projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ixgbe: Populate the prio_tc_map in ixgbe_setup_tc
[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 int fcoe_pb = 0;
3651
3652 #ifdef CONFIG_IXGBE_DCB
3653 fcoe_pb = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
3654
3655 #endif
3656 if (fcoe_pb == pb && tc < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3657 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3658 }
3659 #endif
3660
3661 /* Calculate delay value for device */
3662 switch (hw->mac.type) {
3663 case ixgbe_mac_X540:
3664 dv_id = IXGBE_DV_X540(link, tc);
3665 break;
3666 default:
3667 dv_id = IXGBE_DV(link, tc);
3668 break;
3669 }
3670
3671 /* Loopback switch introduces additional latency */
3672 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3673 dv_id += IXGBE_B2BT(tc);
3674
3675 /* Delay value is calculated in bit times convert to KB */
3676 kb = IXGBE_BT2KB(dv_id);
3677 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3678
3679 marker = rx_pba - kb;
3680
3681 /* It is possible that the packet buffer is not large enough
3682 * to provide required headroom. In this case throw an error
3683 * to user and a do the best we can.
3684 */
3685 if (marker < 0) {
3686 e_warn(drv, "Packet Buffer(%i) can not provide enough"
3687 "headroom to support flow control."
3688 "Decrease MTU or number of traffic classes\n", pb);
3689 marker = tc + 1;
3690 }
3691
3692 return marker;
3693 }
3694
3695 /**
3696 * ixgbe_lpbthresh - calculate low water mark for for flow control
3697 *
3698 * @adapter: board private structure to calculate for
3699 * @pb: packet buffer to calculate
3700 */
3701 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3702 {
3703 struct ixgbe_hw *hw = &adapter->hw;
3704 struct net_device *dev = adapter->netdev;
3705 int tc;
3706 u32 dv_id;
3707
3708 /* Calculate max LAN frame size */
3709 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3710
3711 /* Calculate delay value for device */
3712 switch (hw->mac.type) {
3713 case ixgbe_mac_X540:
3714 dv_id = IXGBE_LOW_DV_X540(tc);
3715 break;
3716 default:
3717 dv_id = IXGBE_LOW_DV(tc);
3718 break;
3719 }
3720
3721 /* Delay value is calculated in bit times convert to KB */
3722 return IXGBE_BT2KB(dv_id);
3723 }
3724
3725 /*
3726 * ixgbe_pbthresh_setup - calculate and setup high low water marks
3727 */
3728 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3729 {
3730 struct ixgbe_hw *hw = &adapter->hw;
3731 int num_tc = netdev_get_num_tc(adapter->netdev);
3732 int i;
3733
3734 if (!num_tc)
3735 num_tc = 1;
3736
3737 hw->fc.low_water = ixgbe_lpbthresh(adapter);
3738
3739 for (i = 0; i < num_tc; i++) {
3740 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3741
3742 /* Low water marks must not be larger than high water marks */
3743 if (hw->fc.low_water > hw->fc.high_water[i])
3744 hw->fc.low_water = 0;
3745 }
3746 }
3747
3748 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3749 {
3750 struct ixgbe_hw *hw = &adapter->hw;
3751 int hdrm;
3752 u8 tc = netdev_get_num_tc(adapter->netdev);
3753
3754 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3755 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3756 hdrm = 32 << adapter->fdir_pballoc;
3757 else
3758 hdrm = 0;
3759
3760 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3761 ixgbe_pbthresh_setup(adapter);
3762 }
3763
3764 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3765 {
3766 struct ixgbe_hw *hw = &adapter->hw;
3767 struct hlist_node *node, *node2;
3768 struct ixgbe_fdir_filter *filter;
3769
3770 spin_lock(&adapter->fdir_perfect_lock);
3771
3772 if (!hlist_empty(&adapter->fdir_filter_list))
3773 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3774
3775 hlist_for_each_entry_safe(filter, node, node2,
3776 &adapter->fdir_filter_list, fdir_node) {
3777 ixgbe_fdir_write_perfect_filter_82599(hw,
3778 &filter->filter,
3779 filter->sw_idx,
3780 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3781 IXGBE_FDIR_DROP_QUEUE :
3782 adapter->rx_ring[filter->action]->reg_idx);
3783 }
3784
3785 spin_unlock(&adapter->fdir_perfect_lock);
3786 }
3787
3788 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3789 {
3790 struct ixgbe_hw *hw = &adapter->hw;
3791
3792 ixgbe_configure_pb(adapter);
3793 #ifdef CONFIG_IXGBE_DCB
3794 ixgbe_configure_dcb(adapter);
3795 #endif
3796
3797 ixgbe_set_rx_mode(adapter->netdev);
3798 ixgbe_restore_vlan(adapter);
3799
3800 #ifdef IXGBE_FCOE
3801 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3802 ixgbe_configure_fcoe(adapter);
3803
3804 #endif /* IXGBE_FCOE */
3805
3806 switch (hw->mac.type) {
3807 case ixgbe_mac_82599EB:
3808 case ixgbe_mac_X540:
3809 hw->mac.ops.disable_rx_buff(hw);
3810 break;
3811 default:
3812 break;
3813 }
3814
3815 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3816 ixgbe_init_fdir_signature_82599(&adapter->hw,
3817 adapter->fdir_pballoc);
3818 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3819 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3820 adapter->fdir_pballoc);
3821 ixgbe_fdir_filter_restore(adapter);
3822 }
3823
3824 switch (hw->mac.type) {
3825 case ixgbe_mac_82599EB:
3826 case ixgbe_mac_X540:
3827 hw->mac.ops.enable_rx_buff(hw);
3828 break;
3829 default:
3830 break;
3831 }
3832
3833 ixgbe_configure_virtualization(adapter);
3834
3835 ixgbe_configure_tx(adapter);
3836 ixgbe_configure_rx(adapter);
3837 }
3838
3839 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3840 {
3841 switch (hw->phy.type) {
3842 case ixgbe_phy_sfp_avago:
3843 case ixgbe_phy_sfp_ftl:
3844 case ixgbe_phy_sfp_intel:
3845 case ixgbe_phy_sfp_unknown:
3846 case ixgbe_phy_sfp_passive_tyco:
3847 case ixgbe_phy_sfp_passive_unknown:
3848 case ixgbe_phy_sfp_active_unknown:
3849 case ixgbe_phy_sfp_ftl_active:
3850 return true;
3851 case ixgbe_phy_nl:
3852 if (hw->mac.type == ixgbe_mac_82598EB)
3853 return true;
3854 default:
3855 return false;
3856 }
3857 }
3858
3859 /**
3860 * ixgbe_sfp_link_config - set up SFP+ link
3861 * @adapter: pointer to private adapter struct
3862 **/
3863 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3864 {
3865 /*
3866 * We are assuming the worst case scenario here, and that
3867 * is that an SFP was inserted/removed after the reset
3868 * but before SFP detection was enabled. As such the best
3869 * solution is to just start searching as soon as we start
3870 */
3871 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3872 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3873
3874 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3875 }
3876
3877 /**
3878 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3879 * @hw: pointer to private hardware struct
3880 *
3881 * Returns 0 on success, negative on failure
3882 **/
3883 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3884 {
3885 u32 autoneg;
3886 bool negotiation, link_up = false;
3887 u32 ret = IXGBE_ERR_LINK_SETUP;
3888
3889 if (hw->mac.ops.check_link)
3890 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3891
3892 if (ret)
3893 goto link_cfg_out;
3894
3895 autoneg = hw->phy.autoneg_advertised;
3896 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3897 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3898 &negotiation);
3899 if (ret)
3900 goto link_cfg_out;
3901
3902 if (hw->mac.ops.setup_link)
3903 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3904 link_cfg_out:
3905 return ret;
3906 }
3907
3908 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3909 {
3910 struct ixgbe_hw *hw = &adapter->hw;
3911 u32 gpie = 0;
3912
3913 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3914 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3915 IXGBE_GPIE_OCD;
3916 gpie |= IXGBE_GPIE_EIAME;
3917 /*
3918 * use EIAM to auto-mask when MSI-X interrupt is asserted
3919 * this saves a register write for every interrupt
3920 */
3921 switch (hw->mac.type) {
3922 case ixgbe_mac_82598EB:
3923 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3924 break;
3925 case ixgbe_mac_82599EB:
3926 case ixgbe_mac_X540:
3927 default:
3928 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3929 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3930 break;
3931 }
3932 } else {
3933 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3934 * specifically only auto mask tx and rx interrupts */
3935 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3936 }
3937
3938 /* XXX: to interrupt immediately for EICS writes, enable this */
3939 /* gpie |= IXGBE_GPIE_EIMEN; */
3940
3941 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3942 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3943 gpie |= IXGBE_GPIE_VTMODE_64;
3944 }
3945
3946 /* Enable Thermal over heat sensor interrupt */
3947 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3948 switch (adapter->hw.mac.type) {
3949 case ixgbe_mac_82599EB:
3950 gpie |= IXGBE_SDP0_GPIEN;
3951 break;
3952 case ixgbe_mac_X540:
3953 gpie |= IXGBE_EIMS_TS;
3954 break;
3955 default:
3956 break;
3957 }
3958 }
3959
3960 /* Enable fan failure interrupt */
3961 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3962 gpie |= IXGBE_SDP1_GPIEN;
3963
3964 if (hw->mac.type == ixgbe_mac_82599EB) {
3965 gpie |= IXGBE_SDP1_GPIEN;
3966 gpie |= IXGBE_SDP2_GPIEN;
3967 }
3968
3969 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3970 }
3971
3972 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
3973 {
3974 struct ixgbe_hw *hw = &adapter->hw;
3975 int err;
3976 u32 ctrl_ext;
3977
3978 ixgbe_get_hw_control(adapter);
3979 ixgbe_setup_gpie(adapter);
3980
3981 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3982 ixgbe_configure_msix(adapter);
3983 else
3984 ixgbe_configure_msi_and_legacy(adapter);
3985
3986 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3987 if (hw->mac.ops.enable_tx_laser &&
3988 ((hw->phy.multispeed_fiber) ||
3989 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3990 (hw->mac.type == ixgbe_mac_82599EB))))
3991 hw->mac.ops.enable_tx_laser(hw);
3992
3993 clear_bit(__IXGBE_DOWN, &adapter->state);
3994 ixgbe_napi_enable_all(adapter);
3995
3996 if (ixgbe_is_sfp(hw)) {
3997 ixgbe_sfp_link_config(adapter);
3998 } else {
3999 err = ixgbe_non_sfp_link_config(hw);
4000 if (err)
4001 e_err(probe, "link_config FAILED %d\n", err);
4002 }
4003
4004 /* clear any pending interrupts, may auto mask */
4005 IXGBE_READ_REG(hw, IXGBE_EICR);
4006 ixgbe_irq_enable(adapter, true, true);
4007
4008 /*
4009 * If this adapter has a fan, check to see if we had a failure
4010 * before we enabled the interrupt.
4011 */
4012 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4013 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4014 if (esdp & IXGBE_ESDP_SDP1)
4015 e_crit(drv, "Fan has stopped, replace the adapter\n");
4016 }
4017
4018 /* enable transmits */
4019 netif_tx_start_all_queues(adapter->netdev);
4020
4021 /* bring the link up in the watchdog, this could race with our first
4022 * link up interrupt but shouldn't be a problem */
4023 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4024 adapter->link_check_timeout = jiffies;
4025 mod_timer(&adapter->service_timer, jiffies);
4026
4027 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4028 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4029 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4030 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4031 }
4032
4033 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4034 {
4035 WARN_ON(in_interrupt());
4036 /* put off any impending NetWatchDogTimeout */
4037 adapter->netdev->trans_start = jiffies;
4038
4039 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4040 usleep_range(1000, 2000);
4041 ixgbe_down(adapter);
4042 /*
4043 * If SR-IOV enabled then wait a bit before bringing the adapter
4044 * back up to give the VFs time to respond to the reset. The
4045 * two second wait is based upon the watchdog timer cycle in
4046 * the VF driver.
4047 */
4048 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4049 msleep(2000);
4050 ixgbe_up(adapter);
4051 clear_bit(__IXGBE_RESETTING, &adapter->state);
4052 }
4053
4054 void ixgbe_up(struct ixgbe_adapter *adapter)
4055 {
4056 /* hardware has been reset, we need to reload some things */
4057 ixgbe_configure(adapter);
4058
4059 ixgbe_up_complete(adapter);
4060 }
4061
4062 void ixgbe_reset(struct ixgbe_adapter *adapter)
4063 {
4064 struct ixgbe_hw *hw = &adapter->hw;
4065 int err;
4066
4067 /* lock SFP init bit to prevent race conditions with the watchdog */
4068 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4069 usleep_range(1000, 2000);
4070
4071 /* clear all SFP and link config related flags while holding SFP_INIT */
4072 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4073 IXGBE_FLAG2_SFP_NEEDS_RESET);
4074 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4075
4076 err = hw->mac.ops.init_hw(hw);
4077 switch (err) {
4078 case 0:
4079 case IXGBE_ERR_SFP_NOT_PRESENT:
4080 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4081 break;
4082 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4083 e_dev_err("master disable timed out\n");
4084 break;
4085 case IXGBE_ERR_EEPROM_VERSION:
4086 /* We are running on a pre-production device, log a warning */
4087 e_dev_warn("This device is a pre-production adapter/LOM. "
4088 "Please be aware there may be issues associated with "
4089 "your hardware. If you are experiencing problems "
4090 "please contact your Intel or hardware "
4091 "representative who provided you with this "
4092 "hardware.\n");
4093 break;
4094 default:
4095 e_dev_err("Hardware Error: %d\n", err);
4096 }
4097
4098 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4099
4100 /* reprogram the RAR[0] in case user changed it. */
4101 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
4102 IXGBE_RAH_AV);
4103 }
4104
4105 /**
4106 * ixgbe_init_rx_page_offset - initialize page offset values for Rx buffers
4107 * @rx_ring: ring to setup
4108 *
4109 * On many IA platforms the L1 cache has a critical stride of 4K, this
4110 * results in each receive buffer starting in the same cache set. To help
4111 * reduce the pressure on this cache set we can interleave the offsets so
4112 * that only every other buffer will be in the same cache set.
