projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ixgbe: Clean up priority based flow control
[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 8
67 #define BUILD 21
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 /* free the skb */
793 dev_kfree_skb_any(tx_buffer->skb);
794
795 /* unmap skb header data */
796 dma_unmap_single(tx_ring->dev,
797 dma_unmap_addr(tx_buffer, dma),
798 dma_unmap_len(tx_buffer, len),
799 DMA_TO_DEVICE);
800
801 /* clear tx_buffer data */
802 tx_buffer->skb = NULL;
803 dma_unmap_len_set(tx_buffer, len, 0);
804
805 /* unmap remaining buffers */
806 while (tx_desc != eop_desc) {
807 tx_buffer++;
808 tx_desc++;
809 i++;
810 if (unlikely(!i)) {
811 i -= tx_ring->count;
812 tx_buffer = tx_ring->tx_buffer_info;
813 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
814 }
815
816 /* unmap any remaining paged data */
817 if (dma_unmap_len(tx_buffer, len)) {
818 dma_unmap_page(tx_ring->dev,
819 dma_unmap_addr(tx_buffer, dma),
820 dma_unmap_len(tx_buffer, len),
821 DMA_TO_DEVICE);
822 dma_unmap_len_set(tx_buffer, len, 0);
823 }
824 }
825
826 /* move us one more past the eop_desc for start of next pkt */
827 tx_buffer++;
828 tx_desc++;
829 i++;
830 if (unlikely(!i)) {
831 i -= tx_ring->count;
832 tx_buffer = tx_ring->tx_buffer_info;
833 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
834 }
835
836 /* issue prefetch for next Tx descriptor */
837 prefetch(tx_desc);
838
839 /* update budget accounting */
840 budget--;
841 } while (likely(budget));
842
843 i += tx_ring->count;
844 tx_ring->next_to_clean = i;
845 u64_stats_update_begin(&tx_ring->syncp);
846 tx_ring->stats.bytes += total_bytes;
847 tx_ring->stats.packets += total_packets;
848 u64_stats_update_end(&tx_ring->syncp);
849 q_vector->tx.total_bytes += total_bytes;
850 q_vector->tx.total_packets += total_packets;
851
852 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
853 /* schedule immediate reset if we believe we hung */
854 struct ixgbe_hw *hw = &adapter->hw;
855 e_err(drv, "Detected Tx Unit Hang\n"
856 " Tx Queue <%d>\n"
857 " TDH, TDT <%x>, <%x>\n"
858 " next_to_use <%x>\n"
859 " next_to_clean <%x>\n"
860 "tx_buffer_info[next_to_clean]\n"
861 " time_stamp <%lx>\n"
862 " jiffies <%lx>\n",
863 tx_ring->queue_index,
864 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
865 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
866 tx_ring->next_to_use, i,
867 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
868
869 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
870
871 e_info(probe,
872 "tx hang %d detected on queue %d, resetting adapter\n",
873 adapter->tx_timeout_count + 1, tx_ring->queue_index);
874
875 /* schedule immediate reset if we believe we hung */
876 ixgbe_tx_timeout_reset(adapter);
877
878 /* the adapter is about to reset, no point in enabling stuff */
879 return true;
880 }
881
882 netdev_tx_completed_queue(txring_txq(tx_ring),
883 total_packets, total_bytes);
884
885 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
886 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
887 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
888 /* Make sure that anybody stopping the queue after this
889 * sees the new next_to_clean.
890 */
891 smp_mb();
892 if (__netif_subqueue_stopped(tx_ring->netdev,
893 tx_ring->queue_index)
894 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
895 netif_wake_subqueue(tx_ring->netdev,
896 tx_ring->queue_index);
897 ++tx_ring->tx_stats.restart_queue;
898 }
899 }
900
901 return !!budget;
902 }
903
904 #ifdef CONFIG_IXGBE_DCA
905 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
906 struct ixgbe_ring *tx_ring,
907 int cpu)
908 {
909 struct ixgbe_hw *hw = &adapter->hw;
910 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
911 u16 reg_offset;
912
913 switch (hw->mac.type) {
914 case ixgbe_mac_82598EB:
915 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
916 break;
917 case ixgbe_mac_82599EB:
918 case ixgbe_mac_X540:
919 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
920 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
921 break;
922 default:
923 /* for unknown hardware do not write register */
924 return;
925 }
926
927 /*
928 * We can enable relaxed ordering for reads, but not writes when
929 * DCA is enabled. This is due to a known issue in some chipsets
930 * which will cause the DCA tag to be cleared.
931 */
932 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
933 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
934 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
935
936 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
937 }
938
939 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
940 struct ixgbe_ring *rx_ring,
941 int cpu)
942 {
943 struct ixgbe_hw *hw = &adapter->hw;
944 u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
945 u8 reg_idx = rx_ring->reg_idx;
946
947
948 switch (hw->mac.type) {
949 case ixgbe_mac_82599EB:
950 case ixgbe_mac_X540:
951 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
952 break;
953 default:
954 break;
955 }
956
957 /*
958 * We can enable relaxed ordering for reads, but not writes when
959 * DCA is enabled. This is due to a known issue in some chipsets
960 * which will cause the DCA tag to be cleared.
961 */
962 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
963 IXGBE_DCA_RXCTRL_DATA_DCA_EN |
964 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
965
966 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
967 }
968
969 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
970 {
971 struct ixgbe_adapter *adapter = q_vector->adapter;
972 struct ixgbe_ring *ring;
973 int cpu = get_cpu();
974
975 if (q_vector->cpu == cpu)
976 goto out_no_update;
977
978 ixgbe_for_each_ring(ring, q_vector->tx)
979 ixgbe_update_tx_dca(adapter, ring, cpu);
980
981 ixgbe_for_each_ring(ring, q_vector->rx)
982 ixgbe_update_rx_dca(adapter, ring, cpu);
983
984 q_vector->cpu = cpu;
985 out_no_update:
986 put_cpu();
987 }
988
989 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
990 {
991 int num_q_vectors;
992 int i;
993
994 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
995 return;
996
997 /* always use CB2 mode, difference is masked in the CB driver */
998 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
999
1000 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
1001 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1002 else
1003 num_q_vectors = 1;
1004
1005 for (i = 0; i < num_q_vectors; i++) {
1006 adapter->q_vector[i]->cpu = -1;
1007 ixgbe_update_dca(adapter->q_vector[i]);
1008 }
1009 }
1010
1011 static int __ixgbe_notify_dca(struct device *dev, void *data)
1012 {
1013 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1014 unsigned long event = *(unsigned long *)data;
1015
1016 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1017 return 0;
1018
1019 switch (event) {
1020 case DCA_PROVIDER_ADD:
1021 /* if we're already enabled, don't do it again */
1022 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1023 break;
1024 if (dca_add_requester(dev) == 0) {
1025 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1026 ixgbe_setup_dca(adapter);
1027 break;
1028 }
1029 /* Fall Through since DCA is disabled. */
1030 case DCA_PROVIDER_REMOVE:
1031 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1032 dca_remove_requester(dev);
1033 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1034 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1035 }
1036 break;
1037 }
1038
1039 return 0;
1040 }
1041
1042 #endif /* CONFIG_IXGBE_DCA */
1043 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1044 union ixgbe_adv_rx_desc *rx_desc,
1045 struct sk_buff *skb)
1046 {
1047 if (ring->netdev->features & NETIF_F_RXHASH)
1048 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1049 }
1050
1051 #ifdef IXGBE_FCOE
1052 /**
1053 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1054 * @adapter: address of board private structure
1055 * @rx_desc: advanced rx descriptor
1056 *
1057 * Returns : true if it is FCoE pkt
1058 */
1059 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_adapter *adapter,
1060 union ixgbe_adv_rx_desc *rx_desc)
1061 {
1062 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1063
1064 return (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
1065 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1066 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1067 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1068 }
1069
1070 #endif /* IXGBE_FCOE */
1071 /**
1072 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1073 * @ring: structure containing ring specific data
1074 * @rx_desc: current Rx descriptor being processed
1075 * @skb: skb currently being received and modified
1076 **/
1077 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1078 union ixgbe_adv_rx_desc *rx_desc,
1079 struct sk_buff *skb)
1080 {
1081 skb_checksum_none_assert(skb);
1082
1083 /* Rx csum disabled */
1084 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1085 return;
1086
1087 /* if IP and error */
1088 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1089 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1090 ring->rx_stats.csum_err++;
1091 return;
1092 }
1093
1094 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1095 return;
1096
1097 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1098 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1099
1100 /*
1101 * 82599 errata, UDP frames with a 0 checksum can be marked as
1102 * checksum errors.
1103 */
1104 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1105 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1106 return;
1107
1108 ring->rx_stats.csum_err++;
1109 return;
1110 }
1111
1112 /* It must be a TCP or UDP packet with a valid checksum */
1113 skb->ip_summed = CHECKSUM_UNNECESSARY;
1114 }
1115
1116 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1117 {
1118 rx_ring->next_to_use = val;
1119
1120 /* update next to alloc since we have filled the ring */
1121 rx_ring->next_to_alloc = val;
1122 /*
1123 * Force memory writes to complete before letting h/w
1124 * know there are new descriptors to fetch. (Only
1125 * applicable for weak-ordered memory model archs,
1126 * such as IA-64).
1127 */
1128 wmb();
1129 writel(val, rx_ring->tail);
1130 }
1131
1132 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1133 struct ixgbe_rx_buffer *bi)
1134 {
1135 struct page *page = bi->page;
1136 dma_addr_t dma = bi->dma;
1137
1138 /* since we are recycling buffers we should seldom need to alloc */
1139 if (likely(dma))
1140 return true;
1141
1142 /* alloc new page for storage */
1143 if (likely(!page)) {
1144 page = alloc_pages(GFP_ATOMIC | __GFP_COLD,
1145 ixgbe_rx_pg_order(rx_ring));
1146 if (unlikely(!page)) {
1147 rx_ring->rx_stats.alloc_rx_page_failed++;
1148 return false;
1149 }
1150 bi->page = page;
1151 }
1152
1153 /* map page for use */
1154 dma = dma_map_page(rx_ring->dev, page, 0,
1155 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1156
1157 /*
1158 * if mapping failed free memory back to system since
1159 * there isn't much point in holding memory we can't use
1160 */
1161 if (dma_mapping_error(rx_ring->dev, dma)) {
1162 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1163 bi->page = NULL;
1164
1165 rx_ring->rx_stats.alloc_rx_page_failed++;
1166 return false;
1167 }
1168
1169 bi->dma = dma;
1170 bi->page_offset ^= ixgbe_rx_bufsz(rx_ring);
1171
1172 return true;
1173 }
1174
1175 /**
1176 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1177 * @rx_ring: ring to place buffers on
1178 * @cleaned_count: number of buffers to replace
1179 **/
1180 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1181 {
1182 union ixgbe_adv_rx_desc *rx_desc;
1183 struct ixgbe_rx_buffer *bi;
1184 u16 i = rx_ring->next_to_use;
1185
1186 /* nothing to do */
1187 if (!cleaned_count)
1188 return;
1189
1190 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1191 bi = &rx_ring->rx_buffer_info[i];
1192 i -= rx_ring->count;
1193
1194 do {
1195 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1196 break;
1197
1198 /*
1199 * Refresh the desc even if buffer_addrs didn't change
1200 * because each write-back erases this info.
1201 */
1202 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1203
1204 rx_desc++;
1205 bi++;
1206 i++;
1207 if (unlikely(!i)) {
1208 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1209 bi = rx_ring->rx_buffer_info;
1210 i -= rx_ring->count;
1211 }
1212
1213 /* clear the hdr_addr for the next_to_use descriptor */
1214 rx_desc->read.hdr_addr = 0;
1215
1216 cleaned_count--;
1217 } while (cleaned_count);
1218
1219 i += rx_ring->count;
1220
1221 if (rx_ring->next_to_use != i)
1222 ixgbe_release_rx_desc(rx_ring, i);
1223 }
1224
1225 /**
1226 * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1227 * @data: pointer to the start of the headers
1228 * @max_len: total length of section to find headers in
1229 *
1230 * This function is meant to determine the length of headers that will
1231 * be recognized by hardware for LRO, GRO, and RSC offloads. The main
1232 * motivation of doing this is to only perform one pull for IPv4 TCP
1233 * packets so that we can do basic things like calculating the gso_size
1234 * based on the average data per packet.
1235 **/
1236 static unsigned int ixgbe_get_headlen(unsigned char *data,
1237 unsigned int max_len)
1238 {
1239 union {
1240 unsigned char *network;
1241 /* l2 headers */
1242 struct ethhdr *eth;
1243 struct vlan_hdr *vlan;
1244 /* l3 headers */
1245 struct iphdr *ipv4;
1246 } hdr;
1247 __be16 protocol;
1248 u8 nexthdr = 0; /* default to not TCP */
1249 u8 hlen;
1250
1251 /* this should never happen, but better safe than sorry */
1252 if (max_len < ETH_HLEN)
1253 return max_len;
1254
1255 /* initialize network frame pointer */
1256 hdr.network = data;
1257
1258 /* set first protocol and move network header forward */
1259 protocol = hdr.eth->h_proto;
1260 hdr.network += ETH_HLEN;
1261
1262 /* handle any vlan tag if present */
1263 if (protocol == __constant_htons(ETH_P_8021Q)) {
1264 if ((hdr.network - data) > (max_len - VLAN_HLEN))
1265 return max_len;
1266
1267 protocol = hdr.vlan->h_vlan_encapsulated_proto;
1268 hdr.network += VLAN_HLEN;
1269 }
1270
1271 /* handle L3 protocols */
1272 if (protocol == __constant_htons(ETH_P_IP)) {
1273 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1274 return max_len;
1275
1276 /* access ihl as a u8 to avoid unaligned access on ia64 */
1277 hlen = (hdr.network[0] & 0x0F) << 2;
1278
1279 /* verify hlen meets minimum size requirements */
1280 if (hlen < sizeof(struct iphdr))
1281 return hdr.network - data;
1282
1283 /* record next protocol */
1284 nexthdr = hdr.ipv4->protocol;
1285 hdr.network += hlen;
1286 #ifdef IXGBE_FCOE
1287 } else if (protocol == __constant_htons(ETH_P_FCOE)) {
1288 if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1289 return max_len;
1290 hdr.network += FCOE_HEADER_LEN;
1291 #endif
1292 } else {
1293 return hdr.network - data;
1294 }
1295
1296 /* finally sort out TCP */
1297 if (nexthdr == IPPROTO_TCP) {
1298 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1299 return max_len;
1300
1301 /* access doff as a u8 to avoid unaligned access on ia64 */
1302 hlen = (hdr.network[12] & 0xF0) >> 2;
1303
1304 /* verify hlen meets minimum size requirements */
1305 if (hlen < sizeof(struct tcphdr))
1306 return hdr.network - data;
1307
1308 hdr.network += hlen;
1309 }
1310
1311 /*
1312 * If everything has gone correctly hdr.network should be the
1313 * data section of the packet and will be the end of the header.
1314 * If not then it probably represents the end of the last recognized
1315 * header.
1316 */
1317 if ((hdr.network - data) < max_len)
1318 return hdr.network - data;
1319 else
1320 return max_len;
1321 }
1322
1323 static void ixgbe_get_rsc_cnt(struct ixgbe_ring *rx_ring,
1324 union ixgbe_adv_rx_desc *rx_desc,
1325 struct sk_buff *skb)
1326 {
1327 __le32 rsc_enabled;
1328 u32 rsc_cnt;
1329
1330 if (!ring_is_rsc_enabled(rx_ring))
1331 return;
1332
1333 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1334 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1335
1336 /* If this is an RSC frame rsc_cnt should be non-zero */
1337 if (!rsc_enabled)
1338 return;
1339
1340 rsc_cnt = le32_to_cpu(rsc_enabled);
1341 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1342
1343 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1344 }
1345
1346 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1347 struct sk_buff *skb)
1348 {
1349 u16 hdr_len = skb_headlen(skb);
1350
1351 /* set gso_size to avoid messing up TCP MSS */
1352 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1353 IXGBE_CB(skb)->append_cnt);
1354 }
1355
1356 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1357 struct sk_buff *skb)
1358 {
1359 /* if append_cnt is 0 then frame is not RSC */
1360 if (!IXGBE_CB(skb)->append_cnt)
1361 return;
1362
1363 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1364 rx_ring->rx_stats.rsc_flush++;
1365
1366 ixgbe_set_rsc_gso_size(rx_ring, skb);
1367
1368 /* gso_size is computed using append_cnt so always clear it last */
1369 IXGBE_CB(skb)->append_cnt = 0;
1370 }
1371
1372 /**
1373 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1374 * @rx_ring: rx descriptor ring packet is being transacted on
1375 * @rx_desc: pointer to the EOP Rx descriptor
1376 * @skb: pointer to current skb being populated
1377 *
1378 * This function checks the ring, descriptor, and packet information in
1379 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1380 * other fields within the skb.
1381 **/
1382 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1383 union ixgbe_adv_rx_desc *rx_desc,
1384 struct sk_buff *skb)
1385 {
1386 ixgbe_update_rsc_stats(rx_ring, skb);
1387
1388 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1389
1390 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1391
1392 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1393 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1394 __vlan_hwaccel_put_tag(skb, vid);
1395 }
1396
1397 skb_record_rx_queue(skb, rx_ring->queue_index);
1398
1399 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1400 }
1401
1402 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1403 struct sk_buff *skb)
1404 {
1405 struct ixgbe_adapter *adapter = q_vector->adapter;
1406
1407 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1408 napi_gro_receive(&q_vector->napi, skb);
1409 else
1410 netif_rx(skb);
1411 }
1412
1413 /**
1414 * ixgbe_is_non_eop - process handling of non-EOP buffers
1415 * @rx_ring: Rx ring being processed
1416 * @rx_desc: Rx descriptor for current buffer
1417 * @skb: Current socket buffer containing buffer in progress
1418 *
1419 * This function updates next to clean. If the buffer is an EOP buffer
1420 * this function exits returning false, otherwise it will place the
1421 * sk_buff in the next buffer to be chained and return true indicating
1422 * that this is in fact a non-EOP buffer.
1423 **/
1424 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1425 union ixgbe_adv_rx_desc *rx_desc,
1426 struct sk_buff *skb)
1427 {
1428 u32 ntc = rx_ring->next_to_clean + 1;
1429
1430 /* fetch, update, and store next to clean */
1431 ntc = (ntc < rx_ring->count) ? ntc : 0;
1432 rx_ring->next_to_clean = ntc;
1433
1434 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1435
1436 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1437 return false;
1438
1439 /* append_cnt indicates packet is RSC, if so fetch nextp */
1440 if (IXGBE_CB(skb)->append_cnt) {
1441 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1442 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1443 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1444 }
1445
1446 /* place skb in next buffer to be received */
1447 rx_ring->rx_buffer_info[ntc].skb = skb;
1448 rx_ring->rx_stats.non_eop_descs++;
1449
1450 return true;
1451 }
1452
1453 /**
1454 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1455 * @rx_ring: rx descriptor ring packet is being transacted on
1456 * @rx_desc: pointer to the EOP Rx descriptor
1457 * @skb: pointer to current skb being fixed
1458 *
1459 * Check for corrupted packet headers caused by senders on the local L2
1460 * embedded NIC switch not setting up their Tx Descriptors right. These
1461 * should be very rare.
1462 *
1463 * Also address the case where we are pulling data in on pages only
1464 * and as such no data is present in the skb header.
1465 *
1466 * In addition if skb is not at least 60 bytes we need to pad it so that
1467 * it is large enough to qualify as a valid Ethernet frame.
1468 *
1469 * Returns true if an error was encountered and skb was freed.
1470 **/
1471 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1472 union ixgbe_adv_rx_desc *rx_desc,
1473 struct sk_buff *skb)
1474 {
1475 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1476 struct net_device *netdev = rx_ring->netdev;
1477 unsigned char *va;
1478 unsigned int pull_len;
1479
1480 /* if the page was released unmap it, else just sync our portion */
1481 if (unlikely(IXGBE_CB(skb)->page_released)) {
1482 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1483 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1484 IXGBE_CB(skb)->page_released = false;
1485 } else {
1486 dma_sync_single_range_for_cpu(rx_ring->dev,
1487 IXGBE_CB(skb)->dma,
1488 frag->page_offset,
1489 ixgbe_rx_bufsz(rx_ring),
1490 DMA_FROM_DEVICE);
1491 }
1492 IXGBE_CB(skb)->dma = 0;
1493
1494 /* verify that the packet does not have any known errors */
1495 if (unlikely(ixgbe_test_staterr(rx_desc,
1496 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1497 !(netdev->features & NETIF_F_RXALL))) {
1498 dev_kfree_skb_any(skb);
1499 return true;
1500 }
1501
1502 /*
1503 * it is valid to use page_address instead of kmap since we are
1504 * working with pages allocated out of the lomem pool per
1505 * alloc_page(GFP_ATOMIC)
1506 */
1507 va = skb_frag_address(frag);
1508
1509 /*
1510 * we need the header to contain the greater of either ETH_HLEN or
1511 * 60 bytes if the skb->len is less than 60 for skb_pad.
1512 */
1513 pull_len = skb_frag_size(frag);
1514 if (pull_len > 256)
1515 pull_len = ixgbe_get_headlen(va, pull_len);
1516
1517 /* align pull length to size of long to optimize memcpy performance */
1518 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1519
1520 /* update all of the pointers */
1521 skb_frag_size_sub(frag, pull_len);
1522 frag->page_offset += pull_len;
1523 skb->data_len -= pull_len;
1524 skb->tail += pull_len;
1525
1526 /*
1527 * if we sucked the frag empty then we should free it,
1528 * if there are other frags here something is screwed up in hardware
1529 */
1530 if (skb_frag_size(frag) == 0) {
1531 BUG_ON(skb_shinfo(skb)->nr_frags != 1);
1532 skb_shinfo(skb)->nr_frags = 0;
1533 __skb_frag_unref(frag);
1534 skb->truesize -= ixgbe_rx_bufsz(rx_ring);
1535 }
1536
1537 /* if skb_pad returns an error the skb was freed */
1538 if (unlikely(skb->len < 60)) {
1539 int pad_len = 60 - skb->len;
1540
1541 if (skb_pad(skb, pad_len))
1542 return true;
1543 __skb_put(skb, pad_len);
1544 }
1545
1546 return false;
1547 }
1548
1549 /**
1550 * ixgbe_can_reuse_page - determine if we can reuse a page
1551 * @rx_buffer: pointer to rx_buffer containing the page we want to reuse
1552 *
1553 * Returns true if page can be reused in another Rx buffer
1554 **/
1555 static inline bool ixgbe_can_reuse_page(struct ixgbe_rx_buffer *rx_buffer)
1556 {
1557 struct page *page = rx_buffer->page;
1558
1559 /* if we are only owner of page and it is local we can reuse it */
1560 return likely(page_count(page) == 1) &&
1561 likely(page_to_nid(page) == numa_node_id());
1562 }
1563
1564 /**
1565 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1566 * @rx_ring: rx descriptor ring to store buffers on
1567 * @old_buff: donor buffer to have page reused
1568 *
1569 * Syncronizes page for reuse by the adapter
1570 **/
1571 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1572 struct ixgbe_rx_buffer *old_buff)
1573 {
1574 struct ixgbe_rx_buffer *new_buff;
1575 u16 nta = rx_ring->next_to_alloc;
1576 u16 bufsz = ixgbe_rx_bufsz(rx_ring);
1577
1578 new_buff = &rx_ring->rx_buffer_info[nta];
1579
1580 /* update, and store next to alloc */
1581 nta++;
1582 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1583
1584 /* transfer page from old buffer to new buffer */
1585 new_buff->page = old_buff->page;
1586 new_buff->dma = old_buff->dma;
1587
1588 /* flip page offset to other buffer and store to new_buff */
1589 new_buff->page_offset = old_buff->page_offset ^ bufsz;
1590
1591 /* sync the buffer for use by the device */
1592 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1593 new_buff->page_offset, bufsz,
1594 DMA_FROM_DEVICE);
1595
1596 /* bump ref count on page before it is given to the stack */
1597 get_page(new_buff->page);
1598 }
1599
1600 /**
1601 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1602 * @rx_ring: rx descriptor ring to transact packets on
1603 * @rx_buffer: buffer containing page to add
1604 * @rx_desc: descriptor containing length of buffer written by hardware
1605 * @skb: sk_buff to place the data into
1606 *
1607 * This function is based on skb_add_rx_frag. I would have used that
1608 * function however it doesn't handle the truesize case correctly since we
1609 * are allocating more memory than might be used for a single receive.
