?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ixgbe: bump version number
[linux-2.6/libata-dev.git] / drivers / net / ethernet / intel / ixgbe / ixgbe_main.c
blob484bbedffe2abd2876a27fe53a2ae7d4e13ff313
1 /*******************************************************************************
2
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2012 Intel Corporation.
5
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
9
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 more details.
14
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
21
22 Contact Information:
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
28 #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/interrupt.h>
36 #include <linux/ip.h>
37 #include <linux/tcp.h>
38 #include <linux/sctp.h>
39 #include <linux/pkt_sched.h>
40 #include <linux/ipv6.h>
41 #include <linux/slab.h>
42 #include <net/checksum.h>
43 #include <net/ip6_checksum.h>
44 #include <linux/ethtool.h>
45 #include <linux/if.h>
46 #include <linux/if_vlan.h>
47 #include <linux/if_bridge.h>
48 #include <linux/prefetch.h>
49 #include <scsi/fc/fc_fcoe.h>
50
51 #include "ixgbe.h"
52 #include "ixgbe_common.h"
53 #include "ixgbe_dcb_82599.h"
54 #include "ixgbe_sriov.h"
55
56 char ixgbe_driver_name[] = "ixgbe";
57 static const char ixgbe_driver_string[] =
58 "Intel(R) 10 Gigabit PCI Express Network Driver";
59 #ifdef IXGBE_FCOE
60 char ixgbe_default_device_descr[] =
61 "Intel(R) 10 Gigabit Network Connection";
62 #else
63 static char ixgbe_default_device_descr[] =
64 "Intel(R) 10 Gigabit Network Connection";
65 #endif
66 #define DRV_VERSION "3.11.33-k"
67 const char ixgbe_driver_version[] = DRV_VERSION;
68 static const char ixgbe_copyright[] =
69 "Copyright (c) 1999-2012 Intel Corporation.";
70
71 static const struct ixgbe_info *ixgbe_info_tbl[] = {
72 [board_82598] = &ixgbe_82598_info,
73 [board_82599] = &ixgbe_82599_info,
74 [board_X540] = &ixgbe_X540_info,
75 };
76
77 /* ixgbe_pci_tbl - PCI Device ID Table
78 *
79 * Wildcard entries (PCI_ANY_ID) should come last
80 * Last entry must be all 0s
81 *
82 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
83 * Class, Class Mask, private data (not used) }
84 */
85 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
115 /* required last entry */
116 {0, }
117 };
118 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
119
120 #ifdef CONFIG_IXGBE_DCA
121 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
122 void *p);
123 static struct notifier_block dca_notifier = {
124 .notifier_call = ixgbe_notify_dca,
125 .next = NULL,
126 .priority = 0
127 };
128 #endif
129
130 #ifdef CONFIG_PCI_IOV
131 static unsigned int max_vfs;
132 module_param(max_vfs, uint, 0);
133 MODULE_PARM_DESC(max_vfs,
134 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63");
135 #endif /* CONFIG_PCI_IOV */
136
137 static unsigned int allow_unsupported_sfp;
138 module_param(allow_unsupported_sfp, uint, 0);
139 MODULE_PARM_DESC(allow_unsupported_sfp,
140 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
141
142 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
143 static int debug = -1;
144 module_param(debug, int, 0);
145 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
146
147 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
148 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
149 MODULE_LICENSE("GPL");
150 MODULE_VERSION(DRV_VERSION);
151
152 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
153 {
154 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
155 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
156 schedule_work(&adapter->service_task);
157 }
158
159 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
160 {
161 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
162
163 /* flush memory to make sure state is correct before next watchdog */
164 smp_mb__before_clear_bit();
165 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
166 }
167
168 struct ixgbe_reg_info {
169 u32 ofs;
170 char *name;
171 };
172
173 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
174
175 /* General Registers */
176 {IXGBE_CTRL, "CTRL"},
177 {IXGBE_STATUS, "STATUS"},
178 {IXGBE_CTRL_EXT, "CTRL_EXT"},
179
180 /* Interrupt Registers */
181 {IXGBE_EICR, "EICR"},
182
183 /* RX Registers */
184 {IXGBE_SRRCTL(0), "SRRCTL"},
185 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
186 {IXGBE_RDLEN(0), "RDLEN"},
187 {IXGBE_RDH(0), "RDH"},
188 {IXGBE_RDT(0), "RDT"},
189 {IXGBE_RXDCTL(0), "RXDCTL"},
190 {IXGBE_RDBAL(0), "RDBAL"},
191 {IXGBE_RDBAH(0), "RDBAH"},
192
193 /* TX Registers */
194 {IXGBE_TDBAL(0), "TDBAL"},
195 {IXGBE_TDBAH(0), "TDBAH"},
196 {IXGBE_TDLEN(0), "TDLEN"},
197 {IXGBE_TDH(0), "TDH"},
198 {IXGBE_TDT(0), "TDT"},
199 {IXGBE_TXDCTL(0), "TXDCTL"},
200
201 /* List Terminator */
202 {}
203 };
204
205
206 /*
207 * ixgbe_regdump - register printout routine
208 */
209 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
210 {
211 int i = 0, j = 0;
212 char rname[16];
213 u32 regs[64];
214
215 switch (reginfo->ofs) {
216 case IXGBE_SRRCTL(0):
217 for (i = 0; i < 64; i++)
218 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
219 break;
220 case IXGBE_DCA_RXCTRL(0):
221 for (i = 0; i < 64; i++)
222 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
223 break;
224 case IXGBE_RDLEN(0):
225 for (i = 0; i < 64; i++)
226 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
227 break;
228 case IXGBE_RDH(0):
229 for (i = 0; i < 64; i++)
230 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
231 break;
232 case IXGBE_RDT(0):
233 for (i = 0; i < 64; i++)
234 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
235 break;
236 case IXGBE_RXDCTL(0):
237 for (i = 0; i < 64; i++)
238 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
239 break;
240 case IXGBE_RDBAL(0):
241 for (i = 0; i < 64; i++)
242 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
243 break;
244 case IXGBE_RDBAH(0):
245 for (i = 0; i < 64; i++)
246 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
247 break;
248 case IXGBE_TDBAL(0):
249 for (i = 0; i < 64; i++)
250 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
251 break;
252 case IXGBE_TDBAH(0):
253 for (i = 0; i < 64; i++)
254 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
255 break;
256 case IXGBE_TDLEN(0):
257 for (i = 0; i < 64; i++)
258 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
259 break;
260 case IXGBE_TDH(0):
261 for (i = 0; i < 64; i++)
262 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
263 break;
264 case IXGBE_TDT(0):
265 for (i = 0; i < 64; i++)
266 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
267 break;
268 case IXGBE_TXDCTL(0):
269 for (i = 0; i < 64; i++)
270 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
271 break;
272 default:
273 pr_info("%-15s %08x\n", reginfo->name,
274 IXGBE_READ_REG(hw, reginfo->ofs));
275 return;
276 }
277
278 for (i = 0; i < 8; i++) {
279 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
280 pr_err("%-15s", rname);
281 for (j = 0; j < 8; j++)
282 pr_cont(" %08x", regs[i*8+j]);
283 pr_cont("\n");
284 }
285
286 }
287
288 /*
289 * ixgbe_dump - Print registers, tx-rings and rx-rings
290 */
291 static void ixgbe_dump(struct ixgbe_adapter *adapter)
292 {
293 struct net_device *netdev = adapter->netdev;
294 struct ixgbe_hw *hw = &adapter->hw;
295 struct ixgbe_reg_info *reginfo;
296 int n = 0;
297 struct ixgbe_ring *tx_ring;
298 struct ixgbe_tx_buffer *tx_buffer;
299 union ixgbe_adv_tx_desc *tx_desc;
300 struct my_u0 { u64 a; u64 b; } *u0;
301 struct ixgbe_ring *rx_ring;
302 union ixgbe_adv_rx_desc *rx_desc;
303 struct ixgbe_rx_buffer *rx_buffer_info;
304 u32 staterr;
305 int i = 0;
306
307 if (!netif_msg_hw(adapter))
308 return;
309
310 /* Print netdevice Info */
311 if (netdev) {
312 dev_info(&adapter->pdev->dev, "Net device Info\n");
313 pr_info("Device Name state "
314 "trans_start last_rx\n");
315 pr_info("%-15s %016lX %016lX %016lX\n",
316 netdev->name,
317 netdev->state,
318 netdev->trans_start,
319 netdev->last_rx);
320 }
321
322 /* Print Registers */
323 dev_info(&adapter->pdev->dev, "Register Dump\n");
324 pr_info(" Register Name Value\n");
325 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
326 reginfo->name; reginfo++) {
327 ixgbe_regdump(hw, reginfo);
328 }
329
330 /* Print TX Ring Summary */
331 if (!netdev || !netif_running(netdev))
332 goto exit;
333
334 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
335 pr_info(" %s %s %s %s\n",
336 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
337 "leng", "ntw", "timestamp");
338 for (n = 0; n < adapter->num_tx_queues; n++) {
339 tx_ring = adapter->tx_ring[n];
340 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
341 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
342 n, tx_ring->next_to_use, tx_ring->next_to_clean,
343 (u64)dma_unmap_addr(tx_buffer, dma),
344 dma_unmap_len(tx_buffer, len),
345 tx_buffer->next_to_watch,
346 (u64)tx_buffer->time_stamp);
347 }
348
349 /* Print TX Rings */
350 if (!netif_msg_tx_done(adapter))
351 goto rx_ring_summary;
352
353 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
354
355 /* Transmit Descriptor Formats
356 *
357 * 82598 Advanced Transmit Descriptor
358 * +--------------------------------------------------------------+
359 * 0 | Buffer Address [63:0] |
360 * +--------------------------------------------------------------+
361 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
362 * +--------------------------------------------------------------+
363 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
364 *
365 * 82598 Advanced Transmit Descriptor (Write-Back Format)
366 * +--------------------------------------------------------------+
367 * 0 | RSV [63:0] |
368 * +--------------------------------------------------------------+
369 * 8 | RSV | STA | NXTSEQ |
370 * +--------------------------------------------------------------+
371 * 63 36 35 32 31 0
372 *
373 * 82599+ Advanced Transmit Descriptor
374 * +--------------------------------------------------------------+
375 * 0 | Buffer Address [63:0] |
376 * +--------------------------------------------------------------+
377 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
378 * +--------------------------------------------------------------+
379 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
380 *
381 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
382 * +--------------------------------------------------------------+
383 * 0 | RSV [63:0] |
384 * +--------------------------------------------------------------+
385 * 8 | RSV | STA | RSV |
386 * +--------------------------------------------------------------+
387 * 63 36 35 32 31 0
388 */
389
390 for (n = 0; n < adapter->num_tx_queues; n++) {
391 tx_ring = adapter->tx_ring[n];
392 pr_info("------------------------------------\n");
393 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
394 pr_info("------------------------------------\n");
395 pr_info("%s%s %s %s %s %s\n",
396 "T [desc] [address 63:0 ] ",
397 "[PlPOIdStDDt Ln] [bi->dma ] ",
398 "leng", "ntw", "timestamp", "bi->skb");
399
400 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
401 tx_desc = IXGBE_TX_DESC(tx_ring, i);
402 tx_buffer = &tx_ring->tx_buffer_info[i];
403 u0 = (struct my_u0 *)tx_desc;
404 if (dma_unmap_len(tx_buffer, len) > 0) {
405 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p",
406 i,
407 le64_to_cpu(u0->a),
408 le64_to_cpu(u0->b),
409 (u64)dma_unmap_addr(tx_buffer, dma),
410 dma_unmap_len(tx_buffer, len),
411 tx_buffer->next_to_watch,
412 (u64)tx_buffer->time_stamp,
413 tx_buffer->skb);
414 if (i == tx_ring->next_to_use &&
415 i == tx_ring->next_to_clean)
416 pr_cont(" NTC/U\n");
417 else if (i == tx_ring->next_to_use)
418 pr_cont(" NTU\n");
419 else if (i == tx_ring->next_to_clean)
420 pr_cont(" NTC\n");
421 else
422 pr_cont("\n");
423
424 if (netif_msg_pktdata(adapter) &&
425 tx_buffer->skb)
426 print_hex_dump(KERN_INFO, "",
427 DUMP_PREFIX_ADDRESS, 16, 1,
428 tx_buffer->skb->data,
429 dma_unmap_len(tx_buffer, len),
430 true);
431 }
432 }
433 }
434
435 /* Print RX Rings Summary */
436 rx_ring_summary:
437 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
438 pr_info("Queue [NTU] [NTC]\n");
439 for (n = 0; n < adapter->num_rx_queues; n++) {
440 rx_ring = adapter->rx_ring[n];
441 pr_info("%5d %5X %5X\n",
442 n, rx_ring->next_to_use, rx_ring->next_to_clean);
443 }
444
445 /* Print RX Rings */
446 if (!netif_msg_rx_status(adapter))
447 goto exit;
448
449 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
450
451 /* Receive Descriptor Formats
452 *
453 * 82598 Advanced Receive Descriptor (Read) Format
454 * 63 1 0
455 * +-----------------------------------------------------+
456 * 0 | Packet Buffer Address [63:1] |A0/NSE|
457 * +----------------------------------------------+------+
458 * 8 | Header Buffer Address [63:1] | DD |
459 * +-----------------------------------------------------+
460 *
461 *
462 * 82598 Advanced Receive Descriptor (Write-Back) Format
463 *
464 * 63 48 47 32 31 30 21 20 16 15 4 3 0
465 * +------------------------------------------------------+
466 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
467 * | Packet | IP | | | | Type | Type |
468 * | Checksum | Ident | | | | | |
469 * +------------------------------------------------------+
470 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
471 * +------------------------------------------------------+
472 * 63 48 47 32 31 20 19 0
473 *
474 * 82599+ Advanced Receive Descriptor (Read) Format
475 * 63 1 0
476 * +-----------------------------------------------------+
477 * 0 | Packet Buffer Address [63:1] |A0/NSE|
478 * +----------------------------------------------+------+
479 * 8 | Header Buffer Address [63:1] | DD |
480 * +-----------------------------------------------------+
481 *
482 *
483 * 82599+ Advanced Receive Descriptor (Write-Back) Format
484 *
485 * 63 48 47 32 31 30 21 20 17 16 4 3 0
486 * +------------------------------------------------------+
487 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
488 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
489 * |/ Flow Dir Flt ID | | | | | |
490 * +------------------------------------------------------+
491 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
492 * +------------------------------------------------------+
493 * 63 48 47 32 31 20 19 0
494 */
495
496 for (n = 0; n < adapter->num_rx_queues; n++) {
497 rx_ring = adapter->rx_ring[n];
498 pr_info("------------------------------------\n");
499 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
500 pr_info("------------------------------------\n");
501 pr_info("%s%s%s",
502 "R [desc] [ PktBuf A0] ",
503 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
504 "<-- Adv Rx Read format\n");
505 pr_info("%s%s%s",
506 "RWB[desc] [PcsmIpSHl PtRs] ",
507 "[vl er S cks ln] ---------------- [bi->skb ] ",
508 "<-- Adv Rx Write-Back format\n");
509
510 for (i = 0; i < rx_ring->count; i++) {
511 rx_buffer_info = &rx_ring->rx_buffer_info[i];
512 rx_desc = IXGBE_RX_DESC(rx_ring, i);
513 u0 = (struct my_u0 *)rx_desc;
514 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
515 if (staterr & IXGBE_RXD_STAT_DD) {
516 /* Descriptor Done */
517 pr_info("RWB[0x%03X] %016llX "
518 "%016llX ---------------- %p", i,
519 le64_to_cpu(u0->a),
520 le64_to_cpu(u0->b),
521 rx_buffer_info->skb);
522 } else {
523 pr_info("R [0x%03X] %016llX "
524 "%016llX %016llX %p", i,
525 le64_to_cpu(u0->a),
526 le64_to_cpu(u0->b),
527 (u64)rx_buffer_info->dma,
528 rx_buffer_info->skb);
529
530 if (netif_msg_pktdata(adapter) &&
531 rx_buffer_info->dma) {
532 print_hex_dump(KERN_INFO, "",
533 DUMP_PREFIX_ADDRESS, 16, 1,
534 page_address(rx_buffer_info->page) +
535 rx_buffer_info->page_offset,
536 ixgbe_rx_bufsz(rx_ring), true);
537 }
538 }
539
540 if (i == rx_ring->next_to_use)
541 pr_cont(" NTU\n");
542 else if (i == rx_ring->next_to_clean)
543 pr_cont(" NTC\n");
544 else
545 pr_cont("\n");
546
547 }
548 }
549
550 exit:
551 return;
552 }
553
554 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
555 {
556 u32 ctrl_ext;
557
558 /* Let firmware take over control of h/w */
559 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
560 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
561 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
562 }
563
564 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
565 {
566 u32 ctrl_ext;
567
568 /* Let firmware know the driver has taken over */
569 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
570 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
571 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
572 }
573
574 /**
575 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
576 * @adapter: pointer to adapter struct
577 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
578 * @queue: queue to map the corresponding interrupt to
579 * @msix_vector: the vector to map to the corresponding queue
580 *
581 */
582 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
583 u8 queue, u8 msix_vector)
584 {
585 u32 ivar, index;
586 struct ixgbe_hw *hw = &adapter->hw;
587 switch (hw->mac.type) {
588 case ixgbe_mac_82598EB:
589 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
590 if (direction == -1)
591 direction = 0;
592 index = (((direction * 64) + queue) >> 2) & 0x1F;
593 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
594 ivar &= ~(0xFF << (8 * (queue & 0x3)));
595 ivar |= (msix_vector << (8 * (queue & 0x3)));
596 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
597 break;
598 case ixgbe_mac_82599EB:
599 case ixgbe_mac_X540:
600 if (direction == -1) {
601 /* other causes */
602 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
603 index = ((queue & 1) * 8);
604 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
605 ivar &= ~(0xFF << index);
606 ivar |= (msix_vector << index);
607 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
608 break;
609 } else {
610 /* tx or rx causes */
611 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
612 index = ((16 * (queue & 1)) + (8 * direction));
613 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
614 ivar &= ~(0xFF << index);
615 ivar |= (msix_vector << index);
616 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
617 break;
618 }
619 default:
620 break;
621 }
622 }
623
624 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
625 u64 qmask)
626 {
627 u32 mask;
628
629 switch (adapter->hw.mac.type) {
630 case ixgbe_mac_82598EB:
631 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
632 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
633 break;
634 case ixgbe_mac_82599EB:
635 case ixgbe_mac_X540:
636 mask = (qmask & 0xFFFFFFFF);
637 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
638 mask = (qmask >> 32);
639 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
640 break;
641 default:
642 break;
643 }
644 }
645
646 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
647 struct ixgbe_tx_buffer *tx_buffer)
648 {
649 if (tx_buffer->skb) {
650 dev_kfree_skb_any(tx_buffer->skb);
651 if (dma_unmap_len(tx_buffer, len))
652 dma_unmap_single(ring->dev,
653 dma_unmap_addr(tx_buffer, dma),
654 dma_unmap_len(tx_buffer, len),
655 DMA_TO_DEVICE);
656 } else if (dma_unmap_len(tx_buffer, len)) {
657 dma_unmap_page(ring->dev,
658 dma_unmap_addr(tx_buffer, dma),
659 dma_unmap_len(tx_buffer, len),
660 DMA_TO_DEVICE);
661 }
662 tx_buffer->next_to_watch = NULL;
663 tx_buffer->skb = NULL;
664 dma_unmap_len_set(tx_buffer, len, 0);
665 /* tx_buffer must be completely set up in the transmit path */
666 }
667
668 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
669 {
670 struct ixgbe_hw *hw = &adapter->hw;
671 struct ixgbe_hw_stats *hwstats = &adapter->stats;
672 int i;
673 u32 data;
674
675 if ((hw->fc.current_mode != ixgbe_fc_full) &&
676 (hw->fc.current_mode != ixgbe_fc_rx_pause))
677 return;
678
679 switch (hw->mac.type) {
680 case ixgbe_mac_82598EB:
681 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
682 break;
683 default:
684 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
685 }
686 hwstats->lxoffrxc += data;
687
688 /* refill credits (no tx hang) if we received xoff */
689 if (!data)
690 return;
691
692 for (i = 0; i < adapter->num_tx_queues; i++)
693 clear_bit(__IXGBE_HANG_CHECK_ARMED,
694 &adapter->tx_ring[i]->state);
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 xoff[8] = {0};
702 u8 tc;
703 int i;
704 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
705
706 if (adapter->ixgbe_ieee_pfc)
707 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
708
709 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
710 ixgbe_update_xoff_rx_lfc(adapter);
711 return;
712 }
713
714 /* update stats for each tc, only valid with PFC enabled */
715 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
716 u32 pxoffrxc;
717
718 switch (hw->mac.type) {
719 case ixgbe_mac_82598EB:
720 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
721 break;
722 default:
723 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
724 }
725 hwstats->pxoffrxc[i] += pxoffrxc;
726 /* Get the TC for given UP */
727 tc = netdev_get_prio_tc_map(adapter->netdev, i);
728 xoff[tc] += pxoffrxc;
729 }
730
731 /* disarm tx queues that have received xoff frames */
732 for (i = 0; i < adapter->num_tx_queues; i++) {
733 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
734
735 tc = tx_ring->dcb_tc;
736 if (xoff[tc])
737 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
738 }
739 }
740
741 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
742 {
743 return ring->stats.packets;
744 }
745
746 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
747 {
748 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
749 struct ixgbe_hw *hw = &adapter->hw;
750
751 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
752 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
753
754 if (head != tail)
755 return (head < tail) ?
756 tail - head : (tail + ring->count - head);
757
758 return 0;
759 }
760
761 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
762 {
763 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
764 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
765 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
766 bool ret = false;
767
768 clear_check_for_tx_hang(tx_ring);
769
770 /*
771 * Check for a hung queue, but be thorough. This verifies
772 * that a transmit has been completed since the previous
773 * check AND there is at least one packet pending. The
774 * ARMED bit is set to indicate a potential hang. The
775 * bit is cleared if a pause frame is received to remove
776 * false hang detection due to PFC or 802.3x frames. By
777 * requiring this to fail twice we avoid races with
778 * pfc clearing the ARMED bit and conditions where we
779 * run the check_tx_hang logic with a transmit completion
780 * pending but without time to complete it yet.
781 */
782 if ((tx_done_old == tx_done) && tx_pending) {
783 /* make sure it is true for two checks in a row */
784 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
785 &tx_ring->state);
786 } else {
787 /* update completed stats and continue */
788 tx_ring->tx_stats.tx_done_old = tx_done;
789 /* reset the countdown */
790 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
791 }
792
793 return ret;
794 }
795
796 /**
797 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
798 * @adapter: driver private struct
799 **/
800 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
801 {
802
803 /* Do the reset outside of interrupt context */
804 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
805 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
806 ixgbe_service_event_schedule(adapter);
807 }
808 }
809
810 /**
811 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
812 * @q_vector: structure containing interrupt and ring information
813 * @tx_ring: tx ring to clean
814 **/
815 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
816 struct ixgbe_ring *tx_ring)
817 {
818 struct ixgbe_adapter *adapter = q_vector->adapter;
819 struct ixgbe_tx_buffer *tx_buffer;
820 union ixgbe_adv_tx_desc *tx_desc;
821 unsigned int total_bytes = 0, total_packets = 0;
822 unsigned int budget = q_vector->tx.work_limit;
823 unsigned int i = tx_ring->next_to_clean;
824
825 if (test_bit(__IXGBE_DOWN, &adapter->state))
826 return true;
827
828 tx_buffer = &tx_ring->tx_buffer_info[i];
829 tx_desc = IXGBE_TX_DESC(tx_ring, i);
830 i -= tx_ring->count;
831
832 do {
833 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
834
835 /* if next_to_watch is not set then there is no work pending */
836 if (!eop_desc)
837 break;
838
839 /* prevent any other reads prior to eop_desc */
840 rmb();
841
842 /* if DD is not set pending work has not been completed */
843 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
844 break;
845
846 /* clear next_to_watch to prevent false hangs */
847 tx_buffer->next_to_watch = NULL;
848
849 /* update the statistics for this packet */
850 total_bytes += tx_buffer->bytecount;
851 total_packets += tx_buffer->gso_segs;
852
853 if (unlikely(tx_buffer->tx_flags & IXGBE_TX_FLAGS_TSTAMP))
854 ixgbe_ptp_tx_hwtstamp(q_vector, tx_buffer->skb);
855
856 /* free the skb */
857 dev_kfree_skb_any(tx_buffer->skb);
858
859 /* unmap skb header data */
860 dma_unmap_single(tx_ring->dev,
861 dma_unmap_addr(tx_buffer, dma),
862 dma_unmap_len(tx_buffer, len),
863 DMA_TO_DEVICE);
864
865 /* clear tx_buffer data */
866 tx_buffer->skb = NULL;
867 dma_unmap_len_set(tx_buffer, len, 0);
868
869 /* unmap remaining buffers */
870 while (tx_desc != eop_desc) {
871 tx_buffer++;
872 tx_desc++;
873 i++;
874 if (unlikely(!i)) {
875 i -= tx_ring->count;
876 tx_buffer = tx_ring->tx_buffer_info;
877 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
878 }
879
880 /* unmap any remaining paged data */
881 if (dma_unmap_len(tx_buffer, len)) {
882 dma_unmap_page(tx_ring->dev,
883 dma_unmap_addr(tx_buffer, dma),
884 dma_unmap_len(tx_buffer, len),
885 DMA_TO_DEVICE);
886 dma_unmap_len_set(tx_buffer, len, 0);
887 }
888 }
889
890 /* move us one more past the eop_desc for start of next pkt */
891 tx_buffer++;
892 tx_desc++;
893 i++;
894 if (unlikely(!i)) {
895 i -= tx_ring->count;
896 tx_buffer = tx_ring->tx_buffer_info;
897 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
898 }
899
900 /* issue prefetch for next Tx descriptor */
901 prefetch(tx_desc);
902
903 /* update budget accounting */
904 budget--;
905 } while (likely(budget));
906
907 i += tx_ring->count;
908 tx_ring->next_to_clean = i;
909 u64_stats_update_begin(&tx_ring->syncp);
910 tx_ring->stats.bytes += total_bytes;
911 tx_ring->stats.packets += total_packets;
912 u64_stats_update_end(&tx_ring->syncp);
913 q_vector->tx.total_bytes += total_bytes;
914 q_vector->tx.total_packets += total_packets;
915
916 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
917 /* schedule immediate reset if we believe we hung */
918 struct ixgbe_hw *hw = &adapter->hw;
919 e_err(drv, "Detected Tx Unit Hang\n"
920 " Tx Queue <%d>\n"
921 " TDH, TDT <%x>, <%x>\n"
922 " next_to_use <%x>\n"
923 " next_to_clean <%x>\n"
924 "tx_buffer_info[next_to_clean]\n"
925 " time_stamp <%lx>\n"
926 " jiffies <%lx>\n",
927 tx_ring->queue_index,
928 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
929 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
930 tx_ring->next_to_use, i,
931 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
932
933 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
934
935 e_info(probe,
936 "tx hang %d detected on queue %d, resetting adapter\n",
937 adapter->tx_timeout_count + 1, tx_ring->queue_index);
938
939 /* schedule immediate reset if we believe we hung */
940 ixgbe_tx_timeout_reset(adapter);
941
942 /* the adapter is about to reset, no point in enabling stuff */
943 return true;
944 }
945
946 netdev_tx_completed_queue(txring_txq(tx_ring),
947 total_packets, total_bytes);
948
949 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
950 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
951 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
952 /* Make sure that anybody stopping the queue after this
953 * sees the new next_to_clean.
954 */
955 smp_mb();
956 if (__netif_subqueue_stopped(tx_ring->netdev,
957 tx_ring->queue_index)
958 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
959 netif_wake_subqueue(tx_ring->netdev,
960 tx_ring->queue_index);
961 ++tx_ring->tx_stats.restart_queue;
962 }
963 }
964
965 return !!budget;
966 }
967
968 #ifdef CONFIG_IXGBE_DCA
969 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
970 struct ixgbe_ring *tx_ring,
971 int cpu)
972 {
973 struct ixgbe_hw *hw = &adapter->hw;
974 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
975 u16 reg_offset;
976
977 switch (hw->mac.type) {
978 case ixgbe_mac_82598EB:
979 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
980 break;
981 case ixgbe_mac_82599EB:
982 case ixgbe_mac_X540:
983 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
984 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
985 break;
986 default:
987 /* for unknown hardware do not write register */
988 return;
989 }
990
991 /*
992 * We can enable relaxed ordering for reads, but not writes when
993 * DCA is enabled. This is due to a known issue in some chipsets
994 * which will cause the DCA tag to be cleared.
