1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2015 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"
68 #ifdef CONFIG_IXGBE_VXLAN
69 #include <net/vxlan.h>
72 char ixgbe_driver_name[] = "ixgbe";
73 static const char ixgbe_driver_string[] =
74 "Intel(R) 10 Gigabit PCI Express Network Driver";
76 char ixgbe_default_device_descr[] =
77 "Intel(R) 10 Gigabit Network Connection";
79 static char ixgbe_default_device_descr[] =
80 "Intel(R) 10 Gigabit Network Connection";
82 #define DRV_VERSION "4.0.1-k"
83 const char ixgbe_driver_version[] = DRV_VERSION;
84 static const char ixgbe_copyright[] =
85 "Copyright (c) 1999-2015 Intel Corporation.";
87 static const char ixgbe_overheat_msg[] = "Network adapter has been stopped because it has over heated. Restart the computer. If the problem persists, power off the system and replace the adapter";
89 static const struct ixgbe_info *ixgbe_info_tbl[] = {
90 [board_82598] = &ixgbe_82598_info,
91 [board_82599] = &ixgbe_82599_info,
92 [board_X540] = &ixgbe_X540_info,
93 [board_X550] = &ixgbe_X550_info,
94 [board_X550EM_x] = &ixgbe_X550EM_x_info,
97 /* ixgbe_pci_tbl - PCI Device ID Table
99 * Wildcard entries (PCI_ANY_ID) should come last
100 * Last entry must be all 0s
102 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
103 * Class, Class Mask, private data (not used) }
105 static const struct pci_device_id ixgbe_pci_tbl[] = {
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
117 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
118 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
119 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
120 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
121 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
122 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
123 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
124 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
125 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
126 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
127 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
128 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
129 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
130 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
131 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
132 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
133 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
134 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
135 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
136 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
137 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
138 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
139 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
140 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
141 /* required last entry */
144 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
146 #ifdef CONFIG_IXGBE_DCA
147 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
149 static struct notifier_block dca_notifier = {
150 .notifier_call = ixgbe_notify_dca,
156 #ifdef CONFIG_PCI_IOV
157 static unsigned int max_vfs;
158 module_param(max_vfs, uint, 0);
159 MODULE_PARM_DESC(max_vfs,
160 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
161 #endif /* CONFIG_PCI_IOV */
163 static unsigned int allow_unsupported_sfp;
164 module_param(allow_unsupported_sfp, uint, 0);
165 MODULE_PARM_DESC(allow_unsupported_sfp,
166 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
168 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
169 static int debug = -1;
170 module_param(debug, int, 0);
171 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
173 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
174 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
175 MODULE_LICENSE("GPL");
176 MODULE_VERSION(DRV_VERSION);
178 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
180 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
183 struct pci_dev *parent_dev;
184 struct pci_bus *parent_bus;
186 parent_bus = adapter->pdev->bus->parent;
190 parent_dev = parent_bus->self;
194 if (!pci_is_pcie(parent_dev))
197 pcie_capability_read_word(parent_dev, reg, value);
198 if (*value == IXGBE_FAILED_READ_CFG_WORD &&
199 ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
204 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
206 struct ixgbe_hw *hw = &adapter->hw;
210 hw->bus.type = ixgbe_bus_type_pci_express;
212 /* Get the negotiated link width and speed from PCI config space of the
213 * parent, as this device is behind a switch
215 err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
217 /* assume caller will handle error case */
221 hw->bus.width = ixgbe_convert_bus_width(link_status);
222 hw->bus.speed = ixgbe_convert_bus_speed(link_status);
228 * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
229 * @hw: hw specific details
231 * This function is used by probe to determine whether a device's PCI-Express
232 * bandwidth details should be gathered from the parent bus instead of from the
233 * device. Used to ensure that various locations all have the correct device ID
236 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
238 switch (hw->device_id) {
239 case IXGBE_DEV_ID_82599_SFP_SF_QP:
240 case IXGBE_DEV_ID_82599_QSFP_SF_QP:
247 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
250 struct ixgbe_hw *hw = &adapter->hw;
252 enum pci_bus_speed speed = PCI_SPEED_UNKNOWN;
253 enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN;
254 struct pci_dev *pdev;
256 /* Some devices are not connected over PCIe and thus do not negotiate
257 * speed. These devices do not have valid bus info, and thus any report
258 * we generate may not be correct.
260 if (hw->bus.type == ixgbe_bus_type_internal)
263 /* determine whether to use the parent device */
264 if (ixgbe_pcie_from_parent(&adapter->hw))
265 pdev = adapter->pdev->bus->parent->self;
267 pdev = adapter->pdev;
269 if (pcie_get_minimum_link(pdev, &speed, &width) ||
270 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
271 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
276 case PCIE_SPEED_2_5GT:
277 /* 8b/10b encoding reduces max throughput by 20% */
280 case PCIE_SPEED_5_0GT:
281 /* 8b/10b encoding reduces max throughput by 20% */
284 case PCIE_SPEED_8_0GT:
285 /* 128b/130b encoding reduces throughput by less than 2% */
289 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
293 e_dev_info("PCI Express bandwidth of %dGT/s available\n",
295 e_dev_info("(Speed:%s, Width: x%d, Encoding Loss:%s)\n",
296 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" :
297 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" :
298 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" :
301 (speed == PCIE_SPEED_2_5GT ? "20%" :
302 speed == PCIE_SPEED_5_0GT ? "20%" :
303 speed == PCIE_SPEED_8_0GT ? "<2%" :
306 if (max_gts < expected_gts) {
307 e_dev_warn("This is not sufficient for optimal performance of this card.\n");
308 e_dev_warn("For optimal performance, at least %dGT/s of bandwidth is required.\n",
310 e_dev_warn("A slot with more lanes and/or higher speed is suggested.\n");
314 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
316 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
317 !test_bit(__IXGBE_REMOVING, &adapter->state) &&
318 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
319 schedule_work(&adapter->service_task);
322 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
324 struct ixgbe_adapter *adapter = hw->back;
329 e_dev_err("Adapter removed\n");
330 if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
331 ixgbe_service_event_schedule(adapter);
334 static void ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
338 /* The following check not only optimizes a bit by not
339 * performing a read on the status register when the
340 * register just read was a status register read that
341 * returned IXGBE_FAILED_READ_REG. It also blocks any
342 * potential recursion.
344 if (reg == IXGBE_STATUS) {
345 ixgbe_remove_adapter(hw);
348 value = ixgbe_read_reg(hw, IXGBE_STATUS);
349 if (value == IXGBE_FAILED_READ_REG)
350 ixgbe_remove_adapter(hw);
354 * ixgbe_read_reg - Read from device register
355 * @hw: hw specific details
356 * @reg: offset of register to read
358 * Returns : value read or IXGBE_FAILED_READ_REG if removed
360 * This function is used to read device registers. It checks for device
361 * removal by confirming any read that returns all ones by checking the
362 * status register value for all ones. This function avoids reading from
363 * the hardware if a removal was previously detected in which case it
364 * returns IXGBE_FAILED_READ_REG (all ones).
366 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
368 u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
371 if (ixgbe_removed(reg_addr))
372 return IXGBE_FAILED_READ_REG;
373 value = readl(reg_addr + reg);
374 if (unlikely(value == IXGBE_FAILED_READ_REG))
375 ixgbe_check_remove(hw, reg);
379 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
383 pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
384 if (value == IXGBE_FAILED_READ_CFG_WORD) {
385 ixgbe_remove_adapter(hw);
391 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
393 struct ixgbe_adapter *adapter = hw->back;
396 if (ixgbe_removed(hw->hw_addr))
397 return IXGBE_FAILED_READ_CFG_WORD;
398 pci_read_config_word(adapter->pdev, reg, &value);
399 if (value == IXGBE_FAILED_READ_CFG_WORD &&
400 ixgbe_check_cfg_remove(hw, adapter->pdev))
401 return IXGBE_FAILED_READ_CFG_WORD;
405 #ifdef CONFIG_PCI_IOV
406 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
408 struct ixgbe_adapter *adapter = hw->back;
411 if (ixgbe_removed(hw->hw_addr))
412 return IXGBE_FAILED_READ_CFG_DWORD;
413 pci_read_config_dword(adapter->pdev, reg, &value);
414 if (value == IXGBE_FAILED_READ_CFG_DWORD &&
415 ixgbe_check_cfg_remove(hw, adapter->pdev))
416 return IXGBE_FAILED_READ_CFG_DWORD;
419 #endif /* CONFIG_PCI_IOV */
421 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
423 struct ixgbe_adapter *adapter = hw->back;
425 if (ixgbe_removed(hw->hw_addr))
427 pci_write_config_word(adapter->pdev, reg, value);
430 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
432 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
434 /* flush memory to make sure state is correct before next watchdog */
435 smp_mb__before_atomic();
436 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
439 struct ixgbe_reg_info {
444 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
446 /* General Registers */
447 {IXGBE_CTRL, "CTRL"},
448 {IXGBE_STATUS, "STATUS"},
449 {IXGBE_CTRL_EXT, "CTRL_EXT"},
451 /* Interrupt Registers */
452 {IXGBE_EICR, "EICR"},
455 {IXGBE_SRRCTL(0), "SRRCTL"},
456 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
457 {IXGBE_RDLEN(0), "RDLEN"},
458 {IXGBE_RDH(0), "RDH"},
459 {IXGBE_RDT(0), "RDT"},
460 {IXGBE_RXDCTL(0), "RXDCTL"},
461 {IXGBE_RDBAL(0), "RDBAL"},
462 {IXGBE_RDBAH(0), "RDBAH"},
465 {IXGBE_TDBAL(0), "TDBAL"},
466 {IXGBE_TDBAH(0), "TDBAH"},
467 {IXGBE_TDLEN(0), "TDLEN"},
468 {IXGBE_TDH(0), "TDH"},
469 {IXGBE_TDT(0), "TDT"},
470 {IXGBE_TXDCTL(0), "TXDCTL"},
472 /* List Terminator */
478 * ixgbe_regdump - register printout routine
480 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
486 switch (reginfo->ofs) {
487 case IXGBE_SRRCTL(0):
488 for (i = 0; i < 64; i++)
489 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
491 case IXGBE_DCA_RXCTRL(0):
492 for (i = 0; i < 64; i++)
493 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
496 for (i = 0; i < 64; i++)
497 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
500 for (i = 0; i < 64; i++)
501 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
504 for (i = 0; i < 64; i++)
505 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
507 case IXGBE_RXDCTL(0):
508 for (i = 0; i < 64; i++)
509 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
512 for (i = 0; i < 64; i++)
513 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
516 for (i = 0; i < 64; i++)
517 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
520 for (i = 0; i < 64; i++)
521 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
524 for (i = 0; i < 64; i++)
525 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
528 for (i = 0; i < 64; i++)
529 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
532 for (i = 0; i < 64; i++)
533 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
536 for (i = 0; i < 64; i++)
537 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
539 case IXGBE_TXDCTL(0):
540 for (i = 0; i < 64; i++)
541 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
544 pr_info("%-15s %08x\n", reginfo->name,
545 IXGBE_READ_REG(hw, reginfo->ofs));
549 for (i = 0; i < 8; i++) {
550 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
551 pr_err("%-15s", rname);
552 for (j = 0; j < 8; j++)
553 pr_cont(" %08x", regs[i*8+j]);
560 * ixgbe_dump - Print registers, tx-rings and rx-rings
562 static void ixgbe_dump(struct ixgbe_adapter *adapter)
564 struct net_device *netdev = adapter->netdev;
565 struct ixgbe_hw *hw = &adapter->hw;
566 struct ixgbe_reg_info *reginfo;
568 struct ixgbe_ring *tx_ring;
569 struct ixgbe_tx_buffer *tx_buffer;
570 union ixgbe_adv_tx_desc *tx_desc;
571 struct my_u0 { u64 a; u64 b; } *u0;
572 struct ixgbe_ring *rx_ring;
573 union ixgbe_adv_rx_desc *rx_desc;
574 struct ixgbe_rx_buffer *rx_buffer_info;
578 if (!netif_msg_hw(adapter))
581 /* Print netdevice Info */
583 dev_info(&adapter->pdev->dev, "Net device Info\n");
584 pr_info("Device Name state "
585 "trans_start last_rx\n");
586 pr_info("%-15s %016lX %016lX %016lX\n",
593 /* Print Registers */
594 dev_info(&adapter->pdev->dev, "Register Dump\n");
595 pr_info(" Register Name Value\n");
596 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
597 reginfo->name; reginfo++) {
598 ixgbe_regdump(hw, reginfo);
601 /* Print TX Ring Summary */
602 if (!netdev || !netif_running(netdev))
605 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
606 pr_info(" %s %s %s %s\n",
607 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
608 "leng", "ntw", "timestamp");
609 for (n = 0; n < adapter->num_tx_queues; n++) {
610 tx_ring = adapter->tx_ring[n];
611 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
612 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
613 n, tx_ring->next_to_use, tx_ring->next_to_clean,
614 (u64)dma_unmap_addr(tx_buffer, dma),
615 dma_unmap_len(tx_buffer, len),
616 tx_buffer->next_to_watch,
617 (u64)tx_buffer->time_stamp);
621 if (!netif_msg_tx_done(adapter))
622 goto rx_ring_summary;
624 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
626 /* Transmit Descriptor Formats
628 * 82598 Advanced Transmit Descriptor
629 * +--------------------------------------------------------------+
630 * 0 | Buffer Address [63:0] |
631 * +--------------------------------------------------------------+
632 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
633 * +--------------------------------------------------------------+
634 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
636 * 82598 Advanced Transmit Descriptor (Write-Back Format)
637 * +--------------------------------------------------------------+
639 * +--------------------------------------------------------------+
640 * 8 | RSV | STA | NXTSEQ |
641 * +--------------------------------------------------------------+
644 * 82599+ Advanced Transmit Descriptor
645 * +--------------------------------------------------------------+
646 * 0 | Buffer Address [63:0] |
647 * +--------------------------------------------------------------+
648 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
649 * +--------------------------------------------------------------+
650 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
652 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
653 * +--------------------------------------------------------------+
655 * +--------------------------------------------------------------+
656 * 8 | RSV | STA | RSV |
657 * +--------------------------------------------------------------+
661 for (n = 0; n < adapter->num_tx_queues; n++) {
662 tx_ring = adapter->tx_ring[n];
663 pr_info("------------------------------------\n");
664 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
665 pr_info("------------------------------------\n");
666 pr_info("%s%s %s %s %s %s\n",
667 "T [desc] [address 63:0 ] ",
668 "[PlPOIdStDDt Ln] [bi->dma ] ",
669 "leng", "ntw", "timestamp", "bi->skb");
671 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
672 tx_desc = IXGBE_TX_DESC(tx_ring, i);
673 tx_buffer = &tx_ring->tx_buffer_info[i];
674 u0 = (struct my_u0 *)tx_desc;
675 if (dma_unmap_len(tx_buffer, len) > 0) {
676 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p",
680 (u64)dma_unmap_addr(tx_buffer, dma),
681 dma_unmap_len(tx_buffer, len),
682 tx_buffer->next_to_watch,
683 (u64)tx_buffer->time_stamp,
685 if (i == tx_ring->next_to_use &&
686 i == tx_ring->next_to_clean)
688 else if (i == tx_ring->next_to_use)
690 else if (i == tx_ring->next_to_clean)
695 if (netif_msg_pktdata(adapter) &&
697 print_hex_dump(KERN_INFO, "",
698 DUMP_PREFIX_ADDRESS, 16, 1,
699 tx_buffer->skb->data,
700 dma_unmap_len(tx_buffer, len),
706 /* Print RX Rings Summary */
708 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
709 pr_info("Queue [NTU] [NTC]\n");
710 for (n = 0; n < adapter->num_rx_queues; n++) {
711 rx_ring = adapter->rx_ring[n];
712 pr_info("%5d %5X %5X\n",
713 n, rx_ring->next_to_use, rx_ring->next_to_clean);
717 if (!netif_msg_rx_status(adapter))
720 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
722 /* Receive Descriptor Formats
724 * 82598 Advanced Receive Descriptor (Read) Format
726 * +-----------------------------------------------------+
727 * 0 | Packet Buffer Address [63:1] |A0/NSE|
728 * +----------------------------------------------+------+
729 * 8 | Header Buffer Address [63:1] | DD |
730 * +-----------------------------------------------------+
733 * 82598 Advanced Receive Descriptor (Write-Back) Format
735 * 63 48 47 32 31 30 21 20 16 15 4 3 0
736 * +------------------------------------------------------+
737 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
738 * | Packet | IP | | | | Type | Type |
739 * | Checksum | Ident | | | | | |
740 * +------------------------------------------------------+
741 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
742 * +------------------------------------------------------+
743 * 63 48 47 32 31 20 19 0
745 * 82599+ Advanced Receive Descriptor (Read) Format
747 * +-----------------------------------------------------+
748 * 0 | Packet Buffer Address [63:1] |A0/NSE|
749 * +----------------------------------------------+------+
750 * 8 | Header Buffer Address [63:1] | DD |
751 * +-----------------------------------------------------+
754 * 82599+ Advanced Receive Descriptor (Write-Back) Format
756 * 63 48 47 32 31 30 21 20 17 16 4 3 0
757 * +------------------------------------------------------+
758 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
759 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
760 * |/ Flow Dir Flt ID | | | | | |
761 * +------------------------------------------------------+
762 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
763 * +------------------------------------------------------+
764 * 63 48 47 32 31 20 19 0
767 for (n = 0; n < adapter->num_rx_queues; n++) {
768 rx_ring = adapter->rx_ring[n];
769 pr_info("------------------------------------\n");
770 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
771 pr_info("------------------------------------\n");
773 "R [desc] [ PktBuf A0] ",
774 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
775 "<-- Adv Rx Read format\n");
777 "RWB[desc] [PcsmIpSHl PtRs] ",
778 "[vl er S cks ln] ---------------- [bi->skb ] ",
779 "<-- Adv Rx Write-Back format\n");
781 for (i = 0; i < rx_ring->count; i++) {
782 rx_buffer_info = &rx_ring->rx_buffer_info[i];
783 rx_desc = IXGBE_RX_DESC(rx_ring, i);
784 u0 = (struct my_u0 *)rx_desc;
785 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
786 if (staterr & IXGBE_RXD_STAT_DD) {
787 /* Descriptor Done */
788 pr_info("RWB[0x%03X] %016llX "
789 "%016llX ---------------- %p", i,
792 rx_buffer_info->skb);
794 pr_info("R [0x%03X] %016llX "
795 "%016llX %016llX %p", i,
798 (u64)rx_buffer_info->dma,
799 rx_buffer_info->skb);
801 if (netif_msg_pktdata(adapter) &&
802 rx_buffer_info->dma) {
803 print_hex_dump(KERN_INFO, "",
804 DUMP_PREFIX_ADDRESS, 16, 1,
805 page_address(rx_buffer_info->page) +
806 rx_buffer_info->page_offset,
807 ixgbe_rx_bufsz(rx_ring), true);
811 if (i == rx_ring->next_to_use)
813 else if (i == rx_ring->next_to_clean)
822 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
826 /* Let firmware take over control of h/w */
827 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
828 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
829 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
832 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
836 /* Let firmware know the driver has taken over */
837 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
838 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
839 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
843 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
844 * @adapter: pointer to adapter struct
845 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
846 * @queue: queue to map the corresponding interrupt to
847 * @msix_vector: the vector to map to the corresponding queue
850 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
851 u8 queue, u8 msix_vector)
854 struct ixgbe_hw *hw = &adapter->hw;
855 switch (hw->mac.type) {
856 case ixgbe_mac_82598EB:
857 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
860 index = (((direction * 64) + queue) >> 2) & 0x1F;
861 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
862 ivar &= ~(0xFF << (8 * (queue & 0x3)));
863 ivar |= (msix_vector << (8 * (queue & 0x3)));
864 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
866 case ixgbe_mac_82599EB:
869 case ixgbe_mac_X550EM_x:
870 if (direction == -1) {
872 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
873 index = ((queue & 1) * 8);
874 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
875 ivar &= ~(0xFF << index);
876 ivar |= (msix_vector << index);
877 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
880 /* tx or rx causes */
881 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
882 index = ((16 * (queue & 1)) + (8 * direction));
883 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
884 ivar &= ~(0xFF << index);
885 ivar |= (msix_vector << index);
886 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
894 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
899 switch (adapter->hw.mac.type) {
900 case ixgbe_mac_82598EB:
901 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
902 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
904 case ixgbe_mac_82599EB:
907 case ixgbe_mac_X550EM_x:
908 mask = (qmask & 0xFFFFFFFF);
909 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
910 mask = (qmask >> 32);
911 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
918 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
919 struct ixgbe_tx_buffer *tx_buffer)
921 if (tx_buffer->skb) {
922 dev_kfree_skb_any(tx_buffer->skb);
923 if (dma_unmap_len(tx_buffer, len))
924 dma_unmap_single(ring->dev,
925 dma_unmap_addr(tx_buffer, dma),
926 dma_unmap_len(tx_buffer, len),
928 } else if (dma_unmap_len(tx_buffer, len)) {
929 dma_unmap_page(ring->dev,
930 dma_unmap_addr(tx_buffer, dma),
931 dma_unmap_len(tx_buffer, len),
934 tx_buffer->next_to_watch = NULL;
935 tx_buffer->skb = NULL;
936 dma_unmap_len_set(tx_buffer, len, 0);
937 /* tx_buffer must be completely set up in the transmit path */
940 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
942 struct ixgbe_hw *hw = &adapter->hw;
943 struct ixgbe_hw_stats *hwstats = &adapter->stats;
947 if ((hw->fc.current_mode != ixgbe_fc_full) &&
948 (hw->fc.current_mode != ixgbe_fc_rx_pause))
951 switch (hw->mac.type) {
952 case ixgbe_mac_82598EB:
953 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
956 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
958 hwstats->lxoffrxc += data;
960 /* refill credits (no tx hang) if we received xoff */
964 for (i = 0; i < adapter->num_tx_queues; i++)
965 clear_bit(__IXGBE_HANG_CHECK_ARMED,
966 &adapter->tx_ring[i]->state);
969 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
971 struct ixgbe_hw *hw = &adapter->hw;
972 struct ixgbe_hw_stats *hwstats = &adapter->stats;
976 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
978 if (adapter->ixgbe_ieee_pfc)
979 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
981 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
982 ixgbe_update_xoff_rx_lfc(adapter);
986 /* update stats for each tc, only valid with PFC enabled */
987 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
990 switch (hw->mac.type) {
991 case ixgbe_mac_82598EB:
992 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
995 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
997 hwstats->pxoffrxc[i] += pxoffrxc;
998 /* Get the TC for given UP */
999 tc = netdev_get_prio_tc_map(adapter->netdev, i);
1000 xoff[tc] += pxoffrxc;
1003 /* disarm tx queues that have received xoff frames */
1004 for (i = 0; i < adapter->num_tx_queues; i++) {
1005 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
1007 tc = tx_ring->dcb_tc;
1009 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1013 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1015 return ring->stats.packets;
1018 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1020 struct ixgbe_adapter *adapter;
1021 struct ixgbe_hw *hw;
1024 if (ring->l2_accel_priv)
1025 adapter = ring->l2_accel_priv->real_adapter;
1027 adapter = netdev_priv(ring->netdev);
1030 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
1031 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
1034 return (head < tail) ?
1035 tail - head : (tail + ring->count - head);
1040 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1042 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1043 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1044 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1046 clear_check_for_tx_hang(tx_ring);
1049 * Check for a hung queue, but be thorough. This verifies
1050 * that a transmit has been completed since the previous
1051 * check AND there is at least one packet pending. The
1052 * ARMED bit is set to indicate a potential hang. The
1053 * bit is cleared if a pause frame is received to remove
1054 * false hang detection due to PFC or 802.3x frames. By
1055 * requiring this to fail twice we avoid races with
1056 * pfc clearing the ARMED bit and conditions where we
1057 * run the check_tx_hang logic with a transmit completion
1058 * pending but without time to complete it yet.
1060 if (tx_done_old == tx_done && tx_pending)
1061 /* make sure it is true for two checks in a row */
1062 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1064 /* update completed stats and continue */
1065 tx_ring->tx_stats.tx_done_old = tx_done;
1066 /* reset the countdown */
1067 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1073 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1074 * @adapter: driver private struct
1076 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1079 /* Do the reset outside of interrupt context */
1080 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1081 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
1082 e_warn(drv, "initiating reset due to tx timeout\n");
1083 ixgbe_service_event_schedule(adapter);
1088 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1089 * @q_vector: structure containing interrupt and ring information
1090 * @tx_ring: tx ring to clean
1092 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1093 struct ixgbe_ring *tx_ring)
1095 struct ixgbe_adapter *adapter = q_vector->adapter;
1096 struct ixgbe_tx_buffer *tx_buffer;
1097 union ixgbe_adv_tx_desc *tx_desc;
1098 unsigned int total_bytes = 0, total_packets = 0;
1099 unsigned int budget = q_vector->tx.work_limit;
1100 unsigned int i = tx_ring->next_to_clean;
1102 if (test_bit(__IXGBE_DOWN, &adapter->state))
1105 tx_buffer = &tx_ring->tx_buffer_info[i];
1106 tx_desc = IXGBE_TX_DESC(tx_ring, i);
1107 i -= tx_ring->count;
1110 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1112 /* if next_to_watch is not set then there is no work pending */
1116 /* prevent any other reads prior to eop_desc */
1117 read_barrier_depends();
1119 /* if DD is not set pending work has not been completed */
1120 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1123 /* clear next_to_watch to prevent false hangs */
1124 tx_buffer->next_to_watch = NULL;
1126 /* update the statistics for this packet */
1127 total_bytes += tx_buffer->bytecount;
1128 total_packets += tx_buffer->gso_segs;
1131 dev_consume_skb_any(tx_buffer->skb);
1133 /* unmap skb header data */
1134 dma_unmap_single(tx_ring->dev,
1135 dma_unmap_addr(tx_buffer, dma),
1136 dma_unmap_len(tx_buffer, len),
1139 /* clear tx_buffer data */
1140 tx_buffer->skb = NULL;
1141 dma_unmap_len_set(tx_buffer, len, 0);
1143 /* unmap remaining buffers */
1144 while (tx_desc != eop_desc) {
1149 i -= tx_ring->count;
1150 tx_buffer = tx_ring->tx_buffer_info;
1151 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1154 /* unmap any remaining paged data */
1155 if (dma_unmap_len(tx_buffer, len)) {
1156 dma_unmap_page(tx_ring->dev,
1157 dma_unmap_addr(tx_buffer, dma),
1158 dma_unmap_len(tx_buffer, len),
1160 dma_unmap_len_set(tx_buffer, len, 0);
1164 /* move us one more past the eop_desc for start of next pkt */
1169 i -= tx_ring->count;
1170 tx_buffer = tx_ring->tx_buffer_info;
1171 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1174 /* issue prefetch for next Tx descriptor */
1177 /* update budget accounting */
1179 } while (likely(budget));
1181 i += tx_ring->count;
1182 tx_ring->next_to_clean = i;
1183 u64_stats_update_begin(&tx_ring->syncp);
1184 tx_ring->stats.bytes += total_bytes;
1185 tx_ring->stats.packets += total_packets;
1186 u64_stats_update_end(&tx_ring->syncp);
1187 q_vector->tx.total_bytes += total_bytes;
1188 q_vector->tx.total_packets += total_packets;
1190 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1191 /* schedule immediate reset if we believe we hung */
1192 struct ixgbe_hw *hw = &adapter->hw;
1193 e_err(drv, "Detected Tx Unit Hang\n"
1195 " TDH, TDT <%x>, <%x>\n"
1196 " next_to_use <%x>\n"
1197 " next_to_clean <%x>\n"
1198 "tx_buffer_info[next_to_clean]\n"
1199 " time_stamp <%lx>\n"
1201 tx_ring->queue_index,
1202 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1203 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1204 tx_ring->next_to_use, i,
1205 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1207 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1210 "tx hang %d detected on queue %d, resetting adapter\n",
1211 adapter->tx_timeout_count + 1, tx_ring->queue_index);
1213 /* schedule immediate reset if we believe we hung */
1214 ixgbe_tx_timeout_reset(adapter);
1216 /* the adapter is about to reset, no point in enabling stuff */
1220 netdev_tx_completed_queue(txring_txq(tx_ring),
1221 total_packets, total_bytes);
1223 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1224 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1225 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1226 /* Make sure that anybody stopping the queue after this
1227 * sees the new next_to_clean.
