->hw;
1425 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1426 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1427 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1431 * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1432 * @adapter: board private structure
1433 * @ring: structure containing ring specific data
1435 * Configure the Tx descriptor ring after a reset.
1437 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1438 struct ixgbevf_ring *ring)
1440 struct ixgbe_hw *hw = &adapter->hw;
1441 u64 tdba = ring->dma;
1443 u32 txdctl = IXGBE_TXDCTL_ENABLE;
1444 u8 reg_idx = ring->reg_idx;
1446 /* disable queue to avoid issues while updating state */
1447 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1448 IXGBE_WRITE_FLUSH(hw);
1450 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1451 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1452 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1453 ring->count * sizeof(union ixgbe_adv_tx_desc));
1455 /* disable head writeback */
1456 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1457 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1459 /* enable relaxed ordering */
1460 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1461 (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1462 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1464 /* reset head and tail pointers */
1465 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1466 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1467 ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1469 /* reset ntu and ntc to place SW in sync with hardwdare */
1470 ring->next_to_clean = 0;
1471 ring->next_to_use = 0;
1473 /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1474 * to or less than the number of on chip descriptors, which is
1477 txdctl |= (8 << 16); /* WTHRESH = 8 */
1479 /* Setting PTHRESH to 32 both improves performance */
1480 txdctl |= (1 << 8) | /* HTHRESH = 1 */
1481 32; /* PTHRESH = 32 */
1483 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1485 /* poll to verify queue is enabled */
1487 usleep_range(1000, 2000);
1488 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1489 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1491 pr_err("Could not enable Tx Queue %d\n", reg_idx);
1495 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1496 * @adapter: board private structure
1498 * Configure the Tx unit of the MAC after a reset.
1500 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1504 /* Setup the HW Tx Head and Tail descriptor pointers */
1505 for (i = 0; i < adapter->num_tx_queues; i++)
1506 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1509 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1511 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1513 struct ixgbe_hw *hw = &adapter->hw;
1516 srrctl = IXGBE_SRRCTL_DROP_EN;
1518 srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1519 srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1520 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1522 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1525 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1527 struct ixgbe_hw *hw = &adapter->hw;
1529 /* PSRTYPE must be initialized in 82599 */
1530 u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1531 IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1532 IXGBE_PSRTYPE_L2HDR;
1534 if (adapter->num_rx_queues > 1)
1537 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1540 #define IXGBEVF_MAX_RX_DESC_POLL 10
1541 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1542 struct ixgbevf_ring *ring)
1544 struct ixgbe_hw *hw = &adapter->hw;
1545 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1547 u8 reg_idx = ring->reg_idx;
1549 if (IXGBE_REMOVED(hw->hw_addr))
1551 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1552 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1554 /* write value back with RXDCTL.ENABLE bit cleared */
1555 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1557 /* the hardware may take up to 100us to really disable the rx queue */
1560 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1561 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1564 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1568 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1569 struct ixgbevf_ring *ring)
1571 struct ixgbe_hw *hw = &adapter->hw;
1572 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1574 u8 reg_idx = ring->reg_idx;
1576 if (IXGBE_REMOVED(hw->hw_addr))
1579 usleep_range(1000, 2000);
1580 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1581 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1584 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1588 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1589 struct ixgbevf_ring *ring)
1591 struct ixgbe_hw *hw = &adapter->hw;
1592 u64 rdba = ring->dma;
1594 u8 reg_idx = ring->reg_idx;
1596 /* disable queue to avoid issues while updating state */
1597 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1598 ixgbevf_disable_rx_queue(adapter, ring);
1600 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1601 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1602 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1603 ring->count * sizeof(union ixgbe_adv_rx_desc));
1605 /* enable relaxed ordering */
1606 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1607 IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1609 /* reset head and tail pointers */
1610 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1611 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1612 ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1614 /* reset ntu and ntc to place SW in sync with hardwdare */
1615 ring->next_to_clean = 0;
1616 ring->next_to_use = 0;
1617 ring->next_to_alloc = 0;
1619 ixgbevf_configure_srrctl(adapter, reg_idx);
1621 /* allow any size packet since we can handle overflow */
1622 rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1624 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1625 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1627 ixgbevf_rx_desc_queue_enable(adapter, ring);
1628 ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1632 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1633 * @adapter: board private structure
1635 * Configure the Rx unit of the MAC after a reset.
1637 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1640 struct ixgbe_hw *hw = &adapter->hw;
1641 struct net_device *netdev = adapter->netdev;
1643 ixgbevf_setup_psrtype(adapter);
1645 /* notify the PF of our intent to use this size of frame */
1646 ixgbevf_rlpml_set_vf(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1648 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1649 * the Base and Length of the Rx Descriptor Ring */
1650 for (i = 0; i < adapter->num_rx_queues; i++)
1651 ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1654 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1655 __be16 proto, u16 vid)
1657 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1658 struct ixgbe_hw *hw = &adapter->hw;
1661 spin_lock_bh(&adapter->mbx_lock);
1663 /* add VID to filter table */
1664 err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1666 spin_unlock_bh(&adapter->mbx_lock);
1668 /* translate error return types so error makes sense */
1669 if (err == IXGBE_ERR_MBX)
1672 if (err == IXGBE_ERR_INVALID_ARGUMENT)
1675 set_bit(vid, adapter->active_vlans);
1680 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1681 __be16 proto, u16 vid)
1683 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1684 struct ixgbe_hw *hw = &adapter->hw;
1685 int err = -EOPNOTSUPP;
1687 spin_lock_bh(&adapter->mbx_lock);
1689 /* remove VID from filter table */
1690 err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1692 spin_unlock_bh(&adapter->mbx_lock);
1694 clear_bit(vid, adapter->active_vlans);
1699 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1703 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1704 ixgbevf_vlan_rx_add_vid(adapter->netdev,
1705 htons(ETH_P_8021Q), vid);
1708 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1710 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1711 struct ixgbe_hw *hw = &adapter->hw;
1714 if ((netdev_uc_count(netdev)) > 10) {
1715 pr_err("Too many unicast filters - No Space\n");
1719 if (!netdev_uc_empty(netdev)) {
1720 struct netdev_hw_addr *ha;
1721 netdev_for_each_uc_addr(ha, netdev) {
1722 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1727 * If the list is empty then send message to PF driver to
1728 * clear all macvlans on this VF.
1730 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1737 * ixgbevf_set_rx_mode - Multicast and unicast set
1738 * @netdev: network interface device structure
1740 * The set_rx_method entry point is called whenever the multicast address
1741 * list, unicast address list or the network interface flags are updated.
1742 * This routine is responsible for configuring the hardware for proper
1743 * multicast mode and configuring requested unicast filters.
