Merge remote-tracking branch 'moduleh/module.h-split'
[linux-2.6/next.git] / drivers / net / ethernet / intel / ixgbe / ixgbe_main.c
blobd20e8040d85521e0dab05723d4614757d276cbbb
1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2011 Intel Corporation.
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.
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.
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.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
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
26 *******************************************************************************/
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>
50 #include "ixgbe.h"
51 #include "ixgbe_common.h"
52 #include "ixgbe_dcb_82599.h"
53 #include "ixgbe_sriov.h"
55 char ixgbe_driver_name[] = "ixgbe";
56 static const char ixgbe_driver_string[] =
57 "Intel(R) 10 Gigabit PCI Express Network Driver";
58 #define MAJ 3
59 #define MIN 4
60 #define BUILD 8
61 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
62 __stringify(BUILD) "-k"
63 const char ixgbe_driver_version[] = DRV_VERSION;
64 static const char ixgbe_copyright[] =
65 "Copyright (c) 1999-2011 Intel Corporation.";
67 static const struct ixgbe_info *ixgbe_info_tbl[] = {
68 [board_82598] = &ixgbe_82598_info,
69 [board_82599] = &ixgbe_82599_info,
70 [board_X540] = &ixgbe_X540_info,
73 /* ixgbe_pci_tbl - PCI Device ID Table
75 * Wildcard entries (PCI_ANY_ID) should come last
76 * Last entry must be all 0s
78 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
79 * Class, Class Mask, private data (not used) }
81 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
82 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
83 board_82598 },
84 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
85 board_82598 },
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
87 board_82598 },
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
89 board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
91 board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
93 board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
95 board_82598 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
97 board_82598 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
99 board_82598 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
101 board_82598 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
103 board_82598 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
105 board_82598 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
107 board_82599 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
109 board_82599 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
111 board_82599 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
113 board_82599 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
115 board_82599 },
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
117 board_82599 },
118 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
119 board_82599 },
120 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE),
121 board_82599 },
122 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE),
123 board_82599 },
124 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM),
125 board_82599 },
126 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
127 board_82599 },
128 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T),
129 board_X540 },
130 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2),
131 board_82599 },
132 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS),
133 board_82599 },
135 /* required last entry */
136 {0, }
138 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
140 #ifdef CONFIG_IXGBE_DCA
141 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
142 void *p);
143 static struct notifier_block dca_notifier = {
144 .notifier_call = ixgbe_notify_dca,
145 .next = NULL,
146 .priority = 0
148 #endif
150 #ifdef CONFIG_PCI_IOV
151 static unsigned int max_vfs;
152 module_param(max_vfs, uint, 0);
153 MODULE_PARM_DESC(max_vfs,
154 "Maximum number of virtual functions to allocate per physical function");
155 #endif /* CONFIG_PCI_IOV */
157 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
158 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
159 MODULE_LICENSE("GPL");
160 MODULE_VERSION(DRV_VERSION);
162 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
164 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
166 struct ixgbe_hw *hw = &adapter->hw;
167 u32 gcr;
168 u32 gpie;
169 u32 vmdctl;
171 #ifdef CONFIG_PCI_IOV
172 /* disable iov and allow time for transactions to clear */
173 pci_disable_sriov(adapter->pdev);
174 #endif
176 /* turn off device IOV mode */
177 gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
178 gcr &= ~(IXGBE_GCR_EXT_SRIOV);
179 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
180 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
181 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
182 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
184 /* set default pool back to 0 */
185 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
186 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
187 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
188 IXGBE_WRITE_FLUSH(hw);
190 /* take a breather then clean up driver data */
191 msleep(100);
193 kfree(adapter->vfinfo);
194 adapter->vfinfo = NULL;
196 adapter->num_vfs = 0;
197 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
200 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
202 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
203 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
204 schedule_work(&adapter->service_task);
207 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
209 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
211 /* flush memory to make sure state is correct before next watchog */
212 smp_mb__before_clear_bit();
213 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
216 struct ixgbe_reg_info {
217 u32 ofs;
218 char *name;
221 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
223 /* General Registers */
224 {IXGBE_CTRL, "CTRL"},
225 {IXGBE_STATUS, "STATUS"},
226 {IXGBE_CTRL_EXT, "CTRL_EXT"},
228 /* Interrupt Registers */
229 {IXGBE_EICR, "EICR"},
231 /* RX Registers */
232 {IXGBE_SRRCTL(0), "SRRCTL"},
233 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
234 {IXGBE_RDLEN(0), "RDLEN"},
235 {IXGBE_RDH(0), "RDH"},
236 {IXGBE_RDT(0), "RDT"},
237 {IXGBE_RXDCTL(0), "RXDCTL"},
238 {IXGBE_RDBAL(0), "RDBAL"},
239 {IXGBE_RDBAH(0), "RDBAH"},
241 /* TX Registers */
242 {IXGBE_TDBAL(0), "TDBAL"},
243 {IXGBE_TDBAH(0), "TDBAH"},
244 {IXGBE_TDLEN(0), "TDLEN"},
245 {IXGBE_TDH(0), "TDH"},
246 {IXGBE_TDT(0), "TDT"},
247 {IXGBE_TXDCTL(0), "TXDCTL"},
249 /* List Terminator */
255 * ixgbe_regdump - register printout routine
257 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
259 int i = 0, j = 0;
260 char rname[16];
261 u32 regs[64];
263 switch (reginfo->ofs) {
264 case IXGBE_SRRCTL(0):
265 for (i = 0; i < 64; i++)
266 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
267 break;
268 case IXGBE_DCA_RXCTRL(0):
269 for (i = 0; i < 64; i++)
270 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
271 break;
272 case IXGBE_RDLEN(0):
273 for (i = 0; i < 64; i++)
274 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
275 break;
276 case IXGBE_RDH(0):
277 for (i = 0; i < 64; i++)
278 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
279 break;
280 case IXGBE_RDT(0):
281 for (i = 0; i < 64; i++)
282 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
283 break;
284 case IXGBE_RXDCTL(0):
285 for (i = 0; i < 64; i++)
286 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
287 break;
288 case IXGBE_RDBAL(0):
289 for (i = 0; i < 64; i++)
290 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
291 break;
292 case IXGBE_RDBAH(0):
293 for (i = 0; i < 64; i++)
294 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
295 break;
296 case IXGBE_TDBAL(0):
297 for (i = 0; i < 64; i++)
298 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
299 break;
300 case IXGBE_TDBAH(0):
301 for (i = 0; i < 64; i++)
302 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
303 break;
304 case IXGBE_TDLEN(0):
305 for (i = 0; i < 64; i++)
306 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
307 break;
308 case IXGBE_TDH(0):
309 for (i = 0; i < 64; i++)
310 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
311 break;
312 case IXGBE_TDT(0):
313 for (i = 0; i < 64; i++)
314 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
315 break;
316 case IXGBE_TXDCTL(0):
317 for (i = 0; i < 64; i++)
318 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
319 break;
320 default:
321 pr_info("%-15s %08x\n", reginfo->name,
322 IXGBE_READ_REG(hw, reginfo->ofs));
323 return;
326 for (i = 0; i < 8; i++) {
327 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
328 pr_err("%-15s", rname);
329 for (j = 0; j < 8; j++)
330 pr_cont(" %08x", regs[i*8+j]);
331 pr_cont("\n");
337 * ixgbe_dump - Print registers, tx-rings and rx-rings
339 static void ixgbe_dump(struct ixgbe_adapter *adapter)
341 struct net_device *netdev = adapter->netdev;
342 struct ixgbe_hw *hw = &adapter->hw;
343 struct ixgbe_reg_info *reginfo;
344 int n = 0;
345 struct ixgbe_ring *tx_ring;
346 struct ixgbe_tx_buffer *tx_buffer_info;
347 union ixgbe_adv_tx_desc *tx_desc;
348 struct my_u0 { u64 a; u64 b; } *u0;
349 struct ixgbe_ring *rx_ring;
350 union ixgbe_adv_rx_desc *rx_desc;
351 struct ixgbe_rx_buffer *rx_buffer_info;
352 u32 staterr;
353 int i = 0;
355 if (!netif_msg_hw(adapter))
356 return;
358 /* Print netdevice Info */
359 if (netdev) {
360 dev_info(&adapter->pdev->dev, "Net device Info\n");
361 pr_info("Device Name state "
362 "trans_start last_rx\n");
363 pr_info("%-15s %016lX %016lX %016lX\n",
364 netdev->name,
365 netdev->state,
366 netdev->trans_start,
367 netdev->last_rx);
370 /* Print Registers */
371 dev_info(&adapter->pdev->dev, "Register Dump\n");
372 pr_info(" Register Name Value\n");
373 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
374 reginfo->name; reginfo++) {
375 ixgbe_regdump(hw, reginfo);
378 /* Print TX Ring Summary */
379 if (!netdev || !netif_running(netdev))
380 goto exit;
382 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
383 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
384 for (n = 0; n < adapter->num_tx_queues; n++) {
385 tx_ring = adapter->tx_ring[n];
386 tx_buffer_info =
387 &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
388 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
389 n, tx_ring->next_to_use, tx_ring->next_to_clean,
390 (u64)tx_buffer_info->dma,
391 tx_buffer_info->length,
392 tx_buffer_info->next_to_watch,
393 (u64)tx_buffer_info->time_stamp);
396 /* Print TX Rings */
397 if (!netif_msg_tx_done(adapter))
398 goto rx_ring_summary;
400 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
402 /* Transmit Descriptor Formats
404 * Advanced Transmit Descriptor
405 * +--------------------------------------------------------------+
406 * 0 | Buffer Address [63:0] |
407 * +--------------------------------------------------------------+
408 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
409 * +--------------------------------------------------------------+
410 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
413 for (n = 0; n < adapter->num_tx_queues; n++) {
414 tx_ring = adapter->tx_ring[n];
415 pr_info("------------------------------------\n");
416 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
417 pr_info("------------------------------------\n");
418 pr_info("T [desc] [address 63:0 ] "
419 "[PlPOIdStDDt Ln] [bi->dma ] "
420 "leng ntw timestamp bi->skb\n");
422 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
423 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
424 tx_buffer_info = &tx_ring->tx_buffer_info[i];
425 u0 = (struct my_u0 *)tx_desc;
426 pr_info("T [0x%03X] %016llX %016llX %016llX"
427 " %04X %p %016llX %p", i,
428 le64_to_cpu(u0->a),
429 le64_to_cpu(u0->b),
430 (u64)tx_buffer_info->dma,
431 tx_buffer_info->length,
432 tx_buffer_info->next_to_watch,
433 (u64)tx_buffer_info->time_stamp,
434 tx_buffer_info->skb);
435 if (i == tx_ring->next_to_use &&
436 i == tx_ring->next_to_clean)
437 pr_cont(" NTC/U\n");
438 else if (i == tx_ring->next_to_use)
439 pr_cont(" NTU\n");
440 else if (i == tx_ring->next_to_clean)
441 pr_cont(" NTC\n");
442 else
443 pr_cont("\n");
445 if (netif_msg_pktdata(adapter) &&
446 tx_buffer_info->dma != 0)
447 print_hex_dump(KERN_INFO, "",
448 DUMP_PREFIX_ADDRESS, 16, 1,
449 phys_to_virt(tx_buffer_info->dma),
450 tx_buffer_info->length, true);
454 /* Print RX Rings Summary */
455 rx_ring_summary:
456 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
457 pr_info("Queue [NTU] [NTC]\n");
458 for (n = 0; n < adapter->num_rx_queues; n++) {
459 rx_ring = adapter->rx_ring[n];
460 pr_info("%5d %5X %5X\n",
461 n, rx_ring->next_to_use, rx_ring->next_to_clean);
464 /* Print RX Rings */
465 if (!netif_msg_rx_status(adapter))
466 goto exit;
468 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
470 /* Advanced Receive Descriptor (Read) Format
471 * 63 1 0
472 * +-----------------------------------------------------+
473 * 0 | Packet Buffer Address [63:1] |A0/NSE|
474 * +----------------------------------------------+------+
475 * 8 | Header Buffer Address [63:1] | DD |
476 * +-----------------------------------------------------+
479 * Advanced Receive Descriptor (Write-Back) Format
481 * 63 48 47 32 31 30 21 20 16 15 4 3 0
482 * +------------------------------------------------------+
483 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
484 * | Checksum Ident | | | | Type | Type |
485 * +------------------------------------------------------+
486 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
487 * +------------------------------------------------------+
488 * 63 48 47 32 31 20 19 0
490 for (n = 0; n < adapter->num_rx_queues; n++) {
491 rx_ring = adapter->rx_ring[n];
492 pr_info("------------------------------------\n");
493 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
494 pr_info("------------------------------------\n");
495 pr_info("R [desc] [ PktBuf A0] "
496 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
497 "<-- Adv Rx Read format\n");
498 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
499 "[vl er S cks ln] ---------------- [bi->skb] "
500 "<-- Adv Rx Write-Back format\n");
502 for (i = 0; i < rx_ring->count; i++) {
503 rx_buffer_info = &rx_ring->rx_buffer_info[i];
504 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
505 u0 = (struct my_u0 *)rx_desc;
506 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
507 if (staterr & IXGBE_RXD_STAT_DD) {
508 /* Descriptor Done */
509 pr_info("RWB[0x%03X] %016llX "
510 "%016llX ---------------- %p", i,
511 le64_to_cpu(u0->a),
512 le64_to_cpu(u0->b),
513 rx_buffer_info->skb);
514 } else {
515 pr_info("R [0x%03X] %016llX "
516 "%016llX %016llX %p", i,
517 le64_to_cpu(u0->a),
518 le64_to_cpu(u0->b),
519 (u64)rx_buffer_info->dma,
520 rx_buffer_info->skb);
522 if (netif_msg_pktdata(adapter)) {
523 print_hex_dump(KERN_INFO, "",
524 DUMP_PREFIX_ADDRESS, 16, 1,
525 phys_to_virt(rx_buffer_info->dma),
526 rx_ring->rx_buf_len, true);
528 if (rx_ring->rx_buf_len
529 < IXGBE_RXBUFFER_2048)
530 print_hex_dump(KERN_INFO, "",
531 DUMP_PREFIX_ADDRESS, 16, 1,
532 phys_to_virt(
533 rx_buffer_info->page_dma +
534 rx_buffer_info->page_offset
536 PAGE_SIZE/2, true);
540 if (i == rx_ring->next_to_use)
541 pr_cont(" NTU\n");
542 else if (i == rx_ring->next_to_clean)
543 pr_cont(" NTC\n");
544 else
545 pr_cont("\n");
550 exit:
551 return;
554 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
556 u32 ctrl_ext;
558 /* Let firmware take over control of h/w */
559 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
560 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
561 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
564 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
566 u32 ctrl_ext;
568 /* Let firmware know the driver has taken over */
569 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
570 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
571 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
575 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
576 * @adapter: pointer to adapter struct
577 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
578 * @queue: queue to map the corresponding interrupt to
579 * @msix_vector: the vector to map to the corresponding queue
582 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
583 u8 queue, u8 msix_vector)
585 u32 ivar, index;
586 struct ixgbe_hw *hw = &adapter->hw;
587 switch (hw->mac.type) {
588 case ixgbe_mac_82598EB:
589 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
590 if (direction == -1)
591 direction = 0;
592 index = (((direction * 64) + queue) >> 2) & 0x1F;
593 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
594 ivar &= ~(0xFF << (8 * (queue & 0x3)));
595 ivar |= (msix_vector << (8 * (queue & 0x3)));
596 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
597 break;
598 case ixgbe_mac_82599EB:
599 case ixgbe_mac_X540:
600 if (direction == -1) {
601 /* other causes */
602 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
603 index = ((queue & 1) * 8);
604 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
605 ivar &= ~(0xFF << index);
606 ivar |= (msix_vector << index);
607 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
608 break;
609 } else {
610 /* tx or rx causes */
611 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
612 index = ((16 * (queue & 1)) + (8 * direction));
613 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
614 ivar &= ~(0xFF << index);
615 ivar |= (msix_vector << index);
616 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
617 break;
619 default:
620 break;
624 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
625 u64 qmask)
627 u32 mask;
629 switch (adapter->hw.mac.type) {
630 case ixgbe_mac_82598EB:
631 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
632 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
633 break;
634 case ixgbe_mac_82599EB:
635 case ixgbe_mac_X540:
636 mask = (qmask & 0xFFFFFFFF);
637 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
638 mask = (qmask >> 32);
639 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
640 break;
641 default:
642 break;
646 static inline void ixgbe_unmap_tx_resource(struct ixgbe_ring *ring,
647 struct ixgbe_tx_buffer *tx_buffer)
649 if (tx_buffer->dma) {
650 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_MAPPED_AS_PAGE)
651 dma_unmap_page(ring->dev,
652 tx_buffer->dma,
653 tx_buffer->length,
654 DMA_TO_DEVICE);
655 else
656 dma_unmap_single(ring->dev,
657 tx_buffer->dma,
658 tx_buffer->length,
659 DMA_TO_DEVICE);
661 tx_buffer->dma = 0;
664 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
665 struct ixgbe_tx_buffer *tx_buffer_info)
667 ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
668 if (tx_buffer_info->skb)
669 dev_kfree_skb_any(tx_buffer_info->skb);
670 tx_buffer_info->skb = NULL;
671 /* tx_buffer_info must be completely set up in the transmit path */
674 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
676 struct ixgbe_hw *hw = &adapter->hw;
677 struct ixgbe_hw_stats *hwstats = &adapter->stats;
678 u32 data = 0;
679 u32 xoff[8] = {0};
680 int i;
682 if ((hw->fc.current_mode == ixgbe_fc_full) ||
683 (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
684 switch (hw->mac.type) {
685 case ixgbe_mac_82598EB:
686 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
687 break;
688 default:
689 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
691 hwstats->lxoffrxc += data;
693 /* refill credits (no tx hang) if we received xoff */
694 if (!data)
695 return;
697 for (i = 0; i < adapter->num_tx_queues; i++)
698 clear_bit(__IXGBE_HANG_CHECK_ARMED,
699 &adapter->tx_ring[i]->state);
700 return;
701 } else if (!(adapter->dcb_cfg.pfc_mode_enable))
702 return;
704 /* update stats for each tc, only valid with PFC enabled */
705 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
706 switch (hw->mac.type) {
707 case ixgbe_mac_82598EB:
708 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
709 break;
710 default:
711 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
713 hwstats->pxoffrxc[i] += xoff[i];
716 /* disarm tx queues that have received xoff frames */
717 for (i = 0; i < adapter->num_tx_queues; i++) {
718 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
719 u8 tc = tx_ring->dcb_tc;
721 if (xoff[tc])
722 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
726 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
728 return ring->tx_stats.completed;
731 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
733 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
734 struct ixgbe_hw *hw = &adapter->hw;
736 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
737 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
739 if (head != tail)
740 return (head < tail) ?
741 tail - head : (tail + ring->count - head);
743 return 0;
746 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
748 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
749 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
750 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
751 bool ret = false;
753 clear_check_for_tx_hang(tx_ring);
756 * Check for a hung queue, but be thorough. This verifies
757 * that a transmit has been completed since the previous
758 * check AND there is at least one packet pending. The
759 * ARMED bit is set to indicate a potential hang. The
760 * bit is cleared if a pause frame is received to remove
761 * false hang detection due to PFC or 802.3x frames. By
762 * requiring this to fail twice we avoid races with
763 * pfc clearing the ARMED bit and conditions where we
764 * run the check_tx_hang logic with a transmit completion
765 * pending but without time to complete it yet.
767 if ((tx_done_old == tx_done) && tx_pending) {
768 /* make sure it is true for two checks in a row */
769 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
770 &tx_ring->state);
771 } else {
772 /* update completed stats and continue */
773 tx_ring->tx_stats.tx_done_old = tx_done;
774 /* reset the countdown */
775 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
778 return ret;
782 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
783 * @adapter: driver private struct
785 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
788 /* Do the reset outside of interrupt context */
789 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
790 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
791 ixgbe_service_event_schedule(adapter);
796 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
797 * @q_vector: structure containing interrupt and ring information
798 * @tx_ring: tx ring to clean
800 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
801 struct ixgbe_ring *tx_ring)
803 struct ixgbe_adapter *adapter = q_vector->adapter;
804 struct ixgbe_tx_buffer *tx_buffer;
805 union ixgbe_adv_tx_desc *tx_desc;
806 unsigned int total_bytes = 0, total_packets = 0;
807 u16 budget = q_vector->tx.work_limit;
808 u16 i = tx_ring->next_to_clean;
810 tx_buffer = &tx_ring->tx_buffer_info[i];
811 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
813 for (; budget; budget--) {
814 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
816 /* if next_to_watch is not set then there is no work pending */
817 if (!eop_desc)
818 break;
820 /* if DD is not set pending work has not been completed */
821 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
822 break;
824 /* count the packet as being completed */
825 tx_ring->tx_stats.completed++;
827 /* clear next_to_watch to prevent false hangs */
828 tx_buffer->next_to_watch = NULL;
830 /* prevent any other reads prior to eop_desc being verified */
831 rmb();
833 do {
834 ixgbe_unmap_tx_resource(tx_ring, tx_buffer);
835 tx_desc->wb.status = 0;
836 if (likely(tx_desc == eop_desc)) {
837 eop_desc = NULL;
838 dev_kfree_skb_any(tx_buffer->skb);
839 tx_buffer->skb = NULL;
841 total_bytes += tx_buffer->bytecount;
842 total_packets += tx_buffer->gso_segs;
845 tx_buffer++;
846 tx_desc++;
847 i++;
848 if (unlikely(i == tx_ring->count)) {
849 i = 0;
851 tx_buffer = tx_ring->tx_buffer_info;
852 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
855 } while (eop_desc);
858 tx_ring->next_to_clean = i;
859 u64_stats_update_begin(&tx_ring->syncp);
860 tx_ring->stats.bytes += total_bytes;
861 tx_ring->stats.packets += total_packets;
862 u64_stats_update_end(&tx_ring->syncp);
863 q_vector->tx.total_bytes += total_bytes;
864 q_vector->tx.total_packets += total_packets;
866 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
867 /* schedule immediate reset if we believe we hung */
868 struct ixgbe_hw *hw = &adapter->hw;
869 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
870 e_err(drv, "Detected Tx Unit Hang\n"
871 " Tx Queue <%d>\n"
872 " TDH, TDT <%x>, <%x>\n"
873 " next_to_use <%x>\n"
874 " next_to_clean <%x>\n"
875 "tx_buffer_info[next_to_clean]\n"
876 " time_stamp <%lx>\n"
877 " jiffies <%lx>\n",
878 tx_ring->queue_index,
879 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
880 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
881 tx_ring->next_to_use, i,
882 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
884 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
886 e_info(probe,
887 "tx hang %d detected on queue %d, resetting adapter\n",
888 adapter->tx_timeout_count + 1, tx_ring->queue_index);
890 /* schedule immediate reset if we believe we hung */
891 ixgbe_tx_timeout_reset(adapter);
893 /* the adapter is about to reset, no point in enabling stuff */
894 return budget;
897 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
898 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
899 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
900 /* Make sure that anybody stopping the queue after this
901 * sees the new next_to_clean.
903 smp_mb();
904 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
905 !test_bit(__IXGBE_DOWN, &adapter->state)) {
906 netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
907 ++tx_ring->tx_stats.restart_queue;
911 return budget;
914 #ifdef CONFIG_IXGBE_DCA
915 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
916 struct ixgbe_ring *rx_ring,
917 int cpu)
919 struct ixgbe_hw *hw = &adapter->hw;
920 u32 rxctrl;
921 u8 reg_idx = rx_ring->reg_idx;
923 rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
924 switch (hw->mac.type) {
925 case ixgbe_mac_82598EB:
926 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
927 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
928 break;
929 case ixgbe_mac_82599EB:
930 case ixgbe_mac_X540:
931 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
932 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
933 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
934 break;
935 default:
936 break;
938 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
939 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
940 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
941 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
944 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
945 struct ixgbe_ring *tx_ring,
946 int cpu)
948 struct ixgbe_hw *hw = &adapter->hw;
949 u32 txctrl;
950 u8 reg_idx = tx_ring->reg_idx;
952 switch (hw->mac.type) {
953 case ixgbe_mac_82598EB:
954 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
955 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
956 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
957 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
958 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
959 break;
960 case ixgbe_mac_82599EB:
961 case ixgbe_mac_X540:
962 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
963 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
964 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
965 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
966 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
967 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
968 break;
969 default:
970 break;
974 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
976 struct ixgbe_adapter *adapter = q_vector->adapter;
977 struct ixgbe_ring *ring;
978 int cpu = get_cpu();
980 if (q_vector->cpu == cpu)
981 goto out_no_update;
983 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
984 ixgbe_update_tx_dca(adapter, ring, cpu);
986 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
987 ixgbe_update_rx_dca(adapter, ring, cpu);
989 q_vector->cpu = cpu;
990 out_no_update:
991 put_cpu();
994 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
996 int num_q_vectors;
997 int i;
999 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1000 return;
1002 /* always use CB2 mode, difference is masked in the CB driver */
1003 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1005 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
1006 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1007 else
1008 num_q_vectors = 1;
1010 for (i = 0; i < num_q_vectors; i++) {
1011 adapter->q_vector[i]->cpu = -1;
1012 ixgbe_update_dca(adapter->q_vector[i]);
1016 static int __ixgbe_notify_dca(struct device *dev, void *data)
1018 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1019 unsigned long event = *(unsigned long *)data;
1021 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1022 return 0;
1024 switch (event) {
1025 case DCA_PROVIDER_ADD:
1026 /* if we're already enabled, don't do it again */
1027 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1028 break;
1029 if (dca_add_requester(dev) == 0) {
1030 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1031 ixgbe_setup_dca(adapter);
1032 break;
1034 /* Fall Through since DCA is disabled. */
1035 case DCA_PROVIDER_REMOVE:
1036 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1037 dca_remove_requester(dev);
1038 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1039 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1041 break;
1044 return 0;
1046 #endif /* CONFIG_IXGBE_DCA */
1048 static inline void ixgbe_rx_hash(union ixgbe_adv_rx_desc *rx_desc,
1049 struct sk_buff *skb)
1051 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1055 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1056 * @adapter: address of board private structure
1057 * @rx_desc: advanced rx descriptor
1059 * Returns : true if it is FCoE pkt
1061 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_adapter *adapter,
1062 union ixgbe_adv_rx_desc *rx_desc)
1064 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1066 return (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
1067 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1068 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1069 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1073 * ixgbe_receive_skb - Send a completed packet up the stack
1074 * @adapter: board private structure
1075 * @skb: packet to send up
1076 * @status: hardware indication of status of receive
1077 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1078 * @rx_desc: rx descriptor
1080 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
1081 struct sk_buff *skb, u8 status,
1082 struct ixgbe_ring *ring,
1083 union ixgbe_adv_rx_desc *rx_desc)
1085 struct ixgbe_adapter *adapter = q_vector->adapter;
1086 struct napi_struct *napi = &q_vector->napi;
1087 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
1088 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
1090 if (is_vlan && (tag & VLAN_VID_MASK))
1091 __vlan_hwaccel_put_tag(skb, tag);
1093 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1094 napi_gro_receive(napi, skb);
1095 else
1096 netif_rx(skb);
1100 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1101 * @adapter: address of board private structure
1102 * @status_err: hardware indication of status of receive
1103 * @skb: skb currently being received and modified
1104 * @status_err: status error value of last descriptor in packet
1106 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
1107 union ixgbe_adv_rx_desc *rx_desc,
1108 struct sk_buff *skb,
1109 u32 status_err)
1111 skb->ip_summed = CHECKSUM_NONE;
1113 /* Rx csum disabled */
1114 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1115 return;
1117 /* if IP and error */
1118 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1119 (status_err & IXGBE_RXDADV_ERR_IPE)) {
1120 adapter->hw_csum_rx_error++;
1121 return;
1124 if (!(status_err & IXGBE_RXD_STAT_L4CS))
1125 return;
1127 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1128 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1131 * 82599 errata, UDP frames with a 0 checksum can be marked as
1132 * checksum errors.