995 */
996 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
997 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
998 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
999
1000 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1001 }
1002
1003 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1004 struct ixgbe_ring *rx_ring,
1005 int cpu)
1006 {
1007 struct ixgbe_hw *hw = &adapter->hw;
1008 u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1009 u8 reg_idx = rx_ring->reg_idx;
1010
1011
1012 switch (hw->mac.type) {
1013 case ixgbe_mac_82599EB:
1014 case ixgbe_mac_X540:
1015 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1016 break;
1017 default:
1018 break;
1019 }
1020
1021 /*
1022 * We can enable relaxed ordering for reads, but not writes when
1023 * DCA is enabled. This is due to a known issue in some chipsets
1024 * which will cause the DCA tag to be cleared.
1025 */
1026 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1027 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1028
1029 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1030 }
1031
1032 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1033 {
1034 struct ixgbe_adapter *adapter = q_vector->adapter;
1035 struct ixgbe_ring *ring;
1036 int cpu = get_cpu();
1037
1038 if (q_vector->cpu == cpu)
1039 goto out_no_update;
1040
1041 ixgbe_for_each_ring(ring, q_vector->tx)
1042 ixgbe_update_tx_dca(adapter, ring, cpu);
1043
1044 ixgbe_for_each_ring(ring, q_vector->rx)
1045 ixgbe_update_rx_dca(adapter, ring, cpu);
1046
1047 q_vector->cpu = cpu;
1048 out_no_update:
1049 put_cpu();
1050 }
1051
1052 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1053 {
1054 int i;
1055
1056 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1057 return;
1058
1059 /* always use CB2 mode, difference is masked in the CB driver */
1060 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1061
1062 for (i = 0; i < adapter->num_q_vectors; i++) {
1063 adapter->q_vector[i]->cpu = -1;
1064 ixgbe_update_dca(adapter->q_vector[i]);
1065 }
1066 }
1067
1068 static int __ixgbe_notify_dca(struct device *dev, void *data)
1069 {
1070 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1071 unsigned long event = *(unsigned long *)data;
1072
1073 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1074 return 0;
1075
1076 switch (event) {
1077 case DCA_PROVIDER_ADD:
1078 /* if we're already enabled, don't do it again */
1079 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1080 break;
1081 if (dca_add_requester(dev) == 0) {
1082 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1083 ixgbe_setup_dca(adapter);
1084 break;
1085 }
1086 /* Fall Through since DCA is disabled. */
1087 case DCA_PROVIDER_REMOVE:
1088 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1089 dca_remove_requester(dev);
1090 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1091 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1092 }
1093 break;
1094 }
1095
1096 return 0;
1097 }
1098
1099 #endif /* CONFIG_IXGBE_DCA */
1100 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1101 union ixgbe_adv_rx_desc *rx_desc,
1102 struct sk_buff *skb)
1103 {
1104 if (ring->netdev->features & NETIF_F_RXHASH)
1105 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1106 }
1107
1108 #ifdef IXGBE_FCOE
1109 /**
1110 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1111 * @ring: structure containing ring specific data
1112 * @rx_desc: advanced rx descriptor
1113 *
1114 * Returns : true if it is FCoE pkt
1115 */
1116 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1117 union ixgbe_adv_rx_desc *rx_desc)
1118 {
1119 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1120
1121 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1122 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1123 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1124 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1125 }
1126
1127 #endif /* IXGBE_FCOE */
1128 /**
1129 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1130 * @ring: structure containing ring specific data
1131 * @rx_desc: current Rx descriptor being processed
1132 * @skb: skb currently being received and modified
1133 **/
1134 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1135 union ixgbe_adv_rx_desc *rx_desc,
1136 struct sk_buff *skb)
1137 {
1138 skb_checksum_none_assert(skb);
1139
1140 /* Rx csum disabled */
1141 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1142 return;
1143
1144 /* if IP and error */
1145 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1146 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1147 ring->rx_stats.csum_err++;
1148 return;
1149 }
1150
1151 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1152 return;
1153
1154 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1155 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1156
1157 /*
1158 * 82599 errata, UDP frames with a 0 checksum can be marked as
1159 * checksum errors.
1160 */
1161 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1162 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1163 return;
1164
1165 ring->rx_stats.csum_err++;
1166 return;
1167 }
1168
1169 /* It must be a TCP or UDP packet with a valid checksum */
1170 skb->ip_summed = CHECKSUM_UNNECESSARY;
1171 }
1172
1173 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1174 {
1175 rx_ring->next_to_use = val;
1176
1177 /* update next to alloc since we have filled the ring */
1178 rx_ring->next_to_alloc = val;
1179 /*
1180 * Force memory writes to complete before letting h/w
1181 * know there are new descriptors to fetch. (Only
1182 * applicable for weak-ordered memory model archs,
1183 * such as IA-64).
1184 */
1185 wmb();
1186 writel(val, rx_ring->tail);
1187 }
1188
1189 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1190 struct ixgbe_rx_buffer *bi)
1191 {
1192 struct page *page = bi->page;
1193 dma_addr_t dma = bi->dma;
1194
1195 /* since we are recycling buffers we should seldom need to alloc */
1196 if (likely(dma))
1197 return true;
1198
1199 /* alloc new page for storage */
1200 if (likely(!page)) {
1201 page = __skb_alloc_pages(GFP_ATOMIC | __GFP_COLD | __GFP_COMP,
1202 bi->skb, ixgbe_rx_pg_order(rx_ring));
1203 if (unlikely(!page)) {
1204 rx_ring->rx_stats.alloc_rx_page_failed++;
1205 return false;
1206 }
1207 bi->page = page;
1208 }
1209
1210 /* map page for use */
1211 dma = dma_map_page(rx_ring->dev, page, 0,
1212 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1213
1214 /*
1215 * if mapping failed free memory back to system since
1216 * there isn't much point in holding memory we can't use
1217 */
1218 if (dma_mapping_error(rx_ring->dev, dma)) {
1219 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1220 bi->page = NULL;
1221
1222 rx_ring->rx_stats.alloc_rx_page_failed++;
1223 return false;
1224 }
1225
1226 bi->dma = dma;
1227 bi->page_offset = 0;
1228
1229 return true;
1230 }
1231
1232 /**
1233 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1234 * @rx_ring: ring to place buffers on
1235 * @cleaned_count: number of buffers to replace
1236 **/
1237 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1238 {
1239 union ixgbe_adv_rx_desc *rx_desc;
1240 struct ixgbe_rx_buffer *bi;
1241 u16 i = rx_ring->next_to_use;
1242
1243 /* nothing to do */
1244 if (!cleaned_count)
1245 return;
1246
1247 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1248 bi = &rx_ring->rx_buffer_info[i];
1249 i -= rx_ring->count;
1250
1251 do {
1252 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1253 break;
1254
1255 /*
1256 * Refresh the desc even if buffer_addrs didn't change
1257 * because each write-back erases this info.
1258 */
1259 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1260
1261 rx_desc++;
1262 bi++;
1263 i++;
1264 if (unlikely(!i)) {
1265 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1266 bi = rx_ring->rx_buffer_info;
1267 i -= rx_ring->count;
1268 }
1269
1270 /* clear the hdr_addr for the next_to_use descriptor */
1271 rx_desc->read.hdr_addr = 0;
1272
1273 cleaned_count--;
1274 } while (cleaned_count);
1275
1276 i += rx_ring->count;
1277
1278 if (rx_ring->next_to_use != i)
1279 ixgbe_release_rx_desc(rx_ring, i);
1280 }
1281
1282 /**
1283 * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1284 * @data: pointer to the start of the headers
1285 * @max_len: total length of section to find headers in
1286 *
1287 * This function is meant to determine the length of headers that will
1288 * be recognized by hardware for LRO, GRO, and RSC offloads. The main
1289 * motivation of doing this is to only perform one pull for IPv4 TCP
1290 * packets so that we can do basic things like calculating the gso_size
1291 * based on the average data per packet.
1292 **/
1293 static unsigned int ixgbe_get_headlen(unsigned char *data,
1294 unsigned int max_len)
1295 {
1296 union {
1297 unsigned char *network;
1298 /* l2 headers */
1299 struct ethhdr *eth;
1300 struct vlan_hdr *vlan;
1301 /* l3 headers */
1302 struct iphdr *ipv4;
1303 struct ipv6hdr *ipv6;
1304 } hdr;
1305 __be16 protocol;
1306 u8 nexthdr = 0; /* default to not TCP */
1307 u8 hlen;
1308
1309 /* this should never happen, but better safe than sorry */
1310 if (max_len < ETH_HLEN)
1311 return max_len;
1312
1313 /* initialize network frame pointer */
1314 hdr.network = data;
1315
1316 /* set first protocol and move network header forward */
1317 protocol = hdr.eth->h_proto;
1318 hdr.network += ETH_HLEN;
1319
1320 /* handle any vlan tag if present */
1321 if (protocol == __constant_htons(ETH_P_8021Q)) {
1322 if ((hdr.network - data) > (max_len - VLAN_HLEN))
1323 return max_len;
1324
1325 protocol = hdr.vlan->h_vlan_encapsulated_proto;
1326 hdr.network += VLAN_HLEN;
1327 }
1328
1329 /* handle L3 protocols */
1330 if (protocol == __constant_htons(ETH_P_IP)) {
1331 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1332 return max_len;
1333
1334 /* access ihl as a u8 to avoid unaligned access on ia64 */
1335 hlen = (hdr.network[0] & 0x0F) << 2;
1336
1337 /* verify hlen meets minimum size requirements */
1338 if (hlen < sizeof(struct iphdr))
1339 return hdr.network - data;
1340
1341 /* record next protocol */
1342 nexthdr = hdr.ipv4->protocol;
1343 hdr.network += hlen;
1344 } else if (protocol == __constant_htons(ETH_P_IPV6)) {
1345 if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
1346 return max_len;
1347
1348 /* record next protocol */
1349 nexthdr = hdr.ipv6->nexthdr;
1350 hdr.network += sizeof(struct ipv6hdr);
1351 #ifdef IXGBE_FCOE
1352 } else if (protocol == __constant_htons(ETH_P_FCOE)) {
1353 if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1354 return max_len;
1355 hdr.network += FCOE_HEADER_LEN;
1356 #endif
1357 } else {
1358 return hdr.network - data;
1359 }
1360
1361 /* finally sort out TCP/UDP */
1362 if (nexthdr == IPPROTO_TCP) {
1363 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1364 return max_len;
1365
1366 /* access doff as a u8 to avoid unaligned access on ia64 */
1367 hlen = (hdr.network[12] & 0xF0) >> 2;
1368
1369 /* verify hlen meets minimum size requirements */
1370 if (hlen < sizeof(struct tcphdr))
1371 return hdr.network - data;
1372
1373 hdr.network += hlen;
1374 } else if (nexthdr == IPPROTO_UDP) {
1375 if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
1376 return max_len;
1377
1378 hdr.network += sizeof(struct udphdr);
1379 }
1380
1381 /*
1382 * If everything has gone correctly hdr.network should be the
1383 * data section of the packet and will be the end of the header.
1384 * If not then it probably represents the end of the last recognized
1385 * header.
1386 */
1387 if ((hdr.network - data) < max_len)
1388 return hdr.network - data;
1389 else
1390 return max_len;
1391 }
1392
1393 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1394 struct sk_buff *skb)
1395 {
1396 u16 hdr_len = skb_headlen(skb);
1397
1398 /* set gso_size to avoid messing up TCP MSS */
1399 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1400 IXGBE_CB(skb)->append_cnt);
1401 }
1402
1403 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1404 struct sk_buff *skb)
1405 {
1406 /* if append_cnt is 0 then frame is not RSC */
1407 if (!IXGBE_CB(skb)->append_cnt)
1408 return;
1409
1410 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1411 rx_ring->rx_stats.rsc_flush++;
1412
1413 ixgbe_set_rsc_gso_size(rx_ring, skb);
1414
1415 /* gso_size is computed using append_cnt so always clear it last */
1416 IXGBE_CB(skb)->append_cnt = 0;
1417 }
1418
1419 /**
1420 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1421 * @rx_ring: rx descriptor ring packet is being transacted on
1422 * @rx_desc: pointer to the EOP Rx descriptor
1423 * @skb: pointer to current skb being populated
1424 *
1425 * This function checks the ring, descriptor, and packet information in
1426 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1427 * other fields within the skb.
1428 **/
1429 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1430 union ixgbe_adv_rx_desc *rx_desc,
1431 struct sk_buff *skb)
1432 {
1433 struct net_device *dev = rx_ring->netdev;
1434
1435 ixgbe_update_rsc_stats(rx_ring, skb);
1436
1437 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1438
1439 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1440
1441 ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector, rx_desc, skb);
1442
1443 if ((dev->features & NETIF_F_HW_VLAN_RX) &&
1444 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1445 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1446 __vlan_hwaccel_put_tag(skb, vid);
1447 }
1448
1449 skb_record_rx_queue(skb, rx_ring->queue_index);
1450
1451 skb->protocol = eth_type_trans(skb, dev);
1452 }
1453
1454 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1455 struct sk_buff *skb)
1456 {
1457 struct ixgbe_adapter *adapter = q_vector->adapter;
1458
1459 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1460 napi_gro_receive(&q_vector->napi, skb);
1461 else
1462 netif_rx(skb);
1463 }
1464
1465 /**
1466 * ixgbe_is_non_eop - process handling of non-EOP buffers
1467 * @rx_ring: Rx ring being processed
1468 * @rx_desc: Rx descriptor for current buffer
1469 * @skb: Current socket buffer containing buffer in progress
1470 *
1471 * This function updates next to clean. If the buffer is an EOP buffer
1472 * this function exits returning false, otherwise it will place the
1473 * sk_buff in the next buffer to be chained and return true indicating
1474 * that this is in fact a non-EOP buffer.
1475 **/
1476 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1477 union ixgbe_adv_rx_desc *rx_desc,
1478 struct sk_buff *skb)
1479 {
1480 u32 ntc = rx_ring->next_to_clean + 1;
1481
1482 /* fetch, update, and store next to clean */
1483 ntc = (ntc < rx_ring->count) ? ntc : 0;
1484 rx_ring->next_to_clean = ntc;
1485
1486 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1487
1488 /* update RSC append count if present */
1489 if (ring_is_rsc_enabled(rx_ring)) {
1490 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1491 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1492
1493 if (unlikely(rsc_enabled)) {
1494 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1495
1496 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1497 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1498
1499 /* update ntc based on RSC value */
1500 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1501 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1502 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1503 }
1504 }
1505
1506 /* if we are the last buffer then there is nothing else to do */
1507 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1508 return false;
1509
1510 /* place skb in next buffer to be received */
1511 rx_ring->rx_buffer_info[ntc].skb = skb;
1512 rx_ring->rx_stats.non_eop_descs++;
1513
1514 return true;
1515 }
1516
1517 /**
1518 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1519 * @rx_ring: rx descriptor ring packet is being transacted on
1520 * @skb: pointer to current skb being adjusted
1521 *
1522 * This function is an ixgbe specific version of __pskb_pull_tail. The
1523 * main difference between this version and the original function is that
1524 * this function can make several assumptions about the state of things
1525 * that allow for significant optimizations versus the standard function.
1526 * As a result we can do things like drop a frag and maintain an accurate
1527 * truesize for the skb.
1528 */
1529 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1530 struct sk_buff *skb)
1531 {
1532 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1533 unsigned char *va;
1534 unsigned int pull_len;
1535
1536 /*
1537 * it is valid to use page_address instead of kmap since we are
1538 * working with pages allocated out of the lomem pool per
1539 * alloc_page(GFP_ATOMIC)
1540 */
1541 va = skb_frag_address(frag);
1542
1543 /*
1544 * we need the header to contain the greater of either ETH_HLEN or
1545 * 60 bytes if the skb->len is less than 60 for skb_pad.
1546 */
1547 pull_len = ixgbe_get_headlen(va, IXGBE_RX_HDR_SIZE);
1548
1549 /* align pull length to size of long to optimize memcpy performance */
1550 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1551
1552 /* update all of the pointers */
1553 skb_frag_size_sub(frag, pull_len);
1554 frag->page_offset += pull_len;
1555 skb->data_len -= pull_len;
1556 skb->tail += pull_len;
1557 }
1558
1559 /**
1560 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1561 * @rx_ring: rx descriptor ring packet is being transacted on
1562 * @skb: pointer to current skb being updated
1563 *
1564 * This function provides a basic DMA sync up for the first fragment of an
1565 * skb. The reason for doing this is that the first fragment cannot be
1566 * unmapped until we have reached the end of packet descriptor for a buffer
1567 * chain.
1568 */
1569 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1570 struct sk_buff *skb)
1571 {
1572 /* if the page was released unmap it, else just sync our portion */
1573 if (unlikely(IXGBE_CB(skb)->page_released)) {
1574 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1575 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1576 IXGBE_CB(skb)->page_released = false;
1577 } else {
1578 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1579
1580 dma_sync_single_range_for_cpu(rx_ring->dev,
1581 IXGBE_CB(skb)->dma,
1582 frag->page_offset,
1583 ixgbe_rx_bufsz(rx_ring),
1584 DMA_FROM_DEVICE);
1585 }
1586 IXGBE_CB(skb)->dma = 0;
1587 }
1588
1589 /**
1590 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1591 * @rx_ring: rx descriptor ring packet is being transacted on
1592 * @rx_desc: pointer to the EOP Rx descriptor
1593 * @skb: pointer to current skb being fixed
1594 *
1595 * Check for corrupted packet headers caused by senders on the local L2
1596 * embedded NIC switch not setting up their Tx Descriptors right. These
1597 * should be very rare.
1598 *
1599 * Also address the case where we are pulling data in on pages only
1600 * and as such no data is present in the skb header.
1601 *
1602 * In addition if skb is not at least 60 bytes we need to pad it so that
1603 * it is large enough to qualify as a valid Ethernet frame.
1604 *
1605 * Returns true if an error was encountered and skb was freed.
1606 **/
1607 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1608 union ixgbe_adv_rx_desc *rx_desc,
1609 struct sk_buff *skb)
1610 {
1611 struct net_device *netdev = rx_ring->netdev;
1612
1613 /* verify that the packet does not have any known errors */
1614 if (unlikely(ixgbe_test_staterr(rx_desc,
1615 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1616 !(netdev->features & NETIF_F_RXALL))) {
1617 dev_kfree_skb_any(skb);
1618 return true;
1619 }
1620
1621 /* place header in linear portion of buffer */
1622 if (skb_is_nonlinear(skb))
1623 ixgbe_pull_tail(rx_ring, skb);
1624
1625 #ifdef IXGBE_FCOE
1626 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1627 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1628 return false;
1629
1630 #endif
1631 /* if skb_pad returns an error the skb was freed */
1632 if (unlikely(skb->len < 60)) {
1633 int pad_len = 60 - skb->len;
1634
1635 if (skb_pad(skb, pad_len))
1636 return true;
1637 __skb_put(skb, pad_len);
1638 }
1639
1640 return false;
1641 }
1642
1643 /**
1644 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1645 * @rx_ring: rx descriptor ring to store buffers on
1646 * @old_buff: donor buffer to have page reused
1647 *
1648 * Synchronizes page for reuse by the adapter
1649 **/
1650 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1651 struct ixgbe_rx_buffer *old_buff)
1652 {
1653 struct ixgbe_rx_buffer *new_buff;
1654 u16 nta = rx_ring->next_to_alloc;
1655
1656 new_buff = &rx_ring->rx_buffer_info[nta];
1657
1658 /* update, and store next to alloc */
1659 nta++;
1660 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1661
1662 /* transfer page from old buffer to new buffer */
1663 new_buff->page = old_buff->page;
1664 new_buff->dma = old_buff->dma;
1665 new_buff->page_offset = old_buff->page_offset;
1666
1667 /* sync the buffer for use by the device */
1668 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1669 new_buff->page_offset,
1670 ixgbe_rx_bufsz(rx_ring),
1671 DMA_FROM_DEVICE);
1672 }
1673
1674 /**
1675 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1676 * @rx_ring: rx descriptor ring to transact packets on
1677 * @rx_buffer: buffer containing page to add
1678 * @rx_desc: descriptor containing length of buffer written by hardware
1679 * @skb: sk_buff to place the data into
1680 *
1681 * This function will add the data contained in rx_buffer->page to the skb.
1682 * This is done either through a direct copy if the data in the buffer is
1683 * less than the skb header size, otherwise it will just attach the page as
1684 * a frag to the skb.
1685 *
1686 * The function will then update the page offset if necessary and return
1687 * true if the buffer can be reused by the adapter.
1688 **/
1689 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1690 struct ixgbe_rx_buffer *rx_buffer,
1691 union ixgbe_adv_rx_desc *rx_desc,
1692 struct sk_buff *skb)
1693 {
1694 struct page *page = rx_buffer->page;
1695 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1696 #if (PAGE_SIZE < 8192)
1697 unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1698 #else
1699 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1700 unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1701 ixgbe_rx_bufsz(rx_ring);
1702 #endif
1703
1704 if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1705 unsigned char *va = page_address(page) + rx_buffer->page_offset;
1706
1707 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1708
1709 /* we can reuse buffer as-is, just make sure it is local */
1710 if (likely(page_to_nid(page) == numa_node_id()))
1711 return true;
1712
1713 /* this page cannot be reused so discard it */
1714 put_page(page);
1715 return false;
1716 }
1717
1718 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1719 rx_buffer->page_offset, size, truesize);
1720
1721 /* avoid re-using remote pages */
1722 if (unlikely(page_to_nid(page) != numa_node_id()))
1723 return false;
1724
1725 #if (PAGE_SIZE < 8192)
1726 /* if we are only owner of page we can reuse it */
1727 if (unlikely(page_count(page) != 1))
1728 return false;
1729
1730 /* flip page offset to other buffer */
1731 rx_buffer->page_offset ^= truesize;
1732
1733 /*
1734 * since we are the only owner of the page and we need to
1735 * increment it, just set the value to 2 in order to avoid
1736 * an unecessary locked operation
1737 */
1738 atomic_set(&page->_count, 2);
1739 #else
1740 /* move offset up to the next cache line */
1741 rx_buffer->page_offset += truesize;
1742
1743 if (rx_buffer->page_offset > last_offset)
1744 return false;
1745
1746 /* bump ref count on page before it is given to the stack */
1747 get_page(page);
1748 #endif
1749
1750 return true;
1751 }
1752
1753 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1754 union ixgbe_adv_rx_desc *rx_desc)
1755 {
1756 struct ixgbe_rx_buffer *rx_buffer;
1757 struct sk_buff *skb;
1758 struct page *page;
1759
1760 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1761 page = rx_buffer->page;
1762 prefetchw(page);
1763
1764 skb = rx_buffer->skb;
1765
1766 if (likely(!skb)) {
1767 void *page_addr = page_address(page) +
1768 rx_buffer->page_offset;
1769
1770 /* prefetch first cache line of first page */
1771 prefetch(page_addr);
1772 #if L1_CACHE_BYTES < 128
1773 prefetch(page_addr + L1_CACHE_BYTES);
1774 #endif
1775
1776 /* allocate a skb to store the frags */
1777 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1778 IXGBE_RX_HDR_SIZE);
1779 if (unlikely(!skb)) {
1780 rx_ring->rx_stats.alloc_rx_buff_failed++;
1781 return NULL;
1782 }
1783
1784 /*
1785 * we will be copying header into skb->data in
1786 * pskb_may_pull so it is in our interest to prefetch
1787 * it now to avoid a possible cache miss
1788 */
1789 prefetchw(skb->data);
1790
1791 /*
1792 * Delay unmapping of the first packet. It carries the
1793 * header information, HW may still access the header
1794 * after the writeback. Only unmap it when EOP is
1795 * reached
1796 */
1797 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1798 goto dma_sync;
1799
1800 IXGBE_CB(skb)->dma = rx_buffer->dma;
1801 } else {
1802 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
1803 ixgbe_dma_sync_frag(rx_ring, skb);
1804
1805 dma_sync:
1806 /* we are reusing so sync this buffer for CPU use */
1807 dma_sync_single_range_for_cpu(rx_ring->dev,
1808 rx_buffer->dma,
1809 rx_buffer->page_offset,
1810 ixgbe_rx_bufsz(rx_ring),
1811 DMA_FROM_DEVICE);
1812 }
1813
1814 /* pull page into skb */
1815 if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
1816 /* hand second half of page back to the ring */
1817 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1818 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1819 /* the page has been released from the ring */
1820 IXGBE_CB(skb)->page_released = true;
1821 } else {
1822 /* we are not reusing the buffer so unmap it */
1823 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1824 ixgbe_rx_pg_size(rx_ring),
1825 DMA_FROM_DEVICE);
1826 }
1827
1828 /* clear contents of buffer_info */
1829 rx_buffer->skb = NULL;
1830 rx_buffer->dma = 0;
1831 rx_buffer->page = NULL;
1832
1833 return skb;
1834 }
1835
1836 /**
1837 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1838 * @q_vector: structure containing interrupt and ring information
1839 * @rx_ring: rx descriptor ring to transact packets on
1840 * @budget: Total limit on number of packets to process
1841 *
1842 * This function provides a "bounce buffer" approach to Rx interrupt
1843 * processing. The advantage to this is that on systems that have
1844 * expensive overhead for IOMMU access this provides a means of avoiding
1845 * it by maintaining the mapping of the page to the syste.
1846 *
1847 * Returns true if all work is completed without reaching budget
1848 **/
1849 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1850 struct ixgbe_ring *rx_ring,
1851 const int budget)
1852 {
1853 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1854 #ifdef IXGBE_FCOE
1855 struct ixgbe_adapter *adapter = q_vector->adapter;
1856 int ddp_bytes;
1857 unsigned int mss = 0;
1858 #endif /* IXGBE_FCOE */
1859 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1860
1861 do {
1862 union ixgbe_adv_rx_desc *rx_desc;
1863 struct sk_buff *skb;
1864
1865 /* return some buffers to hardware, one at a time is too slow */
1866 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1867 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1868 cleaned_count = 0;
1869 }
1870
1871 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
1872
1873 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
1874 break;
1875
1876 /*
1877 * This memory barrier is needed to keep us from reading
1878 * any other fields out of the rx_desc until we know the
1879 * RXD_STAT_DD bit is set
1880 */
1881 rmb();
1882
1883 /* retrieve a buffer from the ring */
1884 skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
1885
1886 /* exit if we failed to retrieve a buffer */
1887 if (!skb)
1888 break;
1889
1890 cleaned_count++;
1891
1892 /* place incomplete frames back on ring for completion */
1893 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
1894 continue;
1895
1896 /* verify the packet layout is correct */
1897 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
1898 continue;
1899
1900 /* probably a little skewed due to removing CRC */
1901 total_rx_bytes += skb->len;
1902
1903 /* populate checksum, timestamp, VLAN, and protocol */
1904 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
1905
1906 #ifdef IXGBE_FCOE
1907 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1908 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
1909 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1910 /* include DDPed FCoE data */
1911 if (ddp_bytes > 0) {
1912 if (!mss) {
1913 mss = rx_ring->netdev->mtu -
1914 sizeof(struct fcoe_hdr) -
1915 sizeof(struct fc_frame_header) -
1916 sizeof(struct fcoe_crc_eof);
1917 if (mss > 512)
1918 mss &= ~511;
1919 }
1920 total_rx_bytes += ddp_bytes;
1921 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
1922 mss);
1923 }
1924 if (!ddp_bytes) {
1925 dev_kfree_skb_any(skb);
1926 continue;
1927 }
1928 }
1929
1930 #endif /* IXGBE_FCOE */
1931 ixgbe_rx_skb(q_vector, skb);
1932
1933 /* update budget accounting */
1934 total_rx_packets++;
1935 } while (likely(total_rx_packets < budget));
1936
1937 u64_stats_update_begin(&rx_ring->syncp);
1938 rx_ring->stats.packets += total_rx_packets;
1939 rx_ring->stats.bytes += total_rx_bytes;
1940 u64_stats_update_end(&rx_ring->syncp);
1941 q_vector->rx.total_packets += total_rx_packets;
1942 q_vector->rx.total_bytes += total_rx_bytes;
1943
1944 if (cleaned_count)
1945 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1946
1947 return (total_rx_packets < budget);
1948 }
1949
1950 /**
1951 * ixgbe_configure_msix - Configure MSI-X hardware
1952 * @adapter: board private structure
1953 *
1954 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1955 * interrupts.