1745 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1747 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1748 struct ixgbe_hw *hw = &adapter->hw;
1750 spin_lock_bh(&adapter->mbx_lock);
1752 /* reprogram multicast list */
1753 hw->mac.ops.update_mc_addr_list(hw, netdev);
1755 ixgbevf_write_uc_addr_list(netdev);
1757 spin_unlock_bh(&adapter->mbx_lock);
1760 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1763 struct ixgbevf_q_vector *q_vector;
1764 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1766 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1767 q_vector = adapter->q_vector[q_idx];
1768 #ifdef CONFIG_NET_RX_BUSY_POLL
1769 ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1771 napi_enable(&q_vector->napi);
1775 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1778 struct ixgbevf_q_vector *q_vector;
1779 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1781 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1782 q_vector = adapter->q_vector[q_idx];
1783 napi_disable(&q_vector->napi);
1784 #ifdef CONFIG_NET_RX_BUSY_POLL
1785 while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1786 pr_info("QV %d locked\n", q_idx);
1787 usleep_range(1000, 20000);
1789 #endif /* CONFIG_NET_RX_BUSY_POLL */
1793 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1795 struct ixgbe_hw *hw = &adapter->hw;
1796 unsigned int def_q = 0;
1797 unsigned int num_tcs = 0;
1798 unsigned int num_rx_queues = 1;
1801 spin_lock_bh(&adapter->mbx_lock);
1803 /* fetch queue configuration from the PF */
1804 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1806 spin_unlock_bh(&adapter->mbx_lock);
1812 /* update default Tx ring register index */
1813 adapter->tx_ring[0]->reg_idx = def_q;
1815 /* we need as many queues as traffic classes */
1816 num_rx_queues = num_tcs;
1819 /* if we have a bad config abort request queue reset */
1820 if (adapter->num_rx_queues != num_rx_queues) {
1821 /* force mailbox timeout to prevent further messages */
1822 hw->mbx.timeout = 0;
1824 /* wait for watchdog to come around and bail us out */
1825 adapter->flags |= IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
1831 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1833 ixgbevf_configure_dcb(adapter);
1835 ixgbevf_set_rx_mode(adapter->netdev);
1837 ixgbevf_restore_vlan(adapter);
1839 ixgbevf_configure_tx(adapter);
1840 ixgbevf_configure_rx(adapter);
1843 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1845 /* Only save pre-reset stats if there are some */
1846 if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1847 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1848 adapter->stats.base_vfgprc;
1849 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1850 adapter->stats.base_vfgptc;
1851 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1852 adapter->stats.base_vfgorc;
1853 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1854 adapter->stats.base_vfgotc;
1855 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1856 adapter->stats.base_vfmprc;
1860 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1862 struct ixgbe_hw *hw = &adapter->hw;
1864 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1865 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1866 adapter->stats.last_vfgorc |=
1867 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1868 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1869 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1870 adapter->stats.last_vfgotc |=
1871 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1872 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1874 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1875 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1876 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1877 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1878 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
1881 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
1883 struct ixgbe_hw *hw = &adapter->hw;
1884 int api[] = { ixgbe_mbox_api_11,
1886 ixgbe_mbox_api_unknown };
1887 int err = 0, idx = 0;
1889 spin_lock_bh(&adapter->mbx_lock);
1891 while (api[idx] != ixgbe_mbox_api_unknown) {
1892 err = ixgbevf_negotiate_api_version(hw, api[idx]);
1898 spin_unlock_bh(&adapter->mbx_lock);
1901 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
1903 struct net_device *netdev = adapter->netdev;
1904 struct ixgbe_hw *hw = &adapter->hw;
1906 ixgbevf_configure_msix(adapter);
1908 spin_lock_bh(&adapter->mbx_lock);
1910 if (is_valid_ether_addr(hw->mac.addr))
1911 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
1913 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
1915 spin_unlock_bh(&adapter->mbx_lock);
1917 smp_mb__before_atomic();
1918 clear_bit(__IXGBEVF_DOWN, &adapter->state);
1919 ixgbevf_napi_enable_all(adapter);
1921 /* enable transmits */
1922 netif_tx_start_all_queues(netdev);
1924 ixgbevf_save_reset_stats(adapter);
1925 ixgbevf_init_last_counter_stats(adapter);
1927 hw->mac.get_link_status = 1;
1928 mod_timer(&adapter->watchdog_timer, jiffies);
1931 void ixgbevf_up(struct ixgbevf_adapter *adapter)
1933 struct ixgbe_hw *hw = &adapter->hw;
1935 ixgbevf_configure(adapter);
1937 ixgbevf_up_complete(adapter);
1939 /* clear any pending interrupts, may auto mask */
1940 IXGBE_READ_REG(hw, IXGBE_VTEICR);
1942 ixgbevf_irq_enable(adapter);
1946 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
1947 * @rx_ring: ring to free buffers from
1949 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
1951 struct device *dev = rx_ring->dev;
1955 /* Free Rx ring sk_buff */
1957 dev_kfree_skb(rx_ring->skb);
1958 rx_ring->skb = NULL;
1961 /* ring already cleared, nothing to do */
1962 if (!rx_ring->rx_buffer_info)
1965 /* Free all the Rx ring pages */
1966 for (i = 0; i < rx_ring->count; i++) {
1967 struct ixgbevf_rx_buffer *rx_buffer;
1969 rx_buffer = &rx_ring->rx_buffer_info[i];
1971 dma_unmap_page(dev, rx_buffer->dma,
1972 PAGE_SIZE, DMA_FROM_DEVICE);
1974 if (rx_buffer->page)
1975 __free_page(rx_buffer->page);
1976 rx_buffer->page = NULL;
1979 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
1980 memset(rx_ring->rx_buffer_info, 0, size);
1982 /* Zero out the descriptor ring */
1983 memset(rx_ring->desc, 0, rx_ring->size);
1987 * ixgbevf_clean_tx_ring - Free Tx Buffers
1988 * @tx_ring: ring to be cleaned
1990 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
1992 struct ixgbevf_tx_buffer *tx_buffer_info;
1996 if (!tx_ring->tx_buffer_info)
1999 /* Free all the Tx ring sk_buffs */
2000 for (i = 0; i < tx_ring->count; i++) {
2001 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2002 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2005 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2006 memset(tx_ring->tx_buffer_info, 0, size);
2008 memset(tx_ring->desc, 0, tx_ring->size);
2012 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2013 * @adapter: board private structure
2015 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2019 for (i = 0; i < adapter->num_rx_queues; i++)
2020 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2024 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2025 * @adapter: board private structure
2027 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2031 for (i = 0; i < adapter->num_tx_queues; i++)
2032 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2035 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2037 struct net_device *netdev = adapter->netdev;
2038 struct ixgbe_hw *hw = &adapter->hw;
2041 /* signal that we are down to the interrupt handler */
2042 if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2043 return; /* do nothing if already down */
2045 /* disable all enabled rx queues */
2046 for (i = 0; i < adapter->num_rx_queues; i++)
2047 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2049 netif_tx_disable(netdev);
2053 netif_tx_stop_all_queues(netdev);
2055 ixgbevf_irq_disable(adapter);
2057 ixgbevf_napi_disable_all(adapter);
2059 del_timer_sync(&adapter->watchdog_timer);
2060 /* can't call flush scheduled work here because it can deadlock
2061 * if linkwatch_event tries to acquire the rtnl_lock which we are
2063 while (adapter->flags & IXGBE_FLAG_IN_WATCHDOG_TASK)
2066 /* disable transmits in the hardware now that interrupts are off */
2067 for (i = 0; i < adapter->num_tx_queues; i++) {
2068 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2070 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2071 IXGBE_TXDCTL_SWFLSH);
2074 netif_carrier_off(netdev);
2076 if (!pci_channel_offline(adapter->pdev))
2077 ixgbevf_reset(adapter);
2079 ixgbevf_clean_all_tx_rings(adapter);
2080 ixgbevf_clean_all_rx_rings(adapter);
2083 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2085 WARN_ON(in_interrupt());
2087 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2090 ixgbevf_down(adapter);
2091 ixgbevf_up(adapter);
2093 clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2096 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2098 struct ixgbe_hw *hw = &adapter->hw;
2099 struct net_device *netdev = adapter->netdev;
2101 if (hw->mac.ops.reset_hw(hw)) {
2102 hw_dbg(hw, "PF still resetting\n");
2104 hw->mac.ops.init_hw(hw);
2105 ixgbevf_negotiate_api(adapter);
2108 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2109 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2111 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
2116 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2119 int vector_threshold;
2121 /* We'll want at least 2 (vector_threshold):
2122 * 1) TxQ[0] + RxQ[0] handler
2123 * 2) Other (Link Status Change, etc.)
