ixgbe: move all Rx DMA control register writes to one central location
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / net / ixgbe / ixgbe_main.c
blobf8cdc992f24de990d5f3b868dc17bce56ae8c673
1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2010 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/ip.h>
36 #include <linux/tcp.h>
37 #include <linux/pkt_sched.h>
38 #include <linux/ipv6.h>
39 #include <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <scsi/fc/fc_fcoe.h>
46 #include "ixgbe.h"
47 #include "ixgbe_common.h"
48 #include "ixgbe_dcb_82599.h"
49 #include "ixgbe_sriov.h"
51 char ixgbe_driver_name[] = "ixgbe";
52 static const char ixgbe_driver_string[] =
53 "Intel(R) 10 Gigabit PCI Express Network Driver";
55 #define DRV_VERSION "2.0.84-k2"
56 const char ixgbe_driver_version[] = DRV_VERSION;
57 static char ixgbe_copyright[] = "Copyright (c) 1999-2010 Intel Corporation.";
59 static const struct ixgbe_info *ixgbe_info_tbl[] = {
60 [board_82598] = &ixgbe_82598_info,
61 [board_82599] = &ixgbe_82599_info,
64 /* ixgbe_pci_tbl - PCI Device ID Table
66 * Wildcard entries (PCI_ANY_ID) should come last
67 * Last entry must be all 0s
69 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
70 * Class, Class Mask, private data (not used) }
72 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
73 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
74 board_82598 },
75 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
76 board_82598 },
77 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
78 board_82598 },
79 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
80 board_82598 },
81 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
82 board_82598 },
83 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
84 board_82598 },
85 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
86 board_82598 },
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
88 board_82598 },
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
90 board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
92 board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
94 board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
96 board_82598 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
98 board_82599 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
100 board_82599 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
102 board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
104 board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
106 board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
108 board_82599 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
110 board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM),
112 board_82599 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
114 board_82599 },
116 /* required last entry */
117 {0, }
119 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
121 #ifdef CONFIG_IXGBE_DCA
122 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
123 void *p);
124 static struct notifier_block dca_notifier = {
125 .notifier_call = ixgbe_notify_dca,
126 .next = NULL,
127 .priority = 0
129 #endif
131 #ifdef CONFIG_PCI_IOV
132 static unsigned int max_vfs;
133 module_param(max_vfs, uint, 0);
134 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
135 "per physical function");
136 #endif /* CONFIG_PCI_IOV */
138 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
139 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
140 MODULE_LICENSE("GPL");
141 MODULE_VERSION(DRV_VERSION);
143 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
145 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
147 struct ixgbe_hw *hw = &adapter->hw;
148 u32 gcr;
149 u32 gpie;
150 u32 vmdctl;
152 #ifdef CONFIG_PCI_IOV
153 /* disable iov and allow time for transactions to clear */
154 pci_disable_sriov(adapter->pdev);
155 #endif
157 /* turn off device IOV mode */
158 gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
159 gcr &= ~(IXGBE_GCR_EXT_SRIOV);
160 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
161 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
162 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
163 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
165 /* set default pool back to 0 */
166 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
167 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
168 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
170 /* take a breather then clean up driver data */
171 msleep(100);
172 if (adapter->vfinfo)
173 kfree(adapter->vfinfo);
174 adapter->vfinfo = NULL;
176 adapter->num_vfs = 0;
177 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
180 struct ixgbe_reg_info {
181 u32 ofs;
182 char *name;
185 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
187 /* General Registers */
188 {IXGBE_CTRL, "CTRL"},
189 {IXGBE_STATUS, "STATUS"},
190 {IXGBE_CTRL_EXT, "CTRL_EXT"},
192 /* Interrupt Registers */
193 {IXGBE_EICR, "EICR"},
195 /* RX Registers */
196 {IXGBE_SRRCTL(0), "SRRCTL"},
197 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
198 {IXGBE_RDLEN(0), "RDLEN"},
199 {IXGBE_RDH(0), "RDH"},
200 {IXGBE_RDT(0), "RDT"},
201 {IXGBE_RXDCTL(0), "RXDCTL"},
202 {IXGBE_RDBAL(0), "RDBAL"},
203 {IXGBE_RDBAH(0), "RDBAH"},
205 /* TX Registers */
206 {IXGBE_TDBAL(0), "TDBAL"},
207 {IXGBE_TDBAH(0), "TDBAH"},
208 {IXGBE_TDLEN(0), "TDLEN"},
209 {IXGBE_TDH(0), "TDH"},
210 {IXGBE_TDT(0), "TDT"},
211 {IXGBE_TXDCTL(0), "TXDCTL"},
213 /* List Terminator */
219 * ixgbe_regdump - register printout routine
221 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
223 int i = 0, j = 0;
224 char rname[16];
225 u32 regs[64];
227 switch (reginfo->ofs) {
228 case IXGBE_SRRCTL(0):
229 for (i = 0; i < 64; i++)
230 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
231 break;
232 case IXGBE_DCA_RXCTRL(0):
233 for (i = 0; i < 64; i++)
234 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
235 break;
236 case IXGBE_RDLEN(0):
237 for (i = 0; i < 64; i++)
238 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
239 break;
240 case IXGBE_RDH(0):
241 for (i = 0; i < 64; i++)
242 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
243 break;
244 case IXGBE_RDT(0):
245 for (i = 0; i < 64; i++)
246 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
247 break;
248 case IXGBE_RXDCTL(0):
249 for (i = 0; i < 64; i++)
250 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
251 break;
252 case IXGBE_RDBAL(0):
253 for (i = 0; i < 64; i++)
254 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
255 break;
256 case IXGBE_RDBAH(0):
257 for (i = 0; i < 64; i++)
258 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
259 break;
260 case IXGBE_TDBAL(0):
261 for (i = 0; i < 64; i++)
262 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
263 break;
264 case IXGBE_TDBAH(0):
265 for (i = 0; i < 64; i++)
266 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
267 break;
268 case IXGBE_TDLEN(0):
269 for (i = 0; i < 64; i++)
270 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
271 break;
272 case IXGBE_TDH(0):
273 for (i = 0; i < 64; i++)
274 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
275 break;
276 case IXGBE_TDT(0):
277 for (i = 0; i < 64; i++)
278 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
279 break;
280 case IXGBE_TXDCTL(0):
281 for (i = 0; i < 64; i++)
282 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
283 break;
284 default:
285 printk(KERN_INFO "%-15s %08x\n", reginfo->name,
286 IXGBE_READ_REG(hw, reginfo->ofs));
287 return;
290 for (i = 0; i < 8; i++) {
291 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
292 printk(KERN_ERR "%-15s ", rname);
293 for (j = 0; j < 8; j++)
294 printk(KERN_CONT "%08x ", regs[i*8+j]);
295 printk(KERN_CONT "\n");
301 * ixgbe_dump - Print registers, tx-rings and rx-rings
303 static void ixgbe_dump(struct ixgbe_adapter *adapter)
305 struct net_device *netdev = adapter->netdev;
306 struct ixgbe_hw *hw = &adapter->hw;
307 struct ixgbe_reg_info *reginfo;
308 int n = 0;
309 struct ixgbe_ring *tx_ring;
310 struct ixgbe_tx_buffer *tx_buffer_info;
311 union ixgbe_adv_tx_desc *tx_desc;
312 struct my_u0 { u64 a; u64 b; } *u0;
313 struct ixgbe_ring *rx_ring;
314 union ixgbe_adv_rx_desc *rx_desc;
315 struct ixgbe_rx_buffer *rx_buffer_info;
316 u32 staterr;
317 int i = 0;
319 if (!netif_msg_hw(adapter))
320 return;
322 /* Print netdevice Info */
323 if (netdev) {
324 dev_info(&adapter->pdev->dev, "Net device Info\n");
325 printk(KERN_INFO "Device Name state "
326 "trans_start last_rx\n");
327 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
328 netdev->name,
329 netdev->state,
330 netdev->trans_start,
331 netdev->last_rx);
334 /* Print Registers */
335 dev_info(&adapter->pdev->dev, "Register Dump\n");
336 printk(KERN_INFO " Register Name Value\n");
337 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
338 reginfo->name; reginfo++) {
339 ixgbe_regdump(hw, reginfo);
342 /* Print TX Ring Summary */
343 if (!netdev || !netif_running(netdev))
344 goto exit;
346 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
347 printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma ] "
348 "leng ntw timestamp\n");
349 for (n = 0; n < adapter->num_tx_queues; n++) {
350 tx_ring = adapter->tx_ring[n];
351 tx_buffer_info =
352 &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
353 printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
354 n, tx_ring->next_to_use, tx_ring->next_to_clean,
355 (u64)tx_buffer_info->dma,
356 tx_buffer_info->length,
357 tx_buffer_info->next_to_watch,
358 (u64)tx_buffer_info->time_stamp);
361 /* Print TX Rings */
362 if (!netif_msg_tx_done(adapter))
363 goto rx_ring_summary;
365 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
367 /* Transmit Descriptor Formats
369 * Advanced Transmit Descriptor
370 * +--------------------------------------------------------------+
371 * 0 | Buffer Address [63:0] |
372 * +--------------------------------------------------------------+
373 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
374 * +--------------------------------------------------------------+
375 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
378 for (n = 0; n < adapter->num_tx_queues; n++) {
379 tx_ring = adapter->tx_ring[n];
380 printk(KERN_INFO "------------------------------------\n");
381 printk(KERN_INFO "TX QUEUE INDEX = %d\n", tx_ring->queue_index);
382 printk(KERN_INFO "------------------------------------\n");
383 printk(KERN_INFO "T [desc] [address 63:0 ] "
384 "[PlPOIdStDDt Ln] [bi->dma ] "
385 "leng ntw timestamp bi->skb\n");
387 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
388 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
389 tx_buffer_info = &tx_ring->tx_buffer_info[i];
390 u0 = (struct my_u0 *)tx_desc;
391 printk(KERN_INFO "T [0x%03X] %016llX %016llX %016llX"
392 " %04X %3X %016llX %p", i,
393 le64_to_cpu(u0->a),
394 le64_to_cpu(u0->b),
395 (u64)tx_buffer_info->dma,
396 tx_buffer_info->length,
397 tx_buffer_info->next_to_watch,
398 (u64)tx_buffer_info->time_stamp,
399 tx_buffer_info->skb);
400 if (i == tx_ring->next_to_use &&
401 i == tx_ring->next_to_clean)
402 printk(KERN_CONT " NTC/U\n");
403 else if (i == tx_ring->next_to_use)
404 printk(KERN_CONT " NTU\n");
405 else if (i == tx_ring->next_to_clean)
406 printk(KERN_CONT " NTC\n");
407 else
408 printk(KERN_CONT "\n");
410 if (netif_msg_pktdata(adapter) &&
411 tx_buffer_info->dma != 0)
412 print_hex_dump(KERN_INFO, "",
413 DUMP_PREFIX_ADDRESS, 16, 1,
414 phys_to_virt(tx_buffer_info->dma),
415 tx_buffer_info->length, true);
419 /* Print RX Rings Summary */
420 rx_ring_summary:
421 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
422 printk(KERN_INFO "Queue [NTU] [NTC]\n");
423 for (n = 0; n < adapter->num_rx_queues; n++) {
424 rx_ring = adapter->rx_ring[n];
425 printk(KERN_INFO "%5d %5X %5X\n", n,
426 rx_ring->next_to_use, rx_ring->next_to_clean);
429 /* Print RX Rings */
430 if (!netif_msg_rx_status(adapter))
431 goto exit;
433 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
435 /* Advanced Receive Descriptor (Read) Format
436 * 63 1 0
437 * +-----------------------------------------------------+
438 * 0 | Packet Buffer Address [63:1] |A0/NSE|
439 * +----------------------------------------------+------+
440 * 8 | Header Buffer Address [63:1] | DD |
441 * +-----------------------------------------------------+
444 * Advanced Receive Descriptor (Write-Back) Format
446 * 63 48 47 32 31 30 21 20 16 15 4 3 0
447 * +------------------------------------------------------+
448 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
449 * | Checksum Ident | | | | Type | Type |
450 * +------------------------------------------------------+
451 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
452 * +------------------------------------------------------+
453 * 63 48 47 32 31 20 19 0
455 for (n = 0; n < adapter->num_rx_queues; n++) {
456 rx_ring = adapter->rx_ring[n];
457 printk(KERN_INFO "------------------------------------\n");
458 printk(KERN_INFO "RX QUEUE INDEX = %d\n", rx_ring->queue_index);
459 printk(KERN_INFO "------------------------------------\n");
460 printk(KERN_INFO "R [desc] [ PktBuf A0] "
461 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
462 "<-- Adv Rx Read format\n");
463 printk(KERN_INFO "RWB[desc] [PcsmIpSHl PtRs] "
464 "[vl er S cks ln] ---------------- [bi->skb] "
465 "<-- Adv Rx Write-Back format\n");
467 for (i = 0; i < rx_ring->count; i++) {
468 rx_buffer_info = &rx_ring->rx_buffer_info[i];
469 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
470 u0 = (struct my_u0 *)rx_desc;
471 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
472 if (staterr & IXGBE_RXD_STAT_DD) {
473 /* Descriptor Done */
474 printk(KERN_INFO "RWB[0x%03X] %016llX "
475 "%016llX ---------------- %p", i,
476 le64_to_cpu(u0->a),
477 le64_to_cpu(u0->b),
478 rx_buffer_info->skb);
479 } else {
480 printk(KERN_INFO "R [0x%03X] %016llX "
481 "%016llX %016llX %p", i,
482 le64_to_cpu(u0->a),
483 le64_to_cpu(u0->b),
484 (u64)rx_buffer_info->dma,
485 rx_buffer_info->skb);
487 if (netif_msg_pktdata(adapter)) {
488 print_hex_dump(KERN_INFO, "",
489 DUMP_PREFIX_ADDRESS, 16, 1,
490 phys_to_virt(rx_buffer_info->dma),
491 rx_ring->rx_buf_len, true);
493 if (rx_ring->rx_buf_len
494 < IXGBE_RXBUFFER_2048)
495 print_hex_dump(KERN_INFO, "",
496 DUMP_PREFIX_ADDRESS, 16, 1,
497 phys_to_virt(
498 rx_buffer_info->page_dma +
499 rx_buffer_info->page_offset
501 PAGE_SIZE/2, true);
505 if (i == rx_ring->next_to_use)
506 printk(KERN_CONT " NTU\n");
507 else if (i == rx_ring->next_to_clean)
508 printk(KERN_CONT " NTC\n");
509 else
510 printk(KERN_CONT "\n");
515 exit:
516 return;
519 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
521 u32 ctrl_ext;
523 /* Let firmware take over control of h/w */
524 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
525 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
526 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
529 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
531 u32 ctrl_ext;
533 /* Let firmware know the driver has taken over */
534 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
535 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
536 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
540 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
541 * @adapter: pointer to adapter struct
542 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
543 * @queue: queue to map the corresponding interrupt to
544 * @msix_vector: the vector to map to the corresponding queue
547 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
548 u8 queue, u8 msix_vector)
550 u32 ivar, index;
551 struct ixgbe_hw *hw = &adapter->hw;
552 switch (hw->mac.type) {
553 case ixgbe_mac_82598EB:
554 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
555 if (direction == -1)
556 direction = 0;
557 index = (((direction * 64) + queue) >> 2) & 0x1F;
558 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
559 ivar &= ~(0xFF << (8 * (queue & 0x3)));
560 ivar |= (msix_vector << (8 * (queue & 0x3)));
561 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
562 break;
563 case ixgbe_mac_82599EB:
564 if (direction == -1) {
565 /* other causes */
566 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
567 index = ((queue & 1) * 8);
568 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
569 ivar &= ~(0xFF << index);
570 ivar |= (msix_vector << index);
571 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
572 break;
573 } else {
574 /* tx or rx causes */
575 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
576 index = ((16 * (queue & 1)) + (8 * direction));
577 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
578 ivar &= ~(0xFF << index);
579 ivar |= (msix_vector << index);
580 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
581 break;
583 default:
584 break;
588 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
589 u64 qmask)
591 u32 mask;
593 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
594 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
595 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
596 } else {
597 mask = (qmask & 0xFFFFFFFF);
598 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
599 mask = (qmask >> 32);
600 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
604 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
605 struct ixgbe_tx_buffer
606 *tx_buffer_info)
608 if (tx_buffer_info->dma) {
609 if (tx_buffer_info->mapped_as_page)
610 dma_unmap_page(&adapter->pdev->dev,
611 tx_buffer_info->dma,
612 tx_buffer_info->length,
613 DMA_TO_DEVICE);
614 else
615 dma_unmap_single(&adapter->pdev->dev,
616 tx_buffer_info->dma,
617 tx_buffer_info->length,
618 DMA_TO_DEVICE);
619 tx_buffer_info->dma = 0;
621 if (tx_buffer_info->skb) {
622 dev_kfree_skb_any(tx_buffer_info->skb);
623 tx_buffer_info->skb = NULL;
625 tx_buffer_info->time_stamp = 0;
626 /* tx_buffer_info must be completely set up in the transmit path */
630 * ixgbe_tx_xon_state - check the tx ring xon state
631 * @adapter: the ixgbe adapter
632 * @tx_ring: the corresponding tx_ring
634 * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
635 * corresponding TC of this tx_ring when checking TFCS.
637 * Returns : true if in xon state (currently not paused)
639 static inline bool ixgbe_tx_xon_state(struct ixgbe_adapter *adapter,
640 struct ixgbe_ring *tx_ring)
642 u32 txoff = IXGBE_TFCS_TXOFF;
644 #ifdef CONFIG_IXGBE_DCB
645 if (adapter->dcb_cfg.pfc_mode_enable) {
646 int tc;
647 int reg_idx = tx_ring->reg_idx;
648 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
650 switch (adapter->hw.mac.type) {
651 case ixgbe_mac_82598EB:
652 tc = reg_idx >> 2;
653 txoff = IXGBE_TFCS_TXOFF0;
654 break;
655 case ixgbe_mac_82599EB:
656 tc = 0;
657 txoff = IXGBE_TFCS_TXOFF;
658 if (dcb_i == 8) {
659 /* TC0, TC1 */
660 tc = reg_idx >> 5;
661 if (tc == 2) /* TC2, TC3 */
662 tc += (reg_idx - 64) >> 4;
663 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
664 tc += 1 + ((reg_idx - 96) >> 3);
665 } else if (dcb_i == 4) {
666 /* TC0, TC1 */
667 tc = reg_idx >> 6;
668 if (tc == 1) {
669 tc += (reg_idx - 64) >> 5;
670 if (tc == 2) /* TC2, TC3 */
671 tc += (reg_idx - 96) >> 4;
674 break;
675 default:
676 tc = 0;
678 txoff <<= tc;
680 #endif
681 return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
684 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
685 struct ixgbe_ring *tx_ring,
686 unsigned int eop)
688 struct ixgbe_hw *hw = &adapter->hw;
690 /* Detect a transmit hang in hardware, this serializes the
691 * check with the clearing of time_stamp and movement of eop */
692 adapter->detect_tx_hung = false;
693 if (tx_ring->tx_buffer_info[eop].time_stamp &&
694 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
695 ixgbe_tx_xon_state(adapter, tx_ring)) {
696 /* detected Tx unit hang */
697 union ixgbe_adv_tx_desc *tx_desc;
698 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
699 e_err(drv, "Detected Tx Unit Hang\n"
700 " Tx Queue <%d>\n"
701 " TDH, TDT <%x>, <%x>\n"
702 " next_to_use <%x>\n"
703 " next_to_clean <%x>\n"
704 "tx_buffer_info[next_to_clean]\n"
705 " time_stamp <%lx>\n"
706 " jiffies <%lx>\n",
707 tx_ring->queue_index,
708 IXGBE_READ_REG(hw, tx_ring->head),
709 IXGBE_READ_REG(hw, tx_ring->tail),
710 tx_ring->next_to_use, eop,
711 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
712 return true;
715 return false;
718 #define IXGBE_MAX_TXD_PWR 14
719 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
721 /* Tx Descriptors needed, worst case */
722 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
723 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
724 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
725 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
727 static void ixgbe_tx_timeout(struct net_device *netdev);
730 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
731 * @q_vector: structure containing interrupt and ring information
732 * @tx_ring: tx ring to clean
734 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
735 struct ixgbe_ring *tx_ring)
737 struct ixgbe_adapter *adapter = q_vector->adapter;
738 struct net_device *netdev = adapter->netdev;
739 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
740 struct ixgbe_tx_buffer *tx_buffer_info;
741 unsigned int i, eop, count = 0;
742 unsigned int total_bytes = 0, total_packets = 0;
744 i = tx_ring->next_to_clean;
745 eop = tx_ring->tx_buffer_info[i].next_to_watch;
746 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
748 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
749 (count < tx_ring->work_limit)) {
750 bool cleaned = false;
751 rmb(); /* read buffer_info after eop_desc */
752 for ( ; !cleaned; count++) {
753 struct sk_buff *skb;
754 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
755 tx_buffer_info = &tx_ring->tx_buffer_info[i];
756 cleaned = (i == eop);
757 skb = tx_buffer_info->skb;
759 if (cleaned && skb) {
760 unsigned int segs, bytecount;
761 unsigned int hlen = skb_headlen(skb);
763 /* gso_segs is currently only valid for tcp */
764 segs = skb_shinfo(skb)->gso_segs ?: 1;
765 #ifdef IXGBE_FCOE
766 /* adjust for FCoE Sequence Offload */
767 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
768 && (skb->protocol == htons(ETH_P_FCOE)) &&
769 skb_is_gso(skb)) {
770 hlen = skb_transport_offset(skb) +
771 sizeof(struct fc_frame_header) +
772 sizeof(struct fcoe_crc_eof);
773 segs = DIV_ROUND_UP(skb->len - hlen,
774 skb_shinfo(skb)->gso_size);
776 #endif /* IXGBE_FCOE */
777 /* multiply data chunks by size of headers */
778 bytecount = ((segs - 1) * hlen) + skb->len;
779 total_packets += segs;
780 total_bytes += bytecount;
783 ixgbe_unmap_and_free_tx_resource(adapter,
784 tx_buffer_info);
786 tx_desc->wb.status = 0;
788 i++;
789 if (i == tx_ring->count)
790 i = 0;
793 eop = tx_ring->tx_buffer_info[i].next_to_watch;
794 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
797 tx_ring->next_to_clean = i;
799 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
800 if (unlikely(count && netif_carrier_ok(netdev) &&
801 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
802 /* Make sure that anybody stopping the queue after this
803 * sees the new next_to_clean.
