Merge branch 'devel-stable' of master.kernel.org:/home/rmk/linux-2.6-arm
[pandora-kernel.git] / drivers / net / ixgbe / ixgbe_main.c
1 /*******************************************************************************
2
3   Intel 10 Gigabit PCI Express Linux driver
4   Copyright(c) 1999 - 2010 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 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
34 #include <linux/in.h>
35 #include <linux/ip.h>
36 #include <linux/tcp.h>
37 #include <linux/pkt_sched.h>
38 #include <linux/ipv6.h>
39 #include <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <scsi/fc/fc_fcoe.h>
45
46 #include "ixgbe.h"
47 #include "ixgbe_common.h"
48 #include "ixgbe_dcb_82599.h"
49 #include "ixgbe_sriov.h"
50
51 char ixgbe_driver_name[] = "ixgbe";
52 static const char ixgbe_driver_string[] =
53                               "Intel(R) 10 Gigabit PCI Express Network Driver";
54
55 #define DRV_VERSION "3.0.12-k2"
56 const char ixgbe_driver_version[] = DRV_VERSION;
57 static char ixgbe_copyright[] = "Copyright (c) 1999-2010 Intel Corporation.";
58
59 static const struct ixgbe_info *ixgbe_info_tbl[] = {
60         [board_82598] = &ixgbe_82598_info,
61         [board_82599] = &ixgbe_82599_info,
62         [board_X540] = &ixgbe_X540_info,
63 };
64
65 /* ixgbe_pci_tbl - PCI Device ID Table
66  *
67  * Wildcard entries (PCI_ANY_ID) should come last
68  * Last entry must be all 0s
69  *
70  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
71  *   Class, Class Mask, private data (not used) }
72  */
73 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
74         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
75          board_82598 },
76         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
77          board_82598 },
78         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
79          board_82598 },
80         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
81          board_82598 },
82         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
83          board_82598 },
84         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
85          board_82598 },
86         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
87          board_82598 },
88         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
89          board_82598 },
90         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
91          board_82598 },
92         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
93          board_82598 },
94         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
95          board_82598 },
96         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
97          board_82598 },
98         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
99          board_82599 },
100         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
101          board_82599 },
102         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
103          board_82599 },
104         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
105          board_82599 },
106         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
107          board_82599 },
108         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
109          board_82599 },
110         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
111          board_82599 },
112         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE),
113          board_82599 },
114         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE),
115          board_82599 },
116         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM),
117          board_82599 },
118         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
119          board_82599 },
120         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T),
121          board_X540 },
122
123         /* required last entry */
124         {0, }
125 };
126 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
127
128 #ifdef CONFIG_IXGBE_DCA
129 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
130                             void *p);
131 static struct notifier_block dca_notifier = {
132         .notifier_call = ixgbe_notify_dca,
133         .next          = NULL,
134         .priority      = 0
135 };
136 #endif
137
138 #ifdef CONFIG_PCI_IOV
139 static unsigned int max_vfs;
140 module_param(max_vfs, uint, 0);
141 MODULE_PARM_DESC(max_vfs,
142                  "Maximum number of virtual functions to allocate per physical function");
143 #endif /* CONFIG_PCI_IOV */
144
145 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
146 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
147 MODULE_LICENSE("GPL");
148 MODULE_VERSION(DRV_VERSION);
149
150 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
151
152 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
153 {
154         struct ixgbe_hw *hw = &adapter->hw;
155         u32 gcr;
156         u32 gpie;
157         u32 vmdctl;
158
159 #ifdef CONFIG_PCI_IOV
160         /* disable iov and allow time for transactions to clear */
161         pci_disable_sriov(adapter->pdev);
162 #endif
163
164         /* turn off device IOV mode */
165         gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
166         gcr &= ~(IXGBE_GCR_EXT_SRIOV);
167         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
168         gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
169         gpie &= ~IXGBE_GPIE_VTMODE_MASK;
170         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
171
172         /* set default pool back to 0 */
173         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
174         vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
175         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
176
177         /* take a breather then clean up driver data */
178         msleep(100);
179
180         kfree(adapter->vfinfo);
181         adapter->vfinfo = NULL;
182
183         adapter->num_vfs = 0;
184         adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
185 }
186
187 struct ixgbe_reg_info {
188         u32 ofs;
189         char *name;
190 };
191
192 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
193
194         /* General Registers */
195         {IXGBE_CTRL, "CTRL"},
196         {IXGBE_STATUS, "STATUS"},
197         {IXGBE_CTRL_EXT, "CTRL_EXT"},
198
199         /* Interrupt Registers */
200         {IXGBE_EICR, "EICR"},
201
202         /* RX Registers */
203         {IXGBE_SRRCTL(0), "SRRCTL"},
204         {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
205         {IXGBE_RDLEN(0), "RDLEN"},
206         {IXGBE_RDH(0), "RDH"},
207         {IXGBE_RDT(0), "RDT"},
208         {IXGBE_RXDCTL(0), "RXDCTL"},
209         {IXGBE_RDBAL(0), "RDBAL"},
210         {IXGBE_RDBAH(0), "RDBAH"},
211
212         /* TX Registers */
213         {IXGBE_TDBAL(0), "TDBAL"},
214         {IXGBE_TDBAH(0), "TDBAH"},
215         {IXGBE_TDLEN(0), "TDLEN"},
216         {IXGBE_TDH(0), "TDH"},
217         {IXGBE_TDT(0), "TDT"},
218         {IXGBE_TXDCTL(0), "TXDCTL"},
219
220         /* List Terminator */
221         {}
222 };
223
224
225 /*
226  * ixgbe_regdump - register printout routine
227  */
228 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
229 {
230         int i = 0, j = 0;
231         char rname[16];
232         u32 regs[64];
233
234         switch (reginfo->ofs) {
235         case IXGBE_SRRCTL(0):
236                 for (i = 0; i < 64; i++)
237                         regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
238                 break;
239         case IXGBE_DCA_RXCTRL(0):
240                 for (i = 0; i < 64; i++)
241                         regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
242                 break;
243         case IXGBE_RDLEN(0):
244                 for (i = 0; i < 64; i++)
245                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
246                 break;
247         case IXGBE_RDH(0):
248                 for (i = 0; i < 64; i++)
249                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
250                 break;
251         case IXGBE_RDT(0):
252                 for (i = 0; i < 64; i++)
253                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
254                 break;
255         case IXGBE_RXDCTL(0):
256                 for (i = 0; i < 64; i++)
257                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
258                 break;
259         case IXGBE_RDBAL(0):
260                 for (i = 0; i < 64; i++)
261                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
262                 break;
263         case IXGBE_RDBAH(0):
264                 for (i = 0; i < 64; i++)
265                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
266                 break;
267         case IXGBE_TDBAL(0):
268                 for (i = 0; i < 64; i++)
269                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
270                 break;
271         case IXGBE_TDBAH(0):
272                 for (i = 0; i < 64; i++)
273                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
274                 break;
275         case IXGBE_TDLEN(0):
276                 for (i = 0; i < 64; i++)
277                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
278                 break;
279         case IXGBE_TDH(0):
280                 for (i = 0; i < 64; i++)
281                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
282                 break;
283         case IXGBE_TDT(0):
284                 for (i = 0; i < 64; i++)
285                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
286                 break;
287         case IXGBE_TXDCTL(0):
288                 for (i = 0; i < 64; i++)
289                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
290                 break;
291         default:
292                 pr_info("%-15s %08x\n", reginfo->name,
293                         IXGBE_READ_REG(hw, reginfo->ofs));
294                 return;
295         }
296
297         for (i = 0; i < 8; i++) {
298                 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
299                 pr_err("%-15s", rname);
300                 for (j = 0; j < 8; j++)
301                         pr_cont(" %08x", regs[i*8+j]);
302                 pr_cont("\n");
303         }
304
305 }
306
307 /*
308  * ixgbe_dump - Print registers, tx-rings and rx-rings
309  */
310 static void ixgbe_dump(struct ixgbe_adapter *adapter)
311 {
312         struct net_device *netdev = adapter->netdev;
313         struct ixgbe_hw *hw = &adapter->hw;
314         struct ixgbe_reg_info *reginfo;
315         int n = 0;
316         struct ixgbe_ring *tx_ring;
317         struct ixgbe_tx_buffer *tx_buffer_info;
318         union ixgbe_adv_tx_desc *tx_desc;
319         struct my_u0 { u64 a; u64 b; } *u0;
320         struct ixgbe_ring *rx_ring;
321         union ixgbe_adv_rx_desc *rx_desc;
322         struct ixgbe_rx_buffer *rx_buffer_info;
323         u32 staterr;
324         int i = 0;
325
326         if (!netif_msg_hw(adapter))
327                 return;
328
329         /* Print netdevice Info */
330         if (netdev) {
331                 dev_info(&adapter->pdev->dev, "Net device Info\n");
332                 pr_info("Device Name     state            "
333                         "trans_start      last_rx\n");
334                 pr_info("%-15s %016lX %016lX %016lX\n",
335                         netdev->name,
336                         netdev->state,
337                         netdev->trans_start,
338                         netdev->last_rx);
339         }
340
341         /* Print Registers */
342         dev_info(&adapter->pdev->dev, "Register Dump\n");
343         pr_info(" Register Name   Value\n");
344         for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
345              reginfo->name; reginfo++) {
346                 ixgbe_regdump(hw, reginfo);
347         }
348
349         /* Print TX Ring Summary */
350         if (!netdev || !netif_running(netdev))
351                 goto exit;
352
353         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
354         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
355         for (n = 0; n < adapter->num_tx_queues; n++) {
356                 tx_ring = adapter->tx_ring[n];
357                 tx_buffer_info =
358                         &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
359                 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
360                            n, tx_ring->next_to_use, tx_ring->next_to_clean,
361                            (u64)tx_buffer_info->dma,
362                            tx_buffer_info->length,
363                            tx_buffer_info->next_to_watch,
364                            (u64)tx_buffer_info->time_stamp);
365         }
366
367         /* Print TX Rings */
368         if (!netif_msg_tx_done(adapter))
369                 goto rx_ring_summary;
370
371         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
372
373         /* Transmit Descriptor Formats
374          *
375          * Advanced Transmit Descriptor
376          *   +--------------------------------------------------------------+
377          * 0 |         Buffer Address [63:0]                                |
378          *   +--------------------------------------------------------------+
379          * 8 |  PAYLEN  | PORTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
380          *   +--------------------------------------------------------------+
381          *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
382          */
383
384         for (n = 0; n < adapter->num_tx_queues; n++) {
385                 tx_ring = adapter->tx_ring[n];
386                 pr_info("------------------------------------\n");
387                 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
388                 pr_info("------------------------------------\n");
389                 pr_info("T [desc]     [address 63:0  ] "
390                         "[PlPOIdStDDt Ln] [bi->dma       ] "
391                         "leng  ntw timestamp        bi->skb\n");
392
393                 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
394                         tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
395                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
396                         u0 = (struct my_u0 *)tx_desc;
397                         pr_info("T [0x%03X]    %016llX %016llX %016llX"
398                                 " %04X  %3X %016llX %p", i,
399                                 le64_to_cpu(u0->a),
400                                 le64_to_cpu(u0->b),
401                                 (u64)tx_buffer_info->dma,
402                                 tx_buffer_info->length,
403                                 tx_buffer_info->next_to_watch,
404                                 (u64)tx_buffer_info->time_stamp,
405                                 tx_buffer_info->skb);
406                         if (i == tx_ring->next_to_use &&
407                                 i == tx_ring->next_to_clean)
408                                 pr_cont(" NTC/U\n");
409                         else if (i == tx_ring->next_to_use)
410                                 pr_cont(" NTU\n");
411                         else if (i == tx_ring->next_to_clean)
412                                 pr_cont(" NTC\n");
413                         else
414                                 pr_cont("\n");
415
416                         if (netif_msg_pktdata(adapter) &&
417                                 tx_buffer_info->dma != 0)
418                                 print_hex_dump(KERN_INFO, "",
419                                         DUMP_PREFIX_ADDRESS, 16, 1,
420                                         phys_to_virt(tx_buffer_info->dma),
421                                         tx_buffer_info->length, true);
422                 }
423         }
424
425         /* Print RX Rings Summary */
426 rx_ring_summary:
427         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
428         pr_info("Queue [NTU] [NTC]\n");
429         for (n = 0; n < adapter->num_rx_queues; n++) {
430                 rx_ring = adapter->rx_ring[n];
431                 pr_info("%5d %5X %5X\n",
432                         n, rx_ring->next_to_use, rx_ring->next_to_clean);
433         }
434
435         /* Print RX Rings */
436         if (!netif_msg_rx_status(adapter))
437                 goto exit;
438
439         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
440
441         /* Advanced Receive Descriptor (Read) Format
442          *    63                                           1        0
443          *    +-----------------------------------------------------+
444          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
445          *    +----------------------------------------------+------+
446          *  8 |       Header Buffer Address [63:1]           |  DD  |
447          *    +-----------------------------------------------------+
448          *
449          *
450          * Advanced Receive Descriptor (Write-Back) Format
451          *
452          *   63       48 47    32 31  30      21 20 16 15   4 3     0
453          *   +------------------------------------------------------+
454          * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
455          *   | Checksum   Ident  |   |           |    | Type | Type |
456          *   +------------------------------------------------------+
457          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
458          *   +------------------------------------------------------+
459          *   63       48 47    32 31            20 19               0
460          */
461         for (n = 0; n < adapter->num_rx_queues; n++) {
462                 rx_ring = adapter->rx_ring[n];
463                 pr_info("------------------------------------\n");
464                 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
465                 pr_info("------------------------------------\n");
466                 pr_info("R  [desc]      [ PktBuf     A0] "
467                         "[  HeadBuf   DD] [bi->dma       ] [bi->skb] "
468                         "<-- Adv Rx Read format\n");
469                 pr_info("RWB[desc]      [PcsmIpSHl PtRs] "
470                         "[vl er S cks ln] ---------------- [bi->skb] "
471                         "<-- Adv Rx Write-Back format\n");
472
473                 for (i = 0; i < rx_ring->count; i++) {
474                         rx_buffer_info = &rx_ring->rx_buffer_info[i];
475                         rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
476                         u0 = (struct my_u0 *)rx_desc;
477                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
478                         if (staterr & IXGBE_RXD_STAT_DD) {
479                                 /* Descriptor Done */
480                                 pr_info("RWB[0x%03X]     %016llX "
481                                         "%016llX ---------------- %p", i,
482                                         le64_to_cpu(u0->a),
483                                         le64_to_cpu(u0->b),
484                                         rx_buffer_info->skb);
485                         } else {
486                                 pr_info("R  [0x%03X]     %016llX "
487                                         "%016llX %016llX %p", i,
488                                         le64_to_cpu(u0->a),
489                                         le64_to_cpu(u0->b),
490                                         (u64)rx_buffer_info->dma,
491                                         rx_buffer_info->skb);
492
493                                 if (netif_msg_pktdata(adapter)) {
494                                         print_hex_dump(KERN_INFO, "",
495                                            DUMP_PREFIX_ADDRESS, 16, 1,
496                                            phys_to_virt(rx_buffer_info->dma),
497                                            rx_ring->rx_buf_len, true);
498
499                                         if (rx_ring->rx_buf_len
500                                                 < IXGBE_RXBUFFER_2048)
501                                                 print_hex_dump(KERN_INFO, "",
502                                                   DUMP_PREFIX_ADDRESS, 16, 1,
503                                                   phys_to_virt(
504                                                     rx_buffer_info->page_dma +
505                                                     rx_buffer_info->page_offset
506                                                   ),
507                                                   PAGE_SIZE/2, true);
508                                 }
509                         }
510
511                         if (i == rx_ring->next_to_use)
512                                 pr_cont(" NTU\n");
513                         else if (i == rx_ring->next_to_clean)
514                                 pr_cont(" NTC\n");
515                         else
516                                 pr_cont("\n");
517
518                 }
519         }
520
521 exit:
522         return;
523 }
524
525 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
526 {
527         u32 ctrl_ext;
528
529         /* Let firmware take over control of h/w */
530         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
531         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
532                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
533 }
534
535 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
536 {
537         u32 ctrl_ext;
538
539         /* Let firmware know the driver has taken over */
540         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
541         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
542                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
543 }
544
545 /*
546  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
547  * @adapter: pointer to adapter struct
548  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
549  * @queue: queue to map the corresponding interrupt to
550  * @msix_vector: the vector to map to the corresponding queue
551  *
552  */
553 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
554                            u8 queue, u8 msix_vector)
555 {
556         u32 ivar, index;
557         struct ixgbe_hw *hw = &adapter->hw;
558         switch (hw->mac.type) {
559         case ixgbe_mac_82598EB:
560                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
561                 if (direction == -1)
562                         direction = 0;
563                 index = (((direction * 64) + queue) >> 2) & 0x1F;
564                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
565                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
566                 ivar |= (msix_vector << (8 * (queue & 0x3)));
567                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
568                 break;
569         case ixgbe_mac_82599EB:
570         case ixgbe_mac_X540:
571                 if (direction == -1) {
572                         /* other causes */
573                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
574                         index = ((queue & 1) * 8);
575                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
576                         ivar &= ~(0xFF << index);
577                         ivar |= (msix_vector << index);
578                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
579                         break;
580                 } else {
581                         /* tx or rx causes */
582                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
583                         index = ((16 * (queue & 1)) + (8 * direction));
584                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
585                         ivar &= ~(0xFF << index);
586                         ivar |= (msix_vector << index);
587                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
588                         break;
589                 }
590         default:
591                 break;
592         }
593 }
594
595 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
596                                           u64 qmask)
597 {
598         u32 mask;
599
600         switch (adapter->hw.mac.type) {
601         case ixgbe_mac_82598EB:
602                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
603                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
604                 break;
605         case ixgbe_mac_82599EB:
606         case ixgbe_mac_X540:
607                 mask = (qmask & 0xFFFFFFFF);
608                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
609                 mask = (qmask >> 32);
610                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
611                 break;
612         default:
613                 break;
614         }
615 }
616
617 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
618                                       struct ixgbe_tx_buffer *tx_buffer_info)
619 {
620         if (tx_buffer_info->dma) {
621                 if (tx_buffer_info->mapped_as_page)
622                         dma_unmap_page(tx_ring->dev,
623                                        tx_buffer_info->dma,
624                                        tx_buffer_info->length,
625                                        DMA_TO_DEVICE);
626                 else
627                         dma_unmap_single(tx_ring->dev,
628                                          tx_buffer_info->dma,
629                                          tx_buffer_info->length,
630                                          DMA_TO_DEVICE);
631                 tx_buffer_info->dma = 0;
632         }
633         if (tx_buffer_info->skb) {
634                 dev_kfree_skb_any(tx_buffer_info->skb);
635                 tx_buffer_info->skb = NULL;
636         }
637         tx_buffer_info->time_stamp = 0;
638         /* tx_buffer_info must be completely set up in the transmit path */
639 }
640
641 /**
642  * ixgbe_dcb_txq_to_tc - convert a reg index to a traffic class
643  * @adapter: driver private struct
644  * @index: reg idx of queue to query (0-127)
645  *
646  * Helper function to determine the traffic index for a paticular
647  * register index.
648  *
649  * Returns : a tc index for use in range 0-7, or 0-3
650  */
651 u8 ixgbe_dcb_txq_to_tc(struct ixgbe_adapter *adapter, u8 reg_idx)
652 {
653         int tc = -1;
654         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
655
656         /* if DCB is not enabled the queues have no TC */
657         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
658                 return tc;
659
660         /* check valid range */
661         if (reg_idx >= adapter->hw.mac.max_tx_queues)
662                 return tc;
663
664         switch (adapter->hw.mac.type) {
665         case ixgbe_mac_82598EB:
666                 tc = reg_idx >> 2;
667                 break;
668         default:
669                 if (dcb_i != 4 && dcb_i != 8)
670                         break;
671
672                 /* if VMDq is enabled the lowest order bits determine TC */
673                 if (adapter->flags & (IXGBE_FLAG_SRIOV_ENABLED |
674                                       IXGBE_FLAG_VMDQ_ENABLED)) {
675                         tc = reg_idx & (dcb_i - 1);
676                         break;
677                 }
678
679                 /*
680                  * Convert the reg_idx into the correct TC. This bitmask
681                  * targets the last full 32 ring traffic class and assigns
682                  * it a value of 1. From there the rest of the rings are
683                  * based on shifting the mask further up to include the
684                  * reg_idx / 16 and then reg_idx / 8. It assumes dcB_i
685                  * will only ever be 8 or 4 and that reg_idx will never
686                  * be greater then 128. The code without the power of 2
687                  * optimizations would be:
688                  * (((reg_idx % 32) + 32) * dcb_i) >> (9 - reg_idx / 32)
689                  */
690                 tc = ((reg_idx & 0X1F) + 0x20) * dcb_i;
691                 tc >>= 9 - (reg_idx >> 5);
692         }
693
694         return tc;
695 }
696
697 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
698 {
699         struct ixgbe_hw *hw = &adapter->hw;
700         struct ixgbe_hw_stats *hwstats = &adapter->stats;
701         u32 data = 0;
702         u32 xoff[8] = {0};
703         int i;
704
705         if ((hw->fc.current_mode == ixgbe_fc_full) ||
706             (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
707                 switch (hw->mac.type) {
708                 case ixgbe_mac_82598EB:
709                         data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
710                         break;
711                 default:
712                         data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
713                 }
714                 hwstats->lxoffrxc += data;
715
716                 /* refill credits (no tx hang) if we received xoff */
717                 if (!data)
718                         return;
719
720                 for (i = 0; i < adapter->num_tx_queues; i++)
721                         clear_bit(__IXGBE_HANG_CHECK_ARMED,
722                                   &adapter->tx_ring[i]->state);
723                 return;
724         } else if (!(adapter->dcb_cfg.pfc_mode_enable))
725                 return;
726
727         /* update stats for each tc, only valid with PFC enabled */
728         for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
729                 switch (hw->mac.type) {
730                 case ixgbe_mac_82598EB:
731                         xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
732                         break;
733                 default:
734                         xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
735                 }
736                 hwstats->pxoffrxc[i] += xoff[i];
737         }
738
739         /* disarm tx queues that have received xoff frames */
740         for (i = 0; i < adapter->num_tx_queues; i++) {
741                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
742                 u32 tc = ixgbe_dcb_txq_to_tc(adapter, tx_ring->reg_idx);
743
744                 if (xoff[tc])
745                         clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
746         }
747 }
748
749 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
750 {
751         return ring->tx_stats.completed;
752 }
753
754 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
755 {
756         struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
757         struct ixgbe_hw *hw = &adapter->hw;
758
759         u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
760         u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
761
762         if (head != tail)
763                 return (head < tail) ?
764                         tail - head : (tail + ring->count - head);
765
766         return 0;
767 }
768
769 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
770 {
771         u32 tx_done = ixgbe_get_tx_completed(tx_ring);
772         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
773         u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
774         bool ret = false;
775
776         clear_check_for_tx_hang(tx_ring);
777
778         /*
779          * Check for a hung queue, but be thorough. This verifies
780          * that a transmit has been completed since the previous
781          * check AND there is at least one packet pending. The
782          * ARMED bit is set to indicate a potential hang. The
783          * bit is cleared if a pause frame is received to remove
784          * false hang detection due to PFC or 802.3x frames. By
785          * requiring this to fail twice we avoid races with
786          * pfc clearing the ARMED bit and conditions where we
787          * run the check_tx_hang logic with a transmit completion
788          * pending but without time to complete it yet.
789          */
790         if ((tx_done_old == tx_done) && tx_pending) {
791                 /* make sure it is true for two checks in a row */
792                 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
793                                        &tx_ring->state);
794         } else {
795                 /* update completed stats and continue */
796                 tx_ring->tx_stats.tx_done_old = tx_done;
797                 /* reset the countdown */
798                 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
799         }
800
801         return ret;
802 }
803
804 #define IXGBE_MAX_TXD_PWR       14
805 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
806
807 /* Tx Descriptors needed, worst case */
808 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
809                          (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
810 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
811         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
812
813 static void ixgbe_tx_timeout(struct net_device *netdev);
814
815 /**
816  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
817  * @q_vector: structure containing interrupt and ring information
818  * @tx_ring: tx ring to clean
819  **/
820 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
821                                struct ixgbe_ring *tx_ring)
822 {
823         struct ixgbe_adapter *adapter = q_vector->adapter;
824         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
825         struct ixgbe_tx_buffer *tx_buffer_info;
826         unsigned int total_bytes = 0, total_packets = 0;
827         u16 i, eop, count = 0;
828
829         i = tx_ring->next_to_clean;
830         eop = tx_ring->tx_buffer_info[i].next_to_watch;
831         eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
832
833         while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
834                (count < tx_ring->work_limit)) {
835                 bool cleaned = false;
836                 rmb(); /* read buffer_info after eop_desc */
837                 for ( ; !cleaned; count++) {
838                         tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
839                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
840
841                         tx_desc->wb.status = 0;
842                         cleaned = (i == eop);
843
844                         i++;
845                         if (i == tx_ring->count)
846                                 i = 0;
847
848                         if (cleaned && tx_buffer_info->skb) {
849                                 total_bytes += tx_buffer_info->bytecount;
850                                 total_packets += tx_buffer_info->gso_segs;
851                         }
852
853                         ixgbe_unmap_and_free_tx_resource(tx_ring,
854                                                          tx_buffer_info);
855                 }
856
857                 tx_ring->tx_stats.completed++;
858                 eop = tx_ring->tx_buffer_info[i].next_to_watch;
859                 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
860         }
861
862         tx_ring->next_to_clean = i;
863         tx_ring->total_bytes += total_bytes;
864         tx_ring->total_packets += total_packets;
865         u64_stats_update_begin(&tx_ring->syncp);
866         tx_ring->stats.packets += total_packets;
867         tx_ring->stats.bytes += total_bytes;
868         u64_stats_update_end(&tx_ring->syncp);
869
870         if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
871                 /* schedule immediate reset if we believe we hung */
872                 struct ixgbe_hw *hw = &adapter->hw;
873                 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
874                 e_err(drv, "Detected Tx Unit Hang\n"
875                         "  Tx Queue             <%d>\n"
876                         "  TDH, TDT             <%x>, <%x>\n"
877                         "  next_to_use          <%x>\n"
878                         "  next_to_clean        <%x>\n"
879                         "tx_buffer_info[next_to_clean]\n"
880                         "  time_stamp           <%lx>\n"
881                         "  jiffies              <%lx>\n",
882                         tx_ring->queue_index,
883                         IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
884                         IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
885                         tx_ring->next_to_use, eop,
886                         tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
887
888                 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
889
890                 e_info(probe,
891                        "tx hang %d detected on queue %d, resetting adapter\n",
892                         adapter->tx_timeout_count + 1, tx_ring->queue_index);
893
894                 /* schedule immediate reset if we believe we hung */
895                 ixgbe_tx_timeout(adapter->netdev);
896
897                 /* the adapter is about to reset, no point in enabling stuff */
898                 return true;
899         }
900
901 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
902         if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
903                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
904                 /* Make sure that anybody stopping the queue after this
905                  * sees the new next_to_clean.
