1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2014 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 Linux NICS <linux.nics@intel.com>
24 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
27 *******************************************************************************/
29 #include <linux/types.h>
30 #include <linux/module.h>
31 #include <linux/pci.h>
32 #include <linux/netdevice.h>
33 #include <linux/vmalloc.h>
34 #include <linux/string.h>
36 #include <linux/interrupt.h>
38 #include <linux/tcp.h>
39 #include <linux/sctp.h>
40 #include <linux/pkt_sched.h>
41 #include <linux/ipv6.h>
42 #include <linux/slab.h>
43 #include <net/checksum.h>
44 #include <net/ip6_checksum.h>
45 #include <linux/ethtool.h>
47 #include <linux/if_vlan.h>
48 #include <linux/if_macvlan.h>
49 #include <linux/if_bridge.h>
50 #include <linux/prefetch.h>
51 #include <scsi/fc/fc_fcoe.h>
54 #include "ixgbe_common.h"
55 #include "ixgbe_dcb_82599.h"
56 #include "ixgbe_sriov.h"
58 char ixgbe_driver_name
[] = "ixgbe";
59 static const char ixgbe_driver_string
[] =
60 "Intel(R) 10 Gigabit PCI Express Network Driver";
62 char ixgbe_default_device_descr
[] =
63 "Intel(R) 10 Gigabit Network Connection";
65 static char ixgbe_default_device_descr
[] =
66 "Intel(R) 10 Gigabit Network Connection";
68 #define DRV_VERSION "3.19.1-k"
69 const char ixgbe_driver_version
[] = DRV_VERSION
;
70 static const char ixgbe_copyright
[] =
71 "Copyright (c) 1999-2014 Intel Corporation.";
73 static const struct ixgbe_info
*ixgbe_info_tbl
[] = {
74 [board_82598
] = &ixgbe_82598_info
,
75 [board_82599
] = &ixgbe_82599_info
,
76 [board_X540
] = &ixgbe_X540_info
,
79 /* ixgbe_pci_tbl - PCI Device ID Table
81 * Wildcard entries (PCI_ANY_ID) should come last
82 * Last entry must be all 0s
84 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
85 * Class, Class Mask, private data (not used) }
87 static const struct pci_device_id ixgbe_pci_tbl
[] = {
88 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598
), board_82598
},
89 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AF_DUAL_PORT
), board_82598
},
90 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AF_SINGLE_PORT
), board_82598
},
91 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AT
), board_82598
},
92 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AT2
), board_82598
},
93 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598EB_CX4
), board_82598
},
94 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_CX4_DUAL_PORT
), board_82598
},
95 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_DA_DUAL_PORT
), board_82598
},
96 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM
), board_82598
},
97 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598EB_XF_LR
), board_82598
},
98 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598EB_SFP_LOM
), board_82598
},
99 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_BX
), board_82598
},
100 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_KX4
), board_82599
},
101 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_XAUI_LOM
), board_82599
},
102 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_KR
), board_82599
},
103 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP
), board_82599
},
104 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP_EM
), board_82599
},
105 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_KX4_MEZZ
), board_82599
},
106 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_CX4
), board_82599
},
107 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_BACKPLANE_FCOE
), board_82599
},
108 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP_FCOE
), board_82599
},
109 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_T3_LOM
), board_82599
},
110 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_COMBO_BACKPLANE
), board_82599
},
111 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X540T
), board_X540
},
112 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP_SF2
), board_82599
},
113 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_LS
), board_82599
},
114 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_QSFP_SF_QP
), board_82599
},
115 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599EN_SFP
), board_82599
},
116 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP_SF_QP
), board_82599
},
117 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X540T1
), board_X540
},
118 /* required last entry */
121 MODULE_DEVICE_TABLE(pci
, ixgbe_pci_tbl
);
123 #ifdef CONFIG_IXGBE_DCA
124 static int ixgbe_notify_dca(struct notifier_block
*, unsigned long event
,
126 static struct notifier_block dca_notifier
= {
127 .notifier_call
= ixgbe_notify_dca
,
133 #ifdef CONFIG_PCI_IOV
134 static unsigned int max_vfs
;
135 module_param(max_vfs
, uint
, 0);
136 MODULE_PARM_DESC(max_vfs
,
137 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
138 #endif /* CONFIG_PCI_IOV */
140 static unsigned int allow_unsupported_sfp
;
141 module_param(allow_unsupported_sfp
, uint
, 0);
142 MODULE_PARM_DESC(allow_unsupported_sfp
,
143 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
145 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
146 static int debug
= -1;
147 module_param(debug
, int, 0);
148 MODULE_PARM_DESC(debug
, "Debug level (0=none,...,16=all)");
150 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
151 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
152 MODULE_LICENSE("GPL");
153 MODULE_VERSION(DRV_VERSION
);
155 static bool ixgbe_check_cfg_remove(struct ixgbe_hw
*hw
, struct pci_dev
*pdev
);
157 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter
*adapter
,
160 struct pci_dev
*parent_dev
;
161 struct pci_bus
*parent_bus
;
163 parent_bus
= adapter
->pdev
->bus
->parent
;
167 parent_dev
= parent_bus
->self
;
171 if (!pci_is_pcie(parent_dev
))
174 pcie_capability_read_word(parent_dev
, reg
, value
);
175 if (*value
== IXGBE_FAILED_READ_CFG_WORD
&&
176 ixgbe_check_cfg_remove(&adapter
->hw
, parent_dev
))
181 static s32
ixgbe_get_parent_bus_info(struct ixgbe_adapter
*adapter
)
183 struct ixgbe_hw
*hw
= &adapter
->hw
;
187 hw
->bus
.type
= ixgbe_bus_type_pci_express
;
189 /* Get the negotiated link width and speed from PCI config space of the
190 * parent, as this device is behind a switch
192 err
= ixgbe_read_pci_cfg_word_parent(adapter
, 18, &link_status
);
194 /* assume caller will handle error case */
198 hw
->bus
.width
= ixgbe_convert_bus_width(link_status
);
199 hw
->bus
.speed
= ixgbe_convert_bus_speed(link_status
);
205 * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
206 * @hw: hw specific details
208 * This function is used by probe to determine whether a device's PCI-Express
209 * bandwidth details should be gathered from the parent bus instead of from the
210 * device. Used to ensure that various locations all have the correct device ID
213 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw
*hw
)
215 switch (hw
->device_id
) {
216 case IXGBE_DEV_ID_82599_SFP_SF_QP
:
217 case IXGBE_DEV_ID_82599_QSFP_SF_QP
:
224 static void ixgbe_check_minimum_link(struct ixgbe_adapter
*adapter
,
228 enum pci_bus_speed speed
= PCI_SPEED_UNKNOWN
;
229 enum pcie_link_width width
= PCIE_LNK_WIDTH_UNKNOWN
;
230 struct pci_dev
*pdev
;
232 /* determine whether to use the the parent device
234 if (ixgbe_pcie_from_parent(&adapter
->hw
))
235 pdev
= adapter
->pdev
->bus
->parent
->self
;
237 pdev
= adapter
->pdev
;
239 if (pcie_get_minimum_link(pdev
, &speed
, &width
) ||
240 speed
== PCI_SPEED_UNKNOWN
|| width
== PCIE_LNK_WIDTH_UNKNOWN
) {
241 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
246 case PCIE_SPEED_2_5GT
:
247 /* 8b/10b encoding reduces max throughput by 20% */
250 case PCIE_SPEED_5_0GT
:
251 /* 8b/10b encoding reduces max throughput by 20% */
254 case PCIE_SPEED_8_0GT
:
255 /* 128b/130b encoding reduces throughput by less than 2% */
259 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
263 e_dev_info("PCI Express bandwidth of %dGT/s available\n",
265 e_dev_info("(Speed:%s, Width: x%d, Encoding Loss:%s)\n",
266 (speed
== PCIE_SPEED_8_0GT
? "8.0GT/s" :
267 speed
== PCIE_SPEED_5_0GT
? "5.0GT/s" :
268 speed
== PCIE_SPEED_2_5GT
? "2.5GT/s" :
271 (speed
== PCIE_SPEED_2_5GT
? "20%" :
272 speed
== PCIE_SPEED_5_0GT
? "20%" :
273 speed
== PCIE_SPEED_8_0GT
? "<2%" :
276 if (max_gts
< expected_gts
) {
277 e_dev_warn("This is not sufficient for optimal performance of this card.\n");
278 e_dev_warn("For optimal performance, at least %dGT/s of bandwidth is required.\n",
280 e_dev_warn("A slot with more lanes and/or higher speed is suggested.\n");
284 static void ixgbe_service_event_schedule(struct ixgbe_adapter
*adapter
)
286 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
) &&
287 !test_bit(__IXGBE_REMOVING
, &adapter
->state
) &&
288 !test_and_set_bit(__IXGBE_SERVICE_SCHED
, &adapter
->state
))
289 schedule_work(&adapter
->service_task
);
292 static void ixgbe_remove_adapter(struct ixgbe_hw
*hw
)
294 struct ixgbe_adapter
*adapter
= hw
->back
;
299 e_dev_err("Adapter removed\n");
300 if (test_bit(__IXGBE_SERVICE_INITED
, &adapter
->state
))
301 ixgbe_service_event_schedule(adapter
);
304 static void ixgbe_check_remove(struct ixgbe_hw
*hw
, u32 reg
)
308 /* The following check not only optimizes a bit by not
309 * performing a read on the status register when the
310 * register just read was a status register read that
311 * returned IXGBE_FAILED_READ_REG. It also blocks any
312 * potential recursion.
314 if (reg
== IXGBE_STATUS
) {
315 ixgbe_remove_adapter(hw
);
318 value
= ixgbe_read_reg(hw
, IXGBE_STATUS
);
319 if (value
== IXGBE_FAILED_READ_REG
)
320 ixgbe_remove_adapter(hw
);
324 * ixgbe_read_reg - Read from device register
325 * @hw: hw specific details
326 * @reg: offset of register to read
328 * Returns : value read or IXGBE_FAILED_READ_REG if removed
330 * This function is used to read device registers. It checks for device
331 * removal by confirming any read that returns all ones by checking the
332 * status register value for all ones. This function avoids reading from
333 * the hardware if a removal was previously detected in which case it
334 * returns IXGBE_FAILED_READ_REG (all ones).
336 u32
ixgbe_read_reg(struct ixgbe_hw
*hw
, u32 reg
)
338 u8 __iomem
*reg_addr
= ACCESS_ONCE(hw
->hw_addr
);
341 if (ixgbe_removed(reg_addr
))
342 return IXGBE_FAILED_READ_REG
;
343 value
= readl(reg_addr
+ reg
);
344 if (unlikely(value
== IXGBE_FAILED_READ_REG
))
345 ixgbe_check_remove(hw
, reg
);
349 static bool ixgbe_check_cfg_remove(struct ixgbe_hw
*hw
, struct pci_dev
*pdev
)
353 pci_read_config_word(pdev
, PCI_VENDOR_ID
, &value
);
354 if (value
== IXGBE_FAILED_READ_CFG_WORD
) {
355 ixgbe_remove_adapter(hw
);
361 u16
ixgbe_read_pci_cfg_word(struct ixgbe_hw
*hw
, u32 reg
)
363 struct ixgbe_adapter
*adapter
= hw
->back
;
366 if (ixgbe_removed(hw
->hw_addr
))
367 return IXGBE_FAILED_READ_CFG_WORD
;
368 pci_read_config_word(adapter
->pdev
, reg
, &value
);
369 if (value
== IXGBE_FAILED_READ_CFG_WORD
&&
370 ixgbe_check_cfg_remove(hw
, adapter
->pdev
))
371 return IXGBE_FAILED_READ_CFG_WORD
;
375 #ifdef CONFIG_PCI_IOV
376 static u32
ixgbe_read_pci_cfg_dword(struct ixgbe_hw
*hw
, u32 reg
)
378 struct ixgbe_adapter
*adapter
= hw
->back
;
381 if (ixgbe_removed(hw
->hw_addr
))
382 return IXGBE_FAILED_READ_CFG_DWORD
;
383 pci_read_config_dword(adapter
->pdev
, reg
, &value
);
384 if (value
== IXGBE_FAILED_READ_CFG_DWORD
&&
385 ixgbe_check_cfg_remove(hw
, adapter
->pdev
))
386 return IXGBE_FAILED_READ_CFG_DWORD
;
389 #endif /* CONFIG_PCI_IOV */
391 void ixgbe_write_pci_cfg_word(struct ixgbe_hw
*hw
, u32 reg
, u16 value
)
393 struct ixgbe_adapter
*adapter
= hw
->back
;
395 if (ixgbe_removed(hw
->hw_addr
))
397 pci_write_config_word(adapter
->pdev
, reg
, value
);
400 static void ixgbe_service_event_complete(struct ixgbe_adapter
*adapter
)
402 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED
, &adapter
->state
));
404 /* flush memory to make sure state is correct before next watchdog */
405 smp_mb__before_atomic();
406 clear_bit(__IXGBE_SERVICE_SCHED
, &adapter
->state
);
409 struct ixgbe_reg_info
{
414 static const struct ixgbe_reg_info ixgbe_reg_info_tbl
[] = {
416 /* General Registers */
417 {IXGBE_CTRL
, "CTRL"},
418 {IXGBE_STATUS
, "STATUS"},
419 {IXGBE_CTRL_EXT
, "CTRL_EXT"},
421 /* Interrupt Registers */
422 {IXGBE_EICR
, "EICR"},
425 {IXGBE_SRRCTL(0), "SRRCTL"},
426 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
427 {IXGBE_RDLEN(0), "RDLEN"},
428 {IXGBE_RDH(0), "RDH"},
429 {IXGBE_RDT(0), "RDT"},
430 {IXGBE_RXDCTL(0), "RXDCTL"},
431 {IXGBE_RDBAL(0), "RDBAL"},
432 {IXGBE_RDBAH(0), "RDBAH"},
435 {IXGBE_TDBAL(0), "TDBAL"},
436 {IXGBE_TDBAH(0), "TDBAH"},
437 {IXGBE_TDLEN(0), "TDLEN"},
438 {IXGBE_TDH(0), "TDH"},
439 {IXGBE_TDT(0), "TDT"},
440 {IXGBE_TXDCTL(0), "TXDCTL"},
442 /* List Terminator */
448 * ixgbe_regdump - register printout routine
450 static void ixgbe_regdump(struct ixgbe_hw
*hw
, struct ixgbe_reg_info
*reginfo
)
456 switch (reginfo
->ofs
) {
457 case IXGBE_SRRCTL(0):
458 for (i
= 0; i
< 64; i
++)
459 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_SRRCTL(i
));
461 case IXGBE_DCA_RXCTRL(0):
462 for (i
= 0; i
< 64; i
++)
463 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_DCA_RXCTRL(i
));
466 for (i
= 0; i
< 64; i
++)
467 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDLEN(i
));
470 for (i
= 0; i
< 64; i
++)
471 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDH(i
));
474 for (i
= 0; i
< 64; i
++)
475 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDT(i
));
477 case IXGBE_RXDCTL(0):
478 for (i
= 0; i
< 64; i
++)
479 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RXDCTL(i
));
482 for (i
= 0; i
< 64; i
++)
483 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDBAL(i
));
486 for (i
= 0; i
< 64; i
++)
487 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDBAH(i
));
490 for (i
= 0; i
< 64; i
++)
491 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDBAL(i
));
494 for (i
= 0; i
< 64; i
++)
495 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDBAH(i
));
498 for (i
= 0; i
< 64; i
++)
499 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDLEN(i
));
502 for (i
= 0; i
< 64; i
++)
503 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDH(i
));
506 for (i
= 0; i
< 64; i
++)
507 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDT(i
));
509 case IXGBE_TXDCTL(0):
510 for (i
= 0; i
< 64; i
++)
511 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TXDCTL(i
));
514 pr_info("%-15s %08x\n", reginfo
->name
,
515 IXGBE_READ_REG(hw
, reginfo
->ofs
));
519 for (i
= 0; i
< 8; i
++) {
520 snprintf(rname
, 16, "%s[%d-%d]", reginfo
->name
, i
*8, i
*8+7);
521 pr_err("%-15s", rname
);
522 for (j
= 0; j
< 8; j
++)
523 pr_cont(" %08x", regs
[i
*8+j
]);
530 * ixgbe_dump - Print registers, tx-rings and rx-rings
532 static void ixgbe_dump(struct ixgbe_adapter
*adapter
)
534 struct net_device
*netdev
= adapter
->netdev
;
535 struct ixgbe_hw
*hw
= &adapter
->hw
;
536 struct ixgbe_reg_info
*reginfo
;
538 struct ixgbe_ring
*tx_ring
;
539 struct ixgbe_tx_buffer
*tx_buffer
;
540 union ixgbe_adv_tx_desc
*tx_desc
;
541 struct my_u0
{ u64 a
; u64 b
; } *u0
;
542 struct ixgbe_ring
*rx_ring
;
543 union ixgbe_adv_rx_desc
*rx_desc
;
544 struct ixgbe_rx_buffer
*rx_buffer_info
;
548 if (!netif_msg_hw(adapter
))
551 /* Print netdevice Info */
553 dev_info(&adapter
->pdev
->dev
, "Net device Info\n");
554 pr_info("Device Name state "
555 "trans_start last_rx\n");
556 pr_info("%-15s %016lX %016lX %016lX\n",
563 /* Print Registers */
564 dev_info(&adapter
->pdev
->dev
, "Register Dump\n");
565 pr_info(" Register Name Value\n");
566 for (reginfo
= (struct ixgbe_reg_info
*)ixgbe_reg_info_tbl
;
567 reginfo
->name
; reginfo
++) {
568 ixgbe_regdump(hw
, reginfo
);
571 /* Print TX Ring Summary */
572 if (!netdev
|| !netif_running(netdev
))
575 dev_info(&adapter
->pdev
->dev
, "TX Rings Summary\n");
576 pr_info(" %s %s %s %s\n",
577 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
578 "leng", "ntw", "timestamp");
579 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
580 tx_ring
= adapter
->tx_ring
[n
];
581 tx_buffer
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_clean
];
582 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
583 n
, tx_ring
->next_to_use
, tx_ring
->next_to_clean
,
584 (u64
)dma_unmap_addr(tx_buffer
, dma
),
585 dma_unmap_len(tx_buffer
, len
),
586 tx_buffer
->next_to_watch
,
587 (u64
)tx_buffer
->time_stamp
);
591 if (!netif_msg_tx_done(adapter
))
592 goto rx_ring_summary
;
594 dev_info(&adapter
->pdev
->dev
, "TX Rings Dump\n");
596 /* Transmit Descriptor Formats
598 * 82598 Advanced Transmit Descriptor
599 * +--------------------------------------------------------------+
600 * 0 | Buffer Address [63:0] |
601 * +--------------------------------------------------------------+
602 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
603 * +--------------------------------------------------------------+
604 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
606 * 82598 Advanced Transmit Descriptor (Write-Back Format)
607 * +--------------------------------------------------------------+
609 * +--------------------------------------------------------------+
610 * 8 | RSV | STA | NXTSEQ |
611 * +--------------------------------------------------------------+
614 * 82599+ Advanced Transmit Descriptor
615 * +--------------------------------------------------------------+
616 * 0 | Buffer Address [63:0] |
617 * +--------------------------------------------------------------+
618 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
619 * +--------------------------------------------------------------+
620 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
622 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
623 * +--------------------------------------------------------------+
625 * +--------------------------------------------------------------+
626 * 8 | RSV | STA | RSV |
627 * +--------------------------------------------------------------+
631 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
632 tx_ring
= adapter
->tx_ring
[n
];
633 pr_info("------------------------------------\n");
634 pr_info("TX QUEUE INDEX = %d\n", tx_ring
->queue_index
);
635 pr_info("------------------------------------\n");
636 pr_info("%s%s %s %s %s %s\n",
637 "T [desc] [address 63:0 ] ",
638 "[PlPOIdStDDt Ln] [bi->dma ] ",
639 "leng", "ntw", "timestamp", "bi->skb");
641 for (i
= 0; tx_ring
->desc
&& (i
< tx_ring
->count
); i
++) {
642 tx_desc
= IXGBE_TX_DESC(tx_ring
, i
);
643 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
644 u0
= (struct my_u0
*)tx_desc
;
645 if (dma_unmap_len(tx_buffer
, len
) > 0) {
646 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p",
650 (u64
)dma_unmap_addr(tx_buffer
, dma
),
651 dma_unmap_len(tx_buffer
, len
),
652 tx_buffer
->next_to_watch
,
653 (u64
)tx_buffer
->time_stamp
,
655 if (i
== tx_ring
->next_to_use
&&
656 i
== tx_ring
->next_to_clean
)
658 else if (i
== tx_ring
->next_to_use
)
660 else if (i
== tx_ring
->next_to_clean
)
665 if (netif_msg_pktdata(adapter
) &&
667 print_hex_dump(KERN_INFO
, "",
668 DUMP_PREFIX_ADDRESS
, 16, 1,
669 tx_buffer
->skb
->data
,
670 dma_unmap_len(tx_buffer
, len
),
676 /* Print RX Rings Summary */
678 dev_info(&adapter
->pdev
->dev
, "RX Rings Summary\n");
679 pr_info("Queue [NTU] [NTC]\n");
680 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
681 rx_ring
= adapter
->rx_ring
[n
];
682 pr_info("%5d %5X %5X\n",
683 n
, rx_ring
->next_to_use
, rx_ring
->next_to_clean
);
687 if (!netif_msg_rx_status(adapter
))
690 dev_info(&adapter
->pdev
->dev
, "RX Rings Dump\n");
692 /* Receive Descriptor Formats
694 * 82598 Advanced Receive Descriptor (Read) Format
696 * +-----------------------------------------------------+
697 * 0 | Packet Buffer Address [63:1] |A0/NSE|
698 * +----------------------------------------------+------+
699 * 8 | Header Buffer Address [63:1] | DD |
700 * +-----------------------------------------------------+
703 * 82598 Advanced Receive Descriptor (Write-Back) Format
705 * 63 48 47 32 31 30 21 20 16 15 4 3 0
706 * +------------------------------------------------------+
707 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
708 * | Packet | IP | | | | Type | Type |
709 * | Checksum | Ident | | | | | |
710 * +------------------------------------------------------+
711 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
712 * +------------------------------------------------------+
713 * 63 48 47 32 31 20 19 0
715 * 82599+ Advanced Receive Descriptor (Read) Format
717 * +-----------------------------------------------------+
718 * 0 | Packet Buffer Address [63:1] |A0/NSE|
719 * +----------------------------------------------+------+
720 * 8 | Header Buffer Address [63:1] | DD |
721 * +-----------------------------------------------------+
724 * 82599+ Advanced Receive Descriptor (Write-Back) Format
726 * 63 48 47 32 31 30 21 20 17 16 4 3 0
727 * +------------------------------------------------------+
728 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
729 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
730 * |/ Flow Dir Flt ID | | | | | |
731 * +------------------------------------------------------+
732 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
733 * +------------------------------------------------------+
734 * 63 48 47 32 31 20 19 0
737 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
738 rx_ring
= adapter
->rx_ring
[n
];
739 pr_info("------------------------------------\n");
740 pr_info("RX QUEUE INDEX = %d\n", rx_ring
->queue_index
);
741 pr_info("------------------------------------\n");
743 "R [desc] [ PktBuf A0] ",
744 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
745 "<-- Adv Rx Read format\n");
747 "RWB[desc] [PcsmIpSHl PtRs] ",
748 "[vl er S cks ln] ---------------- [bi->skb ] ",
749 "<-- Adv Rx Write-Back format\n");
751 for (i
= 0; i
< rx_ring
->count
; i
++) {
752 rx_buffer_info
= &rx_ring
->rx_buffer_info
[i
];
753 rx_desc
= IXGBE_RX_DESC(rx_ring
, i
);
754 u0
= (struct my_u0
*)rx_desc
;
755 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
756 if (staterr
& IXGBE_RXD_STAT_DD
) {
757 /* Descriptor Done */
758 pr_info("RWB[0x%03X] %016llX "
759 "%016llX ---------------- %p", i
,
762 rx_buffer_info
->skb
);
764 pr_info("R [0x%03X] %016llX "
765 "%016llX %016llX %p", i
,
768 (u64
)rx_buffer_info
->dma
,
769 rx_buffer_info
->skb
);
771 if (netif_msg_pktdata(adapter
) &&
772 rx_buffer_info
->dma
) {
773 print_hex_dump(KERN_INFO
, "",
774 DUMP_PREFIX_ADDRESS
, 16, 1,
775 page_address(rx_buffer_info
->page
) +
776 rx_buffer_info
->page_offset
,
777 ixgbe_rx_bufsz(rx_ring
), true);
781 if (i
== rx_ring
->next_to_use
)
783 else if (i
== rx_ring
->next_to_clean
)
792 static void ixgbe_release_hw_control(struct ixgbe_adapter
*adapter
)
796 /* Let firmware take over control of h/w */
797 ctrl_ext
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_CTRL_EXT
);
798 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_CTRL_EXT
,
799 ctrl_ext
& ~IXGBE_CTRL_EXT_DRV_LOAD
);
802 static void ixgbe_get_hw_control(struct ixgbe_adapter
*adapter
)
806 /* Let firmware know the driver has taken over */
807 ctrl_ext
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_CTRL_EXT
);
808 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_CTRL_EXT
,
809 ctrl_ext
| IXGBE_CTRL_EXT_DRV_LOAD
);
813 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
814 * @adapter: pointer to adapter struct
815 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
816 * @queue: queue to map the corresponding interrupt to
817 * @msix_vector: the vector to map to the corresponding queue
820 static void ixgbe_set_ivar(struct ixgbe_adapter
*adapter
, s8 direction
,
821 u8 queue
, u8 msix_vector
)
824 struct ixgbe_hw
*hw
= &adapter
->hw
;
825 switch (hw
->mac
.type
) {
826 case ixgbe_mac_82598EB
:
827 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
830 index
= (((direction
* 64) + queue
) >> 2) & 0x1F;
831 ivar
= IXGBE_READ_REG(hw
, IXGBE_IVAR(index
));
832 ivar
&= ~(0xFF << (8 * (queue
& 0x3)));
833 ivar
|= (msix_vector
<< (8 * (queue
& 0x3)));
834 IXGBE_WRITE_REG(hw
, IXGBE_IVAR(index
), ivar
);
836 case ixgbe_mac_82599EB
:
838 if (direction
== -1) {
840 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
841 index
= ((queue
& 1) * 8);
842 ivar
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_IVAR_MISC
);
843 ivar
&= ~(0xFF << index
);
844 ivar
|= (msix_vector
<< index
);
845 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_IVAR_MISC
, ivar
);
848 /* tx or rx causes */
849 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
850 index
= ((16 * (queue
& 1)) + (8 * direction
));
851 ivar
= IXGBE_READ_REG(hw
, IXGBE_IVAR(queue
>> 1));
852 ivar
&= ~(0xFF << index
);
853 ivar
|= (msix_vector
<< index
);
854 IXGBE_WRITE_REG(hw
, IXGBE_IVAR(queue
>> 1), ivar
);
862 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter
*adapter
,
867 switch (adapter
->hw
.mac
.type
) {
868 case ixgbe_mac_82598EB
:
869 mask
= (IXGBE_EIMS_RTX_QUEUE
& qmask
);
870 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICS
, mask
);
872 case ixgbe_mac_82599EB
:
874 mask
= (qmask
& 0xFFFFFFFF);
875 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICS_EX(0), mask
);
876 mask
= (qmask
>> 32);
877 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICS_EX(1), mask
);
884 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring
*ring
,
885 struct ixgbe_tx_buffer
*tx_buffer
)
887 if (tx_buffer
->skb
) {
888 dev_kfree_skb_any(tx_buffer
->skb
);
889 if (dma_unmap_len(tx_buffer
, len
))
890 dma_unmap_single(ring
->dev
,
891 dma_unmap_addr(tx_buffer
, dma
),
892 dma_unmap_len(tx_buffer
, len
),
894 } else if (dma_unmap_len(tx_buffer
, len
)) {
895 dma_unmap_page(ring
->dev
,
896 dma_unmap_addr(tx_buffer
, dma
),
897 dma_unmap_len(tx_buffer
, len
),
900 tx_buffer
->next_to_watch
= NULL
;
901 tx_buffer
->skb
= NULL
;
902 dma_unmap_len_set(tx_buffer
, len
, 0);
903 /* tx_buffer must be completely set up in the transmit path */
906 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter
*adapter
)
908 struct ixgbe_hw
*hw
= &adapter
->hw
;
909 struct ixgbe_hw_stats
*hwstats
= &adapter
->stats
;
913 if ((hw
->fc
.current_mode
!= ixgbe_fc_full
) &&
914 (hw
->fc
.current_mode
!= ixgbe_fc_rx_pause
))
917 switch (hw
->mac
.type
) {
918 case ixgbe_mac_82598EB
:
919 data
= IXGBE_READ_REG(hw
, IXGBE_LXOFFRXC
);
922 data
= IXGBE_READ_REG(hw
, IXGBE_LXOFFRXCNT
);
924 hwstats
->lxoffrxc
+= data
;
926 /* refill credits (no tx hang) if we received xoff */
930 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
931 clear_bit(__IXGBE_HANG_CHECK_ARMED
,
932 &adapter
->tx_ring
[i
]->state
);
935 static void ixgbe_update_xoff_received(struct ixgbe_adapter
*adapter
)
937 struct ixgbe_hw
*hw
= &adapter
->hw
;
938 struct ixgbe_hw_stats
*hwstats
= &adapter
->stats
;
942 bool pfc_en
= adapter
->dcb_cfg
.pfc_mode_enable
;
944 if (adapter
->ixgbe_ieee_pfc
)
945 pfc_en
|= !!(adapter
->ixgbe_ieee_pfc
->pfc_en
);
947 if (!(adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) || !pfc_en
) {
948 ixgbe_update_xoff_rx_lfc(adapter
);
952 /* update stats for each tc, only valid with PFC enabled */
953 for (i
= 0; i
< MAX_TX_PACKET_BUFFERS
; i
++) {
956 switch (hw
->mac
.type
) {
957 case ixgbe_mac_82598EB
:
958 pxoffrxc
= IXGBE_READ_REG(hw
, IXGBE_PXOFFRXC(i
));
961 pxoffrxc
= IXGBE_READ_REG(hw
, IXGBE_PXOFFRXCNT(i
));
963 hwstats
->pxoffrxc
[i
] += pxoffrxc
;
964 /* Get the TC for given UP */
965 tc
= netdev_get_prio_tc_map(adapter
->netdev
, i
);
966 xoff
[tc
] += pxoffrxc
;
969 /* disarm tx queues that have received xoff frames */
970 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
971 struct ixgbe_ring
*tx_ring
= adapter
->tx_ring
[i
];
973 tc
= tx_ring
->dcb_tc
;
975 clear_bit(__IXGBE_HANG_CHECK_ARMED
, &tx_ring
->state
);
979 static u64
ixgbe_get_tx_completed(struct ixgbe_ring
*ring
)
981 return ring
->stats
.packets
;
984 static u64
ixgbe_get_tx_pending(struct ixgbe_ring
*ring
)
986 struct ixgbe_adapter
*adapter
;
990 if (ring
->l2_accel_priv
)
991 adapter
= ring
->l2_accel_priv
->real_adapter
;
993 adapter
= netdev_priv(ring
->netdev
);
996 head
= IXGBE_READ_REG(hw
, IXGBE_TDH(ring
->reg_idx
));
997 tail
= IXGBE_READ_REG(hw
, IXGBE_TDT(ring
->reg_idx
));
1000 return (head
< tail
) ?
1001 tail
- head
: (tail
+ ring
->count
- head
);
1006 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring
*tx_ring
)
1008 u32 tx_done
= ixgbe_get_tx_completed(tx_ring
);
1009 u32 tx_done_old
= tx_ring
->tx_stats
.tx_done_old
;
1010 u32 tx_pending
= ixgbe_get_tx_pending(tx_ring
);
1012 clear_check_for_tx_hang(tx_ring
);
1015 * Check for a hung queue, but be thorough. This verifies
1016 * that a transmit has been completed since the previous
1017 * check AND there is at least one packet pending. The
1018 * ARMED bit is set to indicate a potential hang. The
1019 * bit is cleared if a pause frame is received to remove
1020 * false hang detection due to PFC or 802.3x frames. By
1021 * requiring this to fail twice we avoid races with
1022 * pfc clearing the ARMED bit and conditions where we
1023 * run the check_tx_hang logic with a transmit completion
1024 * pending but without time to complete it yet.
1026 if (tx_done_old
== tx_done
&& tx_pending
)
1027 /* make sure it is true for two checks in a row */
1028 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED
,
1030 /* update completed stats and continue */
1031 tx_ring
->tx_stats
.tx_done_old
= tx_done
;
1032 /* reset the countdown */
1033 clear_bit(__IXGBE_HANG_CHECK_ARMED
, &tx_ring
->state
);
1039 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1040 * @adapter: driver private struct
1042 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter
*adapter
)
1045 /* Do the reset outside of interrupt context */
1046 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
1047 adapter
->flags2
|= IXGBE_FLAG2_RESET_REQUESTED
;
1048 e_warn(drv
, "initiating reset due to tx timeout\n");
1049 ixgbe_service_event_schedule(adapter
);
1054 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1055 * @q_vector: structure containing interrupt and ring information
1056 * @tx_ring: tx ring to clean
1058 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector
*q_vector
,
1059 struct ixgbe_ring
*tx_ring
)
1061 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1062 struct ixgbe_tx_buffer
*tx_buffer
;
1063 union ixgbe_adv_tx_desc
*tx_desc
;
1064 unsigned int total_bytes
= 0, total_packets
= 0;
1065 unsigned int budget
= q_vector
->tx
.work_limit
;
1066 unsigned int i
= tx_ring
->next_to_clean
;
1068 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
1071 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
1072 tx_desc
= IXGBE_TX_DESC(tx_ring
, i
);
1073 i
-= tx_ring
->count
;
1076 union ixgbe_adv_tx_desc
*eop_desc
= tx_buffer
->next_to_watch
;
1078 /* if next_to_watch is not set then there is no work pending */
1082 /* prevent any other reads prior to eop_desc */
1083 read_barrier_depends();
1085 /* if DD is not set pending work has not been completed */
1086 if (!(eop_desc
->wb
.status
& cpu_to_le32(IXGBE_TXD_STAT_DD
)))
1089 /* clear next_to_watch to prevent false hangs */
1090 tx_buffer
->next_to_watch
= NULL
;
1092 /* update the statistics for this packet */
1093 total_bytes
+= tx_buffer
->bytecount
;
1094 total_packets
+= tx_buffer
->gso_segs
;
1097 dev_consume_skb_any(tx_buffer
->skb
);
1099 /* unmap skb header data */
1100 dma_unmap_single(tx_ring
->dev
,
1101 dma_unmap_addr(tx_buffer
, dma
),
1102 dma_unmap_len(tx_buffer
, len
),
1105 /* clear tx_buffer data */
1106 tx_buffer
->skb
= NULL
;
1107 dma_unmap_len_set(tx_buffer
, len
, 0);
1109 /* unmap remaining buffers */
1110 while (tx_desc
!= eop_desc
) {
1115 i
-= tx_ring
->count
;
1116 tx_buffer
= tx_ring
->tx_buffer_info
;
1117 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
1120 /* unmap any remaining paged data */
1121 if (dma_unmap_len(tx_buffer
, len
)) {
1122 dma_unmap_page(tx_ring
->dev
,
1123 dma_unmap_addr(tx_buffer
, dma
),
1124 dma_unmap_len(tx_buffer
, len
),
1126 dma_unmap_len_set(tx_buffer
, len
, 0);
1130 /* move us one more past the eop_desc for start of next pkt */
1135 i
-= tx_ring
->count
;
1136 tx_buffer
= tx_ring
->tx_buffer_info
;
1137 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
1140 /* issue prefetch for next Tx descriptor */
1143 /* update budget accounting */
1145 } while (likely(budget
));
1147 i
+= tx_ring
->count
;
1148 tx_ring
->next_to_clean
= i
;
1149 u64_stats_update_begin(&tx_ring
->syncp
);
1150 tx_ring
->stats
.bytes
+= total_bytes
;
1151 tx_ring
->stats
.packets
+= total_packets
;
1152 u64_stats_update_end(&tx_ring
->syncp
);
1153 q_vector
->tx
.total_bytes
+= total_bytes
;
1154 q_vector
->tx
.total_packets
+= total_packets
;
1156 if (check_for_tx_hang(tx_ring
) && ixgbe_check_tx_hang(tx_ring
)) {
1157 /* schedule immediate reset if we believe we hung */
1158 struct ixgbe_hw
*hw
= &adapter
->hw
;
1159 e_err(drv
, "Detected Tx Unit Hang\n"
1161 " TDH, TDT <%x>, <%x>\n"
1162 " next_to_use <%x>\n"
1163 " next_to_clean <%x>\n"
1164 "tx_buffer_info[next_to_clean]\n"
1165 " time_stamp <%lx>\n"
1167 tx_ring
->queue_index
,
1168 IXGBE_READ_REG(hw
, IXGBE_TDH(tx_ring
->reg_idx
)),
1169 IXGBE_READ_REG(hw
, IXGBE_TDT(tx_ring
->reg_idx
)),
1170 tx_ring
->next_to_use
, i
,
1171 tx_ring
->tx_buffer_info
[i
].time_stamp
, jiffies
);
1173 netif_stop_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
1176 "tx hang %d detected on queue %d, resetting adapter\n",
1177 adapter
->tx_timeout_count
+ 1, tx_ring
->queue_index
);
1179 /* schedule immediate reset if we believe we hung */
1180 ixgbe_tx_timeout_reset(adapter
);
1182 /* the adapter is about to reset, no point in enabling stuff */
1186 netdev_tx_completed_queue(txring_txq(tx_ring
),
1187 total_packets
, total_bytes
);
1189 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1190 if (unlikely(total_packets
&& netif_carrier_ok(tx_ring
->netdev
) &&
1191 (ixgbe_desc_unused(tx_ring
) >= TX_WAKE_THRESHOLD
))) {
1192 /* Make sure that anybody stopping the queue after this
1193 * sees the new next_to_clean.
