1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2013 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/if_bridge.h>
48 #include <linux/prefetch.h>
49 #include <scsi/fc/fc_fcoe.h>
52 #include "ixgbe_common.h"
53 #include "ixgbe_dcb_82599.h"
54 #include "ixgbe_sriov.h"
56 char ixgbe_driver_name[] = "ixgbe";
57 static const char ixgbe_driver_string[] =
58 "Intel(R) 10 Gigabit PCI Express Network Driver";
60 char ixgbe_default_device_descr[] =
61 "Intel(R) 10 Gigabit Network Connection";
63 static char ixgbe_default_device_descr[] =
64 "Intel(R) 10 Gigabit Network Connection";
66 #define DRV_VERSION "3.15.1-k"
67 const char ixgbe_driver_version[] = DRV_VERSION;
68 static const char ixgbe_copyright[] =
69 "Copyright (c) 1999-2013 Intel Corporation.";
71 static const struct ixgbe_info *ixgbe_info_tbl[] = {
72 [board_82598] = &ixgbe_82598_info,
73 [board_82599] = &ixgbe_82599_info,
74 [board_X540] = &ixgbe_X540_info,
77 /* ixgbe_pci_tbl - PCI Device ID Table
79 * Wildcard entries (PCI_ANY_ID) should come last
80 * Last entry must be all 0s
82 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
83 * Class, Class Mask, private data (not used) }
85 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
116 /* required last entry */
119 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
121 #ifdef CONFIG_IXGBE_DCA
122 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
124 static struct notifier_block dca_notifier = {
125 .notifier_call = ixgbe_notify_dca,
131 #ifdef CONFIG_PCI_IOV
132 static unsigned int max_vfs;
133 module_param(max_vfs, uint, 0);
134 MODULE_PARM_DESC(max_vfs,
135 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63");
136 #endif /* CONFIG_PCI_IOV */
138 static unsigned int allow_unsupported_sfp;
139 module_param(allow_unsupported_sfp, uint, 0);
140 MODULE_PARM_DESC(allow_unsupported_sfp,
141 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
143 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
144 static int debug = -1;
145 module_param(debug, int, 0);
146 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
148 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
149 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
150 MODULE_LICENSE("GPL");
151 MODULE_VERSION(DRV_VERSION);
153 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
157 struct pci_dev *parent_dev;
158 struct pci_bus *parent_bus;
160 parent_bus = adapter->pdev->bus->parent;
164 parent_dev = parent_bus->self;
168 pos = pci_find_capability(parent_dev, PCI_CAP_ID_EXP);
172 pci_read_config_word(parent_dev, pos + reg, value);
176 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
178 struct ixgbe_hw *hw = &adapter->hw;
182 hw->bus.type = ixgbe_bus_type_pci_express;
184 /* Get the negotiated link width and speed from PCI config space of the
185 * parent, as this device is behind a switch
187 err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
189 /* assume caller will handle error case */
193 hw->bus.width = ixgbe_convert_bus_width(link_status);
194 hw->bus.speed = ixgbe_convert_bus_speed(link_status);
200 * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
201 * @hw: hw specific details
203 * This function is used by probe to determine whether a device's PCI-Express
204 * bandwidth details should be gathered from the parent bus instead of from the
205 * device. Used to ensure that various locations all have the correct device ID
208 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
210 switch (hw->device_id) {
211 case IXGBE_DEV_ID_82599_SFP_SF_QP:
212 case IXGBE_DEV_ID_82599_QSFP_SF_QP:
219 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
223 enum pci_bus_speed speed = PCI_SPEED_UNKNOWN;
224 enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN;
225 struct pci_dev *pdev;
227 /* determine whether to use the the parent device
229 if (ixgbe_pcie_from_parent(&adapter->hw))
230 pdev = adapter->pdev->bus->parent->self;
232 pdev = adapter->pdev;
234 if (pcie_get_minimum_link(pdev, &speed, &width) ||
235 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
236 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
241 case PCIE_SPEED_2_5GT:
242 /* 8b/10b encoding reduces max throughput by 20% */
245 case PCIE_SPEED_5_0GT:
246 /* 8b/10b encoding reduces max throughput by 20% */
249 case PCIE_SPEED_8_0GT:
250 /* 128b/130b encoding only reduces throughput by 1% */
254 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
258 e_dev_info("PCI Express bandwidth of %dGT/s available\n",
260 e_dev_info("(Speed:%s, Width: x%d, Encoding Loss:%s)\n",
261 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" :
262 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" :
263 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" :
266 (speed == PCIE_SPEED_2_5GT ? "20%" :
267 speed == PCIE_SPEED_5_0GT ? "20%" :
268 speed == PCIE_SPEED_8_0GT ? "N/a" :
271 if (max_gts < expected_gts) {
272 e_dev_warn("This is not sufficient for optimal performance of this card.\n");
273 e_dev_warn("For optimal performance, at least %dGT/s of bandwidth is required.\n",
275 e_dev_warn("A slot with more lanes and/or higher speed is suggested.\n");
279 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
281 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
282 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
283 schedule_work(&adapter->service_task);
286 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
288 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
290 /* flush memory to make sure state is correct before next watchdog */
291 smp_mb__before_clear_bit();
292 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
295 struct ixgbe_reg_info {
300 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
302 /* General Registers */
303 {IXGBE_CTRL, "CTRL"},
304 {IXGBE_STATUS, "STATUS"},
305 {IXGBE_CTRL_EXT, "CTRL_EXT"},
307 /* Interrupt Registers */
308 {IXGBE_EICR, "EICR"},
311 {IXGBE_SRRCTL(0), "SRRCTL"},
312 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
313 {IXGBE_RDLEN(0), "RDLEN"},
314 {IXGBE_RDH(0), "RDH"},
315 {IXGBE_RDT(0), "RDT"},
316 {IXGBE_RXDCTL(0), "RXDCTL"},
317 {IXGBE_RDBAL(0), "RDBAL"},
318 {IXGBE_RDBAH(0), "RDBAH"},
321 {IXGBE_TDBAL(0), "TDBAL"},
322 {IXGBE_TDBAH(0), "TDBAH"},
323 {IXGBE_TDLEN(0), "TDLEN"},
324 {IXGBE_TDH(0), "TDH"},
325 {IXGBE_TDT(0), "TDT"},
326 {IXGBE_TXDCTL(0), "TXDCTL"},
328 /* List Terminator */
334 * ixgbe_regdump - register printout routine
336 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
342 switch (reginfo->ofs) {
343 case IXGBE_SRRCTL(0):
344 for (i = 0; i < 64; i++)
345 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
347 case IXGBE_DCA_RXCTRL(0):
348 for (i = 0; i < 64; i++)
349 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
352 for (i = 0; i < 64; i++)
353 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
356 for (i = 0; i < 64; i++)
357 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
360 for (i = 0; i < 64; i++)
361 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
363 case IXGBE_RXDCTL(0):
364 for (i = 0; i < 64; i++)
365 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
368 for (i = 0; i < 64; i++)
369 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
372 for (i = 0; i < 64; i++)
373 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
376 for (i = 0; i < 64; i++)
377 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
380 for (i = 0; i < 64; i++)
381 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
384 for (i = 0; i < 64; i++)
385 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
388 for (i = 0; i < 64; i++)
389 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
392 for (i = 0; i < 64; i++)
393 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
395 case IXGBE_TXDCTL(0):
396 for (i = 0; i < 64; i++)
397 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
400 pr_info("%-15s %08x\n", reginfo->name,
401 IXGBE_READ_REG(hw, reginfo->ofs));
405 for (i = 0; i < 8; i++) {
406 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
407 pr_err("%-15s", rname);
408 for (j = 0; j < 8; j++)
409 pr_cont(" %08x", regs[i*8+j]);
416 * ixgbe_dump - Print registers, tx-rings and rx-rings
418 static void ixgbe_dump(struct ixgbe_adapter *adapter)
420 struct net_device *netdev = adapter->netdev;
421 struct ixgbe_hw *hw = &adapter->hw;
422 struct ixgbe_reg_info *reginfo;
424 struct ixgbe_ring *tx_ring;
425 struct ixgbe_tx_buffer *tx_buffer;
426 union ixgbe_adv_tx_desc *tx_desc;
427 struct my_u0 { u64 a; u64 b; } *u0;
428 struct ixgbe_ring *rx_ring;
429 union ixgbe_adv_rx_desc *rx_desc;
430 struct ixgbe_rx_buffer *rx_buffer_info;
434 if (!netif_msg_hw(adapter))
437 /* Print netdevice Info */
439 dev_info(&adapter->pdev->dev, "Net device Info\n");
440 pr_info("Device Name state "
441 "trans_start last_rx\n");
442 pr_info("%-15s %016lX %016lX %016lX\n",
449 /* Print Registers */
450 dev_info(&adapter->pdev->dev, "Register Dump\n");
451 pr_info(" Register Name Value\n");
452 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
453 reginfo->name; reginfo++) {
454 ixgbe_regdump(hw, reginfo);
457 /* Print TX Ring Summary */
458 if (!netdev || !netif_running(netdev))
461 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
462 pr_info(" %s %s %s %s\n",
463 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
464 "leng", "ntw", "timestamp");
465 for (n = 0; n < adapter->num_tx_queues; n++) {
466 tx_ring = adapter->tx_ring[n];
467 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
468 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
469 n, tx_ring->next_to_use, tx_ring->next_to_clean,
470 (u64)dma_unmap_addr(tx_buffer, dma),
471 dma_unmap_len(tx_buffer, len),
472 tx_buffer->next_to_watch,
473 (u64)tx_buffer->time_stamp);
477 if (!netif_msg_tx_done(adapter))
478 goto rx_ring_summary;
480 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
482 /* Transmit Descriptor Formats
484 * 82598 Advanced Transmit Descriptor
485 * +--------------------------------------------------------------+
486 * 0 | Buffer Address [63:0] |
487 * +--------------------------------------------------------------+
488 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
489 * +--------------------------------------------------------------+
490 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
492 * 82598 Advanced Transmit Descriptor (Write-Back Format)
493 * +--------------------------------------------------------------+
495 * +--------------------------------------------------------------+
496 * 8 | RSV | STA | NXTSEQ |
497 * +--------------------------------------------------------------+
500 * 82599+ Advanced Transmit Descriptor
501 * +--------------------------------------------------------------+
502 * 0 | Buffer Address [63:0] |
503 * +--------------------------------------------------------------+
504 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
505 * +--------------------------------------------------------------+
506 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
508 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
509 * +--------------------------------------------------------------+
511 * +--------------------------------------------------------------+
512 * 8 | RSV | STA | RSV |
513 * +--------------------------------------------------------------+
517 for (n = 0; n < adapter->num_tx_queues; n++) {
518 tx_ring = adapter->tx_ring[n];
519 pr_info("------------------------------------\n");
520 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
521 pr_info("------------------------------------\n");
522 pr_info("%s%s %s %s %s %s\n",
523 "T [desc] [address 63:0 ] ",
524 "[PlPOIdStDDt Ln] [bi->dma ] ",
525 "leng", "ntw", "timestamp", "bi->skb");
527 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
528 tx_desc = IXGBE_TX_DESC(tx_ring, i);
529 tx_buffer = &tx_ring->tx_buffer_info[i];
530 u0 = (struct my_u0 *)tx_desc;
531 if (dma_unmap_len(tx_buffer, len) > 0) {
532 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p",
536 (u64)dma_unmap_addr(tx_buffer, dma),
537 dma_unmap_len(tx_buffer, len),
538 tx_buffer->next_to_watch,
539 (u64)tx_buffer->time_stamp,
541 if (i == tx_ring->next_to_use &&
542 i == tx_ring->next_to_clean)
544 else if (i == tx_ring->next_to_use)
546 else if (i == tx_ring->next_to_clean)
551 if (netif_msg_pktdata(adapter) &&
553 print_hex_dump(KERN_INFO, "",
554 DUMP_PREFIX_ADDRESS, 16, 1,
555 tx_buffer->skb->data,
556 dma_unmap_len(tx_buffer, len),
562 /* Print RX Rings Summary */
564 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
565 pr_info("Queue [NTU] [NTC]\n");
566 for (n = 0; n < adapter->num_rx_queues; n++) {
567 rx_ring = adapter->rx_ring[n];
568 pr_info("%5d %5X %5X\n",
569 n, rx_ring->next_to_use, rx_ring->next_to_clean);
573 if (!netif_msg_rx_status(adapter))
576 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
578 /* Receive Descriptor Formats
580 * 82598 Advanced Receive Descriptor (Read) Format
582 * +-----------------------------------------------------+
583 * 0 | Packet Buffer Address [63:1] |A0/NSE|
584 * +----------------------------------------------+------+
585 * 8 | Header Buffer Address [63:1] | DD |
586 * +-----------------------------------------------------+
589 * 82598 Advanced Receive Descriptor (Write-Back) Format
591 * 63 48 47 32 31 30 21 20 16 15 4 3 0
592 * +------------------------------------------------------+
593 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
594 * | Packet | IP | | | | Type | Type |
595 * | Checksum | Ident | | | | | |
596 * +------------------------------------------------------+
597 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
598 * +------------------------------------------------------+
599 * 63 48 47 32 31 20 19 0
601 * 82599+ Advanced Receive Descriptor (Read) Format
603 * +-----------------------------------------------------+
604 * 0 | Packet Buffer Address [63:1] |A0/NSE|
605 * +----------------------------------------------+------+
606 * 8 | Header Buffer Address [63:1] | DD |
607 * +-----------------------------------------------------+
610 * 82599+ Advanced Receive Descriptor (Write-Back) Format
612 * 63 48 47 32 31 30 21 20 17 16 4 3 0
613 * +------------------------------------------------------+
614 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
615 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
616 * |/ Flow Dir Flt ID | | | | | |
617 * +------------------------------------------------------+
618 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
619 * +------------------------------------------------------+
620 * 63 48 47 32 31 20 19 0
623 for (n = 0; n < adapter->num_rx_queues; n++) {
624 rx_ring = adapter->rx_ring[n];
625 pr_info("------------------------------------\n");
626 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
627 pr_info("------------------------------------\n");
629 "R [desc] [ PktBuf A0] ",
630 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
631 "<-- Adv Rx Read format\n");
633 "RWB[desc] [PcsmIpSHl PtRs] ",
634 "[vl er S cks ln] ---------------- [bi->skb ] ",
635 "<-- Adv Rx Write-Back format\n");
637 for (i = 0; i < rx_ring->count; i++) {
638 rx_buffer_info = &rx_ring->rx_buffer_info[i];
639 rx_desc = IXGBE_RX_DESC(rx_ring, i);
640 u0 = (struct my_u0 *)rx_desc;
641 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
642 if (staterr & IXGBE_RXD_STAT_DD) {
643 /* Descriptor Done */
644 pr_info("RWB[0x%03X] %016llX "
645 "%016llX ---------------- %p", i,
648 rx_buffer_info->skb);
650 pr_info("R [0x%03X] %016llX "
651 "%016llX %016llX %p", i,
654 (u64)rx_buffer_info->dma,
655 rx_buffer_info->skb);
657 if (netif_msg_pktdata(adapter) &&
658 rx_buffer_info->dma) {
659 print_hex_dump(KERN_INFO, "",
660 DUMP_PREFIX_ADDRESS, 16, 1,
661 page_address(rx_buffer_info->page) +
662 rx_buffer_info->page_offset,
663 ixgbe_rx_bufsz(rx_ring), true);
667 if (i == rx_ring->next_to_use)
669 else if (i == rx_ring->next_to_clean)
681 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
685 /* Let firmware take over control of h/w */
686 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
687 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
688 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
691 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
695 /* Let firmware know the driver has taken over */
696 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
697 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
698 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
702 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
703 * @adapter: pointer to adapter struct
704 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
705 * @queue: queue to map the corresponding interrupt to
706 * @msix_vector: the vector to map to the corresponding queue
709 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
710 u8 queue, u8 msix_vector)
713 struct ixgbe_hw *hw = &adapter->hw;
714 switch (hw->mac.type) {
715 case ixgbe_mac_82598EB:
716 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
719 index = (((direction * 64) + queue) >> 2) & 0x1F;
720 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
721 ivar &= ~(0xFF << (8 * (queue & 0x3)));
722 ivar |= (msix_vector << (8 * (queue & 0x3)));
723 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
725 case ixgbe_mac_82599EB:
727 if (direction == -1) {
729 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
730 index = ((queue & 1) * 8);
731 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
732 ivar &= ~(0xFF << index);
733 ivar |= (msix_vector << index);
734 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
737 /* tx or rx causes */
738 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
739 index = ((16 * (queue & 1)) + (8 * direction));
740 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
741 ivar &= ~(0xFF << index);
742 ivar |= (msix_vector << index);
743 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
751 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
756 switch (adapter->hw.mac.type) {
757 case ixgbe_mac_82598EB:
758 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
759 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
761 case ixgbe_mac_82599EB:
763 mask = (qmask & 0xFFFFFFFF);
764 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
765 mask = (qmask >> 32);
766 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
773 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
774 struct ixgbe_tx_buffer *tx_buffer)
776 if (tx_buffer->skb) {
777 dev_kfree_skb_any(tx_buffer->skb);
778 if (dma_unmap_len(tx_buffer, len))
779 dma_unmap_single(ring->dev,
780 dma_unmap_addr(tx_buffer, dma),
781 dma_unmap_len(tx_buffer, len),
783 } else if (dma_unmap_len(tx_buffer, len)) {
784 dma_unmap_page(ring->dev,
785 dma_unmap_addr(tx_buffer, dma),
786 dma_unmap_len(tx_buffer, len),
789 tx_buffer->next_to_watch = NULL;
790 tx_buffer->skb = NULL;
791 dma_unmap_len_set(tx_buffer, len, 0);
792 /* tx_buffer must be completely set up in the transmit path */
795 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
797 struct ixgbe_hw *hw = &adapter->hw;
798 struct ixgbe_hw_stats *hwstats = &adapter->stats;
802 if ((hw->fc.current_mode != ixgbe_fc_full) &&
803 (hw->fc.current_mode != ixgbe_fc_rx_pause))
806 switch (hw->mac.type) {
807 case ixgbe_mac_82598EB:
808 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
811 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
813 hwstats->lxoffrxc += data;
815 /* refill credits (no tx hang) if we received xoff */
819 for (i = 0; i < adapter->num_tx_queues; i++)
820 clear_bit(__IXGBE_HANG_CHECK_ARMED,
821 &adapter->tx_ring[i]->state);
824 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
826 struct ixgbe_hw *hw = &adapter->hw;
827 struct ixgbe_hw_stats *hwstats = &adapter->stats;
831 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
833 if (adapter->ixgbe_ieee_pfc)
834 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
836 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
837 ixgbe_update_xoff_rx_lfc(adapter);
841 /* update stats for each tc, only valid with PFC enabled */
842 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
845 switch (hw->mac.type) {
846 case ixgbe_mac_82598EB:
847 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
850 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
852 hwstats->pxoffrxc[i] += pxoffrxc;
853 /* Get the TC for given UP */
854 tc = netdev_get_prio_tc_map(adapter->netdev, i);
855 xoff[tc] += pxoffrxc;
858 /* disarm tx queues that have received xoff frames */
859 for (i = 0; i < adapter->num_tx_queues; i++) {
860 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
862 tc = tx_ring->dcb_tc;
864 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
868 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
870 return ring->stats.packets;
873 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
875 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
876 struct ixgbe_hw *hw = &adapter->hw;
878 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
879 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
882 return (head < tail) ?
883 tail - head : (tail + ring->count - head);
888 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
890 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
891 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
892 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
895 clear_check_for_tx_hang(tx_ring);
898 * Check for a hung queue, but be thorough. This verifies
899 * that a transmit has been completed since the previous
900 * check AND there is at least one packet pending. The
901 * ARMED bit is set to indicate a potential hang. The
902 * bit is cleared if a pause frame is received to remove
903 * false hang detection due to PFC or 802.3x frames. By
904 * requiring this to fail twice we avoid races with
905 * pfc clearing the ARMED bit and conditions where we
906 * run the check_tx_hang logic with a transmit completion
907 * pending but without time to complete it yet.
909 if ((tx_done_old == tx_done) && tx_pending) {
910 /* make sure it is true for two checks in a row */
911 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
914 /* update completed stats and continue */
915 tx_ring->tx_stats.tx_done_old = tx_done;
916 /* reset the countdown */
917 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
924 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
925 * @adapter: driver private struct
927 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
930 /* Do the reset outside of interrupt context */
931 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
932 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
933 e_warn(drv, "initiating reset due to tx timeout\n");
934 ixgbe_service_event_schedule(adapter);
939 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
940 * @q_vector: structure containing interrupt and ring information
941 * @tx_ring: tx ring to clean
943 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
944 struct ixgbe_ring *tx_ring)
946 struct ixgbe_adapter *adapter = q_vector->adapter;
947 struct ixgbe_tx_buffer *tx_buffer;
948 union ixgbe_adv_tx_desc *tx_desc;
949 unsigned int total_bytes = 0, total_packets = 0;
950 unsigned int budget = q_vector->tx.work_limit;
951 unsigned int i = tx_ring->next_to_clean;
953 if (test_bit(__IXGBE_DOWN, &adapter->state))
956 tx_buffer = &tx_ring->tx_buffer_info[i];
957 tx_desc = IXGBE_TX_DESC(tx_ring, i);
961 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
963 /* if next_to_watch is not set then there is no work pending */
967 /* prevent any other reads prior to eop_desc */
968 read_barrier_depends();
970 /* if DD is not set pending work has not been completed */
971 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
974 /* clear next_to_watch to prevent false hangs */
975 tx_buffer->next_to_watch = NULL;
977 /* update the statistics for this packet */
978 total_bytes += tx_buffer->bytecount;
979 total_packets += tx_buffer->gso_segs;
982 dev_kfree_skb_any(tx_buffer->skb);
984 /* unmap skb header data */
985 dma_unmap_single(tx_ring->dev,
986 dma_unmap_addr(tx_buffer, dma),
987 dma_unmap_len(tx_buffer, len),
990 /* clear tx_buffer data */
991 tx_buffer->skb = NULL;
992 dma_unmap_len_set(tx_buffer, len, 0);
994 /* unmap remaining buffers */
995 while (tx_desc != eop_desc) {
1000 i -= tx_ring->count;
1001 tx_buffer = tx_ring->tx_buffer_info;
1002 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1005 /* unmap any remaining paged data */
1006 if (dma_unmap_len(tx_buffer, len)) {
1007 dma_unmap_page(tx_ring->dev,
1008 dma_unmap_addr(tx_buffer, dma),
1009 dma_unmap_len(tx_buffer, len),
1011 dma_unmap_len_set(tx_buffer, len, 0);
1015 /* move us one more past the eop_desc for start of next pkt */
1020 i -= tx_ring->count;
1021 tx_buffer = tx_ring->tx_buffer_info;
1022 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1025 /* issue prefetch for next Tx descriptor */
1028 /* update budget accounting */
1030 } while (likely(budget));
1032 i += tx_ring->count;
1033 tx_ring->next_to_clean = i;
1034 u64_stats_update_begin(&tx_ring->syncp);
1035 tx_ring->stats.bytes += total_bytes;
1036 tx_ring->stats.packets += total_packets;
1037 u64_stats_update_end(&tx_ring->syncp);
1038 q_vector->tx.total_bytes += total_bytes;
1039 q_vector->tx.total_packets += total_packets;
1041 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1042 /* schedule immediate reset if we believe we hung */
1043 struct ixgbe_hw *hw = &adapter->hw;
1044 e_err(drv, "Detected Tx Unit Hang\n"
1046 " TDH, TDT <%x>, <%x>\n"
1047 " next_to_use <%x>\n"
1048 " next_to_clean <%x>\n"
1049 "tx_buffer_info[next_to_clean]\n"
1050 " time_stamp <%lx>\n"
1052 tx_ring->queue_index,
1053 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1054 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1055 tx_ring->next_to_use, i,
1056 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1058 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1061 "tx hang %d detected on queue %d, resetting adapter\n",
1062 adapter->tx_timeout_count + 1, tx_ring->queue_index);
1064 /* schedule immediate reset if we believe we hung */
1065 ixgbe_tx_timeout_reset(adapter);
1067 /* the adapter is about to reset, no point in enabling stuff */
1071 netdev_tx_completed_queue(txring_txq(tx_ring),
1072 total_packets, total_bytes);
1074 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1075 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1076 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1077 /* Make sure that anybody stopping the queue after this
1078 * sees the new next_to_clean.