805 smp_mb();
806 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
807 !test_bit(__IXGBE_DOWN, &adapter->state)) {
808 netif_wake_subqueue(netdev, tx_ring->queue_index);
809 ++tx_ring->restart_queue;
813 if (adapter->detect_tx_hung) {
814 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
815 /* schedule immediate reset if we believe we hung */
816 e_info(probe, "tx hang %d detected, resetting "
817 "adapter\n", adapter->tx_timeout_count + 1);
818 ixgbe_tx_timeout(adapter->netdev);
822 /* re-arm the interrupt */
823 if (count >= tx_ring->work_limit)
824 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
826 tx_ring->total_bytes += total_bytes;
827 tx_ring->total_packets += total_packets;
828 tx_ring->stats.packets += total_packets;
829 tx_ring->stats.bytes += total_bytes;
830 return (count < tx_ring->work_limit);
833 #ifdef CONFIG_IXGBE_DCA
834 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
835 struct ixgbe_ring *rx_ring)
837 u32 rxctrl;
838 int cpu = get_cpu();
839 int q = rx_ring->reg_idx;
841 if (rx_ring->cpu != cpu) {
842 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
843 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
844 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
845 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
846 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
847 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
848 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
849 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
851 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
852 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
853 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
854 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
855 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
856 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
857 rx_ring->cpu = cpu;
859 put_cpu();
862 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
863 struct ixgbe_ring *tx_ring)
865 u32 txctrl;
866 int cpu = get_cpu();
867 int q = tx_ring->reg_idx;
868 struct ixgbe_hw *hw = &adapter->hw;
870 if (tx_ring->cpu != cpu) {
871 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
872 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(q));
873 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
874 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
875 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
876 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(q), txctrl);
877 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
878 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(q));
879 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
880 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
881 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
882 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
883 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(q), txctrl);
885 tx_ring->cpu = cpu;
887 put_cpu();
890 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
892 int i;
894 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
895 return;
897 /* always use CB2 mode, difference is masked in the CB driver */
898 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
900 for (i = 0; i < adapter->num_tx_queues; i++) {
901 adapter->tx_ring[i]->cpu = -1;
902 ixgbe_update_tx_dca(adapter, adapter->tx_ring[i]);
904 for (i = 0; i < adapter->num_rx_queues; i++) {
905 adapter->rx_ring[i]->cpu = -1;
906 ixgbe_update_rx_dca(adapter, adapter->rx_ring[i]);
910 static int __ixgbe_notify_dca(struct device *dev, void *data)
912 struct net_device *netdev = dev_get_drvdata(dev);
913 struct ixgbe_adapter *adapter = netdev_priv(netdev);
914 unsigned long event = *(unsigned long *)data;
916 switch (event) {
917 case DCA_PROVIDER_ADD:
918 /* if we're already enabled, don't do it again */
919 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
920 break;
921 if (dca_add_requester(dev) == 0) {
922 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
923 ixgbe_setup_dca(adapter);
924 break;
926 /* Fall Through since DCA is disabled. */
927 case DCA_PROVIDER_REMOVE:
928 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
929 dca_remove_requester(dev);
930 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
931 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
933 break;
936 return 0;
939 #endif /* CONFIG_IXGBE_DCA */
941 * ixgbe_receive_skb - Send a completed packet up the stack
942 * @adapter: board private structure
943 * @skb: packet to send up
944 * @status: hardware indication of status of receive
945 * @rx_ring: rx descriptor ring (for a specific queue) to setup
946 * @rx_desc: rx descriptor
948 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
949 struct sk_buff *skb, u8 status,
950 struct ixgbe_ring *ring,
951 union ixgbe_adv_rx_desc *rx_desc)
953 struct ixgbe_adapter *adapter = q_vector->adapter;
954 struct napi_struct *napi = &q_vector->napi;
955 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
956 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
958 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
959 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
960 vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
961 else
962 napi_gro_receive(napi, skb);
963 } else {
964 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
965 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
966 else
967 netif_rx(skb);
972 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
973 * @adapter: address of board private structure
974 * @status_err: hardware indication of status of receive
975 * @skb: skb currently being received and modified
977 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
978 union ixgbe_adv_rx_desc *rx_desc,
979 struct sk_buff *skb)
981 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
983 skb->ip_summed = CHECKSUM_NONE;
985 /* Rx csum disabled */
986 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
987 return;
989 /* if IP and error */
990 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
991 (status_err & IXGBE_RXDADV_ERR_IPE)) {
992 adapter->hw_csum_rx_error++;
993 return;
996 if (!(status_err & IXGBE_RXD_STAT_L4CS))
997 return;
999 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1000 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1003 * 82599 errata, UDP frames with a 0 checksum can be marked as
1004 * checksum errors.
1006 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1007 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1008 return;
1010 adapter->hw_csum_rx_error++;
1011 return;
1014 /* It must be a TCP or UDP packet with a valid checksum */
1015 skb->ip_summed = CHECKSUM_UNNECESSARY;
1018 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
1019 struct ixgbe_ring *rx_ring, u32 val)
1022 * Force memory writes to complete before letting h/w
1023 * know there are new descriptors to fetch. (Only
1024 * applicable for weak-ordered memory model archs,
1025 * such as IA-64).
1027 wmb();
1028 IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
1032 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1033 * @adapter: address of board private structure
1035 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
1036 struct ixgbe_ring *rx_ring,
1037 int cleaned_count)
1039 struct net_device *netdev = adapter->netdev;
1040 struct pci_dev *pdev = adapter->pdev;
1041 union ixgbe_adv_rx_desc *rx_desc;
1042 struct ixgbe_rx_buffer *bi;
1043 unsigned int i;
1044 unsigned int bufsz = rx_ring->rx_buf_len;
1046 i = rx_ring->next_to_use;
1047 bi = &rx_ring->rx_buffer_info[i];
1049 while (cleaned_count--) {
1050 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
1052 if (!bi->page_dma &&
1053 (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
1054 if (!bi->page) {
1055 bi->page = netdev_alloc_page(netdev);
1056 if (!bi->page) {
1057 adapter->alloc_rx_page_failed++;
1058 goto no_buffers;
1060 bi->page_offset = 0;
1061 } else {
1062 /* use a half page if we're re-using */
1063 bi->page_offset ^= (PAGE_SIZE / 2);
1066 bi->page_dma = dma_map_page(&pdev->dev, bi->page,
1067 bi->page_offset,
1068 (PAGE_SIZE / 2),
1069 DMA_FROM_DEVICE);
1072 if (!bi->skb) {
1073 struct sk_buff *skb = netdev_alloc_skb_ip_align(netdev,
1074 bufsz);
1075 bi->skb = skb;
1077 if (!skb) {
1078 adapter->alloc_rx_buff_failed++;
1079 goto no_buffers;
1081 /* initialize queue mapping */
1082 skb_record_rx_queue(skb, rx_ring->queue_index);
1085 if (!bi->dma) {
1086 bi->dma = dma_map_single(&pdev->dev,
1087 bi->skb->data,
1088 rx_ring->rx_buf_len,
1089 DMA_FROM_DEVICE);
1091 /* Refresh the desc even if buffer_addrs didn't change because
1092 * each write-back erases this info. */
1093 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1094 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1095 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1096 } else {
1097 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1100 i++;
1101 if (i == rx_ring->count)
1102 i = 0;
1103 bi = &rx_ring->rx_buffer_info[i];
1106 no_buffers:
1107 if (rx_ring->next_to_use != i) {
1108 rx_ring->next_to_use = i;
1109 if (i-- == 0)
1110 i = (rx_ring->count - 1);
1112 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
1116 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
1118 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1121 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
1123 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1126 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
1128 return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1129 IXGBE_RXDADV_RSCCNT_MASK) >>
1130 IXGBE_RXDADV_RSCCNT_SHIFT;
1134 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1135 * @skb: pointer to the last skb in the rsc queue
1136 * @count: pointer to number of packets coalesced in this context
1138 * This function changes a queue full of hw rsc buffers into a completed
1139 * packet. It uses the ->prev pointers to find the first packet and then
1140 * turns it into the frag list owner.
1142 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
1143 u64 *count)
1145 unsigned int frag_list_size = 0;
1147 while (skb->prev) {
1148 struct sk_buff *prev = skb->prev;
1149 frag_list_size += skb->len;
1150 skb->prev = NULL;
1151 skb = prev;
1152 *count += 1;
1155 skb_shinfo(skb)->frag_list = skb->next;
1156 skb->next = NULL;
1157 skb->len += frag_list_size;
1158 skb->data_len += frag_list_size;
1159 skb->truesize += frag_list_size;
1160 return skb;
1163 struct ixgbe_rsc_cb {
1164 dma_addr_t dma;
1165 bool delay_unmap;
1168 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
1170 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1171 struct ixgbe_ring *rx_ring,
1172 int *work_done, int work_to_do)
1174 struct ixgbe_adapter *adapter = q_vector->adapter;
1175 struct net_device *netdev = adapter->netdev;
1176 struct pci_dev *pdev = adapter->pdev;
1177 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1178 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1179 struct sk_buff *skb;
1180 unsigned int i, rsc_count = 0;
1181 u32 len, staterr;
1182 u16 hdr_info;
1183 bool cleaned = false;
1184 int cleaned_count = 0;
1185 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1186 #ifdef IXGBE_FCOE
1187 int ddp_bytes = 0;
1188 #endif /* IXGBE_FCOE */
1190 i = rx_ring->next_to_clean;
1191 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
1192 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1193 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1195 while (staterr & IXGBE_RXD_STAT_DD) {
1196 u32 upper_len = 0;
1197 if (*work_done >= work_to_do)
1198 break;
1199 (*work_done)++;
1201 rmb(); /* read descriptor and rx_buffer_info after status DD */
1202 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1203 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
1204 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1205 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1206 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1207 if ((len > IXGBE_RX_HDR_SIZE) ||
1208 (upper_len && !(hdr_info & IXGBE_RXDADV_SPH)))
1209 len = IXGBE_RX_HDR_SIZE;
1210 } else {
1211 len = le16_to_cpu(rx_desc->wb.upper.length);
1214 cleaned = true;
1215 skb = rx_buffer_info->skb;
1216 prefetch(skb->data);
1217 rx_buffer_info->skb = NULL;
1219 if (rx_buffer_info->dma) {
1220 if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
1221 (!(staterr & IXGBE_RXD_STAT_EOP)) &&
1222 (!(skb->prev))) {
1224 * When HWRSC is enabled, delay unmapping
1225 * of the first packet. It carries the
1226 * header information, HW may still
1227 * access the header after the writeback.
1228 * Only unmap it when EOP is reached
1230 IXGBE_RSC_CB(skb)->delay_unmap = true;
1231 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1232 } else {
1233 dma_unmap_single(&pdev->dev,
1234 rx_buffer_info->dma,
1235 rx_ring->rx_buf_len,
1236 DMA_FROM_DEVICE);
1238 rx_buffer_info->dma = 0;
1239 skb_put(skb, len);
1242 if (upper_len) {
1243 dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
1244 PAGE_SIZE / 2, DMA_FROM_DEVICE);
1245 rx_buffer_info->page_dma = 0;
1246 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1247 rx_buffer_info->page,
1248 rx_buffer_info->page_offset,
1249 upper_len);
1251 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
1252 (page_count(rx_buffer_info->page) != 1))
1253 rx_buffer_info->page = NULL;
1254 else
1255 get_page(rx_buffer_info->page);
1257 skb->len += upper_len;
1258 skb->data_len += upper_len;
1259 skb->truesize += upper_len;
1262 i++;
1263 if (i == rx_ring->count)
1264 i = 0;
1266 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
1267 prefetch(next_rxd);
1268 cleaned_count++;
1270 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
1271 rsc_count = ixgbe_get_rsc_count(rx_desc);
1273 if (rsc_count) {
1274 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1275 IXGBE_RXDADV_NEXTP_SHIFT;
1276 next_buffer = &rx_ring->rx_buffer_info[nextp];
1277 } else {
1278 next_buffer = &rx_ring->rx_buffer_info[i];
1281 if (staterr & IXGBE_RXD_STAT_EOP) {
1282 if (skb->prev)
1283 skb = ixgbe_transform_rsc_queue(skb, &(rx_ring->rsc_count));
1284 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
1285 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1286 dma_unmap_single(&pdev->dev,
1287 IXGBE_RSC_CB(skb)->dma,
1288 rx_ring->rx_buf_len,
1289 DMA_FROM_DEVICE);
1290 IXGBE_RSC_CB(skb)->dma = 0;
1291 IXGBE_RSC_CB(skb)->delay_unmap = false;
1293 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)
1294 rx_ring->rsc_count += skb_shinfo(skb)->nr_frags;
1295 else
1296 rx_ring->rsc_count++;
1297 rx_ring->rsc_flush++;
1299 rx_ring->stats.packets++;
1300 rx_ring->stats.bytes += skb->len;
1301 } else {
1302 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1303 rx_buffer_info->skb = next_buffer->skb;
1304 rx_buffer_info->dma = next_buffer->dma;
1305 next_buffer->skb = skb;
1306 next_buffer->dma = 0;
1307 } else {
1308 skb->next = next_buffer->skb;
1309 skb->next->prev = skb;
1311 rx_ring->non_eop_descs++;
1312 goto next_desc;
1315 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1316 dev_kfree_skb_irq(skb);
1317 goto next_desc;
1320 ixgbe_rx_checksum(adapter, rx_desc, skb);
1322 /* probably a little skewed due to removing CRC */
1323 total_rx_bytes += skb->len;
1324 total_rx_packets++;
1326 skb->protocol = eth_type_trans(skb, adapter->netdev);
1327 #ifdef IXGBE_FCOE
1328 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1329 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1330 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1331 if (!ddp_bytes)
1332 goto next_desc;
1334 #endif /* IXGBE_FCOE */
1335 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1337 next_desc:
1338 rx_desc->wb.upper.status_error = 0;
1340 /* return some buffers to hardware, one at a time is too slow */
1341 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1342 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1343 cleaned_count = 0;
1346 /* use prefetched values */
1347 rx_desc = next_rxd;
1348 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1350 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1353 rx_ring->next_to_clean = i;
1354 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1356 if (cleaned_count)
1357 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1359 #ifdef IXGBE_FCOE
1360 /* include DDPed FCoE data */
1361 if (ddp_bytes > 0) {
1362 unsigned int mss;
1364 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
1365 sizeof(struct fc_frame_header) -
1366 sizeof(struct fcoe_crc_eof);
1367 if (mss > 512)
1368 mss &= ~511;
1369 total_rx_bytes += ddp_bytes;
1370 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1372 #endif /* IXGBE_FCOE */
1374 rx_ring->total_packets += total_rx_packets;
1375 rx_ring->total_bytes += total_rx_bytes;
1376 netdev->stats.rx_bytes += total_rx_bytes;
1377 netdev->stats.rx_packets += total_rx_packets;
1379 return cleaned;
1382 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1384 * ixgbe_configure_msix - Configure MSI-X hardware
1385 * @adapter: board private structure
1387 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1388 * interrupts.
1390 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1392 struct ixgbe_q_vector *q_vector;
1393 int i, j, q_vectors, v_idx, r_idx;
1394 u32 mask;
1396 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1399 * Populate the IVAR table and set the ITR values to the
1400 * corresponding register.
1402 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1403 q_vector = adapter->q_vector[v_idx];
1404 /* XXX for_each_set_bit(...) */
1405 r_idx = find_first_bit(q_vector->rxr_idx,
1406 adapter->num_rx_queues);
1408 for (i = 0; i < q_vector->rxr_count; i++) {
1409 j = adapter->rx_ring[r_idx]->reg_idx;
1410 ixgbe_set_ivar(adapter, 0, j, v_idx);
1411 r_idx = find_next_bit(q_vector->rxr_idx,
1412 adapter->num_rx_queues,
1413 r_idx + 1);
1415 r_idx = find_first_bit(q_vector->txr_idx,
1416 adapter->num_tx_queues);
1418 for (i = 0; i < q_vector->txr_count; i++) {
1419 j = adapter->tx_ring[r_idx]->reg_idx;
1420 ixgbe_set_ivar(adapter, 1, j, v_idx);
1421 r_idx = find_next_bit(q_vector->txr_idx,
1422 adapter->num_tx_queues,
1423 r_idx + 1);
1426 if (q_vector->txr_count && !q_vector->rxr_count)
1427 /* tx only */
1428 q_vector->eitr = adapter->tx_eitr_param;
1429 else if (q_vector->rxr_count)
1430 /* rx or mixed */
1431 q_vector->eitr = adapter->rx_eitr_param;
1433 ixgbe_write_eitr(q_vector);
1436 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1437 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1438 v_idx);
1439 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1440 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1441 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1443 /* set up to autoclear timer, and the vectors */
1444 mask = IXGBE_EIMS_ENABLE_MASK;
1445 if (adapter->num_vfs)
1446 mask &= ~(IXGBE_EIMS_OTHER |
1447 IXGBE_EIMS_MAILBOX |
1448 IXGBE_EIMS_LSC);
1449 else
1450 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1451 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1454 enum latency_range {
1455 lowest_latency = 0,
1456 low_latency = 1,
1457 bulk_latency = 2,
1458 latency_invalid = 255
1462 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1463 * @adapter: pointer to adapter
1464 * @eitr: eitr setting (ints per sec) to give last timeslice
1465 * @itr_setting: current throttle rate in ints/second
1466 * @packets: the number of packets during this measurement interval
1467 * @bytes: the number of bytes during this measurement interval
1469 * Stores a new ITR value based on packets and byte
1470 * counts during the last interrupt. The advantage of per interrupt
1471 * computation is faster updates and more accurate ITR for the current
1472 * traffic pattern. Constants in this function were computed
1473 * based on theoretical maximum wire speed and thresholds were set based
1474 * on testing data as well as attempting to minimize response time
1475 * while increasing bulk throughput.
1476 * this functionality is controlled by the InterruptThrottleRate module
1477 * parameter (see ixgbe_param.c)
1479 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1480 u32 eitr, u8 itr_setting,
1481 int packets, int bytes)
1483 unsigned int retval = itr_setting;
1484 u32 timepassed_us;
1485 u64 bytes_perint;
1487 if (packets == 0)
1488 goto update_itr_done;
1491 /* simple throttlerate management
1492 * 0-20MB/s lowest (100000 ints/s)
1493 * 20-100MB/s low (20000 ints/s)
1494 * 100-1249MB/s bulk (8000 ints/s)
1496 /* what was last interrupt timeslice? */
1497 timepassed_us = 1000000/eitr;
1498 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1500 switch (itr_setting) {
1501 case lowest_latency:
1502 if (bytes_perint > adapter->eitr_low)
1503 retval = low_latency;
1504 break;
1505 case low_latency:
1506 if (bytes_perint > adapter->eitr_high)
1507 retval = bulk_latency;
1508 else if (bytes_perint <= adapter->eitr_low)
1509 retval = lowest_latency;
1510 break;
1511 case bulk_latency:
1512 if (bytes_perint <= adapter->eitr_high)
1513 retval = low_latency;
1514 break;
1517 update_itr_done:
1518 return retval;
1522 * ixgbe_write_eitr - write EITR register in hardware specific way
1523 * @q_vector: structure containing interrupt and ring information
1525 * This function is made to be called by ethtool and by the driver
1526 * when it needs to update EITR registers at runtime. Hardware
1527 * specific quirks/differences are taken care of here.
1529 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1531 struct ixgbe_adapter *adapter = q_vector->adapter;
1532 struct ixgbe_hw *hw = &adapter->hw;
1533 int v_idx = q_vector->v_idx;
1534 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1536 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1537 /* must write high and low 16 bits to reset counter */
1538 itr_reg |= (itr_reg << 16);
1539 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1541 * 82599 can support a value of zero, so allow it for
1542 * max interrupt rate, but there is an errata where it can
1543 * not be zero with RSC
1545 if (itr_reg == 8 &&
1546 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1547 itr_reg = 0;
1550 * set the WDIS bit to not clear the timer bits and cause an
1551 * immediate assertion of the interrupt
1553 itr_reg |= IXGBE_EITR_CNT_WDIS;
1555 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1558 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1560 struct ixgbe_adapter *adapter = q_vector->adapter;
1561 u32 new_itr;
1562 u8 current_itr, ret_itr;
1563 int i, r_idx;
1564 struct ixgbe_ring *rx_ring, *tx_ring;
1566 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1567 for (i = 0; i < q_vector->txr_count; i++) {
1568 tx_ring = adapter->tx_ring[r_idx];
1569 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1570 q_vector->tx_itr,
1571 tx_ring->total_packets,
1572 tx_ring->total_bytes);
1573 /* if the result for this queue would decrease interrupt
1574 * rate for this vector then use that result */
1575 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1576 q_vector->tx_itr - 1 : ret_itr);
1577 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1578 r_idx + 1);
1581 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1582 for (i = 0; i < q_vector->rxr_count; i++) {
1583 rx_ring = adapter->rx_ring[r_idx];
1584 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1585 q_vector->rx_itr,
1586 rx_ring->total_packets,
1587 rx_ring->total_bytes);
1588 /* if the result for this queue would decrease interrupt
1589 * rate for this vector then use that result */
1590 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1591 q_vector->rx_itr - 1 : ret_itr);
1592 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1593 r_idx + 1);
1596 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1598 switch (current_itr) {
1599 /* counts and packets in update_itr are dependent on these numbers */
1600 case lowest_latency:
1601 new_itr = 100000;
1602 break;
1603 case low_latency:
1604 new_itr = 20000; /* aka hwitr = ~200 */
1605 break;
1606 case bulk_latency:
1607 default:
1608 new_itr = 8000;
1609 break;
1612 if (new_itr != q_vector->eitr) {
1613 /* do an exponential smoothing */
1614 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1616 /* save the algorithm value here, not the smoothed one */
1617 q_vector->eitr = new_itr;
1619 ixgbe_write_eitr(q_vector);
1624 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1625 * @work: pointer to work_struct containing our data
1627 static void ixgbe_check_overtemp_task(struct work_struct *work)
1629 struct ixgbe_adapter *adapter = container_of(work,
1630 struct ixgbe_adapter,
1631 check_overtemp_task);
1632 struct ixgbe_hw *hw = &adapter->hw;
1633 u32 eicr = adapter->interrupt_event;
1635 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
1636 switch (hw->device_id) {
1637 case IXGBE_DEV_ID_82599_T3_LOM: {
1638 u32 autoneg;
1639 bool link_up = false;
1641 if (hw->mac.ops.check_link)
1642 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1644 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1645 (eicr & IXGBE_EICR_LSC))
1646 /* Check if this is due to overtemp */
1647 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1648 break;
1650 return;
1651 default:
1652 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1653 return;
1654 break;
1656 e_crit(drv, "Network adapter has been stopped because it has "
1657 "over heated. Restart the computer. If the problem "
1658 "persists, power off the system and replace the "
1659 "adapter\n");
1660 /* write to clear the interrupt */
1661 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1665 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1667 struct ixgbe_hw *hw = &adapter->hw;
1669 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1670 (eicr & IXGBE_EICR_GPI_SDP1)) {
1671 e_crit(probe, "Fan has stopped, replace the adapter\n");
1672 /* write to clear the interrupt */
1673 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1677 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1679 struct ixgbe_hw *hw = &adapter->hw;
1681 if (eicr & IXGBE_EICR_GPI_SDP1) {
1682 /* Clear the interrupt */
1683 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1684 schedule_work(&adapter->multispeed_fiber_task);
1685 } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1686 /* Clear the interrupt */
1687 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1688 schedule_work(&adapter->sfp_config_module_task);
1689 } else {
1690 /* Interrupt isn't for us... */
1691 return;
1695 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1697 struct ixgbe_hw *hw = &adapter->hw;
1699 adapter->lsc_int++;
1700 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1701 adapter->link_check_timeout = jiffies;
1702 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1703 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1704 IXGBE_WRITE_FLUSH(hw);
1705 schedule_work(&adapter->watchdog_task);
1709 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1711 struct net_device *netdev = data;
1712 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1713 struct ixgbe_hw *hw = &adapter->hw;
1714 u32 eicr;
1717 * Workaround for Silicon errata. Use clear-by-write instead
1718 * of clear-by-read. Reading with EICS will return the
1719 * interrupt causes without clearing, which later be done
1720 * with the write to EICR.