2125 vector_threshold = MIN_MSIX_COUNT;
2127 /* The more we get, the more we will assign to Tx/Rx Cleanup
2128 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2129 * Right now, we simply care about how many we'll get; we'll
2130 * set them up later while requesting irq's.
2132 vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2133 vector_threshold, vectors);
2136 dev_err(&adapter->pdev->dev,
2137 "Unable to allocate MSI-X interrupts\n");
2138 kfree(adapter->msix_entries);
2139 adapter->msix_entries = NULL;
2143 /* Adjust for only the vectors we'll use, which is minimum
2144 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2145 * vectors we were allocated.
2147 adapter->num_msix_vectors = vectors;
2153 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2154 * @adapter: board private structure to initialize
2156 * This is the top level queue allocation routine. The order here is very
2157 * important, starting with the "most" number of features turned on at once,
2158 * and ending with the smallest set of features. This way large combinations
2159 * can be allocated if they're turned on, and smaller combinations are the
2160 * fallthrough conditions.
2163 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2165 struct ixgbe_hw *hw = &adapter->hw;
2166 unsigned int def_q = 0;
2167 unsigned int num_tcs = 0;
2170 /* Start with base case */
2171 adapter->num_rx_queues = 1;
2172 adapter->num_tx_queues = 1;
2174 spin_lock_bh(&adapter->mbx_lock);
2176 /* fetch queue configuration from the PF */
2177 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2179 spin_unlock_bh(&adapter->mbx_lock);
2184 /* we need as many queues as traffic classes */
2186 adapter->num_rx_queues = num_tcs;
2190 * ixgbevf_alloc_queues - Allocate memory for all rings
2191 * @adapter: board private structure to initialize
2193 * We allocate one ring per queue at run-time since we don't know the
2194 * number of queues at compile-time. The polling_netdev array is
2195 * intended for Multiqueue, but should work fine with a single queue.
2197 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2199 struct ixgbevf_ring *ring;
2202 for (; tx < adapter->num_tx_queues; tx++) {
2203 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2205 goto err_allocation;
2207 ring->dev = &adapter->pdev->dev;
2208 ring->netdev = adapter->netdev;
2209 ring->count = adapter->tx_ring_count;
2210 ring->queue_index = tx;
2213 adapter->tx_ring[tx] = ring;
2216 for (; rx < adapter->num_rx_queues; rx++) {
2217 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2219 goto err_allocation;
2221 ring->dev = &adapter->pdev->dev;
2222 ring->netdev = adapter->netdev;
2224 ring->count = adapter->rx_ring_count;
2225 ring->queue_index = rx;
2228 adapter->rx_ring[rx] = ring;
2235 kfree(adapter->tx_ring[--tx]);
2236 adapter->tx_ring[tx] = NULL;
2240 kfree(adapter->rx_ring[--rx]);
2241 adapter->rx_ring[rx] = NULL;
2247 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2248 * @adapter: board private structure to initialize
2250 * Attempt to configure the interrupts using the best available
2251 * capabilities of the hardware and the kernel.
2253 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2255 struct net_device *netdev = adapter->netdev;
2257 int vector, v_budget;
2260 * It's easy to be greedy for MSI-X vectors, but it really
2261 * doesn't do us much good if we have a lot more vectors
2262 * than CPU's. So let's be conservative and only ask for
2263 * (roughly) the same number of vectors as there are CPU's.
2264 * The default is to use pairs of vectors.
2266 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2267 v_budget = min_t(int, v_budget, num_online_cpus());
2268 v_budget += NON_Q_VECTORS;
2270 /* A failure in MSI-X entry allocation isn't fatal, but it does
2271 * mean we disable MSI-X capabilities of the adapter. */
2272 adapter->msix_entries = kcalloc(v_budget,
2273 sizeof(struct msix_entry), GFP_KERNEL);
2274 if (!adapter->msix_entries) {
2279 for (vector = 0; vector < v_budget; vector++)
2280 adapter->msix_entries[vector].entry = vector;
2282 err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2286 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2290 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2297 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2298 * @adapter: board private structure to initialize
2300 * We allocate one q_vector per queue interrupt. If allocation fails we
2303 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2305 int q_idx, num_q_vectors;
2306 struct ixgbevf_q_vector *q_vector;
2308 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2310 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2311 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2314 q_vector->adapter = adapter;
2315 q_vector->v_idx = q_idx;
2316 netif_napi_add(adapter->netdev, &q_vector->napi,
2318 #ifdef CONFIG_NET_RX_BUSY_POLL
2319 napi_hash_add(&q_vector->napi);
2321 adapter->q_vector[q_idx] = q_vector;
2329 q_vector = adapter->q_vector[q_idx];
2330 #ifdef CONFIG_NET_RX_BUSY_POLL
2331 napi_hash_del(&q_vector->napi);
2333 netif_napi_del(&q_vector->napi);
2335 adapter->q_vector[q_idx] = NULL;
2341 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2342 * @adapter: board private structure to initialize
2344 * This function frees the memory allocated to the q_vectors. In addition if
2345 * NAPI is enabled it will delete any references to the NAPI struct prior
2346 * to freeing the q_vector.
2348 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2350 int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2352 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2353 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2355 adapter->q_vector[q_idx] = NULL;
2356 #ifdef CONFIG_NET_RX_BUSY_POLL
2357 napi_hash_del(&q_vector->napi);
2359 netif_napi_del(&q_vector->napi);
2365 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2366 * @adapter: board private structure
2369 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2371 pci_disable_msix(adapter->pdev);
2372 kfree(adapter->msix_entries);
2373 adapter->msix_entries = NULL;
2377 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2378 * @adapter: board private structure to initialize
2381 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2385 /* Number of supported queues */
2386 ixgbevf_set_num_queues(adapter);
2388 err = ixgbevf_set_interrupt_capability(adapter);
2390 hw_dbg(&adapter->hw,
2391 "Unable to setup interrupt capabilities\n");
2392 goto err_set_interrupt;
2395 err = ixgbevf_alloc_q_vectors(adapter);
2397 hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
2399 goto err_alloc_q_vectors;
2402 err = ixgbevf_alloc_queues(adapter);
2404 pr_err("Unable to allocate memory for queues\n");
2405 goto err_alloc_queues;
2408 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
2409 "Tx Queue count = %u\n",
2410 (adapter->num_rx_queues > 1) ? "Enabled" :
2411 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2413 set_bit(__IXGBEVF_DOWN, &adapter->state);
2417 ixgbevf_free_q_vectors(adapter);
2418 err_alloc_q_vectors:
2419 ixgbevf_reset_interrupt_capability(adapter);
2425 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2426 * @adapter: board private structure to clear interrupt scheme on
2428 * We go through and clear interrupt specific resources and reset the structure
2429 * to pre-load conditions
2431 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2435 for (i = 0; i < adapter->num_tx_queues; i++) {
2436 kfree(adapter->tx_ring[i]);
2437 adapter->tx_ring[i] = NULL;
2439 for (i = 0; i < adapter->num_rx_queues; i++) {
2440 kfree(adapter->rx_ring[i]);
2441 adapter->rx_ring[i] = NULL;
2444 adapter->num_tx_queues = 0;
2445 adapter->num_rx_queues = 0;
2447 ixgbevf_free_q_vectors(adapter);
2448 ixgbevf_reset_interrupt_capability(adapter);
2452 * ixgbevf_sw_init - Initialize general software structures
2453 * (struct ixgbevf_adapter)
2454 * @adapter: board private structure to initialize
2456 * ixgbevf_sw_init initializes the Adapter private data structure.
2457 * Fields are initialized based on PCI device information and
2458 * OS network device settings (MTU size).