1196 if (__netif_subqueue_stopped(tx_ring
->netdev
,
1197 tx_ring
->queue_index
)
1198 && !test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
1199 netif_wake_subqueue(tx_ring
->netdev
,
1200 tx_ring
->queue_index
);
1201 ++tx_ring
->tx_stats
.restart_queue
;
1208 #ifdef CONFIG_IXGBE_DCA
1209 static void ixgbe_update_tx_dca(struct ixgbe_adapter
*adapter
,
1210 struct ixgbe_ring
*tx_ring
,
1213 struct ixgbe_hw
*hw
= &adapter
->hw
;
1214 u32 txctrl
= dca3_get_tag(tx_ring
->dev
, cpu
);
1217 switch (hw
->mac
.type
) {
1218 case ixgbe_mac_82598EB
:
1219 reg_offset
= IXGBE_DCA_TXCTRL(tx_ring
->reg_idx
);
1221 case ixgbe_mac_82599EB
:
1222 case ixgbe_mac_X540
:
1223 reg_offset
= IXGBE_DCA_TXCTRL_82599(tx_ring
->reg_idx
);
1224 txctrl
<<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599
;
1227 /* for unknown hardware do not write register */
1232 * We can enable relaxed ordering for reads, but not writes when
1233 * DCA is enabled. This is due to a known issue in some chipsets
1234 * which will cause the DCA tag to be cleared.
1236 txctrl
|= IXGBE_DCA_TXCTRL_DESC_RRO_EN
|
1237 IXGBE_DCA_TXCTRL_DATA_RRO_EN
|
1238 IXGBE_DCA_TXCTRL_DESC_DCA_EN
;
1240 IXGBE_WRITE_REG(hw
, reg_offset
, txctrl
);
1243 static void ixgbe_update_rx_dca(struct ixgbe_adapter
*adapter
,
1244 struct ixgbe_ring
*rx_ring
,
1247 struct ixgbe_hw
*hw
= &adapter
->hw
;
1248 u32 rxctrl
= dca3_get_tag(rx_ring
->dev
, cpu
);
1249 u8 reg_idx
= rx_ring
->reg_idx
;
1252 switch (hw
->mac
.type
) {
1253 case ixgbe_mac_82599EB
:
1254 case ixgbe_mac_X540
:
1255 rxctrl
<<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599
;
1262 * We can enable relaxed ordering for reads, but not writes when
1263 * DCA is enabled. This is due to a known issue in some chipsets
1264 * which will cause the DCA tag to be cleared.
1266 rxctrl
|= IXGBE_DCA_RXCTRL_DESC_RRO_EN
|
1267 IXGBE_DCA_RXCTRL_DESC_DCA_EN
;
1269 IXGBE_WRITE_REG(hw
, IXGBE_DCA_RXCTRL(reg_idx
), rxctrl
);
1272 static void ixgbe_update_dca(struct ixgbe_q_vector
*q_vector
)
1274 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1275 struct ixgbe_ring
*ring
;
1276 int cpu
= get_cpu();
1278 if (q_vector
->cpu
== cpu
)
1281 ixgbe_for_each_ring(ring
, q_vector
->tx
)
1282 ixgbe_update_tx_dca(adapter
, ring
, cpu
);
1284 ixgbe_for_each_ring(ring
, q_vector
->rx
)
1285 ixgbe_update_rx_dca(adapter
, ring
, cpu
);
1287 q_vector
->cpu
= cpu
;
1292 static void ixgbe_setup_dca(struct ixgbe_adapter
*adapter
)
1296 if (!(adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
))
1299 /* always use CB2 mode, difference is masked in the CB driver */
1300 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
, 2);
1302 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
1303 adapter
->q_vector
[i
]->cpu
= -1;
1304 ixgbe_update_dca(adapter
->q_vector
[i
]);
1308 static int __ixgbe_notify_dca(struct device
*dev
, void *data
)
1310 struct ixgbe_adapter
*adapter
= dev_get_drvdata(dev
);
1311 unsigned long event
= *(unsigned long *)data
;
1313 if (!(adapter
->flags
& IXGBE_FLAG_DCA_CAPABLE
))
1317 case DCA_PROVIDER_ADD
:
1318 /* if we're already enabled, don't do it again */
1319 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
1321 if (dca_add_requester(dev
) == 0) {
1322 adapter
->flags
|= IXGBE_FLAG_DCA_ENABLED
;
1323 ixgbe_setup_dca(adapter
);
1326 /* Fall Through since DCA is disabled. */
1327 case DCA_PROVIDER_REMOVE
:
1328 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
) {
1329 dca_remove_requester(dev
);
1330 adapter
->flags
&= ~IXGBE_FLAG_DCA_ENABLED
;
1331 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
, 1);
1339 #endif /* CONFIG_IXGBE_DCA */
1340 static inline void ixgbe_rx_hash(struct ixgbe_ring
*ring
,
1341 union ixgbe_adv_rx_desc
*rx_desc
,
1342 struct sk_buff
*skb
)
1344 if (ring
->netdev
->features
& NETIF_F_RXHASH
)
1346 le32_to_cpu(rx_desc
->wb
.lower
.hi_dword
.rss
),
1352 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1353 * @ring: structure containing ring specific data
1354 * @rx_desc: advanced rx descriptor
1356 * Returns : true if it is FCoE pkt
1358 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring
*ring
,
1359 union ixgbe_adv_rx_desc
*rx_desc
)
1361 __le16 pkt_info
= rx_desc
->wb
.lower
.lo_dword
.hs_rss
.pkt_info
;
1363 return test_bit(__IXGBE_RX_FCOE
, &ring
->state
) &&
1364 ((pkt_info
& cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK
)) ==
1365 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE
<<
1366 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT
)));
1369 #endif /* IXGBE_FCOE */
1371 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1372 * @ring: structure containing ring specific data
1373 * @rx_desc: current Rx descriptor being processed
1374 * @skb: skb currently being received and modified
1376 static inline void ixgbe_rx_checksum(struct ixgbe_ring
*ring
,
1377 union ixgbe_adv_rx_desc
*rx_desc
,
1378 struct sk_buff
*skb
)
1380 skb_checksum_none_assert(skb
);
1382 /* Rx csum disabled */
1383 if (!(ring
->netdev
->features
& NETIF_F_RXCSUM
))
1386 /* if IP and error */
1387 if (ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_IPCS
) &&
1388 ixgbe_test_staterr(rx_desc
, IXGBE_RXDADV_ERR_IPE
)) {
1389 ring
->rx_stats
.csum_err
++;
1393 if (!ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_L4CS
))
1396 if (ixgbe_test_staterr(rx_desc
, IXGBE_RXDADV_ERR_TCPE
)) {
1397 __le16 pkt_info
= rx_desc
->wb
.lower
.lo_dword
.hs_rss
.pkt_info
;
1400 * 82599 errata, UDP frames with a 0 checksum can be marked as
1403 if ((pkt_info
& cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP
)) &&
1404 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR
, &ring
->state
))
1407 ring
->rx_stats
.csum_err
++;
1411 /* It must be a TCP or UDP packet with a valid checksum */
1412 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1415 static inline void ixgbe_release_rx_desc(struct ixgbe_ring
*rx_ring
, u32 val
)
1417 rx_ring
->next_to_use
= val
;
1419 /* update next to alloc since we have filled the ring */
1420 rx_ring
->next_to_alloc
= val
;
1422 * Force memory writes to complete before letting h/w
1423 * know there are new descriptors to fetch. (Only
1424 * applicable for weak-ordered memory model archs,
1428 ixgbe_write_tail(rx_ring
, val
);
1431 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring
*rx_ring
,
1432 struct ixgbe_rx_buffer
*bi
)
1434 struct page
*page
= bi
->page
;
1435 dma_addr_t dma
= bi
->dma
;
1437 /* since we are recycling buffers we should seldom need to alloc */
1441 /* alloc new page for storage */
1442 if (likely(!page
)) {
1443 page
= __skb_alloc_pages(GFP_ATOMIC
| __GFP_COLD
| __GFP_COMP
,
1444 bi
->skb
, ixgbe_rx_pg_order(rx_ring
));
1445 if (unlikely(!page
)) {
1446 rx_ring
->rx_stats
.alloc_rx_page_failed
++;
1452 /* map page for use */
1453 dma
= dma_map_page(rx_ring
->dev
, page
, 0,
1454 ixgbe_rx_pg_size(rx_ring
), DMA_FROM_DEVICE
);
1457 * if mapping failed free memory back to system since
1458 * there isn't much point in holding memory we can't use
1460 if (dma_mapping_error(rx_ring
->dev
, dma
)) {
1461 __free_pages(page
, ixgbe_rx_pg_order(rx_ring
));
1464 rx_ring
->rx_stats
.alloc_rx_page_failed
++;
1469 bi
->page_offset
= 0;
1475 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1476 * @rx_ring: ring to place buffers on
1477 * @cleaned_count: number of buffers to replace
1479 void ixgbe_alloc_rx_buffers(struct ixgbe_ring
*rx_ring
, u16 cleaned_count
)
1481 union ixgbe_adv_rx_desc
*rx_desc
;
1482 struct ixgbe_rx_buffer
*bi
;
1483 u16 i
= rx_ring
->next_to_use
;
1489 rx_desc
= IXGBE_RX_DESC(rx_ring
, i
);
1490 bi
= &rx_ring
->rx_buffer_info
[i
];
1491 i
-= rx_ring
->count
;
1494 if (!ixgbe_alloc_mapped_page(rx_ring
, bi
))
1498 * Refresh the desc even if buffer_addrs didn't change
1499 * because each write-back erases this info.
1501 rx_desc
->read
.pkt_addr
= cpu_to_le64(bi
->dma
+ bi
->page_offset
);
1507 rx_desc
= IXGBE_RX_DESC(rx_ring
, 0);
1508 bi
= rx_ring
->rx_buffer_info
;
1509 i
-= rx_ring
->count
;
1512 /* clear the hdr_addr for the next_to_use descriptor */
1513 rx_desc
->read
.hdr_addr
= 0;
1516 } while (cleaned_count
);
1518 i
+= rx_ring
->count
;
1520 if (rx_ring
->next_to_use
!= i
)
1521 ixgbe_release_rx_desc(rx_ring
, i
);
1524 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring
*ring
,
1525 struct sk_buff
*skb
)
1527 u16 hdr_len
= skb_headlen(skb
);
1529 /* set gso_size to avoid messing up TCP MSS */
1530 skb_shinfo(skb
)->gso_size
= DIV_ROUND_UP((skb
->len
- hdr_len
),
1531 IXGBE_CB(skb
)->append_cnt
);
1532 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
1535 static void ixgbe_update_rsc_stats(struct ixgbe_ring
*rx_ring
,
1536 struct sk_buff
*skb
)
1538 /* if append_cnt is 0 then frame is not RSC */
1539 if (!IXGBE_CB(skb
)->append_cnt
)
1542 rx_ring
->rx_stats
.rsc_count
+= IXGBE_CB(skb
)->append_cnt
;
1543 rx_ring
->rx_stats
.rsc_flush
++;
1545 ixgbe_set_rsc_gso_size(rx_ring
, skb
);
1547 /* gso_size is computed using append_cnt so always clear it last */
1548 IXGBE_CB(skb
)->append_cnt
= 0;
1552 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1553 * @rx_ring: rx descriptor ring packet is being transacted on
1554 * @rx_desc: pointer to the EOP Rx descriptor
1555 * @skb: pointer to current skb being populated
1557 * This function checks the ring, descriptor, and packet information in
1558 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1559 * other fields within the skb.
1561 static void ixgbe_process_skb_fields(struct ixgbe_ring
*rx_ring
,
1562 union ixgbe_adv_rx_desc
*rx_desc
,
1563 struct sk_buff
*skb
)
1565 struct net_device
*dev
= rx_ring
->netdev
;
1567 ixgbe_update_rsc_stats(rx_ring
, skb
);
1569 ixgbe_rx_hash(rx_ring
, rx_desc
, skb
);
1571 ixgbe_rx_checksum(rx_ring
, rx_desc
, skb
);
1573 if (unlikely(ixgbe_test_staterr(rx_desc
, IXGBE_RXDADV_STAT_TS
)))
1574 ixgbe_ptp_rx_hwtstamp(rx_ring
->q_vector
->adapter
, skb
);
1576 if ((dev
->features
& NETIF_F_HW_VLAN_CTAG_RX
) &&
1577 ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_VP
)) {
1578 u16 vid
= le16_to_cpu(rx_desc
->wb
.upper
.vlan
);
1579 __vlan_hwaccel_put_tag(skb
, htons(ETH_P_8021Q
), vid
);
1582 skb_record_rx_queue(skb
, rx_ring
->queue_index
);
1584 skb
->protocol
= eth_type_trans(skb
, dev
);
1587 static void ixgbe_rx_skb(struct ixgbe_q_vector
*q_vector
,
1588 struct sk_buff
*skb
)
1590 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1592 if (ixgbe_qv_busy_polling(q_vector
))
1593 netif_receive_skb(skb
);
1594 else if (!(adapter
->flags
& IXGBE_FLAG_IN_NETPOLL
))
1595 napi_gro_receive(&q_vector
->napi
, skb
);
1601 * ixgbe_is_non_eop - process handling of non-EOP buffers
1602 * @rx_ring: Rx ring being processed
1603 * @rx_desc: Rx descriptor for current buffer
1604 * @skb: Current socket buffer containing buffer in progress
1606 * This function updates next to clean. If the buffer is an EOP buffer
1607 * this function exits returning false, otherwise it will place the
1608 * sk_buff in the next buffer to be chained and return true indicating
1609 * that this is in fact a non-EOP buffer.
1611 static bool ixgbe_is_non_eop(struct ixgbe_ring
*rx_ring
,
1612 union ixgbe_adv_rx_desc
*rx_desc
,
1613 struct sk_buff
*skb
)
1615 u32 ntc
= rx_ring
->next_to_clean
+ 1;
1617 /* fetch, update, and store next to clean */
1618 ntc
= (ntc
< rx_ring
->count
) ? ntc
: 0;
1619 rx_ring
->next_to_clean
= ntc
;
1621 prefetch(IXGBE_RX_DESC(rx_ring
, ntc
));
1623 /* update RSC append count if present */
1624 if (ring_is_rsc_enabled(rx_ring
)) {
1625 __le32 rsc_enabled
= rx_desc
->wb
.lower
.lo_dword
.data
&
1626 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK
);
1628 if (unlikely(rsc_enabled
)) {
1629 u32 rsc_cnt
= le32_to_cpu(rsc_enabled
);
1631 rsc_cnt
>>= IXGBE_RXDADV_RSCCNT_SHIFT
;
1632 IXGBE_CB(skb
)->append_cnt
+= rsc_cnt
- 1;
1634 /* update ntc based on RSC value */
1635 ntc
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
1636 ntc
&= IXGBE_RXDADV_NEXTP_MASK
;
1637 ntc
>>= IXGBE_RXDADV_NEXTP_SHIFT
;
1641 /* if we are the last buffer then there is nothing else to do */
1642 if (likely(ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_EOP
)))
1645 /* place skb in next buffer to be received */
1646 rx_ring
->rx_buffer_info
[ntc
].skb
= skb
;
1647 rx_ring
->rx_stats
.non_eop_descs
++;
1653 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1654 * @rx_ring: rx descriptor ring packet is being transacted on
1655 * @skb: pointer to current skb being adjusted
1657 * This function is an ixgbe specific version of __pskb_pull_tail. The
1658 * main difference between this version and the original function is that
1659 * this function can make several assumptions about the state of things
1660 * that allow for significant optimizations versus the standard function.
1661 * As a result we can do things like drop a frag and maintain an accurate
1662 * truesize for the skb.
1664 static void ixgbe_pull_tail(struct ixgbe_ring
*rx_ring
,
1665 struct sk_buff
*skb
)
1667 struct skb_frag_struct
*frag
= &skb_shinfo(skb
)->frags
[0];
1669 unsigned int pull_len
;
1672 * it is valid to use page_address instead of kmap since we are
1673 * working with pages allocated out of the lomem pool per
1674 * alloc_page(GFP_ATOMIC)
1676 va
= skb_frag_address(frag
);
1679 * we need the header to contain the greater of either ETH_HLEN or
1680 * 60 bytes if the skb->len is less than 60 for skb_pad.
1682 pull_len
= eth_get_headlen(va
, IXGBE_RX_HDR_SIZE
);
1684 /* align pull length to size of long to optimize memcpy performance */
1685 skb_copy_to_linear_data(skb
, va
, ALIGN(pull_len
, sizeof(long)));
1687 /* update all of the pointers */
1688 skb_frag_size_sub(frag
, pull_len
);
1689 frag
->page_offset
+= pull_len
;
1690 skb
->data_len
-= pull_len
;
1691 skb
->tail
+= pull_len
;
1695 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1696 * @rx_ring: rx descriptor ring packet is being transacted on
1697 * @skb: pointer to current skb being updated
1699 * This function provides a basic DMA sync up for the first fragment of an
1700 * skb. The reason for doing this is that the first fragment cannot be
1701 * unmapped until we have reached the end of packet descriptor for a buffer
1704 static void ixgbe_dma_sync_frag(struct ixgbe_ring
*rx_ring
,
1705 struct sk_buff
*skb
)
1707 /* if the page was released unmap it, else just sync our portion */
1708 if (unlikely(IXGBE_CB(skb
)->page_released
)) {
1709 dma_unmap_page(rx_ring
->dev
, IXGBE_CB(skb
)->dma
,
1710 ixgbe_rx_pg_size(rx_ring
), DMA_FROM_DEVICE
);
1711 IXGBE_CB(skb
)->page_released
= false;
1713 struct skb_frag_struct
*frag
= &skb_shinfo(skb
)->frags
[0];
1715 dma_sync_single_range_for_cpu(rx_ring
->dev
,
1718 ixgbe_rx_bufsz(rx_ring
),
1721 IXGBE_CB(skb
)->dma
= 0;
1725 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1726 * @rx_ring: rx descriptor ring packet is being transacted on
1727 * @rx_desc: pointer to the EOP Rx descriptor
1728 * @skb: pointer to current skb being fixed
1730 * Check for corrupted packet headers caused by senders on the local L2
1731 * embedded NIC switch not setting up their Tx Descriptors right. These
1732 * should be very rare.
1734 * Also address the case where we are pulling data in on pages only
1735 * and as such no data is present in the skb header.
1737 * In addition if skb is not at least 60 bytes we need to pad it so that
1738 * it is large enough to qualify as a valid Ethernet frame.
1740 * Returns true if an error was encountered and skb was freed.
1742 static bool ixgbe_cleanup_headers(struct ixgbe_ring
*rx_ring
,
1743 union ixgbe_adv_rx_desc
*rx_desc
,
1744 struct sk_buff
*skb
)
1746 struct net_device
*netdev
= rx_ring
->netdev
;
1748 /* verify that the packet does not have any known errors */
1749 if (unlikely(ixgbe_test_staterr(rx_desc
,
1750 IXGBE_RXDADV_ERR_FRAME_ERR_MASK
) &&
1751 !(netdev
->features
& NETIF_F_RXALL
))) {
1752 dev_kfree_skb_any(skb
);
1756 /* place header in linear portion of buffer */
1757 if (skb_is_nonlinear(skb
))
1758 ixgbe_pull_tail(rx_ring
, skb
);
1761 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1762 if (ixgbe_rx_is_fcoe(rx_ring
, rx_desc
))
1766 /* if skb_pad returns an error the skb was freed */
1767 if (unlikely(skb
->len
< 60)) {
1768 int pad_len
= 60 - skb
->len
;
1770 if (skb_pad(skb
, pad_len
))
1772 __skb_put(skb
, pad_len
);
1779 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1780 * @rx_ring: rx descriptor ring to store buffers on
1781 * @old_buff: donor buffer to have page reused
1783 * Synchronizes page for reuse by the adapter
1785 static void ixgbe_reuse_rx_page(struct ixgbe_ring
*rx_ring
,
1786 struct ixgbe_rx_buffer
*old_buff
)
1788 struct ixgbe_rx_buffer
*new_buff
;
1789 u16 nta
= rx_ring
->next_to_alloc
;
1791 new_buff
= &rx_ring
->rx_buffer_info
[nta
];
1793 /* update, and store next to alloc */
1795 rx_ring
->next_to_alloc
= (nta
< rx_ring
->count
) ? nta
: 0;
1797 /* transfer page from old buffer to new buffer */
1798 new_buff
->page
= old_buff
->page
;
1799 new_buff
->dma
= old_buff
->dma
;
1800 new_buff
->page_offset
= old_buff
->page_offset
;
1802 /* sync the buffer for use by the device */
1803 dma_sync_single_range_for_device(rx_ring
->dev
, new_buff
->dma
,
1804 new_buff
->page_offset
,
1805 ixgbe_rx_bufsz(rx_ring
),
1810 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1811 * @rx_ring: rx descriptor ring to transact packets on
1812 * @rx_buffer: buffer containing page to add
1813 * @rx_desc: descriptor containing length of buffer written by hardware
1814 * @skb: sk_buff to place the data into
1816 * This function will add the data contained in rx_buffer->page to the skb.
1817 * This is done either through a direct copy if the data in the buffer is
1818 * less than the skb header size, otherwise it will just attach the page as
1819 * a frag to the skb.
1821 * The function will then update the page offset if necessary and return
1822 * true if the buffer can be reused by the adapter.
1824 static bool ixgbe_add_rx_frag(struct ixgbe_ring
*rx_ring
,
1825 struct ixgbe_rx_buffer
*rx_buffer
,
1826 union ixgbe_adv_rx_desc
*rx_desc
,
1827 struct sk_buff
*skb
)
1829 struct page
*page
= rx_buffer
->page
;
1830 unsigned int size
= le16_to_cpu(rx_desc
->wb
.upper
.length
);
1831 #if (PAGE_SIZE < 8192)
1832 unsigned int truesize
= ixgbe_rx_bufsz(rx_ring
);
1834 unsigned int truesize
= ALIGN(size
, L1_CACHE_BYTES
);
1835 unsigned int last_offset
= ixgbe_rx_pg_size(rx_ring
) -
1836 ixgbe_rx_bufsz(rx_ring
);
1839 if ((size
<= IXGBE_RX_HDR_SIZE
) && !skb_is_nonlinear(skb
)) {
1840 unsigned char *va
= page_address(page
) + rx_buffer
->page_offset
;
1842 memcpy(__skb_put(skb
, size
), va
, ALIGN(size
, sizeof(long)));
1844 /* we can reuse buffer as-is, just make sure it is local */
1845 if (likely(page_to_nid(page
) == numa_node_id()))
1848 /* this page cannot be reused so discard it */
1853 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
,
1854 rx_buffer
->page_offset
, size
, truesize
);
1856 /* avoid re-using remote pages */
1857 if (unlikely(page_to_nid(page
) != numa_node_id()))
1860 #if (PAGE_SIZE < 8192)
1861 /* if we are only owner of page we can reuse it */
1862 if (unlikely(page_count(page
) != 1))
1865 /* flip page offset to other buffer */
1866 rx_buffer
->page_offset
^= truesize
;
1869 * since we are the only owner of the page and we need to
1870 * increment it, just set the value to 2 in order to avoid
1871 * an unecessary locked operation
1873 atomic_set(&page
->_count
, 2);
1875 /* move offset up to the next cache line */
1876 rx_buffer
->page_offset
+= truesize
;
1878 if (rx_buffer
->page_offset
> last_offset
)
1881 /* bump ref count on page before it is given to the stack */
1888 static struct sk_buff
*ixgbe_fetch_rx_buffer(struct ixgbe_ring
*rx_ring
,
1889 union ixgbe_adv_rx_desc
*rx_desc
)
1891 struct ixgbe_rx_buffer
*rx_buffer
;
1892 struct sk_buff
*skb
;
1895 rx_buffer
= &rx_ring
->rx_buffer_info
[rx_ring
->next_to_clean
];
1896 page
= rx_buffer
->page
;
1899 skb
= rx_buffer
->skb
;
1902 void *page_addr
= page_address(page
) +
1903 rx_buffer
->page_offset
;
1905 /* prefetch first cache line of first page */
1906 prefetch(page_addr
);
1907 #if L1_CACHE_BYTES < 128
1908 prefetch(page_addr
+ L1_CACHE_BYTES
);
1911 /* allocate a skb to store the frags */
1912 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
1914 if (unlikely(!skb
)) {
1915 rx_ring
->rx_stats
.alloc_rx_buff_failed
++;
1920 * we will be copying header into skb->data in
1921 * pskb_may_pull so it is in our interest to prefetch
1922 * it now to avoid a possible cache miss
1924 prefetchw(skb
->data
);
1927 * Delay unmapping of the first packet. It carries the
1928 * header information, HW may still access the header
1929 * after the writeback. Only unmap it when EOP is
1932 if (likely(ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_EOP
)))
1935 IXGBE_CB(skb
)->dma
= rx_buffer
->dma
;
1937 if (ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_EOP
))
1938 ixgbe_dma_sync_frag(rx_ring
, skb
);
1941 /* we are reusing so sync this buffer for CPU use */
1942 dma_sync_single_range_for_cpu(rx_ring
->dev
,
1944 rx_buffer
->page_offset
,
1945 ixgbe_rx_bufsz(rx_ring
),
1949 /* pull page into skb */
1950 if (ixgbe_add_rx_frag(rx_ring
, rx_buffer
, rx_desc
, skb
)) {
1951 /* hand second half of page back to the ring */
1952 ixgbe_reuse_rx_page(rx_ring
, rx_buffer
);
1953 } else if (IXGBE_CB(skb
)->dma
== rx_buffer
->dma
) {
1954 /* the page has been released from the ring */
1955 IXGBE_CB(skb
)->page_released
= true;
1957 /* we are not reusing the buffer so unmap it */
1958 dma_unmap_page(rx_ring
->dev
, rx_buffer
->dma
,
1959 ixgbe_rx_pg_size(rx_ring
),
1963 /* clear contents of buffer_info */
1964 rx_buffer
->skb
= NULL
;
1966 rx_buffer
->page
= NULL
;
1972 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1973 * @q_vector: structure containing interrupt and ring information
1974 * @rx_ring: rx descriptor ring to transact packets on
1975 * @budget: Total limit on number of packets to process
1977 * This function provides a "bounce buffer" approach to Rx interrupt
1978 * processing. The advantage to this is that on systems that have
1979 * expensive overhead for IOMMU access this provides a means of avoiding
1980 * it by maintaining the mapping of the page to the syste.
1982 * Returns amount of work completed
1984 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector
*q_vector
,
1985 struct ixgbe_ring
*rx_ring
,
1988 unsigned int total_rx_bytes
= 0, total_rx_packets
= 0;
1990 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1992 unsigned int mss
= 0;
1993 #endif /* IXGBE_FCOE */
1994 u16 cleaned_count
= ixgbe_desc_unused(rx_ring
);
1996 while (likely(total_rx_packets
< budget
)) {
1997 union ixgbe_adv_rx_desc
*rx_desc
;
1998 struct sk_buff
*skb
;
2000 /* return some buffers to hardware, one at a time is too slow */
2001 if (cleaned_count
>= IXGBE_RX_BUFFER_WRITE
) {
2002 ixgbe_alloc_rx_buffers(rx_ring
, cleaned_count
);
2006 rx_desc
= IXGBE_RX_DESC(rx_ring
, rx_ring
->next_to_clean
);
2008 if (!ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_DD
))
2012 * This memory barrier is needed to keep us from reading
2013 * any other fields out of the rx_desc until we know the
2014 * RXD_STAT_DD bit is set
2018 /* retrieve a buffer from the ring */
2019 skb
= ixgbe_fetch_rx_buffer(rx_ring
, rx_desc
);
2021 /* exit if we failed to retrieve a buffer */
2027 /* place incomplete frames back on ring for completion */
2028 if (ixgbe_is_non_eop(rx_ring
, rx_desc
, skb
))
2031 /* verify the packet layout is correct */
2032 if (ixgbe_cleanup_headers(rx_ring
, rx_desc
, skb
))
2035 /* probably a little skewed due to removing CRC */
2036 total_rx_bytes
+= skb
->len
;
2038 /* populate checksum, timestamp, VLAN, and protocol */
2039 ixgbe_process_skb_fields(rx_ring
, rx_desc
, skb
);
2042 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2043 if (ixgbe_rx_is_fcoe(rx_ring
, rx_desc
)) {
2044 ddp_bytes
= ixgbe_fcoe_ddp(adapter
, rx_desc
, skb
);
2045 /* include DDPed FCoE data */
2046 if (ddp_bytes
> 0) {
2048 mss
= rx_ring
->netdev
->mtu
-
2049 sizeof(struct fcoe_hdr
) -
2050 sizeof(struct fc_frame_header
) -
2051 sizeof(struct fcoe_crc_eof
);
2055 total_rx_bytes
+= ddp_bytes
;
2056 total_rx_packets
+= DIV_ROUND_UP(ddp_bytes
,
2060 dev_kfree_skb_any(skb
);
2065 #endif /* IXGBE_FCOE */
2066 skb_mark_napi_id(skb
, &q_vector
->napi
);
2067 ixgbe_rx_skb(q_vector
, skb
);
2069 /* update budget accounting */
2073 u64_stats_update_begin(&rx_ring
->syncp
);
2074 rx_ring
->stats
.packets
+= total_rx_packets
;
2075 rx_ring
->stats
.bytes
+= total_rx_bytes
;
2076 u64_stats_update_end(&rx_ring
->syncp
);
2077 q_vector
->rx
.total_packets
+= total_rx_packets
;
2078 q_vector
->rx
.total_bytes
+= total_rx_bytes
;
2080 return total_rx_packets
;
2083 #ifdef CONFIG_NET_RX_BUSY_POLL
2084 /* must be called with local_bh_disable()d */
2085 static int ixgbe_low_latency_recv(struct napi_struct
*napi
)
2087 struct ixgbe_q_vector
*q_vector
=
2088 container_of(napi
, struct ixgbe_q_vector
, napi
);
2089 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
2090 struct ixgbe_ring
*ring
;
2093 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
2094 return LL_FLUSH_FAILED
;
2096 if (!ixgbe_qv_lock_poll(q_vector
))
2097 return LL_FLUSH_BUSY
;
2099 ixgbe_for_each_ring(ring
, q_vector
->rx
) {
2100 found
= ixgbe_clean_rx_irq(q_vector
, ring
, 4);
2101 #ifdef BP_EXTENDED_STATS
2103 ring
->stats
.cleaned
+= found
;
2105 ring
->stats
.misses
++;
2111 ixgbe_qv_unlock_poll(q_vector
);
2115 #endif /* CONFIG_NET_RX_BUSY_POLL */
2118 * ixgbe_configure_msix - Configure MSI-X hardware
2119 * @adapter: board private structure
2121 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2124 static void ixgbe_configure_msix(struct ixgbe_adapter
*adapter
)
2126 struct ixgbe_q_vector
*q_vector
;
2130 /* Populate MSIX to EITR Select */
2131 if (adapter
->num_vfs
> 32) {
2132 u32 eitrsel
= (1 << (adapter
->num_vfs
- 32)) - 1;
2133 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EITRSEL
, eitrsel
);
2137 * Populate the IVAR table and set the ITR values to the
2138 * corresponding register.
2140 for (v_idx
= 0; v_idx
< adapter
->num_q_vectors
; v_idx
++) {
2141 struct ixgbe_ring
*ring
;
2142 q_vector
= adapter
->q_vector
[v_idx
];
2144 ixgbe_for_each_ring(ring
, q_vector
->rx
)
2145 ixgbe_set_ivar(adapter
, 0, ring
->reg_idx
, v_idx
);
2147 ixgbe_for_each_ring(ring
, q_vector
->tx
)
2148 ixgbe_set_ivar(adapter
, 1, ring
->reg_idx
, v_idx
);
2150 ixgbe_write_eitr(q_vector
);
2153 switch (adapter
->hw
.mac
.type
) {
2154 case ixgbe_mac_82598EB
:
2155 ixgbe_set_ivar(adapter
, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX
,
2158 case ixgbe_mac_82599EB
:
2159 case ixgbe_mac_X540
:
2160 ixgbe_set_ivar(adapter
, -1, 1, v_idx
);
2165 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EITR(v_idx
), 1950);
2167 /* set up to autoclear timer, and the vectors */
2168 mask
= IXGBE_EIMS_ENABLE_MASK
;
2169 mask
&= ~(IXGBE_EIMS_OTHER
|
2170 IXGBE_EIMS_MAILBOX
|
2173 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIAC
, mask
);
2176 enum latency_range
{
2180 latency_invalid
= 255
2184 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2185 * @q_vector: structure containing interrupt and ring information
2186 * @ring_container: structure containing ring performance data
2188 * Stores a new ITR value based on packets and byte
2189 * counts during the last interrupt. The advantage of per interrupt
2190 * computation is faster updates and more accurate ITR for the current
2191 * traffic pattern. Constants in this function were computed
2192 * based on theoretical maximum wire speed and thresholds were set based
2193 * on testing data as well as attempting to minimize response time
2194 * while increasing bulk throughput.
2195 * this functionality is controlled by the InterruptThrottleRate module
2196 * parameter (see ixgbe_param.c)
2198 static void ixgbe_update_itr(struct ixgbe_q_vector
*q_vector
,
2199 struct ixgbe_ring_container
*ring_container
)
2201 int bytes
= ring_container
->total_bytes
;
2202 int packets
= ring_container
->total_packets
;
2205 u8 itr_setting
= ring_container
->itr
;
2210 /* simple throttlerate management
2211 * 0-10MB/s lowest (100000 ints/s)
2212 * 10-20MB/s low (20000 ints/s)
2213 * 20-1249MB/s bulk (8000 ints/s)
2215 /* what was last interrupt timeslice? */
2216 timepassed_us
= q_vector
->itr
>> 2;
2217 if (timepassed_us
== 0)
2220 bytes_perint
= bytes
/ timepassed_us
; /* bytes/usec */
2222 switch (itr_setting
) {
2223 case lowest_latency
:
2224 if (bytes_perint
> 10)
2225 itr_setting
= low_latency
;
2228 if (bytes_perint
> 20)
2229 itr_setting
= bulk_latency
;
2230 else if (bytes_perint
<= 10)
2231 itr_setting
= lowest_latency
;
2234 if (bytes_perint
<= 20)
2235 itr_setting
= low_latency
;
2239 /* clear work counters since we have the values we need */
2240 ring_container
->total_bytes
= 0;
2241 ring_container
->total_packets
= 0;
2243 /* write updated itr to ring container */
2244 ring_container
->itr
= itr_setting
;
2248 * ixgbe_write_eitr - write EITR register in hardware specific way
2249 * @q_vector: structure containing interrupt and ring information
2251 * This function is made to be called by ethtool and by the driver
2252 * when it needs to update EITR registers at runtime. Hardware
2253 * specific quirks/differences are taken care of here.
2255 void ixgbe_write_eitr(struct ixgbe_q_vector
*q_vector
)
2257 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
2258 struct ixgbe_hw
*hw
= &adapter
->hw
;
2259 int v_idx
= q_vector
->v_idx
;
2260 u32 itr_reg
= q_vector
->itr
& IXGBE_MAX_EITR
;
2262 switch (adapter
->hw
.mac
.type
) {
2263 case ixgbe_mac_82598EB
:
2264 /* must write high and low 16 bits to reset counter */
2265 itr_reg
|= (itr_reg
<< 16);
2267 case ixgbe_mac_82599EB
:
2268 case ixgbe_mac_X540
:
2270 * set the WDIS bit to not clear the timer bits and cause an
2271 * immediate assertion of the interrupt
2273 itr_reg
|= IXGBE_EITR_CNT_WDIS
;
2278 IXGBE_WRITE_REG(hw
, IXGBE_EITR(v_idx
), itr_reg
);
2281 static void ixgbe_set_itr(struct ixgbe_q_vector
*q_vector
)
2283 u32 new_itr
= q_vector
->itr
;
2286 ixgbe_update_itr(q_vector
, &q_vector
->tx
);
2287 ixgbe_update_itr(q_vector
, &q_vector
->rx
);
2289 current_itr
= max(q_vector
->rx
.itr
, q_vector
->tx
.itr
);
2291 switch (current_itr
) {
2292 /* counts and packets in update_itr are dependent on these numbers */
2293 case lowest_latency
:
2294 new_itr
= IXGBE_100K_ITR
;
2297 new_itr
= IXGBE_20K_ITR
;
2300 new_itr
= IXGBE_8K_ITR
;
2306 if (new_itr
!= q_vector
->itr
) {
2307 /* do an exponential smoothing */
2308 new_itr
= (10 * new_itr
* q_vector
->itr
) /
2309 ((9 * new_itr
) + q_vector
->itr
);
2311 /* save the algorithm value here */
2312 q_vector
->itr
= new_itr
;
2314 ixgbe_write_eitr(q_vector
);
2319 * ixgbe_check_overtemp_subtask - check for over temperature
2320 * @adapter: pointer to adapter
2322 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter
*adapter
)
2324 struct ixgbe_hw
*hw
= &adapter
->hw
;
2325 u32 eicr
= adapter
->interrupt_event
;
2327 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
2330 if (!(adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
) &&
2331 !(adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_EVENT
))
2334 adapter
->flags2
&= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT
;
2336 switch (hw
->device_id
) {
2337 case IXGBE_DEV_ID_82599_T3_LOM
:
2339 * Since the warning interrupt is for both ports
2340 * we don't have to check if:
2341 * - This interrupt wasn't for our port.