1230 if (__netif_subqueue_stopped(tx_ring->netdev,
1231 tx_ring->queue_index)
1232 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1233 netif_wake_subqueue(tx_ring->netdev,
1234 tx_ring->queue_index);
1235 ++tx_ring->tx_stats.restart_queue;
1242 #ifdef CONFIG_IXGBE_DCA
1243 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1244 struct ixgbe_ring *tx_ring,
1247 struct ixgbe_hw *hw = &adapter->hw;
1248 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1251 switch (hw->mac.type) {
1252 case ixgbe_mac_82598EB:
1253 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1255 case ixgbe_mac_82599EB:
1256 case ixgbe_mac_X540:
1257 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1258 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1261 /* for unknown hardware do not write register */
1266 * We can enable relaxed ordering for reads, but not writes when
1267 * DCA is enabled. This is due to a known issue in some chipsets
1268 * which will cause the DCA tag to be cleared.
1270 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1271 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1272 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1274 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1277 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1278 struct ixgbe_ring *rx_ring,
1281 struct ixgbe_hw *hw = &adapter->hw;
1282 u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1283 u8 reg_idx = rx_ring->reg_idx;
1286 switch (hw->mac.type) {
1287 case ixgbe_mac_82599EB:
1288 case ixgbe_mac_X540:
1289 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1296 * We can enable relaxed ordering for reads, but not writes when
1297 * DCA is enabled. This is due to a known issue in some chipsets
1298 * which will cause the DCA tag to be cleared.
1300 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1301 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1303 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1306 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1308 struct ixgbe_adapter *adapter = q_vector->adapter;
1309 struct ixgbe_ring *ring;
1310 int cpu = get_cpu();
1312 if (q_vector->cpu == cpu)
1315 ixgbe_for_each_ring(ring, q_vector->tx)
1316 ixgbe_update_tx_dca(adapter, ring, cpu);
1318 ixgbe_for_each_ring(ring, q_vector->rx)
1319 ixgbe_update_rx_dca(adapter, ring, cpu);
1321 q_vector->cpu = cpu;
1326 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1330 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1333 /* always use CB2 mode, difference is masked in the CB driver */
1334 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1336 for (i = 0; i < adapter->num_q_vectors; i++) {
1337 adapter->q_vector[i]->cpu = -1;
1338 ixgbe_update_dca(adapter->q_vector[i]);
1342 static int __ixgbe_notify_dca(struct device *dev, void *data)
1344 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1345 unsigned long event = *(unsigned long *)data;
1347 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1351 case DCA_PROVIDER_ADD:
1352 /* if we're already enabled, don't do it again */
1353 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1355 if (dca_add_requester(dev) == 0) {
1356 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1357 ixgbe_setup_dca(adapter);
1360 /* Fall Through since DCA is disabled. */
1361 case DCA_PROVIDER_REMOVE:
1362 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1363 dca_remove_requester(dev);
1364 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1365 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1373 #endif /* CONFIG_IXGBE_DCA */
1375 #define IXGBE_RSS_L4_TYPES_MASK \
1376 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1377 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1378 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1379 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1381 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1382 union ixgbe_adv_rx_desc *rx_desc,
1383 struct sk_buff *skb)
1387 if (!(ring->netdev->features & NETIF_F_RXHASH))
1390 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1391 IXGBE_RXDADV_RSSTYPE_MASK;
1396 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1397 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1398 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1403 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1404 * @ring: structure containing ring specific data
1405 * @rx_desc: advanced rx descriptor
1407 * Returns : true if it is FCoE pkt
1409 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1410 union ixgbe_adv_rx_desc *rx_desc)
1412 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1414 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1415 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1416 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1417 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1420 #endif /* IXGBE_FCOE */
1422 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1423 * @ring: structure containing ring specific data
1424 * @rx_desc: current Rx descriptor being processed
1425 * @skb: skb currently being received and modified
1427 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1428 union ixgbe_adv_rx_desc *rx_desc,
1429 struct sk_buff *skb)
1431 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1432 __le16 hdr_info = rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1433 bool encap_pkt = false;
1435 skb_checksum_none_assert(skb);
1437 /* Rx csum disabled */
1438 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1441 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) &&
1442 (hdr_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_TUNNEL >> 16))) {
1444 skb->encapsulation = 1;
1447 /* if IP and error */
1448 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1449 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1450 ring->rx_stats.csum_err++;
1454 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1457 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1459 * 82599 errata, UDP frames with a 0 checksum can be marked as
1462 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1463 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1466 ring->rx_stats.csum_err++;
1470 /* It must be a TCP or UDP packet with a valid checksum */
1471 skb->ip_summed = CHECKSUM_UNNECESSARY;
1473 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1476 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1477 ring->rx_stats.csum_err++;
1480 /* If we checked the outer header let the stack know */
1481 skb->csum_level = 1;
1485 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1486 struct ixgbe_rx_buffer *bi)
1488 struct page *page = bi->page;
1491 /* since we are recycling buffers we should seldom need to alloc */
1495 /* alloc new page for storage */
1496 page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1497 if (unlikely(!page)) {
1498 rx_ring->rx_stats.alloc_rx_page_failed++;
1502 /* map page for use */
1503 dma = dma_map_page(rx_ring->dev, page, 0,
1504 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1507 * if mapping failed free memory back to system since
1508 * there isn't much point in holding memory we can't use
1510 if (dma_mapping_error(rx_ring->dev, dma)) {
1511 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1513 rx_ring->rx_stats.alloc_rx_page_failed++;
1519 bi->page_offset = 0;
1525 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1526 * @rx_ring: ring to place buffers on
1527 * @cleaned_count: number of buffers to replace
1529 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1531 union ixgbe_adv_rx_desc *rx_desc;
1532 struct ixgbe_rx_buffer *bi;
1533 u16 i = rx_ring->next_to_use;
1539 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1540 bi = &rx_ring->rx_buffer_info[i];
1541 i -= rx_ring->count;
1544 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1548 * Refresh the desc even if buffer_addrs didn't change
1549 * because each write-back erases this info.
1551 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1557 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1558 bi = rx_ring->rx_buffer_info;
1559 i -= rx_ring->count;
1562 /* clear the status bits for the next_to_use descriptor */
1563 rx_desc->wb.upper.status_error = 0;
1566 } while (cleaned_count);
1568 i += rx_ring->count;
1570 if (rx_ring->next_to_use != i) {
1571 rx_ring->next_to_use = i;
1573 /* update next to alloc since we have filled the ring */
1574 rx_ring->next_to_alloc = i;
1576 /* Force memory writes to complete before letting h/w
1577 * know there are new descriptors to fetch. (Only
1578 * applicable for weak-ordered memory model archs,
1582 writel(i, rx_ring->tail);
1586 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1587 struct sk_buff *skb)
1589 u16 hdr_len = skb_headlen(skb);
1591 /* set gso_size to avoid messing up TCP MSS */
1592 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1593 IXGBE_CB(skb)->append_cnt);
1594 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1597 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1598 struct sk_buff *skb)
1600 /* if append_cnt is 0 then frame is not RSC */
1601 if (!IXGBE_CB(skb)->append_cnt)
1604 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1605 rx_ring->rx_stats.rsc_flush++;
1607 ixgbe_set_rsc_gso_size(rx_ring, skb);
1609 /* gso_size is computed using append_cnt so always clear it last */
1610 IXGBE_CB(skb)->append_cnt = 0;
1614 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1615 * @rx_ring: rx descriptor ring packet is being transacted on
1616 * @rx_desc: pointer to the EOP Rx descriptor
1617 * @skb: pointer to current skb being populated
1619 * This function checks the ring, descriptor, and packet information in
1620 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1621 * other fields within the skb.
1623 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1624 union ixgbe_adv_rx_desc *rx_desc,
1625 struct sk_buff *skb)
1627 struct net_device *dev = rx_ring->netdev;
1629 ixgbe_update_rsc_stats(rx_ring, skb);
1631 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1633 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1635 if (unlikely(ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_TS)))
1636 ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector->adapter, skb);
1638 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1639 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1640 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1641 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1644 skb_record_rx_queue(skb, rx_ring->queue_index);
1646 skb->protocol = eth_type_trans(skb, dev);
1649 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1650 struct sk_buff *skb)
1652 if (ixgbe_qv_busy_polling(q_vector))
1653 netif_receive_skb(skb);
1655 napi_gro_receive(&q_vector->napi, skb);
1659 * ixgbe_is_non_eop - process handling of non-EOP buffers
1660 * @rx_ring: Rx ring being processed
1661 * @rx_desc: Rx descriptor for current buffer
1662 * @skb: Current socket buffer containing buffer in progress
1664 * This function updates next to clean. If the buffer is an EOP buffer
1665 * this function exits returning false, otherwise it will place the
1666 * sk_buff in the next buffer to be chained and return true indicating
1667 * that this is in fact a non-EOP buffer.
1669 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1670 union ixgbe_adv_rx_desc *rx_desc,
1671 struct sk_buff *skb)
1673 u32 ntc = rx_ring->next_to_clean + 1;
1675 /* fetch, update, and store next to clean */
1676 ntc = (ntc < rx_ring->count) ? ntc : 0;
1677 rx_ring->next_to_clean = ntc;
1679 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1681 /* update RSC append count if present */
1682 if (ring_is_rsc_enabled(rx_ring)) {
1683 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1684 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1686 if (unlikely(rsc_enabled)) {
1687 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1689 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1690 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1692 /* update ntc based on RSC value */
1693 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1694 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1695 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1699 /* if we are the last buffer then there is nothing else to do */
1700 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1703 /* place skb in next buffer to be received */
1704 rx_ring->rx_buffer_info[ntc].skb = skb;
1705 rx_ring->rx_stats.non_eop_descs++;
1711 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1712 * @rx_ring: rx descriptor ring packet is being transacted on
1713 * @skb: pointer to current skb being adjusted
1715 * This function is an ixgbe specific version of __pskb_pull_tail. The
1716 * main difference between this version and the original function is that
1717 * this function can make several assumptions about the state of things
1718 * that allow for significant optimizations versus the standard function.
1719 * As a result we can do things like drop a frag and maintain an accurate
1720 * truesize for the skb.
1722 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1723 struct sk_buff *skb)
1725 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1727 unsigned int pull_len;
1730 * it is valid to use page_address instead of kmap since we are
1731 * working with pages allocated out of the lomem pool per
1732 * alloc_page(GFP_ATOMIC)
1734 va = skb_frag_address(frag);
1737 * we need the header to contain the greater of either ETH_HLEN or
1738 * 60 bytes if the skb->len is less than 60 for skb_pad.
1740 pull_len = eth_get_headlen(va, IXGBE_RX_HDR_SIZE);
1742 /* align pull length to size of long to optimize memcpy performance */
1743 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1745 /* update all of the pointers */
1746 skb_frag_size_sub(frag, pull_len);
1747 frag->page_offset += pull_len;
1748 skb->data_len -= pull_len;
1749 skb->tail += pull_len;
1753 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1754 * @rx_ring: rx descriptor ring packet is being transacted on
1755 * @skb: pointer to current skb being updated
1757 * This function provides a basic DMA sync up for the first fragment of an
1758 * skb. The reason for doing this is that the first fragment cannot be
1759 * unmapped until we have reached the end of packet descriptor for a buffer
1762 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1763 struct sk_buff *skb)
1765 /* if the page was released unmap it, else just sync our portion */
1766 if (unlikely(IXGBE_CB(skb)->page_released)) {
1767 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1768 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1769 IXGBE_CB(skb)->page_released = false;
1771 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1773 dma_sync_single_range_for_cpu(rx_ring->dev,
1776 ixgbe_rx_bufsz(rx_ring),
1779 IXGBE_CB(skb)->dma = 0;
1783 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1784 * @rx_ring: rx descriptor ring packet is being transacted on
1785 * @rx_desc: pointer to the EOP Rx descriptor
1786 * @skb: pointer to current skb being fixed
1788 * Check for corrupted packet headers caused by senders on the local L2
1789 * embedded NIC switch not setting up their Tx Descriptors right. These
1790 * should be very rare.
1792 * Also address the case where we are pulling data in on pages only
1793 * and as such no data is present in the skb header.
1795 * In addition if skb is not at least 60 bytes we need to pad it so that
1796 * it is large enough to qualify as a valid Ethernet frame.
1798 * Returns true if an error was encountered and skb was freed.
1800 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1801 union ixgbe_adv_rx_desc *rx_desc,
1802 struct sk_buff *skb)
1804 struct net_device *netdev = rx_ring->netdev;
1806 /* verify that the packet does not have any known errors */
1807 if (unlikely(ixgbe_test_staterr(rx_desc,
1808 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1809 !(netdev->features & NETIF_F_RXALL))) {
1810 dev_kfree_skb_any(skb);
1814 /* place header in linear portion of buffer */
1815 if (skb_is_nonlinear(skb))
1816 ixgbe_pull_tail(rx_ring, skb);
1819 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1820 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1824 /* if eth_skb_pad returns an error the skb was freed */
1825 if (eth_skb_pad(skb))
1832 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1833 * @rx_ring: rx descriptor ring to store buffers on
1834 * @old_buff: donor buffer to have page reused
1836 * Synchronizes page for reuse by the adapter
1838 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1839 struct ixgbe_rx_buffer *old_buff)
1841 struct ixgbe_rx_buffer *new_buff;
1842 u16 nta = rx_ring->next_to_alloc;
1844 new_buff = &rx_ring->rx_buffer_info[nta];
1846 /* update, and store next to alloc */
1848 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1850 /* transfer page from old buffer to new buffer */
1851 *new_buff = *old_buff;
1853 /* sync the buffer for use by the device */
1854 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1855 new_buff->page_offset,
1856 ixgbe_rx_bufsz(rx_ring),
1860 static inline bool ixgbe_page_is_reserved(struct page *page)
1862 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1866 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1867 * @rx_ring: rx descriptor ring to transact packets on
1868 * @rx_buffer: buffer containing page to add
1869 * @rx_desc: descriptor containing length of buffer written by hardware
1870 * @skb: sk_buff to place the data into
1872 * This function will add the data contained in rx_buffer->page to the skb.
1873 * This is done either through a direct copy if the data in the buffer is
1874 * less than the skb header size, otherwise it will just attach the page as
1875 * a frag to the skb.
1877 * The function will then update the page offset if necessary and return
1878 * true if the buffer can be reused by the adapter.
1880 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1881 struct ixgbe_rx_buffer *rx_buffer,
1882 union ixgbe_adv_rx_desc *rx_desc,
1883 struct sk_buff *skb)
1885 struct page *page = rx_buffer->page;
1886 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1887 #if (PAGE_SIZE < 8192)
1888 unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1890 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1891 unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1892 ixgbe_rx_bufsz(rx_ring);
1895 if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1896 unsigned char *va = page_address(page) + rx_buffer->page_offset;
1898 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1900 /* page is not reserved, we can reuse buffer as-is */
1901 if (likely(!ixgbe_page_is_reserved(page)))
1904 /* this page cannot be reused so discard it */
1905 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1909 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1910 rx_buffer->page_offset, size, truesize);
1912 /* avoid re-using remote pages */
1913 if (unlikely(ixgbe_page_is_reserved(page)))
1916 #if (PAGE_SIZE < 8192)
1917 /* if we are only owner of page we can reuse it */
1918 if (unlikely(page_count(page) != 1))
1921 /* flip page offset to other buffer */
1922 rx_buffer->page_offset ^= truesize;
1924 /* move offset up to the next cache line */
1925 rx_buffer->page_offset += truesize;
1927 if (rx_buffer->page_offset > last_offset)
1931 /* Even if we own the page, we are not allowed to use atomic_set()
1932 * This would break get_page_unless_zero() users.
1934 atomic_inc(&page->_count);
1939 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1940 union ixgbe_adv_rx_desc *rx_desc)
1942 struct ixgbe_rx_buffer *rx_buffer;
1943 struct sk_buff *skb;
1946 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1947 page = rx_buffer->page;
1950 skb = rx_buffer->skb;
1953 void *page_addr = page_address(page) +
1954 rx_buffer->page_offset;
1956 /* prefetch first cache line of first page */
1957 prefetch(page_addr);
1958 #if L1_CACHE_BYTES < 128
1959 prefetch(page_addr + L1_CACHE_BYTES);
1962 /* allocate a skb to store the frags */
1963 skb = napi_alloc_skb(&rx_ring->q_vector->napi,
1965 if (unlikely(!skb)) {
1966 rx_ring->rx_stats.alloc_rx_buff_failed++;
1971 * we will be copying header into skb->data in
1972 * pskb_may_pull so it is in our interest to prefetch
1973 * it now to avoid a possible cache miss
1975 prefetchw(skb->data);
1978 * Delay unmapping of the first packet. It carries the
1979 * header information, HW may still access the header
1980 * after the writeback. Only unmap it when EOP is
1983 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1986 IXGBE_CB(skb)->dma = rx_buffer->dma;
1988 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
1989 ixgbe_dma_sync_frag(rx_ring, skb);
1992 /* we are reusing so sync this buffer for CPU use */
1993 dma_sync_single_range_for_cpu(rx_ring->dev,
1995 rx_buffer->page_offset,
1996 ixgbe_rx_bufsz(rx_ring),
1999 rx_buffer->skb = NULL;
2002 /* pull page into skb */
2003 if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
2004 /* hand second half of page back to the ring */
2005 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2006 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
2007 /* the page has been released from the ring */
2008 IXGBE_CB(skb)->page_released = true;
2010 /* we are not reusing the buffer so unmap it */
2011 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
2012 ixgbe_rx_pg_size(rx_ring),
2016 /* clear contents of buffer_info */
2017 rx_buffer->page = NULL;
2023 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2024 * @q_vector: structure containing interrupt and ring information
2025 * @rx_ring: rx descriptor ring to transact packets on
2026 * @budget: Total limit on number of packets to process
2028 * This function provides a "bounce buffer" approach to Rx interrupt
2029 * processing. The advantage to this is that on systems that have
2030 * expensive overhead for IOMMU access this provides a means of avoiding
2031 * it by maintaining the mapping of the page to the syste.
2033 * Returns amount of work completed
2035 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2036 struct ixgbe_ring *rx_ring,
2039 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2041 struct ixgbe_adapter *adapter = q_vector->adapter;
2043 unsigned int mss = 0;
2044 #endif /* IXGBE_FCOE */
2045 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2047 while (likely(total_rx_packets < budget)) {
2048 union ixgbe_adv_rx_desc *rx_desc;
2049 struct sk_buff *skb;
2051 /* return some buffers to hardware, one at a time is too slow */
2052 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2053 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2057 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2059 if (!rx_desc->wb.upper.status_error)
2062 /* This memory barrier is needed to keep us from reading
2063 * any other fields out of the rx_desc until we know the
2064 * descriptor has been written back
2068 /* retrieve a buffer from the ring */
2069 skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
2071 /* exit if we failed to retrieve a buffer */
2077 /* place incomplete frames back on ring for completion */
2078 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2081 /* verify the packet layout is correct */
2082 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2085 /* probably a little skewed due to removing CRC */
2086 total_rx_bytes += skb->len;
2088 /* populate checksum, timestamp, VLAN, and protocol */
2089 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2092 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2093 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2094 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2095 /* include DDPed FCoE data */
2096 if (ddp_bytes > 0) {
2098 mss = rx_ring->netdev->mtu -
2099 sizeof(struct fcoe_hdr) -
2100 sizeof(struct fc_frame_header) -
2101 sizeof(struct fcoe_crc_eof);
2105 total_rx_bytes += ddp_bytes;
2106 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2110 dev_kfree_skb_any(skb);
2115 #endif /* IXGBE_FCOE */
2116 skb_mark_napi_id(skb, &q_vector->napi);
2117 ixgbe_rx_skb(q_vector, skb);
2119 /* update budget accounting */
2123 u64_stats_update_begin(&rx_ring->syncp);
2124 rx_ring->stats.packets += total_rx_packets;
2125 rx_ring->stats.bytes += total_rx_bytes;
2126 u64_stats_update_end(&rx_ring->syncp);
2127 q_vector->rx.total_packets += total_rx_packets;
2128 q_vector->rx.total_bytes += total_rx_bytes;
2130 return total_rx_packets;
2133 #ifdef CONFIG_NET_RX_BUSY_POLL
2134 /* must be called with local_bh_disable()d */
2135 static int ixgbe_low_latency_recv(struct napi_struct *napi)
2137 struct ixgbe_q_vector *q_vector =
2138 container_of(napi, struct ixgbe_q_vector, napi);
2139 struct ixgbe_adapter *adapter = q_vector->adapter;
2140 struct ixgbe_ring *ring;
2143 if (test_bit(__IXGBE_DOWN, &adapter->state))
2144 return LL_FLUSH_FAILED;
2146 if (!ixgbe_qv_lock_poll(q_vector))
2147 return LL_FLUSH_BUSY;
2149 ixgbe_for_each_ring(ring, q_vector->rx) {
2150 found = ixgbe_clean_rx_irq(q_vector, ring, 4);
2151 #ifdef BP_EXTENDED_STATS
2153 ring->stats.cleaned += found;
2155 ring->stats.misses++;
2161 ixgbe_qv_unlock_poll(q_vector);
2165 #endif /* CONFIG_NET_RX_BUSY_POLL */
2168 * ixgbe_configure_msix - Configure MSI-X hardware
2169 * @adapter: board private structure
2171 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2174 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2176 struct ixgbe_q_vector *q_vector;
2180 /* Populate MSIX to EITR Select */
2181 if (adapter->num_vfs > 32) {
2182 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2183 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2187 * Populate the IVAR table and set the ITR values to the
2188 * corresponding register.
2190 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2191 struct ixgbe_ring *ring;
2192 q_vector = adapter->q_vector[v_idx];
2194 ixgbe_for_each_ring(ring, q_vector->rx)
2195 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2197 ixgbe_for_each_ring(ring, q_vector->tx)
2198 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2200 ixgbe_write_eitr(q_vector);
2203 switch (adapter->hw.mac.type) {
2204 case ixgbe_mac_82598EB:
2205 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2208 case ixgbe_mac_82599EB:
2209 case ixgbe_mac_X540:
2210 case ixgbe_mac_X550:
2211 case ixgbe_mac_X550EM_x:
2212 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2217 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2219 /* set up to autoclear timer, and the vectors */
2220 mask = IXGBE_EIMS_ENABLE_MASK;
2221 mask &= ~(IXGBE_EIMS_OTHER |
2222 IXGBE_EIMS_MAILBOX |
2225 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2228 enum latency_range {
2232 latency_invalid = 255
2236 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2237 * @q_vector: structure containing interrupt and ring information
2238 * @ring_container: structure containing ring performance data
2240 * Stores a new ITR value based on packets and byte
2241 * counts during the last interrupt. The advantage of per interrupt
2242 * computation is faster updates and more accurate ITR for the current
2243 * traffic pattern. Constants in this function were computed
2244 * based on theoretical maximum wire speed and thresholds were set based
2245 * on testing data as well as attempting to minimize response time
2246 * while increasing bulk throughput.
2247 * this functionality is controlled by the InterruptThrottleRate module
2248 * parameter (see ixgbe_param.c)
2250 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2251 struct ixgbe_ring_container *ring_container)
2253 int bytes = ring_container->total_bytes;
2254 int packets = ring_container->total_packets;
2257 u8 itr_setting = ring_container->itr;
2262 /* simple throttlerate management
2263 * 0-10MB/s lowest (100000 ints/s)
2264 * 10-20MB/s low (20000 ints/s)
2265 * 20-1249MB/s bulk (12000 ints/s)
2267 /* what was last interrupt timeslice? */
2268 timepassed_us = q_vector->itr >> 2;
2269 if (timepassed_us == 0)
2272 bytes_perint = bytes / timepassed_us; /* bytes/usec */
2274 switch (itr_setting) {
2275 case lowest_latency:
2276 if (bytes_perint > 10)
2277 itr_setting = low_latency;
2280 if (bytes_perint > 20)
2281 itr_setting = bulk_latency;
2282 else if (bytes_perint <= 10)
2283 itr_setting = lowest_latency;
2286 if (bytes_perint <= 20)
2287 itr_setting = low_latency;
2291 /* clear work counters since we have the values we need */
2292 ring_container->total_bytes = 0;
2293 ring_container->total_packets = 0;
2295 /* write updated itr to ring container */
2296 ring_container->itr = itr_setting;
2300 * ixgbe_write_eitr - write EITR register in hardware specific way
2301 * @q_vector: structure containing interrupt and ring information
2303 * This function is made to be called by ethtool and by the driver
2304 * when it needs to update EITR registers at runtime. Hardware
2305 * specific quirks/differences are taken care of here.
2307 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2309 struct ixgbe_adapter *adapter = q_vector->adapter;
2310 struct ixgbe_hw *hw = &adapter->hw;
2311 int v_idx = q_vector->v_idx;
2312 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2314 switch (adapter->hw.mac.type) {
2315 case ixgbe_mac_82598EB:
2316 /* must write high and low 16 bits to reset counter */
2317 itr_reg |= (itr_reg << 16);
2319 case ixgbe_mac_82599EB:
2320 case ixgbe_mac_X540:
2321 case ixgbe_mac_X550:
2322 case ixgbe_mac_X550EM_x:
2324 * set the WDIS bit to not clear the timer bits and cause an
2325 * immediate assertion of the interrupt
2327 itr_reg |= IXGBE_EITR_CNT_WDIS;
2332 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2335 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2337 u32 new_itr = q_vector->itr;
2340 ixgbe_update_itr(q_vector, &q_vector->tx);
2341 ixgbe_update_itr(q_vector, &q_vector->rx);
2343 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2345 switch (current_itr) {
2346 /* counts and packets in update_itr are dependent on these numbers */
2347 case lowest_latency:
2348 new_itr = IXGBE_100K_ITR;
2351 new_itr = IXGBE_20K_ITR;
2354 new_itr = IXGBE_12K_ITR;
2360 if (new_itr != q_vector->itr) {
2361 /* do an exponential smoothing */
2362 new_itr = (10 * new_itr * q_vector->itr) /
2363 ((9 * new_itr) + q_vector->itr);
2365 /* save the algorithm value here */
2366 q_vector->itr = new_itr;
2368 ixgbe_write_eitr(q_vector);
2373 * ixgbe_check_overtemp_subtask - check for over temperature
2374 * @adapter: pointer to adapter
2376 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2378 struct ixgbe_hw *hw = &adapter->hw;
2379 u32 eicr = adapter->interrupt_event;
2381 if (test_bit(__IXGBE_DOWN, &adapter->state))
2384 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2385 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2388 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2390 switch (hw->device_id) {
2391 case IXGBE_DEV_ID_82599_T3_LOM:
2393 * Since the warning interrupt is for both ports
2394 * we don't have to check if:
2395 * - This interrupt wasn't for our port.