4113 **/
4114 static void ixgbe_init_rx_page_offset(struct ixgbe_ring *rx_ring)
4115 {
4116 struct ixgbe_rx_buffer *rx_buffer = rx_ring->rx_buffer_info;
4117 u16 i;
4118
4119 for (i = 0; i < rx_ring->count; i += 2) {
4120 rx_buffer[0].page_offset = 0;
4121 rx_buffer[1].page_offset = ixgbe_rx_bufsz(rx_ring);
4122 rx_buffer = &rx_buffer[2];
4123 }
4124 }
4125
4126 /**
4127 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4128 * @rx_ring: ring to free buffers from
4129 **/
4130 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4131 {
4132 struct device *dev = rx_ring->dev;
4133 unsigned long size;
4134 u16 i;
4135
4136 /* ring already cleared, nothing to do */
4137 if (!rx_ring->rx_buffer_info)
4138 return;
4139
4140 /* Free all the Rx ring sk_buffs */
4141 for (i = 0; i < rx_ring->count; i++) {
4142 struct ixgbe_rx_buffer *rx_buffer;
4143
4144 rx_buffer = &rx_ring->rx_buffer_info[i];
4145 if (rx_buffer->skb) {
4146 struct sk_buff *skb = rx_buffer->skb;
4147 if (IXGBE_CB(skb)->page_released) {
4148 dma_unmap_page(dev,
4149 IXGBE_CB(skb)->dma,
4150 ixgbe_rx_bufsz(rx_ring),
4151 DMA_FROM_DEVICE);
4152 IXGBE_CB(skb)->page_released = false;
4153 }
4154 dev_kfree_skb(skb);
4155 }
4156 rx_buffer->skb = NULL;
4157 if (rx_buffer->dma)
4158 dma_unmap_page(dev, rx_buffer->dma,
4159 ixgbe_rx_pg_size(rx_ring),
4160 DMA_FROM_DEVICE);
4161 rx_buffer->dma = 0;
4162 if (rx_buffer->page)
4163 __free_pages(rx_buffer->page,
4164 ixgbe_rx_pg_order(rx_ring));
4165 rx_buffer->page = NULL;
4166 }
4167
4168 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4169 memset(rx_ring->rx_buffer_info, 0, size);
4170
4171 ixgbe_init_rx_page_offset(rx_ring);
4172
4173 /* Zero out the descriptor ring */
4174 memset(rx_ring->desc, 0, rx_ring->size);
4175
4176 rx_ring->next_to_alloc = 0;
4177 rx_ring->next_to_clean = 0;
4178 rx_ring->next_to_use = 0;
4179 }
4180
4181 /**
4182 * ixgbe_clean_tx_ring - Free Tx Buffers
4183 * @tx_ring: ring to be cleaned
4184 **/
4185 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4186 {
4187 struct ixgbe_tx_buffer *tx_buffer_info;
4188 unsigned long size;
4189 u16 i;
4190
4191 /* ring already cleared, nothing to do */
4192 if (!tx_ring->tx_buffer_info)
4193 return;
4194
4195 /* Free all the Tx ring sk_buffs */
4196 for (i = 0; i < tx_ring->count; i++) {
4197 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4198 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4199 }
4200
4201 netdev_tx_reset_queue(txring_txq(tx_ring));
4202
4203 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4204 memset(tx_ring->tx_buffer_info, 0, size);
4205
4206 /* Zero out the descriptor ring */
4207 memset(tx_ring->desc, 0, tx_ring->size);
4208
4209 tx_ring->next_to_use = 0;
4210 tx_ring->next_to_clean = 0;
4211 }
4212
4213 /**
4214 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4215 * @adapter: board private structure
4216 **/
4217 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4218 {
4219 int i;
4220
4221 for (i = 0; i < adapter->num_rx_queues; i++)
4222 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4223 }
4224
4225 /**
4226 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4227 * @adapter: board private structure
4228 **/
4229 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4230 {
4231 int i;
4232
4233 for (i = 0; i < adapter->num_tx_queues; i++)
4234 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4235 }
4236
4237 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4238 {
4239 struct hlist_node *node, *node2;
4240 struct ixgbe_fdir_filter *filter;
4241
4242 spin_lock(&adapter->fdir_perfect_lock);
4243
4244 hlist_for_each_entry_safe(filter, node, node2,
4245 &adapter->fdir_filter_list, fdir_node) {
4246 hlist_del(&filter->fdir_node);
4247 kfree(filter);
4248 }
4249 adapter->fdir_filter_count = 0;
4250
4251 spin_unlock(&adapter->fdir_perfect_lock);
4252 }
4253
4254 void ixgbe_down(struct ixgbe_adapter *adapter)
4255 {
4256 struct net_device *netdev = adapter->netdev;
4257 struct ixgbe_hw *hw = &adapter->hw;
4258 u32 rxctrl;
4259 int i;
4260
4261 /* signal that we are down to the interrupt handler */
4262 set_bit(__IXGBE_DOWN, &adapter->state);
4263
4264 /* disable receives */
4265 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4266 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4267
4268 /* disable all enabled rx queues */
4269 for (i = 0; i < adapter->num_rx_queues; i++)
4270 /* this call also flushes the previous write */
4271 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4272
4273 usleep_range(10000, 20000);
4274
4275 netif_tx_stop_all_queues(netdev);
4276
4277 /* call carrier off first to avoid false dev_watchdog timeouts */
4278 netif_carrier_off(netdev);
4279 netif_tx_disable(netdev);
4280
4281 ixgbe_irq_disable(adapter);
4282
4283 ixgbe_napi_disable_all(adapter);
4284
4285 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4286 IXGBE_FLAG2_RESET_REQUESTED);
4287 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4288
4289 del_timer_sync(&adapter->service_timer);
4290
4291 if (adapter->num_vfs) {
4292 /* Clear EITR Select mapping */
4293 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4294
4295 /* Mark all the VFs as inactive */
4296 for (i = 0 ; i < adapter->num_vfs; i++)
4297 adapter->vfinfo[i].clear_to_send = false;
4298
4299 /* ping all the active vfs to let them know we are going down */
4300 ixgbe_ping_all_vfs(adapter);
4301
4302 /* Disable all VFTE/VFRE TX/RX */
4303 ixgbe_disable_tx_rx(adapter);
4304 }
4305
4306 /* disable transmits in the hardware now that interrupts are off */
4307 for (i = 0; i < adapter->num_tx_queues; i++) {
4308 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4309 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4310 }
4311
4312 /* Disable the Tx DMA engine on 82599 and X540 */
4313 switch (hw->mac.type) {
4314 case ixgbe_mac_82599EB:
4315 case ixgbe_mac_X540:
4316 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4317 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4318 ~IXGBE_DMATXCTL_TE));
4319 break;
4320 default:
4321 break;
4322 }
4323
4324 if (!pci_channel_offline(adapter->pdev))
4325 ixgbe_reset(adapter);
4326
4327 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4328 if (hw->mac.ops.disable_tx_laser &&
4329 ((hw->phy.multispeed_fiber) ||
4330 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4331 (hw->mac.type == ixgbe_mac_82599EB))))
4332 hw->mac.ops.disable_tx_laser(hw);
4333
4334 ixgbe_clean_all_tx_rings(adapter);
4335 ixgbe_clean_all_rx_rings(adapter);
4336
4337 #ifdef CONFIG_IXGBE_DCA
4338 /* since we reset the hardware DCA settings were cleared */
4339 ixgbe_setup_dca(adapter);
4340 #endif
4341 }
4342
4343 /**
4344 * ixgbe_tx_timeout - Respond to a Tx Hang
4345 * @netdev: network interface device structure
4346 **/
4347 static void ixgbe_tx_timeout(struct net_device *netdev)
4348 {
4349 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4350
4351 /* Do the reset outside of interrupt context */
4352 ixgbe_tx_timeout_reset(adapter);
4353 }
4354
4355 /**
4356 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4357 * @adapter: board private structure to initialize
4358 *
4359 * ixgbe_sw_init initializes the Adapter private data structure.
4360 * Fields are initialized based on PCI device information and
4361 * OS network device settings (MTU size).
4362 **/
4363 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4364 {
4365 struct ixgbe_hw *hw = &adapter->hw;
4366 struct pci_dev *pdev = adapter->pdev;
4367 unsigned int rss;
4368 #ifdef CONFIG_IXGBE_DCB
4369 int j;
4370 struct tc_configuration *tc;
4371 #endif
4372
4373 /* PCI config space info */
4374
4375 hw->vendor_id = pdev->vendor;
4376 hw->device_id = pdev->device;
4377 hw->revision_id = pdev->revision;
4378 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4379 hw->subsystem_device_id = pdev->subsystem_device;
4380
4381 /* Set capability flags */
4382 rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4383 adapter->ring_feature[RING_F_RSS].limit = rss;
4384 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4385 switch (hw->mac.type) {
4386 case ixgbe_mac_82598EB:
4387 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4388 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4389 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
4390 break;
4391 case ixgbe_mac_X540:
4392 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4393 case ixgbe_mac_82599EB:
4394 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
4395 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4396 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4397 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4398 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4399 /* Flow Director hash filters enabled */
4400 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4401 adapter->atr_sample_rate = 20;
4402 adapter->ring_feature[RING_F_FDIR].limit =
4403 IXGBE_MAX_FDIR_INDICES;
4404 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4405 #ifdef IXGBE_FCOE
4406 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4407 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4408 #ifdef CONFIG_IXGBE_DCB
4409 /* Default traffic class to use for FCoE */
4410 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4411 #endif
4412 #endif /* IXGBE_FCOE */
4413 break;
4414 default:
4415 break;
4416 }
4417
4418 /* n-tuple support exists, always init our spinlock */
4419 spin_lock_init(&adapter->fdir_perfect_lock);
4420
4421 #ifdef CONFIG_IXGBE_DCB
4422 switch (hw->mac.type) {
4423 case ixgbe_mac_X540:
4424 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4425 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4426 break;
4427 default:
4428 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4429 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4430 break;
4431 }
4432
4433 /* Configure DCB traffic classes */
4434 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4435 tc = &adapter->dcb_cfg.tc_config[j];
4436 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4437 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4438 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4439 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4440 tc->dcb_pfc = pfc_disabled;
4441 }
4442
4443 /* Initialize default user to priority mapping, UPx->TC0 */
4444 tc = &adapter->dcb_cfg.tc_config[0];
4445 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
4446 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
4447
4448 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4449 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4450 adapter->dcb_cfg.pfc_mode_enable = false;
4451 adapter->dcb_set_bitmap = 0x00;
4452 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4453 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
4454 sizeof(adapter->temp_dcb_cfg));
4455
4456 #endif
4457
4458 /* default flow control settings */
4459 hw->fc.requested_mode = ixgbe_fc_full;
4460 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4461 ixgbe_pbthresh_setup(adapter);
4462 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4463 hw->fc.send_xon = true;
4464 hw->fc.disable_fc_autoneg = false;
4465
4466 /* enable itr by default in dynamic mode */
4467 adapter->rx_itr_setting = 1;
4468 adapter->tx_itr_setting = 1;
4469
4470 /* set default ring sizes */
4471 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4472 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4473
4474 /* set default work limits */
4475 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4476
4477 /* initialize eeprom parameters */
4478 if (ixgbe_init_eeprom_params_generic(hw)) {
4479 e_dev_err("EEPROM initialization failed\n");
4480 return -EIO;
4481 }
4482
4483 set_bit(__IXGBE_DOWN, &adapter->state);
4484
4485 return 0;
4486 }
4487
4488 /**
4489 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4490 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4491 *
4492 * Return 0 on success, negative on failure
4493 **/
4494 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4495 {
4496 struct device *dev = tx_ring->dev;
4497 int orig_node = dev_to_node(dev);
4498 int numa_node = -1;
4499 int size;
4500
4501 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4502
4503 if (tx_ring->q_vector)
4504 numa_node = tx_ring->q_vector->numa_node;
4505
4506 tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
4507 if (!tx_ring->tx_buffer_info)
4508 tx_ring->tx_buffer_info = vzalloc(size);
4509 if (!tx_ring->tx_buffer_info)
4510 goto err;
4511
4512 /* round up to nearest 4K */
4513 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4514 tx_ring->size = ALIGN(tx_ring->size, 4096);
4515
4516 set_dev_node(dev, numa_node);
4517 tx_ring->desc = dma_alloc_coherent(dev,
4518 tx_ring->size,
4519 &tx_ring->dma,
4520 GFP_KERNEL);
4521 set_dev_node(dev, orig_node);
4522 if (!tx_ring->desc)
4523 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4524 &tx_ring->dma, GFP_KERNEL);
4525 if (!tx_ring->desc)
4526 goto err;
4527
4528 tx_ring->next_to_use = 0;
4529 tx_ring->next_to_clean = 0;
4530 return 0;
4531
4532 err:
4533 vfree(tx_ring->tx_buffer_info);
4534 tx_ring->tx_buffer_info = NULL;
4535 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4536 return -ENOMEM;
4537 }
4538
4539 /**
4540 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4541 * @adapter: board private structure
4542 *
4543 * If this function returns with an error, then it's possible one or
4544 * more of the rings is populated (while the rest are not). It is the
4545 * callers duty to clean those orphaned rings.
4546 *
4547 * Return 0 on success, negative on failure
4548 **/
4549 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4550 {
4551 int i, err = 0;
4552
4553 for (i = 0; i < adapter->num_tx_queues; i++) {
4554 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4555 if (!err)
4556 continue;
4557 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4558 break;
4559 }
4560
4561 return err;
4562 }
4563
4564 /**
4565 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4566 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4567 *
4568 * Returns 0 on success, negative on failure
4569 **/
4570 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4571 {
4572 struct device *dev = rx_ring->dev;
4573 int orig_node = dev_to_node(dev);
4574 int numa_node = -1;
4575 int size;
4576
4577 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4578
4579 if (rx_ring->q_vector)
4580 numa_node = rx_ring->q_vector->numa_node;
4581
4582 rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
4583 if (!rx_ring->rx_buffer_info)
4584 rx_ring->rx_buffer_info = vzalloc(size);
4585 if (!rx_ring->rx_buffer_info)
4586 goto err;
4587
4588 /* Round up to nearest 4K */
4589 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4590 rx_ring->size = ALIGN(rx_ring->size, 4096);
4591
4592 set_dev_node(dev, numa_node);
4593 rx_ring->desc = dma_alloc_coherent(dev,
4594 rx_ring->size,
4595 &rx_ring->dma,
4596 GFP_KERNEL);
4597 set_dev_node(dev, orig_node);
4598 if (!rx_ring->desc)
4599 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4600 &rx_ring->dma, GFP_KERNEL);
4601 if (!rx_ring->desc)
4602 goto err;
4603
4604 rx_ring->next_to_clean = 0;
4605 rx_ring->next_to_use = 0;
4606
4607 ixgbe_init_rx_page_offset(rx_ring);
4608
4609 return 0;
4610 err:
4611 vfree(rx_ring->rx_buffer_info);
4612 rx_ring->rx_buffer_info = NULL;
4613 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4614 return -ENOMEM;
4615 }
4616
4617 /**
4618 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4619 * @adapter: board private structure
4620 *
4621 * If this function returns with an error, then it's possible one or
4622 * more of the rings is populated (while the rest are not). It is the
4623 * callers duty to clean those orphaned rings.