906                  */
907                 smp_mb();
908                 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
909                     !test_bit(__IXGBE_DOWN, &adapter->state)) {
910                         netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
911                         ++tx_ring->tx_stats.restart_queue;
912                 }
913         }
914
915         return count < tx_ring->work_limit;
916 }
917
918 #ifdef CONFIG_IXGBE_DCA
919 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
920                                 struct ixgbe_ring *rx_ring,
921                                 int cpu)
922 {
923         struct ixgbe_hw *hw = &adapter->hw;
924         u32 rxctrl;
925         u8 reg_idx = rx_ring->reg_idx;
926
927         rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
928         switch (hw->mac.type) {
929         case ixgbe_mac_82598EB:
930                 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
931                 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
932                 break;
933         case ixgbe_mac_82599EB:
934         case ixgbe_mac_X540:
935                 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
936                 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
937                            IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
938                 break;
939         default:
940                 break;
941         }
942         rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
943         rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
944         rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
945         rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
946                     IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
947         IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
948 }
949
950 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
951                                 struct ixgbe_ring *tx_ring,
952                                 int cpu)
953 {
954         struct ixgbe_hw *hw = &adapter->hw;
955         u32 txctrl;
956         u8 reg_idx = tx_ring->reg_idx;
957
958         switch (hw->mac.type) {
959         case ixgbe_mac_82598EB:
960                 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
961                 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
962                 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
963                 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
964                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
965                 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
966                 break;
967         case ixgbe_mac_82599EB:
968         case ixgbe_mac_X540:
969                 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
970                 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
971                 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
972                            IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
973                 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
974                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
975                 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
976                 break;
977         default:
978                 break;
979         }
980 }
981
982 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
983 {
984         struct ixgbe_adapter *adapter = q_vector->adapter;
985         int cpu = get_cpu();
986         long r_idx;
987         int i;
988
989         if (q_vector->cpu == cpu)
990                 goto out_no_update;
991
992         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
993         for (i = 0; i < q_vector->txr_count; i++) {
994                 ixgbe_update_tx_dca(adapter, adapter->tx_ring[r_idx], cpu);
995                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
996                                       r_idx + 1);
997         }
998
999         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1000         for (i = 0; i < q_vector->rxr_count; i++) {
1001                 ixgbe_update_rx_dca(adapter, adapter->rx_ring[r_idx], cpu);
1002                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1003                                       r_idx + 1);
1004         }
1005
1006         q_vector->cpu = cpu;
1007 out_no_update:
1008         put_cpu();
1009 }
1010
1011 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1012 {
1013         int num_q_vectors;
1014         int i;
1015
1016         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1017                 return;
1018
1019         /* always use CB2 mode, difference is masked in the CB driver */
1020         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1021
1022         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
1023                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1024         else
1025                 num_q_vectors = 1;
1026
1027         for (i = 0; i < num_q_vectors; i++) {
1028                 adapter->q_vector[i]->cpu = -1;
1029                 ixgbe_update_dca(adapter->q_vector[i]);
1030         }
1031 }
1032
1033 static int __ixgbe_notify_dca(struct device *dev, void *data)
1034 {
1035         struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1036         unsigned long event = *(unsigned long *)data;
1037
1038         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1039                 return 0;
1040
1041         switch (event) {
1042         case DCA_PROVIDER_ADD:
1043                 /* if we're already enabled, don't do it again */
1044                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1045                         break;
1046                 if (dca_add_requester(dev) == 0) {
1047                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1048                         ixgbe_setup_dca(adapter);
1049                         break;
1050                 }
1051                 /* Fall Through since DCA is disabled. */
1052         case DCA_PROVIDER_REMOVE:
1053                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1054                         dca_remove_requester(dev);
1055                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1056                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1057                 }
1058                 break;
1059         }
1060
1061         return 0;
1062 }
1063
1064 #endif /* CONFIG_IXGBE_DCA */
1065 /**
1066  * ixgbe_receive_skb - Send a completed packet up the stack
1067  * @adapter: board private structure
1068  * @skb: packet to send up
1069  * @status: hardware indication of status of receive
1070  * @rx_ring: rx descriptor ring (for a specific queue) to setup
1071  * @rx_desc: rx descriptor
1072  **/
1073 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
1074                               struct sk_buff *skb, u8 status,
1075                               struct ixgbe_ring *ring,
1076                               union ixgbe_adv_rx_desc *rx_desc)
1077 {
1078         struct ixgbe_adapter *adapter = q_vector->adapter;
1079         struct napi_struct *napi = &q_vector->napi;
1080         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
1081         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
1082
1083         if (is_vlan && (tag & VLAN_VID_MASK))
1084                 __vlan_hwaccel_put_tag(skb, tag);
1085
1086         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1087                 napi_gro_receive(napi, skb);
1088         else
1089                 netif_rx(skb);
1090 }
1091
1092 /**
1093  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1094  * @adapter: address of board private structure
1095  * @status_err: hardware indication of status of receive
1096  * @skb: skb currently being received and modified
1097  **/
1098 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
1099                                      union ixgbe_adv_rx_desc *rx_desc,
1100                                      struct sk_buff *skb)
1101 {
1102         u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
1103
1104         skb_checksum_none_assert(skb);
1105
1106         /* Rx csum disabled */
1107         if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1108                 return;
1109
1110         /* if IP and error */
1111         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1112             (status_err & IXGBE_RXDADV_ERR_IPE)) {
1113                 adapter->hw_csum_rx_error++;
1114                 return;
1115         }
1116
1117         if (!(status_err & IXGBE_RXD_STAT_L4CS))
1118                 return;
1119
1120         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1121                 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1122
1123                 /*
1124                  * 82599 errata, UDP frames with a 0 checksum can be marked as
1125                  * checksum errors.
1126                  */
1127                 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1128                     (adapter->hw.mac.type == ixgbe_mac_82599EB))
1129                         return;
1130
1131                 adapter->hw_csum_rx_error++;
1132                 return;
1133         }
1134
1135         /* It must be a TCP or UDP packet with a valid checksum */
1136         skb->ip_summed = CHECKSUM_UNNECESSARY;
1137 }
1138
1139 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1140 {
1141         /*
1142          * Force memory writes to complete before letting h/w
1143          * know there are new descriptors to fetch.  (Only
1144          * applicable for weak-ordered memory model archs,
1145          * such as IA-64).
1146          */
1147         wmb();
1148         writel(val, rx_ring->tail);
1149 }
1150
1151 /**
1152  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1153  * @rx_ring: ring to place buffers on
1154  * @cleaned_count: number of buffers to replace
1155  **/
1156 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1157 {
1158         union ixgbe_adv_rx_desc *rx_desc;
1159         struct ixgbe_rx_buffer *bi;
1160         struct sk_buff *skb;
1161         u16 i = rx_ring->next_to_use;
1162
1163         /* do nothing if no valid netdev defined */
1164         if (!rx_ring->netdev)
1165                 return;
1166
1167         while (cleaned_count--) {
1168                 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1169                 bi = &rx_ring->rx_buffer_info[i];
1170                 skb = bi->skb;
1171
1172                 if (!skb) {
1173                         skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1174                                                         rx_ring->rx_buf_len);
1175                         if (!skb) {
1176                                 rx_ring->rx_stats.alloc_rx_buff_failed++;
1177                                 goto no_buffers;
1178                         }
1179                         /* initialize queue mapping */
1180                         skb_record_rx_queue(skb, rx_ring->queue_index);
1181                         bi->skb = skb;
1182                 }
1183
1184                 if (!bi->dma) {
1185                         bi->dma = dma_map_single(rx_ring->dev,
1186                                                  skb->data,
1187                                                  rx_ring->rx_buf_len,
1188                                                  DMA_FROM_DEVICE);
1189                         if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1190                                 rx_ring->rx_stats.alloc_rx_buff_failed++;
1191                                 bi->dma = 0;
1192                                 goto no_buffers;
1193                         }
1194                 }
1195
1196                 if (ring_is_ps_enabled(rx_ring)) {
1197                         if (!bi->page) {
1198                                 bi->page = netdev_alloc_page(rx_ring->netdev);
1199                                 if (!bi->page) {
1200                                         rx_ring->rx_stats.alloc_rx_page_failed++;
1201                                         goto no_buffers;
1202                                 }
1203                         }
1204
1205                         if (!bi->page_dma) {
1206                                 /* use a half page if we're re-using */
1207                                 bi->page_offset ^= PAGE_SIZE / 2;
1208                                 bi->page_dma = dma_map_page(rx_ring->dev,
1209                                                             bi->page,
1210                                                             bi->page_offset,
1211                                                             PAGE_SIZE / 2,
1212                                                             DMA_FROM_DEVICE);
1213                                 if (dma_mapping_error(rx_ring->dev,
1214                                                       bi->page_dma)) {
1215                                         rx_ring->rx_stats.alloc_rx_page_failed++;
1216                                         bi->page_dma = 0;
1217                                         goto no_buffers;
1218                                 }
1219                         }
1220
1221                         /* Refresh the desc even if buffer_addrs didn't change
1222                          * because each write-back erases this info. */
1223                         rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1224                         rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1225                 } else {
1226                         rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1227                         rx_desc->read.hdr_addr = 0;
1228                 }
1229
1230                 i++;
1231                 if (i == rx_ring->count)
1232                         i = 0;
1233         }
1234
1235 no_buffers:
1236         if (rx_ring->next_to_use != i) {
1237                 rx_ring->next_to_use = i;
1238                 ixgbe_release_rx_desc(rx_ring, i);
1239         }
1240 }
1241
1242 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1243 {
1244         /* HW will not DMA in data larger than the given buffer, even if it
1245          * parses the (NFS, of course) header to be larger.  In that case, it
1246          * fills the header buffer and spills the rest into the page.
1247          */
1248         u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1249         u16 hlen = (hdr_info &  IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1250                     IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1251         if (hlen > IXGBE_RX_HDR_SIZE)
1252                 hlen = IXGBE_RX_HDR_SIZE;
1253         return hlen;
1254 }
1255
1256 /**
1257  * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1258  * @skb: pointer to the last skb in the rsc queue
1259  *
1260  * This function changes a queue full of hw rsc buffers into a completed
1261  * packet.  It uses the ->prev pointers to find the first packet and then
1262  * turns it into the frag list owner.
1263  **/
1264 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1265 {
1266         unsigned int frag_list_size = 0;
1267         unsigned int skb_cnt = 1;
1268
1269         while (skb->prev) {
1270                 struct sk_buff *prev = skb->prev;
1271                 frag_list_size += skb->len;
1272                 skb->prev = NULL;
1273                 skb = prev;
1274                 skb_cnt++;
1275         }
1276
1277         skb_shinfo(skb)->frag_list = skb->next;
1278         skb->next = NULL;
1279         skb->len += frag_list_size;
1280         skb->data_len += frag_list_size;
1281         skb->truesize += frag_list_size;
1282         IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1283
1284         return skb;
1285 }
1286
1287 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1288 {
1289         return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1290                 IXGBE_RXDADV_RSCCNT_MASK);
1291 }
1292
1293 static void ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1294                                struct ixgbe_ring *rx_ring,
1295                                int *work_done, int work_to_do)
1296 {
1297         struct ixgbe_adapter *adapter = q_vector->adapter;
1298         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1299         struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1300         struct sk_buff *skb;
1301         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1302         const int current_node = numa_node_id();
1303 #ifdef IXGBE_FCOE
1304         int ddp_bytes = 0;
1305 #endif /* IXGBE_FCOE */
1306         u32 staterr;
1307         u16 i;
1308         u16 cleaned_count = 0;
1309         bool pkt_is_rsc = false;
1310
1311         i = rx_ring->next_to_clean;
1312         rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1313         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1314
1315         while (staterr & IXGBE_RXD_STAT_DD) {
1316                 u32 upper_len = 0;
1317
1318                 rmb(); /* read descriptor and rx_buffer_info after status DD */
1319
1320                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1321
1322                 skb = rx_buffer_info->skb;
1323                 rx_buffer_info->skb = NULL;
1324                 prefetch(skb->data);
1325
1326                 if (ring_is_rsc_enabled(rx_ring))
1327                         pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1328
1329                 /* if this is a skb from previous receive DMA will be 0 */
1330                 if (rx_buffer_info->dma) {
1331                         u16 hlen;
1332                         if (pkt_is_rsc &&
1333                             !(staterr & IXGBE_RXD_STAT_EOP) &&
1334                             !skb->prev) {
1335                                 /*
1336                                  * When HWRSC is enabled, delay unmapping
1337                                  * of the first packet. It carries the
1338                                  * header information, HW may still
1339                                  * access the header after the writeback.
1340                                  * Only unmap it when EOP is reached
1341                                  */
1342                                 IXGBE_RSC_CB(skb)->delay_unmap = true;
1343                                 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1344                         } else {
1345                                 dma_unmap_single(rx_ring->dev,
1346                                                  rx_buffer_info->dma,
1347                                                  rx_ring->rx_buf_len,
1348                                                  DMA_FROM_DEVICE);
1349                         }
1350                         rx_buffer_info->dma = 0;
1351
1352                         if (ring_is_ps_enabled(rx_ring)) {
1353                                 hlen = ixgbe_get_hlen(rx_desc);
1354                                 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1355                         } else {
1356                                 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1357                         }
1358
1359                         skb_put(skb, hlen);
1360                 } else {
1361                         /* assume packet split since header is unmapped */
1362                         upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1363                 }
1364
1365                 if (upper_len) {
1366                         dma_unmap_page(rx_ring->dev,
1367                                        rx_buffer_info->page_dma,
1368                                        PAGE_SIZE / 2,
1369                                        DMA_FROM_DEVICE);
1370                         rx_buffer_info->page_dma = 0;
1371                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1372                                            rx_buffer_info->page,
1373                                            rx_buffer_info->page_offset,
1374                                            upper_len);
1375
1376                         if ((page_count(rx_buffer_info->page) == 1) &&
1377                             (page_to_nid(rx_buffer_info->page) == current_node))
1378                                 get_page(rx_buffer_info->page);
1379                         else
1380                                 rx_buffer_info->page = NULL;
1381
1382                         skb->len += upper_len;
1383                         skb->data_len += upper_len;
1384                         skb->truesize += upper_len;
1385                 }
1386
1387                 i++;
1388                 if (i == rx_ring->count)
1389                         i = 0;
1390
1391                 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1392                 prefetch(next_rxd);
1393                 cleaned_count++;
1394
1395                 if (pkt_is_rsc) {
1396                         u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1397                                      IXGBE_RXDADV_NEXTP_SHIFT;
1398                         next_buffer = &rx_ring->rx_buffer_info[nextp];
1399                 } else {
1400                         next_buffer = &rx_ring->rx_buffer_info[i];
1401                 }
1402
1403                 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1404                         if (ring_is_ps_enabled(rx_ring)) {
1405                                 rx_buffer_info->skb = next_buffer->skb;
1406                                 rx_buffer_info->dma = next_buffer->dma;
1407                                 next_buffer->skb = skb;
1408                                 next_buffer->dma = 0;
1409                         } else {
1410                                 skb->next = next_buffer->skb;
1411                                 skb->next->prev = skb;
1412                         }
1413                         rx_ring->rx_stats.non_eop_descs++;
1414                         goto next_desc;
1415                 }
1416
1417                 if (skb->prev) {
1418                         skb = ixgbe_transform_rsc_queue(skb);
1419                         /* if we got here without RSC the packet is invalid */
1420                         if (!pkt_is_rsc) {
1421                                 __pskb_trim(skb, 0);
1422                                 rx_buffer_info->skb = skb;
1423                                 goto next_desc;
1424                         }
1425                 }
1426
1427                 if (ring_is_rsc_enabled(rx_ring)) {
1428                         if (IXGBE_RSC_CB(skb)->delay_unmap) {
1429                                 dma_unmap_single(rx_ring->dev,
1430                                                  IXGBE_RSC_CB(skb)->dma,
1431                                                  rx_ring->rx_buf_len,
1432                                                  DMA_FROM_DEVICE);
1433                                 IXGBE_RSC_CB(skb)->dma = 0;
1434                                 IXGBE_RSC_CB(skb)->delay_unmap = false;
1435                         }
1436                 }
1437                 if (pkt_is_rsc) {
1438                         if (ring_is_ps_enabled(rx_ring))
1439                                 rx_ring->rx_stats.rsc_count +=
1440                                         skb_shinfo(skb)->nr_frags;
1441                         else
1442                                 rx_ring->rx_stats.rsc_count +=
1443                                         IXGBE_RSC_CB(skb)->skb_cnt;
1444                         rx_ring->rx_stats.rsc_flush++;
1445                 }
1446
1447                 /* ERR_MASK will only have valid bits if EOP set */
1448                 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1449                         /* trim packet back to size 0 and recycle it */
1450                         __pskb_trim(skb, 0);
1451                         rx_buffer_info->skb = skb;
1452                         goto next_desc;
1453                 }
1454
1455                 ixgbe_rx_checksum(adapter, rx_desc, skb);
1456
1457                 /* probably a little skewed due to removing CRC */
1458                 total_rx_bytes += skb->len;
1459                 total_rx_packets++;
1460
1461                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1462 #ifdef IXGBE_FCOE
1463                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1464                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1465                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1466                         if (!ddp_bytes)
1467                                 goto next_desc;
1468                 }
1469 #endif /* IXGBE_FCOE */
1470                 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1471
1472 next_desc:
1473                 rx_desc->wb.upper.status_error = 0;
1474
1475                 (*work_done)++;
1476                 if (*work_done >= work_to_do)
1477                         break;
1478
1479                 /* return some buffers to hardware, one at a time is too slow */
1480                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1481                         ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1482                         cleaned_count = 0;
1483                 }
1484
1485                 /* use prefetched values */
1486                 rx_desc = next_rxd;
1487                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1488         }
1489
1490         rx_ring->next_to_clean = i;
1491         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1492
1493         if (cleaned_count)
1494                 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1495
1496 #ifdef IXGBE_FCOE
1497         /* include DDPed FCoE data */
1498         if (ddp_bytes > 0) {
1499                 unsigned int mss;
1500
1501                 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1502                         sizeof(struct fc_frame_header) -
1503                         sizeof(struct fcoe_crc_eof);
1504                 if (mss > 512)
1505                         mss &= ~511;
1506                 total_rx_bytes += ddp_bytes;
1507                 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1508         }
1509 #endif /* IXGBE_FCOE */
1510
1511         rx_ring->total_packets += total_rx_packets;
1512         rx_ring->total_bytes += total_rx_bytes;
1513         u64_stats_update_begin(&rx_ring->syncp);
1514         rx_ring->stats.packets += total_rx_packets;
1515         rx_ring->stats.bytes += total_rx_bytes;
1516         u64_stats_update_end(&rx_ring->syncp);
1517 }
1518
1519 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1520 /**
1521  * ixgbe_configure_msix - Configure MSI-X hardware
1522  * @adapter: board private structure
1523  *
1524  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1525  * interrupts.
1526  **/
1527 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1528 {
1529         struct ixgbe_q_vector *q_vector;
1530         int i, q_vectors, v_idx, r_idx;
1531         u32 mask;
1532
1533         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1534
1535         /*
1536          * Populate the IVAR table and set the ITR values to the
1537          * corresponding register.
1538          */
1539         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1540                 q_vector = adapter->q_vector[v_idx];
1541                 /* XXX for_each_set_bit(...) */
1542                 r_idx = find_first_bit(q_vector->rxr_idx,
1543                                        adapter->num_rx_queues);
1544
1545                 for (i = 0; i < q_vector->rxr_count; i++) {
1546                         u8 reg_idx = adapter->rx_ring[r_idx]->reg_idx;
1547                         ixgbe_set_ivar(adapter, 0, reg_idx, v_idx);
1548                         r_idx = find_next_bit(q_vector->rxr_idx,
1549                                               adapter->num_rx_queues,
1550                                               r_idx + 1);
1551                 }
1552                 r_idx = find_first_bit(q_vector->txr_idx,
1553                                        adapter->num_tx_queues);
1554
1555                 for (i = 0; i < q_vector->txr_count; i++) {
1556                         u8 reg_idx = adapter->tx_ring[r_idx]->reg_idx;
1557                         ixgbe_set_ivar(adapter, 1, reg_idx, v_idx);
1558                         r_idx = find_next_bit(q_vector->txr_idx,
1559                                               adapter->num_tx_queues,
1560                                               r_idx + 1);
1561                 }
1562
1563                 if (q_vector->txr_count && !q_vector->rxr_count)
1564                         /* tx only */
1565                         q_vector->eitr = adapter->tx_eitr_param;
1566                 else if (q_vector->rxr_count)
1567                         /* rx or mixed */
1568                         q_vector->eitr = adapter->rx_eitr_param;
1569
1570                 ixgbe_write_eitr(q_vector);
1571                 /* If Flow Director is enabled, set interrupt affinity */
1572                 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1573                     (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1574                         /*
1575                          * Allocate the affinity_hint cpumask, assign the mask
1576                          * for this vector, and set our affinity_hint for
1577                          * this irq.
1578                          */
1579                         if (!alloc_cpumask_var(&q_vector->affinity_mask,
1580                                                GFP_KERNEL))
1581                                 return;
1582                         cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1583                         irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1584                                               q_vector->affinity_mask);
1585                 }
1586         }
1587
1588         switch (adapter->hw.mac.type) {
1589         case ixgbe_mac_82598EB:
1590                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1591                                v_idx);
1592                 break;
1593         case ixgbe_mac_82599EB:
1594         case ixgbe_mac_X540:
1595                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1596                 break;
1597
1598         default:
1599                 break;
1600         }
1601         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1602
1603         /* set up to autoclear timer, and the vectors */
1604         mask = IXGBE_EIMS_ENABLE_MASK;
1605         if (adapter->num_vfs)
1606                 mask &= ~(IXGBE_EIMS_OTHER |
1607                           IXGBE_EIMS_MAILBOX |
1608                           IXGBE_EIMS_LSC);
1609         else
1610                 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1611         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1612 }
1613
1614 enum latency_range {
1615         lowest_latency = 0,
1616         low_latency = 1,
1617         bulk_latency = 2,
1618         latency_invalid = 255
1619 };
1620
1621 /**
1622  * ixgbe_update_itr - update the dynamic ITR value based on statistics
1623  * @adapter: pointer to adapter
1624  * @eitr: eitr setting (ints per sec) to give last timeslice
1625  * @itr_setting: current throttle rate in ints/second
1626  * @packets: the number of packets during this measurement interval
1627  * @bytes: the number of bytes during this measurement interval
1628  *
1629  *      Stores a new ITR value based on packets and byte
1630  *      counts during the last interrupt.  The advantage of per interrupt
1631  *      computation is faster updates and more accurate ITR for the current
1632  *      traffic pattern.  Constants in this function were computed
1633  *      based on theoretical maximum wire speed and thresholds were set based
1634  *      on testing data as well as attempting to minimize response time
1635  *      while increasing bulk throughput.
1636  *      this functionality is controlled by the InterruptThrottleRate module
1637  *      parameter (see ixgbe_param.c)
1638  **/
1639 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1640                            u32 eitr, u8 itr_setting,
1641                            int packets, int bytes)
1642 {
1643         unsigned int retval = itr_setting;
1644         u32 timepassed_us;
1645         u64 bytes_perint;
1646
1647         if (packets == 0)
1648                 goto update_itr_done;
1649
1650
1651         /* simple throttlerate management
1652          *    0-20MB/s lowest (100000 ints/s)
1653          *   20-100MB/s low   (20000 ints/s)
1654          *  100-1249MB/s bulk (8000 ints/s)
1655          */
1656         /* what was last interrupt timeslice? */
1657         timepassed_us = 1000000/eitr;
1658         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1659
1660         switch (itr_setting) {
1661         case lowest_latency:
1662                 if (bytes_perint > adapter->eitr_low)
1663                         retval = low_latency;
1664                 break;
1665         case low_latency:
1666                 if (bytes_perint > adapter->eitr_high)
1667                         retval = bulk_latency;
1668                 else if (bytes_perint <= adapter->eitr_low)
1669                         retval = lowest_latency;
1670                 break;
1671         case bulk_latency:
1672                 if (bytes_perint <= adapter->eitr_high)
1673                         retval = low_latency;
1674                 break;
1675         }
1676
1677 update_itr_done:
1678         return retval;
1679 }
1680
1681 /**
1682  * ixgbe_write_eitr - write EITR register in hardware specific way
1683  * @q_vector: structure containing interrupt and ring information
1684  *
1685  * This function is made to be called by ethtool and by the driver
1686  * when it needs to update EITR registers at runtime.  Hardware
1687  * specific quirks/differences are taken care of here.
1688  */
1689 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1690 {
1691         struct ixgbe_adapter *adapter = q_vector->adapter;
1692         struct ixgbe_hw *hw = &adapter->hw;
1693         int v_idx = q_vector->v_idx;
1694         u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1695
1696         switch (adapter->hw.mac.type) {
1697         case ixgbe_mac_82598EB:
1698                 /* must write high and low 16 bits to reset counter */
1699                 itr_reg |= (itr_reg << 16);
1700                 break;
1701         case ixgbe_mac_82599EB:
1702         case ixgbe_mac_X540:
1703                 /*
1704                  * 82599 and X540 can support a value of zero, so allow it for
1705                  * max interrupt rate, but there is an errata where it can
1706                  * not be zero with RSC
1707                  */
1708                 if (itr_reg == 8 &&
1709                     !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1710                         itr_reg = 0;
1711
1712                 /*
1713                  * set the WDIS bit to not clear the timer bits and cause an
1714                  * immediate assertion of the interrupt
1715                  */
1716                 itr_reg |= IXGBE_EITR_CNT_WDIS;
1717                 break;
1718         default:
1719                 break;
1720         }
1721         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1722 }
1723
1724 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1725 {
1726         struct ixgbe_adapter *adapter = q_vector->adapter;
1727         int i, r_idx;
1728         u32 new_itr;
1729         u8 current_itr, ret_itr;
1730
1731         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1732         for (i = 0; i < q_vector->txr_count; i++) {
1733                 struct ixgbe_ring *tx_ring = adapter->tx_ring[r_idx];
1734                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1735                                            q_vector->tx_itr,
1736                                            tx_ring->total_packets,
1737                                            tx_ring->total_bytes);
1738                 /* if the result for this queue would decrease interrupt
1739                  * rate for this vector then use that result */
1740                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1741                                     q_vector->tx_itr - 1 : ret_itr);
1742                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1743                                       r_idx + 1);
1744         }
1745
1746         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1747         for (i = 0; i < q_vector->rxr_count; i++) {
1748                 struct ixgbe_ring *rx_ring = adapter->rx_ring[r_idx];
1749                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1750                                            q_vector->rx_itr,
1751                                            rx_ring->total_packets,
1752                                            rx_ring->total_bytes);
1753                 /* if the result for this queue would decrease interrupt
1754                  * rate for this vector then use that result */
1755                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1756                                     q_vector->rx_itr - 1 : ret_itr);
1757                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1758                                       r_idx + 1);
1759         }
1760
1761         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1762
1763         switch (current_itr) {
1764         /* counts and packets in update_itr are dependent on these numbers */
1765         case lowest_latency:
1766                 new_itr = 100000;
1767                 break;
1768         case low_latency:
1769                 new_itr = 20000; /* aka hwitr = ~200 */
1770                 break;
1771         case bulk_latency:
1772         default:
1773                 new_itr = 8000;
1774                 break;
1775         }
1776
1777         if (new_itr != q_vector->eitr) {
1778                 /* do an exponential smoothing */
1779                 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
1780
1781                 /* save the algorithm value here, not the smoothed one */
1782                 q_vector->eitr = new_itr;
1783
1784                 ixgbe_write_eitr(q_vector);
1785         }
1786 }
1787
1788 /**
1789  * ixgbe_check_overtemp_task - worker thread to check over tempurature
1790  * @work: pointer to work_struct containing our data
1791  **/
1792 static void ixgbe_check_overtemp_task(struct work_struct *work)
1793 {
1794         struct ixgbe_adapter *adapter = container_of(work,
1795                                                      struct ixgbe_adapter,
1796                                                      check_overtemp_task);
1797         struct ixgbe_hw *hw = &adapter->hw;
1798         u32 eicr = adapter->interrupt_event;
1799
1800         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1801                 return;
1802
1803         switch (hw->device_id) {
1804         case IXGBE_DEV_ID_82599_T3_LOM: {
1805                 u32 autoneg;
1806                 bool link_up = false;
1807
1808                 if (hw->mac.ops.check_link)
1809                         hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1810
1811                 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1812                     (eicr & IXGBE_EICR_LSC))
1813                         /* Check if this is due to overtemp */
1814                         if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1815                                 break;
1816                 return;
1817         }
1818         default:
1819                 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1820                         return;
1821                 break;
1822         }
1823         e_crit(drv,
1824                "Network adapter has been stopped because it has over heated. "
1825                "Restart the computer. If the problem persists, "
1826                "power off the system and replace the adapter\n");
1827         /* write to clear the interrupt */
1828         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1829 }
1830
1831 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1832 {
1833         struct ixgbe_hw *hw = &adapter->hw;
1834
1835         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1836             (eicr & IXGBE_EICR_GPI_SDP1)) {
1837                 e_crit(probe, "Fan has stopped, replace the adapter\n");
1838                 /* write to clear the interrupt */
1839                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1840         }
1841 }
1842
1843 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1844 {
1845         struct ixgbe_hw *hw = &adapter->hw;
1846
1847         if (eicr & IXGBE_EICR_GPI_SDP2) {
1848                 /* Clear the interrupt */
1849                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1850                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1851                         schedule_work(&adapter->sfp_config_module_task);
1852         }
1853
1854         if (eicr & IXGBE_EICR_GPI_SDP1) {
1855                 /* Clear the interrupt */
1856                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1857                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1858                         schedule_work(&adapter->multispeed_fiber_task);
1859         }
1860 }
1861
1862 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1863 {
1864         struct ixgbe_hw *hw = &adapter->hw;
1865
1866         adapter->lsc_int++;
1867         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1868         adapter->link_check_timeout = jiffies;
1869         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1870                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1871                 IXGBE_WRITE_FLUSH(hw);
1872                 schedule_work(&adapter->watchdog_task);
1873         }
1874 }
1875
1876 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1877 {
1878         struct net_device *netdev = data;
1879         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1880         struct ixgbe_hw *hw = &adapter->hw;
1881         u32 eicr;
1882
1883         /*
1884          * Workaround for Silicon errata.  Use clear-by-write instead
1885          * of clear-by-read.  Reading with EICS will return the
1886          * interrupt causes without clearing, which later be done
1887          * with the write to EICR.