2460 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2462 struct ixgbe_hw *hw = &adapter->hw;
2463 struct pci_dev *pdev = adapter->pdev;
2464 struct net_device *netdev = adapter->netdev;
2467 /* PCI config space info */
2469 hw->vendor_id = pdev->vendor;
2470 hw->device_id = pdev->device;
2471 hw->revision_id = pdev->revision;
2472 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2473 hw->subsystem_device_id = pdev->subsystem_device;
2475 hw->mbx.ops.init_params(hw);
2477 /* assume legacy case in which PF would only give VF 2 queues */
2478 hw->mac.max_tx_queues = 2;
2479 hw->mac.max_rx_queues = 2;
2481 /* lock to protect mailbox accesses */
2482 spin_lock_init(&adapter->mbx_lock);
2484 err = hw->mac.ops.reset_hw(hw);
2486 dev_info(&pdev->dev,
2487 "PF still in reset state. Is the PF interface up?\n");
2489 err = hw->mac.ops.init_hw(hw);
2491 pr_err("init_shared_code failed: %d\n", err);
2494 ixgbevf_negotiate_api(adapter);
2495 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2497 dev_info(&pdev->dev, "Error reading MAC address\n");
2498 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2499 dev_info(&pdev->dev,
2500 "MAC address not assigned by administrator.\n");
2501 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2504 if (!is_valid_ether_addr(netdev->dev_addr)) {
2505 dev_info(&pdev->dev, "Assigning random MAC address\n");
2506 eth_hw_addr_random(netdev);
2507 memcpy(hw->mac.addr, netdev->dev_addr, netdev->addr_len);
2510 /* Enable dynamic interrupt throttling rates */
2511 adapter->rx_itr_setting = 1;
2512 adapter->tx_itr_setting = 1;
2514 /* set default ring sizes */
2515 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2516 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2518 set_bit(__IXGBEVF_DOWN, &adapter->state);
2525 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
2527 u32 current_counter = IXGBE_READ_REG(hw, reg); \
2528 if (current_counter < last_counter) \
2529 counter += 0x100000000LL; \
2530 last_counter = current_counter; \
2531 counter &= 0xFFFFFFFF00000000LL; \
2532 counter |= current_counter; \
2535 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2537 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
2538 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
2539 u64 current_counter = (current_counter_msb << 32) | \
2540 current_counter_lsb; \
2541 if (current_counter < last_counter) \
2542 counter += 0x1000000000LL; \
2543 last_counter = current_counter; \
2544 counter &= 0xFFFFFFF000000000LL; \
2545 counter |= current_counter; \
2548 * ixgbevf_update_stats - Update the board statistics counters.
2549 * @adapter: board private structure
2551 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2553 struct ixgbe_hw *hw = &adapter->hw;
2556 if (!adapter->link_up)
2559 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2560 adapter->stats.vfgprc);
2561 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2562 adapter->stats.vfgptc);
2563 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2564 adapter->stats.last_vfgorc,
2565 adapter->stats.vfgorc);
2566 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2567 adapter->stats.last_vfgotc,
2568 adapter->stats.vfgotc);
2569 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2570 adapter->stats.vfmprc);
2572 for (i = 0; i < adapter->num_rx_queues; i++) {
2573 adapter->hw_csum_rx_error +=
2574 adapter->rx_ring[i]->hw_csum_rx_error;
2575 adapter->rx_ring[i]->hw_csum_rx_error = 0;
2580 * ixgbevf_watchdog - Timer Call-back
2581 * @data: pointer to adapter cast into an unsigned long
2583 static void ixgbevf_watchdog(unsigned long data)
2585 struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2586 struct ixgbe_hw *hw = &adapter->hw;
2591 * Do the watchdog outside of interrupt context due to the lovely
2592 * delays that some of the newer hardware requires
2595 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
2596 goto watchdog_short_circuit;
2598 /* get one bit for every active tx/rx interrupt vector */
2599 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2600 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2601 if (qv->rx.ring || qv->tx.ring)
2605 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2607 watchdog_short_circuit:
2608 schedule_work(&adapter->watchdog_task);
2612 * ixgbevf_tx_timeout - Respond to a Tx Hang
2613 * @netdev: network interface device structure
2615 static void ixgbevf_tx_timeout(struct net_device *netdev)
2617 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2619 /* Do the reset outside of interrupt context */
2620 schedule_work(&adapter->reset_task);
2623 static void ixgbevf_reset_task(struct work_struct *work)
2625 struct ixgbevf_adapter *adapter;
2626 adapter = container_of(work, struct ixgbevf_adapter, reset_task);
2628 /* If we're already down or resetting, just bail */
2629 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2630 test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
2631 test_bit(__IXGBEVF_RESETTING, &adapter->state))
2634 adapter->tx_timeout_count++;
2636 ixgbevf_reinit_locked(adapter);
2640 * ixgbevf_watchdog_task - worker thread to bring link up
2641 * @work: pointer to work_struct containing our data
2643 static void ixgbevf_watchdog_task(struct work_struct *work)
2645 struct ixgbevf_adapter *adapter = container_of(work,
2646 struct ixgbevf_adapter,
2648 struct net_device *netdev = adapter->netdev;
2649 struct ixgbe_hw *hw = &adapter->hw;
2650 u32 link_speed = adapter->link_speed;
2651 bool link_up = adapter->link_up;
2654 if (IXGBE_REMOVED(hw->hw_addr)) {
2655 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2657 ixgbevf_down(adapter);
2662 ixgbevf_queue_reset_subtask(adapter);
2664 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
2667 * Always check the link on the watchdog because we have
2670 spin_lock_bh(&adapter->mbx_lock);
2672 need_reset = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2674 spin_unlock_bh(&adapter->mbx_lock);
2677 adapter->link_up = link_up;
2678 adapter->link_speed = link_speed;
2679 netif_carrier_off(netdev);
2680 netif_tx_stop_all_queues(netdev);
2681 schedule_work(&adapter->reset_task);
2684 adapter->link_up = link_up;
2685 adapter->link_speed = link_speed;
2688 if (!netif_carrier_ok(netdev)) {
2689 char *link_speed_string;
2690 switch (link_speed) {
2691 case IXGBE_LINK_SPEED_10GB_FULL:
2692 link_speed_string = "10 Gbps";
2694 case IXGBE_LINK_SPEED_1GB_FULL:
2695 link_speed_string = "1 Gbps";
2697 case IXGBE_LINK_SPEED_100_FULL:
2698 link_speed_string = "100 Mbps";
2701 link_speed_string = "unknown speed";
2704 dev_info(&adapter->pdev->dev,
2705 "NIC Link is Up, %s\n", link_speed_string);
2706 netif_carrier_on(netdev);
2707 netif_tx_wake_all_queues(netdev);
2710 adapter->link_up = false;
2711 adapter->link_speed = 0;
2712 if (netif_carrier_ok(netdev)) {
2713 dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2714 netif_carrier_off(netdev);
2715 netif_tx_stop_all_queues(netdev);
2719 ixgbevf_update_stats(adapter);
2722 /* Reset the timer */
2723 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
2724 !test_bit(__IXGBEVF_REMOVING, &adapter->state))
2725 mod_timer(&adapter->watchdog_timer,
2726 round_jiffies(jiffies + (2 * HZ)));
2728 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
2732 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2733 * @tx_ring: Tx descriptor ring for a specific queue
2735 * Free all transmit software resources
2737 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2739 ixgbevf_clean_tx_ring(tx_ring);
2741 vfree(tx_ring->tx_buffer_info);
2742 tx_ring->tx_buffer_info = NULL;
2744 /* if not set, then don't free */
2748 dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2751 tx_ring->desc = NULL;
2755 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2756 * @adapter: board private structure
2758 * Free all transmit software resources
2760 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2764 for (i = 0; i < adapter->num_tx_queues; i++)
2765 if (adapter->tx_ring[i]->desc)
2766 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2770 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2771 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2773 * Return 0 on success, negative on failure
2775 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2779 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2780 tx_ring->tx_buffer_info = vzalloc(size);
2781 if (!tx_ring->tx_buffer_info)
2784 /* round up to nearest 4K */
2785 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2786 tx_ring->size = ALIGN(tx_ring->size, 4096);
2788 tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
2789 &tx_ring->dma, GFP_KERNEL);
2796 vfree(tx_ring->tx_buffer_info);
2797 tx_ring->tx_buffer_info = NULL;
2798 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2799 "descriptor ring\n");
2804 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2805 * @adapter: board private structure
2807 * If this function returns with an error, then it's possible one or
2808 * more of the rings is populated (while the rest are not). It is the
2809 * callers duty to clean those orphaned rings.
