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 "4.2.1-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;
1251 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1252 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1254 switch (hw->mac.type) {
1255 case ixgbe_mac_82598EB:
1256 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1258 case ixgbe_mac_82599EB:
1259 case ixgbe_mac_X540:
1260 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1261 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1264 /* for unknown hardware do not write register */
1269 * We can enable relaxed ordering for reads, but not writes when
1270 * DCA is enabled. This is due to a known issue in some chipsets
1271 * which will cause the DCA tag to be cleared.
1273 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1274 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1275 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1277 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1280 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1281 struct ixgbe_ring *rx_ring,
1284 struct ixgbe_hw *hw = &adapter->hw;
1286 u8 reg_idx = rx_ring->reg_idx;
1288 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1289 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1291 switch (hw->mac.type) {
1292 case ixgbe_mac_82599EB:
1293 case ixgbe_mac_X540:
1294 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1301 * We can enable relaxed ordering for reads, but not writes when
1302 * DCA is enabled. This is due to a known issue in some chipsets
1303 * which will cause the DCA tag to be cleared.
1305 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1306 IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1307 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1309 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1312 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1314 struct ixgbe_adapter *adapter = q_vector->adapter;
1315 struct ixgbe_ring *ring;
1316 int cpu = get_cpu();
1318 if (q_vector->cpu == cpu)
1321 ixgbe_for_each_ring(ring, q_vector->tx)
1322 ixgbe_update_tx_dca(adapter, ring, cpu);
1324 ixgbe_for_each_ring(ring, q_vector->rx)
1325 ixgbe_update_rx_dca(adapter, ring, cpu);
1327 q_vector->cpu = cpu;
1332 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1336 /* always use CB2 mode, difference is masked in the CB driver */
1337 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1338 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1339 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1341 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1342 IXGBE_DCA_CTRL_DCA_DISABLE);
1344 for (i = 0; i < adapter->num_q_vectors; i++) {
1345 adapter->q_vector[i]->cpu = -1;
1346 ixgbe_update_dca(adapter->q_vector[i]);
1350 static int __ixgbe_notify_dca(struct device *dev, void *data)
1352 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1353 unsigned long event = *(unsigned long *)data;
1355 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1359 case DCA_PROVIDER_ADD:
1360 /* if we're already enabled, don't do it again */
1361 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1363 if (dca_add_requester(dev) == 0) {
1364 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1365 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1366 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1369 /* Fall Through since DCA is disabled. */
1370 case DCA_PROVIDER_REMOVE:
1371 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1372 dca_remove_requester(dev);
1373 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1374 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1375 IXGBE_DCA_CTRL_DCA_DISABLE);
1383 #endif /* CONFIG_IXGBE_DCA */
1385 #define IXGBE_RSS_L4_TYPES_MASK \
1386 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1387 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1388 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1389 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1391 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1392 union ixgbe_adv_rx_desc *rx_desc,
1393 struct sk_buff *skb)
1397 if (!(ring->netdev->features & NETIF_F_RXHASH))
1400 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1401 IXGBE_RXDADV_RSSTYPE_MASK;
1406 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1407 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1408 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1413 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1414 * @ring: structure containing ring specific data
1415 * @rx_desc: advanced rx descriptor
1417 * Returns : true if it is FCoE pkt
1419 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1420 union ixgbe_adv_rx_desc *rx_desc)
1422 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1424 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1425 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1426 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1427 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1430 #endif /* IXGBE_FCOE */
1432 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1433 * @ring: structure containing ring specific data
1434 * @rx_desc: current Rx descriptor being processed
1435 * @skb: skb currently being received and modified
1437 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1438 union ixgbe_adv_rx_desc *rx_desc,
1439 struct sk_buff *skb)
1441 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1442 __le16 hdr_info = rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1443 bool encap_pkt = false;
1445 skb_checksum_none_assert(skb);
1447 /* Rx csum disabled */
1448 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1451 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) &&
1452 (hdr_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_TUNNEL >> 16))) {
1454 skb->encapsulation = 1;
1457 /* if IP and error */
1458 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1459 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1460 ring->rx_stats.csum_err++;
1464 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1467 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1469 * 82599 errata, UDP frames with a 0 checksum can be marked as
1472 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1473 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1476 ring->rx_stats.csum_err++;
1480 /* It must be a TCP or UDP packet with a valid checksum */
1481 skb->ip_summed = CHECKSUM_UNNECESSARY;
1483 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1486 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1487 ring->rx_stats.csum_err++;
1490 /* If we checked the outer header let the stack know */
1491 skb->csum_level = 1;
1495 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1496 struct ixgbe_rx_buffer *bi)
1498 struct page *page = bi->page;
1501 /* since we are recycling buffers we should seldom need to alloc */
1505 /* alloc new page for storage */
1506 page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1507 if (unlikely(!page)) {
1508 rx_ring->rx_stats.alloc_rx_page_failed++;
1512 /* map page for use */
1513 dma = dma_map_page(rx_ring->dev, page, 0,
1514 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1517 * if mapping failed free memory back to system since
1518 * there isn't much point in holding memory we can't use
1520 if (dma_mapping_error(rx_ring->dev, dma)) {
1521 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1523 rx_ring->rx_stats.alloc_rx_page_failed++;
1529 bi->page_offset = 0;
1535 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1536 * @rx_ring: ring to place buffers on
1537 * @cleaned_count: number of buffers to replace
1539 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1541 union ixgbe_adv_rx_desc *rx_desc;
1542 struct ixgbe_rx_buffer *bi;
1543 u16 i = rx_ring->next_to_use;
1549 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1550 bi = &rx_ring->rx_buffer_info[i];
1551 i -= rx_ring->count;
1554 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1558 * Refresh the desc even if buffer_addrs didn't change
1559 * because each write-back erases this info.
1561 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1567 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1568 bi = rx_ring->rx_buffer_info;
1569 i -= rx_ring->count;
1572 /* clear the status bits for the next_to_use descriptor */
1573 rx_desc->wb.upper.status_error = 0;
1576 } while (cleaned_count);
1578 i += rx_ring->count;
1580 if (rx_ring->next_to_use != i) {
1581 rx_ring->next_to_use = i;
1583 /* update next to alloc since we have filled the ring */
1584 rx_ring->next_to_alloc = i;
1586 /* Force memory writes to complete before letting h/w
1587 * know there are new descriptors to fetch. (Only
1588 * applicable for weak-ordered memory model archs,
1592 writel(i, rx_ring->tail);
1596 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1597 struct sk_buff *skb)
1599 u16 hdr_len = skb_headlen(skb);
1601 /* set gso_size to avoid messing up TCP MSS */
1602 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1603 IXGBE_CB(skb)->append_cnt);
1604 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1607 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1608 struct sk_buff *skb)
1610 /* if append_cnt is 0 then frame is not RSC */
1611 if (!IXGBE_CB(skb)->append_cnt)
1614 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1615 rx_ring->rx_stats.rsc_flush++;
1617 ixgbe_set_rsc_gso_size(rx_ring, skb);
1619 /* gso_size is computed using append_cnt so always clear it last */
1620 IXGBE_CB(skb)->append_cnt = 0;
1624 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1625 * @rx_ring: rx descriptor ring packet is being transacted on
1626 * @rx_desc: pointer to the EOP Rx descriptor
1627 * @skb: pointer to current skb being populated
1629 * This function checks the ring, descriptor, and packet information in
1630 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1631 * other fields within the skb.
1633 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1634 union ixgbe_adv_rx_desc *rx_desc,
1635 struct sk_buff *skb)
1637 struct net_device *dev = rx_ring->netdev;
1639 ixgbe_update_rsc_stats(rx_ring, skb);
1641 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1643 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1645 if (unlikely(ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_TS)))
1646 ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector->adapter, skb);
1648 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1649 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1650 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1651 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1654 skb_record_rx_queue(skb, rx_ring->queue_index);
1656 skb->protocol = eth_type_trans(skb, dev);
1659 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1660 struct sk_buff *skb)
1662 if (ixgbe_qv_busy_polling(q_vector))
1663 netif_receive_skb(skb);
1665 napi_gro_receive(&q_vector->napi, skb);
1669 * ixgbe_is_non_eop - process handling of non-EOP buffers
1670 * @rx_ring: Rx ring being processed
1671 * @rx_desc: Rx descriptor for current buffer
1672 * @skb: Current socket buffer containing buffer in progress
1674 * This function updates next to clean. If the buffer is an EOP buffer
1675 * this function exits returning false, otherwise it will place the
1676 * sk_buff in the next buffer to be chained and return true indicating
1677 * that this is in fact a non-EOP buffer.
1679 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1680 union ixgbe_adv_rx_desc *rx_desc,
1681 struct sk_buff *skb)
1683 u32 ntc = rx_ring->next_to_clean + 1;
1685 /* fetch, update, and store next to clean */
1686 ntc = (ntc < rx_ring->count) ? ntc : 0;
1687 rx_ring->next_to_clean = ntc;
1689 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1691 /* update RSC append count if present */
1692 if (ring_is_rsc_enabled(rx_ring)) {
1693 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1694 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1696 if (unlikely(rsc_enabled)) {
1697 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1699 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1700 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1702 /* update ntc based on RSC value */
1703 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1704 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1705 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1709 /* if we are the last buffer then there is nothing else to do */
1710 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1713 /* place skb in next buffer to be received */
1714 rx_ring->rx_buffer_info[ntc].skb = skb;
1715 rx_ring->rx_stats.non_eop_descs++;
1721 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1722 * @rx_ring: rx descriptor ring packet is being transacted on
1723 * @skb: pointer to current skb being adjusted
1725 * This function is an ixgbe specific version of __pskb_pull_tail. The
1726 * main difference between this version and the original function is that
1727 * this function can make several assumptions about the state of things
1728 * that allow for significant optimizations versus the standard function.
1729 * As a result we can do things like drop a frag and maintain an accurate
1730 * truesize for the skb.
1732 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1733 struct sk_buff *skb)
1735 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1737 unsigned int pull_len;
1740 * it is valid to use page_address instead of kmap since we are
1741 * working with pages allocated out of the lomem pool per
1742 * alloc_page(GFP_ATOMIC)
1744 va = skb_frag_address(frag);
1747 * we need the header to contain the greater of either ETH_HLEN or
1748 * 60 bytes if the skb->len is less than 60 for skb_pad.
1750 pull_len = eth_get_headlen(va, IXGBE_RX_HDR_SIZE);
1752 /* align pull length to size of long to optimize memcpy performance */
1753 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1755 /* update all of the pointers */
1756 skb_frag_size_sub(frag, pull_len);
1757 frag->page_offset += pull_len;
1758 skb->data_len -= pull_len;
1759 skb->tail += pull_len;
1763 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1764 * @rx_ring: rx descriptor ring packet is being transacted on
1765 * @skb: pointer to current skb being updated
1767 * This function provides a basic DMA sync up for the first fragment of an
1768 * skb. The reason for doing this is that the first fragment cannot be
1769 * unmapped until we have reached the end of packet descriptor for a buffer
1772 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1773 struct sk_buff *skb)
1775 /* if the page was released unmap it, else just sync our portion */
1776 if (unlikely(IXGBE_CB(skb)->page_released)) {
1777 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1778 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1779 IXGBE_CB(skb)->page_released = false;
1781 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1783 dma_sync_single_range_for_cpu(rx_ring->dev,
1786 ixgbe_rx_bufsz(rx_ring),
1789 IXGBE_CB(skb)->dma = 0;
1793 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1794 * @rx_ring: rx descriptor ring packet is being transacted on
1795 * @rx_desc: pointer to the EOP Rx descriptor
1796 * @skb: pointer to current skb being fixed
1798 * Check for corrupted packet headers caused by senders on the local L2
1799 * embedded NIC switch not setting up their Tx Descriptors right. These
1800 * should be very rare.
1802 * Also address the case where we are pulling data in on pages only
1803 * and as such no data is present in the skb header.
1805 * In addition if skb is not at least 60 bytes we need to pad it so that
1806 * it is large enough to qualify as a valid Ethernet frame.
1808 * Returns true if an error was encountered and skb was freed.
1810 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1811 union ixgbe_adv_rx_desc *rx_desc,
1812 struct sk_buff *skb)
1814 struct net_device *netdev = rx_ring->netdev;
1816 /* verify that the packet does not have any known errors */
1817 if (unlikely(ixgbe_test_staterr(rx_desc,
1818 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1819 !(netdev->features & NETIF_F_RXALL))) {
1820 dev_kfree_skb_any(skb);
1824 /* place header in linear portion of buffer */
1825 if (skb_is_nonlinear(skb))
1826 ixgbe_pull_tail(rx_ring, skb);
1829 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1830 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1834 /* if eth_skb_pad returns an error the skb was freed */
1835 if (eth_skb_pad(skb))
1842 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1843 * @rx_ring: rx descriptor ring to store buffers on
1844 * @old_buff: donor buffer to have page reused
1846 * Synchronizes page for reuse by the adapter
1848 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1849 struct ixgbe_rx_buffer *old_buff)
1851 struct ixgbe_rx_buffer *new_buff;
1852 u16 nta = rx_ring->next_to_alloc;
1854 new_buff = &rx_ring->rx_buffer_info[nta];
1856 /* update, and store next to alloc */
1858 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1860 /* transfer page from old buffer to new buffer */
1861 *new_buff = *old_buff;
1863 /* sync the buffer for use by the device */
1864 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1865 new_buff->page_offset,
1866 ixgbe_rx_bufsz(rx_ring),
1870 static inline bool ixgbe_page_is_reserved(struct page *page)
1872 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1876 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1877 * @rx_ring: rx descriptor ring to transact packets on
1878 * @rx_buffer: buffer containing page to add
1879 * @rx_desc: descriptor containing length of buffer written by hardware
1880 * @skb: sk_buff to place the data into
1882 * This function will add the data contained in rx_buffer->page to the skb.
1883 * This is done either through a direct copy if the data in the buffer is
1884 * less than the skb header size, otherwise it will just attach the page as
1885 * a frag to the skb.
1887 * The function will then update the page offset if necessary and return
1888 * true if the buffer can be reused by the adapter.
1890 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1891 struct ixgbe_rx_buffer *rx_buffer,
1892 union ixgbe_adv_rx_desc *rx_desc,
1893 struct sk_buff *skb)
1895 struct page *page = rx_buffer->page;
1896 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1897 #if (PAGE_SIZE < 8192)
1898 unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1900 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1901 unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1902 ixgbe_rx_bufsz(rx_ring);
1905 if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1906 unsigned char *va = page_address(page) + rx_buffer->page_offset;
1908 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1910 /* page is not reserved, we can reuse buffer as-is */
1911 if (likely(!ixgbe_page_is_reserved(page)))
1914 /* this page cannot be reused so discard it */
1915 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1919 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1920 rx_buffer->page_offset, size, truesize);
1922 /* avoid re-using remote pages */
1923 if (unlikely(ixgbe_page_is_reserved(page)))
1926 #if (PAGE_SIZE < 8192)
1927 /* if we are only owner of page we can reuse it */
1928 if (unlikely(page_count(page) != 1))
1931 /* flip page offset to other buffer */
1932 rx_buffer->page_offset ^= truesize;
1934 /* move offset up to the next cache line */
1935 rx_buffer->page_offset += truesize;
1937 if (rx_buffer->page_offset > last_offset)
1941 /* Even if we own the page, we are not allowed to use atomic_set()
1942 * This would break get_page_unless_zero() users.
1944 atomic_inc(&page->_count);
1949 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1950 union ixgbe_adv_rx_desc *rx_desc)
1952 struct ixgbe_rx_buffer *rx_buffer;
1953 struct sk_buff *skb;
1956 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1957 page = rx_buffer->page;
1960 skb = rx_buffer->skb;
1963 void *page_addr = page_address(page) +
1964 rx_buffer->page_offset;
1966 /* prefetch first cache line of first page */
1967 prefetch(page_addr);
1968 #if L1_CACHE_BYTES < 128
1969 prefetch(page_addr + L1_CACHE_BYTES);
1972 /* allocate a skb to store the frags */
1973 skb = napi_alloc_skb(&rx_ring->q_vector->napi,
1975 if (unlikely(!skb)) {
1976 rx_ring->rx_stats.alloc_rx_buff_failed++;
1981 * we will be copying header into skb->data in
1982 * pskb_may_pull so it is in our interest to prefetch
1983 * it now to avoid a possible cache miss
1985 prefetchw(skb->data);
1988 * Delay unmapping of the first packet. It carries the
1989 * header information, HW may still access the header
1990 * after the writeback. Only unmap it when EOP is
1993 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1996 IXGBE_CB(skb)->dma = rx_buffer->dma;
1998 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
1999 ixgbe_dma_sync_frag(rx_ring, skb);
2002 /* we are reusing so sync this buffer for CPU use */
2003 dma_sync_single_range_for_cpu(rx_ring->dev,
2005 rx_buffer->page_offset,
2006 ixgbe_rx_bufsz(rx_ring),
2009 rx_buffer->skb = NULL;
2012 /* pull page into skb */
2013 if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
2014 /* hand second half of page back to the ring */
2015 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2016 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
2017 /* the page has been released from the ring */
2018 IXGBE_CB(skb)->page_released = true;
2020 /* we are not reusing the buffer so unmap it */
2021 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
2022 ixgbe_rx_pg_size(rx_ring),
2026 /* clear contents of buffer_info */
2027 rx_buffer->page = NULL;
2033 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2034 * @q_vector: structure containing interrupt and ring information
2035 * @rx_ring: rx descriptor ring to transact packets on
2036 * @budget: Total limit on number of packets to process
2038 * This function provides a "bounce buffer" approach to Rx interrupt
2039 * processing. The advantage to this is that on systems that have
2040 * expensive overhead for IOMMU access this provides a means of avoiding
2041 * it by maintaining the mapping of the page to the syste.
2043 * Returns amount of work completed
2045 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2046 struct ixgbe_ring *rx_ring,
2049 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2051 struct ixgbe_adapter *adapter = q_vector->adapter;
2053 unsigned int mss = 0;
2054 #endif /* IXGBE_FCOE */
2055 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2057 while (likely(total_rx_packets < budget)) {
2058 union ixgbe_adv_rx_desc *rx_desc;
2059 struct sk_buff *skb;
2061 /* return some buffers to hardware, one at a time is too slow */
2062 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2063 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2067 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2069 if (!rx_desc->wb.upper.status_error)
2072 /* This memory barrier is needed to keep us from reading
2073 * any other fields out of the rx_desc until we know the
2074 * descriptor has been written back
2078 /* retrieve a buffer from the ring */
2079 skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
2081 /* exit if we failed to retrieve a buffer */
2087 /* place incomplete frames back on ring for completion */
2088 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2091 /* verify the packet layout is correct */
2092 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2095 /* probably a little skewed due to removing CRC */
2096 total_rx_bytes += skb->len;
2098 /* populate checksum, timestamp, VLAN, and protocol */
2099 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2102 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2103 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2104 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2105 /* include DDPed FCoE data */
2106 if (ddp_bytes > 0) {
2108 mss = rx_ring->netdev->mtu -
2109 sizeof(struct fcoe_hdr) -
2110 sizeof(struct fc_frame_header) -
2111 sizeof(struct fcoe_crc_eof);
2115 total_rx_bytes += ddp_bytes;
2116 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2120 dev_kfree_skb_any(skb);
2125 #endif /* IXGBE_FCOE */
2126 skb_mark_napi_id(skb, &q_vector->napi);
2127 ixgbe_rx_skb(q_vector, skb);
2129 /* update budget accounting */
2133 u64_stats_update_begin(&rx_ring->syncp);
2134 rx_ring->stats.packets += total_rx_packets;
2135 rx_ring->stats.bytes += total_rx_bytes;
2136 u64_stats_update_end(&rx_ring->syncp);
2137 q_vector->rx.total_packets += total_rx_packets;
2138 q_vector->rx.total_bytes += total_rx_bytes;
2140 return total_rx_packets;
2143 #ifdef CONFIG_NET_RX_BUSY_POLL
2144 /* must be called with local_bh_disable()d */
2145 static int ixgbe_low_latency_recv(struct napi_struct *napi)
2147 struct ixgbe_q_vector *q_vector =
2148 container_of(napi, struct ixgbe_q_vector, napi);
2149 struct ixgbe_adapter *adapter = q_vector->adapter;
2150 struct ixgbe_ring *ring;
2153 if (test_bit(__IXGBE_DOWN, &adapter->state))
2154 return LL_FLUSH_FAILED;
2156 if (!ixgbe_qv_lock_poll(q_vector))
2157 return LL_FLUSH_BUSY;
2159 ixgbe_for_each_ring(ring, q_vector->rx) {
2160 found = ixgbe_clean_rx_irq(q_vector, ring, 4);
2161 #ifdef BP_EXTENDED_STATS
2163 ring->stats.cleaned += found;
2165 ring->stats.misses++;
2171 ixgbe_qv_unlock_poll(q_vector);
2175 #endif /* CONFIG_NET_RX_BUSY_POLL */
2178 * ixgbe_configure_msix - Configure MSI-X hardware
2179 * @adapter: board private structure
2181 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2184 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2186 struct ixgbe_q_vector *q_vector;
2190 /* Populate MSIX to EITR Select */
2191 if (adapter->num_vfs > 32) {
2192 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2193 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2197 * Populate the IVAR table and set the ITR values to the
2198 * corresponding register.
2200 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2201 struct ixgbe_ring *ring;
2202 q_vector = adapter->q_vector[v_idx];
2204 ixgbe_for_each_ring(ring, q_vector->rx)
2205 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2207 ixgbe_for_each_ring(ring, q_vector->tx)
2208 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2210 ixgbe_write_eitr(q_vector);
2213 switch (adapter->hw.mac.type) {
2214 case ixgbe_mac_82598EB:
2215 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2218 case ixgbe_mac_82599EB:
2219 case ixgbe_mac_X540:
2220 case ixgbe_mac_X550:
2221 case ixgbe_mac_X550EM_x:
2222 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2227 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2229 /* set up to autoclear timer, and the vectors */
2230 mask = IXGBE_EIMS_ENABLE_MASK;
2231 mask &= ~(IXGBE_EIMS_OTHER |
2232 IXGBE_EIMS_MAILBOX |
2235 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2238 enum latency_range {
2242 latency_invalid = 255
2246 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2247 * @q_vector: structure containing interrupt and ring information
2248 * @ring_container: structure containing ring performance data
2250 * Stores a new ITR value based on packets and byte
2251 * counts during the last interrupt. The advantage of per interrupt
2252 * computation is faster updates and more accurate ITR for the current
2253 * traffic pattern. Constants in this function were computed
2254 * based on theoretical maximum wire speed and thresholds were set based
2255 * on testing data as well as attempting to minimize response time
2256 * while increasing bulk throughput.
2257 * this functionality is controlled by the InterruptThrottleRate module
2258 * parameter (see ixgbe_param.c)
2260 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2261 struct ixgbe_ring_container *ring_container)
2263 int bytes = ring_container->total_bytes;
2264 int packets = ring_container->total_packets;
2267 u8 itr_setting = ring_container->itr;
2272 /* simple throttlerate management
2273 * 0-10MB/s lowest (100000 ints/s)
2274 * 10-20MB/s low (20000 ints/s)
2275 * 20-1249MB/s bulk (12000 ints/s)
2277 /* what was last interrupt timeslice? */
2278 timepassed_us = q_vector->itr >> 2;
2279 if (timepassed_us == 0)
2282 bytes_perint = bytes / timepassed_us; /* bytes/usec */
2284 switch (itr_setting) {
2285 case lowest_latency:
2286 if (bytes_perint > 10)
2287 itr_setting = low_latency;
2290 if (bytes_perint > 20)
2291 itr_setting = bulk_latency;
2292 else if (bytes_perint <= 10)
2293 itr_setting = lowest_latency;
2296 if (bytes_perint <= 20)
2297 itr_setting = low_latency;
2301 /* clear work counters since we have the values we need */
2302 ring_container->total_bytes = 0;
2303 ring_container->total_packets = 0;
2305 /* write updated itr to ring container */
2306 ring_container->itr = itr_setting;
2310 * ixgbe_write_eitr - write EITR register in hardware specific way
2311 * @q_vector: structure containing interrupt and ring information
2313 * This function is made to be called by ethtool and by the driver
2314 * when it needs to update EITR registers at runtime. Hardware
2315 * specific quirks/differences are taken care of here.
