1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2014 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 Linux NICS <linux.nics@intel.com>
24 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
27 *******************************************************************************/
29 #include <linux/types.h>
30 #include <linux/module.h>
31 #include <linux/pci.h>
32 #include <linux/netdevice.h>
33 #include <linux/vmalloc.h>
34 #include <linux/string.h>
36 #include <linux/interrupt.h>
38 #include <linux/tcp.h>
39 #include <linux/sctp.h>
40 #include <linux/pkt_sched.h>
41 #include <linux/ipv6.h>
42 #include <linux/slab.h>
43 #include <net/checksum.h>
44 #include <net/ip6_checksum.h>
45 #include <linux/etherdevice.h>
46 #include <linux/ethtool.h>
48 #include <linux/if_vlan.h>
49 #include <linux/if_macvlan.h>
50 #include <linux/if_bridge.h>
51 #include <linux/prefetch.h>
52 #include <scsi/fc/fc_fcoe.h>
53 #include <net/vxlan.h>
56 #include <linux/of_net.h>
60 #include <asm/idprom.h>
65 #include "ixgbe_common.h"
66 #include "ixgbe_dcb_82599.h"
67 #include "ixgbe_sriov.h"
69 char ixgbe_driver_name[] = "ixgbe";
70 static const char ixgbe_driver_string[] =
71 "Intel(R) 10 Gigabit PCI Express Network Driver";
73 char ixgbe_default_device_descr[] =
74 "Intel(R) 10 Gigabit Network Connection";
76 static char ixgbe_default_device_descr[] =
77 "Intel(R) 10 Gigabit Network Connection";
79 #define DRV_VERSION "4.0.1-k"
80 const char ixgbe_driver_version[] = DRV_VERSION;
81 static const char ixgbe_copyright[] =
82 "Copyright (c) 1999-2014 Intel Corporation.";
84 static const struct ixgbe_info *ixgbe_info_tbl[] = {
85 [board_82598] = &ixgbe_82598_info,
86 [board_82599] = &ixgbe_82599_info,
87 [board_X540] = &ixgbe_X540_info,
88 [board_X550] = &ixgbe_X550_info,
89 [board_X550EM_x] = &ixgbe_X550EM_x_info,
92 /* ixgbe_pci_tbl - PCI Device ID Table
94 * Wildcard entries (PCI_ANY_ID) should come last
95 * Last entry must be all 0s
97 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
98 * Class, Class Mask, private data (not used) }
100 static const struct pci_device_id ixgbe_pci_tbl[] = {
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
117 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
118 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
119 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
120 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
121 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
122 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
123 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
124 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
125 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
126 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
127 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
128 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
129 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
130 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
131 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
132 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
133 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
134 /* required last entry */
137 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
139 #ifdef CONFIG_IXGBE_DCA
140 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
142 static struct notifier_block dca_notifier = {
143 .notifier_call = ixgbe_notify_dca,
149 #ifdef CONFIG_PCI_IOV
150 static unsigned int max_vfs;
151 module_param(max_vfs, uint, 0);
152 MODULE_PARM_DESC(max_vfs,
153 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
154 #endif /* CONFIG_PCI_IOV */
156 static unsigned int allow_unsupported_sfp;
157 module_param(allow_unsupported_sfp, uint, 0);
158 MODULE_PARM_DESC(allow_unsupported_sfp,
159 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
161 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
162 static int debug = -1;
163 module_param(debug, int, 0);
164 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
166 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
167 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
168 MODULE_LICENSE("GPL");
169 MODULE_VERSION(DRV_VERSION);
171 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
173 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
176 struct pci_dev *parent_dev;
177 struct pci_bus *parent_bus;
179 parent_bus = adapter->pdev->bus->parent;
183 parent_dev = parent_bus->self;
187 if (!pci_is_pcie(parent_dev))
190 pcie_capability_read_word(parent_dev, reg, value);
191 if (*value == IXGBE_FAILED_READ_CFG_WORD &&
192 ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
197 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
199 struct ixgbe_hw *hw = &adapter->hw;
203 hw->bus.type = ixgbe_bus_type_pci_express;
205 /* Get the negotiated link width and speed from PCI config space of the
206 * parent, as this device is behind a switch
208 err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
210 /* assume caller will handle error case */
214 hw->bus.width = ixgbe_convert_bus_width(link_status);
215 hw->bus.speed = ixgbe_convert_bus_speed(link_status);
221 * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
222 * @hw: hw specific details
224 * This function is used by probe to determine whether a device's PCI-Express
225 * bandwidth details should be gathered from the parent bus instead of from the
226 * device. Used to ensure that various locations all have the correct device ID
229 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
231 switch (hw->device_id) {
232 case IXGBE_DEV_ID_82599_SFP_SF_QP:
233 case IXGBE_DEV_ID_82599_QSFP_SF_QP:
240 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
244 enum pci_bus_speed speed = PCI_SPEED_UNKNOWN;
245 enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN;
246 struct pci_dev *pdev;
248 /* determine whether to use the the parent device
250 if (ixgbe_pcie_from_parent(&adapter->hw))
251 pdev = adapter->pdev->bus->parent->self;
253 pdev = adapter->pdev;
255 if (pcie_get_minimum_link(pdev, &speed, &width) ||
256 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
257 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
262 case PCIE_SPEED_2_5GT:
263 /* 8b/10b encoding reduces max throughput by 20% */
266 case PCIE_SPEED_5_0GT:
267 /* 8b/10b encoding reduces max throughput by 20% */
270 case PCIE_SPEED_8_0GT:
271 /* 128b/130b encoding reduces throughput by less than 2% */
275 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
279 e_dev_info("PCI Express bandwidth of %dGT/s available\n",
281 e_dev_info("(Speed:%s, Width: x%d, Encoding Loss:%s)\n",
282 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" :
283 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" :
284 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" :
287 (speed == PCIE_SPEED_2_5GT ? "20%" :
288 speed == PCIE_SPEED_5_0GT ? "20%" :
289 speed == PCIE_SPEED_8_0GT ? "<2%" :
292 if (max_gts < expected_gts) {
293 e_dev_warn("This is not sufficient for optimal performance of this card.\n");
294 e_dev_warn("For optimal performance, at least %dGT/s of bandwidth is required.\n",
296 e_dev_warn("A slot with more lanes and/or higher speed is suggested.\n");
300 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
302 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
303 !test_bit(__IXGBE_REMOVING, &adapter->state) &&
304 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
305 schedule_work(&adapter->service_task);
308 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
310 struct ixgbe_adapter *adapter = hw->back;
315 e_dev_err("Adapter removed\n");
316 if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
317 ixgbe_service_event_schedule(adapter);
320 static void ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
324 /* The following check not only optimizes a bit by not
325 * performing a read on the status register when the
326 * register just read was a status register read that
327 * returned IXGBE_FAILED_READ_REG. It also blocks any
328 * potential recursion.
330 if (reg == IXGBE_STATUS) {
331 ixgbe_remove_adapter(hw);
334 value = ixgbe_read_reg(hw, IXGBE_STATUS);
335 if (value == IXGBE_FAILED_READ_REG)
336 ixgbe_remove_adapter(hw);
340 * ixgbe_read_reg - Read from device register
341 * @hw: hw specific details
342 * @reg: offset of register to read
344 * Returns : value read or IXGBE_FAILED_READ_REG if removed
346 * This function is used to read device registers. It checks for device
347 * removal by confirming any read that returns all ones by checking the
348 * status register value for all ones. This function avoids reading from
349 * the hardware if a removal was previously detected in which case it
350 * returns IXGBE_FAILED_READ_REG (all ones).
352 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
354 u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
357 if (ixgbe_removed(reg_addr))
358 return IXGBE_FAILED_READ_REG;
359 value = readl(reg_addr + reg);
360 if (unlikely(value == IXGBE_FAILED_READ_REG))
361 ixgbe_check_remove(hw, reg);
365 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
369 pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
370 if (value == IXGBE_FAILED_READ_CFG_WORD) {
371 ixgbe_remove_adapter(hw);
377 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
379 struct ixgbe_adapter *adapter = hw->back;
382 if (ixgbe_removed(hw->hw_addr))
383 return IXGBE_FAILED_READ_CFG_WORD;
384 pci_read_config_word(adapter->pdev, reg, &value);
385 if (value == IXGBE_FAILED_READ_CFG_WORD &&
386 ixgbe_check_cfg_remove(hw, adapter->pdev))
387 return IXGBE_FAILED_READ_CFG_WORD;
391 #ifdef CONFIG_PCI_IOV
392 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
394 struct ixgbe_adapter *adapter = hw->back;
397 if (ixgbe_removed(hw->hw_addr))
398 return IXGBE_FAILED_READ_CFG_DWORD;
399 pci_read_config_dword(adapter->pdev, reg, &value);
400 if (value == IXGBE_FAILED_READ_CFG_DWORD &&
401 ixgbe_check_cfg_remove(hw, adapter->pdev))
402 return IXGBE_FAILED_READ_CFG_DWORD;
405 #endif /* CONFIG_PCI_IOV */
407 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
409 struct ixgbe_adapter *adapter = hw->back;
411 if (ixgbe_removed(hw->hw_addr))
413 pci_write_config_word(adapter->pdev, reg, value);
416 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
418 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
420 /* flush memory to make sure state is correct before next watchdog */
421 smp_mb__before_atomic();
422 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
425 struct ixgbe_reg_info {
430 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
432 /* General Registers */
433 {IXGBE_CTRL, "CTRL"},
434 {IXGBE_STATUS, "STATUS"},
435 {IXGBE_CTRL_EXT, "CTRL_EXT"},
437 /* Interrupt Registers */
438 {IXGBE_EICR, "EICR"},
441 {IXGBE_SRRCTL(0), "SRRCTL"},
442 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
443 {IXGBE_RDLEN(0), "RDLEN"},
444 {IXGBE_RDH(0), "RDH"},
445 {IXGBE_RDT(0), "RDT"},
446 {IXGBE_RXDCTL(0), "RXDCTL"},
447 {IXGBE_RDBAL(0), "RDBAL"},
448 {IXGBE_RDBAH(0), "RDBAH"},
451 {IXGBE_TDBAL(0), "TDBAL"},
452 {IXGBE_TDBAH(0), "TDBAH"},
453 {IXGBE_TDLEN(0), "TDLEN"},
454 {IXGBE_TDH(0), "TDH"},
455 {IXGBE_TDT(0), "TDT"},
456 {IXGBE_TXDCTL(0), "TXDCTL"},
458 /* List Terminator */
464 * ixgbe_regdump - register printout routine
466 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
472 switch (reginfo->ofs) {
473 case IXGBE_SRRCTL(0):
474 for (i = 0; i < 64; i++)
475 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
477 case IXGBE_DCA_RXCTRL(0):
478 for (i = 0; i < 64; i++)
479 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
482 for (i = 0; i < 64; i++)
483 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
486 for (i = 0; i < 64; i++)
487 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
490 for (i = 0; i < 64; i++)
491 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
493 case IXGBE_RXDCTL(0):
494 for (i = 0; i < 64; i++)
495 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
498 for (i = 0; i < 64; i++)
499 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
502 for (i = 0; i < 64; i++)
503 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
506 for (i = 0; i < 64; i++)
507 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
510 for (i = 0; i < 64; i++)
511 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
514 for (i = 0; i < 64; i++)
515 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
518 for (i = 0; i < 64; i++)
519 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
522 for (i = 0; i < 64; i++)
523 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
525 case IXGBE_TXDCTL(0):
526 for (i = 0; i < 64; i++)
527 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
530 pr_info("%-15s %08x\n", reginfo->name,
531 IXGBE_READ_REG(hw, reginfo->ofs));
535 for (i = 0; i < 8; i++) {
536 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
537 pr_err("%-15s", rname);
538 for (j = 0; j < 8; j++)
539 pr_cont(" %08x", regs[i*8+j]);
546 * ixgbe_dump - Print registers, tx-rings and rx-rings
548 static void ixgbe_dump(struct ixgbe_adapter *adapter)
550 struct net_device *netdev = adapter->netdev;
551 struct ixgbe_hw *hw = &adapter->hw;
552 struct ixgbe_reg_info *reginfo;
554 struct ixgbe_ring *tx_ring;
555 struct ixgbe_tx_buffer *tx_buffer;
556 union ixgbe_adv_tx_desc *tx_desc;
557 struct my_u0 { u64 a; u64 b; } *u0;
558 struct ixgbe_ring *rx_ring;
559 union ixgbe_adv_rx_desc *rx_desc;
560 struct ixgbe_rx_buffer *rx_buffer_info;
564 if (!netif_msg_hw(adapter))
567 /* Print netdevice Info */
569 dev_info(&adapter->pdev->dev, "Net device Info\n");
570 pr_info("Device Name state "
571 "trans_start last_rx\n");
572 pr_info("%-15s %016lX %016lX %016lX\n",
579 /* Print Registers */
580 dev_info(&adapter->pdev->dev, "Register Dump\n");
581 pr_info(" Register Name Value\n");
582 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
583 reginfo->name; reginfo++) {
584 ixgbe_regdump(hw, reginfo);
587 /* Print TX Ring Summary */
588 if (!netdev || !netif_running(netdev))
591 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
592 pr_info(" %s %s %s %s\n",
593 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
594 "leng", "ntw", "timestamp");
595 for (n = 0; n < adapter->num_tx_queues; n++) {
596 tx_ring = adapter->tx_ring[n];
597 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
598 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
599 n, tx_ring->next_to_use, tx_ring->next_to_clean,
600 (u64)dma_unmap_addr(tx_buffer, dma),
601 dma_unmap_len(tx_buffer, len),
602 tx_buffer->next_to_watch,
603 (u64)tx_buffer->time_stamp);
607 if (!netif_msg_tx_done(adapter))
608 goto rx_ring_summary;
610 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
612 /* Transmit Descriptor Formats
614 * 82598 Advanced Transmit Descriptor
615 * +--------------------------------------------------------------+
616 * 0 | Buffer Address [63:0] |
617 * +--------------------------------------------------------------+
618 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
619 * +--------------------------------------------------------------+
620 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
622 * 82598 Advanced Transmit Descriptor (Write-Back Format)
623 * +--------------------------------------------------------------+
625 * +--------------------------------------------------------------+
626 * 8 | RSV | STA | NXTSEQ |
627 * +--------------------------------------------------------------+
630 * 82599+ Advanced Transmit Descriptor
631 * +--------------------------------------------------------------+
632 * 0 | Buffer Address [63:0] |
633 * +--------------------------------------------------------------+
634 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
635 * +--------------------------------------------------------------+
636 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
638 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
639 * +--------------------------------------------------------------+
641 * +--------------------------------------------------------------+
642 * 8 | RSV | STA | RSV |
643 * +--------------------------------------------------------------+
647 for (n = 0; n < adapter->num_tx_queues; n++) {
648 tx_ring = adapter->tx_ring[n];
649 pr_info("------------------------------------\n");
650 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
651 pr_info("------------------------------------\n");
652 pr_info("%s%s %s %s %s %s\n",
653 "T [desc] [address 63:0 ] ",
654 "[PlPOIdStDDt Ln] [bi->dma ] ",
655 "leng", "ntw", "timestamp", "bi->skb");
657 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
658 tx_desc = IXGBE_TX_DESC(tx_ring, i);
659 tx_buffer = &tx_ring->tx_buffer_info[i];
660 u0 = (struct my_u0 *)tx_desc;
661 if (dma_unmap_len(tx_buffer, len) > 0) {
662 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p",
666 (u64)dma_unmap_addr(tx_buffer, dma),
667 dma_unmap_len(tx_buffer, len),
668 tx_buffer->next_to_watch,
669 (u64)tx_buffer->time_stamp,
671 if (i == tx_ring->next_to_use &&
672 i == tx_ring->next_to_clean)
674 else if (i == tx_ring->next_to_use)
676 else if (i == tx_ring->next_to_clean)
681 if (netif_msg_pktdata(adapter) &&
683 print_hex_dump(KERN_INFO, "",
684 DUMP_PREFIX_ADDRESS, 16, 1,
685 tx_buffer->skb->data,
686 dma_unmap_len(tx_buffer, len),
692 /* Print RX Rings Summary */
694 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
695 pr_info("Queue [NTU] [NTC]\n");
696 for (n = 0; n < adapter->num_rx_queues; n++) {
697 rx_ring = adapter->rx_ring[n];
698 pr_info("%5d %5X %5X\n",
699 n, rx_ring->next_to_use, rx_ring->next_to_clean);
703 if (!netif_msg_rx_status(adapter))
706 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
708 /* Receive Descriptor Formats
710 * 82598 Advanced Receive Descriptor (Read) Format
712 * +-----------------------------------------------------+
713 * 0 | Packet Buffer Address [63:1] |A0/NSE|
714 * +----------------------------------------------+------+
715 * 8 | Header Buffer Address [63:1] | DD |
716 * +-----------------------------------------------------+
719 * 82598 Advanced Receive Descriptor (Write-Back) Format
721 * 63 48 47 32 31 30 21 20 16 15 4 3 0
722 * +------------------------------------------------------+
723 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
724 * | Packet | IP | | | | Type | Type |
725 * | Checksum | Ident | | | | | |
726 * +------------------------------------------------------+
727 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
728 * +------------------------------------------------------+
729 * 63 48 47 32 31 20 19 0
731 * 82599+ Advanced Receive Descriptor (Read) Format
733 * +-----------------------------------------------------+
734 * 0 | Packet Buffer Address [63:1] |A0/NSE|
735 * +----------------------------------------------+------+
736 * 8 | Header Buffer Address [63:1] | DD |
737 * +-----------------------------------------------------+
740 * 82599+ Advanced Receive Descriptor (Write-Back) Format
742 * 63 48 47 32 31 30 21 20 17 16 4 3 0
743 * +------------------------------------------------------+
744 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
745 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
746 * |/ Flow Dir Flt ID | | | | | |
747 * +------------------------------------------------------+
748 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
749 * +------------------------------------------------------+
750 * 63 48 47 32 31 20 19 0
753 for (n = 0; n < adapter->num_rx_queues; n++) {
754 rx_ring = adapter->rx_ring[n];
755 pr_info("------------------------------------\n");
756 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
757 pr_info("------------------------------------\n");
759 "R [desc] [ PktBuf A0] ",
760 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
761 "<-- Adv Rx Read format\n");
763 "RWB[desc] [PcsmIpSHl PtRs] ",
764 "[vl er S cks ln] ---------------- [bi->skb ] ",
765 "<-- Adv Rx Write-Back format\n");
767 for (i = 0; i < rx_ring->count; i++) {
768 rx_buffer_info = &rx_ring->rx_buffer_info[i];
769 rx_desc = IXGBE_RX_DESC(rx_ring, i);
770 u0 = (struct my_u0 *)rx_desc;
771 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
772 if (staterr & IXGBE_RXD_STAT_DD) {
773 /* Descriptor Done */
774 pr_info("RWB[0x%03X] %016llX "
775 "%016llX ---------------- %p", i,
778 rx_buffer_info->skb);
780 pr_info("R [0x%03X] %016llX "
781 "%016llX %016llX %p", i,
784 (u64)rx_buffer_info->dma,
785 rx_buffer_info->skb);
787 if (netif_msg_pktdata(adapter) &&
788 rx_buffer_info->dma) {
789 print_hex_dump(KERN_INFO, "",
790 DUMP_PREFIX_ADDRESS, 16, 1,
791 page_address(rx_buffer_info->page) +
792 rx_buffer_info->page_offset,
793 ixgbe_rx_bufsz(rx_ring), true);
797 if (i == rx_ring->next_to_use)
799 else if (i == rx_ring->next_to_clean)
808 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
812 /* Let firmware take over control of h/w */
813 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
814 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
815 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
818 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
822 /* Let firmware know the driver has taken over */
823 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
824 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
825 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
829 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
830 * @adapter: pointer to adapter struct
831 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
832 * @queue: queue to map the corresponding interrupt to
833 * @msix_vector: the vector to map to the corresponding queue
836 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
837 u8 queue, u8 msix_vector)
840 struct ixgbe_hw *hw = &adapter->hw;
841 switch (hw->mac.type) {
842 case ixgbe_mac_82598EB:
843 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
846 index = (((direction * 64) + queue) >> 2) & 0x1F;
847 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
848 ivar &= ~(0xFF << (8 * (queue & 0x3)));
849 ivar |= (msix_vector << (8 * (queue & 0x3)));
850 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
852 case ixgbe_mac_82599EB:
855 case ixgbe_mac_X550EM_x:
856 if (direction == -1) {
858 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
859 index = ((queue & 1) * 8);
860 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
861 ivar &= ~(0xFF << index);
862 ivar |= (msix_vector << index);
863 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
866 /* tx or rx causes */
867 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
868 index = ((16 * (queue & 1)) + (8 * direction));
869 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
870 ivar &= ~(0xFF << index);
871 ivar |= (msix_vector << index);
872 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
880 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
885 switch (adapter->hw.mac.type) {
886 case ixgbe_mac_82598EB:
887 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
888 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
890 case ixgbe_mac_82599EB:
893 case ixgbe_mac_X550EM_x:
894 mask = (qmask & 0xFFFFFFFF);
895 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
896 mask = (qmask >> 32);
897 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
904 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
905 struct ixgbe_tx_buffer *tx_buffer)
907 if (tx_buffer->skb) {
908 dev_kfree_skb_any(tx_buffer->skb);
909 if (dma_unmap_len(tx_buffer, len))
910 dma_unmap_single(ring->dev,
911 dma_unmap_addr(tx_buffer, dma),
912 dma_unmap_len(tx_buffer, len),
914 } else if (dma_unmap_len(tx_buffer, len)) {
915 dma_unmap_page(ring->dev,
916 dma_unmap_addr(tx_buffer, dma),
917 dma_unmap_len(tx_buffer, len),
920 tx_buffer->next_to_watch = NULL;
921 tx_buffer->skb = NULL;
922 dma_unmap_len_set(tx_buffer, len, 0);
923 /* tx_buffer must be completely set up in the transmit path */
926 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
928 struct ixgbe_hw *hw = &adapter->hw;
929 struct ixgbe_hw_stats *hwstats = &adapter->stats;
933 if ((hw->fc.current_mode != ixgbe_fc_full) &&
934 (hw->fc.current_mode != ixgbe_fc_rx_pause))
937 switch (hw->mac.type) {
938 case ixgbe_mac_82598EB:
939 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
942 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
944 hwstats->lxoffrxc += data;
946 /* refill credits (no tx hang) if we received xoff */
950 for (i = 0; i < adapter->num_tx_queues; i++)
951 clear_bit(__IXGBE_HANG_CHECK_ARMED,
952 &adapter->tx_ring[i]->state);
955 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
957 struct ixgbe_hw *hw = &adapter->hw;
958 struct ixgbe_hw_stats *hwstats = &adapter->stats;
962 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
964 if (adapter->ixgbe_ieee_pfc)
965 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
967 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
968 ixgbe_update_xoff_rx_lfc(adapter);
972 /* update stats for each tc, only valid with PFC enabled */
973 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
976 switch (hw->mac.type) {
977 case ixgbe_mac_82598EB:
978 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
981 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
983 hwstats->pxoffrxc[i] += pxoffrxc;
984 /* Get the TC for given UP */
985 tc = netdev_get_prio_tc_map(adapter->netdev, i);
986 xoff[tc] += pxoffrxc;
989 /* disarm tx queues that have received xoff frames */
990 for (i = 0; i < adapter->num_tx_queues; i++) {
991 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
993 tc = tx_ring->dcb_tc;
995 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
999 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1001 return ring->stats.packets;
1004 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1006 struct ixgbe_adapter *adapter;
1007 struct ixgbe_hw *hw;
1010 if (ring->l2_accel_priv)
1011 adapter = ring->l2_accel_priv->real_adapter;
1013 adapter = netdev_priv(ring->netdev);
1016 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
1017 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
1020 return (head < tail) ?
1021 tail - head : (tail + ring->count - head);
1026 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1028 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1029 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1030 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1032 clear_check_for_tx_hang(tx_ring);
1035 * Check for a hung queue, but be thorough. This verifies
1036 * that a transmit has been completed since the previous
1037 * check AND there is at least one packet pending. The
1038 * ARMED bit is set to indicate a potential hang. The
1039 * bit is cleared if a pause frame is received to remove
1040 * false hang detection due to PFC or 802.3x frames. By
1041 * requiring this to fail twice we avoid races with
1042 * pfc clearing the ARMED bit and conditions where we
1043 * run the check_tx_hang logic with a transmit completion
1044 * pending but without time to complete it yet.
1046 if (tx_done_old == tx_done && tx_pending)
1047 /* make sure it is true for two checks in a row */
1048 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1050 /* update completed stats and continue */
1051 tx_ring->tx_stats.tx_done_old = tx_done;
1052 /* reset the countdown */
1053 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1059 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1060 * @adapter: driver private struct
1062 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1065 /* Do the reset outside of interrupt context */
1066 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1067 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
1068 e_warn(drv, "initiating reset due to tx timeout\n");
1069 ixgbe_service_event_schedule(adapter);
1074 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1075 * @q_vector: structure containing interrupt and ring information
1076 * @tx_ring: tx ring to clean
1078 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1079 struct ixgbe_ring *tx_ring)
1081 struct ixgbe_adapter *adapter = q_vector->adapter;
1082 struct ixgbe_tx_buffer *tx_buffer;
1083 union ixgbe_adv_tx_desc *tx_desc;
1084 unsigned int total_bytes = 0, total_packets = 0;
1085 unsigned int budget = q_vector->tx.work_limit;
1086 unsigned int i = tx_ring->next_to_clean;
1088 if (test_bit(__IXGBE_DOWN, &adapter->state))
1091 tx_buffer = &tx_ring->tx_buffer_info[i];
1092 tx_desc = IXGBE_TX_DESC(tx_ring, i);
1093 i -= tx_ring->count;
1096 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1098 /* if next_to_watch is not set then there is no work pending */
1102 /* prevent any other reads prior to eop_desc */
1103 read_barrier_depends();
1105 /* if DD is not set pending work has not been completed */
1106 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1109 /* clear next_to_watch to prevent false hangs */
1110 tx_buffer->next_to_watch = NULL;
1112 /* update the statistics for this packet */
1113 total_bytes += tx_buffer->bytecount;
1114 total_packets += tx_buffer->gso_segs;
1117 dev_consume_skb_any(tx_buffer->skb);
1119 /* unmap skb header data */
1120 dma_unmap_single(tx_ring->dev,
1121 dma_unmap_addr(tx_buffer, dma),
1122 dma_unmap_len(tx_buffer, len),
1125 /* clear tx_buffer data */
1126 tx_buffer->skb = NULL;
1127 dma_unmap_len_set(tx_buffer, len, 0);
1129 /* unmap remaining buffers */
1130 while (tx_desc != eop_desc) {
1135 i -= tx_ring->count;
1136 tx_buffer = tx_ring->tx_buffer_info;
1137 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1140 /* unmap any remaining paged data */
1141 if (dma_unmap_len(tx_buffer, len)) {
1142 dma_unmap_page(tx_ring->dev,
1143 dma_unmap_addr(tx_buffer, dma),
1144 dma_unmap_len(tx_buffer, len),
1146 dma_unmap_len_set(tx_buffer, len, 0);
1150 /* move us one more past the eop_desc for start of next pkt */
1155 i -= tx_ring->count;
1156 tx_buffer = tx_ring->tx_buffer_info;
1157 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1160 /* issue prefetch for next Tx descriptor */
1163 /* update budget accounting */
1165 } while (likely(budget));
1167 i += tx_ring->count;
1168 tx_ring->next_to_clean = i;
1169 u64_stats_update_begin(&tx_ring->syncp);
1170 tx_ring->stats.bytes += total_bytes;
1171 tx_ring->stats.packets += total_packets;
1172 u64_stats_update_end(&tx_ring->syncp);
1173 q_vector->tx.total_bytes += total_bytes;
1174 q_vector->tx.total_packets += total_packets;
1176 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1177 /* schedule immediate reset if we believe we hung */
1178 struct ixgbe_hw *hw = &adapter->hw;
1179 e_err(drv, "Detected Tx Unit Hang\n"
1181 " TDH, TDT <%x>, <%x>\n"
1182 " next_to_use <%x>\n"
1183 " next_to_clean <%x>\n"
1184 "tx_buffer_info[next_to_clean]\n"
1185 " time_stamp <%lx>\n"
1187 tx_ring->queue_index,
1188 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1189 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1190 tx_ring->next_to_use, i,
1191 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1193 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1196 "tx hang %d detected on queue %d, resetting adapter\n",
1197 adapter->tx_timeout_count + 1, tx_ring->queue_index);
1199 /* schedule immediate reset if we believe we hung */
1200 ixgbe_tx_timeout_reset(adapter);
1202 /* the adapter is about to reset, no point in enabling stuff */
1206 netdev_tx_completed_queue(txring_txq(tx_ring),
1207 total_packets, total_bytes);
1209 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1210 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1211 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1212 /* Make sure that anybody stopping the queue after this
1213 * sees the new next_to_clean.