1610 **/
1611 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1612 struct ixgbe_rx_buffer *rx_buffer,
1613 struct sk_buff *skb, int size)
1614 {
1615 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1616 rx_buffer->page, rx_buffer->page_offset,
1617 size);
1618 skb->len += size;
1619 skb->data_len += size;
1620 skb->truesize += ixgbe_rx_bufsz(rx_ring);
1621 }
1622
1623 /**
1624 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1625 * @q_vector: structure containing interrupt and ring information
1626 * @rx_ring: rx descriptor ring to transact packets on
1627 * @budget: Total limit on number of packets to process
1628 *
1629 * This function provides a "bounce buffer" approach to Rx interrupt
1630 * processing. The advantage to this is that on systems that have
1631 * expensive overhead for IOMMU access this provides a means of avoiding
1632 * it by maintaining the mapping of the page to the syste.
1633 *
1634 * Returns true if all work is completed without reaching budget
1635 **/
1636 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1637 struct ixgbe_ring *rx_ring,
1638 int budget)
1639 {
1640 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1641 #ifdef IXGBE_FCOE
1642 struct ixgbe_adapter *adapter = q_vector->adapter;
1643 int ddp_bytes = 0;
1644 #endif /* IXGBE_FCOE */
1645 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1646
1647 do {
1648 struct ixgbe_rx_buffer *rx_buffer;
1649 union ixgbe_adv_rx_desc *rx_desc;
1650 struct sk_buff *skb;
1651 struct page *page;
1652 u16 ntc;
1653
1654 /* return some buffers to hardware, one at a time is too slow */
1655 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1656 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1657 cleaned_count = 0;
1658 }
1659
1660 ntc = rx_ring->next_to_clean;
1661 rx_desc = IXGBE_RX_DESC(rx_ring, ntc);
1662 rx_buffer = &rx_ring->rx_buffer_info[ntc];
1663
1664 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
1665 break;
1666
1667 /*
1668 * This memory barrier is needed to keep us from reading
1669 * any other fields out of the rx_desc until we know the
1670 * RXD_STAT_DD bit is set
1671 */
1672 rmb();
1673
1674 page = rx_buffer->page;
1675 prefetchw(page);
1676
1677 skb = rx_buffer->skb;
1678
1679 if (likely(!skb)) {
1680 void *page_addr = page_address(page) +
1681 rx_buffer->page_offset;
1682
1683 /* prefetch first cache line of first page */
1684 prefetch(page_addr);
1685 #if L1_CACHE_BYTES < 128
1686 prefetch(page_addr + L1_CACHE_BYTES);
1687 #endif
1688
1689 /* allocate a skb to store the frags */
1690 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1691 IXGBE_RX_HDR_SIZE);
1692 if (unlikely(!skb)) {
1693 rx_ring->rx_stats.alloc_rx_buff_failed++;
1694 break;
1695 }
1696
1697 /*
1698 * we will be copying header into skb->data in
1699 * pskb_may_pull so it is in our interest to prefetch
1700 * it now to avoid a possible cache miss
1701 */
1702 prefetchw(skb->data);
1703
1704 /*
1705 * Delay unmapping of the first packet. It carries the
1706 * header information, HW may still access the header
1707 * after the writeback. Only unmap it when EOP is
1708 * reached
1709 */
1710 IXGBE_CB(skb)->dma = rx_buffer->dma;
1711 } else {
1712 /* we are reusing so sync this buffer for CPU use */
1713 dma_sync_single_range_for_cpu(rx_ring->dev,
1714 rx_buffer->dma,
1715 rx_buffer->page_offset,
1716 ixgbe_rx_bufsz(rx_ring),
1717 DMA_FROM_DEVICE);
1718 }
1719
1720 /* pull page into skb */
1721 ixgbe_add_rx_frag(rx_ring, rx_buffer, skb,
1722 le16_to_cpu(rx_desc->wb.upper.length));
1723
1724 if (ixgbe_can_reuse_page(rx_buffer)) {
1725 /* hand second half of page back to the ring */
1726 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1727 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1728 /* the page has been released from the ring */
1729 IXGBE_CB(skb)->page_released = true;
1730 } else {
1731 /* we are not reusing the buffer so unmap it */
1732 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1733 ixgbe_rx_pg_size(rx_ring),
1734 DMA_FROM_DEVICE);
1735 }
1736
1737 /* clear contents of buffer_info */
1738 rx_buffer->skb = NULL;
1739 rx_buffer->dma = 0;
1740 rx_buffer->page = NULL;
1741
1742 ixgbe_get_rsc_cnt(rx_ring, rx_desc, skb);
1743
1744 cleaned_count++;
1745
1746 /* place incomplete frames back on ring for completion */
1747 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
1748 continue;
1749
1750 /* verify the packet layout is correct */
1751 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
1752 continue;
1753
1754 /* probably a little skewed due to removing CRC */
1755 total_rx_bytes += skb->len;
1756 total_rx_packets++;
1757
1758 /* populate checksum, timestamp, VLAN, and protocol */
1759 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
1760
1761 #ifdef IXGBE_FCOE
1762 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1763 if (ixgbe_rx_is_fcoe(adapter, rx_desc)) {
1764 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1765 if (!ddp_bytes) {
1766 dev_kfree_skb_any(skb);
1767 continue;
1768 }
1769 }
1770
1771 #endif /* IXGBE_FCOE */
1772 ixgbe_rx_skb(q_vector, skb);
1773
1774 /* update budget accounting */
1775 budget--;
1776 } while (likely(budget));
1777
1778 #ifdef IXGBE_FCOE
1779 /* include DDPed FCoE data */
1780 if (ddp_bytes > 0) {
1781 unsigned int mss;
1782
1783 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1784 sizeof(struct fc_frame_header) -
1785 sizeof(struct fcoe_crc_eof);
1786 if (mss > 512)
1787 mss &= ~511;
1788 total_rx_bytes += ddp_bytes;
1789 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1790 }
1791
1792 #endif /* IXGBE_FCOE */
1793 u64_stats_update_begin(&rx_ring->syncp);
1794 rx_ring->stats.packets += total_rx_packets;
1795 rx_ring->stats.bytes += total_rx_bytes;
1796 u64_stats_update_end(&rx_ring->syncp);
1797 q_vector->rx.total_packets += total_rx_packets;
1798 q_vector->rx.total_bytes += total_rx_bytes;
1799
1800 if (cleaned_count)
1801 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1802
1803 return !!budget;
1804 }
1805
1806 /**
1807 * ixgbe_configure_msix - Configure MSI-X hardware
1808 * @adapter: board private structure
1809 *
1810 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1811 * interrupts.
1812 **/
1813 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1814 {
1815 struct ixgbe_q_vector *q_vector;
1816 int q_vectors, v_idx;
1817 u32 mask;
1818
1819 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1820
1821 /* Populate MSIX to EITR Select */
1822 if (adapter->num_vfs > 32) {
1823 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1824 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1825 }
1826
1827 /*
1828 * Populate the IVAR table and set the ITR values to the
1829 * corresponding register.
1830 */
1831 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1832 struct ixgbe_ring *ring;
1833 q_vector = adapter->q_vector[v_idx];
1834
1835 ixgbe_for_each_ring(ring, q_vector->rx)
1836 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1837
1838 ixgbe_for_each_ring(ring, q_vector->tx)
1839 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1840
1841 if (q_vector->tx.ring && !q_vector->rx.ring) {
1842 /* tx only vector */
1843 if (adapter->tx_itr_setting == 1)
1844 q_vector->itr = IXGBE_10K_ITR;
1845 else
1846 q_vector->itr = adapter->tx_itr_setting;
1847 } else {
1848 /* rx or rx/tx vector */
1849 if (adapter->rx_itr_setting == 1)
1850 q_vector->itr = IXGBE_20K_ITR;
1851 else
1852 q_vector->itr = adapter->rx_itr_setting;
1853 }
1854
1855 ixgbe_write_eitr(q_vector);
1856 }
1857
1858 switch (adapter->hw.mac.type) {
1859 case ixgbe_mac_82598EB:
1860 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1861 v_idx);
1862 break;
1863 case ixgbe_mac_82599EB:
1864 case ixgbe_mac_X540:
1865 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1866 break;
1867 default:
1868 break;
1869 }
1870 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1871
1872 /* set up to autoclear timer, and the vectors */
1873 mask = IXGBE_EIMS_ENABLE_MASK;
1874 mask &= ~(IXGBE_EIMS_OTHER |
1875 IXGBE_EIMS_MAILBOX |
1876 IXGBE_EIMS_LSC);
1877
1878 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1879 }
1880
1881 enum latency_range {
1882 lowest_latency = 0,
1883 low_latency = 1,
1884 bulk_latency = 2,
1885 latency_invalid = 255
1886 };
1887
1888 /**
1889 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1890 * @q_vector: structure containing interrupt and ring information
1891 * @ring_container: structure containing ring performance data
1892 *
1893 * Stores a new ITR value based on packets and byte
1894 * counts during the last interrupt. The advantage of per interrupt
1895 * computation is faster updates and more accurate ITR for the current
1896 * traffic pattern. Constants in this function were computed
1897 * based on theoretical maximum wire speed and thresholds were set based
1898 * on testing data as well as attempting to minimize response time
1899 * while increasing bulk throughput.
1900 * this functionality is controlled by the InterruptThrottleRate module
1901 * parameter (see ixgbe_param.c)
1902 **/
1903 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1904 struct ixgbe_ring_container *ring_container)
1905 {
1906 int bytes = ring_container->total_bytes;
1907 int packets = ring_container->total_packets;
1908 u32 timepassed_us;
1909 u64 bytes_perint;
1910 u8 itr_setting = ring_container->itr;
1911
1912 if (packets == 0)
1913 return;
1914
1915 /* simple throttlerate management
1916 * 0-10MB/s lowest (100000 ints/s)
1917 * 10-20MB/s low (20000 ints/s)
1918 * 20-1249MB/s bulk (8000 ints/s)
1919 */
1920 /* what was last interrupt timeslice? */
1921 timepassed_us = q_vector->itr >> 2;
1922 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1923
1924 switch (itr_setting) {
1925 case lowest_latency:
1926 if (bytes_perint > 10)
1927 itr_setting = low_latency;
1928 break;
1929 case low_latency:
1930 if (bytes_perint > 20)
1931 itr_setting = bulk_latency;
1932 else if (bytes_perint <= 10)
1933 itr_setting = lowest_latency;
1934 break;
1935 case bulk_latency:
1936 if (bytes_perint <= 20)
1937 itr_setting = low_latency;
1938 break;
1939 }
1940
1941 /* clear work counters since we have the values we need */
1942 ring_container->total_bytes = 0;
1943 ring_container->total_packets = 0;
1944
1945 /* write updated itr to ring container */
1946 ring_container->itr = itr_setting;
1947 }
1948
1949 /**
1950 * ixgbe_write_eitr - write EITR register in hardware specific way
1951 * @q_vector: structure containing interrupt and ring information
1952 *
1953 * This function is made to be called by ethtool and by the driver
1954 * when it needs to update EITR registers at runtime. Hardware
1955 * specific quirks/differences are taken care of here.
1956 */
1957 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1958 {
1959 struct ixgbe_adapter *adapter = q_vector->adapter;
1960 struct ixgbe_hw *hw = &adapter->hw;
1961 int v_idx = q_vector->v_idx;
1962 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1963
1964 switch (adapter->hw.mac.type) {
1965 case ixgbe_mac_82598EB:
1966 /* must write high and low 16 bits to reset counter */
1967 itr_reg |= (itr_reg << 16);
1968 break;
1969 case ixgbe_mac_82599EB:
1970 case ixgbe_mac_X540:
1971 /*
1972 * set the WDIS bit to not clear the timer bits and cause an
1973 * immediate assertion of the interrupt
1974 */
1975 itr_reg |= IXGBE_EITR_CNT_WDIS;
1976 break;
1977 default:
1978 break;
1979 }
1980 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1981 }
1982
1983 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1984 {
1985 u32 new_itr = q_vector->itr;
1986 u8 current_itr;
1987
1988 ixgbe_update_itr(q_vector, &q_vector->tx);
1989 ixgbe_update_itr(q_vector, &q_vector->rx);
1990
1991 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1992
1993 switch (current_itr) {
1994 /* counts and packets in update_itr are dependent on these numbers */
1995 case lowest_latency:
1996 new_itr = IXGBE_100K_ITR;
1997 break;
1998 case low_latency:
1999 new_itr = IXGBE_20K_ITR;
2000 break;
2001 case bulk_latency:
2002 new_itr = IXGBE_8K_ITR;
2003 break;
2004 default:
2005 break;
2006 }
2007
2008 if (new_itr != q_vector->itr) {
2009 /* do an exponential smoothing */
2010 new_itr = (10 * new_itr * q_vector->itr) /
2011 ((9 * new_itr) + q_vector->itr);
2012
2013 /* save the algorithm value here */
2014 q_vector->itr = new_itr;
2015
2016 ixgbe_write_eitr(q_vector);
2017 }
2018 }
2019
2020 /**
2021 * ixgbe_check_overtemp_subtask - check for over temperature
2022 * @adapter: pointer to adapter
2023 **/
2024 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2025 {
2026 struct ixgbe_hw *hw = &adapter->hw;
2027 u32 eicr = adapter->interrupt_event;
2028
2029 if (test_bit(__IXGBE_DOWN, &adapter->state))
2030 return;
2031
2032 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2033 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2034 return;
2035
2036 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2037
2038 switch (hw->device_id) {
2039 case IXGBE_DEV_ID_82599_T3_LOM:
2040 /*
2041 * Since the warning interrupt is for both ports
2042 * we don't have to check if:
2043 * - This interrupt wasn't for our port.
2044 * - We may have missed the interrupt so always have to
2045 * check if we got a LSC
2046 */
2047 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2048 !(eicr & IXGBE_EICR_LSC))
2049 return;
2050
2051 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2052 u32 autoneg;
2053 bool link_up = false;
2054
2055 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2056
2057 if (link_up)
2058 return;
2059 }
2060
2061 /* Check if this is not due to overtemp */
2062 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2063 return;
2064
2065 break;
2066 default:
2067 if (!(eicr & IXGBE_EICR_GPI_SDP0))
2068 return;
2069 break;
2070 }
2071 e_crit(drv,
2072 "Network adapter has been stopped because it has over heated. "
2073 "Restart the computer. If the problem persists, "
2074 "power off the system and replace the adapter\n");
2075
2076 adapter->interrupt_event = 0;
2077 }
2078
2079 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2080 {
2081 struct ixgbe_hw *hw = &adapter->hw;
2082
2083 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2084 (eicr & IXGBE_EICR_GPI_SDP1)) {
2085 e_crit(probe, "Fan has stopped, replace the adapter\n");
2086 /* write to clear the interrupt */
2087 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2088 }
2089 }
2090
2091 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2092 {
2093 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2094 return;
2095
2096 switch (adapter->hw.mac.type) {
2097 case ixgbe_mac_82599EB:
2098 /*
2099 * Need to check link state so complete overtemp check
2100 * on service task
2101 */
2102 if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2103 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2104 adapter->interrupt_event = eicr;
2105 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2106 ixgbe_service_event_schedule(adapter);
2107 return;
2108 }
2109 return;
2110 case ixgbe_mac_X540:
2111 if (!(eicr & IXGBE_EICR_TS))
2112 return;
2113 break;
2114 default:
2115 return;
2116 }
2117
2118 e_crit(drv,
2119 "Network adapter has been stopped because it has over heated. "
2120 "Restart the computer. If the problem persists, "
2121 "power off the system and replace the adapter\n");
2122 }
2123
2124 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2125 {
2126 struct ixgbe_hw *hw = &adapter->hw;
2127
2128 if (eicr & IXGBE_EICR_GPI_SDP2) {
2129 /* Clear the interrupt */
2130 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2131 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2132 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2133 ixgbe_service_event_schedule(adapter);
2134 }
2135 }
2136
2137 if (eicr & IXGBE_EICR_GPI_SDP1) {
2138 /* Clear the interrupt */
2139 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2140 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2141 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2142 ixgbe_service_event_schedule(adapter);
2143 }
2144 }
2145 }
2146
2147 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2148 {
2149 struct ixgbe_hw *hw = &adapter->hw;
2150
2151 adapter->lsc_int++;
2152 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2153 adapter->link_check_timeout = jiffies;
2154 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2155 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2156 IXGBE_WRITE_FLUSH(hw);
2157 ixgbe_service_event_schedule(adapter);
2158 }
2159 }
2160
2161 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2162 u64 qmask)
2163 {
2164 u32 mask;
2165 struct ixgbe_hw *hw = &adapter->hw;
2166
2167 switch (hw->mac.type) {
2168 case ixgbe_mac_82598EB:
2169 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2170 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2171 break;
2172 case ixgbe_mac_82599EB:
2173 case ixgbe_mac_X540:
2174 mask = (qmask & 0xFFFFFFFF);
2175 if (mask)
2176 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2177 mask = (qmask >> 32);
2178 if (mask)
2179 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2180 break;
2181 default:
2182 break;
2183 }
2184 /* skip the flush */
2185 }
2186
2187 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2188 u64 qmask)
2189 {
2190 u32 mask;
2191 struct ixgbe_hw *hw = &adapter->hw;
2192
2193 switch (hw->mac.type) {
2194 case ixgbe_mac_82598EB:
2195 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2196 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2197 break;
2198 case ixgbe_mac_82599EB:
2199 case ixgbe_mac_X540:
2200 mask = (qmask & 0xFFFFFFFF);
2201 if (mask)
2202 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2203 mask = (qmask >> 32);
2204 if (mask)
2205 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2206 break;
2207 default:
2208 break;
2209 }
2210 /* skip the flush */
2211 }
2212
2213 /**
2214 * ixgbe_irq_enable - Enable default interrupt generation settings
2215 * @adapter: board private structure
2216 **/
2217 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2218 bool flush)
2219 {
2220 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2221
2222 /* don't reenable LSC while waiting for link */
2223 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2224 mask &= ~IXGBE_EIMS_LSC;
2225
2226 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2227 switch (adapter->hw.mac.type) {
2228 case ixgbe_mac_82599EB:
2229 mask |= IXGBE_EIMS_GPI_SDP0;
2230 break;
2231 case ixgbe_mac_X540:
2232 mask |= IXGBE_EIMS_TS;
2233 break;
2234 default:
2235 break;
2236 }
2237 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2238 mask |= IXGBE_EIMS_GPI_SDP1;
2239 switch (adapter->hw.mac.type) {
2240 case ixgbe_mac_82599EB:
2241 mask |= IXGBE_EIMS_GPI_SDP1;
2242 mask |= IXGBE_EIMS_GPI_SDP2;
2243 case ixgbe_mac_X540:
2244 mask |= IXGBE_EIMS_ECC;
2245 mask |= IXGBE_EIMS_MAILBOX;
2246 break;
2247 default:
2248 break;
2249 }
2250 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2251 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2252 mask |= IXGBE_EIMS_FLOW_DIR;
2253
2254 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2255 if (queues)
2256 ixgbe_irq_enable_queues(adapter, ~0);
2257 if (flush)
2258 IXGBE_WRITE_FLUSH(&adapter->hw);
2259 }
2260
2261 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2262 {
2263 struct ixgbe_adapter *adapter = data;
2264 struct ixgbe_hw *hw = &adapter->hw;
2265 u32 eicr;
2266
2267 /*
2268 * Workaround for Silicon errata. Use clear-by-write instead
2269 * of clear-by-read. Reading with EICS will return the
2270 * interrupt causes without clearing, which later be done
2271 * with the write to EICR.
2272 */
2273 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2274 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2275
2276 if (eicr & IXGBE_EICR_LSC)
2277 ixgbe_check_lsc(adapter);
2278
2279 if (eicr & IXGBE_EICR_MAILBOX)
2280 ixgbe_msg_task(adapter);
2281
2282 switch (hw->mac.type) {
2283 case ixgbe_mac_82599EB:
2284 case ixgbe_mac_X540:
2285 if (eicr & IXGBE_EICR_ECC)
2286 e_info(link, "Received unrecoverable ECC Err, please "
2287 "reboot\n");
2288 /* Handle Flow Director Full threshold interrupt */
2289 if (eicr & IXGBE_EICR_FLOW_DIR) {
2290 int reinit_count = 0;
2291 int i;
2292 for (i = 0; i < adapter->num_tx_queues; i++) {
2293 struct ixgbe_ring *ring = adapter->tx_ring[i];
2294 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2295 &ring->state))
2296 reinit_count++;
2297 }
2298 if (reinit_count) {
2299 /* no more flow director interrupts until after init */
2300 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2301 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2302 ixgbe_service_event_schedule(adapter);
2303 }
2304 }
2305 ixgbe_check_sfp_event(adapter, eicr);
2306 ixgbe_check_overtemp_event(adapter, eicr);
2307 break;
2308 default:
2309 break;
2310 }
2311
2312 ixgbe_check_fan_failure(adapter, eicr);
2313
2314 /* re-enable the original interrupt state, no lsc, no queues */
2315 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2316 ixgbe_irq_enable(adapter, false, false);
2317
2318 return IRQ_HANDLED;
2319 }
2320
2321 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2322 {
2323 struct ixgbe_q_vector *q_vector = data;
2324
2325 /* EIAM disabled interrupts (on this vector) for us */
2326
2327 if (q_vector->rx.ring || q_vector->tx.ring)
2328 napi_schedule(&q_vector->napi);
2329
2330 return IRQ_HANDLED;
2331 }
2332
2333 /**
2334 * ixgbe_poll - NAPI Rx polling callback
2335 * @napi: structure for representing this polling device
2336 * @budget: how many packets driver is allowed to clean
2337 *
2338 * This function is used for legacy and MSI, NAPI mode
2339 **/
2340 int ixgbe_poll(struct napi_struct *napi, int budget)
2341 {
2342 struct ixgbe_q_vector *q_vector =
2343 container_of(napi, struct ixgbe_q_vector, napi);
2344 struct ixgbe_adapter *adapter = q_vector->adapter;
2345 struct ixgbe_ring *ring;
2346 int per_ring_budget;
2347 bool clean_complete = true;
2348
2349 #ifdef CONFIG_IXGBE_DCA
2350 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2351 ixgbe_update_dca(q_vector);
2352 #endif
2353
2354 ixgbe_for_each_ring(ring, q_vector->tx)
2355 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2356
2357 /* attempt to distribute budget to each queue fairly, but don't allow
2358 * the budget to go below 1 because we'll exit polling */
2359 if (q_vector->rx.count > 1)
2360 per_ring_budget = max(budget/q_vector->rx.count, 1);
2361 else
2362 per_ring_budget = budget;
2363
2364 ixgbe_for_each_ring(ring, q_vector->rx)
2365 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
2366 per_ring_budget);
2367
2368 /* If all work not completed, return budget and keep polling */
2369 if (!clean_complete)
2370 return budget;
2371
2372 /* all work done, exit the polling mode */
2373 napi_complete(napi);
2374 if (adapter->rx_itr_setting & 1)
2375 ixgbe_set_itr(q_vector);
2376 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2377 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2378
2379 return 0;
2380 }
2381
2382 /**
2383 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2384 * @adapter: board private structure
2385 *
2386 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2387 * interrupts from the kernel.
2388 **/
2389 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2390 {
2391 struct net_device *netdev = adapter->netdev;
2392 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2393 int vector, err;
2394 int ri = 0, ti = 0;
2395
2396 for (vector = 0; vector < q_vectors; vector++) {
2397 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2398 struct msix_entry *entry = &adapter->msix_entries[vector];
2399
2400 if (q_vector->tx.ring && q_vector->rx.ring) {
2401 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2402 "%s-%s-%d", netdev->name, "TxRx", ri++);
2403 ti++;
2404 } else if (q_vector->rx.ring) {
2405 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2406 "%s-%s-%d", netdev->name, "rx", ri++);
2407 } else if (q_vector->tx.ring) {
2408 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2409 "%s-%s-%d", netdev->name, "tx", ti++);
2410 } else {
2411 /* skip this unused q_vector */
2412 continue;
2413 }
2414 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2415 q_vector->name, q_vector);
2416 if (err) {
2417 e_err(probe, "request_irq failed for MSIX interrupt "
2418 "Error: %d\n", err);
2419 goto free_queue_irqs;
2420 }
2421 /* If Flow Director is enabled, set interrupt affinity */
2422 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2423 /* assign the mask for this irq */
2424 irq_set_affinity_hint(entry->vector,
2425 &q_vector->affinity_mask);
2426 }
2427 }
2428
2429 err = request_irq(adapter->msix_entries[vector].vector,
2430 ixgbe_msix_other, 0, netdev->name, adapter);
2431 if (err) {
2432 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2433 goto free_queue_irqs;
2434 }
2435
2436 return 0;
2437
2438 free_queue_irqs:
2439 while (vector) {
2440 vector--;
2441 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2442 NULL);
2443 free_irq(adapter->msix_entries[vector].vector,
2444 adapter->q_vector[vector]);
2445 }
2446 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2447 pci_disable_msix(adapter->pdev);
2448 kfree(adapter->msix_entries);
2449 adapter->msix_entries = NULL;
2450 return err;
2451 }
2452
2453 /**
2454 * ixgbe_intr - legacy mode Interrupt Handler
2455 * @irq: interrupt number
2456 * @data: pointer to a network interface device structure
2457 **/
2458 static irqreturn_t ixgbe_intr(int irq, void *data)
2459 {
2460 struct ixgbe_adapter *adapter = data;
2461 struct ixgbe_hw *hw = &adapter->hw;
2462 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2463 u32 eicr;
2464
2465 /*
2466 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2467 * before the read of EICR.