2317 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2319 struct ixgbe_adapter *adapter = q_vector->adapter;
2320 struct ixgbe_hw *hw = &adapter->hw;
2321 int v_idx = q_vector->v_idx;
2322 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2324 switch (adapter->hw.mac.type) {
2325 case ixgbe_mac_82598EB:
2326 /* must write high and low 16 bits to reset counter */
2327 itr_reg |= (itr_reg << 16);
2329 case ixgbe_mac_82599EB:
2330 case ixgbe_mac_X540:
2331 case ixgbe_mac_X550:
2332 case ixgbe_mac_X550EM_x:
2334 * set the WDIS bit to not clear the timer bits and cause an
2335 * immediate assertion of the interrupt
2337 itr_reg |= IXGBE_EITR_CNT_WDIS;
2342 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2345 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2347 u32 new_itr = q_vector->itr;
2350 ixgbe_update_itr(q_vector, &q_vector->tx);
2351 ixgbe_update_itr(q_vector, &q_vector->rx);
2353 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2355 switch (current_itr) {
2356 /* counts and packets in update_itr are dependent on these numbers */
2357 case lowest_latency:
2358 new_itr = IXGBE_100K_ITR;
2361 new_itr = IXGBE_20K_ITR;
2364 new_itr = IXGBE_12K_ITR;
2370 if (new_itr != q_vector->itr) {
2371 /* do an exponential smoothing */
2372 new_itr = (10 * new_itr * q_vector->itr) /
2373 ((9 * new_itr) + q_vector->itr);
2375 /* save the algorithm value here */
2376 q_vector->itr = new_itr;
2378 ixgbe_write_eitr(q_vector);
2383 * ixgbe_check_overtemp_subtask - check for over temperature
2384 * @adapter: pointer to adapter
2386 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2388 struct ixgbe_hw *hw = &adapter->hw;
2389 u32 eicr = adapter->interrupt_event;
2391 if (test_bit(__IXGBE_DOWN, &adapter->state))
2394 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2395 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2398 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2400 switch (hw->device_id) {
2401 case IXGBE_DEV_ID_82599_T3_LOM:
2403 * Since the warning interrupt is for both ports
2404 * we don't have to check if:
2405 * - This interrupt wasn't for our port.
2406 * - We may have missed the interrupt so always have to
2407 * check if we got a LSC
2409 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2410 !(eicr & IXGBE_EICR_LSC))
2413 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2415 bool link_up = false;
2417 hw->mac.ops.check_link(hw, &speed, &link_up, false);
2423 /* Check if this is not due to overtemp */
2424 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2429 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2431 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2435 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2437 adapter->interrupt_event = 0;
2440 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2442 struct ixgbe_hw *hw = &adapter->hw;
2444 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2445 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2446 e_crit(probe, "Fan has stopped, replace the adapter\n");
2447 /* write to clear the interrupt */
2448 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2452 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2454 struct ixgbe_hw *hw = &adapter->hw;
2456 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2459 switch (adapter->hw.mac.type) {
2460 case ixgbe_mac_82599EB:
2462 * Need to check link state so complete overtemp check
2465 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2466 (eicr & IXGBE_EICR_LSC)) &&
2467 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2468 adapter->interrupt_event = eicr;
2469 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2470 ixgbe_service_event_schedule(adapter);
2474 case ixgbe_mac_X540:
2475 if (!(eicr & IXGBE_EICR_TS))
2482 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2485 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2487 switch (hw->mac.type) {
2488 case ixgbe_mac_82598EB:
2489 if (hw->phy.type == ixgbe_phy_nl)
2492 case ixgbe_mac_82599EB:
2493 case ixgbe_mac_X550EM_x:
2494 switch (hw->mac.ops.get_media_type(hw)) {
2495 case ixgbe_media_type_fiber:
2496 case ixgbe_media_type_fiber_qsfp:
2506 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2508 struct ixgbe_hw *hw = &adapter->hw;
2509 u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2511 if (!ixgbe_is_sfp(hw))
2514 /* Later MAC's use different SDP */
2515 if (hw->mac.type >= ixgbe_mac_X540)
2516 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2518 if (eicr & eicr_mask) {
2519 /* Clear the interrupt */
2520 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2521 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2522 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2523 adapter->sfp_poll_time = 0;
2524 ixgbe_service_event_schedule(adapter);
2528 if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2529 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2530 /* Clear the interrupt */
2531 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2532 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2533 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2534 ixgbe_service_event_schedule(adapter);
2539 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2541 struct ixgbe_hw *hw = &adapter->hw;
2544 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2545 adapter->link_check_timeout = jiffies;
2546 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2547 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2548 IXGBE_WRITE_FLUSH(hw);
2549 ixgbe_service_event_schedule(adapter);
2553 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2557 struct ixgbe_hw *hw = &adapter->hw;
2559 switch (hw->mac.type) {
2560 case ixgbe_mac_82598EB:
2561 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2562 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2564 case ixgbe_mac_82599EB:
2565 case ixgbe_mac_X540:
2566 case ixgbe_mac_X550:
2567 case ixgbe_mac_X550EM_x:
2568 mask = (qmask & 0xFFFFFFFF);
2570 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2571 mask = (qmask >> 32);
2573 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2578 /* skip the flush */
2581 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2585 struct ixgbe_hw *hw = &adapter->hw;
2587 switch (hw->mac.type) {
2588 case ixgbe_mac_82598EB:
2589 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2590 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2592 case ixgbe_mac_82599EB:
2593 case ixgbe_mac_X540:
2594 case ixgbe_mac_X550:
2595 case ixgbe_mac_X550EM_x:
2596 mask = (qmask & 0xFFFFFFFF);
2598 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2599 mask = (qmask >> 32);
2601 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2606 /* skip the flush */
2610 * ixgbe_irq_enable - Enable default interrupt generation settings
2611 * @adapter: board private structure
2613 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2616 struct ixgbe_hw *hw = &adapter->hw;
2617 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2619 /* don't reenable LSC while waiting for link */
2620 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2621 mask &= ~IXGBE_EIMS_LSC;
2623 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2624 switch (adapter->hw.mac.type) {
2625 case ixgbe_mac_82599EB:
2626 mask |= IXGBE_EIMS_GPI_SDP0(hw);
2628 case ixgbe_mac_X540:
2629 case ixgbe_mac_X550:
2630 case ixgbe_mac_X550EM_x:
2631 mask |= IXGBE_EIMS_TS;
2636 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2637 mask |= IXGBE_EIMS_GPI_SDP1(hw);
2638 switch (adapter->hw.mac.type) {
2639 case ixgbe_mac_82599EB:
2640 mask |= IXGBE_EIMS_GPI_SDP1(hw);
2641 mask |= IXGBE_EIMS_GPI_SDP2(hw);
2643 case ixgbe_mac_X540:
2644 case ixgbe_mac_X550:
2645 case ixgbe_mac_X550EM_x:
2646 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP)
2647 mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
2648 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
2649 mask |= IXGBE_EICR_GPI_SDP0_X540;
2650 mask |= IXGBE_EIMS_ECC;
2651 mask |= IXGBE_EIMS_MAILBOX;
2657 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2658 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2659 mask |= IXGBE_EIMS_FLOW_DIR;
2661 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2663 ixgbe_irq_enable_queues(adapter, ~0);
2665 IXGBE_WRITE_FLUSH(&adapter->hw);
2668 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2670 struct ixgbe_adapter *adapter = data;
2671 struct ixgbe_hw *hw = &adapter->hw;
2675 * Workaround for Silicon errata. Use clear-by-write instead
2676 * of clear-by-read. Reading with EICS will return the
2677 * interrupt causes without clearing, which later be done
2678 * with the write to EICR.
2680 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2682 /* The lower 16bits of the EICR register are for the queue interrupts
2683 * which should be masked here in order to not accidentally clear them if
2684 * the bits are high when ixgbe_msix_other is called. There is a race
2685 * condition otherwise which results in possible performance loss
2686 * especially if the ixgbe_msix_other interrupt is triggering
2687 * consistently (as it would when PPS is turned on for the X540 device)
2691 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2693 if (eicr & IXGBE_EICR_LSC)
2694 ixgbe_check_lsc(adapter);
2696 if (eicr & IXGBE_EICR_MAILBOX)
2697 ixgbe_msg_task(adapter);
2699 switch (hw->mac.type) {
2700 case ixgbe_mac_82599EB:
2701 case ixgbe_mac_X540:
2702 case ixgbe_mac_X550:
2703 case ixgbe_mac_X550EM_x:
2704 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
2705 (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
2706 adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
2707 ixgbe_service_event_schedule(adapter);
2708 IXGBE_WRITE_REG(hw, IXGBE_EICR,
2709 IXGBE_EICR_GPI_SDP0_X540);
2711 if (eicr & IXGBE_EICR_ECC) {
2712 e_info(link, "Received ECC Err, initiating reset\n");
2713 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2714 ixgbe_service_event_schedule(adapter);
2715 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2717 /* Handle Flow Director Full threshold interrupt */
2718 if (eicr & IXGBE_EICR_FLOW_DIR) {
2719 int reinit_count = 0;
2721 for (i = 0; i < adapter->num_tx_queues; i++) {
2722 struct ixgbe_ring *ring = adapter->tx_ring[i];
2723 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2728 /* no more flow director interrupts until after init */
2729 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2730 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2731 ixgbe_service_event_schedule(adapter);
2734 ixgbe_check_sfp_event(adapter, eicr);
2735 ixgbe_check_overtemp_event(adapter, eicr);
2741 ixgbe_check_fan_failure(adapter, eicr);
2743 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2744 ixgbe_ptp_check_pps_event(adapter, eicr);
2746 /* re-enable the original interrupt state, no lsc, no queues */
2747 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2748 ixgbe_irq_enable(adapter, false, false);
2753 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2755 struct ixgbe_q_vector *q_vector = data;
2757 /* EIAM disabled interrupts (on this vector) for us */
2759 if (q_vector->rx.ring || q_vector->tx.ring)
2760 napi_schedule(&q_vector->napi);
2766 * ixgbe_poll - NAPI Rx polling callback
2767 * @napi: structure for representing this polling device
2768 * @budget: how many packets driver is allowed to clean
2770 * This function is used for legacy and MSI, NAPI mode
2772 int ixgbe_poll(struct napi_struct *napi, int budget)
2774 struct ixgbe_q_vector *q_vector =
2775 container_of(napi, struct ixgbe_q_vector, napi);
2776 struct ixgbe_adapter *adapter = q_vector->adapter;
2777 struct ixgbe_ring *ring;
2778 int per_ring_budget, work_done = 0;
2779 bool clean_complete = true;
2781 #ifdef CONFIG_IXGBE_DCA
2782 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2783 ixgbe_update_dca(q_vector);
2786 ixgbe_for_each_ring(ring, q_vector->tx)
2787 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2789 if (!ixgbe_qv_lock_napi(q_vector))
2792 /* attempt to distribute budget to each queue fairly, but don't allow
2793 * the budget to go below 1 because we'll exit polling */
2794 if (q_vector->rx.count > 1)
2795 per_ring_budget = max(budget/q_vector->rx.count, 1);
2797 per_ring_budget = budget;
2799 ixgbe_for_each_ring(ring, q_vector->rx) {
2800 int cleaned = ixgbe_clean_rx_irq(q_vector, ring,
2803 work_done += cleaned;
2804 clean_complete &= (cleaned < per_ring_budget);
2807 ixgbe_qv_unlock_napi(q_vector);
2808 /* If all work not completed, return budget and keep polling */
2809 if (!clean_complete)
2812 /* all work done, exit the polling mode */
2813 napi_complete_done(napi, work_done);
2814 if (adapter->rx_itr_setting & 1)
2815 ixgbe_set_itr(q_vector);
2816 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2817 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2823 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2824 * @adapter: board private structure
2826 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2827 * interrupts from the kernel.
2829 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2831 struct net_device *netdev = adapter->netdev;
2835 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2836 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2837 struct msix_entry *entry = &adapter->msix_entries[vector];
2839 if (q_vector->tx.ring && q_vector->rx.ring) {
2840 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2841 "%s-%s-%d", netdev->name, "TxRx", ri++);
2843 } else if (q_vector->rx.ring) {
2844 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2845 "%s-%s-%d", netdev->name, "rx", ri++);
2846 } else if (q_vector->tx.ring) {
2847 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2848 "%s-%s-%d", netdev->name, "tx", ti++);
2850 /* skip this unused q_vector */
2853 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2854 q_vector->name, q_vector);
2856 e_err(probe, "request_irq failed for MSIX interrupt "
2857 "Error: %d\n", err);
2858 goto free_queue_irqs;
2860 /* If Flow Director is enabled, set interrupt affinity */
2861 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2862 /* assign the mask for this irq */
2863 irq_set_affinity_hint(entry->vector,
2864 &q_vector->affinity_mask);
2868 err = request_irq(adapter->msix_entries[vector].vector,
2869 ixgbe_msix_other, 0, netdev->name, adapter);
2871 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2872 goto free_queue_irqs;
2880 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2882 free_irq(adapter->msix_entries[vector].vector,
2883 adapter->q_vector[vector]);
2885 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2886 pci_disable_msix(adapter->pdev);
2887 kfree(adapter->msix_entries);
2888 adapter->msix_entries = NULL;
2893 * ixgbe_intr - legacy mode Interrupt Handler
2894 * @irq: interrupt number
2895 * @data: pointer to a network interface device structure
2897 static irqreturn_t ixgbe_intr(int irq, void *data)
2899 struct ixgbe_adapter *adapter = data;
2900 struct ixgbe_hw *hw = &adapter->hw;
2901 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2905 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2906 * before the read of EICR.
2908 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2910 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2911 * therefore no explicit interrupt disable is necessary */
2912 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2915 * shared interrupt alert!
2916 * make sure interrupts are enabled because the read will
2917 * have disabled interrupts due to EIAM
2918 * finish the workaround of silicon errata on 82598. Unmask
2919 * the interrupt that we masked before the EICR read.
2921 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2922 ixgbe_irq_enable(adapter, true, true);
2923 return IRQ_NONE; /* Not our interrupt */
2926 if (eicr & IXGBE_EICR_LSC)
2927 ixgbe_check_lsc(adapter);
2929 switch (hw->mac.type) {
2930 case ixgbe_mac_82599EB:
2931 ixgbe_check_sfp_event(adapter, eicr);
2933 case ixgbe_mac_X540:
2934 case ixgbe_mac_X550:
2935 case ixgbe_mac_X550EM_x:
2936 if (eicr & IXGBE_EICR_ECC) {
2937 e_info(link, "Received ECC Err, initiating reset\n");
2938 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2939 ixgbe_service_event_schedule(adapter);
2940 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2942 ixgbe_check_overtemp_event(adapter, eicr);
2948 ixgbe_check_fan_failure(adapter, eicr);
2949 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2950 ixgbe_ptp_check_pps_event(adapter, eicr);
2952 /* would disable interrupts here but EIAM disabled it */
2953 napi_schedule(&q_vector->napi);
2956 * re-enable link(maybe) and non-queue interrupts, no flush.
2957 * ixgbe_poll will re-enable the queue interrupts
2959 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2960 ixgbe_irq_enable(adapter, false, false);
2966 * ixgbe_request_irq - initialize interrupts
2967 * @adapter: board private structure
2969 * Attempts to configure interrupts using the best available
2970 * capabilities of the hardware and kernel.
2972 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2974 struct net_device *netdev = adapter->netdev;
2977 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2978 err = ixgbe_request_msix_irqs(adapter);
2979 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2980 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2981 netdev->name, adapter);
2983 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2984 netdev->name, adapter);
2987 e_err(probe, "request_irq failed, Error %d\n", err);
2992 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2996 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2997 free_irq(adapter->pdev->irq, adapter);
3001 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3002 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3003 struct msix_entry *entry = &adapter->msix_entries[vector];
3005 /* free only the irqs that were actually requested */
3006 if (!q_vector->rx.ring && !q_vector->tx.ring)
3009 /* clear the affinity_mask in the IRQ descriptor */
3010 irq_set_affinity_hint(entry->vector, NULL);
3012 free_irq(entry->vector, q_vector);
3015 free_irq(adapter->msix_entries[vector++].vector, adapter);
3019 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3020 * @adapter: board private structure
3022 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3024 switch (adapter->hw.mac.type) {
3025 case ixgbe_mac_82598EB:
3026 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3028 case ixgbe_mac_82599EB:
3029 case ixgbe_mac_X540:
3030 case ixgbe_mac_X550:
3031 case ixgbe_mac_X550EM_x:
3032 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3033 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3034 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3039 IXGBE_WRITE_FLUSH(&adapter->hw);
3040 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3043 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3044 synchronize_irq(adapter->msix_entries[vector].vector);
3046 synchronize_irq(adapter->msix_entries[vector++].vector);
3048 synchronize_irq(adapter->pdev->irq);
3053 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3056 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3058 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3060 ixgbe_write_eitr(q_vector);
3062 ixgbe_set_ivar(adapter, 0, 0, 0);
3063 ixgbe_set_ivar(adapter, 1, 0, 0);
3065 e_info(hw, "Legacy interrupt IVAR setup done\n");
3069 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3070 * @adapter: board private structure
3071 * @ring: structure containing ring specific data
3073 * Configure the Tx descriptor ring after a reset.
3075 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3076 struct ixgbe_ring *ring)
3078 struct ixgbe_hw *hw = &adapter->hw;
3079 u64 tdba = ring->dma;
3081 u32 txdctl = IXGBE_TXDCTL_ENABLE;
3082 u8 reg_idx = ring->reg_idx;
3084 /* disable queue to avoid issues while updating state */
3085 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3086 IXGBE_WRITE_FLUSH(hw);
3088 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3089 (tdba & DMA_BIT_MASK(32)));
3090 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3091 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3092 ring->count * sizeof(union ixgbe_adv_tx_desc));
3093 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3094 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3095 ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3098 * set WTHRESH to encourage burst writeback, it should not be set
3099 * higher than 1 when:
3100 * - ITR is 0 as it could cause false TX hangs
3101 * - ITR is set to > 100k int/sec and BQL is enabled
3103 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3104 * to or less than the number of on chip descriptors, which is
3107 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3108 txdctl |= (1 << 16); /* WTHRESH = 1 */
3110 txdctl |= (8 << 16); /* WTHRESH = 8 */
3113 * Setting PTHRESH to 32 both improves performance
3114 * and avoids a TX hang with DFP enabled
3116 txdctl |= (1 << 8) | /* HTHRESH = 1 */
3117 32; /* PTHRESH = 32 */
3119 /* reinitialize flowdirector state */
3120 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3121 ring->atr_sample_rate = adapter->atr_sample_rate;
3122 ring->atr_count = 0;
3123 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3125 ring->atr_sample_rate = 0;
3128 /* initialize XPS */
3129 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3130 struct ixgbe_q_vector *q_vector = ring->q_vector;
3133 netif_set_xps_queue(ring->netdev,
3134 &q_vector->affinity_mask,
3138 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3141 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3143 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3144 if (hw->mac.type == ixgbe_mac_82598EB &&
3145 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3148 /* poll to verify queue is enabled */
3150 usleep_range(1000, 2000);
3151 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3152 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3154 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
3157 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3159 struct ixgbe_hw *hw = &adapter->hw;
3161 u8 tcs = netdev_get_num_tc(adapter->netdev);
3163 if (hw->mac.type == ixgbe_mac_82598EB)
3166 /* disable the arbiter while setting MTQC */
3167 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3168 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3169 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3171 /* set transmit pool layout */
3172 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3173 mtqc = IXGBE_MTQC_VT_ENA;
3175 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3177 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3178 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3179 mtqc |= IXGBE_MTQC_32VF;
3181 mtqc |= IXGBE_MTQC_64VF;
3184 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3186 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3188 mtqc = IXGBE_MTQC_64Q_1PB;
3191 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3193 /* Enable Security TX Buffer IFG for multiple pb */
3195 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3196 sectx |= IXGBE_SECTX_DCB;
3197 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3200 /* re-enable the arbiter */
3201 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3202 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3206 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3207 * @adapter: board private structure
3209 * Configure the Tx unit of the MAC after a reset.
3211 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3213 struct ixgbe_hw *hw = &adapter->hw;
3217 ixgbe_setup_mtqc(adapter);
3219 if (hw->mac.type != ixgbe_mac_82598EB) {
3220 /* DMATXCTL.EN must be before Tx queues are enabled */
3221 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3222 dmatxctl |= IXGBE_DMATXCTL_TE;
3223 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3226 /* Setup the HW Tx Head and Tail descriptor pointers */
3227 for (i = 0; i < adapter->num_tx_queues; i++)
3228 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3231 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3232 struct ixgbe_ring *ring)
3234 struct ixgbe_hw *hw = &adapter->hw;
3235 u8 reg_idx = ring->reg_idx;
3236 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3238 srrctl |= IXGBE_SRRCTL_DROP_EN;
3240 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3243 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3244 struct ixgbe_ring *ring)
3246 struct ixgbe_hw *hw = &adapter->hw;
3247 u8 reg_idx = ring->reg_idx;
3248 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3250 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3252 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3255 #ifdef CONFIG_IXGBE_DCB
3256 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3258 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3262 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3264 if (adapter->ixgbe_ieee_pfc)
3265 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3268 * We should set the drop enable bit if:
3271 * Number of Rx queues > 1 and flow control is disabled
3273 * This allows us to avoid head of line blocking for security
3274 * and performance reasons.
3276 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3277 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3278 for (i = 0; i < adapter->num_rx_queues; i++)
3279 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3281 for (i = 0; i < adapter->num_rx_queues; i++)
3282 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3286 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3288 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3289 struct ixgbe_ring *rx_ring)
3291 struct ixgbe_hw *hw = &adapter->hw;
3293 u8 reg_idx = rx_ring->reg_idx;
3295 if (hw->mac.type == ixgbe_mac_82598EB) {
3296 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3299 * if VMDq is not active we must program one srrctl register
3300 * per RSS queue since we have enabled RDRXCTL.MVMEN
3305 /* configure header buffer length, needed for RSC */
3306 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3308 /* configure the packet buffer length */
3309 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3311 /* configure descriptor type */
3312 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3314 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3318 * Return a number of entries in the RSS indirection table
3320 * @adapter: device handle
3322 * - 82598/82599/X540: 128
3323 * - X550(non-SRIOV mode): 512
3324 * - X550(SRIOV mode): 64
3326 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3328 if (adapter->hw.mac.type < ixgbe_mac_X550)
3330 else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3337 * Write the RETA table to HW
3339 * @adapter: device handle
3341 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3343 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3345 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3346 struct ixgbe_hw *hw = &adapter->hw;
3349 u8 *indir_tbl = adapter->rss_indir_tbl;
3351 /* Fill out the redirection table as follows:
3352 * - 82598: 8 bit wide entries containing pair of 4 bit RSS
3354 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3355 * - X550: 8 bit wide entries containing 6 bit RSS index
3357 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3358 indices_multi = 0x11;
3360 indices_multi = 0x1;
3362 /* Write redirection table to HW */
3363 for (i = 0; i < reta_entries; i++) {
3364 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3367 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3369 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3377 * Write the RETA table to HW (for x550 devices in SRIOV mode)
3379 * @adapter: device handle
3381 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3383 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3385 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3386 struct ixgbe_hw *hw = &adapter->hw;
3388 unsigned int pf_pool = adapter->num_vfs;
3390 /* Write redirection table to HW */
3391 for (i = 0; i < reta_entries; i++) {
3392 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3394 IXGBE_WRITE_REG(hw, IXGBE_PFVFRETA(i >> 2, pf_pool),
3401 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3403 struct ixgbe_hw *hw = &adapter->hw;
3405 u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3406 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3408 /* Program table for at least 2 queues w/ SR-IOV so that VFs can
3409 * make full use of any rings they may have. We will use the
3410 * PSRTYPE register to control how many rings we use within the PF.
