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";
59 char ixgbe_default_device_descr[] =
60 "Intel(R) 10 Gigabit Network Connection";
62 static char ixgbe_default_device_descr[] =
63 "Intel(R) 10 Gigabit Network Connection";
68 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
69 __stringify(BUILD) "-k"
70 const char ixgbe_driver_version[] = DRV_VERSION;
71 static const char ixgbe_copyright[] =
72 "Copyright (c) 1999-2012 Intel Corporation.";
74 static const struct ixgbe_info *ixgbe_info_tbl[] = {
75 [board_82598] = &ixgbe_82598_info,
76 [board_82599] = &ixgbe_82599_info,
77 [board_X540] = &ixgbe_X540_info,
80 /* ixgbe_pci_tbl - PCI Device ID Table
82 * Wildcard entries (PCI_ANY_ID) should come last
83 * Last entry must be all 0s
85 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
86 * Class, Class Mask, private data (not used) }
88 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
117 /* required last entry */
120 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
122 #ifdef CONFIG_IXGBE_DCA
123 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
125 static struct notifier_block dca_notifier = {
126 .notifier_call = ixgbe_notify_dca,
132 #ifdef CONFIG_PCI_IOV
133 static unsigned int max_vfs;
134 module_param(max_vfs, uint, 0);
135 MODULE_PARM_DESC(max_vfs,
136 "Maximum number of virtual functions to allocate per physical function");
137 #endif /* CONFIG_PCI_IOV */
139 static unsigned int allow_unsupported_sfp;
140 module_param(allow_unsupported_sfp, uint, 0);
141 MODULE_PARM_DESC(allow_unsupported_sfp,
142 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
144 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
145 static int debug = -1;
146 module_param(debug, int, 0);
147 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
149 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
150 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
151 MODULE_LICENSE("GPL");
152 MODULE_VERSION(DRV_VERSION);
154 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
156 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
157 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
158 schedule_work(&adapter->service_task);
161 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
163 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
165 /* flush memory to make sure state is correct before next watchdog */
166 smp_mb__before_clear_bit();
167 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
170 struct ixgbe_reg_info {
175 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
177 /* General Registers */
178 {IXGBE_CTRL, "CTRL"},
179 {IXGBE_STATUS, "STATUS"},
180 {IXGBE_CTRL_EXT, "CTRL_EXT"},
182 /* Interrupt Registers */
183 {IXGBE_EICR, "EICR"},
186 {IXGBE_SRRCTL(0), "SRRCTL"},
187 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
188 {IXGBE_RDLEN(0), "RDLEN"},
189 {IXGBE_RDH(0), "RDH"},
190 {IXGBE_RDT(0), "RDT"},
191 {IXGBE_RXDCTL(0), "RXDCTL"},
192 {IXGBE_RDBAL(0), "RDBAL"},
193 {IXGBE_RDBAH(0), "RDBAH"},
196 {IXGBE_TDBAL(0), "TDBAL"},
197 {IXGBE_TDBAH(0), "TDBAH"},
198 {IXGBE_TDLEN(0), "TDLEN"},
199 {IXGBE_TDH(0), "TDH"},
200 {IXGBE_TDT(0), "TDT"},
201 {IXGBE_TXDCTL(0), "TXDCTL"},
203 /* List Terminator */
209 * ixgbe_regdump - register printout routine
211 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
217 switch (reginfo->ofs) {
218 case IXGBE_SRRCTL(0):
219 for (i = 0; i < 64; i++)
220 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
222 case IXGBE_DCA_RXCTRL(0):
223 for (i = 0; i < 64; i++)
224 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
227 for (i = 0; i < 64; i++)
228 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
231 for (i = 0; i < 64; i++)
232 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
235 for (i = 0; i < 64; i++)
236 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
238 case IXGBE_RXDCTL(0):
239 for (i = 0; i < 64; i++)
240 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
243 for (i = 0; i < 64; i++)
244 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
247 for (i = 0; i < 64; i++)
248 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
251 for (i = 0; i < 64; i++)
252 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
255 for (i = 0; i < 64; i++)
256 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
259 for (i = 0; i < 64; i++)
260 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
263 for (i = 0; i < 64; i++)
264 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
267 for (i = 0; i < 64; i++)
268 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
270 case IXGBE_TXDCTL(0):
271 for (i = 0; i < 64; i++)
272 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
275 pr_info("%-15s %08x\n", reginfo->name,
276 IXGBE_READ_REG(hw, reginfo->ofs));
280 for (i = 0; i < 8; i++) {
281 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
282 pr_err("%-15s", rname);
283 for (j = 0; j < 8; j++)
284 pr_cont(" %08x", regs[i*8+j]);
291 * ixgbe_dump - Print registers, tx-rings and rx-rings
293 static void ixgbe_dump(struct ixgbe_adapter *adapter)
295 struct net_device *netdev = adapter->netdev;
296 struct ixgbe_hw *hw = &adapter->hw;
297 struct ixgbe_reg_info *reginfo;
299 struct ixgbe_ring *tx_ring;
300 struct ixgbe_tx_buffer *tx_buffer;
301 union ixgbe_adv_tx_desc *tx_desc;
302 struct my_u0 { u64 a; u64 b; } *u0;
303 struct ixgbe_ring *rx_ring;
304 union ixgbe_adv_rx_desc *rx_desc;
305 struct ixgbe_rx_buffer *rx_buffer_info;
309 if (!netif_msg_hw(adapter))
312 /* Print netdevice Info */
314 dev_info(&adapter->pdev->dev, "Net device Info\n");
315 pr_info("Device Name state "
316 "trans_start last_rx\n");
317 pr_info("%-15s %016lX %016lX %016lX\n",
324 /* Print Registers */
325 dev_info(&adapter->pdev->dev, "Register Dump\n");
326 pr_info(" Register Name Value\n");
327 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
328 reginfo->name; reginfo++) {
329 ixgbe_regdump(hw, reginfo);
332 /* Print TX Ring Summary */
333 if (!netdev || !netif_running(netdev))
336 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
337 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
338 for (n = 0; n < adapter->num_tx_queues; n++) {
339 tx_ring = adapter->tx_ring[n];
340 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
341 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
342 n, tx_ring->next_to_use, tx_ring->next_to_clean,
343 (u64)dma_unmap_addr(tx_buffer, dma),
344 dma_unmap_len(tx_buffer, len),
345 tx_buffer->next_to_watch,
346 (u64)tx_buffer->time_stamp);
350 if (!netif_msg_tx_done(adapter))
351 goto rx_ring_summary;
353 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
355 /* Transmit Descriptor Formats
357 * Advanced Transmit Descriptor
358 * +--------------------------------------------------------------+
359 * 0 | Buffer Address [63:0] |
360 * +--------------------------------------------------------------+
361 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
362 * +--------------------------------------------------------------+
363 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
366 for (n = 0; n < adapter->num_tx_queues; n++) {
367 tx_ring = adapter->tx_ring[n];
368 pr_info("------------------------------------\n");
369 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
370 pr_info("------------------------------------\n");
371 pr_info("T [desc] [address 63:0 ] "
372 "[PlPOIdStDDt Ln] [bi->dma ] "
373 "leng ntw timestamp bi->skb\n");
375 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
376 tx_desc = IXGBE_TX_DESC(tx_ring, i);
377 tx_buffer = &tx_ring->tx_buffer_info[i];
378 u0 = (struct my_u0 *)tx_desc;
379 pr_info("T [0x%03X] %016llX %016llX %016llX"
380 " %04X %p %016llX %p", i,
383 (u64)dma_unmap_addr(tx_buffer, dma),
384 dma_unmap_len(tx_buffer, len),
385 tx_buffer->next_to_watch,
386 (u64)tx_buffer->time_stamp,
388 if (i == tx_ring->next_to_use &&
389 i == tx_ring->next_to_clean)
391 else if (i == tx_ring->next_to_use)
393 else if (i == tx_ring->next_to_clean)
398 if (netif_msg_pktdata(adapter) &&
399 dma_unmap_len(tx_buffer, len) != 0)
400 print_hex_dump(KERN_INFO, "",
401 DUMP_PREFIX_ADDRESS, 16, 1,
402 phys_to_virt(dma_unmap_addr(tx_buffer,
404 dma_unmap_len(tx_buffer, len),
409 /* Print RX Rings Summary */
411 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
412 pr_info("Queue [NTU] [NTC]\n");
413 for (n = 0; n < adapter->num_rx_queues; n++) {
414 rx_ring = adapter->rx_ring[n];
415 pr_info("%5d %5X %5X\n",
416 n, rx_ring->next_to_use, rx_ring->next_to_clean);
420 if (!netif_msg_rx_status(adapter))
423 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
425 /* Advanced Receive Descriptor (Read) Format
427 * +-----------------------------------------------------+
428 * 0 | Packet Buffer Address [63:1] |A0/NSE|
429 * +----------------------------------------------+------+
430 * 8 | Header Buffer Address [63:1] | DD |
431 * +-----------------------------------------------------+
434 * Advanced Receive Descriptor (Write-Back) Format
436 * 63 48 47 32 31 30 21 20 16 15 4 3 0
437 * +------------------------------------------------------+
438 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
439 * | Checksum Ident | | | | Type | Type |
440 * +------------------------------------------------------+
441 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
442 * +------------------------------------------------------+
443 * 63 48 47 32 31 20 19 0
445 for (n = 0; n < adapter->num_rx_queues; n++) {
446 rx_ring = adapter->rx_ring[n];
447 pr_info("------------------------------------\n");
448 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
449 pr_info("------------------------------------\n");
450 pr_info("R [desc] [ PktBuf A0] "
451 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
452 "<-- Adv Rx Read format\n");
453 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
454 "[vl er S cks ln] ---------------- [bi->skb] "
455 "<-- Adv Rx Write-Back format\n");
457 for (i = 0; i < rx_ring->count; i++) {
458 rx_buffer_info = &rx_ring->rx_buffer_info[i];
459 rx_desc = IXGBE_RX_DESC(rx_ring, i);
460 u0 = (struct my_u0 *)rx_desc;
461 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
462 if (staterr & IXGBE_RXD_STAT_DD) {
463 /* Descriptor Done */
464 pr_info("RWB[0x%03X] %016llX "
465 "%016llX ---------------- %p", i,
468 rx_buffer_info->skb);
470 pr_info("R [0x%03X] %016llX "
471 "%016llX %016llX %p", i,
474 (u64)rx_buffer_info->dma,
475 rx_buffer_info->skb);
477 if (netif_msg_pktdata(adapter)) {
478 print_hex_dump(KERN_INFO, "",
479 DUMP_PREFIX_ADDRESS, 16, 1,
480 phys_to_virt(rx_buffer_info->dma),
481 ixgbe_rx_bufsz(rx_ring), true);
485 if (i == rx_ring->next_to_use)
487 else if (i == rx_ring->next_to_clean)
499 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
503 /* Let firmware take over control of h/w */
504 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
505 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
506 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
509 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
513 /* Let firmware know the driver has taken over */
514 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
515 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
516 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
520 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
521 * @adapter: pointer to adapter struct
522 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
523 * @queue: queue to map the corresponding interrupt to
524 * @msix_vector: the vector to map to the corresponding queue
527 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
528 u8 queue, u8 msix_vector)
531 struct ixgbe_hw *hw = &adapter->hw;
532 switch (hw->mac.type) {
533 case ixgbe_mac_82598EB:
534 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
537 index = (((direction * 64) + queue) >> 2) & 0x1F;
538 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
539 ivar &= ~(0xFF << (8 * (queue & 0x3)));
540 ivar |= (msix_vector << (8 * (queue & 0x3)));
541 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
543 case ixgbe_mac_82599EB:
545 if (direction == -1) {
547 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
548 index = ((queue & 1) * 8);
549 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
550 ivar &= ~(0xFF << index);
551 ivar |= (msix_vector << index);
552 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
555 /* tx or rx causes */
556 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
557 index = ((16 * (queue & 1)) + (8 * direction));
558 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
559 ivar &= ~(0xFF << index);
560 ivar |= (msix_vector << index);
561 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
569 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
574 switch (adapter->hw.mac.type) {
575 case ixgbe_mac_82598EB:
576 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
577 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
579 case ixgbe_mac_82599EB:
581 mask = (qmask & 0xFFFFFFFF);
582 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
583 mask = (qmask >> 32);
584 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
591 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
592 struct ixgbe_tx_buffer *tx_buffer)
594 if (tx_buffer->skb) {
595 dev_kfree_skb_any(tx_buffer->skb);
596 if (dma_unmap_len(tx_buffer, len))
597 dma_unmap_single(ring->dev,
598 dma_unmap_addr(tx_buffer, dma),
599 dma_unmap_len(tx_buffer, len),
601 } else if (dma_unmap_len(tx_buffer, len)) {
602 dma_unmap_page(ring->dev,
603 dma_unmap_addr(tx_buffer, dma),
604 dma_unmap_len(tx_buffer, len),
607 tx_buffer->next_to_watch = NULL;
608 tx_buffer->skb = NULL;
609 dma_unmap_len_set(tx_buffer, len, 0);
610 /* tx_buffer must be completely set up in the transmit path */
613 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
615 struct ixgbe_hw *hw = &adapter->hw;
616 struct ixgbe_hw_stats *hwstats = &adapter->stats;
621 if ((hw->fc.current_mode == ixgbe_fc_full) ||
622 (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
623 switch (hw->mac.type) {
624 case ixgbe_mac_82598EB:
625 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
628 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
630 hwstats->lxoffrxc += data;
632 /* refill credits (no tx hang) if we received xoff */
636 for (i = 0; i < adapter->num_tx_queues; i++)
637 clear_bit(__IXGBE_HANG_CHECK_ARMED,
638 &adapter->tx_ring[i]->state);
640 } else if (!(adapter->dcb_cfg.pfc_mode_enable))
643 /* update stats for each tc, only valid with PFC enabled */
644 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
645 switch (hw->mac.type) {
646 case ixgbe_mac_82598EB:
647 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
650 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
652 hwstats->pxoffrxc[i] += xoff[i];
655 /* disarm tx queues that have received xoff frames */
656 for (i = 0; i < adapter->num_tx_queues; i++) {
657 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
658 u8 tc = tx_ring->dcb_tc;
661 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
665 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
667 return ring->stats.packets;
670 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
672 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
673 struct ixgbe_hw *hw = &adapter->hw;
675 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
676 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
679 return (head < tail) ?
680 tail - head : (tail + ring->count - head);
685 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
687 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
688 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
689 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
692 clear_check_for_tx_hang(tx_ring);
695 * Check for a hung queue, but be thorough. This verifies
696 * that a transmit has been completed since the previous
697 * check AND there is at least one packet pending. The
698 * ARMED bit is set to indicate a potential hang. The
699 * bit is cleared if a pause frame is received to remove
700 * false hang detection due to PFC or 802.3x frames. By
701 * requiring this to fail twice we avoid races with
702 * pfc clearing the ARMED bit and conditions where we
703 * run the check_tx_hang logic with a transmit completion
704 * pending but without time to complete it yet.
706 if ((tx_done_old == tx_done) && tx_pending) {
707 /* make sure it is true for two checks in a row */
708 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
711 /* update completed stats and continue */
712 tx_ring->tx_stats.tx_done_old = tx_done;
713 /* reset the countdown */
714 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
721 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
722 * @adapter: driver private struct
724 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
727 /* Do the reset outside of interrupt context */
728 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
729 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
730 ixgbe_service_event_schedule(adapter);
735 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
736 * @q_vector: structure containing interrupt and ring information
737 * @tx_ring: tx ring to clean
739 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
740 struct ixgbe_ring *tx_ring)
742 struct ixgbe_adapter *adapter = q_vector->adapter;
743 struct ixgbe_tx_buffer *tx_buffer;
744 union ixgbe_adv_tx_desc *tx_desc;
745 unsigned int total_bytes = 0, total_packets = 0;
746 unsigned int budget = q_vector->tx.work_limit;
747 unsigned int i = tx_ring->next_to_clean;
749 if (test_bit(__IXGBE_DOWN, &adapter->state))
752 tx_buffer = &tx_ring->tx_buffer_info[i];
753 tx_desc = IXGBE_TX_DESC(tx_ring, i);
757 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
759 /* if next_to_watch is not set then there is no work pending */
763 /* prevent any other reads prior to eop_desc */
766 /* if DD is not set pending work has not been completed */
767 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
770 /* clear next_to_watch to prevent false hangs */
771 tx_buffer->next_to_watch = NULL;
773 /* update the statistics for this packet */
774 total_bytes += tx_buffer->bytecount;
775 total_packets += tx_buffer->gso_segs;
778 dev_kfree_skb_any(tx_buffer->skb);
780 /* unmap skb header data */
781 dma_unmap_single(tx_ring->dev,
782 dma_unmap_addr(tx_buffer, dma),
783 dma_unmap_len(tx_buffer, len),
786 /* clear tx_buffer data */
787 tx_buffer->skb = NULL;
788 dma_unmap_len_set(tx_buffer, len, 0);
790 /* unmap remaining buffers */
791 while (tx_desc != eop_desc) {
797 tx_buffer = tx_ring->tx_buffer_info;
798 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
801 /* unmap any remaining paged data */
802 if (dma_unmap_len(tx_buffer, len)) {
803 dma_unmap_page(tx_ring->dev,
804 dma_unmap_addr(tx_buffer, dma),
805 dma_unmap_len(tx_buffer, len),
807 dma_unmap_len_set(tx_buffer, len, 0);
811 /* move us one more past the eop_desc for start of next pkt */
817 tx_buffer = tx_ring->tx_buffer_info;
818 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
821 /* issue prefetch for next Tx descriptor */
824 /* update budget accounting */
826 } while (likely(budget));
829 tx_ring->next_to_clean = i;
830 u64_stats_update_begin(&tx_ring->syncp);
831 tx_ring->stats.bytes += total_bytes;
832 tx_ring->stats.packets += total_packets;
833 u64_stats_update_end(&tx_ring->syncp);
834 q_vector->tx.total_bytes += total_bytes;
835 q_vector->tx.total_packets += total_packets;
837 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
838 /* schedule immediate reset if we believe we hung */
839 struct ixgbe_hw *hw = &adapter->hw;
840 e_err(drv, "Detected Tx Unit Hang\n"
842 " TDH, TDT <%x>, <%x>\n"
843 " next_to_use <%x>\n"
844 " next_to_clean <%x>\n"
845 "tx_buffer_info[next_to_clean]\n"
846 " time_stamp <%lx>\n"
848 tx_ring->queue_index,
849 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
850 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
851 tx_ring->next_to_use, i,
852 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
854 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
857 "tx hang %d detected on queue %d, resetting adapter\n",
858 adapter->tx_timeout_count + 1, tx_ring->queue_index);
860 /* schedule immediate reset if we believe we hung */
861 ixgbe_tx_timeout_reset(adapter);
863 /* the adapter is about to reset, no point in enabling stuff */
867 netdev_tx_completed_queue(txring_txq(tx_ring),
868 total_packets, total_bytes);
870 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
871 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
872 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
873 /* Make sure that anybody stopping the queue after this
874 * sees the new next_to_clean.
877 if (__netif_subqueue_stopped(tx_ring->netdev,
878 tx_ring->queue_index)
879 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
880 netif_wake_subqueue(tx_ring->netdev,
881 tx_ring->queue_index);
882 ++tx_ring->tx_stats.restart_queue;
889 #ifdef CONFIG_IXGBE_DCA
890 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
891 struct ixgbe_ring *tx_ring,
894 struct ixgbe_hw *hw = &adapter->hw;
895 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
898 switch (hw->mac.type) {
899 case ixgbe_mac_82598EB:
900 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
902 case ixgbe_mac_82599EB:
904 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
905 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
908 /* for unknown hardware do not write register */
913 * We can enable relaxed ordering for reads, but not writes when
914 * DCA is enabled. This is due to a known issue in some chipsets
915 * which will cause the DCA tag to be cleared.
917 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
918 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
919 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
921 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
924 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
925 struct ixgbe_ring *rx_ring,
928 struct ixgbe_hw *hw = &adapter->hw;
929 u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
930 u8 reg_idx = rx_ring->reg_idx;
933 switch (hw->mac.type) {
934 case ixgbe_mac_82599EB:
936 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
943 * We can enable relaxed ordering for reads, but not writes when
944 * DCA is enabled. This is due to a known issue in some chipsets
945 * which will cause the DCA tag to be cleared.
