1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2012 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
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".
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>
35 #include <linux/interrupt.h>
37 #include <linux/tcp.h>
38 #include <linux/sctp.h>
39 #include <linux/pkt_sched.h>
40 #include <linux/ipv6.h>
41 #include <linux/slab.h>
42 #include <net/checksum.h>
43 #include <net/ip6_checksum.h>
44 #include <linux/ethtool.h>
46 #include <linux/if_vlan.h>
47 #include <linux/prefetch.h>
48 #include <scsi/fc/fc_fcoe.h>
51 #include "ixgbe_common.h"
52 #include "ixgbe_dcb_82599.h"
53 #include "ixgbe_sriov.h"
55 char ixgbe_driver_name[] = "ixgbe";
56 static const char ixgbe_driver_string[] =
57 "Intel(R) 10 Gigabit PCI Express Network Driver";
58 char ixgbe_default_device_descr[] =
59 "Intel(R) 10 Gigabit Network Connection";
63 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
64 __stringify(BUILD) "-k"
65 const char ixgbe_driver_version[] = DRV_VERSION;
66 static const char ixgbe_copyright[] =
67 "Copyright (c) 1999-2012 Intel Corporation.";
69 static const struct ixgbe_info *ixgbe_info_tbl[] = {
70 [board_82598] = &ixgbe_82598_info,
71 [board_82599] = &ixgbe_82599_info,
72 [board_X540] = &ixgbe_X540_info,
75 /* ixgbe_pci_tbl - PCI Device ID Table
77 * Wildcard entries (PCI_ANY_ID) should come last
78 * Last entry must be all 0s
80 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
81 * Class, Class Mask, private data (not used) }
83 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
84 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
85 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
112 /* required last entry */
115 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
117 #ifdef CONFIG_IXGBE_DCA
118 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
120 static struct notifier_block dca_notifier = {
121 .notifier_call = ixgbe_notify_dca,
127 #ifdef CONFIG_PCI_IOV
128 static unsigned int max_vfs;
129 module_param(max_vfs, uint, 0);
130 MODULE_PARM_DESC(max_vfs,
131 "Maximum number of virtual functions to allocate per physical function");
132 #endif /* CONFIG_PCI_IOV */
134 static unsigned int allow_unsupported_sfp;
135 module_param(allow_unsupported_sfp, uint, 0);
136 MODULE_PARM_DESC(allow_unsupported_sfp,
137 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
139 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
140 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
141 MODULE_LICENSE("GPL");
142 MODULE_VERSION(DRV_VERSION);
144 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
146 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
148 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
149 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
150 schedule_work(&adapter->service_task);
153 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
155 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
157 /* flush memory to make sure state is correct before next watchdog */
158 smp_mb__before_clear_bit();
159 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
162 struct ixgbe_reg_info {
167 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
169 /* General Registers */
170 {IXGBE_CTRL, "CTRL"},
171 {IXGBE_STATUS, "STATUS"},
172 {IXGBE_CTRL_EXT, "CTRL_EXT"},
174 /* Interrupt Registers */
175 {IXGBE_EICR, "EICR"},
178 {IXGBE_SRRCTL(0), "SRRCTL"},
179 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
180 {IXGBE_RDLEN(0), "RDLEN"},
181 {IXGBE_RDH(0), "RDH"},
182 {IXGBE_RDT(0), "RDT"},
183 {IXGBE_RXDCTL(0), "RXDCTL"},
184 {IXGBE_RDBAL(0), "RDBAL"},
185 {IXGBE_RDBAH(0), "RDBAH"},
188 {IXGBE_TDBAL(0), "TDBAL"},
189 {IXGBE_TDBAH(0), "TDBAH"},
190 {IXGBE_TDLEN(0), "TDLEN"},
191 {IXGBE_TDH(0), "TDH"},
192 {IXGBE_TDT(0), "TDT"},
193 {IXGBE_TXDCTL(0), "TXDCTL"},
195 /* List Terminator */
201 * ixgbe_regdump - register printout routine
203 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
209 switch (reginfo->ofs) {
210 case IXGBE_SRRCTL(0):
211 for (i = 0; i < 64; i++)
212 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
214 case IXGBE_DCA_RXCTRL(0):
215 for (i = 0; i < 64; i++)
216 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
219 for (i = 0; i < 64; i++)
220 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
223 for (i = 0; i < 64; i++)
224 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
227 for (i = 0; i < 64; i++)
228 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
230 case IXGBE_RXDCTL(0):
231 for (i = 0; i < 64; i++)
232 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
235 for (i = 0; i < 64; i++)
236 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
239 for (i = 0; i < 64; i++)
240 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
243 for (i = 0; i < 64; i++)
244 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
247 for (i = 0; i < 64; i++)
248 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
251 for (i = 0; i < 64; i++)
252 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
255 for (i = 0; i < 64; i++)
256 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
259 for (i = 0; i < 64; i++)
260 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
262 case IXGBE_TXDCTL(0):
263 for (i = 0; i < 64; i++)
264 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
267 pr_info("%-15s %08x\n", reginfo->name,
268 IXGBE_READ_REG(hw, reginfo->ofs));
272 for (i = 0; i < 8; i++) {
273 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
274 pr_err("%-15s", rname);
275 for (j = 0; j < 8; j++)
276 pr_cont(" %08x", regs[i*8+j]);
283 * ixgbe_dump - Print registers, tx-rings and rx-rings
285 static void ixgbe_dump(struct ixgbe_adapter *adapter)
287 struct net_device *netdev = adapter->netdev;
288 struct ixgbe_hw *hw = &adapter->hw;
289 struct ixgbe_reg_info *reginfo;
291 struct ixgbe_ring *tx_ring;
292 struct ixgbe_tx_buffer *tx_buffer;
293 union ixgbe_adv_tx_desc *tx_desc;
294 struct my_u0 { u64 a; u64 b; } *u0;
295 struct ixgbe_ring *rx_ring;
296 union ixgbe_adv_rx_desc *rx_desc;
297 struct ixgbe_rx_buffer *rx_buffer_info;
301 if (!netif_msg_hw(adapter))
304 /* Print netdevice Info */
306 dev_info(&adapter->pdev->dev, "Net device Info\n");
307 pr_info("Device Name state "
308 "trans_start last_rx\n");
309 pr_info("%-15s %016lX %016lX %016lX\n",
316 /* Print Registers */
317 dev_info(&adapter->pdev->dev, "Register Dump\n");
318 pr_info(" Register Name Value\n");
319 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
320 reginfo->name; reginfo++) {
321 ixgbe_regdump(hw, reginfo);
324 /* Print TX Ring Summary */
325 if (!netdev || !netif_running(netdev))
328 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
329 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
330 for (n = 0; n < adapter->num_tx_queues; n++) {
331 tx_ring = adapter->tx_ring[n];
332 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
333 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
334 n, tx_ring->next_to_use, tx_ring->next_to_clean,
335 (u64)dma_unmap_addr(tx_buffer, dma),
336 dma_unmap_len(tx_buffer, len),
337 tx_buffer->next_to_watch,
338 (u64)tx_buffer->time_stamp);
342 if (!netif_msg_tx_done(adapter))
343 goto rx_ring_summary;
345 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
347 /* Transmit Descriptor Formats
349 * Advanced Transmit Descriptor
350 * +--------------------------------------------------------------+
351 * 0 | Buffer Address [63:0] |
352 * +--------------------------------------------------------------+
353 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
354 * +--------------------------------------------------------------+
355 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
358 for (n = 0; n < adapter->num_tx_queues; n++) {
359 tx_ring = adapter->tx_ring[n];
360 pr_info("------------------------------------\n");
361 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
362 pr_info("------------------------------------\n");
363 pr_info("T [desc] [address 63:0 ] "
364 "[PlPOIdStDDt Ln] [bi->dma ] "
365 "leng ntw timestamp bi->skb\n");
367 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
368 tx_desc = IXGBE_TX_DESC(tx_ring, i);
369 tx_buffer = &tx_ring->tx_buffer_info[i];
370 u0 = (struct my_u0 *)tx_desc;
371 pr_info("T [0x%03X] %016llX %016llX %016llX"
372 " %04X %p %016llX %p", i,
375 (u64)dma_unmap_addr(tx_buffer, dma),
376 dma_unmap_len(tx_buffer, len),
377 tx_buffer->next_to_watch,
378 (u64)tx_buffer->time_stamp,
380 if (i == tx_ring->next_to_use &&
381 i == tx_ring->next_to_clean)
383 else if (i == tx_ring->next_to_use)
385 else if (i == tx_ring->next_to_clean)
390 if (netif_msg_pktdata(adapter) &&
391 dma_unmap_len(tx_buffer, len) != 0)
392 print_hex_dump(KERN_INFO, "",
393 DUMP_PREFIX_ADDRESS, 16, 1,
394 phys_to_virt(dma_unmap_addr(tx_buffer,
396 dma_unmap_len(tx_buffer, len),
401 /* Print RX Rings Summary */
403 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
404 pr_info("Queue [NTU] [NTC]\n");
405 for (n = 0; n < adapter->num_rx_queues; n++) {
406 rx_ring = adapter->rx_ring[n];
407 pr_info("%5d %5X %5X\n",
408 n, rx_ring->next_to_use, rx_ring->next_to_clean);
412 if (!netif_msg_rx_status(adapter))
415 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
417 /* Advanced Receive Descriptor (Read) Format
419 * +-----------------------------------------------------+
420 * 0 | Packet Buffer Address [63:1] |A0/NSE|
421 * +----------------------------------------------+------+
422 * 8 | Header Buffer Address [63:1] | DD |
423 * +-----------------------------------------------------+
426 * Advanced Receive Descriptor (Write-Back) Format
428 * 63 48 47 32 31 30 21 20 16 15 4 3 0
429 * +------------------------------------------------------+
430 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
431 * | Checksum Ident | | | | Type | Type |
432 * +------------------------------------------------------+
433 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
434 * +------------------------------------------------------+
435 * 63 48 47 32 31 20 19 0
437 for (n = 0; n < adapter->num_rx_queues; n++) {
438 rx_ring = adapter->rx_ring[n];
439 pr_info("------------------------------------\n");
440 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
441 pr_info("------------------------------------\n");
442 pr_info("R [desc] [ PktBuf A0] "
443 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
444 "<-- Adv Rx Read format\n");
445 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
446 "[vl er S cks ln] ---------------- [bi->skb] "
447 "<-- Adv Rx Write-Back format\n");
449 for (i = 0; i < rx_ring->count; i++) {
450 rx_buffer_info = &rx_ring->rx_buffer_info[i];
451 rx_desc = IXGBE_RX_DESC(rx_ring, i);
452 u0 = (struct my_u0 *)rx_desc;
453 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
454 if (staterr & IXGBE_RXD_STAT_DD) {
455 /* Descriptor Done */
456 pr_info("RWB[0x%03X] %016llX "
457 "%016llX ---------------- %p", i,
460 rx_buffer_info->skb);
462 pr_info("R [0x%03X] %016llX "
463 "%016llX %016llX %p", i,
466 (u64)rx_buffer_info->dma,
467 rx_buffer_info->skb);
469 if (netif_msg_pktdata(adapter)) {
470 print_hex_dump(KERN_INFO, "",
471 DUMP_PREFIX_ADDRESS, 16, 1,
472 phys_to_virt(rx_buffer_info->dma),
473 ixgbe_rx_bufsz(rx_ring), true);
477 if (i == rx_ring->next_to_use)
479 else if (i == rx_ring->next_to_clean)
491 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
495 /* Let firmware take over control of h/w */
496 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
497 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
498 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
501 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
505 /* Let firmware know the driver has taken over */
506 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
507 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
508 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
512 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
513 * @adapter: pointer to adapter struct
514 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
515 * @queue: queue to map the corresponding interrupt to
516 * @msix_vector: the vector to map to the corresponding queue
519 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
520 u8 queue, u8 msix_vector)
523 struct ixgbe_hw *hw = &adapter->hw;
524 switch (hw->mac.type) {
525 case ixgbe_mac_82598EB:
526 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
529 index = (((direction * 64) + queue) >> 2) & 0x1F;
530 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
531 ivar &= ~(0xFF << (8 * (queue & 0x3)));
532 ivar |= (msix_vector << (8 * (queue & 0x3)));
533 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
535 case ixgbe_mac_82599EB:
537 if (direction == -1) {
539 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
540 index = ((queue & 1) * 8);
541 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
542 ivar &= ~(0xFF << index);
543 ivar |= (msix_vector << index);
544 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
547 /* tx or rx causes */
548 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
549 index = ((16 * (queue & 1)) + (8 * direction));
550 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
551 ivar &= ~(0xFF << index);
552 ivar |= (msix_vector << index);
553 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
561 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
566 switch (adapter->hw.mac.type) {
567 case ixgbe_mac_82598EB:
568 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
569 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
571 case ixgbe_mac_82599EB:
573 mask = (qmask & 0xFFFFFFFF);
574 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
575 mask = (qmask >> 32);
576 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
583 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
584 struct ixgbe_tx_buffer *tx_buffer)
586 if (tx_buffer->skb) {
587 dev_kfree_skb_any(tx_buffer->skb);
588 if (dma_unmap_len(tx_buffer, len))
589 dma_unmap_single(ring->dev,
590 dma_unmap_addr(tx_buffer, dma),
591 dma_unmap_len(tx_buffer, len),
593 } else if (dma_unmap_len(tx_buffer, len)) {
594 dma_unmap_page(ring->dev,
595 dma_unmap_addr(tx_buffer, dma),
596 dma_unmap_len(tx_buffer, len),
599 tx_buffer->next_to_watch = NULL;
600 tx_buffer->skb = NULL;
601 dma_unmap_len_set(tx_buffer, len, 0);
602 /* tx_buffer must be completely set up in the transmit path */
605 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
607 struct ixgbe_hw *hw = &adapter->hw;
608 struct ixgbe_hw_stats *hwstats = &adapter->stats;
613 if ((hw->fc.current_mode == ixgbe_fc_full) ||
614 (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
615 switch (hw->mac.type) {
616 case ixgbe_mac_82598EB:
617 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
620 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
622 hwstats->lxoffrxc += data;
624 /* refill credits (no tx hang) if we received xoff */
628 for (i = 0; i < adapter->num_tx_queues; i++)
629 clear_bit(__IXGBE_HANG_CHECK_ARMED,
630 &adapter->tx_ring[i]->state);
632 } else if (!(adapter->dcb_cfg.pfc_mode_enable))
635 /* update stats for each tc, only valid with PFC enabled */
636 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
637 switch (hw->mac.type) {
638 case ixgbe_mac_82598EB:
639 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
642 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
644 hwstats->pxoffrxc[i] += xoff[i];
647 /* disarm tx queues that have received xoff frames */
648 for (i = 0; i < adapter->num_tx_queues; i++) {
649 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
650 u8 tc = tx_ring->dcb_tc;
653 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
657 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
659 return ring->stats.packets;
662 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
664 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
665 struct ixgbe_hw *hw = &adapter->hw;
667 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
668 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
671 return (head < tail) ?
672 tail - head : (tail + ring->count - head);
677 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
679 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
680 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
681 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
684 clear_check_for_tx_hang(tx_ring);
687 * Check for a hung queue, but be thorough. This verifies
688 * that a transmit has been completed since the previous
689 * check AND there is at least one packet pending. The
690 * ARMED bit is set to indicate a potential hang. The
691 * bit is cleared if a pause frame is received to remove
692 * false hang detection due to PFC or 802.3x frames. By
693 * requiring this to fail twice we avoid races with
694 * pfc clearing the ARMED bit and conditions where we
695 * run the check_tx_hang logic with a transmit completion
696 * pending but without time to complete it yet.
698 if ((tx_done_old == tx_done) && tx_pending) {
699 /* make sure it is true for two checks in a row */
700 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
703 /* update completed stats and continue */
704 tx_ring->tx_stats.tx_done_old = tx_done;
705 /* reset the countdown */
706 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
713 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
714 * @adapter: driver private struct
716 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
719 /* Do the reset outside of interrupt context */
720 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
721 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
722 ixgbe_service_event_schedule(adapter);
727 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
728 * @q_vector: structure containing interrupt and ring information
729 * @tx_ring: tx ring to clean
731 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
732 struct ixgbe_ring *tx_ring)
734 struct ixgbe_adapter *adapter = q_vector->adapter;
735 struct ixgbe_tx_buffer *tx_buffer;
736 union ixgbe_adv_tx_desc *tx_desc;
737 unsigned int total_bytes = 0, total_packets = 0;
738 unsigned int budget = q_vector->tx.work_limit;
739 unsigned int i = tx_ring->next_to_clean;
741 if (test_bit(__IXGBE_DOWN, &adapter->state))
744 tx_buffer = &tx_ring->tx_buffer_info[i];
745 tx_desc = IXGBE_TX_DESC(tx_ring, i);
749 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
751 /* if next_to_watch is not set then there is no work pending */
755 /* prevent any other reads prior to eop_desc */
758 /* if DD is not set pending work has not been completed */
759 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
762 /* clear next_to_watch to prevent false hangs */
763 tx_buffer->next_to_watch = NULL;
765 /* update the statistics for this packet */
766 total_bytes += tx_buffer->bytecount;
767 total_packets += tx_buffer->gso_segs;
770 dev_kfree_skb_any(tx_buffer->skb);
772 /* unmap skb header data */
773 dma_unmap_single(tx_ring->dev,
774 dma_unmap_addr(tx_buffer, dma),
775 dma_unmap_len(tx_buffer, len),
778 /* clear tx_buffer data */
779 tx_buffer->skb = NULL;
780 dma_unmap_len_set(tx_buffer, len, 0);
782 /* unmap remaining buffers */
783 while (tx_desc != eop_desc) {
789 tx_buffer = tx_ring->tx_buffer_info;
790 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
793 /* unmap any remaining paged data */
794 if (dma_unmap_len(tx_buffer, len)) {
795 dma_unmap_page(tx_ring->dev,
796 dma_unmap_addr(tx_buffer, dma),
797 dma_unmap_len(tx_buffer, len),
799 dma_unmap_len_set(tx_buffer, len, 0);
803 /* move us one more past the eop_desc for start of next pkt */
809 tx_buffer = tx_ring->tx_buffer_info;
810 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
813 /* issue prefetch for next Tx descriptor */
816 /* update budget accounting */
818 } while (likely(budget));
821 tx_ring->next_to_clean = i;
822 u64_stats_update_begin(&tx_ring->syncp);
823 tx_ring->stats.bytes += total_bytes;
824 tx_ring->stats.packets += total_packets;
825 u64_stats_update_end(&tx_ring->syncp);
826 q_vector->tx.total_bytes += total_bytes;
827 q_vector->tx.total_packets += total_packets;
829 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
830 /* schedule immediate reset if we believe we hung */
831 struct ixgbe_hw *hw = &adapter->hw;
832 e_err(drv, "Detected Tx Unit Hang\n"
834 " TDH, TDT <%x>, <%x>\n"
835 " next_to_use <%x>\n"
836 " next_to_clean <%x>\n"
837 "tx_buffer_info[next_to_clean]\n"
838 " time_stamp <%lx>\n"
840 tx_ring->queue_index,
841 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
842 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
843 tx_ring->next_to_use, i,
844 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
846 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
849 "tx hang %d detected on queue %d, resetting adapter\n",
850 adapter->tx_timeout_count + 1, tx_ring->queue_index);
852 /* schedule immediate reset if we believe we hung */
853 ixgbe_tx_timeout_reset(adapter);
855 /* the adapter is about to reset, no point in enabling stuff */
859 netdev_tx_completed_queue(txring_txq(tx_ring),
860 total_packets, total_bytes);
862 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
863 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
864 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
865 /* Make sure that anybody stopping the queue after this
866 * sees the new next_to_clean.
869 if (__netif_subqueue_stopped(tx_ring->netdev,
870 tx_ring->queue_index)
871 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
872 netif_wake_subqueue(tx_ring->netdev,
873 tx_ring->queue_index);
874 ++tx_ring->tx_stats.restart_queue;
881 #ifdef CONFIG_IXGBE_DCA
882 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
883 struct ixgbe_ring *tx_ring,
886 struct ixgbe_hw *hw = &adapter->hw;
887 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
890 switch (hw->mac.type) {
891 case ixgbe_mac_82598EB:
892 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
894 case ixgbe_mac_82599EB:
896 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
897 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
900 /* for unknown hardware do not write register */
905 * We can enable relaxed ordering for reads, but not writes when
906 * DCA is enabled. This is due to a known issue in some chipsets
907 * which will cause the DCA tag to be cleared.
909 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
910 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
911 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
913 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
916 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
917 struct ixgbe_ring *rx_ring,
920 struct ixgbe_hw *hw = &adapter->hw;
921 u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
922 u8 reg_idx = rx_ring->reg_idx;
925 switch (hw->mac.type) {
926 case ixgbe_mac_82599EB:
928 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
935 * We can enable relaxed ordering for reads, but not writes when
936 * DCA is enabled. This is due to a known issue in some chipsets
937 * which will cause the DCA tag to be cleared.
939 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
940 IXGBE_DCA_RXCTRL_DATA_DCA_EN |
941 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
943 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
946 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
948 struct ixgbe_adapter *adapter = q_vector->adapter;
949 struct ixgbe_ring *ring;
952 if (q_vector->cpu == cpu)
955 ixgbe_for_each_ring(ring, q_vector->tx)
956 ixgbe_update_tx_dca(adapter, ring, cpu);
958 ixgbe_for_each_ring(ring, q_vector->rx)
959 ixgbe_update_rx_dca(adapter, ring, cpu);
966 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
971 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
974 /* always use CB2 mode, difference is masked in the CB driver */
975 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
977 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
978 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
982 for (i = 0; i < num_q_vectors; i++) {
983 adapter->q_vector[i]->cpu = -1;
984 ixgbe_update_dca(adapter->q_vector[i]);
988 static int __ixgbe_notify_dca(struct device *dev, void *data)
990 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
991 unsigned long event = *(unsigned long *)data;
993 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
997 case DCA_PROVIDER_ADD:
998 /* if we're already enabled, don't do it again */
999 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1001 if (dca_add_requester(dev) == 0) {
1002 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1003 ixgbe_setup_dca(adapter);
1006 /* Fall Through since DCA is disabled. */
1007 case DCA_PROVIDER_REMOVE:
1008 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1009 dca_remove_requester(dev);
1010 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1011 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1019 #endif /* CONFIG_IXGBE_DCA */
1020 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1021 union ixgbe_adv_rx_desc *rx_desc,
1022 struct sk_buff *skb)
1024 if (ring->netdev->features & NETIF_F_RXHASH)
1025 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1030 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1031 * @adapter: address of board private structure
1032 * @rx_desc: advanced rx descriptor
1034 * Returns : true if it is FCoE pkt
1036 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_adapter *adapter,
1037 union ixgbe_adv_rx_desc *rx_desc)
1039 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1041 return (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
1042 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1043 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1044 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1047 #endif /* IXGBE_FCOE */
1049 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1050 * @ring: structure containing ring specific data
1051 * @rx_desc: current Rx descriptor being processed
1052 * @skb: skb currently being received and modified
1054 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1055 union ixgbe_adv_rx_desc *rx_desc,
1056 struct sk_buff *skb)
1058 skb_checksum_none_assert(skb);
1060 /* Rx csum disabled */
1061 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1064 /* if IP and error */
1065 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1066 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1067 ring->rx_stats.csum_err++;
1071 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1074 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1075 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1078 * 82599 errata, UDP frames with a 0 checksum can be marked as
1081 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1082 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1085 ring->rx_stats.csum_err++;
1089 /* It must be a TCP or UDP packet with a valid checksum */
1090 skb->ip_summed = CHECKSUM_UNNECESSARY;
1093 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1095 rx_ring->next_to_use = val;
1097 /* update next to alloc since we have filled the ring */
1098 rx_ring->next_to_alloc = val;
1100 * Force memory writes to complete before letting h/w
1101 * know there are new descriptors to fetch. (Only
1102 * applicable for weak-ordered memory model archs,
1106 writel(val, rx_ring->tail);
1109 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1110 struct ixgbe_rx_buffer *bi)
1112 struct page *page = bi->page;
1113 dma_addr_t dma = bi->dma;
1115 /* since we are recycling buffers we should seldom need to alloc */
1119 /* alloc new page for storage */
1120 if (likely(!page)) {
1121 page = alloc_pages(GFP_ATOMIC | __GFP_COLD,
1122 ixgbe_rx_pg_order(rx_ring));
1123 if (unlikely(!page)) {
1124 rx_ring->rx_stats.alloc_rx_page_failed++;
1130 /* map page for use */
1131 dma = dma_map_page(rx_ring->dev, page, 0,
1132 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1135 * if mapping failed free memory back to system since
1136 * there isn't much point in holding memory we can't use
1138 if (dma_mapping_error(rx_ring->dev, dma)) {
1142 rx_ring->rx_stats.alloc_rx_page_failed++;
1147 bi->page_offset ^= ixgbe_rx_bufsz(rx_ring);
1153 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1154 * @rx_ring: ring to place buffers on
1155 * @cleaned_count: number of buffers to replace
1157 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1159 union ixgbe_adv_rx_desc *rx_desc;
1160 struct ixgbe_rx_buffer *bi;
1161 u16 i = rx_ring->next_to_use;
1167 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1168 bi = &rx_ring->rx_buffer_info[i];
1169 i -= rx_ring->count;
1172 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1176 * Refresh the desc even if buffer_addrs didn't change
1177 * because each write-back erases this info.