3412 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
3415 /* Fill out hash function seeds */
3416 for (i = 0; i < 10; i++)
3417 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3419 /* Fill out redirection table */
3420 memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3422 for (i = 0, j = 0; i < reta_entries; i++, j++) {
3426 adapter->rss_indir_tbl[i] = j;
3429 ixgbe_store_reta(adapter);
3432 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3434 struct ixgbe_hw *hw = &adapter->hw;
3435 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3436 unsigned int pf_pool = adapter->num_vfs;
3439 /* Fill out hash function seeds */
3440 for (i = 0; i < 10; i++)
3441 IXGBE_WRITE_REG(hw, IXGBE_PFVFRSSRK(i, pf_pool),
3442 adapter->rss_key[i]);
3444 /* Fill out the redirection table */
3445 for (i = 0, j = 0; i < 64; i++, j++) {
3449 adapter->rss_indir_tbl[i] = j;
3452 ixgbe_store_vfreta(adapter);
3455 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3457 struct ixgbe_hw *hw = &adapter->hw;
3458 u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3461 /* Disable indicating checksum in descriptor, enables RSS hash */
3462 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3463 rxcsum |= IXGBE_RXCSUM_PCSD;
3464 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3466 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3467 if (adapter->ring_feature[RING_F_RSS].mask)
3468 mrqc = IXGBE_MRQC_RSSEN;
3470 u8 tcs = netdev_get_num_tc(adapter->netdev);
3472 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3474 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3476 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3477 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3478 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3480 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3483 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3485 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3487 mrqc = IXGBE_MRQC_RSSEN;
3491 /* Perform hash on these packet types */
3492 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3493 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3494 IXGBE_MRQC_RSS_FIELD_IPV6 |
3495 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3497 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3498 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3499 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3500 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3502 netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
3503 if ((hw->mac.type >= ixgbe_mac_X550) &&
3504 (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3505 unsigned int pf_pool = adapter->num_vfs;
3507 /* Enable VF RSS mode */
3508 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3509 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3511 /* Setup RSS through the VF registers */
3512 ixgbe_setup_vfreta(adapter);
3513 vfmrqc = IXGBE_MRQC_RSSEN;
3514 vfmrqc |= rss_field;
3515 IXGBE_WRITE_REG(hw, IXGBE_PFVFMRQC(pf_pool), vfmrqc);
3517 ixgbe_setup_reta(adapter);
3519 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3524 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3525 * @adapter: address of board private structure
3526 * @index: index of ring to set
3528 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3529 struct ixgbe_ring *ring)
3531 struct ixgbe_hw *hw = &adapter->hw;
3533 u8 reg_idx = ring->reg_idx;
3535 if (!ring_is_rsc_enabled(ring))
3538 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3539 rscctrl |= IXGBE_RSCCTL_RSCEN;
3541 * we must limit the number of descriptors so that the
3542 * total size of max desc * buf_len is not greater
3545 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3546 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3549 #define IXGBE_MAX_RX_DESC_POLL 10
3550 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3551 struct ixgbe_ring *ring)
3553 struct ixgbe_hw *hw = &adapter->hw;
3554 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3556 u8 reg_idx = ring->reg_idx;
3558 if (ixgbe_removed(hw->hw_addr))
3560 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3561 if (hw->mac.type == ixgbe_mac_82598EB &&
3562 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3566 usleep_range(1000, 2000);
3567 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3568 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3571 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3572 "the polling period\n", reg_idx);
3576 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3577 struct ixgbe_ring *ring)
3579 struct ixgbe_hw *hw = &adapter->hw;
3580 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3582 u8 reg_idx = ring->reg_idx;
3584 if (ixgbe_removed(hw->hw_addr))
3586 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3587 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3589 /* write value back with RXDCTL.ENABLE bit cleared */
3590 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3592 if (hw->mac.type == ixgbe_mac_82598EB &&
3593 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3596 /* the hardware may take up to 100us to really disable the rx queue */
3599 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3600 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3603 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3604 "the polling period\n", reg_idx);
3608 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3609 struct ixgbe_ring *ring)
3611 struct ixgbe_hw *hw = &adapter->hw;
3612 u64 rdba = ring->dma;
3614 u8 reg_idx = ring->reg_idx;
3616 /* disable queue to avoid issues while updating state */
3617 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3618 ixgbe_disable_rx_queue(adapter, ring);
3620 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3621 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3622 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3623 ring->count * sizeof(union ixgbe_adv_rx_desc));
3624 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3625 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3626 ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
3628 ixgbe_configure_srrctl(adapter, ring);
3629 ixgbe_configure_rscctl(adapter, ring);
3631 if (hw->mac.type == ixgbe_mac_82598EB) {
3633 * enable cache line friendly hardware writes:
3634 * PTHRESH=32 descriptors (half the internal cache),
3635 * this also removes ugly rx_no_buffer_count increment
3636 * HTHRESH=4 descriptors (to minimize latency on fetch)
3637 * WTHRESH=8 burst writeback up to two cache lines
3639 rxdctl &= ~0x3FFFFF;
3643 /* enable receive descriptor ring */
3644 rxdctl |= IXGBE_RXDCTL_ENABLE;
3645 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3647 ixgbe_rx_desc_queue_enable(adapter, ring);
3648 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3651 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3653 struct ixgbe_hw *hw = &adapter->hw;
3654 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3657 /* PSRTYPE must be initialized in non 82598 adapters */
3658 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3659 IXGBE_PSRTYPE_UDPHDR |
3660 IXGBE_PSRTYPE_IPV4HDR |
3661 IXGBE_PSRTYPE_L2HDR |
3662 IXGBE_PSRTYPE_IPV6HDR;
3664 if (hw->mac.type == ixgbe_mac_82598EB)
3672 for_each_set_bit(pool, &adapter->fwd_bitmask, 32)
3673 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
3676 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3678 struct ixgbe_hw *hw = &adapter->hw;
3679 u32 reg_offset, vf_shift;
3680 u32 gcr_ext, vmdctl;
3683 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3686 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3687 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3688 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3689 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3690 vmdctl |= IXGBE_VT_CTL_REPLEN;
3691 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3693 vf_shift = VMDQ_P(0) % 32;
3694 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3696 /* Enable only the PF's pool for Tx/Rx */
3697 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3698 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3699 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3700 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3701 if (adapter->bridge_mode == BRIDGE_MODE_VEB)
3702 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3704 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3705 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3708 * Set up VF register offsets for selected VT Mode,
3709 * i.e. 32 or 64 VFs for SR-IOV
3711 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3712 case IXGBE_82599_VMDQ_8Q_MASK:
3713 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3715 case IXGBE_82599_VMDQ_4Q_MASK:
3716 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3719 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3723 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3726 /* Enable MAC Anti-Spoofing */
3727 hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3730 /* Ensure LLDP and FC is set for Ethertype Antispoofing if we will be
3731 * calling set_ethertype_anti_spoofing for each VF in loop below
3733 if (hw->mac.ops.set_ethertype_anti_spoofing) {
3734 IXGBE_WRITE_REG(hw, IXGBE_ETQF(IXGBE_ETQF_FILTER_LLDP),
3735 (IXGBE_ETQF_FILTER_EN |
3736 IXGBE_ETQF_TX_ANTISPOOF |
3739 IXGBE_WRITE_REG(hw, IXGBE_ETQF(IXGBE_ETQF_FILTER_FC),
3740 (IXGBE_ETQF_FILTER_EN |
3741 IXGBE_ETQF_TX_ANTISPOOF |
3745 /* For VFs that have spoof checking turned off */
3746 for (i = 0; i < adapter->num_vfs; i++) {
3747 if (!adapter->vfinfo[i].spoofchk_enabled)
3748 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3750 /* enable ethertype anti spoofing if hw supports it */
3751 if (hw->mac.ops.set_ethertype_anti_spoofing)
3752 hw->mac.ops.set_ethertype_anti_spoofing(hw, true, i);
3754 /* Enable/Disable RSS query feature */
3755 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
3756 adapter->vfinfo[i].rss_query_enabled);
3760 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3762 struct ixgbe_hw *hw = &adapter->hw;
3763 struct net_device *netdev = adapter->netdev;
3764 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3765 struct ixgbe_ring *rx_ring;
3770 /* adjust max frame to be able to do baby jumbo for FCoE */
3771 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3772 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3773 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3775 #endif /* IXGBE_FCOE */
3777 /* adjust max frame to be at least the size of a standard frame */
3778 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3779 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3781 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3782 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3783 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3784 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3786 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3789 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3790 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3791 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3792 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3795 * Setup the HW Rx Head and Tail Descriptor Pointers and
3796 * the Base and Length of the Rx Descriptor Ring
3798 for (i = 0; i < adapter->num_rx_queues; i++) {
3799 rx_ring = adapter->rx_ring[i];
3800 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3801 set_ring_rsc_enabled(rx_ring);
3803 clear_ring_rsc_enabled(rx_ring);
3807 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3809 struct ixgbe_hw *hw = &adapter->hw;
3810 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3812 switch (hw->mac.type) {
3813 case ixgbe_mac_82598EB:
3815 * For VMDq support of different descriptor types or
3816 * buffer sizes through the use of multiple SRRCTL
3817 * registers, RDRXCTL.MVMEN must be set to 1
3819 * also, the manual doesn't mention it clearly but DCA hints
3820 * will only use queue 0's tags unless this bit is set. Side
3821 * effects of setting this bit are only that SRRCTL must be
3822 * fully programmed [0..15]
3824 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3826 case ixgbe_mac_X550:
3827 case ixgbe_mac_X550EM_x:
3828 if (adapter->num_vfs)
3829 rdrxctl |= IXGBE_RDRXCTL_PSP;
3830 /* fall through for older HW */
3831 case ixgbe_mac_82599EB:
3832 case ixgbe_mac_X540:
3833 /* Disable RSC for ACK packets */
3834 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3835 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3836 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3837 /* hardware requires some bits to be set by default */
3838 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3839 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3842 /* We should do nothing since we don't know this hardware */
3846 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3850 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3851 * @adapter: board private structure
3853 * Configure the Rx unit of the MAC after a reset.
3855 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3857 struct ixgbe_hw *hw = &adapter->hw;
3861 /* disable receives while setting up the descriptors */
3862 hw->mac.ops.disable_rx(hw);
3864 ixgbe_setup_psrtype(adapter);
3865 ixgbe_setup_rdrxctl(adapter);
3868 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
3869 rfctl &= ~IXGBE_RFCTL_RSC_DIS;
3870 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
3871 rfctl |= IXGBE_RFCTL_RSC_DIS;
3872 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
3874 /* Program registers for the distribution of queues */
3875 ixgbe_setup_mrqc(adapter);
3877 /* set_rx_buffer_len must be called before ring initialization */
3878 ixgbe_set_rx_buffer_len(adapter);
3881 * Setup the HW Rx Head and Tail Descriptor Pointers and
3882 * the Base and Length of the Rx Descriptor Ring
3884 for (i = 0; i < adapter->num_rx_queues; i++)
3885 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3887 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3888 /* disable drop enable for 82598 parts */
3889 if (hw->mac.type == ixgbe_mac_82598EB)
3890 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3892 /* enable all receives */
3893 rxctrl |= IXGBE_RXCTRL_RXEN;
3894 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3897 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
3898 __be16 proto, u16 vid)
3900 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3901 struct ixgbe_hw *hw = &adapter->hw;
3903 /* add VID to filter table */
3904 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3905 set_bit(vid, adapter->active_vlans);
3910 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
3911 __be16 proto, u16 vid)
3913 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3914 struct ixgbe_hw *hw = &adapter->hw;
3916 /* remove VID from filter table */
3917 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3918 clear_bit(vid, adapter->active_vlans);
3924 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3925 * @adapter: driver data
3927 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3929 struct ixgbe_hw *hw = &adapter->hw;
3933 switch (hw->mac.type) {
3934 case ixgbe_mac_82598EB:
3935 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3936 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3937 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3939 case ixgbe_mac_82599EB:
3940 case ixgbe_mac_X540:
3941 case ixgbe_mac_X550:
3942 case ixgbe_mac_X550EM_x:
3943 for (i = 0; i < adapter->num_rx_queues; i++) {
3944 struct ixgbe_ring *ring = adapter->rx_ring[i];
3946 if (ring->l2_accel_priv)
3949 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3950 vlnctrl &= ~IXGBE_RXDCTL_VME;
3951 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3960 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3961 * @adapter: driver data
3963 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3965 struct ixgbe_hw *hw = &adapter->hw;
3969 switch (hw->mac.type) {
3970 case ixgbe_mac_82598EB:
3971 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3972 vlnctrl |= IXGBE_VLNCTRL_VME;
3973 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3975 case ixgbe_mac_82599EB:
3976 case ixgbe_mac_X540:
3977 case ixgbe_mac_X550:
3978 case ixgbe_mac_X550EM_x:
3979 for (i = 0; i < adapter->num_rx_queues; i++) {
3980 struct ixgbe_ring *ring = adapter->rx_ring[i];
3982 if (ring->l2_accel_priv)
3985 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3986 vlnctrl |= IXGBE_RXDCTL_VME;
3987 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3995 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3999 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4001 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4002 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4006 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4007 * @netdev: network interface device structure
4009 * Writes multicast address list to the MTA hash table.
4010 * Returns: -ENOMEM on failure
4011 * 0 on no addresses written
4012 * X on writing X addresses to MTA
4014 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4016 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4017 struct ixgbe_hw *hw = &adapter->hw;
4019 if (!netif_running(netdev))
4022 if (hw->mac.ops.update_mc_addr_list)
4023 hw->mac.ops.update_mc_addr_list(hw, netdev);
4027 #ifdef CONFIG_PCI_IOV
4028 ixgbe_restore_vf_multicasts(adapter);
4031 return netdev_mc_count(netdev);
4034 #ifdef CONFIG_PCI_IOV
4035 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4037 struct ixgbe_hw *hw = &adapter->hw;
4039 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4040 if (adapter->mac_table[i].state & IXGBE_MAC_STATE_IN_USE)
4041 hw->mac.ops.set_rar(hw, i, 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 &= ~(IXGBE_MAC_STATE_MODIFIED);
4052 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4054 struct ixgbe_hw *hw = &adapter->hw;
4056 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4057 if (adapter->mac_table[i].state & IXGBE_MAC_STATE_MODIFIED) {
4058 if (adapter->mac_table[i].state &
4059 IXGBE_MAC_STATE_IN_USE)
4060 hw->mac.ops.set_rar(hw, i,
4061 adapter->mac_table[i].addr,
4062 adapter->mac_table[i].queue,
4065 hw->mac.ops.clear_rar(hw, i);
4067 adapter->mac_table[i].state &=
4068 ~(IXGBE_MAC_STATE_MODIFIED);
4073 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4076 struct ixgbe_hw *hw = &adapter->hw;
4078 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4079 adapter->mac_table[i].state |= IXGBE_MAC_STATE_MODIFIED;
4080 adapter->mac_table[i].state &= ~IXGBE_MAC_STATE_IN_USE;
4081 eth_zero_addr(adapter->mac_table[i].addr);
4082 adapter->mac_table[i].queue = 0;
4084 ixgbe_sync_mac_table(adapter);
4087 static int ixgbe_available_rars(struct ixgbe_adapter *adapter)
4089 struct ixgbe_hw *hw = &adapter->hw;
4092 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4093 if (adapter->mac_table[i].state == 0)
4099 /* this function destroys the first RAR entry */
4100 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter,
4103 struct ixgbe_hw *hw = &adapter->hw;
4105 memcpy(&adapter->mac_table[0].addr, addr, ETH_ALEN);
4106 adapter->mac_table[0].queue = VMDQ_P(0);
4107 adapter->mac_table[0].state = (IXGBE_MAC_STATE_DEFAULT |
4108 IXGBE_MAC_STATE_IN_USE);
4109 hw->mac.ops.set_rar(hw, 0, adapter->mac_table[0].addr,
4110 adapter->mac_table[0].queue,
4114 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter, u8 *addr, u16 queue)
4116 struct ixgbe_hw *hw = &adapter->hw;
4119 if (is_zero_ether_addr(addr))
4122 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4123 if (adapter->mac_table[i].state & IXGBE_MAC_STATE_IN_USE)
4125 adapter->mac_table[i].state |= (IXGBE_MAC_STATE_MODIFIED |
4126 IXGBE_MAC_STATE_IN_USE);
4127 ether_addr_copy(adapter->mac_table[i].addr, addr);
4128 adapter->mac_table[i].queue = queue;
4129 ixgbe_sync_mac_table(adapter);
4135 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter, u8 *addr, u16 queue)
4137 /* search table for addr, if found, set to 0 and sync */
4139 struct ixgbe_hw *hw = &adapter->hw;
4141 if (is_zero_ether_addr(addr))
4144 for (i = 0; i < hw->mac.num_rar_entries; i++) {
4145 if (ether_addr_equal(addr, adapter->mac_table[i].addr) &&
4146 adapter->mac_table[i].queue == queue) {
4147 adapter->mac_table[i].state |= IXGBE_MAC_STATE_MODIFIED;
4148 adapter->mac_table[i].state &= ~IXGBE_MAC_STATE_IN_USE;
4149 eth_zero_addr(adapter->mac_table[i].addr);
4150 adapter->mac_table[i].queue = 0;
4151 ixgbe_sync_mac_table(adapter);
4158 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
4159 * @netdev: network interface device structure
4161 * Writes unicast address list to the RAR table.
4162 * Returns: -ENOMEM on failure/insufficient address space
4163 * 0 on no addresses written
4164 * X on writing X addresses to the RAR table
4166 static int ixgbe_write_uc_addr_list(struct net_device *netdev, int vfn)
4168 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4171 /* return ENOMEM indicating insufficient memory for addresses */
4172 if (netdev_uc_count(netdev) > ixgbe_available_rars(adapter))
4175 if (!netdev_uc_empty(netdev)) {
4176 struct netdev_hw_addr *ha;
4177 netdev_for_each_uc_addr(ha, netdev) {
4178 ixgbe_del_mac_filter(adapter, ha->addr, vfn);
4179 ixgbe_add_mac_filter(adapter, ha->addr, vfn);
4187 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4188 * @netdev: network interface device structure
4190 * The set_rx_method entry point is called whenever the unicast/multicast
4191 * address list or the network interface flags are updated. This routine is
4192 * responsible for configuring the hardware for proper unicast, multicast and
4195 void ixgbe_set_rx_mode(struct net_device *netdev)
4197 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4198 struct ixgbe_hw *hw = &adapter->hw;
4199 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4203 /* Check for Promiscuous and All Multicast modes */
4204 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4205 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4207 /* set all bits that we expect to always be set */
4208 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4209 fctrl |= IXGBE_FCTRL_BAM;
4210 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4211 fctrl |= IXGBE_FCTRL_PMCF;
4213 /* clear the bits we are changing the status of */
4214 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4215 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
4216 if (netdev->flags & IFF_PROMISC) {
4217 hw->addr_ctrl.user_set_promisc = true;
4218 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4219 vmolr |= IXGBE_VMOLR_MPE;
4220 /* Only disable hardware filter vlans in promiscuous mode
4221 * if SR-IOV and VMDQ are disabled - otherwise ensure
4222 * that hardware VLAN filters remain enabled.
4224 if (adapter->flags & (IXGBE_FLAG_VMDQ_ENABLED |
4225 IXGBE_FLAG_SRIOV_ENABLED))
4226 vlnctrl |= (IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
4228 if (netdev->flags & IFF_ALLMULTI) {
4229 fctrl |= IXGBE_FCTRL_MPE;
4230 vmolr |= IXGBE_VMOLR_MPE;
4232 vlnctrl |= IXGBE_VLNCTRL_VFE;
4233 hw->addr_ctrl.user_set_promisc = false;
4237 * Write addresses to available RAR registers, if there is not
4238 * sufficient space to store all the addresses then enable
4239 * unicast promiscuous mode
4241 count = ixgbe_write_uc_addr_list(netdev, VMDQ_P(0));
4243 fctrl |= IXGBE_FCTRL_UPE;
4244 vmolr |= IXGBE_VMOLR_ROPE;
4247 /* Write addresses to the MTA, if the attempt fails
4248 * then we should just turn on promiscuous mode so
4249 * that we can at least receive multicast traffic
4251 count = ixgbe_write_mc_addr_list(netdev);
4253 fctrl |= IXGBE_FCTRL_MPE;
4254 vmolr |= IXGBE_VMOLR_MPE;
4256 vmolr |= IXGBE_VMOLR_ROMPE;
4259 if (hw->mac.type != ixgbe_mac_82598EB) {
4260 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4261 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4263 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4266 /* This is useful for sniffing bad packets. */
4267 if (adapter->netdev->features & NETIF_F_RXALL) {
4268 /* UPE and MPE will be handled by normal PROMISC logic
4269 * in e1000e_set_rx_mode */
4270 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4271 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4272 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4274 fctrl &= ~(IXGBE_FCTRL_DPF);
4275 /* NOTE: VLAN filtering is disabled by setting PROMISC */
4278 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4279 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4281 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
4282 ixgbe_vlan_strip_enable(adapter);
4284 ixgbe_vlan_strip_disable(adapter);
4287 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4291 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4292 ixgbe_qv_init_lock(adapter->q_vector[q_idx]);
4293 napi_enable(&adapter->q_vector[q_idx]->napi);
4297 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4301 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4302 napi_disable(&adapter->q_vector[q_idx]->napi);
4303 while (!ixgbe_qv_disable(adapter->q_vector[q_idx])) {
4304 pr_info("QV %d locked\n", q_idx);
4305 usleep_range(1000, 20000);
4310 static void ixgbe_clear_vxlan_port(struct ixgbe_adapter *adapter)
4312 switch (adapter->hw.mac.type) {
4313 case ixgbe_mac_X550:
4314 case ixgbe_mac_X550EM_x:
4315 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VXLANCTRL, 0);
4316 #ifdef CONFIG_IXGBE_VXLAN
4317 adapter->vxlan_port = 0;
4325 #ifdef CONFIG_IXGBE_DCB
4327 * ixgbe_configure_dcb - Configure DCB hardware
4328 * @adapter: ixgbe adapter struct
4330 * This is called by the driver on open to configure the DCB hardware.
4331 * This is also called by the gennetlink interface when reconfiguring
4334 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
4336 struct ixgbe_hw *hw = &adapter->hw;
4337 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4339 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
4340 if (hw->mac.type == ixgbe_mac_82598EB)
4341 netif_set_gso_max_size(adapter->netdev, 65536);
4345 if (hw->mac.type == ixgbe_mac_82598EB)
4346 netif_set_gso_max_size(adapter->netdev, 32768);
4349 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
4350 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
4353 /* reconfigure the hardware */
4354 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
4355 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4357 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4359 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
4360 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
4361 ixgbe_dcb_hw_ets(&adapter->hw,
4362 adapter->ixgbe_ieee_ets,
4364 ixgbe_dcb_hw_pfc_config(&adapter->hw,
4365 adapter->ixgbe_ieee_pfc->pfc_en,
4366 adapter->ixgbe_ieee_ets->prio_tc);
4369 /* Enable RSS Hash per TC */
4370 if (hw->mac.type != ixgbe_mac_82598EB) {
4372 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
4379 /* write msb to all 8 TCs in one write */
4380 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
4385 /* Additional bittime to account for IXGBE framing */
4386 #define IXGBE_ETH_FRAMING 20
4389 * ixgbe_hpbthresh - calculate high water mark for flow control
4391 * @adapter: board private structure to calculate for
4392 * @pb: packet buffer to calculate
4394 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
4396 struct ixgbe_hw *hw = &adapter->hw;
4397 struct net_device *dev = adapter->netdev;
4398 int link, tc, kb, marker;
4401 /* Calculate max LAN frame size */
4402 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
4405 /* FCoE traffic class uses FCOE jumbo frames */
4406 if ((dev->features & NETIF_F_FCOE_MTU) &&
4407 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4408 (pb == ixgbe_fcoe_get_tc(adapter)))
4409 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4412 /* Calculate delay value for device */
4413 switch (hw->mac.type) {
4414 case ixgbe_mac_X540:
4415 case ixgbe_mac_X550:
4416 case ixgbe_mac_X550EM_x:
4417 dv_id = IXGBE_DV_X540(link, tc);
4420 dv_id = IXGBE_DV(link, tc);
4424 /* Loopback switch introduces additional latency */
4425 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4426 dv_id += IXGBE_B2BT(tc);
4428 /* Delay value is calculated in bit times convert to KB */
4429 kb = IXGBE_BT2KB(dv_id);
4430 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
4432 marker = rx_pba - kb;
4434 /* It is possible that the packet buffer is not large enough
4435 * to provide required headroom. In this case throw an error
4436 * to user and a do the best we can.
