1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2012 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
35 #include <linux/interrupt.h>
37 #include <linux/tcp.h>
38 #include <linux/sctp.h>
39 #include <linux/pkt_sched.h>
40 #include <linux/ipv6.h>
41 #include <linux/slab.h>
42 #include <net/checksum.h>
43 #include <net/ip6_checksum.h>
44 #include <linux/ethtool.h>
46 #include <linux/if_vlan.h>
47 #include <linux/if_bridge.h>
48 #include <linux/prefetch.h>
49 #include <scsi/fc/fc_fcoe.h>
52 #include "ixgbe_common.h"
53 #include "ixgbe_dcb_82599.h"
54 #include "ixgbe_sriov.h"
56 char ixgbe_driver_name
[] = "ixgbe";
57 static const char ixgbe_driver_string
[] =
58 "Intel(R) 10 Gigabit PCI Express Network Driver";
60 char ixgbe_default_device_descr
[] =
61 "Intel(R) 10 Gigabit Network Connection";
63 static char ixgbe_default_device_descr
[] =
64 "Intel(R) 10 Gigabit Network Connection";
69 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
70 __stringify(BUILD) "-k"
71 const char ixgbe_driver_version
[] = DRV_VERSION
;
72 static const char ixgbe_copyright
[] =
73 "Copyright (c) 1999-2012 Intel Corporation.";
75 static const struct ixgbe_info
*ixgbe_info_tbl
[] = {
76 [board_82598
] = &ixgbe_82598_info
,
77 [board_82599
] = &ixgbe_82599_info
,
78 [board_X540
] = &ixgbe_X540_info
,
81 /* ixgbe_pci_tbl - PCI Device ID Table
83 * Wildcard entries (PCI_ANY_ID) should come last
84 * Last entry must be all 0s
86 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
87 * Class, Class Mask, private data (not used) }
89 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl
) = {
90 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598
), board_82598
},
91 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AF_DUAL_PORT
), board_82598
},
92 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AF_SINGLE_PORT
), board_82598
},
93 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AT
), board_82598
},
94 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AT2
), board_82598
},
95 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598EB_CX4
), board_82598
},
96 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_CX4_DUAL_PORT
), board_82598
},
97 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_DA_DUAL_PORT
), board_82598
},
98 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM
), board_82598
},
99 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598EB_XF_LR
), board_82598
},
100 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598EB_SFP_LOM
), board_82598
},
101 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_BX
), board_82598
},
102 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_KX4
), board_82599
},
103 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_XAUI_LOM
), board_82599
},
104 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_KR
), board_82599
},
105 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP
), board_82599
},
106 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP_EM
), board_82599
},
107 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_KX4_MEZZ
), board_82599
},
108 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_CX4
), board_82599
},
109 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_BACKPLANE_FCOE
), board_82599
},
110 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP_FCOE
), board_82599
},
111 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_T3_LOM
), board_82599
},
112 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_COMBO_BACKPLANE
), board_82599
},
113 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X540T
), board_X540
},
114 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP_SF2
), board_82599
},
115 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_LS
), board_82599
},
116 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599EN_SFP
), board_82599
},
117 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP_SF_QP
), board_82599
},
118 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X540T1
), board_X540
},
119 /* required last entry */
122 MODULE_DEVICE_TABLE(pci
, ixgbe_pci_tbl
);
124 #ifdef CONFIG_IXGBE_DCA
125 static int ixgbe_notify_dca(struct notifier_block
*, unsigned long event
,
127 static struct notifier_block dca_notifier
= {
128 .notifier_call
= ixgbe_notify_dca
,
134 #ifdef CONFIG_PCI_IOV
135 static unsigned int max_vfs
;
136 module_param(max_vfs
, uint
, 0);
137 MODULE_PARM_DESC(max_vfs
,
138 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63");
139 #endif /* CONFIG_PCI_IOV */
141 static unsigned int allow_unsupported_sfp
;
142 module_param(allow_unsupported_sfp
, uint
, 0);
143 MODULE_PARM_DESC(allow_unsupported_sfp
,
144 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
146 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
147 static int debug
= -1;
148 module_param(debug
, int, 0);
149 MODULE_PARM_DESC(debug
, "Debug level (0=none,...,16=all)");
151 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
152 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
153 MODULE_LICENSE("GPL");
154 MODULE_VERSION(DRV_VERSION
);
156 static void ixgbe_service_event_schedule(struct ixgbe_adapter
*adapter
)
158 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
) &&
159 !test_and_set_bit(__IXGBE_SERVICE_SCHED
, &adapter
->state
))
160 schedule_work(&adapter
->service_task
);
163 static void ixgbe_service_event_complete(struct ixgbe_adapter
*adapter
)
165 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED
, &adapter
->state
));
167 /* flush memory to make sure state is correct before next watchdog */
168 smp_mb__before_clear_bit();
169 clear_bit(__IXGBE_SERVICE_SCHED
, &adapter
->state
);
172 struct ixgbe_reg_info
{
177 static const struct ixgbe_reg_info ixgbe_reg_info_tbl
[] = {
179 /* General Registers */
180 {IXGBE_CTRL
, "CTRL"},
181 {IXGBE_STATUS
, "STATUS"},
182 {IXGBE_CTRL_EXT
, "CTRL_EXT"},
184 /* Interrupt Registers */
185 {IXGBE_EICR
, "EICR"},
188 {IXGBE_SRRCTL(0), "SRRCTL"},
189 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
190 {IXGBE_RDLEN(0), "RDLEN"},
191 {IXGBE_RDH(0), "RDH"},
192 {IXGBE_RDT(0), "RDT"},
193 {IXGBE_RXDCTL(0), "RXDCTL"},
194 {IXGBE_RDBAL(0), "RDBAL"},
195 {IXGBE_RDBAH(0), "RDBAH"},
198 {IXGBE_TDBAL(0), "TDBAL"},
199 {IXGBE_TDBAH(0), "TDBAH"},
200 {IXGBE_TDLEN(0), "TDLEN"},
201 {IXGBE_TDH(0), "TDH"},
202 {IXGBE_TDT(0), "TDT"},
203 {IXGBE_TXDCTL(0), "TXDCTL"},
205 /* List Terminator */
211 * ixgbe_regdump - register printout routine
213 static void ixgbe_regdump(struct ixgbe_hw
*hw
, struct ixgbe_reg_info
*reginfo
)
219 switch (reginfo
->ofs
) {
220 case IXGBE_SRRCTL(0):
221 for (i
= 0; i
< 64; i
++)
222 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_SRRCTL(i
));
224 case IXGBE_DCA_RXCTRL(0):
225 for (i
= 0; i
< 64; i
++)
226 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_DCA_RXCTRL(i
));
229 for (i
= 0; i
< 64; i
++)
230 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDLEN(i
));
233 for (i
= 0; i
< 64; i
++)
234 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDH(i
));
237 for (i
= 0; i
< 64; i
++)
238 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDT(i
));
240 case IXGBE_RXDCTL(0):
241 for (i
= 0; i
< 64; i
++)
242 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RXDCTL(i
));
245 for (i
= 0; i
< 64; i
++)
246 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDBAL(i
));
249 for (i
= 0; i
< 64; i
++)
250 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_RDBAH(i
));
253 for (i
= 0; i
< 64; i
++)
254 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDBAL(i
));
257 for (i
