Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[linux-3.10.git] / drivers / net / ethernet / intel / ixgbevf / ixgbevf_main.c
1 /*******************************************************************************
2
3   Intel 82599 Virtual Function driver
4   Copyright(c) 1999 - 2012 Intel Corporation.
5
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.
9
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
13   more details.
14
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.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
28
29 /******************************************************************************
30  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
31 ******************************************************************************/
32
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34
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>
42 #include <linux/in.h>
43 #include <linux/ip.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>
51 #include <linux/if.h>
52 #include <linux/if_vlan.h>
53 #include <linux/prefetch.h>
54
55 #include "ixgbevf.h"
56
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";
60
61 #define DRV_VERSION "2.7.12-k"
62 const char ixgbevf_driver_version[] = DRV_VERSION;
63 static char ixgbevf_copyright[] =
64         "Copyright (c) 2009 - 2012 Intel Corporation.";
65
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,
69 };
70
71 /* ixgbevf_pci_tbl - PCI Device ID Table
72  *
73  * Wildcard entries (PCI_ANY_ID) should come last
74  * Last entry must be all 0s
75  *
76  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
77  *   Class, Class Mask, private data (not used) }
78  */
79 static DEFINE_PCI_DEVICE_TABLE(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 */
83         {0, }
84 };
85 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
86
87 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
88 MODULE_DESCRIPTION("Intel(R) 82599 Virtual Function Driver");
89 MODULE_LICENSE("GPL");
90 MODULE_VERSION(DRV_VERSION);
91
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)");
96
97 /* forward decls */
98 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
99 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
100
101 static inline void ixgbevf_release_rx_desc(struct ixgbe_hw *hw,
102                                            struct ixgbevf_ring *rx_ring,
103                                            u32 val)
104 {
105         /*
106          * Force memory writes to complete before letting h/w
107          * know there are new descriptors to fetch.  (Only
108          * applicable for weak-ordered memory model archs,
109          * such as IA-64).
110          */
111         wmb();
112         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rx_ring->reg_idx), val);
113 }
114
115 /**
116  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
117  * @adapter: pointer to adapter struct
118  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
119  * @queue: queue to map the corresponding interrupt to
120  * @msix_vector: the vector to map to the corresponding queue
121  */
122 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
123                              u8 queue, u8 msix_vector)
124 {
125         u32 ivar, index;
126         struct ixgbe_hw *hw = &adapter->hw;
127         if (direction == -1) {
128                 /* other causes */
129                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
130                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
131                 ivar &= ~0xFF;
132                 ivar |= msix_vector;
133                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
134         } else {
135                 /* tx or rx causes */
136                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
137                 index = ((16 * (queue & 1)) + (8 * direction));
138                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
139                 ivar &= ~(0xFF << index);
140                 ivar |= (msix_vector << index);
141                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
142         }
143 }
144
145 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
146                                                struct ixgbevf_tx_buffer
147                                                *tx_buffer_info)
148 {
149         if (tx_buffer_info->dma) {
150                 if (tx_buffer_info->mapped_as_page)
151                         dma_unmap_page(tx_ring->dev,
152                                        tx_buffer_info->dma,
153                                        tx_buffer_info->length,
154                                        DMA_TO_DEVICE);
155                 else
156                         dma_unmap_single(tx_ring->dev,
157                                          tx_buffer_info->dma,
158                                          tx_buffer_info->length,
159                                          DMA_TO_DEVICE);
160                 tx_buffer_info->dma = 0;
161         }
162         if (tx_buffer_info->skb) {
163                 dev_kfree_skb_any(tx_buffer_info->skb);
164                 tx_buffer_info->skb = NULL;
165         }
166         tx_buffer_info->time_stamp = 0;
167         /* tx_buffer_info must be completely set up in the transmit path */
168 }
169
170 #define IXGBE_MAX_TXD_PWR       14
171 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
172
173 /* Tx Descriptors needed, worst case */
174 #define TXD_USE_COUNT(S) DIV_ROUND_UP((S), IXGBE_MAX_DATA_PER_TXD)
175 #define DESC_NEEDED (MAX_SKB_FRAGS + 4)
176
177 static void ixgbevf_tx_timeout(struct net_device *netdev);
178
179 /**
180  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
181  * @q_vector: board private structure
182  * @tx_ring: tx ring to clean
183  **/
184 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
185                                  struct ixgbevf_ring *tx_ring)
186 {
187         struct ixgbevf_adapter *adapter = q_vector->adapter;
188         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
189         struct ixgbevf_tx_buffer *tx_buffer_info;
190         unsigned int i, count = 0;
191         unsigned int total_bytes = 0, total_packets = 0;
192
193         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
194                 return true;
195
196         i = tx_ring->next_to_clean;
197         tx_buffer_info = &tx_ring->tx_buffer_info[i];
198         eop_desc = tx_buffer_info->next_to_watch;
199
200         do {
201                 bool cleaned = false;
202
203                 /* if next_to_watch is not set then there is no work pending */
204                 if (!eop_desc)
205                         break;
206
207                 /* prevent any other reads prior to eop_desc */
208                 read_barrier_depends();
209
210                 /* if DD is not set pending work has not been completed */
211                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
212                         break;
213
214                 /* clear next_to_watch to prevent false hangs */
215                 tx_buffer_info->next_to_watch = NULL;
216
217                 for ( ; !cleaned; count++) {
218                         struct sk_buff *skb;
219                         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
220                         cleaned = (tx_desc == eop_desc);
221                         skb = tx_buffer_info->skb;
222
223                         if (cleaned && skb) {
224                                 unsigned int segs, bytecount;
225
226                                 /* gso_segs is currently only valid for tcp */
227                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
228                                 /* multiply data chunks by size of headers */
229                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
230                                             skb->len;
231                                 total_packets += segs;
232                                 total_bytes += bytecount;
233                         }
234
235                         ixgbevf_unmap_and_free_tx_resource(tx_ring,
236                                                            tx_buffer_info);
237
238                         tx_desc->wb.status = 0;
239
240                         i++;
241                         if (i == tx_ring->count)
242                                 i = 0;
243
244                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
245                 }
246
247                 eop_desc = tx_buffer_info->next_to_watch;
248         } while (count < tx_ring->count);
249
250         tx_ring->next_to_clean = i;
251
252 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
253         if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
254                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
255                 /* Make sure that anybody stopping the queue after this
256                  * sees the new next_to_clean.
257                  */
258                 smp_mb();
259                 if (__netif_subqueue_stopped(tx_ring->netdev,
260                                              tx_ring->queue_index) &&
261                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
262                         netif_wake_subqueue(tx_ring->netdev,
263                                             tx_ring->queue_index);
264                         ++adapter->restart_queue;
265                 }
266         }
267
268         u64_stats_update_begin(&tx_ring->syncp);
269         tx_ring->total_bytes += total_bytes;
270         tx_ring->total_packets += total_packets;
271         u64_stats_update_end(&tx_ring->syncp);
272         q_vector->tx.total_bytes += total_bytes;
273         q_vector->tx.total_packets += total_packets;
274
275         return count < tx_ring->count;
276 }
277
278 /**
279  * ixgbevf_receive_skb - Send a completed packet up the stack
280  * @q_vector: structure containing interrupt and ring information
281  * @skb: packet to send up
282  * @status: hardware indication of status of receive
283  * @rx_desc: rx descriptor
284  **/
285 static void ixgbevf_receive_skb(struct ixgbevf_q_vector *q_vector,
286                                 struct sk_buff *skb, u8 status,
287                                 union ixgbe_adv_rx_desc *rx_desc)
288 {
289         struct ixgbevf_adapter *adapter = q_vector->adapter;
290         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
291         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
292
293         if (is_vlan && test_bit(tag & VLAN_VID_MASK, adapter->active_vlans))
294                 __vlan_hwaccel_put_tag(skb, tag);
295
296         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
297                 napi_gro_receive(&q_vector->napi, skb);
298         else
299                 netif_rx(skb);
300 }
301
302 /**
303  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
304  * @ring: pointer to Rx descriptor ring structure
305  * @status_err: hardware indication of status of receive
306  * @skb: skb currently being received and modified
307  **/
308 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
309                                        u32 status_err, struct sk_buff *skb)
310 {
311         skb_checksum_none_assert(skb);
312
313         /* Rx csum disabled */
314         if (!(ring->netdev->features & NETIF_F_RXCSUM))
315                 return;
316
317         /* if IP and error */
318         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
319             (status_err & IXGBE_RXDADV_ERR_IPE)) {
320                 ring->hw_csum_rx_error++;
321                 return;
322         }
323
324         if (!(status_err & IXGBE_RXD_STAT_L4CS))
325                 return;
326
327         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
328                 ring->hw_csum_rx_error++;
329                 return;
330         }
331
332         /* It must be a TCP or UDP packet with a valid checksum */
333         skb->ip_summed = CHECKSUM_UNNECESSARY;
334         ring->hw_csum_rx_good++;
335 }
336
337 /**
338  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
339  * @adapter: address of board private structure
340  **/
341 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_adapter *adapter,
342                                      struct ixgbevf_ring *rx_ring,
343                                      int cleaned_count)
344 {
345         struct pci_dev *pdev = adapter->pdev;
346         union ixgbe_adv_rx_desc *rx_desc;
347         struct ixgbevf_rx_buffer *bi;
348         unsigned int i = rx_ring->next_to_use;
349
350         bi = &rx_ring->rx_buffer_info[i];
351
352         while (cleaned_count--) {
353                 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
354
355                 if (!bi->skb) {
356                         struct sk_buff *skb;
357
358                         skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
359                                                         rx_ring->rx_buf_len);
360                         if (!skb) {
361                                 adapter->alloc_rx_buff_failed++;
362                                 goto no_buffers;
363                         }
364                         bi->skb = skb;
365
366                         bi->dma = dma_map_single(&pdev->dev, skb->data,
367                                                  rx_ring->rx_buf_len,
368                                                  DMA_FROM_DEVICE);
369                         if (dma_mapping_error(&pdev->dev, bi->dma)) {
370                                 dev_kfree_skb(skb);
371                                 bi->skb = NULL;
372                                 dev_err(&pdev->dev, "RX DMA map failed\n");
373                                 break;
374                         }
375                 }
376                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
377
378                 i++;
379                 if (i == rx_ring->count)
380                         i = 0;
381                 bi = &rx_ring->rx_buffer_info[i];
382         }
383
384 no_buffers:
385         if (rx_ring->next_to_use != i) {
386                 rx_ring->next_to_use = i;
387                 ixgbevf_release_rx_desc(&adapter->hw, rx_ring, i);
388         }
389 }
390
391 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
392                                              u32 qmask)
393 {
394         struct ixgbe_hw *hw = &adapter->hw;
395
396         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
397 }
398
399 static bool ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
400                                  struct ixgbevf_ring *rx_ring,
401                                  int budget)
402 {
403         struct ixgbevf_adapter *adapter = q_vector->adapter;
404         struct pci_dev *pdev = adapter->pdev;
405         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
406         struct ixgbevf_rx_buffer *rx_buffer_info, *next_buffer;
407         struct sk_buff *skb;
408         unsigned int i;
409         u32 len, staterr;
410         int cleaned_count = 0;
411         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
412
413         i = rx_ring->next_to_clean;
414         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
415         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
416         rx_buffer_info = &rx_ring->rx_buffer_info[i];
417
418         while (staterr & IXGBE_RXD_STAT_DD) {
419                 if (!budget)
420                         break;
421                 budget--;
422
423                 rmb(); /* read descriptor and rx_buffer_info after status DD */
424                 len = le16_to_cpu(rx_desc->wb.upper.length);
425                 skb = rx_buffer_info->skb;
426                 prefetch(skb->data - NET_IP_ALIGN);
427                 rx_buffer_info->skb = NULL;
428
429                 if (rx_buffer_info->dma) {
430                         dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
431                                          rx_ring->rx_buf_len,
432                                          DMA_FROM_DEVICE);
433                         rx_buffer_info->dma = 0;
434                         skb_put(skb, len);
435                 }
436
437                 i++;
438                 if (i == rx_ring->count)
439                         i = 0;
440
441                 next_rxd = IXGBEVF_RX_DESC(rx_ring, i);
442                 prefetch(next_rxd);
443                 cleaned_count++;
444
445                 next_buffer = &rx_ring->rx_buffer_info[i];
446
447                 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
448                         skb->next = next_buffer->skb;
449                         IXGBE_CB(skb->next)->prev = skb;
450                         adapter->non_eop_descs++;
451                         goto next_desc;
452                 }
453
454                 /* we should not be chaining buffers, if we did drop the skb */
455                 if (IXGBE_CB(skb)->prev) {
456                         do {
457                                 struct sk_buff *this = skb;
458                                 skb = IXGBE_CB(skb)->prev;
459                                 dev_kfree_skb(this);
460                         } while (skb);
461                         goto next_desc;
462                 }
463
464                 /* ERR_MASK will only have valid bits if EOP set */
465                 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
466                         dev_kfree_skb_irq(skb);
467                         goto next_desc;
468                 }
469
470                 ixgbevf_rx_checksum(rx_ring, staterr, skb);
471
472                 /* probably a little skewed due to removing CRC */
473                 total_rx_bytes += skb->len;
474                 total_rx_packets++;
475
476                 /*
477                  * Work around issue of some types of VM to VM loop back
478                  * packets not getting split correctly
479                  */
480                 if (staterr & IXGBE_RXD_STAT_LB) {
481                         u32 header_fixup_len = skb_headlen(skb);
482                         if (header_fixup_len < 14)
483                                 skb_push(skb, header_fixup_len);
484                 }
485                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
486
487                 /* Workaround hardware that can't do proper VEPA multicast
488                  * source pruning.
