projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ixgbe: Add function for obtaining FCoE TC based on FCoE user priority
[deliverable/linux.git] / drivers / net / ethernet / intel / ixgbe / ixgbe_main.c
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/prefetch.h>
48 #include <scsi/fc/fc_fcoe.h>
49
50 #include "ixgbe.h"
51 #include "ixgbe_common.h"
52 #include "ixgbe_dcb_82599.h"
53 #include "ixgbe_sriov.h"
54
55 char ixgbe_driver_name[] = "ixgbe";
56 static const char ixgbe_driver_string[] =
57 "Intel(R) 10 Gigabit PCI Express Network Driver";
58 #ifdef IXGBE_FCOE
59 char ixgbe_default_device_descr[] =
60 "Intel(R) 10 Gigabit Network Connection";
61 #else
62 static char ixgbe_default_device_descr[] =
63 "Intel(R) 10 Gigabit Network Connection";
64 #endif
65 #define MAJ 3
66 #define MIN 9
67 #define BUILD 15
68 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
69 __stringify(BUILD) "-k"
70 const char ixgbe_driver_version[] = DRV_VERSION;
71 static const char ixgbe_copyright[] =
72 "Copyright (c) 1999-2012 Intel Corporation.";
73
74 static const struct ixgbe_info *ixgbe_info_tbl[] = {
75 [board_82598] = &ixgbe_82598_info,
76 [board_82599] = &ixgbe_82599_info,
77 [board_X540] = &ixgbe_X540_info,
78 };
79
80 /* ixgbe_pci_tbl - PCI Device ID Table
81 *
82 * Wildcard entries (PCI_ANY_ID) should come last
83 * Last entry must be all 0s
84 *
85 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
86 * Class, Class Mask, private data (not used) }
87 */
88 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
117 /* required last entry */
118 {0, }
119 };
120 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
121
122 #ifdef CONFIG_IXGBE_DCA
123 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
124 void *p);
125 static struct notifier_block dca_notifier = {
126 .notifier_call = ixgbe_notify_dca,
127 .next = NULL,
128 .priority = 0
129 };
130 #endif
131
132 #ifdef CONFIG_PCI_IOV
133 static unsigned int max_vfs;
134 module_param(max_vfs, uint, 0);
135 MODULE_PARM_DESC(max_vfs,
136 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63");
137 #endif /* CONFIG_PCI_IOV */
138
139 static unsigned int allow_unsupported_sfp;
140 module_param(allow_unsupported_sfp, uint, 0);
141 MODULE_PARM_DESC(allow_unsupported_sfp,
142 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
143
144 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
145 static int debug = -1;
146 module_param(debug, int, 0);
147 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
148
149 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
150 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
151 MODULE_LICENSE("GPL");
152 MODULE_VERSION(DRV_VERSION);
153
154 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
155 {
156 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
157 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
158 schedule_work(&adapter->service_task);
159 }
160
161 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
162 {
163 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
164
165 /* flush memory to make sure state is correct before next watchdog */
166 smp_mb__before_clear_bit();
167 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
168 }
169
170 struct ixgbe_reg_info {
171 u32 ofs;
172 char *name;
173 };
174
175 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
176
177 /* General Registers */
178 {IXGBE_CTRL, "CTRL"},
179 {IXGBE_STATUS, "STATUS"},
180 {IXGBE_CTRL_EXT, "CTRL_EXT"},
181
182 /* Interrupt Registers */
183 {IXGBE_EICR, "EICR"},
184
185 /* RX Registers */
186 {IXGBE_SRRCTL(0), "SRRCTL"},
187 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
188 {IXGBE_RDLEN(0), "RDLEN"},
189 {IXGBE_RDH(0), "RDH"},
190 {IXGBE_RDT(0), "RDT"},
191 {IXGBE_RXDCTL(0), "RXDCTL"},
192 {IXGBE_RDBAL(0), "RDBAL"},
193 {IXGBE_RDBAH(0), "RDBAH"},
194
195 /* TX Registers */
196 {IXGBE_TDBAL(0), "TDBAL"},
197 {IXGBE_TDBAH(0), "TDBAH"},
198 {IXGBE_TDLEN(0), "TDLEN"},
199 {IXGBE_TDH(0), "TDH"},
200 {IXGBE_TDT(0), "TDT"},
201 {IXGBE_TXDCTL(0), "TXDCTL"},
202
203 /* List Terminator */
204 {}
205 };
206
207
208 /*
209 * ixgbe_regdump - register printout routine
210 */
211 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
212 {
213 int i = 0, j = 0;
214 char rname[16];
215 u32 regs[64];
216
217 switch (reginfo->ofs) {
218 case IXGBE_SRRCTL(0):
219 for (i = 0; i < 64; i++)
220 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
221 break;
222 case IXGBE_DCA_RXCTRL(0):
223 for (i = 0; i < 64; i++)
224 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
225 break;
226 case IXGBE_RDLEN(0):
227 for (i = 0; i < 64; i++)
228 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
229 break;
230 case IXGBE_RDH(0):
231 for (i = 0; i < 64; i++)
232 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
233 break;
234 case IXGBE_RDT(0):
235 for (i = 0; i < 64; i++)
236 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
237 break;
238 case IXGBE_RXDCTL(0):
239 for (i = 0; i < 64; i++)
240 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
241 break;
242 case IXGBE_RDBAL(0):
243 for (i = 0; i < 64; i++)
244 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
245 break;
246 case IXGBE_RDBAH(0):
247 for (i = 0; i < 64; i++)
248 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
249 break;
250 case IXGBE_TDBAL(0):
251 for (i = 0; i < 64; i++)
252 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
253 break;
254 case IXGBE_TDBAH(0):
255 for (i = 0; i < 64; i++)
256 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
257 break;
258 case IXGBE_TDLEN(0):
259 for (i = 0; i < 64; i++)
260 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
261 break;
262 case IXGBE_TDH(0):
263 for (i = 0; i < 64; i++)
264 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
265 break;
266 case IXGBE_TDT(0):
267 for (i = 0; i < 64; i++)
268 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
269 break;
270 case IXGBE_TXDCTL(0):
271 for (i = 0; i < 64; i++)
272 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
273 break;
274 default:
275 pr_info("%-15s %08x\n", reginfo->name,
276 IXGBE_READ_REG(hw, reginfo->ofs));
277 return;
278 }
279
280 for (i = 0; i < 8; i++) {
281 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
282 pr_err("%-15s", rname);
283 for (j = 0; j < 8; j++)
284 pr_cont(" %08x", regs[i*8+j]);
285 pr_cont("\n");
286 }
287
288 }
289
290 /*
291 * ixgbe_dump - Print registers, tx-rings and rx-rings
292 */
293 static void ixgbe_dump(struct ixgbe_adapter *adapter)
294 {
295 struct net_device *netdev = adapter->netdev;
296 struct ixgbe_hw *hw = &adapter->hw;
297 struct ixgbe_reg_info *reginfo;
298 int n = 0;
299 struct ixgbe_ring *tx_ring;
300 struct ixgbe_tx_buffer *tx_buffer;
301 union ixgbe_adv_tx_desc *tx_desc;
302 struct my_u0 { u64 a; u64 b; } *u0;
303 struct ixgbe_ring *rx_ring;
304 union ixgbe_adv_rx_desc *rx_desc;
305 struct ixgbe_rx_buffer *rx_buffer_info;
306 u32 staterr;
307 int i = 0;
308
309 if (!netif_msg_hw(adapter))
310 return;
311
312 /* Print netdevice Info */
313 if (netdev) {
314 dev_info(&adapter->pdev->dev, "Net device Info\n");
315 pr_info("Device Name state "
316 "trans_start last_rx\n");
317 pr_info("%-15s %016lX %016lX %016lX\n",
318 netdev->name,
319 netdev->state,
320 netdev->trans_start,
321 netdev->last_rx);
322 }
323
324 /* Print Registers */
325 dev_info(&adapter->pdev->dev, "Register Dump\n");
326 pr_info(" Register Name Value\n");
327 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
328 reginfo->name; reginfo++) {
329 ixgbe_regdump(hw, reginfo);
330 }
331
332 /* Print TX Ring Summary */
333 if (!netdev || !netif_running(netdev))
334 goto exit;
335
336 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
337 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
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 %04X %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 * Advanced Transmit Descriptor
358 * +--------------------------------------------------------------+
359 * 0 | Buffer Address [63:0] |
360 * +--------------------------------------------------------------+
361 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
362 * +--------------------------------------------------------------+
363 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
364 */
365
366 for (n = 0; n < adapter->num_tx_queues; n++) {
367 tx_ring = adapter->tx_ring[n];
368 pr_info("------------------------------------\n");
369 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
370 pr_info("------------------------------------\n");
371 pr_info("T [desc] [address 63:0 ] "
372 "[PlPOIdStDDt Ln] [bi->dma ] "
373 "leng ntw timestamp bi->skb\n");
374
375 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
376 tx_desc = IXGBE_TX_DESC(tx_ring, i);
377 tx_buffer = &tx_ring->tx_buffer_info[i];
378 u0 = (struct my_u0 *)tx_desc;
379 pr_info("T [0x%03X] %016llX %016llX %016llX"
380 " %04X %p %016llX %p", i,
381 le64_to_cpu(u0->a),
382 le64_to_cpu(u0->b),
383 (u64)dma_unmap_addr(tx_buffer, dma),
384 dma_unmap_len(tx_buffer, len),
385 tx_buffer->next_to_watch,
386 (u64)tx_buffer->time_stamp,
387 tx_buffer->skb);
388 if (i == tx_ring->next_to_use &&
389 i == tx_ring->next_to_clean)
390 pr_cont(" NTC/U\n");
391 else if (i == tx_ring->next_to_use)
392 pr_cont(" NTU\n");
393 else if (i == tx_ring->next_to_clean)
394 pr_cont(" NTC\n");
395 else
396 pr_cont("\n");
397
398 if (netif_msg_pktdata(adapter) &&
399 dma_unmap_len(tx_buffer, len) != 0)
400 print_hex_dump(KERN_INFO, "",
401 DUMP_PREFIX_ADDRESS, 16, 1,
402 phys_to_virt(dma_unmap_addr(tx_buffer,
403 dma)),
404 dma_unmap_len(tx_buffer, len),
405 true);
406 }
407 }
408
409 /* Print RX Rings Summary */
410 rx_ring_summary:
411 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
412 pr_info("Queue [NTU] [NTC]\n");
413 for (n = 0; n < adapter->num_rx_queues; n++) {
414 rx_ring = adapter->rx_ring[n];
415 pr_info("%5d %5X %5X\n",
416 n, rx_ring->next_to_use, rx_ring->next_to_clean);
417 }
418
419 /* Print RX Rings */
420 if (!netif_msg_rx_status(adapter))
421 goto exit;
422
423 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
424
425 /* Advanced Receive Descriptor (Read) Format
426 * 63 1 0
427 * +-----------------------------------------------------+
428 * 0 | Packet Buffer Address [63:1] |A0/NSE|
429 * +----------------------------------------------+------+
430 * 8 | Header Buffer Address [63:1] | DD |
431 * +-----------------------------------------------------+
432 *
433 *
434 * Advanced Receive Descriptor (Write-Back) Format
435 *
436 * 63 48 47 32 31 30 21 20 16 15 4 3 0
437 * +------------------------------------------------------+
438 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
439 * | Checksum Ident | | | | Type | Type |
440 * +------------------------------------------------------+
441 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
442 * +------------------------------------------------------+
443 * 63 48 47 32 31 20 19 0
444 */
445 for (n = 0; n < adapter->num_rx_queues; n++) {
446 rx_ring = adapter->rx_ring[n];
447 pr_info("------------------------------------\n");
448 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
449 pr_info("------------------------------------\n");
450 pr_info("R [desc] [ PktBuf A0] "
451 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
452 "<-- Adv Rx Read format\n");
453 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
454 "[vl er S cks ln] ---------------- [bi->skb] "
455 "<-- Adv Rx Write-Back format\n");
456
457 for (i = 0; i < rx_ring->count; i++) {
458 rx_buffer_info = &rx_ring->rx_buffer_info[i];
459 rx_desc = IXGBE_RX_DESC(rx_ring, i);
460 u0 = (struct my_u0 *)rx_desc;
461 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
462 if (staterr & IXGBE_RXD_STAT_DD) {
463 /* Descriptor Done */
464 pr_info("RWB[0x%03X] %016llX "
465 "%016llX ---------------- %p", i,
466 le64_to_cpu(u0->a),
467 le64_to_cpu(u0->b),
468 rx_buffer_info->skb);
469 } else {
470 pr_info("R [0x%03X] %016llX "
471 "%016llX %016llX %p", i,
472 le64_to_cpu(u0->a),
473 le64_to_cpu(u0->b),
474 (u64)rx_buffer_info->dma,
475 rx_buffer_info->skb);
476
477 if (netif_msg_pktdata(adapter)) {
478 print_hex_dump(KERN_INFO, "",
479 DUMP_PREFIX_ADDRESS, 16, 1,
480 phys_to_virt(rx_buffer_info->dma),
481 ixgbe_rx_bufsz(rx_ring), true);
482 }
483 }
484
485 if (i == rx_ring->next_to_use)
486 pr_cont(" NTU\n");
487 else if (i == rx_ring->next_to_clean)
488 pr_cont(" NTC\n");
489 else
490 pr_cont("\n");
491
492 }
493 }
494
495 exit:
496 return;
497 }
498
499 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
500 {
501 u32 ctrl_ext;
502
503 /* Let firmware take over control of h/w */
504 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
505 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
506 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
507 }
508
509 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
510 {
511 u32 ctrl_ext;
512
513 /* Let firmware know the driver has taken over */
514 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
515 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
516 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
517 }
518
519 /**
520 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
521 * @adapter: pointer to adapter struct
522 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
523 * @queue: queue to map the corresponding interrupt to
524 * @msix_vector: the vector to map to the corresponding queue
525 *
526 */
527 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
528 u8 queue, u8 msix_vector)
529 {
530 u32 ivar, index;
531 struct ixgbe_hw *hw = &adapter->hw;
532 switch (hw->mac.type) {
533 case ixgbe_mac_82598EB:
534 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
535 if (direction == -1)
536 direction = 0;
537 index = (((direction * 64) + queue) >> 2) & 0x1F;
538 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
539 ivar &= ~(0xFF << (8 * (queue & 0x3)));
540 ivar |= (msix_vector << (8 * (queue & 0x3)));
541 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
542 break;
543 case ixgbe_mac_82599EB:
544 case ixgbe_mac_X540:
545 if (direction == -1) {
546 /* other causes */
547 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
548 index = ((queue & 1) * 8);
549 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
550 ivar &= ~(0xFF << index);
551 ivar |= (msix_vector << index);
552 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
553 break;
554 } else {
555 /* tx or rx causes */
556 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
557 index = ((16 * (queue & 1)) + (8 * direction));
558 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
559 ivar &= ~(0xFF << index);
560 ivar |= (msix_vector << index);
561 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
562 break;
563 }
564 default:
565 break;
566 }
567 }
568
569 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
570 u64 qmask)
571 {
572 u32 mask;
573
574 switch (adapter->hw.mac.type) {
575 case ixgbe_mac_82598EB:
576 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
577 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
578 break;
579 case ixgbe_mac_82599EB:
580 case ixgbe_mac_X540:
581 mask = (qmask & 0xFFFFFFFF);
582 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
583 mask = (qmask >> 32);
584 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
585 break;
586 default:
587 break;
588 }
589 }
590
591 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
592 struct ixgbe_tx_buffer *tx_buffer)
593 {
594 if (tx_buffer->skb) {
595 dev_kfree_skb_any(tx_buffer->skb);
596 if (dma_unmap_len(tx_buffer, len))
597 dma_unmap_single(ring->dev,
598 dma_unmap_addr(tx_buffer, dma),
599 dma_unmap_len(tx_buffer, len),
600 DMA_TO_DEVICE);
601 } else if (dma_unmap_len(tx_buffer, len)) {
602 dma_unmap_page(ring->dev,
603 dma_unmap_addr(tx_buffer, dma),
604 dma_unmap_len(tx_buffer, len),
605 DMA_TO_DEVICE);
606 }
607 tx_buffer->next_to_watch = NULL;
608 tx_buffer->skb = NULL;
609 dma_unmap_len_set(tx_buffer, len, 0);
610 /* tx_buffer must be completely set up in the transmit path */
611 }
612
613 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
614 {
615 struct ixgbe_hw *hw = &adapter->hw;
616 struct ixgbe_hw_stats *hwstats = &adapter->stats;
617 int i;
618 u32 data;
619
620 if ((hw->fc.current_mode != ixgbe_fc_full) &&
621 (hw->fc.current_mode != ixgbe_fc_rx_pause))
622 return;
623
624 switch (hw->mac.type) {
625 case ixgbe_mac_82598EB:
626 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
627 break;
628 default:
629 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
630 }
631 hwstats->lxoffrxc += data;
632
633 /* refill credits (no tx hang) if we received xoff */
634 if (!data)
635 return;
636
637 for (i = 0; i < adapter->num_tx_queues; i++)
638 clear_bit(__IXGBE_HANG_CHECK_ARMED,
639 &adapter->tx_ring[i]->state);
640 }
641
642 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
643 {
644 struct ixgbe_hw *hw = &adapter->hw;
645 struct ixgbe_hw_stats *hwstats = &adapter->stats;
646 u32 xoff[8] = {0};
647 int i;
648 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
649
650 if (adapter->ixgbe_ieee_pfc)
651 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
652
653 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
654 ixgbe_update_xoff_rx_lfc(adapter);
655 return;
656 }
657
658 /* update stats for each tc, only valid with PFC enabled */
659 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
660 switch (hw->mac.type) {
661 case ixgbe_mac_82598EB:
662 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
663 break;
664 default:
665 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
666 }
667 hwstats->pxoffrxc[i] += xoff[i];
668 }
669
670 /* disarm tx queues that have received xoff frames */
671 for (i = 0; i < adapter->num_tx_queues; i++) {
672 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
673 u8 tc = tx_ring->dcb_tc;
674
675 if (xoff[tc])
676 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
677 }
678 }
679
680 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
681 {
682 return ring->stats.packets;
683 }
684
685 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
686 {
687 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
688 struct ixgbe_hw *hw = &adapter->hw;
689
690 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
691 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
692
693 if (head != tail)
694 return (head < tail) ?
695 tail - head : (tail + ring->count - head);
696
697 return 0;
698 }
699
700 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
701 {
702 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
703 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
704 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
705 bool ret = false;
706
707 clear_check_for_tx_hang(tx_ring);
708
709 /*
710 * Check for a hung queue, but be thorough. This verifies
711 * that a transmit has been completed since the previous
712 * check AND there is at least one packet pending. The
713 * ARMED bit is set to indicate a potential hang. The
714 * bit is cleared if a pause frame is received to remove
715 * false hang detection due to PFC or 802.3x frames. By
716 * requiring this to fail twice we avoid races with
717 * pfc clearing the ARMED bit and conditions where we
718 * run the check_tx_hang logic with a transmit completion
719 * pending but without time to complete it yet.
720 */
721 if ((tx_done_old == tx_done) && tx_pending) {
722 /* make sure it is true for two checks in a row */
723 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
724 &tx_ring->state);
725 } else {
726 /* update completed stats and continue */
727 tx_ring->tx_stats.tx_done_old = tx_done;
728 /* reset the countdown */
729 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
730 }
731
732 return ret;
733 }
734
735 /**
736 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
737 * @adapter: driver private struct
738 **/
739 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
740 {
741
742 /* Do the reset outside of interrupt context */
743 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
744 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
745 ixgbe_service_event_schedule(adapter);
746 }
747 }
748
749 /**
750 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
751 * @q_vector: structure containing interrupt and ring information
752 * @tx_ring: tx ring to clean
753 **/
754 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
755 struct ixgbe_ring *tx_ring)
756 {
757 struct ixgbe_adapter *adapter = q_vector->adapter;
758 struct ixgbe_tx_buffer *tx_buffer;
759 union ixgbe_adv_tx_desc *tx_desc;
760 unsigned int total_bytes = 0, total_packets = 0;
761 unsigned int budget = q_vector->tx.work_limit;
762 unsigned int i = tx_ring->next_to_clean;
763
764 if (test_bit(__IXGBE_DOWN, &adapter->state))
765 return true;
766
767 tx_buffer = &tx_ring->tx_buffer_info[i];
768 tx_desc = IXGBE_TX_DESC(tx_ring, i);
769 i -= tx_ring->count;
770
771 do {
772 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
773
774 /* if next_to_watch is not set then there is no work pending */
775 if (!eop_desc)
776 break;
777
778 /* prevent any other reads prior to eop_desc */
779 rmb();
780
781 /* if DD is not set pending work has not been completed */
782 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
783 break;
784
785 /* clear next_to_watch to prevent false hangs */
786 tx_buffer->next_to_watch = NULL;
787
788 /* update the statistics for this packet */
789 total_bytes += tx_buffer->bytecount;
790 total_packets += tx_buffer->gso_segs;
791
792 #ifdef CONFIG_IXGBE_PTP
793 if (unlikely(tx_buffer->tx_flags & IXGBE_TX_FLAGS_TSTAMP))
794 ixgbe_ptp_tx_hwtstamp(q_vector, tx_buffer->skb);
795 #endif
796
797 /* free the skb */
798 dev_kfree_skb_any(tx_buffer->skb);
799
800 /* unmap skb header data */
801 dma_unmap_single(tx_ring->dev,
802 dma_unmap_addr(tx_buffer, dma),
803 dma_unmap_len(tx_buffer, len),
804 DMA_TO_DEVICE);
805
806 /* clear tx_buffer data */
807 tx_buffer->skb = NULL;
808 dma_unmap_len_set(tx_buffer, len, 0);
809
810 /* unmap remaining buffers */
811 while (tx_desc != eop_desc) {
812 tx_buffer++;
813 tx_desc++;
814 i++;
815 if (unlikely(!i)) {
816 i -= tx_ring->count;
817 tx_buffer = tx_ring->tx_buffer_info;
818 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
819 }
820
821 /* unmap any remaining paged data */
822 if (dma_unmap_len(tx_buffer, len)) {
823 dma_unmap_page(tx_ring->dev,
824 dma_unmap_addr(tx_buffer, dma),
825 dma_unmap_len(tx_buffer, len),
826 DMA_TO_DEVICE);
827 dma_unmap_len_set(tx_buffer, len, 0);
828 }
829 }
830
831 /* move us one more past the eop_desc for start of next pkt */
832 tx_buffer++;
833 tx_desc++;
834 i++;
835 if (unlikely(!i)) {
836 i -= tx_ring->count;
837 tx_buffer = tx_ring->tx_buffer_info;
838 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
839 }
840
841 /* issue prefetch for next Tx descriptor */
842 prefetch(tx_desc);
843
844 /* update budget accounting */
845 budget--;
846 } while (likely(budget));
847
848 i += tx_ring->count;
849 tx_ring->next_to_clean = i;
850 u64_stats_update_begin(&tx_ring->syncp);
851 tx_ring->stats.bytes += total_bytes;
852 tx_ring->stats.packets += total_packets;
853 u64_stats_update_end(&tx_ring->syncp);
854 q_vector->tx.total_bytes += total_bytes;
855 q_vector->tx.total_packets += total_packets;
856
857 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
858 /* schedule immediate reset if we believe we hung */
859 struct ixgbe_hw *hw = &adapter->hw;
860 e_err(drv, "Detected Tx Unit Hang\n"
861 " Tx Queue <%d>\n"
862 " TDH, TDT <%x>, <%x>\n"
863 " next_to_use <%x>\n"
864 " next_to_clean <%x>\n"
865 "tx_buffer_info[next_to_clean]\n"
866 " time_stamp <%lx>\n"
867 " jiffies <%lx>\n",
868 tx_ring->queue_index,
869 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
870 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
871 tx_ring->next_to_use, i,
872 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
873
874 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
875
876 e_info(probe,
877 "tx hang %d detected on queue %d, resetting adapter\n",
878 adapter->tx_timeout_count + 1, tx_ring->queue_index);
879
880 /* schedule immediate reset if we believe we hung */
881 ixgbe_tx_timeout_reset(adapter);
882
883 /* the adapter is about to reset, no point in enabling stuff */
884 return true;
885 }
886
887 netdev_tx_completed_queue(txring_txq(tx_ring),
888 total_packets, total_bytes);
889
890 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
891 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
892 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
893 /* Make sure that anybody stopping the queue after this
894 * sees the new next_to_clean.
895 */
896 smp_mb();
897 if (__netif_subqueue_stopped(tx_ring->netdev,
898 tx_ring->queue_index)
899 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
900 netif_wake_subqueue(tx_ring->netdev,
901 tx_ring->queue_index);
902 ++tx_ring->tx_stats.restart_queue;
903 }
904 }
905
906 return !!budget;
907 }
908
909 #ifdef CONFIG_IXGBE_DCA
910 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
911 struct ixgbe_ring *tx_ring,
912 int cpu)
913 {
914 struct ixgbe_hw *hw = &adapter->hw;
915 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
916 u16 reg_offset;
917
918 switch (hw->mac.type) {
919 case ixgbe_mac_82598EB:
920 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
921 break;
922 case ixgbe_mac_82599EB:
923 case ixgbe_mac_X540:
924 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
925 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
926 break;
927 default:
928 /* for unknown hardware do not write register */
929 return;
930 }
931
932 /*
933 * We can enable relaxed ordering for reads, but not writes when
934 * DCA is enabled. This is due to a known issue in some chipsets
935 * which will cause the DCA tag to be cleared.
936 */
937 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
938 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
939 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
940
941 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
942 }
943
944 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
945 struct ixgbe_ring *rx_ring,
946 int cpu)
947 {
948 struct ixgbe_hw *hw = &adapter->hw;
949 u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
950 u8 reg_idx = rx_ring->reg_idx;
951
952
953 switch (hw->mac.type) {
954 case ixgbe_mac_82599EB:
955 case ixgbe_mac_X540:
956 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
957 break;
958 default:
959 break;
960 }
961
962 /*
963 * We can enable relaxed ordering for reads, but not writes when
964 * DCA is enabled. This is due to a known issue in some chipsets
965 * which will cause the DCA tag to be cleared.