1134 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1135 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1136 return;
1138 adapter->hw_csum_rx_error++;
1139 return;
1142 /* It must be a TCP or UDP packet with a valid checksum */
1143 skb->ip_summed = CHECKSUM_UNNECESSARY;
1146 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1149 * Force memory writes to complete before letting h/w
1150 * know there are new descriptors to fetch. (Only
1151 * applicable for weak-ordered memory model archs,
1152 * such as IA-64).
1154 wmb();
1155 writel(val, rx_ring->tail);
1159 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1160 * @rx_ring: ring to place buffers on
1161 * @cleaned_count: number of buffers to replace
1163 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1165 union ixgbe_adv_rx_desc *rx_desc;
1166 struct ixgbe_rx_buffer *bi;
1167 struct sk_buff *skb;
1168 u16 i = rx_ring->next_to_use;
1170 /* do nothing if no valid netdev defined */
1171 if (!rx_ring->netdev)
1172 return;
1174 while (cleaned_count--) {
1175 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1176 bi = &rx_ring->rx_buffer_info[i];
1177 skb = bi->skb;
1179 if (!skb) {
1180 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1181 rx_ring->rx_buf_len);
1182 if (!skb) {
1183 rx_ring->rx_stats.alloc_rx_buff_failed++;
1184 goto no_buffers;
1186 /* initialize queue mapping */
1187 skb_record_rx_queue(skb, rx_ring->queue_index);
1188 bi->skb = skb;
1191 if (!bi->dma) {
1192 bi->dma = dma_map_single(rx_ring->dev,
1193 skb->data,
1194 rx_ring->rx_buf_len,
1195 DMA_FROM_DEVICE);
1196 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1197 rx_ring->rx_stats.alloc_rx_buff_failed++;
1198 bi->dma = 0;
1199 goto no_buffers;
1203 if (ring_is_ps_enabled(rx_ring)) {
1204 if (!bi->page) {
1205 bi->page = netdev_alloc_page(rx_ring->netdev);
1206 if (!bi->page) {
1207 rx_ring->rx_stats.alloc_rx_page_failed++;
1208 goto no_buffers;
1212 if (!bi->page_dma) {
1213 /* use a half page if we're re-using */
1214 bi->page_offset ^= PAGE_SIZE / 2;
1215 bi->page_dma = dma_map_page(rx_ring->dev,
1216 bi->page,
1217 bi->page_offset,
1218 PAGE_SIZE / 2,
1219 DMA_FROM_DEVICE);
1220 if (dma_mapping_error(rx_ring->dev,
1221 bi->page_dma)) {
1222 rx_ring->rx_stats.alloc_rx_page_failed++;
1223 bi->page_dma = 0;
1224 goto no_buffers;
1228 /* Refresh the desc even if buffer_addrs didn't change
1229 * because each write-back erases this info. */
1230 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1231 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1232 } else {
1233 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1234 rx_desc->read.hdr_addr = 0;
1237 i++;
1238 if (i == rx_ring->count)
1239 i = 0;
1242 no_buffers:
1243 if (rx_ring->next_to_use != i) {
1244 rx_ring->next_to_use = i;
1245 ixgbe_release_rx_desc(rx_ring, i);
1249 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1251 /* HW will not DMA in data larger than the given buffer, even if it
1252 * parses the (NFS, of course) header to be larger. In that case, it
1253 * fills the header buffer and spills the rest into the page.
1255 u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1256 u16 hlen = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1257 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1258 if (hlen > IXGBE_RX_HDR_SIZE)
1259 hlen = IXGBE_RX_HDR_SIZE;
1260 return hlen;
1264 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1265 * @skb: pointer to the last skb in the rsc queue
1267 * This function changes a queue full of hw rsc buffers into a completed
1268 * packet. It uses the ->prev pointers to find the first packet and then
1269 * turns it into the frag list owner.
1271 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1273 unsigned int frag_list_size = 0;
1274 unsigned int skb_cnt = 1;
1276 while (skb->prev) {
1277 struct sk_buff *prev = skb->prev;
1278 frag_list_size += skb->len;
1279 skb->prev = NULL;
1280 skb = prev;
1281 skb_cnt++;
1284 skb_shinfo(skb)->frag_list = skb->next;
1285 skb->next = NULL;
1286 skb->len += frag_list_size;
1287 skb->data_len += frag_list_size;
1288 skb->truesize += frag_list_size;
1289 IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1291 return skb;
1294 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1296 return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1297 IXGBE_RXDADV_RSCCNT_MASK);
1300 static void ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1301 struct ixgbe_ring *rx_ring,
1302 int *work_done, int work_to_do)
1304 struct ixgbe_adapter *adapter = q_vector->adapter;
1305 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1306 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1307 struct sk_buff *skb;
1308 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1309 const int current_node = numa_node_id();
1310 #ifdef IXGBE_FCOE
1311 int ddp_bytes = 0;
1312 #endif /* IXGBE_FCOE */
1313 u32 staterr;
1314 u16 i;
1315 u16 cleaned_count = 0;
1316 bool pkt_is_rsc = false;
1318 i = rx_ring->next_to_clean;
1319 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1320 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1322 while (staterr & IXGBE_RXD_STAT_DD) {
1323 u32 upper_len = 0;
1325 rmb(); /* read descriptor and rx_buffer_info after status DD */
1327 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1329 skb = rx_buffer_info->skb;
1330 rx_buffer_info->skb = NULL;
1331 prefetch(skb->data);
1333 if (ring_is_rsc_enabled(rx_ring))
1334 pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1336 /* if this is a skb from previous receive DMA will be 0 */
1337 if (rx_buffer_info->dma) {
1338 u16 hlen;
1339 if (pkt_is_rsc &&
1340 !(staterr & IXGBE_RXD_STAT_EOP) &&
1341 !skb->prev) {
1343 * When HWRSC is enabled, delay unmapping
1344 * of the first packet. It carries the
1345 * header information, HW may still
1346 * access the header after the writeback.
1347 * Only unmap it when EOP is reached
1349 IXGBE_RSC_CB(skb)->delay_unmap = true;
1350 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1351 } else {
1352 dma_unmap_single(rx_ring->dev,
1353 rx_buffer_info->dma,
1354 rx_ring->rx_buf_len,
1355 DMA_FROM_DEVICE);
1357 rx_buffer_info->dma = 0;
1359 if (ring_is_ps_enabled(rx_ring)) {
1360 hlen = ixgbe_get_hlen(rx_desc);
1361 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1362 } else {
1363 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1366 skb_put(skb, hlen);
1367 } else {
1368 /* assume packet split since header is unmapped */
1369 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1372 if (upper_len) {
1373 dma_unmap_page(rx_ring->dev,
1374 rx_buffer_info->page_dma,
1375 PAGE_SIZE / 2,
1376 DMA_FROM_DEVICE);
1377 rx_buffer_info->page_dma = 0;
1378 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1379 rx_buffer_info->page,
1380 rx_buffer_info->page_offset,
1381 upper_len);
1383 if ((page_count(rx_buffer_info->page) == 1) &&
1384 (page_to_nid(rx_buffer_info->page) == current_node))
1385 get_page(rx_buffer_info->page);
1386 else
1387 rx_buffer_info->page = NULL;
1389 skb->len += upper_len;
1390 skb->data_len += upper_len;
1391 skb->truesize += upper_len;
1394 i++;
1395 if (i == rx_ring->count)
1396 i = 0;
1398 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1399 prefetch(next_rxd);
1400 cleaned_count++;
1402 if (pkt_is_rsc) {
1403 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1404 IXGBE_RXDADV_NEXTP_SHIFT;
1405 next_buffer = &rx_ring->rx_buffer_info[nextp];
1406 } else {
1407 next_buffer = &rx_ring->rx_buffer_info[i];
1410 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1411 if (ring_is_ps_enabled(rx_ring)) {
1412 rx_buffer_info->skb = next_buffer->skb;
1413 rx_buffer_info->dma = next_buffer->dma;
1414 next_buffer->skb = skb;
1415 next_buffer->dma = 0;
1416 } else {
1417 skb->next = next_buffer->skb;
1418 skb->next->prev = skb;
1420 rx_ring->rx_stats.non_eop_descs++;
1421 goto next_desc;
1424 if (skb->prev) {
1425 skb = ixgbe_transform_rsc_queue(skb);
1426 /* if we got here without RSC the packet is invalid */
1427 if (!pkt_is_rsc) {
1428 __pskb_trim(skb, 0);
1429 rx_buffer_info->skb = skb;
1430 goto next_desc;
1434 if (ring_is_rsc_enabled(rx_ring)) {
1435 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1436 dma_unmap_single(rx_ring->dev,
1437 IXGBE_RSC_CB(skb)->dma,
1438 rx_ring->rx_buf_len,
1439 DMA_FROM_DEVICE);
1440 IXGBE_RSC_CB(skb)->dma = 0;
1441 IXGBE_RSC_CB(skb)->delay_unmap = false;
1444 if (pkt_is_rsc) {
1445 if (ring_is_ps_enabled(rx_ring))
1446 rx_ring->rx_stats.rsc_count +=
1447 skb_shinfo(skb)->nr_frags;
1448 else
1449 rx_ring->rx_stats.rsc_count +=
1450 IXGBE_RSC_CB(skb)->skb_cnt;
1451 rx_ring->rx_stats.rsc_flush++;
1454 /* ERR_MASK will only have valid bits if EOP set */
1455 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
1456 dev_kfree_skb_any(skb);
1457 goto next_desc;
1460 ixgbe_rx_checksum(adapter, rx_desc, skb, staterr);
1461 if (adapter->netdev->features & NETIF_F_RXHASH)
1462 ixgbe_rx_hash(rx_desc, skb);
1464 /* probably a little skewed due to removing CRC */
1465 total_rx_bytes += skb->len;
1466 total_rx_packets++;
1468 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1469 #ifdef IXGBE_FCOE
1470 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1471 if (ixgbe_rx_is_fcoe(adapter, rx_desc)) {
1472 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb,
1473 staterr);
1474 if (!ddp_bytes) {
1475 dev_kfree_skb_any(skb);
1476 goto next_desc;
1479 #endif /* IXGBE_FCOE */
1480 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1482 next_desc:
1483 rx_desc->wb.upper.status_error = 0;
1485 (*work_done)++;
1486 if (*work_done >= work_to_do)
1487 break;
1489 /* return some buffers to hardware, one at a time is too slow */
1490 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1491 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1492 cleaned_count = 0;
1495 /* use prefetched values */
1496 rx_desc = next_rxd;
1497 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1500 rx_ring->next_to_clean = i;
1501 cleaned_count = ixgbe_desc_unused(rx_ring);
1503 if (cleaned_count)
1504 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1506 #ifdef IXGBE_FCOE
1507 /* include DDPed FCoE data */
1508 if (ddp_bytes > 0) {
1509 unsigned int mss;
1511 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1512 sizeof(struct fc_frame_header) -
1513 sizeof(struct fcoe_crc_eof);
1514 if (mss > 512)
1515 mss &= ~511;
1516 total_rx_bytes += ddp_bytes;
1517 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1519 #endif /* IXGBE_FCOE */
1521 u64_stats_update_begin(&rx_ring->syncp);
1522 rx_ring->stats.packets += total_rx_packets;
1523 rx_ring->stats.bytes += total_rx_bytes;
1524 u64_stats_update_end(&rx_ring->syncp);
1525 q_vector->rx.total_packets += total_rx_packets;
1526 q_vector->rx.total_bytes += total_rx_bytes;
1529 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1531 * ixgbe_configure_msix - Configure MSI-X hardware
1532 * @adapter: board private structure
1534 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1535 * interrupts.
1537 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1539 struct ixgbe_q_vector *q_vector;
1540 int q_vectors, v_idx;
1541 u32 mask;
1543 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1546 * Populate the IVAR table and set the ITR values to the
1547 * corresponding register.
1549 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1550 struct ixgbe_ring *ring;
1551 q_vector = adapter->q_vector[v_idx];
1553 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
1554 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1556 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
1557 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1559 if (q_vector->tx.ring && !q_vector->rx.ring)
1560 /* tx only */
1561 q_vector->eitr = adapter->tx_eitr_param;
1562 else if (q_vector->rx.ring)
1563 /* rx or mixed */
1564 q_vector->eitr = adapter->rx_eitr_param;
1566 ixgbe_write_eitr(q_vector);
1567 /* If ATR is enabled, set interrupt affinity */
1568 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
1570 * Allocate the affinity_hint cpumask, assign the mask
1571 * for this vector, and set our affinity_hint for
1572 * this irq.
1574 if (!alloc_cpumask_var(&q_vector->affinity_mask,
1575 GFP_KERNEL))
1576 return;
1577 cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1578 irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1579 q_vector->affinity_mask);
1583 switch (adapter->hw.mac.type) {
1584 case ixgbe_mac_82598EB:
1585 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1586 v_idx);
1587 break;
1588 case ixgbe_mac_82599EB:
1589 case ixgbe_mac_X540:
1590 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1591 break;
1593 default:
1594 break;
1596 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1598 /* set up to autoclear timer, and the vectors */
1599 mask = IXGBE_EIMS_ENABLE_MASK;
1600 if (adapter->num_vfs)
1601 mask &= ~(IXGBE_EIMS_OTHER |
1602 IXGBE_EIMS_MAILBOX |
1603 IXGBE_EIMS_LSC);
1604 else
1605 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1606 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1609 enum latency_range {
1610 lowest_latency = 0,
1611 low_latency = 1,
1612 bulk_latency = 2,
1613 latency_invalid = 255
1617 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1618 * @q_vector: structure containing interrupt and ring information
1619 * @ring_container: structure containing ring performance data
1621 * Stores a new ITR value based on packets and byte
1622 * counts during the last interrupt. The advantage of per interrupt
1623 * computation is faster updates and more accurate ITR for the current
1624 * traffic pattern. Constants in this function were computed
1625 * based on theoretical maximum wire speed and thresholds were set based
1626 * on testing data as well as attempting to minimize response time
1627 * while increasing bulk throughput.
1628 * this functionality is controlled by the InterruptThrottleRate module
1629 * parameter (see ixgbe_param.c)
1631 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1632 struct ixgbe_ring_container *ring_container)
1634 u64 bytes_perint;
1635 struct ixgbe_adapter *adapter = q_vector->adapter;
1636 int bytes = ring_container->total_bytes;
1637 int packets = ring_container->total_packets;
1638 u32 timepassed_us;
1639 u8 itr_setting = ring_container->itr;
1641 if (packets == 0)
1642 return;
1644 /* simple throttlerate management
1645 * 0-20MB/s lowest (100000 ints/s)
1646 * 20-100MB/s low (20000 ints/s)
1647 * 100-1249MB/s bulk (8000 ints/s)
1649 /* what was last interrupt timeslice? */
1650 timepassed_us = 1000000/q_vector->eitr;
1651 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1653 switch (itr_setting) {
1654 case lowest_latency:
1655 if (bytes_perint > adapter->eitr_low)
1656 itr_setting = low_latency;
1657 break;
1658 case low_latency:
1659 if (bytes_perint > adapter->eitr_high)
1660 itr_setting = bulk_latency;
1661 else if (bytes_perint <= adapter->eitr_low)
1662 itr_setting = lowest_latency;
1663 break;
1664 case bulk_latency:
1665 if (bytes_perint <= adapter->eitr_high)
1666 itr_setting = low_latency;
1667 break;
1670 /* clear work counters since we have the values we need */
1671 ring_container->total_bytes = 0;
1672 ring_container->total_packets = 0;
1674 /* write updated itr to ring container */
1675 ring_container->itr = itr_setting;
1679 * ixgbe_write_eitr - write EITR register in hardware specific way
1680 * @q_vector: structure containing interrupt and ring information
1682 * This function is made to be called by ethtool and by the driver
1683 * when it needs to update EITR registers at runtime. Hardware
1684 * specific quirks/differences are taken care of here.
1686 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1688 struct ixgbe_adapter *adapter = q_vector->adapter;
1689 struct ixgbe_hw *hw = &adapter->hw;
1690 int v_idx = q_vector->v_idx;
1691 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1693 switch (adapter->hw.mac.type) {
1694 case ixgbe_mac_82598EB:
1695 /* must write high and low 16 bits to reset counter */
1696 itr_reg |= (itr_reg << 16);
1697 break;
1698 case ixgbe_mac_82599EB:
1699 case ixgbe_mac_X540:
1701 * 82599 and X540 can support a value of zero, so allow it for
1702 * max interrupt rate, but there is an errata where it can
1703 * not be zero with RSC
1705 if (itr_reg == 8 &&
1706 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1707 itr_reg = 0;
1710 * set the WDIS bit to not clear the timer bits and cause an
1711 * immediate assertion of the interrupt
1713 itr_reg |= IXGBE_EITR_CNT_WDIS;
1714 break;
1715 default:
1716 break;
1718 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1721 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1723 u32 new_itr = q_vector->eitr;
1724 u8 current_itr;
1726 ixgbe_update_itr(q_vector, &q_vector->tx);
1727 ixgbe_update_itr(q_vector, &q_vector->rx);
1729 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1731 switch (current_itr) {
1732 /* counts and packets in update_itr are dependent on these numbers */
1733 case lowest_latency:
1734 new_itr = 100000;
1735 break;
1736 case low_latency:
1737 new_itr = 20000; /* aka hwitr = ~200 */
1738 break;
1739 case bulk_latency:
1740 new_itr = 8000;
1741 break;
1742 default:
1743 break;
1746 if (new_itr != q_vector->eitr) {
1747 /* do an exponential smoothing */
1748 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
1750 /* save the algorithm value here */
1751 q_vector->eitr = new_itr;
1753 ixgbe_write_eitr(q_vector);
1758 * ixgbe_check_overtemp_subtask - check for over tempurature
1759 * @adapter: pointer to adapter
1761 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
1763 struct ixgbe_hw *hw = &adapter->hw;
1764 u32 eicr = adapter->interrupt_event;
1766 if (test_bit(__IXGBE_DOWN, &adapter->state))
1767 return;
1769 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1770 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
1771 return;
1773 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
1775 switch (hw->device_id) {
1776 case IXGBE_DEV_ID_82599_T3_LOM:
1778 * Since the warning interrupt is for both ports
1779 * we don't have to check if:
1780 * - This interrupt wasn't for our port.
1781 * - We may have missed the interrupt so always have to
1782 * check if we got a LSC
1784 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
1785 !(eicr & IXGBE_EICR_LSC))
1786 return;
1788 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
1789 u32 autoneg;
1790 bool link_up = false;
1792 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1794 if (link_up)
1795 return;
1798 /* Check if this is not due to overtemp */
1799 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
1800 return;
1802 break;
1803 default:
1804 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1805 return;
1806 break;
1808 e_crit(drv,
1809 "Network adapter has been stopped because it has over heated. "
1810 "Restart the computer. If the problem persists, "
1811 "power off the system and replace the adapter\n");
1813 adapter->interrupt_event = 0;
1816 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1818 struct ixgbe_hw *hw = &adapter->hw;
1820 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1821 (eicr & IXGBE_EICR_GPI_SDP1)) {
1822 e_crit(probe, "Fan has stopped, replace the adapter\n");
1823 /* write to clear the interrupt */
1824 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1828 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1830 struct ixgbe_hw *hw = &adapter->hw;
1832 if (eicr & IXGBE_EICR_GPI_SDP2) {
1833 /* Clear the interrupt */
1834 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1835 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1836 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
1837 ixgbe_service_event_schedule(adapter);
1841 if (eicr & IXGBE_EICR_GPI_SDP1) {
1842 /* Clear the interrupt */
1843 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1844 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1845 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
1846 ixgbe_service_event_schedule(adapter);
1851 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1853 struct ixgbe_hw *hw = &adapter->hw;
1855 adapter->lsc_int++;
1856 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1857 adapter->link_check_timeout = jiffies;
1858 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1859 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1860 IXGBE_WRITE_FLUSH(hw);
1861 ixgbe_service_event_schedule(adapter);
1865 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1867 struct ixgbe_adapter *adapter = data;
1868 struct ixgbe_hw *hw = &adapter->hw;
1869 u32 eicr;
1872 * Workaround for Silicon errata. Use clear-by-write instead
1873 * of clear-by-read. Reading with EICS will return the
1874 * interrupt causes without clearing, which later be done
1875 * with the write to EICR.
1877 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1878 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1880 if (eicr & IXGBE_EICR_LSC)
1881 ixgbe_check_lsc(adapter);
1883 if (eicr & IXGBE_EICR_MAILBOX)
1884 ixgbe_msg_task(adapter);
1886 switch (hw->mac.type) {
1887 case ixgbe_mac_82599EB:
1888 case ixgbe_mac_X540:
1889 /* Handle Flow Director Full threshold interrupt */
1890 if (eicr & IXGBE_EICR_FLOW_DIR) {
1891 int reinit_count = 0;
1892 int i;
1893 for (i = 0; i < adapter->num_tx_queues; i++) {
1894 struct ixgbe_ring *ring = adapter->tx_ring[i];
1895 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1896 &ring->state))
1897 reinit_count++;
1899 if (reinit_count) {
1900 /* no more flow director interrupts until after init */
1901 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
1902 eicr &= ~IXGBE_EICR_FLOW_DIR;
1903 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
1904 ixgbe_service_event_schedule(adapter);
1907 ixgbe_check_sfp_event(adapter, eicr);
1908 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1909 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
1910 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1911 adapter->interrupt_event = eicr;
1912 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
1913 ixgbe_service_event_schedule(adapter);
1916 break;
1917 default:
1918 break;
1921 ixgbe_check_fan_failure(adapter, eicr);
1923 /* re-enable the original interrupt state, no lsc, no queues */
1924 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1925 IXGBE_WRITE_REG(hw, IXGBE_EIMS, eicr &
1926 ~(IXGBE_EIMS_LSC | IXGBE_EIMS_RTX_QUEUE));
1928 return IRQ_HANDLED;
1931 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1932 u64 qmask)
1934 u32 mask;
1935 struct ixgbe_hw *hw = &adapter->hw;
1937 switch (hw->mac.type) {
1938 case ixgbe_mac_82598EB:
1939 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1940 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1941 break;
1942 case ixgbe_mac_82599EB:
1943 case ixgbe_mac_X540:
1944 mask = (qmask & 0xFFFFFFFF);
1945 if (mask)
1946 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1947 mask = (qmask >> 32);
1948 if (mask)
1949 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
1950 break;
1951 default:
1952 break;
1954 /* skip the flush */
1957 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1958 u64 qmask)
1960 u32 mask;
1961 struct ixgbe_hw *hw = &adapter->hw;
1963 switch (hw->mac.type) {
1964 case ixgbe_mac_82598EB:
1965 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1966 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
1967 break;
1968 case ixgbe_mac_82599EB:
1969 case ixgbe_mac_X540:
1970 mask = (qmask & 0xFFFFFFFF);
1971 if (mask)
1972 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
1973 mask = (qmask >> 32);
1974 if (mask)
1975 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
1976 break;
1977 default:
1978 break;
1980 /* skip the flush */
1983 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1985 struct ixgbe_q_vector *q_vector = data;
1987 if (!q_vector->tx.count)
1988 return IRQ_HANDLED;
1990 /* EIAM disabled interrupts (on this vector) for us */
1991 napi_schedule(&q_vector->napi);
1993 return IRQ_HANDLED;
1997 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1998 * @irq: unused
1999 * @data: pointer to our q_vector struct for this interrupt vector
2001 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
2003 struct ixgbe_q_vector *q_vector = data;
2005 if (!q_vector->rx.count)
2006 return IRQ_HANDLED;
2008 /* EIAM disabled interrupts (on this vector) for us */
2009 napi_schedule(&q_vector->napi);
2011 return IRQ_HANDLED;
2014 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
2016 struct ixgbe_q_vector *q_vector = data;
2018 if (!q_vector->tx.count && !q_vector->rx.count)
2019 return IRQ_HANDLED;
2021 /* EIAM disabled interrupts (on this vector) for us */
2022 napi_schedule(&q_vector->napi);
2024 return IRQ_HANDLED;
2028 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
2029 * @napi: napi struct with our devices info in it
2030 * @budget: amount of work driver is allowed to do this pass, in packets
2032 * This function is optimized for cleaning one queue only on a single
2033 * q_vector!!!