1722 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1723 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1725 if (eicr & IXGBE_EICR_LSC)
1726 ixgbe_check_lsc(adapter);
1728 if (eicr & IXGBE_EICR_MAILBOX)
1729 ixgbe_msg_task(adapter);
1731 if (hw->mac.type == ixgbe_mac_82598EB)
1732 ixgbe_check_fan_failure(adapter, eicr);
1734 if (hw->mac.type == ixgbe_mac_82599EB) {
1735 ixgbe_check_sfp_event(adapter, eicr);
1736 adapter->interrupt_event = eicr;
1737 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1738 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
1739 schedule_work(&adapter->check_overtemp_task);
1741 /* Handle Flow Director Full threshold interrupt */
1742 if (eicr & IXGBE_EICR_FLOW_DIR) {
1743 int i;
1744 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1745 /* Disable transmits before FDIR Re-initialization */
1746 netif_tx_stop_all_queues(netdev);
1747 for (i = 0; i < adapter->num_tx_queues; i++) {
1748 struct ixgbe_ring *tx_ring =
1749 adapter->tx_ring[i];
1750 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1751 &tx_ring->reinit_state))
1752 schedule_work(&adapter->fdir_reinit_task);
1756 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1757 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1759 return IRQ_HANDLED;
1762 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1763 u64 qmask)
1765 u32 mask;
1767 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1768 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1769 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1770 } else {
1771 mask = (qmask & 0xFFFFFFFF);
1772 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1773 mask = (qmask >> 32);
1774 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1776 /* skip the flush */
1779 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1780 u64 qmask)
1782 u32 mask;
1784 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1785 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1786 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1787 } else {
1788 mask = (qmask & 0xFFFFFFFF);
1789 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1790 mask = (qmask >> 32);
1791 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1793 /* skip the flush */
1796 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1798 struct ixgbe_q_vector *q_vector = data;
1799 struct ixgbe_adapter *adapter = q_vector->adapter;
1800 struct ixgbe_ring *tx_ring;
1801 int i, r_idx;
1803 if (!q_vector->txr_count)
1804 return IRQ_HANDLED;
1806 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1807 for (i = 0; i < q_vector->txr_count; i++) {
1808 tx_ring = adapter->tx_ring[r_idx];
1809 tx_ring->total_bytes = 0;
1810 tx_ring->total_packets = 0;
1811 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1812 r_idx + 1);
1815 /* EIAM disabled interrupts (on this vector) for us */
1816 napi_schedule(&q_vector->napi);
1818 return IRQ_HANDLED;
1822 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1823 * @irq: unused
1824 * @data: pointer to our q_vector struct for this interrupt vector
1826 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1828 struct ixgbe_q_vector *q_vector = data;
1829 struct ixgbe_adapter *adapter = q_vector->adapter;
1830 struct ixgbe_ring *rx_ring;
1831 int r_idx;
1832 int i;
1834 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1835 for (i = 0; i < q_vector->rxr_count; i++) {
1836 rx_ring = adapter->rx_ring[r_idx];
1837 rx_ring->total_bytes = 0;
1838 rx_ring->total_packets = 0;
1839 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1840 r_idx + 1);
1843 if (!q_vector->rxr_count)
1844 return IRQ_HANDLED;
1846 /* disable interrupts on this vector only */
1847 /* EIAM disabled interrupts (on this vector) for us */
1848 napi_schedule(&q_vector->napi);
1850 return IRQ_HANDLED;
1853 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1855 struct ixgbe_q_vector *q_vector = data;
1856 struct ixgbe_adapter *adapter = q_vector->adapter;
1857 struct ixgbe_ring *ring;
1858 int r_idx;
1859 int i;
1861 if (!q_vector->txr_count && !q_vector->rxr_count)
1862 return IRQ_HANDLED;
1864 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1865 for (i = 0; i < q_vector->txr_count; i++) {
1866 ring = adapter->tx_ring[r_idx];
1867 ring->total_bytes = 0;
1868 ring->total_packets = 0;
1869 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1870 r_idx + 1);
1873 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1874 for (i = 0; i < q_vector->rxr_count; i++) {
1875 ring = adapter->rx_ring[r_idx];
1876 ring->total_bytes = 0;
1877 ring->total_packets = 0;
1878 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1879 r_idx + 1);
1882 /* EIAM disabled interrupts (on this vector) for us */
1883 napi_schedule(&q_vector->napi);
1885 return IRQ_HANDLED;
1889 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1890 * @napi: napi struct with our devices info in it
1891 * @budget: amount of work driver is allowed to do this pass, in packets
1893 * This function is optimized for cleaning one queue only on a single
1894 * q_vector!!!
1896 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1898 struct ixgbe_q_vector *q_vector =
1899 container_of(napi, struct ixgbe_q_vector, napi);
1900 struct ixgbe_adapter *adapter = q_vector->adapter;
1901 struct ixgbe_ring *rx_ring = NULL;
1902 int work_done = 0;
1903 long r_idx;
1905 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1906 rx_ring = adapter->rx_ring[r_idx];
1907 #ifdef CONFIG_IXGBE_DCA
1908 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1909 ixgbe_update_rx_dca(adapter, rx_ring);
1910 #endif
1912 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1914 /* If all Rx work done, exit the polling mode */
1915 if (work_done < budget) {
1916 napi_complete(napi);
1917 if (adapter->rx_itr_setting & 1)
1918 ixgbe_set_itr_msix(q_vector);
1919 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1920 ixgbe_irq_enable_queues(adapter,
1921 ((u64)1 << q_vector->v_idx));
1924 return work_done;
1928 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1929 * @napi: napi struct with our devices info in it
1930 * @budget: amount of work driver is allowed to do this pass, in packets
1932 * This function will clean more than one rx queue associated with a
1933 * q_vector.
1935 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1937 struct ixgbe_q_vector *q_vector =
1938 container_of(napi, struct ixgbe_q_vector, napi);
1939 struct ixgbe_adapter *adapter = q_vector->adapter;
1940 struct ixgbe_ring *ring = NULL;
1941 int work_done = 0, i;
1942 long r_idx;
1943 bool tx_clean_complete = true;
1945 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1946 for (i = 0; i < q_vector->txr_count; i++) {
1947 ring = adapter->tx_ring[r_idx];
1948 #ifdef CONFIG_IXGBE_DCA
1949 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1950 ixgbe_update_tx_dca(adapter, ring);
1951 #endif
1952 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1953 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1954 r_idx + 1);
1957 /* attempt to distribute budget to each queue fairly, but don't allow
1958 * the budget to go below 1 because we'll exit polling */
1959 budget /= (q_vector->rxr_count ?: 1);
1960 budget = max(budget, 1);
1961 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1962 for (i = 0; i < q_vector->rxr_count; i++) {
1963 ring = adapter->rx_ring[r_idx];
1964 #ifdef CONFIG_IXGBE_DCA
1965 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1966 ixgbe_update_rx_dca(adapter, ring);
1967 #endif
1968 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1969 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1970 r_idx + 1);
1973 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1974 ring = adapter->rx_ring[r_idx];
1975 /* If all Rx work done, exit the polling mode */
1976 if (work_done < budget) {
1977 napi_complete(napi);
1978 if (adapter->rx_itr_setting & 1)
1979 ixgbe_set_itr_msix(q_vector);
1980 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1981 ixgbe_irq_enable_queues(adapter,
1982 ((u64)1 << q_vector->v_idx));
1983 return 0;
1986 return work_done;
1990 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1991 * @napi: napi struct with our devices info in it
1992 * @budget: amount of work driver is allowed to do this pass, in packets
1994 * This function is optimized for cleaning one queue only on a single
1995 * q_vector!!!
1997 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1999 struct ixgbe_q_vector *q_vector =
2000 container_of(napi, struct ixgbe_q_vector, napi);
2001 struct ixgbe_adapter *adapter = q_vector->adapter;
2002 struct ixgbe_ring *tx_ring = NULL;
2003 int work_done = 0;
2004 long r_idx;
2006 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2007 tx_ring = adapter->tx_ring[r_idx];
2008 #ifdef CONFIG_IXGBE_DCA
2009 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2010 ixgbe_update_tx_dca(adapter, tx_ring);
2011 #endif
2013 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2014 work_done = budget;
2016 /* If all Tx work done, exit the polling mode */
2017 if (work_done < budget) {
2018 napi_complete(napi);
2019 if (adapter->tx_itr_setting & 1)
2020 ixgbe_set_itr_msix(q_vector);
2021 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2022 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2025 return work_done;
2028 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2029 int r_idx)
2031 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2033 set_bit(r_idx, q_vector->rxr_idx);
2034 q_vector->rxr_count++;
2037 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2038 int t_idx)
2040 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2042 set_bit(t_idx, q_vector->txr_idx);
2043 q_vector->txr_count++;
2047 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2048 * @adapter: board private structure to initialize
2049 * @vectors: allotted vector count for descriptor rings
2051 * This function maps descriptor rings to the queue-specific vectors
2052 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2053 * one vector per ring/queue, but on a constrained vector budget, we
2054 * group the rings as "efficiently" as possible. You would add new
2055 * mapping configurations in here.
2057 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
2058 int vectors)
2060 int v_start = 0;
2061 int rxr_idx = 0, txr_idx = 0;
2062 int rxr_remaining = adapter->num_rx_queues;
2063 int txr_remaining = adapter->num_tx_queues;
2064 int i, j;
2065 int rqpv, tqpv;
2066 int err = 0;
2068 /* No mapping required if MSI-X is disabled. */
2069 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2070 goto out;
2073 * The ideal configuration...
2074 * We have enough vectors to map one per queue.
2076 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2077 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2078 map_vector_to_rxq(adapter, v_start, rxr_idx);
2080 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2081 map_vector_to_txq(adapter, v_start, txr_idx);
2083 goto out;
2087 * If we don't have enough vectors for a 1-to-1
2088 * mapping, we'll have to group them so there are
2089 * multiple queues per vector.
2091 /* Re-adjusting *qpv takes care of the remainder. */
2092 for (i = v_start; i < vectors; i++) {
2093 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
2094 for (j = 0; j < rqpv; j++) {
2095 map_vector_to_rxq(adapter, i, rxr_idx);
2096 rxr_idx++;
2097 rxr_remaining--;
2100 for (i = v_start; i < vectors; i++) {
2101 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
2102 for (j = 0; j < tqpv; j++) {
2103 map_vector_to_txq(adapter, i, txr_idx);
2104 txr_idx++;
2105 txr_remaining--;
2109 out:
2110 return err;
2114 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2115 * @adapter: board private structure
2117 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2118 * interrupts from the kernel.
2120 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2122 struct net_device *netdev = adapter->netdev;
2123 irqreturn_t (*handler)(int, void *);
2124 int i, vector, q_vectors, err;
2125 int ri=0, ti=0;
2127 /* Decrement for Other and TCP Timer vectors */
2128 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2130 /* Map the Tx/Rx rings to the vectors we were allotted. */
2131 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
2132 if (err)
2133 goto out;
2135 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
2136 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
2137 &ixgbe_msix_clean_many)
2138 for (vector = 0; vector < q_vectors; vector++) {
2139 handler = SET_HANDLER(adapter->q_vector[vector]);
2141 if(handler == &ixgbe_msix_clean_rx) {
2142 sprintf(adapter->name[vector], "%s-%s-%d",
2143 netdev->name, "rx", ri++);
2145 else if(handler == &ixgbe_msix_clean_tx) {
2146 sprintf(adapter->name[vector], "%s-%s-%d",
2147 netdev->name, "tx", ti++);
2149 else
2150 sprintf(adapter->name[vector], "%s-%s-%d",
2151 netdev->name, "TxRx", vector);
2153 err = request_irq(adapter->msix_entries[vector].vector,
2154 handler, 0, adapter->name[vector],
2155 adapter->q_vector[vector]);
2156 if (err) {
2157 e_err(probe, "request_irq failed for MSIX interrupt "
2158 "Error: %d\n", err);
2159 goto free_queue_irqs;
2163 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
2164 err = request_irq(adapter->msix_entries[vector].vector,
2165 ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
2166 if (err) {
2167 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2168 goto free_queue_irqs;
2171 return 0;
2173 free_queue_irqs:
2174 for (i = vector - 1; i >= 0; i--)
2175 free_irq(adapter->msix_entries[--vector].vector,
2176 adapter->q_vector[i]);
2177 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2178 pci_disable_msix(adapter->pdev);
2179 kfree(adapter->msix_entries);
2180 adapter->msix_entries = NULL;
2181 out:
2182 return err;
2185 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2187 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2188 u8 current_itr;
2189 u32 new_itr = q_vector->eitr;
2190 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2191 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2193 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2194 q_vector->tx_itr,
2195 tx_ring->total_packets,
2196 tx_ring->total_bytes);
2197 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2198 q_vector->rx_itr,
2199 rx_ring->total_packets,
2200 rx_ring->total_bytes);
2202 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2204 switch (current_itr) {
2205 /* counts and packets in update_itr are dependent on these numbers */
2206 case lowest_latency:
2207 new_itr = 100000;
2208 break;
2209 case low_latency:
2210 new_itr = 20000; /* aka hwitr = ~200 */
2211 break;
2212 case bulk_latency:
2213 new_itr = 8000;
2214 break;
2215 default:
2216 break;
2219 if (new_itr != q_vector->eitr) {
2220 /* do an exponential smoothing */
2221 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
2223 /* save the algorithm value here, not the smoothed one */
2224 q_vector->eitr = new_itr;
2226 ixgbe_write_eitr(q_vector);
2231 * ixgbe_irq_enable - Enable default interrupt generation settings
2232 * @adapter: board private structure
2234 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
2236 u32 mask;
2238 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2239 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2240 mask |= IXGBE_EIMS_GPI_SDP0;
2241 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2242 mask |= IXGBE_EIMS_GPI_SDP1;
2243 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2244 mask |= IXGBE_EIMS_ECC;
2245 mask |= IXGBE_EIMS_GPI_SDP1;
2246 mask |= IXGBE_EIMS_GPI_SDP2;
2247 if (adapter->num_vfs)
2248 mask |= IXGBE_EIMS_MAILBOX;
2250 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2251 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2252 mask |= IXGBE_EIMS_FLOW_DIR;
2254 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2255 ixgbe_irq_enable_queues(adapter, ~0);
2256 IXGBE_WRITE_FLUSH(&adapter->hw);
2258 if (adapter->num_vfs > 32) {
2259 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2260 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2265 * ixgbe_intr - legacy mode Interrupt Handler
2266 * @irq: interrupt number
2267 * @data: pointer to a network interface device structure
2269 static irqreturn_t ixgbe_intr(int irq, void *data)
2271 struct net_device *netdev = data;
2272 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2273 struct ixgbe_hw *hw = &adapter->hw;
2274 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2275 u32 eicr;
2278 * Workaround for silicon errata. Mask the interrupts
2279 * before the read of EICR.
2281 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2283 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2284 * therefore no explict interrupt disable is necessary */
2285 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2286 if (!eicr) {
2287 /* shared interrupt alert!
2288 * make sure interrupts are enabled because the read will
2289 * have disabled interrupts due to EIAM */
2290 ixgbe_irq_enable(adapter);
2291 return IRQ_NONE; /* Not our interrupt */
2294 if (eicr & IXGBE_EICR_LSC)
2295 ixgbe_check_lsc(adapter);
2297 if (hw->mac.type == ixgbe_mac_82599EB)
2298 ixgbe_check_sfp_event(adapter, eicr);
2300 ixgbe_check_fan_failure(adapter, eicr);
2301 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2302 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
2303 schedule_work(&adapter->check_overtemp_task);
2305 if (napi_schedule_prep(&(q_vector->napi))) {
2306 adapter->tx_ring[0]->total_packets = 0;
2307 adapter->tx_ring[0]->total_bytes = 0;
2308 adapter->rx_ring[0]->total_packets = 0;
2309 adapter->rx_ring[0]->total_bytes = 0;
2310 /* would disable interrupts here but EIAM disabled it */
2311 __napi_schedule(&(q_vector->napi));
2314 return IRQ_HANDLED;
2317 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2319 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2321 for (i = 0; i < q_vectors; i++) {
2322 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2323 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2324 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2325 q_vector->rxr_count = 0;
2326 q_vector->txr_count = 0;
2331 * ixgbe_request_irq - initialize interrupts
2332 * @adapter: board private structure
2334 * Attempts to configure interrupts using the best available
2335 * capabilities of the hardware and kernel.
2337 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2339 struct net_device *netdev = adapter->netdev;
2340 int err;
2342 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2343 err = ixgbe_request_msix_irqs(adapter);
2344 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2345 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2346 netdev->name, netdev);
2347 } else {
2348 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2349 netdev->name, netdev);
2352 if (err)
2353 e_err(probe, "request_irq failed, Error %d\n", err);
2355 return err;
2358 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2360 struct net_device *netdev = adapter->netdev;
2362 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2363 int i, q_vectors;
2365 q_vectors = adapter->num_msix_vectors;
2367 i = q_vectors - 1;
2368 free_irq(adapter->msix_entries[i].vector, netdev);
2370 i--;
2371 for (; i >= 0; i--) {
2372 free_irq(adapter->msix_entries[i].vector,
2373 adapter->q_vector[i]);
2376 ixgbe_reset_q_vectors(adapter);
2377 } else {
2378 free_irq(adapter->pdev->irq, netdev);
2383 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2384 * @adapter: board private structure
2386 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2388 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2389 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2390 } else {
2391 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2392 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2393 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2394 if (adapter->num_vfs > 32)
2395 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2397 IXGBE_WRITE_FLUSH(&adapter->hw);
2398 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2399 int i;
2400 for (i = 0; i < adapter->num_msix_vectors; i++)
2401 synchronize_irq(adapter->msix_entries[i].vector);
2402 } else {
2403 synchronize_irq(adapter->pdev->irq);
2408 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2411 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2413 struct ixgbe_hw *hw = &adapter->hw;
2415 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2416 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2418 ixgbe_set_ivar(adapter, 0, 0, 0);
2419 ixgbe_set_ivar(adapter, 1, 0, 0);
2421 map_vector_to_rxq(adapter, 0, 0);
2422 map_vector_to_txq(adapter, 0, 0);
2424 e_info(hw, "Legacy interrupt IVAR setup done\n");
2428 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2429 * @adapter: board private structure
2430 * @ring: structure containing ring specific data
2432 * Configure the Tx descriptor ring after a reset.
2434 static void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2435 struct ixgbe_ring *ring)
2437 struct ixgbe_hw *hw = &adapter->hw;
2438 u64 tdba = ring->dma;
2439 u16 reg_idx = ring->reg_idx;
2441 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2442 (tdba & DMA_BIT_MASK(32)));
2443 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2444 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2445 ring->count * sizeof(union ixgbe_adv_tx_desc));
2446 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2447 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2448 ring->head = IXGBE_TDH(reg_idx);
2449 ring->tail = IXGBE_TDT(reg_idx);
2453 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2455 struct ixgbe_hw *hw = &adapter->hw;
2456 u32 rttdcs;
2457 u32 mask;
2459 if (hw->mac.type == ixgbe_mac_82598EB)
2460 return;
2462 /* disable the arbiter while setting MTQC */
2463 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2464 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2465 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2467 /* set transmit pool layout */
2468 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2469 switch (adapter->flags & mask) {
2471 case (IXGBE_FLAG_SRIOV_ENABLED):
2472 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2473 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2474 break;
2476 case (IXGBE_FLAG_DCB_ENABLED):
2477 /* We enable 8 traffic classes, DCB only */
2478 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2479 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2480 break;
2482 default:
2483 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2484 break;
2487 /* re-enable the arbiter */
2488 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2489 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2493 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2494 * @adapter: board private structure
2496 * Configure the Tx unit of the MAC after a reset.
2498 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2500 u32 i;
2502 /* Setup the HW Tx Head and Tail descriptor pointers */
2503 for (i = 0; i < adapter->num_tx_queues; i++)
2504 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2506 ixgbe_setup_mtqc(adapter);
2509 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2511 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2512 struct ixgbe_ring *rx_ring)
2514 u32 srrctl;
2515 int index;
2516 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2518 index = rx_ring->reg_idx;
2519 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2520 unsigned long mask;
2521 mask = (unsigned long) feature[RING_F_RSS].mask;
2522 index = index & mask;
2524 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2526 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2527 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2529 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2530 IXGBE_SRRCTL_BSIZEHDR_MASK;
2532 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2533 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2534 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2535 #else
2536 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2537 #endif
2538 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2539 } else {
2540 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2541 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2542 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2545 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2548 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2550 struct ixgbe_hw *hw = &adapter->hw;
2551 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2552 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2553 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2554 u32 mrqc = 0, reta = 0;
2555 u32 rxcsum;
2556 int i, j;
2557 int mask;
2559 /* Fill out hash function seeds */
2560 for (i = 0; i < 10; i++)
2561 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2563 /* Fill out redirection table */
2564 for (i = 0, j = 0; i < 128; i++, j++) {
2565 if (j == adapter->ring_feature[RING_F_RSS].indices)
2566 j = 0;
2567 /* reta = 4-byte sliding window of
2568 * 0x00..(indices-1)(indices-1)00..etc. */
2569 reta = (reta << 8) | (j * 0x11);
2570 if ((i & 3) == 3)
2571 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2574 /* Disable indicating checksum in descriptor, enables RSS hash */
2575 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2576 rxcsum |= IXGBE_RXCSUM_PCSD;
2577 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2579 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2580 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2581 else
2582 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2583 #ifdef CONFIG_IXGBE_DCB
2584 | IXGBE_FLAG_DCB_ENABLED
2585 #endif
2586 | IXGBE_FLAG_SRIOV_ENABLED
2589 switch (mask) {
2590 case (IXGBE_FLAG_RSS_ENABLED):
2591 mrqc = IXGBE_MRQC_RSSEN;
2592 break;
2593 case (IXGBE_FLAG_SRIOV_ENABLED):
2594 mrqc = IXGBE_MRQC_VMDQEN;
2595 break;
2596 #ifdef CONFIG_IXGBE_DCB
2597 case (IXGBE_FLAG_DCB_ENABLED):
2598 mrqc = IXGBE_MRQC_RT8TCEN;
2599 break;
2600 #endif /* CONFIG_IXGBE_DCB */
2601 default:
2602 break;
2605 /* Perform hash on these packet types */
2606 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2607 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2608 | IXGBE_MRQC_RSS_FIELD_IPV6
2609 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2611 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2615 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2616 * @adapter: address of board private structure
2617 * @index: index of ring to set
2619 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2620 struct ixgbe_ring *ring)
2622 struct ixgbe_hw *hw = &adapter->hw;
2623 u32 rscctrl;
2624 int rx_buf_len;
2625 u16 reg_idx = ring->reg_idx;
2627 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
2628 return;
2630 rx_buf_len = ring->rx_buf_len;
2631 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2632 rscctrl |= IXGBE_RSCCTL_RSCEN;
2634 * we must limit the number of descriptors so that the
2635 * total size of max desc * buf_len is not greater
2636 * than 65535
2638 if (ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2639 #if (MAX_SKB_FRAGS > 16)
2640 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2641 #elif (MAX_SKB_FRAGS > 8)
2642 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2643 #elif (MAX_SKB_FRAGS > 4)
2644 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2645 #else
2646 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2647 #endif
2648 } else {
2649 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2650 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2651 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2652 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2653 else
2654 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2656 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2659 static void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2660 struct ixgbe_ring *ring)
2662 struct ixgbe_hw *hw = &adapter->hw;
2663 u64 rdba = ring->dma;
2664 u16 reg_idx = ring->reg_idx;
2666 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2667 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2668 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2669 ring->count * sizeof(union ixgbe_adv_rx_desc));
2670 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2671 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2672 ring->head = IXGBE_RDH(reg_idx);
2673 ring->tail = IXGBE_RDT(reg_idx);
2676 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2678 struct ixgbe_hw *hw = &adapter->hw;
2679 int p;
2681 /* PSRTYPE must be initialized in non 82598 adapters */
2682 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2683 IXGBE_PSRTYPE_UDPHDR |
2684 IXGBE_PSRTYPE_IPV4HDR |
2685 IXGBE_PSRTYPE_L2HDR |
2686 IXGBE_PSRTYPE_IPV6HDR;
2688 if (hw->mac.type == ixgbe_mac_82598EB)
2689 return;
2691 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2692 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2694 for (p = 0; p < adapter->num_rx_pools; p++)
2695 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2696 psrtype);
2699 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2701 struct ixgbe_hw *hw = &adapter->hw;
2702 struct net_device *netdev = adapter->netdev;
2703 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2704 int rx_buf_len;
2705 struct ixgbe_ring *rx_ring;
2706 int i;
2707 u32 mhadd, hlreg0;
2709 /* Decide whether to use packet split mode or not */
2710 /* Do not use packet split if we're in SR-IOV Mode */
2711 if (!adapter->num_vfs)
2712 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2714 /* Set the RX buffer length according to the mode */
2715 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2716 rx_buf_len = IXGBE_RX_HDR_SIZE;
2717 } else {
2718 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2719 (netdev->mtu <= ETH_DATA_LEN))
2720 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2721 else
2722 rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
2725 #ifdef IXGBE_FCOE
2726 /* adjust max frame to be able to do baby jumbo for FCoE */
2727 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2728 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2729 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2731 #endif /* IXGBE_FCOE */
2732 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2733 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2734 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2735 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2737 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2740 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2741 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2742 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2743 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2746 * Setup the HW Rx Head and Tail Descriptor Pointers and
2747 * the Base and Length of the Rx Descriptor Ring
2749 for (i = 0; i < adapter->num_rx_queues; i++) {
2750 rx_ring = adapter->rx_ring[i];
2751 rx_ring->rx_buf_len = rx_buf_len;
2753 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2754 rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2755 else
2756 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2758 #ifdef IXGBE_FCOE
2759 if (netdev->features & NETIF_F_FCOE_MTU)
2761 struct ixgbe_ring_feature *f;
2762 f = &adapter->ring_feature[RING_F_FCOE];
2763 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2764 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2765 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2766 rx_ring->rx_buf_len =
2767 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2770 #endif /* IXGBE_FCOE */
2775 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
2777 struct ixgbe_hw *hw = &adapter->hw;
2778 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2780 switch (hw->mac.type) {
2781 case ixgbe_mac_82598EB:
2783 * For VMDq support of different descriptor types or
2784 * buffer sizes through the use of multiple SRRCTL
2785 * registers, RDRXCTL.MVMEN must be set to 1
2787 * also, the manual doesn't mention it clearly but DCA hints
2788 * will only use queue 0's tags unless this bit is set. Side
2789 * effects of setting this bit are only that SRRCTL must be
2790 * fully programmed [0..15]
2792 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2793 break;
2794 case ixgbe_mac_82599EB:
2795 /* Disable RSC for ACK packets */
2796 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2797 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2798 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2799 /* hardware requires some bits to be set by default */
2800 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
2801 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2802 break;
2803 default:
2804 /* We should do nothing since we don't know this hardware */
2805 return;
2808 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2812 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2813 * @adapter: board private structure
2815 * Configure the Rx unit of the MAC after a reset.