1179 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1185 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1186 bi = rx_ring->rx_buffer_info;
1187 i -= rx_ring->count;
1190 /* clear the hdr_addr for the next_to_use descriptor */
1191 rx_desc->read.hdr_addr = 0;
1194 } while (cleaned_count);
1196 i += rx_ring->count;
1198 if (rx_ring->next_to_use != i)
1199 ixgbe_release_rx_desc(rx_ring, i);
1203 * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1204 * @data: pointer to the start of the headers
1205 * @max_len: total length of section to find headers in
1207 * This function is meant to determine the length of headers that will
1208 * be recognized by hardware for LRO, GRO, and RSC offloads. The main
1209 * motivation of doing this is to only perform one pull for IPv4 TCP
1210 * packets so that we can do basic things like calculating the gso_size
1211 * based on the average data per packet.
1213 static unsigned int ixgbe_get_headlen(unsigned char *data,
1214 unsigned int max_len)
1217 unsigned char *network;
1220 struct vlan_hdr *vlan;
1225 u8 nexthdr = 0; /* default to not TCP */
1228 /* this should never happen, but better safe than sorry */
1229 if (max_len < ETH_HLEN)
1232 /* initialize network frame pointer */
1235 /* set first protocol and move network header forward */
1236 protocol = hdr.eth->h_proto;
1237 hdr.network += ETH_HLEN;
1239 /* handle any vlan tag if present */
1240 if (protocol == __constant_htons(ETH_P_8021Q)) {
1241 if ((hdr.network - data) > (max_len - VLAN_HLEN))
1244 protocol = hdr.vlan->h_vlan_encapsulated_proto;
1245 hdr.network += VLAN_HLEN;
1248 /* handle L3 protocols */
1249 if (protocol == __constant_htons(ETH_P_IP)) {
1250 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1253 /* access ihl as a u8 to avoid unaligned access on ia64 */
1254 hlen = (hdr.network[0] & 0x0F) << 2;
1256 /* verify hlen meets minimum size requirements */
1257 if (hlen < sizeof(struct iphdr))
1258 return hdr.network - data;
1260 /* record next protocol */
1261 nexthdr = hdr.ipv4->protocol;
1262 hdr.network += hlen;
1264 } else if (protocol == __constant_htons(ETH_P_FCOE)) {
1265 if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1267 hdr.network += FCOE_HEADER_LEN;
1270 return hdr.network - data;
1273 /* finally sort out TCP */
1274 if (nexthdr == IPPROTO_TCP) {
1275 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1278 /* access doff as a u8 to avoid unaligned access on ia64 */
1279 hlen = (hdr.network[12] & 0xF0) >> 2;
1281 /* verify hlen meets minimum size requirements */
1282 if (hlen < sizeof(struct tcphdr))
1283 return hdr.network - data;
1285 hdr.network += hlen;
1289 * If everything has gone correctly hdr.network should be the
1290 * data section of the packet and will be the end of the header.
1291 * If not then it probably represents the end of the last recognized
1294 if ((hdr.network - data) < max_len)
1295 return hdr.network - data;
1300 static void ixgbe_get_rsc_cnt(struct ixgbe_ring *rx_ring,
1301 union ixgbe_adv_rx_desc *rx_desc,
1302 struct sk_buff *skb)
1307 if (!ring_is_rsc_enabled(rx_ring))
1310 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1311 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1313 /* If this is an RSC frame rsc_cnt should be non-zero */
1317 rsc_cnt = le32_to_cpu(rsc_enabled);
1318 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1320 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1323 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1324 struct sk_buff *skb)
1326 u16 hdr_len = skb_headlen(skb);
1328 /* set gso_size to avoid messing up TCP MSS */
1329 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1330 IXGBE_CB(skb)->append_cnt);
1333 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1334 struct sk_buff *skb)
1336 /* if append_cnt is 0 then frame is not RSC */
1337 if (!IXGBE_CB(skb)->append_cnt)
1340 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1341 rx_ring->rx_stats.rsc_flush++;
1343 ixgbe_set_rsc_gso_size(rx_ring, skb);
1345 /* gso_size is computed using append_cnt so always clear it last */
1346 IXGBE_CB(skb)->append_cnt = 0;
1350 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1351 * @rx_ring: rx descriptor ring packet is being transacted on
1352 * @rx_desc: pointer to the EOP Rx descriptor
1353 * @skb: pointer to current skb being populated
1355 * This function checks the ring, descriptor, and packet information in
1356 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1357 * other fields within the skb.
1359 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1360 union ixgbe_adv_rx_desc *rx_desc,
1361 struct sk_buff *skb)
1363 ixgbe_update_rsc_stats(rx_ring, skb);
1365 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1367 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1369 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1370 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1371 __vlan_hwaccel_put_tag(skb, vid);
1374 skb_record_rx_queue(skb, rx_ring->queue_index);
1376 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1379 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1380 struct sk_buff *skb)
1382 struct ixgbe_adapter *adapter = q_vector->adapter;
1384 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1385 napi_gro_receive(&q_vector->napi, skb);
1391 * ixgbe_is_non_eop - process handling of non-EOP buffers
1392 * @rx_ring: Rx ring being processed
1393 * @rx_desc: Rx descriptor for current buffer
1394 * @skb: Current socket buffer containing buffer in progress
1396 * This function updates next to clean. If the buffer is an EOP buffer
1397 * this function exits returning false, otherwise it will place the
1398 * sk_buff in the next buffer to be chained and return true indicating
1399 * that this is in fact a non-EOP buffer.
1401 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1402 union ixgbe_adv_rx_desc *rx_desc,
1403 struct sk_buff *skb)
1405 u32 ntc = rx_ring->next_to_clean + 1;
1407 /* fetch, update, and store next to clean */
1408 ntc = (ntc < rx_ring->count) ? ntc : 0;
1409 rx_ring->next_to_clean = ntc;
1411 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1413 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1416 /* append_cnt indicates packet is RSC, if so fetch nextp */
1417 if (IXGBE_CB(skb)->append_cnt) {
1418 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1419 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1420 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1423 /* place skb in next buffer to be received */
1424 rx_ring->rx_buffer_info[ntc].skb = skb;
1425 rx_ring->rx_stats.non_eop_descs++;
1431 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1432 * @rx_ring: rx descriptor ring packet is being transacted on
1433 * @rx_desc: pointer to the EOP Rx descriptor
1434 * @skb: pointer to current skb being fixed
1436 * Check for corrupted packet headers caused by senders on the local L2
1437 * embedded NIC switch not setting up their Tx Descriptors right. These
1438 * should be very rare.
1440 * Also address the case where we are pulling data in on pages only
1441 * and as such no data is present in the skb header.
1443 * In addition if skb is not at least 60 bytes we need to pad it so that
1444 * it is large enough to qualify as a valid Ethernet frame.
1446 * Returns true if an error was encountered and skb was freed.
1448 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1449 union ixgbe_adv_rx_desc *rx_desc,
1450 struct sk_buff *skb)
1452 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1453 struct net_device *netdev = rx_ring->netdev;
1455 unsigned int pull_len;
1457 /* if the page was released unmap it, else just sync our portion */
1458 if (unlikely(IXGBE_CB(skb)->page_released)) {
1459 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1460 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1461 IXGBE_CB(skb)->page_released = false;
1463 dma_sync_single_range_for_cpu(rx_ring->dev,
1466 ixgbe_rx_bufsz(rx_ring),
1469 IXGBE_CB(skb)->dma = 0;
1471 /* verify that the packet does not have any known errors */
1472 if (unlikely(ixgbe_test_staterr(rx_desc,
1473 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1474 !(netdev->features & NETIF_F_RXALL))) {
1475 dev_kfree_skb_any(skb);
1480 * it is valid to use page_address instead of kmap since we are
1481 * working with pages allocated out of the lomem pool per
1482 * alloc_page(GFP_ATOMIC)
1484 va = skb_frag_address(frag);
1487 * we need the header to contain the greater of either ETH_HLEN or
1488 * 60 bytes if the skb->len is less than 60 for skb_pad.
1490 pull_len = skb_frag_size(frag);
1492 pull_len = ixgbe_get_headlen(va, pull_len);
1494 /* align pull length to size of long to optimize memcpy performance */
1495 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1497 /* update all of the pointers */
1498 skb_frag_size_sub(frag, pull_len);
1499 frag->page_offset += pull_len;
1500 skb->data_len -= pull_len;
1501 skb->tail += pull_len;
1504 * if we sucked the frag empty then we should free it,
1505 * if there are other frags here something is screwed up in hardware
1507 if (skb_frag_size(frag) == 0) {
1508 BUG_ON(skb_shinfo(skb)->nr_frags != 1);
1509 skb_shinfo(skb)->nr_frags = 0;
1510 __skb_frag_unref(frag);
1511 skb->truesize -= ixgbe_rx_bufsz(rx_ring);
1514 /* if skb_pad returns an error the skb was freed */
1515 if (unlikely(skb->len < 60)) {
1516 int pad_len = 60 - skb->len;
1518 if (skb_pad(skb, pad_len))
1520 __skb_put(skb, pad_len);
1527 * ixgbe_can_reuse_page - determine if we can reuse a page
1528 * @rx_buffer: pointer to rx_buffer containing the page we want to reuse
1530 * Returns true if page can be reused in another Rx buffer
1532 static inline bool ixgbe_can_reuse_page(struct ixgbe_rx_buffer *rx_buffer)
1534 struct page *page = rx_buffer->page;
1536 /* if we are only owner of page and it is local we can reuse it */
1537 return likely(page_count(page) == 1) &&
1538 likely(page_to_nid(page) == numa_node_id());
1542 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1543 * @rx_ring: rx descriptor ring to store buffers on
1544 * @old_buff: donor buffer to have page reused
1546 * Syncronizes page for reuse by the adapter
1548 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1549 struct ixgbe_rx_buffer *old_buff)
1551 struct ixgbe_rx_buffer *new_buff;
1552 u16 nta = rx_ring->next_to_alloc;
1553 u16 bufsz = ixgbe_rx_bufsz(rx_ring);
1555 new_buff = &rx_ring->rx_buffer_info[nta];
1557 /* update, and store next to alloc */
1559 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1561 /* transfer page from old buffer to new buffer */
1562 new_buff->page = old_buff->page;
1563 new_buff->dma = old_buff->dma;
1565 /* flip page offset to other buffer and store to new_buff */
1566 new_buff->page_offset = old_buff->page_offset ^ bufsz;
1568 /* sync the buffer for use by the device */
1569 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1570 new_buff->page_offset, bufsz,
1573 /* bump ref count on page before it is given to the stack */
1574 get_page(new_buff->page);
1578 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1579 * @rx_ring: rx descriptor ring to transact packets on
1580 * @rx_buffer: buffer containing page to add
1581 * @rx_desc: descriptor containing length of buffer written by hardware
1582 * @skb: sk_buff to place the data into
1584 * This function is based on skb_add_rx_frag. I would have used that
1585 * function however it doesn't handle the truesize case correctly since we
1586 * are allocating more memory than might be used for a single receive.
1588 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1589 struct ixgbe_rx_buffer *rx_buffer,
1590 struct sk_buff *skb, int size)
1592 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1593 rx_buffer->page, rx_buffer->page_offset,
1596 skb->data_len += size;
1597 skb->truesize += ixgbe_rx_bufsz(rx_ring);
1601 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1602 * @q_vector: structure containing interrupt and ring information
1603 * @rx_ring: rx descriptor ring to transact packets on
1604 * @budget: Total limit on number of packets to process
1606 * This function provides a "bounce buffer" approach to Rx interrupt
1607 * processing. The advantage to this is that on systems that have
1608 * expensive overhead for IOMMU access this provides a means of avoiding
1609 * it by maintaining the mapping of the page to the syste.
1611 * Returns true if all work is completed without reaching budget
1613 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1614 struct ixgbe_ring *rx_ring,
1617 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1619 struct ixgbe_adapter *adapter = q_vector->adapter;
1621 #endif /* IXGBE_FCOE */
1622 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1625 struct ixgbe_rx_buffer *rx_buffer;
1626 union ixgbe_adv_rx_desc *rx_desc;
1627 struct sk_buff *skb;
1631 /* return some buffers to hardware, one at a time is too slow */
1632 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1633 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1637 ntc = rx_ring->next_to_clean;
1638 rx_desc = IXGBE_RX_DESC(rx_ring, ntc);
1639 rx_buffer = &rx_ring->rx_buffer_info[ntc];
1641 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
1645 * This memory barrier is needed to keep us from reading
1646 * any other fields out of the rx_desc until we know the
1647 * RXD_STAT_DD bit is set
1651 page = rx_buffer->page;
1654 skb = rx_buffer->skb;
1657 void *page_addr = page_address(page) +
1658 rx_buffer->page_offset;
1660 /* prefetch first cache line of first page */
1661 prefetch(page_addr);
1662 #if L1_CACHE_BYTES < 128
1663 prefetch(page_addr + L1_CACHE_BYTES);
1666 /* allocate a skb to store the frags */
1667 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1669 if (unlikely(!skb)) {
1670 rx_ring->rx_stats.alloc_rx_buff_failed++;
1675 * we will be copying header into skb->data in
1676 * pskb_may_pull so it is in our interest to prefetch
1677 * it now to avoid a possible cache miss
1679 prefetchw(skb->data);
1682 * Delay unmapping of the first packet. It carries the
1683 * header information, HW may still access the header
1684 * after the writeback. Only unmap it when EOP is
1687 IXGBE_CB(skb)->dma = rx_buffer->dma;
1689 /* we are reusing so sync this buffer for CPU use */
1690 dma_sync_single_range_for_cpu(rx_ring->dev,
1692 rx_buffer->page_offset,
1693 ixgbe_rx_bufsz(rx_ring),
1697 /* pull page into skb */
1698 ixgbe_add_rx_frag(rx_ring, rx_buffer, skb,
1699 le16_to_cpu(rx_desc->wb.upper.length));
1701 if (ixgbe_can_reuse_page(rx_buffer)) {
1702 /* hand second half of page back to the ring */
1703 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1704 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1705 /* the page has been released from the ring */
1706 IXGBE_CB(skb)->page_released = true;
1708 /* we are not reusing the buffer so unmap it */
1709 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1710 ixgbe_rx_pg_size(rx_ring),
1714 /* clear contents of buffer_info */
1715 rx_buffer->skb = NULL;
1717 rx_buffer->page = NULL;
1719 ixgbe_get_rsc_cnt(rx_ring, rx_desc, skb);
1723 /* place incomplete frames back on ring for completion */
1724 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
1727 /* verify the packet layout is correct */
1728 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
1731 /* probably a little skewed due to removing CRC */
1732 total_rx_bytes += skb->len;
1735 /* populate checksum, timestamp, VLAN, and protocol */
1736 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
1739 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1740 if (ixgbe_rx_is_fcoe(adapter, rx_desc)) {
1741 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1743 dev_kfree_skb_any(skb);
1748 #endif /* IXGBE_FCOE */
1749 ixgbe_rx_skb(q_vector, skb);
1751 /* update budget accounting */
1753 } while (likely(budget));
1756 /* include DDPed FCoE data */
1757 if (ddp_bytes > 0) {
1760 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1761 sizeof(struct fc_frame_header) -
1762 sizeof(struct fcoe_crc_eof);
1765 total_rx_bytes += ddp_bytes;
1766 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1769 #endif /* IXGBE_FCOE */
1770 u64_stats_update_begin(&rx_ring->syncp);
1771 rx_ring->stats.packets += total_rx_packets;
1772 rx_ring->stats.bytes += total_rx_bytes;
1773 u64_stats_update_end(&rx_ring->syncp);
1774 q_vector->rx.total_packets += total_rx_packets;
1775 q_vector->rx.total_bytes += total_rx_bytes;
1778 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1784 * ixgbe_configure_msix - Configure MSI-X hardware
1785 * @adapter: board private structure
1787 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1790 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1792 struct ixgbe_q_vector *q_vector;
1793 int q_vectors, v_idx;
1796 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1798 /* Populate MSIX to EITR Select */
1799 if (adapter->num_vfs > 32) {
1800 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1801 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1805 * Populate the IVAR table and set the ITR values to the
1806 * corresponding register.
1808 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1809 struct ixgbe_ring *ring;
1810 q_vector = adapter->q_vector[v_idx];
1812 ixgbe_for_each_ring(ring, q_vector->rx)
1813 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1815 ixgbe_for_each_ring(ring, q_vector->tx)
1816 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1818 if (q_vector->tx.ring && !q_vector->rx.ring) {
1819 /* tx only vector */
1820 if (adapter->tx_itr_setting == 1)
1821 q_vector->itr = IXGBE_10K_ITR;
1823 q_vector->itr = adapter->tx_itr_setting;
1825 /* rx or rx/tx vector */
1826 if (adapter->rx_itr_setting == 1)
1827 q_vector->itr = IXGBE_20K_ITR;
1829 q_vector->itr = adapter->rx_itr_setting;
1832 ixgbe_write_eitr(q_vector);
1835 switch (adapter->hw.mac.type) {
1836 case ixgbe_mac_82598EB:
1837 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1840 case ixgbe_mac_82599EB:
1841 case ixgbe_mac_X540:
1842 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1847 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1849 /* set up to autoclear timer, and the vectors */
1850 mask = IXGBE_EIMS_ENABLE_MASK;
1851 mask &= ~(IXGBE_EIMS_OTHER |
1852 IXGBE_EIMS_MAILBOX |
1855 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1858 enum latency_range {
1862 latency_invalid = 255
1866 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1867 * @q_vector: structure containing interrupt and ring information
1868 * @ring_container: structure containing ring performance data
1870 * Stores a new ITR value based on packets and byte
1871 * counts during the last interrupt. The advantage of per interrupt
1872 * computation is faster updates and more accurate ITR for the current
1873 * traffic pattern. Constants in this function were computed
1874 * based on theoretical maximum wire speed and thresholds were set based
1875 * on testing data as well as attempting to minimize response time
1876 * while increasing bulk throughput.
1877 * this functionality is controlled by the InterruptThrottleRate module
1878 * parameter (see ixgbe_param.c)
1880 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1881 struct ixgbe_ring_container *ring_container)
1883 int bytes = ring_container->total_bytes;
1884 int packets = ring_container->total_packets;
1887 u8 itr_setting = ring_container->itr;
1892 /* simple throttlerate management
1893 * 0-10MB/s lowest (100000 ints/s)
1894 * 10-20MB/s low (20000 ints/s)
1895 * 20-1249MB/s bulk (8000 ints/s)
1897 /* what was last interrupt timeslice? */
1898 timepassed_us = q_vector->itr >> 2;
1899 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1901 switch (itr_setting) {
1902 case lowest_latency:
1903 if (bytes_perint > 10)
1904 itr_setting = low_latency;
1907 if (bytes_perint > 20)
1908 itr_setting = bulk_latency;
1909 else if (bytes_perint <= 10)
1910 itr_setting = lowest_latency;
1913 if (bytes_perint <= 20)
1914 itr_setting = low_latency;
1918 /* clear work counters since we have the values we need */
1919 ring_container->total_bytes = 0;
1920 ring_container->total_packets = 0;
1922 /* write updated itr to ring container */
1923 ring_container->itr = itr_setting;
1927 * ixgbe_write_eitr - write EITR register in hardware specific way
1928 * @q_vector: structure containing interrupt and ring information
1930 * This function is made to be called by ethtool and by the driver
1931 * when it needs to update EITR registers at runtime. Hardware
1932 * specific quirks/differences are taken care of here.
1934 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1936 struct ixgbe_adapter *adapter = q_vector->adapter;
1937 struct ixgbe_hw *hw = &adapter->hw;
1938 int v_idx = q_vector->v_idx;
1939 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1941 switch (adapter->hw.mac.type) {
1942 case ixgbe_mac_82598EB:
1943 /* must write high and low 16 bits to reset counter */
1944 itr_reg |= (itr_reg << 16);
1946 case ixgbe_mac_82599EB:
1947 case ixgbe_mac_X540:
1949 * set the WDIS bit to not clear the timer bits and cause an
1950 * immediate assertion of the interrupt
1952 itr_reg |= IXGBE_EITR_CNT_WDIS;
1957 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1960 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1962 u32 new_itr = q_vector->itr;
1965 ixgbe_update_itr(q_vector, &q_vector->tx);
1966 ixgbe_update_itr(q_vector, &q_vector->rx);
1968 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1970 switch (current_itr) {
1971 /* counts and packets in update_itr are dependent on these numbers */
1972 case lowest_latency:
1973 new_itr = IXGBE_100K_ITR;
1976 new_itr = IXGBE_20K_ITR;
1979 new_itr = IXGBE_8K_ITR;
1985 if (new_itr != q_vector->itr) {
1986 /* do an exponential smoothing */
1987 new_itr = (10 * new_itr * q_vector->itr) /
1988 ((9 * new_itr) + q_vector->itr);
1990 /* save the algorithm value here */
1991 q_vector->itr = new_itr;
1993 ixgbe_write_eitr(q_vector);
1998 * ixgbe_check_overtemp_subtask - check for over temperature
1999 * @adapter: pointer to adapter
2001 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2003 struct ixgbe_hw *hw = &adapter->hw;
2004 u32 eicr = adapter->interrupt_event;
2006 if (test_bit(__IXGBE_DOWN, &adapter->state))
2009 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2010 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2013 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2015 switch (hw->device_id) {
2016 case IXGBE_DEV_ID_82599_T3_LOM:
2018 * Since the warning interrupt is for both ports
2019 * we don't have to check if:
2020 * - This interrupt wasn't for our port.