947 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
948 IXGBE_DCA_RXCTRL_DATA_DCA_EN |
949 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
951 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
954 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
956 struct ixgbe_adapter *adapter = q_vector->adapter;
957 struct ixgbe_ring *ring;
960 if (q_vector->cpu == cpu)
963 ixgbe_for_each_ring(ring, q_vector->tx)
964 ixgbe_update_tx_dca(adapter, ring, cpu);
966 ixgbe_for_each_ring(ring, q_vector->rx)
967 ixgbe_update_rx_dca(adapter, ring, cpu);
974 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
979 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
982 /* always use CB2 mode, difference is masked in the CB driver */
983 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
985 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
986 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
990 for (i = 0; i < num_q_vectors; i++) {
991 adapter->q_vector[i]->cpu = -1;
992 ixgbe_update_dca(adapter->q_vector[i]);
996 static int __ixgbe_notify_dca(struct device *dev, void *data)
998 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
999 unsigned long event = *(unsigned long *)data;
1001 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1005 case DCA_PROVIDER_ADD:
1006 /* if we're already enabled, don't do it again */
1007 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1009 if (dca_add_requester(dev) == 0) {
1010 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1011 ixgbe_setup_dca(adapter);
1014 /* Fall Through since DCA is disabled. */
1015 case DCA_PROVIDER_REMOVE:
1016 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1017 dca_remove_requester(dev);
1018 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1019 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1027 #endif /* CONFIG_IXGBE_DCA */
1028 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1029 union ixgbe_adv_rx_desc *rx_desc,
1030 struct sk_buff *skb)
1032 if (ring->netdev->features & NETIF_F_RXHASH)
1033 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1038 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1039 * @adapter: address of board private structure
1040 * @rx_desc: advanced rx descriptor
1042 * Returns : true if it is FCoE pkt
1044 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_adapter *adapter,
1045 union ixgbe_adv_rx_desc *rx_desc)
1047 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1049 return (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
1050 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1051 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1052 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1055 #endif /* IXGBE_FCOE */
1057 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1058 * @ring: structure containing ring specific data
1059 * @rx_desc: current Rx descriptor being processed
1060 * @skb: skb currently being received and modified
1062 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1063 union ixgbe_adv_rx_desc *rx_desc,
1064 struct sk_buff *skb)
1066 skb_checksum_none_assert(skb);
1068 /* Rx csum disabled */
1069 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1072 /* if IP and error */
1073 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1074 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1075 ring->rx_stats.csum_err++;
1079 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1082 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1083 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1086 * 82599 errata, UDP frames with a 0 checksum can be marked as
1089 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1090 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1093 ring->rx_stats.csum_err++;
1097 /* It must be a TCP or UDP packet with a valid checksum */
1098 skb->ip_summed = CHECKSUM_UNNECESSARY;
1101 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1103 rx_ring->next_to_use = val;
1105 /* update next to alloc since we have filled the ring */
1106 rx_ring->next_to_alloc = val;
1108 * Force memory writes to complete before letting h/w
1109 * know there are new descriptors to fetch. (Only
1110 * applicable for weak-ordered memory model archs,
1114 writel(val, rx_ring->tail);
1117 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1118 struct ixgbe_rx_buffer *bi)
1120 struct page *page = bi->page;
1121 dma_addr_t dma = bi->dma;
1123 /* since we are recycling buffers we should seldom need to alloc */
1127 /* alloc new page for storage */
1128 if (likely(!page)) {
1129 page = alloc_pages(GFP_ATOMIC | __GFP_COLD,
1130 ixgbe_rx_pg_order(rx_ring));
1131 if (unlikely(!page)) {
1132 rx_ring->rx_stats.alloc_rx_page_failed++;
1138 /* map page for use */
1139 dma = dma_map_page(rx_ring->dev, page, 0,
1140 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1143 * if mapping failed free memory back to system since
1144 * there isn't much point in holding memory we can't use
1146 if (dma_mapping_error(rx_ring->dev, dma)) {
1150 rx_ring->rx_stats.alloc_rx_page_failed++;
1155 bi->page_offset ^= ixgbe_rx_bufsz(rx_ring);
1161 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1162 * @rx_ring: ring to place buffers on
1163 * @cleaned_count: number of buffers to replace
1165 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1167 union ixgbe_adv_rx_desc *rx_desc;
1168 struct ixgbe_rx_buffer *bi;
1169 u16 i = rx_ring->next_to_use;
1175 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1176 bi = &rx_ring->rx_buffer_info[i];
1177 i -= rx_ring->count;
1180 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1184 * Refresh the desc even if buffer_addrs didn't change
1185 * because each write-back erases this info.
1187 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1193 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1194 bi = rx_ring->rx_buffer_info;
1195 i -= rx_ring->count;
1198 /* clear the hdr_addr for the next_to_use descriptor */
1199 rx_desc->read.hdr_addr = 0;
1202 } while (cleaned_count);
1204 i += rx_ring->count;
1206 if (rx_ring->next_to_use != i)
1207 ixgbe_release_rx_desc(rx_ring, i);
1211 * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1212 * @data: pointer to the start of the headers
1213 * @max_len: total length of section to find headers in
1215 * This function is meant to determine the length of headers that will
1216 * be recognized by hardware for LRO, GRO, and RSC offloads. The main
1217 * motivation of doing this is to only perform one pull for IPv4 TCP
1218 * packets so that we can do basic things like calculating the gso_size
1219 * based on the average data per packet.
1221 static unsigned int ixgbe_get_headlen(unsigned char *data,
1222 unsigned int max_len)
1225 unsigned char *network;
1228 struct vlan_hdr *vlan;
1233 u8 nexthdr = 0; /* default to not TCP */
1236 /* this should never happen, but better safe than sorry */
1237 if (max_len < ETH_HLEN)
1240 /* initialize network frame pointer */
1243 /* set first protocol and move network header forward */
1244 protocol = hdr.eth->h_proto;
1245 hdr.network += ETH_HLEN;
1247 /* handle any vlan tag if present */
1248 if (protocol == __constant_htons(ETH_P_8021Q)) {
1249 if ((hdr.network - data) > (max_len - VLAN_HLEN))
1252 protocol = hdr.vlan->h_vlan_encapsulated_proto;
1253 hdr.network += VLAN_HLEN;
1256 /* handle L3 protocols */
1257 if (protocol == __constant_htons(ETH_P_IP)) {
1258 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1261 /* access ihl as a u8 to avoid unaligned access on ia64 */
1262 hlen = (hdr.network[0] & 0x0F) << 2;
1264 /* verify hlen meets minimum size requirements */
1265 if (hlen < sizeof(struct iphdr))
1266 return hdr.network - data;
1268 /* record next protocol */
1269 nexthdr = hdr.ipv4->protocol;
1270 hdr.network += hlen;
1272 } else if (protocol == __constant_htons(ETH_P_FCOE)) {
1273 if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1275 hdr.network += FCOE_HEADER_LEN;
1278 return hdr.network - data;
1281 /* finally sort out TCP */
1282 if (nexthdr == IPPROTO_TCP) {
1283 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1286 /* access doff as a u8 to avoid unaligned access on ia64 */
1287 hlen = (hdr.network[12] & 0xF0) >> 2;
1289 /* verify hlen meets minimum size requirements */
1290 if (hlen < sizeof(struct tcphdr))
1291 return hdr.network - data;
1293 hdr.network += hlen;
1297 * If everything has gone correctly hdr.network should be the
1298 * data section of the packet and will be the end of the header.
1299 * If not then it probably represents the end of the last recognized
1302 if ((hdr.network - data) < max_len)
1303 return hdr.network - data;
1308 static void ixgbe_get_rsc_cnt(struct ixgbe_ring *rx_ring,
1309 union ixgbe_adv_rx_desc *rx_desc,
1310 struct sk_buff *skb)
1315 if (!ring_is_rsc_enabled(rx_ring))
1318 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1319 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1321 /* If this is an RSC frame rsc_cnt should be non-zero */
1325 rsc_cnt = le32_to_cpu(rsc_enabled);
1326 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1328 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1331 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1332 struct sk_buff *skb)
1334 u16 hdr_len = skb_headlen(skb);
1336 /* set gso_size to avoid messing up TCP MSS */
1337 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1338 IXGBE_CB(skb)->append_cnt);
1341 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1342 struct sk_buff *skb)
1344 /* if append_cnt is 0 then frame is not RSC */
1345 if (!IXGBE_CB(skb)->append_cnt)
1348 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1349 rx_ring->rx_stats.rsc_flush++;
1351 ixgbe_set_rsc_gso_size(rx_ring, skb);
1353 /* gso_size is computed using append_cnt so always clear it last */
1354 IXGBE_CB(skb)->append_cnt = 0;
1358 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1359 * @rx_ring: rx descriptor ring packet is being transacted on
1360 * @rx_desc: pointer to the EOP Rx descriptor
1361 * @skb: pointer to current skb being populated
1363 * This function checks the ring, descriptor, and packet information in
1364 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1365 * other fields within the skb.
1367 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1368 union ixgbe_adv_rx_desc *rx_desc,
1369 struct sk_buff *skb)
1371 ixgbe_update_rsc_stats(rx_ring, skb);
1373 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1375 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1377 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1378 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1379 __vlan_hwaccel_put_tag(skb, vid);
1382 skb_record_rx_queue(skb, rx_ring->queue_index);
1384 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1387 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1388 struct sk_buff *skb)
1390 struct ixgbe_adapter *adapter = q_vector->adapter;
1392 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1393 napi_gro_receive(&q_vector->napi, skb);
1399 * ixgbe_is_non_eop - process handling of non-EOP buffers
1400 * @rx_ring: Rx ring being processed
1401 * @rx_desc: Rx descriptor for current buffer
1402 * @skb: Current socket buffer containing buffer in progress
1404 * This function updates next to clean. If the buffer is an EOP buffer
1405 * this function exits returning false, otherwise it will place the
1406 * sk_buff in the next buffer to be chained and return true indicating
1407 * that this is in fact a non-EOP buffer.
1409 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1410 union ixgbe_adv_rx_desc *rx_desc,
1411 struct sk_buff *skb)
1413 u32 ntc = rx_ring->next_to_clean + 1;
1415 /* fetch, update, and store next to clean */
1416 ntc = (ntc < rx_ring->count) ? ntc : 0;
1417 rx_ring->next_to_clean = ntc;
1419 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1421 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1424 /* append_cnt indicates packet is RSC, if so fetch nextp */
1425 if (IXGBE_CB(skb)->append_cnt) {
1426 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1427 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1428 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1431 /* place skb in next buffer to be received */
1432 rx_ring->rx_buffer_info[ntc].skb = skb;
1433 rx_ring->rx_stats.non_eop_descs++;
1439 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1440 * @rx_ring: rx descriptor ring packet is being transacted on
1441 * @rx_desc: pointer to the EOP Rx descriptor
1442 * @skb: pointer to current skb being fixed
1444 * Check for corrupted packet headers caused by senders on the local L2
1445 * embedded NIC switch not setting up their Tx Descriptors right. These
1446 * should be very rare.
1448 * Also address the case where we are pulling data in on pages only
1449 * and as such no data is present in the skb header.
1451 * In addition if skb is not at least 60 bytes we need to pad it so that
1452 * it is large enough to qualify as a valid Ethernet frame.
1454 * Returns true if an error was encountered and skb was freed.
1456 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1457 union ixgbe_adv_rx_desc *rx_desc,
1458 struct sk_buff *skb)
1460 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1461 struct net_device *netdev = rx_ring->netdev;
1463 unsigned int pull_len;
1465 /* if the page was released unmap it, else just sync our portion */
1466 if (unlikely(IXGBE_CB(skb)->page_released)) {
1467 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1468 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1469 IXGBE_CB(skb)->page_released = false;
1471 dma_sync_single_range_for_cpu(rx_ring->dev,
1474 ixgbe_rx_bufsz(rx_ring),
1477 IXGBE_CB(skb)->dma = 0;
1479 /* verify that the packet does not have any known errors */
1480 if (unlikely(ixgbe_test_staterr(rx_desc,
1481 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1482 !(netdev->features & NETIF_F_RXALL))) {
1483 dev_kfree_skb_any(skb);
1488 * it is valid to use page_address instead of kmap since we are
1489 * working with pages allocated out of the lomem pool per
1490 * alloc_page(GFP_ATOMIC)
1492 va = skb_frag_address(frag);
1495 * we need the header to contain the greater of either ETH_HLEN or
1496 * 60 bytes if the skb->len is less than 60 for skb_pad.
1498 pull_len = skb_frag_size(frag);
1500 pull_len = ixgbe_get_headlen(va, pull_len);
1502 /* align pull length to size of long to optimize memcpy performance */
1503 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1505 /* update all of the pointers */
1506 skb_frag_size_sub(frag, pull_len);
1507 frag->page_offset += pull_len;
1508 skb->data_len -= pull_len;
1509 skb->tail += pull_len;
1512 * if we sucked the frag empty then we should free it,
1513 * if there are other frags here something is screwed up in hardware
1515 if (skb_frag_size(frag) == 0) {
1516 BUG_ON(skb_shinfo(skb)->nr_frags != 1);
1517 skb_shinfo(skb)->nr_frags = 0;
1518 __skb_frag_unref(frag);
1519 skb->truesize -= ixgbe_rx_bufsz(rx_ring);
1522 /* if skb_pad returns an error the skb was freed */
1523 if (unlikely(skb->len < 60)) {
1524 int pad_len = 60 - skb->len;
1526 if (skb_pad(skb, pad_len))
1528 __skb_put(skb, pad_len);
1535 * ixgbe_can_reuse_page - determine if we can reuse a page
1536 * @rx_buffer: pointer to rx_buffer containing the page we want to reuse
1538 * Returns true if page can be reused in another Rx buffer
1540 static inline bool ixgbe_can_reuse_page(struct ixgbe_rx_buffer *rx_buffer)
1542 struct page *page = rx_buffer->page;
1544 /* if we are only owner of page and it is local we can reuse it */
1545 return likely(page_count(page) == 1) &&
1546 likely(page_to_nid(page) == numa_node_id());
1550 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1551 * @rx_ring: rx descriptor ring to store buffers on
1552 * @old_buff: donor buffer to have page reused
1554 * Syncronizes page for reuse by the adapter
1556 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1557 struct ixgbe_rx_buffer *old_buff)
1559 struct ixgbe_rx_buffer *new_buff;
1560 u16 nta = rx_ring->next_to_alloc;
1561 u16 bufsz = ixgbe_rx_bufsz(rx_ring);
1563 new_buff = &rx_ring->rx_buffer_info[nta];
1565 /* update, and store next to alloc */
1567 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1569 /* transfer page from old buffer to new buffer */
1570 new_buff->page = old_buff->page;
1571 new_buff->dma = old_buff->dma;
1573 /* flip page offset to other buffer and store to new_buff */
1574 new_buff->page_offset = old_buff->page_offset ^ bufsz;
1576 /* sync the buffer for use by the device */
1577 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1578 new_buff->page_offset, bufsz,
1581 /* bump ref count on page before it is given to the stack */
1582 get_page(new_buff->page);
1586 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1587 * @rx_ring: rx descriptor ring to transact packets on
1588 * @rx_buffer: buffer containing page to add
1589 * @rx_desc: descriptor containing length of buffer written by hardware
1590 * @skb: sk_buff to place the data into
1592 * This function is based on skb_add_rx_frag. I would have used that
1593 * function however it doesn't handle the truesize case correctly since we
1594 * are allocating more memory than might be used for a single receive.
1596 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1597 struct ixgbe_rx_buffer *rx_buffer,
1598 struct sk_buff *skb, int size)
1600 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1601 rx_buffer->page, rx_buffer->page_offset,
1604 skb->data_len += size;
1605 skb->truesize += ixgbe_rx_bufsz(rx_ring);
1609 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1610 * @q_vector: structure containing interrupt and ring information
1611 * @rx_ring: rx descriptor ring to transact packets on
1612 * @budget: Total limit on number of packets to process
1614 * This function provides a "bounce buffer" approach to Rx interrupt
1615 * processing. The advantage to this is that on systems that have
1616 * expensive overhead for IOMMU access this provides a means of avoiding
1617 * it by maintaining the mapping of the page to the syste.
1619 * Returns true if all work is completed without reaching budget
1621 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1622 struct ixgbe_ring *rx_ring,
1625 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1627 struct ixgbe_adapter *adapter = q_vector->adapter;
1629 #endif /* IXGBE_FCOE */
1630 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1633 struct ixgbe_rx_buffer *rx_buffer;
1634 union ixgbe_adv_rx_desc *rx_desc;
1635 struct sk_buff *skb;
1639 /* return some buffers to hardware, one at a time is too slow */
1640 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1641 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1645 ntc = rx_ring->next_to_clean;
1646 rx_desc = IXGBE_RX_DESC(rx_ring, ntc);
1647 rx_buffer = &rx_ring->rx_buffer_info[ntc];
1649 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
1653 * This memory barrier is needed to keep us from reading
1654 * any other fields out of the rx_desc until we know the
1655 * RXD_STAT_DD bit is set
1659 page = rx_buffer->page;
1662 skb = rx_buffer->skb;
1665 void *page_addr = page_address(page) +
1666 rx_buffer->page_offset;
1668 /* prefetch first cache line of first page */
1669 prefetch(page_addr);
1670 #if L1_CACHE_BYTES < 128
1671 prefetch(page_addr + L1_CACHE_BYTES);
1674 /* allocate a skb to store the frags */
1675 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1677 if (unlikely(!skb)) {
1678 rx_ring->rx_stats.alloc_rx_buff_failed++;
1683 * we will be copying header into skb->data in
1684 * pskb_may_pull so it is in our interest to prefetch
1685 * it now to avoid a possible cache miss
1687 prefetchw(skb->data);
1690 * Delay unmapping of the first packet. It carries the
1691 * header information, HW may still access the header
1692 * after the writeback. Only unmap it when EOP is
1695 IXGBE_CB(skb)->dma = rx_buffer->dma;
1697 /* we are reusing so sync this buffer for CPU use */
1698 dma_sync_single_range_for_cpu(rx_ring->dev,
1700 rx_buffer->page_offset,
1701 ixgbe_rx_bufsz(rx_ring),
1705 /* pull page into skb */
1706 ixgbe_add_rx_frag(rx_ring, rx_buffer, skb,
1707 le16_to_cpu(rx_desc->wb.upper.length));
1709 if (ixgbe_can_reuse_page(rx_buffer)) {
1710 /* hand second half of page back to the ring */
1711 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1712 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1713 /* the page has been released from the ring */
1714 IXGBE_CB(skb)->page_released = true;
1716 /* we are not reusing the buffer so unmap it */
1717 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1718 ixgbe_rx_pg_size(rx_ring),
1722 /* clear contents of buffer_info */
1723 rx_buffer->skb = NULL;
1725 rx_buffer->page = NULL;
1727 ixgbe_get_rsc_cnt(rx_ring, rx_desc, skb);
1731 /* place incomplete frames back on ring for completion */
1732 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
1735 /* verify the packet layout is correct */
1736 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
1739 /* probably a little skewed due to removing CRC */
1740 total_rx_bytes += skb->len;
1743 /* populate checksum, timestamp, VLAN, and protocol */
1744 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
1747 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1748 if (ixgbe_rx_is_fcoe(adapter, rx_desc)) {
1749 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1751 dev_kfree_skb_any(skb);
1756 #endif /* IXGBE_FCOE */
1757 ixgbe_rx_skb(q_vector, skb);
1759 /* update budget accounting */
1761 } while (likely(budget));
1764 /* include DDPed FCoE data */
1765 if (ddp_bytes > 0) {
1768 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1769 sizeof(struct fc_frame_header) -
1770 sizeof(struct fcoe_crc_eof);
1773 total_rx_bytes += ddp_bytes;
1774 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1777 #endif /* IXGBE_FCOE */
1778 u64_stats_update_begin(&rx_ring->syncp);
1779 rx_ring->stats.packets += total_rx_packets;
1780 rx_ring->stats.bytes += total_rx_bytes;
1781 u64_stats_update_end(&rx_ring->syncp);
1782 q_vector->rx.total_packets += total_rx_packets;
1783 q_vector->rx.total_bytes += total_rx_bytes;
1786 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1792 * ixgbe_configure_msix - Configure MSI-X hardware
1793 * @adapter: board private structure
1795 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1798 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1800 struct ixgbe_q_vector *q_vector;
1801 int q_vectors, v_idx;
1804 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1806 /* Populate MSIX to EITR Select */
1807 if (adapter->num_vfs > 32) {
1808 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1809 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1813 * Populate the IVAR table and set the ITR values to the
1814 * corresponding register.
1816 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1817 struct ixgbe_ring *ring;
1818 q_vector = adapter->q_vector[v_idx];
1820 ixgbe_for_each_ring(ring, q_vector->rx)
1821 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1823 ixgbe_for_each_ring(ring, q_vector->tx)
1824 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1826 if (q_vector->tx.ring && !q_vector->rx.ring) {
1827 /* tx only vector */
1828 if (adapter->tx_itr_setting == 1)
1829 q_vector->itr = IXGBE_10K_ITR;
1831 q_vector->itr = adapter->tx_itr_setting;
1833 /* rx or rx/tx vector */
1834 if (adapter->rx_itr_setting == 1)
1835 q_vector->itr = IXGBE_20K_ITR;
1837 q_vector->itr = adapter->rx_itr_setting;
1840 ixgbe_write_eitr(q_vector);
1843 switch (adapter->hw.mac.type) {
1844 case ixgbe_mac_82598EB:
1845 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1848 case ixgbe_mac_82599EB:
1849 case ixgbe_mac_X540:
1850 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1855 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1857 /* set up to autoclear timer, and the vectors */
1858 mask = IXGBE_EIMS_ENABLE_MASK;
1859 mask &= ~(IXGBE_EIMS_OTHER |
1860 IXGBE_EIMS_MAILBOX |
1863 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1866 enum latency_range {
1870 latency_invalid = 255
1874 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1875 * @q_vector: structure containing interrupt and ring information
1876 * @ring_container: structure containing ring performance data
1878 * Stores a new ITR value based on packets and byte
1879 * counts during the last interrupt. The advantage of per interrupt
1880 * computation is faster updates and more accurate ITR for the current
1881 * traffic pattern. Constants in this function were computed
1882 * based on theoretical maximum wire speed and thresholds were set based
1883 * on testing data as well as attempting to minimize response time
1884 * while increasing bulk throughput.