2468 */
2469 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2470
2471 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2472 * therefore no explicit interrupt disable is necessary */
2473 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2474 if (!eicr) {
2475 /*
2476 * shared interrupt alert!
2477 * make sure interrupts are enabled because the read will
2478 * have disabled interrupts due to EIAM
2479 * finish the workaround of silicon errata on 82598. Unmask
2480 * the interrupt that we masked before the EICR read.
2481 */
2482 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2483 ixgbe_irq_enable(adapter, true, true);
2484 return IRQ_NONE; /* Not our interrupt */
2485 }
2486
2487 if (eicr & IXGBE_EICR_LSC)
2488 ixgbe_check_lsc(adapter);
2489
2490 switch (hw->mac.type) {
2491 case ixgbe_mac_82599EB:
2492 ixgbe_check_sfp_event(adapter, eicr);
2493 /* Fall through */
2494 case ixgbe_mac_X540:
2495 if (eicr & IXGBE_EICR_ECC)
2496 e_info(link, "Received unrecoverable ECC err, please "
2497 "reboot\n");
2498 ixgbe_check_overtemp_event(adapter, eicr);
2499 break;
2500 default:
2501 break;
2502 }
2503
2504 ixgbe_check_fan_failure(adapter, eicr);
2505
2506 /* would disable interrupts here but EIAM disabled it */
2507 napi_schedule(&q_vector->napi);
2508
2509 /*
2510 * re-enable link(maybe) and non-queue interrupts, no flush.
2511 * ixgbe_poll will re-enable the queue interrupts
2512 */
2513 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2514 ixgbe_irq_enable(adapter, false, false);
2515
2516 return IRQ_HANDLED;
2517 }
2518
2519 /**
2520 * ixgbe_request_irq - initialize interrupts
2521 * @adapter: board private structure
2522 *
2523 * Attempts to configure interrupts using the best available
2524 * capabilities of the hardware and kernel.
2525 **/
2526 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2527 {
2528 struct net_device *netdev = adapter->netdev;
2529 int err;
2530
2531 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2532 err = ixgbe_request_msix_irqs(adapter);
2533 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2534 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2535 netdev->name, adapter);
2536 else
2537 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2538 netdev->name, adapter);
2539
2540 if (err)
2541 e_err(probe, "request_irq failed, Error %d\n", err);
2542
2543 return err;
2544 }
2545
2546 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2547 {
2548 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2549 int i, q_vectors;
2550
2551 q_vectors = adapter->num_msix_vectors;
2552 i = q_vectors - 1;
2553 free_irq(adapter->msix_entries[i].vector, adapter);
2554 i--;
2555
2556 for (; i >= 0; i--) {
2557 /* free only the irqs that were actually requested */
2558 if (!adapter->q_vector[i]->rx.ring &&
2559 !adapter->q_vector[i]->tx.ring)
2560 continue;
2561
2562 /* clear the affinity_mask in the IRQ descriptor */
2563 irq_set_affinity_hint(adapter->msix_entries[i].vector,
2564 NULL);
2565
2566 free_irq(adapter->msix_entries[i].vector,
2567 adapter->q_vector[i]);
2568 }
2569 } else {
2570 free_irq(adapter->pdev->irq, adapter);
2571 }
2572 }
2573
2574 /**
2575 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2576 * @adapter: board private structure
2577 **/
2578 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2579 {
2580 switch (adapter->hw.mac.type) {
2581 case ixgbe_mac_82598EB:
2582 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2583 break;
2584 case ixgbe_mac_82599EB:
2585 case ixgbe_mac_X540:
2586 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2587 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2588 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2589 break;
2590 default:
2591 break;
2592 }
2593 IXGBE_WRITE_FLUSH(&adapter->hw);
2594 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2595 int i;
2596 for (i = 0; i < adapter->num_msix_vectors; i++)
2597 synchronize_irq(adapter->msix_entries[i].vector);
2598 } else {
2599 synchronize_irq(adapter->pdev->irq);
2600 }
2601 }
2602
2603 /**
2604 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2605 *
2606 **/
2607 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2608 {
2609 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2610
2611 /* rx/tx vector */
2612 if (adapter->rx_itr_setting == 1)
2613 q_vector->itr = IXGBE_20K_ITR;
2614 else
2615 q_vector->itr = adapter->rx_itr_setting;
2616
2617 ixgbe_write_eitr(q_vector);
2618
2619 ixgbe_set_ivar(adapter, 0, 0, 0);
2620 ixgbe_set_ivar(adapter, 1, 0, 0);
2621
2622 e_info(hw, "Legacy interrupt IVAR setup done\n");
2623 }
2624
2625 /**
2626 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2627 * @adapter: board private structure
2628 * @ring: structure containing ring specific data
2629 *
2630 * Configure the Tx descriptor ring after a reset.
2631 **/
2632 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2633 struct ixgbe_ring *ring)
2634 {
2635 struct ixgbe_hw *hw = &adapter->hw;
2636 u64 tdba = ring->dma;
2637 int wait_loop = 10;
2638 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2639 u8 reg_idx = ring->reg_idx;
2640
2641 /* disable queue to avoid issues while updating state */
2642 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2643 IXGBE_WRITE_FLUSH(hw);
2644
2645 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2646 (tdba & DMA_BIT_MASK(32)));
2647 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2648 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2649 ring->count * sizeof(union ixgbe_adv_tx_desc));
2650 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2651 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2652 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2653
2654 /*
2655 * set WTHRESH to encourage burst writeback, it should not be set
2656 * higher than 1 when ITR is 0 as it could cause false TX hangs
2657 *
2658 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2659 * to or less than the number of on chip descriptors, which is
2660 * currently 40.
2661 */
2662 if (!ring->q_vector || (ring->q_vector->itr < 8))
2663 txdctl |= (1 << 16); /* WTHRESH = 1 */
2664 else
2665 txdctl |= (8 << 16); /* WTHRESH = 8 */
2666
2667 /*
2668 * Setting PTHRESH to 32 both improves performance
2669 * and avoids a TX hang with DFP enabled
2670 */
2671 txdctl |= (1 << 8) | /* HTHRESH = 1 */
2672 32; /* PTHRESH = 32 */
2673
2674 /* reinitialize flowdirector state */
2675 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2676 adapter->atr_sample_rate) {
2677 ring->atr_sample_rate = adapter->atr_sample_rate;
2678 ring->atr_count = 0;
2679 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2680 } else {
2681 ring->atr_sample_rate = 0;
2682 }
2683
2684 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2685
2686 /* enable queue */
2687 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2688
2689 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2690 if (hw->mac.type == ixgbe_mac_82598EB &&
2691 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2692 return;
2693
2694 /* poll to verify queue is enabled */
2695 do {
2696 usleep_range(1000, 2000);
2697 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2698 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2699 if (!wait_loop)
2700 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2701 }
2702
2703 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2704 {
2705 struct ixgbe_hw *hw = &adapter->hw;
2706 u32 rttdcs;
2707 u32 reg;
2708 u8 tcs = netdev_get_num_tc(adapter->netdev);
2709
2710 if (hw->mac.type == ixgbe_mac_82598EB)
2711 return;
2712
2713 /* disable the arbiter while setting MTQC */
2714 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2715 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2716 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2717
2718 /* set transmit pool layout */
2719 switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2720 case (IXGBE_FLAG_SRIOV_ENABLED):
2721 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2722 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2723 break;
2724 default:
2725 if (!tcs)
2726 reg = IXGBE_MTQC_64Q_1PB;
2727 else if (tcs <= 4)
2728 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2729 else
2730 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2731
2732 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2733
2734 /* Enable Security TX Buffer IFG for multiple pb */
2735 if (tcs) {
2736 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2737 reg |= IXGBE_SECTX_DCB;
2738 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2739 }
2740 break;
2741 }
2742
2743 /* re-enable the arbiter */
2744 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2745 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2746 }
2747
2748 /**
2749 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2750 * @adapter: board private structure
2751 *
2752 * Configure the Tx unit of the MAC after a reset.
2753 **/
2754 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2755 {
2756 struct ixgbe_hw *hw = &adapter->hw;
2757 u32 dmatxctl;
2758 u32 i;
2759
2760 ixgbe_setup_mtqc(adapter);
2761
2762 if (hw->mac.type != ixgbe_mac_82598EB) {
2763 /* DMATXCTL.EN must be before Tx queues are enabled */
2764 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2765 dmatxctl |= IXGBE_DMATXCTL_TE;
2766 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2767 }
2768
2769 /* Setup the HW Tx Head and Tail descriptor pointers */
2770 for (i = 0; i < adapter->num_tx_queues; i++)
2771 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2772 }
2773
2774 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2775
2776 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2777 struct ixgbe_ring *rx_ring)
2778 {
2779 u32 srrctl;
2780 u8 reg_idx = rx_ring->reg_idx;
2781
2782 switch (adapter->hw.mac.type) {
2783 case ixgbe_mac_82598EB: {
2784 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2785 const int mask = feature[RING_F_RSS].mask;
2786 reg_idx = reg_idx & mask;
2787 }
2788 break;
2789 case ixgbe_mac_82599EB:
2790 case ixgbe_mac_X540:
2791 default:
2792 break;
2793 }
2794
2795 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2796
2797 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2798 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2799 if (adapter->num_vfs)
2800 srrctl |= IXGBE_SRRCTL_DROP_EN;
2801
2802 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2803 IXGBE_SRRCTL_BSIZEHDR_MASK;
2804
2805 #if PAGE_SIZE > IXGBE_MAX_RXBUFFER
2806 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2807 #else
2808 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2809 #endif
2810 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2811
2812 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2813 }
2814
2815 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2816 {
2817 struct ixgbe_hw *hw = &adapter->hw;
2818 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2819 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2820 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2821 u32 mrqc = 0, reta = 0;
2822 u32 rxcsum;
2823 int i, j;
2824 u8 tcs = netdev_get_num_tc(adapter->netdev);
2825 int maxq = adapter->ring_feature[RING_F_RSS].indices;
2826
2827 if (tcs)
2828 maxq = min(maxq, adapter->num_tx_queues / tcs);
2829
2830 /* Fill out hash function seeds */
2831 for (i = 0; i < 10; i++)
2832 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2833
2834 /* Fill out redirection table */
2835 for (i = 0, j = 0; i < 128; i++, j++) {
2836 if (j == maxq)
2837 j = 0;
2838 /* reta = 4-byte sliding window of
2839 * 0x00..(indices-1)(indices-1)00..etc. */
2840 reta = (reta << 8) | (j * 0x11);
2841 if ((i & 3) == 3)
2842 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2843 }
2844
2845 /* Disable indicating checksum in descriptor, enables RSS hash */
2846 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2847 rxcsum |= IXGBE_RXCSUM_PCSD;
2848 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2849
2850 if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2851 (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2852 mrqc = IXGBE_MRQC_RSSEN;
2853 } else {
2854 int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2855 | IXGBE_FLAG_SRIOV_ENABLED);
2856
2857 switch (mask) {
2858 case (IXGBE_FLAG_RSS_ENABLED):
2859 if (!tcs)
2860 mrqc = IXGBE_MRQC_RSSEN;
2861 else if (tcs <= 4)
2862 mrqc = IXGBE_MRQC_RTRSS4TCEN;
2863 else
2864 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2865 break;
2866 case (IXGBE_FLAG_SRIOV_ENABLED):
2867 mrqc = IXGBE_MRQC_VMDQEN;
2868 break;
2869 default:
2870 break;
2871 }
2872 }
2873
2874 /* Perform hash on these packet types */
2875 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2876 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2877 | IXGBE_MRQC_RSS_FIELD_IPV6
2878 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2879
2880 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
2881 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
2882 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
2883 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2884
2885 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2886 }
2887
2888 /**
2889 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2890 * @adapter: address of board private structure
2891 * @index: index of ring to set
2892 **/
2893 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2894 struct ixgbe_ring *ring)
2895 {
2896 struct ixgbe_hw *hw = &adapter->hw;
2897 u32 rscctrl;
2898 u8 reg_idx = ring->reg_idx;
2899
2900 if (!ring_is_rsc_enabled(ring))
2901 return;
2902
2903 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2904 rscctrl |= IXGBE_RSCCTL_RSCEN;
2905 /*
2906 * we must limit the number of descriptors so that the
2907 * total size of max desc * buf_len is not greater
2908 * than 65536
2909 */
2910 #if (PAGE_SIZE <= 8192)
2911 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2912 #elif (PAGE_SIZE <= 16384)
2913 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2914 #else
2915 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2916 #endif
2917 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2918 }
2919
2920 #define IXGBE_MAX_RX_DESC_POLL 10
2921 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2922 struct ixgbe_ring *ring)
2923 {
2924 struct ixgbe_hw *hw = &adapter->hw;
2925 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2926 u32 rxdctl;
2927 u8 reg_idx = ring->reg_idx;
2928
2929 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2930 if (hw->mac.type == ixgbe_mac_82598EB &&
2931 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2932 return;
2933
2934 do {
2935 usleep_range(1000, 2000);
2936 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2937 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2938
2939 if (!wait_loop) {
2940 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2941 "the polling period\n", reg_idx);
2942 }
2943 }
2944
2945 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
2946 struct ixgbe_ring *ring)
2947 {
2948 struct ixgbe_hw *hw = &adapter->hw;
2949 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2950 u32 rxdctl;
2951 u8 reg_idx = ring->reg_idx;
2952
2953 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2954 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
2955
2956 /* write value back with RXDCTL.ENABLE bit cleared */
2957 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2958
2959 if (hw->mac.type == ixgbe_mac_82598EB &&
2960 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2961 return;
2962
2963 /* the hardware may take up to 100us to really disable the rx queue */
2964 do {
2965 udelay(10);
2966 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2967 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
2968
2969 if (!wait_loop) {
2970 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
2971 "the polling period\n", reg_idx);
2972 }
2973 }
2974
2975 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2976 struct ixgbe_ring *ring)
2977 {
2978 struct ixgbe_hw *hw = &adapter->hw;
2979 u64 rdba = ring->dma;
2980 u32 rxdctl;
2981 u8 reg_idx = ring->reg_idx;
2982
2983 /* disable queue to avoid issues while updating state */
2984 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2985 ixgbe_disable_rx_queue(adapter, ring);
2986
2987 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2988 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2989 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2990 ring->count * sizeof(union ixgbe_adv_rx_desc));
2991 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2992 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2993 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2994
2995 ixgbe_configure_srrctl(adapter, ring);
2996 ixgbe_configure_rscctl(adapter, ring);
2997
2998 /* If operating in IOV mode set RLPML for X540 */
2999 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3000 hw->mac.type == ixgbe_mac_X540) {
3001 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3002 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3003 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3004 }
3005
3006 if (hw->mac.type == ixgbe_mac_82598EB) {
3007 /*
3008 * enable cache line friendly hardware writes:
3009 * PTHRESH=32 descriptors (half the internal cache),
3010 * this also removes ugly rx_no_buffer_count increment
3011 * HTHRESH=4 descriptors (to minimize latency on fetch)
3012 * WTHRESH=8 burst writeback up to two cache lines
3013 */
3014 rxdctl &= ~0x3FFFFF;
3015 rxdctl |= 0x080420;
3016 }
3017
3018 /* enable receive descriptor ring */
3019 rxdctl |= IXGBE_RXDCTL_ENABLE;
3020 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3021
3022 ixgbe_rx_desc_queue_enable(adapter, ring);
3023 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3024 }
3025
3026 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3027 {
3028 struct ixgbe_hw *hw = &adapter->hw;
3029 int p;
3030
3031 /* PSRTYPE must be initialized in non 82598 adapters */
3032 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3033 IXGBE_PSRTYPE_UDPHDR |
3034 IXGBE_PSRTYPE_IPV4HDR |
3035 IXGBE_PSRTYPE_L2HDR |
3036 IXGBE_PSRTYPE_IPV6HDR;
3037
3038 if (hw->mac.type == ixgbe_mac_82598EB)
3039 return;
3040
3041 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3042 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3043
3044 for (p = 0; p < adapter->num_rx_pools; p++)
3045 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3046 psrtype);
3047 }
3048
3049 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3050 {
3051 struct ixgbe_hw *hw = &adapter->hw;
3052 u32 gcr_ext;
3053 u32 vt_reg_bits;
3054 u32 reg_offset, vf_shift;
3055 u32 vmdctl;
3056 int i;
3057
3058 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3059 return;
3060
3061 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3062 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3063 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3064 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3065
3066 vf_shift = adapter->num_vfs % 32;
3067 reg_offset = (adapter->num_vfs >= 32) ? 1 : 0;
3068
3069 /* Enable only the PF's pool for Tx/Rx */
3070 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3071 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3072 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3073 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3074 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3075
3076 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3077 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3078
3079 /*
3080 * Set up VF register offsets for selected VT Mode,
3081 * i.e. 32 or 64 VFs for SR-IOV
3082 */
3083 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3084 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3085 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3086 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3087
3088 /* enable Tx loopback for VF/PF communication */
3089 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3090 /* Enable MAC Anti-Spoofing */
3091 hw->mac.ops.set_mac_anti_spoofing(hw,
3092 (adapter->num_vfs != 0),
3093 adapter->num_vfs);
3094 /* For VFs that have spoof checking turned off */
3095 for (i = 0; i < adapter->num_vfs; i++) {
3096 if (!adapter->vfinfo[i].spoofchk_enabled)
3097 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3098 }
3099 }
3100
3101 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3102 {
3103 struct ixgbe_hw *hw = &adapter->hw;
3104 struct net_device *netdev = adapter->netdev;
3105 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3106 struct ixgbe_ring *rx_ring;
3107 int i;
3108 u32 mhadd, hlreg0;
3109
3110 #ifdef IXGBE_FCOE
3111 /* adjust max frame to be able to do baby jumbo for FCoE */
3112 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3113 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3114 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3115
3116 #endif /* IXGBE_FCOE */
3117 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3118 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3119 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3120 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3121
3122 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3123 }
3124
3125 /* MHADD will allow an extra 4 bytes past for vlan tagged frames */
3126 max_frame += VLAN_HLEN;
3127
3128 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3129 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3130 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3131 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3132
3133 /*
3134 * Setup the HW Rx Head and Tail Descriptor Pointers and
3135 * the Base and Length of the Rx Descriptor Ring
3136 */
3137 for (i = 0; i < adapter->num_rx_queues; i++) {
3138 rx_ring = adapter->rx_ring[i];
3139 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3140 set_ring_rsc_enabled(rx_ring);
3141 else
3142 clear_ring_rsc_enabled(rx_ring);
3143 }
3144 }
3145
3146 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3147 {
3148 struct ixgbe_hw *hw = &adapter->hw;
3149 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3150
3151 switch (hw->mac.type) {
3152 case ixgbe_mac_82598EB:
3153 /*
3154 * For VMDq support of different descriptor types or
3155 * buffer sizes through the use of multiple SRRCTL
3156 * registers, RDRXCTL.MVMEN must be set to 1
3157 *
3158 * also, the manual doesn't mention it clearly but DCA hints
3159 * will only use queue 0's tags unless this bit is set. Side
3160 * effects of setting this bit are only that SRRCTL must be
3161 * fully programmed [0..15]
3162 */
3163 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3164 break;
3165 case ixgbe_mac_82599EB:
3166 case ixgbe_mac_X540:
3167 /* Disable RSC for ACK packets */
3168 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3169 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3170 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3171 /* hardware requires some bits to be set by default */
3172 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3173 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3174 break;
3175 default:
3176 /* We should do nothing since we don't know this hardware */
3177 return;
3178 }
3179
3180 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3181 }
3182
3183 /**
3184 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3185 * @adapter: board private structure
3186 *
3187 * Configure the Rx unit of the MAC after a reset.
3188 **/
3189 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3190 {
3191 struct ixgbe_hw *hw = &adapter->hw;
3192 int i;
3193 u32 rxctrl;
3194
3195 /* disable receives while setting up the descriptors */
3196 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3197 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3198
3199 ixgbe_setup_psrtype(adapter);
3200 ixgbe_setup_rdrxctl(adapter);
3201
3202 /* Program registers for the distribution of queues */
3203 ixgbe_setup_mrqc(adapter);
3204
3205 /* set_rx_buffer_len must be called before ring initialization */
3206 ixgbe_set_rx_buffer_len(adapter);
3207
3208 /*
3209 * Setup the HW Rx Head and Tail Descriptor Pointers and
3210 * the Base and Length of the Rx Descriptor Ring
3211 */
3212 for (i = 0; i < adapter->num_rx_queues; i++)
3213 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3214
3215 /* disable drop enable for 82598 parts */
3216 if (hw->mac.type == ixgbe_mac_82598EB)
3217 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3218
3219 /* enable all receives */
3220 rxctrl |= IXGBE_RXCTRL_RXEN;
3221 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3222 }
3223
3224 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3225 {
3226 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3227 struct ixgbe_hw *hw = &adapter->hw;
3228 int pool_ndx = adapter->num_vfs;
3229
3230 /* add VID to filter table */
3231 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3232 set_bit(vid, adapter->active_vlans);
3233
3234 return 0;
3235 }
3236
3237 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3238 {
3239 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3240 struct ixgbe_hw *hw = &adapter->hw;
3241 int pool_ndx = adapter->num_vfs;
3242
3243 /* remove VID from filter table */
3244 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3245 clear_bit(vid, adapter->active_vlans);
3246
3247 return 0;
3248 }
3249
3250 /**
3251 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3252 * @adapter: driver data
3253 */
3254 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3255 {
3256 struct ixgbe_hw *hw = &adapter->hw;
3257 u32 vlnctrl;
3258
3259 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3260 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3261 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3262 }
3263
3264 /**
3265 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3266 * @adapter: driver data
3267 */
3268 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3269 {
3270 struct ixgbe_hw *hw = &adapter->hw;
3271 u32 vlnctrl;
3272
3273 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3274 vlnctrl |= IXGBE_VLNCTRL_VFE;
3275 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3276 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3277 }
3278
3279 /**
3280 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3281 * @adapter: driver data
3282 */
3283 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3284 {
3285 struct ixgbe_hw *hw = &adapter->hw;
3286 u32 vlnctrl;
3287 int i, j;
3288
3289 switch (hw->mac.type) {
3290 case ixgbe_mac_82598EB:
3291 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3292 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3293 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3294 break;
3295 case ixgbe_mac_82599EB:
3296 case ixgbe_mac_X540:
3297 for (i = 0; i < adapter->num_rx_queues; i++) {
3298 j = adapter->rx_ring[i]->reg_idx;
3299 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3300 vlnctrl &= ~IXGBE_RXDCTL_VME;
3301 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3302 }
3303 break;
3304 default:
3305 break;
3306 }
3307 }
3308
3309 /**
3310 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3311 * @adapter: driver data
3312 */
3313 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3314 {
3315 struct ixgbe_hw *hw = &adapter->hw;
3316 u32 vlnctrl;
3317 int i, j;
3318
3319 switch (hw->mac.type) {
3320 case ixgbe_mac_82598EB:
3321 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3322 vlnctrl |= IXGBE_VLNCTRL_VME;
3323 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3324 break;
3325 case ixgbe_mac_82599EB:
3326 case ixgbe_mac_X540:
3327 for (i = 0; i < adapter->num_rx_queues; i++) {
3328 j = adapter->rx_ring[i]->reg_idx;
3329 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3330 vlnctrl |= IXGBE_RXDCTL_VME;
3331 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3332 }
3333 break;
3334 default:
3335 break;
3336 }
3337 }
3338
3339 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3340 {
3341 u16 vid;
3342
3343 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3344
3345 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3346 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3347 }
3348
3349 /**
3350 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3351 * @netdev: network interface device structure
3352 *
3353 * Writes unicast address list to the RAR table.