1888          */
1889         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1890         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1891
1892         if (eicr & IXGBE_EICR_LSC)
1893                 ixgbe_check_lsc(adapter);
1894
1895         if (eicr & IXGBE_EICR_MAILBOX)
1896                 ixgbe_msg_task(adapter);
1897
1898         switch (hw->mac.type) {
1899         case ixgbe_mac_82599EB:
1900                 ixgbe_check_sfp_event(adapter, eicr);
1901                 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1902                     ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
1903                         adapter->interrupt_event = eicr;
1904                         schedule_work(&adapter->check_overtemp_task);
1905                 }
1906                 /* now fallthrough to handle Flow Director */
1907         case ixgbe_mac_X540:
1908                 /* Handle Flow Director Full threshold interrupt */
1909                 if (eicr & IXGBE_EICR_FLOW_DIR) {
1910                         int i;
1911                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1912                         /* Disable transmits before FDIR Re-initialization */
1913                         netif_tx_stop_all_queues(netdev);
1914                         for (i = 0; i < adapter->num_tx_queues; i++) {
1915                                 struct ixgbe_ring *tx_ring =
1916                                                             adapter->tx_ring[i];
1917                                 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1918                                                        &tx_ring->state))
1919                                         schedule_work(&adapter->fdir_reinit_task);
1920                         }
1921                 }
1922                 break;
1923         default:
1924                 break;
1925         }
1926
1927         ixgbe_check_fan_failure(adapter, eicr);
1928
1929         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1930                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1931
1932         return IRQ_HANDLED;
1933 }
1934
1935 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1936                                            u64 qmask)
1937 {
1938         u32 mask;
1939         struct ixgbe_hw *hw = &adapter->hw;
1940
1941         switch (hw->mac.type) {
1942         case ixgbe_mac_82598EB:
1943                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1944                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1945                 break;
1946         case ixgbe_mac_82599EB:
1947         case ixgbe_mac_X540:
1948                 mask = (qmask & 0xFFFFFFFF);
1949                 if (mask)
1950                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1951                 mask = (qmask >> 32);
1952                 if (mask)
1953                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
1954                 break;
1955         default:
1956                 break;
1957         }
1958         /* skip the flush */
1959 }
1960
1961 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1962                                             u64 qmask)
1963 {
1964         u32 mask;
1965         struct ixgbe_hw *hw = &adapter->hw;
1966
1967         switch (hw->mac.type) {
1968         case ixgbe_mac_82598EB:
1969                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1970                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
1971                 break;
1972         case ixgbe_mac_82599EB:
1973         case ixgbe_mac_X540:
1974                 mask = (qmask & 0xFFFFFFFF);
1975                 if (mask)
1976                         IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
1977                 mask = (qmask >> 32);
1978                 if (mask)
1979                         IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
1980                 break;
1981         default:
1982                 break;
1983         }
1984         /* skip the flush */
1985 }
1986
1987 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1988 {
1989         struct ixgbe_q_vector *q_vector = data;
1990         struct ixgbe_adapter  *adapter = q_vector->adapter;
1991         struct ixgbe_ring     *tx_ring;
1992         int i, r_idx;
1993
1994         if (!q_vector->txr_count)
1995                 return IRQ_HANDLED;
1996
1997         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1998         for (i = 0; i < q_vector->txr_count; i++) {
1999                 tx_ring = adapter->tx_ring[r_idx];
2000                 tx_ring->total_bytes = 0;
2001                 tx_ring->total_packets = 0;
2002                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2003                                       r_idx + 1);
2004         }
2005
2006         /* EIAM disabled interrupts (on this vector) for us */
2007         napi_schedule(&q_vector->napi);
2008
2009         return IRQ_HANDLED;
2010 }
2011
2012 /**
2013  * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
2014  * @irq: unused
2015  * @data: pointer to our q_vector struct for this interrupt vector
2016  **/
2017 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
2018 {
2019         struct ixgbe_q_vector *q_vector = data;
2020         struct ixgbe_adapter  *adapter = q_vector->adapter;
2021         struct ixgbe_ring  *rx_ring;
2022         int r_idx;
2023         int i;
2024
2025 #ifdef CONFIG_IXGBE_DCA
2026         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2027                 ixgbe_update_dca(q_vector);
2028 #endif
2029
2030         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2031         for (i = 0; i < q_vector->rxr_count; i++) {
2032                 rx_ring = adapter->rx_ring[r_idx];
2033                 rx_ring->total_bytes = 0;
2034                 rx_ring->total_packets = 0;
2035                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2036                                       r_idx + 1);
2037         }
2038
2039         if (!q_vector->rxr_count)
2040                 return IRQ_HANDLED;
2041
2042         /* EIAM disabled interrupts (on this vector) for us */
2043         napi_schedule(&q_vector->napi);
2044
2045         return IRQ_HANDLED;
2046 }
2047
2048 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
2049 {
2050         struct ixgbe_q_vector *q_vector = data;
2051         struct ixgbe_adapter  *adapter = q_vector->adapter;
2052         struct ixgbe_ring  *ring;
2053         int r_idx;
2054         int i;
2055
2056         if (!q_vector->txr_count && !q_vector->rxr_count)
2057                 return IRQ_HANDLED;
2058
2059         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2060         for (i = 0; i < q_vector->txr_count; i++) {
2061                 ring = adapter->tx_ring[r_idx];
2062                 ring->total_bytes = 0;
2063                 ring->total_packets = 0;
2064                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2065                                       r_idx + 1);
2066         }
2067
2068         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2069         for (i = 0; i < q_vector->rxr_count; i++) {
2070                 ring = adapter->rx_ring[r_idx];
2071                 ring->total_bytes = 0;
2072                 ring->total_packets = 0;
2073                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2074                                       r_idx + 1);
2075         }
2076
2077         /* EIAM disabled interrupts (on this vector) for us */
2078         napi_schedule(&q_vector->napi);
2079
2080         return IRQ_HANDLED;
2081 }
2082
2083 /**
2084  * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
2085  * @napi: napi struct with our devices info in it
2086  * @budget: amount of work driver is allowed to do this pass, in packets
2087  *
2088  * This function is optimized for cleaning one queue only on a single
2089  * q_vector!!!
2090  **/
2091 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
2092 {
2093         struct ixgbe_q_vector *q_vector =
2094                                container_of(napi, struct ixgbe_q_vector, napi);
2095         struct ixgbe_adapter *adapter = q_vector->adapter;
2096         struct ixgbe_ring *rx_ring = NULL;
2097         int work_done = 0;
2098         long r_idx;
2099
2100 #ifdef CONFIG_IXGBE_DCA
2101         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2102                 ixgbe_update_dca(q_vector);
2103 #endif
2104
2105         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2106         rx_ring = adapter->rx_ring[r_idx];
2107
2108         ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
2109
2110         /* If all Rx work done, exit the polling mode */
2111         if (work_done < budget) {
2112                 napi_complete(napi);
2113                 if (adapter->rx_itr_setting & 1)
2114                         ixgbe_set_itr_msix(q_vector);
2115                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2116                         ixgbe_irq_enable_queues(adapter,
2117                                                 ((u64)1 << q_vector->v_idx));
2118         }
2119
2120         return work_done;
2121 }
2122
2123 /**
2124  * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
2125  * @napi: napi struct with our devices info in it
2126  * @budget: amount of work driver is allowed to do this pass, in packets
2127  *
2128  * This function will clean more than one rx queue associated with a
2129  * q_vector.
2130  **/
2131 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
2132 {
2133         struct ixgbe_q_vector *q_vector =
2134                                container_of(napi, struct ixgbe_q_vector, napi);
2135         struct ixgbe_adapter *adapter = q_vector->adapter;
2136         struct ixgbe_ring *ring = NULL;
2137         int work_done = 0, i;
2138         long r_idx;
2139         bool tx_clean_complete = true;
2140
2141 #ifdef CONFIG_IXGBE_DCA
2142         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2143                 ixgbe_update_dca(q_vector);
2144 #endif
2145
2146         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2147         for (i = 0; i < q_vector->txr_count; i++) {
2148                 ring = adapter->tx_ring[r_idx];
2149                 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
2150                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2151                                       r_idx + 1);
2152         }
2153
2154         /* attempt to distribute budget to each queue fairly, but don't allow
2155          * the budget to go below 1 because we'll exit polling */
2156         budget /= (q_vector->rxr_count ?: 1);
2157         budget = max(budget, 1);
2158         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2159         for (i = 0; i < q_vector->rxr_count; i++) {
2160                 ring = adapter->rx_ring[r_idx];
2161                 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
2162                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2163                                       r_idx + 1);
2164         }
2165
2166         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2167         ring = adapter->rx_ring[r_idx];
2168         /* If all Rx work done, exit the polling mode */
2169         if (work_done < budget) {
2170                 napi_complete(napi);
2171                 if (adapter->rx_itr_setting & 1)
2172                         ixgbe_set_itr_msix(q_vector);
2173                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2174                         ixgbe_irq_enable_queues(adapter,
2175                                                 ((u64)1 << q_vector->v_idx));
2176                 return 0;
2177         }
2178
2179         return work_done;
2180 }
2181
2182 /**
2183  * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2184  * @napi: napi struct with our devices info in it
2185  * @budget: amount of work driver is allowed to do this pass, in packets
2186  *
2187  * This function is optimized for cleaning one queue only on a single
2188  * q_vector!!!
2189  **/
2190 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2191 {
2192         struct ixgbe_q_vector *q_vector =
2193                                container_of(napi, struct ixgbe_q_vector, napi);
2194         struct ixgbe_adapter *adapter = q_vector->adapter;
2195         struct ixgbe_ring *tx_ring = NULL;
2196         int work_done = 0;
2197         long r_idx;
2198
2199 #ifdef CONFIG_IXGBE_DCA
2200         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2201                 ixgbe_update_dca(q_vector);
2202 #endif
2203
2204         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2205         tx_ring = adapter->tx_ring[r_idx];
2206
2207         if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2208                 work_done = budget;
2209
2210         /* If all Tx work done, exit the polling mode */
2211         if (work_done < budget) {
2212                 napi_complete(napi);
2213                 if (adapter->tx_itr_setting & 1)
2214                         ixgbe_set_itr_msix(q_vector);
2215                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2216                         ixgbe_irq_enable_queues(adapter,
2217                                                 ((u64)1 << q_vector->v_idx));
2218         }
2219
2220         return work_done;
2221 }
2222
2223 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2224                                      int r_idx)
2225 {
2226         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2227         struct ixgbe_ring *rx_ring = a->rx_ring[r_idx];
2228
2229         set_bit(r_idx, q_vector->rxr_idx);
2230         q_vector->rxr_count++;
2231         rx_ring->q_vector = q_vector;
2232 }
2233
2234 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2235                                      int t_idx)
2236 {
2237         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2238         struct ixgbe_ring *tx_ring = a->tx_ring[t_idx];
2239
2240         set_bit(t_idx, q_vector->txr_idx);
2241         q_vector->txr_count++;
2242         tx_ring->q_vector = q_vector;
2243 }
2244
2245 /**
2246  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2247  * @adapter: board private structure to initialize
2248  *
2249  * This function maps descriptor rings to the queue-specific vectors
2250  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
2251  * one vector per ring/queue, but on a constrained vector budget, we
2252  * group the rings as "efficiently" as possible.  You would add new
2253  * mapping configurations in here.
2254  **/
2255 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2256 {
2257         int q_vectors;
2258         int v_start = 0;
2259         int rxr_idx = 0, txr_idx = 0;
2260         int rxr_remaining = adapter->num_rx_queues;
2261         int txr_remaining = adapter->num_tx_queues;
2262         int i, j;
2263         int rqpv, tqpv;
2264         int err = 0;
2265
2266         /* No mapping required if MSI-X is disabled. */
2267         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2268                 goto out;
2269
2270         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2271
2272         /*
2273          * The ideal configuration...
2274          * We have enough vectors to map one per queue.
2275          */
2276         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2277                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2278                         map_vector_to_rxq(adapter, v_start, rxr_idx);
2279
2280                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2281                         map_vector_to_txq(adapter, v_start, txr_idx);
2282
2283                 goto out;
2284         }
2285
2286         /*
2287          * If we don't have enough vectors for a 1-to-1
2288          * mapping, we'll have to group them so there are
2289          * multiple queues per vector.
2290          */
2291         /* Re-adjusting *qpv takes care of the remainder. */
2292         for (i = v_start; i < q_vectors; i++) {
2293                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
2294                 for (j = 0; j < rqpv; j++) {
2295                         map_vector_to_rxq(adapter, i, rxr_idx);
2296                         rxr_idx++;
2297                         rxr_remaining--;
2298                 }
2299                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
2300                 for (j = 0; j < tqpv; j++) {
2301                         map_vector_to_txq(adapter, i, txr_idx);
2302                         txr_idx++;
2303                         txr_remaining--;
2304                 }
2305         }
2306 out:
2307         return err;
2308 }
2309
2310 /**
2311  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2312  * @adapter: board private structure
2313  *
2314  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2315  * interrupts from the kernel.
2316  **/
2317 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2318 {
2319         struct net_device *netdev = adapter->netdev;
2320         irqreturn_t (*handler)(int, void *);
2321         int i, vector, q_vectors, err;
2322         int ri = 0, ti = 0;
2323
2324         /* Decrement for Other and TCP Timer vectors */
2325         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2326
2327         err = ixgbe_map_rings_to_vectors(adapter);
2328         if (err)
2329                 return err;
2330
2331 #define SET_HANDLER(_v) (((_v)->rxr_count && (_v)->txr_count)        \
2332                                           ? &ixgbe_msix_clean_many : \
2333                           (_v)->rxr_count ? &ixgbe_msix_clean_rx   : \
2334                           (_v)->txr_count ? &ixgbe_msix_clean_tx   : \
2335                           NULL)
2336         for (vector = 0; vector < q_vectors; vector++) {
2337                 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2338                 handler = SET_HANDLER(q_vector);
2339
2340                 if (handler == &ixgbe_msix_clean_rx) {
2341                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2342                                  "%s-%s-%d", netdev->name, "rx", ri++);
2343                 } else if (handler == &ixgbe_msix_clean_tx) {
2344                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2345                                  "%s-%s-%d", netdev->name, "tx", ti++);
2346                 } else if (handler == &ixgbe_msix_clean_many) {
2347                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2348                                  "%s-%s-%d", netdev->name, "TxRx", ri++);
2349                         ti++;
2350                 } else {
2351                         /* skip this unused q_vector */
2352                         continue;
2353                 }
2354                 err = request_irq(adapter->msix_entries[vector].vector,
2355                                   handler, 0, q_vector->name,
2356                                   q_vector);
2357                 if (err) {
2358                         e_err(probe, "request_irq failed for MSIX interrupt "
2359                               "Error: %d\n", err);
2360                         goto free_queue_irqs;
2361                 }
2362         }
2363
2364         sprintf(adapter->lsc_int_name, "%s:lsc", netdev->name);
2365         err = request_irq(adapter->msix_entries[vector].vector,
2366                           ixgbe_msix_lsc, 0, adapter->lsc_int_name, netdev);
2367         if (err) {
2368                 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2369                 goto free_queue_irqs;
2370         }
2371
2372         return 0;
2373
2374 free_queue_irqs:
2375         for (i = vector - 1; i >= 0; i--)
2376                 free_irq(adapter->msix_entries[--vector].vector,
2377                          adapter->q_vector[i]);
2378         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2379         pci_disable_msix(adapter->pdev);
2380         kfree(adapter->msix_entries);
2381         adapter->msix_entries = NULL;
2382         return err;
2383 }
2384
2385 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2386 {
2387         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2388         struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2389         struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2390         u32 new_itr = q_vector->eitr;
2391         u8 current_itr;
2392
2393         q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2394                                             q_vector->tx_itr,
2395                                             tx_ring->total_packets,
2396                                             tx_ring->total_bytes);
2397         q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2398                                             q_vector->rx_itr,
2399                                             rx_ring->total_packets,
2400                                             rx_ring->total_bytes);
2401
2402         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2403
2404         switch (current_itr) {
2405         /* counts and packets in update_itr are dependent on these numbers */
2406         case lowest_latency:
2407                 new_itr = 100000;
2408                 break;
2409         case low_latency:
2410                 new_itr = 20000; /* aka hwitr = ~200 */
2411                 break;
2412         case bulk_latency:
2413                 new_itr = 8000;
2414                 break;
2415         default:
2416                 break;
2417         }
2418
2419         if (new_itr != q_vector->eitr) {
2420                 /* do an exponential smoothing */
2421                 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
2422
2423                 /* save the algorithm value here */
2424                 q_vector->eitr = new_itr;
2425
2426                 ixgbe_write_eitr(q_vector);
2427         }
2428 }
2429
2430 /**
2431  * ixgbe_irq_enable - Enable default interrupt generation settings
2432  * @adapter: board private structure
2433  **/
2434 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2435                                     bool flush)
2436 {
2437         u32 mask;
2438
2439         mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2440         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2441                 mask |= IXGBE_EIMS_GPI_SDP0;
2442         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2443                 mask |= IXGBE_EIMS_GPI_SDP1;
2444         switch (adapter->hw.mac.type) {
2445         case ixgbe_mac_82599EB:
2446         case ixgbe_mac_X540:
2447                 mask |= IXGBE_EIMS_ECC;
2448                 mask |= IXGBE_EIMS_GPI_SDP1;
2449                 mask |= IXGBE_EIMS_GPI_SDP2;
2450                 if (adapter->num_vfs)
2451                         mask |= IXGBE_EIMS_MAILBOX;
2452                 break;
2453         default:
2454                 break;
2455         }
2456         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2457             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2458                 mask |= IXGBE_EIMS_FLOW_DIR;
2459
2460         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2461         if (queues)
2462                 ixgbe_irq_enable_queues(adapter, ~0);
2463         if (flush)
2464                 IXGBE_WRITE_FLUSH(&adapter->hw);
2465
2466         if (adapter->num_vfs > 32) {
2467                 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2468                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2469         }
2470 }
2471
2472 /**
2473  * ixgbe_intr - legacy mode Interrupt Handler
2474  * @irq: interrupt number
2475  * @data: pointer to a network interface device structure
2476  **/
2477 static irqreturn_t ixgbe_intr(int irq, void *data)
2478 {
2479         struct net_device *netdev = data;
2480         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2481         struct ixgbe_hw *hw = &adapter->hw;
2482         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2483         u32 eicr;
2484
2485         /*
2486          * Workaround for silicon errata on 82598.  Mask the interrupts
2487          * before the read of EICR.
2488          */
2489         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2490
2491         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2492          * therefore no explict interrupt disable is necessary */
2493         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2494         if (!eicr) {
2495                 /*
2496                  * shared interrupt alert!
2497                  * make sure interrupts are enabled because the read will
2498                  * have disabled interrupts due to EIAM
2499                  * finish the workaround of silicon errata on 82598.  Unmask
2500                  * the interrupt that we masked before the EICR read.
2501                  */
2502                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2503                         ixgbe_irq_enable(adapter, true, true);
2504                 return IRQ_NONE;        /* Not our interrupt */
2505         }
2506
2507         if (eicr & IXGBE_EICR_LSC)
2508                 ixgbe_check_lsc(adapter);
2509
2510         switch (hw->mac.type) {
2511         case ixgbe_mac_82599EB:
2512                 ixgbe_check_sfp_event(adapter, eicr);
2513                 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2514                     ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
2515                         adapter->interrupt_event = eicr;
2516                         schedule_work(&adapter->check_overtemp_task);
2517                 }
2518                 break;
2519         default:
2520                 break;
2521         }
2522
2523         ixgbe_check_fan_failure(adapter, eicr);
2524
2525         if (napi_schedule_prep(&(q_vector->napi))) {
2526                 adapter->tx_ring[0]->total_packets = 0;
2527                 adapter->tx_ring[0]->total_bytes = 0;
2528                 adapter->rx_ring[0]->total_packets = 0;
2529                 adapter->rx_ring[0]->total_bytes = 0;
2530                 /* would disable interrupts here but EIAM disabled it */
2531                 __napi_schedule(&(q_vector->napi));
2532         }
2533
2534         /*
2535          * re-enable link(maybe) and non-queue interrupts, no flush.
2536          * ixgbe_poll will re-enable the queue interrupts
2537          */
2538
2539         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2540                 ixgbe_irq_enable(adapter, false, false);
2541
2542         return IRQ_HANDLED;
2543 }
2544
2545 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2546 {
2547         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2548
2549         for (i = 0; i < q_vectors; i++) {
2550                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2551                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2552                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2553                 q_vector->rxr_count = 0;
2554                 q_vector->txr_count = 0;
2555         }
2556 }
2557
2558 /**
2559  * ixgbe_request_irq - initialize interrupts
2560  * @adapter: board private structure
2561  *
2562  * Attempts to configure interrupts using the best available
2563  * capabilities of the hardware and kernel.
2564  **/
2565 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2566 {
2567         struct net_device *netdev = adapter->netdev;
2568         int err;
2569
2570         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2571                 err = ixgbe_request_msix_irqs(adapter);
2572         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2573                 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2574                                   netdev->name, netdev);
2575         } else {
2576                 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2577                                   netdev->name, netdev);
2578         }
2579
2580         if (err)
2581                 e_err(probe, "request_irq failed, Error %d\n", err);
2582
2583         return err;
2584 }
2585
2586 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2587 {
2588         struct net_device *netdev = adapter->netdev;
2589
2590         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2591                 int i, q_vectors;
2592
2593                 q_vectors = adapter->num_msix_vectors;
2594
2595                 i = q_vectors - 1;
2596                 free_irq(adapter->msix_entries[i].vector, netdev);
2597
2598                 i--;
2599                 for (; i >= 0; i--) {
2600                         free_irq(adapter->msix_entries[i].vector,
2601                                  adapter->q_vector[i]);
2602                 }
2603
2604                 ixgbe_reset_q_vectors(adapter);
2605         } else {
2606                 free_irq(adapter->pdev->irq, netdev);
2607         }
2608 }
2609
2610 /**
2611  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2612  * @adapter: board private structure
2613  **/
2614 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2615 {
2616         switch (adapter->hw.mac.type) {
2617         case ixgbe_mac_82598EB:
2618                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2619                 break;
2620         case ixgbe_mac_82599EB:
2621         case ixgbe_mac_X540:
2622                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2623                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2624                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2625                 if (adapter->num_vfs > 32)
2626                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2627                 break;
2628         default:
2629                 break;
2630         }
2631         IXGBE_WRITE_FLUSH(&adapter->hw);
2632         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2633                 int i;
2634                 for (i = 0; i < adapter->num_msix_vectors; i++)
2635                         synchronize_irq(adapter->msix_entries[i].vector);
2636         } else {
2637                 synchronize_irq(adapter->pdev->irq);
2638         }
2639 }
2640
2641 /**
2642  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2643  *
2644  **/
2645 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2646 {
2647         struct ixgbe_hw *hw = &adapter->hw;
2648
2649         IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2650                         EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2651
2652         ixgbe_set_ivar(adapter, 0, 0, 0);
2653         ixgbe_set_ivar(adapter, 1, 0, 0);
2654
2655         map_vector_to_rxq(adapter, 0, 0);
2656         map_vector_to_txq(adapter, 0, 0);
2657
2658         e_info(hw, "Legacy interrupt IVAR setup done\n");
2659 }
2660
2661 /**
2662  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2663  * @adapter: board private structure
2664  * @ring: structure containing ring specific data
2665  *
2666  * Configure the Tx descriptor ring after a reset.
2667  **/
2668 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2669                              struct ixgbe_ring *ring)
2670 {
2671         struct ixgbe_hw *hw = &adapter->hw;
2672         u64 tdba = ring->dma;
2673         int wait_loop = 10;
2674         u32 txdctl;
2675         u8 reg_idx = ring->reg_idx;
2676
2677         /* disable queue to avoid issues while updating state */
2678         txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2679         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2680                         txdctl & ~IXGBE_TXDCTL_ENABLE);
2681         IXGBE_WRITE_FLUSH(hw);
2682
2683         IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2684                         (tdba & DMA_BIT_MASK(32)));
2685         IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2686         IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2687                         ring->count * sizeof(union ixgbe_adv_tx_desc));
2688         IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2689         IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2690         ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2691
2692         /* configure fetching thresholds */
2693         if (adapter->rx_itr_setting == 0) {
2694                 /* cannot set wthresh when itr==0 */
2695                 txdctl &= ~0x007F0000;
2696         } else {
2697                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2698                 txdctl |= (8 << 16);
2699         }
2700         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2701                 /* PThresh workaround for Tx hang with DFP enabled. */
2702                 txdctl |= 32;
2703         }
2704
2705         /* reinitialize flowdirector state */
2706         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2707             adapter->atr_sample_rate) {
2708                 ring->atr_sample_rate = adapter->atr_sample_rate;
2709                 ring->atr_count = 0;
2710                 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2711         } else {
2712                 ring->atr_sample_rate = 0;
2713         }
2714
2715         clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2716
2717         /* enable queue */
2718         txdctl |= IXGBE_TXDCTL_ENABLE;
2719         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2720
2721         /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2722         if (hw->mac.type == ixgbe_mac_82598EB &&
2723             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2724                 return;
2725
2726         /* poll to verify queue is enabled */
2727         do {
2728                 msleep(1);
2729                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2730         } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2731         if (!wait_loop)
2732                 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2733 }
2734
2735 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2736 {
2737         struct ixgbe_hw *hw = &adapter->hw;
2738         u32 rttdcs;
2739         u32 mask;
2740
2741         if (hw->mac.type == ixgbe_mac_82598EB)
2742                 return;
2743
2744         /* disable the arbiter while setting MTQC */
2745         rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2746         rttdcs |= IXGBE_RTTDCS_ARBDIS;
2747         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2748
2749         /* set transmit pool layout */
2750         mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2751         switch (adapter->flags & mask) {
2752
2753         case (IXGBE_FLAG_SRIOV_ENABLED):
2754                 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2755                                 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2756                 break;
2757
2758         case (IXGBE_FLAG_DCB_ENABLED):
2759                 /* We enable 8 traffic classes, DCB only */
2760                 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2761                               (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2762                 break;
2763
2764         default:
2765                 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2766                 break;
2767         }
2768
2769         /* re-enable the arbiter */
2770         rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2771         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2772 }
2773
2774 /**
2775  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2776  * @adapter: board private structure
2777  *
2778  * Configure the Tx unit of the MAC after a reset.
2779  **/
2780 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2781 {
2782         struct ixgbe_hw *hw = &adapter->hw;
2783         u32 dmatxctl;
2784         u32 i;
2785
2786         ixgbe_setup_mtqc(adapter);
2787
2788         if (hw->mac.type != ixgbe_mac_82598EB) {
2789                 /* DMATXCTL.EN must be before Tx queues are enabled */
2790                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2791                 dmatxctl |= IXGBE_DMATXCTL_TE;
2792                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2793         }
2794
2795         /* Setup the HW Tx Head and Tail descriptor pointers */
2796         for (i = 0; i < adapter->num_tx_queues; i++)
2797                 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2798 }
2799
2800 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2801
2802 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2803                                    struct ixgbe_ring *rx_ring)
2804 {
2805         u32 srrctl;
2806         u8 reg_idx = rx_ring->reg_idx;
2807
2808         switch (adapter->hw.mac.type) {
2809         case ixgbe_mac_82598EB: {
2810                 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2811                 const int mask = feature[RING_F_RSS].mask;
2812                 reg_idx = reg_idx & mask;
2813         }
2814                 break;
2815         case ixgbe_mac_82599EB:
2816         case ixgbe_mac_X540:
2817         default:
2818                 break;
2819         }
2820
2821         srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2822
2823         srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2824         srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2825         if (adapter->num_vfs)
2826                 srrctl |= IXGBE_SRRCTL_DROP_EN;
2827
2828         srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2829                   IXGBE_SRRCTL_BSIZEHDR_MASK;
2830
2831         if (ring_is_ps_enabled(rx_ring)) {
2832 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2833                 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2834 #else
2835                 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2836 #endif
2837                 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2838         } else {
2839                 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2840                           IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2841                 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2842         }
2843
2844         IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2845 }
2846
2847 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2848 {
2849         struct ixgbe_hw *hw = &adapter->hw;
2850         static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2851                           0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2852                           0x6A3E67EA, 0x14364D17, 0x3BED200D};
2853         u32 mrqc = 0, reta = 0;
2854         u32 rxcsum;
2855         int i, j;
2856         int mask;
2857
2858         /* Fill out hash function seeds */
2859         for (i = 0; i < 10; i++)
2860                 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2861
2862         /* Fill out redirection table */
2863         for (i = 0, j = 0; i < 128; i++, j++) {
2864                 if (j == adapter->ring_feature[RING_F_RSS].indices)
2865                         j = 0;
2866                 /* reta = 4-byte sliding window of
2867                  * 0x00..(indices-1)(indices-1)00..etc. */
2868                 reta = (reta << 8) | (j * 0x11);
2869                 if ((i & 3) == 3)
2870                         IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2871         }
2872
2873         /* Disable indicating checksum in descriptor, enables RSS hash */
2874         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2875         rxcsum |= IXGBE_RXCSUM_PCSD;
2876         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2877
2878         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2879                 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2880         else
2881                 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2882 #ifdef CONFIG_IXGBE_DCB
2883                                          | IXGBE_FLAG_DCB_ENABLED
2884 #endif
2885                                          | IXGBE_FLAG_SRIOV_ENABLED
2886                                         );
2887
2888         switch (mask) {
2889         case (IXGBE_FLAG_RSS_ENABLED):
2890                 mrqc = IXGBE_MRQC_RSSEN;
2891                 break;
2892         case (IXGBE_FLAG_SRIOV_ENABLED):
2893                 mrqc = IXGBE_MRQC_VMDQEN;
2894                 break;
2895 #ifdef CONFIG_IXGBE_DCB
2896         case (IXGBE_FLAG_DCB_ENABLED):
2897                 mrqc = IXGBE_MRQC_RT8TCEN;
2898                 break;
2899 #endif /* CONFIG_IXGBE_DCB */
2900         default:
2901                 break;
2902         }
2903
2904         /* Perform hash on these packet types */
2905         mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2906               | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2907               | IXGBE_MRQC_RSS_FIELD_IPV6
2908               | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2909
2910         IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2911 }
2912
2913 /**
2914  * ixgbe_clear_rscctl - disable RSC for the indicated ring
2915  * @adapter: address of board private structure
2916  * @ring: structure containing ring specific data
2917  **/
2918 void ixgbe_clear_rscctl(struct ixgbe_adapter *adapter,
2919                         struct ixgbe_ring *ring)
2920 {
2921         struct ixgbe_hw *hw = &adapter->hw;
2922         u32 rscctrl;
2923         u8 reg_idx = ring->reg_idx;
2924
2925         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2926         rscctrl &= ~IXGBE_RSCCTL_RSCEN;
2927         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2928 }
2929
2930 /**
2931  * ixgbe_configure_rscctl - enable RSC for the indicated ring
2932  * @adapter:    address of board private structure
2933  * @index:      index of ring to set
2934  **/
2935 void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2936                                    struct ixgbe_ring *ring)
2937 {
2938         struct ixgbe_hw *hw = &adapter->hw;
2939         u32 rscctrl;
2940         int rx_buf_len;
2941         u8 reg_idx = ring->reg_idx;
2942
2943         if (!ring_is_rsc_enabled(ring))
2944                 return;
2945
2946         rx_buf_len = ring->rx_buf_len;
2947         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2948         rscctrl |= IXGBE_RSCCTL_RSCEN;
2949         /*
2950          * we must limit the number of descriptors so that the
2951          * total size of max desc * buf_len is not greater
2952          * than 65535
2953          */
2954         if (ring_is_ps_enabled(ring)) {
2955 #if (MAX_SKB_FRAGS > 16)
2956                 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2957 #elif (MAX_SKB_FRAGS > 8)
2958                 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2959 #elif (MAX_SKB_FRAGS > 4)
2960                 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2961 #else
2962                 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2963 #endif
2964         } else {
2965                 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2966                         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2967                 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2968                         rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2969                 else
2970                         rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2971         }
2972         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2973 }
2974
2975 /**
2976  *  ixgbe_set_uta - Set unicast filter table address
2977  *  @adapter: board private structure
2978  *
2979  *  The unicast table address is a register array of 32-bit registers.