489                  */
490                 if ((skb->pkt_type & (PACKET_BROADCAST | PACKET_MULTICAST)) &&
491                     !(compare_ether_addr(adapter->netdev->dev_addr,
492                                         eth_hdr(skb)->h_source))) {
493                         dev_kfree_skb_irq(skb);
494                         goto next_desc;
495                 }
496
497                 ixgbevf_receive_skb(q_vector, skb, staterr, rx_desc);
498
499 next_desc:
500                 rx_desc->wb.upper.status_error = 0;
501
502                 /* return some buffers to hardware, one at a time is too slow */
503                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
504                         ixgbevf_alloc_rx_buffers(adapter, rx_ring,
505                                                  cleaned_count);
506                         cleaned_count = 0;
507                 }
508
509                 /* use prefetched values */
510                 rx_desc = next_rxd;
511                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
512
513                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
514         }
515
516         rx_ring->next_to_clean = i;
517         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
518
519         if (cleaned_count)
520                 ixgbevf_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
521
522         u64_stats_update_begin(&rx_ring->syncp);
523         rx_ring->total_packets += total_rx_packets;
524         rx_ring->total_bytes += total_rx_bytes;
525         u64_stats_update_end(&rx_ring->syncp);
526         q_vector->rx.total_packets += total_rx_packets;
527         q_vector->rx.total_bytes += total_rx_bytes;
528
529         return !!budget;
530 }
531
532 /**
533  * ixgbevf_poll - NAPI polling calback
534  * @napi: napi struct with our devices info in it
535  * @budget: amount of work driver is allowed to do this pass, in packets
536  *
537  * This function will clean more than one or more rings associated with a
538  * q_vector.
539  **/
540 static int ixgbevf_poll(struct napi_struct *napi, int budget)
541 {
542         struct ixgbevf_q_vector *q_vector =
543                 container_of(napi, struct ixgbevf_q_vector, napi);
544         struct ixgbevf_adapter *adapter = q_vector->adapter;
545         struct ixgbevf_ring *ring;
546         int per_ring_budget;
547         bool clean_complete = true;
548
549         ixgbevf_for_each_ring(ring, q_vector->tx)
550                 clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
551
552         /* attempt to distribute budget to each queue fairly, but don't allow
553          * the budget to go below 1 because we'll exit polling */
554         if (q_vector->rx.count > 1)
555                 per_ring_budget = max(budget/q_vector->rx.count, 1);
556         else
557                 per_ring_budget = budget;
558
559         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
560         ixgbevf_for_each_ring(ring, q_vector->rx)
561                 clean_complete &= ixgbevf_clean_rx_irq(q_vector, ring,
562                                                        per_ring_budget);
563         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
564
565         /* If all work not completed, return budget and keep polling */
566         if (!clean_complete)
567                 return budget;
568         /* all work done, exit the polling mode */
569         napi_complete(napi);
570         if (adapter->rx_itr_setting & 1)
571                 ixgbevf_set_itr(q_vector);
572         if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
573                 ixgbevf_irq_enable_queues(adapter,
574                                           1 << q_vector->v_idx);
575
576         return 0;
577 }
578
579 /**
580  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
581  * @q_vector: structure containing interrupt and ring information
582  */
583 static void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
584 {
585         struct ixgbevf_adapter *adapter = q_vector->adapter;
586         struct ixgbe_hw *hw = &adapter->hw;
587         int v_idx = q_vector->v_idx;
588         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
589
590         /*
591          * set the WDIS bit to not clear the timer bits and cause an
592          * immediate assertion of the interrupt
593          */
594         itr_reg |= IXGBE_EITR_CNT_WDIS;
595
596         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
597 }
598
599 /**
600  * ixgbevf_configure_msix - Configure MSI-X hardware
601  * @adapter: board private structure
602  *
603  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
604  * interrupts.
605  **/
606 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
607 {
608         struct ixgbevf_q_vector *q_vector;
609         int q_vectors, v_idx;
610
611         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
612         adapter->eims_enable_mask = 0;
613
614         /*
615          * Populate the IVAR table and set the ITR values to the
616          * corresponding register.
617          */
618         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
619                 struct ixgbevf_ring *ring;
620                 q_vector = adapter->q_vector[v_idx];
621
622                 ixgbevf_for_each_ring(ring, q_vector->rx)
623                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
624
625                 ixgbevf_for_each_ring(ring, q_vector->tx)
626                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
627
628                 if (q_vector->tx.ring && !q_vector->rx.ring) {
629                         /* tx only vector */
630                         if (adapter->tx_itr_setting == 1)
631                                 q_vector->itr = IXGBE_10K_ITR;
632                         else
633                                 q_vector->itr = adapter->tx_itr_setting;
634                 } else {
635                         /* rx or rx/tx vector */
636                         if (adapter->rx_itr_setting == 1)
637                                 q_vector->itr = IXGBE_20K_ITR;
638                         else
639                                 q_vector->itr = adapter->rx_itr_setting;
640                 }
641
642                 /* add q_vector eims value to global eims_enable_mask */
643                 adapter->eims_enable_mask |= 1 << v_idx;
644
645                 ixgbevf_write_eitr(q_vector);
646         }
647
648         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
649         /* setup eims_other and add value to global eims_enable_mask */
650         adapter->eims_other = 1 << v_idx;
651         adapter->eims_enable_mask |= adapter->eims_other;
652 }
653
654 enum latency_range {
655         lowest_latency = 0,
656         low_latency = 1,
657         bulk_latency = 2,
658         latency_invalid = 255
659 };
660
661 /**
662  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
663  * @q_vector: structure containing interrupt and ring information
664  * @ring_container: structure containing ring performance data
665  *
666  *      Stores a new ITR value based on packets and byte
667  *      counts during the last interrupt.  The advantage of per interrupt
668  *      computation is faster updates and more accurate ITR for the current
669  *      traffic pattern.  Constants in this function were computed
670  *      based on theoretical maximum wire speed and thresholds were set based
671  *      on testing data as well as attempting to minimize response time
672  *      while increasing bulk throughput.
673  **/
674 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
675                                struct ixgbevf_ring_container *ring_container)
676 {
677         int bytes = ring_container->total_bytes;
678         int packets = ring_container->total_packets;
679         u32 timepassed_us;
680         u64 bytes_perint;
681         u8 itr_setting = ring_container->itr;
682
683         if (packets == 0)
684                 return;
685
686         /* simple throttlerate management
687          *    0-20MB/s lowest (100000 ints/s)
688          *   20-100MB/s low   (20000 ints/s)
689          *  100-1249MB/s bulk (8000 ints/s)
690          */
691         /* what was last interrupt timeslice? */
692         timepassed_us = q_vector->itr >> 2;
693         bytes_perint = bytes / timepassed_us; /* bytes/usec */
694
695         switch (itr_setting) {
696         case lowest_latency:
697                 if (bytes_perint > 10)
698                         itr_setting = low_latency;
699                 break;
700         case low_latency:
701                 if (bytes_perint > 20)
702                         itr_setting = bulk_latency;
703                 else if (bytes_perint <= 10)
704                         itr_setting = lowest_latency;
705                 break;
706         case bulk_latency:
707                 if (bytes_perint <= 20)
708                         itr_setting = low_latency;
709                 break;
710         }
711
712         /* clear work counters since we have the values we need */
713         ring_container->total_bytes = 0;
714         ring_container->total_packets = 0;
715
716         /* write updated itr to ring container */
717         ring_container->itr = itr_setting;
718 }
719
720 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
721 {
722         u32 new_itr = q_vector->itr;
723         u8 current_itr;
724
725         ixgbevf_update_itr(q_vector, &q_vector->tx);
726         ixgbevf_update_itr(q_vector, &q_vector->rx);
727
728         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
729
730         switch (current_itr) {
731         /* counts and packets in update_itr are dependent on these numbers */
732         case lowest_latency:
733                 new_itr = IXGBE_100K_ITR;
734                 break;
735         case low_latency:
736                 new_itr = IXGBE_20K_ITR;
737                 break;
738         case bulk_latency:
739         default:
740                 new_itr = IXGBE_8K_ITR;
741                 break;
742         }
743
744         if (new_itr != q_vector->itr) {
745                 /* do an exponential smoothing */
746                 new_itr = (10 * new_itr * q_vector->itr) /
747                           ((9 * new_itr) + q_vector->itr);
748
749                 /* save the algorithm value here */
750                 q_vector->itr = new_itr;
751
752                 ixgbevf_write_eitr(q_vector);
753         }
754 }
755
756 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
757 {
758         struct ixgbevf_adapter *adapter = data;
759         struct pci_dev *pdev = adapter->pdev;
760         struct ixgbe_hw *hw = &adapter->hw;
761         u32 msg;
762         bool got_ack = false;
763
764         hw->mac.get_link_status = 1;
765         if (!hw->mbx.ops.check_for_ack(hw))
766                 got_ack = true;
767
768         if (!hw->mbx.ops.check_for_msg(hw)) {
769                 hw->mbx.ops.read(hw, &msg, 1);
770
771                 if ((msg & IXGBE_MBVFICR_VFREQ_MASK) == IXGBE_PF_CONTROL_MSG) {
772                         mod_timer(&adapter->watchdog_timer,
773                                   round_jiffies(jiffies + 1));
774                         adapter->link_up = false;
775                 }
776
777                 if (msg & IXGBE_VT_MSGTYPE_NACK)
778                         dev_info(&pdev->dev,
779                                  "Last Request of type %2.2x to PF Nacked\n",
780                                  msg & 0xFF);
781                 hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFSTS;
782         }
783
784         /* checking for the ack clears the PFACK bit.  Place
785          * it back in the v2p_mailbox cache so that anyone
786          * polling for an ack will not miss it
787          */
788         if (got_ack)
789                 hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFACK;
790
791         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
792
793         return IRQ_HANDLED;
794 }
795
796 /**
797  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
798  * @irq: unused
799  * @data: pointer to our q_vector struct for this interrupt vector
800  **/
801 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
802 {
803         struct ixgbevf_q_vector *q_vector = data;
804
805         /* EIAM disabled interrupts (on this vector) for us */
806         if (q_vector->rx.ring || q_vector->tx.ring)
807                 napi_schedule(&q_vector->napi);
808
809         return IRQ_HANDLED;
810 }
811
812 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
813                                      int r_idx)
814 {
815         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
816
817         a->rx_ring[r_idx].next = q_vector->rx.ring;
818         q_vector->rx.ring = &a->rx_ring[r_idx];
819         q_vector->rx.count++;
820 }
821
822 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
823                                      int t_idx)
824 {
825         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
826
827         a->tx_ring[t_idx].next = q_vector->tx.ring;
828         q_vector->tx.ring = &a->tx_ring[t_idx];
829         q_vector->tx.count++;
830 }
831
832 /**
833  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
834  * @adapter: board private structure to initialize
835  *
836  * This function maps descriptor rings to the queue-specific vectors
837  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
838  * one vector per ring/queue, but on a constrained vector budget, we
839  * group the rings as "efficiently" as possible.  You would add new
840  * mapping configurations in here.
841  **/
842 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
843 {
844         int q_vectors;
845         int v_start = 0;
846         int rxr_idx = 0, txr_idx = 0;
847         int rxr_remaining = adapter->num_rx_queues;
848         int txr_remaining = adapter->num_tx_queues;
849         int i, j;
850         int rqpv, tqpv;
851         int err = 0;
852
853         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
854
855         /*
856          * The ideal configuration...
857          * We have enough vectors to map one per queue.
858          */
859         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
860                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
861                         map_vector_to_rxq(adapter, v_start, rxr_idx);
862
863                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
864                         map_vector_to_txq(adapter, v_start, txr_idx);
865                 goto out;
866         }
867
868         /*
869          * If we don't have enough vectors for a 1-to-1
870          * mapping, we'll have to group them so there are
871          * multiple queues per vector.
872          */
873         /* Re-adjusting *qpv takes care of the remainder. */
874         for (i = v_start; i < q_vectors; i++) {
875                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
876                 for (j = 0; j < rqpv; j++) {
877                         map_vector_to_rxq(adapter, i, rxr_idx);
878                         rxr_idx++;
879                         rxr_remaining--;
880                 }
881         }
882         for (i = v_start; i < q_vectors; i++) {
883                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
884                 for (j = 0; j < tqpv; j++) {
885                         map_vector_to_txq(adapter, i, txr_idx);
886                         txr_idx++;
887                         txr_remaining--;
888                 }
889         }
890
891 out:
892         return err;
893 }
894
895 /**
896  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
897  * @adapter: board private structure
898  *
899  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
900  * interrupts from the kernel.
901  **/
902 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
903 {
904         struct net_device *netdev = adapter->netdev;
905         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
906         int vector, err;
907         int ri = 0, ti = 0;
908
909         for (vector = 0; vector < q_vectors; vector++) {
910                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
911                 struct msix_entry *entry = &adapter->msix_entries[vector];
912
913                 if (q_vector->tx.ring && q_vector->rx.ring) {
914                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
915                                  "%s-%s-%d", netdev->name, "TxRx", ri++);
916                         ti++;
917                 } else if (q_vector->rx.ring) {
918                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
919                                  "%s-%s-%d", netdev->name, "rx", ri++);
920                 } else if (q_vector->tx.ring) {
921                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
922                                  "%s-%s-%d", netdev->name, "tx", ti++);
923                 } else {
924                         /* skip this unused q_vector */
925                         continue;
926                 }
927                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
928                                   q_vector->name, q_vector);
929                 if (err) {
930                         hw_dbg(&adapter->hw,
931                                "request_irq failed for MSIX interrupt "
932                                "Error: %d\n", err);
933                         goto free_queue_irqs;
934                 }
935         }
936
937         err = request_irq(adapter->msix_entries[vector].vector,
938                           &ixgbevf_msix_other, 0, netdev->name, adapter);
939         if (err) {
940                 hw_dbg(&adapter->hw,
941                        "request_irq for msix_other failed: %d\n", err);
942                 goto free_queue_irqs;
943         }
944
945         return 0;
946
947 free_queue_irqs:
948         while (vector) {
949                 vector--;
950                 free_irq(adapter->msix_entries[vector].vector,
951                          adapter->q_vector[vector]);
952         }
953         /* This failure is non-recoverable - it indicates the system is
954          * out of MSIX vector resources and the VF driver cannot run
955          * without them.  Set the number of msix vectors to zero
956          * indicating that not enough can be allocated.  The error
957          * will be returned to the user indicating device open failed.