966 */
967 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
968 IXGBE_DCA_RXCTRL_DATA_DCA_EN |
969 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
970
971 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
972 }
973
974 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
975 {
976 struct ixgbe_adapter *adapter = q_vector->adapter;
977 struct ixgbe_ring *ring;
978 int cpu = get_cpu();
979
980 if (q_vector->cpu == cpu)
981 goto out_no_update;
982
983 ixgbe_for_each_ring(ring, q_vector->tx)
984 ixgbe_update_tx_dca(adapter, ring, cpu);
985
986 ixgbe_for_each_ring(ring, q_vector->rx)
987 ixgbe_update_rx_dca(adapter, ring, cpu);
988
989 q_vector->cpu = cpu;
990 out_no_update:
991 put_cpu();
992 }
993
994 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
995 {
996 int i;
997
998 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
999 return;
1000
1001 /* always use CB2 mode, difference is masked in the CB driver */
1002 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1003
1004 for (i = 0; i < adapter->num_q_vectors; i++) {
1005 adapter->q_vector[i]->cpu = -1;
1006 ixgbe_update_dca(adapter->q_vector[i]);
1007 }
1008 }
1009
1010 static int __ixgbe_notify_dca(struct device *dev, void *data)
1011 {
1012 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1013 unsigned long event = *(unsigned long *)data;
1014
1015 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1016 return 0;
1017
1018 switch (event) {
1019 case DCA_PROVIDER_ADD:
1020 /* if we're already enabled, don't do it again */
1021 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1022 break;
1023 if (dca_add_requester(dev) == 0) {
1024 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1025 ixgbe_setup_dca(adapter);
1026 break;
1027 }
1028 /* Fall Through since DCA is disabled. */
1029 case DCA_PROVIDER_REMOVE:
1030 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1031 dca_remove_requester(dev);
1032 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1033 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1034 }
1035 break;
1036 }
1037
1038 return 0;
1039 }
1040
1041 #endif /* CONFIG_IXGBE_DCA */
1042 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1043 union ixgbe_adv_rx_desc *rx_desc,
1044 struct sk_buff *skb)
1045 {
1046 if (ring->netdev->features & NETIF_F_RXHASH)
1047 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1048 }
1049
1050 #ifdef IXGBE_FCOE
1051 /**
1052 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1053 * @ring: structure containing ring specific data
1054 * @rx_desc: advanced rx descriptor
1055 *
1056 * Returns : true if it is FCoE pkt
1057 */
1058 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1059 union ixgbe_adv_rx_desc *rx_desc)
1060 {
1061 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1062
1063 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1064 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1065 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1066 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1067 }
1068
1069 #endif /* IXGBE_FCOE */
1070 /**
1071 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1072 * @ring: structure containing ring specific data
1073 * @rx_desc: current Rx descriptor being processed
1074 * @skb: skb currently being received and modified
1075 **/
1076 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1077 union ixgbe_adv_rx_desc *rx_desc,
1078 struct sk_buff *skb)
1079 {
1080 skb_checksum_none_assert(skb);
1081
1082 /* Rx csum disabled */
1083 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1084 return;
1085
1086 /* if IP and error */
1087 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1088 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1089 ring->rx_stats.csum_err++;
1090 return;
1091 }
1092
1093 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1094 return;
1095
1096 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1097 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1098
1099 /*
1100 * 82599 errata, UDP frames with a 0 checksum can be marked as
1101 * checksum errors.
1102 */
1103 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1104 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1105 return;
1106
1107 ring->rx_stats.csum_err++;
1108 return;
1109 }
1110
1111 /* It must be a TCP or UDP packet with a valid checksum */
1112 skb->ip_summed = CHECKSUM_UNNECESSARY;
1113 }
1114
1115 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1116 {
1117 rx_ring->next_to_use = val;
1118
1119 /* update next to alloc since we have filled the ring */
1120 rx_ring->next_to_alloc = val;
1121 /*
1122 * Force memory writes to complete before letting h/w
1123 * know there are new descriptors to fetch. (Only
1124 * applicable for weak-ordered memory model archs,
1125 * such as IA-64).
1126 */
1127 wmb();
1128 writel(val, rx_ring->tail);
1129 }
1130
1131 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1132 struct ixgbe_rx_buffer *bi)
1133 {
1134 struct page *page = bi->page;
1135 dma_addr_t dma = bi->dma;
1136
1137 /* since we are recycling buffers we should seldom need to alloc */
1138 if (likely(dma))
1139 return true;
1140
1141 /* alloc new page for storage */
1142 if (likely(!page)) {
1143 page = alloc_pages(GFP_ATOMIC | __GFP_COLD | __GFP_COMP,
1144 ixgbe_rx_pg_order(rx_ring));
1145 if (unlikely(!page)) {
1146 rx_ring->rx_stats.alloc_rx_page_failed++;
1147 return false;
1148 }
1149 bi->page = page;
1150 }
1151
1152 /* map page for use */
1153 dma = dma_map_page(rx_ring->dev, page, 0,
1154 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1155
1156 /*
1157 * if mapping failed free memory back to system since
1158 * there isn't much point in holding memory we can't use
1159 */
1160 if (dma_mapping_error(rx_ring->dev, dma)) {
1161 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1162 bi->page = NULL;
1163
1164 rx_ring->rx_stats.alloc_rx_page_failed++;
1165 return false;
1166 }
1167
1168 bi->dma = dma;
1169 bi->page_offset ^= ixgbe_rx_bufsz(rx_ring);
1170
1171 return true;
1172 }
1173
1174 /**
1175 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1176 * @rx_ring: ring to place buffers on
1177 * @cleaned_count: number of buffers to replace
1178 **/
1179 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1180 {
1181 union ixgbe_adv_rx_desc *rx_desc;
1182 struct ixgbe_rx_buffer *bi;
1183 u16 i = rx_ring->next_to_use;
1184
1185 /* nothing to do */
1186 if (!cleaned_count)
1187 return;
1188
1189 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1190 bi = &rx_ring->rx_buffer_info[i];
1191 i -= rx_ring->count;
1192
1193 do {
1194 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1195 break;
1196
1197 /*
1198 * Refresh the desc even if buffer_addrs didn't change
1199 * because each write-back erases this info.
1200 */
1201 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1202
1203 rx_desc++;
1204 bi++;
1205 i++;
1206 if (unlikely(!i)) {
1207 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1208 bi = rx_ring->rx_buffer_info;
1209 i -= rx_ring->count;
1210 }
1211
1212 /* clear the hdr_addr for the next_to_use descriptor */
1213 rx_desc->read.hdr_addr = 0;
1214
1215 cleaned_count--;
1216 } while (cleaned_count);
1217
1218 i += rx_ring->count;
1219
1220 if (rx_ring->next_to_use != i)
1221 ixgbe_release_rx_desc(rx_ring, i);
1222 }
1223
1224 /**
1225 * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1226 * @data: pointer to the start of the headers
1227 * @max_len: total length of section to find headers in
1228 *
1229 * This function is meant to determine the length of headers that will
1230 * be recognized by hardware for LRO, GRO, and RSC offloads. The main
1231 * motivation of doing this is to only perform one pull for IPv4 TCP
1232 * packets so that we can do basic things like calculating the gso_size
1233 * based on the average data per packet.
1234 **/
1235 static unsigned int ixgbe_get_headlen(unsigned char *data,
1236 unsigned int max_len)
1237 {
1238 union {
1239 unsigned char *network;
1240 /* l2 headers */
1241 struct ethhdr *eth;
1242 struct vlan_hdr *vlan;
1243 /* l3 headers */
1244 struct iphdr *ipv4;
1245 } hdr;
1246 __be16 protocol;
1247 u8 nexthdr = 0; /* default to not TCP */
1248 u8 hlen;
1249
1250 /* this should never happen, but better safe than sorry */
1251 if (max_len < ETH_HLEN)
1252 return max_len;
1253
1254 /* initialize network frame pointer */
1255 hdr.network = data;
1256
1257 /* set first protocol and move network header forward */
1258 protocol = hdr.eth->h_proto;
1259 hdr.network += ETH_HLEN;
1260
1261 /* handle any vlan tag if present */
1262 if (protocol == __constant_htons(ETH_P_8021Q)) {
1263 if ((hdr.network - data) > (max_len - VLAN_HLEN))
1264 return max_len;
1265
1266 protocol = hdr.vlan->h_vlan_encapsulated_proto;
1267 hdr.network += VLAN_HLEN;
1268 }
1269
1270 /* handle L3 protocols */
1271 if (protocol == __constant_htons(ETH_P_IP)) {
1272 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1273 return max_len;
1274
1275 /* access ihl as a u8 to avoid unaligned access on ia64 */
1276 hlen = (hdr.network[0] & 0x0F) << 2;
1277
1278 /* verify hlen meets minimum size requirements */
1279 if (hlen < sizeof(struct iphdr))
1280 return hdr.network - data;
1281
1282 /* record next protocol */
1283 nexthdr = hdr.ipv4->protocol;
1284 hdr.network += hlen;
1285 #ifdef IXGBE_FCOE
1286 } else if (protocol == __constant_htons(ETH_P_FCOE)) {
1287 if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1288 return max_len;
1289 hdr.network += FCOE_HEADER_LEN;
1290 #endif
1291 } else {
1292 return hdr.network - data;
1293 }
1294
1295 /* finally sort out TCP */
1296 if (nexthdr == IPPROTO_TCP) {
1297 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1298 return max_len;
1299
1300 /* access doff as a u8 to avoid unaligned access on ia64 */
1301 hlen = (hdr.network[12] & 0xF0) >> 2;
1302
1303 /* verify hlen meets minimum size requirements */
1304 if (hlen < sizeof(struct tcphdr))
1305 return hdr.network - data;
1306
1307 hdr.network += hlen;
1308 }
1309
1310 /*
1311 * If everything has gone correctly hdr.network should be the
1312 * data section of the packet and will be the end of the header.
1313 * If not then it probably represents the end of the last recognized
1314 * header.
1315 */
1316 if ((hdr.network - data) < max_len)
1317 return hdr.network - data;
1318 else
1319 return max_len;
1320 }
1321
1322 static void ixgbe_get_rsc_cnt(struct ixgbe_ring *rx_ring,
1323 union ixgbe_adv_rx_desc *rx_desc,
1324 struct sk_buff *skb)
1325 {
1326 __le32 rsc_enabled;
1327 u32 rsc_cnt;
1328
1329 if (!ring_is_rsc_enabled(rx_ring))
1330 return;
1331
1332 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1333 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1334
1335 /* If this is an RSC frame rsc_cnt should be non-zero */
1336 if (!rsc_enabled)
1337 return;
1338
1339 rsc_cnt = le32_to_cpu(rsc_enabled);
1340 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1341
1342 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1343 }
1344
1345 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1346 struct sk_buff *skb)
1347 {
1348 u16 hdr_len = skb_headlen(skb);
1349
1350 /* set gso_size to avoid messing up TCP MSS */
1351 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1352 IXGBE_CB(skb)->append_cnt);
1353 }
1354
1355 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1356 struct sk_buff *skb)
1357 {
1358 /* if append_cnt is 0 then frame is not RSC */
1359 if (!IXGBE_CB(skb)->append_cnt)
1360 return;
1361
1362 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1363 rx_ring->rx_stats.rsc_flush++;
1364
1365 ixgbe_set_rsc_gso_size(rx_ring, skb);
1366
1367 /* gso_size is computed using append_cnt so always clear it last */
1368 IXGBE_CB(skb)->append_cnt = 0;
1369 }
1370
1371 /**
1372 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1373 * @rx_ring: rx descriptor ring packet is being transacted on
1374 * @rx_desc: pointer to the EOP Rx descriptor
1375 * @skb: pointer to current skb being populated
1376 *
1377 * This function checks the ring, descriptor, and packet information in
1378 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1379 * other fields within the skb.
1380 **/
1381 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1382 union ixgbe_adv_rx_desc *rx_desc,
1383 struct sk_buff *skb)
1384 {
1385 struct net_device *dev = rx_ring->netdev;
1386
1387 ixgbe_update_rsc_stats(rx_ring, skb);
1388
1389 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1390
1391 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1392
1393 #ifdef CONFIG_IXGBE_PTP
1394 ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector, rx_desc, skb);
1395 #endif
1396
1397 if ((dev->features & NETIF_F_HW_VLAN_RX) &&
1398 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1399 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1400 __vlan_hwaccel_put_tag(skb, vid);
1401 }
1402
1403 skb_record_rx_queue(skb, rx_ring->queue_index);
1404
1405 skb->protocol = eth_type_trans(skb, dev);
1406 }
1407
1408 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1409 struct sk_buff *skb)
1410 {
1411 struct ixgbe_adapter *adapter = q_vector->adapter;
1412
1413 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1414 napi_gro_receive(&q_vector->napi, skb);
1415 else
1416 netif_rx(skb);
1417 }
1418
1419 /**
1420 * ixgbe_is_non_eop - process handling of non-EOP buffers
1421 * @rx_ring: Rx ring being processed
1422 * @rx_desc: Rx descriptor for current buffer
1423 * @skb: Current socket buffer containing buffer in progress
1424 *
1425 * This function updates next to clean. If the buffer is an EOP buffer
1426 * this function exits returning false, otherwise it will place the
1427 * sk_buff in the next buffer to be chained and return true indicating
1428 * that this is in fact a non-EOP buffer.
1429 **/
1430 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1431 union ixgbe_adv_rx_desc *rx_desc,
1432 struct sk_buff *skb)
1433 {
1434 u32 ntc = rx_ring->next_to_clean + 1;
1435
1436 /* fetch, update, and store next to clean */
1437 ntc = (ntc < rx_ring->count) ? ntc : 0;
1438 rx_ring->next_to_clean = ntc;
1439
1440 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1441
1442 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1443 return false;
1444
1445 /* append_cnt indicates packet is RSC, if so fetch nextp */
1446 if (IXGBE_CB(skb)->append_cnt) {
1447 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1448 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1449 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1450 }
1451
1452 /* place skb in next buffer to be received */
1453 rx_ring->rx_buffer_info[ntc].skb = skb;
1454 rx_ring->rx_stats.non_eop_descs++;
1455
1456 return true;
1457 }
1458
1459 /**
1460 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1461 * @rx_ring: rx descriptor ring packet is being transacted on
1462 * @rx_desc: pointer to the EOP Rx descriptor
1463 * @skb: pointer to current skb being fixed
1464 *
1465 * Check for corrupted packet headers caused by senders on the local L2
1466 * embedded NIC switch not setting up their Tx Descriptors right. These
1467 * should be very rare.
1468 *
1469 * Also address the case where we are pulling data in on pages only
1470 * and as such no data is present in the skb header.
1471 *
1472 * In addition if skb is not at least 60 bytes we need to pad it so that
1473 * it is large enough to qualify as a valid Ethernet frame.
1474 *
1475 * Returns true if an error was encountered and skb was freed.
1476 **/
1477 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1478 union ixgbe_adv_rx_desc *rx_desc,
1479 struct sk_buff *skb)
1480 {
1481 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1482 struct net_device *netdev = rx_ring->netdev;
1483 unsigned char *va;
1484 unsigned int pull_len;
1485
1486 /* if the page was released unmap it, else just sync our portion */
1487 if (unlikely(IXGBE_CB(skb)->page_released)) {
1488 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1489 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1490 IXGBE_CB(skb)->page_released = false;
1491 } else {
1492 dma_sync_single_range_for_cpu(rx_ring->dev,
1493 IXGBE_CB(skb)->dma,
1494 frag->page_offset,
1495 ixgbe_rx_bufsz(rx_ring),
1496 DMA_FROM_DEVICE);
1497 }
1498 IXGBE_CB(skb)->dma = 0;
1499
1500 /* verify that the packet does not have any known errors */
1501 if (unlikely(ixgbe_test_staterr(rx_desc,
1502 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1503 !(netdev->features & NETIF_F_RXALL))) {
1504 dev_kfree_skb_any(skb);
1505 return true;
1506 }
1507
1508 /*
1509 * it is valid to use page_address instead of kmap since we are
1510 * working with pages allocated out of the lomem pool per
1511 * alloc_page(GFP_ATOMIC)
1512 */
1513 va = skb_frag_address(frag);
1514
1515 /*
1516 * we need the header to contain the greater of either ETH_HLEN or
1517 * 60 bytes if the skb->len is less than 60 for skb_pad.
1518 */
1519 pull_len = skb_frag_size(frag);
1520 if (pull_len > 256)
1521 pull_len = ixgbe_get_headlen(va, pull_len);
1522
1523 /* align pull length to size of long to optimize memcpy performance */
1524 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1525
1526 /* update all of the pointers */
1527 skb_frag_size_sub(frag, pull_len);
1528 frag->page_offset += pull_len;
1529 skb->data_len -= pull_len;
1530 skb->tail += pull_len;
1531
1532 /*
1533 * if we sucked the frag empty then we should free it,
1534 * if there are other frags here something is screwed up in hardware
1535 */
1536 if (skb_frag_size(frag) == 0) {
1537 BUG_ON(skb_shinfo(skb)->nr_frags != 1);
1538 skb_shinfo(skb)->nr_frags = 0;
1539 __skb_frag_unref(frag);
1540 skb->truesize -= ixgbe_rx_bufsz(rx_ring);
1541 }
1542
1543 #ifdef IXGBE_FCOE
1544 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1545 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1546 return false;
1547
1548 #endif
1549 /* if skb_pad returns an error the skb was freed */
1550 if (unlikely(skb->len < 60)) {
1551 int pad_len = 60 - skb->len;
1552
1553 if (skb_pad(skb, pad_len))
1554 return true;
1555 __skb_put(skb, pad_len);
1556 }
1557
1558 return false;
1559 }
1560
1561 /**
1562 * ixgbe_can_reuse_page - determine if we can reuse a page
1563 * @rx_buffer: pointer to rx_buffer containing the page we want to reuse
1564 *
1565 * Returns true if page can be reused in another Rx buffer
1566 **/
1567 static inline bool ixgbe_can_reuse_page(struct ixgbe_rx_buffer *rx_buffer)
1568 {
1569 struct page *page = rx_buffer->page;
1570
1571 /* if we are only owner of page and it is local we can reuse it */
1572 return likely(page_count(page) == 1) &&
1573 likely(page_to_nid(page) == numa_node_id());
1574 }
1575
1576 /**
1577 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1578 * @rx_ring: rx descriptor ring to store buffers on
1579 * @old_buff: donor buffer to have page reused
1580 *
1581 * Syncronizes page for reuse by the adapter
1582 **/
1583 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1584 struct ixgbe_rx_buffer *old_buff)
1585 {
1586 struct ixgbe_rx_buffer *new_buff;
1587 u16 nta = rx_ring->next_to_alloc;
1588 u16 bufsz = ixgbe_rx_bufsz(rx_ring);
1589
1590 new_buff = &rx_ring->rx_buffer_info[nta];
1591
1592 /* update, and store next to alloc */
1593 nta++;
1594 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1595
1596 /* transfer page from old buffer to new buffer */
1597 new_buff->page = old_buff->page;
1598 new_buff->dma = old_buff->dma;
1599
1600 /* flip page offset to other buffer and store to new_buff */
1601 new_buff->page_offset = old_buff->page_offset ^ bufsz;
1602
1603 /* sync the buffer for use by the device */
1604 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1605 new_buff->page_offset, bufsz,
1606 DMA_FROM_DEVICE);
1607
1608 /* bump ref count on page before it is given to the stack */
1609 get_page(new_buff->page);
1610 }
1611
1612 /**
1613 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1614 * @rx_ring: rx descriptor ring to transact packets on
1615 * @rx_buffer: buffer containing page to add
1616 * @rx_desc: descriptor containing length of buffer written by hardware
1617 * @skb: sk_buff to place the data into
1618 *
1619 * This function is based on skb_add_rx_frag. I would have used that
1620 * function however it doesn't handle the truesize case correctly since we
1621 * are allocating more memory than might be used for a single receive.
1622 **/
1623 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1624 struct ixgbe_rx_buffer *rx_buffer,
1625 struct sk_buff *skb, int size)
1626 {
1627 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1628 rx_buffer->page, rx_buffer->page_offset,
1629 size);
1630 skb->len += size;
1631 skb->data_len += size;
1632 skb->truesize += ixgbe_rx_bufsz(rx_ring);
1633 }
1634
1635 /**
1636 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1637 * @q_vector: structure containing interrupt and ring information
1638 * @rx_ring: rx descriptor ring to transact packets on
1639 * @budget: Total limit on number of packets to process
1640 *
1641 * This function provides a "bounce buffer" approach to Rx interrupt
1642 * processing. The advantage to this is that on systems that have
1643 * expensive overhead for IOMMU access this provides a means of avoiding
1644 * it by maintaining the mapping of the page to the syste.
1645 *
1646 * Returns true if all work is completed without reaching budget
1647 **/
1648 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1649 struct ixgbe_ring *rx_ring,
1650 int budget)
1651 {
1652 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1653 #ifdef IXGBE_FCOE
1654 struct ixgbe_adapter *adapter = q_vector->adapter;
1655 int ddp_bytes = 0;
1656 #endif /* IXGBE_FCOE */
1657 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1658
1659 do {
1660 struct ixgbe_rx_buffer *rx_buffer;
1661 union ixgbe_adv_rx_desc *rx_desc;
1662 struct sk_buff *skb;
1663 struct page *page;
1664 u16 ntc;
1665
1666 /* return some buffers to hardware, one at a time is too slow */
1667 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1668 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1669 cleaned_count = 0;
1670 }
1671
1672 ntc = rx_ring->next_to_clean;
1673 rx_desc = IXGBE_RX_DESC(rx_ring, ntc);
1674 rx_buffer = &rx_ring->rx_buffer_info[ntc];
1675
1676 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
1677 break;
1678
1679 /*
1680 * This memory barrier is needed to keep us from reading
1681 * any other fields out of the rx_desc until we know the
1682 * RXD_STAT_DD bit is set
1683 */
1684 rmb();
1685
1686 page = rx_buffer->page;
1687 prefetchw(page);
1688
1689 skb = rx_buffer->skb;
1690
1691 if (likely(!skb)) {
1692 void *page_addr = page_address(page) +
1693 rx_buffer->page_offset;
1694
1695 /* prefetch first cache line of first page */
1696 prefetch(page_addr);
1697 #if L1_CACHE_BYTES < 128
1698 prefetch(page_addr + L1_CACHE_BYTES);
1699 #endif
1700
1701 /* allocate a skb to store the frags */
1702 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1703 IXGBE_RX_HDR_SIZE);
1704 if (unlikely(!skb)) {
1705 rx_ring->rx_stats.alloc_rx_buff_failed++;
1706 break;
1707 }
1708
1709 /*
1710 * we will be copying header into skb->data in
1711 * pskb_may_pull so it is in our interest to prefetch
1712 * it now to avoid a possible cache miss
1713 */
1714 prefetchw(skb->data);
1715
1716 /*
1717 * Delay unmapping of the first packet. It carries the
1718 * header information, HW may still access the header
1719 * after the writeback. Only unmap it when EOP is
1720 * reached
1721 */
1722 IXGBE_CB(skb)->dma = rx_buffer->dma;
1723 } else {
1724 /* we are reusing so sync this buffer for CPU use */
1725 dma_sync_single_range_for_cpu(rx_ring->dev,
1726 rx_buffer->dma,
1727 rx_buffer->page_offset,
1728 ixgbe_rx_bufsz(rx_ring),
1729 DMA_FROM_DEVICE);
1730 }
1731
1732 /* pull page into skb */
1733 ixgbe_add_rx_frag(rx_ring, rx_buffer, skb,
1734 le16_to_cpu(rx_desc->wb.upper.length));
1735
1736 if (ixgbe_can_reuse_page(rx_buffer)) {
1737 /* hand second half of page back to the ring */
1738 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1739 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1740 /* the page has been released from the ring */
1741 IXGBE_CB(skb)->page_released = true;
1742 } else {
1743 /* we are not reusing the buffer so unmap it */
1744 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1745 ixgbe_rx_pg_size(rx_ring),
1746 DMA_FROM_DEVICE);
1747 }
1748
1749 /* clear contents of buffer_info */
1750 rx_buffer->skb = NULL;
1751 rx_buffer->dma = 0;
1752 rx_buffer->page = NULL;
1753
1754 ixgbe_get_rsc_cnt(rx_ring, rx_desc, skb);
1755
1756 cleaned_count++;
1757
1758 /* place incomplete frames back on ring for completion */
1759 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
1760 continue;
1761
1762 /* verify the packet layout is correct */
1763 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
1764 continue;
1765
1766 /* probably a little skewed due to removing CRC */
1767 total_rx_bytes += skb->len;
1768 total_rx_packets++;
1769
1770 /* populate checksum, timestamp, VLAN, and protocol */
1771 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
1772
1773 #ifdef IXGBE_FCOE
1774 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1775 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
1776 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1777 if (!ddp_bytes) {
1778 dev_kfree_skb_any(skb);
1779 continue;
1780 }
1781 }
1782
1783 #endif /* IXGBE_FCOE */
1784 ixgbe_rx_skb(q_vector, skb);
1785
1786 /* update budget accounting */
1787 budget--;
1788 } while (likely(budget));
1789
1790 #ifdef IXGBE_FCOE
1791 /* include DDPed FCoE data */
1792 if (ddp_bytes > 0) {
1793 unsigned int mss;
1794
1795 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1796 sizeof(struct fc_frame_header) -
1797 sizeof(struct fcoe_crc_eof);
1798 if (mss > 512)
1799 mss &= ~511;
1800 total_rx_bytes += ddp_bytes;
1801 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1802 }
1803
1804 #endif /* IXGBE_FCOE */
1805 u64_stats_update_begin(&rx_ring->syncp);
1806 rx_ring->stats.packets += total_rx_packets;
1807 rx_ring->stats.bytes += total_rx_bytes;
1808 u64_stats_update_end(&rx_ring->syncp);
1809 q_vector->rx.total_packets += total_rx_packets;
1810 q_vector->rx.total_bytes += total_rx_bytes;
1811
1812 if (cleaned_count)
1813 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1814
1815 return !!budget;
1816 }
1817
1818 /**
1819 * ixgbe_configure_msix - Configure MSI-X hardware
1820 * @adapter: board private structure
1821 *
1822 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1823 * interrupts.
1824 **/
1825 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1826 {
1827 struct ixgbe_q_vector *q_vector;
1828 int v_idx;
1829 u32 mask;
1830
1831 /* Populate MSIX to EITR Select */
1832 if (adapter->num_vfs > 32) {
1833 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1834 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1835 }
1836
1837 /*
1838 * Populate the IVAR table and set the ITR values to the
1839 * corresponding register.
1840 */
1841 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
1842 struct ixgbe_ring *ring;
1843 q_vector = adapter->q_vector[v_idx];
1844
1845 ixgbe_for_each_ring(ring, q_vector->rx)
1846 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1847
1848 ixgbe_for_each_ring(ring, q_vector->tx)
1849 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1850
1851 if (q_vector->tx.ring && !q_vector->rx.ring) {
1852 /* tx only vector */
1853 if (adapter->tx_itr_setting == 1)
1854 q_vector->itr = IXGBE_10K_ITR;
1855 else
1856 q_vector->itr = adapter->tx_itr_setting;
1857 } else {
1858 /* rx or rx/tx vector */
1859 if (adapter->rx_itr_setting == 1)
1860 q_vector->itr = IXGBE_20K_ITR;
1861 else
1862 q_vector->itr = adapter->rx_itr_setting;
1863 }
1864
1865 ixgbe_write_eitr(q_vector);
1866 }
1867
1868 switch (adapter->hw.mac.type) {
1869 case ixgbe_mac_82598EB:
1870 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1871 v_idx);
1872 break;
1873 case ixgbe_mac_82599EB:
1874 case ixgbe_mac_X540:
1875 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1876 break;
1877 default:
1878 break;
1879 }
1880 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1881
1882 /* set up to autoclear timer, and the vectors */
1883 mask = IXGBE_EIMS_ENABLE_MASK;
1884 mask &= ~(IXGBE_EIMS_OTHER |
1885 IXGBE_EIMS_MAILBOX |
1886 IXGBE_EIMS_LSC);
1887
1888 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1889 }
1890
1891 enum latency_range {
1892 lowest_latency = 0,
1893 low_latency = 1,
1894 bulk_latency = 2,
1895 latency_invalid = 255
1896 };
1897
1898 /**
1899 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1900 * @q_vector: structure containing interrupt and ring information
1901 * @ring_container: structure containing ring performance data
1902 *
1903 * Stores a new ITR value based on packets and byte
1904 * counts during the last interrupt. The advantage of per interrupt
1905 * computation is faster updates and more accurate ITR for the current
1906 * traffic pattern. Constants in this function were computed
1907 * based on theoretical maximum wire speed and thresholds were set based
1908 * on testing data as well as attempting to minimize response time
1909 * while increasing bulk throughput.