4624 *
4625 * Return 0 on success, negative on failure
4626 **/
4627 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4628 {
4629 int i, err = 0;
4630
4631 for (i = 0; i < adapter->num_rx_queues; i++) {
4632 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
4633 if (!err)
4634 continue;
4635 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4636 break;
4637 }
4638
4639 return err;
4640 }
4641
4642 /**
4643 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4644 * @tx_ring: Tx descriptor ring for a specific queue
4645 *
4646 * Free all transmit software resources
4647 **/
4648 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
4649 {
4650 ixgbe_clean_tx_ring(tx_ring);
4651
4652 vfree(tx_ring->tx_buffer_info);
4653 tx_ring->tx_buffer_info = NULL;
4654
4655 /* if not set, then don't free */
4656 if (!tx_ring->desc)
4657 return;
4658
4659 dma_free_coherent(tx_ring->dev, tx_ring->size,
4660 tx_ring->desc, tx_ring->dma);
4661
4662 tx_ring->desc = NULL;
4663 }
4664
4665 /**
4666 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4667 * @adapter: board private structure
4668 *
4669 * Free all transmit software resources
4670 **/
4671 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4672 {
4673 int i;
4674
4675 for (i = 0; i < adapter->num_tx_queues; i++)
4676 if (adapter->tx_ring[i]->desc)
4677 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4678 }
4679
4680 /**
4681 * ixgbe_free_rx_resources - Free Rx Resources
4682 * @rx_ring: ring to clean the resources from
4683 *
4684 * Free all receive software resources
4685 **/
4686 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
4687 {
4688 ixgbe_clean_rx_ring(rx_ring);
4689
4690 vfree(rx_ring->rx_buffer_info);
4691 rx_ring->rx_buffer_info = NULL;
4692
4693 /* if not set, then don't free */
4694 if (!rx_ring->desc)
4695 return;
4696
4697 dma_free_coherent(rx_ring->dev, rx_ring->size,
4698 rx_ring->desc, rx_ring->dma);
4699
4700 rx_ring->desc = NULL;
4701 }
4702
4703 /**
4704 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4705 * @adapter: board private structure
4706 *
4707 * Free all receive software resources
4708 **/
4709 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4710 {
4711 int i;
4712
4713 for (i = 0; i < adapter->num_rx_queues; i++)
4714 if (adapter->rx_ring[i]->desc)
4715 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4716 }
4717
4718 /**
4719 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4720 * @netdev: network interface device structure
4721 * @new_mtu: new value for maximum frame size
4722 *
4723 * Returns 0 on success, negative on failure
4724 **/
4725 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4726 {
4727 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4728 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4729
4730 /* MTU < 68 is an error and causes problems on some kernels */
4731 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4732 return -EINVAL;
4733
4734 /*
4735 * For 82599EB we cannot allow PF to change MTU greater than 1500
4736 * in SR-IOV mode as it may cause buffer overruns in guest VFs that
4737 * don't allocate and chain buffers correctly.
4738 */
4739 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
4740 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
4741 (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
4742 return -EINVAL;
4743
4744 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4745
4746 /* must set new MTU before calling down or up */
4747 netdev->mtu = new_mtu;
4748
4749 if (netif_running(netdev))
4750 ixgbe_reinit_locked(adapter);
4751
4752 return 0;
4753 }
4754
4755 /**
4756 * ixgbe_open - Called when a network interface is made active
4757 * @netdev: network interface device structure
4758 *
4759 * Returns 0 on success, negative value on failure
4760 *
4761 * The open entry point is called when a network interface is made
4762 * active by the system (IFF_UP). At this point all resources needed
4763 * for transmit and receive operations are allocated, the interrupt
4764 * handler is registered with the OS, the watchdog timer is started,
4765 * and the stack is notified that the interface is ready.
4766 **/
4767 static int ixgbe_open(struct net_device *netdev)
4768 {
4769 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4770 int err;
4771
4772 /* disallow open during test */
4773 if (test_bit(__IXGBE_TESTING, &adapter->state))
4774 return -EBUSY;
4775
4776 netif_carrier_off(netdev);
4777
4778 /* allocate transmit descriptors */
4779 err = ixgbe_setup_all_tx_resources(adapter);
4780 if (err)
4781 goto err_setup_tx;
4782
4783 /* allocate receive descriptors */
4784 err = ixgbe_setup_all_rx_resources(adapter);
4785 if (err)
4786 goto err_setup_rx;
4787
4788 ixgbe_configure(adapter);
4789
4790 err = ixgbe_request_irq(adapter);
4791 if (err)
4792 goto err_req_irq;
4793
4794 ixgbe_up_complete(adapter);
4795
4796 return 0;
4797
4798 err_req_irq:
4799 err_setup_rx:
4800 ixgbe_free_all_rx_resources(adapter);
4801 err_setup_tx:
4802 ixgbe_free_all_tx_resources(adapter);
4803 ixgbe_reset(adapter);
4804
4805 return err;
4806 }
4807
4808 /**
4809 * ixgbe_close - Disables a network interface
4810 * @netdev: network interface device structure
4811 *
4812 * Returns 0, this is not allowed to fail
4813 *
4814 * The close entry point is called when an interface is de-activated
4815 * by the OS. The hardware is still under the drivers control, but
4816 * needs to be disabled. A global MAC reset is issued to stop the
4817 * hardware, and all transmit and receive resources are freed.
4818 **/
4819 static int ixgbe_close(struct net_device *netdev)
4820 {
4821 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4822
4823 ixgbe_down(adapter);
4824 ixgbe_free_irq(adapter);
4825
4826 ixgbe_fdir_filter_exit(adapter);
4827
4828 ixgbe_free_all_tx_resources(adapter);
4829 ixgbe_free_all_rx_resources(adapter);
4830
4831 ixgbe_release_hw_control(adapter);
4832
4833 return 0;
4834 }
4835
4836 #ifdef CONFIG_PM
4837 static int ixgbe_resume(struct pci_dev *pdev)
4838 {
4839 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4840 struct net_device *netdev = adapter->netdev;
4841 u32 err;
4842
4843 pci_set_power_state(pdev, PCI_D0);
4844 pci_restore_state(pdev);
4845 /*
4846 * pci_restore_state clears dev->state_saved so call
4847 * pci_save_state to restore it.
4848 */
4849 pci_save_state(pdev);
4850
4851 err = pci_enable_device_mem(pdev);
4852 if (err) {
4853 e_dev_err("Cannot enable PCI device from suspend\n");
4854 return err;
4855 }
4856 pci_set_master(pdev);
4857
4858 pci_wake_from_d3(pdev, false);
4859
4860 rtnl_lock();
4861 err = ixgbe_init_interrupt_scheme(adapter);
4862 rtnl_unlock();
4863 if (err) {
4864 e_dev_err("Cannot initialize interrupts for device\n");
4865 return err;
4866 }
4867
4868 ixgbe_reset(adapter);
4869
4870 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4871
4872 if (netif_running(netdev)) {
4873 err = ixgbe_open(netdev);
4874 if (err)
4875 return err;
4876 }
4877
4878 netif_device_attach(netdev);
4879
4880 return 0;
4881 }
4882 #endif /* CONFIG_PM */
4883
4884 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4885 {
4886 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4887 struct net_device *netdev = adapter->netdev;
4888 struct ixgbe_hw *hw = &adapter->hw;
4889 u32 ctrl, fctrl;
4890 u32 wufc = adapter->wol;
4891 #ifdef CONFIG_PM
4892 int retval = 0;
4893 #endif
4894
4895 netif_device_detach(netdev);
4896
4897 if (netif_running(netdev)) {
4898 rtnl_lock();
4899 ixgbe_down(adapter);
4900 ixgbe_free_irq(adapter);
4901 ixgbe_free_all_tx_resources(adapter);
4902 ixgbe_free_all_rx_resources(adapter);
4903 rtnl_unlock();
4904 }
4905
4906 ixgbe_clear_interrupt_scheme(adapter);
4907
4908 #ifdef CONFIG_PM
4909 retval = pci_save_state(pdev);
4910 if (retval)
4911 return retval;
4912
4913 #endif
4914 if (wufc) {
4915 ixgbe_set_rx_mode(netdev);
4916
4917 /*
4918 * enable the optics for both mult-speed fiber and
4919 * 82599 SFP+ fiber as we can WoL.
4920 */
4921 if (hw->mac.ops.enable_tx_laser &&
4922 (hw->phy.multispeed_fiber ||
4923 (hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber &&
4924 hw->mac.type == ixgbe_mac_82599EB)))
4925 hw->mac.ops.enable_tx_laser(hw);
4926
4927 /* turn on all-multi mode if wake on multicast is enabled */
4928 if (wufc & IXGBE_WUFC_MC) {
4929 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4930 fctrl |= IXGBE_FCTRL_MPE;
4931 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4932 }
4933
4934 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4935 ctrl |= IXGBE_CTRL_GIO_DIS;
4936 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4937
4938 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4939 } else {
4940 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4941 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4942 }
4943
4944 switch (hw->mac.type) {
4945 case ixgbe_mac_82598EB:
4946 pci_wake_from_d3(pdev, false);
4947 break;
4948 case ixgbe_mac_82599EB:
4949 case ixgbe_mac_X540:
4950 pci_wake_from_d3(pdev, !!wufc);
4951 break;
4952 default:
4953 break;
4954 }
4955
4956 *enable_wake = !!wufc;
4957
4958 ixgbe_release_hw_control(adapter);
4959
4960 pci_disable_device(pdev);
4961
4962 return 0;
4963 }
4964
4965 #ifdef CONFIG_PM
4966 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4967 {
4968 int retval;
4969 bool wake;
4970
4971 retval = __ixgbe_shutdown(pdev, &wake);
4972 if (retval)
4973 return retval;
4974
4975 if (wake) {
4976 pci_prepare_to_sleep(pdev);
4977 } else {
4978 pci_wake_from_d3(pdev, false);
4979 pci_set_power_state(pdev, PCI_D3hot);
4980 }
4981
4982 return 0;
4983 }
4984 #endif /* CONFIG_PM */
4985
4986 static void ixgbe_shutdown(struct pci_dev *pdev)
4987 {
4988 bool wake;
4989
4990 __ixgbe_shutdown(pdev, &wake);
4991
4992 if (system_state == SYSTEM_POWER_OFF) {
4993 pci_wake_from_d3(pdev, wake);
4994 pci_set_power_state(pdev, PCI_D3hot);
4995 }
4996 }
4997
4998 /**
4999 * ixgbe_update_stats - Update the board statistics counters.
5000 * @adapter: board private structure
5001 **/
5002 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5003 {
5004 struct net_device *netdev = adapter->netdev;
5005 struct ixgbe_hw *hw = &adapter->hw;
5006 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5007 u64 total_mpc = 0;
5008 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5009 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5010 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5011 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5012 #ifdef IXGBE_FCOE
5013 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5014 unsigned int cpu;
5015 u64 fcoe_noddp_counts_sum = 0, fcoe_noddp_ext_buff_counts_sum = 0;
5016 #endif /* IXGBE_FCOE */
5017
5018 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5019 test_bit(__IXGBE_RESETTING, &adapter->state))
5020 return;
5021
5022 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5023 u64 rsc_count = 0;
5024 u64 rsc_flush = 0;
5025 for (i = 0; i < adapter->num_rx_queues; i++) {
5026 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5027 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5028 }
5029 adapter->rsc_total_count = rsc_count;
5030 adapter->rsc_total_flush = rsc_flush;
5031 }
5032
5033 for (i = 0; i < adapter->num_rx_queues; i++) {
5034 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5035 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5036 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5037 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5038 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5039 bytes += rx_ring->stats.bytes;
5040 packets += rx_ring->stats.packets;
5041 }
5042 adapter->non_eop_descs = non_eop_descs;
5043 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5044 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5045 adapter->hw_csum_rx_error = hw_csum_rx_error;
5046 netdev->stats.rx_bytes = bytes;
5047 netdev->stats.rx_packets = packets;
5048
5049 bytes = 0;
5050 packets = 0;
5051 /* gather some stats to the adapter struct that are per queue */
5052 for (i = 0; i < adapter->num_tx_queues; i++) {
5053 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5054 restart_queue += tx_ring->tx_stats.restart_queue;
5055 tx_busy += tx_ring->tx_stats.tx_busy;
5056 bytes += tx_ring->stats.bytes;
5057 packets += tx_ring->stats.packets;
5058 }
5059 adapter->restart_queue = restart_queue;
5060 adapter->tx_busy = tx_busy;
5061 netdev->stats.tx_bytes = bytes;
5062 netdev->stats.tx_packets = packets;
5063
5064 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5065
5066 /* 8 register reads */
5067 for (i = 0; i < 8; i++) {
5068 /* for packet buffers not used, the register should read 0 */
5069 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5070 missed_rx += mpc;
5071 hwstats->mpc[i] += mpc;
5072 total_mpc += hwstats->mpc[i];
5073 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5074 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5075 switch (hw->mac.type) {
5076 case ixgbe_mac_82598EB:
5077 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5078 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5079 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5080 hwstats->pxonrxc[i] +=
5081 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5082 break;
5083 case ixgbe_mac_82599EB:
5084 case ixgbe_mac_X540:
5085 hwstats->pxonrxc[i] +=
5086 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5087 break;
5088 default:
5089 break;
5090 }
5091 }
5092
5093 /*16 register reads */
5094 for (i = 0; i < 16; i++) {
5095 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5096 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5097 if ((hw->mac.type == ixgbe_mac_82599EB) ||
5098 (hw->mac.type == ixgbe_mac_X540)) {
5099 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5100 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5101 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5102 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5103 }
5104 }
5105
5106 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5107 /* work around hardware counting issue */
5108 hwstats->gprc -= missed_rx;
5109
5110 ixgbe_update_xoff_received(adapter);
5111
5112 /* 82598 hardware only has a 32 bit counter in the high register */
5113 switch (hw->mac.type) {
5114 case ixgbe_mac_82598EB:
5115 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5116 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5117 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5118 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5119 break;
5120 case ixgbe_mac_X540:
5121 /* OS2BMC stats are X540 only*/
5122 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5123 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5124 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5125 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5126 case ixgbe_mac_82599EB:
5127 for (i = 0; i < 16; i++)
5128 adapter->hw_rx_no_dma_resources +=
5129 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5130 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5131 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5132 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5133 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5134 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5135 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5136 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5137 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5138 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5139 #ifdef IXGBE_FCOE
5140 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5141 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5142 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5143 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5144 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5145 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5146 /* Add up per cpu counters for total ddp aloc fail */
5147 if (fcoe->pcpu_noddp && fcoe->pcpu_noddp_ext_buff) {
5148 for_each_possible_cpu(cpu) {
5149 fcoe_noddp_counts_sum +=
5150 *per_cpu_ptr(fcoe->pcpu_noddp, cpu);
5151 fcoe_noddp_ext_buff_counts_sum +=
5152 *per_cpu_ptr(fcoe->
5153 pcpu_noddp_ext_buff, cpu);
5154 }
5155 }
5156 hwstats->fcoe_noddp = fcoe_noddp_counts_sum;
5157 hwstats->fcoe_noddp_ext_buff = fcoe_noddp_ext_buff_counts_sum;
5158 #endif /* IXGBE_FCOE */
5159 break;
5160 default:
5161 break;
5162 }
5163 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5164 hwstats->bprc += bprc;
5165 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5166 if (hw->mac.type == ixgbe_mac_82598EB)
5167 hwstats->mprc -= bprc;
5168 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5169 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5170 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5171 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5172 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5173 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5174 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5175 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5176 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5177 hwstats->lxontxc += lxon;
5178 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5179 hwstats->lxofftxc += lxoff;
5180 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5181 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5182 /*
5183 * 82598 errata - tx of flow control packets is included in tx counters
5184 */
5185 xon_off_tot = lxon + lxoff;
5186 hwstats->gptc -= xon_off_tot;
5187 hwstats->mptc -= xon_off_tot;
5188 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5189 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5190 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5191 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5192 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5193 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5194 hwstats->ptc64 -= xon_off_tot;
5195 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5196 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5197 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5198 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5199 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5200 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5201
5202 /* Fill out the OS statistics structure */
5203 netdev->stats.multicast = hwstats->mprc;
5204
5205 /* Rx Errors */
5206 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5207 netdev->stats.rx_dropped = 0;
5208 netdev->stats.rx_length_errors = hwstats->rlec;
5209 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5210 netdev->stats.rx_missed_errors = total_mpc;
5211 }
5212
5213 /**
5214 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5215 * @adapter: pointer to the device adapter structure
5216 **/
5217 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5218 {
5219 struct ixgbe_hw *hw = &adapter->hw;
5220 int i;
5221
5222 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5223 return;
5224
5225 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5226
5227 /* if interface is down do nothing */
5228 if (test_bit(__IXGBE_DOWN, &adapter->state))
5229 return;
5230
5231 /* do nothing if we are not using signature filters */
5232 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5233 return;
5234
5235 adapter->fdir_overflow++;
5236
5237 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5238 for (i = 0; i < adapter->num_tx_queues; i++)
5239 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5240 &(adapter->tx_ring[i]->state));
5241 /* re-enable flow director interrupts */
5242 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5243 } else {
5244 e_err(probe, "failed to finish FDIR re-initialization, "
5245 "ignored adding FDIR ATR filters\n");
5246 }
5247 }
5248
5249 /**
5250 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5251 * @adapter: pointer to the device adapter structure
5252 *
5253 * This function serves two purposes. First it strobes the interrupt lines
5254 * in order to make certain interrupts are occurring. Secondly it sets the
5255 * bits needed to check for TX hangs. As a result we should immediately
5256 * determine if a hang has occurred.