2396 * - We may have missed the interrupt so always have to
2397 * check if we got a LSC
2399 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2400 !(eicr & IXGBE_EICR_LSC))
2403 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2405 bool link_up = false;
2407 hw->mac.ops.check_link(hw, &speed, &link_up, false);
2413 /* Check if this is not due to overtemp */
2414 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2419 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2421 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2425 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2427 adapter->interrupt_event = 0;
2430 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2432 struct ixgbe_hw *hw = &adapter->hw;
2434 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2435 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2436 e_crit(probe, "Fan has stopped, replace the adapter\n");
2437 /* write to clear the interrupt */
2438 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2442 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2444 struct ixgbe_hw *hw = &adapter->hw;
2446 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2449 switch (adapter->hw.mac.type) {
2450 case ixgbe_mac_82599EB:
2452 * Need to check link state so complete overtemp check
2455 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2456 (eicr & IXGBE_EICR_LSC)) &&
2457 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2458 adapter->interrupt_event = eicr;
2459 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2460 ixgbe_service_event_schedule(adapter);
2464 case ixgbe_mac_X540:
2465 if (!(eicr & IXGBE_EICR_TS))
2472 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2475 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2477 switch (hw->mac.type) {
2478 case ixgbe_mac_82598EB:
2479 if (hw->phy.type == ixgbe_phy_nl)
2482 case ixgbe_mac_82599EB:
2483 case ixgbe_mac_X550EM_x:
2484 switch (hw->mac.ops.get_media_type(hw)) {
2485 case ixgbe_media_type_fiber:
2486 case ixgbe_media_type_fiber_qsfp:
2496 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2498 struct ixgbe_hw *hw = &adapter->hw;
2499 u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2501 if (!ixgbe_is_sfp(hw))
2504 /* Later MAC's use different SDP */
2505 if (hw->mac.type >= ixgbe_mac_X540)
2506 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2508 if (eicr & eicr_mask) {
2509 /* Clear the interrupt */
2510 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2511 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2512 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2513 adapter->sfp_poll_time = 0;
2514 ixgbe_service_event_schedule(adapter);
2518 if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2519 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2520 /* Clear the interrupt */
2521 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2522 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2523 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2524 ixgbe_service_event_schedule(adapter);
2529 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2531 struct ixgbe_hw *hw = &adapter->hw;
2534 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2535 adapter->link_check_timeout = jiffies;
2536 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2537 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2538 IXGBE_WRITE_FLUSH(hw);
2539 ixgbe_service_event_schedule(adapter);
2543 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2547 struct ixgbe_hw *hw = &adapter->hw;
2549 switch (hw->mac.type) {
2550 case ixgbe_mac_82598EB:
2551 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2552 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2554 case ixgbe_mac_82599EB:
2555 case ixgbe_mac_X540:
2556 case ixgbe_mac_X550:
2557 case ixgbe_mac_X550EM_x:
2558 mask = (qmask & 0xFFFFFFFF);
2560 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2561 mask = (qmask >> 32);
2563 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2568 /* skip the flush */
2571 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2575 struct ixgbe_hw *hw = &adapter->hw;
2577 switch (hw->mac.type) {
2578 case ixgbe_mac_82598EB:
2579 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2580 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2582 case ixgbe_mac_82599EB:
2583 case ixgbe_mac_X540:
2584 case ixgbe_mac_X550:
2585 case ixgbe_mac_X550EM_x:
2586 mask = (qmask & 0xFFFFFFFF);
2588 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2589 mask = (qmask >> 32);
2591 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2596 /* skip the flush */
2600 * ixgbe_irq_enable - Enable default interrupt generation settings
2601 * @adapter: board private structure
2603 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2606 struct ixgbe_hw *hw = &adapter->hw;
2607 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2609 /* don't reenable LSC while waiting for link */
2610 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2611 mask &= ~IXGBE_EIMS_LSC;
2613 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2614 switch (adapter->hw.mac.type) {
2615 case ixgbe_mac_82599EB:
2616 mask |= IXGBE_EIMS_GPI_SDP0(hw);
2618 case ixgbe_mac_X540:
2619 case ixgbe_mac_X550:
2620 case ixgbe_mac_X550EM_x:
2621 mask |= IXGBE_EIMS_TS;
2626 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2627 mask |= IXGBE_EIMS_GPI_SDP1(hw);
2628 switch (adapter->hw.mac.type) {
2629 case ixgbe_mac_82599EB:
2630 mask |= IXGBE_EIMS_GPI_SDP1(hw);
2631 mask |= IXGBE_EIMS_GPI_SDP2(hw);
2633 case ixgbe_mac_X540:
2634 case ixgbe_mac_X550:
2635 case ixgbe_mac_X550EM_x:
2636 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP)
2637 mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
2638 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
2639 mask |= IXGBE_EICR_GPI_SDP0_X540;
2640 mask |= IXGBE_EIMS_ECC;
2641 mask |= IXGBE_EIMS_MAILBOX;
2647 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2648 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2649 mask |= IXGBE_EIMS_FLOW_DIR;
2651 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2653 ixgbe_irq_enable_queues(adapter, ~0);
2655 IXGBE_WRITE_FLUSH(&adapter->hw);
2658 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2660 struct ixgbe_adapter *adapter = data;
2661 struct ixgbe_hw *hw = &adapter->hw;
2665 * Workaround for Silicon errata. Use clear-by-write instead
2666 * of clear-by-read. Reading with EICS will return the
2667 * interrupt causes without clearing, which later be done
2668 * with the write to EICR.
2670 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2672 /* The lower 16bits of the EICR register are for the queue interrupts
2673 * which should be masked here in order to not accidentally clear them if
2674 * the bits are high when ixgbe_msix_other is called. There is a race
2675 * condition otherwise which results in possible performance loss
2676 * especially if the ixgbe_msix_other interrupt is triggering
2677 * consistently (as it would when PPS is turned on for the X540 device)
2681 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2683 if (eicr & IXGBE_EICR_LSC)
2684 ixgbe_check_lsc(adapter);
2686 if (eicr & IXGBE_EICR_MAILBOX)
2687 ixgbe_msg_task(adapter);
2689 switch (hw->mac.type) {
2690 case ixgbe_mac_82599EB:
2691 case ixgbe_mac_X540:
2692 case ixgbe_mac_X550:
2693 case ixgbe_mac_X550EM_x:
2694 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
2695 (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
2696 adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
2697 ixgbe_service_event_schedule(adapter);
2698 IXGBE_WRITE_REG(hw, IXGBE_EICR,
2699 IXGBE_EICR_GPI_SDP0_X540);
2701 if (eicr & IXGBE_EICR_ECC) {
2702 e_info(link, "Received ECC Err, initiating reset\n");
2703 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2704 ixgbe_service_event_schedule(adapter);
2705 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2707 /* Handle Flow Director Full threshold interrupt */
2708 if (eicr & IXGBE_EICR_FLOW_DIR) {
2709 int reinit_count = 0;
2711 for (i = 0; i < adapter->num_tx_queues; i++) {
2712 struct ixgbe_ring *ring = adapter->tx_ring[i];
2713 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2718 /* no more flow director interrupts until after init */
2719 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2720 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2721 ixgbe_service_event_schedule(adapter);
2724 ixgbe_check_sfp_event(adapter, eicr);
2725 ixgbe_check_overtemp_event(adapter, eicr);
2731 ixgbe_check_fan_failure(adapter, eicr);
2733 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2734 ixgbe_ptp_check_pps_event(adapter, eicr);
2736 /* re-enable the original interrupt state, no lsc, no queues */
2737 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2738 ixgbe_irq_enable(adapter, false, false);
2743 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2745 struct ixgbe_q_vector *q_vector = data;
2747 /* EIAM disabled interrupts (on this vector) for us */
2749 if (q_vector->rx.ring || q_vector->tx.ring)
2750 napi_schedule(&q_vector->napi);
2756 * ixgbe_poll - NAPI Rx polling callback
2757 * @napi: structure for representing this polling device
2758 * @budget: how many packets driver is allowed to clean
2760 * This function is used for legacy and MSI, NAPI mode
2762 int ixgbe_poll(struct napi_struct *napi, int budget)
2764 struct ixgbe_q_vector *q_vector =
2765 container_of(napi, struct ixgbe_q_vector, napi);
2766 struct ixgbe_adapter *adapter = q_vector->adapter;
2767 struct ixgbe_ring *ring;
2768 int per_ring_budget;
2769 bool clean_complete = true;
2771 #ifdef CONFIG_IXGBE_DCA
2772 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2773 ixgbe_update_dca(q_vector);
2776 ixgbe_for_each_ring(ring, q_vector->tx)
2777 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2779 if (!ixgbe_qv_lock_napi(q_vector))
2782 /* attempt to distribute budget to each queue fairly, but don't allow
2783 * the budget to go below 1 because we'll exit polling */
2784 if (q_vector->rx.count > 1)
2785 per_ring_budget = max(budget/q_vector->rx.count, 1);
2787 per_ring_budget = budget;
2789 ixgbe_for_each_ring(ring, q_vector->rx)
2790 clean_complete &= (ixgbe_clean_rx_irq(q_vector, ring,
2791 per_ring_budget) < per_ring_budget);
2793 ixgbe_qv_unlock_napi(q_vector);
2794 /* If all work not completed, return budget and keep polling */
2795 if (!clean_complete)
2798 /* all work done, exit the polling mode */
2799 napi_complete(napi);
2800 if (adapter->rx_itr_setting & 1)
2801 ixgbe_set_itr(q_vector);
2802 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2803 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2809 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2810 * @adapter: board private structure
2812 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2813 * interrupts from the kernel.
2815 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2817 struct net_device *netdev = adapter->netdev;
2821 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2822 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2823 struct msix_entry *entry = &adapter->msix_entries[vector];
2825 if (q_vector->tx.ring && q_vector->rx.ring) {
2826 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2827 "%s-%s-%d", netdev->name, "TxRx", ri++);
2829 } else if (q_vector->rx.ring) {
2830 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2831 "%s-%s-%d", netdev->name, "rx", ri++);
2832 } else if (q_vector->tx.ring) {
2833 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2834 "%s-%s-%d", netdev->name, "tx", ti++);
2836 /* skip this unused q_vector */
2839 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2840 q_vector->name, q_vector);
2842 e_err(probe, "request_irq failed for MSIX interrupt "
2843 "Error: %d\n", err);
2844 goto free_queue_irqs;
2846 /* If Flow Director is enabled, set interrupt affinity */
2847 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2848 /* assign the mask for this irq */
2849 irq_set_affinity_hint(entry->vector,
2850 &q_vector->affinity_mask);
2854 err = request_irq(adapter->msix_entries[vector].vector,
2855 ixgbe_msix_other, 0, netdev->name, adapter);
2857 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2858 goto free_queue_irqs;
2866 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2868 free_irq(adapter->msix_entries[vector].vector,
2869 adapter->q_vector[vector]);
2871 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2872 pci_disable_msix(adapter->pdev);
2873 kfree(adapter->msix_entries);
2874 adapter->msix_entries = NULL;
2879 * ixgbe_intr - legacy mode Interrupt Handler
2880 * @irq: interrupt number
2881 * @data: pointer to a network interface device structure
2883 static irqreturn_t ixgbe_intr(int irq, void *data)
2885 struct ixgbe_adapter *adapter = data;
2886 struct ixgbe_hw *hw = &adapter->hw;
2887 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2891 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2892 * before the read of EICR.
2894 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2896 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2897 * therefore no explicit interrupt disable is necessary */
2898 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2901 * shared interrupt alert!
2902 * make sure interrupts are enabled because the read will
2903 * have disabled interrupts due to EIAM
2904 * finish the workaround of silicon errata on 82598. Unmask
2905 * the interrupt that we masked before the EICR read.
2907 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2908 ixgbe_irq_enable(adapter, true, true);
2909 return IRQ_NONE; /* Not our interrupt */
2912 if (eicr & IXGBE_EICR_LSC)
2913 ixgbe_check_lsc(adapter);
2915 switch (hw->mac.type) {
2916 case ixgbe_mac_82599EB:
2917 ixgbe_check_sfp_event(adapter, eicr);
2919 case ixgbe_mac_X540:
2920 case ixgbe_mac_X550:
2921 case ixgbe_mac_X550EM_x:
2922 if (eicr & IXGBE_EICR_ECC) {
2923 e_info(link, "Received ECC Err, initiating reset\n");
2924 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2925 ixgbe_service_event_schedule(adapter);
2926 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2928 ixgbe_check_overtemp_event(adapter, eicr);
2934 ixgbe_check_fan_failure(adapter, eicr);
2935 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2936 ixgbe_ptp_check_pps_event(adapter, eicr);
2938 /* would disable interrupts here but EIAM disabled it */
2939 napi_schedule(&q_vector->napi);
2942 * re-enable link(maybe) and non-queue interrupts, no flush.
2943 * ixgbe_poll will re-enable the queue interrupts
2945 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2946 ixgbe_irq_enable(adapter, false, false);
2952 * ixgbe_request_irq - initialize interrupts
2953 * @adapter: board private structure
2955 * Attempts to configure interrupts using the best available
2956 * capabilities of the hardware and kernel.
2958 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2960 struct net_device *netdev = adapter->netdev;
2963 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2964 err = ixgbe_request_msix_irqs(adapter);
2965 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2966 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2967 netdev->name, adapter);
2969 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2970 netdev->name, adapter);
2973 e_err(probe, "request_irq failed, Error %d\n", err);
2978 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2982 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2983 free_irq(adapter->pdev->irq, adapter);
2987 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2988 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2989 struct msix_entry *entry = &adapter->msix_entries[vector];
2991 /* free only the irqs that were actually requested */
2992 if (!q_vector->rx.ring && !q_vector->tx.ring)
2995 /* clear the affinity_mask in the IRQ descriptor */
2996 irq_set_affinity_hint(entry->vector, NULL);
2998 free_irq(entry->vector, q_vector);
3001 free_irq(adapter->msix_entries[vector++].vector, adapter);
3005 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3006 * @adapter: board private structure
3008 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3010 switch (adapter->hw.mac.type) {
3011 case ixgbe_mac_82598EB:
3012 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3014 case ixgbe_mac_82599EB:
3015 case ixgbe_mac_X540:
3016 case ixgbe_mac_X550:
3017 case ixgbe_mac_X550EM_x:
3018 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3019 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3020 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3025 IXGBE_WRITE_FLUSH(&adapter->hw);
3026 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3029 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3030 synchronize_irq(adapter->msix_entries[vector].vector);
3032 synchronize_irq(adapter->msix_entries[vector++].vector);
3034 synchronize_irq(adapter->pdev->irq);
3039 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3042 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3044 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3046 ixgbe_write_eitr(q_vector);
3048 ixgbe_set_ivar(adapter, 0, 0, 0);
3049 ixgbe_set_ivar(adapter, 1, 0, 0);
3051 e_info(hw, "Legacy interrupt IVAR setup done\n");
3055 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3056 * @adapter: board private structure
3057 * @ring: structure containing ring specific data
3059 * Configure the Tx descriptor ring after a reset.
3061 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3062 struct ixgbe_ring *ring)
3064 struct ixgbe_hw *hw = &adapter->hw;
3065 u64 tdba = ring->dma;
3067 u32 txdctl = IXGBE_TXDCTL_ENABLE;
3068 u8 reg_idx = ring->reg_idx;
3070 /* disable queue to avoid issues while updating state */
3071 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3072 IXGBE_WRITE_FLUSH(hw);
3074 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3075 (tdba & DMA_BIT_MASK(32)));
3076 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3077 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3078 ring->count * sizeof(union ixgbe_adv_tx_desc));
3079 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3080 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3081 ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3084 * set WTHRESH to encourage burst writeback, it should not be set
3085 * higher than 1 when:
3086 * - ITR is 0 as it could cause false TX hangs
3087 * - ITR is set to > 100k int/sec and BQL is enabled
3089 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3090 * to or less than the number of on chip descriptors, which is
3093 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3094 txdctl |= (1 << 16); /* WTHRESH = 1 */
3096 txdctl |= (8 << 16); /* WTHRESH = 8 */
3099 * Setting PTHRESH to 32 both improves performance
3100 * and avoids a TX hang with DFP enabled
3102 txdctl |= (1 << 8) | /* HTHRESH = 1 */
3103 32; /* PTHRESH = 32 */
3105 /* reinitialize flowdirector state */
3106 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3107 ring->atr_sample_rate = adapter->atr_sample_rate;
3108 ring->atr_count = 0;
3109 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3111 ring->atr_sample_rate = 0;
3114 /* initialize XPS */
3115 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3116 struct ixgbe_q_vector *q_vector = ring->q_vector;
3119 netif_set_xps_queue(ring->netdev,
3120 &q_vector->affinity_mask,
3124 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3127 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3129 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3130 if (hw->mac.type == ixgbe_mac_82598EB &&
3131 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3134 /* poll to verify queue is enabled */
3136 usleep_range(1000, 2000);
3137 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3138 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3140 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
3143 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3145 struct ixgbe_hw *hw = &adapter->hw;
3147 u8 tcs = netdev_get_num_tc(adapter->netdev);
3149 if (hw->mac.type == ixgbe_mac_82598EB)
3152 /* disable the arbiter while setting MTQC */
3153 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3154 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3155 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3157 /* set transmit pool layout */
3158 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3159 mtqc = IXGBE_MTQC_VT_ENA;
3161 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3163 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3164 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3165 mtqc |= IXGBE_MTQC_32VF;
3167 mtqc |= IXGBE_MTQC_64VF;
3170 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3172 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3174 mtqc = IXGBE_MTQC_64Q_1PB;
3177 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3179 /* Enable Security TX Buffer IFG for multiple pb */
3181 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3182 sectx |= IXGBE_SECTX_DCB;
3183 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3186 /* re-enable the arbiter */
3187 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3188 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3192 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3193 * @adapter: board private structure
3195 * Configure the Tx unit of the MAC after a reset.
3197 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3199 struct ixgbe_hw *hw = &adapter->hw;
3203 ixgbe_setup_mtqc(adapter);
3205 if (hw->mac.type != ixgbe_mac_82598EB) {
3206 /* DMATXCTL.EN must be before Tx queues are enabled */
3207 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3208 dmatxctl |= IXGBE_DMATXCTL_TE;
3209 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3212 /* Setup the HW Tx Head and Tail descriptor pointers */
3213 for (i = 0; i < adapter->num_tx_queues; i++)
3214 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3217 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3218 struct ixgbe_ring *ring)
3220 struct ixgbe_hw *hw = &adapter->hw;
3221 u8 reg_idx = ring->reg_idx;
3222 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3224 srrctl |= IXGBE_SRRCTL_DROP_EN;
3226 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3229 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3230 struct ixgbe_ring *ring)
3232 struct ixgbe_hw *hw = &adapter->hw;
3233 u8 reg_idx = ring->reg_idx;
3234 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3236 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3238 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3241 #ifdef CONFIG_IXGBE_DCB
3242 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3244 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3248 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3250 if (adapter->ixgbe_ieee_pfc)
3251 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3254 * We should set the drop enable bit if:
3257 * Number of Rx queues > 1 and flow control is disabled
3259 * This allows us to avoid head of line blocking for security
3260 * and performance reasons.
3262 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3263 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3264 for (i = 0; i < adapter->num_rx_queues; i++)
3265 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3267 for (i = 0; i < adapter->num_rx_queues; i++)
3268 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3272 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3274 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3275 struct ixgbe_ring *rx_ring)
3277 struct ixgbe_hw *hw = &adapter->hw;
3279 u8 reg_idx = rx_ring->reg_idx;
3281 if (hw->mac.type == ixgbe_mac_82598EB) {
3282 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3285 * if VMDq is not active we must program one srrctl register
3286 * per RSS queue since we have enabled RDRXCTL.MVMEN
3291 /* configure header buffer length, needed for RSC */
3292 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3294 /* configure the packet buffer length */
3295 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3297 /* configure descriptor type */
3298 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3300 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3304 * Return a number of entries in the RSS indirection table
3306 * @adapter: device handle
3308 * - 82598/82599/X540: 128
3309 * - X550(non-SRIOV mode): 512
3310 * - X550(SRIOV mode): 64
3312 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3314 if (adapter->hw.mac.type < ixgbe_mac_X550)
3316 else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3323 * Write the RETA table to HW
3325 * @adapter: device handle
3327 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3329 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3331 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3332 struct ixgbe_hw *hw = &adapter->hw;
3335 u8 *indir_tbl = adapter->rss_indir_tbl;
3337 /* Fill out the redirection table as follows:
3338 * - 82598: 8 bit wide entries containing pair of 4 bit RSS
3340 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3341 * - X550: 8 bit wide entries containing 6 bit RSS index
3343 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3344 indices_multi = 0x11;
3346 indices_multi = 0x1;
3348 /* Write redirection table to HW */
3349 for (i = 0; i < reta_entries; i++) {
3350 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3353 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3355 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3363 * Write the RETA table to HW (for x550 devices in SRIOV mode)
3365 * @adapter: device handle
3367 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3369 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3371 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3372 struct ixgbe_hw *hw = &adapter->hw;
3374 unsigned int pf_pool = adapter->num_vfs;
3376 /* Write redirection table to HW */
3377 for (i = 0; i < reta_entries; i++) {
3378 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3380 IXGBE_WRITE_REG(hw, IXGBE_PFVFRETA(i >> 2, pf_pool),
3387 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3389 struct ixgbe_hw *hw = &adapter->hw;
3391 u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3392 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3394 /* Program table for at least 2 queues w/ SR-IOV so that VFs can
3395 * make full use of any rings they may have. We will use the
3396 * PSRTYPE register to control how many rings we use within the PF.
3398 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
3401 /* Fill out hash function seeds */
3402 for (i = 0; i < 10; i++)
3403 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3405 /* Fill out redirection table */
3406 memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3408 for (i = 0, j = 0; i < reta_entries; i++, j++) {
3412 adapter->rss_indir_tbl[i] = j;
3415 ixgbe_store_reta(adapter);
3418 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3420 struct ixgbe_hw *hw = &adapter->hw;
3421 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3422 unsigned int pf_pool = adapter->num_vfs;
3425 /* Fill out hash function seeds */
3426 for (i = 0; i < 10; i++)
3427 IXGBE_WRITE_REG(hw, IXGBE_PFVFRSSRK(i, pf_pool),
3428 adapter->rss_key[i]);
3430 /* Fill out the redirection table */
3431 for (i = 0, j = 0; i < 64; i++, j++) {
3435 adapter->rss_indir_tbl[i] = j;
3438 ixgbe_store_vfreta(adapter);
3441 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3443 struct ixgbe_hw *hw = &adapter->hw;
3444 u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3447 /* Disable indicating checksum in descriptor, enables RSS hash */
3448 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3449 rxcsum |= IXGBE_RXCSUM_PCSD;
3450 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3452 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3453 if (adapter->ring_feature[RING_F_RSS].mask)
3454 mrqc = IXGBE_MRQC_RSSEN;
3456 u8 tcs = netdev_get_num_tc(adapter->netdev);
3458 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3460 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3462 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3463 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3464 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3466 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3469 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3471 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3473 mrqc = IXGBE_MRQC_RSSEN;
3477 /* Perform hash on these packet types */
3478 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3479 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3480 IXGBE_MRQC_RSS_FIELD_IPV6 |
3481 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3483 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3484 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3485 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3486 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3488 netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
3489 if ((hw->mac.type >= ixgbe_mac_X550) &&
3490 (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3491 unsigned int pf_pool = adapter->num_vfs;
3493 /* Enable VF RSS mode */
3494 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3495 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3497 /* Setup RSS through the VF registers */
3498 ixgbe_setup_vfreta(adapter);
3499 vfmrqc = IXGBE_MRQC_RSSEN;
3500 vfmrqc |= rss_field;
3501 IXGBE_WRITE_REG(hw, IXGBE_PFVFMRQC(pf_pool), vfmrqc);
3503 ixgbe_setup_reta(adapter);
3505 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3510 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3511 * @adapter: address of board private structure
3512 * @index: index of ring to set
3514 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3515 struct ixgbe_ring *ring)
3517 struct ixgbe_hw *hw = &adapter->hw;
3519 u8 reg_idx = ring->reg_idx;
3521 if (!ring_is_rsc_enabled(ring))
3524 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3525 rscctrl |= IXGBE_RSCCTL_RSCEN;
3527 * we must limit the number of descriptors so that the
3528 * total size of max desc * buf_len is not greater
3531 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3532 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3535 #define IXGBE_MAX_RX_DESC_POLL 10
3536 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3537 struct ixgbe_ring *ring)
3539 struct ixgbe_hw *hw = &adapter->hw;
3540 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3542 u8 reg_idx = ring->reg_idx;
3544 if (ixgbe_removed(hw->hw_addr))
3546 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3547 if (hw->mac.type == ixgbe_mac_82598EB &&
3548 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3552 usleep_range(1000, 2000);
3553 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3554 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3557 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3558 "the polling period\n", reg_idx);
3562 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3563 struct ixgbe_ring *ring)
3565 struct ixgbe_hw *hw = &adapter->hw;
3566 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3568 u8 reg_idx = ring->reg_idx;
3570 if (ixgbe_removed(hw->hw_addr))
3572 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3573 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3575 /* write value back with RXDCTL.ENABLE bit cleared */
3576 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3578 if (hw->mac.type == ixgbe_mac_82598EB &&
3579 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3582 /* the hardware may take up to 100us to really disable the rx queue */
3585 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3586 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3589 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3590 "the polling period\n", reg_idx);
3594 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3595 struct ixgbe_ring *ring)
3597 struct ixgbe_hw *hw = &adapter->hw;
3598 u64 rdba = ring->dma;
3600 u8 reg_idx = ring->reg_idx;
3602 /* disable queue to avoid issues while updating state */
3603 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3604 ixgbe_disable_rx_queue(adapter, ring);
3606 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3607 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3608 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3609 ring->count * sizeof(union ixgbe_adv_rx_desc));
3610 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3611 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3612 ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
3614 ixgbe_configure_srrctl(adapter, ring);
3615 ixgbe_configure_rscctl(adapter, ring);
3617 if (hw->mac.type == ixgbe_mac_82598EB) {
3619 * enable cache line friendly hardware writes:
3620 * PTHRESH=32 descriptors (half the internal cache),
3621 * this also removes ugly rx_no_buffer_count increment
3622 * HTHRESH=4 descriptors (to minimize latency on fetch)
3623 * WTHRESH=8 burst writeback up to two cache lines
3625 rxdctl &= ~0x3FFFFF;
3629 /* enable receive descriptor ring */
3630 rxdctl |= IXGBE_RXDCTL_ENABLE;
3631 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3633 ixgbe_rx_desc_queue_enable(adapter, ring);
3634 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3637 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3639 struct ixgbe_hw *hw = &adapter->hw;
3640 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3643 /* PSRTYPE must be initialized in non 82598 adapters */
3644 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3645 IXGBE_PSRTYPE_UDPHDR |
3646 IXGBE_PSRTYPE_IPV4HDR |
3647 IXGBE_PSRTYPE_L2HDR |
3648 IXGBE_PSRTYPE_IPV6HDR;
3650 if (hw->mac.type == ixgbe_mac_82598EB)
3658 for_each_set_bit(pool, &adapter->fwd_bitmask, 32)
3659 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
3662 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3664 struct ixgbe_hw *hw = &adapter->hw;
3665 u32 reg_offset, vf_shift;
3666 u32 gcr_ext, vmdctl;
3669 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3672 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3673 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3674 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3675 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3676 vmdctl |= IXGBE_VT_CTL_REPLEN;
3677 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3679 vf_shift = VMDQ_P(0) % 32;
3680 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3682 /* Enable only the PF's pool for Tx/Rx */
3683 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3684 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3685 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3686 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3687 if (adapter->bridge_mode == BRIDGE_MODE_VEB)
3688 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3690 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3691 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3694 * Set up VF register offsets for selected VT Mode,
3695 * i.e. 32 or 64 VFs for SR-IOV
3697 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3698 case IXGBE_82599_VMDQ_8Q_MASK:
3699 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3701 case IXGBE_82599_VMDQ_4Q_MASK:
3702 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3705 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3709 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3712 /* Enable MAC Anti-Spoofing */
3713 hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3716 /* Ensure LLDP is set for Ethertype Antispoofing if we will be
3717 * calling set_ethertype_anti_spoofing for each VF in loop below
3719 if (hw->mac.ops.set_ethertype_anti_spoofing)
3720 IXGBE_WRITE_REG(hw, IXGBE_ETQF(IXGBE_ETQF_FILTER_LLDP),
3721 (IXGBE_ETQF_FILTER_EN | /* enable filter */
3722 IXGBE_ETQF_TX_ANTISPOOF | /* tx antispoof */
3723 IXGBE_ETH_P_LLDP)); /* LLDP eth type */
3725 /* For VFs that have spoof checking turned off */
3726 for (i = 0; i < adapter->num_vfs; i++) {
3727 if (!adapter->vfinfo[i].spoofchk_enabled)
3728 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3730 /* enable ethertype anti spoofing if hw supports it */
3731 if (hw->mac.ops.set_ethertype_anti_spoofing)
3732 hw->mac.ops.set_ethertype_anti_spoofing(hw, true, i);
3734 /* Enable/Disable RSS query feature */
3735 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
3736 adapter->vfinfo[i].rss_query_enabled);
3740 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3742 struct ixgbe_hw *hw = &adapter->hw;
3743 struct net_device *netdev = adapter->netdev;
3744 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3745 struct ixgbe_ring *rx_ring;
3750 /* adjust max frame to be able to do baby jumbo for FCoE */
3751 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3752 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3753 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3755 #endif /* IXGBE_FCOE */
3757 /* adjust max frame to be at least the size of a standard frame */
3758 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3759 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3761 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3762 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3763 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3764 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3766 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3769 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3770 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3771 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3772 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3775 * Setup the HW Rx Head and Tail Descriptor Pointers and
3776 * the Base and Length of the Rx Descriptor Ring
3778 for (i = 0; i < adapter->num_rx_queues; i++) {
3779 rx_ring = adapter->rx_ring[i];
3780 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3781 set_ring_rsc_enabled(rx_ring);
3783 clear_ring_rsc_enabled(rx_ring);
3787 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3789 struct ixgbe_hw *hw = &adapter->hw;
3790 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3792 switch (hw->mac.type) {
3793 case ixgbe_mac_82598EB:
3795 * For VMDq support of different descriptor types or
3796 * buffer sizes through the use of multiple SRRCTL
3797 * registers, RDRXCTL.MVMEN must be set to 1
3799 * also, the manual doesn't mention it clearly but DCA hints
3800 * will only use queue 0's tags unless this bit is set. Side
3801 * effects of setting this bit are only that SRRCTL must be
3802 * fully programmed [0..15]
3804 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3806 case ixgbe_mac_X550:
3807 case ixgbe_mac_X550EM_x:
3808 if (adapter->num_vfs)
3809 rdrxctl |= IXGBE_RDRXCTL_PSP;
3810 /* fall through for older HW */
3811 case ixgbe_mac_82599EB:
3812 case ixgbe_mac_X540:
3813 /* Disable RSC for ACK packets */
3814 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3815 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3816 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3817 /* hardware requires some bits to be set by default */
3818 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3819 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3822 /* We should do nothing since we don't know this hardware */
3826 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3830 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3831 * @adapter: board private structure
3833 * Configure the Rx unit of the MAC after a reset.