1081 if (__netif_subqueue_stopped(tx_ring->netdev,
1082 tx_ring->queue_index)
1083 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1084 netif_wake_subqueue(tx_ring->netdev,
1085 tx_ring->queue_index);
1086 ++tx_ring->tx_stats.restart_queue;
1093 #ifdef CONFIG_IXGBE_DCA
1094 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1095 struct ixgbe_ring *tx_ring,
1098 struct ixgbe_hw *hw = &adapter->hw;
1099 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1102 switch (hw->mac.type) {
1103 case ixgbe_mac_82598EB:
1104 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1106 case ixgbe_mac_82599EB:
1107 case ixgbe_mac_X540:
1108 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1109 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1112 /* for unknown hardware do not write register */
1117 * We can enable relaxed ordering for reads, but not writes when
1118 * DCA is enabled. This is due to a known issue in some chipsets
1119 * which will cause the DCA tag to be cleared.
1121 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1122 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1123 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1125 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1128 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1129 struct ixgbe_ring *rx_ring,
1132 struct ixgbe_hw *hw = &adapter->hw;
1133 u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1134 u8 reg_idx = rx_ring->reg_idx;
1137 switch (hw->mac.type) {
1138 case ixgbe_mac_82599EB:
1139 case ixgbe_mac_X540:
1140 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1147 * We can enable relaxed ordering for reads, but not writes when
1148 * DCA is enabled. This is due to a known issue in some chipsets
1149 * which will cause the DCA tag to be cleared.
1151 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1152 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1154 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1157 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1159 struct ixgbe_adapter *adapter = q_vector->adapter;
1160 struct ixgbe_ring *ring;
1161 int cpu = get_cpu();
1163 if (q_vector->cpu == cpu)
1166 ixgbe_for_each_ring(ring, q_vector->tx)
1167 ixgbe_update_tx_dca(adapter, ring, cpu);
1169 ixgbe_for_each_ring(ring, q_vector->rx)
1170 ixgbe_update_rx_dca(adapter, ring, cpu);
1172 q_vector->cpu = cpu;
1177 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1181 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1184 /* always use CB2 mode, difference is masked in the CB driver */
1185 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1187 for (i = 0; i < adapter->num_q_vectors; i++) {
1188 adapter->q_vector[i]->cpu = -1;
1189 ixgbe_update_dca(adapter->q_vector[i]);
1193 static int __ixgbe_notify_dca(struct device *dev, void *data)
1195 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1196 unsigned long event = *(unsigned long *)data;
1198 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1202 case DCA_PROVIDER_ADD:
1203 /* if we're already enabled, don't do it again */
1204 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1206 if (dca_add_requester(dev) == 0) {
1207 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1208 ixgbe_setup_dca(adapter);
1211 /* Fall Through since DCA is disabled. */
1212 case DCA_PROVIDER_REMOVE:
1213 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1214 dca_remove_requester(dev);
1215 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1216 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1224 #endif /* CONFIG_IXGBE_DCA */
1225 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1226 union ixgbe_adv_rx_desc *rx_desc,
1227 struct sk_buff *skb)
1229 if (ring->netdev->features & NETIF_F_RXHASH)
1230 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1235 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1236 * @ring: structure containing ring specific data
1237 * @rx_desc: advanced rx descriptor
1239 * Returns : true if it is FCoE pkt
1241 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1242 union ixgbe_adv_rx_desc *rx_desc)
1244 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1246 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1247 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1248 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1249 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1252 #endif /* IXGBE_FCOE */
1254 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1255 * @ring: structure containing ring specific data
1256 * @rx_desc: current Rx descriptor being processed
1257 * @skb: skb currently being received and modified
1259 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1260 union ixgbe_adv_rx_desc *rx_desc,
1261 struct sk_buff *skb)
1263 skb_checksum_none_assert(skb);
1265 /* Rx csum disabled */
1266 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1269 /* if IP and error */
1270 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1271 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1272 ring->rx_stats.csum_err++;
1276 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1279 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1280 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1283 * 82599 errata, UDP frames with a 0 checksum can be marked as
1286 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1287 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1290 ring->rx_stats.csum_err++;
1294 /* It must be a TCP or UDP packet with a valid checksum */
1295 skb->ip_summed = CHECKSUM_UNNECESSARY;
1298 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1300 rx_ring->next_to_use = val;
1302 /* update next to alloc since we have filled the ring */
1303 rx_ring->next_to_alloc = val;
1305 * Force memory writes to complete before letting h/w
1306 * know there are new descriptors to fetch. (Only
1307 * applicable for weak-ordered memory model archs,
1311 writel(val, rx_ring->tail);
1314 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1315 struct ixgbe_rx_buffer *bi)
1317 struct page *page = bi->page;
1318 dma_addr_t dma = bi->dma;
1320 /* since we are recycling buffers we should seldom need to alloc */
1324 /* alloc new page for storage */
1325 if (likely(!page)) {
1326 page = __skb_alloc_pages(GFP_ATOMIC | __GFP_COLD | __GFP_COMP,
1327 bi->skb, ixgbe_rx_pg_order(rx_ring));
1328 if (unlikely(!page)) {
1329 rx_ring->rx_stats.alloc_rx_page_failed++;
1335 /* map page for use */
1336 dma = dma_map_page(rx_ring->dev, page, 0,
1337 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1340 * if mapping failed free memory back to system since
1341 * there isn't much point in holding memory we can't use
1343 if (dma_mapping_error(rx_ring->dev, dma)) {
1344 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1347 rx_ring->rx_stats.alloc_rx_page_failed++;
1352 bi->page_offset = 0;
1358 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1359 * @rx_ring: ring to place buffers on
1360 * @cleaned_count: number of buffers to replace
1362 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1364 union ixgbe_adv_rx_desc *rx_desc;
1365 struct ixgbe_rx_buffer *bi;
1366 u16 i = rx_ring->next_to_use;
1372 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1373 bi = &rx_ring->rx_buffer_info[i];
1374 i -= rx_ring->count;
1377 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1381 * Refresh the desc even if buffer_addrs didn't change
1382 * because each write-back erases this info.
1384 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1390 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1391 bi = rx_ring->rx_buffer_info;
1392 i -= rx_ring->count;
1395 /* clear the hdr_addr for the next_to_use descriptor */
1396 rx_desc->read.hdr_addr = 0;
1399 } while (cleaned_count);
1401 i += rx_ring->count;
1403 if (rx_ring->next_to_use != i)
1404 ixgbe_release_rx_desc(rx_ring, i);
1408 * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1409 * @data: pointer to the start of the headers
1410 * @max_len: total length of section to find headers in
1412 * This function is meant to determine the length of headers that will
1413 * be recognized by hardware for LRO, GRO, and RSC offloads. The main
1414 * motivation of doing this is to only perform one pull for IPv4 TCP
1415 * packets so that we can do basic things like calculating the gso_size
1416 * based on the average data per packet.
1418 static unsigned int ixgbe_get_headlen(unsigned char *data,
1419 unsigned int max_len)
1422 unsigned char *network;
1425 struct vlan_hdr *vlan;
1428 struct ipv6hdr *ipv6;
1431 u8 nexthdr = 0; /* default to not TCP */
1434 /* this should never happen, but better safe than sorry */
1435 if (max_len < ETH_HLEN)
1438 /* initialize network frame pointer */
1441 /* set first protocol and move network header forward */
1442 protocol = hdr.eth->h_proto;
1443 hdr.network += ETH_HLEN;
1445 /* handle any vlan tag if present */
1446 if (protocol == __constant_htons(ETH_P_8021Q)) {
1447 if ((hdr.network - data) > (max_len - VLAN_HLEN))
1450 protocol = hdr.vlan->h_vlan_encapsulated_proto;
1451 hdr.network += VLAN_HLEN;
1454 /* handle L3 protocols */
1455 if (protocol == __constant_htons(ETH_P_IP)) {
1456 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1459 /* access ihl as a u8 to avoid unaligned access on ia64 */
1460 hlen = (hdr.network[0] & 0x0F) << 2;
1462 /* verify hlen meets minimum size requirements */
1463 if (hlen < sizeof(struct iphdr))
1464 return hdr.network - data;
1466 /* record next protocol if header is present */
1467 if (!(hdr.ipv4->frag_off & htons(IP_OFFSET)))
1468 nexthdr = hdr.ipv4->protocol;
1469 } else if (protocol == __constant_htons(ETH_P_IPV6)) {
1470 if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
1473 /* record next protocol */
1474 nexthdr = hdr.ipv6->nexthdr;
1475 hlen = sizeof(struct ipv6hdr);
1477 } else if (protocol == __constant_htons(ETH_P_FCOE)) {
1478 if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1480 hlen = FCOE_HEADER_LEN;
1483 return hdr.network - data;
1486 /* relocate pointer to start of L4 header */
1487 hdr.network += hlen;
1489 /* finally sort out TCP/UDP */
1490 if (nexthdr == IPPROTO_TCP) {
1491 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1494 /* access doff as a u8 to avoid unaligned access on ia64 */
1495 hlen = (hdr.network[12] & 0xF0) >> 2;
1497 /* verify hlen meets minimum size requirements */
1498 if (hlen < sizeof(struct tcphdr))
1499 return hdr.network - data;
1501 hdr.network += hlen;
1502 } else if (nexthdr == IPPROTO_UDP) {
1503 if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
1506 hdr.network += sizeof(struct udphdr);
1510 * If everything has gone correctly hdr.network should be the
1511 * data section of the packet and will be the end of the header.
1512 * If not then it probably represents the end of the last recognized
1515 if ((hdr.network - data) < max_len)
1516 return hdr.network - data;
1521 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1522 struct sk_buff *skb)
1524 u16 hdr_len = skb_headlen(skb);
1526 /* set gso_size to avoid messing up TCP MSS */
1527 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1528 IXGBE_CB(skb)->append_cnt);
1529 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1532 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1533 struct sk_buff *skb)
1535 /* if append_cnt is 0 then frame is not RSC */
1536 if (!IXGBE_CB(skb)->append_cnt)
1539 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1540 rx_ring->rx_stats.rsc_flush++;
1542 ixgbe_set_rsc_gso_size(rx_ring, skb);
1544 /* gso_size is computed using append_cnt so always clear it last */
1545 IXGBE_CB(skb)->append_cnt = 0;
1549 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1550 * @rx_ring: rx descriptor ring packet is being transacted on
1551 * @rx_desc: pointer to the EOP Rx descriptor
1552 * @skb: pointer to current skb being populated
1554 * This function checks the ring, descriptor, and packet information in
1555 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1556 * other fields within the skb.
1558 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1559 union ixgbe_adv_rx_desc *rx_desc,
1560 struct sk_buff *skb)
1562 struct net_device *dev = rx_ring->netdev;
1564 ixgbe_update_rsc_stats(rx_ring, skb);
1566 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1568 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1570 ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1572 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1573 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1574 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1575 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1578 skb_record_rx_queue(skb, rx_ring->queue_index);
1580 skb->protocol = eth_type_trans(skb, dev);
1583 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1584 struct sk_buff *skb)
1586 struct ixgbe_adapter *adapter = q_vector->adapter;
1588 if (ixgbe_qv_ll_polling(q_vector))
1589 netif_receive_skb(skb);
1590 else if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1591 napi_gro_receive(&q_vector->napi, skb);
1597 * ixgbe_is_non_eop - process handling of non-EOP buffers
1598 * @rx_ring: Rx ring being processed
1599 * @rx_desc: Rx descriptor for current buffer
1600 * @skb: Current socket buffer containing buffer in progress
1602 * This function updates next to clean. If the buffer is an EOP buffer
1603 * this function exits returning false, otherwise it will place the
1604 * sk_buff in the next buffer to be chained and return true indicating
1605 * that this is in fact a non-EOP buffer.
1607 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1608 union ixgbe_adv_rx_desc *rx_desc,
1609 struct sk_buff *skb)
1611 u32 ntc = rx_ring->next_to_clean + 1;
1613 /* fetch, update, and store next to clean */
1614 ntc = (ntc < rx_ring->count) ? ntc : 0;
1615 rx_ring->next_to_clean = ntc;
1617 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1619 /* update RSC append count if present */
1620 if (ring_is_rsc_enabled(rx_ring)) {
1621 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1622 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1624 if (unlikely(rsc_enabled)) {
1625 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1627 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1628 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1630 /* update ntc based on RSC value */
1631 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1632 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1633 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1637 /* if we are the last buffer then there is nothing else to do */
1638 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1641 /* place skb in next buffer to be received */
1642 rx_ring->rx_buffer_info[ntc].skb = skb;
1643 rx_ring->rx_stats.non_eop_descs++;
1649 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1650 * @rx_ring: rx descriptor ring packet is being transacted on
1651 * @skb: pointer to current skb being adjusted
1653 * This function is an ixgbe specific version of __pskb_pull_tail. The
1654 * main difference between this version and the original function is that
1655 * this function can make several assumptions about the state of things
1656 * that allow for significant optimizations versus the standard function.
1657 * As a result we can do things like drop a frag and maintain an accurate
1658 * truesize for the skb.
1660 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1661 struct sk_buff *skb)
1663 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1665 unsigned int pull_len;
1668 * it is valid to use page_address instead of kmap since we are
1669 * working with pages allocated out of the lomem pool per
1670 * alloc_page(GFP_ATOMIC)
1672 va = skb_frag_address(frag);
1675 * we need the header to contain the greater of either ETH_HLEN or
1676 * 60 bytes if the skb->len is less than 60 for skb_pad.
1678 pull_len = ixgbe_get_headlen(va, IXGBE_RX_HDR_SIZE);
1680 /* align pull length to size of long to optimize memcpy performance */
1681 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1683 /* update all of the pointers */
1684 skb_frag_size_sub(frag, pull_len);
1685 frag->page_offset += pull_len;
1686 skb->data_len -= pull_len;
1687 skb->tail += pull_len;
1691 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1692 * @rx_ring: rx descriptor ring packet is being transacted on
1693 * @skb: pointer to current skb being updated
1695 * This function provides a basic DMA sync up for the first fragment of an
1696 * skb. The reason for doing this is that the first fragment cannot be
1697 * unmapped until we have reached the end of packet descriptor for a buffer
1700 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1701 struct sk_buff *skb)
1703 /* if the page was released unmap it, else just sync our portion */
1704 if (unlikely(IXGBE_CB(skb)->page_released)) {
1705 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1706 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1707 IXGBE_CB(skb)->page_released = false;
1709 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1711 dma_sync_single_range_for_cpu(rx_ring->dev,
1714 ixgbe_rx_bufsz(rx_ring),
1717 IXGBE_CB(skb)->dma = 0;
1721 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1722 * @rx_ring: rx descriptor ring packet is being transacted on
1723 * @rx_desc: pointer to the EOP Rx descriptor
1724 * @skb: pointer to current skb being fixed
1726 * Check for corrupted packet headers caused by senders on the local L2
1727 * embedded NIC switch not setting up their Tx Descriptors right. These
1728 * should be very rare.
1730 * Also address the case where we are pulling data in on pages only
1731 * and as such no data is present in the skb header.
1733 * In addition if skb is not at least 60 bytes we need to pad it so that
1734 * it is large enough to qualify as a valid Ethernet frame.
1736 * Returns true if an error was encountered and skb was freed.
1738 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1739 union ixgbe_adv_rx_desc *rx_desc,
1740 struct sk_buff *skb)
1742 struct net_device *netdev = rx_ring->netdev;
1744 /* verify that the packet does not have any known errors */
1745 if (unlikely(ixgbe_test_staterr(rx_desc,
1746 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1747 !(netdev->features & NETIF_F_RXALL))) {
1748 dev_kfree_skb_any(skb);
1752 /* place header in linear portion of buffer */
1753 if (skb_is_nonlinear(skb))
1754 ixgbe_pull_tail(rx_ring, skb);
1757 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1758 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1762 /* if skb_pad returns an error the skb was freed */
1763 if (unlikely(skb->len < 60)) {
1764 int pad_len = 60 - skb->len;
1766 if (skb_pad(skb, pad_len))
1768 __skb_put(skb, pad_len);
1775 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1776 * @rx_ring: rx descriptor ring to store buffers on
1777 * @old_buff: donor buffer to have page reused
1779 * Synchronizes page for reuse by the adapter
1781 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1782 struct ixgbe_rx_buffer *old_buff)
1784 struct ixgbe_rx_buffer *new_buff;
1785 u16 nta = rx_ring->next_to_alloc;
1787 new_buff = &rx_ring->rx_buffer_info[nta];
1789 /* update, and store next to alloc */
1791 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1793 /* transfer page from old buffer to new buffer */
1794 new_buff->page = old_buff->page;
1795 new_buff->dma = old_buff->dma;
1796 new_buff->page_offset = old_buff->page_offset;
1798 /* sync the buffer for use by the device */
1799 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1800 new_buff->page_offset,
1801 ixgbe_rx_bufsz(rx_ring),
1806 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1807 * @rx_ring: rx descriptor ring to transact packets on
1808 * @rx_buffer: buffer containing page to add
1809 * @rx_desc: descriptor containing length of buffer written by hardware
1810 * @skb: sk_buff to place the data into
1812 * This function will add the data contained in rx_buffer->page to the skb.
1813 * This is done either through a direct copy if the data in the buffer is
1814 * less than the skb header size, otherwise it will just attach the page as
1815 * a frag to the skb.
1817 * The function will then update the page offset if necessary and return
1818 * true if the buffer can be reused by the adapter.
1820 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1821 struct ixgbe_rx_buffer *rx_buffer,
1822 union ixgbe_adv_rx_desc *rx_desc,
1823 struct sk_buff *skb)
1825 struct page *page = rx_buffer->page;
1826 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1827 #if (PAGE_SIZE < 8192)
1828 unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1830 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1831 unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1832 ixgbe_rx_bufsz(rx_ring);
1835 if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1836 unsigned char *va = page_address(page) + rx_buffer->page_offset;
1838 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1840 /* we can reuse buffer as-is, just make sure it is local */
1841 if (likely(page_to_nid(page) == numa_node_id()))
1844 /* this page cannot be reused so discard it */
1849 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1850 rx_buffer->page_offset, size, truesize);
1852 /* avoid re-using remote pages */
1853 if (unlikely(page_to_nid(page) != numa_node_id()))
1856 #if (PAGE_SIZE < 8192)
1857 /* if we are only owner of page we can reuse it */
1858 if (unlikely(page_count(page) != 1))
1861 /* flip page offset to other buffer */
1862 rx_buffer->page_offset ^= truesize;
1865 * since we are the only owner of the page and we need to
1866 * increment it, just set the value to 2 in order to avoid
1867 * an unecessary locked operation
1869 atomic_set(&page->_count, 2);
1871 /* move offset up to the next cache line */
1872 rx_buffer->page_offset += truesize;
1874 if (rx_buffer->page_offset > last_offset)
1877 /* bump ref count on page before it is given to the stack */
1884 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1885 union ixgbe_adv_rx_desc *rx_desc)
1887 struct ixgbe_rx_buffer *rx_buffer;
1888 struct sk_buff *skb;
1891 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1892 page = rx_buffer->page;
1895 skb = rx_buffer->skb;
1898 void *page_addr = page_address(page) +
1899 rx_buffer->page_offset;
1901 /* prefetch first cache line of first page */
1902 prefetch(page_addr);
1903 #if L1_CACHE_BYTES < 128
1904 prefetch(page_addr + L1_CACHE_BYTES);
1907 /* allocate a skb to store the frags */
1908 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1910 if (unlikely(!skb)) {
1911 rx_ring->rx_stats.alloc_rx_buff_failed++;
1916 * we will be copying header into skb->data in
1917 * pskb_may_pull so it is in our interest to prefetch
1918 * it now to avoid a possible cache miss
1920 prefetchw(skb->data);
1923 * Delay unmapping of the first packet. It carries the
1924 * header information, HW may still access the header
1925 * after the writeback. Only unmap it when EOP is
1928 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1931 IXGBE_CB(skb)->dma = rx_buffer->dma;
1933 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
1934 ixgbe_dma_sync_frag(rx_ring, skb);
1937 /* we are reusing so sync this buffer for CPU use */
1938 dma_sync_single_range_for_cpu(rx_ring->dev,
1940 rx_buffer->page_offset,
1941 ixgbe_rx_bufsz(rx_ring),
1945 /* pull page into skb */
1946 if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
1947 /* hand second half of page back to the ring */
1948 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1949 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1950 /* the page has been released from the ring */
1951 IXGBE_CB(skb)->page_released = true;
1953 /* we are not reusing the buffer so unmap it */
1954 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1955 ixgbe_rx_pg_size(rx_ring),
1959 /* clear contents of buffer_info */
1960 rx_buffer->skb = NULL;
1962 rx_buffer->page = NULL;
1968 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1969 * @q_vector: structure containing interrupt and ring information
1970 * @rx_ring: rx descriptor ring to transact packets on
1971 * @budget: Total limit on number of packets to process
1973 * This function provides a "bounce buffer" approach to Rx interrupt
1974 * processing. The advantage to this is that on systems that have
1975 * expensive overhead for IOMMU access this provides a means of avoiding
1976 * it by maintaining the mapping of the page to the syste.