4439 e_warn(drv, "Packet Buffer(%i) can not provide enough"
4440 "headroom to support flow control."
4441 "Decrease MTU or number of traffic classes\n", pb);
4449 * ixgbe_lpbthresh - calculate low water mark for for flow control
4451 * @adapter: board private structure to calculate for
4452 * @pb: packet buffer to calculate
4454 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
4456 struct ixgbe_hw *hw = &adapter->hw;
4457 struct net_device *dev = adapter->netdev;
4461 /* Calculate max LAN frame size */
4462 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4465 /* FCoE traffic class uses FCOE jumbo frames */
4466 if ((dev->features & NETIF_F_FCOE_MTU) &&
4467 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4468 (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
4469 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4472 /* Calculate delay value for device */
4473 switch (hw->mac.type) {
4474 case ixgbe_mac_X540:
4475 case ixgbe_mac_X550:
4476 case ixgbe_mac_X550EM_x:
4477 dv_id = IXGBE_LOW_DV_X540(tc);
4480 dv_id = IXGBE_LOW_DV(tc);
4484 /* Delay value is calculated in bit times convert to KB */
4485 return IXGBE_BT2KB(dv_id);
4489 * ixgbe_pbthresh_setup - calculate and setup high low water marks
4491 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
4493 struct ixgbe_hw *hw = &adapter->hw;
4494 int num_tc = netdev_get_num_tc(adapter->netdev);
4500 for (i = 0; i < num_tc; i++) {
4501 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
4502 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
4504 /* Low water marks must not be larger than high water marks */
4505 if (hw->fc.low_water[i] > hw->fc.high_water[i])
4506 hw->fc.low_water[i] = 0;
4509 for (; i < MAX_TRAFFIC_CLASS; i++)
4510 hw->fc.high_water[i] = 0;
4513 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
4515 struct ixgbe_hw *hw = &adapter->hw;
4517 u8 tc = netdev_get_num_tc(adapter->netdev);
4519 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4520 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4521 hdrm = 32 << adapter->fdir_pballoc;
4525 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
4526 ixgbe_pbthresh_setup(adapter);
4529 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
4531 struct ixgbe_hw *hw = &adapter->hw;
4532 struct hlist_node *node2;
4533 struct ixgbe_fdir_filter *filter;
4535 spin_lock(&adapter->fdir_perfect_lock);
4537 if (!hlist_empty(&adapter->fdir_filter_list))
4538 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
4540 hlist_for_each_entry_safe(filter, node2,
4541 &adapter->fdir_filter_list, fdir_node) {
4542 ixgbe_fdir_write_perfect_filter_82599(hw,
4545 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
4546 IXGBE_FDIR_DROP_QUEUE :
4547 adapter->rx_ring[filter->action]->reg_idx);
4550 spin_unlock(&adapter->fdir_perfect_lock);
4553 static void ixgbe_macvlan_set_rx_mode(struct net_device *dev, unsigned int pool,
4554 struct ixgbe_adapter *adapter)
4556 struct ixgbe_hw *hw = &adapter->hw;
4559 /* No unicast promiscuous support for VMDQ devices. */
4560 vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(pool));
4561 vmolr |= (IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
4563 /* clear the affected bit */
4564 vmolr &= ~IXGBE_VMOLR_MPE;
4566 if (dev->flags & IFF_ALLMULTI) {
4567 vmolr |= IXGBE_VMOLR_MPE;
4569 vmolr |= IXGBE_VMOLR_ROMPE;
4570 hw->mac.ops.update_mc_addr_list(hw, dev);
4572 ixgbe_write_uc_addr_list(adapter->netdev, pool);
4573 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(pool), vmolr);
4576 static void ixgbe_fwd_psrtype(struct ixgbe_fwd_adapter *vadapter)
4578 struct ixgbe_adapter *adapter = vadapter->real_adapter;
4579 int rss_i = adapter->num_rx_queues_per_pool;
4580 struct ixgbe_hw *hw = &adapter->hw;
4581 u16 pool = vadapter->pool;
4582 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4583 IXGBE_PSRTYPE_UDPHDR |
4584 IXGBE_PSRTYPE_IPV4HDR |
4585 IXGBE_PSRTYPE_L2HDR |
4586 IXGBE_PSRTYPE_IPV6HDR;
4588 if (hw->mac.type == ixgbe_mac_82598EB)
4596 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4600 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4601 * @rx_ring: ring to free buffers from
4603 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4605 struct device *dev = rx_ring->dev;
4609 /* ring already cleared, nothing to do */
4610 if (!rx_ring->rx_buffer_info)
4613 /* Free all the Rx ring sk_buffs */
4614 for (i = 0; i < rx_ring->count; i++) {
4615 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
4617 if (rx_buffer->skb) {
4618 struct sk_buff *skb = rx_buffer->skb;
4619 if (IXGBE_CB(skb)->page_released)
4622 ixgbe_rx_bufsz(rx_ring),
4625 rx_buffer->skb = NULL;
4628 if (!rx_buffer->page)
4631 dma_unmap_page(dev, rx_buffer->dma,
4632 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
4633 __free_pages(rx_buffer->page, ixgbe_rx_pg_order(rx_ring));
4635 rx_buffer->page = NULL;
4638 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4639 memset(rx_ring->rx_buffer_info, 0, size);
4641 /* Zero out the descriptor ring */
4642 memset(rx_ring->desc, 0, rx_ring->size);
4644 rx_ring->next_to_alloc = 0;
4645 rx_ring->next_to_clean = 0;
4646 rx_ring->next_to_use = 0;
4649 static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter *vadapter,
4650 struct ixgbe_ring *rx_ring)
4652 struct ixgbe_adapter *adapter = vadapter->real_adapter;
4653 int index = rx_ring->queue_index + vadapter->rx_base_queue;
4655 /* shutdown specific queue receive and wait for dma to settle */
4656 ixgbe_disable_rx_queue(adapter, rx_ring);
4657 usleep_range(10000, 20000);
4658 ixgbe_irq_disable_queues(adapter, ((u64)1 << index));
4659 ixgbe_clean_rx_ring(rx_ring);
4660 rx_ring->l2_accel_priv = NULL;
4663 static int ixgbe_fwd_ring_down(struct net_device *vdev,
4664 struct ixgbe_fwd_adapter *accel)
4666 struct ixgbe_adapter *adapter = accel->real_adapter;
4667 unsigned int rxbase = accel->rx_base_queue;
4668 unsigned int txbase = accel->tx_base_queue;
4671 netif_tx_stop_all_queues(vdev);
4673 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4674 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4675 adapter->rx_ring[rxbase + i]->netdev = adapter->netdev;
4678 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4679 adapter->tx_ring[txbase + i]->l2_accel_priv = NULL;
4680 adapter->tx_ring[txbase + i]->netdev = adapter->netdev;
4687 static int ixgbe_fwd_ring_up(struct net_device *vdev,
4688 struct ixgbe_fwd_adapter *accel)
4690 struct ixgbe_adapter *adapter = accel->real_adapter;
4691 unsigned int rxbase, txbase, queues;
4692 int i, baseq, err = 0;
4694 if (!test_bit(accel->pool, &adapter->fwd_bitmask))
4697 baseq = accel->pool * adapter->num_rx_queues_per_pool;
4698 netdev_dbg(vdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
4699 accel->pool, adapter->num_rx_pools,
4700 baseq, baseq + adapter->num_rx_queues_per_pool,
4701 adapter->fwd_bitmask);
4703 accel->netdev = vdev;
4704 accel->rx_base_queue = rxbase = baseq;
4705 accel->tx_base_queue = txbase = baseq;
4707 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
4708 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4710 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4711 adapter->rx_ring[rxbase + i]->netdev = vdev;
4712 adapter->rx_ring[rxbase + i]->l2_accel_priv = accel;
4713 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[rxbase + i]);
4716 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4717 adapter->tx_ring[txbase + i]->netdev = vdev;
4718 adapter->tx_ring[txbase + i]->l2_accel_priv = accel;
4721 queues = min_t(unsigned int,
4722 adapter->num_rx_queues_per_pool, vdev->num_tx_queues);
4723 err = netif_set_real_num_tx_queues(vdev, queues);
4727 err = netif_set_real_num_rx_queues(vdev, queues);
4731 if (is_valid_ether_addr(vdev->dev_addr))
4732 ixgbe_add_mac_filter(adapter, vdev->dev_addr, accel->pool);
4734 ixgbe_fwd_psrtype(accel);
4735 ixgbe_macvlan_set_rx_mode(vdev, accel->pool, adapter);
4738 ixgbe_fwd_ring_down(vdev, accel);
4742 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
4744 struct net_device *upper;
4745 struct list_head *iter;
4748 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
4749 if (netif_is_macvlan(upper)) {
4750 struct macvlan_dev *dfwd = netdev_priv(upper);
4751 struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv;
4753 if (dfwd->fwd_priv) {
4754 err = ixgbe_fwd_ring_up(upper, vadapter);
4762 static void ixgbe_configure(struct ixgbe_adapter *adapter)
4764 struct ixgbe_hw *hw = &adapter->hw;
4766 ixgbe_configure_pb(adapter);
4767 #ifdef CONFIG_IXGBE_DCB
4768 ixgbe_configure_dcb(adapter);
4771 * We must restore virtualization before VLANs or else
4772 * the VLVF registers will not be populated
4774 ixgbe_configure_virtualization(adapter);
4776 ixgbe_set_rx_mode(adapter->netdev);
4777 ixgbe_restore_vlan(adapter);
4779 switch (hw->mac.type) {
4780 case ixgbe_mac_82599EB:
4781 case ixgbe_mac_X540:
4782 hw->mac.ops.disable_rx_buff(hw);
4788 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4789 ixgbe_init_fdir_signature_82599(&adapter->hw,
4790 adapter->fdir_pballoc);
4791 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
4792 ixgbe_init_fdir_perfect_82599(&adapter->hw,
4793 adapter->fdir_pballoc);
4794 ixgbe_fdir_filter_restore(adapter);
4797 switch (hw->mac.type) {
4798 case ixgbe_mac_82599EB:
4799 case ixgbe_mac_X540:
4800 hw->mac.ops.enable_rx_buff(hw);
4806 #ifdef CONFIG_IXGBE_DCA
4808 if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
4809 ixgbe_setup_dca(adapter);
4810 #endif /* CONFIG_IXGBE_DCA */
4813 /* configure FCoE L2 filters, redirection table, and Rx control */
4814 ixgbe_configure_fcoe(adapter);
4816 #endif /* IXGBE_FCOE */
4817 ixgbe_configure_tx(adapter);
4818 ixgbe_configure_rx(adapter);
4819 ixgbe_configure_dfwd(adapter);
4823 * ixgbe_sfp_link_config - set up SFP+ link
4824 * @adapter: pointer to private adapter struct
4826 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
4829 * We are assuming the worst case scenario here, and that
4830 * is that an SFP was inserted/removed after the reset
4831 * but before SFP detection was enabled. As such the best
4832 * solution is to just start searching as soon as we start
4834 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
4835 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
4837 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
4838 adapter->sfp_poll_time = 0;
4842 * ixgbe_non_sfp_link_config - set up non-SFP+ link
4843 * @hw: pointer to private hardware struct
4845 * Returns 0 on success, negative on failure
4847 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
4850 bool autoneg, link_up = false;
4851 int ret = IXGBE_ERR_LINK_SETUP;
4853 if (hw->mac.ops.check_link)
4854 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
4859 speed = hw->phy.autoneg_advertised;
4860 if ((!speed) && (hw->mac.ops.get_link_capabilities))
4861 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
4866 if (hw->mac.ops.setup_link)
4867 ret = hw->mac.ops.setup_link(hw, speed, link_up);
4872 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
4874 struct ixgbe_hw *hw = &adapter->hw;
4877 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4878 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
4880 gpie |= IXGBE_GPIE_EIAME;
4882 * use EIAM to auto-mask when MSI-X interrupt is asserted
4883 * this saves a register write for every interrupt
4885 switch (hw->mac.type) {
4886 case ixgbe_mac_82598EB:
4887 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4889 case ixgbe_mac_82599EB:
4890 case ixgbe_mac_X540:
4891 case ixgbe_mac_X550:
4892 case ixgbe_mac_X550EM_x:
4894 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4895 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4899 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
4900 * specifically only auto mask tx and rx interrupts */
4901 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4904 /* XXX: to interrupt immediately for EICS writes, enable this */
4905 /* gpie |= IXGBE_GPIE_EIMEN; */
4907 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
4908 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
4910 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4911 case IXGBE_82599_VMDQ_8Q_MASK:
4912 gpie |= IXGBE_GPIE_VTMODE_16;
4914 case IXGBE_82599_VMDQ_4Q_MASK:
4915 gpie |= IXGBE_GPIE_VTMODE_32;
4918 gpie |= IXGBE_GPIE_VTMODE_64;
4923 /* Enable Thermal over heat sensor interrupt */
4924 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
4925 switch (adapter->hw.mac.type) {
4926 case ixgbe_mac_82599EB:
4927 gpie |= IXGBE_SDP0_GPIEN_8259X;
4934 /* Enable fan failure interrupt */
4935 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
4936 gpie |= IXGBE_SDP1_GPIEN(hw);
4938 switch (hw->mac.type) {
4939 case ixgbe_mac_82599EB:
4940 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
4942 case ixgbe_mac_X550EM_x:
4943 gpie |= IXGBE_SDP0_GPIEN_X540;
4949 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4952 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
4954 struct ixgbe_hw *hw = &adapter->hw;
4958 ixgbe_get_hw_control(adapter);
4959 ixgbe_setup_gpie(adapter);
4961 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4962 ixgbe_configure_msix(adapter);
4964 ixgbe_configure_msi_and_legacy(adapter);
4966 /* enable the optics for 82599 SFP+ fiber */
4967 if (hw->mac.ops.enable_tx_laser)
4968 hw->mac.ops.enable_tx_laser(hw);
4970 if (hw->phy.ops.set_phy_power)
4971 hw->phy.ops.set_phy_power(hw, true);
4973 smp_mb__before_atomic();
4974 clear_bit(__IXGBE_DOWN, &adapter->state);
4975 ixgbe_napi_enable_all(adapter);
4977 if (ixgbe_is_sfp(hw)) {
4978 ixgbe_sfp_link_config(adapter);
4980 err = ixgbe_non_sfp_link_config(hw);
4982 e_err(probe, "link_config FAILED %d\n", err);
4985 /* clear any pending interrupts, may auto mask */
4986 IXGBE_READ_REG(hw, IXGBE_EICR);
4987 ixgbe_irq_enable(adapter, true, true);
4990 * If this adapter has a fan, check to see if we had a failure
4991 * before we enabled the interrupt.
4993 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4994 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4995 if (esdp & IXGBE_ESDP_SDP1)
4996 e_crit(drv, "Fan has stopped, replace the adapter\n");
4999 /* bring the link up in the watchdog, this could race with our first
5000 * link up interrupt but shouldn't be a problem */
5001 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5002 adapter->link_check_timeout = jiffies;
5003 mod_timer(&adapter->service_timer, jiffies);
5005 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5006 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5007 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5008 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5011 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5013 WARN_ON(in_interrupt());
5014 /* put off any impending NetWatchDogTimeout */
5015 adapter->netdev->trans_start = jiffies;
5017 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5018 usleep_range(1000, 2000);
5019 ixgbe_down(adapter);
5021 * If SR-IOV enabled then wait a bit before bringing the adapter
5022 * back up to give the VFs time to respond to the reset. The
5023 * two second wait is based upon the watchdog timer cycle in
5026 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5029 clear_bit(__IXGBE_RESETTING, &adapter->state);
5032 void ixgbe_up(struct ixgbe_adapter *adapter)
5034 /* hardware has been reset, we need to reload some things */
5035 ixgbe_configure(adapter);
5037 ixgbe_up_complete(adapter);
5040 void ixgbe_reset(struct ixgbe_adapter *adapter)
5042 struct ixgbe_hw *hw = &adapter->hw;
5043 struct net_device *netdev = adapter->netdev;
5045 u8 old_addr[ETH_ALEN];
5047 if (ixgbe_removed(hw->hw_addr))
5049 /* lock SFP init bit to prevent race conditions with the watchdog */
5050 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5051 usleep_range(1000, 2000);
5053 /* clear all SFP and link config related flags while holding SFP_INIT */
5054 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5055 IXGBE_FLAG2_SFP_NEEDS_RESET);
5056 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5058 err = hw->mac.ops.init_hw(hw);
5061 case IXGBE_ERR_SFP_NOT_PRESENT:
5062 case IXGBE_ERR_SFP_NOT_SUPPORTED:
5064 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5065 e_dev_err("master disable timed out\n");
5067 case IXGBE_ERR_EEPROM_VERSION:
5068 /* We are running on a pre-production device, log a warning */
5069 e_dev_warn("This device is a pre-production adapter/LOM. "
5070 "Please be aware there may be issues associated with "
5071 "your hardware. If you are experiencing problems "
5072 "please contact your Intel or hardware "
5073 "representative who provided you with this "
5077 e_dev_err("Hardware Error: %d\n", err);
5080 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5081 /* do not flush user set addresses */
5082 memcpy(old_addr, &adapter->mac_table[0].addr, netdev->addr_len);
5083 ixgbe_flush_sw_mac_table(adapter);
5084 ixgbe_mac_set_default_filter(adapter, old_addr);
5086 /* update SAN MAC vmdq pool selection */
5087 if (hw->mac.san_mac_rar_index)
5088 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5090 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5091 ixgbe_ptp_reset(adapter);
5093 if (hw->phy.ops.set_phy_power) {
5094 if (!netif_running(adapter->netdev) && !adapter->wol)
5095 hw->phy.ops.set_phy_power(hw, false);
5097 hw->phy.ops.set_phy_power(hw, true);
5102 * ixgbe_clean_tx_ring - Free Tx Buffers
5103 * @tx_ring: ring to be cleaned
5105 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5107 struct ixgbe_tx_buffer *tx_buffer_info;
5111 /* ring already cleared, nothing to do */
5112 if (!tx_ring->tx_buffer_info)
5115 /* Free all the Tx ring sk_buffs */
5116 for (i = 0; i < tx_ring->count; i++) {
5117 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5118 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
5121 netdev_tx_reset_queue(txring_txq(tx_ring));
5123 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5124 memset(tx_ring->tx_buffer_info, 0, size);
5126 /* Zero out the descriptor ring */
5127 memset(tx_ring->desc, 0, tx_ring->size);
5129 tx_ring->next_to_use = 0;
5130 tx_ring->next_to_clean = 0;
5134 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
5135 * @adapter: board private structure
5137 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
5141 for (i = 0; i < adapter->num_rx_queues; i++)
5142 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
5146 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
5147 * @adapter: board private structure
5149 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
5153 for (i = 0; i < adapter->num_tx_queues; i++)
5154 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
5157 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
5159 struct hlist_node *node2;
5160 struct ixgbe_fdir_filter *filter;
5162 spin_lock(&adapter->fdir_perfect_lock);
5164 hlist_for_each_entry_safe(filter, node2,
5165 &adapter->fdir_filter_list, fdir_node) {
5166 hlist_del(&filter->fdir_node);
5169 adapter->fdir_filter_count = 0;
5171 spin_unlock(&adapter->fdir_perfect_lock);
5174 void ixgbe_down(struct ixgbe_adapter *adapter)
5176 struct net_device *netdev = adapter->netdev;
5177 struct ixgbe_hw *hw = &adapter->hw;
5178 struct net_device *upper;
5179 struct list_head *iter;
5182 /* signal that we are down to the interrupt handler */
5183 if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
5184 return; /* do nothing if already down */
5186 /* disable receives */
5187 hw->mac.ops.disable_rx(hw);
5189 /* disable all enabled rx queues */
5190 for (i = 0; i < adapter->num_rx_queues; i++)
5191 /* this call also flushes the previous write */
5192 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
5194 usleep_range(10000, 20000);
5196 netif_tx_stop_all_queues(netdev);
5198 /* call carrier off first to avoid false dev_watchdog timeouts */
5199 netif_carrier_off(netdev);
5200 netif_tx_disable(netdev);
5202 /* disable any upper devices */
5203 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
5204 if (netif_is_macvlan(upper)) {
5205 struct macvlan_dev *vlan = netdev_priv(upper);
5207 if (vlan->fwd_priv) {
5208 netif_tx_stop_all_queues(upper);
5209 netif_carrier_off(upper);
5210 netif_tx_disable(upper);
5215 ixgbe_irq_disable(adapter);
5217 ixgbe_napi_disable_all(adapter);
5219 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
5220 IXGBE_FLAG2_RESET_REQUESTED);
5221 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5223 del_timer_sync(&adapter->service_timer);
5225 if (adapter->num_vfs) {
5226 /* Clear EITR Select mapping */
5227 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
5229 /* Mark all the VFs as inactive */
5230 for (i = 0 ; i < adapter->num_vfs; i++)
5231 adapter->vfinfo[i].clear_to_send = false;
5233 /* ping all the active vfs to let them know we are going down */
5234 ixgbe_ping_all_vfs(adapter);
5236 /* Disable all VFTE/VFRE TX/RX */
5237 ixgbe_disable_tx_rx(adapter);
5240 /* disable transmits in the hardware now that interrupts are off */
5241 for (i = 0; i < adapter->num_tx_queues; i++) {
5242 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
5243 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5246 /* Disable the Tx DMA engine on 82599 and later MAC */
5247 switch (hw->mac.type) {
5248 case ixgbe_mac_82599EB:
5249 case ixgbe_mac_X540:
5250 case ixgbe_mac_X550:
5251 case ixgbe_mac_X550EM_x:
5252 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5253 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5254 ~IXGBE_DMATXCTL_TE));
5260 if (!pci_channel_offline(adapter->pdev))
5261 ixgbe_reset(adapter);
5263 /* power down the optics for 82599 SFP+ fiber */
5264 if (hw->mac.ops.disable_tx_laser)
5265 hw->mac.ops.disable_tx_laser(hw);
5267 ixgbe_clean_all_tx_rings(adapter);
5268 ixgbe_clean_all_rx_rings(adapter);
5272 * ixgbe_tx_timeout - Respond to a Tx Hang
5273 * @netdev: network interface device structure
5275 static void ixgbe_tx_timeout(struct net_device *netdev)
5277 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5279 /* Do the reset outside of interrupt context */
5280 ixgbe_tx_timeout_reset(adapter);
5284 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5285 * @adapter: board private structure to initialize
5287 * ixgbe_sw_init initializes the Adapter private data structure.
5288 * Fields are initialized based on PCI device information and
5289 * OS network device settings (MTU size).