1910 * this functionality is controlled by the InterruptThrottleRate module
1911 * parameter (see ixgbe_param.c)
1912 **/
1913 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1914 struct ixgbe_ring_container *ring_container)
1915 {
1916 int bytes = ring_container->total_bytes;
1917 int packets = ring_container->total_packets;
1918 u32 timepassed_us;
1919 u64 bytes_perint;
1920 u8 itr_setting = ring_container->itr;
1921
1922 if (packets == 0)
1923 return;
1924
1925 /* simple throttlerate management
1926 * 0-10MB/s lowest (100000 ints/s)
1927 * 10-20MB/s low (20000 ints/s)
1928 * 20-1249MB/s bulk (8000 ints/s)
1929 */
1930 /* what was last interrupt timeslice? */
1931 timepassed_us = q_vector->itr >> 2;
1932 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1933
1934 switch (itr_setting) {
1935 case lowest_latency:
1936 if (bytes_perint > 10)
1937 itr_setting = low_latency;
1938 break;
1939 case low_latency:
1940 if (bytes_perint > 20)
1941 itr_setting = bulk_latency;
1942 else if (bytes_perint <= 10)
1943 itr_setting = lowest_latency;
1944 break;
1945 case bulk_latency:
1946 if (bytes_perint <= 20)
1947 itr_setting = low_latency;
1948 break;
1949 }
1950
1951 /* clear work counters since we have the values we need */
1952 ring_container->total_bytes = 0;
1953 ring_container->total_packets = 0;
1954
1955 /* write updated itr to ring container */
1956 ring_container->itr = itr_setting;
1957 }
1958
1959 /**
1960 * ixgbe_write_eitr - write EITR register in hardware specific way
1961 * @q_vector: structure containing interrupt and ring information
1962 *
1963 * This function is made to be called by ethtool and by the driver
1964 * when it needs to update EITR registers at runtime. Hardware
1965 * specific quirks/differences are taken care of here.
1966 */
1967 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1968 {
1969 struct ixgbe_adapter *adapter = q_vector->adapter;
1970 struct ixgbe_hw *hw = &adapter->hw;
1971 int v_idx = q_vector->v_idx;
1972 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1973
1974 switch (adapter->hw.mac.type) {
1975 case ixgbe_mac_82598EB:
1976 /* must write high and low 16 bits to reset counter */
1977 itr_reg |= (itr_reg << 16);
1978 break;
1979 case ixgbe_mac_82599EB:
1980 case ixgbe_mac_X540:
1981 /*
1982 * set the WDIS bit to not clear the timer bits and cause an
1983 * immediate assertion of the interrupt
1984 */
1985 itr_reg |= IXGBE_EITR_CNT_WDIS;
1986 break;
1987 default:
1988 break;
1989 }
1990 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1991 }
1992
1993 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1994 {
1995 u32 new_itr = q_vector->itr;
1996 u8 current_itr;
1997
1998 ixgbe_update_itr(q_vector, &q_vector->tx);
1999 ixgbe_update_itr(q_vector, &q_vector->rx);
2000
2001 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2002
2003 switch (current_itr) {
2004 /* counts and packets in update_itr are dependent on these numbers */
2005 case lowest_latency:
2006 new_itr = IXGBE_100K_ITR;
2007 break;
2008 case low_latency:
2009 new_itr = IXGBE_20K_ITR;
2010 break;
2011 case bulk_latency:
2012 new_itr = IXGBE_8K_ITR;
2013 break;
2014 default:
2015 break;
2016 }
2017
2018 if (new_itr != q_vector->itr) {
2019 /* do an exponential smoothing */
2020 new_itr = (10 * new_itr * q_vector->itr) /
2021 ((9 * new_itr) + q_vector->itr);
2022
2023 /* save the algorithm value here */
2024 q_vector->itr = new_itr;
2025
2026 ixgbe_write_eitr(q_vector);
2027 }
2028 }
2029
2030 /**
2031 * ixgbe_check_overtemp_subtask - check for over temperature
2032 * @adapter: pointer to adapter
2033 **/
2034 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2035 {
2036 struct ixgbe_hw *hw = &adapter->hw;
2037 u32 eicr = adapter->interrupt_event;
2038
2039 if (test_bit(__IXGBE_DOWN, &adapter->state))
2040 return;
2041
2042 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2043 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2044 return;
2045
2046 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2047
2048 switch (hw->device_id) {
2049 case IXGBE_DEV_ID_82599_T3_LOM:
2050 /*
2051 * Since the warning interrupt is for both ports
2052 * we don't have to check if:
2053 * - This interrupt wasn't for our port.
2054 * - We may have missed the interrupt so always have to
2055 * check if we got a LSC
2056 */
2057 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2058 !(eicr & IXGBE_EICR_LSC))
2059 return;
2060
2061 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2062 u32 autoneg;
2063 bool link_up = false;
2064
2065 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2066
2067 if (link_up)
2068 return;
2069 }
2070
2071 /* Check if this is not due to overtemp */
2072 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2073 return;
2074
2075 break;
2076 default:
2077 if (!(eicr & IXGBE_EICR_GPI_SDP0))
2078 return;
2079 break;
2080 }
2081 e_crit(drv,
2082 "Network adapter has been stopped because it has over heated. "
2083 "Restart the computer. If the problem persists, "
2084 "power off the system and replace the adapter\n");
2085
2086 adapter->interrupt_event = 0;
2087 }
2088
2089 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2090 {
2091 struct ixgbe_hw *hw = &adapter->hw;
2092
2093 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2094 (eicr & IXGBE_EICR_GPI_SDP1)) {
2095 e_crit(probe, "Fan has stopped, replace the adapter\n");
2096 /* write to clear the interrupt */
2097 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2098 }
2099 }
2100
2101 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2102 {
2103 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2104 return;
2105
2106 switch (adapter->hw.mac.type) {
2107 case ixgbe_mac_82599EB:
2108 /*
2109 * Need to check link state so complete overtemp check
2110 * on service task
2111 */
2112 if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2113 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2114 adapter->interrupt_event = eicr;
2115 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2116 ixgbe_service_event_schedule(adapter);
2117 return;
2118 }
2119 return;
2120 case ixgbe_mac_X540:
2121 if (!(eicr & IXGBE_EICR_TS))
2122 return;
2123 break;
2124 default:
2125 return;
2126 }
2127
2128 e_crit(drv,
2129 "Network adapter has been stopped because it has over heated. "
2130 "Restart the computer. If the problem persists, "
2131 "power off the system and replace the adapter\n");
2132 }
2133
2134 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2135 {
2136 struct ixgbe_hw *hw = &adapter->hw;
2137
2138 if (eicr & IXGBE_EICR_GPI_SDP2) {
2139 /* Clear the interrupt */
2140 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2141 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2142 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2143 ixgbe_service_event_schedule(adapter);
2144 }
2145 }
2146
2147 if (eicr & IXGBE_EICR_GPI_SDP1) {
2148 /* Clear the interrupt */
2149 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2150 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2151 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2152 ixgbe_service_event_schedule(adapter);
2153 }
2154 }
2155 }
2156
2157 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2158 {
2159 struct ixgbe_hw *hw = &adapter->hw;
2160
2161 adapter->lsc_int++;
2162 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2163 adapter->link_check_timeout = jiffies;
2164 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2165 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2166 IXGBE_WRITE_FLUSH(hw);
2167 ixgbe_service_event_schedule(adapter);
2168 }
2169 }
2170
2171 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2172 u64 qmask)
2173 {
2174 u32 mask;
2175 struct ixgbe_hw *hw = &adapter->hw;
2176
2177 switch (hw->mac.type) {
2178 case ixgbe_mac_82598EB:
2179 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2180 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2181 break;
2182 case ixgbe_mac_82599EB:
2183 case ixgbe_mac_X540:
2184 mask = (qmask & 0xFFFFFFFF);
2185 if (mask)
2186 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2187 mask = (qmask >> 32);
2188 if (mask)
2189 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2190 break;
2191 default:
2192 break;
2193 }
2194 /* skip the flush */
2195 }
2196
2197 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2198 u64 qmask)
2199 {
2200 u32 mask;
2201 struct ixgbe_hw *hw = &adapter->hw;
2202
2203 switch (hw->mac.type) {
2204 case ixgbe_mac_82598EB:
2205 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2206 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2207 break;
2208 case ixgbe_mac_82599EB:
2209 case ixgbe_mac_X540:
2210 mask = (qmask & 0xFFFFFFFF);
2211 if (mask)
2212 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2213 mask = (qmask >> 32);
2214 if (mask)
2215 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2216 break;
2217 default:
2218 break;
2219 }
2220 /* skip the flush */
2221 }
2222
2223 /**
2224 * ixgbe_irq_enable - Enable default interrupt generation settings
2225 * @adapter: board private structure
2226 **/
2227 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2228 bool flush)
2229 {
2230 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2231
2232 /* don't reenable LSC while waiting for link */
2233 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2234 mask &= ~IXGBE_EIMS_LSC;
2235
2236 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2237 switch (adapter->hw.mac.type) {
2238 case ixgbe_mac_82599EB:
2239 mask |= IXGBE_EIMS_GPI_SDP0;
2240 break;
2241 case ixgbe_mac_X540:
2242 mask |= IXGBE_EIMS_TS;
2243 break;
2244 default:
2245 break;
2246 }
2247 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2248 mask |= IXGBE_EIMS_GPI_SDP1;
2249 switch (adapter->hw.mac.type) {
2250 case ixgbe_mac_82599EB:
2251 mask |= IXGBE_EIMS_GPI_SDP1;
2252 mask |= IXGBE_EIMS_GPI_SDP2;
2253 case ixgbe_mac_X540:
2254 mask |= IXGBE_EIMS_ECC;
2255 mask |= IXGBE_EIMS_MAILBOX;
2256 break;
2257 default:
2258 break;
2259 }
2260 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2261 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2262 mask |= IXGBE_EIMS_FLOW_DIR;
2263
2264 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2265 if (queues)
2266 ixgbe_irq_enable_queues(adapter, ~0);
2267 if (flush)
2268 IXGBE_WRITE_FLUSH(&adapter->hw);
2269 }
2270
2271 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2272 {
2273 struct ixgbe_adapter *adapter = data;
2274 struct ixgbe_hw *hw = &adapter->hw;
2275 u32 eicr;
2276
2277 /*
2278 * Workaround for Silicon errata. Use clear-by-write instead
2279 * of clear-by-read. Reading with EICS will return the
2280 * interrupt causes without clearing, which later be done
2281 * with the write to EICR.
2282 */
2283 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2284 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2285
2286 if (eicr & IXGBE_EICR_LSC)
2287 ixgbe_check_lsc(adapter);
2288
2289 if (eicr & IXGBE_EICR_MAILBOX)
2290 ixgbe_msg_task(adapter);
2291
2292 switch (hw->mac.type) {
2293 case ixgbe_mac_82599EB:
2294 case ixgbe_mac_X540:
2295 if (eicr & IXGBE_EICR_ECC)
2296 e_info(link, "Received unrecoverable ECC Err, please "
2297 "reboot\n");
2298 /* Handle Flow Director Full threshold interrupt */
2299 if (eicr & IXGBE_EICR_FLOW_DIR) {
2300 int reinit_count = 0;
2301 int i;
2302 for (i = 0; i < adapter->num_tx_queues; i++) {
2303 struct ixgbe_ring *ring = adapter->tx_ring[i];
2304 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2305 &ring->state))
2306 reinit_count++;
2307 }
2308 if (reinit_count) {
2309 /* no more flow director interrupts until after init */
2310 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2311 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2312 ixgbe_service_event_schedule(adapter);
2313 }
2314 }
2315 ixgbe_check_sfp_event(adapter, eicr);
2316 ixgbe_check_overtemp_event(adapter, eicr);
2317 break;
2318 default:
2319 break;
2320 }
2321
2322 ixgbe_check_fan_failure(adapter, eicr);
2323 #ifdef CONFIG_IXGBE_PTP
2324 ixgbe_ptp_check_pps_event(adapter, eicr);
2325 #endif
2326
2327 /* re-enable the original interrupt state, no lsc, no queues */
2328 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2329 ixgbe_irq_enable(adapter, false, false);
2330
2331 return IRQ_HANDLED;
2332 }
2333
2334 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2335 {
2336 struct ixgbe_q_vector *q_vector = data;
2337
2338 /* EIAM disabled interrupts (on this vector) for us */
2339
2340 if (q_vector->rx.ring || q_vector->tx.ring)
2341 napi_schedule(&q_vector->napi);
2342
2343 return IRQ_HANDLED;
2344 }
2345
2346 /**
2347 * ixgbe_poll - NAPI Rx polling callback
2348 * @napi: structure for representing this polling device
2349 * @budget: how many packets driver is allowed to clean
2350 *
2351 * This function is used for legacy and MSI, NAPI mode
2352 **/
2353 int ixgbe_poll(struct napi_struct *napi, int budget)
2354 {
2355 struct ixgbe_q_vector *q_vector =
2356 container_of(napi, struct ixgbe_q_vector, napi);
2357 struct ixgbe_adapter *adapter = q_vector->adapter;
2358 struct ixgbe_ring *ring;
2359 int per_ring_budget;
2360 bool clean_complete = true;
2361
2362 #ifdef CONFIG_IXGBE_DCA
2363 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2364 ixgbe_update_dca(q_vector);
2365 #endif
2366
2367 ixgbe_for_each_ring(ring, q_vector->tx)
2368 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2369
2370 /* attempt to distribute budget to each queue fairly, but don't allow
2371 * the budget to go below 1 because we'll exit polling */
2372 if (q_vector->rx.count > 1)
2373 per_ring_budget = max(budget/q_vector->rx.count, 1);
2374 else
2375 per_ring_budget = budget;
2376
2377 ixgbe_for_each_ring(ring, q_vector->rx)
2378 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
2379 per_ring_budget);
2380
2381 /* If all work not completed, return budget and keep polling */
2382 if (!clean_complete)
2383 return budget;
2384
2385 /* all work done, exit the polling mode */
2386 napi_complete(napi);
2387 if (adapter->rx_itr_setting & 1)
2388 ixgbe_set_itr(q_vector);
2389 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2390 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2391
2392 return 0;
2393 }
2394
2395 /**
2396 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2397 * @adapter: board private structure
2398 *
2399 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2400 * interrupts from the kernel.
2401 **/
2402 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2403 {
2404 struct net_device *netdev = adapter->netdev;
2405 int vector, err;
2406 int ri = 0, ti = 0;
2407
2408 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2409 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2410 struct msix_entry *entry = &adapter->msix_entries[vector];
2411
2412 if (q_vector->tx.ring && q_vector->rx.ring) {
2413 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2414 "%s-%s-%d", netdev->name, "TxRx", ri++);
2415 ti++;
2416 } else if (q_vector->rx.ring) {
2417 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2418 "%s-%s-%d", netdev->name, "rx", ri++);
2419 } else if (q_vector->tx.ring) {
2420 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2421 "%s-%s-%d", netdev->name, "tx", ti++);
2422 } else {
2423 /* skip this unused q_vector */
2424 continue;
2425 }
2426 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2427 q_vector->name, q_vector);
2428 if (err) {
2429 e_err(probe, "request_irq failed for MSIX interrupt "
2430 "Error: %d\n", err);
2431 goto free_queue_irqs;
2432 }
2433 /* If Flow Director is enabled, set interrupt affinity */
2434 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2435 /* assign the mask for this irq */
2436 irq_set_affinity_hint(entry->vector,
2437 &q_vector->affinity_mask);
2438 }
2439 }
2440
2441 err = request_irq(adapter->msix_entries[vector].vector,
2442 ixgbe_msix_other, 0, netdev->name, adapter);
2443 if (err) {
2444 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2445 goto free_queue_irqs;
2446 }
2447
2448 return 0;
2449
2450 free_queue_irqs:
2451 while (vector) {
2452 vector--;
2453 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2454 NULL);
2455 free_irq(adapter->msix_entries[vector].vector,
2456 adapter->q_vector[vector]);
2457 }
2458 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2459 pci_disable_msix(adapter->pdev);
2460 kfree(adapter->msix_entries);
2461 adapter->msix_entries = NULL;
2462 return err;
2463 }
2464
2465 /**
2466 * ixgbe_intr - legacy mode Interrupt Handler
2467 * @irq: interrupt number
2468 * @data: pointer to a network interface device structure
2469 **/
2470 static irqreturn_t ixgbe_intr(int irq, void *data)
2471 {
2472 struct ixgbe_adapter *adapter = data;
2473 struct ixgbe_hw *hw = &adapter->hw;
2474 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2475 u32 eicr;
2476
2477 /*
2478 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2479 * before the read of EICR.
2480 */
2481 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2482
2483 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2484 * therefore no explicit interrupt disable is necessary */
2485 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2486 if (!eicr) {
2487 /*
2488 * shared interrupt alert!
2489 * make sure interrupts are enabled because the read will
2490 * have disabled interrupts due to EIAM
2491 * finish the workaround of silicon errata on 82598. Unmask
2492 * the interrupt that we masked before the EICR read.
2493 */
2494 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2495 ixgbe_irq_enable(adapter, true, true);
2496 return IRQ_NONE; /* Not our interrupt */
2497 }
2498
2499 if (eicr & IXGBE_EICR_LSC)
2500 ixgbe_check_lsc(adapter);
2501
2502 switch (hw->mac.type) {
2503 case ixgbe_mac_82599EB:
2504 ixgbe_check_sfp_event(adapter, eicr);
2505 /* Fall through */
2506 case ixgbe_mac_X540:
2507 if (eicr & IXGBE_EICR_ECC)
2508 e_info(link, "Received unrecoverable ECC err, please "
2509 "reboot\n");
2510 ixgbe_check_overtemp_event(adapter, eicr);
2511 break;
2512 default:
2513 break;
2514 }
2515
2516 ixgbe_check_fan_failure(adapter, eicr);
2517 #ifdef CONFIG_IXGBE_PTP
2518 ixgbe_ptp_check_pps_event(adapter, eicr);
2519 #endif
2520
2521 /* would disable interrupts here but EIAM disabled it */
2522 napi_schedule(&q_vector->napi);
2523
2524 /*
2525 * re-enable link(maybe) and non-queue interrupts, no flush.
2526 * ixgbe_poll will re-enable the queue interrupts
2527 */
2528 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2529 ixgbe_irq_enable(adapter, false, false);
2530
2531 return IRQ_HANDLED;
2532 }
2533
2534 /**
2535 * ixgbe_request_irq - initialize interrupts
2536 * @adapter: board private structure
2537 *
2538 * Attempts to configure interrupts using the best available
2539 * capabilities of the hardware and kernel.
2540 **/
2541 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2542 {
2543 struct net_device *netdev = adapter->netdev;
2544 int err;
2545
2546 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2547 err = ixgbe_request_msix_irqs(adapter);
2548 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2549 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2550 netdev->name, adapter);
2551 else
2552 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2553 netdev->name, adapter);
2554
2555 if (err)
2556 e_err(probe, "request_irq failed, Error %d\n", err);
2557
2558 return err;
2559 }
2560
2561 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2562 {
2563 int vector;
2564
2565 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2566 free_irq(adapter->pdev->irq, adapter);
2567 return;
2568 }
2569
2570 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2571 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2572 struct msix_entry *entry = &adapter->msix_entries[vector];
2573
2574 /* free only the irqs that were actually requested */
2575 if (!q_vector->rx.ring && !q_vector->tx.ring)
2576 continue;
2577
2578 /* clear the affinity_mask in the IRQ descriptor */
2579 irq_set_affinity_hint(entry->vector, NULL);
2580
2581 free_irq(entry->vector, q_vector);
2582 }
2583
2584 free_irq(adapter->msix_entries[vector++].vector, adapter);
2585 }
2586
2587 /**
2588 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2589 * @adapter: board private structure
2590 **/
2591 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2592 {
2593 switch (adapter->hw.mac.type) {
2594 case ixgbe_mac_82598EB:
2595 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2596 break;
2597 case ixgbe_mac_82599EB:
2598 case ixgbe_mac_X540:
2599 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2600 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2601 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2602 break;
2603 default:
2604 break;
2605 }
2606 IXGBE_WRITE_FLUSH(&adapter->hw);
2607 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2608 int vector;
2609
2610 for (vector = 0; vector < adapter->num_q_vectors; vector++)
2611 synchronize_irq(adapter->msix_entries[vector].vector);
2612
2613 synchronize_irq(adapter->msix_entries[vector++].vector);
2614 } else {
2615 synchronize_irq(adapter->pdev->irq);
2616 }
2617 }
2618
2619 /**
2620 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2621 *
2622 **/
2623 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2624 {
2625 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2626
2627 /* rx/tx vector */
2628 if (adapter->rx_itr_setting == 1)
2629 q_vector->itr = IXGBE_20K_ITR;
2630 else
2631 q_vector->itr = adapter->rx_itr_setting;
2632
2633 ixgbe_write_eitr(q_vector);
2634
2635 ixgbe_set_ivar(adapter, 0, 0, 0);
2636 ixgbe_set_ivar(adapter, 1, 0, 0);
2637
2638 e_info(hw, "Legacy interrupt IVAR setup done\n");
2639 }
2640
2641 /**
2642 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2643 * @adapter: board private structure
2644 * @ring: structure containing ring specific data
2645 *
2646 * Configure the Tx descriptor ring after a reset.
2647 **/
2648 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2649 struct ixgbe_ring *ring)
2650 {
2651 struct ixgbe_hw *hw = &adapter->hw;
2652 u64 tdba = ring->dma;
2653 int wait_loop = 10;
2654 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2655 u8 reg_idx = ring->reg_idx;
2656
2657 /* disable queue to avoid issues while updating state */
2658 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2659 IXGBE_WRITE_FLUSH(hw);
2660
2661 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2662 (tdba & DMA_BIT_MASK(32)));
2663 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2664 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2665 ring->count * sizeof(union ixgbe_adv_tx_desc));
2666 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2667 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2668 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2669
2670 /*
2671 * set WTHRESH to encourage burst writeback, it should not be set
2672 * higher than 1 when ITR is 0 as it could cause false TX hangs
2673 *
2674 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2675 * to or less than the number of on chip descriptors, which is
2676 * currently 40.
2677 */
2678 if (!ring->q_vector || (ring->q_vector->itr < 8))
2679 txdctl |= (1 << 16); /* WTHRESH = 1 */
2680 else
2681 txdctl |= (8 << 16); /* WTHRESH = 8 */
2682
2683 /*
2684 * Setting PTHRESH to 32 both improves performance
2685 * and avoids a TX hang with DFP enabled
2686 */
2687 txdctl |= (1 << 8) | /* HTHRESH = 1 */
2688 32; /* PTHRESH = 32 */
2689
2690 /* reinitialize flowdirector state */
2691 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2692 adapter->atr_sample_rate) {
2693 ring->atr_sample_rate = adapter->atr_sample_rate;
2694 ring->atr_count = 0;
2695 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2696 } else {
2697 ring->atr_sample_rate = 0;
2698 }
2699
2700 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2701
2702 /* enable queue */
2703 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2704
2705 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2706 if (hw->mac.type == ixgbe_mac_82598EB &&
2707 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2708 return;
2709
2710 /* poll to verify queue is enabled */
2711 do {
2712 usleep_range(1000, 2000);
2713 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2714 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2715 if (!wait_loop)
2716 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2717 }
2718
2719 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2720 {
2721 struct ixgbe_hw *hw = &adapter->hw;
2722 u32 rttdcs;
2723 u32 reg;
2724 u8 tcs = netdev_get_num_tc(adapter->netdev);
2725
2726 if (hw->mac.type == ixgbe_mac_82598EB)
2727 return;
2728
2729 /* disable the arbiter while setting MTQC */
2730 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2731 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2732 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2733
2734 /* set transmit pool layout */
2735 switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2736 case (IXGBE_FLAG_SRIOV_ENABLED):
2737 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2738 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2739 break;
2740 default:
2741 if (!tcs)
2742 reg = IXGBE_MTQC_64Q_1PB;
2743 else if (tcs <= 4)
2744 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2745 else
2746 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2747
2748 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2749
2750 /* Enable Security TX Buffer IFG for multiple pb */
2751 if (tcs) {
2752 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2753 reg |= IXGBE_SECTX_DCB;
2754 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2755 }
2756 break;
2757 }
2758
2759 /* re-enable the arbiter */
2760 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2761 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2762 }
2763
2764 /**
2765 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2766 * @adapter: board private structure
2767 *
2768 * Configure the Tx unit of the MAC after a reset.
2769 **/
2770 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2771 {
2772 struct ixgbe_hw *hw = &adapter->hw;
2773 u32 dmatxctl;
2774 u32 i;
2775
2776 ixgbe_setup_mtqc(adapter);
2777
2778 if (hw->mac.type != ixgbe_mac_82598EB) {
2779 /* DMATXCTL.EN must be before Tx queues are enabled */
2780 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2781 dmatxctl |= IXGBE_DMATXCTL_TE;
2782 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2783 }
2784
2785 /* Setup the HW Tx Head and Tail descriptor pointers */
2786 for (i = 0; i < adapter->num_tx_queues; i++)
2787 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2788 }
2789
2790 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
2791 struct ixgbe_ring *ring)
2792 {
2793 struct ixgbe_hw *hw = &adapter->hw;
2794 u8 reg_idx = ring->reg_idx;
2795 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2796
2797 srrctl |= IXGBE_SRRCTL_DROP_EN;
2798
2799 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2800 }
2801
2802 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
2803 struct ixgbe_ring *ring)
2804 {
2805 struct ixgbe_hw *hw = &adapter->hw;
2806 u8 reg_idx = ring->reg_idx;
2807 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2808
2809 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
2810
2811 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2812 }
2813
2814 #ifdef CONFIG_IXGBE_DCB
2815 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2816 #else
2817 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2818 #endif
2819 {
2820 int i;
2821 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
2822
2823 if (adapter->ixgbe_ieee_pfc)
2824 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
2825
2826 /*
2827 * We should set the drop enable bit if:
2828 * SR-IOV is enabled
2829 * or
2830 * Number of Rx queues > 1 and flow control is disabled
2831 *
2832 * This allows us to avoid head of line blocking for security
2833 * and performance reasons.