1885 * this functionality is controlled by the InterruptThrottleRate module
1886 * parameter (see ixgbe_param.c)
1888 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1889 struct ixgbe_ring_container *ring_container)
1891 int bytes = ring_container->total_bytes;
1892 int packets = ring_container->total_packets;
1895 u8 itr_setting = ring_container->itr;
1900 /* simple throttlerate management
1901 * 0-10MB/s lowest (100000 ints/s)
1902 * 10-20MB/s low (20000 ints/s)
1903 * 20-1249MB/s bulk (8000 ints/s)
1905 /* what was last interrupt timeslice? */
1906 timepassed_us = q_vector->itr >> 2;
1907 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1909 switch (itr_setting) {
1910 case lowest_latency:
1911 if (bytes_perint > 10)
1912 itr_setting = low_latency;
1915 if (bytes_perint > 20)
1916 itr_setting = bulk_latency;
1917 else if (bytes_perint <= 10)
1918 itr_setting = lowest_latency;
1921 if (bytes_perint <= 20)
1922 itr_setting = low_latency;
1926 /* clear work counters since we have the values we need */
1927 ring_container->total_bytes = 0;
1928 ring_container->total_packets = 0;
1930 /* write updated itr to ring container */
1931 ring_container->itr = itr_setting;
1935 * ixgbe_write_eitr - write EITR register in hardware specific way
1936 * @q_vector: structure containing interrupt and ring information
1938 * This function is made to be called by ethtool and by the driver
1939 * when it needs to update EITR registers at runtime. Hardware
1940 * specific quirks/differences are taken care of here.
1942 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1944 struct ixgbe_adapter *adapter = q_vector->adapter;
1945 struct ixgbe_hw *hw = &adapter->hw;
1946 int v_idx = q_vector->v_idx;
1947 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1949 switch (adapter->hw.mac.type) {
1950 case ixgbe_mac_82598EB:
1951 /* must write high and low 16 bits to reset counter */
1952 itr_reg |= (itr_reg << 16);
1954 case ixgbe_mac_82599EB:
1955 case ixgbe_mac_X540:
1957 * set the WDIS bit to not clear the timer bits and cause an
1958 * immediate assertion of the interrupt
1960 itr_reg |= IXGBE_EITR_CNT_WDIS;
1965 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1968 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1970 u32 new_itr = q_vector->itr;
1973 ixgbe_update_itr(q_vector, &q_vector->tx);
1974 ixgbe_update_itr(q_vector, &q_vector->rx);
1976 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1978 switch (current_itr) {
1979 /* counts and packets in update_itr are dependent on these numbers */
1980 case lowest_latency:
1981 new_itr = IXGBE_100K_ITR;
1984 new_itr = IXGBE_20K_ITR;
1987 new_itr = IXGBE_8K_ITR;
1993 if (new_itr != q_vector->itr) {
1994 /* do an exponential smoothing */
1995 new_itr = (10 * new_itr * q_vector->itr) /
1996 ((9 * new_itr) + q_vector->itr);
1998 /* save the algorithm value here */
1999 q_vector->itr = new_itr;
2001 ixgbe_write_eitr(q_vector);
2006 * ixgbe_check_overtemp_subtask - check for over temperature
2007 * @adapter: pointer to adapter
2009 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2011 struct ixgbe_hw *hw = &adapter->hw;
2012 u32 eicr = adapter->interrupt_event;
2014 if (test_bit(__IXGBE_DOWN, &adapter->state))
2017 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2018 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2021 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2023 switch (hw->device_id) {
2024 case IXGBE_DEV_ID_82599_T3_LOM:
2026 * Since the warning interrupt is for both ports
2027 * we don't have to check if:
2028 * - This interrupt wasn't for our port.
2029 * - We may have missed the interrupt so always have to
2030 * check if we got a LSC
2032 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2033 !(eicr & IXGBE_EICR_LSC))
2036 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2038 bool link_up = false;
2040 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2046 /* Check if this is not due to overtemp */
2047 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2052 if (!(eicr & IXGBE_EICR_GPI_SDP0))
2057 "Network adapter has been stopped because it has over heated. "
2058 "Restart the computer. If the problem persists, "
2059 "power off the system and replace the adapter\n");
2061 adapter->interrupt_event = 0;
2064 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2066 struct ixgbe_hw *hw = &adapter->hw;
2068 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2069 (eicr & IXGBE_EICR_GPI_SDP1)) {
2070 e_crit(probe, "Fan has stopped, replace the adapter\n");
2071 /* write to clear the interrupt */
2072 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2076 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2078 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2081 switch (adapter->hw.mac.type) {
2082 case ixgbe_mac_82599EB:
2084 * Need to check link state so complete overtemp check
2087 if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2088 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2089 adapter->interrupt_event = eicr;
2090 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2091 ixgbe_service_event_schedule(adapter);
2095 case ixgbe_mac_X540:
2096 if (!(eicr & IXGBE_EICR_TS))
2104 "Network adapter has been stopped because it has over heated. "
2105 "Restart the computer. If the problem persists, "
2106 "power off the system and replace the adapter\n");
2109 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2111 struct ixgbe_hw *hw = &adapter->hw;
2113 if (eicr & IXGBE_EICR_GPI_SDP2) {
2114 /* Clear the interrupt */
2115 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2116 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2117 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2118 ixgbe_service_event_schedule(adapter);
2122 if (eicr & IXGBE_EICR_GPI_SDP1) {
2123 /* Clear the interrupt */
2124 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2125 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2126 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2127 ixgbe_service_event_schedule(adapter);
2132 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2134 struct ixgbe_hw *hw = &adapter->hw;
2137 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2138 adapter->link_check_timeout = jiffies;
2139 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2140 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2141 IXGBE_WRITE_FLUSH(hw);
2142 ixgbe_service_event_schedule(adapter);
2146 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2150 struct ixgbe_hw *hw = &adapter->hw;
2152 switch (hw->mac.type) {
2153 case ixgbe_mac_82598EB:
2154 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2155 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2157 case ixgbe_mac_82599EB:
2158 case ixgbe_mac_X540:
2159 mask = (qmask & 0xFFFFFFFF);
2161 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2162 mask = (qmask >> 32);
2164 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2169 /* skip the flush */
2172 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2176 struct ixgbe_hw *hw = &adapter->hw;
2178 switch (hw->mac.type) {
2179 case ixgbe_mac_82598EB:
2180 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2181 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2183 case ixgbe_mac_82599EB:
2184 case ixgbe_mac_X540:
2185 mask = (qmask & 0xFFFFFFFF);
2187 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2188 mask = (qmask >> 32);
2190 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2195 /* skip the flush */
2199 * ixgbe_irq_enable - Enable default interrupt generation settings
2200 * @adapter: board private structure
2202 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2205 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2207 /* don't reenable LSC while waiting for link */
2208 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2209 mask &= ~IXGBE_EIMS_LSC;
2211 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2212 switch (adapter->hw.mac.type) {
2213 case ixgbe_mac_82599EB:
2214 mask |= IXGBE_EIMS_GPI_SDP0;
2216 case ixgbe_mac_X540:
2217 mask |= IXGBE_EIMS_TS;
2222 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2223 mask |= IXGBE_EIMS_GPI_SDP1;
2224 switch (adapter->hw.mac.type) {
2225 case ixgbe_mac_82599EB:
2226 mask |= IXGBE_EIMS_GPI_SDP1;
2227 mask |= IXGBE_EIMS_GPI_SDP2;
2228 case ixgbe_mac_X540:
2229 mask |= IXGBE_EIMS_ECC;
2230 mask |= IXGBE_EIMS_MAILBOX;
2235 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2236 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2237 mask |= IXGBE_EIMS_FLOW_DIR;
2239 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2241 ixgbe_irq_enable_queues(adapter, ~0);
2243 IXGBE_WRITE_FLUSH(&adapter->hw);
2246 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2248 struct ixgbe_adapter *adapter = data;
2249 struct ixgbe_hw *hw = &adapter->hw;
2253 * Workaround for Silicon errata. Use clear-by-write instead
2254 * of clear-by-read. Reading with EICS will return the
2255 * interrupt causes without clearing, which later be done
2256 * with the write to EICR.
2258 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2259 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2261 if (eicr & IXGBE_EICR_LSC)
2262 ixgbe_check_lsc(adapter);
2264 if (eicr & IXGBE_EICR_MAILBOX)
2265 ixgbe_msg_task(adapter);
2267 switch (hw->mac.type) {
2268 case ixgbe_mac_82599EB:
2269 case ixgbe_mac_X540:
2270 if (eicr & IXGBE_EICR_ECC)
2271 e_info(link, "Received unrecoverable ECC Err, please "
2273 /* Handle Flow Director Full threshold interrupt */
2274 if (eicr & IXGBE_EICR_FLOW_DIR) {
2275 int reinit_count = 0;
2277 for (i = 0; i < adapter->num_tx_queues; i++) {
2278 struct ixgbe_ring *ring = adapter->tx_ring[i];
2279 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2284 /* no more flow director interrupts until after init */
2285 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2286 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2287 ixgbe_service_event_schedule(adapter);
2290 ixgbe_check_sfp_event(adapter, eicr);
2291 ixgbe_check_overtemp_event(adapter, eicr);
2297 ixgbe_check_fan_failure(adapter, eicr);
2299 /* re-enable the original interrupt state, no lsc, no queues */
2300 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2301 ixgbe_irq_enable(adapter, false, false);
2306 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2308 struct ixgbe_q_vector *q_vector = data;
2310 /* EIAM disabled interrupts (on this vector) for us */
2312 if (q_vector->rx.ring || q_vector->tx.ring)
2313 napi_schedule(&q_vector->napi);
2319 * ixgbe_poll - NAPI Rx polling callback
2320 * @napi: structure for representing this polling device
2321 * @budget: how many packets driver is allowed to clean
2323 * This function is used for legacy and MSI, NAPI mode
2325 int ixgbe_poll(struct napi_struct *napi, int budget)
2327 struct ixgbe_q_vector *q_vector =
2328 container_of(napi, struct ixgbe_q_vector, napi);
2329 struct ixgbe_adapter *adapter = q_vector->adapter;
2330 struct ixgbe_ring *ring;
2331 int per_ring_budget;
2332 bool clean_complete = true;
2334 #ifdef CONFIG_IXGBE_DCA
2335 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2336 ixgbe_update_dca(q_vector);
2339 ixgbe_for_each_ring(ring, q_vector->tx)
2340 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2342 /* attempt to distribute budget to each queue fairly, but don't allow
2343 * the budget to go below 1 because we'll exit polling */
2344 if (q_vector->rx.count > 1)
2345 per_ring_budget = max(budget/q_vector->rx.count, 1);
2347 per_ring_budget = budget;
2349 ixgbe_for_each_ring(ring, q_vector->rx)
2350 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
2353 /* If all work not completed, return budget and keep polling */
2354 if (!clean_complete)
2357 /* all work done, exit the polling mode */
2358 napi_complete(napi);
2359 if (adapter->rx_itr_setting & 1)
2360 ixgbe_set_itr(q_vector);
2361 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2362 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2368 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2369 * @adapter: board private structure
2371 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2372 * interrupts from the kernel.
2374 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2376 struct net_device *netdev = adapter->netdev;
2377 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2381 for (vector = 0; vector < q_vectors; vector++) {
2382 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2383 struct msix_entry *entry = &adapter->msix_entries[vector];
2385 if (q_vector->tx.ring && q_vector->rx.ring) {
2386 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2387 "%s-%s-%d", netdev->name, "TxRx", ri++);
2389 } else if (q_vector->rx.ring) {
2390 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2391 "%s-%s-%d", netdev->name, "rx", ri++);
2392 } else if (q_vector->tx.ring) {
2393 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2394 "%s-%s-%d", netdev->name, "tx", ti++);
2396 /* skip this unused q_vector */
2399 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2400 q_vector->name, q_vector);
2402 e_err(probe, "request_irq failed for MSIX interrupt "
2403 "Error: %d\n", err);
2404 goto free_queue_irqs;
2406 /* If Flow Director is enabled, set interrupt affinity */
2407 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2408 /* assign the mask for this irq */
2409 irq_set_affinity_hint(entry->vector,
2410 &q_vector->affinity_mask);
2414 err = request_irq(adapter->msix_entries[vector].vector,
2415 ixgbe_msix_other, 0, netdev->name, adapter);
2417 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2418 goto free_queue_irqs;
2426 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2428 free_irq(adapter->msix_entries[vector].vector,
2429 adapter->q_vector[vector]);
2431 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2432 pci_disable_msix(adapter->pdev);
2433 kfree(adapter->msix_entries);
2434 adapter->msix_entries = NULL;
2439 * ixgbe_intr - legacy mode Interrupt Handler
2440 * @irq: interrupt number
2441 * @data: pointer to a network interface device structure
2443 static irqreturn_t ixgbe_intr(int irq, void *data)
2445 struct ixgbe_adapter *adapter = data;
2446 struct ixgbe_hw *hw = &adapter->hw;
2447 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2451 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2452 * before the read of EICR.
2454 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2456 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2457 * therefore no explicit interrupt disable is necessary */
2458 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2461 * shared interrupt alert!
2462 * make sure interrupts are enabled because the read will
2463 * have disabled interrupts due to EIAM
2464 * finish the workaround of silicon errata on 82598. Unmask
2465 * the interrupt that we masked before the EICR read.
2467 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2468 ixgbe_irq_enable(adapter, true, true);
2469 return IRQ_NONE; /* Not our interrupt */
2472 if (eicr & IXGBE_EICR_LSC)
2473 ixgbe_check_lsc(adapter);
2475 switch (hw->mac.type) {
2476 case ixgbe_mac_82599EB:
2477 ixgbe_check_sfp_event(adapter, eicr);
2479 case ixgbe_mac_X540:
2480 if (eicr & IXGBE_EICR_ECC)
2481 e_info(link, "Received unrecoverable ECC err, please "
2483 ixgbe_check_overtemp_event(adapter, eicr);
2489 ixgbe_check_fan_failure(adapter, eicr);
2491 /* would disable interrupts here but EIAM disabled it */
2492 napi_schedule(&q_vector->napi);
2495 * re-enable link(maybe) and non-queue interrupts, no flush.
2496 * ixgbe_poll will re-enable the queue interrupts
2498 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2499 ixgbe_irq_enable(adapter, false, false);
2505 * ixgbe_request_irq - initialize interrupts
2506 * @adapter: board private structure
2508 * Attempts to configure interrupts using the best available
2509 * capabilities of the hardware and kernel.
2511 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2513 struct net_device *netdev = adapter->netdev;
2516 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2517 err = ixgbe_request_msix_irqs(adapter);
2518 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2519 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2520 netdev->name, adapter);
2522 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2523 netdev->name, adapter);
2526 e_err(probe, "request_irq failed, Error %d\n", err);
2531 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2533 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2536 q_vectors = adapter->num_msix_vectors;
2538 free_irq(adapter->msix_entries[i].vector, adapter);
2541 for (; i >= 0; i--) {
2542 /* free only the irqs that were actually requested */
2543 if (!adapter->q_vector[i]->rx.ring &&
2544 !adapter->q_vector[i]->tx.ring)
2547 /* clear the affinity_mask in the IRQ descriptor */
2548 irq_set_affinity_hint(adapter->msix_entries[i].vector,
2551 free_irq(adapter->msix_entries[i].vector,
2552 adapter->q_vector[i]);
2555 free_irq(adapter->pdev->irq, adapter);
2560 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2561 * @adapter: board private structure
2563 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2565 switch (adapter->hw.mac.type) {
2566 case ixgbe_mac_82598EB:
2567 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2569 case ixgbe_mac_82599EB:
2570 case ixgbe_mac_X540:
2571 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2572 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2573 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2578 IXGBE_WRITE_FLUSH(&adapter->hw);
2579 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2581 for (i = 0; i < adapter->num_msix_vectors; i++)
2582 synchronize_irq(adapter->msix_entries[i].vector);
2584 synchronize_irq(adapter->pdev->irq);
2589 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2592 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2594 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2597 if (adapter->rx_itr_setting == 1)
2598 q_vector->itr = IXGBE_20K_ITR;
2600 q_vector->itr = adapter->rx_itr_setting;
2602 ixgbe_write_eitr(q_vector);
2604 ixgbe_set_ivar(adapter, 0, 0, 0);
2605 ixgbe_set_ivar(adapter, 1, 0, 0);
2607 e_info(hw, "Legacy interrupt IVAR setup done\n");
2611 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2612 * @adapter: board private structure
2613 * @ring: structure containing ring specific data
2615 * Configure the Tx descriptor ring after a reset.
2617 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2618 struct ixgbe_ring *ring)
2620 struct ixgbe_hw *hw = &adapter->hw;
2621 u64 tdba = ring->dma;
2623 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2624 u8 reg_idx = ring->reg_idx;
2626 /* disable queue to avoid issues while updating state */
2627 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2628 IXGBE_WRITE_FLUSH(hw);
2630 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2631 (tdba & DMA_BIT_MASK(32)));
2632 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2633 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2634 ring->count * sizeof(union ixgbe_adv_tx_desc));
2635 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2636 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2637 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2640 * set WTHRESH to encourage burst writeback, it should not be set
2641 * higher than 1 when ITR is 0 as it could cause false TX hangs
2643 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2644 * to or less than the number of on chip descriptors, which is
2647 if (!ring->q_vector || (ring->q_vector->itr < 8))
2648 txdctl |= (1 << 16); /* WTHRESH = 1 */
2650 txdctl |= (8 << 16); /* WTHRESH = 8 */
2653 * Setting PTHRESH to 32 both improves performance
2654 * and avoids a TX hang with DFP enabled
2656 txdctl |= (1 << 8) | /* HTHRESH = 1 */
2657 32; /* PTHRESH = 32 */
2659 /* reinitialize flowdirector state */
2660 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2661 adapter->atr_sample_rate) {
2662 ring->atr_sample_rate = adapter->atr_sample_rate;
2663 ring->atr_count = 0;
2664 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2666 ring->atr_sample_rate = 0;
2669 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2672 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2674 netdev_tx_reset_queue(txring_txq(ring));
2676 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2677 if (hw->mac.type == ixgbe_mac_82598EB &&
2678 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2681 /* poll to verify queue is enabled */
2683 usleep_range(1000, 2000);
2684 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2685 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2687 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2690 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2692 struct ixgbe_hw *hw = &adapter->hw;
2695 u8 tcs = netdev_get_num_tc(adapter->netdev);
2697 if (hw->mac.type == ixgbe_mac_82598EB)
2700 /* disable the arbiter while setting MTQC */
2701 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2702 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2703 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2705 /* set transmit pool layout */
2706 switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2707 case (IXGBE_FLAG_SRIOV_ENABLED):
2708 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2709 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2713 reg = IXGBE_MTQC_64Q_1PB;
2715 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2717 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2719 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2721 /* Enable Security TX Buffer IFG for multiple pb */
2723 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2724 reg |= IXGBE_SECTX_DCB;
2725 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2730 /* re-enable the arbiter */
2731 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2732 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2736 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2737 * @adapter: board private structure
2739 * Configure the Tx unit of the MAC after a reset.