5257 */
5258 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5259 {
5260 struct ixgbe_hw *hw = &adapter->hw;
5261 u64 eics = 0;
5262 int i;
5263
5264 /* If we're down or resetting, just bail */
5265 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5266 test_bit(__IXGBE_RESETTING, &adapter->state))
5267 return;
5268
5269 /* Force detection of hung controller */
5270 if (netif_carrier_ok(adapter->netdev)) {
5271 for (i = 0; i < adapter->num_tx_queues; i++)
5272 set_check_for_tx_hang(adapter->tx_ring[i]);
5273 }
5274
5275 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5276 /*
5277 * for legacy and MSI interrupts don't set any bits
5278 * that are enabled for EIAM, because this operation
5279 * would set *both* EIMS and EICS for any bit in EIAM
5280 */
5281 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5282 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5283 } else {
5284 /* get one bit for every active tx/rx interrupt vector */
5285 for (i = 0; i < adapter->num_q_vectors; i++) {
5286 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5287 if (qv->rx.ring || qv->tx.ring)
5288 eics |= ((u64)1 << i);
5289 }
5290 }
5291
5292 /* Cause software interrupt to ensure rings are cleaned */
5293 ixgbe_irq_rearm_queues(adapter, eics);
5294
5295 }
5296
5297 /**
5298 * ixgbe_watchdog_update_link - update the link status
5299 * @adapter: pointer to the device adapter structure
5300 * @link_speed: pointer to a u32 to store the link_speed
5301 **/
5302 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5303 {
5304 struct ixgbe_hw *hw = &adapter->hw;
5305 u32 link_speed = adapter->link_speed;
5306 bool link_up = adapter->link_up;
5307 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
5308
5309 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5310 return;
5311
5312 if (hw->mac.ops.check_link) {
5313 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5314 } else {
5315 /* always assume link is up, if no check link function */
5316 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5317 link_up = true;
5318 }
5319
5320 if (adapter->ixgbe_ieee_pfc)
5321 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
5322
5323 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
5324 hw->mac.ops.fc_enable(hw);
5325 ixgbe_set_rx_drop_en(adapter);
5326 }
5327
5328 if (link_up ||
5329 time_after(jiffies, (adapter->link_check_timeout +
5330 IXGBE_TRY_LINK_TIMEOUT))) {
5331 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5332 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5333 IXGBE_WRITE_FLUSH(hw);
5334 }
5335
5336 adapter->link_up = link_up;
5337 adapter->link_speed = link_speed;
5338 }
5339
5340 /**
5341 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5342 * print link up message
5343 * @adapter: pointer to the device adapter structure
5344 **/
5345 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5346 {
5347 struct net_device *netdev = adapter->netdev;
5348 struct ixgbe_hw *hw = &adapter->hw;
5349 u32 link_speed = adapter->link_speed;
5350 bool flow_rx, flow_tx;
5351
5352 /* only continue if link was previously down */
5353 if (netif_carrier_ok(netdev))
5354 return;
5355
5356 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5357
5358 switch (hw->mac.type) {
5359 case ixgbe_mac_82598EB: {
5360 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5361 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5362 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5363 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5364 }
5365 break;
5366 case ixgbe_mac_X540:
5367 case ixgbe_mac_82599EB: {
5368 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5369 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5370 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5371 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5372 }
5373 break;
5374 default:
5375 flow_tx = false;
5376 flow_rx = false;
5377 break;
5378 }
5379
5380 #ifdef CONFIG_IXGBE_PTP
5381 ixgbe_ptp_start_cyclecounter(adapter);
5382 #endif
5383
5384 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5385 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5386 "10 Gbps" :
5387 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5388 "1 Gbps" :
5389 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5390 "100 Mbps" :
5391 "unknown speed"))),
5392 ((flow_rx && flow_tx) ? "RX/TX" :
5393 (flow_rx ? "RX" :
5394 (flow_tx ? "TX" : "None"))));
5395
5396 netif_carrier_on(netdev);
5397 ixgbe_check_vf_rate_limit(adapter);
5398 }
5399
5400 /**
5401 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5402 * print link down message
5403 * @adapter: pointer to the adapter structure
5404 **/
5405 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
5406 {
5407 struct net_device *netdev = adapter->netdev;
5408 struct ixgbe_hw *hw = &adapter->hw;
5409
5410 adapter->link_up = false;
5411 adapter->link_speed = 0;
5412
5413 /* only continue if link was up previously */
5414 if (!netif_carrier_ok(netdev))
5415 return;
5416
5417 /* poll for SFP+ cable when link is down */
5418 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5419 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5420
5421 #ifdef CONFIG_IXGBE_PTP
5422 ixgbe_ptp_start_cyclecounter(adapter);
5423 #endif
5424
5425 e_info(drv, "NIC Link is Down\n");
5426 netif_carrier_off(netdev);
5427 }
5428
5429 /**
5430 * ixgbe_watchdog_flush_tx - flush queues on link down
5431 * @adapter: pointer to the device adapter structure
5432 **/
5433 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5434 {
5435 int i;
5436 int some_tx_pending = 0;
5437
5438 if (!netif_carrier_ok(adapter->netdev)) {
5439 for (i = 0; i < adapter->num_tx_queues; i++) {
5440 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5441 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5442 some_tx_pending = 1;
5443 break;
5444 }
5445 }
5446
5447 if (some_tx_pending) {
5448 /* We've lost link, so the controller stops DMA,
5449 * but we've got queued Tx work that's never going
5450 * to get done, so reset controller to flush Tx.
5451 * (Do the reset outside of interrupt context).
5452 */
5453 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5454 }
5455 }
5456 }
5457
5458 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5459 {
5460 u32 ssvpc;
5461
5462 /* Do not perform spoof check for 82598 */
5463 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5464 return;
5465
5466 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5467
5468 /*
5469 * ssvpc register is cleared on read, if zero then no
5470 * spoofed packets in the last interval.
5471 */
5472 if (!ssvpc)
5473 return;
5474
5475 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
5476 }
5477
5478 /**
5479 * ixgbe_watchdog_subtask - check and bring link up
5480 * @adapter: pointer to the device adapter structure
5481 **/
5482 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5483 {
5484 /* if interface is down do nothing */
5485 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5486 test_bit(__IXGBE_RESETTING, &adapter->state))
5487 return;
5488
5489 ixgbe_watchdog_update_link(adapter);
5490
5491 if (adapter->link_up)
5492 ixgbe_watchdog_link_is_up(adapter);
5493 else
5494 ixgbe_watchdog_link_is_down(adapter);
5495
5496 ixgbe_spoof_check(adapter);
5497 ixgbe_update_stats(adapter);
5498
5499 ixgbe_watchdog_flush_tx(adapter);
5500 }
5501
5502 /**
5503 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5504 * @adapter: the ixgbe adapter structure
5505 **/
5506 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5507 {
5508 struct ixgbe_hw *hw = &adapter->hw;
5509 s32 err;
5510
5511 /* not searching for SFP so there is nothing to do here */
5512 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5513 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5514 return;
5515
5516 /* someone else is in init, wait until next service event */
5517 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5518 return;
5519
5520 err = hw->phy.ops.identify_sfp(hw);
5521 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5522 goto sfp_out;
5523
5524 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5525 /* If no cable is present, then we need to reset
5526 * the next time we find a good cable. */
5527 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5528 }
5529
5530 /* exit on error */
5531 if (err)
5532 goto sfp_out;
5533
5534 /* exit if reset not needed */
5535 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5536 goto sfp_out;
5537
5538 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5539
5540 /*
5541 * A module may be identified correctly, but the EEPROM may not have
5542 * support for that module. setup_sfp() will fail in that case, so
5543 * we should not allow that module to load.
5544 */
5545 if (hw->mac.type == ixgbe_mac_82598EB)
5546 err = hw->phy.ops.reset(hw);
5547 else
5548 err = hw->mac.ops.setup_sfp(hw);
5549
5550 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5551 goto sfp_out;
5552
5553 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5554 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5555
5556 sfp_out:
5557 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5558
5559 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5560 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5561 e_dev_err("failed to initialize because an unsupported "
5562 "SFP+ module type was detected.\n");
5563 e_dev_err("Reload the driver after installing a "
5564 "supported module.\n");
5565 unregister_netdev(adapter->netdev);
5566 }
5567 }
5568
5569 /**
5570 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5571 * @adapter: the ixgbe adapter structure
5572 **/
5573 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5574 {
5575 struct ixgbe_hw *hw = &adapter->hw;
5576 u32 autoneg;
5577 bool negotiation;
5578
5579 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5580 return;
5581
5582 /* someone else is in init, wait until next service event */
5583 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5584 return;
5585
5586 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5587
5588 autoneg = hw->phy.autoneg_advertised;
5589 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5590 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5591 if (hw->mac.ops.setup_link)
5592 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5593
5594 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5595 adapter->link_check_timeout = jiffies;
5596 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5597 }
5598
5599 #ifdef CONFIG_PCI_IOV
5600 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
5601 {
5602 int vf;
5603 struct ixgbe_hw *hw = &adapter->hw;
5604 struct net_device *netdev = adapter->netdev;
5605 u32 gpc;
5606 u32 ciaa, ciad;
5607
5608 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
5609 if (gpc) /* If incrementing then no need for the check below */
5610 return;
5611 /*
5612 * Check to see if a bad DMA write target from an errant or
5613 * malicious VF has caused a PCIe error. If so then we can
5614 * issue a VFLR to the offending VF(s) and then resume without
5615 * requesting a full slot reset.