2834 */
2835 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
2836 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
2837 for (i = 0; i < adapter->num_rx_queues; i++)
2838 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
2839 } else {
2840 for (i = 0; i < adapter->num_rx_queues; i++)
2841 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
2842 }
2843 }
2844
2845 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2846
2847 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2848 struct ixgbe_ring *rx_ring)
2849 {
2850 struct ixgbe_hw *hw = &adapter->hw;
2851 u32 srrctl;
2852 u8 reg_idx = rx_ring->reg_idx;
2853
2854 if (hw->mac.type == ixgbe_mac_82598EB) {
2855 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
2856
2857 /*
2858 * if VMDq is not active we must program one srrctl register
2859 * per RSS queue since we have enabled RDRXCTL.MVMEN
2860 */
2861 reg_idx &= mask;
2862 }
2863
2864 /* configure header buffer length, needed for RSC */
2865 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
2866
2867 /* configure the packet buffer length */
2868 #if PAGE_SIZE > IXGBE_MAX_RXBUFFER
2869 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2870 #else
2871 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2872 #endif
2873
2874 /* configure descriptor type */
2875 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2876
2877 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2878 }
2879
2880 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2881 {
2882 struct ixgbe_hw *hw = &adapter->hw;
2883 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2884 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2885 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2886 u32 mrqc = 0, reta = 0;
2887 u32 rxcsum;
2888 int i, j;
2889 u8 tcs = netdev_get_num_tc(adapter->netdev);
2890 int maxq = adapter->ring_feature[RING_F_RSS].indices;
2891
2892 if (tcs)
2893 maxq = min(maxq, adapter->num_tx_queues / tcs);
2894
2895 /* Fill out hash function seeds */
2896 for (i = 0; i < 10; i++)
2897 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2898
2899 /* Fill out redirection table */
2900 for (i = 0, j = 0; i < 128; i++, j++) {
2901 if (j == maxq)
2902 j = 0;
2903 /* reta = 4-byte sliding window of
2904 * 0x00..(indices-1)(indices-1)00..etc. */
2905 reta = (reta << 8) | (j * 0x11);
2906 if ((i & 3) == 3)
2907 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2908 }
2909
2910 /* Disable indicating checksum in descriptor, enables RSS hash */
2911 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2912 rxcsum |= IXGBE_RXCSUM_PCSD;
2913 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2914
2915 if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2916 (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2917 mrqc = IXGBE_MRQC_RSSEN;
2918 } else {
2919 int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2920 | IXGBE_FLAG_SRIOV_ENABLED);
2921
2922 switch (mask) {
2923 case (IXGBE_FLAG_RSS_ENABLED):
2924 if (!tcs)
2925 mrqc = IXGBE_MRQC_RSSEN;
2926 else if (tcs <= 4)
2927 mrqc = IXGBE_MRQC_RTRSS4TCEN;
2928 else
2929 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2930 break;
2931 case (IXGBE_FLAG_SRIOV_ENABLED):
2932 mrqc = IXGBE_MRQC_VMDQEN;
2933 break;
2934 default:
2935 break;
2936 }
2937 }
2938
2939 /* Perform hash on these packet types */
2940 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2941 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2942 | IXGBE_MRQC_RSS_FIELD_IPV6
2943 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2944
2945 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
2946 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
2947 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
2948 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2949
2950 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2951 }
2952
2953 /**
2954 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2955 * @adapter: address of board private structure
2956 * @index: index of ring to set
2957 **/
2958 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2959 struct ixgbe_ring *ring)
2960 {
2961 struct ixgbe_hw *hw = &adapter->hw;
2962 u32 rscctrl;
2963 u8 reg_idx = ring->reg_idx;
2964
2965 if (!ring_is_rsc_enabled(ring))
2966 return;
2967
2968 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2969 rscctrl |= IXGBE_RSCCTL_RSCEN;
2970 /*
2971 * we must limit the number of descriptors so that the
2972 * total size of max desc * buf_len is not greater
2973 * than 65536
2974 */
2975 #if (PAGE_SIZE <= 8192)
2976 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2977 #elif (PAGE_SIZE <= 16384)
2978 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2979 #else
2980 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2981 #endif
2982 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2983 }
2984
2985 #define IXGBE_MAX_RX_DESC_POLL 10
2986 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2987 struct ixgbe_ring *ring)
2988 {
2989 struct ixgbe_hw *hw = &adapter->hw;
2990 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2991 u32 rxdctl;
2992 u8 reg_idx = ring->reg_idx;
2993
2994 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2995 if (hw->mac.type == ixgbe_mac_82598EB &&
2996 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2997 return;
2998
2999 do {
3000 usleep_range(1000, 2000);
3001 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3002 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3003
3004 if (!wait_loop) {
3005 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3006 "the polling period\n", reg_idx);
3007 }
3008 }
3009
3010 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3011 struct ixgbe_ring *ring)
3012 {
3013 struct ixgbe_hw *hw = &adapter->hw;
3014 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3015 u32 rxdctl;
3016 u8 reg_idx = ring->reg_idx;
3017
3018 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3019 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3020
3021 /* write value back with RXDCTL.ENABLE bit cleared */
3022 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3023
3024 if (hw->mac.type == ixgbe_mac_82598EB &&
3025 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3026 return;
3027
3028 /* the hardware may take up to 100us to really disable the rx queue */
3029 do {
3030 udelay(10);
3031 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3032 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3033
3034 if (!wait_loop) {
3035 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3036 "the polling period\n", reg_idx);
3037 }
3038 }
3039
3040 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3041 struct ixgbe_ring *ring)
3042 {
3043 struct ixgbe_hw *hw = &adapter->hw;
3044 u64 rdba = ring->dma;
3045 u32 rxdctl;
3046 u8 reg_idx = ring->reg_idx;
3047
3048 /* disable queue to avoid issues while updating state */
3049 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3050 ixgbe_disable_rx_queue(adapter, ring);
3051
3052 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3053 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3054 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3055 ring->count * sizeof(union ixgbe_adv_rx_desc));
3056 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3057 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3058 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3059
3060 ixgbe_configure_srrctl(adapter, ring);
3061 ixgbe_configure_rscctl(adapter, ring);
3062
3063 /* If operating in IOV mode set RLPML for X540 */
3064 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3065 hw->mac.type == ixgbe_mac_X540) {
3066 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3067 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3068 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3069 }
3070
3071 if (hw->mac.type == ixgbe_mac_82598EB) {
3072 /*
3073 * enable cache line friendly hardware writes:
3074 * PTHRESH=32 descriptors (half the internal cache),
3075 * this also removes ugly rx_no_buffer_count increment
3076 * HTHRESH=4 descriptors (to minimize latency on fetch)
3077 * WTHRESH=8 burst writeback up to two cache lines
3078 */
3079 rxdctl &= ~0x3FFFFF;
3080 rxdctl |= 0x080420;
3081 }
3082
3083 /* enable receive descriptor ring */
3084 rxdctl |= IXGBE_RXDCTL_ENABLE;
3085 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3086
3087 ixgbe_rx_desc_queue_enable(adapter, ring);
3088 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3089 }
3090
3091 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3092 {
3093 struct ixgbe_hw *hw = &adapter->hw;
3094 int p;
3095
3096 /* PSRTYPE must be initialized in non 82598 adapters */
3097 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3098 IXGBE_PSRTYPE_UDPHDR |
3099 IXGBE_PSRTYPE_IPV4HDR |
3100 IXGBE_PSRTYPE_L2HDR |
3101 IXGBE_PSRTYPE_IPV6HDR;
3102
3103 if (hw->mac.type == ixgbe_mac_82598EB)
3104 return;
3105
3106 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3107 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3108
3109 for (p = 0; p < adapter->num_rx_pools; p++)
3110 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3111 psrtype);
3112 }
3113
3114 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3115 {
3116 struct ixgbe_hw *hw = &adapter->hw;
3117 u32 gcr_ext;
3118 u32 vt_reg_bits;
3119 u32 reg_offset, vf_shift;
3120 u32 vmdctl;
3121 int i;
3122
3123 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3124 return;
3125
3126 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3127 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3128 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3129 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3130
3131 vf_shift = adapter->num_vfs % 32;
3132 reg_offset = (adapter->num_vfs >= 32) ? 1 : 0;
3133
3134 /* Enable only the PF's pool for Tx/Rx */
3135 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3136 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3137 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3138 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3139 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3140
3141 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3142 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3143
3144 /*
3145 * Set up VF register offsets for selected VT Mode,
3146 * i.e. 32 or 64 VFs for SR-IOV
3147 */
3148 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3149 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3150 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3151 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3152
3153 /* enable Tx loopback for VF/PF communication */
3154 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3155 /* Enable MAC Anti-Spoofing */
3156 hw->mac.ops.set_mac_anti_spoofing(hw,
3157 (adapter->num_vfs != 0),
3158 adapter->num_vfs);
3159 /* For VFs that have spoof checking turned off */
3160 for (i = 0; i < adapter->num_vfs; i++) {
3161 if (!adapter->vfinfo[i].spoofchk_enabled)
3162 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3163 }
3164 }
3165
3166 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3167 {
3168 struct ixgbe_hw *hw = &adapter->hw;
3169 struct net_device *netdev = adapter->netdev;
3170 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3171 struct ixgbe_ring *rx_ring;
3172 int i;
3173 u32 mhadd, hlreg0;
3174
3175 #ifdef IXGBE_FCOE
3176 /* adjust max frame to be able to do baby jumbo for FCoE */
3177 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3178 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3179 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3180
3181 #endif /* IXGBE_FCOE */
3182 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3183 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3184 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3185 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3186
3187 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3188 }
3189
3190 /* MHADD will allow an extra 4 bytes past for vlan tagged frames */
3191 max_frame += VLAN_HLEN;
3192
3193 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3194 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3195 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3196 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3197
3198 /*
3199 * Setup the HW Rx Head and Tail Descriptor Pointers and
3200 * the Base and Length of the Rx Descriptor Ring
3201 */
3202 for (i = 0; i < adapter->num_rx_queues; i++) {
3203 rx_ring = adapter->rx_ring[i];
3204 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3205 set_ring_rsc_enabled(rx_ring);
3206 else
3207 clear_ring_rsc_enabled(rx_ring);
3208 }
3209 }
3210
3211 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3212 {
3213 struct ixgbe_hw *hw = &adapter->hw;
3214 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3215
3216 switch (hw->mac.type) {
3217 case ixgbe_mac_82598EB:
3218 /*
3219 * For VMDq support of different descriptor types or
3220 * buffer sizes through the use of multiple SRRCTL
3221 * registers, RDRXCTL.MVMEN must be set to 1
3222 *
3223 * also, the manual doesn't mention it clearly but DCA hints
3224 * will only use queue 0's tags unless this bit is set. Side
3225 * effects of setting this bit are only that SRRCTL must be
3226 * fully programmed [0..15]
3227 */
3228 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3229 break;
3230 case ixgbe_mac_82599EB:
3231 case ixgbe_mac_X540:
3232 /* Disable RSC for ACK packets */
3233 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3234 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3235 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3236 /* hardware requires some bits to be set by default */
3237 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3238 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3239 break;
3240 default:
3241 /* We should do nothing since we don't know this hardware */
3242 return;
3243 }
3244
3245 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3246 }
3247
3248 /**
3249 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3250 * @adapter: board private structure
3251 *
3252 * Configure the Rx unit of the MAC after a reset.
3253 **/
3254 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3255 {
3256 struct ixgbe_hw *hw = &adapter->hw;
3257 int i;
3258 u32 rxctrl;
3259
3260 /* disable receives while setting up the descriptors */
3261 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3262 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3263
3264 ixgbe_setup_psrtype(adapter);
3265 ixgbe_setup_rdrxctl(adapter);
3266
3267 /* Program registers for the distribution of queues */
3268 ixgbe_setup_mrqc(adapter);
3269
3270 /* set_rx_buffer_len must be called before ring initialization */
3271 ixgbe_set_rx_buffer_len(adapter);
3272
3273 /*
3274 * Setup the HW Rx Head and Tail Descriptor Pointers and
3275 * the Base and Length of the Rx Descriptor Ring
3276 */
3277 for (i = 0; i < adapter->num_rx_queues; i++)
3278 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3279
3280 /* disable drop enable for 82598 parts */
3281 if (hw->mac.type == ixgbe_mac_82598EB)
3282 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3283
3284 /* enable all receives */
3285 rxctrl |= IXGBE_RXCTRL_RXEN;
3286 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3287 }
3288
3289 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3290 {
3291 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3292 struct ixgbe_hw *hw = &adapter->hw;
3293 int pool_ndx = adapter->num_vfs;
3294
3295 /* add VID to filter table */
3296 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3297 set_bit(vid, adapter->active_vlans);
3298
3299 return 0;
3300 }
3301
3302 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3303 {
3304 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3305 struct ixgbe_hw *hw = &adapter->hw;
3306 int pool_ndx = adapter->num_vfs;
3307
3308 /* remove VID from filter table */
3309 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3310 clear_bit(vid, adapter->active_vlans);
3311
3312 return 0;
3313 }
3314
3315 /**
3316 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3317 * @adapter: driver data
3318 */
3319 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3320 {
3321 struct ixgbe_hw *hw = &adapter->hw;
3322 u32 vlnctrl;
3323
3324 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3325 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3326 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3327 }
3328
3329 /**
3330 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3331 * @adapter: driver data
3332 */
3333 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3334 {
3335 struct ixgbe_hw *hw = &adapter->hw;
3336 u32 vlnctrl;
3337
3338 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3339 vlnctrl |= IXGBE_VLNCTRL_VFE;
3340 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3341 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3342 }
3343
3344 /**
3345 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3346 * @adapter: driver data
3347 */
3348 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3349 {
3350 struct ixgbe_hw *hw = &adapter->hw;
3351 u32 vlnctrl;
3352 int i, j;
3353
3354 switch (hw->mac.type) {
3355 case ixgbe_mac_82598EB:
3356 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3357 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3358 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3359 break;
3360 case ixgbe_mac_82599EB:
3361 case ixgbe_mac_X540:
3362 for (i = 0; i < adapter->num_rx_queues; i++) {
3363 j = adapter->rx_ring[i]->reg_idx;
3364 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3365 vlnctrl &= ~IXGBE_RXDCTL_VME;
3366 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3367 }
3368 break;
3369 default:
3370 break;
3371 }
3372 }
3373
3374 /**
3375 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3376 * @adapter: driver data
3377 */
3378 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3379 {
3380 struct ixgbe_hw *hw = &adapter->hw;
3381 u32 vlnctrl;
3382 int i, j;
3383
3384 switch (hw->mac.type) {
3385 case ixgbe_mac_82598EB:
3386 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3387 vlnctrl |= IXGBE_VLNCTRL_VME;
3388 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3389 break;
3390 case ixgbe_mac_82599EB:
3391 case ixgbe_mac_X540:
3392 for (i = 0; i < adapter->num_rx_queues; i++) {
3393 j = adapter->rx_ring[i]->reg_idx;
3394 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3395 vlnctrl |= IXGBE_RXDCTL_VME;
3396 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3397 }
3398 break;
3399 default:
3400 break;
3401 }
3402 }
3403
3404 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3405 {
3406 u16 vid;
3407
3408 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3409
3410 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3411 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3412 }
3413
3414 /**
3415 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3416 * @netdev: network interface device structure
3417 *
3418 * Writes unicast address list to the RAR table.
3419 * Returns: -ENOMEM on failure/insufficient address space
3420 * 0 on no addresses written
3421 * X on writing X addresses to the RAR table
3422 **/
3423 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3424 {
3425 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3426 struct ixgbe_hw *hw = &adapter->hw;
3427 unsigned int vfn = adapter->num_vfs;
3428 unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3429 int count = 0;
3430
3431 /* return ENOMEM indicating insufficient memory for addresses */
3432 if (netdev_uc_count(netdev) > rar_entries)
3433 return -ENOMEM;
3434
3435 if (!netdev_uc_empty(netdev) && rar_entries) {
3436 struct netdev_hw_addr *ha;
3437 /* return error if we do not support writing to RAR table */
3438 if (!hw->mac.ops.set_rar)
3439 return -ENOMEM;
3440
3441 netdev_for_each_uc_addr(ha, netdev) {
3442 if (!rar_entries)
3443 break;
3444 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3445 vfn, IXGBE_RAH_AV);
3446 count++;
3447 }
3448 }
3449 /* write the addresses in reverse order to avoid write combining */
3450 for (; rar_entries > 0 ; rar_entries--)
3451 hw->mac.ops.clear_rar(hw, rar_entries);
3452
3453 return count;
3454 }
3455
3456 /**
3457 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3458 * @netdev: network interface device structure
3459 *
3460 * The set_rx_method entry point is called whenever the unicast/multicast
3461 * address list or the network interface flags are updated. This routine is
3462 * responsible for configuring the hardware for proper unicast, multicast and
3463 * promiscuous mode.
3464 **/
3465 void ixgbe_set_rx_mode(struct net_device *netdev)
3466 {
3467 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3468 struct ixgbe_hw *hw = &adapter->hw;
3469 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3470 int count;
3471
3472 /* Check for Promiscuous and All Multicast modes */
3473
3474 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3475
3476 /* set all bits that we expect to always be set */
3477 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3478 fctrl |= IXGBE_FCTRL_BAM;
3479 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3480 fctrl |= IXGBE_FCTRL_PMCF;
3481
3482 /* clear the bits we are changing the status of */
3483 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3484
3485 if (netdev->flags & IFF_PROMISC) {
3486 hw->addr_ctrl.user_set_promisc = true;
3487 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3488 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3489 /* don't hardware filter vlans in promisc mode */
3490 ixgbe_vlan_filter_disable(adapter);
3491 } else {
3492 if (netdev->flags & IFF_ALLMULTI) {
3493 fctrl |= IXGBE_FCTRL_MPE;
3494 vmolr |= IXGBE_VMOLR_MPE;
3495 } else {
3496 /*
3497 * Write addresses to the MTA, if the attempt fails
3498 * then we should just turn on promiscuous mode so
3499 * that we can at least receive multicast traffic
3500 */
3501 hw->mac.ops.update_mc_addr_list(hw, netdev);
3502 vmolr |= IXGBE_VMOLR_ROMPE;
3503 }
3504 ixgbe_vlan_filter_enable(adapter);
3505 hw->addr_ctrl.user_set_promisc = false;
3506 }
3507
3508 /*
3509 * Write addresses to available RAR registers, if there is not
3510 * sufficient space to store all the addresses then enable
3511 * unicast promiscuous mode
3512 */
3513 count = ixgbe_write_uc_addr_list(netdev);
3514 if (count < 0) {
3515 fctrl |= IXGBE_FCTRL_UPE;
3516 vmolr |= IXGBE_VMOLR_ROPE;
3517 }
3518
3519 if (adapter->num_vfs) {
3520 ixgbe_restore_vf_multicasts(adapter);
3521 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3522 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3523 IXGBE_VMOLR_ROPE);
3524 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3525 }
3526
3527 /* This is useful for sniffing bad packets. */
3528 if (adapter->netdev->features & NETIF_F_RXALL) {
3529 /* UPE and MPE will be handled by normal PROMISC logic
3530 * in e1000e_set_rx_mode */
3531 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3532 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3533 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3534
3535 fctrl &= ~(IXGBE_FCTRL_DPF);
3536 /* NOTE: VLAN filtering is disabled by setting PROMISC */
3537 }
3538
3539 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3540
3541 if (netdev->features & NETIF_F_HW_VLAN_RX)
3542 ixgbe_vlan_strip_enable(adapter);
3543 else
3544 ixgbe_vlan_strip_disable(adapter);
3545 }
3546
3547 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3548 {
3549 int q_idx;
3550
3551 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3552 napi_enable(&adapter->q_vector[q_idx]->napi);
3553 }
3554
3555 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3556 {
3557 int q_idx;
3558
3559 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3560 napi_disable(&adapter->q_vector[q_idx]->napi);
3561 }
3562
3563 #ifdef CONFIG_IXGBE_DCB
3564 /**
3565 * ixgbe_configure_dcb - Configure DCB hardware
3566 * @adapter: ixgbe adapter struct
3567 *
3568 * This is called by the driver on open to configure the DCB hardware.
3569 * This is also called by the gennetlink interface when reconfiguring
3570 * the DCB state.
3571 */
3572 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3573 {
3574 struct ixgbe_hw *hw = &adapter->hw;
3575 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3576
3577 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3578 if (hw->mac.type == ixgbe_mac_82598EB)
3579 netif_set_gso_max_size(adapter->netdev, 65536);
3580 return;
3581 }
3582
3583 if (hw->mac.type == ixgbe_mac_82598EB)
3584 netif_set_gso_max_size(adapter->netdev, 32768);
3585
3586 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3587
3588 #ifdef IXGBE_FCOE
3589 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3590 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3591 #endif
3592
3593 /* reconfigure the hardware */
3594 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3595 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3596 DCB_TX_CONFIG);
3597 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3598 DCB_RX_CONFIG);
3599 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3600 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3601 ixgbe_dcb_hw_ets(&adapter->hw,
3602 adapter->ixgbe_ieee_ets,
3603 max_frame);
3604 ixgbe_dcb_hw_pfc_config(&adapter->hw,
3605 adapter->ixgbe_ieee_pfc->pfc_en,
3606 adapter->ixgbe_ieee_ets->prio_tc);
3607 }
3608
3609 /* Enable RSS Hash per TC */
3610 if (hw->mac.type != ixgbe_mac_82598EB) {
3611 int i;
3612 u32 reg = 0;
3613
3614 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3615 u8 msb = 0;
3616 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3617
3618 while (cnt >>= 1)
3619 msb++;
3620
3621 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3622 }
3623 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3624 }
3625 }
3626 #endif
3627
3628 /* Additional bittime to account for IXGBE framing */
3629 #define IXGBE_ETH_FRAMING 20
3630
3631 /**
3632 * ixgbe_hpbthresh - calculate high water mark for flow control
3633 *
3634 * @adapter: board private structure to calculate for
3635 * @pb: packet buffer to calculate
3636 */
3637 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3638 {
3639 struct ixgbe_hw *hw = &adapter->hw;
3640 struct net_device *dev = adapter->netdev;
3641 int link, tc, kb, marker;
3642 u32 dv_id, rx_pba;
3643
3644 /* Calculate max LAN frame size */
3645 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3646
3647 #ifdef IXGBE_FCOE
3648 /* FCoE traffic class uses FCOE jumbo frames */
3649 if ((dev->features & NETIF_F_FCOE_MTU) &&
3650 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
3651 (pb == ixgbe_fcoe_get_tc(adapter)))
3652 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3653
3654 #endif
3655 /* Calculate delay value for device */
3656 switch (hw->mac.type) {
3657 case ixgbe_mac_X540:
3658 dv_id = IXGBE_DV_X540(link, tc);
3659 break;
3660 default:
3661 dv_id = IXGBE_DV(link, tc);
3662 break;
3663 }
3664
3665 /* Loopback switch introduces additional latency */
3666 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3667 dv_id += IXGBE_B2BT(tc);
3668
3669 /* Delay value is calculated in bit times convert to KB */
3670 kb = IXGBE_BT2KB(dv_id);
3671 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3672
3673 marker = rx_pba - kb;
3674
3675 /* It is possible that the packet buffer is not large enough
3676 * to provide required headroom. In this case throw an error
3677 * to user and a do the best we can.
3678 */
3679 if (marker < 0) {
3680 e_warn(drv, "Packet Buffer(%i) can not provide enough"
3681 "headroom to support flow control."
3682 "Decrease MTU or number of traffic classes\n", pb);
3683 marker = tc + 1;
3684 }
3685
3686 return marker;
3687 }
3688
3689 /**
3690 * ixgbe_lpbthresh - calculate low water mark for for flow control
3691 *
3692 * @adapter: board private structure to calculate for
3693 * @pb: packet buffer to calculate
3694 */
3695 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3696 {
3697 struct ixgbe_hw *hw = &adapter->hw;
3698 struct net_device *dev = adapter->netdev;
3699 int tc;
3700 u32 dv_id;
3701
3702 /* Calculate max LAN frame size */
3703 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3704
3705 /* Calculate delay value for device */
3706 switch (hw->mac.type) {
3707 case ixgbe_mac_X540:
3708 dv_id = IXGBE_LOW_DV_X540(tc);
3709 break;
3710 default:
3711 dv_id = IXGBE_LOW_DV(tc);
3712 break;
3713 }
3714
3715 /* Delay value is calculated in bit times convert to KB */
3716 return IXGBE_BT2KB(dv_id);
3717 }
3718
3719 /*
3720 * ixgbe_pbthresh_setup - calculate and setup high low water marks
3721 */
3722 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3723 {
3724 struct ixgbe_hw *hw = &adapter->hw;
3725 int num_tc = netdev_get_num_tc(adapter->netdev);
3726 int i;
3727
3728 if (!num_tc)
3729 num_tc = 1;
3730
3731 hw->fc.low_water = ixgbe_lpbthresh(adapter);
3732
3733 for (i = 0; i < num_tc; i++) {
3734 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3735
3736 /* Low water marks must not be larger than high water marks */
3737 if (hw->fc.low_water > hw->fc.high_water[i])
3738 hw->fc.low_water = 0;
3739 }
3740 }
3741
3742 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3743 {
3744 struct ixgbe_hw *hw = &adapter->hw;
3745 int hdrm;
3746 u8 tc = netdev_get_num_tc(adapter->netdev);
3747
3748 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3749 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3750 hdrm = 32 << adapter->fdir_pballoc;
3751 else
3752 hdrm = 0;
3753
3754 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3755 ixgbe_pbthresh_setup(adapter);
3756 }
3757
3758 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3759 {
3760 struct ixgbe_hw *hw = &adapter->hw;
3761 struct hlist_node *node, *node2;
3762 struct ixgbe_fdir_filter *filter;
3763
3764 spin_lock(&adapter->fdir_perfect_lock);
3765
3766 if (!hlist_empty(&adapter->fdir_filter_list))
3767 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3768
3769 hlist_for_each_entry_safe(filter, node, node2,
3770 &adapter->fdir_filter_list, fdir_node) {
3771 ixgbe_fdir_write_perfect_filter_82599(hw,
3772 &filter->filter,
3773 filter->sw_idx,
3774 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3775 IXGBE_FDIR_DROP_QUEUE :
3776 adapter->rx_ring[filter->action]->reg_idx);
3777 }
3778
3779 spin_unlock(&adapter->fdir_perfect_lock);
3780 }
3781
3782 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3783 {
3784 struct ixgbe_hw *hw = &adapter->hw;
3785
3786 ixgbe_configure_pb(adapter);
3787 #ifdef CONFIG_IXGBE_DCB
3788 ixgbe_configure_dcb(adapter);
3789 #endif
3790
3791 ixgbe_set_rx_mode(adapter->netdev);
3792 ixgbe_restore_vlan(adapter);
3793
3794 #ifdef IXGBE_FCOE
3795 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3796 ixgbe_configure_fcoe(adapter);
3797
3798 #endif /* IXGBE_FCOE */
3799
3800 switch (hw->mac.type) {
3801 case ixgbe_mac_82599EB:
3802 case ixgbe_mac_X540:
3803 hw->mac.ops.disable_rx_buff(hw);
3804 break;
3805 default:
3806 break;
3807 }
3808
3809 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3810 ixgbe_init_fdir_signature_82599(&adapter->hw,
3811 adapter->fdir_pballoc);
3812 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3813 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3814 adapter->fdir_pballoc);
3815 ixgbe_fdir_filter_restore(adapter);
3816 }
3817
3818 switch (hw->mac.type) {
3819 case ixgbe_mac_82599EB:
3820 case ixgbe_mac_X540:
3821 hw->mac.ops.enable_rx_buff(hw);
3822 break;
3823 default:
3824 break;
3825 }
3826
3827 ixgbe_configure_virtualization(adapter);
3828
3829 ixgbe_configure_tx(adapter);
3830 ixgbe_configure_rx(adapter);
3831 }
3832
3833 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3834 {
3835 switch (hw->phy.type) {
3836 case ixgbe_phy_sfp_avago:
3837 case ixgbe_phy_sfp_ftl:
3838 case ixgbe_phy_sfp_intel:
3839 case ixgbe_phy_sfp_unknown:
3840 case ixgbe_phy_sfp_passive_tyco:
3841 case ixgbe_phy_sfp_passive_unknown:
3842 case ixgbe_phy_sfp_active_unknown:
3843 case ixgbe_phy_sfp_ftl_active:
3844 return true;
3845 case ixgbe_phy_nl:
3846 if (hw->mac.type == ixgbe_mac_82598EB)
3847 return true;
3848 default:
3849 return false;
3850 }
3851 }
3852
3853 /**
3854 * ixgbe_sfp_link_config - set up SFP+ link
3855 * @adapter: pointer to private adapter struct
3856 **/
3857 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3858 {
3859 /*
3860 * We are assuming the worst case scenario here, and that
3861 * is that an SFP was inserted/removed after the reset
3862 * but before SFP detection was enabled. As such the best
3863 * solution is to just start searching as soon as we start
3864 */
3865 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3866 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3867
3868 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3869 }
3870
3871 /**
3872 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3873 * @hw: pointer to private hardware struct
3874 *
3875 * Returns 0 on success, negative on failure
3876 **/
3877 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3878 {
3879 u32 autoneg;
3880 bool negotiation, link_up = false;
3881 u32 ret = IXGBE_ERR_LINK_SETUP;
3882
3883 if (hw->mac.ops.check_link)
3884 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3885
3886 if (ret)
3887 goto link_cfg_out;
3888
3889 autoneg = hw->phy.autoneg_advertised;
3890 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3891 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3892 &negotiation);
3893 if (ret)
3894 goto link_cfg_out;
3895
3896 if (hw->mac.ops.setup_link)
3897 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3898 link_cfg_out:
3899 return ret;
3900 }
3901
3902 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3903 {
3904 struct ixgbe_hw *hw = &adapter->hw;
3905 u32 gpie = 0;
3906
3907 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3908 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3909 IXGBE_GPIE_OCD;
3910 gpie |= IXGBE_GPIE_EIAME;
3911 /*
3912 * use EIAM to auto-mask when MSI-X interrupt is asserted
3913 * this saves a register write for every interrupt
3914 */
3915 switch (hw->mac.type) {
3916 case ixgbe_mac_82598EB:
3917 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3918 break;
3919 case ixgbe_mac_82599EB:
3920 case ixgbe_mac_X540:
3921 default:
3922 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3923 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3924 break;
3925 }
3926 } else {
3927 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3928 * specifically only auto mask tx and rx interrupts */
3929 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3930 }
3931
3932 /* XXX: to interrupt immediately for EICS writes, enable this */
3933 /* gpie |= IXGBE_GPIE_EIMEN; */
3934
3935 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3936 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3937 gpie |= IXGBE_GPIE_VTMODE_64;
3938 }
3939
3940 /* Enable Thermal over heat sensor interrupt */
3941 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3942 switch (adapter->hw.mac.type) {
3943 case ixgbe_mac_82599EB:
3944 gpie |= IXGBE_SDP0_GPIEN;
3945 break;
3946 case ixgbe_mac_X540:
3947 gpie |= IXGBE_EIMS_TS;
3948 break;
3949 default:
3950 break;
3951 }
3952 }
3953
3954 /* Enable fan failure interrupt */
3955 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3956 gpie |= IXGBE_SDP1_GPIEN;
3957
3958 if (hw->mac.type == ixgbe_mac_82599EB) {
3959 gpie |= IXGBE_SDP1_GPIEN;
3960 gpie |= IXGBE_SDP2_GPIEN;
3961 }
3962
3963 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3964 }
3965
3966 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
3967 {
3968 struct ixgbe_hw *hw = &adapter->hw;
3969 int err;
3970 u32 ctrl_ext;
3971
3972 ixgbe_get_hw_control(adapter);
3973 ixgbe_setup_gpie(adapter);
3974
3975 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3976 ixgbe_configure_msix(adapter);
3977 else
3978 ixgbe_configure_msi_and_legacy(adapter);
3979
3980 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3981 if (hw->mac.ops.enable_tx_laser &&
3982 ((hw->phy.multispeed_fiber) ||
3983 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3984 (hw->mac.type == ixgbe_mac_82599EB))))
3985 hw->mac.ops.enable_tx_laser(hw);
3986
3987 clear_bit(__IXGBE_DOWN, &adapter->state);
3988 ixgbe_napi_enable_all(adapter);
3989
3990 if (ixgbe_is_sfp(hw)) {
3991 ixgbe_sfp_link_config(adapter);
3992 } else {
3993 err = ixgbe_non_sfp_link_config(hw);
3994 if (err)
3995 e_err(probe, "link_config FAILED %d\n", err);
3996 }
3997
3998 /* clear any pending interrupts, may auto mask */
3999 IXGBE_READ_REG(hw, IXGBE_EICR);
4000 ixgbe_irq_enable(adapter, true, true);
4001
4002 /*
4003 * If this adapter has a fan, check to see if we had a failure
4004 * before we enabled the interrupt.