5291 static int ixgbe_sw_init(struct ixgbe_adapter *adapter)
5293 struct ixgbe_hw *hw = &adapter->hw;
5294 struct pci_dev *pdev = adapter->pdev;
5295 unsigned int rss, fdir;
5297 #ifdef CONFIG_IXGBE_DCB
5299 struct tc_configuration *tc;
5302 /* PCI config space info */
5304 hw->vendor_id = pdev->vendor;
5305 hw->device_id = pdev->device;
5306 hw->revision_id = pdev->revision;
5307 hw->subsystem_vendor_id = pdev->subsystem_vendor;
5308 hw->subsystem_device_id = pdev->subsystem_device;
5310 /* Set common capability flags and settings */
5311 rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
5312 adapter->ring_feature[RING_F_RSS].limit = rss;
5313 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5314 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
5315 adapter->atr_sample_rate = 20;
5316 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
5317 adapter->ring_feature[RING_F_FDIR].limit = fdir;
5318 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
5319 #ifdef CONFIG_IXGBE_DCA
5320 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
5323 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5324 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5325 #ifdef CONFIG_IXGBE_DCB
5326 /* Default traffic class to use for FCoE */
5327 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5328 #endif /* CONFIG_IXGBE_DCB */
5329 #endif /* IXGBE_FCOE */
5331 adapter->mac_table = kzalloc(sizeof(struct ixgbe_mac_addr) *
5332 hw->mac.num_rar_entries,
5335 /* Set MAC specific capability flags and exceptions */
5336 switch (hw->mac.type) {
5337 case ixgbe_mac_82598EB:
5338 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
5340 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5341 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5343 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
5344 adapter->ring_feature[RING_F_FDIR].limit = 0;
5345 adapter->atr_sample_rate = 0;
5346 adapter->fdir_pballoc = 0;
5348 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5349 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5350 #ifdef CONFIG_IXGBE_DCB
5351 adapter->fcoe.up = 0;
5352 #endif /* IXGBE_DCB */
5353 #endif /* IXGBE_FCOE */
5355 case ixgbe_mac_82599EB:
5356 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5357 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5359 case ixgbe_mac_X540:
5360 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
5361 if (fwsm & IXGBE_FWSM_TS_ENABLED)
5362 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5364 case ixgbe_mac_X550EM_x:
5365 case ixgbe_mac_X550:
5366 #ifdef CONFIG_IXGBE_DCA
5367 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
5369 #ifdef CONFIG_IXGBE_VXLAN
5370 adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
5378 /* FCoE support exists, always init the FCoE lock */
5379 spin_lock_init(&adapter->fcoe.lock);
5382 /* n-tuple support exists, always init our spinlock */
5383 spin_lock_init(&adapter->fdir_perfect_lock);
5385 #ifdef CONFIG_IXGBE_DCB
5386 switch (hw->mac.type) {
5387 case ixgbe_mac_X540:
5388 case ixgbe_mac_X550:
5389 case ixgbe_mac_X550EM_x:
5390 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
5391 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
5394 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
5395 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
5399 /* Configure DCB traffic classes */
5400 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5401 tc = &adapter->dcb_cfg.tc_config[j];
5402 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5403 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5404 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5405 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5406 tc->dcb_pfc = pfc_disabled;
5409 /* Initialize default user to priority mapping, UPx->TC0 */
5410 tc = &adapter->dcb_cfg.tc_config[0];
5411 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5412 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5414 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5415 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5416 adapter->dcb_cfg.pfc_mode_enable = false;
5417 adapter->dcb_set_bitmap = 0x00;
5418 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5419 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
5420 sizeof(adapter->temp_dcb_cfg));
5424 /* default flow control settings */
5425 hw->fc.requested_mode = ixgbe_fc_full;
5426 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5427 ixgbe_pbthresh_setup(adapter);
5428 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5429 hw->fc.send_xon = true;
5430 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
5432 #ifdef CONFIG_PCI_IOV
5434 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
5436 /* assign number of SR-IOV VFs */
5437 if (hw->mac.type != ixgbe_mac_82598EB) {
5438 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
5439 adapter->num_vfs = 0;
5440 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
5442 adapter->num_vfs = max_vfs;
5445 #endif /* CONFIG_PCI_IOV */
5447 /* enable itr by default in dynamic mode */
5448 adapter->rx_itr_setting = 1;
5449 adapter->tx_itr_setting = 1;
5451 /* set default ring sizes */
5452 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5453 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5455 /* set default work limits */
5456 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5458 /* initialize eeprom parameters */
5459 if (ixgbe_init_eeprom_params_generic(hw)) {
5460 e_dev_err("EEPROM initialization failed\n");
5464 /* PF holds first pool slot */
5465 set_bit(0, &adapter->fwd_bitmask);
5466 set_bit(__IXGBE_DOWN, &adapter->state);
5472 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5473 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5475 * Return 0 on success, negative on failure
5477 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5479 struct device *dev = tx_ring->dev;
5480 int orig_node = dev_to_node(dev);
5484 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5486 if (tx_ring->q_vector)
5487 ring_node = tx_ring->q_vector->numa_node;
5489 tx_ring->tx_buffer_info = vzalloc_node(size, ring_node);
5490 if (!tx_ring->tx_buffer_info)
5491 tx_ring->tx_buffer_info = vzalloc(size);
5492 if (!tx_ring->tx_buffer_info)
5495 u64_stats_init(&tx_ring->syncp);
5497 /* round up to nearest 4K */
5498 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5499 tx_ring->size = ALIGN(tx_ring->size, 4096);
5501 set_dev_node(dev, ring_node);
5502 tx_ring->desc = dma_alloc_coherent(dev,
5506 set_dev_node(dev, orig_node);
5508 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5509 &tx_ring->dma, GFP_KERNEL);
5513 tx_ring->next_to_use = 0;
5514 tx_ring->next_to_clean = 0;
5518 vfree(tx_ring->tx_buffer_info);
5519 tx_ring->tx_buffer_info = NULL;
5520 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5525 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5526 * @adapter: board private structure
5528 * If this function returns with an error, then it's possible one or
5529 * more of the rings is populated (while the rest are not). It is the
5530 * callers duty to clean those orphaned rings.
5532 * Return 0 on success, negative on failure
5534 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5538 for (i = 0; i < adapter->num_tx_queues; i++) {
5539 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5543 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5549 /* rewind the index freeing the rings as we go */
5551 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5556 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5557 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5559 * Returns 0 on success, negative on failure
5561 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5563 struct device *dev = rx_ring->dev;
5564 int orig_node = dev_to_node(dev);
5568 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5570 if (rx_ring->q_vector)
5571 ring_node = rx_ring->q_vector->numa_node;
5573 rx_ring->rx_buffer_info = vzalloc_node(size, ring_node);
5574 if (!rx_ring->rx_buffer_info)
5575 rx_ring->rx_buffer_info = vzalloc(size);
5576 if (!rx_ring->rx_buffer_info)
5579 u64_stats_init(&rx_ring->syncp);
5581 /* Round up to nearest 4K */
5582 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5583 rx_ring->size = ALIGN(rx_ring->size, 4096);
5585 set_dev_node(dev, ring_node);
5586 rx_ring->desc = dma_alloc_coherent(dev,
5590 set_dev_node(dev, orig_node);
5592 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5593 &rx_ring->dma, GFP_KERNEL);
5597 rx_ring->next_to_clean = 0;
5598 rx_ring->next_to_use = 0;
5602 vfree(rx_ring->rx_buffer_info);
5603 rx_ring->rx_buffer_info = NULL;
5604 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5609 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5610 * @adapter: board private structure
5612 * If this function returns with an error, then it's possible one or
5613 * more of the rings is populated (while the rest are not). It is the
5614 * callers duty to clean those orphaned rings.
5616 * Return 0 on success, negative on failure
5618 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5622 for (i = 0; i < adapter->num_rx_queues; i++) {
5623 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5627 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5632 err = ixgbe_setup_fcoe_ddp_resources(adapter);
5637 /* rewind the index freeing the rings as we go */
5639 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5644 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5645 * @tx_ring: Tx descriptor ring for a specific queue
5647 * Free all transmit software resources
5649 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5651 ixgbe_clean_tx_ring(tx_ring);
5653 vfree(tx_ring->tx_buffer_info);
5654 tx_ring->tx_buffer_info = NULL;
5656 /* if not set, then don't free */
5660 dma_free_coherent(tx_ring->dev, tx_ring->size,
5661 tx_ring->desc, tx_ring->dma);
5663 tx_ring->desc = NULL;
5667 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5668 * @adapter: board private structure
5670 * Free all transmit software resources
5672 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5676 for (i = 0; i < adapter->num_tx_queues; i++)
5677 if (adapter->tx_ring[i]->desc)
5678 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5682 * ixgbe_free_rx_resources - Free Rx Resources
5683 * @rx_ring: ring to clean the resources from
5685 * Free all receive software resources
5687 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5689 ixgbe_clean_rx_ring(rx_ring);
5691 vfree(rx_ring->rx_buffer_info);
5692 rx_ring->rx_buffer_info = NULL;
5694 /* if not set, then don't free */
5698 dma_free_coherent(rx_ring->dev, rx_ring->size,
5699 rx_ring->desc, rx_ring->dma);
5701 rx_ring->desc = NULL;
5705 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5706 * @adapter: board private structure
5708 * Free all receive software resources
5710 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5715 ixgbe_free_fcoe_ddp_resources(adapter);
5718 for (i = 0; i < adapter->num_rx_queues; i++)
5719 if (adapter->rx_ring[i]->desc)
5720 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5724 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5725 * @netdev: network interface device structure
5726 * @new_mtu: new value for maximum frame size
5728 * Returns 0 on success, negative on failure
5730 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5732 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5733 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5735 /* MTU < 68 is an error and causes problems on some kernels */
5736 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5740 * For 82599EB we cannot allow legacy VFs to enable their receive
5741 * paths when MTU greater than 1500 is configured. So display a
5742 * warning that legacy VFs will be disabled.
5744 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
5745 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
5746 (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
5747 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
5749 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5751 /* must set new MTU before calling down or up */
5752 netdev->mtu = new_mtu;
5754 if (netif_running(netdev))
5755 ixgbe_reinit_locked(adapter);
5761 * ixgbe_open - Called when a network interface is made active
5762 * @netdev: network interface device structure
5764 * Returns 0 on success, negative value on failure
5766 * The open entry point is called when a network interface is made
5767 * active by the system (IFF_UP). At this point all resources needed
5768 * for transmit and receive operations are allocated, the interrupt
5769 * handler is registered with the OS, the watchdog timer is started,
5770 * and the stack is notified that the interface is ready.
5772 static int ixgbe_open(struct net_device *netdev)
5774 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5775 struct ixgbe_hw *hw = &adapter->hw;
5778 /* disallow open during test */
5779 if (test_bit(__IXGBE_TESTING, &adapter->state))
5782 netif_carrier_off(netdev);
5784 /* allocate transmit descriptors */
5785 err = ixgbe_setup_all_tx_resources(adapter);
5789 /* allocate receive descriptors */
5790 err = ixgbe_setup_all_rx_resources(adapter);
5794 ixgbe_configure(adapter);
5796 err = ixgbe_request_irq(adapter);
5800 /* Notify the stack of the actual queue counts. */
5801 if (adapter->num_rx_pools > 1)
5802 queues = adapter->num_rx_queues_per_pool;
5804 queues = adapter->num_tx_queues;
5806 err = netif_set_real_num_tx_queues(netdev, queues);
5808 goto err_set_queues;
5810 if (adapter->num_rx_pools > 1 &&
5811 adapter->num_rx_queues > IXGBE_MAX_L2A_QUEUES)
5812 queues = IXGBE_MAX_L2A_QUEUES;
5814 queues = adapter->num_rx_queues;
5815 err = netif_set_real_num_rx_queues(netdev, queues);
5817 goto err_set_queues;
5819 ixgbe_ptp_init(adapter);
5821 ixgbe_up_complete(adapter);
5823 ixgbe_clear_vxlan_port(adapter);
5824 #ifdef CONFIG_IXGBE_VXLAN
5825 vxlan_get_rx_port(netdev);
5831 ixgbe_free_irq(adapter);
5833 ixgbe_free_all_rx_resources(adapter);
5834 if (hw->phy.ops.set_phy_power && !adapter->wol)
5835 hw->phy.ops.set_phy_power(&adapter->hw, false);
5837 ixgbe_free_all_tx_resources(adapter);
5839 ixgbe_reset(adapter);
5844 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
5846 ixgbe_ptp_suspend(adapter);
5848 if (adapter->hw.phy.ops.enter_lplu) {
5849 adapter->hw.phy.reset_disable = true;
5850 ixgbe_down(adapter);
5851 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
5852 adapter->hw.phy.reset_disable = false;
5854 ixgbe_down(adapter);
5857 ixgbe_free_irq(adapter);
5859 ixgbe_free_all_tx_resources(adapter);
5860 ixgbe_free_all_rx_resources(adapter);
5864 * ixgbe_close - Disables a network interface
5865 * @netdev: network interface device structure
5867 * Returns 0, this is not allowed to fail
5869 * The close entry point is called when an interface is de-activated
5870 * by the OS. The hardware is still under the drivers control, but
5871 * needs to be disabled. A global MAC reset is issued to stop the
5872 * hardware, and all transmit and receive resources are freed.
5874 static int ixgbe_close(struct net_device *netdev)
5876 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5878 ixgbe_ptp_stop(adapter);
5880 ixgbe_close_suspend(adapter);
5882 ixgbe_fdir_filter_exit(adapter);
5884 ixgbe_release_hw_control(adapter);
5890 static int ixgbe_resume(struct pci_dev *pdev)
5892 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5893 struct net_device *netdev = adapter->netdev;
5896 adapter->hw.hw_addr = adapter->io_addr;
5897 pci_set_power_state(pdev, PCI_D0);
5898 pci_restore_state(pdev);
5900 * pci_restore_state clears dev->state_saved so call
5901 * pci_save_state to restore it.
5903 pci_save_state(pdev);
5905 err = pci_enable_device_mem(pdev);
5907 e_dev_err("Cannot enable PCI device from suspend\n");
5910 smp_mb__before_atomic();
5911 clear_bit(__IXGBE_DISABLED, &adapter->state);
5912 pci_set_master(pdev);
5914 pci_wake_from_d3(pdev, false);
5916 ixgbe_reset(adapter);
5918 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5921 err = ixgbe_init_interrupt_scheme(adapter);
5922 if (!err && netif_running(netdev))
5923 err = ixgbe_open(netdev);
5930 netif_device_attach(netdev);
5934 #endif /* CONFIG_PM */
5936 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5938 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5939 struct net_device *netdev = adapter->netdev;
5940 struct ixgbe_hw *hw = &adapter->hw;
5942 u32 wufc = adapter->wol;
5947 netif_device_detach(netdev);
5950 if (netif_running(netdev))
5951 ixgbe_close_suspend(adapter);
5954 ixgbe_clear_interrupt_scheme(adapter);
5957 retval = pci_save_state(pdev);
5962 if (hw->mac.ops.stop_link_on_d3)
5963 hw->mac.ops.stop_link_on_d3(hw);
5966 ixgbe_set_rx_mode(netdev);
5968 /* enable the optics for 82599 SFP+ fiber as we can WoL */
5969 if (hw->mac.ops.enable_tx_laser)
5970 hw->mac.ops.enable_tx_laser(hw);
5972 /* turn on all-multi mode if wake on multicast is enabled */
5973 if (wufc & IXGBE_WUFC_MC) {
5974 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5975 fctrl |= IXGBE_FCTRL_MPE;
5976 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5979 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5980 ctrl |= IXGBE_CTRL_GIO_DIS;
5981 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5983 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5985 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5986 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5989 switch (hw->mac.type) {
5990 case ixgbe_mac_82598EB:
5991 pci_wake_from_d3(pdev, false);
5993 case ixgbe_mac_82599EB:
5994 case ixgbe_mac_X540:
5995 case ixgbe_mac_X550:
5996 case ixgbe_mac_X550EM_x:
5997 pci_wake_from_d3(pdev, !!wufc);
6003 *enable_wake = !!wufc;
6004 if (hw->phy.ops.set_phy_power && !*enable_wake)
6005 hw->phy.ops.set_phy_power(hw, false);
6007 ixgbe_release_hw_control(adapter);
6009 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6010 pci_disable_device(pdev);
6016 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
6021 retval = __ixgbe_shutdown(pdev, &wake);
6026 pci_prepare_to_sleep(pdev);
6028 pci_wake_from_d3(pdev, false);
6029 pci_set_power_state(pdev, PCI_D3hot);
6034 #endif /* CONFIG_PM */
6036 static void ixgbe_shutdown(struct pci_dev *pdev)
6040 __ixgbe_shutdown(pdev, &wake);
6042 if (system_state == SYSTEM_POWER_OFF) {
6043 pci_wake_from_d3(pdev, wake);
6044 pci_set_power_state(pdev, PCI_D3hot);
6049 * ixgbe_update_stats - Update the board statistics counters.
6050 * @adapter: board private structure
6052 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
6054 struct net_device *netdev = adapter->netdev;
6055 struct ixgbe_hw *hw = &adapter->hw;
6056 struct ixgbe_hw_stats *hwstats = &adapter->stats;
6058 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
6059 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
6060 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
6061 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
6063 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6064 test_bit(__IXGBE_RESETTING, &adapter->state))
6067 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
6070 for (i = 0; i < adapter->num_rx_queues; i++) {
6071 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
6072 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
6074 adapter->rsc_total_count = rsc_count;
6075 adapter->rsc_total_flush = rsc_flush;
6078 for (i = 0; i < adapter->num_rx_queues; i++) {
6079 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
6080 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
6081 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
6082 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
6083 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
6084 bytes += rx_ring->stats.bytes;
6085 packets += rx_ring->stats.packets;
6087 adapter->non_eop_descs = non_eop_descs;
6088 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
6089 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
6090 adapter->hw_csum_rx_error = hw_csum_rx_error;
6091 netdev->stats.rx_bytes = bytes;
6092 netdev->stats.rx_packets = packets;
6096 /* gather some stats to the adapter struct that are per queue */
6097 for (i = 0; i < adapter->num_tx_queues; i++) {
6098 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6099 restart_queue += tx_ring->tx_stats.restart_queue;
6100 tx_busy += tx_ring->tx_stats.tx_busy;
6101 bytes += tx_ring->stats.bytes;
6102 packets += tx_ring->stats.packets;
6104 adapter->restart_queue = restart_queue;
6105 adapter->tx_busy = tx_busy;
6106 netdev->stats.tx_bytes = bytes;
6107 netdev->stats.tx_packets = packets;
6109 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
6111 /* 8 register reads */
6112 for (i = 0; i < 8; i++) {
6113 /* for packet buffers not used, the register should read 0 */
6114 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
6116 hwstats->mpc[i] += mpc;
6117 total_mpc += hwstats->mpc[i];
6118 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
6119 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
6120 switch (hw->mac.type) {
6121 case ixgbe_mac_82598EB:
6122 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
6123 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
6124 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
6125 hwstats->pxonrxc[i] +=
6126 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
6128 case ixgbe_mac_82599EB:
6129 case ixgbe_mac_X540:
6130 case ixgbe_mac_X550:
6131 case ixgbe_mac_X550EM_x:
6132 hwstats->pxonrxc[i] +=
6133 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
6140 /*16 register reads */
6141 for (i = 0; i < 16; i++) {
6142 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
6143 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
6144 if ((hw->mac.type == ixgbe_mac_82599EB) ||
6145 (hw->mac.type == ixgbe_mac_X540) ||
6146 (hw->mac.type == ixgbe_mac_X550) ||
6147 (hw->mac.type == ixgbe_mac_X550EM_x)) {
6148 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
6149 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
6150 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
6151 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
6155 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
6156 /* work around hardware counting issue */
6157 hwstats->gprc -= missed_rx;
6159 ixgbe_update_xoff_received(adapter);
6161 /* 82598 hardware only has a 32 bit counter in the high register */
6162 switch (hw->mac.type) {
6163 case ixgbe_mac_82598EB:
6164 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
6165 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
6166 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
6167 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
6169 case ixgbe_mac_X540:
6170 case ixgbe_mac_X550:
6171 case ixgbe_mac_X550EM_x:
6172 /* OS2BMC stats are X540 and later */
6173 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
6174 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
6175 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
6176 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
6177 case ixgbe_mac_82599EB:
6178 for (i = 0; i < 16; i++)
6179 adapter->hw_rx_no_dma_resources +=
6180 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
6181 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
6182 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
6183 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
6184 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
6185 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
6186 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
6187 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
6188 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
6189 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
6191 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
6192 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
6193 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
6194 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
6195 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
6196 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
6197 /* Add up per cpu counters for total ddp aloc fail */
6198 if (adapter->fcoe.ddp_pool) {
6199 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
6200 struct ixgbe_fcoe_ddp_pool *ddp_pool;
6202 u64 noddp = 0, noddp_ext_buff = 0;
6203 for_each_possible_cpu(cpu) {
6204 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
6205 noddp += ddp_pool->noddp;
6206 noddp_ext_buff += ddp_pool->noddp_ext_buff;
6208 hwstats->fcoe_noddp = noddp;
6209 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
6211 #endif /* IXGBE_FCOE */
6216 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
6217 hwstats->bprc += bprc;
6218 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
6219 if (hw->mac.type == ixgbe_mac_82598EB)
6220 hwstats->mprc -= bprc;
6221 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
6222 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
6223 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
6224 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
6225 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
6226 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
6227 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
6228 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
6229 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
6230 hwstats->lxontxc += lxon;
6231 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
6232 hwstats->lxofftxc += lxoff;
6233 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
6234 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
6236 * 82598 errata - tx of flow control packets is included in tx counters
6238 xon_off_tot = lxon + lxoff;
6239 hwstats->gptc -= xon_off_tot;
6240 hwstats->mptc -= xon_off_tot;
6241 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
6242 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
6243 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
6244 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
6245 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
6246 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
6247 hwstats->ptc64 -= xon_off_tot;
6248 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
6249 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
6250 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
6251 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
6252 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
6253 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
6255 /* Fill out the OS statistics structure */
6256 netdev->stats.multicast = hwstats->mprc;
6259 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
6260 netdev->stats.rx_dropped = 0;
6261 netdev->stats.rx_length_errors = hwstats->rlec;
6262 netdev->stats.rx_crc_errors = hwstats->crcerrs;
6263 netdev->stats.rx_missed_errors = total_mpc;
6267 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
6268 * @adapter: pointer to the device adapter structure
6270 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
6272 struct ixgbe_hw *hw = &adapter->hw;
6275 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
6278 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6280 /* if interface is down do nothing */
6281 if (test_bit(__IXGBE_DOWN, &adapter->state))
6284 /* do nothing if we are not using signature filters */
6285 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
6288 adapter->fdir_overflow++;
6290 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
6291 for (i = 0; i < adapter->num_tx_queues; i++)
6292 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
6293 &(adapter->tx_ring[i]->state));
6294 /* re-enable flow director interrupts */
6295 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
6297 e_err(probe, "failed to finish FDIR re-initialization, "
6298 "ignored adding FDIR ATR filters\n");
6303 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
6304 * @adapter: pointer to the device adapter structure
6306 * This function serves two purposes. First it strobes the interrupt lines
6307 * in order to make certain interrupts are occurring. Secondly it sets the
6308 * bits needed to check for TX hangs. As a result we should immediately
6309 * determine if a hang has occurred.