1956 **/
1957 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1958 {
1959 struct ixgbe_q_vector *q_vector;
1960 int v_idx;
1961 u32 mask;
1962
1963 /* Populate MSIX to EITR Select */
1964 if (adapter->num_vfs > 32) {
1965 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1966 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1967 }
1968
1969 /*
1970 * Populate the IVAR table and set the ITR values to the
1971 * corresponding register.
1972 */
1973 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
1974 struct ixgbe_ring *ring;
1975 q_vector = adapter->q_vector[v_idx];
1976
1977 ixgbe_for_each_ring(ring, q_vector->rx)
1978 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1979
1980 ixgbe_for_each_ring(ring, q_vector->tx)
1981 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1982
1983 ixgbe_write_eitr(q_vector);
1984 }
1985
1986 switch (adapter->hw.mac.type) {
1987 case ixgbe_mac_82598EB:
1988 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1989 v_idx);
1990 break;
1991 case ixgbe_mac_82599EB:
1992 case ixgbe_mac_X540:
1993 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1994 break;
1995 default:
1996 break;
1997 }
1998 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1999
2000 /* set up to autoclear timer, and the vectors */
2001 mask = IXGBE_EIMS_ENABLE_MASK;
2002 mask &= ~(IXGBE_EIMS_OTHER |
2003 IXGBE_EIMS_MAILBOX |
2004 IXGBE_EIMS_LSC);
2005
2006 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2007 }
2008
2009 enum latency_range {
2010 lowest_latency = 0,
2011 low_latency = 1,
2012 bulk_latency = 2,
2013 latency_invalid = 255
2014 };
2015
2016 /**
2017 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2018 * @q_vector: structure containing interrupt and ring information
2019 * @ring_container: structure containing ring performance data
2020 *
2021 * Stores a new ITR value based on packets and byte
2022 * counts during the last interrupt. The advantage of per interrupt
2023 * computation is faster updates and more accurate ITR for the current
2024 * traffic pattern. Constants in this function were computed
2025 * based on theoretical maximum wire speed and thresholds were set based
2026 * on testing data as well as attempting to minimize response time
2027 * while increasing bulk throughput.
2028 * this functionality is controlled by the InterruptThrottleRate module
2029 * parameter (see ixgbe_param.c)
2030 **/
2031 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2032 struct ixgbe_ring_container *ring_container)
2033 {
2034 int bytes = ring_container->total_bytes;
2035 int packets = ring_container->total_packets;
2036 u32 timepassed_us;
2037 u64 bytes_perint;
2038 u8 itr_setting = ring_container->itr;
2039
2040 if (packets == 0)
2041 return;
2042
2043 /* simple throttlerate management
2044 * 0-10MB/s lowest (100000 ints/s)
2045 * 10-20MB/s low (20000 ints/s)
2046 * 20-1249MB/s bulk (8000 ints/s)
2047 */
2048 /* what was last interrupt timeslice? */
2049 timepassed_us = q_vector->itr >> 2;
2050 bytes_perint = bytes / timepassed_us; /* bytes/usec */
2051
2052 switch (itr_setting) {
2053 case lowest_latency:
2054 if (bytes_perint > 10)
2055 itr_setting = low_latency;
2056 break;
2057 case low_latency:
2058 if (bytes_perint > 20)
2059 itr_setting = bulk_latency;
2060 else if (bytes_perint <= 10)
2061 itr_setting = lowest_latency;
2062 break;
2063 case bulk_latency:
2064 if (bytes_perint <= 20)
2065 itr_setting = low_latency;
2066 break;
2067 }
2068
2069 /* clear work counters since we have the values we need */
2070 ring_container->total_bytes = 0;
2071 ring_container->total_packets = 0;
2072
2073 /* write updated itr to ring container */
2074 ring_container->itr = itr_setting;
2075 }
2076
2077 /**
2078 * ixgbe_write_eitr - write EITR register in hardware specific way
2079 * @q_vector: structure containing interrupt and ring information
2080 *
2081 * This function is made to be called by ethtool and by the driver
2082 * when it needs to update EITR registers at runtime. Hardware
2083 * specific quirks/differences are taken care of here.
2084 */
2085 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2086 {
2087 struct ixgbe_adapter *adapter = q_vector->adapter;
2088 struct ixgbe_hw *hw = &adapter->hw;
2089 int v_idx = q_vector->v_idx;
2090 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2091
2092 switch (adapter->hw.mac.type) {
2093 case ixgbe_mac_82598EB:
2094 /* must write high and low 16 bits to reset counter */
2095 itr_reg |= (itr_reg << 16);
2096 break;
2097 case ixgbe_mac_82599EB:
2098 case ixgbe_mac_X540:
2099 /*
2100 * set the WDIS bit to not clear the timer bits and cause an
2101 * immediate assertion of the interrupt
2102 */
2103 itr_reg |= IXGBE_EITR_CNT_WDIS;
2104 break;
2105 default:
2106 break;
2107 }
2108 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2109 }
2110
2111 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2112 {
2113 u32 new_itr = q_vector->itr;
2114 u8 current_itr;
2115
2116 ixgbe_update_itr(q_vector, &q_vector->tx);
2117 ixgbe_update_itr(q_vector, &q_vector->rx);
2118
2119 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2120
2121 switch (current_itr) {
2122 /* counts and packets in update_itr are dependent on these numbers */
2123 case lowest_latency:
2124 new_itr = IXGBE_100K_ITR;
2125 break;
2126 case low_latency:
2127 new_itr = IXGBE_20K_ITR;
2128 break;
2129 case bulk_latency:
2130 new_itr = IXGBE_8K_ITR;
2131 break;
2132 default:
2133 break;
2134 }
2135
2136 if (new_itr != q_vector->itr) {
2137 /* do an exponential smoothing */
2138 new_itr = (10 * new_itr * q_vector->itr) /
2139 ((9 * new_itr) + q_vector->itr);
2140
2141 /* save the algorithm value here */
2142 q_vector->itr = new_itr;
2143
2144 ixgbe_write_eitr(q_vector);
2145 }
2146 }
2147
2148 /**
2149 * ixgbe_check_overtemp_subtask - check for over temperature
2150 * @adapter: pointer to adapter
2151 **/
2152 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2153 {
2154 struct ixgbe_hw *hw = &adapter->hw;
2155 u32 eicr = adapter->interrupt_event;
2156
2157 if (test_bit(__IXGBE_DOWN, &adapter->state))
2158 return;
2159
2160 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2161 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2162 return;
2163
2164 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2165
2166 switch (hw->device_id) {
2167 case IXGBE_DEV_ID_82599_T3_LOM:
2168 /*
2169 * Since the warning interrupt is for both ports
2170 * we don't have to check if:
2171 * - This interrupt wasn't for our port.
2172 * - We may have missed the interrupt so always have to
2173 * check if we got a LSC
2174 */
2175 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2176 !(eicr & IXGBE_EICR_LSC))
2177 return;
2178
2179 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2180 u32 autoneg;
2181 bool link_up = false;
2182
2183 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2184
2185 if (link_up)
2186 return;
2187 }
2188
2189 /* Check if this is not due to overtemp */
2190 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2191 return;
2192
2193 break;
2194 default:
2195 if (!(eicr & IXGBE_EICR_GPI_SDP0))
2196 return;
2197 break;
2198 }
2199 e_crit(drv,
2200 "Network adapter has been stopped because it has over heated. "
2201 "Restart the computer. If the problem persists, "
2202 "power off the system and replace the adapter\n");
2203
2204 adapter->interrupt_event = 0;
2205 }
2206
2207 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2208 {
2209 struct ixgbe_hw *hw = &adapter->hw;
2210
2211 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2212 (eicr & IXGBE_EICR_GPI_SDP1)) {
2213 e_crit(probe, "Fan has stopped, replace the adapter\n");
2214 /* write to clear the interrupt */
2215 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2216 }
2217 }
2218
2219 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2220 {
2221 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2222 return;
2223
2224 switch (adapter->hw.mac.type) {
2225 case ixgbe_mac_82599EB:
2226 /*
2227 * Need to check link state so complete overtemp check
2228 * on service task
2229 */
2230 if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2231 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2232 adapter->interrupt_event = eicr;
2233 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2234 ixgbe_service_event_schedule(adapter);
2235 return;
2236 }
2237 return;
2238 case ixgbe_mac_X540:
2239 if (!(eicr & IXGBE_EICR_TS))
2240 return;
2241 break;
2242 default:
2243 return;
2244 }
2245
2246 e_crit(drv,
2247 "Network adapter has been stopped because it has over heated. "
2248 "Restart the computer. If the problem persists, "
2249 "power off the system and replace the adapter\n");
2250 }
2251
2252 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2253 {
2254 struct ixgbe_hw *hw = &adapter->hw;
2255
2256 if (eicr & IXGBE_EICR_GPI_SDP2) {
2257 /* Clear the interrupt */
2258 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2259 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2260 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2261 ixgbe_service_event_schedule(adapter);
2262 }
2263 }
2264
2265 if (eicr & IXGBE_EICR_GPI_SDP1) {
2266 /* Clear the interrupt */
2267 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2268 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2269 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2270 ixgbe_service_event_schedule(adapter);
2271 }
2272 }
2273 }
2274
2275 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2276 {
2277 struct ixgbe_hw *hw = &adapter->hw;
2278
2279 adapter->lsc_int++;
2280 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2281 adapter->link_check_timeout = jiffies;
2282 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2283 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2284 IXGBE_WRITE_FLUSH(hw);
2285 ixgbe_service_event_schedule(adapter);
2286 }
2287 }
2288
2289 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2290 u64 qmask)
2291 {
2292 u32 mask;
2293 struct ixgbe_hw *hw = &adapter->hw;
2294
2295 switch (hw->mac.type) {
2296 case ixgbe_mac_82598EB:
2297 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2298 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2299 break;
2300 case ixgbe_mac_82599EB:
2301 case ixgbe_mac_X540:
2302 mask = (qmask & 0xFFFFFFFF);
2303 if (mask)
2304 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2305 mask = (qmask >> 32);
2306 if (mask)
2307 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2308 break;
2309 default:
2310 break;
2311 }
2312 /* skip the flush */
2313 }
2314
2315 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2316 u64 qmask)
2317 {
2318 u32 mask;
2319 struct ixgbe_hw *hw = &adapter->hw;
2320
2321 switch (hw->mac.type) {
2322 case ixgbe_mac_82598EB:
2323 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2324 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2325 break;
2326 case ixgbe_mac_82599EB:
2327 case ixgbe_mac_X540:
2328 mask = (qmask & 0xFFFFFFFF);
2329 if (mask)
2330 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2331 mask = (qmask >> 32);
2332 if (mask)
2333 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2334 break;
2335 default:
2336 break;
2337 }
2338 /* skip the flush */
2339 }
2340
2341 /**
2342 * ixgbe_irq_enable - Enable default interrupt generation settings
2343 * @adapter: board private structure
2344 **/
2345 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2346 bool flush)
2347 {
2348 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2349
2350 /* don't reenable LSC while waiting for link */
2351 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2352 mask &= ~IXGBE_EIMS_LSC;
2353
2354 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2355 switch (adapter->hw.mac.type) {
2356 case ixgbe_mac_82599EB:
2357 mask |= IXGBE_EIMS_GPI_SDP0;
2358 break;
2359 case ixgbe_mac_X540:
2360 mask |= IXGBE_EIMS_TS;
2361 break;
2362 default:
2363 break;
2364 }
2365 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2366 mask |= IXGBE_EIMS_GPI_SDP1;
2367 switch (adapter->hw.mac.type) {
2368 case ixgbe_mac_82599EB:
2369 mask |= IXGBE_EIMS_GPI_SDP1;
2370 mask |= IXGBE_EIMS_GPI_SDP2;
2371 case ixgbe_mac_X540:
2372 mask |= IXGBE_EIMS_ECC;
2373 mask |= IXGBE_EIMS_MAILBOX;
2374 break;
2375 default:
2376 break;
2377 }
2378
2379 if (adapter->hw.mac.type == ixgbe_mac_X540)
2380 mask |= IXGBE_EIMS_TIMESYNC;
2381
2382 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2383 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2384 mask |= IXGBE_EIMS_FLOW_DIR;
2385
2386 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2387 if (queues)
2388 ixgbe_irq_enable_queues(adapter, ~0);
2389 if (flush)
2390 IXGBE_WRITE_FLUSH(&adapter->hw);
2391 }
2392
2393 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2394 {
2395 struct ixgbe_adapter *adapter = data;
2396 struct ixgbe_hw *hw = &adapter->hw;
2397 u32 eicr;
2398
2399 /*
2400 * Workaround for Silicon errata. Use clear-by-write instead
2401 * of clear-by-read. Reading with EICS will return the
2402 * interrupt causes without clearing, which later be done
2403 * with the write to EICR.
2404 */
2405 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2406 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2407
2408 if (eicr & IXGBE_EICR_LSC)
2409 ixgbe_check_lsc(adapter);
2410
2411 if (eicr & IXGBE_EICR_MAILBOX)
2412 ixgbe_msg_task(adapter);
2413
2414 switch (hw->mac.type) {
2415 case ixgbe_mac_82599EB:
2416 case ixgbe_mac_X540:
2417 if (eicr & IXGBE_EICR_ECC)
2418 e_info(link, "Received unrecoverable ECC Err, please "
2419 "reboot\n");
2420 /* Handle Flow Director Full threshold interrupt */
2421 if (eicr & IXGBE_EICR_FLOW_DIR) {
2422 int reinit_count = 0;
2423 int i;
2424 for (i = 0; i < adapter->num_tx_queues; i++) {
2425 struct ixgbe_ring *ring = adapter->tx_ring[i];
2426 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2427 &ring->state))
2428 reinit_count++;
2429 }
2430 if (reinit_count) {
2431 /* no more flow director interrupts until after init */
2432 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2433 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2434 ixgbe_service_event_schedule(adapter);
2435 }
2436 }
2437 ixgbe_check_sfp_event(adapter, eicr);
2438 ixgbe_check_overtemp_event(adapter, eicr);
2439 break;
2440 default:
2441 break;
2442 }
2443
2444 ixgbe_check_fan_failure(adapter, eicr);
2445
2446 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2447 ixgbe_ptp_check_pps_event(adapter, eicr);
2448
2449 /* re-enable the original interrupt state, no lsc, no queues */
2450 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2451 ixgbe_irq_enable(adapter, false, false);
2452
2453 return IRQ_HANDLED;
2454 }
2455
2456 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2457 {
2458 struct ixgbe_q_vector *q_vector = data;
2459
2460 /* EIAM disabled interrupts (on this vector) for us */
2461
2462 if (q_vector->rx.ring || q_vector->tx.ring)
2463 napi_schedule(&q_vector->napi);
2464
2465 return IRQ_HANDLED;
2466 }
2467
2468 /**
2469 * ixgbe_poll - NAPI Rx polling callback
2470 * @napi: structure for representing this polling device
2471 * @budget: how many packets driver is allowed to clean
2472 *
2473 * This function is used for legacy and MSI, NAPI mode
2474 **/
2475 int ixgbe_poll(struct napi_struct *napi, int budget)
2476 {
2477 struct ixgbe_q_vector *q_vector =
2478 container_of(napi, struct ixgbe_q_vector, napi);
2479 struct ixgbe_adapter *adapter = q_vector->adapter;
2480 struct ixgbe_ring *ring;
2481 int per_ring_budget;
2482 bool clean_complete = true;
2483
2484 #ifdef CONFIG_IXGBE_DCA
2485 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2486 ixgbe_update_dca(q_vector);
2487 #endif
2488
2489 ixgbe_for_each_ring(ring, q_vector->tx)
2490 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2491
2492 /* attempt to distribute budget to each queue fairly, but don't allow
2493 * the budget to go below 1 because we'll exit polling */
2494 if (q_vector->rx.count > 1)
2495 per_ring_budget = max(budget/q_vector->rx.count, 1);
2496 else
2497 per_ring_budget = budget;
2498
2499 ixgbe_for_each_ring(ring, q_vector->rx)
2500 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
2501 per_ring_budget);
2502
2503 /* If all work not completed, return budget and keep polling */
2504 if (!clean_complete)
2505 return budget;
2506
2507 /* all work done, exit the polling mode */
2508 napi_complete(napi);
2509 if (adapter->rx_itr_setting & 1)
2510 ixgbe_set_itr(q_vector);
2511 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2512 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2513
2514 return 0;
2515 }
2516
2517 /**
2518 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2519 * @adapter: board private structure
2520 *
2521 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2522 * interrupts from the kernel.
2523 **/
2524 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2525 {
2526 struct net_device *netdev = adapter->netdev;
2527 int vector, err;
2528 int ri = 0, ti = 0;
2529
2530 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2531 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2532 struct msix_entry *entry = &adapter->msix_entries[vector];
2533
2534 if (q_vector->tx.ring && q_vector->rx.ring) {
2535 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2536 "%s-%s-%d", netdev->name, "TxRx", ri++);
2537 ti++;
2538 } else if (q_vector->rx.ring) {
2539 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2540 "%s-%s-%d", netdev->name, "rx", ri++);
2541 } else if (q_vector->tx.ring) {
2542 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2543 "%s-%s-%d", netdev->name, "tx", ti++);
2544 } else {
2545 /* skip this unused q_vector */
2546 continue;
2547 }
2548 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2549 q_vector->name, q_vector);
2550 if (err) {
2551 e_err(probe, "request_irq failed for MSIX interrupt "
2552 "Error: %d\n", err);
2553 goto free_queue_irqs;
2554 }
2555 /* If Flow Director is enabled, set interrupt affinity */
2556 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2557 /* assign the mask for this irq */
2558 irq_set_affinity_hint(entry->vector,
2559 &q_vector->affinity_mask);
2560 }
2561 }
2562
2563 err = request_irq(adapter->msix_entries[vector].vector,
2564 ixgbe_msix_other, 0, netdev->name, adapter);
2565 if (err) {
2566 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2567 goto free_queue_irqs;
2568 }
2569
2570 return 0;
2571
2572 free_queue_irqs:
2573 while (vector) {
2574 vector--;
2575 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2576 NULL);
2577 free_irq(adapter->msix_entries[vector].vector,
2578 adapter->q_vector[vector]);
2579 }
2580 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2581 pci_disable_msix(adapter->pdev);
2582 kfree(adapter->msix_entries);
2583 adapter->msix_entries = NULL;
2584 return err;
2585 }
2586
2587 /**
2588 * ixgbe_intr - legacy mode Interrupt Handler
2589 * @irq: interrupt number
2590 * @data: pointer to a network interface device structure
2591 **/
2592 static irqreturn_t ixgbe_intr(int irq, void *data)
2593 {
2594 struct ixgbe_adapter *adapter = data;
2595 struct ixgbe_hw *hw = &adapter->hw;
2596 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2597 u32 eicr;
2598
2599 /*
2600 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2601 * before the read of EICR.
2602 */
2603 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2604
2605 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2606 * therefore no explicit interrupt disable is necessary */
2607 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2608 if (!eicr) {
2609 /*
2610 * shared interrupt alert!
2611 * make sure interrupts are enabled because the read will
2612 * have disabled interrupts due to EIAM
2613 * finish the workaround of silicon errata on 82598. Unmask
2614 * the interrupt that we masked before the EICR read.
2615 */
2616 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2617 ixgbe_irq_enable(adapter, true, true);
2618 return IRQ_NONE; /* Not our interrupt */
2619 }
2620
2621 if (eicr & IXGBE_EICR_LSC)
2622 ixgbe_check_lsc(adapter);
2623
2624 switch (hw->mac.type) {
2625 case ixgbe_mac_82599EB:
2626 ixgbe_check_sfp_event(adapter, eicr);
2627 /* Fall through */
2628 case ixgbe_mac_X540:
2629 if (eicr & IXGBE_EICR_ECC)
2630 e_info(link, "Received unrecoverable ECC err, please "
2631 "reboot\n");
2632 ixgbe_check_overtemp_event(adapter, eicr);
2633 break;
2634 default:
2635 break;
2636 }
2637
2638 ixgbe_check_fan_failure(adapter, eicr);
2639 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2640 ixgbe_ptp_check_pps_event(adapter, eicr);
2641
2642 /* would disable interrupts here but EIAM disabled it */
2643 napi_schedule(&q_vector->napi);
2644
2645 /*
2646 * re-enable link(maybe) and non-queue interrupts, no flush.
2647 * ixgbe_poll will re-enable the queue interrupts
2648 */
2649 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2650 ixgbe_irq_enable(adapter, false, false);
2651
2652 return IRQ_HANDLED;
2653 }
2654
2655 /**
2656 * ixgbe_request_irq - initialize interrupts
2657 * @adapter: board private structure
2658 *
2659 * Attempts to configure interrupts using the best available
2660 * capabilities of the hardware and kernel.
2661 **/
2662 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2663 {
2664 struct net_device *netdev = adapter->netdev;
2665 int err;
2666
2667 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2668 err = ixgbe_request_msix_irqs(adapter);
2669 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2670 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2671 netdev->name, adapter);
2672 else
2673 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2674 netdev->name, adapter);
2675
2676 if (err)
2677 e_err(probe, "request_irq failed, Error %d\n", err);
2678
2679 return err;
2680 }
2681
2682 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2683 {
2684 int vector;
2685
2686 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2687 free_irq(adapter->pdev->irq, adapter);
2688 return;
2689 }
2690
2691 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2692 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2693 struct msix_entry *entry = &adapter->msix_entries[vector];
2694
2695 /* free only the irqs that were actually requested */
2696 if (!q_vector->rx.ring && !q_vector->tx.ring)
2697 continue;
2698
2699 /* clear the affinity_mask in the IRQ descriptor */
2700 irq_set_affinity_hint(entry->vector, NULL);
2701
2702 free_irq(entry->vector, q_vector);
2703 }
2704
2705 free_irq(adapter->msix_entries[vector++].vector, adapter);
2706 }
2707
2708 /**
2709 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2710 * @adapter: board private structure
2711 **/
2712 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2713 {
2714 switch (adapter->hw.mac.type) {
2715 case ixgbe_mac_82598EB:
2716 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2717 break;
2718 case ixgbe_mac_82599EB:
2719 case ixgbe_mac_X540:
2720 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2721 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2722 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2723 break;
2724 default:
2725 break;
2726 }
2727 IXGBE_WRITE_FLUSH(&adapter->hw);
2728 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2729 int vector;
2730
2731 for (vector = 0; vector < adapter->num_q_vectors; vector++)
2732 synchronize_irq(adapter->msix_entries[vector].vector);
2733
2734 synchronize_irq(adapter->msix_entries[vector++].vector);
2735 } else {
2736 synchronize_irq(adapter->pdev->irq);
2737 }
2738 }
2739
2740 /**
2741 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2742 *
2743 **/
2744 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2745 {
2746 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2747
2748 ixgbe_write_eitr(q_vector);
2749
2750 ixgbe_set_ivar(adapter, 0, 0, 0);
2751 ixgbe_set_ivar(adapter, 1, 0, 0);
2752
2753 e_info(hw, "Legacy interrupt IVAR setup done\n");
2754 }
2755
2756 /**
2757 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2758 * @adapter: board private structure
2759 * @ring: structure containing ring specific data
2760 *
2761 * Configure the Tx descriptor ring after a reset.
2762 **/
2763 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2764 struct ixgbe_ring *ring)
2765 {
2766 struct ixgbe_hw *hw = &adapter->hw;
2767 u64 tdba = ring->dma;
2768 int wait_loop = 10;
2769 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2770 u8 reg_idx = ring->reg_idx;
2771
2772 /* disable queue to avoid issues while updating state */
2773 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2774 IXGBE_WRITE_FLUSH(hw);
2775
2776 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2777 (tdba & DMA_BIT_MASK(32)));
2778 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2779 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2780 ring->count * sizeof(union ixgbe_adv_tx_desc));
2781 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2782 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2783 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2784
2785 /*
2786 * set WTHRESH to encourage burst writeback, it should not be set
2787 * higher than 1 when ITR is 0 as it could cause false TX hangs
2788 *
2789 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2790 * to or less than the number of on chip descriptors, which is
2791 * currently 40.
2792 */
2793 if (!ring->q_vector || (ring->q_vector->itr < 8))
2794 txdctl |= (1 << 16); /* WTHRESH = 1 */
2795 else
2796 txdctl |= (8 << 16); /* WTHRESH = 8 */
2797
2798 /*
2799 * Setting PTHRESH to 32 both improves performance
2800 * and avoids a TX hang with DFP enabled
2801 */
2802 txdctl |= (1 << 8) | /* HTHRESH = 1 */
2803 32; /* PTHRESH = 32 */
2804
2805 /* reinitialize flowdirector state */
2806 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2807 ring->atr_sample_rate = adapter->atr_sample_rate;
2808 ring->atr_count = 0;
2809 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2810 } else {
2811 ring->atr_sample_rate = 0;
2812 }
2813
2814 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2815
2816 /* enable queue */
2817 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2818
2819 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2820 if (hw->mac.type == ixgbe_mac_82598EB &&
2821 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2822 return;
2823
2824 /* poll to verify queue is enabled */
2825 do {
2826 usleep_range(1000, 2000);
2827 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2828 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2829 if (!wait_loop)
2830 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2831 }
2832
2833 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2834 {
2835 struct ixgbe_hw *hw = &adapter->hw;
2836 u32 rttdcs, mtqc;
2837 u8 tcs = netdev_get_num_tc(adapter->netdev);
2838
2839 if (hw->mac.type == ixgbe_mac_82598EB)
2840 return;
2841
2842 /* disable the arbiter while setting MTQC */
2843 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2844 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2845 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2846
2847 /* set transmit pool layout */
2848 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2849 mtqc = IXGBE_MTQC_VT_ENA;
2850 if (tcs > 4)
2851 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2852 else if (tcs > 1)
2853 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2854 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
2855 mtqc |= IXGBE_MTQC_32VF;
2856 else
2857 mtqc |= IXGBE_MTQC_64VF;
2858 } else {
2859 if (tcs > 4)
2860 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2861 else if (tcs > 1)
2862 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2863 else
2864 mtqc = IXGBE_MTQC_64Q_1PB;
2865 }
2866
2867 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
2868
2869 /* Enable Security TX Buffer IFG for multiple pb */
2870 if (tcs) {
2871 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2872 sectx |= IXGBE_SECTX_DCB;
2873 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
2874 }
2875
2876 /* re-enable the arbiter */
2877 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2878 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2879 }
2880
2881 /**
2882 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2883 * @adapter: board private structure
2884 *
2885 * Configure the Tx unit of the MAC after a reset.
2886 **/
2887 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2888 {
2889 struct ixgbe_hw *hw = &adapter->hw;
2890 u32 dmatxctl;
2891 u32 i;
2892
2893 ixgbe_setup_mtqc(adapter);
2894
2895 if (hw->mac.type != ixgbe_mac_82598EB) {
2896 /* DMATXCTL.EN must be before Tx queues are enabled */
2897 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2898 dmatxctl |= IXGBE_DMATXCTL_TE;
2899 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2900 }
2901
2902 /* Setup the HW Tx Head and Tail descriptor pointers */
2903 for (i = 0; i < adapter->num_tx_queues; i++)
2904 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2905 }
2906
2907 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
2908 struct ixgbe_ring *ring)
2909 {
2910 struct ixgbe_hw *hw = &adapter->hw;
2911 u8 reg_idx = ring->reg_idx;
2912 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2913
2914 srrctl |= IXGBE_SRRCTL_DROP_EN;
2915
2916 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2917 }
2918
2919 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
2920 struct ixgbe_ring *ring)
2921 {
2922 struct ixgbe_hw *hw = &adapter->hw;
2923 u8 reg_idx = ring->reg_idx;
2924 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2925
2926 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
2927
2928 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2929 }
2930
2931 #ifdef CONFIG_IXGBE_DCB
2932 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2933 #else
2934 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2935 #endif
2936 {
2937 int i;
2938 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
2939
2940 if (adapter->ixgbe_ieee_pfc)
2941 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
2942
2943 /*
2944 * We should set the drop enable bit if:
2945 * SR-IOV is enabled
2946 * or
2947 * Number of Rx queues > 1 and flow control is disabled
2948 *
2949 * This allows us to avoid head of line blocking for security
2950 * and performance reasons.