2342 * - We may have missed the interrupt so always have to
2343 * check if we got a LSC
2345 if (!(eicr
& IXGBE_EICR_GPI_SDP0
) &&
2346 !(eicr
& IXGBE_EICR_LSC
))
2349 if (!(eicr
& IXGBE_EICR_LSC
) && hw
->mac
.ops
.check_link
) {
2351 bool link_up
= false;
2353 hw
->mac
.ops
.check_link(hw
, &speed
, &link_up
, false);
2359 /* Check if this is not due to overtemp */
2360 if (hw
->phy
.ops
.check_overtemp(hw
) != IXGBE_ERR_OVERTEMP
)
2365 if (!(eicr
& IXGBE_EICR_GPI_SDP0
))
2370 "Network adapter has been stopped because it has over heated. "
2371 "Restart the computer. If the problem persists, "
2372 "power off the system and replace the adapter\n");
2374 adapter
->interrupt_event
= 0;
2377 static void ixgbe_check_fan_failure(struct ixgbe_adapter
*adapter
, u32 eicr
)
2379 struct ixgbe_hw
*hw
= &adapter
->hw
;
2381 if ((adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) &&
2382 (eicr
& IXGBE_EICR_GPI_SDP1
)) {
2383 e_crit(probe
, "Fan has stopped, replace the adapter\n");
2384 /* write to clear the interrupt */
2385 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_GPI_SDP1
);
2389 static void ixgbe_check_overtemp_event(struct ixgbe_adapter
*adapter
, u32 eicr
)
2391 if (!(adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
))
2394 switch (adapter
->hw
.mac
.type
) {
2395 case ixgbe_mac_82599EB
:
2397 * Need to check link state so complete overtemp check
2400 if (((eicr
& IXGBE_EICR_GPI_SDP0
) || (eicr
& IXGBE_EICR_LSC
)) &&
2401 (!test_bit(__IXGBE_DOWN
, &adapter
->state
))) {
2402 adapter
->interrupt_event
= eicr
;
2403 adapter
->flags2
|= IXGBE_FLAG2_TEMP_SENSOR_EVENT
;
2404 ixgbe_service_event_schedule(adapter
);
2408 case ixgbe_mac_X540
:
2409 if (!(eicr
& IXGBE_EICR_TS
))
2417 "Network adapter has been stopped because it has over heated. "
2418 "Restart the computer. If the problem persists, "
2419 "power off the system and replace the adapter\n");
2422 static void ixgbe_check_sfp_event(struct ixgbe_adapter
*adapter
, u32 eicr
)
2424 struct ixgbe_hw
*hw
= &adapter
->hw
;
2426 if (eicr
& IXGBE_EICR_GPI_SDP2
) {
2427 /* Clear the interrupt */
2428 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_GPI_SDP2
);
2429 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
2430 adapter
->flags2
|= IXGBE_FLAG2_SFP_NEEDS_RESET
;
2431 ixgbe_service_event_schedule(adapter
);
2435 if (eicr
& IXGBE_EICR_GPI_SDP1
) {
2436 /* Clear the interrupt */
2437 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_GPI_SDP1
);
2438 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
2439 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_CONFIG
;
2440 ixgbe_service_event_schedule(adapter
);
2445 static void ixgbe_check_lsc(struct ixgbe_adapter
*adapter
)
2447 struct ixgbe_hw
*hw
= &adapter
->hw
;
2450 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_UPDATE
;
2451 adapter
->link_check_timeout
= jiffies
;
2452 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
2453 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, IXGBE_EIMC_LSC
);
2454 IXGBE_WRITE_FLUSH(hw
);
2455 ixgbe_service_event_schedule(adapter
);
2459 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter
*adapter
,
2463 struct ixgbe_hw
*hw
= &adapter
->hw
;
2465 switch (hw
->mac
.type
) {
2466 case ixgbe_mac_82598EB
:
2467 mask
= (IXGBE_EIMS_RTX_QUEUE
& qmask
);
2468 IXGBE_WRITE_REG(hw
, IXGBE_EIMS
, mask
);
2470 case ixgbe_mac_82599EB
:
2471 case ixgbe_mac_X540
:
2472 mask
= (qmask
& 0xFFFFFFFF);
2474 IXGBE_WRITE_REG(hw
, IXGBE_EIMS_EX(0), mask
);
2475 mask
= (qmask
>> 32);
2477 IXGBE_WRITE_REG(hw
, IXGBE_EIMS_EX(1), mask
);
2482 /* skip the flush */
2485 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter
*adapter
,
2489 struct ixgbe_hw
*hw
= &adapter
->hw
;
2491 switch (hw
->mac
.type
) {
2492 case ixgbe_mac_82598EB
:
2493 mask
= (IXGBE_EIMS_RTX_QUEUE
& qmask
);
2494 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, mask
);
2496 case ixgbe_mac_82599EB
:
2497 case ixgbe_mac_X540
:
2498 mask
= (qmask
& 0xFFFFFFFF);
2500 IXGBE_WRITE_REG(hw
, IXGBE_EIMC_EX(0), mask
);
2501 mask
= (qmask
>> 32);
2503 IXGBE_WRITE_REG(hw
, IXGBE_EIMC_EX(1), mask
);
2508 /* skip the flush */
2512 * ixgbe_irq_enable - Enable default interrupt generation settings
2513 * @adapter: board private structure
2515 static inline void ixgbe_irq_enable(struct ixgbe_adapter
*adapter
, bool queues
,
2518 u32 mask
= (IXGBE_EIMS_ENABLE_MASK
& ~IXGBE_EIMS_RTX_QUEUE
);
2520 /* don't reenable LSC while waiting for link */
2521 if (adapter
->flags
& IXGBE_FLAG_NEED_LINK_UPDATE
)
2522 mask
&= ~IXGBE_EIMS_LSC
;
2524 if (adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
)
2525 switch (adapter
->hw
.mac
.type
) {
2526 case ixgbe_mac_82599EB
:
2527 mask
|= IXGBE_EIMS_GPI_SDP0
;
2529 case ixgbe_mac_X540
:
2530 mask
|= IXGBE_EIMS_TS
;
2535 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
)
2536 mask
|= IXGBE_EIMS_GPI_SDP1
;
2537 switch (adapter
->hw
.mac
.type
) {
2538 case ixgbe_mac_82599EB
:
2539 mask
|= IXGBE_EIMS_GPI_SDP1
;
2540 mask
|= IXGBE_EIMS_GPI_SDP2
;
2541 case ixgbe_mac_X540
:
2542 mask
|= IXGBE_EIMS_ECC
;
2543 mask
|= IXGBE_EIMS_MAILBOX
;
2549 if (adapter
->hw
.mac
.type
== ixgbe_mac_X540
)
2550 mask
|= IXGBE_EIMS_TIMESYNC
;
2552 if ((adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) &&
2553 !(adapter
->flags2
& IXGBE_FLAG2_FDIR_REQUIRES_REINIT
))
2554 mask
|= IXGBE_EIMS_FLOW_DIR
;
2556 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMS
, mask
);
2558 ixgbe_irq_enable_queues(adapter
, ~0);
2560 IXGBE_WRITE_FLUSH(&adapter
->hw
);
2563 static irqreturn_t
ixgbe_msix_other(int irq
, void *data
)
2565 struct ixgbe_adapter
*adapter
= data
;
2566 struct ixgbe_hw
*hw
= &adapter
->hw
;
2570 * Workaround for Silicon errata. Use clear-by-write instead
2571 * of clear-by-read. Reading with EICS will return the
2572 * interrupt causes without clearing, which later be done
2573 * with the write to EICR.
2575 eicr
= IXGBE_READ_REG(hw
, IXGBE_EICS
);
2577 /* The lower 16bits of the EICR register are for the queue interrupts
2578 * which should be masked here in order to not accidently clear them if
2579 * the bits are high when ixgbe_msix_other is called. There is a race
2580 * condition otherwise which results in possible performance loss
2581 * especially if the ixgbe_msix_other interrupt is triggering
2582 * consistently (as it would when PPS is turned on for the X540 device)
2586 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, eicr
);
2588 if (eicr
& IXGBE_EICR_LSC
)
2589 ixgbe_check_lsc(adapter
);
2591 if (eicr
& IXGBE_EICR_MAILBOX
)
2592 ixgbe_msg_task(adapter
);
2594 switch (hw
->mac
.type
) {
2595 case ixgbe_mac_82599EB
:
2596 case ixgbe_mac_X540
:
2597 if (eicr
& IXGBE_EICR_ECC
) {
2598 e_info(link
, "Received ECC Err, initiating reset\n");
2599 adapter
->flags2
|= IXGBE_FLAG2_RESET_REQUESTED
;
2600 ixgbe_service_event_schedule(adapter
);
2601 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_ECC
);
2603 /* Handle Flow Director Full threshold interrupt */
2604 if (eicr
& IXGBE_EICR_FLOW_DIR
) {
2605 int reinit_count
= 0;
2607 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
2608 struct ixgbe_ring
*ring
= adapter
->tx_ring
[i
];
2609 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE
,
2614 /* no more flow director interrupts until after init */
2615 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, IXGBE_EIMC_FLOW_DIR
);
2616 adapter
->flags2
|= IXGBE_FLAG2_FDIR_REQUIRES_REINIT
;
2617 ixgbe_service_event_schedule(adapter
);
2620 ixgbe_check_sfp_event(adapter
, eicr
);
2621 ixgbe_check_overtemp_event(adapter
, eicr
);
2627 ixgbe_check_fan_failure(adapter
, eicr
);
2629 if (unlikely(eicr
& IXGBE_EICR_TIMESYNC
))
2630 ixgbe_ptp_check_pps_event(adapter
, eicr
);
2632 /* re-enable the original interrupt state, no lsc, no queues */
2633 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2634 ixgbe_irq_enable(adapter
, false, false);
2639 static irqreturn_t
ixgbe_msix_clean_rings(int irq
, void *data
)
2641 struct ixgbe_q_vector
*q_vector
= data
;
2643 /* EIAM disabled interrupts (on this vector) for us */
2645 if (q_vector
->rx
.ring
|| q_vector
->tx
.ring
)
2646 napi_schedule(&q_vector
->napi
);
2652 * ixgbe_poll - NAPI Rx polling callback
2653 * @napi: structure for representing this polling device
2654 * @budget: how many packets driver is allowed to clean
2656 * This function is used for legacy and MSI, NAPI mode
2658 int ixgbe_poll(struct napi_struct
*napi
, int budget
)
2660 struct ixgbe_q_vector
*q_vector
=
2661 container_of(napi
, struct ixgbe_q_vector
, napi
);
2662 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
2663 struct ixgbe_ring
*ring
;
2664 int per_ring_budget
;
2665 bool clean_complete
= true;
2667 #ifdef CONFIG_IXGBE_DCA
2668 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
2669 ixgbe_update_dca(q_vector
);
2672 ixgbe_for_each_ring(ring
, q_vector
->tx
)
2673 clean_complete
&= !!ixgbe_clean_tx_irq(q_vector
, ring
);
2675 if (!ixgbe_qv_lock_napi(q_vector
))
2678 /* attempt to distribute budget to each queue fairly, but don't allow
2679 * the budget to go below 1 because we'll exit polling */
2680 if (q_vector
->rx
.count
> 1)
2681 per_ring_budget
= max(budget
/q_vector
->rx
.count
, 1);
2683 per_ring_budget
= budget
;
2685 ixgbe_for_each_ring(ring
, q_vector
->rx
)
2686 clean_complete
&= (ixgbe_clean_rx_irq(q_vector
, ring
,
2687 per_ring_budget
) < per_ring_budget
);
2689 ixgbe_qv_unlock_napi(q_vector
);
2690 /* If all work not completed, return budget and keep polling */
2691 if (!clean_complete
)
2694 /* all work done, exit the polling mode */
2695 napi_complete(napi
);
2696 if (adapter
->rx_itr_setting
& 1)
2697 ixgbe_set_itr(q_vector
);
2698 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2699 ixgbe_irq_enable_queues(adapter
, ((u64
)1 << q_vector
->v_idx
));
2705 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2706 * @adapter: board private structure
2708 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2709 * interrupts from the kernel.
2711 static int ixgbe_request_msix_irqs(struct ixgbe_adapter
*adapter
)
2713 struct net_device
*netdev
= adapter
->netdev
;
2717 for (vector
= 0; vector
< adapter
->num_q_vectors
; vector
++) {
2718 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[vector
];
2719 struct msix_entry
*entry
= &adapter
->msix_entries
[vector
];
2721 if (q_vector
->tx
.ring
&& q_vector
->rx
.ring
) {
2722 snprintf(q_vector
->name
, sizeof(q_vector
->name
) - 1,
2723 "%s-%s-%d", netdev
->name
, "TxRx", ri
++);
2725 } else if (q_vector
->rx
.ring
) {
2726 snprintf(q_vector
->name
, sizeof(q_vector
->name
) - 1,
2727 "%s-%s-%d", netdev
->name
, "rx", ri
++);
2728 } else if (q_vector
->tx
.ring
) {
2729 snprintf(q_vector
->name
, sizeof(q_vector
->name
) - 1,
2730 "%s-%s-%d", netdev
->name
, "tx", ti
++);
2732 /* skip this unused q_vector */
2735 err
= request_irq(entry
->vector
, &ixgbe_msix_clean_rings
, 0,
2736 q_vector
->name
, q_vector
);
2738 e_err(probe
, "request_irq failed for MSIX interrupt "
2739 "Error: %d\n", err
);
2740 goto free_queue_irqs
;
2742 /* If Flow Director is enabled, set interrupt affinity */
2743 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) {
2744 /* assign the mask for this irq */
2745 irq_set_affinity_hint(entry
->vector
,
2746 &q_vector
->affinity_mask
);
2750 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
2751 ixgbe_msix_other
, 0, netdev
->name
, adapter
);
2753 e_err(probe
, "request_irq for msix_other failed: %d\n", err
);
2754 goto free_queue_irqs
;
2762 irq_set_affinity_hint(adapter
->msix_entries
[vector
].vector
,
2764 free_irq(adapter
->msix_entries
[vector
].vector
,
2765 adapter
->q_vector
[vector
]);
2767 adapter
->flags
&= ~IXGBE_FLAG_MSIX_ENABLED
;
2768 pci_disable_msix(adapter
->pdev
);
2769 kfree(adapter
->msix_entries
);
2770 adapter
->msix_entries
= NULL
;
2775 * ixgbe_intr - legacy mode Interrupt Handler
2776 * @irq: interrupt number
2777 * @data: pointer to a network interface device structure
2779 static irqreturn_t
ixgbe_intr(int irq
, void *data
)
2781 struct ixgbe_adapter
*adapter
= data
;
2782 struct ixgbe_hw
*hw
= &adapter
->hw
;
2783 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[0];
2787 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2788 * before the read of EICR.
2790 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, IXGBE_IRQ_CLEAR_MASK
);
2792 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2793 * therefore no explicit interrupt disable is necessary */
2794 eicr
= IXGBE_READ_REG(hw
, IXGBE_EICR
);
2797 * shared interrupt alert!
2798 * make sure interrupts are enabled because the read will
2799 * have disabled interrupts due to EIAM
2800 * finish the workaround of silicon errata on 82598. Unmask
2801 * the interrupt that we masked before the EICR read.
2803 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2804 ixgbe_irq_enable(adapter
, true, true);
2805 return IRQ_NONE
; /* Not our interrupt */
2808 if (eicr
& IXGBE_EICR_LSC
)
2809 ixgbe_check_lsc(adapter
);
2811 switch (hw
->mac
.type
) {
2812 case ixgbe_mac_82599EB
:
2813 ixgbe_check_sfp_event(adapter
, eicr
);
2815 case ixgbe_mac_X540
:
2816 if (eicr
& IXGBE_EICR_ECC
) {
2817 e_info(link
, "Received ECC Err, initiating reset\n");
2818 adapter
->flags2
|= IXGBE_FLAG2_RESET_REQUESTED
;
2819 ixgbe_service_event_schedule(adapter
);
2820 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_ECC
);
2822 ixgbe_check_overtemp_event(adapter
, eicr
);
2828 ixgbe_check_fan_failure(adapter
, eicr
);
2829 if (unlikely(eicr
& IXGBE_EICR_TIMESYNC
))
2830 ixgbe_ptp_check_pps_event(adapter
, eicr
);
2832 /* would disable interrupts here but EIAM disabled it */
2833 napi_schedule(&q_vector
->napi
);
2836 * re-enable link(maybe) and non-queue interrupts, no flush.
2837 * ixgbe_poll will re-enable the queue interrupts
2839 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2840 ixgbe_irq_enable(adapter
, false, false);
2846 * ixgbe_request_irq - initialize interrupts
2847 * @adapter: board private structure
2849 * Attempts to configure interrupts using the best available
2850 * capabilities of the hardware and kernel.
2852 static int ixgbe_request_irq(struct ixgbe_adapter
*adapter
)
2854 struct net_device
*netdev
= adapter
->netdev
;
2857 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)
2858 err
= ixgbe_request_msix_irqs(adapter
);
2859 else if (adapter
->flags
& IXGBE_FLAG_MSI_ENABLED
)
2860 err
= request_irq(adapter
->pdev
->irq
, ixgbe_intr
, 0,
2861 netdev
->name
, adapter
);
2863 err
= request_irq(adapter
->pdev
->irq
, ixgbe_intr
, IRQF_SHARED
,
2864 netdev
->name
, adapter
);
2867 e_err(probe
, "request_irq failed, Error %d\n", err
);
2872 static void ixgbe_free_irq(struct ixgbe_adapter
*adapter
)
2876 if (!(adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)) {
2877 free_irq(adapter
->pdev
->irq
, adapter
);
2881 for (vector
= 0; vector
< adapter
->num_q_vectors
; vector
++) {
2882 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[vector
];
2883 struct msix_entry
*entry
= &adapter
->msix_entries
[vector
];
2885 /* free only the irqs that were actually requested */
2886 if (!q_vector
->rx
.ring
&& !q_vector
->tx
.ring
)
2889 /* clear the affinity_mask in the IRQ descriptor */
2890 irq_set_affinity_hint(entry
->vector
, NULL
);
2892 free_irq(entry
->vector
, q_vector
);
2895 free_irq(adapter
->msix_entries
[vector
++].vector
, adapter
);
2899 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2900 * @adapter: board private structure
2902 static inline void ixgbe_irq_disable(struct ixgbe_adapter
*adapter
)
2904 switch (adapter
->hw
.mac
.type
) {
2905 case ixgbe_mac_82598EB
:
2906 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC
, ~0);
2908 case ixgbe_mac_82599EB
:
2909 case ixgbe_mac_X540
:
2910 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC
, 0xFFFF0000);
2911 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC_EX(0), ~0);
2912 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC_EX(1), ~0);
2917 IXGBE_WRITE_FLUSH(&adapter
->hw
);
2918 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
2921 for (vector
= 0; vector
< adapter
->num_q_vectors
; vector
++)
2922 synchronize_irq(adapter
->msix_entries
[vector
].vector
);
2924 synchronize_irq(adapter
->msix_entries
[vector
++].vector
);
2926 synchronize_irq(adapter
->pdev
->irq
);
2931 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2934 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter
*adapter
)
2936 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[0];
2938 ixgbe_write_eitr(q_vector
);
2940 ixgbe_set_ivar(adapter
, 0, 0, 0);
2941 ixgbe_set_ivar(adapter
, 1, 0, 0);
2943 e_info(hw
, "Legacy interrupt IVAR setup done\n");
2947 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2948 * @adapter: board private structure
2949 * @ring: structure containing ring specific data
2951 * Configure the Tx descriptor ring after a reset.
2953 void ixgbe_configure_tx_ring(struct ixgbe_adapter
*adapter
,
2954 struct ixgbe_ring
*ring
)
2956 struct ixgbe_hw
*hw
= &adapter
->hw
;
2957 u64 tdba
= ring
->dma
;
2959 u32 txdctl
= IXGBE_TXDCTL_ENABLE
;
2960 u8 reg_idx
= ring
->reg_idx
;
2962 /* disable queue to avoid issues while updating state */
2963 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(reg_idx
), 0);
2964 IXGBE_WRITE_FLUSH(hw
);
2966 IXGBE_WRITE_REG(hw
, IXGBE_TDBAL(reg_idx
),
2967 (tdba
& DMA_BIT_MASK(32)));
2968 IXGBE_WRITE_REG(hw
, IXGBE_TDBAH(reg_idx
), (tdba
>> 32));
2969 IXGBE_WRITE_REG(hw
, IXGBE_TDLEN(reg_idx
),
2970 ring
->count
* sizeof(union ixgbe_adv_tx_desc
));
2971 IXGBE_WRITE_REG(hw
, IXGBE_TDH(reg_idx
), 0);
2972 IXGBE_WRITE_REG(hw
, IXGBE_TDT(reg_idx
), 0);
2973 ring
->tail
= adapter
->io_addr
+ IXGBE_TDT(reg_idx
);
2976 * set WTHRESH to encourage burst writeback, it should not be set
2977 * higher than 1 when:
2978 * - ITR is 0 as it could cause false TX hangs
2979 * - ITR is set to > 100k int/sec and BQL is enabled
2981 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2982 * to or less than the number of on chip descriptors, which is
2985 if (!ring
->q_vector
|| (ring
->q_vector
->itr
< IXGBE_100K_ITR
))
2986 txdctl
|= (1 << 16); /* WTHRESH = 1 */
2988 txdctl
|= (8 << 16); /* WTHRESH = 8 */
2991 * Setting PTHRESH to 32 both improves performance
2992 * and avoids a TX hang with DFP enabled
2994 txdctl
|= (1 << 8) | /* HTHRESH = 1 */
2995 32; /* PTHRESH = 32 */
2997 /* reinitialize flowdirector state */
2998 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) {
2999 ring
->atr_sample_rate
= adapter
->atr_sample_rate
;
3000 ring
->atr_count
= 0;
3001 set_bit(__IXGBE_TX_FDIR_INIT_DONE
, &ring
->state
);
3003 ring
->atr_sample_rate
= 0;
3006 /* initialize XPS */
3007 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE
, &ring
->state
)) {
3008 struct ixgbe_q_vector
*q_vector
= ring
->q_vector
;
3011 netif_set_xps_queue(ring
->netdev
,
3012 &q_vector
->affinity_mask
,
3016 clear_bit(__IXGBE_HANG_CHECK_ARMED
, &ring
->state
);
3019 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(reg_idx
), txdctl
);
3021 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3022 if (hw
->mac
.type
== ixgbe_mac_82598EB
&&
3023 !(IXGBE_READ_REG(hw
, IXGBE_LINKS
) & IXGBE_LINKS_UP
))
3026 /* poll to verify queue is enabled */
3028 usleep_range(1000, 2000);
3029 txdctl
= IXGBE_READ_REG(hw
, IXGBE_TXDCTL(reg_idx
));
3030 } while (--wait_loop
&& !(txdctl
& IXGBE_TXDCTL_ENABLE
));
3032 e_err(drv
, "Could not enable Tx Queue %d\n", reg_idx
);
3035 static void ixgbe_setup_mtqc(struct ixgbe_adapter
*adapter
)
3037 struct ixgbe_hw
*hw
= &adapter
->hw
;
3039 u8 tcs
= netdev_get_num_tc(adapter
->netdev
);
3041 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
3044 /* disable the arbiter while setting MTQC */
3045 rttdcs
= IXGBE_READ_REG(hw
, IXGBE_RTTDCS
);
3046 rttdcs
|= IXGBE_RTTDCS_ARBDIS
;
3047 IXGBE_WRITE_REG(hw
, IXGBE_RTTDCS
, rttdcs
);
3049 /* set transmit pool layout */
3050 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
3051 mtqc
= IXGBE_MTQC_VT_ENA
;
3053 mtqc
|= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_8TC_8TQ
;
3055 mtqc
|= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_4TC_4TQ
;
3056 else if (adapter
->ring_feature
[RING_F_RSS
].indices
== 4)
3057 mtqc
|= IXGBE_MTQC_32VF
;
3059 mtqc
|= IXGBE_MTQC_64VF
;
3062 mtqc
= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_8TC_8TQ
;
3064 mtqc
= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_4TC_4TQ
;
3066 mtqc
= IXGBE_MTQC_64Q_1PB
;
3069 IXGBE_WRITE_REG(hw
, IXGBE_MTQC
, mtqc
);
3071 /* Enable Security TX Buffer IFG for multiple pb */
3073 u32 sectx
= IXGBE_READ_REG(hw
, IXGBE_SECTXMINIFG
);
3074 sectx
|= IXGBE_SECTX_DCB
;
3075 IXGBE_WRITE_REG(hw
, IXGBE_SECTXMINIFG
, sectx
);
3078 /* re-enable the arbiter */
3079 rttdcs
&= ~IXGBE_RTTDCS_ARBDIS
;
3080 IXGBE_WRITE_REG(hw
, IXGBE_RTTDCS
, rttdcs
);
3084 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3085 * @adapter: board private structure
3087 * Configure the Tx unit of the MAC after a reset.
3089 static void ixgbe_configure_tx(struct ixgbe_adapter
*adapter
)
3091 struct ixgbe_hw
*hw
= &adapter
->hw
;
3095 ixgbe_setup_mtqc(adapter
);
3097 if (hw
->mac
.type
!= ixgbe_mac_82598EB
) {
3098 /* DMATXCTL.EN must be before Tx queues are enabled */
3099 dmatxctl
= IXGBE_READ_REG(hw
, IXGBE_DMATXCTL
);
3100 dmatxctl
|= IXGBE_DMATXCTL_TE
;
3101 IXGBE_WRITE_REG(hw
, IXGBE_DMATXCTL
, dmatxctl
);
3104 /* Setup the HW Tx Head and Tail descriptor pointers */
3105 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3106 ixgbe_configure_tx_ring(adapter
, adapter
->tx_ring
[i
]);
3109 static void ixgbe_enable_rx_drop(struct ixgbe_adapter
*adapter
,
3110 struct ixgbe_ring
*ring
)
3112 struct ixgbe_hw
*hw
= &adapter
->hw
;
3113 u8 reg_idx
= ring
->reg_idx
;
3114 u32 srrctl
= IXGBE_READ_REG(hw
, IXGBE_SRRCTL(reg_idx
));
3116 srrctl
|= IXGBE_SRRCTL_DROP_EN
;
3118 IXGBE_WRITE_REG(hw
, IXGBE_SRRCTL(reg_idx
), srrctl
);
3121 static void ixgbe_disable_rx_drop(struct ixgbe_adapter
*adapter
,
3122 struct ixgbe_ring
*ring
)
3124 struct ixgbe_hw
*hw
= &adapter
->hw
;
3125 u8 reg_idx
= ring
->reg_idx
;
3126 u32 srrctl
= IXGBE_READ_REG(hw
, IXGBE_SRRCTL(reg_idx
));
3128 srrctl
&= ~IXGBE_SRRCTL_DROP_EN
;
3130 IXGBE_WRITE_REG(hw
, IXGBE_SRRCTL(reg_idx
), srrctl
);
3133 #ifdef CONFIG_IXGBE_DCB
3134 void ixgbe_set_rx_drop_en(struct ixgbe_adapter
*adapter
)
3136 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter
*adapter
)
3140 bool pfc_en
= adapter
->dcb_cfg
.pfc_mode_enable
;
3142 if (adapter
->ixgbe_ieee_pfc
)
3143 pfc_en
|= !!(adapter
->ixgbe_ieee_pfc
->pfc_en
);
3146 * We should set the drop enable bit if:
3149 * Number of Rx queues > 1 and flow control is disabled
3151 * This allows us to avoid head of line blocking for security
3152 * and performance reasons.
3154 if (adapter
->num_vfs
|| (adapter
->num_rx_queues
> 1 &&
3155 !(adapter
->hw
.fc
.current_mode
& ixgbe_fc_tx_pause
) && !pfc_en
)) {
3156 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3157 ixgbe_enable_rx_drop(adapter
, adapter
->rx_ring
[i
]);
3159 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3160 ixgbe_disable_rx_drop(adapter
, adapter
->rx_ring
[i
]);
3164 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3166 static void ixgbe_configure_srrctl(struct ixgbe_adapter
*adapter
,
3167 struct ixgbe_ring
*rx_ring
)
3169 struct ixgbe_hw
*hw
= &adapter
->hw
;
3171 u8 reg_idx
= rx_ring
->reg_idx
;
3173 if (hw
->mac
.type
== ixgbe_mac_82598EB
) {
3174 u16 mask
= adapter
->ring_feature
[RING_F_RSS
].mask
;
3177 * if VMDq is not active we must program one srrctl register
3178 * per RSS queue since we have enabled RDRXCTL.MVMEN
3183 /* configure header buffer length, needed for RSC */
3184 srrctl
= IXGBE_RX_HDR_SIZE
<< IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT
;
3186 /* configure the packet buffer length */
3187 srrctl
|= ixgbe_rx_bufsz(rx_ring
) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT
;
3189 /* configure descriptor type */
3190 srrctl
|= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF
;
3192 IXGBE_WRITE_REG(hw
, IXGBE_SRRCTL(reg_idx
), srrctl
);
3195 static void ixgbe_setup_mrqc(struct ixgbe_adapter
*adapter
)
3197 struct ixgbe_hw
*hw
= &adapter
->hw
;
3198 static const u32 seed
[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
3199 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
3200 0x6A3E67EA, 0x14364D17, 0x3BED200D};
3201 u32 mrqc
= 0, reta
= 0;
3204 u16 rss_i
= adapter
->ring_feature
[RING_F_RSS
].indices
;
3207 * Program table for at least 2 queues w/ SR-IOV so that VFs can
3208 * make full use of any rings they may have. We will use the
3209 * PSRTYPE register to control how many rings we use within the PF.
3211 if ((adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) && (rss_i
< 2))
3214 /* Fill out hash function seeds */
3215 for (i
= 0; i
< 10; i
++)
3216 IXGBE_WRITE_REG(hw
, IXGBE_RSSRK(i
), seed
[i
]);
3218 /* Fill out redirection table */
3219 for (i
= 0, j
= 0; i
< 128; i
++, j
++) {
3222 /* reta = 4-byte sliding window of
3223 * 0x00..(indices-1)(indices-1)00..etc. */
3224 reta
= (reta
<< 8) | (j
* 0x11);
3226 IXGBE_WRITE_REG(hw
, IXGBE_RETA(i
>> 2), reta
);
3229 /* Disable indicating checksum in descriptor, enables RSS hash */
3230 rxcsum
= IXGBE_READ_REG(hw
, IXGBE_RXCSUM
);
3231 rxcsum
|= IXGBE_RXCSUM_PCSD
;
3232 IXGBE_WRITE_REG(hw
, IXGBE_RXCSUM
, rxcsum
);
3234 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
3235 if (adapter
->ring_feature
[RING_F_RSS
].mask
)
3236 mrqc
= IXGBE_MRQC_RSSEN
;
3238 u8 tcs
= netdev_get_num_tc(adapter
->netdev
);
3240 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
3242 mrqc
= IXGBE_MRQC_VMDQRT8TCEN
; /* 8 TCs */
3244 mrqc
= IXGBE_MRQC_VMDQRT4TCEN
; /* 4 TCs */
3245 else if (adapter
->ring_feature
[RING_F_RSS
].indices
== 4)
3246 mrqc
= IXGBE_MRQC_VMDQRSS32EN
;
3248 mrqc
= IXGBE_MRQC_VMDQRSS64EN
;
3251 mrqc
= IXGBE_MRQC_RTRSS8TCEN
;
3253 mrqc
= IXGBE_MRQC_RTRSS4TCEN
;
3255 mrqc
= IXGBE_MRQC_RSSEN
;
3259 /* Perform hash on these packet types */
3260 mrqc
|= IXGBE_MRQC_RSS_FIELD_IPV4
|
3261 IXGBE_MRQC_RSS_FIELD_IPV4_TCP
|
3262 IXGBE_MRQC_RSS_FIELD_IPV6
|
3263 IXGBE_MRQC_RSS_FIELD_IPV6_TCP
;
3265 if (adapter
->flags2
& IXGBE_FLAG2_RSS_FIELD_IPV4_UDP
)
3266 mrqc
|= IXGBE_MRQC_RSS_FIELD_IPV4_UDP
;
3267 if (adapter
->flags2
& IXGBE_FLAG2_RSS_FIELD_IPV6_UDP
)
3268 mrqc
|= IXGBE_MRQC_RSS_FIELD_IPV6_UDP
;
3270 IXGBE_WRITE_REG(hw
, IXGBE_MRQC
, mrqc
);
3274 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3275 * @adapter: address of board private structure
3276 * @index: index of ring to set
3278 static void ixgbe_configure_rscctl(struct ixgbe_adapter
*adapter
,
3279 struct ixgbe_ring
*ring
)
3281 struct ixgbe_hw
*hw
= &adapter
->hw
;
3283 u8 reg_idx
= ring
->reg_idx
;
3285 if (!ring_is_rsc_enabled(ring
))
3288 rscctrl
= IXGBE_READ_REG(hw
, IXGBE_RSCCTL(reg_idx
));
3289 rscctrl
|= IXGBE_RSCCTL_RSCEN
;
3291 * we must limit the number of descriptors so that the
3292 * total size of max desc * buf_len is not greater
3295 rscctrl
|= IXGBE_RSCCTL_MAXDESC_16
;
3296 IXGBE_WRITE_REG(hw
, IXGBE_RSCCTL(reg_idx
), rscctrl
);
3299 #define IXGBE_MAX_RX_DESC_POLL 10
3300 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter
*adapter
,
3301 struct ixgbe_ring
*ring
)
3303 struct ixgbe_hw
*hw
= &adapter
->hw
;
3304 int wait_loop
= IXGBE_MAX_RX_DESC_POLL
;
3306 u8 reg_idx
= ring
->reg_idx
;
3308 if (ixgbe_removed(hw
->hw_addr
))
3310 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3311 if (hw
->mac
.type
== ixgbe_mac_82598EB
&&
3312 !(IXGBE_READ_REG(hw
, IXGBE_LINKS
) & IXGBE_LINKS_UP
))
3316 usleep_range(1000, 2000);
3317 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
3318 } while (--wait_loop
&& !(rxdctl
& IXGBE_RXDCTL_ENABLE
));
3321 e_err(drv
, "RXDCTL.ENABLE on Rx queue %d not set within "
3322 "the polling period\n", reg_idx
);
3326 void ixgbe_disable_rx_queue(struct ixgbe_adapter
*adapter
,
3327 struct ixgbe_ring
*ring
)
3329 struct ixgbe_hw
*hw
= &adapter
->hw
;
3330 int wait_loop
= IXGBE_MAX_RX_DESC_POLL
;
3332 u8 reg_idx
= ring
->reg_idx
;
3334 if (ixgbe_removed(hw
->hw_addr
))
3336 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
3337 rxdctl
&= ~IXGBE_RXDCTL_ENABLE
;
3339 /* write value back with RXDCTL.ENABLE bit cleared */
3340 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(reg_idx
), rxdctl
);
3342 if (hw
->mac
.type
== ixgbe_mac_82598EB
&&
3343 !(IXGBE_READ_REG(hw
, IXGBE_LINKS
) & IXGBE_LINKS_UP
))
3346 /* the hardware may take up to 100us to really disable the rx queue */
3349 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
3350 } while (--wait_loop
&& (rxdctl
& IXGBE_RXDCTL_ENABLE
));
3353 e_err(drv
, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3354 "the polling period\n", reg_idx
);
3358 void ixgbe_configure_rx_ring(struct ixgbe_adapter
*adapter
,
3359 struct ixgbe_ring
*ring
)
3361 struct ixgbe_hw
*hw
= &adapter
->hw
;
3362 u64 rdba
= ring
->dma
;
3364 u8 reg_idx
= ring
->reg_idx
;
3366 /* disable queue to avoid issues while updating state */
3367 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
3368 ixgbe_disable_rx_queue(adapter
, ring
);
3370 IXGBE_WRITE_REG(hw
, IXGBE_RDBAL(reg_idx
), (rdba
& DMA_BIT_MASK(32)));
3371 IXGBE_WRITE_REG(hw
, IXGBE_RDBAH(reg_idx
), (rdba
>> 32));
3372 IXGBE_WRITE_REG(hw
, IXGBE_RDLEN(reg_idx
),
3373 ring
->count
* sizeof(union ixgbe_adv_rx_desc
));
3374 IXGBE_WRITE_REG(hw
, IXGBE_RDH(reg_idx
), 0);
3375 IXGBE_WRITE_REG(hw
, IXGBE_RDT(reg_idx
), 0);
3376 ring
->tail
= adapter
->io_addr
+ IXGBE_RDT(reg_idx
);
3378 ixgbe_configure_srrctl(adapter
, ring
);
3379 ixgbe_configure_rscctl(adapter
, ring
);
3381 if (hw
->mac
.type
== ixgbe_mac_82598EB
) {
3383 * enable cache line friendly hardware writes:
3384 * PTHRESH=32 descriptors (half the internal cache),
3385 * this also removes ugly rx_no_buffer_count increment
3386 * HTHRESH=4 descriptors (to minimize latency on fetch)
3387 * WTHRESH=8 burst writeback up to two cache lines
3389 rxdctl
&= ~0x3FFFFF;
3393 /* enable receive descriptor ring */
3394 rxdctl
|= IXGBE_RXDCTL_ENABLE
;
3395 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(reg_idx
), rxdctl
);
3397 ixgbe_rx_desc_queue_enable(adapter
, ring
);
3398 ixgbe_alloc_rx_buffers(ring
, ixgbe_desc_unused(ring
));
3401 static void ixgbe_setup_psrtype(struct ixgbe_adapter
*adapter
)
3403 struct ixgbe_hw
*hw
= &adapter
->hw
;
3404 int rss_i
= adapter
->ring_feature
[RING_F_RSS
].indices
;
3407 /* PSRTYPE must be initialized in non 82598 adapters */
3408 u32 psrtype
= IXGBE_PSRTYPE_TCPHDR
|
3409 IXGBE_PSRTYPE_UDPHDR
|
3410 IXGBE_PSRTYPE_IPV4HDR
|
3411 IXGBE_PSRTYPE_L2HDR
|
3412 IXGBE_PSRTYPE_IPV6HDR
;
3414 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
3422 for_each_set_bit(pool
, &adapter
->fwd_bitmask
, 32)
3423 IXGBE_WRITE_REG(hw
, IXGBE_PSRTYPE(VMDQ_P(pool
)), psrtype
);
3426 static void ixgbe_configure_virtualization(struct ixgbe_adapter
*adapter
)
3428 struct ixgbe_hw
*hw
= &adapter
->hw
;
3429 u32 reg_offset
, vf_shift
;
3430 u32 gcr_ext
, vmdctl
;
3433 if (!(adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
))
3436 vmdctl
= IXGBE_READ_REG(hw
, IXGBE_VT_CTL
);
3437 vmdctl
|= IXGBE_VMD_CTL_VMDQ_EN
;
3438 vmdctl
&= ~IXGBE_VT_CTL_POOL_MASK
;
3439 vmdctl
|= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT
;
3440 vmdctl
|= IXGBE_VT_CTL_REPLEN
;
3441 IXGBE_WRITE_REG(hw
, IXGBE_VT_CTL
, vmdctl
);
3443 vf_shift
= VMDQ_P(0) % 32;
3444 reg_offset
= (VMDQ_P(0) >= 32) ? 1 : 0;
3446 /* Enable only the PF's pool for Tx/Rx */
3447 IXGBE_WRITE_REG(hw
, IXGBE_VFRE(reg_offset
), (~0) << vf_shift
);
3448 IXGBE_WRITE_REG(hw
, IXGBE_VFRE(reg_offset
^ 1), reg_offset
- 1);
3449 IXGBE_WRITE_REG(hw
, IXGBE_VFTE(reg_offset
), (~0) << vf_shift
);
3450 IXGBE_WRITE_REG(hw
, IXGBE_VFTE(reg_offset
^ 1), reg_offset
- 1);
3451 if (adapter
->flags2
& IXGBE_FLAG2_BRIDGE_MODE_VEB
)
3452 IXGBE_WRITE_REG(hw
, IXGBE_PFDTXGSWC
, IXGBE_PFDTXGSWC_VT_LBEN
);
3454 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3455 hw
->mac
.ops
.set_vmdq(hw
, 0, VMDQ_P(0));
3458 * Set up VF register offsets for selected VT Mode,
3459 * i.e. 32 or 64 VFs for SR-IOV
3461 switch (adapter
->ring_feature
[RING_F_VMDQ
].mask
) {
3462 case IXGBE_82599_VMDQ_8Q_MASK
:
3463 gcr_ext
= IXGBE_GCR_EXT_VT_MODE_16
;
3465 case IXGBE_82599_VMDQ_4Q_MASK
:
3466 gcr_ext
= IXGBE_GCR_EXT_VT_MODE_32
;
3469 gcr_ext
= IXGBE_GCR_EXT_VT_MODE_64
;
3473 IXGBE_WRITE_REG(hw
, IXGBE_GCR_EXT
, gcr_ext
);
3476 /* Enable MAC Anti-Spoofing */
3477 hw
->mac
.ops
.set_mac_anti_spoofing(hw
, (adapter
->num_vfs
!= 0),
3479 /* For VFs that have spoof checking turned off */
3480 for (i
= 0; i
< adapter
->num_vfs
; i
++) {
3481 if (!adapter
->vfinfo
[i
].spoofchk_enabled
)
3482 ixgbe_ndo_set_vf_spoofchk(adapter
->netdev
, i
, false);
3486 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter
*adapter
)
3488 struct ixgbe_hw
*hw
= &adapter
->hw
;
3489 struct net_device
*netdev
= adapter
->netdev
;
3490 int max_frame
= netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
3491 struct ixgbe_ring
*rx_ring
;
3496 /* adjust max frame to be able to do baby jumbo for FCoE */
3497 if ((adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
) &&
3498 (max_frame
< IXGBE_FCOE_JUMBO_FRAME_SIZE
))
3499 max_frame
= IXGBE_FCOE_JUMBO_FRAME_SIZE
;
3501 #endif /* IXGBE_FCOE */
3503 /* adjust max frame to be at least the size of a standard frame */
3504 if (max_frame
< (ETH_FRAME_LEN
+ ETH_FCS_LEN
))
3505 max_frame
= (ETH_FRAME_LEN
+ ETH_FCS_LEN
);
3507 mhadd
= IXGBE_READ_REG(hw
, IXGBE_MHADD
);
3508 if (max_frame
!= (mhadd
>> IXGBE_MHADD_MFS_SHIFT
)) {
3509 mhadd
&= ~IXGBE_MHADD_MFS_MASK
;
3510 mhadd
|= max_frame
<< IXGBE_MHADD_MFS_SHIFT
;
3512 IXGBE_WRITE_REG(hw
, IXGBE_MHADD
, mhadd
);
3515 hlreg0
= IXGBE_READ_REG(hw
, IXGBE_HLREG0
);
3516 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3517 hlreg0
|= IXGBE_HLREG0_JUMBOEN
;
3518 IXGBE_WRITE_REG(hw
, IXGBE_HLREG0
, hlreg0
);
3521 * Setup the HW Rx Head and Tail Descriptor Pointers and
3522 * the Base and Length of the Rx Descriptor Ring
3524 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3525 rx_ring
= adapter
->rx_ring
[i
];
3526 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)
3527 set_ring_rsc_enabled(rx_ring
);
3529 clear_ring_rsc_enabled(rx_ring
);
3533 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter
*adapter
)
3535 struct ixgbe_hw
*hw
= &adapter
->hw
;
3536 u32 rdrxctl
= IXGBE_READ_REG(hw
, IXGBE_RDRXCTL
);
3538 switch (hw
->mac
.type
) {
3539 case ixgbe_mac_82598EB
:
3541 * For VMDq support of different descriptor types or
3542 * buffer sizes through the use of multiple SRRCTL
3543 * registers, RDRXCTL.MVMEN must be set to 1
3545 * also, the manual doesn't mention it clearly but DCA hints
3546 * will only use queue 0's tags unless this bit is set. Side
3547 * effects of setting this bit are only that SRRCTL must be
3548 * fully programmed [0..15]
3550 rdrxctl
|= IXGBE_RDRXCTL_MVMEN
;
3552 case ixgbe_mac_82599EB
:
3553 case ixgbe_mac_X540
:
3554 /* Disable RSC for ACK packets */
3555 IXGBE_WRITE_REG(hw
, IXGBE_RSCDBU
,
3556 (IXGBE_RSCDBU_RSCACKDIS
| IXGBE_READ_REG(hw
, IXGBE_RSCDBU
)));
3557 rdrxctl
&= ~IXGBE_RDRXCTL_RSCFRSTSIZE
;
3558 /* hardware requires some bits to be set by default */
3559 rdrxctl
|= (IXGBE_RDRXCTL_RSCACKC
| IXGBE_RDRXCTL_FCOE_WRFIX
);
3560 rdrxctl
|= IXGBE_RDRXCTL_CRCSTRIP
;
3563 /* We should do nothing since we don't know this hardware */
3567 IXGBE_WRITE_REG(hw
, IXGBE_RDRXCTL
, rdrxctl
);
3571 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3572 * @adapter: board private structure
3574 * Configure the Rx unit of the MAC after a reset.