= 0; i
< 64; i
++)
258 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDBAH(i
));
261 for (i
= 0; i
< 64; i
++)
262 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDLEN(i
));
265 for (i
= 0; i
< 64; i
++)
266 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDH(i
));
269 for (i
= 0; i
< 64; i
++)
270 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TDT(i
));
272 case IXGBE_TXDCTL(0):
273 for (i
= 0; i
< 64; i
++)
274 regs
[i
] = IXGBE_READ_REG(hw
, IXGBE_TXDCTL(i
));
277 pr_info("%-15s %08x\n", reginfo
->name
,
278 IXGBE_READ_REG(hw
, reginfo
->ofs
));
282 for (i
= 0; i
< 8; i
++) {
283 snprintf(rname
, 16, "%s[%d-%d]", reginfo
->name
, i
*8, i
*8+7);
284 pr_err("%-15s", rname
);
285 for (j
= 0; j
< 8; j
++)
286 pr_cont(" %08x", regs
[i
*8+j
]);
293 * ixgbe_dump - Print registers, tx-rings and rx-rings
295 static void ixgbe_dump(struct ixgbe_adapter
*adapter
)
297 struct net_device
*netdev
= adapter
->netdev
;
298 struct ixgbe_hw
*hw
= &adapter
->hw
;
299 struct ixgbe_reg_info
*reginfo
;
301 struct ixgbe_ring
*tx_ring
;
302 struct ixgbe_tx_buffer
*tx_buffer
;
303 union ixgbe_adv_tx_desc
*tx_desc
;
304 struct my_u0
{ u64 a
; u64 b
; } *u0
;
305 struct ixgbe_ring
*rx_ring
;
306 union ixgbe_adv_rx_desc
*rx_desc
;
307 struct ixgbe_rx_buffer
*rx_buffer_info
;
311 if (!netif_msg_hw(adapter
))
314 /* Print netdevice Info */
316 dev_info(&adapter
->pdev
->dev
, "Net device Info\n");
317 pr_info("Device Name state "
318 "trans_start last_rx\n");
319 pr_info("%-15s %016lX %016lX %016lX\n",
326 /* Print Registers */
327 dev_info(&adapter
->pdev
->dev
, "Register Dump\n");
328 pr_info(" Register Name Value\n");
329 for (reginfo
= (struct ixgbe_reg_info
*)ixgbe_reg_info_tbl
;
330 reginfo
->name
; reginfo
++) {
331 ixgbe_regdump(hw
, reginfo
);
334 /* Print TX Ring Summary */
335 if (!netdev
|| !netif_running(netdev
))
338 dev_info(&adapter
->pdev
->dev
, "TX Rings Summary\n");
339 pr_info(" %s %s %s %s\n",
340 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
341 "leng", "ntw", "timestamp");
342 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
343 tx_ring
= adapter
->tx_ring
[n
];
344 tx_buffer
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_clean
];
345 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
346 n
, tx_ring
->next_to_use
, tx_ring
->next_to_clean
,
347 (u64
)dma_unmap_addr(tx_buffer
, dma
),
348 dma_unmap_len(tx_buffer
, len
),
349 tx_buffer
->next_to_watch
,
350 (u64
)tx_buffer
->time_stamp
);
354 if (!netif_msg_tx_done(adapter
))
355 goto rx_ring_summary
;
357 dev_info(&adapter
->pdev
->dev
, "TX Rings Dump\n");
359 /* Transmit Descriptor Formats
361 * 82598 Advanced Transmit Descriptor
362 * +--------------------------------------------------------------+
363 * 0 | Buffer Address [63:0] |
364 * +--------------------------------------------------------------+
365 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
366 * +--------------------------------------------------------------+
367 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
369 * 82598 Advanced Transmit Descriptor (Write-Back Format)
370 * +--------------------------------------------------------------+
372 * +--------------------------------------------------------------+
373 * 8 | RSV | STA | NXTSEQ |
374 * +--------------------------------------------------------------+
377 * 82599+ Advanced Transmit Descriptor
378 * +--------------------------------------------------------------+
379 * 0 | Buffer Address [63:0] |
380 * +--------------------------------------------------------------+
381 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
382 * +--------------------------------------------------------------+
383 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
385 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
386 * +--------------------------------------------------------------+
388 * +--------------------------------------------------------------+
389 * 8 | RSV | STA | RSV |
390 * +--------------------------------------------------------------+
394 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
395 tx_ring
= adapter
->tx_ring
[n
];
396 pr_info("------------------------------------\n");
397 pr_info("TX QUEUE INDEX = %d\n", tx_ring
->queue_index
);
398 pr_info("------------------------------------\n");
399 pr_info("%s%s %s %s %s %s\n",
400 "T [desc] [address 63:0 ] ",
401 "[PlPOIdStDDt Ln] [bi->dma ] ",
402 "leng", "ntw", "timestamp", "bi->skb");
404 for (i
= 0; tx_ring
->desc
&& (i
< tx_ring
->count
); i
++) {
405 tx_desc
= IXGBE_TX_DESC(tx_ring
, i
);
406 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
407 u0
= (struct my_u0
*)tx_desc
;
408 if (dma_unmap_len(tx_buffer
, len
) > 0) {
409 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p",
413 (u64
)dma_unmap_addr(tx_buffer
, dma
),
414 dma_unmap_len(tx_buffer
, len
),
415 tx_buffer
->next_to_watch
,
416 (u64
)tx_buffer
->time_stamp
,
418 if (i
== tx_ring
->next_to_use
&&
419 i
== tx_ring
->next_to_clean
)
421 else if (i
== tx_ring
->next_to_use
)
423 else if (i
== tx_ring
->next_to_clean
)
428 if (netif_msg_pktdata(adapter
) &&
430 print_hex_dump(KERN_INFO
, "",
431 DUMP_PREFIX_ADDRESS
, 16, 1,
432 tx_buffer
->skb
->data
,
433 dma_unmap_len(tx_buffer
, len
),
439 /* Print RX Rings Summary */
441 dev_info(&adapter
->pdev
->dev
, "RX Rings Summary\n");
442 pr_info("Queue [NTU] [NTC]\n");
443 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
444 rx_ring
= adapter
->rx_ring
[n
];
445 pr_info("%5d %5X %5X\n",
446 n
, rx_ring
->next_to_use
, rx_ring
->next_to_clean
);
450 if (!netif_msg_rx_status(adapter
))
453 dev_info(&adapter
->pdev
->dev
, "RX Rings Dump\n");
455 /* Receive Descriptor Formats
457 * 82598 Advanced Receive Descriptor (Read) Format
459 * +-----------------------------------------------------+
460 * 0 | Packet Buffer Address [63:1] |A0/NSE|
461 * +----------------------------------------------+------+
462 * 8 | Header Buffer Address [63:1] | DD |
463 * +-----------------------------------------------------+
466 * 82598 Advanced Receive Descriptor (Write-Back) Format
468 * 63 48 47 32 31 30 21 20 16 15 4 3 0
469 * +------------------------------------------------------+
470 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
471 * | Packet | IP | | | | Type | Type |
472 * | Checksum | Ident | | | | | |
473 * +------------------------------------------------------+
474 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
475 * +------------------------------------------------------+
476 * 63 48 47 32 31 20 19 0
478 * 82599+ Advanced Receive Descriptor (Read) Format
480 * +-----------------------------------------------------+
481 * 0 | Packet Buffer Address [63:1] |A0/NSE|
482 * +----------------------------------------------+------+
483 * 8 | Header Buffer Address [63:1] | DD |
484 * +-----------------------------------------------------+
487 * 82599+ Advanced Receive Descriptor (Write-Back) Format
489 * 63 48 47 32 31 30 21 20 17 16 4 3 0
490 * +------------------------------------------------------+
491 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
492 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
493 * |/ Flow Dir Flt ID | | | | | |
494 * +------------------------------------------------------+
495 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
496 * +------------------------------------------------------+
497 * 63 48 47 32 31 20 19 0
500 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
501 rx_ring
= adapter
->rx_ring
[n
];
502 pr_info("------------------------------------\n");