5616 */
5617
5618 for (vf = 0; vf < adapter->num_vfs; vf++) {
5619 ciaa = (vf << 16) | 0x80000000;
5620 /* 32 bit read so align, we really want status at offset 6 */
5621 ciaa |= PCI_COMMAND;
5622 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5623 ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
5624 ciaa &= 0x7FFFFFFF;
5625 /* disable debug mode asap after reading data */
5626 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5627 /* Get the upper 16 bits which will be the PCI status reg */
5628 ciad >>= 16;
5629 if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
5630 netdev_err(netdev, "VF %d Hung DMA\n", vf);
5631 /* Issue VFLR */
5632 ciaa = (vf << 16) | 0x80000000;
5633 ciaa |= 0xA8;
5634 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5635 ciad = 0x00008000; /* VFLR */
5636 IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
5637 ciaa &= 0x7FFFFFFF;
5638 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5639 }
5640 }
5641 }
5642
5643 #endif
5644 /**
5645 * ixgbe_service_timer - Timer Call-back
5646 * @data: pointer to adapter cast into an unsigned long
5647 **/
5648 static void ixgbe_service_timer(unsigned long data)
5649 {
5650 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5651 unsigned long next_event_offset;
5652 bool ready = true;
5653
5654 /* poll faster when waiting for link */
5655 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5656 next_event_offset = HZ / 10;
5657 else
5658 next_event_offset = HZ * 2;
5659
5660 #ifdef CONFIG_PCI_IOV
5661 /*
5662 * don't bother with SR-IOV VF DMA hang check if there are
5663 * no VFs or the link is down
5664 */
5665 if (!adapter->num_vfs ||
5666 (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5667 goto normal_timer_service;
5668
5669 /* If we have VFs allocated then we must check for DMA hangs */
5670 ixgbe_check_for_bad_vf(adapter);
5671 next_event_offset = HZ / 50;
5672 adapter->timer_event_accumulator++;
5673
5674 if (adapter->timer_event_accumulator >= 100)
5675 adapter->timer_event_accumulator = 0;
5676 else
5677 ready = false;
5678
5679 normal_timer_service:
5680 #endif
5681 /* Reset the timer */
5682 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5683
5684 if (ready)
5685 ixgbe_service_event_schedule(adapter);
5686 }
5687
5688 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5689 {
5690 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
5691 return;
5692
5693 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
5694
5695 /* If we're already down or resetting, just bail */
5696 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5697 test_bit(__IXGBE_RESETTING, &adapter->state))
5698 return;
5699
5700 ixgbe_dump(adapter);
5701 netdev_err(adapter->netdev, "Reset adapter\n");
5702 adapter->tx_timeout_count++;
5703
5704 ixgbe_reinit_locked(adapter);
5705 }
5706
5707 /**
5708 * ixgbe_service_task - manages and runs subtasks
5709 * @work: pointer to work_struct containing our data
5710 **/
5711 static void ixgbe_service_task(struct work_struct *work)
5712 {
5713 struct ixgbe_adapter *adapter = container_of(work,
5714 struct ixgbe_adapter,
5715 service_task);
5716
5717 ixgbe_reset_subtask(adapter);
5718 ixgbe_sfp_detection_subtask(adapter);
5719 ixgbe_sfp_link_config_subtask(adapter);
5720 ixgbe_check_overtemp_subtask(adapter);
5721 ixgbe_watchdog_subtask(adapter);
5722 ixgbe_fdir_reinit_subtask(adapter);
5723 ixgbe_check_hang_subtask(adapter);
5724 #ifdef CONFIG_IXGBE_PTP
5725 ixgbe_ptp_overflow_check(adapter);
5726 #endif
5727
5728 ixgbe_service_event_complete(adapter);
5729 }
5730
5731 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
5732 struct ixgbe_tx_buffer *first,
5733 u8 *hdr_len)
5734 {
5735 struct sk_buff *skb = first->skb;
5736 u32 vlan_macip_lens, type_tucmd;
5737 u32 mss_l4len_idx, l4len;
5738
5739 if (!skb_is_gso(skb))
5740 return 0;
5741
5742 if (skb_header_cloned(skb)) {
5743 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5744 if (err)
5745 return err;
5746 }
5747
5748 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5749 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
5750
5751 if (first->protocol == __constant_htons(ETH_P_IP)) {
5752 struct iphdr *iph = ip_hdr(skb);
5753 iph->tot_len = 0;
5754 iph->check = 0;
5755 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5756 iph->daddr, 0,
5757 IPPROTO_TCP,
5758 0);
5759 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5760 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5761 IXGBE_TX_FLAGS_CSUM |
5762 IXGBE_TX_FLAGS_IPV4;
5763 } else if (skb_is_gso_v6(skb)) {
5764 ipv6_hdr(skb)->payload_len = 0;
5765 tcp_hdr(skb)->check =
5766 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5767 &ipv6_hdr(skb)->daddr,
5768 0, IPPROTO_TCP, 0);
5769 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5770 IXGBE_TX_FLAGS_CSUM;
5771 }
5772
5773 /* compute header lengths */
5774 l4len = tcp_hdrlen(skb);
5775 *hdr_len = skb_transport_offset(skb) + l4len;
5776
5777 /* update gso size and bytecount with header size */
5778 first->gso_segs = skb_shinfo(skb)->gso_segs;
5779 first->bytecount += (first->gso_segs - 1) * *hdr_len;
5780
5781 /* mss_l4len_id: use 1 as index for TSO */
5782 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
5783 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
5784 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
5785
5786 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
5787 vlan_macip_lens = skb_network_header_len(skb);
5788 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5789 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5790
5791 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
5792 mss_l4len_idx);
5793
5794 return 1;
5795 }
5796
5797 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
5798 struct ixgbe_tx_buffer *first)
5799 {
5800 struct sk_buff *skb = first->skb;
5801 u32 vlan_macip_lens = 0;
5802 u32 mss_l4len_idx = 0;
5803 u32 type_tucmd = 0;
5804
5805 if (skb->ip_summed != CHECKSUM_PARTIAL) {
5806 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
5807 !(first->tx_flags & IXGBE_TX_FLAGS_TXSW))
5808 return;
5809 } else {
5810 u8 l4_hdr = 0;
5811 switch (first->protocol) {
5812 case __constant_htons(ETH_P_IP):
5813 vlan_macip_lens |= skb_network_header_len(skb);
5814 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5815 l4_hdr = ip_hdr(skb)->protocol;
5816 break;
5817 case __constant_htons(ETH_P_IPV6):
5818 vlan_macip_lens |= skb_network_header_len(skb);
5819 l4_hdr = ipv6_hdr(skb)->nexthdr;
5820 break;
5821 default:
5822 if (unlikely(net_ratelimit())) {
5823 dev_warn(tx_ring->dev,
5824 "partial checksum but proto=%x!\n",
5825 first->protocol);
5826 }
5827 break;
5828 }
5829
5830 switch (l4_hdr) {
5831 case IPPROTO_TCP:
5832 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5833 mss_l4len_idx = tcp_hdrlen(skb) <<
5834 IXGBE_ADVTXD_L4LEN_SHIFT;
5835 break;
5836 case IPPROTO_SCTP:
5837 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5838 mss_l4len_idx = sizeof(struct sctphdr) <<
5839 IXGBE_ADVTXD_L4LEN_SHIFT;
5840 break;
5841 case IPPROTO_UDP:
5842 mss_l4len_idx = sizeof(struct udphdr) <<
5843 IXGBE_ADVTXD_L4LEN_SHIFT;
5844 break;
5845 default:
5846 if (unlikely(net_ratelimit())) {
5847 dev_warn(tx_ring->dev,
5848 "partial checksum but l4 proto=%x!\n",
5849 l4_hdr);
5850 }
5851 break;
5852 }
5853
5854 /* update TX checksum flag */
5855 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
5856 }
5857
5858 /* vlan_macip_lens: MACLEN, VLAN tag */
5859 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5860 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5861
5862 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
5863 type_tucmd, mss_l4len_idx);
5864 }
5865
5866 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
5867 {
5868 /* set type for advanced descriptor with frame checksum insertion */
5869 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
5870 IXGBE_ADVTXD_DCMD_IFCS |
5871 IXGBE_ADVTXD_DCMD_DEXT);
5872
5873 /* set HW vlan bit if vlan is present */
5874 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
5875 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
5876
5877 #ifdef CONFIG_IXGBE_PTP
5878 if (tx_flags & IXGBE_TX_FLAGS_TSTAMP)
5879 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_MAC_TSTAMP);
5880 #endif
5881
5882 /* set segmentation enable bits for TSO/FSO */
5883 #ifdef IXGBE_FCOE
5884 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FSO))
5885 #else
5886 if (tx_flags & IXGBE_TX_FLAGS_TSO)
5887 #endif
5888 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
5889
5890 return cmd_type;
5891 }
5892
5893 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
5894 u32 tx_flags, unsigned int paylen)
5895 {
5896 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
5897
5898 /* enable L4 checksum for TSO and TX checksum offload */
5899 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5900 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
5901
5902 /* enble IPv4 checksum for TSO */
5903 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5904 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
5905
5906 /* use index 1 context for TSO/FSO/FCOE */
5907 #ifdef IXGBE_FCOE
5908 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FCOE))
5909 #else
5910 if (tx_flags & IXGBE_TX_FLAGS_TSO)
5911 #endif
5912 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
5913
5914 /*
5915 * Check Context must be set if Tx switch is enabled, which it
5916 * always is for case where virtual functions are running
5917 */
5918 #ifdef IXGBE_FCOE
5919 if (tx_flags & (IXGBE_TX_FLAGS_TXSW | IXGBE_TX_FLAGS_FCOE))
5920 #else
5921 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
5922 #endif
5923 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
5924
5925 tx_desc->read.olinfo_status = olinfo_status;
5926 }
5927
5928 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
5929 IXGBE_TXD_CMD_RS)
5930
5931 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
5932 struct ixgbe_tx_buffer *first,
5933 const u8 hdr_len)
5934 {
5935 dma_addr_t dma;
5936 struct sk_buff *skb = first->skb;
5937 struct ixgbe_tx_buffer *tx_buffer;
5938 union ixgbe_adv_tx_desc *tx_desc;
5939 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
5940 unsigned int data_len = skb->data_len;
5941 unsigned int size = skb_headlen(skb);
5942 unsigned int paylen = skb->len - hdr_len;
5943 u32 tx_flags = first->tx_flags;
5944 __le32 cmd_type;
5945 u16 i = tx_ring->next_to_use;
5946
5947 tx_desc = IXGBE_TX_DESC(tx_ring, i);
5948
5949 ixgbe_tx_olinfo_status(tx_desc, tx_flags, paylen);
5950 cmd_type = ixgbe_tx_cmd_type(tx_flags);
5951
5952 #ifdef IXGBE_FCOE
5953 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5954 if (data_len < sizeof(struct fcoe_crc_eof)) {
5955 size -= sizeof(struct fcoe_crc_eof) - data_len;
5956 data_len = 0;
5957 } else {
5958 data_len -= sizeof(struct fcoe_crc_eof);
5959 }
5960 }
5961
5962 #endif
5963 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
5964 if (dma_mapping_error(tx_ring->dev, dma))
5965 goto dma_error;
5966
5967 /* record length, and DMA address */
5968 dma_unmap_len_set(first, len, size);
5969 dma_unmap_addr_set(first, dma, dma);
5970
5971 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5972
5973 for (;;) {
5974 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
5975 tx_desc->read.cmd_type_len =
5976 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
5977
5978 i++;
5979 tx_desc++;
5980 if (i == tx_ring->count) {
5981 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5982 i = 0;
5983 }
5984
5985 dma += IXGBE_MAX_DATA_PER_TXD;
5986 size -= IXGBE_MAX_DATA_PER_TXD;
5987
5988 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5989 tx_desc->read.olinfo_status = 0;
5990 }
5991
5992 if (likely(!data_len))
5993 break;
5994
5995 if (unlikely(skb->no_fcs))
5996 cmd_type &= ~(cpu_to_le32(IXGBE_ADVTXD_DCMD_IFCS));
5997 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
5998
5999 i++;
6000 tx_desc++;
6001 if (i == tx_ring->count) {
6002 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6003 i = 0;
6004 }
6005
6006 #ifdef IXGBE_FCOE
6007 size = min_t(unsigned int, data_len, skb_frag_size(frag));
6008 #else
6009 size = skb_frag_size(frag);
6010 #endif
6011 data_len -= size;
6012
6013 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6014 DMA_TO_DEVICE);
6015 if (dma_mapping_error(tx_ring->dev, dma))
6016 goto dma_error;
6017
6018 tx_buffer = &tx_ring->tx_buffer_info[i];
6019 dma_unmap_len_set(tx_buffer, len, size);
6020 dma_unmap_addr_set(tx_buffer, dma, dma);
6021
6022 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6023 tx_desc->read.olinfo_status = 0;
6024
6025 frag++;
6026 }
6027
6028 /* write last descriptor with RS and EOP bits */
6029 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
6030 tx_desc->read.cmd_type_len = cmd_type;
6031
6032 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6033
6034 /* set the timestamp */
6035 first->time_stamp = jiffies;
6036
6037 /*
6038 * Force memory writes to complete before letting h/w know there
6039 * are new descriptors to fetch. (Only applicable for weak-ordered
6040 * memory model archs, such as IA-64).
6041 *
6042 * We also need this memory barrier to make certain all of the
6043 * status bits have been updated before next_to_watch is written.
6044 */
6045 wmb();
6046
6047 /* set next_to_watch value indicating a packet is present */
6048 first->next_to_watch = tx_desc;
6049
6050 i++;
6051 if (i == tx_ring->count)
6052 i = 0;
6053
6054 tx_ring->next_to_use = i;
6055
6056 /* notify HW of packet */
6057 writel(i, tx_ring->tail);
6058
6059 return;
6060 dma_error:
6061 dev_err(tx_ring->dev, "TX DMA map failed\n");
6062
6063 /* clear dma mappings for failed tx_buffer_info map */
6064 for (;;) {
6065 tx_buffer = &tx_ring->tx_buffer_info[i];
6066 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6067 if (tx_buffer == first)
6068 break;
6069 if (i == 0)
6070 i = tx_ring->count;
6071 i--;
6072 }
6073
6074 tx_ring->next_to_use = i;
6075 }
6076
6077 static void ixgbe_atr(struct ixgbe_ring *ring,
6078 struct ixgbe_tx_buffer *first)
6079 {
6080 struct ixgbe_q_vector *q_vector = ring->q_vector;
6081 union ixgbe_atr_hash_dword input = { .dword = 0 };
6082 union ixgbe_atr_hash_dword common = { .dword = 0 };
6083 union {
6084 unsigned char *network;
6085 struct iphdr *ipv4;
6086 struct ipv6hdr *ipv6;
6087 } hdr;
6088 struct tcphdr *th;
6089 __be16 vlan_id;
6090
6091 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6092 if (!q_vector)
6093 return;
6094
6095 /* do nothing if sampling is disabled */
6096 if (!ring->atr_sample_rate)
6097 return;
6098
6099 ring->atr_count++;
6100
6101 /* snag network header to get L4 type and address */
6102 hdr.network = skb_network_header(first->skb);
6103
6104 /* Currently only IPv4/IPv6 with TCP is supported */
6105 if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6106 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6107 (first->protocol != __constant_htons(ETH_P_IP) ||
6108 hdr.ipv4->protocol != IPPROTO_TCP))
6109 return;
6110
6111 th = tcp_hdr(first->skb);
6112
6113 /* skip this packet since it is invalid or the socket is closing */
6114 if (!th || th->fin)
6115 return;
6116
6117 /* sample on all syn packets or once every atr sample count */
6118 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6119 return;
6120
6121 /* reset sample count */
6122 ring->atr_count = 0;
6123
6124 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6125
6126 /*
6127 * src and dst are inverted, think how the receiver sees them
6128 *
6129 * The input is broken into two sections, a non-compressed section
6130 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6131 * is XORed together and stored in the compressed dword.