3576 static void ixgbe_configure_rx(struct ixgbe_adapter
*adapter
)
3578 struct ixgbe_hw
*hw
= &adapter
->hw
;
3582 /* disable receives while setting up the descriptors */
3583 rxctrl
= IXGBE_READ_REG(hw
, IXGBE_RXCTRL
);
3584 IXGBE_WRITE_REG(hw
, IXGBE_RXCTRL
, rxctrl
& ~IXGBE_RXCTRL_RXEN
);
3586 ixgbe_setup_psrtype(adapter
);
3587 ixgbe_setup_rdrxctl(adapter
);
3590 rfctl
= IXGBE_READ_REG(hw
, IXGBE_RFCTL
);
3591 rfctl
&= ~IXGBE_RFCTL_RSC_DIS
;
3592 if (!(adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
))
3593 rfctl
|= IXGBE_RFCTL_RSC_DIS
;
3594 IXGBE_WRITE_REG(hw
, IXGBE_RFCTL
, rfctl
);
3596 /* Program registers for the distribution of queues */
3597 ixgbe_setup_mrqc(adapter
);
3599 /* set_rx_buffer_len must be called before ring initialization */
3600 ixgbe_set_rx_buffer_len(adapter
);
3603 * Setup the HW Rx Head and Tail Descriptor Pointers and
3604 * the Base and Length of the Rx Descriptor Ring
3606 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3607 ixgbe_configure_rx_ring(adapter
, adapter
->rx_ring
[i
]);
3609 /* disable drop enable for 82598 parts */
3610 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
3611 rxctrl
|= IXGBE_RXCTRL_DMBYPS
;
3613 /* enable all receives */
3614 rxctrl
|= IXGBE_RXCTRL_RXEN
;
3615 hw
->mac
.ops
.enable_rx_dma(hw
, rxctrl
);
3618 static int ixgbe_vlan_rx_add_vid(struct net_device
*netdev
,
3619 __be16 proto
, u16 vid
)
3621 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
3622 struct ixgbe_hw
*hw
= &adapter
->hw
;
3624 /* add VID to filter table */
3625 hw
->mac
.ops
.set_vfta(&adapter
->hw
, vid
, VMDQ_P(0), true);
3626 set_bit(vid
, adapter
->active_vlans
);
3631 static int ixgbe_vlan_rx_kill_vid(struct net_device
*netdev
,
3632 __be16 proto
, u16 vid
)
3634 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
3635 struct ixgbe_hw
*hw
= &adapter
->hw
;
3637 /* remove VID from filter table */
3638 hw
->mac
.ops
.set_vfta(&adapter
->hw
, vid
, VMDQ_P(0), false);
3639 clear_bit(vid
, adapter
->active_vlans
);
3645 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3646 * @adapter: driver data
3648 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter
*adapter
)
3650 struct ixgbe_hw
*hw
= &adapter
->hw
;
3654 switch (hw
->mac
.type
) {
3655 case ixgbe_mac_82598EB
:
3656 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
3657 vlnctrl
&= ~IXGBE_VLNCTRL_VME
;
3658 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
3660 case ixgbe_mac_82599EB
:
3661 case ixgbe_mac_X540
:
3662 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3663 struct ixgbe_ring
*ring
= adapter
->rx_ring
[i
];
3665 if (ring
->l2_accel_priv
)
3668 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(j
));
3669 vlnctrl
&= ~IXGBE_RXDCTL_VME
;
3670 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(j
), vlnctrl
);
3679 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3680 * @adapter: driver data
3682 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter
*adapter
)
3684 struct ixgbe_hw
*hw
= &adapter
->hw
;
3688 switch (hw
->mac
.type
) {
3689 case ixgbe_mac_82598EB
:
3690 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
3691 vlnctrl
|= IXGBE_VLNCTRL_VME
;
3692 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
3694 case ixgbe_mac_82599EB
:
3695 case ixgbe_mac_X540
:
3696 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3697 struct ixgbe_ring
*ring
= adapter
->rx_ring
[i
];
3699 if (ring
->l2_accel_priv
)
3702 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(j
));
3703 vlnctrl
|= IXGBE_RXDCTL_VME
;
3704 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(j
), vlnctrl
);
3712 static void ixgbe_restore_vlan(struct ixgbe_adapter
*adapter
)
3716 ixgbe_vlan_rx_add_vid(adapter
->netdev
, htons(ETH_P_8021Q
), 0);
3718 for_each_set_bit(vid
, adapter
->active_vlans
, VLAN_N_VID
)
3719 ixgbe_vlan_rx_add_vid(adapter
->netdev
, htons(ETH_P_8021Q
), vid
);
3723 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
3724 * @netdev: network interface device structure
3726 * Writes multicast address list to the MTA hash table.
3727 * Returns: -ENOMEM on failure
3728 * 0 on no addresses written
3729 * X on writing X addresses to MTA
3731 static int ixgbe_write_mc_addr_list(struct net_device
*netdev
)
3733 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
3734 struct ixgbe_hw
*hw
= &adapter
->hw
;
3736 if (!netif_running(netdev
))
3739 if (hw
->mac
.ops
.update_mc_addr_list
)
3740 hw
->mac
.ops
.update_mc_addr_list(hw
, netdev
);
3744 #ifdef CONFIG_PCI_IOV
3745 ixgbe_restore_vf_multicasts(adapter
);
3748 return netdev_mc_count(netdev
);
3751 #ifdef CONFIG_PCI_IOV
3752 void ixgbe_full_sync_mac_table(struct ixgbe_adapter
*adapter
)
3754 struct ixgbe_hw
*hw
= &adapter
->hw
;
3756 for (i
= 0; i
< hw
->mac
.num_rar_entries
; i
++) {
3757 if (adapter
->mac_table
[i
].state
& IXGBE_MAC_STATE_IN_USE
)
3758 hw
->mac
.ops
.set_rar(hw
, i
, adapter
->mac_table
[i
].addr
,
3759 adapter
->mac_table
[i
].queue
,
3762 hw
->mac
.ops
.clear_rar(hw
, i
);
3764 adapter
->mac_table
[i
].state
&= ~(IXGBE_MAC_STATE_MODIFIED
);
3769 static void ixgbe_sync_mac_table(struct ixgbe_adapter
*adapter
)
3771 struct ixgbe_hw
*hw
= &adapter
->hw
;
3773 for (i
= 0; i
< hw
->mac
.num_rar_entries
; i
++) {
3774 if (adapter
->mac_table
[i
].state
& IXGBE_MAC_STATE_MODIFIED
) {
3775 if (adapter
->mac_table
[i
].state
&
3776 IXGBE_MAC_STATE_IN_USE
)
3777 hw
->mac
.ops
.set_rar(hw
, i
,
3778 adapter
->mac_table
[i
].addr
,
3779 adapter
->mac_table
[i
].queue
,
3782 hw
->mac
.ops
.clear_rar(hw
, i
);
3784 adapter
->mac_table
[i
].state
&=
3785 ~(IXGBE_MAC_STATE_MODIFIED
);
3790 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter
*adapter
)
3793 struct ixgbe_hw
*hw
= &adapter
->hw
;
3795 for (i
= 0; i
< hw
->mac
.num_rar_entries
; i
++) {
3796 adapter
->mac_table
[i
].state
|= IXGBE_MAC_STATE_MODIFIED
;
3797 adapter
->mac_table
[i
].state
&= ~IXGBE_MAC_STATE_IN_USE
;
3798 memset(adapter
->mac_table
[i
].addr
, 0, ETH_ALEN
);
3799 adapter
->mac_table
[i
].queue
= 0;
3801 ixgbe_sync_mac_table(adapter
);
3804 static int ixgbe_available_rars(struct ixgbe_adapter
*adapter
)
3806 struct ixgbe_hw
*hw
= &adapter
->hw
;
3809 for (i
= 0; i
< hw
->mac
.num_rar_entries
; i
++) {
3810 if (adapter
->mac_table
[i
].state
== 0)
3816 /* this function destroys the first RAR entry */
3817 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter
*adapter
,
3820 struct ixgbe_hw
*hw
= &adapter
->hw
;
3822 memcpy(&adapter
->mac_table
[0].addr
, addr
, ETH_ALEN
);
3823 adapter
->mac_table
[0].queue
= VMDQ_P(0);
3824 adapter
->mac_table
[0].state
= (IXGBE_MAC_STATE_DEFAULT
|
3825 IXGBE_MAC_STATE_IN_USE
);
3826 hw
->mac
.ops
.set_rar(hw
, 0, adapter
->mac_table
[0].addr
,
3827 adapter
->mac_table
[0].queue
,
3831 int ixgbe_add_mac_filter(struct ixgbe_adapter
*adapter
, u8
*addr
, u16 queue
)
3833 struct ixgbe_hw
*hw
= &adapter
->hw
;
3836 if (is_zero_ether_addr(addr
))
3839 for (i
= 0; i
< hw
->mac
.num_rar_entries
; i
++) {
3840 if (adapter
->mac_table
[i
].state
& IXGBE_MAC_STATE_IN_USE
)
3842 adapter
->mac_table
[i
].state
|= (IXGBE_MAC_STATE_MODIFIED
|
3843 IXGBE_MAC_STATE_IN_USE
);
3844 ether_addr_copy(adapter
->mac_table
[i
].addr
, addr
);
3845 adapter
->mac_table
[i
].queue
= queue
;
3846 ixgbe_sync_mac_table(adapter
);
3852 int ixgbe_del_mac_filter(struct ixgbe_adapter
*adapter
, u8
*addr
, u16 queue
)
3854 /* search table for addr, if found, set to 0 and sync */
3856 struct ixgbe_hw
*hw
= &adapter
->hw
;
3858 if (is_zero_ether_addr(addr
))
3861 for (i
= 0; i
< hw
->mac
.num_rar_entries
; i
++) {
3862 if (ether_addr_equal(addr
, adapter
->mac_table
[i
].addr
) &&
3863 adapter
->mac_table
[i
].queue
== queue
) {
3864 adapter
->mac_table
[i
].state
|= IXGBE_MAC_STATE_MODIFIED
;
3865 adapter
->mac_table
[i
].state
&= ~IXGBE_MAC_STATE_IN_USE
;
3866 memset(adapter
->mac_table
[i
].addr
, 0, ETH_ALEN
);
3867 adapter
->mac_table
[i
].queue
= 0;
3868 ixgbe_sync_mac_table(adapter
);
3875 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3876 * @netdev: network interface device structure
3878 * Writes unicast address list to the RAR table.
3879 * Returns: -ENOMEM on failure/insufficient address space
3880 * 0 on no addresses written
3881 * X on writing X addresses to the RAR table
3883 static int ixgbe_write_uc_addr_list(struct net_device
*netdev
, int vfn
)
3885 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
3888 /* return ENOMEM indicating insufficient memory for addresses */
3889 if (netdev_uc_count(netdev
) > ixgbe_available_rars(adapter
))
3892 if (!netdev_uc_empty(netdev
)) {
3893 struct netdev_hw_addr
*ha
;
3894 netdev_for_each_uc_addr(ha
, netdev
) {
3895 ixgbe_del_mac_filter(adapter
, ha
->addr
, vfn
);
3896 ixgbe_add_mac_filter(adapter
, ha
->addr
, vfn
);
3904 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3905 * @netdev: network interface device structure
3907 * The set_rx_method entry point is called whenever the unicast/multicast
3908 * address list or the network interface flags are updated. This routine is
3909 * responsible for configuring the hardware for proper unicast, multicast and
3912 void ixgbe_set_rx_mode(struct net_device
*netdev
)
3914 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
3915 struct ixgbe_hw
*hw
= &adapter
->hw
;
3916 u32 fctrl
, vmolr
= IXGBE_VMOLR_BAM
| IXGBE_VMOLR_AUPE
;
3920 /* Check for Promiscuous and All Multicast modes */
3921 fctrl
= IXGBE_READ_REG(hw
, IXGBE_FCTRL
);
3922 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
3924 /* set all bits that we expect to always be set */
3925 fctrl
&= ~IXGBE_FCTRL_SBP
; /* disable store-bad-packets */
3926 fctrl
|= IXGBE_FCTRL_BAM
;
3927 fctrl
|= IXGBE_FCTRL_DPF
; /* discard pause frames when FC enabled */
3928 fctrl
|= IXGBE_FCTRL_PMCF
;
3930 /* clear the bits we are changing the status of */
3931 fctrl
&= ~(IXGBE_FCTRL_UPE
| IXGBE_FCTRL_MPE
);
3932 vlnctrl
&= ~(IXGBE_VLNCTRL_VFE
| IXGBE_VLNCTRL_CFIEN
);
3933 if (netdev
->flags
& IFF_PROMISC
) {
3934 hw
->addr_ctrl
.user_set_promisc
= true;
3935 fctrl
|= (IXGBE_FCTRL_UPE
| IXGBE_FCTRL_MPE
);
3936 vmolr
|= IXGBE_VMOLR_MPE
;
3937 /* Only disable hardware filter vlans in promiscuous mode
3938 * if SR-IOV and VMDQ are disabled - otherwise ensure
3939 * that hardware VLAN filters remain enabled.
3941 if (!(adapter
->flags
& (IXGBE_FLAG_VMDQ_ENABLED
|
3942 IXGBE_FLAG_SRIOV_ENABLED
)))
3943 vlnctrl
|= (IXGBE_VLNCTRL_VFE
| IXGBE_VLNCTRL_CFIEN
);
3945 if (netdev
->flags
& IFF_ALLMULTI
) {
3946 fctrl
|= IXGBE_FCTRL_MPE
;
3947 vmolr
|= IXGBE_VMOLR_MPE
;
3949 vlnctrl
|= IXGBE_VLNCTRL_VFE
;
3950 hw
->addr_ctrl
.user_set_promisc
= false;
3954 * Write addresses to available RAR registers, if there is not
3955 * sufficient space to store all the addresses then enable
3956 * unicast promiscuous mode
3958 count
= ixgbe_write_uc_addr_list(netdev
, VMDQ_P(0));
3960 fctrl
|= IXGBE_FCTRL_UPE
;
3961 vmolr
|= IXGBE_VMOLR_ROPE
;
3964 /* Write addresses to the MTA, if the attempt fails
3965 * then we should just turn on promiscuous mode so
3966 * that we can at least receive multicast traffic
3968 count
= ixgbe_write_mc_addr_list(netdev
);
3970 fctrl
|= IXGBE_FCTRL_MPE
;
3971 vmolr
|= IXGBE_VMOLR_MPE
;
3973 vmolr
|= IXGBE_VMOLR_ROMPE
;
3976 if (hw
->mac
.type
!= ixgbe_mac_82598EB
) {
3977 vmolr
|= IXGBE_READ_REG(hw
, IXGBE_VMOLR(VMDQ_P(0))) &
3978 ~(IXGBE_VMOLR_MPE
| IXGBE_VMOLR_ROMPE
|
3980 IXGBE_WRITE_REG(hw
, IXGBE_VMOLR(VMDQ_P(0)), vmolr
);
3983 /* This is useful for sniffing bad packets. */
3984 if (adapter
->netdev
->features
& NETIF_F_RXALL
) {
3985 /* UPE and MPE will be handled by normal PROMISC logic
3986 * in e1000e_set_rx_mode */
3987 fctrl
|= (IXGBE_FCTRL_SBP
| /* Receive bad packets */
3988 IXGBE_FCTRL_BAM
| /* RX All Bcast Pkts */
3989 IXGBE_FCTRL_PMCF
); /* RX All MAC Ctrl Pkts */
3991 fctrl
&= ~(IXGBE_FCTRL_DPF
);
3992 /* NOTE: VLAN filtering is disabled by setting PROMISC */
3995 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
3996 IXGBE_WRITE_REG(hw
, IXGBE_FCTRL
, fctrl
);
3998 if (netdev
->features
& NETIF_F_HW_VLAN_CTAG_RX
)
3999 ixgbe_vlan_strip_enable(adapter
);
4001 ixgbe_vlan_strip_disable(adapter
);
4004 static void ixgbe_napi_enable_all(struct ixgbe_adapter
*adapter
)
4008 for (q_idx
= 0; q_idx
< adapter
->num_q_vectors
; q_idx
++) {
4009 ixgbe_qv_init_lock(adapter
->q_vector
[q_idx
]);
4010 napi_enable(&adapter
->q_vector
[q_idx
]->napi
);
4014 static void ixgbe_napi_disable_all(struct ixgbe_adapter
*adapter
)
4018 for (q_idx
= 0; q_idx
< adapter
->num_q_vectors
; q_idx
++) {
4019 napi_disable(&adapter
->q_vector
[q_idx
]->napi
);
4020 while (!ixgbe_qv_disable(adapter
->q_vector
[q_idx
])) {
4021 pr_info("QV %d locked\n", q_idx
);
4022 usleep_range(1000, 20000);
4027 #ifdef CONFIG_IXGBE_DCB
4029 * ixgbe_configure_dcb - Configure DCB hardware
4030 * @adapter: ixgbe adapter struct
4032 * This is called by the driver on open to configure the DCB hardware.
4033 * This is also called by the gennetlink interface when reconfiguring
4036 static void ixgbe_configure_dcb(struct ixgbe_adapter
*adapter
)
4038 struct ixgbe_hw
*hw
= &adapter
->hw
;
4039 int max_frame
= adapter
->netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
4041 if (!(adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
)) {
4042 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
4043 netif_set_gso_max_size(adapter
->netdev
, 65536);
4047 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
4048 netif_set_gso_max_size(adapter
->netdev
, 32768);
4051 if (adapter
->netdev
->features
& NETIF_F_FCOE_MTU
)
4052 max_frame
= max(max_frame
, IXGBE_FCOE_JUMBO_FRAME_SIZE
);
4055 /* reconfigure the hardware */
4056 if (adapter
->dcbx_cap
& DCB_CAP_DCBX_VER_CEE
) {
4057 ixgbe_dcb_calculate_tc_credits(hw
, &adapter
->dcb_cfg
, max_frame
,
4059 ixgbe_dcb_calculate_tc_credits(hw
, &adapter
->dcb_cfg
, max_frame
,
4061 ixgbe_dcb_hw_config(hw
, &adapter
->dcb_cfg
);
4062 } else if (adapter
->ixgbe_ieee_ets
&& adapter
->ixgbe_ieee_pfc
) {
4063 ixgbe_dcb_hw_ets(&adapter
->hw
,
4064 adapter
->ixgbe_ieee_ets
,
4066 ixgbe_dcb_hw_pfc_config(&adapter
->hw
,
4067 adapter
->ixgbe_ieee_pfc
->pfc_en
,
4068 adapter
->ixgbe_ieee_ets
->prio_tc
);
4071 /* Enable RSS Hash per TC */
4072 if (hw
->mac
.type
!= ixgbe_mac_82598EB
) {
4074 u16 rss_i
= adapter
->ring_feature
[RING_F_RSS
].indices
- 1;
4081 /* write msb to all 8 TCs in one write */
4082 IXGBE_WRITE_REG(hw
, IXGBE_RQTC
, msb
* 0x11111111);
4087 /* Additional bittime to account for IXGBE framing */
4088 #define IXGBE_ETH_FRAMING 20
4091 * ixgbe_hpbthresh - calculate high water mark for flow control
4093 * @adapter: board private structure to calculate for
4094 * @pb: packet buffer to calculate
4096 static int ixgbe_hpbthresh(struct ixgbe_adapter
*adapter
, int pb
)
4098 struct ixgbe_hw
*hw
= &adapter
->hw
;
4099 struct net_device
*dev
= adapter
->netdev
;
4100 int link
, tc
, kb
, marker
;
4103 /* Calculate max LAN frame size */
4104 tc
= link
= dev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+ IXGBE_ETH_FRAMING
;
4107 /* FCoE traffic class uses FCOE jumbo frames */
4108 if ((dev
->features
& NETIF_F_FCOE_MTU
) &&
4109 (tc
< IXGBE_FCOE_JUMBO_FRAME_SIZE
) &&
4110 (pb
== ixgbe_fcoe_get_tc(adapter
)))
4111 tc
= IXGBE_FCOE_JUMBO_FRAME_SIZE
;
4114 /* Calculate delay value for device */
4115 switch (hw
->mac
.type
) {
4116 case ixgbe_mac_X540
:
4117 dv_id
= IXGBE_DV_X540(link
, tc
);
4120 dv_id
= IXGBE_DV(link
, tc
);
4124 /* Loopback switch introduces additional latency */
4125 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
4126 dv_id
+= IXGBE_B2BT(tc
);
4128 /* Delay value is calculated in bit times convert to KB */
4129 kb
= IXGBE_BT2KB(dv_id
);
4130 rx_pba
= IXGBE_READ_REG(hw
, IXGBE_RXPBSIZE(pb
)) >> 10;
4132 marker
= rx_pba
- kb
;
4134 /* It is possible that the packet buffer is not large enough
4135 * to provide required headroom. In this case throw an error
4136 * to user and a do the best we can.
4139 e_warn(drv
, "Packet Buffer(%i) can not provide enough"
4140 "headroom to support flow control."
4141 "Decrease MTU or number of traffic classes\n", pb
);
4149 * ixgbe_lpbthresh - calculate low water mark for for flow control
4151 * @adapter: board private structure to calculate for
4152 * @pb: packet buffer to calculate
4154 static int ixgbe_lpbthresh(struct ixgbe_adapter
*adapter
, int pb
)
4156 struct ixgbe_hw
*hw
= &adapter
->hw
;
4157 struct net_device
*dev
= adapter
->netdev
;
4161 /* Calculate max LAN frame size */
4162 tc
= dev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
4165 /* FCoE traffic class uses FCOE jumbo frames */
4166 if ((dev
->features
& NETIF_F_FCOE_MTU
) &&
4167 (tc
< IXGBE_FCOE_JUMBO_FRAME_SIZE
) &&
4168 (pb
== netdev_get_prio_tc_map(dev
, adapter
->fcoe
.up
)))
4169 tc
= IXGBE_FCOE_JUMBO_FRAME_SIZE
;
4172 /* Calculate delay value for device */
4173 switch (hw
->mac
.type
) {
4174 case ixgbe_mac_X540
:
4175 dv_id
= IXGBE_LOW_DV_X540(tc
);
4178 dv_id
= IXGBE_LOW_DV(tc
);
4182 /* Delay value is calculated in bit times convert to KB */
4183 return IXGBE_BT2KB(dv_id
);
4187 * ixgbe_pbthresh_setup - calculate and setup high low water marks
4189 static void ixgbe_pbthresh_setup(struct ixgbe_adapter
*adapter
)
4191 struct ixgbe_hw
*hw
= &adapter
->hw
;
4192 int num_tc
= netdev_get_num_tc(adapter
->netdev
);
4198 for (i
= 0; i
< num_tc
; i
++) {
4199 hw
->fc
.high_water
[i
] = ixgbe_hpbthresh(adapter
, i
);
4200 hw
->fc
.low_water
[i
] = ixgbe_lpbthresh(adapter
, i
);
4202 /* Low water marks must not be larger than high water marks */
4203 if (hw
->fc
.low_water
[i
] > hw
->fc
.high_water
[i
])
4204 hw
->fc
.low_water
[i
] = 0;
4207 for (; i
< MAX_TRAFFIC_CLASS
; i
++)
4208 hw
->fc
.high_water
[i
] = 0;
4211 static void ixgbe_configure_pb(struct ixgbe_adapter
*adapter
)
4213 struct ixgbe_hw
*hw
= &adapter
->hw
;
4215 u8 tc
= netdev_get_num_tc(adapter
->netdev
);
4217 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
||
4218 adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
)
4219 hdrm
= 32 << adapter
->fdir_pballoc
;
4223 hw
->mac
.ops
.set_rxpba(hw
, tc
, hdrm
, PBA_STRATEGY_EQUAL
);
4224 ixgbe_pbthresh_setup(adapter
);
4227 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter
*adapter
)
4229 struct ixgbe_hw
*hw
= &adapter
->hw
;
4230 struct hlist_node
*node2
;
4231 struct ixgbe_fdir_filter
*filter
;
4233 spin_lock(&adapter
->fdir_perfect_lock
);
4235 if (!hlist_empty(&adapter
->fdir_filter_list
))
4236 ixgbe_fdir_set_input_mask_82599(hw
, &adapter
->fdir_mask
);
4238 hlist_for_each_entry_safe(filter
, node2
,
4239 &adapter
->fdir_filter_list
, fdir_node
) {
4240 ixgbe_fdir_write_perfect_filter_82599(hw
,
4243 (filter
->action
== IXGBE_FDIR_DROP_QUEUE
) ?