1216 if (__netif_subqueue_stopped(tx_ring->netdev,
1217 tx_ring->queue_index)
1218 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1219 netif_wake_subqueue(tx_ring->netdev,
1220 tx_ring->queue_index);
1221 ++tx_ring->tx_stats.restart_queue;
1228 #ifdef CONFIG_IXGBE_DCA
1229 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1230 struct ixgbe_ring *tx_ring,
1233 struct ixgbe_hw *hw = &adapter->hw;
1234 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1237 switch (hw->mac.type) {
1238 case ixgbe_mac_82598EB:
1239 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1241 case ixgbe_mac_82599EB:
1242 case ixgbe_mac_X540:
1243 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1244 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1247 /* for unknown hardware do not write register */
1252 * We can enable relaxed ordering for reads, but not writes when
1253 * DCA is enabled. This is due to a known issue in some chipsets
1254 * which will cause the DCA tag to be cleared.
1256 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1257 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1258 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1260 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1263 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1264 struct ixgbe_ring *rx_ring,
1267 struct ixgbe_hw *hw = &adapter->hw;
1268 u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1269 u8 reg_idx = rx_ring->reg_idx;
1272 switch (hw->mac.type) {
1273 case ixgbe_mac_82599EB:
1274 case ixgbe_mac_X540:
1275 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1282 * We can enable relaxed ordering for reads, but not writes when
1283 * DCA is enabled. This is due to a known issue in some chipsets
1284 * which will cause the DCA tag to be cleared.
1286 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1287 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1289 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1292 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1294 struct ixgbe_adapter *adapter = q_vector->adapter;
1295 struct ixgbe_ring *ring;
1296 int cpu = get_cpu();
1298 if (q_vector->cpu == cpu)
1301 ixgbe_for_each_ring(ring, q_vector->tx)
1302 ixgbe_update_tx_dca(adapter, ring, cpu);
1304 ixgbe_for_each_ring(ring, q_vector->rx)
1305 ixgbe_update_rx_dca(adapter, ring, cpu);
1307 q_vector->cpu = cpu;
1312 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1316 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1319 /* always use CB2 mode, difference is masked in the CB driver */
1320 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1322 for (i = 0; i < adapter->num_q_vectors; i++) {
1323 adapter->q_vector[i]->cpu = -1;
1324 ixgbe_update_dca(adapter->q_vector[i]);
1328 static int __ixgbe_notify_dca(struct device *dev, void *data)
1330 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1331 unsigned long event = *(unsigned long *)data;
1333 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1337 case DCA_PROVIDER_ADD:
1338 /* if we're already enabled, don't do it again */
1339 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1341 if (dca_add_requester(dev) == 0) {
1342 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1343 ixgbe_setup_dca(adapter);
1346 /* Fall Through since DCA is disabled. */
1347 case DCA_PROVIDER_REMOVE:
1348 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1349 dca_remove_requester(dev);
1350 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1351 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1359 #endif /* CONFIG_IXGBE_DCA */
1360 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1361 union ixgbe_adv_rx_desc *rx_desc,
1362 struct sk_buff *skb)
1364 if (ring->netdev->features & NETIF_F_RXHASH)
1366 le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1372 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1373 * @ring: structure containing ring specific data
1374 * @rx_desc: advanced rx descriptor
1376 * Returns : true if it is FCoE pkt
1378 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1379 union ixgbe_adv_rx_desc *rx_desc)
1381 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1383 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1384 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1385 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1386 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1389 #endif /* IXGBE_FCOE */
1391 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1392 * @ring: structure containing ring specific data
1393 * @rx_desc: current Rx descriptor being processed
1394 * @skb: skb currently being received and modified
1396 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1397 union ixgbe_adv_rx_desc *rx_desc,
1398 struct sk_buff *skb)
1400 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1401 __le16 hdr_info = rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1402 bool encap_pkt = false;
1404 skb_checksum_none_assert(skb);
1406 /* Rx csum disabled */
1407 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1410 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) &&
1411 (hdr_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_TUNNEL >> 16))) {
1413 skb->encapsulation = 1;
1414 skb->ip_summed = CHECKSUM_NONE;
1417 /* if IP and error */
1418 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1419 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1420 ring->rx_stats.csum_err++;
1424 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1427 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1429 * 82599 errata, UDP frames with a 0 checksum can be marked as
1432 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1433 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1436 ring->rx_stats.csum_err++;
1440 /* It must be a TCP or UDP packet with a valid checksum */
1441 skb->ip_summed = CHECKSUM_UNNECESSARY;
1443 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1446 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1447 ring->rx_stats.csum_err++;
1450 /* If we checked the outer header let the stack know */
1451 skb->csum_level = 1;
1455 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1456 struct ixgbe_rx_buffer *bi)
1458 struct page *page = bi->page;
1461 /* since we are recycling buffers we should seldom need to alloc */
1465 /* alloc new page for storage */
1466 page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1467 if (unlikely(!page)) {
1468 rx_ring->rx_stats.alloc_rx_page_failed++;
1472 /* map page for use */
1473 dma = dma_map_page(rx_ring->dev, page, 0,
1474 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1477 * if mapping failed free memory back to system since
1478 * there isn't much point in holding memory we can't use
1480 if (dma_mapping_error(rx_ring->dev, dma)) {
1481 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1483 rx_ring->rx_stats.alloc_rx_page_failed++;
1489 bi->page_offset = 0;
1495 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1496 * @rx_ring: ring to place buffers on
1497 * @cleaned_count: number of buffers to replace
1499 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1501 union ixgbe_adv_rx_desc *rx_desc;
1502 struct ixgbe_rx_buffer *bi;
1503 u16 i = rx_ring->next_to_use;
1509 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1510 bi = &rx_ring->rx_buffer_info[i];
1511 i -= rx_ring->count;
1514 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1518 * Refresh the desc even if buffer_addrs didn't change
1519 * because each write-back erases this info.
1521 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1527 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1528 bi = rx_ring->rx_buffer_info;
1529 i -= rx_ring->count;
1532 /* clear the status bits for the next_to_use descriptor */
1533 rx_desc->wb.upper.status_error = 0;
1536 } while (cleaned_count);
1538 i += rx_ring->count;
1540 if (rx_ring->next_to_use != i) {
1541 rx_ring->next_to_use = i;
1543 /* update next to alloc since we have filled the ring */
1544 rx_ring->next_to_alloc = i;
1546 /* Force memory writes to complete before letting h/w
1547 * know there are new descriptors to fetch. (Only
1548 * applicable for weak-ordered memory model archs,
1552 writel(i, rx_ring->tail);
1556 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1557 struct sk_buff *skb)
1559 u16 hdr_len = skb_headlen(skb);
1561 /* set gso_size to avoid messing up TCP MSS */
1562 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1563 IXGBE_CB(skb)->append_cnt);
1564 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1567 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1568 struct sk_buff *skb)
1570 /* if append_cnt is 0 then frame is not RSC */
1571 if (!IXGBE_CB(skb)->append_cnt)
1574 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1575 rx_ring->rx_stats.rsc_flush++;
1577 ixgbe_set_rsc_gso_size(rx_ring, skb);
1579 /* gso_size is computed using append_cnt so always clear it last */
1580 IXGBE_CB(skb)->append_cnt = 0;
1584 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1585 * @rx_ring: rx descriptor ring packet is being transacted on
1586 * @rx_desc: pointer to the EOP Rx descriptor
1587 * @skb: pointer to current skb being populated
1589 * This function checks the ring, descriptor, and packet information in
1590 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1591 * other fields within the skb.
1593 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1594 union ixgbe_adv_rx_desc *rx_desc,
1595 struct sk_buff *skb)
1597 struct net_device *dev = rx_ring->netdev;
1599 ixgbe_update_rsc_stats(rx_ring, skb);
1601 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1603 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1605 if (unlikely(ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_TS)))
1606 ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector->adapter, skb);
1608 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1609 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1610 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1611 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1614 skb_record_rx_queue(skb, rx_ring->queue_index);
1616 skb->protocol = eth_type_trans(skb, dev);
1619 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1620 struct sk_buff *skb)
1622 if (ixgbe_qv_busy_polling(q_vector))
1623 netif_receive_skb(skb);
1625 napi_gro_receive(&q_vector->napi, skb);
1629 * ixgbe_is_non_eop - process handling of non-EOP buffers
1630 * @rx_ring: Rx ring being processed
1631 * @rx_desc: Rx descriptor for current buffer
1632 * @skb: Current socket buffer containing buffer in progress
1634 * This function updates next to clean. If the buffer is an EOP buffer
1635 * this function exits returning false, otherwise it will place the
1636 * sk_buff in the next buffer to be chained and return true indicating
1637 * that this is in fact a non-EOP buffer.
1639 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1640 union ixgbe_adv_rx_desc *rx_desc,
1641 struct sk_buff *skb)
1643 u32 ntc = rx_ring->next_to_clean + 1;
1645 /* fetch, update, and store next to clean */
1646 ntc = (ntc < rx_ring->count) ? ntc : 0;
1647 rx_ring->next_to_clean = ntc;
1649 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1651 /* update RSC append count if present */
1652 if (ring_is_rsc_enabled(rx_ring)) {
1653 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1654 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1656 if (unlikely(rsc_enabled)) {
1657 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1659 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1660 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1662 /* update ntc based on RSC value */
1663 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1664 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1665 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1669 /* if we are the last buffer then there is nothing else to do */
1670 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1673 /* place skb in next buffer to be received */
1674 rx_ring->rx_buffer_info[ntc].skb = skb;
1675 rx_ring->rx_stats.non_eop_descs++;
1681 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1682 * @rx_ring: rx descriptor ring packet is being transacted on
1683 * @skb: pointer to current skb being adjusted
1685 * This function is an ixgbe specific version of __pskb_pull_tail. The
1686 * main difference between this version and the original function is that
1687 * this function can make several assumptions about the state of things
1688 * that allow for significant optimizations versus the standard function.
1689 * As a result we can do things like drop a frag and maintain an accurate
1690 * truesize for the skb.
1692 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1693 struct sk_buff *skb)
1695 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1697 unsigned int pull_len;
1700 * it is valid to use page_address instead of kmap since we are
1701 * working with pages allocated out of the lomem pool per
1702 * alloc_page(GFP_ATOMIC)
1704 va = skb_frag_address(frag);
1707 * we need the header to contain the greater of either ETH_HLEN or
1708 * 60 bytes if the skb->len is less than 60 for skb_pad.
1710 pull_len = eth_get_headlen(va, IXGBE_RX_HDR_SIZE);
1712 /* align pull length to size of long to optimize memcpy performance */
1713 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1715 /* update all of the pointers */
1716 skb_frag_size_sub(frag, pull_len);
1717 frag->page_offset += pull_len;
1718 skb->data_len -= pull_len;
1719 skb->tail += pull_len;
1723 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1724 * @rx_ring: rx descriptor ring packet is being transacted on
1725 * @skb: pointer to current skb being updated
1727 * This function provides a basic DMA sync up for the first fragment of an
1728 * skb. The reason for doing this is that the first fragment cannot be
1729 * unmapped until we have reached the end of packet descriptor for a buffer
1732 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1733 struct sk_buff *skb)
1735 /* if the page was released unmap it, else just sync our portion */
1736 if (unlikely(IXGBE_CB(skb)->page_released)) {
1737 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1738 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1739 IXGBE_CB(skb)->page_released = false;
1741 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1743 dma_sync_single_range_for_cpu(rx_ring->dev,
1746 ixgbe_rx_bufsz(rx_ring),
1749 IXGBE_CB(skb)->dma = 0;
1753 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1754 * @rx_ring: rx descriptor ring packet is being transacted on
1755 * @rx_desc: pointer to the EOP Rx descriptor
1756 * @skb: pointer to current skb being fixed
1758 * Check for corrupted packet headers caused by senders on the local L2
1759 * embedded NIC switch not setting up their Tx Descriptors right. These
1760 * should be very rare.
1762 * Also address the case where we are pulling data in on pages only
1763 * and as such no data is present in the skb header.
1765 * In addition if skb is not at least 60 bytes we need to pad it so that
1766 * it is large enough to qualify as a valid Ethernet frame.
1768 * Returns true if an error was encountered and skb was freed.
1770 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1771 union ixgbe_adv_rx_desc *rx_desc,
1772 struct sk_buff *skb)
1774 struct net_device *netdev = rx_ring->netdev;
1776 /* verify that the packet does not have any known errors */
1777 if (unlikely(ixgbe_test_staterr(rx_desc,
1778 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1779 !(netdev->features & NETIF_F_RXALL))) {
1780 dev_kfree_skb_any(skb);
1784 /* place header in linear portion of buffer */
1785 if (skb_is_nonlinear(skb))
1786 ixgbe_pull_tail(rx_ring, skb);
1789 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1790 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1794 /* if eth_skb_pad returns an error the skb was freed */
1795 if (eth_skb_pad(skb))
1802 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1803 * @rx_ring: rx descriptor ring to store buffers on
1804 * @old_buff: donor buffer to have page reused
1806 * Synchronizes page for reuse by the adapter
1808 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1809 struct ixgbe_rx_buffer *old_buff)
1811 struct ixgbe_rx_buffer *new_buff;
1812 u16 nta = rx_ring->next_to_alloc;
1814 new_buff = &rx_ring->rx_buffer_info[nta];
1816 /* update, and store next to alloc */
1818 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1820 /* transfer page from old buffer to new buffer */
1821 *new_buff = *old_buff;
1823 /* sync the buffer for use by the device */
1824 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1825 new_buff->page_offset,
1826 ixgbe_rx_bufsz(rx_ring),
1830 static inline bool ixgbe_page_is_reserved(struct page *page)
1832 return (page_to_nid(page) != numa_mem_id()) || page->pfmemalloc;
1836 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1837 * @rx_ring: rx descriptor ring to transact packets on
1838 * @rx_buffer: buffer containing page to add
1839 * @rx_desc: descriptor containing length of buffer written by hardware
1840 * @skb: sk_buff to place the data into
1842 * This function will add the data contained in rx_buffer->page to the skb.
1843 * This is done either through a direct copy if the data in the buffer is
1844 * less than the skb header size, otherwise it will just attach the page as
1845 * a frag to the skb.
1847 * The function will then update the page offset if necessary and return
1848 * true if the buffer can be reused by the adapter.
1850 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1851 struct ixgbe_rx_buffer *rx_buffer,
1852 union ixgbe_adv_rx_desc *rx_desc,
1853 struct sk_buff *skb)
1855 struct page *page = rx_buffer->page;
1856 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1857 #if (PAGE_SIZE < 8192)
1858 unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1860 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1861 unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1862 ixgbe_rx_bufsz(rx_ring);
1865 if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1866 unsigned char *va = page_address(page) + rx_buffer->page_offset;
1868 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1870 /* page is not reserved, we can reuse buffer as-is */
1871 if (likely(!ixgbe_page_is_reserved(page)))
1874 /* this page cannot be reused so discard it */
1875 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1879 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1880 rx_buffer->page_offset, size, truesize);
1882 /* avoid re-using remote pages */
1883 if (unlikely(ixgbe_page_is_reserved(page)))
1886 #if (PAGE_SIZE < 8192)
1887 /* if we are only owner of page we can reuse it */
1888 if (unlikely(page_count(page) != 1))
1891 /* flip page offset to other buffer */
1892 rx_buffer->page_offset ^= truesize;
1894 /* move offset up to the next cache line */
1895 rx_buffer->page_offset += truesize;
1897 if (rx_buffer->page_offset > last_offset)
1901 /* Even if we own the page, we are not allowed to use atomic_set()
1902 * This would break get_page_unless_zero() users.
1904 atomic_inc(&page->_count);
1909 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1910 union ixgbe_adv_rx_desc *rx_desc)
1912 struct ixgbe_rx_buffer *rx_buffer;
1913 struct sk_buff *skb;
1916 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1917 page = rx_buffer->page;
1920 skb = rx_buffer->skb;
1923 void *page_addr = page_address(page) +
1924 rx_buffer->page_offset;
1926 /* prefetch first cache line of first page */
1927 prefetch(page_addr);
1928 #if L1_CACHE_BYTES < 128
1929 prefetch(page_addr + L1_CACHE_BYTES);
1932 /* allocate a skb to store the frags */
1933 skb = napi_alloc_skb(&rx_ring->q_vector->napi,
1935 if (unlikely(!skb)) {
1936 rx_ring->rx_stats.alloc_rx_buff_failed++;
1941 * we will be copying header into skb->data in
1942 * pskb_may_pull so it is in our interest to prefetch
1943 * it now to avoid a possible cache miss
1945 prefetchw(skb->data);
1948 * Delay unmapping of the first packet. It carries the
1949 * header information, HW may still access the header
1950 * after the writeback. Only unmap it when EOP is
1953 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1956 IXGBE_CB(skb)->dma = rx_buffer->dma;
1958 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
1959 ixgbe_dma_sync_frag(rx_ring, skb);
1962 /* we are reusing so sync this buffer for CPU use */
1963 dma_sync_single_range_for_cpu(rx_ring->dev,
1965 rx_buffer->page_offset,
1966 ixgbe_rx_bufsz(rx_ring),
1969 rx_buffer->skb = NULL;
1972 /* pull page into skb */
1973 if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
1974 /* hand second half of page back to the ring */
1975 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1976 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1977 /* the page has been released from the ring */
1978 IXGBE_CB(skb)->page_released = true;
1980 /* we are not reusing the buffer so unmap it */
1981 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1982 ixgbe_rx_pg_size(rx_ring),
1986 /* clear contents of buffer_info */
1987 rx_buffer->page = NULL;
1993 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1994 * @q_vector: structure containing interrupt and ring information
1995 * @rx_ring: rx descriptor ring to transact packets on
1996 * @budget: Total limit on number of packets to process
1998 * This function provides a "bounce buffer" approach to Rx interrupt
1999 * processing. The advantage to this is that on systems that have
2000 * expensive overhead for IOMMU access this provides a means of avoiding
2001 * it by maintaining the mapping of the page to the syste.
2003 * Returns amount of work completed
2005 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2006 struct ixgbe_ring *rx_ring,
2009 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2011 struct ixgbe_adapter *adapter = q_vector->adapter;
2013 unsigned int mss = 0;
2014 #endif /* IXGBE_FCOE */
2015 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2017 while (likely(total_rx_packets < budget)) {
2018 union ixgbe_adv_rx_desc *rx_desc;
2019 struct sk_buff *skb;
2021 /* return some buffers to hardware, one at a time is too slow */
2022 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2023 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2027 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2029 if (!rx_desc->wb.upper.status_error)
2032 /* This memory barrier is needed to keep us from reading
2033 * any other fields out of the rx_desc until we know the
2034 * descriptor has been written back
2038 /* retrieve a buffer from the ring */
2039 skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
2041 /* exit if we failed to retrieve a buffer */
2047 /* place incomplete frames back on ring for completion */
2048 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2051 /* verify the packet layout is correct */
2052 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2055 /* probably a little skewed due to removing CRC */
2056 total_rx_bytes += skb->len;
2058 /* populate checksum, timestamp, VLAN, and protocol */
2059 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2062 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2063 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2064 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2065 /* include DDPed FCoE data */
2066 if (ddp_bytes > 0) {
2068 mss = rx_ring->netdev->mtu -
2069 sizeof(struct fcoe_hdr) -
2070 sizeof(struct fc_frame_header) -
2071 sizeof(struct fcoe_crc_eof);
2075 total_rx_bytes += ddp_bytes;
2076 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2080 dev_kfree_skb_any(skb);
2085 #endif /* IXGBE_FCOE */
2086 skb_mark_napi_id(skb, &q_vector->napi);
2087 ixgbe_rx_skb(q_vector, skb);
2089 /* update budget accounting */
2093 u64_stats_update_begin(&rx_ring->syncp);
2094 rx_ring->stats.packets += total_rx_packets;
2095 rx_ring->stats.bytes += total_rx_bytes;
2096 u64_stats_update_end(&rx_ring->syncp);
2097 q_vector->rx.total_packets += total_rx_packets;
2098 q_vector->rx.total_bytes += total_rx_bytes;
2100 return total_rx_packets;
2103 #ifdef CONFIG_NET_RX_BUSY_POLL
2104 /* must be called with local_bh_disable()d */
2105 static int ixgbe_low_latency_recv(struct napi_struct *napi)
2107 struct ixgbe_q_vector *q_vector =
2108 container_of(napi, struct ixgbe_q_vector, napi);
2109 struct ixgbe_adapter *adapter = q_vector->adapter;
2110 struct ixgbe_ring *ring;
2113 if (test_bit(__IXGBE_DOWN, &adapter->state))
2114 return LL_FLUSH_FAILED;
2116 if (!ixgbe_qv_lock_poll(q_vector))
2117 return LL_FLUSH_BUSY;
2119 ixgbe_for_each_ring(ring, q_vector->rx) {
2120 found = ixgbe_clean_rx_irq(q_vector, ring, 4);
2121 #ifdef BP_EXTENDED_STATS
2123 ring->stats.cleaned += found;
2125 ring->stats.misses++;
2131 ixgbe_qv_unlock_poll(q_vector);
2135 #endif /* CONFIG_NET_RX_BUSY_POLL */
2138 * ixgbe_configure_msix - Configure MSI-X hardware
2139 * @adapter: board private structure
2141 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2144 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2146 struct ixgbe_q_vector *q_vector;
2150 /* Populate MSIX to EITR Select */
2151 if (adapter->num_vfs > 32) {
2152 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2153 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2157 * Populate the IVAR table and set the ITR values to the
2158 * corresponding register.
2160 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2161 struct ixgbe_ring *ring;
2162 q_vector = adapter->q_vector[v_idx];
2164 ixgbe_for_each_ring(ring, q_vector->rx)
2165 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2167 ixgbe_for_each_ring(ring, q_vector->tx)
2168 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2170 ixgbe_write_eitr(q_vector);
2173 switch (adapter->hw.mac.type) {
2174 case ixgbe_mac_82598EB:
2175 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2178 case ixgbe_mac_82599EB:
2179 case ixgbe_mac_X540:
2180 case ixgbe_mac_X550:
2181 case ixgbe_mac_X550EM_x:
2182 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2187 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2189 /* set up to autoclear timer, and the vectors */
2190 mask = IXGBE_EIMS_ENABLE_MASK;
2191 mask &= ~(IXGBE_EIMS_OTHER |
2192 IXGBE_EIMS_MAILBOX |
2195 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2198 enum latency_range {
2202 latency_invalid = 255
2206 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2207 * @q_vector: structure containing interrupt and ring information
2208 * @ring_container: structure containing ring performance data
2210 * Stores a new ITR value based on packets and byte
2211 * counts during the last interrupt. The advantage of per interrupt
2212 * computation is faster updates and more accurate ITR for the current
2213 * traffic pattern. Constants in this function were computed
2214 * based on theoretical maximum wire speed and thresholds were set based
2215 * on testing data as well as attempting to minimize response time
2216 * while increasing bulk throughput.
2217 * this functionality is controlled by the InterruptThrottleRate module
2218 * parameter (see ixgbe_param.c)
2220 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2221 struct ixgbe_ring_container *ring_container)
2223 int bytes = ring_container->total_bytes;
2224 int packets = ring_container->total_packets;
2227 u8 itr_setting = ring_container->itr;
2232 /* simple throttlerate management
2233 * 0-10MB/s lowest (100000 ints/s)
2234 * 10-20MB/s low (20000 ints/s)
2235 * 20-1249MB/s bulk (8000 ints/s)
2237 /* what was last interrupt timeslice? */
2238 timepassed_us = q_vector->itr >> 2;
2239 if (timepassed_us == 0)
2242 bytes_perint = bytes / timepassed_us; /* bytes/usec */
2244 switch (itr_setting) {
2245 case lowest_latency:
2246 if (bytes_perint > 10)
2247 itr_setting = low_latency;
2250 if (bytes_perint > 20)
2251 itr_setting = bulk_latency;
2252 else if (bytes_perint <= 10)
2253 itr_setting = lowest_latency;
2256 if (bytes_perint <= 20)
2257 itr_setting = low_latency;
2261 /* clear work counters since we have the values we need */
2262 ring_container->total_bytes = 0;
2263 ring_container->total_packets = 0;
2265 /* write updated itr to ring container */
2266 ring_container->itr = itr_setting;
2270 * ixgbe_write_eitr - write EITR register in hardware specific way
2271 * @q_vector: structure containing interrupt and ring information
2273 * This function is made to be called by ethtool and by the driver
2274 * when it needs to update EITR registers at runtime. Hardware
2275 * specific quirks/differences are taken care of here.