3354 * Returns: -ENOMEM on failure/insufficient address space
3355 * 0 on no addresses written
3356 * X on writing X addresses to the RAR table
3357 **/
3358 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3359 {
3360 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3361 struct ixgbe_hw *hw = &adapter->hw;
3362 unsigned int vfn = adapter->num_vfs;
3363 unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3364 int count = 0;
3365
3366 /* return ENOMEM indicating insufficient memory for addresses */
3367 if (netdev_uc_count(netdev) > rar_entries)
3368 return -ENOMEM;
3369
3370 if (!netdev_uc_empty(netdev) && rar_entries) {
3371 struct netdev_hw_addr *ha;
3372 /* return error if we do not support writing to RAR table */
3373 if (!hw->mac.ops.set_rar)
3374 return -ENOMEM;
3375
3376 netdev_for_each_uc_addr(ha, netdev) {
3377 if (!rar_entries)
3378 break;
3379 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3380 vfn, IXGBE_RAH_AV);
3381 count++;
3382 }
3383 }
3384 /* write the addresses in reverse order to avoid write combining */
3385 for (; rar_entries > 0 ; rar_entries--)
3386 hw->mac.ops.clear_rar(hw, rar_entries);
3387
3388 return count;
3389 }
3390
3391 /**
3392 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3393 * @netdev: network interface device structure
3394 *
3395 * The set_rx_method entry point is called whenever the unicast/multicast
3396 * address list or the network interface flags are updated. This routine is
3397 * responsible for configuring the hardware for proper unicast, multicast and
3398 * promiscuous mode.
3399 **/
3400 void ixgbe_set_rx_mode(struct net_device *netdev)
3401 {
3402 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3403 struct ixgbe_hw *hw = &adapter->hw;
3404 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3405 int count;
3406
3407 /* Check for Promiscuous and All Multicast modes */
3408
3409 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3410
3411 /* set all bits that we expect to always be set */
3412 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3413 fctrl |= IXGBE_FCTRL_BAM;
3414 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3415 fctrl |= IXGBE_FCTRL_PMCF;
3416
3417 /* clear the bits we are changing the status of */
3418 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3419
3420 if (netdev->flags & IFF_PROMISC) {
3421 hw->addr_ctrl.user_set_promisc = true;
3422 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3423 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3424 /* don't hardware filter vlans in promisc mode */
3425 ixgbe_vlan_filter_disable(adapter);
3426 } else {
3427 if (netdev->flags & IFF_ALLMULTI) {
3428 fctrl |= IXGBE_FCTRL_MPE;
3429 vmolr |= IXGBE_VMOLR_MPE;
3430 } else {
3431 /*
3432 * Write addresses to the MTA, if the attempt fails
3433 * then we should just turn on promiscuous mode so
3434 * that we can at least receive multicast traffic
3435 */
3436 hw->mac.ops.update_mc_addr_list(hw, netdev);
3437 vmolr |= IXGBE_VMOLR_ROMPE;
3438 }
3439 ixgbe_vlan_filter_enable(adapter);
3440 hw->addr_ctrl.user_set_promisc = false;
3441 }
3442
3443 /*
3444 * Write addresses to available RAR registers, if there is not
3445 * sufficient space to store all the addresses then enable
3446 * unicast promiscuous mode
3447 */
3448 count = ixgbe_write_uc_addr_list(netdev);
3449 if (count < 0) {
3450 fctrl |= IXGBE_FCTRL_UPE;
3451 vmolr |= IXGBE_VMOLR_ROPE;
3452 }
3453
3454 if (adapter->num_vfs) {
3455 ixgbe_restore_vf_multicasts(adapter);
3456 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3457 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3458 IXGBE_VMOLR_ROPE);
3459 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3460 }
3461
3462 /* This is useful for sniffing bad packets. */
3463 if (adapter->netdev->features & NETIF_F_RXALL) {
3464 /* UPE and MPE will be handled by normal PROMISC logic
3465 * in e1000e_set_rx_mode */
3466 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3467 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3468 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3469
3470 fctrl &= ~(IXGBE_FCTRL_DPF);
3471 /* NOTE: VLAN filtering is disabled by setting PROMISC */
3472 }
3473
3474 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3475
3476 if (netdev->features & NETIF_F_HW_VLAN_RX)
3477 ixgbe_vlan_strip_enable(adapter);
3478 else
3479 ixgbe_vlan_strip_disable(adapter);
3480 }
3481
3482 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3483 {
3484 int q_idx;
3485 struct ixgbe_q_vector *q_vector;
3486 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3487
3488 /* legacy and MSI only use one vector */
3489 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3490 q_vectors = 1;
3491
3492 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3493 q_vector = adapter->q_vector[q_idx];
3494 napi_enable(&q_vector->napi);
3495 }
3496 }
3497
3498 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3499 {
3500 int q_idx;
3501 struct ixgbe_q_vector *q_vector;
3502 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3503
3504 /* legacy and MSI only use one vector */
3505 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3506 q_vectors = 1;
3507
3508 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3509 q_vector = adapter->q_vector[q_idx];
3510 napi_disable(&q_vector->napi);
3511 }
3512 }
3513
3514 #ifdef CONFIG_IXGBE_DCB
3515 /*
3516 * ixgbe_configure_dcb - Configure DCB hardware
3517 * @adapter: ixgbe adapter struct
3518 *
3519 * This is called by the driver on open to configure the DCB hardware.
3520 * This is also called by the gennetlink interface when reconfiguring
3521 * the DCB state.
3522 */
3523 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3524 {
3525 struct ixgbe_hw *hw = &adapter->hw;
3526 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3527
3528 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3529 if (hw->mac.type == ixgbe_mac_82598EB)
3530 netif_set_gso_max_size(adapter->netdev, 65536);
3531 return;
3532 }
3533
3534 if (hw->mac.type == ixgbe_mac_82598EB)
3535 netif_set_gso_max_size(adapter->netdev, 32768);
3536
3537
3538 /* Enable VLAN tag insert/strip */
3539 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3540
3541 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3542
3543 #ifdef IXGBE_FCOE
3544 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3545 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3546 #endif
3547
3548 /* reconfigure the hardware */
3549 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3550 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3551 DCB_TX_CONFIG);
3552 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3553 DCB_RX_CONFIG);
3554 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3555 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3556 ixgbe_dcb_hw_ets(&adapter->hw,
3557 adapter->ixgbe_ieee_ets,
3558 max_frame);
3559 ixgbe_dcb_hw_pfc_config(&adapter->hw,
3560 adapter->ixgbe_ieee_pfc->pfc_en,
3561 adapter->ixgbe_ieee_ets->prio_tc);
3562 }
3563
3564 /* Enable RSS Hash per TC */
3565 if (hw->mac.type != ixgbe_mac_82598EB) {
3566 int i;
3567 u32 reg = 0;
3568
3569 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3570 u8 msb = 0;
3571 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3572
3573 while (cnt >>= 1)
3574 msb++;
3575
3576 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3577 }
3578 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3579 }
3580 }
3581 #endif
3582
3583 /* Additional bittime to account for IXGBE framing */
3584 #define IXGBE_ETH_FRAMING 20
3585
3586 /*
3587 * ixgbe_hpbthresh - calculate high water mark for flow control
3588 *
3589 * @adapter: board private structure to calculate for
3590 * @pb - packet buffer to calculate
3591 */
3592 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3593 {
3594 struct ixgbe_hw *hw = &adapter->hw;
3595 struct net_device *dev = adapter->netdev;
3596 int link, tc, kb, marker;
3597 u32 dv_id, rx_pba;
3598
3599 /* Calculate max LAN frame size */
3600 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3601
3602 #ifdef IXGBE_FCOE
3603 /* FCoE traffic class uses FCOE jumbo frames */
3604 if (dev->features & NETIF_F_FCOE_MTU) {
3605 int fcoe_pb = 0;
3606
3607 #ifdef CONFIG_IXGBE_DCB
3608 fcoe_pb = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
3609
3610 #endif
3611 if (fcoe_pb == pb && tc < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3612 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3613 }
3614 #endif
3615
3616 /* Calculate delay value for device */
3617 switch (hw->mac.type) {
3618 case ixgbe_mac_X540:
3619 dv_id = IXGBE_DV_X540(link, tc);
3620 break;
3621 default:
3622 dv_id = IXGBE_DV(link, tc);
3623 break;
3624 }
3625
3626 /* Loopback switch introduces additional latency */
3627 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3628 dv_id += IXGBE_B2BT(tc);
3629
3630 /* Delay value is calculated in bit times convert to KB */
3631 kb = IXGBE_BT2KB(dv_id);
3632 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3633
3634 marker = rx_pba - kb;
3635
3636 /* It is possible that the packet buffer is not large enough
3637 * to provide required headroom. In this case throw an error
3638 * to user and a do the best we can.
3639 */
3640 if (marker < 0) {
3641 e_warn(drv, "Packet Buffer(%i) can not provide enough"
3642 "headroom to support flow control."
3643 "Decrease MTU or number of traffic classes\n", pb);
3644 marker = tc + 1;
3645 }
3646
3647 return marker;
3648 }
3649
3650 /*
3651 * ixgbe_lpbthresh - calculate low water mark for for flow control
3652 *
3653 * @adapter: board private structure to calculate for
3654 * @pb - packet buffer to calculate
3655 */
3656 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3657 {
3658 struct ixgbe_hw *hw = &adapter->hw;
3659 struct net_device *dev = adapter->netdev;
3660 int tc;
3661 u32 dv_id;
3662
3663 /* Calculate max LAN frame size */
3664 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3665
3666 /* Calculate delay value for device */
3667 switch (hw->mac.type) {
3668 case ixgbe_mac_X540:
3669 dv_id = IXGBE_LOW_DV_X540(tc);
3670 break;
3671 default:
3672 dv_id = IXGBE_LOW_DV(tc);
3673 break;
3674 }
3675
3676 /* Delay value is calculated in bit times convert to KB */
3677 return IXGBE_BT2KB(dv_id);
3678 }
3679
3680 /*
3681 * ixgbe_pbthresh_setup - calculate and setup high low water marks
3682 */
3683 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3684 {
3685 struct ixgbe_hw *hw = &adapter->hw;
3686 int num_tc = netdev_get_num_tc(adapter->netdev);
3687 int i;
3688
3689 if (!num_tc)
3690 num_tc = 1;
3691
3692 hw->fc.low_water = ixgbe_lpbthresh(adapter);
3693
3694 for (i = 0; i < num_tc; i++) {
3695 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3696
3697 /* Low water marks must not be larger than high water marks */
3698 if (hw->fc.low_water > hw->fc.high_water[i])
3699 hw->fc.low_water = 0;
3700 }
3701 }
3702
3703 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3704 {
3705 struct ixgbe_hw *hw = &adapter->hw;
3706 int hdrm;
3707 u8 tc = netdev_get_num_tc(adapter->netdev);
3708
3709 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3710 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3711 hdrm = 32 << adapter->fdir_pballoc;
3712 else
3713 hdrm = 0;
3714
3715 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3716 ixgbe_pbthresh_setup(adapter);
3717 }
3718
3719 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3720 {
3721 struct ixgbe_hw *hw = &adapter->hw;
3722 struct hlist_node *node, *node2;
3723 struct ixgbe_fdir_filter *filter;
3724
3725 spin_lock(&adapter->fdir_perfect_lock);
3726
3727 if (!hlist_empty(&adapter->fdir_filter_list))
3728 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3729
3730 hlist_for_each_entry_safe(filter, node, node2,
3731 &adapter->fdir_filter_list, fdir_node) {
3732 ixgbe_fdir_write_perfect_filter_82599(hw,
3733 &filter->filter,
3734 filter->sw_idx,
3735 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3736 IXGBE_FDIR_DROP_QUEUE :
3737 adapter->rx_ring[filter->action]->reg_idx);
3738 }
3739
3740 spin_unlock(&adapter->fdir_perfect_lock);
3741 }
3742
3743 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3744 {
3745 struct ixgbe_hw *hw = &adapter->hw;
3746
3747 ixgbe_configure_pb(adapter);
3748 #ifdef CONFIG_IXGBE_DCB
3749 ixgbe_configure_dcb(adapter);
3750 #endif
3751
3752 ixgbe_set_rx_mode(adapter->netdev);
3753 ixgbe_restore_vlan(adapter);
3754
3755 #ifdef IXGBE_FCOE
3756 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3757 ixgbe_configure_fcoe(adapter);
3758
3759 #endif /* IXGBE_FCOE */
3760
3761 switch (hw->mac.type) {
3762 case ixgbe_mac_82599EB:
3763 case ixgbe_mac_X540:
3764 hw->mac.ops.disable_rx_buff(hw);
3765 break;
3766 default:
3767 break;
3768 }
3769
3770 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3771 ixgbe_init_fdir_signature_82599(&adapter->hw,
3772 adapter->fdir_pballoc);
3773 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3774 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3775 adapter->fdir_pballoc);
3776 ixgbe_fdir_filter_restore(adapter);
3777 }
3778
3779 switch (hw->mac.type) {
3780 case ixgbe_mac_82599EB:
3781 case ixgbe_mac_X540:
3782 hw->mac.ops.enable_rx_buff(hw);
3783 break;
3784 default:
3785 break;
3786 }
3787
3788 ixgbe_configure_virtualization(adapter);
3789
3790 ixgbe_configure_tx(adapter);
3791 ixgbe_configure_rx(adapter);
3792 }
3793
3794 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3795 {
3796 switch (hw->phy.type) {
3797 case ixgbe_phy_sfp_avago:
3798 case ixgbe_phy_sfp_ftl:
3799 case ixgbe_phy_sfp_intel:
3800 case ixgbe_phy_sfp_unknown:
3801 case ixgbe_phy_sfp_passive_tyco:
3802 case ixgbe_phy_sfp_passive_unknown:
3803 case ixgbe_phy_sfp_active_unknown:
3804 case ixgbe_phy_sfp_ftl_active:
3805 return true;
3806 case ixgbe_phy_nl:
3807 if (hw->mac.type == ixgbe_mac_82598EB)
3808 return true;
3809 default:
3810 return false;
3811 }
3812 }
3813
3814 /**
3815 * ixgbe_sfp_link_config - set up SFP+ link
3816 * @adapter: pointer to private adapter struct
3817 **/
3818 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3819 {
3820 /*
3821 * We are assuming the worst case scenario here, and that
3822 * is that an SFP was inserted/removed after the reset
3823 * but before SFP detection was enabled. As such the best
3824 * solution is to just start searching as soon as we start
3825 */
3826 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3827 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3828
3829 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3830 }
3831
3832 /**
3833 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3834 * @hw: pointer to private hardware struct
3835 *
3836 * Returns 0 on success, negative on failure
3837 **/
3838 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3839 {
3840 u32 autoneg;
3841 bool negotiation, link_up = false;
3842 u32 ret = IXGBE_ERR_LINK_SETUP;
3843
3844 if (hw->mac.ops.check_link)
3845 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3846
3847 if (ret)
3848 goto link_cfg_out;
3849
3850 autoneg = hw->phy.autoneg_advertised;
3851 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3852 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3853 &negotiation);
3854 if (ret)
3855 goto link_cfg_out;
3856
3857 if (hw->mac.ops.setup_link)
3858 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3859 link_cfg_out:
3860 return ret;
3861 }
3862
3863 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3864 {
3865 struct ixgbe_hw *hw = &adapter->hw;
3866 u32 gpie = 0;
3867
3868 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3869 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3870 IXGBE_GPIE_OCD;
3871 gpie |= IXGBE_GPIE_EIAME;
3872 /*
3873 * use EIAM to auto-mask when MSI-X interrupt is asserted
3874 * this saves a register write for every interrupt
3875 */
3876 switch (hw->mac.type) {
3877 case ixgbe_mac_82598EB:
3878 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3879 break;
3880 case ixgbe_mac_82599EB:
3881 case ixgbe_mac_X540:
3882 default:
3883 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3884 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3885 break;
3886 }
3887 } else {
3888 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3889 * specifically only auto mask tx and rx interrupts */
3890 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3891 }
3892
3893 /* XXX: to interrupt immediately for EICS writes, enable this */
3894 /* gpie |= IXGBE_GPIE_EIMEN; */
3895
3896 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3897 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3898 gpie |= IXGBE_GPIE_VTMODE_64;
3899 }
3900
3901 /* Enable Thermal over heat sensor interrupt */
3902 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3903 switch (adapter->hw.mac.type) {
3904 case ixgbe_mac_82599EB:
3905 gpie |= IXGBE_SDP0_GPIEN;
3906 break;
3907 case ixgbe_mac_X540:
3908 gpie |= IXGBE_EIMS_TS;
3909 break;
3910 default:
3911 break;
3912 }
3913 }
3914
3915 /* Enable fan failure interrupt */
3916 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3917 gpie |= IXGBE_SDP1_GPIEN;
3918
3919 if (hw->mac.type == ixgbe_mac_82599EB) {
3920 gpie |= IXGBE_SDP1_GPIEN;
3921 gpie |= IXGBE_SDP2_GPIEN;
3922 }
3923
3924 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3925 }
3926
3927 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
3928 {
3929 struct ixgbe_hw *hw = &adapter->hw;
3930 int err;
3931 u32 ctrl_ext;
3932
3933 ixgbe_get_hw_control(adapter);
3934 ixgbe_setup_gpie(adapter);
3935
3936 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3937 ixgbe_configure_msix(adapter);
3938 else
3939 ixgbe_configure_msi_and_legacy(adapter);
3940
3941 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3942 if (hw->mac.ops.enable_tx_laser &&
3943 ((hw->phy.multispeed_fiber) ||
3944 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3945 (hw->mac.type == ixgbe_mac_82599EB))))
3946 hw->mac.ops.enable_tx_laser(hw);
3947
3948 clear_bit(__IXGBE_DOWN, &adapter->state);
3949 ixgbe_napi_enable_all(adapter);
3950
3951 if (ixgbe_is_sfp(hw)) {
3952 ixgbe_sfp_link_config(adapter);
3953 } else {
3954 err = ixgbe_non_sfp_link_config(hw);
3955 if (err)
3956 e_err(probe, "link_config FAILED %d\n", err);
3957 }
3958
3959 /* clear any pending interrupts, may auto mask */
3960 IXGBE_READ_REG(hw, IXGBE_EICR);
3961 ixgbe_irq_enable(adapter, true, true);
3962
3963 /*
3964 * If this adapter has a fan, check to see if we had a failure
3965 * before we enabled the interrupt.
3966 */
3967 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3968 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3969 if (esdp & IXGBE_ESDP_SDP1)
3970 e_crit(drv, "Fan has stopped, replace the adapter\n");
3971 }
3972
3973 /* enable transmits */
3974 netif_tx_start_all_queues(adapter->netdev);
3975
3976 /* bring the link up in the watchdog, this could race with our first
3977 * link up interrupt but shouldn't be a problem */
3978 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3979 adapter->link_check_timeout = jiffies;
3980 mod_timer(&adapter->service_timer, jiffies);
3981
3982 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3983 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3984 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3985 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3986 }
3987
3988 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3989 {
3990 WARN_ON(in_interrupt());
3991 /* put off any impending NetWatchDogTimeout */
3992 adapter->netdev->trans_start = jiffies;
3993
3994 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3995 usleep_range(1000, 2000);
3996 ixgbe_down(adapter);
3997 /*
3998 * If SR-IOV enabled then wait a bit before bringing the adapter
3999 * back up to give the VFs time to respond to the reset. The
4000 * two second wait is based upon the watchdog timer cycle in
4001 * the VF driver.
4002 */
4003 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4004 msleep(2000);
4005 ixgbe_up(adapter);
4006 clear_bit(__IXGBE_RESETTING, &adapter->state);
4007 }
4008
4009 void ixgbe_up(struct ixgbe_adapter *adapter)
4010 {
4011 /* hardware has been reset, we need to reload some things */
4012 ixgbe_configure(adapter);
4013
4014 ixgbe_up_complete(adapter);
4015 }
4016
4017 void ixgbe_reset(struct ixgbe_adapter *adapter)
4018 {
4019 struct ixgbe_hw *hw = &adapter->hw;
4020 int err;
4021
4022 /* lock SFP init bit to prevent race conditions with the watchdog */
4023 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4024 usleep_range(1000, 2000);
4025
4026 /* clear all SFP and link config related flags while holding SFP_INIT */
4027 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4028 IXGBE_FLAG2_SFP_NEEDS_RESET);
4029 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4030
4031 err = hw->mac.ops.init_hw(hw);
4032 switch (err) {
4033 case 0:
4034 case IXGBE_ERR_SFP_NOT_PRESENT:
4035 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4036 break;
4037 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4038 e_dev_err("master disable timed out\n");
4039 break;
4040 case IXGBE_ERR_EEPROM_VERSION:
4041 /* We are running on a pre-production device, log a warning */
4042 e_dev_warn("This device is a pre-production adapter/LOM. "
4043 "Please be aware there may be issues associated with "
4044 "your hardware. If you are experiencing problems "
4045 "please contact your Intel or hardware "
4046 "representative who provided you with this "
4047 "hardware.\n");
4048 break;
4049 default:
4050 e_dev_err("Hardware Error: %d\n", err);
4051 }
4052
4053 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4054
4055 /* reprogram the RAR[0] in case user changed it. */
4056 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
4057 IXGBE_RAH_AV);
4058 }
4059
4060 /**
4061 * ixgbe_init_rx_page_offset - initialize page offset values for Rx buffers
4062 * @rx_ring: ring to setup
4063 *
4064 * On many IA platforms the L1 cache has a critical stride of 4K, this
4065 * results in each receive buffer starting in the same cache set. To help
4066 * reduce the pressure on this cache set we can interleave the offsets so
4067 * that only every other buffer will be in the same cache set.