4005 */
4006 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4007 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4008 if (esdp & IXGBE_ESDP_SDP1)
4009 e_crit(drv, "Fan has stopped, replace the adapter\n");
4010 }
4011
4012 /* enable transmits */
4013 netif_tx_start_all_queues(adapter->netdev);
4014
4015 /* bring the link up in the watchdog, this could race with our first
4016 * link up interrupt but shouldn't be a problem */
4017 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4018 adapter->link_check_timeout = jiffies;
4019 mod_timer(&adapter->service_timer, jiffies);
4020
4021 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4022 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4023 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4024 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4025 }
4026
4027 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4028 {
4029 WARN_ON(in_interrupt());
4030 /* put off any impending NetWatchDogTimeout */
4031 adapter->netdev->trans_start = jiffies;
4032
4033 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4034 usleep_range(1000, 2000);
4035 ixgbe_down(adapter);
4036 /*
4037 * If SR-IOV enabled then wait a bit before bringing the adapter
4038 * back up to give the VFs time to respond to the reset. The
4039 * two second wait is based upon the watchdog timer cycle in
4040 * the VF driver.
4041 */
4042 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4043 msleep(2000);
4044 ixgbe_up(adapter);
4045 clear_bit(__IXGBE_RESETTING, &adapter->state);
4046 }
4047
4048 void ixgbe_up(struct ixgbe_adapter *adapter)
4049 {
4050 /* hardware has been reset, we need to reload some things */
4051 ixgbe_configure(adapter);
4052
4053 ixgbe_up_complete(adapter);
4054 }
4055
4056 void ixgbe_reset(struct ixgbe_adapter *adapter)
4057 {
4058 struct ixgbe_hw *hw = &adapter->hw;
4059 int err;
4060
4061 /* lock SFP init bit to prevent race conditions with the watchdog */
4062 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4063 usleep_range(1000, 2000);
4064
4065 /* clear all SFP and link config related flags while holding SFP_INIT */
4066 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4067 IXGBE_FLAG2_SFP_NEEDS_RESET);
4068 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4069
4070 err = hw->mac.ops.init_hw(hw);
4071 switch (err) {
4072 case 0:
4073 case IXGBE_ERR_SFP_NOT_PRESENT:
4074 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4075 break;
4076 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4077 e_dev_err("master disable timed out\n");
4078 break;
4079 case IXGBE_ERR_EEPROM_VERSION:
4080 /* We are running on a pre-production device, log a warning */
4081 e_dev_warn("This device is a pre-production adapter/LOM. "
4082 "Please be aware there may be issues associated with "
4083 "your hardware. If you are experiencing problems "
4084 "please contact your Intel or hardware "
4085 "representative who provided you with this "
4086 "hardware.\n");
4087 break;
4088 default:
4089 e_dev_err("Hardware Error: %d\n", err);
4090 }
4091
4092 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4093
4094 /* reprogram the RAR[0] in case user changed it. */
4095 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
4096 IXGBE_RAH_AV);
4097 }
4098
4099 /**
4100 * ixgbe_init_rx_page_offset - initialize page offset values for Rx buffers
4101 * @rx_ring: ring to setup
4102 *
4103 * On many IA platforms the L1 cache has a critical stride of 4K, this
4104 * results in each receive buffer starting in the same cache set. To help
4105 * reduce the pressure on this cache set we can interleave the offsets so
4106 * that only every other buffer will be in the same cache set.
4107 **/
4108 static void ixgbe_init_rx_page_offset(struct ixgbe_ring *rx_ring)
4109 {
4110 struct ixgbe_rx_buffer *rx_buffer = rx_ring->rx_buffer_info;
4111 u16 i;
4112
4113 for (i = 0; i < rx_ring->count; i += 2) {
4114 rx_buffer[0].page_offset = 0;
4115 rx_buffer[1].page_offset = ixgbe_rx_bufsz(rx_ring);
4116 rx_buffer = &rx_buffer[2];
4117 }
4118 }
4119
4120 /**
4121 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4122 * @rx_ring: ring to free buffers from
4123 **/
4124 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4125 {
4126 struct device *dev = rx_ring->dev;
4127 unsigned long size;
4128 u16 i;
4129
4130 /* ring already cleared, nothing to do */
4131 if (!rx_ring->rx_buffer_info)
4132 return;
4133
4134 /* Free all the Rx ring sk_buffs */
4135 for (i = 0; i < rx_ring->count; i++) {
4136 struct ixgbe_rx_buffer *rx_buffer;
4137
4138 rx_buffer = &rx_ring->rx_buffer_info[i];
4139 if (rx_buffer->skb) {
4140 struct sk_buff *skb = rx_buffer->skb;
4141 if (IXGBE_CB(skb)->page_released) {
4142 dma_unmap_page(dev,
4143 IXGBE_CB(skb)->dma,
4144 ixgbe_rx_bufsz(rx_ring),
4145 DMA_FROM_DEVICE);
4146 IXGBE_CB(skb)->page_released = false;
4147 }
4148 dev_kfree_skb(skb);
4149 }
4150 rx_buffer->skb = NULL;
4151 if (rx_buffer->dma)
4152 dma_unmap_page(dev, rx_buffer->dma,
4153 ixgbe_rx_pg_size(rx_ring),
4154 DMA_FROM_DEVICE);
4155 rx_buffer->dma = 0;
4156 if (rx_buffer->page)
4157 __free_pages(rx_buffer->page,
4158 ixgbe_rx_pg_order(rx_ring));
4159 rx_buffer->page = NULL;
4160 }
4161
4162 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4163 memset(rx_ring->rx_buffer_info, 0, size);
4164
4165 ixgbe_init_rx_page_offset(rx_ring);
4166
4167 /* Zero out the descriptor ring */
4168 memset(rx_ring->desc, 0, rx_ring->size);
4169
4170 rx_ring->next_to_alloc = 0;
4171 rx_ring->next_to_clean = 0;
4172 rx_ring->next_to_use = 0;
4173 }
4174
4175 /**
4176 * ixgbe_clean_tx_ring - Free Tx Buffers
4177 * @tx_ring: ring to be cleaned
4178 **/
4179 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4180 {
4181 struct ixgbe_tx_buffer *tx_buffer_info;
4182 unsigned long size;
4183 u16 i;
4184
4185 /* ring already cleared, nothing to do */
4186 if (!tx_ring->tx_buffer_info)
4187 return;
4188
4189 /* Free all the Tx ring sk_buffs */
4190 for (i = 0; i < tx_ring->count; i++) {
4191 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4192 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4193 }
4194
4195 netdev_tx_reset_queue(txring_txq(tx_ring));
4196
4197 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4198 memset(tx_ring->tx_buffer_info, 0, size);
4199
4200 /* Zero out the descriptor ring */
4201 memset(tx_ring->desc, 0, tx_ring->size);
4202
4203 tx_ring->next_to_use = 0;
4204 tx_ring->next_to_clean = 0;
4205 }
4206
4207 /**
4208 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4209 * @adapter: board private structure
4210 **/
4211 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4212 {
4213 int i;
4214
4215 for (i = 0; i < adapter->num_rx_queues; i++)
4216 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4217 }
4218
4219 /**
4220 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4221 * @adapter: board private structure
4222 **/
4223 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4224 {
4225 int i;
4226
4227 for (i = 0; i < adapter->num_tx_queues; i++)
4228 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4229 }
4230
4231 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4232 {
4233 struct hlist_node *node, *node2;
4234 struct ixgbe_fdir_filter *filter;
4235
4236 spin_lock(&adapter->fdir_perfect_lock);
4237
4238 hlist_for_each_entry_safe(filter, node, node2,
4239 &adapter->fdir_filter_list, fdir_node) {
4240 hlist_del(&filter->fdir_node);
4241 kfree(filter);
4242 }
4243 adapter->fdir_filter_count = 0;
4244
4245 spin_unlock(&adapter->fdir_perfect_lock);
4246 }
4247
4248 void ixgbe_down(struct ixgbe_adapter *adapter)
4249 {
4250 struct net_device *netdev = adapter->netdev;
4251 struct ixgbe_hw *hw = &adapter->hw;
4252 u32 rxctrl;
4253 int i;
4254
4255 /* signal that we are down to the interrupt handler */
4256 set_bit(__IXGBE_DOWN, &adapter->state);
4257
4258 /* disable receives */
4259 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4260 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4261
4262 /* disable all enabled rx queues */
4263 for (i = 0; i < adapter->num_rx_queues; i++)
4264 /* this call also flushes the previous write */
4265 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4266
4267 usleep_range(10000, 20000);
4268
4269 netif_tx_stop_all_queues(netdev);
4270
4271 /* call carrier off first to avoid false dev_watchdog timeouts */
4272 netif_carrier_off(netdev);
4273 netif_tx_disable(netdev);
4274
4275 ixgbe_irq_disable(adapter);
4276
4277 ixgbe_napi_disable_all(adapter);
4278
4279 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4280 IXGBE_FLAG2_RESET_REQUESTED);
4281 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4282
4283 del_timer_sync(&adapter->service_timer);
4284
4285 if (adapter->num_vfs) {
4286 /* Clear EITR Select mapping */
4287 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4288
4289 /* Mark all the VFs as inactive */
4290 for (i = 0 ; i < adapter->num_vfs; i++)
4291 adapter->vfinfo[i].clear_to_send = false;
4292
4293 /* ping all the active vfs to let them know we are going down */
4294 ixgbe_ping_all_vfs(adapter);
4295
4296 /* Disable all VFTE/VFRE TX/RX */
4297 ixgbe_disable_tx_rx(adapter);
4298 }
4299
4300 /* disable transmits in the hardware now that interrupts are off */
4301 for (i = 0; i < adapter->num_tx_queues; i++) {
4302 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4303 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4304 }
4305
4306 /* Disable the Tx DMA engine on 82599 and X540 */
4307 switch (hw->mac.type) {
4308 case ixgbe_mac_82599EB:
4309 case ixgbe_mac_X540:
4310 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4311 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4312 ~IXGBE_DMATXCTL_TE));
4313 break;
4314 default:
4315 break;
4316 }
4317
4318 if (!pci_channel_offline(adapter->pdev))
4319 ixgbe_reset(adapter);
4320
4321 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4322 if (hw->mac.ops.disable_tx_laser &&
4323 ((hw->phy.multispeed_fiber) ||
4324 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4325 (hw->mac.type == ixgbe_mac_82599EB))))
4326 hw->mac.ops.disable_tx_laser(hw);
4327
4328 ixgbe_clean_all_tx_rings(adapter);
4329 ixgbe_clean_all_rx_rings(adapter);
4330
4331 #ifdef CONFIG_IXGBE_DCA
4332 /* since we reset the hardware DCA settings were cleared */
4333 ixgbe_setup_dca(adapter);
4334 #endif
4335 }
4336
4337 /**
4338 * ixgbe_tx_timeout - Respond to a Tx Hang
4339 * @netdev: network interface device structure
4340 **/
4341 static void ixgbe_tx_timeout(struct net_device *netdev)
4342 {
4343 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4344
4345 /* Do the reset outside of interrupt context */
4346 ixgbe_tx_timeout_reset(adapter);
4347 }
4348
4349 /**
4350 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4351 * @adapter: board private structure to initialize
4352 *
4353 * ixgbe_sw_init initializes the Adapter private data structure.
4354 * Fields are initialized based on PCI device information and
4355 * OS network device settings (MTU size).
4356 **/
4357 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4358 {
4359 struct ixgbe_hw *hw = &adapter->hw;
4360 struct pci_dev *pdev = adapter->pdev;
4361 unsigned int rss;
4362 #ifdef CONFIG_IXGBE_DCB
4363 int j;
4364 struct tc_configuration *tc;
4365 #endif
4366
4367 /* PCI config space info */
4368
4369 hw->vendor_id = pdev->vendor;
4370 hw->device_id = pdev->device;
4371 hw->revision_id = pdev->revision;
4372 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4373 hw->subsystem_device_id = pdev->subsystem_device;
4374
4375 /* Set capability flags */
4376 rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4377 adapter->ring_feature[RING_F_RSS].limit = rss;
4378 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4379 switch (hw->mac.type) {
4380 case ixgbe_mac_82598EB:
4381 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4382 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4383 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
4384 break;
4385 case ixgbe_mac_X540:
4386 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4387 case ixgbe_mac_82599EB:
4388 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
4389 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4390 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4391 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4392 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4393 /* Flow Director hash filters enabled */
4394 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4395 adapter->atr_sample_rate = 20;
4396 adapter->ring_feature[RING_F_FDIR].limit =
4397 IXGBE_MAX_FDIR_INDICES;
4398 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4399 #ifdef IXGBE_FCOE
4400 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4401 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4402 #ifdef CONFIG_IXGBE_DCB
4403 /* Default traffic class to use for FCoE */
4404 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4405 #endif
4406 #endif /* IXGBE_FCOE */
4407 break;
4408 default:
4409 break;
4410 }
4411
4412 /* n-tuple support exists, always init our spinlock */
4413 spin_lock_init(&adapter->fdir_perfect_lock);
4414
4415 #ifdef CONFIG_IXGBE_DCB
4416 switch (hw->mac.type) {
4417 case ixgbe_mac_X540:
4418 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4419 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4420 break;
4421 default:
4422 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4423 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4424 break;
4425 }
4426
4427 /* Configure DCB traffic classes */
4428 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4429 tc = &adapter->dcb_cfg.tc_config[j];
4430 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4431 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4432 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4433 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4434 tc->dcb_pfc = pfc_disabled;
4435 }
4436
4437 /* Initialize default user to priority mapping, UPx->TC0 */
4438 tc = &adapter->dcb_cfg.tc_config[0];
4439 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
4440 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
4441
4442 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4443 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4444 adapter->dcb_cfg.pfc_mode_enable = false;
4445 adapter->dcb_set_bitmap = 0x00;
4446 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4447 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
4448 sizeof(adapter->temp_dcb_cfg));
4449
4450 #endif
4451
4452 /* default flow control settings */
4453 hw->fc.requested_mode = ixgbe_fc_full;
4454 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4455 ixgbe_pbthresh_setup(adapter);
4456 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4457 hw->fc.send_xon = true;
4458 hw->fc.disable_fc_autoneg = false;
4459
4460 /* enable itr by default in dynamic mode */
4461 adapter->rx_itr_setting = 1;
4462 adapter->tx_itr_setting = 1;
4463
4464 /* set default ring sizes */
4465 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4466 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4467
4468 /* set default work limits */
4469 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4470
4471 /* initialize eeprom parameters */
4472 if (ixgbe_init_eeprom_params_generic(hw)) {
4473 e_dev_err("EEPROM initialization failed\n");
4474 return -EIO;
4475 }
4476
4477 set_bit(__IXGBE_DOWN, &adapter->state);
4478
4479 return 0;
4480 }
4481
4482 /**
4483 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4484 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4485 *
4486 * Return 0 on success, negative on failure
4487 **/
4488 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4489 {
4490 struct device *dev = tx_ring->dev;
4491 int orig_node = dev_to_node(dev);
4492 int numa_node = -1;
4493 int size;
4494
4495 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4496
4497 if (tx_ring->q_vector)
4498 numa_node = tx_ring->q_vector->numa_node;
4499
4500 tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
4501 if (!tx_ring->tx_buffer_info)
4502 tx_ring->tx_buffer_info = vzalloc(size);
4503 if (!tx_ring->tx_buffer_info)
4504 goto err;
4505
4506 /* round up to nearest 4K */
4507 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4508 tx_ring->size = ALIGN(tx_ring->size, 4096);
4509
4510 set_dev_node(dev, numa_node);
4511 tx_ring->desc = dma_alloc_coherent(dev,
4512 tx_ring->size,
4513 &tx_ring->dma,
4514 GFP_KERNEL);
4515 set_dev_node(dev, orig_node);
4516 if (!tx_ring->desc)
4517 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4518 &tx_ring->dma, GFP_KERNEL);
4519 if (!tx_ring->desc)
4520 goto err;
4521
4522 tx_ring->next_to_use = 0;
4523 tx_ring->next_to_clean = 0;
4524 return 0;
4525
4526 err:
4527 vfree(tx_ring->tx_buffer_info);
4528 tx_ring->tx_buffer_info = NULL;
4529 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4530 return -ENOMEM;
4531 }
4532
4533 /**
4534 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4535 * @adapter: board private structure
4536 *
4537 * If this function returns with an error, then it's possible one or
4538 * more of the rings is populated (while the rest are not). It is the
4539 * callers duty to clean those orphaned rings.
4540 *
4541 * Return 0 on success, negative on failure
4542 **/
4543 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4544 {
4545 int i, err = 0;
4546
4547 for (i = 0; i < adapter->num_tx_queues; i++) {
4548 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4549 if (!err)
4550 continue;
4551 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4552 break;
4553 }
4554
4555 return err;
4556 }
4557
4558 /**
4559 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4560 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4561 *
4562 * Returns 0 on success, negative on failure
4563 **/
4564 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4565 {
4566 struct device *dev = rx_ring->dev;
4567 int orig_node = dev_to_node(dev);
4568 int numa_node = -1;
4569 int size;
4570
4571 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4572
4573 if (rx_ring->q_vector)
4574 numa_node = rx_ring->q_vector->numa_node;
4575
4576 rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
4577 if (!rx_ring->rx_buffer_info)
4578 rx_ring->rx_buffer_info = vzalloc(size);
4579 if (!rx_ring->rx_buffer_info)
4580 goto err;
4581
4582 /* Round up to nearest 4K */
4583 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4584 rx_ring->size = ALIGN(rx_ring->size, 4096);
4585
4586 set_dev_node(dev, numa_node);
4587 rx_ring->desc = dma_alloc_coherent(dev,
4588 rx_ring->size,
4589 &rx_ring->dma,
4590 GFP_KERNEL);
4591 set_dev_node(dev, orig_node);
4592 if (!rx_ring->desc)
4593 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4594 &rx_ring->dma, GFP_KERNEL);
4595 if (!rx_ring->desc)
4596 goto err;
4597
4598 rx_ring->next_to_clean = 0;
4599 rx_ring->next_to_use = 0;
4600
4601 ixgbe_init_rx_page_offset(rx_ring);
4602
4603 return 0;
4604 err:
4605 vfree(rx_ring->rx_buffer_info);
4606 rx_ring->rx_buffer_info = NULL;
4607 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4608 return -ENOMEM;
4609 }
4610
4611 /**
4612 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4613 * @adapter: board private structure
4614 *
4615 * If this function returns with an error, then it's possible one or
4616 * more of the rings is populated (while the rest are not). It is the
4617 * callers duty to clean those orphaned rings.
4618 *
4619 * Return 0 on success, negative on failure
4620 **/
4621 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4622 {
4623 int i, err = 0;
4624
4625 for (i = 0; i < adapter->num_rx_queues; i++) {
4626 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
4627 if (!err)
4628 continue;
4629 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4630 break;
4631 }
4632
4633 return err;
4634 }
4635
4636 /**
4637 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4638 * @tx_ring: Tx descriptor ring for a specific queue
4639 *
4640 * Free all transmit software resources
4641 **/
4642 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
4643 {
4644 ixgbe_clean_tx_ring(tx_ring);
4645
4646 vfree(tx_ring->tx_buffer_info);
4647 tx_ring->tx_buffer_info = NULL;
4648
4649 /* if not set, then don't free */
4650 if (!tx_ring->desc)
4651 return;
4652
4653 dma_free_coherent(tx_ring->dev, tx_ring->size,
4654 tx_ring->desc, tx_ring->dma);
4655
4656 tx_ring->desc = NULL;
4657 }
4658
4659 /**
4660 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4661 * @adapter: board private structure
4662 *
4663 * Free all transmit software resources
4664 **/
4665 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4666 {
4667 int i;
4668
4669 for (i = 0; i < adapter->num_tx_queues; i++)
4670 if (adapter->tx_ring[i]->desc)
4671 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4672 }
4673
4674 /**
4675 * ixgbe_free_rx_resources - Free Rx Resources
4676 * @rx_ring: ring to clean the resources from
4677 *
4678 * Free all receive software resources
4679 **/
4680 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
4681 {
4682 ixgbe_clean_rx_ring(rx_ring);
4683
4684 vfree(rx_ring->rx_buffer_info);
4685 rx_ring->rx_buffer_info = NULL;
4686
4687 /* if not set, then don't free */
4688 if (!rx_ring->desc)
4689 return;
4690
4691 dma_free_coherent(rx_ring->dev, rx_ring->size,
4692 rx_ring->desc, rx_ring->dma);
4693
4694 rx_ring->desc = NULL;
4695 }
4696
4697 /**
4698 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4699 * @adapter: board private structure
4700 *
4701 * Free all receive software resources
4702 **/
4703 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4704 {
4705 int i;
4706
4707 for (i = 0; i < adapter->num_rx_queues; i++)
4708 if (adapter->rx_ring[i]->desc)
4709 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4710 }
4711
4712 /**
4713 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4714 * @netdev: network interface device structure
4715 * @new_mtu: new value for maximum frame size
4716 *
4717 * Returns 0 on success, negative on failure
4718 **/
4719 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4720 {
4721 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4722 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4723
4724 /* MTU < 68 is an error and causes problems on some kernels */
4725 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4726 return -EINVAL;
4727
4728 /*
4729 * For 82599EB we cannot allow PF to change MTU greater than 1500
4730 * in SR-IOV mode as it may cause buffer overruns in guest VFs that
4731 * don't allocate and chain buffers correctly.
4732 */
4733 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
4734 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
4735 (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
4736 return -EINVAL;
4737
4738 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4739
4740 /* must set new MTU before calling down or up */
4741 netdev->mtu = new_mtu;
4742
4743 if (netif_running(netdev))
4744 ixgbe_reinit_locked(adapter);
4745
4746 return 0;
4747 }
4748
4749 /**
4750 * ixgbe_open - Called when a network interface is made active
4751 * @netdev: network interface device structure
4752 *
4753 * Returns 0 on success, negative value on failure
4754 *
4755 * The open entry point is called when a network interface is made
4756 * active by the system (IFF_UP). At this point all resources needed
4757 * for transmit and receive operations are allocated, the interrupt
4758 * handler is registered with the OS, the watchdog timer is started,
4759 * and the stack is notified that the interface is ready.
4760 **/
4761 static int ixgbe_open(struct net_device *netdev)
4762 {
4763 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4764 int err;
4765
4766 /* disallow open during test */
4767 if (test_bit(__IXGBE_TESTING, &adapter->state))
4768 return -EBUSY;
4769
4770 netif_carrier_off(netdev);
4771
4772 /* allocate transmit descriptors */
4773 err = ixgbe_setup_all_tx_resources(adapter);
4774 if (err)
4775 goto err_setup_tx;
4776
4777 /* allocate receive descriptors */
4778 err = ixgbe_setup_all_rx_resources(adapter);
4779 if (err)
4780 goto err_setup_rx;
4781
4782 ixgbe_configure(adapter);
4783
4784 err = ixgbe_request_irq(adapter);
4785 if (err)
4786 goto err_req_irq;
4787
4788 ixgbe_up_complete(adapter);
4789
4790 return 0;
4791
4792 err_req_irq:
4793 err_setup_rx:
4794 ixgbe_free_all_rx_resources(adapter);
4795 err_setup_tx:
4796 ixgbe_free_all_tx_resources(adapter);
4797 ixgbe_reset(adapter);
4798
4799 return err;
4800 }
4801
4802 /**
4803 * ixgbe_close - Disables a network interface
4804 * @netdev: network interface device structure
4805 *
4806 * Returns 0, this is not allowed to fail
4807 *
4808 * The close entry point is called when an interface is de-activated
4809 * by the OS. The hardware is still under the drivers control, but
4810 * needs to be disabled. A global MAC reset is issued to stop the
4811 * hardware, and all transmit and receive resources are freed.
4812 **/
4813 static int ixgbe_close(struct net_device *netdev)
4814 {
4815 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4816
4817 ixgbe_down(adapter);
4818 ixgbe_free_irq(adapter);
4819
4820 ixgbe_fdir_filter_exit(adapter);
4821
4822 ixgbe_free_all_tx_resources(adapter);
4823 ixgbe_free_all_rx_resources(adapter);
4824
4825 ixgbe_release_hw_control(adapter);
4826
4827 return 0;
4828 }
4829
4830 #ifdef CONFIG_PM
4831 static int ixgbe_resume(struct pci_dev *pdev)
4832 {
4833 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4834 struct net_device *netdev = adapter->netdev;
4835 u32 err;
4836
4837 pci_set_power_state(pdev, PCI_D0);
4838 pci_restore_state(pdev);
4839 /*
4840 * pci_restore_state clears dev->state_saved so call
4841 * pci_save_state to restore it.