1978 * Returns amount of work completed
1980 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1981 struct ixgbe_ring *rx_ring,
1984 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1986 struct ixgbe_adapter *adapter = q_vector->adapter;
1988 unsigned int mss = 0;
1989 #endif /* IXGBE_FCOE */
1990 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1993 union ixgbe_adv_rx_desc *rx_desc;
1994 struct sk_buff *skb;
1996 /* return some buffers to hardware, one at a time is too slow */
1997 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1998 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2002 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2004 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
2008 * This memory barrier is needed to keep us from reading
2009 * any other fields out of the rx_desc until we know the
2010 * RXD_STAT_DD bit is set
2014 /* retrieve a buffer from the ring */
2015 skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
2017 /* exit if we failed to retrieve a buffer */
2023 /* place incomplete frames back on ring for completion */
2024 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2027 /* verify the packet layout is correct */
2028 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2031 /* probably a little skewed due to removing CRC */
2032 total_rx_bytes += skb->len;
2034 /* populate checksum, timestamp, VLAN, and protocol */
2035 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2038 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2039 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2040 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2041 /* include DDPed FCoE data */
2042 if (ddp_bytes > 0) {
2044 mss = rx_ring->netdev->mtu -
2045 sizeof(struct fcoe_hdr) -
2046 sizeof(struct fc_frame_header) -
2047 sizeof(struct fcoe_crc_eof);
2051 total_rx_bytes += ddp_bytes;
2052 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2056 dev_kfree_skb_any(skb);
2061 #endif /* IXGBE_FCOE */
2062 skb_mark_napi_id(skb, &q_vector->napi);
2063 ixgbe_rx_skb(q_vector, skb);
2065 /* update budget accounting */
2067 } while (likely(total_rx_packets < budget));
2069 u64_stats_update_begin(&rx_ring->syncp);
2070 rx_ring->stats.packets += total_rx_packets;
2071 rx_ring->stats.bytes += total_rx_bytes;
2072 u64_stats_update_end(&rx_ring->syncp);
2073 q_vector->rx.total_packets += total_rx_packets;
2074 q_vector->rx.total_bytes += total_rx_bytes;
2077 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2079 return total_rx_packets;
2082 #ifdef CONFIG_NET_RX_BUSY_POLL
2083 /* must be called with local_bh_disable()d */
2084 static int ixgbe_low_latency_recv(struct napi_struct *napi)
2086 struct ixgbe_q_vector *q_vector =
2087 container_of(napi, struct ixgbe_q_vector, napi);
2088 struct ixgbe_adapter *adapter = q_vector->adapter;
2089 struct ixgbe_ring *ring;
2092 if (test_bit(__IXGBE_DOWN, &adapter->state))
2093 return LL_FLUSH_FAILED;
2095 if (!ixgbe_qv_lock_poll(q_vector))
2096 return LL_FLUSH_BUSY;
2098 ixgbe_for_each_ring(ring, q_vector->rx) {
2099 found = ixgbe_clean_rx_irq(q_vector, ring, 4);
2100 #ifdef LL_EXTENDED_STATS
2102 ring->stats.cleaned += found;
2104 ring->stats.misses++;
2110 ixgbe_qv_unlock_poll(q_vector);
2114 #endif /* CONFIG_NET_RX_BUSY_POLL */
2117 * ixgbe_configure_msix - Configure MSI-X hardware
2118 * @adapter: board private structure
2120 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2123 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2125 struct ixgbe_q_vector *q_vector;
2129 /* Populate MSIX to EITR Select */
2130 if (adapter->num_vfs > 32) {
2131 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2132 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2136 * Populate the IVAR table and set the ITR values to the
2137 * corresponding register.
2139 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2140 struct ixgbe_ring *ring;
2141 q_vector = adapter->q_vector[v_idx];
2143 ixgbe_for_each_ring(ring, q_vector->rx)
2144 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2146 ixgbe_for_each_ring(ring, q_vector->tx)
2147 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2149 ixgbe_write_eitr(q_vector);
2152 switch (adapter->hw.mac.type) {
2153 case ixgbe_mac_82598EB:
2154 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2157 case ixgbe_mac_82599EB:
2158 case ixgbe_mac_X540:
2159 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2164 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2166 /* set up to autoclear timer, and the vectors */
2167 mask = IXGBE_EIMS_ENABLE_MASK;
2168 mask &= ~(IXGBE_EIMS_OTHER |
2169 IXGBE_EIMS_MAILBOX |
2172 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2175 enum latency_range {
2179 latency_invalid = 255
2183 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2184 * @q_vector: structure containing interrupt and ring information
2185 * @ring_container: structure containing ring performance data
2187 * Stores a new ITR value based on packets and byte
2188 * counts during the last interrupt. The advantage of per interrupt
2189 * computation is faster updates and more accurate ITR for the current
2190 * traffic pattern. Constants in this function were computed
2191 * based on theoretical maximum wire speed and thresholds were set based
2192 * on testing data as well as attempting to minimize response time
2193 * while increasing bulk throughput.
2194 * this functionality is controlled by the InterruptThrottleRate module
2195 * parameter (see ixgbe_param.c)
2197 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2198 struct ixgbe_ring_container *ring_container)
2200 int bytes = ring_container->total_bytes;
2201 int packets = ring_container->total_packets;
2204 u8 itr_setting = ring_container->itr;
2209 /* simple throttlerate management
2210 * 0-10MB/s lowest (100000 ints/s)
2211 * 10-20MB/s low (20000 ints/s)
2212 * 20-1249MB/s bulk (8000 ints/s)
2214 /* what was last interrupt timeslice? */
2215 timepassed_us = q_vector->itr >> 2;
2216 if (timepassed_us == 0)
2219 bytes_perint = bytes / timepassed_us; /* bytes/usec */
2221 switch (itr_setting) {
2222 case lowest_latency:
2223 if (bytes_perint > 10)
2224 itr_setting = low_latency;
2227 if (bytes_perint > 20)
2228 itr_setting = bulk_latency;
2229 else if (bytes_perint <= 10)
2230 itr_setting = lowest_latency;
2233 if (bytes_perint <= 20)
2234 itr_setting = low_latency;
2238 /* clear work counters since we have the values we need */
2239 ring_container->total_bytes = 0;
2240 ring_container->total_packets = 0;
2242 /* write updated itr to ring container */
2243 ring_container->itr = itr_setting;
2247 * ixgbe_write_eitr - write EITR register in hardware specific way
2248 * @q_vector: structure containing interrupt and ring information
2250 * This function is made to be called by ethtool and by the driver
2251 * when it needs to update EITR registers at runtime. Hardware
2252 * specific quirks/differences are taken care of here.
2254 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2256 struct ixgbe_adapter *adapter = q_vector->adapter;
2257 struct ixgbe_hw *hw = &adapter->hw;
2258 int v_idx = q_vector->v_idx;
2259 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2261 switch (adapter->hw.mac.type) {
2262 case ixgbe_mac_82598EB:
2263 /* must write high and low 16 bits to reset counter */
2264 itr_reg |= (itr_reg << 16);
2266 case ixgbe_mac_82599EB:
2267 case ixgbe_mac_X540:
2269 * set the WDIS bit to not clear the timer bits and cause an
2270 * immediate assertion of the interrupt
2272 itr_reg |= IXGBE_EITR_CNT_WDIS;
2277 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2280 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2282 u32 new_itr = q_vector->itr;
2285 ixgbe_update_itr(q_vector, &q_vector->tx);
2286 ixgbe_update_itr(q_vector, &q_vector->rx);
2288 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2290 switch (current_itr) {
2291 /* counts and packets in update_itr are dependent on these numbers */
2292 case lowest_latency:
2293 new_itr = IXGBE_100K_ITR;
2296 new_itr = IXGBE_20K_ITR;
2299 new_itr = IXGBE_8K_ITR;
2305 if (new_itr != q_vector->itr) {
2306 /* do an exponential smoothing */
2307 new_itr = (10 * new_itr * q_vector->itr) /
2308 ((9 * new_itr) + q_vector->itr);
2310 /* save the algorithm value here */
2311 q_vector->itr = new_itr;
2313 ixgbe_write_eitr(q_vector);
2318 * ixgbe_check_overtemp_subtask - check for over temperature
2319 * @adapter: pointer to adapter
2321 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2323 struct ixgbe_hw *hw = &adapter->hw;
2324 u32 eicr = adapter->interrupt_event;
2326 if (test_bit(__IXGBE_DOWN, &adapter->state))
2329 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2330 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2333 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2335 switch (hw->device_id) {
2336 case IXGBE_DEV_ID_82599_T3_LOM:
2338 * Since the warning interrupt is for both ports
2339 * we don't have to check if:
2340 * - This interrupt wasn't for our port.
2341 * - We may have missed the interrupt so always have to
2342 * check if we got a LSC
2344 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2345 !(eicr & IXGBE_EICR_LSC))
2348 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2350 bool link_up = false;
2352 hw->mac.ops.check_link(hw, &speed, &link_up, false);
2358 /* Check if this is not due to overtemp */
2359 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2364 if (!(eicr & IXGBE_EICR_GPI_SDP0))
2369 "Network adapter has been stopped because it has over heated. "
2370 "Restart the computer. If the problem persists, "
2371 "power off the system and replace the adapter\n");
2373 adapter->interrupt_event = 0;
2376 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2378 struct ixgbe_hw *hw = &adapter->hw;
2380 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2381 (eicr & IXGBE_EICR_GPI_SDP1)) {
2382 e_crit(probe, "Fan has stopped, replace the adapter\n");
2383 /* write to clear the interrupt */
2384 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2388 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2390 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2393 switch (adapter->hw.mac.type) {
2394 case ixgbe_mac_82599EB:
2396 * Need to check link state so complete overtemp check
2399 if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2400 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2401 adapter->interrupt_event = eicr;
2402 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2403 ixgbe_service_event_schedule(adapter);
2407 case ixgbe_mac_X540:
2408 if (!(eicr & IXGBE_EICR_TS))
2416 "Network adapter has been stopped because it has over heated. "
2417 "Restart the computer. If the problem persists, "
2418 "power off the system and replace the adapter\n");
2421 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2423 struct ixgbe_hw *hw = &adapter->hw;
2425 if (eicr & IXGBE_EICR_GPI_SDP2) {
2426 /* Clear the interrupt */
2427 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2428 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2429 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2430 ixgbe_service_event_schedule(adapter);
2434 if (eicr & IXGBE_EICR_GPI_SDP1) {
2435 /* Clear the interrupt */
2436 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2437 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2438 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2439 ixgbe_service_event_schedule(adapter);
2444 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2446 struct ixgbe_hw *hw = &adapter->hw;
2449 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2450 adapter->link_check_timeout = jiffies;
2451 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2452 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2453 IXGBE_WRITE_FLUSH(hw);
2454 ixgbe_service_event_schedule(adapter);
2458 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2462 struct ixgbe_hw *hw = &adapter->hw;
2464 switch (hw->mac.type) {
2465 case ixgbe_mac_82598EB:
2466 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2467 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2469 case ixgbe_mac_82599EB:
2470 case ixgbe_mac_X540:
2471 mask = (qmask & 0xFFFFFFFF);
2473 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2474 mask = (qmask >> 32);
2476 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2481 /* skip the flush */
2484 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2488 struct ixgbe_hw *hw = &adapter->hw;
2490 switch (hw->mac.type) {
2491 case ixgbe_mac_82598EB:
2492 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2493 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2495 case ixgbe_mac_82599EB:
2496 case ixgbe_mac_X540:
2497 mask = (qmask & 0xFFFFFFFF);
2499 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2500 mask = (qmask >> 32);
2502 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2507 /* skip the flush */
2511 * ixgbe_irq_enable - Enable default interrupt generation settings
2512 * @adapter: board private structure
2514 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2517 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2519 /* don't reenable LSC while waiting for link */
2520 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2521 mask &= ~IXGBE_EIMS_LSC;
2523 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2524 switch (adapter->hw.mac.type) {
2525 case ixgbe_mac_82599EB:
2526 mask |= IXGBE_EIMS_GPI_SDP0;
2528 case ixgbe_mac_X540:
2529 mask |= IXGBE_EIMS_TS;
2534 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2535 mask |= IXGBE_EIMS_GPI_SDP1;
2536 switch (adapter->hw.mac.type) {
2537 case ixgbe_mac_82599EB:
2538 mask |= IXGBE_EIMS_GPI_SDP1;
2539 mask |= IXGBE_EIMS_GPI_SDP2;
2540 case ixgbe_mac_X540:
2541 mask |= IXGBE_EIMS_ECC;
2542 mask |= IXGBE_EIMS_MAILBOX;
2548 if (adapter->hw.mac.type == ixgbe_mac_X540)
2549 mask |= IXGBE_EIMS_TIMESYNC;
2551 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2552 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2553 mask |= IXGBE_EIMS_FLOW_DIR;
2555 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2557 ixgbe_irq_enable_queues(adapter, ~0);
2559 IXGBE_WRITE_FLUSH(&adapter->hw);
2562 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2564 struct ixgbe_adapter *adapter = data;
2565 struct ixgbe_hw *hw = &adapter->hw;
2569 * Workaround for Silicon errata. Use clear-by-write instead
2570 * of clear-by-read. Reading with EICS will return the
2571 * interrupt causes without clearing, which later be done
2572 * with the write to EICR.
2574 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2576 /* The lower 16bits of the EICR register are for the queue interrupts
2577 * which should be masked here in order to not accidently clear them if
2578 * the bits are high when ixgbe_msix_other is called. There is a race
2579 * condition otherwise which results in possible performance loss
2580 * especially if the ixgbe_msix_other interrupt is triggering
2581 * consistently (as it would when PPS is turned on for the X540 device)
2585 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2587 if (eicr & IXGBE_EICR_LSC)
2588 ixgbe_check_lsc(adapter);
2590 if (eicr & IXGBE_EICR_MAILBOX)
2591 ixgbe_msg_task(adapter);
2593 switch (hw->mac.type) {
2594 case ixgbe_mac_82599EB:
2595 case ixgbe_mac_X540:
2596 if (eicr & IXGBE_EICR_ECC)
2597 e_info(link, "Received unrecoverable ECC Err, please "
2599 /* Handle Flow Director Full threshold interrupt */
2600 if (eicr & IXGBE_EICR_FLOW_DIR) {
2601 int reinit_count = 0;
2603 for (i = 0; i < adapter->num_tx_queues; i++) {
2604 struct ixgbe_ring *ring = adapter->tx_ring[i];
2605 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2610 /* no more flow director interrupts until after init */
2611 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2612 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2613 ixgbe_service_event_schedule(adapter);
2616 ixgbe_check_sfp_event(adapter, eicr);
2617 ixgbe_check_overtemp_event(adapter, eicr);
2623 ixgbe_check_fan_failure(adapter, eicr);
2625 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2626 ixgbe_ptp_check_pps_event(adapter, eicr);
2628 /* re-enable the original interrupt state, no lsc, no queues */
2629 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2630 ixgbe_irq_enable(adapter, false, false);
2635 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2637 struct ixgbe_q_vector *q_vector = data;
2639 /* EIAM disabled interrupts (on this vector) for us */
2641 if (q_vector->rx.ring || q_vector->tx.ring)
2642 napi_schedule(&q_vector->napi);
2648 * ixgbe_poll - NAPI Rx polling callback
2649 * @napi: structure for representing this polling device
2650 * @budget: how many packets driver is allowed to clean
2652 * This function is used for legacy and MSI, NAPI mode
2654 int ixgbe_poll(struct napi_struct *napi, int budget)
2656 struct ixgbe_q_vector *q_vector =
2657 container_of(napi, struct ixgbe_q_vector, napi);
2658 struct ixgbe_adapter *adapter = q_vector->adapter;
2659 struct ixgbe_ring *ring;
2660 int per_ring_budget;
2661 bool clean_complete = true;
2663 #ifdef CONFIG_IXGBE_DCA
2664 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2665 ixgbe_update_dca(q_vector);
2668 ixgbe_for_each_ring(ring, q_vector->tx)
2669 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2671 if (!ixgbe_qv_lock_napi(q_vector))
2674 /* attempt to distribute budget to each queue fairly, but don't allow
2675 * the budget to go below 1 because we'll exit polling */
2676 if (q_vector->rx.count > 1)
2677 per_ring_budget = max(budget/q_vector->rx.count, 1);
2679 per_ring_budget = budget;
2681 ixgbe_for_each_ring(ring, q_vector->rx)
2682 clean_complete &= (ixgbe_clean_rx_irq(q_vector, ring,
2683 per_ring_budget) < per_ring_budget);
2685 ixgbe_qv_unlock_napi(q_vector);
2686 /* If all work not completed, return budget and keep polling */
2687 if (!clean_complete)
2690 /* all work done, exit the polling mode */
2691 napi_complete(napi);
2692 if (adapter->rx_itr_setting & 1)
2693 ixgbe_set_itr(q_vector);
2694 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2695 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2701 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2702 * @adapter: board private structure
2704 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2705 * interrupts from the kernel.
2707 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2709 struct net_device *netdev = adapter->netdev;
2713 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2714 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2715 struct msix_entry *entry = &adapter->msix_entries[vector];
2717 if (q_vector->tx.ring && q_vector->rx.ring) {
2718 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2719 "%s-%s-%d", netdev->name, "TxRx", ri++);
2721 } else if (q_vector->rx.ring) {
2722 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2723 "%s-%s-%d", netdev->name, "rx", ri++);
2724 } else if (q_vector->tx.ring) {
2725 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2726 "%s-%s-%d", netdev->name, "tx", ti++);
2728 /* skip this unused q_vector */
2731 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2732 q_vector->name, q_vector);
2734 e_err(probe, "request_irq failed for MSIX interrupt "
2735 "Error: %d\n", err);
2736 goto free_queue_irqs;
2738 /* If Flow Director is enabled, set interrupt affinity */
2739 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2740 /* assign the mask for this irq */
2741 irq_set_affinity_hint(entry->vector,
2742 &q_vector->affinity_mask);
2746 err = request_irq(adapter->msix_entries[vector].vector,
2747 ixgbe_msix_other, 0, netdev->name, adapter);
2749 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2750 goto free_queue_irqs;
2758 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2760 free_irq(adapter->msix_entries[vector].vector,
2761 adapter->q_vector[vector]);
2763 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2764 pci_disable_msix(adapter->pdev);
2765 kfree(adapter->msix_entries);
2766 adapter->msix_entries = NULL;
2771 * ixgbe_intr - legacy mode Interrupt Handler
2772 * @irq: interrupt number
2773 * @data: pointer to a network interface device structure
2775 static irqreturn_t ixgbe_intr(int irq, void *data)
2777 struct ixgbe_adapter *adapter = data;
2778 struct ixgbe_hw *hw = &adapter->hw;
2779 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2783 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2784 * before the read of EICR.
2786 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2788 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2789 * therefore no explicit interrupt disable is necessary */
2790 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2793 * shared interrupt alert!
2794 * make sure interrupts are enabled because the read will
2795 * have disabled interrupts due to EIAM
2796 * finish the workaround of silicon errata on 82598. Unmask
2797 * the interrupt that we masked before the EICR read.
2799 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2800 ixgbe_irq_enable(adapter, true, true);
2801 return IRQ_NONE; /* Not our interrupt */
2804 if (eicr & IXGBE_EICR_LSC)
2805 ixgbe_check_lsc(adapter);
2807 switch (hw->mac.type) {
2808 case ixgbe_mac_82599EB:
2809 ixgbe_check_sfp_event(adapter, eicr);
2811 case ixgbe_mac_X540:
2812 if (eicr & IXGBE_EICR_ECC)
2813 e_info(link, "Received unrecoverable ECC err, please "
2815 ixgbe_check_overtemp_event(adapter, eicr);
2821 ixgbe_check_fan_failure(adapter, eicr);
2822 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2823 ixgbe_ptp_check_pps_event(adapter, eicr);
2825 /* would disable interrupts here but EIAM disabled it */
2826 napi_schedule(&q_vector->napi);
2829 * re-enable link(maybe) and non-queue interrupts, no flush.
2830 * ixgbe_poll will re-enable the queue interrupts
2832 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2833 ixgbe_irq_enable(adapter, false, false);
2839 * ixgbe_request_irq - initialize interrupts
2840 * @adapter: board private structure
2842 * Attempts to configure interrupts using the best available
2843 * capabilities of the hardware and kernel.
2845 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2847 struct net_device *netdev = adapter->netdev;
2850 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2851 err = ixgbe_request_msix_irqs(adapter);
2852 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2853 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2854 netdev->name, adapter);
2856 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2857 netdev->name, adapter);
2860 e_err(probe, "request_irq failed, Error %d\n", err);
2865 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2869 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2870 free_irq(adapter->pdev->irq, adapter);
2874 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2875 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2876 struct msix_entry *entry = &adapter->msix_entries[vector];
2878 /* free only the irqs that were actually requested */
2879 if (!q_vector->rx.ring && !q_vector->tx.ring)
2882 /* clear the affinity_mask in the IRQ descriptor */
2883 irq_set_affinity_hint(entry->vector, NULL);
2885 free_irq(entry->vector, q_vector);
2888 free_irq(adapter->msix_entries[vector++].vector, adapter);
2892 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2893 * @adapter: board private structure
2895 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2897 switch (adapter->hw.mac.type) {
2898 case ixgbe_mac_82598EB:
2899 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2901 case ixgbe_mac_82599EB:
2902 case ixgbe_mac_X540:
2903 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2904 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2905 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2910 IXGBE_WRITE_FLUSH(&adapter->hw);
2911 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2914 for (vector = 0; vector < adapter->num_q_vectors; vector++)
2915 synchronize_irq(adapter->msix_entries[vector].vector);
2917 synchronize_irq(adapter->msix_entries[vector++].vector);
2919 synchronize_irq(adapter->pdev->irq);
2924 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2927 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2929 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2931 ixgbe_write_eitr(q_vector);
2933 ixgbe_set_ivar(adapter, 0, 0, 0);
2934 ixgbe_set_ivar(adapter, 1, 0, 0);
2936 e_info(hw, "Legacy interrupt IVAR setup done\n");
2940 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2941 * @adapter: board private structure
2942 * @ring: structure containing ring specific data
2944 * Configure the Tx descriptor ring after a reset.
2946 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2947 struct ixgbe_ring *ring)
2949 struct ixgbe_hw *hw = &adapter->hw;
2950 u64 tdba = ring->dma;
2952 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2953 u8 reg_idx = ring->reg_idx;
2955 /* disable queue to avoid issues while updating state */
2956 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2957 IXGBE_WRITE_FLUSH(hw);
2959 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2960 (tdba & DMA_BIT_MASK(32)));
2961 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2962 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2963 ring->count * sizeof(union ixgbe_adv_tx_desc));
2964 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2965 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2966 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2969 * set WTHRESH to encourage burst writeback, it should not be set
2970 * higher than 1 when:
2971 * - ITR is 0 as it could cause false TX hangs
2972 * - ITR is set to > 100k int/sec and BQL is enabled
2974 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2975 * to or less than the number of on chip descriptors, which is
2978 #if IS_ENABLED(CONFIG_BQL)
2979 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
2981 if (!ring->q_vector || (ring->q_vector->itr < 8))
2983 txdctl |= (1 << 16); /* WTHRESH = 1 */
2985 txdctl |= (8 << 16); /* WTHRESH = 8 */
2988 * Setting PTHRESH to 32 both improves performance
2989 * and avoids a TX hang with DFP enabled
2991 txdctl |= (1 << 8) | /* HTHRESH = 1 */
2992 32; /* PTHRESH = 32 */
2994 /* reinitialize flowdirector state */
2995 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2996 ring->atr_sample_rate = adapter->atr_sample_rate;
2997 ring->atr_count = 0;
2998 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3000 ring->atr_sample_rate = 0;
3003 /* initialize XPS */
3004 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3005 struct ixgbe_q_vector *q_vector = ring->q_vector;
3008 netif_set_xps_queue(adapter->netdev,
3009 &q_vector->affinity_mask,
3013 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3016 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3018 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3019 if (hw->mac.type == ixgbe_mac_82598EB &&
3020 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3023 /* poll to verify queue is enabled */
3025 usleep_range(1000, 2000);
3026 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3027 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3029 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
3032 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3034 struct ixgbe_hw *hw = &adapter->hw;
3036 u8 tcs = netdev_get_num_tc(adapter->netdev);
3038 if (hw->mac.type == ixgbe_mac_82598EB)
3041 /* disable the arbiter while setting MTQC */
3042 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3043 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3044 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3046 /* set transmit pool layout */
3047 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3048 mtqc = IXGBE_MTQC_VT_ENA;
3050 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3052 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3053 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3054 mtqc |= IXGBE_MTQC_32VF;
3056 mtqc |= IXGBE_MTQC_64VF;
3059 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3061 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3063 mtqc = IXGBE_MTQC_64Q_1PB;
3066 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3068 /* Enable Security TX Buffer IFG for multiple pb */
3070 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3071 sectx |= IXGBE_SECTX_DCB;
3072 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3075 /* re-enable the arbiter */
3076 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3077 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3081 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3082 * @adapter: board private structure
3084 * Configure the Tx unit of the MAC after a reset.