2021 * - We may have missed the interrupt so always have to
2022 * check if we got a LSC
2024 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2025 !(eicr & IXGBE_EICR_LSC))
2028 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2030 bool link_up = false;
2032 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2038 /* Check if this is not due to overtemp */
2039 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2044 if (!(eicr & IXGBE_EICR_GPI_SDP0))
2049 "Network adapter has been stopped because it has over heated. "
2050 "Restart the computer. If the problem persists, "
2051 "power off the system and replace the adapter\n");
2053 adapter->interrupt_event = 0;
2056 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2058 struct ixgbe_hw *hw = &adapter->hw;
2060 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2061 (eicr & IXGBE_EICR_GPI_SDP1)) {
2062 e_crit(probe, "Fan has stopped, replace the adapter\n");
2063 /* write to clear the interrupt */
2064 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2068 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2070 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2073 switch (adapter->hw.mac.type) {
2074 case ixgbe_mac_82599EB:
2076 * Need to check link state so complete overtemp check
2079 if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2080 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2081 adapter->interrupt_event = eicr;
2082 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2083 ixgbe_service_event_schedule(adapter);
2087 case ixgbe_mac_X540:
2088 if (!(eicr & IXGBE_EICR_TS))
2096 "Network adapter has been stopped because it has over heated. "
2097 "Restart the computer. If the problem persists, "
2098 "power off the system and replace the adapter\n");
2101 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2103 struct ixgbe_hw *hw = &adapter->hw;
2105 if (eicr & IXGBE_EICR_GPI_SDP2) {
2106 /* Clear the interrupt */
2107 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2108 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2109 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2110 ixgbe_service_event_schedule(adapter);
2114 if (eicr & IXGBE_EICR_GPI_SDP1) {
2115 /* Clear the interrupt */
2116 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2117 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2118 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2119 ixgbe_service_event_schedule(adapter);
2124 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2126 struct ixgbe_hw *hw = &adapter->hw;
2129 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2130 adapter->link_check_timeout = jiffies;
2131 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2132 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2133 IXGBE_WRITE_FLUSH(hw);
2134 ixgbe_service_event_schedule(adapter);
2138 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2142 struct ixgbe_hw *hw = &adapter->hw;
2144 switch (hw->mac.type) {
2145 case ixgbe_mac_82598EB:
2146 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2147 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2149 case ixgbe_mac_82599EB:
2150 case ixgbe_mac_X540:
2151 mask = (qmask & 0xFFFFFFFF);
2153 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2154 mask = (qmask >> 32);
2156 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2161 /* skip the flush */
2164 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2168 struct ixgbe_hw *hw = &adapter->hw;
2170 switch (hw->mac.type) {
2171 case ixgbe_mac_82598EB:
2172 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2173 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2175 case ixgbe_mac_82599EB:
2176 case ixgbe_mac_X540:
2177 mask = (qmask & 0xFFFFFFFF);
2179 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2180 mask = (qmask >> 32);
2182 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2187 /* skip the flush */
2191 * ixgbe_irq_enable - Enable default interrupt generation settings
2192 * @adapter: board private structure
2194 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2197 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2199 /* don't reenable LSC while waiting for link */
2200 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2201 mask &= ~IXGBE_EIMS_LSC;
2203 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2204 switch (adapter->hw.mac.type) {
2205 case ixgbe_mac_82599EB:
2206 mask |= IXGBE_EIMS_GPI_SDP0;
2208 case ixgbe_mac_X540:
2209 mask |= IXGBE_EIMS_TS;
2214 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2215 mask |= IXGBE_EIMS_GPI_SDP1;
2216 switch (adapter->hw.mac.type) {
2217 case ixgbe_mac_82599EB:
2218 mask |= IXGBE_EIMS_GPI_SDP1;
2219 mask |= IXGBE_EIMS_GPI_SDP2;
2220 case ixgbe_mac_X540:
2221 mask |= IXGBE_EIMS_ECC;
2222 mask |= IXGBE_EIMS_MAILBOX;
2227 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2228 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2229 mask |= IXGBE_EIMS_FLOW_DIR;
2231 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2233 ixgbe_irq_enable_queues(adapter, ~0);
2235 IXGBE_WRITE_FLUSH(&adapter->hw);
2238 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2240 struct ixgbe_adapter *adapter = data;
2241 struct ixgbe_hw *hw = &adapter->hw;
2245 * Workaround for Silicon errata. Use clear-by-write instead
2246 * of clear-by-read. Reading with EICS will return the
2247 * interrupt causes without clearing, which later be done
2248 * with the write to EICR.
2250 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2251 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2253 if (eicr & IXGBE_EICR_LSC)
2254 ixgbe_check_lsc(adapter);
2256 if (eicr & IXGBE_EICR_MAILBOX)
2257 ixgbe_msg_task(adapter);
2259 switch (hw->mac.type) {
2260 case ixgbe_mac_82599EB:
2261 case ixgbe_mac_X540:
2262 if (eicr & IXGBE_EICR_ECC)
2263 e_info(link, "Received unrecoverable ECC Err, please "
2265 /* Handle Flow Director Full threshold interrupt */
2266 if (eicr & IXGBE_EICR_FLOW_DIR) {
2267 int reinit_count = 0;
2269 for (i = 0; i < adapter->num_tx_queues; i++) {
2270 struct ixgbe_ring *ring = adapter->tx_ring[i];
2271 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2276 /* no more flow director interrupts until after init */
2277 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2278 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2279 ixgbe_service_event_schedule(adapter);
2282 ixgbe_check_sfp_event(adapter, eicr);
2283 ixgbe_check_overtemp_event(adapter, eicr);
2289 ixgbe_check_fan_failure(adapter, eicr);
2291 /* re-enable the original interrupt state, no lsc, no queues */
2292 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2293 ixgbe_irq_enable(adapter, false, false);
2298 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2300 struct ixgbe_q_vector *q_vector = data;
2302 /* EIAM disabled interrupts (on this vector) for us */
2304 if (q_vector->rx.ring || q_vector->tx.ring)
2305 napi_schedule(&q_vector->napi);
2311 * ixgbe_poll - NAPI Rx polling callback
2312 * @napi: structure for representing this polling device
2313 * @budget: how many packets driver is allowed to clean
2315 * This function is used for legacy and MSI, NAPI mode
2317 static int ixgbe_poll(struct napi_struct *napi, int budget)
2319 struct ixgbe_q_vector *q_vector =
2320 container_of(napi, struct ixgbe_q_vector, napi);
2321 struct ixgbe_adapter *adapter = q_vector->adapter;
2322 struct ixgbe_ring *ring;
2323 int per_ring_budget;
2324 bool clean_complete = true;
2326 #ifdef CONFIG_IXGBE_DCA
2327 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2328 ixgbe_update_dca(q_vector);
2331 ixgbe_for_each_ring(ring, q_vector->tx)
2332 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2334 /* attempt to distribute budget to each queue fairly, but don't allow
2335 * the budget to go below 1 because we'll exit polling */
2336 if (q_vector->rx.count > 1)
2337 per_ring_budget = max(budget/q_vector->rx.count, 1);
2339 per_ring_budget = budget;
2341 ixgbe_for_each_ring(ring, q_vector->rx)
2342 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
2345 /* If all work not completed, return budget and keep polling */
2346 if (!clean_complete)
2349 /* all work done, exit the polling mode */
2350 napi_complete(napi);
2351 if (adapter->rx_itr_setting & 1)
2352 ixgbe_set_itr(q_vector);
2353 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2354 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2360 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2361 * @adapter: board private structure
2363 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2364 * interrupts from the kernel.
2366 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2368 struct net_device *netdev = adapter->netdev;
2369 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2373 for (vector = 0; vector < q_vectors; vector++) {
2374 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2375 struct msix_entry *entry = &adapter->msix_entries[vector];
2377 if (q_vector->tx.ring && q_vector->rx.ring) {
2378 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2379 "%s-%s-%d", netdev->name, "TxRx", ri++);
2381 } else if (q_vector->rx.ring) {
2382 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2383 "%s-%s-%d", netdev->name, "rx", ri++);
2384 } else if (q_vector->tx.ring) {
2385 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2386 "%s-%s-%d", netdev->name, "tx", ti++);
2388 /* skip this unused q_vector */
2391 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2392 q_vector->name, q_vector);
2394 e_err(probe, "request_irq failed for MSIX interrupt "
2395 "Error: %d\n", err);
2396 goto free_queue_irqs;
2398 /* If Flow Director is enabled, set interrupt affinity */
2399 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2400 /* assign the mask for this irq */
2401 irq_set_affinity_hint(entry->vector,
2402 &q_vector->affinity_mask);
2406 err = request_irq(adapter->msix_entries[vector].vector,
2407 ixgbe_msix_other, 0, netdev->name, adapter);
2409 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2410 goto free_queue_irqs;
2418 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2420 free_irq(adapter->msix_entries[vector].vector,
2421 adapter->q_vector[vector]);
2423 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2424 pci_disable_msix(adapter->pdev);
2425 kfree(adapter->msix_entries);
2426 adapter->msix_entries = NULL;
2431 * ixgbe_intr - legacy mode Interrupt Handler
2432 * @irq: interrupt number
2433 * @data: pointer to a network interface device structure
2435 static irqreturn_t ixgbe_intr(int irq, void *data)
2437 struct ixgbe_adapter *adapter = data;
2438 struct ixgbe_hw *hw = &adapter->hw;
2439 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2443 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2444 * before the read of EICR.
2446 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2448 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2449 * therefore no explicit interrupt disable is necessary */
2450 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2453 * shared interrupt alert!
2454 * make sure interrupts are enabled because the read will
2455 * have disabled interrupts due to EIAM
2456 * finish the workaround of silicon errata on 82598. Unmask
2457 * the interrupt that we masked before the EICR read.
2459 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2460 ixgbe_irq_enable(adapter, true, true);
2461 return IRQ_NONE; /* Not our interrupt */
2464 if (eicr & IXGBE_EICR_LSC)
2465 ixgbe_check_lsc(adapter);
2467 switch (hw->mac.type) {
2468 case ixgbe_mac_82599EB:
2469 ixgbe_check_sfp_event(adapter, eicr);
2471 case ixgbe_mac_X540:
2472 if (eicr & IXGBE_EICR_ECC)
2473 e_info(link, "Received unrecoverable ECC err, please "
2475 ixgbe_check_overtemp_event(adapter, eicr);
2481 ixgbe_check_fan_failure(adapter, eicr);
2483 /* would disable interrupts here but EIAM disabled it */
2484 napi_schedule(&q_vector->napi);
2487 * re-enable link(maybe) and non-queue interrupts, no flush.
2488 * ixgbe_poll will re-enable the queue interrupts
2490 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2491 ixgbe_irq_enable(adapter, false, false);
2497 * ixgbe_request_irq - initialize interrupts
2498 * @adapter: board private structure
2500 * Attempts to configure interrupts using the best available
2501 * capabilities of the hardware and kernel.
2503 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2505 struct net_device *netdev = adapter->netdev;
2508 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2509 err = ixgbe_request_msix_irqs(adapter);
2510 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2511 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2512 netdev->name, adapter);
2514 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2515 netdev->name, adapter);
2518 e_err(probe, "request_irq failed, Error %d\n", err);
2523 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2525 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2528 q_vectors = adapter->num_msix_vectors;
2530 free_irq(adapter->msix_entries[i].vector, adapter);
2533 for (; i >= 0; i--) {
2534 /* free only the irqs that were actually requested */
2535 if (!adapter->q_vector[i]->rx.ring &&
2536 !adapter->q_vector[i]->tx.ring)
2539 /* clear the affinity_mask in the IRQ descriptor */
2540 irq_set_affinity_hint(adapter->msix_entries[i].vector,
2543 free_irq(adapter->msix_entries[i].vector,
2544 adapter->q_vector[i]);
2547 free_irq(adapter->pdev->irq, adapter);
2552 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2553 * @adapter: board private structure
2555 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2557 switch (adapter->hw.mac.type) {
2558 case ixgbe_mac_82598EB:
2559 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2561 case ixgbe_mac_82599EB:
2562 case ixgbe_mac_X540:
2563 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2564 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2565 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2570 IXGBE_WRITE_FLUSH(&adapter->hw);
2571 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2573 for (i = 0; i < adapter->num_msix_vectors; i++)
2574 synchronize_irq(adapter->msix_entries[i].vector);
2576 synchronize_irq(adapter->pdev->irq);
2581 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2584 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2586 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2589 if (adapter->rx_itr_setting == 1)
2590 q_vector->itr = IXGBE_20K_ITR;
2592 q_vector->itr = adapter->rx_itr_setting;
2594 ixgbe_write_eitr(q_vector);
2596 ixgbe_set_ivar(adapter, 0, 0, 0);
2597 ixgbe_set_ivar(adapter, 1, 0, 0);
2599 e_info(hw, "Legacy interrupt IVAR setup done\n");
2603 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2604 * @adapter: board private structure
2605 * @ring: structure containing ring specific data
2607 * Configure the Tx descriptor ring after a reset.
2609 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2610 struct ixgbe_ring *ring)
2612 struct ixgbe_hw *hw = &adapter->hw;
2613 u64 tdba = ring->dma;
2615 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2616 u8 reg_idx = ring->reg_idx;
2618 /* disable queue to avoid issues while updating state */
2619 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2620 IXGBE_WRITE_FLUSH(hw);
2622 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2623 (tdba & DMA_BIT_MASK(32)));
2624 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2625 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2626 ring->count * sizeof(union ixgbe_adv_tx_desc));
2627 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2628 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2629 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2632 * set WTHRESH to encourage burst writeback, it should not be set
2633 * higher than 1 when ITR is 0 as it could cause false TX hangs
2635 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2636 * to or less than the number of on chip descriptors, which is
2639 if (!ring->q_vector || (ring->q_vector->itr < 8))
2640 txdctl |= (1 << 16); /* WTHRESH = 1 */
2642 txdctl |= (8 << 16); /* WTHRESH = 8 */
2645 * Setting PTHRESH to 32 both improves performance
2646 * and avoids a TX hang with DFP enabled
2648 txdctl |= (1 << 8) | /* HTHRESH = 1 */
2649 32; /* PTHRESH = 32 */
2651 /* reinitialize flowdirector state */
2652 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2653 adapter->atr_sample_rate) {
2654 ring->atr_sample_rate = adapter->atr_sample_rate;
2655 ring->atr_count = 0;
2656 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2658 ring->atr_sample_rate = 0;
2661 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2664 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2666 netdev_tx_reset_queue(txring_txq(ring));
2668 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2669 if (hw->mac.type == ixgbe_mac_82598EB &&
2670 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2673 /* poll to verify queue is enabled */
2675 usleep_range(1000, 2000);
2676 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2677 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2679 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2682 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2684 struct ixgbe_hw *hw = &adapter->hw;
2687 u8 tcs = netdev_get_num_tc(adapter->netdev);
2689 if (hw->mac.type == ixgbe_mac_82598EB)
2692 /* disable the arbiter while setting MTQC */
2693 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2694 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2695 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2697 /* set transmit pool layout */
2698 switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2699 case (IXGBE_FLAG_SRIOV_ENABLED):
2700 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2701 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2705 reg = IXGBE_MTQC_64Q_1PB;
2707 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2709 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2711 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2713 /* Enable Security TX Buffer IFG for multiple pb */
2715 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2716 reg |= IXGBE_SECTX_DCB;
2717 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2722 /* re-enable the arbiter */
2723 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2724 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2728 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2729 * @adapter: board private structure
2731 * Configure the Tx unit of the MAC after a reset.
2733 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2735 struct ixgbe_hw *hw = &adapter->hw;
2739 ixgbe_setup_mtqc(adapter);
2741 if (hw->mac.type != ixgbe_mac_82598EB) {
2742 /* DMATXCTL.EN must be before Tx queues are enabled */
2743 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2744 dmatxctl |= IXGBE_DMATXCTL_TE;
2745 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2748 /* Setup the HW Tx Head and Tail descriptor pointers */
2749 for (i = 0; i < adapter->num_tx_queues; i++)
2750 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2753 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2755 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2756 struct ixgbe_ring *rx_ring)
2759 u8 reg_idx = rx_ring->reg_idx;
2761 switch (adapter->hw.mac.type) {
2762 case ixgbe_mac_82598EB: {
2763 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2764 const int mask = feature[RING_F_RSS].mask;
2765 reg_idx = reg_idx & mask;
2768 case ixgbe_mac_82599EB:
2769 case ixgbe_mac_X540:
2774 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2776 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2777 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2778 if (adapter->num_vfs)
2779 srrctl |= IXGBE_SRRCTL_DROP_EN;
2781 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2782 IXGBE_SRRCTL_BSIZEHDR_MASK;
2784 #if PAGE_SIZE > IXGBE_MAX_RXBUFFER
2785 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2787 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2789 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2791 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2794 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2796 struct ixgbe_hw *hw = &adapter->hw;
2797 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2798 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2799 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2800 u32 mrqc = 0, reta = 0;
2803 u8 tcs = netdev_get_num_tc(adapter->netdev);
2804 int maxq = adapter->ring_feature[RING_F_RSS].indices;
2807 maxq = min(maxq, adapter->num_tx_queues / tcs);
2809 /* Fill out hash function seeds */
2810 for (i = 0; i < 10; i++)
2811 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2813 /* Fill out redirection table */
2814 for (i = 0, j = 0; i < 128; i++, j++) {
2817 /* reta = 4-byte sliding window of
2818 * 0x00..(indices-1)(indices-1)00..etc. */
2819 reta = (reta << 8) | (j * 0x11);
2821 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2824 /* Disable indicating checksum in descriptor, enables RSS hash */
2825 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2826 rxcsum |= IXGBE_RXCSUM_PCSD;
2827 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2829 if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2830 (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2831 mrqc = IXGBE_MRQC_RSSEN;
2833 int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2834 | IXGBE_FLAG_SRIOV_ENABLED);
2837 case (IXGBE_FLAG_RSS_ENABLED):
2839 mrqc = IXGBE_MRQC_RSSEN;
2841 mrqc = IXGBE_MRQC_RTRSS4TCEN;
2843 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2845 case (IXGBE_FLAG_SRIOV_ENABLED):
2846 mrqc = IXGBE_MRQC_VMDQEN;
2853 /* Perform hash on these packet types */
2854 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2855 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2856 | IXGBE_MRQC_RSS_FIELD_IPV6
2857 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2859 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2863 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2864 * @adapter: address of board private structure
2865 * @index: index of ring to set
2867 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2868 struct ixgbe_ring *ring)
2870 struct ixgbe_hw *hw = &adapter->hw;
2872 u8 reg_idx = ring->reg_idx;
2874 if (!ring_is_rsc_enabled(ring))
2877 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2878 rscctrl |= IXGBE_RSCCTL_RSCEN;
2880 * we must limit the number of descriptors so that the
2881 * total size of max desc * buf_len is not greater
2884 #if (PAGE_SIZE <= 8192)
2885 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2886 #elif (PAGE_SIZE <= 16384)
2887 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2889 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2891 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2895 * ixgbe_set_uta - Set unicast filter table address
2896 * @adapter: board private structure
2898 * The unicast table address is a register array of 32-bit registers.
2899 * The table is meant to be used in a way similar to how the MTA is used
2900 * however due to certain limitations in the hardware it is necessary to
2901 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2902 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2904 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2906 struct ixgbe_hw *hw = &adapter->hw;
2909 /* The UTA table only exists on 82599 hardware and newer */
2910 if (hw->mac.type < ixgbe_mac_82599EB)
2913 /* we only need to do this if VMDq is enabled */
2914 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2917 for (i = 0; i < 128; i++)
2918 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2921 #define IXGBE_MAX_RX_DESC_POLL 10
2922 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2923 struct ixgbe_ring *ring)
2925 struct ixgbe_hw *hw = &adapter->hw;
2926 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2928 u8 reg_idx = ring->reg_idx;
2930 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2931 if (hw->mac.type == ixgbe_mac_82598EB &&
2932 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2936 usleep_range(1000, 2000);
2937 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2938 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2941 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2942 "the polling period\n", reg_idx);
2946 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
2947 struct ixgbe_ring *ring)
2949 struct ixgbe_hw *hw = &adapter->hw;
2950 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2952 u8 reg_idx = ring->reg_idx;
2954 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2955 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
2957 /* write value back with RXDCTL.ENABLE bit cleared */
2958 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2960 if (hw->mac.type == ixgbe_mac_82598EB &&
2961 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2964 /* the hardware may take up to 100us to really disable the rx queue */
2967 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2968 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
2971 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
2972 "the polling period\n", reg_idx);
2976 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2977 struct ixgbe_ring *ring)
2979 struct ixgbe_hw *hw = &adapter->hw;
2980 u64 rdba = ring->dma;
2982 u8 reg_idx = ring->reg_idx;
2984 /* disable queue to avoid issues while updating state */
2985 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2986 ixgbe_disable_rx_queue(adapter, ring);
2988 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2989 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2990 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2991 ring->count * sizeof(union ixgbe_adv_rx_desc));
2992 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2993 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2994 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2996 ixgbe_configure_srrctl(adapter, ring);
2997 ixgbe_configure_rscctl(adapter, ring);
2999 /* If operating in IOV mode set RLPML for X540 */
3000 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3001 hw->mac.type == ixgbe_mac_X540) {
3002 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3003 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3004 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3007 if (hw->mac.type == ixgbe_mac_82598EB) {
3009 * enable cache line friendly hardware writes:
3010 * PTHRESH=32 descriptors (half the internal cache),
3011 * this also removes ugly rx_no_buffer_count increment
3012 * HTHRESH=4 descriptors (to minimize latency on fetch)
3013 * WTHRESH=8 burst writeback up to two cache lines
3015 rxdctl &= ~0x3FFFFF;
3019 /* enable receive descriptor ring */
3020 rxdctl |= IXGBE_RXDCTL_ENABLE;
3021 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3023 ixgbe_rx_desc_queue_enable(adapter, ring);
3024 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3027 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3029 struct ixgbe_hw *hw = &adapter->hw;
3032 /* PSRTYPE must be initialized in non 82598 adapters */
3033 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3034 IXGBE_PSRTYPE_UDPHDR |
3035 IXGBE_PSRTYPE_IPV4HDR |
3036 IXGBE_PSRTYPE_L2HDR |
3037 IXGBE_PSRTYPE_IPV6HDR;
3039 if (hw->mac.type == ixgbe_mac_82598EB)
3042 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3043 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3045 for (p = 0; p < adapter->num_rx_pools; p++)
3046 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3050 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3052 struct ixgbe_hw *hw = &adapter->hw;
3055 u32 reg_offset, vf_shift;
3059 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3062 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3063 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3064 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3065 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3067 vf_shift = adapter->num_vfs % 32;
3068 reg_offset = (adapter->num_vfs >= 32) ? 1 : 0;
3070 /* Enable only the PF's pool for Tx/Rx */
3071 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3072 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3073 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3074 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3075 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3077 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3078 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3081 * Set up VF register offsets for selected VT Mode,
3082 * i.e. 32 or 64 VFs for SR-IOV
3084 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3085 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3086 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3087 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3089 /* enable Tx loopback for VF/PF communication */
3090 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3091 /* Enable MAC Anti-Spoofing */
3092 hw->mac.ops.set_mac_anti_spoofing(hw,
3093 (adapter->num_vfs != 0),
3095 /* For VFs that have spoof checking turned off */
3096 for (i = 0; i < adapter->num_vfs; i++) {
3097 if (!adapter->vfinfo[i].spoofchk_enabled)
3098 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3102 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3104 struct ixgbe_hw *hw = &adapter->hw;
3105 struct net_device *netdev = adapter->netdev;
3106 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3107 struct ixgbe_ring *rx_ring;
3112 /* adjust max frame to be able to do baby jumbo for FCoE */
3113 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3114 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3115 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3117 #endif /* IXGBE_FCOE */
3118 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3119 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3120 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3121 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3123 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3126 /* MHADD will allow an extra 4 bytes past for vlan tagged frames */
3127 max_frame += VLAN_HLEN;
3129 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3130 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3131 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3132 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3135 * Setup the HW Rx Head and Tail Descriptor Pointers and
3136 * the Base and Length of the Rx Descriptor Ring
3138 for (i = 0; i < adapter->num_rx_queues; i++) {
3139 rx_ring = adapter->rx_ring[i];
3140 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3141 set_ring_rsc_enabled(rx_ring);
3143 clear_ring_rsc_enabled(rx_ring);
3145 if (netdev->features & NETIF_F_FCOE_MTU) {
3146 struct ixgbe_ring_feature *f;
3147 f = &adapter->ring_feature[RING_F_FCOE];
3148 if ((i >= f->mask) && (i < f->mask + f->indices))
3149 set_bit(__IXGBE_RX_FCOE_BUFSZ, &rx_ring->state);
3151 #endif /* IXGBE_FCOE */
3155 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3157 struct ixgbe_hw *hw = &adapter->hw;
3158 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3160 switch (hw->mac.type) {
3161 case ixgbe_mac_82598EB:
3163 * For VMDq support of different descriptor types or
3164 * buffer sizes through the use of multiple SRRCTL
3165 * registers, RDRXCTL.MVMEN must be set to 1
3167 * also, the manual doesn't mention it clearly but DCA hints
3168 * will only use queue 0's tags unless this bit is set. Side
3169 * effects of setting this bit are only that SRRCTL must be
3170 * fully programmed [0..15]
3172 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3174 case ixgbe_mac_82599EB:
3175 case ixgbe_mac_X540:
3176 /* Disable RSC for ACK packets */
3177 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3178 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3179 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3180 /* hardware requires some bits to be set by default */
3181 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3182 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3185 /* We should do nothing since we don't know this hardware */
3189 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3193 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3194 * @adapter: board private structure
3196 * Configure the Rx unit of the MAC after a reset.