503 pr_info("RX QUEUE INDEX = %d\n", rx_ring
->queue_index
);
504 pr_info("------------------------------------\n");
506 "R [desc] [ PktBuf A0] ",
507 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
508 "<-- Adv Rx Read format\n");
510 "RWB[desc] [PcsmIpSHl PtRs] ",
511 "[vl er S cks ln] ---------------- [bi->skb ] ",
512 "<-- Adv Rx Write-Back format\n");
514 for (i
= 0; i
< rx_ring
->count
; i
++) {
515 rx_buffer_info
= &rx_ring
->rx_buffer_info
[i
];
516 rx_desc
= IXGBE_RX_DESC(rx_ring
, i
);
517 u0
= (struct my_u0
*)rx_desc
;
518 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
519 if (staterr
& IXGBE_RXD_STAT_DD
) {
520 /* Descriptor Done */
521 pr_info("RWB[0x%03X] %016llX "
522 "%016llX ---------------- %p", i
,
525 rx_buffer_info
->skb
);
527 pr_info("R [0x%03X] %016llX "
528 "%016llX %016llX %p", i
,
531 (u64
)rx_buffer_info
->dma
,
532 rx_buffer_info
->skb
);
534 if (netif_msg_pktdata(adapter
) &&
535 rx_buffer_info
->dma
) {
536 print_hex_dump(KERN_INFO
, "",
537 DUMP_PREFIX_ADDRESS
, 16, 1,
538 page_address(rx_buffer_info
->page
) +
539 rx_buffer_info
->page_offset
,
540 ixgbe_rx_bufsz(rx_ring
), true);
544 if (i
== rx_ring
->next_to_use
)
546 else if (i
== rx_ring
->next_to_clean
)
558 static void ixgbe_release_hw_control(struct ixgbe_adapter
*adapter
)
562 /* Let firmware take over control of h/w */
563 ctrl_ext
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_CTRL_EXT
);
564 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_CTRL_EXT
,
565 ctrl_ext
& ~IXGBE_CTRL_EXT_DRV_LOAD
);
568 static void ixgbe_get_hw_control(struct ixgbe_adapter
*adapter
)
572 /* Let firmware know the driver has taken over */
573 ctrl_ext
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_CTRL_EXT
);
574 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_CTRL_EXT
,
575 ctrl_ext
| IXGBE_CTRL_EXT_DRV_LOAD
);
579 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
580 * @adapter: pointer to adapter struct
581 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
582 * @queue: queue to map the corresponding interrupt to
583 * @msix_vector: the vector to map to the corresponding queue
586 static void ixgbe_set_ivar(struct ixgbe_adapter
*adapter
, s8 direction
,
587 u8 queue
, u8 msix_vector
)
590 struct ixgbe_hw
*hw
= &adapter
->hw
;
591 switch (hw
->mac
.type
) {
592 case ixgbe_mac_82598EB
:
593 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
596 index
= (((direction
* 64) + queue
) >> 2) & 0x1F;
597 ivar
= IXGBE_READ_REG(hw
, IXGBE_IVAR(index
));
598 ivar
&= ~(0xFF << (8 * (queue
& 0x3)));
599 ivar
|= (msix_vector
<< (8 * (queue
& 0x3)));
600 IXGBE_WRITE_REG(hw
, IXGBE_IVAR(index
), ivar
);
602 case ixgbe_mac_82599EB
:
604 if (direction
== -1) {
606 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
607 index
= ((queue
& 1) * 8);
608 ivar
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_IVAR_MISC
);
609 ivar
&= ~(0xFF << index
);
610 ivar
|= (msix_vector
<< index
);
611 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_IVAR_MISC
, ivar
);
614 /* tx or rx causes */
615 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
616 index
= ((16 * (queue
& 1)) + (8 * direction
));
617 ivar
= IXGBE_READ_REG(hw
, IXGBE_IVAR(queue
>> 1));
618 ivar
&= ~(0xFF << index
);
619 ivar
|= (msix_vector
<< index
);
620 IXGBE_WRITE_REG(hw
, IXGBE_IVAR(queue
>> 1), ivar
);
628 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter
*adapter
,
633 switch (adapter
->hw
.mac
.type
) {
634 case ixgbe_mac_82598EB
:
635 mask
= (IXGBE_EIMS_RTX_QUEUE
& qmask
);
636 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICS
, mask
);
638 case ixgbe_mac_82599EB
:
640 mask
= (qmask
& 0xFFFFFFFF);
641 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICS_EX(0), mask
);
642 mask
= (qmask
>> 32);
643 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICS_EX(1), mask
);
650 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring
*ring
,
651 struct ixgbe_tx_buffer
*tx_buffer
)
653 if (tx_buffer
->skb
) {
654 dev_kfree_skb_any(tx_buffer
->skb
);
655 if (dma_unmap_len(tx_buffer
, len
))
656 dma_unmap_single(ring
->dev
,
657 dma_unmap_addr(tx_buffer
, dma
),
658 dma_unmap_len(tx_buffer
, len
),
660 } else if (dma_unmap_len(tx_buffer
, len
)) {
661 dma_unmap_page(ring
->dev
,
662 dma_unmap_addr(tx_buffer
, dma
),
663 dma_unmap_len(tx_buffer
, len
),
666 tx_buffer
->next_to_watch
= NULL
;
667 tx_buffer
->skb
= NULL
;
668 dma_unmap_len_set(tx_buffer
, len
, 0);
669 /* tx_buffer must be completely set up in the transmit path */
672 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter
*adapter
)
674 struct ixgbe_hw
*hw
= &adapter
->hw
;
675 struct ixgbe_hw_stats
*hwstats
= &adapter
->stats
;
679 if ((hw
->fc
.current_mode
!= ixgbe_fc_full
) &&
680 (hw
->fc
.current_mode
!= ixgbe_fc_rx_pause
))
683 switch (hw
->mac
.type
) {
684 case ixgbe_mac_82598EB
:
685 data
= IXGBE_READ_REG(hw
, IXGBE_LXOFFRXC
);
688 data
= IXGBE_READ_REG(hw
, IXGBE_LXOFFRXCNT
);
690 hwstats
->lxoffrxc
+= data
;
692 /* refill credits (no tx hang) if we received xoff */
696 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
697 clear_bit(__IXGBE_HANG_CHECK_ARMED
,
698 &adapter
->tx_ring
[i
]->state
);
701 static void ixgbe_update_xoff_received(struct ixgbe_adapter
*adapter
)
703 struct ixgbe_hw
*hw
= &adapter
->hw
;
704 struct ixgbe_hw_stats
*hwstats
= &adapter
->stats
;
707 bool pfc_en
= adapter
->dcb_cfg
.pfc_mode_enable
;
709 if (adapter
->ixgbe_ieee_pfc
)
710 pfc_en
|= !!(adapter
->ixgbe_ieee_pfc
->pfc_en
);
712 if (!(adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) || !pfc_en
) {
713 ixgbe_update_xoff_rx_lfc(adapter
);
717 /* update stats for each tc, only valid with PFC enabled */
718 for (i
= 0; i
< MAX_TX_PACKET_BUFFERS
; i
++) {
719 switch (hw
->mac
.type
) {
720 case ixgbe_mac_82598EB
:
721 xoff
[i
] = IXGBE_READ_REG(hw
, IXGBE_PXOFFRXC(i
));
724 xoff
[i
] = IXGBE_READ_REG(hw
, IXGBE_PXOFFRXCNT(i
));
726 hwstats
->pxoffrxc
[i
] += xoff
[i
];
729 /* disarm tx queues that have received xoff frames */
730 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
731 struct ixgbe_ring
*tx_ring
= adapter
->tx_ring
[i
];
732 u8 tc
= tx_ring
->dcb_tc
;
735 clear_bit(__IXGBE_HANG_CHECK_ARMED
, &tx_ring
->state
);
739 static u64
ixgbe_get_tx_completed(struct ixgbe_ring
*ring
)
741 return ring
->stats
.packets
;
744 static u64
ixgbe_get_tx_pending(struct ixgbe_ring
*ring
)
746 struct ixgbe_adapter
*adapter
= netdev_priv(ring
->netdev
);
747 struct ixgbe_hw
*hw
= &adapter
->hw
;
749 u32 head
= IXGBE_READ_REG(hw
, IXGBE_TDH(ring
->reg_idx
));
750 u32 tail
= IXGBE_READ_REG(hw
, IXGBE_TDT(ring
->reg_idx
));
753 return (head
< tail
) ?
754 tail
- head
: (tail
+ ring
->count
- head
);
759 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring
*tx_ring
)
761 u32 tx_done
= ixgbe_get_tx_completed(tx_ring
);
762 u32 tx_done_old
= tx_ring
->tx_stats
.tx_done_old
;
763 u32 tx_pending
= ixgbe_get_tx_pending(tx_ring
);
766 clear_check_for_tx_hang(tx_ring
);
769 * Check for a hung queue, but be thorough. This verifies
770 * that a transmit has been completed since the previous
771 * check AND there is at least one packet pending. The
772 * ARMED bit is set to indicate a potential hang. The
773 * bit is cleared if a pause frame is received to remove
774 * false hang detection due to PFC or 802.3x frames. By
775 * requiring this to fail twice we avoid races with
776 * pfc clearing the ARMED bit and conditions where we
777 * run the check_tx_hang logic with a transmit completion
778 * pending but without time to complete it yet.