6132 */
6133 input.formatted.vlan_id = vlan_id;
6134
6135 /*
6136 * since src port and flex bytes occupy the same word XOR them together
6137 * and write the value to source port portion of compressed dword
6138 */
6139 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6140 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6141 else
6142 common.port.src ^= th->dest ^ first->protocol;
6143 common.port.dst ^= th->source;
6144
6145 if (first->protocol == __constant_htons(ETH_P_IP)) {
6146 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6147 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6148 } else {
6149 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6150 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6151 hdr.ipv6->saddr.s6_addr32[1] ^
6152 hdr.ipv6->saddr.s6_addr32[2] ^
6153 hdr.ipv6->saddr.s6_addr32[3] ^
6154 hdr.ipv6->daddr.s6_addr32[0] ^
6155 hdr.ipv6->daddr.s6_addr32[1] ^
6156 hdr.ipv6->daddr.s6_addr32[2] ^
6157 hdr.ipv6->daddr.s6_addr32[3];
6158 }
6159
6160 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6161 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6162 input, common, ring->queue_index);
6163 }
6164
6165 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6166 {
6167 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6168 /* Herbert's original patch had:
6169 * smp_mb__after_netif_stop_queue();
6170 * but since that doesn't exist yet, just open code it. */
6171 smp_mb();
6172
6173 /* We need to check again in a case another CPU has just
6174 * made room available. */
6175 if (likely(ixgbe_desc_unused(tx_ring) < size))
6176 return -EBUSY;
6177
6178 /* A reprieve! - use start_queue because it doesn't call schedule */
6179 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6180 ++tx_ring->tx_stats.restart_queue;
6181 return 0;
6182 }
6183
6184 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6185 {
6186 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6187 return 0;
6188 return __ixgbe_maybe_stop_tx(tx_ring, size);
6189 }
6190
6191 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6192 {
6193 struct ixgbe_adapter *adapter = netdev_priv(dev);
6194 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6195 smp_processor_id();
6196 #ifdef IXGBE_FCOE
6197 __be16 protocol = vlan_get_protocol(skb);
6198
6199 if (((protocol == htons(ETH_P_FCOE)) ||
6200 (protocol == htons(ETH_P_FIP))) &&
6201 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6202 struct ixgbe_ring_feature *f;
6203
6204 f = &adapter->ring_feature[RING_F_FCOE];
6205
6206 while (txq >= f->indices)
6207 txq -= f->indices;
6208 txq += adapter->ring_feature[RING_F_FCOE].offset;
6209
6210 return txq;
6211 }
6212 #endif
6213
6214 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6215 while (unlikely(txq >= dev->real_num_tx_queues))
6216 txq -= dev->real_num_tx_queues;
6217 return txq;
6218 }
6219
6220 return skb_tx_hash(dev, skb);
6221 }
6222
6223 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6224 struct ixgbe_adapter *adapter,
6225 struct ixgbe_ring *tx_ring)
6226 {
6227 struct ixgbe_tx_buffer *first;
6228 int tso;
6229 u32 tx_flags = 0;
6230 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6231 unsigned short f;
6232 #endif
6233 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6234 __be16 protocol = skb->protocol;
6235 u8 hdr_len = 0;
6236
6237 /*
6238 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6239 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6240 * + 2 desc gap to keep tail from touching head,
6241 * + 1 desc for context descriptor,
6242 * otherwise try next time
6243 */
6244 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6245 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6246 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6247 #else
6248 count += skb_shinfo(skb)->nr_frags;
6249 #endif
6250 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6251 tx_ring->tx_stats.tx_busy++;
6252 return NETDEV_TX_BUSY;
6253 }
6254
6255 /* record the location of the first descriptor for this packet */
6256 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6257 first->skb = skb;
6258 first->bytecount = skb->len;
6259 first->gso_segs = 1;
6260
6261 /* if we have a HW VLAN tag being added default to the HW one */
6262 if (vlan_tx_tag_present(skb)) {
6263 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6264 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6265 /* else if it is a SW VLAN check the next protocol and store the tag */
6266 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6267 struct vlan_hdr *vhdr, _vhdr;
6268 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6269 if (!vhdr)
6270 goto out_drop;
6271
6272 protocol = vhdr->h_vlan_encapsulated_proto;
6273 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6274 IXGBE_TX_FLAGS_VLAN_SHIFT;
6275 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6276 }
6277
6278 skb_tx_timestamp(skb);
6279
6280 #ifdef CONFIG_IXGBE_PTP
6281 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6282 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6283 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
6284 }
6285 #endif
6286
6287 #ifdef CONFIG_PCI_IOV
6288 /*
6289 * Use the l2switch_enable flag - would be false if the DMA
6290 * Tx switch had been disabled.
6291 */
6292 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6293 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6294
6295 #endif
6296 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6297 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6298 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6299 (skb->priority != TC_PRIO_CONTROL))) {
6300 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6301 tx_flags |= (skb->priority & 0x7) <<
6302 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6303 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6304 struct vlan_ethhdr *vhdr;
6305 if (skb_header_cloned(skb) &&
6306 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6307 goto out_drop;
6308 vhdr = (struct vlan_ethhdr *)skb->data;
6309 vhdr->h_vlan_TCI = htons(tx_flags >>
6310 IXGBE_TX_FLAGS_VLAN_SHIFT);
6311 } else {
6312 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6313 }
6314 }
6315
6316 /* record initial flags and protocol */
6317 first->tx_flags = tx_flags;
6318 first->protocol = protocol;
6319
6320 #ifdef IXGBE_FCOE
6321 /* setup tx offload for FCoE */
6322 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6323 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6324 tso = ixgbe_fso(tx_ring, first, &hdr_len);
6325 if (tso < 0)
6326 goto out_drop;
6327
6328 goto xmit_fcoe;
6329 }
6330
6331 #endif /* IXGBE_FCOE */
6332 tso = ixgbe_tso(tx_ring, first, &hdr_len);
6333 if (tso < 0)
6334 goto out_drop;
6335 else if (!tso)
6336 ixgbe_tx_csum(tx_ring, first);
6337
6338 /* add the ATR filter if ATR is on */
6339 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6340 ixgbe_atr(tx_ring, first);
6341
6342 #ifdef IXGBE_FCOE
6343 xmit_fcoe:
6344 #endif /* IXGBE_FCOE */
6345 ixgbe_tx_map(tx_ring, first, hdr_len);
6346
6347 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6348
6349 return NETDEV_TX_OK;
6350
6351 out_drop:
6352 dev_kfree_skb_any(first->skb);
6353 first->skb = NULL;
6354
6355 return NETDEV_TX_OK;
6356 }
6357
6358 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6359 struct net_device *netdev)
6360 {
6361 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6362 struct ixgbe_ring *tx_ring;
6363
6364 /*
6365 * The minimum packet size for olinfo paylen is 17 so pad the skb
6366 * in order to meet this minimum size requirement.
6367 */
6368 if (unlikely(skb->len < 17)) {
6369 if (skb_pad(skb, 17 - skb->len))
6370 return NETDEV_TX_OK;
6371 skb->len = 17;
6372 }
6373
6374 tx_ring = adapter->tx_ring[skb->queue_mapping];
6375 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6376 }
6377
6378 /**
6379 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6380 * @netdev: network interface device structure
6381 * @p: pointer to an address structure
6382 *
6383 * Returns 0 on success, negative on failure
6384 **/
6385 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6386 {
6387 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6388 struct ixgbe_hw *hw = &adapter->hw;
6389 struct sockaddr *addr = p;
6390
6391 if (!is_valid_ether_addr(addr->sa_data))
6392 return -EADDRNOTAVAIL;
6393
6394 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6395 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6396
6397 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6398 IXGBE_RAH_AV);
6399
6400 return 0;
6401 }
6402
6403 static int
6404 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6405 {
6406 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6407 struct ixgbe_hw *hw = &adapter->hw;
6408 u16 value;
6409 int rc;
6410
6411 if (prtad != hw->phy.mdio.prtad)
6412 return -EINVAL;
6413 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6414 if (!rc)
6415 rc = value;
6416 return rc;
6417 }
6418
6419 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6420 u16 addr, u16 value)
6421 {
6422 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6423 struct ixgbe_hw *hw = &adapter->hw;
6424
6425 if (prtad != hw->phy.mdio.prtad)
6426 return -EINVAL;
6427 return hw->phy.ops.write_reg(hw, addr, devad, value);
6428 }
6429
6430 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6431 {
6432 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6433
6434 switch (cmd) {
6435 #ifdef CONFIG_IXGBE_PTP
6436 case SIOCSHWTSTAMP:
6437 return ixgbe_ptp_hwtstamp_ioctl(adapter, req, cmd);
6438 #endif
6439 default:
6440 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6441 }
6442 }
6443
6444 /**
6445 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6446 * netdev->dev_addrs
6447 * @netdev: network interface device structure
6448 *
6449 * Returns non-zero on failure
6450 **/
6451 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6452 {
6453 int err = 0;
6454 struct ixgbe_adapter *adapter = netdev_priv(dev);
6455 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6456
6457 if (is_valid_ether_addr(mac->san_addr)) {
6458 rtnl_lock();
6459 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6460 rtnl_unlock();
6461 }
6462 return err;
6463 }
6464
6465 /**
6466 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6467 * netdev->dev_addrs
6468 * @netdev: network interface device structure
6469 *
6470 * Returns non-zero on failure
6471 **/
6472 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6473 {
6474 int err = 0;
6475 struct ixgbe_adapter *adapter = netdev_priv(dev);
6476 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6477
6478 if (is_valid_ether_addr(mac->san_addr)) {
6479 rtnl_lock();
6480 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6481 rtnl_unlock();
6482 }
6483 return err;
6484 }
6485
6486 #ifdef CONFIG_NET_POLL_CONTROLLER
6487 /*
6488 * Polling 'interrupt' - used by things like netconsole to send skbs
6489 * without having to re-enable interrupts. It's not called while
6490 * the interrupt routine is executing.
6491 */
6492 static void ixgbe_netpoll(struct net_device *netdev)
6493 {
6494 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6495 int i;
6496
6497 /* if interface is down do nothing */
6498 if (test_bit(__IXGBE_DOWN, &adapter->state))
6499 return;
6500
6501 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6502 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6503 for (i = 0; i < adapter->num_q_vectors; i++)
6504 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
6505 } else {
6506 ixgbe_intr(adapter->pdev->irq, netdev);
6507 }
6508 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6509 }
6510
6511 #endif
6512 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6513 struct rtnl_link_stats64 *stats)
6514 {
6515 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6516 int i;
6517
6518 rcu_read_lock();
6519 for (i = 0; i < adapter->num_rx_queues; i++) {
6520 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6521 u64 bytes, packets;
6522 unsigned int start;
6523
6524 if (ring) {
6525 do {
6526 start = u64_stats_fetch_begin_bh(&ring->syncp);
6527 packets = ring->stats.packets;
6528 bytes = ring->stats.bytes;
6529 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6530 stats->rx_packets += packets;
6531 stats->rx_bytes += bytes;
6532 }
6533 }
6534
6535 for (i = 0; i < adapter->num_tx_queues; i++) {
6536 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6537 u64 bytes, packets;
6538 unsigned int start;
6539
6540 if (ring) {
6541 do {
6542 start = u64_stats_fetch_begin_bh(&ring->syncp);
6543 packets = ring->stats.packets;
6544 bytes = ring->stats.bytes;
6545 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6546 stats->tx_packets += packets;
6547 stats->tx_bytes += bytes;
6548 }
6549 }
6550 rcu_read_unlock();
6551 /* following stats updated by ixgbe_watchdog_task() */
6552 stats->multicast = netdev->stats.multicast;
6553 stats->rx_errors = netdev->stats.rx_errors;
6554 stats->rx_length_errors = netdev->stats.rx_length_errors;
6555 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6556 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6557 return stats;
6558 }
6559
6560 #ifdef CONFIG_IXGBE_DCB
6561 /**
6562 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6563 * @adapter: pointer to ixgbe_adapter
6564 * @tc: number of traffic classes currently enabled
6565 *
6566 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6567 * 802.1Q priority maps to a packet buffer that exists.
6568 */
6569 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6570 {
6571 struct ixgbe_hw *hw = &adapter->hw;
6572 u32 reg, rsave;
6573 int i;
6574
6575 /* 82598 have a static priority to TC mapping that can not
6576 * be changed so no validation is needed.
6577 */
6578 if (hw->mac.type == ixgbe_mac_82598EB)
6579 return;
6580
6581 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6582 rsave = reg;
6583
6584 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6585 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6586
6587 /* If up2tc is out of bounds default to zero */
6588 if (up2tc > tc)
6589 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6590 }
6591
6592 if (reg != rsave)
6593 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6594
6595 return;
6596 }
6597
6598 /**
6599 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
6600 * @adapter: Pointer to adapter struct
6601 *
6602 * Populate the netdev user priority to tc map
6603 */
6604 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
6605 {
6606 struct net_device *dev = adapter->netdev;
6607 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
6608 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
6609 u8 prio;
6610
6611 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
6612 u8 tc = 0;
6613
6614 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
6615 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
6616 else if (ets)
6617 tc = ets->prio_tc[prio];
6618
6619 netdev_set_prio_tc_map(dev, prio, tc);
6620 }
6621 }
6622
6623 /**
6624 * ixgbe_setup_tc - configure net_device for multiple traffic classes
6625 *
6626 * @netdev: net device to configure
6627 * @tc: number of traffic classes to enable
6628 */
6629 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6630 {
6631 struct ixgbe_adapter *adapter = netdev_priv(dev);
6632 struct ixgbe_hw *hw = &adapter->hw;
6633
6634 /* Multiple traffic classes requires multiple queues */
6635 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6636 e_err(drv, "Enable failed, needs MSI-X\n");
6637 return -EINVAL;
6638 }
6639
6640 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6641 e_err(drv, "Enable failed, SR-IOV enabled\n");
6642 return -EINVAL;
6643 }
6644
6645 /* Hardware supports up to 8 traffic classes */
6646 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
6647 (hw->mac.type == ixgbe_mac_82598EB &&
6648 tc < MAX_TRAFFIC_CLASS))
6649 return -EINVAL;
6650
6651 /* Hardware has to reinitialize queues and interrupts to
6652 * match packet buffer alignment. Unfortunately, the
6653 * hardware is not flexible enough to do this dynamically.
6654 */
6655 if (netif_running(dev))
6656 ixgbe_close(dev);
6657 ixgbe_clear_interrupt_scheme(adapter);
6658
6659 if (tc) {
6660 netdev_set_num_tc(dev, tc);
6661 ixgbe_set_prio_tc_map(adapter);
6662
6663 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6664 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6665
6666 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
6667 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
6668 adapter->hw.fc.requested_mode = ixgbe_fc_none;
6669 }
6670 } else {
6671 netdev_reset_tc(dev);
6672
6673 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6674 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6675
6676 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6677 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6678
6679 adapter->temp_dcb_cfg.pfc_mode_enable = false;
6680 adapter->dcb_cfg.pfc_mode_enable = false;
6681 }
6682
6683 ixgbe_init_interrupt_scheme(adapter);
6684 ixgbe_validate_rtr(adapter, tc);
6685 if (netif_running(dev))
6686 ixgbe_open(dev);
6687
6688 return 0;
6689 }
6690
6691 #endif /* CONFIG_IXGBE_DCB */
6692 void ixgbe_do_reset(struct net_device *netdev)
6693 {
6694 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6695
6696 if (netif_running(netdev))
6697 ixgbe_reinit_locked(adapter);
6698 else
6699 ixgbe_reset(adapter);
6700 }
6701
6702 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
6703 netdev_features_t features)
6704 {
6705 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6706
6707 /* return error if RXHASH is being enabled when RSS is not supported */
6708 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
6709 features &= ~NETIF_F_RXHASH;
6710
6711 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6712 if (!(features & NETIF_F_RXCSUM))
6713 features &= ~NETIF_F_LRO;
6714
6715 /* Turn off LRO if not RSC capable */
6716 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
6717 features &= ~NETIF_F_LRO;
6718
6719 return features;
6720 }
6721
6722 static int ixgbe_set_features(struct net_device *netdev,
6723 netdev_features_t features)
6724 {
6725 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6726 netdev_features_t changed = netdev->features ^ features;
6727 bool need_reset = false;
6728
6729 /* Make sure RSC matches LRO, reset if change */
6730 if (!(features & NETIF_F_LRO)) {
6731 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6732 need_reset = true;
6733 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
6734 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
6735 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6736 if (adapter->rx_itr_setting == 1 ||
6737 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
6738 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
6739 need_reset = true;
6740 } else if ((changed ^ features) & NETIF_F_LRO) {
6741 e_info(probe, "rx-usecs set too low, "
6742 "disabling RSC\n");
6743 }
6744 }
6745
6746 /*
6747 * Check if Flow Director n-tuple support was enabled or disabled. If
6748 * the state changed, we need to reset.