3198 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3200 struct ixgbe_hw *hw = &adapter->hw;
3204 /* disable receives while setting up the descriptors */
3205 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3206 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3208 ixgbe_setup_psrtype(adapter);
3209 ixgbe_setup_rdrxctl(adapter);
3211 /* Program registers for the distribution of queues */
3212 ixgbe_setup_mrqc(adapter);
3214 ixgbe_set_uta(adapter);
3216 /* set_rx_buffer_len must be called before ring initialization */
3217 ixgbe_set_rx_buffer_len(adapter);
3220 * Setup the HW Rx Head and Tail Descriptor Pointers and
3221 * the Base and Length of the Rx Descriptor Ring
3223 for (i = 0; i < adapter->num_rx_queues; i++)
3224 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3226 /* disable drop enable for 82598 parts */
3227 if (hw->mac.type == ixgbe_mac_82598EB)
3228 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3230 /* enable all receives */
3231 rxctrl |= IXGBE_RXCTRL_RXEN;
3232 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3235 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3237 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3238 struct ixgbe_hw *hw = &adapter->hw;
3239 int pool_ndx = adapter->num_vfs;
3241 /* add VID to filter table */
3242 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3243 set_bit(vid, adapter->active_vlans);
3248 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3250 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3251 struct ixgbe_hw *hw = &adapter->hw;
3252 int pool_ndx = adapter->num_vfs;
3254 /* remove VID from filter table */
3255 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3256 clear_bit(vid, adapter->active_vlans);
3262 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3263 * @adapter: driver data
3265 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3267 struct ixgbe_hw *hw = &adapter->hw;
3270 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3271 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3272 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3276 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3277 * @adapter: driver data
3279 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3281 struct ixgbe_hw *hw = &adapter->hw;
3284 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3285 vlnctrl |= IXGBE_VLNCTRL_VFE;
3286 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3287 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3291 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3292 * @adapter: driver data
3294 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3296 struct ixgbe_hw *hw = &adapter->hw;
3300 switch (hw->mac.type) {
3301 case ixgbe_mac_82598EB:
3302 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3303 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3304 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3306 case ixgbe_mac_82599EB:
3307 case ixgbe_mac_X540:
3308 for (i = 0; i < adapter->num_rx_queues; i++) {
3309 j = adapter->rx_ring[i]->reg_idx;
3310 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3311 vlnctrl &= ~IXGBE_RXDCTL_VME;
3312 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3321 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3322 * @adapter: driver data
3324 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3326 struct ixgbe_hw *hw = &adapter->hw;
3330 switch (hw->mac.type) {
3331 case ixgbe_mac_82598EB:
3332 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3333 vlnctrl |= IXGBE_VLNCTRL_VME;
3334 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3336 case ixgbe_mac_82599EB:
3337 case ixgbe_mac_X540:
3338 for (i = 0; i < adapter->num_rx_queues; i++) {
3339 j = adapter->rx_ring[i]->reg_idx;
3340 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3341 vlnctrl |= IXGBE_RXDCTL_VME;
3342 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3350 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3354 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3356 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3357 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3361 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3362 * @netdev: network interface device structure
3364 * Writes unicast address list to the RAR table.
3365 * Returns: -ENOMEM on failure/insufficient address space
3366 * 0 on no addresses written
3367 * X on writing X addresses to the RAR table
3369 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3371 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3372 struct ixgbe_hw *hw = &adapter->hw;
3373 unsigned int vfn = adapter->num_vfs;
3374 unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3377 /* return ENOMEM indicating insufficient memory for addresses */
3378 if (netdev_uc_count(netdev) > rar_entries)
3381 if (!netdev_uc_empty(netdev) && rar_entries) {
3382 struct netdev_hw_addr *ha;
3383 /* return error if we do not support writing to RAR table */
3384 if (!hw->mac.ops.set_rar)
3387 netdev_for_each_uc_addr(ha, netdev) {
3390 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3395 /* write the addresses in reverse order to avoid write combining */
3396 for (; rar_entries > 0 ; rar_entries--)
3397 hw->mac.ops.clear_rar(hw, rar_entries);
3403 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3404 * @netdev: network interface device structure
3406 * The set_rx_method entry point is called whenever the unicast/multicast
3407 * address list or the network interface flags are updated. This routine is
3408 * responsible for configuring the hardware for proper unicast, multicast and
3411 void ixgbe_set_rx_mode(struct net_device *netdev)
3413 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3414 struct ixgbe_hw *hw = &adapter->hw;
3415 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3418 /* Check for Promiscuous and All Multicast modes */
3420 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3422 /* set all bits that we expect to always be set */
3423 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3424 fctrl |= IXGBE_FCTRL_BAM;
3425 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3426 fctrl |= IXGBE_FCTRL_PMCF;
3428 /* clear the bits we are changing the status of */
3429 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3431 if (netdev->flags & IFF_PROMISC) {
3432 hw->addr_ctrl.user_set_promisc = true;
3433 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3434 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3435 /* don't hardware filter vlans in promisc mode */
3436 ixgbe_vlan_filter_disable(adapter);
3438 if (netdev->flags & IFF_ALLMULTI) {
3439 fctrl |= IXGBE_FCTRL_MPE;
3440 vmolr |= IXGBE_VMOLR_MPE;
3443 * Write addresses to the MTA, if the attempt fails
3444 * then we should just turn on promiscuous mode so
3445 * that we can at least receive multicast traffic
3447 hw->mac.ops.update_mc_addr_list(hw, netdev);
3448 vmolr |= IXGBE_VMOLR_ROMPE;
3450 ixgbe_vlan_filter_enable(adapter);
3451 hw->addr_ctrl.user_set_promisc = false;
3453 * Write addresses to available RAR registers, if there is not
3454 * sufficient space to store all the addresses then enable
3455 * unicast promiscuous mode
3457 count = ixgbe_write_uc_addr_list(netdev);
3459 fctrl |= IXGBE_FCTRL_UPE;
3460 vmolr |= IXGBE_VMOLR_ROPE;
3464 if (adapter->num_vfs) {
3465 ixgbe_restore_vf_multicasts(adapter);
3466 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3467 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3469 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3472 /* This is useful for sniffing bad packets. */
3473 if (adapter->netdev->features & NETIF_F_RXALL) {
3474 /* UPE and MPE will be handled by normal PROMISC logic
3475 * in e1000e_set_rx_mode */
3476 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3477 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3478 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3480 fctrl &= ~(IXGBE_FCTRL_DPF);
3481 /* NOTE: VLAN filtering is disabled by setting PROMISC */
3484 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3486 if (netdev->features & NETIF_F_HW_VLAN_RX)
3487 ixgbe_vlan_strip_enable(adapter);
3489 ixgbe_vlan_strip_disable(adapter);
3492 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3495 struct ixgbe_q_vector *q_vector;
3496 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3498 /* legacy and MSI only use one vector */
3499 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3502 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3503 q_vector = adapter->q_vector[q_idx];
3504 napi_enable(&q_vector->napi);
3508 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3511 struct ixgbe_q_vector *q_vector;
3512 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3514 /* legacy and MSI only use one vector */
3515 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3518 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3519 q_vector = adapter->q_vector[q_idx];
3520 napi_disable(&q_vector->napi);
3524 #ifdef CONFIG_IXGBE_DCB
3526 * ixgbe_configure_dcb - Configure DCB hardware
3527 * @adapter: ixgbe adapter struct
3529 * This is called by the driver on open to configure the DCB hardware.
3530 * This is also called by the gennetlink interface when reconfiguring
3533 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3535 struct ixgbe_hw *hw = &adapter->hw;
3536 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3538 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3539 if (hw->mac.type == ixgbe_mac_82598EB)
3540 netif_set_gso_max_size(adapter->netdev, 65536);
3544 if (hw->mac.type == ixgbe_mac_82598EB)
3545 netif_set_gso_max_size(adapter->netdev, 32768);
3548 /* Enable VLAN tag insert/strip */
3549 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3551 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3554 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3555 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3558 /* reconfigure the hardware */
3559 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3560 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3562 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3564 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3565 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3566 ixgbe_dcb_hw_ets(&adapter->hw,
3567 adapter->ixgbe_ieee_ets,
3569 ixgbe_dcb_hw_pfc_config(&adapter->hw,
3570 adapter->ixgbe_ieee_pfc->pfc_en,
3571 adapter->ixgbe_ieee_ets->prio_tc);
3574 /* Enable RSS Hash per TC */
3575 if (hw->mac.type != ixgbe_mac_82598EB) {
3579 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3581 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3586 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3588 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3593 /* Additional bittime to account for IXGBE framing */
3594 #define IXGBE_ETH_FRAMING 20
3597 * ixgbe_hpbthresh - calculate high water mark for flow control
3599 * @adapter: board private structure to calculate for
3600 * @pb - packet buffer to calculate
3602 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3604 struct ixgbe_hw *hw = &adapter->hw;
3605 struct net_device *dev = adapter->netdev;
3606 int link, tc, kb, marker;
3609 /* Calculate max LAN frame size */
3610 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3613 /* FCoE traffic class uses FCOE jumbo frames */
3614 if (dev->features & NETIF_F_FCOE_MTU) {
3617 #ifdef CONFIG_IXGBE_DCB
3618 fcoe_pb = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
3621 if (fcoe_pb == pb && tc < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3622 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3626 /* Calculate delay value for device */
3627 switch (hw->mac.type) {
3628 case ixgbe_mac_X540:
3629 dv_id = IXGBE_DV_X540(link, tc);
3632 dv_id = IXGBE_DV(link, tc);
3636 /* Loopback switch introduces additional latency */
3637 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3638 dv_id += IXGBE_B2BT(tc);
3640 /* Delay value is calculated in bit times convert to KB */
3641 kb = IXGBE_BT2KB(dv_id);
3642 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3644 marker = rx_pba - kb;
3646 /* It is possible that the packet buffer is not large enough
3647 * to provide required headroom. In this case throw an error
3648 * to user and a do the best we can.
3651 e_warn(drv, "Packet Buffer(%i) can not provide enough"
3652 "headroom to support flow control."
3653 "Decrease MTU or number of traffic classes\n", pb);
3661 * ixgbe_lpbthresh - calculate low water mark for for flow control
3663 * @adapter: board private structure to calculate for
3664 * @pb - packet buffer to calculate
3666 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3668 struct ixgbe_hw *hw = &adapter->hw;
3669 struct net_device *dev = adapter->netdev;
3673 /* Calculate max LAN frame size */
3674 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3676 /* Calculate delay value for device */
3677 switch (hw->mac.type) {
3678 case ixgbe_mac_X540:
3679 dv_id = IXGBE_LOW_DV_X540(tc);
3682 dv_id = IXGBE_LOW_DV(tc);
3686 /* Delay value is calculated in bit times convert to KB */
3687 return IXGBE_BT2KB(dv_id);
3691 * ixgbe_pbthresh_setup - calculate and setup high low water marks
3693 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3695 struct ixgbe_hw *hw = &adapter->hw;
3696 int num_tc = netdev_get_num_tc(adapter->netdev);
3702 hw->fc.low_water = ixgbe_lpbthresh(adapter);
3704 for (i = 0; i < num_tc; i++) {
3705 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3707 /* Low water marks must not be larger than high water marks */
3708 if (hw->fc.low_water > hw->fc.high_water[i])
3709 hw->fc.low_water = 0;
3713 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3715 struct ixgbe_hw *hw = &adapter->hw;
3717 u8 tc = netdev_get_num_tc(adapter->netdev);
3719 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3720 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3721 hdrm = 32 << adapter->fdir_pballoc;
3725 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3726 ixgbe_pbthresh_setup(adapter);
3729 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3731 struct ixgbe_hw *hw = &adapter->hw;
3732 struct hlist_node *node, *node2;
3733 struct ixgbe_fdir_filter *filter;
3735 spin_lock(&adapter->fdir_perfect_lock);
3737 if (!hlist_empty(&adapter->fdir_filter_list))
3738 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3740 hlist_for_each_entry_safe(filter, node, node2,
3741 &adapter->fdir_filter_list, fdir_node) {
3742 ixgbe_fdir_write_perfect_filter_82599(hw,
3745 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3746 IXGBE_FDIR_DROP_QUEUE :
3747 adapter->rx_ring[filter->action]->reg_idx);
3750 spin_unlock(&adapter->fdir_perfect_lock);
3753 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3755 struct ixgbe_hw *hw = &adapter->hw;
3757 ixgbe_configure_pb(adapter);
3758 #ifdef CONFIG_IXGBE_DCB
3759 ixgbe_configure_dcb(adapter);
3762 ixgbe_set_rx_mode(adapter->netdev);
3763 ixgbe_restore_vlan(adapter);
3766 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3767 ixgbe_configure_fcoe(adapter);
3769 #endif /* IXGBE_FCOE */
3771 switch (hw->mac.type) {
3772 case ixgbe_mac_82599EB:
3773 case ixgbe_mac_X540:
3774 hw->mac.ops.disable_rx_buff(hw);
3780 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3781 ixgbe_init_fdir_signature_82599(&adapter->hw,
3782 adapter->fdir_pballoc);
3783 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3784 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3785 adapter->fdir_pballoc);
3786 ixgbe_fdir_filter_restore(adapter);
3789 switch (hw->mac.type) {
3790 case ixgbe_mac_82599EB:
3791 case ixgbe_mac_X540:
3792 hw->mac.ops.enable_rx_buff(hw);
3798 ixgbe_configure_virtualization(adapter);
3800 ixgbe_configure_tx(adapter);
3801 ixgbe_configure_rx(adapter);
3804 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3806 switch (hw->phy.type) {
3807 case ixgbe_phy_sfp_avago:
3808 case ixgbe_phy_sfp_ftl:
3809 case ixgbe_phy_sfp_intel:
3810 case ixgbe_phy_sfp_unknown:
3811 case ixgbe_phy_sfp_passive_tyco:
3812 case ixgbe_phy_sfp_passive_unknown:
3813 case ixgbe_phy_sfp_active_unknown:
3814 case ixgbe_phy_sfp_ftl_active:
3817 if (hw->mac.type == ixgbe_mac_82598EB)
3825 * ixgbe_sfp_link_config - set up SFP+ link
3826 * @adapter: pointer to private adapter struct
3828 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3831 * We are assuming the worst case scenario here, and that
3832 * is that an SFP was inserted/removed after the reset
3833 * but before SFP detection was enabled. As such the best
3834 * solution is to just start searching as soon as we start
3836 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3837 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3839 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3843 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3844 * @hw: pointer to private hardware struct
3846 * Returns 0 on success, negative on failure
3848 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3851 bool negotiation, link_up = false;
3852 u32 ret = IXGBE_ERR_LINK_SETUP;
3854 if (hw->mac.ops.check_link)
3855 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3860 autoneg = hw->phy.autoneg_advertised;
3861 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3862 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3867 if (hw->mac.ops.setup_link)
3868 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3873 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3875 struct ixgbe_hw *hw = &adapter->hw;
3878 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3879 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3881 gpie |= IXGBE_GPIE_EIAME;
3883 * use EIAM to auto-mask when MSI-X interrupt is asserted
3884 * this saves a register write for every interrupt
3886 switch (hw->mac.type) {
3887 case ixgbe_mac_82598EB:
3888 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3890 case ixgbe_mac_82599EB:
3891 case ixgbe_mac_X540:
3893 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3894 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3898 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3899 * specifically only auto mask tx and rx interrupts */
3900 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3903 /* XXX: to interrupt immediately for EICS writes, enable this */
3904 /* gpie |= IXGBE_GPIE_EIMEN; */
3906 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3907 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3908 gpie |= IXGBE_GPIE_VTMODE_64;
3911 /* Enable Thermal over heat sensor interrupt */
3912 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3913 switch (adapter->hw.mac.type) {
3914 case ixgbe_mac_82599EB:
3915 gpie |= IXGBE_SDP0_GPIEN;
3917 case ixgbe_mac_X540:
3918 gpie |= IXGBE_EIMS_TS;
3925 /* Enable fan failure interrupt */
3926 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3927 gpie |= IXGBE_SDP1_GPIEN;
3929 if (hw->mac.type == ixgbe_mac_82599EB) {
3930 gpie |= IXGBE_SDP1_GPIEN;
3931 gpie |= IXGBE_SDP2_GPIEN;
3934 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3937 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
3939 struct ixgbe_hw *hw = &adapter->hw;
3943 ixgbe_get_hw_control(adapter);
3944 ixgbe_setup_gpie(adapter);
3946 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3947 ixgbe_configure_msix(adapter);
3949 ixgbe_configure_msi_and_legacy(adapter);
3951 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3952 if (hw->mac.ops.enable_tx_laser &&
3953 ((hw->phy.multispeed_fiber) ||
3954 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3955 (hw->mac.type == ixgbe_mac_82599EB))))
3956 hw->mac.ops.enable_tx_laser(hw);
3958 clear_bit(__IXGBE_DOWN, &adapter->state);
3959 ixgbe_napi_enable_all(adapter);
3961 if (ixgbe_is_sfp(hw)) {
3962 ixgbe_sfp_link_config(adapter);
3964 err = ixgbe_non_sfp_link_config(hw);
3966 e_err(probe, "link_config FAILED %d\n", err);
3969 /* clear any pending interrupts, may auto mask */
3970 IXGBE_READ_REG(hw, IXGBE_EICR);
3971 ixgbe_irq_enable(adapter, true, true);
3974 * If this adapter has a fan, check to see if we had a failure
3975 * before we enabled the interrupt.
3977 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3978 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3979 if (esdp & IXGBE_ESDP_SDP1)
3980 e_crit(drv, "Fan has stopped, replace the adapter\n");
3983 /* enable transmits */
3984 netif_tx_start_all_queues(adapter->netdev);
3986 /* bring the link up in the watchdog, this could race with our first
3987 * link up interrupt but shouldn't be a problem */
3988 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3989 adapter->link_check_timeout = jiffies;
3990 mod_timer(&adapter->service_timer, jiffies);
3992 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3993 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3994 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3995 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3998 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4000 WARN_ON(in_interrupt());
4001 /* put off any impending NetWatchDogTimeout */
4002 adapter->netdev->trans_start = jiffies;
4004 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4005 usleep_range(1000, 2000);
4006 ixgbe_down(adapter);
4008 * If SR-IOV enabled then wait a bit before bringing the adapter
4009 * back up to give the VFs time to respond to the reset. The
4010 * two second wait is based upon the watchdog timer cycle in
4013 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4016 clear_bit(__IXGBE_RESETTING, &adapter->state);
4019 void ixgbe_up(struct ixgbe_adapter *adapter)
4021 /* hardware has been reset, we need to reload some things */
4022 ixgbe_configure(adapter);
4024 ixgbe_up_complete(adapter);
4027 void ixgbe_reset(struct ixgbe_adapter *adapter)
4029 struct ixgbe_hw *hw = &adapter->hw;
4032 /* lock SFP init bit to prevent race conditions with the watchdog */
4033 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4034 usleep_range(1000, 2000);
4036 /* clear all SFP and link config related flags while holding SFP_INIT */
4037 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4038 IXGBE_FLAG2_SFP_NEEDS_RESET);
4039 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4041 err = hw->mac.ops.init_hw(hw);
4044 case IXGBE_ERR_SFP_NOT_PRESENT:
4045 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4047 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4048 e_dev_err("master disable timed out\n");
4050 case IXGBE_ERR_EEPROM_VERSION:
4051 /* We are running on a pre-production device, log a warning */
4052 e_dev_warn("This device is a pre-production adapter/LOM. "
4053 "Please be aware there may be issues associated with "
4054 "your hardware. If you are experiencing problems "
4055 "please contact your Intel or hardware "
4056 "representative who provided you with this "
4060 e_dev_err("Hardware Error: %d\n", err);
4063 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4065 /* reprogram the RAR[0] in case user changed it. */
4066 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
4071 * ixgbe_init_rx_page_offset - initialize page offset values for Rx buffers
4072 * @rx_ring: ring to setup
4074 * On many IA platforms the L1 cache has a critical stride of 4K, this
4075 * results in each receive buffer starting in the same cache set. To help
4076 * reduce the pressure on this cache set we can interleave the offsets so
4077 * that only every other buffer will be in the same cache set.