3086 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3088 struct ixgbe_hw *hw = &adapter->hw;
3092 ixgbe_setup_mtqc(adapter);
3094 if (hw->mac.type != ixgbe_mac_82598EB) {
3095 /* DMATXCTL.EN must be before Tx queues are enabled */
3096 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3097 dmatxctl |= IXGBE_DMATXCTL_TE;
3098 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3101 /* Setup the HW Tx Head and Tail descriptor pointers */
3102 for (i = 0; i < adapter->num_tx_queues; i++)
3103 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3106 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3107 struct ixgbe_ring *ring)
3109 struct ixgbe_hw *hw = &adapter->hw;
3110 u8 reg_idx = ring->reg_idx;
3111 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3113 srrctl |= IXGBE_SRRCTL_DROP_EN;
3115 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3118 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3119 struct ixgbe_ring *ring)
3121 struct ixgbe_hw *hw = &adapter->hw;
3122 u8 reg_idx = ring->reg_idx;
3123 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3125 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3127 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3130 #ifdef CONFIG_IXGBE_DCB
3131 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3133 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3137 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3139 if (adapter->ixgbe_ieee_pfc)
3140 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3143 * We should set the drop enable bit if:
3146 * Number of Rx queues > 1 and flow control is disabled
3148 * This allows us to avoid head of line blocking for security
3149 * and performance reasons.
3151 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3152 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3153 for (i = 0; i < adapter->num_rx_queues; i++)
3154 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3156 for (i = 0; i < adapter->num_rx_queues; i++)
3157 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3161 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3163 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3164 struct ixgbe_ring *rx_ring)
3166 struct ixgbe_hw *hw = &adapter->hw;
3168 u8 reg_idx = rx_ring->reg_idx;
3170 if (hw->mac.type == ixgbe_mac_82598EB) {
3171 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3174 * if VMDq is not active we must program one srrctl register
3175 * per RSS queue since we have enabled RDRXCTL.MVMEN
3180 /* configure header buffer length, needed for RSC */
3181 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3183 /* configure the packet buffer length */
3184 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3186 /* configure descriptor type */
3187 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3189 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3192 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3194 struct ixgbe_hw *hw = &adapter->hw;
3195 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
3196 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
3197 0x6A3E67EA, 0x14364D17, 0x3BED200D};
3198 u32 mrqc = 0, reta = 0;
3201 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3204 * Program table for at least 2 queues w/ SR-IOV so that VFs can
3205 * make full use of any rings they may have. We will use the
3206 * PSRTYPE register to control how many rings we use within the PF.
3208 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
3211 /* Fill out hash function seeds */
3212 for (i = 0; i < 10; i++)
3213 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
3215 /* Fill out redirection table */
3216 for (i = 0, j = 0; i < 128; i++, j++) {
3219 /* reta = 4-byte sliding window of
3220 * 0x00..(indices-1)(indices-1)00..etc. */
3221 reta = (reta << 8) | (j * 0x11);
3223 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3226 /* Disable indicating checksum in descriptor, enables RSS hash */
3227 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3228 rxcsum |= IXGBE_RXCSUM_PCSD;
3229 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3231 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3232 if (adapter->ring_feature[RING_F_RSS].mask)
3233 mrqc = IXGBE_MRQC_RSSEN;
3235 u8 tcs = netdev_get_num_tc(adapter->netdev);
3237 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3239 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3241 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3242 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3243 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3245 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3248 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3250 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3252 mrqc = IXGBE_MRQC_RSSEN;
3256 /* Perform hash on these packet types */
3257 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3258 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3259 IXGBE_MRQC_RSS_FIELD_IPV6 |
3260 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3262 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3263 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3264 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3265 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3267 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3271 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3272 * @adapter: address of board private structure
3273 * @index: index of ring to set
3275 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3276 struct ixgbe_ring *ring)
3278 struct ixgbe_hw *hw = &adapter->hw;
3280 u8 reg_idx = ring->reg_idx;
3282 if (!ring_is_rsc_enabled(ring))
3285 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3286 rscctrl |= IXGBE_RSCCTL_RSCEN;
3288 * we must limit the number of descriptors so that the
3289 * total size of max desc * buf_len is not greater
3292 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3293 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3296 #define IXGBE_MAX_RX_DESC_POLL 10
3297 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3298 struct ixgbe_ring *ring)
3300 struct ixgbe_hw *hw = &adapter->hw;
3301 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3303 u8 reg_idx = ring->reg_idx;
3305 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3306 if (hw->mac.type == ixgbe_mac_82598EB &&
3307 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3311 usleep_range(1000, 2000);
3312 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3313 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3316 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3317 "the polling period\n", reg_idx);
3321 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3322 struct ixgbe_ring *ring)
3324 struct ixgbe_hw *hw = &adapter->hw;
3325 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3327 u8 reg_idx = ring->reg_idx;
3329 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3330 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3332 /* write value back with RXDCTL.ENABLE bit cleared */
3333 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3335 if (hw->mac.type == ixgbe_mac_82598EB &&
3336 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3339 /* the hardware may take up to 100us to really disable the rx queue */
3342 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3343 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3346 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3347 "the polling period\n", reg_idx);
3351 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3352 struct ixgbe_ring *ring)
3354 struct ixgbe_hw *hw = &adapter->hw;
3355 u64 rdba = ring->dma;
3357 u8 reg_idx = ring->reg_idx;
3359 /* disable queue to avoid issues while updating state */
3360 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3361 ixgbe_disable_rx_queue(adapter, ring);
3363 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3364 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3365 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3366 ring->count * sizeof(union ixgbe_adv_rx_desc));
3367 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3368 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3369 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3371 ixgbe_configure_srrctl(adapter, ring);
3372 ixgbe_configure_rscctl(adapter, ring);
3374 if (hw->mac.type == ixgbe_mac_82598EB) {
3376 * enable cache line friendly hardware writes:
3377 * PTHRESH=32 descriptors (half the internal cache),
3378 * this also removes ugly rx_no_buffer_count increment
3379 * HTHRESH=4 descriptors (to minimize latency on fetch)
3380 * WTHRESH=8 burst writeback up to two cache lines
3382 rxdctl &= ~0x3FFFFF;
3386 /* enable receive descriptor ring */
3387 rxdctl |= IXGBE_RXDCTL_ENABLE;
3388 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3390 ixgbe_rx_desc_queue_enable(adapter, ring);
3391 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3394 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3396 struct ixgbe_hw *hw = &adapter->hw;
3397 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3400 /* PSRTYPE must be initialized in non 82598 adapters */
3401 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3402 IXGBE_PSRTYPE_UDPHDR |
3403 IXGBE_PSRTYPE_IPV4HDR |
3404 IXGBE_PSRTYPE_L2HDR |
3405 IXGBE_PSRTYPE_IPV6HDR;
3407 if (hw->mac.type == ixgbe_mac_82598EB)
3415 for (p = 0; p < adapter->num_rx_pools; p++)
3416 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(p)),
3420 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3422 struct ixgbe_hw *hw = &adapter->hw;
3423 u32 reg_offset, vf_shift;
3424 u32 gcr_ext, vmdctl;
3427 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3430 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3431 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3432 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3433 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3434 vmdctl |= IXGBE_VT_CTL_REPLEN;
3435 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3437 vf_shift = VMDQ_P(0) % 32;
3438 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3440 /* Enable only the PF's pool for Tx/Rx */
3441 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3442 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3443 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3444 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3445 if (adapter->flags2 & IXGBE_FLAG2_BRIDGE_MODE_VEB)
3446 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3448 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3449 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3452 * Set up VF register offsets for selected VT Mode,
3453 * i.e. 32 or 64 VFs for SR-IOV
3455 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3456 case IXGBE_82599_VMDQ_8Q_MASK:
3457 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3459 case IXGBE_82599_VMDQ_4Q_MASK:
3460 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3463 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3467 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3470 /* Enable MAC Anti-Spoofing */
3471 hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3473 /* For VFs that have spoof checking turned off */
3474 for (i = 0; i < adapter->num_vfs; i++) {
3475 if (!adapter->vfinfo[i].spoofchk_enabled)
3476 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3480 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3482 struct ixgbe_hw *hw = &adapter->hw;
3483 struct net_device *netdev = adapter->netdev;
3484 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3485 struct ixgbe_ring *rx_ring;
3490 /* adjust max frame to be able to do baby jumbo for FCoE */
3491 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3492 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3493 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3495 #endif /* IXGBE_FCOE */
3497 /* adjust max frame to be at least the size of a standard frame */
3498 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3499 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3501 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3502 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3503 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3504 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3506 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3509 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3510 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3511 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3512 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3515 * Setup the HW Rx Head and Tail Descriptor Pointers and
3516 * the Base and Length of the Rx Descriptor Ring
3518 for (i = 0; i < adapter->num_rx_queues; i++) {
3519 rx_ring = adapter->rx_ring[i];
3520 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3521 set_ring_rsc_enabled(rx_ring);
3523 clear_ring_rsc_enabled(rx_ring);
3527 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3529 struct ixgbe_hw *hw = &adapter->hw;
3530 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3532 switch (hw->mac.type) {
3533 case ixgbe_mac_82598EB:
3535 * For VMDq support of different descriptor types or
3536 * buffer sizes through the use of multiple SRRCTL
3537 * registers, RDRXCTL.MVMEN must be set to 1
3539 * also, the manual doesn't mention it clearly but DCA hints
3540 * will only use queue 0's tags unless this bit is set. Side
3541 * effects of setting this bit are only that SRRCTL must be
3542 * fully programmed [0..15]
3544 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3546 case ixgbe_mac_82599EB:
3547 case ixgbe_mac_X540:
3548 /* Disable RSC for ACK packets */
3549 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3550 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3551 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3552 /* hardware requires some bits to be set by default */
3553 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3554 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3557 /* We should do nothing since we don't know this hardware */
3561 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3565 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3566 * @adapter: board private structure
3568 * Configure the Rx unit of the MAC after a reset.
3570 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3572 struct ixgbe_hw *hw = &adapter->hw;
3576 /* disable receives while setting up the descriptors */
3577 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3578 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3580 ixgbe_setup_psrtype(adapter);
3581 ixgbe_setup_rdrxctl(adapter);
3583 /* Program registers for the distribution of queues */
3584 ixgbe_setup_mrqc(adapter);
3586 /* set_rx_buffer_len must be called before ring initialization */
3587 ixgbe_set_rx_buffer_len(adapter);
3590 * Setup the HW Rx Head and Tail Descriptor Pointers and
3591 * the Base and Length of the Rx Descriptor Ring
3593 for (i = 0; i < adapter->num_rx_queues; i++)
3594 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3596 /* disable drop enable for 82598 parts */
3597 if (hw->mac.type == ixgbe_mac_82598EB)
3598 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3600 /* enable all receives */
3601 rxctrl |= IXGBE_RXCTRL_RXEN;
3602 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3605 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
3606 __be16 proto, u16 vid)
3608 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3609 struct ixgbe_hw *hw = &adapter->hw;
3611 /* add VID to filter table */
3612 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3613 set_bit(vid, adapter->active_vlans);
3618 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
3619 __be16 proto, u16 vid)
3621 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3622 struct ixgbe_hw *hw = &adapter->hw;
3624 /* remove VID from filter table */
3625 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3626 clear_bit(vid, adapter->active_vlans);
3632 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3633 * @adapter: driver data
3635 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3637 struct ixgbe_hw *hw = &adapter->hw;
3640 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3641 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3642 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3646 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3647 * @adapter: driver data
3649 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3651 struct ixgbe_hw *hw = &adapter->hw;
3654 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3655 vlnctrl |= IXGBE_VLNCTRL_VFE;
3656 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3657 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3661 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3662 * @adapter: driver data
3664 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3666 struct ixgbe_hw *hw = &adapter->hw;
3670 switch (hw->mac.type) {
3671 case ixgbe_mac_82598EB:
3672 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3673 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3674 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3676 case ixgbe_mac_82599EB:
3677 case ixgbe_mac_X540:
3678 for (i = 0; i < adapter->num_rx_queues; i++) {
3679 j = adapter->rx_ring[i]->reg_idx;
3680 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3681 vlnctrl &= ~IXGBE_RXDCTL_VME;
3682 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3691 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3692 * @adapter: driver data
3694 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3696 struct ixgbe_hw *hw = &adapter->hw;
3700 switch (hw->mac.type) {
3701 case ixgbe_mac_82598EB:
3702 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3703 vlnctrl |= IXGBE_VLNCTRL_VME;
3704 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3706 case ixgbe_mac_82599EB:
3707 case ixgbe_mac_X540:
3708 for (i = 0; i < adapter->num_rx_queues; i++) {
3709 j = adapter->rx_ring[i]->reg_idx;
3710 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3711 vlnctrl |= IXGBE_RXDCTL_VME;
3712 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3720 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3724 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
3726 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3727 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
3731 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3732 * @netdev: network interface device structure
3734 * Writes unicast address list to the RAR table.
3735 * Returns: -ENOMEM on failure/insufficient address space
3736 * 0 on no addresses written
3737 * X on writing X addresses to the RAR table
3739 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3741 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3742 struct ixgbe_hw *hw = &adapter->hw;
3743 unsigned int rar_entries = hw->mac.num_rar_entries - 1;
3746 /* In SR-IOV mode significantly less RAR entries are available */
3747 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3748 rar_entries = IXGBE_MAX_PF_MACVLANS - 1;
3750 /* return ENOMEM indicating insufficient memory for addresses */
3751 if (netdev_uc_count(netdev) > rar_entries)
3754 if (!netdev_uc_empty(netdev)) {
3755 struct netdev_hw_addr *ha;
3756 /* return error if we do not support writing to RAR table */
3757 if (!hw->mac.ops.set_rar)
3760 netdev_for_each_uc_addr(ha, netdev) {
3763 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3764 VMDQ_P(0), IXGBE_RAH_AV);
3768 /* write the addresses in reverse order to avoid write combining */
3769 for (; rar_entries > 0 ; rar_entries--)
3770 hw->mac.ops.clear_rar(hw, rar_entries);
3776 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3777 * @netdev: network interface device structure
3779 * The set_rx_method entry point is called whenever the unicast/multicast
3780 * address list or the network interface flags are updated. This routine is
3781 * responsible for configuring the hardware for proper unicast, multicast and
3784 void ixgbe_set_rx_mode(struct net_device *netdev)
3786 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3787 struct ixgbe_hw *hw = &adapter->hw;
3788 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3791 /* Check for Promiscuous and All Multicast modes */
3793 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3795 /* set all bits that we expect to always be set */
3796 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3797 fctrl |= IXGBE_FCTRL_BAM;
3798 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3799 fctrl |= IXGBE_FCTRL_PMCF;
3801 /* clear the bits we are changing the status of */
3802 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3804 if (netdev->flags & IFF_PROMISC) {
3805 hw->addr_ctrl.user_set_promisc = true;
3806 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3807 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3808 /* Only disable hardware filter vlans in promiscuous mode
3809 * if SR-IOV and VMDQ are disabled - otherwise ensure
3810 * that hardware VLAN filters remain enabled.
3812 if (!(adapter->flags & (IXGBE_FLAG_VMDQ_ENABLED |
3813 IXGBE_FLAG_SRIOV_ENABLED)))
3814 ixgbe_vlan_filter_disable(adapter);
3816 ixgbe_vlan_filter_enable(adapter);
3818 if (netdev->flags & IFF_ALLMULTI) {
3819 fctrl |= IXGBE_FCTRL_MPE;
3820 vmolr |= IXGBE_VMOLR_MPE;
3823 * Write addresses to the MTA, if the attempt fails
3824 * then we should just turn on promiscuous mode so
3825 * that we can at least receive multicast traffic
3827 hw->mac.ops.update_mc_addr_list(hw, netdev);
3828 vmolr |= IXGBE_VMOLR_ROMPE;
3830 ixgbe_vlan_filter_enable(adapter);
3831 hw->addr_ctrl.user_set_promisc = false;
3835 * Write addresses to available RAR registers, if there is not
3836 * sufficient space to store all the addresses then enable
3837 * unicast promiscuous mode
3839 count = ixgbe_write_uc_addr_list(netdev);
3841 fctrl |= IXGBE_FCTRL_UPE;
3842 vmolr |= IXGBE_VMOLR_ROPE;
3845 if (adapter->num_vfs)
3846 ixgbe_restore_vf_multicasts(adapter);
3848 if (hw->mac.type != ixgbe_mac_82598EB) {
3849 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
3850 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3852 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
3855 /* This is useful for sniffing bad packets. */
3856 if (adapter->netdev->features & NETIF_F_RXALL) {
3857 /* UPE and MPE will be handled by normal PROMISC logic
3858 * in e1000e_set_rx_mode */
3859 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3860 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3861 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3863 fctrl &= ~(IXGBE_FCTRL_DPF);
3864 /* NOTE: VLAN filtering is disabled by setting PROMISC */
3867 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3869 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
3870 ixgbe_vlan_strip_enable(adapter);
3872 ixgbe_vlan_strip_disable(adapter);
3875 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3879 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
3880 ixgbe_qv_init_lock(adapter->q_vector[q_idx]);
3881 napi_enable(&adapter->q_vector[q_idx]->napi);
3885 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3889 local_bh_disable(); /* for ixgbe_qv_lock_napi() */
3890 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
3891 napi_disable(&adapter->q_vector[q_idx]->napi);
3892 while (!ixgbe_qv_lock_napi(adapter->q_vector[q_idx])) {
3893 pr_info("QV %d locked\n", q_idx);
3900 #ifdef CONFIG_IXGBE_DCB
3902 * ixgbe_configure_dcb - Configure DCB hardware
3903 * @adapter: ixgbe adapter struct
3905 * This is called by the driver on open to configure the DCB hardware.
3906 * This is also called by the gennetlink interface when reconfiguring
3909 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3911 struct ixgbe_hw *hw = &adapter->hw;
3912 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3914 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3915 if (hw->mac.type == ixgbe_mac_82598EB)
3916 netif_set_gso_max_size(adapter->netdev, 65536);
3920 if (hw->mac.type == ixgbe_mac_82598EB)
3921 netif_set_gso_max_size(adapter->netdev, 32768);
3924 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3925 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3928 /* reconfigure the hardware */
3929 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3930 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3932 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3934 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3935 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3936 ixgbe_dcb_hw_ets(&adapter->hw,
3937 adapter->ixgbe_ieee_ets,
3939 ixgbe_dcb_hw_pfc_config(&adapter->hw,
3940 adapter->ixgbe_ieee_pfc->pfc_en,
3941 adapter->ixgbe_ieee_ets->prio_tc);
3944 /* Enable RSS Hash per TC */
3945 if (hw->mac.type != ixgbe_mac_82598EB) {
3947 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
3954 /* write msb to all 8 TCs in one write */
3955 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
3960 /* Additional bittime to account for IXGBE framing */
3961 #define IXGBE_ETH_FRAMING 20
3964 * ixgbe_hpbthresh - calculate high water mark for flow control
3966 * @adapter: board private structure to calculate for
3967 * @pb: packet buffer to calculate
3969 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3971 struct ixgbe_hw *hw = &adapter->hw;
3972 struct net_device *dev = adapter->netdev;
3973 int link, tc, kb, marker;
3976 /* Calculate max LAN frame size */
3977 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3980 /* FCoE traffic class uses FCOE jumbo frames */
3981 if ((dev->features & NETIF_F_FCOE_MTU) &&
3982 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
3983 (pb == ixgbe_fcoe_get_tc(adapter)))
3984 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3987 /* Calculate delay value for device */
3988 switch (hw->mac.type) {
3989 case ixgbe_mac_X540:
3990 dv_id = IXGBE_DV_X540(link, tc);
3993 dv_id = IXGBE_DV(link, tc);
3997 /* Loopback switch introduces additional latency */
3998 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3999 dv_id += IXGBE_B2BT(tc);
4001 /* Delay value is calculated in bit times convert to KB */
4002 kb = IXGBE_BT2KB(dv_id);
4003 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
4005 marker = rx_pba - kb;
4007 /* It is possible that the packet buffer is not large enough
4008 * to provide required headroom. In this case throw an error
4009 * to user and a do the best we can.
4012 e_warn(drv, "Packet Buffer(%i) can not provide enough"
4013 "headroom to support flow control."