3835 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3837 struct ixgbe_hw *hw = &adapter->hw;
3841 /* disable receives while setting up the descriptors */
3842 hw->mac.ops.disable_rx(hw);
3844 ixgbe_setup_psrtype(adapter);
3845 ixgbe_setup_rdrxctl(adapter);
3848 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
3849 rfctl &= ~IXGBE_RFCTL_RSC_DIS;
3850 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
3851 rfctl |= IXGBE_RFCTL_RSC_DIS;
3852 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
3854 /* Program registers for the distribution of queues */
3855 ixgbe_setup_mrqc(adapter);
3857 /* set_rx_buffer_len must be called before ring initialization */
3858 ixgbe_set_rx_buffer_len(adapter);
3861 * Setup the HW Rx Head and Tail Descriptor Pointers and
3862 * the Base and Length of the Rx Descriptor Ring
3864 for (i = 0; i < adapter->num_rx_queues; i++)
3865 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3867 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3868 /* disable drop enable for 82598 parts */
3869 if (hw->mac.type == ixgbe_mac_82598EB)
3870 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3872 /* enable all receives */
3873 rxctrl |= IXGBE_RXCTRL_RXEN;
3874 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3877 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
3878 __be16 proto, u16 vid)
3880 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3881 struct ixgbe_hw *hw = &adapter->hw;
3883 /* add VID to filter table */
3884 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3885 set_bit(vid, adapter->active_vlans);
3890 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
3891 __be16 proto, u16 vid)
3893 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3894 struct ixgbe_hw *hw = &adapter->hw;
3896 /* remove VID from filter table */
3897 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3898 clear_bit(vid, adapter->active_vlans);
3904 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3905 * @adapter: driver data
3907 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3909 struct ixgbe_hw *hw = &adapter->hw;
3913 switch (hw->mac.type) {
3914 case ixgbe_mac_82598EB:
3915 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3916 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3917 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3919 case ixgbe_mac_82599EB:
3920 case ixgbe_mac_X540:
3921 case ixgbe_mac_X550:
3922 case ixgbe_mac_X550EM_x:
3923 for (i = 0; i < adapter->num_rx_queues; i++) {
3924 struct ixgbe_ring *ring = adapter->rx_ring[i];
3926 if (ring->l2_accel_priv)
3929 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3930 vlnctrl &= ~IXGBE_RXDCTL_VME;
3931 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3940 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3941 * @adapter: driver data
3943 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3945 struct ixgbe_hw *hw = &adapter->hw;
3949 switch (hw->mac.type) {
3950 case ixgbe_mac_82598EB:
3951 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3952 vlnctrl |= IXGBE_VLNCTRL_VME;
3953 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3955 case ixgbe_mac_82599EB:
3956 case ixgbe_mac_X540:
3957 case ixgbe_mac_X550:
3958 case ixgbe_mac_X550EM_x:
3959 for (i = 0; i < adapter->num_rx_queues; i++) {
3960 struct ixgbe_ring *ring = adapter->rx_ring[i];
3962 if (ring->l2_accel_priv)
3965 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3966 vlnctrl |= IXGBE_RXDCTL_VME;
3967 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3975 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3979 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
3981 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3982 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
3986 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
3987 * @netdev: network interface device structure
3989 * Writes multicast address list to the MTA hash table.
3990 * Returns: -ENOMEM on failure
3991 * 0 on no addresses written
3992 * X on writing X addresses to MTA
3994 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
3996 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3997 struct ixgbe_hw *hw = &adapter->hw;
3999 if (!netif_running(netdev))
4002 if (hw->mac.ops.update_mc_addr_list)
4003 hw->mac.ops.update_mc_addr_list(hw, netdev);
4007 #ifdef CONFIG_PCI_IOV
4008 ixgbe_restore_vf_multicasts(adapter);
4011 return netdev_mc_count(netdev);
4014 #ifdef CONFIG_PCI_IOV
4015 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4017 struct ixgbe_hw *hw = &adapter->hw;
4019 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4020 if (adapter->mac_table[i].state & IXGBE_MAC_STATE_IN_USE)
4021 hw->mac.ops.set_rar(hw, i, adapter->mac_table[i].addr,
4022 adapter->mac_table[i].queue,
4025 hw->mac.ops.clear_rar(hw, i);
4027 adapter->mac_table[i].state &= ~(IXGBE_MAC_STATE_MODIFIED);
4032 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4034 struct ixgbe_hw *hw = &adapter->hw;
4036 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4037 if (adapter->mac_table[i].state & IXGBE_MAC_STATE_MODIFIED) {
4038 if (adapter->mac_table[i].state &
4039 IXGBE_MAC_STATE_IN_USE)
4040 hw->mac.ops.set_rar(hw, i,
4041 adapter->mac_table[i].addr,
4042 adapter->mac_table[i].queue,
4045 hw->mac.ops.clear_rar(hw, i);
4047 adapter->mac_table[i].state &=
4048 ~(IXGBE_MAC_STATE_MODIFIED);
4053 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4056 struct ixgbe_hw *hw = &adapter->hw;
4058 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4059 adapter->mac_table[i].state |= IXGBE_MAC_STATE_MODIFIED;
4060 adapter->mac_table[i].state &= ~IXGBE_MAC_STATE_IN_USE;
4061 eth_zero_addr(adapter->mac_table[i].addr);
4062 adapter->mac_table[i].queue = 0;
4064 ixgbe_sync_mac_table(adapter);
4067 static int ixgbe_available_rars(struct ixgbe_adapter *adapter)
4069 struct ixgbe_hw *hw = &adapter->hw;
4072 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4073 if (adapter->mac_table[i].state == 0)
4079 /* this function destroys the first RAR entry */
4080 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter,
4083 struct ixgbe_hw *hw = &adapter->hw;
4085 memcpy(&adapter->mac_table[0].addr, addr, ETH_ALEN);
4086 adapter->mac_table[0].queue = VMDQ_P(0);
4087 adapter->mac_table[0].state = (IXGBE_MAC_STATE_DEFAULT |
4088 IXGBE_MAC_STATE_IN_USE);
4089 hw->mac.ops.set_rar(hw, 0, adapter->mac_table[0].addr,
4090 adapter->mac_table[0].queue,
4094 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter, u8 *addr, u16 queue)
4096 struct ixgbe_hw *hw = &adapter->hw;
4099 if (is_zero_ether_addr(addr))
4102 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4103 if (adapter->mac_table[i].state & IXGBE_MAC_STATE_IN_USE)
4105 adapter->mac_table[i].state |= (IXGBE_MAC_STATE_MODIFIED |
4106 IXGBE_MAC_STATE_IN_USE);
4107 ether_addr_copy(adapter->mac_table[i].addr, addr);
4108 adapter->mac_table[i].queue = queue;
4109 ixgbe_sync_mac_table(adapter);
4115 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter, u8 *addr, u16 queue)
4117 /* search table for addr, if found, set to 0 and sync */
4119 struct ixgbe_hw *hw = &adapter->hw;
4121 if (is_zero_ether_addr(addr))
4124 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4125 if (ether_addr_equal(addr, adapter->mac_table[i].addr) &&
4126 adapter->mac_table[i].queue == queue) {
4127 adapter->mac_table[i].state |= IXGBE_MAC_STATE_MODIFIED;
4128 adapter->mac_table[i].state &= ~IXGBE_MAC_STATE_IN_USE;
4129 eth_zero_addr(adapter->mac_table[i].addr);
4130 adapter->mac_table[i].queue = 0;
4131 ixgbe_sync_mac_table(adapter);
4138 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
4139 * @netdev: network interface device structure
4141 * Writes unicast address list to the RAR table.
4142 * Returns: -ENOMEM on failure/insufficient address space
4143 * 0 on no addresses written
4144 * X on writing X addresses to the RAR table
4146 static int ixgbe_write_uc_addr_list(struct net_device *netdev, int vfn)
4148 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4151 /* return ENOMEM indicating insufficient memory for addresses */
4152 if (netdev_uc_count(netdev) > ixgbe_available_rars(adapter))
4155 if (!netdev_uc_empty(netdev)) {
4156 struct netdev_hw_addr *ha;
4157 netdev_for_each_uc_addr(ha, netdev) {
4158 ixgbe_del_mac_filter(adapter, ha->addr, vfn);
4159 ixgbe_add_mac_filter(adapter, ha->addr, vfn);
4167 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4168 * @netdev: network interface device structure
4170 * The set_rx_method entry point is called whenever the unicast/multicast
4171 * address list or the network interface flags are updated. This routine is
4172 * responsible for configuring the hardware for proper unicast, multicast and
4175 void ixgbe_set_rx_mode(struct net_device *netdev)
4177 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4178 struct ixgbe_hw *hw = &adapter->hw;
4179 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4183 /* Check for Promiscuous and All Multicast modes */
4184 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4185 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4187 /* set all bits that we expect to always be set */
4188 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4189 fctrl |= IXGBE_FCTRL_BAM;
4190 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4191 fctrl |= IXGBE_FCTRL_PMCF;
4193 /* clear the bits we are changing the status of */
4194 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4195 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
4196 if (netdev->flags & IFF_PROMISC) {
4197 hw->addr_ctrl.user_set_promisc = true;
4198 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4199 vmolr |= IXGBE_VMOLR_MPE;
4200 /* Only disable hardware filter vlans in promiscuous mode
4201 * if SR-IOV and VMDQ are disabled - otherwise ensure
4202 * that hardware VLAN filters remain enabled.
4204 if (adapter->flags & (IXGBE_FLAG_VMDQ_ENABLED |
4205 IXGBE_FLAG_SRIOV_ENABLED))
4206 vlnctrl |= (IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
4208 if (netdev->flags & IFF_ALLMULTI) {
4209 fctrl |= IXGBE_FCTRL_MPE;
4210 vmolr |= IXGBE_VMOLR_MPE;
4212 vlnctrl |= IXGBE_VLNCTRL_VFE;
4213 hw->addr_ctrl.user_set_promisc = false;
4217 * Write addresses to available RAR registers, if there is not
4218 * sufficient space to store all the addresses then enable
4219 * unicast promiscuous mode
4221 count = ixgbe_write_uc_addr_list(netdev, VMDQ_P(0));
4223 fctrl |= IXGBE_FCTRL_UPE;
4224 vmolr |= IXGBE_VMOLR_ROPE;
4227 /* Write addresses to the MTA, if the attempt fails
4228 * then we should just turn on promiscuous mode so
4229 * that we can at least receive multicast traffic
4231 count = ixgbe_write_mc_addr_list(netdev);
4233 fctrl |= IXGBE_FCTRL_MPE;
4234 vmolr |= IXGBE_VMOLR_MPE;
4236 vmolr |= IXGBE_VMOLR_ROMPE;
4239 if (hw->mac.type != ixgbe_mac_82598EB) {
4240 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4241 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4243 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4246 /* This is useful for sniffing bad packets. */
4247 if (adapter->netdev->features & NETIF_F_RXALL) {
4248 /* UPE and MPE will be handled by normal PROMISC logic
4249 * in e1000e_set_rx_mode */
4250 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4251 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4252 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4254 fctrl &= ~(IXGBE_FCTRL_DPF);
4255 /* NOTE: VLAN filtering is disabled by setting PROMISC */
4258 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4259 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4261 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
4262 ixgbe_vlan_strip_enable(adapter);
4264 ixgbe_vlan_strip_disable(adapter);
4267 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4271 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4272 ixgbe_qv_init_lock(adapter->q_vector[q_idx]);
4273 napi_enable(&adapter->q_vector[q_idx]->napi);
4277 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4281 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4282 napi_disable(&adapter->q_vector[q_idx]->napi);
4283 while (!ixgbe_qv_disable(adapter->q_vector[q_idx])) {
4284 pr_info("QV %d locked\n", q_idx);
4285 usleep_range(1000, 20000);
4290 static void ixgbe_clear_vxlan_port(struct ixgbe_adapter *adapter)
4292 switch (adapter->hw.mac.type) {
4293 case ixgbe_mac_X550:
4294 case ixgbe_mac_X550EM_x:
4295 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VXLANCTRL, 0);
4296 #ifdef CONFIG_IXGBE_VXLAN
4297 adapter->vxlan_port = 0;
4305 #ifdef CONFIG_IXGBE_DCB
4307 * ixgbe_configure_dcb - Configure DCB hardware
4308 * @adapter: ixgbe adapter struct
4310 * This is called by the driver on open to configure the DCB hardware.
4311 * This is also called by the gennetlink interface when reconfiguring
4314 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
4316 struct ixgbe_hw *hw = &adapter->hw;
4317 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4319 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
4320 if (hw->mac.type == ixgbe_mac_82598EB)
4321 netif_set_gso_max_size(adapter->netdev, 65536);
4325 if (hw->mac.type == ixgbe_mac_82598EB)
4326 netif_set_gso_max_size(adapter->netdev, 32768);
4329 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
4330 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
4333 /* reconfigure the hardware */
4334 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
4335 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4337 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4339 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
4340 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
4341 ixgbe_dcb_hw_ets(&adapter->hw,
4342 adapter->ixgbe_ieee_ets,
4344 ixgbe_dcb_hw_pfc_config(&adapter->hw,
4345 adapter->ixgbe_ieee_pfc->pfc_en,
4346 adapter->ixgbe_ieee_ets->prio_tc);
4349 /* Enable RSS Hash per TC */
4350 if (hw->mac.type != ixgbe_mac_82598EB) {
4352 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
4359 /* write msb to all 8 TCs in one write */
4360 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
4365 /* Additional bittime to account for IXGBE framing */
4366 #define IXGBE_ETH_FRAMING 20
4369 * ixgbe_hpbthresh - calculate high water mark for flow control
4371 * @adapter: board private structure to calculate for
4372 * @pb: packet buffer to calculate
4374 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
4376 struct ixgbe_hw *hw = &adapter->hw;
4377 struct net_device *dev = adapter->netdev;
4378 int link, tc, kb, marker;
4381 /* Calculate max LAN frame size */
4382 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
4385 /* FCoE traffic class uses FCOE jumbo frames */
4386 if ((dev->features & NETIF_F_FCOE_MTU) &&
4387 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4388 (pb == ixgbe_fcoe_get_tc(adapter)))
4389 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4392 /* Calculate delay value for device */
4393 switch (hw->mac.type) {
4394 case ixgbe_mac_X540:
4395 case ixgbe_mac_X550:
4396 case ixgbe_mac_X550EM_x:
4397 dv_id = IXGBE_DV_X540(link, tc);
4400 dv_id = IXGBE_DV(link, tc);
4404 /* Loopback switch introduces additional latency */
4405 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4406 dv_id += IXGBE_B2BT(tc);
4408 /* Delay value is calculated in bit times convert to KB */
4409 kb = IXGBE_BT2KB(dv_id);
4410 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
4412 marker = rx_pba - kb;
4414 /* It is possible that the packet buffer is not large enough
4415 * to provide required headroom. In this case throw an error
4416 * to user and a do the best we can.
4419 e_warn(drv, "Packet Buffer(%i) can not provide enough"
4420 "headroom to support flow control."
4421 "Decrease MTU or number of traffic classes\n", pb);
4429 * ixgbe_lpbthresh - calculate low water mark for for flow control
4431 * @adapter: board private structure to calculate for
4432 * @pb: packet buffer to calculate
4434 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
4436 struct ixgbe_hw *hw = &adapter->hw;
4437 struct net_device *dev = adapter->netdev;
4441 /* Calculate max LAN frame size */
4442 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4445 /* FCoE traffic class uses FCOE jumbo frames */
4446 if ((dev->features & NETIF_F_FCOE_MTU) &&
4447 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4448 (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
4449 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4452 /* Calculate delay value for device */
4453 switch (hw->mac.type) {
4454 case ixgbe_mac_X540:
4455 case ixgbe_mac_X550:
4456 case ixgbe_mac_X550EM_x:
4457 dv_id = IXGBE_LOW_DV_X540(tc);
4460 dv_id = IXGBE_LOW_DV(tc);
4464 /* Delay value is calculated in bit times convert to KB */
4465 return IXGBE_BT2KB(dv_id);
4469 * ixgbe_pbthresh_setup - calculate and setup high low water marks
4471 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
4473 struct ixgbe_hw *hw = &adapter->hw;
4474 int num_tc = netdev_get_num_tc(adapter->netdev);
4480 for (i = 0; i < num_tc; i++) {
4481 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
4482 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
4484 /* Low water marks must not be larger than high water marks */
4485 if (hw->fc.low_water[i] > hw->fc.high_water[i])
4486 hw->fc.low_water[i] = 0;
4489 for (; i < MAX_TRAFFIC_CLASS; i++)
4490 hw->fc.high_water[i] = 0;
4493 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
4495 struct ixgbe_hw *hw = &adapter->hw;
4497 u8 tc = netdev_get_num_tc(adapter->netdev);
4499 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4500 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4501 hdrm = 32 << adapter->fdir_pballoc;
4505 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
4506 ixgbe_pbthresh_setup(adapter);
4509 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
4511 struct ixgbe_hw *hw = &adapter->hw;
4512 struct hlist_node *node2;
4513 struct ixgbe_fdir_filter *filter;
4515 spin_lock(&adapter->fdir_perfect_lock);
4517 if (!hlist_empty(&adapter->fdir_filter_list))
4518 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
4520 hlist_for_each_entry_safe(filter, node2,
4521 &adapter->fdir_filter_list, fdir_node) {
4522 ixgbe_fdir_write_perfect_filter_82599(hw,
4525 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
4526 IXGBE_FDIR_DROP_QUEUE :
4527 adapter->rx_ring[filter->action]->reg_idx);
4530 spin_unlock(&adapter->fdir_perfect_lock);
4533 static void ixgbe_macvlan_set_rx_mode(struct net_device *dev, unsigned int pool,
4534 struct ixgbe_adapter *adapter)
4536 struct ixgbe_hw *hw = &adapter->hw;
4539 /* No unicast promiscuous support for VMDQ devices. */
4540 vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(pool));
4541 vmolr |= (IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
4543 /* clear the affected bit */
4544 vmolr &= ~IXGBE_VMOLR_MPE;
4546 if (dev->flags & IFF_ALLMULTI) {
4547 vmolr |= IXGBE_VMOLR_MPE;
4549 vmolr |= IXGBE_VMOLR_ROMPE;
4550 hw->mac.ops.update_mc_addr_list(hw, dev);
4552 ixgbe_write_uc_addr_list(adapter->netdev, pool);
4553 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(pool), vmolr);
4556 static void ixgbe_fwd_psrtype(struct ixgbe_fwd_adapter *vadapter)
4558 struct ixgbe_adapter *adapter = vadapter->real_adapter;
4559 int rss_i = adapter->num_rx_queues_per_pool;
4560 struct ixgbe_hw *hw = &adapter->hw;
4561 u16 pool = vadapter->pool;
4562 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4563 IXGBE_PSRTYPE_UDPHDR |
4564 IXGBE_PSRTYPE_IPV4HDR |
4565 IXGBE_PSRTYPE_L2HDR |
4566 IXGBE_PSRTYPE_IPV6HDR;
4568 if (hw->mac.type == ixgbe_mac_82598EB)
4576 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4580 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4581 * @rx_ring: ring to free buffers from
4583 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4585 struct device *dev = rx_ring->dev;
4589 /* ring already cleared, nothing to do */
4590 if (!rx_ring->rx_buffer_info)
4593 /* Free all the Rx ring sk_buffs */
4594 for (i = 0; i < rx_ring->count; i++) {
4595 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
4597 if (rx_buffer->skb) {
4598 struct sk_buff *skb = rx_buffer->skb;
4599 if (IXGBE_CB(skb)->page_released)
4602 ixgbe_rx_bufsz(rx_ring),
4605 rx_buffer->skb = NULL;
4608 if (!rx_buffer->page)
4611 dma_unmap_page(dev, rx_buffer->dma,
4612 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
4613 __free_pages(rx_buffer->page, ixgbe_rx_pg_order(rx_ring));
4615 rx_buffer->page = NULL;
4618 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4619 memset(rx_ring->rx_buffer_info, 0, size);
4621 /* Zero out the descriptor ring */
4622 memset(rx_ring->desc, 0, rx_ring->size);
4624 rx_ring->next_to_alloc = 0;
4625 rx_ring->next_to_clean = 0;
4626 rx_ring->next_to_use = 0;
4629 static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter *vadapter,
4630 struct ixgbe_ring *rx_ring)
4632 struct ixgbe_adapter *adapter = vadapter->real_adapter;
4633 int index = rx_ring->queue_index + vadapter->rx_base_queue;
4635 /* shutdown specific queue receive and wait for dma to settle */
4636 ixgbe_disable_rx_queue(adapter, rx_ring);
4637 usleep_range(10000, 20000);
4638 ixgbe_irq_disable_queues(adapter, ((u64)1 << index));
4639 ixgbe_clean_rx_ring(rx_ring);
4640 rx_ring->l2_accel_priv = NULL;
4643 static int ixgbe_fwd_ring_down(struct net_device *vdev,
4644 struct ixgbe_fwd_adapter *accel)
4646 struct ixgbe_adapter *adapter = accel->real_adapter;
4647 unsigned int rxbase = accel->rx_base_queue;
4648 unsigned int txbase = accel->tx_base_queue;
4651 netif_tx_stop_all_queues(vdev);
4653 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4654 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4655 adapter->rx_ring[rxbase + i]->netdev = adapter->netdev;
4658 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4659 adapter->tx_ring[txbase + i]->l2_accel_priv = NULL;
4660 adapter->tx_ring[txbase + i]->netdev = adapter->netdev;
4667 static int ixgbe_fwd_ring_up(struct net_device *vdev,
4668 struct ixgbe_fwd_adapter *accel)
4670 struct ixgbe_adapter *adapter = accel->real_adapter;
4671 unsigned int rxbase, txbase, queues;
4672 int i, baseq, err = 0;
4674 if (!test_bit(accel->pool, &adapter->fwd_bitmask))
4677 baseq = accel->pool * adapter->num_rx_queues_per_pool;
4678 netdev_dbg(vdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
4679 accel->pool, adapter->num_rx_pools,
4680 baseq, baseq + adapter->num_rx_queues_per_pool,
4681 adapter->fwd_bitmask);
4683 accel->netdev = vdev;
4684 accel->rx_base_queue = rxbase = baseq;
4685 accel->tx_base_queue = txbase = baseq;
4687 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
4688 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4690 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4691 adapter->rx_ring[rxbase + i]->netdev = vdev;
4692 adapter->rx_ring[rxbase + i]->l2_accel_priv = accel;
4693 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[rxbase + i]);
4696 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4697 adapter->tx_ring[txbase + i]->netdev = vdev;
4698 adapter->tx_ring[txbase + i]->l2_accel_priv = accel;
4701 queues = min_t(unsigned int,
4702 adapter->num_rx_queues_per_pool, vdev->num_tx_queues);
4703 err = netif_set_real_num_tx_queues(vdev, queues);
4707 err = netif_set_real_num_rx_queues(vdev, queues);
4711 if (is_valid_ether_addr(vdev->dev_addr))
4712 ixgbe_add_mac_filter(adapter, vdev->dev_addr, accel->pool);
4714 ixgbe_fwd_psrtype(accel);
4715 ixgbe_macvlan_set_rx_mode(vdev, accel->pool, adapter);
4718 ixgbe_fwd_ring_down(vdev, accel);
4722 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
4724 struct net_device *upper;
4725 struct list_head *iter;
4728 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
4729 if (netif_is_macvlan(upper)) {
4730 struct macvlan_dev *dfwd = netdev_priv(upper);
4731 struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv;
4733 if (dfwd->fwd_priv) {
4734 err = ixgbe_fwd_ring_up(upper, vadapter);
4742 static void ixgbe_configure(struct ixgbe_adapter *adapter)
4744 struct ixgbe_hw *hw = &adapter->hw;
4746 ixgbe_configure_pb(adapter);
4747 #ifdef CONFIG_IXGBE_DCB
4748 ixgbe_configure_dcb(adapter);
4751 * We must restore virtualization before VLANs or else
4752 * the VLVF registers will not be populated
4754 ixgbe_configure_virtualization(adapter);
4756 ixgbe_set_rx_mode(adapter->netdev);
4757 ixgbe_restore_vlan(adapter);
4759 switch (hw->mac.type) {
4760 case ixgbe_mac_82599EB:
4761 case ixgbe_mac_X540:
4762 hw->mac.ops.disable_rx_buff(hw);
4768 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4769 ixgbe_init_fdir_signature_82599(&adapter->hw,
4770 adapter->fdir_pballoc);
4771 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
4772 ixgbe_init_fdir_perfect_82599(&adapter->hw,
4773 adapter->fdir_pballoc);
4774 ixgbe_fdir_filter_restore(adapter);
4777 switch (hw->mac.type) {
4778 case ixgbe_mac_82599EB:
4779 case ixgbe_mac_X540:
4780 hw->mac.ops.enable_rx_buff(hw);
4787 /* configure FCoE L2 filters, redirection table, and Rx control */
4788 ixgbe_configure_fcoe(adapter);
4790 #endif /* IXGBE_FCOE */
4791 ixgbe_configure_tx(adapter);
4792 ixgbe_configure_rx(adapter);
4793 ixgbe_configure_dfwd(adapter);
4797 * ixgbe_sfp_link_config - set up SFP+ link
4798 * @adapter: pointer to private adapter struct
4800 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
4803 * We are assuming the worst case scenario here, and that
4804 * is that an SFP was inserted/removed after the reset
4805 * but before SFP detection was enabled. As such the best
4806 * solution is to just start searching as soon as we start
4808 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
4809 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
4811 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
4812 adapter->sfp_poll_time = 0;
4816 * ixgbe_non_sfp_link_config - set up non-SFP+ link
4817 * @hw: pointer to private hardware struct
4819 * Returns 0 on success, negative on failure
4821 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
4824 bool autoneg, link_up = false;
4825 int ret = IXGBE_ERR_LINK_SETUP;
4827 if (hw->mac.ops.check_link)
4828 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
4833 speed = hw->phy.autoneg_advertised;
4834 if ((!speed) && (hw->mac.ops.get_link_capabilities))
4835 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
4840 if (hw->mac.ops.setup_link)
4841 ret = hw->mac.ops.setup_link(hw, speed, link_up);
4846 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
4848 struct ixgbe_hw *hw = &adapter->hw;
4851 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4852 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
4854 gpie |= IXGBE_GPIE_EIAME;
4856 * use EIAM to auto-mask when MSI-X interrupt is asserted
4857 * this saves a register write for every interrupt
4859 switch (hw->mac.type) {
4860 case ixgbe_mac_82598EB:
4861 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4863 case ixgbe_mac_82599EB:
4864 case ixgbe_mac_X540:
4865 case ixgbe_mac_X550:
4866 case ixgbe_mac_X550EM_x:
4868 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4869 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4873 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
4874 * specifically only auto mask tx and rx interrupts */
4875 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4878 /* XXX: to interrupt immediately for EICS writes, enable this */
4879 /* gpie |= IXGBE_GPIE_EIMEN; */
4881 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
4882 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
4884 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4885 case IXGBE_82599_VMDQ_8Q_MASK:
4886 gpie |= IXGBE_GPIE_VTMODE_16;
4888 case IXGBE_82599_VMDQ_4Q_MASK:
4889 gpie |= IXGBE_GPIE_VTMODE_32;
4892 gpie |= IXGBE_GPIE_VTMODE_64;
4897 /* Enable Thermal over heat sensor interrupt */
4898 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
4899 switch (adapter->hw.mac.type) {
4900 case ixgbe_mac_82599EB:
4901 gpie |= IXGBE_SDP0_GPIEN_8259X;
4903 case ixgbe_mac_X540:
4904 gpie |= IXGBE_EIMS_TS;
4911 /* Enable fan failure interrupt */
4912 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
4913 gpie |= IXGBE_SDP1_GPIEN(hw);
4915 switch (hw->mac.type) {
4916 case ixgbe_mac_82599EB:
4917 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
4919 case ixgbe_mac_X550EM_x:
4920 gpie |= IXGBE_SDP0_GPIEN_X540;
4926 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4929 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
4931 struct ixgbe_hw *hw = &adapter->hw;
4935 ixgbe_get_hw_control(adapter);
4936 ixgbe_setup_gpie(adapter);
4938 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4939 ixgbe_configure_msix(adapter);
4941 ixgbe_configure_msi_and_legacy(adapter);
4943 /* enable the optics for 82599 SFP+ fiber */
4944 if (hw->mac.ops.enable_tx_laser)
4945 hw->mac.ops.enable_tx_laser(hw);
4947 if (hw->phy.ops.set_phy_power)
4948 hw->phy.ops.set_phy_power(hw, true);
4950 smp_mb__before_atomic();
4951 clear_bit(__IXGBE_DOWN, &adapter->state);
4952 ixgbe_napi_enable_all(adapter);
4954 if (ixgbe_is_sfp(hw)) {
4955 ixgbe_sfp_link_config(adapter);
4957 err = ixgbe_non_sfp_link_config(hw);
4959 e_err(probe, "link_config FAILED %d\n", err);
4962 /* clear any pending interrupts, may auto mask */
4963 IXGBE_READ_REG(hw, IXGBE_EICR);
4964 ixgbe_irq_enable(adapter, true, true);
4967 * If this adapter has a fan, check to see if we had a failure
4968 * before we enabled the interrupt.