2951 */
2952 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
2953 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
2954 for (i = 0; i < adapter->num_rx_queues; i++)
2955 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
2956 } else {
2957 for (i = 0; i < adapter->num_rx_queues; i++)
2958 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
2959 }
2960 }
2961
2962 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2963
2964 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2965 struct ixgbe_ring *rx_ring)
2966 {
2967 struct ixgbe_hw *hw = &adapter->hw;
2968 u32 srrctl;
2969 u8 reg_idx = rx_ring->reg_idx;
2970
2971 if (hw->mac.type == ixgbe_mac_82598EB) {
2972 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
2973
2974 /*
2975 * if VMDq is not active we must program one srrctl register
2976 * per RSS queue since we have enabled RDRXCTL.MVMEN
2977 */
2978 reg_idx &= mask;
2979 }
2980
2981 /* configure header buffer length, needed for RSC */
2982 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
2983
2984 /* configure the packet buffer length */
2985 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2986
2987 /* configure descriptor type */
2988 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2989
2990 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2991 }
2992
2993 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2994 {
2995 struct ixgbe_hw *hw = &adapter->hw;
2996 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2997 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2998 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2999 u32 mrqc = 0, reta = 0;
3000 u32 rxcsum;
3001 int i, j;
3002 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3003
3004 /*
3005 * Program table for at least 2 queues w/ SR-IOV so that VFs can
3006 * make full use of any rings they may have. We will use the
3007 * PSRTYPE register to control how many rings we use within the PF.
3008 */
3009 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
3010 rss_i = 2;
3011
3012 /* Fill out hash function seeds */
3013 for (i = 0; i < 10; i++)
3014 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
3015
3016 /* Fill out redirection table */
3017 for (i = 0, j = 0; i < 128; i++, j++) {
3018 if (j == rss_i)
3019 j = 0;
3020 /* reta = 4-byte sliding window of
3021 * 0x00..(indices-1)(indices-1)00..etc. */
3022 reta = (reta << 8) | (j * 0x11);
3023 if ((i & 3) == 3)
3024 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3025 }
3026
3027 /* Disable indicating checksum in descriptor, enables RSS hash */
3028 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3029 rxcsum |= IXGBE_RXCSUM_PCSD;
3030 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3031
3032 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3033 if (adapter->ring_feature[RING_F_RSS].mask)
3034 mrqc = IXGBE_MRQC_RSSEN;
3035 } else {
3036 u8 tcs = netdev_get_num_tc(adapter->netdev);
3037
3038 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3039 if (tcs > 4)
3040 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3041 else if (tcs > 1)
3042 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3043 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3044 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3045 else
3046 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3047 } else {
3048 if (tcs > 4)
3049 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3050 else if (tcs > 1)
3051 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3052 else
3053 mrqc = IXGBE_MRQC_RSSEN;
3054 }
3055 }
3056
3057 /* Perform hash on these packet types */
3058 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3059 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3060 IXGBE_MRQC_RSS_FIELD_IPV6 |
3061 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3062
3063 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3064 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3065 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3066 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3067
3068 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3069 }
3070
3071 /**
3072 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3073 * @adapter: address of board private structure
3074 * @index: index of ring to set
3075 **/
3076 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3077 struct ixgbe_ring *ring)
3078 {
3079 struct ixgbe_hw *hw = &adapter->hw;
3080 u32 rscctrl;
3081 u8 reg_idx = ring->reg_idx;
3082
3083 if (!ring_is_rsc_enabled(ring))
3084 return;
3085
3086 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3087 rscctrl |= IXGBE_RSCCTL_RSCEN;
3088 /*
3089 * we must limit the number of descriptors so that the
3090 * total size of max desc * buf_len is not greater
3091 * than 65536
3092 */
3093 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3094 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3095 }
3096
3097 #define IXGBE_MAX_RX_DESC_POLL 10
3098 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3099 struct ixgbe_ring *ring)
3100 {
3101 struct ixgbe_hw *hw = &adapter->hw;
3102 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3103 u32 rxdctl;
3104 u8 reg_idx = ring->reg_idx;
3105
3106 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3107 if (hw->mac.type == ixgbe_mac_82598EB &&
3108 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3109 return;
3110
3111 do {
3112 usleep_range(1000, 2000);
3113 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3114 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3115
3116 if (!wait_loop) {
3117 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3118 "the polling period\n", reg_idx);
3119 }
3120 }
3121
3122 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3123 struct ixgbe_ring *ring)
3124 {
3125 struct ixgbe_hw *hw = &adapter->hw;
3126 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3127 u32 rxdctl;
3128 u8 reg_idx = ring->reg_idx;
3129
3130 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3131 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3132
3133 /* write value back with RXDCTL.ENABLE bit cleared */
3134 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3135
3136 if (hw->mac.type == ixgbe_mac_82598EB &&
3137 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3138 return;
3139
3140 /* the hardware may take up to 100us to really disable the rx queue */
3141 do {
3142 udelay(10);
3143 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3144 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3145
3146 if (!wait_loop) {
3147 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3148 "the polling period\n", reg_idx);
3149 }
3150 }
3151
3152 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3153 struct ixgbe_ring *ring)
3154 {
3155 struct ixgbe_hw *hw = &adapter->hw;
3156 u64 rdba = ring->dma;
3157 u32 rxdctl;
3158 u8 reg_idx = ring->reg_idx;
3159
3160 /* disable queue to avoid issues while updating state */
3161 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3162 ixgbe_disable_rx_queue(adapter, ring);
3163
3164 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3165 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3166 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3167 ring->count * sizeof(union ixgbe_adv_rx_desc));
3168 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3169 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3170 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3171
3172 ixgbe_configure_srrctl(adapter, ring);
3173 ixgbe_configure_rscctl(adapter, ring);
3174
3175 if (hw->mac.type == ixgbe_mac_82598EB) {
3176 /*
3177 * enable cache line friendly hardware writes:
3178 * PTHRESH=32 descriptors (half the internal cache),
3179 * this also removes ugly rx_no_buffer_count increment
3180 * HTHRESH=4 descriptors (to minimize latency on fetch)
3181 * WTHRESH=8 burst writeback up to two cache lines
3182 */
3183 rxdctl &= ~0x3FFFFF;
3184 rxdctl |= 0x080420;
3185 }
3186
3187 /* enable receive descriptor ring */
3188 rxdctl |= IXGBE_RXDCTL_ENABLE;
3189 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3190
3191 ixgbe_rx_desc_queue_enable(adapter, ring);
3192 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3193 }
3194
3195 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3196 {
3197 struct ixgbe_hw *hw = &adapter->hw;
3198 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3199 int p;
3200
3201 /* PSRTYPE must be initialized in non 82598 adapters */
3202 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3203 IXGBE_PSRTYPE_UDPHDR |
3204 IXGBE_PSRTYPE_IPV4HDR |
3205 IXGBE_PSRTYPE_L2HDR |
3206 IXGBE_PSRTYPE_IPV6HDR;
3207
3208 if (hw->mac.type == ixgbe_mac_82598EB)
3209 return;
3210
3211 if (rss_i > 3)
3212 psrtype |= 2 << 29;
3213 else if (rss_i > 1)
3214 psrtype |= 1 << 29;
3215
3216 for (p = 0; p < adapter->num_rx_pools; p++)
3217 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(p)),
3218 psrtype);
3219 }
3220
3221 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3222 {
3223 struct ixgbe_hw *hw = &adapter->hw;
3224 u32 reg_offset, vf_shift;
3225 u32 gcr_ext, vmdctl;
3226 int i;
3227
3228 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3229 return;
3230
3231 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3232 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3233 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3234 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3235 vmdctl |= IXGBE_VT_CTL_REPLEN;
3236 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3237
3238 vf_shift = VMDQ_P(0) % 32;
3239 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3240
3241 /* Enable only the PF's pool for Tx/Rx */
3242 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3243 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3244 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3245 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3246 if (adapter->flags2 & IXGBE_FLAG2_BRIDGE_MODE_VEB)
3247 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3248
3249 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3250 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3251
3252 /*
3253 * Set up VF register offsets for selected VT Mode,
3254 * i.e. 32 or 64 VFs for SR-IOV
3255 */
3256 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3257 case IXGBE_82599_VMDQ_8Q_MASK:
3258 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3259 break;
3260 case IXGBE_82599_VMDQ_4Q_MASK:
3261 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3262 break;
3263 default:
3264 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3265 break;
3266 }
3267
3268 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3269
3270
3271 /* Enable MAC Anti-Spoofing */
3272 hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3273 adapter->num_vfs);
3274 /* For VFs that have spoof checking turned off */
3275 for (i = 0; i < adapter->num_vfs; i++) {
3276 if (!adapter->vfinfo[i].spoofchk_enabled)
3277 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3278 }
3279 }
3280
3281 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3282 {
3283 struct ixgbe_hw *hw = &adapter->hw;
3284 struct net_device *netdev = adapter->netdev;
3285 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3286 struct ixgbe_ring *rx_ring;
3287 int i;
3288 u32 mhadd, hlreg0;
3289
3290 #ifdef IXGBE_FCOE
3291 /* adjust max frame to be able to do baby jumbo for FCoE */
3292 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3293 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3294 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3295
3296 #endif /* IXGBE_FCOE */
3297
3298 /* adjust max frame to be at least the size of a standard frame */
3299 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3300 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3301
3302 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3303 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3304 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3305 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3306
3307 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3308 }
3309
3310 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3311 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3312 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3313 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
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 rx_ring = adapter->rx_ring[i];
3321 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3322 set_ring_rsc_enabled(rx_ring);
3323 else
3324 clear_ring_rsc_enabled(rx_ring);
3325 }
3326 }
3327
3328 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3329 {
3330 struct ixgbe_hw *hw = &adapter->hw;
3331 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3332
3333 switch (hw->mac.type) {
3334 case ixgbe_mac_82598EB:
3335 /*
3336 * For VMDq support of different descriptor types or
3337 * buffer sizes through the use of multiple SRRCTL
3338 * registers, RDRXCTL.MVMEN must be set to 1
3339 *
3340 * also, the manual doesn't mention it clearly but DCA hints
3341 * will only use queue 0's tags unless this bit is set. Side
3342 * effects of setting this bit are only that SRRCTL must be
3343 * fully programmed [0..15]
3344 */
3345 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3346 break;
3347 case ixgbe_mac_82599EB:
3348 case ixgbe_mac_X540:
3349 /* Disable RSC for ACK packets */
3350 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3351 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3352 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3353 /* hardware requires some bits to be set by default */
3354 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3355 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3356 break;
3357 default:
3358 /* We should do nothing since we don't know this hardware */
3359 return;
3360 }
3361
3362 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3363 }
3364
3365 /**
3366 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3367 * @adapter: board private structure
3368 *
3369 * Configure the Rx unit of the MAC after a reset.
3370 **/
3371 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3372 {
3373 struct ixgbe_hw *hw = &adapter->hw;
3374 int i;
3375 u32 rxctrl;
3376
3377 /* disable receives while setting up the descriptors */
3378 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3379 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3380
3381 ixgbe_setup_psrtype(adapter);
3382 ixgbe_setup_rdrxctl(adapter);
3383
3384 /* Program registers for the distribution of queues */
3385 ixgbe_setup_mrqc(adapter);
3386
3387 /* set_rx_buffer_len must be called before ring initialization */
3388 ixgbe_set_rx_buffer_len(adapter);
3389
3390 /*
3391 * Setup the HW Rx Head and Tail Descriptor Pointers and
3392 * the Base and Length of the Rx Descriptor Ring
3393 */
3394 for (i = 0; i < adapter->num_rx_queues; i++)
3395 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3396
3397 /* disable drop enable for 82598 parts */
3398 if (hw->mac.type == ixgbe_mac_82598EB)
3399 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3400
3401 /* enable all receives */
3402 rxctrl |= IXGBE_RXCTRL_RXEN;
3403 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3404 }
3405
3406 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3407 {
3408 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3409 struct ixgbe_hw *hw = &adapter->hw;
3410
3411 /* add VID to filter table */
3412 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3413 set_bit(vid, adapter->active_vlans);
3414
3415 return 0;
3416 }
3417
3418 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3419 {
3420 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3421 struct ixgbe_hw *hw = &adapter->hw;
3422
3423 /* remove VID from filter table */
3424 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3425 clear_bit(vid, adapter->active_vlans);
3426
3427 return 0;
3428 }
3429
3430 /**
3431 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3432 * @adapter: driver data
3433 */
3434 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3435 {
3436 struct ixgbe_hw *hw = &adapter->hw;
3437 u32 vlnctrl;
3438
3439 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3440 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3441 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3442 }
3443
3444 /**
3445 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3446 * @adapter: driver data
3447 */
3448 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3449 {
3450 struct ixgbe_hw *hw = &adapter->hw;
3451 u32 vlnctrl;
3452
3453 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3454 vlnctrl |= IXGBE_VLNCTRL_VFE;
3455 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3456 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3457 }
3458
3459 /**
3460 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3461 * @adapter: driver data
3462 */
3463 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3464 {
3465 struct ixgbe_hw *hw = &adapter->hw;
3466 u32 vlnctrl;
3467 int i, j;
3468
3469 switch (hw->mac.type) {
3470 case ixgbe_mac_82598EB:
3471 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3472 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3473 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3474 break;
3475 case ixgbe_mac_82599EB:
3476 case ixgbe_mac_X540:
3477 for (i = 0; i < adapter->num_rx_queues; i++) {
3478 j = adapter->rx_ring[i]->reg_idx;
3479 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3480 vlnctrl &= ~IXGBE_RXDCTL_VME;
3481 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3482 }
3483 break;
3484 default:
3485 break;
3486 }
3487 }
3488
3489 /**
3490 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3491 * @adapter: driver data
3492 */
3493 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3494 {
3495 struct ixgbe_hw *hw = &adapter->hw;
3496 u32 vlnctrl;
3497 int i, j;
3498
3499 switch (hw->mac.type) {
3500 case ixgbe_mac_82598EB:
3501 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3502 vlnctrl |= IXGBE_VLNCTRL_VME;
3503 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3504 break;
3505 case ixgbe_mac_82599EB:
3506 case ixgbe_mac_X540:
3507 for (i = 0; i < adapter->num_rx_queues; i++) {
3508 j = adapter->rx_ring[i]->reg_idx;
3509 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3510 vlnctrl |= IXGBE_RXDCTL_VME;
3511 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3512 }
3513 break;
3514 default:
3515 break;
3516 }
3517 }
3518
3519 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3520 {
3521 u16 vid;
3522
3523 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3524
3525 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3526 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3527 }
3528
3529 /**
3530 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3531 * @netdev: network interface device structure
3532 *
3533 * Writes unicast address list to the RAR table.
3534 * Returns: -ENOMEM on failure/insufficient address space
3535 * 0 on no addresses written
3536 * X on writing X addresses to the RAR table
3537 **/
3538 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3539 {
3540 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3541 struct ixgbe_hw *hw = &adapter->hw;
3542 unsigned int rar_entries = hw->mac.num_rar_entries - 1;
3543 int count = 0;
3544
3545 /* In SR-IOV mode significantly less RAR entries are available */
3546 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3547 rar_entries = IXGBE_MAX_PF_MACVLANS - 1;
3548
3549 /* return ENOMEM indicating insufficient memory for addresses */
3550 if (netdev_uc_count(netdev) > rar_entries)
3551 return -ENOMEM;
3552
3553 if (!netdev_uc_empty(netdev)) {
3554 struct netdev_hw_addr *ha;
3555 /* return error if we do not support writing to RAR table */
3556 if (!hw->mac.ops.set_rar)
3557 return -ENOMEM;
3558
3559 netdev_for_each_uc_addr(ha, netdev) {
3560 if (!rar_entries)
3561 break;
3562 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3563 VMDQ_P(0), IXGBE_RAH_AV);
3564 count++;
3565 }
3566 }
3567 /* write the addresses in reverse order to avoid write combining */
3568 for (; rar_entries > 0 ; rar_entries--)
3569 hw->mac.ops.clear_rar(hw, rar_entries);
3570
3571 return count;
3572 }
3573
3574 /**
3575 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3576 * @netdev: network interface device structure
3577 *
3578 * The set_rx_method entry point is called whenever the unicast/multicast
3579 * address list or the network interface flags are updated. This routine is
3580 * responsible for configuring the hardware for proper unicast, multicast and
3581 * promiscuous mode.
3582 **/
3583 void ixgbe_set_rx_mode(struct net_device *netdev)
3584 {
3585 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3586 struct ixgbe_hw *hw = &adapter->hw;
3587 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3588 int count;
3589
3590 /* Check for Promiscuous and All Multicast modes */
3591
3592 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3593
3594 /* set all bits that we expect to always be set */
3595 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3596 fctrl |= IXGBE_FCTRL_BAM;
3597 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3598 fctrl |= IXGBE_FCTRL_PMCF;
3599
3600 /* clear the bits we are changing the status of */
3601 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3602
3603 if (netdev->flags & IFF_PROMISC) {
3604 hw->addr_ctrl.user_set_promisc = true;
3605 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3606 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3607 /* don't hardware filter vlans in promisc mode */
3608 ixgbe_vlan_filter_disable(adapter);
3609 } else {
3610 if (netdev->flags & IFF_ALLMULTI) {
3611 fctrl |= IXGBE_FCTRL_MPE;
3612 vmolr |= IXGBE_VMOLR_MPE;
3613 } else {
3614 /*
3615 * Write addresses to the MTA, if the attempt fails
3616 * then we should just turn on promiscuous mode so
3617 * that we can at least receive multicast traffic
3618 */
3619 hw->mac.ops.update_mc_addr_list(hw, netdev);
3620 vmolr |= IXGBE_VMOLR_ROMPE;
3621 }
3622 ixgbe_vlan_filter_enable(adapter);
3623 hw->addr_ctrl.user_set_promisc = false;
3624 }
3625
3626 /*
3627 * Write addresses to available RAR registers, if there is not
3628 * sufficient space to store all the addresses then enable
3629 * unicast promiscuous mode
3630 */
3631 count = ixgbe_write_uc_addr_list(netdev);
3632 if (count < 0) {
3633 fctrl |= IXGBE_FCTRL_UPE;
3634 vmolr |= IXGBE_VMOLR_ROPE;
3635 }
3636
3637 if (adapter->num_vfs)
3638 ixgbe_restore_vf_multicasts(adapter);
3639
3640 if (hw->mac.type != ixgbe_mac_82598EB) {
3641 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
3642 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3643 IXGBE_VMOLR_ROPE);
3644 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
3645 }
3646
3647 /* This is useful for sniffing bad packets. */
3648 if (adapter->netdev->features & NETIF_F_RXALL) {
3649 /* UPE and MPE will be handled by normal PROMISC logic
3650 * in e1000e_set_rx_mode */
3651 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3652 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3653 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3654
3655 fctrl &= ~(IXGBE_FCTRL_DPF);
3656 /* NOTE: VLAN filtering is disabled by setting PROMISC */
3657 }
3658
3659 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3660
3661 if (netdev->features & NETIF_F_HW_VLAN_RX)
3662 ixgbe_vlan_strip_enable(adapter);
3663 else
3664 ixgbe_vlan_strip_disable(adapter);
3665 }
3666
3667 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3668 {
3669 int q_idx;
3670
3671 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3672 napi_enable(&adapter->q_vector[q_idx]->napi);
3673 }
3674
3675 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3676 {
3677 int q_idx;
3678
3679 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3680 napi_disable(&adapter->q_vector[q_idx]->napi);
3681 }
3682
3683 #ifdef CONFIG_IXGBE_DCB
3684 /**
3685 * ixgbe_configure_dcb - Configure DCB hardware
3686 * @adapter: ixgbe adapter struct
3687 *
3688 * This is called by the driver on open to configure the DCB hardware.
3689 * This is also called by the gennetlink interface when reconfiguring
3690 * the DCB state.
3691 */
3692 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3693 {
3694 struct ixgbe_hw *hw = &adapter->hw;
3695 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3696
3697 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3698 if (hw->mac.type == ixgbe_mac_82598EB)
3699 netif_set_gso_max_size(adapter->netdev, 65536);
3700 return;
3701 }
3702
3703 if (hw->mac.type == ixgbe_mac_82598EB)
3704 netif_set_gso_max_size(adapter->netdev, 32768);
3705
3706 #ifdef IXGBE_FCOE
3707 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3708 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3709 #endif
3710
3711 /* reconfigure the hardware */
3712 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3713 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3714 DCB_TX_CONFIG);
3715 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3716 DCB_RX_CONFIG);
3717 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3718 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3719 ixgbe_dcb_hw_ets(&adapter->hw,
3720 adapter->ixgbe_ieee_ets,
3721 max_frame);
3722 ixgbe_dcb_hw_pfc_config(&adapter->hw,
3723 adapter->ixgbe_ieee_pfc->pfc_en,
3724 adapter->ixgbe_ieee_ets->prio_tc);
3725 }
3726
3727 /* Enable RSS Hash per TC */
3728 if (hw->mac.type != ixgbe_mac_82598EB) {
3729 u32 msb = 0;
3730 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
3731
3732 while (rss_i) {
3733 msb++;
3734 rss_i >>= 1;
3735 }
3736
3737 /* write msb to all 8 TCs in one write */
3738 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
3739 }
3740 }
3741 #endif
3742
3743 /* Additional bittime to account for IXGBE framing */
3744 #define IXGBE_ETH_FRAMING 20
3745
3746 /**
3747 * ixgbe_hpbthresh - calculate high water mark for flow control
3748 *
3749 * @adapter: board private structure to calculate for
3750 * @pb: packet buffer to calculate
3751 */
3752 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3753 {
3754 struct ixgbe_hw *hw = &adapter->hw;
3755 struct net_device *dev = adapter->netdev;
3756 int link, tc, kb, marker;
3757 u32 dv_id, rx_pba;
3758
3759 /* Calculate max LAN frame size */
3760 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3761
3762 #ifdef IXGBE_FCOE
3763 /* FCoE traffic class uses FCOE jumbo frames */
3764 if ((dev->features & NETIF_F_FCOE_MTU) &&
3765 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
3766 (pb == ixgbe_fcoe_get_tc(adapter)))
3767 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3768
3769 #endif
3770 /* Calculate delay value for device */
3771 switch (hw->mac.type) {
3772 case ixgbe_mac_X540:
3773 dv_id = IXGBE_DV_X540(link, tc);
3774 break;
3775 default:
3776 dv_id = IXGBE_DV(link, tc);
3777 break;
3778 }
3779
3780 /* Loopback switch introduces additional latency */
3781 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3782 dv_id += IXGBE_B2BT(tc);
3783
3784 /* Delay value is calculated in bit times convert to KB */
3785 kb = IXGBE_BT2KB(dv_id);
3786 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3787
3788 marker = rx_pba - kb;
3789
3790 /* It is possible that the packet buffer is not large enough
3791 * to provide required headroom. In this case throw an error
3792 * to user and a do the best we can.
3793 */
3794 if (marker < 0) {
3795 e_warn(drv, "Packet Buffer(%i) can not provide enough"
3796 "headroom to support flow control."
3797 "Decrease MTU or number of traffic classes\n", pb);
3798 marker = tc + 1;
3799 }
3800
3801 return marker;
3802 }
3803
3804 /**
3805 * ixgbe_lpbthresh - calculate low water mark for for flow control
3806 *
3807 * @adapter: board private structure to calculate for
3808 * @pb: packet buffer to calculate
3809 */
3810 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3811 {
3812 struct ixgbe_hw *hw = &adapter->hw;
3813 struct net_device *dev = adapter->netdev;
3814 int tc;
3815 u32 dv_id;
3816
3817 /* Calculate max LAN frame size */
3818 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3819
3820 /* Calculate delay value for device */
3821 switch (hw->mac.type) {
3822 case ixgbe_mac_X540:
3823 dv_id = IXGBE_LOW_DV_X540(tc);
3824 break;
3825 default:
3826 dv_id = IXGBE_LOW_DV(tc);
3827 break;
3828 }
3829
3830 /* Delay value is calculated in bit times convert to KB */
3831 return IXGBE_BT2KB(dv_id);
3832 }
3833
3834 /*
3835 * ixgbe_pbthresh_setup - calculate and setup high low water marks
3836 */
3837 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3838 {
3839 struct ixgbe_hw *hw = &adapter->hw;
3840 int num_tc = netdev_get_num_tc(adapter->netdev);
3841 int i;
3842
3843 if (!num_tc)
3844 num_tc = 1;
3845
3846 hw->fc.low_water = ixgbe_lpbthresh(adapter);
3847
3848 for (i = 0; i < num_tc; i++) {
3849 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3850
3851 /* Low water marks must not be larger than high water marks */
3852 if (hw->fc.low_water > hw->fc.high_water[i])
3853 hw->fc.low_water = 0;
3854 }
3855 }
3856
3857 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3858 {
3859 struct ixgbe_hw *hw = &adapter->hw;
3860 int hdrm;
3861 u8 tc = netdev_get_num_tc(adapter->netdev);
3862
3863 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3864 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3865 hdrm = 32 << adapter->fdir_pballoc;
3866 else
3867 hdrm = 0;
3868
3869 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3870 ixgbe_pbthresh_setup(adapter);
3871 }
3872
3873 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3874 {
3875 struct ixgbe_hw *hw = &adapter->hw;
3876 struct hlist_node *node, *node2;
3877 struct ixgbe_fdir_filter *filter;
3878
3879 spin_lock(&adapter->fdir_perfect_lock);
3880
3881 if (!hlist_empty(&adapter->fdir_filter_list))
3882 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3883
3884 hlist_for_each_entry_safe(filter, node, node2,
3885 &adapter->fdir_filter_list, fdir_node) {
3886 ixgbe_fdir_write_perfect_filter_82599(hw,
3887 &filter->filter,
3888 filter->sw_idx,
3889 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3890 IXGBE_FDIR_DROP_QUEUE :
3891 adapter->rx_ring[filter->action]->reg_idx);
3892 }
3893
3894 spin_unlock(&adapter->fdir_perfect_lock);
3895 }
3896
3897 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3898 {
3899 struct ixgbe_hw *hw = &adapter->hw;
3900
3901 ixgbe_configure_pb(adapter);
3902 #ifdef CONFIG_IXGBE_DCB
3903 ixgbe_configure_dcb(adapter);
3904 #endif
3905 /*
3906 * We must restore virtualization before VLANs or else
3907 * the VLVF registers will not be populated
3908 */
3909 ixgbe_configure_virtualization(adapter);
3910
3911 ixgbe_set_rx_mode(adapter->netdev);
3912 ixgbe_restore_vlan(adapter);
3913
3914 switch (hw->mac.type) {
3915 case ixgbe_mac_82599EB:
3916 case ixgbe_mac_X540:
3917 hw->mac.ops.disable_rx_buff(hw);
3918 break;
3919 default:
3920 break;
3921 }
3922
3923 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3924 ixgbe_init_fdir_signature_82599(&adapter->hw,
3925 adapter->fdir_pballoc);
3926 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3927 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3928 adapter->fdir_pballoc);
3929 ixgbe_fdir_filter_restore(adapter);
3930 }
3931
3932 switch (hw->mac.type) {
3933 case ixgbe_mac_82599EB:
3934 case ixgbe_mac_X540:
3935 hw->mac.ops.enable_rx_buff(hw);
3936 break;
3937 default:
3938 break;
3939 }
3940
3941 #ifdef IXGBE_FCOE
3942 /* configure FCoE L2 filters, redirection table, and Rx control */
3943 ixgbe_configure_fcoe(adapter);
3944
3945 #endif /* IXGBE_FCOE */
3946 ixgbe_configure_tx(adapter);
3947 ixgbe_configure_rx(adapter);
3948 }
3949
3950 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3951 {
3952 switch (hw->phy.type) {
3953 case ixgbe_phy_sfp_avago:
3954 case ixgbe_phy_sfp_ftl:
3955 case ixgbe_phy_sfp_intel:
3956 case ixgbe_phy_sfp_unknown:
3957 case ixgbe_phy_sfp_passive_tyco:
3958 case ixgbe_phy_sfp_passive_unknown:
3959 case ixgbe_phy_sfp_active_unknown:
3960 case ixgbe_phy_sfp_ftl_active:
3961 return true;
3962 case ixgbe_phy_nl:
3963 if (hw->mac.type == ixgbe_mac_82598EB)
3964 return true;
3965 default:
3966 return false;
3967 }
3968 }
3969
3970 /**
3971 * ixgbe_sfp_link_config - set up SFP+ link
3972 * @adapter: pointer to private adapter struct
3973 **/
3974 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3975 {
3976 /*
3977 * We are assuming the worst case scenario here, and that
3978 * is that an SFP was inserted/removed after the reset
3979 * but before SFP detection was enabled. As such the best
3980 * solution is to just start searching as soon as we start
3981 */
3982 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3983 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3984
3985 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3986 }
3987
3988 /**
3989 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3990 * @hw: pointer to private hardware struct
3991 *
3992 * Returns 0 on success, negative on failure
3993 **/
3994 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3995 {
3996 u32 autoneg;
3997 bool negotiation, link_up = false;
3998 u32 ret = IXGBE_ERR_LINK_SETUP;
3999
4000 if (hw->mac.ops.check_link)
4001 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
4002
4003 if (ret)
4004 goto link_cfg_out;
4005
4006 autoneg = hw->phy.autoneg_advertised;
4007 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
4008 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
4009 &negotiation);
4010 if (ret)
4011 goto link_cfg_out;
4012
4013 if (hw->mac.ops.setup_link)
4014 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
4015 link_cfg_out:
4016 return ret;
4017 }
4018
4019 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
4020 {
4021 struct ixgbe_hw *hw = &adapter->hw;
4022 u32 gpie = 0;
4023
4024 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4025 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
4026 IXGBE_GPIE_OCD;
4027 gpie |= IXGBE_GPIE_EIAME;
4028 /*
4029 * use EIAM to auto-mask when MSI-X interrupt is asserted
4030 * this saves a register write for every interrupt
4031 */
4032 switch (hw->mac.type) {
4033 case ixgbe_mac_82598EB:
4034 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4035 break;
4036 case ixgbe_mac_82599EB:
4037 case ixgbe_mac_X540:
4038 default:
4039 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4040 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4041 break;
4042 }
4043 } else {
4044 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
4045 * specifically only auto mask tx and rx interrupts */
4046 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4047 }
4048
4049 /* XXX: to interrupt immediately for EICS writes, enable this */
4050 /* gpie |= IXGBE_GPIE_EIMEN; */
4051
4052 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
4053 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
4054
4055 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4056 case IXGBE_82599_VMDQ_8Q_MASK:
4057 gpie |= IXGBE_GPIE_VTMODE_16;
4058 break;
4059 case IXGBE_82599_VMDQ_4Q_MASK:
4060 gpie |= IXGBE_GPIE_VTMODE_32;
4061 break;
4062 default:
4063 gpie |= IXGBE_GPIE_VTMODE_64;
4064 break;
4065 }
4066 }
4067
4068 /* Enable Thermal over heat sensor interrupt */
4069 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
4070 switch (adapter->hw.mac.type) {
4071 case ixgbe_mac_82599EB:
4072 gpie |= IXGBE_SDP0_GPIEN;
4073 break;
4074 case ixgbe_mac_X540:
4075 gpie |= IXGBE_EIMS_TS;
4076 break;
4077 default:
4078 break;
4079 }
4080 }
4081
4082 /* Enable fan failure interrupt */
4083 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
4084 gpie |= IXGBE_SDP1_GPIEN;
4085
4086 if (hw->mac.type == ixgbe_mac_82599EB) {
4087 gpie |= IXGBE_SDP1_GPIEN;
4088 gpie |= IXGBE_SDP2_GPIEN;
4089 }
4090
4091 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4092 }
4093
4094 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
4095 {
4096 struct ixgbe_hw *hw = &adapter->hw;
4097 int err;
4098 u32 ctrl_ext;
4099
4100 ixgbe_get_hw_control(adapter);
4101 ixgbe_setup_gpie(adapter);
4102
4103 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4104 ixgbe_configure_msix(adapter);
4105 else
4106 ixgbe_configure_msi_and_legacy(adapter);
4107
4108 /* enable the optics for 82599 SFP+ fiber */
4109 if (hw->mac.ops.enable_tx_laser)
4110 hw->mac.ops.enable_tx_laser(hw);
4111
4112 clear_bit(__IXGBE_DOWN, &adapter->state);
4113 ixgbe_napi_enable_all(adapter);
4114
4115 if (ixgbe_is_sfp(hw)) {
4116 ixgbe_sfp_link_config(adapter);
4117 } else {
4118 err = ixgbe_non_sfp_link_config(hw);
4119 if (err)
4120 e_err(probe, "link_config FAILED %d\n", err);
4121 }
4122
4123 /* clear any pending interrupts, may auto mask */
4124 IXGBE_READ_REG(hw, IXGBE_EICR);
4125 ixgbe_irq_enable(adapter, true, true);
4126
4127 /*
4128 * If this adapter has a fan, check to see if we had a failure
4129 * before we enabled the interrupt.