2811 * Return 0 on success, negative on failure
2813 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2817 for (i = 0; i < adapter->num_tx_queues; i++) {
2818 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
2821 hw_dbg(&adapter->hw,
2822 "Allocation for Tx Queue %u failed\n", i);
2830 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2831 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2833 * Returns 0 on success, negative on failure
2835 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
2839 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2840 rx_ring->rx_buffer_info = vzalloc(size);
2841 if (!rx_ring->rx_buffer_info)
2844 /* Round up to nearest 4K */
2845 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2846 rx_ring->size = ALIGN(rx_ring->size, 4096);
2848 rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
2849 &rx_ring->dma, GFP_KERNEL);
2856 vfree(rx_ring->rx_buffer_info);
2857 rx_ring->rx_buffer_info = NULL;
2858 dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
2863 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
2864 * @adapter: board private structure
2866 * If this function returns with an error, then it's possible one or
2867 * more of the rings is populated (while the rest are not). It is the
2868 * callers duty to clean those orphaned rings.
2870 * Return 0 on success, negative on failure
2872 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
2876 for (i = 0; i < adapter->num_rx_queues; i++) {
2877 err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
2880 hw_dbg(&adapter->hw,
2881 "Allocation for Rx Queue %u failed\n", i);
2888 * ixgbevf_free_rx_resources - Free Rx Resources
2889 * @rx_ring: ring to clean the resources from
2891 * Free all receive software resources
2893 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
2895 ixgbevf_clean_rx_ring(rx_ring);
2897 vfree(rx_ring->rx_buffer_info);
2898 rx_ring->rx_buffer_info = NULL;
2900 dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
2903 rx_ring->desc = NULL;
2907 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
2908 * @adapter: board private structure
2910 * Free all receive software resources
2912 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
2916 for (i = 0; i < adapter->num_rx_queues; i++)
2917 if (adapter->rx_ring[i]->desc)
2918 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
2922 * ixgbevf_open - Called when a network interface is made active
2923 * @netdev: network interface device structure
2925 * Returns 0 on success, negative value on failure
2927 * The open entry point is called when a network interface is made
2928 * active by the system (IFF_UP). At this point all resources needed
2929 * for transmit and receive operations are allocated, the interrupt
2930 * handler is registered with the OS, the watchdog timer is started,
2931 * and the stack is notified that the interface is ready.
2933 static int ixgbevf_open(struct net_device *netdev)
2935 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2936 struct ixgbe_hw *hw = &adapter->hw;
2939 /* A previous failure to open the device because of a lack of
2940 * available MSIX vector resources may have reset the number
2941 * of msix vectors variable to zero. The only way to recover
2942 * is to unload/reload the driver and hope that the system has
2943 * been able to recover some MSIX vector resources.
2945 if (!adapter->num_msix_vectors)
2948 /* disallow open during test */
2949 if (test_bit(__IXGBEVF_TESTING, &adapter->state))
2952 if (hw->adapter_stopped) {
2953 ixgbevf_reset(adapter);
2954 /* if adapter is still stopped then PF isn't up and
2955 * the vf can't start. */
2956 if (hw->adapter_stopped) {
2957 err = IXGBE_ERR_MBX;
2958 pr_err("Unable to start - perhaps the PF Driver isn't "
2960 goto err_setup_reset;
2964 /* allocate transmit descriptors */
2965 err = ixgbevf_setup_all_tx_resources(adapter);
2969 /* allocate receive descriptors */
2970 err = ixgbevf_setup_all_rx_resources(adapter);
2974 ixgbevf_configure(adapter);
2977 * Map the Tx/Rx rings to the vectors we were allotted.
2978 * if request_irq will be called in this function map_rings
2979 * must be called *before* up_complete
2981 ixgbevf_map_rings_to_vectors(adapter);
2983 ixgbevf_up_complete(adapter);
2985 /* clear any pending interrupts, may auto mask */
2986 IXGBE_READ_REG(hw, IXGBE_VTEICR);
2987 err = ixgbevf_request_irq(adapter);
2991 ixgbevf_irq_enable(adapter);
2996 ixgbevf_down(adapter);
2998 ixgbevf_free_all_rx_resources(adapter);
3000 ixgbevf_free_all_tx_resources(adapter);
3001 ixgbevf_reset(adapter);
3009 * ixgbevf_close - Disables a network interface
3010 * @netdev: network interface device structure
3012 * Returns 0, this is not allowed to fail
3014 * The close entry point is called when an interface is de-activated
3015 * by the OS. The hardware is still under the drivers control, but
3016 * needs to be disabled. A global MAC reset is issued to stop the
3017 * hardware, and all transmit and receive resources are freed.
3019 static int ixgbevf_close(struct net_device *netdev)
3021 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3023 ixgbevf_down(adapter);
3024 ixgbevf_free_irq(adapter);
3026 ixgbevf_free_all_tx_resources(adapter);
3027 ixgbevf_free_all_rx_resources(adapter);
3032 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3034 struct net_device *dev = adapter->netdev;
3036 if (!(adapter->flags & IXGBEVF_FLAG_QUEUE_RESET_REQUESTED))
3039 adapter->flags &= ~IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
3041 /* if interface is down do nothing */
3042 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3043 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3046 /* Hardware has to reinitialize queues and interrupts to
3047 * match packet buffer alignment. Unfortunately, the
3048 * hardware is not flexible enough to do this dynamically.