2035 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
2037 struct ixgbe_q_vector *q_vector =
2038 container_of(napi, struct ixgbe_q_vector, napi);
2039 struct ixgbe_adapter *adapter = q_vector->adapter;
2040 int work_done = 0;
2042 #ifdef CONFIG_IXGBE_DCA
2043 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2044 ixgbe_update_dca(q_vector);
2045 #endif
2047 ixgbe_clean_rx_irq(q_vector, q_vector->rx.ring, &work_done, budget);
2049 /* If all Rx work done, exit the polling mode */
2050 if (work_done < budget) {
2051 napi_complete(napi);
2052 if (adapter->rx_itr_setting & 1)
2053 ixgbe_set_itr(q_vector);
2054 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2055 ixgbe_irq_enable_queues(adapter,
2056 ((u64)1 << q_vector->v_idx));
2059 return work_done;
2063 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
2064 * @napi: napi struct with our devices info in it
2065 * @budget: amount of work driver is allowed to do this pass, in packets
2067 * This function will clean more than one rx queue associated with a
2068 * q_vector.
2070 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
2072 struct ixgbe_q_vector *q_vector =
2073 container_of(napi, struct ixgbe_q_vector, napi);
2074 struct ixgbe_adapter *adapter = q_vector->adapter;
2075 struct ixgbe_ring *ring;
2076 int work_done = 0;
2077 bool clean_complete = true;
2079 #ifdef CONFIG_IXGBE_DCA
2080 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2081 ixgbe_update_dca(q_vector);
2082 #endif
2084 for (ring = q_vector->tx.ring; ring != NULL; ring = ring->next)
2085 clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
2087 /* attempt to distribute budget to each queue fairly, but don't allow
2088 * the budget to go below 1 because we'll exit polling */
2089 budget /= (q_vector->rx.count ?: 1);
2090 budget = max(budget, 1);
2092 for (ring = q_vector->rx.ring; ring != NULL; ring = ring->next)
2093 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
2095 if (!clean_complete)
2096 work_done = budget;
2098 /* If all Rx work done, exit the polling mode */
2099 if (work_done < budget) {
2100 napi_complete(napi);
2101 if (adapter->rx_itr_setting & 1)
2102 ixgbe_set_itr(q_vector);
2103 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2104 ixgbe_irq_enable_queues(adapter,
2105 ((u64)1 << q_vector->v_idx));
2106 return 0;
2109 return work_done;
2113 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2114 * @napi: napi struct with our devices info in it
2115 * @budget: amount of work driver is allowed to do this pass, in packets
2117 * This function is optimized for cleaning one queue only on a single
2118 * q_vector!!!
2120 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2122 struct ixgbe_q_vector *q_vector =
2123 container_of(napi, struct ixgbe_q_vector, napi);
2124 struct ixgbe_adapter *adapter = q_vector->adapter;
2126 #ifdef CONFIG_IXGBE_DCA
2127 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2128 ixgbe_update_dca(q_vector);
2129 #endif
2131 if (!ixgbe_clean_tx_irq(q_vector, q_vector->tx.ring))
2132 return budget;
2134 /* If all Tx work done, exit the polling mode */
2135 napi_complete(napi);
2136 if (adapter->tx_itr_setting & 1)
2137 ixgbe_set_itr(q_vector);
2138 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2139 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2141 return 0;
2144 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2145 int r_idx)
2147 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2148 struct ixgbe_ring *rx_ring = a->rx_ring[r_idx];
2150 rx_ring->q_vector = q_vector;
2151 rx_ring->next = q_vector->rx.ring;
2152 q_vector->rx.ring = rx_ring;
2153 q_vector->rx.count++;
2156 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2157 int t_idx)
2159 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2160 struct ixgbe_ring *tx_ring = a->tx_ring[t_idx];
2162 tx_ring->q_vector = q_vector;
2163 tx_ring->next = q_vector->tx.ring;
2164 q_vector->tx.ring = tx_ring;
2165 q_vector->tx.count++;
2166 q_vector->tx.work_limit = a->tx_work_limit;
2170 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2171 * @adapter: board private structure to initialize
2173 * This function maps descriptor rings to the queue-specific vectors
2174 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2175 * one vector per ring/queue, but on a constrained vector budget, we
2176 * group the rings as "efficiently" as possible. You would add new
2177 * mapping configurations in here.
2179 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2181 int q_vectors;
2182 int v_start = 0;
2183 int rxr_idx = 0, txr_idx = 0;
2184 int rxr_remaining = adapter->num_rx_queues;
2185 int txr_remaining = adapter->num_tx_queues;
2186 int i, j;
2187 int rqpv, tqpv;
2188 int err = 0;
2190 /* No mapping required if MSI-X is disabled. */
2191 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2192 goto out;
2194 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2197 * The ideal configuration...
2198 * We have enough vectors to map one per queue.
2200 if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2201 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2202 map_vector_to_rxq(adapter, v_start, rxr_idx);
2204 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2205 map_vector_to_txq(adapter, v_start, txr_idx);
2207 goto out;
2211 * If we don't have enough vectors for a 1-to-1
2212 * mapping, we'll have to group them so there are
2213 * multiple queues per vector.
2215 /* Re-adjusting *qpv takes care of the remainder. */
2216 for (i = v_start; i < q_vectors; i++) {
2217 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
2218 for (j = 0; j < rqpv; j++) {
2219 map_vector_to_rxq(adapter, i, rxr_idx);
2220 rxr_idx++;
2221 rxr_remaining--;
2223 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
2224 for (j = 0; j < tqpv; j++) {
2225 map_vector_to_txq(adapter, i, txr_idx);
2226 txr_idx++;
2227 txr_remaining--;
2230 out:
2231 return err;
2235 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2236 * @adapter: board private structure
2238 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2239 * interrupts from the kernel.
2241 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2243 struct net_device *netdev = adapter->netdev;
2244 irqreturn_t (*handler)(int, void *);
2245 int i, vector, q_vectors, err;
2246 int ri = 0, ti = 0;
2248 /* Decrement for Other and TCP Timer vectors */
2249 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2251 err = ixgbe_map_rings_to_vectors(adapter);
2252 if (err)
2253 return err;
2255 #define SET_HANDLER(_v) (((_v)->rx.count && (_v)->tx.count) \
2256 ? &ixgbe_msix_clean_many : \
2257 (_v)->rx.count ? &ixgbe_msix_clean_rx : \
2258 (_v)->tx.count ? &ixgbe_msix_clean_tx : \
2259 NULL)
2260 for (vector = 0; vector < q_vectors; vector++) {
2261 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2262 handler = SET_HANDLER(q_vector);
2264 if (handler == &ixgbe_msix_clean_rx) {
2265 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2266 "%s-%s-%d", netdev->name, "rx", ri++);
2267 } else if (handler == &ixgbe_msix_clean_tx) {
2268 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2269 "%s-%s-%d", netdev->name, "tx", ti++);
2270 } else if (handler == &ixgbe_msix_clean_many) {
2271 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2272 "%s-%s-%d", netdev->name, "TxRx", ri++);
2273 ti++;
2274 } else {
2275 /* skip this unused q_vector */
2276 continue;
2278 err = request_irq(adapter->msix_entries[vector].vector,
2279 handler, 0, q_vector->name,
2280 q_vector);
2281 if (err) {
2282 e_err(probe, "request_irq failed for MSIX interrupt "
2283 "Error: %d\n", err);
2284 goto free_queue_irqs;
2288 sprintf(adapter->lsc_int_name, "%s:lsc", netdev->name);
2289 err = request_irq(adapter->msix_entries[vector].vector,
2290 ixgbe_msix_lsc, 0, adapter->lsc_int_name, adapter);
2291 if (err) {
2292 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2293 goto free_queue_irqs;
2296 return 0;
2298 free_queue_irqs:
2299 for (i = vector - 1; i >= 0; i--)
2300 free_irq(adapter->msix_entries[--vector].vector,
2301 adapter->q_vector[i]);
2302 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2303 pci_disable_msix(adapter->pdev);
2304 kfree(adapter->msix_entries);
2305 adapter->msix_entries = NULL;
2306 return err;
2310 * ixgbe_irq_enable - Enable default interrupt generation settings
2311 * @adapter: board private structure
2313 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2314 bool flush)
2316 u32 mask;
2318 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2319 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2320 mask |= IXGBE_EIMS_GPI_SDP0;
2321 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2322 mask |= IXGBE_EIMS_GPI_SDP1;
2323 switch (adapter->hw.mac.type) {
2324 case ixgbe_mac_82599EB:
2325 case ixgbe_mac_X540:
2326 mask |= IXGBE_EIMS_ECC;
2327 mask |= IXGBE_EIMS_GPI_SDP1;
2328 mask |= IXGBE_EIMS_GPI_SDP2;
2329 if (adapter->num_vfs)
2330 mask |= IXGBE_EIMS_MAILBOX;
2331 break;
2332 default:
2333 break;
2335 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
2336 mask |= IXGBE_EIMS_FLOW_DIR;
2338 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2339 if (queues)
2340 ixgbe_irq_enable_queues(adapter, ~0);
2341 if (flush)
2342 IXGBE_WRITE_FLUSH(&adapter->hw);
2344 if (adapter->num_vfs > 32) {
2345 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2346 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2351 * ixgbe_intr - legacy mode Interrupt Handler
2352 * @irq: interrupt number
2353 * @data: pointer to a network interface device structure
2355 static irqreturn_t ixgbe_intr(int irq, void *data)
2357 struct ixgbe_adapter *adapter = data;
2358 struct ixgbe_hw *hw = &adapter->hw;
2359 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2360 u32 eicr;
2363 * Workaround for silicon errata on 82598. Mask the interrupts
2364 * before the read of EICR.
2366 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2368 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2369 * therefore no explict interrupt disable is necessary */
2370 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2371 if (!eicr) {
2373 * shared interrupt alert!
2374 * make sure interrupts are enabled because the read will
2375 * have disabled interrupts due to EIAM
2376 * finish the workaround of silicon errata on 82598. Unmask
2377 * the interrupt that we masked before the EICR read.
2379 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2380 ixgbe_irq_enable(adapter, true, true);
2381 return IRQ_NONE; /* Not our interrupt */
2384 if (eicr & IXGBE_EICR_LSC)
2385 ixgbe_check_lsc(adapter);
2387 switch (hw->mac.type) {
2388 case ixgbe_mac_82599EB:
2389 ixgbe_check_sfp_event(adapter, eicr);
2390 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2391 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
2392 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2393 adapter->interrupt_event = eicr;
2394 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2395 ixgbe_service_event_schedule(adapter);
2398 break;
2399 default:
2400 break;
2403 ixgbe_check_fan_failure(adapter, eicr);
2405 if (napi_schedule_prep(&(q_vector->napi))) {
2406 /* would disable interrupts here but EIAM disabled it */
2407 __napi_schedule(&(q_vector->napi));
2411 * re-enable link(maybe) and non-queue interrupts, no flush.
2412 * ixgbe_poll will re-enable the queue interrupts
2415 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2416 ixgbe_irq_enable(adapter, false, false);
2418 return IRQ_HANDLED;
2421 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2423 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2424 int i;
2426 /* legacy and MSI only use one vector */
2427 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2428 q_vectors = 1;
2430 for (i = 0; i < adapter->num_rx_queues; i++) {
2431 adapter->rx_ring[i]->q_vector = NULL;
2432 adapter->rx_ring[i]->next = NULL;
2434 for (i = 0; i < adapter->num_tx_queues; i++) {
2435 adapter->tx_ring[i]->q_vector = NULL;
2436 adapter->tx_ring[i]->next = NULL;
2439 for (i = 0; i < q_vectors; i++) {
2440 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2441 memset(&q_vector->rx, 0, sizeof(struct ixgbe_ring_container));
2442 memset(&q_vector->tx, 0, sizeof(struct ixgbe_ring_container));
2447 * ixgbe_request_irq - initialize interrupts
2448 * @adapter: board private structure
2450 * Attempts to configure interrupts using the best available
2451 * capabilities of the hardware and kernel.
2453 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2455 struct net_device *netdev = adapter->netdev;
2456 int err;
2458 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2459 err = ixgbe_request_msix_irqs(adapter);
2460 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2461 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2462 netdev->name, adapter);
2463 } else {
2464 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2465 netdev->name, adapter);
2468 if (err)
2469 e_err(probe, "request_irq failed, Error %d\n", err);
2471 return err;
2474 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2476 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2477 int i, q_vectors;
2479 q_vectors = adapter->num_msix_vectors;
2481 i = q_vectors - 1;
2482 free_irq(adapter->msix_entries[i].vector, adapter);
2484 i--;
2485 for (; i >= 0; i--) {
2486 /* free only the irqs that were actually requested */
2487 if (!adapter->q_vector[i]->rx.count &&
2488 !adapter->q_vector[i]->tx.count)
2489 continue;
2491 free_irq(adapter->msix_entries[i].vector,
2492 adapter->q_vector[i]);
2495 ixgbe_reset_q_vectors(adapter);
2496 } else {
2497 free_irq(adapter->pdev->irq, adapter);
2502 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2503 * @adapter: board private structure
2505 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2507 switch (adapter->hw.mac.type) {
2508 case ixgbe_mac_82598EB:
2509 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2510 break;
2511 case ixgbe_mac_82599EB:
2512 case ixgbe_mac_X540:
2513 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2514 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2515 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2516 if (adapter->num_vfs > 32)
2517 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2518 break;
2519 default:
2520 break;
2522 IXGBE_WRITE_FLUSH(&adapter->hw);
2523 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2524 int i;
2525 for (i = 0; i < adapter->num_msix_vectors; i++)
2526 synchronize_irq(adapter->msix_entries[i].vector);
2527 } else {
2528 synchronize_irq(adapter->pdev->irq);
2533 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2536 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2538 struct ixgbe_hw *hw = &adapter->hw;
2540 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2541 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2543 ixgbe_set_ivar(adapter, 0, 0, 0);
2544 ixgbe_set_ivar(adapter, 1, 0, 0);
2546 map_vector_to_rxq(adapter, 0, 0);
2547 map_vector_to_txq(adapter, 0, 0);
2549 e_info(hw, "Legacy interrupt IVAR setup done\n");
2553 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2554 * @adapter: board private structure
2555 * @ring: structure containing ring specific data
2557 * Configure the Tx descriptor ring after a reset.
2559 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2560 struct ixgbe_ring *ring)
2562 struct ixgbe_hw *hw = &adapter->hw;
2563 u64 tdba = ring->dma;
2564 int wait_loop = 10;
2565 u32 txdctl;
2566 u8 reg_idx = ring->reg_idx;
2568 /* disable queue to avoid issues while updating state */
2569 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2570 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2571 txdctl & ~IXGBE_TXDCTL_ENABLE);
2572 IXGBE_WRITE_FLUSH(hw);
2574 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2575 (tdba & DMA_BIT_MASK(32)));
2576 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2577 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2578 ring->count * sizeof(union ixgbe_adv_tx_desc));
2579 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2580 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2581 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2583 /* configure fetching thresholds */
2584 if (adapter->rx_itr_setting == 0) {
2585 /* cannot set wthresh when itr==0 */
2586 txdctl &= ~0x007F0000;
2587 } else {
2588 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2589 txdctl |= (8 << 16);
2591 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2592 /* PThresh workaround for Tx hang with DFP enabled. */
2593 txdctl |= 32;
2596 /* reinitialize flowdirector state */
2597 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2598 adapter->atr_sample_rate) {
2599 ring->atr_sample_rate = adapter->atr_sample_rate;
2600 ring->atr_count = 0;
2601 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2602 } else {
2603 ring->atr_sample_rate = 0;
2606 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2608 /* enable queue */
2609 txdctl |= IXGBE_TXDCTL_ENABLE;
2610 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2612 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2613 if (hw->mac.type == ixgbe_mac_82598EB &&
2614 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2615 return;
2617 /* poll to verify queue is enabled */
2618 do {
2619 usleep_range(1000, 2000);
2620 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2621 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2622 if (!wait_loop)
2623 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2626 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2628 struct ixgbe_hw *hw = &adapter->hw;
2629 u32 rttdcs;
2630 u32 reg;
2631 u8 tcs = netdev_get_num_tc(adapter->netdev);
2633 if (hw->mac.type == ixgbe_mac_82598EB)
2634 return;
2636 /* disable the arbiter while setting MTQC */
2637 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2638 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2639 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2641 /* set transmit pool layout */
2642 switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2643 case (IXGBE_FLAG_SRIOV_ENABLED):
2644 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2645 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2646 break;
2647 default:
2648 if (!tcs)
2649 reg = IXGBE_MTQC_64Q_1PB;
2650 else if (tcs <= 4)
2651 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2652 else
2653 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2655 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2657 /* Enable Security TX Buffer IFG for multiple pb */
2658 if (tcs) {
2659 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2660 reg |= IXGBE_SECTX_DCB;
2661 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2663 break;
2666 /* re-enable the arbiter */
2667 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2668 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2672 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2673 * @adapter: board private structure
2675 * Configure the Tx unit of the MAC after a reset.
2677 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2679 struct ixgbe_hw *hw = &adapter->hw;
2680 u32 dmatxctl;
2681 u32 i;
2683 ixgbe_setup_mtqc(adapter);
2685 if (hw->mac.type != ixgbe_mac_82598EB) {
2686 /* DMATXCTL.EN must be before Tx queues are enabled */
2687 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2688 dmatxctl |= IXGBE_DMATXCTL_TE;
2689 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2692 /* Setup the HW Tx Head and Tail descriptor pointers */
2693 for (i = 0; i < adapter->num_tx_queues; i++)
2694 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2697 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2699 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2700 struct ixgbe_ring *rx_ring)
2702 u32 srrctl;
2703 u8 reg_idx = rx_ring->reg_idx;
2705 switch (adapter->hw.mac.type) {
2706 case ixgbe_mac_82598EB: {
2707 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2708 const int mask = feature[RING_F_RSS].mask;
2709 reg_idx = reg_idx & mask;
2711 break;
2712 case ixgbe_mac_82599EB:
2713 case ixgbe_mac_X540:
2714 default:
2715 break;
2718 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2720 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2721 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2722 if (adapter->num_vfs)
2723 srrctl |= IXGBE_SRRCTL_DROP_EN;
2725 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2726 IXGBE_SRRCTL_BSIZEHDR_MASK;
2728 if (ring_is_ps_enabled(rx_ring)) {
2729 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2730 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2731 #else
2732 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2733 #endif
2734 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2735 } else {
2736 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2737 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2738 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2741 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2744 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2746 struct ixgbe_hw *hw = &adapter->hw;
2747 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2748 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2749 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2750 u32 mrqc = 0, reta = 0;
2751 u32 rxcsum;
2752 int i, j;
2753 u8 tcs = netdev_get_num_tc(adapter->netdev);
2754 int maxq = adapter->ring_feature[RING_F_RSS].indices;
2756 if (tcs)
2757 maxq = min(maxq, adapter->num_tx_queues / tcs);
2759 /* Fill out hash function seeds */
2760 for (i = 0; i < 10; i++)
2761 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2763 /* Fill out redirection table */
2764 for (i = 0, j = 0; i < 128; i++, j++) {
2765 if (j == maxq)
2766 j = 0;
2767 /* reta = 4-byte sliding window of
2768 * 0x00..(indices-1)(indices-1)00..etc. */
2769 reta = (reta << 8) | (j * 0x11);
2770 if ((i & 3) == 3)
2771 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2774 /* Disable indicating checksum in descriptor, enables RSS hash */
2775 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2776 rxcsum |= IXGBE_RXCSUM_PCSD;
2777 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2779 if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2780 (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2781 mrqc = IXGBE_MRQC_RSSEN;
2782 } else {
2783 int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2784 | IXGBE_FLAG_SRIOV_ENABLED);
2786 switch (mask) {
2787 case (IXGBE_FLAG_RSS_ENABLED):
2788 if (!tcs)
2789 mrqc = IXGBE_MRQC_RSSEN;
2790 else if (tcs <= 4)
2791 mrqc = IXGBE_MRQC_RTRSS4TCEN;
2792 else
2793 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2794 break;
2795 case (IXGBE_FLAG_SRIOV_ENABLED):
2796 mrqc = IXGBE_MRQC_VMDQEN;
2797 break;
2798 default:
2799 break;
2803 /* Perform hash on these packet types */
2804 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2805 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2806 | IXGBE_MRQC_RSS_FIELD_IPV6
2807 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2809 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2813 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2814 * @adapter: address of board private structure
2815 * @index: index of ring to set
2817 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2818 struct ixgbe_ring *ring)
2820 struct ixgbe_hw *hw = &adapter->hw;
2821 u32 rscctrl;
2822 int rx_buf_len;
2823 u8 reg_idx = ring->reg_idx;
2825 if (!ring_is_rsc_enabled(ring))
2826 return;
2828 rx_buf_len = ring->rx_buf_len;
2829 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2830 rscctrl |= IXGBE_RSCCTL_RSCEN;
2832 * we must limit the number of descriptors so that the
2833 * total size of max desc * buf_len is not greater
2834 * than 65535
2836 if (ring_is_ps_enabled(ring)) {
2837 #if (MAX_SKB_FRAGS > 16)
2838 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2839 #elif (MAX_SKB_FRAGS > 8)
2840 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2841 #elif (MAX_SKB_FRAGS > 4)
2842 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2843 #else
2844 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2845 #endif
2846 } else {
2847 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2848 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2849 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2850 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2851 else
2852 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2854 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2858 * ixgbe_set_uta - Set unicast filter table address
2859 * @adapter: board private structure
2861 * The unicast table address is a register array of 32-bit registers.