4244 IXGBE_FDIR_DROP_QUEUE
:
4245 adapter
->rx_ring
[filter
->action
]->reg_idx
);
4248 spin_unlock(&adapter
->fdir_perfect_lock
);
4251 static void ixgbe_macvlan_set_rx_mode(struct net_device
*dev
, unsigned int pool
,
4252 struct ixgbe_adapter
*adapter
)
4254 struct ixgbe_hw
*hw
= &adapter
->hw
;
4257 /* No unicast promiscuous support for VMDQ devices. */
4258 vmolr
= IXGBE_READ_REG(hw
, IXGBE_VMOLR(pool
));
4259 vmolr
|= (IXGBE_VMOLR_ROMPE
| IXGBE_VMOLR_BAM
| IXGBE_VMOLR_AUPE
);
4261 /* clear the affected bit */
4262 vmolr
&= ~IXGBE_VMOLR_MPE
;
4264 if (dev
->flags
& IFF_ALLMULTI
) {
4265 vmolr
|= IXGBE_VMOLR_MPE
;
4267 vmolr
|= IXGBE_VMOLR_ROMPE
;
4268 hw
->mac
.ops
.update_mc_addr_list(hw
, dev
);
4270 ixgbe_write_uc_addr_list(adapter
->netdev
, pool
);
4271 IXGBE_WRITE_REG(hw
, IXGBE_VMOLR(pool
), vmolr
);
4274 static void ixgbe_fwd_psrtype(struct ixgbe_fwd_adapter
*vadapter
)
4276 struct ixgbe_adapter
*adapter
= vadapter
->real_adapter
;
4277 int rss_i
= adapter
->num_rx_queues_per_pool
;
4278 struct ixgbe_hw
*hw
= &adapter
->hw
;
4279 u16 pool
= vadapter
->pool
;
4280 u32 psrtype
= IXGBE_PSRTYPE_TCPHDR
|
4281 IXGBE_PSRTYPE_UDPHDR
|
4282 IXGBE_PSRTYPE_IPV4HDR
|
4283 IXGBE_PSRTYPE_L2HDR
|
4284 IXGBE_PSRTYPE_IPV6HDR
;
4286 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
4294 IXGBE_WRITE_REG(hw
, IXGBE_PSRTYPE(VMDQ_P(pool
)), psrtype
);
4298 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4299 * @rx_ring: ring to free buffers from
4301 static void ixgbe_clean_rx_ring(struct ixgbe_ring
*rx_ring
)
4303 struct device
*dev
= rx_ring
->dev
;
4307 /* ring already cleared, nothing to do */
4308 if (!rx_ring
->rx_buffer_info
)
4311 /* Free all the Rx ring sk_buffs */
4312 for (i
= 0; i
< rx_ring
->count
; i
++) {
4313 struct ixgbe_rx_buffer
*rx_buffer
;
4315 rx_buffer
= &rx_ring
->rx_buffer_info
[i
];
4316 if (rx_buffer
->skb
) {
4317 struct sk_buff
*skb
= rx_buffer
->skb
;
4318 if (IXGBE_CB(skb
)->page_released
) {
4321 ixgbe_rx_bufsz(rx_ring
),
4323 IXGBE_CB(skb
)->page_released
= false;
4327 rx_buffer
->skb
= NULL
;
4329 dma_unmap_page(dev
, rx_buffer
->dma
,
4330 ixgbe_rx_pg_size(rx_ring
),
4333 if (rx_buffer
->page
)
4334 __free_pages(rx_buffer
->page
,
4335 ixgbe_rx_pg_order(rx_ring
));
4336 rx_buffer
->page
= NULL
;
4339 size
= sizeof(struct ixgbe_rx_buffer
) * rx_ring
->count
;
4340 memset(rx_ring
->rx_buffer_info
, 0, size
);
4342 /* Zero out the descriptor ring */
4343 memset(rx_ring
->desc
, 0, rx_ring
->size
);
4345 rx_ring
->next_to_alloc
= 0;
4346 rx_ring
->next_to_clean
= 0;
4347 rx_ring
->next_to_use
= 0;
4350 static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter
*vadapter
,
4351 struct ixgbe_ring
*rx_ring
)
4353 struct ixgbe_adapter
*adapter
= vadapter
->real_adapter
;
4354 int index
= rx_ring
->queue_index
+ vadapter
->rx_base_queue
;
4356 /* shutdown specific queue receive and wait for dma to settle */
4357 ixgbe_disable_rx_queue(adapter
, rx_ring
);
4358 usleep_range(10000, 20000);
4359 ixgbe_irq_disable_queues(adapter
, ((u64
)1 << index
));
4360 ixgbe_clean_rx_ring(rx_ring
);
4361 rx_ring
->l2_accel_priv
= NULL
;
4364 static int ixgbe_fwd_ring_down(struct net_device
*vdev
,
4365 struct ixgbe_fwd_adapter
*accel
)
4367 struct ixgbe_adapter
*adapter
= accel
->real_adapter
;
4368 unsigned int rxbase
= accel
->rx_base_queue
;
4369 unsigned int txbase
= accel
->tx_base_queue
;
4372 netif_tx_stop_all_queues(vdev
);
4374 for (i
= 0; i
< adapter
->num_rx_queues_per_pool
; i
++) {
4375 ixgbe_disable_fwd_ring(accel
, adapter
->rx_ring
[rxbase
+ i
]);
4376 adapter
->rx_ring
[rxbase
+ i
]->netdev
= adapter
->netdev
;
4379 for (i
= 0; i
< adapter
->num_rx_queues_per_pool
; i
++) {
4380 adapter
->tx_ring
[txbase
+ i
]->l2_accel_priv
= NULL
;
4381 adapter
->tx_ring
[txbase
+ i
]->netdev
= adapter
->netdev
;
4388 static int ixgbe_fwd_ring_up(struct net_device
*vdev
,
4389 struct ixgbe_fwd_adapter
*accel
)
4391 struct ixgbe_adapter
*adapter
= accel
->real_adapter
;
4392 unsigned int rxbase
, txbase
, queues
;
4393 int i
, baseq
, err
= 0;
4395 if (!test_bit(accel
->pool
, &adapter
->fwd_bitmask
))
4398 baseq
= accel
->pool
* adapter
->num_rx_queues_per_pool
;
4399 netdev_dbg(vdev
, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
4400 accel
->pool
, adapter
->num_rx_pools
,
4401 baseq
, baseq
+ adapter
->num_rx_queues_per_pool
,
4402 adapter
->fwd_bitmask
);
4404 accel
->netdev
= vdev
;
4405 accel
->rx_base_queue
= rxbase
= baseq
;
4406 accel
->tx_base_queue
= txbase
= baseq
;
4408 for (i
= 0; i
< adapter
->num_rx_queues_per_pool
; i
++)
4409 ixgbe_disable_fwd_ring(accel
, adapter
->rx_ring
[rxbase
+ i
]);
4411 for (i
= 0; i
< adapter
->num_rx_queues_per_pool
; i
++) {
4412 adapter
->rx_ring
[rxbase
+ i
]->netdev
= vdev
;
4413 adapter
->rx_ring
[rxbase
+ i
]->l2_accel_priv
= accel
;
4414 ixgbe_configure_rx_ring(adapter
, adapter
->rx_ring
[rxbase
+ i
]);
4417 for (i
= 0; i
< adapter
->num_rx_queues_per_pool
; i
++) {
4418 adapter
->tx_ring
[txbase
+ i
]->netdev
= vdev
;
4419 adapter
->tx_ring
[txbase
+ i
]->l2_accel_priv
= accel
;
4422 queues
= min_t(unsigned int,
4423 adapter
->num_rx_queues_per_pool
, vdev
->num_tx_queues
);
4424 err
= netif_set_real_num_tx_queues(vdev
, queues
);
4428 err
= netif_set_real_num_rx_queues(vdev
, queues
);
4432 if (is_valid_ether_addr(vdev
->dev_addr
))
4433 ixgbe_add_mac_filter(adapter
, vdev
->dev_addr
, accel
->pool
);
4435 ixgbe_fwd_psrtype(accel
);
4436 ixgbe_macvlan_set_rx_mode(vdev
, accel
->pool
, adapter
);
4439 ixgbe_fwd_ring_down(vdev
, accel
);
4443 static void ixgbe_configure_dfwd(struct ixgbe_adapter
*adapter
)
4445 struct net_device
*upper
;
4446 struct list_head
*iter
;
4449 netdev_for_each_all_upper_dev_rcu(adapter
->netdev
, upper
, iter
) {
4450 if (netif_is_macvlan(upper
)) {
4451 struct macvlan_dev
*dfwd
= netdev_priv(upper
);
4452 struct ixgbe_fwd_adapter
*vadapter
= dfwd
->fwd_priv
;
4454 if (dfwd
->fwd_priv
) {
4455 err
= ixgbe_fwd_ring_up(upper
, vadapter
);
4463 static void ixgbe_configure(struct ixgbe_adapter
*adapter
)
4465 struct ixgbe_hw
*hw
= &adapter
->hw
;
4467 ixgbe_configure_pb(adapter
);
4468 #ifdef CONFIG_IXGBE_DCB
4469 ixgbe_configure_dcb(adapter
);
4472 * We must restore virtualization before VLANs or else
4473 * the VLVF registers will not be populated
4475 ixgbe_configure_virtualization(adapter
);
4477 ixgbe_set_rx_mode(adapter
->netdev
);
4478 ixgbe_restore_vlan(adapter
);
4480 switch (hw
->mac
.type
) {
4481 case ixgbe_mac_82599EB
:
4482 case ixgbe_mac_X540
:
4483 hw
->mac
.ops
.disable_rx_buff(hw
);
4489 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) {
4490 ixgbe_init_fdir_signature_82599(&adapter
->hw
,
4491 adapter
->fdir_pballoc
);
4492 } else if (adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
) {
4493 ixgbe_init_fdir_perfect_82599(&adapter
->hw
,
4494 adapter
->fdir_pballoc
);
4495 ixgbe_fdir_filter_restore(adapter
);
4498 switch (hw
->mac
.type
) {
4499 case ixgbe_mac_82599EB
:
4500 case ixgbe_mac_X540
:
4501 hw
->mac
.ops
.enable_rx_buff(hw
);
4508 /* configure FCoE L2 filters, redirection table, and Rx control */
4509 ixgbe_configure_fcoe(adapter
);
4511 #endif /* IXGBE_FCOE */
4512 ixgbe_configure_tx(adapter
);
4513 ixgbe_configure_rx(adapter
);
4514 ixgbe_configure_dfwd(adapter
);
4517 static inline bool ixgbe_is_sfp(struct ixgbe_hw
*hw
)
4519 switch (hw
->phy
.type
) {
4520 case ixgbe_phy_sfp_avago
:
4521 case ixgbe_phy_sfp_ftl
:
4522 case ixgbe_phy_sfp_intel
:
4523 case ixgbe_phy_sfp_unknown
:
4524 case ixgbe_phy_sfp_passive_tyco
:
4525 case ixgbe_phy_sfp_passive_unknown
:
4526 case ixgbe_phy_sfp_active_unknown
:
4527 case ixgbe_phy_sfp_ftl_active
:
4528 case ixgbe_phy_qsfp_passive_unknown
:
4529 case ixgbe_phy_qsfp_active_unknown
:
4530 case ixgbe_phy_qsfp_intel
:
4531 case ixgbe_phy_qsfp_unknown
:
4532 /* ixgbe_phy_none is set when no SFP module is present */
4533 case ixgbe_phy_none
:
4536 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
4544 * ixgbe_sfp_link_config - set up SFP+ link
4545 * @adapter: pointer to private adapter struct
4547 static void ixgbe_sfp_link_config(struct ixgbe_adapter
*adapter
)
4550 * We are assuming the worst case scenario here, and that
4551 * is that an SFP was inserted/removed after the reset
4552 * but before SFP detection was enabled. As such the best
4553 * solution is to just start searching as soon as we start
4555 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
)
4556 adapter
->flags2
|= IXGBE_FLAG2_SEARCH_FOR_SFP
;
4558 adapter
->flags2
|= IXGBE_FLAG2_SFP_NEEDS_RESET
;
4562 * ixgbe_non_sfp_link_config - set up non-SFP+ link
4563 * @hw: pointer to private hardware struct
4565 * Returns 0 on success, negative on failure
4567 static int ixgbe_non_sfp_link_config(struct ixgbe_hw
*hw
)
4570 bool autoneg
, link_up
= false;
4571 u32 ret
= IXGBE_ERR_LINK_SETUP
;
4573 if (hw
->mac
.ops
.check_link
)
4574 ret
= hw
->mac
.ops
.check_link(hw
, &speed
, &link_up
, false);
4579 speed
= hw
->phy
.autoneg_advertised
;
4580 if ((!speed
) && (hw
->mac
.ops
.get_link_capabilities
))
4581 ret
= hw
->mac
.ops
.get_link_capabilities(hw
, &speed
,
4586 if (hw
->mac
.ops
.setup_link
)
4587 ret
= hw
->mac
.ops
.setup_link(hw
, speed
, link_up
);
4592 static void ixgbe_setup_gpie(struct ixgbe_adapter
*adapter
)
4594 struct ixgbe_hw
*hw
= &adapter
->hw
;
4597 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
4598 gpie
= IXGBE_GPIE_MSIX_MODE
| IXGBE_GPIE_PBA_SUPPORT
|
4600 gpie
|= IXGBE_GPIE_EIAME
;
4602 * use EIAM to auto-mask when MSI-X interrupt is asserted
4603 * this saves a register write for every interrupt
4605 switch (hw
->mac
.type
) {
4606 case ixgbe_mac_82598EB
:
4607 IXGBE_WRITE_REG(hw
, IXGBE_EIAM
, IXGBE_EICS_RTX_QUEUE
);
4609 case ixgbe_mac_82599EB
:
4610 case ixgbe_mac_X540
:
4612 IXGBE_WRITE_REG(hw
, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4613 IXGBE_WRITE_REG(hw
, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4617 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
4618 * specifically only auto mask tx and rx interrupts */
4619 IXGBE_WRITE_REG(hw
, IXGBE_EIAM
, IXGBE_EICS_RTX_QUEUE
);
4622 /* XXX: to interrupt immediately for EICS writes, enable this */
4623 /* gpie |= IXGBE_GPIE_EIMEN; */
4625 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
4626 gpie
&= ~IXGBE_GPIE_VTMODE_MASK
;
4628 switch (adapter
->ring_feature
[RING_F_VMDQ
].mask
) {
4629 case IXGBE_82599_VMDQ_8Q_MASK
:
4630 gpie
|= IXGBE_GPIE_VTMODE_16
;
4632 case IXGBE_82599_VMDQ_4Q_MASK
:
4633 gpie
|= IXGBE_GPIE_VTMODE_32
;
4636 gpie
|= IXGBE_GPIE_VTMODE_64
;
4641 /* Enable Thermal over heat sensor interrupt */
4642 if (adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
) {
4643 switch (adapter
->hw
.mac
.type
) {
4644 case ixgbe_mac_82599EB
:
4645 gpie
|= IXGBE_SDP0_GPIEN
;
4647 case ixgbe_mac_X540
:
4648 gpie
|= IXGBE_EIMS_TS
;
4655 /* Enable fan failure interrupt */
4656 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
)
4657 gpie
|= IXGBE_SDP1_GPIEN
;
4659 if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
4660 gpie
|= IXGBE_SDP1_GPIEN
;
4661 gpie
|= IXGBE_SDP2_GPIEN
;
4664 IXGBE_WRITE_REG(hw
, IXGBE_GPIE
, gpie
);
4667 static void ixgbe_up_complete(struct ixgbe_adapter
*adapter
)
4669 struct ixgbe_hw
*hw
= &adapter
->hw
;
4673 ixgbe_get_hw_control(adapter
);
4674 ixgbe_setup_gpie(adapter
);
4676 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)
4677 ixgbe_configure_msix(adapter
);
4679 ixgbe_configure_msi_and_legacy(adapter
);
4681 /* enable the optics for 82599 SFP+ fiber */
4682 if (hw
->mac
.ops
.enable_tx_laser
)
4683 hw
->mac
.ops
.enable_tx_laser(hw
);
4685 smp_mb__before_atomic();
4686 clear_bit(__IXGBE_DOWN
, &adapter
->state
);
4687 ixgbe_napi_enable_all(adapter
);
4689 if (ixgbe_is_sfp(hw
)) {
4690 ixgbe_sfp_link_config(adapter
);
4692 err
= ixgbe_non_sfp_link_config(hw
);
4694 e_err(probe
, "link_config FAILED %d\n", err
);
4697 /* clear any pending interrupts, may auto mask */
4698 IXGBE_READ_REG(hw
, IXGBE_EICR
);
4699 ixgbe_irq_enable(adapter
, true, true);
4702 * If this adapter has a fan, check to see if we had a failure
4703 * before we enabled the interrupt.
4705 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) {
4706 u32 esdp
= IXGBE_READ_REG(hw
, IXGBE_ESDP
);
4707 if (esdp
& IXGBE_ESDP_SDP1
)
4708 e_crit(drv
, "Fan has stopped, replace the adapter\n");
4711 /* bring the link up in the watchdog, this could race with our first
4712 * link up interrupt but shouldn't be a problem */
4713 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_UPDATE
;
4714 adapter
->link_check_timeout
= jiffies
;
4715 mod_timer(&adapter
->service_timer
, jiffies
);
4717 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4718 ctrl_ext
= IXGBE_READ_REG(hw
, IXGBE_CTRL_EXT
);
4719 ctrl_ext
|= IXGBE_CTRL_EXT_PFRSTD
;
4720 IXGBE_WRITE_REG(hw
, IXGBE_CTRL_EXT
, ctrl_ext
);
4723 void ixgbe_reinit_locked(struct ixgbe_adapter
*adapter
)
4725 WARN_ON(in_interrupt());
4726 /* put off any impending NetWatchDogTimeout */
4727 adapter
->netdev
->trans_start
= jiffies
;
4729 while (test_and_set_bit(__IXGBE_RESETTING
, &adapter
->state
))
4730 usleep_range(1000, 2000);
4731 ixgbe_down(adapter
);
4733 * If SR-IOV enabled then wait a bit before bringing the adapter
4734 * back up to give the VFs time to respond to the reset. The
4735 * two second wait is based upon the watchdog timer cycle in
4738 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
4741 clear_bit(__IXGBE_RESETTING
, &adapter
->state
);
4744 void ixgbe_up(struct ixgbe_adapter
*adapter
)
4746 /* hardware has been reset, we need to reload some things */
4747 ixgbe_configure(adapter
);
4749 ixgbe_up_complete(adapter
);
4752 void ixgbe_reset(struct ixgbe_adapter
*adapter
)
4754 struct ixgbe_hw
*hw
= &adapter
->hw
;
4755 struct net_device
*netdev
= adapter
->netdev
;
4757 u8 old_addr
[ETH_ALEN
];
4759 if (ixgbe_removed(hw
->hw_addr
))
4761 /* lock SFP init bit to prevent race conditions with the watchdog */
4762 while (test_and_set_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
))
4763 usleep_range(1000, 2000);
4765 /* clear all SFP and link config related flags while holding SFP_INIT */
4766 adapter
->flags2
&= ~(IXGBE_FLAG2_SEARCH_FOR_SFP
|
4767 IXGBE_FLAG2_SFP_NEEDS_RESET
);
4768 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_CONFIG
;
4770 err
= hw
->mac
.ops
.init_hw(hw
);
4773 case IXGBE_ERR_SFP_NOT_PRESENT
:
4774 case IXGBE_ERR_SFP_NOT_SUPPORTED
:
4776 case IXGBE_ERR_MASTER_REQUESTS_PENDING
:
4777 e_dev_err("master disable timed out\n");
4779 case IXGBE_ERR_EEPROM_VERSION
:
4780 /* We are running on a pre-production device, log a warning */
4781 e_dev_warn("This device is a pre-production adapter/LOM. "
4782 "Please be aware there may be issues associated with "
4783 "your hardware. If you are experiencing problems "
4784 "please contact your Intel or hardware "
4785 "representative who provided you with this "
4789 e_dev_err("Hardware Error: %d\n", err
);
4792 clear_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
);
4793 /* do not flush user set addresses */
4794 memcpy(old_addr
, &adapter
->mac_table
[0].addr
, netdev
->addr_len
);
4795 ixgbe_flush_sw_mac_table(adapter
);
4796 ixgbe_mac_set_default_filter(adapter
, old_addr
);
4798 /* update SAN MAC vmdq pool selection */
4799 if (hw
->mac
.san_mac_rar_index
)
4800 hw
->mac
.ops
.set_vmdq_san_mac(hw
, VMDQ_P(0));
4802 if (test_bit(__IXGBE_PTP_RUNNING
, &adapter
->state
))
4803 ixgbe_ptp_reset(adapter
);
4807 * ixgbe_clean_tx_ring - Free Tx Buffers
4808 * @tx_ring: ring to be cleaned
4810 static void ixgbe_clean_tx_ring(struct ixgbe_ring
*tx_ring
)
4812 struct ixgbe_tx_buffer
*tx_buffer_info
;
4816 /* ring already cleared, nothing to do */
4817 if (!tx_ring
->tx_buffer_info
)
4820 /* Free all the Tx ring sk_buffs */
4821 for (i
= 0; i
< tx_ring
->count
; i
++) {
4822 tx_buffer_info
= &tx_ring
->tx_buffer_info
[i
];
4823 ixgbe_unmap_and_free_tx_resource(tx_ring
, tx_buffer_info
);
4826 netdev_tx_reset_queue(txring_txq(tx_ring
));
4828 size
= sizeof(struct ixgbe_tx_buffer
) * tx_ring
->count
;
4829 memset(tx_ring
->tx_buffer_info
, 0, size
);
4831 /* Zero out the descriptor ring */
4832 memset(tx_ring
->desc
, 0, tx_ring
->size
);
4834 tx_ring
->next_to_use
= 0;
4835 tx_ring
->next_to_clean
= 0;
4839 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4840 * @adapter: board private structure
4842 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter
*adapter
)
4846 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
4847 ixgbe_clean_rx_ring(adapter
->rx_ring
[i
]);
4851 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4852 * @adapter: board private structure
4854 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter
*adapter
)
4858 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
4859 ixgbe_clean_tx_ring(adapter
->tx_ring
[i
]);
4862 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter
*adapter
)
4864 struct hlist_node
*node2
;
4865 struct ixgbe_fdir_filter
*filter
;
4867 spin_lock(&adapter
->fdir_perfect_lock
);
4869 hlist_for_each_entry_safe(filter
, node2
,
4870 &adapter
->fdir_filter_list
, fdir_node
) {
4871 hlist_del(&filter
->fdir_node
);
4874 adapter
->fdir_filter_count
= 0;
4876 spin_unlock(&adapter
->fdir_perfect_lock
);
4879 void ixgbe_down(struct ixgbe_adapter
*adapter
)
4881 struct net_device
*netdev
= adapter
->netdev
;
4882 struct ixgbe_hw
*hw
= &adapter
->hw
;
4883 struct net_device
*upper
;
4884 struct list_head
*iter
;
4888 /* signal that we are down to the interrupt handler */
4889 if (test_and_set_bit(__IXGBE_DOWN
, &adapter
->state
))
4890 return; /* do nothing if already down */
4892 /* disable receives */
4893 rxctrl
= IXGBE_READ_REG(hw
, IXGBE_RXCTRL
);
4894 IXGBE_WRITE_REG(hw
, IXGBE_RXCTRL
, rxctrl
& ~IXGBE_RXCTRL_RXEN
);
4896 /* disable all enabled rx queues */
4897 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
4898 /* this call also flushes the previous write */
4899 ixgbe_disable_rx_queue(adapter
, adapter
->rx_ring
[i
]);
4901 usleep_range(10000, 20000);
4903 netif_tx_stop_all_queues(netdev
);
4905 /* call carrier off first to avoid false dev_watchdog timeouts */
4906 netif_carrier_off(netdev
);
4907 netif_tx_disable(netdev
);
4909 /* disable any upper devices */
4910 netdev_for_each_all_upper_dev_rcu(adapter
->netdev
, upper
, iter
) {
4911 if (netif_is_macvlan(upper
)) {
4912 struct macvlan_dev
*vlan
= netdev_priv(upper
);
4914 if (vlan
->fwd_priv
) {
4915 netif_tx_stop_all_queues(upper
);
4916 netif_carrier_off(upper
);
4917 netif_tx_disable(upper
);
4922 ixgbe_irq_disable(adapter
);
4924 ixgbe_napi_disable_all(adapter
);
4926 adapter
->flags2
&= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT
|
4927 IXGBE_FLAG2_RESET_REQUESTED
);
4928 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_UPDATE
;
4930 del_timer_sync(&adapter
->service_timer
);
4932 if (adapter
->num_vfs
) {
4933 /* Clear EITR Select mapping */
4934 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EITRSEL
, 0);
4936 /* Mark all the VFs as inactive */
4937 for (i
= 0 ; i
< adapter
->num_vfs
; i
++)
4938 adapter
->vfinfo
[i
].clear_to_send
= false;
4940 /* ping all the active vfs to let them know we are going down */
4941 ixgbe_ping_all_vfs(adapter
);
4943 /* Disable all VFTE/VFRE TX/RX */
4944 ixgbe_disable_tx_rx(adapter
);
4947 /* disable transmits in the hardware now that interrupts are off */
4948 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
4949 u8 reg_idx
= adapter
->tx_ring
[i
]->reg_idx
;
4950 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(reg_idx
), IXGBE_TXDCTL_SWFLSH
);
4953 /* Disable the Tx DMA engine on 82599 and X540 */
4954 switch (hw
->mac
.type
) {
4955 case ixgbe_mac_82599EB
:
4956 case ixgbe_mac_X540
:
4957 IXGBE_WRITE_REG(hw
, IXGBE_DMATXCTL
,
4958 (IXGBE_READ_REG(hw
, IXGBE_DMATXCTL
) &
4959 ~IXGBE_DMATXCTL_TE
));
4965 if (!pci_channel_offline(adapter
->pdev
))
4966 ixgbe_reset(adapter
);
4968 /* power down the optics for 82599 SFP+ fiber */
4969 if (hw
->mac
.ops
.disable_tx_laser
)
4970 hw
->mac
.ops
.disable_tx_laser(hw
);
4972 ixgbe_clean_all_tx_rings(adapter
);
4973 ixgbe_clean_all_rx_rings(adapter
);
4975 #ifdef CONFIG_IXGBE_DCA
4976 /* since we reset the hardware DCA settings were cleared */
4977 ixgbe_setup_dca(adapter
);
4982 * ixgbe_tx_timeout - Respond to a Tx Hang
4983 * @netdev: network interface device structure
4985 static void ixgbe_tx_timeout(struct net_device
*netdev
)
4987 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4989 /* Do the reset outside of interrupt context */
4990 ixgbe_tx_timeout_reset(adapter
);
4994 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4995 * @adapter: board private structure to initialize
4997 * ixgbe_sw_init initializes the Adapter private data structure.
4998 * Fields are initialized based on PCI device information and
4999 * OS network device settings (MTU size).
5001 static int ixgbe_sw_init(struct ixgbe_adapter
*adapter
)
5003 struct ixgbe_hw
*hw
= &adapter
->hw
;
5004 struct pci_dev
*pdev
= adapter
->pdev
;
5005 unsigned int rss
, fdir
;
5007 #ifdef CONFIG_IXGBE_DCB
5009 struct tc_configuration
*tc
;
5012 /* PCI config space info */
5014 hw
->vendor_id
= pdev
->vendor
;
5015 hw
->device_id
= pdev
->device
;
5016 hw
->revision_id
= pdev
->revision
;
5017 hw
->subsystem_vendor_id
= pdev
->subsystem_vendor
;
5018 hw
->subsystem_device_id
= pdev
->subsystem_device
;
5020 /* Set common capability flags and settings */
5021 rss
= min_t(int, IXGBE_MAX_RSS_INDICES
, num_online_cpus());
5022 adapter
->ring_feature
[RING_F_RSS
].limit
= rss
;
5023 adapter
->flags2
|= IXGBE_FLAG2_RSC_CAPABLE
;
5024 adapter
->flags2
|= IXGBE_FLAG2_RSC_ENABLED
;
5025 adapter
->max_q_vectors
= MAX_Q_VECTORS_82599
;
5026 adapter
->atr_sample_rate
= 20;
5027 fdir
= min_t(int, IXGBE_MAX_FDIR_INDICES
, num_online_cpus());
5028 adapter
->ring_feature
[RING_F_FDIR
].limit
= fdir
;
5029 adapter
->fdir_pballoc
= IXGBE_FDIR_PBALLOC_64K
;
5030 #ifdef CONFIG_IXGBE_DCA
5031 adapter
->flags
|= IXGBE_FLAG_DCA_CAPABLE
;
5034 adapter
->flags
|= IXGBE_FLAG_FCOE_CAPABLE
;
5035 adapter
->flags
&= ~IXGBE_FLAG_FCOE_ENABLED
;
5036 #ifdef CONFIG_IXGBE_DCB
5037 /* Default traffic class to use for FCoE */
5038 adapter
->fcoe
.up
= IXGBE_FCOE_DEFTC
;
5039 #endif /* CONFIG_IXGBE_DCB */
5040 #endif /* IXGBE_FCOE */
5042 adapter
->mac_table
= kzalloc(sizeof(struct ixgbe_mac_addr
) *
5043 hw
->mac
.num_rar_entries
,
5046 /* Set MAC specific capability flags and exceptions */
5047 switch (hw
->mac
.type
) {
5048 case ixgbe_mac_82598EB
:
5049 adapter
->flags2
&= ~IXGBE_FLAG2_RSC_CAPABLE
;
5050 adapter
->flags2
&= ~IXGBE_FLAG2_RSC_ENABLED
;
5052 if (hw
->device_id
== IXGBE_DEV_ID_82598AT
)
5053 adapter
->flags
|= IXGBE_FLAG_FAN_FAIL_CAPABLE
;
5055 adapter
->max_q_vectors
= MAX_Q_VECTORS_82598
;
5056 adapter
->ring_feature
[RING_F_FDIR
].limit
= 0;
5057 adapter
->atr_sample_rate
= 0;
5058 adapter
->fdir_pballoc
= 0;
5060 adapter
->flags
&= ~IXGBE_FLAG_FCOE_CAPABLE
;
5061 adapter
->flags
&= ~IXGBE_FLAG_FCOE_ENABLED
;
5062 #ifdef CONFIG_IXGBE_DCB
5063 adapter
->fcoe
.up
= 0;
5064 #endif /* IXGBE_DCB */
5065 #endif /* IXGBE_FCOE */
5067 case ixgbe_mac_82599EB
:
5068 if (hw
->device_id
== IXGBE_DEV_ID_82599_T3_LOM
)
5069 adapter
->flags2
|= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
;
5071 case ixgbe_mac_X540
:
5072 fwsm
= IXGBE_READ_REG(hw
, IXGBE_FWSM
);
5073 if (fwsm
& IXGBE_FWSM_TS_ENABLED
)
5074 adapter
->flags2
|= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
;
5081 /* FCoE support exists, always init the FCoE lock */
5082 spin_lock_init(&adapter
->fcoe
.lock
);
5085 /* n-tuple support exists, always init our spinlock */
5086 spin_lock_init(&adapter
->fdir_perfect_lock
);
5088 #ifdef CONFIG_IXGBE_DCB
5089 switch (hw
->mac
.type
) {
5090 case ixgbe_mac_X540
:
5091 adapter
->dcb_cfg
.num_tcs
.pg_tcs
= X540_TRAFFIC_CLASS
;
5092 adapter
->dcb_cfg
.num_tcs
.pfc_tcs
= X540_TRAFFIC_CLASS
;
5095 adapter
->dcb_cfg
.num_tcs
.pg_tcs
= MAX_TRAFFIC_CLASS
;
5096 adapter
->dcb_cfg
.num_tcs
.pfc_tcs
= MAX_TRAFFIC_CLASS
;
5100 /* Configure DCB traffic classes */
5101 for (j
= 0; j
< MAX_TRAFFIC_CLASS
; j
++) {
5102 tc
= &adapter
->dcb_cfg
.tc_config
[j
];
5103 tc
->path
[DCB_TX_CONFIG
].bwg_id
= 0;
5104 tc
->path
[DCB_TX_CONFIG
].bwg_percent
= 12 + (j
& 1);
5105 tc
->path
[DCB_RX_CONFIG
].bwg_id
= 0;
5106 tc
->path
[DCB_RX_CONFIG
].bwg_percent
= 12 + (j
& 1);
5107 tc
->dcb_pfc
= pfc_disabled
;
5110 /* Initialize default user to priority mapping, UPx->TC0 */
5111 tc
= &adapter
->dcb_cfg
.tc_config
[0];
5112 tc
->path
[DCB_TX_CONFIG
].up_to_tc_bitmap
= 0xFF;
5113 tc
->path
[DCB_RX_CONFIG
].up_to_tc_bitmap
= 0xFF;
5115 adapter
->dcb_cfg
.bw_percentage
[DCB_TX_CONFIG
][0] = 100;
5116 adapter
->dcb_cfg
.bw_percentage
[DCB_RX_CONFIG
][0] = 100;
5117 adapter
->dcb_cfg
.pfc_mode_enable
= false;
5118 adapter
->dcb_set_bitmap
= 0x00;
5119 adapter
->dcbx_cap
= DCB_CAP_DCBX_HOST
| DCB_CAP_DCBX_VER_CEE
;
5120 memcpy(&adapter
->temp_dcb_cfg
, &adapter
->dcb_cfg
,
5121 sizeof(adapter
->temp_dcb_cfg
));
5125 /* default flow control settings */
5126 hw
->fc
.requested_mode
= ixgbe_fc_full
;
5127 hw
->fc
.current_mode
= ixgbe_fc_full
; /* init for ethtool output */
5128 ixgbe_pbthresh_setup(adapter
);
5129 hw
->fc
.pause_time
= IXGBE_DEFAULT_FCPAUSE
;
5130 hw
->fc
.send_xon
= true;
5131 hw
->fc
.disable_fc_autoneg
= ixgbe_device_supports_autoneg_fc(hw
);
5133 #ifdef CONFIG_PCI_IOV
5135 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
5137 /* assign number of SR-IOV VFs */
5138 if (hw
->mac
.type
!= ixgbe_mac_82598EB
) {
5139 if (max_vfs
> IXGBE_MAX_VFS_DRV_LIMIT
) {
5140 adapter
->num_vfs
= 0;
5141 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
5143 adapter
->num_vfs
= max_vfs
;
5146 #endif /* CONFIG_PCI_IOV */
5148 /* enable itr by default in dynamic mode */
5149 adapter
->rx_itr_setting
= 1;
5150 adapter
->tx_itr_setting
= 1;
5152 /* set default ring sizes */
5153 adapter
->tx_ring_count
= IXGBE_DEFAULT_TXD
;
5154 adapter
->rx_ring_count
= IXGBE_DEFAULT_RXD
;
5156 /* set default work limits */
5157 adapter
->tx_work_limit
= IXGBE_DEFAULT_TX_WORK
;
5159 /* initialize eeprom parameters */
5160 if (ixgbe_init_eeprom_params_generic(hw
)) {
5161 e_dev_err("EEPROM initialization failed\n");
5165 /* PF holds first pool slot */
5166 set_bit(0, &adapter
->fwd_bitmask
);
5167 set_bit(__IXGBE_DOWN
, &adapter
->state
);
5173 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5174 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5176 * Return 0 on success, negative on failure
5178 int ixgbe_setup_tx_resources(struct ixgbe_ring
*tx_ring
)
5180 struct device
*dev
= tx_ring
->dev
;
5181 int orig_node
= dev_to_node(dev
);
5185 size
= sizeof(struct ixgbe_tx_buffer
) * tx_ring
->count
;
5187 if (tx_ring
->q_vector
)
5188 ring_node
= tx_ring
->q_vector
->numa_node
;
5190 tx_ring
->tx_buffer_info
= vzalloc_node(size
, ring_node
);
5191 if (!tx_ring
->tx_buffer_info
)
5192 tx_ring
->tx_buffer_info
= vzalloc(size
);
5193 if (!tx_ring
->tx_buffer_info
)
5196 u64_stats_init(&tx_ring
->syncp
);
5198 /* round up to nearest 4K */
5199 tx_ring
->size
= tx_ring
->count
* sizeof(union ixgbe_adv_tx_desc
);
5200 tx_ring
->size
= ALIGN(tx_ring
->size
, 4096);
5202 set_dev_node(dev
, ring_node
);
5203 tx_ring
->desc
= dma_alloc_coherent(dev
,
5207 set_dev_node(dev
, orig_node
);
5209 tx_ring
->desc
= dma_alloc_coherent(dev
, tx_ring
->size
,
5210 &tx_ring
->dma
, GFP_KERNEL
);
5214 tx_ring
->next_to_use
= 0;
5215 tx_ring
->next_to_clean
= 0;
5219 vfree(tx_ring
->tx_buffer_info
);
5220 tx_ring
->tx_buffer_info
= NULL
;
5221 dev_err(dev
, "Unable to allocate memory for the Tx descriptor ring\n");
5226 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5227 * @adapter: board private structure
5229 * If this function returns with an error, then it's possible one or
5230 * more of the rings is populated (while the rest are not). It is the
5231 * callers duty to clean those orphaned rings.
5233 * Return 0 on success, negative on failure
5235 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter
*adapter
)
5239 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
5240 err
= ixgbe_setup_tx_resources(adapter
->tx_ring
[i
]);
5244 e_err(probe
, "Allocation for Tx Queue %u failed\n", i
);
5250 /* rewind the index freeing the rings as we go */
5252 ixgbe_free_tx_resources(adapter
->tx_ring
[i
]);
5257 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5258 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5260 * Returns 0 on success, negative on failure
5262 int ixgbe_setup_rx_resources(struct ixgbe_ring
*rx_ring
)
5264 struct device
*dev
= rx_ring
->dev
;
5265 int orig_node
= dev_to_node(dev
);
5269 size
= sizeof(struct ixgbe_rx_buffer
) * rx_ring
->count
;
5271 if (rx_ring
->q_vector
)
5272 ring_node
= rx_ring
->q_vector
->numa_node
;
5274 rx_ring
->rx_buffer_info
= vzalloc_node(size
, ring_node
);
5275 if (!rx_ring
->rx_buffer_info
)
5276 rx_ring
->rx_buffer_info
= vzalloc(size
);
5277 if (!rx_ring
->rx_buffer_info
)
5280 u64_stats_init(&rx_ring
->syncp
);
5282 /* Round up to nearest 4K */
5283 rx_ring
->size
= rx_ring
->count
* sizeof(union ixgbe_adv_rx_desc
);
5284 rx_ring
->size
= ALIGN(rx_ring
->size
, 4096);
5286 set_dev_node(dev
, ring_node
);
5287 rx_ring
->desc
= dma_alloc_coherent(dev
,
5291 set_dev_node(dev
, orig_node
);
5293 rx_ring
->desc
= dma_alloc_coherent(dev
, rx_ring
->size
,
5294 &rx_ring
->dma
, GFP_KERNEL
);
5298 rx_ring
->next_to_clean
= 0;
5299 rx_ring
->next_to_use
= 0;
5303 vfree(rx_ring
->rx_buffer_info
);
5304 rx_ring
->rx_buffer_info
= NULL
;
5305 dev_err(dev
, "Unable to allocate memory for the Rx descriptor ring\n");
5310 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5311 * @adapter: board private structure
5313 * If this function returns with an error, then it's possible one or
5314 * more of the rings is populated (while the rest are not). It is the
5315 * callers duty to clean those orphaned rings.
5317 * Return 0 on success, negative on failure
5319 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter
*adapter
)
5323 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
5324 err
= ixgbe_setup_rx_resources(adapter
->rx_ring
[i
]);
5328 e_err(probe
, "Allocation for Rx Queue %u failed\n", i
);
5333 err
= ixgbe_setup_fcoe_ddp_resources(adapter
);
5338 /* rewind the index freeing the rings as we go */
5340 ixgbe_free_rx_resources(adapter
->rx_ring
[i
]);
5345 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5346 * @tx_ring: Tx descriptor ring for a specific queue
5348 * Free all transmit software resources
5350 void ixgbe_free_tx_resources(struct ixgbe_ring
*tx_ring
)
5352 ixgbe_clean_tx_ring(tx_ring
);
5354 vfree(tx_ring
->tx_buffer_info
);
5355 tx_ring
->tx_buffer_info
= NULL
;
5357 /* if not set, then don't free */
5361 dma_free_coherent(tx_ring
->dev
, tx_ring
->size
,
5362 tx_ring
->desc
, tx_ring
->dma
);
5364 tx_ring
->desc
= NULL
;
5368 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5369 * @adapter: board private structure
5371 * Free all transmit software resources
5373 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter
*adapter
)
5377 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
5378 if (adapter
->tx_ring
[i
]->desc
)
5379 ixgbe_free_tx_resources(adapter
->tx_ring
[i
]);
5383 * ixgbe_free_rx_resources - Free Rx Resources
5384 * @rx_ring: ring to clean the resources from
5386 * Free all receive software resources
5388 void ixgbe_free_rx_resources(struct ixgbe_ring
*rx_ring
)
5390 ixgbe_clean_rx_ring(rx_ring
);
5392 vfree(rx_ring
->rx_buffer_info
);
5393 rx_ring
->rx_buffer_info
= NULL
;
5395 /* if not set, then don't free */
5399 dma_free_coherent(rx_ring
->dev
, rx_ring
->size
,
5400 rx_ring
->desc
, rx_ring
->dma
);
5402 rx_ring
->desc
= NULL
;
5406 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5407 * @adapter: board private structure
5409 * Free all receive software resources
5411 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter
*adapter
)
5416 ixgbe_free_fcoe_ddp_resources(adapter
);
5419 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
5420 if (adapter
->rx_ring
[i
]->desc
)
5421 ixgbe_free_rx_resources(adapter
->rx_ring
[i
]);
5425 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5426 * @netdev: network interface device structure
5427 * @new_mtu: new value for maximum frame size
5429 * Returns 0 on success, negative on failure
5431 static int ixgbe_change_mtu(struct net_device
*netdev
, int new_mtu
)
5433 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5434 int max_frame
= new_mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
5436 /* MTU < 68 is an error and causes problems on some kernels */
5437 if ((new_mtu
< 68) || (max_frame
> IXGBE_MAX_JUMBO_FRAME_SIZE
))
5441 * For 82599EB we cannot allow legacy VFs to enable their receive
5442 * paths when MTU greater than 1500 is configured. So display a
5443 * warning that legacy VFs will be disabled.
5445 if ((adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) &&
5446 (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
) &&
5447 (max_frame
> (ETH_FRAME_LEN
+ ETH_FCS_LEN
)))
5448 e_warn(probe
, "Setting MTU > 1500 will disable legacy VFs\n");
5450 e_info(probe
, "changing MTU from %d to %d\n", netdev
->mtu
, new_mtu
);
5452 /* must set new MTU before calling down or up */
5453 netdev
->mtu
= new_mtu
;
5455 if (netif_running(netdev
))
5456 ixgbe_reinit_locked(adapter
);
5462 * ixgbe_open - Called when a network interface is made active
5463 * @netdev: network interface device structure
5465 * Returns 0 on success, negative value on failure
5467 * The open entry point is called when a network interface is made
5468 * active by the system (IFF_UP). At this point all resources needed
5469 * for transmit and receive operations are allocated, the interrupt
5470 * handler is registered with the OS, the watchdog timer is started,
5471 * and the stack is notified that the interface is ready.
5473 static int ixgbe_open(struct net_device
*netdev
)
5475 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5478 /* disallow open during test */
5479 if (test_bit(__IXGBE_TESTING
, &adapter
->state
))
5482 netif_carrier_off(netdev
);
5484 /* allocate transmit descriptors */
5485 err
= ixgbe_setup_all_tx_resources(adapter
);
5489 /* allocate receive descriptors */
5490 err
= ixgbe_setup_all_rx_resources(adapter
);
5494 ixgbe_configure(adapter
);
5496 err
= ixgbe_request_irq(adapter
);
5500 /* Notify the stack of the actual queue counts. */
5501 if (adapter
->num_rx_pools
> 1)
5502 queues
= adapter
->num_rx_queues_per_pool
;
5504 queues
= adapter
->num_tx_queues
;
5506 err
= netif_set_real_num_tx_queues(netdev
, queues
);
5508 goto err_set_queues
;
5510 if (adapter
->num_rx_pools
> 1 &&
5511 adapter
->num_rx_queues
> IXGBE_MAX_L2A_QUEUES
)
5512 queues
= IXGBE_MAX_L2A_QUEUES
;
5514 queues
= adapter
->num_rx_queues
;
5515 err
= netif_set_real_num_rx_queues(netdev
, queues
);
5517 goto err_set_queues
;
5519 ixgbe_ptp_init(adapter
);
5521 ixgbe_up_complete(adapter
);
5526 ixgbe_free_irq(adapter
);
5528 ixgbe_free_all_rx_resources(adapter
);
5530 ixgbe_free_all_tx_resources(adapter
);
5532 ixgbe_reset(adapter
);
5537 static void ixgbe_close_suspend(struct ixgbe_adapter
*adapter
)
5539 ixgbe_ptp_suspend(adapter
);
5541 ixgbe_down(adapter
);
5542 ixgbe_free_irq(adapter
);
5544 ixgbe_free_all_tx_resources(adapter
);
5545 ixgbe_free_all_rx_resources(adapter
);
5549 * ixgbe_close - Disables a network interface
5550 * @netdev: network interface device structure
5552 * Returns 0, this is not allowed to fail
5554 * The close entry point is called when an interface is de-activated
5555 * by the OS. The hardware is still under the drivers control, but
5556 * needs to be disabled. A global MAC reset is issued to stop the
5557 * hardware, and all transmit and receive resources are freed.