2741 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2743 struct ixgbe_hw *hw = &adapter->hw;
2747 ixgbe_setup_mtqc(adapter);
2749 if (hw->mac.type != ixgbe_mac_82598EB) {
2750 /* DMATXCTL.EN must be before Tx queues are enabled */
2751 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2752 dmatxctl |= IXGBE_DMATXCTL_TE;
2753 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2756 /* Setup the HW Tx Head and Tail descriptor pointers */
2757 for (i = 0; i < adapter->num_tx_queues; i++)
2758 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2761 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2763 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2764 struct ixgbe_ring *rx_ring)
2767 u8 reg_idx = rx_ring->reg_idx;
2769 switch (adapter->hw.mac.type) {
2770 case ixgbe_mac_82598EB: {
2771 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2772 const int mask = feature[RING_F_RSS].mask;
2773 reg_idx = reg_idx & mask;
2776 case ixgbe_mac_82599EB:
2777 case ixgbe_mac_X540:
2782 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2784 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2785 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2786 if (adapter->num_vfs)
2787 srrctl |= IXGBE_SRRCTL_DROP_EN;
2789 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2790 IXGBE_SRRCTL_BSIZEHDR_MASK;
2792 #if PAGE_SIZE > IXGBE_MAX_RXBUFFER
2793 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2795 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2797 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2799 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2802 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2804 struct ixgbe_hw *hw = &adapter->hw;
2805 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2806 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2807 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2808 u32 mrqc = 0, reta = 0;
2811 u8 tcs = netdev_get_num_tc(adapter->netdev);
2812 int maxq = adapter->ring_feature[RING_F_RSS].indices;
2815 maxq = min(maxq, adapter->num_tx_queues / tcs);
2817 /* Fill out hash function seeds */
2818 for (i = 0; i < 10; i++)
2819 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2821 /* Fill out redirection table */
2822 for (i = 0, j = 0; i < 128; i++, j++) {
2825 /* reta = 4-byte sliding window of
2826 * 0x00..(indices-1)(indices-1)00..etc. */
2827 reta = (reta << 8) | (j * 0x11);
2829 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2832 /* Disable indicating checksum in descriptor, enables RSS hash */
2833 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2834 rxcsum |= IXGBE_RXCSUM_PCSD;
2835 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2837 if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2838 (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2839 mrqc = IXGBE_MRQC_RSSEN;
2841 int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2842 | IXGBE_FLAG_SRIOV_ENABLED);
2845 case (IXGBE_FLAG_RSS_ENABLED):
2847 mrqc = IXGBE_MRQC_RSSEN;
2849 mrqc = IXGBE_MRQC_RTRSS4TCEN;
2851 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2853 case (IXGBE_FLAG_SRIOV_ENABLED):
2854 mrqc = IXGBE_MRQC_VMDQEN;
2861 /* Perform hash on these packet types */
2862 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2863 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2864 | IXGBE_MRQC_RSS_FIELD_IPV6
2865 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2867 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
2868 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
2869 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
2870 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2872 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2876 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2877 * @adapter: address of board private structure
2878 * @index: index of ring to set
2880 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2881 struct ixgbe_ring *ring)
2883 struct ixgbe_hw *hw = &adapter->hw;
2885 u8 reg_idx = ring->reg_idx;
2887 if (!ring_is_rsc_enabled(ring))
2890 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2891 rscctrl |= IXGBE_RSCCTL_RSCEN;
2893 * we must limit the number of descriptors so that the
2894 * total size of max desc * buf_len is not greater
2897 #if (PAGE_SIZE <= 8192)
2898 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2899 #elif (PAGE_SIZE <= 16384)
2900 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2902 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2904 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2908 * ixgbe_set_uta - Set unicast filter table address
2909 * @adapter: board private structure
2911 * The unicast table address is a register array of 32-bit registers.
2912 * The table is meant to be used in a way similar to how the MTA is used
2913 * however due to certain limitations in the hardware it is necessary to
2914 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2915 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
2917 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2919 struct ixgbe_hw *hw = &adapter->hw;
2922 /* The UTA table only exists on 82599 hardware and newer */
2923 if (hw->mac.type < ixgbe_mac_82599EB)
2926 /* we only need to do this if VMDq is enabled */
2927 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2930 for (i = 0; i < 128; i++)
2931 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2934 #define IXGBE_MAX_RX_DESC_POLL 10
2935 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2936 struct ixgbe_ring *ring)
2938 struct ixgbe_hw *hw = &adapter->hw;
2939 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2941 u8 reg_idx = ring->reg_idx;
2943 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2944 if (hw->mac.type == ixgbe_mac_82598EB &&
2945 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2949 usleep_range(1000, 2000);
2950 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2951 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2954 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2955 "the polling period\n", reg_idx);
2959 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
2960 struct ixgbe_ring *ring)
2962 struct ixgbe_hw *hw = &adapter->hw;
2963 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2965 u8 reg_idx = ring->reg_idx;
2967 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2968 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
2970 /* write value back with RXDCTL.ENABLE bit cleared */
2971 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2973 if (hw->mac.type == ixgbe_mac_82598EB &&
2974 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2977 /* the hardware may take up to 100us to really disable the rx queue */
2980 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2981 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
2984 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
2985 "the polling period\n", reg_idx);
2989 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2990 struct ixgbe_ring *ring)
2992 struct ixgbe_hw *hw = &adapter->hw;
2993 u64 rdba = ring->dma;
2995 u8 reg_idx = ring->reg_idx;
2997 /* disable queue to avoid issues while updating state */
2998 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2999 ixgbe_disable_rx_queue(adapter, ring);
3001 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3002 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3003 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3004 ring->count * sizeof(union ixgbe_adv_rx_desc));
3005 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3006 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3007 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3009 ixgbe_configure_srrctl(adapter, ring);
3010 ixgbe_configure_rscctl(adapter, ring);
3012 /* If operating in IOV mode set RLPML for X540 */
3013 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3014 hw->mac.type == ixgbe_mac_X540) {
3015 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3016 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3017 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3020 if (hw->mac.type == ixgbe_mac_82598EB) {
3022 * enable cache line friendly hardware writes:
3023 * PTHRESH=32 descriptors (half the internal cache),
3024 * this also removes ugly rx_no_buffer_count increment
3025 * HTHRESH=4 descriptors (to minimize latency on fetch)
3026 * WTHRESH=8 burst writeback up to two cache lines
3028 rxdctl &= ~0x3FFFFF;
3032 /* enable receive descriptor ring */
3033 rxdctl |= IXGBE_RXDCTL_ENABLE;
3034 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3036 ixgbe_rx_desc_queue_enable(adapter, ring);
3037 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3040 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3042 struct ixgbe_hw *hw = &adapter->hw;
3045 /* PSRTYPE must be initialized in non 82598 adapters */
3046 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3047 IXGBE_PSRTYPE_UDPHDR |
3048 IXGBE_PSRTYPE_IPV4HDR |
3049 IXGBE_PSRTYPE_L2HDR |
3050 IXGBE_PSRTYPE_IPV6HDR;
3052 if (hw->mac.type == ixgbe_mac_82598EB)
3055 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3056 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3058 for (p = 0; p < adapter->num_rx_pools; p++)
3059 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3063 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3065 struct ixgbe_hw *hw = &adapter->hw;
3068 u32 reg_offset, vf_shift;
3072 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3075 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3076 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3077 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3078 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3080 vf_shift = adapter->num_vfs % 32;
3081 reg_offset = (adapter->num_vfs >= 32) ? 1 : 0;
3083 /* Enable only the PF's pool for Tx/Rx */
3084 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3085 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3086 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3087 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3088 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3090 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3091 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3094 * Set up VF register offsets for selected VT Mode,
3095 * i.e. 32 or 64 VFs for SR-IOV
3097 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3098 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3099 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3100 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3102 /* enable Tx loopback for VF/PF communication */
3103 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3104 /* Enable MAC Anti-Spoofing */
3105 hw->mac.ops.set_mac_anti_spoofing(hw,
3106 (adapter->num_vfs != 0),
3108 /* For VFs that have spoof checking turned off */
3109 for (i = 0; i < adapter->num_vfs; i++) {
3110 if (!adapter->vfinfo[i].spoofchk_enabled)
3111 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3115 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3117 struct ixgbe_hw *hw = &adapter->hw;
3118 struct net_device *netdev = adapter->netdev;
3119 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3120 struct ixgbe_ring *rx_ring;
3125 /* adjust max frame to be able to do baby jumbo for FCoE */
3126 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3127 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3128 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3130 #endif /* IXGBE_FCOE */
3131 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3132 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3133 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3134 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3136 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3139 /* MHADD will allow an extra 4 bytes past for vlan tagged frames */
3140 max_frame += VLAN_HLEN;
3142 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3143 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3144 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3145 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3148 * Setup the HW Rx Head and Tail Descriptor Pointers and
3149 * the Base and Length of the Rx Descriptor Ring
3151 for (i = 0; i < adapter->num_rx_queues; i++) {
3152 rx_ring = adapter->rx_ring[i];
3153 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3154 set_ring_rsc_enabled(rx_ring);
3156 clear_ring_rsc_enabled(rx_ring);
3158 if (netdev->features & NETIF_F_FCOE_MTU) {
3159 struct ixgbe_ring_feature *f;
3160 f = &adapter->ring_feature[RING_F_FCOE];
3161 if ((i >= f->mask) && (i < f->mask + f->indices))
3162 set_bit(__IXGBE_RX_FCOE_BUFSZ, &rx_ring->state);
3164 #endif /* IXGBE_FCOE */
3168 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3170 struct ixgbe_hw *hw = &adapter->hw;
3171 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3173 switch (hw->mac.type) {
3174 case ixgbe_mac_82598EB:
3176 * For VMDq support of different descriptor types or
3177 * buffer sizes through the use of multiple SRRCTL
3178 * registers, RDRXCTL.MVMEN must be set to 1
3180 * also, the manual doesn't mention it clearly but DCA hints
3181 * will only use queue 0's tags unless this bit is set. Side
3182 * effects of setting this bit are only that SRRCTL must be
3183 * fully programmed [0..15]
3185 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3187 case ixgbe_mac_82599EB:
3188 case ixgbe_mac_X540:
3189 /* Disable RSC for ACK packets */
3190 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3191 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3192 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3193 /* hardware requires some bits to be set by default */
3194 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3195 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3198 /* We should do nothing since we don't know this hardware */
3202 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3206 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3207 * @adapter: board private structure
3209 * Configure the Rx unit of the MAC after a reset.
3211 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3213 struct ixgbe_hw *hw = &adapter->hw;
3217 /* disable receives while setting up the descriptors */
3218 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3219 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3221 ixgbe_setup_psrtype(adapter);
3222 ixgbe_setup_rdrxctl(adapter);
3224 /* Program registers for the distribution of queues */
3225 ixgbe_setup_mrqc(adapter);
3227 ixgbe_set_uta(adapter);
3229 /* set_rx_buffer_len must be called before ring initialization */
3230 ixgbe_set_rx_buffer_len(adapter);
3233 * Setup the HW Rx Head and Tail Descriptor Pointers and
3234 * the Base and Length of the Rx Descriptor Ring
3236 for (i = 0; i < adapter->num_rx_queues; i++)
3237 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3239 /* disable drop enable for 82598 parts */
3240 if (hw->mac.type == ixgbe_mac_82598EB)
3241 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3243 /* enable all receives */
3244 rxctrl |= IXGBE_RXCTRL_RXEN;
3245 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3248 static int ixgbe_vlan_rx_add_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 /* add VID to filter table */
3255 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3256 set_bit(vid, adapter->active_vlans);
3261 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3263 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3264 struct ixgbe_hw *hw = &adapter->hw;
3265 int pool_ndx = adapter->num_vfs;
3267 /* remove VID from filter table */
3268 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3269 clear_bit(vid, adapter->active_vlans);
3275 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3276 * @adapter: driver data
3278 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3280 struct ixgbe_hw *hw = &adapter->hw;
3283 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3284 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3285 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3289 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3290 * @adapter: driver data
3292 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3294 struct ixgbe_hw *hw = &adapter->hw;
3297 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3298 vlnctrl |= IXGBE_VLNCTRL_VFE;
3299 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3300 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3304 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3305 * @adapter: driver data
3307 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3309 struct ixgbe_hw *hw = &adapter->hw;
3313 switch (hw->mac.type) {
3314 case ixgbe_mac_82598EB:
3315 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3316 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3317 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3319 case ixgbe_mac_82599EB:
3320 case ixgbe_mac_X540:
3321 for (i = 0; i < adapter->num_rx_queues; i++) {
3322 j = adapter->rx_ring[i]->reg_idx;
3323 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3324 vlnctrl &= ~IXGBE_RXDCTL_VME;
3325 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3334 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3335 * @adapter: driver data
3337 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3339 struct ixgbe_hw *hw = &adapter->hw;
3343 switch (hw->mac.type) {
3344 case ixgbe_mac_82598EB:
3345 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3346 vlnctrl |= IXGBE_VLNCTRL_VME;
3347 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3349 case ixgbe_mac_82599EB:
3350 case ixgbe_mac_X540:
3351 for (i = 0; i < adapter->num_rx_queues; i++) {
3352 j = adapter->rx_ring[i]->reg_idx;
3353 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3354 vlnctrl |= IXGBE_RXDCTL_VME;
3355 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3363 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3367 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3369 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3370 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3374 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3375 * @netdev: network interface device structure
3377 * Writes unicast address list to the RAR table.
3378 * Returns: -ENOMEM on failure/insufficient address space
3379 * 0 on no addresses written
3380 * X on writing X addresses to the RAR table
3382 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3384 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3385 struct ixgbe_hw *hw = &adapter->hw;
3386 unsigned int vfn = adapter->num_vfs;
3387 unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3390 /* return ENOMEM indicating insufficient memory for addresses */
3391 if (netdev_uc_count(netdev) > rar_entries)
3394 if (!netdev_uc_empty(netdev) && rar_entries) {
3395 struct netdev_hw_addr *ha;
3396 /* return error if we do not support writing to RAR table */
3397 if (!hw->mac.ops.set_rar)
3400 netdev_for_each_uc_addr(ha, netdev) {
3403 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3408 /* write the addresses in reverse order to avoid write combining */
3409 for (; rar_entries > 0 ; rar_entries--)
3410 hw->mac.ops.clear_rar(hw, rar_entries);
3416 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3417 * @netdev: network interface device structure
3419 * The set_rx_method entry point is called whenever the unicast/multicast
3420 * address list or the network interface flags are updated. This routine is
3421 * responsible for configuring the hardware for proper unicast, multicast and
3424 void ixgbe_set_rx_mode(struct net_device *netdev)
3426 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3427 struct ixgbe_hw *hw = &adapter->hw;
3428 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3431 /* Check for Promiscuous and All Multicast modes */
3433 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3435 /* set all bits that we expect to always be set */
3436 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3437 fctrl |= IXGBE_FCTRL_BAM;
3438 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3439 fctrl |= IXGBE_FCTRL_PMCF;
3441 /* clear the bits we are changing the status of */
3442 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3444 if (netdev->flags & IFF_PROMISC) {
3445 hw->addr_ctrl.user_set_promisc = true;
3446 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3447 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3448 /* don't hardware filter vlans in promisc mode */
3449 ixgbe_vlan_filter_disable(adapter);
3451 if (netdev->flags & IFF_ALLMULTI) {
3452 fctrl |= IXGBE_FCTRL_MPE;
3453 vmolr |= IXGBE_VMOLR_MPE;
3456 * Write addresses to the MTA, if the attempt fails
3457 * then we should just turn on promiscuous mode so
3458 * that we can at least receive multicast traffic
3460 hw->mac.ops.update_mc_addr_list(hw, netdev);
3461 vmolr |= IXGBE_VMOLR_ROMPE;
3463 ixgbe_vlan_filter_enable(adapter);
3464 hw->addr_ctrl.user_set_promisc = false;
3466 * Write addresses to available RAR registers, if there is not
3467 * sufficient space to store all the addresses then enable
3468 * unicast promiscuous mode
3470 count = ixgbe_write_uc_addr_list(netdev);
3472 fctrl |= IXGBE_FCTRL_UPE;
3473 vmolr |= IXGBE_VMOLR_ROPE;
3477 if (adapter->num_vfs) {
3478 ixgbe_restore_vf_multicasts(adapter);
3479 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3480 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3482 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3485 /* This is useful for sniffing bad packets. */
3486 if (adapter->netdev->features & NETIF_F_RXALL) {
3487 /* UPE and MPE will be handled by normal PROMISC logic
3488 * in e1000e_set_rx_mode */
3489 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3490 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3491 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3493 fctrl &= ~(IXGBE_FCTRL_DPF);
3494 /* NOTE: VLAN filtering is disabled by setting PROMISC */
3497 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3499 if (netdev->features & NETIF_F_HW_VLAN_RX)
3500 ixgbe_vlan_strip_enable(adapter);
3502 ixgbe_vlan_strip_disable(adapter);
3505 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3508 struct ixgbe_q_vector *q_vector;
3509 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3511 /* legacy and MSI only use one vector */
3512 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3515 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3516 q_vector = adapter->q_vector[q_idx];
3517 napi_enable(&q_vector->napi);
3521 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3524 struct ixgbe_q_vector *q_vector;
3525 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3527 /* legacy and MSI only use one vector */
3528 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3531 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3532 q_vector = adapter->q_vector[q_idx];
3533 napi_disable(&q_vector->napi);
3537 #ifdef CONFIG_IXGBE_DCB
3539 * ixgbe_configure_dcb - Configure DCB hardware
3540 * @adapter: ixgbe adapter struct
3542 * This is called by the driver on open to configure the DCB hardware.
3543 * This is also called by the gennetlink interface when reconfiguring
3546 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3548 struct ixgbe_hw *hw = &adapter->hw;
3549 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3551 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3552 if (hw->mac.type == ixgbe_mac_82598EB)
3553 netif_set_gso_max_size(adapter->netdev, 65536);
3557 if (hw->mac.type == ixgbe_mac_82598EB)
3558 netif_set_gso_max_size(adapter->netdev, 32768);
3561 /* Enable VLAN tag insert/strip */
3562 adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3564 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3567 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3568 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3571 /* reconfigure the hardware */
3572 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3573 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3575 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3577 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3578 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3579 ixgbe_dcb_hw_ets(&adapter->hw,
3580 adapter->ixgbe_ieee_ets,
3582 ixgbe_dcb_hw_pfc_config(&adapter->hw,
3583 adapter->ixgbe_ieee_pfc->pfc_en,
3584 adapter->ixgbe_ieee_ets->prio_tc);
3587 /* Enable RSS Hash per TC */
3588 if (hw->mac.type != ixgbe_mac_82598EB) {
3592 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3594 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3599 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3601 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3606 /* Additional bittime to account for IXGBE framing */
3607 #define IXGBE_ETH_FRAMING 20
3610 * ixgbe_hpbthresh - calculate high water mark for flow control
3612 * @adapter: board private structure to calculate for
3613 * @pb - packet buffer to calculate
3615 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3617 struct ixgbe_hw *hw = &adapter->hw;
3618 struct net_device *dev = adapter->netdev;
3619 int link, tc, kb, marker;
3622 /* Calculate max LAN frame size */
3623 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3626 /* FCoE traffic class uses FCOE jumbo frames */
3627 if (dev->features & NETIF_F_FCOE_MTU) {
3630 #ifdef CONFIG_IXGBE_DCB
3631 fcoe_pb = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
3634 if (fcoe_pb == pb && tc < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3635 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3639 /* Calculate delay value for device */
3640 switch (hw->mac.type) {
3641 case ixgbe_mac_X540:
3642 dv_id = IXGBE_DV_X540(link, tc);
3645 dv_id = IXGBE_DV(link, tc);
3649 /* Loopback switch introduces additional latency */
3650 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3651 dv_id += IXGBE_B2BT(tc);
3653 /* Delay value is calculated in bit times convert to KB */
3654 kb = IXGBE_BT2KB(dv_id);
3655 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3657 marker = rx_pba - kb;
3659 /* It is possible that the packet buffer is not large enough
3660 * to provide required headroom. In this case throw an error
3661 * to user and a do the best we can.
3664 e_warn(drv, "Packet Buffer(%i) can not provide enough"
3665 "headroom to support flow control."