2277 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2279 struct ixgbe_adapter *adapter = q_vector->adapter;
2280 struct ixgbe_hw *hw = &adapter->hw;
2281 int v_idx = q_vector->v_idx;
2282 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2284 switch (adapter->hw.mac.type) {
2285 case ixgbe_mac_82598EB:
2286 /* must write high and low 16 bits to reset counter */
2287 itr_reg |= (itr_reg << 16);
2289 case ixgbe_mac_82599EB:
2290 case ixgbe_mac_X540:
2291 case ixgbe_mac_X550:
2292 case ixgbe_mac_X550EM_x:
2294 * set the WDIS bit to not clear the timer bits and cause an
2295 * immediate assertion of the interrupt
2297 itr_reg |= IXGBE_EITR_CNT_WDIS;
2302 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2305 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2307 u32 new_itr = q_vector->itr;
2310 ixgbe_update_itr(q_vector, &q_vector->tx);
2311 ixgbe_update_itr(q_vector, &q_vector->rx);
2313 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2315 switch (current_itr) {
2316 /* counts and packets in update_itr are dependent on these numbers */
2317 case lowest_latency:
2318 new_itr = IXGBE_100K_ITR;
2321 new_itr = IXGBE_20K_ITR;
2324 new_itr = IXGBE_8K_ITR;
2330 if (new_itr != q_vector->itr) {
2331 /* do an exponential smoothing */
2332 new_itr = (10 * new_itr * q_vector->itr) /
2333 ((9 * new_itr) + q_vector->itr);
2335 /* save the algorithm value here */
2336 q_vector->itr = new_itr;
2338 ixgbe_write_eitr(q_vector);
2343 * ixgbe_check_overtemp_subtask - check for over temperature
2344 * @adapter: pointer to adapter
2346 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2348 struct ixgbe_hw *hw = &adapter->hw;
2349 u32 eicr = adapter->interrupt_event;
2351 if (test_bit(__IXGBE_DOWN, &adapter->state))
2354 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2355 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2358 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2360 switch (hw->device_id) {
2361 case IXGBE_DEV_ID_82599_T3_LOM:
2363 * Since the warning interrupt is for both ports
2364 * we don't have to check if:
2365 * - This interrupt wasn't for our port.
2366 * - We may have missed the interrupt so always have to
2367 * check if we got a LSC
2369 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2370 !(eicr & IXGBE_EICR_LSC))
2373 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2375 bool link_up = false;
2377 hw->mac.ops.check_link(hw, &speed, &link_up, false);
2383 /* Check if this is not due to overtemp */
2384 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2389 if (!(eicr & IXGBE_EICR_GPI_SDP0))
2394 "Network adapter has been stopped because it has over heated. "
2395 "Restart the computer. If the problem persists, "
2396 "power off the system and replace the adapter\n");
2398 adapter->interrupt_event = 0;
2401 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2403 struct ixgbe_hw *hw = &adapter->hw;
2405 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2406 (eicr & IXGBE_EICR_GPI_SDP1)) {
2407 e_crit(probe, "Fan has stopped, replace the adapter\n");
2408 /* write to clear the interrupt */
2409 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2413 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2415 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2418 switch (adapter->hw.mac.type) {
2419 case ixgbe_mac_82599EB:
2421 * Need to check link state so complete overtemp check
2424 if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2425 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2426 adapter->interrupt_event = eicr;
2427 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2428 ixgbe_service_event_schedule(adapter);
2432 case ixgbe_mac_X540:
2433 if (!(eicr & IXGBE_EICR_TS))
2441 "Network adapter has been stopped because it has over heated. "
2442 "Restart the computer. If the problem persists, "
2443 "power off the system and replace the adapter\n");
2446 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2448 struct ixgbe_hw *hw = &adapter->hw;
2450 if (eicr & IXGBE_EICR_GPI_SDP2) {
2451 /* Clear the interrupt */
2452 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2453 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2454 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2455 ixgbe_service_event_schedule(adapter);
2459 if (eicr & IXGBE_EICR_GPI_SDP1) {
2460 /* Clear the interrupt */
2461 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2462 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2463 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2464 ixgbe_service_event_schedule(adapter);
2469 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2471 struct ixgbe_hw *hw = &adapter->hw;
2474 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2475 adapter->link_check_timeout = jiffies;
2476 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2477 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2478 IXGBE_WRITE_FLUSH(hw);
2479 ixgbe_service_event_schedule(adapter);
2483 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2487 struct ixgbe_hw *hw = &adapter->hw;
2489 switch (hw->mac.type) {
2490 case ixgbe_mac_82598EB:
2491 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2492 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2494 case ixgbe_mac_82599EB:
2495 case ixgbe_mac_X540:
2496 case ixgbe_mac_X550:
2497 case ixgbe_mac_X550EM_x:
2498 mask = (qmask & 0xFFFFFFFF);
2500 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2501 mask = (qmask >> 32);
2503 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2508 /* skip the flush */
2511 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2515 struct ixgbe_hw *hw = &adapter->hw;
2517 switch (hw->mac.type) {
2518 case ixgbe_mac_82598EB:
2519 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2520 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2522 case ixgbe_mac_82599EB:
2523 case ixgbe_mac_X540:
2524 case ixgbe_mac_X550:
2525 case ixgbe_mac_X550EM_x:
2526 mask = (qmask & 0xFFFFFFFF);
2528 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2529 mask = (qmask >> 32);
2531 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2536 /* skip the flush */
2540 * ixgbe_irq_enable - Enable default interrupt generation settings
2541 * @adapter: board private structure
2543 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2546 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2548 /* don't reenable LSC while waiting for link */
2549 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2550 mask &= ~IXGBE_EIMS_LSC;
2552 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2553 switch (adapter->hw.mac.type) {
2554 case ixgbe_mac_82599EB:
2555 mask |= IXGBE_EIMS_GPI_SDP0;
2557 case ixgbe_mac_X540:
2558 case ixgbe_mac_X550:
2559 case ixgbe_mac_X550EM_x:
2560 mask |= IXGBE_EIMS_TS;
2565 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2566 mask |= IXGBE_EIMS_GPI_SDP1;
2567 switch (adapter->hw.mac.type) {
2568 case ixgbe_mac_82599EB:
2569 mask |= IXGBE_EIMS_GPI_SDP1;
2570 mask |= IXGBE_EIMS_GPI_SDP2;
2572 case ixgbe_mac_X540:
2573 case ixgbe_mac_X550:
2574 case ixgbe_mac_X550EM_x:
2575 mask |= IXGBE_EIMS_ECC;
2576 mask |= IXGBE_EIMS_MAILBOX;
2582 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2583 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2584 mask |= IXGBE_EIMS_FLOW_DIR;
2586 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2588 ixgbe_irq_enable_queues(adapter, ~0);
2590 IXGBE_WRITE_FLUSH(&adapter->hw);
2593 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2595 struct ixgbe_adapter *adapter = data;
2596 struct ixgbe_hw *hw = &adapter->hw;
2600 * Workaround for Silicon errata. Use clear-by-write instead
2601 * of clear-by-read. Reading with EICS will return the
2602 * interrupt causes without clearing, which later be done
2603 * with the write to EICR.
2605 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2607 /* The lower 16bits of the EICR register are for the queue interrupts
2608 * which should be masked here in order to not accidentally clear them if
2609 * the bits are high when ixgbe_msix_other is called. There is a race
2610 * condition otherwise which results in possible performance loss
2611 * especially if the ixgbe_msix_other interrupt is triggering
2612 * consistently (as it would when PPS is turned on for the X540 device)
2616 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2618 if (eicr & IXGBE_EICR_LSC)
2619 ixgbe_check_lsc(adapter);
2621 if (eicr & IXGBE_EICR_MAILBOX)
2622 ixgbe_msg_task(adapter);
2624 switch (hw->mac.type) {
2625 case ixgbe_mac_82599EB:
2626 case ixgbe_mac_X540:
2627 case ixgbe_mac_X550:
2628 case ixgbe_mac_X550EM_x:
2629 if (eicr & IXGBE_EICR_ECC) {
2630 e_info(link, "Received ECC Err, initiating reset\n");
2631 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2632 ixgbe_service_event_schedule(adapter);
2633 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2635 /* Handle Flow Director Full threshold interrupt */
2636 if (eicr & IXGBE_EICR_FLOW_DIR) {
2637 int reinit_count = 0;
2639 for (i = 0; i < adapter->num_tx_queues; i++) {
2640 struct ixgbe_ring *ring = adapter->tx_ring[i];
2641 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2646 /* no more flow director interrupts until after init */
2647 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2648 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2649 ixgbe_service_event_schedule(adapter);
2652 ixgbe_check_sfp_event(adapter, eicr);
2653 ixgbe_check_overtemp_event(adapter, eicr);
2659 ixgbe_check_fan_failure(adapter, eicr);
2661 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2662 ixgbe_ptp_check_pps_event(adapter, eicr);
2664 /* re-enable the original interrupt state, no lsc, no queues */
2665 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2666 ixgbe_irq_enable(adapter, false, false);
2671 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2673 struct ixgbe_q_vector *q_vector = data;
2675 /* EIAM disabled interrupts (on this vector) for us */
2677 if (q_vector->rx.ring || q_vector->tx.ring)
2678 napi_schedule(&q_vector->napi);
2684 * ixgbe_poll - NAPI Rx polling callback
2685 * @napi: structure for representing this polling device
2686 * @budget: how many packets driver is allowed to clean
2688 * This function is used for legacy and MSI, NAPI mode
2690 int ixgbe_poll(struct napi_struct *napi, int budget)
2692 struct ixgbe_q_vector *q_vector =
2693 container_of(napi, struct ixgbe_q_vector, napi);
2694 struct ixgbe_adapter *adapter = q_vector->adapter;
2695 struct ixgbe_ring *ring;
2696 int per_ring_budget;
2697 bool clean_complete = true;
2699 #ifdef CONFIG_IXGBE_DCA
2700 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2701 ixgbe_update_dca(q_vector);
2704 ixgbe_for_each_ring(ring, q_vector->tx)
2705 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2707 if (!ixgbe_qv_lock_napi(q_vector))
2710 /* attempt to distribute budget to each queue fairly, but don't allow
2711 * the budget to go below 1 because we'll exit polling */
2712 if (q_vector->rx.count > 1)
2713 per_ring_budget = max(budget/q_vector->rx.count, 1);
2715 per_ring_budget = budget;
2717 ixgbe_for_each_ring(ring, q_vector->rx)
2718 clean_complete &= (ixgbe_clean_rx_irq(q_vector, ring,
2719 per_ring_budget) < per_ring_budget);
2721 ixgbe_qv_unlock_napi(q_vector);
2722 /* If all work not completed, return budget and keep polling */
2723 if (!clean_complete)
2726 /* all work done, exit the polling mode */
2727 napi_complete(napi);
2728 if (adapter->rx_itr_setting & 1)
2729 ixgbe_set_itr(q_vector);
2730 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2731 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2737 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2738 * @adapter: board private structure
2740 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2741 * interrupts from the kernel.
2743 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2745 struct net_device *netdev = adapter->netdev;
2749 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2750 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2751 struct msix_entry *entry = &adapter->msix_entries[vector];
2753 if (q_vector->tx.ring && q_vector->rx.ring) {
2754 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2755 "%s-%s-%d", netdev->name, "TxRx", ri++);
2757 } else if (q_vector->rx.ring) {
2758 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2759 "%s-%s-%d", netdev->name, "rx", ri++);
2760 } else if (q_vector->tx.ring) {
2761 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2762 "%s-%s-%d", netdev->name, "tx", ti++);
2764 /* skip this unused q_vector */
2767 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2768 q_vector->name, q_vector);
2770 e_err(probe, "request_irq failed for MSIX interrupt "
2771 "Error: %d\n", err);
2772 goto free_queue_irqs;
2774 /* If Flow Director is enabled, set interrupt affinity */
2775 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2776 /* assign the mask for this irq */
2777 irq_set_affinity_hint(entry->vector,
2778 &q_vector->affinity_mask);
2782 err = request_irq(adapter->msix_entries[vector].vector,
2783 ixgbe_msix_other, 0, netdev->name, adapter);
2785 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2786 goto free_queue_irqs;
2794 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2796 free_irq(adapter->msix_entries[vector].vector,
2797 adapter->q_vector[vector]);
2799 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2800 pci_disable_msix(adapter->pdev);
2801 kfree(adapter->msix_entries);
2802 adapter->msix_entries = NULL;
2807 * ixgbe_intr - legacy mode Interrupt Handler
2808 * @irq: interrupt number
2809 * @data: pointer to a network interface device structure
2811 static irqreturn_t ixgbe_intr(int irq, void *data)
2813 struct ixgbe_adapter *adapter = data;
2814 struct ixgbe_hw *hw = &adapter->hw;
2815 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2819 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2820 * before the read of EICR.
2822 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2824 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2825 * therefore no explicit interrupt disable is necessary */
2826 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2829 * shared interrupt alert!
2830 * make sure interrupts are enabled because the read will
2831 * have disabled interrupts due to EIAM
2832 * finish the workaround of silicon errata on 82598. Unmask
2833 * the interrupt that we masked before the EICR read.
2835 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2836 ixgbe_irq_enable(adapter, true, true);
2837 return IRQ_NONE; /* Not our interrupt */
2840 if (eicr & IXGBE_EICR_LSC)
2841 ixgbe_check_lsc(adapter);
2843 switch (hw->mac.type) {
2844 case ixgbe_mac_82599EB:
2845 ixgbe_check_sfp_event(adapter, eicr);
2847 case ixgbe_mac_X540:
2848 case ixgbe_mac_X550:
2849 case ixgbe_mac_X550EM_x:
2850 if (eicr & IXGBE_EICR_ECC) {
2851 e_info(link, "Received ECC Err, initiating reset\n");
2852 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2853 ixgbe_service_event_schedule(adapter);
2854 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2856 ixgbe_check_overtemp_event(adapter, eicr);
2862 ixgbe_check_fan_failure(adapter, eicr);
2863 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2864 ixgbe_ptp_check_pps_event(adapter, eicr);
2866 /* would disable interrupts here but EIAM disabled it */
2867 napi_schedule(&q_vector->napi);
2870 * re-enable link(maybe) and non-queue interrupts, no flush.
2871 * ixgbe_poll will re-enable the queue interrupts
2873 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2874 ixgbe_irq_enable(adapter, false, false);
2880 * ixgbe_request_irq - initialize interrupts
2881 * @adapter: board private structure
2883 * Attempts to configure interrupts using the best available
2884 * capabilities of the hardware and kernel.
2886 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2888 struct net_device *netdev = adapter->netdev;
2891 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2892 err = ixgbe_request_msix_irqs(adapter);
2893 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2894 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2895 netdev->name, adapter);
2897 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2898 netdev->name, adapter);
2901 e_err(probe, "request_irq failed, Error %d\n", err);
2906 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2910 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2911 free_irq(adapter->pdev->irq, adapter);
2915 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2916 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2917 struct msix_entry *entry = &adapter->msix_entries[vector];
2919 /* free only the irqs that were actually requested */
2920 if (!q_vector->rx.ring && !q_vector->tx.ring)
2923 /* clear the affinity_mask in the IRQ descriptor */
2924 irq_set_affinity_hint(entry->vector, NULL);
2926 free_irq(entry->vector, q_vector);
2929 free_irq(adapter->msix_entries[vector++].vector, adapter);
2933 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2934 * @adapter: board private structure
2936 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2938 switch (adapter->hw.mac.type) {
2939 case ixgbe_mac_82598EB:
2940 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2942 case ixgbe_mac_82599EB:
2943 case ixgbe_mac_X540:
2944 case ixgbe_mac_X550:
2945 case ixgbe_mac_X550EM_x:
2946 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2947 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2948 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2953 IXGBE_WRITE_FLUSH(&adapter->hw);
2954 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2957 for (vector = 0; vector < adapter->num_q_vectors; vector++)
2958 synchronize_irq(adapter->msix_entries[vector].vector);
2960 synchronize_irq(adapter->msix_entries[vector++].vector);
2962 synchronize_irq(adapter->pdev->irq);
2967 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2970 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2972 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2974 ixgbe_write_eitr(q_vector);
2976 ixgbe_set_ivar(adapter, 0, 0, 0);
2977 ixgbe_set_ivar(adapter, 1, 0, 0);
2979 e_info(hw, "Legacy interrupt IVAR setup done\n");
2983 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2984 * @adapter: board private structure
2985 * @ring: structure containing ring specific data
2987 * Configure the Tx descriptor ring after a reset.
2989 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2990 struct ixgbe_ring *ring)
2992 struct ixgbe_hw *hw = &adapter->hw;
2993 u64 tdba = ring->dma;
2995 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2996 u8 reg_idx = ring->reg_idx;
2998 /* disable queue to avoid issues while updating state */
2999 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3000 IXGBE_WRITE_FLUSH(hw);
3002 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3003 (tdba & DMA_BIT_MASK(32)));
3004 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3005 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3006 ring->count * sizeof(union ixgbe_adv_tx_desc));
3007 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3008 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3009 ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3012 * set WTHRESH to encourage burst writeback, it should not be set
3013 * higher than 1 when:
3014 * - ITR is 0 as it could cause false TX hangs
3015 * - ITR is set to > 100k int/sec and BQL is enabled
3017 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3018 * to or less than the number of on chip descriptors, which is
3021 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3022 txdctl |= (1 << 16); /* WTHRESH = 1 */
3024 txdctl |= (8 << 16); /* WTHRESH = 8 */
3027 * Setting PTHRESH to 32 both improves performance
3028 * and avoids a TX hang with DFP enabled
3030 txdctl |= (1 << 8) | /* HTHRESH = 1 */
3031 32; /* PTHRESH = 32 */
3033 /* reinitialize flowdirector state */
3034 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3035 ring->atr_sample_rate = adapter->atr_sample_rate;
3036 ring->atr_count = 0;
3037 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3039 ring->atr_sample_rate = 0;
3042 /* initialize XPS */
3043 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3044 struct ixgbe_q_vector *q_vector = ring->q_vector;
3047 netif_set_xps_queue(ring->netdev,
3048 &q_vector->affinity_mask,
3052 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3055 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3057 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3058 if (hw->mac.type == ixgbe_mac_82598EB &&
3059 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3062 /* poll to verify queue is enabled */
3064 usleep_range(1000, 2000);
3065 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3066 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3068 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
3071 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3073 struct ixgbe_hw *hw = &adapter->hw;
3075 u8 tcs = netdev_get_num_tc(adapter->netdev);
3077 if (hw->mac.type == ixgbe_mac_82598EB)
3080 /* disable the arbiter while setting MTQC */
3081 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3082 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3083 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3085 /* set transmit pool layout */
3086 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3087 mtqc = IXGBE_MTQC_VT_ENA;
3089 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3091 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3092 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3093 mtqc |= IXGBE_MTQC_32VF;
3095 mtqc |= IXGBE_MTQC_64VF;
3098 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3100 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3102 mtqc = IXGBE_MTQC_64Q_1PB;
3105 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3107 /* Enable Security TX Buffer IFG for multiple pb */
3109 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3110 sectx |= IXGBE_SECTX_DCB;
3111 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3114 /* re-enable the arbiter */
3115 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3116 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3120 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3121 * @adapter: board private structure
3123 * Configure the Tx unit of the MAC after a reset.
3125 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3127 struct ixgbe_hw *hw = &adapter->hw;
3131 ixgbe_setup_mtqc(adapter);
3133 if (hw->mac.type != ixgbe_mac_82598EB) {
3134 /* DMATXCTL.EN must be before Tx queues are enabled */
3135 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3136 dmatxctl |= IXGBE_DMATXCTL_TE;
3137 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3140 /* Setup the HW Tx Head and Tail descriptor pointers */
3141 for (i = 0; i < adapter->num_tx_queues; i++)
3142 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3145 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3146 struct ixgbe_ring *ring)
3148 struct ixgbe_hw *hw = &adapter->hw;
3149 u8 reg_idx = ring->reg_idx;
3150 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3152 srrctl |= IXGBE_SRRCTL_DROP_EN;
3154 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3157 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3158 struct ixgbe_ring *ring)
3160 struct ixgbe_hw *hw = &adapter->hw;
3161 u8 reg_idx = ring->reg_idx;
3162 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3164 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3166 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3169 #ifdef CONFIG_IXGBE_DCB
3170 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3172 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3176 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3178 if (adapter->ixgbe_ieee_pfc)
3179 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3182 * We should set the drop enable bit if:
3185 * Number of Rx queues > 1 and flow control is disabled
3187 * This allows us to avoid head of line blocking for security
3188 * and performance reasons.
3190 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3191 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3192 for (i = 0; i < adapter->num_rx_queues; i++)
3193 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3195 for (i = 0; i < adapter->num_rx_queues; i++)
3196 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3200 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3202 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3203 struct ixgbe_ring *rx_ring)
3205 struct ixgbe_hw *hw = &adapter->hw;
3207 u8 reg_idx = rx_ring->reg_idx;
3209 if (hw->mac.type == ixgbe_mac_82598EB) {
3210 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3213 * if VMDq is not active we must program one srrctl register
3214 * per RSS queue since we have enabled RDRXCTL.MVMEN
3219 /* configure header buffer length, needed for RSC */
3220 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3222 /* configure the packet buffer length */
3223 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3225 /* configure descriptor type */
3226 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3228 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3232 * Return a number of entries in the RSS indirection table
3234 * @adapter: device handle
3236 * - 82598/82599/X540: 128
3237 * - X550(non-SRIOV mode): 512
3238 * - X550(SRIOV mode): 64
3240 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3242 if (adapter->hw.mac.type < ixgbe_mac_X550)
3244 else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3251 * Write the RETA table to HW
3253 * @adapter: device handle
3255 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3257 static void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3259 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3260 struct ixgbe_hw *hw = &adapter->hw;
3263 u8 *indir_tbl = adapter->rss_indir_tbl;
3265 /* Fill out the redirection table as follows:
3266 * - 82598: 8 bit wide entries containing pair of 4 bit RSS
3268 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3269 * - X550: 8 bit wide entries containing 6 bit RSS index
3271 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3272 indices_multi = 0x11;
3274 indices_multi = 0x1;
3276 /* Write redirection table to HW */
3277 for (i = 0; i < reta_entries; i++) {
3278 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3281 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3283 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3291 * Write the RETA table to HW (for x550 devices in SRIOV mode)
3293 * @adapter: device handle
3295 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3297 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3299 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3300 struct ixgbe_hw *hw = &adapter->hw;
3302 unsigned int pf_pool = adapter->num_vfs;
3304 /* Write redirection table to HW */
3305 for (i = 0; i < reta_entries; i++) {
3306 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3308 IXGBE_WRITE_REG(hw, IXGBE_PFVFRETA(i >> 2, pf_pool),
3315 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3317 struct ixgbe_hw *hw = &adapter->hw;
3319 u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3320 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3322 /* Program table for at least 2 queues w/ SR-IOV so that VFs can
3323 * make full use of any rings they may have. We will use the
3324 * PSRTYPE register to control how many rings we use within the PF.