2817 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2819 struct ixgbe_hw *hw = &adapter->hw;
2820 struct ixgbe_ring *rx_ring;
2821 int i;
2822 u32 rxctrl;
2823 u32 gcr_ext;
2825 /* disable receives while setting up the descriptors */
2826 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2827 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2829 ixgbe_setup_psrtype(adapter);
2830 ixgbe_setup_rdrxctl(adapter);
2832 /* set_rx_buffer_len must be called before ring initialization */
2833 ixgbe_set_rx_buffer_len(adapter);
2836 * Setup the HW Rx Head and Tail Descriptor Pointers and
2837 * the Base and Length of the Rx Descriptor Ring
2839 for (i = 0; i < adapter->num_rx_queues; i++) {
2840 rx_ring = adapter->rx_ring[i];
2841 ixgbe_configure_rx_ring(adapter, rx_ring);
2842 ixgbe_configure_srrctl(adapter, rx_ring);
2843 ixgbe_configure_rscctl(adapter, rx_ring);
2846 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2847 u32 vt_reg_bits;
2848 u32 reg_offset, vf_shift;
2849 u32 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2850 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN
2851 | IXGBE_VT_CTL_REPLEN;
2852 vt_reg_bits |= (adapter->num_vfs <<
2853 IXGBE_VT_CTL_POOL_SHIFT);
2854 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2855 IXGBE_WRITE_REG(hw, IXGBE_MRQC, 0);
2857 vf_shift = adapter->num_vfs % 32;
2858 reg_offset = adapter->num_vfs / 32;
2859 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), 0);
2860 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), 0);
2861 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), 0);
2862 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), 0);
2863 /* Enable only the PF's pool for Tx/Rx */
2864 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2865 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2866 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2869 /* Program MRQC for the distribution of queues */
2870 ixgbe_setup_mrqc(adapter);
2872 if (adapter->num_vfs) {
2873 /* Map PF MAC address in RAR Entry 0 to first pool
2874 * following VFs */
2875 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2877 /* Set up VF register offsets for selected VT Mode, i.e.
2878 * 64 VFs for SR-IOV */
2879 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2880 gcr_ext |= IXGBE_GCR_EXT_SRIOV;
2881 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2885 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2887 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2888 struct ixgbe_hw *hw = &adapter->hw;
2889 int pool_ndx = adapter->num_vfs;
2891 /* add VID to filter table */
2892 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
2895 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2897 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2898 struct ixgbe_hw *hw = &adapter->hw;
2899 int pool_ndx = adapter->num_vfs;
2901 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2902 ixgbe_irq_disable(adapter);
2904 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2906 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2907 ixgbe_irq_enable(adapter);
2909 /* remove VID from filter table */
2910 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
2914 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
2915 * @adapter: driver data
2917 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
2919 struct ixgbe_hw *hw = &adapter->hw;
2920 u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2921 int i, j;
2923 switch (hw->mac.type) {
2924 case ixgbe_mac_82598EB:
2925 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2926 #ifdef CONFIG_IXGBE_DCB
2927 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
2928 vlnctrl &= ~IXGBE_VLNCTRL_VME;
2929 #endif
2930 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2931 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2932 break;
2933 case ixgbe_mac_82599EB:
2934 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2935 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2936 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2937 #ifdef CONFIG_IXGBE_DCB
2938 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
2939 break;
2940 #endif
2941 for (i = 0; i < adapter->num_rx_queues; i++) {
2942 j = adapter->rx_ring[i]->reg_idx;
2943 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2944 vlnctrl &= ~IXGBE_RXDCTL_VME;
2945 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2947 break;
2948 default:
2949 break;
2954 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
2955 * @adapter: driver data
2957 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
2959 struct ixgbe_hw *hw = &adapter->hw;
2960 u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2961 int i, j;
2963 switch (hw->mac.type) {
2964 case ixgbe_mac_82598EB:
2965 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2966 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2967 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2968 break;
2969 case ixgbe_mac_82599EB:
2970 vlnctrl |= IXGBE_VLNCTRL_VFE;
2971 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2972 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2973 for (i = 0; i < adapter->num_rx_queues; i++) {
2974 j = adapter->rx_ring[i]->reg_idx;
2975 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2976 vlnctrl |= IXGBE_RXDCTL_VME;
2977 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2979 break;
2980 default:
2981 break;
2985 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2986 struct vlan_group *grp)
2988 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2990 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2991 ixgbe_irq_disable(adapter);
2992 adapter->vlgrp = grp;
2995 * For a DCB driver, always enable VLAN tag stripping so we can
2996 * still receive traffic from a DCB-enabled host even if we're
2997 * not in DCB mode.
2999 ixgbe_vlan_filter_enable(adapter);
3001 ixgbe_vlan_rx_add_vid(netdev, 0);
3003 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3004 ixgbe_irq_enable(adapter);
3007 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3009 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
3011 if (adapter->vlgrp) {
3012 u16 vid;
3013 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
3014 if (!vlan_group_get_device(adapter->vlgrp, vid))
3015 continue;
3016 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3022 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3023 * @netdev: network interface device structure
3025 * Writes unicast address list to the RAR table.
3026 * Returns: -ENOMEM on failure/insufficient address space
3027 * 0 on no addresses written
3028 * X on writing X addresses to the RAR table
3030 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3032 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3033 struct ixgbe_hw *hw = &adapter->hw;
3034 unsigned int vfn = adapter->num_vfs;
3035 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3036 int count = 0;
3038 /* return ENOMEM indicating insufficient memory for addresses */
3039 if (netdev_uc_count(netdev) > rar_entries)
3040 return -ENOMEM;
3042 if (!netdev_uc_empty(netdev) && rar_entries) {
3043 struct netdev_hw_addr *ha;
3044 /* return error if we do not support writing to RAR table */
3045 if (!hw->mac.ops.set_rar)
3046 return -ENOMEM;
3048 netdev_for_each_uc_addr(ha, netdev) {
3049 if (!rar_entries)
3050 break;
3051 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3052 vfn, IXGBE_RAH_AV);
3053 count++;
3056 /* write the addresses in reverse order to avoid write combining */
3057 for (; rar_entries > 0 ; rar_entries--)
3058 hw->mac.ops.clear_rar(hw, rar_entries);
3060 return count;
3064 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3065 * @netdev: network interface device structure
3067 * The set_rx_method entry point is called whenever the unicast/multicast
3068 * address list or the network interface flags are updated. This routine is
3069 * responsible for configuring the hardware for proper unicast, multicast and
3070 * promiscuous mode.
3072 void ixgbe_set_rx_mode(struct net_device *netdev)
3074 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3075 struct ixgbe_hw *hw = &adapter->hw;
3076 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3077 int count;
3079 /* Check for Promiscuous and All Multicast modes */
3081 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3083 /* set all bits that we expect to always be set */
3084 fctrl |= IXGBE_FCTRL_BAM;
3085 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3086 fctrl |= IXGBE_FCTRL_PMCF;
3088 /* clear the bits we are changing the status of */
3089 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3091 if (netdev->flags & IFF_PROMISC) {
3092 hw->addr_ctrl.user_set_promisc = true;
3093 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3094 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3095 /* don't hardware filter vlans in promisc mode */
3096 ixgbe_vlan_filter_disable(adapter);
3097 } else {
3098 if (netdev->flags & IFF_ALLMULTI) {
3099 fctrl |= IXGBE_FCTRL_MPE;
3100 vmolr |= IXGBE_VMOLR_MPE;
3101 } else {
3103 * Write addresses to the MTA, if the attempt fails
3104 * then we should just turn on promiscous mode so
3105 * that we can at least receive multicast traffic
3107 hw->mac.ops.update_mc_addr_list(hw, netdev);
3108 vmolr |= IXGBE_VMOLR_ROMPE;
3110 ixgbe_vlan_filter_enable(adapter);
3111 hw->addr_ctrl.user_set_promisc = false;
3113 * Write addresses to available RAR registers, if there is not
3114 * sufficient space to store all the addresses then enable
3115 * unicast promiscous mode
3117 count = ixgbe_write_uc_addr_list(netdev);
3118 if (count < 0) {
3119 fctrl |= IXGBE_FCTRL_UPE;
3120 vmolr |= IXGBE_VMOLR_ROPE;
3124 if (adapter->num_vfs) {
3125 ixgbe_restore_vf_multicasts(adapter);
3126 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3127 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3128 IXGBE_VMOLR_ROPE);
3129 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3132 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3135 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3137 int q_idx;
3138 struct ixgbe_q_vector *q_vector;
3139 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3141 /* legacy and MSI only use one vector */
3142 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3143 q_vectors = 1;
3145 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3146 struct napi_struct *napi;
3147 q_vector = adapter->q_vector[q_idx];
3148 napi = &q_vector->napi;
3149 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3150 if (!q_vector->rxr_count || !q_vector->txr_count) {
3151 if (q_vector->txr_count == 1)
3152 napi->poll = &ixgbe_clean_txonly;
3153 else if (q_vector->rxr_count == 1)
3154 napi->poll = &ixgbe_clean_rxonly;
3158 napi_enable(napi);
3162 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3164 int q_idx;
3165 struct ixgbe_q_vector *q_vector;
3166 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3168 /* legacy and MSI only use one vector */
3169 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3170 q_vectors = 1;
3172 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3173 q_vector = adapter->q_vector[q_idx];
3174 napi_disable(&q_vector->napi);
3178 #ifdef CONFIG_IXGBE_DCB
3180 * ixgbe_configure_dcb - Configure DCB hardware
3181 * @adapter: ixgbe adapter struct
3183 * This is called by the driver on open to configure the DCB hardware.
3184 * This is also called by the gennetlink interface when reconfiguring
3185 * the DCB state.
3187 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3189 struct ixgbe_hw *hw = &adapter->hw;
3190 u32 txdctl;
3191 int i, j;
3193 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3194 if (hw->mac.type == ixgbe_mac_82598EB)
3195 netif_set_gso_max_size(adapter->netdev, 65536);
3196 return;
3199 if (hw->mac.type == ixgbe_mac_82598EB)
3200 netif_set_gso_max_size(adapter->netdev, 32768);
3202 ixgbe_dcb_check_config(&adapter->dcb_cfg);
3203 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
3204 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
3206 /* reconfigure the hardware */
3207 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
3209 for (i = 0; i < adapter->num_tx_queues; i++) {
3210 j = adapter->tx_ring[i]->reg_idx;
3211 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3212 /* PThresh workaround for Tx hang with DFP enabled. */
3213 txdctl |= 32;
3214 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3216 /* Enable VLAN tag insert/strip */
3217 ixgbe_vlan_filter_enable(adapter);
3219 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3222 #endif
3223 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3225 struct net_device *netdev = adapter->netdev;
3226 struct ixgbe_hw *hw = &adapter->hw;
3227 int i;
3229 ixgbe_set_rx_mode(netdev);
3231 ixgbe_restore_vlan(adapter);
3232 #ifdef CONFIG_IXGBE_DCB
3233 ixgbe_configure_dcb(adapter);
3234 #endif
3236 #ifdef IXGBE_FCOE
3237 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3238 ixgbe_configure_fcoe(adapter);
3240 #endif /* IXGBE_FCOE */
3241 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3242 for (i = 0; i < adapter->num_tx_queues; i++)
3243 adapter->tx_ring[i]->atr_sample_rate =
3244 adapter->atr_sample_rate;
3245 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3246 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3247 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3250 ixgbe_configure_tx(adapter);
3251 ixgbe_configure_rx(adapter);
3252 for (i = 0; i < adapter->num_rx_queues; i++)
3253 ixgbe_alloc_rx_buffers(adapter, adapter->rx_ring[i],
3254 (adapter->rx_ring[i]->count - 1));
3257 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3259 switch (hw->phy.type) {
3260 case ixgbe_phy_sfp_avago:
3261 case ixgbe_phy_sfp_ftl:
3262 case ixgbe_phy_sfp_intel:
3263 case ixgbe_phy_sfp_unknown:
3264 case ixgbe_phy_sfp_passive_tyco:
3265 case ixgbe_phy_sfp_passive_unknown:
3266 case ixgbe_phy_sfp_active_unknown:
3267 case ixgbe_phy_sfp_ftl_active:
3268 return true;
3269 default:
3270 return false;
3275 * ixgbe_sfp_link_config - set up SFP+ link
3276 * @adapter: pointer to private adapter struct
3278 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3280 struct ixgbe_hw *hw = &adapter->hw;
3282 if (hw->phy.multispeed_fiber) {
3284 * In multispeed fiber setups, the device may not have
3285 * had a physical connection when the driver loaded.
3286 * If that's the case, the initial link configuration
3287 * couldn't get the MAC into 10G or 1G mode, so we'll
3288 * never have a link status change interrupt fire.
3289 * We need to try and force an autonegotiation
3290 * session, then bring up link.
3292 hw->mac.ops.setup_sfp(hw);
3293 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3294 schedule_work(&adapter->multispeed_fiber_task);
3295 } else {
3297 * Direct Attach Cu and non-multispeed fiber modules
3298 * still need to be configured properly prior to
3299 * attempting link.
3301 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3302 schedule_work(&adapter->sfp_config_module_task);
3307 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3308 * @hw: pointer to private hardware struct
3310 * Returns 0 on success, negative on failure
3312 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3314 u32 autoneg;
3315 bool negotiation, link_up = false;
3316 u32 ret = IXGBE_ERR_LINK_SETUP;
3318 if (hw->mac.ops.check_link)
3319 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3321 if (ret)
3322 goto link_cfg_out;
3324 if (hw->mac.ops.get_link_capabilities)
3325 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
3326 if (ret)
3327 goto link_cfg_out;
3329 if (hw->mac.ops.setup_link)
3330 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3331 link_cfg_out:
3332 return ret;
3335 #define IXGBE_MAX_RX_DESC_POLL 10
3336 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3337 int rxr)
3339 int j = adapter->rx_ring[rxr]->reg_idx;
3340 int k;
3342 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
3343 if (IXGBE_READ_REG(&adapter->hw,
3344 IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
3345 break;
3346 else
3347 msleep(1);
3349 if (k >= IXGBE_MAX_RX_DESC_POLL) {
3350 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3351 "the polling period\n", rxr);
3353 ixgbe_release_rx_desc(&adapter->hw, adapter->rx_ring[rxr],
3354 (adapter->rx_ring[rxr]->count - 1));
3357 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3359 struct ixgbe_hw *hw = &adapter->hw;
3360 int i, j = 0;
3361 int num_rx_rings = adapter->num_rx_queues;
3362 int err;
3363 u32 txdctl, rxdctl;
3364 u32 dmatxctl;
3365 u32 gpie;
3366 u32 ctrl_ext;
3368 ixgbe_get_hw_control(adapter);
3370 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
3371 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
3372 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3373 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
3374 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
3375 } else {
3376 /* MSI only */
3377 gpie = 0;
3379 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3380 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3381 gpie |= IXGBE_GPIE_VTMODE_64;
3383 /* XXX: to interrupt immediately for EICS writes, enable this */
3384 /* gpie |= IXGBE_GPIE_EIMEN; */
3385 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3388 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3390 * use EIAM to auto-mask when MSI-X interrupt is asserted
3391 * this saves a register write for every interrupt
3393 switch (hw->mac.type) {
3394 case ixgbe_mac_82598EB:
3395 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3396 break;
3397 default:
3398 case ixgbe_mac_82599EB:
3399 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3400 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3401 break;
3403 } else {
3404 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3405 * specifically only auto mask tx and rx interrupts */
3406 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3409 /* Enable Thermal over heat sensor interrupt */
3410 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3411 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3412 gpie |= IXGBE_SDP0_GPIEN;
3413 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3416 /* Enable fan failure interrupt if media type is copper */
3417 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3418 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3419 gpie |= IXGBE_SDP1_GPIEN;
3420 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3423 if (hw->mac.type == ixgbe_mac_82599EB) {
3424 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3425 gpie |= IXGBE_SDP1_GPIEN;
3426 gpie |= IXGBE_SDP2_GPIEN;
3427 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3430 if (hw->mac.type == ixgbe_mac_82599EB) {
3431 /* DMATXCTL.EN must be set after all Tx queue config is done */
3432 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3433 dmatxctl |= IXGBE_DMATXCTL_TE;
3434 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3436 for (i = 0; i < adapter->num_tx_queues; i++) {
3437 j = adapter->tx_ring[i]->reg_idx;
3438 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3439 if (adapter->rx_itr_setting == 0) {
3440 /* cannot set wthresh when itr==0 */
3441 txdctl &= ~0x007F0000;
3442 } else {
3443 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
3444 txdctl |= (8 << 16);
3446 txdctl |= IXGBE_TXDCTL_ENABLE;
3447 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3448 if (hw->mac.type == ixgbe_mac_82599EB) {
3449 int wait_loop = 10;
3450 /* poll for Tx Enable ready */
3451 do {
3452 msleep(1);
3453 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3454 } while (--wait_loop &&
3455 !(txdctl & IXGBE_TXDCTL_ENABLE));
3456 if (!wait_loop)
3457 e_err(drv, "Could not enable Tx Queue %d\n", j);
3461 for (i = 0; i < num_rx_rings; i++) {
3462 j = adapter->rx_ring[i]->reg_idx;
3463 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3464 /* enable PTHRESH=32 descriptors (half the internal cache)
3465 * and HTHRESH=0 descriptors (to minimize latency on fetch),
3466 * this also removes a pesky rx_no_buffer_count increment */
3467 rxdctl |= 0x0020;
3468 rxdctl |= IXGBE_RXDCTL_ENABLE;
3469 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
3470 if (hw->mac.type == ixgbe_mac_82599EB)
3471 ixgbe_rx_desc_queue_enable(adapter, i);
3473 /* enable all receives */
3474 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3475 if (hw->mac.type == ixgbe_mac_82598EB)
3476 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
3477 else
3478 rxdctl |= IXGBE_RXCTRL_RXEN;
3479 hw->mac.ops.enable_rx_dma(hw, rxdctl);
3481 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3482 ixgbe_configure_msix(adapter);
3483 else
3484 ixgbe_configure_msi_and_legacy(adapter);
3486 /* enable the optics */
3487 if (hw->phy.multispeed_fiber)
3488 hw->mac.ops.enable_tx_laser(hw);
3490 clear_bit(__IXGBE_DOWN, &adapter->state);
3491 ixgbe_napi_enable_all(adapter);
3493 /* clear any pending interrupts, may auto mask */
3494 IXGBE_READ_REG(hw, IXGBE_EICR);
3496 ixgbe_irq_enable(adapter);
3499 * If this adapter has a fan, check to see if we had a failure
3500 * before we enabled the interrupt.
3502 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3503 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3504 if (esdp & IXGBE_ESDP_SDP1)
3505 e_crit(drv, "Fan has stopped, replace the adapter\n");
3509 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3510 * arrived before interrupts were enabled but after probe. Such
3511 * devices wouldn't have their type identified yet. We need to
3512 * kick off the SFP+ module setup first, then try to bring up link.
3513 * If we're not hot-pluggable SFP+, we just need to configure link
3514 * and bring it up.
3516 if (hw->phy.type == ixgbe_phy_unknown) {
3517 err = hw->phy.ops.identify(hw);
3518 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3520 * Take the device down and schedule the sfp tasklet
3521 * which will unregister_netdev and log it.
3523 ixgbe_down(adapter);
3524 schedule_work(&adapter->sfp_config_module_task);
3525 return err;
3529 if (ixgbe_is_sfp(hw)) {
3530 ixgbe_sfp_link_config(adapter);
3531 } else {
3532 err = ixgbe_non_sfp_link_config(hw);
3533 if (err)
3534 e_err(probe, "link_config FAILED %d\n", err);
3537 for (i = 0; i < adapter->num_tx_queues; i++)
3538 set_bit(__IXGBE_FDIR_INIT_DONE,
3539 &(adapter->tx_ring[i]->reinit_state));
3541 /* enable transmits */
3542 netif_tx_start_all_queues(adapter->netdev);
3544 /* bring the link up in the watchdog, this could race with our first
3545 * link up interrupt but shouldn't be a problem */
3546 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3547 adapter->link_check_timeout = jiffies;
3548 mod_timer(&adapter->watchdog_timer, jiffies);
3550 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3551 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3552 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3553 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3555 return 0;
3558 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3560 WARN_ON(in_interrupt());
3561 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3562 msleep(1);
3563 ixgbe_down(adapter);
3565 * If SR-IOV enabled then wait a bit before bringing the adapter
3566 * back up to give the VFs time to respond to the reset. The
3567 * two second wait is based upon the watchdog timer cycle in
3568 * the VF driver.