3666 "Decrease MTU or number of traffic classes\n", pb);
3674 * ixgbe_lpbthresh - calculate low water mark for for flow control
3676 * @adapter: board private structure to calculate for
3677 * @pb - packet buffer to calculate
3679 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3681 struct ixgbe_hw *hw = &adapter->hw;
3682 struct net_device *dev = adapter->netdev;
3686 /* Calculate max LAN frame size */
3687 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3689 /* Calculate delay value for device */
3690 switch (hw->mac.type) {
3691 case ixgbe_mac_X540:
3692 dv_id = IXGBE_LOW_DV_X540(tc);
3695 dv_id = IXGBE_LOW_DV(tc);
3699 /* Delay value is calculated in bit times convert to KB */
3700 return IXGBE_BT2KB(dv_id);
3704 * ixgbe_pbthresh_setup - calculate and setup high low water marks
3706 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3708 struct ixgbe_hw *hw = &adapter->hw;
3709 int num_tc = netdev_get_num_tc(adapter->netdev);
3715 hw->fc.low_water = ixgbe_lpbthresh(adapter);
3717 for (i = 0; i < num_tc; i++) {
3718 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3720 /* Low water marks must not be larger than high water marks */
3721 if (hw->fc.low_water > hw->fc.high_water[i])
3722 hw->fc.low_water = 0;
3726 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3728 struct ixgbe_hw *hw = &adapter->hw;
3730 u8 tc = netdev_get_num_tc(adapter->netdev);
3732 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3733 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3734 hdrm = 32 << adapter->fdir_pballoc;
3738 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3739 ixgbe_pbthresh_setup(adapter);
3742 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3744 struct ixgbe_hw *hw = &adapter->hw;
3745 struct hlist_node *node, *node2;
3746 struct ixgbe_fdir_filter *filter;
3748 spin_lock(&adapter->fdir_perfect_lock);
3750 if (!hlist_empty(&adapter->fdir_filter_list))
3751 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3753 hlist_for_each_entry_safe(filter, node, node2,
3754 &adapter->fdir_filter_list, fdir_node) {
3755 ixgbe_fdir_write_perfect_filter_82599(hw,
3758 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3759 IXGBE_FDIR_DROP_QUEUE :
3760 adapter->rx_ring[filter->action]->reg_idx);
3763 spin_unlock(&adapter->fdir_perfect_lock);
3766 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3768 struct ixgbe_hw *hw = &adapter->hw;
3770 ixgbe_configure_pb(adapter);
3771 #ifdef CONFIG_IXGBE_DCB
3772 ixgbe_configure_dcb(adapter);
3775 ixgbe_set_rx_mode(adapter->netdev);
3776 ixgbe_restore_vlan(adapter);
3779 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3780 ixgbe_configure_fcoe(adapter);
3782 #endif /* IXGBE_FCOE */
3784 switch (hw->mac.type) {
3785 case ixgbe_mac_82599EB:
3786 case ixgbe_mac_X540:
3787 hw->mac.ops.disable_rx_buff(hw);
3793 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3794 ixgbe_init_fdir_signature_82599(&adapter->hw,
3795 adapter->fdir_pballoc);
3796 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3797 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3798 adapter->fdir_pballoc);
3799 ixgbe_fdir_filter_restore(adapter);
3802 switch (hw->mac.type) {
3803 case ixgbe_mac_82599EB:
3804 case ixgbe_mac_X540:
3805 hw->mac.ops.enable_rx_buff(hw);
3811 ixgbe_configure_virtualization(adapter);
3813 ixgbe_configure_tx(adapter);
3814 ixgbe_configure_rx(adapter);
3817 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3819 switch (hw->phy.type) {
3820 case ixgbe_phy_sfp_avago:
3821 case ixgbe_phy_sfp_ftl:
3822 case ixgbe_phy_sfp_intel:
3823 case ixgbe_phy_sfp_unknown:
3824 case ixgbe_phy_sfp_passive_tyco:
3825 case ixgbe_phy_sfp_passive_unknown:
3826 case ixgbe_phy_sfp_active_unknown:
3827 case ixgbe_phy_sfp_ftl_active:
3830 if (hw->mac.type == ixgbe_mac_82598EB)
3838 * ixgbe_sfp_link_config - set up SFP+ link
3839 * @adapter: pointer to private adapter struct
3841 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3844 * We are assuming the worst case scenario here, and that
3845 * is that an SFP was inserted/removed after the reset
3846 * but before SFP detection was enabled. As such the best
3847 * solution is to just start searching as soon as we start
3849 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3850 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3852 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3856 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3857 * @hw: pointer to private hardware struct
3859 * Returns 0 on success, negative on failure
3861 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3864 bool negotiation, link_up = false;
3865 u32 ret = IXGBE_ERR_LINK_SETUP;
3867 if (hw->mac.ops.check_link)
3868 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3873 autoneg = hw->phy.autoneg_advertised;
3874 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3875 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3880 if (hw->mac.ops.setup_link)
3881 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3886 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3888 struct ixgbe_hw *hw = &adapter->hw;
3891 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3892 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3894 gpie |= IXGBE_GPIE_EIAME;
3896 * use EIAM to auto-mask when MSI-X interrupt is asserted
3897 * this saves a register write for every interrupt
3899 switch (hw->mac.type) {
3900 case ixgbe_mac_82598EB:
3901 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3903 case ixgbe_mac_82599EB:
3904 case ixgbe_mac_X540:
3906 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3907 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3911 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3912 * specifically only auto mask tx and rx interrupts */
3913 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3916 /* XXX: to interrupt immediately for EICS writes, enable this */
3917 /* gpie |= IXGBE_GPIE_EIMEN; */
3919 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3920 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3921 gpie |= IXGBE_GPIE_VTMODE_64;
3924 /* Enable Thermal over heat sensor interrupt */
3925 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3926 switch (adapter->hw.mac.type) {
3927 case ixgbe_mac_82599EB:
3928 gpie |= IXGBE_SDP0_GPIEN;
3930 case ixgbe_mac_X540:
3931 gpie |= IXGBE_EIMS_TS;
3938 /* Enable fan failure interrupt */
3939 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3940 gpie |= IXGBE_SDP1_GPIEN;
3942 if (hw->mac.type == ixgbe_mac_82599EB) {
3943 gpie |= IXGBE_SDP1_GPIEN;
3944 gpie |= IXGBE_SDP2_GPIEN;
3947 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3950 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
3952 struct ixgbe_hw *hw = &adapter->hw;
3956 ixgbe_get_hw_control(adapter);
3957 ixgbe_setup_gpie(adapter);
3959 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3960 ixgbe_configure_msix(adapter);
3962 ixgbe_configure_msi_and_legacy(adapter);
3964 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3965 if (hw->mac.ops.enable_tx_laser &&
3966 ((hw->phy.multispeed_fiber) ||
3967 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3968 (hw->mac.type == ixgbe_mac_82599EB))))
3969 hw->mac.ops.enable_tx_laser(hw);
3971 clear_bit(__IXGBE_DOWN, &adapter->state);
3972 ixgbe_napi_enable_all(adapter);
3974 if (ixgbe_is_sfp(hw)) {
3975 ixgbe_sfp_link_config(adapter);
3977 err = ixgbe_non_sfp_link_config(hw);
3979 e_err(probe, "link_config FAILED %d\n", err);
3982 /* clear any pending interrupts, may auto mask */
3983 IXGBE_READ_REG(hw, IXGBE_EICR);
3984 ixgbe_irq_enable(adapter, true, true);
3987 * If this adapter has a fan, check to see if we had a failure
3988 * before we enabled the interrupt.
3990 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3991 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3992 if (esdp & IXGBE_ESDP_SDP1)
3993 e_crit(drv, "Fan has stopped, replace the adapter\n");
3996 /* enable transmits */
3997 netif_tx_start_all_queues(adapter->netdev);
3999 /* bring the link up in the watchdog, this could race with our first
4000 * link up interrupt but shouldn't be a problem */
4001 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4002 adapter->link_check_timeout = jiffies;
4003 mod_timer(&adapter->service_timer, jiffies);
4005 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4006 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4007 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4008 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4011 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4013 WARN_ON(in_interrupt());
4014 /* put off any impending NetWatchDogTimeout */
4015 adapter->netdev->trans_start = jiffies;
4017 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4018 usleep_range(1000, 2000);
4019 ixgbe_down(adapter);
4021 * If SR-IOV enabled then wait a bit before bringing the adapter
4022 * back up to give the VFs time to respond to the reset. The
4023 * two second wait is based upon the watchdog timer cycle in
4026 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4029 clear_bit(__IXGBE_RESETTING, &adapter->state);
4032 void ixgbe_up(struct ixgbe_adapter *adapter)
4034 /* hardware has been reset, we need to reload some things */
4035 ixgbe_configure(adapter);
4037 ixgbe_up_complete(adapter);
4040 void ixgbe_reset(struct ixgbe_adapter *adapter)
4042 struct ixgbe_hw *hw = &adapter->hw;
4045 /* lock SFP init bit to prevent race conditions with the watchdog */
4046 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4047 usleep_range(1000, 2000);
4049 /* clear all SFP and link config related flags while holding SFP_INIT */
4050 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4051 IXGBE_FLAG2_SFP_NEEDS_RESET);
4052 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4054 err = hw->mac.ops.init_hw(hw);
4057 case IXGBE_ERR_SFP_NOT_PRESENT:
4058 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4060 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4061 e_dev_err("master disable timed out\n");
4063 case IXGBE_ERR_EEPROM_VERSION:
4064 /* We are running on a pre-production device, log a warning */
4065 e_dev_warn("This device is a pre-production adapter/LOM. "
4066 "Please be aware there may be issues associated with "
4067 "your hardware. If you are experiencing problems "
4068 "please contact your Intel or hardware "
4069 "representative who provided you with this "
4073 e_dev_err("Hardware Error: %d\n", err);
4076 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4078 /* reprogram the RAR[0] in case user changed it. */
4079 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
4084 * ixgbe_init_rx_page_offset - initialize page offset values for Rx buffers
4085 * @rx_ring: ring to setup
4087 * On many IA platforms the L1 cache has a critical stride of 4K, this
4088 * results in each receive buffer starting in the same cache set. To help
4089 * reduce the pressure on this cache set we can interleave the offsets so
4090 * that only every other buffer will be in the same cache set.
4092 static void ixgbe_init_rx_page_offset(struct ixgbe_ring *rx_ring)
4094 struct ixgbe_rx_buffer *rx_buffer = rx_ring->rx_buffer_info;
4097 for (i = 0; i < rx_ring->count; i += 2) {
4098 rx_buffer[0].page_offset = 0;
4099 rx_buffer[1].page_offset = ixgbe_rx_bufsz(rx_ring);
4100 rx_buffer = &rx_buffer[2];
4105 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4106 * @rx_ring: ring to free buffers from
4108 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4110 struct device *dev = rx_ring->dev;
4114 /* ring already cleared, nothing to do */
4115 if (!rx_ring->rx_buffer_info)
4118 /* Free all the Rx ring sk_buffs */
4119 for (i = 0; i < rx_ring->count; i++) {
4120 struct ixgbe_rx_buffer *rx_buffer;
4122 rx_buffer = &rx_ring->rx_buffer_info[i];
4123 if (rx_buffer->skb) {
4124 struct sk_buff *skb = rx_buffer->skb;
4125 if (IXGBE_CB(skb)->page_released) {
4128 ixgbe_rx_bufsz(rx_ring),
4130 IXGBE_CB(skb)->page_released = false;
4134 rx_buffer->skb = NULL;
4136 dma_unmap_page(dev, rx_buffer->dma,
4137 ixgbe_rx_pg_size(rx_ring),
4140 if (rx_buffer->page)
4141 put_page(rx_buffer->page);
4142 rx_buffer->page = NULL;
4145 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4146 memset(rx_ring->rx_buffer_info, 0, size);
4148 ixgbe_init_rx_page_offset(rx_ring);
4150 /* Zero out the descriptor ring */
4151 memset(rx_ring->desc, 0, rx_ring->size);
4153 rx_ring->next_to_alloc = 0;
4154 rx_ring->next_to_clean = 0;
4155 rx_ring->next_to_use = 0;
4159 * ixgbe_clean_tx_ring - Free Tx Buffers
4160 * @tx_ring: ring to be cleaned
4162 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4164 struct ixgbe_tx_buffer *tx_buffer_info;
4168 /* ring already cleared, nothing to do */
4169 if (!tx_ring->tx_buffer_info)
4172 /* Free all the Tx ring sk_buffs */
4173 for (i = 0; i < tx_ring->count; i++) {
4174 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4175 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4178 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4179 memset(tx_ring->tx_buffer_info, 0, size);
4181 /* Zero out the descriptor ring */
4182 memset(tx_ring->desc, 0, tx_ring->size);
4184 tx_ring->next_to_use = 0;
4185 tx_ring->next_to_clean = 0;
4189 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4190 * @adapter: board private structure
4192 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4196 for (i = 0; i < adapter->num_rx_queues; i++)
4197 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4201 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4202 * @adapter: board private structure
4204 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4208 for (i = 0; i < adapter->num_tx_queues; i++)
4209 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4212 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4214 struct hlist_node *node, *node2;
4215 struct ixgbe_fdir_filter *filter;
4217 spin_lock(&adapter->fdir_perfect_lock);
4219 hlist_for_each_entry_safe(filter, node, node2,
4220 &adapter->fdir_filter_list, fdir_node) {
4221 hlist_del(&filter->fdir_node);
4224 adapter->fdir_filter_count = 0;
4226 spin_unlock(&adapter->fdir_perfect_lock);
4229 void ixgbe_down(struct ixgbe_adapter *adapter)
4231 struct net_device *netdev = adapter->netdev;
4232 struct ixgbe_hw *hw = &adapter->hw;
4236 /* signal that we are down to the interrupt handler */
4237 set_bit(__IXGBE_DOWN, &adapter->state);
4239 /* disable receives */
4240 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4241 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4243 /* disable all enabled rx queues */
4244 for (i = 0; i < adapter->num_rx_queues; i++)
4245 /* this call also flushes the previous write */
4246 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4248 usleep_range(10000, 20000);
4250 netif_tx_stop_all_queues(netdev);
4252 /* call carrier off first to avoid false dev_watchdog timeouts */
4253 netif_carrier_off(netdev);
4254 netif_tx_disable(netdev);
4256 ixgbe_irq_disable(adapter);
4258 ixgbe_napi_disable_all(adapter);
4260 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4261 IXGBE_FLAG2_RESET_REQUESTED);
4262 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4264 del_timer_sync(&adapter->service_timer);
4266 if (adapter->num_vfs) {
4267 /* Clear EITR Select mapping */
4268 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4270 /* Mark all the VFs as inactive */
4271 for (i = 0 ; i < adapter->num_vfs; i++)
4272 adapter->vfinfo[i].clear_to_send = false;
4274 /* ping all the active vfs to let them know we are going down */
4275 ixgbe_ping_all_vfs(adapter);
4277 /* Disable all VFTE/VFRE TX/RX */
4278 ixgbe_disable_tx_rx(adapter);
4281 /* disable transmits in the hardware now that interrupts are off */
4282 for (i = 0; i < adapter->num_tx_queues; i++) {
4283 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4284 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4287 /* Disable the Tx DMA engine on 82599 and X540 */
4288 switch (hw->mac.type) {
4289 case ixgbe_mac_82599EB:
4290 case ixgbe_mac_X540:
4291 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4292 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4293 ~IXGBE_DMATXCTL_TE));
4299 if (!pci_channel_offline(adapter->pdev))
4300 ixgbe_reset(adapter);
4302 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4303 if (hw->mac.ops.disable_tx_laser &&
4304 ((hw->phy.multispeed_fiber) ||
4305 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4306 (hw->mac.type == ixgbe_mac_82599EB))))
4307 hw->mac.ops.disable_tx_laser(hw);
4309 ixgbe_clean_all_tx_rings(adapter);
4310 ixgbe_clean_all_rx_rings(adapter);
4312 #ifdef CONFIG_IXGBE_DCA
4313 /* since we reset the hardware DCA settings were cleared */
4314 ixgbe_setup_dca(adapter);
4319 * ixgbe_tx_timeout - Respond to a Tx Hang
4320 * @netdev: network interface device structure
4322 static void ixgbe_tx_timeout(struct net_device *netdev)
4324 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4326 /* Do the reset outside of interrupt context */
4327 ixgbe_tx_timeout_reset(adapter);
4331 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4332 * @adapter: board private structure to initialize
4334 * ixgbe_sw_init initializes the Adapter private data structure.
4335 * Fields are initialized based on PCI device information and
4336 * OS network device settings (MTU size).
4338 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4340 struct ixgbe_hw *hw = &adapter->hw;
4341 struct pci_dev *pdev = adapter->pdev;
4343 #ifdef CONFIG_IXGBE_DCB
4345 struct tc_configuration *tc;
4348 /* PCI config space info */
4350 hw->vendor_id = pdev->vendor;
4351 hw->device_id = pdev->device;
4352 hw->revision_id = pdev->revision;
4353 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4354 hw->subsystem_device_id = pdev->subsystem_device;
4356 /* Set capability flags */
4357 rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4358 adapter->ring_feature[RING_F_RSS].indices = rss;
4359 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4360 switch (hw->mac.type) {
4361 case ixgbe_mac_82598EB:
4362 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4363 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4364 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4366 case ixgbe_mac_X540:
4367 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4368 case ixgbe_mac_82599EB:
4369 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4370 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4371 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4372 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4373 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4374 /* Flow Director hash filters enabled */
4375 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4376 adapter->atr_sample_rate = 20;
4377 adapter->ring_feature[RING_F_FDIR].indices =
4378 IXGBE_MAX_FDIR_INDICES;
4379 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4381 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4382 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4383 adapter->ring_feature[RING_F_FCOE].indices = 0;
4384 #ifdef CONFIG_IXGBE_DCB
4385 /* Default traffic class to use for FCoE */
4386 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4388 #endif /* IXGBE_FCOE */
4394 /* n-tuple support exists, always init our spinlock */
4395 spin_lock_init(&adapter->fdir_perfect_lock);
4397 #ifdef CONFIG_IXGBE_DCB
4398 switch (hw->mac.type) {
4399 case ixgbe_mac_X540:
4400 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4401 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4404 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4405 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4409 /* Configure DCB traffic classes */
4410 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4411 tc = &adapter->dcb_cfg.tc_config[j];
4412 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4413 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4414 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4415 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4416 tc->dcb_pfc = pfc_disabled;
4419 /* Initialize default user to priority mapping, UPx->TC0 */
4420 tc = &adapter->dcb_cfg.tc_config[0];
4421 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
4422 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
4424 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4425 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4426 adapter->dcb_cfg.pfc_mode_enable = false;
4427 adapter->dcb_set_bitmap = 0x00;
4428 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4429 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4434 /* default flow control settings */
4435 hw->fc.requested_mode = ixgbe_fc_full;
4436 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4438 adapter->last_lfc_mode = hw->fc.current_mode;
4440 ixgbe_pbthresh_setup(adapter);
4441 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4442 hw->fc.send_xon = true;
4443 hw->fc.disable_fc_autoneg = false;
4445 /* enable itr by default in dynamic mode */
4446 adapter->rx_itr_setting = 1;
4447 adapter->tx_itr_setting = 1;
4449 /* set default ring sizes */
4450 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4451 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4453 /* set default work limits */
4454 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4456 /* initialize eeprom parameters */
4457 if (ixgbe_init_eeprom_params_generic(hw)) {
4458 e_dev_err("EEPROM initialization failed\n");
4462 set_bit(__IXGBE_DOWN, &adapter->state);
4468 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4469 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4471 * Return 0 on success, negative on failure
4473 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4475 struct device *dev = tx_ring->dev;
4476 int orig_node = dev_to_node(dev);
4480 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4482 if (tx_ring->q_vector)
4483 numa_node = tx_ring->q_vector->numa_node;
4485 tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
4486 if (!tx_ring->tx_buffer_info)
4487 tx_ring->tx_buffer_info = vzalloc(size);
4488 if (!tx_ring->tx_buffer_info)
4491 /* round up to nearest 4K */
4492 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4493 tx_ring->size = ALIGN(tx_ring->size, 4096);
4495 set_dev_node(dev, numa_node);
4496 tx_ring->desc = dma_alloc_coherent(dev,
4500 set_dev_node(dev, orig_node);
4502 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4503 &tx_ring->dma, GFP_KERNEL);
4507 tx_ring->next_to_use = 0;
4508 tx_ring->next_to_clean = 0;
4512 vfree(tx_ring->tx_buffer_info);
4513 tx_ring->tx_buffer_info = NULL;
4514 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4519 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4520 * @adapter: board private structure
4522 * If this function returns with an error, then it's possible one or
4523 * more of the rings is populated (while the rest are not). It is the
4524 * callers duty to clean those orphaned rings.
4526 * Return 0 on success, negative on failure
4528 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4532 for (i = 0; i < adapter->num_tx_queues; i++) {
4533 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4536 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4544 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4545 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4547 * Returns 0 on success, negative on failure
4549 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4551 struct device *dev = rx_ring->dev;
4552 int orig_node = dev_to_node(dev);
4556 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4558 if (rx_ring->q_vector)
4559 numa_node = rx_ring->q_vector->numa_node;
4561 rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
4562 if (!rx_ring->rx_buffer_info)
4563 rx_ring->rx_buffer_info = vzalloc(size);
4564 if (!rx_ring->rx_buffer_info)
4567 /* Round up to nearest 4K */
4568 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4569 rx_ring->size = ALIGN(rx_ring->size, 4096);
4571 set_dev_node(dev, numa_node);
4572 rx_ring->desc = dma_alloc_coherent(dev,
4576 set_dev_node(dev, orig_node);
4578 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4579 &rx_ring->dma, GFP_KERNEL);
4583 rx_ring->next_to_clean = 0;
4584 rx_ring->next_to_use = 0;
4586 ixgbe_init_rx_page_offset(rx_ring);
4590 vfree(rx_ring->rx_buffer_info);
4591 rx_ring->rx_buffer_info = NULL;
4592 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4597 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4598 * @adapter: board private structure
4600 * If this function returns with an error, then it's possible one or
4601 * more of the rings is populated (while the rest are not). It is the
4602 * callers duty to clean those orphaned rings.