3050 if (netif_running(dev))
3053 ixgbevf_clear_interrupt_scheme(adapter);
3054 ixgbevf_init_interrupt_scheme(adapter);
3056 if (netif_running(dev))
3060 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3061 u32 vlan_macip_lens, u32 type_tucmd,
3064 struct ixgbe_adv_tx_context_desc *context_desc;
3065 u16 i = tx_ring->next_to_use;
3067 context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3070 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3072 /* set bits to identify this as an advanced context descriptor */
3073 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3075 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3076 context_desc->seqnum_seed = 0;
3077 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
3078 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3081 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3082 struct ixgbevf_tx_buffer *first,
3085 struct sk_buff *skb = first->skb;
3086 u32 vlan_macip_lens, type_tucmd;
3087 u32 mss_l4len_idx, l4len;
3090 if (skb->ip_summed != CHECKSUM_PARTIAL)
3093 if (!skb_is_gso(skb))
3096 err = skb_cow_head(skb, 0);
3100 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3101 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3103 if (skb->protocol == htons(ETH_P_IP)) {
3104 struct iphdr *iph = ip_hdr(skb);
3107 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3111 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3112 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3113 IXGBE_TX_FLAGS_CSUM |
3114 IXGBE_TX_FLAGS_IPV4;
3115 } else if (skb_is_gso_v6(skb)) {
3116 ipv6_hdr(skb)->payload_len = 0;
3117 tcp_hdr(skb)->check =
3118 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3119 &ipv6_hdr(skb)->daddr,
3121 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3122 IXGBE_TX_FLAGS_CSUM;
3125 /* compute header lengths */
3126 l4len = tcp_hdrlen(skb);
3128 *hdr_len = skb_transport_offset(skb) + l4len;
3130 /* update gso size and bytecount with header size */
3131 first->gso_segs = skb_shinfo(skb)->gso_segs;
3132 first->bytecount += (first->gso_segs - 1) * *hdr_len;
3134 /* mss_l4len_id: use 1 as index for TSO */
3135 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
3136 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3137 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
3139 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3140 vlan_macip_lens = skb_network_header_len(skb);
3141 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3142 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3144 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3145 type_tucmd, mss_l4len_idx);
3150 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3151 struct ixgbevf_tx_buffer *first)
3153 struct sk_buff *skb = first->skb;
3154 u32 vlan_macip_lens = 0;
3155 u32 mss_l4len_idx = 0;
3158 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3160 switch (skb->protocol) {
3161 case htons(ETH_P_IP):
3162 vlan_macip_lens |= skb_network_header_len(skb);
3163 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3164 l4_hdr = ip_hdr(skb)->protocol;
3166 case htons(ETH_P_IPV6):
3167 vlan_macip_lens |= skb_network_header_len(skb);
3168 l4_hdr = ipv6_hdr(skb)->nexthdr;
3171 if (unlikely(net_ratelimit())) {
3172 dev_warn(tx_ring->dev,
3173 "partial checksum but proto=%x!\n",
3181 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3182 mss_l4len_idx = tcp_hdrlen(skb) <<
3183 IXGBE_ADVTXD_L4LEN_SHIFT;
3186 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3187 mss_l4len_idx = sizeof(struct sctphdr) <<
3188 IXGBE_ADVTXD_L4LEN_SHIFT;
3191 mss_l4len_idx = sizeof(struct udphdr) <<
3192 IXGBE_ADVTXD_L4LEN_SHIFT;
3195 if (unlikely(net_ratelimit())) {
3196 dev_warn(tx_ring->dev,
3197 "partial checksum but l4 proto=%x!\n",
3203 /* update TX checksum flag */
3204 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3207 /* vlan_macip_lens: MACLEN, VLAN tag */
3208 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3209 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3211 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3212 type_tucmd, mss_l4len_idx);
3215 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3217 /* set type for advanced descriptor with frame checksum insertion */
3218 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3219 IXGBE_ADVTXD_DCMD_IFCS |
3220 IXGBE_ADVTXD_DCMD_DEXT);
3222 /* set HW vlan bit if vlan is present */
3223 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3224 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3226 /* set segmentation enable bits for TSO/FSO */
3227 if (tx_flags & IXGBE_TX_FLAGS_TSO)
3228 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3233 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3234 u32 tx_flags, unsigned int paylen)
3236 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3238 /* enable L4 checksum for TSO and TX checksum offload */
3239 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3240 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3242 /* enble IPv4 checksum for TSO */
3243 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3244 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3246 /* use index 1 context for TSO/FSO/FCOE */
3247 if (tx_flags & IXGBE_TX_FLAGS_TSO)
3248 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
3250 /* Check Context must be set if Tx switch is enabled, which it
3251 * always is for case where virtual functions are running
3253 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3255 tx_desc->read.olinfo_status = olinfo_status;
3258 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3259 struct ixgbevf_tx_buffer *first,
3263 struct sk_buff *skb = first->skb;
3264 struct ixgbevf_tx_buffer *tx_buffer;
3265 union ixgbe_adv_tx_desc *tx_desc;
3266 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3267 unsigned int data_len = skb->data_len;
3268 unsigned int size = skb_headlen(skb);
3269 unsigned int paylen = skb->len - hdr_len;
3270 u32 tx_flags = first->tx_flags;
3272 u16 i = tx_ring->next_to_use;
3274 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3276 ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3277 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3279 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3280 if (dma_mapping_error(tx_ring->dev, dma))
3283 /* record length, and DMA address */
3284 dma_unmap_len_set(first, len, size);
3285 dma_unmap_addr_set(first, dma, dma);
3287 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3290 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3291 tx_desc->read.cmd_type_len =
3292 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3296 if (i == tx_ring->count) {
3297 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3301 dma += IXGBE_MAX_DATA_PER_TXD;
3302 size -= IXGBE_MAX_DATA_PER_TXD;
3304 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3305 tx_desc->read.olinfo_status = 0;
3308 if (likely(!data_len))
3311 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3315 if (i == tx_ring->count) {
3316 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3320 size = skb_frag_size(frag);
3323 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3325 if (dma_mapping_error(tx_ring->dev, dma))
3328 tx_buffer = &tx_ring->tx_buffer_info[i];
3329 dma_unmap_len_set(tx_buffer, len, size);
3330 dma_unmap_addr_set(tx_buffer, dma, dma);
3332 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3333 tx_desc->read.olinfo_status = 0;
3338 /* write last descriptor with RS and EOP bits */
3339 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3340 tx_desc->read.cmd_type_len = cmd_type;
3342 /* set the timestamp */
3343 first->time_stamp = jiffies;
3345 /* Force memory writes to complete before letting h/w know there
3346 * are new descriptors to fetch. (Only applicable for weak-ordered
3347 * memory model archs, such as IA-64).
3349 * We also need this memory barrier (wmb) to make certain all of the
3350 * status bits have been updated before next_to_watch is written.
3354 /* set next_to_watch value indicating a packet is present */
3355 first->next_to_watch = tx_desc;
3358 if (i == tx_ring->count)
3361 tx_ring->next_to_use = i;
3363 /* notify HW of packet */
3364 ixgbevf_write_tail(tx_ring, i);
3368 dev_err(tx_ring->dev, "TX DMA map failed\n");
3370 /* clear dma mappings for failed tx_buffer_info map */
3372 tx_buffer = &tx_ring->tx_buffer_info[i];
3373 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3374 if (tx_buffer == first)
3381 tx_ring->next_to_use = i;
3384 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3386 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3387 /* Herbert's original patch had:
3388 * smp_mb__after_netif_stop_queue();
3389 * but since that doesn't exist yet, just open code it. */
3392 /* We need to check again in a case another CPU has just
3393 * made room available. */
3394 if (likely(ixgbevf_desc_unused(tx_ring) < size))
3397 /* A reprieve! - use start_queue because it doesn't call schedule */
3398 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3399 ++tx_ring->tx_stats.restart_queue;
3404 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3406 if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3408 return __ixgbevf_maybe_stop_tx(tx_ring, size);
3411 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3413 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3414 struct ixgbevf_tx_buffer *first;
3415 struct ixgbevf_ring *tx_ring;
3418 u16 count = TXD_USE_COUNT(skb_headlen(skb));
3419 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3423 u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3425 if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3427 return NETDEV_TX_OK;
3430 tx_ring = adapter->tx_ring[skb->queue_mapping];
3433 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3434 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3435 * + 2 desc gap to keep tail from touching head,
3436 * + 1 desc for context descriptor,
3437 * otherwise try next time
3439 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3440 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3441 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3443 count += skb_shinfo(skb)->nr_frags;
3445 if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3446 tx_ring->tx_stats.tx_busy++;
3447 return NETDEV_TX_BUSY;
3450 /* record the location of the first descriptor for this packet */
3451 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3453 first->bytecount = skb->len;
3454 first->gso_segs = 1;
3456 if (vlan_tx_tag_present(skb)) {
3457 tx_flags |= vlan_tx_tag_get(skb);
3458 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3459 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3462 /* record initial flags and protocol */
3463 first->tx_flags = tx_flags;
3464 first->protocol = vlan_get_protocol(skb);
3466 tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3470 ixgbevf_tx_csum(tx_ring, first);
3472 ixgbevf_tx_map(tx_ring, first, hdr_len);
3474 ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3476 return NETDEV_TX_OK;
3479 dev_kfree_skb_any(first->skb);
3482 return NETDEV_TX_OK;
3486 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3487 * @netdev: network interface device structure
3488 * @p: pointer to an address structure
3490 * Returns 0 on success, negative on failure
3492 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3494 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3495 struct ixgbe_hw *hw = &adapter->hw;
3496 struct sockaddr *addr = p;
3498 if (!is_valid_ether_addr(addr->sa_data))
3499 return -EADDRNOTAVAIL;
3501 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3502 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3504 spin_lock_bh(&adapter->mbx_lock);
3506 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3508 spin_unlock_bh(&adapter->mbx_lock);
3514 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3515 * @netdev: network interface device structure
3516 * @new_mtu: new value for maximum frame size
3518 * Returns 0 on success, negative on failure
3520 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3522 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3523 struct ixgbe_hw *hw = &adapter->hw;
3524 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3525 int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3527 switch (adapter->hw.api_version) {
3528 case ixgbe_mbox_api_11:
3529 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3532 if (adapter->hw.mac.type == ixgbe_mac_X540_vf)
3533 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3537 /* MTU < 68 is an error and causes problems on some kernels */
3538 if ((new_mtu < 68) || (max_frame > max_possible_frame))
3541 hw_dbg(hw, "changing MTU from %d to %d\n",
3542 netdev->mtu, new_mtu);
3543 /* must set new MTU before calling down or up */
3544 netdev->mtu = new_mtu;
3546 /* notify the PF of our intent to use this size of frame */
3547 ixgbevf_rlpml_set_vf(hw, max_frame);
3552 #ifdef CONFIG_NET_POLL_CONTROLLER
3553 /* Polling 'interrupt' - used by things like netconsole to send skbs
3554 * without having to re-enable interrupts. It's not called while
3555 * the interrupt routine is executing.