4014 "Decrease MTU or number of traffic classes\n", pb);
4022 * ixgbe_lpbthresh - calculate low water mark for for flow control
4024 * @adapter: board private structure to calculate for
4025 * @pb: packet buffer to calculate
4027 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
4029 struct ixgbe_hw *hw = &adapter->hw;
4030 struct net_device *dev = adapter->netdev;
4034 /* Calculate max LAN frame size */
4035 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4037 /* Calculate delay value for device */
4038 switch (hw->mac.type) {
4039 case ixgbe_mac_X540:
4040 dv_id = IXGBE_LOW_DV_X540(tc);
4043 dv_id = IXGBE_LOW_DV(tc);
4047 /* Delay value is calculated in bit times convert to KB */
4048 return IXGBE_BT2KB(dv_id);
4052 * ixgbe_pbthresh_setup - calculate and setup high low water marks
4054 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
4056 struct ixgbe_hw *hw = &adapter->hw;
4057 int num_tc = netdev_get_num_tc(adapter->netdev);
4063 hw->fc.low_water = ixgbe_lpbthresh(adapter);
4065 for (i = 0; i < num_tc; i++) {
4066 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
4068 /* Low water marks must not be larger than high water marks */
4069 if (hw->fc.low_water > hw->fc.high_water[i])
4070 hw->fc.low_water = 0;
4074 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
4076 struct ixgbe_hw *hw = &adapter->hw;
4078 u8 tc = netdev_get_num_tc(adapter->netdev);
4080 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4081 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4082 hdrm = 32 << adapter->fdir_pballoc;
4086 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
4087 ixgbe_pbthresh_setup(adapter);
4090 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
4092 struct ixgbe_hw *hw = &adapter->hw;
4093 struct hlist_node *node2;
4094 struct ixgbe_fdir_filter *filter;
4096 spin_lock(&adapter->fdir_perfect_lock);
4098 if (!hlist_empty(&adapter->fdir_filter_list))
4099 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
4101 hlist_for_each_entry_safe(filter, node2,
4102 &adapter->fdir_filter_list, fdir_node) {
4103 ixgbe_fdir_write_perfect_filter_82599(hw,
4106 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
4107 IXGBE_FDIR_DROP_QUEUE :
4108 adapter->rx_ring[filter->action]->reg_idx);
4111 spin_unlock(&adapter->fdir_perfect_lock);
4114 static void ixgbe_configure(struct ixgbe_adapter *adapter)
4116 struct ixgbe_hw *hw = &adapter->hw;
4118 ixgbe_configure_pb(adapter);
4119 #ifdef CONFIG_IXGBE_DCB
4120 ixgbe_configure_dcb(adapter);
4123 * We must restore virtualization before VLANs or else
4124 * the VLVF registers will not be populated
4126 ixgbe_configure_virtualization(adapter);
4128 ixgbe_set_rx_mode(adapter->netdev);
4129 ixgbe_restore_vlan(adapter);
4131 switch (hw->mac.type) {
4132 case ixgbe_mac_82599EB:
4133 case ixgbe_mac_X540:
4134 hw->mac.ops.disable_rx_buff(hw);
4140 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4141 ixgbe_init_fdir_signature_82599(&adapter->hw,
4142 adapter->fdir_pballoc);
4143 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
4144 ixgbe_init_fdir_perfect_82599(&adapter->hw,
4145 adapter->fdir_pballoc);
4146 ixgbe_fdir_filter_restore(adapter);
4149 switch (hw->mac.type) {
4150 case ixgbe_mac_82599EB:
4151 case ixgbe_mac_X540:
4152 hw->mac.ops.enable_rx_buff(hw);
4159 /* configure FCoE L2 filters, redirection table, and Rx control */
4160 ixgbe_configure_fcoe(adapter);
4162 #endif /* IXGBE_FCOE */
4163 ixgbe_configure_tx(adapter);
4164 ixgbe_configure_rx(adapter);
4167 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
4169 switch (hw->phy.type) {
4170 case ixgbe_phy_sfp_avago:
4171 case ixgbe_phy_sfp_ftl:
4172 case ixgbe_phy_sfp_intel:
4173 case ixgbe_phy_sfp_unknown:
4174 case ixgbe_phy_sfp_passive_tyco:
4175 case ixgbe_phy_sfp_passive_unknown:
4176 case ixgbe_phy_sfp_active_unknown:
4177 case ixgbe_phy_sfp_ftl_active:
4178 case ixgbe_phy_qsfp_passive_unknown:
4179 case ixgbe_phy_qsfp_active_unknown:
4180 case ixgbe_phy_qsfp_intel:
4181 case ixgbe_phy_qsfp_unknown:
4184 if (hw->mac.type == ixgbe_mac_82598EB)
4192 * ixgbe_sfp_link_config - set up SFP+ link
4193 * @adapter: pointer to private adapter struct
4195 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
4198 * We are assuming the worst case scenario here, and that
4199 * is that an SFP was inserted/removed after the reset
4200 * but before SFP detection was enabled. As such the best
4201 * solution is to just start searching as soon as we start
4203 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
4204 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
4206 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
4210 * ixgbe_non_sfp_link_config - set up non-SFP+ link
4211 * @hw: pointer to private hardware struct
4213 * Returns 0 on success, negative on failure
4215 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
4218 bool autoneg, link_up = false;
4219 u32 ret = IXGBE_ERR_LINK_SETUP;
4221 if (hw->mac.ops.check_link)
4222 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
4227 speed = hw->phy.autoneg_advertised;
4228 if ((!speed) && (hw->mac.ops.get_link_capabilities))
4229 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
4234 if (hw->mac.ops.setup_link)
4235 ret = hw->mac.ops.setup_link(hw, speed, link_up);
4240 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
4242 struct ixgbe_hw *hw = &adapter->hw;
4245 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4246 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
4248 gpie |= IXGBE_GPIE_EIAME;
4250 * use EIAM to auto-mask when MSI-X interrupt is asserted
4251 * this saves a register write for every interrupt
4253 switch (hw->mac.type) {
4254 case ixgbe_mac_82598EB:
4255 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4257 case ixgbe_mac_82599EB:
4258 case ixgbe_mac_X540:
4260 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4261 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4265 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
4266 * specifically only auto mask tx and rx interrupts */
4267 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4270 /* XXX: to interrupt immediately for EICS writes, enable this */
4271 /* gpie |= IXGBE_GPIE_EIMEN; */
4273 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
4274 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
4276 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4277 case IXGBE_82599_VMDQ_8Q_MASK:
4278 gpie |= IXGBE_GPIE_VTMODE_16;
4280 case IXGBE_82599_VMDQ_4Q_MASK:
4281 gpie |= IXGBE_GPIE_VTMODE_32;
4284 gpie |= IXGBE_GPIE_VTMODE_64;
4289 /* Enable Thermal over heat sensor interrupt */
4290 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
4291 switch (adapter->hw.mac.type) {
4292 case ixgbe_mac_82599EB:
4293 gpie |= IXGBE_SDP0_GPIEN;
4295 case ixgbe_mac_X540:
4296 gpie |= IXGBE_EIMS_TS;
4303 /* Enable fan failure interrupt */
4304 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
4305 gpie |= IXGBE_SDP1_GPIEN;
4307 if (hw->mac.type == ixgbe_mac_82599EB) {
4308 gpie |= IXGBE_SDP1_GPIEN;
4309 gpie |= IXGBE_SDP2_GPIEN;
4312 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4315 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
4317 struct ixgbe_hw *hw = &adapter->hw;
4321 ixgbe_get_hw_control(adapter);
4322 ixgbe_setup_gpie(adapter);
4324 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4325 ixgbe_configure_msix(adapter);
4327 ixgbe_configure_msi_and_legacy(adapter);
4329 /* enable the optics for 82599 SFP+ fiber */
4330 if (hw->mac.ops.enable_tx_laser)
4331 hw->mac.ops.enable_tx_laser(hw);
4333 clear_bit(__IXGBE_DOWN, &adapter->state);
4334 ixgbe_napi_enable_all(adapter);
4336 if (ixgbe_is_sfp(hw)) {
4337 ixgbe_sfp_link_config(adapter);
4339 err = ixgbe_non_sfp_link_config(hw);
4341 e_err(probe, "link_config FAILED %d\n", err);
4344 /* clear any pending interrupts, may auto mask */
4345 IXGBE_READ_REG(hw, IXGBE_EICR);
4346 ixgbe_irq_enable(adapter, true, true);
4349 * If this adapter has a fan, check to see if we had a failure
4350 * before we enabled the interrupt.
4352 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4353 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4354 if (esdp & IXGBE_ESDP_SDP1)
4355 e_crit(drv, "Fan has stopped, replace the adapter\n");
4358 /* enable transmits */
4359 netif_tx_start_all_queues(adapter->netdev);
4361 /* bring the link up in the watchdog, this could race with our first
4362 * link up interrupt but shouldn't be a problem */
4363 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4364 adapter->link_check_timeout = jiffies;
4365 mod_timer(&adapter->service_timer, jiffies);
4367 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4368 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4369 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4370 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4373 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4375 WARN_ON(in_interrupt());
4376 /* put off any impending NetWatchDogTimeout */
4377 adapter->netdev->trans_start = jiffies;
4379 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4380 usleep_range(1000, 2000);
4381 ixgbe_down(adapter);
4383 * If SR-IOV enabled then wait a bit before bringing the adapter
4384 * back up to give the VFs time to respond to the reset. The
4385 * two second wait is based upon the watchdog timer cycle in
4388 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4391 clear_bit(__IXGBE_RESETTING, &adapter->state);
4394 void ixgbe_up(struct ixgbe_adapter *adapter)
4396 /* hardware has been reset, we need to reload some things */
4397 ixgbe_configure(adapter);
4399 ixgbe_up_complete(adapter);
4402 void ixgbe_reset(struct ixgbe_adapter *adapter)
4404 struct ixgbe_hw *hw = &adapter->hw;
4407 /* lock SFP init bit to prevent race conditions with the watchdog */
4408 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4409 usleep_range(1000, 2000);
4411 /* clear all SFP and link config related flags while holding SFP_INIT */
4412 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4413 IXGBE_FLAG2_SFP_NEEDS_RESET);
4414 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4416 err = hw->mac.ops.init_hw(hw);
4419 case IXGBE_ERR_SFP_NOT_PRESENT:
4420 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4422 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4423 e_dev_err("master disable timed out\n");
4425 case IXGBE_ERR_EEPROM_VERSION:
4426 /* We are running on a pre-production device, log a warning */
4427 e_dev_warn("This device is a pre-production adapter/LOM. "
4428 "Please be aware there may be issues associated with "
4429 "your hardware. If you are experiencing problems "
4430 "please contact your Intel or hardware "
4431 "representative who provided you with this "
4435 e_dev_err("Hardware Error: %d\n", err);
4438 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4440 /* reprogram the RAR[0] in case user changed it. */
4441 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
4443 /* update SAN MAC vmdq pool selection */
4444 if (hw->mac.san_mac_rar_index)
4445 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
4447 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
4448 ixgbe_ptp_reset(adapter);
4452 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4453 * @rx_ring: ring to free buffers from
4455 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4457 struct device *dev = rx_ring->dev;
4461 /* ring already cleared, nothing to do */
4462 if (!rx_ring->rx_buffer_info)
4465 /* Free all the Rx ring sk_buffs */
4466 for (i = 0; i < rx_ring->count; i++) {
4467 struct ixgbe_rx_buffer *rx_buffer;
4469 rx_buffer = &rx_ring->rx_buffer_info[i];
4470 if (rx_buffer->skb) {
4471 struct sk_buff *skb = rx_buffer->skb;
4472 if (IXGBE_CB(skb)->page_released) {
4475 ixgbe_rx_bufsz(rx_ring),
4477 IXGBE_CB(skb)->page_released = false;
4481 rx_buffer->skb = NULL;
4483 dma_unmap_page(dev, rx_buffer->dma,
4484 ixgbe_rx_pg_size(rx_ring),
4487 if (rx_buffer->page)
4488 __free_pages(rx_buffer->page,
4489 ixgbe_rx_pg_order(rx_ring));
4490 rx_buffer->page = NULL;
4493 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4494 memset(rx_ring->rx_buffer_info, 0, size);
4496 /* Zero out the descriptor ring */
4497 memset(rx_ring->desc, 0, rx_ring->size);
4499 rx_ring->next_to_alloc = 0;
4500 rx_ring->next_to_clean = 0;
4501 rx_ring->next_to_use = 0;
4505 * ixgbe_clean_tx_ring - Free Tx Buffers
4506 * @tx_ring: ring to be cleaned
4508 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4510 struct ixgbe_tx_buffer *tx_buffer_info;
4514 /* ring already cleared, nothing to do */
4515 if (!tx_ring->tx_buffer_info)
4518 /* Free all the Tx ring sk_buffs */
4519 for (i = 0; i < tx_ring->count; i++) {
4520 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4521 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4524 netdev_tx_reset_queue(txring_txq(tx_ring));
4526 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4527 memset(tx_ring->tx_buffer_info, 0, size);
4529 /* Zero out the descriptor ring */
4530 memset(tx_ring->desc, 0, tx_ring->size);
4532 tx_ring->next_to_use = 0;
4533 tx_ring->next_to_clean = 0;
4537 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4538 * @adapter: board private structure
4540 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4544 for (i = 0; i < adapter->num_rx_queues; i++)
4545 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4549 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4550 * @adapter: board private structure
4552 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4556 for (i = 0; i < adapter->num_tx_queues; i++)
4557 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4560 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4562 struct hlist_node *node2;
4563 struct ixgbe_fdir_filter *filter;
4565 spin_lock(&adapter->fdir_perfect_lock);
4567 hlist_for_each_entry_safe(filter, node2,
4568 &adapter->fdir_filter_list, fdir_node) {
4569 hlist_del(&filter->fdir_node);
4572 adapter->fdir_filter_count = 0;
4574 spin_unlock(&adapter->fdir_perfect_lock);
4577 void ixgbe_down(struct ixgbe_adapter *adapter)
4579 struct net_device *netdev = adapter->netdev;
4580 struct ixgbe_hw *hw = &adapter->hw;
4584 /* signal that we are down to the interrupt handler */
4585 set_bit(__IXGBE_DOWN, &adapter->state);
4587 /* disable receives */
4588 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4589 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4591 /* disable all enabled rx queues */
4592 for (i = 0; i < adapter->num_rx_queues; i++)
4593 /* this call also flushes the previous write */
4594 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4596 usleep_range(10000, 20000);
4598 netif_tx_stop_all_queues(netdev);
4600 /* call carrier off first to avoid false dev_watchdog timeouts */
4601 netif_carrier_off(netdev);
4602 netif_tx_disable(netdev);
4604 ixgbe_irq_disable(adapter);
4606 ixgbe_napi_disable_all(adapter);
4608 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4609 IXGBE_FLAG2_RESET_REQUESTED);
4610 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4612 del_timer_sync(&adapter->service_timer);
4614 if (adapter->num_vfs) {
4615 /* Clear EITR Select mapping */
4616 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4618 /* Mark all the VFs as inactive */
4619 for (i = 0 ; i < adapter->num_vfs; i++)
4620 adapter->vfinfo[i].clear_to_send = false;
4622 /* ping all the active vfs to let them know we are going down */
4623 ixgbe_ping_all_vfs(adapter);
4625 /* Disable all VFTE/VFRE TX/RX */
4626 ixgbe_disable_tx_rx(adapter);
4629 /* disable transmits in the hardware now that interrupts are off */
4630 for (i = 0; i < adapter->num_tx_queues; i++) {
4631 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4632 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4635 /* Disable the Tx DMA engine on 82599 and X540 */
4636 switch (hw->mac.type) {
4637 case ixgbe_mac_82599EB:
4638 case ixgbe_mac_X540:
4639 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4640 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4641 ~IXGBE_DMATXCTL_TE));
4647 if (!pci_channel_offline(adapter->pdev))
4648 ixgbe_reset(adapter);
4650 /* power down the optics for 82599 SFP+ fiber */
4651 if (hw->mac.ops.disable_tx_laser)
4652 hw->mac.ops.disable_tx_laser(hw);
4654 ixgbe_clean_all_tx_rings(adapter);
4655 ixgbe_clean_all_rx_rings(adapter);
4657 #ifdef CONFIG_IXGBE_DCA
4658 /* since we reset the hardware DCA settings were cleared */
4659 ixgbe_setup_dca(adapter);
4664 * ixgbe_tx_timeout - Respond to a Tx Hang
4665 * @netdev: network interface device structure
4667 static void ixgbe_tx_timeout(struct net_device *netdev)
4669 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4671 /* Do the reset outside of interrupt context */
4672 ixgbe_tx_timeout_reset(adapter);
4676 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4677 * @adapter: board private structure to initialize
4679 * ixgbe_sw_init initializes the Adapter private data structure.
4680 * Fields are initialized based on PCI device information and
4681 * OS network device settings (MTU size).
4683 static int ixgbe_sw_init(struct ixgbe_adapter *adapter)
4685 struct ixgbe_hw *hw = &adapter->hw;
4686 struct pci_dev *pdev = adapter->pdev;
4687 unsigned int rss, fdir;
4689 #ifdef CONFIG_IXGBE_DCB
4691 struct tc_configuration *tc;
4694 /* PCI config space info */
4696 hw->vendor_id = pdev->vendor;
4697 hw->device_id = pdev->device;
4698 hw->revision_id = pdev->revision;
4699 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4700 hw->subsystem_device_id = pdev->subsystem_device;
4702 /* Set common capability flags and settings */
4703 rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4704 adapter->ring_feature[RING_F_RSS].limit = rss;
4705 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4706 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4707 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
4708 adapter->atr_sample_rate = 20;
4709 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
4710 adapter->ring_feature[RING_F_FDIR].limit = fdir;
4711 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4712 #ifdef CONFIG_IXGBE_DCA
4713 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
4716 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4717 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4718 #ifdef CONFIG_IXGBE_DCB
4719 /* Default traffic class to use for FCoE */
4720 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4721 #endif /* CONFIG_IXGBE_DCB */
4722 #endif /* IXGBE_FCOE */
4724 /* Set MAC specific capability flags and exceptions */
4725 switch (hw->mac.type) {
4726 case ixgbe_mac_82598EB:
4727 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
4728 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
4730 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4731 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4733 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
4734 adapter->ring_feature[RING_F_FDIR].limit = 0;
4735 adapter->atr_sample_rate = 0;
4736 adapter->fdir_pballoc = 0;
4738 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
4739 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4740 #ifdef CONFIG_IXGBE_DCB
4741 adapter->fcoe.up = 0;
4742 #endif /* IXGBE_DCB */
4743 #endif /* IXGBE_FCOE */
4745 case ixgbe_mac_82599EB:
4746 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4747 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4749 case ixgbe_mac_X540:
4750 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM);
4751 if (fwsm & IXGBE_FWSM_TS_ENABLED)
4752 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4759 /* FCoE support exists, always init the FCoE lock */
4760 spin_lock_init(&adapter->fcoe.lock);
4763 /* n-tuple support exists, always init our spinlock */
4764 spin_lock_init(&adapter->fdir_perfect_lock);
4766 #ifdef CONFIG_IXGBE_DCB
4767 switch (hw->mac.type) {
4768 case ixgbe_mac_X540:
4769 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4770 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4773 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4774 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4778 /* Configure DCB traffic classes */
4779 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4780 tc = &adapter->dcb_cfg.tc_config[j];
4781 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4782 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4783 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4784 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4785 tc->dcb_pfc = pfc_disabled;
4788 /* Initialize default user to priority mapping, UPx->TC0 */
4789 tc = &adapter->dcb_cfg.tc_config[0];
4790 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
4791 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
4793 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4794 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4795 adapter->dcb_cfg.pfc_mode_enable = false;
4796 adapter->dcb_set_bitmap = 0x00;
4797 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4798 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
4799 sizeof(adapter->temp_dcb_cfg));
4803 /* default flow control settings */
4804 hw->fc.requested_mode = ixgbe_fc_full;
4805 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4806 ixgbe_pbthresh_setup(adapter);
4807 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4808 hw->fc.send_xon = true;
4809 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
4811 #ifdef CONFIG_PCI_IOV
4812 /* assign number of SR-IOV VFs */
4813 if (hw->mac.type != ixgbe_mac_82598EB)
4814 adapter->num_vfs = (max_vfs > 63) ? 0 : max_vfs;
4817 /* enable itr by default in dynamic mode */
4818 adapter->rx_itr_setting = 1;
4819 adapter->tx_itr_setting = 1;
4821 /* set default ring sizes */
4822 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4823 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4825 /* set default work limits */
4826 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4828 /* initialize eeprom parameters */
4829 if (ixgbe_init_eeprom_params_generic(hw)) {
4830 e_dev_err("EEPROM initialization failed\n");
4834 set_bit(__IXGBE_DOWN, &adapter->state);
4840 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4841 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4843 * Return 0 on success, negative on failure
4845 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4847 struct device *dev = tx_ring->dev;
4848 int orig_node = dev_to_node(dev);
4852 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4854 if (tx_ring->q_vector)
4855 numa_node = tx_ring->q_vector->numa_node;
4857 tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
4858 if (!tx_ring->tx_buffer_info)
4859 tx_ring->tx_buffer_info = vzalloc(size);
4860 if (!tx_ring->tx_buffer_info)
4863 /* round up to nearest 4K */
4864 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4865 tx_ring->size = ALIGN(tx_ring->size, 4096);
4867 set_dev_node(dev, numa_node);
4868 tx_ring->desc = dma_alloc_coherent(dev,
4872 set_dev_node(dev, orig_node);
4874 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4875 &tx_ring->dma, GFP_KERNEL);
4879 tx_ring->next_to_use = 0;
4880 tx_ring->next_to_clean = 0;
4884 vfree(tx_ring->tx_buffer_info);
4885 tx_ring->tx_buffer_info = NULL;
4886 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4891 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4892 * @adapter: board private structure
4894 * If this function returns with an error, then it's possible one or
4895 * more of the rings is populated (while the rest are not). It is the
4896 * callers duty to clean those orphaned rings.
4898 * Return 0 on success, negative on failure
4900 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4904 for (i = 0; i < adapter->num_tx_queues; i++) {
4905 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4909 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4915 /* rewind the index freeing the rings as we go */
4917 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4922 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4923 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4925 * Returns 0 on success, negative on failure
4927 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4929 struct device *dev = rx_ring->dev;
4930 int orig_node = dev_to_node(dev);
4934 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4936 if (rx_ring->q_vector)
4937 numa_node = rx_ring->q_vector->numa_node;
4939 rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
4940 if (!rx_ring->rx_buffer_info)
4941 rx_ring->rx_buffer_info = vzalloc(size);
4942 if (!rx_ring->rx_buffer_info)
4945 /* Round up to nearest 4K */
4946 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4947 rx_ring->size = ALIGN(rx_ring->size, 4096);
4949 set_dev_node(dev, numa_node);
4950 rx_ring->desc = dma_alloc_coherent(dev,
4954 set_dev_node(dev, orig_node);
4956 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4957 &rx_ring->dma, GFP_KERNEL);
4961 rx_ring->next_to_clean = 0;
4962 rx_ring->next_to_use = 0;
4966 vfree(rx_ring->rx_buffer_info);
4967 rx_ring->rx_buffer_info = NULL;
4968 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4973 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4974 * @adapter: board private structure
4976 * If this function returns with an error, then it's possible one or
4977 * more of the rings is populated (while the rest are not). It is the
4978 * callers duty to clean those orphaned rings.
4980 * Return 0 on success, negative on failure
4982 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4986 for (i = 0; i < adapter->num_rx_queues; i++) {
4987 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
4991 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4996 err = ixgbe_setup_fcoe_ddp_resources(adapter);
5001 /* rewind the index freeing the rings as we go */
5003 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5008 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5009 * @tx_ring: Tx descriptor ring for a specific queue
5011 * Free all transmit software resources
5013 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5015 ixgbe_clean_tx_ring(tx_ring);
5017 vfree(tx_ring->tx_buffer_info);
5018 tx_ring->tx_buffer_info = NULL;
5020 /* if not set, then don't free */
5024 dma_free_coherent(tx_ring->dev, tx_ring->size,
5025 tx_ring->desc, tx_ring->dma);
5027 tx_ring->desc = NULL;
5031 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5032 * @adapter: board private structure
5034 * Free all transmit software resources
5036 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5040 for (i = 0; i < adapter->num_tx_queues; i++)
5041 if (adapter->tx_ring[i]->desc)
5042 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5046 * ixgbe_free_rx_resources - Free Rx Resources
5047 * @rx_ring: ring to clean the resources from
5049 * Free all receive software resources
5051 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5053 ixgbe_clean_rx_ring(rx_ring);
5055 vfree(rx_ring->rx_buffer_info);
5056 rx_ring->rx_buffer_info = NULL;
5058 /* if not set, then don't free */
5062 dma_free_coherent(rx_ring->dev, rx_ring->size,
5063 rx_ring->desc, rx_ring->dma);
5065 rx_ring->desc = NULL;
5069 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5070 * @adapter: board private structure
5072 * Free all receive software resources
5074 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5079 ixgbe_free_fcoe_ddp_resources(adapter);
5082 for (i = 0; i < adapter->num_rx_queues; i++)
5083 if (adapter->rx_ring[i]->desc)
5084 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5088 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5089 * @netdev: network interface device structure
5090 * @new_mtu: new value for maximum frame size
5092 * Returns 0 on success, negative on failure
5094 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5096 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5097 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5099 /* MTU < 68 is an error and causes problems on some kernels */
5100 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5104 * For 82599EB we cannot allow legacy VFs to enable their receive
5105 * paths when MTU greater than 1500 is configured. So display a
5106 * warning that legacy VFs will be disabled.