2862 * The table is meant to be used in a way similar to how the MTA is used
2863 * however due to certain limitations in the hardware it is necessary to
2864 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2865 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2867 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2869 struct ixgbe_hw *hw = &adapter->hw;
2870 int i;
2872 /* The UTA table only exists on 82599 hardware and newer */
2873 if (hw->mac.type < ixgbe_mac_82599EB)
2874 return;
2876 /* we only need to do this if VMDq is enabled */
2877 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2878 return;
2880 for (i = 0; i < 128; i++)
2881 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2884 #define IXGBE_MAX_RX_DESC_POLL 10
2885 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2886 struct ixgbe_ring *ring)
2888 struct ixgbe_hw *hw = &adapter->hw;
2889 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2890 u32 rxdctl;
2891 u8 reg_idx = ring->reg_idx;
2893 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2894 if (hw->mac.type == ixgbe_mac_82598EB &&
2895 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2896 return;
2898 do {
2899 usleep_range(1000, 2000);
2900 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2901 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2903 if (!wait_loop) {
2904 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2905 "the polling period\n", reg_idx);
2909 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
2910 struct ixgbe_ring *ring)
2912 struct ixgbe_hw *hw = &adapter->hw;
2913 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2914 u32 rxdctl;
2915 u8 reg_idx = ring->reg_idx;
2917 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2918 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
2920 /* write value back with RXDCTL.ENABLE bit cleared */
2921 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2923 if (hw->mac.type == ixgbe_mac_82598EB &&
2924 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2925 return;
2927 /* the hardware may take up to 100us to really disable the rx queue */
2928 do {
2929 udelay(10);
2930 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2931 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
2933 if (!wait_loop) {
2934 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
2935 "the polling period\n", reg_idx);
2939 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2940 struct ixgbe_ring *ring)
2942 struct ixgbe_hw *hw = &adapter->hw;
2943 u64 rdba = ring->dma;
2944 u32 rxdctl;
2945 u8 reg_idx = ring->reg_idx;
2947 /* disable queue to avoid issues while updating state */
2948 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2949 ixgbe_disable_rx_queue(adapter, ring);
2951 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2952 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2953 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2954 ring->count * sizeof(union ixgbe_adv_rx_desc));
2955 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2956 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2957 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2959 ixgbe_configure_srrctl(adapter, ring);
2960 ixgbe_configure_rscctl(adapter, ring);
2962 /* If operating in IOV mode set RLPML for X540 */
2963 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
2964 hw->mac.type == ixgbe_mac_X540) {
2965 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
2966 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
2967 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
2970 if (hw->mac.type == ixgbe_mac_82598EB) {
2972 * enable cache line friendly hardware writes:
2973 * PTHRESH=32 descriptors (half the internal cache),
2974 * this also removes ugly rx_no_buffer_count increment
2975 * HTHRESH=4 descriptors (to minimize latency on fetch)
2976 * WTHRESH=8 burst writeback up to two cache lines
2978 rxdctl &= ~0x3FFFFF;
2979 rxdctl |= 0x080420;
2982 /* enable receive descriptor ring */
2983 rxdctl |= IXGBE_RXDCTL_ENABLE;
2984 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2986 ixgbe_rx_desc_queue_enable(adapter, ring);
2987 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
2990 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2992 struct ixgbe_hw *hw = &adapter->hw;
2993 int p;
2995 /* PSRTYPE must be initialized in non 82598 adapters */
2996 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2997 IXGBE_PSRTYPE_UDPHDR |
2998 IXGBE_PSRTYPE_IPV4HDR |
2999 IXGBE_PSRTYPE_L2HDR |
3000 IXGBE_PSRTYPE_IPV6HDR;
3002 if (hw->mac.type == ixgbe_mac_82598EB)
3003 return;
3005 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3006 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3008 for (p = 0; p < adapter->num_rx_pools; p++)
3009 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3010 psrtype);
3013 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3015 struct ixgbe_hw *hw = &adapter->hw;
3016 u32 gcr_ext;
3017 u32 vt_reg_bits;
3018 u32 reg_offset, vf_shift;
3019 u32 vmdctl;
3021 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3022 return;
3024 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3025 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3026 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3027 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3029 vf_shift = adapter->num_vfs % 32;
3030 reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
3032 /* Enable only the PF's pool for Tx/Rx */
3033 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3034 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3035 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3036 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3037 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3039 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3040 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3043 * Set up VF register offsets for selected VT Mode,
3044 * i.e. 32 or 64 VFs for SR-IOV
3046 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3047 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3048 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3049 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3051 /* enable Tx loopback for VF/PF communication */
3052 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3053 /* Enable MAC Anti-Spoofing */
3054 hw->mac.ops.set_mac_anti_spoofing(hw,
3055 (adapter->antispoofing_enabled =
3056 (adapter->num_vfs != 0)),
3057 adapter->num_vfs);
3060 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3062 struct ixgbe_hw *hw = &adapter->hw;
3063 struct net_device *netdev = adapter->netdev;
3064 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3065 int rx_buf_len;
3066 struct ixgbe_ring *rx_ring;
3067 int i;
3068 u32 mhadd, hlreg0;
3070 /* Decide whether to use packet split mode or not */
3071 /* On by default */
3072 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
3074 /* Do not use packet split if we're in SR-IOV Mode */
3075 if (adapter->num_vfs)
3076 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3078 /* Disable packet split due to 82599 erratum #45 */
3079 if (hw->mac.type == ixgbe_mac_82599EB)
3080 adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3082 /* Set the RX buffer length according to the mode */
3083 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
3084 rx_buf_len = IXGBE_RX_HDR_SIZE;
3085 } else {
3086 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
3087 (netdev->mtu <= ETH_DATA_LEN))
3088 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3089 else
3090 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
3093 #ifdef IXGBE_FCOE
3094 /* adjust max frame to be able to do baby jumbo for FCoE */
3095 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3096 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3097 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3099 #endif /* IXGBE_FCOE */
3100 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3101 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3102 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3103 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3105 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3108 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3109 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3110 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3111 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3114 * Setup the HW Rx Head and Tail Descriptor Pointers and
3115 * the Base and Length of the Rx Descriptor Ring
3117 for (i = 0; i < adapter->num_rx_queues; i++) {
3118 rx_ring = adapter->rx_ring[i];
3119 rx_ring->rx_buf_len = rx_buf_len;
3121 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
3122 set_ring_ps_enabled(rx_ring);
3123 else
3124 clear_ring_ps_enabled(rx_ring);
3126 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3127 set_ring_rsc_enabled(rx_ring);
3128 else
3129 clear_ring_rsc_enabled(rx_ring);
3131 #ifdef IXGBE_FCOE
3132 if (netdev->features & NETIF_F_FCOE_MTU) {
3133 struct ixgbe_ring_feature *f;
3134 f = &adapter->ring_feature[RING_F_FCOE];
3135 if ((i >= f->mask) && (i < f->mask + f->indices)) {
3136 clear_ring_ps_enabled(rx_ring);
3137 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3138 rx_ring->rx_buf_len =
3139 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3140 } else if (!ring_is_rsc_enabled(rx_ring) &&
3141 !ring_is_ps_enabled(rx_ring)) {
3142 rx_ring->rx_buf_len =
3143 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3146 #endif /* IXGBE_FCOE */
3150 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3152 struct ixgbe_hw *hw = &adapter->hw;
3153 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3155 switch (hw->mac.type) {
3156 case ixgbe_mac_82598EB:
3158 * For VMDq support of different descriptor types or
3159 * buffer sizes through the use of multiple SRRCTL
3160 * registers, RDRXCTL.MVMEN must be set to 1
3162 * also, the manual doesn't mention it clearly but DCA hints
3163 * will only use queue 0's tags unless this bit is set. Side
3164 * effects of setting this bit are only that SRRCTL must be
3165 * fully programmed [0..15]
3167 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3168 break;
3169 case ixgbe_mac_82599EB:
3170 case ixgbe_mac_X540:
3171 /* Disable RSC for ACK packets */
3172 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3173 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3174 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3175 /* hardware requires some bits to be set by default */
3176 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3177 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3178 break;
3179 default:
3180 /* We should do nothing since we don't know this hardware */
3181 return;
3184 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3188 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3189 * @adapter: board private structure
3191 * Configure the Rx unit of the MAC after a reset.
3193 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3195 struct ixgbe_hw *hw = &adapter->hw;
3196 int i;
3197 u32 rxctrl;
3199 /* disable receives while setting up the descriptors */
3200 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3201 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3203 ixgbe_setup_psrtype(adapter);
3204 ixgbe_setup_rdrxctl(adapter);
3206 /* Program registers for the distribution of queues */
3207 ixgbe_setup_mrqc(adapter);
3209 ixgbe_set_uta(adapter);
3211 /* set_rx_buffer_len must be called before ring initialization */
3212 ixgbe_set_rx_buffer_len(adapter);
3215 * Setup the HW Rx Head and Tail Descriptor Pointers and
3216 * the Base and Length of the Rx Descriptor Ring
3218 for (i = 0; i < adapter->num_rx_queues; i++)
3219 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3221 /* disable drop enable for 82598 parts */
3222 if (hw->mac.type == ixgbe_mac_82598EB)
3223 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3225 /* enable all receives */
3226 rxctrl |= IXGBE_RXCTRL_RXEN;
3227 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3230 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3232 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3233 struct ixgbe_hw *hw = &adapter->hw;
3234 int pool_ndx = adapter->num_vfs;
3236 /* add VID to filter table */
3237 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3238 set_bit(vid, adapter->active_vlans);
3241 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3243 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3244 struct ixgbe_hw *hw = &adapter->hw;
3245 int pool_ndx = adapter->num_vfs;
3247 /* remove VID from filter table */
3248 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3249 clear_bit(vid, adapter->active_vlans);
3253 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3254 * @adapter: driver data
3256 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3258 struct ixgbe_hw *hw = &adapter->hw;
3259 u32 vlnctrl;
3261 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3262 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3263 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3267 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3268 * @adapter: driver data
3270 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3272 struct ixgbe_hw *hw = &adapter->hw;
3273 u32 vlnctrl;
3275 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3276 vlnctrl |= IXGBE_VLNCTRL_VFE;
3277 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3278 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3282 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3283 * @adapter: driver data
3285 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3287 struct ixgbe_hw *hw = &adapter->hw;
3288 u32 vlnctrl;
3289 int i, j;
3291 switch (hw->mac.type) {
3292 case ixgbe_mac_82598EB:
3293 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3294 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3295 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3296 break;
3297 case ixgbe_mac_82599EB:
3298 case ixgbe_mac_X540:
3299 for (i = 0; i < adapter->num_rx_queues; i++) {
3300 j = adapter->rx_ring[i]->reg_idx;
3301 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3302 vlnctrl &= ~IXGBE_RXDCTL_VME;
3303 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3305 break;
3306 default:
3307 break;
3312 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3313 * @adapter: driver data
3315 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3317 struct ixgbe_hw *hw = &adapter->hw;
3318 u32 vlnctrl;
3319 int i, j;
3321 switch (hw->mac.type) {
3322 case ixgbe_mac_82598EB:
3323 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3324 vlnctrl |= IXGBE_VLNCTRL_VME;
3325 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3326 break;
3327 case ixgbe_mac_82599EB:
3328 case ixgbe_mac_X540:
3329 for (i = 0; i < adapter->num_rx_queues; i++) {
3330 j = adapter->rx_ring[i]->reg_idx;
3331 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3332 vlnctrl |= IXGBE_RXDCTL_VME;
3333 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3335 break;
3336 default:
3337 break;
3341 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3343 u16 vid;
3345 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3347 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3348 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3352 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3353 * @netdev: network interface device structure
3355 * Writes unicast address list to the RAR table.
3356 * Returns: -ENOMEM on failure/insufficient address space
3357 * 0 on no addresses written
3358 * X on writing X addresses to the RAR table
3360 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3362 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3363 struct ixgbe_hw *hw = &adapter->hw;
3364 unsigned int vfn = adapter->num_vfs;
3365 unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3366 int count = 0;
3368 /* return ENOMEM indicating insufficient memory for addresses */
3369 if (netdev_uc_count(netdev) > rar_entries)
3370 return -ENOMEM;
3372 if (!netdev_uc_empty(netdev) && rar_entries) {
3373 struct netdev_hw_addr *ha;
3374 /* return error if we do not support writing to RAR table */
3375 if (!hw->mac.ops.set_rar)
3376 return -ENOMEM;
3378 netdev_for_each_uc_addr(ha, netdev) {
3379 if (!rar_entries)
3380 break;
3381 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3382 vfn, IXGBE_RAH_AV);
3383 count++;
3386 /* write the addresses in reverse order to avoid write combining */
3387 for (; rar_entries > 0 ; rar_entries--)
3388 hw->mac.ops.clear_rar(hw, rar_entries);
3390 return count;
3394 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3395 * @netdev: network interface device structure
3397 * The set_rx_method entry point is called whenever the unicast/multicast
3398 * address list or the network interface flags are updated. This routine is
3399 * responsible for configuring the hardware for proper unicast, multicast and
3400 * promiscuous mode.
3402 void ixgbe_set_rx_mode(struct net_device *netdev)
3404 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3405 struct ixgbe_hw *hw = &adapter->hw;
3406 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3407 int count;
3409 /* Check for Promiscuous and All Multicast modes */
3411 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3413 /* set all bits that we expect to always be set */
3414 fctrl |= IXGBE_FCTRL_BAM;
3415 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3416 fctrl |= IXGBE_FCTRL_PMCF;
3418 /* clear the bits we are changing the status of */
3419 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3421 if (netdev->flags & IFF_PROMISC) {
3422 hw->addr_ctrl.user_set_promisc = true;
3423 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3424 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3425 /* don't hardware filter vlans in promisc mode */
3426 ixgbe_vlan_filter_disable(adapter);
3427 } else {
3428 if (netdev->flags & IFF_ALLMULTI) {
3429 fctrl |= IXGBE_FCTRL_MPE;
3430 vmolr |= IXGBE_VMOLR_MPE;
3431 } else {
3433 * Write addresses to the MTA, if the attempt fails
3434 * then we should just turn on promiscuous mode so
3435 * that we can at least receive multicast traffic
3437 hw->mac.ops.update_mc_addr_list(hw, netdev);
3438 vmolr |= IXGBE_VMOLR_ROMPE;
3440 ixgbe_vlan_filter_enable(adapter);
3441 hw->addr_ctrl.user_set_promisc = false;
3443 * Write addresses to available RAR registers, if there is not
3444 * sufficient space to store all the addresses then enable
3445 * unicast promiscuous mode
3447 count = ixgbe_write_uc_addr_list(netdev);
3448 if (count < 0) {
3449 fctrl |= IXGBE_FCTRL_UPE;
3450 vmolr |= IXGBE_VMOLR_ROPE;
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);
3462 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3464 if (netdev->features & NETIF_F_HW_VLAN_RX)
3465 ixgbe_vlan_strip_enable(adapter);
3466 else
3467 ixgbe_vlan_strip_disable(adapter);
3470 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3472 int q_idx;
3473 struct ixgbe_q_vector *q_vector;
3474 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3476 /* legacy and MSI only use one vector */
3477 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3478 q_vectors = 1;
3480 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3481 struct napi_struct *napi;
3482 q_vector = adapter->q_vector[q_idx];
3483 napi = &q_vector->napi;
3484 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3485 if (!q_vector->rx.count || !q_vector->tx.count) {
3486 if (q_vector->tx.count == 1)
3487 napi->poll = &ixgbe_clean_txonly;
3488 else if (q_vector->rx.count == 1)
3489 napi->poll = &ixgbe_clean_rxonly;
3493 napi_enable(napi);
3497 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3499 int q_idx;
3500 struct ixgbe_q_vector *q_vector;
3501 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3503 /* legacy and MSI only use one vector */
3504 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3505 q_vectors = 1;
3507 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3508 q_vector = adapter->q_vector[q_idx];
3509 napi_disable(&q_vector->napi);
3513 #ifdef CONFIG_IXGBE_DCB
3515 * ixgbe_configure_dcb - Configure DCB hardware
3516 * @adapter: ixgbe adapter struct
3518 * This is called by the driver on open to configure the DCB hardware.
3519 * This is also called by the gennetlink interface when reconfiguring
3520 * the DCB state.
3522 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3524 struct ixgbe_hw *hw = &adapter->hw;
3525 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3527 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3528 if (hw->mac.type == ixgbe_mac_82598EB)
3529 netif_set_gso_max_size(adapter->netdev, 65536);
3530 return;
3533 if (hw->mac.type == ixgbe_mac_82598EB)
3534 netif_set_gso_max_size(adapter->netdev, 32768);
3537 /* Enable VLAN tag insert/strip */
3538 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3540 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3542 /* reconfigure the hardware */
3543 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3544 #ifdef IXGBE_FCOE
3545 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3546 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3547 #endif
3548 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3549 DCB_TX_CONFIG);
3550 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3551 DCB_RX_CONFIG);
3552 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3553 } else {
3554 struct net_device *dev = adapter->netdev;
3556 if (adapter->ixgbe_ieee_ets)
3557 dev->dcbnl_ops->ieee_setets(dev,
3558 adapter->ixgbe_ieee_ets);
3559 if (adapter->ixgbe_ieee_pfc)
3560 dev->dcbnl_ops->ieee_setpfc(dev,
3561 adapter->ixgbe_ieee_pfc);
3564 /* Enable RSS Hash per TC */
3565 if (hw->mac.type != ixgbe_mac_82598EB) {
3566 int i;
3567 u32 reg = 0;
3569 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3570 u8 msb = 0;
3571 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3573 while (cnt >>= 1)
3574 msb++;
3576 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3578 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3582 #endif
3584 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3586 int hdrm = 0;
3587 int num_tc = netdev_get_num_tc(adapter->netdev);
3588 struct ixgbe_hw *hw = &adapter->hw;
3590 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3591 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3592 hdrm = 64 << adapter->fdir_pballoc;
3594 hw->mac.ops.set_rxpba(&adapter->hw, num_tc, hdrm, PBA_STRATEGY_EQUAL);
3597 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3599 struct ixgbe_hw *hw = &adapter->hw;
3600 struct hlist_node *node, *node2;
3601 struct ixgbe_fdir_filter *filter;
3603 spin_lock(&adapter->fdir_perfect_lock);
3605 if (!hlist_empty(&adapter->fdir_filter_list))
3606 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3608 hlist_for_each_entry_safe(filter, node, node2,
3609 &adapter->fdir_filter_list, fdir_node) {
3610 ixgbe_fdir_write_perfect_filter_82599(hw,
3611 &filter->filter,
3612 filter->sw_idx,
3613 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3614 IXGBE_FDIR_DROP_QUEUE :
3615 adapter->rx_ring[filter->action]->reg_idx);
3618 spin_unlock(&adapter->fdir_perfect_lock);
3621 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3623 struct net_device *netdev = adapter->netdev;
3624 struct ixgbe_hw *hw = &adapter->hw;
3625 int i;
3627 ixgbe_configure_pb(adapter);
3628 #ifdef CONFIG_IXGBE_DCB
3629 ixgbe_configure_dcb(adapter);
3630 #endif
3632 ixgbe_set_rx_mode(netdev);
3633 ixgbe_restore_vlan(adapter);
3635 #ifdef IXGBE_FCOE
3636 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3637 ixgbe_configure_fcoe(adapter);
3639 #endif /* IXGBE_FCOE */
3640 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3641 for (i = 0; i < adapter->num_tx_queues; i++)
3642 adapter->tx_ring[i]->atr_sample_rate =
3643 adapter->atr_sample_rate;
3644 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3645 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3646 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3647 adapter->fdir_pballoc);
3648 ixgbe_fdir_filter_restore(adapter);
3650 ixgbe_configure_virtualization(adapter);
3652 ixgbe_configure_tx(adapter);
3653 ixgbe_configure_rx(adapter);
3656 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3658 switch (hw->phy.type) {
3659 case ixgbe_phy_sfp_avago:
3660 case ixgbe_phy_sfp_ftl:
3661 case ixgbe_phy_sfp_intel:
3662 case ixgbe_phy_sfp_unknown:
3663 case ixgbe_phy_sfp_passive_tyco:
3664 case ixgbe_phy_sfp_passive_unknown:
3665 case ixgbe_phy_sfp_active_unknown:
3666 case ixgbe_phy_sfp_ftl_active:
3667 return true;
3668 default:
3669 return false;
3674 * ixgbe_sfp_link_config - set up SFP+ link
3675 * @adapter: pointer to private adapter struct
3677 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3680 * We are assuming the worst case scenerio here, and that
3681 * is that an SFP was inserted/removed after the reset
3682 * but before SFP detection was enabled. As such the best
3683 * solution is to just start searching as soon as we start
3685 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3686 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3688 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3692 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3693 * @hw: pointer to private hardware struct
3695 * Returns 0 on success, negative on failure
3697 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3699 u32 autoneg;
3700 bool negotiation, link_up = false;
3701 u32 ret = IXGBE_ERR_LINK_SETUP;
3703 if (hw->mac.ops.check_link)
3704 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3706 if (ret)
3707 goto link_cfg_out;
3709 autoneg = hw->phy.autoneg_advertised;
3710 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3711 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3712 &negotiation);
3713 if (ret)
3714 goto link_cfg_out;
3716 if (hw->mac.ops.setup_link)
3717 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3718 link_cfg_out:
3719 return ret;
3722 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3724 struct ixgbe_hw *hw = &adapter->hw;
3725 u32 gpie = 0;
3727 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3728 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3729 IXGBE_GPIE_OCD;
3730 gpie |= IXGBE_GPIE_EIAME;
3732 * use EIAM to auto-mask when MSI-X interrupt is asserted
3733 * this saves a register write for every interrupt
3735 switch (hw->mac.type) {
3736 case ixgbe_mac_82598EB:
3737 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3738 break;
3739 case ixgbe_mac_82599EB:
3740 case ixgbe_mac_X540:
3741 default:
3742 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3743 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3744 break;
3746 } else {
3747 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3748 * specifically only auto mask tx and rx interrupts */
3749 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3752 /* XXX: to interrupt immediately for EICS writes, enable this */
3753 /* gpie |= IXGBE_GPIE_EIMEN; */
3755 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3756 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3757 gpie |= IXGBE_GPIE_VTMODE_64;
3760 /* Enable fan failure interrupt */
3761 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3762 gpie |= IXGBE_SDP1_GPIEN;
3764 if (hw->mac.type == ixgbe_mac_82599EB) {
3765 gpie |= IXGBE_SDP1_GPIEN;
3766 gpie |= IXGBE_SDP2_GPIEN;
3769 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3772 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3774 struct ixgbe_hw *hw = &adapter->hw;
3775 int err;
3776 u32 ctrl_ext;
3778 ixgbe_get_hw_control(adapter);
3779 ixgbe_setup_gpie(adapter);
3781 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3782 ixgbe_configure_msix(adapter);
3783 else
3784 ixgbe_configure_msi_and_legacy(adapter);
3786 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3787 if (hw->mac.ops.enable_tx_laser &&
3788 ((hw->phy.multispeed_fiber) ||
3789 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3790 (hw->mac.type == ixgbe_mac_82599EB))))
3791 hw->mac.ops.enable_tx_laser(hw);
3793 clear_bit(__IXGBE_DOWN, &adapter->state);
3794 ixgbe_napi_enable_all(adapter);
3796 if (ixgbe_is_sfp(hw)) {
3797 ixgbe_sfp_link_config(adapter);
3798 } else {
3799 err = ixgbe_non_sfp_link_config(hw);
3800 if (err)
3801 e_err(probe, "link_config FAILED %d\n", err);
3804 /* clear any pending interrupts, may auto mask */
3805 IXGBE_READ_REG(hw, IXGBE_EICR);
3806 ixgbe_irq_enable(adapter, true, true);
3809 * If this adapter has a fan, check to see if we had a failure
3810 * before we enabled the interrupt.
3812 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3813 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3814 if (esdp & IXGBE_ESDP_SDP1)
3815 e_crit(drv, "Fan has stopped, replace the adapter\n");
3818 /* enable transmits */
3819 netif_tx_start_all_queues(adapter->netdev);
3821 /* bring the link up in the watchdog, this could race with our first
3822 * link up interrupt but shouldn't be a problem */
3823 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3824 adapter->link_check_timeout = jiffies;
3825 mod_timer(&adapter->service_timer, jiffies);
3827 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3828 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3829 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3830 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3832 return 0;
3835 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3837 WARN_ON(in_interrupt());
3838 /* put off any impending NetWatchDogTimeout */
3839 adapter->netdev->trans_start = jiffies;
3841 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3842 usleep_range(1000, 2000);
3843 ixgbe_down(adapter);
3845 * If SR-IOV enabled then wait a bit before bringing the adapter
3846 * back up to give the VFs time to respond to the reset. The
3847 * two second wait is based upon the watchdog timer cycle in
3848 * the VF driver.