6311 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
6313 struct ixgbe_hw *hw = &adapter->hw;
6317 /* If we're down, removing or resetting, just bail */
6318 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6319 test_bit(__IXGBE_REMOVING, &adapter->state) ||
6320 test_bit(__IXGBE_RESETTING, &adapter->state))
6323 /* Force detection of hung controller */
6324 if (netif_carrier_ok(adapter->netdev)) {
6325 for (i = 0; i < adapter->num_tx_queues; i++)
6326 set_check_for_tx_hang(adapter->tx_ring[i]);
6329 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6331 * for legacy and MSI interrupts don't set any bits
6332 * that are enabled for EIAM, because this operation
6333 * would set *both* EIMS and EICS for any bit in EIAM
6335 IXGBE_WRITE_REG(hw, IXGBE_EICS,
6336 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
6338 /* get one bit for every active tx/rx interrupt vector */
6339 for (i = 0; i < adapter->num_q_vectors; i++) {
6340 struct ixgbe_q_vector *qv = adapter->q_vector[i];
6341 if (qv->rx.ring || qv->tx.ring)
6342 eics |= ((u64)1 << i);
6346 /* Cause software interrupt to ensure rings are cleaned */
6347 ixgbe_irq_rearm_queues(adapter, eics);
6351 * ixgbe_watchdog_update_link - update the link status
6352 * @adapter: pointer to the device adapter structure
6353 * @link_speed: pointer to a u32 to store the link_speed
6355 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
6357 struct ixgbe_hw *hw = &adapter->hw;
6358 u32 link_speed = adapter->link_speed;
6359 bool link_up = adapter->link_up;
6360 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
6362 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6365 if (hw->mac.ops.check_link) {
6366 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6368 /* always assume link is up, if no check link function */
6369 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
6373 if (adapter->ixgbe_ieee_pfc)
6374 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
6376 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
6377 hw->mac.ops.fc_enable(hw);
6378 ixgbe_set_rx_drop_en(adapter);
6382 time_after(jiffies, (adapter->link_check_timeout +
6383 IXGBE_TRY_LINK_TIMEOUT))) {
6384 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6385 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6386 IXGBE_WRITE_FLUSH(hw);
6389 adapter->link_up = link_up;
6390 adapter->link_speed = link_speed;
6393 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
6395 #ifdef CONFIG_IXGBE_DCB
6396 struct net_device *netdev = adapter->netdev;
6397 struct dcb_app app = {
6398 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
6403 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
6404 up = dcb_ieee_getapp_mask(netdev, &app);
6406 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
6411 * ixgbe_watchdog_link_is_up - update netif_carrier status and
6412 * print link up message
6413 * @adapter: pointer to the device adapter structure
6415 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
6417 struct net_device *netdev = adapter->netdev;
6418 struct ixgbe_hw *hw = &adapter->hw;
6419 struct net_device *upper;
6420 struct list_head *iter;
6421 u32 link_speed = adapter->link_speed;
6422 const char *speed_str;
6423 bool flow_rx, flow_tx;
6425 /* only continue if link was previously down */
6426 if (netif_carrier_ok(netdev))
6429 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
6431 switch (hw->mac.type) {
6432 case ixgbe_mac_82598EB: {
6433 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6434 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6435 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6436 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6439 case ixgbe_mac_X540:
6440 case ixgbe_mac_X550:
6441 case ixgbe_mac_X550EM_x:
6442 case ixgbe_mac_82599EB: {
6443 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6444 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6445 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6446 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6455 adapter->last_rx_ptp_check = jiffies;
6457 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6458 ixgbe_ptp_start_cyclecounter(adapter);
6460 switch (link_speed) {
6461 case IXGBE_LINK_SPEED_10GB_FULL:
6462 speed_str = "10 Gbps";
6464 case IXGBE_LINK_SPEED_2_5GB_FULL:
6465 speed_str = "2.5 Gbps";
6467 case IXGBE_LINK_SPEED_1GB_FULL:
6468 speed_str = "1 Gbps";
6470 case IXGBE_LINK_SPEED_100_FULL:
6471 speed_str = "100 Mbps";
6474 speed_str = "unknown speed";
6477 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
6478 ((flow_rx && flow_tx) ? "RX/TX" :
6480 (flow_tx ? "TX" : "None"))));
6482 netif_carrier_on(netdev);
6483 ixgbe_check_vf_rate_limit(adapter);
6485 /* enable transmits */
6486 netif_tx_wake_all_queues(adapter->netdev);
6488 /* enable any upper devices */
6490 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
6491 if (netif_is_macvlan(upper)) {
6492 struct macvlan_dev *vlan = netdev_priv(upper);
6495 netif_tx_wake_all_queues(upper);
6500 /* update the default user priority for VFs */
6501 ixgbe_update_default_up(adapter);
6503 /* ping all the active vfs to let them know link has changed */
6504 ixgbe_ping_all_vfs(adapter);
6508 * ixgbe_watchdog_link_is_down - update netif_carrier status and
6509 * print link down message
6510 * @adapter: pointer to the adapter structure
6512 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
6514 struct net_device *netdev = adapter->netdev;
6515 struct ixgbe_hw *hw = &adapter->hw;
6517 adapter->link_up = false;
6518 adapter->link_speed = 0;
6520 /* only continue if link was up previously */
6521 if (!netif_carrier_ok(netdev))
6524 /* poll for SFP+ cable when link is down */
6525 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
6526 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
6528 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6529 ixgbe_ptp_start_cyclecounter(adapter);
6531 e_info(drv, "NIC Link is Down\n");
6532 netif_carrier_off(netdev);
6534 /* ping all the active vfs to let them know link has changed */
6535 ixgbe_ping_all_vfs(adapter);
6538 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
6542 for (i = 0; i < adapter->num_tx_queues; i++) {
6543 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6545 if (tx_ring->next_to_use != tx_ring->next_to_clean)
6552 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
6554 struct ixgbe_hw *hw = &adapter->hw;
6555 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
6556 u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
6560 if (!adapter->num_vfs)
6563 /* resetting the PF is only needed for MAC before X550 */
6564 if (hw->mac.type >= ixgbe_mac_X550)
6567 for (i = 0; i < adapter->num_vfs; i++) {
6568 for (j = 0; j < q_per_pool; j++) {
6571 h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
6572 t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
6583 * ixgbe_watchdog_flush_tx - flush queues on link down
6584 * @adapter: pointer to the device adapter structure
6586 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
6588 if (!netif_carrier_ok(adapter->netdev)) {
6589 if (ixgbe_ring_tx_pending(adapter) ||
6590 ixgbe_vf_tx_pending(adapter)) {
6591 /* We've lost link, so the controller stops DMA,
6592 * but we've got queued Tx work that's never going
6593 * to get done, so reset controller to flush Tx.
6594 * (Do the reset outside of interrupt context).
6596 e_warn(drv, "initiating reset to clear Tx work after link loss\n");
6597 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
6602 #ifdef CONFIG_PCI_IOV
6603 static inline void ixgbe_issue_vf_flr(struct ixgbe_adapter *adapter,
6604 struct pci_dev *vfdev)
6606 if (!pci_wait_for_pending_transaction(vfdev))
6607 e_dev_warn("Issuing VFLR with pending transactions\n");
6609 e_dev_err("Issuing VFLR for VF %s\n", pci_name(vfdev));
6610 pcie_capability_set_word(vfdev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
6615 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6617 struct ixgbe_hw *hw = &adapter->hw;
6618 struct pci_dev *pdev = adapter->pdev;
6619 struct pci_dev *vfdev;
6622 unsigned short vf_id;
6624 if (!(netif_carrier_ok(adapter->netdev)))
6627 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6628 if (gpc) /* If incrementing then no need for the check below */
6630 /* Check to see if a bad DMA write target from an errant or
6631 * malicious VF has caused a PCIe error. If so then we can
6632 * issue a VFLR to the offending VF(s) and then resume without
6633 * requesting a full slot reset.
6639 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
6643 /* get the device ID for the VF */
6644 pci_read_config_word(pdev, pos + PCI_SRIOV_VF_DID, &vf_id);
6646 /* check status reg for all VFs owned by this PF */
6647 vfdev = pci_get_device(pdev->vendor, vf_id, NULL);
6649 if (vfdev->is_virtfn && (vfdev->physfn == pdev)) {
6652 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
6653 if (status_reg & PCI_STATUS_REC_MASTER_ABORT)
6655 ixgbe_issue_vf_flr(adapter, vfdev);
6658 vfdev = pci_get_device(pdev->vendor, vf_id, vfdev);
6662 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6666 /* Do not perform spoof check for 82598 or if not in IOV mode */
6667 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
6668 adapter->num_vfs == 0)
6671 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6674 * ssvpc register is cleared on read, if zero then no
6675 * spoofed packets in the last interval.
6680 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
6683 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
6688 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
6691 #endif /* CONFIG_PCI_IOV */
6695 * ixgbe_watchdog_subtask - check and bring link up
6696 * @adapter: pointer to the device adapter structure
6698 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6700 /* if interface is down, removing or resetting, do nothing */
6701 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6702 test_bit(__IXGBE_REMOVING, &adapter->state) ||
6703 test_bit(__IXGBE_RESETTING, &adapter->state))
6706 ixgbe_watchdog_update_link(adapter);
6708 if (adapter->link_up)
6709 ixgbe_watchdog_link_is_up(adapter);
6711 ixgbe_watchdog_link_is_down(adapter);
6713 ixgbe_check_for_bad_vf(adapter);
6714 ixgbe_spoof_check(adapter);
6715 ixgbe_update_stats(adapter);
6717 ixgbe_watchdog_flush_tx(adapter);
6721 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6722 * @adapter: the ixgbe adapter structure
6724 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6726 struct ixgbe_hw *hw = &adapter->hw;
6729 /* not searching for SFP so there is nothing to do here */
6730 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6731 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6734 if (adapter->sfp_poll_time &&
6735 time_after(adapter->sfp_poll_time, jiffies))
6736 return; /* If not yet time to poll for SFP */
6738 /* someone else is in init, wait until next service event */
6739 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6742 adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
6744 err = hw->phy.ops.identify_sfp(hw);
6745 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6748 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6749 /* If no cable is present, then we need to reset
6750 * the next time we find a good cable. */
6751 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6758 /* exit if reset not needed */
6759 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6762 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6765 * A module may be identified correctly, but the EEPROM may not have
6766 * support for that module. setup_sfp() will fail in that case, so
6767 * we should not allow that module to load.
6769 if (hw->mac.type == ixgbe_mac_82598EB)
6770 err = hw->phy.ops.reset(hw);
6772 err = hw->mac.ops.setup_sfp(hw);
6774 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6777 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6778 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6781 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6783 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6784 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6785 e_dev_err("failed to initialize because an unsupported "
6786 "SFP+ module type was detected.\n");
6787 e_dev_err("Reload the driver after installing a "
6788 "supported module.\n");
6789 unregister_netdev(adapter->netdev);
6794 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6795 * @adapter: the ixgbe adapter structure
6797 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6799 struct ixgbe_hw *hw = &adapter->hw;
6801 bool autoneg = false;
6803 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6806 /* someone else is in init, wait until next service event */
6807 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6810 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6812 speed = hw->phy.autoneg_advertised;
6813 if ((!speed) && (hw->mac.ops.get_link_capabilities)) {
6814 hw->mac.ops.get_link_capabilities(hw, &speed, &autoneg);
6816 /* setup the highest link when no autoneg */
6818 if (speed & IXGBE_LINK_SPEED_10GB_FULL)
6819 speed = IXGBE_LINK_SPEED_10GB_FULL;
6823 if (hw->mac.ops.setup_link)
6824 hw->mac.ops.setup_link(hw, speed, true);
6826 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6827 adapter->link_check_timeout = jiffies;
6828 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6832 * ixgbe_service_timer - Timer Call-back
6833 * @data: pointer to adapter cast into an unsigned long
6835 static void ixgbe_service_timer(unsigned long data)
6837 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6838 unsigned long next_event_offset;
6840 /* poll faster when waiting for link */
6841 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6842 next_event_offset = HZ / 10;
6844 next_event_offset = HZ * 2;
6846 /* Reset the timer */
6847 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6849 ixgbe_service_event_schedule(adapter);
6852 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
6854 struct ixgbe_hw *hw = &adapter->hw;
6857 if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
6860 adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
6862 if (!hw->phy.ops.handle_lasi)
6865 status = hw->phy.ops.handle_lasi(&adapter->hw);
6866 if (status != IXGBE_ERR_OVERTEMP)
6869 e_crit(drv, "%s\n", ixgbe_overheat_msg);
6872 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6874 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6877 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6879 /* If we're already down, removing or resetting, just bail */
6880 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6881 test_bit(__IXGBE_REMOVING, &adapter->state) ||
6882 test_bit(__IXGBE_RESETTING, &adapter->state))
6885 ixgbe_dump(adapter);
6886 netdev_err(adapter->netdev, "Reset adapter\n");
6887 adapter->tx_timeout_count++;
6890 ixgbe_reinit_locked(adapter);
6895 * ixgbe_service_task - manages and runs subtasks
6896 * @work: pointer to work_struct containing our data
6898 static void ixgbe_service_task(struct work_struct *work)
6900 struct ixgbe_adapter *adapter = container_of(work,
6901 struct ixgbe_adapter,
6903 if (ixgbe_removed(adapter->hw.hw_addr)) {
6904 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
6906 ixgbe_down(adapter);
6909 ixgbe_service_event_complete(adapter);
6912 #ifdef CONFIG_IXGBE_VXLAN
6913 if (adapter->flags2 & IXGBE_FLAG2_VXLAN_REREG_NEEDED) {
6914 adapter->flags2 &= ~IXGBE_FLAG2_VXLAN_REREG_NEEDED;
6915 vxlan_get_rx_port(adapter->netdev);
6917 #endif /* CONFIG_IXGBE_VXLAN */
6918 ixgbe_reset_subtask(adapter);
6919 ixgbe_phy_interrupt_subtask(adapter);
6920 ixgbe_sfp_detection_subtask(adapter);
6921 ixgbe_sfp_link_config_subtask(adapter);
6922 ixgbe_check_overtemp_subtask(adapter);
6923 ixgbe_watchdog_subtask(adapter);
6924 ixgbe_fdir_reinit_subtask(adapter);
6925 ixgbe_check_hang_subtask(adapter);
6927 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
6928 ixgbe_ptp_overflow_check(adapter);
6929 ixgbe_ptp_rx_hang(adapter);
6932 ixgbe_service_event_complete(adapter);
6935 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
6936 struct ixgbe_tx_buffer *first,
6939 struct sk_buff *skb = first->skb;
6940 u32 vlan_macip_lens, type_tucmd;
6941 u32 mss_l4len_idx, l4len;
6944 if (skb->ip_summed != CHECKSUM_PARTIAL)
6947 if (!skb_is_gso(skb))
6950 err = skb_cow_head(skb, 0);
6954 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6955 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6957 if (first->protocol == htons(ETH_P_IP)) {
6958 struct iphdr *iph = ip_hdr(skb);
6961 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6965 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6966 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6967 IXGBE_TX_FLAGS_CSUM |
6968 IXGBE_TX_FLAGS_IPV4;
6969 } else if (skb_is_gso_v6(skb)) {
6970 ipv6_hdr(skb)->payload_len = 0;
6971 tcp_hdr(skb)->check =
6972 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6973 &ipv6_hdr(skb)->daddr,
6975 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6976 IXGBE_TX_FLAGS_CSUM;
6979 /* compute header lengths */
6980 l4len = tcp_hdrlen(skb);
6981 *hdr_len = skb_transport_offset(skb) + l4len;
6983 /* update gso size and bytecount with header size */
6984 first->gso_segs = skb_shinfo(skb)->gso_segs;
6985 first->bytecount += (first->gso_segs - 1) * *hdr_len;
6987 /* mss_l4len_id: use 0 as index for TSO */
6988 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6989 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6991 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6992 vlan_macip_lens = skb_network_header_len(skb);
6993 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6994 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6996 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
7002 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
7003 struct ixgbe_tx_buffer *first)
7005 struct sk_buff *skb = first->skb;
7006 u32 vlan_macip_lens = 0;
7007 u32 mss_l4len_idx = 0;
7010 if (skb->ip_summed != CHECKSUM_PARTIAL) {
7011 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
7012 !(first->tx_flags & IXGBE_TX_FLAGS_CC))
7014 vlan_macip_lens = skb_network_offset(skb) <<
7015 IXGBE_ADVTXD_MACLEN_SHIFT;
7020 struct ipv6hdr *ipv6;
7024 struct tcphdr *tcphdr;
7028 if (skb->encapsulation) {
7029 network_hdr.raw = skb_inner_network_header(skb);
7030 transport_hdr.raw = skb_inner_transport_header(skb);
7031 vlan_macip_lens = skb_inner_network_offset(skb) <<
7032 IXGBE_ADVTXD_MACLEN_SHIFT;
7034 network_hdr.raw = skb_network_header(skb);
7035 transport_hdr.raw = skb_transport_header(skb);
7036 vlan_macip_lens = skb_network_offset(skb) <<
7037 IXGBE_ADVTXD_MACLEN_SHIFT;
7040 /* use first 4 bits to determine IP version */
7041 switch (network_hdr.ipv4->version) {
7043 vlan_macip_lens |= transport_hdr.raw - network_hdr.raw;
7044 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7045 l4_hdr = network_hdr.ipv4->protocol;
7048 vlan_macip_lens |= transport_hdr.raw - network_hdr.raw;
7049 l4_hdr = network_hdr.ipv6->nexthdr;
7052 if (unlikely(net_ratelimit())) {
7053 dev_warn(tx_ring->dev,
7054 "partial checksum but version=%d\n",
7055 network_hdr.ipv4->version);
7061 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
7062 mss_l4len_idx = (transport_hdr.tcphdr->doff * 4) <<
7063 IXGBE_ADVTXD_L4LEN_SHIFT;
7066 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
7067 mss_l4len_idx = sizeof(struct sctphdr) <<
7068 IXGBE_ADVTXD_L4LEN_SHIFT;
7071 mss_l4len_idx = sizeof(struct udphdr) <<
7072 IXGBE_ADVTXD_L4LEN_SHIFT;
7075 if (unlikely(net_ratelimit())) {
7076 dev_warn(tx_ring->dev,
7077 "partial checksum but l4 proto=%x!\n",
7083 /* update TX checksum flag */
7084 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
7087 /* vlan_macip_lens: MACLEN, VLAN tag */
7088 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
7090 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
7091 type_tucmd, mss_l4len_idx);
7094 #define IXGBE_SET_FLAG(_input, _flag, _result) \
7095 ((_flag <= _result) ? \
7096 ((u32)(_input & _flag) * (_result / _flag)) : \
7097 ((u32)(_input & _flag) / (_flag / _result)))
7099 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
7101 /* set type for advanced descriptor with frame checksum insertion */
7102 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
7103 IXGBE_ADVTXD_DCMD_DEXT |
7104 IXGBE_ADVTXD_DCMD_IFCS;
7106 /* set HW vlan bit if vlan is present */
7107 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
7108 IXGBE_ADVTXD_DCMD_VLE);
7110 /* set segmentation enable bits for TSO/FSO */
7111 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
7112 IXGBE_ADVTXD_DCMD_TSE);
7114 /* set timestamp bit if present */
7115 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
7116 IXGBE_ADVTXD_MAC_TSTAMP);
7118 /* insert frame checksum */
7119 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
7124 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
7125 u32 tx_flags, unsigned int paylen)
7127 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
7129 /* enable L4 checksum for TSO and TX checksum offload */
7130 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7131 IXGBE_TX_FLAGS_CSUM,
7132 IXGBE_ADVTXD_POPTS_TXSM);
7134 /* enble IPv4 checksum for TSO */
7135 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7136 IXGBE_TX_FLAGS_IPV4,
7137 IXGBE_ADVTXD_POPTS_IXSM);
7140 * Check Context must be set if Tx switch is enabled, which it
7141 * always is for case where virtual functions are running
7143 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7147 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
7150 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7152 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
7154 /* Herbert's original patch had:
7155 * smp_mb__after_netif_stop_queue();
7156 * but since that doesn't exist yet, just open code it.
7160 /* We need to check again in a case another CPU has just
7161 * made room available.
7163 if (likely(ixgbe_desc_unused(tx_ring) < size))
7166 /* A reprieve! - use start_queue because it doesn't call schedule */
7167 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
7168 ++tx_ring->tx_stats.restart_queue;
7172 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7174 if (likely(ixgbe_desc_unused(tx_ring) >= size))
7177 return __ixgbe_maybe_stop_tx(tx_ring, size);
7180 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
7183 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
7184 struct ixgbe_tx_buffer *first,
7187 struct sk_buff *skb = first->skb;
7188 struct ixgbe_tx_buffer *tx_buffer;
7189 union ixgbe_adv_tx_desc *tx_desc;
7190 struct skb_frag_struct *frag;
7192 unsigned int data_len, size;
7193 u32 tx_flags = first->tx_flags;
7194 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
7195 u16 i = tx_ring->next_to_use;
7197 tx_desc = IXGBE_TX_DESC(tx_ring, i);
7199 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
7201 size = skb_headlen(skb);
7202 data_len = skb->data_len;
7205 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
7206 if (data_len < sizeof(struct fcoe_crc_eof)) {
7207 size -= sizeof(struct fcoe_crc_eof) - data_len;
7210 data_len -= sizeof(struct fcoe_crc_eof);
7215 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
7219 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
7220 if (dma_mapping_error(tx_ring->dev, dma))
7223 /* record length, and DMA address */
7224 dma_unmap_len_set(tx_buffer, len, size);
7225 dma_unmap_addr_set(tx_buffer, dma, dma);
7227 tx_desc->read.buffer_addr = cpu_to_le64(dma);
7229 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
7230 tx_desc->read.cmd_type_len =
7231 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
7235 if (i == tx_ring->count) {
7236 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
7239 tx_desc->read.olinfo_status = 0;
7241 dma += IXGBE_MAX_DATA_PER_TXD;
7242 size -= IXGBE_MAX_DATA_PER_TXD;
7244 tx_desc->read.buffer_addr = cpu_to_le64(dma);
7247 if (likely(!data_len))
7250 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
7254 if (i == tx_ring->count) {
7255 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
7258 tx_desc->read.olinfo_status = 0;
7261 size = min_t(unsigned int, data_len, skb_frag_size(frag));
7263 size = skb_frag_size(frag);
7267 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
7270 tx_buffer = &tx_ring->tx_buffer_info[i];
7273 /* write last descriptor with RS and EOP bits */
7274 cmd_type |= size | IXGBE_TXD_CMD;
7275 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
7277 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
7279 /* set the timestamp */
7280 first->time_stamp = jiffies;
7283 * Force memory writes to complete before letting h/w know there
7284 * are new descriptors to fetch. (Only applicable for weak-ordered
7285 * memory model archs, such as IA-64).
7287 * We also need this memory barrier to make certain all of the
7288 * status bits have been updated before next_to_watch is written.
7292 /* set next_to_watch value indicating a packet is present */
7293 first->next_to_watch = tx_desc;
7296 if (i == tx_ring->count)
7299 tx_ring->next_to_use = i;
7301 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
7303 if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
7304 writel(i, tx_ring->tail);
7306 /* we need this if more than one processor can write to our tail
7307 * at a time, it synchronizes IO on IA64/Altix systems
7314 dev_err(tx_ring->dev, "TX DMA map failed\n");
7316 /* clear dma mappings for failed tx_buffer_info map */
7318 tx_buffer = &tx_ring->tx_buffer_info[i];
7319 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
7320 if (tx_buffer == first)
7327 tx_ring->next_to_use = i;
7330 static void ixgbe_atr(struct ixgbe_ring *ring,
7331 struct ixgbe_tx_buffer *first)
7333 struct ixgbe_q_vector *q_vector = ring->q_vector;
7334 union ixgbe_atr_hash_dword input = { .dword = 0 };
7335 union ixgbe_atr_hash_dword common = { .dword = 0 };
7337 unsigned char *network;
7339 struct ipv6hdr *ipv6;
7342 struct sk_buff *skb;
7343 #ifdef CONFIG_IXGBE_VXLAN
7345 #endif /* CONFIG_IXGBE_VXLAN */
7348 /* if ring doesn't have a interrupt vector, cannot perform ATR */
7352 /* do nothing if sampling is disabled */
7353 if (!ring->atr_sample_rate)
7358 /* snag network header to get L4 type and address */
7360 hdr.network = skb_network_header(skb);
7361 if (skb->encapsulation) {
7362 #ifdef CONFIG_IXGBE_VXLAN
7363 struct ixgbe_adapter *adapter = q_vector->adapter;
7365 if (!adapter->vxlan_port)
7367 if (first->protocol != htons(ETH_P_IP) ||
7368 hdr.ipv4->version != IPVERSION ||
7369 hdr.ipv4->protocol != IPPROTO_UDP) {
7372 if (ntohs(udp_hdr(skb)->dest) != adapter->vxlan_port)
7375 hdr.network = skb_inner_network_header(skb);
7376 th = inner_tcp_hdr(skb);
7379 #endif /* CONFIG_IXGBE_VXLAN */
7381 /* Currently only IPv4/IPv6 with TCP is supported */
7382 if ((first->protocol != htons(ETH_P_IPV6) ||
7383 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
7384 (first->protocol != htons(ETH_P_IP) ||
7385 hdr.ipv4->protocol != IPPROTO_TCP))
7390 /* skip this packet since it is invalid or the socket is closing */
7394 /* sample on all syn packets or once every atr sample count */
7395 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
7398 /* reset sample count */
7399 ring->atr_count = 0;
7401 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
7404 * src and dst are inverted, think how the receiver sees them
7406 * The input is broken into two sections, a non-compressed section
7407 * containing vm_pool, vlan_id, and flow_type. The rest of the data
7408 * is XORed together and stored in the compressed dword.