4842 */
4843 pci_save_state(pdev);
4844
4845 err = pci_enable_device_mem(pdev);
4846 if (err) {
4847 e_dev_err("Cannot enable PCI device from suspend\n");
4848 return err;
4849 }
4850 pci_set_master(pdev);
4851
4852 pci_wake_from_d3(pdev, false);
4853
4854 rtnl_lock();
4855 err = ixgbe_init_interrupt_scheme(adapter);
4856 rtnl_unlock();
4857 if (err) {
4858 e_dev_err("Cannot initialize interrupts for device\n");
4859 return err;
4860 }
4861
4862 ixgbe_reset(adapter);
4863
4864 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4865
4866 if (netif_running(netdev)) {
4867 err = ixgbe_open(netdev);
4868 if (err)
4869 return err;
4870 }
4871
4872 netif_device_attach(netdev);
4873
4874 return 0;
4875 }
4876 #endif /* CONFIG_PM */
4877
4878 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4879 {
4880 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4881 struct net_device *netdev = adapter->netdev;
4882 struct ixgbe_hw *hw = &adapter->hw;
4883 u32 ctrl, fctrl;
4884 u32 wufc = adapter->wol;
4885 #ifdef CONFIG_PM
4886 int retval = 0;
4887 #endif
4888
4889 netif_device_detach(netdev);
4890
4891 if (netif_running(netdev)) {
4892 rtnl_lock();
4893 ixgbe_down(adapter);
4894 ixgbe_free_irq(adapter);
4895 ixgbe_free_all_tx_resources(adapter);
4896 ixgbe_free_all_rx_resources(adapter);
4897 rtnl_unlock();
4898 }
4899
4900 ixgbe_clear_interrupt_scheme(adapter);
4901
4902 #ifdef CONFIG_PM
4903 retval = pci_save_state(pdev);
4904 if (retval)
4905 return retval;
4906
4907 #endif
4908 if (wufc) {
4909 ixgbe_set_rx_mode(netdev);
4910
4911 /*
4912 * enable the optics for both mult-speed fiber and
4913 * 82599 SFP+ fiber as we can WoL.
4914 */
4915 if (hw->mac.ops.enable_tx_laser &&
4916 (hw->phy.multispeed_fiber ||
4917 (hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber &&
4918 hw->mac.type == ixgbe_mac_82599EB)))
4919 hw->mac.ops.enable_tx_laser(hw);
4920
4921 /* turn on all-multi mode if wake on multicast is enabled */
4922 if (wufc & IXGBE_WUFC_MC) {
4923 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4924 fctrl |= IXGBE_FCTRL_MPE;
4925 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4926 }
4927
4928 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4929 ctrl |= IXGBE_CTRL_GIO_DIS;
4930 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4931
4932 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4933 } else {
4934 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4935 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4936 }
4937
4938 switch (hw->mac.type) {
4939 case ixgbe_mac_82598EB:
4940 pci_wake_from_d3(pdev, false);
4941 break;
4942 case ixgbe_mac_82599EB:
4943 case ixgbe_mac_X540:
4944 pci_wake_from_d3(pdev, !!wufc);
4945 break;
4946 default:
4947 break;
4948 }
4949
4950 *enable_wake = !!wufc;
4951
4952 ixgbe_release_hw_control(adapter);
4953
4954 pci_disable_device(pdev);
4955
4956 return 0;
4957 }
4958
4959 #ifdef CONFIG_PM
4960 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4961 {
4962 int retval;
4963 bool wake;
4964
4965 retval = __ixgbe_shutdown(pdev, &wake);
4966 if (retval)
4967 return retval;
4968
4969 if (wake) {
4970 pci_prepare_to_sleep(pdev);
4971 } else {
4972 pci_wake_from_d3(pdev, false);
4973 pci_set_power_state(pdev, PCI_D3hot);
4974 }
4975
4976 return 0;
4977 }
4978 #endif /* CONFIG_PM */
4979
4980 static void ixgbe_shutdown(struct pci_dev *pdev)
4981 {
4982 bool wake;
4983
4984 __ixgbe_shutdown(pdev, &wake);
4985
4986 if (system_state == SYSTEM_POWER_OFF) {
4987 pci_wake_from_d3(pdev, wake);
4988 pci_set_power_state(pdev, PCI_D3hot);
4989 }
4990 }
4991
4992 /**
4993 * ixgbe_update_stats - Update the board statistics counters.
4994 * @adapter: board private structure
4995 **/
4996 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4997 {
4998 struct net_device *netdev = adapter->netdev;
4999 struct ixgbe_hw *hw = &adapter->hw;
5000 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5001 u64 total_mpc = 0;
5002 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5003 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5004 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5005 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5006 #ifdef IXGBE_FCOE
5007 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5008 unsigned int cpu;
5009 u64 fcoe_noddp_counts_sum = 0, fcoe_noddp_ext_buff_counts_sum = 0;
5010 #endif /* IXGBE_FCOE */
5011
5012 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5013 test_bit(__IXGBE_RESETTING, &adapter->state))
5014 return;
5015
5016 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5017 u64 rsc_count = 0;
5018 u64 rsc_flush = 0;
5019 for (i = 0; i < adapter->num_rx_queues; i++) {
5020 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5021 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5022 }
5023 adapter->rsc_total_count = rsc_count;
5024 adapter->rsc_total_flush = rsc_flush;
5025 }
5026
5027 for (i = 0; i < adapter->num_rx_queues; i++) {
5028 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5029 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5030 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5031 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5032 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5033 bytes += rx_ring->stats.bytes;
5034 packets += rx_ring->stats.packets;
5035 }
5036 adapter->non_eop_descs = non_eop_descs;
5037 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5038 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5039 adapter->hw_csum_rx_error = hw_csum_rx_error;
5040 netdev->stats.rx_bytes = bytes;
5041 netdev->stats.rx_packets = packets;
5042
5043 bytes = 0;
5044 packets = 0;
5045 /* gather some stats to the adapter struct that are per queue */
5046 for (i = 0; i < adapter->num_tx_queues; i++) {
5047 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5048 restart_queue += tx_ring->tx_stats.restart_queue;
5049 tx_busy += tx_ring->tx_stats.tx_busy;
5050 bytes += tx_ring->stats.bytes;
5051 packets += tx_ring->stats.packets;
5052 }
5053 adapter->restart_queue = restart_queue;
5054 adapter->tx_busy = tx_busy;
5055 netdev->stats.tx_bytes = bytes;
5056 netdev->stats.tx_packets = packets;
5057
5058 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5059
5060 /* 8 register reads */
5061 for (i = 0; i < 8; i++) {
5062 /* for packet buffers not used, the register should read 0 */
5063 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5064 missed_rx += mpc;
5065 hwstats->mpc[i] += mpc;
5066 total_mpc += hwstats->mpc[i];
5067 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5068 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5069 switch (hw->mac.type) {
5070 case ixgbe_mac_82598EB:
5071 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5072 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5073 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5074 hwstats->pxonrxc[i] +=
5075 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5076 break;
5077 case ixgbe_mac_82599EB:
5078 case ixgbe_mac_X540:
5079 hwstats->pxonrxc[i] +=
5080 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5081 break;
5082 default:
5083 break;
5084 }
5085 }
5086
5087 /*16 register reads */
5088 for (i = 0; i < 16; i++) {
5089 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5090 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5091 if ((hw->mac.type == ixgbe_mac_82599EB) ||
5092 (hw->mac.type == ixgbe_mac_X540)) {
5093 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5094 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5095 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5096 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5097 }
5098 }
5099
5100 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5101 /* work around hardware counting issue */
5102 hwstats->gprc -= missed_rx;
5103
5104 ixgbe_update_xoff_received(adapter);
5105
5106 /* 82598 hardware only has a 32 bit counter in the high register */
5107 switch (hw->mac.type) {
5108 case ixgbe_mac_82598EB:
5109 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5110 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5111 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5112 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5113 break;
5114 case ixgbe_mac_X540:
5115 /* OS2BMC stats are X540 only*/
5116 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5117 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5118 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5119 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5120 case ixgbe_mac_82599EB:
5121 for (i = 0; i < 16; i++)
5122 adapter->hw_rx_no_dma_resources +=
5123 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5124 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5125 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5126 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5127 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5128 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5129 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5130 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5131 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5132 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5133 #ifdef IXGBE_FCOE
5134 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5135 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5136 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5137 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5138 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5139 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5140 /* Add up per cpu counters for total ddp aloc fail */
5141 if (fcoe->pcpu_noddp && fcoe->pcpu_noddp_ext_buff) {
5142 for_each_possible_cpu(cpu) {
5143 fcoe_noddp_counts_sum +=
5144 *per_cpu_ptr(fcoe->pcpu_noddp, cpu);
5145 fcoe_noddp_ext_buff_counts_sum +=
5146 *per_cpu_ptr(fcoe->
5147 pcpu_noddp_ext_buff, cpu);
5148 }
5149 }
5150 hwstats->fcoe_noddp = fcoe_noddp_counts_sum;
5151 hwstats->fcoe_noddp_ext_buff = fcoe_noddp_ext_buff_counts_sum;
5152 #endif /* IXGBE_FCOE */
5153 break;
5154 default:
5155 break;
5156 }
5157 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5158 hwstats->bprc += bprc;
5159 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5160 if (hw->mac.type == ixgbe_mac_82598EB)
5161 hwstats->mprc -= bprc;
5162 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5163 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5164 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5165 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5166 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5167 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5168 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5169 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5170 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5171 hwstats->lxontxc += lxon;
5172 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5173 hwstats->lxofftxc += lxoff;
5174 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5175 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5176 /*
5177 * 82598 errata - tx of flow control packets is included in tx counters
5178 */
5179 xon_off_tot = lxon + lxoff;
5180 hwstats->gptc -= xon_off_tot;
5181 hwstats->mptc -= xon_off_tot;
5182 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5183 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5184 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5185 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5186 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5187 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5188 hwstats->ptc64 -= xon_off_tot;
5189 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5190 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5191 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5192 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5193 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5194 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5195
5196 /* Fill out the OS statistics structure */
5197 netdev->stats.multicast = hwstats->mprc;
5198
5199 /* Rx Errors */
5200 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5201 netdev->stats.rx_dropped = 0;
5202 netdev->stats.rx_length_errors = hwstats->rlec;
5203 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5204 netdev->stats.rx_missed_errors = total_mpc;
5205 }
5206
5207 /**
5208 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5209 * @adapter: pointer to the device adapter structure
5210 **/
5211 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5212 {
5213 struct ixgbe_hw *hw = &adapter->hw;
5214 int i;
5215
5216 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5217 return;
5218
5219 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5220
5221 /* if interface is down do nothing */
5222 if (test_bit(__IXGBE_DOWN, &adapter->state))
5223 return;
5224
5225 /* do nothing if we are not using signature filters */
5226 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5227 return;
5228
5229 adapter->fdir_overflow++;
5230
5231 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5232 for (i = 0; i < adapter->num_tx_queues; i++)
5233 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5234 &(adapter->tx_ring[i]->state));
5235 /* re-enable flow director interrupts */
5236 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5237 } else {
5238 e_err(probe, "failed to finish FDIR re-initialization, "
5239 "ignored adding FDIR ATR filters\n");
5240 }
5241 }
5242
5243 /**
5244 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5245 * @adapter: pointer to the device adapter structure
5246 *
5247 * This function serves two purposes. First it strobes the interrupt lines
5248 * in order to make certain interrupts are occurring. Secondly it sets the
5249 * bits needed to check for TX hangs. As a result we should immediately
5250 * determine if a hang has occurred.
5251 */
5252 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5253 {
5254 struct ixgbe_hw *hw = &adapter->hw;
5255 u64 eics = 0;
5256 int i;
5257
5258 /* If we're down or resetting, just bail */
5259 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5260 test_bit(__IXGBE_RESETTING, &adapter->state))
5261 return;
5262
5263 /* Force detection of hung controller */
5264 if (netif_carrier_ok(adapter->netdev)) {
5265 for (i = 0; i < adapter->num_tx_queues; i++)
5266 set_check_for_tx_hang(adapter->tx_ring[i]);
5267 }
5268
5269 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5270 /*
5271 * for legacy and MSI interrupts don't set any bits
5272 * that are enabled for EIAM, because this operation
5273 * would set *both* EIMS and EICS for any bit in EIAM
5274 */
5275 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5276 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5277 } else {
5278 /* get one bit for every active tx/rx interrupt vector */
5279 for (i = 0; i < adapter->num_q_vectors; i++) {
5280 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5281 if (qv->rx.ring || qv->tx.ring)
5282 eics |= ((u64)1 << i);
5283 }
5284 }
5285
5286 /* Cause software interrupt to ensure rings are cleaned */
5287 ixgbe_irq_rearm_queues(adapter, eics);
5288
5289 }
5290
5291 /**
5292 * ixgbe_watchdog_update_link - update the link status
5293 * @adapter: pointer to the device adapter structure
5294 * @link_speed: pointer to a u32 to store the link_speed
5295 **/
5296 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5297 {
5298 struct ixgbe_hw *hw = &adapter->hw;
5299 u32 link_speed = adapter->link_speed;
5300 bool link_up = adapter->link_up;
5301 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
5302
5303 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5304 return;
5305
5306 if (hw->mac.ops.check_link) {
5307 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5308 } else {
5309 /* always assume link is up, if no check link function */
5310 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5311 link_up = true;
5312 }
5313
5314 if (adapter->ixgbe_ieee_pfc)
5315 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
5316
5317 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
5318 hw->mac.ops.fc_enable(hw);
5319 ixgbe_set_rx_drop_en(adapter);
5320 }
5321
5322 if (link_up ||
5323 time_after(jiffies, (adapter->link_check_timeout +
5324 IXGBE_TRY_LINK_TIMEOUT))) {
5325 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5326 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5327 IXGBE_WRITE_FLUSH(hw);
5328 }
5329
5330 adapter->link_up = link_up;
5331 adapter->link_speed = link_speed;
5332 }
5333
5334 /**
5335 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5336 * print link up message
5337 * @adapter: pointer to the device adapter structure
5338 **/
5339 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5340 {
5341 struct net_device *netdev = adapter->netdev;
5342 struct ixgbe_hw *hw = &adapter->hw;
5343 u32 link_speed = adapter->link_speed;
5344 bool flow_rx, flow_tx;
5345
5346 /* only continue if link was previously down */
5347 if (netif_carrier_ok(netdev))
5348 return;
5349
5350 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5351
5352 switch (hw->mac.type) {
5353 case ixgbe_mac_82598EB: {
5354 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5355 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5356 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5357 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5358 }
5359 break;
5360 case ixgbe_mac_X540:
5361 case ixgbe_mac_82599EB: {
5362 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5363 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5364 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5365 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5366 }
5367 break;
5368 default:
5369 flow_tx = false;
5370 flow_rx = false;
5371 break;
5372 }
5373
5374 #ifdef CONFIG_IXGBE_PTP
5375 ixgbe_ptp_start_cyclecounter(adapter);
5376 #endif
5377
5378 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5379 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5380 "10 Gbps" :
5381 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5382 "1 Gbps" :
5383 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5384 "100 Mbps" :
5385 "unknown speed"))),
5386 ((flow_rx && flow_tx) ? "RX/TX" :
5387 (flow_rx ? "RX" :
5388 (flow_tx ? "TX" : "None"))));
5389
5390 netif_carrier_on(netdev);
5391 ixgbe_check_vf_rate_limit(adapter);
5392 }
5393
5394 /**
5395 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5396 * print link down message
5397 * @adapter: pointer to the adapter structure
5398 **/
5399 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
5400 {
5401 struct net_device *netdev = adapter->netdev;
5402 struct ixgbe_hw *hw = &adapter->hw;
5403
5404 adapter->link_up = false;
5405 adapter->link_speed = 0;
5406
5407 /* only continue if link was up previously */
5408 if (!netif_carrier_ok(netdev))
5409 return;
5410
5411 /* poll for SFP+ cable when link is down */
5412 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5413 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5414
5415 #ifdef CONFIG_IXGBE_PTP
5416 ixgbe_ptp_start_cyclecounter(adapter);
5417 #endif
5418
5419 e_info(drv, "NIC Link is Down\n");
5420 netif_carrier_off(netdev);
5421 }
5422
5423 /**
5424 * ixgbe_watchdog_flush_tx - flush queues on link down
5425 * @adapter: pointer to the device adapter structure
5426 **/
5427 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5428 {
5429 int i;
5430 int some_tx_pending = 0;
5431
5432 if (!netif_carrier_ok(adapter->netdev)) {
5433 for (i = 0; i < adapter->num_tx_queues; i++) {
5434 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5435 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5436 some_tx_pending = 1;
5437 break;
5438 }
5439 }
5440
5441 if (some_tx_pending) {
5442 /* We've lost link, so the controller stops DMA,
5443 * but we've got queued Tx work that's never going
5444 * to get done, so reset controller to flush Tx.
5445 * (Do the reset outside of interrupt context).
5446 */
5447 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5448 }
5449 }
5450 }
5451
5452 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5453 {
5454 u32 ssvpc;
5455
5456 /* Do not perform spoof check for 82598 */
5457 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5458 return;
5459
5460 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5461
5462 /*
5463 * ssvpc register is cleared on read, if zero then no
5464 * spoofed packets in the last interval.
5465 */
5466 if (!ssvpc)
5467 return;
5468
5469 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
5470 }
5471
5472 /**
5473 * ixgbe_watchdog_subtask - check and bring link up
5474 * @adapter: pointer to the device adapter structure
5475 **/
5476 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5477 {
5478 /* if interface is down do nothing */
5479 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5480 test_bit(__IXGBE_RESETTING, &adapter->state))
5481 return;
5482
5483 ixgbe_watchdog_update_link(adapter);
5484
5485 if (adapter->link_up)
5486 ixgbe_watchdog_link_is_up(adapter);
5487 else
5488 ixgbe_watchdog_link_is_down(adapter);
5489
5490 ixgbe_spoof_check(adapter);
5491 ixgbe_update_stats(adapter);
5492
5493 ixgbe_watchdog_flush_tx(adapter);
5494 }
5495
5496 /**
5497 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5498 * @adapter: the ixgbe adapter structure
5499 **/
5500 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5501 {
5502 struct ixgbe_hw *hw = &adapter->hw;
5503 s32 err;
5504
5505 /* not searching for SFP so there is nothing to do here */
5506 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5507 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5508 return;
5509
5510 /* someone else is in init, wait until next service event */
5511 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5512 return;
5513
5514 err = hw->phy.ops.identify_sfp(hw);
5515 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5516 goto sfp_out;
5517
5518 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5519 /* If no cable is present, then we need to reset
5520 * the next time we find a good cable. */
5521 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5522 }
5523
5524 /* exit on error */
5525 if (err)
5526 goto sfp_out;
5527
5528 /* exit if reset not needed */
5529 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5530 goto sfp_out;
5531
5532 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5533
5534 /*
5535 * A module may be identified correctly, but the EEPROM may not have
5536 * support for that module. setup_sfp() will fail in that case, so
5537 * we should not allow that module to load.
5538 */
5539 if (hw->mac.type == ixgbe_mac_82598EB)
5540 err = hw->phy.ops.reset(hw);
5541 else
5542 err = hw->mac.ops.setup_sfp(hw);
5543
5544 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5545 goto sfp_out;
5546
5547 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5548 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5549
5550 sfp_out:
5551 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5552
5553 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5554 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5555 e_dev_err("failed to initialize because an unsupported "
5556 "SFP+ module type was detected.\n");
5557 e_dev_err("Reload the driver after installing a "
5558 "supported module.\n");
5559 unregister_netdev(adapter->netdev);
5560 }
5561 }
5562
5563 /**
5564 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5565 * @adapter: the ixgbe adapter structure
5566 **/
5567 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5568 {
5569 struct ixgbe_hw *hw = &adapter->hw;
5570 u32 autoneg;
5571 bool negotiation;
5572
5573 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5574 return;
5575
5576 /* someone else is in init, wait until next service event */
5577 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5578 return;
5579
5580 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5581
5582 autoneg = hw->phy.autoneg_advertised;
5583 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5584 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5585 if (hw->mac.ops.setup_link)
5586 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5587
5588 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5589 adapter->link_check_timeout = jiffies;
5590 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5591 }
5592
5593 #ifdef CONFIG_PCI_IOV
5594 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
5595 {
5596 int vf;
5597 struct ixgbe_hw *hw = &adapter->hw;
5598 struct net_device *netdev = adapter->netdev;
5599 u32 gpc;
5600 u32 ciaa, ciad;
5601
5602 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
5603 if (gpc) /* If incrementing then no need for the check below */
5604 return;
5605 /*
5606 * Check to see if a bad DMA write target from an errant or
5607 * malicious VF has caused a PCIe error. If so then we can
5608 * issue a VFLR to the offending VF(s) and then resume without
5609 * requesting a full slot reset.