4079 static void ixgbe_init_rx_page_offset(struct ixgbe_ring *rx_ring)
4081 struct ixgbe_rx_buffer *rx_buffer = rx_ring->rx_buffer_info;
4084 for (i = 0; i < rx_ring->count; i += 2) {
4085 rx_buffer[0].page_offset = 0;
4086 rx_buffer[1].page_offset = ixgbe_rx_bufsz(rx_ring);
4087 rx_buffer = &rx_buffer[2];
4092 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4093 * @rx_ring: ring to free buffers from
4095 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4097 struct device *dev = rx_ring->dev;
4101 /* ring already cleared, nothing to do */
4102 if (!rx_ring->rx_buffer_info)
4105 /* Free all the Rx ring sk_buffs */
4106 for (i = 0; i < rx_ring->count; i++) {
4107 struct ixgbe_rx_buffer *rx_buffer;
4109 rx_buffer = &rx_ring->rx_buffer_info[i];
4110 if (rx_buffer->skb) {
4111 struct sk_buff *skb = rx_buffer->skb;
4112 if (IXGBE_CB(skb)->page_released) {
4115 ixgbe_rx_bufsz(rx_ring),
4117 IXGBE_CB(skb)->page_released = false;
4121 rx_buffer->skb = NULL;
4123 dma_unmap_page(dev, rx_buffer->dma,
4124 ixgbe_rx_pg_size(rx_ring),
4127 if (rx_buffer->page)
4128 put_page(rx_buffer->page);
4129 rx_buffer->page = NULL;
4132 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4133 memset(rx_ring->rx_buffer_info, 0, size);
4135 ixgbe_init_rx_page_offset(rx_ring);
4137 /* Zero out the descriptor ring */
4138 memset(rx_ring->desc, 0, rx_ring->size);
4140 rx_ring->next_to_alloc = 0;
4141 rx_ring->next_to_clean = 0;
4142 rx_ring->next_to_use = 0;
4146 * ixgbe_clean_tx_ring - Free Tx Buffers
4147 * @tx_ring: ring to be cleaned
4149 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4151 struct ixgbe_tx_buffer *tx_buffer_info;
4155 /* ring already cleared, nothing to do */
4156 if (!tx_ring->tx_buffer_info)
4159 /* Free all the Tx ring sk_buffs */
4160 for (i = 0; i < tx_ring->count; i++) {
4161 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4162 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4165 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4166 memset(tx_ring->tx_buffer_info, 0, size);
4168 /* Zero out the descriptor ring */
4169 memset(tx_ring->desc, 0, tx_ring->size);
4171 tx_ring->next_to_use = 0;
4172 tx_ring->next_to_clean = 0;
4176 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4177 * @adapter: board private structure
4179 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4183 for (i = 0; i < adapter->num_rx_queues; i++)
4184 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4188 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4189 * @adapter: board private structure
4191 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4195 for (i = 0; i < adapter->num_tx_queues; i++)
4196 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4199 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4201 struct hlist_node *node, *node2;
4202 struct ixgbe_fdir_filter *filter;
4204 spin_lock(&adapter->fdir_perfect_lock);
4206 hlist_for_each_entry_safe(filter, node, node2,
4207 &adapter->fdir_filter_list, fdir_node) {
4208 hlist_del(&filter->fdir_node);
4211 adapter->fdir_filter_count = 0;
4213 spin_unlock(&adapter->fdir_perfect_lock);
4216 void ixgbe_down(struct ixgbe_adapter *adapter)
4218 struct net_device *netdev = adapter->netdev;
4219 struct ixgbe_hw *hw = &adapter->hw;
4223 /* signal that we are down to the interrupt handler */
4224 set_bit(__IXGBE_DOWN, &adapter->state);
4226 /* disable receives */
4227 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4228 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4230 /* disable all enabled rx queues */
4231 for (i = 0; i < adapter->num_rx_queues; i++)
4232 /* this call also flushes the previous write */
4233 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4235 usleep_range(10000, 20000);
4237 netif_tx_stop_all_queues(netdev);
4239 /* call carrier off first to avoid false dev_watchdog timeouts */
4240 netif_carrier_off(netdev);
4241 netif_tx_disable(netdev);
4243 ixgbe_irq_disable(adapter);
4245 ixgbe_napi_disable_all(adapter);
4247 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4248 IXGBE_FLAG2_RESET_REQUESTED);
4249 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4251 del_timer_sync(&adapter->service_timer);
4253 if (adapter->num_vfs) {
4254 /* Clear EITR Select mapping */
4255 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4257 /* Mark all the VFs as inactive */
4258 for (i = 0 ; i < adapter->num_vfs; i++)
4259 adapter->vfinfo[i].clear_to_send = false;
4261 /* ping all the active vfs to let them know we are going down */
4262 ixgbe_ping_all_vfs(adapter);
4264 /* Disable all VFTE/VFRE TX/RX */
4265 ixgbe_disable_tx_rx(adapter);
4268 /* disable transmits in the hardware now that interrupts are off */
4269 for (i = 0; i < adapter->num_tx_queues; i++) {
4270 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4271 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4274 /* Disable the Tx DMA engine on 82599 and X540 */
4275 switch (hw->mac.type) {
4276 case ixgbe_mac_82599EB:
4277 case ixgbe_mac_X540:
4278 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4279 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4280 ~IXGBE_DMATXCTL_TE));
4286 if (!pci_channel_offline(adapter->pdev))
4287 ixgbe_reset(adapter);
4289 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4290 if (hw->mac.ops.disable_tx_laser &&
4291 ((hw->phy.multispeed_fiber) ||
4292 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4293 (hw->mac.type == ixgbe_mac_82599EB))))
4294 hw->mac.ops.disable_tx_laser(hw);
4296 ixgbe_clean_all_tx_rings(adapter);
4297 ixgbe_clean_all_rx_rings(adapter);
4299 #ifdef CONFIG_IXGBE_DCA
4300 /* since we reset the hardware DCA settings were cleared */
4301 ixgbe_setup_dca(adapter);
4306 * ixgbe_tx_timeout - Respond to a Tx Hang
4307 * @netdev: network interface device structure
4309 static void ixgbe_tx_timeout(struct net_device *netdev)
4311 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4313 /* Do the reset outside of interrupt context */
4314 ixgbe_tx_timeout_reset(adapter);
4318 * ixgbe_set_rss_queues: Allocate queues for RSS
4319 * @adapter: board private structure to initialize
4321 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
4322 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4325 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4328 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4330 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4332 adapter->num_rx_queues = f->indices;
4333 adapter->num_tx_queues = f->indices;
4341 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4342 * @adapter: board private structure to initialize
4344 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4345 * to the original CPU that initiated the Tx session. This runs in addition
4346 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4347 * Rx load across CPUs using RSS.
4350 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4353 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4355 f_fdir->indices = min_t(int, num_online_cpus(), f_fdir->indices);
4359 * Use RSS in addition to Flow Director to ensure the best
4360 * distribution of flows across cores, even when an FDIR flow
4363 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4364 (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4365 adapter->num_tx_queues = f_fdir->indices;
4366 adapter->num_rx_queues = f_fdir->indices;
4369 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4376 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4377 * @adapter: board private structure to initialize
4379 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4380 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4381 * rx queues out of the max number of rx queues, instead, it is used as the
4382 * index of the first rx queue used by FCoE.
4385 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4387 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4389 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4392 f->indices = min_t(int, num_online_cpus(), f->indices);
4394 adapter->num_rx_queues = 1;
4395 adapter->num_tx_queues = 1;
4397 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4398 e_info(probe, "FCoE enabled with RSS\n");
4399 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4400 ixgbe_set_fdir_queues(adapter);
4402 ixgbe_set_rss_queues(adapter);
4405 /* adding FCoE rx rings to the end */
4406 f->mask = adapter->num_rx_queues;
4407 adapter->num_rx_queues += f->indices;
4408 adapter->num_tx_queues += f->indices;
4412 #endif /* IXGBE_FCOE */
4414 /* Artificial max queue cap per traffic class in DCB mode */
4415 #define DCB_QUEUE_CAP 8
4417 #ifdef CONFIG_IXGBE_DCB
4418 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4420 int per_tc_q, q, i, offset = 0;
4421 struct net_device *dev = adapter->netdev;
4422 int tcs = netdev_get_num_tc(dev);
4427 /* Map queue offset and counts onto allocated tx queues */
4428 per_tc_q = min_t(unsigned int, dev->num_tx_queues / tcs, DCB_QUEUE_CAP);
4429 q = min_t(int, num_online_cpus(), per_tc_q);
4431 for (i = 0; i < tcs; i++) {
4432 netdev_set_tc_queue(dev, i, q, offset);
4436 adapter->num_tx_queues = q * tcs;
4437 adapter->num_rx_queues = q * tcs;
4440 /* FCoE enabled queues require special configuration indexed
4441 * by feature specific indices and mask. Here we map FCoE
4442 * indices onto the DCB queue pairs allowing FCoE to own
4443 * configuration later.
4445 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4446 u8 prio_tc[MAX_USER_PRIORITY] = {0};
4448 struct ixgbe_ring_feature *f =
4449 &adapter->ring_feature[RING_F_FCOE];
4451 ixgbe_dcb_unpack_map(&adapter->dcb_cfg, DCB_TX_CONFIG, prio_tc);
4452 tc = prio_tc[adapter->fcoe.up];
4453 f->indices = dev->tc_to_txq[tc].count;
4454 f->mask = dev->tc_to_txq[tc].offset;
4463 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4464 * @adapter: board private structure to initialize
4466 * IOV doesn't actually use anything, so just NAK the
4467 * request for now and let the other queue routines
4468 * figure out what to do.
4470 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4476 * ixgbe_set_num_queues: Allocate queues for device, feature dependent
4477 * @adapter: board private structure to initialize
4479 * This is the top level queue allocation routine. The order here is very
4480 * important, starting with the "most" number of features turned on at once,
4481 * and ending with the smallest set of features. This way large combinations
4482 * can be allocated if they're turned on, and smaller combinations are the
4483 * fallthrough conditions.
4486 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4488 /* Start with base case */
4489 adapter->num_rx_queues = 1;
4490 adapter->num_tx_queues = 1;
4491 adapter->num_rx_pools = adapter->num_rx_queues;
4492 adapter->num_rx_queues_per_pool = 1;
4494 if (ixgbe_set_sriov_queues(adapter))
4497 #ifdef CONFIG_IXGBE_DCB
4498 if (ixgbe_set_dcb_queues(adapter))
4503 if (ixgbe_set_fcoe_queues(adapter))
4506 #endif /* IXGBE_FCOE */
4507 if (ixgbe_set_fdir_queues(adapter))
4510 if (ixgbe_set_rss_queues(adapter))
4513 /* fallback to base case */
4514 adapter->num_rx_queues = 1;
4515 adapter->num_tx_queues = 1;
4518 if ((adapter->netdev->reg_state == NETREG_UNREGISTERED) ||
4519 (adapter->netdev->reg_state == NETREG_UNREGISTERING))
4522 /* Notify the stack of the (possibly) reduced queue counts. */
4523 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4524 return netif_set_real_num_rx_queues(adapter->netdev,
4525 adapter->num_rx_queues);
4528 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4531 int err, vector_threshold;
4533 /* We'll want at least 2 (vector_threshold):
4534 * 1) TxQ[0] + RxQ[0] handler
4535 * 2) Other (Link Status Change, etc.)
4537 vector_threshold = MIN_MSIX_COUNT;
4540 * The more we get, the more we will assign to Tx/Rx Cleanup
4541 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4542 * Right now, we simply care about how many we'll get; we'll
4543 * set them up later while requesting irq's.
4545 while (vectors >= vector_threshold) {
4546 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4548 if (!err) /* Success in acquiring all requested vectors. */
4551 vectors = 0; /* Nasty failure, quit now */
4552 else /* err == number of vectors we should try again with */
4556 if (vectors < vector_threshold) {
4557 /* Can't allocate enough MSI-X interrupts? Oh well.
4558 * This just means we'll go with either a single MSI
4559 * vector or fall back to legacy interrupts.
4561 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4562 "Unable to allocate MSI-X interrupts\n");
4563 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4564 kfree(adapter->msix_entries);
4565 adapter->msix_entries = NULL;
4567 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4569 * Adjust for only the vectors we'll use, which is minimum
4570 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4571 * vectors we were allocated.
4573 adapter->num_msix_vectors = min(vectors,
4574 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4579 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4580 * @adapter: board private structure to initialize
4582 * Cache the descriptor ring offsets for RSS to the assigned rings.
4585 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4589 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4592 for (i = 0; i < adapter->num_rx_queues; i++)
4593 adapter->rx_ring[i]->reg_idx = i;
4594 for (i = 0; i < adapter->num_tx_queues; i++)
4595 adapter->tx_ring[i]->reg_idx = i;
4600 #ifdef CONFIG_IXGBE_DCB
4602 /* ixgbe_get_first_reg_idx - Return first register index associated with ring */
4603 static void ixgbe_get_first_reg_idx(struct ixgbe_adapter *adapter, u8 tc,
4604 unsigned int *tx, unsigned int *rx)
4606 struct net_device *dev = adapter->netdev;
4607 struct ixgbe_hw *hw = &adapter->hw;
4608 u8 num_tcs = netdev_get_num_tc(dev);
4613 switch (hw->mac.type) {
4614 case ixgbe_mac_82598EB:
4618 case ixgbe_mac_82599EB:
4619 case ixgbe_mac_X540:
4624 } else if (tc < 5) {
4625 *tx = ((tc + 2) << 4);
4627 } else if (tc < num_tcs) {
4628 *tx = ((tc + 8) << 3);
4657 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4658 * @adapter: board private structure to initialize
4660 * Cache the descriptor ring offsets for DCB to the assigned rings.
4663 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4665 struct net_device *dev = adapter->netdev;
4667 u8 num_tcs = netdev_get_num_tc(dev);
4672 for (i = 0, k = 0; i < num_tcs; i++) {
4673 unsigned int tx_s, rx_s;
4674 u16 count = dev->tc_to_txq[i].count;
4676 ixgbe_get_first_reg_idx(adapter, i, &tx_s, &rx_s);
4677 for (j = 0; j < count; j++, k++) {
4678 adapter->tx_ring[k]->reg_idx = tx_s + j;
4679 adapter->rx_ring[k]->reg_idx = rx_s + j;
4680 adapter->tx_ring[k]->dcb_tc = i;
4681 adapter->rx_ring[k]->dcb_tc = i;
4690 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4691 * @adapter: board private structure to initialize
4693 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4696 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4701 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
4702 (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) {
4703 for (i = 0; i < adapter->num_rx_queues; i++)
4704 adapter->rx_ring[i]->reg_idx = i;
4705 for (i = 0; i < adapter->num_tx_queues; i++)
4706 adapter->tx_ring[i]->reg_idx = i;
4715 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4716 * @adapter: board private structure to initialize
4718 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4721 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4723 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4725 u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4727 if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4730 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4731 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
4732 ixgbe_cache_ring_fdir(adapter);
4734 ixgbe_cache_ring_rss(adapter);
4736 fcoe_rx_i = f->mask;
4737 fcoe_tx_i = f->mask;
4739 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4740 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4741 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4746 #endif /* IXGBE_FCOE */
4748 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4749 * @adapter: board private structure to initialize
4751 * SR-IOV doesn't use any descriptor rings but changes the default if
4752 * no other mapping is used.
4755 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4757 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4758 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4759 if (adapter->num_vfs)
4766 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4767 * @adapter: board private structure to initialize
4769 * Once we know the feature-set enabled for the device, we'll cache
4770 * the register offset the descriptor ring is assigned to.
4772 * Note, the order the various feature calls is important. It must start with
4773 * the "most" features enabled at the same time, then trickle down to the
4774 * least amount of features turned on at once.
4776 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4778 /* start with default case */
4779 adapter->rx_ring[0]->reg_idx = 0;
4780 adapter->tx_ring[0]->reg_idx = 0;
4782 if (ixgbe_cache_ring_sriov(adapter))
4785 #ifdef CONFIG_IXGBE_DCB
4786 if (ixgbe_cache_ring_dcb(adapter))
4791 if (ixgbe_cache_ring_fcoe(adapter))
4793 #endif /* IXGBE_FCOE */
4795 if (ixgbe_cache_ring_fdir(adapter))
4798 if (ixgbe_cache_ring_rss(adapter))
4803 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4804 * @adapter: board private structure to initialize
4806 * Attempt to configure the interrupts using the best available
4807 * capabilities of the hardware and the kernel.
4809 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4811 struct ixgbe_hw *hw = &adapter->hw;
4813 int vector, v_budget;
4816 * It's easy to be greedy for MSI-X vectors, but it really
4817 * doesn't do us much good if we have a lot more vectors
4818 * than CPU's. So let's be conservative and only ask for
4819 * (roughly) the same number of vectors as there are CPU's.
4820 * The default is to use pairs of vectors.
4822 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
4823 v_budget = min_t(int, v_budget, num_online_cpus());
4824 v_budget += NON_Q_VECTORS;
4827 * At the same time, hardware can only support a maximum of
4828 * hw.mac->max_msix_vectors vectors. With features
4829 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4830 * descriptor queues supported by our device. Thus, we cap it off in
4831 * those rare cases where the cpu count also exceeds our vector limit.
4833 v_budget = min_t(int, v_budget, hw->mac.max_msix_vectors);
4835 /* A failure in MSI-X entry allocation isn't fatal, but it does
4836 * mean we disable MSI-X capabilities of the adapter. */
4837 adapter->msix_entries = kcalloc(v_budget,
4838 sizeof(struct msix_entry), GFP_KERNEL);
4839 if (adapter->msix_entries) {
4840 for (vector = 0; vector < v_budget; vector++)
4841 adapter->msix_entries[vector].entry = vector;
4843 ixgbe_acquire_msix_vectors(adapter, v_budget);
4845 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4849 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4850 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4851 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4853 "ATR is not supported while multiple "
4854 "queues are disabled. Disabling Flow Director\n");
4856 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4857 adapter->atr_sample_rate = 0;
4858 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4859 ixgbe_disable_sriov(adapter);
4861 err = ixgbe_set_num_queues(adapter);
4865 err = pci_enable_msi(adapter->pdev);
4867 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4869 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4870 "Unable to allocate MSI interrupt, "
4871 "falling back to legacy. Error: %d\n", err);
4880 static void ixgbe_add_ring(struct ixgbe_ring *ring,
4881 struct ixgbe_ring_container *head)
4883 ring->next = head->ring;
4889 * ixgbe_alloc_q_vector - Allocate memory for a single interrupt vector
4890 * @adapter: board private structure to initialize
4891 * @v_idx: index of vector in adapter struct
4893 * We allocate one q_vector. If allocation fails we return -ENOMEM.
4895 static int ixgbe_alloc_q_vector(struct ixgbe_adapter *adapter, int v_idx,
4896 int txr_count, int txr_idx,
4897 int rxr_count, int rxr_idx)
4899 struct ixgbe_q_vector *q_vector;
4900 struct ixgbe_ring *ring;
4903 int ring_count, size;
4905 ring_count = txr_count + rxr_count;
4906 size = sizeof(struct ixgbe_q_vector) +
4907 (sizeof(struct ixgbe_ring) * ring_count);
4909 /* customize cpu for Flow Director mapping */
4910 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4911 if (cpu_online(v_idx)) {
4913 node = cpu_to_node(cpu);
4917 /* allocate q_vector and rings */
4918 q_vector = kzalloc_node(size, GFP_KERNEL, node);
4920 q_vector = kzalloc(size, GFP_KERNEL);
4924 /* setup affinity mask and node */
4926 cpumask_set_cpu(cpu, &q_vector->affinity_mask);
4928 cpumask_copy(&q_vector->affinity_mask, cpu_online_mask);
4929 q_vector->numa_node = node;
4931 /* initialize NAPI */
4932 netif_napi_add(adapter->netdev, &q_vector->napi,
4935 /* tie q_vector and adapter together */
4936 adapter->q_vector[v_idx] = q_vector;
4937 q_vector->adapter = adapter;
4938 q_vector->v_idx = v_idx;
4940 /* initialize work limits */
4941 q_vector->tx.work_limit = adapter->tx_work_limit;
4943 /* initialize pointer to rings */
4944 ring = q_vector->ring;
4947 /* assign generic ring traits */
4948 ring->dev = &adapter->pdev->dev;
4949 ring->netdev = adapter->netdev;
4951 /* configure backlink on ring */
4952 ring->q_vector = q_vector;
4954 /* update q_vector Tx values */
4955 ixgbe_add_ring(ring, &q_vector->tx);
4957 /* apply Tx specific ring traits */
4958 ring->count = adapter->tx_ring_count;
4959 ring->queue_index = txr_idx;
4961 /* assign ring to adapter */
4962 adapter->tx_ring[txr_idx] = ring;
4964 /* update count and index */
4968 /* push pointer to next ring */
4973 /* assign generic ring traits */
4974 ring->dev = &adapter->pdev->dev;
4975 ring->netdev = adapter->netdev;
4977 /* configure backlink on ring */
4978 ring->q_vector = q_vector;
4980 /* update q_vector Rx values */
4981 ixgbe_add_ring(ring, &q_vector->rx);
4984 * 82599 errata, UDP frames with a 0 checksum
4985 * can be marked as checksum errors.
4987 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
4988 set_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state);
4990 /* apply Rx specific ring traits */
4991 ring->count = adapter->rx_ring_count;
4992 ring->queue_index = rxr_idx;
4994 /* assign ring to adapter */
4995 adapter->rx_ring[rxr_idx] = ring;
4997 /* update count and index */
5001 /* push pointer to next ring */
5009 * ixgbe_free_q_vector - Free memory allocated for specific interrupt vector
5010 * @adapter: board private structure to initialize
5011 * @v_idx: Index of vector to be freed
5013 * This function frees the memory allocated to the q_vector. In addition if
5014 * NAPI is enabled it will delete any references to the NAPI struct prior
5015 * to freeing the q_vector.
5017 static void ixgbe_free_q_vector(struct ixgbe_adapter *adapter, int v_idx)
5019 struct ixgbe_q_vector *q_vector = adapter->q_vector[v_idx];
5020 struct ixgbe_ring *ring;
5022 ixgbe_for_each_ring(ring, q_vector->tx)
5023 adapter->tx_ring[ring->queue_index] = NULL;
5025 ixgbe_for_each_ring(ring, q_vector->rx)
5026 adapter->rx_ring[ring->queue_index] = NULL;
5028 adapter->q_vector[v_idx] = NULL;
5029 netif_napi_del(&q_vector->napi);
5032 * ixgbe_get_stats64() might access the rings on this vector,
5033 * we must wait a grace period before freeing it.
5035 kfree_rcu(q_vector, rcu);
5039 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
5040 * @adapter: board private structure to initialize
5042 * We allocate one q_vector per queue interrupt. If allocation fails we
5045 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
5047 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5048 int rxr_remaining = adapter->num_rx_queues;
5049 int txr_remaining = adapter->num_tx_queues;
5050 int rxr_idx = 0, txr_idx = 0, v_idx = 0;
5053 /* only one q_vector if MSI-X is disabled. */
5054 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
5057 if (q_vectors >= (rxr_remaining + txr_remaining)) {
5058 for (; rxr_remaining; v_idx++, q_vectors--) {
5059 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
5060 err = ixgbe_alloc_q_vector(adapter, v_idx,
5061 0, 0, rqpv, rxr_idx);
5066 /* update counts and index */
5067 rxr_remaining -= rqpv;
5072 for (; q_vectors; v_idx++, q_vectors--) {
5073 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
5074 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors);
5075 err = ixgbe_alloc_q_vector(adapter, v_idx,
5082 /* update counts and index */
5083 rxr_remaining -= rqpv;
5085 txr_remaining -= tqpv;
5094 ixgbe_free_q_vector(adapter, v_idx);
5101 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
5102 * @adapter: board private structure to initialize
5104 * This function frees the memory allocated to the q_vectors. In addition if
5105 * NAPI is enabled it will delete any references to the NAPI struct prior
5106 * to freeing the q_vector.
5108 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
5110 int v_idx, q_vectors;
5112 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5113 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5117 for (v_idx = 0; v_idx < q_vectors; v_idx++)
5118 ixgbe_free_q_vector(adapter, v_idx);
5121 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
5123 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5124 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
5125 pci_disable_msix(adapter->pdev);
5126 kfree(adapter->msix_entries);
5127 adapter->msix_entries = NULL;
5128 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
5129 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
5130 pci_disable_msi(adapter->pdev);
5135 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
5136 * @adapter: board private structure to initialize
5138 * We determine which interrupt scheme to use based on...
5139 * - Kernel support (MSI, MSI-X)
5140 * - which can be user-defined (via MODULE_PARAM)
5141 * - Hardware queue count (num_*_queues)
5142 * - defined by miscellaneous hardware support/features (RSS, etc.)
5144 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
5148 /* Number of supported queues */
5149 err = ixgbe_set_num_queues(adapter);
5153 err = ixgbe_set_interrupt_capability(adapter);
5155 e_dev_err("Unable to setup interrupt capabilities\n");
5156 goto err_set_interrupt;
5159 err = ixgbe_alloc_q_vectors(adapter);
5161 e_dev_err("Unable to allocate memory for queue vectors\n");
5162 goto err_alloc_q_vectors;
5165 ixgbe_cache_ring_register(adapter);
5167 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
5168 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
5169 adapter->num_rx_queues, adapter->num_tx_queues);
5171 set_bit(__IXGBE_DOWN, &adapter->state);
5175 err_alloc_q_vectors:
5176 ixgbe_reset_interrupt_capability(adapter);
5182 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
5183 * @adapter: board private structure to clear interrupt scheme on
5185 * We go through and clear interrupt specific resources and reset the structure
5186 * to pre-load conditions
5188 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
5190 adapter->num_tx_queues = 0;
5191 adapter->num_rx_queues = 0;
5193 ixgbe_free_q_vectors(adapter);
5194 ixgbe_reset_interrupt_capability(adapter);
5198 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5199 * @adapter: board private structure to initialize
5201 * ixgbe_sw_init initializes the Adapter private data structure.