3557 static void ixgbevf_netpoll(struct net_device *netdev)
3559 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3562 /* if interface is down do nothing */
3563 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
3565 for (i = 0; i < adapter->num_rx_queues; i++)
3566 ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
3568 #endif /* CONFIG_NET_POLL_CONTROLLER */
3570 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3572 struct net_device *netdev = pci_get_drvdata(pdev);
3573 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3578 netif_device_detach(netdev);
3580 if (netif_running(netdev)) {
3582 ixgbevf_down(adapter);
3583 ixgbevf_free_irq(adapter);
3584 ixgbevf_free_all_tx_resources(adapter);
3585 ixgbevf_free_all_rx_resources(adapter);
3589 ixgbevf_clear_interrupt_scheme(adapter);
3592 retval = pci_save_state(pdev);
3597 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3598 pci_disable_device(pdev);
3604 static int ixgbevf_resume(struct pci_dev *pdev)
3606 struct net_device *netdev = pci_get_drvdata(pdev);
3607 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3610 pci_restore_state(pdev);
3612 * pci_restore_state clears dev->state_saved so call
3613 * pci_save_state to restore it.
3615 pci_save_state(pdev);
3617 err = pci_enable_device_mem(pdev);
3619 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3622 smp_mb__before_atomic();
3623 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3624 pci_set_master(pdev);
3626 ixgbevf_reset(adapter);
3629 err = ixgbevf_init_interrupt_scheme(adapter);
3632 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3636 if (netif_running(netdev)) {
3637 err = ixgbevf_open(netdev);
3642 netif_device_attach(netdev);
3647 #endif /* CONFIG_PM */
3648 static void ixgbevf_shutdown(struct pci_dev *pdev)
3650 ixgbevf_suspend(pdev, PMSG_SUSPEND);
3653 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3654 struct rtnl_link_stats64 *stats)
3656 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3659 const struct ixgbevf_ring *ring;
3662 ixgbevf_update_stats(adapter);
3664 stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3666 for (i = 0; i < adapter->num_rx_queues; i++) {
3667 ring = adapter->rx_ring[i];
3669 start = u64_stats_fetch_begin_irq(&ring->syncp);
3670 bytes = ring->stats.bytes;
3671 packets = ring->stats.packets;
3672 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3673 stats->rx_bytes += bytes;
3674 stats->rx_packets += packets;
3677 for (i = 0; i < adapter->num_tx_queues; i++) {
3678 ring = adapter->tx_ring[i];
3680 start = u64_stats_fetch_begin_irq(&ring->syncp);
3681 bytes = ring->stats.bytes;
3682 packets = ring->stats.packets;
3683 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3684 stats->tx_bytes += bytes;
3685 stats->tx_packets += packets;
3691 static const struct net_device_ops ixgbevf_netdev_ops = {
3692 .ndo_open = ixgbevf_open,
3693 .ndo_stop = ixgbevf_close,
3694 .ndo_start_xmit = ixgbevf_xmit_frame,
3695 .ndo_set_rx_mode = ixgbevf_set_rx_mode,
3696 .ndo_get_stats64 = ixgbevf_get_stats,
3697 .ndo_validate_addr = eth_validate_addr,
3698 .ndo_set_mac_address = ixgbevf_set_mac,
3699 .ndo_change_mtu = ixgbevf_change_mtu,
3700 .ndo_tx_timeout = ixgbevf_tx_timeout,
3701 .ndo_vlan_rx_add_vid = ixgbevf_vlan_rx_add_vid,
3702 .ndo_vlan_rx_kill_vid = ixgbevf_vlan_rx_kill_vid,
3703 #ifdef CONFIG_NET_RX_BUSY_POLL
3704 .ndo_busy_poll = ixgbevf_busy_poll_recv,
3706 #ifdef CONFIG_NET_POLL_CONTROLLER
3707 .ndo_poll_controller = ixgbevf_netpoll,
3711 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3713 dev->netdev_ops = &ixgbevf_netdev_ops;
3714 ixgbevf_set_ethtool_ops(dev);
3715 dev->watchdog_timeo = 5 * HZ;
3719 * ixgbevf_probe - Device Initialization Routine
3720 * @pdev: PCI device information struct
3721 * @ent: entry in ixgbevf_pci_tbl
3723 * Returns 0 on success, negative on failure
3725 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3726 * The OS initialization, configuring of the adapter private structure,
3727 * and a hardware reset occur.