4970 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4971 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4972 if (esdp & IXGBE_ESDP_SDP1)
4973 e_crit(drv, "Fan has stopped, replace the adapter\n");
4976 /* bring the link up in the watchdog, this could race with our first
4977 * link up interrupt but shouldn't be a problem */
4978 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4979 adapter->link_check_timeout = jiffies;
4980 mod_timer(&adapter->service_timer, jiffies);
4982 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4983 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4984 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4985 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4988 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4990 WARN_ON(in_interrupt());
4991 /* put off any impending NetWatchDogTimeout */
4992 adapter->netdev->trans_start = jiffies;
4994 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4995 usleep_range(1000, 2000);
4996 ixgbe_down(adapter);
4998 * If SR-IOV enabled then wait a bit before bringing the adapter
4999 * back up to give the VFs time to respond to the reset. The
5000 * two second wait is based upon the watchdog timer cycle in
5003 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5006 clear_bit(__IXGBE_RESETTING, &adapter->state);
5009 void ixgbe_up(struct ixgbe_adapter *adapter)
5011 /* hardware has been reset, we need to reload some things */
5012 ixgbe_configure(adapter);
5014 ixgbe_up_complete(adapter);
5017 void ixgbe_reset(struct ixgbe_adapter *adapter)
5019 struct ixgbe_hw *hw = &adapter->hw;
5020 struct net_device *netdev = adapter->netdev;
5022 u8 old_addr[ETH_ALEN];
5024 if (ixgbe_removed(hw->hw_addr))
5026 /* lock SFP init bit to prevent race conditions with the watchdog */
5027 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5028 usleep_range(1000, 2000);
5030 /* clear all SFP and link config related flags while holding SFP_INIT */
5031 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5032 IXGBE_FLAG2_SFP_NEEDS_RESET);
5033 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5035 err = hw->mac.ops.init_hw(hw);
5038 case IXGBE_ERR_SFP_NOT_PRESENT:
5039 case IXGBE_ERR_SFP_NOT_SUPPORTED:
5041 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5042 e_dev_err("master disable timed out\n");
5044 case IXGBE_ERR_EEPROM_VERSION:
5045 /* We are running on a pre-production device, log a warning */
5046 e_dev_warn("This device is a pre-production adapter/LOM. "
5047 "Please be aware there may be issues associated with "
5048 "your hardware. If you are experiencing problems "
5049 "please contact your Intel or hardware "
5050 "representative who provided you with this "
5054 e_dev_err("Hardware Error: %d\n", err);
5057 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5058 /* do not flush user set addresses */
5059 memcpy(old_addr, &adapter->mac_table[0].addr, netdev->addr_len);
5060 ixgbe_flush_sw_mac_table(adapter);
5061 ixgbe_mac_set_default_filter(adapter, old_addr);
5063 /* update SAN MAC vmdq pool selection */
5064 if (hw->mac.san_mac_rar_index)
5065 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5067 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5068 ixgbe_ptp_reset(adapter);
5070 if (hw->phy.ops.set_phy_power) {
5071 if (!netif_running(adapter->netdev) && !adapter->wol)
5072 hw->phy.ops.set_phy_power(hw, false);
5074 hw->phy.ops.set_phy_power(hw, true);
5079 * ixgbe_clean_tx_ring - Free Tx Buffers
5080 * @tx_ring: ring to be cleaned
5082 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5084 struct ixgbe_tx_buffer *tx_buffer_info;
5088 /* ring already cleared, nothing to do */
5089 if (!tx_ring->tx_buffer_info)
5092 /* Free all the Tx ring sk_buffs */
5093 for (i = 0; i < tx_ring->count; i++) {
5094 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5095 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
5098 netdev_tx_reset_queue(txring_txq(tx_ring));
5100 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5101 memset(tx_ring->tx_buffer_info, 0, size);
5103 /* Zero out the descriptor ring */
5104 memset(tx_ring->desc, 0, tx_ring->size);
5106 tx_ring->next_to_use = 0;
5107 tx_ring->next_to_clean = 0;
5111 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
5112 * @adapter: board private structure
5114 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
5118 for (i = 0; i < adapter->num_rx_queues; i++)
5119 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
5123 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
5124 * @adapter: board private structure
5126 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
5130 for (i = 0; i < adapter->num_tx_queues; i++)
5131 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
5134 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
5136 struct hlist_node *node2;
5137 struct ixgbe_fdir_filter *filter;
5139 spin_lock(&adapter->fdir_perfect_lock);
5141 hlist_for_each_entry_safe(filter, node2,
5142 &adapter->fdir_filter_list, fdir_node) {
5143 hlist_del(&filter->fdir_node);
5146 adapter->fdir_filter_count = 0;
5148 spin_unlock(&adapter->fdir_perfect_lock);
5151 void ixgbe_down(struct ixgbe_adapter *adapter)
5153 struct net_device *netdev = adapter->netdev;
5154 struct ixgbe_hw *hw = &adapter->hw;
5155 struct net_device *upper;
5156 struct list_head *iter;
5159 /* signal that we are down to the interrupt handler */
5160 if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
5161 return; /* do nothing if already down */
5163 /* disable receives */
5164 hw->mac.ops.disable_rx(hw);
5166 /* disable all enabled rx queues */
5167 for (i = 0; i < adapter->num_rx_queues; i++)
5168 /* this call also flushes the previous write */
5169 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
5171 usleep_range(10000, 20000);
5173 netif_tx_stop_all_queues(netdev);
5175 /* call carrier off first to avoid false dev_watchdog timeouts */
5176 netif_carrier_off(netdev);
5177 netif_tx_disable(netdev);
5179 /* disable any upper devices */
5180 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
5181 if (netif_is_macvlan(upper)) {
5182 struct macvlan_dev *vlan = netdev_priv(upper);
5184 if (vlan->fwd_priv) {
5185 netif_tx_stop_all_queues(upper);
5186 netif_carrier_off(upper);
5187 netif_tx_disable(upper);
5192 ixgbe_irq_disable(adapter);
5194 ixgbe_napi_disable_all(adapter);
5196 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
5197 IXGBE_FLAG2_RESET_REQUESTED);
5198 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5200 del_timer_sync(&adapter->service_timer);
5202 if (adapter->num_vfs) {
5203 /* Clear EITR Select mapping */
5204 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
5206 /* Mark all the VFs as inactive */
5207 for (i = 0 ; i < adapter->num_vfs; i++)
5208 adapter->vfinfo[i].clear_to_send = false;
5210 /* ping all the active vfs to let them know we are going down */
5211 ixgbe_ping_all_vfs(adapter);
5213 /* Disable all VFTE/VFRE TX/RX */
5214 ixgbe_disable_tx_rx(adapter);
5217 /* disable transmits in the hardware now that interrupts are off */
5218 for (i = 0; i < adapter->num_tx_queues; i++) {
5219 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
5220 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5223 /* Disable the Tx DMA engine on 82599 and later MAC */
5224 switch (hw->mac.type) {
5225 case ixgbe_mac_82599EB:
5226 case ixgbe_mac_X540:
5227 case ixgbe_mac_X550:
5228 case ixgbe_mac_X550EM_x:
5229 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5230 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5231 ~IXGBE_DMATXCTL_TE));
5237 if (!pci_channel_offline(adapter->pdev))
5238 ixgbe_reset(adapter);
5240 /* power down the optics for 82599 SFP+ fiber */
5241 if (hw->mac.ops.disable_tx_laser)
5242 hw->mac.ops.disable_tx_laser(hw);
5244 ixgbe_clean_all_tx_rings(adapter);
5245 ixgbe_clean_all_rx_rings(adapter);
5247 #ifdef CONFIG_IXGBE_DCA
5248 /* since we reset the hardware DCA settings were cleared */
5249 ixgbe_setup_dca(adapter);
5254 * ixgbe_tx_timeout - Respond to a Tx Hang
5255 * @netdev: network interface device structure
5257 static void ixgbe_tx_timeout(struct net_device *netdev)
5259 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5261 /* Do the reset outside of interrupt context */
5262 ixgbe_tx_timeout_reset(adapter);
5266 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5267 * @adapter: board private structure to initialize
5269 * ixgbe_sw_init initializes the Adapter private data structure.
5270 * Fields are initialized based on PCI device information and
5271 * OS network device settings (MTU size).
5273 static int ixgbe_sw_init(struct ixgbe_adapter *adapter)
5275 struct ixgbe_hw *hw = &adapter->hw;
5276 struct pci_dev *pdev = adapter->pdev;
5277 unsigned int rss, fdir;
5279 #ifdef CONFIG_IXGBE_DCB
5281 struct tc_configuration *tc;
5284 /* PCI config space info */
5286 hw->vendor_id = pdev->vendor;
5287 hw->device_id = pdev->device;
5288 hw->revision_id = pdev->revision;
5289 hw->subsystem_vendor_id = pdev->subsystem_vendor;
5290 hw->subsystem_device_id = pdev->subsystem_device;
5292 /* Set common capability flags and settings */
5293 rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
5294 adapter->ring_feature[RING_F_RSS].limit = rss;
5295 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5296 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5297 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
5298 adapter->atr_sample_rate = 20;
5299 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
5300 adapter->ring_feature[RING_F_FDIR].limit = fdir;
5301 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
5302 #ifdef CONFIG_IXGBE_DCA
5303 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
5306 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5307 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5308 #ifdef CONFIG_IXGBE_DCB
5309 /* Default traffic class to use for FCoE */
5310 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5311 #endif /* CONFIG_IXGBE_DCB */
5312 #endif /* IXGBE_FCOE */
5314 adapter->mac_table = kzalloc(sizeof(struct ixgbe_mac_addr) *
5315 hw->mac.num_rar_entries,
5318 /* Set MAC specific capability flags and exceptions */
5319 switch (hw->mac.type) {
5320 case ixgbe_mac_82598EB:
5321 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
5322 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
5324 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5325 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5327 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
5328 adapter->ring_feature[RING_F_FDIR].limit = 0;
5329 adapter->atr_sample_rate = 0;
5330 adapter->fdir_pballoc = 0;
5332 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5333 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5334 #ifdef CONFIG_IXGBE_DCB
5335 adapter->fcoe.up = 0;
5336 #endif /* IXGBE_DCB */
5337 #endif /* IXGBE_FCOE */
5339 case ixgbe_mac_82599EB:
5340 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5341 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5343 case ixgbe_mac_X540:
5344 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
5345 if (fwsm & IXGBE_FWSM_TS_ENABLED)
5346 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5348 case ixgbe_mac_X550EM_x:
5349 case ixgbe_mac_X550:
5350 #ifdef CONFIG_IXGBE_DCA
5351 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
5353 #ifdef CONFIG_IXGBE_VXLAN
5354 adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
5362 /* FCoE support exists, always init the FCoE lock */
5363 spin_lock_init(&adapter->fcoe.lock);
5366 /* n-tuple support exists, always init our spinlock */
5367 spin_lock_init(&adapter->fdir_perfect_lock);
5369 #ifdef CONFIG_IXGBE_DCB
5370 switch (hw->mac.type) {
5371 case ixgbe_mac_X540:
5372 case ixgbe_mac_X550:
5373 case ixgbe_mac_X550EM_x:
5374 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
5375 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
5378 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
5379 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
5383 /* Configure DCB traffic classes */
5384 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5385 tc = &adapter->dcb_cfg.tc_config[j];
5386 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5387 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5388 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5389 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5390 tc->dcb_pfc = pfc_disabled;
5393 /* Initialize default user to priority mapping, UPx->TC0 */
5394 tc = &adapter->dcb_cfg.tc_config[0];
5395 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5396 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5398 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5399 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5400 adapter->dcb_cfg.pfc_mode_enable = false;
5401 adapter->dcb_set_bitmap = 0x00;
5402 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5403 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
5404 sizeof(adapter->temp_dcb_cfg));
5408 /* default flow control settings */
5409 hw->fc.requested_mode = ixgbe_fc_full;
5410 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5411 ixgbe_pbthresh_setup(adapter);
5412 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5413 hw->fc.send_xon = true;
5414 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
5416 #ifdef CONFIG_PCI_IOV
5418 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
5420 /* assign number of SR-IOV VFs */
5421 if (hw->mac.type != ixgbe_mac_82598EB) {
5422 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
5423 adapter->num_vfs = 0;
5424 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
5426 adapter->num_vfs = max_vfs;
5429 #endif /* CONFIG_PCI_IOV */
5431 /* enable itr by default in dynamic mode */
5432 adapter->rx_itr_setting = 1;
5433 adapter->tx_itr_setting = 1;
5435 /* set default ring sizes */
5436 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5437 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5439 /* set default work limits */
5440 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5442 /* initialize eeprom parameters */
5443 if (ixgbe_init_eeprom_params_generic(hw)) {
5444 e_dev_err("EEPROM initialization failed\n");
5448 /* PF holds first pool slot */
5449 set_bit(0, &adapter->fwd_bitmask);
5450 set_bit(__IXGBE_DOWN, &adapter->state);
5456 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5457 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5459 * Return 0 on success, negative on failure
5461 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5463 struct device *dev = tx_ring->dev;
5464 int orig_node = dev_to_node(dev);
5468 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5470 if (tx_ring->q_vector)
5471 ring_node = tx_ring->q_vector->numa_node;
5473 tx_ring->tx_buffer_info = vzalloc_node(size, ring_node);
5474 if (!tx_ring->tx_buffer_info)
5475 tx_ring->tx_buffer_info = vzalloc(size);
5476 if (!tx_ring->tx_buffer_info)
5479 u64_stats_init(&tx_ring->syncp);
5481 /* round up to nearest 4K */
5482 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5483 tx_ring->size = ALIGN(tx_ring->size, 4096);
5485 set_dev_node(dev, ring_node);
5486 tx_ring->desc = dma_alloc_coherent(dev,
5490 set_dev_node(dev, orig_node);
5492 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5493 &tx_ring->dma, GFP_KERNEL);
5497 tx_ring->next_to_use = 0;
5498 tx_ring->next_to_clean = 0;
5502 vfree(tx_ring->tx_buffer_info);
5503 tx_ring->tx_buffer_info = NULL;
5504 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5509 * ixgbe_setup_all_tx_resources - allocate all queues Tx 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_tx_resources(struct ixgbe_adapter *adapter)
5522 for (i = 0; i < adapter->num_tx_queues; i++) {
5523 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5527 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5533 /* rewind the index freeing the rings as we go */
5535 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5540 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5541 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5543 * Returns 0 on success, negative on failure
5545 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5547 struct device *dev = rx_ring->dev;
5548 int orig_node = dev_to_node(dev);
5552 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5554 if (rx_ring->q_vector)
5555 ring_node = rx_ring->q_vector->numa_node;
5557 rx_ring->rx_buffer_info = vzalloc_node(size, ring_node);
5558 if (!rx_ring->rx_buffer_info)
5559 rx_ring->rx_buffer_info = vzalloc(size);
5560 if (!rx_ring->rx_buffer_info)
5563 u64_stats_init(&rx_ring->syncp);
5565 /* Round up to nearest 4K */
5566 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5567 rx_ring->size = ALIGN(rx_ring->size, 4096);
5569 set_dev_node(dev, ring_node);
5570 rx_ring->desc = dma_alloc_coherent(dev,
5574 set_dev_node(dev, orig_node);
5576 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5577 &rx_ring->dma, GFP_KERNEL);
5581 rx_ring->next_to_clean = 0;
5582 rx_ring->next_to_use = 0;
5586 vfree(rx_ring->rx_buffer_info);
5587 rx_ring->rx_buffer_info = NULL;
5588 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5593 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5594 * @adapter: board private structure
5596 * If this function returns with an error, then it's possible one or
5597 * more of the rings is populated (while the rest are not). It is the
5598 * callers duty to clean those orphaned rings.
5600 * Return 0 on success, negative on failure
5602 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5606 for (i = 0; i < adapter->num_rx_queues; i++) {
5607 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5611 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5616 err = ixgbe_setup_fcoe_ddp_resources(adapter);
5621 /* rewind the index freeing the rings as we go */
5623 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5628 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5629 * @tx_ring: Tx descriptor ring for a specific queue
5631 * Free all transmit software resources
5633 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5635 ixgbe_clean_tx_ring(tx_ring);
5637 vfree(tx_ring->tx_buffer_info);
5638 tx_ring->tx_buffer_info = NULL;
5640 /* if not set, then don't free */
5644 dma_free_coherent(tx_ring->dev, tx_ring->size,
5645 tx_ring->desc, tx_ring->dma);
5647 tx_ring->desc = NULL;
5651 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5652 * @adapter: board private structure
5654 * Free all transmit software resources
5656 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5660 for (i = 0; i < adapter->num_tx_queues; i++)
5661 if (adapter->tx_ring[i]->desc)
5662 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5666 * ixgbe_free_rx_resources - Free Rx Resources
5667 * @rx_ring: ring to clean the resources from
5669 * Free all receive software resources
5671 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5673 ixgbe_clean_rx_ring(rx_ring);
5675 vfree(rx_ring->rx_buffer_info);
5676 rx_ring->rx_buffer_info = NULL;
5678 /* if not set, then don't free */
5682 dma_free_coherent(rx_ring->dev, rx_ring->size,
5683 rx_ring->desc, rx_ring->dma);
5685 rx_ring->desc = NULL;
5689 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5690 * @adapter: board private structure
5692 * Free all receive software resources
5694 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5699 ixgbe_free_fcoe_ddp_resources(adapter);
5702 for (i = 0; i < adapter->num_rx_queues; i++)
5703 if (adapter->rx_ring[i]->desc)
5704 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5708 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5709 * @netdev: network interface device structure
5710 * @new_mtu: new value for maximum frame size
5712 * Returns 0 on success, negative on failure
5714 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5716 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5717 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5719 /* MTU < 68 is an error and causes problems on some kernels */
5720 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5724 * For 82599EB we cannot allow legacy VFs to enable their receive
5725 * paths when MTU greater than 1500 is configured. So display a
5726 * warning that legacy VFs will be disabled.
5728 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
5729 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
5730 (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
5731 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
5733 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5735 /* must set new MTU before calling down or up */
5736 netdev->mtu = new_mtu;
5738 if (netif_running(netdev))
5739 ixgbe_reinit_locked(adapter);
5745 * ixgbe_open - Called when a network interface is made active
5746 * @netdev: network interface device structure
5748 * Returns 0 on success, negative value on failure
5750 * The open entry point is called when a network interface is made
5751 * active by the system (IFF_UP). At this point all resources needed
5752 * for transmit and receive operations are allocated, the interrupt
5753 * handler is registered with the OS, the watchdog timer is started,
5754 * and the stack is notified that the interface is ready.
5756 static int ixgbe_open(struct net_device *netdev)
5758 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5759 struct ixgbe_hw *hw = &adapter->hw;
5762 /* disallow open during test */
5763 if (test_bit(__IXGBE_TESTING, &adapter->state))
5766 netif_carrier_off(netdev);
5768 /* allocate transmit descriptors */
5769 err = ixgbe_setup_all_tx_resources(adapter);
5773 /* allocate receive descriptors */
5774 err = ixgbe_setup_all_rx_resources(adapter);
5778 ixgbe_configure(adapter);
5780 err = ixgbe_request_irq(adapter);
5784 /* Notify the stack of the actual queue counts. */
5785 if (adapter->num_rx_pools > 1)
5786 queues = adapter->num_rx_queues_per_pool;
5788 queues = adapter->num_tx_queues;
5790 err = netif_set_real_num_tx_queues(netdev, queues);
5792 goto err_set_queues;
5794 if (adapter->num_rx_pools > 1 &&
5795 adapter->num_rx_queues > IXGBE_MAX_L2A_QUEUES)
5796 queues = IXGBE_MAX_L2A_QUEUES;
5798 queues = adapter->num_rx_queues;
5799 err = netif_set_real_num_rx_queues(netdev, queues);
5801 goto err_set_queues;
5803 ixgbe_ptp_init(adapter);
5805 ixgbe_up_complete(adapter);
5807 ixgbe_clear_vxlan_port(adapter);
5808 #ifdef CONFIG_IXGBE_VXLAN
5809 vxlan_get_rx_port(netdev);
5815 ixgbe_free_irq(adapter);
5817 ixgbe_free_all_rx_resources(adapter);
5818 if (hw->phy.ops.set_phy_power && !adapter->wol)
5819 hw->phy.ops.set_phy_power(&adapter->hw, false);
5821 ixgbe_free_all_tx_resources(adapter);
5823 ixgbe_reset(adapter);
5828 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
5830 ixgbe_ptp_suspend(adapter);
5832 if (adapter->hw.phy.ops.enter_lplu) {
5833 adapter->hw.phy.reset_disable = true;
5834 ixgbe_down(adapter);
5835 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
5836 adapter->hw.phy.reset_disable = false;
5838 ixgbe_down(adapter);
5841 ixgbe_free_irq(adapter);
5843 ixgbe_free_all_tx_resources(adapter);
5844 ixgbe_free_all_rx_resources(adapter);
5848 * ixgbe_close - Disables a network interface
5849 * @netdev: network interface device structure
5851 * Returns 0, this is not allowed to fail
5853 * The close entry point is called when an interface is de-activated
5854 * by the OS. The hardware is still under the drivers control, but
5855 * needs to be disabled. A global MAC reset is issued to stop the
5856 * hardware, and all transmit and receive resources are freed.
5858 static int ixgbe_close(struct net_device *netdev)
5860 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5862 ixgbe_ptp_stop(adapter);
5864 ixgbe_close_suspend(adapter);
5866 ixgbe_fdir_filter_exit(adapter);
5868 ixgbe_release_hw_control(adapter);
5874 static int ixgbe_resume(struct pci_dev *pdev)
5876 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5877 struct net_device *netdev = adapter->netdev;
5880 adapter->hw.hw_addr = adapter->io_addr;
5881 pci_set_power_state(pdev, PCI_D0);
5882 pci_restore_state(pdev);
5884 * pci_restore_state clears dev->state_saved so call
5885 * pci_save_state to restore it.