780 if ((tx_done_old
== tx_done
) && tx_pending
) {
781 /* make sure it is true for two checks in a row */
782 ret
= test_and_set_bit(__IXGBE_HANG_CHECK_ARMED
,
785 /* update completed stats and continue */
786 tx_ring
->tx_stats
.tx_done_old
= tx_done
;
787 /* reset the countdown */
788 clear_bit(__IXGBE_HANG_CHECK_ARMED
, &tx_ring
->state
);
795 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
796 * @adapter: driver private struct
798 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter
*adapter
)
801 /* Do the reset outside of interrupt context */
802 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
803 adapter
->flags2
|= IXGBE_FLAG2_RESET_REQUESTED
;
804 ixgbe_service_event_schedule(adapter
);
809 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
810 * @q_vector: structure containing interrupt and ring information
811 * @tx_ring: tx ring to clean
813 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector
*q_vector
,
814 struct ixgbe_ring
*tx_ring
)
816 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
817 struct ixgbe_tx_buffer
*tx_buffer
;
818 union ixgbe_adv_tx_desc
*tx_desc
;
819 unsigned int total_bytes
= 0, total_packets
= 0;
820 unsigned int budget
= q_vector
->tx
.work_limit
;
821 unsigned int i
= tx_ring
->next_to_clean
;
823 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
826 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
827 tx_desc
= IXGBE_TX_DESC(tx_ring
, i
);
831 union ixgbe_adv_tx_desc
*eop_desc
= tx_buffer
->next_to_watch
;
833 /* if next_to_watch is not set then there is no work pending */
837 /* prevent any other reads prior to eop_desc */
840 /* if DD is not set pending work has not been completed */
841 if (!(eop_desc
->wb
.status
& cpu_to_le32(IXGBE_TXD_STAT_DD
)))
844 /* clear next_to_watch to prevent false hangs */
845 tx_buffer
->next_to_watch
= NULL
;
847 /* update the statistics for this packet */
848 total_bytes
+= tx_buffer
->bytecount
;
849 total_packets
+= tx_buffer
->gso_segs
;
851 if (unlikely(tx_buffer
->tx_flags
& IXGBE_TX_FLAGS_TSTAMP
))
852 ixgbe_ptp_tx_hwtstamp(q_vector
, tx_buffer
->skb
);
855 dev_kfree_skb_any(tx_buffer
->skb
);
857 /* unmap skb header data */
858 dma_unmap_single(tx_ring
->dev
,
859 dma_unmap_addr(tx_buffer
, dma
),
860 dma_unmap_len(tx_buffer
, len
),
863 /* clear tx_buffer data */
864 tx_buffer
->skb
= NULL
;
865 dma_unmap_len_set(tx_buffer
, len
, 0);
867 /* unmap remaining buffers */
868 while (tx_desc
!= eop_desc
) {
874 tx_buffer
= tx_ring
->tx_buffer_info
;
875 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
878 /* unmap any remaining paged data */
879 if (dma_unmap_len(tx_buffer
, len
)) {
880 dma_unmap_page(tx_ring
->dev
,
881 dma_unmap_addr(tx_buffer
, dma
),
882 dma_unmap_len(tx_buffer
, len
),
884 dma_unmap_len_set(tx_buffer
, len
, 0);
888 /* move us one more past the eop_desc for start of next pkt */
894 tx_buffer
= tx_ring
->tx_buffer_info
;
895 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
898 /* issue prefetch for next Tx descriptor */
901 /* update budget accounting */
903 } while (likely(budget
));
906 tx_ring
->next_to_clean
= i
;
907 u64_stats_update_begin(&tx_ring
->syncp
);
908 tx_ring
->stats
.bytes
+= total_bytes
;
909 tx_ring
->stats
.packets
+= total_packets
;
910 u64_stats_update_end(&tx_ring
->syncp
);
911 q_vector
->tx
.total_bytes
+= total_bytes
;
912 q_vector
->tx
.total_packets
+= total_packets
;
914 if (check_for_tx_hang(tx_ring
) && ixgbe_check_tx_hang(tx_ring
)) {
915 /* schedule immediate reset if we believe we hung */
916 struct ixgbe_hw
*hw
= &adapter
->hw
;
917 e_err(drv
, "Detected Tx Unit Hang\n"
919 " TDH, TDT <%x>, <%x>\n"
920 " next_to_use <%x>\n"
921 " next_to_clean <%x>\n"
922 "tx_buffer_info[next_to_clean]\n"
923 " time_stamp <%lx>\n"
925 tx_ring
->queue_index
,
926 IXGBE_READ_REG(hw
, IXGBE_TDH(tx_ring
->reg_idx
)),
927 IXGBE_READ_REG(hw
, IXGBE_TDT(tx_ring
->reg_idx
)),
928 tx_ring
->next_to_use
, i
,
929 tx_ring
->tx_buffer_info
[i
].time_stamp
, jiffies
);
931 netif_stop_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
934 "tx hang %d detected on queue %d, resetting adapter\n",
935 adapter
->tx_timeout_count
+ 1, tx_ring
->queue_index
);
937 /* schedule immediate reset if we believe we hung */
938 ixgbe_tx_timeout_reset(adapter
);
940 /* the adapter is about to reset, no point in enabling stuff */
944 netdev_tx_completed_queue(txring_txq(tx_ring
),
945 total_packets
, total_bytes
);
947 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
948 if (unlikely(total_packets
&& netif_carrier_ok(tx_ring
->netdev
) &&
949 (ixgbe_desc_unused(tx_ring
) >= TX_WAKE_THRESHOLD
))) {
950 /* Make sure that anybody stopping the queue after this
951 * sees the new next_to_clean.
954 if (__netif_subqueue_stopped(tx_ring
->netdev
,
955 tx_ring
->queue_index
)
956 && !test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
957 netif_wake_subqueue(tx_ring
->netdev
,
958 tx_ring
->queue_index
);
959 ++tx_ring
->tx_stats
.restart_queue
;
966 #ifdef CONFIG_IXGBE_DCA
967 static void ixgbe_update_tx_dca(struct ixgbe_adapter
*adapter
,
968 struct ixgbe_ring
*tx_ring
,
971 struct ixgbe_hw
*hw
= &adapter
->hw
;
972 u32 txctrl
= dca3_get_tag(tx_ring
->dev
, cpu
);
975 switch (hw
->mac
.type
) {
976 case ixgbe_mac_82598EB
:
977 reg_offset
= IXGBE_DCA_TXCTRL(tx_ring
->reg_idx
);
979 case ixgbe_mac_82599EB
:
981 reg_offset
= IXGBE_DCA_TXCTRL_82599(tx_ring
->reg_idx
);
982 txctrl
<<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599
;
985 /* for unknown hardware do not write register */
990 * We can enable relaxed ordering for reads, but not writes when
991 * DCA is enabled. This is due to a known issue in some chipsets
992 * which will cause the DCA tag to be cleared.
994 txctrl
|= IXGBE_DCA_TXCTRL_DESC_RRO_EN
|
995 IXGBE_DCA_TXCTRL_DATA_RRO_EN
|
996 IXGBE_DCA_TXCTRL_DESC_DCA_EN
;
998 IXGBE_WRITE_REG(hw
, reg_offset
, txctrl
);
1001 static void ixgbe_update_rx_dca(struct ixgbe_adapter
*adapter
,
1002 struct ixgbe_ring
*rx_ring
,
1005 struct ixgbe_hw
*hw
= &adapter
->hw
;
1006 u32 rxctrl
= dca3_get_tag(rx_ring
->dev
, cpu
);
1007 u8 reg_idx
= rx_ring
->reg_idx
;
1010 switch (hw
->mac
.type
) {
1011 case ixgbe_mac_82599EB
:
1012 case ixgbe_mac_X540
:
1013 rxctrl
<<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599
;
1020 * We can enable relaxed ordering for reads, but not writes when
1021 * DCA is enabled. This is due to a known issue in some chipsets
1022 * which will cause the DCA tag to be cleared.