3570 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3571 msleep(2000);
3572 ixgbe_up(adapter);
3573 clear_bit(__IXGBE_RESETTING, &adapter->state);
3576 int ixgbe_up(struct ixgbe_adapter *adapter)
3578 /* hardware has been reset, we need to reload some things */
3579 ixgbe_configure(adapter);
3581 return ixgbe_up_complete(adapter);
3584 void ixgbe_reset(struct ixgbe_adapter *adapter)
3586 struct ixgbe_hw *hw = &adapter->hw;
3587 int err;
3589 err = hw->mac.ops.init_hw(hw);
3590 switch (err) {
3591 case 0:
3592 case IXGBE_ERR_SFP_NOT_PRESENT:
3593 break;
3594 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3595 e_dev_err("master disable timed out\n");
3596 break;
3597 case IXGBE_ERR_EEPROM_VERSION:
3598 /* We are running on a pre-production device, log a warning */
3599 e_dev_warn("This device is a pre-production adapter/LOM. "
3600 "Please be aware there may be issuesassociated with "
3601 "your hardware. If you are experiencing problems "
3602 "please contact your Intel or hardware "
3603 "representative who provided you with this "
3604 "hardware.\n");
3605 break;
3606 default:
3607 e_dev_err("Hardware Error: %d\n", err);
3610 /* reprogram the RAR[0] in case user changed it. */
3611 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3612 IXGBE_RAH_AV);
3616 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3617 * @adapter: board private structure
3618 * @rx_ring: ring to free buffers from
3620 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
3621 struct ixgbe_ring *rx_ring)
3623 struct pci_dev *pdev = adapter->pdev;
3624 unsigned long size;
3625 unsigned int i;
3627 /* Free all the Rx ring sk_buffs */
3629 for (i = 0; i < rx_ring->count; i++) {
3630 struct ixgbe_rx_buffer *rx_buffer_info;
3632 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3633 if (rx_buffer_info->dma) {
3634 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
3635 rx_ring->rx_buf_len,
3636 DMA_FROM_DEVICE);
3637 rx_buffer_info->dma = 0;
3639 if (rx_buffer_info->skb) {
3640 struct sk_buff *skb = rx_buffer_info->skb;
3641 rx_buffer_info->skb = NULL;
3642 do {
3643 struct sk_buff *this = skb;
3644 if (IXGBE_RSC_CB(this)->delay_unmap) {
3645 dma_unmap_single(&pdev->dev,
3646 IXGBE_RSC_CB(this)->dma,
3647 rx_ring->rx_buf_len,
3648 DMA_FROM_DEVICE);
3649 IXGBE_RSC_CB(this)->dma = 0;
3650 IXGBE_RSC_CB(skb)->delay_unmap = false;
3652 skb = skb->prev;
3653 dev_kfree_skb(this);
3654 } while (skb);
3656 if (!rx_buffer_info->page)
3657 continue;
3658 if (rx_buffer_info->page_dma) {
3659 dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
3660 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3661 rx_buffer_info->page_dma = 0;
3663 put_page(rx_buffer_info->page);
3664 rx_buffer_info->page = NULL;
3665 rx_buffer_info->page_offset = 0;
3668 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3669 memset(rx_ring->rx_buffer_info, 0, size);
3671 /* Zero out the descriptor ring */
3672 memset(rx_ring->desc, 0, rx_ring->size);
3674 rx_ring->next_to_clean = 0;
3675 rx_ring->next_to_use = 0;
3677 if (rx_ring->head)
3678 writel(0, adapter->hw.hw_addr + rx_ring->head);
3679 if (rx_ring->tail)
3680 writel(0, adapter->hw.hw_addr + rx_ring->tail);
3684 * ixgbe_clean_tx_ring - Free Tx Buffers
3685 * @adapter: board private structure
3686 * @tx_ring: ring to be cleaned
3688 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
3689 struct ixgbe_ring *tx_ring)
3691 struct ixgbe_tx_buffer *tx_buffer_info;
3692 unsigned long size;
3693 unsigned int i;
3695 /* Free all the Tx ring sk_buffs */
3697 for (i = 0; i < tx_ring->count; i++) {
3698 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3699 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3702 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3703 memset(tx_ring->tx_buffer_info, 0, size);
3705 /* Zero out the descriptor ring */
3706 memset(tx_ring->desc, 0, tx_ring->size);
3708 tx_ring->next_to_use = 0;
3709 tx_ring->next_to_clean = 0;
3711 if (tx_ring->head)
3712 writel(0, adapter->hw.hw_addr + tx_ring->head);
3713 if (tx_ring->tail)
3714 writel(0, adapter->hw.hw_addr + tx_ring->tail);
3718 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3719 * @adapter: board private structure
3721 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3723 int i;
3725 for (i = 0; i < adapter->num_rx_queues; i++)
3726 ixgbe_clean_rx_ring(adapter, adapter->rx_ring[i]);
3730 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3731 * @adapter: board private structure
3733 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3735 int i;
3737 for (i = 0; i < adapter->num_tx_queues; i++)
3738 ixgbe_clean_tx_ring(adapter, adapter->tx_ring[i]);
3741 void ixgbe_down(struct ixgbe_adapter *adapter)
3743 struct net_device *netdev = adapter->netdev;
3744 struct ixgbe_hw *hw = &adapter->hw;
3745 u32 rxctrl;
3746 u32 txdctl;
3747 int i, j;
3749 /* signal that we are down to the interrupt handler */
3750 set_bit(__IXGBE_DOWN, &adapter->state);
3752 /* disable receive for all VFs and wait one second */
3753 if (adapter->num_vfs) {
3754 /* ping all the active vfs to let them know we are going down */
3755 ixgbe_ping_all_vfs(adapter);
3757 /* Disable all VFTE/VFRE TX/RX */
3758 ixgbe_disable_tx_rx(adapter);
3760 /* Mark all the VFs as inactive */
3761 for (i = 0 ; i < adapter->num_vfs; i++)
3762 adapter->vfinfo[i].clear_to_send = 0;
3765 /* disable receives */
3766 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3767 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3769 IXGBE_WRITE_FLUSH(hw);
3770 msleep(10);
3772 netif_tx_stop_all_queues(netdev);
3774 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3775 del_timer_sync(&adapter->sfp_timer);
3776 del_timer_sync(&adapter->watchdog_timer);
3777 cancel_work_sync(&adapter->watchdog_task);
3779 netif_carrier_off(netdev);
3780 netif_tx_disable(netdev);
3782 ixgbe_irq_disable(adapter);
3784 ixgbe_napi_disable_all(adapter);
3786 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3787 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3788 cancel_work_sync(&adapter->fdir_reinit_task);
3790 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3791 cancel_work_sync(&adapter->check_overtemp_task);
3793 /* disable transmits in the hardware now that interrupts are off */
3794 for (i = 0; i < adapter->num_tx_queues; i++) {
3795 j = adapter->tx_ring[i]->reg_idx;
3796 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3797 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3798 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3800 /* Disable the Tx DMA engine on 82599 */
3801 if (hw->mac.type == ixgbe_mac_82599EB)
3802 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3803 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3804 ~IXGBE_DMATXCTL_TE));
3806 /* power down the optics */
3807 if (hw->phy.multispeed_fiber)
3808 hw->mac.ops.disable_tx_laser(hw);
3810 /* clear n-tuple filters that are cached */
3811 ethtool_ntuple_flush(netdev);
3813 if (!pci_channel_offline(adapter->pdev))
3814 ixgbe_reset(adapter);
3815 ixgbe_clean_all_tx_rings(adapter);
3816 ixgbe_clean_all_rx_rings(adapter);
3818 #ifdef CONFIG_IXGBE_DCA
3819 /* since we reset the hardware DCA settings were cleared */
3820 ixgbe_setup_dca(adapter);
3821 #endif
3825 * ixgbe_poll - NAPI Rx polling callback
3826 * @napi: structure for representing this polling device
3827 * @budget: how many packets driver is allowed to clean
3829 * This function is used for legacy and MSI, NAPI mode
3831 static int ixgbe_poll(struct napi_struct *napi, int budget)
3833 struct ixgbe_q_vector *q_vector =
3834 container_of(napi, struct ixgbe_q_vector, napi);
3835 struct ixgbe_adapter *adapter = q_vector->adapter;
3836 int tx_clean_complete, work_done = 0;
3838 #ifdef CONFIG_IXGBE_DCA
3839 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3840 ixgbe_update_tx_dca(adapter, adapter->tx_ring[0]);
3841 ixgbe_update_rx_dca(adapter, adapter->rx_ring[0]);
3843 #endif
3845 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3846 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3848 if (!tx_clean_complete)
3849 work_done = budget;
3851 /* If budget not fully consumed, exit the polling mode */
3852 if (work_done < budget) {
3853 napi_complete(napi);
3854 if (adapter->rx_itr_setting & 1)
3855 ixgbe_set_itr(adapter);
3856 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3857 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3859 return work_done;
3863 * ixgbe_tx_timeout - Respond to a Tx Hang
3864 * @netdev: network interface device structure
3866 static void ixgbe_tx_timeout(struct net_device *netdev)
3868 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3870 /* Do the reset outside of interrupt context */
3871 schedule_work(&adapter->reset_task);
3874 static void ixgbe_reset_task(struct work_struct *work)
3876 struct ixgbe_adapter *adapter;
3877 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3879 /* If we're already down or resetting, just bail */
3880 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3881 test_bit(__IXGBE_RESETTING, &adapter->state))
3882 return;
3884 adapter->tx_timeout_count++;
3886 ixgbe_dump(adapter);
3887 netdev_err(adapter->netdev, "Reset adapter\n");
3888 ixgbe_reinit_locked(adapter);
3891 #ifdef CONFIG_IXGBE_DCB
3892 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3894 bool ret = false;
3895 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3897 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3898 return ret;
3900 f->mask = 0x7 << 3;
3901 adapter->num_rx_queues = f->indices;
3902 adapter->num_tx_queues = f->indices;
3903 ret = true;
3905 return ret;
3907 #endif
3910 * ixgbe_set_rss_queues: Allocate queues for RSS
3911 * @adapter: board private structure to initialize
3913 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3914 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3917 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3919 bool ret = false;
3920 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3922 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3923 f->mask = 0xF;
3924 adapter->num_rx_queues = f->indices;
3925 adapter->num_tx_queues = f->indices;
3926 ret = true;
3927 } else {
3928 ret = false;
3931 return ret;
3935 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3936 * @adapter: board private structure to initialize
3938 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3939 * to the original CPU that initiated the Tx session. This runs in addition
3940 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3941 * Rx load across CPUs using RSS.
3944 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3946 bool ret = false;
3947 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3949 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3950 f_fdir->mask = 0;
3952 /* Flow Director must have RSS enabled */
3953 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3954 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3955 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3956 adapter->num_tx_queues = f_fdir->indices;
3957 adapter->num_rx_queues = f_fdir->indices;
3958 ret = true;
3959 } else {
3960 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3961 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3963 return ret;
3966 #ifdef IXGBE_FCOE
3968 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3969 * @adapter: board private structure to initialize
3971 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3972 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3973 * rx queues out of the max number of rx queues, instead, it is used as the
3974 * index of the first rx queue used by FCoE.
3977 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3979 bool ret = false;
3980 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3982 f->indices = min((int)num_online_cpus(), f->indices);
3983 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3984 adapter->num_rx_queues = 1;
3985 adapter->num_tx_queues = 1;
3986 #ifdef CONFIG_IXGBE_DCB
3987 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3988 e_info(probe, "FCoE enabled with DCB\n");
3989 ixgbe_set_dcb_queues(adapter);
3991 #endif
3992 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3993 e_info(probe, "FCoE enabled with RSS\n");
3994 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3995 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3996 ixgbe_set_fdir_queues(adapter);
3997 else
3998 ixgbe_set_rss_queues(adapter);
4000 /* adding FCoE rx rings to the end */
4001 f->mask = adapter->num_rx_queues;
4002 adapter->num_rx_queues += f->indices;
4003 adapter->num_tx_queues += f->indices;
4005 ret = true;
4008 return ret;
4011 #endif /* IXGBE_FCOE */
4013 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4014 * @adapter: board private structure to initialize
4016 * IOV doesn't actually use anything, so just NAK the
4017 * request for now and let the other queue routines
4018 * figure out what to do.
4020 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4022 return false;
4026 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4027 * @adapter: board private structure to initialize
4029 * This is the top level queue allocation routine. The order here is very
4030 * important, starting with the "most" number of features turned on at once,
4031 * and ending with the smallest set of features. This way large combinations
4032 * can be allocated if they're turned on, and smaller combinations are the
4033 * fallthrough conditions.
4036 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4038 /* Start with base case */
4039 adapter->num_rx_queues = 1;
4040 adapter->num_tx_queues = 1;
4041 adapter->num_rx_pools = adapter->num_rx_queues;
4042 adapter->num_rx_queues_per_pool = 1;
4044 if (ixgbe_set_sriov_queues(adapter))
4045 return;
4047 #ifdef IXGBE_FCOE
4048 if (ixgbe_set_fcoe_queues(adapter))
4049 goto done;
4051 #endif /* IXGBE_FCOE */
4052 #ifdef CONFIG_IXGBE_DCB
4053 if (ixgbe_set_dcb_queues(adapter))
4054 goto done;
4056 #endif
4057 if (ixgbe_set_fdir_queues(adapter))
4058 goto done;
4060 if (ixgbe_set_rss_queues(adapter))
4061 goto done;
4063 /* fallback to base case */
4064 adapter->num_rx_queues = 1;
4065 adapter->num_tx_queues = 1;
4067 done:
4068 /* Notify the stack of the (possibly) reduced Tx Queue count. */
4069 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4072 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4073 int vectors)
4075 int err, vector_threshold;
4077 /* We'll want at least 3 (vector_threshold):
4078 * 1) TxQ[0] Cleanup
4079 * 2) RxQ[0] Cleanup
4080 * 3) Other (Link Status Change, etc.)
4081 * 4) TCP Timer (optional)
4083 vector_threshold = MIN_MSIX_COUNT;
4085 /* The more we get, the more we will assign to Tx/Rx Cleanup
4086 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4087 * Right now, we simply care about how many we'll get; we'll
4088 * set them up later while requesting irq's.
4090 while (vectors >= vector_threshold) {
4091 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4092 vectors);
4093 if (!err) /* Success in acquiring all requested vectors. */
4094 break;
4095 else if (err < 0)
4096 vectors = 0; /* Nasty failure, quit now */
4097 else /* err == number of vectors we should try again with */
4098 vectors = err;
4101 if (vectors < vector_threshold) {
4102 /* Can't allocate enough MSI-X interrupts? Oh well.
4103 * This just means we'll go with either a single MSI
4104 * vector or fall back to legacy interrupts.
4106 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4107 "Unable to allocate MSI-X interrupts\n");
4108 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4109 kfree(adapter->msix_entries);
4110 adapter->msix_entries = NULL;
4111 } else {
4112 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4114 * Adjust for only the vectors we'll use, which is minimum
4115 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4116 * vectors we were allocated.
4118 adapter->num_msix_vectors = min(vectors,
4119 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4124 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4125 * @adapter: board private structure to initialize
4127 * Cache the descriptor ring offsets for RSS to the assigned rings.
4130 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4132 int i;
4133 bool ret = false;
4135 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4136 for (i = 0; i < adapter->num_rx_queues; i++)
4137 adapter->rx_ring[i]->reg_idx = i;
4138 for (i = 0; i < adapter->num_tx_queues; i++)
4139 adapter->tx_ring[i]->reg_idx = i;
4140 ret = true;
4141 } else {
4142 ret = false;
4145 return ret;
4148 #ifdef CONFIG_IXGBE_DCB
4150 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4151 * @adapter: board private structure to initialize
4153 * Cache the descriptor ring offsets for DCB to the assigned rings.
4156 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4158 int i;
4159 bool ret = false;
4160 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4162 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4163 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
4164 /* the number of queues is assumed to be symmetric */
4165 for (i = 0; i < dcb_i; i++) {
4166 adapter->rx_ring[i]->reg_idx = i << 3;
4167 adapter->tx_ring[i]->reg_idx = i << 2;
4169 ret = true;
4170 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
4171 if (dcb_i == 8) {
4173 * Tx TC0 starts at: descriptor queue 0
4174 * Tx TC1 starts at: descriptor queue 32
4175 * Tx TC2 starts at: descriptor queue 64
4176 * Tx TC3 starts at: descriptor queue 80
4177 * Tx TC4 starts at: descriptor queue 96
4178 * Tx TC5 starts at: descriptor queue 104
4179 * Tx TC6 starts at: descriptor queue 112
4180 * Tx TC7 starts at: descriptor queue 120
4182 * Rx TC0-TC7 are offset by 16 queues each
4184 for (i = 0; i < 3; i++) {
4185 adapter->tx_ring[i]->reg_idx = i << 5;
4186 adapter->rx_ring[i]->reg_idx = i << 4;
4188 for ( ; i < 5; i++) {
4189 adapter->tx_ring[i]->reg_idx =
4190 ((i + 2) << 4);
4191 adapter->rx_ring[i]->reg_idx = i << 4;
4193 for ( ; i < dcb_i; i++) {
4194 adapter->tx_ring[i]->reg_idx =
4195 ((i + 8) << 3);
4196 adapter->rx_ring[i]->reg_idx = i << 4;
4199 ret = true;
4200 } else if (dcb_i == 4) {
4202 * Tx TC0 starts at: descriptor queue 0
4203 * Tx TC1 starts at: descriptor queue 64
4204 * Tx TC2 starts at: descriptor queue 96
4205 * Tx TC3 starts at: descriptor queue 112
4207 * Rx TC0-TC3 are offset by 32 queues each
4209 adapter->tx_ring[0]->reg_idx = 0;
4210 adapter->tx_ring[1]->reg_idx = 64;
4211 adapter->tx_ring[2]->reg_idx = 96;
4212 adapter->tx_ring[3]->reg_idx = 112;
4213 for (i = 0 ; i < dcb_i; i++)
4214 adapter->rx_ring[i]->reg_idx = i << 5;
4216 ret = true;
4217 } else {
4218 ret = false;
4220 } else {
4221 ret = false;
4223 } else {
4224 ret = false;
4227 return ret;
4229 #endif
4232 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4233 * @adapter: board private structure to initialize
4235 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4238 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4240 int i;
4241 bool ret = false;
4243 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4244 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4245 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4246 for (i = 0; i < adapter->num_rx_queues; i++)
4247 adapter->rx_ring[i]->reg_idx = i;
4248 for (i = 0; i < adapter->num_tx_queues; i++)
4249 adapter->tx_ring[i]->reg_idx = i;
4250 ret = true;
4253 return ret;
4256 #ifdef IXGBE_FCOE
4258 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4259 * @adapter: board private structure to initialize
4261 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4264 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4266 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
4267 bool ret = false;
4268 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4270 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4271 #ifdef CONFIG_IXGBE_DCB
4272 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4273 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4275 ixgbe_cache_ring_dcb(adapter);
4276 /* find out queues in TC for FCoE */
4277 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4278 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4280 * In 82599, the number of Tx queues for each traffic
4281 * class for both 8-TC and 4-TC modes are:
4282 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4283 * 8 TCs: 32 32 16 16 8 8 8 8
4284 * 4 TCs: 64 64 32 32
4285 * We have max 8 queues for FCoE, where 8 the is
4286 * FCoE redirection table size. If TC for FCoE is
4287 * less than or equal to TC3, we have enough queues
4288 * to add max of 8 queues for FCoE, so we start FCoE
4289 * tx descriptor from the next one, i.e., reg_idx + 1.
4290 * If TC for FCoE is above TC3, implying 8 TC mode,
4291 * and we need 8 for FCoE, we have to take all queues
4292 * in that traffic class for FCoE.
4294 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4295 fcoe_tx_i--;
4297 #endif /* CONFIG_IXGBE_DCB */
4298 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4299 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4300 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4301 ixgbe_cache_ring_fdir(adapter);
4302 else
4303 ixgbe_cache_ring_rss(adapter);
4305 fcoe_rx_i = f->mask;
4306 fcoe_tx_i = f->mask;
4308 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4309 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4310 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4312 ret = true;
4314 return ret;
4317 #endif /* IXGBE_FCOE */
4319 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4320 * @adapter: board private structure to initialize
4322 * SR-IOV doesn't use any descriptor rings but changes the default if
4323 * no other mapping is used.
4326 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4328 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4329 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4330 if (adapter->num_vfs)
4331 return true;
4332 else
4333 return false;
4337 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4338 * @adapter: board private structure to initialize
4340 * Once we know the feature-set enabled for the device, we'll cache
4341 * the register offset the descriptor ring is assigned to.
4343 * Note, the order the various feature calls is important. It must start with
4344 * the "most" features enabled at the same time, then trickle down to the
4345 * least amount of features turned on at once.
4347 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4349 /* start with default case */
4350 adapter->rx_ring[0]->reg_idx = 0;
4351 adapter->tx_ring[0]->reg_idx = 0;
4353 if (ixgbe_cache_ring_sriov(adapter))
4354 return;
4356 #ifdef IXGBE_FCOE
4357 if (ixgbe_cache_ring_fcoe(adapter))
4358 return;
4360 #endif /* IXGBE_FCOE */
4361 #ifdef CONFIG_IXGBE_DCB
4362 if (ixgbe_cache_ring_dcb(adapter))
4363 return;
4365 #endif
4366 if (ixgbe_cache_ring_fdir(adapter))
4367 return;
4369 if (ixgbe_cache_ring_rss(adapter))
4370 return;
4374 * ixgbe_alloc_queues - Allocate memory for all rings
4375 * @adapter: board private structure to initialize
4377 * We allocate one ring per queue at run-time since we don't know the
4378 * number of queues at compile-time. The polling_netdev array is
4379 * intended for Multiqueue, but should work fine with a single queue.
4381 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4383 int i;
4384 int orig_node = adapter->node;
4386 for (i = 0; i < adapter->num_tx_queues; i++) {
4387 struct ixgbe_ring *ring = adapter->tx_ring[i];
4388 if (orig_node == -1) {
4389 int cur_node = next_online_node(adapter->node);
4390 if (cur_node == MAX_NUMNODES)
4391 cur_node = first_online_node;
4392 adapter->node = cur_node;
4394 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4395 adapter->node);
4396 if (!ring)
4397 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4398 if (!ring)
4399 goto err_tx_ring_allocation;
4400 ring->count = adapter->tx_ring_count;
4401 ring->queue_index = i;
4402 ring->numa_node = adapter->node;
4404 adapter->tx_ring[i] = ring;
4407 /* Restore the adapter's original node */
4408 adapter->node = orig_node;
4410 for (i = 0; i < adapter->num_rx_queues; i++) {
4411 struct ixgbe_ring *ring = adapter->rx_ring[i];
4412 if (orig_node == -1) {
4413 int cur_node = next_online_node(adapter->node);
4414 if (cur_node == MAX_NUMNODES)
4415 cur_node = first_online_node;
4416 adapter->node = cur_node;
4418 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4419 adapter->node);
4420 if (!ring)
4421 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4422 if (!ring)
4423 goto err_rx_ring_allocation;
4424 ring->count = adapter->rx_ring_count;
4425 ring->queue_index = i;
4426 ring->numa_node = adapter->node;
4428 adapter->rx_ring[i] = ring;
4431 /* Restore the adapter's original node */
4432 adapter->node = orig_node;
4434 ixgbe_cache_ring_register(adapter);
4436 return 0;
4438 err_rx_ring_allocation:
4439 for (i = 0; i < adapter->num_tx_queues; i++)
4440 kfree(adapter->tx_ring[i]);
4441 err_tx_ring_allocation:
4442 return -ENOMEM;
4446 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4447 * @adapter: board private structure to initialize
4449 * Attempt to configure the interrupts using the best available
4450 * capabilities of the hardware and the kernel.
4452 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4454 struct ixgbe_hw *hw = &adapter->hw;
4455 int err = 0;
4456 int vector, v_budget;
4459 * It's easy to be greedy for MSI-X vectors, but it really
4460 * doesn't do us much good if we have a lot more vectors
4461 * than CPU's. So let's be conservative and only ask for
4462 * (roughly) the same number of vectors as there are CPU's.