2980  *  The table is meant to be used in a way similar to how the MTA is used
2981  *  however due to certain limitations in the hardware it is necessary to
2982  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2983  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
2984  **/
2985 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2986 {
2987         struct ixgbe_hw *hw = &adapter->hw;
2988         int i;
2989
2990         /* The UTA table only exists on 82599 hardware and newer */
2991         if (hw->mac.type < ixgbe_mac_82599EB)
2992                 return;
2993
2994         /* we only need to do this if VMDq is enabled */
2995         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2996                 return;
2997
2998         for (i = 0; i < 128; i++)
2999                 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
3000 }
3001
3002 #define IXGBE_MAX_RX_DESC_POLL 10
3003 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3004                                        struct ixgbe_ring *ring)
3005 {
3006         struct ixgbe_hw *hw = &adapter->hw;
3007         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3008         u32 rxdctl;
3009         u8 reg_idx = ring->reg_idx;
3010
3011         /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3012         if (hw->mac.type == ixgbe_mac_82598EB &&
3013             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3014                 return;
3015
3016         do {
3017                 msleep(1);
3018                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3019         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3020
3021         if (!wait_loop) {
3022                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3023                       "the polling period\n", reg_idx);
3024         }
3025 }
3026
3027 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3028                             struct ixgbe_ring *ring)
3029 {
3030         struct ixgbe_hw *hw = &adapter->hw;
3031         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3032         u32 rxdctl;
3033         u8 reg_idx = ring->reg_idx;
3034
3035         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3036         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3037
3038         /* write value back with RXDCTL.ENABLE bit cleared */
3039         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3040
3041         if (hw->mac.type == ixgbe_mac_82598EB &&
3042             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3043                 return;
3044
3045         /* the hardware may take up to 100us to really disable the rx queue */
3046         do {
3047                 udelay(10);
3048                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3049         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3050
3051         if (!wait_loop) {
3052                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3053                       "the polling period\n", reg_idx);
3054         }
3055 }
3056
3057 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3058                              struct ixgbe_ring *ring)
3059 {
3060         struct ixgbe_hw *hw = &adapter->hw;
3061         u64 rdba = ring->dma;
3062         u32 rxdctl;
3063         u8 reg_idx = ring->reg_idx;
3064
3065         /* disable queue to avoid issues while updating state */
3066         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3067         ixgbe_disable_rx_queue(adapter, ring);
3068
3069         IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3070         IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3071         IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3072                         ring->count * sizeof(union ixgbe_adv_rx_desc));
3073         IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3074         IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3075         ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3076
3077         ixgbe_configure_srrctl(adapter, ring);
3078         ixgbe_configure_rscctl(adapter, ring);
3079
3080         if (hw->mac.type == ixgbe_mac_82598EB) {
3081                 /*
3082                  * enable cache line friendly hardware writes:
3083                  * PTHRESH=32 descriptors (half the internal cache),
3084                  * this also removes ugly rx_no_buffer_count increment
3085                  * HTHRESH=4 descriptors (to minimize latency on fetch)
3086                  * WTHRESH=8 burst writeback up to two cache lines
3087                  */
3088                 rxdctl &= ~0x3FFFFF;
3089                 rxdctl |=  0x080420;
3090         }
3091
3092         /* enable receive descriptor ring */
3093         rxdctl |= IXGBE_RXDCTL_ENABLE;
3094         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3095
3096         ixgbe_rx_desc_queue_enable(adapter, ring);
3097         ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
3098 }
3099
3100 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3101 {
3102         struct ixgbe_hw *hw = &adapter->hw;
3103         int p;
3104
3105         /* PSRTYPE must be initialized in non 82598 adapters */
3106         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3107                       IXGBE_PSRTYPE_UDPHDR |
3108                       IXGBE_PSRTYPE_IPV4HDR |
3109                       IXGBE_PSRTYPE_L2HDR |
3110                       IXGBE_PSRTYPE_IPV6HDR;
3111
3112         if (hw->mac.type == ixgbe_mac_82598EB)
3113                 return;
3114
3115         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3116                 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3117
3118         for (p = 0; p < adapter->num_rx_pools; p++)
3119                 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3120                                 psrtype);
3121 }
3122
3123 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3124 {
3125         struct ixgbe_hw *hw = &adapter->hw;
3126         u32 gcr_ext;
3127         u32 vt_reg_bits;
3128         u32 reg_offset, vf_shift;
3129         u32 vmdctl;
3130
3131         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3132                 return;
3133
3134         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3135         vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3136         vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3137         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3138
3139         vf_shift = adapter->num_vfs % 32;
3140         reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
3141
3142         /* Enable only the PF's pool for Tx/Rx */
3143         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3144         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3145         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3146         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3147         IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3148
3149         /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3150         hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3151
3152         /*
3153          * Set up VF register offsets for selected VT Mode,
3154          * i.e. 32 or 64 VFs for SR-IOV
3155          */
3156         gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3157         gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3158         gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3159         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3160
3161         /* enable Tx loopback for VF/PF communication */
3162         IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3163         /* Enable MAC Anti-Spoofing */
3164         hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3165                                           adapter->num_vfs);
3166 }
3167
3168 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3169 {
3170         struct ixgbe_hw *hw = &adapter->hw;
3171         struct net_device *netdev = adapter->netdev;
3172         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3173         int rx_buf_len;
3174         struct ixgbe_ring *rx_ring;
3175         int i;
3176         u32 mhadd, hlreg0;
3177
3178         /* Decide whether to use packet split mode or not */
3179         /* Do not use packet split if we're in SR-IOV Mode */
3180         if (!adapter->num_vfs)
3181                 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
3182
3183         /* Set the RX buffer length according to the mode */
3184         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
3185                 rx_buf_len = IXGBE_RX_HDR_SIZE;
3186         } else {
3187                 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
3188                     (netdev->mtu <= ETH_DATA_LEN))
3189                         rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3190                 else
3191                         rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
3192         }
3193
3194 #ifdef IXGBE_FCOE
3195         /* adjust max frame to be able to do baby jumbo for FCoE */
3196         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3197             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3198                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3199
3200 #endif /* IXGBE_FCOE */
3201         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3202         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3203                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3204                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3205
3206                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3207         }
3208
3209         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3210         /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3211         hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3212         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3213
3214         /*
3215          * Setup the HW Rx Head and Tail Descriptor Pointers and
3216          * the Base and Length of the Rx Descriptor Ring
3217          */
3218         for (i = 0; i < adapter->num_rx_queues; i++) {
3219                 rx_ring = adapter->rx_ring[i];
3220                 rx_ring->rx_buf_len = rx_buf_len;
3221
3222                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
3223                         set_ring_ps_enabled(rx_ring);
3224                 else
3225                         clear_ring_ps_enabled(rx_ring);
3226
3227                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3228                         set_ring_rsc_enabled(rx_ring);
3229                 else
3230                         clear_ring_rsc_enabled(rx_ring);
3231
3232 #ifdef IXGBE_FCOE
3233                 if (netdev->features & NETIF_F_FCOE_MTU) {
3234                         struct ixgbe_ring_feature *f;
3235                         f = &adapter->ring_feature[RING_F_FCOE];
3236                         if ((i >= f->mask) && (i < f->mask + f->indices)) {
3237                                 clear_ring_ps_enabled(rx_ring);
3238                                 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3239                                         rx_ring->rx_buf_len =
3240                                                 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3241                         } else if (!ring_is_rsc_enabled(rx_ring) &&
3242                                    !ring_is_ps_enabled(rx_ring)) {
3243                                 rx_ring->rx_buf_len =
3244                                                 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3245                         }
3246                 }
3247 #endif /* IXGBE_FCOE */
3248         }
3249 }
3250
3251 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3252 {
3253         struct ixgbe_hw *hw = &adapter->hw;
3254         u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3255
3256         switch (hw->mac.type) {
3257         case ixgbe_mac_82598EB:
3258                 /*
3259                  * For VMDq support of different descriptor types or
3260                  * buffer sizes through the use of multiple SRRCTL
3261                  * registers, RDRXCTL.MVMEN must be set to 1
3262                  *
3263                  * also, the manual doesn't mention it clearly but DCA hints
3264                  * will only use queue 0's tags unless this bit is set.  Side
3265                  * effects of setting this bit are only that SRRCTL must be
3266                  * fully programmed [0..15]
3267                  */
3268                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3269                 break;
3270         case ixgbe_mac_82599EB:
3271         case ixgbe_mac_X540:
3272                 /* Disable RSC for ACK packets */
3273                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3274                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3275                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3276                 /* hardware requires some bits to be set by default */
3277                 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3278                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3279                 break;
3280         default:
3281                 /* We should do nothing since we don't know this hardware */
3282                 return;
3283         }
3284
3285         IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3286 }
3287
3288 /**
3289  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3290  * @adapter: board private structure
3291  *
3292  * Configure the Rx unit of the MAC after a reset.
3293  **/
3294 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3295 {
3296         struct ixgbe_hw *hw = &adapter->hw;
3297         int i;
3298         u32 rxctrl;
3299
3300         /* disable receives while setting up the descriptors */
3301         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3302         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3303
3304         ixgbe_setup_psrtype(adapter);
3305         ixgbe_setup_rdrxctl(adapter);
3306
3307         /* Program registers for the distribution of queues */
3308         ixgbe_setup_mrqc(adapter);
3309
3310         ixgbe_set_uta(adapter);
3311
3312         /* set_rx_buffer_len must be called before ring initialization */
3313         ixgbe_set_rx_buffer_len(adapter);
3314
3315         /*
3316          * Setup the HW Rx Head and Tail Descriptor Pointers and
3317          * the Base and Length of the Rx Descriptor Ring
3318          */
3319         for (i = 0; i < adapter->num_rx_queues; i++)
3320                 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3321
3322         /* disable drop enable for 82598 parts */
3323         if (hw->mac.type == ixgbe_mac_82598EB)
3324                 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3325
3326         /* enable all receives */
3327         rxctrl |= IXGBE_RXCTRL_RXEN;
3328         hw->mac.ops.enable_rx_dma(hw, rxctrl);
3329 }
3330
3331 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3332 {
3333         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3334         struct ixgbe_hw *hw = &adapter->hw;
3335         int pool_ndx = adapter->num_vfs;
3336
3337         /* add VID to filter table */
3338         hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3339         set_bit(vid, adapter->active_vlans);
3340 }
3341
3342 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3343 {
3344         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3345         struct ixgbe_hw *hw = &adapter->hw;
3346         int pool_ndx = adapter->num_vfs;
3347
3348         /* remove VID from filter table */
3349         hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3350         clear_bit(vid, adapter->active_vlans);
3351 }
3352
3353 /**
3354  * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3355  * @adapter: driver data
3356  */
3357 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3358 {
3359         struct ixgbe_hw *hw = &adapter->hw;
3360         u32 vlnctrl;
3361
3362         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3363         vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3364         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3365 }
3366
3367 /**
3368  * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3369  * @adapter: driver data
3370  */
3371 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3372 {
3373         struct ixgbe_hw *hw = &adapter->hw;
3374         u32 vlnctrl;
3375
3376         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3377         vlnctrl |= IXGBE_VLNCTRL_VFE;
3378         vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3379         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3380 }
3381
3382 /**
3383  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3384  * @adapter: driver data
3385  */
3386 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3387 {
3388         struct ixgbe_hw *hw = &adapter->hw;
3389         u32 vlnctrl;
3390         int i, j;
3391
3392         switch (hw->mac.type) {
3393         case ixgbe_mac_82598EB:
3394                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3395                 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3396                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3397                 break;
3398         case ixgbe_mac_82599EB:
3399         case ixgbe_mac_X540:
3400                 for (i = 0; i < adapter->num_rx_queues; i++) {
3401                         j = adapter->rx_ring[i]->reg_idx;
3402                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3403                         vlnctrl &= ~IXGBE_RXDCTL_VME;
3404                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3405                 }
3406                 break;
3407         default:
3408                 break;
3409         }
3410 }
3411
3412 /**
3413  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3414  * @adapter: driver data
3415  */
3416 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3417 {
3418         struct ixgbe_hw *hw = &adapter->hw;
3419         u32 vlnctrl;
3420         int i, j;
3421
3422         switch (hw->mac.type) {
3423         case ixgbe_mac_82598EB:
3424                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3425                 vlnctrl |= IXGBE_VLNCTRL_VME;
3426                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3427                 break;
3428         case ixgbe_mac_82599EB:
3429         case ixgbe_mac_X540:
3430                 for (i = 0; i < adapter->num_rx_queues; i++) {
3431                         j = adapter->rx_ring[i]->reg_idx;
3432                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3433                         vlnctrl |= IXGBE_RXDCTL_VME;
3434                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3435                 }
3436                 break;
3437         default:
3438                 break;
3439         }
3440 }
3441
3442 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3443 {
3444         u16 vid;
3445
3446         ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3447
3448         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3449                 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3450 }
3451
3452 /**
3453  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3454  * @netdev: network interface device structure
3455  *
3456  * Writes unicast address list to the RAR table.
3457  * Returns: -ENOMEM on failure/insufficient address space
3458  *                0 on no addresses written
3459  *                X on writing X addresses to the RAR table
3460  **/
3461 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3462 {
3463         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3464         struct ixgbe_hw *hw = &adapter->hw;
3465         unsigned int vfn = adapter->num_vfs;
3466         unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3467         int count = 0;
3468
3469         /* return ENOMEM indicating insufficient memory for addresses */
3470         if (netdev_uc_count(netdev) > rar_entries)
3471                 return -ENOMEM;
3472
3473         if (!netdev_uc_empty(netdev) && rar_entries) {
3474                 struct netdev_hw_addr *ha;
3475                 /* return error if we do not support writing to RAR table */
3476                 if (!hw->mac.ops.set_rar)
3477                         return -ENOMEM;
3478
3479                 netdev_for_each_uc_addr(ha, netdev) {
3480                         if (!rar_entries)
3481                                 break;
3482                         hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3483                                             vfn, IXGBE_RAH_AV);
3484                         count++;
3485                 }
3486         }
3487         /* write the addresses in reverse order to avoid write combining */
3488         for (; rar_entries > 0 ; rar_entries--)
3489                 hw->mac.ops.clear_rar(hw, rar_entries);
3490
3491         return count;
3492 }
3493
3494 /**
3495  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3496  * @netdev: network interface device structure
3497  *
3498  * The set_rx_method entry point is called whenever the unicast/multicast
3499  * address list or the network interface flags are updated.  This routine is
3500  * responsible for configuring the hardware for proper unicast, multicast and
3501  * promiscuous mode.
3502  **/
3503 void ixgbe_set_rx_mode(struct net_device *netdev)
3504 {
3505         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3506         struct ixgbe_hw *hw = &adapter->hw;
3507         u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3508         int count;
3509
3510         /* Check for Promiscuous and All Multicast modes */
3511
3512         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3513
3514         /* set all bits that we expect to always be set */
3515         fctrl |= IXGBE_FCTRL_BAM;
3516         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3517         fctrl |= IXGBE_FCTRL_PMCF;
3518
3519         /* clear the bits we are changing the status of */
3520         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3521
3522         if (netdev->flags & IFF_PROMISC) {
3523                 hw->addr_ctrl.user_set_promisc = true;
3524                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3525                 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3526                 /* don't hardware filter vlans in promisc mode */
3527                 ixgbe_vlan_filter_disable(adapter);
3528         } else {
3529                 if (netdev->flags & IFF_ALLMULTI) {
3530                         fctrl |= IXGBE_FCTRL_MPE;
3531                         vmolr |= IXGBE_VMOLR_MPE;
3532                 } else {
3533                         /*
3534                          * Write addresses to the MTA, if the attempt fails
3535                          * then we should just turn on promiscous mode so
3536                          * that we can at least receive multicast traffic
3537                          */
3538                         hw->mac.ops.update_mc_addr_list(hw, netdev);
3539                         vmolr |= IXGBE_VMOLR_ROMPE;
3540                 }
3541                 ixgbe_vlan_filter_enable(adapter);
3542                 hw->addr_ctrl.user_set_promisc = false;
3543                 /*
3544                  * Write addresses to available RAR registers, if there is not
3545                  * sufficient space to store all the addresses then enable
3546                  * unicast promiscous mode
3547                  */
3548                 count = ixgbe_write_uc_addr_list(netdev);
3549                 if (count < 0) {
3550                         fctrl |= IXGBE_FCTRL_UPE;
3551                         vmolr |= IXGBE_VMOLR_ROPE;
3552                 }
3553         }
3554
3555         if (adapter->num_vfs) {
3556                 ixgbe_restore_vf_multicasts(adapter);
3557                 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3558                          ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3559                            IXGBE_VMOLR_ROPE);
3560                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3561         }
3562
3563         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3564
3565         if (netdev->features & NETIF_F_HW_VLAN_RX)
3566                 ixgbe_vlan_strip_enable(adapter);
3567         else
3568                 ixgbe_vlan_strip_disable(adapter);
3569 }
3570
3571 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3572 {
3573         int q_idx;
3574         struct ixgbe_q_vector *q_vector;
3575         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3576
3577         /* legacy and MSI only use one vector */
3578         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3579                 q_vectors = 1;
3580
3581         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3582                 struct napi_struct *napi;
3583                 q_vector = adapter->q_vector[q_idx];
3584                 napi = &q_vector->napi;
3585                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3586                         if (!q_vector->rxr_count || !q_vector->txr_count) {
3587                                 if (q_vector->txr_count == 1)
3588                                         napi->poll = &ixgbe_clean_txonly;
3589                                 else if (q_vector->rxr_count == 1)
3590                                         napi->poll = &ixgbe_clean_rxonly;
3591                         }
3592                 }
3593
3594                 napi_enable(napi);
3595         }
3596 }
3597
3598 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3599 {
3600         int q_idx;
3601         struct ixgbe_q_vector *q_vector;
3602         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3603
3604         /* legacy and MSI only use one vector */
3605         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3606                 q_vectors = 1;
3607
3608         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3609                 q_vector = adapter->q_vector[q_idx];
3610                 napi_disable(&q_vector->napi);
3611         }
3612 }
3613
3614 #ifdef CONFIG_IXGBE_DCB
3615 /*
3616  * ixgbe_configure_dcb - Configure DCB hardware
3617  * @adapter: ixgbe adapter struct
3618  *
3619  * This is called by the driver on open to configure the DCB hardware.
3620  * This is also called by the gennetlink interface when reconfiguring
3621  * the DCB state.
3622  */
3623 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3624 {
3625         struct ixgbe_hw *hw = &adapter->hw;
3626         int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3627
3628         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3629                 if (hw->mac.type == ixgbe_mac_82598EB)
3630                         netif_set_gso_max_size(adapter->netdev, 65536);
3631                 return;
3632         }
3633
3634         if (hw->mac.type == ixgbe_mac_82598EB)
3635                 netif_set_gso_max_size(adapter->netdev, 32768);
3636
3637 #ifdef CONFIG_FCOE
3638         if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3639                 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3640 #endif
3641
3642         ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3643                                         DCB_TX_CONFIG);
3644         ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3645                                         DCB_RX_CONFIG);
3646
3647         /* Enable VLAN tag insert/strip */
3648         adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3649
3650         hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3651
3652         /* reconfigure the hardware */
3653         ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3654 }
3655
3656 #endif
3657 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3658 {
3659         struct net_device *netdev = adapter->netdev;
3660         struct ixgbe_hw *hw = &adapter->hw;
3661         int i;
3662
3663 #ifdef CONFIG_IXGBE_DCB
3664         ixgbe_configure_dcb(adapter);
3665 #endif
3666
3667         ixgbe_set_rx_mode(netdev);
3668         ixgbe_restore_vlan(adapter);
3669
3670 #ifdef IXGBE_FCOE
3671         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3672                 ixgbe_configure_fcoe(adapter);
3673
3674 #endif /* IXGBE_FCOE */
3675         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3676                 for (i = 0; i < adapter->num_tx_queues; i++)
3677                         adapter->tx_ring[i]->atr_sample_rate =
3678                                                        adapter->atr_sample_rate;
3679                 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3680         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3681                 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3682         }
3683         ixgbe_configure_virtualization(adapter);
3684
3685         ixgbe_configure_tx(adapter);
3686         ixgbe_configure_rx(adapter);
3687 }
3688
3689 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3690 {
3691         switch (hw->phy.type) {
3692         case ixgbe_phy_sfp_avago:
3693         case ixgbe_phy_sfp_ftl:
3694         case ixgbe_phy_sfp_intel:
3695         case ixgbe_phy_sfp_unknown:
3696         case ixgbe_phy_sfp_passive_tyco:
3697         case ixgbe_phy_sfp_passive_unknown:
3698         case ixgbe_phy_sfp_active_unknown:
3699         case ixgbe_phy_sfp_ftl_active:
3700                 return true;
3701         default:
3702                 return false;
3703         }
3704 }
3705
3706 /**
3707  * ixgbe_sfp_link_config - set up SFP+ link
3708  * @adapter: pointer to private adapter struct
3709  **/
3710 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3711 {
3712         struct ixgbe_hw *hw = &adapter->hw;
3713
3714                 if (hw->phy.multispeed_fiber) {
3715                         /*
3716                          * In multispeed fiber setups, the device may not have
3717                          * had a physical connection when the driver loaded.
3718                          * If that's the case, the initial link configuration
3719                          * couldn't get the MAC into 10G or 1G mode, so we'll
3720                          * never have a link status change interrupt fire.
3721                          * We need to try and force an autonegotiation
3722                          * session, then bring up link.
3723                          */
3724                         hw->mac.ops.setup_sfp(hw);
3725                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3726                                 schedule_work(&adapter->multispeed_fiber_task);
3727                 } else {
3728                         /*
3729                          * Direct Attach Cu and non-multispeed fiber modules
3730                          * still need to be configured properly prior to
3731                          * attempting link.
3732                          */
3733                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3734                                 schedule_work(&adapter->sfp_config_module_task);
3735                 }
3736 }
3737
3738 /**
3739  * ixgbe_non_sfp_link_config - set up non-SFP+ link
3740  * @hw: pointer to private hardware struct
3741  *
3742  * Returns 0 on success, negative on failure
3743  **/
3744 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3745 {
3746         u32 autoneg;
3747         bool negotiation, link_up = false;
3748         u32 ret = IXGBE_ERR_LINK_SETUP;
3749
3750         if (hw->mac.ops.check_link)
3751                 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3752
3753         if (ret)
3754                 goto link_cfg_out;
3755
3756         if (hw->mac.ops.get_link_capabilities)
3757                 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3758                                                         &negotiation);
3759         if (ret)
3760                 goto link_cfg_out;
3761
3762         if (hw->mac.ops.setup_link)
3763                 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3764 link_cfg_out:
3765         return ret;
3766 }
3767
3768 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3769 {
3770         struct ixgbe_hw *hw = &adapter->hw;
3771         u32 gpie = 0;
3772
3773         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3774                 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3775                        IXGBE_GPIE_OCD;
3776                 gpie |= IXGBE_GPIE_EIAME;
3777                 /*
3778                  * use EIAM to auto-mask when MSI-X interrupt is asserted
3779                  * this saves a register write for every interrupt
3780                  */
3781                 switch (hw->mac.type) {
3782                 case ixgbe_mac_82598EB:
3783                         IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3784                         break;
3785                 case ixgbe_mac_82599EB:
3786                 case ixgbe_mac_X540:
3787                 default:
3788                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3789                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3790                         break;
3791                 }
3792         } else {
3793                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3794                  * specifically only auto mask tx and rx interrupts */
3795                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3796         }
3797
3798         /* XXX: to interrupt immediately for EICS writes, enable this */
3799         /* gpie |= IXGBE_GPIE_EIMEN; */
3800
3801         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3802                 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3803                 gpie |= IXGBE_GPIE_VTMODE_64;
3804         }
3805
3806         /* Enable fan failure interrupt */
3807         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3808                 gpie |= IXGBE_SDP1_GPIEN;
3809
3810         if (hw->mac.type == ixgbe_mac_82599EB)
3811                 gpie |= IXGBE_SDP1_GPIEN;
3812                 gpie |= IXGBE_SDP2_GPIEN;
3813
3814         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3815 }
3816
3817 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3818 {
3819         struct ixgbe_hw *hw = &adapter->hw;
3820         int err;
3821         u32 ctrl_ext;
3822
3823         ixgbe_get_hw_control(adapter);
3824         ixgbe_setup_gpie(adapter);
3825
3826         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3827                 ixgbe_configure_msix(adapter);
3828         else
3829                 ixgbe_configure_msi_and_legacy(adapter);
3830
3831         /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3832         if (hw->mac.ops.enable_tx_laser &&
3833             ((hw->phy.multispeed_fiber) ||
3834              ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3835               (hw->mac.type == ixgbe_mac_82599EB))))
3836                 hw->mac.ops.enable_tx_laser(hw);
3837
3838         clear_bit(__IXGBE_DOWN, &adapter->state);
3839         ixgbe_napi_enable_all(adapter);
3840
3841         if (ixgbe_is_sfp(hw)) {
3842                 ixgbe_sfp_link_config(adapter);
3843         } else {
3844                 err = ixgbe_non_sfp_link_config(hw);
3845                 if (err)
3846                         e_err(probe, "link_config FAILED %d\n", err);
3847         }
3848
3849         /* clear any pending interrupts, may auto mask */
3850         IXGBE_READ_REG(hw, IXGBE_EICR);
3851         ixgbe_irq_enable(adapter, true, true);
3852
3853         /*
3854          * If this adapter has a fan, check to see if we had a failure
3855          * before we enabled the interrupt.
3856          */
3857         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3858                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3859                 if (esdp & IXGBE_ESDP_SDP1)
3860                         e_crit(drv, "Fan has stopped, replace the adapter\n");
3861         }
3862
3863         /*
3864          * For hot-pluggable SFP+ devices, a new SFP+ module may have
3865          * arrived before interrupts were enabled but after probe.  Such
3866          * devices wouldn't have their type identified yet. We need to
3867          * kick off the SFP+ module setup first, then try to bring up link.
3868          * If we're not hot-pluggable SFP+, we just need to configure link
3869          * and bring it up.