3326 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
3329 /* Fill out hash function seeds */
3330 for (i = 0; i < 10; i++)
3331 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3333 /* Fill out redirection table */
3334 memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3336 for (i = 0, j = 0; i < reta_entries; i++, j++) {
3340 adapter->rss_indir_tbl[i] = j;
3343 ixgbe_store_reta(adapter);
3346 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3348 struct ixgbe_hw *hw = &adapter->hw;
3349 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3350 unsigned int pf_pool = adapter->num_vfs;
3353 /* Fill out hash function seeds */
3354 for (i = 0; i < 10; i++)
3355 IXGBE_WRITE_REG(hw, IXGBE_PFVFRSSRK(i, pf_pool),
3356 adapter->rss_key[i]);
3358 /* Fill out the redirection table */
3359 for (i = 0, j = 0; i < 64; i++, j++) {
3363 adapter->rss_indir_tbl[i] = j;
3366 ixgbe_store_vfreta(adapter);
3369 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3371 struct ixgbe_hw *hw = &adapter->hw;
3372 u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3375 /* Disable indicating checksum in descriptor, enables RSS hash */
3376 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3377 rxcsum |= IXGBE_RXCSUM_PCSD;
3378 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3380 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3381 if (adapter->ring_feature[RING_F_RSS].mask)
3382 mrqc = IXGBE_MRQC_RSSEN;
3384 u8 tcs = netdev_get_num_tc(adapter->netdev);
3386 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3388 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3390 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3391 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3392 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3394 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3397 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3399 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3401 mrqc = IXGBE_MRQC_RSSEN;
3405 /* Perform hash on these packet types */
3406 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3407 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3408 IXGBE_MRQC_RSS_FIELD_IPV6 |
3409 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3411 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3412 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3413 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3414 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3416 netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
3417 if ((hw->mac.type >= ixgbe_mac_X550) &&
3418 (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3419 unsigned int pf_pool = adapter->num_vfs;
3421 /* Enable VF RSS mode */
3422 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3423 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3425 /* Setup RSS through the VF registers */
3426 ixgbe_setup_vfreta(adapter);
3427 vfmrqc = IXGBE_MRQC_RSSEN;
3428 vfmrqc |= rss_field;
3429 IXGBE_WRITE_REG(hw, IXGBE_PFVFMRQC(pf_pool), vfmrqc);
3431 ixgbe_setup_reta(adapter);
3433 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3438 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3439 * @adapter: address of board private structure
3440 * @index: index of ring to set
3442 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3443 struct ixgbe_ring *ring)
3445 struct ixgbe_hw *hw = &adapter->hw;
3447 u8 reg_idx = ring->reg_idx;
3449 if (!ring_is_rsc_enabled(ring))
3452 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3453 rscctrl |= IXGBE_RSCCTL_RSCEN;
3455 * we must limit the number of descriptors so that the
3456 * total size of max desc * buf_len is not greater
3459 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3460 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3463 #define IXGBE_MAX_RX_DESC_POLL 10
3464 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3465 struct ixgbe_ring *ring)
3467 struct ixgbe_hw *hw = &adapter->hw;
3468 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3470 u8 reg_idx = ring->reg_idx;
3472 if (ixgbe_removed(hw->hw_addr))
3474 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3475 if (hw->mac.type == ixgbe_mac_82598EB &&
3476 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3480 usleep_range(1000, 2000);
3481 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3482 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3485 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3486 "the polling period\n", reg_idx);
3490 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3491 struct ixgbe_ring *ring)
3493 struct ixgbe_hw *hw = &adapter->hw;
3494 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3496 u8 reg_idx = ring->reg_idx;
3498 if (ixgbe_removed(hw->hw_addr))
3500 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3501 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3503 /* write value back with RXDCTL.ENABLE bit cleared */
3504 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3506 if (hw->mac.type == ixgbe_mac_82598EB &&
3507 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3510 /* the hardware may take up to 100us to really disable the rx queue */
3513 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3514 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3517 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3518 "the polling period\n", reg_idx);
3522 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3523 struct ixgbe_ring *ring)
3525 struct ixgbe_hw *hw = &adapter->hw;
3526 u64 rdba = ring->dma;
3528 u8 reg_idx = ring->reg_idx;
3530 /* disable queue to avoid issues while updating state */
3531 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3532 ixgbe_disable_rx_queue(adapter, ring);
3534 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3535 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3536 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3537 ring->count * sizeof(union ixgbe_adv_rx_desc));
3538 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3539 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3540 ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
3542 ixgbe_configure_srrctl(adapter, ring);
3543 ixgbe_configure_rscctl(adapter, ring);
3545 if (hw->mac.type == ixgbe_mac_82598EB) {
3547 * enable cache line friendly hardware writes:
3548 * PTHRESH=32 descriptors (half the internal cache),
3549 * this also removes ugly rx_no_buffer_count increment
3550 * HTHRESH=4 descriptors (to minimize latency on fetch)
3551 * WTHRESH=8 burst writeback up to two cache lines
3553 rxdctl &= ~0x3FFFFF;
3557 /* enable receive descriptor ring */
3558 rxdctl |= IXGBE_RXDCTL_ENABLE;
3559 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3561 ixgbe_rx_desc_queue_enable(adapter, ring);
3562 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3565 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3567 struct ixgbe_hw *hw = &adapter->hw;
3568 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3571 /* PSRTYPE must be initialized in non 82598 adapters */
3572 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3573 IXGBE_PSRTYPE_UDPHDR |
3574 IXGBE_PSRTYPE_IPV4HDR |
3575 IXGBE_PSRTYPE_L2HDR |
3576 IXGBE_PSRTYPE_IPV6HDR;
3578 if (hw->mac.type == ixgbe_mac_82598EB)
3586 for_each_set_bit(pool, &adapter->fwd_bitmask, 32)
3587 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
3590 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3592 struct ixgbe_hw *hw = &adapter->hw;
3593 u32 reg_offset, vf_shift;
3594 u32 gcr_ext, vmdctl;
3597 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3600 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3601 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3602 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3603 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3604 vmdctl |= IXGBE_VT_CTL_REPLEN;
3605 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3607 vf_shift = VMDQ_P(0) % 32;
3608 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3610 /* Enable only the PF's pool for Tx/Rx */
3611 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3612 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3613 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3614 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3615 if (adapter->bridge_mode == BRIDGE_MODE_VEB)
3616 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3618 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3619 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3622 * Set up VF register offsets for selected VT Mode,
3623 * i.e. 32 or 64 VFs for SR-IOV
3625 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3626 case IXGBE_82599_VMDQ_8Q_MASK:
3627 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3629 case IXGBE_82599_VMDQ_4Q_MASK:
3630 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3633 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3637 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3640 /* Enable MAC Anti-Spoofing */
3641 hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3644 /* Ensure LLDP is set for Ethertype Antispoofing if we will be
3645 * calling set_ethertype_anti_spoofing for each VF in loop below
3647 if (hw->mac.ops.set_ethertype_anti_spoofing)
3648 IXGBE_WRITE_REG(hw, IXGBE_ETQF(IXGBE_ETQF_FILTER_LLDP),
3649 (IXGBE_ETQF_FILTER_EN | /* enable filter */
3650 IXGBE_ETQF_TX_ANTISPOOF | /* tx antispoof */
3651 IXGBE_ETH_P_LLDP)); /* LLDP eth type */
3653 /* For VFs that have spoof checking turned off */
3654 for (i = 0; i < adapter->num_vfs; i++) {
3655 if (!adapter->vfinfo[i].spoofchk_enabled)
3656 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3658 /* enable ethertype anti spoofing if hw supports it */
3659 if (hw->mac.ops.set_ethertype_anti_spoofing)
3660 hw->mac.ops.set_ethertype_anti_spoofing(hw, true, i);
3662 /* Enable/Disable RSS query feature */
3663 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
3664 adapter->vfinfo[i].rss_query_enabled);
3668 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3670 struct ixgbe_hw *hw = &adapter->hw;
3671 struct net_device *netdev = adapter->netdev;
3672 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3673 struct ixgbe_ring *rx_ring;
3678 /* adjust max frame to be able to do baby jumbo for FCoE */
3679 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3680 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3681 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3683 #endif /* IXGBE_FCOE */
3685 /* adjust max frame to be at least the size of a standard frame */
3686 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3687 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3689 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3690 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3691 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3692 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3694 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3697 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3698 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3699 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3700 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3703 * Setup the HW Rx Head and Tail Descriptor Pointers and
3704 * the Base and Length of the Rx Descriptor Ring
3706 for (i = 0; i < adapter->num_rx_queues; i++) {
3707 rx_ring = adapter->rx_ring[i];
3708 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3709 set_ring_rsc_enabled(rx_ring);
3711 clear_ring_rsc_enabled(rx_ring);
3715 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3717 struct ixgbe_hw *hw = &adapter->hw;
3718 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3720 switch (hw->mac.type) {
3721 case ixgbe_mac_X550:
3722 case ixgbe_mac_X550EM_x:
3723 case ixgbe_mac_82598EB:
3725 * For VMDq support of different descriptor types or
3726 * buffer sizes through the use of multiple SRRCTL
3727 * registers, RDRXCTL.MVMEN must be set to 1
3729 * also, the manual doesn't mention it clearly but DCA hints
3730 * will only use queue 0's tags unless this bit is set. Side
3731 * effects of setting this bit are only that SRRCTL must be
3732 * fully programmed [0..15]
3734 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3736 case ixgbe_mac_82599EB:
3737 case ixgbe_mac_X540:
3738 /* Disable RSC for ACK packets */
3739 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3740 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3741 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3742 /* hardware requires some bits to be set by default */
3743 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3744 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3747 /* We should do nothing since we don't know this hardware */
3751 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3755 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3756 * @adapter: board private structure
3758 * Configure the Rx unit of the MAC after a reset.
3760 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3762 struct ixgbe_hw *hw = &adapter->hw;
3766 /* disable receives while setting up the descriptors */
3767 hw->mac.ops.disable_rx(hw);
3769 ixgbe_setup_psrtype(adapter);
3770 ixgbe_setup_rdrxctl(adapter);
3773 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
3774 rfctl &= ~IXGBE_RFCTL_RSC_DIS;
3775 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
3776 rfctl |= IXGBE_RFCTL_RSC_DIS;
3777 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
3779 /* Program registers for the distribution of queues */
3780 ixgbe_setup_mrqc(adapter);
3782 /* set_rx_buffer_len must be called before ring initialization */
3783 ixgbe_set_rx_buffer_len(adapter);
3786 * Setup the HW Rx Head and Tail Descriptor Pointers and
3787 * the Base and Length of the Rx Descriptor Ring
3789 for (i = 0; i < adapter->num_rx_queues; i++)
3790 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3792 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3793 /* disable drop enable for 82598 parts */
3794 if (hw->mac.type == ixgbe_mac_82598EB)
3795 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3797 /* enable all receives */
3798 rxctrl |= IXGBE_RXCTRL_RXEN;
3799 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3802 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
3803 __be16 proto, u16 vid)
3805 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3806 struct ixgbe_hw *hw = &adapter->hw;
3808 /* add VID to filter table */
3809 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3810 set_bit(vid, adapter->active_vlans);
3815 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
3816 __be16 proto, u16 vid)
3818 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3819 struct ixgbe_hw *hw = &adapter->hw;
3821 /* remove VID from filter table */
3822 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3823 clear_bit(vid, adapter->active_vlans);
3829 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3830 * @adapter: driver data
3832 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3834 struct ixgbe_hw *hw = &adapter->hw;
3838 switch (hw->mac.type) {
3839 case ixgbe_mac_82598EB:
3840 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3841 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3842 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3844 case ixgbe_mac_82599EB:
3845 case ixgbe_mac_X540:
3846 case ixgbe_mac_X550:
3847 case ixgbe_mac_X550EM_x:
3848 for (i = 0; i < adapter->num_rx_queues; i++) {
3849 struct ixgbe_ring *ring = adapter->rx_ring[i];
3851 if (ring->l2_accel_priv)
3854 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3855 vlnctrl &= ~IXGBE_RXDCTL_VME;
3856 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3865 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3866 * @adapter: driver data
3868 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3870 struct ixgbe_hw *hw = &adapter->hw;
3874 switch (hw->mac.type) {
3875 case ixgbe_mac_82598EB:
3876 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3877 vlnctrl |= IXGBE_VLNCTRL_VME;
3878 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3880 case ixgbe_mac_82599EB:
3881 case ixgbe_mac_X540:
3882 case ixgbe_mac_X550:
3883 case ixgbe_mac_X550EM_x:
3884 for (i = 0; i < adapter->num_rx_queues; i++) {
3885 struct ixgbe_ring *ring = adapter->rx_ring[i];
3887 if (ring->l2_accel_priv)
3890 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3891 vlnctrl |= IXGBE_RXDCTL_VME;
3892 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3900 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3904 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
3906 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3907 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
3911 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
3912 * @netdev: network interface device structure
3914 * Writes multicast address list to the MTA hash table.
3915 * Returns: -ENOMEM on failure
3916 * 0 on no addresses written
3917 * X on writing X addresses to MTA
3919 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
3921 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3922 struct ixgbe_hw *hw = &adapter->hw;
3924 if (!netif_running(netdev))
3927 if (hw->mac.ops.update_mc_addr_list)
3928 hw->mac.ops.update_mc_addr_list(hw, netdev);
3932 #ifdef CONFIG_PCI_IOV
3933 ixgbe_restore_vf_multicasts(adapter);
3936 return netdev_mc_count(netdev);
3939 #ifdef CONFIG_PCI_IOV
3940 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
3942 struct ixgbe_hw *hw = &adapter->hw;
3944 for (i = 0; i < hw->mac.num_rar_entries; i++) {
3945 if (adapter->mac_table[i].state & IXGBE_MAC_STATE_IN_USE)
3946 hw->mac.ops.set_rar(hw, i, adapter->mac_table[i].addr,
3947 adapter->mac_table[i].queue,
3950 hw->mac.ops.clear_rar(hw, i);
3952 adapter->mac_table[i].state &= ~(IXGBE_MAC_STATE_MODIFIED);
3957 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
3959 struct ixgbe_hw *hw = &adapter->hw;
3961 for (i = 0; i < hw->mac.num_rar_entries; i++) {
3962 if (adapter->mac_table[i].state & IXGBE_MAC_STATE_MODIFIED) {
3963 if (adapter->mac_table[i].state &
3964 IXGBE_MAC_STATE_IN_USE)
3965 hw->mac.ops.set_rar(hw, i,
3966 adapter->mac_table[i].addr,
3967 adapter->mac_table[i].queue,
3970 hw->mac.ops.clear_rar(hw, i);
3972 adapter->mac_table[i].state &=
3973 ~(IXGBE_MAC_STATE_MODIFIED);
3978 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
3981 struct ixgbe_hw *hw = &adapter->hw;
3983 for (i = 0; i < hw->mac.num_rar_entries; i++) {
3984 adapter->mac_table[i].state |= IXGBE_MAC_STATE_MODIFIED;
3985 adapter->mac_table[i].state &= ~IXGBE_MAC_STATE_IN_USE;
3986 eth_zero_addr(adapter->mac_table[i].addr);
3987 adapter->mac_table[i].queue = 0;
3989 ixgbe_sync_mac_table(adapter);
3992 static int ixgbe_available_rars(struct ixgbe_adapter *adapter)
3994 struct ixgbe_hw *hw = &adapter->hw;
3997 for (i = 0; i < hw->mac.num_rar_entries; i++) {
3998 if (adapter->mac_table[i].state == 0)
4004 /* this function destroys the first RAR entry */
4005 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter,
4008 struct ixgbe_hw *hw = &adapter->hw;
4010 memcpy(&adapter->mac_table[0].addr, addr, ETH_ALEN);
4011 adapter->mac_table[0].queue = VMDQ_P(0);
4012 adapter->mac_table[0].state = (IXGBE_MAC_STATE_DEFAULT |
4013 IXGBE_MAC_STATE_IN_USE);
4014 hw->mac.ops.set_rar(hw, 0, adapter->mac_table[0].addr,
4015 adapter->mac_table[0].queue,
4019 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter, u8 *addr, u16 queue)
4021 struct ixgbe_hw *hw = &adapter->hw;
4024 if (is_zero_ether_addr(addr))
4027 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4028 if (adapter->mac_table[i].state & IXGBE_MAC_STATE_IN_USE)
4030 adapter->mac_table[i].state |= (IXGBE_MAC_STATE_MODIFIED |
4031 IXGBE_MAC_STATE_IN_USE);
4032 ether_addr_copy(adapter->mac_table[i].addr, addr);
4033 adapter->mac_table[i].queue = queue;
4034 ixgbe_sync_mac_table(adapter);
4040 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter, u8 *addr, u16 queue)
4042 /* search table for addr, if found, set to 0 and sync */
4044 struct ixgbe_hw *hw = &adapter->hw;
4046 if (is_zero_ether_addr(addr))
4049 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4050 if (ether_addr_equal(addr, adapter->mac_table[i].addr) &&
4051 adapter->mac_table[i].queue == queue) {
4052 adapter->mac_table[i].state |= IXGBE_MAC_STATE_MODIFIED;
4053 adapter->mac_table[i].state &= ~IXGBE_MAC_STATE_IN_USE;
4054 eth_zero_addr(adapter->mac_table[i].addr);
4055 adapter->mac_table[i].queue = 0;
4056 ixgbe_sync_mac_table(adapter);
4063 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
4064 * @netdev: network interface device structure
4066 * Writes unicast address list to the RAR table.
4067 * Returns: -ENOMEM on failure/insufficient address space
4068 * 0 on no addresses written
4069 * X on writing X addresses to the RAR table
4071 static int ixgbe_write_uc_addr_list(struct net_device *netdev, int vfn)
4073 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4076 /* return ENOMEM indicating insufficient memory for addresses */
4077 if (netdev_uc_count(netdev) > ixgbe_available_rars(adapter))
4080 if (!netdev_uc_empty(netdev)) {
4081 struct netdev_hw_addr *ha;
4082 netdev_for_each_uc_addr(ha, netdev) {
4083 ixgbe_del_mac_filter(adapter, ha->addr, vfn);
4084 ixgbe_add_mac_filter(adapter, ha->addr, vfn);
4092 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4093 * @netdev: network interface device structure
4095 * The set_rx_method entry point is called whenever the unicast/multicast
4096 * address list or the network interface flags are updated. This routine is
4097 * responsible for configuring the hardware for proper unicast, multicast and
4100 void ixgbe_set_rx_mode(struct net_device *netdev)
4102 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4103 struct ixgbe_hw *hw = &adapter->hw;
4104 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4108 /* Check for Promiscuous and All Multicast modes */
4109 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4110 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4112 /* set all bits that we expect to always be set */
4113 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4114 fctrl |= IXGBE_FCTRL_BAM;
4115 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4116 fctrl |= IXGBE_FCTRL_PMCF;
4118 /* clear the bits we are changing the status of */
4119 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4120 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
4121 if (netdev->flags & IFF_PROMISC) {
4122 hw->addr_ctrl.user_set_promisc = true;
4123 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4124 vmolr |= IXGBE_VMOLR_MPE;
4125 /* Only disable hardware filter vlans in promiscuous mode
4126 * if SR-IOV and VMDQ are disabled - otherwise ensure
4127 * that hardware VLAN filters remain enabled.
4129 if (adapter->flags & (IXGBE_FLAG_VMDQ_ENABLED |
4130 IXGBE_FLAG_SRIOV_ENABLED))
4131 vlnctrl |= (IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
4133 if (netdev->flags & IFF_ALLMULTI) {
4134 fctrl |= IXGBE_FCTRL_MPE;
4135 vmolr |= IXGBE_VMOLR_MPE;
4137 vlnctrl |= IXGBE_VLNCTRL_VFE;
4138 hw->addr_ctrl.user_set_promisc = false;
4142 * Write addresses to available RAR registers, if there is not
4143 * sufficient space to store all the addresses then enable
4144 * unicast promiscuous mode
4146 count = ixgbe_write_uc_addr_list(netdev, VMDQ_P(0));
4148 fctrl |= IXGBE_FCTRL_UPE;
4149 vmolr |= IXGBE_VMOLR_ROPE;
4152 /* Write addresses to the MTA, if the attempt fails
4153 * then we should just turn on promiscuous mode so
4154 * that we can at least receive multicast traffic
4156 count = ixgbe_write_mc_addr_list(netdev);
4158 fctrl |= IXGBE_FCTRL_MPE;
4159 vmolr |= IXGBE_VMOLR_MPE;
4161 vmolr |= IXGBE_VMOLR_ROMPE;
4164 if (hw->mac.type != ixgbe_mac_82598EB) {
4165 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4166 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4168 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4171 /* This is useful for sniffing bad packets. */
4172 if (adapter->netdev->features & NETIF_F_RXALL) {
4173 /* UPE and MPE will be handled by normal PROMISC logic
4174 * in e1000e_set_rx_mode */
4175 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4176 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4177 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4179 fctrl &= ~(IXGBE_FCTRL_DPF);
4180 /* NOTE: VLAN filtering is disabled by setting PROMISC */
4183 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4184 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4186 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
4187 ixgbe_vlan_strip_enable(adapter);
4189 ixgbe_vlan_strip_disable(adapter);
4192 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4196 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4197 ixgbe_qv_init_lock(adapter->q_vector[q_idx]);
4198 napi_enable(&adapter->q_vector[q_idx]->napi);
4202 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4206 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4207 napi_disable(&adapter->q_vector[q_idx]->napi);
4208 while (!ixgbe_qv_disable(adapter->q_vector[q_idx])) {
4209 pr_info("QV %d locked\n", q_idx);
4210 usleep_range(1000, 20000);
4215 #ifdef CONFIG_IXGBE_DCB
4217 * ixgbe_configure_dcb - Configure DCB hardware
4218 * @adapter: ixgbe adapter struct
4220 * This is called by the driver on open to configure the DCB hardware.
4221 * This is also called by the gennetlink interface when reconfiguring
4224 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
4226 struct ixgbe_hw *hw = &adapter->hw;
4227 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4229 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
4230 if (hw->mac.type == ixgbe_mac_82598EB)
4231 netif_set_gso_max_size(adapter->netdev, 65536);
4235 if (hw->mac.type == ixgbe_mac_82598EB)
4236 netif_set_gso_max_size(adapter->netdev, 32768);
4239 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
4240 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
4243 /* reconfigure the hardware */
4244 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
4245 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4247 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4249 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
4250 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
4251 ixgbe_dcb_hw_ets(&adapter->hw,
4252 adapter->ixgbe_ieee_ets,
4254 ixgbe_dcb_hw_pfc_config(&adapter->hw,
4255 adapter->ixgbe_ieee_pfc->pfc_en,
4256 adapter->ixgbe_ieee_ets->prio_tc);
4259 /* Enable RSS Hash per TC */
4260 if (hw->mac.type != ixgbe_mac_82598EB) {
4262 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
4269 /* write msb to all 8 TCs in one write */
4270 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
4275 /* Additional bittime to account for IXGBE framing */
4276 #define IXGBE_ETH_FRAMING 20
4279 * ixgbe_hpbthresh - calculate high water mark for flow control
4281 * @adapter: board private structure to calculate for
4282 * @pb: packet buffer to calculate
4284 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
4286 struct ixgbe_hw *hw = &adapter->hw;
4287 struct net_device *dev = adapter->netdev;
4288 int link, tc, kb, marker;
4291 /* Calculate max LAN frame size */
4292 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
4295 /* FCoE traffic class uses FCOE jumbo frames */
4296 if ((dev->features & NETIF_F_FCOE_MTU) &&
4297 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4298 (pb == ixgbe_fcoe_get_tc(adapter)))
4299 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4302 /* Calculate delay value for device */
4303 switch (hw->mac.type) {
4304 case ixgbe_mac_X540:
4305 case ixgbe_mac_X550:
4306 case ixgbe_mac_X550EM_x:
4307 dv_id = IXGBE_DV_X540(link, tc);
4310 dv_id = IXGBE_DV(link, tc);
4314 /* Loopback switch introduces additional latency */
4315 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4316 dv_id += IXGBE_B2BT(tc);
4318 /* Delay value is calculated in bit times convert to KB */
4319 kb = IXGBE_BT2KB(dv_id);
4320 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
4322 marker = rx_pba - kb;
4324 /* It is possible that the packet buffer is not large enough
4325 * to provide required headroom. In this case throw an error
4326 * to user and a do the best we can.
4329 e_warn(drv, "Packet Buffer(%i) can not provide enough"
4330 "headroom to support flow control."