958          * Any further attempts to force the driver to open will also
959          * fail.  The only way to recover is to unload the driver and
960          * reload it again.  If the system has recovered some MSIX
961          * vectors then it may succeed.
962          */
963         adapter->num_msix_vectors = 0;
964         return err;
965 }
966
967 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
968 {
969         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
970
971         for (i = 0; i < q_vectors; i++) {
972                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
973                 q_vector->rx.ring = NULL;
974                 q_vector->tx.ring = NULL;
975                 q_vector->rx.count = 0;
976                 q_vector->tx.count = 0;
977         }
978 }
979
980 /**
981  * ixgbevf_request_irq - initialize interrupts
982  * @adapter: board private structure
983  *
984  * Attempts to configure interrupts using the best available
985  * capabilities of the hardware and kernel.
986  **/
987 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
988 {
989         int err = 0;
990
991         err = ixgbevf_request_msix_irqs(adapter);
992
993         if (err)
994                 hw_dbg(&adapter->hw,
995                        "request_irq failed, Error %d\n", err);
996
997         return err;
998 }
999
1000 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1001 {
1002         int i, q_vectors;
1003
1004         q_vectors = adapter->num_msix_vectors;
1005         i = q_vectors - 1;
1006
1007         free_irq(adapter->msix_entries[i].vector, adapter);
1008         i--;
1009
1010         for (; i >= 0; i--) {
1011                 /* free only the irqs that were actually requested */
1012                 if (!adapter->q_vector[i]->rx.ring &&
1013                     !adapter->q_vector[i]->tx.ring)
1014                         continue;
1015
1016                 free_irq(adapter->msix_entries[i].vector,
1017                          adapter->q_vector[i]);
1018         }
1019
1020         ixgbevf_reset_q_vectors(adapter);
1021 }
1022
1023 /**
1024  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1025  * @adapter: board private structure
1026  **/
1027 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1028 {
1029         struct ixgbe_hw *hw = &adapter->hw;
1030         int i;
1031
1032         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1033         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1034         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1035
1036         IXGBE_WRITE_FLUSH(hw);
1037
1038         for (i = 0; i < adapter->num_msix_vectors; i++)
1039                 synchronize_irq(adapter->msix_entries[i].vector);
1040 }
1041
1042 /**
1043  * ixgbevf_irq_enable - Enable default interrupt generation settings
1044  * @adapter: board private structure
1045  **/
1046 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1047 {
1048         struct ixgbe_hw *hw = &adapter->hw;
1049
1050         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1051         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1052         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1053 }
1054
1055 /**
1056  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1057  * @adapter: board private structure
1058  *
1059  * Configure the Tx unit of the MAC after a reset.
1060  **/
1061 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1062 {
1063         u64 tdba;
1064         struct ixgbe_hw *hw = &adapter->hw;
1065         u32 i, j, tdlen, txctrl;
1066
1067         /* Setup the HW Tx Head and Tail descriptor pointers */
1068         for (i = 0; i < adapter->num_tx_queues; i++) {
1069                 struct ixgbevf_ring *ring = &adapter->tx_ring[i];
1070                 j = ring->reg_idx;
1071                 tdba = ring->dma;
1072                 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1073                 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1074                                 (tdba & DMA_BIT_MASK(32)));
1075                 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1076                 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j), tdlen);
1077                 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(j), 0);
1078                 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(j), 0);
1079                 adapter->tx_ring[i].head = IXGBE_VFTDH(j);
1080                 adapter->tx_ring[i].tail = IXGBE_VFTDT(j);
1081                 /* Disable Tx Head Writeback RO bit, since this hoses
1082                  * bookkeeping if things aren't delivered in order.
1083                  */
1084                 txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1085                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1086                 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1087         }
1088 }
1089
1090 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1091
1092 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1093 {
1094         struct ixgbevf_ring *rx_ring;
1095         struct ixgbe_hw *hw = &adapter->hw;
1096         u32 srrctl;
1097
1098         rx_ring = &adapter->rx_ring[index];
1099
1100         srrctl = IXGBE_SRRCTL_DROP_EN;
1101
1102         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1103
1104         srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
1105                   IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1106
1107         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1108 }
1109
1110 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter)
1111 {
1112         struct ixgbe_hw *hw = &adapter->hw;
1113         struct net_device *netdev = adapter->netdev;
1114         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1115         int i;
1116         u16 rx_buf_len;
1117
1118         /* notify the PF of our intent to use this size of frame */
1119         ixgbevf_rlpml_set_vf(hw, max_frame);
1120
1121         /* PF will allow an extra 4 bytes past for vlan tagged frames */
1122         max_frame += VLAN_HLEN;
1123
1124         /*
1125          * Allocate buffer sizes that fit well into 32K and
1126          * take into account max frame size of 9.5K
1127          */
1128         if ((hw->mac.type == ixgbe_mac_X540_vf) &&
1129             (max_frame <= MAXIMUM_ETHERNET_VLAN_SIZE))
1130                 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1131         else if (max_frame <= IXGBEVF_RXBUFFER_2K)
1132                 rx_buf_len = IXGBEVF_RXBUFFER_2K;
1133         else if (max_frame <= IXGBEVF_RXBUFFER_4K)
1134                 rx_buf_len = IXGBEVF_RXBUFFER_4K;
1135         else if (max_frame <= IXGBEVF_RXBUFFER_8K)
1136                 rx_buf_len = IXGBEVF_RXBUFFER_8K;
1137         else
1138                 rx_buf_len = IXGBEVF_RXBUFFER_10K;
1139
1140         for (i = 0; i < adapter->num_rx_queues; i++)
1141                 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1142 }
1143
1144 /**
1145  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1146  * @adapter: board private structure
1147  *
1148  * Configure the Rx unit of the MAC after a reset.
1149  **/
1150 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1151 {
1152         u64 rdba;
1153         struct ixgbe_hw *hw = &adapter->hw;
1154         int i, j;
1155         u32 rdlen;
1156
1157         /* PSRTYPE must be initialized in 82599 */
1158         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, 0);
1159
1160         /* set_rx_buffer_len must be called before ring initialization */
1161         ixgbevf_set_rx_buffer_len(adapter);
1162
1163         rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1164         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1165          * the Base and Length of the Rx Descriptor Ring */
1166         for (i = 0; i < adapter->num_rx_queues; i++) {
1167                 rdba = adapter->rx_ring[i].dma;
1168                 j = adapter->rx_ring[i].reg_idx;
1169                 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1170                                 (rdba & DMA_BIT_MASK(32)));
1171                 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1172                 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j), rdlen);
1173                 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(j), 0);
1174                 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(j), 0);
1175                 adapter->rx_ring[i].head = IXGBE_VFRDH(j);
1176                 adapter->rx_ring[i].tail = IXGBE_VFRDT(j);
1177
1178                 ixgbevf_configure_srrctl(adapter, j);
1179         }
1180 }
1181
1182 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1183 {
1184         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1185         struct ixgbe_hw *hw = &adapter->hw;
1186         int err;
1187
1188         spin_lock_bh(&adapter->mbx_lock);
1189
1190         /* add VID to filter table */
1191         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1192
1193         spin_unlock_bh(&adapter->mbx_lock);
1194
1195         /* translate error return types so error makes sense */
1196         if (err == IXGBE_ERR_MBX)
1197                 return -EIO;
1198
1199         if (err == IXGBE_ERR_INVALID_ARGUMENT)
1200                 return -EACCES;
1201
1202         set_bit(vid, adapter->active_vlans);
1203
1204         return err;
1205 }
1206
1207 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1208 {
1209         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1210         struct ixgbe_hw *hw = &adapter->hw;
1211         int err = -EOPNOTSUPP;
1212
1213         spin_lock_bh(&adapter->mbx_lock);
1214
1215         /* remove VID from filter table */
1216         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1217
1218         spin_unlock_bh(&adapter->mbx_lock);
1219
1220         clear_bit(vid, adapter->active_vlans);
1221
1222         return err;
1223 }
1224
1225 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1226 {
1227         u16 vid;
1228
1229         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1230                 ixgbevf_vlan_rx_add_vid(adapter->netdev, vid);
1231 }
1232
1233 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1234 {
1235         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1236         struct ixgbe_hw *hw = &adapter->hw;
1237         int count = 0;
1238
1239         if ((netdev_uc_count(netdev)) > 10) {
1240                 pr_err("Too many unicast filters - No Space\n");
1241                 return -ENOSPC;
1242         }
1243
1244         if (!netdev_uc_empty(netdev)) {
1245                 struct netdev_hw_addr *ha;
1246                 netdev_for_each_uc_addr(ha, netdev) {
1247                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1248                         udelay(200);
1249                 }
1250         } else {
1251                 /*
1252                  * If the list is empty then send message to PF driver to
1253                  * clear all macvlans on this VF.
1254                  */
1255                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1256         }
1257
1258         return count;
1259 }
1260
1261 /**
1262  * ixgbevf_set_rx_mode - Multicast and unicast set
1263  * @netdev: network interface device structure
1264  *
1265  * The set_rx_method entry point is called whenever the multicast address
1266  * list, unicast address list or the network interface flags are updated.
1267  * This routine is responsible for configuring the hardware for proper
1268  * multicast mode and configuring requested unicast filters.
1269  **/
1270 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1271 {
1272         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1273         struct ixgbe_hw *hw = &adapter->hw;
1274
1275         spin_lock_bh(&adapter->mbx_lock);
1276
1277         /* reprogram multicast list */
1278         hw->mac.ops.update_mc_addr_list(hw, netdev);
1279
1280         ixgbevf_write_uc_addr_list(netdev);
1281
1282         spin_unlock_bh(&adapter->mbx_lock);
1283 }
1284
1285 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1286 {
1287         int q_idx;
1288         struct ixgbevf_q_vector *q_vector;
1289         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1290
1291         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1292                 q_vector = adapter->q_vector[q_idx];
1293                 napi_enable(&q_vector->napi);
1294         }
1295 }
1296
1297 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1298 {
1299         int q_idx;
1300         struct ixgbevf_q_vector *q_vector;
1301         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1302
1303         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1304                 q_vector = adapter->q_vector[q_idx];
1305                 napi_disable(&q_vector->napi);
1306         }
1307 }
1308
1309 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1310 {
1311         struct net_device *netdev = adapter->netdev;
1312         int i;
1313
1314         ixgbevf_set_rx_mode(netdev);
1315
1316         ixgbevf_restore_vlan(adapter);
1317
1318         ixgbevf_configure_tx(adapter);
1319         ixgbevf_configure_rx(adapter);
1320         for (i = 0; i < adapter->num_rx_queues; i++) {
1321                 struct ixgbevf_ring *ring = &adapter->rx_ring[i];
1322                 ixgbevf_alloc_rx_buffers(adapter, ring,
1323                                          IXGBE_DESC_UNUSED(ring));
1324         }
1325 }
1326
1327 #define IXGBE_MAX_RX_DESC_POLL 10
1328 static inline void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1329                                                 int rxr)
1330 {
1331         struct ixgbe_hw *hw = &adapter->hw;
1332         int j = adapter->rx_ring[rxr].reg_idx;
1333         int k;
1334
1335         for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
1336                 if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
1337                         break;
1338                 else
1339                         msleep(1);
1340         }
1341         if (k >= IXGBE_MAX_RX_DESC_POLL) {
1342                 hw_dbg(hw, "RXDCTL.ENABLE on Rx queue %d "
1343                        "not set within the polling period\n", rxr);
1344         }
1345
1346         ixgbevf_release_rx_desc(hw, &adapter->rx_ring[rxr],
1347                                 adapter->rx_ring[rxr].count - 1);
1348 }
1349
1350 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1351 {
1352         /* Only save pre-reset stats if there are some */
1353         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1354                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1355                         adapter->stats.base_vfgprc;
1356                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1357                         adapter->stats.base_vfgptc;
1358                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1359                         adapter->stats.base_vfgorc;
1360                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1361                         adapter->stats.base_vfgotc;
1362                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1363                         adapter->stats.base_vfmprc;
1364         }
1365 }
1366
1367 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1368 {
1369         struct ixgbe_hw *hw = &adapter->hw;
1370
1371         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1372         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1373         adapter->stats.last_vfgorc |=
1374                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1375         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1376         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1377         adapter->stats.last_vfgotc |=
1378                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1379         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1380
1381         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1382         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1383         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1384         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1385         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
1386 }
1387
1388 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
1389 {
1390         struct ixgbe_hw *hw = &adapter->hw;
1391         int api[] = { ixgbe_mbox_api_11,
1392                       ixgbe_mbox_api_10,
1393                       ixgbe_mbox_api_unknown };
1394         int err = 0, idx = 0;
1395
1396         spin_lock_bh(&adapter->mbx_lock);
1397
1398         while (api[idx] != ixgbe_mbox_api_unknown) {
1399                 err = ixgbevf_negotiate_api_version(hw, api[idx]);
1400                 if (!err)
1401                         break;
1402                 idx++;
1403         }
1404
1405         spin_unlock_bh(&adapter->mbx_lock);
1406 }
1407
1408 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
1409 {
1410         struct net_device *netdev = adapter->netdev;