3729 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3731 struct net_device *netdev;
3732 struct ixgbevf_adapter *adapter = NULL;
3733 struct ixgbe_hw *hw = NULL;
3734 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3735 int err, pci_using_dac;
3737 err = pci_enable_device(pdev);
3741 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3744 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3746 dev_err(&pdev->dev, "No usable DMA "
3747 "configuration, aborting\n");
3753 err = pci_request_regions(pdev, ixgbevf_driver_name);
3755 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3759 pci_set_master(pdev);
3761 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3765 goto err_alloc_etherdev;
3768 SET_NETDEV_DEV(netdev, &pdev->dev);
3770 pci_set_drvdata(pdev, netdev);
3771 adapter = netdev_priv(netdev);
3773 adapter->netdev = netdev;
3774 adapter->pdev = pdev;
3777 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3780 * call save state here in standalone driver because it relies on
3781 * adapter struct to exist, and needs to call netdev_priv
3783 pci_save_state(pdev);
3785 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3786 pci_resource_len(pdev, 0));
3787 adapter->io_addr = hw->hw_addr;
3793 ixgbevf_assign_netdev_ops(netdev);
3796 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3797 hw->mac.type = ii->mac;
3799 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3800 sizeof(struct ixgbe_mbx_operations));
3802 /* setup the private structure */
3803 err = ixgbevf_sw_init(adapter);
3807 /* The HW MAC address was set and/or determined in sw_init */
3808 if (!is_valid_ether_addr(netdev->dev_addr)) {
3809 pr_err("invalid MAC address\n");
3814 netdev->hw_features = NETIF_F_SG |
3821 netdev->features = netdev->hw_features |
3822 NETIF_F_HW_VLAN_CTAG_TX |
3823 NETIF_F_HW_VLAN_CTAG_RX |
3824 NETIF_F_HW_VLAN_CTAG_FILTER;
3826 netdev->vlan_features |= NETIF_F_TSO;
3827 netdev->vlan_features |= NETIF_F_TSO6;
3828 netdev->vlan_features |= NETIF_F_IP_CSUM;
3829 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3830 netdev->vlan_features |= NETIF_F_SG;
3833 netdev->features |= NETIF_F_HIGHDMA;
3835 netdev->priv_flags |= IFF_UNICAST_FLT;
3837 init_timer(&adapter->watchdog_timer);
3838 adapter->watchdog_timer.function = ixgbevf_watchdog;
3839 adapter->watchdog_timer.data = (unsigned long)adapter;
3841 if (IXGBE_REMOVED(hw->hw_addr)) {
3845 INIT_WORK(&adapter->reset_task, ixgbevf_reset_task);
3846 INIT_WORK(&adapter->watchdog_task, ixgbevf_watchdog_task);
3847 set_bit(__IXGBEVF_WORK_INIT, &adapter->state);
3849 err = ixgbevf_init_interrupt_scheme(adapter);
3853 strcpy(netdev->name, "eth%d");
3855 err = register_netdev(netdev);
3859 netif_carrier_off(netdev);
3861 ixgbevf_init_last_counter_stats(adapter);
3863 /* print the MAC address */
3864 hw_dbg(hw, "%pM\n", netdev->dev_addr);
3866 hw_dbg(hw, "MAC: %d\n", hw->mac.type);
3868 hw_dbg(hw, "Intel(R) 82599 Virtual Function\n");
3872 ixgbevf_clear_interrupt_scheme(adapter);
3874 ixgbevf_reset_interrupt_capability(adapter);
3875 iounmap(adapter->io_addr);
3877 free_netdev(netdev);
3879 pci_release_regions(pdev);
3882 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3883 pci_disable_device(pdev);
3888 * ixgbevf_remove - Device Removal Routine
3889 * @pdev: PCI device information struct
3891 * ixgbevf_remove is called by the PCI subsystem to alert the driver
3892 * that it should release a PCI device. The could be caused by a
3893 * Hot-Plug event, or because the driver is going to be removed from
3896 static void ixgbevf_remove(struct pci_dev *pdev)
3898 struct net_device *netdev = pci_get_drvdata(pdev);
3899 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3901 set_bit(__IXGBEVF_REMOVING, &adapter->state);
3903 del_timer_sync(&adapter->watchdog_timer);
3905 cancel_work_sync(&adapter->reset_task);
3906 cancel_work_sync(&adapter->watchdog_task);
3908 if (netdev->reg_state == NETREG_REGISTERED)
3909 unregister_netdev(netdev);
3911 ixgbevf_clear_interrupt_scheme(adapter);
3912 ixgbevf_reset_interrupt_capability(adapter);
3914 iounmap(adapter->io_addr);
3915 pci_release_regions(pdev);
3917 hw_dbg(&adapter->hw, "Remove complete\n");
3919 free_netdev(netdev);
3921 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3922 pci_disable_device(pdev);
3926 * ixgbevf_io_error_detected - called when PCI error is detected
3927 * @pdev: Pointer to PCI device
3928 * @state: The current pci connection state
3930 * This function is called after a PCI bus error affecting
3931 * this device has been detected.
3933 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
3934 pci_channel_state_t state)
3936 struct net_device *netdev = pci_get_drvdata(pdev);
3937 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3939 if (!test_bit(__IXGBEVF_WORK_INIT, &adapter->state))
3940 return PCI_ERS_RESULT_DISCONNECT;
3943 netif_device_detach(netdev);
3945 if (state == pci_channel_io_perm_failure) {
3947 return PCI_ERS_RESULT_DISCONNECT;
3950 if (netif_running(netdev))
3951 ixgbevf_down(adapter);
3953 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3954 pci_disable_device(pdev);
3957 /* Request a slot slot reset. */
3958 return PCI_ERS_RESULT_NEED_RESET;
3962 * ixgbevf_io_slot_reset - called after the pci bus has been reset.
3963 * @pdev: Pointer to PCI device
3965 * Restart the card from scratch, as if from a cold-boot. Implementation
3966 * resembles the first-half of the ixgbevf_resume routine.
3968 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
3970 struct net_device *netdev = pci_get_drvdata(pdev);
3971 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3973 if (pci_enable_device_mem(pdev)) {
3975 "Cannot re-enable PCI device after reset.\n");
3976 return PCI_ERS_RESULT_DISCONNECT;
3979 smp_mb__before_atomic();
3980 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3981 pci_set_master(pdev);
3983 ixgbevf_reset(adapter);
3985 return PCI_ERS_RESULT_RECOVERED;
3989 * ixgbevf_io_resume - called when traffic can start flowing again.
3990 * @pdev: Pointer to PCI device
3992 * This callback is called when the error recovery driver tells us that
3993 * its OK to resume normal operation. Implementation resembles the
3994 * second-half of the ixgbevf_resume routine.
3996 static void ixgbevf_io_resume(struct pci_dev *pdev)
3998 struct net_device *netdev = pci_get_drvdata(pdev);
3999 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4001 if (netif_running(netdev))
4002 ixgbevf_up(adapter);
4004 netif_device_attach(netdev);
4007 /* PCI Error Recovery (ERS) */
4008 static const struct pci_error_handlers ixgbevf_err_handler = {
4009 .error_detected = ixgbevf_io_error_detected,
4010 .slot_reset = ixgbevf_io_slot_reset,
4011 .resume = ixgbevf_io_resume,
4014 static struct pci_driver ixgbevf_driver = {
4015 .name = ixgbevf_driver_name,
4016 .id_table = ixgbevf_pci_tbl,
4017 .probe = ixgbevf_probe,
4018 .remove = ixgbevf_remove,
4020 /* Power Management Hooks */
4021 .suspend = ixgbevf_suspend,
4022 .resume = ixgbevf_resume,
4024 .shutdown = ixgbevf_shutdown,
4025 .err_handler = &ixgbevf_err_handler
4029 * ixgbevf_init_module - Driver Registration Routine
4031 * ixgbevf_init_module is the first routine called when the driver is
4032 * loaded. All it does is register with the PCI subsystem.
4034 static int __init ixgbevf_init_module(void)
4037 pr_info("%s - version %s\n", ixgbevf_driver_string,
4038 ixgbevf_driver_version);
4040 pr_info("%s\n", ixgbevf_copyright);
4042 ret = pci_register_driver(&ixgbevf_driver);
4046 module_init(ixgbevf_init_module);
4049 * ixgbevf_exit_module - Driver Exit Cleanup Routine
4051 * ixgbevf_exit_module is called just before the driver is removed
4054 static void __exit ixgbevf_exit_module(void)
4056 pci_unregister_driver(&ixgbevf_driver);
4061 * ixgbevf_get_hw_dev_name - return device name string
4062 * used by hardware layer to print debugging information
4064 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4066 struct ixgbevf_adapter *adapter = hw->back;
4067 return adapter->netdev->name;
4071 module_exit(ixgbevf_exit_module);
4073 /* ixgbevf_main.c */