4130 */
4131 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4132 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4133 if (esdp & IXGBE_ESDP_SDP1)
4134 e_crit(drv, "Fan has stopped, replace the adapter\n");
4135 }
4136
4137 /* enable transmits */
4138 netif_tx_start_all_queues(adapter->netdev);
4139
4140 /* bring the link up in the watchdog, this could race with our first
4141 * link up interrupt but shouldn't be a problem */
4142 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4143 adapter->link_check_timeout = jiffies;
4144 mod_timer(&adapter->service_timer, jiffies);
4145
4146 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4147 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4148 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4149 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4150 }
4151
4152 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4153 {
4154 WARN_ON(in_interrupt());
4155 /* put off any impending NetWatchDogTimeout */
4156 adapter->netdev->trans_start = jiffies;
4157
4158 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4159 usleep_range(1000, 2000);
4160 ixgbe_down(adapter);
4161 /*
4162 * If SR-IOV enabled then wait a bit before bringing the adapter
4163 * back up to give the VFs time to respond to the reset. The
4164 * two second wait is based upon the watchdog timer cycle in
4165 * the VF driver.
4166 */
4167 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4168 msleep(2000);
4169 ixgbe_up(adapter);
4170 clear_bit(__IXGBE_RESETTING, &adapter->state);
4171 }
4172
4173 void ixgbe_up(struct ixgbe_adapter *adapter)
4174 {
4175 /* hardware has been reset, we need to reload some things */
4176 ixgbe_configure(adapter);
4177
4178 ixgbe_up_complete(adapter);
4179 }
4180
4181 void ixgbe_reset(struct ixgbe_adapter *adapter)
4182 {
4183 struct ixgbe_hw *hw = &adapter->hw;
4184 int err;
4185
4186 /* lock SFP init bit to prevent race conditions with the watchdog */
4187 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4188 usleep_range(1000, 2000);
4189
4190 /* clear all SFP and link config related flags while holding SFP_INIT */
4191 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4192 IXGBE_FLAG2_SFP_NEEDS_RESET);
4193 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4194
4195 err = hw->mac.ops.init_hw(hw);
4196 switch (err) {
4197 case 0:
4198 case IXGBE_ERR_SFP_NOT_PRESENT:
4199 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4200 break;
4201 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4202 e_dev_err("master disable timed out\n");
4203 break;
4204 case IXGBE_ERR_EEPROM_VERSION:
4205 /* We are running on a pre-production device, log a warning */
4206 e_dev_warn("This device is a pre-production adapter/LOM. "
4207 "Please be aware there may be issues associated with "
4208 "your hardware. If you are experiencing problems "
4209 "please contact your Intel or hardware "
4210 "representative who provided you with this "
4211 "hardware.\n");
4212 break;
4213 default:
4214 e_dev_err("Hardware Error: %d\n", err);
4215 }
4216
4217 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4218
4219 /* reprogram the RAR[0] in case user changed it. */
4220 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
4221
4222 /* update SAN MAC vmdq pool selection */
4223 if (hw->mac.san_mac_rar_index)
4224 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
4225
4226 if (adapter->flags2 & IXGBE_FLAG2_PTP_ENABLED)
4227 ixgbe_ptp_reset(adapter);
4228 }
4229
4230 /**
4231 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4232 * @rx_ring: ring to free buffers from
4233 **/
4234 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4235 {
4236 struct device *dev = rx_ring->dev;
4237 unsigned long size;
4238 u16 i;
4239
4240 /* ring already cleared, nothing to do */
4241 if (!rx_ring->rx_buffer_info)
4242 return;
4243
4244 /* Free all the Rx ring sk_buffs */
4245 for (i = 0; i < rx_ring->count; i++) {
4246 struct ixgbe_rx_buffer *rx_buffer;
4247
4248 rx_buffer = &rx_ring->rx_buffer_info[i];
4249 if (rx_buffer->skb) {
4250 struct sk_buff *skb = rx_buffer->skb;
4251 if (IXGBE_CB(skb)->page_released) {
4252 dma_unmap_page(dev,
4253 IXGBE_CB(skb)->dma,
4254 ixgbe_rx_bufsz(rx_ring),
4255 DMA_FROM_DEVICE);
4256 IXGBE_CB(skb)->page_released = false;
4257 }
4258 dev_kfree_skb(skb);
4259 }
4260 rx_buffer->skb = NULL;
4261 if (rx_buffer->dma)
4262 dma_unmap_page(dev, rx_buffer->dma,
4263 ixgbe_rx_pg_size(rx_ring),
4264 DMA_FROM_DEVICE);
4265 rx_buffer->dma = 0;
4266 if (rx_buffer->page)
4267 __free_pages(rx_buffer->page,
4268 ixgbe_rx_pg_order(rx_ring));
4269 rx_buffer->page = NULL;
4270 }
4271
4272 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4273 memset(rx_ring->rx_buffer_info, 0, size);
4274
4275 /* Zero out the descriptor ring */
4276 memset(rx_ring->desc, 0, rx_ring->size);
4277
4278 rx_ring->next_to_alloc = 0;
4279 rx_ring->next_to_clean = 0;
4280 rx_ring->next_to_use = 0;
4281 }
4282
4283 /**
4284 * ixgbe_clean_tx_ring - Free Tx Buffers
4285 * @tx_ring: ring to be cleaned
4286 **/
4287 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4288 {
4289 struct ixgbe_tx_buffer *tx_buffer_info;
4290 unsigned long size;
4291 u16 i;
4292
4293 /* ring already cleared, nothing to do */
4294 if (!tx_ring->tx_buffer_info)
4295 return;
4296
4297 /* Free all the Tx ring sk_buffs */
4298 for (i = 0; i < tx_ring->count; i++) {
4299 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4300 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4301 }
4302
4303 netdev_tx_reset_queue(txring_txq(tx_ring));
4304
4305 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4306 memset(tx_ring->tx_buffer_info, 0, size);
4307
4308 /* Zero out the descriptor ring */
4309 memset(tx_ring->desc, 0, tx_ring->size);
4310
4311 tx_ring->next_to_use = 0;
4312 tx_ring->next_to_clean = 0;
4313 }
4314
4315 /**
4316 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4317 * @adapter: board private structure
4318 **/
4319 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4320 {
4321 int i;
4322
4323 for (i = 0; i < adapter->num_rx_queues; i++)
4324 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4325 }
4326
4327 /**
4328 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4329 * @adapter: board private structure
4330 **/
4331 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4332 {
4333 int i;
4334
4335 for (i = 0; i < adapter->num_tx_queues; i++)
4336 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4337 }
4338
4339 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4340 {
4341 struct hlist_node *node, *node2;
4342 struct ixgbe_fdir_filter *filter;
4343
4344 spin_lock(&adapter->fdir_perfect_lock);
4345
4346 hlist_for_each_entry_safe(filter, node, node2,
4347 &adapter->fdir_filter_list, fdir_node) {
4348 hlist_del(&filter->fdir_node);
4349 kfree(filter);
4350 }
4351 adapter->fdir_filter_count = 0;
4352
4353 spin_unlock(&adapter->fdir_perfect_lock);
4354 }
4355
4356 void ixgbe_down(struct ixgbe_adapter *adapter)
4357 {
4358 struct net_device *netdev = adapter->netdev;
4359 struct ixgbe_hw *hw = &adapter->hw;
4360 u32 rxctrl;
4361 int i;
4362
4363 /* signal that we are down to the interrupt handler */
4364 set_bit(__IXGBE_DOWN, &adapter->state);
4365
4366 /* disable receives */
4367 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4368 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4369
4370 /* disable all enabled rx queues */
4371 for (i = 0; i < adapter->num_rx_queues; i++)
4372 /* this call also flushes the previous write */
4373 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4374
4375 usleep_range(10000, 20000);
4376
4377 netif_tx_stop_all_queues(netdev);
4378
4379 /* call carrier off first to avoid false dev_watchdog timeouts */
4380 netif_carrier_off(netdev);
4381 netif_tx_disable(netdev);
4382
4383 ixgbe_irq_disable(adapter);
4384
4385 ixgbe_napi_disable_all(adapter);
4386
4387 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4388 IXGBE_FLAG2_RESET_REQUESTED);
4389 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4390
4391 del_timer_sync(&adapter->service_timer);
4392
4393 if (adapter->num_vfs) {
4394 /* Clear EITR Select mapping */
4395 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4396
4397 /* Mark all the VFs as inactive */
4398 for (i = 0 ; i < adapter->num_vfs; i++)
4399 adapter->vfinfo[i].clear_to_send = false;
4400
4401 /* ping all the active vfs to let them know we are going down */
4402 ixgbe_ping_all_vfs(adapter);
4403
4404 /* Disable all VFTE/VFRE TX/RX */
4405 ixgbe_disable_tx_rx(adapter);
4406 }
4407
4408 /* disable transmits in the hardware now that interrupts are off */
4409 for (i = 0; i < adapter->num_tx_queues; i++) {
4410 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4411 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4412 }
4413
4414 /* Disable the Tx DMA engine on 82599 and X540 */
4415 switch (hw->mac.type) {
4416 case ixgbe_mac_82599EB:
4417 case ixgbe_mac_X540:
4418 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4419 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4420 ~IXGBE_DMATXCTL_TE));
4421 break;
4422 default:
4423 break;
4424 }
4425
4426 if (!pci_channel_offline(adapter->pdev))
4427 ixgbe_reset(adapter);
4428
4429 /* power down the optics for 82599 SFP+ fiber */
4430 if (hw->mac.ops.disable_tx_laser)
4431 hw->mac.ops.disable_tx_laser(hw);
4432
4433 ixgbe_clean_all_tx_rings(adapter);
4434 ixgbe_clean_all_rx_rings(adapter);
4435
4436 #ifdef CONFIG_IXGBE_DCA
4437 /* since we reset the hardware DCA settings were cleared */
4438 ixgbe_setup_dca(adapter);
4439 #endif
4440 }
4441
4442 /**
4443 * ixgbe_tx_timeout - Respond to a Tx Hang
4444 * @netdev: network interface device structure
4445 **/
4446 static void ixgbe_tx_timeout(struct net_device *netdev)
4447 {
4448 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4449
4450 /* Do the reset outside of interrupt context */
4451 ixgbe_tx_timeout_reset(adapter);
4452 }
4453
4454 /**
4455 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4456 * @adapter: board private structure to initialize
4457 *
4458 * ixgbe_sw_init initializes the Adapter private data structure.
4459 * Fields are initialized based on PCI device information and
4460 * OS network device settings (MTU size).
4461 **/
4462 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4463 {
4464 struct ixgbe_hw *hw = &adapter->hw;
4465 struct pci_dev *pdev = adapter->pdev;
4466 unsigned int rss;
4467 #ifdef CONFIG_IXGBE_DCB
4468 int j;
4469 struct tc_configuration *tc;
4470 #endif
4471
4472 /* PCI config space info */
4473
4474 hw->vendor_id = pdev->vendor;
4475 hw->device_id = pdev->device;
4476 hw->revision_id = pdev->revision;
4477 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4478 hw->subsystem_device_id = pdev->subsystem_device;
4479
4480 /* Set capability flags */
4481 rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4482 adapter->ring_feature[RING_F_RSS].limit = rss;
4483 switch (hw->mac.type) {
4484 case ixgbe_mac_82598EB:
4485 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4486 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4487 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
4488 break;
4489 case ixgbe_mac_X540:
4490 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4491 case ixgbe_mac_82599EB:
4492 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
4493 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4494 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4495 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4496 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4497 /* Flow Director hash filters enabled */
4498 adapter->atr_sample_rate = 20;
4499 adapter->ring_feature[RING_F_FDIR].limit =
4500 IXGBE_MAX_FDIR_INDICES;
4501 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4502 #ifdef IXGBE_FCOE
4503 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4504 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4505 #ifdef CONFIG_IXGBE_DCB
4506 /* Default traffic class to use for FCoE */
4507 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4508 #endif
4509 #endif /* IXGBE_FCOE */
4510 break;
4511 default:
4512 break;
4513 }
4514
4515 #ifdef IXGBE_FCOE
4516 /* FCoE support exists, always init the FCoE lock */
4517 spin_lock_init(&adapter->fcoe.lock);
4518
4519 #endif
4520 /* n-tuple support exists, always init our spinlock */
4521 spin_lock_init(&adapter->fdir_perfect_lock);
4522
4523 #ifdef CONFIG_IXGBE_DCB
4524 switch (hw->mac.type) {
4525 case ixgbe_mac_X540:
4526 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4527 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4528 break;
4529 default:
4530 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4531 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4532 break;
4533 }
4534
4535 /* Configure DCB traffic classes */
4536 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4537 tc = &adapter->dcb_cfg.tc_config[j];
4538 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4539 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4540 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4541 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4542 tc->dcb_pfc = pfc_disabled;
4543 }
4544
4545 /* Initialize default user to priority mapping, UPx->TC0 */
4546 tc = &adapter->dcb_cfg.tc_config[0];
4547 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
4548 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
4549
4550 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4551 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4552 adapter->dcb_cfg.pfc_mode_enable = false;
4553 adapter->dcb_set_bitmap = 0x00;
4554 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4555 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
4556 sizeof(adapter->temp_dcb_cfg));
4557
4558 #endif
4559
4560 /* default flow control settings */
4561 hw->fc.requested_mode = ixgbe_fc_full;
4562 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4563 ixgbe_pbthresh_setup(adapter);
4564 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4565 hw->fc.send_xon = true;
4566 hw->fc.disable_fc_autoneg =
4567 (ixgbe_device_supports_autoneg_fc(hw) == 0) ? false : true;
4568
4569 #ifdef CONFIG_PCI_IOV
4570 /* assign number of SR-IOV VFs */
4571 if (hw->mac.type != ixgbe_mac_82598EB)
4572 adapter->num_vfs = (max_vfs > 63) ? 0 : max_vfs;
4573
4574 #endif
4575 /* enable itr by default in dynamic mode */
4576 adapter->rx_itr_setting = 1;
4577 adapter->tx_itr_setting = 1;
4578
4579 /* set default ring sizes */
4580 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4581 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4582
4583 /* set default work limits */
4584 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4585
4586 /* initialize eeprom parameters */
4587 if (ixgbe_init_eeprom_params_generic(hw)) {
4588 e_dev_err("EEPROM initialization failed\n");
4589 return -EIO;
4590 }
4591
4592 set_bit(__IXGBE_DOWN, &adapter->state);
4593
4594 return 0;
4595 }
4596
4597 /**
4598 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4599 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4600 *
4601 * Return 0 on success, negative on failure
4602 **/
4603 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4604 {
4605 struct device *dev = tx_ring->dev;
4606 int orig_node = dev_to_node(dev);
4607 int numa_node = -1;
4608 int size;
4609
4610 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4611
4612 if (tx_ring->q_vector)
4613 numa_node = tx_ring->q_vector->numa_node;
4614
4615 tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
4616 if (!tx_ring->tx_buffer_info)
4617 tx_ring->tx_buffer_info = vzalloc(size);
4618 if (!tx_ring->tx_buffer_info)
4619 goto err;
4620
4621 /* round up to nearest 4K */
4622 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4623 tx_ring->size = ALIGN(tx_ring->size, 4096);
4624
4625 set_dev_node(dev, numa_node);
4626 tx_ring->desc = dma_alloc_coherent(dev,
4627 tx_ring->size,
4628 &tx_ring->dma,
4629 GFP_KERNEL);
4630 set_dev_node(dev, orig_node);
4631 if (!tx_ring->desc)
4632 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4633 &tx_ring->dma, GFP_KERNEL);
4634 if (!tx_ring->desc)
4635 goto err;
4636
4637 tx_ring->next_to_use = 0;
4638 tx_ring->next_to_clean = 0;
4639 return 0;
4640
4641 err:
4642 vfree(tx_ring->tx_buffer_info);
4643 tx_ring->tx_buffer_info = NULL;
4644 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4645 return -ENOMEM;
4646 }
4647
4648 /**
4649 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4650 * @adapter: board private structure
4651 *
4652 * If this function returns with an error, then it's possible one or
4653 * more of the rings is populated (while the rest are not). It is the
4654 * callers duty to clean those orphaned rings.
4655 *
4656 * Return 0 on success, negative on failure
4657 **/
4658 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4659 {
4660 int i, err = 0;
4661
4662 for (i = 0; i < adapter->num_tx_queues; i++) {
4663 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4664 if (!err)
4665 continue;
4666
4667 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4668 goto err_setup_tx;
4669 }
4670
4671 return 0;
4672 err_setup_tx:
4673 /* rewind the index freeing the rings as we go */
4674 while (i--)
4675 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4676 return err;
4677 }
4678
4679 /**
4680 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4681 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4682 *
4683 * Returns 0 on success, negative on failure
4684 **/
4685 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4686 {
4687 struct device *dev = rx_ring->dev;
4688 int orig_node = dev_to_node(dev);
4689 int numa_node = -1;
4690 int size;
4691
4692 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4693
4694 if (rx_ring->q_vector)
4695 numa_node = rx_ring->q_vector->numa_node;
4696
4697 rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
4698 if (!rx_ring->rx_buffer_info)
4699 rx_ring->rx_buffer_info = vzalloc(size);
4700 if (!rx_ring->rx_buffer_info)
4701 goto err;
4702
4703 /* Round up to nearest 4K */
4704 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4705 rx_ring->size = ALIGN(rx_ring->size, 4096);
4706
4707 set_dev_node(dev, numa_node);
4708 rx_ring->desc = dma_alloc_coherent(dev,
4709 rx_ring->size,
4710 &rx_ring->dma,
4711 GFP_KERNEL);
4712 set_dev_node(dev, orig_node);
4713 if (!rx_ring->desc)
4714 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4715 &rx_ring->dma, GFP_KERNEL);
4716 if (!rx_ring->desc)
4717 goto err;
4718
4719 rx_ring->next_to_clean = 0;
4720 rx_ring->next_to_use = 0;
4721
4722 return 0;
4723 err:
4724 vfree(rx_ring->rx_buffer_info);
4725 rx_ring->rx_buffer_info = NULL;
4726 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4727 return -ENOMEM;
4728 }
4729
4730 /**
4731 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4732 * @adapter: board private structure
4733 *
4734 * If this function returns with an error, then it's possible one or
4735 * more of the rings is populated (while the rest are not). It is the
4736 * callers duty to clean those orphaned rings.
4737 *
4738 * Return 0 on success, negative on failure
4739 **/
4740 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4741 {
4742 int i, err = 0;
4743
4744 for (i = 0; i < adapter->num_rx_queues; i++) {
4745 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
4746 if (!err)
4747 continue;
4748
4749 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4750 goto err_setup_rx;
4751 }
4752
4753 #ifdef IXGBE_FCOE
4754 err = ixgbe_setup_fcoe_ddp_resources(adapter);
4755 if (!err)
4756 #endif
4757 return 0;
4758 err_setup_rx:
4759 /* rewind the index freeing the rings as we go */
4760 while (i--)
4761 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4762 return err;
4763 }
4764
4765 /**
4766 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4767 * @tx_ring: Tx descriptor ring for a specific queue
4768 *
4769 * Free all transmit software resources
4770 **/
4771 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
4772 {
4773 ixgbe_clean_tx_ring(tx_ring);
4774
4775 vfree(tx_ring->tx_buffer_info);
4776 tx_ring->tx_buffer_info = NULL;
4777
4778 /* if not set, then don't free */
4779 if (!tx_ring->desc)
4780 return;
4781
4782 dma_free_coherent(tx_ring->dev, tx_ring->size,
4783 tx_ring->desc, tx_ring->dma);
4784
4785 tx_ring->desc = NULL;
4786 }
4787
4788 /**
4789 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4790 * @adapter: board private structure
4791 *
4792 * Free all transmit software resources
4793 **/
4794 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4795 {
4796 int i;
4797
4798 for (i = 0; i < adapter->num_tx_queues; i++)
4799 if (adapter->tx_ring[i]->desc)
4800 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4801 }
4802
4803 /**
4804 * ixgbe_free_rx_resources - Free Rx Resources
4805 * @rx_ring: ring to clean the resources from
4806 *
4807 * Free all receive software resources
4808 **/
4809 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
4810 {
4811 ixgbe_clean_rx_ring(rx_ring);
4812
4813 vfree(rx_ring->rx_buffer_info);
4814 rx_ring->rx_buffer_info = NULL;
4815
4816 /* if not set, then don't free */
4817 if (!rx_ring->desc)
4818 return;
4819
4820 dma_free_coherent(rx_ring->dev, rx_ring->size,
4821 rx_ring->desc, rx_ring->dma);
4822
4823 rx_ring->desc = NULL;
4824 }
4825
4826 /**
4827 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4828 * @adapter: board private structure
4829 *
4830 * Free all receive software resources
4831 **/
4832 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4833 {
4834 int i;
4835
4836 #ifdef IXGBE_FCOE
4837 ixgbe_free_fcoe_ddp_resources(adapter);
4838
4839 #endif
4840 for (i = 0; i < adapter->num_rx_queues; i++)
4841 if (adapter->rx_ring[i]->desc)
4842 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4843 }
4844
4845 /**
4846 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4847 * @netdev: network interface device structure
4848 * @new_mtu: new value for maximum frame size
4849 *
4850 * Returns 0 on success, negative on failure
4851 **/
4852 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4853 {
4854 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4855 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4856
4857 /* MTU < 68 is an error and causes problems on some kernels */
4858 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4859 return -EINVAL;
4860
4861 /*
4862 * For 82599EB we cannot allow legacy VFs to enable their receive
4863 * paths when MTU greater than 1500 is configured. So display a
4864 * warning that legacy VFs will be disabled.
4865 */
4866 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
4867 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
4868 (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
4869 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
4870
4871 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4872
4873 /* must set new MTU before calling down or up */
4874 netdev->mtu = new_mtu;
4875
4876 if (netif_running(netdev))
4877 ixgbe_reinit_locked(adapter);
4878
4879 return 0;
4880 }
4881
4882 /**
4883 * ixgbe_open - Called when a network interface is made active
4884 * @netdev: network interface device structure
4885 *
4886 * Returns 0 on success, negative value on failure
4887 *
4888 * The open entry point is called when a network interface is made
4889 * active by the system (IFF_UP). At this point all resources needed
4890 * for transmit and receive operations are allocated, the interrupt
4891 * handler is registered with the OS, the watchdog timer is started,
4892 * and the stack is notified that the interface is ready.
4893 **/
4894 static int ixgbe_open(struct net_device *netdev)
4895 {
4896 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4897 int err;
4898
4899 /* disallow open during test */
4900 if (test_bit(__IXGBE_TESTING, &adapter->state))
4901 return -EBUSY;
4902
4903 netif_carrier_off(netdev);
4904
4905 /* allocate transmit descriptors */
4906 err = ixgbe_setup_all_tx_resources(adapter);
4907 if (err)
4908 goto err_setup_tx;
4909
4910 /* allocate receive descriptors */
4911 err = ixgbe_setup_all_rx_resources(adapter);
4912 if (err)
4913 goto err_setup_rx;
4914
4915 ixgbe_configure(adapter);
4916
4917 err = ixgbe_request_irq(adapter);
4918 if (err)
4919 goto err_req_irq;
4920
4921 /* Notify the stack of the actual queue counts. */
4922 err = netif_set_real_num_tx_queues(netdev,
4923 adapter->num_rx_pools > 1 ? 1 :
4924 adapter->num_tx_queues);
4925 if (err)
4926 goto err_set_queues;
4927
4928
4929 err = netif_set_real_num_rx_queues(netdev,
4930 adapter->num_rx_pools > 1 ? 1 :
4931 adapter->num_rx_queues);
4932 if (err)
4933 goto err_set_queues;
4934
4935 ixgbe_ptp_init(adapter);
4936
4937 ixgbe_up_complete(adapter);
4938
4939 return 0;
4940
4941 err_set_queues:
4942 ixgbe_free_irq(adapter);
4943 err_req_irq:
4944 ixgbe_free_all_rx_resources(adapter);
4945 err_setup_rx:
4946 ixgbe_free_all_tx_resources(adapter);
4947 err_setup_tx:
4948 ixgbe_reset(adapter);
4949
4950 return err;
4951 }
4952
4953 /**
4954 * ixgbe_close - Disables a network interface
4955 * @netdev: network interface device structure
4956 *
4957 * Returns 0, this is not allowed to fail
4958 *
4959 * The close entry point is called when an interface is de-activated
4960 * by the OS. The hardware is still under the drivers control, but
4961 * needs to be disabled. A global MAC reset is issued to stop the
4962 * hardware, and all transmit and receive resources are freed.