1411         struct ixgbe_hw *hw = &adapter->hw;
1412         int i, j = 0;
1413         int num_rx_rings = adapter->num_rx_queues;
1414         u32 txdctl, rxdctl;
1415
1416         for (i = 0; i < adapter->num_tx_queues; i++) {
1417                 j = adapter->tx_ring[i].reg_idx;
1418                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1419                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
1420                 txdctl |= (8 << 16);
1421                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1422         }
1423
1424         for (i = 0; i < adapter->num_tx_queues; i++) {
1425                 j = adapter->tx_ring[i].reg_idx;
1426                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1427                 txdctl |= IXGBE_TXDCTL_ENABLE;
1428                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1429         }
1430
1431         for (i = 0; i < num_rx_rings; i++) {
1432                 j = adapter->rx_ring[i].reg_idx;
1433                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1434                 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1435                 if (hw->mac.type == ixgbe_mac_X540_vf) {
1436                         rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
1437                         rxdctl |= ((netdev->mtu + ETH_HLEN + ETH_FCS_LEN) |
1438                                    IXGBE_RXDCTL_RLPML_EN);
1439                 }
1440                 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1441                 ixgbevf_rx_desc_queue_enable(adapter, i);
1442         }
1443
1444         ixgbevf_configure_msix(adapter);
1445
1446         spin_lock_bh(&adapter->mbx_lock);
1447
1448         if (is_valid_ether_addr(hw->mac.addr))
1449                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
1450         else
1451                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
1452
1453         spin_unlock_bh(&adapter->mbx_lock);
1454
1455         clear_bit(__IXGBEVF_DOWN, &adapter->state);
1456         ixgbevf_napi_enable_all(adapter);
1457
1458         /* enable transmits */
1459         netif_tx_start_all_queues(netdev);
1460
1461         ixgbevf_save_reset_stats(adapter);
1462         ixgbevf_init_last_counter_stats(adapter);
1463
1464         hw->mac.get_link_status = 1;
1465         mod_timer(&adapter->watchdog_timer, jiffies);
1466 }
1467
1468 static int ixgbevf_reset_queues(struct ixgbevf_adapter *adapter)
1469 {
1470         struct ixgbe_hw *hw = &adapter->hw;
1471         struct ixgbevf_ring *rx_ring;
1472         unsigned int def_q = 0;
1473         unsigned int num_tcs = 0;
1474         unsigned int num_rx_queues = 1;
1475         int err, i;
1476
1477         spin_lock_bh(&adapter->mbx_lock);
1478
1479         /* fetch queue configuration from the PF */
1480         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1481
1482         spin_unlock_bh(&adapter->mbx_lock);
1483
1484         if (err)
1485                 return err;
1486
1487         if (num_tcs > 1) {
1488                 /* update default Tx ring register index */
1489                 adapter->tx_ring[0].reg_idx = def_q;
1490
1491                 /* we need as many queues as traffic classes */
1492                 num_rx_queues = num_tcs;
1493         }
1494
1495         /* nothing to do if we have the correct number of queues */
1496         if (adapter->num_rx_queues == num_rx_queues)
1497                 return 0;
1498
1499         /* allocate new rings */
1500         rx_ring = kcalloc(num_rx_queues,
1501                           sizeof(struct ixgbevf_ring), GFP_KERNEL);
1502         if (!rx_ring)
1503                 return -ENOMEM;
1504
1505         /* setup ring fields */
1506         for (i = 0; i < num_rx_queues; i++) {
1507                 rx_ring[i].count = adapter->rx_ring_count;
1508                 rx_ring[i].queue_index = i;
1509                 rx_ring[i].reg_idx = i;
1510                 rx_ring[i].dev = &adapter->pdev->dev;
1511                 rx_ring[i].netdev = adapter->netdev;
1512
1513                 /* allocate resources on the ring */
1514                 err = ixgbevf_setup_rx_resources(adapter, &rx_ring[i]);
1515                 if (err) {
1516                         while (i) {
1517                                 i--;
1518                                 ixgbevf_free_rx_resources(adapter, &rx_ring[i]);
1519                         }
1520                         kfree(rx_ring);
1521                         return err;
1522                 }
1523         }
1524
1525         /* free the existing rings and queues */
1526         ixgbevf_free_all_rx_resources(adapter);
1527         adapter->num_rx_queues = 0;
1528         kfree(adapter->rx_ring);
1529
1530         /* move new rings into position on the adapter struct */
1531         adapter->rx_ring = rx_ring;
1532         adapter->num_rx_queues = num_rx_queues;
1533
1534         /* reset ring to vector mapping */
1535         ixgbevf_reset_q_vectors(adapter);
1536         ixgbevf_map_rings_to_vectors(adapter);
1537
1538         return 0;
1539 }
1540
1541 void ixgbevf_up(struct ixgbevf_adapter *adapter)
1542 {
1543         struct ixgbe_hw *hw = &adapter->hw;
1544
1545         ixgbevf_negotiate_api(adapter);
1546
1547         ixgbevf_reset_queues(adapter);
1548
1549         ixgbevf_configure(adapter);
1550
1551         ixgbevf_up_complete(adapter);
1552
1553         /* clear any pending interrupts, may auto mask */
1554         IXGBE_READ_REG(hw, IXGBE_VTEICR);
1555
1556         ixgbevf_irq_enable(adapter);
1557 }
1558
1559 /**
1560  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
1561  * @adapter: board private structure
1562  * @rx_ring: ring to free buffers from
1563  **/
1564 static void ixgbevf_clean_rx_ring(struct ixgbevf_adapter *adapter,
1565                                   struct ixgbevf_ring *rx_ring)
1566 {
1567         struct pci_dev *pdev = adapter->pdev;
1568         unsigned long size;
1569         unsigned int i;
1570
1571         if (!rx_ring->rx_buffer_info)
1572                 return;
1573
1574         /* Free all the Rx ring sk_buffs */
1575         for (i = 0; i < rx_ring->count; i++) {
1576                 struct ixgbevf_rx_buffer *rx_buffer_info;
1577
1578                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1579                 if (rx_buffer_info->dma) {
1580                         dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
1581                                          rx_ring->rx_buf_len,
1582                                          DMA_FROM_DEVICE);
1583                         rx_buffer_info->dma = 0;
1584                 }
1585                 if (rx_buffer_info->skb) {
1586                         struct sk_buff *skb = rx_buffer_info->skb;
1587                         rx_buffer_info->skb = NULL;
1588                         do {
1589                                 struct sk_buff *this = skb;
1590                                 skb = IXGBE_CB(skb)->prev;
1591                                 dev_kfree_skb(this);
1592                         } while (skb);
1593                 }
1594         }
1595
1596         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
1597         memset(rx_ring->rx_buffer_info, 0, size);
1598
1599         /* Zero out the descriptor ring */
1600         memset(rx_ring->desc, 0, rx_ring->size);
1601
1602         rx_ring->next_to_clean = 0;
1603         rx_ring->next_to_use = 0;
1604
1605         if (rx_ring->head)
1606                 writel(0, adapter->hw.hw_addr + rx_ring->head);
1607         if (rx_ring->tail)
1608                 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1609 }
1610
1611 /**
1612  * ixgbevf_clean_tx_ring - Free Tx Buffers
1613  * @adapter: board private structure
1614  * @tx_ring: ring to be cleaned
1615  **/
1616 static void ixgbevf_clean_tx_ring(struct ixgbevf_adapter *adapter,
1617                                   struct ixgbevf_ring *tx_ring)
1618 {
1619         struct ixgbevf_tx_buffer *tx_buffer_info;
1620         unsigned long size;
1621         unsigned int i;
1622
1623         if (!tx_ring->tx_buffer_info)
1624                 return;
1625
1626         /* Free all the Tx ring sk_buffs */
1627         for (i = 0; i < tx_ring->count; i++) {
1628                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
1629                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
1630         }
1631
1632         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
1633         memset(tx_ring->tx_buffer_info, 0, size);
1634
1635         memset(tx_ring->desc, 0, tx_ring->size);
1636
1637         tx_ring->next_to_use = 0;
1638         tx_ring->next_to_clean = 0;
1639
1640         if (tx_ring->head)
1641                 writel(0, adapter->hw.hw_addr + tx_ring->head);
1642         if (tx_ring->tail)
1643                 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1644 }
1645
1646 /**
1647  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
1648  * @adapter: board private structure
1649  **/
1650 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
1651 {
1652         int i;
1653
1654         for (i = 0; i < adapter->num_rx_queues; i++)
1655                 ixgbevf_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1656 }
1657
1658 /**
1659  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
1660  * @adapter: board private structure
1661  **/
1662 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
1663 {
1664         int i;
1665
1666         for (i = 0; i < adapter->num_tx_queues; i++)
1667                 ixgbevf_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1668 }
1669
1670 void ixgbevf_down(struct ixgbevf_adapter *adapter)
1671 {
1672         struct net_device *netdev = adapter->netdev;
1673         struct ixgbe_hw *hw = &adapter->hw;
1674         u32 txdctl;
1675         int i, j;
1676
1677         /* signal that we are down to the interrupt handler */
1678         set_bit(__IXGBEVF_DOWN, &adapter->state);
1679         /* disable receives */
1680
1681         netif_tx_disable(netdev);
1682
1683         msleep(10);
1684
1685         netif_tx_stop_all_queues(netdev);
1686
1687         ixgbevf_irq_disable(adapter);
1688
1689         ixgbevf_napi_disable_all(adapter);
1690
1691         del_timer_sync(&adapter->watchdog_timer);
1692         /* can't call flush scheduled work here because it can deadlock
1693          * if linkwatch_event tries to acquire the rtnl_lock which we are
1694          * holding */
1695         while (adapter->flags & IXGBE_FLAG_IN_WATCHDOG_TASK)
1696                 msleep(1);
1697
1698         /* disable transmits in the hardware now that interrupts are off */
1699         for (i = 0; i < adapter->num_tx_queues; i++) {
1700                 j = adapter->tx_ring[i].reg_idx;
1701                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1702                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j),
1703                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
1704         }
1705
1706         netif_carrier_off(netdev);
1707
1708         if (!pci_channel_offline(adapter->pdev))
1709                 ixgbevf_reset(adapter);
1710
1711         ixgbevf_clean_all_tx_rings(adapter);
1712         ixgbevf_clean_all_rx_rings(adapter);
1713 }
1714
1715 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
1716 {
1717         WARN_ON(in_interrupt());
1718
1719         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
1720                 msleep(1);
1721
1722         ixgbevf_down(adapter);
1723         ixgbevf_up(adapter);
1724
1725         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
1726 }
1727
1728 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
1729 {
1730         struct ixgbe_hw *hw = &adapter->hw;
1731         struct net_device *netdev = adapter->netdev;
1732
1733         if (hw->mac.ops.reset_hw(hw))
1734                 hw_dbg(hw, "PF still resetting\n");
1735         else
1736                 hw->mac.ops.init_hw(hw);
1737
1738         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1739                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1740                        netdev->addr_len);
1741                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1742                        netdev->addr_len);
1743         }
1744 }
1745
1746 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
1747                                         int vectors)
1748 {
1749         int err = 0;
1750         int vector_threshold;
1751
1752         /* We'll want at least 2 (vector_threshold):
1753          * 1) TxQ[0] + RxQ[0] handler
1754          * 2) Other (Link Status Change, etc.)
1755          */
1756         vector_threshold = MIN_MSIX_COUNT;
1757
1758         /* The more we get, the more we will assign to Tx/Rx Cleanup
1759          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
1760          * Right now, we simply care about how many we'll get; we'll
1761          * set them up later while requesting irq's.
1762          */
1763         while (vectors >= vector_threshold) {
1764                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
1765                                       vectors);
1766                 if (!err || err < 0) /* Success or a nasty failure. */
1767                         break;
1768                 else /* err == number of vectors we should try again with */
1769                         vectors = err;
1770         }
1771
1772         if (vectors < vector_threshold)
1773                 err = -ENOMEM;
1774
1775         if (err) {
1776                 dev_err(&adapter->pdev->dev,
1777                         "Unable to allocate MSI-X interrupts\n");
1778                 kfree(adapter->msix_entries);
1779                 adapter->msix_entries = NULL;
1780         } else {
1781                 /*
1782                  * Adjust for only the vectors we'll use, which is minimum
1783                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
1784                  * vectors we were allocated.
1785                  */
1786                 adapter->num_msix_vectors = vectors;
1787         }
1788
1789         return err;
1790 }
1791
1792 /**
1793  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
1794  * @adapter: board private structure to initialize
1795  *
1796  * This is the top level queue allocation routine.  The order here is very
1797  * important, starting with the "most" number of features turned on at once,
1798  * and ending with the smallest set of features.  This way large combinations
1799  * can be allocated if they're turned on, and smaller combinations are the
1800  * fallthrough conditions.
1801  *
1802  **/
1803 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
1804 {
1805         /* Start with base case */
1806         adapter->num_rx_queues = 1;
1807         adapter->num_tx_queues = 1;
1808 }
1809
1810 /**
1811  * ixgbevf_alloc_queues - Allocate memory for all rings
1812  * @adapter: board private structure to initialize
1813  *
1814  * We allocate one ring per queue at run-time since we don't know the
1815  * number of queues at compile-time.  The polling_netdev array is
1816  * intended for Multiqueue, but should work fine with a single queue.
1817  **/
1818 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
1819 {
1820         int i;
1821
1822         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1823                                    sizeof(struct ixgbevf_ring), GFP_KERNEL);
1824         if (!adapter->tx_ring)
1825                 goto err_tx_ring_allocation;
1826
1827         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1828                                    sizeof(struct ixgbevf_ring), GFP_KERNEL);
1829         if (!adapter->rx_ring)
1830                 goto err_rx_ring_allocation;
1831
1832         for (i = 0; i < adapter->num_tx_queues; i++) {
1833                 adapter->tx_ring[i].count = adapter->tx_ring_count;
1834                 adapter->tx_ring[i].queue_index = i;
1835                 /* reg_idx may be remapped later by DCB config */
1836                 adapter->tx_ring[i].reg_idx = i;
1837                 adapter->tx_ring[i].dev = &adapter->pdev->dev;
1838                 adapter->tx_ring[i].netdev = adapter->netdev;
1839         }
1840
1841         for (i = 0; i < adapter->num_rx_queues; i++) {
1842                 adapter->rx_ring[i].count = adapter->rx_ring_count;
1843                 adapter->rx_ring[i].queue_index = i;
1844                 adapter->rx_ring[i].reg_idx = i;
1845                 adapter->rx_ring[i].dev = &adapter->pdev->dev;
1846                 adapter->rx_ring[i].netdev = adapter->netdev;
1847         }
1848
1849         return 0;
1850
1851 err_rx_ring_allocation:
1852         kfree(adapter->tx_ring);
1853 err_tx_ring_allocation:
1854         return -ENOMEM;
1855 }
1856
1857 /**
1858  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
1859  * @adapter: board private structure to initialize
1860  *
1861  * Attempt to configure the interrupts using the best available
1862  * capabilities of the hardware and the kernel.