4068 **/
4069 static void ixgbe_init_rx_page_offset(struct ixgbe_ring *rx_ring)
4070 {
4071 struct ixgbe_rx_buffer *rx_buffer = rx_ring->rx_buffer_info;
4072 u16 i;
4073
4074 for (i = 0; i < rx_ring->count; i += 2) {
4075 rx_buffer[0].page_offset = 0;
4076 rx_buffer[1].page_offset = ixgbe_rx_bufsz(rx_ring);
4077 rx_buffer = &rx_buffer[2];
4078 }
4079 }
4080
4081 /**
4082 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4083 * @rx_ring: ring to free buffers from
4084 **/
4085 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4086 {
4087 struct device *dev = rx_ring->dev;
4088 unsigned long size;
4089 u16 i;
4090
4091 /* ring already cleared, nothing to do */
4092 if (!rx_ring->rx_buffer_info)
4093 return;
4094
4095 /* Free all the Rx ring sk_buffs */
4096 for (i = 0; i < rx_ring->count; i++) {
4097 struct ixgbe_rx_buffer *rx_buffer;
4098
4099 rx_buffer = &rx_ring->rx_buffer_info[i];
4100 if (rx_buffer->skb) {
4101 struct sk_buff *skb = rx_buffer->skb;
4102 if (IXGBE_CB(skb)->page_released) {
4103 dma_unmap_page(dev,
4104 IXGBE_CB(skb)->dma,
4105 ixgbe_rx_bufsz(rx_ring),
4106 DMA_FROM_DEVICE);
4107 IXGBE_CB(skb)->page_released = false;
4108 }
4109 dev_kfree_skb(skb);
4110 }
4111 rx_buffer->skb = NULL;
4112 if (rx_buffer->dma)
4113 dma_unmap_page(dev, rx_buffer->dma,
4114 ixgbe_rx_pg_size(rx_ring),
4115 DMA_FROM_DEVICE);
4116 rx_buffer->dma = 0;
4117 if (rx_buffer->page)
4118 __free_pages(rx_buffer->page,
4119 ixgbe_rx_pg_order(rx_ring));
4120 rx_buffer->page = NULL;
4121 }
4122
4123 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4124 memset(rx_ring->rx_buffer_info, 0, size);
4125
4126 ixgbe_init_rx_page_offset(rx_ring);
4127
4128 /* Zero out the descriptor ring */
4129 memset(rx_ring->desc, 0, rx_ring->size);
4130
4131 rx_ring->next_to_alloc = 0;
4132 rx_ring->next_to_clean = 0;
4133 rx_ring->next_to_use = 0;
4134 }
4135
4136 /**
4137 * ixgbe_clean_tx_ring - Free Tx Buffers
4138 * @tx_ring: ring to be cleaned
4139 **/
4140 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4141 {
4142 struct ixgbe_tx_buffer *tx_buffer_info;
4143 unsigned long size;
4144 u16 i;
4145
4146 /* ring already cleared, nothing to do */
4147 if (!tx_ring->tx_buffer_info)
4148 return;
4149
4150 /* Free all the Tx ring sk_buffs */
4151 for (i = 0; i < tx_ring->count; i++) {
4152 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4153 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4154 }
4155
4156 netdev_tx_reset_queue(txring_txq(tx_ring));
4157
4158 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4159 memset(tx_ring->tx_buffer_info, 0, size);
4160
4161 /* Zero out the descriptor ring */
4162 memset(tx_ring->desc, 0, tx_ring->size);
4163
4164 tx_ring->next_to_use = 0;
4165 tx_ring->next_to_clean = 0;
4166 }
4167
4168 /**
4169 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4170 * @adapter: board private structure
4171 **/
4172 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4173 {
4174 int i;
4175
4176 for (i = 0; i < adapter->num_rx_queues; i++)
4177 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4178 }
4179
4180 /**
4181 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4182 * @adapter: board private structure
4183 **/
4184 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4185 {
4186 int i;
4187
4188 for (i = 0; i < adapter->num_tx_queues; i++)
4189 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4190 }
4191
4192 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4193 {
4194 struct hlist_node *node, *node2;
4195 struct ixgbe_fdir_filter *filter;
4196
4197 spin_lock(&adapter->fdir_perfect_lock);
4198
4199 hlist_for_each_entry_safe(filter, node, node2,
4200 &adapter->fdir_filter_list, fdir_node) {
4201 hlist_del(&filter->fdir_node);
4202 kfree(filter);
4203 }
4204 adapter->fdir_filter_count = 0;
4205
4206 spin_unlock(&adapter->fdir_perfect_lock);
4207 }
4208
4209 void ixgbe_down(struct ixgbe_adapter *adapter)
4210 {
4211 struct net_device *netdev = adapter->netdev;
4212 struct ixgbe_hw *hw = &adapter->hw;
4213 u32 rxctrl;
4214 int i;
4215
4216 /* signal that we are down to the interrupt handler */
4217 set_bit(__IXGBE_DOWN, &adapter->state);
4218
4219 /* disable receives */
4220 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4221 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4222
4223 /* disable all enabled rx queues */
4224 for (i = 0; i < adapter->num_rx_queues; i++)
4225 /* this call also flushes the previous write */
4226 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4227
4228 usleep_range(10000, 20000);
4229
4230 netif_tx_stop_all_queues(netdev);
4231
4232 /* call carrier off first to avoid false dev_watchdog timeouts */
4233 netif_carrier_off(netdev);
4234 netif_tx_disable(netdev);
4235
4236 ixgbe_irq_disable(adapter);
4237
4238 ixgbe_napi_disable_all(adapter);
4239
4240 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4241 IXGBE_FLAG2_RESET_REQUESTED);
4242 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4243
4244 del_timer_sync(&adapter->service_timer);
4245
4246 if (adapter->num_vfs) {
4247 /* Clear EITR Select mapping */
4248 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4249
4250 /* Mark all the VFs as inactive */
4251 for (i = 0 ; i < adapter->num_vfs; i++)
4252 adapter->vfinfo[i].clear_to_send = false;
4253
4254 /* ping all the active vfs to let them know we are going down */
4255 ixgbe_ping_all_vfs(adapter);
4256
4257 /* Disable all VFTE/VFRE TX/RX */
4258 ixgbe_disable_tx_rx(adapter);
4259 }
4260
4261 /* disable transmits in the hardware now that interrupts are off */
4262 for (i = 0; i < adapter->num_tx_queues; i++) {
4263 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4264 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4265 }
4266
4267 /* Disable the Tx DMA engine on 82599 and X540 */
4268 switch (hw->mac.type) {
4269 case ixgbe_mac_82599EB:
4270 case ixgbe_mac_X540:
4271 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4272 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4273 ~IXGBE_DMATXCTL_TE));
4274 break;
4275 default:
4276 break;
4277 }
4278
4279 if (!pci_channel_offline(adapter->pdev))
4280 ixgbe_reset(adapter);
4281
4282 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4283 if (hw->mac.ops.disable_tx_laser &&
4284 ((hw->phy.multispeed_fiber) ||
4285 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4286 (hw->mac.type == ixgbe_mac_82599EB))))
4287 hw->mac.ops.disable_tx_laser(hw);
4288
4289 ixgbe_clean_all_tx_rings(adapter);
4290 ixgbe_clean_all_rx_rings(adapter);
4291
4292 #ifdef CONFIG_IXGBE_DCA
4293 /* since we reset the hardware DCA settings were cleared */
4294 ixgbe_setup_dca(adapter);
4295 #endif
4296 }
4297
4298 /**
4299 * ixgbe_tx_timeout - Respond to a Tx Hang
4300 * @netdev: network interface device structure
4301 **/
4302 static void ixgbe_tx_timeout(struct net_device *netdev)
4303 {
4304 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4305
4306 /* Do the reset outside of interrupt context */
4307 ixgbe_tx_timeout_reset(adapter);
4308 }
4309
4310 /**
4311 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4312 * @adapter: board private structure to initialize
4313 *
4314 * ixgbe_sw_init initializes the Adapter private data structure.
4315 * Fields are initialized based on PCI device information and
4316 * OS network device settings (MTU size).
4317 **/
4318 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4319 {
4320 struct ixgbe_hw *hw = &adapter->hw;
4321 struct pci_dev *pdev = adapter->pdev;
4322 unsigned int rss;
4323 #ifdef CONFIG_IXGBE_DCB
4324 int j;
4325 struct tc_configuration *tc;
4326 #endif
4327
4328 /* PCI config space info */
4329
4330 hw->vendor_id = pdev->vendor;
4331 hw->device_id = pdev->device;
4332 hw->revision_id = pdev->revision;
4333 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4334 hw->subsystem_device_id = pdev->subsystem_device;
4335
4336 /* Set capability flags */
4337 rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4338 adapter->ring_feature[RING_F_RSS].indices = rss;
4339 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4340 switch (hw->mac.type) {
4341 case ixgbe_mac_82598EB:
4342 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4343 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4344 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4345 break;
4346 case ixgbe_mac_X540:
4347 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4348 case ixgbe_mac_82599EB:
4349 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4350 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4351 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4352 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4353 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4354 /* Flow Director hash filters enabled */
4355 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4356 adapter->atr_sample_rate = 20;
4357 adapter->ring_feature[RING_F_FDIR].indices =
4358 IXGBE_MAX_FDIR_INDICES;
4359 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4360 #ifdef IXGBE_FCOE
4361 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4362 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4363 adapter->ring_feature[RING_F_FCOE].indices = 0;
4364 #ifdef CONFIG_IXGBE_DCB
4365 /* Default traffic class to use for FCoE */
4366 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4367 #endif
4368 #endif /* IXGBE_FCOE */
4369 break;
4370 default:
4371 break;
4372 }
4373
4374 /* n-tuple support exists, always init our spinlock */
4375 spin_lock_init(&adapter->fdir_perfect_lock);
4376
4377 #ifdef CONFIG_IXGBE_DCB
4378 switch (hw->mac.type) {
4379 case ixgbe_mac_X540:
4380 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4381 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4382 break;
4383 default:
4384 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4385 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4386 break;
4387 }
4388
4389 /* Configure DCB traffic classes */
4390 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4391 tc = &adapter->dcb_cfg.tc_config[j];
4392 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4393 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4394 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4395 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4396 tc->dcb_pfc = pfc_disabled;
4397 }
4398
4399 /* Initialize default user to priority mapping, UPx->TC0 */
4400 tc = &adapter->dcb_cfg.tc_config[0];
4401 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
4402 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
4403
4404 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4405 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4406 adapter->dcb_cfg.pfc_mode_enable = false;
4407 adapter->dcb_set_bitmap = 0x00;
4408 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4409 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
4410 sizeof(adapter->temp_dcb_cfg));
4411
4412 #endif
4413
4414 /* default flow control settings */
4415 hw->fc.requested_mode = ixgbe_fc_full;
4416 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4417 ixgbe_pbthresh_setup(adapter);
4418 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4419 hw->fc.send_xon = true;
4420 hw->fc.disable_fc_autoneg = false;
4421
4422 /* enable itr by default in dynamic mode */
4423 adapter->rx_itr_setting = 1;
4424 adapter->tx_itr_setting = 1;
4425
4426 /* set default ring sizes */
4427 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4428 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4429
4430 /* set default work limits */
4431 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4432
4433 /* initialize eeprom parameters */
4434 if (ixgbe_init_eeprom_params_generic(hw)) {
4435 e_dev_err("EEPROM initialization failed\n");
4436 return -EIO;
4437 }
4438
4439 set_bit(__IXGBE_DOWN, &adapter->state);
4440
4441 return 0;
4442 }
4443
4444 /**
4445 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4446 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4447 *
4448 * Return 0 on success, negative on failure
4449 **/
4450 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4451 {
4452 struct device *dev = tx_ring->dev;
4453 int orig_node = dev_to_node(dev);
4454 int numa_node = -1;
4455 int size;
4456
4457 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4458
4459 if (tx_ring->q_vector)
4460 numa_node = tx_ring->q_vector->numa_node;
4461
4462 tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
4463 if (!tx_ring->tx_buffer_info)
4464 tx_ring->tx_buffer_info = vzalloc(size);
4465 if (!tx_ring->tx_buffer_info)
4466 goto err;
4467
4468 /* round up to nearest 4K */
4469 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4470 tx_ring->size = ALIGN(tx_ring->size, 4096);
4471
4472 set_dev_node(dev, numa_node);
4473 tx_ring->desc = dma_alloc_coherent(dev,
4474 tx_ring->size,
4475 &tx_ring->dma,
4476 GFP_KERNEL);
4477 set_dev_node(dev, orig_node);
4478 if (!tx_ring->desc)
4479 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4480 &tx_ring->dma, GFP_KERNEL);
4481 if (!tx_ring->desc)
4482 goto err;
4483
4484 tx_ring->next_to_use = 0;
4485 tx_ring->next_to_clean = 0;
4486 return 0;
4487
4488 err:
4489 vfree(tx_ring->tx_buffer_info);
4490 tx_ring->tx_buffer_info = NULL;
4491 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4492 return -ENOMEM;
4493 }
4494
4495 /**
4496 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4497 * @adapter: board private structure
4498 *
4499 * If this function returns with an error, then it's possible one or
4500 * more of the rings is populated (while the rest are not). It is the
4501 * callers duty to clean those orphaned rings.
4502 *
4503 * Return 0 on success, negative on failure
4504 **/
4505 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4506 {
4507 int i, err = 0;
4508
4509 for (i = 0; i < adapter->num_tx_queues; i++) {
4510 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4511 if (!err)
4512 continue;
4513 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4514 break;
4515 }
4516
4517 return err;
4518 }
4519
4520 /**
4521 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4522 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4523 *
4524 * Returns 0 on success, negative on failure
4525 **/
4526 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4527 {
4528 struct device *dev = rx_ring->dev;
4529 int orig_node = dev_to_node(dev);
4530 int numa_node = -1;
4531 int size;
4532
4533 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4534
4535 if (rx_ring->q_vector)
4536 numa_node = rx_ring->q_vector->numa_node;
4537
4538 rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
4539 if (!rx_ring->rx_buffer_info)
4540 rx_ring->rx_buffer_info = vzalloc(size);
4541 if (!rx_ring->rx_buffer_info)
4542 goto err;
4543
4544 /* Round up to nearest 4K */
4545 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4546 rx_ring->size = ALIGN(rx_ring->size, 4096);
4547
4548 set_dev_node(dev, numa_node);
4549 rx_ring->desc = dma_alloc_coherent(dev,
4550 rx_ring->size,
4551 &rx_ring->dma,
4552 GFP_KERNEL);
4553 set_dev_node(dev, orig_node);
4554 if (!rx_ring->desc)
4555 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4556 &rx_ring->dma, GFP_KERNEL);
4557 if (!rx_ring->desc)
4558 goto err;
4559
4560 rx_ring->next_to_clean = 0;
4561 rx_ring->next_to_use = 0;
4562
4563 ixgbe_init_rx_page_offset(rx_ring);
4564
4565 return 0;
4566 err:
4567 vfree(rx_ring->rx_buffer_info);
4568 rx_ring->rx_buffer_info = NULL;
4569 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4570 return -ENOMEM;
4571 }
4572
4573 /**
4574 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4575 * @adapter: board private structure
4576 *
4577 * If this function returns with an error, then it's possible one or
4578 * more of the rings is populated (while the rest are not). It is the
4579 * callers duty to clean those orphaned rings.
4580 *
4581 * Return 0 on success, negative on failure
4582 **/
4583 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4584 {
4585 int i, err = 0;
4586
4587 for (i = 0; i < adapter->num_rx_queues; i++) {
4588 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
4589 if (!err)
4590 continue;
4591 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4592 break;
4593 }
4594
4595 return err;
4596 }
4597
4598 /**
4599 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4600 * @tx_ring: Tx descriptor ring for a specific queue
4601 *
4602 * Free all transmit software resources
4603 **/
4604 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
4605 {
4606 ixgbe_clean_tx_ring(tx_ring);
4607
4608 vfree(tx_ring->tx_buffer_info);
4609 tx_ring->tx_buffer_info = NULL;
4610
4611 /* if not set, then don't free */
4612 if (!tx_ring->desc)
4613 return;
4614
4615 dma_free_coherent(tx_ring->dev, tx_ring->size,
4616 tx_ring->desc, tx_ring->dma);
4617
4618 tx_ring->desc = NULL;
4619 }
4620
4621 /**
4622 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4623 * @adapter: board private structure
4624 *
4625 * Free all transmit software resources
4626 **/
4627 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4628 {
4629 int i;
4630
4631 for (i = 0; i < adapter->num_tx_queues; i++)
4632 if (adapter->tx_ring[i]->desc)
4633 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4634 }
4635
4636 /**
4637 * ixgbe_free_rx_resources - Free Rx Resources
4638 * @rx_ring: ring to clean the resources from
4639 *
4640 * Free all receive software resources
4641 **/
4642 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
4643 {
4644 ixgbe_clean_rx_ring(rx_ring);
4645
4646 vfree(rx_ring->rx_buffer_info);
4647 rx_ring->rx_buffer_info = NULL;
4648
4649 /* if not set, then don't free */
4650 if (!rx_ring->desc)
4651 return;
4652
4653 dma_free_coherent(rx_ring->dev, rx_ring->size,
4654 rx_ring->desc, rx_ring->dma);
4655
4656 rx_ring->desc = NULL;
4657 }
4658
4659 /**
4660 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4661 * @adapter: board private structure
4662 *
4663 * Free all receive software resources
4664 **/
4665 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4666 {
4667 int i;
4668
4669 for (i = 0; i < adapter->num_rx_queues; i++)
4670 if (adapter->rx_ring[i]->desc)
4671 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4672 }
4673
4674 /**
4675 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4676 * @netdev: network interface device structure
4677 * @new_mtu: new value for maximum frame size
4678 *
4679 * Returns 0 on success, negative on failure
4680 **/
4681 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4682 {
4683 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4684 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4685
4686 /* MTU < 68 is an error and causes problems on some kernels */
4687 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4688 return -EINVAL;
4689
4690 /*
4691 * For 82599EB we cannot allow PF to change MTU greater than 1500
4692 * in SR-IOV mode as it may cause buffer overruns in guest VFs that
4693 * don't allocate and chain buffers correctly.
4694 */
4695 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
4696 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
4697 (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
4698 return -EINVAL;
4699
4700 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4701
4702 /* must set new MTU before calling down or up */
4703 netdev->mtu = new_mtu;
4704
4705 if (netif_running(netdev))
4706 ixgbe_reinit_locked(adapter);
4707
4708 return 0;
4709 }
4710
4711 /**
4712 * ixgbe_open - Called when a network interface is made active
4713 * @netdev: network interface device structure
4714 *
4715 * Returns 0 on success, negative value on failure
4716 *
4717 * The open entry point is called when a network interface is made
4718 * active by the system (IFF_UP). At this point all resources needed
4719 * for transmit and receive operations are allocated, the interrupt
4720 * handler is registered with the OS, the watchdog timer is started,
4721 * and the stack is notified that the interface is ready.
4722 **/
4723 static int ixgbe_open(struct net_device *netdev)
4724 {
4725 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4726 int err;
4727
4728 /* disallow open during test */
4729 if (test_bit(__IXGBE_TESTING, &adapter->state))
4730 return -EBUSY;
4731
4732 netif_carrier_off(netdev);
4733
4734 /* allocate transmit descriptors */
4735 err = ixgbe_setup_all_tx_resources(adapter);
4736 if (err)
4737 goto err_setup_tx;
4738
4739 /* allocate receive descriptors */
4740 err = ixgbe_setup_all_rx_resources(adapter);
4741 if (err)
4742 goto err_setup_rx;
4743
4744 ixgbe_configure(adapter);
4745
4746 err = ixgbe_request_irq(adapter);
4747 if (err)
4748 goto err_req_irq;
4749
4750 ixgbe_up_complete(adapter);
4751
4752 return 0;
4753
4754 err_req_irq:
4755 err_setup_rx:
4756 ixgbe_free_all_rx_resources(adapter);
4757 err_setup_tx:
4758 ixgbe_free_all_tx_resources(adapter);
4759 ixgbe_reset(adapter);
4760
4761 return err;
4762 }
4763
4764 /**
4765 * ixgbe_close - Disables a network interface
4766 * @netdev: network interface device structure
4767 *
4768 * Returns 0, this is not allowed to fail
4769 *
4770 * The close entry point is called when an interface is de-activated
4771 * by the OS. The hardware is still under the drivers control, but
4772 * needs to be disabled. A global MAC reset is issued to stop the
4773 * hardware, and all transmit and receive resources are freed.
4774 **/
4775 static int ixgbe_close(struct net_device *netdev)
4776 {
4777 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4778
4779 ixgbe_down(adapter);
4780 ixgbe_free_irq(adapter);
4781
4782 ixgbe_fdir_filter_exit(adapter);
4783
4784 ixgbe_free_all_tx_resources(adapter);
4785 ixgbe_free_all_rx_resources(adapter);
4786
4787 ixgbe_release_hw_control(adapter);
4788
4789 return 0;
4790 }
4791
4792 #ifdef CONFIG_PM
4793 static int ixgbe_resume(struct pci_dev *pdev)
4794 {
4795 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4796 struct net_device *netdev = adapter->netdev;
4797 u32 err;
4798
4799 pci_set_power_state(pdev, PCI_D0);
4800 pci_restore_state(pdev);
4801 /*
4802 * pci_restore_state clears dev->state_saved so call
4803 * pci_save_state to restore it.
4804 */
4805 pci_save_state(pdev);
4806
4807 err = pci_enable_device_mem(pdev);
4808 if (err) {
4809 e_dev_err("Cannot enable PCI device from suspend\n");
4810 return err;
4811 }
4812 pci_set_master(pdev);
4813
4814 pci_wake_from_d3(pdev, false);
4815
4816 rtnl_lock();
4817 err = ixgbe_init_interrupt_scheme(adapter);
4818 rtnl_unlock();
4819 if (err) {
4820 e_dev_err("Cannot initialize interrupts for device\n");
4821 return err;
4822 }
4823
4824 ixgbe_reset(adapter);
4825
4826 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4827
4828 if (netif_running(netdev)) {
4829 err = ixgbe_open(netdev);
4830 if (err)
4831 return err;
4832 }
4833
4834 netif_device_attach(netdev);
4835
4836 return 0;
4837 }
4838 #endif /* CONFIG_PM */
4839
4840 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4841 {
4842 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4843 struct net_device *netdev = adapter->netdev;
4844 struct ixgbe_hw *hw = &adapter->hw;
4845 u32 ctrl, fctrl;
4846 u32 wufc = adapter->wol;
4847 #ifdef CONFIG_PM
4848 int retval = 0;
4849 #endif
4850
4851 netif_device_detach(netdev);
4852
4853 if (netif_running(netdev)) {
4854 rtnl_lock();
4855 ixgbe_down(adapter);
4856 ixgbe_free_irq(adapter);
4857 ixgbe_free_all_tx_resources(adapter);
4858 ixgbe_free_all_rx_resources(adapter);
4859 rtnl_unlock();
4860 }
4861
4862 ixgbe_clear_interrupt_scheme(adapter);
4863
4864 #ifdef CONFIG_PM
4865 retval = pci_save_state(pdev);
4866 if (retval)
4867 return retval;
4868
4869 #endif
4870 if (wufc) {
4871 ixgbe_set_rx_mode(netdev);
4872
4873 /*
4874 * enable the optics for both mult-speed fiber and
4875 * 82599 SFP+ fiber as we can WoL.
4876 */
4877 if (hw->mac.ops.enable_tx_laser &&
4878 (hw->phy.multispeed_fiber ||
4879 (hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber &&
4880 hw->mac.type == ixgbe_mac_82599EB)))
4881 hw->mac.ops.enable_tx_laser(hw);
4882
4883 /* turn on all-multi mode if wake on multicast is enabled */
4884 if (wufc & IXGBE_WUFC_MC) {
4885 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4886 fctrl |= IXGBE_FCTRL_MPE;
4887 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4888 }
4889
4890 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4891 ctrl |= IXGBE_CTRL_GIO_DIS;
4892 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4893
4894 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4895 } else {
4896 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4897 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4898 }
4899
4900 switch (hw->mac.type) {
4901 case ixgbe_mac_82598EB:
4902 pci_wake_from_d3(pdev, false);
4903 break;
4904 case ixgbe_mac_82599EB:
4905 case ixgbe_mac_X540:
4906 pci_wake_from_d3(pdev, !!wufc);
4907 break;
4908 default:
4909 break;
4910 }
4911
4912 *enable_wake = !!wufc;
4913
4914 ixgbe_release_hw_control(adapter);
4915
4916 pci_disable_device(pdev);
4917
4918 return 0;
4919 }
4920
4921 #ifdef CONFIG_PM
4922 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4923 {
4924 int retval;
4925 bool wake;
4926
4927 retval = __ixgbe_shutdown(pdev, &wake);
4928 if (retval)
4929 return retval;
4930
4931 if (wake) {
4932 pci_prepare_to_sleep(pdev);
4933 } else {
4934 pci_wake_from_d3(pdev, false);
4935 pci_set_power_state(pdev, PCI_D3hot);
4936 }
4937
4938 return 0;
4939 }
4940 #endif /* CONFIG_PM */
4941
4942 static void ixgbe_shutdown(struct pci_dev *pdev)
4943 {
4944 bool wake;
4945
4946 __ixgbe_shutdown(pdev, &wake);
4947
4948 if (system_state == SYSTEM_POWER_OFF) {
4949 pci_wake_from_d3(pdev, wake);
4950 pci_set_power_state(pdev, PCI_D3hot);
4951 }
4952 }
4953
4954 /**
4955 * ixgbe_update_stats - Update the board statistics counters.