6749 */
6750 if (!(features & NETIF_F_NTUPLE)) {
6751 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
6752 /* turn off Flow Director, set ATR and reset */
6753 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
6754 !(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
6755 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6756 need_reset = true;
6757 }
6758 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6759 } else if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
6760 /* turn off ATR, enable perfect filters and reset */
6761 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6762 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6763 need_reset = true;
6764 }
6765
6766 if (features & NETIF_F_HW_VLAN_RX)
6767 ixgbe_vlan_strip_enable(adapter);
6768 else
6769 ixgbe_vlan_strip_disable(adapter);
6770
6771 if (changed & NETIF_F_RXALL)
6772 need_reset = true;
6773
6774 netdev->features = features;
6775 if (need_reset)
6776 ixgbe_do_reset(netdev);
6777
6778 return 0;
6779 }
6780
6781 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm,
6782 struct net_device *dev,
6783 unsigned char *addr,
6784 u16 flags)
6785 {
6786 struct ixgbe_adapter *adapter = netdev_priv(dev);
6787 int err = -EOPNOTSUPP;
6788
6789 if (ndm->ndm_state & NUD_PERMANENT) {
6790 pr_info("%s: FDB only supports static addresses\n",
6791 ixgbe_driver_name);
6792 return -EINVAL;
6793 }
6794
6795 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6796 if (is_unicast_ether_addr(addr))
6797 err = dev_uc_add_excl(dev, addr);
6798 else if (is_multicast_ether_addr(addr))
6799 err = dev_mc_add_excl(dev, addr);
6800 else
6801 err = -EINVAL;
6802 }
6803
6804 /* Only return duplicate errors if NLM_F_EXCL is set */
6805 if (err == -EEXIST && !(flags & NLM_F_EXCL))
6806 err = 0;
6807
6808 return err;
6809 }
6810
6811 static int ixgbe_ndo_fdb_del(struct ndmsg *ndm,
6812 struct net_device *dev,
6813 unsigned char *addr)
6814 {
6815 struct ixgbe_adapter *adapter = netdev_priv(dev);
6816 int err = -EOPNOTSUPP;
6817
6818 if (ndm->ndm_state & NUD_PERMANENT) {
6819 pr_info("%s: FDB only supports static addresses\n",
6820 ixgbe_driver_name);
6821 return -EINVAL;
6822 }
6823
6824 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6825 if (is_unicast_ether_addr(addr))
6826 err = dev_uc_del(dev, addr);
6827 else if (is_multicast_ether_addr(addr))
6828 err = dev_mc_del(dev, addr);
6829 else
6830 err = -EINVAL;
6831 }
6832
6833 return err;
6834 }
6835
6836 static int ixgbe_ndo_fdb_dump(struct sk_buff *skb,
6837 struct netlink_callback *cb,
6838 struct net_device *dev,
6839 int idx)
6840 {
6841 struct ixgbe_adapter *adapter = netdev_priv(dev);
6842
6843 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6844 idx = ndo_dflt_fdb_dump(skb, cb, dev, idx);
6845
6846 return idx;
6847 }
6848
6849 static const struct net_device_ops ixgbe_netdev_ops = {
6850 .ndo_open = ixgbe_open,
6851 .ndo_stop = ixgbe_close,
6852 .ndo_start_xmit = ixgbe_xmit_frame,
6853 .ndo_select_queue = ixgbe_select_queue,
6854 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6855 .ndo_validate_addr = eth_validate_addr,
6856 .ndo_set_mac_address = ixgbe_set_mac,
6857 .ndo_change_mtu = ixgbe_change_mtu,
6858 .ndo_tx_timeout = ixgbe_tx_timeout,
6859 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6860 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6861 .ndo_do_ioctl = ixgbe_ioctl,
6862 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6863 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6864 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6865 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
6866 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6867 .ndo_get_stats64 = ixgbe_get_stats64,
6868 #ifdef CONFIG_IXGBE_DCB
6869 .ndo_setup_tc = ixgbe_setup_tc,
6870 #endif
6871 #ifdef CONFIG_NET_POLL_CONTROLLER
6872 .ndo_poll_controller = ixgbe_netpoll,
6873 #endif
6874 #ifdef IXGBE_FCOE
6875 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6876 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
6877 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6878 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6879 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6880 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6881 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
6882 #endif /* IXGBE_FCOE */
6883 .ndo_set_features = ixgbe_set_features,
6884 .ndo_fix_features = ixgbe_fix_features,
6885 .ndo_fdb_add = ixgbe_ndo_fdb_add,
6886 .ndo_fdb_del = ixgbe_ndo_fdb_del,
6887 .ndo_fdb_dump = ixgbe_ndo_fdb_dump,
6888 };
6889
6890 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6891 const struct ixgbe_info *ii)
6892 {
6893 #ifdef CONFIG_PCI_IOV
6894 struct ixgbe_hw *hw = &adapter->hw;
6895
6896 if (hw->mac.type == ixgbe_mac_82598EB)
6897 return;
6898
6899 /* The 82599 supports up to 64 VFs per physical function
6900 * but this implementation limits allocation to 63 so that
6901 * basic networking resources are still available to the
6902 * physical function. If the user requests greater thn
6903 * 63 VFs then it is an error - reset to default of zero.
6904 */
6905 adapter->num_vfs = (max_vfs > 63) ? 0 : max_vfs;
6906 ixgbe_enable_sriov(adapter, ii);
6907 #endif /* CONFIG_PCI_IOV */
6908 }
6909
6910 /**
6911 * ixgbe_wol_supported - Check whether device supports WoL
6912 * @hw: hw specific details
6913 * @device_id: the device ID
6914 * @subdev_id: the subsystem device ID
6915 *
6916 * This function is used by probe and ethtool to determine
6917 * which devices have WoL support
6918 *
6919 **/
6920 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
6921 u16 subdevice_id)
6922 {
6923 struct ixgbe_hw *hw = &adapter->hw;
6924 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
6925 int is_wol_supported = 0;
6926
6927 switch (device_id) {
6928 case IXGBE_DEV_ID_82599_SFP:
6929 /* Only these subdevices could supports WOL */
6930 switch (subdevice_id) {
6931 case IXGBE_SUBDEV_ID_82599_560FLR:
6932 /* only support first port */
6933 if (hw->bus.func != 0)
6934 break;
6935 case IXGBE_SUBDEV_ID_82599_SFP:
6936 is_wol_supported = 1;
6937 break;
6938 }
6939 break;
6940 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
6941 /* All except this subdevice support WOL */
6942 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
6943 is_wol_supported = 1;
6944 break;
6945 case IXGBE_DEV_ID_82599_KX4:
6946 is_wol_supported = 1;
6947 break;
6948 case IXGBE_DEV_ID_X540T:
6949 /* check eeprom to see if enabled wol */
6950 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
6951 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
6952 (hw->bus.func == 0))) {
6953 is_wol_supported = 1;
6954 }
6955 break;
6956 }
6957
6958 return is_wol_supported;
6959 }
6960
6961 /**
6962 * ixgbe_probe - Device Initialization Routine
6963 * @pdev: PCI device information struct
6964 * @ent: entry in ixgbe_pci_tbl
6965 *
6966 * Returns 0 on success, negative on failure
6967 *
6968 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6969 * The OS initialization, configuring of the adapter private structure,
6970 * and a hardware reset occur.
6971 **/
6972 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6973 const struct pci_device_id *ent)
6974 {
6975 struct net_device *netdev;
6976 struct ixgbe_adapter *adapter = NULL;
6977 struct ixgbe_hw *hw;
6978 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6979 static int cards_found;
6980 int i, err, pci_using_dac;
6981 u8 part_str[IXGBE_PBANUM_LENGTH];
6982 unsigned int indices = num_possible_cpus();
6983 #ifdef IXGBE_FCOE
6984 u16 device_caps;
6985 #endif
6986 u32 eec;
6987
6988 /* Catch broken hardware that put the wrong VF device ID in
6989 * the PCIe SR-IOV capability.
6990 */
6991 if (pdev->is_virtfn) {
6992 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6993 pci_name(pdev), pdev->vendor, pdev->device);
6994 return -EINVAL;
6995 }
6996
6997 err = pci_enable_device_mem(pdev);
6998 if (err)
6999 return err;
7000
7001 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7002 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7003 pci_using_dac = 1;
7004 } else {
7005 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7006 if (err) {
7007 err = dma_set_coherent_mask(&pdev->dev,
7008 DMA_BIT_MASK(32));
7009 if (err) {
7010 dev_err(&pdev->dev,
7011 "No usable DMA configuration, aborting\n");
7012 goto err_dma;
7013 }
7014 }
7015 pci_using_dac = 0;
7016 }
7017
7018 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7019 IORESOURCE_MEM), ixgbe_driver_name);
7020 if (err) {
7021 dev_err(&pdev->dev,
7022 "pci_request_selected_regions failed 0x%x\n", err);
7023 goto err_pci_reg;
7024 }
7025
7026 pci_enable_pcie_error_reporting(pdev);
7027
7028 pci_set_master(pdev);
7029 pci_save_state(pdev);
7030
7031 #ifdef CONFIG_IXGBE_DCB
7032 indices *= MAX_TRAFFIC_CLASS;
7033 #endif
7034
7035 if (ii->mac == ixgbe_mac_82598EB)
7036 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7037 else
7038 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7039
7040 #ifdef IXGBE_FCOE
7041 indices += min_t(unsigned int, num_possible_cpus(),
7042 IXGBE_MAX_FCOE_INDICES);
7043 #endif
7044 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7045 if (!netdev) {
7046 err = -ENOMEM;
7047 goto err_alloc_etherdev;
7048 }
7049
7050 SET_NETDEV_DEV(netdev, &pdev->dev);
7051
7052 adapter = netdev_priv(netdev);
7053 pci_set_drvdata(pdev, adapter);
7054
7055 adapter->netdev = netdev;
7056 adapter->pdev = pdev;
7057 hw = &adapter->hw;
7058 hw->back = adapter;
7059 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7060
7061 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7062 pci_resource_len(pdev, 0));
7063 if (!hw->hw_addr) {
7064 err = -EIO;
7065 goto err_ioremap;
7066 }
7067
7068 for (i = 1; i <= 5; i++) {
7069 if (pci_resource_len(pdev, i) == 0)
7070 continue;
7071 }
7072
7073 netdev->netdev_ops = &ixgbe_netdev_ops;
7074 ixgbe_set_ethtool_ops(netdev);
7075 netdev->watchdog_timeo = 5 * HZ;
7076 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7077
7078 adapter->bd_number = cards_found;
7079
7080 /* Setup hw api */
7081 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7082 hw->mac.type = ii->mac;
7083
7084 /* EEPROM */
7085 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7086 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7087 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7088 if (!(eec & (1 << 8)))
7089 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7090
7091 /* PHY */
7092 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7093 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7094 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7095 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7096 hw->phy.mdio.mmds = 0;
7097 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7098 hw->phy.mdio.dev = netdev;
7099 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7100 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7101
7102 ii->get_invariants(hw);
7103
7104 /* setup the private structure */
7105 err = ixgbe_sw_init(adapter);
7106 if (err)
7107 goto err_sw_init;
7108
7109 /* Make it possible the adapter to be woken up via WOL */
7110 switch (adapter->hw.mac.type) {
7111 case ixgbe_mac_82599EB:
7112 case ixgbe_mac_X540:
7113 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7114 break;
7115 default:
7116 break;
7117 }
7118
7119 /*
7120 * If there is a fan on this device and it has failed log the
7121 * failure.