4331 "Decrease MTU or number of traffic classes\n", pb);
4339 * ixgbe_lpbthresh - calculate low water mark for for flow control
4341 * @adapter: board private structure to calculate for
4342 * @pb: packet buffer to calculate
4344 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
4346 struct ixgbe_hw *hw = &adapter->hw;
4347 struct net_device *dev = adapter->netdev;
4351 /* Calculate max LAN frame size */
4352 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4355 /* FCoE traffic class uses FCOE jumbo frames */
4356 if ((dev->features & NETIF_F_FCOE_MTU) &&
4357 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4358 (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
4359 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4362 /* Calculate delay value for device */
4363 switch (hw->mac.type) {
4364 case ixgbe_mac_X540:
4365 case ixgbe_mac_X550:
4366 case ixgbe_mac_X550EM_x:
4367 dv_id = IXGBE_LOW_DV_X540(tc);
4370 dv_id = IXGBE_LOW_DV(tc);
4374 /* Delay value is calculated in bit times convert to KB */
4375 return IXGBE_BT2KB(dv_id);
4379 * ixgbe_pbthresh_setup - calculate and setup high low water marks
4381 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
4383 struct ixgbe_hw *hw = &adapter->hw;
4384 int num_tc = netdev_get_num_tc(adapter->netdev);
4390 for (i = 0; i < num_tc; i++) {
4391 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
4392 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
4394 /* Low water marks must not be larger than high water marks */
4395 if (hw->fc.low_water[i] > hw->fc.high_water[i])
4396 hw->fc.low_water[i] = 0;
4399 for (; i < MAX_TRAFFIC_CLASS; i++)
4400 hw->fc.high_water[i] = 0;
4403 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
4405 struct ixgbe_hw *hw = &adapter->hw;
4407 u8 tc = netdev_get_num_tc(adapter->netdev);
4409 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4410 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4411 hdrm = 32 << adapter->fdir_pballoc;
4415 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
4416 ixgbe_pbthresh_setup(adapter);
4419 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
4421 struct ixgbe_hw *hw = &adapter->hw;
4422 struct hlist_node *node2;
4423 struct ixgbe_fdir_filter *filter;
4425 spin_lock(&adapter->fdir_perfect_lock);
4427 if (!hlist_empty(&adapter->fdir_filter_list))
4428 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
4430 hlist_for_each_entry_safe(filter, node2,
4431 &adapter->fdir_filter_list, fdir_node) {
4432 ixgbe_fdir_write_perfect_filter_82599(hw,
4435 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
4436 IXGBE_FDIR_DROP_QUEUE :
4437 adapter->rx_ring[filter->action]->reg_idx);
4440 spin_unlock(&adapter->fdir_perfect_lock);
4443 static void ixgbe_macvlan_set_rx_mode(struct net_device *dev, unsigned int pool,
4444 struct ixgbe_adapter *adapter)
4446 struct ixgbe_hw *hw = &adapter->hw;
4449 /* No unicast promiscuous support for VMDQ devices. */
4450 vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(pool));
4451 vmolr |= (IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
4453 /* clear the affected bit */
4454 vmolr &= ~IXGBE_VMOLR_MPE;
4456 if (dev->flags & IFF_ALLMULTI) {
4457 vmolr |= IXGBE_VMOLR_MPE;
4459 vmolr |= IXGBE_VMOLR_ROMPE;
4460 hw->mac.ops.update_mc_addr_list(hw, dev);
4462 ixgbe_write_uc_addr_list(adapter->netdev, pool);
4463 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(pool), vmolr);
4466 static void ixgbe_fwd_psrtype(struct ixgbe_fwd_adapter *vadapter)
4468 struct ixgbe_adapter *adapter = vadapter->real_adapter;
4469 int rss_i = adapter->num_rx_queues_per_pool;
4470 struct ixgbe_hw *hw = &adapter->hw;
4471 u16 pool = vadapter->pool;
4472 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4473 IXGBE_PSRTYPE_UDPHDR |
4474 IXGBE_PSRTYPE_IPV4HDR |
4475 IXGBE_PSRTYPE_L2HDR |
4476 IXGBE_PSRTYPE_IPV6HDR;
4478 if (hw->mac.type == ixgbe_mac_82598EB)
4486 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4490 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4491 * @rx_ring: ring to free buffers from
4493 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4495 struct device *dev = rx_ring->dev;
4499 /* ring already cleared, nothing to do */
4500 if (!rx_ring->rx_buffer_info)
4503 /* Free all the Rx ring sk_buffs */
4504 for (i = 0; i < rx_ring->count; i++) {
4505 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
4507 if (rx_buffer->skb) {
4508 struct sk_buff *skb = rx_buffer->skb;
4509 if (IXGBE_CB(skb)->page_released)
4512 ixgbe_rx_bufsz(rx_ring),
4515 rx_buffer->skb = NULL;
4518 if (!rx_buffer->page)
4521 dma_unmap_page(dev, rx_buffer->dma,
4522 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
4523 __free_pages(rx_buffer->page, ixgbe_rx_pg_order(rx_ring));
4525 rx_buffer->page = NULL;
4528 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4529 memset(rx_ring->rx_buffer_info, 0, size);
4531 /* Zero out the descriptor ring */
4532 memset(rx_ring->desc, 0, rx_ring->size);
4534 rx_ring->next_to_alloc = 0;
4535 rx_ring->next_to_clean = 0;
4536 rx_ring->next_to_use = 0;
4539 static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter *vadapter,
4540 struct ixgbe_ring *rx_ring)
4542 struct ixgbe_adapter *adapter = vadapter->real_adapter;
4543 int index = rx_ring->queue_index + vadapter->rx_base_queue;
4545 /* shutdown specific queue receive and wait for dma to settle */
4546 ixgbe_disable_rx_queue(adapter, rx_ring);
4547 usleep_range(10000, 20000);
4548 ixgbe_irq_disable_queues(adapter, ((u64)1 << index));
4549 ixgbe_clean_rx_ring(rx_ring);
4550 rx_ring->l2_accel_priv = NULL;
4553 static int ixgbe_fwd_ring_down(struct net_device *vdev,
4554 struct ixgbe_fwd_adapter *accel)
4556 struct ixgbe_adapter *adapter = accel->real_adapter;
4557 unsigned int rxbase = accel->rx_base_queue;
4558 unsigned int txbase = accel->tx_base_queue;
4561 netif_tx_stop_all_queues(vdev);
4563 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4564 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4565 adapter->rx_ring[rxbase + i]->netdev = adapter->netdev;
4568 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4569 adapter->tx_ring[txbase + i]->l2_accel_priv = NULL;
4570 adapter->tx_ring[txbase + i]->netdev = adapter->netdev;
4577 static int ixgbe_fwd_ring_up(struct net_device *vdev,
4578 struct ixgbe_fwd_adapter *accel)
4580 struct ixgbe_adapter *adapter = accel->real_adapter;
4581 unsigned int rxbase, txbase, queues;
4582 int i, baseq, err = 0;
4584 if (!test_bit(accel->pool, &adapter->fwd_bitmask))
4587 baseq = accel->pool * adapter->num_rx_queues_per_pool;
4588 netdev_dbg(vdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
4589 accel->pool, adapter->num_rx_pools,
4590 baseq, baseq + adapter->num_rx_queues_per_pool,
4591 adapter->fwd_bitmask);
4593 accel->netdev = vdev;
4594 accel->rx_base_queue = rxbase = baseq;
4595 accel->tx_base_queue = txbase = baseq;
4597 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
4598 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4600 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4601 adapter->rx_ring[rxbase + i]->netdev = vdev;
4602 adapter->rx_ring[rxbase + i]->l2_accel_priv = accel;
4603 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[rxbase + i]);
4606 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4607 adapter->tx_ring[txbase + i]->netdev = vdev;
4608 adapter->tx_ring[txbase + i]->l2_accel_priv = accel;
4611 queues = min_t(unsigned int,
4612 adapter->num_rx_queues_per_pool, vdev->num_tx_queues);
4613 err = netif_set_real_num_tx_queues(vdev, queues);
4617 err = netif_set_real_num_rx_queues(vdev, queues);
4621 if (is_valid_ether_addr(vdev->dev_addr))
4622 ixgbe_add_mac_filter(adapter, vdev->dev_addr, accel->pool);
4624 ixgbe_fwd_psrtype(accel);
4625 ixgbe_macvlan_set_rx_mode(vdev, accel->pool, adapter);
4628 ixgbe_fwd_ring_down(vdev, accel);
4632 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
4634 struct net_device *upper;
4635 struct list_head *iter;
4638 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
4639 if (netif_is_macvlan(upper)) {
4640 struct macvlan_dev *dfwd = netdev_priv(upper);
4641 struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv;
4643 if (dfwd->fwd_priv) {
4644 err = ixgbe_fwd_ring_up(upper, vadapter);
4652 static void ixgbe_configure(struct ixgbe_adapter *adapter)
4654 struct ixgbe_hw *hw = &adapter->hw;
4656 ixgbe_configure_pb(adapter);
4657 #ifdef CONFIG_IXGBE_DCB
4658 ixgbe_configure_dcb(adapter);
4661 * We must restore virtualization before VLANs or else
4662 * the VLVF registers will not be populated
4664 ixgbe_configure_virtualization(adapter);
4666 ixgbe_set_rx_mode(adapter->netdev);
4667 ixgbe_restore_vlan(adapter);
4669 switch (hw->mac.type) {
4670 case ixgbe_mac_82599EB:
4671 case ixgbe_mac_X540:
4672 hw->mac.ops.disable_rx_buff(hw);
4678 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4679 ixgbe_init_fdir_signature_82599(&adapter->hw,
4680 adapter->fdir_pballoc);
4681 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
4682 ixgbe_init_fdir_perfect_82599(&adapter->hw,
4683 adapter->fdir_pballoc);
4684 ixgbe_fdir_filter_restore(adapter);
4687 switch (hw->mac.type) {
4688 case ixgbe_mac_82599EB:
4689 case ixgbe_mac_X540:
4690 hw->mac.ops.enable_rx_buff(hw);
4697 /* configure FCoE L2 filters, redirection table, and Rx control */
4698 ixgbe_configure_fcoe(adapter);
4700 #endif /* IXGBE_FCOE */
4701 ixgbe_configure_tx(adapter);
4702 ixgbe_configure_rx(adapter);
4703 ixgbe_configure_dfwd(adapter);
4706 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
4708 switch (hw->phy.type) {
4709 case ixgbe_phy_sfp_avago:
4710 case ixgbe_phy_sfp_ftl:
4711 case ixgbe_phy_sfp_intel:
4712 case ixgbe_phy_sfp_unknown:
4713 case ixgbe_phy_sfp_passive_tyco:
4714 case ixgbe_phy_sfp_passive_unknown:
4715 case ixgbe_phy_sfp_active_unknown:
4716 case ixgbe_phy_sfp_ftl_active:
4717 case ixgbe_phy_qsfp_passive_unknown:
4718 case ixgbe_phy_qsfp_active_unknown:
4719 case ixgbe_phy_qsfp_intel:
4720 case ixgbe_phy_qsfp_unknown:
4721 /* ixgbe_phy_none is set when no SFP module is present */
4722 case ixgbe_phy_none:
4725 if (hw->mac.type == ixgbe_mac_82598EB)
4733 * ixgbe_sfp_link_config - set up SFP+ link
4734 * @adapter: pointer to private adapter struct
4736 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
4739 * We are assuming the worst case scenario here, and that
4740 * is that an SFP was inserted/removed after the reset
4741 * but before SFP detection was enabled. As such the best
4742 * solution is to just start searching as soon as we start
4744 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
4745 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
4747 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
4751 * ixgbe_non_sfp_link_config - set up non-SFP+ link
4752 * @hw: pointer to private hardware struct
4754 * Returns 0 on success, negative on failure
4756 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
4759 bool autoneg, link_up = false;
4760 u32 ret = IXGBE_ERR_LINK_SETUP;
4762 if (hw->mac.ops.check_link)
4763 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
4768 speed = hw->phy.autoneg_advertised;
4769 if ((!speed) && (hw->mac.ops.get_link_capabilities))
4770 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
4775 if (hw->mac.ops.setup_link)
4776 ret = hw->mac.ops.setup_link(hw, speed, link_up);
4781 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
4783 struct ixgbe_hw *hw = &adapter->hw;
4786 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4787 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
4789 gpie |= IXGBE_GPIE_EIAME;
4791 * use EIAM to auto-mask when MSI-X interrupt is asserted
4792 * this saves a register write for every interrupt
4794 switch (hw->mac.type) {
4795 case ixgbe_mac_82598EB:
4796 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4798 case ixgbe_mac_82599EB:
4799 case ixgbe_mac_X540:
4800 case ixgbe_mac_X550:
4801 case ixgbe_mac_X550EM_x:
4803 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4804 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4808 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
4809 * specifically only auto mask tx and rx interrupts */
4810 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4813 /* XXX: to interrupt immediately for EICS writes, enable this */
4814 /* gpie |= IXGBE_GPIE_EIMEN; */
4816 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
4817 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
4819 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4820 case IXGBE_82599_VMDQ_8Q_MASK:
4821 gpie |= IXGBE_GPIE_VTMODE_16;
4823 case IXGBE_82599_VMDQ_4Q_MASK:
4824 gpie |= IXGBE_GPIE_VTMODE_32;
4827 gpie |= IXGBE_GPIE_VTMODE_64;
4832 /* Enable Thermal over heat sensor interrupt */
4833 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
4834 switch (adapter->hw.mac.type) {
4835 case ixgbe_mac_82599EB:
4836 gpie |= IXGBE_SDP0_GPIEN;
4838 case ixgbe_mac_X540:
4839 gpie |= IXGBE_EIMS_TS;
4846 /* Enable fan failure interrupt */
4847 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
4848 gpie |= IXGBE_SDP1_GPIEN;
4850 if (hw->mac.type == ixgbe_mac_82599EB) {
4851 gpie |= IXGBE_SDP1_GPIEN;
4852 gpie |= IXGBE_SDP2_GPIEN;
4855 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4858 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
4860 struct ixgbe_hw *hw = &adapter->hw;
4864 ixgbe_get_hw_control(adapter);
4865 ixgbe_setup_gpie(adapter);
4867 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4868 ixgbe_configure_msix(adapter);
4870 ixgbe_configure_msi_and_legacy(adapter);
4872 /* enable the optics for 82599 SFP+ fiber */
4873 if (hw->mac.ops.enable_tx_laser)
4874 hw->mac.ops.enable_tx_laser(hw);
4876 smp_mb__before_atomic();
4877 clear_bit(__IXGBE_DOWN, &adapter->state);
4878 ixgbe_napi_enable_all(adapter);
4880 if (ixgbe_is_sfp(hw)) {
4881 ixgbe_sfp_link_config(adapter);
4883 err = ixgbe_non_sfp_link_config(hw);
4885 e_err(probe, "link_config FAILED %d\n", err);
4888 /* clear any pending interrupts, may auto mask */
4889 IXGBE_READ_REG(hw, IXGBE_EICR);
4890 ixgbe_irq_enable(adapter, true, true);
4893 * If this adapter has a fan, check to see if we had a failure
4894 * before we enabled the interrupt.
4896 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4897 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4898 if (esdp & IXGBE_ESDP_SDP1)
4899 e_crit(drv, "Fan has stopped, replace the adapter\n");
4902 /* bring the link up in the watchdog, this could race with our first
4903 * link up interrupt but shouldn't be a problem */
4904 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4905 adapter->link_check_timeout = jiffies;
4906 mod_timer(&adapter->service_timer, jiffies);
4908 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4909 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4910 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4911 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4914 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4916 WARN_ON(in_interrupt());
4917 /* put off any impending NetWatchDogTimeout */
4918 adapter->netdev->trans_start = jiffies;
4920 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4921 usleep_range(1000, 2000);
4922 ixgbe_down(adapter);
4924 * If SR-IOV enabled then wait a bit before bringing the adapter
4925 * back up to give the VFs time to respond to the reset. The
4926 * two second wait is based upon the watchdog timer cycle in
4929 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4932 clear_bit(__IXGBE_RESETTING, &adapter->state);
4935 void ixgbe_up(struct ixgbe_adapter *adapter)
4937 /* hardware has been reset, we need to reload some things */
4938 ixgbe_configure(adapter);
4940 ixgbe_up_complete(adapter);
4943 void ixgbe_reset(struct ixgbe_adapter *adapter)
4945 struct ixgbe_hw *hw = &adapter->hw;
4946 struct net_device *netdev = adapter->netdev;
4948 u8 old_addr[ETH_ALEN];
4950 if (ixgbe_removed(hw->hw_addr))
4952 /* lock SFP init bit to prevent race conditions with the watchdog */
4953 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4954 usleep_range(1000, 2000);
4956 /* clear all SFP and link config related flags while holding SFP_INIT */
4957 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4958 IXGBE_FLAG2_SFP_NEEDS_RESET);
4959 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4961 err = hw->mac.ops.init_hw(hw);
4964 case IXGBE_ERR_SFP_NOT_PRESENT:
4965 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4967 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4968 e_dev_err("master disable timed out\n");
4970 case IXGBE_ERR_EEPROM_VERSION:
4971 /* We are running on a pre-production device, log a warning */
4972 e_dev_warn("This device is a pre-production adapter/LOM. "
4973 "Please be aware there may be issues associated with "
4974 "your hardware. If you are experiencing problems "
4975 "please contact your Intel or hardware "
4976 "representative who provided you with this "
4980 e_dev_err("Hardware Error: %d\n", err);
4983 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4984 /* do not flush user set addresses */
4985 memcpy(old_addr, &adapter->mac_table[0].addr, netdev->addr_len);
4986 ixgbe_flush_sw_mac_table(adapter);
4987 ixgbe_mac_set_default_filter(adapter, old_addr);
4989 /* update SAN MAC vmdq pool selection */
4990 if (hw->mac.san_mac_rar_index)
4991 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
4993 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
4994 ixgbe_ptp_reset(adapter);
4998 * ixgbe_clean_tx_ring - Free Tx Buffers
4999 * @tx_ring: ring to be cleaned
5001 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5003 struct ixgbe_tx_buffer *tx_buffer_info;
5007 /* ring already cleared, nothing to do */
5008 if (!tx_ring->tx_buffer_info)
5011 /* Free all the Tx ring sk_buffs */
5012 for (i = 0; i < tx_ring->count; i++) {
5013 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5014 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
5017 netdev_tx_reset_queue(txring_txq(tx_ring));
5019 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5020 memset(tx_ring->tx_buffer_info, 0, size);
5022 /* Zero out the descriptor ring */
5023 memset(tx_ring->desc, 0, tx_ring->size);
5025 tx_ring->next_to_use = 0;
5026 tx_ring->next_to_clean = 0;
5030 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
5031 * @adapter: board private structure
5033 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
5037 for (i = 0; i < adapter->num_rx_queues; i++)
5038 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
5042 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
5043 * @adapter: board private structure
5045 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
5049 for (i = 0; i < adapter->num_tx_queues; i++)
5050 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
5053 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
5055 struct hlist_node *node2;
5056 struct ixgbe_fdir_filter *filter;
5058 spin_lock(&adapter->fdir_perfect_lock);
5060 hlist_for_each_entry_safe(filter, node2,
5061 &adapter->fdir_filter_list, fdir_node) {
5062 hlist_del(&filter->fdir_node);
5065 adapter->fdir_filter_count = 0;
5067 spin_unlock(&adapter->fdir_perfect_lock);
5070 void ixgbe_down(struct ixgbe_adapter *adapter)
5072 struct net_device *netdev = adapter->netdev;
5073 struct ixgbe_hw *hw = &adapter->hw;
5074 struct net_device *upper;
5075 struct list_head *iter;
5078 /* signal that we are down to the interrupt handler */
5079 if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
5080 return; /* do nothing if already down */
5082 /* disable receives */
5083 hw->mac.ops.disable_rx(hw);
5085 /* disable all enabled rx queues */
5086 for (i = 0; i < adapter->num_rx_queues; i++)
5087 /* this call also flushes the previous write */
5088 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
5090 usleep_range(10000, 20000);
5092 netif_tx_stop_all_queues(netdev);
5094 /* call carrier off first to avoid false dev_watchdog timeouts */
5095 netif_carrier_off(netdev);
5096 netif_tx_disable(netdev);
5098 /* disable any upper devices */
5099 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
5100 if (netif_is_macvlan(upper)) {
5101 struct macvlan_dev *vlan = netdev_priv(upper);
5103 if (vlan->fwd_priv) {
5104 netif_tx_stop_all_queues(upper);
5105 netif_carrier_off(upper);
5106 netif_tx_disable(upper);
5111 ixgbe_irq_disable(adapter);
5113 ixgbe_napi_disable_all(adapter);
5115 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
5116 IXGBE_FLAG2_RESET_REQUESTED);
5117 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5119 del_timer_sync(&adapter->service_timer);
5121 if (adapter->num_vfs) {
5122 /* Clear EITR Select mapping */
5123 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
5125 /* Mark all the VFs as inactive */
5126 for (i = 0 ; i < adapter->num_vfs; i++)
5127 adapter->vfinfo[i].clear_to_send = false;
5129 /* ping all the active vfs to let them know we are going down */
5130 ixgbe_ping_all_vfs(adapter);
5132 /* Disable all VFTE/VFRE TX/RX */
5133 ixgbe_disable_tx_rx(adapter);
5136 /* disable transmits in the hardware now that interrupts are off */
5137 for (i = 0; i < adapter->num_tx_queues; i++) {
5138 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
5139 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5142 /* Disable the Tx DMA engine on 82599 and later MAC */
5143 switch (hw->mac.type) {
5144 case ixgbe_mac_82599EB:
5145 case ixgbe_mac_X540:
5146 case ixgbe_mac_X550:
5147 case ixgbe_mac_X550EM_x:
5148 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5149 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5150 ~IXGBE_DMATXCTL_TE));
5156 if (!pci_channel_offline(adapter->pdev))
5157 ixgbe_reset(adapter);
5159 /* power down the optics for 82599 SFP+ fiber */
5160 if (hw->mac.ops.disable_tx_laser)
5161 hw->mac.ops.disable_tx_laser(hw);
5163 ixgbe_clean_all_tx_rings(adapter);
5164 ixgbe_clean_all_rx_rings(adapter);
5166 #ifdef CONFIG_IXGBE_DCA
5167 /* since we reset the hardware DCA settings were cleared */
5168 ixgbe_setup_dca(adapter);
5173 * ixgbe_tx_timeout - Respond to a Tx Hang
5174 * @netdev: network interface device structure
5176 static void ixgbe_tx_timeout(struct net_device *netdev)
5178 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5180 /* Do the reset outside of interrupt context */
5181 ixgbe_tx_timeout_reset(adapter);
5185 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5186 * @adapter: board private structure to initialize
5188 * ixgbe_sw_init initializes the Adapter private data structure.
5189 * Fields are initialized based on PCI device information and
5190 * OS network device settings (MTU size).
5192 static int ixgbe_sw_init(struct ixgbe_adapter *adapter)
5194 struct ixgbe_hw *hw = &adapter->hw;
5195 struct pci_dev *pdev = adapter->pdev;
5196 unsigned int rss, fdir;
5198 #ifdef CONFIG_IXGBE_DCB
5200 struct tc_configuration *tc;
5203 /* PCI config space info */
5205 hw->vendor_id = pdev->vendor;
5206 hw->device_id = pdev->device;
5207 hw->revision_id = pdev->revision;
5208 hw->subsystem_vendor_id = pdev->subsystem_vendor;
5209 hw->subsystem_device_id = pdev->subsystem_device;
5211 /* Set common capability flags and settings */
5212 rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
5213 adapter->ring_feature[RING_F_RSS].limit = rss;
5214 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5215 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5216 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
5217 adapter->atr_sample_rate = 20;
5218 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
5219 adapter->ring_feature[RING_F_FDIR].limit = fdir;
5220 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
5221 #ifdef CONFIG_IXGBE_DCA
5222 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
5225 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5226 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5227 #ifdef CONFIG_IXGBE_DCB
5228 /* Default traffic class to use for FCoE */
5229 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5230 #endif /* CONFIG_IXGBE_DCB */
5231 #endif /* IXGBE_FCOE */
5233 adapter->mac_table = kzalloc(sizeof(struct ixgbe_mac_addr) *
5234 hw->mac.num_rar_entries,
5237 /* Set MAC specific capability flags and exceptions */
5238 switch (hw->mac.type) {
5239 case ixgbe_mac_82598EB:
5240 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
5241 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
5243 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5244 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5246 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
5247 adapter->ring_feature[RING_F_FDIR].limit = 0;
5248 adapter->atr_sample_rate = 0;
5249 adapter->fdir_pballoc = 0;
5251 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5252 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5253 #ifdef CONFIG_IXGBE_DCB
5254 adapter->fcoe.up = 0;
5255 #endif /* IXGBE_DCB */
5256 #endif /* IXGBE_FCOE */
5258 case ixgbe_mac_82599EB:
5259 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5260 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5262 case ixgbe_mac_X540:
5263 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM);
5264 if (fwsm & IXGBE_FWSM_TS_ENABLED)
5265 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5267 case ixgbe_mac_X550EM_x:
5268 case ixgbe_mac_X550:
5269 #ifdef CONFIG_IXGBE_DCA
5270 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
5278 /* FCoE support exists, always init the FCoE lock */
5279 spin_lock_init(&adapter->fcoe.lock);
5282 /* n-tuple support exists, always init our spinlock */
5283 spin_lock_init(&adapter->fdir_perfect_lock);
5285 #ifdef CONFIG_IXGBE_DCB
5286 switch (hw->mac.type) {
5287 case ixgbe_mac_X540:
5288 case ixgbe_mac_X550:
5289 case ixgbe_mac_X550EM_x:
5290 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
5291 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
5294 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
5295 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
5299 /* Configure DCB traffic classes */
5300 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5301 tc = &adapter->dcb_cfg.tc_config[j];
5302 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5303 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5304 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5305 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5306 tc->dcb_pfc = pfc_disabled;
5309 /* Initialize default user to priority mapping, UPx->TC0 */
5310 tc = &adapter->dcb_cfg.tc_config[0];
5311 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5312 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5314 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5315 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5316 adapter->dcb_cfg.pfc_mode_enable = false;
5317 adapter->dcb_set_bitmap = 0x00;
5318 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5319 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
5320 sizeof(adapter->temp_dcb_cfg));
5324 /* default flow control settings */
5325 hw->fc.requested_mode = ixgbe_fc_full;
5326 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5327 ixgbe_pbthresh_setup(adapter);
5328 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5329 hw->fc.send_xon = true;
5330 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
5332 #ifdef CONFIG_PCI_IOV
5334 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
5336 /* assign number of SR-IOV VFs */
5337 if (hw->mac.type != ixgbe_mac_82598EB) {
5338 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
5339 adapter->num_vfs = 0;
5340 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
5342 adapter->num_vfs = max_vfs;
5345 #endif /* CONFIG_PCI_IOV */
5347 /* enable itr by default in dynamic mode */
5348 adapter->rx_itr_setting = 1;
5349 adapter->tx_itr_setting = 1;
5351 /* set default ring sizes */
5352 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5353 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5355 /* set default work limits */
5356 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5358 /* initialize eeprom parameters */
5359 if (ixgbe_init_eeprom_params_generic(hw)) {
5360 e_dev_err("EEPROM initialization failed\n");
5364 /* PF holds first pool slot */
5365 set_bit(0, &adapter->fwd_bitmask);
5366 set_bit(__IXGBE_DOWN, &adapter->state);
5372 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5373 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5375 * Return 0 on success, negative on failure
5377 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5379 struct device *dev = tx_ring->dev;
5380 int orig_node = dev_to_node(dev);
5384 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5386 if (tx_ring->q_vector)
5387 ring_node = tx_ring->q_vector->numa_node;
5389 tx_ring->tx_buffer_info = vzalloc_node(size, ring_node);
5390 if (!tx_ring->tx_buffer_info)
5391 tx_ring->tx_buffer_info = vzalloc(size);
5392 if (!tx_ring->tx_buffer_info)
5395 u64_stats_init(&tx_ring->syncp);
5397 /* round up to nearest 4K */
5398 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5399 tx_ring->size = ALIGN(tx_ring->size, 4096);
5401 set_dev_node(dev, ring_node);
5402 tx_ring->desc = dma_alloc_coherent(dev,
5406 set_dev_node(dev, orig_node);
5408 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5409 &tx_ring->dma, GFP_KERNEL);
5413 tx_ring->next_to_use = 0;
5414 tx_ring->next_to_clean = 0;
5418 vfree(tx_ring->tx_buffer_info);
5419 tx_ring->tx_buffer_info = NULL;
5420 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5425 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5426 * @adapter: board private structure
5428 * If this function returns with an error, then it's possible one or
5429 * more of the rings is populated (while the rest are not). It is the
5430 * callers duty to clean those orphaned rings.
5432 * Return 0 on success, negative on failure
5434 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5438 for (i = 0; i < adapter->num_tx_queues; i++) {
5439 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5443 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5449 /* rewind the index freeing the rings as we go */
5451 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5456 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5457 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5459 * Returns 0 on success, negative on failure
5461 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5463 struct device *dev = rx_ring->dev;
5464 int orig_node = dev_to_node(dev);
5468 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5470 if (rx_ring->q_vector)
5471 ring_node = rx_ring->q_vector->numa_node;
5473 rx_ring->rx_buffer_info = vzalloc_node(size, ring_node);
5474 if (!rx_ring->rx_buffer_info)
5475 rx_ring->rx_buffer_info = vzalloc(size);
5476 if (!rx_ring->rx_buffer_info)
5479 u64_stats_init(&rx_ring->syncp);
5481 /* Round up to nearest 4K */
5482 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5483 rx_ring->size = ALIGN(rx_ring->size, 4096);
5485 set_dev_node(dev, ring_node);
5486 rx_ring->desc = dma_alloc_coherent(dev,
5490 set_dev_node(dev, orig_node);
5492 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5493 &rx_ring->dma, GFP_KERNEL);
5497 rx_ring->next_to_clean = 0;
5498 rx_ring->next_to_use = 0;
5502 vfree(rx_ring->rx_buffer_info);
5503 rx_ring->rx_buffer_info = NULL;
5504 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5509 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5510 * @adapter: board private structure
5512 * If this function returns with an error, then it's possible one or
5513 * more of the rings is populated (while the rest are not). It is the
5514 * callers duty to clean those orphaned rings.
5516 * Return 0 on success, negative on failure
5518 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5522 for (i = 0; i < adapter->num_rx_queues; i++) {
5523 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5527 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5532 err = ixgbe_setup_fcoe_ddp_resources(adapter);
5537 /* rewind the index freeing the rings as we go */
5539 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5544 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5545 * @tx_ring: Tx descriptor ring for a specific queue
5547 * Free all transmit software resources
5549 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5551 ixgbe_clean_tx_ring(tx_ring);
5553 vfree(tx_ring->tx_buffer_info);
5554 tx_ring->tx_buffer_info = NULL;
5556 /* if not set, then don't free */
5560 dma_free_coherent(tx_ring->dev, tx_ring->size,
5561 tx_ring->desc, tx_ring->dma);
5563 tx_ring->desc = NULL;
5567 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5568 * @adapter: board private structure
5570 * Free all transmit software resources
5572 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5576 for (i = 0; i < adapter->num_tx_queues; i++)
5577 if (adapter->tx_ring[i]->desc)
5578 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5582 * ixgbe_free_rx_resources - Free Rx Resources
5583 * @rx_ring: ring to clean the resources from
5585 * Free all receive software resources
5587 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5589 ixgbe_clean_rx_ring(rx_ring);
5591 vfree(rx_ring->rx_buffer_info);
5592 rx_ring->rx_buffer_info = NULL;
5594 /* if not set, then don't free */
5598 dma_free_coherent(rx_ring->dev, rx_ring->size,
5599 rx_ring->desc, rx_ring->dma);
5601 rx_ring->desc = NULL;
5605 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5606 * @adapter: board private structure
5608 * Free all receive software resources
5610 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5615 ixgbe_free_fcoe_ddp_resources(adapter);
5618 for (i = 0; i < adapter->num_rx_queues; i++)
5619 if (adapter->rx_ring[i]->desc)
5620 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5624 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5625 * @netdev: network interface device structure
5626 * @new_mtu: new value for maximum frame size
5628 * Returns 0 on success, negative on failure
5630 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5632 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5633 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5635 /* MTU < 68 is an error and causes problems on some kernels */
5636 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5640 * For 82599EB we cannot allow legacy VFs to enable their receive
5641 * paths when MTU greater than 1500 is configured. So display a
5642 * warning that legacy VFs will be disabled.
5644 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
5645 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
5646 (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
5647 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
5649 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5651 /* must set new MTU before calling down or up */
5652 netdev->mtu = new_mtu;
5654 if (netif_running(netdev))
5655 ixgbe_reinit_locked(adapter);
5661 * ixgbe_open - Called when a network interface is made active
5662 * @netdev: network interface device structure
5664 * Returns 0 on success, negative value on failure
5666 * The open entry point is called when a network interface is made
5667 * active by the system (IFF_UP). At this point all resources needed
5668 * for transmit and receive operations are allocated, the interrupt
5669 * handler is registered with the OS, the watchdog timer is started,
5670 * and the stack is notified that the interface is ready.