1863  **/
1864 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
1865 {
1866         struct net_device *netdev = adapter->netdev;
1867         int err = 0;
1868         int vector, v_budget;
1869
1870         /*
1871          * It's easy to be greedy for MSI-X vectors, but it really
1872          * doesn't do us much good if we have a lot more vectors
1873          * than CPU's.  So let's be conservative and only ask for
1874          * (roughly) the same number of vectors as there are CPU's.
1875          * The default is to use pairs of vectors.
1876          */
1877         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
1878         v_budget = min_t(int, v_budget, num_online_cpus());
1879         v_budget += NON_Q_VECTORS;
1880
1881         /* A failure in MSI-X entry allocation isn't fatal, but it does
1882          * mean we disable MSI-X capabilities of the adapter. */
1883         adapter->msix_entries = kcalloc(v_budget,
1884                                         sizeof(struct msix_entry), GFP_KERNEL);
1885         if (!adapter->msix_entries) {
1886                 err = -ENOMEM;
1887                 goto out;
1888         }
1889
1890         for (vector = 0; vector < v_budget; vector++)
1891                 adapter->msix_entries[vector].entry = vector;
1892
1893         err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
1894         if (err)
1895                 goto out;
1896
1897         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
1898         if (err)
1899                 goto out;
1900
1901         err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
1902
1903 out:
1904         return err;
1905 }
1906
1907 /**
1908  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
1909  * @adapter: board private structure to initialize
1910  *
1911  * We allocate one q_vector per queue interrupt.  If allocation fails we
1912  * return -ENOMEM.
1913  **/
1914 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
1915 {
1916         int q_idx, num_q_vectors;
1917         struct ixgbevf_q_vector *q_vector;
1918
1919         num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1920
1921         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1922                 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
1923                 if (!q_vector)
1924                         goto err_out;
1925                 q_vector->adapter = adapter;
1926                 q_vector->v_idx = q_idx;
1927                 netif_napi_add(adapter->netdev, &q_vector->napi,
1928                                ixgbevf_poll, 64);
1929                 adapter->q_vector[q_idx] = q_vector;
1930         }
1931
1932         return 0;
1933
1934 err_out:
1935         while (q_idx) {
1936                 q_idx--;
1937                 q_vector = adapter->q_vector[q_idx];
1938                 netif_napi_del(&q_vector->napi);
1939                 kfree(q_vector);
1940                 adapter->q_vector[q_idx] = NULL;
1941         }
1942         return -ENOMEM;
1943 }
1944
1945 /**
1946  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
1947  * @adapter: board private structure to initialize
1948  *
1949  * This function frees the memory allocated to the q_vectors.  In addition if
1950  * NAPI is enabled it will delete any references to the NAPI struct prior
1951  * to freeing the q_vector.
1952  **/
1953 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
1954 {
1955         int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1956
1957         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1958                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
1959
1960                 adapter->q_vector[q_idx] = NULL;
1961                 netif_napi_del(&q_vector->napi);
1962                 kfree(q_vector);
1963         }
1964 }
1965
1966 /**
1967  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
1968  * @adapter: board private structure
1969  *
1970  **/
1971 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
1972 {
1973         pci_disable_msix(adapter->pdev);
1974         kfree(adapter->msix_entries);
1975         adapter->msix_entries = NULL;
1976 }
1977
1978 /**
1979  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
1980  * @adapter: board private structure to initialize
1981  *
1982  **/
1983 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
1984 {
1985         int err;
1986
1987         /* Number of supported queues */
1988         ixgbevf_set_num_queues(adapter);
1989
1990         err = ixgbevf_set_interrupt_capability(adapter);
1991         if (err) {
1992                 hw_dbg(&adapter->hw,
1993                        "Unable to setup interrupt capabilities\n");
1994                 goto err_set_interrupt;
1995         }
1996
1997         err = ixgbevf_alloc_q_vectors(adapter);
1998         if (err) {
1999                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
2000                        "vectors\n");
2001                 goto err_alloc_q_vectors;
2002         }
2003
2004         err = ixgbevf_alloc_queues(adapter);
2005         if (err) {
2006                 pr_err("Unable to allocate memory for queues\n");
2007                 goto err_alloc_queues;
2008         }
2009
2010         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
2011                "Tx Queue count = %u\n",
2012                (adapter->num_rx_queues > 1) ? "Enabled" :
2013                "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2014
2015         set_bit(__IXGBEVF_DOWN, &adapter->state);
2016
2017         return 0;
2018 err_alloc_queues:
2019         ixgbevf_free_q_vectors(adapter);
2020 err_alloc_q_vectors:
2021         ixgbevf_reset_interrupt_capability(adapter);
2022 err_set_interrupt:
2023         return err;
2024 }
2025
2026 /**
2027  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2028  * @adapter: board private structure to clear interrupt scheme on
2029  *
2030  * We go through and clear interrupt specific resources and reset the structure
2031  * to pre-load conditions
2032  **/
2033 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2034 {
2035         adapter->num_tx_queues = 0;
2036         adapter->num_rx_queues = 0;
2037
2038         ixgbevf_free_q_vectors(adapter);
2039         ixgbevf_reset_interrupt_capability(adapter);
2040 }
2041
2042 /**
2043  * ixgbevf_sw_init - Initialize general software structures
2044  * (struct ixgbevf_adapter)
2045  * @adapter: board private structure to initialize
2046  *
2047  * ixgbevf_sw_init initializes the Adapter private data structure.
2048  * Fields are initialized based on PCI device information and
2049  * OS network device settings (MTU size).
2050  **/
2051 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2052 {
2053         struct ixgbe_hw *hw = &adapter->hw;
2054         struct pci_dev *pdev = adapter->pdev;
2055         int err;
2056
2057         /* PCI config space info */
2058
2059         hw->vendor_id = pdev->vendor;
2060         hw->device_id = pdev->device;
2061         hw->revision_id = pdev->revision;
2062         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2063         hw->subsystem_device_id = pdev->subsystem_device;
2064
2065         hw->mbx.ops.init_params(hw);
2066
2067         /* assume legacy case in which PF would only give VF 2 queues */
2068         hw->mac.max_tx_queues = 2;
2069         hw->mac.max_rx_queues = 2;
2070
2071         err = hw->mac.ops.reset_hw(hw);
2072         if (err) {
2073                 dev_info(&pdev->dev,
2074                          "PF still in reset state, assigning new address\n");
2075                 eth_hw_addr_random(adapter->netdev);
2076                 memcpy(adapter->hw.mac.addr, adapter->netdev->dev_addr,
2077                         adapter->netdev->addr_len);
2078         } else {
2079                 err = hw->mac.ops.init_hw(hw);
2080                 if (err) {
2081                         pr_err("init_shared_code failed: %d\n", err);
2082                         goto out;
2083                 }
2084                 memcpy(adapter->netdev->dev_addr, adapter->hw.mac.addr,
2085                        adapter->netdev->addr_len);
2086         }
2087
2088         /* lock to protect mailbox accesses */
2089         spin_lock_init(&adapter->mbx_lock);
2090
2091         /* Enable dynamic interrupt throttling rates */
2092         adapter->rx_itr_setting = 1;
2093         adapter->tx_itr_setting = 1;
2094
2095         /* set default ring sizes */
2096         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2097         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2098
2099         set_bit(__IXGBEVF_DOWN, &adapter->state);
2100         return 0;
2101
2102 out:
2103         return err;
2104 }
2105
2106 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
2107         {                                                       \
2108                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
2109                 if (current_counter < last_counter)             \
2110                         counter += 0x100000000LL;               \
2111                 last_counter = current_counter;                 \
2112                 counter &= 0xFFFFFFFF00000000LL;                \
2113                 counter |= current_counter;                     \
2114         }
2115
2116 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2117         {                                                                \
2118                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
2119                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
2120                 u64 current_counter = (current_counter_msb << 32) |      \
2121                         current_counter_lsb;                             \
2122                 if (current_counter < last_counter)                      \
2123                         counter += 0x1000000000LL;                       \
2124                 last_counter = current_counter;                          \
2125                 counter &= 0xFFFFFFF000000000LL;                         \
2126                 counter |= current_counter;                              \
2127         }
2128 /**
2129  * ixgbevf_update_stats - Update the board statistics counters.
2130  * @adapter: board private structure
2131  **/
2132 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2133 {
2134         struct ixgbe_hw *hw = &adapter->hw;
2135         int i;
2136
2137         if (!adapter->link_up)
2138                 return;
2139
2140         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2141                                 adapter->stats.vfgprc);
2142         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2143                                 adapter->stats.vfgptc);
2144         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2145                                 adapter->stats.last_vfgorc,
2146                                 adapter->stats.vfgorc);
2147         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2148                                 adapter->stats.last_vfgotc,
2149                                 adapter->stats.vfgotc);
2150         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2151                                 adapter->stats.vfmprc);
2152
2153         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2154                 adapter->hw_csum_rx_error +=
2155                         adapter->rx_ring[i].hw_csum_rx_error;
2156                 adapter->hw_csum_rx_good +=
2157                         adapter->rx_ring[i].hw_csum_rx_good;
2158                 adapter->rx_ring[i].hw_csum_rx_error = 0;
2159                 adapter->rx_ring[i].hw_csum_rx_good = 0;
2160         }
2161 }
2162
2163 /**
2164  * ixgbevf_watchdog - Timer Call-back
2165  * @data: pointer to adapter cast into an unsigned long
2166  **/
2167 static void ixgbevf_watchdog(unsigned long data)
2168 {
2169         struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2170         struct ixgbe_hw *hw = &adapter->hw;
2171         u32 eics = 0;
2172         int i;
2173
2174         /*
2175          * Do the watchdog outside of interrupt context due to the lovely
2176          * delays that some of the newer hardware requires
2177          */
2178
2179         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
2180                 goto watchdog_short_circuit;
2181
2182         /* get one bit for every active tx/rx interrupt vector */
2183         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2184                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2185                 if (qv->rx.ring || qv->tx.ring)
2186                         eics |= 1 << i;
2187         }
2188
2189         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2190
2191 watchdog_short_circuit:
2192         schedule_work(&adapter->watchdog_task);
2193 }
2194
2195 /**
2196  * ixgbevf_tx_timeout - Respond to a Tx Hang
2197  * @netdev: network interface device structure
2198  **/
2199 static void ixgbevf_tx_timeout(struct net_device *netdev)
2200 {
2201         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2202
2203         /* Do the reset outside of interrupt context */
2204         schedule_work(&adapter->reset_task);
2205 }
2206
2207 static void ixgbevf_reset_task(struct work_struct *work)
2208 {
2209         struct ixgbevf_adapter *adapter;
2210         adapter = container_of(work, struct ixgbevf_adapter, reset_task);
2211
2212         /* If we're already down or resetting, just bail */
2213         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2214             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2215                 return;
2216
2217         adapter->tx_timeout_count++;
2218
2219         ixgbevf_reinit_locked(adapter);
2220 }
2221
2222 /**
2223  * ixgbevf_watchdog_task - worker thread to bring link up
2224  * @work: pointer to work_struct containing our data
2225  **/
2226 static void ixgbevf_watchdog_task(struct work_struct *work)
2227 {
2228         struct ixgbevf_adapter *adapter = container_of(work,
2229                                                        struct ixgbevf_adapter,
2230                                                        watchdog_task);
2231         struct net_device *netdev = adapter->netdev;
2232         struct ixgbe_hw *hw = &adapter->hw;
2233         u32 link_speed = adapter->link_speed;
2234         bool link_up = adapter->link_up;
2235         s32 need_reset;
2236
2237         adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
2238
2239         /*
2240          * Always check the link on the watchdog because we have
2241          * no LSC interrupt
2242          */
2243         spin_lock_bh(&adapter->mbx_lock);
2244
2245         need_reset = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2246
2247         spin_unlock_bh(&adapter->mbx_lock);
2248
2249         if (need_reset) {
2250                 adapter->link_up = link_up;
2251                 adapter->link_speed = link_speed;
2252                 netif_carrier_off(netdev);
2253                 netif_tx_stop_all_queues(netdev);
2254                 schedule_work(&adapter->reset_task);
2255                 goto pf_has_reset;
2256         }
2257         adapter->link_up = link_up;
2258         adapter->link_speed = link_speed;
2259
2260         if (link_up) {
2261                 if (!netif_carrier_ok(netdev)) {
2262                         char *link_speed_string;
2263                         switch (link_speed) {
2264                         case IXGBE_LINK_SPEED_10GB_FULL:
2265                                 link_speed_string = "10 Gbps";
2266                                 break;
2267                         case IXGBE_LINK_SPEED_1GB_FULL:
2268                                 link_speed_string = "1 Gbps";
2269                                 break;
2270                         case IXGBE_LINK_SPEED_100_FULL:
2271                                 link_speed_string = "100 Mbps";
2272                                 break;
2273                         default:
2274                                 link_speed_string = "unknown speed";
2275                                 break;
2276                         }
2277                         dev_info(&adapter->pdev->dev,
2278                                 "NIC Link is Up, %s\n", link_speed_string);
2279                         netif_carrier_on(netdev);
2280                         netif_tx_wake_all_queues(netdev);
2281                 }
2282         } else {
2283                 adapter->link_up = false;
2284                 adapter->link_speed = 0;
2285                 if (netif_carrier_ok(netdev)) {
2286                         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2287                         netif_carrier_off(netdev);
2288                         netif_tx_stop_all_queues(netdev);
2289                 }
2290         }
2291
2292         ixgbevf_update_stats(adapter);
2293
2294 pf_has_reset:
2295         /* Reset the timer */
2296         if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
2297                 mod_timer(&adapter->watchdog_timer,
2298                           round_jiffies(jiffies + (2 * HZ)));
2299
2300         adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
2301 }
2302
2303 /**
2304  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2305  * @adapter: board private structure
2306  * @tx_ring: Tx descriptor ring for a specific queue
2307  *
2308  * Free all transmit software resources
2309  **/
2310 void ixgbevf_free_tx_resources(struct ixgbevf_adapter *adapter,
2311                                struct ixgbevf_ring *tx_ring)
2312 {
2313         struct pci_dev *pdev = adapter->pdev;
2314
2315         ixgbevf_clean_tx_ring(adapter, tx_ring);
2316
2317         vfree(tx_ring->tx_buffer_info);
2318         tx_ring->tx_buffer_info = NULL;
2319
2320         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2321                           tx_ring->dma);
2322
2323         tx_ring->desc = NULL;
2324 }
2325
2326 /**
2327  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2328  * @adapter: board private structure
2329  *
2330  * Free all transmit software resources
2331  **/
2332 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2333 {
2334         int i;
2335
2336         for (i = 0; i < adapter->num_tx_queues; i++)
2337                 if (adapter->tx_ring[i].desc)
2338                         ixgbevf_free_tx_resources(adapter,
2339                                                   &adapter->tx_ring[i]);
2340
2341 }
2342
2343 /**
2344  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2345  * @adapter: board private structure
2346  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
2347  *
2348  * Return 0 on success, negative on failure
2349  **/
2350 int ixgbevf_setup_tx_resources(struct ixgbevf_adapter *adapter,
2351                                struct ixgbevf_ring *tx_ring)
2352 {
2353         struct pci_dev *pdev = adapter->pdev;
2354         int size;
2355
2356         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2357         tx_ring->tx_buffer_info = vzalloc(size);
2358         if (!tx_ring->tx_buffer_info)
2359                 goto err;
2360
2361         /* round up to nearest 4K */
2362         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2363         tx_ring->size = ALIGN(tx_ring->size, 4096);
2364
2365         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
2366                                            &tx_ring->dma, GFP_KERNEL);
2367         if (!tx_ring->desc)
2368                 goto err;
2369
2370         tx_ring->next_to_use = 0;
2371         tx_ring->next_to_clean = 0;
2372         return 0;
2373
2374 err:
2375         vfree(tx_ring->tx_buffer_info);
2376         tx_ring->tx_buffer_info = NULL;
2377         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2378                "descriptor ring\n");
2379         return -ENOMEM;
2380 }
2381
2382 /**
2383  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2384  * @adapter: board private structure
2385  *
2386  * If this function returns with an error, then it's possible one or
2387  * more of the rings is populated (while the rest are not).  It is the
2388  * callers duty to clean those orphaned rings.