1024 rxctrl
|= IXGBE_DCA_RXCTRL_DESC_RRO_EN
|
1025 IXGBE_DCA_RXCTRL_DESC_DCA_EN
;
1027 IXGBE_WRITE_REG(hw
, IXGBE_DCA_RXCTRL(reg_idx
), rxctrl
);
1030 static void ixgbe_update_dca(struct ixgbe_q_vector
*q_vector
)
1032 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1033 struct ixgbe_ring
*ring
;
1034 int cpu
= get_cpu();
1036 if (q_vector
->cpu
== cpu
)
1039 ixgbe_for_each_ring(ring
, q_vector
->tx
)
1040 ixgbe_update_tx_dca(adapter
, ring
, cpu
);
1042 ixgbe_for_each_ring(ring
, q_vector
->rx
)
1043 ixgbe_update_rx_dca(adapter
, ring
, cpu
);
1045 q_vector
->cpu
= cpu
;
1050 static void ixgbe_setup_dca(struct ixgbe_adapter
*adapter
)
1054 if (!(adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
))
1057 /* always use CB2 mode, difference is masked in the CB driver */
1058 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
, 2);
1060 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
1061 adapter
->q_vector
[i
]->cpu
= -1;
1062 ixgbe_update_dca(adapter
->q_vector
[i
]);
1066 static int __ixgbe_notify_dca(struct device
*dev
, void *data
)
1068 struct ixgbe_adapter
*adapter
= dev_get_drvdata(dev
);
1069 unsigned long event
= *(unsigned long *)data
;
1071 if (!(adapter
->flags
& IXGBE_FLAG_DCA_CAPABLE
))
1075 case DCA_PROVIDER_ADD
:
1076 /* if we're already enabled, don't do it again */
1077 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
1079 if (dca_add_requester(dev
) == 0) {
1080 adapter
->flags
|= IXGBE_FLAG_DCA_ENABLED
;
1081 ixgbe_setup_dca(adapter
);
1084 /* Fall Through since DCA is disabled. */
1085 case DCA_PROVIDER_REMOVE
:
1086 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
) {
1087 dca_remove_requester(dev
);
1088 adapter
->flags
&= ~IXGBE_FLAG_DCA_ENABLED
;
1089 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
, 1);
1097 #endif /* CONFIG_IXGBE_DCA */
1098 static inline void ixgbe_rx_hash(struct ixgbe_ring
*ring
,
1099 union ixgbe_adv_rx_desc
*rx_desc
,
1100 struct sk_buff
*skb
)
1102 if (ring
->netdev
->features
& NETIF_F_RXHASH
)
1103 skb
->rxhash
= le32_to_cpu(rx_desc
->wb
.lower
.hi_dword
.rss
);
1108 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1109 * @ring: structure containing ring specific data
1110 * @rx_desc: advanced rx descriptor
1112 * Returns : true if it is FCoE pkt
1114 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring
*ring
,
1115 union ixgbe_adv_rx_desc
*rx_desc
)
1117 __le16 pkt_info
= rx_desc
->wb
.lower
.lo_dword
.hs_rss
.pkt_info
;
1119 return test_bit(__IXGBE_RX_FCOE
, &ring
->state
) &&
1120 ((pkt_info
& cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK
)) ==
1121 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE
<<
1122 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT
)));
1125 #endif /* IXGBE_FCOE */
1127 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1128 * @ring: structure containing ring specific data
1129 * @rx_desc: current Rx descriptor being processed
1130 * @skb: skb currently being received and modified
1132 static inline void ixgbe_rx_checksum(struct ixgbe_ring
*ring
,
1133 union ixgbe_adv_rx_desc
*rx_desc
,
1134 struct sk_buff
*skb
)
1136 skb_checksum_none_assert(skb
);
1138 /* Rx csum disabled */
1139 if (!(ring
->netdev
->features
& NETIF_F_RXCSUM
))
1142 /* if IP and error */
1143 if (ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_IPCS
) &&
1144 ixgbe_test_staterr(rx_desc
, IXGBE_RXDADV_ERR_IPE
)) {
1145 ring
->rx_stats
.csum_err
++;
1149 if (!ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_L4CS
))
1152 if (ixgbe_test_staterr(rx_desc
, IXGBE_RXDADV_ERR_TCPE
)) {
1153 __le16 pkt_info
= rx_desc
->wb
.lower
.lo_dword
.hs_rss
.pkt_info
;
1156 * 82599 errata, UDP frames with a 0 checksum can be marked as
1159 if ((pkt_info
& cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP
)) &&
1160 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR
, &ring
->state
))
1163 ring
->rx_stats
.csum_err
++;
1167 /* It must be a TCP or UDP packet with a valid checksum */
1168 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1171 static inline void ixgbe_release_rx_desc(struct ixgbe_ring
*rx_ring
, u32 val
)
1173 rx_ring
->next_to_use
= val
;
1175 /* update next to alloc since we have filled the ring */
1176 rx_ring
->next_to_alloc
= val
;
1178 * Force memory writes to complete before letting h/w
1179 * know there are new descriptors to fetch. (Only
1180 * applicable for weak-ordered memory model archs,
1184 writel(val
, rx_ring
->tail
);
1187 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring
*rx_ring
,
1188 struct ixgbe_rx_buffer
*bi
)
1190 struct page
*page
= bi
->page
;
1191 dma_addr_t dma
= bi
->dma
;
1193 /* since we are recycling buffers we should seldom need to alloc */
1197 /* alloc new page for storage */
1198 if (likely(!page
)) {
1199 page
= __skb_alloc_pages(GFP_ATOMIC
| __GFP_COLD
| __GFP_COMP
,
1200 bi
->skb
, ixgbe_rx_pg_order(rx_ring
));
1201 if (unlikely(!page
)) {
1202 rx_ring
->rx_stats
.alloc_rx_page_failed
++;
1208 /* map page for use */
1209 dma
= dma_map_page(rx_ring
->dev
, page
, 0,
1210 ixgbe_rx_pg_size(rx_ring
), DMA_FROM_DEVICE
);
1213 * if mapping failed free memory back to system since
1214 * there isn't much point in holding memory we can't use
1216 if (dma_mapping_error(rx_ring
->dev
, dma
)) {
1217 __free_pages(page
, ixgbe_rx_pg_order(rx_ring
));
1220 rx_ring
->rx_stats
.alloc_rx_page_failed
++;
1225 bi
->page_offset
= 0;
1231 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1232 * @rx_ring: ring to place buffers on
1233 * @cleaned_count: number of buffers to replace
1235 void ixgbe_alloc_rx_buffers(struct ixgbe_ring
*rx_ring
, u16 cleaned_count
)
1237 union ixgbe_adv_rx_desc
*rx_desc
;
1238 struct ixgbe_rx_buffer
*bi
;
1239 u16 i
= rx_ring
->next_to_use
;
1245 rx_desc
= IXGBE_RX_DESC(rx_ring
, i
);
1246 bi
= &rx_ring
->rx_buffer_info
[i
];
1247 i
-= rx_ring
->count
;
1250 if (!ixgbe_alloc_mapped_page(rx_ring
, bi
))
1254 * Refresh the desc even if buffer_addrs didn't change
1255 * because each write-back erases this info.
1257 rx_desc
->read
.pkt_addr
= cpu_to_le64(bi
->dma
+ bi
->page_offset
);
1263 rx_desc
= IXGBE_RX_DESC(rx_ring
, 0);
1264 bi
= rx_ring
->rx_buffer_info
;
1265 i
-= rx_ring
->count
;
1268 /* clear the hdr_addr for the next_to_use descriptor */
1269 rx_desc
->read
.hdr_addr
= 0;
1272 } while (cleaned_count
);
1274 i
+= rx_ring
->count
;
1276 if (rx_ring
->next_to_use
!= i
)
1277 ixgbe_release_rx_desc(rx_ring
, i
);
1281 * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1282 * @data: pointer to the start of the headers
1283 * @max_len: total length of section to find headers in
1285 * This function is meant to determine the length of headers that will
1286 * be recognized by hardware for LRO, GRO, and RSC offloads. The main
1287 * motivation of doing this is to only perform one pull for IPv4 TCP
1288 * packets so that we can do basic things like calculating the gso_size
1289 * based on the average data per packet.