5610 */
5611
5612 for (vf = 0; vf < adapter->num_vfs; vf++) {
5613 ciaa = (vf << 16) | 0x80000000;
5614 /* 32 bit read so align, we really want status at offset 6 */
5615 ciaa |= PCI_COMMAND;
5616 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5617 ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
5618 ciaa &= 0x7FFFFFFF;
5619 /* disable debug mode asap after reading data */
5620 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5621 /* Get the upper 16 bits which will be the PCI status reg */
5622 ciad >>= 16;
5623 if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
5624 netdev_err(netdev, "VF %d Hung DMA\n", vf);
5625 /* Issue VFLR */
5626 ciaa = (vf << 16) | 0x80000000;
5627 ciaa |= 0xA8;
5628 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5629 ciad = 0x00008000; /* VFLR */
5630 IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
5631 ciaa &= 0x7FFFFFFF;
5632 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5633 }
5634 }
5635 }
5636
5637 #endif
5638 /**
5639 * ixgbe_service_timer - Timer Call-back
5640 * @data: pointer to adapter cast into an unsigned long
5641 **/
5642 static void ixgbe_service_timer(unsigned long data)
5643 {
5644 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5645 unsigned long next_event_offset;
5646 bool ready = true;
5647
5648 /* poll faster when waiting for link */
5649 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5650 next_event_offset = HZ / 10;
5651 else
5652 next_event_offset = HZ * 2;
5653
5654 #ifdef CONFIG_PCI_IOV
5655 /*
5656 * don't bother with SR-IOV VF DMA hang check if there are
5657 * no VFs or the link is down
5658 */
5659 if (!adapter->num_vfs ||
5660 (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5661 goto normal_timer_service;
5662
5663 /* If we have VFs allocated then we must check for DMA hangs */
5664 ixgbe_check_for_bad_vf(adapter);
5665 next_event_offset = HZ / 50;
5666 adapter->timer_event_accumulator++;
5667
5668 if (adapter->timer_event_accumulator >= 100)
5669 adapter->timer_event_accumulator = 0;
5670 else
5671 ready = false;
5672
5673 normal_timer_service:
5674 #endif
5675 /* Reset the timer */
5676 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5677
5678 if (ready)
5679 ixgbe_service_event_schedule(adapter);
5680 }
5681
5682 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5683 {
5684 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
5685 return;
5686
5687 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
5688
5689 /* If we're already down or resetting, just bail */
5690 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5691 test_bit(__IXGBE_RESETTING, &adapter->state))
5692 return;
5693
5694 ixgbe_dump(adapter);
5695 netdev_err(adapter->netdev, "Reset adapter\n");
5696 adapter->tx_timeout_count++;
5697
5698 ixgbe_reinit_locked(adapter);
5699 }
5700
5701 /**
5702 * ixgbe_service_task - manages and runs subtasks
5703 * @work: pointer to work_struct containing our data
5704 **/
5705 static void ixgbe_service_task(struct work_struct *work)
5706 {
5707 struct ixgbe_adapter *adapter = container_of(work,
5708 struct ixgbe_adapter,
5709 service_task);
5710
5711 ixgbe_reset_subtask(adapter);
5712 ixgbe_sfp_detection_subtask(adapter);
5713 ixgbe_sfp_link_config_subtask(adapter);
5714 ixgbe_check_overtemp_subtask(adapter);
5715 ixgbe_watchdog_subtask(adapter);
5716 ixgbe_fdir_reinit_subtask(adapter);
5717 ixgbe_check_hang_subtask(adapter);
5718 #ifdef CONFIG_IXGBE_PTP
5719 ixgbe_ptp_overflow_check(adapter);
5720 #endif
5721
5722 ixgbe_service_event_complete(adapter);
5723 }
5724
5725 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
5726 struct ixgbe_tx_buffer *first,
5727 u8 *hdr_len)
5728 {
5729 struct sk_buff *skb = first->skb;
5730 u32 vlan_macip_lens, type_tucmd;
5731 u32 mss_l4len_idx, l4len;
5732
5733 if (!skb_is_gso(skb))
5734 return 0;
5735
5736 if (skb_header_cloned(skb)) {
5737 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5738 if (err)
5739 return err;
5740 }
5741
5742 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5743 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
5744
5745 if (first->protocol == __constant_htons(ETH_P_IP)) {
5746 struct iphdr *iph = ip_hdr(skb);
5747 iph->tot_len = 0;
5748 iph->check = 0;
5749 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5750 iph->daddr, 0,
5751 IPPROTO_TCP,
5752 0);
5753 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5754 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5755 IXGBE_TX_FLAGS_CSUM |
5756 IXGBE_TX_FLAGS_IPV4;
5757 } else if (skb_is_gso_v6(skb)) {
5758 ipv6_hdr(skb)->payload_len = 0;
5759 tcp_hdr(skb)->check =
5760 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5761 &ipv6_hdr(skb)->daddr,
5762 0, IPPROTO_TCP, 0);
5763 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5764 IXGBE_TX_FLAGS_CSUM;
5765 }
5766
5767 /* compute header lengths */
5768 l4len = tcp_hdrlen(skb);
5769 *hdr_len = skb_transport_offset(skb) + l4len;
5770
5771 /* update gso size and bytecount with header size */
5772 first->gso_segs = skb_shinfo(skb)->gso_segs;
5773 first->bytecount += (first->gso_segs - 1) * *hdr_len;
5774
5775 /* mss_l4len_id: use 1 as index for TSO */
5776 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
5777 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
5778 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
5779
5780 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
5781 vlan_macip_lens = skb_network_header_len(skb);
5782 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5783 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5784
5785 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
5786 mss_l4len_idx);
5787
5788 return 1;
5789 }
5790
5791 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
5792 struct ixgbe_tx_buffer *first)
5793 {
5794 struct sk_buff *skb = first->skb;
5795 u32 vlan_macip_lens = 0;
5796 u32 mss_l4len_idx = 0;
5797 u32 type_tucmd = 0;
5798
5799 if (skb->ip_summed != CHECKSUM_PARTIAL) {
5800 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
5801 !(first->tx_flags & IXGBE_TX_FLAGS_TXSW))
5802 return;
5803 } else {
5804 u8 l4_hdr = 0;
5805 switch (first->protocol) {
5806 case __constant_htons(ETH_P_IP):
5807 vlan_macip_lens |= skb_network_header_len(skb);
5808 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5809 l4_hdr = ip_hdr(skb)->protocol;
5810 break;
5811 case __constant_htons(ETH_P_IPV6):
5812 vlan_macip_lens |= skb_network_header_len(skb);
5813 l4_hdr = ipv6_hdr(skb)->nexthdr;
5814 break;
5815 default:
5816 if (unlikely(net_ratelimit())) {
5817 dev_warn(tx_ring->dev,
5818 "partial checksum but proto=%x!\n",
5819 first->protocol);
5820 }
5821 break;
5822 }
5823
5824 switch (l4_hdr) {
5825 case IPPROTO_TCP:
5826 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5827 mss_l4len_idx = tcp_hdrlen(skb) <<
5828 IXGBE_ADVTXD_L4LEN_SHIFT;
5829 break;
5830 case IPPROTO_SCTP:
5831 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5832 mss_l4len_idx = sizeof(struct sctphdr) <<
5833 IXGBE_ADVTXD_L4LEN_SHIFT;
5834 break;
5835 case IPPROTO_UDP:
5836 mss_l4len_idx = sizeof(struct udphdr) <<
5837 IXGBE_ADVTXD_L4LEN_SHIFT;
5838 break;
5839 default:
5840 if (unlikely(net_ratelimit())) {
5841 dev_warn(tx_ring->dev,
5842 "partial checksum but l4 proto=%x!\n",
5843 l4_hdr);
5844 }
5845 break;
5846 }
5847
5848 /* update TX checksum flag */
5849 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
5850 }
5851
5852 /* vlan_macip_lens: MACLEN, VLAN tag */
5853 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5854 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5855
5856 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
5857 type_tucmd, mss_l4len_idx);
5858 }
5859
5860 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
5861 {
5862 /* set type for advanced descriptor with frame checksum insertion */
5863 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
5864 IXGBE_ADVTXD_DCMD_IFCS |
5865 IXGBE_ADVTXD_DCMD_DEXT);
5866
5867 /* set HW vlan bit if vlan is present */
5868 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
5869 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
5870
5871 #ifdef CONFIG_IXGBE_PTP
5872 if (tx_flags & IXGBE_TX_FLAGS_TSTAMP)
5873 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_MAC_TSTAMP);
5874 #endif
5875
5876 /* set segmentation enable bits for TSO/FSO */
5877 #ifdef IXGBE_FCOE
5878 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FSO))
5879 #else
5880 if (tx_flags & IXGBE_TX_FLAGS_TSO)
5881 #endif
5882 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
5883
5884 return cmd_type;
5885 }
5886
5887 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
5888 u32 tx_flags, unsigned int paylen)
5889 {
5890 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
5891
5892 /* enable L4 checksum for TSO and TX checksum offload */
5893 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5894 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
5895
5896 /* enble IPv4 checksum for TSO */
5897 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5898 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
5899
5900 /* use index 1 context for TSO/FSO/FCOE */
5901 #ifdef IXGBE_FCOE
5902 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FCOE))
5903 #else
5904 if (tx_flags & IXGBE_TX_FLAGS_TSO)
5905 #endif
5906 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
5907
5908 /*
5909 * Check Context must be set if Tx switch is enabled, which it
5910 * always is for case where virtual functions are running
5911 */
5912 #ifdef IXGBE_FCOE
5913 if (tx_flags & (IXGBE_TX_FLAGS_TXSW | IXGBE_TX_FLAGS_FCOE))
5914 #else
5915 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
5916 #endif
5917 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
5918
5919 tx_desc->read.olinfo_status = olinfo_status;
5920 }
5921
5922 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
5923 IXGBE_TXD_CMD_RS)
5924
5925 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
5926 struct ixgbe_tx_buffer *first,
5927 const u8 hdr_len)
5928 {
5929 dma_addr_t dma;
5930 struct sk_buff *skb = first->skb;
5931 struct ixgbe_tx_buffer *tx_buffer;
5932 union ixgbe_adv_tx_desc *tx_desc;
5933 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
5934 unsigned int data_len = skb->data_len;
5935 unsigned int size = skb_headlen(skb);
5936 unsigned int paylen = skb->len - hdr_len;
5937 u32 tx_flags = first->tx_flags;
5938 __le32 cmd_type;
5939 u16 i = tx_ring->next_to_use;
5940
5941 tx_desc = IXGBE_TX_DESC(tx_ring, i);
5942
5943 ixgbe_tx_olinfo_status(tx_desc, tx_flags, paylen);
5944 cmd_type = ixgbe_tx_cmd_type(tx_flags);
5945
5946 #ifdef IXGBE_FCOE
5947 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5948 if (data_len < sizeof(struct fcoe_crc_eof)) {
5949 size -= sizeof(struct fcoe_crc_eof) - data_len;
5950 data_len = 0;
5951 } else {
5952 data_len -= sizeof(struct fcoe_crc_eof);
5953 }
5954 }
5955
5956 #endif
5957 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
5958 if (dma_mapping_error(tx_ring->dev, dma))
5959 goto dma_error;
5960
5961 /* record length, and DMA address */
5962 dma_unmap_len_set(first, len, size);
5963 dma_unmap_addr_set(first, dma, dma);
5964
5965 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5966
5967 for (;;) {
5968 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
5969 tx_desc->read.cmd_type_len =
5970 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
5971
5972 i++;
5973 tx_desc++;
5974 if (i == tx_ring->count) {
5975 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5976 i = 0;
5977 }
5978
5979 dma += IXGBE_MAX_DATA_PER_TXD;
5980 size -= IXGBE_MAX_DATA_PER_TXD;
5981
5982 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5983 tx_desc->read.olinfo_status = 0;
5984 }
5985
5986 if (likely(!data_len))
5987 break;
5988
5989 if (unlikely(skb->no_fcs))
5990 cmd_type &= ~(cpu_to_le32(IXGBE_ADVTXD_DCMD_IFCS));
5991 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
5992
5993 i++;
5994 tx_desc++;
5995 if (i == tx_ring->count) {
5996 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5997 i = 0;
5998 }
5999
6000 #ifdef IXGBE_FCOE
6001 size = min_t(unsigned int, data_len, skb_frag_size(frag));
6002 #else
6003 size = skb_frag_size(frag);
6004 #endif
6005 data_len -= size;
6006
6007 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6008 DMA_TO_DEVICE);
6009 if (dma_mapping_error(tx_ring->dev, dma))
6010 goto dma_error;
6011
6012 tx_buffer = &tx_ring->tx_buffer_info[i];
6013 dma_unmap_len_set(tx_buffer, len, size);
6014 dma_unmap_addr_set(tx_buffer, dma, dma);
6015
6016 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6017 tx_desc->read.olinfo_status = 0;
6018
6019 frag++;
6020 }
6021
6022 /* write last descriptor with RS and EOP bits */
6023 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
6024 tx_desc->read.cmd_type_len = cmd_type;
6025
6026 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6027
6028 /* set the timestamp */
6029 first->time_stamp = jiffies;
6030
6031 /*
6032 * Force memory writes to complete before letting h/w know there
6033 * are new descriptors to fetch. (Only applicable for weak-ordered
6034 * memory model archs, such as IA-64).
6035 *
6036 * We also need this memory barrier to make certain all of the
6037 * status bits have been updated before next_to_watch is written.
6038 */
6039 wmb();
6040
6041 /* set next_to_watch value indicating a packet is present */
6042 first->next_to_watch = tx_desc;
6043
6044 i++;
6045 if (i == tx_ring->count)
6046 i = 0;
6047
6048 tx_ring->next_to_use = i;
6049
6050 /* notify HW of packet */
6051 writel(i, tx_ring->tail);
6052
6053 return;
6054 dma_error:
6055 dev_err(tx_ring->dev, "TX DMA map failed\n");
6056
6057 /* clear dma mappings for failed tx_buffer_info map */
6058 for (;;) {
6059 tx_buffer = &tx_ring->tx_buffer_info[i];
6060 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6061 if (tx_buffer == first)
6062 break;
6063 if (i == 0)
6064 i = tx_ring->count;
6065 i--;
6066 }
6067
6068 tx_ring->next_to_use = i;
6069 }
6070
6071 static void ixgbe_atr(struct ixgbe_ring *ring,
6072 struct ixgbe_tx_buffer *first)
6073 {
6074 struct ixgbe_q_vector *q_vector = ring->q_vector;
6075 union ixgbe_atr_hash_dword input = { .dword = 0 };
6076 union ixgbe_atr_hash_dword common = { .dword = 0 };
6077 union {
6078 unsigned char *network;
6079 struct iphdr *ipv4;
6080 struct ipv6hdr *ipv6;
6081 } hdr;
6082 struct tcphdr *th;
6083 __be16 vlan_id;
6084
6085 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6086 if (!q_vector)
6087 return;
6088
6089 /* do nothing if sampling is disabled */
6090 if (!ring->atr_sample_rate)
6091 return;
6092
6093 ring->atr_count++;
6094
6095 /* snag network header to get L4 type and address */
6096 hdr.network = skb_network_header(first->skb);
6097
6098 /* Currently only IPv4/IPv6 with TCP is supported */
6099 if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6100 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6101 (first->protocol != __constant_htons(ETH_P_IP) ||
6102 hdr.ipv4->protocol != IPPROTO_TCP))
6103 return;
6104
6105 th = tcp_hdr(first->skb);
6106
6107 /* skip this packet since it is invalid or the socket is closing */
6108 if (!th || th->fin)
6109 return;
6110
6111 /* sample on all syn packets or once every atr sample count */
6112 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6113 return;
6114
6115 /* reset sample count */
6116 ring->atr_count = 0;
6117
6118 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6119
6120 /*
6121 * src and dst are inverted, think how the receiver sees them
6122 *
6123 * The input is broken into two sections, a non-compressed section
6124 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6125 * is XORed together and stored in the compressed dword.
6126 */
6127 input.formatted.vlan_id = vlan_id;
6128
6129 /*
6130 * since src port and flex bytes occupy the same word XOR them together
6131 * and write the value to source port portion of compressed dword
6132 */
6133 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6134 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6135 else
6136 common.port.src ^= th->dest ^ first->protocol;
6137 common.port.dst ^= th->source;
6138
6139 if (first->protocol == __constant_htons(ETH_P_IP)) {
6140 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6141 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6142 } else {
6143 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6144 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6145 hdr.ipv6->saddr.s6_addr32[1] ^
6146 hdr.ipv6->saddr.s6_addr32[2] ^
6147 hdr.ipv6->saddr.s6_addr32[3] ^
6148 hdr.ipv6->daddr.s6_addr32[0] ^
6149 hdr.ipv6->daddr.s6_addr32[1] ^
6150 hdr.ipv6->daddr.s6_addr32[2] ^
6151 hdr.ipv6->daddr.s6_addr32[3];
6152 }
6153
6154 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6155 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6156 input, common, ring->queue_index);
6157 }
6158
6159 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6160 {
6161 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6162 /* Herbert's original patch had:
6163 * smp_mb__after_netif_stop_queue();
6164 * but since that doesn't exist yet, just open code it. */
6165 smp_mb();
6166
6167 /* We need to check again in a case another CPU has just
6168 * made room available. */
6169 if (likely(ixgbe_desc_unused(tx_ring) < size))
6170 return -EBUSY;
6171
6172 /* A reprieve! - use start_queue because it doesn't call schedule */
6173 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6174 ++tx_ring->tx_stats.restart_queue;
6175 return 0;
6176 }
6177
6178 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6179 {
6180 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6181 return 0;
6182 return __ixgbe_maybe_stop_tx(tx_ring, size);
6183 }
6184
6185 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6186 {
6187 struct ixgbe_adapter *adapter = netdev_priv(dev);
6188 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6189 smp_processor_id();
6190 #ifdef IXGBE_FCOE
6191 __be16 protocol = vlan_get_protocol(skb);
6192
6193 if (((protocol == htons(ETH_P_FCOE)) ||
6194 (protocol == htons(ETH_P_FIP))) &&
6195 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6196 struct ixgbe_ring_feature *f;
6197
6198 f = &adapter->ring_feature[RING_F_FCOE];
6199
6200 while (txq >= f->indices)
6201 txq -= f->indices;
6202 txq += adapter->ring_feature[RING_F_FCOE].offset;
6203
6204 return txq;
6205 }
6206 #endif
6207
6208 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6209 while (unlikely(txq >= dev->real_num_tx_queues))
6210 txq -= dev->real_num_tx_queues;
6211 return txq;
6212 }
6213
6214 return skb_tx_hash(dev, skb);
6215 }
6216
6217 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6218 struct ixgbe_adapter *adapter,
6219 struct ixgbe_ring *tx_ring)
6220 {
6221 struct ixgbe_tx_buffer *first;
6222 int tso;
6223 u32 tx_flags = 0;
6224 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6225 unsigned short f;
6226 #endif
6227 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6228 __be16 protocol = skb->protocol;
6229 u8 hdr_len = 0;
6230
6231 /*
6232 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6233 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6234 * + 2 desc gap to keep tail from touching head,
6235 * + 1 desc for context descriptor,
6236 * otherwise try next time
6237 */
6238 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6239 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6240 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6241 #else
6242 count += skb_shinfo(skb)->nr_frags;
6243 #endif
6244 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6245 tx_ring->tx_stats.tx_busy++;
6246 return NETDEV_TX_BUSY;
6247 }
6248
6249 /* record the location of the first descriptor for this packet */
6250 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6251 first->skb = skb;
6252 first->bytecount = skb->len;
6253 first->gso_segs = 1;
6254
6255 /* if we have a HW VLAN tag being added default to the HW one */
6256 if (vlan_tx_tag_present(skb)) {
6257 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6258 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6259 /* else if it is a SW VLAN check the next protocol and store the tag */
6260 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6261 struct vlan_hdr *vhdr, _vhdr;
6262 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6263 if (!vhdr)
6264 goto out_drop;
6265
6266 protocol = vhdr->h_vlan_encapsulated_proto;
6267 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6268 IXGBE_TX_FLAGS_VLAN_SHIFT;
6269 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6270 }
6271
6272 skb_tx_timestamp(skb);
6273
6274 #ifdef CONFIG_IXGBE_PTP
6275 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6276 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6277 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
6278 }
6279 #endif
6280
6281 #ifdef CONFIG_PCI_IOV
6282 /*
6283 * Use the l2switch_enable flag - would be false if the DMA
6284 * Tx switch had been disabled.
6285 */
6286 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6287 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6288
6289 #endif
6290 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6291 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6292 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6293 (skb->priority != TC_PRIO_CONTROL))) {
6294 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6295 tx_flags |= (skb->priority & 0x7) <<
6296 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6297 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6298 struct vlan_ethhdr *vhdr;
6299 if (skb_header_cloned(skb) &&
6300 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6301 goto out_drop;
6302 vhdr = (struct vlan_ethhdr *)skb->data;
6303 vhdr->h_vlan_TCI = htons(tx_flags >>
6304 IXGBE_TX_FLAGS_VLAN_SHIFT);
6305 } else {
6306 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6307 }
6308 }
6309
6310 /* record initial flags and protocol */
6311 first->tx_flags = tx_flags;
6312 first->protocol = protocol;
6313
6314 #ifdef IXGBE_FCOE
6315 /* setup tx offload for FCoE */
6316 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6317 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6318 tso = ixgbe_fso(tx_ring, first, &hdr_len);
6319 if (tso < 0)
6320 goto out_drop;
6321
6322 goto xmit_fcoe;
6323 }
6324
6325 #endif /* IXGBE_FCOE */
6326 tso = ixgbe_tso(tx_ring, first, &hdr_len);
6327 if (tso < 0)
6328 goto out_drop;
6329 else if (!tso)
6330 ixgbe_tx_csum(tx_ring, first);
6331
6332 /* add the ATR filter if ATR is on */
6333 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6334 ixgbe_atr(tx_ring, first);
6335
6336 #ifdef IXGBE_FCOE
6337 xmit_fcoe:
6338 #endif /* IXGBE_FCOE */
6339 ixgbe_tx_map(tx_ring, first, hdr_len);
6340
6341 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6342
6343 return NETDEV_TX_OK;
6344
6345 out_drop:
6346 dev_kfree_skb_any(first->skb);
6347 first->skb = NULL;
6348
6349 return NETDEV_TX_OK;
6350 }
6351
6352 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6353 struct net_device *netdev)
6354 {
6355 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6356 struct ixgbe_ring *tx_ring;
6357
6358 /*
6359 * The minimum packet size for olinfo paylen is 17 so pad the skb
6360 * in order to meet this minimum size requirement.
6361 */
6362 if (unlikely(skb->len < 17)) {
6363 if (skb_pad(skb, 17 - skb->len))
6364 return NETDEV_TX_OK;
6365 skb->len = 17;
6366 }
6367
6368 tx_ring = adapter->tx_ring[skb->queue_mapping];
6369 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6370 }
6371
6372 /**
6373 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6374 * @netdev: network interface device structure
6375 * @p: pointer to an address structure
6376 *
6377 * Returns 0 on success, negative on failure
6378 **/
6379 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6380 {
6381 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6382 struct ixgbe_hw *hw = &adapter->hw;
6383 struct sockaddr *addr = p;
6384
6385 if (!is_valid_ether_addr(addr->sa_data))
6386 return -EADDRNOTAVAIL;
6387
6388 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6389 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6390
6391 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6392 IXGBE_RAH_AV);
6393
6394 return 0;
6395 }
6396
6397 static int
6398 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6399 {
6400 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6401 struct ixgbe_hw *hw = &adapter->hw;
6402 u16 value;
6403 int rc;
6404
6405 if (prtad != hw->phy.mdio.prtad)
6406 return -EINVAL;
6407 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6408 if (!rc)
6409 rc = value;
6410 return rc;
6411 }
6412
6413 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6414 u16 addr, u16 value)
6415 {
6416 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6417 struct ixgbe_hw *hw = &adapter->hw;
6418
6419 if (prtad != hw->phy.mdio.prtad)
6420 return -EINVAL;
6421 return hw->phy.ops.write_reg(hw, addr, devad, value);
6422 }
6423
6424 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6425 {
6426 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6427
6428 switch (cmd) {
6429 #ifdef CONFIG_IXGBE_PTP
6430 case SIOCSHWTSTAMP:
6431 return ixgbe_ptp_hwtstamp_ioctl(adapter, req, cmd);
6432 #endif
6433 default:
6434 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6435 }
6436 }
6437
6438 /**
6439 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6440 * netdev->dev_addrs
6441 * @netdev: network interface device structure
6442 *
6443 * Returns non-zero on failure
6444 **/
6445 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6446 {
6447 int err = 0;
6448 struct ixgbe_adapter *adapter = netdev_priv(dev);
6449 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6450
6451 if (is_valid_ether_addr(mac->san_addr)) {
6452 rtnl_lock();
6453 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6454 rtnl_unlock();
6455 }
6456 return err;
6457 }
6458
6459 /**
6460 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6461 * netdev->dev_addrs
6462 * @netdev: network interface device structure
6463 *
6464 * Returns non-zero on failure
6465 **/
6466 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6467 {
6468 int err = 0;
6469 struct ixgbe_adapter *adapter = netdev_priv(dev);
6470 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6471
6472 if (is_valid_ether_addr(mac->san_addr)) {
6473 rtnl_lock();
6474 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6475 rtnl_unlock();
6476 }
6477 return err;
6478 }
6479
6480 #ifdef CONFIG_NET_POLL_CONTROLLER
6481 /*
6482 * Polling 'interrupt' - used by things like netconsole to send skbs
6483 * without having to re-enable interrupts. It's not called while
6484 * the interrupt routine is executing.
6485 */
6486 static void ixgbe_netpoll(struct net_device *netdev)
6487 {
6488 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6489 int i;
6490
6491 /* if interface is down do nothing */
6492 if (test_bit(__IXGBE_DOWN, &adapter->state))
6493 return;
6494
6495 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6496 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6497 for (i = 0; i < adapter->num_q_vectors; i++)
6498 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
6499 } else {
6500 ixgbe_intr(adapter->pdev->irq, netdev);
6501 }
6502 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6503 }
6504
6505 #endif
6506 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6507 struct rtnl_link_stats64 *stats)
6508 {
6509 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6510 int i;
6511
6512 rcu_read_lock();
6513 for (i = 0; i < adapter->num_rx_queues; i++) {
6514 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6515 u64 bytes, packets;
6516 unsigned int start;
6517
6518 if (ring) {
6519 do {
6520 start = u64_stats_fetch_begin_bh(&ring->syncp);
6521 packets = ring->stats.packets;
6522 bytes = ring->stats.bytes;
6523 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6524 stats->rx_packets += packets;
6525 stats->rx_bytes += bytes;
6526 }
6527 }
6528
6529 for (i = 0; i < adapter->num_tx_queues; i++) {
6530 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6531 u64 bytes, packets;
6532 unsigned int start;
6533
6534 if (ring) {
6535 do {
6536 start = u64_stats_fetch_begin_bh(&ring->syncp);
6537 packets = ring->stats.packets;
6538 bytes = ring->stats.bytes;
6539 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6540 stats->tx_packets += packets;
6541 stats->tx_bytes += bytes;
6542 }
6543 }
6544 rcu_read_unlock();
6545 /* following stats updated by ixgbe_watchdog_task() */
6546 stats->multicast = netdev->stats.multicast;
6547 stats->rx_errors = netdev->stats.rx_errors;
6548 stats->rx_length_errors = netdev->stats.rx_length_errors;
6549 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6550 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6551 return stats;
6552 }
6553
6554 #ifdef CONFIG_IXGBE_DCB
6555 /**
6556 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6557 * @adapter: pointer to ixgbe_adapter
6558 * @tc: number of traffic classes currently enabled
6559 *
6560 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6561 * 802.1Q priority maps to a packet buffer that exists.
6562 */
6563 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6564 {
6565 struct ixgbe_hw *hw = &adapter->hw;
6566 u32 reg, rsave;
6567 int i;
6568
6569 /* 82598 have a static priority to TC mapping that can not
6570 * be changed so no validation is needed.
6571 */
6572 if (hw->mac.type == ixgbe_mac_82598EB)
6573 return;
6574
6575 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6576 rsave = reg;
6577
6578 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6579 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6580
6581 /* If up2tc is out of bounds default to zero */
6582 if (up2tc > tc)
6583 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6584 }
6585
6586 if (reg != rsave)
6587 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6588
6589 return;
6590 }
6591
6592 /**
6593 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
6594 * @adapter: Pointer to adapter struct
6595 *
6596 * Populate the netdev user priority to tc map
6597 */
6598 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
6599 {
6600 struct net_device *dev = adapter->netdev;
6601 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
6602 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
6603 u8 prio;
6604
6605 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
6606 u8 tc = 0;
6607
6608 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
6609 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
6610 else if (ets)
6611 tc = ets->prio_tc[prio];
6612
6613 netdev_set_prio_tc_map(dev, prio, tc);
6614 }
6615 }
6616
6617 /**
6618 * ixgbe_setup_tc - configure net_device for multiple traffic classes
6619 *
6620 * @netdev: net device to configure
6621 * @tc: number of traffic classes to enable
6622 */
6623 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6624 {
6625 struct ixgbe_adapter *adapter = netdev_priv(dev);
6626 struct ixgbe_hw *hw = &adapter->hw;
6627
6628 /* Multiple traffic classes requires multiple queues */
6629 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6630 e_err(drv, "Enable failed, needs MSI-X\n");
6631 return -EINVAL;
6632 }
6633
6634 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6635 e_err(drv, "Enable failed, SR-IOV enabled\n");
6636 return -EINVAL;
6637 }
6638
6639 /* Hardware supports up to 8 traffic classes */
6640 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
6641 (hw->mac.type == ixgbe_mac_82598EB &&
6642 tc < MAX_TRAFFIC_CLASS))
6643 return -EINVAL;
6644
6645 /* Hardware has to reinitialize queues and interrupts to
6646 * match packet buffer alignment. Unfortunately, the
6647 * hardware is not flexible enough to do this dynamically.
6648 */
6649 if (netif_running(dev))
6650 ixgbe_close(dev);
6651 ixgbe_clear_interrupt_scheme(adapter);
6652
6653 if (tc) {
6654 netdev_set_num_tc(dev, tc);
6655 ixgbe_set_prio_tc_map(adapter);
6656
6657 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6658 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6659
6660 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
6661 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
6662 adapter->hw.fc.requested_mode = ixgbe_fc_none;
6663 }
6664 } else {
6665 netdev_reset_tc(dev);
6666
6667 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6668 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6669
6670 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6671 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6672
6673 adapter->temp_dcb_cfg.pfc_mode_enable = false;
6674 adapter->dcb_cfg.pfc_mode_enable = false;
6675 }
6676
6677 ixgbe_init_interrupt_scheme(adapter);
6678 ixgbe_validate_rtr(adapter, tc);
6679 if (netif_running(dev))
6680 ixgbe_open(dev);
6681
6682 return 0;
6683 }
6684
6685 #endif /* CONFIG_IXGBE_DCB */
6686 void ixgbe_do_reset(struct net_device *netdev)
6687 {
6688 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6689
6690 if (netif_running(netdev))
6691 ixgbe_reinit_locked(adapter);
6692 else
6693 ixgbe_reset(adapter);
6694 }
6695
6696 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
6697 netdev_features_t features)
6698 {
6699 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6700
6701 /* return error if RXHASH is being enabled when RSS is not supported */
6702 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
6703 features &= ~NETIF_F_RXHASH;
6704
6705 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6706 if (!(features & NETIF_F_RXCSUM))
6707 features &= ~NETIF_F_LRO;
6708
6709 /* Turn off LRO if not RSC capable */
6710 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
6711 features &= ~NETIF_F_LRO;
6712
6713 return features;
6714 }
6715
6716 static int ixgbe_set_features(struct net_device *netdev,
6717 netdev_features_t features)
6718 {
6719 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6720 netdev_features_t changed = netdev->features ^ features;
6721 bool need_reset = false;
6722
6723 /* Make sure RSC matches LRO, reset if change */
6724 if (!(features & NETIF_F_LRO)) {
6725 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6726 need_reset = true;
6727 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
6728 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
6729 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6730 if (adapter->rx_itr_setting == 1 ||
6731 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
6732 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
6733 need_reset = true;
6734 } else if ((changed ^ features) & NETIF_F_LRO) {
6735 e_info(probe, "rx-usecs set too low, "
6736 "disabling RSC\n");
6737 }
6738 }
6739
6740 /*
6741 * Check if Flow Director n-tuple support was enabled or disabled. If
6742 * the state changed, we need to reset.