5202 * Fields are initialized based on PCI device information and
5203 * OS network device settings (MTU size).
5205 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
5207 struct ixgbe_hw *hw = &adapter->hw;
5208 struct pci_dev *pdev = adapter->pdev;
5210 #ifdef CONFIG_IXGBE_DCB
5212 struct tc_configuration *tc;
5215 /* PCI config space info */
5217 hw->vendor_id = pdev->vendor;
5218 hw->device_id = pdev->device;
5219 hw->revision_id = pdev->revision;
5220 hw->subsystem_vendor_id = pdev->subsystem_vendor;
5221 hw->subsystem_device_id = pdev->subsystem_device;
5223 /* Set capability flags */
5224 rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
5225 adapter->ring_feature[RING_F_RSS].indices = rss;
5226 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
5227 switch (hw->mac.type) {
5228 case ixgbe_mac_82598EB:
5229 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5230 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5231 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
5233 case ixgbe_mac_X540:
5234 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5235 case ixgbe_mac_82599EB:
5236 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
5237 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5238 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5239 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5240 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5241 /* Flow Director hash filters enabled */
5242 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
5243 adapter->atr_sample_rate = 20;
5244 adapter->ring_feature[RING_F_FDIR].indices =
5245 IXGBE_MAX_FDIR_INDICES;
5246 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
5248 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5249 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5250 adapter->ring_feature[RING_F_FCOE].indices = 0;
5251 #ifdef CONFIG_IXGBE_DCB
5252 /* Default traffic class to use for FCoE */
5253 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5255 #endif /* IXGBE_FCOE */
5261 /* n-tuple support exists, always init our spinlock */
5262 spin_lock_init(&adapter->fdir_perfect_lock);
5264 #ifdef CONFIG_IXGBE_DCB
5265 switch (hw->mac.type) {
5266 case ixgbe_mac_X540:
5267 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
5268 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
5271 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
5272 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
5276 /* Configure DCB traffic classes */
5277 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5278 tc = &adapter->dcb_cfg.tc_config[j];
5279 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5280 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5281 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5282 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5283 tc->dcb_pfc = pfc_disabled;
5286 /* Initialize default user to priority mapping, UPx->TC0 */
5287 tc = &adapter->dcb_cfg.tc_config[0];
5288 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5289 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5291 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5292 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5293 adapter->dcb_cfg.pfc_mode_enable = false;
5294 adapter->dcb_set_bitmap = 0x00;
5295 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5296 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5301 /* default flow control settings */
5302 hw->fc.requested_mode = ixgbe_fc_full;
5303 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5305 adapter->last_lfc_mode = hw->fc.current_mode;
5307 ixgbe_pbthresh_setup(adapter);
5308 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5309 hw->fc.send_xon = true;
5310 hw->fc.disable_fc_autoneg = false;
5312 /* enable itr by default in dynamic mode */
5313 adapter->rx_itr_setting = 1;
5314 adapter->tx_itr_setting = 1;
5316 /* set default ring sizes */
5317 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5318 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5320 /* set default work limits */
5321 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5323 /* initialize eeprom parameters */
5324 if (ixgbe_init_eeprom_params_generic(hw)) {
5325 e_dev_err("EEPROM initialization failed\n");
5329 set_bit(__IXGBE_DOWN, &adapter->state);
5335 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5336 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5338 * Return 0 on success, negative on failure
5340 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5342 struct device *dev = tx_ring->dev;
5343 int orig_node = dev_to_node(dev);
5347 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5349 if (tx_ring->q_vector)
5350 numa_node = tx_ring->q_vector->numa_node;
5352 tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
5353 if (!tx_ring->tx_buffer_info)
5354 tx_ring->tx_buffer_info = vzalloc(size);
5355 if (!tx_ring->tx_buffer_info)
5358 /* round up to nearest 4K */
5359 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5360 tx_ring->size = ALIGN(tx_ring->size, 4096);
5362 set_dev_node(dev, numa_node);
5363 tx_ring->desc = dma_alloc_coherent(dev,
5367 set_dev_node(dev, orig_node);
5369 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5370 &tx_ring->dma, GFP_KERNEL);
5374 tx_ring->next_to_use = 0;
5375 tx_ring->next_to_clean = 0;
5379 vfree(tx_ring->tx_buffer_info);
5380 tx_ring->tx_buffer_info = NULL;
5381 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5386 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5387 * @adapter: board private structure
5389 * If this function returns with an error, then it's possible one or
5390 * more of the rings is populated (while the rest are not). It is the
5391 * callers duty to clean those orphaned rings.
5393 * Return 0 on success, negative on failure
5395 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5399 for (i = 0; i < adapter->num_tx_queues; i++) {
5400 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5403 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5411 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5412 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5414 * Returns 0 on success, negative on failure
5416 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5418 struct device *dev = rx_ring->dev;
5419 int orig_node = dev_to_node(dev);
5423 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5425 if (rx_ring->q_vector)
5426 numa_node = rx_ring->q_vector->numa_node;
5428 rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
5429 if (!rx_ring->rx_buffer_info)
5430 rx_ring->rx_buffer_info = vzalloc(size);
5431 if (!rx_ring->rx_buffer_info)
5434 /* Round up to nearest 4K */
5435 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5436 rx_ring->size = ALIGN(rx_ring->size, 4096);
5438 set_dev_node(dev, numa_node);
5439 rx_ring->desc = dma_alloc_coherent(dev,
5443 set_dev_node(dev, orig_node);
5445 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5446 &rx_ring->dma, GFP_KERNEL);
5450 rx_ring->next_to_clean = 0;
5451 rx_ring->next_to_use = 0;
5453 ixgbe_init_rx_page_offset(rx_ring);
5457 vfree(rx_ring->rx_buffer_info);
5458 rx_ring->rx_buffer_info = NULL;
5459 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5464 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5465 * @adapter: board private structure
5467 * If this function returns with an error, then it's possible one or
5468 * more of the rings is populated (while the rest are not). It is the
5469 * callers duty to clean those orphaned rings.
5471 * Return 0 on success, negative on failure
5473 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5477 for (i = 0; i < adapter->num_rx_queues; i++) {
5478 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5481 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5489 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5490 * @tx_ring: Tx descriptor ring for a specific queue
5492 * Free all transmit software resources
5494 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5496 ixgbe_clean_tx_ring(tx_ring);
5498 vfree(tx_ring->tx_buffer_info);
5499 tx_ring->tx_buffer_info = NULL;
5501 /* if not set, then don't free */
5505 dma_free_coherent(tx_ring->dev, tx_ring->size,
5506 tx_ring->desc, tx_ring->dma);
5508 tx_ring->desc = NULL;
5512 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5513 * @adapter: board private structure
5515 * Free all transmit software resources
5517 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5521 for (i = 0; i < adapter->num_tx_queues; i++)
5522 if (adapter->tx_ring[i]->desc)
5523 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5527 * ixgbe_free_rx_resources - Free Rx Resources
5528 * @rx_ring: ring to clean the resources from
5530 * Free all receive software resources
5532 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5534 ixgbe_clean_rx_ring(rx_ring);
5536 vfree(rx_ring->rx_buffer_info);
5537 rx_ring->rx_buffer_info = NULL;
5539 /* if not set, then don't free */
5543 dma_free_coherent(rx_ring->dev, rx_ring->size,
5544 rx_ring->desc, rx_ring->dma);
5546 rx_ring->desc = NULL;
5550 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5551 * @adapter: board private structure
5553 * Free all receive software resources
5555 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5559 for (i = 0; i < adapter->num_rx_queues; i++)
5560 if (adapter->rx_ring[i]->desc)
5561 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5565 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5566 * @netdev: network interface device structure
5567 * @new_mtu: new value for maximum frame size
5569 * Returns 0 on success, negative on failure
5571 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5573 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5574 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5576 /* MTU < 68 is an error and causes problems on some kernels */
5577 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5581 * For 82599EB we cannot allow PF to change MTU greater than 1500
5582 * in SR-IOV mode as it may cause buffer overruns in guest VFs that
5583 * don't allocate and chain buffers correctly.
5585 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
5586 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
5587 (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
5590 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5592 /* must set new MTU before calling down or up */
5593 netdev->mtu = new_mtu;
5595 if (netif_running(netdev))
5596 ixgbe_reinit_locked(adapter);
5602 * ixgbe_open - Called when a network interface is made active
5603 * @netdev: network interface device structure
5605 * Returns 0 on success, negative value on failure
5607 * The open entry point is called when a network interface is made
5608 * active by the system (IFF_UP). At this point all resources needed
5609 * for transmit and receive operations are allocated, the interrupt
5610 * handler is registered with the OS, the watchdog timer is started,
5611 * and the stack is notified that the interface is ready.
5613 static int ixgbe_open(struct net_device *netdev)
5615 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5618 /* disallow open during test */
5619 if (test_bit(__IXGBE_TESTING, &adapter->state))
5622 netif_carrier_off(netdev);
5624 /* allocate transmit descriptors */
5625 err = ixgbe_setup_all_tx_resources(adapter);
5629 /* allocate receive descriptors */
5630 err = ixgbe_setup_all_rx_resources(adapter);
5634 ixgbe_configure(adapter);
5636 err = ixgbe_request_irq(adapter);
5640 ixgbe_up_complete(adapter);
5646 ixgbe_free_all_rx_resources(adapter);
5648 ixgbe_free_all_tx_resources(adapter);
5649 ixgbe_reset(adapter);
5655 * ixgbe_close - Disables a network interface
5656 * @netdev: network interface device structure
5658 * Returns 0, this is not allowed to fail
5660 * The close entry point is called when an interface is de-activated
5661 * by the OS. The hardware is still under the drivers control, but
5662 * needs to be disabled. A global MAC reset is issued to stop the
5663 * hardware, and all transmit and receive resources are freed.
5665 static int ixgbe_close(struct net_device *netdev)
5667 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5669 ixgbe_down(adapter);
5670 ixgbe_free_irq(adapter);
5672 ixgbe_fdir_filter_exit(adapter);
5674 ixgbe_free_all_tx_resources(adapter);
5675 ixgbe_free_all_rx_resources(adapter);
5677 ixgbe_release_hw_control(adapter);
5683 static int ixgbe_resume(struct pci_dev *pdev)
5685 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5686 struct net_device *netdev = adapter->netdev;
5689 pci_set_power_state(pdev, PCI_D0);
5690 pci_restore_state(pdev);
5692 * pci_restore_state clears dev->state_saved so call
5693 * pci_save_state to restore it.
5695 pci_save_state(pdev);
5697 err = pci_enable_device_mem(pdev);
5699 e_dev_err("Cannot enable PCI device from suspend\n");
5702 pci_set_master(pdev);
5704 pci_wake_from_d3(pdev, false);
5706 err = ixgbe_init_interrupt_scheme(adapter);
5708 e_dev_err("Cannot initialize interrupts for device\n");
5712 ixgbe_reset(adapter);
5714 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5716 if (netif_running(netdev)) {
5717 err = ixgbe_open(netdev);
5722 netif_device_attach(netdev);
5726 #endif /* CONFIG_PM */
5728 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5730 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5731 struct net_device *netdev = adapter->netdev;
5732 struct ixgbe_hw *hw = &adapter->hw;
5734 u32 wufc = adapter->wol;
5739 netif_device_detach(netdev);
5741 if (netif_running(netdev)) {
5742 ixgbe_down(adapter);
5743 ixgbe_free_irq(adapter);
5744 ixgbe_free_all_tx_resources(adapter);
5745 ixgbe_free_all_rx_resources(adapter);
5748 ixgbe_clear_interrupt_scheme(adapter);
5750 kfree(adapter->ixgbe_ieee_pfc);
5751 kfree(adapter->ixgbe_ieee_ets);
5755 retval = pci_save_state(pdev);
5761 ixgbe_set_rx_mode(netdev);
5763 /* turn on all-multi mode if wake on multicast is enabled */
5764 if (wufc & IXGBE_WUFC_MC) {
5765 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5766 fctrl |= IXGBE_FCTRL_MPE;
5767 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5770 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5771 ctrl |= IXGBE_CTRL_GIO_DIS;
5772 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5774 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5776 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5777 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5780 switch (hw->mac.type) {
5781 case ixgbe_mac_82598EB:
5782 pci_wake_from_d3(pdev, false);
5784 case ixgbe_mac_82599EB:
5785 case ixgbe_mac_X540:
5786 pci_wake_from_d3(pdev, !!wufc);
5792 *enable_wake = !!wufc;
5794 ixgbe_release_hw_control(adapter);
5796 pci_disable_device(pdev);
5802 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5807 retval = __ixgbe_shutdown(pdev, &wake);
5812 pci_prepare_to_sleep(pdev);
5814 pci_wake_from_d3(pdev, false);
5815 pci_set_power_state(pdev, PCI_D3hot);
5820 #endif /* CONFIG_PM */
5822 static void ixgbe_shutdown(struct pci_dev *pdev)
5826 __ixgbe_shutdown(pdev, &wake);
5828 if (system_state == SYSTEM_POWER_OFF) {
5829 pci_wake_from_d3(pdev, wake);
5830 pci_set_power_state(pdev, PCI_D3hot);
5835 * ixgbe_update_stats - Update the board statistics counters.
5836 * @adapter: board private structure
5838 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5840 struct net_device *netdev = adapter->netdev;
5841 struct ixgbe_hw *hw = &adapter->hw;
5842 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5844 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5845 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5846 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5847 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5849 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5851 u64 fcoe_noddp_counts_sum = 0, fcoe_noddp_ext_buff_counts_sum = 0;
5852 #endif /* IXGBE_FCOE */
5854 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5855 test_bit(__IXGBE_RESETTING, &adapter->state))
5858 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5861 for (i = 0; i < 16; i++)
5862 adapter->hw_rx_no_dma_resources +=
5863 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5864 for (i = 0; i < adapter->num_rx_queues; i++) {
5865 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5866 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5868 adapter->rsc_total_count = rsc_count;
5869 adapter->rsc_total_flush = rsc_flush;
5872 for (i = 0; i < adapter->num_rx_queues; i++) {
5873 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5874 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5875 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5876 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5877 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5878 bytes += rx_ring->stats.bytes;
5879 packets += rx_ring->stats.packets;
5881 adapter->non_eop_descs = non_eop_descs;
5882 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5883 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5884 adapter->hw_csum_rx_error = hw_csum_rx_error;
5885 netdev->stats.rx_bytes = bytes;
5886 netdev->stats.rx_packets = packets;
5890 /* gather some stats to the adapter struct that are per queue */
5891 for (i = 0; i < adapter->num_tx_queues; i++) {
5892 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5893 restart_queue += tx_ring->tx_stats.restart_queue;
5894 tx_busy += tx_ring->tx_stats.tx_busy;
5895 bytes += tx_ring->stats.bytes;
5896 packets += tx_ring->stats.packets;
5898 adapter->restart_queue = restart_queue;
5899 adapter->tx_busy = tx_busy;
5900 netdev->stats.tx_bytes = bytes;
5901 netdev->stats.tx_packets = packets;
5903 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5905 /* 8 register reads */
5906 for (i = 0; i < 8; i++) {
5907 /* for packet buffers not used, the register should read 0 */
5908 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5910 hwstats->mpc[i] += mpc;
5911 total_mpc += hwstats->mpc[i];
5912 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5913 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5914 switch (hw->mac.type) {
5915 case ixgbe_mac_82598EB:
5916 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5917 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5918 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5919 hwstats->pxonrxc[i] +=
5920 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5922 case ixgbe_mac_82599EB:
5923 case ixgbe_mac_X540:
5924 hwstats->pxonrxc[i] +=
5925 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5932 /*16 register reads */
5933 for (i = 0; i < 16; i++) {
5934 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5935 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5936 if ((hw->mac.type == ixgbe_mac_82599EB) ||
5937 (hw->mac.type == ixgbe_mac_X540)) {
5938 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5939 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5940 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5941 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5945 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5946 /* work around hardware counting issue */
5947 hwstats->gprc -= missed_rx;
5949 ixgbe_update_xoff_received(adapter);
5951 /* 82598 hardware only has a 32 bit counter in the high register */
5952 switch (hw->mac.type) {
5953 case ixgbe_mac_82598EB:
5954 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5955 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5956 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5957 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5959 case ixgbe_mac_X540:
5960 /* OS2BMC stats are X540 only*/
5961 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5962 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5963 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5964 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5965 case ixgbe_mac_82599EB:
5966 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5967 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5968 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5969 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5970 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5971 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5972 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5973 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5974 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5976 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5977 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5978 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5979 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5980 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5981 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5982 /* Add up per cpu counters for total ddp aloc fail */
5983 if (fcoe->pcpu_noddp && fcoe->pcpu_noddp_ext_buff) {
5984 for_each_possible_cpu(cpu) {
5985 fcoe_noddp_counts_sum +=
5986 *per_cpu_ptr(fcoe->pcpu_noddp, cpu);
5987 fcoe_noddp_ext_buff_counts_sum +=
5989 pcpu_noddp_ext_buff, cpu);
5992 hwstats->fcoe_noddp = fcoe_noddp_counts_sum;
5993 hwstats->fcoe_noddp_ext_buff = fcoe_noddp_ext_buff_counts_sum;
5994 #endif /* IXGBE_FCOE */
5999 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
6000 hwstats->bprc += bprc;
6001 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
6002 if (hw->mac.type == ixgbe_mac_82598EB)
6003 hwstats->mprc -= bprc;
6004 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
6005 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
6006 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
6007 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
6008 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
6009 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
6010 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
6011 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
6012 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
6013 hwstats->lxontxc += lxon;
6014 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
6015 hwstats->lxofftxc += lxoff;
6016 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
6017 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
6019 * 82598 errata - tx of flow control packets is included in tx counters
6021 xon_off_tot = lxon + lxoff;
6022 hwstats->gptc -= xon_off_tot;
6023 hwstats->mptc -= xon_off_tot;
6024 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
6025 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
6026 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
6027 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
6028 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
6029 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
6030 hwstats->ptc64 -= xon_off_tot;
6031 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
6032 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
6033 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
6034 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
6035 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
6036 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
6038 /* Fill out the OS statistics structure */
6039 netdev->stats.multicast = hwstats->mprc;
6042 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
6043 netdev->stats.rx_dropped = 0;
6044 netdev->stats.rx_length_errors = hwstats->rlec;
6045 netdev->stats.rx_crc_errors = hwstats->crcerrs;
6046 netdev->stats.rx_missed_errors = total_mpc;
6050 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
6051 * @adapter - pointer to the device adapter structure
6053 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
6055 struct ixgbe_hw *hw = &adapter->hw;
6058 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
6061 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6063 /* if interface is down do nothing */
6064 if (test_bit(__IXGBE_DOWN, &adapter->state))
6067 /* do nothing if we are not using signature filters */
6068 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
6071 adapter->fdir_overflow++;
6073 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
6074 for (i = 0; i < adapter->num_tx_queues; i++)
6075 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
6076 &(adapter->tx_ring[i]->state));
6077 /* re-enable flow director interrupts */
6078 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
6080 e_err(probe, "failed to finish FDIR re-initialization, "
6081 "ignored adding FDIR ATR filters\n");
6086 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
6087 * @adapter - pointer to the device adapter structure
6089 * This function serves two purposes. First it strobes the interrupt lines
6090 * in order to make certain interrupts are occurring. Secondly it sets the
6091 * bits needed to check for TX hangs. As a result we should immediately
6092 * determine if a hang has occurred.
6094 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
6096 struct ixgbe_hw *hw = &adapter->hw;
6100 /* If we're down or resetting, just bail */
6101 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6102 test_bit(__IXGBE_RESETTING, &adapter->state))
6105 /* Force detection of hung controller */
6106 if (netif_carrier_ok(adapter->netdev)) {
6107 for (i = 0; i < adapter->num_tx_queues; i++)
6108 set_check_for_tx_hang(adapter->tx_ring[i]);
6111 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6113 * for legacy and MSI interrupts don't set any bits
6114 * that are enabled for EIAM, because this operation
6115 * would set *both* EIMS and EICS for any bit in EIAM
6117 IXGBE_WRITE_REG(hw, IXGBE_EICS,
6118 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
6120 /* get one bit for every active tx/rx interrupt vector */
6121 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
6122 struct ixgbe_q_vector *qv = adapter->q_vector[i];
6123 if (qv->rx.ring || qv->tx.ring)
6124 eics |= ((u64)1 << i);
6128 /* Cause software interrupt to ensure rings are cleaned */
6129 ixgbe_irq_rearm_queues(adapter, eics);
6134 * ixgbe_watchdog_update_link - update the link status
6135 * @adapter - pointer to the device adapter structure
6136 * @link_speed - pointer to a u32 to store the link_speed
6138 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
6140 struct ixgbe_hw *hw = &adapter->hw;
6141 u32 link_speed = adapter->link_speed;
6142 bool link_up = adapter->link_up;
6145 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6148 if (hw->mac.ops.check_link) {
6149 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6151 /* always assume link is up, if no check link function */
6152 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
6156 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6157 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
6158 hw->mac.ops.fc_enable(hw, i);
6160 hw->mac.ops.fc_enable(hw, 0);
6165 time_after(jiffies, (adapter->link_check_timeout +
6166 IXGBE_TRY_LINK_TIMEOUT))) {
6167 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6168 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6169 IXGBE_WRITE_FLUSH(hw);
6172 adapter->link_up = link_up;
6173 adapter->link_speed = link_speed;
6177 * ixgbe_watchdog_link_is_up - update netif_carrier status and
6178 * print link up message
6179 * @adapter - pointer to the device adapter structure
6181 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
6183 struct net_device *netdev = adapter->netdev;
6184 struct ixgbe_hw *hw = &adapter->hw;
6185 u32 link_speed = adapter->link_speed;
6186 bool flow_rx, flow_tx;
6188 /* only continue if link was previously down */
6189 if (netif_carrier_ok(netdev))
6192 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
6194 switch (hw->mac.type) {
6195 case ixgbe_mac_82598EB: {
6196 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6197 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6198 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6199 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6202 case ixgbe_mac_X540:
6203 case ixgbe_mac_82599EB: {
6204 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6205 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6206 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6207 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6215 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
6216 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
6218 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
6220 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
6223 ((flow_rx && flow_tx) ? "RX/TX" :
6225 (flow_tx ? "TX" : "None"))));
6227 netif_carrier_on(netdev);
6228 ixgbe_check_vf_rate_limit(adapter);
6232 * ixgbe_watchdog_link_is_down - update netif_carrier status and
6233 * print link down message
6234 * @adapter - pointer to the adapter structure
6236 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
6238 struct net_device *netdev = adapter->netdev;
6239 struct ixgbe_hw *hw = &adapter->hw;
6241 adapter->link_up = false;
6242 adapter->link_speed = 0;
6244 /* only continue if link was up previously */
6245 if (!netif_carrier_ok(netdev))
6248 /* poll for SFP+ cable when link is down */
6249 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
6250 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
6252 e_info(drv, "NIC Link is Down\n");
6253 netif_carrier_off(netdev);
6257 * ixgbe_watchdog_flush_tx - flush queues on link down
6258 * @adapter - pointer to the device adapter structure
6260 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
6263 int some_tx_pending = 0;
6265 if (!netif_carrier_ok(adapter->netdev)) {
6266 for (i = 0; i < adapter->num_tx_queues; i++) {
6267 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6268 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
6269 some_tx_pending = 1;
6274 if (some_tx_pending) {
6275 /* We've lost link, so the controller stops DMA,
6276 * but we've got queued Tx work that's never going
6277 * to get done, so reset controller to flush Tx.
6278 * (Do the reset outside of interrupt context).