4464 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4465 (int)num_online_cpus()) + NON_Q_VECTORS;
4468 * At the same time, hardware can only support a maximum of
4469 * hw.mac->max_msix_vectors vectors. With features
4470 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4471 * descriptor queues supported by our device. Thus, we cap it off in
4472 * those rare cases where the cpu count also exceeds our vector limit.
4474 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4476 /* A failure in MSI-X entry allocation isn't fatal, but it does
4477 * mean we disable MSI-X capabilities of the adapter. */
4478 adapter->msix_entries = kcalloc(v_budget,
4479 sizeof(struct msix_entry), GFP_KERNEL);
4480 if (adapter->msix_entries) {
4481 for (vector = 0; vector < v_budget; vector++)
4482 adapter->msix_entries[vector].entry = vector;
4484 ixgbe_acquire_msix_vectors(adapter, v_budget);
4486 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4487 goto out;
4490 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4491 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4492 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4493 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4494 adapter->atr_sample_rate = 0;
4495 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4496 ixgbe_disable_sriov(adapter);
4498 ixgbe_set_num_queues(adapter);
4500 err = pci_enable_msi(adapter->pdev);
4501 if (!err) {
4502 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4503 } else {
4504 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4505 "Unable to allocate MSI interrupt, "
4506 "falling back to legacy. Error: %d\n", err);
4507 /* reset err */
4508 err = 0;
4511 out:
4512 return err;
4516 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4517 * @adapter: board private structure to initialize
4519 * We allocate one q_vector per queue interrupt. If allocation fails we
4520 * return -ENOMEM.
4522 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4524 int q_idx, num_q_vectors;
4525 struct ixgbe_q_vector *q_vector;
4526 int napi_vectors;
4527 int (*poll)(struct napi_struct *, int);
4529 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4530 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4531 napi_vectors = adapter->num_rx_queues;
4532 poll = &ixgbe_clean_rxtx_many;
4533 } else {
4534 num_q_vectors = 1;
4535 napi_vectors = 1;
4536 poll = &ixgbe_poll;
4539 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4540 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4541 GFP_KERNEL, adapter->node);
4542 if (!q_vector)
4543 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4544 GFP_KERNEL);
4545 if (!q_vector)
4546 goto err_out;
4547 q_vector->adapter = adapter;
4548 if (q_vector->txr_count && !q_vector->rxr_count)
4549 q_vector->eitr = adapter->tx_eitr_param;
4550 else
4551 q_vector->eitr = adapter->rx_eitr_param;
4552 q_vector->v_idx = q_idx;
4553 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4554 adapter->q_vector[q_idx] = q_vector;
4557 return 0;
4559 err_out:
4560 while (q_idx) {
4561 q_idx--;
4562 q_vector = adapter->q_vector[q_idx];
4563 netif_napi_del(&q_vector->napi);
4564 kfree(q_vector);
4565 adapter->q_vector[q_idx] = NULL;
4567 return -ENOMEM;
4571 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4572 * @adapter: board private structure to initialize
4574 * This function frees the memory allocated to the q_vectors. In addition if
4575 * NAPI is enabled it will delete any references to the NAPI struct prior
4576 * to freeing the q_vector.
4578 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4580 int q_idx, num_q_vectors;
4582 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4583 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4584 else
4585 num_q_vectors = 1;
4587 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4588 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4589 adapter->q_vector[q_idx] = NULL;
4590 netif_napi_del(&q_vector->napi);
4591 kfree(q_vector);
4595 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4597 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4598 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4599 pci_disable_msix(adapter->pdev);
4600 kfree(adapter->msix_entries);
4601 adapter->msix_entries = NULL;
4602 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4603 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4604 pci_disable_msi(adapter->pdev);
4609 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4610 * @adapter: board private structure to initialize
4612 * We determine which interrupt scheme to use based on...
4613 * - Kernel support (MSI, MSI-X)
4614 * - which can be user-defined (via MODULE_PARAM)
4615 * - Hardware queue count (num_*_queues)
4616 * - defined by miscellaneous hardware support/features (RSS, etc.)
4618 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4620 int err;
4622 /* Number of supported queues */
4623 ixgbe_set_num_queues(adapter);
4625 err = ixgbe_set_interrupt_capability(adapter);
4626 if (err) {
4627 e_dev_err("Unable to setup interrupt capabilities\n");
4628 goto err_set_interrupt;
4631 err = ixgbe_alloc_q_vectors(adapter);
4632 if (err) {
4633 e_dev_err("Unable to allocate memory for queue vectors\n");
4634 goto err_alloc_q_vectors;
4637 err = ixgbe_alloc_queues(adapter);
4638 if (err) {
4639 e_dev_err("Unable to allocate memory for queues\n");
4640 goto err_alloc_queues;
4643 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4644 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4645 adapter->num_rx_queues, adapter->num_tx_queues);
4647 set_bit(__IXGBE_DOWN, &adapter->state);
4649 return 0;
4651 err_alloc_queues:
4652 ixgbe_free_q_vectors(adapter);
4653 err_alloc_q_vectors:
4654 ixgbe_reset_interrupt_capability(adapter);
4655 err_set_interrupt:
4656 return err;
4660 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4661 * @adapter: board private structure to clear interrupt scheme on
4663 * We go through and clear interrupt specific resources and reset the structure
4664 * to pre-load conditions
4666 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4668 int i;
4670 for (i = 0; i < adapter->num_tx_queues; i++) {
4671 kfree(adapter->tx_ring[i]);
4672 adapter->tx_ring[i] = NULL;
4674 for (i = 0; i < adapter->num_rx_queues; i++) {
4675 kfree(adapter->rx_ring[i]);
4676 adapter->rx_ring[i] = NULL;
4679 ixgbe_free_q_vectors(adapter);
4680 ixgbe_reset_interrupt_capability(adapter);
4684 * ixgbe_sfp_timer - worker thread to find a missing module
4685 * @data: pointer to our adapter struct
4687 static void ixgbe_sfp_timer(unsigned long data)
4689 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4692 * Do the sfp_timer outside of interrupt context due to the
4693 * delays that sfp+ detection requires
4695 schedule_work(&adapter->sfp_task);
4699 * ixgbe_sfp_task - worker thread to find a missing module
4700 * @work: pointer to work_struct containing our data
4702 static void ixgbe_sfp_task(struct work_struct *work)
4704 struct ixgbe_adapter *adapter = container_of(work,
4705 struct ixgbe_adapter,
4706 sfp_task);
4707 struct ixgbe_hw *hw = &adapter->hw;
4709 if ((hw->phy.type == ixgbe_phy_nl) &&
4710 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4711 s32 ret = hw->phy.ops.identify_sfp(hw);
4712 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4713 goto reschedule;
4714 ret = hw->phy.ops.reset(hw);
4715 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4716 e_dev_err("failed to initialize because an unsupported "
4717 "SFP+ module type was detected.\n");
4718 e_dev_err("Reload the driver after installing a "
4719 "supported module.\n");
4720 unregister_netdev(adapter->netdev);
4721 } else {
4722 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
4724 /* don't need this routine any more */
4725 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4727 return;
4728 reschedule:
4729 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4730 mod_timer(&adapter->sfp_timer,
4731 round_jiffies(jiffies + (2 * HZ)));
4735 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4736 * @adapter: board private structure to initialize
4738 * ixgbe_sw_init initializes the Adapter private data structure.
4739 * Fields are initialized based on PCI device information and
4740 * OS network device settings (MTU size).
4742 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4744 struct ixgbe_hw *hw = &adapter->hw;
4745 struct pci_dev *pdev = adapter->pdev;
4746 struct net_device *dev = adapter->netdev;
4747 unsigned int rss;
4748 #ifdef CONFIG_IXGBE_DCB
4749 int j;
4750 struct tc_configuration *tc;
4751 #endif
4753 /* PCI config space info */
4755 hw->vendor_id = pdev->vendor;
4756 hw->device_id = pdev->device;
4757 hw->revision_id = pdev->revision;
4758 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4759 hw->subsystem_device_id = pdev->subsystem_device;
4761 /* Set capability flags */
4762 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4763 adapter->ring_feature[RING_F_RSS].indices = rss;
4764 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4765 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4766 if (hw->mac.type == ixgbe_mac_82598EB) {
4767 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4768 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4769 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4770 } else if (hw->mac.type == ixgbe_mac_82599EB) {
4771 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4772 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4773 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4774 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4775 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4776 if (dev->features & NETIF_F_NTUPLE) {
4777 /* Flow Director perfect filter enabled */
4778 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4779 adapter->atr_sample_rate = 0;
4780 spin_lock_init(&adapter->fdir_perfect_lock);
4781 } else {
4782 /* Flow Director hash filters enabled */
4783 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4784 adapter->atr_sample_rate = 20;
4786 adapter->ring_feature[RING_F_FDIR].indices =
4787 IXGBE_MAX_FDIR_INDICES;
4788 adapter->fdir_pballoc = 0;
4789 #ifdef IXGBE_FCOE
4790 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4791 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4792 adapter->ring_feature[RING_F_FCOE].indices = 0;
4793 #ifdef CONFIG_IXGBE_DCB
4794 /* Default traffic class to use for FCoE */
4795 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4796 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4797 #endif
4798 #endif /* IXGBE_FCOE */
4801 #ifdef CONFIG_IXGBE_DCB
4802 /* Configure DCB traffic classes */
4803 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4804 tc = &adapter->dcb_cfg.tc_config[j];
4805 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4806 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4807 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4808 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4809 tc->dcb_pfc = pfc_disabled;
4811 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4812 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4813 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4814 adapter->dcb_cfg.pfc_mode_enable = false;
4815 adapter->dcb_cfg.round_robin_enable = false;
4816 adapter->dcb_set_bitmap = 0x00;
4817 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4818 adapter->ring_feature[RING_F_DCB].indices);
4820 #endif
4822 /* default flow control settings */
4823 hw->fc.requested_mode = ixgbe_fc_full;
4824 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4825 #ifdef CONFIG_DCB
4826 adapter->last_lfc_mode = hw->fc.current_mode;
4827 #endif
4828 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
4829 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
4830 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4831 hw->fc.send_xon = true;
4832 hw->fc.disable_fc_autoneg = false;
4834 /* enable itr by default in dynamic mode */
4835 adapter->rx_itr_setting = 1;
4836 adapter->rx_eitr_param = 20000;
4837 adapter->tx_itr_setting = 1;
4838 adapter->tx_eitr_param = 10000;
4840 /* set defaults for eitr in MegaBytes */
4841 adapter->eitr_low = 10;
4842 adapter->eitr_high = 20;
4844 /* set default ring sizes */
4845 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4846 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4848 /* initialize eeprom parameters */
4849 if (ixgbe_init_eeprom_params_generic(hw)) {
4850 e_dev_err("EEPROM initialization failed\n");
4851 return -EIO;
4854 /* enable rx csum by default */
4855 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4857 /* get assigned NUMA node */
4858 adapter->node = dev_to_node(&pdev->dev);
4860 set_bit(__IXGBE_DOWN, &adapter->state);
4862 return 0;
4866 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4867 * @adapter: board private structure
4868 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4870 * Return 0 on success, negative on failure
4872 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4873 struct ixgbe_ring *tx_ring)
4875 struct pci_dev *pdev = adapter->pdev;
4876 int size;
4878 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4879 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4880 if (!tx_ring->tx_buffer_info)
4881 tx_ring->tx_buffer_info = vmalloc(size);
4882 if (!tx_ring->tx_buffer_info)
4883 goto err;
4884 memset(tx_ring->tx_buffer_info, 0, size);
4886 /* round up to nearest 4K */
4887 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4888 tx_ring->size = ALIGN(tx_ring->size, 4096);
4890 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
4891 &tx_ring->dma, GFP_KERNEL);
4892 if (!tx_ring->desc)
4893 goto err;
4895 tx_ring->next_to_use = 0;
4896 tx_ring->next_to_clean = 0;
4897 tx_ring->work_limit = tx_ring->count;
4898 return 0;
4900 err:
4901 vfree(tx_ring->tx_buffer_info);
4902 tx_ring->tx_buffer_info = NULL;
4903 e_err(probe, "Unable to allocate memory for the Tx descriptor ring\n");
4904 return -ENOMEM;
4908 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4909 * @adapter: board private structure
4911 * If this function returns with an error, then it's possible one or
4912 * more of the rings is populated (while the rest are not). It is the
4913 * callers duty to clean those orphaned rings.
4915 * Return 0 on success, negative on failure
4917 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4919 int i, err = 0;
4921 for (i = 0; i < adapter->num_tx_queues; i++) {
4922 err = ixgbe_setup_tx_resources(adapter, adapter->tx_ring[i]);
4923 if (!err)
4924 continue;
4925 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4926 break;
4929 return err;
4933 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4934 * @adapter: board private structure
4935 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4937 * Returns 0 on success, negative on failure
4939 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4940 struct ixgbe_ring *rx_ring)
4942 struct pci_dev *pdev = adapter->pdev;
4943 int size;
4945 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4946 rx_ring->rx_buffer_info = vmalloc_node(size, adapter->node);
4947 if (!rx_ring->rx_buffer_info)
4948 rx_ring->rx_buffer_info = vmalloc(size);
4949 if (!rx_ring->rx_buffer_info) {
4950 e_err(probe, "vmalloc allocation failed for the Rx "
4951 "descriptor ring\n");
4952 goto alloc_failed;
4954 memset(rx_ring->rx_buffer_info, 0, size);
4956 /* Round up to nearest 4K */
4957 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4958 rx_ring->size = ALIGN(rx_ring->size, 4096);
4960 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
4961 &rx_ring->dma, GFP_KERNEL);
4963 if (!rx_ring->desc) {
4964 e_err(probe, "Memory allocation failed for the Rx "
4965 "descriptor ring\n");
4966 vfree(rx_ring->rx_buffer_info);
4967 goto alloc_failed;
4970 rx_ring->next_to_clean = 0;
4971 rx_ring->next_to_use = 0;
4973 return 0;
4975 alloc_failed:
4976 return -ENOMEM;
4980 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4981 * @adapter: board private structure
4983 * If this function returns with an error, then it's possible one or
4984 * more of the rings is populated (while the rest are not). It is the
4985 * callers duty to clean those orphaned rings.
4987 * Return 0 on success, negative on failure
4990 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4992 int i, err = 0;
4994 for (i = 0; i < adapter->num_rx_queues; i++) {
4995 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
4996 if (!err)
4997 continue;
4998 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4999 break;
5002 return err;
5006 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5007 * @adapter: board private structure
5008 * @tx_ring: Tx descriptor ring for a specific queue
5010 * Free all transmit software resources
5012 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
5013 struct ixgbe_ring *tx_ring)
5015 struct pci_dev *pdev = adapter->pdev;
5017 ixgbe_clean_tx_ring(adapter, tx_ring);
5019 vfree(tx_ring->tx_buffer_info);
5020 tx_ring->tx_buffer_info = NULL;
5022 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
5023 tx_ring->dma);
5025 tx_ring->desc = NULL;
5029 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5030 * @adapter: board private structure
5032 * Free all transmit software resources
5034 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5036 int i;
5038 for (i = 0; i < adapter->num_tx_queues; i++)
5039 if (adapter->tx_ring[i]->desc)
5040 ixgbe_free_tx_resources(adapter, adapter->tx_ring[i]);
5044 * ixgbe_free_rx_resources - Free Rx Resources
5045 * @adapter: board private structure
5046 * @rx_ring: ring to clean the resources from
5048 * Free all receive software resources
5050 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
5051 struct ixgbe_ring *rx_ring)
5053 struct pci_dev *pdev = adapter->pdev;
5055 ixgbe_clean_rx_ring(adapter, rx_ring);
5057 vfree(rx_ring->rx_buffer_info);
5058 rx_ring->rx_buffer_info = NULL;
5060 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
5061 rx_ring->dma);
5063 rx_ring->desc = NULL;
5067 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5068 * @adapter: board private structure
5070 * Free all receive software resources
5072 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5074 int i;
5076 for (i = 0; i < adapter->num_rx_queues; i++)
5077 if (adapter->rx_ring[i]->desc)
5078 ixgbe_free_rx_resources(adapter, adapter->rx_ring[i]);
5082 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5083 * @netdev: network interface device structure
5084 * @new_mtu: new value for maximum frame size
5086 * Returns 0 on success, negative on failure
5088 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5090 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5091 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5093 /* MTU < 68 is an error and causes problems on some kernels */
5094 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5095 return -EINVAL;
5097 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5098 /* must set new MTU before calling down or up */
5099 netdev->mtu = new_mtu;
5101 if (netif_running(netdev))
5102 ixgbe_reinit_locked(adapter);
5104 return 0;
5108 * ixgbe_open - Called when a network interface is made active
5109 * @netdev: network interface device structure
5111 * Returns 0 on success, negative value on failure
5113 * The open entry point is called when a network interface is made
5114 * active by the system (IFF_UP). At this point all resources needed
5115 * for transmit and receive operations are allocated, the interrupt
5116 * handler is registered with the OS, the watchdog timer is started,
5117 * and the stack is notified that the interface is ready.
5119 static int ixgbe_open(struct net_device *netdev)
5121 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5122 int err;
5124 /* disallow open during test */
5125 if (test_bit(__IXGBE_TESTING, &adapter->state))
5126 return -EBUSY;
5128 netif_carrier_off(netdev);
5130 /* allocate transmit descriptors */
5131 err = ixgbe_setup_all_tx_resources(adapter);
5132 if (err)
5133 goto err_setup_tx;
5135 /* allocate receive descriptors */
5136 err = ixgbe_setup_all_rx_resources(adapter);
5137 if (err)
5138 goto err_setup_rx;
5140 ixgbe_configure(adapter);
5142 err = ixgbe_request_irq(adapter);
5143 if (err)
5144 goto err_req_irq;
5146 err = ixgbe_up_complete(adapter);
5147 if (err)
5148 goto err_up;
5150 netif_tx_start_all_queues(netdev);
5152 return 0;
5154 err_up:
5155 ixgbe_release_hw_control(adapter);
5156 ixgbe_free_irq(adapter);
5157 err_req_irq:
5158 err_setup_rx:
5159 ixgbe_free_all_rx_resources(adapter);
5160 err_setup_tx:
5161 ixgbe_free_all_tx_resources(adapter);
5162 ixgbe_reset(adapter);
5164 return err;
5168 * ixgbe_close - Disables a network interface
5169 * @netdev: network interface device structure
5171 * Returns 0, this is not allowed to fail
5173 * The close entry point is called when an interface is de-activated
5174 * by the OS. The hardware is still under the drivers control, but
5175 * needs to be disabled. A global MAC reset is issued to stop the
5176 * hardware, and all transmit and receive resources are freed.
5178 static int ixgbe_close(struct net_device *netdev)
5180 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5182 ixgbe_down(adapter);
5183 ixgbe_free_irq(adapter);
5185 ixgbe_free_all_tx_resources(adapter);
5186 ixgbe_free_all_rx_resources(adapter);
5188 ixgbe_release_hw_control(adapter);
5190 return 0;
5193 #ifdef CONFIG_PM
5194 static int ixgbe_resume(struct pci_dev *pdev)
5196 struct net_device *netdev = pci_get_drvdata(pdev);
5197 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5198 u32 err;
5200 pci_set_power_state(pdev, PCI_D0);
5201 pci_restore_state(pdev);
5203 * pci_restore_state clears dev->state_saved so call
5204 * pci_save_state to restore it.
5206 pci_save_state(pdev);
5208 err = pci_enable_device_mem(pdev);
5209 if (err) {
5210 e_dev_err("Cannot enable PCI device from suspend\n");
5211 return err;
5213 pci_set_master(pdev);
5215 pci_wake_from_d3(pdev, false);
5217 err = ixgbe_init_interrupt_scheme(adapter);
5218 if (err) {
5219 e_dev_err("Cannot initialize interrupts for device\n");
5220 return err;
5223 ixgbe_reset(adapter);
5225 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5227 if (netif_running(netdev)) {
5228 err = ixgbe_open(adapter->netdev);
5229 if (err)
5230 return err;
5233 netif_device_attach(netdev);
5235 return 0;
5237 #endif /* CONFIG_PM */
5239 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5241 struct net_device *netdev = pci_get_drvdata(pdev);
5242 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5243 struct ixgbe_hw *hw = &adapter->hw;
5244 u32 ctrl, fctrl;
5245 u32 wufc = adapter->wol;
5246 #ifdef CONFIG_PM
5247 int retval = 0;
5248 #endif
5250 netif_device_detach(netdev);
5252 if (netif_running(netdev)) {
5253 ixgbe_down(adapter);
5254 ixgbe_free_irq(adapter);
5255 ixgbe_free_all_tx_resources(adapter);
5256 ixgbe_free_all_rx_resources(adapter);
5259 #ifdef CONFIG_PM
5260 retval = pci_save_state(pdev);
5261 if (retval)
5262 return retval;
5264 #endif
5265 if (wufc) {
5266 ixgbe_set_rx_mode(netdev);
5268 /* turn on all-multi mode if wake on multicast is enabled */
5269 if (wufc & IXGBE_WUFC_MC) {
5270 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5271 fctrl |= IXGBE_FCTRL_MPE;
5272 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5275 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5276 ctrl |= IXGBE_CTRL_GIO_DIS;
5277 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5279 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5280 } else {
5281 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5282 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5285 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
5286 pci_wake_from_d3(pdev, true);
5287 else
5288 pci_wake_from_d3(pdev, false);
5290 *enable_wake = !!wufc;
5292 ixgbe_clear_interrupt_scheme(adapter);
5294 ixgbe_release_hw_control(adapter);
5296 pci_disable_device(pdev);
5298 return 0;
5301 #ifdef CONFIG_PM
5302 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5304 int retval;
5305 bool wake;
5307 retval = __ixgbe_shutdown(pdev, &wake);
5308 if (retval)
5309 return retval;
5311 if (wake) {
5312 pci_prepare_to_sleep(pdev);
5313 } else {
5314 pci_wake_from_d3(pdev, false);
5315 pci_set_power_state(pdev, PCI_D3hot);
5318 return 0;
5320 #endif /* CONFIG_PM */
5322 static void ixgbe_shutdown(struct pci_dev *pdev)
5324 bool wake;
5326 __ixgbe_shutdown(pdev, &wake);
5328 if (system_state == SYSTEM_POWER_OFF) {
5329 pci_wake_from_d3(pdev, wake);
5330 pci_set_power_state(pdev, PCI_D3hot);
5335 * ixgbe_update_stats - Update the board statistics counters.