3850 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3851 msleep(2000);
3852 ixgbe_up(adapter);
3853 clear_bit(__IXGBE_RESETTING, &adapter->state);
3856 int ixgbe_up(struct ixgbe_adapter *adapter)
3858 /* hardware has been reset, we need to reload some things */
3859 ixgbe_configure(adapter);
3861 return ixgbe_up_complete(adapter);
3864 void ixgbe_reset(struct ixgbe_adapter *adapter)
3866 struct ixgbe_hw *hw = &adapter->hw;
3867 int err;
3869 /* lock SFP init bit to prevent race conditions with the watchdog */
3870 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
3871 usleep_range(1000, 2000);
3873 /* clear all SFP and link config related flags while holding SFP_INIT */
3874 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
3875 IXGBE_FLAG2_SFP_NEEDS_RESET);
3876 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
3878 err = hw->mac.ops.init_hw(hw);
3879 switch (err) {
3880 case 0:
3881 case IXGBE_ERR_SFP_NOT_PRESENT:
3882 case IXGBE_ERR_SFP_NOT_SUPPORTED:
3883 break;
3884 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3885 e_dev_err("master disable timed out\n");
3886 break;
3887 case IXGBE_ERR_EEPROM_VERSION:
3888 /* We are running on a pre-production device, log a warning */
3889 e_dev_warn("This device is a pre-production adapter/LOM. "
3890 "Please be aware there may be issuesassociated with "
3891 "your hardware. If you are experiencing problems "
3892 "please contact your Intel or hardware "
3893 "representative who provided you with this "
3894 "hardware.\n");
3895 break;
3896 default:
3897 e_dev_err("Hardware Error: %d\n", err);
3900 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
3902 /* reprogram the RAR[0] in case user changed it. */
3903 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3904 IXGBE_RAH_AV);
3908 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3909 * @rx_ring: ring to free buffers from
3911 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3913 struct device *dev = rx_ring->dev;
3914 unsigned long size;
3915 u16 i;
3917 /* ring already cleared, nothing to do */
3918 if (!rx_ring->rx_buffer_info)
3919 return;
3921 /* Free all the Rx ring sk_buffs */
3922 for (i = 0; i < rx_ring->count; i++) {
3923 struct ixgbe_rx_buffer *rx_buffer_info;
3925 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3926 if (rx_buffer_info->dma) {
3927 dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3928 rx_ring->rx_buf_len,
3929 DMA_FROM_DEVICE);
3930 rx_buffer_info->dma = 0;
3932 if (rx_buffer_info->skb) {
3933 struct sk_buff *skb = rx_buffer_info->skb;
3934 rx_buffer_info->skb = NULL;
3935 do {
3936 struct sk_buff *this = skb;
3937 if (IXGBE_RSC_CB(this)->delay_unmap) {
3938 dma_unmap_single(dev,
3939 IXGBE_RSC_CB(this)->dma,
3940 rx_ring->rx_buf_len,
3941 DMA_FROM_DEVICE);
3942 IXGBE_RSC_CB(this)->dma = 0;
3943 IXGBE_RSC_CB(skb)->delay_unmap = false;
3945 skb = skb->prev;
3946 dev_kfree_skb(this);
3947 } while (skb);
3949 if (!rx_buffer_info->page)
3950 continue;
3951 if (rx_buffer_info->page_dma) {
3952 dma_unmap_page(dev, rx_buffer_info->page_dma,
3953 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3954 rx_buffer_info->page_dma = 0;
3956 put_page(rx_buffer_info->page);
3957 rx_buffer_info->page = NULL;
3958 rx_buffer_info->page_offset = 0;
3961 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3962 memset(rx_ring->rx_buffer_info, 0, size);
3964 /* Zero out the descriptor ring */
3965 memset(rx_ring->desc, 0, rx_ring->size);
3967 rx_ring->next_to_clean = 0;
3968 rx_ring->next_to_use = 0;
3972 * ixgbe_clean_tx_ring - Free Tx Buffers
3973 * @tx_ring: ring to be cleaned
3975 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3977 struct ixgbe_tx_buffer *tx_buffer_info;
3978 unsigned long size;
3979 u16 i;
3981 /* ring already cleared, nothing to do */
3982 if (!tx_ring->tx_buffer_info)
3983 return;
3985 /* Free all the Tx ring sk_buffs */
3986 for (i = 0; i < tx_ring->count; i++) {
3987 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3988 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3991 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3992 memset(tx_ring->tx_buffer_info, 0, size);
3994 /* Zero out the descriptor ring */
3995 memset(tx_ring->desc, 0, tx_ring->size);
3997 tx_ring->next_to_use = 0;
3998 tx_ring->next_to_clean = 0;
4002 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4003 * @adapter: board private structure
4005 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4007 int i;
4009 for (i = 0; i < adapter->num_rx_queues; i++)
4010 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4014 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4015 * @adapter: board private structure
4017 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4019 int i;
4021 for (i = 0; i < adapter->num_tx_queues; i++)
4022 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4025 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4027 struct hlist_node *node, *node2;
4028 struct ixgbe_fdir_filter *filter;
4030 spin_lock(&adapter->fdir_perfect_lock);
4032 hlist_for_each_entry_safe(filter, node, node2,
4033 &adapter->fdir_filter_list, fdir_node) {
4034 hlist_del(&filter->fdir_node);
4035 kfree(filter);
4037 adapter->fdir_filter_count = 0;
4039 spin_unlock(&adapter->fdir_perfect_lock);
4042 void ixgbe_down(struct ixgbe_adapter *adapter)
4044 struct net_device *netdev = adapter->netdev;
4045 struct ixgbe_hw *hw = &adapter->hw;
4046 u32 rxctrl;
4047 int i;
4048 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4050 /* signal that we are down to the interrupt handler */
4051 set_bit(__IXGBE_DOWN, &adapter->state);
4053 /* disable receives */
4054 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4055 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4057 /* disable all enabled rx queues */
4058 for (i = 0; i < adapter->num_rx_queues; i++)
4059 /* this call also flushes the previous write */
4060 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4062 usleep_range(10000, 20000);
4064 netif_tx_stop_all_queues(netdev);
4066 /* call carrier off first to avoid false dev_watchdog timeouts */
4067 netif_carrier_off(netdev);
4068 netif_tx_disable(netdev);
4070 ixgbe_irq_disable(adapter);
4072 ixgbe_napi_disable_all(adapter);
4074 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4075 IXGBE_FLAG2_RESET_REQUESTED);
4076 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4078 del_timer_sync(&adapter->service_timer);
4080 /* disable receive for all VFs and wait one second */
4081 if (adapter->num_vfs) {
4082 /* ping all the active vfs to let them know we are going down */
4083 ixgbe_ping_all_vfs(adapter);
4085 /* Disable all VFTE/VFRE TX/RX */
4086 ixgbe_disable_tx_rx(adapter);
4088 /* Mark all the VFs as inactive */
4089 for (i = 0 ; i < adapter->num_vfs; i++)
4090 adapter->vfinfo[i].clear_to_send = 0;
4093 /* Cleanup the affinity_hint CPU mask memory and callback */
4094 for (i = 0; i < num_q_vectors; i++) {
4095 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
4096 /* clear the affinity_mask in the IRQ descriptor */
4097 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
4098 /* release the CPU mask memory */
4099 free_cpumask_var(q_vector->affinity_mask);
4102 /* disable transmits in the hardware now that interrupts are off */
4103 for (i = 0; i < adapter->num_tx_queues; i++) {
4104 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4105 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4108 /* Disable the Tx DMA engine on 82599 and X540 */
4109 switch (hw->mac.type) {
4110 case ixgbe_mac_82599EB:
4111 case ixgbe_mac_X540:
4112 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4113 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4114 ~IXGBE_DMATXCTL_TE));
4115 break;
4116 default:
4117 break;
4120 if (!pci_channel_offline(adapter->pdev))
4121 ixgbe_reset(adapter);
4123 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4124 if (hw->mac.ops.disable_tx_laser &&
4125 ((hw->phy.multispeed_fiber) ||
4126 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4127 (hw->mac.type == ixgbe_mac_82599EB))))
4128 hw->mac.ops.disable_tx_laser(hw);
4130 ixgbe_clean_all_tx_rings(adapter);
4131 ixgbe_clean_all_rx_rings(adapter);
4133 #ifdef CONFIG_IXGBE_DCA
4134 /* since we reset the hardware DCA settings were cleared */
4135 ixgbe_setup_dca(adapter);
4136 #endif
4140 * ixgbe_poll - NAPI Rx polling callback
4141 * @napi: structure for representing this polling device
4142 * @budget: how many packets driver is allowed to clean
4144 * This function is used for legacy and MSI, NAPI mode
4146 static int ixgbe_poll(struct napi_struct *napi, int budget)
4148 struct ixgbe_q_vector *q_vector =
4149 container_of(napi, struct ixgbe_q_vector, napi);
4150 struct ixgbe_adapter *adapter = q_vector->adapter;
4151 int tx_clean_complete, work_done = 0;
4153 #ifdef CONFIG_IXGBE_DCA
4154 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
4155 ixgbe_update_dca(q_vector);
4156 #endif
4158 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
4159 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
4161 if (!tx_clean_complete)
4162 work_done = budget;
4164 /* If budget not fully consumed, exit the polling mode */
4165 if (work_done < budget) {
4166 napi_complete(napi);
4167 if (adapter->rx_itr_setting & 1)
4168 ixgbe_set_itr(q_vector);
4169 if (!test_bit(__IXGBE_DOWN, &adapter->state))
4170 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
4172 return work_done;
4176 * ixgbe_tx_timeout - Respond to a Tx Hang
4177 * @netdev: network interface device structure
4179 static void ixgbe_tx_timeout(struct net_device *netdev)
4181 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4183 /* Do the reset outside of interrupt context */
4184 ixgbe_tx_timeout_reset(adapter);
4188 * ixgbe_set_rss_queues: Allocate queues for RSS
4189 * @adapter: board private structure to initialize
4191 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4192 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4195 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4197 bool ret = false;
4198 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4200 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4201 f->mask = 0xF;
4202 adapter->num_rx_queues = f->indices;
4203 adapter->num_tx_queues = f->indices;
4204 ret = true;
4205 } else {
4206 ret = false;
4209 return ret;
4213 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4214 * @adapter: board private structure to initialize
4216 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4217 * to the original CPU that initiated the Tx session. This runs in addition
4218 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4219 * Rx load across CPUs using RSS.
4222 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4224 bool ret = false;
4225 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4227 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4228 f_fdir->mask = 0;
4230 /* Flow Director must have RSS enabled */
4231 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4232 (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4233 adapter->num_tx_queues = f_fdir->indices;
4234 adapter->num_rx_queues = f_fdir->indices;
4235 ret = true;
4236 } else {
4237 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4239 return ret;
4242 #ifdef IXGBE_FCOE
4244 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4245 * @adapter: board private structure to initialize
4247 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4248 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4249 * rx queues out of the max number of rx queues, instead, it is used as the
4250 * index of the first rx queue used by FCoE.
4253 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4255 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4257 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4258 return false;
4260 f->indices = min((int)num_online_cpus(), f->indices);
4262 adapter->num_rx_queues = 1;
4263 adapter->num_tx_queues = 1;
4265 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4266 e_info(probe, "FCoE enabled with RSS\n");
4267 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4268 ixgbe_set_fdir_queues(adapter);
4269 else
4270 ixgbe_set_rss_queues(adapter);
4273 /* adding FCoE rx rings to the end */
4274 f->mask = adapter->num_rx_queues;
4275 adapter->num_rx_queues += f->indices;
4276 adapter->num_tx_queues += f->indices;
4278 return true;
4280 #endif /* IXGBE_FCOE */
4282 /* Artificial max queue cap per traffic class in DCB mode */
4283 #define DCB_QUEUE_CAP 8
4285 #ifdef CONFIG_IXGBE_DCB
4286 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4288 int per_tc_q, q, i, offset = 0;
4289 struct net_device *dev = adapter->netdev;
4290 int tcs = netdev_get_num_tc(dev);
4292 if (!tcs)
4293 return false;
4295 /* Map queue offset and counts onto allocated tx queues */
4296 per_tc_q = min(dev->num_tx_queues / tcs, (unsigned int)DCB_QUEUE_CAP);
4297 q = min((int)num_online_cpus(), per_tc_q);
4299 for (i = 0; i < tcs; i++) {
4300 netdev_set_prio_tc_map(dev, i, i);
4301 netdev_set_tc_queue(dev, i, q, offset);
4302 offset += q;
4305 adapter->num_tx_queues = q * tcs;
4306 adapter->num_rx_queues = q * tcs;
4308 #ifdef IXGBE_FCOE
4309 /* FCoE enabled queues require special configuration indexed
4310 * by feature specific indices and mask. Here we map FCoE
4311 * indices onto the DCB queue pairs allowing FCoE to own
4312 * configuration later.
4314 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4315 int tc;
4316 struct ixgbe_ring_feature *f =
4317 &adapter->ring_feature[RING_F_FCOE];
4319 tc = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
4320 f->indices = dev->tc_to_txq[tc].count;
4321 f->mask = dev->tc_to_txq[tc].offset;
4323 #endif
4325 return true;
4327 #endif
4330 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4331 * @adapter: board private structure to initialize
4333 * IOV doesn't actually use anything, so just NAK the
4334 * request for now and let the other queue routines
4335 * figure out what to do.
4337 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4339 return false;
4343 * ixgbe_set_num_queues: Allocate queues for device, feature dependent
4344 * @adapter: board private structure to initialize
4346 * This is the top level queue allocation routine. The order here is very
4347 * important, starting with the "most" number of features turned on at once,
4348 * and ending with the smallest set of features. This way large combinations
4349 * can be allocated if they're turned on, and smaller combinations are the
4350 * fallthrough conditions.
4353 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4355 /* Start with base case */
4356 adapter->num_rx_queues = 1;
4357 adapter->num_tx_queues = 1;
4358 adapter->num_rx_pools = adapter->num_rx_queues;
4359 adapter->num_rx_queues_per_pool = 1;
4361 if (ixgbe_set_sriov_queues(adapter))
4362 goto done;
4364 #ifdef CONFIG_IXGBE_DCB
4365 if (ixgbe_set_dcb_queues(adapter))
4366 goto done;
4368 #endif
4369 #ifdef IXGBE_FCOE
4370 if (ixgbe_set_fcoe_queues(adapter))
4371 goto done;
4373 #endif /* IXGBE_FCOE */
4374 if (ixgbe_set_fdir_queues(adapter))
4375 goto done;
4377 if (ixgbe_set_rss_queues(adapter))
4378 goto done;
4380 /* fallback to base case */
4381 adapter->num_rx_queues = 1;
4382 adapter->num_tx_queues = 1;
4384 done:
4385 /* Notify the stack of the (possibly) reduced queue counts. */
4386 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4387 return netif_set_real_num_rx_queues(adapter->netdev,
4388 adapter->num_rx_queues);
4391 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4392 int vectors)
4394 int err, vector_threshold;
4396 /* We'll want at least 3 (vector_threshold):
4397 * 1) TxQ[0] Cleanup
4398 * 2) RxQ[0] Cleanup
4399 * 3) Other (Link Status Change, etc.)
4400 * 4) TCP Timer (optional)
4402 vector_threshold = MIN_MSIX_COUNT;
4404 /* The more we get, the more we will assign to Tx/Rx Cleanup
4405 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4406 * Right now, we simply care about how many we'll get; we'll
4407 * set them up later while requesting irq's.
4409 while (vectors >= vector_threshold) {
4410 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4411 vectors);
4412 if (!err) /* Success in acquiring all requested vectors. */
4413 break;
4414 else if (err < 0)
4415 vectors = 0; /* Nasty failure, quit now */
4416 else /* err == number of vectors we should try again with */
4417 vectors = err;
4420 if (vectors < vector_threshold) {
4421 /* Can't allocate enough MSI-X interrupts? Oh well.
4422 * This just means we'll go with either a single MSI
4423 * vector or fall back to legacy interrupts.
4425 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4426 "Unable to allocate MSI-X interrupts\n");
4427 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4428 kfree(adapter->msix_entries);
4429 adapter->msix_entries = NULL;
4430 } else {
4431 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4433 * Adjust for only the vectors we'll use, which is minimum
4434 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4435 * vectors we were allocated.
4437 adapter->num_msix_vectors = min(vectors,
4438 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4443 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4444 * @adapter: board private structure to initialize
4446 * Cache the descriptor ring offsets for RSS to the assigned rings.
4449 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4451 int i;
4453 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4454 return false;
4456 for (i = 0; i < adapter->num_rx_queues; i++)
4457 adapter->rx_ring[i]->reg_idx = i;
4458 for (i = 0; i < adapter->num_tx_queues; i++)
4459 adapter->tx_ring[i]->reg_idx = i;
4461 return true;
4464 #ifdef CONFIG_IXGBE_DCB
4466 /* ixgbe_get_first_reg_idx - Return first register index associated with ring */
4467 static void ixgbe_get_first_reg_idx(struct ixgbe_adapter *adapter, u8 tc,
4468 unsigned int *tx, unsigned int *rx)
4470 struct net_device *dev = adapter->netdev;
4471 struct ixgbe_hw *hw = &adapter->hw;
4472 u8 num_tcs = netdev_get_num_tc(dev);
4474 *tx = 0;
4475 *rx = 0;
4477 switch (hw->mac.type) {
4478 case ixgbe_mac_82598EB:
4479 *tx = tc << 2;
4480 *rx = tc << 3;
4481 break;
4482 case ixgbe_mac_82599EB:
4483 case ixgbe_mac_X540:
4484 if (num_tcs > 4) {
4485 if (tc < 3) {
4486 *tx = tc << 5;
4487 *rx = tc << 4;
4488 } else if (tc < 5) {
4489 *tx = ((tc + 2) << 4);
4490 *rx = tc << 4;
4491 } else if (tc < num_tcs) {
4492 *tx = ((tc + 8) << 3);
4493 *rx = tc << 4;
4495 } else {
4496 *rx = tc << 5;
4497 switch (tc) {
4498 case 0:
4499 *tx = 0;
4500 break;
4501 case 1:
4502 *tx = 64;
4503 break;
4504 case 2:
4505 *tx = 96;
4506 break;
4507 case 3:
4508 *tx = 112;
4509 break;
4510 default:
4511 break;
4514 break;
4515 default:
4516 break;
4521 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4522 * @adapter: board private structure to initialize
4524 * Cache the descriptor ring offsets for DCB to the assigned rings.
4527 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4529 struct net_device *dev = adapter->netdev;
4530 int i, j, k;
4531 u8 num_tcs = netdev_get_num_tc(dev);
4533 if (!num_tcs)
4534 return false;
4536 for (i = 0, k = 0; i < num_tcs; i++) {
4537 unsigned int tx_s, rx_s;
4538 u16 count = dev->tc_to_txq[i].count;
4540 ixgbe_get_first_reg_idx(adapter, i, &tx_s, &rx_s);
4541 for (j = 0; j < count; j++, k++) {
4542 adapter->tx_ring[k]->reg_idx = tx_s + j;
4543 adapter->rx_ring[k]->reg_idx = rx_s + j;
4544 adapter->tx_ring[k]->dcb_tc = i;
4545 adapter->rx_ring[k]->dcb_tc = i;
4549 return true;
4551 #endif
4554 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4555 * @adapter: board private structure to initialize
4557 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4560 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4562 int i;
4563 bool ret = false;
4565 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4566 (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4567 for (i = 0; i < adapter->num_rx_queues; i++)
4568 adapter->rx_ring[i]->reg_idx = i;
4569 for (i = 0; i < adapter->num_tx_queues; i++)
4570 adapter->tx_ring[i]->reg_idx = i;
4571 ret = true;
4574 return ret;
4577 #ifdef IXGBE_FCOE
4579 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4580 * @adapter: board private structure to initialize
4582 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4585 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4587 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4588 int i;
4589 u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4591 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4592 return false;
4594 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4595 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4596 ixgbe_cache_ring_fdir(adapter);
4597 else
4598 ixgbe_cache_ring_rss(adapter);
4600 fcoe_rx_i = f->mask;
4601 fcoe_tx_i = f->mask;
4603 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4604 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4605 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4607 return true;
4610 #endif /* IXGBE_FCOE */
4612 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4613 * @adapter: board private structure to initialize
4615 * SR-IOV doesn't use any descriptor rings but changes the default if
4616 * no other mapping is used.
4619 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4621 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4622 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4623 if (adapter->num_vfs)
4624 return true;
4625 else
4626 return false;
4630 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4631 * @adapter: board private structure to initialize
4633 * Once we know the feature-set enabled for the device, we'll cache
4634 * the register offset the descriptor ring is assigned to.
4636 * Note, the order the various feature calls is important. It must start with
4637 * the "most" features enabled at the same time, then trickle down to the
4638 * least amount of features turned on at once.
4640 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4642 /* start with default case */
4643 adapter->rx_ring[0]->reg_idx = 0;
4644 adapter->tx_ring[0]->reg_idx = 0;
4646 if (ixgbe_cache_ring_sriov(adapter))
4647 return;
4649 #ifdef CONFIG_IXGBE_DCB
4650 if (ixgbe_cache_ring_dcb(adapter))
4651 return;
4652 #endif
4654 #ifdef IXGBE_FCOE
4655 if (ixgbe_cache_ring_fcoe(adapter))
4656 return;
4657 #endif /* IXGBE_FCOE */
4659 if (ixgbe_cache_ring_fdir(adapter))
4660 return;
4662 if (ixgbe_cache_ring_rss(adapter))
4663 return;
4667 * ixgbe_alloc_queues - Allocate memory for all rings
4668 * @adapter: board private structure to initialize
4670 * We allocate one ring per queue at run-time since we don't know the
4671 * number of queues at compile-time. The polling_netdev array is
4672 * intended for Multiqueue, but should work fine with a single queue.
4674 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4676 int rx = 0, tx = 0, nid = adapter->node;
4678 if (nid < 0 || !node_online(nid))
4679 nid = first_online_node;
4681 for (; tx < adapter->num_tx_queues; tx++) {
4682 struct ixgbe_ring *ring;
4684 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4685 if (!ring)
4686 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4687 if (!ring)
4688 goto err_allocation;
4689 ring->count = adapter->tx_ring_count;
4690 ring->queue_index = tx;
4691 ring->numa_node = nid;
4692 ring->dev = &adapter->pdev->dev;
4693 ring->netdev = adapter->netdev;
4695 adapter->tx_ring[tx] = ring;
4698 for (; rx < adapter->num_rx_queues; rx++) {
4699 struct ixgbe_ring *ring;
4701 ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, nid);
4702 if (!ring)
4703 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
4704 if (!ring)
4705 goto err_allocation;
4706 ring->count = adapter->rx_ring_count;
4707 ring->queue_index = rx;
4708 ring->numa_node = nid;
4709 ring->dev = &adapter->pdev->dev;
4710 ring->netdev = adapter->netdev;
4712 adapter->rx_ring[rx] = ring;
4715 ixgbe_cache_ring_register(adapter);
4717 return 0;
4719 err_allocation:
4720 while (tx)
4721 kfree(adapter->tx_ring[--tx]);
4723 while (rx)
4724 kfree(adapter->rx_ring[--rx]);
4725 return -ENOMEM;
4729 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4730 * @adapter: board private structure to initialize
4732 * Attempt to configure the interrupts using the best available
4733 * capabilities of the hardware and the kernel.
4735 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4737 struct ixgbe_hw *hw = &adapter->hw;
4738 int err = 0;
4739 int vector, v_budget;
4742 * It's easy to be greedy for MSI-X vectors, but it really
4743 * doesn't do us much good if we have a lot more vectors
4744 * than CPU's. So let's be conservative and only ask for
4745 * (roughly) the same number of vectors as there are CPU's.
4747 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4748 (int)num_online_cpus()) + NON_Q_VECTORS;
4751 * At the same time, hardware can only support a maximum of
4752 * hw.mac->max_msix_vectors vectors. With features
4753 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4754 * descriptor queues supported by our device. Thus, we cap it off in
4755 * those rare cases where the cpu count also exceeds our vector limit.
4757 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4759 /* A failure in MSI-X entry allocation isn't fatal, but it does
4760 * mean we disable MSI-X capabilities of the adapter. */
4761 adapter->msix_entries = kcalloc(v_budget,
4762 sizeof(struct msix_entry), GFP_KERNEL);
4763 if (adapter->msix_entries) {
4764 for (vector = 0; vector < v_budget; vector++)
4765 adapter->msix_entries[vector].entry = vector;
4767 ixgbe_acquire_msix_vectors(adapter, v_budget);
4769 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4770 goto out;
4773 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4774 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4775 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4776 e_err(probe,
4777 "ATR is not supported while multiple "
4778 "queues are disabled. Disabling Flow Director\n");
4780 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4781 adapter->atr_sample_rate = 0;
4782 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4783 ixgbe_disable_sriov(adapter);
4785 err = ixgbe_set_num_queues(adapter);
4786 if (err)
4787 return err;
4789 err = pci_enable_msi(adapter->pdev);
4790 if (!err) {
4791 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4792 } else {
4793 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4794 "Unable to allocate MSI interrupt, "
4795 "falling back to legacy. Error: %d\n", err);
4796 /* reset err */
4797 err = 0;
4800 out:
4801 return err;
4805 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4806 * @adapter: board private structure to initialize
4808 * We allocate one q_vector per queue interrupt. If allocation fails we
4809 * return -ENOMEM.
4811 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4813 int q_idx, num_q_vectors;
4814 struct ixgbe_q_vector *q_vector;
4815 int (*poll)(struct napi_struct *, int);
4817 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4818 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4819 poll = &ixgbe_clean_rxtx_many;
4820 } else {
4821 num_q_vectors = 1;
4822 poll = &ixgbe_poll;
4825 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4826 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4827 GFP_KERNEL, adapter->node);
4828 if (!q_vector)
4829 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4830 GFP_KERNEL);
4831 if (!q_vector)
4832 goto err_out;
4833 q_vector->adapter = adapter;
4834 if (q_vector->tx.count && !q_vector->rx.count)
4835 q_vector->eitr = adapter->tx_eitr_param;
4836 else
4837 q_vector->eitr = adapter->rx_eitr_param;
4838 q_vector->v_idx = q_idx;
4839 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4840 adapter->q_vector[q_idx] = q_vector;
4843 return 0;
4845 err_out:
4846 while (q_idx) {
4847 q_idx--;
4848 q_vector = adapter->q_vector[q_idx];
4849 netif_napi_del(&q_vector->napi);
4850 kfree(q_vector);
4851 adapter->q_vector[q_idx] = NULL;
4853 return -ENOMEM;
4857 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4858 * @adapter: board private structure to initialize
4860 * This function frees the memory allocated to the q_vectors. In addition if
4861 * NAPI is enabled it will delete any references to the NAPI struct prior
4862 * to freeing the q_vector.
4864 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4866 int q_idx, num_q_vectors;
4868 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4869 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4870 else
4871 num_q_vectors = 1;
4873 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4874 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4875 adapter->q_vector[q_idx] = NULL;
4876 netif_napi_del(&q_vector->napi);
4877 kfree(q_vector);
4881 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4883 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4884 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4885 pci_disable_msix(adapter->pdev);
4886 kfree(adapter->msix_entries);
4887 adapter->msix_entries = NULL;
4888 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4889 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4890 pci_disable_msi(adapter->pdev);
4895 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4896 * @adapter: board private structure to initialize
4898 * We determine which interrupt scheme to use based on...
4899 * - Kernel support (MSI, MSI-X)
4900 * - which can be user-defined (via MODULE_PARAM)
4901 * - Hardware queue count (num_*_queues)
4902 * - defined by miscellaneous hardware support/features (RSS, etc.)
4904 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4906 int err;
4908 /* Number of supported queues */
4909 err = ixgbe_set_num_queues(adapter);
4910 if (err)
4911 return err;
4913 err = ixgbe_set_interrupt_capability(adapter);
4914 if (err) {
4915 e_dev_err("Unable to setup interrupt capabilities\n");
4916 goto err_set_interrupt;
4919 err = ixgbe_alloc_q_vectors(adapter);
4920 if (err) {
4921 e_dev_err("Unable to allocate memory for queue vectors\n");
4922 goto err_alloc_q_vectors;
4925 err = ixgbe_alloc_queues(adapter);
4926 if (err) {
4927 e_dev_err("Unable to allocate memory for queues\n");
4928 goto err_alloc_queues;
4931 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4932 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4933 adapter->num_rx_queues, adapter->num_tx_queues);
4935 set_bit(__IXGBE_DOWN, &adapter->state);
4937 return 0;
4939 err_alloc_queues:
4940 ixgbe_free_q_vectors(adapter);
4941 err_alloc_q_vectors:
4942 ixgbe_reset_interrupt_capability(adapter);
4943 err_set_interrupt:
4944 return err;
4948 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4949 * @adapter: board private structure to clear interrupt scheme on
4951 * We go through and clear interrupt specific resources and reset the structure
4952 * to pre-load conditions
4954 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4956 int i;
4958 for (i = 0; i < adapter->num_tx_queues; i++) {
4959 kfree(adapter->tx_ring[i]);
4960 adapter->tx_ring[i] = NULL;
4962 for (i = 0; i < adapter->num_rx_queues; i++) {
4963 struct ixgbe_ring *ring = adapter->rx_ring[i];
4965 /* ixgbe_get_stats64() might access this ring, we must wait
4966 * a grace period before freeing it.