5672 static int ixgbe_open(struct net_device *netdev)
5674 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5677 /* disallow open during test */
5678 if (test_bit(__IXGBE_TESTING, &adapter->state))
5681 netif_carrier_off(netdev);
5683 /* allocate transmit descriptors */
5684 err = ixgbe_setup_all_tx_resources(adapter);
5688 /* allocate receive descriptors */
5689 err = ixgbe_setup_all_rx_resources(adapter);
5693 ixgbe_configure(adapter);
5695 err = ixgbe_request_irq(adapter);
5699 /* Notify the stack of the actual queue counts. */
5700 if (adapter->num_rx_pools > 1)
5701 queues = adapter->num_rx_queues_per_pool;
5703 queues = adapter->num_tx_queues;
5705 err = netif_set_real_num_tx_queues(netdev, queues);
5707 goto err_set_queues;
5709 if (adapter->num_rx_pools > 1 &&
5710 adapter->num_rx_queues > IXGBE_MAX_L2A_QUEUES)
5711 queues = IXGBE_MAX_L2A_QUEUES;
5713 queues = adapter->num_rx_queues;
5714 err = netif_set_real_num_rx_queues(netdev, queues);
5716 goto err_set_queues;
5718 ixgbe_ptp_init(adapter);
5720 ixgbe_up_complete(adapter);
5722 #if IS_ENABLED(CONFIG_IXGBE_VXLAN)
5723 vxlan_get_rx_port(netdev);
5729 ixgbe_free_irq(adapter);
5731 ixgbe_free_all_rx_resources(adapter);
5733 ixgbe_free_all_tx_resources(adapter);
5735 ixgbe_reset(adapter);
5740 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
5742 ixgbe_ptp_suspend(adapter);
5744 ixgbe_down(adapter);
5745 ixgbe_free_irq(adapter);
5747 ixgbe_free_all_tx_resources(adapter);
5748 ixgbe_free_all_rx_resources(adapter);
5752 * ixgbe_close - Disables a network interface
5753 * @netdev: network interface device structure
5755 * Returns 0, this is not allowed to fail
5757 * The close entry point is called when an interface is de-activated
5758 * by the OS. The hardware is still under the drivers control, but
5759 * needs to be disabled. A global MAC reset is issued to stop the
5760 * hardware, and all transmit and receive resources are freed.
5762 static int ixgbe_close(struct net_device *netdev)
5764 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5766 ixgbe_ptp_stop(adapter);
5768 ixgbe_close_suspend(adapter);
5770 ixgbe_fdir_filter_exit(adapter);
5772 ixgbe_release_hw_control(adapter);
5778 static int ixgbe_resume(struct pci_dev *pdev)
5780 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5781 struct net_device *netdev = adapter->netdev;
5784 adapter->hw.hw_addr = adapter->io_addr;
5785 pci_set_power_state(pdev, PCI_D0);
5786 pci_restore_state(pdev);
5788 * pci_restore_state clears dev->state_saved so call
5789 * pci_save_state to restore it.
5791 pci_save_state(pdev);
5793 err = pci_enable_device_mem(pdev);
5795 e_dev_err("Cannot enable PCI device from suspend\n");
5798 smp_mb__before_atomic();
5799 clear_bit(__IXGBE_DISABLED, &adapter->state);
5800 pci_set_master(pdev);
5802 pci_wake_from_d3(pdev, false);
5804 ixgbe_reset(adapter);
5806 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5809 err = ixgbe_init_interrupt_scheme(adapter);
5810 if (!err && netif_running(netdev))
5811 err = ixgbe_open(netdev);
5818 netif_device_attach(netdev);
5822 #endif /* CONFIG_PM */
5824 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5826 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5827 struct net_device *netdev = adapter->netdev;
5828 struct ixgbe_hw *hw = &adapter->hw;
5830 u32 wufc = adapter->wol;
5835 netif_device_detach(netdev);
5838 if (netif_running(netdev))
5839 ixgbe_close_suspend(adapter);
5842 ixgbe_clear_interrupt_scheme(adapter);
5845 retval = pci_save_state(pdev);
5850 if (hw->mac.ops.stop_link_on_d3)
5851 hw->mac.ops.stop_link_on_d3(hw);
5854 ixgbe_set_rx_mode(netdev);
5856 /* enable the optics for 82599 SFP+ fiber as we can WoL */
5857 if (hw->mac.ops.enable_tx_laser)
5858 hw->mac.ops.enable_tx_laser(hw);
5860 /* turn on all-multi mode if wake on multicast is enabled */
5861 if (wufc & IXGBE_WUFC_MC) {
5862 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5863 fctrl |= IXGBE_FCTRL_MPE;
5864 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5867 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5868 ctrl |= IXGBE_CTRL_GIO_DIS;
5869 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5871 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5873 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5874 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5877 switch (hw->mac.type) {
5878 case ixgbe_mac_82598EB:
5879 pci_wake_from_d3(pdev, false);
5881 case ixgbe_mac_82599EB:
5882 case ixgbe_mac_X540:
5883 case ixgbe_mac_X550:
5884 case ixgbe_mac_X550EM_x:
5885 pci_wake_from_d3(pdev, !!wufc);
5891 *enable_wake = !!wufc;
5893 ixgbe_release_hw_control(adapter);
5895 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
5896 pci_disable_device(pdev);
5902 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5907 retval = __ixgbe_shutdown(pdev, &wake);
5912 pci_prepare_to_sleep(pdev);
5914 pci_wake_from_d3(pdev, false);
5915 pci_set_power_state(pdev, PCI_D3hot);
5920 #endif /* CONFIG_PM */
5922 static void ixgbe_shutdown(struct pci_dev *pdev)
5926 __ixgbe_shutdown(pdev, &wake);
5928 if (system_state == SYSTEM_POWER_OFF) {
5929 pci_wake_from_d3(pdev, wake);
5930 pci_set_power_state(pdev, PCI_D3hot);
5935 * ixgbe_update_stats - Update the board statistics counters.
5936 * @adapter: board private structure
5938 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5940 struct net_device *netdev = adapter->netdev;
5941 struct ixgbe_hw *hw = &adapter->hw;
5942 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5944 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5945 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5946 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5947 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5949 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5950 test_bit(__IXGBE_RESETTING, &adapter->state))
5953 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5956 for (i = 0; i < adapter->num_rx_queues; i++) {
5957 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5958 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5960 adapter->rsc_total_count = rsc_count;
5961 adapter->rsc_total_flush = rsc_flush;
5964 for (i = 0; i < adapter->num_rx_queues; i++) {
5965 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5966 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5967 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5968 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5969 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5970 bytes += rx_ring->stats.bytes;
5971 packets += rx_ring->stats.packets;
5973 adapter->non_eop_descs = non_eop_descs;
5974 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5975 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5976 adapter->hw_csum_rx_error = hw_csum_rx_error;
5977 netdev->stats.rx_bytes = bytes;
5978 netdev->stats.rx_packets = packets;
5982 /* gather some stats to the adapter struct that are per queue */
5983 for (i = 0; i < adapter->num_tx_queues; i++) {
5984 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5985 restart_queue += tx_ring->tx_stats.restart_queue;
5986 tx_busy += tx_ring->tx_stats.tx_busy;
5987 bytes += tx_ring->stats.bytes;
5988 packets += tx_ring->stats.packets;
5990 adapter->restart_queue = restart_queue;
5991 adapter->tx_busy = tx_busy;
5992 netdev->stats.tx_bytes = bytes;
5993 netdev->stats.tx_packets = packets;
5995 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5997 /* 8 register reads */
5998 for (i = 0; i < 8; i++) {
5999 /* for packet buffers not used, the register should read 0 */
6000 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
6002 hwstats->mpc[i] += mpc;
6003 total_mpc += hwstats->mpc[i];
6004 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
6005 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
6006 switch (hw->mac.type) {
6007 case ixgbe_mac_82598EB:
6008 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
6009 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
6010 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
6011 hwstats->pxonrxc[i] +=
6012 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
6014 case ixgbe_mac_82599EB:
6015 case ixgbe_mac_X540:
6016 case ixgbe_mac_X550:
6017 case ixgbe_mac_X550EM_x:
6018 hwstats->pxonrxc[i] +=
6019 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
6026 /*16 register reads */
6027 for (i = 0; i < 16; i++) {
6028 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
6029 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
6030 if ((hw->mac.type == ixgbe_mac_82599EB) ||
6031 (hw->mac.type == ixgbe_mac_X540) ||
6032 (hw->mac.type == ixgbe_mac_X550) ||
6033 (hw->mac.type == ixgbe_mac_X550EM_x)) {
6034 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
6035 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
6036 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
6037 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
6041 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
6042 /* work around hardware counting issue */
6043 hwstats->gprc -= missed_rx;
6045 ixgbe_update_xoff_received(adapter);
6047 /* 82598 hardware only has a 32 bit counter in the high register */
6048 switch (hw->mac.type) {
6049 case ixgbe_mac_82598EB:
6050 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
6051 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
6052 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
6053 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
6055 case ixgbe_mac_X540:
6056 case ixgbe_mac_X550:
6057 case ixgbe_mac_X550EM_x:
6058 /* OS2BMC stats are X540 and later */
6059 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
6060 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
6061 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
6062 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
6063 case ixgbe_mac_82599EB:
6064 for (i = 0; i < 16; i++)
6065 adapter->hw_rx_no_dma_resources +=
6066 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
6067 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
6068 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
6069 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
6070 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
6071 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
6072 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
6073 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
6074 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
6075 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
6077 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
6078 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
6079 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
6080 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
6081 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
6082 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
6083 /* Add up per cpu counters for total ddp aloc fail */
6084 if (adapter->fcoe.ddp_pool) {
6085 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
6086 struct ixgbe_fcoe_ddp_pool *ddp_pool;
6088 u64 noddp = 0, noddp_ext_buff = 0;
6089 for_each_possible_cpu(cpu) {
6090 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
6091 noddp += ddp_pool->noddp;
6092 noddp_ext_buff += ddp_pool->noddp_ext_buff;
6094 hwstats->fcoe_noddp = noddp;
6095 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
6097 #endif /* IXGBE_FCOE */
6102 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
6103 hwstats->bprc += bprc;
6104 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
6105 if (hw->mac.type == ixgbe_mac_82598EB)
6106 hwstats->mprc -= bprc;
6107 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
6108 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
6109 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
6110 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
6111 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
6112 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
6113 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
6114 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
6115 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
6116 hwstats->lxontxc += lxon;
6117 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
6118 hwstats->lxofftxc += lxoff;
6119 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
6120 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
6122 * 82598 errata - tx of flow control packets is included in tx counters
6124 xon_off_tot = lxon + lxoff;
6125 hwstats->gptc -= xon_off_tot;
6126 hwstats->mptc -= xon_off_tot;
6127 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
6128 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
6129 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
6130 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
6131 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
6132 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
6133 hwstats->ptc64 -= xon_off_tot;
6134 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
6135 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
6136 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
6137 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
6138 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
6139 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
6141 /* Fill out the OS statistics structure */
6142 netdev->stats.multicast = hwstats->mprc;
6145 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
6146 netdev->stats.rx_dropped = 0;
6147 netdev->stats.rx_length_errors = hwstats->rlec;
6148 netdev->stats.rx_crc_errors = hwstats->crcerrs;
6149 netdev->stats.rx_missed_errors = total_mpc;
6153 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
6154 * @adapter: pointer to the device adapter structure
6156 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
6158 struct ixgbe_hw *hw = &adapter->hw;
6161 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
6164 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6166 /* if interface is down do nothing */
6167 if (test_bit(__IXGBE_DOWN, &adapter->state))
6170 /* do nothing if we are not using signature filters */
6171 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
6174 adapter->fdir_overflow++;
6176 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
6177 for (i = 0; i < adapter->num_tx_queues; i++)
6178 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
6179 &(adapter->tx_ring[i]->state));
6180 /* re-enable flow director interrupts */
6181 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
6183 e_err(probe, "failed to finish FDIR re-initialization, "
6184 "ignored adding FDIR ATR filters\n");
6189 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
6190 * @adapter: pointer to the device adapter structure
6192 * This function serves two purposes. First it strobes the interrupt lines
6193 * in order to make certain interrupts are occurring. Secondly it sets the
6194 * bits needed to check for TX hangs. As a result we should immediately
6195 * determine if a hang has occurred.
6197 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
6199 struct ixgbe_hw *hw = &adapter->hw;
6203 /* If we're down, removing or resetting, just bail */
6204 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6205 test_bit(__IXGBE_REMOVING, &adapter->state) ||
6206 test_bit(__IXGBE_RESETTING, &adapter->state))
6209 /* Force detection of hung controller */
6210 if (netif_carrier_ok(adapter->netdev)) {
6211 for (i = 0; i < adapter->num_tx_queues; i++)
6212 set_check_for_tx_hang(adapter->tx_ring[i]);
6215 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6217 * for legacy and MSI interrupts don't set any bits
6218 * that are enabled for EIAM, because this operation
6219 * would set *both* EIMS and EICS for any bit in EIAM
6221 IXGBE_WRITE_REG(hw, IXGBE_EICS,
6222 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
6224 /* get one bit for every active tx/rx interrupt vector */
6225 for (i = 0; i < adapter->num_q_vectors; i++) {
6226 struct ixgbe_q_vector *qv = adapter->q_vector[i];
6227 if (qv->rx.ring || qv->tx.ring)
6228 eics |= ((u64)1 << i);
6232 /* Cause software interrupt to ensure rings are cleaned */
6233 ixgbe_irq_rearm_queues(adapter, eics);
6237 * ixgbe_watchdog_update_link - update the link status
6238 * @adapter: pointer to the device adapter structure
6239 * @link_speed: pointer to a u32 to store the link_speed
6241 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
6243 struct ixgbe_hw *hw = &adapter->hw;
6244 u32 link_speed = adapter->link_speed;
6245 bool link_up = adapter->link_up;
6246 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
6248 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6251 if (hw->mac.ops.check_link) {
6252 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6254 /* always assume link is up, if no check link function */
6255 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
6259 if (adapter->ixgbe_ieee_pfc)
6260 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
6262 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
6263 hw->mac.ops.fc_enable(hw);
6264 ixgbe_set_rx_drop_en(adapter);
6268 time_after(jiffies, (adapter->link_check_timeout +
6269 IXGBE_TRY_LINK_TIMEOUT))) {
6270 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6271 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6272 IXGBE_WRITE_FLUSH(hw);
6275 adapter->link_up = link_up;
6276 adapter->link_speed = link_speed;
6279 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
6281 #ifdef CONFIG_IXGBE_DCB
6282 struct net_device *netdev = adapter->netdev;
6283 struct dcb_app app = {
6284 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
6289 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
6290 up = dcb_ieee_getapp_mask(netdev, &app);
6292 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
6297 * ixgbe_watchdog_link_is_up - update netif_carrier status and
6298 * print link up message
6299 * @adapter: pointer to the device adapter structure
6301 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
6303 struct net_device *netdev = adapter->netdev;
6304 struct ixgbe_hw *hw = &adapter->hw;
6305 struct net_device *upper;
6306 struct list_head *iter;
6307 u32 link_speed = adapter->link_speed;
6308 bool flow_rx, flow_tx;
6310 /* only continue if link was previously down */
6311 if (netif_carrier_ok(netdev))
6314 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
6316 switch (hw->mac.type) {
6317 case ixgbe_mac_82598EB: {
6318 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6319 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6320 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6321 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6324 case ixgbe_mac_X540:
6325 case ixgbe_mac_X550:
6326 case ixgbe_mac_X550EM_x:
6327 case ixgbe_mac_82599EB: {
6328 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6329 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6330 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6331 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6340 adapter->last_rx_ptp_check = jiffies;
6342 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6343 ixgbe_ptp_start_cyclecounter(adapter);
6345 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
6346 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
6348 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
6350 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
6353 ((flow_rx && flow_tx) ? "RX/TX" :
6355 (flow_tx ? "TX" : "None"))));
6357 netif_carrier_on(netdev);
6358 ixgbe_check_vf_rate_limit(adapter);
6360 /* enable transmits */
6361 netif_tx_wake_all_queues(adapter->netdev);
6363 /* enable any upper devices */
6365 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
6366 if (netif_is_macvlan(upper)) {
6367 struct macvlan_dev *vlan = netdev_priv(upper);
6370 netif_tx_wake_all_queues(upper);
6375 /* update the default user priority for VFs */
6376 ixgbe_update_default_up(adapter);
6378 /* ping all the active vfs to let them know link has changed */
6379 ixgbe_ping_all_vfs(adapter);
6383 * ixgbe_watchdog_link_is_down - update netif_carrier status and
6384 * print link down message
6385 * @adapter: pointer to the adapter structure
6387 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
6389 struct net_device *netdev = adapter->netdev;
6390 struct ixgbe_hw *hw = &adapter->hw;
6392 adapter->link_up = false;
6393 adapter->link_speed = 0;
6395 /* only continue if link was up previously */
6396 if (!netif_carrier_ok(netdev))
6399 /* poll for SFP+ cable when link is down */
6400 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
6401 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
6403 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6404 ixgbe_ptp_start_cyclecounter(adapter);
6406 e_info(drv, "NIC Link is Down\n");
6407 netif_carrier_off(netdev);
6409 /* ping all the active vfs to let them know link has changed */
6410 ixgbe_ping_all_vfs(adapter);
6413 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
6417 for (i = 0; i < adapter->num_tx_queues; i++) {
6418 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6420 if (tx_ring->next_to_use != tx_ring->next_to_clean)
6427 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
6429 struct ixgbe_hw *hw = &adapter->hw;
6430 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
6431 u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
6435 if (!adapter->num_vfs)
6438 /* resetting the PF is only needed for MAC before X550 */
6439 if (hw->mac.type >= ixgbe_mac_X550)
6442 for (i = 0; i < adapter->num_vfs; i++) {
6443 for (j = 0; j < q_per_pool; j++) {
6446 h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
6447 t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
6458 * ixgbe_watchdog_flush_tx - flush queues on link down
6459 * @adapter: pointer to the device adapter structure
6461 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
6463 if (!netif_carrier_ok(adapter->netdev)) {
6464 if (ixgbe_ring_tx_pending(adapter) ||
6465 ixgbe_vf_tx_pending(adapter)) {
6466 /* We've lost link, so the controller stops DMA,
6467 * but we've got queued Tx work that's never going
6468 * to get done, so reset controller to flush Tx.
6469 * (Do the reset outside of interrupt context).
6471 e_warn(drv, "initiating reset to clear Tx work after link loss\n");
6472 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
6477 #ifdef CONFIG_PCI_IOV
6478 static inline void ixgbe_issue_vf_flr(struct ixgbe_adapter *adapter,
6479 struct pci_dev *vfdev)
6481 if (!pci_wait_for_pending_transaction(vfdev))
6482 e_dev_warn("Issuing VFLR with pending transactions\n");
6484 e_dev_err("Issuing VFLR for VF %s\n", pci_name(vfdev));
6485 pcie_capability_set_word(vfdev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
6490 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6492 struct ixgbe_hw *hw = &adapter->hw;
6493 struct pci_dev *pdev = adapter->pdev;
6494 struct pci_dev *vfdev;
6497 unsigned short vf_id;
6499 if (!(netif_carrier_ok(adapter->netdev)))
6502 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6503 if (gpc) /* If incrementing then no need for the check below */
6505 /* Check to see if a bad DMA write target from an errant or
6506 * malicious VF has caused a PCIe error. If so then we can
6507 * issue a VFLR to the offending VF(s) and then resume without
6508 * requesting a full slot reset.
6514 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
6518 /* get the device ID for the VF */
6519 pci_read_config_word(pdev, pos + PCI_SRIOV_VF_DID, &vf_id);
6521 /* check status reg for all VFs owned by this PF */
6522 vfdev = pci_get_device(pdev->vendor, vf_id, NULL);
6524 if (vfdev->is_virtfn && (vfdev->physfn == pdev)) {
6527 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
6528 if (status_reg & PCI_STATUS_REC_MASTER_ABORT)
6530 ixgbe_issue_vf_flr(adapter, vfdev);
6533 vfdev = pci_get_device(pdev->vendor, vf_id, vfdev);
6537 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6541 /* Do not perform spoof check for 82598 or if not in IOV mode */
6542 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
6543 adapter->num_vfs == 0)
6546 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6549 * ssvpc register is cleared on read, if zero then no
6550 * spoofed packets in the last interval.
6555 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
6558 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
6563 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
6566 #endif /* CONFIG_PCI_IOV */
6570 * ixgbe_watchdog_subtask - check and bring link up
6571 * @adapter: pointer to the device adapter structure
6573 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6575 /* if interface is down, removing or resetting, do nothing */
6576 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6577 test_bit(__IXGBE_REMOVING, &adapter->state) ||
6578 test_bit(__IXGBE_RESETTING, &adapter->state))
6581 ixgbe_watchdog_update_link(adapter);
6583 if (adapter->link_up)
6584 ixgbe_watchdog_link_is_up(adapter);
6586 ixgbe_watchdog_link_is_down(adapter);
6588 ixgbe_check_for_bad_vf(adapter);
6589 ixgbe_spoof_check(adapter);
6590 ixgbe_update_stats(adapter);
6592 ixgbe_watchdog_flush_tx(adapter);
6596 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6597 * @adapter: the ixgbe adapter structure
6599 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6601 struct ixgbe_hw *hw = &adapter->hw;
6604 /* not searching for SFP so there is nothing to do here */
6605 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6606 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6609 /* someone else is in init, wait until next service event */
6610 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6613 err = hw->phy.ops.identify_sfp(hw);
6614 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6617 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6618 /* If no cable is present, then we need to reset
6619 * the next time we find a good cable. */
6620 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6627 /* exit if reset not needed */
6628 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6631 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6634 * A module may be identified correctly, but the EEPROM may not have
6635 * support for that module. setup_sfp() will fail in that case, so
6636 * we should not allow that module to load.
6638 if (hw->mac.type == ixgbe_mac_82598EB)
6639 err = hw->phy.ops.reset(hw);
6641 err = hw->mac.ops.setup_sfp(hw);
6643 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6646 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6647 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6650 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6652 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6653 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6654 e_dev_err("failed to initialize because an unsupported "
6655 "SFP+ module type was detected.\n");
6656 e_dev_err("Reload the driver after installing a "
6657 "supported module.\n");
6658 unregister_netdev(adapter->netdev);
6663 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6664 * @adapter: the ixgbe adapter structure
6666 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6668 struct ixgbe_hw *hw = &adapter->hw;
6670 bool autoneg = false;
6672 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6675 /* someone else is in init, wait until next service event */
6676 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6679 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6681 speed = hw->phy.autoneg_advertised;
6682 if ((!speed) && (hw->mac.ops.get_link_capabilities)) {
6683 hw->mac.ops.get_link_capabilities(hw, &speed, &autoneg);
6685 /* setup the highest link when no autoneg */
6687 if (speed & IXGBE_LINK_SPEED_10GB_FULL)
6688 speed = IXGBE_LINK_SPEED_10GB_FULL;
6692 if (hw->mac.ops.setup_link)
6693 hw->mac.ops.setup_link(hw, speed, true);
6695 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6696 adapter->link_check_timeout = jiffies;
6697 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6701 * ixgbe_service_timer - Timer Call-back
6702 * @data: pointer to adapter cast into an unsigned long
6704 static void ixgbe_service_timer(unsigned long data)
6706 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6707 unsigned long next_event_offset;
6709 /* poll faster when waiting for link */
6710 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6711 next_event_offset = HZ / 10;
6713 next_event_offset = HZ * 2;
6715 /* Reset the timer */
6716 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6718 ixgbe_service_event_schedule(adapter);
6721 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6723 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6726 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6728 /* If we're already down, removing or resetting, just bail */
6729 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6730 test_bit(__IXGBE_REMOVING, &adapter->state) ||
6731 test_bit(__IXGBE_RESETTING, &adapter->state))
6734 ixgbe_dump(adapter);
6735 netdev_err(adapter->netdev, "Reset adapter\n");
6736 adapter->tx_timeout_count++;
6739 ixgbe_reinit_locked(adapter);
6744 * ixgbe_service_task - manages and runs subtasks
6745 * @work: pointer to work_struct containing our data
6747 static void ixgbe_service_task(struct work_struct *work)
6749 struct ixgbe_adapter *adapter = container_of(work,
6750 struct ixgbe_adapter,
6752 if (ixgbe_removed(adapter->hw.hw_addr)) {
6753 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
6755 ixgbe_down(adapter);
6758 ixgbe_service_event_complete(adapter);
6761 ixgbe_reset_subtask(adapter);
6762 ixgbe_sfp_detection_subtask(adapter);
6763 ixgbe_sfp_link_config_subtask(adapter);
6764 ixgbe_check_overtemp_subtask(adapter);
6765 ixgbe_watchdog_subtask(adapter);
6766 ixgbe_fdir_reinit_subtask(adapter);
6767 ixgbe_check_hang_subtask(adapter);
6769 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
6770 ixgbe_ptp_overflow_check(adapter);
6771 ixgbe_ptp_rx_hang(adapter);
6774 ixgbe_service_event_complete(adapter);
6777 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
6778 struct ixgbe_tx_buffer *first,
6781 struct sk_buff *skb = first->skb;
6782 u32 vlan_macip_lens, type_tucmd;
6783 u32 mss_l4len_idx, l4len;
6786 if (skb->ip_summed != CHECKSUM_PARTIAL)
6789 if (!skb_is_gso(skb))
6792 err = skb_cow_head(skb, 0);
6796 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6797 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6799 if (first->protocol == htons(ETH_P_IP)) {
6800 struct iphdr *iph = ip_hdr(skb);
6803 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6807 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6808 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6809 IXGBE_TX_FLAGS_CSUM |
6810 IXGBE_TX_FLAGS_IPV4;
6811 } else if (skb_is_gso_v6(skb)) {
6812 ipv6_hdr(skb)->payload_len = 0;
6813 tcp_hdr(skb)->check =
6814 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6815 &ipv6_hdr(skb)->daddr,
6817 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6818 IXGBE_TX_FLAGS_CSUM;
6821 /* compute header lengths */
6822 l4len = tcp_hdrlen(skb);
6823 *hdr_len = skb_transport_offset(skb) + l4len;
6825 /* update gso size and bytecount with header size */
6826 first->gso_segs = skb_shinfo(skb)->gso_segs;
6827 first->bytecount += (first->gso_segs - 1) * *hdr_len;
6829 /* mss_l4len_id: use 0 as index for TSO */
6830 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6831 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6833 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6834 vlan_macip_lens = skb_network_header_len(skb);
6835 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6836 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6838 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6844 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6845 struct ixgbe_tx_buffer *first)
6847 struct sk_buff *skb = first->skb;
6848 u32 vlan_macip_lens = 0;
6849 u32 mss_l4len_idx = 0;
6852 if (skb->ip_summed != CHECKSUM_PARTIAL) {
6853 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6854 !(first->tx_flags & IXGBE_TX_FLAGS_CC))
6858 switch (first->protocol) {
6859 case htons(ETH_P_IP):
6860 vlan_macip_lens |= skb_network_header_len(skb);
6861 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6862 l4_hdr = ip_hdr(skb)->protocol;
6864 case htons(ETH_P_IPV6):
6865 vlan_macip_lens |= skb_network_header_len(skb);
6866 l4_hdr = ipv6_hdr(skb)->nexthdr;
6869 if (unlikely(net_ratelimit())) {
6870 dev_warn(tx_ring->dev,
6871 "partial checksum but proto=%x!\n",
6879 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6880 mss_l4len_idx = tcp_hdrlen(skb) <<
6881 IXGBE_ADVTXD_L4LEN_SHIFT;
6884 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6885 mss_l4len_idx = sizeof(struct sctphdr) <<
6886 IXGBE_ADVTXD_L4LEN_SHIFT;
6889 mss_l4len_idx = sizeof(struct udphdr) <<
6890 IXGBE_ADVTXD_L4LEN_SHIFT;
6893 if (unlikely(net_ratelimit())) {
6894 dev_warn(tx_ring->dev,
6895 "partial checksum but l4 proto=%x!\n",
6901 /* update TX checksum flag */
6902 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
6905 /* vlan_macip_lens: MACLEN, VLAN tag */
6906 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6907 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6909 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
6910 type_tucmd, mss_l4len_idx);
6913 #define IXGBE_SET_FLAG(_input, _flag, _result) \
6914 ((_flag <= _result) ? \
6915 ((u32)(_input & _flag) * (_result / _flag)) : \
6916 ((u32)(_input & _flag) / (_flag / _result)))
6918 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
6920 /* set type for advanced descriptor with frame checksum insertion */
6921 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
6922 IXGBE_ADVTXD_DCMD_DEXT |
6923 IXGBE_ADVTXD_DCMD_IFCS;
6925 /* set HW vlan bit if vlan is present */
6926 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
6927 IXGBE_ADVTXD_DCMD_VLE);
6929 /* set segmentation enable bits for TSO/FSO */
6930 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
6931 IXGBE_ADVTXD_DCMD_TSE);
6933 /* set timestamp bit if present */
6934 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
6935 IXGBE_ADVTXD_MAC_TSTAMP);
6937 /* insert frame checksum */
6938 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
6943 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
6944 u32 tx_flags, unsigned int paylen)
6946 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
6948 /* enable L4 checksum for TSO and TX checksum offload */
6949 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6950 IXGBE_TX_FLAGS_CSUM,
6951 IXGBE_ADVTXD_POPTS_TXSM);
6953 /* enble IPv4 checksum for TSO */
6954 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6955 IXGBE_TX_FLAGS_IPV4,
6956 IXGBE_ADVTXD_POPTS_IXSM);
6959 * Check Context must be set if Tx switch is enabled, which it
6960 * always is for case where virtual functions are running
6962 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
6966 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6969 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6971 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6973 /* Herbert's original patch had:
6974 * smp_mb__after_netif_stop_queue();
6975 * but since that doesn't exist yet, just open code it.