4963 **/
4964 static int ixgbe_close(struct net_device *netdev)
4965 {
4966 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4967
4968 ixgbe_ptp_stop(adapter);
4969
4970 ixgbe_down(adapter);
4971 ixgbe_free_irq(adapter);
4972
4973 ixgbe_fdir_filter_exit(adapter);
4974
4975 ixgbe_free_all_tx_resources(adapter);
4976 ixgbe_free_all_rx_resources(adapter);
4977
4978 ixgbe_release_hw_control(adapter);
4979
4980 return 0;
4981 }
4982
4983 #ifdef CONFIG_PM
4984 static int ixgbe_resume(struct pci_dev *pdev)
4985 {
4986 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4987 struct net_device *netdev = adapter->netdev;
4988 u32 err;
4989
4990 pci_set_power_state(pdev, PCI_D0);
4991 pci_restore_state(pdev);
4992 /*
4993 * pci_restore_state clears dev->state_saved so call
4994 * pci_save_state to restore it.
4995 */
4996 pci_save_state(pdev);
4997
4998 err = pci_enable_device_mem(pdev);
4999 if (err) {
5000 e_dev_err("Cannot enable PCI device from suspend\n");
5001 return err;
5002 }
5003 pci_set_master(pdev);
5004
5005 pci_wake_from_d3(pdev, false);
5006
5007 ixgbe_reset(adapter);
5008
5009 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5010
5011 rtnl_lock();
5012 err = ixgbe_init_interrupt_scheme(adapter);
5013 if (!err && netif_running(netdev))
5014 err = ixgbe_open(netdev);
5015
5016 rtnl_unlock();
5017
5018 if (err)
5019 return err;
5020
5021 netif_device_attach(netdev);
5022
5023 return 0;
5024 }
5025 #endif /* CONFIG_PM */
5026
5027 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5028 {
5029 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5030 struct net_device *netdev = adapter->netdev;
5031 struct ixgbe_hw *hw = &adapter->hw;
5032 u32 ctrl, fctrl;
5033 u32 wufc = adapter->wol;
5034 #ifdef CONFIG_PM
5035 int retval = 0;
5036 #endif
5037
5038 netif_device_detach(netdev);
5039
5040 if (netif_running(netdev)) {
5041 rtnl_lock();
5042 ixgbe_down(adapter);
5043 ixgbe_free_irq(adapter);
5044 ixgbe_free_all_tx_resources(adapter);
5045 ixgbe_free_all_rx_resources(adapter);
5046 rtnl_unlock();
5047 }
5048
5049 ixgbe_clear_interrupt_scheme(adapter);
5050
5051 #ifdef CONFIG_PM
5052 retval = pci_save_state(pdev);
5053 if (retval)
5054 return retval;
5055
5056 #endif
5057 if (wufc) {
5058 ixgbe_set_rx_mode(netdev);
5059
5060 /* enable the optics for 82599 SFP+ fiber as we can WoL */
5061 if (hw->mac.ops.enable_tx_laser)
5062 hw->mac.ops.enable_tx_laser(hw);
5063
5064 /* turn on all-multi mode if wake on multicast is enabled */
5065 if (wufc & IXGBE_WUFC_MC) {
5066 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5067 fctrl |= IXGBE_FCTRL_MPE;
5068 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5069 }
5070
5071 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5072 ctrl |= IXGBE_CTRL_GIO_DIS;
5073 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5074
5075 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5076 } else {
5077 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5078 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5079 }
5080
5081 switch (hw->mac.type) {
5082 case ixgbe_mac_82598EB:
5083 pci_wake_from_d3(pdev, false);
5084 break;
5085 case ixgbe_mac_82599EB:
5086 case ixgbe_mac_X540:
5087 pci_wake_from_d3(pdev, !!wufc);
5088 break;
5089 default:
5090 break;
5091 }
5092
5093 *enable_wake = !!wufc;
5094
5095 ixgbe_release_hw_control(adapter);
5096
5097 pci_disable_device(pdev);
5098
5099 return 0;
5100 }
5101
5102 #ifdef CONFIG_PM
5103 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5104 {
5105 int retval;
5106 bool wake;
5107
5108 retval = __ixgbe_shutdown(pdev, &wake);
5109 if (retval)
5110 return retval;
5111
5112 if (wake) {
5113 pci_prepare_to_sleep(pdev);
5114 } else {
5115 pci_wake_from_d3(pdev, false);
5116 pci_set_power_state(pdev, PCI_D3hot);
5117 }
5118
5119 return 0;
5120 }
5121 #endif /* CONFIG_PM */
5122
5123 static void ixgbe_shutdown(struct pci_dev *pdev)
5124 {
5125 bool wake;
5126
5127 __ixgbe_shutdown(pdev, &wake);
5128
5129 if (system_state == SYSTEM_POWER_OFF) {
5130 pci_wake_from_d3(pdev, wake);
5131 pci_set_power_state(pdev, PCI_D3hot);
5132 }
5133 }
5134
5135 /**
5136 * ixgbe_update_stats - Update the board statistics counters.
5137 * @adapter: board private structure
5138 **/
5139 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5140 {
5141 struct net_device *netdev = adapter->netdev;
5142 struct ixgbe_hw *hw = &adapter->hw;
5143 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5144 u64 total_mpc = 0;
5145 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5146 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5147 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5148 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5149
5150 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5151 test_bit(__IXGBE_RESETTING, &adapter->state))
5152 return;
5153
5154 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5155 u64 rsc_count = 0;
5156 u64 rsc_flush = 0;
5157 for (i = 0; i < adapter->num_rx_queues; i++) {
5158 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5159 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5160 }
5161 adapter->rsc_total_count = rsc_count;
5162 adapter->rsc_total_flush = rsc_flush;
5163 }
5164
5165 for (i = 0; i < adapter->num_rx_queues; i++) {
5166 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5167 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5168 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5169 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5170 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5171 bytes += rx_ring->stats.bytes;
5172 packets += rx_ring->stats.packets;
5173 }
5174 adapter->non_eop_descs = non_eop_descs;
5175 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5176 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5177 adapter->hw_csum_rx_error = hw_csum_rx_error;
5178 netdev->stats.rx_bytes = bytes;
5179 netdev->stats.rx_packets = packets;
5180
5181 bytes = 0;
5182 packets = 0;
5183 /* gather some stats to the adapter struct that are per queue */
5184 for (i = 0; i < adapter->num_tx_queues; i++) {
5185 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5186 restart_queue += tx_ring->tx_stats.restart_queue;
5187 tx_busy += tx_ring->tx_stats.tx_busy;
5188 bytes += tx_ring->stats.bytes;
5189 packets += tx_ring->stats.packets;
5190 }
5191 adapter->restart_queue = restart_queue;
5192 adapter->tx_busy = tx_busy;
5193 netdev->stats.tx_bytes = bytes;
5194 netdev->stats.tx_packets = packets;
5195
5196 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5197
5198 /* 8 register reads */
5199 for (i = 0; i < 8; i++) {
5200 /* for packet buffers not used, the register should read 0 */
5201 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5202 missed_rx += mpc;
5203 hwstats->mpc[i] += mpc;
5204 total_mpc += hwstats->mpc[i];
5205 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5206 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5207 switch (hw->mac.type) {
5208 case ixgbe_mac_82598EB:
5209 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5210 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5211 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5212 hwstats->pxonrxc[i] +=
5213 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5214 break;
5215 case ixgbe_mac_82599EB:
5216 case ixgbe_mac_X540:
5217 hwstats->pxonrxc[i] +=
5218 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5219 break;
5220 default:
5221 break;
5222 }
5223 }
5224
5225 /*16 register reads */
5226 for (i = 0; i < 16; i++) {
5227 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5228 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5229 if ((hw->mac.type == ixgbe_mac_82599EB) ||
5230 (hw->mac.type == ixgbe_mac_X540)) {
5231 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5232 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5233 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5234 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5235 }
5236 }
5237
5238 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5239 /* work around hardware counting issue */
5240 hwstats->gprc -= missed_rx;
5241
5242 ixgbe_update_xoff_received(adapter);
5243
5244 /* 82598 hardware only has a 32 bit counter in the high register */
5245 switch (hw->mac.type) {
5246 case ixgbe_mac_82598EB:
5247 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5248 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5249 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5250 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5251 break;
5252 case ixgbe_mac_X540:
5253 /* OS2BMC stats are X540 only*/
5254 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5255 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5256 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5257 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5258 case ixgbe_mac_82599EB:
5259 for (i = 0; i < 16; i++)
5260 adapter->hw_rx_no_dma_resources +=
5261 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5262 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5263 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5264 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5265 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5266 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5267 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5268 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5269 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5270 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5271 #ifdef IXGBE_FCOE
5272 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5273 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5274 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5275 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5276 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5277 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5278 /* Add up per cpu counters for total ddp aloc fail */
5279 if (adapter->fcoe.ddp_pool) {
5280 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5281 struct ixgbe_fcoe_ddp_pool *ddp_pool;
5282 unsigned int cpu;
5283 u64 noddp = 0, noddp_ext_buff = 0;
5284 for_each_possible_cpu(cpu) {
5285 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
5286 noddp += ddp_pool->noddp;
5287 noddp_ext_buff += ddp_pool->noddp_ext_buff;
5288 }
5289 hwstats->fcoe_noddp = noddp;
5290 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
5291 }
5292 #endif /* IXGBE_FCOE */
5293 break;
5294 default:
5295 break;
5296 }
5297 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5298 hwstats->bprc += bprc;
5299 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5300 if (hw->mac.type == ixgbe_mac_82598EB)
5301 hwstats->mprc -= bprc;
5302 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5303 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5304 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5305 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5306 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5307 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5308 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5309 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5310 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5311 hwstats->lxontxc += lxon;
5312 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5313 hwstats->lxofftxc += lxoff;
5314 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5315 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5316 /*
5317 * 82598 errata - tx of flow control packets is included in tx counters
5318 */
5319 xon_off_tot = lxon + lxoff;
5320 hwstats->gptc -= xon_off_tot;
5321 hwstats->mptc -= xon_off_tot;
5322 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5323 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5324 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5325 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5326 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5327 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5328 hwstats->ptc64 -= xon_off_tot;
5329 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5330 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5331 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5332 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5333 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5334 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5335
5336 /* Fill out the OS statistics structure */
5337 netdev->stats.multicast = hwstats->mprc;
5338
5339 /* Rx Errors */
5340 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5341 netdev->stats.rx_dropped = 0;
5342 netdev->stats.rx_length_errors = hwstats->rlec;
5343 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5344 netdev->stats.rx_missed_errors = total_mpc;
5345 }
5346
5347 /**
5348 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5349 * @adapter: pointer to the device adapter structure
5350 **/
5351 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5352 {
5353 struct ixgbe_hw *hw = &adapter->hw;
5354 int i;
5355
5356 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5357 return;
5358
5359 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5360
5361 /* if interface is down do nothing */
5362 if (test_bit(__IXGBE_DOWN, &adapter->state))
5363 return;
5364
5365 /* do nothing if we are not using signature filters */
5366 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5367 return;
5368
5369 adapter->fdir_overflow++;
5370
5371 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5372 for (i = 0; i < adapter->num_tx_queues; i++)
5373 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5374 &(adapter->tx_ring[i]->state));
5375 /* re-enable flow director interrupts */
5376 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5377 } else {
5378 e_err(probe, "failed to finish FDIR re-initialization, "
5379 "ignored adding FDIR ATR filters\n");
5380 }
5381 }
5382
5383 /**
5384 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5385 * @adapter: pointer to the device adapter structure
5386 *
5387 * This function serves two purposes. First it strobes the interrupt lines
5388 * in order to make certain interrupts are occurring. Secondly it sets the
5389 * bits needed to check for TX hangs. As a result we should immediately
5390 * determine if a hang has occurred.
5391 */
5392 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5393 {
5394 struct ixgbe_hw *hw = &adapter->hw;
5395 u64 eics = 0;
5396 int i;
5397
5398 /* If we're down or resetting, just bail */
5399 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5400 test_bit(__IXGBE_RESETTING, &adapter->state))
5401 return;
5402
5403 /* Force detection of hung controller */
5404 if (netif_carrier_ok(adapter->netdev)) {
5405 for (i = 0; i < adapter->num_tx_queues; i++)
5406 set_check_for_tx_hang(adapter->tx_ring[i]);
5407 }
5408
5409 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5410 /*
5411 * for legacy and MSI interrupts don't set any bits
5412 * that are enabled for EIAM, because this operation
5413 * would set *both* EIMS and EICS for any bit in EIAM
5414 */
5415 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5416 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5417 } else {
5418 /* get one bit for every active tx/rx interrupt vector */
5419 for (i = 0; i < adapter->num_q_vectors; i++) {
5420 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5421 if (qv->rx.ring || qv->tx.ring)
5422 eics |= ((u64)1 << i);
5423 }
5424 }
5425
5426 /* Cause software interrupt to ensure rings are cleaned */
5427 ixgbe_irq_rearm_queues(adapter, eics);
5428
5429 }
5430
5431 /**
5432 * ixgbe_watchdog_update_link - update the link status
5433 * @adapter: pointer to the device adapter structure
5434 * @link_speed: pointer to a u32 to store the link_speed
5435 **/
5436 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5437 {
5438 struct ixgbe_hw *hw = &adapter->hw;
5439 u32 link_speed = adapter->link_speed;
5440 bool link_up = adapter->link_up;
5441 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
5442
5443 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5444 return;
5445
5446 if (hw->mac.ops.check_link) {
5447 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5448 } else {
5449 /* always assume link is up, if no check link function */
5450 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5451 link_up = true;
5452 }
5453
5454 if (adapter->ixgbe_ieee_pfc)
5455 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
5456
5457 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
5458 hw->mac.ops.fc_enable(hw);
5459 ixgbe_set_rx_drop_en(adapter);
5460 }
5461
5462 if (link_up ||
5463 time_after(jiffies, (adapter->link_check_timeout +
5464 IXGBE_TRY_LINK_TIMEOUT))) {
5465 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5466 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5467 IXGBE_WRITE_FLUSH(hw);
5468 }
5469
5470 adapter->link_up = link_up;
5471 adapter->link_speed = link_speed;
5472 }
5473
5474 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
5475 {
5476 #ifdef CONFIG_IXGBE_DCB
5477 struct net_device *netdev = adapter->netdev;
5478 struct dcb_app app = {
5479 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
5480 .protocol = 0,
5481 };
5482 u8 up = 0;
5483
5484 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
5485 up = dcb_ieee_getapp_mask(netdev, &app);
5486
5487 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
5488 #endif
5489 }
5490
5491 /**
5492 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5493 * print link up message
5494 * @adapter: pointer to the device adapter structure
5495 **/
5496 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5497 {
5498 struct net_device *netdev = adapter->netdev;
5499 struct ixgbe_hw *hw = &adapter->hw;
5500 u32 link_speed = adapter->link_speed;
5501 bool flow_rx, flow_tx;
5502
5503 /* only continue if link was previously down */
5504 if (netif_carrier_ok(netdev))
5505 return;
5506
5507 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5508
5509 switch (hw->mac.type) {
5510 case ixgbe_mac_82598EB: {
5511 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5512 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5513 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5514 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5515 }
5516 break;
5517 case ixgbe_mac_X540:
5518 case ixgbe_mac_82599EB: {
5519 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5520 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5521 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5522 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5523 }
5524 break;
5525 default:
5526 flow_tx = false;
5527 flow_rx = false;
5528 break;
5529 }
5530
5531 if (adapter->flags2 & IXGBE_FLAG2_PTP_ENABLED)
5532 ixgbe_ptp_start_cyclecounter(adapter);
5533
5534 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5535 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5536 "10 Gbps" :
5537 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5538 "1 Gbps" :
5539 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5540 "100 Mbps" :
5541 "unknown speed"))),
5542 ((flow_rx && flow_tx) ? "RX/TX" :
5543 (flow_rx ? "RX" :
5544 (flow_tx ? "TX" : "None"))));
5545
5546 netif_carrier_on(netdev);
5547 ixgbe_check_vf_rate_limit(adapter);
5548
5549 /* update the default user priority for VFs */
5550 ixgbe_update_default_up(adapter);
5551
5552 /* ping all the active vfs to let them know link has changed */
5553 ixgbe_ping_all_vfs(adapter);
5554 }
5555
5556 /**
5557 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5558 * print link down message
5559 * @adapter: pointer to the adapter structure
5560 **/
5561 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
5562 {
5563 struct net_device *netdev = adapter->netdev;
5564 struct ixgbe_hw *hw = &adapter->hw;
5565
5566 adapter->link_up = false;
5567 adapter->link_speed = 0;
5568
5569 /* only continue if link was up previously */
5570 if (!netif_carrier_ok(netdev))
5571 return;
5572
5573 /* poll for SFP+ cable when link is down */
5574 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5575 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5576
5577 if (adapter->flags2 & IXGBE_FLAG2_PTP_ENABLED)
5578 ixgbe_ptp_start_cyclecounter(adapter);
5579
5580 e_info(drv, "NIC Link is Down\n");
5581 netif_carrier_off(netdev);
5582
5583 /* ping all the active vfs to let them know link has changed */
5584 ixgbe_ping_all_vfs(adapter);
5585 }
5586
5587 /**
5588 * ixgbe_watchdog_flush_tx - flush queues on link down
5589 * @adapter: pointer to the device adapter structure
5590 **/
5591 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5592 {
5593 int i;
5594 int some_tx_pending = 0;
5595
5596 if (!netif_carrier_ok(adapter->netdev)) {
5597 for (i = 0; i < adapter->num_tx_queues; i++) {
5598 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5599 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5600 some_tx_pending = 1;
5601 break;
5602 }
5603 }
5604
5605 if (some_tx_pending) {
5606 /* We've lost link, so the controller stops DMA,
5607 * but we've got queued Tx work that's never going
5608 * to get done, so reset controller to flush Tx.
5609 * (Do the reset outside of interrupt context).
5610 */
5611 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5612 }
5613 }
5614 }
5615
5616 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5617 {
5618 u32 ssvpc;
5619
5620 /* Do not perform spoof check for 82598 or if not in IOV mode */
5621 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
5622 adapter->num_vfs == 0)
5623 return;
5624
5625 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5626
5627 /*
5628 * ssvpc register is cleared on read, if zero then no
5629 * spoofed packets in the last interval.
5630 */
5631 if (!ssvpc)
5632 return;
5633
5634 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
5635 }
5636
5637 /**
5638 * ixgbe_watchdog_subtask - check and bring link up
5639 * @adapter: pointer to the device adapter structure
5640 **/
5641 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5642 {
5643 /* if interface is down do nothing */
5644 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5645 test_bit(__IXGBE_RESETTING, &adapter->state))
5646 return;
5647
5648 ixgbe_watchdog_update_link(adapter);
5649
5650 if (adapter->link_up)
5651 ixgbe_watchdog_link_is_up(adapter);
5652 else
5653 ixgbe_watchdog_link_is_down(adapter);
5654
5655 ixgbe_spoof_check(adapter);
5656 ixgbe_update_stats(adapter);
5657
5658 ixgbe_watchdog_flush_tx(adapter);
5659 }
5660
5661 /**
5662 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5663 * @adapter: the ixgbe adapter structure
5664 **/
5665 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5666 {
5667 struct ixgbe_hw *hw = &adapter->hw;
5668 s32 err;
5669
5670 /* not searching for SFP so there is nothing to do here */
5671 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5672 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5673 return;
5674
5675 /* someone else is in init, wait until next service event */
5676 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5677 return;
5678
5679 err = hw->phy.ops.identify_sfp(hw);
5680 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5681 goto sfp_out;
5682
5683 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5684 /* If no cable is present, then we need to reset
5685 * the next time we find a good cable. */
5686 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5687 }
5688
5689 /* exit on error */
5690 if (err)
5691 goto sfp_out;
5692
5693 /* exit if reset not needed */
5694 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5695 goto sfp_out;
5696
5697 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5698
5699 /*
5700 * A module may be identified correctly, but the EEPROM may not have
5701 * support for that module. setup_sfp() will fail in that case, so
5702 * we should not allow that module to load.
5703 */
5704 if (hw->mac.type == ixgbe_mac_82598EB)
5705 err = hw->phy.ops.reset(hw);
5706 else
5707 err = hw->mac.ops.setup_sfp(hw);
5708
5709 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5710 goto sfp_out;
5711
5712 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5713 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5714
5715 sfp_out:
5716 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5717
5718 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5719 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5720 e_dev_err("failed to initialize because an unsupported "
5721 "SFP+ module type was detected.\n");
5722 e_dev_err("Reload the driver after installing a "
5723 "supported module.\n");
5724 unregister_netdev(adapter->netdev);
5725 }
5726 }
5727
5728 /**
5729 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5730 * @adapter: the ixgbe adapter structure
5731 **/
5732 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5733 {
5734 struct ixgbe_hw *hw = &adapter->hw;
5735 u32 autoneg;
5736 bool negotiation;
5737
5738 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5739 return;
5740
5741 /* someone else is in init, wait until next service event */
5742 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5743 return;
5744
5745 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5746
5747 autoneg = hw->phy.autoneg_advertised;
5748 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5749 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5750 if (hw->mac.ops.setup_link)
5751 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5752
5753 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5754 adapter->link_check_timeout = jiffies;
5755 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5756 }
5757
5758 #ifdef CONFIG_PCI_IOV
5759 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
5760 {
5761 int vf;
5762 struct ixgbe_hw *hw = &adapter->hw;
5763 struct net_device *netdev = adapter->netdev;
5764 u32 gpc;
5765 u32 ciaa, ciad;
5766
5767 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
5768 if (gpc) /* If incrementing then no need for the check below */
5769 return;
5770 /*
5771 * Check to see if a bad DMA write target from an errant or
5772 * malicious VF has caused a PCIe error. If so then we can
5773 * issue a VFLR to the offending VF(s) and then resume without
5774 * requesting a full slot reset.