4968 kfree_rcu(ring, rcu);
4969 adapter->rx_ring[i] = NULL;
4972 adapter->num_tx_queues = 0;
4973 adapter->num_rx_queues = 0;
4975 ixgbe_free_q_vectors(adapter);
4976 ixgbe_reset_interrupt_capability(adapter);
4980 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4981 * @adapter: board private structure to initialize
4983 * ixgbe_sw_init initializes the Adapter private data structure.
4984 * Fields are initialized based on PCI device information and
4985 * OS network device settings (MTU size).
4987 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4989 struct ixgbe_hw *hw = &adapter->hw;
4990 struct pci_dev *pdev = adapter->pdev;
4991 struct net_device *dev = adapter->netdev;
4992 unsigned int rss;
4993 #ifdef CONFIG_IXGBE_DCB
4994 int j;
4995 struct tc_configuration *tc;
4996 #endif
4997 int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4999 /* PCI config space info */
5001 hw->vendor_id = pdev->vendor;
5002 hw->device_id = pdev->device;
5003 hw->revision_id = pdev->revision;
5004 hw->subsystem_vendor_id = pdev->subsystem_vendor;
5005 hw->subsystem_device_id = pdev->subsystem_device;
5007 /* Set capability flags */
5008 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
5009 adapter->ring_feature[RING_F_RSS].indices = rss;
5010 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
5011 switch (hw->mac.type) {
5012 case ixgbe_mac_82598EB:
5013 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5014 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5015 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
5016 break;
5017 case ixgbe_mac_82599EB:
5018 case ixgbe_mac_X540:
5019 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
5020 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5021 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5022 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5023 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5024 /* Flow Director hash filters enabled */
5025 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
5026 adapter->atr_sample_rate = 20;
5027 adapter->ring_feature[RING_F_FDIR].indices =
5028 IXGBE_MAX_FDIR_INDICES;
5029 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
5030 #ifdef IXGBE_FCOE
5031 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5032 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5033 adapter->ring_feature[RING_F_FCOE].indices = 0;
5034 #ifdef CONFIG_IXGBE_DCB
5035 /* Default traffic class to use for FCoE */
5036 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5037 #endif
5038 #endif /* IXGBE_FCOE */
5039 break;
5040 default:
5041 break;
5044 /* n-tuple support exists, always init our spinlock */
5045 spin_lock_init(&adapter->fdir_perfect_lock);
5047 #ifdef CONFIG_IXGBE_DCB
5048 /* Configure DCB traffic classes */
5049 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5050 tc = &adapter->dcb_cfg.tc_config[j];
5051 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5052 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5053 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5054 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5055 tc->dcb_pfc = pfc_disabled;
5057 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5058 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5059 adapter->dcb_cfg.pfc_mode_enable = false;
5060 adapter->dcb_set_bitmap = 0x00;
5061 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5062 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5063 MAX_TRAFFIC_CLASS);
5065 #endif
5067 /* default flow control settings */
5068 hw->fc.requested_mode = ixgbe_fc_full;
5069 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5070 #ifdef CONFIG_DCB
5071 adapter->last_lfc_mode = hw->fc.current_mode;
5072 #endif
5073 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5074 hw->fc.low_water = FC_LOW_WATER(max_frame);
5075 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5076 hw->fc.send_xon = true;
5077 hw->fc.disable_fc_autoneg = false;
5079 /* enable itr by default in dynamic mode */
5080 adapter->rx_itr_setting = 1;
5081 adapter->rx_eitr_param = 20000;
5082 adapter->tx_itr_setting = 1;
5083 adapter->tx_eitr_param = 10000;
5085 /* set defaults for eitr in MegaBytes */
5086 adapter->eitr_low = 10;
5087 adapter->eitr_high = 20;
5089 /* set default ring sizes */
5090 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5091 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5093 /* set default work limits */
5094 adapter->tx_work_limit = adapter->tx_ring_count;
5096 /* initialize eeprom parameters */
5097 if (ixgbe_init_eeprom_params_generic(hw)) {
5098 e_dev_err("EEPROM initialization failed\n");
5099 return -EIO;
5102 /* enable rx csum by default */
5103 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
5105 /* get assigned NUMA node */
5106 adapter->node = dev_to_node(&pdev->dev);
5108 set_bit(__IXGBE_DOWN, &adapter->state);
5110 return 0;
5114 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5115 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5117 * Return 0 on success, negative on failure
5119 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5121 struct device *dev = tx_ring->dev;
5122 int size;
5124 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5125 tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
5126 if (!tx_ring->tx_buffer_info)
5127 tx_ring->tx_buffer_info = vzalloc(size);
5128 if (!tx_ring->tx_buffer_info)
5129 goto err;
5131 /* round up to nearest 4K */
5132 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5133 tx_ring->size = ALIGN(tx_ring->size, 4096);
5135 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5136 &tx_ring->dma, GFP_KERNEL);
5137 if (!tx_ring->desc)
5138 goto err;
5140 tx_ring->next_to_use = 0;
5141 tx_ring->next_to_clean = 0;
5142 return 0;
5144 err:
5145 vfree(tx_ring->tx_buffer_info);
5146 tx_ring->tx_buffer_info = NULL;
5147 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5148 return -ENOMEM;
5152 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5153 * @adapter: board private structure
5155 * If this function returns with an error, then it's possible one or
5156 * more of the rings is populated (while the rest are not). It is the
5157 * callers duty to clean those orphaned rings.
5159 * Return 0 on success, negative on failure
5161 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5163 int i, err = 0;
5165 for (i = 0; i < adapter->num_tx_queues; i++) {
5166 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5167 if (!err)
5168 continue;
5169 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5170 break;
5173 return err;
5177 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5178 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5180 * Returns 0 on success, negative on failure
5182 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5184 struct device *dev = rx_ring->dev;
5185 int size;
5187 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5188 rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
5189 if (!rx_ring->rx_buffer_info)
5190 rx_ring->rx_buffer_info = vzalloc(size);
5191 if (!rx_ring->rx_buffer_info)
5192 goto err;
5194 /* Round up to nearest 4K */
5195 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5196 rx_ring->size = ALIGN(rx_ring->size, 4096);
5198 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5199 &rx_ring->dma, GFP_KERNEL);
5201 if (!rx_ring->desc)
5202 goto err;
5204 rx_ring->next_to_clean = 0;
5205 rx_ring->next_to_use = 0;
5207 return 0;
5208 err:
5209 vfree(rx_ring->rx_buffer_info);
5210 rx_ring->rx_buffer_info = NULL;
5211 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5212 return -ENOMEM;
5216 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5217 * @adapter: board private structure
5219 * If this function returns with an error, then it's possible one or
5220 * more of the rings is populated (while the rest are not). It is the
5221 * callers duty to clean those orphaned rings.
5223 * Return 0 on success, negative on failure
5225 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5227 int i, err = 0;
5229 for (i = 0; i < adapter->num_rx_queues; i++) {
5230 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5231 if (!err)
5232 continue;
5233 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5234 break;
5237 return err;
5241 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5242 * @tx_ring: Tx descriptor ring for a specific queue
5244 * Free all transmit software resources
5246 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5248 ixgbe_clean_tx_ring(tx_ring);
5250 vfree(tx_ring->tx_buffer_info);
5251 tx_ring->tx_buffer_info = NULL;
5253 /* if not set, then don't free */
5254 if (!tx_ring->desc)
5255 return;
5257 dma_free_coherent(tx_ring->dev, tx_ring->size,
5258 tx_ring->desc, tx_ring->dma);
5260 tx_ring->desc = NULL;
5264 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5265 * @adapter: board private structure
5267 * Free all transmit software resources
5269 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5271 int i;
5273 for (i = 0; i < adapter->num_tx_queues; i++)
5274 if (adapter->tx_ring[i]->desc)
5275 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5279 * ixgbe_free_rx_resources - Free Rx Resources
5280 * @rx_ring: ring to clean the resources from
5282 * Free all receive software resources
5284 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5286 ixgbe_clean_rx_ring(rx_ring);
5288 vfree(rx_ring->rx_buffer_info);
5289 rx_ring->rx_buffer_info = NULL;
5291 /* if not set, then don't free */
5292 if (!rx_ring->desc)
5293 return;
5295 dma_free_coherent(rx_ring->dev, rx_ring->size,
5296 rx_ring->desc, rx_ring->dma);
5298 rx_ring->desc = NULL;
5302 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5303 * @adapter: board private structure
5305 * Free all receive software resources
5307 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5309 int i;
5311 for (i = 0; i < adapter->num_rx_queues; i++)
5312 if (adapter->rx_ring[i]->desc)
5313 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5317 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5318 * @netdev: network interface device structure
5319 * @new_mtu: new value for maximum frame size
5321 * Returns 0 on success, negative on failure
5323 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5325 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5326 struct ixgbe_hw *hw = &adapter->hw;
5327 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5329 /* MTU < 68 is an error and causes problems on some kernels */
5330 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED &&
5331 hw->mac.type != ixgbe_mac_X540) {
5332 if ((new_mtu < 68) || (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
5333 return -EINVAL;
5334 } else {
5335 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5336 return -EINVAL;
5339 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5340 /* must set new MTU before calling down or up */
5341 netdev->mtu = new_mtu;
5343 hw->fc.high_water = FC_HIGH_WATER(max_frame);
5344 hw->fc.low_water = FC_LOW_WATER(max_frame);
5346 if (netif_running(netdev))
5347 ixgbe_reinit_locked(adapter);
5349 return 0;
5353 * ixgbe_open - Called when a network interface is made active
5354 * @netdev: network interface device structure
5356 * Returns 0 on success, negative value on failure
5358 * The open entry point is called when a network interface is made
5359 * active by the system (IFF_UP). At this point all resources needed
5360 * for transmit and receive operations are allocated, the interrupt
5361 * handler is registered with the OS, the watchdog timer is started,
5362 * and the stack is notified that the interface is ready.
5364 static int ixgbe_open(struct net_device *netdev)
5366 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5367 int err;
5369 /* disallow open during test */
5370 if (test_bit(__IXGBE_TESTING, &adapter->state))
5371 return -EBUSY;
5373 netif_carrier_off(netdev);
5375 /* allocate transmit descriptors */
5376 err = ixgbe_setup_all_tx_resources(adapter);
5377 if (err)
5378 goto err_setup_tx;
5380 /* allocate receive descriptors */
5381 err = ixgbe_setup_all_rx_resources(adapter);
5382 if (err)
5383 goto err_setup_rx;
5385 ixgbe_configure(adapter);
5387 err = ixgbe_request_irq(adapter);
5388 if (err)
5389 goto err_req_irq;
5391 err = ixgbe_up_complete(adapter);
5392 if (err)
5393 goto err_up;
5395 netif_tx_start_all_queues(netdev);
5397 return 0;
5399 err_up:
5400 ixgbe_release_hw_control(adapter);
5401 ixgbe_free_irq(adapter);
5402 err_req_irq:
5403 err_setup_rx:
5404 ixgbe_free_all_rx_resources(adapter);
5405 err_setup_tx:
5406 ixgbe_free_all_tx_resources(adapter);
5407 ixgbe_reset(adapter);
5409 return err;
5413 * ixgbe_close - Disables a network interface
5414 * @netdev: network interface device structure
5416 * Returns 0, this is not allowed to fail
5418 * The close entry point is called when an interface is de-activated
5419 * by the OS. The hardware is still under the drivers control, but
5420 * needs to be disabled. A global MAC reset is issued to stop the
5421 * hardware, and all transmit and receive resources are freed.
5423 static int ixgbe_close(struct net_device *netdev)
5425 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5427 ixgbe_down(adapter);
5428 ixgbe_free_irq(adapter);
5430 ixgbe_fdir_filter_exit(adapter);
5432 ixgbe_free_all_tx_resources(adapter);
5433 ixgbe_free_all_rx_resources(adapter);
5435 ixgbe_release_hw_control(adapter);
5437 return 0;
5440 #ifdef CONFIG_PM
5441 static int ixgbe_resume(struct pci_dev *pdev)
5443 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5444 struct net_device *netdev = adapter->netdev;
5445 u32 err;
5447 pci_set_power_state(pdev, PCI_D0);
5448 pci_restore_state(pdev);
5450 * pci_restore_state clears dev->state_saved so call
5451 * pci_save_state to restore it.
5453 pci_save_state(pdev);
5455 err = pci_enable_device_mem(pdev);
5456 if (err) {
5457 e_dev_err("Cannot enable PCI device from suspend\n");
5458 return err;
5460 pci_set_master(pdev);
5462 pci_wake_from_d3(pdev, false);
5464 err = ixgbe_init_interrupt_scheme(adapter);
5465 if (err) {
5466 e_dev_err("Cannot initialize interrupts for device\n");
5467 return err;
5470 ixgbe_reset(adapter);
5472 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5474 if (netif_running(netdev)) {
5475 err = ixgbe_open(netdev);
5476 if (err)
5477 return err;
5480 netif_device_attach(netdev);
5482 return 0;
5484 #endif /* CONFIG_PM */
5486 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5488 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5489 struct net_device *netdev = adapter->netdev;
5490 struct ixgbe_hw *hw = &adapter->hw;
5491 u32 ctrl, fctrl;
5492 u32 wufc = adapter->wol;
5493 #ifdef CONFIG_PM
5494 int retval = 0;
5495 #endif
5497 netif_device_detach(netdev);
5499 if (netif_running(netdev)) {
5500 ixgbe_down(adapter);
5501 ixgbe_free_irq(adapter);
5502 ixgbe_free_all_tx_resources(adapter);
5503 ixgbe_free_all_rx_resources(adapter);
5506 ixgbe_clear_interrupt_scheme(adapter);
5507 #ifdef CONFIG_DCB
5508 kfree(adapter->ixgbe_ieee_pfc);
5509 kfree(adapter->ixgbe_ieee_ets);
5510 #endif
5512 #ifdef CONFIG_PM
5513 retval = pci_save_state(pdev);
5514 if (retval)
5515 return retval;
5517 #endif
5518 if (wufc) {
5519 ixgbe_set_rx_mode(netdev);
5521 /* turn on all-multi mode if wake on multicast is enabled */
5522 if (wufc & IXGBE_WUFC_MC) {
5523 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5524 fctrl |= IXGBE_FCTRL_MPE;
5525 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5528 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5529 ctrl |= IXGBE_CTRL_GIO_DIS;
5530 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5532 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5533 } else {
5534 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5535 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5538 switch (hw->mac.type) {
5539 case ixgbe_mac_82598EB:
5540 pci_wake_from_d3(pdev, false);
5541 break;
5542 case ixgbe_mac_82599EB:
5543 case ixgbe_mac_X540:
5544 pci_wake_from_d3(pdev, !!wufc);
5545 break;
5546 default:
5547 break;
5550 *enable_wake = !!wufc;
5552 ixgbe_release_hw_control(adapter);
5554 pci_disable_device(pdev);
5556 return 0;
5559 #ifdef CONFIG_PM
5560 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5562 int retval;
5563 bool wake;
5565 retval = __ixgbe_shutdown(pdev, &wake);
5566 if (retval)
5567 return retval;
5569 if (wake) {
5570 pci_prepare_to_sleep(pdev);
5571 } else {
5572 pci_wake_from_d3(pdev, false);
5573 pci_set_power_state(pdev, PCI_D3hot);
5576 return 0;
5578 #endif /* CONFIG_PM */
5580 static void ixgbe_shutdown(struct pci_dev *pdev)
5582 bool wake;
5584 __ixgbe_shutdown(pdev, &wake);
5586 if (system_state == SYSTEM_POWER_OFF) {
5587 pci_wake_from_d3(pdev, wake);
5588 pci_set_power_state(pdev, PCI_D3hot);
5593 * ixgbe_update_stats - Update the board statistics counters.
5594 * @adapter: board private structure
5596 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5598 struct net_device *netdev = adapter->netdev;
5599 struct ixgbe_hw *hw = &adapter->hw;
5600 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5601 u64 total_mpc = 0;
5602 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5603 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5604 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5605 u64 bytes = 0, packets = 0;
5607 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5608 test_bit(__IXGBE_RESETTING, &adapter->state))
5609 return;
5611 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5612 u64 rsc_count = 0;
5613 u64 rsc_flush = 0;
5614 for (i = 0; i < 16; i++)
5615 adapter->hw_rx_no_dma_resources +=
5616 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5617 for (i = 0; i < adapter->num_rx_queues; i++) {
5618 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5619 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5621 adapter->rsc_total_count = rsc_count;
5622 adapter->rsc_total_flush = rsc_flush;
5625 for (i = 0; i < adapter->num_rx_queues; i++) {
5626 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5627 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5628 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5629 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5630 bytes += rx_ring->stats.bytes;
5631 packets += rx_ring->stats.packets;
5633 adapter->non_eop_descs = non_eop_descs;
5634 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5635 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5636 netdev->stats.rx_bytes = bytes;
5637 netdev->stats.rx_packets = packets;
5639 bytes = 0;
5640 packets = 0;
5641 /* gather some stats to the adapter struct that are per queue */
5642 for (i = 0; i < adapter->num_tx_queues; i++) {
5643 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5644 restart_queue += tx_ring->tx_stats.restart_queue;
5645 tx_busy += tx_ring->tx_stats.tx_busy;
5646 bytes += tx_ring->stats.bytes;
5647 packets += tx_ring->stats.packets;
5649 adapter->restart_queue = restart_queue;
5650 adapter->tx_busy = tx_busy;
5651 netdev->stats.tx_bytes = bytes;
5652 netdev->stats.tx_packets = packets;
5654 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5655 for (i = 0; i < 8; i++) {
5656 /* for packet buffers not used, the register should read 0 */
5657 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5658 missed_rx += mpc;
5659 hwstats->mpc[i] += mpc;
5660 total_mpc += hwstats->mpc[i];
5661 if (hw->mac.type == ixgbe_mac_82598EB)
5662 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5663 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5664 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5665 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5666 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5667 switch (hw->mac.type) {
5668 case ixgbe_mac_82598EB:
5669 hwstats->pxonrxc[i] +=
5670 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5671 break;
5672 case ixgbe_mac_82599EB:
5673 case ixgbe_mac_X540:
5674 hwstats->pxonrxc[i] +=
5675 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5676 break;
5677 default:
5678 break;
5680 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5681 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5683 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5684 /* work around hardware counting issue */
5685 hwstats->gprc -= missed_rx;
5687 ixgbe_update_xoff_received(adapter);
5689 /* 82598 hardware only has a 32 bit counter in the high register */
5690 switch (hw->mac.type) {
5691 case ixgbe_mac_82598EB:
5692 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5693 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5694 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5695 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5696 break;
5697 case ixgbe_mac_X540:
5698 /* OS2BMC stats are X540 only*/
5699 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5700 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5701 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5702 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5703 case ixgbe_mac_82599EB:
5704 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5705 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5706 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5707 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5708 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5709 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5710 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5711 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5712 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5713 #ifdef IXGBE_FCOE
5714 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5715 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5716 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5717 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5718 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5719 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5720 #endif /* IXGBE_FCOE */
5721 break;
5722 default:
5723 break;
5725 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5726 hwstats->bprc += bprc;
5727 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5728 if (hw->mac.type == ixgbe_mac_82598EB)
5729 hwstats->mprc -= bprc;
5730 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5731 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5732 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5733 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5734 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5735 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5736 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5737 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5738 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5739 hwstats->lxontxc += lxon;
5740 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5741 hwstats->lxofftxc += lxoff;
5742 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5743 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5744 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5746 * 82598 errata - tx of flow control packets is included in tx counters
5748 xon_off_tot = lxon + lxoff;
5749 hwstats->gptc -= xon_off_tot;
5750 hwstats->mptc -= xon_off_tot;
5751 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5752 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5753 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5754 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5755 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5756 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5757 hwstats->ptc64 -= xon_off_tot;
5758 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5759 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5760 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5761 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5762 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5763 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5765 /* Fill out the OS statistics structure */
5766 netdev->stats.multicast = hwstats->mprc;
5768 /* Rx Errors */
5769 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5770 netdev->stats.rx_dropped = 0;
5771 netdev->stats.rx_length_errors = hwstats->rlec;
5772 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5773 netdev->stats.rx_missed_errors = total_mpc;
5777 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5778 * @adapter - pointer to the device adapter structure
5780 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5782 struct ixgbe_hw *hw = &adapter->hw;
5783 int i;
5785 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5786 return;
5788 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5790 /* if interface is down do nothing */
5791 if (test_bit(__IXGBE_DOWN, &adapter->state))
5792 return;
5794 /* do nothing if we are not using signature filters */
5795 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5796 return;
5798 adapter->fdir_overflow++;
5800 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5801 for (i = 0; i < adapter->num_tx_queues; i++)
5802 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5803 &(adapter->tx_ring[i]->state));
5804 /* re-enable flow director interrupts */
5805 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5806 } else {
5807 e_err(probe, "failed to finish FDIR re-initialization, "
5808 "ignored adding FDIR ATR filters\n");
5813 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5814 * @adapter - pointer to the device adapter structure
5816 * This function serves two purposes. First it strobes the interrupt lines
5817 * in order to make certain interrupts are occuring. Secondly it sets the
5818 * bits needed to check for TX hangs. As a result we should immediately
5819 * determine if a hang has occured.
5821 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5823 struct ixgbe_hw *hw = &adapter->hw;
5824 u64 eics = 0;
5825 int i;
5827 /* If we're down or resetting, just bail */
5828 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5829 test_bit(__IXGBE_RESETTING, &adapter->state))
5830 return;
5832 /* Force detection of hung controller */
5833 if (netif_carrier_ok(adapter->netdev)) {
5834 for (i = 0; i < adapter->num_tx_queues; i++)
5835 set_check_for_tx_hang(adapter->tx_ring[i]);
5838 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5840 * for legacy and MSI interrupts don't set any bits
5841 * that are enabled for EIAM, because this operation
5842 * would set *both* EIMS and EICS for any bit in EIAM
5844 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5845 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5846 } else {
5847 /* get one bit for every active tx/rx interrupt vector */
5848 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5849 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5850 if (qv->rx.ring || qv->tx.ring)
5851 eics |= ((u64)1 << i);
5855 /* Cause software interrupt to ensure rings are cleaned */
5856 ixgbe_irq_rearm_queues(adapter, eics);
5861 * ixgbe_watchdog_update_link - update the link status
5862 * @adapter - pointer to the device adapter structure
5863 * @link_speed - pointer to a u32 to store the link_speed
5865 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5867 struct ixgbe_hw *hw = &adapter->hw;
5868 u32 link_speed = adapter->link_speed;
5869 bool link_up = adapter->link_up;
5870 int i;
5872 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5873 return;
5875 if (hw->mac.ops.check_link) {
5876 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5877 } else {
5878 /* always assume link is up, if no check link function */
5879 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5880 link_up = true;
5882 if (link_up) {
5883 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5884 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5885 hw->mac.ops.fc_enable(hw, i);
5886 } else {
5887 hw->mac.ops.fc_enable(hw, 0);
5891 if (link_up ||
5892 time_after(jiffies, (adapter->link_check_timeout +
5893 IXGBE_TRY_LINK_TIMEOUT))) {
5894 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5895 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5896 IXGBE_WRITE_FLUSH(hw);
5899 adapter->link_up = link_up;
5900 adapter->link_speed = link_speed;
5904 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5905 * print link up message
5906 * @adapter - pointer to the device adapter structure
5908 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5910 struct net_device *netdev = adapter->netdev;
5911 struct ixgbe_hw *hw = &adapter->hw;
5912 u32 link_speed = adapter->link_speed;
5913 bool flow_rx, flow_tx;
5915 /* only continue if link was previously down */
5916 if (netif_carrier_ok(netdev))
5917 return;
5919 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5921 switch (hw->mac.type) {
5922 case ixgbe_mac_82598EB: {
5923 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5924 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5925 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5926 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5928 break;
5929 case ixgbe_mac_X540:
5930 case ixgbe_mac_82599EB: {
5931 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5932 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5933 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5934 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5936 break;
5937 default:
5938 flow_tx = false;
5939 flow_rx = false;
5940 break;
5942 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5943 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5944 "10 Gbps" :
5945 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5946 "1 Gbps" :
5947 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5948 "100 Mbps" :
5949 "unknown speed"))),
5950 ((flow_rx && flow_tx) ? "RX/TX" :
5951 (flow_rx ? "RX" :
5952 (flow_tx ? "TX" : "None"))));
5954 netif_carrier_on(netdev);
5955 ixgbe_check_vf_rate_limit(adapter);
5959 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5960 * print link down message
5961 * @adapter - pointer to the adapter structure
5963 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter* adapter)
5965 struct net_device *netdev = adapter->netdev;
5966 struct ixgbe_hw *hw = &adapter->hw;
5968 adapter->link_up = false;
5969 adapter->link_speed = 0;
5971 /* only continue if link was up previously */
5972 if (!netif_carrier_ok(netdev))
5973 return;
5975 /* poll for SFP+ cable when link is down */
5976 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5977 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5979 e_info(drv, "NIC Link is Down\n");
5980 netif_carrier_off(netdev);
5984 * ixgbe_watchdog_flush_tx - flush queues on link down
5985 * @adapter - pointer to the device adapter structure
5987 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5989 int i;
5990 int some_tx_pending = 0;
5992 if (!netif_carrier_ok(adapter->netdev)) {
5993 for (i = 0; i < adapter->num_tx_queues; i++) {
5994 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5995 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5996 some_tx_pending = 1;
5997 break;
6001 if (some_tx_pending) {
6002 /* We've lost link, so the controller stops DMA,
6003 * but we've got queued Tx work that's never going
6004 * to get done, so reset controller to flush Tx.
6005 * (Do the reset outside of interrupt context).
6007 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
6012 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6014 u32 ssvpc;
6016 /* Do not perform spoof check for 82598 */
6017 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6018 return;
6020 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6023 * ssvpc register is cleared on read, if zero then no
6024 * spoofed packets in the last interval.