5559 static int ixgbe_close(struct net_device
*netdev
)
5561 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5563 ixgbe_ptp_stop(adapter
);
5565 ixgbe_close_suspend(adapter
);
5567 ixgbe_fdir_filter_exit(adapter
);
5569 ixgbe_release_hw_control(adapter
);
5575 static int ixgbe_resume(struct pci_dev
*pdev
)
5577 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
5578 struct net_device
*netdev
= adapter
->netdev
;
5581 adapter
->hw
.hw_addr
= adapter
->io_addr
;
5582 pci_set_power_state(pdev
, PCI_D0
);
5583 pci_restore_state(pdev
);
5585 * pci_restore_state clears dev->state_saved so call
5586 * pci_save_state to restore it.
5588 pci_save_state(pdev
);
5590 err
= pci_enable_device_mem(pdev
);
5592 e_dev_err("Cannot enable PCI device from suspend\n");
5595 smp_mb__before_atomic();
5596 clear_bit(__IXGBE_DISABLED
, &adapter
->state
);
5597 pci_set_master(pdev
);
5599 pci_wake_from_d3(pdev
, false);
5601 ixgbe_reset(adapter
);
5603 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_WUS
, ~0);
5606 err
= ixgbe_init_interrupt_scheme(adapter
);
5607 if (!err
&& netif_running(netdev
))
5608 err
= ixgbe_open(netdev
);
5615 netif_device_attach(netdev
);
5619 #endif /* CONFIG_PM */
5621 static int __ixgbe_shutdown(struct pci_dev
*pdev
, bool *enable_wake
)
5623 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
5624 struct net_device
*netdev
= adapter
->netdev
;
5625 struct ixgbe_hw
*hw
= &adapter
->hw
;
5627 u32 wufc
= adapter
->wol
;
5632 netif_device_detach(netdev
);
5635 if (netif_running(netdev
))
5636 ixgbe_close_suspend(adapter
);
5639 ixgbe_clear_interrupt_scheme(adapter
);
5642 retval
= pci_save_state(pdev
);
5647 if (hw
->mac
.ops
.stop_link_on_d3
)
5648 hw
->mac
.ops
.stop_link_on_d3(hw
);
5651 ixgbe_set_rx_mode(netdev
);
5653 /* enable the optics for 82599 SFP+ fiber as we can WoL */
5654 if (hw
->mac
.ops
.enable_tx_laser
)
5655 hw
->mac
.ops
.enable_tx_laser(hw
);
5657 /* turn on all-multi mode if wake on multicast is enabled */
5658 if (wufc
& IXGBE_WUFC_MC
) {
5659 fctrl
= IXGBE_READ_REG(hw
, IXGBE_FCTRL
);
5660 fctrl
|= IXGBE_FCTRL_MPE
;
5661 IXGBE_WRITE_REG(hw
, IXGBE_FCTRL
, fctrl
);
5664 ctrl
= IXGBE_READ_REG(hw
, IXGBE_CTRL
);
5665 ctrl
|= IXGBE_CTRL_GIO_DIS
;
5666 IXGBE_WRITE_REG(hw
, IXGBE_CTRL
, ctrl
);
5668 IXGBE_WRITE_REG(hw
, IXGBE_WUFC
, wufc
);
5670 IXGBE_WRITE_REG(hw
, IXGBE_WUC
, 0);
5671 IXGBE_WRITE_REG(hw
, IXGBE_WUFC
, 0);
5674 switch (hw
->mac
.type
) {
5675 case ixgbe_mac_82598EB
:
5676 pci_wake_from_d3(pdev
, false);
5678 case ixgbe_mac_82599EB
:
5679 case ixgbe_mac_X540
:
5680 pci_wake_from_d3(pdev
, !!wufc
);
5686 *enable_wake
= !!wufc
;
5688 ixgbe_release_hw_control(adapter
);
5690 if (!test_and_set_bit(__IXGBE_DISABLED
, &adapter
->state
))
5691 pci_disable_device(pdev
);
5697 static int ixgbe_suspend(struct pci_dev
*pdev
, pm_message_t state
)
5702 retval
= __ixgbe_shutdown(pdev
, &wake
);
5707 pci_prepare_to_sleep(pdev
);
5709 pci_wake_from_d3(pdev
, false);
5710 pci_set_power_state(pdev
, PCI_D3hot
);
5715 #endif /* CONFIG_PM */
5717 static void ixgbe_shutdown(struct pci_dev
*pdev
)
5721 __ixgbe_shutdown(pdev
, &wake
);
5723 if (system_state
== SYSTEM_POWER_OFF
) {
5724 pci_wake_from_d3(pdev
, wake
);
5725 pci_set_power_state(pdev
, PCI_D3hot
);
5730 * ixgbe_update_stats - Update the board statistics counters.
5731 * @adapter: board private structure
5733 void ixgbe_update_stats(struct ixgbe_adapter
*adapter
)
5735 struct net_device
*netdev
= adapter
->netdev
;
5736 struct ixgbe_hw
*hw
= &adapter
->hw
;
5737 struct ixgbe_hw_stats
*hwstats
= &adapter
->stats
;
5739 u32 i
, missed_rx
= 0, mpc
, bprc
, lxon
, lxoff
, xon_off_tot
;
5740 u64 non_eop_descs
= 0, restart_queue
= 0, tx_busy
= 0;
5741 u64 alloc_rx_page_failed
= 0, alloc_rx_buff_failed
= 0;
5742 u64 bytes
= 0, packets
= 0, hw_csum_rx_error
= 0;
5744 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
5745 test_bit(__IXGBE_RESETTING
, &adapter
->state
))
5748 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
) {
5751 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
5752 rsc_count
+= adapter
->rx_ring
[i
]->rx_stats
.rsc_count
;
5753 rsc_flush
+= adapter
->rx_ring
[i
]->rx_stats
.rsc_flush
;
5755 adapter
->rsc_total_count
= rsc_count
;
5756 adapter
->rsc_total_flush
= rsc_flush
;
5759 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
5760 struct ixgbe_ring
*rx_ring
= adapter
->rx_ring
[i
];
5761 non_eop_descs
+= rx_ring
->rx_stats
.non_eop_descs
;
5762 alloc_rx_page_failed
+= rx_ring
->rx_stats
.alloc_rx_page_failed
;
5763 alloc_rx_buff_failed
+= rx_ring
->rx_stats
.alloc_rx_buff_failed
;
5764 hw_csum_rx_error
+= rx_ring
->rx_stats
.csum_err
;
5765 bytes
+= rx_ring
->stats
.bytes
;
5766 packets
+= rx_ring
->stats
.packets
;
5768 adapter
->non_eop_descs
= non_eop_descs
;
5769 adapter
->alloc_rx_page_failed
= alloc_rx_page_failed
;
5770 adapter
->alloc_rx_buff_failed
= alloc_rx_buff_failed
;
5771 adapter
->hw_csum_rx_error
= hw_csum_rx_error
;
5772 netdev
->stats
.rx_bytes
= bytes
;
5773 netdev
->stats
.rx_packets
= packets
;
5777 /* gather some stats to the adapter struct that are per queue */
5778 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
5779 struct ixgbe_ring
*tx_ring
= adapter
->tx_ring
[i
];
5780 restart_queue
+= tx_ring
->tx_stats
.restart_queue
;
5781 tx_busy
+= tx_ring
->tx_stats
.tx_busy
;
5782 bytes
+= tx_ring
->stats
.bytes
;
5783 packets
+= tx_ring
->stats
.packets
;
5785 adapter
->restart_queue
= restart_queue
;
5786 adapter
->tx_busy
= tx_busy
;
5787 netdev
->stats
.tx_bytes
= bytes
;
5788 netdev
->stats
.tx_packets
= packets
;
5790 hwstats
->crcerrs
+= IXGBE_READ_REG(hw
, IXGBE_CRCERRS
);
5792 /* 8 register reads */
5793 for (i
= 0; i
< 8; i
++) {
5794 /* for packet buffers not used, the register should read 0 */
5795 mpc
= IXGBE_READ_REG(hw
, IXGBE_MPC(i
));
5797 hwstats
->mpc
[i
] += mpc
;
5798 total_mpc
+= hwstats
->mpc
[i
];
5799 hwstats
->pxontxc
[i
] += IXGBE_READ_REG(hw
, IXGBE_PXONTXC(i
));
5800 hwstats
->pxofftxc
[i
] += IXGBE_READ_REG(hw
, IXGBE_PXOFFTXC(i
));
5801 switch (hw
->mac
.type
) {
5802 case ixgbe_mac_82598EB
:
5803 hwstats
->rnbc
[i
] += IXGBE_READ_REG(hw
, IXGBE_RNBC(i
));
5804 hwstats
->qbtc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBTC(i
));
5805 hwstats
->qbrc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBRC(i
));
5806 hwstats
->pxonrxc
[i
] +=
5807 IXGBE_READ_REG(hw
, IXGBE_PXONRXC(i
));
5809 case ixgbe_mac_82599EB
:
5810 case ixgbe_mac_X540
:
5811 hwstats
->pxonrxc
[i
] +=
5812 IXGBE_READ_REG(hw
, IXGBE_PXONRXCNT(i
));
5819 /*16 register reads */
5820 for (i
= 0; i
< 16; i
++) {
5821 hwstats
->qptc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QPTC(i
));
5822 hwstats
->qprc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QPRC(i
));
5823 if ((hw
->mac
.type
== ixgbe_mac_82599EB
) ||
5824 (hw
->mac
.type
== ixgbe_mac_X540
)) {
5825 hwstats
->qbtc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBTC_L(i
));
5826 IXGBE_READ_REG(hw
, IXGBE_QBTC_H(i
)); /* to clear */
5827 hwstats
->qbrc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBRC_L(i
));
5828 IXGBE_READ_REG(hw
, IXGBE_QBRC_H(i
)); /* to clear */
5832 hwstats
->gprc
+= IXGBE_READ_REG(hw
, IXGBE_GPRC
);
5833 /* work around hardware counting issue */
5834 hwstats
->gprc
-= missed_rx
;
5836 ixgbe_update_xoff_received(adapter
);
5838 /* 82598 hardware only has a 32 bit counter in the high register */
5839 switch (hw
->mac
.type
) {
5840 case ixgbe_mac_82598EB
:
5841 hwstats
->lxonrxc
+= IXGBE_READ_REG(hw
, IXGBE_LXONRXC
);
5842 hwstats
->gorc
+= IXGBE_READ_REG(hw
, IXGBE_GORCH
);
5843 hwstats
->gotc
+= IXGBE_READ_REG(hw
, IXGBE_GOTCH
);
5844 hwstats
->tor
+= IXGBE_READ_REG(hw
, IXGBE_TORH
);
5846 case ixgbe_mac_X540
:
5847 /* OS2BMC stats are X540 only*/
5848 hwstats
->o2bgptc
+= IXGBE_READ_REG(hw
, IXGBE_O2BGPTC
);
5849 hwstats
->o2bspc
+= IXGBE_READ_REG(hw
, IXGBE_O2BSPC
);
5850 hwstats
->b2ospc
+= IXGBE_READ_REG(hw
, IXGBE_B2OSPC
);
5851 hwstats
->b2ogprc
+= IXGBE_READ_REG(hw
, IXGBE_B2OGPRC
);
5852 case ixgbe_mac_82599EB
:
5853 for (i
= 0; i
< 16; i
++)
5854 adapter
->hw_rx_no_dma_resources
+=
5855 IXGBE_READ_REG(hw
, IXGBE_QPRDC(i
));
5856 hwstats
->gorc
+= IXGBE_READ_REG(hw
, IXGBE_GORCL
);
5857 IXGBE_READ_REG(hw
, IXGBE_GORCH
); /* to clear */
5858 hwstats
->gotc
+= IXGBE_READ_REG(hw
, IXGBE_GOTCL
);
5859 IXGBE_READ_REG(hw
, IXGBE_GOTCH
); /* to clear */
5860 hwstats
->tor
+= IXGBE_READ_REG(hw
, IXGBE_TORL
);
5861 IXGBE_READ_REG(hw
, IXGBE_TORH
); /* to clear */
5862 hwstats
->lxonrxc
+= IXGBE_READ_REG(hw
, IXGBE_LXONRXCNT
);
5863 hwstats
->fdirmatch
+= IXGBE_READ_REG(hw
, IXGBE_FDIRMATCH
);
5864 hwstats
->fdirmiss
+= IXGBE_READ_REG(hw
, IXGBE_FDIRMISS
);
5866 hwstats
->fccrc
+= IXGBE_READ_REG(hw
, IXGBE_FCCRC
);
5867 hwstats
->fcoerpdc
+= IXGBE_READ_REG(hw
, IXGBE_FCOERPDC
);
5868 hwstats
->fcoeprc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEPRC
);
5869 hwstats
->fcoeptc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEPTC
);
5870 hwstats
->fcoedwrc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEDWRC
);
5871 hwstats
->fcoedwtc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEDWTC
);
5872 /* Add up per cpu counters for total ddp aloc fail */
5873 if (adapter
->fcoe
.ddp_pool
) {
5874 struct ixgbe_fcoe
*fcoe
= &adapter
->fcoe
;
5875 struct ixgbe_fcoe_ddp_pool
*ddp_pool
;
5877 u64 noddp
= 0, noddp_ext_buff
= 0;
5878 for_each_possible_cpu(cpu
) {
5879 ddp_pool
= per_cpu_ptr(fcoe
->ddp_pool
, cpu
);
5880 noddp
+= ddp_pool
->noddp
;
5881 noddp_ext_buff
+= ddp_pool
->noddp_ext_buff
;
5883 hwstats
->fcoe_noddp
= noddp
;
5884 hwstats
->fcoe_noddp_ext_buff
= noddp_ext_buff
;
5886 #endif /* IXGBE_FCOE */
5891 bprc
= IXGBE_READ_REG(hw
, IXGBE_BPRC
);
5892 hwstats
->bprc
+= bprc
;
5893 hwstats
->mprc
+= IXGBE_READ_REG(hw
, IXGBE_MPRC
);
5894 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
5895 hwstats
->mprc
-= bprc
;
5896 hwstats
->roc
+= IXGBE_READ_REG(hw
, IXGBE_ROC
);
5897 hwstats
->prc64
+= IXGBE_READ_REG(hw
, IXGBE_PRC64
);
5898 hwstats
->prc127
+= IXGBE_READ_REG(hw
, IXGBE_PRC127
);
5899 hwstats
->prc255
+= IXGBE_READ_REG(hw
, IXGBE_PRC255
);
5900 hwstats
->prc511
+= IXGBE_READ_REG(hw
, IXGBE_PRC511
);
5901 hwstats
->prc1023
+= IXGBE_READ_REG(hw
, IXGBE_PRC1023
);
5902 hwstats
->prc1522
+= IXGBE_READ_REG(hw
, IXGBE_PRC1522
);
5903 hwstats
->rlec
+= IXGBE_READ_REG(hw
, IXGBE_RLEC
);
5904 lxon
= IXGBE_READ_REG(hw
, IXGBE_LXONTXC
);
5905 hwstats
->lxontxc
+= lxon
;
5906 lxoff
= IXGBE_READ_REG(hw
, IXGBE_LXOFFTXC
);
5907 hwstats
->lxofftxc
+= lxoff
;
5908 hwstats
->gptc
+= IXGBE_READ_REG(hw
, IXGBE_GPTC
);
5909 hwstats
->mptc
+= IXGBE_READ_REG(hw
, IXGBE_MPTC
);
5911 * 82598 errata - tx of flow control packets is included in tx counters
5913 xon_off_tot
= lxon
+ lxoff
;
5914 hwstats
->gptc
-= xon_off_tot
;
5915 hwstats
->mptc
-= xon_off_tot
;
5916 hwstats
->gotc
-= (xon_off_tot
* (ETH_ZLEN
+ ETH_FCS_LEN
));
5917 hwstats
->ruc
+= IXGBE_READ_REG(hw
, IXGBE_RUC
);
5918 hwstats
->rfc
+= IXGBE_READ_REG(hw
, IXGBE_RFC
);
5919 hwstats
->rjc
+= IXGBE_READ_REG(hw
, IXGBE_RJC
);
5920 hwstats
->tpr
+= IXGBE_READ_REG(hw
, IXGBE_TPR
);
5921 hwstats
->ptc64
+= IXGBE_READ_REG(hw
, IXGBE_PTC64
);
5922 hwstats
->ptc64
-= xon_off_tot
;
5923 hwstats
->ptc127
+= IXGBE_READ_REG(hw
, IXGBE_PTC127
);
5924 hwstats
->ptc255
+= IXGBE_READ_REG(hw
, IXGBE_PTC255
);
5925 hwstats
->ptc511
+= IXGBE_READ_REG(hw
, IXGBE_PTC511
);
5926 hwstats
->ptc1023
+= IXGBE_READ_REG(hw
, IXGBE_PTC1023
);
5927 hwstats
->ptc1522
+= IXGBE_READ_REG(hw
, IXGBE_PTC1522
);
5928 hwstats
->bptc
+= IXGBE_READ_REG(hw
, IXGBE_BPTC
);
5930 /* Fill out the OS statistics structure */
5931 netdev
->stats
.multicast
= hwstats
->mprc
;
5934 netdev
->stats
.rx_errors
= hwstats
->crcerrs
+ hwstats
->rlec
;
5935 netdev
->stats
.rx_dropped
= 0;
5936 netdev
->stats
.rx_length_errors
= hwstats
->rlec
;
5937 netdev
->stats
.rx_crc_errors
= hwstats
->crcerrs
;
5938 netdev
->stats
.rx_missed_errors
= total_mpc
;
5942 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5943 * @adapter: pointer to the device adapter structure
5945 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter
*adapter
)
5947 struct ixgbe_hw
*hw
= &adapter
->hw
;
5950 if (!(adapter
->flags2
& IXGBE_FLAG2_FDIR_REQUIRES_REINIT
))
5953 adapter
->flags2
&= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT
;
5955 /* if interface is down do nothing */
5956 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
5959 /* do nothing if we are not using signature filters */
5960 if (!(adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
))
5963 adapter
->fdir_overflow
++;
5965 if (ixgbe_reinit_fdir_tables_82599(hw
) == 0) {
5966 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
5967 set_bit(__IXGBE_TX_FDIR_INIT_DONE
,
5968 &(adapter
->tx_ring
[i
]->state
));
5969 /* re-enable flow director interrupts */
5970 IXGBE_WRITE_REG(hw
, IXGBE_EIMS
, IXGBE_EIMS_FLOW_DIR
);
5972 e_err(probe
, "failed to finish FDIR re-initialization, "
5973 "ignored adding FDIR ATR filters\n");
5978 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5979 * @adapter: pointer to the device adapter structure
5981 * This function serves two purposes. First it strobes the interrupt lines
5982 * in order to make certain interrupts are occurring. Secondly it sets the
5983 * bits needed to check for TX hangs. As a result we should immediately
5984 * determine if a hang has occurred.
5986 static void ixgbe_check_hang_subtask(struct ixgbe_adapter
*adapter
)
5988 struct ixgbe_hw
*hw
= &adapter
->hw
;
5992 /* If we're down, removing or resetting, just bail */
5993 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
5994 test_bit(__IXGBE_REMOVING
, &adapter
->state
) ||
5995 test_bit(__IXGBE_RESETTING
, &adapter
->state
))
5998 /* Force detection of hung controller */
5999 if (netif_carrier_ok(adapter
->netdev
)) {
6000 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
6001 set_check_for_tx_hang(adapter
->tx_ring
[i
]);
6004 if (!(adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)) {
6006 * for legacy and MSI interrupts don't set any bits
6007 * that are enabled for EIAM, because this operation
6008 * would set *both* EIMS and EICS for any bit in EIAM
6010 IXGBE_WRITE_REG(hw
, IXGBE_EICS
,
6011 (IXGBE_EICS_TCP_TIMER
| IXGBE_EICS_OTHER
));
6013 /* get one bit for every active tx/rx interrupt vector */
6014 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
6015 struct ixgbe_q_vector
*qv
= adapter
->q_vector
[i
];
6016 if (qv
->rx
.ring
|| qv
->tx
.ring
)
6017 eics
|= ((u64
)1 << i
);
6021 /* Cause software interrupt to ensure rings are cleaned */
6022 ixgbe_irq_rearm_queues(adapter
, eics
);
6027 * ixgbe_watchdog_update_link - update the link status
6028 * @adapter: pointer to the device adapter structure
6029 * @link_speed: pointer to a u32 to store the link_speed
6031 static void ixgbe_watchdog_update_link(struct ixgbe_adapter
*adapter
)
6033 struct ixgbe_hw
*hw
= &adapter
->hw
;
6034 u32 link_speed
= adapter
->link_speed
;
6035 bool link_up
= adapter
->link_up
;
6036 bool pfc_en
= adapter
->dcb_cfg
.pfc_mode_enable
;
6038 if (!(adapter
->flags
& IXGBE_FLAG_NEED_LINK_UPDATE
))
6041 if (hw
->mac
.ops
.check_link
) {
6042 hw
->mac
.ops
.check_link(hw
, &link_speed
, &link_up
, false);
6044 /* always assume link is up, if no check link function */
6045 link_speed
= IXGBE_LINK_SPEED_10GB_FULL
;
6049 if (adapter
->ixgbe_ieee_pfc
)
6050 pfc_en
|= !!(adapter
->ixgbe_ieee_pfc
->pfc_en
);
6052 if (link_up
&& !((adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) && pfc_en
)) {
6053 hw
->mac
.ops
.fc_enable(hw
);
6054 ixgbe_set_rx_drop_en(adapter
);
6058 time_after(jiffies
, (adapter
->link_check_timeout
+
6059 IXGBE_TRY_LINK_TIMEOUT
))) {
6060 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_UPDATE
;
6061 IXGBE_WRITE_REG(hw
, IXGBE_EIMS
, IXGBE_EIMC_LSC
);
6062 IXGBE_WRITE_FLUSH(hw
);
6065 adapter
->link_up
= link_up
;
6066 adapter
->link_speed
= link_speed
;
6069 static void ixgbe_update_default_up(struct ixgbe_adapter
*adapter
)
6071 #ifdef CONFIG_IXGBE_DCB
6072 struct net_device
*netdev
= adapter
->netdev
;
6073 struct dcb_app app
= {
6074 .selector
= IEEE_8021QAZ_APP_SEL_ETHERTYPE
,
6079 if (adapter
->dcbx_cap
& DCB_CAP_DCBX_VER_IEEE
)
6080 up
= dcb_ieee_getapp_mask(netdev
, &app
);
6082 adapter
->default_up
= (up
> 1) ? (ffs(up
) - 1) : 0;
6087 * ixgbe_watchdog_link_is_up - update netif_carrier status and
6088 * print link up message
6089 * @adapter: pointer to the device adapter structure
6091 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter
*adapter
)
6093 struct net_device
*netdev
= adapter
->netdev
;
6094 struct ixgbe_hw
*hw
= &adapter
->hw
;
6095 struct net_device
*upper
;
6096 struct list_head
*iter
;
6097 u32 link_speed
= adapter
->link_speed
;
6098 bool flow_rx
, flow_tx
;
6100 /* only continue if link was previously down */
6101 if (netif_carrier_ok(netdev
))
6104 adapter
->flags2
&= ~IXGBE_FLAG2_SEARCH_FOR_SFP
;
6106 switch (hw
->mac
.type
) {
6107 case ixgbe_mac_82598EB
: {
6108 u32 frctl
= IXGBE_READ_REG(hw
, IXGBE_FCTRL
);
6109 u32 rmcs
= IXGBE_READ_REG(hw
, IXGBE_RMCS
);
6110 flow_rx
= !!(frctl
& IXGBE_FCTRL_RFCE
);
6111 flow_tx
= !!(rmcs
& IXGBE_RMCS_TFCE_802_3X
);
6114 case ixgbe_mac_X540
:
6115 case ixgbe_mac_82599EB
: {
6116 u32 mflcn
= IXGBE_READ_REG(hw
, IXGBE_MFLCN
);
6117 u32 fccfg
= IXGBE_READ_REG(hw
, IXGBE_FCCFG
);
6118 flow_rx
= !!(mflcn
& IXGBE_MFLCN_RFCE
);
6119 flow_tx
= !!(fccfg
& IXGBE_FCCFG_TFCE_802_3X
);
6128 adapter
->last_rx_ptp_check
= jiffies
;
6130 if (test_bit(__IXGBE_PTP_RUNNING
, &adapter
->state
))
6131 ixgbe_ptp_start_cyclecounter(adapter
);
6133 e_info(drv
, "NIC Link is Up %s, Flow Control: %s\n",
6134 (link_speed
== IXGBE_LINK_SPEED_10GB_FULL
?
6136 (link_speed
== IXGBE_LINK_SPEED_1GB_FULL
?
6138 (link_speed
== IXGBE_LINK_SPEED_100_FULL
?
6141 ((flow_rx
&& flow_tx
) ? "RX/TX" :
6143 (flow_tx
? "TX" : "None"))));
6145 netif_carrier_on(netdev
);
6146 ixgbe_check_vf_rate_limit(adapter
);
6148 /* enable transmits */
6149 netif_tx_wake_all_queues(adapter
->netdev
);
6151 /* enable any upper devices */
6153 netdev_for_each_all_upper_dev_rcu(adapter
->netdev
, upper
, iter
) {
6154 if (netif_is_macvlan(upper
)) {
6155 struct macvlan_dev
*vlan
= netdev_priv(upper
);
6158 netif_tx_wake_all_queues(upper
);
6163 /* update the default user priority for VFs */
6164 ixgbe_update_default_up(adapter
);
6166 /* ping all the active vfs to let them know link has changed */
6167 ixgbe_ping_all_vfs(adapter
);
6171 * ixgbe_watchdog_link_is_down - update netif_carrier status and
6172 * print link down message
6173 * @adapter: pointer to the adapter structure
6175 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter
*adapter
)
6177 struct net_device
*netdev
= adapter
->netdev
;
6178 struct ixgbe_hw
*hw
= &adapter
->hw
;
6180 adapter
->link_up
= false;
6181 adapter
->link_speed
= 0;
6183 /* only continue if link was up previously */
6184 if (!netif_carrier_ok(netdev
))
6187 /* poll for SFP+ cable when link is down */
6188 if (ixgbe_is_sfp(hw
) && hw
->mac
.type
== ixgbe_mac_82598EB
)
6189 adapter
->flags2
|= IXGBE_FLAG2_SEARCH_FOR_SFP
;
6191 if (test_bit(__IXGBE_PTP_RUNNING
, &adapter
->state
))
6192 ixgbe_ptp_start_cyclecounter(adapter
);
6194 e_info(drv
, "NIC Link is Down\n");
6195 netif_carrier_off(netdev
);
6197 /* ping all the active vfs to let them know link has changed */
6198 ixgbe_ping_all_vfs(adapter
);
6201 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter
*adapter
)
6205 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
6206 struct ixgbe_ring
*tx_ring
= adapter
->tx_ring
[i
];
6208 if (tx_ring
->next_to_use
!= tx_ring
->next_to_clean
)
6215 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter
*adapter
)
6217 struct ixgbe_hw
*hw
= &adapter
->hw
;
6218 struct ixgbe_ring_feature
*vmdq
= &adapter
->ring_feature
[RING_F_VMDQ
];
6219 u32 q_per_pool
= __ALIGN_MASK(1, ~vmdq
->mask
);
6223 if (!adapter
->num_vfs
)
6226 for (i
= 0; i
< adapter
->num_vfs
; i
++) {
6227 for (j
= 0; j
< q_per_pool
; j
++) {
6230 h
= IXGBE_READ_REG(hw
, IXGBE_PVFTDHN(q_per_pool
, i
, j
));
6231 t
= IXGBE_READ_REG(hw
, IXGBE_PVFTDTN(q_per_pool
, i
, j
));
6242 * ixgbe_watchdog_flush_tx - flush queues on link down
6243 * @adapter: pointer to the device adapter structure
6245 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter
*adapter
)
6247 if (!netif_carrier_ok(adapter
->netdev
)) {
6248 if (ixgbe_ring_tx_pending(adapter
) ||
6249 ixgbe_vf_tx_pending(adapter
)) {
6250 /* We've lost link, so the controller stops DMA,
6251 * but we've got queued Tx work that's never going
6252 * to get done, so reset controller to flush Tx.
6253 * (Do the reset outside of interrupt context).
6255 e_warn(drv
, "initiating reset to clear Tx work after link loss\n");
6256 adapter
->flags2
|= IXGBE_FLAG2_RESET_REQUESTED
;
6261 static void ixgbe_spoof_check(struct ixgbe_adapter
*adapter
)
6265 /* Do not perform spoof check for 82598 or if not in IOV mode */
6266 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
||
6267 adapter
->num_vfs
== 0)
6270 ssvpc
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_SSVPC
);
6273 * ssvpc register is cleared on read, if zero then no
6274 * spoofed packets in the last interval.
6279 e_warn(drv
, "%u Spoofed packets detected\n", ssvpc
);
6283 * ixgbe_watchdog_subtask - check and bring link up
6284 * @adapter: pointer to the device adapter structure
6286 static void ixgbe_watchdog_subtask(struct ixgbe_adapter
*adapter
)
6288 /* if interface is down, removing or resetting, do nothing */
6289 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
6290 test_bit(__IXGBE_REMOVING
, &adapter
->state
) ||
6291 test_bit(__IXGBE_RESETTING
, &adapter
->state
))
6294 ixgbe_watchdog_update_link(adapter
);
6296 if (adapter
->link_up
)
6297 ixgbe_watchdog_link_is_up(adapter
);
6299 ixgbe_watchdog_link_is_down(adapter
);
6301 ixgbe_spoof_check(adapter
);
6302 ixgbe_update_stats(adapter
);
6304 ixgbe_watchdog_flush_tx(adapter
);
6308 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6309 * @adapter: the ixgbe adapter structure
6311 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter
*adapter
)
6313 struct ixgbe_hw
*hw
= &adapter
->hw
;
6316 /* not searching for SFP so there is nothing to do here */
6317 if (!(adapter
->flags2
& IXGBE_FLAG2_SEARCH_FOR_SFP
) &&
6318 !(adapter
->flags2
& IXGBE_FLAG2_SFP_NEEDS_RESET
))
6321 /* someone else is in init, wait until next service event */
6322 if (test_and_set_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
))
6325 err
= hw
->phy
.ops
.identify_sfp(hw
);
6326 if (err
== IXGBE_ERR_SFP_NOT_SUPPORTED
)
6329 if (err
== IXGBE_ERR_SFP_NOT_PRESENT
) {
6330 /* If no cable is present, then we need to reset
6331 * the next time we find a good cable. */
6332 adapter
->flags2
|= IXGBE_FLAG2_SFP_NEEDS_RESET
;
6339 /* exit if reset not needed */
6340 if (!(adapter
->flags2
& IXGBE_FLAG2_SFP_NEEDS_RESET
))
6343 adapter
->flags2
&= ~IXGBE_FLAG2_SFP_NEEDS_RESET
;
6346 * A module may be identified correctly, but the EEPROM may not have
6347 * support for that module. setup_sfp() will fail in that case, so
6348 * we should not allow that module to load.
6350 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
6351 err
= hw
->phy
.ops
.reset(hw
);
6353 err
= hw
->mac
.ops
.setup_sfp(hw
);
6355 if (err
== IXGBE_ERR_SFP_NOT_SUPPORTED
)
6358 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_CONFIG
;
6359 e_info(probe
, "detected SFP+: %d\n", hw
->phy
.sfp_type
);
6362 clear_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
);
6364 if ((err
== IXGBE_ERR_SFP_NOT_SUPPORTED
) &&
6365 (adapter
->netdev
->reg_state
== NETREG_REGISTERED
)) {
6366 e_dev_err("failed to initialize because an unsupported "
6367 "SFP+ module type was detected.\n");
6368 e_dev_err("Reload the driver after installing a "
6369 "supported module.\n");
6370 unregister_netdev(adapter
->netdev
);
6375 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6376 * @adapter: the ixgbe adapter structure
6378 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter
*adapter
)
6380 struct ixgbe_hw
*hw
= &adapter
->hw
;
6382 bool autoneg
= false;
6384 if (!(adapter
->flags
& IXGBE_FLAG_NEED_LINK_CONFIG
))
6387 /* someone else is in init, wait until next service event */
6388 if (test_and_set_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
))
6391 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_CONFIG
;
6393 speed
= hw
->phy
.autoneg_advertised
;
6394 if ((!speed
) && (hw
->mac
.ops
.get_link_capabilities
)) {
6395 hw
->mac
.ops
.get_link_capabilities(hw
, &speed
, &autoneg
);
6397 /* setup the highest link when no autoneg */
6399 if (speed
& IXGBE_LINK_SPEED_10GB_FULL
)
6400 speed
= IXGBE_LINK_SPEED_10GB_FULL
;
6404 if (hw
->mac
.ops
.setup_link
)
6405 hw
->mac
.ops
.setup_link(hw
, speed
, true);
6407 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_UPDATE
;
6408 adapter
->link_check_timeout
= jiffies
;
6409 clear_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
);
6412 #ifdef CONFIG_PCI_IOV
6413 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter
*adapter
)
6416 struct ixgbe_hw
*hw
= &adapter
->hw
;
6417 struct net_device
*netdev
= adapter
->netdev
;
6421 gpc
= IXGBE_READ_REG(hw
, IXGBE_TXDGPC
);
6422 if (gpc
) /* If incrementing then no need for the check below */
6425 * Check to see if a bad DMA write target from an errant or
6426 * malicious VF has caused a PCIe error. If so then we can
6427 * issue a VFLR to the offending VF(s) and then resume without
6428 * requesting a full slot reset.