7122 */
7123 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7124 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7125 if (esdp & IXGBE_ESDP_SDP1)
7126 e_crit(probe, "Fan has stopped, replace the adapter\n");
7127 }
7128
7129 if (allow_unsupported_sfp)
7130 hw->allow_unsupported_sfp = allow_unsupported_sfp;
7131
7132 /* reset_hw fills in the perm_addr as well */
7133 hw->phy.reset_if_overtemp = true;
7134 err = hw->mac.ops.reset_hw(hw);
7135 hw->phy.reset_if_overtemp = false;
7136 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7137 hw->mac.type == ixgbe_mac_82598EB) {
7138 err = 0;
7139 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7140 e_dev_err("failed to load because an unsupported SFP+ "
7141 "module type was detected.\n");
7142 e_dev_err("Reload the driver after installing a supported "
7143 "module.\n");
7144 goto err_sw_init;
7145 } else if (err) {
7146 e_dev_err("HW Init failed: %d\n", err);
7147 goto err_sw_init;
7148 }
7149
7150 ixgbe_probe_vf(adapter, ii);
7151
7152 netdev->features = NETIF_F_SG |
7153 NETIF_F_IP_CSUM |
7154 NETIF_F_IPV6_CSUM |
7155 NETIF_F_HW_VLAN_TX |
7156 NETIF_F_HW_VLAN_RX |
7157 NETIF_F_HW_VLAN_FILTER |
7158 NETIF_F_TSO |
7159 NETIF_F_TSO6 |
7160 NETIF_F_RXHASH |
7161 NETIF_F_RXCSUM;
7162
7163 netdev->hw_features = netdev->features;
7164
7165 switch (adapter->hw.mac.type) {
7166 case ixgbe_mac_82599EB:
7167 case ixgbe_mac_X540:
7168 netdev->features |= NETIF_F_SCTP_CSUM;
7169 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7170 NETIF_F_NTUPLE;
7171 break;
7172 default:
7173 break;
7174 }
7175
7176 netdev->hw_features |= NETIF_F_RXALL;
7177
7178 netdev->vlan_features |= NETIF_F_TSO;
7179 netdev->vlan_features |= NETIF_F_TSO6;
7180 netdev->vlan_features |= NETIF_F_IP_CSUM;
7181 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7182 netdev->vlan_features |= NETIF_F_SG;
7183
7184 netdev->priv_flags |= IFF_UNICAST_FLT;
7185 netdev->priv_flags |= IFF_SUPP_NOFCS;
7186
7187 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7188 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7189 IXGBE_FLAG_DCB_ENABLED);
7190
7191 #ifdef CONFIG_IXGBE_DCB
7192 netdev->dcbnl_ops = &dcbnl_ops;
7193 #endif
7194
7195 #ifdef IXGBE_FCOE
7196 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7197 if (hw->mac.ops.get_device_caps) {
7198 hw->mac.ops.get_device_caps(hw, &device_caps);
7199 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7200 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7201 }
7202 }
7203 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7204 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7205 netdev->vlan_features |= NETIF_F_FSO;
7206 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7207 }
7208 #endif /* IXGBE_FCOE */
7209 if (pci_using_dac) {
7210 netdev->features |= NETIF_F_HIGHDMA;
7211 netdev->vlan_features |= NETIF_F_HIGHDMA;
7212 }
7213
7214 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7215 netdev->hw_features |= NETIF_F_LRO;
7216 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7217 netdev->features |= NETIF_F_LRO;
7218
7219 /* make sure the EEPROM is good */
7220 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7221 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7222 err = -EIO;
7223 goto err_sw_init;
7224 }
7225
7226 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7227 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7228
7229 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7230 e_dev_err("invalid MAC address\n");
7231 err = -EIO;
7232 goto err_sw_init;
7233 }
7234
7235 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7236 (unsigned long) adapter);
7237
7238 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7239 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7240
7241 err = ixgbe_init_interrupt_scheme(adapter);
7242 if (err)
7243 goto err_sw_init;
7244
7245 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7246 netdev->hw_features &= ~NETIF_F_RXHASH;
7247 netdev->features &= ~NETIF_F_RXHASH;
7248 }
7249
7250 /* WOL not supported for all devices */
7251 adapter->wol = 0;
7252 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7253 if (ixgbe_wol_supported(adapter, pdev->device, pdev->subsystem_device))
7254 adapter->wol = IXGBE_WUFC_MAG;
7255
7256 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7257
7258 #ifdef CONFIG_IXGBE_PTP
7259 ixgbe_ptp_init(adapter);
7260 #endif /* CONFIG_IXGBE_PTP*/
7261
7262 /* save off EEPROM version number */
7263 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7264 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7265
7266 /* pick up the PCI bus settings for reporting later */
7267 hw->mac.ops.get_bus_info(hw);
7268
7269 /* print bus type/speed/width info */
7270 e_dev_info("(PCI Express:%s:%s) %pM\n",
7271 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7272 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7273 "Unknown"),
7274 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7275 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7276 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7277 "Unknown"),
7278 netdev->dev_addr);
7279
7280 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7281 if (err)
7282 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7283 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7284 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7285 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7286 part_str);
7287 else
7288 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7289 hw->mac.type, hw->phy.type, part_str);
7290
7291 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7292 e_dev_warn("PCI-Express bandwidth available for this card is "
7293 "not sufficient for optimal performance.\n");
7294 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7295 "is required.\n");
7296 }
7297
7298 /* reset the hardware with the new settings */
7299 err = hw->mac.ops.start_hw(hw);
7300 if (err == IXGBE_ERR_EEPROM_VERSION) {
7301 /* We are running on a pre-production device, log a warning */
7302 e_dev_warn("This device is a pre-production adapter/LOM. "
7303 "Please be aware there may be issues associated "
7304 "with your hardware. If you are experiencing "
7305 "problems please contact your Intel or hardware "
7306 "representative who provided you with this "
7307 "hardware.\n");
7308 }
7309 strcpy(netdev->name, "eth%d");
7310 err = register_netdev(netdev);
7311 if (err)
7312 goto err_register;
7313
7314 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7315 if (hw->mac.ops.disable_tx_laser &&
7316 ((hw->phy.multispeed_fiber) ||
7317 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7318 (hw->mac.type == ixgbe_mac_82599EB))))
7319 hw->mac.ops.disable_tx_laser(hw);
7320
7321 /* carrier off reporting is important to ethtool even BEFORE open */
7322 netif_carrier_off(netdev);
7323
7324 #ifdef CONFIG_IXGBE_DCA
7325 if (dca_add_requester(&pdev->dev) == 0) {
7326 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7327 ixgbe_setup_dca(adapter);
7328 }
7329 #endif
7330 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7331 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7332 for (i = 0; i < adapter->num_vfs; i++)
7333 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7334 }
7335
7336 /* firmware requires driver version to be 0xFFFFFFFF
7337 * since os does not support feature
7338 */
7339 if (hw->mac.ops.set_fw_drv_ver)
7340 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7341 0xFF);
7342
7343 /* add san mac addr to netdev */
7344 ixgbe_add_sanmac_netdev(netdev);
7345
7346 e_dev_info("%s\n", ixgbe_default_device_descr);
7347 cards_found++;
7348
7349 #ifdef CONFIG_IXGBE_HWMON
7350 if (ixgbe_sysfs_init(adapter))
7351 e_err(probe, "failed to allocate sysfs resources\n");
7352 #endif /* CONFIG_IXGBE_HWMON */
7353
7354 return 0;
7355
7356 err_register:
7357 ixgbe_release_hw_control(adapter);
7358 ixgbe_clear_interrupt_scheme(adapter);
7359 err_sw_init:
7360 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7361 ixgbe_disable_sriov(adapter);
7362 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7363 iounmap(hw->hw_addr);
7364 err_ioremap:
7365 free_netdev(netdev);
7366 err_alloc_etherdev:
7367 pci_release_selected_regions(pdev,
7368 pci_select_bars(pdev, IORESOURCE_MEM));
7369 err_pci_reg:
7370 err_dma:
7371 pci_disable_device(pdev);
7372 return err;
7373 }
7374
7375 /**
7376 * ixgbe_remove - Device Removal Routine
7377 * @pdev: PCI device information struct
7378 *
7379 * ixgbe_remove is called by the PCI subsystem to alert the driver
7380 * that it should release a PCI device. The could be caused by a
7381 * Hot-Plug event, or because the driver is going to be removed from
7382 * memory.
7383 **/
7384 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7385 {
7386 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7387 struct net_device *netdev = adapter->netdev;
7388
7389 set_bit(__IXGBE_DOWN, &adapter->state);
7390 cancel_work_sync(&adapter->service_task);
7391
7392 #ifdef CONFIG_IXGBE_PTP
7393 ixgbe_ptp_stop(adapter);
7394 #endif
7395
7396 #ifdef CONFIG_IXGBE_DCA
7397 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7398 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7399 dca_remove_requester(&pdev->dev);
7400 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7401 }
7402
7403 #endif
7404 #ifdef CONFIG_IXGBE_HWMON
7405 ixgbe_sysfs_exit(adapter);
7406 #endif /* CONFIG_IXGBE_HWMON */
7407
7408 #ifdef IXGBE_FCOE
7409 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7410 ixgbe_cleanup_fcoe(adapter);
7411
7412 #endif /* IXGBE_FCOE */
7413
7414 /* remove the added san mac */
7415 ixgbe_del_sanmac_netdev(netdev);
7416
7417 if (netdev->reg_state == NETREG_REGISTERED)
7418 unregister_netdev(netdev);
7419
7420 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7421 if (!(ixgbe_check_vf_assignment(adapter)))
7422 ixgbe_disable_sriov(adapter);
7423 else
7424 e_dev_warn("Unloading driver while VFs are assigned "
7425 "- VFs will not be deallocated\n");
7426 }
7427
7428 ixgbe_clear_interrupt_scheme(adapter);
7429
7430 ixgbe_release_hw_control(adapter);
7431
7432 #ifdef CONFIG_DCB
7433 kfree(adapter->ixgbe_ieee_pfc);
7434 kfree(adapter->ixgbe_ieee_ets);
7435
7436 #endif
7437 iounmap(adapter->hw.hw_addr);
7438 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7439 IORESOURCE_MEM));
7440
7441 e_dev_info("complete\n");
7442
7443 free_netdev(netdev);
7444
7445 pci_disable_pcie_error_reporting(pdev);
7446
7447 pci_disable_device(pdev);
7448 }
7449
7450 /**
7451 * ixgbe_io_error_detected - called when PCI error is detected
7452 * @pdev: Pointer to PCI device
7453 * @state: The current pci connection state
7454 *
7455 * This function is called after a PCI bus error affecting
7456 * this device has been detected.
7457 */
7458 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7459 pci_channel_state_t state)
7460 {
7461 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7462 struct net_device *netdev = adapter->netdev;
7463
7464 #ifdef CONFIG_PCI_IOV
7465 struct pci_dev *bdev, *vfdev;
7466 u32 dw0, dw1, dw2, dw3;
7467 int vf, pos;
7468 u16 req_id, pf_func;
7469
7470 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7471 adapter->num_vfs == 0)
7472 goto skip_bad_vf_detection;
7473
7474 bdev = pdev->bus->self;
7475 while (bdev && (bdev->pcie_type != PCI_EXP_TYPE_ROOT_PORT))
7476 bdev = bdev->bus->self;
7477
7478 if (!bdev)
7479 goto skip_bad_vf_detection;
7480
7481 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7482 if (!pos)
7483 goto skip_bad_vf_detection;
7484
7485 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7486 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7487 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7488 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7489
7490 req_id = dw1 >> 16;
7491 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7492 if (!(req_id & 0x0080))
7493 goto skip_bad_vf_detection;
7494
7495 pf_func = req_id & 0x01;
7496 if ((pf_func & 1) == (pdev->devfn & 1)) {
7497 unsigned int device_id;
7498
7499 vf = (req_id & 0x7F) >> 1;
7500 e_dev_err("VF %d has caused a PCIe error\n", vf);
7501 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7502 "%8.8x\tdw3: %8.8x\n",
7503 dw0, dw1, dw2, dw3);
7504 switch (adapter->hw.mac.type) {
7505 case ixgbe_mac_82599EB:
7506 device_id = IXGBE_82599_VF_DEVICE_ID;
7507 break;
7508 case ixgbe_mac_X540:
7509 device_id = IXGBE_X540_VF_DEVICE_ID;
7510 break;
7511 default:
7512 device_id = 0;
7513 break;
7514 }
7515
7516 /* Find the pci device of the offending VF */
7517 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID, device_id, NULL);
7518 while (vfdev) {
7519 if (vfdev->devfn == (req_id & 0xFF))
7520 break;
7521 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID,
7522 device_id, vfdev);
7523 }
7524 /*
7525 * There's a slim chance the VF could have been hot plugged,
7526 * so if it is no longer present we don't need to issue the
7527 * VFLR. Just clean up the AER in that case.
7528 */
7529 if (vfdev) {
7530 e_dev_err("Issuing VFLR to VF %d\n", vf);
7531 pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7532 }
7533
7534 pci_cleanup_aer_uncorrect_error_status(pdev);
7535 }
7536
7537 /*
7538 * Even though the error may have occurred on the other port
7539 * we still need to increment the vf error reference count for
7540 * both ports because the I/O resume function will be called
7541 * for both of them.
7542 */
7543 adapter->vferr_refcount++;
7544
7545 return PCI_ERS_RESULT_RECOVERED;
7546
7547 skip_bad_vf_detection:
7548 #endif /* CONFIG_PCI_IOV */
7549 netif_device_detach(netdev);
7550
7551 if (state == pci_channel_io_perm_failure)
7552 return PCI_ERS_RESULT_DISCONNECT;
7553
7554 if (netif_running(netdev))
7555 ixgbe_down(adapter);
7556 pci_disable_device(pdev);
7557
7558 /* Request a slot reset. */
7559 return PCI_ERS_RESULT_NEED_RESET;
7560 }
7561
7562 /**
7563 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7564 * @pdev: Pointer to PCI device
7565 *
7566 * Restart the card from scratch, as if from a cold-boot.
7567 */
7568 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7569 {
7570 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7571 pci_ers_result_t result;
7572 int err;
7573
7574 if (pci_enable_device_mem(pdev)) {
7575 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7576 result = PCI_ERS_RESULT_DISCONNECT;
7577 } else {
7578 pci_set_master(pdev);
7579 pci_restore_state(pdev);
7580 pci_save_state(pdev);
7581
7582 pci_wake_from_d3(pdev, false);
7583
7584 ixgbe_reset(adapter);
7585 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7586 result = PCI_ERS_RESULT_RECOVERED;
7587 }
7588
7589 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7590 if (err) {
7591 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7592 "failed 0x%0x\n", err);
7593 /* non-fatal, continue */
7594 }
7595
7596 return result;
7597 }
7598
7599 /**
7600 * ixgbe_io_resume - called when traffic can start flowing again.
7601 * @pdev: Pointer to PCI device
7602 *
7603 * This callback is called when the error recovery driver tells us that
7604 * its OK to resume normal operation.
7605 */
7606 static void ixgbe_io_resume(struct pci_dev *pdev)
7607 {
7608 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7609 struct net_device *netdev = adapter->netdev;
7610
7611 #ifdef CONFIG_PCI_IOV
7612 if (adapter->vferr_refcount) {
7613 e_info(drv, "Resuming after VF err\n");
7614 adapter->vferr_refcount--;
7615 return;
7616 }
7617
7618 #endif
7619 if (netif_running(netdev))
7620 ixgbe_up(adapter);
7621
7622 netif_device_attach(netdev);
7623 }
7624
7625 static struct pci_error_handlers ixgbe_err_handler = {
7626 .error_detected = ixgbe_io_error_detected,
7627 .slot_reset = ixgbe_io_slot_reset,
7628 .resume = ixgbe_io_resume,
7629 };
7630
7631 static struct pci_driver ixgbe_driver = {
7632 .name = ixgbe_driver_name,
7633 .id_table = ixgbe_pci_tbl,
7634 .probe = ixgbe_probe,
7635 .remove = __devexit_p(ixgbe_remove),
7636 #ifdef CONFIG_PM
7637 .suspend = ixgbe_suspend,
7638 .resume = ixgbe_resume,
7639 #endif
7640 .shutdown = ixgbe_shutdown,
7641 .err_handler = &ixgbe_err_handler
7642 };
7643
7644 /**
7645 * ixgbe_init_module - Driver Registration Routine
7646 *
7647 * ixgbe_init_module is the first routine called when the driver is
7648 * loaded. All it does is register with the PCI subsystem.
7649 **/
7650 static int __init ixgbe_init_module(void)
7651 {
7652 int ret;
7653 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7654 pr_info("%s\n", ixgbe_copyright);
7655
7656 #ifdef CONFIG_IXGBE_DCA
7657 dca_register_notify(&dca_notifier);
7658 #endif
7659
7660 ret = pci_register_driver(&ixgbe_driver);
7661 return ret;
7662 }
7663
7664 module_init(ixgbe_init_module);
7665
7666 /**
7667 * ixgbe_exit_module - Driver Exit Cleanup Routine
7668 *
7669 * ixgbe_exit_module is called just before the driver is removed
7670 * from memory.
7671 **/
7672 static void __exit ixgbe_exit_module(void)
7673 {
7674 #ifdef CONFIG_IXGBE_DCA
7675 dca_unregister_notify(&dca_notifier);
7676 #endif
7677 pci_unregister_driver(&ixgbe_driver);
7678 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7679 }
7680
7681 #ifdef CONFIG_IXGBE_DCA
7682 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7683 void *p)
7684 {
7685 int ret_val;
7686
7687 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7688 __ixgbe_notify_dca);
7689
7690 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7691 }
7692
7693 #endif /* CONFIG_IXGBE_DCA */
7694
7695 module_exit(ixgbe_exit_module);
7696
7697 /* ixgbe_main.c */
Linux kernel - Deliverables
This page took 0.184059 seconds and 6 git commands to generate.