5108 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
5109 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
5110 (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
5111 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
5113 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5115 /* must set new MTU before calling down or up */
5116 netdev->mtu = new_mtu;
5118 if (netif_running(netdev))
5119 ixgbe_reinit_locked(adapter);
5125 * ixgbe_open - Called when a network interface is made active
5126 * @netdev: network interface device structure
5128 * Returns 0 on success, negative value on failure
5130 * The open entry point is called when a network interface is made
5131 * active by the system (IFF_UP). At this point all resources needed
5132 * for transmit and receive operations are allocated, the interrupt
5133 * handler is registered with the OS, the watchdog timer is started,
5134 * and the stack is notified that the interface is ready.
5136 static int ixgbe_open(struct net_device *netdev)
5138 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5141 /* disallow open during test */
5142 if (test_bit(__IXGBE_TESTING, &adapter->state))
5145 netif_carrier_off(netdev);
5147 /* allocate transmit descriptors */
5148 err = ixgbe_setup_all_tx_resources(adapter);
5152 /* allocate receive descriptors */
5153 err = ixgbe_setup_all_rx_resources(adapter);
5157 ixgbe_configure(adapter);
5159 err = ixgbe_request_irq(adapter);
5163 /* Notify the stack of the actual queue counts. */
5164 err = netif_set_real_num_tx_queues(netdev,
5165 adapter->num_rx_pools > 1 ? 1 :
5166 adapter->num_tx_queues);
5168 goto err_set_queues;
5171 err = netif_set_real_num_rx_queues(netdev,
5172 adapter->num_rx_pools > 1 ? 1 :
5173 adapter->num_rx_queues);
5175 goto err_set_queues;
5177 ixgbe_ptp_init(adapter);
5179 ixgbe_up_complete(adapter);
5184 ixgbe_free_irq(adapter);
5186 ixgbe_free_all_rx_resources(adapter);
5188 ixgbe_free_all_tx_resources(adapter);
5190 ixgbe_reset(adapter);
5196 * ixgbe_close - Disables a network interface
5197 * @netdev: network interface device structure
5199 * Returns 0, this is not allowed to fail
5201 * The close entry point is called when an interface is de-activated
5202 * by the OS. The hardware is still under the drivers control, but
5203 * needs to be disabled. A global MAC reset is issued to stop the
5204 * hardware, and all transmit and receive resources are freed.
5206 static int ixgbe_close(struct net_device *netdev)
5208 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5210 ixgbe_ptp_stop(adapter);
5212 ixgbe_down(adapter);
5213 ixgbe_free_irq(adapter);
5215 ixgbe_fdir_filter_exit(adapter);
5217 ixgbe_free_all_tx_resources(adapter);
5218 ixgbe_free_all_rx_resources(adapter);
5220 ixgbe_release_hw_control(adapter);
5226 static int ixgbe_resume(struct pci_dev *pdev)
5228 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5229 struct net_device *netdev = adapter->netdev;
5232 pci_set_power_state(pdev, PCI_D0);
5233 pci_restore_state(pdev);
5235 * pci_restore_state clears dev->state_saved so call
5236 * pci_save_state to restore it.
5238 pci_save_state(pdev);
5240 err = pci_enable_device_mem(pdev);
5242 e_dev_err("Cannot enable PCI device from suspend\n");
5245 pci_set_master(pdev);
5247 pci_wake_from_d3(pdev, false);
5249 ixgbe_reset(adapter);
5251 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5254 err = ixgbe_init_interrupt_scheme(adapter);
5255 if (!err && netif_running(netdev))
5256 err = ixgbe_open(netdev);
5263 netif_device_attach(netdev);
5267 #endif /* CONFIG_PM */
5269 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5271 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5272 struct net_device *netdev = adapter->netdev;
5273 struct ixgbe_hw *hw = &adapter->hw;
5275 u32 wufc = adapter->wol;
5280 netif_device_detach(netdev);
5283 if (netif_running(netdev)) {
5284 ixgbe_down(adapter);
5285 ixgbe_free_irq(adapter);
5286 ixgbe_free_all_tx_resources(adapter);
5287 ixgbe_free_all_rx_resources(adapter);
5291 ixgbe_clear_interrupt_scheme(adapter);
5294 retval = pci_save_state(pdev);
5299 if (hw->mac.ops.stop_link_on_d3)
5300 hw->mac.ops.stop_link_on_d3(hw);
5303 ixgbe_set_rx_mode(netdev);
5305 /* enable the optics for 82599 SFP+ fiber as we can WoL */
5306 if (hw->mac.ops.enable_tx_laser)
5307 hw->mac.ops.enable_tx_laser(hw);
5309 /* turn on all-multi mode if wake on multicast is enabled */
5310 if (wufc & IXGBE_WUFC_MC) {
5311 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5312 fctrl |= IXGBE_FCTRL_MPE;
5313 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5316 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5317 ctrl |= IXGBE_CTRL_GIO_DIS;
5318 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5320 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5322 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5323 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5326 switch (hw->mac.type) {
5327 case ixgbe_mac_82598EB:
5328 pci_wake_from_d3(pdev, false);
5330 case ixgbe_mac_82599EB:
5331 case ixgbe_mac_X540:
5332 pci_wake_from_d3(pdev, !!wufc);
5338 *enable_wake = !!wufc;
5340 ixgbe_release_hw_control(adapter);
5342 pci_disable_device(pdev);
5348 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5353 retval = __ixgbe_shutdown(pdev, &wake);
5358 pci_prepare_to_sleep(pdev);
5360 pci_wake_from_d3(pdev, false);
5361 pci_set_power_state(pdev, PCI_D3hot);
5366 #endif /* CONFIG_PM */
5368 static void ixgbe_shutdown(struct pci_dev *pdev)
5372 __ixgbe_shutdown(pdev, &wake);
5374 if (system_state == SYSTEM_POWER_OFF) {
5375 pci_wake_from_d3(pdev, wake);
5376 pci_set_power_state(pdev, PCI_D3hot);
5381 * ixgbe_update_stats - Update the board statistics counters.
5382 * @adapter: board private structure
5384 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5386 struct net_device *netdev = adapter->netdev;
5387 struct ixgbe_hw *hw = &adapter->hw;
5388 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5390 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5391 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5392 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5393 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5395 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5396 test_bit(__IXGBE_RESETTING, &adapter->state))
5399 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5402 for (i = 0; i < adapter->num_rx_queues; i++) {
5403 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5404 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5406 adapter->rsc_total_count = rsc_count;
5407 adapter->rsc_total_flush = rsc_flush;
5410 for (i = 0; i < adapter->num_rx_queues; i++) {
5411 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5412 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5413 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5414 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5415 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5416 bytes += rx_ring->stats.bytes;
5417 packets += rx_ring->stats.packets;
5419 adapter->non_eop_descs = non_eop_descs;
5420 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5421 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5422 adapter->hw_csum_rx_error = hw_csum_rx_error;
5423 netdev->stats.rx_bytes = bytes;
5424 netdev->stats.rx_packets = packets;
5428 /* gather some stats to the adapter struct that are per queue */
5429 for (i = 0; i < adapter->num_tx_queues; i++) {
5430 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5431 restart_queue += tx_ring->tx_stats.restart_queue;
5432 tx_busy += tx_ring->tx_stats.tx_busy;
5433 bytes += tx_ring->stats.bytes;
5434 packets += tx_ring->stats.packets;
5436 adapter->restart_queue = restart_queue;
5437 adapter->tx_busy = tx_busy;
5438 netdev->stats.tx_bytes = bytes;
5439 netdev->stats.tx_packets = packets;
5441 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5443 /* 8 register reads */
5444 for (i = 0; i < 8; i++) {
5445 /* for packet buffers not used, the register should read 0 */
5446 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5448 hwstats->mpc[i] += mpc;
5449 total_mpc += hwstats->mpc[i];
5450 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5451 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5452 switch (hw->mac.type) {
5453 case ixgbe_mac_82598EB:
5454 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5455 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5456 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5457 hwstats->pxonrxc[i] +=
5458 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5460 case ixgbe_mac_82599EB:
5461 case ixgbe_mac_X540:
5462 hwstats->pxonrxc[i] +=
5463 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5470 /*16 register reads */
5471 for (i = 0; i < 16; i++) {
5472 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5473 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5474 if ((hw->mac.type == ixgbe_mac_82599EB) ||
5475 (hw->mac.type == ixgbe_mac_X540)) {
5476 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5477 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5478 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5479 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5483 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5484 /* work around hardware counting issue */
5485 hwstats->gprc -= missed_rx;
5487 ixgbe_update_xoff_received(adapter);
5489 /* 82598 hardware only has a 32 bit counter in the high register */
5490 switch (hw->mac.type) {
5491 case ixgbe_mac_82598EB:
5492 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5493 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5494 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5495 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5497 case ixgbe_mac_X540:
5498 /* OS2BMC stats are X540 only*/
5499 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5500 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5501 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5502 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5503 case ixgbe_mac_82599EB:
5504 for (i = 0; i < 16; i++)
5505 adapter->hw_rx_no_dma_resources +=
5506 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5507 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5508 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5509 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5510 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5511 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5512 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5513 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5514 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5515 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5517 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5518 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5519 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5520 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5521 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5522 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5523 /* Add up per cpu counters for total ddp aloc fail */
5524 if (adapter->fcoe.ddp_pool) {
5525 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5526 struct ixgbe_fcoe_ddp_pool *ddp_pool;
5528 u64 noddp = 0, noddp_ext_buff = 0;
5529 for_each_possible_cpu(cpu) {
5530 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
5531 noddp += ddp_pool->noddp;
5532 noddp_ext_buff += ddp_pool->noddp_ext_buff;
5534 hwstats->fcoe_noddp = noddp;
5535 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
5537 #endif /* IXGBE_FCOE */
5542 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5543 hwstats->bprc += bprc;
5544 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5545 if (hw->mac.type == ixgbe_mac_82598EB)
5546 hwstats->mprc -= bprc;
5547 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5548 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5549 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5550 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5551 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5552 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5553 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5554 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5555 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5556 hwstats->lxontxc += lxon;
5557 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5558 hwstats->lxofftxc += lxoff;
5559 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5560 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5562 * 82598 errata - tx of flow control packets is included in tx counters
5564 xon_off_tot = lxon + lxoff;
5565 hwstats->gptc -= xon_off_tot;
5566 hwstats->mptc -= xon_off_tot;
5567 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5568 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5569 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5570 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5571 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5572 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5573 hwstats->ptc64 -= xon_off_tot;
5574 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5575 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5576 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5577 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5578 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5579 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5581 /* Fill out the OS statistics structure */
5582 netdev->stats.multicast = hwstats->mprc;
5585 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5586 netdev->stats.rx_dropped = 0;
5587 netdev->stats.rx_length_errors = hwstats->rlec;
5588 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5589 netdev->stats.rx_missed_errors = total_mpc;
5593 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5594 * @adapter: pointer to the device adapter structure
5596 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5598 struct ixgbe_hw *hw = &adapter->hw;
5601 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5604 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5606 /* if interface is down do nothing */
5607 if (test_bit(__IXGBE_DOWN, &adapter->state))
5610 /* do nothing if we are not using signature filters */
5611 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5614 adapter->fdir_overflow++;
5616 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5617 for (i = 0; i < adapter->num_tx_queues; i++)
5618 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5619 &(adapter->tx_ring[i]->state));
5620 /* re-enable flow director interrupts */
5621 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5623 e_err(probe, "failed to finish FDIR re-initialization, "
5624 "ignored adding FDIR ATR filters\n");
5629 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5630 * @adapter: pointer to the device adapter structure
5632 * This function serves two purposes. First it strobes the interrupt lines
5633 * in order to make certain interrupts are occurring. Secondly it sets the
5634 * bits needed to check for TX hangs. As a result we should immediately
5635 * determine if a hang has occurred.
5637 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5639 struct ixgbe_hw *hw = &adapter->hw;
5643 /* If we're down or resetting, just bail */
5644 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5645 test_bit(__IXGBE_RESETTING, &adapter->state))
5648 /* Force detection of hung controller */
5649 if (netif_carrier_ok(adapter->netdev)) {
5650 for (i = 0; i < adapter->num_tx_queues; i++)
5651 set_check_for_tx_hang(adapter->tx_ring[i]);
5654 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5656 * for legacy and MSI interrupts don't set any bits
5657 * that are enabled for EIAM, because this operation
5658 * would set *both* EIMS and EICS for any bit in EIAM
5660 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5661 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5663 /* get one bit for every active tx/rx interrupt vector */
5664 for (i = 0; i < adapter->num_q_vectors; i++) {
5665 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5666 if (qv->rx.ring || qv->tx.ring)
5667 eics |= ((u64)1 << i);
5671 /* Cause software interrupt to ensure rings are cleaned */
5672 ixgbe_irq_rearm_queues(adapter, eics);
5677 * ixgbe_watchdog_update_link - update the link status
5678 * @adapter: pointer to the device adapter structure
5679 * @link_speed: pointer to a u32 to store the link_speed
5681 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5683 struct ixgbe_hw *hw = &adapter->hw;
5684 u32 link_speed = adapter->link_speed;
5685 bool link_up = adapter->link_up;
5686 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
5688 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5691 if (hw->mac.ops.check_link) {
5692 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5694 /* always assume link is up, if no check link function */
5695 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5699 if (adapter->ixgbe_ieee_pfc)
5700 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
5702 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
5703 hw->mac.ops.fc_enable(hw);
5704 ixgbe_set_rx_drop_en(adapter);
5708 time_after(jiffies, (adapter->link_check_timeout +
5709 IXGBE_TRY_LINK_TIMEOUT))) {
5710 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5711 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5712 IXGBE_WRITE_FLUSH(hw);
5715 adapter->link_up = link_up;
5716 adapter->link_speed = link_speed;
5719 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
5721 #ifdef CONFIG_IXGBE_DCB
5722 struct net_device *netdev = adapter->netdev;
5723 struct dcb_app app = {
5724 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
5729 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
5730 up = dcb_ieee_getapp_mask(netdev, &app);
5732 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
5737 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5738 * print link up message
5739 * @adapter: pointer to the device adapter structure
5741 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5743 struct net_device *netdev = adapter->netdev;
5744 struct ixgbe_hw *hw = &adapter->hw;
5745 u32 link_speed = adapter->link_speed;
5746 bool flow_rx, flow_tx;
5748 /* only continue if link was previously down */
5749 if (netif_carrier_ok(netdev))
5752 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5754 switch (hw->mac.type) {
5755 case ixgbe_mac_82598EB: {
5756 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5757 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5758 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5759 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5762 case ixgbe_mac_X540:
5763 case ixgbe_mac_82599EB: {
5764 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5765 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5766 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5767 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5776 adapter->last_rx_ptp_check = jiffies;
5778 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5779 ixgbe_ptp_start_cyclecounter(adapter);
5781 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5782 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5784 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5786 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5789 ((flow_rx && flow_tx) ? "RX/TX" :
5791 (flow_tx ? "TX" : "None"))));
5793 netif_carrier_on(netdev);
5794 ixgbe_check_vf_rate_limit(adapter);
5796 /* update the default user priority for VFs */
5797 ixgbe_update_default_up(adapter);
5799 /* ping all the active vfs to let them know link has changed */
5800 ixgbe_ping_all_vfs(adapter);
5804 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5805 * print link down message
5806 * @adapter: pointer to the adapter structure
5808 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
5810 struct net_device *netdev = adapter->netdev;
5811 struct ixgbe_hw *hw = &adapter->hw;
5813 adapter->link_up = false;
5814 adapter->link_speed = 0;
5816 /* only continue if link was up previously */
5817 if (!netif_carrier_ok(netdev))
5820 /* poll for SFP+ cable when link is down */
5821 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5822 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5824 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5825 ixgbe_ptp_start_cyclecounter(adapter);
5827 e_info(drv, "NIC Link is Down\n");
5828 netif_carrier_off(netdev);
5830 /* ping all the active vfs to let them know link has changed */
5831 ixgbe_ping_all_vfs(adapter);
5835 * ixgbe_watchdog_flush_tx - flush queues on link down
5836 * @adapter: pointer to the device adapter structure
5838 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5841 int some_tx_pending = 0;
5843 if (!netif_carrier_ok(adapter->netdev)) {
5844 for (i = 0; i < adapter->num_tx_queues; i++) {
5845 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5846 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5847 some_tx_pending = 1;
5852 if (some_tx_pending) {
5853 /* We've lost link, so the controller stops DMA,
5854 * but we've got queued Tx work that's never going
5855 * to get done, so reset controller to flush Tx.
5856 * (Do the reset outside of interrupt context).
5858 e_warn(drv, "initiating reset to clear Tx work after link loss\n");
5859 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5864 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5868 /* Do not perform spoof check for 82598 or if not in IOV mode */
5869 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
5870 adapter->num_vfs == 0)
5873 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5876 * ssvpc register is cleared on read, if zero then no
5877 * spoofed packets in the last interval.
5882 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
5886 * ixgbe_watchdog_subtask - check and bring link up
5887 * @adapter: pointer to the device adapter structure
5889 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5891 /* if interface is down do nothing */
5892 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5893 test_bit(__IXGBE_RESETTING, &adapter->state))
5896 ixgbe_watchdog_update_link(adapter);
5898 if (adapter->link_up)
5899 ixgbe_watchdog_link_is_up(adapter);
5901 ixgbe_watchdog_link_is_down(adapter);
5903 ixgbe_spoof_check(adapter);
5904 ixgbe_update_stats(adapter);
5906 ixgbe_watchdog_flush_tx(adapter);
5910 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5911 * @adapter: the ixgbe adapter structure
5913 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5915 struct ixgbe_hw *hw = &adapter->hw;
5918 /* not searching for SFP so there is nothing to do here */
5919 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5920 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5923 /* someone else is in init, wait until next service event */
5924 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5927 err = hw->phy.ops.identify_sfp(hw);
5928 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5931 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5932 /* If no cable is present, then we need to reset
5933 * the next time we find a good cable. */
5934 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5941 /* exit if reset not needed */
5942 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5945 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5948 * A module may be identified correctly, but the EEPROM may not have
5949 * support for that module. setup_sfp() will fail in that case, so
5950 * we should not allow that module to load.
5952 if (hw->mac.type == ixgbe_mac_82598EB)
5953 err = hw->phy.ops.reset(hw);
5955 err = hw->mac.ops.setup_sfp(hw);
5957 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5960 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5961 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5964 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5966 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5967 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5968 e_dev_err("failed to initialize because an unsupported "
5969 "SFP+ module type was detected.\n");
5970 e_dev_err("Reload the driver after installing a "
5971 "supported module.\n");
5972 unregister_netdev(adapter->netdev);
5977 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5978 * @adapter: the ixgbe adapter structure
5980 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5982 struct ixgbe_hw *hw = &adapter->hw;
5984 bool autoneg = false;
5986 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5989 /* someone else is in init, wait until next service event */
5990 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5993 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5995 speed = hw->phy.autoneg_advertised;
5996 if ((!speed) && (hw->mac.ops.get_link_capabilities))
5997 hw->mac.ops.get_link_capabilities(hw, &speed, &autoneg);
5998 if (hw->mac.ops.setup_link)
5999 hw->mac.ops.setup_link(hw, speed, true);
6001 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6002 adapter->link_check_timeout = jiffies;
6003 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6006 #ifdef CONFIG_PCI_IOV
6007 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6010 struct ixgbe_hw *hw = &adapter->hw;
6011 struct net_device *netdev = adapter->netdev;
6015 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6016 if (gpc) /* If incrementing then no need for the check below */
6019 * Check to see if a bad DMA write target from an errant or
6020 * malicious VF has caused a PCIe error. If so then we can
6021 * issue a VFLR to the offending VF(s) and then resume without
6022 * requesting a full slot reset.