7410 input.formatted.vlan_id = vlan_id;
7413 * since src port and flex bytes occupy the same word XOR them together
7414 * and write the value to source port portion of compressed dword
7416 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
7417 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
7419 common.port.src ^= th->dest ^ first->protocol;
7420 common.port.dst ^= th->source;
7422 if (first->protocol == htons(ETH_P_IP)) {
7423 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
7424 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
7426 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
7427 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
7428 hdr.ipv6->saddr.s6_addr32[1] ^
7429 hdr.ipv6->saddr.s6_addr32[2] ^
7430 hdr.ipv6->saddr.s6_addr32[3] ^
7431 hdr.ipv6->daddr.s6_addr32[0] ^
7432 hdr.ipv6->daddr.s6_addr32[1] ^
7433 hdr.ipv6->daddr.s6_addr32[2] ^
7434 hdr.ipv6->daddr.s6_addr32[3];
7437 #ifdef CONFIG_IXGBE_VXLAN
7439 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
7440 #endif /* CONFIG_IXGBE_VXLAN */
7442 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
7443 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
7444 input, common, ring->queue_index);
7447 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
7448 void *accel_priv, select_queue_fallback_t fallback)
7450 struct ixgbe_fwd_adapter *fwd_adapter = accel_priv;
7452 struct ixgbe_adapter *adapter;
7453 struct ixgbe_ring_feature *f;
7458 return skb->queue_mapping + fwd_adapter->tx_base_queue;
7463 * only execute the code below if protocol is FCoE
7464 * or FIP and we have FCoE enabled on the adapter
7466 switch (vlan_get_protocol(skb)) {
7467 case htons(ETH_P_FCOE):
7468 case htons(ETH_P_FIP):
7469 adapter = netdev_priv(dev);
7471 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7474 return fallback(dev, skb);
7477 f = &adapter->ring_feature[RING_F_FCOE];
7479 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
7482 while (txq >= f->indices)
7485 return txq + f->offset;
7487 return fallback(dev, skb);
7491 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
7492 struct ixgbe_adapter *adapter,
7493 struct ixgbe_ring *tx_ring)
7495 struct ixgbe_tx_buffer *first;
7499 u16 count = TXD_USE_COUNT(skb_headlen(skb));
7500 __be16 protocol = skb->protocol;
7504 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
7505 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
7506 * + 2 desc gap to keep tail from touching head,
7507 * + 1 desc for context descriptor,
7508 * otherwise try next time
7510 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
7511 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
7513 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
7514 tx_ring->tx_stats.tx_busy++;
7515 return NETDEV_TX_BUSY;
7518 /* record the location of the first descriptor for this packet */
7519 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
7521 first->bytecount = skb->len;
7522 first->gso_segs = 1;
7524 /* if we have a HW VLAN tag being added default to the HW one */
7525 if (skb_vlan_tag_present(skb)) {
7526 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
7527 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7528 /* else if it is a SW VLAN check the next protocol and store the tag */
7529 } else if (protocol == htons(ETH_P_8021Q)) {
7530 struct vlan_hdr *vhdr, _vhdr;
7531 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
7535 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
7536 IXGBE_TX_FLAGS_VLAN_SHIFT;
7537 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
7539 protocol = vlan_get_protocol(skb);
7541 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
7542 adapter->ptp_clock &&
7543 !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
7545 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
7546 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
7548 /* schedule check for Tx timestamp */
7549 adapter->ptp_tx_skb = skb_get(skb);
7550 adapter->ptp_tx_start = jiffies;
7551 schedule_work(&adapter->ptp_tx_work);
7554 skb_tx_timestamp(skb);
7556 #ifdef CONFIG_PCI_IOV
7558 * Use the l2switch_enable flag - would be false if the DMA
7559 * Tx switch had been disabled.
7561 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7562 tx_flags |= IXGBE_TX_FLAGS_CC;
7565 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
7566 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7567 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
7568 (skb->priority != TC_PRIO_CONTROL))) {
7569 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
7570 tx_flags |= (skb->priority & 0x7) <<
7571 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
7572 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
7573 struct vlan_ethhdr *vhdr;
7575 if (skb_cow_head(skb, 0))
7577 vhdr = (struct vlan_ethhdr *)skb->data;
7578 vhdr->h_vlan_TCI = htons(tx_flags >>
7579 IXGBE_TX_FLAGS_VLAN_SHIFT);
7581 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7585 /* record initial flags and protocol */
7586 first->tx_flags = tx_flags;
7587 first->protocol = protocol;
7590 /* setup tx offload for FCoE */
7591 if ((protocol == htons(ETH_P_FCOE)) &&
7592 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
7593 tso = ixgbe_fso(tx_ring, first, &hdr_len);
7600 #endif /* IXGBE_FCOE */
7601 tso = ixgbe_tso(tx_ring, first, &hdr_len);
7605 ixgbe_tx_csum(tx_ring, first);
7607 /* add the ATR filter if ATR is on */
7608 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
7609 ixgbe_atr(tx_ring, first);
7613 #endif /* IXGBE_FCOE */
7614 ixgbe_tx_map(tx_ring, first, hdr_len);
7616 return NETDEV_TX_OK;
7619 dev_kfree_skb_any(first->skb);
7622 return NETDEV_TX_OK;
7625 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
7626 struct net_device *netdev,
7627 struct ixgbe_ring *ring)
7629 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7630 struct ixgbe_ring *tx_ring;
7633 * The minimum packet size for olinfo paylen is 17 so pad the skb
7634 * in order to meet this minimum size requirement.
7636 if (skb_put_padto(skb, 17))
7637 return NETDEV_TX_OK;
7639 tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
7641 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
7644 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
7645 struct net_device *netdev)
7647 return __ixgbe_xmit_frame(skb, netdev, NULL);
7651 * ixgbe_set_mac - Change the Ethernet Address of the NIC
7652 * @netdev: network interface device structure
7653 * @p: pointer to an address structure
7655 * Returns 0 on success, negative on failure
7657 static int ixgbe_set_mac(struct net_device *netdev, void *p)
7659 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7660 struct ixgbe_hw *hw = &adapter->hw;
7661 struct sockaddr *addr = p;
7664 if (!is_valid_ether_addr(addr->sa_data))
7665 return -EADDRNOTAVAIL;
7667 ixgbe_del_mac_filter(adapter, hw->mac.addr, VMDQ_P(0));
7668 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
7669 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
7671 ret = ixgbe_add_mac_filter(adapter, hw->mac.addr, VMDQ_P(0));
7672 return ret > 0 ? 0 : ret;
7676 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
7678 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7679 struct ixgbe_hw *hw = &adapter->hw;
7683 if (prtad != hw->phy.mdio.prtad)
7685 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
7691 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
7692 u16 addr, u16 value)
7694 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7695 struct ixgbe_hw *hw = &adapter->hw;
7697 if (prtad != hw->phy.mdio.prtad)
7699 return hw->phy.ops.write_reg(hw, addr, devad, value);
7702 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
7704 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7708 return ixgbe_ptp_set_ts_config(adapter, req);
7710 return ixgbe_ptp_get_ts_config(adapter, req);
7712 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
7717 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
7719 * @netdev: network interface device structure
7721 * Returns non-zero on failure
7723 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
7726 struct ixgbe_adapter *adapter = netdev_priv(dev);
7727 struct ixgbe_hw *hw = &adapter->hw;
7729 if (is_valid_ether_addr(hw->mac.san_addr)) {
7731 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
7734 /* update SAN MAC vmdq pool selection */
7735 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
7741 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
7743 * @netdev: network interface device structure
7745 * Returns non-zero on failure
7747 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
7750 struct ixgbe_adapter *adapter = netdev_priv(dev);
7751 struct ixgbe_mac_info *mac = &adapter->hw.mac;
7753 if (is_valid_ether_addr(mac->san_addr)) {
7755 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
7761 #ifdef CONFIG_NET_POLL_CONTROLLER
7763 * Polling 'interrupt' - used by things like netconsole to send skbs
7764 * without having to re-enable interrupts. It's not called while
7765 * the interrupt routine is executing.
7767 static void ixgbe_netpoll(struct net_device *netdev)
7769 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7772 /* if interface is down do nothing */
7773 if (test_bit(__IXGBE_DOWN, &adapter->state))
7776 /* loop through and schedule all active queues */
7777 for (i = 0; i < adapter->num_q_vectors; i++)
7778 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
7782 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7783 struct rtnl_link_stats64 *stats)
7785 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7789 for (i = 0; i < adapter->num_rx_queues; i++) {
7790 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_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->rx_packets += packets;
7801 stats->rx_bytes += bytes;
7805 for (i = 0; i < adapter->num_tx_queues; i++) {
7806 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7812 start = u64_stats_fetch_begin_irq(&ring->syncp);
7813 packets = ring->stats.packets;
7814 bytes = ring->stats.bytes;
7815 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
7816 stats->tx_packets += packets;
7817 stats->tx_bytes += bytes;
7821 /* following stats updated by ixgbe_watchdog_task() */
7822 stats->multicast = netdev->stats.multicast;
7823 stats->rx_errors = netdev->stats.rx_errors;
7824 stats->rx_length_errors = netdev->stats.rx_length_errors;
7825 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
7826 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
7830 #ifdef CONFIG_IXGBE_DCB
7832 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7833 * @adapter: pointer to ixgbe_adapter
7834 * @tc: number of traffic classes currently enabled
7836 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7837 * 802.1Q priority maps to a packet buffer that exists.
7839 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
7841 struct ixgbe_hw *hw = &adapter->hw;
7845 /* 82598 have a static priority to TC mapping that can not
7846 * be changed so no validation is needed.
7848 if (hw->mac.type == ixgbe_mac_82598EB)
7851 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
7854 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
7855 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
7857 /* If up2tc is out of bounds default to zero */
7859 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
7863 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
7869 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
7870 * @adapter: Pointer to adapter struct
7872 * Populate the netdev user priority to tc map
7874 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
7876 struct net_device *dev = adapter->netdev;
7877 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
7878 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
7881 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
7884 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
7885 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
7887 tc = ets->prio_tc[prio];
7889 netdev_set_prio_tc_map(dev, prio, tc);
7893 #endif /* CONFIG_IXGBE_DCB */
7895 * ixgbe_setup_tc - configure net_device for multiple traffic classes
7897 * @netdev: net device to configure
7898 * @tc: number of traffic classes to enable
7900 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7902 struct ixgbe_adapter *adapter = netdev_priv(dev);
7903 struct ixgbe_hw *hw = &adapter->hw;
7906 /* Hardware supports up to 8 traffic classes */
7907 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
7910 if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
7913 pools = (find_first_zero_bit(&adapter->fwd_bitmask, 32) > 1);
7914 if (tc && pools && adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS)
7917 /* Hardware has to reinitialize queues and interrupts to
7918 * match packet buffer alignment. Unfortunately, the
7919 * hardware is not flexible enough to do this dynamically.
7921 if (netif_running(dev))
7923 ixgbe_clear_interrupt_scheme(adapter);
7925 #ifdef CONFIG_IXGBE_DCB
7927 netdev_set_num_tc(dev, tc);
7928 ixgbe_set_prio_tc_map(adapter);
7930 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7932 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
7933 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
7934 adapter->hw.fc.requested_mode = ixgbe_fc_none;
7937 netdev_reset_tc(dev);
7939 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7940 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7942 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7944 adapter->temp_dcb_cfg.pfc_mode_enable = false;
7945 adapter->dcb_cfg.pfc_mode_enable = false;
7948 ixgbe_validate_rtr(adapter, tc);
7950 #endif /* CONFIG_IXGBE_DCB */
7951 ixgbe_init_interrupt_scheme(adapter);
7953 if (netif_running(dev))
7954 return ixgbe_open(dev);
7959 #ifdef CONFIG_PCI_IOV
7960 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
7962 struct net_device *netdev = adapter->netdev;
7965 ixgbe_setup_tc(netdev, netdev_get_num_tc(netdev));
7970 void ixgbe_do_reset(struct net_device *netdev)
7972 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7974 if (netif_running(netdev))
7975 ixgbe_reinit_locked(adapter);
7977 ixgbe_reset(adapter);
7980 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
7981 netdev_features_t features)
7983 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7985 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7986 if (!(features & NETIF_F_RXCSUM))
7987 features &= ~NETIF_F_LRO;
7989 /* Turn off LRO if not RSC capable */
7990 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
7991 features &= ~NETIF_F_LRO;
7996 static int ixgbe_set_features(struct net_device *netdev,
7997 netdev_features_t features)
7999 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8000 netdev_features_t changed = netdev->features ^ features;
8001 bool need_reset = false;
8003 /* Make sure RSC matches LRO, reset if change */
8004 if (!(features & NETIF_F_LRO)) {
8005 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
8007 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
8008 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
8009 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
8010 if (adapter->rx_itr_setting == 1 ||
8011 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
8012 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
8014 } else if ((changed ^ features) & NETIF_F_LRO) {
8015 e_info(probe, "rx-usecs set too low, "
8021 * Check if Flow Director n-tuple support was enabled or disabled. If
8022 * the state changed, we need to reset.
8024 switch (features & NETIF_F_NTUPLE) {
8025 case NETIF_F_NTUPLE:
8026 /* turn off ATR, enable perfect filters and reset */
8027 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
8030 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
8031 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
8034 /* turn off perfect filters, enable ATR and reset */
8035 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
8038 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
8040 /* We cannot enable ATR if SR-IOV is enabled */
8041 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8044 /* We cannot enable ATR if we have 2 or more traffic classes */
8045 if (netdev_get_num_tc(netdev) > 1)
8048 /* We cannot enable ATR if RSS is disabled */
8049 if (adapter->ring_feature[RING_F_RSS].limit <= 1)
8052 /* A sample rate of 0 indicates ATR disabled */
8053 if (!adapter->atr_sample_rate)
8056 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
8060 if (features & NETIF_F_HW_VLAN_CTAG_RX)
8061 ixgbe_vlan_strip_enable(adapter);
8063 ixgbe_vlan_strip_disable(adapter);
8065 if (changed & NETIF_F_RXALL)
8068 netdev->features = features;
8070 #ifdef CONFIG_IXGBE_VXLAN
8071 if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
8072 if (features & NETIF_F_RXCSUM)
8073 adapter->flags2 |= IXGBE_FLAG2_VXLAN_REREG_NEEDED;
8075 ixgbe_clear_vxlan_port(adapter);
8077 #endif /* CONFIG_IXGBE_VXLAN */
8080 ixgbe_do_reset(netdev);
8085 #ifdef CONFIG_IXGBE_VXLAN
8087 * ixgbe_add_vxlan_port - Get notifications about VXLAN ports that come up
8088 * @dev: The port's netdev
8089 * @sa_family: Socket Family that VXLAN is notifiying us about
8090 * @port: New UDP port number that VXLAN started listening to
8092 static void ixgbe_add_vxlan_port(struct net_device *dev, sa_family_t sa_family,
8095 struct ixgbe_adapter *adapter = netdev_priv(dev);
8096 struct ixgbe_hw *hw = &adapter->hw;
8097 u16 new_port = ntohs(port);
8099 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
8102 if (sa_family == AF_INET6)
8105 if (adapter->vxlan_port == new_port)
8108 if (adapter->vxlan_port) {
8110 "Hit Max num of VXLAN ports, not adding port %d\n",
8115 adapter->vxlan_port = new_port;
8116 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, new_port);
8120 * ixgbe_del_vxlan_port - Get notifications about VXLAN ports that go away
8121 * @dev: The port's netdev
8122 * @sa_family: Socket Family that VXLAN is notifying us about
8123 * @port: UDP port number that VXLAN stopped listening to
8125 static void ixgbe_del_vxlan_port(struct net_device *dev, sa_family_t sa_family,
8128 struct ixgbe_adapter *adapter = netdev_priv(dev);
8129 u16 new_port = ntohs(port);
8131 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
8134 if (sa_family == AF_INET6)
8137 if (adapter->vxlan_port != new_port) {
8138 netdev_info(dev, "Port %d was not found, not deleting\n",
8143 ixgbe_clear_vxlan_port(adapter);
8144 adapter->flags2 |= IXGBE_FLAG2_VXLAN_REREG_NEEDED;
8146 #endif /* CONFIG_IXGBE_VXLAN */
8148 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
8149 struct net_device *dev,
8150 const unsigned char *addr, u16 vid,
8153 /* guarantee we can provide a unique filter for the unicast address */
8154 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
8155 if (IXGBE_MAX_PF_MACVLANS <= netdev_uc_count(dev))
8159 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
8163 * ixgbe_configure_bridge_mode - set various bridge modes
8164 * @adapter - the private structure
8165 * @mode - requested bridge mode
8167 * Configure some settings require for various bridge modes.
8169 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
8172 struct ixgbe_hw *hw = &adapter->hw;
8173 unsigned int p, num_pools;
8177 case BRIDGE_MODE_VEPA:
8178 /* disable Tx loopback, rely on switch hairpin mode */
8179 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
8181 /* must enable Rx switching replication to allow multicast
8182 * packet reception on all VFs, and to enable source address
8185 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
8186 vmdctl |= IXGBE_VT_CTL_REPLEN;
8187 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
8189 /* enable Rx source address pruning. Note, this requires
8190 * replication to be enabled or else it does nothing.
8192 num_pools = adapter->num_vfs + adapter->num_rx_pools;
8193 for (p = 0; p < num_pools; p++) {
8194 if (hw->mac.ops.set_source_address_pruning)
8195 hw->mac.ops.set_source_address_pruning(hw,
8200 case BRIDGE_MODE_VEB:
8201 /* enable Tx loopback for internal VF/PF communication */
8202 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
8203 IXGBE_PFDTXGSWC_VT_LBEN);
8205 /* disable Rx switching replication unless we have SR-IOV
8208 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
8209 if (!adapter->num_vfs)
8210 vmdctl &= ~IXGBE_VT_CTL_REPLEN;
8211 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
8213 /* disable Rx source address pruning, since we don't expect to
8214 * be receiving external loopback of our transmitted frames.
8216 num_pools = adapter->num_vfs + adapter->num_rx_pools;
8217 for (p = 0; p < num_pools; p++) {
8218 if (hw->mac.ops.set_source_address_pruning)
8219 hw->mac.ops.set_source_address_pruning(hw,
8228 adapter->bridge_mode = mode;
8230 e_info(drv, "enabling bridge mode: %s\n",
8231 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
8236 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
8237 struct nlmsghdr *nlh, u16 flags)
8239 struct ixgbe_adapter *adapter = netdev_priv(dev);
8240 struct nlattr *attr, *br_spec;
8243 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
8246 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
8250 nla_for_each_nested(attr, br_spec, rem) {
8254 if (nla_type(attr) != IFLA_BRIDGE_MODE)
8257 if (nla_len(attr) < sizeof(mode))
8260 mode = nla_get_u16(attr);
8261 status = ixgbe_configure_bridge_mode(adapter, mode);
8271 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
8272 struct net_device *dev,
8273 u32 filter_mask, int nlflags)
8275 struct ixgbe_adapter *adapter = netdev_priv(dev);
8277 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
8280 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
8281 adapter->bridge_mode, 0, 0, nlflags,
8285 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
8287 struct ixgbe_fwd_adapter *fwd_adapter = NULL;
8288 struct ixgbe_adapter *adapter = netdev_priv(pdev);
8289 int used_pools = adapter->num_vfs + adapter->num_rx_pools;
8293 /* Hardware has a limited number of available pools. Each VF, and the
8294 * PF require a pool. Check to ensure we don't attempt to use more
8295 * then the available number of pools.
8297 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
8298 return ERR_PTR(-EINVAL);
8301 if (vdev->num_rx_queues != vdev->num_tx_queues) {
8302 netdev_info(pdev, "%s: Only supports a single queue count for TX and RX\n",
8304 return ERR_PTR(-EINVAL);
8307 /* Check for hardware restriction on number of rx/tx queues */
8308 if (vdev->num_tx_queues > IXGBE_MAX_L2A_QUEUES ||
8309 vdev->num_tx_queues == IXGBE_BAD_L2A_QUEUE) {
8311 "%s: Supports RX/TX Queue counts 1,2, and 4\n",
8313 return ERR_PTR(-EINVAL);
8316 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8317 adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS - 1) ||
8318 (adapter->num_rx_pools > IXGBE_MAX_MACVLANS))
8319 return ERR_PTR(-EBUSY);
8321 fwd_adapter = kzalloc(sizeof(*fwd_adapter), GFP_KERNEL);
8323 return ERR_PTR(-ENOMEM);
8325 pool = find_first_zero_bit(&adapter->fwd_bitmask, 32);
8326 adapter->num_rx_pools++;
8327 set_bit(pool, &adapter->fwd_bitmask);
8328 limit = find_last_bit(&adapter->fwd_bitmask, 32);
8330 /* Enable VMDq flag so device will be set in VM mode */
8331 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED | IXGBE_FLAG_SRIOV_ENABLED;
8332 adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
8333 adapter->ring_feature[RING_F_RSS].limit = vdev->num_tx_queues;
8335 /* Force reinit of ring allocation with VMDQ enabled */
8336 err = ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
8339 fwd_adapter->pool = pool;
8340 fwd_adapter->real_adapter = adapter;
8341 err = ixgbe_fwd_ring_up(vdev, fwd_adapter);
8344 netif_tx_start_all_queues(vdev);
8347 /* unwind counter and free adapter struct */
8349 "%s: dfwd hardware acceleration failed\n", vdev->name);
8350 clear_bit(pool, &adapter->fwd_bitmask);
8351 adapter->num_rx_pools--;
8353 return ERR_PTR(err);
8356 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
8358 struct ixgbe_fwd_adapter *fwd_adapter = priv;
8359 struct ixgbe_adapter *adapter = fwd_adapter->real_adapter;
8362 clear_bit(fwd_adapter->pool, &adapter->fwd_bitmask);
8363 adapter->num_rx_pools--;
8365 limit = find_last_bit(&adapter->fwd_bitmask, 32);
8366 adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
8367 ixgbe_fwd_ring_down(fwd_adapter->netdev, fwd_adapter);
8368 ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
8369 netdev_dbg(pdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
8370 fwd_adapter->pool, adapter->num_rx_pools,
8371 fwd_adapter->rx_base_queue,
8372 fwd_adapter->rx_base_queue + adapter->num_rx_queues_per_pool,
8373 adapter->fwd_bitmask);
8377 #define IXGBE_MAX_TUNNEL_HDR_LEN 80
8378 static netdev_features_t
8379 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
8380 netdev_features_t features)
8382 if (!skb->encapsulation)
8385 if (unlikely(skb_inner_mac_header(skb) - skb_transport_header(skb) >
8386 IXGBE_MAX_TUNNEL_HDR_LEN))
8387 return features & ~NETIF_F_ALL_CSUM;
8392 static const struct net_device_ops ixgbe_netdev_ops = {
8393 .ndo_open = ixgbe_open,
8394 .ndo_stop = ixgbe_close,
8395 .ndo_start_xmit = ixgbe_xmit_frame,
8396 .ndo_select_queue = ixgbe_select_queue,
8397 .ndo_set_rx_mode = ixgbe_set_rx_mode,
8398 .ndo_validate_addr = eth_validate_addr,
8399 .ndo_set_mac_address = ixgbe_set_mac,
8400 .ndo_change_mtu = ixgbe_change_mtu,
8401 .ndo_tx_timeout = ixgbe_tx_timeout,
8402 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
8403 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
8404 .ndo_do_ioctl = ixgbe_ioctl,
8405 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
8406 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
8407 .ndo_set_vf_rate = ixgbe_ndo_set_vf_bw,
8408 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
8409 .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
8410 .ndo_set_vf_trust = ixgbe_ndo_set_vf_trust,
8411 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
8412 .ndo_get_stats64 = ixgbe_get_stats64,
8413 #ifdef CONFIG_IXGBE_DCB
8414 .ndo_setup_tc = ixgbe_setup_tc,
8416 #ifdef CONFIG_NET_POLL_CONTROLLER
8417 .ndo_poll_controller = ixgbe_netpoll,
8419 #ifdef CONFIG_NET_RX_BUSY_POLL
8420 .ndo_busy_poll = ixgbe_low_latency_recv,
8423 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
8424 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
8425 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
8426 .ndo_fcoe_enable = ixgbe_fcoe_enable,
8427 .ndo_fcoe_disable = ixgbe_fcoe_disable,
8428 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
8429 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
8430 #endif /* IXGBE_FCOE */
8431 .ndo_set_features = ixgbe_set_features,
8432 .ndo_fix_features = ixgbe_fix_features,
8433 .ndo_fdb_add = ixgbe_ndo_fdb_add,
8434 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
8435 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
8436 .ndo_dfwd_add_station = ixgbe_fwd_add,
8437 .ndo_dfwd_del_station = ixgbe_fwd_del,
8438 #ifdef CONFIG_IXGBE_VXLAN
8439 .ndo_add_vxlan_port = ixgbe_add_vxlan_port,
8440 .ndo_del_vxlan_port = ixgbe_del_vxlan_port,
8441 #endif /* CONFIG_IXGBE_VXLAN */
8442 .ndo_features_check = ixgbe_features_check,
8446 * ixgbe_enumerate_functions - Get the number of ports this device has
8447 * @adapter: adapter structure
8449 * This function enumerates the phsyical functions co-located on a single slot,
8450 * in order to determine how many ports a device has. This is most useful in
8451 * determining the required GT/s of PCIe bandwidth necessary for optimal
8454 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
8456 struct pci_dev *entry, *pdev = adapter->pdev;
8459 /* Some cards can not use the generic count PCIe functions method,
8460 * because they are behind a parent switch, so we hardcode these with
8461 * the correct number of functions.