6431 for (vf
= 0; vf
< adapter
->num_vfs
; vf
++) {
6432 ciaa
= (vf
<< 16) | 0x80000000;
6433 /* 32 bit read so align, we really want status at offset 6 */
6434 ciaa
|= PCI_COMMAND
;
6435 IXGBE_WRITE_REG(hw
, IXGBE_CIAA_82599
, ciaa
);
6436 ciad
= IXGBE_READ_REG(hw
, IXGBE_CIAD_82599
);
6438 /* disable debug mode asap after reading data */
6439 IXGBE_WRITE_REG(hw
, IXGBE_CIAA_82599
, ciaa
);
6440 /* Get the upper 16 bits which will be the PCI status reg */
6442 if (ciad
& PCI_STATUS_REC_MASTER_ABORT
) {
6443 netdev_err(netdev
, "VF %d Hung DMA\n", vf
);
6445 ciaa
= (vf
<< 16) | 0x80000000;
6447 IXGBE_WRITE_REG(hw
, IXGBE_CIAA_82599
, ciaa
);
6448 ciad
= 0x00008000; /* VFLR */
6449 IXGBE_WRITE_REG(hw
, IXGBE_CIAD_82599
, ciad
);
6451 IXGBE_WRITE_REG(hw
, IXGBE_CIAA_82599
, ciaa
);
6458 * ixgbe_service_timer - Timer Call-back
6459 * @data: pointer to adapter cast into an unsigned long
6461 static void ixgbe_service_timer(unsigned long data
)
6463 struct ixgbe_adapter
*adapter
= (struct ixgbe_adapter
*)data
;
6464 unsigned long next_event_offset
;
6467 /* poll faster when waiting for link */
6468 if (adapter
->flags
& IXGBE_FLAG_NEED_LINK_UPDATE
)
6469 next_event_offset
= HZ
/ 10;
6471 next_event_offset
= HZ
* 2;
6473 #ifdef CONFIG_PCI_IOV
6475 * don't bother with SR-IOV VF DMA hang check if there are
6476 * no VFs or the link is down
6478 if (!adapter
->num_vfs
||
6479 (adapter
->flags
& IXGBE_FLAG_NEED_LINK_UPDATE
))
6480 goto normal_timer_service
;
6482 /* If we have VFs allocated then we must check for DMA hangs */
6483 ixgbe_check_for_bad_vf(adapter
);
6484 next_event_offset
= HZ
/ 50;
6485 adapter
->timer_event_accumulator
++;
6487 if (adapter
->timer_event_accumulator
>= 100)
6488 adapter
->timer_event_accumulator
= 0;
6492 normal_timer_service
:
6494 /* Reset the timer */
6495 mod_timer(&adapter
->service_timer
, next_event_offset
+ jiffies
);
6498 ixgbe_service_event_schedule(adapter
);
6501 static void ixgbe_reset_subtask(struct ixgbe_adapter
*adapter
)
6503 if (!(adapter
->flags2
& IXGBE_FLAG2_RESET_REQUESTED
))
6506 adapter
->flags2
&= ~IXGBE_FLAG2_RESET_REQUESTED
;
6508 /* If we're already down, removing or resetting, just bail */
6509 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
6510 test_bit(__IXGBE_REMOVING
, &adapter
->state
) ||
6511 test_bit(__IXGBE_RESETTING
, &adapter
->state
))
6514 ixgbe_dump(adapter
);
6515 netdev_err(adapter
->netdev
, "Reset adapter\n");
6516 adapter
->tx_timeout_count
++;
6519 ixgbe_reinit_locked(adapter
);
6524 * ixgbe_service_task - manages and runs subtasks
6525 * @work: pointer to work_struct containing our data
6527 static void ixgbe_service_task(struct work_struct
*work
)
6529 struct ixgbe_adapter
*adapter
= container_of(work
,
6530 struct ixgbe_adapter
,
6532 if (ixgbe_removed(adapter
->hw
.hw_addr
)) {
6533 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
6535 ixgbe_down(adapter
);
6538 ixgbe_service_event_complete(adapter
);
6541 ixgbe_reset_subtask(adapter
);
6542 ixgbe_sfp_detection_subtask(adapter
);
6543 ixgbe_sfp_link_config_subtask(adapter
);
6544 ixgbe_check_overtemp_subtask(adapter
);
6545 ixgbe_watchdog_subtask(adapter
);
6546 ixgbe_fdir_reinit_subtask(adapter
);
6547 ixgbe_check_hang_subtask(adapter
);
6549 if (test_bit(__IXGBE_PTP_RUNNING
, &adapter
->state
)) {
6550 ixgbe_ptp_overflow_check(adapter
);
6551 ixgbe_ptp_rx_hang(adapter
);
6554 ixgbe_service_event_complete(adapter
);
6557 static int ixgbe_tso(struct ixgbe_ring
*tx_ring
,
6558 struct ixgbe_tx_buffer
*first
,
6561 struct sk_buff
*skb
= first
->skb
;
6562 u32 vlan_macip_lens
, type_tucmd
;
6563 u32 mss_l4len_idx
, l4len
;
6566 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
6569 if (!skb_is_gso(skb
))
6572 err
= skb_cow_head(skb
, 0);
6576 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6577 type_tucmd
= IXGBE_ADVTXD_TUCMD_L4T_TCP
;
6579 if (first
->protocol
== htons(ETH_P_IP
)) {
6580 struct iphdr
*iph
= ip_hdr(skb
);
6583 tcp_hdr(skb
)->check
= ~csum_tcpudp_magic(iph
->saddr
,
6587 type_tucmd
|= IXGBE_ADVTXD_TUCMD_IPV4
;
6588 first
->tx_flags
|= IXGBE_TX_FLAGS_TSO
|
6589 IXGBE_TX_FLAGS_CSUM
|
6590 IXGBE_TX_FLAGS_IPV4
;
6591 } else if (skb_is_gso_v6(skb
)) {
6592 ipv6_hdr(skb
)->payload_len
= 0;
6593 tcp_hdr(skb
)->check
=
6594 ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
6595 &ipv6_hdr(skb
)->daddr
,
6597 first
->tx_flags
|= IXGBE_TX_FLAGS_TSO
|
6598 IXGBE_TX_FLAGS_CSUM
;
6601 /* compute header lengths */
6602 l4len
= tcp_hdrlen(skb
);
6603 *hdr_len
= skb_transport_offset(skb
) + l4len
;
6605 /* update gso size and bytecount with header size */
6606 first
->gso_segs
= skb_shinfo(skb
)->gso_segs
;
6607 first
->bytecount
+= (first
->gso_segs
- 1) * *hdr_len
;
6609 /* mss_l4len_id: use 0 as index for TSO */
6610 mss_l4len_idx
= l4len
<< IXGBE_ADVTXD_L4LEN_SHIFT
;
6611 mss_l4len_idx
|= skb_shinfo(skb
)->gso_size
<< IXGBE_ADVTXD_MSS_SHIFT
;
6613 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6614 vlan_macip_lens
= skb_network_header_len(skb
);
6615 vlan_macip_lens
|= skb_network_offset(skb
) << IXGBE_ADVTXD_MACLEN_SHIFT
;
6616 vlan_macip_lens
|= first
->tx_flags
& IXGBE_TX_FLAGS_VLAN_MASK
;
6618 ixgbe_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, 0, type_tucmd
,
6624 static void ixgbe_tx_csum(struct ixgbe_ring
*tx_ring
,
6625 struct ixgbe_tx_buffer
*first
)
6627 struct sk_buff
*skb
= first
->skb
;
6628 u32 vlan_macip_lens
= 0;
6629 u32 mss_l4len_idx
= 0;
6632 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
) {
6633 if (!(first
->tx_flags
& IXGBE_TX_FLAGS_HW_VLAN
) &&
6634 !(first
->tx_flags
& IXGBE_TX_FLAGS_CC
))
6638 switch (first
->protocol
) {
6639 case htons(ETH_P_IP
):
6640 vlan_macip_lens
|= skb_network_header_len(skb
);
6641 type_tucmd
|= IXGBE_ADVTXD_TUCMD_IPV4
;
6642 l4_hdr
= ip_hdr(skb
)->protocol
;
6644 case htons(ETH_P_IPV6
):
6645 vlan_macip_lens
|= skb_network_header_len(skb
);
6646 l4_hdr
= ipv6_hdr(skb
)->nexthdr
;
6649 if (unlikely(net_ratelimit())) {
6650 dev_warn(tx_ring
->dev
,
6651 "partial checksum but proto=%x!\n",
6659 type_tucmd
|= IXGBE_ADVTXD_TUCMD_L4T_TCP
;
6660 mss_l4len_idx
= tcp_hdrlen(skb
) <<
6661 IXGBE_ADVTXD_L4LEN_SHIFT
;
6664 type_tucmd
|= IXGBE_ADVTXD_TUCMD_L4T_SCTP
;
6665 mss_l4len_idx
= sizeof(struct sctphdr
) <<
6666 IXGBE_ADVTXD_L4LEN_SHIFT
;
6669 mss_l4len_idx
= sizeof(struct udphdr
) <<
6670 IXGBE_ADVTXD_L4LEN_SHIFT
;
6673 if (unlikely(net_ratelimit())) {
6674 dev_warn(tx_ring
->dev
,
6675 "partial checksum but l4 proto=%x!\n",
6681 /* update TX checksum flag */
6682 first
->tx_flags
|= IXGBE_TX_FLAGS_CSUM
;
6685 /* vlan_macip_lens: MACLEN, VLAN tag */
6686 vlan_macip_lens
|= skb_network_offset(skb
) << IXGBE_ADVTXD_MACLEN_SHIFT
;
6687 vlan_macip_lens
|= first
->tx_flags
& IXGBE_TX_FLAGS_VLAN_MASK
;
6689 ixgbe_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, 0,
6690 type_tucmd
, mss_l4len_idx
);
6693 #define IXGBE_SET_FLAG(_input, _flag, _result) \
6694 ((_flag <= _result) ? \
6695 ((u32)(_input & _flag) * (_result / _flag)) : \
6696 ((u32)(_input & _flag) / (_flag / _result)))
6698 static u32
ixgbe_tx_cmd_type(struct sk_buff
*skb
, u32 tx_flags
)
6700 /* set type for advanced descriptor with frame checksum insertion */
6701 u32 cmd_type
= IXGBE_ADVTXD_DTYP_DATA
|
6702 IXGBE_ADVTXD_DCMD_DEXT
|
6703 IXGBE_ADVTXD_DCMD_IFCS
;
6705 /* set HW vlan bit if vlan is present */
6706 cmd_type
|= IXGBE_SET_FLAG(tx_flags
, IXGBE_TX_FLAGS_HW_VLAN
,
6707 IXGBE_ADVTXD_DCMD_VLE
);
6709 /* set segmentation enable bits for TSO/FSO */
6710 cmd_type
|= IXGBE_SET_FLAG(tx_flags
, IXGBE_TX_FLAGS_TSO
,
6711 IXGBE_ADVTXD_DCMD_TSE
);
6713 /* set timestamp bit if present */
6714 cmd_type
|= IXGBE_SET_FLAG(tx_flags
, IXGBE_TX_FLAGS_TSTAMP
,
6715 IXGBE_ADVTXD_MAC_TSTAMP
);
6717 /* insert frame checksum */
6718 cmd_type
^= IXGBE_SET_FLAG(skb
->no_fcs
, 1, IXGBE_ADVTXD_DCMD_IFCS
);
6723 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc
*tx_desc
,
6724 u32 tx_flags
, unsigned int paylen
)
6726 u32 olinfo_status
= paylen
<< IXGBE_ADVTXD_PAYLEN_SHIFT
;
6728 /* enable L4 checksum for TSO and TX checksum offload */
6729 olinfo_status
|= IXGBE_SET_FLAG(tx_flags
,
6730 IXGBE_TX_FLAGS_CSUM
,
6731 IXGBE_ADVTXD_POPTS_TXSM
);
6733 /* enble IPv4 checksum for TSO */
6734 olinfo_status
|= IXGBE_SET_FLAG(tx_flags
,
6735 IXGBE_TX_FLAGS_IPV4
,
6736 IXGBE_ADVTXD_POPTS_IXSM
);
6739 * Check Context must be set if Tx switch is enabled, which it
6740 * always is for case where virtual functions are running
6742 olinfo_status
|= IXGBE_SET_FLAG(tx_flags
,
6746 tx_desc
->read
.olinfo_status
= cpu_to_le32(olinfo_status
);
6749 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring
*tx_ring
, u16 size
)
6751 netif_stop_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
6753 /* Herbert's original patch had:
6754 * smp_mb__after_netif_stop_queue();
6755 * but since that doesn't exist yet, just open code it.
6759 /* We need to check again in a case another CPU has just
6760 * made room available.
6762 if (likely(ixgbe_desc_unused(tx_ring
) < size
))
6765 /* A reprieve! - use start_queue because it doesn't call schedule */
6766 netif_start_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
6767 ++tx_ring
->tx_stats
.restart_queue
;
6771 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring
*tx_ring
, u16 size
)
6773 if (likely(ixgbe_desc_unused(tx_ring
) >= size
))
6776 return __ixgbe_maybe_stop_tx(tx_ring
, size
);
6779 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6782 static void ixgbe_tx_map(struct ixgbe_ring
*tx_ring
,
6783 struct ixgbe_tx_buffer
*first
,
6786 struct sk_buff
*skb
= first
->skb
;
6787 struct ixgbe_tx_buffer
*tx_buffer
;
6788 union ixgbe_adv_tx_desc
*tx_desc
;
6789 struct skb_frag_struct
*frag
;
6791 unsigned int data_len
, size
;
6792 u32 tx_flags
= first
->tx_flags
;
6793 u32 cmd_type
= ixgbe_tx_cmd_type(skb
, tx_flags
);
6794 u16 i
= tx_ring
->next_to_use
;
6796 tx_desc
= IXGBE_TX_DESC(tx_ring
, i
);
6798 ixgbe_tx_olinfo_status(tx_desc
, tx_flags
, skb
->len
- hdr_len
);
6800 size
= skb_headlen(skb
);
6801 data_len
= skb
->data_len
;
6804 if (tx_flags
& IXGBE_TX_FLAGS_FCOE
) {
6805 if (data_len
< sizeof(struct fcoe_crc_eof
)) {
6806 size
-= sizeof(struct fcoe_crc_eof
) - data_len
;
6809 data_len
-= sizeof(struct fcoe_crc_eof
);
6814 dma
= dma_map_single(tx_ring
->dev
, skb
->data
, size
, DMA_TO_DEVICE
);
6818 for (frag
= &skb_shinfo(skb
)->frags
[0];; frag
++) {
6819 if (dma_mapping_error(tx_ring
->dev
, dma
))
6822 /* record length, and DMA address */
6823 dma_unmap_len_set(tx_buffer
, len
, size
);
6824 dma_unmap_addr_set(tx_buffer
, dma
, dma
);
6826 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
6828 while (unlikely(size
> IXGBE_MAX_DATA_PER_TXD
)) {
6829 tx_desc
->read
.cmd_type_len
=
6830 cpu_to_le32(cmd_type
^ IXGBE_MAX_DATA_PER_TXD
);
6834 if (i
== tx_ring
->count
) {
6835 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
6838 tx_desc
->read
.olinfo_status
= 0;
6840 dma
+= IXGBE_MAX_DATA_PER_TXD
;
6841 size
-= IXGBE_MAX_DATA_PER_TXD
;
6843 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
6846 if (likely(!data_len
))
6849 tx_desc
->read
.cmd_type_len
= cpu_to_le32(cmd_type
^ size
);
6853 if (i
== tx_ring
->count
) {
6854 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
6857 tx_desc
->read
.olinfo_status
= 0;
6860 size
= min_t(unsigned int, data_len
, skb_frag_size(frag
));
6862 size
= skb_frag_size(frag
);
6866 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0, size
,
6869 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
6872 /* write last descriptor with RS and EOP bits */
6873 cmd_type
|= size
| IXGBE_TXD_CMD
;
6874 tx_desc
->read
.cmd_type_len
= cpu_to_le32(cmd_type
);
6876 netdev_tx_sent_queue(txring_txq(tx_ring
), first
->bytecount
);
6878 /* set the timestamp */
6879 first
->time_stamp
= jiffies
;
6882 * Force memory writes to complete before letting h/w know there
6883 * are new descriptors to fetch. (Only applicable for weak-ordered
6884 * memory model archs, such as IA-64).
6886 * We also need this memory barrier to make certain all of the
6887 * status bits have been updated before next_to_watch is written.
6891 /* set next_to_watch value indicating a packet is present */
6892 first
->next_to_watch
= tx_desc
;
6895 if (i
== tx_ring
->count
)
6898 tx_ring
->next_to_use
= i
;
6900 ixgbe_maybe_stop_tx(tx_ring
, DESC_NEEDED
);
6902 if (netif_xmit_stopped(txring_txq(tx_ring
)) || !skb
->xmit_more
) {
6903 /* notify HW of packet */
6904 ixgbe_write_tail(tx_ring
, i
);
6909 dev_err(tx_ring
->dev
, "TX DMA map failed\n");
6911 /* clear dma mappings for failed tx_buffer_info map */
6913 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
6914 ixgbe_unmap_and_free_tx_resource(tx_ring
, tx_buffer
);
6915 if (tx_buffer
== first
)
6922 tx_ring
->next_to_use
= i
;
6925 static void ixgbe_atr(struct ixgbe_ring
*ring
,
6926 struct ixgbe_tx_buffer
*first
)
6928 struct ixgbe_q_vector
*q_vector
= ring
->q_vector
;
6929 union ixgbe_atr_hash_dword input
= { .dword
= 0 };
6930 union ixgbe_atr_hash_dword common
= { .dword
= 0 };
6932 unsigned char *network
;
6934 struct ipv6hdr
*ipv6
;
6939 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6943 /* do nothing if sampling is disabled */
6944 if (!ring
->atr_sample_rate
)
6949 /* snag network header to get L4 type and address */
6950 hdr
.network
= skb_network_header(first
->skb
);
6952 /* Currently only IPv4/IPv6 with TCP is supported */
6953 if ((first
->protocol
!= htons(ETH_P_IPV6
) ||
6954 hdr
.ipv6
->nexthdr
!= IPPROTO_TCP
) &&
6955 (first
->protocol
!= htons(ETH_P_IP
) ||
6956 hdr
.ipv4
->protocol
!= IPPROTO_TCP
))
6959 th
= tcp_hdr(first
->skb
);
6961 /* skip this packet since it is invalid or the socket is closing */
6965 /* sample on all syn packets or once every atr sample count */
6966 if (!th
->syn
&& (ring
->atr_count
< ring
->atr_sample_rate
))
6969 /* reset sample count */
6970 ring
->atr_count
= 0;
6972 vlan_id
= htons(first
->tx_flags
>> IXGBE_TX_FLAGS_VLAN_SHIFT
);
6975 * src and dst are inverted, think how the receiver sees them
6977 * The input is broken into two sections, a non-compressed section
6978 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6979 * is XORed together and stored in the compressed dword.
6981 input
.formatted
.vlan_id
= vlan_id
;
6984 * since src port and flex bytes occupy the same word XOR them together
6985 * and write the value to source port portion of compressed dword
6987 if (first
->tx_flags
& (IXGBE_TX_FLAGS_SW_VLAN
| IXGBE_TX_FLAGS_HW_VLAN
))
6988 common
.port
.src
^= th
->dest
^ htons(ETH_P_8021Q
);
6990 common
.port
.src
^= th
->dest
^ first
->protocol
;
6991 common
.port
.dst
^= th
->source
;
6993 if (first
->protocol
== htons(ETH_P_IP
)) {
6994 input
.formatted
.flow_type
= IXGBE_ATR_FLOW_TYPE_TCPV4
;
6995 common
.ip
^= hdr
.ipv4
->saddr
^ hdr
.ipv4
->daddr
;
6997 input
.formatted
.flow_type
= IXGBE_ATR_FLOW_TYPE_TCPV6
;
6998 common
.ip
^= hdr
.ipv6
->saddr
.s6_addr32
[0] ^
6999 hdr
.ipv6
->saddr
.s6_addr32
[1] ^
7000 hdr
.ipv6
->saddr
.s6_addr32
[2] ^
7001 hdr
.ipv6
->saddr
.s6_addr32
[3] ^
7002 hdr
.ipv6
->daddr
.s6_addr32
[0] ^
7003 hdr
.ipv6
->daddr
.s6_addr32
[1] ^
7004 hdr
.ipv6
->daddr
.s6_addr32
[2] ^
7005 hdr
.ipv6
->daddr
.s6_addr32
[3];
7008 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
7009 ixgbe_fdir_add_signature_filter_82599(&q_vector
->adapter
->hw
,
7010 input
, common
, ring
->queue_index
);
7013 static u16
ixgbe_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
7014 void *accel_priv
, select_queue_fallback_t fallback
)
7016 struct ixgbe_fwd_adapter
*fwd_adapter
= accel_priv
;
7018 struct ixgbe_adapter
*adapter
;
7019 struct ixgbe_ring_feature
*f
;
7024 return skb
->queue_mapping
+ fwd_adapter
->tx_base_queue
;
7029 * only execute the code below if protocol is FCoE
7030 * or FIP and we have FCoE enabled on the adapter
7032 switch (vlan_get_protocol(skb
)) {
7033 case htons(ETH_P_FCOE
):
7034 case htons(ETH_P_FIP
):
7035 adapter
= netdev_priv(dev
);
7037 if (adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
)
7040 return fallback(dev
, skb
);
7043 f
= &adapter
->ring_feature
[RING_F_FCOE
];
7045 txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) :
7048 while (txq
>= f
->indices
)
7051 return txq
+ f
->offset
;
7053 return fallback(dev
, skb
);
7057 netdev_tx_t
ixgbe_xmit_frame_ring(struct sk_buff
*skb
,
7058 struct ixgbe_adapter
*adapter
,
7059 struct ixgbe_ring
*tx_ring
)
7061 struct ixgbe_tx_buffer
*first
;
7065 u16 count
= TXD_USE_COUNT(skb_headlen(skb
));
7066 __be16 protocol
= skb
->protocol
;
7070 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
7071 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
7072 * + 2 desc gap to keep tail from touching head,
7073 * + 1 desc for context descriptor,
7074 * otherwise try next time
7076 for (f
= 0; f
< skb_shinfo(skb
)->nr_frags
; f
++)
7077 count
+= TXD_USE_COUNT(skb_shinfo(skb
)->frags
[f
].size
);
7079 if (ixgbe_maybe_stop_tx(tx_ring
, count
+ 3)) {
7080 tx_ring
->tx_stats
.tx_busy
++;
7081 return NETDEV_TX_BUSY
;
7084 /* record the location of the first descriptor for this packet */
7085 first
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_use
];
7087 first
->bytecount
= skb
->len
;
7088 first
->gso_segs
= 1;
7090 /* if we have a HW VLAN tag being added default to the HW one */
7091 if (vlan_tx_tag_present(skb
)) {
7092 tx_flags
|= vlan_tx_tag_get(skb
) << IXGBE_TX_FLAGS_VLAN_SHIFT
;
7093 tx_flags
|= IXGBE_TX_FLAGS_HW_VLAN
;
7094 /* else if it is a SW VLAN check the next protocol and store the tag */
7095 } else if (protocol
== htons(ETH_P_8021Q
)) {
7096 struct vlan_hdr
*vhdr
, _vhdr
;
7097 vhdr
= skb_header_pointer(skb
, ETH_HLEN
, sizeof(_vhdr
), &_vhdr
);
7101 protocol
= vhdr
->h_vlan_encapsulated_proto
;
7102 tx_flags
|= ntohs(vhdr
->h_vlan_TCI
) <<
7103 IXGBE_TX_FLAGS_VLAN_SHIFT
;
7104 tx_flags
|= IXGBE_TX_FLAGS_SW_VLAN
;
7107 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_HW_TSTAMP
) &&
7108 adapter
->ptp_clock
&&
7109 !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS
,
7111 skb_shinfo(skb
)->tx_flags
|= SKBTX_IN_PROGRESS
;
7112 tx_flags
|= IXGBE_TX_FLAGS_TSTAMP
;
7114 /* schedule check for Tx timestamp */
7115 adapter
->ptp_tx_skb
= skb_get(skb
);
7116 adapter
->ptp_tx_start
= jiffies
;
7117 schedule_work(&adapter
->ptp_tx_work
);
7120 skb_tx_timestamp(skb
);
7122 #ifdef CONFIG_PCI_IOV
7124 * Use the l2switch_enable flag - would be false if the DMA
7125 * Tx switch had been disabled.
7127 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
7128 tx_flags
|= IXGBE_TX_FLAGS_CC
;
7131 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
7132 if ((adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) &&
7133 ((tx_flags
& (IXGBE_TX_FLAGS_HW_VLAN
| IXGBE_TX_FLAGS_SW_VLAN
)) ||
7134 (skb
->priority
!= TC_PRIO_CONTROL
))) {
7135 tx_flags
&= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK
;
7136 tx_flags
|= (skb
->priority
& 0x7) <<
7137 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT
;
7138 if (tx_flags
& IXGBE_TX_FLAGS_SW_VLAN
) {
7139 struct vlan_ethhdr
*vhdr
;
7141 if (skb_cow_head(skb
, 0))
7143 vhdr
= (struct vlan_ethhdr
*)skb
->data
;
7144 vhdr
->h_vlan_TCI
= htons(tx_flags
>>
7145 IXGBE_TX_FLAGS_VLAN_SHIFT
);
7147 tx_flags
|= IXGBE_TX_FLAGS_HW_VLAN
;
7151 /* record initial flags and protocol */
7152 first
->tx_flags
= tx_flags
;
7153 first
->protocol
= protocol
;
7156 /* setup tx offload for FCoE */
7157 if ((protocol
== htons(ETH_P_FCOE
)) &&
7158 (tx_ring
->netdev
->features
& (NETIF_F_FSO
| NETIF_F_FCOE_CRC
))) {
7159 tso
= ixgbe_fso(tx_ring
, first
, &hdr_len
);
7166 #endif /* IXGBE_FCOE */
7167 tso
= ixgbe_tso(tx_ring
, first
, &hdr_len
);
7171 ixgbe_tx_csum(tx_ring
, first
);
7173 /* add the ATR filter if ATR is on */
7174 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE
, &tx_ring
->state
))
7175 ixgbe_atr(tx_ring
, first
);
7179 #endif /* IXGBE_FCOE */
7180 ixgbe_tx_map(tx_ring
, first
, hdr_len
);
7182 return NETDEV_TX_OK
;
7185 dev_kfree_skb_any(first
->skb
);
7188 return NETDEV_TX_OK
;
7191 static netdev_tx_t
__ixgbe_xmit_frame(struct sk_buff
*skb
,
7192 struct net_device
*netdev
,
7193 struct ixgbe_ring
*ring
)
7195 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
7196 struct ixgbe_ring
*tx_ring
;
7199 * The minimum packet size for olinfo paylen is 17 so pad the skb
7200 * in order to meet this minimum size requirement.
7202 if (unlikely(skb
->len
< 17)) {
7203 if (skb_pad(skb
, 17 - skb
->len
))
7204 return NETDEV_TX_OK
;
7206 skb_set_tail_pointer(skb
, 17);
7209 tx_ring
= ring
? ring
: adapter
->tx_ring
[skb
->queue_mapping
];
7211 return ixgbe_xmit_frame_ring(skb
, adapter
, tx_ring
);
7214 static netdev_tx_t
ixgbe_xmit_frame(struct sk_buff
*skb
,
7215 struct net_device
*netdev
)
7217 return __ixgbe_xmit_frame(skb
, netdev
, NULL
);
7221 * ixgbe_set_mac - Change the Ethernet Address of the NIC
7222 * @netdev: network interface device structure
7223 * @p: pointer to an address structure
7225 * Returns 0 on success, negative on failure
7227 static int ixgbe_set_mac(struct net_device
*netdev
, void *p
)
7229 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
7230 struct ixgbe_hw
*hw
= &adapter
->hw
;
7231 struct sockaddr
*addr
= p
;
7234 if (!is_valid_ether_addr(addr
->sa_data
))
7235 return -EADDRNOTAVAIL
;
7237 ixgbe_del_mac_filter(adapter
, hw
->mac
.addr
, VMDQ_P(0));
7238 memcpy(netdev
->dev_addr
, addr
->sa_data
, netdev
->addr_len
);
7239 memcpy(hw
->mac
.addr
, addr
->sa_data
, netdev
->addr_len
);
7241 ret
= ixgbe_add_mac_filter(adapter
, hw
->mac
.addr
, VMDQ_P(0));
7242 return ret
> 0 ? 0 : ret
;
7246 ixgbe_mdio_read(struct net_device
*netdev
, int prtad
, int devad
, u16 addr
)
7248 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
7249 struct ixgbe_hw
*hw
= &adapter
->hw
;
7253 if (prtad
!= hw
->phy
.mdio
.prtad
)
7255 rc
= hw
->phy
.ops
.read_reg(hw
, addr
, devad
, &value
);
7261 static int ixgbe_mdio_write(struct net_device
*netdev
, int prtad
, int devad
,
7262 u16 addr
, u16 value
)
7264 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
7265 struct ixgbe_hw
*hw
= &adapter
->hw
;
7267 if (prtad
!= hw
->phy
.mdio
.prtad
)
7269 return hw
->phy
.ops
.write_reg(hw
, addr
, devad
, value
);
7272 static int ixgbe_ioctl(struct net_device
*netdev
, struct ifreq
*req
, int cmd
)
7274 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
7278 return ixgbe_ptp_set_ts_config(adapter
, req
);
7280 return ixgbe_ptp_get_ts_config(adapter
, req
);
7282 return mdio_mii_ioctl(&adapter
->hw
.phy
.mdio
, if_mii(req
), cmd
);
7287 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
7289 * @netdev: network interface device structure
7291 * Returns non-zero on failure
7293 static int ixgbe_add_sanmac_netdev(struct net_device
*dev
)
7296 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
7297 struct ixgbe_hw
*hw
= &adapter
->hw
;
7299 if (is_valid_ether_addr(hw
->mac
.san_addr
)) {
7301 err
= dev_addr_add(dev
, hw
->mac
.san_addr
, NETDEV_HW_ADDR_T_SAN
);
7304 /* update SAN MAC vmdq pool selection */
7305 hw
->mac
.ops
.set_vmdq_san_mac(hw
, VMDQ_P(0));
7311 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
7313 * @netdev: network interface device structure
7315 * Returns non-zero on failure
7317 static int ixgbe_del_sanmac_netdev(struct net_device
*dev
)
7320 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
7321 struct ixgbe_mac_info
*mac
= &adapter
->hw
.mac
;
7323 if (is_valid_ether_addr(mac
->san_addr
)) {
7325 err
= dev_addr_del(dev
, mac
->san_addr
, NETDEV_HW_ADDR_T_SAN
);
7331 #ifdef CONFIG_NET_POLL_CONTROLLER
7333 * Polling 'interrupt' - used by things like netconsole to send skbs
7334 * without having to re-enable interrupts. It's not called while
7335 * the interrupt routine is executing.
7337 static void ixgbe_netpoll(struct net_device
*netdev
)
7339 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
7342 /* if interface is down do nothing */
7343 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
7346 adapter
->flags
|= IXGBE_FLAG_IN_NETPOLL
;
7347 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
7348 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
7349 ixgbe_msix_clean_rings(0, adapter
->q_vector
[i
]);
7351 ixgbe_intr(adapter
->pdev
->irq
, netdev
);
7353 adapter
->flags
&= ~IXGBE_FLAG_IN_NETPOLL
;
7357 static struct rtnl_link_stats64
*ixgbe_get_stats64(struct net_device
*netdev
,
7358 struct rtnl_link_stats64
*stats
)
7360 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
7364 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
7365 struct ixgbe_ring
*ring
= ACCESS_ONCE(adapter
->rx_ring
[i
]);
7371 start
= u64_stats_fetch_begin_irq(&ring
->syncp
);
7372 packets
= ring
->stats
.packets
;
7373 bytes
= ring
->stats
.bytes
;
7374 } while (u64_stats_fetch_retry_irq(&ring
->syncp
, start
));
7375 stats
->rx_packets
+= packets
;
7376 stats
->rx_bytes
+= bytes
;
7380 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
7381 struct ixgbe_ring
*ring
= ACCESS_ONCE(adapter
->tx_ring
[i
]);
7387 start
= u64_stats_fetch_begin_irq(&ring
->syncp
);
7388 packets
= ring
->stats
.packets
;
7389 bytes
= ring
->stats
.bytes
;
7390 } while (u64_stats_fetch_retry_irq(&ring
->syncp
, start
));
7391 stats
->tx_packets
+= packets
;
7392 stats
->tx_bytes
+= bytes
;
7396 /* following stats updated by ixgbe_watchdog_task() */
7397 stats
->multicast
= netdev
->stats
.multicast
;
7398 stats
->rx_errors
= netdev
->stats
.rx_errors
;
7399 stats
->rx_length_errors
= netdev
->stats
.rx_length_errors
;
7400 stats
->rx_crc_errors
= netdev
->stats
.rx_crc_errors
;
7401 stats
->rx_missed_errors
= netdev
->stats
.rx_missed_errors
;
7405 #ifdef CONFIG_IXGBE_DCB
7407 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7408 * @adapter: pointer to ixgbe_adapter
7409 * @tc: number of traffic classes currently enabled
7411 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7412 * 802.1Q priority maps to a packet buffer that exists.
7414 static void ixgbe_validate_rtr(struct ixgbe_adapter
*adapter
, u8 tc
)
7416 struct ixgbe_hw
*hw
= &adapter
->hw
;
7420 /* 82598 have a static priority to TC mapping that can not
7421 * be changed so no validation is needed.
7423 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
7426 reg
= IXGBE_READ_REG(hw
, IXGBE_RTRUP2TC
);
7429 for (i
= 0; i
< MAX_TRAFFIC_CLASS
; i
++) {
7430 u8 up2tc
= reg
>> (i
* IXGBE_RTRUP2TC_UP_SHIFT
);
7432 /* If up2tc is out of bounds default to zero */
7434 reg
&= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT
);
7438 IXGBE_WRITE_REG(hw
, IXGBE_RTRUP2TC
, reg
);
7444 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
7445 * @adapter: Pointer to adapter struct
7447 * Populate the netdev user priority to tc map
7449 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter
*adapter
)
7451 struct net_device
*dev
= adapter
->netdev
;
7452 struct ixgbe_dcb_config
*dcb_cfg
= &adapter
->dcb_cfg
;
7453 struct ieee_ets
*ets
= adapter
->ixgbe_ieee_ets
;
7456 for (prio
= 0; prio
< MAX_USER_PRIORITY
; prio
++) {
7459 if (adapter
->dcbx_cap
& DCB_CAP_DCBX_VER_CEE
)
7460 tc
= ixgbe_dcb_get_tc_from_up(dcb_cfg
, 0, prio
);
7462 tc
= ets
->prio_tc
[prio
];
7464 netdev_set_prio_tc_map(dev
, prio
, tc
);
7468 #endif /* CONFIG_IXGBE_DCB */
7470 * ixgbe_setup_tc - configure net_device for multiple traffic classes
7472 * @netdev: net device to configure
7473 * @tc: number of traffic classes to enable
7475 int ixgbe_setup_tc(struct net_device
*dev
, u8 tc
)
7477 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
7478 struct ixgbe_hw
*hw
= &adapter
->hw
;
7481 /* Hardware supports up to 8 traffic classes */
7482 if (tc
> adapter
->dcb_cfg
.num_tcs
.pg_tcs
||
7483 (hw
->mac
.type
== ixgbe_mac_82598EB
&&
7484 tc
< MAX_TRAFFIC_CLASS
))
7487 pools
= (find_first_zero_bit(&adapter
->fwd_bitmask
, 32) > 1);
7488 if (tc
&& pools
&& adapter
->num_rx_pools
> IXGBE_MAX_DCBMACVLANS
)
7491 /* Hardware has to reinitialize queues and interrupts to
7492 * match packet buffer alignment. Unfortunately, the
7493 * hardware is not flexible enough to do this dynamically.
7495 if (netif_running(dev
))
7497 ixgbe_clear_interrupt_scheme(adapter
);
7499 #ifdef CONFIG_IXGBE_DCB
7501 netdev_set_num_tc(dev
, tc
);
7502 ixgbe_set_prio_tc_map(adapter
);
7504 adapter
->flags
|= IXGBE_FLAG_DCB_ENABLED
;
7506 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
7507 adapter
->last_lfc_mode
= adapter
->hw
.fc
.requested_mode
;
7508 adapter
->hw
.fc
.requested_mode
= ixgbe_fc_none
;
7511 netdev_reset_tc(dev
);
7513 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
)
7514 adapter
->hw
.fc
.requested_mode
= adapter
->last_lfc_mode
;
7516 adapter
->flags
&= ~IXGBE_FLAG_DCB_ENABLED
;
7518 adapter
->temp_dcb_cfg
.pfc_mode_enable
= false;
7519 adapter
->dcb_cfg
.pfc_mode_enable
= false;
7522 ixgbe_validate_rtr(adapter
, tc
);
7524 #endif /* CONFIG_IXGBE_DCB */
7525 ixgbe_init_interrupt_scheme(adapter
);
7527 if (netif_running(dev
))
7528 return ixgbe_open(dev
);
7533 #ifdef CONFIG_PCI_IOV
7534 void ixgbe_sriov_reinit(struct ixgbe_adapter
*adapter
)
7536 struct net_device
*netdev
= adapter
->netdev
;
7539 ixgbe_setup_tc(netdev
, netdev_get_num_tc(netdev
));
7544 void ixgbe_do_reset(struct net_device
*netdev
)
7546 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
7548 if (netif_running(netdev
))
7549 ixgbe_reinit_locked(adapter
);
7551 ixgbe_reset(adapter
);
7554 static netdev_features_t
ixgbe_fix_features(struct net_device
*netdev
,
7555 netdev_features_t features
)
7557 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
7559 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7560 if (!(features
& NETIF_F_RXCSUM
))
7561 features
&= ~NETIF_F_LRO
;
7563 /* Turn off LRO if not RSC capable */
7564 if (!(adapter
->flags2
& IXGBE_FLAG2_RSC_CAPABLE
))
7565 features
&= ~NETIF_F_LRO
;
7570 static int ixgbe_set_features(struct net_device
*netdev
,
7571 netdev_features_t features
)
7573 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
7574 netdev_features_t changed
= netdev
->features
^ features
;
7575 bool need_reset
= false;
7577 /* Make sure RSC matches LRO, reset if change */
7578 if (!(features
& NETIF_F_LRO
)) {
7579 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)
7581 adapter
->flags2
&= ~IXGBE_FLAG2_RSC_ENABLED
;
7582 } else if ((adapter
->flags2
& IXGBE_FLAG2_RSC_CAPABLE
) &&
7583 !(adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)) {
7584 if (adapter
->rx_itr_setting
== 1 ||
7585 adapter
->rx_itr_setting
> IXGBE_MIN_RSC_ITR
) {
7586 adapter
->flags2
|= IXGBE_FLAG2_RSC_ENABLED
;
7588 } else if ((changed
^ features
) & NETIF_F_LRO
) {
7589 e_info(probe
, "rx-usecs set too low, "
7595 * Check if Flow Director n-tuple support was enabled or disabled. If
7596 * the state changed, we need to reset.
7598 switch (features
& NETIF_F_NTUPLE
) {
7599 case NETIF_F_NTUPLE
:
7600 /* turn off ATR, enable perfect filters and reset */
7601 if (!(adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
))
7604 adapter
->flags
&= ~IXGBE_FLAG_FDIR_HASH_CAPABLE
;
7605 adapter
->flags
|= IXGBE_FLAG_FDIR_PERFECT_CAPABLE
;
7608 /* turn off perfect filters, enable ATR and reset */
7609 if (adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
)
7612 adapter
->flags
&= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE
;
7614 /* We cannot enable ATR if SR-IOV is enabled */
7615 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
7618 /* We cannot enable ATR if we have 2 or more traffic classes */
7619 if (netdev_get_num_tc(netdev
) > 1)
7622 /* We cannot enable ATR if RSS is disabled */
7623 if (adapter
->ring_feature
[RING_F_RSS
].limit
<= 1)
7626 /* A sample rate of 0 indicates ATR disabled */
7627 if (!adapter
->atr_sample_rate
)
7630 adapter
->flags
|= IXGBE_FLAG_FDIR_HASH_CAPABLE
;
7634 if (features
& NETIF_F_HW_VLAN_CTAG_RX
)
7635 ixgbe_vlan_strip_enable(adapter
);
7637 ixgbe_vlan_strip_disable(adapter
);
7639 if (changed
& NETIF_F_RXALL
)
7642 netdev
->features
= features
;
7644 ixgbe_do_reset(netdev
);
7649 static int ixgbe_ndo_fdb_add(struct ndmsg
*ndm
, struct nlattr
*tb
[],
7650 struct net_device
*dev
,
7651 const unsigned char *addr
,
7654 /* guarantee we can provide a unique filter for the unicast address */
7655 if (is_unicast_ether_addr(addr
) || is_link_local_ether_addr(addr
)) {
7656 if (IXGBE_MAX_PF_MACVLANS
<= netdev_uc_count(dev
))
7660 return ndo_dflt_fdb_add(ndm
, tb
, dev
, addr
, flags
);
7663 static int ixgbe_ndo_bridge_setlink(struct net_device
*dev
,
7664 struct nlmsghdr
*nlh
)
7666 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
7667 struct nlattr
*attr
, *br_spec
;
7670 if (!(adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
))
7673 br_spec
= nlmsg_find_attr(nlh
, sizeof(struct ifinfomsg
), IFLA_AF_SPEC
);
7675 nla_for_each_nested(attr
, br_spec
, rem
) {
7679 if (nla_type(attr
) != IFLA_BRIDGE_MODE
)
7682 mode
= nla_get_u16(attr
);
7683 if (mode
== BRIDGE_MODE_VEPA
) {
7685 adapter
->flags2
&= ~IXGBE_FLAG2_BRIDGE_MODE_VEB
;
7686 } else if (mode
== BRIDGE_MODE_VEB
) {
7687 reg
= IXGBE_PFDTXGSWC_VT_LBEN
;
7688 adapter
->flags2
|= IXGBE_FLAG2_BRIDGE_MODE_VEB
;
7692 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_PFDTXGSWC
, reg
);
7694 e_info(drv
, "enabling bridge mode: %s\n",
7695 mode
== BRIDGE_MODE_VEPA
? "VEPA" : "VEB");
7701 static int ixgbe_ndo_bridge_getlink(struct sk_buff
*skb
, u32 pid
, u32 seq
,
7702 struct net_device
*dev
,
7705 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
7708 if (!(adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
))
7711 if (adapter
->flags2
& IXGBE_FLAG2_BRIDGE_MODE_VEB
)
7712 mode
= BRIDGE_MODE_VEB
;
7714 mode
= BRIDGE_MODE_VEPA
;
7716 return ndo_dflt_bridge_getlink(skb
, pid
, seq
, dev
, mode
);
7719 static void *ixgbe_fwd_add(struct net_device
*pdev
, struct net_device
*vdev
)
7721 struct ixgbe_fwd_adapter
*fwd_adapter
= NULL
;
7722 struct ixgbe_adapter
*adapter
= netdev_priv(pdev
);
7723 int used_pools
= adapter
->num_vfs
+ adapter
->num_rx_pools
;
7727 /* Hardware has a limited number of available pools. Each VF, and the
7728 * PF require a pool. Check to ensure we don't attempt to use more
7729 * then the available number of pools.