6979 /* We need to check again in a case another CPU has just
6980 * made room available.
6982 if (likely(ixgbe_desc_unused(tx_ring) < size))
6985 /* A reprieve! - use start_queue because it doesn't call schedule */
6986 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6987 ++tx_ring->tx_stats.restart_queue;
6991 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6993 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6996 return __ixgbe_maybe_stop_tx(tx_ring, size);
6999 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
7002 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
7003 struct ixgbe_tx_buffer *first,
7006 struct sk_buff *skb = first->skb;
7007 struct ixgbe_tx_buffer *tx_buffer;
7008 union ixgbe_adv_tx_desc *tx_desc;
7009 struct skb_frag_struct *frag;
7011 unsigned int data_len, size;
7012 u32 tx_flags = first->tx_flags;
7013 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
7014 u16 i = tx_ring->next_to_use;
7016 tx_desc = IXGBE_TX_DESC(tx_ring, i);
7018 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
7020 size = skb_headlen(skb);
7021 data_len = skb->data_len;
7024 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
7025 if (data_len < sizeof(struct fcoe_crc_eof)) {
7026 size -= sizeof(struct fcoe_crc_eof) - data_len;
7029 data_len -= sizeof(struct fcoe_crc_eof);
7034 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
7038 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
7039 if (dma_mapping_error(tx_ring->dev, dma))
7042 /* record length, and DMA address */
7043 dma_unmap_len_set(tx_buffer, len, size);
7044 dma_unmap_addr_set(tx_buffer, dma, dma);
7046 tx_desc->read.buffer_addr = cpu_to_le64(dma);
7048 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
7049 tx_desc->read.cmd_type_len =
7050 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
7054 if (i == tx_ring->count) {
7055 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
7058 tx_desc->read.olinfo_status = 0;
7060 dma += IXGBE_MAX_DATA_PER_TXD;
7061 size -= IXGBE_MAX_DATA_PER_TXD;
7063 tx_desc->read.buffer_addr = cpu_to_le64(dma);
7066 if (likely(!data_len))
7069 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
7073 if (i == tx_ring->count) {
7074 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
7077 tx_desc->read.olinfo_status = 0;
7080 size = min_t(unsigned int, data_len, skb_frag_size(frag));
7082 size = skb_frag_size(frag);
7086 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
7089 tx_buffer = &tx_ring->tx_buffer_info[i];
7092 /* write last descriptor with RS and EOP bits */
7093 cmd_type |= size | IXGBE_TXD_CMD;
7094 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
7096 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
7098 /* set the timestamp */
7099 first->time_stamp = jiffies;
7102 * Force memory writes to complete before letting h/w know there
7103 * are new descriptors to fetch. (Only applicable for weak-ordered
7104 * memory model archs, such as IA-64).
7106 * We also need this memory barrier to make certain all of the
7107 * status bits have been updated before next_to_watch is written.
7111 /* set next_to_watch value indicating a packet is present */
7112 first->next_to_watch = tx_desc;
7115 if (i == tx_ring->count)
7118 tx_ring->next_to_use = i;
7120 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
7122 if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
7123 writel(i, tx_ring->tail);
7125 /* we need this if more than one processor can write to our tail
7126 * at a time, it synchronizes IO on IA64/Altix systems
7133 dev_err(tx_ring->dev, "TX DMA map failed\n");
7135 /* clear dma mappings for failed tx_buffer_info map */
7137 tx_buffer = &tx_ring->tx_buffer_info[i];
7138 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
7139 if (tx_buffer == first)
7146 tx_ring->next_to_use = i;
7149 static void ixgbe_atr(struct ixgbe_ring *ring,
7150 struct ixgbe_tx_buffer *first)
7152 struct ixgbe_q_vector *q_vector = ring->q_vector;
7153 union ixgbe_atr_hash_dword input = { .dword = 0 };
7154 union ixgbe_atr_hash_dword common = { .dword = 0 };
7156 unsigned char *network;
7158 struct ipv6hdr *ipv6;
7163 /* if ring doesn't have a interrupt vector, cannot perform ATR */
7167 /* do nothing if sampling is disabled */
7168 if (!ring->atr_sample_rate)
7173 /* snag network header to get L4 type and address */
7174 hdr.network = skb_network_header(first->skb);
7176 /* Currently only IPv4/IPv6 with TCP is supported */
7177 if ((first->protocol != htons(ETH_P_IPV6) ||
7178 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
7179 (first->protocol != htons(ETH_P_IP) ||
7180 hdr.ipv4->protocol != IPPROTO_TCP))
7183 th = tcp_hdr(first->skb);
7185 /* skip this packet since it is invalid or the socket is closing */
7189 /* sample on all syn packets or once every atr sample count */
7190 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
7193 /* reset sample count */
7194 ring->atr_count = 0;
7196 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
7199 * src and dst are inverted, think how the receiver sees them
7201 * The input is broken into two sections, a non-compressed section
7202 * containing vm_pool, vlan_id, and flow_type. The rest of the data
7203 * is XORed together and stored in the compressed dword.
7205 input.formatted.vlan_id = vlan_id;
7208 * since src port and flex bytes occupy the same word XOR them together
7209 * and write the value to source port portion of compressed dword
7211 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
7212 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
7214 common.port.src ^= th->dest ^ first->protocol;
7215 common.port.dst ^= th->source;
7217 if (first->protocol == htons(ETH_P_IP)) {
7218 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
7219 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
7221 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
7222 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
7223 hdr.ipv6->saddr.s6_addr32[1] ^
7224 hdr.ipv6->saddr.s6_addr32[2] ^
7225 hdr.ipv6->saddr.s6_addr32[3] ^
7226 hdr.ipv6->daddr.s6_addr32[0] ^
7227 hdr.ipv6->daddr.s6_addr32[1] ^
7228 hdr.ipv6->daddr.s6_addr32[2] ^
7229 hdr.ipv6->daddr.s6_addr32[3];
7232 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
7233 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
7234 input, common, ring->queue_index);
7237 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
7238 void *accel_priv, select_queue_fallback_t fallback)
7240 struct ixgbe_fwd_adapter *fwd_adapter = accel_priv;
7242 struct ixgbe_adapter *adapter;
7243 struct ixgbe_ring_feature *f;
7248 return skb->queue_mapping + fwd_adapter->tx_base_queue;
7253 * only execute the code below if protocol is FCoE
7254 * or FIP and we have FCoE enabled on the adapter
7256 switch (vlan_get_protocol(skb)) {
7257 case htons(ETH_P_FCOE):
7258 case htons(ETH_P_FIP):
7259 adapter = netdev_priv(dev);
7261 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7264 return fallback(dev, skb);
7267 f = &adapter->ring_feature[RING_F_FCOE];
7269 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
7272 while (txq >= f->indices)
7275 return txq + f->offset;
7277 return fallback(dev, skb);
7281 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
7282 struct ixgbe_adapter *adapter,
7283 struct ixgbe_ring *tx_ring)
7285 struct ixgbe_tx_buffer *first;
7289 u16 count = TXD_USE_COUNT(skb_headlen(skb));
7290 __be16 protocol = skb->protocol;
7294 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
7295 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
7296 * + 2 desc gap to keep tail from touching head,
7297 * + 1 desc for context descriptor,
7298 * otherwise try next time
7300 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
7301 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
7303 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
7304 tx_ring->tx_stats.tx_busy++;
7305 return NETDEV_TX_BUSY;
7308 /* record the location of the first descriptor for this packet */
7309 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
7311 first->bytecount = skb->len;
7312 first->gso_segs = 1;
7314 /* if we have a HW VLAN tag being added default to the HW one */
7315 if (skb_vlan_tag_present(skb)) {
7316 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
7317 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7318 /* else if it is a SW VLAN check the next protocol and store the tag */
7319 } else if (protocol == htons(ETH_P_8021Q)) {
7320 struct vlan_hdr *vhdr, _vhdr;
7321 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
7325 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
7326 IXGBE_TX_FLAGS_VLAN_SHIFT;
7327 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
7329 protocol = vlan_get_protocol(skb);
7331 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
7332 adapter->ptp_clock &&
7333 !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
7335 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
7336 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
7338 /* schedule check for Tx timestamp */
7339 adapter->ptp_tx_skb = skb_get(skb);
7340 adapter->ptp_tx_start = jiffies;
7341 schedule_work(&adapter->ptp_tx_work);
7344 skb_tx_timestamp(skb);
7346 #ifdef CONFIG_PCI_IOV
7348 * Use the l2switch_enable flag - would be false if the DMA
7349 * Tx switch had been disabled.
7351 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7352 tx_flags |= IXGBE_TX_FLAGS_CC;
7355 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
7356 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7357 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
7358 (skb->priority != TC_PRIO_CONTROL))) {
7359 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
7360 tx_flags |= (skb->priority & 0x7) <<
7361 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
7362 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
7363 struct vlan_ethhdr *vhdr;
7365 if (skb_cow_head(skb, 0))
7367 vhdr = (struct vlan_ethhdr *)skb->data;
7368 vhdr->h_vlan_TCI = htons(tx_flags >>
7369 IXGBE_TX_FLAGS_VLAN_SHIFT);
7371 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7375 /* record initial flags and protocol */
7376 first->tx_flags = tx_flags;
7377 first->protocol = protocol;
7380 /* setup tx offload for FCoE */
7381 if ((protocol == htons(ETH_P_FCOE)) &&
7382 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
7383 tso = ixgbe_fso(tx_ring, first, &hdr_len);
7390 #endif /* IXGBE_FCOE */
7391 tso = ixgbe_tso(tx_ring, first, &hdr_len);
7395 ixgbe_tx_csum(tx_ring, first);
7397 /* add the ATR filter if ATR is on */
7398 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
7399 ixgbe_atr(tx_ring, first);
7403 #endif /* IXGBE_FCOE */
7404 ixgbe_tx_map(tx_ring, first, hdr_len);
7406 return NETDEV_TX_OK;
7409 dev_kfree_skb_any(first->skb);
7412 return NETDEV_TX_OK;
7415 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
7416 struct net_device *netdev,
7417 struct ixgbe_ring *ring)
7419 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7420 struct ixgbe_ring *tx_ring;
7423 * The minimum packet size for olinfo paylen is 17 so pad the skb
7424 * in order to meet this minimum size requirement.
7426 if (skb_put_padto(skb, 17))
7427 return NETDEV_TX_OK;
7429 tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
7431 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
7434 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
7435 struct net_device *netdev)
7437 return __ixgbe_xmit_frame(skb, netdev, NULL);
7441 * ixgbe_set_mac - Change the Ethernet Address of the NIC
7442 * @netdev: network interface device structure
7443 * @p: pointer to an address structure
7445 * Returns 0 on success, negative on failure
7447 static int ixgbe_set_mac(struct net_device *netdev, void *p)
7449 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7450 struct ixgbe_hw *hw = &adapter->hw;
7451 struct sockaddr *addr = p;
7454 if (!is_valid_ether_addr(addr->sa_data))
7455 return -EADDRNOTAVAIL;
7457 ixgbe_del_mac_filter(adapter, hw->mac.addr, VMDQ_P(0));
7458 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
7459 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
7461 ret = ixgbe_add_mac_filter(adapter, hw->mac.addr, VMDQ_P(0));
7462 return ret > 0 ? 0 : ret;
7466 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
7468 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7469 struct ixgbe_hw *hw = &adapter->hw;
7473 if (prtad != hw->phy.mdio.prtad)
7475 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
7481 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
7482 u16 addr, u16 value)
7484 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7485 struct ixgbe_hw *hw = &adapter->hw;
7487 if (prtad != hw->phy.mdio.prtad)
7489 return hw->phy.ops.write_reg(hw, addr, devad, value);
7492 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
7494 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7498 return ixgbe_ptp_set_ts_config(adapter, req);
7500 return ixgbe_ptp_get_ts_config(adapter, req);
7502 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
7507 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
7509 * @netdev: network interface device structure
7511 * Returns non-zero on failure
7513 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
7516 struct ixgbe_adapter *adapter = netdev_priv(dev);
7517 struct ixgbe_hw *hw = &adapter->hw;
7519 if (is_valid_ether_addr(hw->mac.san_addr)) {
7521 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
7524 /* update SAN MAC vmdq pool selection */
7525 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
7531 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
7533 * @netdev: network interface device structure
7535 * Returns non-zero on failure
7537 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
7540 struct ixgbe_adapter *adapter = netdev_priv(dev);
7541 struct ixgbe_mac_info *mac = &adapter->hw.mac;
7543 if (is_valid_ether_addr(mac->san_addr)) {
7545 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
7551 #ifdef CONFIG_NET_POLL_CONTROLLER
7553 * Polling 'interrupt' - used by things like netconsole to send skbs
7554 * without having to re-enable interrupts. It's not called while
7555 * the interrupt routine is executing.
7557 static void ixgbe_netpoll(struct net_device *netdev)
7559 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7562 /* if interface is down do nothing */
7563 if (test_bit(__IXGBE_DOWN, &adapter->state))
7566 /* loop through and schedule all active queues */
7567 for (i = 0; i < adapter->num_q_vectors; i++)
7568 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
7572 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7573 struct rtnl_link_stats64 *stats)
7575 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7579 for (i = 0; i < adapter->num_rx_queues; i++) {
7580 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7586 start = u64_stats_fetch_begin_irq(&ring->syncp);
7587 packets = ring->stats.packets;
7588 bytes = ring->stats.bytes;
7589 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
7590 stats->rx_packets += packets;
7591 stats->rx_bytes += bytes;
7595 for (i = 0; i < adapter->num_tx_queues; i++) {
7596 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7602 start = u64_stats_fetch_begin_irq(&ring->syncp);
7603 packets = ring->stats.packets;
7604 bytes = ring->stats.bytes;
7605 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
7606 stats->tx_packets += packets;
7607 stats->tx_bytes += bytes;
7611 /* following stats updated by ixgbe_watchdog_task() */
7612 stats->multicast = netdev->stats.multicast;
7613 stats->rx_errors = netdev->stats.rx_errors;
7614 stats->rx_length_errors = netdev->stats.rx_length_errors;
7615 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
7616 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
7620 #ifdef CONFIG_IXGBE_DCB
7622 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7623 * @adapter: pointer to ixgbe_adapter
7624 * @tc: number of traffic classes currently enabled
7626 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7627 * 802.1Q priority maps to a packet buffer that exists.
7629 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
7631 struct ixgbe_hw *hw = &adapter->hw;
7635 /* 82598 have a static priority to TC mapping that can not
7636 * be changed so no validation is needed.
7638 if (hw->mac.type == ixgbe_mac_82598EB)
7641 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
7644 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
7645 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
7647 /* If up2tc is out of bounds default to zero */
7649 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
7653 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
7659 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
7660 * @adapter: Pointer to adapter struct
7662 * Populate the netdev user priority to tc map
7664 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
7666 struct net_device *dev = adapter->netdev;
7667 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
7668 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
7671 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
7674 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
7675 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
7677 tc = ets->prio_tc[prio];
7679 netdev_set_prio_tc_map(dev, prio, tc);
7683 #endif /* CONFIG_IXGBE_DCB */
7685 * ixgbe_setup_tc - configure net_device for multiple traffic classes
7687 * @netdev: net device to configure
7688 * @tc: number of traffic classes to enable
7690 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7692 struct ixgbe_adapter *adapter = netdev_priv(dev);
7693 struct ixgbe_hw *hw = &adapter->hw;
7696 /* Hardware supports up to 8 traffic classes */
7697 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
7698 (hw->mac.type == ixgbe_mac_82598EB &&
7699 tc < MAX_TRAFFIC_CLASS))
7702 pools = (find_first_zero_bit(&adapter->fwd_bitmask, 32) > 1);
7703 if (tc && pools && adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS)
7706 /* Hardware has to reinitialize queues and interrupts to
7707 * match packet buffer alignment. Unfortunately, the
7708 * hardware is not flexible enough to do this dynamically.
7710 if (netif_running(dev))
7712 ixgbe_clear_interrupt_scheme(adapter);
7714 #ifdef CONFIG_IXGBE_DCB
7716 netdev_set_num_tc(dev, tc);
7717 ixgbe_set_prio_tc_map(adapter);
7719 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7721 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
7722 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
7723 adapter->hw.fc.requested_mode = ixgbe_fc_none;
7726 netdev_reset_tc(dev);
7728 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7729 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7731 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7733 adapter->temp_dcb_cfg.pfc_mode_enable = false;
7734 adapter->dcb_cfg.pfc_mode_enable = false;
7737 ixgbe_validate_rtr(adapter, tc);
7739 #endif /* CONFIG_IXGBE_DCB */
7740 ixgbe_init_interrupt_scheme(adapter);
7742 if (netif_running(dev))
7743 return ixgbe_open(dev);
7748 #ifdef CONFIG_PCI_IOV
7749 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
7751 struct net_device *netdev = adapter->netdev;
7754 ixgbe_setup_tc(netdev, netdev_get_num_tc(netdev));
7759 void ixgbe_do_reset(struct net_device *netdev)
7761 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7763 if (netif_running(netdev))
7764 ixgbe_reinit_locked(adapter);
7766 ixgbe_reset(adapter);
7769 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
7770 netdev_features_t features)
7772 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7774 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7775 if (!(features & NETIF_F_RXCSUM))
7776 features &= ~NETIF_F_LRO;
7778 /* Turn off LRO if not RSC capable */
7779 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
7780 features &= ~NETIF_F_LRO;
7785 static int ixgbe_set_features(struct net_device *netdev,
7786 netdev_features_t features)
7788 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7789 netdev_features_t changed = netdev->features ^ features;
7790 bool need_reset = false;
7792 /* Make sure RSC matches LRO, reset if change */
7793 if (!(features & NETIF_F_LRO)) {
7794 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7796 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
7797 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
7798 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7799 if (adapter->rx_itr_setting == 1 ||
7800 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
7801 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
7803 } else if ((changed ^ features) & NETIF_F_LRO) {
7804 e_info(probe, "rx-usecs set too low, "
7810 * Check if Flow Director n-tuple support was enabled or disabled. If
7811 * the state changed, we need to reset.
7813 switch (features & NETIF_F_NTUPLE) {
7814 case NETIF_F_NTUPLE:
7815 /* turn off ATR, enable perfect filters and reset */
7816 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
7819 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7820 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7823 /* turn off perfect filters, enable ATR and reset */
7824 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7827 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7829 /* We cannot enable ATR if SR-IOV is enabled */
7830 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7833 /* We cannot enable ATR if we have 2 or more traffic classes */
7834 if (netdev_get_num_tc(netdev) > 1)
7837 /* We cannot enable ATR if RSS is disabled */
7838 if (adapter->ring_feature[RING_F_RSS].limit <= 1)
7841 /* A sample rate of 0 indicates ATR disabled */
7842 if (!adapter->atr_sample_rate)
7845 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
7849 if (features & NETIF_F_HW_VLAN_CTAG_RX)
7850 ixgbe_vlan_strip_enable(adapter);
7852 ixgbe_vlan_strip_disable(adapter);
7854 if (changed & NETIF_F_RXALL)
7857 netdev->features = features;
7859 ixgbe_do_reset(netdev);
7865 * ixgbe_add_vxlan_port - Get notifications about VXLAN ports that come up
7866 * @dev: The port's netdev
7867 * @sa_family: Socket Family that VXLAN is notifiying us about
7868 * @port: New UDP port number that VXLAN started listening to
7870 static void ixgbe_add_vxlan_port(struct net_device *dev, sa_family_t sa_family,
7873 struct ixgbe_adapter *adapter = netdev_priv(dev);
7874 struct ixgbe_hw *hw = &adapter->hw;
7875 u16 new_port = ntohs(port);
7877 if (sa_family == AF_INET6)
7880 if (adapter->vxlan_port == new_port) {
7881 netdev_info(dev, "Port %d already offloaded\n", new_port);
7885 if (adapter->vxlan_port) {
7887 "Hit Max num of UDP ports, not adding port %d\n",
7892 adapter->vxlan_port = new_port;
7893 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, new_port);
7897 * ixgbe_del_vxlan_port - Get notifications about VXLAN ports that go away
7898 * @dev: The port's netdev
7899 * @sa_family: Socket Family that VXLAN is notifying us about
7900 * @port: UDP port number that VXLAN stopped listening to
7902 static void ixgbe_del_vxlan_port(struct net_device *dev, sa_family_t sa_family,
7905 struct ixgbe_adapter *adapter = netdev_priv(dev);
7906 struct ixgbe_hw *hw = &adapter->hw;
7907 u16 new_port = ntohs(port);
7909 if (sa_family == AF_INET6)
7912 if (adapter->vxlan_port != new_port) {
7913 netdev_info(dev, "Port %d was not found, not deleting\n",
7918 adapter->vxlan_port = 0;
7919 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, 0);
7922 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
7923 struct net_device *dev,
7924 const unsigned char *addr, u16 vid,
7927 /* guarantee we can provide a unique filter for the unicast address */
7928 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
7929 if (IXGBE_MAX_PF_MACVLANS <= netdev_uc_count(dev))
7933 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
7937 * ixgbe_configure_bridge_mode - set various bridge modes
7938 * @adapter - the private structure
7939 * @mode - requested bridge mode
7941 * Configure some settings require for various bridge modes.
7943 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
7946 struct ixgbe_hw *hw = &adapter->hw;
7947 unsigned int p, num_pools;
7951 case BRIDGE_MODE_VEPA:
7952 /* disable Tx loopback, rely on switch hairpin mode */
7953 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
7955 /* must enable Rx switching replication to allow multicast
7956 * packet reception on all VFs, and to enable source address
7959 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
7960 vmdctl |= IXGBE_VT_CTL_REPLEN;
7961 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
7963 /* enable Rx source address pruning. Note, this requires
7964 * replication to be enabled or else it does nothing.
7966 num_pools = adapter->num_vfs + adapter->num_rx_pools;
7967 for (p = 0; p < num_pools; p++) {
7968 if (hw->mac.ops.set_source_address_pruning)
7969 hw->mac.ops.set_source_address_pruning(hw,
7974 case BRIDGE_MODE_VEB:
7975 /* enable Tx loopback for internal VF/PF communication */
7976 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
7977 IXGBE_PFDTXGSWC_VT_LBEN);
7979 /* disable Rx switching replication unless we have SR-IOV
7982 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
7983 if (!adapter->num_vfs)
7984 vmdctl &= ~IXGBE_VT_CTL_REPLEN;
7985 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
7987 /* disable Rx source address pruning, since we don't expect to
7988 * be receiving external loopback of our transmitted frames.
7990 num_pools = adapter->num_vfs + adapter->num_rx_pools;
7991 for (p = 0; p < num_pools; p++) {
7992 if (hw->mac.ops.set_source_address_pruning)
7993 hw->mac.ops.set_source_address_pruning(hw,
8002 adapter->bridge_mode = mode;
8004 e_info(drv, "enabling bridge mode: %s\n",
8005 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
8010 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
8011 struct nlmsghdr *nlh, u16 flags)
8013 struct ixgbe_adapter *adapter = netdev_priv(dev);
8014 struct nlattr *attr, *br_spec;
8017 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
8020 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
8024 nla_for_each_nested(attr, br_spec, rem) {
8028 if (nla_type(attr) != IFLA_BRIDGE_MODE)
8031 if (nla_len(attr) < sizeof(mode))
8034 mode = nla_get_u16(attr);
8035 status = ixgbe_configure_bridge_mode(adapter, mode);
8045 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
8046 struct net_device *dev,
8047 u32 filter_mask, int nlflags)
8049 struct ixgbe_adapter *adapter = netdev_priv(dev);
8051 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
8054 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
8055 adapter->bridge_mode, 0, 0, nlflags);
8058 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
8060 struct ixgbe_fwd_adapter *fwd_adapter = NULL;
8061 struct ixgbe_adapter *adapter = netdev_priv(pdev);
8062 int used_pools = adapter->num_vfs + adapter->num_rx_pools;
8066 /* Hardware has a limited number of available pools. Each VF, and the
8067 * PF require a pool. Check to ensure we don't attempt to use more
8068 * then the available number of pools.