3870          */
3871         if (hw->phy.type == ixgbe_phy_unknown)
3872                 schedule_work(&adapter->sfp_config_module_task);
3873
3874         /* enable transmits */
3875         netif_tx_start_all_queues(adapter->netdev);
3876
3877         /* bring the link up in the watchdog, this could race with our first
3878          * link up interrupt but shouldn't be a problem */
3879         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3880         adapter->link_check_timeout = jiffies;
3881         mod_timer(&adapter->watchdog_timer, jiffies);
3882
3883         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3884         ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3885         ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3886         IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3887
3888         return 0;
3889 }
3890
3891 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3892 {
3893         WARN_ON(in_interrupt());
3894         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3895                 msleep(1);
3896         ixgbe_down(adapter);
3897         /*
3898          * If SR-IOV enabled then wait a bit before bringing the adapter
3899          * back up to give the VFs time to respond to the reset.  The
3900          * two second wait is based upon the watchdog timer cycle in
3901          * the VF driver.
3902          */
3903         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3904                 msleep(2000);
3905         ixgbe_up(adapter);
3906         clear_bit(__IXGBE_RESETTING, &adapter->state);
3907 }
3908
3909 int ixgbe_up(struct ixgbe_adapter *adapter)
3910 {
3911         /* hardware has been reset, we need to reload some things */
3912         ixgbe_configure(adapter);
3913
3914         return ixgbe_up_complete(adapter);
3915 }
3916
3917 void ixgbe_reset(struct ixgbe_adapter *adapter)
3918 {
3919         struct ixgbe_hw *hw = &adapter->hw;
3920         int err;
3921
3922         err = hw->mac.ops.init_hw(hw);
3923         switch (err) {
3924         case 0:
3925         case IXGBE_ERR_SFP_NOT_PRESENT:
3926                 break;
3927         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3928                 e_dev_err("master disable timed out\n");
3929                 break;
3930         case IXGBE_ERR_EEPROM_VERSION:
3931                 /* We are running on a pre-production device, log a warning */
3932                 e_dev_warn("This device is a pre-production adapter/LOM. "
3933                            "Please be aware there may be issuesassociated with "
3934                            "your hardware.  If you are experiencing problems "
3935                            "please contact your Intel or hardware "
3936                            "representative who provided you with this "
3937                            "hardware.\n");
3938                 break;
3939         default:
3940                 e_dev_err("Hardware Error: %d\n", err);
3941         }
3942
3943         /* reprogram the RAR[0] in case user changed it. */
3944         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3945                             IXGBE_RAH_AV);
3946 }
3947
3948 /**
3949  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3950  * @rx_ring: ring to free buffers from
3951  **/
3952 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3953 {
3954         struct device *dev = rx_ring->dev;
3955         unsigned long size;
3956         u16 i;
3957
3958         /* ring already cleared, nothing to do */
3959         if (!rx_ring->rx_buffer_info)
3960                 return;
3961
3962         /* Free all the Rx ring sk_buffs */
3963         for (i = 0; i < rx_ring->count; i++) {
3964                 struct ixgbe_rx_buffer *rx_buffer_info;
3965
3966                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3967                 if (rx_buffer_info->dma) {
3968                         dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3969                                          rx_ring->rx_buf_len,
3970                                          DMA_FROM_DEVICE);
3971                         rx_buffer_info->dma = 0;
3972                 }
3973                 if (rx_buffer_info->skb) {
3974                         struct sk_buff *skb = rx_buffer_info->skb;
3975                         rx_buffer_info->skb = NULL;
3976                         do {
3977                                 struct sk_buff *this = skb;
3978                                 if (IXGBE_RSC_CB(this)->delay_unmap) {
3979                                         dma_unmap_single(dev,
3980                                                          IXGBE_RSC_CB(this)->dma,
3981                                                          rx_ring->rx_buf_len,
3982                                                          DMA_FROM_DEVICE);
3983                                         IXGBE_RSC_CB(this)->dma = 0;
3984                                         IXGBE_RSC_CB(skb)->delay_unmap = false;
3985                                 }
3986                                 skb = skb->prev;
3987                                 dev_kfree_skb(this);
3988                         } while (skb);
3989                 }
3990                 if (!rx_buffer_info->page)
3991                         continue;
3992                 if (rx_buffer_info->page_dma) {
3993                         dma_unmap_page(dev, rx_buffer_info->page_dma,
3994                                        PAGE_SIZE / 2, DMA_FROM_DEVICE);
3995                         rx_buffer_info->page_dma = 0;
3996                 }
3997                 put_page(rx_buffer_info->page);
3998                 rx_buffer_info->page = NULL;
3999                 rx_buffer_info->page_offset = 0;
4000         }
4001
4002         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4003         memset(rx_ring->rx_buffer_info, 0, size);
4004
4005         /* Zero out the descriptor ring */
4006         memset(rx_ring->desc, 0, rx_ring->size);
4007
4008         rx_ring->next_to_clean = 0;
4009         rx_ring->next_to_use = 0;
4010 }
4011
4012 /**
4013  * ixgbe_clean_tx_ring - Free Tx Buffers
4014  * @tx_ring: ring to be cleaned
4015  **/
4016 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4017 {
4018         struct ixgbe_tx_buffer *tx_buffer_info;
4019         unsigned long size;
4020         u16 i;
4021
4022         /* ring already cleared, nothing to do */
4023         if (!tx_ring->tx_buffer_info)
4024                 return;
4025
4026         /* Free all the Tx ring sk_buffs */
4027         for (i = 0; i < tx_ring->count; i++) {
4028                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4029                 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4030         }
4031
4032         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4033         memset(tx_ring->tx_buffer_info, 0, size);
4034
4035         /* Zero out the descriptor ring */
4036         memset(tx_ring->desc, 0, tx_ring->size);
4037
4038         tx_ring->next_to_use = 0;
4039         tx_ring->next_to_clean = 0;
4040 }
4041
4042 /**
4043  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4044  * @adapter: board private structure
4045  **/
4046 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4047 {
4048         int i;
4049
4050         for (i = 0; i < adapter->num_rx_queues; i++)
4051                 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4052 }
4053
4054 /**
4055  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4056  * @adapter: board private structure
4057  **/
4058 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4059 {
4060         int i;
4061
4062         for (i = 0; i < adapter->num_tx_queues; i++)
4063                 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4064 }
4065
4066 void ixgbe_down(struct ixgbe_adapter *adapter)
4067 {
4068         struct net_device *netdev = adapter->netdev;
4069         struct ixgbe_hw *hw = &adapter->hw;
4070         u32 rxctrl;
4071         u32 txdctl;
4072         int i;
4073         int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4074
4075         /* signal that we are down to the interrupt handler */
4076         set_bit(__IXGBE_DOWN, &adapter->state);
4077
4078         /* disable receive for all VFs and wait one second */
4079         if (adapter->num_vfs) {
4080                 /* ping all the active vfs to let them know we are going down */
4081                 ixgbe_ping_all_vfs(adapter);
4082
4083                 /* Disable all VFTE/VFRE TX/RX */
4084                 ixgbe_disable_tx_rx(adapter);
4085
4086                 /* Mark all the VFs as inactive */
4087                 for (i = 0 ; i < adapter->num_vfs; i++)
4088                         adapter->vfinfo[i].clear_to_send = 0;
4089         }
4090
4091         /* disable receives */
4092         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4093         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4094
4095         /* disable all enabled rx queues */
4096         for (i = 0; i < adapter->num_rx_queues; i++)
4097                 /* this call also flushes the previous write */
4098                 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4099
4100         msleep(10);
4101
4102         netif_tx_stop_all_queues(netdev);
4103
4104         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4105         del_timer_sync(&adapter->sfp_timer);
4106         del_timer_sync(&adapter->watchdog_timer);
4107         cancel_work_sync(&adapter->watchdog_task);
4108
4109         netif_carrier_off(netdev);
4110         netif_tx_disable(netdev);
4111
4112         ixgbe_irq_disable(adapter);
4113
4114         ixgbe_napi_disable_all(adapter);
4115
4116         /* Cleanup the affinity_hint CPU mask memory and callback */
4117         for (i = 0; i < num_q_vectors; i++) {
4118                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
4119                 /* clear the affinity_mask in the IRQ descriptor */
4120                 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
4121                 /* release the CPU mask memory */
4122                 free_cpumask_var(q_vector->affinity_mask);
4123         }
4124
4125         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4126             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4127                 cancel_work_sync(&adapter->fdir_reinit_task);
4128
4129         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
4130                 cancel_work_sync(&adapter->check_overtemp_task);
4131
4132         /* disable transmits in the hardware now that interrupts are off */
4133         for (i = 0; i < adapter->num_tx_queues; i++) {
4134                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4135                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
4136                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
4137                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
4138         }
4139         /* Disable the Tx DMA engine on 82599 */
4140         switch (hw->mac.type) {
4141         case ixgbe_mac_82599EB:
4142         case ixgbe_mac_X540:
4143                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4144                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4145                                  ~IXGBE_DMATXCTL_TE));
4146                 break;
4147         default:
4148                 break;
4149         }
4150
4151         /* clear n-tuple filters that are cached */
4152         ethtool_ntuple_flush(netdev);
4153
4154         if (!pci_channel_offline(adapter->pdev))
4155                 ixgbe_reset(adapter);
4156
4157         /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4158         if (hw->mac.ops.disable_tx_laser &&
4159             ((hw->phy.multispeed_fiber) ||
4160              ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4161               (hw->mac.type == ixgbe_mac_82599EB))))
4162                 hw->mac.ops.disable_tx_laser(hw);
4163
4164         ixgbe_clean_all_tx_rings(adapter);
4165         ixgbe_clean_all_rx_rings(adapter);
4166
4167 #ifdef CONFIG_IXGBE_DCA
4168         /* since we reset the hardware DCA settings were cleared */
4169         ixgbe_setup_dca(adapter);
4170 #endif
4171 }
4172
4173 /**
4174  * ixgbe_poll - NAPI Rx polling callback
4175  * @napi: structure for representing this polling device
4176  * @budget: how many packets driver is allowed to clean
4177  *
4178  * This function is used for legacy and MSI, NAPI mode
4179  **/
4180 static int ixgbe_poll(struct napi_struct *napi, int budget)
4181 {
4182         struct ixgbe_q_vector *q_vector =
4183                                 container_of(napi, struct ixgbe_q_vector, napi);
4184         struct ixgbe_adapter *adapter = q_vector->adapter;
4185         int tx_clean_complete, work_done = 0;
4186
4187 #ifdef CONFIG_IXGBE_DCA
4188         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
4189                 ixgbe_update_dca(q_vector);
4190 #endif
4191
4192         tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
4193         ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
4194
4195         if (!tx_clean_complete)
4196                 work_done = budget;
4197
4198         /* If budget not fully consumed, exit the polling mode */
4199         if (work_done < budget) {
4200                 napi_complete(napi);
4201                 if (adapter->rx_itr_setting & 1)
4202                         ixgbe_set_itr(adapter);
4203                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
4204                         ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
4205         }
4206         return work_done;
4207 }
4208
4209 /**
4210  * ixgbe_tx_timeout - Respond to a Tx Hang
4211  * @netdev: network interface device structure
4212  **/
4213 static void ixgbe_tx_timeout(struct net_device *netdev)
4214 {
4215         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4216
4217         adapter->tx_timeout_count++;
4218
4219         /* Do the reset outside of interrupt context */
4220         schedule_work(&adapter->reset_task);
4221 }
4222
4223 static void ixgbe_reset_task(struct work_struct *work)
4224 {
4225         struct ixgbe_adapter *adapter;
4226         adapter = container_of(work, struct ixgbe_adapter, reset_task);
4227
4228         /* If we're already down or resetting, just bail */
4229         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
4230             test_bit(__IXGBE_RESETTING, &adapter->state))
4231                 return;
4232
4233         ixgbe_dump(adapter);
4234         netdev_err(adapter->netdev, "Reset adapter\n");
4235         ixgbe_reinit_locked(adapter);
4236 }
4237
4238 #ifdef CONFIG_IXGBE_DCB
4239 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4240 {
4241         bool ret = false;
4242         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
4243
4244         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4245                 return ret;
4246
4247         f->mask = 0x7 << 3;
4248         adapter->num_rx_queues = f->indices;
4249         adapter->num_tx_queues = f->indices;
4250         ret = true;
4251
4252         return ret;
4253 }
4254 #endif
4255
4256 /**
4257  * ixgbe_set_rss_queues: Allocate queues for RSS
4258  * @adapter: board private structure to initialize
4259  *
4260  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
4261  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4262  *
4263  **/
4264 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4265 {
4266         bool ret = false;
4267         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4268
4269         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4270                 f->mask = 0xF;
4271                 adapter->num_rx_queues = f->indices;
4272                 adapter->num_tx_queues = f->indices;
4273                 ret = true;
4274         } else {
4275                 ret = false;
4276         }
4277
4278         return ret;
4279 }
4280
4281 /**
4282  * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4283  * @adapter: board private structure to initialize
4284  *
4285  * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4286  * to the original CPU that initiated the Tx session.  This runs in addition
4287  * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4288  * Rx load across CPUs using RSS.
4289  *
4290  **/
4291 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4292 {
4293         bool ret = false;
4294         struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4295
4296         f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4297         f_fdir->mask = 0;
4298
4299         /* Flow Director must have RSS enabled */
4300         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4301             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4302              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4303                 adapter->num_tx_queues = f_fdir->indices;
4304                 adapter->num_rx_queues = f_fdir->indices;
4305                 ret = true;
4306         } else {
4307                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4308                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4309         }
4310         return ret;
4311 }
4312
4313 #ifdef IXGBE_FCOE
4314 /**
4315  * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4316  * @adapter: board private structure to initialize
4317  *
4318  * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4319  * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4320  * rx queues out of the max number of rx queues, instead, it is used as the
4321  * index of the first rx queue used by FCoE.
4322  *
4323  **/
4324 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4325 {
4326         bool ret = false;
4327         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4328
4329         f->indices = min((int)num_online_cpus(), f->indices);
4330         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4331                 adapter->num_rx_queues = 1;
4332                 adapter->num_tx_queues = 1;
4333 #ifdef CONFIG_IXGBE_DCB
4334                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4335                         e_info(probe, "FCoE enabled with DCB\n");
4336                         ixgbe_set_dcb_queues(adapter);
4337                 }
4338 #endif
4339                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4340                         e_info(probe, "FCoE enabled with RSS\n");
4341                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4342                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4343                                 ixgbe_set_fdir_queues(adapter);
4344                         else
4345                                 ixgbe_set_rss_queues(adapter);
4346                 }
4347                 /* adding FCoE rx rings to the end */
4348                 f->mask = adapter->num_rx_queues;
4349                 adapter->num_rx_queues += f->indices;
4350                 adapter->num_tx_queues += f->indices;
4351
4352                 ret = true;
4353         }
4354
4355         return ret;
4356 }
4357
4358 #endif /* IXGBE_FCOE */
4359 /**
4360  * ixgbe_set_sriov_queues: Allocate queues for IOV use
4361  * @adapter: board private structure to initialize
4362  *
4363  * IOV doesn't actually use anything, so just NAK the
4364  * request for now and let the other queue routines
4365  * figure out what to do.
4366  */
4367 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4368 {
4369         return false;
4370 }
4371
4372 /*
4373  * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4374  * @adapter: board private structure to initialize
4375  *
4376  * This is the top level queue allocation routine.  The order here is very
4377  * important, starting with the "most" number of features turned on at once,
4378  * and ending with the smallest set of features.  This way large combinations
4379  * can be allocated if they're turned on, and smaller combinations are the
4380  * fallthrough conditions.
4381  *
4382  **/
4383 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4384 {
4385         /* Start with base case */
4386         adapter->num_rx_queues = 1;
4387         adapter->num_tx_queues = 1;
4388         adapter->num_rx_pools = adapter->num_rx_queues;
4389         adapter->num_rx_queues_per_pool = 1;
4390
4391         if (ixgbe_set_sriov_queues(adapter))
4392                 goto done;
4393
4394 #ifdef IXGBE_FCOE
4395         if (ixgbe_set_fcoe_queues(adapter))
4396                 goto done;
4397
4398 #endif /* IXGBE_FCOE */
4399 #ifdef CONFIG_IXGBE_DCB
4400         if (ixgbe_set_dcb_queues(adapter))
4401                 goto done;
4402
4403 #endif
4404         if (ixgbe_set_fdir_queues(adapter))
4405                 goto done;
4406
4407         if (ixgbe_set_rss_queues(adapter))
4408                 goto done;
4409
4410         /* fallback to base case */
4411         adapter->num_rx_queues = 1;
4412         adapter->num_tx_queues = 1;
4413
4414 done:
4415         /* Notify the stack of the (possibly) reduced queue counts. */
4416         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4417         return netif_set_real_num_rx_queues(adapter->netdev,
4418                                             adapter->num_rx_queues);
4419 }
4420
4421 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4422                                        int vectors)
4423 {
4424         int err, vector_threshold;
4425
4426         /* We'll want at least 3 (vector_threshold):
4427          * 1) TxQ[0] Cleanup
4428          * 2) RxQ[0] Cleanup
4429          * 3) Other (Link Status Change, etc.)
4430          * 4) TCP Timer (optional)
4431          */
4432         vector_threshold = MIN_MSIX_COUNT;
4433
4434         /* The more we get, the more we will assign to Tx/Rx Cleanup
4435          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4436          * Right now, we simply care about how many we'll get; we'll
4437          * set them up later while requesting irq's.
4438          */
4439         while (vectors >= vector_threshold) {
4440                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4441                                       vectors);
4442                 if (!err) /* Success in acquiring all requested vectors. */
4443                         break;
4444                 else if (err < 0)
4445                         vectors = 0; /* Nasty failure, quit now */
4446                 else /* err == number of vectors we should try again with */
4447                         vectors = err;
4448         }
4449
4450         if (vectors < vector_threshold) {
4451                 /* Can't allocate enough MSI-X interrupts?  Oh well.
4452                  * This just means we'll go with either a single MSI
4453                  * vector or fall back to legacy interrupts.
4454                  */
4455                 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4456                              "Unable to allocate MSI-X interrupts\n");
4457                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4458                 kfree(adapter->msix_entries);
4459                 adapter->msix_entries = NULL;
4460         } else {
4461                 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4462                 /*
4463                  * Adjust for only the vectors we'll use, which is minimum
4464                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4465                  * vectors we were allocated.
4466                  */
4467                 adapter->num_msix_vectors = min(vectors,
4468                                    adapter->max_msix_q_vectors + NON_Q_VECTORS);
4469         }
4470 }
4471
4472 /**
4473  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4474  * @adapter: board private structure to initialize
4475  *
4476  * Cache the descriptor ring offsets for RSS to the assigned rings.
4477  *
4478  **/
4479 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4480 {
4481         int i;
4482
4483         if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4484                 return false;
4485
4486         for (i = 0; i < adapter->num_rx_queues; i++)
4487                 adapter->rx_ring[i]->reg_idx = i;
4488         for (i = 0; i < adapter->num_tx_queues; i++)
4489                 adapter->tx_ring[i]->reg_idx = i;
4490
4491         return true;
4492 }
4493
4494 #ifdef CONFIG_IXGBE_DCB
4495 /**
4496  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4497  * @adapter: board private structure to initialize
4498  *
4499  * Cache the descriptor ring offsets for DCB to the assigned rings.
4500  *
4501  **/
4502 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4503 {
4504         int i;
4505         bool ret = false;
4506         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4507
4508         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4509                 return false;
4510
4511         /* the number of queues is assumed to be symmetric */
4512         switch (adapter->hw.mac.type) {
4513         case ixgbe_mac_82598EB:
4514                 for (i = 0; i < dcb_i; i++) {
4515                         adapter->rx_ring[i]->reg_idx = i << 3;
4516                         adapter->tx_ring[i]->reg_idx = i << 2;
4517                 }
4518                 ret = true;
4519                 break;
4520         case ixgbe_mac_82599EB:
4521         case ixgbe_mac_X540:
4522                 if (dcb_i == 8) {
4523                         /*
4524                          * Tx TC0 starts at: descriptor queue 0
4525                          * Tx TC1 starts at: descriptor queue 32
4526                          * Tx TC2 starts at: descriptor queue 64
4527                          * Tx TC3 starts at: descriptor queue 80
4528                          * Tx TC4 starts at: descriptor queue 96
4529                          * Tx TC5 starts at: descriptor queue 104
4530                          * Tx TC6 starts at: descriptor queue 112
4531                          * Tx TC7 starts at: descriptor queue 120
4532                          *
4533                          * Rx TC0-TC7 are offset by 16 queues each
4534                          */
4535                         for (i = 0; i < 3; i++) {
4536                                 adapter->tx_ring[i]->reg_idx = i << 5;
4537                                 adapter->rx_ring[i]->reg_idx = i << 4;
4538                         }
4539                         for ( ; i < 5; i++) {
4540                                 adapter->tx_ring[i]->reg_idx = ((i + 2) << 4);
4541                                 adapter->rx_ring[i]->reg_idx = i << 4;
4542                         }
4543                         for ( ; i < dcb_i; i++) {
4544                                 adapter->tx_ring[i]->reg_idx = ((i + 8) << 3);
4545                                 adapter->rx_ring[i]->reg_idx = i << 4;
4546                         }
4547                         ret = true;
4548                 } else if (dcb_i == 4) {
4549                         /*
4550                          * Tx TC0 starts at: descriptor queue 0
4551                          * Tx TC1 starts at: descriptor queue 64
4552                          * Tx TC2 starts at: descriptor queue 96
4553                          * Tx TC3 starts at: descriptor queue 112
4554                          *
4555                          * Rx TC0-TC3 are offset by 32 queues each
4556                          */
4557                         adapter->tx_ring[0]->reg_idx = 0;
4558                         adapter->tx_ring[1]->reg_idx = 64;
4559                         adapter->tx_ring[2]->reg_idx = 96;
4560                         adapter->tx_ring[3]->reg_idx = 112;
4561                         for (i = 0 ; i < dcb_i; i++)
4562                                 adapter->rx_ring[i]->reg_idx = i << 5;
4563                         ret = true;
4564                 }
4565                 break;
4566         default:
4567                 break;
4568         }
4569         return ret;
4570 }
4571 #endif
4572
4573 /**
4574  * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4575  * @adapter: board private structure to initialize
4576  *
4577  * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4578  *
4579  **/
4580 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4581 {
4582         int i;
4583         bool ret = false;
4584
4585         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4586             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4587              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4588                 for (i = 0; i < adapter->num_rx_queues; i++)
4589                         adapter->rx_ring[i]->reg_idx = i;
4590                 for (i = 0; i < adapter->num_tx_queues; i++)
4591                         adapter->tx_ring[i]->reg_idx = i;
4592                 ret = true;
4593         }
4594
4595         return ret;
4596 }
4597
4598 #ifdef IXGBE_FCOE
4599 /**
4600  * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4601  * @adapter: board private structure to initialize
4602  *
4603  * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4604  *
4605  */
4606 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4607 {
4608         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4609         int i;
4610         u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4611
4612         if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4613                 return false;
4614
4615 #ifdef CONFIG_IXGBE_DCB
4616         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4617                 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4618
4619                 ixgbe_cache_ring_dcb(adapter);
4620                 /* find out queues in TC for FCoE */
4621                 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4622                 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4623                 /*
4624                  * In 82599, the number of Tx queues for each traffic
4625                  * class for both 8-TC and 4-TC modes are:
4626                  * TCs  : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4627                  * 8 TCs:  32  32  16  16   8   8   8   8
4628                  * 4 TCs:  64  64  32  32
4629                  * We have max 8 queues for FCoE, where 8 the is
4630                  * FCoE redirection table size. If TC for FCoE is
4631                  * less than or equal to TC3, we have enough queues
4632                  * to add max of 8 queues for FCoE, so we start FCoE
4633                  * Tx queue from the next one, i.e., reg_idx + 1.
4634                  * If TC for FCoE is above TC3, implying 8 TC mode,
4635                  * and we need 8 for FCoE, we have to take all queues
4636                  * in that traffic class for FCoE.
4637                  */
4638                 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4639                         fcoe_tx_i--;
4640         }
4641 #endif /* CONFIG_IXGBE_DCB */
4642         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4643                 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4644                     (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4645                         ixgbe_cache_ring_fdir(adapter);
4646                 else
4647                         ixgbe_cache_ring_rss(adapter);
4648
4649                 fcoe_rx_i = f->mask;
4650                 fcoe_tx_i = f->mask;
4651         }
4652         for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4653                 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4654                 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4655         }
4656         return true;
4657 }
4658
4659 #endif /* IXGBE_FCOE */
4660 /**
4661  * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4662  * @adapter: board private structure to initialize
4663  *
4664  * SR-IOV doesn't use any descriptor rings but changes the default if
4665  * no other mapping is used.
4666  *
4667  */
4668 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4669 {
4670         adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4671         adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4672         if (adapter->num_vfs)
4673                 return true;
4674         else
4675                 return false;
4676 }
4677
4678 /**
4679  * ixgbe_cache_ring_register - Descriptor ring to register mapping
4680  * @adapter: board private structure to initialize
4681  *
4682  * Once we know the feature-set enabled for the device, we'll cache
4683  * the register offset the descriptor ring is assigned to.
4684  *
4685  * Note, the order the various feature calls is important.  It must start with
4686  * the "most" features enabled at the same time, then trickle down to the
4687  * least amount of features turned on at once.
4688  **/
4689 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4690 {
4691         /* start with default case */
4692         adapter->rx_ring[0]->reg_idx = 0;
4693         adapter->tx_ring[0]->reg_idx = 0;
4694
4695         if (ixgbe_cache_ring_sriov(adapter))
4696                 return;
4697
4698 #ifdef IXGBE_FCOE
4699         if (ixgbe_cache_ring_fcoe(adapter))
4700                 return;
4701
4702 #endif /* IXGBE_FCOE */
4703 #ifdef CONFIG_IXGBE_DCB
4704         if (ixgbe_cache_ring_dcb(adapter))
4705                 return;
4706
4707 #endif
4708         if (ixgbe_cache_ring_fdir(adapter))
4709                 return;
4710
4711         if (ixgbe_cache_ring_rss(adapter))
4712                 return;
4713 }
4714
4715 /**
4716  * ixgbe_alloc_queues - Allocate memory for all rings
4717  * @adapter: board private structure to initialize
4718  *
4719  * We allocate one ring per queue at run-time since we don't know the
4720  * number of queues at compile-time.  The polling_netdev array is
4721  * intended for Multiqueue, but should work fine with a single queue.
4722  **/
4723 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4724 {
4725         int rx = 0, tx = 0, nid = adapter->node;
4726
4727         if (nid < 0 || !node_online(nid))
4728                 nid = first_online_node;
4729
4730         for (; tx < adapter->num_tx_queues; tx++) {
4731                 struct ixgbe_ring *ring;
4732
4733                 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4734                 if (!ring)
4735                         ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4736                 if (!ring)
4737                         goto err_allocation;
4738                 ring->count = adapter->tx_ring_count;
4739                 ring->queue_index = tx;
4740                 ring->numa_node = nid;
4741                 ring->dev = &adapter->pdev->dev;
4742                 ring->netdev = adapter->netdev;
4743
4744                 adapter->tx_ring[tx] = ring;
4745         }
4746
4747         for (; rx < adapter->num_rx_queues; rx++) {
4748                 struct ixgbe_ring *ring;
4749
4750                 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4751                 if (!ring)
4752                         ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4753                 if (!ring)
4754                         goto err_allocation;
4755                 ring->count = adapter->rx_ring_count;
4756                 ring->queue_index = rx;
4757                 ring->numa_node = nid;
4758                 ring->dev = &adapter->pdev->dev;
4759                 ring->netdev = adapter->netdev;
4760
4761                 adapter->rx_ring[rx] = ring;
4762         }
4763
4764         ixgbe_cache_ring_register(adapter);
4765
4766         return 0;
4767
4768 err_allocation:
4769         while (tx)
4770                 kfree(adapter->tx_ring[--tx]);
4771
4772         while (rx)
4773                 kfree(adapter->rx_ring[--rx]);
4774         return -ENOMEM;
4775 }
4776
4777 /**
4778  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4779  * @adapter: board private structure to initialize
4780  *
4781  * Attempt to configure the interrupts using the best available
4782  * capabilities of the hardware and the kernel.
4783  **/
4784 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4785 {
4786         struct ixgbe_hw *hw = &adapter->hw;
4787         int err = 0;
4788         int vector, v_budget;
4789
4790         /*
4791          * It's easy to be greedy for MSI-X vectors, but it really
4792          * doesn't do us much good if we have a lot more vectors
4793          * than CPU's.  So let's be conservative and only ask for
4794          * (roughly) the same number of vectors as there are CPU's.
4795          */
4796         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4797                        (int)num_online_cpus()) + NON_Q_VECTORS;
4798
4799         /*
4800          * At the same time, hardware can only support a maximum of
4801          * hw.mac->max_msix_vectors vectors.  With features
4802          * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4803          * descriptor queues supported by our device.  Thus, we cap it off in
4804          * those rare cases where the cpu count also exceeds our vector limit.