4604 * Return 0 on success, negative on failure
4606 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4610 for (i = 0; i < adapter->num_rx_queues; i++) {
4611 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
4614 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4622 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4623 * @tx_ring: Tx descriptor ring for a specific queue
4625 * Free all transmit software resources
4627 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
4629 ixgbe_clean_tx_ring(tx_ring);
4631 vfree(tx_ring->tx_buffer_info);
4632 tx_ring->tx_buffer_info = NULL;
4634 /* if not set, then don't free */
4638 dma_free_coherent(tx_ring->dev, tx_ring->size,
4639 tx_ring->desc, tx_ring->dma);
4641 tx_ring->desc = NULL;
4645 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4646 * @adapter: board private structure
4648 * Free all transmit software resources
4650 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4654 for (i = 0; i < adapter->num_tx_queues; i++)
4655 if (adapter->tx_ring[i]->desc)
4656 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4660 * ixgbe_free_rx_resources - Free Rx Resources
4661 * @rx_ring: ring to clean the resources from
4663 * Free all receive software resources
4665 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
4667 ixgbe_clean_rx_ring(rx_ring);
4669 vfree(rx_ring->rx_buffer_info);
4670 rx_ring->rx_buffer_info = NULL;
4672 /* if not set, then don't free */
4676 dma_free_coherent(rx_ring->dev, rx_ring->size,
4677 rx_ring->desc, rx_ring->dma);
4679 rx_ring->desc = NULL;
4683 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4684 * @adapter: board private structure
4686 * Free all receive software resources
4688 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4692 for (i = 0; i < adapter->num_rx_queues; i++)
4693 if (adapter->rx_ring[i]->desc)
4694 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4698 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4699 * @netdev: network interface device structure
4700 * @new_mtu: new value for maximum frame size
4702 * Returns 0 on success, negative on failure
4704 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4706 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4707 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4709 /* MTU < 68 is an error and causes problems on some kernels */
4710 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4714 * For 82599EB we cannot allow PF to change MTU greater than 1500
4715 * in SR-IOV mode as it may cause buffer overruns in guest VFs that
4716 * don't allocate and chain buffers correctly.
4718 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
4719 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
4720 (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
4723 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4725 /* must set new MTU before calling down or up */
4726 netdev->mtu = new_mtu;
4728 if (netif_running(netdev))
4729 ixgbe_reinit_locked(adapter);
4735 * ixgbe_open - Called when a network interface is made active
4736 * @netdev: network interface device structure
4738 * Returns 0 on success, negative value on failure
4740 * The open entry point is called when a network interface is made
4741 * active by the system (IFF_UP). At this point all resources needed
4742 * for transmit and receive operations are allocated, the interrupt
4743 * handler is registered with the OS, the watchdog timer is started,
4744 * and the stack is notified that the interface is ready.
4746 static int ixgbe_open(struct net_device *netdev)
4748 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4751 /* disallow open during test */
4752 if (test_bit(__IXGBE_TESTING, &adapter->state))
4755 netif_carrier_off(netdev);
4757 /* allocate transmit descriptors */
4758 err = ixgbe_setup_all_tx_resources(adapter);
4762 /* allocate receive descriptors */
4763 err = ixgbe_setup_all_rx_resources(adapter);
4767 ixgbe_configure(adapter);
4769 err = ixgbe_request_irq(adapter);
4773 ixgbe_up_complete(adapter);
4779 ixgbe_free_all_rx_resources(adapter);
4781 ixgbe_free_all_tx_resources(adapter);
4782 ixgbe_reset(adapter);
4788 * ixgbe_close - Disables a network interface
4789 * @netdev: network interface device structure
4791 * Returns 0, this is not allowed to fail
4793 * The close entry point is called when an interface is de-activated
4794 * by the OS. The hardware is still under the drivers control, but
4795 * needs to be disabled. A global MAC reset is issued to stop the
4796 * hardware, and all transmit and receive resources are freed.
4798 static int ixgbe_close(struct net_device *netdev)
4800 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4802 ixgbe_down(adapter);
4803 ixgbe_free_irq(adapter);
4805 ixgbe_fdir_filter_exit(adapter);
4807 ixgbe_free_all_tx_resources(adapter);
4808 ixgbe_free_all_rx_resources(adapter);
4810 ixgbe_release_hw_control(adapter);
4816 static int ixgbe_resume(struct pci_dev *pdev)
4818 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4819 struct net_device *netdev = adapter->netdev;
4822 pci_set_power_state(pdev, PCI_D0);
4823 pci_restore_state(pdev);
4825 * pci_restore_state clears dev->state_saved so call
4826 * pci_save_state to restore it.
4828 pci_save_state(pdev);
4830 err = pci_enable_device_mem(pdev);
4832 e_dev_err("Cannot enable PCI device from suspend\n");
4835 pci_set_master(pdev);
4837 pci_wake_from_d3(pdev, false);
4839 err = ixgbe_init_interrupt_scheme(adapter);
4841 e_dev_err("Cannot initialize interrupts for device\n");
4845 ixgbe_reset(adapter);
4847 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4849 if (netif_running(netdev)) {
4850 err = ixgbe_open(netdev);
4855 netif_device_attach(netdev);
4859 #endif /* CONFIG_PM */
4861 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4863 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4864 struct net_device *netdev = adapter->netdev;
4865 struct ixgbe_hw *hw = &adapter->hw;
4867 u32 wufc = adapter->wol;
4872 netif_device_detach(netdev);
4874 if (netif_running(netdev)) {
4875 ixgbe_down(adapter);
4876 ixgbe_free_irq(adapter);
4877 ixgbe_free_all_tx_resources(adapter);
4878 ixgbe_free_all_rx_resources(adapter);
4881 ixgbe_clear_interrupt_scheme(adapter);
4883 kfree(adapter->ixgbe_ieee_pfc);
4884 kfree(adapter->ixgbe_ieee_ets);
4888 retval = pci_save_state(pdev);
4894 ixgbe_set_rx_mode(netdev);
4896 /* turn on all-multi mode if wake on multicast is enabled */
4897 if (wufc & IXGBE_WUFC_MC) {
4898 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4899 fctrl |= IXGBE_FCTRL_MPE;
4900 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4903 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4904 ctrl |= IXGBE_CTRL_GIO_DIS;
4905 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4907 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4909 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4910 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4913 switch (hw->mac.type) {
4914 case ixgbe_mac_82598EB:
4915 pci_wake_from_d3(pdev, false);
4917 case ixgbe_mac_82599EB:
4918 case ixgbe_mac_X540:
4919 pci_wake_from_d3(pdev, !!wufc);
4925 *enable_wake = !!wufc;
4927 ixgbe_release_hw_control(adapter);
4929 pci_disable_device(pdev);
4935 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4940 retval = __ixgbe_shutdown(pdev, &wake);
4945 pci_prepare_to_sleep(pdev);
4947 pci_wake_from_d3(pdev, false);
4948 pci_set_power_state(pdev, PCI_D3hot);
4953 #endif /* CONFIG_PM */
4955 static void ixgbe_shutdown(struct pci_dev *pdev)
4959 __ixgbe_shutdown(pdev, &wake);
4961 if (system_state == SYSTEM_POWER_OFF) {
4962 pci_wake_from_d3(pdev, wake);
4963 pci_set_power_state(pdev, PCI_D3hot);
4968 * ixgbe_update_stats - Update the board statistics counters.
4969 * @adapter: board private structure
4971 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4973 struct net_device *netdev = adapter->netdev;
4974 struct ixgbe_hw *hw = &adapter->hw;
4975 struct ixgbe_hw_stats *hwstats = &adapter->stats;
4977 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4978 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
4979 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
4980 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
4982 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4984 u64 fcoe_noddp_counts_sum = 0, fcoe_noddp_ext_buff_counts_sum = 0;
4985 #endif /* IXGBE_FCOE */
4987 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
4988 test_bit(__IXGBE_RESETTING, &adapter->state))
4991 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
4994 for (i = 0; i < 16; i++)
4995 adapter->hw_rx_no_dma_resources +=
4996 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4997 for (i = 0; i < adapter->num_rx_queues; i++) {
4998 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
4999 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5001 adapter->rsc_total_count = rsc_count;
5002 adapter->rsc_total_flush = rsc_flush;
5005 for (i = 0; i < adapter->num_rx_queues; i++) {
5006 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5007 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5008 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5009 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5010 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5011 bytes += rx_ring->stats.bytes;
5012 packets += rx_ring->stats.packets;
5014 adapter->non_eop_descs = non_eop_descs;
5015 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5016 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5017 adapter->hw_csum_rx_error = hw_csum_rx_error;
5018 netdev->stats.rx_bytes = bytes;
5019 netdev->stats.rx_packets = packets;
5023 /* gather some stats to the adapter struct that are per queue */
5024 for (i = 0; i < adapter->num_tx_queues; i++) {
5025 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5026 restart_queue += tx_ring->tx_stats.restart_queue;
5027 tx_busy += tx_ring->tx_stats.tx_busy;
5028 bytes += tx_ring->stats.bytes;
5029 packets += tx_ring->stats.packets;
5031 adapter->restart_queue = restart_queue;
5032 adapter->tx_busy = tx_busy;
5033 netdev->stats.tx_bytes = bytes;
5034 netdev->stats.tx_packets = packets;
5036 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5038 /* 8 register reads */
5039 for (i = 0; i < 8; i++) {
5040 /* for packet buffers not used, the register should read 0 */
5041 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5043 hwstats->mpc[i] += mpc;
5044 total_mpc += hwstats->mpc[i];
5045 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5046 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5047 switch (hw->mac.type) {
5048 case ixgbe_mac_82598EB:
5049 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5050 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5051 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5052 hwstats->pxonrxc[i] +=
5053 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5055 case ixgbe_mac_82599EB:
5056 case ixgbe_mac_X540:
5057 hwstats->pxonrxc[i] +=
5058 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5065 /*16 register reads */
5066 for (i = 0; i < 16; i++) {
5067 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5068 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5069 if ((hw->mac.type == ixgbe_mac_82599EB) ||
5070 (hw->mac.type == ixgbe_mac_X540)) {
5071 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5072 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5073 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5074 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5078 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5079 /* work around hardware counting issue */
5080 hwstats->gprc -= missed_rx;
5082 ixgbe_update_xoff_received(adapter);
5084 /* 82598 hardware only has a 32 bit counter in the high register */
5085 switch (hw->mac.type) {
5086 case ixgbe_mac_82598EB:
5087 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5088 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5089 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5090 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5092 case ixgbe_mac_X540:
5093 /* OS2BMC stats are X540 only*/
5094 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5095 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5096 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5097 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5098 case ixgbe_mac_82599EB:
5099 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5100 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5101 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5102 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5103 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5104 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5105 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5106 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5107 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5109 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5110 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5111 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5112 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5113 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5114 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5115 /* Add up per cpu counters for total ddp aloc fail */
5116 if (fcoe->pcpu_noddp && fcoe->pcpu_noddp_ext_buff) {
5117 for_each_possible_cpu(cpu) {
5118 fcoe_noddp_counts_sum +=
5119 *per_cpu_ptr(fcoe->pcpu_noddp, cpu);
5120 fcoe_noddp_ext_buff_counts_sum +=
5122 pcpu_noddp_ext_buff, cpu);
5125 hwstats->fcoe_noddp = fcoe_noddp_counts_sum;
5126 hwstats->fcoe_noddp_ext_buff = fcoe_noddp_ext_buff_counts_sum;
5127 #endif /* IXGBE_FCOE */
5132 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5133 hwstats->bprc += bprc;
5134 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5135 if (hw->mac.type == ixgbe_mac_82598EB)
5136 hwstats->mprc -= bprc;
5137 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5138 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5139 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5140 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5141 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5142 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5143 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5144 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5145 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5146 hwstats->lxontxc += lxon;
5147 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5148 hwstats->lxofftxc += lxoff;
5149 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5150 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5152 * 82598 errata - tx of flow control packets is included in tx counters
5154 xon_off_tot = lxon + lxoff;
5155 hwstats->gptc -= xon_off_tot;
5156 hwstats->mptc -= xon_off_tot;
5157 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5158 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5159 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5160 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5161 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5162 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5163 hwstats->ptc64 -= xon_off_tot;
5164 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5165 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5166 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5167 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5168 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5169 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5171 /* Fill out the OS statistics structure */
5172 netdev->stats.multicast = hwstats->mprc;
5175 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5176 netdev->stats.rx_dropped = 0;
5177 netdev->stats.rx_length_errors = hwstats->rlec;
5178 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5179 netdev->stats.rx_missed_errors = total_mpc;
5183 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5184 * @adapter - pointer to the device adapter structure
5186 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5188 struct ixgbe_hw *hw = &adapter->hw;
5191 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5194 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5196 /* if interface is down do nothing */
5197 if (test_bit(__IXGBE_DOWN, &adapter->state))
5200 /* do nothing if we are not using signature filters */
5201 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5204 adapter->fdir_overflow++;
5206 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5207 for (i = 0; i < adapter->num_tx_queues; i++)
5208 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5209 &(adapter->tx_ring[i]->state));
5210 /* re-enable flow director interrupts */
5211 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5213 e_err(probe, "failed to finish FDIR re-initialization, "
5214 "ignored adding FDIR ATR filters\n");
5219 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5220 * @adapter - pointer to the device adapter structure
5222 * This function serves two purposes. First it strobes the interrupt lines
5223 * in order to make certain interrupts are occurring. Secondly it sets the
5224 * bits needed to check for TX hangs. As a result we should immediately
5225 * determine if a hang has occurred.
5227 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5229 struct ixgbe_hw *hw = &adapter->hw;
5233 /* If we're down or resetting, just bail */
5234 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5235 test_bit(__IXGBE_RESETTING, &adapter->state))
5238 /* Force detection of hung controller */
5239 if (netif_carrier_ok(adapter->netdev)) {
5240 for (i = 0; i < adapter->num_tx_queues; i++)
5241 set_check_for_tx_hang(adapter->tx_ring[i]);
5244 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5246 * for legacy and MSI interrupts don't set any bits
5247 * that are enabled for EIAM, because this operation
5248 * would set *both* EIMS and EICS for any bit in EIAM
5250 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5251 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5253 /* get one bit for every active tx/rx interrupt vector */
5254 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5255 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5256 if (qv->rx.ring || qv->tx.ring)
5257 eics |= ((u64)1 << i);
5261 /* Cause software interrupt to ensure rings are cleaned */
5262 ixgbe_irq_rearm_queues(adapter, eics);
5267 * ixgbe_watchdog_update_link - update the link status
5268 * @adapter - pointer to the device adapter structure
5269 * @link_speed - pointer to a u32 to store the link_speed
5271 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5273 struct ixgbe_hw *hw = &adapter->hw;
5274 u32 link_speed = adapter->link_speed;
5275 bool link_up = adapter->link_up;
5278 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5281 if (hw->mac.ops.check_link) {
5282 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5284 /* always assume link is up, if no check link function */
5285 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5289 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5290 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5291 hw->mac.ops.fc_enable(hw, i);
5293 hw->mac.ops.fc_enable(hw, 0);
5298 time_after(jiffies, (adapter->link_check_timeout +
5299 IXGBE_TRY_LINK_TIMEOUT))) {
5300 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5301 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5302 IXGBE_WRITE_FLUSH(hw);
5305 adapter->link_up = link_up;
5306 adapter->link_speed = link_speed;
5310 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5311 * print link up message
5312 * @adapter - pointer to the device adapter structure
5314 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5316 struct net_device *netdev = adapter->netdev;
5317 struct ixgbe_hw *hw = &adapter->hw;
5318 u32 link_speed = adapter->link_speed;
5319 bool flow_rx, flow_tx;
5321 /* only continue if link was previously down */
5322 if (netif_carrier_ok(netdev))
5325 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5327 switch (hw->mac.type) {
5328 case ixgbe_mac_82598EB: {
5329 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5330 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5331 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5332 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5335 case ixgbe_mac_X540:
5336 case ixgbe_mac_82599EB: {
5337 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5338 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5339 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5340 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5348 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5349 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5351 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5353 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5356 ((flow_rx && flow_tx) ? "RX/TX" :
5358 (flow_tx ? "TX" : "None"))));
5360 netif_carrier_on(netdev);
5361 ixgbe_check_vf_rate_limit(adapter);
5365 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5366 * print link down message
5367 * @adapter - pointer to the adapter structure
5369 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
5371 struct net_device *netdev = adapter->netdev;
5372 struct ixgbe_hw *hw = &adapter->hw;
5374 adapter->link_up = false;
5375 adapter->link_speed = 0;
5377 /* only continue if link was up previously */
5378 if (!netif_carrier_ok(netdev))
5381 /* poll for SFP+ cable when link is down */
5382 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5383 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5385 e_info(drv, "NIC Link is Down\n");
5386 netif_carrier_off(netdev);
5390 * ixgbe_watchdog_flush_tx - flush queues on link down
5391 * @adapter - pointer to the device adapter structure
5393 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5396 int some_tx_pending = 0;
5398 if (!netif_carrier_ok(adapter->netdev)) {
5399 for (i = 0; i < adapter->num_tx_queues; i++) {
5400 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5401 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5402 some_tx_pending = 1;
5407 if (some_tx_pending) {
5408 /* We've lost link, so the controller stops DMA,
5409 * but we've got queued Tx work that's never going
5410 * to get done, so reset controller to flush Tx.
5411 * (Do the reset outside of interrupt context).
5413 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5418 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5422 /* Do not perform spoof check for 82598 */
5423 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5426 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5429 * ssvpc register is cleared on read, if zero then no
5430 * spoofed packets in the last interval.
5435 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
5439 * ixgbe_watchdog_subtask - check and bring link up
5440 * @adapter - pointer to the device adapter structure
5442 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5444 /* if interface is down do nothing */
5445 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5446 test_bit(__IXGBE_RESETTING, &adapter->state))
5449 ixgbe_watchdog_update_link(adapter);
5451 if (adapter->link_up)
5452 ixgbe_watchdog_link_is_up(adapter);
5454 ixgbe_watchdog_link_is_down(adapter);
5456 ixgbe_spoof_check(adapter);
5457 ixgbe_update_stats(adapter);
5459 ixgbe_watchdog_flush_tx(adapter);
5463 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5464 * @adapter - the ixgbe adapter structure
5466 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5468 struct ixgbe_hw *hw = &adapter->hw;
5471 /* not searching for SFP so there is nothing to do here */
5472 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5473 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5476 /* someone else is in init, wait until next service event */
5477 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5480 err = hw->phy.ops.identify_sfp(hw);
5481 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5484 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5485 /* If no cable is present, then we need to reset
5486 * the next time we find a good cable. */
5487 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5494 /* exit if reset not needed */
5495 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5498 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5501 * A module may be identified correctly, but the EEPROM may not have
5502 * support for that module. setup_sfp() will fail in that case, so
5503 * we should not allow that module to load.
5505 if (hw->mac.type == ixgbe_mac_82598EB)
5506 err = hw->phy.ops.reset(hw);
5508 err = hw->mac.ops.setup_sfp(hw);
5510 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5513 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5514 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5517 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5519 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5520 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5521 e_dev_err("failed to initialize because an unsupported "
5522 "SFP+ module type was detected.\n");
5523 e_dev_err("Reload the driver after installing a "
5524 "supported module.\n");
5525 unregister_netdev(adapter->netdev);
5530 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5531 * @adapter - the ixgbe adapter structure
5533 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5535 struct ixgbe_hw *hw = &adapter->hw;
5539 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5542 /* someone else is in init, wait until next service event */
5543 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5546 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5548 autoneg = hw->phy.autoneg_advertised;
5549 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5550 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5551 if (hw->mac.ops.setup_link)
5552 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5554 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5555 adapter->link_check_timeout = jiffies;
5556 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5559 #ifdef CONFIG_PCI_IOV
5560 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
5563 struct ixgbe_hw *hw = &adapter->hw;
5564 struct net_device *netdev = adapter->netdev;
5568 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
5569 if (gpc) /* If incrementing then no need for the check below */
5572 * Check to see if a bad DMA write target from an errant or
5573 * malicious VF has caused a PCIe error. If so then we can
5574 * issue a VFLR to the offending VF(s) and then resume without
5575 * requesting a full slot reset.