1291 static unsigned int ixgbe_get_headlen(unsigned char *data
,
1292 unsigned int max_len
)
1295 unsigned char *network
;
1298 struct vlan_hdr
*vlan
;
1301 struct ipv6hdr
*ipv6
;
1304 u8 nexthdr
= 0; /* default to not TCP */
1307 /* this should never happen, but better safe than sorry */
1308 if (max_len
< ETH_HLEN
)
1311 /* initialize network frame pointer */
1314 /* set first protocol and move network header forward */
1315 protocol
= hdr
.eth
->h_proto
;
1316 hdr
.network
+= ETH_HLEN
;
1318 /* handle any vlan tag if present */
1319 if (protocol
== __constant_htons(ETH_P_8021Q
)) {
1320 if ((hdr
.network
- data
) > (max_len
- VLAN_HLEN
))
1323 protocol
= hdr
.vlan
->h_vlan_encapsulated_proto
;
1324 hdr
.network
+= VLAN_HLEN
;
1327 /* handle L3 protocols */
1328 if (protocol
== __constant_htons(ETH_P_IP
)) {
1329 if ((hdr
.network
- data
) > (max_len
- sizeof(struct iphdr
)))
1332 /* access ihl as a u8 to avoid unaligned access on ia64 */
1333 hlen
= (hdr
.network
[0] & 0x0F) << 2;
1335 /* verify hlen meets minimum size requirements */
1336 if (hlen
< sizeof(struct iphdr
))
1337 return hdr
.network
- data
;
1339 /* record next protocol */
1340 nexthdr
= hdr
.ipv4
->protocol
;
1341 hdr
.network
+= hlen
;
1342 } else if (protocol
== __constant_htons(ETH_P_IPV6
)) {
1343 if ((hdr
.network
- data
) > (max_len
- sizeof(struct ipv6hdr
)))
1346 /* record next protocol */
1347 nexthdr
= hdr
.ipv6
->nexthdr
;
1348 hdr
.network
+= sizeof(struct ipv6hdr
);
1350 } else if (protocol
== __constant_htons(ETH_P_FCOE
)) {
1351 if ((hdr
.network
- data
) > (max_len
- FCOE_HEADER_LEN
))
1353 hdr
.network
+= FCOE_HEADER_LEN
;
1356 return hdr
.network
- data
;
1359 /* finally sort out TCP/UDP */
1360 if (nexthdr
== IPPROTO_TCP
) {
1361 if ((hdr
.network
- data
) > (max_len
- sizeof(struct tcphdr
)))
1364 /* access doff as a u8 to avoid unaligned access on ia64 */
1365 hlen
= (hdr
.network
[12] & 0xF0) >> 2;
1367 /* verify hlen meets minimum size requirements */
1368 if (hlen
< sizeof(struct tcphdr
))
1369 return hdr
.network
- data
;
1371 hdr
.network
+= hlen
;
1372 } else if (nexthdr
== IPPROTO_UDP
) {
1373 if ((hdr
.network
- data
) > (max_len
- sizeof(struct udphdr
)))
1376 hdr
.network
+= sizeof(struct udphdr
);
1380 * If everything has gone correctly hdr.network should be the
1381 * data section of the packet and will be the end of the header.
1382 * If not then it probably represents the end of the last recognized
1385 if ((hdr
.network
- data
) < max_len
)
1386 return hdr
.network
- data
;
1391 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring
*ring
,
1392 struct sk_buff
*skb
)
1394 u16 hdr_len
= skb_headlen(skb
);
1396 /* set gso_size to avoid messing up TCP MSS */
1397 skb_shinfo(skb
)->gso_size
= DIV_ROUND_UP((skb
->len
- hdr_len
),
1398 IXGBE_CB(skb
)->append_cnt
);
1401 static void ixgbe_update_rsc_stats(struct ixgbe_ring
*rx_ring
,
1402 struct sk_buff
*skb
)
1404 /* if append_cnt is 0 then frame is not RSC */
1405 if (!IXGBE_CB(skb
)->append_cnt
)
1408 rx_ring
->rx_stats
.rsc_count
+= IXGBE_CB(skb
)->append_cnt
;
1409 rx_ring
->rx_stats
.rsc_flush
++;
1411 ixgbe_set_rsc_gso_size(rx_ring
, skb
);
1413 /* gso_size is computed using append_cnt so always clear it last */
1414 IXGBE_CB(skb
)->append_cnt
= 0;
1418 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1419 * @rx_ring: rx descriptor ring packet is being transacted on
1420 * @rx_desc: pointer to the EOP Rx descriptor
1421 * @skb: pointer to current skb being populated
1423 * This function checks the ring, descriptor, and packet information in
1424 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1425 * other fields within the skb.
1427 static void ixgbe_process_skb_fields(struct ixgbe_ring
*rx_ring
,
1428 union ixgbe_adv_rx_desc
*rx_desc
,
1429 struct sk_buff
*skb
)
1431 struct net_device
*dev
= rx_ring
->netdev
;
1433 ixgbe_update_rsc_stats(rx_ring
, skb
);
1435 ixgbe_rx_hash(rx_ring
, rx_desc
, skb
);
1437 ixgbe_rx_checksum(rx_ring
, rx_desc
, skb
);
1439 ixgbe_ptp_rx_hwtstamp(rx_ring
->q_vector
, rx_desc
, skb
);
1441 if ((dev
->features
& NETIF_F_HW_VLAN_RX
) &&
1442 ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_VP
)) {
1443 u16 vid
= le16_to_cpu(rx_desc
->wb
.upper
.vlan
);
1444 __vlan_hwaccel_put_tag(skb
, vid
);
1447 skb_record_rx_queue(skb
, rx_ring
->queue_index
);
1449 skb
->protocol
= eth_type_trans(skb
, dev
);
1452 static void ixgbe_rx_skb(struct ixgbe_q_vector
*q_vector
,
1453 struct sk_buff
*skb
)
1455 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1457 if (!(adapter
->flags
& IXGBE_FLAG_IN_NETPOLL
))
1458 napi_gro_receive(&q_vector
->napi
, skb
);
1464 * ixgbe_is_non_eop - process handling of non-EOP buffers
1465 * @rx_ring: Rx ring being processed
1466 * @rx_desc: Rx descriptor for current buffer
1467 * @skb: Current socket buffer containing buffer in progress
1469 * This function updates next to clean. If the buffer is an EOP buffer
1470 * this function exits returning false, otherwise it will place the
1471 * sk_buff in the next buffer to be chained and return true indicating
1472 * that this is in fact a non-EOP buffer.
1474 static bool ixgbe_is_non_eop(struct ixgbe_ring
*rx_ring
,
1475 union ixgbe_adv_rx_desc
*rx_desc
,
1476 struct sk_buff
*skb
)
1478 u32 ntc
= rx_ring
->next_to_clean
+ 1;
1480 /* fetch, update, and store next to clean */
1481 ntc
= (ntc
< rx_ring
->count
) ? ntc
: 0;
1482 rx_ring
->next_to_clean
= ntc
;
1484 prefetch(IXGBE_RX_DESC(rx_ring
, ntc
));
1486 /* update RSC append count if present */
1487 if (ring_is_rsc_enabled(rx_ring
)) {
1488 __le32 rsc_enabled
= rx_desc
->wb
.lower
.lo_dword
.data
&
1489 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK
);
1491 if (unlikely(rsc_enabled
)) {
1492 u32 rsc_cnt
= le32_to_cpu(rsc_enabled
);
1494 rsc_cnt
>>= IXGBE_RXDADV_RSCCNT_SHIFT
;
1495 IXGBE_CB(skb
)->append_cnt
+= rsc_cnt
- 1;
1497 /* update ntc based on RSC value */
1498 ntc
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
1499 ntc
&= IXGBE_RXDADV_NEXTP_MASK
;
1500 ntc
>>= IXGBE_RXDADV_NEXTP_SHIFT
;
1504 /* if we are the last buffer then there is nothing else to do */
1505 if (likely(ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_EOP
)))
1508 /* place skb in next buffer to be received */
1509 rx_ring
->rx_buffer_info
[ntc
].skb
= skb
;
1510 rx_ring
->rx_stats
.non_eop_descs
++;
1516 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1517 * @rx_ring: rx descriptor ring packet is being transacted on
1518 * @skb: pointer to current skb being adjusted
1520 * This function is an ixgbe specific version of __pskb_pull_tail. The
1521 * main difference between this version and the original function is that
1522 * this function can make several assumptions about the state of things
1523 * that allow for significant optimizations versus the standard function.
1524 * As a result we can do things like drop a frag and maintain an accurate
1525 * truesize for the skb.
1527 static void ixgbe_pull_tail(struct ixgbe_ring
*rx_ring
,
1528 struct sk_buff
*skb
)
1530 struct skb_frag_struct
*frag
= &skb_shinfo(skb
)->frags
[0];
1532 unsigned int pull_len
;
1535 * it is valid to use page_address instead of kmap since we are
1536 * working with pages allocated out of the lomem pool per
1537 * alloc_page(GFP_ATOMIC)
1539 va
= skb_frag_address(frag
);
1542 * we need the header to contain the greater of either ETH_HLEN or
1543 * 60 bytes if the skb->len is less than 60 for skb_pad.