4805          */
4806         v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4807
4808         /* A failure in MSI-X entry allocation isn't fatal, but it does
4809          * mean we disable MSI-X capabilities of the adapter. */
4810         adapter->msix_entries = kcalloc(v_budget,
4811                                         sizeof(struct msix_entry), GFP_KERNEL);
4812         if (adapter->msix_entries) {
4813                 for (vector = 0; vector < v_budget; vector++)
4814                         adapter->msix_entries[vector].entry = vector;
4815
4816                 ixgbe_acquire_msix_vectors(adapter, v_budget);
4817
4818                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4819                         goto out;
4820         }
4821
4822         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4823         adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4824         if (adapter->flags & (IXGBE_FLAG_FDIR_HASH_CAPABLE |
4825                               IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
4826                 e_err(probe,
4827                       "Flow Director is not supported while multiple "
4828                       "queues are disabled.  Disabling Flow Director\n");
4829         }
4830         adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4831         adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4832         adapter->atr_sample_rate = 0;
4833         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4834                 ixgbe_disable_sriov(adapter);
4835
4836         err = ixgbe_set_num_queues(adapter);
4837         if (err)
4838                 return err;
4839
4840         err = pci_enable_msi(adapter->pdev);
4841         if (!err) {
4842                 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4843         } else {
4844                 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4845                              "Unable to allocate MSI interrupt, "
4846                              "falling back to legacy.  Error: %d\n", err);
4847                 /* reset err */
4848                 err = 0;
4849         }
4850
4851 out:
4852         return err;
4853 }
4854
4855 /**
4856  * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4857  * @adapter: board private structure to initialize
4858  *
4859  * We allocate one q_vector per queue interrupt.  If allocation fails we
4860  * return -ENOMEM.
4861  **/
4862 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4863 {
4864         int q_idx, num_q_vectors;
4865         struct ixgbe_q_vector *q_vector;
4866         int napi_vectors;
4867         int (*poll)(struct napi_struct *, int);
4868
4869         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4870                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4871                 napi_vectors = adapter->num_rx_queues;
4872                 poll = &ixgbe_clean_rxtx_many;
4873         } else {
4874                 num_q_vectors = 1;
4875                 napi_vectors = 1;
4876                 poll = &ixgbe_poll;
4877         }
4878
4879         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4880                 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4881                                         GFP_KERNEL, adapter->node);
4882                 if (!q_vector)
4883                         q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4884                                            GFP_KERNEL);
4885                 if (!q_vector)
4886                         goto err_out;
4887                 q_vector->adapter = adapter;
4888                 if (q_vector->txr_count && !q_vector->rxr_count)
4889                         q_vector->eitr = adapter->tx_eitr_param;
4890                 else
4891                         q_vector->eitr = adapter->rx_eitr_param;
4892                 q_vector->v_idx = q_idx;
4893                 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4894                 adapter->q_vector[q_idx] = q_vector;
4895         }
4896
4897         return 0;
4898
4899 err_out:
4900         while (q_idx) {
4901                 q_idx--;
4902                 q_vector = adapter->q_vector[q_idx];
4903                 netif_napi_del(&q_vector->napi);
4904                 kfree(q_vector);
4905                 adapter->q_vector[q_idx] = NULL;
4906         }
4907         return -ENOMEM;
4908 }
4909
4910 /**
4911  * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4912  * @adapter: board private structure to initialize
4913  *
4914  * This function frees the memory allocated to the q_vectors.  In addition if
4915  * NAPI is enabled it will delete any references to the NAPI struct prior
4916  * to freeing the q_vector.
4917  **/
4918 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4919 {
4920         int q_idx, num_q_vectors;
4921
4922         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4923                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4924         else
4925                 num_q_vectors = 1;
4926
4927         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4928                 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4929                 adapter->q_vector[q_idx] = NULL;
4930                 netif_napi_del(&q_vector->napi);
4931                 kfree(q_vector);
4932         }
4933 }
4934
4935 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4936 {
4937         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4938                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4939                 pci_disable_msix(adapter->pdev);
4940                 kfree(adapter->msix_entries);
4941                 adapter->msix_entries = NULL;
4942         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4943                 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4944                 pci_disable_msi(adapter->pdev);
4945         }
4946 }
4947
4948 /**
4949  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4950  * @adapter: board private structure to initialize
4951  *
4952  * We determine which interrupt scheme to use based on...
4953  * - Kernel support (MSI, MSI-X)
4954  *   - which can be user-defined (via MODULE_PARAM)
4955  * - Hardware queue count (num_*_queues)
4956  *   - defined by miscellaneous hardware support/features (RSS, etc.)
4957  **/
4958 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4959 {
4960         int err;
4961
4962         /* Number of supported queues */
4963         err = ixgbe_set_num_queues(adapter);
4964         if (err)
4965                 return err;
4966
4967         err = ixgbe_set_interrupt_capability(adapter);
4968         if (err) {
4969                 e_dev_err("Unable to setup interrupt capabilities\n");
4970                 goto err_set_interrupt;
4971         }
4972
4973         err = ixgbe_alloc_q_vectors(adapter);
4974         if (err) {
4975                 e_dev_err("Unable to allocate memory for queue vectors\n");
4976                 goto err_alloc_q_vectors;
4977         }
4978
4979         err = ixgbe_alloc_queues(adapter);
4980         if (err) {
4981                 e_dev_err("Unable to allocate memory for queues\n");
4982                 goto err_alloc_queues;
4983         }
4984
4985         e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4986                    (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4987                    adapter->num_rx_queues, adapter->num_tx_queues);
4988
4989         set_bit(__IXGBE_DOWN, &adapter->state);
4990
4991         return 0;
4992
4993 err_alloc_queues:
4994         ixgbe_free_q_vectors(adapter);
4995 err_alloc_q_vectors:
4996         ixgbe_reset_interrupt_capability(adapter);
4997 err_set_interrupt:
4998         return err;
4999 }
5000
5001 static void ring_free_rcu(struct rcu_head *head)
5002 {
5003         kfree(container_of(head, struct ixgbe_ring, rcu));
5004 }
5005
5006 /**
5007  * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
5008  * @adapter: board private structure to clear interrupt scheme on
5009  *
5010  * We go through and clear interrupt specific resources and reset the structure
5011  * to pre-load conditions
5012  **/
5013 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
5014 {
5015         int i;
5016
5017         for (i = 0; i < adapter->num_tx_queues; i++) {
5018                 kfree(adapter->tx_ring[i]);
5019                 adapter->tx_ring[i] = NULL;
5020         }
5021         for (i = 0; i < adapter->num_rx_queues; i++) {
5022                 struct ixgbe_ring *ring = adapter->rx_ring[i];
5023
5024                 /* ixgbe_get_stats64() might access this ring, we must wait
5025                  * a grace period before freeing it.
5026                  */
5027                 call_rcu(&ring->rcu, ring_free_rcu);
5028                 adapter->rx_ring[i] = NULL;
5029         }
5030
5031         adapter->num_tx_queues = 0;
5032         adapter->num_rx_queues = 0;
5033
5034         ixgbe_free_q_vectors(adapter);
5035         ixgbe_reset_interrupt_capability(adapter);
5036 }
5037
5038 /**
5039  * ixgbe_sfp_timer - worker thread to find a missing module
5040  * @data: pointer to our adapter struct
5041  **/
5042 static void ixgbe_sfp_timer(unsigned long data)
5043 {
5044         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5045
5046         /*
5047          * Do the sfp_timer outside of interrupt context due to the
5048          * delays that sfp+ detection requires
5049          */
5050         schedule_work(&adapter->sfp_task);
5051 }
5052
5053 /**
5054  * ixgbe_sfp_task - worker thread to find a missing module
5055  * @work: pointer to work_struct containing our data
5056  **/
5057 static void ixgbe_sfp_task(struct work_struct *work)
5058 {
5059         struct ixgbe_adapter *adapter = container_of(work,
5060                                                      struct ixgbe_adapter,
5061                                                      sfp_task);
5062         struct ixgbe_hw *hw = &adapter->hw;
5063
5064         if ((hw->phy.type == ixgbe_phy_nl) &&
5065             (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
5066                 s32 ret = hw->phy.ops.identify_sfp(hw);
5067                 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
5068                         goto reschedule;
5069                 ret = hw->phy.ops.reset(hw);
5070                 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5071                         e_dev_err("failed to initialize because an unsupported "
5072                                   "SFP+ module type was detected.\n");
5073                         e_dev_err("Reload the driver after installing a "
5074                                   "supported module.\n");
5075                         unregister_netdev(adapter->netdev);
5076                 } else {
5077                         e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5078                 }
5079                 /* don't need this routine any more */
5080                 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5081         }
5082         return;
5083 reschedule:
5084         if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
5085                 mod_timer(&adapter->sfp_timer,
5086                           round_jiffies(jiffies + (2 * HZ)));
5087 }
5088
5089 /**
5090  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5091  * @adapter: board private structure to initialize
5092  *
5093  * ixgbe_sw_init initializes the Adapter private data structure.
5094  * Fields are initialized based on PCI device information and
5095  * OS network device settings (MTU size).
5096  **/
5097 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
5098 {
5099         struct ixgbe_hw *hw = &adapter->hw;
5100         struct pci_dev *pdev = adapter->pdev;
5101         struct net_device *dev = adapter->netdev;
5102         unsigned int rss;
5103 #ifdef CONFIG_IXGBE_DCB
5104         int j;
5105         struct tc_configuration *tc;
5106 #endif
5107         int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5108
5109         /* PCI config space info */
5110
5111         hw->vendor_id = pdev->vendor;
5112         hw->device_id = pdev->device;
5113         hw->revision_id = pdev->revision;
5114         hw->subsystem_vendor_id = pdev->subsystem_vendor;
5115         hw->subsystem_device_id = pdev->subsystem_device;
5116
5117         /* Set capability flags */
5118         rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
5119         adapter->ring_feature[RING_F_RSS].indices = rss;
5120         adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
5121         adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
5122         switch (hw->mac.type) {
5123         case ixgbe_mac_82598EB:
5124                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5125                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5126                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
5127                 break;
5128         case ixgbe_mac_82599EB:
5129         case ixgbe_mac_X540:
5130                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
5131                 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5132                 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5133                 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5134                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5135                 /* n-tuple support exists, always init our spinlock */
5136                 spin_lock_init(&adapter->fdir_perfect_lock);
5137                 /* Flow Director hash filters enabled */
5138                 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
5139                 adapter->atr_sample_rate = 20;
5140                 adapter->ring_feature[RING_F_FDIR].indices =
5141                                                          IXGBE_MAX_FDIR_INDICES;
5142                 adapter->fdir_pballoc = 0;
5143 #ifdef IXGBE_FCOE
5144                 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5145                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5146                 adapter->ring_feature[RING_F_FCOE].indices = 0;
5147 #ifdef CONFIG_IXGBE_DCB
5148                 /* Default traffic class to use for FCoE */
5149                 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
5150                 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5151 #endif
5152 #endif /* IXGBE_FCOE */
5153                 break;
5154         default:
5155                 break;
5156         }
5157
5158 #ifdef CONFIG_IXGBE_DCB
5159         /* Configure DCB traffic classes */
5160         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5161                 tc = &adapter->dcb_cfg.tc_config[j];
5162                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5163                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5164                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5165                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5166                 tc->dcb_pfc = pfc_disabled;
5167         }
5168         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5169         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5170         adapter->dcb_cfg.rx_pba_cfg = pba_equal;
5171         adapter->dcb_cfg.pfc_mode_enable = false;
5172         adapter->dcb_cfg.round_robin_enable = false;
5173         adapter->dcb_set_bitmap = 0x00;
5174         ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5175                            adapter->ring_feature[RING_F_DCB].indices);
5176
5177 #endif
5178
5179         /* default flow control settings */
5180         hw->fc.requested_mode = ixgbe_fc_full;
5181         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
5182 #ifdef CONFIG_DCB
5183         adapter->last_lfc_mode = hw->fc.current_mode;
5184 #endif
5185         hw->fc.high_water = FC_HIGH_WATER(max_frame);
5186         hw->fc.low_water = FC_LOW_WATER(max_frame);
5187         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5188         hw->fc.send_xon = true;
5189         hw->fc.disable_fc_autoneg = false;
5190
5191         /* enable itr by default in dynamic mode */
5192         adapter->rx_itr_setting = 1;
5193         adapter->rx_eitr_param = 20000;
5194         adapter->tx_itr_setting = 1;
5195         adapter->tx_eitr_param = 10000;
5196
5197         /* set defaults for eitr in MegaBytes */
5198         adapter->eitr_low = 10;
5199         adapter->eitr_high = 20;
5200
5201         /* set default ring sizes */
5202         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5203         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5204
5205         /* initialize eeprom parameters */
5206         if (ixgbe_init_eeprom_params_generic(hw)) {
5207                 e_dev_err("EEPROM initialization failed\n");
5208                 return -EIO;
5209         }
5210
5211         /* enable rx csum by default */
5212         adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
5213
5214         /* get assigned NUMA node */
5215         adapter->node = dev_to_node(&pdev->dev);
5216
5217         set_bit(__IXGBE_DOWN, &adapter->state);
5218
5219         return 0;
5220 }
5221
5222 /**
5223  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5224  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
5225  *
5226  * Return 0 on success, negative on failure
5227  **/
5228 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5229 {
5230         struct device *dev = tx_ring->dev;
5231         int size;
5232
5233         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5234         tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
5235         if (!tx_ring->tx_buffer_info)
5236                 tx_ring->tx_buffer_info = vzalloc(size);
5237         if (!tx_ring->tx_buffer_info)
5238                 goto err;
5239
5240         /* round up to nearest 4K */
5241         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5242         tx_ring->size = ALIGN(tx_ring->size, 4096);
5243
5244         tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5245                                            &tx_ring->dma, GFP_KERNEL);
5246         if (!tx_ring->desc)
5247                 goto err;
5248
5249         tx_ring->next_to_use = 0;
5250         tx_ring->next_to_clean = 0;
5251         tx_ring->work_limit = tx_ring->count;
5252         return 0;
5253
5254 err:
5255         vfree(tx_ring->tx_buffer_info);
5256         tx_ring->tx_buffer_info = NULL;
5257         dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5258         return -ENOMEM;
5259 }
5260
5261 /**
5262  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5263  * @adapter: board private structure
5264  *
5265  * If this function returns with an error, then it's possible one or
5266  * more of the rings is populated (while the rest are not).  It is the
5267  * callers duty to clean those orphaned rings.
5268  *
5269  * Return 0 on success, negative on failure
5270  **/
5271 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5272 {
5273         int i, err = 0;
5274
5275         for (i = 0; i < adapter->num_tx_queues; i++) {
5276                 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5277                 if (!err)
5278                         continue;
5279                 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5280                 break;
5281         }
5282
5283         return err;
5284 }
5285
5286 /**
5287  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5288  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
5289  *
5290  * Returns 0 on success, negative on failure
5291  **/
5292 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5293 {
5294         struct device *dev = rx_ring->dev;
5295         int size;
5296
5297         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5298         rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
5299         if (!rx_ring->rx_buffer_info)
5300                 rx_ring->rx_buffer_info = vzalloc(size);
5301         if (!rx_ring->rx_buffer_info)
5302                 goto err;
5303
5304         /* Round up to nearest 4K */
5305         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5306         rx_ring->size = ALIGN(rx_ring->size, 4096);
5307
5308         rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5309                                            &rx_ring->dma, GFP_KERNEL);
5310
5311         if (!rx_ring->desc)
5312                 goto err;
5313
5314         rx_ring->next_to_clean = 0;
5315         rx_ring->next_to_use = 0;
5316
5317         return 0;
5318 err:
5319         vfree(rx_ring->rx_buffer_info);
5320         rx_ring->rx_buffer_info = NULL;
5321         dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5322         return -ENOMEM;
5323 }
5324
5325 /**
5326  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5327  * @adapter: board private structure
5328  *
5329  * If this function returns with an error, then it's possible one or
5330  * more of the rings is populated (while the rest are not).  It is the
5331  * callers duty to clean those orphaned rings.
5332  *
5333  * Return 0 on success, negative on failure
5334  **/
5335 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5336 {
5337         int i, err = 0;
5338
5339         for (i = 0; i < adapter->num_rx_queues; i++) {
5340                 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5341                 if (!err)
5342                         continue;
5343                 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5344                 break;
5345         }
5346
5347         return err;
5348 }
5349
5350 /**
5351  * ixgbe_free_tx_resources - Free Tx Resources per Queue
5352  * @tx_ring: Tx descriptor ring for a specific queue
5353  *
5354  * Free all transmit software resources
5355  **/
5356 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5357 {
5358         ixgbe_clean_tx_ring(tx_ring);
5359
5360         vfree(tx_ring->tx_buffer_info);
5361         tx_ring->tx_buffer_info = NULL;
5362
5363         /* if not set, then don't free */
5364         if (!tx_ring->desc)
5365                 return;
5366
5367         dma_free_coherent(tx_ring->dev, tx_ring->size,
5368                           tx_ring->desc, tx_ring->dma);
5369
5370         tx_ring->desc = NULL;
5371 }
5372
5373 /**
5374  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5375  * @adapter: board private structure
5376  *
5377  * Free all transmit software resources
5378  **/
5379 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5380 {
5381         int i;
5382
5383         for (i = 0; i < adapter->num_tx_queues; i++)
5384                 if (adapter->tx_ring[i]->desc)
5385                         ixgbe_free_tx_resources(adapter->tx_ring[i]);
5386 }
5387
5388 /**
5389  * ixgbe_free_rx_resources - Free Rx Resources
5390  * @rx_ring: ring to clean the resources from
5391  *
5392  * Free all receive software resources
5393  **/
5394 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5395 {
5396         ixgbe_clean_rx_ring(rx_ring);
5397
5398         vfree(rx_ring->rx_buffer_info);
5399         rx_ring->rx_buffer_info = NULL;
5400
5401         /* if not set, then don't free */
5402         if (!rx_ring->desc)
5403                 return;
5404
5405         dma_free_coherent(rx_ring->dev, rx_ring->size,
5406                           rx_ring->desc, rx_ring->dma);
5407
5408         rx_ring->desc = NULL;
5409 }
5410
5411 /**
5412  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5413  * @adapter: board private structure
5414  *
5415  * Free all receive software resources
5416  **/
5417 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5418 {
5419         int i;
5420
5421         for (i = 0; i < adapter->num_rx_queues; i++)
5422                 if (adapter->rx_ring[i]->desc)
5423                         ixgbe_free_rx_resources(adapter->rx_ring[i]);
5424 }
5425
5426 /**
5427  * ixgbe_change_mtu - Change the Maximum Transfer Unit
5428  * @netdev: network interface device structure
5429  * @new_mtu: new value for maximum frame size
5430  *
5431  * Returns 0 on success, negative on failure
5432  **/
5433 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5434 {
5435         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5436         struct ixgbe_hw *hw = &adapter->hw;
5437         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5438
5439         /* MTU < 68 is an error and causes problems on some kernels */
5440         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5441                 return -EINVAL;
5442
5443         e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5444         /* must set new MTU before calling down or up */
5445         netdev->mtu = new_mtu;
5446
5447         hw->fc.high_water = FC_HIGH_WATER(max_frame);
5448         hw->fc.low_water = FC_LOW_WATER(max_frame);
5449
5450         if (netif_running(netdev))
5451                 ixgbe_reinit_locked(adapter);
5452
5453         return 0;
5454 }
5455
5456 /**
5457  * ixgbe_open - Called when a network interface is made active
5458  * @netdev: network interface device structure
5459  *
5460  * Returns 0 on success, negative value on failure
5461  *
5462  * The open entry point is called when a network interface is made
5463  * active by the system (IFF_UP).  At this point all resources needed
5464  * for transmit and receive operations are allocated, the interrupt
5465  * handler is registered with the OS, the watchdog timer is started,
5466  * and the stack is notified that the interface is ready.
5467  **/
5468 static int ixgbe_open(struct net_device *netdev)
5469 {
5470         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5471         int err;
5472
5473         /* disallow open during test */
5474         if (test_bit(__IXGBE_TESTING, &adapter->state))
5475                 return -EBUSY;
5476
5477         netif_carrier_off(netdev);
5478
5479         /* allocate transmit descriptors */
5480         err = ixgbe_setup_all_tx_resources(adapter);
5481         if (err)
5482                 goto err_setup_tx;
5483
5484         /* allocate receive descriptors */
5485         err = ixgbe_setup_all_rx_resources(adapter);
5486         if (err)
5487                 goto err_setup_rx;
5488
5489         ixgbe_configure(adapter);
5490
5491         err = ixgbe_request_irq(adapter);
5492         if (err)
5493                 goto err_req_irq;
5494
5495         err = ixgbe_up_complete(adapter);
5496         if (err)
5497                 goto err_up;
5498
5499         netif_tx_start_all_queues(netdev);
5500
5501         return 0;
5502
5503 err_up:
5504         ixgbe_release_hw_control(adapter);
5505         ixgbe_free_irq(adapter);
5506 err_req_irq:
5507 err_setup_rx:
5508         ixgbe_free_all_rx_resources(adapter);
5509 err_setup_tx:
5510         ixgbe_free_all_tx_resources(adapter);
5511         ixgbe_reset(adapter);
5512
5513         return err;
5514 }
5515
5516 /**
5517  * ixgbe_close - Disables a network interface
5518  * @netdev: network interface device structure
5519  *
5520  * Returns 0, this is not allowed to fail
5521  *
5522  * The close entry point is called when an interface is de-activated
5523  * by the OS.  The hardware is still under the drivers control, but
5524  * needs to be disabled.  A global MAC reset is issued to stop the
5525  * hardware, and all transmit and receive resources are freed.
5526  **/
5527 static int ixgbe_close(struct net_device *netdev)
5528 {
5529         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5530
5531         ixgbe_down(adapter);
5532         ixgbe_free_irq(adapter);
5533
5534         ixgbe_free_all_tx_resources(adapter);
5535         ixgbe_free_all_rx_resources(adapter);
5536
5537         ixgbe_release_hw_control(adapter);
5538
5539         return 0;
5540 }
5541
5542 #ifdef CONFIG_PM
5543 static int ixgbe_resume(struct pci_dev *pdev)
5544 {
5545         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5546         struct net_device *netdev = adapter->netdev;
5547         u32 err;
5548
5549         pci_set_power_state(pdev, PCI_D0);
5550         pci_restore_state(pdev);
5551         /*
5552          * pci_restore_state clears dev->state_saved so call
5553          * pci_save_state to restore it.
5554          */
5555         pci_save_state(pdev);
5556
5557         err = pci_enable_device_mem(pdev);
5558         if (err) {
5559                 e_dev_err("Cannot enable PCI device from suspend\n");
5560                 return err;
5561         }
5562         pci_set_master(pdev);
5563
5564         pci_wake_from_d3(pdev, false);
5565
5566         err = ixgbe_init_interrupt_scheme(adapter);
5567         if (err) {
5568                 e_dev_err("Cannot initialize interrupts for device\n");
5569                 return err;
5570         }
5571
5572         ixgbe_reset(adapter);
5573
5574         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5575
5576         if (netif_running(netdev)) {
5577                 err = ixgbe_open(netdev);
5578                 if (err)
5579                         return err;
5580         }
5581
5582         netif_device_attach(netdev);
5583
5584         return 0;
5585 }
5586 #endif /* CONFIG_PM */
5587
5588 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5589 {
5590         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5591         struct net_device *netdev = adapter->netdev;
5592         struct ixgbe_hw *hw = &adapter->hw;
5593         u32 ctrl, fctrl;
5594         u32 wufc = adapter->wol;
5595 #ifdef CONFIG_PM
5596         int retval = 0;
5597 #endif
5598
5599         netif_device_detach(netdev);
5600
5601         if (netif_running(netdev)) {
5602                 ixgbe_down(adapter);
5603                 ixgbe_free_irq(adapter);
5604                 ixgbe_free_all_tx_resources(adapter);
5605                 ixgbe_free_all_rx_resources(adapter);
5606         }
5607
5608         ixgbe_clear_interrupt_scheme(adapter);
5609
5610 #ifdef CONFIG_PM
5611         retval = pci_save_state(pdev);
5612         if (retval)
5613                 return retval;
5614
5615 #endif
5616         if (wufc) {
5617                 ixgbe_set_rx_mode(netdev);
5618
5619                 /* turn on all-multi mode if wake on multicast is enabled */
5620                 if (wufc & IXGBE_WUFC_MC) {
5621                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5622                         fctrl |= IXGBE_FCTRL_MPE;
5623                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5624                 }
5625
5626                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5627                 ctrl |= IXGBE_CTRL_GIO_DIS;
5628                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5629
5630                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5631         } else {
5632                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5633                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5634         }
5635
5636         switch (hw->mac.type) {
5637         case ixgbe_mac_82598EB:
5638                 pci_wake_from_d3(pdev, false);
5639                 break;
5640         case ixgbe_mac_82599EB:
5641         case ixgbe_mac_X540:
5642                 pci_wake_from_d3(pdev, !!wufc);
5643                 break;
5644         default:
5645                 break;
5646         }
5647
5648         *enable_wake = !!wufc;
5649
5650         ixgbe_release_hw_control(adapter);
5651
5652         pci_disable_device(pdev);
5653
5654         return 0;
5655 }
5656
5657 #ifdef CONFIG_PM
5658 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5659 {
5660         int retval;
5661         bool wake;
5662
5663         retval = __ixgbe_shutdown(pdev, &wake);
5664         if (retval)
5665                 return retval;
5666
5667         if (wake) {
5668                 pci_prepare_to_sleep(pdev);
5669         } else {
5670                 pci_wake_from_d3(pdev, false);
5671                 pci_set_power_state(pdev, PCI_D3hot);
5672         }
5673
5674         return 0;
5675 }
5676 #endif /* CONFIG_PM */
5677
5678 static void ixgbe_shutdown(struct pci_dev *pdev)
5679 {
5680         bool wake;
5681
5682         __ixgbe_shutdown(pdev, &wake);
5683
5684         if (system_state == SYSTEM_POWER_OFF) {
5685                 pci_wake_from_d3(pdev, wake);
5686                 pci_set_power_state(pdev, PCI_D3hot);
5687         }
5688 }
5689
5690 /**
5691  * ixgbe_update_stats - Update the board statistics counters.