5887 pci_save_state(pdev);
5889 err = pci_enable_device_mem(pdev);
5891 e_dev_err("Cannot enable PCI device from suspend\n");
5894 smp_mb__before_atomic();
5895 clear_bit(__IXGBE_DISABLED, &adapter->state);
5896 pci_set_master(pdev);
5898 pci_wake_from_d3(pdev, false);
5900 ixgbe_reset(adapter);
5902 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5905 err = ixgbe_init_interrupt_scheme(adapter);
5906 if (!err && netif_running(netdev))
5907 err = ixgbe_open(netdev);
5914 netif_device_attach(netdev);
5918 #endif /* CONFIG_PM */
5920 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5922 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5923 struct net_device *netdev = adapter->netdev;
5924 struct ixgbe_hw *hw = &adapter->hw;
5926 u32 wufc = adapter->wol;
5931 netif_device_detach(netdev);
5934 if (netif_running(netdev))
5935 ixgbe_close_suspend(adapter);
5938 ixgbe_clear_interrupt_scheme(adapter);
5941 retval = pci_save_state(pdev);
5946 if (hw->mac.ops.stop_link_on_d3)
5947 hw->mac.ops.stop_link_on_d3(hw);
5950 ixgbe_set_rx_mode(netdev);
5952 /* enable the optics for 82599 SFP+ fiber as we can WoL */
5953 if (hw->mac.ops.enable_tx_laser)
5954 hw->mac.ops.enable_tx_laser(hw);
5956 /* turn on all-multi mode if wake on multicast is enabled */
5957 if (wufc & IXGBE_WUFC_MC) {
5958 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5959 fctrl |= IXGBE_FCTRL_MPE;
5960 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5963 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5964 ctrl |= IXGBE_CTRL_GIO_DIS;
5965 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5967 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5969 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5970 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5973 switch (hw->mac.type) {
5974 case ixgbe_mac_82598EB:
5975 pci_wake_from_d3(pdev, false);
5977 case ixgbe_mac_82599EB:
5978 case ixgbe_mac_X540:
5979 case ixgbe_mac_X550:
5980 case ixgbe_mac_X550EM_x:
5981 pci_wake_from_d3(pdev, !!wufc);
5987 *enable_wake = !!wufc;
5988 if (hw->phy.ops.set_phy_power && !*enable_wake)
5989 hw->phy.ops.set_phy_power(hw, false);
5991 ixgbe_release_hw_control(adapter);
5993 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
5994 pci_disable_device(pdev);
6000 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
6005 retval = __ixgbe_shutdown(pdev, &wake);
6010 pci_prepare_to_sleep(pdev);
6012 pci_wake_from_d3(pdev, false);
6013 pci_set_power_state(pdev, PCI_D3hot);
6018 #endif /* CONFIG_PM */
6020 static void ixgbe_shutdown(struct pci_dev *pdev)
6024 __ixgbe_shutdown(pdev, &wake);
6026 if (system_state == SYSTEM_POWER_OFF) {
6027 pci_wake_from_d3(pdev, wake);
6028 pci_set_power_state(pdev, PCI_D3hot);
6033 * ixgbe_update_stats - Update the board statistics counters.
6034 * @adapter: board private structure
6036 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
6038 struct net_device *netdev = adapter->netdev;
6039 struct ixgbe_hw *hw = &adapter->hw;
6040 struct ixgbe_hw_stats *hwstats = &adapter->stats;
6042 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
6043 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
6044 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
6045 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
6047 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6048 test_bit(__IXGBE_RESETTING, &adapter->state))
6051 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
6054 for (i = 0; i < adapter->num_rx_queues; i++) {
6055 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
6056 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
6058 adapter->rsc_total_count = rsc_count;
6059 adapter->rsc_total_flush = rsc_flush;
6062 for (i = 0; i < adapter->num_rx_queues; i++) {
6063 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
6064 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
6065 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
6066 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
6067 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
6068 bytes += rx_ring->stats.bytes;
6069 packets += rx_ring->stats.packets;
6071 adapter->non_eop_descs = non_eop_descs;
6072 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
6073 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
6074 adapter->hw_csum_rx_error = hw_csum_rx_error;
6075 netdev->stats.rx_bytes = bytes;
6076 netdev->stats.rx_packets = packets;
6080 /* gather some stats to the adapter struct that are per queue */
6081 for (i = 0; i < adapter->num_tx_queues; i++) {
6082 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6083 restart_queue += tx_ring->tx_stats.restart_queue;
6084 tx_busy += tx_ring->tx_stats.tx_busy;
6085 bytes += tx_ring->stats.bytes;
6086 packets += tx_ring->stats.packets;
6088 adapter->restart_queue = restart_queue;
6089 adapter->tx_busy = tx_busy;
6090 netdev->stats.tx_bytes = bytes;
6091 netdev->stats.tx_packets = packets;
6093 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
6095 /* 8 register reads */
6096 for (i = 0; i < 8; i++) {
6097 /* for packet buffers not used, the register should read 0 */
6098 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
6100 hwstats->mpc[i] += mpc;
6101 total_mpc += hwstats->mpc[i];
6102 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
6103 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
6104 switch (hw->mac.type) {
6105 case ixgbe_mac_82598EB:
6106 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
6107 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
6108 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
6109 hwstats->pxonrxc[i] +=
6110 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
6112 case ixgbe_mac_82599EB:
6113 case ixgbe_mac_X540:
6114 case ixgbe_mac_X550:
6115 case ixgbe_mac_X550EM_x:
6116 hwstats->pxonrxc[i] +=
6117 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
6124 /*16 register reads */
6125 for (i = 0; i < 16; i++) {
6126 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
6127 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
6128 if ((hw->mac.type == ixgbe_mac_82599EB) ||
6129 (hw->mac.type == ixgbe_mac_X540) ||
6130 (hw->mac.type == ixgbe_mac_X550) ||
6131 (hw->mac.type == ixgbe_mac_X550EM_x)) {
6132 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
6133 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
6134 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
6135 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
6139 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
6140 /* work around hardware counting issue */
6141 hwstats->gprc -= missed_rx;
6143 ixgbe_update_xoff_received(adapter);
6145 /* 82598 hardware only has a 32 bit counter in the high register */
6146 switch (hw->mac.type) {
6147 case ixgbe_mac_82598EB:
6148 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
6149 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
6150 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
6151 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
6153 case ixgbe_mac_X540:
6154 case ixgbe_mac_X550:
6155 case ixgbe_mac_X550EM_x:
6156 /* OS2BMC stats are X540 and later */
6157 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
6158 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
6159 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
6160 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
6161 case ixgbe_mac_82599EB:
6162 for (i = 0; i < 16; i++)
6163 adapter->hw_rx_no_dma_resources +=
6164 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
6165 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
6166 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
6167 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
6168 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
6169 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
6170 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
6171 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
6172 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
6173 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
6175 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
6176 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
6177 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
6178 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
6179 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
6180 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
6181 /* Add up per cpu counters for total ddp aloc fail */
6182 if (adapter->fcoe.ddp_pool) {
6183 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
6184 struct ixgbe_fcoe_ddp_pool *ddp_pool;
6186 u64 noddp = 0, noddp_ext_buff = 0;
6187 for_each_possible_cpu(cpu) {
6188 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
6189 noddp += ddp_pool->noddp;
6190 noddp_ext_buff += ddp_pool->noddp_ext_buff;
6192 hwstats->fcoe_noddp = noddp;
6193 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
6195 #endif /* IXGBE_FCOE */
6200 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
6201 hwstats->bprc += bprc;
6202 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
6203 if (hw->mac.type == ixgbe_mac_82598EB)
6204 hwstats->mprc -= bprc;
6205 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
6206 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
6207 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
6208 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
6209 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
6210 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
6211 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
6212 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
6213 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
6214 hwstats->lxontxc += lxon;
6215 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
6216 hwstats->lxofftxc += lxoff;
6217 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
6218 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
6220 * 82598 errata - tx of flow control packets is included in tx counters
6222 xon_off_tot = lxon + lxoff;
6223 hwstats->gptc -= xon_off_tot;
6224 hwstats->mptc -= xon_off_tot;
6225 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
6226 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
6227 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
6228 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
6229 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
6230 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
6231 hwstats->ptc64 -= xon_off_tot;
6232 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
6233 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
6234 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
6235 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
6236 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
6237 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
6239 /* Fill out the OS statistics structure */
6240 netdev->stats.multicast = hwstats->mprc;
6243 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
6244 netdev->stats.rx_dropped = 0;
6245 netdev->stats.rx_length_errors = hwstats->rlec;
6246 netdev->stats.rx_crc_errors = hwstats->crcerrs;
6247 netdev->stats.rx_missed_errors = total_mpc;
6251 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
6252 * @adapter: pointer to the device adapter structure
6254 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
6256 struct ixgbe_hw *hw = &adapter->hw;
6259 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
6262 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6264 /* if interface is down do nothing */
6265 if (test_bit(__IXGBE_DOWN, &adapter->state))
6268 /* do nothing if we are not using signature filters */
6269 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
6272 adapter->fdir_overflow++;
6274 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
6275 for (i = 0; i < adapter->num_tx_queues; i++)
6276 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
6277 &(adapter->tx_ring[i]->state));
6278 /* re-enable flow director interrupts */
6279 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
6281 e_err(probe, "failed to finish FDIR re-initialization, "
6282 "ignored adding FDIR ATR filters\n");
6287 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
6288 * @adapter: pointer to the device adapter structure
6290 * This function serves two purposes. First it strobes the interrupt lines
6291 * in order to make certain interrupts are occurring. Secondly it sets the
6292 * bits needed to check for TX hangs. As a result we should immediately
6293 * determine if a hang has occurred.
6295 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
6297 struct ixgbe_hw *hw = &adapter->hw;
6301 /* If we're down, removing or resetting, just bail */
6302 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6303 test_bit(__IXGBE_REMOVING, &adapter->state) ||
6304 test_bit(__IXGBE_RESETTING, &adapter->state))
6307 /* Force detection of hung controller */
6308 if (netif_carrier_ok(adapter->netdev)) {
6309 for (i = 0; i < adapter->num_tx_queues; i++)
6310 set_check_for_tx_hang(adapter->tx_ring[i]);
6313 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6315 * for legacy and MSI interrupts don't set any bits
6316 * that are enabled for EIAM, because this operation
6317 * would set *both* EIMS and EICS for any bit in EIAM
6319 IXGBE_WRITE_REG(hw, IXGBE_EICS,
6320 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
6322 /* get one bit for every active tx/rx interrupt vector */
6323 for (i = 0; i < adapter->num_q_vectors; i++) {
6324 struct ixgbe_q_vector *qv = adapter->q_vector[i];
6325 if (qv->rx.ring || qv->tx.ring)
6326 eics |= ((u64)1 << i);
6330 /* Cause software interrupt to ensure rings are cleaned */
6331 ixgbe_irq_rearm_queues(adapter, eics);
6335 * ixgbe_watchdog_update_link - update the link status
6336 * @adapter: pointer to the device adapter structure
6337 * @link_speed: pointer to a u32 to store the link_speed
6339 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
6341 struct ixgbe_hw *hw = &adapter->hw;
6342 u32 link_speed = adapter->link_speed;
6343 bool link_up = adapter->link_up;
6344 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
6346 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6349 if (hw->mac.ops.check_link) {
6350 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6352 /* always assume link is up, if no check link function */
6353 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
6357 if (adapter->ixgbe_ieee_pfc)
6358 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
6360 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
6361 hw->mac.ops.fc_enable(hw);
6362 ixgbe_set_rx_drop_en(adapter);
6366 time_after(jiffies, (adapter->link_check_timeout +
6367 IXGBE_TRY_LINK_TIMEOUT))) {
6368 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6369 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6370 IXGBE_WRITE_FLUSH(hw);
6373 adapter->link_up = link_up;
6374 adapter->link_speed = link_speed;
6377 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
6379 #ifdef CONFIG_IXGBE_DCB
6380 struct net_device *netdev = adapter->netdev;
6381 struct dcb_app app = {
6382 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
6387 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
6388 up = dcb_ieee_getapp_mask(netdev, &app);
6390 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
6395 * ixgbe_watchdog_link_is_up - update netif_carrier status and
6396 * print link up message
6397 * @adapter: pointer to the device adapter structure
6399 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
6401 struct net_device *netdev = adapter->netdev;
6402 struct ixgbe_hw *hw = &adapter->hw;
6403 struct net_device *upper;
6404 struct list_head *iter;
6405 u32 link_speed = adapter->link_speed;
6406 const char *speed_str;
6407 bool flow_rx, flow_tx;
6409 /* only continue if link was previously down */
6410 if (netif_carrier_ok(netdev))
6413 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
6415 switch (hw->mac.type) {
6416 case ixgbe_mac_82598EB: {
6417 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6418 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6419 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6420 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6423 case ixgbe_mac_X540:
6424 case ixgbe_mac_X550:
6425 case ixgbe_mac_X550EM_x:
6426 case ixgbe_mac_82599EB: {
6427 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6428 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6429 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6430 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6439 adapter->last_rx_ptp_check = jiffies;
6441 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6442 ixgbe_ptp_start_cyclecounter(adapter);
6444 switch (link_speed) {
6445 case IXGBE_LINK_SPEED_10GB_FULL:
6446 speed_str = "10 Gbps";
6448 case IXGBE_LINK_SPEED_2_5GB_FULL:
6449 speed_str = "2.5 Gbps";
6451 case IXGBE_LINK_SPEED_1GB_FULL:
6452 speed_str = "1 Gbps";
6454 case IXGBE_LINK_SPEED_100_FULL:
6455 speed_str = "100 Mbps";
6458 speed_str = "unknown speed";
6461 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
6462 ((flow_rx && flow_tx) ? "RX/TX" :
6464 (flow_tx ? "TX" : "None"))));
6466 netif_carrier_on(netdev);
6467 ixgbe_check_vf_rate_limit(adapter);
6469 /* enable transmits */
6470 netif_tx_wake_all_queues(adapter->netdev);
6472 /* enable any upper devices */
6474 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
6475 if (netif_is_macvlan(upper)) {
6476 struct macvlan_dev *vlan = netdev_priv(upper);
6479 netif_tx_wake_all_queues(upper);
6484 /* update the default user priority for VFs */
6485 ixgbe_update_default_up(adapter);
6487 /* ping all the active vfs to let them know link has changed */
6488 ixgbe_ping_all_vfs(adapter);
6492 * ixgbe_watchdog_link_is_down - update netif_carrier status and
6493 * print link down message
6494 * @adapter: pointer to the adapter structure
6496 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
6498 struct net_device *netdev = adapter->netdev;
6499 struct ixgbe_hw *hw = &adapter->hw;
6501 adapter->link_up = false;
6502 adapter->link_speed = 0;
6504 /* only continue if link was up previously */
6505 if (!netif_carrier_ok(netdev))
6508 /* poll for SFP+ cable when link is down */
6509 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
6510 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
6512 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6513 ixgbe_ptp_start_cyclecounter(adapter);
6515 e_info(drv, "NIC Link is Down\n");
6516 netif_carrier_off(netdev);
6518 /* ping all the active vfs to let them know link has changed */
6519 ixgbe_ping_all_vfs(adapter);
6522 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
6526 for (i = 0; i < adapter->num_tx_queues; i++) {
6527 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6529 if (tx_ring->next_to_use != tx_ring->next_to_clean)
6536 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
6538 struct ixgbe_hw *hw = &adapter->hw;
6539 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
6540 u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
6544 if (!adapter->num_vfs)
6547 /* resetting the PF is only needed for MAC before X550 */
6548 if (hw->mac.type >= ixgbe_mac_X550)
6551 for (i = 0; i < adapter->num_vfs; i++) {
6552 for (j = 0; j < q_per_pool; j++) {
6555 h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
6556 t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
6567 * ixgbe_watchdog_flush_tx - flush queues on link down
6568 * @adapter: pointer to the device adapter structure
6570 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
6572 if (!netif_carrier_ok(adapter->netdev)) {
6573 if (ixgbe_ring_tx_pending(adapter) ||
6574 ixgbe_vf_tx_pending(adapter)) {
6575 /* We've lost link, so the controller stops DMA,
6576 * but we've got queued Tx work that's never going
6577 * to get done, so reset controller to flush Tx.
6578 * (Do the reset outside of interrupt context).
6580 e_warn(drv, "initiating reset to clear Tx work after link loss\n");
6581 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
6586 #ifdef CONFIG_PCI_IOV
6587 static inline void ixgbe_issue_vf_flr(struct ixgbe_adapter *adapter,
6588 struct pci_dev *vfdev)
6590 if (!pci_wait_for_pending_transaction(vfdev))
6591 e_dev_warn("Issuing VFLR with pending transactions\n");
6593 e_dev_err("Issuing VFLR for VF %s\n", pci_name(vfdev));
6594 pcie_capability_set_word(vfdev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
6599 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6601 struct ixgbe_hw *hw = &adapter->hw;
6602 struct pci_dev *pdev = adapter->pdev;
6603 struct pci_dev *vfdev;
6606 unsigned short vf_id;
6608 if (!(netif_carrier_ok(adapter->netdev)))
6611 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6612 if (gpc) /* If incrementing then no need for the check below */
6614 /* Check to see if a bad DMA write target from an errant or
6615 * malicious VF has caused a PCIe error. If so then we can
6616 * issue a VFLR to the offending VF(s) and then resume without
6617 * requesting a full slot reset.
6623 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
6627 /* get the device ID for the VF */
6628 pci_read_config_word(pdev, pos + PCI_SRIOV_VF_DID, &vf_id);
6630 /* check status reg for all VFs owned by this PF */
6631 vfdev = pci_get_device(pdev->vendor, vf_id, NULL);
6633 if (vfdev->is_virtfn && (vfdev->physfn == pdev)) {
6636 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
6637 if (status_reg & PCI_STATUS_REC_MASTER_ABORT)
6639 ixgbe_issue_vf_flr(adapter, vfdev);
6642 vfdev = pci_get_device(pdev->vendor, vf_id, vfdev);
6646 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6650 /* Do not perform spoof check for 82598 or if not in IOV mode */
6651 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
6652 adapter->num_vfs == 0)
6655 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6658 * ssvpc register is cleared on read, if zero then no
6659 * spoofed packets in the last interval.
6664 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
6667 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
6672 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
6675 #endif /* CONFIG_PCI_IOV */
6679 * ixgbe_watchdog_subtask - check and bring link up
6680 * @adapter: pointer to the device adapter structure
6682 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6684 /* if interface is down, removing or resetting, do nothing */
6685 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6686 test_bit(__IXGBE_REMOVING, &adapter->state) ||
6687 test_bit(__IXGBE_RESETTING, &adapter->state))
6690 ixgbe_watchdog_update_link(adapter);
6692 if (adapter->link_up)
6693 ixgbe_watchdog_link_is_up(adapter);
6695 ixgbe_watchdog_link_is_down(adapter);
6697 ixgbe_check_for_bad_vf(adapter);
6698 ixgbe_spoof_check(adapter);
6699 ixgbe_update_stats(adapter);
6701 ixgbe_watchdog_flush_tx(adapter);
6705 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6706 * @adapter: the ixgbe adapter structure
6708 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6710 struct ixgbe_hw *hw = &adapter->hw;
6713 /* not searching for SFP so there is nothing to do here */
6714 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6715 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6718 if (adapter->sfp_poll_time &&
6719 time_after(adapter->sfp_poll_time, jiffies))
6720 return; /* If not yet time to poll for SFP */
6722 /* someone else is in init, wait until next service event */
6723 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6726 adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
6728 err = hw->phy.ops.identify_sfp(hw);
6729 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6732 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6733 /* If no cable is present, then we need to reset
6734 * the next time we find a good cable. */
6735 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6742 /* exit if reset not needed */
6743 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6746 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6749 * A module may be identified correctly, but the EEPROM may not have
6750 * support for that module. setup_sfp() will fail in that case, so
6751 * we should not allow that module to load.
6753 if (hw->mac.type == ixgbe_mac_82598EB)
6754 err = hw->phy.ops.reset(hw);
6756 err = hw->mac.ops.setup_sfp(hw);
6758 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6761 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6762 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6765 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6767 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6768 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6769 e_dev_err("failed to initialize because an unsupported "
6770 "SFP+ module type was detected.\n");
6771 e_dev_err("Reload the driver after installing a "
6772 "supported module.\n");
6773 unregister_netdev(adapter->netdev);
6778 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6779 * @adapter: the ixgbe adapter structure
6781 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6783 struct ixgbe_hw *hw = &adapter->hw;
6785 bool autoneg = false;
6787 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6790 /* someone else is in init, wait until next service event */
6791 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6794 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6796 speed = hw->phy.autoneg_advertised;
6797 if ((!speed) && (hw->mac.ops.get_link_capabilities)) {
6798 hw->mac.ops.get_link_capabilities(hw, &speed, &autoneg);
6800 /* setup the highest link when no autoneg */
6802 if (speed & IXGBE_LINK_SPEED_10GB_FULL)
6803 speed = IXGBE_LINK_SPEED_10GB_FULL;
6807 if (hw->mac.ops.setup_link)
6808 hw->mac.ops.setup_link(hw, speed, true);
6810 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6811 adapter->link_check_timeout = jiffies;
6812 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6816 * ixgbe_service_timer - Timer Call-back
6817 * @data: pointer to adapter cast into an unsigned long
6819 static void ixgbe_service_timer(unsigned long data)
6821 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6822 unsigned long next_event_offset;
6824 /* poll faster when waiting for link */
6825 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6826 next_event_offset = HZ / 10;
6828 next_event_offset = HZ * 2;
6830 /* Reset the timer */
6831 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6833 ixgbe_service_event_schedule(adapter);
6836 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
6838 struct ixgbe_hw *hw = &adapter->hw;
6841 if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
6844 adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
6846 if (!hw->phy.ops.handle_lasi)
6849 status = hw->phy.ops.handle_lasi(&adapter->hw);
6850 if (status != IXGBE_ERR_OVERTEMP)
6853 e_crit(drv, "%s\n", ixgbe_overheat_msg);
6856 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6858 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6861 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6863 /* If we're already down, removing or resetting, just bail */
6864 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6865 test_bit(__IXGBE_REMOVING, &adapter->state) ||
6866 test_bit(__IXGBE_RESETTING, &adapter->state))
6869 ixgbe_dump(adapter);
6870 netdev_err(adapter->netdev, "Reset adapter\n");
6871 adapter->tx_timeout_count++;
6874 ixgbe_reinit_locked(adapter);
6879 * ixgbe_service_task - manages and runs subtasks
6880 * @work: pointer to work_struct containing our data
6882 static void ixgbe_service_task(struct work_struct *work)
6884 struct ixgbe_adapter *adapter = container_of(work,
6885 struct ixgbe_adapter,
6887 if (ixgbe_removed(adapter->hw.hw_addr)) {
6888 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
6890 ixgbe_down(adapter);
6893 ixgbe_service_event_complete(adapter);
6896 #ifdef CONFIG_IXGBE_VXLAN
6897 if (adapter->flags2 & IXGBE_FLAG2_VXLAN_REREG_NEEDED) {
6898 adapter->flags2 &= ~IXGBE_FLAG2_VXLAN_REREG_NEEDED;
6899 vxlan_get_rx_port(adapter->netdev);
6901 #endif /* CONFIG_IXGBE_VXLAN */
6902 ixgbe_reset_subtask(adapter);
6903 ixgbe_phy_interrupt_subtask(adapter);
6904 ixgbe_sfp_detection_subtask(adapter);
6905 ixgbe_sfp_link_config_subtask(adapter);
6906 ixgbe_check_overtemp_subtask(adapter);
6907 ixgbe_watchdog_subtask(adapter);
6908 ixgbe_fdir_reinit_subtask(adapter);
6909 ixgbe_check_hang_subtask(adapter);
6911 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
6912 ixgbe_ptp_overflow_check(adapter);
6913 ixgbe_ptp_rx_hang(adapter);
6916 ixgbe_service_event_complete(adapter);
6919 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
6920 struct ixgbe_tx_buffer *first,
6923 struct sk_buff *skb = first->skb;
6924 u32 vlan_macip_lens, type_tucmd;
6925 u32 mss_l4len_idx, l4len;
6928 if (skb->ip_summed != CHECKSUM_PARTIAL)
6931 if (!skb_is_gso(skb))
6934 err = skb_cow_head(skb, 0);
6938 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6939 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6941 if (first->protocol == htons(ETH_P_IP)) {
6942 struct iphdr *iph = ip_hdr(skb);
6945 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6949 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6950 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6951 IXGBE_TX_FLAGS_CSUM |
6952 IXGBE_TX_FLAGS_IPV4;
6953 } else if (skb_is_gso_v6(skb)) {
6954 ipv6_hdr(skb)->payload_len = 0;
6955 tcp_hdr(skb)->check =
6956 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6957 &ipv6_hdr(skb)->daddr,
6959 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6960 IXGBE_TX_FLAGS_CSUM;
6963 /* compute header lengths */
6964 l4len = tcp_hdrlen(skb);
6965 *hdr_len = skb_transport_offset(skb) + l4len;
6967 /* update gso size and bytecount with header size */
6968 first->gso_segs = skb_shinfo(skb)->gso_segs;
6969 first->bytecount += (first->gso_segs - 1) * *hdr_len;
6971 /* mss_l4len_id: use 0 as index for TSO */
6972 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6973 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6975 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6976 vlan_macip_lens = skb_network_header_len(skb);
6977 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6978 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6980 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6986 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6987 struct ixgbe_tx_buffer *first)
6989 struct sk_buff *skb = first->skb;
6990 u32 vlan_macip_lens = 0;
6991 u32 mss_l4len_idx = 0;
6994 if (skb->ip_summed != CHECKSUM_PARTIAL) {
6995 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6996 !(first->tx_flags & IXGBE_TX_FLAGS_CC))
6998 vlan_macip_lens = skb_network_offset(skb) <<
6999 IXGBE_ADVTXD_MACLEN_SHIFT;
7004 struct ipv6hdr *ipv6;
7008 struct tcphdr *tcphdr;
7012 if (skb->encapsulation) {
7013 network_hdr.raw = skb_inner_network_header(skb);
7014 transport_hdr.raw = skb_inner_transport_header(skb);
7015 vlan_macip_lens = skb_inner_network_offset(skb) <<
7016 IXGBE_ADVTXD_MACLEN_SHIFT;
7018 network_hdr.raw = skb_network_header(skb);
7019 transport_hdr.raw = skb_transport_header(skb);
7020 vlan_macip_lens = skb_network_offset(skb) <<
7021 IXGBE_ADVTXD_MACLEN_SHIFT;
7024 /* use first 4 bits to determine IP version */
7025 switch (network_hdr.ipv4->version) {
7027 vlan_macip_lens |= transport_hdr.raw - network_hdr.raw;
7028 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7029 l4_hdr = network_hdr.ipv4->protocol;
7032 vlan_macip_lens |= transport_hdr.raw - network_hdr.raw;
7033 l4_hdr = network_hdr.ipv6->nexthdr;
7036 if (unlikely(net_ratelimit())) {
7037 dev_warn(tx_ring->dev,
7038 "partial checksum but version=%d\n",
7039 network_hdr.ipv4->version);
7045 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
7046 mss_l4len_idx = (transport_hdr.tcphdr->doff * 4) <<
7047 IXGBE_ADVTXD_L4LEN_SHIFT;
7050 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
7051 mss_l4len_idx = sizeof(struct sctphdr) <<
7052 IXGBE_ADVTXD_L4LEN_SHIFT;
7055 mss_l4len_idx = sizeof(struct udphdr) <<
7056 IXGBE_ADVTXD_L4LEN_SHIFT;
7059 if (unlikely(net_ratelimit())) {
7060 dev_warn(tx_ring->dev,
7061 "partial checksum but l4 proto=%x!\n",
7067 /* update TX checksum flag */
7068 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
7071 /* vlan_macip_lens: MACLEN, VLAN tag */
7072 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
7074 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
7075 type_tucmd, mss_l4len_idx);
7078 #define IXGBE_SET_FLAG(_input, _flag, _result) \
7079 ((_flag <= _result) ? \
7080 ((u32)(_input & _flag) * (_result / _flag)) : \
7081 ((u32)(_input & _flag) / (_flag / _result)))
7083 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
7085 /* set type for advanced descriptor with frame checksum insertion */
7086 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
7087 IXGBE_ADVTXD_DCMD_DEXT |
7088 IXGBE_ADVTXD_DCMD_IFCS;
7090 /* set HW vlan bit if vlan is present */
7091 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
7092 IXGBE_ADVTXD_DCMD_VLE);
7094 /* set segmentation enable bits for TSO/FSO */
7095 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
7096 IXGBE_ADVTXD_DCMD_TSE);
7098 /* set timestamp bit if present */
7099 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
7100 IXGBE_ADVTXD_MAC_TSTAMP);
7102 /* insert frame checksum */
7103 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
7108 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
7109 u32 tx_flags, unsigned int paylen)
7111 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
7113 /* enable L4 checksum for TSO and TX checksum offload */
7114 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7115 IXGBE_TX_FLAGS_CSUM,
7116 IXGBE_ADVTXD_POPTS_TXSM);
7118 /* enble IPv4 checksum for TSO */
7119 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7120 IXGBE_TX_FLAGS_IPV4,
7121 IXGBE_ADVTXD_POPTS_IXSM);
7124 * Check Context must be set if Tx switch is enabled, which it
7125 * always is for case where virtual functions are running
7127 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7131 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
7134 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7136 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
7138 /* Herbert's original patch had:
7139 * smp_mb__after_netif_stop_queue();
7140 * but since that doesn't exist yet, just open code it.