7731 if (used_pools
>= IXGBE_MAX_VF_FUNCTIONS
)
7732 return ERR_PTR(-EINVAL
);
7735 if (vdev
->num_rx_queues
!= vdev
->num_tx_queues
) {
7736 netdev_info(pdev
, "%s: Only supports a single queue count for TX and RX\n",
7738 return ERR_PTR(-EINVAL
);
7741 /* Check for hardware restriction on number of rx/tx queues */
7742 if (vdev
->num_tx_queues
> IXGBE_MAX_L2A_QUEUES
||
7743 vdev
->num_tx_queues
== IXGBE_BAD_L2A_QUEUE
) {
7745 "%s: Supports RX/TX Queue counts 1,2, and 4\n",
7747 return ERR_PTR(-EINVAL
);
7750 if (((adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) &&
7751 adapter
->num_rx_pools
> IXGBE_MAX_DCBMACVLANS
- 1) ||
7752 (adapter
->num_rx_pools
> IXGBE_MAX_MACVLANS
))
7753 return ERR_PTR(-EBUSY
);
7755 fwd_adapter
= kcalloc(1, sizeof(struct ixgbe_fwd_adapter
), GFP_KERNEL
);
7757 return ERR_PTR(-ENOMEM
);
7759 pool
= find_first_zero_bit(&adapter
->fwd_bitmask
, 32);
7760 adapter
->num_rx_pools
++;
7761 set_bit(pool
, &adapter
->fwd_bitmask
);
7762 limit
= find_last_bit(&adapter
->fwd_bitmask
, 32);
7764 /* Enable VMDq flag so device will be set in VM mode */
7765 adapter
->flags
|= IXGBE_FLAG_VMDQ_ENABLED
| IXGBE_FLAG_SRIOV_ENABLED
;
7766 adapter
->ring_feature
[RING_F_VMDQ
].limit
= limit
+ 1;
7767 adapter
->ring_feature
[RING_F_RSS
].limit
= vdev
->num_tx_queues
;
7769 /* Force reinit of ring allocation with VMDQ enabled */
7770 err
= ixgbe_setup_tc(pdev
, netdev_get_num_tc(pdev
));
7773 fwd_adapter
->pool
= pool
;
7774 fwd_adapter
->real_adapter
= adapter
;
7775 err
= ixgbe_fwd_ring_up(vdev
, fwd_adapter
);
7778 netif_tx_start_all_queues(vdev
);
7781 /* unwind counter and free adapter struct */
7783 "%s: dfwd hardware acceleration failed\n", vdev
->name
);
7784 clear_bit(pool
, &adapter
->fwd_bitmask
);
7785 adapter
->num_rx_pools
--;
7787 return ERR_PTR(err
);
7790 static void ixgbe_fwd_del(struct net_device
*pdev
, void *priv
)
7792 struct ixgbe_fwd_adapter
*fwd_adapter
= priv
;
7793 struct ixgbe_adapter
*adapter
= fwd_adapter
->real_adapter
;
7796 clear_bit(fwd_adapter
->pool
, &adapter
->fwd_bitmask
);
7797 adapter
->num_rx_pools
--;
7799 limit
= find_last_bit(&adapter
->fwd_bitmask
, 32);
7800 adapter
->ring_feature
[RING_F_VMDQ
].limit
= limit
+ 1;
7801 ixgbe_fwd_ring_down(fwd_adapter
->netdev
, fwd_adapter
);
7802 ixgbe_setup_tc(pdev
, netdev_get_num_tc(pdev
));
7803 netdev_dbg(pdev
, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
7804 fwd_adapter
->pool
, adapter
->num_rx_pools
,
7805 fwd_adapter
->rx_base_queue
,
7806 fwd_adapter
->rx_base_queue
+ adapter
->num_rx_queues_per_pool
,
7807 adapter
->fwd_bitmask
);
7811 static const struct net_device_ops ixgbe_netdev_ops
= {
7812 .ndo_open
= ixgbe_open
,
7813 .ndo_stop
= ixgbe_close
,
7814 .ndo_start_xmit
= ixgbe_xmit_frame
,
7815 .ndo_select_queue
= ixgbe_select_queue
,
7816 .ndo_set_rx_mode
= ixgbe_set_rx_mode
,
7817 .ndo_validate_addr
= eth_validate_addr
,
7818 .ndo_set_mac_address
= ixgbe_set_mac
,
7819 .ndo_change_mtu
= ixgbe_change_mtu
,
7820 .ndo_tx_timeout
= ixgbe_tx_timeout
,
7821 .ndo_vlan_rx_add_vid
= ixgbe_vlan_rx_add_vid
,
7822 .ndo_vlan_rx_kill_vid
= ixgbe_vlan_rx_kill_vid
,
7823 .ndo_do_ioctl
= ixgbe_ioctl
,
7824 .ndo_set_vf_mac
= ixgbe_ndo_set_vf_mac
,
7825 .ndo_set_vf_vlan
= ixgbe_ndo_set_vf_vlan
,
7826 .ndo_set_vf_rate
= ixgbe_ndo_set_vf_bw
,
7827 .ndo_set_vf_spoofchk
= ixgbe_ndo_set_vf_spoofchk
,
7828 .ndo_get_vf_config
= ixgbe_ndo_get_vf_config
,
7829 .ndo_get_stats64
= ixgbe_get_stats64
,
7830 #ifdef CONFIG_IXGBE_DCB
7831 .ndo_setup_tc
= ixgbe_setup_tc
,
7833 #ifdef CONFIG_NET_POLL_CONTROLLER
7834 .ndo_poll_controller
= ixgbe_netpoll
,
7836 #ifdef CONFIG_NET_RX_BUSY_POLL
7837 .ndo_busy_poll
= ixgbe_low_latency_recv
,
7840 .ndo_fcoe_ddp_setup
= ixgbe_fcoe_ddp_get
,
7841 .ndo_fcoe_ddp_target
= ixgbe_fcoe_ddp_target
,
7842 .ndo_fcoe_ddp_done
= ixgbe_fcoe_ddp_put
,
7843 .ndo_fcoe_enable
= ixgbe_fcoe_enable
,
7844 .ndo_fcoe_disable
= ixgbe_fcoe_disable
,
7845 .ndo_fcoe_get_wwn
= ixgbe_fcoe_get_wwn
,
7846 .ndo_fcoe_get_hbainfo
= ixgbe_fcoe_get_hbainfo
,
7847 #endif /* IXGBE_FCOE */
7848 .ndo_set_features
= ixgbe_set_features
,
7849 .ndo_fix_features
= ixgbe_fix_features
,
7850 .ndo_fdb_add
= ixgbe_ndo_fdb_add
,
7851 .ndo_bridge_setlink
= ixgbe_ndo_bridge_setlink
,
7852 .ndo_bridge_getlink
= ixgbe_ndo_bridge_getlink
,
7853 .ndo_dfwd_add_station
= ixgbe_fwd_add
,
7854 .ndo_dfwd_del_station
= ixgbe_fwd_del
,
7858 * ixgbe_enumerate_functions - Get the number of ports this device has
7859 * @adapter: adapter structure
7861 * This function enumerates the phsyical functions co-located on a single slot,
7862 * in order to determine how many ports a device has. This is most useful in
7863 * determining the required GT/s of PCIe bandwidth necessary for optimal
7866 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter
*adapter
)
7868 struct pci_dev
*entry
, *pdev
= adapter
->pdev
;
7871 /* Some cards can not use the generic count PCIe functions method,
7872 * because they are behind a parent switch, so we hardcode these with
7873 * the correct number of functions.
7875 if (ixgbe_pcie_from_parent(&adapter
->hw
))
7878 list_for_each_entry(entry
, &adapter
->pdev
->bus
->devices
, bus_list
) {
7879 /* don't count virtual functions */
7880 if (entry
->is_virtfn
)
7883 /* When the devices on the bus don't all match our device ID,
7884 * we can't reliably determine the correct number of
7885 * functions. This can occur if a function has been direct
7886 * attached to a virtual machine using VT-d, for example. In
7887 * this case, simply return -1 to indicate this.
7889 if ((entry
->vendor
!= pdev
->vendor
) ||
7890 (entry
->device
!= pdev
->device
))
7900 * ixgbe_wol_supported - Check whether device supports WoL
7901 * @hw: hw specific details
7902 * @device_id: the device ID
7903 * @subdev_id: the subsystem device ID
7905 * This function is used by probe and ethtool to determine
7906 * which devices have WoL support
7909 int ixgbe_wol_supported(struct ixgbe_adapter
*adapter
, u16 device_id
,
7912 struct ixgbe_hw
*hw
= &adapter
->hw
;
7913 u16 wol_cap
= adapter
->eeprom_cap
& IXGBE_DEVICE_CAPS_WOL_MASK
;
7914 int is_wol_supported
= 0;
7916 switch (device_id
) {
7917 case IXGBE_DEV_ID_82599_SFP
:
7918 /* Only these subdevices could supports WOL */
7919 switch (subdevice_id
) {
7920 case IXGBE_SUBDEV_ID_82599_SFP_WOL0
:
7921 case IXGBE_SUBDEV_ID_82599_560FLR
:
7922 /* only support first port */
7923 if (hw
->bus
.func
!= 0)
7925 case IXGBE_SUBDEV_ID_82599_SP_560FLR
:
7926 case IXGBE_SUBDEV_ID_82599_SFP
:
7927 case IXGBE_SUBDEV_ID_82599_RNDC
:
7928 case IXGBE_SUBDEV_ID_82599_ECNA_DP
:
7929 case IXGBE_SUBDEV_ID_82599_LOM_SFP
:
7930 is_wol_supported
= 1;
7934 case IXGBE_DEV_ID_82599EN_SFP
:
7935 /* Only this subdevice supports WOL */
7936 switch (subdevice_id
) {
7937 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1
:
7938 is_wol_supported
= 1;
7942 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE
:
7943 /* All except this subdevice support WOL */
7944 if (subdevice_id
!= IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ
)
7945 is_wol_supported
= 1;
7947 case IXGBE_DEV_ID_82599_KX4
:
7948 is_wol_supported
= 1;
7950 case IXGBE_DEV_ID_X540T
:
7951 case IXGBE_DEV_ID_X540T1
:
7952 /* check eeprom to see if enabled wol */
7953 if ((wol_cap
== IXGBE_DEVICE_CAPS_WOL_PORT0_1
) ||
7954 ((wol_cap
== IXGBE_DEVICE_CAPS_WOL_PORT0
) &&
7955 (hw
->bus
.func
== 0))) {
7956 is_wol_supported
= 1;
7961 return is_wol_supported
;
7965 * ixgbe_probe - Device Initialization Routine
7966 * @pdev: PCI device information struct
7967 * @ent: entry in ixgbe_pci_tbl
7969 * Returns 0 on success, negative on failure
7971 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7972 * The OS initialization, configuring of the adapter private structure,
7973 * and a hardware reset occur.
7975 static int ixgbe_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
7977 struct net_device
*netdev
;
7978 struct ixgbe_adapter
*adapter
= NULL
;
7979 struct ixgbe_hw
*hw
;
7980 const struct ixgbe_info
*ii
= ixgbe_info_tbl
[ent
->driver_data
];
7981 int i
, err
, pci_using_dac
, expected_gts
;
7982 unsigned int indices
= MAX_TX_QUEUES
;
7983 u8 part_str
[IXGBE_PBANUM_LENGTH
];
7989 /* Catch broken hardware that put the wrong VF device ID in
7990 * the PCIe SR-IOV capability.
7992 if (pdev
->is_virtfn
) {
7993 WARN(1, KERN_ERR
"%s (%hx:%hx) should not be a VF!\n",
7994 pci_name(pdev
), pdev
->vendor
, pdev
->device
);
7998 err
= pci_enable_device_mem(pdev
);
8002 if (!dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(64))) {
8005 err
= dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(32));
8008 "No usable DMA configuration, aborting\n");
8014 err
= pci_request_selected_regions(pdev
, pci_select_bars(pdev
,
8015 IORESOURCE_MEM
), ixgbe_driver_name
);
8018 "pci_request_selected_regions failed 0x%x\n", err
);
8022 pci_enable_pcie_error_reporting(pdev
);
8024 pci_set_master(pdev
);
8025 pci_save_state(pdev
);
8027 if (ii
->mac
== ixgbe_mac_82598EB
) {
8028 #ifdef CONFIG_IXGBE_DCB
8029 /* 8 TC w/ 4 queues per TC */
8030 indices
= 4 * MAX_TRAFFIC_CLASS
;
8032 indices
= IXGBE_MAX_RSS_INDICES
;
8036 netdev
= alloc_etherdev_mq(sizeof(struct ixgbe_adapter
), indices
);
8039 goto err_alloc_etherdev
;
8042 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
8044 adapter
= netdev_priv(netdev
);
8045 pci_set_drvdata(pdev
, adapter
);
8047 adapter
->netdev
= netdev
;
8048 adapter
->pdev
= pdev
;
8051 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
8053 hw
->hw_addr
= ioremap(pci_resource_start(pdev
, 0),
8054 pci_resource_len(pdev
, 0));
8055 adapter
->io_addr
= hw
->hw_addr
;
8061 netdev
->netdev_ops
= &ixgbe_netdev_ops
;
8062 ixgbe_set_ethtool_ops(netdev
);
8063 netdev
->watchdog_timeo
= 5 * HZ
;
8064 strlcpy(netdev
->name
, pci_name(pdev
), sizeof(netdev
->name
));
8067 memcpy(&hw
->mac
.ops
, ii
->mac_ops
, sizeof(hw
->mac
.ops
));
8068 hw
->mac
.type
= ii
->mac
;
8071 memcpy(&hw
->eeprom
.ops
, ii
->eeprom_ops
, sizeof(hw
->eeprom
.ops
));
8072 eec
= IXGBE_READ_REG(hw
, IXGBE_EEC
);
8073 if (ixgbe_removed(hw
->hw_addr
)) {
8077 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
8078 if (!(eec
& (1 << 8)))
8079 hw
->eeprom
.ops
.read
= &ixgbe_read_eeprom_bit_bang_generic
;
8082 memcpy(&hw
->phy
.ops
, ii
->phy_ops
, sizeof(hw
->phy
.ops
));
8083 hw
->phy
.sfp_type
= ixgbe_sfp_type_unknown
;
8084 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
8085 hw
->phy
.mdio
.prtad
= MDIO_PRTAD_NONE
;
8086 hw
->phy
.mdio
.mmds
= 0;
8087 hw
->phy
.mdio
.mode_support
= MDIO_SUPPORTS_C45
| MDIO_EMULATE_C22
;
8088 hw
->phy
.mdio
.dev
= netdev
;
8089 hw
->phy
.mdio
.mdio_read
= ixgbe_mdio_read
;
8090 hw
->phy
.mdio
.mdio_write
= ixgbe_mdio_write
;
8092 ii
->get_invariants(hw
);
8094 /* setup the private structure */
8095 err
= ixgbe_sw_init(adapter
);
8099 /* Make it possible the adapter to be woken up via WOL */
8100 switch (adapter
->hw
.mac
.type
) {
8101 case ixgbe_mac_82599EB
:
8102 case ixgbe_mac_X540
:
8103 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_WUS
, ~0);
8110 * If there is a fan on this device and it has failed log the
8113 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) {
8114 u32 esdp
= IXGBE_READ_REG(hw
, IXGBE_ESDP
);
8115 if (esdp
& IXGBE_ESDP_SDP1
)
8116 e_crit(probe
, "Fan has stopped, replace the adapter\n");
8119 if (allow_unsupported_sfp
)
8120 hw
->allow_unsupported_sfp
= allow_unsupported_sfp
;
8122 /* reset_hw fills in the perm_addr as well */
8123 hw
->phy
.reset_if_overtemp
= true;
8124 err
= hw
->mac
.ops
.reset_hw(hw
);
8125 hw
->phy
.reset_if_overtemp
= false;
8126 if (err
== IXGBE_ERR_SFP_NOT_PRESENT
&&
8127 hw
->mac
.type
== ixgbe_mac_82598EB
) {
8129 } else if (err
== IXGBE_ERR_SFP_NOT_SUPPORTED
) {
8130 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
8131 e_dev_err("Reload the driver after installing a supported module.\n");
8134 e_dev_err("HW Init failed: %d\n", err
);
8138 #ifdef CONFIG_PCI_IOV
8139 /* SR-IOV not supported on the 82598 */
8140 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
)
8143 ixgbe_init_mbx_params_pf(hw
);
8144 memcpy(&hw
->mbx
.ops
, ii
->mbx_ops
, sizeof(hw
->mbx
.ops
));
8145 pci_sriov_set_totalvfs(pdev
, IXGBE_MAX_VFS_DRV_LIMIT
);
8146 ixgbe_enable_sriov(adapter
);
8150 netdev
->features
= NETIF_F_SG
|
8153 NETIF_F_HW_VLAN_CTAG_TX
|
8154 NETIF_F_HW_VLAN_CTAG_RX
|
8155 NETIF_F_HW_VLAN_CTAG_FILTER
|
8161 netdev
->hw_features
= netdev
->features
| NETIF_F_HW_L2FW_DOFFLOAD
;
8163 switch (adapter
->hw
.mac
.type
) {
8164 case ixgbe_mac_82599EB
:
8165 case ixgbe_mac_X540
:
8166 netdev
->features
|= NETIF_F_SCTP_CSUM
;
8167 netdev
->hw_features
|= NETIF_F_SCTP_CSUM
|
8174 netdev
->hw_features
|= NETIF_F_RXALL
;
8176 netdev
->vlan_features
|= NETIF_F_TSO
;
8177 netdev
->vlan_features
|= NETIF_F_TSO6
;
8178 netdev
->vlan_features
|= NETIF_F_IP_CSUM
;
8179 netdev
->vlan_features
|= NETIF_F_IPV6_CSUM
;
8180 netdev
->vlan_features
|= NETIF_F_SG
;
8182 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
8183 netdev
->priv_flags
|= IFF_SUPP_NOFCS
;
8185 #ifdef CONFIG_IXGBE_DCB
8186 netdev
->dcbnl_ops
= &dcbnl_ops
;
8190 if (adapter
->flags
& IXGBE_FLAG_FCOE_CAPABLE
) {
8191 unsigned int fcoe_l
;
8193 if (hw
->mac
.ops
.get_device_caps
) {
8194 hw
->mac
.ops
.get_device_caps(hw
, &device_caps
);
8195 if (device_caps
& IXGBE_DEVICE_CAPS_FCOE_OFFLOADS
)
8196 adapter
->flags
&= ~IXGBE_FLAG_FCOE_CAPABLE
;
8200 fcoe_l
= min_t(int, IXGBE_FCRETA_SIZE
, num_online_cpus());
8201 adapter
->ring_feature
[RING_F_FCOE
].limit
= fcoe_l
;
8203 netdev
->features
|= NETIF_F_FSO
|
8206 netdev
->vlan_features
|= NETIF_F_FSO
|
8210 #endif /* IXGBE_FCOE */
8211 if (pci_using_dac
) {
8212 netdev
->features
|= NETIF_F_HIGHDMA
;
8213 netdev
->vlan_features
|= NETIF_F_HIGHDMA
;
8216 if (adapter
->flags2
& IXGBE_FLAG2_RSC_CAPABLE
)
8217 netdev
->hw_features
|= NETIF_F_LRO
;
8218 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)
8219 netdev
->features
|= NETIF_F_LRO
;
8221 /* make sure the EEPROM is good */
8222 if (hw
->eeprom
.ops
.validate_checksum(hw
, NULL
) < 0) {
8223 e_dev_err("The EEPROM Checksum Is Not Valid\n");
8228 memcpy(netdev
->dev_addr
, hw
->mac
.perm_addr
, netdev
->addr_len
);
8230 if (!is_valid_ether_addr(netdev
->dev_addr
)) {
8231 e_dev_err("invalid MAC address\n");
8236 ixgbe_mac_set_default_filter(adapter
, hw
->mac
.perm_addr
);
8238 setup_timer(&adapter
->service_timer
, &ixgbe_service_timer
,
8239 (unsigned long) adapter
);
8241 if (ixgbe_removed(hw
->hw_addr
)) {
8245 INIT_WORK(&adapter
->service_task
, ixgbe_service_task
);
8246 set_bit(__IXGBE_SERVICE_INITED
, &adapter
->state
);
8247 clear_bit(__IXGBE_SERVICE_SCHED
, &adapter
->state
);
8249 err
= ixgbe_init_interrupt_scheme(adapter
);
8253 /* WOL not supported for all devices */
8255 hw
->eeprom
.ops
.read(hw
, 0x2c, &adapter
->eeprom_cap
);
8256 hw
->wol_enabled
= ixgbe_wol_supported(adapter
, pdev
->device
,
8257 pdev
->subsystem_device
);
8258 if (hw
->wol_enabled
)
8259 adapter
->wol
= IXGBE_WUFC_MAG
;
8261 device_set_wakeup_enable(&adapter
->pdev
->dev
, adapter
->wol
);
8263 /* save off EEPROM version number */
8264 hw
->eeprom
.ops
.read(hw
, 0x2e, &adapter
->eeprom_verh
);
8265 hw
->eeprom
.ops
.read(hw
, 0x2d, &adapter
->eeprom_verl
);
8267 /* pick up the PCI bus settings for reporting later */
8268 hw
->mac
.ops
.get_bus_info(hw
);
8269 if (ixgbe_pcie_from_parent(hw
))
8270 ixgbe_get_parent_bus_info(adapter
);
8272 /* calculate the expected PCIe bandwidth required for optimal
8273 * performance. Note that some older parts will never have enough
8274 * bandwidth due to being older generation PCIe parts. We clamp these
8275 * parts to ensure no warning is displayed if it can't be fixed.
8277 switch (hw
->mac
.type
) {
8278 case ixgbe_mac_82598EB
:
8279 expected_gts
= min(ixgbe_enumerate_functions(adapter
) * 10, 16);
8282 expected_gts
= ixgbe_enumerate_functions(adapter
) * 10;
8286 /* don't check link if we failed to enumerate functions */
8287 if (expected_gts
> 0)
8288 ixgbe_check_minimum_link(adapter
, expected_gts
);
8290 err
= ixgbe_read_pba_string_generic(hw
, part_str
, sizeof(part_str
));
8292 strlcpy(part_str
, "Unknown", sizeof(part_str
));
8293 if (ixgbe_is_sfp(hw
) && hw
->phy
.sfp_type
!= ixgbe_sfp_type_not_present
)
8294 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
8295 hw
->mac
.type
, hw
->phy
.type
, hw
->phy
.sfp_type
,
8298 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
8299 hw
->mac
.type
, hw
->phy
.type
, part_str
);
8301 e_dev_info("%pM\n", netdev
->dev_addr
);
8303 /* reset the hardware with the new settings */
8304 err
= hw
->mac
.ops
.start_hw(hw
);
8305 if (err
== IXGBE_ERR_EEPROM_VERSION
) {
8306 /* We are running on a pre-production device, log a warning */
8307 e_dev_warn("This device is a pre-production adapter/LOM. "
8308 "Please be aware there may be issues associated "
8309 "with your hardware. If you are experiencing "
8310 "problems please contact your Intel or hardware "
8311 "representative who provided you with this "
8314 strcpy(netdev
->name
, "eth%d");
8315 err
= register_netdev(netdev
);
8319 /* power down the optics for 82599 SFP+ fiber */
8320 if (hw
->mac
.ops
.disable_tx_laser
)
8321 hw
->mac
.ops
.disable_tx_laser(hw
);
8323 /* carrier off reporting is important to ethtool even BEFORE open */
8324 netif_carrier_off(netdev
);
8326 #ifdef CONFIG_IXGBE_DCA
8327 if (dca_add_requester(&pdev
->dev
) == 0) {
8328 adapter
->flags
|= IXGBE_FLAG_DCA_ENABLED
;
8329 ixgbe_setup_dca(adapter
);
8332 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
8333 e_info(probe
, "IOV is enabled with %d VFs\n", adapter
->num_vfs
);
8334 for (i
= 0; i
< adapter
->num_vfs
; i
++)
8335 ixgbe_vf_configuration(pdev
, (i
| 0x10000000));
8338 /* firmware requires driver version to be 0xFFFFFFFF
8339 * since os does not support feature
8341 if (hw
->mac
.ops
.set_fw_drv_ver
)
8342 hw
->mac
.ops
.set_fw_drv_ver(hw
, 0xFF, 0xFF, 0xFF,
8345 /* add san mac addr to netdev */
8346 ixgbe_add_sanmac_netdev(netdev
);
8348 e_dev_info("%s\n", ixgbe_default_device_descr
);
8350 #ifdef CONFIG_IXGBE_HWMON
8351 if (ixgbe_sysfs_init(adapter
))
8352 e_err(probe
, "failed to allocate sysfs resources\n");
8353 #endif /* CONFIG_IXGBE_HWMON */
8355 ixgbe_dbg_adapter_init(adapter
);
8357 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
8358 if (ixgbe_mng_enabled(hw
) && ixgbe_is_sfp(hw
) && hw
->mac
.ops
.setup_link
)
8359 hw
->mac
.ops
.setup_link(hw
,
8360 IXGBE_LINK_SPEED_10GB_FULL
| IXGBE_LINK_SPEED_1GB_FULL
,
8366 ixgbe_release_hw_control(adapter
);
8367 ixgbe_clear_interrupt_scheme(adapter
);
8369 ixgbe_disable_sriov(adapter
);
8370 adapter
->flags2
&= ~IXGBE_FLAG2_SEARCH_FOR_SFP
;
8371 iounmap(adapter
->io_addr
);
8372 kfree(adapter
->mac_table
);
8374 free_netdev(netdev
);
8376 pci_release_selected_regions(pdev
,
8377 pci_select_bars(pdev
, IORESOURCE_MEM
));
8380 if (!adapter
|| !test_and_set_bit(__IXGBE_DISABLED
, &adapter
->state
))
8381 pci_disable_device(pdev
);
8386 * ixgbe_remove - Device Removal Routine
8387 * @pdev: PCI device information struct
8389 * ixgbe_remove is called by the PCI subsystem to alert the driver
8390 * that it should release a PCI device. The could be caused by a
8391 * Hot-Plug event, or because the driver is going to be removed from
8394 static void ixgbe_remove(struct pci_dev
*pdev
)
8396 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
8397 struct net_device
*netdev
= adapter
->netdev
;
8399 ixgbe_dbg_adapter_exit(adapter
);
8401 set_bit(__IXGBE_REMOVING
, &adapter
->state
);
8402 cancel_work_sync(&adapter
->service_task
);
8405 #ifdef CONFIG_IXGBE_DCA
8406 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
) {
8407 adapter
->flags
&= ~IXGBE_FLAG_DCA_ENABLED
;
8408 dca_remove_requester(&pdev
->dev
);
8409 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
, 1);
8413 #ifdef CONFIG_IXGBE_HWMON
8414 ixgbe_sysfs_exit(adapter
);
8415 #endif /* CONFIG_IXGBE_HWMON */
8417 /* remove the added san mac */
8418 ixgbe_del_sanmac_netdev(netdev
);
8420 if (netdev
->reg_state
== NETREG_REGISTERED
)
8421 unregister_netdev(netdev
);
8423 #ifdef CONFIG_PCI_IOV
8425 * Only disable SR-IOV on unload if the user specified the now
8426 * deprecated max_vfs module parameter.
8429 ixgbe_disable_sriov(adapter
);
8431 ixgbe_clear_interrupt_scheme(adapter
);
8433 ixgbe_release_hw_control(adapter
);
8436 kfree(adapter
->ixgbe_ieee_pfc
);
8437 kfree(adapter
->ixgbe_ieee_ets
);
8440 iounmap(adapter
->io_addr
);
8441 pci_release_selected_regions(pdev
, pci_select_bars(pdev
,
8444 e_dev_info("complete\n");
8446 kfree(adapter
->mac_table
);
8447 free_netdev(netdev
);
8449 pci_disable_pcie_error_reporting(pdev
);
8451 if (!test_and_set_bit(__IXGBE_DISABLED
, &adapter
->state
))
8452 pci_disable_device(pdev
);
8456 * ixgbe_io_error_detected - called when PCI error is detected
8457 * @pdev: Pointer to PCI device
8458 * @state: The current pci connection state
8460 * This function is called after a PCI bus error affecting
8461 * this device has been detected.
8463 static pci_ers_result_t
ixgbe_io_error_detected(struct pci_dev
*pdev
,
8464 pci_channel_state_t state
)
8466 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
8467 struct net_device
*netdev
= adapter
->netdev
;
8469 #ifdef CONFIG_PCI_IOV
8470 struct ixgbe_hw
*hw
= &adapter
->hw
;
8471 struct pci_dev
*bdev
, *vfdev
;
8472 u32 dw0
, dw1
, dw2
, dw3
;
8474 u16 req_id
, pf_func
;
8476 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
||
8477 adapter
->num_vfs
== 0)
8478 goto skip_bad_vf_detection
;
8480 bdev
= pdev
->bus
->self
;
8481 while (bdev
&& (pci_pcie_type(bdev
) != PCI_EXP_TYPE_ROOT_PORT
))
8482 bdev
= bdev
->bus
->self
;
8485 goto skip_bad_vf_detection
;
8487 pos
= pci_find_ext_capability(bdev
, PCI_EXT_CAP_ID_ERR
);
8489 goto skip_bad_vf_detection
;
8491 dw0
= ixgbe_read_pci_cfg_dword(hw
, pos
+ PCI_ERR_HEADER_LOG
);
8492 dw1
= ixgbe_read_pci_cfg_dword(hw
, pos
+ PCI_ERR_HEADER_LOG
+ 4);
8493 dw2
= ixgbe_read_pci_cfg_dword(hw
, pos
+ PCI_ERR_HEADER_LOG
+ 8);
8494 dw3
= ixgbe_read_pci_cfg_dword(hw
, pos
+ PCI_ERR_HEADER_LOG
+ 12);
8495 if (ixgbe_removed(hw
->hw_addr
))
8496 goto skip_bad_vf_detection
;
8499 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
8500 if (!(req_id
& 0x0080))
8501 goto skip_bad_vf_detection
;
8503 pf_func
= req_id
& 0x01;
8504 if ((pf_func
& 1) == (pdev
->devfn
& 1)) {
8505 unsigned int device_id
;
8507 vf
= (req_id
& 0x7F) >> 1;
8508 e_dev_err("VF %d has caused a PCIe error\n", vf
);
8509 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
8510 "%8.8x\tdw3: %8.8x\n",
8511 dw0
, dw1
, dw2
, dw3
);
8512 switch (adapter
->hw
.mac
.type
) {
8513 case ixgbe_mac_82599EB
:
8514 device_id
= IXGBE_82599_VF_DEVICE_ID
;
8516 case ixgbe_mac_X540
:
8517 device_id
= IXGBE_X540_VF_DEVICE_ID
;
8524 /* Find the pci device of the offending VF */
8525 vfdev
= pci_get_device(PCI_VENDOR_ID_INTEL
, device_id
, NULL
);
8527 if (vfdev
->devfn
== (req_id
& 0xFF))
8529 vfdev
= pci_get_device(PCI_VENDOR_ID_INTEL
,
8533 * There's a slim chance the VF could have been hot plugged,
8534 * so if it is no longer present we don't need to issue the
8535 * VFLR. Just clean up the AER in that case.
8538 e_dev_err("Issuing VFLR to VF %d\n", vf
);
8539 pci_write_config_dword(vfdev
, 0xA8, 0x00008000);
8540 /* Free device reference count */
8544 pci_cleanup_aer_uncorrect_error_status(pdev
);
8548 * Even though the error may have occurred on the other port
8549 * we still need to increment the vf error reference count for
8550 * both ports because the I/O resume function will be called
8553 adapter
->vferr_refcount
++;
8555 return PCI_ERS_RESULT_RECOVERED
;
8557 skip_bad_vf_detection
:
8558 #endif /* CONFIG_PCI_IOV */
8559 if (!test_bit(__IXGBE_SERVICE_INITED
, &adapter
->state
))
8560 return PCI_ERS_RESULT_DISCONNECT
;
8563 netif_device_detach(netdev
);
8565 if (state
== pci_channel_io_perm_failure
) {
8567 return PCI_ERS_RESULT_DISCONNECT
;
8570 if (netif_running(netdev
))
8571 ixgbe_down(adapter
);
8573 if (!test_and_set_bit(__IXGBE_DISABLED
, &adapter
->state
))
8574 pci_disable_device(pdev
);
8577 /* Request a slot reset. */
8578 return PCI_ERS_RESULT_NEED_RESET
;
8582 * ixgbe_io_slot_reset - called after the pci bus has been reset.
8583 * @pdev: Pointer to PCI device
8585 * Restart the card from scratch, as if from a cold-boot.
8587 static pci_ers_result_t
ixgbe_io_slot_reset(struct pci_dev
*pdev
)
8589 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
8590 pci_ers_result_t result
;
8593 if (pci_enable_device_mem(pdev
)) {
8594 e_err(probe
, "Cannot re-enable PCI device after reset.\n");
8595 result
= PCI_ERS_RESULT_DISCONNECT
;
8597 smp_mb__before_atomic();
8598 clear_bit(__IXGBE_DISABLED
, &adapter
->state
);
8599 adapter
->hw
.hw_addr
= adapter
->io_addr
;
8600 pci_set_master(pdev
);
8601 pci_restore_state(pdev
);
8602 pci_save_state(pdev
);
8604 pci_wake_from_d3(pdev
, false);
8606 ixgbe_reset(adapter
);
8607 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_WUS
, ~0);
8608 result
= PCI_ERS_RESULT_RECOVERED
;
8611 err
= pci_cleanup_aer_uncorrect_error_status(pdev
);
8613 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
8614 "failed 0x%0x\n", err
);
8615 /* non-fatal, continue */
8622 * ixgbe_io_resume - called when traffic can start flowing again.
8623 * @pdev: Pointer to PCI device
8625 * This callback is called when the error recovery driver tells us that
8626 * its OK to resume normal operation.
8628 static void ixgbe_io_resume(struct pci_dev
*pdev
)
8630 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
8631 struct net_device
*netdev
= adapter
->netdev
;
8633 #ifdef CONFIG_PCI_IOV
8634 if (adapter
->vferr_refcount
) {
8635 e_info(drv
, "Resuming after VF err\n");
8636 adapter
->vferr_refcount
--;
8641 if (netif_running(netdev
))
8644 netif_device_attach(netdev
);
8647 static const struct pci_error_handlers ixgbe_err_handler
= {
8648 .error_detected
= ixgbe_io_error_detected
,
8649 .slot_reset
= ixgbe_io_slot_reset
,
8650 .resume
= ixgbe_io_resume
,
8653 static struct pci_driver ixgbe_driver
= {
8654 .name
= ixgbe_driver_name
,
8655 .id_table
= ixgbe_pci_tbl
,
8656 .probe
= ixgbe_probe
,
8657 .remove
= ixgbe_remove
,
8659 .suspend
= ixgbe_suspend
,
8660 .resume
= ixgbe_resume
,
8662 .shutdown
= ixgbe_shutdown
,
8663 .sriov_configure
= ixgbe_pci_sriov_configure
,
8664 .err_handler
= &ixgbe_err_handler
8668 * ixgbe_init_module - Driver Registration Routine
8670 * ixgbe_init_module is the first routine called when the driver is
8671 * loaded. All it does is register with the PCI subsystem.
8673 static int __init
ixgbe_init_module(void)
8676 pr_info("%s - version %s\n", ixgbe_driver_string
, ixgbe_driver_version
);
8677 pr_info("%s\n", ixgbe_copyright
);
8681 ret
= pci_register_driver(&ixgbe_driver
);
8687 #ifdef CONFIG_IXGBE_DCA
8688 dca_register_notify(&dca_notifier
);
8694 module_init(ixgbe_init_module
);
8697 * ixgbe_exit_module - Driver Exit Cleanup Routine
8699 * ixgbe_exit_module is called just before the driver is removed
8702 static void __exit
ixgbe_exit_module(void)
8704 #ifdef CONFIG_IXGBE_DCA
8705 dca_unregister_notify(&dca_notifier
);
8707 pci_unregister_driver(&ixgbe_driver
);
8711 rcu_barrier(); /* Wait for completion of call_rcu()'s */
8714 #ifdef CONFIG_IXGBE_DCA
8715 static int ixgbe_notify_dca(struct notifier_block
*nb
, unsigned long event
,
8720 ret_val
= driver_for_each_device(&ixgbe_driver
.driver
, NULL
, &event
,
8721 __ixgbe_notify_dca
);
8723 return ret_val
? NOTIFY_BAD
: NOTIFY_DONE
;
8726 #endif /* CONFIG_IXGBE_DCA */
8728 module_exit(ixgbe_exit_module
);