4956 * @adapter: board private structure
4957 **/
4958 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4959 {
4960 struct net_device *netdev = adapter->netdev;
4961 struct ixgbe_hw *hw = &adapter->hw;
4962 struct ixgbe_hw_stats *hwstats = &adapter->stats;
4963 u64 total_mpc = 0;
4964 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4965 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
4966 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
4967 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
4968 #ifdef IXGBE_FCOE
4969 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4970 unsigned int cpu;
4971 u64 fcoe_noddp_counts_sum = 0, fcoe_noddp_ext_buff_counts_sum = 0;
4972 #endif /* IXGBE_FCOE */
4973
4974 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
4975 test_bit(__IXGBE_RESETTING, &adapter->state))
4976 return;
4977
4978 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
4979 u64 rsc_count = 0;
4980 u64 rsc_flush = 0;
4981 for (i = 0; i < adapter->num_rx_queues; i++) {
4982 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
4983 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
4984 }
4985 adapter->rsc_total_count = rsc_count;
4986 adapter->rsc_total_flush = rsc_flush;
4987 }
4988
4989 for (i = 0; i < adapter->num_rx_queues; i++) {
4990 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
4991 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
4992 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
4993 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
4994 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
4995 bytes += rx_ring->stats.bytes;
4996 packets += rx_ring->stats.packets;
4997 }
4998 adapter->non_eop_descs = non_eop_descs;
4999 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5000 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5001 adapter->hw_csum_rx_error = hw_csum_rx_error;
5002 netdev->stats.rx_bytes = bytes;
5003 netdev->stats.rx_packets = packets;
5004
5005 bytes = 0;
5006 packets = 0;
5007 /* gather some stats to the adapter struct that are per queue */
5008 for (i = 0; i < adapter->num_tx_queues; i++) {
5009 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5010 restart_queue += tx_ring->tx_stats.restart_queue;
5011 tx_busy += tx_ring->tx_stats.tx_busy;
5012 bytes += tx_ring->stats.bytes;
5013 packets += tx_ring->stats.packets;
5014 }
5015 adapter->restart_queue = restart_queue;
5016 adapter->tx_busy = tx_busy;
5017 netdev->stats.tx_bytes = bytes;
5018 netdev->stats.tx_packets = packets;
5019
5020 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5021
5022 /* 8 register reads */
5023 for (i = 0; i < 8; i++) {
5024 /* for packet buffers not used, the register should read 0 */
5025 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5026 missed_rx += mpc;
5027 hwstats->mpc[i] += mpc;
5028 total_mpc += hwstats->mpc[i];
5029 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5030 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5031 switch (hw->mac.type) {
5032 case ixgbe_mac_82598EB:
5033 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5034 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5035 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5036 hwstats->pxonrxc[i] +=
5037 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5038 break;
5039 case ixgbe_mac_82599EB:
5040 case ixgbe_mac_X540:
5041 hwstats->pxonrxc[i] +=
5042 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5043 break;
5044 default:
5045 break;
5046 }
5047 }
5048
5049 /*16 register reads */
5050 for (i = 0; i < 16; i++) {
5051 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5052 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5053 if ((hw->mac.type == ixgbe_mac_82599EB) ||
5054 (hw->mac.type == ixgbe_mac_X540)) {
5055 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5056 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5057 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5058 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5059 }
5060 }
5061
5062 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5063 /* work around hardware counting issue */
5064 hwstats->gprc -= missed_rx;
5065
5066 ixgbe_update_xoff_received(adapter);
5067
5068 /* 82598 hardware only has a 32 bit counter in the high register */
5069 switch (hw->mac.type) {
5070 case ixgbe_mac_82598EB:
5071 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5072 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5073 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5074 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5075 break;
5076 case ixgbe_mac_X540:
5077 /* OS2BMC stats are X540 only*/
5078 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5079 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5080 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5081 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5082 case ixgbe_mac_82599EB:
5083 for (i = 0; i < 16; i++)
5084 adapter->hw_rx_no_dma_resources +=
5085 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5086 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5087 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5088 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5089 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5090 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5091 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5092 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5093 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5094 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5095 #ifdef IXGBE_FCOE
5096 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5097 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5098 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5099 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5100 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5101 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5102 /* Add up per cpu counters for total ddp aloc fail */
5103 if (fcoe->pcpu_noddp && fcoe->pcpu_noddp_ext_buff) {
5104 for_each_possible_cpu(cpu) {
5105 fcoe_noddp_counts_sum +=
5106 *per_cpu_ptr(fcoe->pcpu_noddp, cpu);
5107 fcoe_noddp_ext_buff_counts_sum +=
5108 *per_cpu_ptr(fcoe->
5109 pcpu_noddp_ext_buff, cpu);
5110 }
5111 }
5112 hwstats->fcoe_noddp = fcoe_noddp_counts_sum;
5113 hwstats->fcoe_noddp_ext_buff = fcoe_noddp_ext_buff_counts_sum;
5114 #endif /* IXGBE_FCOE */
5115 break;
5116 default:
5117 break;
5118 }
5119 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5120 hwstats->bprc += bprc;
5121 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5122 if (hw->mac.type == ixgbe_mac_82598EB)
5123 hwstats->mprc -= bprc;
5124 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5125 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5126 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5127 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5128 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5129 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5130 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5131 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5132 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5133 hwstats->lxontxc += lxon;
5134 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5135 hwstats->lxofftxc += lxoff;
5136 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5137 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5138 /*
5139 * 82598 errata - tx of flow control packets is included in tx counters
5140 */
5141 xon_off_tot = lxon + lxoff;
5142 hwstats->gptc -= xon_off_tot;
5143 hwstats->mptc -= xon_off_tot;
5144 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5145 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5146 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5147 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5148 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5149 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5150 hwstats->ptc64 -= xon_off_tot;
5151 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5152 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5153 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5154 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5155 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5156 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5157
5158 /* Fill out the OS statistics structure */
5159 netdev->stats.multicast = hwstats->mprc;
5160
5161 /* Rx Errors */
5162 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5163 netdev->stats.rx_dropped = 0;
5164 netdev->stats.rx_length_errors = hwstats->rlec;
5165 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5166 netdev->stats.rx_missed_errors = total_mpc;
5167 }
5168
5169 /**
5170 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5171 * @adapter - pointer to the device adapter structure
5172 **/
5173 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5174 {
5175 struct ixgbe_hw *hw = &adapter->hw;
5176 int i;
5177
5178 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5179 return;
5180
5181 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5182
5183 /* if interface is down do nothing */
5184 if (test_bit(__IXGBE_DOWN, &adapter->state))
5185 return;
5186
5187 /* do nothing if we are not using signature filters */
5188 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5189 return;
5190
5191 adapter->fdir_overflow++;
5192
5193 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5194 for (i = 0; i < adapter->num_tx_queues; i++)
5195 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5196 &(adapter->tx_ring[i]->state));
5197 /* re-enable flow director interrupts */
5198 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5199 } else {
5200 e_err(probe, "failed to finish FDIR re-initialization, "
5201 "ignored adding FDIR ATR filters\n");
5202 }
5203 }
5204
5205 /**
5206 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5207 * @adapter - pointer to the device adapter structure
5208 *
5209 * This function serves two purposes. First it strobes the interrupt lines
5210 * in order to make certain interrupts are occurring. Secondly it sets the
5211 * bits needed to check for TX hangs. As a result we should immediately
5212 * determine if a hang has occurred.
5213 */
5214 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5215 {
5216 struct ixgbe_hw *hw = &adapter->hw;
5217 u64 eics = 0;
5218 int i;
5219
5220 /* If we're down or resetting, just bail */
5221 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5222 test_bit(__IXGBE_RESETTING, &adapter->state))
5223 return;
5224
5225 /* Force detection of hung controller */
5226 if (netif_carrier_ok(adapter->netdev)) {
5227 for (i = 0; i < adapter->num_tx_queues; i++)
5228 set_check_for_tx_hang(adapter->tx_ring[i]);
5229 }
5230
5231 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5232 /*
5233 * for legacy and MSI interrupts don't set any bits
5234 * that are enabled for EIAM, because this operation
5235 * would set *both* EIMS and EICS for any bit in EIAM
5236 */
5237 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5238 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5239 } else {
5240 /* get one bit for every active tx/rx interrupt vector */
5241 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5242 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5243 if (qv->rx.ring || qv->tx.ring)
5244 eics |= ((u64)1 << i);
5245 }
5246 }
5247
5248 /* Cause software interrupt to ensure rings are cleaned */
5249 ixgbe_irq_rearm_queues(adapter, eics);
5250
5251 }
5252
5253 /**
5254 * ixgbe_watchdog_update_link - update the link status
5255 * @adapter - pointer to the device adapter structure
5256 * @link_speed - pointer to a u32 to store the link_speed
5257 **/
5258 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5259 {
5260 struct ixgbe_hw *hw = &adapter->hw;
5261 u32 link_speed = adapter->link_speed;
5262 bool link_up = adapter->link_up;
5263 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
5264
5265 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5266 return;
5267
5268 if (hw->mac.ops.check_link) {
5269 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5270 } else {
5271 /* always assume link is up, if no check link function */
5272 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5273 link_up = true;
5274 }
5275
5276 if (adapter->ixgbe_ieee_pfc)
5277 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
5278
5279 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en))
5280 hw->mac.ops.fc_enable(hw);
5281
5282 if (link_up ||
5283 time_after(jiffies, (adapter->link_check_timeout +
5284 IXGBE_TRY_LINK_TIMEOUT))) {
5285 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5286 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5287 IXGBE_WRITE_FLUSH(hw);
5288 }
5289
5290 adapter->link_up = link_up;
5291 adapter->link_speed = link_speed;
5292 }
5293
5294 /**
5295 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5296 * print link up message
5297 * @adapter - pointer to the device adapter structure
5298 **/
5299 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5300 {
5301 struct net_device *netdev = adapter->netdev;
5302 struct ixgbe_hw *hw = &adapter->hw;
5303 u32 link_speed = adapter->link_speed;
5304 bool flow_rx, flow_tx;
5305
5306 /* only continue if link was previously down */
5307 if (netif_carrier_ok(netdev))
5308 return;
5309
5310 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5311
5312 switch (hw->mac.type) {
5313 case ixgbe_mac_82598EB: {
5314 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5315 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5316 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5317 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5318 }
5319 break;
5320 case ixgbe_mac_X540:
5321 case ixgbe_mac_82599EB: {
5322 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5323 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5324 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5325 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5326 }
5327 break;
5328 default:
5329 flow_tx = false;
5330 flow_rx = false;
5331 break;
5332 }
5333 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5334 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5335 "10 Gbps" :
5336 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5337 "1 Gbps" :
5338 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5339 "100 Mbps" :
5340 "unknown speed"))),
5341 ((flow_rx && flow_tx) ? "RX/TX" :
5342 (flow_rx ? "RX" :
5343 (flow_tx ? "TX" : "None"))));
5344
5345 netif_carrier_on(netdev);
5346 ixgbe_check_vf_rate_limit(adapter);
5347 }
5348
5349 /**
5350 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5351 * print link down message
5352 * @adapter - pointer to the adapter structure
5353 **/
5354 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
5355 {
5356 struct net_device *netdev = adapter->netdev;
5357 struct ixgbe_hw *hw = &adapter->hw;
5358
5359 adapter->link_up = false;
5360 adapter->link_speed = 0;
5361
5362 /* only continue if link was up previously */
5363 if (!netif_carrier_ok(netdev))
5364 return;
5365
5366 /* poll for SFP+ cable when link is down */
5367 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5368 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5369
5370 e_info(drv, "NIC Link is Down\n");
5371 netif_carrier_off(netdev);
5372 }
5373
5374 /**
5375 * ixgbe_watchdog_flush_tx - flush queues on link down
5376 * @adapter - pointer to the device adapter structure
5377 **/
5378 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5379 {
5380 int i;
5381 int some_tx_pending = 0;
5382
5383 if (!netif_carrier_ok(adapter->netdev)) {
5384 for (i = 0; i < adapter->num_tx_queues; i++) {
5385 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5386 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5387 some_tx_pending = 1;
5388 break;
5389 }
5390 }
5391
5392 if (some_tx_pending) {
5393 /* We've lost link, so the controller stops DMA,
5394 * but we've got queued Tx work that's never going
5395 * to get done, so reset controller to flush Tx.
5396 * (Do the reset outside of interrupt context).
5397 */
5398 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5399 }
5400 }
5401 }
5402
5403 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5404 {
5405 u32 ssvpc;
5406
5407 /* Do not perform spoof check for 82598 */
5408 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5409 return;
5410
5411 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5412
5413 /*
5414 * ssvpc register is cleared on read, if zero then no
5415 * spoofed packets in the last interval.
5416 */
5417 if (!ssvpc)
5418 return;
5419
5420 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
5421 }
5422
5423 /**
5424 * ixgbe_watchdog_subtask - check and bring link up
5425 * @adapter - pointer to the device adapter structure
5426 **/
5427 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5428 {
5429 /* if interface is down do nothing */
5430 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5431 test_bit(__IXGBE_RESETTING, &adapter->state))
5432 return;
5433
5434 ixgbe_watchdog_update_link(adapter);
5435
5436 if (adapter->link_up)
5437 ixgbe_watchdog_link_is_up(adapter);
5438 else
5439 ixgbe_watchdog_link_is_down(adapter);
5440
5441 ixgbe_spoof_check(adapter);
5442 ixgbe_update_stats(adapter);
5443
5444 ixgbe_watchdog_flush_tx(adapter);
5445 }
5446
5447 /**
5448 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5449 * @adapter - the ixgbe adapter structure
5450 **/
5451 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5452 {
5453 struct ixgbe_hw *hw = &adapter->hw;
5454 s32 err;
5455
5456 /* not searching for SFP so there is nothing to do here */
5457 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5458 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5459 return;
5460
5461 /* someone else is in init, wait until next service event */
5462 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5463 return;
5464
5465 err = hw->phy.ops.identify_sfp(hw);
5466 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5467 goto sfp_out;
5468
5469 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5470 /* If no cable is present, then we need to reset
5471 * the next time we find a good cable. */
5472 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5473 }
5474
5475 /* exit on error */
5476 if (err)
5477 goto sfp_out;
5478
5479 /* exit if reset not needed */
5480 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5481 goto sfp_out;
5482
5483 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5484
5485 /*
5486 * A module may be identified correctly, but the EEPROM may not have
5487 * support for that module. setup_sfp() will fail in that case, so
5488 * we should not allow that module to load.
5489 */
5490 if (hw->mac.type == ixgbe_mac_82598EB)
5491 err = hw->phy.ops.reset(hw);
5492 else
5493 err = hw->mac.ops.setup_sfp(hw);
5494
5495 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5496 goto sfp_out;
5497
5498 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5499 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5500
5501 sfp_out:
5502 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5503
5504 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5505 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5506 e_dev_err("failed to initialize because an unsupported "
5507 "SFP+ module type was detected.\n");
5508 e_dev_err("Reload the driver after installing a "
5509 "supported module.\n");
5510 unregister_netdev(adapter->netdev);
5511 }
5512 }
5513
5514 /**
5515 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5516 * @adapter - the ixgbe adapter structure
5517 **/
5518 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5519 {
5520 struct ixgbe_hw *hw = &adapter->hw;
5521 u32 autoneg;
5522 bool negotiation;
5523
5524 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5525 return;
5526
5527 /* someone else is in init, wait until next service event */
5528 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5529 return;
5530
5531 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5532
5533 autoneg = hw->phy.autoneg_advertised;
5534 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5535 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5536 if (hw->mac.ops.setup_link)
5537 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5538
5539 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5540 adapter->link_check_timeout = jiffies;
5541 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5542 }
5543
5544 #ifdef CONFIG_PCI_IOV
5545 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
5546 {
5547 int vf;
5548 struct ixgbe_hw *hw = &adapter->hw;
5549 struct net_device *netdev = adapter->netdev;
5550 u32 gpc;
5551 u32 ciaa, ciad;
5552
5553 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
5554 if (gpc) /* If incrementing then no need for the check below */
5555 return;
5556 /*
5557 * Check to see if a bad DMA write target from an errant or
5558 * malicious VF has caused a PCIe error. If so then we can
5559 * issue a VFLR to the offending VF(s) and then resume without
5560 * requesting a full slot reset.
5561 */
5562
5563 for (vf = 0; vf < adapter->num_vfs; vf++) {
5564 ciaa = (vf << 16) | 0x80000000;
5565 /* 32 bit read so align, we really want status at offset 6 */
5566 ciaa |= PCI_COMMAND;
5567 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5568 ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
5569 ciaa &= 0x7FFFFFFF;
5570 /* disable debug mode asap after reading data */
5571 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5572 /* Get the upper 16 bits which will be the PCI status reg */
5573 ciad >>= 16;
5574 if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
5575 netdev_err(netdev, "VF %d Hung DMA\n", vf);
5576 /* Issue VFLR */
5577 ciaa = (vf << 16) | 0x80000000;
5578 ciaa |= 0xA8;
5579 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5580 ciad = 0x00008000; /* VFLR */
5581 IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
5582 ciaa &= 0x7FFFFFFF;
5583 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5584 }
5585 }
5586 }
5587
5588 #endif
5589 /**
5590 * ixgbe_service_timer - Timer Call-back
5591 * @data: pointer to adapter cast into an unsigned long
5592 **/
5593 static void ixgbe_service_timer(unsigned long data)
5594 {
5595 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5596 unsigned long next_event_offset;
5597 bool ready = true;
5598
5599 /* poll faster when waiting for link */
5600 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5601 next_event_offset = HZ / 10;
5602 else
5603 next_event_offset = HZ * 2;
5604
5605 #ifdef CONFIG_PCI_IOV
5606 /*
5607 * don't bother with SR-IOV VF DMA hang check if there are
5608 * no VFs or the link is down
5609 */
5610 if (!adapter->num_vfs ||
5611 (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5612 goto normal_timer_service;
5613
5614 /* If we have VFs allocated then we must check for DMA hangs */
5615 ixgbe_check_for_bad_vf(adapter);
5616 next_event_offset = HZ / 50;
5617 adapter->timer_event_accumulator++;
5618
5619 if (adapter->timer_event_accumulator >= 100)
5620 adapter->timer_event_accumulator = 0;
5621 else
5622 ready = false;
5623
5624 normal_timer_service:
5625 #endif
5626 /* Reset the timer */
5627 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5628
5629 if (ready)
5630 ixgbe_service_event_schedule(adapter);
5631 }
5632
5633 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5634 {
5635 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
5636 return;
5637
5638 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
5639
5640 /* If we're already down or resetting, just bail */
5641 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5642 test_bit(__IXGBE_RESETTING, &adapter->state))
5643 return;
5644
5645 ixgbe_dump(adapter);
5646 netdev_err(adapter->netdev, "Reset adapter\n");
5647 adapter->tx_timeout_count++;
5648
5649 ixgbe_reinit_locked(adapter);
5650 }
5651
5652 /**
5653 * ixgbe_service_task - manages and runs subtasks
5654 * @work: pointer to work_struct containing our data
5655 **/
5656 static void ixgbe_service_task(struct work_struct *work)
5657 {
5658 struct ixgbe_adapter *adapter = container_of(work,
5659 struct ixgbe_adapter,
5660 service_task);
5661
5662 ixgbe_reset_subtask(adapter);
5663 ixgbe_sfp_detection_subtask(adapter);
5664 ixgbe_sfp_link_config_subtask(adapter);
5665 ixgbe_check_overtemp_subtask(adapter);
5666 ixgbe_watchdog_subtask(adapter);
5667 ixgbe_fdir_reinit_subtask(adapter);
5668 ixgbe_check_hang_subtask(adapter);
5669
5670 ixgbe_service_event_complete(adapter);
5671 }
5672
5673 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
5674 struct ixgbe_tx_buffer *first,
5675 u8 *hdr_len)
5676 {
5677 struct sk_buff *skb = first->skb;
5678 u32 vlan_macip_lens, type_tucmd;
5679 u32 mss_l4len_idx, l4len;
5680
5681 if (!skb_is_gso(skb))
5682 return 0;
5683
5684 if (skb_header_cloned(skb)) {
5685 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5686 if (err)
5687 return err;
5688 }
5689
5690 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5691 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
5692
5693 if (first->protocol == __constant_htons(ETH_P_IP)) {
5694 struct iphdr *iph = ip_hdr(skb);
5695 iph->tot_len = 0;
5696 iph->check = 0;
5697 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5698 iph->daddr, 0,
5699 IPPROTO_TCP,
5700 0);
5701 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5702 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5703 IXGBE_TX_FLAGS_CSUM |
5704 IXGBE_TX_FLAGS_IPV4;
5705 } else if (skb_is_gso_v6(skb)) {
5706 ipv6_hdr(skb)->payload_len = 0;
5707 tcp_hdr(skb)->check =
5708 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5709 &ipv6_hdr(skb)->daddr,
5710 0, IPPROTO_TCP, 0);
5711 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5712 IXGBE_TX_FLAGS_CSUM;
5713 }
5714
5715 /* compute header lengths */
5716 l4len = tcp_hdrlen(skb);
5717 *hdr_len = skb_transport_offset(skb) + l4len;
5718
5719 /* update gso size and bytecount with header size */
5720 first->gso_segs = skb_shinfo(skb)->gso_segs;
5721 first->bytecount += (first->gso_segs - 1) * *hdr_len;
5722
5723 /* mss_l4len_id: use 1 as index for TSO */
5724 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
5725 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
5726 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
5727
5728 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
5729 vlan_macip_lens = skb_network_header_len(skb);
5730 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5731 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5732
5733 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
5734 mss_l4len_idx);
5735
5736 return 1;
5737 }
5738
5739 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
5740 struct ixgbe_tx_buffer *first)
5741 {
5742 struct sk_buff *skb = first->skb;
5743 u32 vlan_macip_lens = 0;
5744 u32 mss_l4len_idx = 0;
5745 u32 type_tucmd = 0;
5746
5747 if (skb->ip_summed != CHECKSUM_PARTIAL) {
5748 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
5749 !(first->tx_flags & IXGBE_TX_FLAGS_TXSW))
5750 return;
5751 } else {
5752 u8 l4_hdr = 0;
5753 switch (first->protocol) {
5754 case __constant_htons(ETH_P_IP):
5755 vlan_macip_lens |= skb_network_header_len(skb);
5756 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5757 l4_hdr = ip_hdr(skb)->protocol;
5758 break;
5759 case __constant_htons(ETH_P_IPV6):
5760 vlan_macip_lens |= skb_network_header_len(skb);
5761 l4_hdr = ipv6_hdr(skb)->nexthdr;
5762 break;
5763 default:
5764 if (unlikely(net_ratelimit())) {
5765 dev_warn(tx_ring->dev,
5766 "partial checksum but proto=%x!\n",
5767 first->protocol);
5768 }
5769 break;
5770 }
5771
5772 switch (l4_hdr) {
5773 case IPPROTO_TCP:
5774 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5775 mss_l4len_idx = tcp_hdrlen(skb) <<
5776 IXGBE_ADVTXD_L4LEN_SHIFT;
5777 break;
5778 case IPPROTO_SCTP:
5779 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5780 mss_l4len_idx = sizeof(struct sctphdr) <<
5781 IXGBE_ADVTXD_L4LEN_SHIFT;
5782 break;
5783 case IPPROTO_UDP:
5784 mss_l4len_idx = sizeof(struct udphdr) <<
5785 IXGBE_ADVTXD_L4LEN_SHIFT;
5786 break;
5787 default:
5788 if (unlikely(net_ratelimit())) {
5789 dev_warn(tx_ring->dev,
5790 "partial checksum but l4 proto=%x!\n",
5791 l4_hdr);
5792 }
5793 break;
5794 }
5795
5796 /* update TX checksum flag */
5797 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
5798 }
5799
5800 /* vlan_macip_lens: MACLEN, VLAN tag */
5801 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5802 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5803
5804 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
5805 type_tucmd, mss_l4len_idx);
5806 }
5807
5808 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
5809 {
5810 /* set type for advanced descriptor with frame checksum insertion */
5811 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
5812 IXGBE_ADVTXD_DCMD_IFCS |
5813 IXGBE_ADVTXD_DCMD_DEXT);
5814
5815 /* set HW vlan bit if vlan is present */
5816 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
5817 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
5818
5819 /* set segmentation enable bits for TSO/FSO */
5820 #ifdef IXGBE_FCOE
5821 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FSO))
5822 #else
5823 if (tx_flags & IXGBE_TX_FLAGS_TSO)
5824 #endif
5825 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
5826
5827 return cmd_type;
5828 }
5829
5830 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
5831 u32 tx_flags, unsigned int paylen)
5832 {
5833 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
5834
5835 /* enable L4 checksum for TSO and TX checksum offload */
5836 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5837 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
5838
5839 /* enble IPv4 checksum for TSO */
5840 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5841 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
5842
5843 /* use index 1 context for TSO/FSO/FCOE */
5844 #ifdef IXGBE_FCOE
5845 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FCOE))
5846 #else
5847 if (tx_flags & IXGBE_TX_FLAGS_TSO)
5848 #endif
5849 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
5850
5851 /*
5852 * Check Context must be set if Tx switch is enabled, which it
5853 * always is for case where virtual functions are running
5854 */
5855 #ifdef IXGBE_FCOE
5856 if (tx_flags & (IXGBE_TX_FLAGS_TXSW | IXGBE_TX_FLAGS_FCOE))
5857 #else
5858 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
5859 #endif
5860 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
5861
5862 tx_desc->read.olinfo_status = olinfo_status;
5863 }
5864
5865 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
5866 IXGBE_TXD_CMD_RS)
5867
5868 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
5869 struct ixgbe_tx_buffer *first,
5870 const u8 hdr_len)
5871 {
5872 dma_addr_t dma;
5873 struct sk_buff *skb = first->skb;
5874 struct ixgbe_tx_buffer *tx_buffer;
5875 union ixgbe_adv_tx_desc *tx_desc;
5876 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
5877 unsigned int data_len = skb->data_len;
5878 unsigned int size = skb_headlen(skb);
5879 unsigned int paylen = skb->len - hdr_len;
5880 u32 tx_flags = first->tx_flags;
5881 __le32 cmd_type;
5882 u16 i = tx_ring->next_to_use;
5883
5884 tx_desc = IXGBE_TX_DESC(tx_ring, i);
5885
5886 ixgbe_tx_olinfo_status(tx_desc, tx_flags, paylen);
5887 cmd_type = ixgbe_tx_cmd_type(tx_flags);
5888
5889 #ifdef IXGBE_FCOE
5890 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5891 if (data_len < sizeof(struct fcoe_crc_eof)) {
5892 size -= sizeof(struct fcoe_crc_eof) - data_len;
5893 data_len = 0;
5894 } else {
5895 data_len -= sizeof(struct fcoe_crc_eof);
5896 }
5897 }
5898
5899 #endif
5900 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
5901 if (dma_mapping_error(tx_ring->dev, dma))
5902 goto dma_error;
5903
5904 /* record length, and DMA address */
5905 dma_unmap_len_set(first, len, size);
5906 dma_unmap_addr_set(first, dma, dma);
5907
5908 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5909
5910 for (;;) {
5911 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
5912 tx_desc->read.cmd_type_len =
5913 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
5914
5915 i++;
5916 tx_desc++;
5917 if (i == tx_ring->count) {
5918 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5919 i = 0;
5920 }
5921
5922 dma += IXGBE_MAX_DATA_PER_TXD;
5923 size -= IXGBE_MAX_DATA_PER_TXD;
5924
5925 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5926 tx_desc->read.olinfo_status = 0;
5927 }
5928
5929 if (likely(!data_len))
5930 break;
5931
5932 if (unlikely(skb->no_fcs))
5933 cmd_type &= ~(cpu_to_le32(IXGBE_ADVTXD_DCMD_IFCS));
5934 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
5935
5936 i++;
5937 tx_desc++;
5938 if (i == tx_ring->count) {
5939 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5940 i = 0;
5941 }
5942
5943 #ifdef IXGBE_FCOE
5944 size = min_t(unsigned int, data_len, skb_frag_size(frag));
5945 #else
5946 size = skb_frag_size(frag);
5947 #endif
5948 data_len -= size;
5949
5950 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
5951 DMA_TO_DEVICE);
5952 if (dma_mapping_error(tx_ring->dev, dma))
5953 goto dma_error;
5954
5955 tx_buffer = &tx_ring->tx_buffer_info[i];
5956 dma_unmap_len_set(tx_buffer, len, size);
5957 dma_unmap_addr_set(tx_buffer, dma, dma);
5958
5959 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5960 tx_desc->read.olinfo_status = 0;
5961
5962 frag++;
5963 }
5964
5965 /* write last descriptor with RS and EOP bits */
5966 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
5967 tx_desc->read.cmd_type_len = cmd_type;
5968
5969 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
5970
5971 /* set the timestamp */
5972 first->time_stamp = jiffies;
5973
5974 /*
5975 * Force memory writes to complete before letting h/w know there
5976 * are new descriptors to fetch. (Only applicable for weak-ordered
5977 * memory model archs, such as IA-64).