2389  *
2390  * Return 0 on success, negative on failure
2391  **/
2392 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2393 {
2394         int i, err = 0;
2395
2396         for (i = 0; i < adapter->num_tx_queues; i++) {
2397                 err = ixgbevf_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2398                 if (!err)
2399                         continue;
2400                 hw_dbg(&adapter->hw,
2401                        "Allocation for Tx Queue %u failed\n", i);
2402                 break;
2403         }
2404
2405         return err;
2406 }
2407
2408 /**
2409  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2410  * @adapter: board private structure
2411  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
2412  *
2413  * Returns 0 on success, negative on failure
2414  **/
2415 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
2416                                struct ixgbevf_ring *rx_ring)
2417 {
2418         struct pci_dev *pdev = adapter->pdev;
2419         int size;
2420
2421         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2422         rx_ring->rx_buffer_info = vzalloc(size);
2423         if (!rx_ring->rx_buffer_info)
2424                 goto alloc_failed;
2425
2426         /* Round up to nearest 4K */
2427         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2428         rx_ring->size = ALIGN(rx_ring->size, 4096);
2429
2430         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
2431                                            &rx_ring->dma, GFP_KERNEL);
2432
2433         if (!rx_ring->desc) {
2434                 vfree(rx_ring->rx_buffer_info);
2435                 rx_ring->rx_buffer_info = NULL;
2436                 goto alloc_failed;
2437         }
2438
2439         rx_ring->next_to_clean = 0;
2440         rx_ring->next_to_use = 0;
2441
2442         return 0;
2443 alloc_failed:
2444         return -ENOMEM;
2445 }
2446
2447 /**
2448  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
2449  * @adapter: board private structure
2450  *
2451  * If this function returns with an error, then it's possible one or
2452  * more of the rings is populated (while the rest are not).  It is the
2453  * callers duty to clean those orphaned rings.
2454  *
2455  * Return 0 on success, negative on failure
2456  **/
2457 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
2458 {
2459         int i, err = 0;
2460
2461         for (i = 0; i < adapter->num_rx_queues; i++) {
2462                 err = ixgbevf_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2463                 if (!err)
2464                         continue;
2465                 hw_dbg(&adapter->hw,
2466                        "Allocation for Rx Queue %u failed\n", i);
2467                 break;
2468         }
2469         return err;
2470 }
2471
2472 /**
2473  * ixgbevf_free_rx_resources - Free Rx Resources
2474  * @adapter: board private structure
2475  * @rx_ring: ring to clean the resources from
2476  *
2477  * Free all receive software resources
2478  **/
2479 void ixgbevf_free_rx_resources(struct ixgbevf_adapter *adapter,
2480                                struct ixgbevf_ring *rx_ring)
2481 {
2482         struct pci_dev *pdev = adapter->pdev;
2483
2484         ixgbevf_clean_rx_ring(adapter, rx_ring);
2485
2486         vfree(rx_ring->rx_buffer_info);
2487         rx_ring->rx_buffer_info = NULL;
2488
2489         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2490                           rx_ring->dma);
2491
2492         rx_ring->desc = NULL;
2493 }
2494
2495 /**
2496  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
2497  * @adapter: board private structure
2498  *
2499  * Free all receive software resources
2500  **/
2501 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
2502 {
2503         int i;
2504
2505         for (i = 0; i < adapter->num_rx_queues; i++)
2506                 if (adapter->rx_ring[i].desc)
2507                         ixgbevf_free_rx_resources(adapter,
2508                                                   &adapter->rx_ring[i]);
2509 }
2510
2511 static int ixgbevf_setup_queues(struct ixgbevf_adapter *adapter)
2512 {
2513         struct ixgbe_hw *hw = &adapter->hw;
2514         struct ixgbevf_ring *rx_ring;
2515         unsigned int def_q = 0;
2516         unsigned int num_tcs = 0;
2517         unsigned int num_rx_queues = 1;
2518         int err, i;
2519
2520         spin_lock_bh(&adapter->mbx_lock);
2521
2522         /* fetch queue configuration from the PF */
2523         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2524
2525         spin_unlock_bh(&adapter->mbx_lock);
2526
2527         if (err)
2528                 return err;
2529
2530         if (num_tcs > 1) {
2531                 /* update default Tx ring register index */
2532                 adapter->tx_ring[0].reg_idx = def_q;
2533
2534                 /* we need as many queues as traffic classes */
2535                 num_rx_queues = num_tcs;
2536         }
2537
2538         /* nothing to do if we have the correct number of queues */
2539         if (adapter->num_rx_queues == num_rx_queues)
2540                 return 0;
2541
2542         /* allocate new rings */
2543         rx_ring = kcalloc(num_rx_queues,
2544                           sizeof(struct ixgbevf_ring), GFP_KERNEL);
2545         if (!rx_ring)
2546                 return -ENOMEM;
2547
2548         /* setup ring fields */
2549         for (i = 0; i < num_rx_queues; i++) {
2550                 rx_ring[i].count = adapter->rx_ring_count;
2551                 rx_ring[i].queue_index = i;
2552                 rx_ring[i].reg_idx = i;
2553                 rx_ring[i].dev = &adapter->pdev->dev;
2554                 rx_ring[i].netdev = adapter->netdev;
2555         }
2556
2557         /* free the existing ring and queues */
2558         adapter->num_rx_queues = 0;
2559         kfree(adapter->rx_ring);
2560
2561         /* move new rings into position on the adapter struct */
2562         adapter->rx_ring = rx_ring;
2563         adapter->num_rx_queues = num_rx_queues;
2564
2565         return 0;
2566 }
2567
2568 /**
2569  * ixgbevf_open - Called when a network interface is made active
2570  * @netdev: network interface device structure
2571  *
2572  * Returns 0 on success, negative value on failure
2573  *
2574  * The open entry point is called when a network interface is made
2575  * active by the system (IFF_UP).  At this point all resources needed
2576  * for transmit and receive operations are allocated, the interrupt
2577  * handler is registered with the OS, the watchdog timer is started,
2578  * and the stack is notified that the interface is ready.
2579  **/
2580 static int ixgbevf_open(struct net_device *netdev)
2581 {
2582         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2583         struct ixgbe_hw *hw = &adapter->hw;
2584         int err;
2585
2586         /* A previous failure to open the device because of a lack of
2587          * available MSIX vector resources may have reset the number
2588          * of msix vectors variable to zero.  The only way to recover
2589          * is to unload/reload the driver and hope that the system has
2590          * been able to recover some MSIX vector resources.
2591          */
2592         if (!adapter->num_msix_vectors)
2593                 return -ENOMEM;
2594
2595         /* disallow open during test */
2596         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
2597                 return -EBUSY;
2598
2599         if (hw->adapter_stopped) {
2600                 ixgbevf_reset(adapter);
2601                 /* if adapter is still stopped then PF isn't up and
2602                  * the vf can't start. */
2603                 if (hw->adapter_stopped) {
2604                         err = IXGBE_ERR_MBX;
2605                         pr_err("Unable to start - perhaps the PF Driver isn't "
2606                                "up yet\n");
2607                         goto err_setup_reset;
2608                 }
2609         }
2610
2611         ixgbevf_negotiate_api(adapter);
2612
2613         /* setup queue reg_idx and Rx queue count */
2614         err = ixgbevf_setup_queues(adapter);
2615         if (err)
2616                 goto err_setup_queues;
2617
2618         /* allocate transmit descriptors */
2619         err = ixgbevf_setup_all_tx_resources(adapter);
2620         if (err)
2621                 goto err_setup_tx;
2622
2623         /* allocate receive descriptors */
2624         err = ixgbevf_setup_all_rx_resources(adapter);
2625         if (err)
2626                 goto err_setup_rx;
2627
2628         ixgbevf_configure(adapter);
2629
2630         /*
2631          * Map the Tx/Rx rings to the vectors we were allotted.
2632          * if request_irq will be called in this function map_rings
2633          * must be called *before* up_complete
2634          */
2635         ixgbevf_map_rings_to_vectors(adapter);
2636
2637         ixgbevf_up_complete(adapter);
2638
2639         /* clear any pending interrupts, may auto mask */
2640         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2641         err = ixgbevf_request_irq(adapter);
2642         if (err)
2643                 goto err_req_irq;
2644
2645         ixgbevf_irq_enable(adapter);
2646
2647         return 0;
2648
2649 err_req_irq:
2650         ixgbevf_down(adapter);
2651 err_setup_rx:
2652         ixgbevf_free_all_rx_resources(adapter);
2653 err_setup_tx:
2654         ixgbevf_free_all_tx_resources(adapter);
2655 err_setup_queues:
2656         ixgbevf_reset(adapter);
2657
2658 err_setup_reset:
2659
2660         return err;
2661 }
2662
2663 /**
2664  * ixgbevf_close - Disables a network interface
2665  * @netdev: network interface device structure
2666  *
2667  * Returns 0, this is not allowed to fail
2668  *
2669  * The close entry point is called when an interface is de-activated
2670  * by the OS.  The hardware is still under the drivers control, but
2671  * needs to be disabled.  A global MAC reset is issued to stop the
2672  * hardware, and all transmit and receive resources are freed.