6026 if (!ssvpc)
6027 return;
6029 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
6033 * ixgbe_watchdog_subtask - check and bring link up
6034 * @adapter - pointer to the device adapter structure
6036 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6038 /* if interface is down do nothing */
6039 if (test_bit(__IXGBE_DOWN, &adapter->state))
6040 return;
6042 ixgbe_watchdog_update_link(adapter);
6044 if (adapter->link_up)
6045 ixgbe_watchdog_link_is_up(adapter);
6046 else
6047 ixgbe_watchdog_link_is_down(adapter);
6049 ixgbe_spoof_check(adapter);
6050 ixgbe_update_stats(adapter);
6052 ixgbe_watchdog_flush_tx(adapter);
6056 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6057 * @adapter - the ixgbe adapter structure
6059 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6061 struct ixgbe_hw *hw = &adapter->hw;
6062 s32 err;
6064 /* not searching for SFP so there is nothing to do here */
6065 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6066 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6067 return;
6069 /* someone else is in init, wait until next service event */
6070 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6071 return;
6073 err = hw->phy.ops.identify_sfp(hw);
6074 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6075 goto sfp_out;
6077 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6078 /* If no cable is present, then we need to reset
6079 * the next time we find a good cable. */
6080 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6083 /* exit on error */
6084 if (err)
6085 goto sfp_out;
6087 /* exit if reset not needed */
6088 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6089 goto sfp_out;
6091 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6094 * A module may be identified correctly, but the EEPROM may not have
6095 * support for that module. setup_sfp() will fail in that case, so
6096 * we should not allow that module to load.
6098 if (hw->mac.type == ixgbe_mac_82598EB)
6099 err = hw->phy.ops.reset(hw);
6100 else
6101 err = hw->mac.ops.setup_sfp(hw);
6103 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6104 goto sfp_out;
6106 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6107 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6109 sfp_out:
6110 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6112 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6113 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6114 e_dev_err("failed to initialize because an unsupported "
6115 "SFP+ module type was detected.\n");
6116 e_dev_err("Reload the driver after installing a "
6117 "supported module.\n");
6118 unregister_netdev(adapter->netdev);
6123 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6124 * @adapter - the ixgbe adapter structure
6126 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6128 struct ixgbe_hw *hw = &adapter->hw;
6129 u32 autoneg;
6130 bool negotiation;
6132 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6133 return;
6135 /* someone else is in init, wait until next service event */
6136 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6137 return;
6139 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6141 autoneg = hw->phy.autoneg_advertised;
6142 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
6143 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
6144 hw->mac.autotry_restart = false;
6145 if (hw->mac.ops.setup_link)
6146 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
6148 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6149 adapter->link_check_timeout = jiffies;
6150 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6154 * ixgbe_service_timer - Timer Call-back
6155 * @data: pointer to adapter cast into an unsigned long
6157 static void ixgbe_service_timer(unsigned long data)
6159 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6160 unsigned long next_event_offset;
6162 /* poll faster when waiting for link */
6163 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6164 next_event_offset = HZ / 10;
6165 else
6166 next_event_offset = HZ * 2;
6168 /* Reset the timer */
6169 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6171 ixgbe_service_event_schedule(adapter);
6174 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6176 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6177 return;
6179 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6181 /* If we're already down or resetting, just bail */
6182 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6183 test_bit(__IXGBE_RESETTING, &adapter->state))
6184 return;
6186 ixgbe_dump(adapter);
6187 netdev_err(adapter->netdev, "Reset adapter\n");
6188 adapter->tx_timeout_count++;
6190 ixgbe_reinit_locked(adapter);
6194 * ixgbe_service_task - manages and runs subtasks
6195 * @work: pointer to work_struct containing our data
6197 static void ixgbe_service_task(struct work_struct *work)
6199 struct ixgbe_adapter *adapter = container_of(work,
6200 struct ixgbe_adapter,
6201 service_task);
6203 ixgbe_reset_subtask(adapter);
6204 ixgbe_sfp_detection_subtask(adapter);
6205 ixgbe_sfp_link_config_subtask(adapter);
6206 ixgbe_check_overtemp_subtask(adapter);
6207 ixgbe_watchdog_subtask(adapter);
6208 ixgbe_fdir_reinit_subtask(adapter);
6209 ixgbe_check_hang_subtask(adapter);
6211 ixgbe_service_event_complete(adapter);
6214 void ixgbe_tx_ctxtdesc(struct ixgbe_ring *tx_ring, u32 vlan_macip_lens,
6215 u32 fcoe_sof_eof, u32 type_tucmd, u32 mss_l4len_idx)
6217 struct ixgbe_adv_tx_context_desc *context_desc;
6218 u16 i = tx_ring->next_to_use;
6220 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6222 i++;
6223 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
6225 /* set bits to identify this as an advanced context descriptor */
6226 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
6228 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6229 context_desc->seqnum_seed = cpu_to_le32(fcoe_sof_eof);
6230 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
6231 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6234 static int ixgbe_tso(struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6235 u32 tx_flags, __be16 protocol, u8 *hdr_len)
6237 int err;
6238 u32 vlan_macip_lens, type_tucmd;
6239 u32 mss_l4len_idx, l4len;
6241 if (!skb_is_gso(skb))
6242 return 0;
6244 if (skb_header_cloned(skb)) {
6245 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6246 if (err)
6247 return err;
6250 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6251 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6253 if (protocol == __constant_htons(ETH_P_IP)) {
6254 struct iphdr *iph = ip_hdr(skb);
6255 iph->tot_len = 0;
6256 iph->check = 0;
6257 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6258 iph->daddr, 0,
6259 IPPROTO_TCP,
6261 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6262 } else if (skb_is_gso_v6(skb)) {
6263 ipv6_hdr(skb)->payload_len = 0;
6264 tcp_hdr(skb)->check =
6265 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6266 &ipv6_hdr(skb)->daddr,
6267 0, IPPROTO_TCP, 0);
6270 l4len = tcp_hdrlen(skb);
6271 *hdr_len = skb_transport_offset(skb) + l4len;
6273 /* mss_l4len_id: use 1 as index for TSO */
6274 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6275 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6276 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
6278 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6279 vlan_macip_lens = skb_network_header_len(skb);
6280 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6281 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6283 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6284 mss_l4len_idx);
6286 return 1;
6289 static bool ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6290 struct sk_buff *skb, u32 tx_flags,
6291 __be16 protocol)
6293 u32 vlan_macip_lens = 0;
6294 u32 mss_l4len_idx = 0;
6295 u32 type_tucmd = 0;
6297 if (skb->ip_summed != CHECKSUM_PARTIAL) {
6298 if (!(tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6299 !(tx_flags & IXGBE_TX_FLAGS_TXSW))
6300 return false;
6301 } else {
6302 u8 l4_hdr = 0;
6303 switch (protocol) {
6304 case __constant_htons(ETH_P_IP):
6305 vlan_macip_lens |= skb_network_header_len(skb);
6306 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6307 l4_hdr = ip_hdr(skb)->protocol;
6308 break;
6309 case __constant_htons(ETH_P_IPV6):
6310 vlan_macip_lens |= skb_network_header_len(skb);
6311 l4_hdr = ipv6_hdr(skb)->nexthdr;
6312 break;
6313 default:
6314 if (unlikely(net_ratelimit())) {
6315 dev_warn(tx_ring->dev,
6316 "partial checksum but proto=%x!\n",
6317 skb->protocol);
6319 break;
6322 switch (l4_hdr) {
6323 case IPPROTO_TCP:
6324 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6325 mss_l4len_idx = tcp_hdrlen(skb) <<
6326 IXGBE_ADVTXD_L4LEN_SHIFT;
6327 break;
6328 case IPPROTO_SCTP:
6329 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6330 mss_l4len_idx = sizeof(struct sctphdr) <<
6331 IXGBE_ADVTXD_L4LEN_SHIFT;
6332 break;
6333 case IPPROTO_UDP:
6334 mss_l4len_idx = sizeof(struct udphdr) <<
6335 IXGBE_ADVTXD_L4LEN_SHIFT;
6336 break;
6337 default:
6338 if (unlikely(net_ratelimit())) {
6339 dev_warn(tx_ring->dev,
6340 "partial checksum but l4 proto=%x!\n",
6341 skb->protocol);
6343 break;
6347 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6348 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6350 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6351 type_tucmd, mss_l4len_idx);
6353 return (skb->ip_summed == CHECKSUM_PARTIAL);
6356 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
6358 /* set type for advanced descriptor with frame checksum insertion */
6359 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
6360 IXGBE_ADVTXD_DCMD_IFCS |
6361 IXGBE_ADVTXD_DCMD_DEXT);
6363 /* set HW vlan bit if vlan is present */
6364 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
6365 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
6367 /* set segmentation enable bits for TSO/FSO */
6368 #ifdef IXGBE_FCOE
6369 if ((tx_flags & IXGBE_TX_FLAGS_TSO) || (tx_flags & IXGBE_TX_FLAGS_FSO))
6370 #else
6371 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6372 #endif
6373 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
6375 return cmd_type;
6378 static __le32 ixgbe_tx_olinfo_status(u32 tx_flags, unsigned int paylen)
6380 __le32 olinfo_status =
6381 cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
6383 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6384 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM |
6385 (1 << IXGBE_ADVTXD_IDX_SHIFT));
6386 /* enble IPv4 checksum for TSO */
6387 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6388 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
6391 /* enable L4 checksum for TSO and TX checksum offload */
6392 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6393 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
6395 #ifdef IXGBE_FCOE
6396 /* use index 1 context for FCOE/FSO */
6397 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6398 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC |
6399 (1 << IXGBE_ADVTXD_IDX_SHIFT));
6401 #endif
6403 * Check Context must be set if Tx switch is enabled, which it
6404 * always is for case where virtual functions are running
6406 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
6407 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
6409 return olinfo_status;
6412 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6413 IXGBE_TXD_CMD_RS)
6415 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6416 struct sk_buff *skb,
6417 struct ixgbe_tx_buffer *first,
6418 u32 tx_flags,
6419 const u8 hdr_len)
6421 struct device *dev = tx_ring->dev;
6422 struct ixgbe_tx_buffer *tx_buffer_info;
6423 union ixgbe_adv_tx_desc *tx_desc;
6424 dma_addr_t dma;
6425 __le32 cmd_type, olinfo_status;
6426 struct skb_frag_struct *frag;
6427 unsigned int f = 0;
6428 unsigned int data_len = skb->data_len;
6429 unsigned int size = skb_headlen(skb);
6430 u32 offset = 0;
6431 u32 paylen = skb->len - hdr_len;
6432 u16 i = tx_ring->next_to_use;
6433 u16 gso_segs;
6435 #ifdef IXGBE_FCOE
6436 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6437 if (data_len >= sizeof(struct fcoe_crc_eof)) {
6438 data_len -= sizeof(struct fcoe_crc_eof);
6439 } else {
6440 size -= sizeof(struct fcoe_crc_eof) - data_len;
6441 data_len = 0;
6445 #endif
6446 dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
6447 if (dma_mapping_error(dev, dma))
6448 goto dma_error;
6450 cmd_type = ixgbe_tx_cmd_type(tx_flags);
6451 olinfo_status = ixgbe_tx_olinfo_status(tx_flags, paylen);
6453 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6455 for (;;) {
6456 while (size > IXGBE_MAX_DATA_PER_TXD) {
6457 tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6458 tx_desc->read.cmd_type_len =
6459 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
6460 tx_desc->read.olinfo_status = olinfo_status;
6462 offset += IXGBE_MAX_DATA_PER_TXD;
6463 size -= IXGBE_MAX_DATA_PER_TXD;
6465 tx_desc++;
6466 i++;
6467 if (i == tx_ring->count) {
6468 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6469 i = 0;
6473 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6474 tx_buffer_info->length = offset + size;
6475 tx_buffer_info->tx_flags = tx_flags;
6476 tx_buffer_info->dma = dma;
6478 tx_desc->read.buffer_addr = cpu_to_le64(dma + offset);
6479 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6480 tx_desc->read.olinfo_status = olinfo_status;
6482 if (!data_len)
6483 break;
6485 frag = &skb_shinfo(skb)->frags[f];
6486 #ifdef IXGBE_FCOE
6487 size = min_t(unsigned int, data_len, frag->size);
6488 #else
6489 size = frag->size;
6490 #endif
6491 data_len -= size;
6492 f++;
6494 offset = 0;
6495 tx_flags |= IXGBE_TX_FLAGS_MAPPED_AS_PAGE;
6497 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
6498 if (dma_mapping_error(dev, dma))
6499 goto dma_error;
6501 tx_desc++;
6502 i++;
6503 if (i == tx_ring->count) {
6504 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, 0);
6505 i = 0;
6509 tx_desc->read.cmd_type_len |= cpu_to_le32(IXGBE_TXD_CMD);
6511 i++;
6512 if (i == tx_ring->count)
6513 i = 0;
6515 tx_ring->next_to_use = i;
6517 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6518 gso_segs = skb_shinfo(skb)->gso_segs;
6519 #ifdef IXGBE_FCOE
6520 /* adjust for FCoE Sequence Offload */
6521 else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6522 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6523 skb_shinfo(skb)->gso_size);
6524 #endif /* IXGBE_FCOE */
6525 else
6526 gso_segs = 1;
6528 /* multiply data chunks by size of headers */
6529 tx_buffer_info->bytecount = paylen + (gso_segs * hdr_len);
6530 tx_buffer_info->gso_segs = gso_segs;
6531 tx_buffer_info->skb = skb;
6533 /* set the timestamp */
6534 first->time_stamp = jiffies;
6537 * Force memory writes to complete before letting h/w
6538 * know there are new descriptors to fetch. (Only
6539 * applicable for weak-ordered memory model archs,
6540 * such as IA-64).
6542 wmb();
6544 /* set next_to_watch value indicating a packet is present */
6545 first->next_to_watch = tx_desc;
6547 /* notify HW of packet */
6548 writel(i, tx_ring->tail);
6550 return;
6551 dma_error:
6552 dev_err(dev, "TX DMA map failed\n");
6554 /* clear dma mappings for failed tx_buffer_info map */
6555 for (;;) {
6556 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6557 ixgbe_unmap_tx_resource(tx_ring, tx_buffer_info);
6558 if (tx_buffer_info == first)
6559 break;
6560 if (i == 0)
6561 i = tx_ring->count;
6562 i--;
6565 dev_kfree_skb_any(skb);
6567 tx_ring->next_to_use = i;
6570 static void ixgbe_atr(struct ixgbe_ring *ring, struct sk_buff *skb,
6571 u32 tx_flags, __be16 protocol)
6573 struct ixgbe_q_vector *q_vector = ring->q_vector;
6574 union ixgbe_atr_hash_dword input = { .dword = 0 };
6575 union ixgbe_atr_hash_dword common = { .dword = 0 };
6576 union {
6577 unsigned char *network;
6578 struct iphdr *ipv4;
6579 struct ipv6hdr *ipv6;
6580 } hdr;
6581 struct tcphdr *th;
6582 __be16 vlan_id;
6584 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6585 if (!q_vector)
6586 return;
6588 /* do nothing if sampling is disabled */
6589 if (!ring->atr_sample_rate)
6590 return;
6592 ring->atr_count++;
6594 /* snag network header to get L4 type and address */
6595 hdr.network = skb_network_header(skb);
6597 /* Currently only IPv4/IPv6 with TCP is supported */
6598 if ((protocol != __constant_htons(ETH_P_IPV6) ||
6599 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6600 (protocol != __constant_htons(ETH_P_IP) ||
6601 hdr.ipv4->protocol != IPPROTO_TCP))
6602 return;
6604 th = tcp_hdr(skb);
6606 /* skip this packet since it is invalid or the socket is closing */
6607 if (!th || th->fin)
6608 return;
6610 /* sample on all syn packets or once every atr sample count */
6611 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6612 return;
6614 /* reset sample count */
6615 ring->atr_count = 0;
6617 vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6620 * src and dst are inverted, think how the receiver sees them
6622 * The input is broken into two sections, a non-compressed section
6623 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6624 * is XORed together and stored in the compressed dword.
6626 input.formatted.vlan_id = vlan_id;
6629 * since src port and flex bytes occupy the same word XOR them together
6630 * and write the value to source port portion of compressed dword
6632 if (tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6633 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6634 else
6635 common.port.src ^= th->dest ^ protocol;
6636 common.port.dst ^= th->source;
6638 if (protocol == __constant_htons(ETH_P_IP)) {
6639 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6640 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6641 } else {
6642 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6643 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6644 hdr.ipv6->saddr.s6_addr32[1] ^
6645 hdr.ipv6->saddr.s6_addr32[2] ^
6646 hdr.ipv6->saddr.s6_addr32[3] ^
6647 hdr.ipv6->daddr.s6_addr32[0] ^
6648 hdr.ipv6->daddr.s6_addr32[1] ^
6649 hdr.ipv6->daddr.s6_addr32[2] ^
6650 hdr.ipv6->daddr.s6_addr32[3];
6653 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6654 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6655 input, common, ring->queue_index);
6658 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6660 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6661 /* Herbert's original patch had:
6662 * smp_mb__after_netif_stop_queue();
6663 * but since that doesn't exist yet, just open code it. */
6664 smp_mb();
6666 /* We need to check again in a case another CPU has just
6667 * made room available. */
6668 if (likely(ixgbe_desc_unused(tx_ring) < size))
6669 return -EBUSY;
6671 /* A reprieve! - use start_queue because it doesn't call schedule */
6672 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6673 ++tx_ring->tx_stats.restart_queue;
6674 return 0;
6677 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6679 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6680 return 0;
6681 return __ixgbe_maybe_stop_tx(tx_ring, size);
6684 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6686 struct ixgbe_adapter *adapter = netdev_priv(dev);
6687 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6688 smp_processor_id();
6689 #ifdef IXGBE_FCOE
6690 __be16 protocol = vlan_get_protocol(skb);
6692 if (((protocol == htons(ETH_P_FCOE)) ||
6693 (protocol == htons(ETH_P_FIP))) &&
6694 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6695 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6696 txq += adapter->ring_feature[RING_F_FCOE].mask;
6697 return txq;
6699 #endif
6701 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6702 while (unlikely(txq >= dev->real_num_tx_queues))
6703 txq -= dev->real_num_tx_queues;
6704 return txq;
6707 return skb_tx_hash(dev, skb);
6710 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6711 struct ixgbe_adapter *adapter,
6712 struct ixgbe_ring *tx_ring)
6714 struct ixgbe_tx_buffer *first;
6715 int tso;
6716 u32 tx_flags = 0;
6717 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6718 unsigned short f;
6719 #endif
6720 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6721 __be16 protocol = skb->protocol;
6722 u8 hdr_len = 0;
6725 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6726 * + 1 desc for skb_head_len/IXGBE_MAX_DATA_PER_TXD,
6727 * + 2 desc gap to keep tail from touching head,
6728 * + 1 desc for context descriptor,
6729 * otherwise try next time
6731 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6732 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6733 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6734 #else
6735 count += skb_shinfo(skb)->nr_frags;
6736 #endif
6737 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6738 tx_ring->tx_stats.tx_busy++;
6739 return NETDEV_TX_BUSY;
6742 #ifdef CONFIG_PCI_IOV
6743 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6744 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6746 #endif
6747 /* if we have a HW VLAN tag being added default to the HW one */
6748 if (vlan_tx_tag_present(skb)) {
6749 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6750 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6751 /* else if it is a SW VLAN check the next protocol and store the tag */
6752 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6753 struct vlan_hdr *vhdr, _vhdr;
6754 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6755 if (!vhdr)
6756 goto out_drop;
6758 protocol = vhdr->h_vlan_encapsulated_proto;
6759 tx_flags |= ntohs(vhdr->h_vlan_TCI) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6760 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6763 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6764 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6765 (skb->priority != TC_PRIO_CONTROL))) {
6766 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6767 tx_flags |= tx_ring->dcb_tc <<
6768 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6769 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6770 struct vlan_ethhdr *vhdr;
6771 if (skb_header_cloned(skb) &&
6772 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6773 goto out_drop;
6774 vhdr = (struct vlan_ethhdr *)skb->data;
6775 vhdr->h_vlan_TCI = htons(tx_flags >>
6776 IXGBE_TX_FLAGS_VLAN_SHIFT);
6777 } else {
6778 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6782 /* record the location of the first descriptor for this packet */
6783 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6785 #ifdef IXGBE_FCOE
6786 /* setup tx offload for FCoE */
6787 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6788 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6789 tso = ixgbe_fso(tx_ring, skb, tx_flags, &hdr_len);
6790 if (tso < 0)
6791 goto out_drop;
6792 else if (tso)
6793 tx_flags |= IXGBE_TX_FLAGS_FSO |
6794 IXGBE_TX_FLAGS_FCOE;
6795 else
6796 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6798 goto xmit_fcoe;
6801 #endif /* IXGBE_FCOE */
6802 /* setup IPv4/IPv6 offloads */
6803 if (protocol == __constant_htons(ETH_P_IP))
6804 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6806 tso = ixgbe_tso(tx_ring, skb, tx_flags, protocol, &hdr_len);
6807 if (tso < 0)
6808 goto out_drop;
6809 else if (tso)
6810 tx_flags |= IXGBE_TX_FLAGS_TSO;
6811 else if (ixgbe_tx_csum(tx_ring, skb, tx_flags, protocol))
6812 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6814 /* add the ATR filter if ATR is on */
6815 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6816 ixgbe_atr(tx_ring, skb, tx_flags, protocol);
6818 #ifdef IXGBE_FCOE
6819 xmit_fcoe:
6820 #endif /* IXGBE_FCOE */
6821 ixgbe_tx_map(tx_ring, skb, first, tx_flags, hdr_len);
6823 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6825 return NETDEV_TX_OK;
6827 out_drop:
6828 dev_kfree_skb_any(skb);
6829 return NETDEV_TX_OK;
6832 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6834 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6835 struct ixgbe_ring *tx_ring;
6837 tx_ring = adapter->tx_ring[skb->queue_mapping];
6838 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6842 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6843 * @netdev: network interface device structure
6844 * @p: pointer to an address structure
6846 * Returns 0 on success, negative on failure
6848 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6850 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6851 struct ixgbe_hw *hw = &adapter->hw;
6852 struct sockaddr *addr = p;
6854 if (!is_valid_ether_addr(addr->sa_data))
6855 return -EADDRNOTAVAIL;
6857 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6858 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6860 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6861 IXGBE_RAH_AV);
6863 return 0;
6866 static int
6867 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6869 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6870 struct ixgbe_hw *hw = &adapter->hw;
6871 u16 value;
6872 int rc;
6874 if (prtad != hw->phy.mdio.prtad)
6875 return -EINVAL;
6876 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6877 if (!rc)
6878 rc = value;
6879 return rc;
6882 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6883 u16 addr, u16 value)
6885 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6886 struct ixgbe_hw *hw = &adapter->hw;
6888 if (prtad != hw->phy.mdio.prtad)
6889 return -EINVAL;
6890 return hw->phy.ops.write_reg(hw, addr, devad, value);
6893 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6895 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6897 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6901 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6902 * netdev->dev_addrs
6903 * @netdev: network interface device structure
6905 * Returns non-zero on failure
6907 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6909 int err = 0;
6910 struct ixgbe_adapter *adapter = netdev_priv(dev);
6911 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6913 if (is_valid_ether_addr(mac->san_addr)) {
6914 rtnl_lock();
6915 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6916 rtnl_unlock();
6918 return err;
6922 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6923 * netdev->dev_addrs
6924 * @netdev: network interface device structure
6926 * Returns non-zero on failure
6928 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6930 int err = 0;
6931 struct ixgbe_adapter *adapter = netdev_priv(dev);
6932 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6934 if (is_valid_ether_addr(mac->san_addr)) {
6935 rtnl_lock();
6936 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6937 rtnl_unlock();
6939 return err;
6942 #ifdef CONFIG_NET_POLL_CONTROLLER
6944 * Polling 'interrupt' - used by things like netconsole to send skbs
6945 * without having to re-enable interrupts. It's not called while
6946 * the interrupt routine is executing.
6948 static void ixgbe_netpoll(struct net_device *netdev)
6950 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6951 int i;
6953 /* if interface is down do nothing */
6954 if (test_bit(__IXGBE_DOWN, &adapter->state))
6955 return;
6957 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6958 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6959 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6960 for (i = 0; i < num_q_vectors; i++) {
6961 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6962 ixgbe_msix_clean_many(0, q_vector);
6964 } else {
6965 ixgbe_intr(adapter->pdev->irq, netdev);
6967 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6969 #endif
6971 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6972 struct rtnl_link_stats64 *stats)
6974 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6975 int i;
6977 rcu_read_lock();
6978 for (i = 0; i < adapter->num_rx_queues; i++) {
6979 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6980 u64 bytes, packets;
6981 unsigned int start;
6983 if (ring) {
6984 do {
6985 start = u64_stats_fetch_begin_bh(&ring->syncp);
6986 packets = ring->stats.packets;
6987 bytes = ring->stats.bytes;
6988 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6989 stats->rx_packets += packets;
6990 stats->rx_bytes += bytes;
6994 for (i = 0; i < adapter->num_tx_queues; i++) {
6995 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6996 u64 bytes, packets;
6997 unsigned int start;
6999 if (ring) {
7000 do {
7001 start = u64_stats_fetch_begin_bh(&ring->syncp);
7002 packets = ring->stats.packets;
7003 bytes = ring->stats.bytes;
7004 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7005 stats->tx_packets += packets;
7006 stats->tx_bytes += bytes;
7009 rcu_read_unlock();
7010 /* following stats updated by ixgbe_watchdog_task() */
7011 stats->multicast = netdev->stats.multicast;
7012 stats->rx_errors = netdev->stats.rx_errors;
7013 stats->rx_length_errors = netdev->stats.rx_length_errors;
7014 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
7015 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
7016 return stats;
7019 /* ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7020 * #adapter: pointer to ixgbe_adapter
7021 * @tc: number of traffic classes currently enabled
7023 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7024 * 802.1Q priority maps to a packet buffer that exists.