6025 for (vf = 0; vf < adapter->num_vfs; vf++) {
6026 ciaa = (vf << 16) | 0x80000000;
6027 /* 32 bit read so align, we really want status at offset 6 */
6028 ciaa |= PCI_COMMAND;
6029 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6030 ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
6032 /* disable debug mode asap after reading data */
6033 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6034 /* Get the upper 16 bits which will be the PCI status reg */
6036 if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
6037 netdev_err(netdev, "VF %d Hung DMA\n", vf);
6039 ciaa = (vf << 16) | 0x80000000;
6041 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6042 ciad = 0x00008000; /* VFLR */
6043 IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
6045 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
6052 * ixgbe_service_timer - Timer Call-back
6053 * @data: pointer to adapter cast into an unsigned long
6055 static void ixgbe_service_timer(unsigned long data)
6057 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6058 unsigned long next_event_offset;
6061 /* poll faster when waiting for link */
6062 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6063 next_event_offset = HZ / 10;
6065 next_event_offset = HZ * 2;
6067 #ifdef CONFIG_PCI_IOV
6069 * don't bother with SR-IOV VF DMA hang check if there are
6070 * no VFs or the link is down
6072 if (!adapter->num_vfs ||
6073 (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6074 goto normal_timer_service;
6076 /* If we have VFs allocated then we must check for DMA hangs */
6077 ixgbe_check_for_bad_vf(adapter);
6078 next_event_offset = HZ / 50;
6079 adapter->timer_event_accumulator++;
6081 if (adapter->timer_event_accumulator >= 100)
6082 adapter->timer_event_accumulator = 0;
6086 normal_timer_service:
6088 /* Reset the timer */
6089 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6092 ixgbe_service_event_schedule(adapter);
6095 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6097 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6100 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6102 /* If we're already down or resetting, just bail */
6103 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6104 test_bit(__IXGBE_RESETTING, &adapter->state))
6107 ixgbe_dump(adapter);
6108 netdev_err(adapter->netdev, "Reset adapter\n");
6109 adapter->tx_timeout_count++;
6111 ixgbe_reinit_locked(adapter);
6115 * ixgbe_service_task - manages and runs subtasks
6116 * @work: pointer to work_struct containing our data
6118 static void ixgbe_service_task(struct work_struct *work)
6120 struct ixgbe_adapter *adapter = container_of(work,
6121 struct ixgbe_adapter,
6123 ixgbe_reset_subtask(adapter);
6124 ixgbe_sfp_detection_subtask(adapter);
6125 ixgbe_sfp_link_config_subtask(adapter);
6126 ixgbe_check_overtemp_subtask(adapter);
6127 ixgbe_watchdog_subtask(adapter);
6128 ixgbe_fdir_reinit_subtask(adapter);
6129 ixgbe_check_hang_subtask(adapter);
6131 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
6132 ixgbe_ptp_overflow_check(adapter);
6133 ixgbe_ptp_rx_hang(adapter);
6136 ixgbe_service_event_complete(adapter);
6139 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
6140 struct ixgbe_tx_buffer *first,
6143 struct sk_buff *skb = first->skb;
6144 u32 vlan_macip_lens, type_tucmd;
6145 u32 mss_l4len_idx, l4len;
6147 if (skb->ip_summed != CHECKSUM_PARTIAL)
6150 if (!skb_is_gso(skb))
6153 if (skb_header_cloned(skb)) {
6154 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6159 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6160 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6162 if (first->protocol == __constant_htons(ETH_P_IP)) {
6163 struct iphdr *iph = ip_hdr(skb);
6166 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6170 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6171 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6172 IXGBE_TX_FLAGS_CSUM |
6173 IXGBE_TX_FLAGS_IPV4;
6174 } else if (skb_is_gso_v6(skb)) {
6175 ipv6_hdr(skb)->payload_len = 0;
6176 tcp_hdr(skb)->check =
6177 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6178 &ipv6_hdr(skb)->daddr,
6180 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6181 IXGBE_TX_FLAGS_CSUM;
6184 /* compute header lengths */
6185 l4len = tcp_hdrlen(skb);
6186 *hdr_len = skb_transport_offset(skb) + l4len;
6188 /* update gso size and bytecount with header size */
6189 first->gso_segs = skb_shinfo(skb)->gso_segs;
6190 first->bytecount += (first->gso_segs - 1) * *hdr_len;
6192 /* mss_l4len_id: use 0 as index for TSO */
6193 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6194 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6196 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6197 vlan_macip_lens = skb_network_header_len(skb);
6198 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6199 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6201 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6207 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6208 struct ixgbe_tx_buffer *first)
6210 struct sk_buff *skb = first->skb;
6211 u32 vlan_macip_lens = 0;
6212 u32 mss_l4len_idx = 0;
6215 if (skb->ip_summed != CHECKSUM_PARTIAL) {
6216 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6217 !(first->tx_flags & IXGBE_TX_FLAGS_CC))
6221 switch (first->protocol) {
6222 case __constant_htons(ETH_P_IP):
6223 vlan_macip_lens |= skb_network_header_len(skb);
6224 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6225 l4_hdr = ip_hdr(skb)->protocol;
6227 case __constant_htons(ETH_P_IPV6):
6228 vlan_macip_lens |= skb_network_header_len(skb);
6229 l4_hdr = ipv6_hdr(skb)->nexthdr;
6232 if (unlikely(net_ratelimit())) {
6233 dev_warn(tx_ring->dev,
6234 "partial checksum but proto=%x!\n",
6242 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6243 mss_l4len_idx = tcp_hdrlen(skb) <<
6244 IXGBE_ADVTXD_L4LEN_SHIFT;
6247 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6248 mss_l4len_idx = sizeof(struct sctphdr) <<
6249 IXGBE_ADVTXD_L4LEN_SHIFT;
6252 mss_l4len_idx = sizeof(struct udphdr) <<
6253 IXGBE_ADVTXD_L4LEN_SHIFT;
6256 if (unlikely(net_ratelimit())) {
6257 dev_warn(tx_ring->dev,
6258 "partial checksum but l4 proto=%x!\n",
6264 /* update TX checksum flag */
6265 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
6268 /* vlan_macip_lens: MACLEN, VLAN tag */
6269 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6270 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6272 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6273 type_tucmd, mss_l4len_idx);
6276 #define IXGBE_SET_FLAG(_input, _flag, _result) \
6277 ((_flag <= _result) ? \
6278 ((u32)(_input & _flag) * (_result / _flag)) : \
6279 ((u32)(_input & _flag) / (_flag / _result)))
6281 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
6283 /* set type for advanced descriptor with frame checksum insertion */
6284 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
6285 IXGBE_ADVTXD_DCMD_DEXT |
6286 IXGBE_ADVTXD_DCMD_IFCS;
6288 /* set HW vlan bit if vlan is present */
6289 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
6290 IXGBE_ADVTXD_DCMD_VLE);
6292 /* set segmentation enable bits for TSO/FSO */
6293 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
6294 IXGBE_ADVTXD_DCMD_TSE);
6296 /* set timestamp bit if present */
6297 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
6298 IXGBE_ADVTXD_MAC_TSTAMP);
6300 /* insert frame checksum */
6301 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
6306 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
6307 u32 tx_flags, unsigned int paylen)
6309 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
6311 /* enable L4 checksum for TSO and TX checksum offload */
6312 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6313 IXGBE_TX_FLAGS_CSUM,
6314 IXGBE_ADVTXD_POPTS_TXSM);
6316 /* enble IPv4 checksum for TSO */
6317 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6318 IXGBE_TX_FLAGS_IPV4,
6319 IXGBE_ADVTXD_POPTS_IXSM);
6322 * Check Context must be set if Tx switch is enabled, which it
6323 * always is for case where virtual functions are running
6325 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6329 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6332 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6335 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
6336 struct ixgbe_tx_buffer *first,
6339 struct sk_buff *skb = first->skb;
6340 struct ixgbe_tx_buffer *tx_buffer;
6341 union ixgbe_adv_tx_desc *tx_desc;
6342 struct skb_frag_struct *frag;
6344 unsigned int data_len, size;
6345 u32 tx_flags = first->tx_flags;
6346 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
6347 u16 i = tx_ring->next_to_use;
6349 tx_desc = IXGBE_TX_DESC(tx_ring, i);
6351 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
6353 size = skb_headlen(skb);
6354 data_len = skb->data_len;
6357 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6358 if (data_len < sizeof(struct fcoe_crc_eof)) {
6359 size -= sizeof(struct fcoe_crc_eof) - data_len;
6362 data_len -= sizeof(struct fcoe_crc_eof);
6367 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6371 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
6372 if (dma_mapping_error(tx_ring->dev, dma))
6375 /* record length, and DMA address */
6376 dma_unmap_len_set(tx_buffer, len, size);
6377 dma_unmap_addr_set(tx_buffer, dma, dma);
6379 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6381 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
6382 tx_desc->read.cmd_type_len =
6383 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
6387 if (i == tx_ring->count) {
6388 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6391 tx_desc->read.olinfo_status = 0;
6393 dma += IXGBE_MAX_DATA_PER_TXD;
6394 size -= IXGBE_MAX_DATA_PER_TXD;
6396 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6399 if (likely(!data_len))
6402 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
6406 if (i == tx_ring->count) {
6407 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6410 tx_desc->read.olinfo_status = 0;
6413 size = min_t(unsigned int, data_len, skb_frag_size(frag));
6415 size = skb_frag_size(frag);
6419 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6422 tx_buffer = &tx_ring->tx_buffer_info[i];
6425 /* write last descriptor with RS and EOP bits */
6426 cmd_type |= size | IXGBE_TXD_CMD;
6427 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6429 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6431 /* set the timestamp */
6432 first->time_stamp = jiffies;
6435 * Force memory writes to complete before letting h/w know there
6436 * are new descriptors to fetch. (Only applicable for weak-ordered
6437 * memory model archs, such as IA-64).
6439 * We also need this memory barrier to make certain all of the
6440 * status bits have been updated before next_to_watch is written.
6444 /* set next_to_watch value indicating a packet is present */
6445 first->next_to_watch = tx_desc;
6448 if (i == tx_ring->count)
6451 tx_ring->next_to_use = i;
6453 /* notify HW of packet */
6454 writel(i, tx_ring->tail);
6458 dev_err(tx_ring->dev, "TX DMA map failed\n");
6460 /* clear dma mappings for failed tx_buffer_info map */
6462 tx_buffer = &tx_ring->tx_buffer_info[i];
6463 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6464 if (tx_buffer == first)
6471 tx_ring->next_to_use = i;
6474 static void ixgbe_atr(struct ixgbe_ring *ring,
6475 struct ixgbe_tx_buffer *first)
6477 struct ixgbe_q_vector *q_vector = ring->q_vector;
6478 union ixgbe_atr_hash_dword input = { .dword = 0 };
6479 union ixgbe_atr_hash_dword common = { .dword = 0 };
6481 unsigned char *network;
6483 struct ipv6hdr *ipv6;
6488 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6492 /* do nothing if sampling is disabled */
6493 if (!ring->atr_sample_rate)
6498 /* snag network header to get L4 type and address */
6499 hdr.network = skb_network_header(first->skb);
6501 /* Currently only IPv4/IPv6 with TCP is supported */
6502 if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6503 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6504 (first->protocol != __constant_htons(ETH_P_IP) ||
6505 hdr.ipv4->protocol != IPPROTO_TCP))
6508 th = tcp_hdr(first->skb);
6510 /* skip this packet since it is invalid or the socket is closing */
6514 /* sample on all syn packets or once every atr sample count */
6515 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6518 /* reset sample count */
6519 ring->atr_count = 0;
6521 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6524 * src and dst are inverted, think how the receiver sees them
6526 * The input is broken into two sections, a non-compressed section
6527 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6528 * is XORed together and stored in the compressed dword.
6530 input.formatted.vlan_id = vlan_id;
6533 * since src port and flex bytes occupy the same word XOR them together
6534 * and write the value to source port portion of compressed dword
6536 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6537 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6539 common.port.src ^= th->dest ^ first->protocol;
6540 common.port.dst ^= th->source;
6542 if (first->protocol == __constant_htons(ETH_P_IP)) {
6543 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6544 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6546 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6547 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6548 hdr.ipv6->saddr.s6_addr32[1] ^
6549 hdr.ipv6->saddr.s6_addr32[2] ^
6550 hdr.ipv6->saddr.s6_addr32[3] ^
6551 hdr.ipv6->daddr.s6_addr32[0] ^
6552 hdr.ipv6->daddr.s6_addr32[1] ^
6553 hdr.ipv6->daddr.s6_addr32[2] ^
6554 hdr.ipv6->daddr.s6_addr32[3];
6557 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6558 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6559 input, common, ring->queue_index);
6562 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6564 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6565 /* Herbert's original patch had:
6566 * smp_mb__after_netif_stop_queue();
6567 * but since that doesn't exist yet, just open code it. */
6570 /* We need to check again in a case another CPU has just
6571 * made room available. */
6572 if (likely(ixgbe_desc_unused(tx_ring) < size))
6575 /* A reprieve! - use start_queue because it doesn't call schedule */
6576 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6577 ++tx_ring->tx_stats.restart_queue;
6581 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6583 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6585 return __ixgbe_maybe_stop_tx(tx_ring, size);
6589 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6591 struct ixgbe_adapter *adapter;
6592 struct ixgbe_ring_feature *f;
6596 * only execute the code below if protocol is FCoE
6597 * or FIP and we have FCoE enabled on the adapter
6599 switch (vlan_get_protocol(skb)) {
6600 case __constant_htons(ETH_P_FCOE):
6601 case __constant_htons(ETH_P_FIP):
6602 adapter = netdev_priv(dev);
6604 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
6607 return __netdev_pick_tx(dev, skb);
6610 f = &adapter->ring_feature[RING_F_FCOE];
6612 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6615 while (txq >= f->indices)
6618 return txq + f->offset;
6622 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6623 struct ixgbe_adapter *adapter,
6624 struct ixgbe_ring *tx_ring)
6626 struct ixgbe_tx_buffer *first;
6630 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6631 __be16 protocol = skb->protocol;
6635 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6636 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6637 * + 2 desc gap to keep tail from touching head,
6638 * + 1 desc for context descriptor,
6639 * otherwise try next time
6641 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6642 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6644 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6645 tx_ring->tx_stats.tx_busy++;
6646 return NETDEV_TX_BUSY;
6649 /* record the location of the first descriptor for this packet */
6650 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6652 first->bytecount = skb->len;
6653 first->gso_segs = 1;
6655 /* if we have a HW VLAN tag being added default to the HW one */
6656 if (vlan_tx_tag_present(skb)) {
6657 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6658 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6659 /* else if it is a SW VLAN check the next protocol and store the tag */
6660 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6661 struct vlan_hdr *vhdr, _vhdr;
6662 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6666 protocol = vhdr->h_vlan_encapsulated_proto;
6667 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6668 IXGBE_TX_FLAGS_VLAN_SHIFT;
6669 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6672 skb_tx_timestamp(skb);
6674 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6675 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6676 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
6678 /* schedule check for Tx timestamp */
6679 adapter->ptp_tx_skb = skb_get(skb);
6680 adapter->ptp_tx_start = jiffies;
6681 schedule_work(&adapter->ptp_tx_work);
6684 #ifdef CONFIG_PCI_IOV
6686 * Use the l2switch_enable flag - would be false if the DMA
6687 * Tx switch had been disabled.
6689 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6690 tx_flags |= IXGBE_TX_FLAGS_CC;
6693 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6694 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6695 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6696 (skb->priority != TC_PRIO_CONTROL))) {
6697 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6698 tx_flags |= (skb->priority & 0x7) <<
6699 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6700 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6701 struct vlan_ethhdr *vhdr;
6702 if (skb_header_cloned(skb) &&
6703 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6705 vhdr = (struct vlan_ethhdr *)skb->data;
6706 vhdr->h_vlan_TCI = htons(tx_flags >>
6707 IXGBE_TX_FLAGS_VLAN_SHIFT);
6709 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6713 /* record initial flags and protocol */
6714 first->tx_flags = tx_flags;
6715 first->protocol = protocol;
6718 /* setup tx offload for FCoE */
6719 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6720 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
6721 tso = ixgbe_fso(tx_ring, first, &hdr_len);
6728 #endif /* IXGBE_FCOE */
6729 tso = ixgbe_tso(tx_ring, first, &hdr_len);
6733 ixgbe_tx_csum(tx_ring, first);
6735 /* add the ATR filter if ATR is on */
6736 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6737 ixgbe_atr(tx_ring, first);
6741 #endif /* IXGBE_FCOE */
6742 ixgbe_tx_map(tx_ring, first, hdr_len);
6744 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6746 return NETDEV_TX_OK;
6749 dev_kfree_skb_any(first->skb);
6752 return NETDEV_TX_OK;
6755 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6756 struct net_device *netdev)
6758 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6759 struct ixgbe_ring *tx_ring;
6762 * The minimum packet size for olinfo paylen is 17 so pad the skb
6763 * in order to meet this minimum size requirement.
6765 if (unlikely(skb->len < 17)) {
6766 if (skb_pad(skb, 17 - skb->len))
6767 return NETDEV_TX_OK;
6769 skb_set_tail_pointer(skb, 17);
6772 tx_ring = adapter->tx_ring[skb->queue_mapping];
6773 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6777 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6778 * @netdev: network interface device structure
6779 * @p: pointer to an address structure
6781 * Returns 0 on success, negative on failure
6783 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6785 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6786 struct ixgbe_hw *hw = &adapter->hw;
6787 struct sockaddr *addr = p;
6789 if (!is_valid_ether_addr(addr->sa_data))
6790 return -EADDRNOTAVAIL;
6792 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6793 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6795 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, VMDQ_P(0), IXGBE_RAH_AV);
6801 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6803 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6804 struct ixgbe_hw *hw = &adapter->hw;
6808 if (prtad != hw->phy.mdio.prtad)
6810 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6816 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6817 u16 addr, u16 value)
6819 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6820 struct ixgbe_hw *hw = &adapter->hw;
6822 if (prtad != hw->phy.mdio.prtad)
6824 return hw->phy.ops.write_reg(hw, addr, devad, value);
6827 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6829 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6833 return ixgbe_ptp_hwtstamp_ioctl(adapter, req, cmd);
6835 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6840 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6842 * @netdev: network interface device structure
6844 * Returns non-zero on failure
6846 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6849 struct ixgbe_adapter *adapter = netdev_priv(dev);
6850 struct ixgbe_hw *hw = &adapter->hw;
6852 if (is_valid_ether_addr(hw->mac.san_addr)) {
6854 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
6857 /* update SAN MAC vmdq pool selection */
6858 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
6864 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6866 * @netdev: network interface device structure
6868 * Returns non-zero on failure
6870 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6873 struct ixgbe_adapter *adapter = netdev_priv(dev);
6874 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6876 if (is_valid_ether_addr(mac->san_addr)) {
6878 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6884 #ifdef CONFIG_NET_POLL_CONTROLLER
6886 * Polling 'interrupt' - used by things like netconsole to send skbs
6887 * without having to re-enable interrupts. It's not called while
6888 * the interrupt routine is executing.
6890 static void ixgbe_netpoll(struct net_device *netdev)
6892 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6895 /* if interface is down do nothing */
6896 if (test_bit(__IXGBE_DOWN, &adapter->state))
6899 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6900 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6901 for (i = 0; i < adapter->num_q_vectors; i++)
6902 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
6904 ixgbe_intr(adapter->pdev->irq, netdev);
6906 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6910 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6911 struct rtnl_link_stats64 *stats)
6913 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6917 for (i = 0; i < adapter->num_rx_queues; i++) {
6918 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6924 start = u64_stats_fetch_begin_bh(&ring->syncp);
6925 packets = ring->stats.packets;
6926 bytes = ring->stats.bytes;
6927 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6928 stats->rx_packets += packets;
6929 stats->rx_bytes += bytes;
6933 for (i = 0; i < adapter->num_tx_queues; i++) {
6934 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6940 start = u64_stats_fetch_begin_bh(&ring->syncp);
6941 packets = ring->stats.packets;
6942 bytes = ring->stats.bytes;
6943 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6944 stats->tx_packets += packets;
6945 stats->tx_bytes += bytes;
6949 /* following stats updated by ixgbe_watchdog_task() */
6950 stats->multicast = netdev->stats.multicast;
6951 stats->rx_errors = netdev->stats.rx_errors;
6952 stats->rx_length_errors = netdev->stats.rx_length_errors;
6953 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6954 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6958 #ifdef CONFIG_IXGBE_DCB
6960 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6961 * @adapter: pointer to ixgbe_adapter
6962 * @tc: number of traffic classes currently enabled
6964 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6965 * 802.1Q priority maps to a packet buffer that exists.
6967 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6969 struct ixgbe_hw *hw = &adapter->hw;
6973 /* 82598 have a static priority to TC mapping that can not
6974 * be changed so no validation is needed.
6976 if (hw->mac.type == ixgbe_mac_82598EB)
6979 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6982 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6983 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6985 /* If up2tc is out of bounds default to zero */
6987 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6991 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6997 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
6998 * @adapter: Pointer to adapter struct
7000 * Populate the netdev user priority to tc map
7002 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
7004 struct net_device *dev = adapter->netdev;
7005 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
7006 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
7009 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
7012 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
7013 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
7015 tc = ets->prio_tc[prio];
7017 netdev_set_prio_tc_map(dev, prio, tc);
7021 #endif /* CONFIG_IXGBE_DCB */
7023 * ixgbe_setup_tc - configure net_device for multiple traffic classes
7025 * @netdev: net device to configure
7026 * @tc: number of traffic classes to enable
7028 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7030 struct ixgbe_adapter *adapter = netdev_priv(dev);
7031 struct ixgbe_hw *hw = &adapter->hw;
7033 /* Hardware supports up to 8 traffic classes */
7034 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
7035 (hw->mac.type == ixgbe_mac_82598EB &&
7036 tc < MAX_TRAFFIC_CLASS))
7039 /* Hardware has to reinitialize queues and interrupts to
7040 * match packet buffer alignment. Unfortunately, the
7041 * hardware is not flexible enough to do this dynamically.
7043 if (netif_running(dev))
7045 ixgbe_clear_interrupt_scheme(adapter);
7047 #ifdef CONFIG_IXGBE_DCB
7049 netdev_set_num_tc(dev, tc);
7050 ixgbe_set_prio_tc_map(adapter);
7052 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7054 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
7055 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
7056 adapter->hw.fc.requested_mode = ixgbe_fc_none;
7059 netdev_reset_tc(dev);
7061 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7062 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7064 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7066 adapter->temp_dcb_cfg.pfc_mode_enable = false;
7067 adapter->dcb_cfg.pfc_mode_enable = false;
7070 ixgbe_validate_rtr(adapter, tc);
7072 #endif /* CONFIG_IXGBE_DCB */
7073 ixgbe_init_interrupt_scheme(adapter);
7075 if (netif_running(dev))
7076 return ixgbe_open(dev);
7081 #ifdef CONFIG_PCI_IOV
7082 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
7084 struct net_device *netdev = adapter->netdev;
7087 ixgbe_setup_tc(netdev, netdev_get_num_tc(netdev));
7092 void ixgbe_do_reset(struct net_device *netdev)
7094 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7096 if (netif_running(netdev))
7097 ixgbe_reinit_locked(adapter);
7099 ixgbe_reset(adapter);
7102 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
7103 netdev_features_t features)
7105 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7107 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7108 if (!(features & NETIF_F_RXCSUM))
7109 features &= ~NETIF_F_LRO;
7111 /* Turn off LRO if not RSC capable */
7112 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
7113 features &= ~NETIF_F_LRO;
7118 static int ixgbe_set_features(struct net_device *netdev,
7119 netdev_features_t features)
7121 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7122 netdev_features_t changed = netdev->features ^ features;
7123 bool need_reset = false;
7125 /* Make sure RSC matches LRO, reset if change */
7126 if (!(features & NETIF_F_LRO)) {
7127 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7129 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
7130 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
7131 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7132 if (adapter->rx_itr_setting == 1 ||
7133 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
7134 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
7136 } else if ((changed ^ features) & NETIF_F_LRO) {
7137 e_info(probe, "rx-usecs set too low, "
7143 * Check if Flow Director n-tuple support was enabled or disabled. If
7144 * the state changed, we need to reset.