5692  * @adapter: board private structure
5693  **/
5694 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5695 {
5696         struct net_device *netdev = adapter->netdev;
5697         struct ixgbe_hw *hw = &adapter->hw;
5698         struct ixgbe_hw_stats *hwstats = &adapter->stats;
5699         u64 total_mpc = 0;
5700         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5701         u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5702         u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5703         u64 bytes = 0, packets = 0;
5704
5705         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5706             test_bit(__IXGBE_RESETTING, &adapter->state))
5707                 return;
5708
5709         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5710                 u64 rsc_count = 0;
5711                 u64 rsc_flush = 0;
5712                 for (i = 0; i < 16; i++)
5713                         adapter->hw_rx_no_dma_resources +=
5714                                 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5715                 for (i = 0; i < adapter->num_rx_queues; i++) {
5716                         rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5717                         rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5718                 }
5719                 adapter->rsc_total_count = rsc_count;
5720                 adapter->rsc_total_flush = rsc_flush;
5721         }
5722
5723         for (i = 0; i < adapter->num_rx_queues; i++) {
5724                 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5725                 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5726                 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5727                 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5728                 bytes += rx_ring->stats.bytes;
5729                 packets += rx_ring->stats.packets;
5730         }
5731         adapter->non_eop_descs = non_eop_descs;
5732         adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5733         adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5734         netdev->stats.rx_bytes = bytes;
5735         netdev->stats.rx_packets = packets;
5736
5737         bytes = 0;
5738         packets = 0;
5739         /* gather some stats to the adapter struct that are per queue */
5740         for (i = 0; i < adapter->num_tx_queues; i++) {
5741                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5742                 restart_queue += tx_ring->tx_stats.restart_queue;
5743                 tx_busy += tx_ring->tx_stats.tx_busy;
5744                 bytes += tx_ring->stats.bytes;
5745                 packets += tx_ring->stats.packets;
5746         }
5747         adapter->restart_queue = restart_queue;
5748         adapter->tx_busy = tx_busy;
5749         netdev->stats.tx_bytes = bytes;
5750         netdev->stats.tx_packets = packets;
5751
5752         hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5753         for (i = 0; i < 8; i++) {
5754                 /* for packet buffers not used, the register should read 0 */
5755                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5756                 missed_rx += mpc;
5757                 hwstats->mpc[i] += mpc;
5758                 total_mpc += hwstats->mpc[i];
5759                 if (hw->mac.type == ixgbe_mac_82598EB)
5760                         hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5761                 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5762                 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5763                 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5764                 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5765                 switch (hw->mac.type) {
5766                 case ixgbe_mac_82598EB:
5767                         hwstats->pxonrxc[i] +=
5768                                 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5769                         break;
5770                 case ixgbe_mac_82599EB:
5771                 case ixgbe_mac_X540:
5772                         hwstats->pxonrxc[i] +=
5773                                 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5774                         break;
5775                 default:
5776                         break;
5777                 }
5778                 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5779                 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5780         }
5781         hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5782         /* work around hardware counting issue */
5783         hwstats->gprc -= missed_rx;
5784
5785         ixgbe_update_xoff_received(adapter);
5786
5787         /* 82598 hardware only has a 32 bit counter in the high register */
5788         switch (hw->mac.type) {
5789         case ixgbe_mac_82598EB:
5790                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5791                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5792                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5793                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5794                 break;
5795         case ixgbe_mac_82599EB:
5796         case ixgbe_mac_X540:
5797                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5798                 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5799                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5800                 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5801                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5802                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5803                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5804                 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5805                 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5806 #ifdef IXGBE_FCOE
5807                 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5808                 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5809                 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5810                 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5811                 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5812                 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5813 #endif /* IXGBE_FCOE */
5814                 break;
5815         default:
5816                 break;
5817         }
5818         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5819         hwstats->bprc += bprc;
5820         hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5821         if (hw->mac.type == ixgbe_mac_82598EB)
5822                 hwstats->mprc -= bprc;
5823         hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5824         hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5825         hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5826         hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5827         hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5828         hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5829         hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5830         hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5831         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5832         hwstats->lxontxc += lxon;
5833         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5834         hwstats->lxofftxc += lxoff;
5835         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5836         hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5837         hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5838         /*
5839          * 82598 errata - tx of flow control packets is included in tx counters
5840          */
5841         xon_off_tot = lxon + lxoff;
5842         hwstats->gptc -= xon_off_tot;
5843         hwstats->mptc -= xon_off_tot;
5844         hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5845         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5846         hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5847         hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5848         hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5849         hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5850         hwstats->ptc64 -= xon_off_tot;
5851         hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5852         hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5853         hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5854         hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5855         hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5856         hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5857
5858         /* Fill out the OS statistics structure */
5859         netdev->stats.multicast = hwstats->mprc;
5860
5861         /* Rx Errors */
5862         netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5863         netdev->stats.rx_dropped = 0;
5864         netdev->stats.rx_length_errors = hwstats->rlec;
5865         netdev->stats.rx_crc_errors = hwstats->crcerrs;
5866         netdev->stats.rx_missed_errors = total_mpc;
5867 }
5868
5869 /**
5870  * ixgbe_watchdog - Timer Call-back
5871  * @data: pointer to adapter cast into an unsigned long
5872  **/
5873 static void ixgbe_watchdog(unsigned long data)
5874 {
5875         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5876         struct ixgbe_hw *hw = &adapter->hw;
5877         u64 eics = 0;
5878         int i;
5879
5880         /*
5881          *  Do the watchdog outside of interrupt context due to the lovely
5882          * delays that some of the newer hardware requires
5883          */
5884
5885         if (test_bit(__IXGBE_DOWN, &adapter->state))
5886                 goto watchdog_short_circuit;
5887
5888         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5889                 /*
5890                  * for legacy and MSI interrupts don't set any bits
5891                  * that are enabled for EIAM, because this operation
5892                  * would set *both* EIMS and EICS for any bit in EIAM
5893                  */
5894                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5895                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5896                 goto watchdog_reschedule;
5897         }
5898
5899         /* get one bit for every active tx/rx interrupt vector */
5900         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5901                 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5902                 if (qv->rxr_count || qv->txr_count)
5903                         eics |= ((u64)1 << i);
5904         }
5905
5906         /* Cause software interrupt to ensure rx rings are cleaned */
5907         ixgbe_irq_rearm_queues(adapter, eics);
5908
5909 watchdog_reschedule:
5910         /* Reset the timer */
5911         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5912
5913 watchdog_short_circuit:
5914         schedule_work(&adapter->watchdog_task);
5915 }
5916
5917 /**
5918  * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5919  * @work: pointer to work_struct containing our data
5920  **/
5921 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5922 {
5923         struct ixgbe_adapter *adapter = container_of(work,
5924                                                      struct ixgbe_adapter,
5925                                                      multispeed_fiber_task);
5926         struct ixgbe_hw *hw = &adapter->hw;
5927         u32 autoneg;
5928         bool negotiation;
5929
5930         adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5931         autoneg = hw->phy.autoneg_advertised;
5932         if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5933                 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5934         hw->mac.autotry_restart = false;
5935         if (hw->mac.ops.setup_link)
5936                 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5937         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5938         adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5939 }
5940
5941 /**
5942  * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5943  * @work: pointer to work_struct containing our data
5944  **/
5945 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5946 {
5947         struct ixgbe_adapter *adapter = container_of(work,
5948                                                      struct ixgbe_adapter,
5949                                                      sfp_config_module_task);
5950         struct ixgbe_hw *hw = &adapter->hw;
5951         u32 err;
5952
5953         adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5954
5955         /* Time for electrical oscillations to settle down */
5956         msleep(100);
5957         err = hw->phy.ops.identify_sfp(hw);
5958
5959         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5960                 e_dev_err("failed to initialize because an unsupported SFP+ "
5961                           "module type was detected.\n");
5962                 e_dev_err("Reload the driver after installing a supported "
5963                           "module.\n");
5964                 unregister_netdev(adapter->netdev);
5965                 return;
5966         }
5967         hw->mac.ops.setup_sfp(hw);
5968
5969         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5970                 /* This will also work for DA Twinax connections */
5971                 schedule_work(&adapter->multispeed_fiber_task);
5972         adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5973 }
5974
5975 /**
5976  * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5977  * @work: pointer to work_struct containing our data
5978  **/
5979 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5980 {
5981         struct ixgbe_adapter *adapter = container_of(work,
5982                                                      struct ixgbe_adapter,
5983                                                      fdir_reinit_task);
5984         struct ixgbe_hw *hw = &adapter->hw;
5985         int i;
5986
5987         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5988                 for (i = 0; i < adapter->num_tx_queues; i++)
5989                         set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5990                                 &(adapter->tx_ring[i]->state));
5991         } else {
5992                 e_err(probe, "failed to finish FDIR re-initialization, "
5993                       "ignored adding FDIR ATR filters\n");
5994         }
5995         /* Done FDIR Re-initialization, enable transmits */
5996         netif_tx_start_all_queues(adapter->netdev);
5997 }
5998
5999 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6000 {
6001         u32 ssvpc;
6002
6003         /* Do not perform spoof check for 82598 */
6004         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6005                 return;
6006
6007         ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6008
6009         /*
6010          * ssvpc register is cleared on read, if zero then no
6011          * spoofed packets in the last interval.
6012          */
6013         if (!ssvpc)
6014                 return;
6015
6016         e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
6017 }
6018
6019 static DEFINE_MUTEX(ixgbe_watchdog_lock);
6020
6021 /**
6022  * ixgbe_watchdog_task - worker thread to bring link up
6023  * @work: pointer to work_struct containing our data
6024  **/
6025 static void ixgbe_watchdog_task(struct work_struct *work)
6026 {
6027         struct ixgbe_adapter *adapter = container_of(work,
6028                                                      struct ixgbe_adapter,
6029                                                      watchdog_task);
6030         struct net_device *netdev = adapter->netdev;
6031         struct ixgbe_hw *hw = &adapter->hw;
6032         u32 link_speed;
6033         bool link_up;
6034         int i;
6035         struct ixgbe_ring *tx_ring;
6036         int some_tx_pending = 0;
6037
6038         mutex_lock(&ixgbe_watchdog_lock);
6039
6040         link_up = adapter->link_up;
6041         link_speed = adapter->link_speed;
6042
6043         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
6044                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6045                 if (link_up) {
6046 #ifdef CONFIG_DCB
6047                         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6048                                 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
6049                                         hw->mac.ops.fc_enable(hw, i);
6050                         } else {
6051                                 hw->mac.ops.fc_enable(hw, 0);
6052                         }
6053 #else
6054                         hw->mac.ops.fc_enable(hw, 0);
6055 #endif
6056                 }
6057
6058                 if (link_up ||
6059                     time_after(jiffies, (adapter->link_check_timeout +
6060                                          IXGBE_TRY_LINK_TIMEOUT))) {
6061                         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6062                         IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6063                 }
6064                 adapter->link_up = link_up;
6065                 adapter->link_speed = link_speed;
6066         }
6067
6068         if (link_up) {
6069                 if (!netif_carrier_ok(netdev)) {
6070                         bool flow_rx, flow_tx;
6071
6072                         switch (hw->mac.type) {
6073                         case ixgbe_mac_82598EB: {
6074                                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6075                                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6076                                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6077                                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6078                         }
6079                                 break;
6080                         case ixgbe_mac_82599EB:
6081                         case ixgbe_mac_X540: {
6082                                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6083                                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6084                                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6085                                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6086                         }
6087                                 break;
6088                         default:
6089                                 flow_tx = false;
6090                                 flow_rx = false;
6091                                 break;
6092                         }
6093
6094                         e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
6095                                (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
6096                                "10 Gbps" :
6097                                (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
6098                                "1 Gbps" : "unknown speed")),
6099                                ((flow_rx && flow_tx) ? "RX/TX" :
6100                                (flow_rx ? "RX" :
6101                                (flow_tx ? "TX" : "None"))));
6102
6103                         netif_carrier_on(netdev);
6104                 } else {
6105                         /* Force detection of hung controller */
6106                         for (i = 0; i < adapter->num_tx_queues; i++) {
6107                                 tx_ring = adapter->tx_ring[i];
6108                                 set_check_for_tx_hang(tx_ring);
6109                         }
6110                 }
6111         } else {
6112                 adapter->link_up = false;
6113                 adapter->link_speed = 0;
6114                 if (netif_carrier_ok(netdev)) {
6115                         e_info(drv, "NIC Link is Down\n");
6116                         netif_carrier_off(netdev);
6117                 }
6118         }
6119
6120         if (!netif_carrier_ok(netdev)) {
6121                 for (i = 0; i < adapter->num_tx_queues; i++) {
6122                         tx_ring = adapter->tx_ring[i];
6123                         if (tx_ring->next_to_use != tx_ring->next_to_clean) {
6124                                 some_tx_pending = 1;
6125                                 break;
6126                         }
6127                 }
6128
6129                 if (some_tx_pending) {
6130                         /* We've lost link, so the controller stops DMA,
6131                          * but we've got queued Tx work that's never going
6132                          * to get done, so reset controller to flush Tx.
6133                          * (Do the reset outside of interrupt context).
6134                          */
6135                          schedule_work(&adapter->reset_task);
6136                 }
6137         }
6138
6139         ixgbe_spoof_check(adapter);
6140         ixgbe_update_stats(adapter);
6141         mutex_unlock(&ixgbe_watchdog_lock);
6142 }
6143
6144 static int ixgbe_tso(struct ixgbe_adapter *adapter,
6145                      struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6146                      u32 tx_flags, u8 *hdr_len, __be16 protocol)
6147 {
6148         struct ixgbe_adv_tx_context_desc *context_desc;
6149         unsigned int i;
6150         int err;
6151         struct ixgbe_tx_buffer *tx_buffer_info;
6152         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
6153         u32 mss_l4len_idx, l4len;
6154
6155         if (skb_is_gso(skb)) {
6156                 if (skb_header_cloned(skb)) {
6157                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6158                         if (err)
6159                                 return err;
6160                 }
6161                 l4len = tcp_hdrlen(skb);
6162                 *hdr_len += l4len;
6163
6164                 if (protocol == htons(ETH_P_IP)) {
6165                         struct iphdr *iph = ip_hdr(skb);
6166                         iph->tot_len = 0;
6167                         iph->check = 0;
6168                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6169                                                                  iph->daddr, 0,
6170                                                                  IPPROTO_TCP,
6171                                                                  0);
6172                 } else if (skb_is_gso_v6(skb)) {
6173                         ipv6_hdr(skb)->payload_len = 0;
6174                         tcp_hdr(skb)->check =
6175                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6176                                              &ipv6_hdr(skb)->daddr,
6177                                              0, IPPROTO_TCP, 0);
6178                 }
6179
6180                 i = tx_ring->next_to_use;
6181
6182                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6183                 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6184
6185                 /* VLAN MACLEN IPLEN */
6186                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6187                         vlan_macip_lens |=
6188                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6189                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
6190                                     IXGBE_ADVTXD_MACLEN_SHIFT);
6191                 *hdr_len += skb_network_offset(skb);
6192                 vlan_macip_lens |=
6193                     (skb_transport_header(skb) - skb_network_header(skb));
6194                 *hdr_len +=
6195                     (skb_transport_header(skb) - skb_network_header(skb));
6196                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6197                 context_desc->seqnum_seed = 0;
6198
6199                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6200                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
6201                                    IXGBE_ADVTXD_DTYP_CTXT);
6202
6203                 if (protocol == htons(ETH_P_IP))
6204                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
6205                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6206                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6207
6208                 /* MSS L4LEN IDX */
6209                 mss_l4len_idx =
6210                     (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
6211                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
6212                 /* use index 1 for TSO */
6213                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6214                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6215
6216                 tx_buffer_info->time_stamp = jiffies;
6217                 tx_buffer_info->next_to_watch = i;
6218
6219                 i++;
6220                 if (i == tx_ring->count)
6221                         i = 0;
6222                 tx_ring->next_to_use = i;
6223
6224                 return true;
6225         }
6226         return false;
6227 }
6228
6229 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6230                       __be16 protocol)
6231 {
6232         u32 rtn = 0;
6233
6234         switch (protocol) {
6235         case cpu_to_be16(ETH_P_IP):
6236                 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
6237                 switch (ip_hdr(skb)->protocol) {
6238                 case IPPROTO_TCP:
6239                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6240                         break;
6241                 case IPPROTO_SCTP:
6242                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6243                         break;
6244                 }
6245                 break;
6246         case cpu_to_be16(ETH_P_IPV6):
6247                 /* XXX what about other V6 headers?? */
6248                 switch (ipv6_hdr(skb)->nexthdr) {
6249                 case IPPROTO_TCP:
6250                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6251                         break;
6252                 case IPPROTO_SCTP:
6253                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6254                         break;
6255                 }
6256                 break;
6257         default:
6258                 if (unlikely(net_ratelimit()))
6259                         e_warn(probe, "partial checksum but proto=%x!\n",
6260                                protocol);
6261                 break;
6262         }
6263
6264         return rtn;
6265 }
6266
6267 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
6268                           struct ixgbe_ring *tx_ring,
6269                           struct sk_buff *skb, u32 tx_flags,
6270                           __be16 protocol)
6271 {
6272         struct ixgbe_adv_tx_context_desc *context_desc;
6273         unsigned int i;
6274         struct ixgbe_tx_buffer *tx_buffer_info;
6275         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
6276
6277         if (skb->ip_summed == CHECKSUM_PARTIAL ||
6278             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
6279                 i = tx_ring->next_to_use;
6280                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6281                 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6282
6283                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6284                         vlan_macip_lens |=
6285                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6286                 vlan_macip_lens |= (skb_network_offset(skb) <<
6287                                     IXGBE_ADVTXD_MACLEN_SHIFT);
6288                 if (skb->ip_summed == CHECKSUM_PARTIAL)
6289                         vlan_macip_lens |= (skb_transport_header(skb) -
6290                                             skb_network_header(skb));
6291
6292                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6293                 context_desc->seqnum_seed = 0;
6294
6295                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
6296                                     IXGBE_ADVTXD_DTYP_CTXT);
6297
6298                 if (skb->ip_summed == CHECKSUM_PARTIAL)
6299                         type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
6300
6301                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6302                 /* use index zero for tx checksum offload */
6303                 context_desc->mss_l4len_idx = 0;
6304
6305                 tx_buffer_info->time_stamp = jiffies;
6306                 tx_buffer_info->next_to_watch = i;
6307
6308                 i++;
6309                 if (i == tx_ring->count)
6310                         i = 0;
6311                 tx_ring->next_to_use = i;
6312
6313                 return true;
6314         }
6315
6316         return false;
6317 }
6318
6319 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6320                         struct ixgbe_ring *tx_ring,
6321                         struct sk_buff *skb, u32 tx_flags,
6322                         unsigned int first, const u8 hdr_len)
6323 {
6324         struct device *dev = tx_ring->dev;
6325         struct ixgbe_tx_buffer *tx_buffer_info;
6326         unsigned int len;
6327         unsigned int total = skb->len;
6328         unsigned int offset = 0, size, count = 0, i;
6329         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6330         unsigned int f;
6331         unsigned int bytecount = skb->len;
6332         u16 gso_segs = 1;
6333
6334         i = tx_ring->next_to_use;
6335
6336         if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6337                 /* excluding fcoe_crc_eof for FCoE */
6338                 total -= sizeof(struct fcoe_crc_eof);
6339
6340         len = min(skb_headlen(skb), total);
6341         while (len) {
6342                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6343                 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6344
6345                 tx_buffer_info->length = size;
6346                 tx_buffer_info->mapped_as_page = false;
6347                 tx_buffer_info->dma = dma_map_single(dev,
6348                                                      skb->data + offset,
6349                                                      size, DMA_TO_DEVICE);
6350                 if (dma_mapping_error(dev, tx_buffer_info->dma))
6351                         goto dma_error;
6352                 tx_buffer_info->time_stamp = jiffies;
6353                 tx_buffer_info->next_to_watch = i;
6354
6355                 len -= size;
6356                 total -= size;
6357                 offset += size;
6358                 count++;
6359
6360                 if (len) {
6361                         i++;
6362                         if (i == tx_ring->count)
6363                                 i = 0;
6364                 }
6365         }
6366
6367         for (f = 0; f < nr_frags; f++) {
6368                 struct skb_frag_struct *frag;
6369
6370                 frag = &skb_shinfo(skb)->frags[f];
6371                 len = min((unsigned int)frag->size, total);
6372                 offset = frag->page_offset;
6373
6374                 while (len) {
6375                         i++;
6376                         if (i == tx_ring->count)
6377                                 i = 0;
6378
6379                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
6380                         size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6381
6382                         tx_buffer_info->length = size;
6383                         tx_buffer_info->dma = dma_map_page(dev,
6384                                                            frag->page,
6385                                                            offset, size,
6386                                                            DMA_TO_DEVICE);
6387                         tx_buffer_info->mapped_as_page = true;
6388                         if (dma_mapping_error(dev, tx_buffer_info->dma))
6389                                 goto dma_error;
6390                         tx_buffer_info->time_stamp = jiffies;
6391                         tx_buffer_info->next_to_watch = i;
6392
6393                         len -= size;
6394                         total -= size;
6395                         offset += size;
6396                         count++;
6397                 }
6398                 if (total == 0)
6399                         break;
6400         }
6401
6402         if (tx_flags & IXGBE_TX_FLAGS_TSO)
6403                 gso_segs = skb_shinfo(skb)->gso_segs;
6404 #ifdef IXGBE_FCOE
6405         /* adjust for FCoE Sequence Offload */
6406         else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6407                 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6408                                         skb_shinfo(skb)->gso_size);
6409 #endif /* IXGBE_FCOE */
6410         bytecount += (gso_segs - 1) * hdr_len;
6411
6412         /* multiply data chunks by size of headers */
6413         tx_ring->tx_buffer_info[i].bytecount = bytecount;
6414         tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6415         tx_ring->tx_buffer_info[i].skb = skb;
6416         tx_ring->tx_buffer_info[first].next_to_watch = i;
6417
6418         return count;
6419
6420 dma_error:
6421         e_dev_err("TX DMA map failed\n");
6422
6423         /* clear timestamp and dma mappings for failed tx_buffer_info map */
6424         tx_buffer_info->dma = 0;
6425         tx_buffer_info->time_stamp = 0;
6426         tx_buffer_info->next_to_watch = 0;
6427         if (count)
6428                 count--;
6429
6430         /* clear timestamp and dma mappings for remaining portion of packet */
6431         while (count--) {
6432                 if (i == 0)
6433                         i += tx_ring->count;
6434                 i--;
6435                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6436                 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6437         }
6438
6439         return 0;
6440 }
6441
6442 static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
6443                            int tx_flags, int count, u32 paylen, u8 hdr_len)
6444 {
6445         union ixgbe_adv_tx_desc *tx_desc = NULL;
6446         struct ixgbe_tx_buffer *tx_buffer_info;
6447         u32 olinfo_status = 0, cmd_type_len = 0;
6448         unsigned int i;
6449         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6450
6451         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6452
6453         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6454
6455         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6456                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6457
6458         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6459                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6460
6461                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6462                                  IXGBE_ADVTXD_POPTS_SHIFT;
6463
6464                 /* use index 1 context for tso */
6465                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6466                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6467                         olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6468                                          IXGBE_ADVTXD_POPTS_SHIFT;
6469
6470         } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6471                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6472                                  IXGBE_ADVTXD_POPTS_SHIFT;
6473
6474         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6475                 olinfo_status |= IXGBE_ADVTXD_CC;
6476                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6477                 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6478                         cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6479         }
6480
6481         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6482
6483         i = tx_ring->next_to_use;
6484         while (count--) {
6485                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6486                 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6487                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6488                 tx_desc->read.cmd_type_len =
6489                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6490                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6491                 i++;
6492                 if (i == tx_ring->count)
6493                         i = 0;
6494         }
6495
6496         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6497
6498         /*
6499          * Force memory writes to complete before letting h/w
6500          * know there are new descriptors to fetch.  (Only
6501          * applicable for weak-ordered memory model archs,
6502          * such as IA-64).
6503          */
6504         wmb();
6505
6506         tx_ring->next_to_use = i;
6507         writel(i, tx_ring->tail);
6508 }
6509
6510 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6511                       u32 tx_flags, __be16 protocol)
6512 {
6513         struct ixgbe_q_vector *q_vector = ring->q_vector;
6514         union ixgbe_atr_hash_dword input = { .dword = 0 };
6515         union ixgbe_atr_hash_dword common = { .dword = 0 };
6516         union {
6517                 unsigned char *network;
6518                 struct iphdr *ipv4;
6519                 struct ipv6hdr *ipv6;
6520         } hdr;
6521         struct tcphdr *th;
6522         __be16 vlan_id;
6523
6524         /* if ring doesn't have a interrupt vector, cannot perform ATR */
6525         if (!q_vector)
6526                 return;
6527
6528         /* do nothing if sampling is disabled */
6529         if (!ring->atr_sample_rate)
6530                 return;
6531
6532         ring->atr_count++;
6533
6534         /* snag network header to get L4 type and address */
6535         hdr.network = skb_network_header(skb);
6536
6537         /* Currently only IPv4/IPv6 with TCP is supported */
6538         if ((protocol != __constant_htons(ETH_P_IPV6) ||
6539              hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6540             (protocol != __constant_htons(ETH_P_IP) ||
6541              hdr.ipv4->protocol != IPPROTO_TCP))
6542                 return;
6543
6544         th = tcp_hdr(skb);
6545
6546         /* skip this packet since the socket is closing */
6547         if (th->fin)
6548                 return;
6549
6550         /* sample on all syn packets or once every atr sample count */
6551         if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6552                 return;
6553
6554         /* reset sample count */
6555         ring->atr_count = 0;
6556
6557         vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6558
6559         /*
6560          * src and dst are inverted, think how the receiver sees them
6561          *
6562          * The input is broken into two sections, a non-compressed section
6563          * containing vm_pool, vlan_id, and flow_type.  The rest of the data
6564          * is XORed together and stored in the compressed dword.
6565          */
6566         input.formatted.vlan_id = vlan_id;
6567
6568         /*
6569          * since src port and flex bytes occupy the same word XOR them together
6570          * and write the value to source port portion of compressed dword
6571          */
6572         if (vlan_id)
6573                 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6574         else
6575                 common.port.src ^= th->dest ^ protocol;
6576         common.port.dst ^= th->source;
6577
6578         if (protocol == __constant_htons(ETH_P_IP)) {
6579                 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6580                 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6581         } else {
6582                 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6583                 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6584                              hdr.ipv6->saddr.s6_addr32[1] ^
6585                              hdr.ipv6->saddr.s6_addr32[2] ^
6586                              hdr.ipv6->saddr.s6_addr32[3] ^
6587                              hdr.ipv6->daddr.s6_addr32[0] ^
6588                              hdr.ipv6->daddr.s6_addr32[1] ^
6589                              hdr.ipv6->daddr.s6_addr32[2] ^
6590                              hdr.ipv6->daddr.s6_addr32[3];
6591         }
6592
6593         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6594         ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6595                                               input, common, ring->queue_index);
6596 }
6597
6598 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6599 {
6600         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6601         /* Herbert's original patch had:
6602          *  smp_mb__after_netif_stop_queue();
6603          * but since that doesn't exist yet, just open code it. */
6604         smp_mb();
6605
6606         /* We need to check again in a case another CPU has just
6607          * made room available. */
6608         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6609                 return -EBUSY;
6610
6611         /* A reprieve! - use start_queue because it doesn't call schedule */
6612         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6613         ++tx_ring->tx_stats.restart_queue;
6614         return 0;
6615 }
6616
6617 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6618 {
6619         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6620                 return 0;
6621         return __ixgbe_maybe_stop_tx(tx_ring, size);
6622 }
6623
6624 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6625 {
6626         struct ixgbe_adapter *adapter = netdev_priv(dev);
6627         int txq = smp_processor_id();
6628 #ifdef IXGBE_FCOE
6629         __be16 protocol;
6630
6631         protocol = vlan_get_protocol(skb);
6632
6633         if ((protocol == htons(ETH_P_FCOE)) ||
6634             (protocol == htons(ETH_P_FIP))) {
6635                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6636                         txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6637                         txq += adapter->ring_feature[RING_F_FCOE].mask;
6638                         return txq;
6639 #ifdef CONFIG_IXGBE_DCB
6640                 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6641                         txq = adapter->fcoe.up;
6642                         return txq;
6643 #endif
6644                 }
6645         }
6646 #endif
6647
6648         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6649                 while (unlikely(txq >= dev->real_num_tx_queues))
6650                         txq -= dev->real_num_tx_queues;
6651                 return txq;
6652         }
6653
6654         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6655                 if (skb->priority == TC_PRIO_CONTROL)
6656                         txq = adapter->ring_feature[RING_F_DCB].indices-1;
6657                 else
6658                         txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6659                                >> 13;
6660                 return txq;
6661         }
6662
6663         return skb_tx_hash(dev, skb);
6664 }
6665
6666 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6667                           struct ixgbe_adapter *adapter,
6668                           struct ixgbe_ring *tx_ring)
6669 {
6670         unsigned int first;
6671         unsigned int tx_flags = 0;
6672         u8 hdr_len = 0;
6673         int tso;
6674         int count = 0;
6675         unsigned int f;
6676         __be16 protocol;
6677
6678         protocol = vlan_get_protocol(skb);
6679
6680         if (vlan_tx_tag_present(skb)) {
6681                 tx_flags |= vlan_tx_tag_get(skb);
6682                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6683                         tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6684                         tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6685                 }
6686                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6687                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6688         } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6689                    skb->priority != TC_PRIO_CONTROL) {
6690                 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6691                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6692                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6693         }
6694
6695 #ifdef IXGBE_FCOE
6696         /* for FCoE with DCB, we force the priority to what
6697          * was specified by the switch */
6698         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6699             (protocol == htons(ETH_P_FCOE) ||
6700              protocol == htons(ETH_P_FIP))) {
6701 #ifdef CONFIG_IXGBE_DCB
6702                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6703                         tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6704                                       << IXGBE_TX_FLAGS_VLAN_SHIFT);
6705                         tx_flags |= ((adapter->fcoe.up << 13)
6706                                       << IXGBE_TX_FLAGS_VLAN_SHIFT);
6707                 }
6708 #endif
6709                 /* flag for FCoE offloads */
6710                 if (protocol == htons(ETH_P_FCOE))
6711                         tx_flags |= IXGBE_TX_FLAGS_FCOE;
6712         }
6713 #endif
6714
6715         /* four things can cause us to need a context descriptor */
6716         if (skb_is_gso(skb) ||
6717             (skb->ip_summed == CHECKSUM_PARTIAL) ||
6718             (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6719             (tx_flags & IXGBE_TX_FLAGS_FCOE))
6720                 count++;
6721
6722         count += TXD_USE_COUNT(skb_headlen(skb));
6723         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6724                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6725
6726         if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6727                 tx_ring->tx_stats.tx_busy++;
6728                 return NETDEV_TX_BUSY;
6729         }
6730
6731         first = tx_ring->next_to_use;
6732         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6733 #ifdef IXGBE_FCOE
6734                 /* setup tx offload for FCoE */
6735                 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6736                 if (tso < 0) {
6737                         dev_kfree_skb_any(skb);
6738                         return NETDEV_TX_OK;
6739                 }
6740                 if (tso)
6741                         tx_flags |= IXGBE_TX_FLAGS_FSO;
6742 #endif /* IXGBE_FCOE */
6743         } else {
6744                 if (protocol == htons(ETH_P_IP))
6745                         tx_flags |= IXGBE_TX_FLAGS_IPV4;
6746                 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6747                                 protocol);
6748                 if (tso < 0) {
6749                         dev_kfree_skb_any(skb);
6750                         return NETDEV_TX_OK;
6751                 }
6752
6753                 if (tso)
6754                         tx_flags |= IXGBE_TX_FLAGS_TSO;
6755                 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6756                                        protocol) &&
6757                          (skb->ip_summed == CHECKSUM_PARTIAL))
6758                         tx_flags |= IXGBE_TX_FLAGS_CSUM;
6759         }
6760
6761         count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6762         if (count) {
6763                 /* add the ATR filter if ATR is on */
6764                 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6765                         ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6766                 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6767                 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6768
6769         } else {
6770                 dev_kfree_skb_any(skb);
6771                 tx_ring->tx_buffer_info[first].time_stamp = 0;
6772                 tx_ring->next_to_use = first;
6773         }
6774
6775         return NETDEV_TX_OK;
6776 }
6777
6778 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6779 {
6780         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6781         struct ixgbe_ring *tx_ring;
6782
6783         tx_ring = adapter->tx_ring[skb->queue_mapping];
6784         return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6785 }
6786
6787 /**
6788  * ixgbe_set_mac - Change the Ethernet Address of the NIC
6789  * @netdev: network interface device structure
6790  * @p: pointer to an address structure
6791  *
6792  * Returns 0 on success, negative on failure
6793  **/
6794 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6795 {
6796         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6797         struct ixgbe_hw *hw = &adapter->hw;
6798         struct sockaddr *addr = p;
6799
6800         if (!is_valid_ether_addr(addr->sa_data))
6801                 return -EADDRNOTAVAIL;
6802
6803         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6804         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6805
6806         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6807                             IXGBE_RAH_AV);
6808
6809         return 0;
6810 }
6811
6812 static int
6813 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6814 {
6815         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6816         struct ixgbe_hw *hw = &adapter->hw;
6817         u16 value;
6818         int rc;
6819
6820         if (prtad != hw->phy.mdio.prtad)
6821                 return -EINVAL;
6822         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6823         if (!rc)
6824                 rc = value;
6825         return rc;
6826 }
6827
6828 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6829                             u16 addr, u16 value)
6830 {
6831         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6832         struct ixgbe_hw *hw = &adapter->hw;
6833
6834         if (prtad != hw->phy.mdio.prtad)
6835                 return -EINVAL;
6836         return hw->phy.ops.write_reg(hw, addr, devad, value);
6837 }
6838
6839 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6840 {
6841         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6842
6843         return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6844 }
6845
6846 /**
6847  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6848  * netdev->dev_addrs
6849  * @netdev: network interface device structure
6850  *
6851  * Returns non-zero on failure
6852  **/
6853 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6854 {
6855         int err = 0;
6856         struct ixgbe_adapter *adapter = netdev_priv(dev);
6857         struct ixgbe_mac_info *mac = &adapter->hw.mac;
6858
6859         if (is_valid_ether_addr(mac->san_addr)) {
6860                 rtnl_lock();
6861                 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6862                 rtnl_unlock();
6863         }
6864         return err;
6865 }
6866
6867 /**
6868  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6869  * netdev->dev_addrs
6870  * @netdev: network interface device structure
6871  *
6872  * Returns non-zero on failure
6873  **/
6874 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6875 {
6876         int err = 0;
6877         struct ixgbe_adapter *adapter = netdev_priv(dev);
6878         struct ixgbe_mac_info *mac = &adapter->hw.mac;
6879
6880         if (is_valid_ether_addr(mac->san_addr)) {
6881                 rtnl_lock();
6882                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6883                 rtnl_unlock();
6884         }
6885         return err;
6886 }
6887
6888 #ifdef CONFIG_NET_POLL_CONTROLLER
6889 /*
6890  * Polling 'interrupt' - used by things like netconsole to send skbs
6891  * without having to re-enable interrupts. It's not called while
6892  * the interrupt routine is executing.