5336 * @adapter: board private structure
5338 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5340 struct net_device *netdev = adapter->netdev;
5341 struct ixgbe_hw *hw = &adapter->hw;
5342 u64 total_mpc = 0;
5343 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5344 u64 non_eop_descs = 0, restart_queue = 0;
5346 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5347 test_bit(__IXGBE_RESETTING, &adapter->state))
5348 return;
5350 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5351 u64 rsc_count = 0;
5352 u64 rsc_flush = 0;
5353 for (i = 0; i < 16; i++)
5354 adapter->hw_rx_no_dma_resources +=
5355 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5356 for (i = 0; i < adapter->num_rx_queues; i++) {
5357 rsc_count += adapter->rx_ring[i]->rsc_count;
5358 rsc_flush += adapter->rx_ring[i]->rsc_flush;
5360 adapter->rsc_total_count = rsc_count;
5361 adapter->rsc_total_flush = rsc_flush;
5364 /* gather some stats to the adapter struct that are per queue */
5365 for (i = 0; i < adapter->num_tx_queues; i++)
5366 restart_queue += adapter->tx_ring[i]->restart_queue;
5367 adapter->restart_queue = restart_queue;
5369 for (i = 0; i < adapter->num_rx_queues; i++)
5370 non_eop_descs += adapter->rx_ring[i]->non_eop_descs;
5371 adapter->non_eop_descs = non_eop_descs;
5373 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5374 for (i = 0; i < 8; i++) {
5375 /* for packet buffers not used, the register should read 0 */
5376 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5377 missed_rx += mpc;
5378 adapter->stats.mpc[i] += mpc;
5379 total_mpc += adapter->stats.mpc[i];
5380 if (hw->mac.type == ixgbe_mac_82598EB)
5381 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5382 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5383 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5384 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5385 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5386 if (hw->mac.type == ixgbe_mac_82599EB) {
5387 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
5388 IXGBE_PXONRXCNT(i));
5389 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
5390 IXGBE_PXOFFRXCNT(i));
5391 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5392 } else {
5393 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
5394 IXGBE_PXONRXC(i));
5395 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
5396 IXGBE_PXOFFRXC(i));
5398 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
5399 IXGBE_PXONTXC(i));
5400 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
5401 IXGBE_PXOFFTXC(i));
5403 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5404 /* work around hardware counting issue */
5405 adapter->stats.gprc -= missed_rx;
5407 /* 82598 hardware only has a 32 bit counter in the high register */
5408 if (hw->mac.type == ixgbe_mac_82599EB) {
5409 u64 tmp;
5410 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5411 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
5412 adapter->stats.gorc += (tmp << 32);
5413 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5414 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
5415 adapter->stats.gotc += (tmp << 32);
5416 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5417 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5418 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5419 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5420 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5421 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5422 #ifdef IXGBE_FCOE
5423 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5424 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5425 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5426 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5427 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5428 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5429 #endif /* IXGBE_FCOE */
5430 } else {
5431 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5432 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5433 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5434 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5435 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5437 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5438 adapter->stats.bprc += bprc;
5439 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5440 if (hw->mac.type == ixgbe_mac_82598EB)
5441 adapter->stats.mprc -= bprc;
5442 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5443 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5444 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5445 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5446 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5447 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5448 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5449 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5450 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5451 adapter->stats.lxontxc += lxon;
5452 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5453 adapter->stats.lxofftxc += lxoff;
5454 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5455 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5456 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5458 * 82598 errata - tx of flow control packets is included in tx counters
5460 xon_off_tot = lxon + lxoff;
5461 adapter->stats.gptc -= xon_off_tot;
5462 adapter->stats.mptc -= xon_off_tot;
5463 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5464 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5465 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5466 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5467 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5468 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5469 adapter->stats.ptc64 -= xon_off_tot;
5470 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5471 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5472 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5473 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5474 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5475 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5477 /* Fill out the OS statistics structure */
5478 netdev->stats.multicast = adapter->stats.mprc;
5480 /* Rx Errors */
5481 netdev->stats.rx_errors = adapter->stats.crcerrs +
5482 adapter->stats.rlec;
5483 netdev->stats.rx_dropped = 0;
5484 netdev->stats.rx_length_errors = adapter->stats.rlec;
5485 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
5486 netdev->stats.rx_missed_errors = total_mpc;
5490 * ixgbe_watchdog - Timer Call-back
5491 * @data: pointer to adapter cast into an unsigned long
5493 static void ixgbe_watchdog(unsigned long data)
5495 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5496 struct ixgbe_hw *hw = &adapter->hw;
5497 u64 eics = 0;
5498 int i;
5501 * Do the watchdog outside of interrupt context due to the lovely
5502 * delays that some of the newer hardware requires
5505 if (test_bit(__IXGBE_DOWN, &adapter->state))
5506 goto watchdog_short_circuit;
5508 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5510 * for legacy and MSI interrupts don't set any bits
5511 * that are enabled for EIAM, because this operation
5512 * would set *both* EIMS and EICS for any bit in EIAM
5514 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5515 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5516 goto watchdog_reschedule;
5519 /* get one bit for every active tx/rx interrupt vector */
5520 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5521 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5522 if (qv->rxr_count || qv->txr_count)
5523 eics |= ((u64)1 << i);
5526 /* Cause software interrupt to ensure rx rings are cleaned */
5527 ixgbe_irq_rearm_queues(adapter, eics);
5529 watchdog_reschedule:
5530 /* Reset the timer */
5531 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5533 watchdog_short_circuit:
5534 schedule_work(&adapter->watchdog_task);
5538 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5539 * @work: pointer to work_struct containing our data
5541 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5543 struct ixgbe_adapter *adapter = container_of(work,
5544 struct ixgbe_adapter,
5545 multispeed_fiber_task);
5546 struct ixgbe_hw *hw = &adapter->hw;
5547 u32 autoneg;
5548 bool negotiation;
5550 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5551 autoneg = hw->phy.autoneg_advertised;
5552 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5553 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5554 hw->mac.autotry_restart = false;
5555 if (hw->mac.ops.setup_link)
5556 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5557 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5558 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5562 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5563 * @work: pointer to work_struct containing our data
5565 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5567 struct ixgbe_adapter *adapter = container_of(work,
5568 struct ixgbe_adapter,
5569 sfp_config_module_task);
5570 struct ixgbe_hw *hw = &adapter->hw;
5571 u32 err;
5573 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5575 /* Time for electrical oscillations to settle down */
5576 msleep(100);
5577 err = hw->phy.ops.identify_sfp(hw);
5579 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5580 e_dev_err("failed to initialize because an unsupported SFP+ "
5581 "module type was detected.\n");
5582 e_dev_err("Reload the driver after installing a supported "
5583 "module.\n");
5584 unregister_netdev(adapter->netdev);
5585 return;
5587 hw->mac.ops.setup_sfp(hw);
5589 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5590 /* This will also work for DA Twinax connections */
5591 schedule_work(&adapter->multispeed_fiber_task);
5592 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5596 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5597 * @work: pointer to work_struct containing our data
5599 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5601 struct ixgbe_adapter *adapter = container_of(work,
5602 struct ixgbe_adapter,
5603 fdir_reinit_task);
5604 struct ixgbe_hw *hw = &adapter->hw;
5605 int i;
5607 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5608 for (i = 0; i < adapter->num_tx_queues; i++)
5609 set_bit(__IXGBE_FDIR_INIT_DONE,
5610 &(adapter->tx_ring[i]->reinit_state));
5611 } else {
5612 e_err(probe, "failed to finish FDIR re-initialization, "
5613 "ignored adding FDIR ATR filters\n");
5615 /* Done FDIR Re-initialization, enable transmits */
5616 netif_tx_start_all_queues(adapter->netdev);
5619 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5622 * ixgbe_watchdog_task - worker thread to bring link up
5623 * @work: pointer to work_struct containing our data
5625 static void ixgbe_watchdog_task(struct work_struct *work)
5627 struct ixgbe_adapter *adapter = container_of(work,
5628 struct ixgbe_adapter,
5629 watchdog_task);
5630 struct net_device *netdev = adapter->netdev;
5631 struct ixgbe_hw *hw = &adapter->hw;
5632 u32 link_speed;
5633 bool link_up;
5634 int i;
5635 struct ixgbe_ring *tx_ring;
5636 int some_tx_pending = 0;
5638 mutex_lock(&ixgbe_watchdog_lock);
5640 link_up = adapter->link_up;
5641 link_speed = adapter->link_speed;
5643 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5644 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5645 if (link_up) {
5646 #ifdef CONFIG_DCB
5647 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5648 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5649 hw->mac.ops.fc_enable(hw, i);
5650 } else {
5651 hw->mac.ops.fc_enable(hw, 0);
5653 #else
5654 hw->mac.ops.fc_enable(hw, 0);
5655 #endif
5658 if (link_up ||
5659 time_after(jiffies, (adapter->link_check_timeout +
5660 IXGBE_TRY_LINK_TIMEOUT))) {
5661 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5662 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5664 adapter->link_up = link_up;
5665 adapter->link_speed = link_speed;
5668 if (link_up) {
5669 if (!netif_carrier_ok(netdev)) {
5670 bool flow_rx, flow_tx;
5672 if (hw->mac.type == ixgbe_mac_82599EB) {
5673 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5674 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5675 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5676 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5677 } else {
5678 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5679 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5680 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5681 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5684 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5685 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5686 "10 Gbps" :
5687 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5688 "1 Gbps" : "unknown speed")),
5689 ((flow_rx && flow_tx) ? "RX/TX" :
5690 (flow_rx ? "RX" :
5691 (flow_tx ? "TX" : "None"))));
5693 netif_carrier_on(netdev);
5694 } else {
5695 /* Force detection of hung controller */
5696 adapter->detect_tx_hung = true;
5698 } else {
5699 adapter->link_up = false;
5700 adapter->link_speed = 0;
5701 if (netif_carrier_ok(netdev)) {
5702 e_info(drv, "NIC Link is Down\n");
5703 netif_carrier_off(netdev);
5707 if (!netif_carrier_ok(netdev)) {
5708 for (i = 0; i < adapter->num_tx_queues; i++) {
5709 tx_ring = adapter->tx_ring[i];
5710 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5711 some_tx_pending = 1;
5712 break;
5716 if (some_tx_pending) {
5717 /* We've lost link, so the controller stops DMA,
5718 * but we've got queued Tx work that's never going
5719 * to get done, so reset controller to flush Tx.
5720 * (Do the reset outside of interrupt context).
5722 schedule_work(&adapter->reset_task);
5726 ixgbe_update_stats(adapter);
5727 mutex_unlock(&ixgbe_watchdog_lock);
5730 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5731 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5732 u32 tx_flags, u8 *hdr_len)
5734 struct ixgbe_adv_tx_context_desc *context_desc;
5735 unsigned int i;
5736 int err;
5737 struct ixgbe_tx_buffer *tx_buffer_info;
5738 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5739 u32 mss_l4len_idx, l4len;
5741 if (skb_is_gso(skb)) {
5742 if (skb_header_cloned(skb)) {
5743 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5744 if (err)
5745 return err;
5747 l4len = tcp_hdrlen(skb);
5748 *hdr_len += l4len;
5750 if (skb->protocol == htons(ETH_P_IP)) {
5751 struct iphdr *iph = ip_hdr(skb);
5752 iph->tot_len = 0;
5753 iph->check = 0;
5754 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5755 iph->daddr, 0,
5756 IPPROTO_TCP,
5758 } else if (skb_is_gso_v6(skb)) {
5759 ipv6_hdr(skb)->payload_len = 0;
5760 tcp_hdr(skb)->check =
5761 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5762 &ipv6_hdr(skb)->daddr,
5763 0, IPPROTO_TCP, 0);
5766 i = tx_ring->next_to_use;
5768 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5769 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5771 /* VLAN MACLEN IPLEN */
5772 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5773 vlan_macip_lens |=
5774 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5775 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5776 IXGBE_ADVTXD_MACLEN_SHIFT);
5777 *hdr_len += skb_network_offset(skb);
5778 vlan_macip_lens |=
5779 (skb_transport_header(skb) - skb_network_header(skb));
5780 *hdr_len +=
5781 (skb_transport_header(skb) - skb_network_header(skb));
5782 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5783 context_desc->seqnum_seed = 0;
5785 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5786 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5787 IXGBE_ADVTXD_DTYP_CTXT);
5789 if (skb->protocol == htons(ETH_P_IP))
5790 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5791 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5792 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5794 /* MSS L4LEN IDX */
5795 mss_l4len_idx =
5796 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5797 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5798 /* use index 1 for TSO */
5799 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5800 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5802 tx_buffer_info->time_stamp = jiffies;
5803 tx_buffer_info->next_to_watch = i;
5805 i++;
5806 if (i == tx_ring->count)
5807 i = 0;
5808 tx_ring->next_to_use = i;
5810 return true;
5812 return false;
5815 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5816 struct ixgbe_ring *tx_ring,
5817 struct sk_buff *skb, u32 tx_flags)
5819 struct ixgbe_adv_tx_context_desc *context_desc;
5820 unsigned int i;
5821 struct ixgbe_tx_buffer *tx_buffer_info;
5822 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5824 if (skb->ip_summed == CHECKSUM_PARTIAL ||
5825 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5826 i = tx_ring->next_to_use;
5827 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5828 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5830 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5831 vlan_macip_lens |=
5832 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5833 vlan_macip_lens |= (skb_network_offset(skb) <<
5834 IXGBE_ADVTXD_MACLEN_SHIFT);
5835 if (skb->ip_summed == CHECKSUM_PARTIAL)
5836 vlan_macip_lens |= (skb_transport_header(skb) -
5837 skb_network_header(skb));
5839 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5840 context_desc->seqnum_seed = 0;
5842 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5843 IXGBE_ADVTXD_DTYP_CTXT);
5845 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5846 __be16 protocol;
5848 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
5849 const struct vlan_ethhdr *vhdr =
5850 (const struct vlan_ethhdr *)skb->data;
5852 protocol = vhdr->h_vlan_encapsulated_proto;
5853 } else {
5854 protocol = skb->protocol;
5857 switch (protocol) {
5858 case cpu_to_be16(ETH_P_IP):
5859 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5860 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5861 type_tucmd_mlhl |=
5862 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5863 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
5864 type_tucmd_mlhl |=
5865 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5866 break;
5867 case cpu_to_be16(ETH_P_IPV6):
5868 /* XXX what about other V6 headers?? */
5869 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5870 type_tucmd_mlhl |=
5871 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5872 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
5873 type_tucmd_mlhl |=
5874 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5875 break;
5876 default:
5877 if (unlikely(net_ratelimit())) {
5878 e_warn(probe, "partial checksum "
5879 "but proto=%x!\n",
5880 skb->protocol);
5882 break;
5886 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5887 /* use index zero for tx checksum offload */
5888 context_desc->mss_l4len_idx = 0;
5890 tx_buffer_info->time_stamp = jiffies;
5891 tx_buffer_info->next_to_watch = i;
5893 i++;
5894 if (i == tx_ring->count)
5895 i = 0;
5896 tx_ring->next_to_use = i;
5898 return true;
5901 return false;
5904 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5905 struct ixgbe_ring *tx_ring,
5906 struct sk_buff *skb, u32 tx_flags,
5907 unsigned int first)
5909 struct pci_dev *pdev = adapter->pdev;
5910 struct ixgbe_tx_buffer *tx_buffer_info;
5911 unsigned int len;
5912 unsigned int total = skb->len;
5913 unsigned int offset = 0, size, count = 0, i;
5914 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5915 unsigned int f;
5917 i = tx_ring->next_to_use;
5919 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5920 /* excluding fcoe_crc_eof for FCoE */
5921 total -= sizeof(struct fcoe_crc_eof);
5923 len = min(skb_headlen(skb), total);
5924 while (len) {
5925 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5926 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5928 tx_buffer_info->length = size;
5929 tx_buffer_info->mapped_as_page = false;
5930 tx_buffer_info->dma = dma_map_single(&pdev->dev,
5931 skb->data + offset,
5932 size, DMA_TO_DEVICE);
5933 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
5934 goto dma_error;
5935 tx_buffer_info->time_stamp = jiffies;
5936 tx_buffer_info->next_to_watch = i;
5938 len -= size;
5939 total -= size;
5940 offset += size;
5941 count++;
5943 if (len) {
5944 i++;
5945 if (i == tx_ring->count)
5946 i = 0;
5950 for (f = 0; f < nr_frags; f++) {
5951 struct skb_frag_struct *frag;
5953 frag = &skb_shinfo(skb)->frags[f];
5954 len = min((unsigned int)frag->size, total);
5955 offset = frag->page_offset;
5957 while (len) {
5958 i++;
5959 if (i == tx_ring->count)
5960 i = 0;
5962 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5963 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5965 tx_buffer_info->length = size;
5966 tx_buffer_info->dma = dma_map_page(&adapter->pdev->dev,
5967 frag->page,
5968 offset, size,
5969 DMA_TO_DEVICE);
5970 tx_buffer_info->mapped_as_page = true;
5971 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
5972 goto dma_error;
5973 tx_buffer_info->time_stamp = jiffies;
5974 tx_buffer_info->next_to_watch = i;
5976 len -= size;
5977 total -= size;
5978 offset += size;
5979 count++;
5981 if (total == 0)
5982 break;
5985 tx_ring->tx_buffer_info[i].skb = skb;
5986 tx_ring->tx_buffer_info[first].next_to_watch = i;
5988 return count;
5990 dma_error:
5991 e_dev_err("TX DMA map failed\n");
5993 /* clear timestamp and dma mappings for failed tx_buffer_info map */
5994 tx_buffer_info->dma = 0;
5995 tx_buffer_info->time_stamp = 0;
5996 tx_buffer_info->next_to_watch = 0;
5997 if (count)
5998 count--;
6000 /* clear timestamp and dma mappings for remaining portion of packet */
6001 while (count--) {
6002 if (i==0)
6003 i += tx_ring->count;
6004 i--;
6005 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6006 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
6009 return 0;
6012 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
6013 struct ixgbe_ring *tx_ring,
6014 int tx_flags, int count, u32 paylen, u8 hdr_len)
6016 union ixgbe_adv_tx_desc *tx_desc = NULL;
6017 struct ixgbe_tx_buffer *tx_buffer_info;
6018 u32 olinfo_status = 0, cmd_type_len = 0;
6019 unsigned int i;
6020 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6022 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6024 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6026 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6027 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6029 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6030 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6032 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6033 IXGBE_ADVTXD_POPTS_SHIFT;
6035 /* use index 1 context for tso */
6036 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6037 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6038 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6039 IXGBE_ADVTXD_POPTS_SHIFT;
6041 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6042 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6043 IXGBE_ADVTXD_POPTS_SHIFT;
6045 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6046 olinfo_status |= IXGBE_ADVTXD_CC;
6047 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6048 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6049 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6052 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6054 i = tx_ring->next_to_use;
6055 while (count--) {
6056 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6057 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
6058 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6059 tx_desc->read.cmd_type_len =
6060 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6061 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6062 i++;
6063 if (i == tx_ring->count)
6064 i = 0;
6067 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6070 * Force memory writes to complete before letting h/w
6071 * know there are new descriptors to fetch. (Only
6072 * applicable for weak-ordered memory model archs,
6073 * such as IA-64).
6075 wmb();
6077 tx_ring->next_to_use = i;
6078 writel(i, adapter->hw.hw_addr + tx_ring->tail);
6081 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6082 int queue, u32 tx_flags)
6084 struct ixgbe_atr_input atr_input;
6085 struct tcphdr *th;
6086 struct iphdr *iph = ip_hdr(skb);
6087 struct ethhdr *eth = (struct ethhdr *)skb->data;
6088 u16 vlan_id, src_port, dst_port, flex_bytes;
6089 u32 src_ipv4_addr, dst_ipv4_addr;
6090 u8 l4type = 0;
6092 /* Right now, we support IPv4 only */
6093 if (skb->protocol != htons(ETH_P_IP))
6094 return;
6095 /* check if we're UDP or TCP */
6096 if (iph->protocol == IPPROTO_TCP) {
6097 th = tcp_hdr(skb);
6098 src_port = th->source;
6099 dst_port = th->dest;
6100 l4type |= IXGBE_ATR_L4TYPE_TCP;
6101 /* l4type IPv4 type is 0, no need to assign */
6102 } else {
6103 /* Unsupported L4 header, just bail here */
6104 return;
6107 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6109 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6110 IXGBE_TX_FLAGS_VLAN_SHIFT;
6111 src_ipv4_addr = iph->saddr;
6112 dst_ipv4_addr = iph->daddr;
6113 flex_bytes = eth->h_proto;
6115 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6116 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
6117 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
6118 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
6119 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
6120 /* src and dst are inverted, think how the receiver sees them */
6121 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
6122 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
6124 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6125 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6128 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
6129 struct ixgbe_ring *tx_ring, int size)
6131 netif_stop_subqueue(netdev, tx_ring->queue_index);
6132 /* Herbert's original patch had:
6133 * smp_mb__after_netif_stop_queue();
6134 * but since that doesn't exist yet, just open code it. */
6135 smp_mb();
6137 /* We need to check again in a case another CPU has just
6138 * made room available. */
6139 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6140 return -EBUSY;
6142 /* A reprieve! - use start_queue because it doesn't call schedule */
6143 netif_start_subqueue(netdev, tx_ring->queue_index);
6144 ++tx_ring->restart_queue;
6145 return 0;
6148 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
6149 struct ixgbe_ring *tx_ring, int size)
6151 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6152 return 0;
6153 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
6156 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6158 struct ixgbe_adapter *adapter = netdev_priv(dev);
6159 int txq = smp_processor_id();
6161 #ifdef IXGBE_FCOE
6162 if ((skb->protocol == htons(ETH_P_FCOE)) ||
6163 (skb->protocol == htons(ETH_P_FIP))) {
6164 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6165 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6166 txq += adapter->ring_feature[RING_F_FCOE].mask;
6167 return txq;
6168 #ifdef CONFIG_IXGBE_DCB
6169 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6170 txq = adapter->fcoe.up;
6171 return txq;
6172 #endif
6175 #endif
6177 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6178 while (unlikely(txq >= dev->real_num_tx_queues))
6179 txq -= dev->real_num_tx_queues;
6180 return txq;
6183 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6184 if (skb->priority == TC_PRIO_CONTROL)
6185 txq = adapter->ring_feature[RING_F_DCB].indices-1;
6186 else
6187 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6188 >> 13;
6189 return txq;
6192 return skb_tx_hash(dev, skb);
6195 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6196 struct net_device *netdev)
6198 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6199 struct ixgbe_ring *tx_ring;
6200 struct netdev_queue *txq;
6201 unsigned int first;
6202 unsigned int tx_flags = 0;
6203 u8 hdr_len = 0;
6204 int tso;
6205 int count = 0;
6206 unsigned int f;
6208 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
6209 tx_flags |= vlan_tx_tag_get(skb);
6210 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6211 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6212 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6214 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6215 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6216 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6217 skb->priority != TC_PRIO_CONTROL) {
6218 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6219 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6220 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6223 tx_ring = adapter->tx_ring[skb->queue_mapping];
6225 #ifdef IXGBE_FCOE
6226 /* for FCoE with DCB, we force the priority to what
6227 * was specified by the switch */
6228 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6229 (skb->protocol == htons(ETH_P_FCOE) ||
6230 skb->protocol == htons(ETH_P_FIP))) {
6231 #ifdef CONFIG_IXGBE_DCB
6232 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6233 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6234 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6235 tx_flags |= ((adapter->fcoe.up << 13)
6236 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6238 #endif
6239 /* flag for FCoE offloads */
6240 if (skb->protocol == htons(ETH_P_FCOE))
6241 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6243 #endif
6245 /* four things can cause us to need a context descriptor */
6246 if (skb_is_gso(skb) ||
6247 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6248 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6249 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6250 count++;
6252 count += TXD_USE_COUNT(skb_headlen(skb));
6253 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6254 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6256 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
6257 adapter->tx_busy++;
6258 return NETDEV_TX_BUSY;
6261 first = tx_ring->next_to_use;
6262 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6263 #ifdef IXGBE_FCOE
6264 /* setup tx offload for FCoE */
6265 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6266 if (tso < 0) {
6267 dev_kfree_skb_any(skb);
6268 return NETDEV_TX_OK;
6270 if (tso)
6271 tx_flags |= IXGBE_TX_FLAGS_FSO;
6272 #endif /* IXGBE_FCOE */
6273 } else {
6274 if (skb->protocol == htons(ETH_P_IP))
6275 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6276 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6277 if (tso < 0) {
6278 dev_kfree_skb_any(skb);
6279 return NETDEV_TX_OK;
6282 if (tso)
6283 tx_flags |= IXGBE_TX_FLAGS_TSO;
6284 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
6285 (skb->ip_summed == CHECKSUM_PARTIAL))
6286 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6289 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
6290 if (count) {
6291 /* add the ATR filter if ATR is on */
6292 if (tx_ring->atr_sample_rate) {
6293 ++tx_ring->atr_count;
6294 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6295 test_bit(__IXGBE_FDIR_INIT_DONE,
6296 &tx_ring->reinit_state)) {
6297 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6298 tx_flags);
6299 tx_ring->atr_count = 0;
6302 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6303 txq->tx_bytes += skb->len;
6304 txq->tx_packets++;
6305 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
6306 hdr_len);
6307 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
6309 } else {
6310 dev_kfree_skb_any(skb);
6311 tx_ring->tx_buffer_info[first].time_stamp = 0;
6312 tx_ring->next_to_use = first;
6315 return NETDEV_TX_OK;
6319 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6320 * @netdev: network interface device structure
6321 * @p: pointer to an address structure
6323 * Returns 0 on success, negative on failure
6325 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6327 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6328 struct ixgbe_hw *hw = &adapter->hw;
6329 struct sockaddr *addr = p;
6331 if (!is_valid_ether_addr(addr->sa_data))
6332 return -EADDRNOTAVAIL;
6334 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6335 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6337 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6338 IXGBE_RAH_AV);
6340 return 0;
6343 static int
6344 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6346 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6347 struct ixgbe_hw *hw = &adapter->hw;
6348 u16 value;
6349 int rc;
6351 if (prtad != hw->phy.mdio.prtad)
6352 return -EINVAL;
6353 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6354 if (!rc)
6355 rc = value;
6356 return rc;
6359 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6360 u16 addr, u16 value)
6362 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6363 struct ixgbe_hw *hw = &adapter->hw;
6365 if (prtad != hw->phy.mdio.prtad)
6366 return -EINVAL;
6367 return hw->phy.ops.write_reg(hw, addr, devad, value);
6370 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6372 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6374 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6378 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6379 * netdev->dev_addrs
6380 * @netdev: network interface device structure
6382 * Returns non-zero on failure
6384 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6386 int err = 0;
6387 struct ixgbe_adapter *adapter = netdev_priv(dev);
6388 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6390 if (is_valid_ether_addr(mac->san_addr)) {
6391 rtnl_lock();
6392 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6393 rtnl_unlock();
6395 return err;
6399 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6400 * netdev->dev_addrs
6401 * @netdev: network interface device structure
6403 * Returns non-zero on failure
6405 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6407 int err = 0;
6408 struct ixgbe_adapter *adapter = netdev_priv(dev);
6409 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6411 if (is_valid_ether_addr(mac->san_addr)) {
6412 rtnl_lock();
6413 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6414 rtnl_unlock();
6416 return err;
6419 #ifdef CONFIG_NET_POLL_CONTROLLER
6421 * Polling 'interrupt' - used by things like netconsole to send skbs
6422 * without having to re-enable interrupts. It's not called while
6423 * the interrupt routine is executing.