8463 if (ixgbe_pcie_from_parent(&adapter->hw))
8466 list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
8467 /* don't count virtual functions */
8468 if (entry->is_virtfn)
8471 /* When the devices on the bus don't all match our device ID,
8472 * we can't reliably determine the correct number of
8473 * functions. This can occur if a function has been direct
8474 * attached to a virtual machine using VT-d, for example. In
8475 * this case, simply return -1 to indicate this.
8477 if ((entry->vendor != pdev->vendor) ||
8478 (entry->device != pdev->device))
8488 * ixgbe_wol_supported - Check whether device supports WoL
8489 * @hw: hw specific details
8490 * @device_id: the device ID
8491 * @subdev_id: the subsystem device ID
8493 * This function is used by probe and ethtool to determine
8494 * which devices have WoL support
8497 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
8500 struct ixgbe_hw *hw = &adapter->hw;
8501 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
8502 int is_wol_supported = 0;
8504 switch (device_id) {
8505 case IXGBE_DEV_ID_82599_SFP:
8506 /* Only these subdevices could supports WOL */
8507 switch (subdevice_id) {
8508 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
8509 case IXGBE_SUBDEV_ID_82599_560FLR:
8510 /* only support first port */
8511 if (hw->bus.func != 0)
8513 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
8514 case IXGBE_SUBDEV_ID_82599_SFP:
8515 case IXGBE_SUBDEV_ID_82599_RNDC:
8516 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
8517 case IXGBE_SUBDEV_ID_82599_LOM_SFP:
8518 is_wol_supported = 1;
8522 case IXGBE_DEV_ID_82599EN_SFP:
8523 /* Only this subdevice supports WOL */
8524 switch (subdevice_id) {
8525 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
8526 is_wol_supported = 1;
8530 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
8531 /* All except this subdevice support WOL */
8532 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
8533 is_wol_supported = 1;
8535 case IXGBE_DEV_ID_82599_KX4:
8536 is_wol_supported = 1;
8538 case IXGBE_DEV_ID_X540T:
8539 case IXGBE_DEV_ID_X540T1:
8540 case IXGBE_DEV_ID_X550T:
8541 case IXGBE_DEV_ID_X550EM_X_KX4:
8542 case IXGBE_DEV_ID_X550EM_X_KR:
8543 case IXGBE_DEV_ID_X550EM_X_10G_T:
8544 /* check eeprom to see if enabled wol */
8545 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
8546 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
8547 (hw->bus.func == 0))) {
8548 is_wol_supported = 1;
8553 return is_wol_supported;
8557 * ixgbe_get_platform_mac_addr - Look up MAC address in Open Firmware / IDPROM
8558 * @adapter: Pointer to adapter struct
8560 static void ixgbe_get_platform_mac_addr(struct ixgbe_adapter *adapter)
8563 struct device_node *dp = pci_device_to_OF_node(adapter->pdev);
8564 struct ixgbe_hw *hw = &adapter->hw;
8565 const unsigned char *addr;
8567 addr = of_get_mac_address(dp);
8569 ether_addr_copy(hw->mac.perm_addr, addr);
8572 #endif /* CONFIG_OF */
8575 ether_addr_copy(hw->mac.perm_addr, idprom->id_ethaddr);
8576 #endif /* CONFIG_SPARC */
8580 * ixgbe_probe - Device Initialization Routine
8581 * @pdev: PCI device information struct
8582 * @ent: entry in ixgbe_pci_tbl
8584 * Returns 0 on success, negative on failure
8586 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
8587 * The OS initialization, configuring of the adapter private structure,
8588 * and a hardware reset occur.
8590 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
8592 struct net_device *netdev;
8593 struct ixgbe_adapter *adapter = NULL;
8594 struct ixgbe_hw *hw;
8595 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
8596 int i, err, pci_using_dac, expected_gts;
8597 unsigned int indices = MAX_TX_QUEUES;
8598 u8 part_str[IXGBE_PBANUM_LENGTH];
8599 bool disable_dev = false;
8605 /* Catch broken hardware that put the wrong VF device ID in
8606 * the PCIe SR-IOV capability.
8608 if (pdev->is_virtfn) {
8609 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
8610 pci_name(pdev), pdev->vendor, pdev->device);
8614 err = pci_enable_device_mem(pdev);
8618 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
8621 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
8624 "No usable DMA configuration, aborting\n");
8630 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
8631 IORESOURCE_MEM), ixgbe_driver_name);
8634 "pci_request_selected_regions failed 0x%x\n", err);
8638 pci_enable_pcie_error_reporting(pdev);
8640 pci_set_master(pdev);
8641 pci_save_state(pdev);
8643 if (ii->mac == ixgbe_mac_82598EB) {
8644 #ifdef CONFIG_IXGBE_DCB
8645 /* 8 TC w/ 4 queues per TC */
8646 indices = 4 * MAX_TRAFFIC_CLASS;
8648 indices = IXGBE_MAX_RSS_INDICES;
8652 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
8655 goto err_alloc_etherdev;
8658 SET_NETDEV_DEV(netdev, &pdev->dev);
8660 adapter = netdev_priv(netdev);
8662 adapter->netdev = netdev;
8663 adapter->pdev = pdev;
8666 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
8668 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
8669 pci_resource_len(pdev, 0));
8670 adapter->io_addr = hw->hw_addr;
8676 netdev->netdev_ops = &ixgbe_netdev_ops;
8677 ixgbe_set_ethtool_ops(netdev);
8678 netdev->watchdog_timeo = 5 * HZ;
8679 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
8682 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
8683 hw->mac.type = ii->mac;
8684 hw->mvals = ii->mvals;
8687 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
8688 eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
8689 if (ixgbe_removed(hw->hw_addr)) {
8693 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
8694 if (!(eec & (1 << 8)))
8695 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
8698 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
8699 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
8700 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
8701 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
8702 hw->phy.mdio.mmds = 0;
8703 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
8704 hw->phy.mdio.dev = netdev;
8705 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
8706 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
8708 ii->get_invariants(hw);
8710 /* setup the private structure */
8711 err = ixgbe_sw_init(adapter);
8715 /* Make it possible the adapter to be woken up via WOL */
8716 switch (adapter->hw.mac.type) {
8717 case ixgbe_mac_82599EB:
8718 case ixgbe_mac_X540:
8719 case ixgbe_mac_X550:
8720 case ixgbe_mac_X550EM_x:
8721 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
8728 * If there is a fan on this device and it has failed log the
8731 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
8732 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
8733 if (esdp & IXGBE_ESDP_SDP1)
8734 e_crit(probe, "Fan has stopped, replace the adapter\n");
8737 if (allow_unsupported_sfp)
8738 hw->allow_unsupported_sfp = allow_unsupported_sfp;
8740 /* reset_hw fills in the perm_addr as well */
8741 hw->phy.reset_if_overtemp = true;
8742 err = hw->mac.ops.reset_hw(hw);
8743 hw->phy.reset_if_overtemp = false;
8744 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
8746 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
8747 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
8748 e_dev_err("Reload the driver after installing a supported module.\n");
8751 e_dev_err("HW Init failed: %d\n", err);
8755 #ifdef CONFIG_PCI_IOV
8756 /* SR-IOV not supported on the 82598 */
8757 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
8760 ixgbe_init_mbx_params_pf(hw);
8761 memcpy(&hw->mbx.ops, ii->mbx_ops, sizeof(hw->mbx.ops));
8762 pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
8763 ixgbe_enable_sriov(adapter);
8767 netdev->features = NETIF_F_SG |
8770 NETIF_F_HW_VLAN_CTAG_TX |
8771 NETIF_F_HW_VLAN_CTAG_RX |
8777 netdev->hw_features = netdev->features | NETIF_F_HW_L2FW_DOFFLOAD;
8779 switch (adapter->hw.mac.type) {
8780 case ixgbe_mac_82599EB:
8781 case ixgbe_mac_X540:
8782 case ixgbe_mac_X550:
8783 case ixgbe_mac_X550EM_x:
8784 netdev->features |= NETIF_F_SCTP_CSUM;
8785 netdev->hw_features |= NETIF_F_SCTP_CSUM |
8792 netdev->hw_features |= NETIF_F_RXALL;
8793 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
8795 netdev->vlan_features |= NETIF_F_TSO;
8796 netdev->vlan_features |= NETIF_F_TSO6;
8797 netdev->vlan_features |= NETIF_F_IP_CSUM;
8798 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
8799 netdev->vlan_features |= NETIF_F_SG;
8801 netdev->hw_enc_features |= NETIF_F_SG | NETIF_F_IP_CSUM |
8804 netdev->priv_flags |= IFF_UNICAST_FLT;
8805 netdev->priv_flags |= IFF_SUPP_NOFCS;
8807 #ifdef CONFIG_IXGBE_VXLAN
8808 switch (adapter->hw.mac.type) {
8809 case ixgbe_mac_X550:
8810 case ixgbe_mac_X550EM_x:
8811 netdev->hw_enc_features |= NETIF_F_RXCSUM |
8818 #endif /* CONFIG_IXGBE_VXLAN */
8820 #ifdef CONFIG_IXGBE_DCB
8821 netdev->dcbnl_ops = &dcbnl_ops;
8825 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
8826 unsigned int fcoe_l;
8828 if (hw->mac.ops.get_device_caps) {
8829 hw->mac.ops.get_device_caps(hw, &device_caps);
8830 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
8831 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
8835 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
8836 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
8838 netdev->features |= NETIF_F_FSO |
8841 netdev->vlan_features |= NETIF_F_FSO |
8845 #endif /* IXGBE_FCOE */
8846 if (pci_using_dac) {
8847 netdev->features |= NETIF_F_HIGHDMA;
8848 netdev->vlan_features |= NETIF_F_HIGHDMA;
8851 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
8852 netdev->hw_features |= NETIF_F_LRO;
8853 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
8854 netdev->features |= NETIF_F_LRO;
8856 /* make sure the EEPROM is good */
8857 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
8858 e_dev_err("The EEPROM Checksum Is Not Valid\n");
8863 ixgbe_get_platform_mac_addr(adapter);
8865 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
8867 if (!is_valid_ether_addr(netdev->dev_addr)) {
8868 e_dev_err("invalid MAC address\n");
8873 ixgbe_mac_set_default_filter(adapter, hw->mac.perm_addr);
8875 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
8876 (unsigned long) adapter);
8878 if (ixgbe_removed(hw->hw_addr)) {
8882 INIT_WORK(&adapter->service_task, ixgbe_service_task);
8883 set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
8884 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
8886 err = ixgbe_init_interrupt_scheme(adapter);
8890 /* WOL not supported for all devices */
8892 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
8893 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
8894 pdev->subsystem_device);
8895 if (hw->wol_enabled)
8896 adapter->wol = IXGBE_WUFC_MAG;
8898 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
8900 /* save off EEPROM version number */
8901 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
8902 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
8904 /* pick up the PCI bus settings for reporting later */
8905 if (ixgbe_pcie_from_parent(hw))
8906 ixgbe_get_parent_bus_info(adapter);
8908 hw->mac.ops.get_bus_info(hw);
8910 /* calculate the expected PCIe bandwidth required for optimal
8911 * performance. Note that some older parts will never have enough
8912 * bandwidth due to being older generation PCIe parts. We clamp these
8913 * parts to ensure no warning is displayed if it can't be fixed.
8915 switch (hw->mac.type) {
8916 case ixgbe_mac_82598EB:
8917 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
8920 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
8924 /* don't check link if we failed to enumerate functions */
8925 if (expected_gts > 0)
8926 ixgbe_check_minimum_link(adapter, expected_gts);
8928 err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
8930 strlcpy(part_str, "Unknown", sizeof(part_str));
8931 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
8932 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
8933 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
8936 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
8937 hw->mac.type, hw->phy.type, part_str);
8939 e_dev_info("%pM\n", netdev->dev_addr);
8941 /* reset the hardware with the new settings */
8942 err = hw->mac.ops.start_hw(hw);
8943 if (err == IXGBE_ERR_EEPROM_VERSION) {
8944 /* We are running on a pre-production device, log a warning */
8945 e_dev_warn("This device is a pre-production adapter/LOM. "
8946 "Please be aware there may be issues associated "
8947 "with your hardware. If you are experiencing "
8948 "problems please contact your Intel or hardware "
8949 "representative who provided you with this "
8952 strcpy(netdev->name, "eth%d");
8953 err = register_netdev(netdev);
8957 pci_set_drvdata(pdev, adapter);
8959 /* power down the optics for 82599 SFP+ fiber */
8960 if (hw->mac.ops.disable_tx_laser)
8961 hw->mac.ops.disable_tx_laser(hw);
8963 /* carrier off reporting is important to ethtool even BEFORE open */
8964 netif_carrier_off(netdev);
8966 #ifdef CONFIG_IXGBE_DCA
8967 if (dca_add_requester(&pdev->dev) == 0) {
8968 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
8969 ixgbe_setup_dca(adapter);
8972 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
8973 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
8974 for (i = 0; i < adapter->num_vfs; i++)
8975 ixgbe_vf_configuration(pdev, (i | 0x10000000));
8978 /* firmware requires driver version to be 0xFFFFFFFF
8979 * since os does not support feature
8981 if (hw->mac.ops.set_fw_drv_ver)
8982 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
8985 /* add san mac addr to netdev */
8986 ixgbe_add_sanmac_netdev(netdev);
8988 e_dev_info("%s\n", ixgbe_default_device_descr);
8990 #ifdef CONFIG_IXGBE_HWMON
8991 if (ixgbe_sysfs_init(adapter))
8992 e_err(probe, "failed to allocate sysfs resources\n");
8993 #endif /* CONFIG_IXGBE_HWMON */
8995 ixgbe_dbg_adapter_init(adapter);
8997 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
8998 if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
8999 hw->mac.ops.setup_link(hw,
9000 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
9006 ixgbe_release_hw_control(adapter);
9007 ixgbe_clear_interrupt_scheme(adapter);
9009 ixgbe_disable_sriov(adapter);
9010 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
9011 iounmap(adapter->io_addr);
9012 kfree(adapter->mac_table);
9014 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
9015 free_netdev(netdev);
9017 pci_release_selected_regions(pdev,
9018 pci_select_bars(pdev, IORESOURCE_MEM));
9021 if (!adapter || disable_dev)
9022 pci_disable_device(pdev);
9027 * ixgbe_remove - Device Removal Routine
9028 * @pdev: PCI device information struct
9030 * ixgbe_remove is called by the PCI subsystem to alert the driver
9031 * that it should release a PCI device. The could be caused by a
9032 * Hot-Plug event, or because the driver is going to be removed from
9035 static void ixgbe_remove(struct pci_dev *pdev)
9037 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9038 struct net_device *netdev;
9041 /* if !adapter then we already cleaned up in probe */
9045 netdev = adapter->netdev;
9046 ixgbe_dbg_adapter_exit(adapter);
9048 set_bit(__IXGBE_REMOVING, &adapter->state);
9049 cancel_work_sync(&adapter->service_task);
9052 #ifdef CONFIG_IXGBE_DCA
9053 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
9054 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
9055 dca_remove_requester(&pdev->dev);
9056 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
9057 IXGBE_DCA_CTRL_DCA_DISABLE);
9061 #ifdef CONFIG_IXGBE_HWMON
9062 ixgbe_sysfs_exit(adapter);
9063 #endif /* CONFIG_IXGBE_HWMON */
9065 /* remove the added san mac */
9066 ixgbe_del_sanmac_netdev(netdev);
9068 #ifdef CONFIG_PCI_IOV
9069 ixgbe_disable_sriov(adapter);
9071 if (netdev->reg_state == NETREG_REGISTERED)
9072 unregister_netdev(netdev);
9074 ixgbe_clear_interrupt_scheme(adapter);
9076 ixgbe_release_hw_control(adapter);
9079 kfree(adapter->ixgbe_ieee_pfc);
9080 kfree(adapter->ixgbe_ieee_ets);
9083 iounmap(adapter->io_addr);
9084 pci_release_selected_regions(pdev, pci_select_bars(pdev,
9087 e_dev_info("complete\n");
9089 kfree(adapter->mac_table);
9090 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
9091 free_netdev(netdev);
9093 pci_disable_pcie_error_reporting(pdev);
9096 pci_disable_device(pdev);
9100 * ixgbe_io_error_detected - called when PCI error is detected
9101 * @pdev: Pointer to PCI device
9102 * @state: The current pci connection state
9104 * This function is called after a PCI bus error affecting
9105 * this device has been detected.
9107 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
9108 pci_channel_state_t state)
9110 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9111 struct net_device *netdev = adapter->netdev;
9113 #ifdef CONFIG_PCI_IOV
9114 struct ixgbe_hw *hw = &adapter->hw;
9115 struct pci_dev *bdev, *vfdev;
9116 u32 dw0, dw1, dw2, dw3;
9118 u16 req_id, pf_func;
9120 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
9121 adapter->num_vfs == 0)
9122 goto skip_bad_vf_detection;
9124 bdev = pdev->bus->self;
9125 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
9126 bdev = bdev->bus->self;
9129 goto skip_bad_vf_detection;
9131 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
9133 goto skip_bad_vf_detection;
9135 dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
9136 dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
9137 dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
9138 dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
9139 if (ixgbe_removed(hw->hw_addr))
9140 goto skip_bad_vf_detection;
9143 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
9144 if (!(req_id & 0x0080))
9145 goto skip_bad_vf_detection;
9147 pf_func = req_id & 0x01;
9148 if ((pf_func & 1) == (pdev->devfn & 1)) {
9149 unsigned int device_id;
9151 vf = (req_id & 0x7F) >> 1;
9152 e_dev_err("VF %d has caused a PCIe error\n", vf);
9153 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
9154 "%8.8x\tdw3: %8.8x\n",
9155 dw0, dw1, dw2, dw3);
9156 switch (adapter->hw.mac.type) {
9157 case ixgbe_mac_82599EB:
9158 device_id = IXGBE_82599_VF_DEVICE_ID;
9160 case ixgbe_mac_X540:
9161 device_id = IXGBE_X540_VF_DEVICE_ID;
9163 case ixgbe_mac_X550:
9164 device_id = IXGBE_DEV_ID_X550_VF;
9166 case ixgbe_mac_X550EM_x:
9167 device_id = IXGBE_DEV_ID_X550EM_X_VF;
9174 /* Find the pci device of the offending VF */
9175 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
9177 if (vfdev->devfn == (req_id & 0xFF))
9179 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
9183 * There's a slim chance the VF could have been hot plugged,
9184 * so if it is no longer present we don't need to issue the
9185 * VFLR. Just clean up the AER in that case.
9188 ixgbe_issue_vf_flr(adapter, vfdev);
9189 /* Free device reference count */
9193 pci_cleanup_aer_uncorrect_error_status(pdev);
9197 * Even though the error may have occurred on the other port
9198 * we still need to increment the vf error reference count for
9199 * both ports because the I/O resume function will be called
9202 adapter->vferr_refcount++;
9204 return PCI_ERS_RESULT_RECOVERED;
9206 skip_bad_vf_detection:
9207 #endif /* CONFIG_PCI_IOV */
9208 if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
9209 return PCI_ERS_RESULT_DISCONNECT;
9212 netif_device_detach(netdev);
9214 if (state == pci_channel_io_perm_failure) {
9216 return PCI_ERS_RESULT_DISCONNECT;
9219 if (netif_running(netdev))
9220 ixgbe_down(adapter);
9222 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
9223 pci_disable_device(pdev);
9226 /* Request a slot reset. */
9227 return PCI_ERS_RESULT_NEED_RESET;
9231 * ixgbe_io_slot_reset - called after the pci bus has been reset.
9232 * @pdev: Pointer to PCI device
9234 * Restart the card from scratch, as if from a cold-boot.
9236 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
9238 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9239 pci_ers_result_t result;
9242 if (pci_enable_device_mem(pdev)) {
9243 e_err(probe, "Cannot re-enable PCI device after reset.\n");
9244 result = PCI_ERS_RESULT_DISCONNECT;
9246 smp_mb__before_atomic();
9247 clear_bit(__IXGBE_DISABLED, &adapter->state);
9248 adapter->hw.hw_addr = adapter->io_addr;
9249 pci_set_master(pdev);
9250 pci_restore_state(pdev);
9251 pci_save_state(pdev);
9253 pci_wake_from_d3(pdev, false);
9255 ixgbe_reset(adapter);
9256 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
9257 result = PCI_ERS_RESULT_RECOVERED;
9260 err = pci_cleanup_aer_uncorrect_error_status(pdev);
9262 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
9263 "failed 0x%0x\n", err);
9264 /* non-fatal, continue */
9271 * ixgbe_io_resume - called when traffic can start flowing again.
9272 * @pdev: Pointer to PCI device
9274 * This callback is called when the error recovery driver tells us that
9275 * its OK to resume normal operation.
9277 static void ixgbe_io_resume(struct pci_dev *pdev)
9279 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9280 struct net_device *netdev = adapter->netdev;
9282 #ifdef CONFIG_PCI_IOV
9283 if (adapter->vferr_refcount) {
9284 e_info(drv, "Resuming after VF err\n");
9285 adapter->vferr_refcount--;
9290 if (netif_running(netdev))
9293 netif_device_attach(netdev);
9296 static const struct pci_error_handlers ixgbe_err_handler = {
9297 .error_detected = ixgbe_io_error_detected,
9298 .slot_reset = ixgbe_io_slot_reset,
9299 .resume = ixgbe_io_resume,
9302 static struct pci_driver ixgbe_driver = {
9303 .name = ixgbe_driver_name,
9304 .id_table = ixgbe_pci_tbl,
9305 .probe = ixgbe_probe,
9306 .remove = ixgbe_remove,
9308 .suspend = ixgbe_suspend,
9309 .resume = ixgbe_resume,
9311 .shutdown = ixgbe_shutdown,
9312 .sriov_configure = ixgbe_pci_sriov_configure,
9313 .err_handler = &ixgbe_err_handler
9317 * ixgbe_init_module - Driver Registration Routine
9319 * ixgbe_init_module is the first routine called when the driver is
9320 * loaded. All it does is register with the PCI subsystem.
9322 static int __init ixgbe_init_module(void)
9325 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
9326 pr_info("%s\n", ixgbe_copyright);
9330 ret = pci_register_driver(&ixgbe_driver);
9336 #ifdef CONFIG_IXGBE_DCA
9337 dca_register_notify(&dca_notifier);
9343 module_init(ixgbe_init_module);
9346 * ixgbe_exit_module - Driver Exit Cleanup Routine
9348 * ixgbe_exit_module is called just before the driver is removed
9351 static void __exit ixgbe_exit_module(void)
9353 #ifdef CONFIG_IXGBE_DCA
9354 dca_unregister_notify(&dca_notifier);
9356 pci_unregister_driver(&ixgbe_driver);
9361 #ifdef CONFIG_IXGBE_DCA
9362 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
9367 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
9368 __ixgbe_notify_dca);
9370 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
9373 #endif /* CONFIG_IXGBE_DCA */
9375 module_exit(ixgbe_exit_module);