7144 /* We need to check again in a case another CPU has just
7145 * made room available.
7147 if (likely(ixgbe_desc_unused(tx_ring) < size))
7150 /* A reprieve! - use start_queue because it doesn't call schedule */
7151 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
7152 ++tx_ring->tx_stats.restart_queue;
7156 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7158 if (likely(ixgbe_desc_unused(tx_ring) >= size))
7161 return __ixgbe_maybe_stop_tx(tx_ring, size);
7164 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
7167 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
7168 struct ixgbe_tx_buffer *first,
7171 struct sk_buff *skb = first->skb;
7172 struct ixgbe_tx_buffer *tx_buffer;
7173 union ixgbe_adv_tx_desc *tx_desc;
7174 struct skb_frag_struct *frag;
7176 unsigned int data_len, size;
7177 u32 tx_flags = first->tx_flags;
7178 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
7179 u16 i = tx_ring->next_to_use;
7181 tx_desc = IXGBE_TX_DESC(tx_ring, i);
7183 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
7185 size = skb_headlen(skb);
7186 data_len = skb->data_len;
7189 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
7190 if (data_len < sizeof(struct fcoe_crc_eof)) {
7191 size -= sizeof(struct fcoe_crc_eof) - data_len;
7194 data_len -= sizeof(struct fcoe_crc_eof);
7199 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
7203 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
7204 if (dma_mapping_error(tx_ring->dev, dma))
7207 /* record length, and DMA address */
7208 dma_unmap_len_set(tx_buffer, len, size);
7209 dma_unmap_addr_set(tx_buffer, dma, dma);
7211 tx_desc->read.buffer_addr = cpu_to_le64(dma);
7213 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
7214 tx_desc->read.cmd_type_len =
7215 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
7219 if (i == tx_ring->count) {
7220 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
7223 tx_desc->read.olinfo_status = 0;
7225 dma += IXGBE_MAX_DATA_PER_TXD;
7226 size -= IXGBE_MAX_DATA_PER_TXD;
7228 tx_desc->read.buffer_addr = cpu_to_le64(dma);
7231 if (likely(!data_len))
7234 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
7238 if (i == tx_ring->count) {
7239 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
7242 tx_desc->read.olinfo_status = 0;
7245 size = min_t(unsigned int, data_len, skb_frag_size(frag));
7247 size = skb_frag_size(frag);
7251 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
7254 tx_buffer = &tx_ring->tx_buffer_info[i];
7257 /* write last descriptor with RS and EOP bits */
7258 cmd_type |= size | IXGBE_TXD_CMD;
7259 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
7261 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
7263 /* set the timestamp */
7264 first->time_stamp = jiffies;
7267 * Force memory writes to complete before letting h/w know there
7268 * are new descriptors to fetch. (Only applicable for weak-ordered
7269 * memory model archs, such as IA-64).
7271 * We also need this memory barrier to make certain all of the
7272 * status bits have been updated before next_to_watch is written.
7276 /* set next_to_watch value indicating a packet is present */
7277 first->next_to_watch = tx_desc;
7280 if (i == tx_ring->count)
7283 tx_ring->next_to_use = i;
7285 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
7287 if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
7288 writel(i, tx_ring->tail);
7290 /* we need this if more than one processor can write to our tail
7291 * at a time, it synchronizes IO on IA64/Altix systems
7298 dev_err(tx_ring->dev, "TX DMA map failed\n");
7300 /* clear dma mappings for failed tx_buffer_info map */
7302 tx_buffer = &tx_ring->tx_buffer_info[i];
7303 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
7304 if (tx_buffer == first)
7311 tx_ring->next_to_use = i;
7314 static void ixgbe_atr(struct ixgbe_ring *ring,
7315 struct ixgbe_tx_buffer *first)
7317 struct ixgbe_q_vector *q_vector = ring->q_vector;
7318 union ixgbe_atr_hash_dword input = { .dword = 0 };
7319 union ixgbe_atr_hash_dword common = { .dword = 0 };
7321 unsigned char *network;
7323 struct ipv6hdr *ipv6;
7326 struct sk_buff *skb;
7327 #ifdef CONFIG_IXGBE_VXLAN
7329 #endif /* CONFIG_IXGBE_VXLAN */
7332 /* if ring doesn't have a interrupt vector, cannot perform ATR */
7336 /* do nothing if sampling is disabled */
7337 if (!ring->atr_sample_rate)
7342 /* snag network header to get L4 type and address */
7344 hdr.network = skb_network_header(skb);
7345 if (skb->encapsulation) {
7346 #ifdef CONFIG_IXGBE_VXLAN
7347 struct ixgbe_adapter *adapter = q_vector->adapter;
7349 if (!adapter->vxlan_port)
7351 if (first->protocol != htons(ETH_P_IP) ||
7352 hdr.ipv4->version != IPVERSION ||
7353 hdr.ipv4->protocol != IPPROTO_UDP) {
7356 if (ntohs(udp_hdr(skb)->dest) != adapter->vxlan_port)
7359 hdr.network = skb_inner_network_header(skb);
7360 th = inner_tcp_hdr(skb);
7363 #endif /* CONFIG_IXGBE_VXLAN */
7365 /* Currently only IPv4/IPv6 with TCP is supported */
7366 if ((first->protocol != htons(ETH_P_IPV6) ||
7367 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
7368 (first->protocol != htons(ETH_P_IP) ||
7369 hdr.ipv4->protocol != IPPROTO_TCP))
7374 /* skip this packet since it is invalid or the socket is closing */
7378 /* sample on all syn packets or once every atr sample count */
7379 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
7382 /* reset sample count */
7383 ring->atr_count = 0;
7385 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
7388 * src and dst are inverted, think how the receiver sees them
7390 * The input is broken into two sections, a non-compressed section
7391 * containing vm_pool, vlan_id, and flow_type. The rest of the data
7392 * is XORed together and stored in the compressed dword.
7394 input.formatted.vlan_id = vlan_id;
7397 * since src port and flex bytes occupy the same word XOR them together
7398 * and write the value to source port portion of compressed dword
7400 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
7401 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
7403 common.port.src ^= th->dest ^ first->protocol;
7404 common.port.dst ^= th->source;
7406 if (first->protocol == htons(ETH_P_IP)) {
7407 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
7408 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
7410 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
7411 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
7412 hdr.ipv6->saddr.s6_addr32[1] ^
7413 hdr.ipv6->saddr.s6_addr32[2] ^
7414 hdr.ipv6->saddr.s6_addr32[3] ^
7415 hdr.ipv6->daddr.s6_addr32[0] ^
7416 hdr.ipv6->daddr.s6_addr32[1] ^
7417 hdr.ipv6->daddr.s6_addr32[2] ^
7418 hdr.ipv6->daddr.s6_addr32[3];
7421 #ifdef CONFIG_IXGBE_VXLAN
7423 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
7424 #endif /* CONFIG_IXGBE_VXLAN */
7426 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
7427 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
7428 input, common, ring->queue_index);
7431 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
7432 void *accel_priv, select_queue_fallback_t fallback)
7434 struct ixgbe_fwd_adapter *fwd_adapter = accel_priv;
7436 struct ixgbe_adapter *adapter;
7437 struct ixgbe_ring_feature *f;
7442 return skb->queue_mapping + fwd_adapter->tx_base_queue;
7447 * only execute the code below if protocol is FCoE
7448 * or FIP and we have FCoE enabled on the adapter
7450 switch (vlan_get_protocol(skb)) {
7451 case htons(ETH_P_FCOE):
7452 case htons(ETH_P_FIP):
7453 adapter = netdev_priv(dev);
7455 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7458 return fallback(dev, skb);
7461 f = &adapter->ring_feature[RING_F_FCOE];
7463 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
7466 while (txq >= f->indices)
7469 return txq + f->offset;
7471 return fallback(dev, skb);
7475 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
7476 struct ixgbe_adapter *adapter,
7477 struct ixgbe_ring *tx_ring)
7479 struct ixgbe_tx_buffer *first;
7483 u16 count = TXD_USE_COUNT(skb_headlen(skb));
7484 __be16 protocol = skb->protocol;
7488 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
7489 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
7490 * + 2 desc gap to keep tail from touching head,
7491 * + 1 desc for context descriptor,
7492 * otherwise try next time
7494 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
7495 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
7497 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
7498 tx_ring->tx_stats.tx_busy++;
7499 return NETDEV_TX_BUSY;
7502 /* record the location of the first descriptor for this packet */
7503 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
7505 first->bytecount = skb->len;
7506 first->gso_segs = 1;
7508 /* if we have a HW VLAN tag being added default to the HW one */
7509 if (skb_vlan_tag_present(skb)) {
7510 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
7511 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7512 /* else if it is a SW VLAN check the next protocol and store the tag */
7513 } else if (protocol == htons(ETH_P_8021Q)) {
7514 struct vlan_hdr *vhdr, _vhdr;
7515 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
7519 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
7520 IXGBE_TX_FLAGS_VLAN_SHIFT;
7521 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
7523 protocol = vlan_get_protocol(skb);
7525 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
7526 adapter->ptp_clock &&
7527 !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
7529 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
7530 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
7532 /* schedule check for Tx timestamp */
7533 adapter->ptp_tx_skb = skb_get(skb);
7534 adapter->ptp_tx_start = jiffies;
7535 schedule_work(&adapter->ptp_tx_work);
7538 skb_tx_timestamp(skb);
7540 #ifdef CONFIG_PCI_IOV
7542 * Use the l2switch_enable flag - would be false if the DMA
7543 * Tx switch had been disabled.
7545 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7546 tx_flags |= IXGBE_TX_FLAGS_CC;
7549 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
7550 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7551 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
7552 (skb->priority != TC_PRIO_CONTROL))) {
7553 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
7554 tx_flags |= (skb->priority & 0x7) <<
7555 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
7556 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
7557 struct vlan_ethhdr *vhdr;
7559 if (skb_cow_head(skb, 0))
7561 vhdr = (struct vlan_ethhdr *)skb->data;
7562 vhdr->h_vlan_TCI = htons(tx_flags >>
7563 IXGBE_TX_FLAGS_VLAN_SHIFT);
7565 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7569 /* record initial flags and protocol */
7570 first->tx_flags = tx_flags;
7571 first->protocol = protocol;
7574 /* setup tx offload for FCoE */
7575 if ((protocol == htons(ETH_P_FCOE)) &&
7576 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
7577 tso = ixgbe_fso(tx_ring, first, &hdr_len);
7584 #endif /* IXGBE_FCOE */
7585 tso = ixgbe_tso(tx_ring, first, &hdr_len);
7589 ixgbe_tx_csum(tx_ring, first);
7591 /* add the ATR filter if ATR is on */
7592 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
7593 ixgbe_atr(tx_ring, first);
7597 #endif /* IXGBE_FCOE */
7598 ixgbe_tx_map(tx_ring, first, hdr_len);
7600 return NETDEV_TX_OK;
7603 dev_kfree_skb_any(first->skb);
7606 return NETDEV_TX_OK;
7609 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
7610 struct net_device *netdev,
7611 struct ixgbe_ring *ring)
7613 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7614 struct ixgbe_ring *tx_ring;
7617 * The minimum packet size for olinfo paylen is 17 so pad the skb
7618 * in order to meet this minimum size requirement.
7620 if (skb_put_padto(skb, 17))
7621 return NETDEV_TX_OK;
7623 tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
7625 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
7628 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
7629 struct net_device *netdev)
7631 return __ixgbe_xmit_frame(skb, netdev, NULL);
7635 * ixgbe_set_mac - Change the Ethernet Address of the NIC
7636 * @netdev: network interface device structure
7637 * @p: pointer to an address structure
7639 * Returns 0 on success, negative on failure
7641 static int ixgbe_set_mac(struct net_device *netdev, void *p)
7643 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7644 struct ixgbe_hw *hw = &adapter->hw;
7645 struct sockaddr *addr = p;
7648 if (!is_valid_ether_addr(addr->sa_data))
7649 return -EADDRNOTAVAIL;
7651 ixgbe_del_mac_filter(adapter, hw->mac.addr, VMDQ_P(0));
7652 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
7653 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
7655 ret = ixgbe_add_mac_filter(adapter, hw->mac.addr, VMDQ_P(0));
7656 return ret > 0 ? 0 : ret;
7660 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
7662 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7663 struct ixgbe_hw *hw = &adapter->hw;
7667 if (prtad != hw->phy.mdio.prtad)
7669 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
7675 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
7676 u16 addr, u16 value)
7678 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7679 struct ixgbe_hw *hw = &adapter->hw;
7681 if (prtad != hw->phy.mdio.prtad)
7683 return hw->phy.ops.write_reg(hw, addr, devad, value);
7686 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
7688 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7692 return ixgbe_ptp_set_ts_config(adapter, req);
7694 return ixgbe_ptp_get_ts_config(adapter, req);
7696 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
7701 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
7703 * @netdev: network interface device structure
7705 * Returns non-zero on failure
7707 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
7710 struct ixgbe_adapter *adapter = netdev_priv(dev);
7711 struct ixgbe_hw *hw = &adapter->hw;
7713 if (is_valid_ether_addr(hw->mac.san_addr)) {
7715 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
7718 /* update SAN MAC vmdq pool selection */
7719 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
7725 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
7727 * @netdev: network interface device structure
7729 * Returns non-zero on failure
7731 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
7734 struct ixgbe_adapter *adapter = netdev_priv(dev);
7735 struct ixgbe_mac_info *mac = &adapter->hw.mac;
7737 if (is_valid_ether_addr(mac->san_addr)) {
7739 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
7745 #ifdef CONFIG_NET_POLL_CONTROLLER
7747 * Polling 'interrupt' - used by things like netconsole to send skbs
7748 * without having to re-enable interrupts. It's not called while
7749 * the interrupt routine is executing.
7751 static void ixgbe_netpoll(struct net_device *netdev)
7753 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7756 /* if interface is down do nothing */
7757 if (test_bit(__IXGBE_DOWN, &adapter->state))
7760 /* loop through and schedule all active queues */
7761 for (i = 0; i < adapter->num_q_vectors; i++)
7762 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
7766 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7767 struct rtnl_link_stats64 *stats)
7769 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7773 for (i = 0; i < adapter->num_rx_queues; i++) {
7774 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7780 start = u64_stats_fetch_begin_irq(&ring->syncp);
7781 packets = ring->stats.packets;
7782 bytes = ring->stats.bytes;
7783 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
7784 stats->rx_packets += packets;
7785 stats->rx_bytes += bytes;
7789 for (i = 0; i < adapter->num_tx_queues; i++) {
7790 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7796 start = u64_stats_fetch_begin_irq(&ring->syncp);
7797 packets = ring->stats.packets;
7798 bytes = ring->stats.bytes;
7799 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
7800 stats->tx_packets += packets;
7801 stats->tx_bytes += bytes;
7805 /* following stats updated by ixgbe_watchdog_task() */
7806 stats->multicast = netdev->stats.multicast;
7807 stats->rx_errors = netdev->stats.rx_errors;
7808 stats->rx_length_errors = netdev->stats.rx_length_errors;
7809 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
7810 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
7814 #ifdef CONFIG_IXGBE_DCB
7816 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7817 * @adapter: pointer to ixgbe_adapter
7818 * @tc: number of traffic classes currently enabled
7820 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7821 * 802.1Q priority maps to a packet buffer that exists.
7823 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
7825 struct ixgbe_hw *hw = &adapter->hw;
7829 /* 82598 have a static priority to TC mapping that can not
7830 * be changed so no validation is needed.
7832 if (hw->mac.type == ixgbe_mac_82598EB)
7835 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
7838 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
7839 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
7841 /* If up2tc is out of bounds default to zero */
7843 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
7847 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
7853 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
7854 * @adapter: Pointer to adapter struct
7856 * Populate the netdev user priority to tc map
7858 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
7860 struct net_device *dev = adapter->netdev;
7861 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
7862 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
7865 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
7868 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
7869 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
7871 tc = ets->prio_tc[prio];
7873 netdev_set_prio_tc_map(dev, prio, tc);
7877 #endif /* CONFIG_IXGBE_DCB */
7879 * ixgbe_setup_tc - configure net_device for multiple traffic classes
7881 * @netdev: net device to configure
7882 * @tc: number of traffic classes to enable
7884 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7886 struct ixgbe_adapter *adapter = netdev_priv(dev);
7887 struct ixgbe_hw *hw = &adapter->hw;
7890 /* Hardware supports up to 8 traffic classes */
7891 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
7894 if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
7897 pools = (find_first_zero_bit(&adapter->fwd_bitmask, 32) > 1);
7898 if (tc && pools && adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS)
7901 /* Hardware has to reinitialize queues and interrupts to
7902 * match packet buffer alignment. Unfortunately, the
7903 * hardware is not flexible enough to do this dynamically.
7905 if (netif_running(dev))
7907 ixgbe_clear_interrupt_scheme(adapter);
7909 #ifdef CONFIG_IXGBE_DCB
7911 netdev_set_num_tc(dev, tc);
7912 ixgbe_set_prio_tc_map(adapter);
7914 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7916 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
7917 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
7918 adapter->hw.fc.requested_mode = ixgbe_fc_none;
7921 netdev_reset_tc(dev);
7923 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7924 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7926 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7928 adapter->temp_dcb_cfg.pfc_mode_enable = false;
7929 adapter->dcb_cfg.pfc_mode_enable = false;
7932 ixgbe_validate_rtr(adapter, tc);
7934 #endif /* CONFIG_IXGBE_DCB */
7935 ixgbe_init_interrupt_scheme(adapter);
7937 if (netif_running(dev))
7938 return ixgbe_open(dev);
7943 #ifdef CONFIG_PCI_IOV
7944 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
7946 struct net_device *netdev = adapter->netdev;
7949 ixgbe_setup_tc(netdev, netdev_get_num_tc(netdev));
7954 void ixgbe_do_reset(struct net_device *netdev)
7956 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7958 if (netif_running(netdev))
7959 ixgbe_reinit_locked(adapter);
7961 ixgbe_reset(adapter);
7964 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
7965 netdev_features_t features)
7967 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7969 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7970 if (!(features & NETIF_F_RXCSUM))
7971 features &= ~NETIF_F_LRO;
7973 /* Turn off LRO if not RSC capable */
7974 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
7975 features &= ~NETIF_F_LRO;
7980 static int ixgbe_set_features(struct net_device *netdev,
7981 netdev_features_t features)
7983 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7984 netdev_features_t changed = netdev->features ^ features;
7985 bool need_reset = false;
7987 /* Make sure RSC matches LRO, reset if change */
7988 if (!(features & NETIF_F_LRO)) {
7989 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7991 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
7992 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
7993 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7994 if (adapter->rx_itr_setting == 1 ||
7995 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
7996 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
7998 } else if ((changed ^ features) & NETIF_F_LRO) {
7999 e_info(probe, "rx-usecs set too low, "
8005 * Check if Flow Director n-tuple support was enabled or disabled. If
8006 * the state changed, we need to reset.
8008 switch (features & NETIF_F_NTUPLE) {
8009 case NETIF_F_NTUPLE:
8010 /* turn off ATR, enable perfect filters and reset */
8011 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
8014 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
8015 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
8018 /* turn off perfect filters, enable ATR and reset */
8019 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
8022 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
8024 /* We cannot enable ATR if SR-IOV is enabled */
8025 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8028 /* We cannot enable ATR if we have 2 or more traffic classes */
8029 if (netdev_get_num_tc(netdev) > 1)
8032 /* We cannot enable ATR if RSS is disabled */
8033 if (adapter->ring_feature[RING_F_RSS].limit <= 1)
8036 /* A sample rate of 0 indicates ATR disabled */
8037 if (!adapter->atr_sample_rate)
8040 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
8044 if (features & NETIF_F_HW_VLAN_CTAG_RX)
8045 ixgbe_vlan_strip_enable(adapter);
8047 ixgbe_vlan_strip_disable(adapter);
8049 if (changed & NETIF_F_RXALL)
8052 netdev->features = features;
8054 #ifdef CONFIG_IXGBE_VXLAN
8055 if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
8056 if (features & NETIF_F_RXCSUM)
8057 adapter->flags2 |= IXGBE_FLAG2_VXLAN_REREG_NEEDED;
8059 ixgbe_clear_vxlan_port(adapter);
8061 #endif /* CONFIG_IXGBE_VXLAN */
8064 ixgbe_do_reset(netdev);
8069 #ifdef CONFIG_IXGBE_VXLAN
8071 * ixgbe_add_vxlan_port - Get notifications about VXLAN ports that come up
8072 * @dev: The port's netdev
8073 * @sa_family: Socket Family that VXLAN is notifiying us about
8074 * @port: New UDP port number that VXLAN started listening to
8076 static void ixgbe_add_vxlan_port(struct net_device *dev, sa_family_t sa_family,
8079 struct ixgbe_adapter *adapter = netdev_priv(dev);
8080 struct ixgbe_hw *hw = &adapter->hw;
8081 u16 new_port = ntohs(port);
8083 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
8086 if (sa_family == AF_INET6)
8089 if (adapter->vxlan_port == new_port)
8092 if (adapter->vxlan_port) {
8094 "Hit Max num of VXLAN ports, not adding port %d\n",
8099 adapter->vxlan_port = new_port;
8100 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, new_port);
8104 * ixgbe_del_vxlan_port - Get notifications about VXLAN ports that go away
8105 * @dev: The port's netdev
8106 * @sa_family: Socket Family that VXLAN is notifying us about
8107 * @port: UDP port number that VXLAN stopped listening to
8109 static void ixgbe_del_vxlan_port(struct net_device *dev, sa_family_t sa_family,
8112 struct ixgbe_adapter *adapter = netdev_priv(dev);
8113 u16 new_port = ntohs(port);
8115 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
8118 if (sa_family == AF_INET6)
8121 if (adapter->vxlan_port != new_port) {
8122 netdev_info(dev, "Port %d was not found, not deleting\n",
8127 ixgbe_clear_vxlan_port(adapter);
8128 adapter->flags2 |= IXGBE_FLAG2_VXLAN_REREG_NEEDED;
8130 #endif /* CONFIG_IXGBE_VXLAN */
8132 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
8133 struct net_device *dev,
8134 const unsigned char *addr, u16 vid,
8137 /* guarantee we can provide a unique filter for the unicast address */
8138 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
8139 if (IXGBE_MAX_PF_MACVLANS <= netdev_uc_count(dev))
8143 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
8147 * ixgbe_configure_bridge_mode - set various bridge modes
8148 * @adapter - the private structure
8149 * @mode - requested bridge mode
8151 * Configure some settings require for various bridge modes.
8153 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
8156 struct ixgbe_hw *hw = &adapter->hw;
8157 unsigned int p, num_pools;
8161 case BRIDGE_MODE_VEPA:
8162 /* disable Tx loopback, rely on switch hairpin mode */
8163 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
8165 /* must enable Rx switching replication to allow multicast
8166 * packet reception on all VFs, and to enable source address
8169 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
8170 vmdctl |= IXGBE_VT_CTL_REPLEN;
8171 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
8173 /* enable Rx source address pruning. Note, this requires
8174 * replication to be enabled or else it does nothing.
8176 num_pools = adapter->num_vfs + adapter->num_rx_pools;
8177 for (p = 0; p < num_pools; p++) {
8178 if (hw->mac.ops.set_source_address_pruning)
8179 hw->mac.ops.set_source_address_pruning(hw,
8184 case BRIDGE_MODE_VEB:
8185 /* enable Tx loopback for internal VF/PF communication */
8186 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
8187 IXGBE_PFDTXGSWC_VT_LBEN);
8189 /* disable Rx switching replication unless we have SR-IOV
8192 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
8193 if (!adapter->num_vfs)
8194 vmdctl &= ~IXGBE_VT_CTL_REPLEN;
8195 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
8197 /* disable Rx source address pruning, since we don't expect to
8198 * be receiving external loopback of our transmitted frames.
8200 num_pools = adapter->num_vfs + adapter->num_rx_pools;
8201 for (p = 0; p < num_pools; p++) {
8202 if (hw->mac.ops.set_source_address_pruning)
8203 hw->mac.ops.set_source_address_pruning(hw,
8212 adapter->bridge_mode = mode;
8214 e_info(drv, "enabling bridge mode: %s\n",
8215 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
8220 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
8221 struct nlmsghdr *nlh, u16 flags)
8223 struct ixgbe_adapter *adapter = netdev_priv(dev);
8224 struct nlattr *attr, *br_spec;
8227 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
8230 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
8234 nla_for_each_nested(attr, br_spec, rem) {
8238 if (nla_type(attr) != IFLA_BRIDGE_MODE)
8241 if (nla_len(attr) < sizeof(mode))
8244 mode = nla_get_u16(attr);
8245 status = ixgbe_configure_bridge_mode(adapter, mode);
8255 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
8256 struct net_device *dev,
8257 u32 filter_mask, int nlflags)
8259 struct ixgbe_adapter *adapter = netdev_priv(dev);
8261 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
8264 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
8265 adapter->bridge_mode, 0, 0, nlflags,
8269 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
8271 struct ixgbe_fwd_adapter *fwd_adapter = NULL;
8272 struct ixgbe_adapter *adapter = netdev_priv(pdev);
8273 int used_pools = adapter->num_vfs + adapter->num_rx_pools;
8277 /* Hardware has a limited number of available pools. Each VF, and the
8278 * PF require a pool. Check to ensure we don't attempt to use more
8279 * then the available number of pools.