7146 switch (features & NETIF_F_NTUPLE) {
7147 case NETIF_F_NTUPLE:
7148 /* turn off ATR, enable perfect filters and reset */
7149 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
7152 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7153 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7156 /* turn off perfect filters, enable ATR and reset */
7157 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7160 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7162 /* We cannot enable ATR if SR-IOV is enabled */
7163 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7166 /* We cannot enable ATR if we have 2 or more traffic classes */
7167 if (netdev_get_num_tc(netdev) > 1)
7170 /* We cannot enable ATR if RSS is disabled */
7171 if (adapter->ring_feature[RING_F_RSS].limit <= 1)
7174 /* A sample rate of 0 indicates ATR disabled */
7175 if (!adapter->atr_sample_rate)
7178 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7182 if (features & NETIF_F_HW_VLAN_CTAG_RX)
7183 ixgbe_vlan_strip_enable(adapter);
7185 ixgbe_vlan_strip_disable(adapter);
7187 if (changed & NETIF_F_RXALL)
7190 netdev->features = features;
7192 ixgbe_do_reset(netdev);
7197 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
7198 struct net_device *dev,
7199 const unsigned char *addr,
7202 struct ixgbe_adapter *adapter = netdev_priv(dev);
7205 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7206 return ndo_dflt_fdb_add(ndm, tb, dev, addr, flags);
7208 /* Hardware does not support aging addresses so if a
7209 * ndm_state is given only allow permanent addresses
7211 if (ndm->ndm_state && !(ndm->ndm_state & NUD_PERMANENT)) {
7212 pr_info("%s: FDB only supports static addresses\n",
7217 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
7218 u32 rar_uc_entries = IXGBE_MAX_PF_MACVLANS;
7220 if (netdev_uc_count(dev) < rar_uc_entries)
7221 err = dev_uc_add_excl(dev, addr);
7224 } else if (is_multicast_ether_addr(addr)) {
7225 err = dev_mc_add_excl(dev, addr);
7230 /* Only return duplicate errors if NLM_F_EXCL is set */
7231 if (err == -EEXIST && !(flags & NLM_F_EXCL))
7237 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
7238 struct nlmsghdr *nlh)
7240 struct ixgbe_adapter *adapter = netdev_priv(dev);
7241 struct nlattr *attr, *br_spec;
7244 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7247 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
7249 nla_for_each_nested(attr, br_spec, rem) {
7253 if (nla_type(attr) != IFLA_BRIDGE_MODE)
7256 mode = nla_get_u16(attr);
7257 if (mode == BRIDGE_MODE_VEPA) {
7259 adapter->flags2 &= ~IXGBE_FLAG2_BRIDGE_MODE_VEB;
7260 } else if (mode == BRIDGE_MODE_VEB) {
7261 reg = IXGBE_PFDTXGSWC_VT_LBEN;
7262 adapter->flags2 |= IXGBE_FLAG2_BRIDGE_MODE_VEB;
7266 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, reg);
7268 e_info(drv, "enabling bridge mode: %s\n",
7269 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
7275 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
7276 struct net_device *dev,
7279 struct ixgbe_adapter *adapter = netdev_priv(dev);
7282 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
7285 if (adapter->flags2 & IXGBE_FLAG2_BRIDGE_MODE_VEB)
7286 mode = BRIDGE_MODE_VEB;
7288 mode = BRIDGE_MODE_VEPA;
7290 return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode);
7293 static const struct net_device_ops ixgbe_netdev_ops = {
7294 .ndo_open = ixgbe_open,
7295 .ndo_stop = ixgbe_close,
7296 .ndo_start_xmit = ixgbe_xmit_frame,
7298 .ndo_select_queue = ixgbe_select_queue,
7300 .ndo_set_rx_mode = ixgbe_set_rx_mode,
7301 .ndo_validate_addr = eth_validate_addr,
7302 .ndo_set_mac_address = ixgbe_set_mac,
7303 .ndo_change_mtu = ixgbe_change_mtu,
7304 .ndo_tx_timeout = ixgbe_tx_timeout,
7305 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
7306 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
7307 .ndo_do_ioctl = ixgbe_ioctl,
7308 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
7309 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
7310 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
7311 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
7312 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
7313 .ndo_get_stats64 = ixgbe_get_stats64,
7314 #ifdef CONFIG_IXGBE_DCB
7315 .ndo_setup_tc = ixgbe_setup_tc,
7317 #ifdef CONFIG_NET_POLL_CONTROLLER
7318 .ndo_poll_controller = ixgbe_netpoll,
7320 #ifdef CONFIG_NET_RX_BUSY_POLL
7321 .ndo_busy_poll = ixgbe_low_latency_recv,
7324 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
7325 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
7326 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
7327 .ndo_fcoe_enable = ixgbe_fcoe_enable,
7328 .ndo_fcoe_disable = ixgbe_fcoe_disable,
7329 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
7330 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
7331 #endif /* IXGBE_FCOE */
7332 .ndo_set_features = ixgbe_set_features,
7333 .ndo_fix_features = ixgbe_fix_features,
7334 .ndo_fdb_add = ixgbe_ndo_fdb_add,
7335 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
7336 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
7340 * ixgbe_enumerate_functions - Get the number of ports this device has
7341 * @adapter: adapter structure
7343 * This function enumerates the phsyical functions co-located on a single slot,
7344 * in order to determine how many ports a device has. This is most useful in
7345 * determining the required GT/s of PCIe bandwidth necessary for optimal
7348 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
7350 struct ixgbe_hw *hw = &adapter->hw;
7351 struct list_head *entry;
7354 /* Some cards can not use the generic count PCIe functions method, and
7355 * so must be hardcoded to the correct value.
7357 switch (hw->device_id) {
7358 case IXGBE_DEV_ID_82599_SFP_SF_QP:
7359 case IXGBE_DEV_ID_82599_QSFP_SF_QP:
7363 list_for_each(entry, &adapter->pdev->bus_list) {
7364 struct pci_dev *pdev =
7365 list_entry(entry, struct pci_dev, bus_list);
7366 /* don't count virtual functions */
7367 if (!pdev->is_virtfn)
7376 * ixgbe_wol_supported - Check whether device supports WoL
7377 * @hw: hw specific details
7378 * @device_id: the device ID
7379 * @subdev_id: the subsystem device ID
7381 * This function is used by probe and ethtool to determine
7382 * which devices have WoL support
7385 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
7388 struct ixgbe_hw *hw = &adapter->hw;
7389 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7390 int is_wol_supported = 0;
7392 switch (device_id) {
7393 case IXGBE_DEV_ID_82599_SFP:
7394 /* Only these subdevices could supports WOL */
7395 switch (subdevice_id) {
7396 case IXGBE_SUBDEV_ID_82599_560FLR:
7397 /* only support first port */
7398 if (hw->bus.func != 0)
7400 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
7401 case IXGBE_SUBDEV_ID_82599_SFP:
7402 case IXGBE_SUBDEV_ID_82599_RNDC:
7403 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
7404 case IXGBE_SUBDEV_ID_82599_LOM_SFP:
7405 is_wol_supported = 1;
7409 case IXGBE_DEV_ID_82599EN_SFP:
7410 /* Only this subdevice supports WOL */
7411 switch (subdevice_id) {
7412 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
7413 is_wol_supported = 1;
7417 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7418 /* All except this subdevice support WOL */
7419 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7420 is_wol_supported = 1;
7422 case IXGBE_DEV_ID_82599_KX4:
7423 is_wol_supported = 1;
7425 case IXGBE_DEV_ID_X540T:
7426 case IXGBE_DEV_ID_X540T1:
7427 /* check eeprom to see if enabled wol */
7428 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7429 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7430 (hw->bus.func == 0))) {
7431 is_wol_supported = 1;
7436 return is_wol_supported;
7440 * ixgbe_probe - Device Initialization Routine
7441 * @pdev: PCI device information struct
7442 * @ent: entry in ixgbe_pci_tbl
7444 * Returns 0 on success, negative on failure
7446 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7447 * The OS initialization, configuring of the adapter private structure,
7448 * and a hardware reset occur.
7450 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
7452 struct net_device *netdev;
7453 struct ixgbe_adapter *adapter = NULL;
7454 struct ixgbe_hw *hw;
7455 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
7456 static int cards_found;
7457 int i, err, pci_using_dac, expected_gts;
7458 unsigned int indices = MAX_TX_QUEUES;
7459 u8 part_str[IXGBE_PBANUM_LENGTH];
7465 /* Catch broken hardware that put the wrong VF device ID in
7466 * the PCIe SR-IOV capability.
7468 if (pdev->is_virtfn) {
7469 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
7470 pci_name(pdev), pdev->vendor, pdev->device);
7474 err = pci_enable_device_mem(pdev);
7478 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
7479 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
7482 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
7484 err = dma_set_coherent_mask(&pdev->dev,
7488 "No usable DMA configuration, aborting\n");
7495 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
7496 IORESOURCE_MEM), ixgbe_driver_name);
7499 "pci_request_selected_regions failed 0x%x\n", err);
7503 pci_enable_pcie_error_reporting(pdev);
7505 pci_set_master(pdev);
7506 pci_save_state(pdev);
7508 if (ii->mac == ixgbe_mac_82598EB) {
7509 #ifdef CONFIG_IXGBE_DCB
7510 /* 8 TC w/ 4 queues per TC */
7511 indices = 4 * MAX_TRAFFIC_CLASS;
7513 indices = IXGBE_MAX_RSS_INDICES;
7517 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7520 goto err_alloc_etherdev;
7523 SET_NETDEV_DEV(netdev, &pdev->dev);
7525 adapter = netdev_priv(netdev);
7526 pci_set_drvdata(pdev, adapter);
7528 adapter->netdev = netdev;
7529 adapter->pdev = pdev;
7532 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7534 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7535 pci_resource_len(pdev, 0));
7541 netdev->netdev_ops = &ixgbe_netdev_ops;
7542 ixgbe_set_ethtool_ops(netdev);
7543 netdev->watchdog_timeo = 5 * HZ;
7544 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7546 adapter->bd_number = cards_found;
7549 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7550 hw->mac.type = ii->mac;
7553 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7554 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7555 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7556 if (!(eec & (1 << 8)))
7557 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7560 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7561 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7562 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7563 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7564 hw->phy.mdio.mmds = 0;
7565 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7566 hw->phy.mdio.dev = netdev;
7567 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7568 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7570 ii->get_invariants(hw);
7572 /* setup the private structure */
7573 err = ixgbe_sw_init(adapter);
7577 /* Cache if MNG FW is up so we don't have to read the REG later */
7578 if (hw->mac.ops.mng_fw_enabled)
7579 hw->mng_fw_enabled = hw->mac.ops.mng_fw_enabled(hw);
7581 /* Make it possible the adapter to be woken up via WOL */
7582 switch (adapter->hw.mac.type) {
7583 case ixgbe_mac_82599EB:
7584 case ixgbe_mac_X540:
7585 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7592 * If there is a fan on this device and it has failed log the
7595 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7596 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7597 if (esdp & IXGBE_ESDP_SDP1)
7598 e_crit(probe, "Fan has stopped, replace the adapter\n");
7601 if (allow_unsupported_sfp)
7602 hw->allow_unsupported_sfp = allow_unsupported_sfp;
7604 /* reset_hw fills in the perm_addr as well */
7605 hw->phy.reset_if_overtemp = true;
7606 err = hw->mac.ops.reset_hw(hw);
7607 hw->phy.reset_if_overtemp = false;
7608 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7609 hw->mac.type == ixgbe_mac_82598EB) {
7611 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7612 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
7613 e_dev_err("Reload the driver after installing a supported module.\n");
7616 e_dev_err("HW Init failed: %d\n", err);
7620 #ifdef CONFIG_PCI_IOV
7621 /* SR-IOV not supported on the 82598 */
7622 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7625 ixgbe_init_mbx_params_pf(hw);
7626 memcpy(&hw->mbx.ops, ii->mbx_ops, sizeof(hw->mbx.ops));
7627 ixgbe_enable_sriov(adapter);
7628 pci_sriov_set_totalvfs(pdev, 63);
7632 netdev->features = NETIF_F_SG |
7635 NETIF_F_HW_VLAN_CTAG_TX |
7636 NETIF_F_HW_VLAN_CTAG_RX |
7637 NETIF_F_HW_VLAN_CTAG_FILTER |
7643 netdev->hw_features = netdev->features;
7645 switch (adapter->hw.mac.type) {
7646 case ixgbe_mac_82599EB:
7647 case ixgbe_mac_X540:
7648 netdev->features |= NETIF_F_SCTP_CSUM;
7649 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7656 netdev->hw_features |= NETIF_F_RXALL;
7658 netdev->vlan_features |= NETIF_F_TSO;
7659 netdev->vlan_features |= NETIF_F_TSO6;
7660 netdev->vlan_features |= NETIF_F_IP_CSUM;
7661 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7662 netdev->vlan_features |= NETIF_F_SG;
7664 netdev->priv_flags |= IFF_UNICAST_FLT;
7665 netdev->priv_flags |= IFF_SUPP_NOFCS;
7667 #ifdef CONFIG_IXGBE_DCB
7668 netdev->dcbnl_ops = &dcbnl_ops;
7672 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7673 unsigned int fcoe_l;
7675 if (hw->mac.ops.get_device_caps) {
7676 hw->mac.ops.get_device_caps(hw, &device_caps);
7677 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7678 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7682 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
7683 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
7685 netdev->features |= NETIF_F_FSO |
7688 netdev->vlan_features |= NETIF_F_FSO |
7692 #endif /* IXGBE_FCOE */
7693 if (pci_using_dac) {
7694 netdev->features |= NETIF_F_HIGHDMA;
7695 netdev->vlan_features |= NETIF_F_HIGHDMA;
7698 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7699 netdev->hw_features |= NETIF_F_LRO;
7700 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7701 netdev->features |= NETIF_F_LRO;
7703 /* make sure the EEPROM is good */
7704 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7705 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7710 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7712 if (!is_valid_ether_addr(netdev->dev_addr)) {
7713 e_dev_err("invalid MAC address\n");
7718 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7719 (unsigned long) adapter);
7721 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7722 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7724 err = ixgbe_init_interrupt_scheme(adapter);
7728 /* WOL not supported for all devices */
7730 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7731 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
7732 pdev->subsystem_device);
7733 if (hw->wol_enabled)
7734 adapter->wol = IXGBE_WUFC_MAG;
7736 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7738 /* save off EEPROM version number */
7739 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7740 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7742 /* pick up the PCI bus settings for reporting later */
7743 hw->mac.ops.get_bus_info(hw);
7744 if (ixgbe_pcie_from_parent(hw))
7745 ixgbe_get_parent_bus_info(adapter);
7747 /* print bus type/speed/width info */
7748 e_dev_info("(PCI Express:%s:%s) %pM\n",
7749 (hw->bus.speed == ixgbe_bus_speed_8000 ? "8.0GT/s" :
7750 hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7751 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7753 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7754 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7755 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7759 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7761 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7762 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7763 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7764 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7767 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7768 hw->mac.type, hw->phy.type, part_str);
7770 /* calculate the expected PCIe bandwidth required for optimal
7771 * performance. Note that some older parts will never have enough
7772 * bandwidth due to being older generation PCIe parts. We clamp these
7773 * parts to ensure no warning is displayed if it can't be fixed.
7775 switch (hw->mac.type) {
7776 case ixgbe_mac_82598EB:
7777 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
7780 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
7783 ixgbe_check_minimum_link(adapter, expected_gts);
7785 /* reset the hardware with the new settings */
7786 err = hw->mac.ops.start_hw(hw);
7787 if (err == IXGBE_ERR_EEPROM_VERSION) {
7788 /* We are running on a pre-production device, log a warning */
7789 e_dev_warn("This device is a pre-production adapter/LOM. "
7790 "Please be aware there may be issues associated "
7791 "with your hardware. If you are experiencing "
7792 "problems please contact your Intel or hardware "
7793 "representative who provided you with this "
7796 strcpy(netdev->name, "eth%d");
7797 err = register_netdev(netdev);
7801 /* power down the optics for 82599 SFP+ fiber */
7802 if (hw->mac.ops.disable_tx_laser)
7803 hw->mac.ops.disable_tx_laser(hw);
7805 /* carrier off reporting is important to ethtool even BEFORE open */
7806 netif_carrier_off(netdev);
7808 #ifdef CONFIG_IXGBE_DCA
7809 if (dca_add_requester(&pdev->dev) == 0) {
7810 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7811 ixgbe_setup_dca(adapter);
7814 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7815 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7816 for (i = 0; i < adapter->num_vfs; i++)
7817 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7820 /* firmware requires driver version to be 0xFFFFFFFF
7821 * since os does not support feature
7823 if (hw->mac.ops.set_fw_drv_ver)
7824 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7827 /* add san mac addr to netdev */
7828 ixgbe_add_sanmac_netdev(netdev);
7830 e_dev_info("%s\n", ixgbe_default_device_descr);
7833 #ifdef CONFIG_IXGBE_HWMON
7834 if (ixgbe_sysfs_init(adapter))
7835 e_err(probe, "failed to allocate sysfs resources\n");
7836 #endif /* CONFIG_IXGBE_HWMON */
7838 ixgbe_dbg_adapter_init(adapter);
7840 /* Need link setup for MNG FW, else wait for IXGBE_UP */
7841 if (hw->mng_fw_enabled && hw->mac.ops.setup_link)
7842 hw->mac.ops.setup_link(hw,
7843 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
7849 ixgbe_release_hw_control(adapter);
7850 ixgbe_clear_interrupt_scheme(adapter);
7852 ixgbe_disable_sriov(adapter);
7853 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7854 iounmap(hw->hw_addr);
7856 free_netdev(netdev);
7858 pci_release_selected_regions(pdev,
7859 pci_select_bars(pdev, IORESOURCE_MEM));
7862 pci_disable_device(pdev);
7867 * ixgbe_remove - Device Removal Routine
7868 * @pdev: PCI device information struct
7870 * ixgbe_remove is called by the PCI subsystem to alert the driver
7871 * that it should release a PCI device. The could be caused by a
7872 * Hot-Plug event, or because the driver is going to be removed from
7875 static void ixgbe_remove(struct pci_dev *pdev)
7877 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7878 struct net_device *netdev = adapter->netdev;
7880 ixgbe_dbg_adapter_exit(adapter);
7882 set_bit(__IXGBE_DOWN, &adapter->state);
7883 cancel_work_sync(&adapter->service_task);
7886 #ifdef CONFIG_IXGBE_DCA
7887 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7888 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7889 dca_remove_requester(&pdev->dev);
7890 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7894 #ifdef CONFIG_IXGBE_HWMON
7895 ixgbe_sysfs_exit(adapter);
7896 #endif /* CONFIG_IXGBE_HWMON */
7898 /* remove the added san mac */
7899 ixgbe_del_sanmac_netdev(netdev);
7901 if (netdev->reg_state == NETREG_REGISTERED)
7902 unregister_netdev(netdev);
7904 #ifdef CONFIG_PCI_IOV
7906 * Only disable SR-IOV on unload if the user specified the now
7907 * deprecated max_vfs module parameter.
7910 ixgbe_disable_sriov(adapter);
7912 ixgbe_clear_interrupt_scheme(adapter);
7914 ixgbe_release_hw_control(adapter);
7917 kfree(adapter->ixgbe_ieee_pfc);
7918 kfree(adapter->ixgbe_ieee_ets);
7921 iounmap(adapter->hw.hw_addr);
7922 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7925 e_dev_info("complete\n");
7927 free_netdev(netdev);
7929 pci_disable_pcie_error_reporting(pdev);
7931 pci_disable_device(pdev);
7935 * ixgbe_io_error_detected - called when PCI error is detected
7936 * @pdev: Pointer to PCI device
7937 * @state: The current pci connection state
7939 * This function is called after a PCI bus error affecting
7940 * this device has been detected.
7942 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7943 pci_channel_state_t state)
7945 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7946 struct net_device *netdev = adapter->netdev;
7948 #ifdef CONFIG_PCI_IOV
7949 struct pci_dev *bdev, *vfdev;
7950 u32 dw0, dw1, dw2, dw3;
7952 u16 req_id, pf_func;
7954 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7955 adapter->num_vfs == 0)
7956 goto skip_bad_vf_detection;
7958 bdev = pdev->bus->self;
7959 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
7960 bdev = bdev->bus->self;
7963 goto skip_bad_vf_detection;
7965 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7967 goto skip_bad_vf_detection;
7969 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7970 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7971 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7972 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7975 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7976 if (!(req_id & 0x0080))
7977 goto skip_bad_vf_detection;
7979 pf_func = req_id & 0x01;
7980 if ((pf_func & 1) == (pdev->devfn & 1)) {
7981 unsigned int device_id;
7983 vf = (req_id & 0x7F) >> 1;
7984 e_dev_err("VF %d has caused a PCIe error\n", vf);
7985 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7986 "%8.8x\tdw3: %8.8x\n",
7987 dw0, dw1, dw2, dw3);
7988 switch (adapter->hw.mac.type) {
7989 case ixgbe_mac_82599EB:
7990 device_id = IXGBE_82599_VF_DEVICE_ID;
7992 case ixgbe_mac_X540:
7993 device_id = IXGBE_X540_VF_DEVICE_ID;
8000 /* Find the pci device of the offending VF */
8001 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
8003 if (vfdev->devfn == (req_id & 0xFF))
8005 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
8009 * There's a slim chance the VF could have been hot plugged,
8010 * so if it is no longer present we don't need to issue the
8011 * VFLR. Just clean up the AER in that case.
8014 e_dev_err("Issuing VFLR to VF %d\n", vf);
8015 pci_write_config_dword(vfdev, 0xA8, 0x00008000);
8016 /* Free device reference count */
8020 pci_cleanup_aer_uncorrect_error_status(pdev);
8024 * Even though the error may have occurred on the other port
8025 * we still need to increment the vf error reference count for
8026 * both ports because the I/O resume function will be called
8029 adapter->vferr_refcount++;
8031 return PCI_ERS_RESULT_RECOVERED;
8033 skip_bad_vf_detection:
8034 #endif /* CONFIG_PCI_IOV */
8035 netif_device_detach(netdev);
8037 if (state == pci_channel_io_perm_failure)
8038 return PCI_ERS_RESULT_DISCONNECT;
8040 if (netif_running(netdev))
8041 ixgbe_down(adapter);
8042 pci_disable_device(pdev);
8044 /* Request a slot reset. */
8045 return PCI_ERS_RESULT_NEED_RESET;
8049 * ixgbe_io_slot_reset - called after the pci bus has been reset.
8050 * @pdev: Pointer to PCI device
8052 * Restart the card from scratch, as if from a cold-boot.
8054 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
8056 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8057 pci_ers_result_t result;
8060 if (pci_enable_device_mem(pdev)) {
8061 e_err(probe, "Cannot re-enable PCI device after reset.\n");
8062 result = PCI_ERS_RESULT_DISCONNECT;
8064 pci_set_master(pdev);
8065 pci_restore_state(pdev);
8066 pci_save_state(pdev);
8068 pci_wake_from_d3(pdev, false);
8070 ixgbe_reset(adapter);
8071 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
8072 result = PCI_ERS_RESULT_RECOVERED;
8075 err = pci_cleanup_aer_uncorrect_error_status(pdev);
8077 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
8078 "failed 0x%0x\n", err);
8079 /* non-fatal, continue */
8086 * ixgbe_io_resume - called when traffic can start flowing again.
8087 * @pdev: Pointer to PCI device
8089 * This callback is called when the error recovery driver tells us that
8090 * its OK to resume normal operation.
8092 static void ixgbe_io_resume(struct pci_dev *pdev)
8094 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8095 struct net_device *netdev = adapter->netdev;
8097 #ifdef CONFIG_PCI_IOV
8098 if (adapter->vferr_refcount) {
8099 e_info(drv, "Resuming after VF err\n");
8100 adapter->vferr_refcount--;
8105 if (netif_running(netdev))
8108 netif_device_attach(netdev);
8111 static const struct pci_error_handlers ixgbe_err_handler = {
8112 .error_detected = ixgbe_io_error_detected,
8113 .slot_reset = ixgbe_io_slot_reset,
8114 .resume = ixgbe_io_resume,
8117 static struct pci_driver ixgbe_driver = {
8118 .name = ixgbe_driver_name,
8119 .id_table = ixgbe_pci_tbl,
8120 .probe = ixgbe_probe,
8121 .remove = ixgbe_remove,
8123 .suspend = ixgbe_suspend,
8124 .resume = ixgbe_resume,
8126 .shutdown = ixgbe_shutdown,
8127 .sriov_configure = ixgbe_pci_sriov_configure,
8128 .err_handler = &ixgbe_err_handler
8132 * ixgbe_init_module - Driver Registration Routine
8134 * ixgbe_init_module is the first routine called when the driver is
8135 * loaded. All it does is register with the PCI subsystem.
8137 static int __init ixgbe_init_module(void)
8140 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
8141 pr_info("%s\n", ixgbe_copyright);
8145 ret = pci_register_driver(&ixgbe_driver);
8151 #ifdef CONFIG_IXGBE_DCA
8152 dca_register_notify(&dca_notifier);
8158 module_init(ixgbe_init_module);
8161 * ixgbe_exit_module - Driver Exit Cleanup Routine
8163 * ixgbe_exit_module is called just before the driver is removed
8166 static void __exit ixgbe_exit_module(void)
8168 #ifdef CONFIG_IXGBE_DCA
8169 dca_unregister_notify(&dca_notifier);
8171 pci_unregister_driver(&ixgbe_driver);
8175 rcu_barrier(); /* Wait for completion of call_rcu()'s */
8178 #ifdef CONFIG_IXGBE_DCA
8179 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
8184 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
8185 __ixgbe_notify_dca);
8187 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
8190 #endif /* CONFIG_IXGBE_DCA */
8192 module_exit(ixgbe_exit_module);