6425 static void ixgbe_netpoll(struct net_device *netdev)
6427 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6428 int i;
6430 /* if interface is down do nothing */
6431 if (test_bit(__IXGBE_DOWN, &adapter->state))
6432 return;
6434 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6435 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6436 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6437 for (i = 0; i < num_q_vectors; i++) {
6438 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6439 ixgbe_msix_clean_many(0, q_vector);
6441 } else {
6442 ixgbe_intr(adapter->pdev->irq, netdev);
6444 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6446 #endif
6448 static const struct net_device_ops ixgbe_netdev_ops = {
6449 .ndo_open = ixgbe_open,
6450 .ndo_stop = ixgbe_close,
6451 .ndo_start_xmit = ixgbe_xmit_frame,
6452 .ndo_select_queue = ixgbe_select_queue,
6453 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6454 .ndo_set_multicast_list = ixgbe_set_rx_mode,
6455 .ndo_validate_addr = eth_validate_addr,
6456 .ndo_set_mac_address = ixgbe_set_mac,
6457 .ndo_change_mtu = ixgbe_change_mtu,
6458 .ndo_tx_timeout = ixgbe_tx_timeout,
6459 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
6460 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6461 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6462 .ndo_do_ioctl = ixgbe_ioctl,
6463 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6464 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6465 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6466 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6467 #ifdef CONFIG_NET_POLL_CONTROLLER
6468 .ndo_poll_controller = ixgbe_netpoll,
6469 #endif
6470 #ifdef IXGBE_FCOE
6471 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6472 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6473 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6474 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6475 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6476 #endif /* IXGBE_FCOE */
6479 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6480 const struct ixgbe_info *ii)
6482 #ifdef CONFIG_PCI_IOV
6483 struct ixgbe_hw *hw = &adapter->hw;
6484 int err;
6486 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6487 return;
6489 /* The 82599 supports up to 64 VFs per physical function
6490 * but this implementation limits allocation to 63 so that
6491 * basic networking resources are still available to the
6492 * physical function
6494 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6495 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6496 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6497 if (err) {
6498 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6499 goto err_novfs;
6501 /* If call to enable VFs succeeded then allocate memory
6502 * for per VF control structures.
6504 adapter->vfinfo =
6505 kcalloc(adapter->num_vfs,
6506 sizeof(struct vf_data_storage), GFP_KERNEL);
6507 if (adapter->vfinfo) {
6508 /* Now that we're sure SR-IOV is enabled
6509 * and memory allocated set up the mailbox parameters
6511 ixgbe_init_mbx_params_pf(hw);
6512 memcpy(&hw->mbx.ops, ii->mbx_ops,
6513 sizeof(hw->mbx.ops));
6515 /* Disable RSC when in SR-IOV mode */
6516 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6517 IXGBE_FLAG2_RSC_ENABLED);
6518 return;
6521 /* Oh oh */
6522 e_err(probe, "Unable to allocate memory for VF Data Storage - "
6523 "SRIOV disabled\n");
6524 pci_disable_sriov(adapter->pdev);
6526 err_novfs:
6527 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6528 adapter->num_vfs = 0;
6529 #endif /* CONFIG_PCI_IOV */
6533 * ixgbe_probe - Device Initialization Routine
6534 * @pdev: PCI device information struct
6535 * @ent: entry in ixgbe_pci_tbl
6537 * Returns 0 on success, negative on failure
6539 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6540 * The OS initialization, configuring of the adapter private structure,
6541 * and a hardware reset occur.
6543 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6544 const struct pci_device_id *ent)
6546 struct net_device *netdev;
6547 struct ixgbe_adapter *adapter = NULL;
6548 struct ixgbe_hw *hw;
6549 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6550 static int cards_found;
6551 int i, err, pci_using_dac;
6552 unsigned int indices = num_possible_cpus();
6553 #ifdef IXGBE_FCOE
6554 u16 device_caps;
6555 #endif
6556 u32 part_num, eec;
6558 /* Catch broken hardware that put the wrong VF device ID in
6559 * the PCIe SR-IOV capability.
6561 if (pdev->is_virtfn) {
6562 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6563 pci_name(pdev), pdev->vendor, pdev->device);
6564 return -EINVAL;
6567 err = pci_enable_device_mem(pdev);
6568 if (err)
6569 return err;
6571 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6572 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6573 pci_using_dac = 1;
6574 } else {
6575 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6576 if (err) {
6577 err = dma_set_coherent_mask(&pdev->dev,
6578 DMA_BIT_MASK(32));
6579 if (err) {
6580 dev_err(&pdev->dev,
6581 "No usable DMA configuration, aborting\n");
6582 goto err_dma;
6585 pci_using_dac = 0;
6588 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6589 IORESOURCE_MEM), ixgbe_driver_name);
6590 if (err) {
6591 dev_err(&pdev->dev,
6592 "pci_request_selected_regions failed 0x%x\n", err);
6593 goto err_pci_reg;
6596 pci_enable_pcie_error_reporting(pdev);
6598 pci_set_master(pdev);
6599 pci_save_state(pdev);
6601 if (ii->mac == ixgbe_mac_82598EB)
6602 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6603 else
6604 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6606 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6607 #ifdef IXGBE_FCOE
6608 indices += min_t(unsigned int, num_possible_cpus(),
6609 IXGBE_MAX_FCOE_INDICES);
6610 #endif
6611 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6612 if (!netdev) {
6613 err = -ENOMEM;
6614 goto err_alloc_etherdev;
6617 SET_NETDEV_DEV(netdev, &pdev->dev);
6619 pci_set_drvdata(pdev, netdev);
6620 adapter = netdev_priv(netdev);
6622 adapter->netdev = netdev;
6623 adapter->pdev = pdev;
6624 hw = &adapter->hw;
6625 hw->back = adapter;
6626 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6628 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6629 pci_resource_len(pdev, 0));
6630 if (!hw->hw_addr) {
6631 err = -EIO;
6632 goto err_ioremap;
6635 for (i = 1; i <= 5; i++) {
6636 if (pci_resource_len(pdev, i) == 0)
6637 continue;
6640 netdev->netdev_ops = &ixgbe_netdev_ops;
6641 ixgbe_set_ethtool_ops(netdev);
6642 netdev->watchdog_timeo = 5 * HZ;
6643 strcpy(netdev->name, pci_name(pdev));
6645 adapter->bd_number = cards_found;
6647 /* Setup hw api */
6648 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6649 hw->mac.type = ii->mac;
6651 /* EEPROM */
6652 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6653 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6654 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6655 if (!(eec & (1 << 8)))
6656 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6658 /* PHY */
6659 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6660 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6661 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6662 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6663 hw->phy.mdio.mmds = 0;
6664 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6665 hw->phy.mdio.dev = netdev;
6666 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6667 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6669 /* set up this timer and work struct before calling get_invariants
6670 * which might start the timer
6672 init_timer(&adapter->sfp_timer);
6673 adapter->sfp_timer.function = &ixgbe_sfp_timer;
6674 adapter->sfp_timer.data = (unsigned long) adapter;
6676 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6678 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6679 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6681 /* a new SFP+ module arrival, called from GPI SDP2 context */
6682 INIT_WORK(&adapter->sfp_config_module_task,
6683 ixgbe_sfp_config_module_task);
6685 ii->get_invariants(hw);
6687 /* setup the private structure */
6688 err = ixgbe_sw_init(adapter);
6689 if (err)
6690 goto err_sw_init;
6692 /* Make it possible the adapter to be woken up via WOL */
6693 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6694 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6697 * If there is a fan on this device and it has failed log the
6698 * failure.
6700 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6701 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6702 if (esdp & IXGBE_ESDP_SDP1)
6703 e_crit(probe, "Fan has stopped, replace the adapter\n");
6706 /* reset_hw fills in the perm_addr as well */
6707 hw->phy.reset_if_overtemp = true;
6708 err = hw->mac.ops.reset_hw(hw);
6709 hw->phy.reset_if_overtemp = false;
6710 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6711 hw->mac.type == ixgbe_mac_82598EB) {
6713 * Start a kernel thread to watch for a module to arrive.
6714 * Only do this for 82598, since 82599 will generate
6715 * interrupts on module arrival.
6717 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6718 mod_timer(&adapter->sfp_timer,
6719 round_jiffies(jiffies + (2 * HZ)));
6720 err = 0;
6721 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6722 e_dev_err("failed to initialize because an unsupported SFP+ "
6723 "module type was detected.\n");
6724 e_dev_err("Reload the driver after installing a supported "
6725 "module.\n");
6726 goto err_sw_init;
6727 } else if (err) {
6728 e_dev_err("HW Init failed: %d\n", err);
6729 goto err_sw_init;
6732 ixgbe_probe_vf(adapter, ii);
6734 netdev->features = NETIF_F_SG |
6735 NETIF_F_IP_CSUM |
6736 NETIF_F_HW_VLAN_TX |
6737 NETIF_F_HW_VLAN_RX |
6738 NETIF_F_HW_VLAN_FILTER;
6740 netdev->features |= NETIF_F_IPV6_CSUM;
6741 netdev->features |= NETIF_F_TSO;
6742 netdev->features |= NETIF_F_TSO6;
6743 netdev->features |= NETIF_F_GRO;
6745 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6746 netdev->features |= NETIF_F_SCTP_CSUM;
6748 netdev->vlan_features |= NETIF_F_TSO;
6749 netdev->vlan_features |= NETIF_F_TSO6;
6750 netdev->vlan_features |= NETIF_F_IP_CSUM;
6751 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6752 netdev->vlan_features |= NETIF_F_SG;
6754 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6755 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6756 IXGBE_FLAG_DCB_ENABLED);
6757 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6758 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6760 #ifdef CONFIG_IXGBE_DCB
6761 netdev->dcbnl_ops = &dcbnl_ops;
6762 #endif
6764 #ifdef IXGBE_FCOE
6765 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6766 if (hw->mac.ops.get_device_caps) {
6767 hw->mac.ops.get_device_caps(hw, &device_caps);
6768 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6769 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6772 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6773 netdev->vlan_features |= NETIF_F_FCOE_CRC;
6774 netdev->vlan_features |= NETIF_F_FSO;
6775 netdev->vlan_features |= NETIF_F_FCOE_MTU;
6777 #endif /* IXGBE_FCOE */
6778 if (pci_using_dac)
6779 netdev->features |= NETIF_F_HIGHDMA;
6781 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6782 netdev->features |= NETIF_F_LRO;
6784 /* make sure the EEPROM is good */
6785 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6786 e_dev_err("The EEPROM Checksum Is Not Valid\n");
6787 err = -EIO;
6788 goto err_eeprom;
6791 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6792 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6794 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6795 e_dev_err("invalid MAC address\n");
6796 err = -EIO;
6797 goto err_eeprom;
6800 /* power down the optics */
6801 if (hw->phy.multispeed_fiber)
6802 hw->mac.ops.disable_tx_laser(hw);
6804 init_timer(&adapter->watchdog_timer);
6805 adapter->watchdog_timer.function = &ixgbe_watchdog;
6806 adapter->watchdog_timer.data = (unsigned long)adapter;
6808 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6809 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6811 err = ixgbe_init_interrupt_scheme(adapter);
6812 if (err)
6813 goto err_sw_init;
6815 switch (pdev->device) {
6816 case IXGBE_DEV_ID_82599_KX4:
6817 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
6818 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
6819 break;
6820 default:
6821 adapter->wol = 0;
6822 break;
6824 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6826 /* pick up the PCI bus settings for reporting later */
6827 hw->mac.ops.get_bus_info(hw);
6829 /* print bus type/speed/width info */
6830 e_dev_info("(PCI Express:%s:%s) %pM\n",
6831 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
6832 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
6833 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
6834 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
6835 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
6836 "Unknown"),
6837 netdev->dev_addr);
6838 ixgbe_read_pba_num_generic(hw, &part_num);
6839 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
6840 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
6841 "PBA No: %06x-%03x\n",
6842 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
6843 (part_num >> 8), (part_num & 0xff));
6844 else
6845 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
6846 hw->mac.type, hw->phy.type,
6847 (part_num >> 8), (part_num & 0xff));
6849 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
6850 e_dev_warn("PCI-Express bandwidth available for this card is "
6851 "not sufficient for optimal performance.\n");
6852 e_dev_warn("For optimal performance a x8 PCI-Express slot "
6853 "is required.\n");
6856 /* save off EEPROM version number */
6857 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
6859 /* reset the hardware with the new settings */
6860 err = hw->mac.ops.start_hw(hw);
6862 if (err == IXGBE_ERR_EEPROM_VERSION) {
6863 /* We are running on a pre-production device, log a warning */
6864 e_dev_warn("This device is a pre-production adapter/LOM. "
6865 "Please be aware there may be issues associated "
6866 "with your hardware. If you are experiencing "
6867 "problems please contact your Intel or hardware "
6868 "representative who provided you with this "
6869 "hardware.\n");
6871 strcpy(netdev->name, "eth%d");
6872 err = register_netdev(netdev);
6873 if (err)
6874 goto err_register;
6876 /* carrier off reporting is important to ethtool even BEFORE open */
6877 netif_carrier_off(netdev);
6879 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6880 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6881 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
6883 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
6884 INIT_WORK(&adapter->check_overtemp_task, ixgbe_check_overtemp_task);
6885 #ifdef CONFIG_IXGBE_DCA
6886 if (dca_add_requester(&pdev->dev) == 0) {
6887 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
6888 ixgbe_setup_dca(adapter);
6890 #endif
6891 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6892 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
6893 for (i = 0; i < adapter->num_vfs; i++)
6894 ixgbe_vf_configuration(pdev, (i | 0x10000000));
6897 /* add san mac addr to netdev */
6898 ixgbe_add_sanmac_netdev(netdev);
6900 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
6901 cards_found++;
6902 return 0;
6904 err_register:
6905 ixgbe_release_hw_control(adapter);
6906 ixgbe_clear_interrupt_scheme(adapter);
6907 err_sw_init:
6908 err_eeprom:
6909 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6910 ixgbe_disable_sriov(adapter);
6911 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6912 del_timer_sync(&adapter->sfp_timer);
6913 cancel_work_sync(&adapter->sfp_task);
6914 cancel_work_sync(&adapter->multispeed_fiber_task);
6915 cancel_work_sync(&adapter->sfp_config_module_task);
6916 iounmap(hw->hw_addr);
6917 err_ioremap:
6918 free_netdev(netdev);
6919 err_alloc_etherdev:
6920 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6921 IORESOURCE_MEM));
6922 err_pci_reg:
6923 err_dma:
6924 pci_disable_device(pdev);
6925 return err;
6929 * ixgbe_remove - Device Removal Routine
6930 * @pdev: PCI device information struct
6932 * ixgbe_remove is called by the PCI subsystem to alert the driver
6933 * that it should release a PCI device. The could be caused by a
6934 * Hot-Plug event, or because the driver is going to be removed from
6935 * memory.
6937 static void __devexit ixgbe_remove(struct pci_dev *pdev)
6939 struct net_device *netdev = pci_get_drvdata(pdev);
6940 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6942 set_bit(__IXGBE_DOWN, &adapter->state);
6943 /* clear the module not found bit to make sure the worker won't
6944 * reschedule
6946 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6947 del_timer_sync(&adapter->watchdog_timer);
6949 del_timer_sync(&adapter->sfp_timer);
6950 cancel_work_sync(&adapter->watchdog_task);
6951 cancel_work_sync(&adapter->sfp_task);
6952 cancel_work_sync(&adapter->multispeed_fiber_task);
6953 cancel_work_sync(&adapter->sfp_config_module_task);
6954 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6955 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6956 cancel_work_sync(&adapter->fdir_reinit_task);
6957 flush_scheduled_work();
6959 #ifdef CONFIG_IXGBE_DCA
6960 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
6961 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
6962 dca_remove_requester(&pdev->dev);
6963 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
6966 #endif
6967 #ifdef IXGBE_FCOE
6968 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
6969 ixgbe_cleanup_fcoe(adapter);
6971 #endif /* IXGBE_FCOE */
6973 /* remove the added san mac */
6974 ixgbe_del_sanmac_netdev(netdev);
6976 if (netdev->reg_state == NETREG_REGISTERED)
6977 unregister_netdev(netdev);
6979 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6980 ixgbe_disable_sriov(adapter);
6982 ixgbe_clear_interrupt_scheme(adapter);
6984 ixgbe_release_hw_control(adapter);
6986 iounmap(adapter->hw.hw_addr);
6987 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6988 IORESOURCE_MEM));
6990 e_dev_info("complete\n");
6992 free_netdev(netdev);
6994 pci_disable_pcie_error_reporting(pdev);
6996 pci_disable_device(pdev);
7000 * ixgbe_io_error_detected - called when PCI error is detected
7001 * @pdev: Pointer to PCI device
7002 * @state: The current pci connection state
7004 * This function is called after a PCI bus error affecting
7005 * this device has been detected.
7007 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7008 pci_channel_state_t state)
7010 struct net_device *netdev = pci_get_drvdata(pdev);
7011 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7013 netif_device_detach(netdev);
7015 if (state == pci_channel_io_perm_failure)
7016 return PCI_ERS_RESULT_DISCONNECT;
7018 if (netif_running(netdev))
7019 ixgbe_down(adapter);
7020 pci_disable_device(pdev);
7022 /* Request a slot reset. */
7023 return PCI_ERS_RESULT_NEED_RESET;
7027 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7028 * @pdev: Pointer to PCI device
7030 * Restart the card from scratch, as if from a cold-boot.
7032 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7034 struct net_device *netdev = pci_get_drvdata(pdev);
7035 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7036 pci_ers_result_t result;
7037 int err;
7039 if (pci_enable_device_mem(pdev)) {
7040 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7041 result = PCI_ERS_RESULT_DISCONNECT;
7042 } else {
7043 pci_set_master(pdev);
7044 pci_restore_state(pdev);
7045 pci_save_state(pdev);
7047 pci_wake_from_d3(pdev, false);
7049 ixgbe_reset(adapter);
7050 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7051 result = PCI_ERS_RESULT_RECOVERED;
7054 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7055 if (err) {
7056 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7057 "failed 0x%0x\n", err);
7058 /* non-fatal, continue */
7061 return result;
7065 * ixgbe_io_resume - called when traffic can start flowing again.
7066 * @pdev: Pointer to PCI device
7068 * This callback is called when the error recovery driver tells us that
7069 * its OK to resume normal operation.
7071 static void ixgbe_io_resume(struct pci_dev *pdev)
7073 struct net_device *netdev = pci_get_drvdata(pdev);
7074 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7076 if (netif_running(netdev)) {
7077 if (ixgbe_up(adapter)) {
7078 e_info(probe, "ixgbe_up failed after reset\n");
7079 return;
7083 netif_device_attach(netdev);
7086 static struct pci_error_handlers ixgbe_err_handler = {
7087 .error_detected = ixgbe_io_error_detected,
7088 .slot_reset = ixgbe_io_slot_reset,
7089 .resume = ixgbe_io_resume,
7092 static struct pci_driver ixgbe_driver = {
7093 .name = ixgbe_driver_name,
7094 .id_table = ixgbe_pci_tbl,
7095 .probe = ixgbe_probe,
7096 .remove = __devexit_p(ixgbe_remove),
7097 #ifdef CONFIG_PM
7098 .suspend = ixgbe_suspend,
7099 .resume = ixgbe_resume,
7100 #endif
7101 .shutdown = ixgbe_shutdown,
7102 .err_handler = &ixgbe_err_handler
7106 * ixgbe_init_module - Driver Registration Routine
7108 * ixgbe_init_module is the first routine called when the driver is
7109 * loaded. All it does is register with the PCI subsystem.
7111 static int __init ixgbe_init_module(void)
7113 int ret;
7114 pr_info("%s - version %s\n", ixgbe_driver_string,
7115 ixgbe_driver_version);
7116 pr_info("%s\n", ixgbe_copyright);
7118 #ifdef CONFIG_IXGBE_DCA
7119 dca_register_notify(&dca_notifier);
7120 #endif
7122 ret = pci_register_driver(&ixgbe_driver);
7123 return ret;
7126 module_init(ixgbe_init_module);
7129 * ixgbe_exit_module - Driver Exit Cleanup Routine
7131 * ixgbe_exit_module is called just before the driver is removed
7132 * from memory.
7134 static void __exit ixgbe_exit_module(void)
7136 #ifdef CONFIG_IXGBE_DCA
7137 dca_unregister_notify(&dca_notifier);
7138 #endif
7139 pci_unregister_driver(&ixgbe_driver);
7142 #ifdef CONFIG_IXGBE_DCA
7143 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7144 void *p)
7146 int ret_val;
7148 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7149 __ixgbe_notify_dca);
7151 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7154 #endif /* CONFIG_IXGBE_DCA */
7157 * ixgbe_get_hw_dev return device
7158 * used by hardware layer to print debugging information
7160 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7162 struct ixgbe_adapter *adapter = hw->back;
7163 return adapter->netdev;
7166 module_exit(ixgbe_exit_module);
7168 /* ixgbe_main.c */