5775 */
5776
5777 for (vf = 0; vf < adapter->num_vfs; vf++) {
5778 ciaa = (vf << 16) | 0x80000000;
5779 /* 32 bit read so align, we really want status at offset 6 */
5780 ciaa |= PCI_COMMAND;
5781 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5782 ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
5783 ciaa &= 0x7FFFFFFF;
5784 /* disable debug mode asap after reading data */
5785 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5786 /* Get the upper 16 bits which will be the PCI status reg */
5787 ciad >>= 16;
5788 if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
5789 netdev_err(netdev, "VF %d Hung DMA\n", vf);
5790 /* Issue VFLR */
5791 ciaa = (vf << 16) | 0x80000000;
5792 ciaa |= 0xA8;
5793 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5794 ciad = 0x00008000; /* VFLR */
5795 IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
5796 ciaa &= 0x7FFFFFFF;
5797 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5798 }
5799 }
5800 }
5801
5802 #endif
5803 /**
5804 * ixgbe_service_timer - Timer Call-back
5805 * @data: pointer to adapter cast into an unsigned long
5806 **/
5807 static void ixgbe_service_timer(unsigned long data)
5808 {
5809 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5810 unsigned long next_event_offset;
5811 bool ready = true;
5812
5813 /* poll faster when waiting for link */
5814 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5815 next_event_offset = HZ / 10;
5816 else
5817 next_event_offset = HZ * 2;
5818
5819 #ifdef CONFIG_PCI_IOV
5820 /*
5821 * don't bother with SR-IOV VF DMA hang check if there are
5822 * no VFs or the link is down
5823 */
5824 if (!adapter->num_vfs ||
5825 (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5826 goto normal_timer_service;
5827
5828 /* If we have VFs allocated then we must check for DMA hangs */
5829 ixgbe_check_for_bad_vf(adapter);
5830 next_event_offset = HZ / 50;
5831 adapter->timer_event_accumulator++;
5832
5833 if (adapter->timer_event_accumulator >= 100)
5834 adapter->timer_event_accumulator = 0;
5835 else
5836 ready = false;
5837
5838 normal_timer_service:
5839 #endif
5840 /* Reset the timer */
5841 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5842
5843 if (ready)
5844 ixgbe_service_event_schedule(adapter);
5845 }
5846
5847 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5848 {
5849 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
5850 return;
5851
5852 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
5853
5854 /* If we're already down or resetting, just bail */
5855 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5856 test_bit(__IXGBE_RESETTING, &adapter->state))
5857 return;
5858
5859 ixgbe_dump(adapter);
5860 netdev_err(adapter->netdev, "Reset adapter\n");
5861 adapter->tx_timeout_count++;
5862
5863 ixgbe_reinit_locked(adapter);
5864 }
5865
5866 /**
5867 * ixgbe_service_task - manages and runs subtasks
5868 * @work: pointer to work_struct containing our data
5869 **/
5870 static void ixgbe_service_task(struct work_struct *work)
5871 {
5872 struct ixgbe_adapter *adapter = container_of(work,
5873 struct ixgbe_adapter,
5874 service_task);
5875
5876 ixgbe_reset_subtask(adapter);
5877 ixgbe_sfp_detection_subtask(adapter);
5878 ixgbe_sfp_link_config_subtask(adapter);
5879 ixgbe_check_overtemp_subtask(adapter);
5880 ixgbe_watchdog_subtask(adapter);
5881 ixgbe_fdir_reinit_subtask(adapter);
5882 ixgbe_check_hang_subtask(adapter);
5883 ixgbe_ptp_overflow_check(adapter);
5884
5885 ixgbe_service_event_complete(adapter);
5886 }
5887
5888 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
5889 struct ixgbe_tx_buffer *first,
5890 u8 *hdr_len)
5891 {
5892 struct sk_buff *skb = first->skb;
5893 u32 vlan_macip_lens, type_tucmd;
5894 u32 mss_l4len_idx, l4len;
5895
5896 if (!skb_is_gso(skb))
5897 return 0;
5898
5899 if (skb_header_cloned(skb)) {
5900 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5901 if (err)
5902 return err;
5903 }
5904
5905 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5906 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
5907
5908 if (first->protocol == __constant_htons(ETH_P_IP)) {
5909 struct iphdr *iph = ip_hdr(skb);
5910 iph->tot_len = 0;
5911 iph->check = 0;
5912 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5913 iph->daddr, 0,
5914 IPPROTO_TCP,
5915 0);
5916 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5917 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5918 IXGBE_TX_FLAGS_CSUM |
5919 IXGBE_TX_FLAGS_IPV4;
5920 } else if (skb_is_gso_v6(skb)) {
5921 ipv6_hdr(skb)->payload_len = 0;
5922 tcp_hdr(skb)->check =
5923 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5924 &ipv6_hdr(skb)->daddr,
5925 0, IPPROTO_TCP, 0);
5926 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5927 IXGBE_TX_FLAGS_CSUM;
5928 }
5929
5930 /* compute header lengths */
5931 l4len = tcp_hdrlen(skb);
5932 *hdr_len = skb_transport_offset(skb) + l4len;
5933
5934 /* update gso size and bytecount with header size */
5935 first->gso_segs = skb_shinfo(skb)->gso_segs;
5936 first->bytecount += (first->gso_segs - 1) * *hdr_len;
5937
5938 /* mss_l4len_id: use 1 as index for TSO */
5939 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
5940 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
5941 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
5942
5943 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
5944 vlan_macip_lens = skb_network_header_len(skb);
5945 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5946 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5947
5948 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
5949 mss_l4len_idx);
5950
5951 return 1;
5952 }
5953
5954 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
5955 struct ixgbe_tx_buffer *first)
5956 {
5957 struct sk_buff *skb = first->skb;
5958 u32 vlan_macip_lens = 0;
5959 u32 mss_l4len_idx = 0;
5960 u32 type_tucmd = 0;
5961
5962 if (skb->ip_summed != CHECKSUM_PARTIAL) {
5963 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN)) {
5964 if (unlikely(skb->no_fcs))
5965 first->tx_flags |= IXGBE_TX_FLAGS_NO_IFCS;
5966 if (!(first->tx_flags & IXGBE_TX_FLAGS_TXSW))
5967 return;
5968 }
5969 } else {
5970 u8 l4_hdr = 0;
5971 switch (first->protocol) {
5972 case __constant_htons(ETH_P_IP):
5973 vlan_macip_lens |= skb_network_header_len(skb);
5974 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5975 l4_hdr = ip_hdr(skb)->protocol;
5976 break;
5977 case __constant_htons(ETH_P_IPV6):
5978 vlan_macip_lens |= skb_network_header_len(skb);
5979 l4_hdr = ipv6_hdr(skb)->nexthdr;
5980 break;
5981 default:
5982 if (unlikely(net_ratelimit())) {
5983 dev_warn(tx_ring->dev,
5984 "partial checksum but proto=%x!\n",
5985 first->protocol);
5986 }
5987 break;
5988 }
5989
5990 switch (l4_hdr) {
5991 case IPPROTO_TCP:
5992 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5993 mss_l4len_idx = tcp_hdrlen(skb) <<
5994 IXGBE_ADVTXD_L4LEN_SHIFT;
5995 break;
5996 case IPPROTO_SCTP:
5997 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5998 mss_l4len_idx = sizeof(struct sctphdr) <<
5999 IXGBE_ADVTXD_L4LEN_SHIFT;
6000 break;
6001 case IPPROTO_UDP:
6002 mss_l4len_idx = sizeof(struct udphdr) <<
6003 IXGBE_ADVTXD_L4LEN_SHIFT;
6004 break;
6005 default:
6006 if (unlikely(net_ratelimit())) {
6007 dev_warn(tx_ring->dev,
6008 "partial checksum but l4 proto=%x!\n",
6009 l4_hdr);
6010 }
6011 break;
6012 }
6013
6014 /* update TX checksum flag */
6015 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
6016 }
6017
6018 /* vlan_macip_lens: MACLEN, VLAN tag */
6019 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6020 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6021
6022 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6023 type_tucmd, mss_l4len_idx);
6024 }
6025
6026 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
6027 {
6028 /* set type for advanced descriptor with frame checksum insertion */
6029 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
6030 IXGBE_ADVTXD_DCMD_DEXT);
6031
6032 /* set HW vlan bit if vlan is present */
6033 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
6034 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
6035
6036 if (tx_flags & IXGBE_TX_FLAGS_TSTAMP)
6037 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_MAC_TSTAMP);
6038
6039 /* set segmentation enable bits for TSO/FSO */
6040 #ifdef IXGBE_FCOE
6041 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FSO))
6042 #else
6043 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6044 #endif
6045 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
6046
6047 /* insert frame checksum */
6048 if (!(tx_flags & IXGBE_TX_FLAGS_NO_IFCS))
6049 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_IFCS);
6050
6051 return cmd_type;
6052 }
6053
6054 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
6055 u32 tx_flags, unsigned int paylen)
6056 {
6057 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
6058
6059 /* enable L4 checksum for TSO and TX checksum offload */
6060 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6061 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
6062
6063 /* enble IPv4 checksum for TSO */
6064 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6065 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
6066
6067 /* use index 1 context for TSO/FSO/FCOE */
6068 #ifdef IXGBE_FCOE
6069 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FCOE))
6070 #else
6071 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6072 #endif
6073 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
6074
6075 /*
6076 * Check Context must be set if Tx switch is enabled, which it
6077 * always is for case where virtual functions are running
6078 */
6079 #ifdef IXGBE_FCOE
6080 if (tx_flags & (IXGBE_TX_FLAGS_TXSW | IXGBE_TX_FLAGS_FCOE))
6081 #else
6082 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
6083 #endif
6084 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
6085
6086 tx_desc->read.olinfo_status = olinfo_status;
6087 }
6088
6089 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6090 IXGBE_TXD_CMD_RS)
6091
6092 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6093 struct ixgbe_tx_buffer *first,
6094 const u8 hdr_len)
6095 {
6096 dma_addr_t dma;
6097 struct sk_buff *skb = first->skb;
6098 struct ixgbe_tx_buffer *tx_buffer;
6099 union ixgbe_adv_tx_desc *tx_desc;
6100 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6101 unsigned int data_len = skb->data_len;
6102 unsigned int size = skb_headlen(skb);
6103 unsigned int paylen = skb->len - hdr_len;
6104 u32 tx_flags = first->tx_flags;
6105 __le32 cmd_type;
6106 u16 i = tx_ring->next_to_use;
6107
6108 tx_desc = IXGBE_TX_DESC(tx_ring, i);
6109
6110 ixgbe_tx_olinfo_status(tx_desc, tx_flags, paylen);
6111 cmd_type = ixgbe_tx_cmd_type(tx_flags);
6112
6113 #ifdef IXGBE_FCOE
6114 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6115 if (data_len < sizeof(struct fcoe_crc_eof)) {
6116 size -= sizeof(struct fcoe_crc_eof) - data_len;
6117 data_len = 0;
6118 } else {
6119 data_len -= sizeof(struct fcoe_crc_eof);
6120 }
6121 }
6122
6123 #endif
6124 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6125 if (dma_mapping_error(tx_ring->dev, dma))
6126 goto dma_error;
6127
6128 /* record length, and DMA address */
6129 dma_unmap_len_set(first, len, size);
6130 dma_unmap_addr_set(first, dma, dma);
6131
6132 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6133
6134 for (;;) {
6135 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
6136 tx_desc->read.cmd_type_len =
6137 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
6138
6139 i++;
6140 tx_desc++;
6141 if (i == tx_ring->count) {
6142 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6143 i = 0;
6144 }
6145
6146 dma += IXGBE_MAX_DATA_PER_TXD;
6147 size -= IXGBE_MAX_DATA_PER_TXD;
6148
6149 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6150 tx_desc->read.olinfo_status = 0;
6151 }
6152
6153 if (likely(!data_len))
6154 break;
6155
6156 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6157
6158 i++;
6159 tx_desc++;
6160 if (i == tx_ring->count) {
6161 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6162 i = 0;
6163 }
6164
6165 #ifdef IXGBE_FCOE
6166 size = min_t(unsigned int, data_len, skb_frag_size(frag));
6167 #else
6168 size = skb_frag_size(frag);
6169 #endif
6170 data_len -= size;
6171
6172 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6173 DMA_TO_DEVICE);
6174 if (dma_mapping_error(tx_ring->dev, dma))
6175 goto dma_error;
6176
6177 tx_buffer = &tx_ring->tx_buffer_info[i];
6178 dma_unmap_len_set(tx_buffer, len, size);
6179 dma_unmap_addr_set(tx_buffer, dma, dma);
6180
6181 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6182 tx_desc->read.olinfo_status = 0;
6183
6184 frag++;
6185 }
6186
6187 /* write last descriptor with RS and EOP bits */
6188 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
6189 tx_desc->read.cmd_type_len = cmd_type;
6190
6191 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6192
6193 /* set the timestamp */
6194 first->time_stamp = jiffies;
6195
6196 /*
6197 * Force memory writes to complete before letting h/w know there
6198 * are new descriptors to fetch. (Only applicable for weak-ordered
6199 * memory model archs, such as IA-64).
6200 *
6201 * We also need this memory barrier to make certain all of the
6202 * status bits have been updated before next_to_watch is written.
6203 */
6204 wmb();
6205
6206 /* set next_to_watch value indicating a packet is present */
6207 first->next_to_watch = tx_desc;
6208
6209 i++;
6210 if (i == tx_ring->count)
6211 i = 0;
6212
6213 tx_ring->next_to_use = i;
6214
6215 /* notify HW of packet */
6216 writel(i, tx_ring->tail);
6217
6218 return;
6219 dma_error:
6220 dev_err(tx_ring->dev, "TX DMA map failed\n");
6221
6222 /* clear dma mappings for failed tx_buffer_info map */
6223 for (;;) {
6224 tx_buffer = &tx_ring->tx_buffer_info[i];
6225 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6226 if (tx_buffer == first)
6227 break;
6228 if (i == 0)
6229 i = tx_ring->count;
6230 i--;
6231 }
6232
6233 tx_ring->next_to_use = i;
6234 }
6235
6236 static void ixgbe_atr(struct ixgbe_ring *ring,
6237 struct ixgbe_tx_buffer *first)
6238 {
6239 struct ixgbe_q_vector *q_vector = ring->q_vector;
6240 union ixgbe_atr_hash_dword input = { .dword = 0 };
6241 union ixgbe_atr_hash_dword common = { .dword = 0 };
6242 union {
6243 unsigned char *network;
6244 struct iphdr *ipv4;
6245 struct ipv6hdr *ipv6;
6246 } hdr;
6247 struct tcphdr *th;
6248 __be16 vlan_id;
6249
6250 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6251 if (!q_vector)
6252 return;
6253
6254 /* do nothing if sampling is disabled */
6255 if (!ring->atr_sample_rate)
6256 return;
6257
6258 ring->atr_count++;
6259
6260 /* snag network header to get L4 type and address */
6261 hdr.network = skb_network_header(first->skb);
6262
6263 /* Currently only IPv4/IPv6 with TCP is supported */
6264 if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6265 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6266 (first->protocol != __constant_htons(ETH_P_IP) ||
6267 hdr.ipv4->protocol != IPPROTO_TCP))
6268 return;
6269
6270 th = tcp_hdr(first->skb);
6271
6272 /* skip this packet since it is invalid or the socket is closing */
6273 if (!th || th->fin)
6274 return;
6275
6276 /* sample on all syn packets or once every atr sample count */
6277 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6278 return;
6279
6280 /* reset sample count */
6281 ring->atr_count = 0;
6282
6283 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6284
6285 /*
6286 * src and dst are inverted, think how the receiver sees them
6287 *
6288 * The input is broken into two sections, a non-compressed section
6289 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6290 * is XORed together and stored in the compressed dword.
6291 */
6292 input.formatted.vlan_id = vlan_id;
6293
6294 /*
6295 * since src port and flex bytes occupy the same word XOR them together
6296 * and write the value to source port portion of compressed dword
6297 */
6298 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6299 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6300 else
6301 common.port.src ^= th->dest ^ first->protocol;
6302 common.port.dst ^= th->source;
6303
6304 if (first->protocol == __constant_htons(ETH_P_IP)) {
6305 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6306 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6307 } else {
6308 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6309 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6310 hdr.ipv6->saddr.s6_addr32[1] ^
6311 hdr.ipv6->saddr.s6_addr32[2] ^
6312 hdr.ipv6->saddr.s6_addr32[3] ^
6313 hdr.ipv6->daddr.s6_addr32[0] ^
6314 hdr.ipv6->daddr.s6_addr32[1] ^
6315 hdr.ipv6->daddr.s6_addr32[2] ^
6316 hdr.ipv6->daddr.s6_addr32[3];
6317 }
6318
6319 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6320 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6321 input, common, ring->queue_index);
6322 }
6323
6324 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6325 {
6326 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6327 /* Herbert's original patch had:
6328 * smp_mb__after_netif_stop_queue();
6329 * but since that doesn't exist yet, just open code it. */
6330 smp_mb();
6331
6332 /* We need to check again in a case another CPU has just
6333 * made room available. */
6334 if (likely(ixgbe_desc_unused(tx_ring) < size))
6335 return -EBUSY;
6336
6337 /* A reprieve! - use start_queue because it doesn't call schedule */
6338 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6339 ++tx_ring->tx_stats.restart_queue;
6340 return 0;
6341 }
6342
6343 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6344 {
6345 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6346 return 0;
6347 return __ixgbe_maybe_stop_tx(tx_ring, size);
6348 }
6349
6350 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6351 {
6352 struct ixgbe_adapter *adapter = netdev_priv(dev);
6353 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6354 smp_processor_id();
6355 #ifdef IXGBE_FCOE
6356 __be16 protocol = vlan_get_protocol(skb);
6357
6358 if (((protocol == htons(ETH_P_FCOE)) ||
6359 (protocol == htons(ETH_P_FIP))) &&
6360 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6361 struct ixgbe_ring_feature *f;
6362
6363 f = &adapter->ring_feature[RING_F_FCOE];
6364
6365 while (txq >= f->indices)
6366 txq -= f->indices;
6367 txq += adapter->ring_feature[RING_F_FCOE].offset;
6368
6369 return txq;
6370 }
6371 #endif
6372
6373 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6374 while (unlikely(txq >= dev->real_num_tx_queues))
6375 txq -= dev->real_num_tx_queues;
6376 return txq;
6377 }
6378
6379 return skb_tx_hash(dev, skb);
6380 }
6381
6382 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6383 struct ixgbe_adapter *adapter,
6384 struct ixgbe_ring *tx_ring)
6385 {
6386 struct ixgbe_tx_buffer *first;
6387 int tso;
6388 u32 tx_flags = 0;
6389 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6390 unsigned short f;
6391 #endif
6392 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6393 __be16 protocol = skb->protocol;
6394 u8 hdr_len = 0;
6395
6396 /*
6397 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6398 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6399 * + 2 desc gap to keep tail from touching head,
6400 * + 1 desc for context descriptor,
6401 * otherwise try next time
6402 */
6403 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6404 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6405 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6406 #else
6407 count += skb_shinfo(skb)->nr_frags;
6408 #endif
6409 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6410 tx_ring->tx_stats.tx_busy++;
6411 return NETDEV_TX_BUSY;
6412 }
6413
6414 /* record the location of the first descriptor for this packet */
6415 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6416 first->skb = skb;
6417 first->bytecount = skb->len;
6418 first->gso_segs = 1;
6419
6420 /* if we have a HW VLAN tag being added default to the HW one */
6421 if (vlan_tx_tag_present(skb)) {
6422 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6423 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6424 /* else if it is a SW VLAN check the next protocol and store the tag */
6425 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6426 struct vlan_hdr *vhdr, _vhdr;
6427 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6428 if (!vhdr)
6429 goto out_drop;
6430
6431 protocol = vhdr->h_vlan_encapsulated_proto;
6432 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6433 IXGBE_TX_FLAGS_VLAN_SHIFT;
6434 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6435 }
6436
6437 skb_tx_timestamp(skb);
6438
6439 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6440 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6441 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
6442 }
6443
6444 #ifdef CONFIG_PCI_IOV
6445 /*
6446 * Use the l2switch_enable flag - would be false if the DMA
6447 * Tx switch had been disabled.
6448 */
6449 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6450 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6451
6452 #endif
6453 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6454 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6455 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6456 (skb->priority != TC_PRIO_CONTROL))) {
6457 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6458 tx_flags |= (skb->priority & 0x7) <<
6459 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6460 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6461 struct vlan_ethhdr *vhdr;
6462 if (skb_header_cloned(skb) &&
6463 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6464 goto out_drop;
6465 vhdr = (struct vlan_ethhdr *)skb->data;
6466 vhdr->h_vlan_TCI = htons(tx_flags >>
6467 IXGBE_TX_FLAGS_VLAN_SHIFT);
6468 } else {
6469 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6470 }
6471 }
6472
6473 /* record initial flags and protocol */
6474 first->tx_flags = tx_flags;
6475 first->protocol = protocol;
6476
6477 #ifdef IXGBE_FCOE
6478 /* setup tx offload for FCoE */
6479 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6480 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
6481 tso = ixgbe_fso(tx_ring, first, &hdr_len);
6482 if (tso < 0)
6483 goto out_drop;
6484
6485 goto xmit_fcoe;
6486 }
6487
6488 #endif /* IXGBE_FCOE */
6489 tso = ixgbe_tso(tx_ring, first, &hdr_len);
6490 if (tso < 0)
6491 goto out_drop;
6492 else if (!tso)
6493 ixgbe_tx_csum(tx_ring, first);
6494
6495 /* add the ATR filter if ATR is on */
6496 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6497 ixgbe_atr(tx_ring, first);
6498
6499 #ifdef IXGBE_FCOE
6500 xmit_fcoe:
6501 #endif /* IXGBE_FCOE */
6502 ixgbe_tx_map(tx_ring, first, hdr_len);
6503
6504 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6505
6506 return NETDEV_TX_OK;
6507
6508 out_drop:
6509 dev_kfree_skb_any(first->skb);
6510 first->skb = NULL;
6511
6512 return NETDEV_TX_OK;
6513 }
6514
6515 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6516 struct net_device *netdev)
6517 {
6518 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6519 struct ixgbe_ring *tx_ring;
6520
6521 /*
6522 * The minimum packet size for olinfo paylen is 17 so pad the skb
6523 * in order to meet this minimum size requirement.
6524 */
6525 if (unlikely(skb->len < 17)) {
6526 if (skb_pad(skb, 17 - skb->len))
6527 return NETDEV_TX_OK;
6528 skb->len = 17;
6529 skb_set_tail_pointer(skb, 17);
6530 }
6531
6532 tx_ring = adapter->tx_ring[skb->queue_mapping];
6533 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6534 }
6535
6536 /**
6537 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6538 * @netdev: network interface device structure
6539 * @p: pointer to an address structure
6540 *
6541 * Returns 0 on success, negative on failure
6542 **/
6543 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6544 {
6545 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6546 struct ixgbe_hw *hw = &adapter->hw;
6547 struct sockaddr *addr = p;
6548
6549 if (!is_valid_ether_addr(addr->sa_data))
6550 return -EADDRNOTAVAIL;
6551
6552 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6553 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6554
6555 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
6556
6557 return 0;
6558 }
6559
6560 static int
6561 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6562 {
6563 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6564 struct ixgbe_hw *hw = &adapter->hw;
6565 u16 value;
6566 int rc;
6567
6568 if (prtad != hw->phy.mdio.prtad)
6569 return -EINVAL;
6570 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6571 if (!rc)
6572 rc = value;
6573 return rc;
6574 }
6575
6576 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6577 u16 addr, u16 value)
6578 {
6579 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6580 struct ixgbe_hw *hw = &adapter->hw;
6581
6582 if (prtad != hw->phy.mdio.prtad)
6583 return -EINVAL;
6584 return hw->phy.ops.write_reg(hw, addr, devad, value);
6585 }
6586
6587 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6588 {
6589 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6590
6591 switch (cmd) {
6592 case SIOCSHWTSTAMP:
6593 return ixgbe_ptp_hwtstamp_ioctl(adapter, req, cmd);
6594 default:
6595 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6596 }
6597 }
6598
6599 /**
6600 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6601 * netdev->dev_addrs
6602 * @netdev: network interface device structure
6603 *
6604 * Returns non-zero on failure
6605 **/
6606 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6607 {
6608 int err = 0;
6609 struct ixgbe_adapter *adapter = netdev_priv(dev);
6610 struct ixgbe_hw *hw = &adapter->hw;
6611
6612 if (is_valid_ether_addr(hw->mac.san_addr)) {
6613 rtnl_lock();
6614 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
6615 rtnl_unlock();
6616
6617 /* update SAN MAC vmdq pool selection */
6618 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
6619 }
6620 return err;
6621 }
6622
6623 /**
6624 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6625 * netdev->dev_addrs
6626 * @netdev: network interface device structure
6627 *
6628 * Returns non-zero on failure
6629 **/
6630 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6631 {
6632 int err = 0;
6633 struct ixgbe_adapter *adapter = netdev_priv(dev);
6634 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6635
6636 if (is_valid_ether_addr(mac->san_addr)) {
6637 rtnl_lock();
6638 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6639 rtnl_unlock();
6640 }
6641 return err;
6642 }
6643
6644 #ifdef CONFIG_NET_POLL_CONTROLLER
6645 /*
6646 * Polling 'interrupt' - used by things like netconsole to send skbs
6647 * without having to re-enable interrupts. It's not called while
6648 * the interrupt routine is executing.
6649 */
6650 static void ixgbe_netpoll(struct net_device *netdev)
6651 {
6652 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6653 int i;
6654
6655 /* if interface is down do nothing */
6656 if (test_bit(__IXGBE_DOWN, &adapter->state))
6657 return;
6658
6659 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6660 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6661 for (i = 0; i < adapter->num_q_vectors; i++)
6662 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
6663 } else {
6664 ixgbe_intr(adapter->pdev->irq, netdev);
6665 }
6666 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6667 }
6668
6669 #endif
6670 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6671 struct rtnl_link_stats64 *stats)
6672 {
6673 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6674 int i;
6675
6676 rcu_read_lock();
6677 for (i = 0; i < adapter->num_rx_queues; i++) {
6678 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6679 u64 bytes, packets;
6680 unsigned int start;
6681
6682 if (ring) {
6683 do {
6684 start = u64_stats_fetch_begin_bh(&ring->syncp);
6685 packets = ring->stats.packets;
6686 bytes = ring->stats.bytes;
6687 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6688 stats->rx_packets += packets;
6689 stats->rx_bytes += bytes;
6690 }
6691 }
6692
6693 for (i = 0; i < adapter->num_tx_queues; i++) {
6694 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6695 u64 bytes, packets;
6696 unsigned int start;
6697
6698 if (ring) {
6699 do {
6700 start = u64_stats_fetch_begin_bh(&ring->syncp);
6701 packets = ring->stats.packets;
6702 bytes = ring->stats.bytes;
6703 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6704 stats->tx_packets += packets;
6705 stats->tx_bytes += bytes;
6706 }
6707 }
6708 rcu_read_unlock();
6709 /* following stats updated by ixgbe_watchdog_task() */
6710 stats->multicast = netdev->stats.multicast;
6711 stats->rx_errors = netdev->stats.rx_errors;
6712 stats->rx_length_errors = netdev->stats.rx_length_errors;
6713 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6714 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6715 return stats;
6716 }
6717
6718 #ifdef CONFIG_IXGBE_DCB
6719 /**
6720 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6721 * @adapter: pointer to ixgbe_adapter
6722 * @tc: number of traffic classes currently enabled
6723 *
6724 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6725 * 802.1Q priority maps to a packet buffer that exists.
6726 */
6727 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6728 {
6729 struct ixgbe_hw *hw = &adapter->hw;
6730 u32 reg, rsave;
6731 int i;
6732
6733 /* 82598 have a static priority to TC mapping that can not
6734 * be changed so no validation is needed.
6735 */
6736 if (hw->mac.type == ixgbe_mac_82598EB)
6737 return;
6738
6739 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6740 rsave = reg;
6741
6742 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6743 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6744
6745 /* If up2tc is out of bounds default to zero */
6746 if (up2tc > tc)
6747 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6748 }
6749
6750 if (reg != rsave)
6751 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6752
6753 return;
6754 }
6755
6756 /**
6757 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
6758 * @adapter: Pointer to adapter struct
6759 *
6760 * Populate the netdev user priority to tc map
6761 */
6762 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
6763 {
6764 struct net_device *dev = adapter->netdev;
6765 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
6766 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
6767 u8 prio;
6768
6769 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
6770 u8 tc = 0;
6771
6772 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
6773 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
6774 else if (ets)
6775 tc = ets->prio_tc[prio];
6776
6777 netdev_set_prio_tc_map(dev, prio, tc);
6778 }
6779 }
6780
6781 /**
6782 * ixgbe_setup_tc - configure net_device for multiple traffic classes
6783 *
6784 * @netdev: net device to configure
6785 * @tc: number of traffic classes to enable
6786 */
6787 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6788 {
6789 struct ixgbe_adapter *adapter = netdev_priv(dev);
6790 struct ixgbe_hw *hw = &adapter->hw;
6791
6792 /* Hardware supports up to 8 traffic classes */
6793 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
6794 (hw->mac.type == ixgbe_mac_82598EB &&
6795 tc < MAX_TRAFFIC_CLASS))
6796 return -EINVAL;
6797
6798 /* Hardware has to reinitialize queues and interrupts to
6799 * match packet buffer alignment. Unfortunately, the
6800 * hardware is not flexible enough to do this dynamically.
6801 */
6802 if (netif_running(dev))
6803 ixgbe_close(dev);
6804 ixgbe_clear_interrupt_scheme(adapter);
6805
6806 if (tc) {
6807 netdev_set_num_tc(dev, tc);
6808 ixgbe_set_prio_tc_map(adapter);
6809
6810 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6811
6812 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
6813 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
6814 adapter->hw.fc.requested_mode = ixgbe_fc_none;
6815 }
6816 } else {
6817 netdev_reset_tc(dev);
6818
6819 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6820 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6821
6822 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6823
6824 adapter->temp_dcb_cfg.pfc_mode_enable = false;
6825 adapter->dcb_cfg.pfc_mode_enable = false;
6826 }
6827
6828 ixgbe_init_interrupt_scheme(adapter);
6829 ixgbe_validate_rtr(adapter, tc);
6830 if (netif_running(dev))
6831 ixgbe_open(dev);
6832
6833 return 0;
6834 }
6835
6836 #endif /* CONFIG_IXGBE_DCB */
6837 void ixgbe_do_reset(struct net_device *netdev)
6838 {
6839 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6840
6841 if (netif_running(netdev))
6842 ixgbe_reinit_locked(adapter);
6843 else
6844 ixgbe_reset(adapter);
6845 }
6846
6847 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
6848 netdev_features_t features)
6849 {
6850 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6851
6852 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6853 if (!(features & NETIF_F_RXCSUM))
6854 features &= ~NETIF_F_LRO;
6855
6856 /* Turn off LRO if not RSC capable */
6857 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
6858 features &= ~NETIF_F_LRO;
6859
6860 return features;
6861 }
6862
6863 static int ixgbe_set_features(struct net_device *netdev,
6864 netdev_features_t features)
6865 {
6866 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6867 netdev_features_t changed = netdev->features ^ features;
6868 bool need_reset = false;
6869
6870 /* Make sure RSC matches LRO, reset if change */
6871 if (!(features & NETIF_F_LRO)) {
6872 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6873 need_reset = true;
6874 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
6875 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
6876 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6877 if (adapter->rx_itr_setting == 1 ||
6878 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
6879 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
6880 need_reset = true;
6881 } else if ((changed ^ features) & NETIF_F_LRO) {
6882 e_info(probe, "rx-usecs set too low, "
6883 "disabling RSC\n");
6884 }
6885 }
6886
6887 /*
6888 * Check if Flow Director n-tuple support was enabled or disabled. If
6889 * the state changed, we need to reset.