5578 for (vf = 0; vf < adapter->num_vfs; vf++) {
5579 ciaa = (vf << 16) | 0x80000000;
5580 /* 32 bit read so align, we really want status at offset 6 */
5581 ciaa |= PCI_COMMAND;
5582 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5583 ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
5585 /* disable debug mode asap after reading data */
5586 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5587 /* Get the upper 16 bits which will be the PCI status reg */
5589 if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
5590 netdev_err(netdev, "VF %d Hung DMA\n", vf);
5592 ciaa = (vf << 16) | 0x80000000;
5594 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5595 ciad = 0x00008000; /* VFLR */
5596 IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
5598 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5605 * ixgbe_service_timer - Timer Call-back
5606 * @data: pointer to adapter cast into an unsigned long
5608 static void ixgbe_service_timer(unsigned long data)
5610 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5611 unsigned long next_event_offset;
5614 /* poll faster when waiting for link */
5615 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5616 next_event_offset = HZ / 10;
5618 next_event_offset = HZ * 2;
5620 #ifdef CONFIG_PCI_IOV
5622 * don't bother with SR-IOV VF DMA hang check if there are
5623 * no VFs or the link is down
5625 if (!adapter->num_vfs ||
5626 (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5627 goto normal_timer_service;
5629 /* If we have VFs allocated then we must check for DMA hangs */
5630 ixgbe_check_for_bad_vf(adapter);
5631 next_event_offset = HZ / 50;
5632 adapter->timer_event_accumulator++;
5634 if (adapter->timer_event_accumulator >= 100)
5635 adapter->timer_event_accumulator = 0;
5639 normal_timer_service:
5641 /* Reset the timer */
5642 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5645 ixgbe_service_event_schedule(adapter);
5648 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5650 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
5653 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
5655 /* If we're already down or resetting, just bail */
5656 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5657 test_bit(__IXGBE_RESETTING, &adapter->state))
5660 ixgbe_dump(adapter);
5661 netdev_err(adapter->netdev, "Reset adapter\n");
5662 adapter->tx_timeout_count++;
5664 ixgbe_reinit_locked(adapter);
5668 * ixgbe_service_task - manages and runs subtasks
5669 * @work: pointer to work_struct containing our data
5671 static void ixgbe_service_task(struct work_struct *work)
5673 struct ixgbe_adapter *adapter = container_of(work,
5674 struct ixgbe_adapter,
5677 ixgbe_reset_subtask(adapter);
5678 ixgbe_sfp_detection_subtask(adapter);
5679 ixgbe_sfp_link_config_subtask(adapter);
5680 ixgbe_check_overtemp_subtask(adapter);
5681 ixgbe_watchdog_subtask(adapter);
5682 ixgbe_fdir_reinit_subtask(adapter);
5683 ixgbe_check_hang_subtask(adapter);
5685 ixgbe_service_event_complete(adapter);
5688 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
5689 struct ixgbe_tx_buffer *first,
5692 struct sk_buff *skb = first->skb;
5693 u32 vlan_macip_lens, type_tucmd;
5694 u32 mss_l4len_idx, l4len;
5696 if (!skb_is_gso(skb))
5699 if (skb_header_cloned(skb)) {
5700 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5705 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5706 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
5708 if (first->protocol == __constant_htons(ETH_P_IP)) {
5709 struct iphdr *iph = ip_hdr(skb);
5712 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5716 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5717 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5718 IXGBE_TX_FLAGS_CSUM |
5719 IXGBE_TX_FLAGS_IPV4;
5720 } else if (skb_is_gso_v6(skb)) {
5721 ipv6_hdr(skb)->payload_len = 0;
5722 tcp_hdr(skb)->check =
5723 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5724 &ipv6_hdr(skb)->daddr,
5726 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5727 IXGBE_TX_FLAGS_CSUM;
5730 /* compute header lengths */
5731 l4len = tcp_hdrlen(skb);
5732 *hdr_len = skb_transport_offset(skb) + l4len;
5734 /* update gso size and bytecount with header size */
5735 first->gso_segs = skb_shinfo(skb)->gso_segs;
5736 first->bytecount += (first->gso_segs - 1) * *hdr_len;
5738 /* mss_l4len_id: use 1 as index for TSO */
5739 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
5740 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
5741 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
5743 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
5744 vlan_macip_lens = skb_network_header_len(skb);
5745 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5746 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5748 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
5754 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
5755 struct ixgbe_tx_buffer *first)
5757 struct sk_buff *skb = first->skb;
5758 u32 vlan_macip_lens = 0;
5759 u32 mss_l4len_idx = 0;
5762 if (skb->ip_summed != CHECKSUM_PARTIAL) {
5763 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
5764 !(first->tx_flags & IXGBE_TX_FLAGS_TXSW))
5768 switch (first->protocol) {
5769 case __constant_htons(ETH_P_IP):
5770 vlan_macip_lens |= skb_network_header_len(skb);
5771 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5772 l4_hdr = ip_hdr(skb)->protocol;
5774 case __constant_htons(ETH_P_IPV6):
5775 vlan_macip_lens |= skb_network_header_len(skb);
5776 l4_hdr = ipv6_hdr(skb)->nexthdr;
5779 if (unlikely(net_ratelimit())) {
5780 dev_warn(tx_ring->dev,
5781 "partial checksum but proto=%x!\n",
5789 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5790 mss_l4len_idx = tcp_hdrlen(skb) <<
5791 IXGBE_ADVTXD_L4LEN_SHIFT;
5794 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5795 mss_l4len_idx = sizeof(struct sctphdr) <<
5796 IXGBE_ADVTXD_L4LEN_SHIFT;
5799 mss_l4len_idx = sizeof(struct udphdr) <<
5800 IXGBE_ADVTXD_L4LEN_SHIFT;
5803 if (unlikely(net_ratelimit())) {
5804 dev_warn(tx_ring->dev,
5805 "partial checksum but l4 proto=%x!\n",
5811 /* update TX checksum flag */
5812 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
5815 /* vlan_macip_lens: MACLEN, VLAN tag */
5816 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5817 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5819 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
5820 type_tucmd, mss_l4len_idx);
5823 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
5825 /* set type for advanced descriptor with frame checksum insertion */
5826 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
5827 IXGBE_ADVTXD_DCMD_IFCS |
5828 IXGBE_ADVTXD_DCMD_DEXT);
5830 /* set HW vlan bit if vlan is present */
5831 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
5832 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
5834 /* set segmentation enable bits for TSO/FSO */
5836 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FSO))
5838 if (tx_flags & IXGBE_TX_FLAGS_TSO)
5840 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
5845 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
5846 u32 tx_flags, unsigned int paylen)
5848 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
5850 /* enable L4 checksum for TSO and TX checksum offload */
5851 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5852 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
5854 /* enble IPv4 checksum for TSO */
5855 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5856 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
5858 /* use index 1 context for TSO/FSO/FCOE */
5860 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FCOE))
5862 if (tx_flags & IXGBE_TX_FLAGS_TSO)
5864 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
5867 * Check Context must be set if Tx switch is enabled, which it
5868 * always is for case where virtual functions are running
5871 if (tx_flags & (IXGBE_TX_FLAGS_TXSW | IXGBE_TX_FLAGS_FCOE))
5873 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
5875 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
5877 tx_desc->read.olinfo_status = olinfo_status;
5880 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
5883 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
5884 struct ixgbe_tx_buffer *first,
5888 struct sk_buff *skb = first->skb;
5889 struct ixgbe_tx_buffer *tx_buffer;
5890 union ixgbe_adv_tx_desc *tx_desc;
5891 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
5892 unsigned int data_len = skb->data_len;
5893 unsigned int size = skb_headlen(skb);
5894 unsigned int paylen = skb->len - hdr_len;
5895 u32 tx_flags = first->tx_flags;
5897 u16 i = tx_ring->next_to_use;
5899 tx_desc = IXGBE_TX_DESC(tx_ring, i);
5901 ixgbe_tx_olinfo_status(tx_desc, tx_flags, paylen);
5902 cmd_type = ixgbe_tx_cmd_type(tx_flags);
5905 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5906 if (data_len < sizeof(struct fcoe_crc_eof)) {
5907 size -= sizeof(struct fcoe_crc_eof) - data_len;
5910 data_len -= sizeof(struct fcoe_crc_eof);
5915 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
5916 if (dma_mapping_error(tx_ring->dev, dma))
5919 /* record length, and DMA address */
5920 dma_unmap_len_set(first, len, size);
5921 dma_unmap_addr_set(first, dma, dma);
5923 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5926 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
5927 tx_desc->read.cmd_type_len =
5928 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
5932 if (i == tx_ring->count) {
5933 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5937 dma += IXGBE_MAX_DATA_PER_TXD;
5938 size -= IXGBE_MAX_DATA_PER_TXD;
5940 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5941 tx_desc->read.olinfo_status = 0;
5944 if (likely(!data_len))
5947 if (unlikely(skb->no_fcs))
5948 cmd_type &= ~(cpu_to_le32(IXGBE_ADVTXD_DCMD_IFCS));
5949 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
5953 if (i == tx_ring->count) {
5954 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5959 size = min_t(unsigned int, data_len, skb_frag_size(frag));
5961 size = skb_frag_size(frag);
5965 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
5967 if (dma_mapping_error(tx_ring->dev, dma))
5970 tx_buffer = &tx_ring->tx_buffer_info[i];
5971 dma_unmap_len_set(tx_buffer, len, size);
5972 dma_unmap_addr_set(tx_buffer, dma, dma);
5974 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5975 tx_desc->read.olinfo_status = 0;
5980 /* write last descriptor with RS and EOP bits */
5981 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
5982 tx_desc->read.cmd_type_len = cmd_type;
5984 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
5986 /* set the timestamp */
5987 first->time_stamp = jiffies;
5990 * Force memory writes to complete before letting h/w know there
5991 * are new descriptors to fetch. (Only applicable for weak-ordered
5992 * memory model archs, such as IA-64).
5994 * We also need this memory barrier to make certain all of the
5995 * status bits have been updated before next_to_watch is written.
5999 /* set next_to_watch value indicating a packet is present */
6000 first->next_to_watch = tx_desc;
6003 if (i == tx_ring->count)
6006 tx_ring->next_to_use = i;
6008 /* notify HW of packet */
6009 writel(i, tx_ring->tail);
6013 dev_err(tx_ring->dev, "TX DMA map failed\n");
6015 /* clear dma mappings for failed tx_buffer_info map */
6017 tx_buffer = &tx_ring->tx_buffer_info[i];
6018 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6019 if (tx_buffer == first)
6026 tx_ring->next_to_use = i;
6029 static void ixgbe_atr(struct ixgbe_ring *ring,
6030 struct ixgbe_tx_buffer *first)
6032 struct ixgbe_q_vector *q_vector = ring->q_vector;
6033 union ixgbe_atr_hash_dword input = { .dword = 0 };
6034 union ixgbe_atr_hash_dword common = { .dword = 0 };
6036 unsigned char *network;
6038 struct ipv6hdr *ipv6;
6043 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6047 /* do nothing if sampling is disabled */
6048 if (!ring->atr_sample_rate)
6053 /* snag network header to get L4 type and address */
6054 hdr.network = skb_network_header(first->skb);
6056 /* Currently only IPv4/IPv6 with TCP is supported */
6057 if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6058 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6059 (first->protocol != __constant_htons(ETH_P_IP) ||
6060 hdr.ipv4->protocol != IPPROTO_TCP))
6063 th = tcp_hdr(first->skb);
6065 /* skip this packet since it is invalid or the socket is closing */
6069 /* sample on all syn packets or once every atr sample count */
6070 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6073 /* reset sample count */
6074 ring->atr_count = 0;
6076 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6079 * src and dst are inverted, think how the receiver sees them
6081 * The input is broken into two sections, a non-compressed section
6082 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6083 * is XORed together and stored in the compressed dword.
6085 input.formatted.vlan_id = vlan_id;
6088 * since src port and flex bytes occupy the same word XOR them together
6089 * and write the value to source port portion of compressed dword
6091 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6092 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6094 common.port.src ^= th->dest ^ first->protocol;
6095 common.port.dst ^= th->source;
6097 if (first->protocol == __constant_htons(ETH_P_IP)) {
6098 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6099 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6101 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6102 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6103 hdr.ipv6->saddr.s6_addr32[1] ^
6104 hdr.ipv6->saddr.s6_addr32[2] ^
6105 hdr.ipv6->saddr.s6_addr32[3] ^
6106 hdr.ipv6->daddr.s6_addr32[0] ^
6107 hdr.ipv6->daddr.s6_addr32[1] ^
6108 hdr.ipv6->daddr.s6_addr32[2] ^
6109 hdr.ipv6->daddr.s6_addr32[3];
6112 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6113 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6114 input, common, ring->queue_index);
6117 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6119 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6120 /* Herbert's original patch had:
6121 * smp_mb__after_netif_stop_queue();
6122 * but since that doesn't exist yet, just open code it. */
6125 /* We need to check again in a case another CPU has just
6126 * made room available. */
6127 if (likely(ixgbe_desc_unused(tx_ring) < size))
6130 /* A reprieve! - use start_queue because it doesn't call schedule */
6131 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6132 ++tx_ring->tx_stats.restart_queue;
6136 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6138 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6140 return __ixgbe_maybe_stop_tx(tx_ring, size);
6143 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6145 struct ixgbe_adapter *adapter = netdev_priv(dev);
6146 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6149 __be16 protocol = vlan_get_protocol(skb);
6151 if (((protocol == htons(ETH_P_FCOE)) ||
6152 (protocol == htons(ETH_P_FIP))) &&
6153 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6154 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6155 txq += adapter->ring_feature[RING_F_FCOE].mask;
6160 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6161 while (unlikely(txq >= dev->real_num_tx_queues))
6162 txq -= dev->real_num_tx_queues;
6166 return skb_tx_hash(dev, skb);
6169 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6170 struct ixgbe_adapter *adapter,
6171 struct ixgbe_ring *tx_ring)
6173 struct ixgbe_tx_buffer *first;
6176 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6179 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6180 __be16 protocol = skb->protocol;
6184 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6185 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6186 * + 2 desc gap to keep tail from touching head,
6187 * + 1 desc for context descriptor,
6188 * otherwise try next time
6190 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6191 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6192 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6194 count += skb_shinfo(skb)->nr_frags;
6196 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6197 tx_ring->tx_stats.tx_busy++;
6198 return NETDEV_TX_BUSY;
6201 /* record the location of the first descriptor for this packet */
6202 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6204 first->bytecount = skb->len;
6205 first->gso_segs = 1;
6207 /* if we have a HW VLAN tag being added default to the HW one */
6208 if (vlan_tx_tag_present(skb)) {
6209 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6210 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6211 /* else if it is a SW VLAN check the next protocol and store the tag */
6212 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6213 struct vlan_hdr *vhdr, _vhdr;
6214 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6218 protocol = vhdr->h_vlan_encapsulated_proto;
6219 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6220 IXGBE_TX_FLAGS_VLAN_SHIFT;
6221 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6224 #ifdef CONFIG_PCI_IOV
6226 * Use the l2switch_enable flag - would be false if the DMA
6227 * Tx switch had been disabled.
6229 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6230 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6233 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6234 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6235 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6236 (skb->priority != TC_PRIO_CONTROL))) {
6237 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6238 tx_flags |= (skb->priority & 0x7) <<
6239 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6240 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6241 struct vlan_ethhdr *vhdr;
6242 if (skb_header_cloned(skb) &&
6243 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6245 vhdr = (struct vlan_ethhdr *)skb->data;
6246 vhdr->h_vlan_TCI = htons(tx_flags >>
6247 IXGBE_TX_FLAGS_VLAN_SHIFT);
6249 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6253 /* record initial flags and protocol */
6254 first->tx_flags = tx_flags;
6255 first->protocol = protocol;
6258 /* setup tx offload for FCoE */
6259 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6260 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6261 tso = ixgbe_fso(tx_ring, first, &hdr_len);
6268 #endif /* IXGBE_FCOE */
6269 tso = ixgbe_tso(tx_ring, first, &hdr_len);
6273 ixgbe_tx_csum(tx_ring, first);
6275 /* add the ATR filter if ATR is on */
6276 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6277 ixgbe_atr(tx_ring, first);
6281 #endif /* IXGBE_FCOE */
6282 ixgbe_tx_map(tx_ring, first, hdr_len);
6284 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6286 return NETDEV_TX_OK;
6289 dev_kfree_skb_any(first->skb);
6292 return NETDEV_TX_OK;
6295 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6296 struct net_device *netdev)
6298 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6299 struct ixgbe_ring *tx_ring;
6301 if (skb->len <= 0) {
6302 dev_kfree_skb_any(skb);
6303 return NETDEV_TX_OK;
6307 * The minimum packet size for olinfo paylen is 17 so pad the skb
6308 * in order to meet this minimum size requirement.
6310 if (skb->len < 17) {
6311 if (skb_padto(skb, 17))
6312 return NETDEV_TX_OK;
6316 tx_ring = adapter->tx_ring[skb->queue_mapping];
6317 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6321 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6322 * @netdev: network interface device structure
6323 * @p: pointer to an address structure
6325 * Returns 0 on success, negative on failure
6327 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6329 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6330 struct ixgbe_hw *hw = &adapter->hw;
6331 struct sockaddr *addr = p;
6333 if (!is_valid_ether_addr(addr->sa_data))
6334 return -EADDRNOTAVAIL;
6336 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6337 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6339 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6346 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6348 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6349 struct ixgbe_hw *hw = &adapter->hw;
6353 if (prtad != hw->phy.mdio.prtad)
6355 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6361 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6362 u16 addr, u16 value)
6364 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6365 struct ixgbe_hw *hw = &adapter->hw;
6367 if (prtad != hw->phy.mdio.prtad)
6369 return hw->phy.ops.write_reg(hw, addr, devad, value);
6372 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6374 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6376 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6380 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6382 * @netdev: network interface device structure
6384 * Returns non-zero on failure
6386 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6389 struct ixgbe_adapter *adapter = netdev_priv(dev);
6390 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6392 if (is_valid_ether_addr(mac->san_addr)) {
6394 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6401 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6403 * @netdev: network interface device structure
6405 * Returns non-zero on failure
6407 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6410 struct ixgbe_adapter *adapter = netdev_priv(dev);
6411 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6413 if (is_valid_ether_addr(mac->san_addr)) {
6415 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6421 #ifdef CONFIG_NET_POLL_CONTROLLER
6423 * Polling 'interrupt' - used by things like netconsole to send skbs
6424 * without having to re-enable interrupts. It's not called while
6425 * the interrupt routine is executing.
6427 static void ixgbe_netpoll(struct net_device *netdev)
6429 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6432 /* if interface is down do nothing */
6433 if (test_bit(__IXGBE_DOWN, &adapter->state))
6436 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6437 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6438 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6439 for (i = 0; i < num_q_vectors; i++) {
6440 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6441 ixgbe_msix_clean_rings(0, q_vector);
6444 ixgbe_intr(adapter->pdev->irq, netdev);
6446 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6450 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6451 struct rtnl_link_stats64 *stats)
6453 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6457 for (i = 0; i < adapter->num_rx_queues; i++) {
6458 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6464 start = u64_stats_fetch_begin_bh(&ring->syncp);
6465 packets = ring->stats.packets;
6466 bytes = ring->stats.bytes;
6467 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6468 stats->rx_packets += packets;
6469 stats->rx_bytes += bytes;
6473 for (i = 0; i < adapter->num_tx_queues; i++) {
6474 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6480 start = u64_stats_fetch_begin_bh(&ring->syncp);
6481 packets = ring->stats.packets;
6482 bytes = ring->stats.bytes;
6483 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6484 stats->tx_packets += packets;
6485 stats->tx_bytes += bytes;
6489 /* following stats updated by ixgbe_watchdog_task() */
6490 stats->multicast = netdev->stats.multicast;
6491 stats->rx_errors = netdev->stats.rx_errors;
6492 stats->rx_length_errors = netdev->stats.rx_length_errors;
6493 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6494 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6498 #ifdef CONFIG_IXGBE_DCB
6499 /* ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6500 * #adapter: pointer to ixgbe_adapter
6501 * @tc: number of traffic classes currently enabled
6503 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6504 * 802.1Q priority maps to a packet buffer that exists.
6506 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6508 struct ixgbe_hw *hw = &adapter->hw;
6512 /* 82598 have a static priority to TC mapping that can not
6513 * be changed so no validation is needed.
6515 if (hw->mac.type == ixgbe_mac_82598EB)
6518 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6521 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6522 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6524 /* If up2tc is out of bounds default to zero */
6526 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6530 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6535 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
6538 * @netdev: net device to configure
6539 * @tc: number of traffic classes to enable
6541 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6543 struct ixgbe_adapter *adapter = netdev_priv(dev);
6544 struct ixgbe_hw *hw = &adapter->hw;
6546 /* Multiple traffic classes requires multiple queues */
6547 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6548 e_err(drv, "Enable failed, needs MSI-X\n");
6552 /* Hardware supports up to 8 traffic classes */
6553 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
6554 (hw->mac.type == ixgbe_mac_82598EB &&
6555 tc < MAX_TRAFFIC_CLASS))
6558 /* Hardware has to reinitialize queues and interrupts to
6559 * match packet buffer alignment. Unfortunately, the
6560 * hardware is not flexible enough to do this dynamically.