5978 *
5979 * We also need this memory barrier to make certain all of the
5980 * status bits have been updated before next_to_watch is written.
5981 */
5982 wmb();
5983
5984 /* set next_to_watch value indicating a packet is present */
5985 first->next_to_watch = tx_desc;
5986
5987 i++;
5988 if (i == tx_ring->count)
5989 i = 0;
5990
5991 tx_ring->next_to_use = i;
5992
5993 /* notify HW of packet */
5994 writel(i, tx_ring->tail);
5995
5996 return;
5997 dma_error:
5998 dev_err(tx_ring->dev, "TX DMA map failed\n");
5999
6000 /* clear dma mappings for failed tx_buffer_info map */
6001 for (;;) {
6002 tx_buffer = &tx_ring->tx_buffer_info[i];
6003 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6004 if (tx_buffer == first)
6005 break;
6006 if (i == 0)
6007 i = tx_ring->count;
6008 i--;
6009 }
6010
6011 tx_ring->next_to_use = i;
6012 }
6013
6014 static void ixgbe_atr(struct ixgbe_ring *ring,
6015 struct ixgbe_tx_buffer *first)
6016 {
6017 struct ixgbe_q_vector *q_vector = ring->q_vector;
6018 union ixgbe_atr_hash_dword input = { .dword = 0 };
6019 union ixgbe_atr_hash_dword common = { .dword = 0 };
6020 union {
6021 unsigned char *network;
6022 struct iphdr *ipv4;
6023 struct ipv6hdr *ipv6;
6024 } hdr;
6025 struct tcphdr *th;
6026 __be16 vlan_id;
6027
6028 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6029 if (!q_vector)
6030 return;
6031
6032 /* do nothing if sampling is disabled */
6033 if (!ring->atr_sample_rate)
6034 return;
6035
6036 ring->atr_count++;
6037
6038 /* snag network header to get L4 type and address */
6039 hdr.network = skb_network_header(first->skb);
6040
6041 /* Currently only IPv4/IPv6 with TCP is supported */
6042 if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6043 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6044 (first->protocol != __constant_htons(ETH_P_IP) ||
6045 hdr.ipv4->protocol != IPPROTO_TCP))
6046 return;
6047
6048 th = tcp_hdr(first->skb);
6049
6050 /* skip this packet since it is invalid or the socket is closing */
6051 if (!th || th->fin)
6052 return;
6053
6054 /* sample on all syn packets or once every atr sample count */
6055 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6056 return;
6057
6058 /* reset sample count */
6059 ring->atr_count = 0;
6060
6061 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6062
6063 /*
6064 * src and dst are inverted, think how the receiver sees them
6065 *
6066 * The input is broken into two sections, a non-compressed section
6067 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6068 * is XORed together and stored in the compressed dword.
6069 */
6070 input.formatted.vlan_id = vlan_id;
6071
6072 /*
6073 * since src port and flex bytes occupy the same word XOR them together
6074 * and write the value to source port portion of compressed dword
6075 */
6076 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6077 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6078 else
6079 common.port.src ^= th->dest ^ first->protocol;
6080 common.port.dst ^= th->source;
6081
6082 if (first->protocol == __constant_htons(ETH_P_IP)) {
6083 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6084 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6085 } else {
6086 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6087 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6088 hdr.ipv6->saddr.s6_addr32[1] ^
6089 hdr.ipv6->saddr.s6_addr32[2] ^
6090 hdr.ipv6->saddr.s6_addr32[3] ^
6091 hdr.ipv6->daddr.s6_addr32[0] ^
6092 hdr.ipv6->daddr.s6_addr32[1] ^
6093 hdr.ipv6->daddr.s6_addr32[2] ^
6094 hdr.ipv6->daddr.s6_addr32[3];
6095 }
6096
6097 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6098 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6099 input, common, ring->queue_index);
6100 }
6101
6102 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6103 {
6104 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6105 /* Herbert's original patch had:
6106 * smp_mb__after_netif_stop_queue();
6107 * but since that doesn't exist yet, just open code it. */
6108 smp_mb();
6109
6110 /* We need to check again in a case another CPU has just
6111 * made room available. */
6112 if (likely(ixgbe_desc_unused(tx_ring) < size))
6113 return -EBUSY;
6114
6115 /* A reprieve! - use start_queue because it doesn't call schedule */
6116 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6117 ++tx_ring->tx_stats.restart_queue;
6118 return 0;
6119 }
6120
6121 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6122 {
6123 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6124 return 0;
6125 return __ixgbe_maybe_stop_tx(tx_ring, size);
6126 }
6127
6128 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6129 {
6130 struct ixgbe_adapter *adapter = netdev_priv(dev);
6131 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6132 smp_processor_id();
6133 #ifdef IXGBE_FCOE
6134 __be16 protocol = vlan_get_protocol(skb);
6135
6136 if (((protocol == htons(ETH_P_FCOE)) ||
6137 (protocol == htons(ETH_P_FIP))) &&
6138 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6139 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6140 txq += adapter->ring_feature[RING_F_FCOE].mask;
6141 return txq;
6142 }
6143 #endif
6144
6145 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6146 while (unlikely(txq >= dev->real_num_tx_queues))
6147 txq -= dev->real_num_tx_queues;
6148 return txq;
6149 }
6150
6151 return skb_tx_hash(dev, skb);
6152 }
6153
6154 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6155 struct ixgbe_adapter *adapter,
6156 struct ixgbe_ring *tx_ring)
6157 {
6158 struct ixgbe_tx_buffer *first;
6159 int tso;
6160 u32 tx_flags = 0;
6161 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6162 unsigned short f;
6163 #endif
6164 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6165 __be16 protocol = skb->protocol;
6166 u8 hdr_len = 0;
6167
6168 /*
6169 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6170 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6171 * + 2 desc gap to keep tail from touching head,
6172 * + 1 desc for context descriptor,
6173 * otherwise try next time
6174 */
6175 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6176 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6177 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6178 #else
6179 count += skb_shinfo(skb)->nr_frags;
6180 #endif
6181 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6182 tx_ring->tx_stats.tx_busy++;
6183 return NETDEV_TX_BUSY;
6184 }
6185
6186 /* record the location of the first descriptor for this packet */
6187 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6188 first->skb = skb;
6189 first->bytecount = skb->len;
6190 first->gso_segs = 1;
6191
6192 /* if we have a HW VLAN tag being added default to the HW one */
6193 if (vlan_tx_tag_present(skb)) {
6194 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6195 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6196 /* else if it is a SW VLAN check the next protocol and store the tag */
6197 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6198 struct vlan_hdr *vhdr, _vhdr;
6199 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6200 if (!vhdr)
6201 goto out_drop;
6202
6203 protocol = vhdr->h_vlan_encapsulated_proto;
6204 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6205 IXGBE_TX_FLAGS_VLAN_SHIFT;
6206 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6207 }
6208
6209 #ifdef CONFIG_PCI_IOV
6210 /*
6211 * Use the l2switch_enable flag - would be false if the DMA
6212 * Tx switch had been disabled.
6213 */
6214 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6215 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6216
6217 #endif
6218 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6219 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6220 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6221 (skb->priority != TC_PRIO_CONTROL))) {
6222 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6223 tx_flags |= (skb->priority & 0x7) <<
6224 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6225 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6226 struct vlan_ethhdr *vhdr;
6227 if (skb_header_cloned(skb) &&
6228 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6229 goto out_drop;
6230 vhdr = (struct vlan_ethhdr *)skb->data;
6231 vhdr->h_vlan_TCI = htons(tx_flags >>
6232 IXGBE_TX_FLAGS_VLAN_SHIFT);
6233 } else {
6234 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6235 }
6236 }
6237
6238 /* record initial flags and protocol */
6239 first->tx_flags = tx_flags;
6240 first->protocol = protocol;
6241
6242 #ifdef IXGBE_FCOE
6243 /* setup tx offload for FCoE */
6244 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6245 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6246 tso = ixgbe_fso(tx_ring, first, &hdr_len);
6247 if (tso < 0)
6248 goto out_drop;
6249
6250 goto xmit_fcoe;
6251 }
6252
6253 #endif /* IXGBE_FCOE */
6254 tso = ixgbe_tso(tx_ring, first, &hdr_len);
6255 if (tso < 0)
6256 goto out_drop;
6257 else if (!tso)
6258 ixgbe_tx_csum(tx_ring, first);
6259
6260 /* add the ATR filter if ATR is on */
6261 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6262 ixgbe_atr(tx_ring, first);
6263
6264 #ifdef IXGBE_FCOE
6265 xmit_fcoe:
6266 #endif /* IXGBE_FCOE */
6267 ixgbe_tx_map(tx_ring, first, hdr_len);
6268
6269 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6270
6271 return NETDEV_TX_OK;
6272
6273 out_drop:
6274 dev_kfree_skb_any(first->skb);
6275 first->skb = NULL;
6276
6277 return NETDEV_TX_OK;
6278 }
6279
6280 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6281 struct net_device *netdev)
6282 {
6283 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6284 struct ixgbe_ring *tx_ring;
6285
6286 if (skb->len <= 0) {
6287 dev_kfree_skb_any(skb);
6288 return NETDEV_TX_OK;
6289 }
6290
6291 /*
6292 * The minimum packet size for olinfo paylen is 17 so pad the skb
6293 * in order to meet this minimum size requirement.
6294 */
6295 if (skb->len < 17) {
6296 if (skb_padto(skb, 17))
6297 return NETDEV_TX_OK;
6298 skb->len = 17;
6299 }
6300
6301 tx_ring = adapter->tx_ring[skb->queue_mapping];
6302 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6303 }
6304
6305 /**
6306 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6307 * @netdev: network interface device structure
6308 * @p: pointer to an address structure
6309 *
6310 * Returns 0 on success, negative on failure
6311 **/
6312 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6313 {
6314 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6315 struct ixgbe_hw *hw = &adapter->hw;
6316 struct sockaddr *addr = p;
6317
6318 if (!is_valid_ether_addr(addr->sa_data))
6319 return -EADDRNOTAVAIL;
6320
6321 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6322 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6323
6324 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6325 IXGBE_RAH_AV);
6326
6327 return 0;
6328 }
6329
6330 static int
6331 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6332 {
6333 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6334 struct ixgbe_hw *hw = &adapter->hw;
6335 u16 value;
6336 int rc;
6337
6338 if (prtad != hw->phy.mdio.prtad)
6339 return -EINVAL;
6340 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6341 if (!rc)
6342 rc = value;
6343 return rc;
6344 }
6345
6346 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6347 u16 addr, u16 value)
6348 {
6349 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6350 struct ixgbe_hw *hw = &adapter->hw;
6351
6352 if (prtad != hw->phy.mdio.prtad)
6353 return -EINVAL;
6354 return hw->phy.ops.write_reg(hw, addr, devad, value);
6355 }
6356
6357 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6358 {
6359 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6360
6361 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6362 }
6363
6364 /**
6365 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6366 * netdev->dev_addrs
6367 * @netdev: network interface device structure
6368 *
6369 * Returns non-zero on failure
6370 **/
6371 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6372 {
6373 int err = 0;
6374 struct ixgbe_adapter *adapter = netdev_priv(dev);
6375 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6376
6377 if (is_valid_ether_addr(mac->san_addr)) {
6378 rtnl_lock();
6379 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6380 rtnl_unlock();
6381 }
6382 return err;
6383 }
6384
6385 /**
6386 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6387 * netdev->dev_addrs
6388 * @netdev: network interface device structure
6389 *
6390 * Returns non-zero on failure
6391 **/
6392 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6393 {
6394 int err = 0;
6395 struct ixgbe_adapter *adapter = netdev_priv(dev);
6396 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6397
6398 if (is_valid_ether_addr(mac->san_addr)) {
6399 rtnl_lock();
6400 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6401 rtnl_unlock();
6402 }
6403 return err;
6404 }
6405
6406 #ifdef CONFIG_NET_POLL_CONTROLLER
6407 /*
6408 * Polling 'interrupt' - used by things like netconsole to send skbs
6409 * without having to re-enable interrupts. It's not called while
6410 * the interrupt routine is executing.
6411 */
6412 static void ixgbe_netpoll(struct net_device *netdev)
6413 {
6414 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6415 int i;
6416
6417 /* if interface is down do nothing */
6418 if (test_bit(__IXGBE_DOWN, &adapter->state))
6419 return;
6420
6421 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6422 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6423 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6424 for (i = 0; i < num_q_vectors; i++) {
6425 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6426 ixgbe_msix_clean_rings(0, q_vector);
6427 }
6428 } else {
6429 ixgbe_intr(adapter->pdev->irq, netdev);
6430 }
6431 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6432 }
6433
6434 #endif
6435 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6436 struct rtnl_link_stats64 *stats)
6437 {
6438 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6439 int i;
6440
6441 rcu_read_lock();
6442 for (i = 0; i < adapter->num_rx_queues; i++) {
6443 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6444 u64 bytes, packets;
6445 unsigned int start;
6446
6447 if (ring) {
6448 do {
6449 start = u64_stats_fetch_begin_bh(&ring->syncp);
6450 packets = ring->stats.packets;
6451 bytes = ring->stats.bytes;
6452 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6453 stats->rx_packets += packets;
6454 stats->rx_bytes += bytes;
6455 }
6456 }
6457
6458 for (i = 0; i < adapter->num_tx_queues; i++) {
6459 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6460 u64 bytes, packets;
6461 unsigned int start;
6462
6463 if (ring) {
6464 do {
6465 start = u64_stats_fetch_begin_bh(&ring->syncp);
6466 packets = ring->stats.packets;
6467 bytes = ring->stats.bytes;
6468 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6469 stats->tx_packets += packets;
6470 stats->tx_bytes += bytes;
6471 }
6472 }
6473 rcu_read_unlock();
6474 /* following stats updated by ixgbe_watchdog_task() */
6475 stats->multicast = netdev->stats.multicast;
6476 stats->rx_errors = netdev->stats.rx_errors;
6477 stats->rx_length_errors = netdev->stats.rx_length_errors;
6478 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6479 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6480 return stats;
6481 }
6482
6483 #ifdef CONFIG_IXGBE_DCB
6484 /* ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6485 * #adapter: pointer to ixgbe_adapter
6486 * @tc: number of traffic classes currently enabled
6487 *
6488 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6489 * 802.1Q priority maps to a packet buffer that exists.
6490 */
6491 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6492 {
6493 struct ixgbe_hw *hw = &adapter->hw;
6494 u32 reg, rsave;
6495 int i;
6496
6497 /* 82598 have a static priority to TC mapping that can not
6498 * be changed so no validation is needed.
6499 */
6500 if (hw->mac.type == ixgbe_mac_82598EB)
6501 return;
6502
6503 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6504 rsave = reg;
6505
6506 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6507 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6508
6509 /* If up2tc is out of bounds default to zero */
6510 if (up2tc > tc)
6511 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6512 }
6513
6514 if (reg != rsave)
6515 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6516
6517 return;
6518 }
6519
6520 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
6521 * classes.
6522 *
6523 * @netdev: net device to configure
6524 * @tc: number of traffic classes to enable
6525 */
6526 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6527 {
6528 struct ixgbe_adapter *adapter = netdev_priv(dev);
6529 struct ixgbe_hw *hw = &adapter->hw;
6530
6531 /* Multiple traffic classes requires multiple queues */
6532 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6533 e_err(drv, "Enable failed, needs MSI-X\n");
6534 return -EINVAL;
6535 }
6536
6537 /* Hardware supports up to 8 traffic classes */
6538 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
6539 (hw->mac.type == ixgbe_mac_82598EB &&
6540 tc < MAX_TRAFFIC_CLASS))
6541 return -EINVAL;
6542
6543 /* Hardware has to reinitialize queues and interrupts to
6544 * match packet buffer alignment. Unfortunately, the
6545 * hardware is not flexible enough to do this dynamically.
6546 */
6547 if (netif_running(dev))
6548 ixgbe_close(dev);
6549 ixgbe_clear_interrupt_scheme(adapter);
6550
6551 if (tc) {
6552 netdev_set_num_tc(dev, tc);
6553 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6554 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6555
6556 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
6557 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
6558 adapter->hw.fc.requested_mode = ixgbe_fc_none;
6559 }
6560 } else {
6561 netdev_reset_tc(dev);
6562 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6563 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6564
6565 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6566 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6567
6568 adapter->temp_dcb_cfg.pfc_mode_enable = false;
6569 adapter->dcb_cfg.pfc_mode_enable = false;
6570 }
6571
6572 ixgbe_init_interrupt_scheme(adapter);
6573 ixgbe_validate_rtr(adapter, tc);
6574 if (netif_running(dev))
6575 ixgbe_open(dev);
6576
6577 return 0;
6578 }
6579
6580 #endif /* CONFIG_IXGBE_DCB */
6581 void ixgbe_do_reset(struct net_device *netdev)
6582 {
6583 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6584
6585 if (netif_running(netdev))
6586 ixgbe_reinit_locked(adapter);
6587 else
6588 ixgbe_reset(adapter);
6589 }
6590
6591 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
6592 netdev_features_t features)
6593 {
6594 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6595
6596 #ifdef CONFIG_DCB
6597 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6598 features &= ~NETIF_F_HW_VLAN_RX;
6599 #endif
6600
6601 /* return error if RXHASH is being enabled when RSS is not supported */
6602 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
6603 features &= ~NETIF_F_RXHASH;
6604
6605 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6606 if (!(features & NETIF_F_RXCSUM))
6607 features &= ~NETIF_F_LRO;
6608
6609 /* Turn off LRO if not RSC capable */
6610 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
6611 features &= ~NETIF_F_LRO;
6612
6613
6614 return features;
6615 }
6616
6617 static int ixgbe_set_features(struct net_device *netdev,
6618 netdev_features_t features)
6619 {
6620 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6621 netdev_features_t changed = netdev->features ^ features;
6622 bool need_reset = false;
6623
6624 /* Make sure RSC matches LRO, reset if change */
6625 if (!(features & NETIF_F_LRO)) {
6626 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6627 need_reset = true;
6628 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
6629 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
6630 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6631 if (adapter->rx_itr_setting == 1 ||
6632 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
6633 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
6634 need_reset = true;
6635 } else if ((changed ^ features) & NETIF_F_LRO) {
6636 e_info(probe, "rx-usecs set too low, "
6637 "disabling RSC\n");
6638 }
6639 }
6640
6641 /*
6642 * Check if Flow Director n-tuple support was enabled or disabled. If
6643 * the state changed, we need to reset.
6644 */
6645 if (!(features & NETIF_F_NTUPLE)) {
6646 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
6647 /* turn off Flow Director, set ATR and reset */
6648 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
6649 !(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
6650 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6651 need_reset = true;
6652 }
6653 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6654 } else if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
6655 /* turn off ATR, enable perfect filters and reset */
6656 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6657 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6658 need_reset = true;
6659 }
6660
6661 if (changed & NETIF_F_RXALL)
6662 need_reset = true;
6663
6664 netdev->features = features;
6665 if (need_reset)
6666 ixgbe_do_reset(netdev);
6667
6668 return 0;
6669 }
6670
6671 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm,
6672 struct net_device *dev,
6673 unsigned char *addr,
6674 u16 flags)
6675 {
6676 struct ixgbe_adapter *adapter = netdev_priv(dev);
6677 int err = -EOPNOTSUPP;
6678
6679 if (ndm->ndm_state & NUD_PERMANENT) {
6680 pr_info("%s: FDB only supports static addresses\n",
6681 ixgbe_driver_name);
6682 return -EINVAL;
6683 }
6684
6685 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6686 if (is_unicast_ether_addr(addr))
6687 err = dev_uc_add_excl(dev, addr);
6688 else if (is_multicast_ether_addr(addr))
6689 err = dev_mc_add_excl(dev, addr);
6690 else
6691 err = -EINVAL;
6692 }
6693
6694 /* Only return duplicate errors if NLM_F_EXCL is set */
6695 if (err == -EEXIST && !(flags & NLM_F_EXCL))
6696 err = 0;
6697
6698 return err;
6699 }
6700
6701 static int ixgbe_ndo_fdb_del(struct ndmsg *ndm,
6702 struct net_device *dev,
6703 unsigned char *addr)
6704 {
6705 struct ixgbe_adapter *adapter = netdev_priv(dev);
6706 int err = -EOPNOTSUPP;
6707
6708 if (ndm->ndm_state & NUD_PERMANENT) {
6709 pr_info("%s: FDB only supports static addresses\n",
6710 ixgbe_driver_name);
6711 return -EINVAL;
6712 }
6713
6714 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6715 if (is_unicast_ether_addr(addr))
6716 err = dev_uc_del(dev, addr);
6717 else if (is_multicast_ether_addr(addr))
6718 err = dev_mc_del(dev, addr);
6719 else
6720 err = -EINVAL;
6721 }
6722
6723 return err;
6724 }
6725
6726 static int ixgbe_ndo_fdb_dump(struct sk_buff *skb,
6727 struct netlink_callback *cb,
6728 struct net_device *dev,
6729 int idx)
6730 {
6731 struct ixgbe_adapter *adapter = netdev_priv(dev);
6732
6733 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6734 idx = ndo_dflt_fdb_dump(skb, cb, dev, idx);
6735
6736 return idx;
6737 }
6738
6739 static const struct net_device_ops ixgbe_netdev_ops = {
6740 .ndo_open = ixgbe_open,
6741 .ndo_stop = ixgbe_close,
6742 .ndo_start_xmit = ixgbe_xmit_frame,
6743 .ndo_select_queue = ixgbe_select_queue,
6744 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6745 .ndo_validate_addr = eth_validate_addr,
6746 .ndo_set_mac_address = ixgbe_set_mac,
6747 .ndo_change_mtu = ixgbe_change_mtu,
6748 .ndo_tx_timeout = ixgbe_tx_timeout,
6749 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6750 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6751 .ndo_do_ioctl = ixgbe_ioctl,
6752 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6753 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6754 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6755 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
6756 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6757 .ndo_get_stats64 = ixgbe_get_stats64,
6758 #ifdef CONFIG_IXGBE_DCB
6759 .ndo_setup_tc = ixgbe_setup_tc,
6760 #endif
6761 #ifdef CONFIG_NET_POLL_CONTROLLER
6762 .ndo_poll_controller = ixgbe_netpoll,
6763 #endif
6764 #ifdef IXGBE_FCOE
6765 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6766 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
6767 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6768 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6769 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6770 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6771 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
6772 #endif /* IXGBE_FCOE */
6773 .ndo_set_features = ixgbe_set_features,
6774 .ndo_fix_features = ixgbe_fix_features,
6775 .ndo_fdb_add = ixgbe_ndo_fdb_add,
6776 .ndo_fdb_del = ixgbe_ndo_fdb_del,
6777 .ndo_fdb_dump = ixgbe_ndo_fdb_dump,
6778 };
6779
6780 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6781 const struct ixgbe_info *ii)
6782 {
6783 #ifdef CONFIG_PCI_IOV
6784 struct ixgbe_hw *hw = &adapter->hw;
6785
6786 if (hw->mac.type == ixgbe_mac_82598EB)
6787 return;
6788
6789 /* The 82599 supports up to 64 VFs per physical function
6790 * but this implementation limits allocation to 63 so that
6791 * basic networking resources are still available to the
6792 * physical function. If the user requests greater thn
6793 * 63 VFs then it is an error - reset to default of zero.