2673  **/
2674 static int ixgbevf_close(struct net_device *netdev)
2675 {
2676         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2677
2678         ixgbevf_down(adapter);
2679         ixgbevf_free_irq(adapter);
2680
2681         ixgbevf_free_all_tx_resources(adapter);
2682         ixgbevf_free_all_rx_resources(adapter);
2683
2684         return 0;
2685 }
2686
2687 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
2688                                 u32 vlan_macip_lens, u32 type_tucmd,
2689                                 u32 mss_l4len_idx)
2690 {
2691         struct ixgbe_adv_tx_context_desc *context_desc;
2692         u16 i = tx_ring->next_to_use;
2693
2694         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
2695
2696         i++;
2697         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2698
2699         /* set bits to identify this as an advanced context descriptor */
2700         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
2701
2702         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
2703         context_desc->seqnum_seed       = 0;
2704         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
2705         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
2706 }
2707
2708 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
2709                        struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2710 {
2711         u32 vlan_macip_lens, type_tucmd;
2712         u32 mss_l4len_idx, l4len;
2713
2714         if (!skb_is_gso(skb))
2715                 return 0;
2716
2717         if (skb_header_cloned(skb)) {
2718                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2719                 if (err)
2720                         return err;
2721         }
2722
2723         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2724         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
2725
2726         if (skb->protocol == htons(ETH_P_IP)) {
2727                 struct iphdr *iph = ip_hdr(skb);
2728                 iph->tot_len = 0;
2729                 iph->check = 0;
2730                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2731                                                          iph->daddr, 0,
2732                                                          IPPROTO_TCP,
2733                                                          0);
2734                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2735         } else if (skb_is_gso_v6(skb)) {
2736                 ipv6_hdr(skb)->payload_len = 0;
2737                 tcp_hdr(skb)->check =
2738                     ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2739                                      &ipv6_hdr(skb)->daddr,
2740                                      0, IPPROTO_TCP, 0);
2741         }
2742
2743         /* compute header lengths */
2744         l4len = tcp_hdrlen(skb);
2745         *hdr_len += l4len;
2746         *hdr_len = skb_transport_offset(skb) + l4len;
2747
2748         /* mss_l4len_id: use 1 as index for TSO */
2749         mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
2750         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
2751         mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
2752
2753         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
2754         vlan_macip_lens = skb_network_header_len(skb);
2755         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2756         vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2757
2758         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2759                             type_tucmd, mss_l4len_idx);
2760
2761         return 1;
2762 }
2763
2764 static bool ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
2765                             struct sk_buff *skb, u32 tx_flags)
2766 {
2767         u32 vlan_macip_lens = 0;
2768         u32 mss_l4len_idx = 0;
2769         u32 type_tucmd = 0;
2770
2771         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2772                 u8 l4_hdr = 0;
2773                 switch (skb->protocol) {
2774                 case __constant_htons(ETH_P_IP):
2775                         vlan_macip_lens |= skb_network_header_len(skb);
2776                         type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2777                         l4_hdr = ip_hdr(skb)->protocol;
2778                         break;
2779                 case __constant_htons(ETH_P_IPV6):
2780                         vlan_macip_lens |= skb_network_header_len(skb);
2781                         l4_hdr = ipv6_hdr(skb)->nexthdr;
2782                         break;
2783                 default:
2784                         if (unlikely(net_ratelimit())) {
2785                                 dev_warn(tx_ring->dev,
2786                                  "partial checksum but proto=%x!\n",
2787                                  skb->protocol);
2788                         }
2789                         break;
2790                 }
2791
2792                 switch (l4_hdr) {
2793                 case IPPROTO_TCP:
2794                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
2795                         mss_l4len_idx = tcp_hdrlen(skb) <<
2796                                         IXGBE_ADVTXD_L4LEN_SHIFT;
2797                         break;
2798                 case IPPROTO_SCTP:
2799                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
2800                         mss_l4len_idx = sizeof(struct sctphdr) <<
2801                                         IXGBE_ADVTXD_L4LEN_SHIFT;
2802                         break;
2803                 case IPPROTO_UDP:
2804                         mss_l4len_idx = sizeof(struct udphdr) <<
2805                                         IXGBE_ADVTXD_L4LEN_SHIFT;
2806                         break;
2807                 default:
2808                         if (unlikely(net_ratelimit())) {
2809                                 dev_warn(tx_ring->dev,
2810                                  "partial checksum but l4 proto=%x!\n",
2811                                  l4_hdr);
2812                         }
2813                         break;
2814                 }
2815         }
2816
2817         /* vlan_macip_lens: MACLEN, VLAN tag */
2818         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2819         vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2820
2821         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2822                             type_tucmd, mss_l4len_idx);
2823
2824         return (skb->ip_summed == CHECKSUM_PARTIAL);
2825 }
2826
2827 static int ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
2828                           struct sk_buff *skb, u32 tx_flags)
2829 {
2830         struct ixgbevf_tx_buffer *tx_buffer_info;
2831         unsigned int len;
2832         unsigned int total = skb->len;
2833         unsigned int offset = 0, size;
2834         int count = 0;
2835         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
2836         unsigned int f;
2837         int i;
2838
2839         i = tx_ring->next_to_use;
2840
2841         len = min(skb_headlen(skb), total);
2842         while (len) {
2843                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2844                 size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2845
2846                 tx_buffer_info->length = size;
2847                 tx_buffer_info->mapped_as_page = false;
2848                 tx_buffer_info->dma = dma_map_single(tx_ring->dev,
2849                                                      skb->data + offset,
2850                                                      size, DMA_TO_DEVICE);
2851                 if (dma_mapping_error(tx_ring->dev, tx_buffer_info->dma))
2852                         goto dma_error;
2853
2854                 len -= size;
2855                 total -= size;
2856                 offset += size;
2857                 count++;
2858                 i++;
2859                 if (i == tx_ring->count)
2860                         i = 0;
2861         }
2862
2863         for (f = 0; f < nr_frags; f++) {
2864                 const struct skb_frag_struct *frag;
2865
2866                 frag = &skb_shinfo(skb)->frags[f];
2867                 len = min((unsigned int)skb_frag_size(frag), total);
2868                 offset = 0;
2869
2870                 while (len) {
2871                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
2872                         size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2873
2874                         tx_buffer_info->length = size;
2875                         tx_buffer_info->dma =
2876                                 skb_frag_dma_map(tx_ring->dev, frag,
2877                                                  offset, size, DMA_TO_DEVICE);
2878                         if (dma_mapping_error(tx_ring->dev,
2879                                               tx_buffer_info->dma))
2880                                 goto dma_error;
2881                         tx_buffer_info->mapped_as_page = true;
2882
2883                         len -= size;
2884                         total -= size;
2885                         offset += size;
2886                         count++;
2887                         i++;
2888                         if (i == tx_ring->count)
2889                                 i = 0;
2890                 }
2891                 if (total == 0)
2892                         break;
2893         }
2894
2895         if (i == 0)
2896                 i = tx_ring->count - 1;
2897         else
2898                 i = i - 1;
2899         tx_ring->tx_buffer_info[i].skb = skb;
2900
2901         return count;
2902
2903 dma_error:
2904         dev_err(tx_ring->dev, "TX DMA map failed\n");
2905
2906         /* clear timestamp and dma mappings for failed tx_buffer_info map */
2907         tx_buffer_info->dma = 0;
2908         count--;
2909
2910         /* clear timestamp and dma mappings for remaining portion of packet */
2911         while (count >= 0) {
2912                 count--;
2913                 i--;
2914                 if (i < 0)
2915                         i += tx_ring->count;
2916                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2917                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2918         }
2919
2920         return count;
2921 }
2922
2923 static void ixgbevf_tx_queue(struct ixgbevf_ring *tx_ring, int tx_flags,
2924                              int count, unsigned int first, u32 paylen,
2925                              u8 hdr_len)
2926 {
2927         union ixgbe_adv_tx_desc *tx_desc = NULL;
2928         struct ixgbevf_tx_buffer *tx_buffer_info;
2929         u32 olinfo_status = 0, cmd_type_len = 0;
2930         unsigned int i;
2931
2932         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
2933
2934         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
2935
2936         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
2937
2938         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2939                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
2940
2941         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
2942                 olinfo_status |= IXGBE_ADVTXD_POPTS_TXSM;
2943
2944         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
2945                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
2946
2947                 /* use index 1 context for tso */
2948                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
2949                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
2950                         olinfo_status |= IXGBE_ADVTXD_POPTS_IXSM;
2951         }
2952
2953         /*
2954          * Check Context must be set if Tx switch is enabled, which it
2955          * always is for case where virtual functions are running
2956          */
2957         olinfo_status |= IXGBE_ADVTXD_CC;
2958
2959         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
2960
2961         i = tx_ring->next_to_use;
2962         while (count--) {
2963                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2964                 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2965                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
2966                 tx_desc->read.cmd_type_len =
2967                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
2968                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2969                 i++;
2970                 if (i == tx_ring->count)
2971                         i = 0;
2972         }
2973
2974         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
2975
2976         tx_ring->tx_buffer_info[first].time_stamp = jiffies;
2977
2978         /* Force memory writes to complete before letting h/w
2979          * know there are new descriptors to fetch.  (Only
2980          * applicable for weak-ordered memory model archs,
2981          * such as IA-64).
2982          */
2983         wmb();
2984
2985         tx_ring->tx_buffer_info[first].next_to_watch = tx_desc;
2986         tx_ring->next_to_use = i;
2987 }
2988
2989 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
2990 {
2991         struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
2992
2993         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
2994         /* Herbert's original patch had:
2995          *  smp_mb__after_netif_stop_queue();
2996          * but since that doesn't exist yet, just open code it. */
2997         smp_mb();
2998
2999         /* We need to check again in a case another CPU has just
3000          * made room available. */
3001         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
3002                 return -EBUSY;
3003
3004         /* A reprieve! - use start_queue because it doesn't call schedule */
3005         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3006         ++adapter->restart_queue;
3007         return 0;
3008 }
3009
3010 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3011 {
3012         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
3013                 return 0;
3014         return __ixgbevf_maybe_stop_tx(tx_ring, size);
3015 }
3016
3017 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3018 {
3019         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3020         struct ixgbevf_ring *tx_ring;
3021         unsigned int first;
3022         unsigned int tx_flags = 0;
3023         u8 hdr_len = 0;
3024         int r_idx = 0, tso;
3025         u16 count = TXD_USE_COUNT(skb_headlen(skb));
3026 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3027         unsigned short f;
3028 #endif
3029         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3030         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3031                 dev_kfree_skb(skb);
3032                 return NETDEV_TX_OK;
3033         }
3034
3035         tx_ring = &adapter->tx_ring[r_idx];
3036
3037         /*
3038          * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3039          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3040          *       + 2 desc gap to keep tail from touching head,
3041          *       + 1 desc for context descriptor,
3042          * otherwise try next time
3043          */
3044 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3045         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3046                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3047 #else
3048         count += skb_shinfo(skb)->nr_frags;
3049 #endif
3050         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3051                 adapter->tx_busy++;
3052                 return NETDEV_TX_BUSY;
3053         }
3054
3055         if (vlan_tx_tag_present(skb)) {
3056                 tx_flags |= vlan_tx_tag_get(skb);
3057                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3058                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3059         }
3060
3061         first = tx_ring->next_to_use;
3062
3063         if (skb->protocol == htons(ETH_P_IP))
3064                 tx_flags |= IXGBE_TX_FLAGS_IPV4;
3065         tso = ixgbevf_tso(tx_ring, skb, tx_flags, &hdr_len);
3066         if (tso < 0) {
3067                 dev_kfree_skb_any(skb);
3068                 return NETDEV_TX_OK;
3069         }
3070
3071         if (tso)
3072                 tx_flags |= IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_CSUM;
3073         else if (ixgbevf_tx_csum(tx_ring, skb, tx_flags))
3074                 tx_flags |= IXGBE_TX_FLAGS_CSUM;
3075
3076         ixgbevf_tx_queue(tx_ring, tx_flags,
3077                          ixgbevf_tx_map(tx_ring, skb, tx_flags),
3078                          first, skb->len, hdr_len);
3079
3080         writel(tx_ring->next_to_use, adapter->hw.hw_addr + tx_ring->tail);
3081
3082         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3083
3084         return NETDEV_TX_OK;
3085 }
3086
3087 /**
3088  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3089  * @netdev: network interface device structure
3090  * @p: pointer to an address structure
3091  *
3092  * Returns 0 on success, negative on failure
3093  **/
3094 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3095 {
3096         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3097         struct ixgbe_hw *hw = &adapter->hw;
3098         struct sockaddr *addr = p;
3099
3100         if (!is_valid_ether_addr(addr->sa_data))
3101                 return -EADDRNOTAVAIL;
3102
3103         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3104         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3105
3106         spin_lock_bh(&adapter->mbx_lock);
3107
3108         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3109
3110         spin_unlock_bh(&adapter->mbx_lock);
3111
3112         return 0;
3113 }
3114
3115 /**
3116  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3117  * @netdev: network interface device structure
3118  * @new_mtu: new value for maximum frame size
3119  *
3120  * Returns 0 on success, negative on failure
3121  **/
3122 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3123 {
3124         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3125         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3126         int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3127
3128         switch (adapter->hw.api_version) {
3129         case ixgbe_mbox_api_11:
3130                 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3131                 break;
3132         default:
3133                 if (adapter->hw.mac.type == ixgbe_mac_X540_vf)
3134                         max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3135                 break;
3136         }
3137
3138         /* MTU < 68 is an error and causes problems on some kernels */
3139         if ((new_mtu < 68) || (max_frame > max_possible_frame))
3140                 return -EINVAL;
3141
3142         hw_dbg(&adapter->hw, "changing MTU from %d to %d\n",
3143                netdev->mtu, new_mtu);
3144         /* must set new MTU before calling down or up */
3145         netdev->mtu = new_mtu;
3146
3147         if (netif_running(netdev))
3148                 ixgbevf_reinit_locked(adapter);
3149
3150         return 0;
3151 }
3152
3153 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3154 {
3155         struct net_device *netdev = pci_get_drvdata(pdev);
3156         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3157 #ifdef CONFIG_PM
3158         int retval = 0;
3159 #endif
3160
3161         netif_device_detach(netdev);
3162
3163         if (netif_running(netdev)) {
3164                 rtnl_lock();
3165                 ixgbevf_down(adapter);
3166                 ixgbevf_free_irq(adapter);
3167                 ixgbevf_free_all_tx_resources(adapter);
3168                 ixgbevf_free_all_rx_resources(adapter);
3169                 rtnl_unlock();
3170         }
3171
3172         ixgbevf_clear_interrupt_scheme(adapter);
3173
3174 #ifdef CONFIG_PM
3175         retval = pci_save_state(pdev);
3176         if (retval)
3177                 return retval;
3178
3179 #endif
3180         pci_disable_device(pdev);
3181
3182         return 0;
3183 }
3184
3185 #ifdef CONFIG_PM
3186 static int ixgbevf_resume(struct pci_dev *pdev)
3187 {
3188         struct ixgbevf_adapter *adapter = pci_get_drvdata(pdev);
3189         struct net_device *netdev = adapter->netdev;
3190         u32 err;
3191
3192         pci_set_power_state(pdev, PCI_D0);
3193         pci_restore_state(pdev);
3194         /*
3195          * pci_restore_state clears dev->state_saved so call
3196          * pci_save_state to restore it.