6280 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
6285 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6289 /* Do not perform spoof check for 82598 */
6290 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6293 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6296 * ssvpc register is cleared on read, if zero then no
6297 * spoofed packets in the last interval.
6302 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
6306 * ixgbe_watchdog_subtask - check and bring link up
6307 * @adapter - pointer to the device adapter structure
6309 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6311 /* if interface is down do nothing */
6312 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6313 test_bit(__IXGBE_RESETTING, &adapter->state))
6316 ixgbe_watchdog_update_link(adapter);
6318 if (adapter->link_up)
6319 ixgbe_watchdog_link_is_up(adapter);
6321 ixgbe_watchdog_link_is_down(adapter);
6323 ixgbe_spoof_check(adapter);
6324 ixgbe_update_stats(adapter);
6326 ixgbe_watchdog_flush_tx(adapter);
6330 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6331 * @adapter - the ixgbe adapter structure
6333 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6335 struct ixgbe_hw *hw = &adapter->hw;
6338 /* not searching for SFP so there is nothing to do here */
6339 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6340 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6343 /* someone else is in init, wait until next service event */
6344 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6347 err = hw->phy.ops.identify_sfp(hw);
6348 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6351 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6352 /* If no cable is present, then we need to reset
6353 * the next time we find a good cable. */
6354 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6361 /* exit if reset not needed */
6362 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6365 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6368 * A module may be identified correctly, but the EEPROM may not have
6369 * support for that module. setup_sfp() will fail in that case, so
6370 * we should not allow that module to load.
6372 if (hw->mac.type == ixgbe_mac_82598EB)
6373 err = hw->phy.ops.reset(hw);
6375 err = hw->mac.ops.setup_sfp(hw);
6377 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6380 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6381 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6384 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6386 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6387 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6388 e_dev_err("failed to initialize because an unsupported "
6389 "SFP+ module type was detected.\n");
6390 e_dev_err("Reload the driver after installing a "
6391 "supported module.\n");
6392 unregister_netdev(adapter->netdev);
6397 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6398 * @adapter - the ixgbe adapter structure
6400 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6402 struct ixgbe_hw *hw = &adapter->hw;
6406 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6409 /* someone else is in init, wait until next service event */
6410 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6413 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6415 autoneg = hw->phy.autoneg_advertised;
6416 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
6417 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
6418 if (hw->mac.ops.setup_link)
6419 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
6421 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6422 adapter->link_check_timeout = jiffies;
6423 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6426 #ifdef CONFIG_PCI_IOV
6427 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6430 struct ixgbe_hw *hw = &adapter->hw;
6431 struct net_device *netdev = adapter->netdev;
6435 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6436 if (gpc) /* If incrementing then no need for the check below */
6439 * Check to see if a bad DMA write target from an errant or
6440 * malicious VF has caused a PCIe error. If so then we can
6441 * issue a VFLR to the offending VF(s) and then resume without
6442 * requesting a full slot reset.
6445 for (vf = 0; vf < adapter->num_vfs; vf++) {
6446 ciaa = (vf << 16) | 0x80000000;
6447 /* 32 bit read so align, we really want status at offset 6 */
6448 ciaa |= PCI_COMMAND;
6449 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6450 ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
6452 /* disable debug mode asap after reading data */
6453 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6454 /* Get the upper 16 bits which will be the PCI status reg */
6456 if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
6457 netdev_err(netdev, "VF %d Hung DMA\n", vf);
6459 ciaa = (vf << 16) | 0x80000000;
6461 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6462 ciad = 0x00008000; /* VFLR */
6463 IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
6465 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6472 * ixgbe_service_timer - Timer Call-back
6473 * @data: pointer to adapter cast into an unsigned long
6475 static void ixgbe_service_timer(unsigned long data)
6477 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6478 unsigned long next_event_offset;
6481 /* poll faster when waiting for link */
6482 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6483 next_event_offset = HZ / 10;
6485 next_event_offset = HZ * 2;
6487 #ifdef CONFIG_PCI_IOV
6489 * don't bother with SR-IOV VF DMA hang check if there are
6490 * no VFs or the link is down
6492 if (!adapter->num_vfs ||
6493 (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6494 goto normal_timer_service;
6496 /* If we have VFs allocated then we must check for DMA hangs */
6497 ixgbe_check_for_bad_vf(adapter);
6498 next_event_offset = HZ / 50;
6499 adapter->timer_event_accumulator++;
6501 if (adapter->timer_event_accumulator >= 100)
6502 adapter->timer_event_accumulator = 0;
6506 normal_timer_service:
6508 /* Reset the timer */
6509 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6512 ixgbe_service_event_schedule(adapter);
6515 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6517 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6520 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6522 /* If we're already down or resetting, just bail */
6523 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6524 test_bit(__IXGBE_RESETTING, &adapter->state))
6527 ixgbe_dump(adapter);
6528 netdev_err(adapter->netdev, "Reset adapter\n");
6529 adapter->tx_timeout_count++;
6531 ixgbe_reinit_locked(adapter);
6535 * ixgbe_service_task - manages and runs subtasks
6536 * @work: pointer to work_struct containing our data
6538 static void ixgbe_service_task(struct work_struct *work)
6540 struct ixgbe_adapter *adapter = container_of(work,
6541 struct ixgbe_adapter,
6544 ixgbe_reset_subtask(adapter);
6545 ixgbe_sfp_detection_subtask(adapter);
6546 ixgbe_sfp_link_config_subtask(adapter);
6547 ixgbe_check_overtemp_subtask(adapter);
6548 ixgbe_watchdog_subtask(adapter);
6549 ixgbe_fdir_reinit_subtask(adapter);
6550 ixgbe_check_hang_subtask(adapter);
6552 ixgbe_service_event_complete(adapter);
6555 void ixgbe_tx_ctxtdesc(struct ixgbe_ring *tx_ring, u32 vlan_macip_lens,
6556 u32 fcoe_sof_eof, u32 type_tucmd, u32 mss_l4len_idx)
6558 struct ixgbe_adv_tx_context_desc *context_desc;
6559 u16 i = tx_ring->next_to_use;
6561 context_desc = IXGBE_TX_CTXTDESC(tx_ring, i);
6564 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
6566 /* set bits to identify this as an advanced context descriptor */
6567 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
6569 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6570 context_desc->seqnum_seed = cpu_to_le32(fcoe_sof_eof);
6571 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
6572 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6575 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
6576 struct ixgbe_tx_buffer *first,
6579 struct sk_buff *skb = first->skb;
6580 u32 vlan_macip_lens, type_tucmd;
6581 u32 mss_l4len_idx, l4len;
6583 if (!skb_is_gso(skb))
6586 if (skb_header_cloned(skb)) {
6587 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6592 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6593 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6595 if (first->protocol == __constant_htons(ETH_P_IP)) {
6596 struct iphdr *iph = ip_hdr(skb);
6599 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6603 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6604 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6605 IXGBE_TX_FLAGS_CSUM |
6606 IXGBE_TX_FLAGS_IPV4;
6607 } else if (skb_is_gso_v6(skb)) {
6608 ipv6_hdr(skb)->payload_len = 0;
6609 tcp_hdr(skb)->check =
6610 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6611 &ipv6_hdr(skb)->daddr,
6613 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6614 IXGBE_TX_FLAGS_CSUM;
6617 /* compute header lengths */
6618 l4len = tcp_hdrlen(skb);
6619 *hdr_len = skb_transport_offset(skb) + l4len;
6621 /* update gso size and bytecount with header size */
6622 first->gso_segs = skb_shinfo(skb)->gso_segs;
6623 first->bytecount += (first->gso_segs - 1) * *hdr_len;
6625 /* mss_l4len_id: use 1 as index for TSO */
6626 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6627 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6628 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
6630 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6631 vlan_macip_lens = skb_network_header_len(skb);
6632 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6633 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6635 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6641 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6642 struct ixgbe_tx_buffer *first)
6644 struct sk_buff *skb = first->skb;
6645 u32 vlan_macip_lens = 0;
6646 u32 mss_l4len_idx = 0;
6649 if (skb->ip_summed != CHECKSUM_PARTIAL) {
6650 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6651 !(first->tx_flags & IXGBE_TX_FLAGS_TXSW))
6655 switch (first->protocol) {
6656 case __constant_htons(ETH_P_IP):
6657 vlan_macip_lens |= skb_network_header_len(skb);
6658 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6659 l4_hdr = ip_hdr(skb)->protocol;
6661 case __constant_htons(ETH_P_IPV6):
6662 vlan_macip_lens |= skb_network_header_len(skb);
6663 l4_hdr = ipv6_hdr(skb)->nexthdr;
6666 if (unlikely(net_ratelimit())) {
6667 dev_warn(tx_ring->dev,
6668 "partial checksum but proto=%x!\n",
6676 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6677 mss_l4len_idx = tcp_hdrlen(skb) <<
6678 IXGBE_ADVTXD_L4LEN_SHIFT;
6681 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6682 mss_l4len_idx = sizeof(struct sctphdr) <<
6683 IXGBE_ADVTXD_L4LEN_SHIFT;
6686 mss_l4len_idx = sizeof(struct udphdr) <<
6687 IXGBE_ADVTXD_L4LEN_SHIFT;
6690 if (unlikely(net_ratelimit())) {
6691 dev_warn(tx_ring->dev,
6692 "partial checksum but l4 proto=%x!\n",
6698 /* update TX checksum flag */
6699 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
6702 /* vlan_macip_lens: MACLEN, VLAN tag */
6703 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6704 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6706 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6707 type_tucmd, mss_l4len_idx);
6710 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
6712 /* set type for advanced descriptor with frame checksum insertion */
6713 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
6714 IXGBE_ADVTXD_DCMD_IFCS |
6715 IXGBE_ADVTXD_DCMD_DEXT);
6717 /* set HW vlan bit if vlan is present */
6718 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
6719 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
6721 /* set segmentation enable bits for TSO/FSO */
6723 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FSO))
6725 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6727 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
6732 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
6733 u32 tx_flags, unsigned int paylen)
6735 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
6737 /* enable L4 checksum for TSO and TX checksum offload */
6738 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6739 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
6741 /* enble IPv4 checksum for TSO */
6742 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6743 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
6745 /* use index 1 context for TSO/FSO/FCOE */
6747 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FCOE))
6749 if (tx_flags & IXGBE_TX_FLAGS_TSO)
6751 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
6754 * Check Context must be set if Tx switch is enabled, which it
6755 * always is for case where virtual functions are running
6758 if (tx_flags & (IXGBE_TX_FLAGS_TXSW | IXGBE_TX_FLAGS_FCOE))
6760 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
6762 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
6764 tx_desc->read.olinfo_status = olinfo_status;
6767 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6770 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6771 struct ixgbe_tx_buffer *first,
6775 struct sk_buff *skb = first->skb;
6776 struct ixgbe_tx_buffer *tx_buffer;
6777 union ixgbe_adv_tx_desc *tx_desc;
6778 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6779 unsigned int data_len = skb->data_len;
6780 unsigned int size = skb_headlen(skb);
6781 unsigned int paylen = skb->len - hdr_len;
6782 u32 tx_flags = first->tx_flags;
6784 u16 i = tx_ring->next_to_use;
6786 tx_desc = IXGBE_TX_DESC(tx_ring, i);
6788 ixgbe_tx_olinfo_status(tx_desc, tx_flags, paylen);
6789 cmd_type = ixgbe_tx_cmd_type(tx_flags);
6792 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6793 if (data_len < sizeof(struct fcoe_crc_eof)) {
6794 size -= sizeof(struct fcoe_crc_eof) - data_len;
6797 data_len -= sizeof(struct fcoe_crc_eof);
6802 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6803 if (dma_mapping_error(tx_ring->dev, dma))
6806 /* record length, and DMA address */
6807 dma_unmap_len_set(first, len, size);
6808 dma_unmap_addr_set(first, dma, dma);
6810 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6813 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
6814 tx_desc->read.cmd_type_len =
6815 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
6819 if (i == tx_ring->count) {
6820 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6824 dma += IXGBE_MAX_DATA_PER_TXD;
6825 size -= IXGBE_MAX_DATA_PER_TXD;
6827 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6828 tx_desc->read.olinfo_status = 0;
6831 if (likely(!data_len))
6834 if (unlikely(skb->no_fcs))
6835 cmd_type &= ~(cpu_to_le32(IXGBE_ADVTXD_DCMD_IFCS));
6836 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6840 if (i == tx_ring->count) {
6841 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6846 size = min_t(unsigned int, data_len, skb_frag_size(frag));
6848 size = skb_frag_size(frag);
6852 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6854 if (dma_mapping_error(tx_ring->dev, dma))
6857 tx_buffer = &tx_ring->tx_buffer_info[i];
6858 dma_unmap_len_set(tx_buffer, len, size);
6859 dma_unmap_addr_set(tx_buffer, dma, dma);
6861 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6862 tx_desc->read.olinfo_status = 0;
6867 /* write last descriptor with RS and EOP bits */
6868 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
6869 tx_desc->read.cmd_type_len = cmd_type;
6871 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6873 /* set the timestamp */
6874 first->time_stamp = jiffies;
6877 * Force memory writes to complete before letting h/w know there
6878 * are new descriptors to fetch. (Only applicable for weak-ordered
6879 * memory model archs, such as IA-64).
6881 * We also need this memory barrier to make certain all of the
6882 * status bits have been updated before next_to_watch is written.
6886 /* set next_to_watch value indicating a packet is present */
6887 first->next_to_watch = tx_desc;
6890 if (i == tx_ring->count)
6893 tx_ring->next_to_use = i;
6895 /* notify HW of packet */
6896 writel(i, tx_ring->tail);
6900 dev_err(tx_ring->dev, "TX DMA map failed\n");
6902 /* clear dma mappings for failed tx_buffer_info map */
6904 tx_buffer = &tx_ring->tx_buffer_info[i];
6905 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6906 if (tx_buffer == first)
6913 tx_ring->next_to_use = i;
6916 static void ixgbe_atr(struct ixgbe_ring *ring,
6917 struct ixgbe_tx_buffer *first)
6919 struct ixgbe_q_vector *q_vector = ring->q_vector;
6920 union ixgbe_atr_hash_dword input = { .dword = 0 };
6921 union ixgbe_atr_hash_dword common = { .dword = 0 };
6923 unsigned char *network;
6925 struct ipv6hdr *ipv6;
6930 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6934 /* do nothing if sampling is disabled */
6935 if (!ring->atr_sample_rate)
6940 /* snag network header to get L4 type and address */
6941 hdr.network = skb_network_header(first->skb);
6943 /* Currently only IPv4/IPv6 with TCP is supported */
6944 if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6945 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6946 (first->protocol != __constant_htons(ETH_P_IP) ||
6947 hdr.ipv4->protocol != IPPROTO_TCP))
6950 th = tcp_hdr(first->skb);
6952 /* skip this packet since it is invalid or the socket is closing */
6956 /* sample on all syn packets or once every atr sample count */
6957 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6960 /* reset sample count */
6961 ring->atr_count = 0;
6963 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6966 * src and dst are inverted, think how the receiver sees them
6968 * The input is broken into two sections, a non-compressed section
6969 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6970 * is XORed together and stored in the compressed dword.
6972 input.formatted.vlan_id = vlan_id;
6975 * since src port and flex bytes occupy the same word XOR them together
6976 * and write the value to source port portion of compressed dword
6978 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6979 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6981 common.port.src ^= th->dest ^ first->protocol;
6982 common.port.dst ^= th->source;
6984 if (first->protocol == __constant_htons(ETH_P_IP)) {
6985 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6986 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6988 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6989 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6990 hdr.ipv6->saddr.s6_addr32[1] ^
6991 hdr.ipv6->saddr.s6_addr32[2] ^
6992 hdr.ipv6->saddr.s6_addr32[3] ^
6993 hdr.ipv6->daddr.s6_addr32[0] ^
6994 hdr.ipv6->daddr.s6_addr32[1] ^
6995 hdr.ipv6->daddr.s6_addr32[2] ^
6996 hdr.ipv6->daddr.s6_addr32[3];
6999 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
7000 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
7001 input, common, ring->queue_index);
7004 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7006 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
7007 /* Herbert's original patch had:
7008 * smp_mb__after_netif_stop_queue();
7009 * but since that doesn't exist yet, just open code it. */
7012 /* We need to check again in a case another CPU has just
7013 * made room available. */
7014 if (likely(ixgbe_desc_unused(tx_ring) < size))
7017 /* A reprieve! - use start_queue because it doesn't call schedule */
7018 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
7019 ++tx_ring->tx_stats.restart_queue;
7023 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7025 if (likely(ixgbe_desc_unused(tx_ring) >= size))
7027 return __ixgbe_maybe_stop_tx(tx_ring, size);
7030 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
7032 struct ixgbe_adapter *adapter = netdev_priv(dev);
7033 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
7036 __be16 protocol = vlan_get_protocol(skb);
7038 if (((protocol == htons(ETH_P_FCOE)) ||
7039 (protocol == htons(ETH_P_FIP))) &&
7040 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
7041 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
7042 txq += adapter->ring_feature[RING_F_FCOE].mask;
7047 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
7048 while (unlikely(txq >= dev->real_num_tx_queues))
7049 txq -= dev->real_num_tx_queues;
7053 return skb_tx_hash(dev, skb);
7056 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
7057 struct ixgbe_adapter *adapter,
7058 struct ixgbe_ring *tx_ring)
7060 struct ixgbe_tx_buffer *first;
7063 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
7066 u16 count = TXD_USE_COUNT(skb_headlen(skb));
7067 __be16 protocol = skb->protocol;
7071 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
7072 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
7073 * + 2 desc gap to keep tail from touching head,
7074 * + 1 desc for context descriptor,
7075 * otherwise try next time
7077 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
7078 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
7079 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
7081 count += skb_shinfo(skb)->nr_frags;
7083 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
7084 tx_ring->tx_stats.tx_busy++;
7085 return NETDEV_TX_BUSY;
7088 /* record the location of the first descriptor for this packet */
7089 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
7091 first->bytecount = skb->len;
7092 first->gso_segs = 1;
7094 /* if we have a HW VLAN tag being added default to the HW one */
7095 if (vlan_tx_tag_present(skb)) {
7096 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
7097 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7098 /* else if it is a SW VLAN check the next protocol and store the tag */
7099 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
7100 struct vlan_hdr *vhdr, _vhdr;
7101 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
7105 protocol = vhdr->h_vlan_encapsulated_proto;
7106 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
7107 IXGBE_TX_FLAGS_VLAN_SHIFT;
7108 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
7111 #ifdef CONFIG_PCI_IOV
7113 * Use the l2switch_enable flag - would be false if the DMA
7114 * Tx switch had been disabled.
7116 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7117 tx_flags |= IXGBE_TX_FLAGS_TXSW;
7120 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
7121 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7122 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
7123 (skb->priority != TC_PRIO_CONTROL))) {
7124 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
7125 tx_flags |= (skb->priority & 0x7) <<
7126 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
7127 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
7128 struct vlan_ethhdr *vhdr;
7129 if (skb_header_cloned(skb) &&
7130 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
7132 vhdr = (struct vlan_ethhdr *)skb->data;
7133 vhdr->h_vlan_TCI = htons(tx_flags >>
7134 IXGBE_TX_FLAGS_VLAN_SHIFT);
7136 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7140 /* record initial flags and protocol */
7141 first->tx_flags = tx_flags;
7142 first->protocol = protocol;
7145 /* setup tx offload for FCoE */
7146 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
7147 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
7148 tso = ixgbe_fso(tx_ring, first, &hdr_len);
7155 #endif /* IXGBE_FCOE */
7156 tso = ixgbe_tso(tx_ring, first, &hdr_len);
7160 ixgbe_tx_csum(tx_ring, first);
7162 /* add the ATR filter if ATR is on */
7163 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
7164 ixgbe_atr(tx_ring, first);
7168 #endif /* IXGBE_FCOE */
7169 ixgbe_tx_map(tx_ring, first, hdr_len);
7171 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
7173 return NETDEV_TX_OK;
7176 dev_kfree_skb_any(first->skb);
7179 return NETDEV_TX_OK;
7182 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
7183 struct net_device *netdev)
7185 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7186 struct ixgbe_ring *tx_ring;
7188 if (skb->len <= 0) {
7189 dev_kfree_skb_any(skb);
7190 return NETDEV_TX_OK;
7194 * The minimum packet size for olinfo paylen is 17 so pad the skb
7195 * in order to meet this minimum size requirement.
7197 if (skb->len < 17) {
7198 if (skb_padto(skb, 17))
7199 return NETDEV_TX_OK;
7203 tx_ring = adapter->tx_ring[skb->queue_mapping];
7204 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
7208 * ixgbe_set_mac - Change the Ethernet Address of the NIC
7209 * @netdev: network interface device structure
7210 * @p: pointer to an address structure
7212 * Returns 0 on success, negative on failure
7214 static int ixgbe_set_mac(struct net_device *netdev, void *p)
7216 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7217 struct ixgbe_hw *hw = &adapter->hw;
7218 struct sockaddr *addr = p;
7220 if (!is_valid_ether_addr(addr->sa_data))
7221 return -EADDRNOTAVAIL;
7223 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
7224 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
7226 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
7233 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
7235 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7236 struct ixgbe_hw *hw = &adapter->hw;
7240 if (prtad != hw->phy.mdio.prtad)
7242 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
7248 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
7249 u16 addr, u16 value)
7251 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7252 struct ixgbe_hw *hw = &adapter->hw;
7254 if (prtad != hw->phy.mdio.prtad)
7256 return hw->phy.ops.write_reg(hw, addr, devad, value);
7259 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
7261 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7263 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
7267 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
7269 * @netdev: network interface device structure
7271 * Returns non-zero on failure
7273 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
7276 struct ixgbe_adapter *adapter = netdev_priv(dev);
7277 struct ixgbe_mac_info *mac = &adapter->hw.mac;
7279 if (is_valid_ether_addr(mac->san_addr)) {
7281 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
7288 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
7290 * @netdev: network interface device structure
7292 * Returns non-zero on failure
7294 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
7297 struct ixgbe_adapter *adapter = netdev_priv(dev);
7298 struct ixgbe_mac_info *mac = &adapter->hw.mac;
7300 if (is_valid_ether_addr(mac->san_addr)) {
7302 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
7308 #ifdef CONFIG_NET_POLL_CONTROLLER
7310 * Polling 'interrupt' - used by things like netconsole to send skbs
7311 * without having to re-enable interrupts. It's not called while
7312 * the interrupt routine is executing.
7314 static void ixgbe_netpoll(struct net_device *netdev)
7316 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7319 /* if interface is down do nothing */
7320 if (test_bit(__IXGBE_DOWN, &adapter->state))
7323 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
7324 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
7325 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
7326 for (i = 0; i < num_q_vectors; i++) {
7327 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
7328 ixgbe_msix_clean_rings(0, q_vector);
7331 ixgbe_intr(adapter->pdev->irq, netdev);
7333 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
7337 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7338 struct rtnl_link_stats64 *stats)
7340 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7344 for (i = 0; i < adapter->num_rx_queues; i++) {
7345 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7351 start = u64_stats_fetch_begin_bh(&ring->syncp);
7352 packets = ring->stats.packets;
7353 bytes = ring->stats.bytes;
7354 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7355 stats->rx_packets += packets;
7356 stats->rx_bytes += bytes;
7360 for (i = 0; i < adapter->num_tx_queues; i++) {
7361 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7367 start = u64_stats_fetch_begin_bh(&ring->syncp);
7368 packets = ring->stats.packets;
7369 bytes = ring->stats.bytes;
7370 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
7371 stats->tx_packets += packets;
7372 stats->tx_bytes += bytes;
7376 /* following stats updated by ixgbe_watchdog_task() */
7377 stats->multicast = netdev->stats.multicast;
7378 stats->rx_errors = netdev->stats.rx_errors;
7379 stats->rx_length_errors = netdev->stats.rx_length_errors;
7380 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
7381 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
7385 /* ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7386 * #adapter: pointer to ixgbe_adapter
7387 * @tc: number of traffic classes currently enabled
7389 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7390 * 802.1Q priority maps to a packet buffer that exists.