8070 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
8071 return ERR_PTR(-EINVAL);
8074 if (vdev->num_rx_queues != vdev->num_tx_queues) {
8075 netdev_info(pdev, "%s: Only supports a single queue count for TX and RX\n",
8077 return ERR_PTR(-EINVAL);
8080 /* Check for hardware restriction on number of rx/tx queues */
8081 if (vdev->num_tx_queues > IXGBE_MAX_L2A_QUEUES ||
8082 vdev->num_tx_queues == IXGBE_BAD_L2A_QUEUE) {
8084 "%s: Supports RX/TX Queue counts 1,2, and 4\n",
8086 return ERR_PTR(-EINVAL);
8089 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8090 adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS - 1) ||
8091 (adapter->num_rx_pools > IXGBE_MAX_MACVLANS))
8092 return ERR_PTR(-EBUSY);
8094 fwd_adapter = kcalloc(1, sizeof(struct ixgbe_fwd_adapter), GFP_KERNEL);
8096 return ERR_PTR(-ENOMEM);
8098 pool = find_first_zero_bit(&adapter->fwd_bitmask, 32);
8099 adapter->num_rx_pools++;
8100 set_bit(pool, &adapter->fwd_bitmask);
8101 limit = find_last_bit(&adapter->fwd_bitmask, 32);
8103 /* Enable VMDq flag so device will be set in VM mode */
8104 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED | IXGBE_FLAG_SRIOV_ENABLED;
8105 adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
8106 adapter->ring_feature[RING_F_RSS].limit = vdev->num_tx_queues;
8108 /* Force reinit of ring allocation with VMDQ enabled */
8109 err = ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
8112 fwd_adapter->pool = pool;
8113 fwd_adapter->real_adapter = adapter;
8114 err = ixgbe_fwd_ring_up(vdev, fwd_adapter);
8117 netif_tx_start_all_queues(vdev);
8120 /* unwind counter and free adapter struct */
8122 "%s: dfwd hardware acceleration failed\n", vdev->name);
8123 clear_bit(pool, &adapter->fwd_bitmask);
8124 adapter->num_rx_pools--;
8126 return ERR_PTR(err);
8129 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
8131 struct ixgbe_fwd_adapter *fwd_adapter = priv;
8132 struct ixgbe_adapter *adapter = fwd_adapter->real_adapter;
8135 clear_bit(fwd_adapter->pool, &adapter->fwd_bitmask);
8136 adapter->num_rx_pools--;
8138 limit = find_last_bit(&adapter->fwd_bitmask, 32);
8139 adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
8140 ixgbe_fwd_ring_down(fwd_adapter->netdev, fwd_adapter);
8141 ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
8142 netdev_dbg(pdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
8143 fwd_adapter->pool, adapter->num_rx_pools,
8144 fwd_adapter->rx_base_queue,
8145 fwd_adapter->rx_base_queue + adapter->num_rx_queues_per_pool,
8146 adapter->fwd_bitmask);
8150 static const struct net_device_ops ixgbe_netdev_ops = {
8151 .ndo_open = ixgbe_open,
8152 .ndo_stop = ixgbe_close,
8153 .ndo_start_xmit = ixgbe_xmit_frame,
8154 .ndo_select_queue = ixgbe_select_queue,
8155 .ndo_set_rx_mode = ixgbe_set_rx_mode,
8156 .ndo_validate_addr = eth_validate_addr,
8157 .ndo_set_mac_address = ixgbe_set_mac,
8158 .ndo_change_mtu = ixgbe_change_mtu,
8159 .ndo_tx_timeout = ixgbe_tx_timeout,
8160 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
8161 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
8162 .ndo_do_ioctl = ixgbe_ioctl,
8163 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
8164 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
8165 .ndo_set_vf_rate = ixgbe_ndo_set_vf_bw,
8166 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
8167 .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
8168 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
8169 .ndo_get_stats64 = ixgbe_get_stats64,
8170 #ifdef CONFIG_IXGBE_DCB
8171 .ndo_setup_tc = ixgbe_setup_tc,
8173 #ifdef CONFIG_NET_POLL_CONTROLLER
8174 .ndo_poll_controller = ixgbe_netpoll,
8176 #ifdef CONFIG_NET_RX_BUSY_POLL
8177 .ndo_busy_poll = ixgbe_low_latency_recv,
8180 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
8181 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
8182 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
8183 .ndo_fcoe_enable = ixgbe_fcoe_enable,
8184 .ndo_fcoe_disable = ixgbe_fcoe_disable,
8185 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
8186 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
8187 #endif /* IXGBE_FCOE */
8188 .ndo_set_features = ixgbe_set_features,
8189 .ndo_fix_features = ixgbe_fix_features,
8190 .ndo_fdb_add = ixgbe_ndo_fdb_add,
8191 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
8192 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
8193 .ndo_dfwd_add_station = ixgbe_fwd_add,
8194 .ndo_dfwd_del_station = ixgbe_fwd_del,
8195 .ndo_add_vxlan_port = ixgbe_add_vxlan_port,
8196 .ndo_del_vxlan_port = ixgbe_del_vxlan_port,
8200 * ixgbe_enumerate_functions - Get the number of ports this device has
8201 * @adapter: adapter structure
8203 * This function enumerates the phsyical functions co-located on a single slot,
8204 * in order to determine how many ports a device has. This is most useful in
8205 * determining the required GT/s of PCIe bandwidth necessary for optimal
8208 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
8210 struct pci_dev *entry, *pdev = adapter->pdev;
8213 /* Some cards can not use the generic count PCIe functions method,
8214 * because they are behind a parent switch, so we hardcode these with
8215 * the correct number of functions.
8217 if (ixgbe_pcie_from_parent(&adapter->hw))
8220 list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
8221 /* don't count virtual functions */
8222 if (entry->is_virtfn)
8225 /* When the devices on the bus don't all match our device ID,
8226 * we can't reliably determine the correct number of
8227 * functions. This can occur if a function has been direct
8228 * attached to a virtual machine using VT-d, for example. In
8229 * this case, simply return -1 to indicate this.
8231 if ((entry->vendor != pdev->vendor) ||
8232 (entry->device != pdev->device))
8242 * ixgbe_wol_supported - Check whether device supports WoL
8243 * @hw: hw specific details
8244 * @device_id: the device ID
8245 * @subdev_id: the subsystem device ID
8247 * This function is used by probe and ethtool to determine
8248 * which devices have WoL support
8251 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
8254 struct ixgbe_hw *hw = &adapter->hw;
8255 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
8256 int is_wol_supported = 0;
8258 switch (device_id) {
8259 case IXGBE_DEV_ID_82599_SFP:
8260 /* Only these subdevices could supports WOL */
8261 switch (subdevice_id) {
8262 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
8263 case IXGBE_SUBDEV_ID_82599_560FLR:
8264 /* only support first port */
8265 if (hw->bus.func != 0)
8267 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
8268 case IXGBE_SUBDEV_ID_82599_SFP:
8269 case IXGBE_SUBDEV_ID_82599_RNDC:
8270 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
8271 case IXGBE_SUBDEV_ID_82599_LOM_SFP:
8272 is_wol_supported = 1;
8276 case IXGBE_DEV_ID_82599EN_SFP:
8277 /* Only this subdevice supports WOL */
8278 switch (subdevice_id) {
8279 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
8280 is_wol_supported = 1;
8284 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
8285 /* All except this subdevice support WOL */
8286 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
8287 is_wol_supported = 1;
8289 case IXGBE_DEV_ID_82599_KX4:
8290 is_wol_supported = 1;
8292 case IXGBE_DEV_ID_X540T:
8293 case IXGBE_DEV_ID_X540T1:
8294 /* check eeprom to see if enabled wol */
8295 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
8296 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
8297 (hw->bus.func == 0))) {
8298 is_wol_supported = 1;
8303 return is_wol_supported;
8307 * ixgbe_get_platform_mac_addr - Look up MAC address in Open Firmware / IDPROM
8308 * @adapter: Pointer to adapter struct
8310 static void ixgbe_get_platform_mac_addr(struct ixgbe_adapter *adapter)
8313 struct device_node *dp = pci_device_to_OF_node(adapter->pdev);
8314 struct ixgbe_hw *hw = &adapter->hw;
8315 const unsigned char *addr;
8317 addr = of_get_mac_address(dp);
8319 ether_addr_copy(hw->mac.perm_addr, addr);
8322 #endif /* CONFIG_OF */
8325 ether_addr_copy(hw->mac.perm_addr, idprom->id_ethaddr);
8326 #endif /* CONFIG_SPARC */
8330 * ixgbe_probe - Device Initialization Routine
8331 * @pdev: PCI device information struct
8332 * @ent: entry in ixgbe_pci_tbl
8334 * Returns 0 on success, negative on failure
8336 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
8337 * The OS initialization, configuring of the adapter private structure,
8338 * and a hardware reset occur.
8340 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
8342 struct net_device *netdev;
8343 struct ixgbe_adapter *adapter = NULL;
8344 struct ixgbe_hw *hw;
8345 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
8346 int i, err, pci_using_dac, expected_gts;
8347 unsigned int indices = MAX_TX_QUEUES;
8348 u8 part_str[IXGBE_PBANUM_LENGTH];
8349 bool disable_dev = false;
8355 /* Catch broken hardware that put the wrong VF device ID in
8356 * the PCIe SR-IOV capability.
8358 if (pdev->is_virtfn) {
8359 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
8360 pci_name(pdev), pdev->vendor, pdev->device);
8364 err = pci_enable_device_mem(pdev);
8368 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
8371 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
8374 "No usable DMA configuration, aborting\n");
8380 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
8381 IORESOURCE_MEM), ixgbe_driver_name);
8384 "pci_request_selected_regions failed 0x%x\n", err);
8388 pci_enable_pcie_error_reporting(pdev);
8390 pci_set_master(pdev);
8391 pci_save_state(pdev);
8393 if (ii->mac == ixgbe_mac_82598EB) {
8394 #ifdef CONFIG_IXGBE_DCB
8395 /* 8 TC w/ 4 queues per TC */
8396 indices = 4 * MAX_TRAFFIC_CLASS;
8398 indices = IXGBE_MAX_RSS_INDICES;
8402 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
8405 goto err_alloc_etherdev;
8408 SET_NETDEV_DEV(netdev, &pdev->dev);
8410 adapter = netdev_priv(netdev);
8412 adapter->netdev = netdev;
8413 adapter->pdev = pdev;
8416 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
8418 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
8419 pci_resource_len(pdev, 0));
8420 adapter->io_addr = hw->hw_addr;
8426 netdev->netdev_ops = &ixgbe_netdev_ops;
8427 ixgbe_set_ethtool_ops(netdev);
8428 netdev->watchdog_timeo = 5 * HZ;
8429 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
8432 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
8433 hw->mac.type = ii->mac;
8436 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
8437 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
8438 if (ixgbe_removed(hw->hw_addr)) {
8442 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
8443 if (!(eec & (1 << 8)))
8444 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
8447 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
8448 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
8449 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
8450 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
8451 hw->phy.mdio.mmds = 0;
8452 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
8453 hw->phy.mdio.dev = netdev;
8454 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
8455 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
8457 ii->get_invariants(hw);
8459 /* setup the private structure */
8460 err = ixgbe_sw_init(adapter);
8464 /* Make it possible the adapter to be woken up via WOL */
8465 switch (adapter->hw.mac.type) {
8466 case ixgbe_mac_82599EB:
8467 case ixgbe_mac_X540:
8468 case ixgbe_mac_X550:
8469 case ixgbe_mac_X550EM_x:
8470 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
8477 * If there is a fan on this device and it has failed log the
8480 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
8481 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
8482 if (esdp & IXGBE_ESDP_SDP1)
8483 e_crit(probe, "Fan has stopped, replace the adapter\n");
8486 if (allow_unsupported_sfp)
8487 hw->allow_unsupported_sfp = allow_unsupported_sfp;
8489 /* reset_hw fills in the perm_addr as well */
8490 hw->phy.reset_if_overtemp = true;
8491 err = hw->mac.ops.reset_hw(hw);
8492 hw->phy.reset_if_overtemp = false;
8493 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
8494 hw->mac.type == ixgbe_mac_82598EB) {
8496 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
8497 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
8498 e_dev_err("Reload the driver after installing a supported module.\n");
8501 e_dev_err("HW Init failed: %d\n", err);
8505 #ifdef CONFIG_PCI_IOV
8506 /* SR-IOV not supported on the 82598 */
8507 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
8510 ixgbe_init_mbx_params_pf(hw);
8511 memcpy(&hw->mbx.ops, ii->mbx_ops, sizeof(hw->mbx.ops));
8512 pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
8513 ixgbe_enable_sriov(adapter);
8517 netdev->features = NETIF_F_SG |
8520 NETIF_F_HW_VLAN_CTAG_TX |
8521 NETIF_F_HW_VLAN_CTAG_RX |
8527 netdev->hw_features = netdev->features | NETIF_F_HW_L2FW_DOFFLOAD;
8529 switch (adapter->hw.mac.type) {
8530 case ixgbe_mac_82599EB:
8531 case ixgbe_mac_X540:
8532 case ixgbe_mac_X550:
8533 case ixgbe_mac_X550EM_x:
8534 netdev->features |= NETIF_F_SCTP_CSUM;
8535 netdev->hw_features |= NETIF_F_SCTP_CSUM |
8542 netdev->hw_features |= NETIF_F_RXALL;
8543 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
8545 netdev->vlan_features |= NETIF_F_TSO;
8546 netdev->vlan_features |= NETIF_F_TSO6;
8547 netdev->vlan_features |= NETIF_F_IP_CSUM;
8548 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
8549 netdev->vlan_features |= NETIF_F_SG;
8551 netdev->priv_flags |= IFF_UNICAST_FLT;
8552 netdev->priv_flags |= IFF_SUPP_NOFCS;
8554 switch (adapter->hw.mac.type) {
8555 case ixgbe_mac_X550:
8556 case ixgbe_mac_X550EM_x:
8557 netdev->hw_enc_features |= NETIF_F_RXCSUM;
8563 #ifdef CONFIG_IXGBE_DCB
8564 netdev->dcbnl_ops = &dcbnl_ops;
8568 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
8569 unsigned int fcoe_l;
8571 if (hw->mac.ops.get_device_caps) {
8572 hw->mac.ops.get_device_caps(hw, &device_caps);
8573 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
8574 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
8578 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
8579 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
8581 netdev->features |= NETIF_F_FSO |
8584 netdev->vlan_features |= NETIF_F_FSO |
8588 #endif /* IXGBE_FCOE */
8589 if (pci_using_dac) {
8590 netdev->features |= NETIF_F_HIGHDMA;
8591 netdev->vlan_features |= NETIF_F_HIGHDMA;
8594 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
8595 netdev->hw_features |= NETIF_F_LRO;
8596 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
8597 netdev->features |= NETIF_F_LRO;
8599 /* make sure the EEPROM is good */
8600 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
8601 e_dev_err("The EEPROM Checksum Is Not Valid\n");
8606 ixgbe_get_platform_mac_addr(adapter);
8608 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
8610 if (!is_valid_ether_addr(netdev->dev_addr)) {
8611 e_dev_err("invalid MAC address\n");
8616 ixgbe_mac_set_default_filter(adapter, hw->mac.perm_addr);
8618 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
8619 (unsigned long) adapter);
8621 if (ixgbe_removed(hw->hw_addr)) {
8625 INIT_WORK(&adapter->service_task, ixgbe_service_task);
8626 set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
8627 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
8629 err = ixgbe_init_interrupt_scheme(adapter);
8633 /* WOL not supported for all devices */
8635 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
8636 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
8637 pdev->subsystem_device);
8638 if (hw->wol_enabled)
8639 adapter->wol = IXGBE_WUFC_MAG;
8641 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
8643 /* save off EEPROM version number */
8644 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
8645 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
8647 /* pick up the PCI bus settings for reporting later */
8648 hw->mac.ops.get_bus_info(hw);
8649 if (ixgbe_pcie_from_parent(hw))
8650 ixgbe_get_parent_bus_info(adapter);
8652 /* calculate the expected PCIe bandwidth required for optimal
8653 * performance. Note that some older parts will never have enough
8654 * bandwidth due to being older generation PCIe parts. We clamp these
8655 * parts to ensure no warning is displayed if it can't be fixed.
8657 switch (hw->mac.type) {
8658 case ixgbe_mac_82598EB:
8659 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
8662 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
8666 /* don't check link if we failed to enumerate functions */
8667 if (expected_gts > 0)
8668 ixgbe_check_minimum_link(adapter, expected_gts);
8670 err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
8672 strlcpy(part_str, "Unknown", sizeof(part_str));
8673 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
8674 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
8675 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
8678 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
8679 hw->mac.type, hw->phy.type, part_str);
8681 e_dev_info("%pM\n", netdev->dev_addr);
8683 /* reset the hardware with the new settings */
8684 err = hw->mac.ops.start_hw(hw);
8685 if (err == IXGBE_ERR_EEPROM_VERSION) {
8686 /* We are running on a pre-production device, log a warning */
8687 e_dev_warn("This device is a pre-production adapter/LOM. "
8688 "Please be aware there may be issues associated "
8689 "with your hardware. If you are experiencing "
8690 "problems please contact your Intel or hardware "
8691 "representative who provided you with this "
8694 strcpy(netdev->name, "eth%d");
8695 err = register_netdev(netdev);
8699 pci_set_drvdata(pdev, adapter);
8701 /* power down the optics for 82599 SFP+ fiber */
8702 if (hw->mac.ops.disable_tx_laser)
8703 hw->mac.ops.disable_tx_laser(hw);
8705 /* carrier off reporting is important to ethtool even BEFORE open */
8706 netif_carrier_off(netdev);
8708 #ifdef CONFIG_IXGBE_DCA
8709 if (dca_add_requester(&pdev->dev) == 0) {
8710 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
8711 ixgbe_setup_dca(adapter);
8714 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
8715 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
8716 for (i = 0; i < adapter->num_vfs; i++)
8717 ixgbe_vf_configuration(pdev, (i | 0x10000000));
8720 /* firmware requires driver version to be 0xFFFFFFFF
8721 * since os does not support feature
8723 if (hw->mac.ops.set_fw_drv_ver)
8724 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
8727 /* add san mac addr to netdev */
8728 ixgbe_add_sanmac_netdev(netdev);
8730 e_dev_info("%s\n", ixgbe_default_device_descr);
8732 #ifdef CONFIG_IXGBE_HWMON
8733 if (ixgbe_sysfs_init(adapter))
8734 e_err(probe, "failed to allocate sysfs resources\n");
8735 #endif /* CONFIG_IXGBE_HWMON */
8737 ixgbe_dbg_adapter_init(adapter);
8739 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
8740 if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
8741 hw->mac.ops.setup_link(hw,
8742 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
8748 ixgbe_release_hw_control(adapter);
8749 ixgbe_clear_interrupt_scheme(adapter);
8751 ixgbe_disable_sriov(adapter);
8752 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
8753 iounmap(adapter->io_addr);
8754 kfree(adapter->mac_table);
8756 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
8757 free_netdev(netdev);
8759 pci_release_selected_regions(pdev,
8760 pci_select_bars(pdev, IORESOURCE_MEM));
8763 if (!adapter || disable_dev)
8764 pci_disable_device(pdev);
8769 * ixgbe_remove - Device Removal Routine
8770 * @pdev: PCI device information struct
8772 * ixgbe_remove is called by the PCI subsystem to alert the driver
8773 * that it should release a PCI device. The could be caused by a
8774 * Hot-Plug event, or because the driver is going to be removed from
8777 static void ixgbe_remove(struct pci_dev *pdev)
8779 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8780 struct net_device *netdev;
8783 /* if !adapter then we already cleaned up in probe */
8787 netdev = adapter->netdev;
8788 ixgbe_dbg_adapter_exit(adapter);
8790 set_bit(__IXGBE_REMOVING, &adapter->state);
8791 cancel_work_sync(&adapter->service_task);
8794 #ifdef CONFIG_IXGBE_DCA
8795 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
8796 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
8797 dca_remove_requester(&pdev->dev);
8798 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
8802 #ifdef CONFIG_IXGBE_HWMON
8803 ixgbe_sysfs_exit(adapter);
8804 #endif /* CONFIG_IXGBE_HWMON */
8806 /* remove the added san mac */
8807 ixgbe_del_sanmac_netdev(netdev);
8809 if (netdev->reg_state == NETREG_REGISTERED)
8810 unregister_netdev(netdev);
8812 #ifdef CONFIG_PCI_IOV
8814 * Only disable SR-IOV on unload if the user specified the now
8815 * deprecated max_vfs module parameter.
8818 ixgbe_disable_sriov(adapter);
8820 ixgbe_clear_interrupt_scheme(adapter);
8822 ixgbe_release_hw_control(adapter);
8825 kfree(adapter->ixgbe_ieee_pfc);
8826 kfree(adapter->ixgbe_ieee_ets);
8829 iounmap(adapter->io_addr);
8830 pci_release_selected_regions(pdev, pci_select_bars(pdev,
8833 e_dev_info("complete\n");
8835 kfree(adapter->mac_table);
8836 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
8837 free_netdev(netdev);
8839 pci_disable_pcie_error_reporting(pdev);
8842 pci_disable_device(pdev);
8846 * ixgbe_io_error_detected - called when PCI error is detected
8847 * @pdev: Pointer to PCI device
8848 * @state: The current pci connection state
8850 * This function is called after a PCI bus error affecting
8851 * this device has been detected.
8853 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
8854 pci_channel_state_t state)
8856 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8857 struct net_device *netdev = adapter->netdev;
8859 #ifdef CONFIG_PCI_IOV
8860 struct ixgbe_hw *hw = &adapter->hw;
8861 struct pci_dev *bdev, *vfdev;
8862 u32 dw0, dw1, dw2, dw3;
8864 u16 req_id, pf_func;
8866 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
8867 adapter->num_vfs == 0)
8868 goto skip_bad_vf_detection;
8870 bdev = pdev->bus->self;
8871 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
8872 bdev = bdev->bus->self;
8875 goto skip_bad_vf_detection;
8877 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
8879 goto skip_bad_vf_detection;
8881 dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
8882 dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
8883 dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
8884 dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
8885 if (ixgbe_removed(hw->hw_addr))
8886 goto skip_bad_vf_detection;
8889 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
8890 if (!(req_id & 0x0080))
8891 goto skip_bad_vf_detection;
8893 pf_func = req_id & 0x01;
8894 if ((pf_func & 1) == (pdev->devfn & 1)) {
8895 unsigned int device_id;
8897 vf = (req_id & 0x7F) >> 1;
8898 e_dev_err("VF %d has caused a PCIe error\n", vf);
8899 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
8900 "%8.8x\tdw3: %8.8x\n",
8901 dw0, dw1, dw2, dw3);
8902 switch (adapter->hw.mac.type) {
8903 case ixgbe_mac_82599EB:
8904 device_id = IXGBE_82599_VF_DEVICE_ID;
8906 case ixgbe_mac_X540:
8907 device_id = IXGBE_X540_VF_DEVICE_ID;
8909 case ixgbe_mac_X550:
8910 device_id = IXGBE_DEV_ID_X550_VF;
8912 case ixgbe_mac_X550EM_x:
8913 device_id = IXGBE_DEV_ID_X550EM_X_VF;
8920 /* Find the pci device of the offending VF */
8921 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
8923 if (vfdev->devfn == (req_id & 0xFF))
8925 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
8929 * There's a slim chance the VF could have been hot plugged,
8930 * so if it is no longer present we don't need to issue the
8931 * VFLR. Just clean up the AER in that case.
8934 ixgbe_issue_vf_flr(adapter, vfdev);
8935 /* Free device reference count */
8939 pci_cleanup_aer_uncorrect_error_status(pdev);
8943 * Even though the error may have occurred on the other port
8944 * we still need to increment the vf error reference count for
8945 * both ports because the I/O resume function will be called
8948 adapter->vferr_refcount++;
8950 return PCI_ERS_RESULT_RECOVERED;
8952 skip_bad_vf_detection:
8953 #endif /* CONFIG_PCI_IOV */
8954 if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
8955 return PCI_ERS_RESULT_DISCONNECT;
8958 netif_device_detach(netdev);
8960 if (state == pci_channel_io_perm_failure) {
8962 return PCI_ERS_RESULT_DISCONNECT;
8965 if (netif_running(netdev))
8966 ixgbe_down(adapter);
8968 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
8969 pci_disable_device(pdev);
8972 /* Request a slot reset. */
8973 return PCI_ERS_RESULT_NEED_RESET;
8977 * ixgbe_io_slot_reset - called after the pci bus has been reset.
8978 * @pdev: Pointer to PCI device
8980 * Restart the card from scratch, as if from a cold-boot.
8982 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
8984 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8985 pci_ers_result_t result;
8988 if (pci_enable_device_mem(pdev)) {
8989 e_err(probe, "Cannot re-enable PCI device after reset.\n");
8990 result = PCI_ERS_RESULT_DISCONNECT;
8992 smp_mb__before_atomic();
8993 clear_bit(__IXGBE_DISABLED, &adapter->state);
8994 adapter->hw.hw_addr = adapter->io_addr;
8995 pci_set_master(pdev);
8996 pci_restore_state(pdev);
8997 pci_save_state(pdev);
8999 pci_wake_from_d3(pdev, false);
9001 ixgbe_reset(adapter);
9002 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
9003 result = PCI_ERS_RESULT_RECOVERED;
9006 err = pci_cleanup_aer_uncorrect_error_status(pdev);
9008 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
9009 "failed 0x%0x\n", err);
9010 /* non-fatal, continue */
9017 * ixgbe_io_resume - called when traffic can start flowing again.
9018 * @pdev: Pointer to PCI device
9020 * This callback is called when the error recovery driver tells us that
9021 * its OK to resume normal operation.
9023 static void ixgbe_io_resume(struct pci_dev *pdev)
9025 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9026 struct net_device *netdev = adapter->netdev;
9028 #ifdef CONFIG_PCI_IOV
9029 if (adapter->vferr_refcount) {
9030 e_info(drv, "Resuming after VF err\n");
9031 adapter->vferr_refcount--;
9036 if (netif_running(netdev))
9039 netif_device_attach(netdev);
9042 static const struct pci_error_handlers ixgbe_err_handler = {
9043 .error_detected = ixgbe_io_error_detected,
9044 .slot_reset = ixgbe_io_slot_reset,
9045 .resume = ixgbe_io_resume,
9048 static struct pci_driver ixgbe_driver = {
9049 .name = ixgbe_driver_name,
9050 .id_table = ixgbe_pci_tbl,
9051 .probe = ixgbe_probe,
9052 .remove = ixgbe_remove,
9054 .suspend = ixgbe_suspend,
9055 .resume = ixgbe_resume,
9057 .shutdown = ixgbe_shutdown,
9058 .sriov_configure = ixgbe_pci_sriov_configure,
9059 .err_handler = &ixgbe_err_handler
9063 * ixgbe_init_module - Driver Registration Routine
9065 * ixgbe_init_module is the first routine called when the driver is
9066 * loaded. All it does is register with the PCI subsystem.
9068 static int __init ixgbe_init_module(void)
9071 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
9072 pr_info("%s\n", ixgbe_copyright);
9076 ret = pci_register_driver(&ixgbe_driver);
9082 #ifdef CONFIG_IXGBE_DCA
9083 dca_register_notify(&dca_notifier);
9089 module_init(ixgbe_init_module);
9092 * ixgbe_exit_module - Driver Exit Cleanup Routine
9094 * ixgbe_exit_module is called just before the driver is removed
9097 static void __exit ixgbe_exit_module(void)
9099 #ifdef CONFIG_IXGBE_DCA
9100 dca_unregister_notify(&dca_notifier);
9102 pci_unregister_driver(&ixgbe_driver);
9107 #ifdef CONFIG_IXGBE_DCA
9108 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
9113 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
9114 __ixgbe_notify_dca);
9116 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
9119 #endif /* CONFIG_IXGBE_DCA */
9121 module_exit(ixgbe_exit_module);