6893  */
6894 static void ixgbe_netpoll(struct net_device *netdev)
6895 {
6896         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6897         int i;
6898
6899         /* if interface is down do nothing */
6900         if (test_bit(__IXGBE_DOWN, &adapter->state))
6901                 return;
6902
6903         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6904         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6905                 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6906                 for (i = 0; i < num_q_vectors; i++) {
6907                         struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6908                         ixgbe_msix_clean_many(0, q_vector);
6909                 }
6910         } else {
6911                 ixgbe_intr(adapter->pdev->irq, netdev);
6912         }
6913         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6914 }
6915 #endif
6916
6917 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6918                                                    struct rtnl_link_stats64 *stats)
6919 {
6920         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6921         int i;
6922
6923         rcu_read_lock();
6924         for (i = 0; i < adapter->num_rx_queues; i++) {
6925                 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6926                 u64 bytes, packets;
6927                 unsigned int start;
6928
6929                 if (ring) {
6930                         do {
6931                                 start = u64_stats_fetch_begin_bh(&ring->syncp);
6932                                 packets = ring->stats.packets;
6933                                 bytes   = ring->stats.bytes;
6934                         } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6935                         stats->rx_packets += packets;
6936                         stats->rx_bytes   += bytes;
6937                 }
6938         }
6939
6940         for (i = 0; i < adapter->num_tx_queues; i++) {
6941                 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6942                 u64 bytes, packets;
6943                 unsigned int start;
6944
6945                 if (ring) {
6946                         do {
6947                                 start = u64_stats_fetch_begin_bh(&ring->syncp);
6948                                 packets = ring->stats.packets;
6949                                 bytes   = ring->stats.bytes;
6950                         } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6951                         stats->tx_packets += packets;
6952                         stats->tx_bytes   += bytes;
6953                 }
6954         }
6955         rcu_read_unlock();
6956         /* following stats updated by ixgbe_watchdog_task() */
6957         stats->multicast        = netdev->stats.multicast;
6958         stats->rx_errors        = netdev->stats.rx_errors;
6959         stats->rx_length_errors = netdev->stats.rx_length_errors;
6960         stats->rx_crc_errors    = netdev->stats.rx_crc_errors;
6961         stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6962         return stats;
6963 }
6964
6965
6966 static const struct net_device_ops ixgbe_netdev_ops = {
6967         .ndo_open               = ixgbe_open,
6968         .ndo_stop               = ixgbe_close,
6969         .ndo_start_xmit         = ixgbe_xmit_frame,
6970         .ndo_select_queue       = ixgbe_select_queue,
6971         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
6972         .ndo_set_multicast_list = ixgbe_set_rx_mode,
6973         .ndo_validate_addr      = eth_validate_addr,
6974         .ndo_set_mac_address    = ixgbe_set_mac,
6975         .ndo_change_mtu         = ixgbe_change_mtu,
6976         .ndo_tx_timeout         = ixgbe_tx_timeout,
6977         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
6978         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
6979         .ndo_do_ioctl           = ixgbe_ioctl,
6980         .ndo_set_vf_mac         = ixgbe_ndo_set_vf_mac,
6981         .ndo_set_vf_vlan        = ixgbe_ndo_set_vf_vlan,
6982         .ndo_set_vf_tx_rate     = ixgbe_ndo_set_vf_bw,
6983         .ndo_get_vf_config      = ixgbe_ndo_get_vf_config,
6984         .ndo_get_stats64        = ixgbe_get_stats64,
6985 #ifdef CONFIG_NET_POLL_CONTROLLER
6986         .ndo_poll_controller    = ixgbe_netpoll,
6987 #endif
6988 #ifdef IXGBE_FCOE
6989         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6990         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6991         .ndo_fcoe_enable = ixgbe_fcoe_enable,
6992         .ndo_fcoe_disable = ixgbe_fcoe_disable,
6993         .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6994 #endif /* IXGBE_FCOE */
6995 };
6996
6997 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6998                            const struct ixgbe_info *ii)
6999 {
7000 #ifdef CONFIG_PCI_IOV
7001         struct ixgbe_hw *hw = &adapter->hw;
7002         int err;
7003
7004         if (hw->mac.type == ixgbe_mac_82598EB || !max_vfs)
7005                 return;
7006
7007         /* The 82599 supports up to 64 VFs per physical function
7008          * but this implementation limits allocation to 63 so that
7009          * basic networking resources are still available to the
7010          * physical function
7011          */
7012         adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7013         adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
7014         err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
7015         if (err) {
7016                 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
7017                 goto err_novfs;
7018         }
7019         /* If call to enable VFs succeeded then allocate memory
7020          * for per VF control structures.
7021          */
7022         adapter->vfinfo =
7023                 kcalloc(adapter->num_vfs,
7024                         sizeof(struct vf_data_storage), GFP_KERNEL);
7025         if (adapter->vfinfo) {
7026                 /* Now that we're sure SR-IOV is enabled
7027                  * and memory allocated set up the mailbox parameters
7028                  */
7029                 ixgbe_init_mbx_params_pf(hw);
7030                 memcpy(&hw->mbx.ops, ii->mbx_ops,
7031                        sizeof(hw->mbx.ops));
7032
7033                 /* Disable RSC when in SR-IOV mode */
7034                 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
7035                                      IXGBE_FLAG2_RSC_ENABLED);
7036                 return;
7037         }
7038
7039         /* Oh oh */
7040         e_err(probe, "Unable to allocate memory for VF Data Storage - "
7041               "SRIOV disabled\n");
7042         pci_disable_sriov(adapter->pdev);
7043
7044 err_novfs:
7045         adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
7046         adapter->num_vfs = 0;
7047 #endif /* CONFIG_PCI_IOV */
7048 }
7049
7050 /**
7051  * ixgbe_probe - Device Initialization Routine
7052  * @pdev: PCI device information struct
7053  * @ent: entry in ixgbe_pci_tbl
7054  *
7055  * Returns 0 on success, negative on failure
7056  *
7057  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7058  * The OS initialization, configuring of the adapter private structure,
7059  * and a hardware reset occur.
7060  **/
7061 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7062                                  const struct pci_device_id *ent)
7063 {
7064         struct net_device *netdev;
7065         struct ixgbe_adapter *adapter = NULL;
7066         struct ixgbe_hw *hw;
7067         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7068         static int cards_found;
7069         int i, err, pci_using_dac;
7070         u8 part_str[IXGBE_PBANUM_LENGTH];
7071         unsigned int indices = num_possible_cpus();
7072 #ifdef IXGBE_FCOE
7073         u16 device_caps;
7074 #endif
7075         u32 eec;
7076
7077         /* Catch broken hardware that put the wrong VF device ID in
7078          * the PCIe SR-IOV capability.
7079          */
7080         if (pdev->is_virtfn) {
7081                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7082                      pci_name(pdev), pdev->vendor, pdev->device);
7083                 return -EINVAL;
7084         }
7085
7086         err = pci_enable_device_mem(pdev);
7087         if (err)
7088                 return err;
7089
7090         if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7091             !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7092                 pci_using_dac = 1;
7093         } else {
7094                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7095                 if (err) {
7096                         err = dma_set_coherent_mask(&pdev->dev,
7097                                                     DMA_BIT_MASK(32));
7098                         if (err) {
7099                                 dev_err(&pdev->dev,
7100                                         "No usable DMA configuration, aborting\n");
7101                                 goto err_dma;
7102                         }
7103                 }
7104                 pci_using_dac = 0;
7105         }
7106
7107         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7108                                            IORESOURCE_MEM), ixgbe_driver_name);
7109         if (err) {
7110                 dev_err(&pdev->dev,
7111                         "pci_request_selected_regions failed 0x%x\n", err);
7112                 goto err_pci_reg;
7113         }
7114
7115         pci_enable_pcie_error_reporting(pdev);
7116
7117         pci_set_master(pdev);
7118         pci_save_state(pdev);
7119
7120         if (ii->mac == ixgbe_mac_82598EB)
7121                 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7122         else
7123                 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7124
7125         indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
7126 #ifdef IXGBE_FCOE
7127         indices += min_t(unsigned int, num_possible_cpus(),
7128                          IXGBE_MAX_FCOE_INDICES);
7129 #endif
7130         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7131         if (!netdev) {
7132                 err = -ENOMEM;
7133                 goto err_alloc_etherdev;
7134         }
7135
7136         SET_NETDEV_DEV(netdev, &pdev->dev);
7137
7138         adapter = netdev_priv(netdev);
7139         pci_set_drvdata(pdev, adapter);
7140
7141         adapter->netdev = netdev;
7142         adapter->pdev = pdev;
7143         hw = &adapter->hw;
7144         hw->back = adapter;
7145         adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7146
7147         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7148                               pci_resource_len(pdev, 0));
7149         if (!hw->hw_addr) {
7150                 err = -EIO;
7151                 goto err_ioremap;
7152         }
7153
7154         for (i = 1; i <= 5; i++) {
7155                 if (pci_resource_len(pdev, i) == 0)
7156                         continue;
7157         }
7158
7159         netdev->netdev_ops = &ixgbe_netdev_ops;
7160         ixgbe_set_ethtool_ops(netdev);
7161         netdev->watchdog_timeo = 5 * HZ;
7162         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7163
7164         adapter->bd_number = cards_found;
7165
7166         /* Setup hw api */
7167         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7168         hw->mac.type  = ii->mac;
7169
7170         /* EEPROM */
7171         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7172         eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7173         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7174         if (!(eec & (1 << 8)))
7175                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7176
7177         /* PHY */
7178         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7179         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7180         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7181         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7182         hw->phy.mdio.mmds = 0;
7183         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7184         hw->phy.mdio.dev = netdev;
7185         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7186         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7187
7188         /* set up this timer and work struct before calling get_invariants
7189          * which might start the timer
7190          */
7191         init_timer(&adapter->sfp_timer);
7192         adapter->sfp_timer.function = ixgbe_sfp_timer;
7193         adapter->sfp_timer.data = (unsigned long) adapter;
7194
7195         INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
7196
7197         /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
7198         INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
7199
7200         /* a new SFP+ module arrival, called from GPI SDP2 context */
7201         INIT_WORK(&adapter->sfp_config_module_task,
7202                   ixgbe_sfp_config_module_task);
7203
7204         ii->get_invariants(hw);
7205
7206         /* setup the private structure */
7207         err = ixgbe_sw_init(adapter);
7208         if (err)
7209                 goto err_sw_init;
7210
7211         /* Make it possible the adapter to be woken up via WOL */
7212         switch (adapter->hw.mac.type) {
7213         case ixgbe_mac_82599EB:
7214         case ixgbe_mac_X540:
7215                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7216                 break;
7217         default:
7218                 break;
7219         }
7220
7221         /*
7222          * If there is a fan on this device and it has failed log the
7223          * failure.
7224          */
7225         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7226                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7227                 if (esdp & IXGBE_ESDP_SDP1)
7228                         e_crit(probe, "Fan has stopped, replace the adapter\n");
7229         }
7230
7231         /* reset_hw fills in the perm_addr as well */
7232         hw->phy.reset_if_overtemp = true;
7233         err = hw->mac.ops.reset_hw(hw);
7234         hw->phy.reset_if_overtemp = false;
7235         if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7236             hw->mac.type == ixgbe_mac_82598EB) {
7237                 /*
7238                  * Start a kernel thread to watch for a module to arrive.
7239                  * Only do this for 82598, since 82599 will generate
7240                  * interrupts on module arrival.
7241                  */
7242                 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7243                 mod_timer(&adapter->sfp_timer,
7244                           round_jiffies(jiffies + (2 * HZ)));
7245                 err = 0;
7246         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7247                 e_dev_err("failed to initialize because an unsupported SFP+ "
7248                           "module type was detected.\n");
7249                 e_dev_err("Reload the driver after installing a supported "
7250                           "module.\n");
7251                 goto err_sw_init;
7252         } else if (err) {
7253                 e_dev_err("HW Init failed: %d\n", err);
7254                 goto err_sw_init;
7255         }
7256
7257         ixgbe_probe_vf(adapter, ii);
7258
7259         netdev->features = NETIF_F_SG |
7260                            NETIF_F_IP_CSUM |
7261                            NETIF_F_HW_VLAN_TX |
7262                            NETIF_F_HW_VLAN_RX |
7263                            NETIF_F_HW_VLAN_FILTER;
7264
7265         netdev->features |= NETIF_F_IPV6_CSUM;
7266         netdev->features |= NETIF_F_TSO;
7267         netdev->features |= NETIF_F_TSO6;
7268         netdev->features |= NETIF_F_GRO;
7269
7270         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
7271                 netdev->features |= NETIF_F_SCTP_CSUM;
7272
7273         netdev->vlan_features |= NETIF_F_TSO;
7274         netdev->vlan_features |= NETIF_F_TSO6;
7275         netdev->vlan_features |= NETIF_F_IP_CSUM;
7276         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7277         netdev->vlan_features |= NETIF_F_SG;
7278
7279         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7280                 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7281                                     IXGBE_FLAG_DCB_ENABLED);
7282         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
7283                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
7284
7285 #ifdef CONFIG_IXGBE_DCB
7286         netdev->dcbnl_ops = &dcbnl_ops;
7287 #endif
7288
7289 #ifdef IXGBE_FCOE
7290         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7291                 if (hw->mac.ops.get_device_caps) {
7292                         hw->mac.ops.get_device_caps(hw, &device_caps);
7293                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7294                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7295                 }
7296         }
7297         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7298                 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7299                 netdev->vlan_features |= NETIF_F_FSO;
7300                 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7301         }
7302 #endif /* IXGBE_FCOE */
7303         if (pci_using_dac) {
7304                 netdev->features |= NETIF_F_HIGHDMA;
7305                 netdev->vlan_features |= NETIF_F_HIGHDMA;
7306         }
7307
7308         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7309                 netdev->features |= NETIF_F_LRO;
7310
7311         /* make sure the EEPROM is good */
7312         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7313                 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7314                 err = -EIO;
7315                 goto err_eeprom;
7316         }
7317
7318         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7319         memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7320
7321         if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7322                 e_dev_err("invalid MAC address\n");
7323                 err = -EIO;
7324                 goto err_eeprom;
7325         }
7326
7327         /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7328         if (hw->mac.ops.disable_tx_laser &&
7329             ((hw->phy.multispeed_fiber) ||
7330              ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7331               (hw->mac.type == ixgbe_mac_82599EB))))
7332                 hw->mac.ops.disable_tx_laser(hw);
7333
7334         init_timer(&adapter->watchdog_timer);
7335         adapter->watchdog_timer.function = ixgbe_watchdog;
7336         adapter->watchdog_timer.data = (unsigned long)adapter;
7337
7338         INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
7339         INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
7340
7341         err = ixgbe_init_interrupt_scheme(adapter);
7342         if (err)
7343                 goto err_sw_init;
7344
7345         switch (pdev->device) {
7346         case IXGBE_DEV_ID_82599_SFP:
7347                 /* Only this subdevice supports WOL */
7348                 if (pdev->subsystem_device == IXGBE_SUBDEV_ID_82599_SFP)
7349                         adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7350                                         IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7351                 break;
7352         case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7353                 /* All except this subdevice support WOL */
7354                 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7355                         adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7356                                         IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7357                 break;
7358         case IXGBE_DEV_ID_82599_KX4:
7359                 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7360                                 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7361                 break;
7362         default:
7363                 adapter->wol = 0;
7364                 break;
7365         }
7366         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7367
7368         /* pick up the PCI bus settings for reporting later */
7369         hw->mac.ops.get_bus_info(hw);
7370
7371         /* print bus type/speed/width info */
7372         e_dev_info("(PCI Express:%s:%s) %pM\n",
7373                    (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
7374                     hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
7375                     "Unknown"),
7376                    (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7377                     hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7378                     hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7379                     "Unknown"),
7380                    netdev->dev_addr);
7381
7382         err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7383         if (err)
7384                 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7385         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7386                 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7387                            hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7388                            part_str);
7389         else
7390                 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7391                            hw->mac.type, hw->phy.type, part_str);
7392
7393         if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7394                 e_dev_warn("PCI-Express bandwidth available for this card is "
7395                            "not sufficient for optimal performance.\n");
7396                 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7397                            "is required.\n");
7398         }
7399
7400         /* save off EEPROM version number */
7401         hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7402
7403         /* reset the hardware with the new settings */
7404         err = hw->mac.ops.start_hw(hw);
7405
7406         if (err == IXGBE_ERR_EEPROM_VERSION) {
7407                 /* We are running on a pre-production device, log a warning */
7408                 e_dev_warn("This device is a pre-production adapter/LOM. "
7409                            "Please be aware there may be issues associated "
7410                            "with your hardware.  If you are experiencing "
7411                            "problems please contact your Intel or hardware "
7412                            "representative who provided you with this "
7413                            "hardware.\n");
7414         }
7415         strcpy(netdev->name, "eth%d");
7416         err = register_netdev(netdev);
7417         if (err)
7418                 goto err_register;
7419
7420         /* carrier off reporting is important to ethtool even BEFORE open */
7421         netif_carrier_off(netdev);
7422
7423         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7424             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7425                 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7426
7427         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7428                 INIT_WORK(&adapter->check_overtemp_task,
7429                           ixgbe_check_overtemp_task);
7430 #ifdef CONFIG_IXGBE_DCA
7431         if (dca_add_requester(&pdev->dev) == 0) {
7432                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7433                 ixgbe_setup_dca(adapter);
7434         }
7435 #endif
7436         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7437                 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7438                 for (i = 0; i < adapter->num_vfs; i++)
7439                         ixgbe_vf_configuration(pdev, (i | 0x10000000));
7440         }
7441
7442         /* add san mac addr to netdev */
7443         ixgbe_add_sanmac_netdev(netdev);
7444
7445         e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7446         cards_found++;
7447         return 0;
7448
7449 err_register:
7450         ixgbe_release_hw_control(adapter);
7451         ixgbe_clear_interrupt_scheme(adapter);
7452 err_sw_init:
7453 err_eeprom:
7454         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7455                 ixgbe_disable_sriov(adapter);
7456         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7457         del_timer_sync(&adapter->sfp_timer);
7458         cancel_work_sync(&adapter->sfp_task);
7459         cancel_work_sync(&adapter->multispeed_fiber_task);
7460         cancel_work_sync(&adapter->sfp_config_module_task);
7461         iounmap(hw->hw_addr);
7462 err_ioremap:
7463         free_netdev(netdev);
7464 err_alloc_etherdev:
7465         pci_release_selected_regions(pdev,
7466                                      pci_select_bars(pdev, IORESOURCE_MEM));
7467 err_pci_reg:
7468 err_dma:
7469         pci_disable_device(pdev);
7470         return err;
7471 }
7472
7473 /**
7474  * ixgbe_remove - Device Removal Routine
7475  * @pdev: PCI device information struct
7476  *
7477  * ixgbe_remove is called by the PCI subsystem to alert the driver
7478  * that it should release a PCI device.  The could be caused by a
7479  * Hot-Plug event, or because the driver is going to be removed from
7480  * memory.
7481  **/
7482 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7483 {
7484         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7485         struct net_device *netdev = adapter->netdev;
7486
7487         set_bit(__IXGBE_DOWN, &adapter->state);
7488
7489         /*
7490          * The timers may be rescheduled, so explicitly disable them
7491          * from being rescheduled.
7492          */
7493         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7494         del_timer_sync(&adapter->watchdog_timer);
7495         del_timer_sync(&adapter->sfp_timer);
7496
7497         cancel_work_sync(&adapter->watchdog_task);
7498         cancel_work_sync(&adapter->sfp_task);
7499         cancel_work_sync(&adapter->multispeed_fiber_task);
7500         cancel_work_sync(&adapter->sfp_config_module_task);
7501         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7502             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7503                 cancel_work_sync(&adapter->fdir_reinit_task);
7504         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7505                 cancel_work_sync(&adapter->check_overtemp_task);
7506
7507 #ifdef CONFIG_IXGBE_DCA
7508         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7509                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7510                 dca_remove_requester(&pdev->dev);
7511                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7512         }
7513
7514 #endif
7515 #ifdef IXGBE_FCOE
7516         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7517                 ixgbe_cleanup_fcoe(adapter);
7518
7519 #endif /* IXGBE_FCOE */
7520
7521         /* remove the added san mac */
7522         ixgbe_del_sanmac_netdev(netdev);
7523
7524         if (netdev->reg_state == NETREG_REGISTERED)
7525                 unregister_netdev(netdev);
7526
7527         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7528                 ixgbe_disable_sriov(adapter);
7529
7530         ixgbe_clear_interrupt_scheme(adapter);
7531
7532         ixgbe_release_hw_control(adapter);
7533
7534         iounmap(adapter->hw.hw_addr);
7535         pci_release_selected_regions(pdev, pci_select_bars(pdev,
7536                                      IORESOURCE_MEM));
7537
7538         e_dev_info("complete\n");
7539
7540         free_netdev(netdev);
7541
7542         pci_disable_pcie_error_reporting(pdev);
7543
7544         pci_disable_device(pdev);
7545 }
7546
7547 /**
7548  * ixgbe_io_error_detected - called when PCI error is detected
7549  * @pdev: Pointer to PCI device
7550  * @state: The current pci connection state
7551  *
7552  * This function is called after a PCI bus error affecting
7553  * this device has been detected.
7554  */
7555 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7556                                                 pci_channel_state_t state)
7557 {
7558         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7559         struct net_device *netdev = adapter->netdev;
7560
7561         netif_device_detach(netdev);
7562
7563         if (state == pci_channel_io_perm_failure)
7564                 return PCI_ERS_RESULT_DISCONNECT;
7565
7566         if (netif_running(netdev))
7567                 ixgbe_down(adapter);
7568         pci_disable_device(pdev);
7569
7570         /* Request a slot reset. */
7571         return PCI_ERS_RESULT_NEED_RESET;
7572 }
7573
7574 /**
7575  * ixgbe_io_slot_reset - called after the pci bus has been reset.
7576  * @pdev: Pointer to PCI device
7577  *
7578  * Restart the card from scratch, as if from a cold-boot.
7579  */
7580 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7581 {
7582         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7583         pci_ers_result_t result;
7584         int err;
7585
7586         if (pci_enable_device_mem(pdev)) {
7587                 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7588                 result = PCI_ERS_RESULT_DISCONNECT;
7589         } else {
7590                 pci_set_master(pdev);
7591                 pci_restore_state(pdev);
7592                 pci_save_state(pdev);
7593
7594                 pci_wake_from_d3(pdev, false);
7595
7596                 ixgbe_reset(adapter);
7597                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7598                 result = PCI_ERS_RESULT_RECOVERED;
7599         }
7600
7601         err = pci_cleanup_aer_uncorrect_error_status(pdev);
7602         if (err) {
7603                 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7604                           "failed 0x%0x\n", err);
7605                 /* non-fatal, continue */
7606         }
7607
7608         return result;
7609 }
7610
7611 /**
7612  * ixgbe_io_resume - called when traffic can start flowing again.
7613  * @pdev: Pointer to PCI device
7614  *
7615  * This callback is called when the error recovery driver tells us that
7616  * its OK to resume normal operation.
7617  */
7618 static void ixgbe_io_resume(struct pci_dev *pdev)
7619 {
7620         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7621         struct net_device *netdev = adapter->netdev;
7622
7623         if (netif_running(netdev)) {
7624                 if (ixgbe_up(adapter)) {
7625                         e_info(probe, "ixgbe_up failed after reset\n");
7626                         return;
7627                 }
7628         }
7629
7630         netif_device_attach(netdev);
7631 }
7632
7633 static struct pci_error_handlers ixgbe_err_handler = {
7634         .error_detected = ixgbe_io_error_detected,
7635         .slot_reset = ixgbe_io_slot_reset,
7636         .resume = ixgbe_io_resume,
7637 };
7638
7639 static struct pci_driver ixgbe_driver = {
7640         .name     = ixgbe_driver_name,
7641         .id_table = ixgbe_pci_tbl,
7642         .probe    = ixgbe_probe,
7643         .remove   = __devexit_p(ixgbe_remove),
7644 #ifdef CONFIG_PM
7645         .suspend  = ixgbe_suspend,
7646         .resume   = ixgbe_resume,
7647 #endif
7648         .shutdown = ixgbe_shutdown,
7649         .err_handler = &ixgbe_err_handler
7650 };
7651
7652 /**
7653  * ixgbe_init_module - Driver Registration Routine
7654  *
7655  * ixgbe_init_module is the first routine called when the driver is
7656  * loaded. All it does is register with the PCI subsystem.
7657  **/
7658 static int __init ixgbe_init_module(void)
7659 {
7660         int ret;
7661         pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7662         pr_info("%s\n", ixgbe_copyright);
7663
7664 #ifdef CONFIG_IXGBE_DCA
7665         dca_register_notify(&dca_notifier);
7666 #endif
7667
7668         ret = pci_register_driver(&ixgbe_driver);
7669         return ret;
7670 }
7671
7672 module_init(ixgbe_init_module);
7673
7674 /**
7675  * ixgbe_exit_module - Driver Exit Cleanup Routine
7676  *
7677  * ixgbe_exit_module is called just before the driver is removed
7678  * from memory.
7679  **/
7680 static void __exit ixgbe_exit_module(void)
7681 {
7682 #ifdef CONFIG_IXGBE_DCA
7683         dca_unregister_notify(&dca_notifier);
7684 #endif
7685         pci_unregister_driver(&ixgbe_driver);
7686         rcu_barrier(); /* Wait for completion of call_rcu()'s */
7687 }
7688
7689 #ifdef CONFIG_IXGBE_DCA
7690 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7691                             void *p)
7692 {
7693         int ret_val;
7694
7695         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7696                                          __ixgbe_notify_dca);
7697
7698         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7699 }
7700
7701 #endif /* CONFIG_IXGBE_DCA */
7702
7703 /**
7704  * ixgbe_get_hw_dev return device
7705  * used by hardware layer to print debugging information
7706  **/
7707 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7708 {
7709         struct ixgbe_adapter *adapter = hw->back;
7710         return adapter->netdev;
7711 }
7712
7713 module_exit(ixgbe_exit_module);
7714
7715 /* ixgbe_main.c */