1545 pull_len
= ixgbe_get_headlen(va
, IXGBE_RX_HDR_SIZE
);
1547 /* align pull length to size of long to optimize memcpy performance */
1548 skb_copy_to_linear_data(skb
, va
, ALIGN(pull_len
, sizeof(long)));
1550 /* update all of the pointers */
1551 skb_frag_size_sub(frag
, pull_len
);
1552 frag
->page_offset
+= pull_len
;
1553 skb
->data_len
-= pull_len
;
1554 skb
->tail
+= pull_len
;
1558 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1559 * @rx_ring: rx descriptor ring packet is being transacted on
1560 * @skb: pointer to current skb being updated
1562 * This function provides a basic DMA sync up for the first fragment of an
1563 * skb. The reason for doing this is that the first fragment cannot be
1564 * unmapped until we have reached the end of packet descriptor for a buffer
1567 static void ixgbe_dma_sync_frag(struct ixgbe_ring
*rx_ring
,
1568 struct sk_buff
*skb
)
1570 /* if the page was released unmap it, else just sync our portion */
1571 if (unlikely(IXGBE_CB(skb
)->page_released
)) {
1572 dma_unmap_page(rx_ring
->dev
, IXGBE_CB(skb
)->dma
,
1573 ixgbe_rx_pg_size(rx_ring
), DMA_FROM_DEVICE
);
1574 IXGBE_CB(skb
)->page_released
= false;
1576 struct skb_frag_struct
*frag
= &skb_shinfo(skb
)->frags
[0];
1578 dma_sync_single_range_for_cpu(rx_ring
->dev
,
1581 ixgbe_rx_bufsz(rx_ring
),
1584 IXGBE_CB(skb
)->dma
= 0;
1588 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1589 * @rx_ring: rx descriptor ring packet is being transacted on
1590 * @rx_desc: pointer to the EOP Rx descriptor
1591 * @skb: pointer to current skb being fixed
1593 * Check for corrupted packet headers caused by senders on the local L2
1594 * embedded NIC switch not setting up their Tx Descriptors right. These
1595 * should be very rare.
1597 * Also address the case where we are pulling data in on pages only
1598 * and as such no data is present in the skb header.
1600 * In addition if skb is not at least 60 bytes we need to pad it so that
1601 * it is large enough to qualify as a valid Ethernet frame.
1603 * Returns true if an error was encountered and skb was freed.
1605 static bool ixgbe_cleanup_headers(struct ixgbe_ring
*rx_ring
,
1606 union ixgbe_adv_rx_desc
*rx_desc
,
1607 struct sk_buff
*skb
)
1609 struct net_device
*netdev
= rx_ring
->netdev
;
1611 /* verify that the packet does not have any known errors */
1612 if (unlikely(ixgbe_test_staterr(rx_desc
,
1613 IXGBE_RXDADV_ERR_FRAME_ERR_MASK
) &&
1614 !(netdev
->features
& NETIF_F_RXALL
))) {
1615 dev_kfree_skb_any(skb
);
1619 /* place header in linear portion of buffer */
1620 if (skb_is_nonlinear(skb
))
1621 ixgbe_pull_tail(rx_ring
, skb
);
1624 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1625 if (ixgbe_rx_is_fcoe(rx_ring
, rx_desc
))
1629 /* if skb_pad returns an error the skb was freed */
1630 if (unlikely(skb
->len
< 60)) {
1631 int pad_len
= 60 - skb
->len
;
1633 if (skb_pad(skb
, pad_len
))
1635 __skb_put(skb
, pad_len
);
1642 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1643 * @rx_ring: rx descriptor ring to store buffers on
1644 * @old_buff: donor buffer to have page reused
1646 * Synchronizes page for reuse by the adapter
1648 static void ixgbe_reuse_rx_page(struct ixgbe_ring
*rx_ring
,
1649 struct ixgbe_rx_buffer
*old_buff
)
1651 struct ixgbe_rx_buffer
*new_buff
;
1652 u16 nta
= rx_ring
->next_to_alloc
;
1654 new_buff
= &rx_ring
->rx_buffer_info
[nta
];
1656 /* update, and store next to alloc */
1658 rx_ring
->next_to_alloc
= (nta
< rx_ring
->count
) ? nta
: 0;
1660 /* transfer page from old buffer to new buffer */
1661 new_buff
->page
= old_buff
->page
;
1662 new_buff
->dma
= old_buff
->dma
;
1663 new_buff
->page_offset
= old_buff
->page_offset
;
1665 /* sync the buffer for use by the device */
1666 dma_sync_single_range_for_device(rx_ring
->dev
, new_buff
->dma
,
1667 new_buff
->page_offset
,
1668 ixgbe_rx_bufsz(rx_ring
),
1673 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1674 * @rx_ring: rx descriptor ring to transact packets on
1675 * @rx_buffer: buffer containing page to add
1676 * @rx_desc: descriptor containing length of buffer written by hardware
1677 * @skb: sk_buff to place the data into
1679 * This function will add the data contained in rx_buffer->page to the skb.
1680 * This is done either through a direct copy if the data in the buffer is
1681 * less than the skb header size, otherwise it will just attach the page as
1682 * a frag to the skb.
1684 * The function will then update the page offset if necessary and return
1685 * true if the buffer can be reused by the adapter.
1687 static bool ixgbe_add_rx_frag(struct ixgbe_ring
*rx_ring
,
1688 struct ixgbe_rx_buffer
*rx_buffer
,
1689 union ixgbe_adv_rx_desc
*rx_desc
,
1690 struct sk_buff
*skb
)
1692 struct page
*page
= rx_buffer
->page
;
1693 unsigned int size
= le16_to_cpu(rx_desc
->wb
.upper
.length
);
1694 #if (PAGE_SIZE < 8192)
1695 unsigned int truesize
= ixgbe_rx_bufsz(rx_ring
);
1697 unsigned int truesize
= ALIGN(size
, L1_CACHE_BYTES
);
1698 unsigned int last_offset
= ixgbe_rx_pg_size(rx_ring
) -
1699 ixgbe_rx_bufsz(rx_ring
);
1702 if ((size
<= IXGBE_RX_HDR_SIZE
) && !skb_is_nonlinear(skb
)) {
1703 unsigned char *va
= page_address(page
) + rx_buffer
->page_offset
;
1705 memcpy(__skb_put(skb
, size
), va
, ALIGN(size
, sizeof(long)));
1707 /* we can reuse buffer as-is, just make sure it is local */
1708 if (likely(page_to_nid(page
) == numa_node_id()))
1711 /* this page cannot be reused so discard it */
1716 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
,
1717 rx_buffer
->page_offset
, size
, truesize
);
1719 /* avoid re-using remote pages */
1720 if (unlikely(page_to_nid(page
) != numa_node_id()))
1723 #if (PAGE_SIZE < 8192)
1724 /* if we are only owner of page we can reuse it */
1725 if (unlikely(page_count(page
) != 1))
1728 /* flip page offset to other buffer */
1729 rx_buffer
->page_offset
^= truesize
;
1732 * since we are the only owner of the page and we need to
1733 * increment it, just set the value to 2 in order to avoid
1734 * an unecessary locked operation
1736 atomic_set(&page
->_count
, 2);
1738 /* move offset up to the next cache line */
1739 rx_buffer
->page_offset
+= truesize
;
1741 if (rx_buffer
->page_offset
> last_offset
)
1744 /* bump ref count on page before it is given to the stack */
1751 static struct sk_buff
*ixgbe_fetch_rx_buffer(struct ixgbe_ring
*rx_ring
,
1752 union ixgbe_adv_rx_desc
*rx_desc
)
1754 struct ixgbe_rx_buffer
*rx_buffer
;
1755 struct sk_buff
*skb
;
1758 rx_buffer
= &rx_ring
->rx_buffer_info
[rx_ring
->next_to_clean
];
1759 page
= rx_buffer
->page
;
1762 skb
= rx_buffer
->skb
;
1765 void *page_addr
= page_address(page
) +
1766 rx_buffer
->page_offset
;
1768 /* prefetch first cache line of first page */
1769 prefetch(page_addr
);
1770 #if L1_CACHE_BYTES < 128
1771 prefetch(page_addr
+ L1_CACHE_BYTES
);
1774 /* allocate a skb to store the frags */
1775 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
1777 if (unlikely(!skb
)) {
1778 rx_ring
->rx_stats
.alloc_rx_buff_failed
++;
1783 * we will be copying header into skb->data in
1784 * pskb_may_pull so it is in our interest to prefetch
1785 * it now to avoid a possible cache miss
1787 prefetchw(skb
->data
);
1790 * Delay unmapping of the first packet. It carries the
1791 * header information, HW may still access the header
1792 * after the writeback. Only unmap it when EOP is
1795 if (likely(ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_EOP
)))
1798 IXGBE_CB(skb
)->dma
= rx_buffer
->dma
;
1800 if (ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_EOP
))
1801 ixgbe_dma_sync_frag(rx_ring
, skb
);
1804 /* we are reusing so sync this buffer for CPU use */
1805 dma_sync_single_range_for_cpu(rx_ring
->dev
,
1807 rx_buffer
->page_offset
,
1808 ixgbe_rx_bufsz(rx_ring
),
1812 /* pull page into skb */
1813 if (ixgbe_add_rx_frag(rx_ring
, rx_buffer
, rx_desc
, skb
)) {
1814 /* hand second half of page back to the ring */
1815 ixgbe_reuse_rx_page(rx_ring
, rx_buffer
);
1816 } else if (IXGBE_CB(skb
)->dma
== rx_buffer
->dma
) {
1817 /* the page has been released from the ring */
1818 IXGBE_CB(skb
)->page_released
= true;
1820 /* we are not reusing the buffer so unmap it */
1821 dma_unmap_page(rx_ring
->dev
, rx_buffer
->dma
,
1822 ixgbe_rx_pg_size(rx_ring
),
1826 /* clear contents of buffer_info */
1827 rx_buffer
->skb
= NULL
;
1829 rx_buffer
->page
= NULL
;
1835 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1836 * @q_vector: structure containing interrupt and ring information
1837 * @rx_ring: rx descriptor ring to transact packets on
1838 * @budget: Total limit on number of packets to process
1840 * This function provides a "bounce buffer" approach to Rx interrupt
1841 * processing. The advantage to this is that on systems that have
1842 * expensive overhead for IOMMU access this provides a means of avoiding
1843 * it by maintaining the mapping of the page to the syste.