7026 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
7028 struct ixgbe_hw *hw = &adapter->hw;
7029 u32 reg, rsave;
7030 int i;
7032 /* 82598 have a static priority to TC mapping that can not
7033 * be changed so no validation is needed.
7035 if (hw->mac.type == ixgbe_mac_82598EB)
7036 return;
7038 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
7039 rsave = reg;
7041 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
7042 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
7044 /* If up2tc is out of bounds default to zero */
7045 if (up2tc > tc)
7046 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
7049 if (reg != rsave)
7050 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
7052 return;
7056 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
7057 * classes.
7059 * @netdev: net device to configure
7060 * @tc: number of traffic classes to enable
7062 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7064 struct ixgbe_adapter *adapter = netdev_priv(dev);
7065 struct ixgbe_hw *hw = &adapter->hw;
7067 /* Multiple traffic classes requires multiple queues */
7068 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7069 e_err(drv, "Enable failed, needs MSI-X\n");
7070 return -EINVAL;
7073 /* Hardware supports up to 8 traffic classes */
7074 if (tc > MAX_TRAFFIC_CLASS ||
7075 (hw->mac.type == ixgbe_mac_82598EB && tc < MAX_TRAFFIC_CLASS))
7076 return -EINVAL;
7078 /* Hardware has to reinitialize queues and interrupts to
7079 * match packet buffer alignment. Unfortunantly, the
7080 * hardware is not flexible enough to do this dynamically.
7082 if (netif_running(dev))
7083 ixgbe_close(dev);
7084 ixgbe_clear_interrupt_scheme(adapter);
7086 if (tc) {
7087 netdev_set_num_tc(dev, tc);
7088 adapter->last_lfc_mode = adapter->hw.fc.current_mode;
7090 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7091 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7093 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7094 adapter->hw.fc.requested_mode = ixgbe_fc_none;
7095 } else {
7096 netdev_reset_tc(dev);
7098 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7100 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7101 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7103 adapter->temp_dcb_cfg.pfc_mode_enable = false;
7104 adapter->dcb_cfg.pfc_mode_enable = false;
7107 ixgbe_init_interrupt_scheme(adapter);
7108 ixgbe_validate_rtr(adapter, tc);
7109 if (netif_running(dev))
7110 ixgbe_open(dev);
7112 return 0;
7115 void ixgbe_do_reset(struct net_device *netdev)
7117 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7119 if (netif_running(netdev))
7120 ixgbe_reinit_locked(adapter);
7121 else
7122 ixgbe_reset(adapter);
7125 static u32 ixgbe_fix_features(struct net_device *netdev, u32 data)
7127 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7129 #ifdef CONFIG_DCB
7130 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
7131 data &= ~NETIF_F_HW_VLAN_RX;
7132 #endif
7134 /* return error if RXHASH is being enabled when RSS is not supported */
7135 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
7136 data &= ~NETIF_F_RXHASH;
7138 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7139 if (!(data & NETIF_F_RXCSUM))
7140 data &= ~NETIF_F_LRO;
7142 /* Turn off LRO if not RSC capable or invalid ITR settings */
7143 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)) {
7144 data &= ~NETIF_F_LRO;
7145 } else if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
7146 (adapter->rx_itr_setting != 1 &&
7147 adapter->rx_itr_setting > IXGBE_MAX_RSC_INT_RATE)) {
7148 data &= ~NETIF_F_LRO;
7149 e_info(probe, "rx-usecs set too low, not enabling RSC\n");
7152 return data;
7155 static int ixgbe_set_features(struct net_device *netdev, u32 data)
7157 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7158 bool need_reset = false;
7160 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7161 if (!(data & NETIF_F_RXCSUM))
7162 adapter->flags &= ~IXGBE_FLAG_RX_CSUM_ENABLED;
7163 else
7164 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
7166 /* Make sure RSC matches LRO, reset if change */
7167 if (!!(data & NETIF_F_LRO) !=
7168 !!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7169 adapter->flags2 ^= IXGBE_FLAG2_RSC_ENABLED;
7170 switch (adapter->hw.mac.type) {
7171 case ixgbe_mac_X540:
7172 case ixgbe_mac_82599EB:
7173 need_reset = true;
7174 break;
7175 default:
7176 break;
7181 * Check if Flow Director n-tuple support was enabled or disabled. If
7182 * the state changed, we need to reset.
7184 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
7185 /* turn off ATR, enable perfect filters and reset */
7186 if (data & NETIF_F_NTUPLE) {
7187 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7188 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7189 need_reset = true;
7191 } else if (!(data & NETIF_F_NTUPLE)) {
7192 /* turn off Flow Director, set ATR and reset */
7193 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7194 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
7195 !(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
7196 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7197 need_reset = true;
7200 if (need_reset)
7201 ixgbe_do_reset(netdev);
7203 return 0;
7207 static const struct net_device_ops ixgbe_netdev_ops = {
7208 .ndo_open = ixgbe_open,
7209 .ndo_stop = ixgbe_close,
7210 .ndo_start_xmit = ixgbe_xmit_frame,
7211 .ndo_select_queue = ixgbe_select_queue,
7212 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7213 .ndo_validate_addr = eth_validate_addr,
7214 .ndo_set_mac_address = ixgbe_set_mac,
7215 .ndo_change_mtu = ixgbe_change_mtu,
7216 .ndo_tx_timeout = ixgbe_tx_timeout,
7217 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7218 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7219 .ndo_do_ioctl = ixgbe_ioctl,
7220 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7221 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7222 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7223 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7224 .ndo_get_stats64 = ixgbe_get_stats64,
7225 .ndo_setup_tc = ixgbe_setup_tc,
7226 #ifdef CONFIG_NET_POLL_CONTROLLER
7227 .ndo_poll_controller = ixgbe_netpoll,
7228 #endif
7229 #ifdef IXGBE_FCOE
7230 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7231 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7232 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7233 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7234 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7235 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7236 #endif /* IXGBE_FCOE */
7237 .ndo_set_features = ixgbe_set_features,
7238 .ndo_fix_features = ixgbe_fix_features,
7241 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7242 const struct ixgbe_info *ii)
7244 #ifdef CONFIG_PCI_IOV
7245 struct ixgbe_hw *hw = &adapter->hw;
7246 int err;
7247 int num_vf_macvlans, i;
7248 struct vf_macvlans *mv_list;
7250 if (hw->mac.type == ixgbe_mac_82598EB || !max_vfs)
7251 return;
7253 /* The 82599 supports up to 64 VFs per physical function
7254 * but this implementation limits allocation to 63 so that
7255 * basic networking resources are still available to the
7256 * physical function
7258 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7259 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
7260 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
7261 if (err) {
7262 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
7263 goto err_novfs;
7266 num_vf_macvlans = hw->mac.num_rar_entries -
7267 (IXGBE_MAX_PF_MACVLANS + 1 + adapter->num_vfs);
7269 adapter->mv_list = mv_list = kcalloc(num_vf_macvlans,
7270 sizeof(struct vf_macvlans),
7271 GFP_KERNEL);
7272 if (mv_list) {
7273 /* Initialize list of VF macvlans */
7274 INIT_LIST_HEAD(&adapter->vf_mvs.l);
7275 for (i = 0; i < num_vf_macvlans; i++) {
7276 mv_list->vf = -1;
7277 mv_list->free = true;
7278 mv_list->rar_entry = hw->mac.num_rar_entries -
7279 (i + adapter->num_vfs + 1);
7280 list_add(&mv_list->l, &adapter->vf_mvs.l);
7281 mv_list++;
7285 /* If call to enable VFs succeeded then allocate memory
7286 * for per VF control structures.
7288 adapter->vfinfo =
7289 kcalloc(adapter->num_vfs,
7290 sizeof(struct vf_data_storage), GFP_KERNEL);
7291 if (adapter->vfinfo) {
7292 /* Now that we're sure SR-IOV is enabled
7293 * and memory allocated set up the mailbox parameters
7295 ixgbe_init_mbx_params_pf(hw);
7296 memcpy(&hw->mbx.ops, ii->mbx_ops,
7297 sizeof(hw->mbx.ops));
7299 /* Disable RSC when in SR-IOV mode */
7300 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
7301 IXGBE_FLAG2_RSC_ENABLED);
7302 return;
7305 /* Oh oh */
7306 e_err(probe, "Unable to allocate memory for VF Data Storage - "
7307 "SRIOV disabled\n");
7308 pci_disable_sriov(adapter->pdev);
7310 err_novfs:
7311 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
7312 adapter->num_vfs = 0;
7313 #endif /* CONFIG_PCI_IOV */
7317 * ixgbe_probe - Device Initialization Routine
7318 * @pdev: PCI device information struct
7319 * @ent: entry in ixgbe_pci_tbl
7321 * Returns 0 on success, negative on failure
7323 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7324 * The OS initialization, configuring of the adapter private structure,
7325 * and a hardware reset occur.
7327 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7328 const struct pci_device_id *ent)
7330 struct net_device *netdev;
7331 struct ixgbe_adapter *adapter = NULL;
7332 struct ixgbe_hw *hw;
7333 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7334 static int cards_found;
7335 int i, err, pci_using_dac;
7336 u8 part_str[IXGBE_PBANUM_LENGTH];
7337 unsigned int indices = num_possible_cpus();
7338 #ifdef IXGBE_FCOE
7339 u16 device_caps;
7340 #endif
7341 u32 eec;
7343 /* Catch broken hardware that put the wrong VF device ID in
7344 * the PCIe SR-IOV capability.
7346 if (pdev->is_virtfn) {
7347 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7348 pci_name(pdev), pdev->vendor, pdev->device);
7349 return -EINVAL;
7352 err = pci_enable_device_mem(pdev);
7353 if (err)
7354 return err;
7356 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7357 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7358 pci_using_dac = 1;
7359 } else {
7360 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7361 if (err) {
7362 err = dma_set_coherent_mask(&pdev->dev,
7363 DMA_BIT_MASK(32));
7364 if (err) {
7365 dev_err(&pdev->dev,
7366 "No usable DMA configuration, aborting\n");
7367 goto err_dma;
7370 pci_using_dac = 0;
7373 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7374 IORESOURCE_MEM), ixgbe_driver_name);
7375 if (err) {
7376 dev_err(&pdev->dev,
7377 "pci_request_selected_regions failed 0x%x\n", err);
7378 goto err_pci_reg;
7381 pci_enable_pcie_error_reporting(pdev);
7383 pci_set_master(pdev);
7384 pci_save_state(pdev);
7386 #ifdef CONFIG_IXGBE_DCB
7387 indices *= MAX_TRAFFIC_CLASS;
7388 #endif
7390 if (ii->mac == ixgbe_mac_82598EB)
7391 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7392 else
7393 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7395 #ifdef IXGBE_FCOE
7396 indices += min_t(unsigned int, num_possible_cpus(),
7397 IXGBE_MAX_FCOE_INDICES);
7398 #endif
7399 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7400 if (!netdev) {
7401 err = -ENOMEM;
7402 goto err_alloc_etherdev;
7405 SET_NETDEV_DEV(netdev, &pdev->dev);
7407 adapter = netdev_priv(netdev);
7408 pci_set_drvdata(pdev, adapter);
7410 adapter->netdev = netdev;
7411 adapter->pdev = pdev;
7412 hw = &adapter->hw;
7413 hw->back = adapter;
7414 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7416 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7417 pci_resource_len(pdev, 0));
7418 if (!hw->hw_addr) {
7419 err = -EIO;
7420 goto err_ioremap;
7423 for (i = 1; i <= 5; i++) {
7424 if (pci_resource_len(pdev, i) == 0)
7425 continue;
7428 netdev->netdev_ops = &ixgbe_netdev_ops;
7429 ixgbe_set_ethtool_ops(netdev);
7430 netdev->watchdog_timeo = 5 * HZ;
7431 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7433 adapter->bd_number = cards_found;
7435 /* Setup hw api */
7436 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7437 hw->mac.type = ii->mac;
7439 /* EEPROM */
7440 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7441 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7442 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7443 if (!(eec & (1 << 8)))
7444 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7446 /* PHY */
7447 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7448 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7449 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7450 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7451 hw->phy.mdio.mmds = 0;
7452 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7453 hw->phy.mdio.dev = netdev;
7454 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7455 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7457 ii->get_invariants(hw);
7459 /* setup the private structure */
7460 err = ixgbe_sw_init(adapter);
7461 if (err)
7462 goto err_sw_init;
7464 /* Make it possible the adapter to be woken up via WOL */
7465 switch (adapter->hw.mac.type) {
7466 case ixgbe_mac_82599EB:
7467 case ixgbe_mac_X540:
7468 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7469 break;
7470 default:
7471 break;
7475 * If there is a fan on this device and it has failed log the
7476 * failure.
7478 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7479 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7480 if (esdp & IXGBE_ESDP_SDP1)
7481 e_crit(probe, "Fan has stopped, replace the adapter\n");
7484 /* reset_hw fills in the perm_addr as well */
7485 hw->phy.reset_if_overtemp = true;
7486 err = hw->mac.ops.reset_hw(hw);
7487 hw->phy.reset_if_overtemp = false;
7488 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7489 hw->mac.type == ixgbe_mac_82598EB) {
7490 err = 0;
7491 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7492 e_dev_err("failed to load because an unsupported SFP+ "
7493 "module type was detected.\n");
7494 e_dev_err("Reload the driver after installing a supported "
7495 "module.\n");
7496 goto err_sw_init;
7497 } else if (err) {
7498 e_dev_err("HW Init failed: %d\n", err);
7499 goto err_sw_init;
7502 ixgbe_probe_vf(adapter, ii);
7504 netdev->features = NETIF_F_SG |
7505 NETIF_F_IP_CSUM |
7506 NETIF_F_IPV6_CSUM |
7507 NETIF_F_HW_VLAN_TX |
7508 NETIF_F_HW_VLAN_RX |
7509 NETIF_F_HW_VLAN_FILTER |
7510 NETIF_F_TSO |
7511 NETIF_F_TSO6 |
7512 NETIF_F_RXHASH |
7513 NETIF_F_RXCSUM;
7515 netdev->hw_features = netdev->features;
7517 switch (adapter->hw.mac.type) {
7518 case ixgbe_mac_82599EB:
7519 case ixgbe_mac_X540:
7520 netdev->features |= NETIF_F_SCTP_CSUM;
7521 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7522 NETIF_F_NTUPLE;
7523 break;
7524 default:
7525 break;
7528 netdev->vlan_features |= NETIF_F_TSO;
7529 netdev->vlan_features |= NETIF_F_TSO6;
7530 netdev->vlan_features |= NETIF_F_IP_CSUM;
7531 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7532 netdev->vlan_features |= NETIF_F_SG;
7534 netdev->priv_flags |= IFF_UNICAST_FLT;
7536 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7537 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7538 IXGBE_FLAG_DCB_ENABLED);
7540 #ifdef CONFIG_IXGBE_DCB
7541 netdev->dcbnl_ops = &dcbnl_ops;
7542 #endif
7544 #ifdef IXGBE_FCOE
7545 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7546 if (hw->mac.ops.get_device_caps) {
7547 hw->mac.ops.get_device_caps(hw, &device_caps);
7548 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7549 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7552 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7553 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7554 netdev->vlan_features |= NETIF_F_FSO;
7555 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7557 #endif /* IXGBE_FCOE */
7558 if (pci_using_dac) {
7559 netdev->features |= NETIF_F_HIGHDMA;
7560 netdev->vlan_features |= NETIF_F_HIGHDMA;
7563 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7564 netdev->hw_features |= NETIF_F_LRO;
7565 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7566 netdev->features |= NETIF_F_LRO;
7568 /* make sure the EEPROM is good */
7569 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7570 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7571 err = -EIO;
7572 goto err_eeprom;
7575 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7576 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7578 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7579 e_dev_err("invalid MAC address\n");
7580 err = -EIO;
7581 goto err_eeprom;
7584 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7585 if (hw->mac.ops.disable_tx_laser &&
7586 ((hw->phy.multispeed_fiber) ||
7587 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7588 (hw->mac.type == ixgbe_mac_82599EB))))
7589 hw->mac.ops.disable_tx_laser(hw);
7591 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7592 (unsigned long) adapter);
7594 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7595 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7597 err = ixgbe_init_interrupt_scheme(adapter);
7598 if (err)
7599 goto err_sw_init;
7601 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7602 netdev->hw_features &= ~NETIF_F_RXHASH;
7603 netdev->features &= ~NETIF_F_RXHASH;
7606 switch (pdev->device) {
7607 case IXGBE_DEV_ID_82599_SFP:
7608 /* Only this subdevice supports WOL */
7609 if (pdev->subsystem_device == IXGBE_SUBDEV_ID_82599_SFP)
7610 adapter->wol = IXGBE_WUFC_MAG;
7611 break;
7612 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7613 /* All except this subdevice support WOL */
7614 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7615 adapter->wol = IXGBE_WUFC_MAG;
7616 break;
7617 case IXGBE_DEV_ID_82599_KX4:
7618 adapter->wol = IXGBE_WUFC_MAG;
7619 break;
7620 default:
7621 adapter->wol = 0;
7622 break;
7624 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7626 /* pick up the PCI bus settings for reporting later */
7627 hw->mac.ops.get_bus_info(hw);
7629 /* print bus type/speed/width info */
7630 e_dev_info("(PCI Express:%s:%s) %pM\n",
7631 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7632 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7633 "Unknown"),
7634 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7635 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7636 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7637 "Unknown"),
7638 netdev->dev_addr);
7640 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7641 if (err)
7642 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7643 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7644 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7645 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7646 part_str);
7647 else
7648 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7649 hw->mac.type, hw->phy.type, part_str);
7651 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7652 e_dev_warn("PCI-Express bandwidth available for this card is "
7653 "not sufficient for optimal performance.\n");
7654 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7655 "is required.\n");
7658 /* save off EEPROM version number */
7659 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7661 /* reset the hardware with the new settings */
7662 err = hw->mac.ops.start_hw(hw);
7664 if (err == IXGBE_ERR_EEPROM_VERSION) {
7665 /* We are running on a pre-production device, log a warning */
7666 e_dev_warn("This device is a pre-production adapter/LOM. "
7667 "Please be aware there may be issues associated "
7668 "with your hardware. If you are experiencing "
7669 "problems please contact your Intel or hardware "
7670 "representative who provided you with this "
7671 "hardware.\n");
7673 strcpy(netdev->name, "eth%d");
7674 err = register_netdev(netdev);
7675 if (err)
7676 goto err_register;
7678 /* carrier off reporting is important to ethtool even BEFORE open */
7679 netif_carrier_off(netdev);
7681 #ifdef CONFIG_IXGBE_DCA
7682 if (dca_add_requester(&pdev->dev) == 0) {
7683 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7684 ixgbe_setup_dca(adapter);
7686 #endif
7687 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7688 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7689 for (i = 0; i < adapter->num_vfs; i++)
7690 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7693 /* Inform firmware of driver version */
7694 if (hw->mac.ops.set_fw_drv_ver)
7695 hw->mac.ops.set_fw_drv_ver(hw, MAJ, MIN, BUILD,
7696 FW_CEM_UNUSED_VER);
7698 /* add san mac addr to netdev */
7699 ixgbe_add_sanmac_netdev(netdev);
7701 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7702 cards_found++;
7703 return 0;
7705 err_register:
7706 ixgbe_release_hw_control(adapter);
7707 ixgbe_clear_interrupt_scheme(adapter);
7708 err_sw_init:
7709 err_eeprom:
7710 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7711 ixgbe_disable_sriov(adapter);
7712 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7713 iounmap(hw->hw_addr);
7714 err_ioremap:
7715 free_netdev(netdev);
7716 err_alloc_etherdev:
7717 pci_release_selected_regions(pdev,
7718 pci_select_bars(pdev, IORESOURCE_MEM));
7719 err_pci_reg:
7720 err_dma:
7721 pci_disable_device(pdev);
7722 return err;
7726 * ixgbe_remove - Device Removal Routine
7727 * @pdev: PCI device information struct
7729 * ixgbe_remove is called by the PCI subsystem to alert the driver
7730 * that it should release a PCI device. The could be caused by a
7731 * Hot-Plug event, or because the driver is going to be removed from
7732 * memory.
7734 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7736 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7737 struct net_device *netdev = adapter->netdev;
7739 set_bit(__IXGBE_DOWN, &adapter->state);
7740 cancel_work_sync(&adapter->service_task);
7742 #ifdef CONFIG_IXGBE_DCA
7743 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7744 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7745 dca_remove_requester(&pdev->dev);
7746 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7749 #endif
7750 #ifdef IXGBE_FCOE
7751 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7752 ixgbe_cleanup_fcoe(adapter);
7754 #endif /* IXGBE_FCOE */
7756 /* remove the added san mac */
7757 ixgbe_del_sanmac_netdev(netdev);
7759 if (netdev->reg_state == NETREG_REGISTERED)
7760 unregister_netdev(netdev);
7762 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7763 ixgbe_disable_sriov(adapter);
7765 ixgbe_clear_interrupt_scheme(adapter);
7767 ixgbe_release_hw_control(adapter);
7769 iounmap(adapter->hw.hw_addr);
7770 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7771 IORESOURCE_MEM));
7773 e_dev_info("complete\n");
7775 free_netdev(netdev);
7777 pci_disable_pcie_error_reporting(pdev);
7779 pci_disable_device(pdev);
7783 * ixgbe_io_error_detected - called when PCI error is detected
7784 * @pdev: Pointer to PCI device
7785 * @state: The current pci connection state
7787 * This function is called after a PCI bus error affecting
7788 * this device has been detected.
7790 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7791 pci_channel_state_t state)
7793 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7794 struct net_device *netdev = adapter->netdev;
7796 netif_device_detach(netdev);
7798 if (state == pci_channel_io_perm_failure)
7799 return PCI_ERS_RESULT_DISCONNECT;
7801 if (netif_running(netdev))
7802 ixgbe_down(adapter);
7803 pci_disable_device(pdev);
7805 /* Request a slot reset. */
7806 return PCI_ERS_RESULT_NEED_RESET;
7810 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7811 * @pdev: Pointer to PCI device
7813 * Restart the card from scratch, as if from a cold-boot.
7815 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7817 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7818 pci_ers_result_t result;
7819 int err;
7821 if (pci_enable_device_mem(pdev)) {
7822 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7823 result = PCI_ERS_RESULT_DISCONNECT;
7824 } else {
7825 pci_set_master(pdev);
7826 pci_restore_state(pdev);
7827 pci_save_state(pdev);
7829 pci_wake_from_d3(pdev, false);
7831 ixgbe_reset(adapter);
7832 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7833 result = PCI_ERS_RESULT_RECOVERED;
7836 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7837 if (err) {
7838 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7839 "failed 0x%0x\n", err);
7840 /* non-fatal, continue */
7843 return result;
7847 * ixgbe_io_resume - called when traffic can start flowing again.
7848 * @pdev: Pointer to PCI device
7850 * This callback is called when the error recovery driver tells us that
7851 * its OK to resume normal operation.
7853 static void ixgbe_io_resume(struct pci_dev *pdev)
7855 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7856 struct net_device *netdev = adapter->netdev;
7858 if (netif_running(netdev)) {
7859 if (ixgbe_up(adapter)) {
7860 e_info(probe, "ixgbe_up failed after reset\n");
7861 return;
7865 netif_device_attach(netdev);
7868 static struct pci_error_handlers ixgbe_err_handler = {
7869 .error_detected = ixgbe_io_error_detected,
7870 .slot_reset = ixgbe_io_slot_reset,
7871 .resume = ixgbe_io_resume,
7874 static struct pci_driver ixgbe_driver = {
7875 .name = ixgbe_driver_name,
7876 .id_table = ixgbe_pci_tbl,
7877 .probe = ixgbe_probe,
7878 .remove = __devexit_p(ixgbe_remove),
7879 #ifdef CONFIG_PM
7880 .suspend = ixgbe_suspend,
7881 .resume = ixgbe_resume,
7882 #endif
7883 .shutdown = ixgbe_shutdown,
7884 .err_handler = &ixgbe_err_handler
7888 * ixgbe_init_module - Driver Registration Routine
7890 * ixgbe_init_module is the first routine called when the driver is
7891 * loaded. All it does is register with the PCI subsystem.
7893 static int __init ixgbe_init_module(void)
7895 int ret;
7896 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7897 pr_info("%s\n", ixgbe_copyright);
7899 #ifdef CONFIG_IXGBE_DCA
7900 dca_register_notify(&dca_notifier);
7901 #endif
7903 ret = pci_register_driver(&ixgbe_driver);
7904 return ret;
7907 module_init(ixgbe_init_module);
7910 * ixgbe_exit_module - Driver Exit Cleanup Routine
7912 * ixgbe_exit_module is called just before the driver is removed
7913 * from memory.
7915 static void __exit ixgbe_exit_module(void)
7917 #ifdef CONFIG_IXGBE_DCA
7918 dca_unregister_notify(&dca_notifier);
7919 #endif
7920 pci_unregister_driver(&ixgbe_driver);
7921 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7924 #ifdef CONFIG_IXGBE_DCA
7925 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7926 void *p)
7928 int ret_val;
7930 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7931 __ixgbe_notify_dca);
7933 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7936 #endif /* CONFIG_IXGBE_DCA */
7938 module_exit(ixgbe_exit_module);
7940 /* ixgbe_main.c */