6562 if (netif_running(dev))
6564 ixgbe_clear_interrupt_scheme(adapter);
6567 netdev_set_num_tc(dev, tc);
6568 adapter->last_lfc_mode = adapter->hw.fc.current_mode;
6569 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6570 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6572 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6573 adapter->hw.fc.requested_mode = ixgbe_fc_none;
6575 netdev_reset_tc(dev);
6576 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6578 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6579 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6581 adapter->temp_dcb_cfg.pfc_mode_enable = false;
6582 adapter->dcb_cfg.pfc_mode_enable = false;
6585 ixgbe_init_interrupt_scheme(adapter);
6586 ixgbe_validate_rtr(adapter, tc);
6587 if (netif_running(dev))
6593 #endif /* CONFIG_IXGBE_DCB */
6594 void ixgbe_do_reset(struct net_device *netdev)
6596 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6598 if (netif_running(netdev))
6599 ixgbe_reinit_locked(adapter);
6601 ixgbe_reset(adapter);
6604 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
6605 netdev_features_t features)
6607 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6610 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6611 features &= ~NETIF_F_HW_VLAN_RX;
6614 /* return error if RXHASH is being enabled when RSS is not supported */
6615 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
6616 features &= ~NETIF_F_RXHASH;
6618 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6619 if (!(features & NETIF_F_RXCSUM))
6620 features &= ~NETIF_F_LRO;
6622 /* Turn off LRO if not RSC capable */
6623 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
6624 features &= ~NETIF_F_LRO;
6630 static int ixgbe_set_features(struct net_device *netdev,
6631 netdev_features_t features)
6633 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6634 netdev_features_t changed = netdev->features ^ features;
6635 bool need_reset = false;
6637 /* Make sure RSC matches LRO, reset if change */
6638 if (!(features & NETIF_F_LRO)) {
6639 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6641 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
6642 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
6643 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6644 if (adapter->rx_itr_setting == 1 ||
6645 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
6646 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
6648 } else if ((changed ^ features) & NETIF_F_LRO) {
6649 e_info(probe, "rx-usecs set too low, "
6655 * Check if Flow Director n-tuple support was enabled or disabled. If
6656 * the state changed, we need to reset.
6658 if (!(features & NETIF_F_NTUPLE)) {
6659 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
6660 /* turn off Flow Director, set ATR and reset */
6661 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
6662 !(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
6663 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6666 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6667 } else if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
6668 /* turn off ATR, enable perfect filters and reset */
6669 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6670 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6674 if (changed & NETIF_F_RXALL)
6677 netdev->features = features;
6679 ixgbe_do_reset(netdev);
6684 static const struct net_device_ops ixgbe_netdev_ops = {
6685 .ndo_open = ixgbe_open,
6686 .ndo_stop = ixgbe_close,
6687 .ndo_start_xmit = ixgbe_xmit_frame,
6688 .ndo_select_queue = ixgbe_select_queue,
6689 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6690 .ndo_validate_addr = eth_validate_addr,
6691 .ndo_set_mac_address = ixgbe_set_mac,
6692 .ndo_change_mtu = ixgbe_change_mtu,
6693 .ndo_tx_timeout = ixgbe_tx_timeout,
6694 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6695 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6696 .ndo_do_ioctl = ixgbe_ioctl,
6697 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6698 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6699 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6700 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
6701 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6702 .ndo_get_stats64 = ixgbe_get_stats64,
6703 #ifdef CONFIG_IXGBE_DCB
6704 .ndo_setup_tc = ixgbe_setup_tc,
6706 #ifdef CONFIG_NET_POLL_CONTROLLER
6707 .ndo_poll_controller = ixgbe_netpoll,
6710 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6711 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
6712 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6713 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6714 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6715 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6716 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
6717 #endif /* IXGBE_FCOE */
6718 .ndo_set_features = ixgbe_set_features,
6719 .ndo_fix_features = ixgbe_fix_features,
6722 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6723 const struct ixgbe_info *ii)
6725 #ifdef CONFIG_PCI_IOV
6726 struct ixgbe_hw *hw = &adapter->hw;
6728 if (hw->mac.type == ixgbe_mac_82598EB)
6731 /* The 82599 supports up to 64 VFs per physical function
6732 * but this implementation limits allocation to 63 so that
6733 * basic networking resources are still available to the
6736 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6737 ixgbe_enable_sriov(adapter, ii);
6738 #endif /* CONFIG_PCI_IOV */
6742 * ixgbe_probe - Device Initialization Routine
6743 * @pdev: PCI device information struct
6744 * @ent: entry in ixgbe_pci_tbl
6746 * Returns 0 on success, negative on failure
6748 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6749 * The OS initialization, configuring of the adapter private structure,
6750 * and a hardware reset occur.
6752 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6753 const struct pci_device_id *ent)
6755 struct net_device *netdev;
6756 struct ixgbe_adapter *adapter = NULL;
6757 struct ixgbe_hw *hw;
6758 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6759 static int cards_found;
6760 int i, err, pci_using_dac;
6761 u8 part_str[IXGBE_PBANUM_LENGTH];
6762 unsigned int indices = num_possible_cpus();
6769 /* Catch broken hardware that put the wrong VF device ID in
6770 * the PCIe SR-IOV capability.
6772 if (pdev->is_virtfn) {
6773 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6774 pci_name(pdev), pdev->vendor, pdev->device);
6778 err = pci_enable_device_mem(pdev);
6782 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6783 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6786 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6788 err = dma_set_coherent_mask(&pdev->dev,
6792 "No usable DMA configuration, aborting\n");
6799 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6800 IORESOURCE_MEM), ixgbe_driver_name);
6803 "pci_request_selected_regions failed 0x%x\n", err);
6807 pci_enable_pcie_error_reporting(pdev);
6809 pci_set_master(pdev);
6810 pci_save_state(pdev);
6812 #ifdef CONFIG_IXGBE_DCB
6813 indices *= MAX_TRAFFIC_CLASS;
6816 if (ii->mac == ixgbe_mac_82598EB)
6817 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6819 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6822 indices += min_t(unsigned int, num_possible_cpus(),
6823 IXGBE_MAX_FCOE_INDICES);
6825 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6828 goto err_alloc_etherdev;
6831 SET_NETDEV_DEV(netdev, &pdev->dev);
6833 adapter = netdev_priv(netdev);
6834 pci_set_drvdata(pdev, adapter);
6836 adapter->netdev = netdev;
6837 adapter->pdev = pdev;
6840 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6842 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6843 pci_resource_len(pdev, 0));
6849 for (i = 1; i <= 5; i++) {
6850 if (pci_resource_len(pdev, i) == 0)
6854 netdev->netdev_ops = &ixgbe_netdev_ops;
6855 ixgbe_set_ethtool_ops(netdev);
6856 netdev->watchdog_timeo = 5 * HZ;
6857 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
6859 adapter->bd_number = cards_found;
6862 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6863 hw->mac.type = ii->mac;
6866 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6867 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6868 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6869 if (!(eec & (1 << 8)))
6870 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6873 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6874 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6875 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6876 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6877 hw->phy.mdio.mmds = 0;
6878 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6879 hw->phy.mdio.dev = netdev;
6880 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6881 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6883 ii->get_invariants(hw);
6885 /* setup the private structure */
6886 err = ixgbe_sw_init(adapter);
6890 /* Make it possible the adapter to be woken up via WOL */
6891 switch (adapter->hw.mac.type) {
6892 case ixgbe_mac_82599EB:
6893 case ixgbe_mac_X540:
6894 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6901 * If there is a fan on this device and it has failed log the
6904 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6905 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6906 if (esdp & IXGBE_ESDP_SDP1)
6907 e_crit(probe, "Fan has stopped, replace the adapter\n");
6910 if (allow_unsupported_sfp)
6911 hw->allow_unsupported_sfp = allow_unsupported_sfp;
6913 /* reset_hw fills in the perm_addr as well */
6914 hw->phy.reset_if_overtemp = true;
6915 err = hw->mac.ops.reset_hw(hw);
6916 hw->phy.reset_if_overtemp = false;
6917 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6918 hw->mac.type == ixgbe_mac_82598EB) {
6920 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6921 e_dev_err("failed to load because an unsupported SFP+ "
6922 "module type was detected.\n");
6923 e_dev_err("Reload the driver after installing a supported "
6927 e_dev_err("HW Init failed: %d\n", err);
6931 ixgbe_probe_vf(adapter, ii);
6933 netdev->features = NETIF_F_SG |
6936 NETIF_F_HW_VLAN_TX |
6937 NETIF_F_HW_VLAN_RX |
6938 NETIF_F_HW_VLAN_FILTER |
6944 netdev->hw_features = netdev->features;
6946 switch (adapter->hw.mac.type) {
6947 case ixgbe_mac_82599EB:
6948 case ixgbe_mac_X540:
6949 netdev->features |= NETIF_F_SCTP_CSUM;
6950 netdev->hw_features |= NETIF_F_SCTP_CSUM |
6957 netdev->hw_features |= NETIF_F_RXALL;
6959 netdev->vlan_features |= NETIF_F_TSO;
6960 netdev->vlan_features |= NETIF_F_TSO6;
6961 netdev->vlan_features |= NETIF_F_IP_CSUM;
6962 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6963 netdev->vlan_features |= NETIF_F_SG;
6965 netdev->priv_flags |= IFF_UNICAST_FLT;
6966 netdev->priv_flags |= IFF_SUPP_NOFCS;
6968 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6969 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6970 IXGBE_FLAG_DCB_ENABLED);
6972 #ifdef CONFIG_IXGBE_DCB
6973 netdev->dcbnl_ops = &dcbnl_ops;
6977 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6978 if (hw->mac.ops.get_device_caps) {
6979 hw->mac.ops.get_device_caps(hw, &device_caps);
6980 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6981 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6984 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6985 netdev->vlan_features |= NETIF_F_FCOE_CRC;
6986 netdev->vlan_features |= NETIF_F_FSO;
6987 netdev->vlan_features |= NETIF_F_FCOE_MTU;
6989 #endif /* IXGBE_FCOE */
6990 if (pci_using_dac) {
6991 netdev->features |= NETIF_F_HIGHDMA;
6992 netdev->vlan_features |= NETIF_F_HIGHDMA;
6995 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
6996 netdev->hw_features |= NETIF_F_LRO;
6997 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6998 netdev->features |= NETIF_F_LRO;
7000 /* make sure the EEPROM is good */
7001 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7002 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7007 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7008 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7010 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7011 e_dev_err("invalid MAC address\n");
7016 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7017 (unsigned long) adapter);
7019 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7020 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7022 err = ixgbe_init_interrupt_scheme(adapter);
7026 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7027 netdev->hw_features &= ~NETIF_F_RXHASH;
7028 netdev->features &= ~NETIF_F_RXHASH;
7031 /* WOL not supported for all but the following */
7033 switch (pdev->device) {
7034 case IXGBE_DEV_ID_82599_SFP:
7035 /* Only these subdevice supports WOL */
7036 switch (pdev->subsystem_device) {
7037 case IXGBE_SUBDEV_ID_82599_560FLR:
7038 /* only support first port */
7039 if (hw->bus.func != 0)
7041 case IXGBE_SUBDEV_ID_82599_SFP:
7042 adapter->wol = IXGBE_WUFC_MAG;
7046 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7047 /* All except this subdevice support WOL */
7048 if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7049 adapter->wol = IXGBE_WUFC_MAG;
7051 case IXGBE_DEV_ID_82599_KX4:
7052 adapter->wol = IXGBE_WUFC_MAG;
7054 case IXGBE_DEV_ID_X540T:
7055 /* Check eeprom to see if it is enabled */
7056 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7057 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7059 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7060 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7061 (hw->bus.func == 0)))
7062 adapter->wol = IXGBE_WUFC_MAG;
7065 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7067 /* save off EEPROM version number */
7068 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7069 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7071 /* pick up the PCI bus settings for reporting later */
7072 hw->mac.ops.get_bus_info(hw);
7074 /* print bus type/speed/width info */
7075 e_dev_info("(PCI Express:%s:%s) %pM\n",
7076 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7077 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7079 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7080 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7081 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7085 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7087 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7088 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7089 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7090 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7093 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7094 hw->mac.type, hw->phy.type, part_str);
7096 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7097 e_dev_warn("PCI-Express bandwidth available for this card is "
7098 "not sufficient for optimal performance.\n");
7099 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7103 /* reset the hardware with the new settings */
7104 err = hw->mac.ops.start_hw(hw);
7105 if (err == IXGBE_ERR_EEPROM_VERSION) {
7106 /* We are running on a pre-production device, log a warning */
7107 e_dev_warn("This device is a pre-production adapter/LOM. "
7108 "Please be aware there may be issues associated "
7109 "with your hardware. If you are experiencing "
7110 "problems please contact your Intel or hardware "
7111 "representative who provided you with this "
7114 strcpy(netdev->name, "eth%d");
7115 err = register_netdev(netdev);
7119 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7120 if (hw->mac.ops.disable_tx_laser &&
7121 ((hw->phy.multispeed_fiber) ||
7122 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7123 (hw->mac.type == ixgbe_mac_82599EB))))
7124 hw->mac.ops.disable_tx_laser(hw);
7126 /* carrier off reporting is important to ethtool even BEFORE open */
7127 netif_carrier_off(netdev);
7129 #ifdef CONFIG_IXGBE_DCA
7130 if (dca_add_requester(&pdev->dev) == 0) {
7131 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7132 ixgbe_setup_dca(adapter);
7135 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7136 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7137 for (i = 0; i < adapter->num_vfs; i++)
7138 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7141 /* firmware requires driver version to be 0xFFFFFFFF
7142 * since os does not support feature
7144 if (hw->mac.ops.set_fw_drv_ver)
7145 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7148 /* add san mac addr to netdev */
7149 ixgbe_add_sanmac_netdev(netdev);
7151 e_dev_info("%s\n", ixgbe_default_device_descr);
7156 ixgbe_release_hw_control(adapter);
7157 ixgbe_clear_interrupt_scheme(adapter);
7159 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7160 ixgbe_disable_sriov(adapter);
7161 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7162 iounmap(hw->hw_addr);
7164 free_netdev(netdev);
7166 pci_release_selected_regions(pdev,
7167 pci_select_bars(pdev, IORESOURCE_MEM));
7170 pci_disable_device(pdev);
7175 * ixgbe_remove - Device Removal Routine
7176 * @pdev: PCI device information struct
7178 * ixgbe_remove is called by the PCI subsystem to alert the driver
7179 * that it should release a PCI device. The could be caused by a
7180 * Hot-Plug event, or because the driver is going to be removed from
7183 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7185 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7186 struct net_device *netdev = adapter->netdev;
7188 set_bit(__IXGBE_DOWN, &adapter->state);
7189 cancel_work_sync(&adapter->service_task);
7191 #ifdef CONFIG_IXGBE_DCA
7192 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7193 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7194 dca_remove_requester(&pdev->dev);
7195 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7200 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7201 ixgbe_cleanup_fcoe(adapter);
7203 #endif /* IXGBE_FCOE */
7205 /* remove the added san mac */
7206 ixgbe_del_sanmac_netdev(netdev);
7208 if (netdev->reg_state == NETREG_REGISTERED)
7209 unregister_netdev(netdev);
7211 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7212 if (!(ixgbe_check_vf_assignment(adapter)))
7213 ixgbe_disable_sriov(adapter);
7215 e_dev_warn("Unloading driver while VFs are assigned "
7216 "- VFs will not be deallocated\n");
7219 ixgbe_clear_interrupt_scheme(adapter);
7221 ixgbe_release_hw_control(adapter);
7223 iounmap(adapter->hw.hw_addr);
7224 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7227 e_dev_info("complete\n");
7229 free_netdev(netdev);
7231 pci_disable_pcie_error_reporting(pdev);
7233 pci_disable_device(pdev);
7237 * ixgbe_io_error_detected - called when PCI error is detected
7238 * @pdev: Pointer to PCI device
7239 * @state: The current pci connection state
7241 * This function is called after a PCI bus error affecting
7242 * this device has been detected.
7244 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7245 pci_channel_state_t state)
7247 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7248 struct net_device *netdev = adapter->netdev;
7250 #ifdef CONFIG_PCI_IOV
7251 struct pci_dev *bdev, *vfdev;
7252 u32 dw0, dw1, dw2, dw3;
7254 u16 req_id, pf_func;
7256 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7257 adapter->num_vfs == 0)
7258 goto skip_bad_vf_detection;
7260 bdev = pdev->bus->self;
7261 while (bdev && (bdev->pcie_type != PCI_EXP_TYPE_ROOT_PORT))
7262 bdev = bdev->bus->self;
7265 goto skip_bad_vf_detection;
7267 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7269 goto skip_bad_vf_detection;
7271 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7272 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7273 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7274 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7277 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7278 if (!(req_id & 0x0080))
7279 goto skip_bad_vf_detection;
7281 pf_func = req_id & 0x01;
7282 if ((pf_func & 1) == (pdev->devfn & 1)) {
7283 unsigned int device_id;
7285 vf = (req_id & 0x7F) >> 1;
7286 e_dev_err("VF %d has caused a PCIe error\n", vf);
7287 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7288 "%8.8x\tdw3: %8.8x\n",
7289 dw0, dw1, dw2, dw3);
7290 switch (adapter->hw.mac.type) {
7291 case ixgbe_mac_82599EB:
7292 device_id = IXGBE_82599_VF_DEVICE_ID;
7294 case ixgbe_mac_X540:
7295 device_id = IXGBE_X540_VF_DEVICE_ID;
7302 /* Find the pci device of the offending VF */
7303 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID, device_id, NULL);
7305 if (vfdev->devfn == (req_id & 0xFF))
7307 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID,
7311 * There's a slim chance the VF could have been hot plugged,
7312 * so if it is no longer present we don't need to issue the
7313 * VFLR. Just clean up the AER in that case.
7316 e_dev_err("Issuing VFLR to VF %d\n", vf);
7317 pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7320 pci_cleanup_aer_uncorrect_error_status(pdev);
7324 * Even though the error may have occurred on the other port
7325 * we still need to increment the vf error reference count for
7326 * both ports because the I/O resume function will be called
7329 adapter->vferr_refcount++;
7331 return PCI_ERS_RESULT_RECOVERED;
7333 skip_bad_vf_detection:
7334 #endif /* CONFIG_PCI_IOV */
7335 netif_device_detach(netdev);
7337 if (state == pci_channel_io_perm_failure)
7338 return PCI_ERS_RESULT_DISCONNECT;
7340 if (netif_running(netdev))
7341 ixgbe_down(adapter);
7342 pci_disable_device(pdev);
7344 /* Request a slot reset. */
7345 return PCI_ERS_RESULT_NEED_RESET;
7349 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7350 * @pdev: Pointer to PCI device
7352 * Restart the card from scratch, as if from a cold-boot.
7354 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7356 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7357 pci_ers_result_t result;
7360 if (pci_enable_device_mem(pdev)) {
7361 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7362 result = PCI_ERS_RESULT_DISCONNECT;
7364 pci_set_master(pdev);
7365 pci_restore_state(pdev);
7366 pci_save_state(pdev);
7368 pci_wake_from_d3(pdev, false);
7370 ixgbe_reset(adapter);
7371 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7372 result = PCI_ERS_RESULT_RECOVERED;
7375 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7377 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7378 "failed 0x%0x\n", err);
7379 /* non-fatal, continue */
7386 * ixgbe_io_resume - called when traffic can start flowing again.
7387 * @pdev: Pointer to PCI device
7389 * This callback is called when the error recovery driver tells us that
7390 * its OK to resume normal operation.
7392 static void ixgbe_io_resume(struct pci_dev *pdev)
7394 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7395 struct net_device *netdev = adapter->netdev;
7397 #ifdef CONFIG_PCI_IOV
7398 if (adapter->vferr_refcount) {
7399 e_info(drv, "Resuming after VF err\n");
7400 adapter->vferr_refcount--;
7405 if (netif_running(netdev))
7408 netif_device_attach(netdev);
7411 static struct pci_error_handlers ixgbe_err_handler = {
7412 .error_detected = ixgbe_io_error_detected,
7413 .slot_reset = ixgbe_io_slot_reset,
7414 .resume = ixgbe_io_resume,
7417 static struct pci_driver ixgbe_driver = {
7418 .name = ixgbe_driver_name,
7419 .id_table = ixgbe_pci_tbl,
7420 .probe = ixgbe_probe,
7421 .remove = __devexit_p(ixgbe_remove),
7423 .suspend = ixgbe_suspend,
7424 .resume = ixgbe_resume,
7426 .shutdown = ixgbe_shutdown,
7427 .err_handler = &ixgbe_err_handler
7431 * ixgbe_init_module - Driver Registration Routine
7433 * ixgbe_init_module is the first routine called when the driver is
7434 * loaded. All it does is register with the PCI subsystem.
7436 static int __init ixgbe_init_module(void)
7439 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7440 pr_info("%s\n", ixgbe_copyright);
7442 #ifdef CONFIG_IXGBE_DCA
7443 dca_register_notify(&dca_notifier);
7446 ret = pci_register_driver(&ixgbe_driver);
7450 module_init(ixgbe_init_module);
7453 * ixgbe_exit_module - Driver Exit Cleanup Routine
7455 * ixgbe_exit_module is called just before the driver is removed
7458 static void __exit ixgbe_exit_module(void)
7460 #ifdef CONFIG_IXGBE_DCA
7461 dca_unregister_notify(&dca_notifier);
7463 pci_unregister_driver(&ixgbe_driver);
7464 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7467 #ifdef CONFIG_IXGBE_DCA
7468 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7473 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7474 __ixgbe_notify_dca);
7476 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7479 #endif /* CONFIG_IXGBE_DCA */
7481 module_exit(ixgbe_exit_module);