7392 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
7394 struct ixgbe_hw *hw = &adapter->hw;
7398 /* 82598 have a static priority to TC mapping that can not
7399 * be changed so no validation is needed.
7401 if (hw->mac.type == ixgbe_mac_82598EB)
7404 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
7407 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
7408 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
7410 /* If up2tc is out of bounds default to zero */
7412 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
7416 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
7421 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
7424 * @netdev: net device to configure
7425 * @tc: number of traffic classes to enable
7427 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7429 struct ixgbe_adapter *adapter = netdev_priv(dev);
7430 struct ixgbe_hw *hw = &adapter->hw;
7432 /* Multiple traffic classes requires multiple queues */
7433 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7434 e_err(drv, "Enable failed, needs MSI-X\n");
7438 /* Hardware supports up to 8 traffic classes */
7439 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
7440 (hw->mac.type == ixgbe_mac_82598EB &&
7441 tc < MAX_TRAFFIC_CLASS))
7444 /* Hardware has to reinitialize queues and interrupts to
7445 * match packet buffer alignment. Unfortunately, the
7446 * hardware is not flexible enough to do this dynamically.
7448 if (netif_running(dev))
7450 ixgbe_clear_interrupt_scheme(adapter);
7453 netdev_set_num_tc(dev, tc);
7454 adapter->last_lfc_mode = adapter->hw.fc.current_mode;
7455 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7456 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7458 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7459 adapter->hw.fc.requested_mode = ixgbe_fc_none;
7461 netdev_reset_tc(dev);
7462 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7464 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7465 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7467 adapter->temp_dcb_cfg.pfc_mode_enable = false;
7468 adapter->dcb_cfg.pfc_mode_enable = false;
7471 ixgbe_init_interrupt_scheme(adapter);
7472 ixgbe_validate_rtr(adapter, tc);
7473 if (netif_running(dev))
7479 void ixgbe_do_reset(struct net_device *netdev)
7481 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7483 if (netif_running(netdev))
7484 ixgbe_reinit_locked(adapter);
7486 ixgbe_reset(adapter);
7489 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
7490 netdev_features_t data)
7492 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7495 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
7496 data &= ~NETIF_F_HW_VLAN_RX;
7499 /* return error if RXHASH is being enabled when RSS is not supported */
7500 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
7501 data &= ~NETIF_F_RXHASH;
7503 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7504 if (!(data & NETIF_F_RXCSUM))
7505 data &= ~NETIF_F_LRO;
7507 /* Turn off LRO if not RSC capable or invalid ITR settings */
7508 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)) {
7509 data &= ~NETIF_F_LRO;
7510 } else if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
7511 (adapter->rx_itr_setting != 1 &&
7512 adapter->rx_itr_setting > IXGBE_MAX_RSC_INT_RATE)) {
7513 data &= ~NETIF_F_LRO;
7514 e_info(probe, "rx-usecs set too low, not enabling RSC\n");
7520 static int ixgbe_set_features(struct net_device *netdev,
7521 netdev_features_t data)
7523 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7524 netdev_features_t changed = netdev->features ^ data;
7525 bool need_reset = false;
7527 /* Make sure RSC matches LRO, reset if change */
7528 if (!!(data & NETIF_F_LRO) !=
7529 !!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7530 adapter->flags2 ^= IXGBE_FLAG2_RSC_ENABLED;
7531 switch (adapter->hw.mac.type) {
7532 case ixgbe_mac_X540:
7533 case ixgbe_mac_82599EB:
7542 * Check if Flow Director n-tuple support was enabled or disabled. If
7543 * the state changed, we need to reset.
7545 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
7546 /* turn off ATR, enable perfect filters and reset */
7547 if (data & NETIF_F_NTUPLE) {
7548 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7549 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7552 } else if (!(data & NETIF_F_NTUPLE)) {
7553 /* turn off Flow Director, set ATR and reset */
7554 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7555 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
7556 !(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
7557 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7561 if (changed & NETIF_F_RXALL)
7564 netdev->features = data;
7566 ixgbe_do_reset(netdev);
7572 static const struct net_device_ops ixgbe_netdev_ops = {
7573 .ndo_open = ixgbe_open,
7574 .ndo_stop = ixgbe_close,
7575 .ndo_start_xmit = ixgbe_xmit_frame,
7576 .ndo_select_queue = ixgbe_select_queue,
7577 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7578 .ndo_validate_addr = eth_validate_addr,
7579 .ndo_set_mac_address = ixgbe_set_mac,
7580 .ndo_change_mtu = ixgbe_change_mtu,
7581 .ndo_tx_timeout = ixgbe_tx_timeout,
7582 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7583 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7584 .ndo_do_ioctl = ixgbe_ioctl,
7585 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7586 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7587 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7588 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
7589 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7590 .ndo_get_stats64 = ixgbe_get_stats64,
7591 .ndo_setup_tc = ixgbe_setup_tc,
7592 #ifdef CONFIG_NET_POLL_CONTROLLER
7593 .ndo_poll_controller = ixgbe_netpoll,
7596 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7597 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7598 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7599 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7600 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7601 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7602 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
7603 #endif /* IXGBE_FCOE */
7604 .ndo_set_features = ixgbe_set_features,
7605 .ndo_fix_features = ixgbe_fix_features,
7608 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
7609 const struct ixgbe_info *ii)
7611 #ifdef CONFIG_PCI_IOV
7612 struct ixgbe_hw *hw = &adapter->hw;
7614 if (hw->mac.type == ixgbe_mac_82598EB)
7617 /* The 82599 supports up to 64 VFs per physical function
7618 * but this implementation limits allocation to 63 so that
7619 * basic networking resources are still available to the
7622 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
7623 ixgbe_enable_sriov(adapter, ii);
7624 #endif /* CONFIG_PCI_IOV */
7628 * ixgbe_probe - Device Initialization Routine
7629 * @pdev: PCI device information struct
7630 * @ent: entry in ixgbe_pci_tbl
7632 * Returns 0 on success, negative on failure
7634 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7635 * The OS initialization, configuring of the adapter private structure,
7636 * and a hardware reset occur.
7638 static int __devinit ixgbe_probe(struct pci_dev *pdev,
7639 const struct pci_device_id *ent)
7641 struct net_device *netdev;
7642 struct ixgbe_adapter *adapter = NULL;
7643 struct ixgbe_hw *hw;
7644 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7645 static int cards_found;
7646 int i, err, pci_using_dac;
7647 u8 part_str[IXGBE_PBANUM_LENGTH];
7648 unsigned int indices = num_possible_cpus();
7655 /* Catch broken hardware that put the wrong VF device ID in
7656 * the PCIe SR-IOV capability.
7658 if (pdev->is_virtfn) {
7659 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7660 pci_name(pdev), pdev->vendor, pdev->device);
7664 err = pci_enable_device_mem(pdev);
7668 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7669 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7672 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7674 err = dma_set_coherent_mask(&pdev->dev,
7678 "No usable DMA configuration, aborting\n");
7685 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7686 IORESOURCE_MEM), ixgbe_driver_name);
7689 "pci_request_selected_regions failed 0x%x\n", err);
7693 pci_enable_pcie_error_reporting(pdev);
7695 pci_set_master(pdev);
7696 pci_save_state(pdev);
7698 #ifdef CONFIG_IXGBE_DCB
7699 indices *= MAX_TRAFFIC_CLASS;
7702 if (ii->mac == ixgbe_mac_82598EB)
7703 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7705 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7708 indices += min_t(unsigned int, num_possible_cpus(),
7709 IXGBE_MAX_FCOE_INDICES);
7711 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7714 goto err_alloc_etherdev;
7717 SET_NETDEV_DEV(netdev, &pdev->dev);
7719 adapter = netdev_priv(netdev);
7720 pci_set_drvdata(pdev, adapter);
7722 adapter->netdev = netdev;
7723 adapter->pdev = pdev;
7726 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7728 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7729 pci_resource_len(pdev, 0));
7735 for (i = 1; i <= 5; i++) {
7736 if (pci_resource_len(pdev, i) == 0)
7740 netdev->netdev_ops = &ixgbe_netdev_ops;
7741 ixgbe_set_ethtool_ops(netdev);
7742 netdev->watchdog_timeo = 5 * HZ;
7743 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7745 adapter->bd_number = cards_found;
7748 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7749 hw->mac.type = ii->mac;
7752 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7753 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7754 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7755 if (!(eec & (1 << 8)))
7756 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7759 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7760 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7761 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7762 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7763 hw->phy.mdio.mmds = 0;
7764 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7765 hw->phy.mdio.dev = netdev;
7766 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7767 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7769 ii->get_invariants(hw);
7771 /* setup the private structure */
7772 err = ixgbe_sw_init(adapter);
7776 /* Make it possible the adapter to be woken up via WOL */
7777 switch (adapter->hw.mac.type) {
7778 case ixgbe_mac_82599EB:
7779 case ixgbe_mac_X540:
7780 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7787 * If there is a fan on this device and it has failed log the
7790 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7791 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7792 if (esdp & IXGBE_ESDP_SDP1)
7793 e_crit(probe, "Fan has stopped, replace the adapter\n");
7796 if (allow_unsupported_sfp)
7797 hw->allow_unsupported_sfp = allow_unsupported_sfp;
7799 /* reset_hw fills in the perm_addr as well */
7800 hw->phy.reset_if_overtemp = true;
7801 err = hw->mac.ops.reset_hw(hw);
7802 hw->phy.reset_if_overtemp = false;
7803 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7804 hw->mac.type == ixgbe_mac_82598EB) {
7806 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7807 e_dev_err("failed to load because an unsupported SFP+ "
7808 "module type was detected.\n");
7809 e_dev_err("Reload the driver after installing a supported "
7813 e_dev_err("HW Init failed: %d\n", err);
7817 ixgbe_probe_vf(adapter, ii);
7819 netdev->features = NETIF_F_SG |
7822 NETIF_F_HW_VLAN_TX |
7823 NETIF_F_HW_VLAN_RX |
7824 NETIF_F_HW_VLAN_FILTER |
7830 netdev->hw_features = netdev->features;
7832 switch (adapter->hw.mac.type) {
7833 case ixgbe_mac_82599EB:
7834 case ixgbe_mac_X540:
7835 netdev->features |= NETIF_F_SCTP_CSUM;
7836 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7843 netdev->hw_features |= NETIF_F_RXALL;
7845 netdev->vlan_features |= NETIF_F_TSO;
7846 netdev->vlan_features |= NETIF_F_TSO6;
7847 netdev->vlan_features |= NETIF_F_IP_CSUM;
7848 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7849 netdev->vlan_features |= NETIF_F_SG;
7851 netdev->priv_flags |= IFF_UNICAST_FLT;
7852 netdev->priv_flags |= IFF_SUPP_NOFCS;
7854 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7855 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7856 IXGBE_FLAG_DCB_ENABLED);
7858 #ifdef CONFIG_IXGBE_DCB
7859 netdev->dcbnl_ops = &dcbnl_ops;
7863 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7864 if (hw->mac.ops.get_device_caps) {
7865 hw->mac.ops.get_device_caps(hw, &device_caps);
7866 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7867 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7870 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7871 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7872 netdev->vlan_features |= NETIF_F_FSO;
7873 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7875 #endif /* IXGBE_FCOE */
7876 if (pci_using_dac) {
7877 netdev->features |= NETIF_F_HIGHDMA;
7878 netdev->vlan_features |= NETIF_F_HIGHDMA;
7881 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7882 netdev->hw_features |= NETIF_F_LRO;
7883 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7884 netdev->features |= NETIF_F_LRO;
7886 /* make sure the EEPROM is good */
7887 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7888 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7893 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7894 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7896 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7897 e_dev_err("invalid MAC address\n");
7902 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7903 (unsigned long) adapter);
7905 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7906 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7908 err = ixgbe_init_interrupt_scheme(adapter);
7912 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7913 netdev->hw_features &= ~NETIF_F_RXHASH;
7914 netdev->features &= ~NETIF_F_RXHASH;
7917 /* WOL not supported for all but the following */
7919 switch (pdev->device) {
7920 case IXGBE_DEV_ID_82599_SFP:
7921 /* Only these subdevice supports WOL */
7922 switch (pdev->subsystem_device) {
7923 case IXGBE_SUBDEV_ID_82599_560FLR:
7924 /* only support first port */
7925 if (hw->bus.func != 0)
7927 case IXGBE_SUBDEV_ID_82599_SFP:
7928 adapter->wol = IXGBE_WUFC_MAG;
7932 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7933 /* All except this subdevice support WOL */
7934 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7935 adapter->wol = IXGBE_WUFC_MAG;
7937 case IXGBE_DEV_ID_82599_KX4:
7938 adapter->wol = IXGBE_WUFC_MAG;
7940 case IXGBE_DEV_ID_X540T:
7941 /* Check eeprom to see if it is enabled */
7942 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7943 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7945 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7946 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7947 (hw->bus.func == 0)))
7948 adapter->wol = IXGBE_WUFC_MAG;
7951 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7953 /* save off EEPROM version number */
7954 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7955 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7957 /* pick up the PCI bus settings for reporting later */
7958 hw->mac.ops.get_bus_info(hw);
7960 /* print bus type/speed/width info */
7961 e_dev_info("(PCI Express:%s:%s) %pM\n",
7962 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7963 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7965 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7966 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7967 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7971 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7973 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7974 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7975 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7976 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7979 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7980 hw->mac.type, hw->phy.type, part_str);
7982 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7983 e_dev_warn("PCI-Express bandwidth available for this card is "
7984 "not sufficient for optimal performance.\n");
7985 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7989 /* reset the hardware with the new settings */
7990 err = hw->mac.ops.start_hw(hw);
7991 if (err == IXGBE_ERR_EEPROM_VERSION) {
7992 /* We are running on a pre-production device, log a warning */
7993 e_dev_warn("This device is a pre-production adapter/LOM. "
7994 "Please be aware there may be issues associated "
7995 "with your hardware. If you are experiencing "
7996 "problems please contact your Intel or hardware "
7997 "representative who provided you with this "
8000 strcpy(netdev->name, "eth%d");
8001 err = register_netdev(netdev);
8005 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
8006 if (hw->mac.ops.disable_tx_laser &&
8007 ((hw->phy.multispeed_fiber) ||
8008 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
8009 (hw->mac.type == ixgbe_mac_82599EB))))
8010 hw->mac.ops.disable_tx_laser(hw);
8012 /* carrier off reporting is important to ethtool even BEFORE open */
8013 netif_carrier_off(netdev);
8015 #ifdef CONFIG_IXGBE_DCA
8016 if (dca_add_requester(&pdev->dev) == 0) {
8017 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
8018 ixgbe_setup_dca(adapter);
8021 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
8022 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
8023 for (i = 0; i < adapter->num_vfs; i++)
8024 ixgbe_vf_configuration(pdev, (i | 0x10000000));
8027 /* firmware requires driver version to be 0xFFFFFFFF
8028 * since os does not support feature
8030 if (hw->mac.ops.set_fw_drv_ver)
8031 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
8034 /* add san mac addr to netdev */
8035 ixgbe_add_sanmac_netdev(netdev);
8037 e_dev_info("%s\n", ixgbe_default_device_descr);
8042 ixgbe_release_hw_control(adapter);
8043 ixgbe_clear_interrupt_scheme(adapter);
8045 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8046 ixgbe_disable_sriov(adapter);
8047 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
8048 iounmap(hw->hw_addr);
8050 free_netdev(netdev);
8052 pci_release_selected_regions(pdev,
8053 pci_select_bars(pdev, IORESOURCE_MEM));
8056 pci_disable_device(pdev);
8061 * ixgbe_remove - Device Removal Routine
8062 * @pdev: PCI device information struct
8064 * ixgbe_remove is called by the PCI subsystem to alert the driver
8065 * that it should release a PCI device. The could be caused by a
8066 * Hot-Plug event, or because the driver is going to be removed from
8069 static void __devexit ixgbe_remove(struct pci_dev *pdev)
8071 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8072 struct net_device *netdev = adapter->netdev;
8074 set_bit(__IXGBE_DOWN, &adapter->state);
8075 cancel_work_sync(&adapter->service_task);
8077 #ifdef CONFIG_IXGBE_DCA
8078 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
8079 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
8080 dca_remove_requester(&pdev->dev);
8081 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
8086 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
8087 ixgbe_cleanup_fcoe(adapter);
8089 #endif /* IXGBE_FCOE */
8091 /* remove the added san mac */
8092 ixgbe_del_sanmac_netdev(netdev);
8094 if (netdev->reg_state == NETREG_REGISTERED)
8095 unregister_netdev(netdev);
8097 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
8098 if (!(ixgbe_check_vf_assignment(adapter)))
8099 ixgbe_disable_sriov(adapter);
8101 e_dev_warn("Unloading driver while VFs are assigned "
8102 "- VFs will not be deallocated\n");
8105 ixgbe_clear_interrupt_scheme(adapter);
8107 ixgbe_release_hw_control(adapter);
8109 iounmap(adapter->hw.hw_addr);
8110 pci_release_selected_regions(pdev, pci_select_bars(pdev,
8113 e_dev_info("complete\n");
8115 free_netdev(netdev);
8117 pci_disable_pcie_error_reporting(pdev);
8119 pci_disable_device(pdev);
8123 * ixgbe_io_error_detected - called when PCI error is detected
8124 * @pdev: Pointer to PCI device
8125 * @state: The current pci connection state
8127 * This function is called after a PCI bus error affecting
8128 * this device has been detected.
8130 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
8131 pci_channel_state_t state)
8133 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8134 struct net_device *netdev = adapter->netdev;
8136 #ifdef CONFIG_PCI_IOV
8137 struct pci_dev *bdev, *vfdev;
8138 u32 dw0, dw1, dw2, dw3;
8140 u16 req_id, pf_func;
8142 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
8143 adapter->num_vfs == 0)
8144 goto skip_bad_vf_detection;
8146 bdev = pdev->bus->self;
8147 while (bdev && (bdev->pcie_type != PCI_EXP_TYPE_ROOT_PORT))
8148 bdev = bdev->bus->self;
8151 goto skip_bad_vf_detection;
8153 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
8155 goto skip_bad_vf_detection;
8157 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
8158 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
8159 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
8160 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
8163 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
8164 if (!(req_id & 0x0080))
8165 goto skip_bad_vf_detection;
8167 pf_func = req_id & 0x01;
8168 if ((pf_func & 1) == (pdev->devfn & 1)) {
8169 unsigned int device_id;
8171 vf = (req_id & 0x7F) >> 1;
8172 e_dev_err("VF %d has caused a PCIe error\n", vf);
8173 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
8174 "%8.8x\tdw3: %8.8x\n",
8175 dw0, dw1, dw2, dw3);
8176 switch (adapter->hw.mac.type) {
8177 case ixgbe_mac_82599EB:
8178 device_id = IXGBE_82599_VF_DEVICE_ID;
8180 case ixgbe_mac_X540:
8181 device_id = IXGBE_X540_VF_DEVICE_ID;
8188 /* Find the pci device of the offending VF */
8189 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID, device_id, NULL);
8191 if (vfdev->devfn == (req_id & 0xFF))
8193 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID,
8197 * There's a slim chance the VF could have been hot plugged,
8198 * so if it is no longer present we don't need to issue the
8199 * VFLR. Just clean up the AER in that case.
8202 e_dev_err("Issuing VFLR to VF %d\n", vf);
8203 pci_write_config_dword(vfdev, 0xA8, 0x00008000);
8206 pci_cleanup_aer_uncorrect_error_status(pdev);
8210 * Even though the error may have occurred on the other port
8211 * we still need to increment the vf error reference count for
8212 * both ports because the I/O resume function will be called
8215 adapter->vferr_refcount++;
8217 return PCI_ERS_RESULT_RECOVERED;
8219 skip_bad_vf_detection:
8220 #endif /* CONFIG_PCI_IOV */
8221 netif_device_detach(netdev);
8223 if (state == pci_channel_io_perm_failure)
8224 return PCI_ERS_RESULT_DISCONNECT;
8226 if (netif_running(netdev))
8227 ixgbe_down(adapter);
8228 pci_disable_device(pdev);
8230 /* Request a slot reset. */
8231 return PCI_ERS_RESULT_NEED_RESET;
8235 * ixgbe_io_slot_reset - called after the pci bus has been reset.
8236 * @pdev: Pointer to PCI device
8238 * Restart the card from scratch, as if from a cold-boot.
8240 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
8242 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8243 pci_ers_result_t result;
8246 if (pci_enable_device_mem(pdev)) {
8247 e_err(probe, "Cannot re-enable PCI device after reset.\n");
8248 result = PCI_ERS_RESULT_DISCONNECT;
8250 pci_set_master(pdev);
8251 pci_restore_state(pdev);
8252 pci_save_state(pdev);
8254 pci_wake_from_d3(pdev, false);
8256 ixgbe_reset(adapter);
8257 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
8258 result = PCI_ERS_RESULT_RECOVERED;
8261 err = pci_cleanup_aer_uncorrect_error_status(pdev);
8263 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
8264 "failed 0x%0x\n", err);
8265 /* non-fatal, continue */
8272 * ixgbe_io_resume - called when traffic can start flowing again.
8273 * @pdev: Pointer to PCI device
8275 * This callback is called when the error recovery driver tells us that
8276 * its OK to resume normal operation.
8278 static void ixgbe_io_resume(struct pci_dev *pdev)
8280 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8281 struct net_device *netdev = adapter->netdev;
8283 #ifdef CONFIG_PCI_IOV
8284 if (adapter->vferr_refcount) {
8285 e_info(drv, "Resuming after VF err\n");
8286 adapter->vferr_refcount--;
8291 if (netif_running(netdev))
8294 netif_device_attach(netdev);
8297 static struct pci_error_handlers ixgbe_err_handler = {
8298 .error_detected = ixgbe_io_error_detected,
8299 .slot_reset = ixgbe_io_slot_reset,
8300 .resume = ixgbe_io_resume,
8303 static struct pci_driver ixgbe_driver = {
8304 .name = ixgbe_driver_name,
8305 .id_table = ixgbe_pci_tbl,
8306 .probe = ixgbe_probe,
8307 .remove = __devexit_p(ixgbe_remove),
8309 .suspend = ixgbe_suspend,
8310 .resume = ixgbe_resume,
8312 .shutdown = ixgbe_shutdown,
8313 .err_handler = &ixgbe_err_handler
8317 * ixgbe_init_module - Driver Registration Routine
8319 * ixgbe_init_module is the first routine called when the driver is
8320 * loaded. All it does is register with the PCI subsystem.
8322 static int __init ixgbe_init_module(void)
8325 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
8326 pr_info("%s\n", ixgbe_copyright);
8328 #ifdef CONFIG_IXGBE_DCA
8329 dca_register_notify(&dca_notifier);
8332 ret = pci_register_driver(&ixgbe_driver);
8336 module_init(ixgbe_init_module);
8339 * ixgbe_exit_module - Driver Exit Cleanup Routine
8341 * ixgbe_exit_module is called just before the driver is removed
8344 static void __exit ixgbe_exit_module(void)
8346 #ifdef CONFIG_IXGBE_DCA
8347 dca_unregister_notify(&dca_notifier);
8349 pci_unregister_driver(&ixgbe_driver);
8350 rcu_barrier(); /* Wait for completion of call_rcu()'s */
8353 #ifdef CONFIG_IXGBE_DCA
8354 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
8359 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
8360 __ixgbe_notify_dca);
8362 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
8365 #endif /* CONFIG_IXGBE_DCA */
8367 module_exit(ixgbe_exit_module);