6794 */
6795 adapter->num_vfs = (max_vfs > 63) ? 0 : max_vfs;
6796 ixgbe_enable_sriov(adapter, ii);
6797 #endif /* CONFIG_PCI_IOV */
6798 }
6799
6800 /**
6801 * ixgbe_wol_supported - Check whether device supports WoL
6802 * @hw: hw specific details
6803 * @device_id: the device ID
6804 * @subdev_id: the subsystem device ID
6805 *
6806 * This function is used by probe and ethtool to determine
6807 * which devices have WoL support
6808 *
6809 **/
6810 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
6811 u16 subdevice_id)
6812 {
6813 struct ixgbe_hw *hw = &adapter->hw;
6814 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
6815 int is_wol_supported = 0;
6816
6817 switch (device_id) {
6818 case IXGBE_DEV_ID_82599_SFP:
6819 /* Only these subdevices could supports WOL */
6820 switch (subdevice_id) {
6821 case IXGBE_SUBDEV_ID_82599_560FLR:
6822 /* only support first port */
6823 if (hw->bus.func != 0)
6824 break;
6825 case IXGBE_SUBDEV_ID_82599_SFP:
6826 is_wol_supported = 1;
6827 break;
6828 }
6829 break;
6830 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
6831 /* All except this subdevice support WOL */
6832 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
6833 is_wol_supported = 1;
6834 break;
6835 case IXGBE_DEV_ID_82599_KX4:
6836 is_wol_supported = 1;
6837 break;
6838 case IXGBE_DEV_ID_X540T:
6839 /* check eeprom to see if enabled wol */
6840 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
6841 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
6842 (hw->bus.func == 0))) {
6843 is_wol_supported = 1;
6844 }
6845 break;
6846 }
6847
6848 return is_wol_supported;
6849 }
6850
6851 /**
6852 * ixgbe_probe - Device Initialization Routine
6853 * @pdev: PCI device information struct
6854 * @ent: entry in ixgbe_pci_tbl
6855 *
6856 * Returns 0 on success, negative on failure
6857 *
6858 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6859 * The OS initialization, configuring of the adapter private structure,
6860 * and a hardware reset occur.
6861 **/
6862 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6863 const struct pci_device_id *ent)
6864 {
6865 struct net_device *netdev;
6866 struct ixgbe_adapter *adapter = NULL;
6867 struct ixgbe_hw *hw;
6868 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6869 static int cards_found;
6870 int i, err, pci_using_dac;
6871 u8 part_str[IXGBE_PBANUM_LENGTH];
6872 unsigned int indices = num_possible_cpus();
6873 #ifdef IXGBE_FCOE
6874 u16 device_caps;
6875 #endif
6876 u32 eec;
6877
6878 /* Catch broken hardware that put the wrong VF device ID in
6879 * the PCIe SR-IOV capability.
6880 */
6881 if (pdev->is_virtfn) {
6882 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6883 pci_name(pdev), pdev->vendor, pdev->device);
6884 return -EINVAL;
6885 }
6886
6887 err = pci_enable_device_mem(pdev);
6888 if (err)
6889 return err;
6890
6891 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6892 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6893 pci_using_dac = 1;
6894 } else {
6895 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6896 if (err) {
6897 err = dma_set_coherent_mask(&pdev->dev,
6898 DMA_BIT_MASK(32));
6899 if (err) {
6900 dev_err(&pdev->dev,
6901 "No usable DMA configuration, aborting\n");
6902 goto err_dma;
6903 }
6904 }
6905 pci_using_dac = 0;
6906 }
6907
6908 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6909 IORESOURCE_MEM), ixgbe_driver_name);
6910 if (err) {
6911 dev_err(&pdev->dev,
6912 "pci_request_selected_regions failed 0x%x\n", err);
6913 goto err_pci_reg;
6914 }
6915
6916 pci_enable_pcie_error_reporting(pdev);
6917
6918 pci_set_master(pdev);
6919 pci_save_state(pdev);
6920
6921 #ifdef CONFIG_IXGBE_DCB
6922 indices *= MAX_TRAFFIC_CLASS;
6923 #endif
6924
6925 if (ii->mac == ixgbe_mac_82598EB)
6926 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6927 else
6928 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6929
6930 #ifdef IXGBE_FCOE
6931 indices += min_t(unsigned int, num_possible_cpus(),
6932 IXGBE_MAX_FCOE_INDICES);
6933 #endif
6934 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6935 if (!netdev) {
6936 err = -ENOMEM;
6937 goto err_alloc_etherdev;
6938 }
6939
6940 SET_NETDEV_DEV(netdev, &pdev->dev);
6941
6942 adapter = netdev_priv(netdev);
6943 pci_set_drvdata(pdev, adapter);
6944
6945 adapter->netdev = netdev;
6946 adapter->pdev = pdev;
6947 hw = &adapter->hw;
6948 hw->back = adapter;
6949 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6950
6951 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6952 pci_resource_len(pdev, 0));
6953 if (!hw->hw_addr) {
6954 err = -EIO;
6955 goto err_ioremap;
6956 }
6957
6958 for (i = 1; i <= 5; i++) {
6959 if (pci_resource_len(pdev, i) == 0)
6960 continue;
6961 }
6962
6963 netdev->netdev_ops = &ixgbe_netdev_ops;
6964 ixgbe_set_ethtool_ops(netdev);
6965 netdev->watchdog_timeo = 5 * HZ;
6966 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
6967
6968 adapter->bd_number = cards_found;
6969
6970 /* Setup hw api */
6971 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6972 hw->mac.type = ii->mac;
6973
6974 /* EEPROM */
6975 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6976 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6977 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6978 if (!(eec & (1 << 8)))
6979 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6980
6981 /* PHY */
6982 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6983 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6984 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6985 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6986 hw->phy.mdio.mmds = 0;
6987 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6988 hw->phy.mdio.dev = netdev;
6989 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6990 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6991
6992 ii->get_invariants(hw);
6993
6994 /* setup the private structure */
6995 err = ixgbe_sw_init(adapter);
6996 if (err)
6997 goto err_sw_init;
6998
6999 /* Make it possible the adapter to be woken up via WOL */
7000 switch (adapter->hw.mac.type) {
7001 case ixgbe_mac_82599EB:
7002 case ixgbe_mac_X540:
7003 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7004 break;
7005 default:
7006 break;
7007 }
7008
7009 /*
7010 * If there is a fan on this device and it has failed log the
7011 * failure.
7012 */
7013 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7014 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7015 if (esdp & IXGBE_ESDP_SDP1)
7016 e_crit(probe, "Fan has stopped, replace the adapter\n");
7017 }
7018
7019 if (allow_unsupported_sfp)
7020 hw->allow_unsupported_sfp = allow_unsupported_sfp;
7021
7022 /* reset_hw fills in the perm_addr as well */
7023 hw->phy.reset_if_overtemp = true;
7024 err = hw->mac.ops.reset_hw(hw);
7025 hw->phy.reset_if_overtemp = false;
7026 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7027 hw->mac.type == ixgbe_mac_82598EB) {
7028 err = 0;
7029 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7030 e_dev_err("failed to load because an unsupported SFP+ "
7031 "module type was detected.\n");
7032 e_dev_err("Reload the driver after installing a supported "
7033 "module.\n");
7034 goto err_sw_init;
7035 } else if (err) {
7036 e_dev_err("HW Init failed: %d\n", err);
7037 goto err_sw_init;
7038 }
7039
7040 ixgbe_probe_vf(adapter, ii);
7041
7042 netdev->features = NETIF_F_SG |
7043 NETIF_F_IP_CSUM |
7044 NETIF_F_IPV6_CSUM |
7045 NETIF_F_HW_VLAN_TX |
7046 NETIF_F_HW_VLAN_RX |
7047 NETIF_F_HW_VLAN_FILTER |
7048 NETIF_F_TSO |
7049 NETIF_F_TSO6 |
7050 NETIF_F_RXHASH |
7051 NETIF_F_RXCSUM;
7052
7053 netdev->hw_features = netdev->features;
7054
7055 switch (adapter->hw.mac.type) {
7056 case ixgbe_mac_82599EB:
7057 case ixgbe_mac_X540:
7058 netdev->features |= NETIF_F_SCTP_CSUM;
7059 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7060 NETIF_F_NTUPLE;
7061 break;
7062 default:
7063 break;
7064 }
7065
7066 netdev->hw_features |= NETIF_F_RXALL;
7067
7068 netdev->vlan_features |= NETIF_F_TSO;
7069 netdev->vlan_features |= NETIF_F_TSO6;
7070 netdev->vlan_features |= NETIF_F_IP_CSUM;
7071 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7072 netdev->vlan_features |= NETIF_F_SG;
7073
7074 netdev->priv_flags |= IFF_UNICAST_FLT;
7075 netdev->priv_flags |= IFF_SUPP_NOFCS;
7076
7077 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7078 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7079 IXGBE_FLAG_DCB_ENABLED);
7080
7081 #ifdef CONFIG_IXGBE_DCB
7082 netdev->dcbnl_ops = &dcbnl_ops;
7083 #endif
7084
7085 #ifdef IXGBE_FCOE
7086 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7087 if (hw->mac.ops.get_device_caps) {
7088 hw->mac.ops.get_device_caps(hw, &device_caps);
7089 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7090 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7091 }
7092 }
7093 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7094 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7095 netdev->vlan_features |= NETIF_F_FSO;
7096 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7097 }
7098 #endif /* IXGBE_FCOE */
7099 if (pci_using_dac) {
7100 netdev->features |= NETIF_F_HIGHDMA;
7101 netdev->vlan_features |= NETIF_F_HIGHDMA;
7102 }
7103
7104 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7105 netdev->hw_features |= NETIF_F_LRO;
7106 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7107 netdev->features |= NETIF_F_LRO;
7108
7109 /* make sure the EEPROM is good */
7110 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7111 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7112 err = -EIO;
7113 goto err_sw_init;
7114 }
7115
7116 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7117 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7118
7119 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7120 e_dev_err("invalid MAC address\n");
7121 err = -EIO;
7122 goto err_sw_init;
7123 }
7124
7125 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7126 (unsigned long) adapter);
7127
7128 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7129 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7130
7131 err = ixgbe_init_interrupt_scheme(adapter);
7132 if (err)
7133 goto err_sw_init;
7134
7135 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7136 netdev->hw_features &= ~NETIF_F_RXHASH;
7137 netdev->features &= ~NETIF_F_RXHASH;
7138 }
7139
7140 /* WOL not supported for all devices */
7141 adapter->wol = 0;
7142 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7143 if (ixgbe_wol_supported(adapter, pdev->device, pdev->subsystem_device))
7144 adapter->wol = IXGBE_WUFC_MAG;
7145
7146 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7147
7148 /* save off EEPROM version number */
7149 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7150 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7151
7152 /* pick up the PCI bus settings for reporting later */
7153 hw->mac.ops.get_bus_info(hw);
7154
7155 /* print bus type/speed/width info */
7156 e_dev_info("(PCI Express:%s:%s) %pM\n",
7157 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7158 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7159 "Unknown"),
7160 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7161 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7162 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7163 "Unknown"),
7164 netdev->dev_addr);
7165
7166 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7167 if (err)
7168 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7169 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7170 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7171 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7172 part_str);
7173 else
7174 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7175 hw->mac.type, hw->phy.type, part_str);
7176
7177 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7178 e_dev_warn("PCI-Express bandwidth available for this card is "
7179 "not sufficient for optimal performance.\n");
7180 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7181 "is required.\n");
7182 }
7183
7184 /* reset the hardware with the new settings */
7185 err = hw->mac.ops.start_hw(hw);
7186 if (err == IXGBE_ERR_EEPROM_VERSION) {
7187 /* We are running on a pre-production device, log a warning */
7188 e_dev_warn("This device is a pre-production adapter/LOM. "
7189 "Please be aware there may be issues associated "
7190 "with your hardware. If you are experiencing "
7191 "problems please contact your Intel or hardware "
7192 "representative who provided you with this "
7193 "hardware.\n");
7194 }
7195 strcpy(netdev->name, "eth%d");
7196 err = register_netdev(netdev);
7197 if (err)
7198 goto err_register;
7199
7200 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7201 if (hw->mac.ops.disable_tx_laser &&
7202 ((hw->phy.multispeed_fiber) ||
7203 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7204 (hw->mac.type == ixgbe_mac_82599EB))))
7205 hw->mac.ops.disable_tx_laser(hw);
7206
7207 /* carrier off reporting is important to ethtool even BEFORE open */
7208 netif_carrier_off(netdev);
7209
7210 #ifdef CONFIG_IXGBE_DCA
7211 if (dca_add_requester(&pdev->dev) == 0) {
7212 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7213 ixgbe_setup_dca(adapter);
7214 }
7215 #endif
7216 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7217 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7218 for (i = 0; i < adapter->num_vfs; i++)
7219 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7220 }
7221
7222 /* firmware requires driver version to be 0xFFFFFFFF
7223 * since os does not support feature
7224 */
7225 if (hw->mac.ops.set_fw_drv_ver)
7226 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7227 0xFF);
7228
7229 /* add san mac addr to netdev */
7230 ixgbe_add_sanmac_netdev(netdev);
7231
7232 e_dev_info("%s\n", ixgbe_default_device_descr);
7233 cards_found++;
7234
7235 if (ixgbe_sysfs_init(adapter))
7236 e_err(probe, "failed to allocate sysfs resources\n");
7237
7238 return 0;
7239
7240 err_register:
7241 ixgbe_release_hw_control(adapter);
7242 ixgbe_clear_interrupt_scheme(adapter);
7243 err_sw_init:
7244 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7245 ixgbe_disable_sriov(adapter);
7246 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7247 iounmap(hw->hw_addr);
7248 err_ioremap:
7249 free_netdev(netdev);
7250 err_alloc_etherdev:
7251 pci_release_selected_regions(pdev,
7252 pci_select_bars(pdev, IORESOURCE_MEM));
7253 err_pci_reg:
7254 err_dma:
7255 pci_disable_device(pdev);
7256 return err;
7257 }
7258
7259 /**
7260 * ixgbe_remove - Device Removal Routine
7261 * @pdev: PCI device information struct
7262 *
7263 * ixgbe_remove is called by the PCI subsystem to alert the driver
7264 * that it should release a PCI device. The could be caused by a
7265 * Hot-Plug event, or because the driver is going to be removed from
7266 * memory.
7267 **/
7268 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7269 {
7270 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7271 struct net_device *netdev = adapter->netdev;
7272
7273 set_bit(__IXGBE_DOWN, &adapter->state);
7274 cancel_work_sync(&adapter->service_task);
7275
7276 #ifdef CONFIG_IXGBE_DCA
7277 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7278 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7279 dca_remove_requester(&pdev->dev);
7280 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7281 }
7282
7283 #endif
7284 ixgbe_sysfs_exit(adapter);
7285
7286 #ifdef IXGBE_FCOE
7287 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7288 ixgbe_cleanup_fcoe(adapter);
7289
7290 #endif /* IXGBE_FCOE */
7291
7292 /* remove the added san mac */
7293 ixgbe_del_sanmac_netdev(netdev);
7294
7295 if (netdev->reg_state == NETREG_REGISTERED)
7296 unregister_netdev(netdev);
7297
7298 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7299 if (!(ixgbe_check_vf_assignment(adapter)))
7300 ixgbe_disable_sriov(adapter);
7301 else
7302 e_dev_warn("Unloading driver while VFs are assigned "
7303 "- VFs will not be deallocated\n");
7304 }
7305
7306 ixgbe_clear_interrupt_scheme(adapter);
7307
7308 ixgbe_release_hw_control(adapter);
7309
7310 #ifdef CONFIG_DCB
7311 kfree(adapter->ixgbe_ieee_pfc);
7312 kfree(adapter->ixgbe_ieee_ets);
7313
7314 #endif
7315 iounmap(adapter->hw.hw_addr);
7316 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7317 IORESOURCE_MEM));
7318
7319 e_dev_info("complete\n");
7320
7321 free_netdev(netdev);
7322
7323 pci_disable_pcie_error_reporting(pdev);
7324
7325 pci_disable_device(pdev);
7326 }
7327
7328 /**
7329 * ixgbe_io_error_detected - called when PCI error is detected
7330 * @pdev: Pointer to PCI device
7331 * @state: The current pci connection state
7332 *
7333 * This function is called after a PCI bus error affecting
7334 * this device has been detected.
7335 */
7336 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7337 pci_channel_state_t state)
7338 {
7339 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7340 struct net_device *netdev = adapter->netdev;
7341
7342 #ifdef CONFIG_PCI_IOV
7343 struct pci_dev *bdev, *vfdev;
7344 u32 dw0, dw1, dw2, dw3;
7345 int vf, pos;
7346 u16 req_id, pf_func;
7347
7348 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7349 adapter->num_vfs == 0)
7350 goto skip_bad_vf_detection;
7351
7352 bdev = pdev->bus->self;
7353 while (bdev && (bdev->pcie_type != PCI_EXP_TYPE_ROOT_PORT))
7354 bdev = bdev->bus->self;
7355
7356 if (!bdev)
7357 goto skip_bad_vf_detection;
7358
7359 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7360 if (!pos)
7361 goto skip_bad_vf_detection;
7362
7363 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7364 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7365 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7366 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7367
7368 req_id = dw1 >> 16;
7369 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7370 if (!(req_id & 0x0080))
7371 goto skip_bad_vf_detection;
7372
7373 pf_func = req_id & 0x01;
7374 if ((pf_func & 1) == (pdev->devfn & 1)) {
7375 unsigned int device_id;
7376
7377 vf = (req_id & 0x7F) >> 1;
7378 e_dev_err("VF %d has caused a PCIe error\n", vf);
7379 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7380 "%8.8x\tdw3: %8.8x\n",
7381 dw0, dw1, dw2, dw3);
7382 switch (adapter->hw.mac.type) {
7383 case ixgbe_mac_82599EB:
7384 device_id = IXGBE_82599_VF_DEVICE_ID;
7385 break;
7386 case ixgbe_mac_X540:
7387 device_id = IXGBE_X540_VF_DEVICE_ID;
7388 break;
7389 default:
7390 device_id = 0;
7391 break;
7392 }
7393
7394 /* Find the pci device of the offending VF */
7395 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID, device_id, NULL);
7396 while (vfdev) {
7397 if (vfdev->devfn == (req_id & 0xFF))
7398 break;
7399 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID,
7400 device_id, vfdev);
7401 }
7402 /*
7403 * There's a slim chance the VF could have been hot plugged,
7404 * so if it is no longer present we don't need to issue the
7405 * VFLR. Just clean up the AER in that case.
7406 */
7407 if (vfdev) {
7408 e_dev_err("Issuing VFLR to VF %d\n", vf);
7409 pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7410 }
7411
7412 pci_cleanup_aer_uncorrect_error_status(pdev);
7413 }
7414
7415 /*
7416 * Even though the error may have occurred on the other port
7417 * we still need to increment the vf error reference count for
7418 * both ports because the I/O resume function will be called
7419 * for both of them.
7420 */
7421 adapter->vferr_refcount++;
7422
7423 return PCI_ERS_RESULT_RECOVERED;
7424
7425 skip_bad_vf_detection:
7426 #endif /* CONFIG_PCI_IOV */
7427 netif_device_detach(netdev);
7428
7429 if (state == pci_channel_io_perm_failure)
7430 return PCI_ERS_RESULT_DISCONNECT;
7431
7432 if (netif_running(netdev))
7433 ixgbe_down(adapter);
7434 pci_disable_device(pdev);
7435
7436 /* Request a slot reset. */
7437 return PCI_ERS_RESULT_NEED_RESET;
7438 }
7439
7440 /**
7441 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7442 * @pdev: Pointer to PCI device
7443 *
7444 * Restart the card from scratch, as if from a cold-boot.
7445 */
7446 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7447 {
7448 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7449 pci_ers_result_t result;
7450 int err;
7451
7452 if (pci_enable_device_mem(pdev)) {
7453 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7454 result = PCI_ERS_RESULT_DISCONNECT;
7455 } else {
7456 pci_set_master(pdev);
7457 pci_restore_state(pdev);
7458 pci_save_state(pdev);
7459
7460 pci_wake_from_d3(pdev, false);
7461
7462 ixgbe_reset(adapter);
7463 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7464 result = PCI_ERS_RESULT_RECOVERED;
7465 }
7466
7467 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7468 if (err) {
7469 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7470 "failed 0x%0x\n", err);
7471 /* non-fatal, continue */
7472 }
7473
7474 return result;
7475 }
7476
7477 /**
7478 * ixgbe_io_resume - called when traffic can start flowing again.
7479 * @pdev: Pointer to PCI device
7480 *
7481 * This callback is called when the error recovery driver tells us that
7482 * its OK to resume normal operation.
7483 */
7484 static void ixgbe_io_resume(struct pci_dev *pdev)
7485 {
7486 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7487 struct net_device *netdev = adapter->netdev;
7488
7489 #ifdef CONFIG_PCI_IOV
7490 if (adapter->vferr_refcount) {
7491 e_info(drv, "Resuming after VF err\n");
7492 adapter->vferr_refcount--;
7493 return;
7494 }
7495
7496 #endif
7497 if (netif_running(netdev))
7498 ixgbe_up(adapter);
7499
7500 netif_device_attach(netdev);
7501 }
7502
7503 static struct pci_error_handlers ixgbe_err_handler = {
7504 .error_detected = ixgbe_io_error_detected,
7505 .slot_reset = ixgbe_io_slot_reset,
7506 .resume = ixgbe_io_resume,
7507 };
7508
7509 static struct pci_driver ixgbe_driver = {
7510 .name = ixgbe_driver_name,
7511 .id_table = ixgbe_pci_tbl,
7512 .probe = ixgbe_probe,
7513 .remove = __devexit_p(ixgbe_remove),
7514 #ifdef CONFIG_PM
7515 .suspend = ixgbe_suspend,
7516 .resume = ixgbe_resume,
7517 #endif
7518 .shutdown = ixgbe_shutdown,
7519 .err_handler = &ixgbe_err_handler
7520 };
7521
7522 /**
7523 * ixgbe_init_module - Driver Registration Routine
7524 *
7525 * ixgbe_init_module is the first routine called when the driver is
7526 * loaded. All it does is register with the PCI subsystem.
7527 **/
7528 static int __init ixgbe_init_module(void)
7529 {
7530 int ret;
7531 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7532 pr_info("%s\n", ixgbe_copyright);
7533
7534 #ifdef CONFIG_IXGBE_DCA
7535 dca_register_notify(&dca_notifier);
7536 #endif
7537
7538 ret = pci_register_driver(&ixgbe_driver);
7539 return ret;
7540 }
7541
7542 module_init(ixgbe_init_module);
7543
7544 /**
7545 * ixgbe_exit_module - Driver Exit Cleanup Routine
7546 *
7547 * ixgbe_exit_module is called just before the driver is removed
7548 * from memory.
7549 **/
7550 static void __exit ixgbe_exit_module(void)
7551 {
7552 #ifdef CONFIG_IXGBE_DCA
7553 dca_unregister_notify(&dca_notifier);
7554 #endif
7555 pci_unregister_driver(&ixgbe_driver);
7556 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7557 }
7558
7559 #ifdef CONFIG_IXGBE_DCA
7560 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7561 void *p)
7562 {
7563 int ret_val;
7564
7565 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7566 __ixgbe_notify_dca);
7567
7568 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7569 }
7570
7571 #endif /* CONFIG_IXGBE_DCA */
7572
7573 module_exit(ixgbe_exit_module);
7574
7575 /* ixgbe_main.c */
Linux kernel - Deliverables
This page took 0.183995 seconds and 6 git commands to generate.