6890 */
6891 switch (features & NETIF_F_NTUPLE) {
6892 case NETIF_F_NTUPLE:
6893 /* turn off ATR, enable perfect filters and reset */
6894 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
6895 need_reset = true;
6896
6897 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6898 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6899 break;
6900 default:
6901 /* turn off perfect filters, enable ATR and reset */
6902 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6903 need_reset = true;
6904
6905 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6906
6907 /* We cannot enable ATR if SR-IOV is enabled */
6908 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6909 break;
6910
6911 /* We cannot enable ATR if we have 2 or more traffic classes */
6912 if (netdev_get_num_tc(netdev) > 1)
6913 break;
6914
6915 /* We cannot enable ATR if RSS is disabled */
6916 if (adapter->ring_feature[RING_F_RSS].limit <= 1)
6917 break;
6918
6919 /* A sample rate of 0 indicates ATR disabled */
6920 if (!adapter->atr_sample_rate)
6921 break;
6922
6923 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6924 break;
6925 }
6926
6927 if (features & NETIF_F_HW_VLAN_RX)
6928 ixgbe_vlan_strip_enable(adapter);
6929 else
6930 ixgbe_vlan_strip_disable(adapter);
6931
6932 if (changed & NETIF_F_RXALL)
6933 need_reset = true;
6934
6935 netdev->features = features;
6936 if (need_reset)
6937 ixgbe_do_reset(netdev);
6938
6939 return 0;
6940 }
6941
6942 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
6943 struct net_device *dev,
6944 const unsigned char *addr,
6945 u16 flags)
6946 {
6947 struct ixgbe_adapter *adapter = netdev_priv(dev);
6948 int err;
6949
6950 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
6951 return -EOPNOTSUPP;
6952
6953 /* Hardware does not support aging addresses so if a
6954 * ndm_state is given only allow permanent addresses
6955 */
6956 if (ndm->ndm_state && !(ndm->ndm_state & NUD_PERMANENT)) {
6957 pr_info("%s: FDB only supports static addresses\n",
6958 ixgbe_driver_name);
6959 return -EINVAL;
6960 }
6961
6962 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
6963 u32 rar_uc_entries = IXGBE_MAX_PF_MACVLANS;
6964
6965 if (netdev_uc_count(dev) < rar_uc_entries)
6966 err = dev_uc_add_excl(dev, addr);
6967 else
6968 err = -ENOMEM;
6969 } else if (is_multicast_ether_addr(addr)) {
6970 err = dev_mc_add_excl(dev, addr);
6971 } else {
6972 err = -EINVAL;
6973 }
6974
6975 /* Only return duplicate errors if NLM_F_EXCL is set */
6976 if (err == -EEXIST && !(flags & NLM_F_EXCL))
6977 err = 0;
6978
6979 return err;
6980 }
6981
6982 static int ixgbe_ndo_fdb_del(struct ndmsg *ndm,
6983 struct net_device *dev,
6984 const unsigned char *addr)
6985 {
6986 struct ixgbe_adapter *adapter = netdev_priv(dev);
6987 int err = -EOPNOTSUPP;
6988
6989 if (ndm->ndm_state & NUD_PERMANENT) {
6990 pr_info("%s: FDB only supports static addresses\n",
6991 ixgbe_driver_name);
6992 return -EINVAL;
6993 }
6994
6995 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6996 if (is_unicast_ether_addr(addr))
6997 err = dev_uc_del(dev, addr);
6998 else if (is_multicast_ether_addr(addr))
6999 err = dev_mc_del(dev, addr);
7000 else
7001 err = -EINVAL;
7002 }
7003
7004 return err;
7005 }
7006
7007 static int ixgbe_ndo_fdb_dump(struct sk_buff *skb,
7008 struct netlink_callback *cb,
7009 struct net_device *dev,
7010 int idx)
7011 {
7012 struct ixgbe_adapter *adapter = netdev_priv(dev);
7013
7014 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7015 idx = ndo_dflt_fdb_dump(skb, cb, dev, idx);
7016
7017 return idx;
7018 }
7019
7020 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
7021 struct nlmsghdr *nlh)
7022 {
7023 struct ixgbe_adapter *adapter = netdev_priv(dev);
7024 struct nlattr *attr, *br_spec;
7025 int rem;
7026
7027 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7028 return -EOPNOTSUPP;
7029
7030 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
7031
7032 nla_for_each_nested(attr, br_spec, rem) {
7033 __u16 mode;
7034 u32 reg = 0;
7035
7036 if (nla_type(attr) != IFLA_BRIDGE_MODE)
7037 continue;
7038
7039 mode = nla_get_u16(attr);
7040 if (mode == BRIDGE_MODE_VEPA) {
7041 reg = 0;
7042 adapter->flags2 &= ~IXGBE_FLAG2_BRIDGE_MODE_VEB;
7043 } else if (mode == BRIDGE_MODE_VEB) {
7044 reg = IXGBE_PFDTXGSWC_VT_LBEN;
7045 adapter->flags2 |= IXGBE_FLAG2_BRIDGE_MODE_VEB;
7046 } else
7047 return -EINVAL;
7048
7049 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, reg);
7050
7051 e_info(drv, "enabling bridge mode: %s\n",
7052 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
7053 }
7054
7055 return 0;
7056 }
7057
7058 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
7059 struct net_device *dev)
7060 {
7061 struct ixgbe_adapter *adapter = netdev_priv(dev);
7062 u16 mode;
7063
7064 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7065 return 0;
7066
7067 if (adapter->flags2 & IXGBE_FLAG2_BRIDGE_MODE_VEB)
7068 mode = BRIDGE_MODE_VEB;
7069 else
7070 mode = BRIDGE_MODE_VEPA;
7071
7072 return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode);
7073 }
7074
7075 static const struct net_device_ops ixgbe_netdev_ops = {
7076 .ndo_open = ixgbe_open,
7077 .ndo_stop = ixgbe_close,
7078 .ndo_start_xmit = ixgbe_xmit_frame,
7079 .ndo_select_queue = ixgbe_select_queue,
7080 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7081 .ndo_validate_addr = eth_validate_addr,
7082 .ndo_set_mac_address = ixgbe_set_mac,
7083 .ndo_change_mtu = ixgbe_change_mtu,
7084 .ndo_tx_timeout = ixgbe_tx_timeout,
7085 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7086 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7087 .ndo_do_ioctl = ixgbe_ioctl,
7088 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7089 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7090 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7091 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
7092 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7093 .ndo_get_stats64 = ixgbe_get_stats64,
7094 #ifdef CONFIG_IXGBE_DCB
7095 .ndo_setup_tc = ixgbe_setup_tc,
7096 #endif
7097 #ifdef CONFIG_NET_POLL_CONTROLLER
7098 .ndo_poll_controller = ixgbe_netpoll,
7099 #endif
7100 #ifdef IXGBE_FCOE
7101 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7102 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7103 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7104 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7105 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7106 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7107 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
7108 #endif /* IXGBE_FCOE */
7109 .ndo_set_features = ixgbe_set_features,
7110 .ndo_fix_features = ixgbe_fix_features,
7111 .ndo_fdb_add = ixgbe_ndo_fdb_add,
7112 .ndo_fdb_del = ixgbe_ndo_fdb_del,
7113 .ndo_fdb_dump = ixgbe_ndo_fdb_dump,
7114 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
7115 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
7116 };
7117
7118 /**
7119 * ixgbe_wol_supported - Check whether device supports WoL
7120 * @hw: hw specific details
7121 * @device_id: the device ID
7122 * @subdev_id: the subsystem device ID
7123 *
7124 * This function is used by probe and ethtool to determine
7125 * which devices have WoL support
7126 *
7127 **/
7128 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
7129 u16 subdevice_id)
7130 {
7131 struct ixgbe_hw *hw = &adapter->hw;
7132 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7133 int is_wol_supported = 0;
7134
7135 switch (device_id) {
7136 case IXGBE_DEV_ID_82599_SFP:
7137 /* Only these subdevices could supports WOL */
7138 switch (subdevice_id) {
7139 case IXGBE_SUBDEV_ID_82599_560FLR:
7140 /* only support first port */
7141 if (hw->bus.func != 0)
7142 break;
7143 case IXGBE_SUBDEV_ID_82599_SFP:
7144 case IXGBE_SUBDEV_ID_82599_RNDC:
7145 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
7146 is_wol_supported = 1;
7147 break;
7148 }
7149 break;
7150 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7151 /* All except this subdevice support WOL */
7152 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7153 is_wol_supported = 1;
7154 break;
7155 case IXGBE_DEV_ID_82599_KX4:
7156 is_wol_supported = 1;
7157 break;
7158 case IXGBE_DEV_ID_X540T:
7159 case IXGBE_DEV_ID_X540T1:
7160 /* check eeprom to see if enabled wol */
7161 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7162 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7163 (hw->bus.func == 0))) {
7164 is_wol_supported = 1;
7165 }
7166 break;
7167 }
7168
7169 return is_wol_supported;
7170 }
7171
7172 /**
7173 * ixgbe_probe - Device Initialization Routine
7174 * @pdev: PCI device information struct
7175 * @ent: entry in ixgbe_pci_tbl
7176 *
7177 * Returns 0 on success, negative on failure
7178 *
7179 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7180 * The OS initialization, configuring of the adapter private structure,
7181 * and a hardware reset occur.
7182 **/
7183 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7184 const struct pci_device_id *ent)
7185 {
7186 struct net_device *netdev;
7187 struct ixgbe_adapter *adapter = NULL;
7188 struct ixgbe_hw *hw;
7189 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7190 static int cards_found;
7191 int i, err, pci_using_dac;
7192 u8 part_str[IXGBE_PBANUM_LENGTH];
7193 unsigned int indices = num_possible_cpus();
7194 unsigned int dcb_max = 0;
7195 #ifdef IXGBE_FCOE
7196 u16 device_caps;
7197 #endif
7198 u32 eec;
7199
7200 /* Catch broken hardware that put the wrong VF device ID in
7201 * the PCIe SR-IOV capability.
7202 */
7203 if (pdev->is_virtfn) {
7204 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7205 pci_name(pdev), pdev->vendor, pdev->device);
7206 return -EINVAL;
7207 }
7208
7209 err = pci_enable_device_mem(pdev);
7210 if (err)
7211 return err;
7212
7213 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7214 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7215 pci_using_dac = 1;
7216 } else {
7217 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7218 if (err) {
7219 err = dma_set_coherent_mask(&pdev->dev,
7220 DMA_BIT_MASK(32));
7221 if (err) {
7222 dev_err(&pdev->dev,
7223 "No usable DMA configuration, aborting\n");
7224 goto err_dma;
7225 }
7226 }
7227 pci_using_dac = 0;
7228 }
7229
7230 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7231 IORESOURCE_MEM), ixgbe_driver_name);
7232 if (err) {
7233 dev_err(&pdev->dev,
7234 "pci_request_selected_regions failed 0x%x\n", err);
7235 goto err_pci_reg;
7236 }
7237
7238 pci_enable_pcie_error_reporting(pdev);
7239
7240 pci_set_master(pdev);
7241 pci_save_state(pdev);
7242
7243 #ifdef CONFIG_IXGBE_DCB
7244 if (ii->mac == ixgbe_mac_82598EB)
7245 dcb_max = min_t(unsigned int, indices * MAX_TRAFFIC_CLASS,
7246 IXGBE_MAX_RSS_INDICES);
7247 else
7248 dcb_max = min_t(unsigned int, indices * MAX_TRAFFIC_CLASS,
7249 IXGBE_MAX_FDIR_INDICES);
7250 #endif
7251
7252 if (ii->mac == ixgbe_mac_82598EB)
7253 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7254 else
7255 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7256
7257 #ifdef IXGBE_FCOE
7258 indices += min_t(unsigned int, num_possible_cpus(),
7259 IXGBE_MAX_FCOE_INDICES);
7260 #endif
7261 indices = max_t(unsigned int, dcb_max, indices);
7262 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7263 if (!netdev) {
7264 err = -ENOMEM;
7265 goto err_alloc_etherdev;
7266 }
7267
7268 SET_NETDEV_DEV(netdev, &pdev->dev);
7269
7270 adapter = netdev_priv(netdev);
7271 pci_set_drvdata(pdev, adapter);
7272
7273 adapter->netdev = netdev;
7274 adapter->pdev = pdev;
7275 hw = &adapter->hw;
7276 hw->back = adapter;
7277 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7278
7279 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7280 pci_resource_len(pdev, 0));
7281 if (!hw->hw_addr) {
7282 err = -EIO;
7283 goto err_ioremap;
7284 }
7285
7286 netdev->netdev_ops = &ixgbe_netdev_ops;
7287 ixgbe_set_ethtool_ops(netdev);
7288 netdev->watchdog_timeo = 5 * HZ;
7289 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7290
7291 adapter->bd_number = cards_found;
7292
7293 /* Setup hw api */
7294 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7295 hw->mac.type = ii->mac;
7296
7297 /* EEPROM */
7298 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7299 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7300 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7301 if (!(eec & (1 << 8)))
7302 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7303
7304 /* PHY */
7305 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7306 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7307 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7308 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7309 hw->phy.mdio.mmds = 0;
7310 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7311 hw->phy.mdio.dev = netdev;
7312 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7313 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7314
7315 ii->get_invariants(hw);
7316
7317 /* setup the private structure */
7318 err = ixgbe_sw_init(adapter);
7319 if (err)
7320 goto err_sw_init;
7321
7322 /* Make it possible the adapter to be woken up via WOL */
7323 switch (adapter->hw.mac.type) {
7324 case ixgbe_mac_82599EB:
7325 case ixgbe_mac_X540:
7326 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7327 break;
7328 default:
7329 break;
7330 }
7331
7332 /*
7333 * If there is a fan on this device and it has failed log the
7334 * failure.
7335 */
7336 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7337 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7338 if (esdp & IXGBE_ESDP_SDP1)
7339 e_crit(probe, "Fan has stopped, replace the adapter\n");
7340 }
7341
7342 if (allow_unsupported_sfp)
7343 hw->allow_unsupported_sfp = allow_unsupported_sfp;
7344
7345 /* reset_hw fills in the perm_addr as well */
7346 hw->phy.reset_if_overtemp = true;
7347 err = hw->mac.ops.reset_hw(hw);
7348 hw->phy.reset_if_overtemp = false;
7349 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7350 hw->mac.type == ixgbe_mac_82598EB) {
7351 err = 0;
7352 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7353 e_dev_err("failed to load because an unsupported SFP+ "
7354 "module type was detected.\n");
7355 e_dev_err("Reload the driver after installing a supported "
7356 "module.\n");
7357 goto err_sw_init;
7358 } else if (err) {
7359 e_dev_err("HW Init failed: %d\n", err);
7360 goto err_sw_init;
7361 }
7362
7363 #ifdef CONFIG_PCI_IOV
7364 ixgbe_enable_sriov(adapter, ii);
7365
7366 #endif
7367 netdev->features = NETIF_F_SG |
7368 NETIF_F_IP_CSUM |
7369 NETIF_F_IPV6_CSUM |
7370 NETIF_F_HW_VLAN_TX |
7371 NETIF_F_HW_VLAN_RX |
7372 NETIF_F_HW_VLAN_FILTER |
7373 NETIF_F_TSO |
7374 NETIF_F_TSO6 |
7375 NETIF_F_RXHASH |
7376 NETIF_F_RXCSUM;
7377
7378 netdev->hw_features = netdev->features;
7379
7380 switch (adapter->hw.mac.type) {
7381 case ixgbe_mac_82599EB:
7382 case ixgbe_mac_X540:
7383 netdev->features |= NETIF_F_SCTP_CSUM;
7384 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7385 NETIF_F_NTUPLE;
7386 break;
7387 default:
7388 break;
7389 }
7390
7391 netdev->hw_features |= NETIF_F_RXALL;
7392
7393 netdev->vlan_features |= NETIF_F_TSO;
7394 netdev->vlan_features |= NETIF_F_TSO6;
7395 netdev->vlan_features |= NETIF_F_IP_CSUM;
7396 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7397 netdev->vlan_features |= NETIF_F_SG;
7398
7399 netdev->priv_flags |= IFF_UNICAST_FLT;
7400 netdev->priv_flags |= IFF_SUPP_NOFCS;
7401
7402 #ifdef CONFIG_IXGBE_DCB
7403 netdev->dcbnl_ops = &dcbnl_ops;
7404 #endif
7405
7406 #ifdef IXGBE_FCOE
7407 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7408 if (hw->mac.ops.get_device_caps) {
7409 hw->mac.ops.get_device_caps(hw, &device_caps);
7410 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7411 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7412 }
7413
7414 adapter->ring_feature[RING_F_FCOE].limit = IXGBE_FCRETA_SIZE;
7415
7416 netdev->features |= NETIF_F_FSO |
7417 NETIF_F_FCOE_CRC;
7418
7419 netdev->vlan_features |= NETIF_F_FSO |
7420 NETIF_F_FCOE_CRC |
7421 NETIF_F_FCOE_MTU;
7422 }
7423 #endif /* IXGBE_FCOE */
7424 if (pci_using_dac) {
7425 netdev->features |= NETIF_F_HIGHDMA;
7426 netdev->vlan_features |= NETIF_F_HIGHDMA;
7427 }
7428
7429 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7430 netdev->hw_features |= NETIF_F_LRO;
7431 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7432 netdev->features |= NETIF_F_LRO;
7433
7434 /* make sure the EEPROM is good */
7435 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7436 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7437 err = -EIO;
7438 goto err_sw_init;
7439 }
7440
7441 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7442 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7443
7444 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7445 e_dev_err("invalid MAC address\n");
7446 err = -EIO;
7447 goto err_sw_init;
7448 }
7449
7450 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7451 (unsigned long) adapter);
7452
7453 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7454 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7455
7456 err = ixgbe_init_interrupt_scheme(adapter);
7457 if (err)
7458 goto err_sw_init;
7459
7460 /* WOL not supported for all devices */
7461 adapter->wol = 0;
7462 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7463 if (ixgbe_wol_supported(adapter, pdev->device, pdev->subsystem_device))
7464 adapter->wol = IXGBE_WUFC_MAG;
7465
7466 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7467
7468 /* save off EEPROM version number */
7469 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7470 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7471
7472 /* pick up the PCI bus settings for reporting later */
7473 hw->mac.ops.get_bus_info(hw);
7474
7475 /* print bus type/speed/width info */
7476 e_dev_info("(PCI Express:%s:%s) %pM\n",
7477 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7478 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7479 "Unknown"),
7480 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7481 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7482 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7483 "Unknown"),
7484 netdev->dev_addr);
7485
7486 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7487 if (err)
7488 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7489 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7490 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7491 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7492 part_str);
7493 else
7494 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7495 hw->mac.type, hw->phy.type, part_str);
7496
7497 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7498 e_dev_warn("PCI-Express bandwidth available for this card is "
7499 "not sufficient for optimal performance.\n");
7500 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7501 "is required.\n");
7502 }
7503
7504 /* reset the hardware with the new settings */
7505 err = hw->mac.ops.start_hw(hw);
7506 if (err == IXGBE_ERR_EEPROM_VERSION) {
7507 /* We are running on a pre-production device, log a warning */
7508 e_dev_warn("This device is a pre-production adapter/LOM. "
7509 "Please be aware there may be issues associated "
7510 "with your hardware. If you are experiencing "
7511 "problems please contact your Intel or hardware "
7512 "representative who provided you with this "
7513 "hardware.\n");
7514 }
7515 strcpy(netdev->name, "eth%d");
7516 err = register_netdev(netdev);
7517 if (err)
7518 goto err_register;
7519
7520 /* power down the optics for 82599 SFP+ fiber */
7521 if (hw->mac.ops.disable_tx_laser)
7522 hw->mac.ops.disable_tx_laser(hw);
7523
7524 /* carrier off reporting is important to ethtool even BEFORE open */
7525 netif_carrier_off(netdev);
7526
7527 #ifdef CONFIG_IXGBE_DCA
7528 if (dca_add_requester(&pdev->dev) == 0) {
7529 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7530 ixgbe_setup_dca(adapter);
7531 }
7532 #endif
7533 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7534 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7535 for (i = 0; i < adapter->num_vfs; i++)
7536 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7537 }
7538
7539 /* firmware requires driver version to be 0xFFFFFFFF
7540 * since os does not support feature
7541 */
7542 if (hw->mac.ops.set_fw_drv_ver)
7543 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7544 0xFF);
7545
7546 /* add san mac addr to netdev */
7547 ixgbe_add_sanmac_netdev(netdev);
7548
7549 e_dev_info("%s\n", ixgbe_default_device_descr);
7550 cards_found++;
7551
7552 #ifdef CONFIG_IXGBE_HWMON
7553 if (ixgbe_sysfs_init(adapter))
7554 e_err(probe, "failed to allocate sysfs resources\n");
7555 #endif /* CONFIG_IXGBE_HWMON */
7556
7557 #ifdef CONFIG_DEBUG_FS
7558 ixgbe_dbg_adapter_init(adapter);
7559 #endif /* CONFIG_DEBUG_FS */
7560
7561 return 0;
7562
7563 err_register:
7564 ixgbe_release_hw_control(adapter);
7565 ixgbe_clear_interrupt_scheme(adapter);
7566 err_sw_init:
7567 ixgbe_disable_sriov(adapter);
7568 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7569 iounmap(hw->hw_addr);
7570 err_ioremap:
7571 free_netdev(netdev);
7572 err_alloc_etherdev:
7573 pci_release_selected_regions(pdev,
7574 pci_select_bars(pdev, IORESOURCE_MEM));
7575 err_pci_reg:
7576 err_dma:
7577 pci_disable_device(pdev);
7578 return err;
7579 }
7580
7581 /**
7582 * ixgbe_remove - Device Removal Routine
7583 * @pdev: PCI device information struct
7584 *
7585 * ixgbe_remove is called by the PCI subsystem to alert the driver
7586 * that it should release a PCI device. The could be caused by a
7587 * Hot-Plug event, or because the driver is going to be removed from
7588 * memory.
7589 **/
7590 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7591 {
7592 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7593 struct net_device *netdev = adapter->netdev;
7594
7595 #ifdef CONFIG_DEBUG_FS
7596 ixgbe_dbg_adapter_exit(adapter);
7597 #endif /*CONFIG_DEBUG_FS */
7598
7599 set_bit(__IXGBE_DOWN, &adapter->state);
7600 cancel_work_sync(&adapter->service_task);
7601
7602
7603 #ifdef CONFIG_IXGBE_DCA
7604 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7605 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7606 dca_remove_requester(&pdev->dev);
7607 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7608 }
7609
7610 #endif
7611 #ifdef CONFIG_IXGBE_HWMON
7612 ixgbe_sysfs_exit(adapter);
7613 #endif /* CONFIG_IXGBE_HWMON */
7614
7615 /* remove the added san mac */
7616 ixgbe_del_sanmac_netdev(netdev);
7617
7618 if (netdev->reg_state == NETREG_REGISTERED)
7619 unregister_netdev(netdev);
7620
7621 ixgbe_disable_sriov(adapter);
7622
7623 ixgbe_clear_interrupt_scheme(adapter);
7624
7625 ixgbe_release_hw_control(adapter);
7626
7627 #ifdef CONFIG_DCB
7628 kfree(adapter->ixgbe_ieee_pfc);
7629 kfree(adapter->ixgbe_ieee_ets);
7630
7631 #endif
7632 iounmap(adapter->hw.hw_addr);
7633 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7634 IORESOURCE_MEM));
7635
7636 e_dev_info("complete\n");
7637
7638 free_netdev(netdev);
7639
7640 pci_disable_pcie_error_reporting(pdev);
7641
7642 pci_disable_device(pdev);
7643 }
7644
7645 /**
7646 * ixgbe_io_error_detected - called when PCI error is detected
7647 * @pdev: Pointer to PCI device
7648 * @state: The current pci connection state
7649 *
7650 * This function is called after a PCI bus error affecting
7651 * this device has been detected.
7652 */
7653 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7654 pci_channel_state_t state)
7655 {
7656 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7657 struct net_device *netdev = adapter->netdev;
7658
7659 #ifdef CONFIG_PCI_IOV
7660 struct pci_dev *bdev, *vfdev;
7661 u32 dw0, dw1, dw2, dw3;
7662 int vf, pos;
7663 u16 req_id, pf_func;
7664
7665 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7666 adapter->num_vfs == 0)
7667 goto skip_bad_vf_detection;
7668
7669 bdev = pdev->bus->self;
7670 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
7671 bdev = bdev->bus->self;
7672
7673 if (!bdev)
7674 goto skip_bad_vf_detection;
7675
7676 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7677 if (!pos)
7678 goto skip_bad_vf_detection;
7679
7680 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7681 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7682 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7683 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7684
7685 req_id = dw1 >> 16;
7686 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7687 if (!(req_id & 0x0080))
7688 goto skip_bad_vf_detection;
7689
7690 pf_func = req_id & 0x01;
7691 if ((pf_func & 1) == (pdev->devfn & 1)) {
7692 unsigned int device_id;
7693
7694 vf = (req_id & 0x7F) >> 1;
7695 e_dev_err("VF %d has caused a PCIe error\n", vf);
7696 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7697 "%8.8x\tdw3: %8.8x\n",
7698 dw0, dw1, dw2, dw3);
7699 switch (adapter->hw.mac.type) {
7700 case ixgbe_mac_82599EB:
7701 device_id = IXGBE_82599_VF_DEVICE_ID;
7702 break;
7703 case ixgbe_mac_X540:
7704 device_id = IXGBE_X540_VF_DEVICE_ID;
7705 break;
7706 default:
7707 device_id = 0;
7708 break;
7709 }
7710
7711 /* Find the pci device of the offending VF */
7712 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
7713 while (vfdev) {
7714 if (vfdev->devfn == (req_id & 0xFF))
7715 break;
7716 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
7717 device_id, vfdev);
7718 }
7719 /*
7720 * There's a slim chance the VF could have been hot plugged,
7721 * so if it is no longer present we don't need to issue the
7722 * VFLR. Just clean up the AER in that case.
7723 */
7724 if (vfdev) {
7725 e_dev_err("Issuing VFLR to VF %d\n", vf);
7726 pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7727 }
7728
7729 pci_cleanup_aer_uncorrect_error_status(pdev);
7730 }
7731
7732 /*
7733 * Even though the error may have occurred on the other port
7734 * we still need to increment the vf error reference count for
7735 * both ports because the I/O resume function will be called
7736 * for both of them.
7737 */
7738 adapter->vferr_refcount++;
7739
7740 return PCI_ERS_RESULT_RECOVERED;
7741
7742 skip_bad_vf_detection:
7743 #endif /* CONFIG_PCI_IOV */
7744 netif_device_detach(netdev);
7745
7746 if (state == pci_channel_io_perm_failure)
7747 return PCI_ERS_RESULT_DISCONNECT;
7748
7749 if (netif_running(netdev))
7750 ixgbe_down(adapter);
7751 pci_disable_device(pdev);
7752
7753 /* Request a slot reset. */
7754 return PCI_ERS_RESULT_NEED_RESET;
7755 }
7756
7757 /**
7758 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7759 * @pdev: Pointer to PCI device
7760 *
7761 * Restart the card from scratch, as if from a cold-boot.
7762 */
7763 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7764 {
7765 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7766 pci_ers_result_t result;
7767 int err;
7768
7769 if (pci_enable_device_mem(pdev)) {
7770 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7771 result = PCI_ERS_RESULT_DISCONNECT;
7772 } else {
7773 pci_set_master(pdev);
7774 pci_restore_state(pdev);
7775 pci_save_state(pdev);
7776
7777 pci_wake_from_d3(pdev, false);
7778
7779 ixgbe_reset(adapter);
7780 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7781 result = PCI_ERS_RESULT_RECOVERED;
7782 }
7783
7784 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7785 if (err) {
7786 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7787 "failed 0x%0x\n", err);
7788 /* non-fatal, continue */
7789 }
7790
7791 return result;
7792 }
7793
7794 /**
7795 * ixgbe_io_resume - called when traffic can start flowing again.
7796 * @pdev: Pointer to PCI device
7797 *
7798 * This callback is called when the error recovery driver tells us that
7799 * its OK to resume normal operation.
7800 */
7801 static void ixgbe_io_resume(struct pci_dev *pdev)
7802 {
7803 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7804 struct net_device *netdev = adapter->netdev;
7805
7806 #ifdef CONFIG_PCI_IOV
7807 if (adapter->vferr_refcount) {
7808 e_info(drv, "Resuming after VF err\n");
7809 adapter->vferr_refcount--;
7810 return;
7811 }
7812
7813 #endif
7814 if (netif_running(netdev))
7815 ixgbe_up(adapter);
7816
7817 netif_device_attach(netdev);
7818 }
7819
7820 static const struct pci_error_handlers ixgbe_err_handler = {
7821 .error_detected = ixgbe_io_error_detected,
7822 .slot_reset = ixgbe_io_slot_reset,
7823 .resume = ixgbe_io_resume,
7824 };
7825
7826 static struct pci_driver ixgbe_driver = {
7827 .name = ixgbe_driver_name,
7828 .id_table = ixgbe_pci_tbl,
7829 .probe = ixgbe_probe,
7830 .remove = __devexit_p(ixgbe_remove),
7831 #ifdef CONFIG_PM
7832 .suspend = ixgbe_suspend,
7833 .resume = ixgbe_resume,
7834 #endif
7835 .shutdown = ixgbe_shutdown,
7836 .err_handler = &ixgbe_err_handler
7837 };
7838
7839 /**
7840 * ixgbe_init_module - Driver Registration Routine
7841 *
7842 * ixgbe_init_module is the first routine called when the driver is
7843 * loaded. All it does is register with the PCI subsystem.
7844 **/
7845 static int __init ixgbe_init_module(void)
7846 {
7847 int ret;
7848 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7849 pr_info("%s\n", ixgbe_copyright);
7850
7851 #ifdef CONFIG_DEBUG_FS
7852 ixgbe_dbg_init();
7853 #endif /* CONFIG_DEBUG_FS */
7854
7855 #ifdef CONFIG_IXGBE_DCA
7856 dca_register_notify(&dca_notifier);
7857 #endif
7858
7859 ret = pci_register_driver(&ixgbe_driver);
7860 return ret;
7861 }
7862
7863 module_init(ixgbe_init_module);
7864
7865 /**
7866 * ixgbe_exit_module - Driver Exit Cleanup Routine
7867 *
7868 * ixgbe_exit_module is called just before the driver is removed
7869 * from memory.
7870 **/
7871 static void __exit ixgbe_exit_module(void)
7872 {
7873 #ifdef CONFIG_IXGBE_DCA
7874 dca_unregister_notify(&dca_notifier);
7875 #endif
7876 pci_unregister_driver(&ixgbe_driver);
7877
7878 #ifdef CONFIG_DEBUG_FS
7879 ixgbe_dbg_exit();
7880 #endif /* CONFIG_DEBUG_FS */
7881
7882 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7883 }
7884
7885 #ifdef CONFIG_IXGBE_DCA
7886 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7887 void *p)
7888 {
7889 int ret_val;
7890
7891 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7892 __ixgbe_notify_dca);
7893
7894 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7895 }
7896
7897 #endif /* CONFIG_IXGBE_DCA */
7898
7899 module_exit(ixgbe_exit_module);
7900
7901 /* ixgbe_main.c */
Linux 2.6 libata-dev (mirror)