3197          */
3198         pci_save_state(pdev);
3199
3200         err = pci_enable_device_mem(pdev);
3201         if (err) {
3202                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3203                 return err;
3204         }
3205         pci_set_master(pdev);
3206
3207         rtnl_lock();
3208         err = ixgbevf_init_interrupt_scheme(adapter);
3209         rtnl_unlock();
3210         if (err) {
3211                 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3212                 return err;
3213         }
3214
3215         ixgbevf_reset(adapter);
3216
3217         if (netif_running(netdev)) {
3218                 err = ixgbevf_open(netdev);
3219                 if (err)
3220                         return err;
3221         }
3222
3223         netif_device_attach(netdev);
3224
3225         return err;
3226 }
3227
3228 #endif /* CONFIG_PM */
3229 static void ixgbevf_shutdown(struct pci_dev *pdev)
3230 {
3231         ixgbevf_suspend(pdev, PMSG_SUSPEND);
3232 }
3233
3234 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3235                                                 struct rtnl_link_stats64 *stats)
3236 {
3237         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3238         unsigned int start;
3239         u64 bytes, packets;
3240         const struct ixgbevf_ring *ring;
3241         int i;
3242
3243         ixgbevf_update_stats(adapter);
3244
3245         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3246
3247         for (i = 0; i < adapter->num_rx_queues; i++) {
3248                 ring = &adapter->rx_ring[i];
3249                 do {
3250                         start = u64_stats_fetch_begin_bh(&ring->syncp);
3251                         bytes = ring->total_bytes;
3252                         packets = ring->total_packets;
3253                 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3254                 stats->rx_bytes += bytes;
3255                 stats->rx_packets += packets;
3256         }
3257
3258         for (i = 0; i < adapter->num_tx_queues; i++) {
3259                 ring = &adapter->tx_ring[i];
3260                 do {
3261                         start = u64_stats_fetch_begin_bh(&ring->syncp);
3262                         bytes = ring->total_bytes;
3263                         packets = ring->total_packets;
3264                 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3265                 stats->tx_bytes += bytes;
3266                 stats->tx_packets += packets;
3267         }
3268
3269         return stats;
3270 }
3271
3272 static const struct net_device_ops ixgbevf_netdev_ops = {
3273         .ndo_open               = ixgbevf_open,
3274         .ndo_stop               = ixgbevf_close,
3275         .ndo_start_xmit         = ixgbevf_xmit_frame,
3276         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
3277         .ndo_get_stats64        = ixgbevf_get_stats,
3278         .ndo_validate_addr      = eth_validate_addr,
3279         .ndo_set_mac_address    = ixgbevf_set_mac,
3280         .ndo_change_mtu         = ixgbevf_change_mtu,
3281         .ndo_tx_timeout         = ixgbevf_tx_timeout,
3282         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
3283         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
3284 };
3285
3286 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3287 {
3288         dev->netdev_ops = &ixgbevf_netdev_ops;
3289         ixgbevf_set_ethtool_ops(dev);
3290         dev->watchdog_timeo = 5 * HZ;
3291 }
3292
3293 /**
3294  * ixgbevf_probe - Device Initialization Routine
3295  * @pdev: PCI device information struct
3296  * @ent: entry in ixgbevf_pci_tbl
3297  *
3298  * Returns 0 on success, negative on failure
3299  *
3300  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3301  * The OS initialization, configuring of the adapter private structure,
3302  * and a hardware reset occur.
3303  **/
3304 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3305 {
3306         struct net_device *netdev;
3307         struct ixgbevf_adapter *adapter = NULL;
3308         struct ixgbe_hw *hw = NULL;
3309         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3310         static int cards_found;
3311         int err, pci_using_dac;
3312
3313         err = pci_enable_device(pdev);
3314         if (err)
3315                 return err;
3316
3317         if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
3318             !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
3319                 pci_using_dac = 1;
3320         } else {
3321                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
3322                 if (err) {
3323                         err = dma_set_coherent_mask(&pdev->dev,
3324                                                     DMA_BIT_MASK(32));
3325                         if (err) {
3326                                 dev_err(&pdev->dev, "No usable DMA "
3327                                         "configuration, aborting\n");
3328                                 goto err_dma;
3329                         }
3330                 }
3331                 pci_using_dac = 0;
3332         }
3333
3334         err = pci_request_regions(pdev, ixgbevf_driver_name);
3335         if (err) {
3336                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3337                 goto err_pci_reg;
3338         }
3339
3340         pci_set_master(pdev);
3341
3342         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3343                                    MAX_TX_QUEUES);
3344         if (!netdev) {
3345                 err = -ENOMEM;
3346                 goto err_alloc_etherdev;
3347         }
3348
3349         SET_NETDEV_DEV(netdev, &pdev->dev);
3350
3351         pci_set_drvdata(pdev, netdev);
3352         adapter = netdev_priv(netdev);
3353
3354         adapter->netdev = netdev;
3355         adapter->pdev = pdev;
3356         hw = &adapter->hw;
3357         hw->back = adapter;
3358         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3359
3360         /*
3361          * call save state here in standalone driver because it relies on
3362          * adapter struct to exist, and needs to call netdev_priv
3363          */
3364         pci_save_state(pdev);
3365
3366         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3367                               pci_resource_len(pdev, 0));
3368         if (!hw->hw_addr) {
3369                 err = -EIO;
3370                 goto err_ioremap;
3371         }
3372
3373         ixgbevf_assign_netdev_ops(netdev);
3374
3375         adapter->bd_number = cards_found;
3376
3377         /* Setup hw api */
3378         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3379         hw->mac.type  = ii->mac;
3380
3381         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3382                sizeof(struct ixgbe_mbx_operations));
3383
3384         /* setup the private structure */
3385         err = ixgbevf_sw_init(adapter);
3386         if (err)
3387                 goto err_sw_init;
3388
3389         /* The HW MAC address was set and/or determined in sw_init */
3390         if (!is_valid_ether_addr(netdev->dev_addr)) {
3391                 pr_err("invalid MAC address\n");
3392                 err = -EIO;
3393                 goto err_sw_init;
3394         }
3395
3396         netdev->hw_features = NETIF_F_SG |
3397                            NETIF_F_IP_CSUM |
3398                            NETIF_F_IPV6_CSUM |
3399                            NETIF_F_TSO |
3400                            NETIF_F_TSO6 |
3401                            NETIF_F_RXCSUM;
3402
3403         netdev->features = netdev->hw_features |
3404                            NETIF_F_HW_VLAN_TX |
3405                            NETIF_F_HW_VLAN_RX |
3406                            NETIF_F_HW_VLAN_FILTER;
3407
3408         netdev->vlan_features |= NETIF_F_TSO;
3409         netdev->vlan_features |= NETIF_F_TSO6;
3410         netdev->vlan_features |= NETIF_F_IP_CSUM;
3411         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3412         netdev->vlan_features |= NETIF_F_SG;
3413
3414         if (pci_using_dac)
3415                 netdev->features |= NETIF_F_HIGHDMA;
3416
3417         netdev->priv_flags |= IFF_UNICAST_FLT;
3418
3419         init_timer(&adapter->watchdog_timer);
3420         adapter->watchdog_timer.function = ixgbevf_watchdog;
3421         adapter->watchdog_timer.data = (unsigned long)adapter;
3422
3423         INIT_WORK(&adapter->reset_task, ixgbevf_reset_task);
3424         INIT_WORK(&adapter->watchdog_task, ixgbevf_watchdog_task);
3425
3426         err = ixgbevf_init_interrupt_scheme(adapter);
3427         if (err)
3428                 goto err_sw_init;
3429
3430         strcpy(netdev->name, "eth%d");
3431
3432         err = register_netdev(netdev);
3433         if (err)
3434                 goto err_register;
3435
3436         netif_carrier_off(netdev);
3437
3438         ixgbevf_init_last_counter_stats(adapter);
3439
3440         /* print the MAC address */
3441         hw_dbg(hw, "%pM\n", netdev->dev_addr);
3442
3443         hw_dbg(hw, "MAC: %d\n", hw->mac.type);
3444
3445         hw_dbg(hw, "Intel(R) 82599 Virtual Function\n");
3446         cards_found++;
3447         return 0;
3448
3449 err_register:
3450         ixgbevf_clear_interrupt_scheme(adapter);
3451 err_sw_init:
3452         ixgbevf_reset_interrupt_capability(adapter);
3453         iounmap(hw->hw_addr);
3454 err_ioremap:
3455         free_netdev(netdev);
3456 err_alloc_etherdev:
3457         pci_release_regions(pdev);
3458 err_pci_reg:
3459 err_dma:
3460         pci_disable_device(pdev);
3461         return err;
3462 }
3463
3464 /**
3465  * ixgbevf_remove - Device Removal Routine
3466  * @pdev: PCI device information struct
3467  *
3468  * ixgbevf_remove is called by the PCI subsystem to alert the driver
3469  * that it should release a PCI device.  The could be caused by a
3470  * Hot-Plug event, or because the driver is going to be removed from
3471  * memory.
3472  **/
3473 static void ixgbevf_remove(struct pci_dev *pdev)
3474 {
3475         struct net_device *netdev = pci_get_drvdata(pdev);
3476         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3477
3478         set_bit(__IXGBEVF_DOWN, &adapter->state);
3479
3480         del_timer_sync(&adapter->watchdog_timer);
3481
3482         cancel_work_sync(&adapter->reset_task);
3483         cancel_work_sync(&adapter->watchdog_task);
3484
3485         if (netdev->reg_state == NETREG_REGISTERED)
3486                 unregister_netdev(netdev);
3487
3488         ixgbevf_clear_interrupt_scheme(adapter);
3489         ixgbevf_reset_interrupt_capability(adapter);
3490
3491         iounmap(adapter->hw.hw_addr);
3492         pci_release_regions(pdev);
3493
3494         hw_dbg(&adapter->hw, "Remove complete\n");
3495
3496         kfree(adapter->tx_ring);
3497         kfree(adapter->rx_ring);
3498
3499         free_netdev(netdev);
3500
3501         pci_disable_device(pdev);
3502 }
3503
3504 /**
3505  * ixgbevf_io_error_detected - called when PCI error is detected
3506  * @pdev: Pointer to PCI device
3507  * @state: The current pci connection state
3508  *
3509  * This function is called after a PCI bus error affecting
3510  * this device has been detected.
3511  */
3512 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
3513                                                   pci_channel_state_t state)
3514 {
3515         struct net_device *netdev = pci_get_drvdata(pdev);
3516         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3517
3518         netif_device_detach(netdev);
3519
3520         if (state == pci_channel_io_perm_failure)
3521                 return PCI_ERS_RESULT_DISCONNECT;
3522
3523         if (netif_running(netdev))
3524                 ixgbevf_down(adapter);
3525
3526         pci_disable_device(pdev);
3527
3528         /* Request a slot slot reset. */
3529         return PCI_ERS_RESULT_NEED_RESET;
3530 }
3531
3532 /**
3533  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
3534  * @pdev: Pointer to PCI device
3535  *
3536  * Restart the card from scratch, as if from a cold-boot. Implementation
3537  * resembles the first-half of the ixgbevf_resume routine.
3538  */
3539 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
3540 {
3541         struct net_device *netdev = pci_get_drvdata(pdev);
3542         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3543
3544         if (pci_enable_device_mem(pdev)) {
3545                 dev_err(&pdev->dev,
3546                         "Cannot re-enable PCI device after reset.\n");
3547                 return PCI_ERS_RESULT_DISCONNECT;
3548         }
3549
3550         pci_set_master(pdev);
3551
3552         ixgbevf_reset(adapter);
3553
3554         return PCI_ERS_RESULT_RECOVERED;
3555 }
3556
3557 /**
3558  * ixgbevf_io_resume - called when traffic can start flowing again.
3559  * @pdev: Pointer to PCI device
3560  *
3561  * This callback is called when the error recovery driver tells us that
3562  * its OK to resume normal operation. Implementation resembles the
3563  * second-half of the ixgbevf_resume routine.
3564  */
3565 static void ixgbevf_io_resume(struct pci_dev *pdev)
3566 {
3567         struct net_device *netdev = pci_get_drvdata(pdev);
3568         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3569
3570         if (netif_running(netdev))
3571                 ixgbevf_up(adapter);
3572
3573         netif_device_attach(netdev);
3574 }
3575
3576 /* PCI Error Recovery (ERS) */
3577 static const struct pci_error_handlers ixgbevf_err_handler = {
3578         .error_detected = ixgbevf_io_error_detected,
3579         .slot_reset = ixgbevf_io_slot_reset,
3580         .resume = ixgbevf_io_resume,
3581 };
3582
3583 static struct pci_driver ixgbevf_driver = {
3584         .name     = ixgbevf_driver_name,
3585         .id_table = ixgbevf_pci_tbl,
3586         .probe    = ixgbevf_probe,
3587         .remove   = ixgbevf_remove,
3588 #ifdef CONFIG_PM
3589         /* Power Management Hooks */
3590         .suspend  = ixgbevf_suspend,
3591         .resume   = ixgbevf_resume,
3592 #endif
3593         .shutdown = ixgbevf_shutdown,
3594         .err_handler = &ixgbevf_err_handler
3595 };
3596
3597 /**
3598  * ixgbevf_init_module - Driver Registration Routine
3599  *
3600  * ixgbevf_init_module is the first routine called when the driver is
3601  * loaded. All it does is register with the PCI subsystem.
3602  **/
3603 static int __init ixgbevf_init_module(void)
3604 {
3605         int ret;
3606         pr_info("%s - version %s\n", ixgbevf_driver_string,
3607                 ixgbevf_driver_version);
3608
3609         pr_info("%s\n", ixgbevf_copyright);
3610
3611         ret = pci_register_driver(&ixgbevf_driver);
3612         return ret;
3613 }
3614
3615 module_init(ixgbevf_init_module);
3616
3617 /**
3618  * ixgbevf_exit_module - Driver Exit Cleanup Routine
3619  *
3620  * ixgbevf_exit_module is called just before the driver is removed
3621  * from memory.
3622  **/
3623 static void __exit ixgbevf_exit_module(void)
3624 {
3625         pci_unregister_driver(&ixgbevf_driver);
3626 }
3627
3628 #ifdef DEBUG
3629 /**
3630  * ixgbevf_get_hw_dev_name - return device name string
3631  * used by hardware layer to print debugging information
3632  **/
3633 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
3634 {
3635         struct ixgbevf_adapter *adapter = hw->back;
3636         return adapter->netdev->name;
3637 }
3638
3639 #endif
3640 module_exit(ixgbevf_exit_module);
3641
3642 /* ixgbevf_main.c */