8281 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
8282 return ERR_PTR(-EINVAL);
8285 if (vdev->num_rx_queues != vdev->num_tx_queues) {
8286 netdev_info(pdev, "%s: Only supports a single queue count for TX and RX\n",
8288 return ERR_PTR(-EINVAL);
8291 /* Check for hardware restriction on number of rx/tx queues */
8292 if (vdev->num_tx_queues > IXGBE_MAX_L2A_QUEUES ||
8293 vdev->num_tx_queues == IXGBE_BAD_L2A_QUEUE) {
8295 "%s: Supports RX/TX Queue counts 1,2, and 4\n",
8297 return ERR_PTR(-EINVAL);
8300 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8301 adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS - 1) ||
8302 (adapter->num_rx_pools > IXGBE_MAX_MACVLANS))
8303 return ERR_PTR(-EBUSY);
8305 fwd_adapter = kzalloc(sizeof(*fwd_adapter), GFP_KERNEL);
8307 return ERR_PTR(-ENOMEM);
8309 pool = find_first_zero_bit(&adapter->fwd_bitmask, 32);
8310 adapter->num_rx_pools++;
8311 set_bit(pool, &adapter->fwd_bitmask);
8312 limit = find_last_bit(&adapter->fwd_bitmask, 32);
8314 /* Enable VMDq flag so device will be set in VM mode */
8315 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED | IXGBE_FLAG_SRIOV_ENABLED;
8316 adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
8317 adapter->ring_feature[RING_F_RSS].limit = vdev->num_tx_queues;
8319 /* Force reinit of ring allocation with VMDQ enabled */
8320 err = ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
8323 fwd_adapter->pool = pool;
8324 fwd_adapter->real_adapter = adapter;
8325 err = ixgbe_fwd_ring_up(vdev, fwd_adapter);
8328 netif_tx_start_all_queues(vdev);
8331 /* unwind counter and free adapter struct */
8333 "%s: dfwd hardware acceleration failed\n", vdev->name);
8334 clear_bit(pool, &adapter->fwd_bitmask);
8335 adapter->num_rx_pools--;
8337 return ERR_PTR(err);
8340 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
8342 struct ixgbe_fwd_adapter *fwd_adapter = priv;
8343 struct ixgbe_adapter *adapter = fwd_adapter->real_adapter;
8346 clear_bit(fwd_adapter->pool, &adapter->fwd_bitmask);
8347 adapter->num_rx_pools--;
8349 limit = find_last_bit(&adapter->fwd_bitmask, 32);
8350 adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
8351 ixgbe_fwd_ring_down(fwd_adapter->netdev, fwd_adapter);
8352 ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
8353 netdev_dbg(pdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
8354 fwd_adapter->pool, adapter->num_rx_pools,
8355 fwd_adapter->rx_base_queue,
8356 fwd_adapter->rx_base_queue + adapter->num_rx_queues_per_pool,
8357 adapter->fwd_bitmask);
8361 #define IXGBE_MAX_TUNNEL_HDR_LEN 80
8362 static netdev_features_t
8363 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
8364 netdev_features_t features)
8366 if (!skb->encapsulation)
8369 if (unlikely(skb_inner_mac_header(skb) - skb_transport_header(skb) >
8370 IXGBE_MAX_TUNNEL_HDR_LEN))
8371 return features & ~NETIF_F_ALL_CSUM;
8376 static const struct net_device_ops ixgbe_netdev_ops = {
8377 .ndo_open = ixgbe_open,
8378 .ndo_stop = ixgbe_close,
8379 .ndo_start_xmit = ixgbe_xmit_frame,
8380 .ndo_select_queue = ixgbe_select_queue,
8381 .ndo_set_rx_mode = ixgbe_set_rx_mode,
8382 .ndo_validate_addr = eth_validate_addr,
8383 .ndo_set_mac_address = ixgbe_set_mac,
8384 .ndo_change_mtu = ixgbe_change_mtu,
8385 .ndo_tx_timeout = ixgbe_tx_timeout,
8386 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
8387 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
8388 .ndo_do_ioctl = ixgbe_ioctl,
8389 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
8390 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
8391 .ndo_set_vf_rate = ixgbe_ndo_set_vf_bw,
8392 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
8393 .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
8394 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
8395 .ndo_get_stats64 = ixgbe_get_stats64,
8396 #ifdef CONFIG_IXGBE_DCB
8397 .ndo_setup_tc = ixgbe_setup_tc,
8399 #ifdef CONFIG_NET_POLL_CONTROLLER
8400 .ndo_poll_controller = ixgbe_netpoll,
8402 #ifdef CONFIG_NET_RX_BUSY_POLL
8403 .ndo_busy_poll = ixgbe_low_latency_recv,
8406 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
8407 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
8408 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
8409 .ndo_fcoe_enable = ixgbe_fcoe_enable,
8410 .ndo_fcoe_disable = ixgbe_fcoe_disable,
8411 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
8412 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
8413 #endif /* IXGBE_FCOE */
8414 .ndo_set_features = ixgbe_set_features,
8415 .ndo_fix_features = ixgbe_fix_features,
8416 .ndo_fdb_add = ixgbe_ndo_fdb_add,
8417 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
8418 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
8419 .ndo_dfwd_add_station = ixgbe_fwd_add,
8420 .ndo_dfwd_del_station = ixgbe_fwd_del,
8421 #ifdef CONFIG_IXGBE_VXLAN
8422 .ndo_add_vxlan_port = ixgbe_add_vxlan_port,
8423 .ndo_del_vxlan_port = ixgbe_del_vxlan_port,
8424 #endif /* CONFIG_IXGBE_VXLAN */
8425 .ndo_features_check = ixgbe_features_check,
8429 * ixgbe_enumerate_functions - Get the number of ports this device has
8430 * @adapter: adapter structure
8432 * This function enumerates the phsyical functions co-located on a single slot,
8433 * in order to determine how many ports a device has. This is most useful in
8434 * determining the required GT/s of PCIe bandwidth necessary for optimal
8437 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
8439 struct pci_dev *entry, *pdev = adapter->pdev;
8442 /* Some cards can not use the generic count PCIe functions method,
8443 * because they are behind a parent switch, so we hardcode these with
8444 * the correct number of functions.
8446 if (ixgbe_pcie_from_parent(&adapter->hw))
8449 list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
8450 /* don't count virtual functions */
8451 if (entry->is_virtfn)
8454 /* When the devices on the bus don't all match our device ID,
8455 * we can't reliably determine the correct number of
8456 * functions. This can occur if a function has been direct
8457 * attached to a virtual machine using VT-d, for example. In
8458 * this case, simply return -1 to indicate this.
8460 if ((entry->vendor != pdev->vendor) ||
8461 (entry->device != pdev->device))
8471 * ixgbe_wol_supported - Check whether device supports WoL
8472 * @hw: hw specific details
8473 * @device_id: the device ID
8474 * @subdev_id: the subsystem device ID
8476 * This function is used by probe and ethtool to determine
8477 * which devices have WoL support
8480 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
8483 struct ixgbe_hw *hw = &adapter->hw;
8484 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
8485 int is_wol_supported = 0;
8487 switch (device_id) {
8488 case IXGBE_DEV_ID_82599_SFP:
8489 /* Only these subdevices could supports WOL */
8490 switch (subdevice_id) {
8491 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
8492 case IXGBE_SUBDEV_ID_82599_560FLR:
8493 /* only support first port */
8494 if (hw->bus.func != 0)
8496 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
8497 case IXGBE_SUBDEV_ID_82599_SFP:
8498 case IXGBE_SUBDEV_ID_82599_RNDC:
8499 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
8500 case IXGBE_SUBDEV_ID_82599_LOM_SFP:
8501 is_wol_supported = 1;
8505 case IXGBE_DEV_ID_82599EN_SFP:
8506 /* Only this subdevice supports WOL */
8507 switch (subdevice_id) {
8508 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
8509 is_wol_supported = 1;
8513 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
8514 /* All except this subdevice support WOL */
8515 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
8516 is_wol_supported = 1;
8518 case IXGBE_DEV_ID_82599_KX4:
8519 is_wol_supported = 1;
8521 case IXGBE_DEV_ID_X540T:
8522 case IXGBE_DEV_ID_X540T1:
8523 case IXGBE_DEV_ID_X550T:
8524 case IXGBE_DEV_ID_X550EM_X_KX4:
8525 case IXGBE_DEV_ID_X550EM_X_KR:
8526 case IXGBE_DEV_ID_X550EM_X_10G_T:
8527 /* check eeprom to see if enabled wol */
8528 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
8529 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
8530 (hw->bus.func == 0))) {
8531 is_wol_supported = 1;
8536 return is_wol_supported;
8540 * ixgbe_get_platform_mac_addr - Look up MAC address in Open Firmware / IDPROM
8541 * @adapter: Pointer to adapter struct
8543 static void ixgbe_get_platform_mac_addr(struct ixgbe_adapter *adapter)
8546 struct device_node *dp = pci_device_to_OF_node(adapter->pdev);
8547 struct ixgbe_hw *hw = &adapter->hw;
8548 const unsigned char *addr;
8550 addr = of_get_mac_address(dp);
8552 ether_addr_copy(hw->mac.perm_addr, addr);
8555 #endif /* CONFIG_OF */
8558 ether_addr_copy(hw->mac.perm_addr, idprom->id_ethaddr);
8559 #endif /* CONFIG_SPARC */
8563 * ixgbe_probe - Device Initialization Routine
8564 * @pdev: PCI device information struct
8565 * @ent: entry in ixgbe_pci_tbl
8567 * Returns 0 on success, negative on failure
8569 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
8570 * The OS initialization, configuring of the adapter private structure,
8571 * and a hardware reset occur.
8573 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
8575 struct net_device *netdev;
8576 struct ixgbe_adapter *adapter = NULL;
8577 struct ixgbe_hw *hw;
8578 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
8579 int i, err, pci_using_dac, expected_gts;
8580 unsigned int indices = MAX_TX_QUEUES;
8581 u8 part_str[IXGBE_PBANUM_LENGTH];
8582 bool disable_dev = false;
8588 /* Catch broken hardware that put the wrong VF device ID in
8589 * the PCIe SR-IOV capability.
8591 if (pdev->is_virtfn) {
8592 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
8593 pci_name(pdev), pdev->vendor, pdev->device);
8597 err = pci_enable_device_mem(pdev);
8601 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
8604 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
8607 "No usable DMA configuration, aborting\n");
8613 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
8614 IORESOURCE_MEM), ixgbe_driver_name);
8617 "pci_request_selected_regions failed 0x%x\n", err);
8621 pci_enable_pcie_error_reporting(pdev);
8623 pci_set_master(pdev);
8624 pci_save_state(pdev);
8626 if (ii->mac == ixgbe_mac_82598EB) {
8627 #ifdef CONFIG_IXGBE_DCB
8628 /* 8 TC w/ 4 queues per TC */
8629 indices = 4 * MAX_TRAFFIC_CLASS;
8631 indices = IXGBE_MAX_RSS_INDICES;
8635 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
8638 goto err_alloc_etherdev;
8641 SET_NETDEV_DEV(netdev, &pdev->dev);
8643 adapter = netdev_priv(netdev);
8645 adapter->netdev = netdev;
8646 adapter->pdev = pdev;
8649 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
8651 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
8652 pci_resource_len(pdev, 0));
8653 adapter->io_addr = hw->hw_addr;
8659 netdev->netdev_ops = &ixgbe_netdev_ops;
8660 ixgbe_set_ethtool_ops(netdev);
8661 netdev->watchdog_timeo = 5 * HZ;
8662 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
8665 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
8666 hw->mac.type = ii->mac;
8667 hw->mvals = ii->mvals;
8670 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
8671 eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
8672 if (ixgbe_removed(hw->hw_addr)) {
8676 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
8677 if (!(eec & (1 << 8)))
8678 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
8681 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
8682 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
8683 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
8684 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
8685 hw->phy.mdio.mmds = 0;
8686 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
8687 hw->phy.mdio.dev = netdev;
8688 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
8689 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
8691 ii->get_invariants(hw);
8693 /* setup the private structure */
8694 err = ixgbe_sw_init(adapter);
8698 /* Make it possible the adapter to be woken up via WOL */
8699 switch (adapter->hw.mac.type) {
8700 case ixgbe_mac_82599EB:
8701 case ixgbe_mac_X540:
8702 case ixgbe_mac_X550:
8703 case ixgbe_mac_X550EM_x:
8704 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
8711 * If there is a fan on this device and it has failed log the
8714 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
8715 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
8716 if (esdp & IXGBE_ESDP_SDP1)
8717 e_crit(probe, "Fan has stopped, replace the adapter\n");
8720 if (allow_unsupported_sfp)
8721 hw->allow_unsupported_sfp = allow_unsupported_sfp;
8723 /* reset_hw fills in the perm_addr as well */
8724 hw->phy.reset_if_overtemp = true;
8725 err = hw->mac.ops.reset_hw(hw);
8726 hw->phy.reset_if_overtemp = false;
8727 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
8729 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
8730 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
8731 e_dev_err("Reload the driver after installing a supported module.\n");
8734 e_dev_err("HW Init failed: %d\n", err);
8738 #ifdef CONFIG_PCI_IOV
8739 /* SR-IOV not supported on the 82598 */
8740 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
8743 ixgbe_init_mbx_params_pf(hw);
8744 memcpy(&hw->mbx.ops, ii->mbx_ops, sizeof(hw->mbx.ops));
8745 pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
8746 ixgbe_enable_sriov(adapter);
8750 netdev->features = NETIF_F_SG |
8753 NETIF_F_HW_VLAN_CTAG_TX |
8754 NETIF_F_HW_VLAN_CTAG_RX |
8760 netdev->hw_features = netdev->features | NETIF_F_HW_L2FW_DOFFLOAD;
8762 switch (adapter->hw.mac.type) {
8763 case ixgbe_mac_82599EB:
8764 case ixgbe_mac_X540:
8765 case ixgbe_mac_X550:
8766 case ixgbe_mac_X550EM_x:
8767 netdev->features |= NETIF_F_SCTP_CSUM;
8768 netdev->hw_features |= NETIF_F_SCTP_CSUM |
8775 netdev->hw_features |= NETIF_F_RXALL;
8776 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
8778 netdev->vlan_features |= NETIF_F_TSO;
8779 netdev->vlan_features |= NETIF_F_TSO6;
8780 netdev->vlan_features |= NETIF_F_IP_CSUM;
8781 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
8782 netdev->vlan_features |= NETIF_F_SG;
8784 netdev->hw_enc_features |= NETIF_F_SG | NETIF_F_IP_CSUM |
8787 netdev->priv_flags |= IFF_UNICAST_FLT;
8788 netdev->priv_flags |= IFF_SUPP_NOFCS;
8790 #ifdef CONFIG_IXGBE_VXLAN
8791 switch (adapter->hw.mac.type) {
8792 case ixgbe_mac_X550:
8793 case ixgbe_mac_X550EM_x:
8794 netdev->hw_enc_features |= NETIF_F_RXCSUM |
8801 #endif /* CONFIG_IXGBE_VXLAN */
8803 #ifdef CONFIG_IXGBE_DCB
8804 netdev->dcbnl_ops = &dcbnl_ops;
8808 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
8809 unsigned int fcoe_l;
8811 if (hw->mac.ops.get_device_caps) {
8812 hw->mac.ops.get_device_caps(hw, &device_caps);
8813 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
8814 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
8818 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
8819 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
8821 netdev->features |= NETIF_F_FSO |
8824 netdev->vlan_features |= NETIF_F_FSO |
8828 #endif /* IXGBE_FCOE */
8829 if (pci_using_dac) {
8830 netdev->features |= NETIF_F_HIGHDMA;
8831 netdev->vlan_features |= NETIF_F_HIGHDMA;
8834 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
8835 netdev->hw_features |= NETIF_F_LRO;
8836 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
8837 netdev->features |= NETIF_F_LRO;
8839 /* make sure the EEPROM is good */
8840 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
8841 e_dev_err("The EEPROM Checksum Is Not Valid\n");
8846 ixgbe_get_platform_mac_addr(adapter);
8848 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
8850 if (!is_valid_ether_addr(netdev->dev_addr)) {
8851 e_dev_err("invalid MAC address\n");
8856 ixgbe_mac_set_default_filter(adapter, hw->mac.perm_addr);
8858 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
8859 (unsigned long) adapter);
8861 if (ixgbe_removed(hw->hw_addr)) {
8865 INIT_WORK(&adapter->service_task, ixgbe_service_task);
8866 set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
8867 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
8869 err = ixgbe_init_interrupt_scheme(adapter);
8873 /* WOL not supported for all devices */
8875 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
8876 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
8877 pdev->subsystem_device);
8878 if (hw->wol_enabled)
8879 adapter->wol = IXGBE_WUFC_MAG;
8881 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
8883 /* save off EEPROM version number */
8884 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
8885 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
8887 /* pick up the PCI bus settings for reporting later */
8888 if (ixgbe_pcie_from_parent(hw))
8889 ixgbe_get_parent_bus_info(adapter);
8891 hw->mac.ops.get_bus_info(hw);
8893 /* calculate the expected PCIe bandwidth required for optimal
8894 * performance. Note that some older parts will never have enough
8895 * bandwidth due to being older generation PCIe parts. We clamp these
8896 * parts to ensure no warning is displayed if it can't be fixed.
8898 switch (hw->mac.type) {
8899 case ixgbe_mac_82598EB:
8900 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
8903 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
8907 /* don't check link if we failed to enumerate functions */
8908 if (expected_gts > 0)
8909 ixgbe_check_minimum_link(adapter, expected_gts);
8911 err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
8913 strlcpy(part_str, "Unknown", sizeof(part_str));
8914 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
8915 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
8916 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
8919 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
8920 hw->mac.type, hw->phy.type, part_str);
8922 e_dev_info("%pM\n", netdev->dev_addr);
8924 /* reset the hardware with the new settings */
8925 err = hw->mac.ops.start_hw(hw);
8926 if (err == IXGBE_ERR_EEPROM_VERSION) {
8927 /* We are running on a pre-production device, log a warning */
8928 e_dev_warn("This device is a pre-production adapter/LOM. "
8929 "Please be aware there may be issues associated "
8930 "with your hardware. If you are experiencing "
8931 "problems please contact your Intel or hardware "
8932 "representative who provided you with this "
8935 strcpy(netdev->name, "eth%d");
8936 err = register_netdev(netdev);
8940 pci_set_drvdata(pdev, adapter);
8942 /* power down the optics for 82599 SFP+ fiber */
8943 if (hw->mac.ops.disable_tx_laser)
8944 hw->mac.ops.disable_tx_laser(hw);
8946 /* carrier off reporting is important to ethtool even BEFORE open */
8947 netif_carrier_off(netdev);
8949 #ifdef CONFIG_IXGBE_DCA
8950 if (dca_add_requester(&pdev->dev) == 0) {
8951 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
8952 ixgbe_setup_dca(adapter);
8955 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
8956 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
8957 for (i = 0; i < adapter->num_vfs; i++)
8958 ixgbe_vf_configuration(pdev, (i | 0x10000000));
8961 /* firmware requires driver version to be 0xFFFFFFFF
8962 * since os does not support feature
8964 if (hw->mac.ops.set_fw_drv_ver)
8965 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
8968 /* add san mac addr to netdev */
8969 ixgbe_add_sanmac_netdev(netdev);
8971 e_dev_info("%s\n", ixgbe_default_device_descr);
8973 #ifdef CONFIG_IXGBE_HWMON
8974 if (ixgbe_sysfs_init(adapter))
8975 e_err(probe, "failed to allocate sysfs resources\n");
8976 #endif /* CONFIG_IXGBE_HWMON */
8978 ixgbe_dbg_adapter_init(adapter);
8980 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
8981 if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
8982 hw->mac.ops.setup_link(hw,
8983 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
8989 ixgbe_release_hw_control(adapter);
8990 ixgbe_clear_interrupt_scheme(adapter);
8992 ixgbe_disable_sriov(adapter);
8993 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
8994 iounmap(adapter->io_addr);
8995 kfree(adapter->mac_table);
8997 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
8998 free_netdev(netdev);
9000 pci_release_selected_regions(pdev,
9001 pci_select_bars(pdev, IORESOURCE_MEM));
9004 if (!adapter || disable_dev)
9005 pci_disable_device(pdev);
9010 * ixgbe_remove - Device Removal Routine
9011 * @pdev: PCI device information struct
9013 * ixgbe_remove is called by the PCI subsystem to alert the driver
9014 * that it should release a PCI device. The could be caused by a
9015 * Hot-Plug event, or because the driver is going to be removed from
9018 static void ixgbe_remove(struct pci_dev *pdev)
9020 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9021 struct net_device *netdev;
9024 /* if !adapter then we already cleaned up in probe */
9028 netdev = adapter->netdev;
9029 ixgbe_dbg_adapter_exit(adapter);
9031 set_bit(__IXGBE_REMOVING, &adapter->state);
9032 cancel_work_sync(&adapter->service_task);
9035 #ifdef CONFIG_IXGBE_DCA
9036 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
9037 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
9038 dca_remove_requester(&pdev->dev);
9039 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
9043 #ifdef CONFIG_IXGBE_HWMON
9044 ixgbe_sysfs_exit(adapter);
9045 #endif /* CONFIG_IXGBE_HWMON */
9047 /* remove the added san mac */
9048 ixgbe_del_sanmac_netdev(netdev);
9050 #ifdef CONFIG_PCI_IOV
9051 ixgbe_disable_sriov(adapter);
9053 if (netdev->reg_state == NETREG_REGISTERED)
9054 unregister_netdev(netdev);
9056 ixgbe_clear_interrupt_scheme(adapter);
9058 ixgbe_release_hw_control(adapter);
9061 kfree(adapter->ixgbe_ieee_pfc);
9062 kfree(adapter->ixgbe_ieee_ets);
9065 iounmap(adapter->io_addr);
9066 pci_release_selected_regions(pdev, pci_select_bars(pdev,
9069 e_dev_info("complete\n");
9071 kfree(adapter->mac_table);
9072 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
9073 free_netdev(netdev);
9075 pci_disable_pcie_error_reporting(pdev);
9078 pci_disable_device(pdev);
9082 * ixgbe_io_error_detected - called when PCI error is detected
9083 * @pdev: Pointer to PCI device
9084 * @state: The current pci connection state
9086 * This function is called after a PCI bus error affecting
9087 * this device has been detected.
9089 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
9090 pci_channel_state_t state)
9092 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9093 struct net_device *netdev = adapter->netdev;
9095 #ifdef CONFIG_PCI_IOV
9096 struct ixgbe_hw *hw = &adapter->hw;
9097 struct pci_dev *bdev, *vfdev;
9098 u32 dw0, dw1, dw2, dw3;
9100 u16 req_id, pf_func;
9102 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
9103 adapter->num_vfs == 0)
9104 goto skip_bad_vf_detection;
9106 bdev = pdev->bus->self;
9107 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
9108 bdev = bdev->bus->self;
9111 goto skip_bad_vf_detection;
9113 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
9115 goto skip_bad_vf_detection;
9117 dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
9118 dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
9119 dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
9120 dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
9121 if (ixgbe_removed(hw->hw_addr))
9122 goto skip_bad_vf_detection;
9125 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
9126 if (!(req_id & 0x0080))
9127 goto skip_bad_vf_detection;
9129 pf_func = req_id & 0x01;
9130 if ((pf_func & 1) == (pdev->devfn & 1)) {
9131 unsigned int device_id;
9133 vf = (req_id & 0x7F) >> 1;
9134 e_dev_err("VF %d has caused a PCIe error\n", vf);
9135 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
9136 "%8.8x\tdw3: %8.8x\n",
9137 dw0, dw1, dw2, dw3);
9138 switch (adapter->hw.mac.type) {
9139 case ixgbe_mac_82599EB:
9140 device_id = IXGBE_82599_VF_DEVICE_ID;
9142 case ixgbe_mac_X540:
9143 device_id = IXGBE_X540_VF_DEVICE_ID;
9145 case ixgbe_mac_X550:
9146 device_id = IXGBE_DEV_ID_X550_VF;
9148 case ixgbe_mac_X550EM_x:
9149 device_id = IXGBE_DEV_ID_X550EM_X_VF;
9156 /* Find the pci device of the offending VF */
9157 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
9159 if (vfdev->devfn == (req_id & 0xFF))
9161 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
9165 * There's a slim chance the VF could have been hot plugged,
9166 * so if it is no longer present we don't need to issue the
9167 * VFLR. Just clean up the AER in that case.
9170 ixgbe_issue_vf_flr(adapter, vfdev);
9171 /* Free device reference count */
9175 pci_cleanup_aer_uncorrect_error_status(pdev);
9179 * Even though the error may have occurred on the other port
9180 * we still need to increment the vf error reference count for
9181 * both ports because the I/O resume function will be called
9184 adapter->vferr_refcount++;
9186 return PCI_ERS_RESULT_RECOVERED;
9188 skip_bad_vf_detection:
9189 #endif /* CONFIG_PCI_IOV */
9190 if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
9191 return PCI_ERS_RESULT_DISCONNECT;
9194 netif_device_detach(netdev);
9196 if (state == pci_channel_io_perm_failure) {
9198 return PCI_ERS_RESULT_DISCONNECT;
9201 if (netif_running(netdev))
9202 ixgbe_down(adapter);
9204 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
9205 pci_disable_device(pdev);
9208 /* Request a slot reset. */
9209 return PCI_ERS_RESULT_NEED_RESET;
9213 * ixgbe_io_slot_reset - called after the pci bus has been reset.
9214 * @pdev: Pointer to PCI device
9216 * Restart the card from scratch, as if from a cold-boot.
9218 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
9220 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9221 pci_ers_result_t result;
9224 if (pci_enable_device_mem(pdev)) {
9225 e_err(probe, "Cannot re-enable PCI device after reset.\n");
9226 result = PCI_ERS_RESULT_DISCONNECT;
9228 smp_mb__before_atomic();
9229 clear_bit(__IXGBE_DISABLED, &adapter->state);
9230 adapter->hw.hw_addr = adapter->io_addr;
9231 pci_set_master(pdev);
9232 pci_restore_state(pdev);
9233 pci_save_state(pdev);
9235 pci_wake_from_d3(pdev, false);
9237 ixgbe_reset(adapter);
9238 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
9239 result = PCI_ERS_RESULT_RECOVERED;
9242 err = pci_cleanup_aer_uncorrect_error_status(pdev);
9244 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
9245 "failed 0x%0x\n", err);
9246 /* non-fatal, continue */
9253 * ixgbe_io_resume - called when traffic can start flowing again.
9254 * @pdev: Pointer to PCI device
9256 * This callback is called when the error recovery driver tells us that
9257 * its OK to resume normal operation.
9259 static void ixgbe_io_resume(struct pci_dev *pdev)
9261 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9262 struct net_device *netdev = adapter->netdev;
9264 #ifdef CONFIG_PCI_IOV
9265 if (adapter->vferr_refcount) {
9266 e_info(drv, "Resuming after VF err\n");
9267 adapter->vferr_refcount--;
9272 if (netif_running(netdev))
9275 netif_device_attach(netdev);
9278 static const struct pci_error_handlers ixgbe_err_handler = {
9279 .error_detected = ixgbe_io_error_detected,
9280 .slot_reset = ixgbe_io_slot_reset,
9281 .resume = ixgbe_io_resume,
9284 static struct pci_driver ixgbe_driver = {
9285 .name = ixgbe_driver_name,
9286 .id_table = ixgbe_pci_tbl,
9287 .probe = ixgbe_probe,
9288 .remove = ixgbe_remove,
9290 .suspend = ixgbe_suspend,
9291 .resume = ixgbe_resume,
9293 .shutdown = ixgbe_shutdown,
9294 .sriov_configure = ixgbe_pci_sriov_configure,
9295 .err_handler = &ixgbe_err_handler
9299 * ixgbe_init_module - Driver Registration Routine
9301 * ixgbe_init_module is the first routine called when the driver is
9302 * loaded. All it does is register with the PCI subsystem.
9304 static int __init ixgbe_init_module(void)
9307 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
9308 pr_info("%s\n", ixgbe_copyright);
9312 ret = pci_register_driver(&ixgbe_driver);
9318 #ifdef CONFIG_IXGBE_DCA
9319 dca_register_notify(&dca_notifier);
9325 module_init(ixgbe_init_module);
9328 * ixgbe_exit_module - Driver Exit Cleanup Routine
9330 * ixgbe_exit_module is called just before the driver is removed
9333 static void __exit ixgbe_exit_module(void)
9335 #ifdef CONFIG_IXGBE_DCA
9336 dca_unregister_notify(&dca_notifier);
9338 pci_unregister_driver(&ixgbe_driver);
9343 #ifdef CONFIG_IXGBE_DCA
9344 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
9349 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
9350 __ixgbe_notify_dca);
9352 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
9355 #endif /* CONFIG_IXGBE_DCA */
9357 module_exit(ixgbe_exit_module);