6743 */
6744 if (!(features & NETIF_F_NTUPLE)) {
6745 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
6746 /* turn off Flow Director, set ATR and reset */
6747 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
6748 !(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
6749 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6750 need_reset = true;
6751 }
6752 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6753 } else if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
6754 /* turn off ATR, enable perfect filters and reset */
6755 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6756 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6757 need_reset = true;
6758 }
6759
6760 if (features & NETIF_F_HW_VLAN_RX)
6761 ixgbe_vlan_strip_enable(adapter);
6762 else
6763 ixgbe_vlan_strip_disable(adapter);
6764
6765 if (changed & NETIF_F_RXALL)
6766 need_reset = true;
6767
6768 netdev->features = features;
6769 if (need_reset)
6770 ixgbe_do_reset(netdev);
6771
6772 return 0;
6773 }
6774
6775 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm,
6776 struct net_device *dev,
6777 unsigned char *addr,
6778 u16 flags)
6779 {
6780 struct ixgbe_adapter *adapter = netdev_priv(dev);
6781 int err = -EOPNOTSUPP;
6782
6783 if (ndm->ndm_state & NUD_PERMANENT) {
6784 pr_info("%s: FDB only supports static addresses\n",
6785 ixgbe_driver_name);
6786 return -EINVAL;
6787 }
6788
6789 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6790 if (is_unicast_ether_addr(addr))
6791 err = dev_uc_add_excl(dev, addr);
6792 else if (is_multicast_ether_addr(addr))
6793 err = dev_mc_add_excl(dev, addr);
6794 else
6795 err = -EINVAL;
6796 }
6797
6798 /* Only return duplicate errors if NLM_F_EXCL is set */
6799 if (err == -EEXIST && !(flags & NLM_F_EXCL))
6800 err = 0;
6801
6802 return err;
6803 }
6804
6805 static int ixgbe_ndo_fdb_del(struct ndmsg *ndm,
6806 struct net_device *dev,
6807 unsigned char *addr)
6808 {
6809 struct ixgbe_adapter *adapter = netdev_priv(dev);
6810 int err = -EOPNOTSUPP;
6811
6812 if (ndm->ndm_state & NUD_PERMANENT) {
6813 pr_info("%s: FDB only supports static addresses\n",
6814 ixgbe_driver_name);
6815 return -EINVAL;
6816 }
6817
6818 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6819 if (is_unicast_ether_addr(addr))
6820 err = dev_uc_del(dev, addr);
6821 else if (is_multicast_ether_addr(addr))
6822 err = dev_mc_del(dev, addr);
6823 else
6824 err = -EINVAL;
6825 }
6826
6827 return err;
6828 }
6829
6830 static int ixgbe_ndo_fdb_dump(struct sk_buff *skb,
6831 struct netlink_callback *cb,
6832 struct net_device *dev,
6833 int idx)
6834 {
6835 struct ixgbe_adapter *adapter = netdev_priv(dev);
6836
6837 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6838 idx = ndo_dflt_fdb_dump(skb, cb, dev, idx);
6839
6840 return idx;
6841 }
6842
6843 static const struct net_device_ops ixgbe_netdev_ops = {
6844 .ndo_open = ixgbe_open,
6845 .ndo_stop = ixgbe_close,
6846 .ndo_start_xmit = ixgbe_xmit_frame,
6847 .ndo_select_queue = ixgbe_select_queue,
6848 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6849 .ndo_validate_addr = eth_validate_addr,
6850 .ndo_set_mac_address = ixgbe_set_mac,
6851 .ndo_change_mtu = ixgbe_change_mtu,
6852 .ndo_tx_timeout = ixgbe_tx_timeout,
6853 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6854 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6855 .ndo_do_ioctl = ixgbe_ioctl,
6856 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6857 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6858 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6859 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
6860 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6861 .ndo_get_stats64 = ixgbe_get_stats64,
6862 #ifdef CONFIG_IXGBE_DCB
6863 .ndo_setup_tc = ixgbe_setup_tc,
6864 #endif
6865 #ifdef CONFIG_NET_POLL_CONTROLLER
6866 .ndo_poll_controller = ixgbe_netpoll,
6867 #endif
6868 #ifdef IXGBE_FCOE
6869 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6870 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
6871 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6872 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6873 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6874 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6875 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
6876 #endif /* IXGBE_FCOE */
6877 .ndo_set_features = ixgbe_set_features,
6878 .ndo_fix_features = ixgbe_fix_features,
6879 .ndo_fdb_add = ixgbe_ndo_fdb_add,
6880 .ndo_fdb_del = ixgbe_ndo_fdb_del,
6881 .ndo_fdb_dump = ixgbe_ndo_fdb_dump,
6882 };
6883
6884 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6885 const struct ixgbe_info *ii)
6886 {
6887 #ifdef CONFIG_PCI_IOV
6888 struct ixgbe_hw *hw = &adapter->hw;
6889
6890 if (hw->mac.type == ixgbe_mac_82598EB)
6891 return;
6892
6893 /* The 82599 supports up to 64 VFs per physical function
6894 * but this implementation limits allocation to 63 so that
6895 * basic networking resources are still available to the
6896 * physical function. If the user requests greater thn
6897 * 63 VFs then it is an error - reset to default of zero.
6898 */
6899 adapter->num_vfs = (max_vfs > 63) ? 0 : max_vfs;
6900 ixgbe_enable_sriov(adapter, ii);
6901 #endif /* CONFIG_PCI_IOV */
6902 }
6903
6904 /**
6905 * ixgbe_wol_supported - Check whether device supports WoL
6906 * @hw: hw specific details
6907 * @device_id: the device ID
6908 * @subdev_id: the subsystem device ID
6909 *
6910 * This function is used by probe and ethtool to determine
6911 * which devices have WoL support
6912 *
6913 **/
6914 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
6915 u16 subdevice_id)
6916 {
6917 struct ixgbe_hw *hw = &adapter->hw;
6918 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
6919 int is_wol_supported = 0;
6920
6921 switch (device_id) {
6922 case IXGBE_DEV_ID_82599_SFP:
6923 /* Only these subdevices could supports WOL */
6924 switch (subdevice_id) {
6925 case IXGBE_SUBDEV_ID_82599_560FLR:
6926 /* only support first port */
6927 if (hw->bus.func != 0)
6928 break;
6929 case IXGBE_SUBDEV_ID_82599_SFP:
6930 is_wol_supported = 1;
6931 break;
6932 }
6933 break;
6934 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
6935 /* All except this subdevice support WOL */
6936 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
6937 is_wol_supported = 1;
6938 break;
6939 case IXGBE_DEV_ID_82599_KX4:
6940 is_wol_supported = 1;
6941 break;
6942 case IXGBE_DEV_ID_X540T:
6943 /* check eeprom to see if enabled wol */
6944 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
6945 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
6946 (hw->bus.func == 0))) {
6947 is_wol_supported = 1;
6948 }
6949 break;
6950 }
6951
6952 return is_wol_supported;
6953 }
6954
6955 /**
6956 * ixgbe_probe - Device Initialization Routine
6957 * @pdev: PCI device information struct
6958 * @ent: entry in ixgbe_pci_tbl
6959 *
6960 * Returns 0 on success, negative on failure
6961 *
6962 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6963 * The OS initialization, configuring of the adapter private structure,
6964 * and a hardware reset occur.
6965 **/
6966 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6967 const struct pci_device_id *ent)
6968 {
6969 struct net_device *netdev;
6970 struct ixgbe_adapter *adapter = NULL;
6971 struct ixgbe_hw *hw;
6972 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6973 static int cards_found;
6974 int i, err, pci_using_dac;
6975 u8 part_str[IXGBE_PBANUM_LENGTH];
6976 unsigned int indices = num_possible_cpus();
6977 #ifdef IXGBE_FCOE
6978 u16 device_caps;
6979 #endif
6980 u32 eec;
6981
6982 /* Catch broken hardware that put the wrong VF device ID in
6983 * the PCIe SR-IOV capability.
6984 */
6985 if (pdev->is_virtfn) {
6986 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6987 pci_name(pdev), pdev->vendor, pdev->device);
6988 return -EINVAL;
6989 }
6990
6991 err = pci_enable_device_mem(pdev);
6992 if (err)
6993 return err;
6994
6995 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6996 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6997 pci_using_dac = 1;
6998 } else {
6999 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7000 if (err) {
7001 err = dma_set_coherent_mask(&pdev->dev,
7002 DMA_BIT_MASK(32));
7003 if (err) {
7004 dev_err(&pdev->dev,
7005 "No usable DMA configuration, aborting\n");
7006 goto err_dma;
7007 }
7008 }
7009 pci_using_dac = 0;
7010 }
7011
7012 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7013 IORESOURCE_MEM), ixgbe_driver_name);
7014 if (err) {
7015 dev_err(&pdev->dev,
7016 "pci_request_selected_regions failed 0x%x\n", err);
7017 goto err_pci_reg;
7018 }
7019
7020 pci_enable_pcie_error_reporting(pdev);
7021
7022 pci_set_master(pdev);
7023 pci_save_state(pdev);
7024
7025 #ifdef CONFIG_IXGBE_DCB
7026 indices *= MAX_TRAFFIC_CLASS;
7027 #endif
7028
7029 if (ii->mac == ixgbe_mac_82598EB)
7030 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7031 else
7032 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7033
7034 #ifdef IXGBE_FCOE
7035 indices += min_t(unsigned int, num_possible_cpus(),
7036 IXGBE_MAX_FCOE_INDICES);
7037 #endif
7038 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7039 if (!netdev) {
7040 err = -ENOMEM;
7041 goto err_alloc_etherdev;
7042 }
7043
7044 SET_NETDEV_DEV(netdev, &pdev->dev);
7045
7046 adapter = netdev_priv(netdev);
7047 pci_set_drvdata(pdev, adapter);
7048
7049 adapter->netdev = netdev;
7050 adapter->pdev = pdev;
7051 hw = &adapter->hw;
7052 hw->back = adapter;
7053 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7054
7055 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7056 pci_resource_len(pdev, 0));
7057 if (!hw->hw_addr) {
7058 err = -EIO;
7059 goto err_ioremap;
7060 }
7061
7062 for (i = 1; i <= 5; i++) {
7063 if (pci_resource_len(pdev, i) == 0)
7064 continue;
7065 }
7066
7067 netdev->netdev_ops = &ixgbe_netdev_ops;
7068 ixgbe_set_ethtool_ops(netdev);
7069 netdev->watchdog_timeo = 5 * HZ;
7070 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7071
7072 adapter->bd_number = cards_found;
7073
7074 /* Setup hw api */
7075 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7076 hw->mac.type = ii->mac;
7077
7078 /* EEPROM */
7079 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7080 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7081 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7082 if (!(eec & (1 << 8)))
7083 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7084
7085 /* PHY */
7086 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7087 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7088 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7089 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7090 hw->phy.mdio.mmds = 0;
7091 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7092 hw->phy.mdio.dev = netdev;
7093 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7094 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7095
7096 ii->get_invariants(hw);
7097
7098 /* setup the private structure */
7099 err = ixgbe_sw_init(adapter);
7100 if (err)
7101 goto err_sw_init;
7102
7103 /* Make it possible the adapter to be woken up via WOL */
7104 switch (adapter->hw.mac.type) {
7105 case ixgbe_mac_82599EB:
7106 case ixgbe_mac_X540:
7107 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7108 break;
7109 default:
7110 break;
7111 }
7112
7113 /*
7114 * If there is a fan on this device and it has failed log the
7115 * failure.
7116 */
7117 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7118 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7119 if (esdp & IXGBE_ESDP_SDP1)
7120 e_crit(probe, "Fan has stopped, replace the adapter\n");
7121 }
7122
7123 if (allow_unsupported_sfp)
7124 hw->allow_unsupported_sfp = allow_unsupported_sfp;
7125
7126 /* reset_hw fills in the perm_addr as well */
7127 hw->phy.reset_if_overtemp = true;
7128 err = hw->mac.ops.reset_hw(hw);
7129 hw->phy.reset_if_overtemp = false;
7130 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7131 hw->mac.type == ixgbe_mac_82598EB) {
7132 err = 0;
7133 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7134 e_dev_err("failed to load because an unsupported SFP+ "
7135 "module type was detected.\n");
7136 e_dev_err("Reload the driver after installing a supported "
7137 "module.\n");
7138 goto err_sw_init;
7139 } else if (err) {
7140 e_dev_err("HW Init failed: %d\n", err);
7141 goto err_sw_init;
7142 }
7143
7144 ixgbe_probe_vf(adapter, ii);
7145
7146 netdev->features = NETIF_F_SG |
7147 NETIF_F_IP_CSUM |
7148 NETIF_F_IPV6_CSUM |
7149 NETIF_F_HW_VLAN_TX |
7150 NETIF_F_HW_VLAN_RX |
7151 NETIF_F_HW_VLAN_FILTER |
7152 NETIF_F_TSO |
7153 NETIF_F_TSO6 |
7154 NETIF_F_RXHASH |
7155 NETIF_F_RXCSUM;
7156
7157 netdev->hw_features = netdev->features;
7158
7159 switch (adapter->hw.mac.type) {
7160 case ixgbe_mac_82599EB:
7161 case ixgbe_mac_X540:
7162 netdev->features |= NETIF_F_SCTP_CSUM;
7163 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7164 NETIF_F_NTUPLE;
7165 break;
7166 default:
7167 break;
7168 }
7169
7170 netdev->hw_features |= NETIF_F_RXALL;
7171
7172 netdev->vlan_features |= NETIF_F_TSO;
7173 netdev->vlan_features |= NETIF_F_TSO6;
7174 netdev->vlan_features |= NETIF_F_IP_CSUM;
7175 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7176 netdev->vlan_features |= NETIF_F_SG;
7177
7178 netdev->priv_flags |= IFF_UNICAST_FLT;
7179 netdev->priv_flags |= IFF_SUPP_NOFCS;
7180
7181 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7182 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7183 IXGBE_FLAG_DCB_ENABLED);
7184
7185 #ifdef CONFIG_IXGBE_DCB
7186 netdev->dcbnl_ops = &dcbnl_ops;
7187 #endif
7188
7189 #ifdef IXGBE_FCOE
7190 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7191 if (hw->mac.ops.get_device_caps) {
7192 hw->mac.ops.get_device_caps(hw, &device_caps);
7193 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7194 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7195 }
7196 }
7197 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7198 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7199 netdev->vlan_features |= NETIF_F_FSO;
7200 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7201 }
7202 #endif /* IXGBE_FCOE */
7203 if (pci_using_dac) {
7204 netdev->features |= NETIF_F_HIGHDMA;
7205 netdev->vlan_features |= NETIF_F_HIGHDMA;
7206 }
7207
7208 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7209 netdev->hw_features |= NETIF_F_LRO;
7210 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7211 netdev->features |= NETIF_F_LRO;
7212
7213 /* make sure the EEPROM is good */
7214 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7215 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7216 err = -EIO;
7217 goto err_sw_init;
7218 }
7219
7220 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7221 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7222
7223 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7224 e_dev_err("invalid MAC address\n");
7225 err = -EIO;
7226 goto err_sw_init;
7227 }
7228
7229 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7230 (unsigned long) adapter);
7231
7232 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7233 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7234
7235 err = ixgbe_init_interrupt_scheme(adapter);
7236 if (err)
7237 goto err_sw_init;
7238
7239 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7240 netdev->hw_features &= ~NETIF_F_RXHASH;
7241 netdev->features &= ~NETIF_F_RXHASH;
7242 }
7243
7244 /* WOL not supported for all devices */
7245 adapter->wol = 0;
7246 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7247 if (ixgbe_wol_supported(adapter, pdev->device, pdev->subsystem_device))
7248 adapter->wol = IXGBE_WUFC_MAG;
7249
7250 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7251
7252 #ifdef CONFIG_IXGBE_PTP
7253 ixgbe_ptp_init(adapter);
7254 #endif /* CONFIG_IXGBE_PTP*/
7255
7256 /* save off EEPROM version number */
7257 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7258 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7259
7260 /* pick up the PCI bus settings for reporting later */
7261 hw->mac.ops.get_bus_info(hw);
7262
7263 /* print bus type/speed/width info */
7264 e_dev_info("(PCI Express:%s:%s) %pM\n",
7265 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7266 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7267 "Unknown"),
7268 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7269 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7270 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7271 "Unknown"),
7272 netdev->dev_addr);
7273
7274 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7275 if (err)
7276 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7277 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7278 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7279 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7280 part_str);
7281 else
7282 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7283 hw->mac.type, hw->phy.type, part_str);
7284
7285 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7286 e_dev_warn("PCI-Express bandwidth available for this card is "
7287 "not sufficient for optimal performance.\n");
7288 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7289 "is required.\n");
7290 }
7291
7292 /* reset the hardware with the new settings */
7293 err = hw->mac.ops.start_hw(hw);
7294 if (err == IXGBE_ERR_EEPROM_VERSION) {
7295 /* We are running on a pre-production device, log a warning */
7296 e_dev_warn("This device is a pre-production adapter/LOM. "
7297 "Please be aware there may be issues associated "
7298 "with your hardware. If you are experiencing "
7299 "problems please contact your Intel or hardware "
7300 "representative who provided you with this "
7301 "hardware.\n");
7302 }
7303 strcpy(netdev->name, "eth%d");
7304 err = register_netdev(netdev);
7305 if (err)
7306 goto err_register;
7307
7308 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7309 if (hw->mac.ops.disable_tx_laser &&
7310 ((hw->phy.multispeed_fiber) ||
7311 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7312 (hw->mac.type == ixgbe_mac_82599EB))))
7313 hw->mac.ops.disable_tx_laser(hw);
7314
7315 /* carrier off reporting is important to ethtool even BEFORE open */
7316 netif_carrier_off(netdev);
7317
7318 #ifdef CONFIG_IXGBE_DCA
7319 if (dca_add_requester(&pdev->dev) == 0) {
7320 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7321 ixgbe_setup_dca(adapter);
7322 }
7323 #endif
7324 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7325 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7326 for (i = 0; i < adapter->num_vfs; i++)
7327 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7328 }
7329
7330 /* firmware requires driver version to be 0xFFFFFFFF
7331 * since os does not support feature
7332 */
7333 if (hw->mac.ops.set_fw_drv_ver)
7334 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7335 0xFF);
7336
7337 /* add san mac addr to netdev */
7338 ixgbe_add_sanmac_netdev(netdev);
7339
7340 e_dev_info("%s\n", ixgbe_default_device_descr);
7341 cards_found++;
7342
7343 #ifdef CONFIG_IXGBE_HWMON
7344 if (ixgbe_sysfs_init(adapter))
7345 e_err(probe, "failed to allocate sysfs resources\n");
7346 #endif /* CONFIG_IXGBE_HWMON */
7347
7348 return 0;
7349
7350 err_register:
7351 ixgbe_release_hw_control(adapter);
7352 ixgbe_clear_interrupt_scheme(adapter);
7353 err_sw_init:
7354 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7355 ixgbe_disable_sriov(adapter);
7356 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7357 iounmap(hw->hw_addr);
7358 err_ioremap:
7359 free_netdev(netdev);
7360 err_alloc_etherdev:
7361 pci_release_selected_regions(pdev,
7362 pci_select_bars(pdev, IORESOURCE_MEM));
7363 err_pci_reg:
7364 err_dma:
7365 pci_disable_device(pdev);
7366 return err;
7367 }
7368
7369 /**
7370 * ixgbe_remove - Device Removal Routine
7371 * @pdev: PCI device information struct
7372 *
7373 * ixgbe_remove is called by the PCI subsystem to alert the driver
7374 * that it should release a PCI device. The could be caused by a
7375 * Hot-Plug event, or because the driver is going to be removed from
7376 * memory.
7377 **/
7378 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7379 {
7380 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7381 struct net_device *netdev = adapter->netdev;
7382
7383 set_bit(__IXGBE_DOWN, &adapter->state);
7384 cancel_work_sync(&adapter->service_task);
7385
7386 #ifdef CONFIG_IXGBE_PTP
7387 ixgbe_ptp_stop(adapter);
7388 #endif
7389
7390 #ifdef CONFIG_IXGBE_DCA
7391 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7392 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7393 dca_remove_requester(&pdev->dev);
7394 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7395 }
7396
7397 #endif
7398 #ifdef CONFIG_IXGBE_HWMON
7399 ixgbe_sysfs_exit(adapter);
7400 #endif /* CONFIG_IXGBE_HWMON */
7401
7402 #ifdef IXGBE_FCOE
7403 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7404 ixgbe_cleanup_fcoe(adapter);
7405
7406 #endif /* IXGBE_FCOE */
7407
7408 /* remove the added san mac */
7409 ixgbe_del_sanmac_netdev(netdev);
7410
7411 if (netdev->reg_state == NETREG_REGISTERED)
7412 unregister_netdev(netdev);
7413
7414 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7415 if (!(ixgbe_check_vf_assignment(adapter)))
7416 ixgbe_disable_sriov(adapter);
7417 else
7418 e_dev_warn("Unloading driver while VFs are assigned "
7419 "- VFs will not be deallocated\n");
7420 }
7421
7422 ixgbe_clear_interrupt_scheme(adapter);
7423
7424 ixgbe_release_hw_control(adapter);
7425
7426 #ifdef CONFIG_DCB
7427 kfree(adapter->ixgbe_ieee_pfc);
7428 kfree(adapter->ixgbe_ieee_ets);
7429
7430 #endif
7431 iounmap(adapter->hw.hw_addr);
7432 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7433 IORESOURCE_MEM));
7434
7435 e_dev_info("complete\n");
7436
7437 free_netdev(netdev);
7438
7439 pci_disable_pcie_error_reporting(pdev);
7440
7441 pci_disable_device(pdev);
7442 }
7443
7444 /**
7445 * ixgbe_io_error_detected - called when PCI error is detected
7446 * @pdev: Pointer to PCI device
7447 * @state: The current pci connection state
7448 *
7449 * This function is called after a PCI bus error affecting
7450 * this device has been detected.
7451 */
7452 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7453 pci_channel_state_t state)
7454 {
7455 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7456 struct net_device *netdev = adapter->netdev;
7457
7458 #ifdef CONFIG_PCI_IOV
7459 struct pci_dev *bdev, *vfdev;
7460 u32 dw0, dw1, dw2, dw3;
7461 int vf, pos;
7462 u16 req_id, pf_func;
7463
7464 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7465 adapter->num_vfs == 0)
7466 goto skip_bad_vf_detection;
7467
7468 bdev = pdev->bus->self;
7469 while (bdev && (bdev->pcie_type != PCI_EXP_TYPE_ROOT_PORT))
7470 bdev = bdev->bus->self;
7471
7472 if (!bdev)
7473 goto skip_bad_vf_detection;
7474
7475 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7476 if (!pos)
7477 goto skip_bad_vf_detection;
7478
7479 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7480 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7481 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7482 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7483
7484 req_id = dw1 >> 16;
7485 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7486 if (!(req_id & 0x0080))
7487 goto skip_bad_vf_detection;
7488
7489 pf_func = req_id & 0x01;
7490 if ((pf_func & 1) == (pdev->devfn & 1)) {
7491 unsigned int device_id;
7492
7493 vf = (req_id & 0x7F) >> 1;
7494 e_dev_err("VF %d has caused a PCIe error\n", vf);
7495 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7496 "%8.8x\tdw3: %8.8x\n",
7497 dw0, dw1, dw2, dw3);
7498 switch (adapter->hw.mac.type) {
7499 case ixgbe_mac_82599EB:
7500 device_id = IXGBE_82599_VF_DEVICE_ID;
7501 break;
7502 case ixgbe_mac_X540:
7503 device_id = IXGBE_X540_VF_DEVICE_ID;
7504 break;
7505 default:
7506 device_id = 0;
7507 break;
7508 }
7509
7510 /* Find the pci device of the offending VF */
7511 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID, device_id, NULL);
7512 while (vfdev) {
7513 if (vfdev->devfn == (req_id & 0xFF))
7514 break;
7515 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID,
7516 device_id, vfdev);
7517 }
7518 /*
7519 * There's a slim chance the VF could have been hot plugged,
7520 * so if it is no longer present we don't need to issue the
7521 * VFLR. Just clean up the AER in that case.
7522 */
7523 if (vfdev) {
7524 e_dev_err("Issuing VFLR to VF %d\n", vf);
7525 pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7526 }
7527
7528 pci_cleanup_aer_uncorrect_error_status(pdev);
7529 }
7530
7531 /*
7532 * Even though the error may have occurred on the other port
7533 * we still need to increment the vf error reference count for
7534 * both ports because the I/O resume function will be called
7535 * for both of them.
7536 */
7537 adapter->vferr_refcount++;
7538
7539 return PCI_ERS_RESULT_RECOVERED;
7540
7541 skip_bad_vf_detection:
7542 #endif /* CONFIG_PCI_IOV */
7543 netif_device_detach(netdev);
7544
7545 if (state == pci_channel_io_perm_failure)
7546 return PCI_ERS_RESULT_DISCONNECT;
7547
7548 if (netif_running(netdev))
7549 ixgbe_down(adapter);
7550 pci_disable_device(pdev);
7551
7552 /* Request a slot reset. */
7553 return PCI_ERS_RESULT_NEED_RESET;
7554 }
7555
7556 /**
7557 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7558 * @pdev: Pointer to PCI device
7559 *
7560 * Restart the card from scratch, as if from a cold-boot.
7561 */
7562 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7563 {
7564 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7565 pci_ers_result_t result;
7566 int err;
7567
7568 if (pci_enable_device_mem(pdev)) {
7569 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7570 result = PCI_ERS_RESULT_DISCONNECT;
7571 } else {
7572 pci_set_master(pdev);
7573 pci_restore_state(pdev);
7574 pci_save_state(pdev);
7575
7576 pci_wake_from_d3(pdev, false);
7577
7578 ixgbe_reset(adapter);
7579 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7580 result = PCI_ERS_RESULT_RECOVERED;
7581 }
7582
7583 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7584 if (err) {
7585 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7586 "failed 0x%0x\n", err);
7587 /* non-fatal, continue */
7588 }
7589
7590 return result;
7591 }
7592
7593 /**
7594 * ixgbe_io_resume - called when traffic can start flowing again.
7595 * @pdev: Pointer to PCI device
7596 *
7597 * This callback is called when the error recovery driver tells us that
7598 * its OK to resume normal operation.
7599 */
7600 static void ixgbe_io_resume(struct pci_dev *pdev)
7601 {
7602 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7603 struct net_device *netdev = adapter->netdev;
7604
7605 #ifdef CONFIG_PCI_IOV
7606 if (adapter->vferr_refcount) {
7607 e_info(drv, "Resuming after VF err\n");
7608 adapter->vferr_refcount--;
7609 return;
7610 }
7611
7612 #endif
7613 if (netif_running(netdev))
7614 ixgbe_up(adapter);
7615
7616 netif_device_attach(netdev);
7617 }
7618
7619 static struct pci_error_handlers ixgbe_err_handler = {
7620 .error_detected = ixgbe_io_error_detected,
7621 .slot_reset = ixgbe_io_slot_reset,
7622 .resume = ixgbe_io_resume,
7623 };
7624
7625 static struct pci_driver ixgbe_driver = {
7626 .name = ixgbe_driver_name,
7627 .id_table = ixgbe_pci_tbl,
7628 .probe = ixgbe_probe,
7629 .remove = __devexit_p(ixgbe_remove),
7630 #ifdef CONFIG_PM
7631 .suspend = ixgbe_suspend,
7632 .resume = ixgbe_resume,
7633 #endif
7634 .shutdown = ixgbe_shutdown,
7635 .err_handler = &ixgbe_err_handler
7636 };
7637
7638 /**
7639 * ixgbe_init_module - Driver Registration Routine
7640 *
7641 * ixgbe_init_module is the first routine called when the driver is
7642 * loaded. All it does is register with the PCI subsystem.
7643 **/
7644 static int __init ixgbe_init_module(void)
7645 {
7646 int ret;
7647 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7648 pr_info("%s\n", ixgbe_copyright);
7649
7650 #ifdef CONFIG_IXGBE_DCA
7651 dca_register_notify(&dca_notifier);
7652 #endif
7653
7654 ret = pci_register_driver(&ixgbe_driver);
7655 return ret;
7656 }
7657
7658 module_init(ixgbe_init_module);
7659
7660 /**
7661 * ixgbe_exit_module - Driver Exit Cleanup Routine
7662 *
7663 * ixgbe_exit_module is called just before the driver is removed
7664 * from memory.
7665 **/
7666 static void __exit ixgbe_exit_module(void)
7667 {
7668 #ifdef CONFIG_IXGBE_DCA
7669 dca_unregister_notify(&dca_notifier);
7670 #endif
7671 pci_unregister_driver(&ixgbe_driver);
7672 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7673 }
7674
7675 #ifdef CONFIG_IXGBE_DCA
7676 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7677 void *p)
7678 {
7679 int ret_val;
7680
7681 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7682 __ixgbe_notify_dca);
7683
7684 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7685 }
7686
7687 #endif /* CONFIG_IXGBE_DCA */
7688
7689 module_exit(ixgbe_exit_module);
7690
7691 /* ixgbe_main.c */
Linux kernel - Deliverables
This page took 0.457815 seconds and 5 git commands to generate.