1845 * Returns true if all work is completed without reaching budget
1847 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector
*q_vector
,
1848 struct ixgbe_ring
*rx_ring
,
1851 unsigned int total_rx_bytes
= 0, total_rx_packets
= 0;
1853 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1855 unsigned int mss
= 0;
1856 #endif /* IXGBE_FCOE */
1857 u16 cleaned_count
= ixgbe_desc_unused(rx_ring
);
1860 union ixgbe_adv_rx_desc
*rx_desc
;
1861 struct sk_buff
*skb
;
1863 /* return some buffers to hardware, one at a time is too slow */
1864 if (cleaned_count
>= IXGBE_RX_BUFFER_WRITE
) {
1865 ixgbe_alloc_rx_buffers(rx_ring
, cleaned_count
);
1869 rx_desc
= IXGBE_RX_DESC(rx_ring
, rx_ring
->next_to_clean
);
1871 if (!ixgbe_test_staterr(rx_desc
, IXGBE_RXD_STAT_DD
))
1875 * This memory barrier is needed to keep us from reading
1876 * any other fields out of the rx_desc until we know the
1877 * RXD_STAT_DD bit is set
1881 /* retrieve a buffer from the ring */
1882 skb
= ixgbe_fetch_rx_buffer(rx_ring
, rx_desc
);
1884 /* exit if we failed to retrieve a buffer */
1890 /* place incomplete frames back on ring for completion */
1891 if (ixgbe_is_non_eop(rx_ring
, rx_desc
, skb
))
1894 /* verify the packet layout is correct */
1895 if (ixgbe_cleanup_headers(rx_ring
, rx_desc
, skb
))
1898 /* probably a little skewed due to removing CRC */
1899 total_rx_bytes
+= skb
->len
;
1901 /* populate checksum, timestamp, VLAN, and protocol */
1902 ixgbe_process_skb_fields(rx_ring
, rx_desc
, skb
);
1905 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1906 if (ixgbe_rx_is_fcoe(rx_ring
, rx_desc
)) {
1907 ddp_bytes
= ixgbe_fcoe_ddp(adapter
, rx_desc
, skb
);
1908 /* include DDPed FCoE data */
1909 if (ddp_bytes
> 0) {
1911 mss
= rx_ring
->netdev
->mtu
-
1912 sizeof(struct fcoe_hdr
) -
1913 sizeof(struct fc_frame_header
) -
1914 sizeof(struct fcoe_crc_eof
);
1918 total_rx_bytes
+= ddp_bytes
;
1919 total_rx_packets
+= DIV_ROUND_UP(ddp_bytes
,
1923 dev_kfree_skb_any(skb
);
1928 #endif /* IXGBE_FCOE */
1929 ixgbe_rx_skb(q_vector
, skb
);
1931 /* update budget accounting */
1933 } while (likely(total_rx_packets
< budget
));
1935 u64_stats_update_begin(&rx_ring
->syncp
);
1936 rx_ring
->stats
.packets
+= total_rx_packets
;
1937 rx_ring
->stats
.bytes
+= total_rx_bytes
;
1938 u64_stats_update_end(&rx_ring
->syncp
);
1939 q_vector
->rx
.total_packets
+= total_rx_packets
;
1940 q_vector
->rx
.total_bytes
+= total_rx_bytes
;
1943 ixgbe_alloc_rx_buffers(rx_ring
, cleaned_count
);
1945 return (total_rx_packets
< budget
);
1949 * ixgbe_configure_msix - Configure MSI-X hardware
1950 * @adapter: board private structure
1952 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1955 static void ixgbe_configure_msix(struct ixgbe_adapter
*adapter
)
1957 struct ixgbe_q_vector
*q_vector
;
1961 /* Populate MSIX to EITR Select */
1962 if (adapter
->num_vfs
> 32) {
1963 u32 eitrsel
= (1 << (adapter
->num_vfs
- 32)) - 1;
1964 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EITRSEL
, eitrsel
);
1968 * Populate the IVAR table and set the ITR values to the
1969 * corresponding register.
1971 for (v_idx
= 0; v_idx
< adapter
->num_q_vectors
; v_idx
++) {
1972 struct ixgbe_ring
*ring
;
1973 q_vector
= adapter
->q_vector
[v_idx
];
1975 ixgbe_for_each_ring(ring
, q_vector
->rx
)
1976 ixgbe_set_ivar(adapter
, 0, ring
->reg_idx
, v_idx
);
1978 ixgbe_for_each_ring(ring
, q_vector
->tx
)
1979 ixgbe_set_ivar(adapter
, 1, ring
->reg_idx
, v_idx
);
1981 ixgbe_write_eitr(q_vector
);
1984 switch (adapter
->hw
.mac
.type
) {
1985 case ixgbe_mac_82598EB
:
1986 ixgbe_set_ivar(adapter
, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX
,
1989 case ixgbe_mac_82599EB
:
1990 case ixgbe_mac_X540
:
1991 ixgbe_set_ivar(adapter
, -1, 1, v_idx
);
1996 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EITR(v_idx
), 1950);
1998 /* set up to autoclear timer, and the vectors */
1999 mask
= IXGBE_EIMS_ENABLE_MASK
;
2000 mask
&= ~(IXGBE_EIMS_OTHER
|
2001 IXGBE_EIMS_MAILBOX
|
2004 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIAC
, mask
);
2007 enum latency_range
{
2011 latency_invalid
= 255
2015 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2016 * @q_vector: structure containing interrupt and ring information
2017 * @ring_container: structure containing ring performance data
2019 * Stores a new ITR value based on packets and byte
2020 * counts during the last interrupt. The advantage of per interrupt
2021 * computation is faster updates and more accurate ITR for the current
2022 * traffic pattern. Constants in this function were computed
2023 * based on theoretical maximum wire speed and thresholds were set based
2024 * on testing data as well as attempting to minimize response time
2025 * while increasing bulk throughput.
2026 * this functionality is controlled by the InterruptThrottleRate module
2027 * parameter (see ixgbe_param.c)
2029 static void ixgbe_update_itr(struct ixgbe_q_vector
*q_vector
,
2030 struct ixgbe_ring_container
*ring_container
)
2032 int bytes
= ring_container
->total_bytes
;
2033 int packets
= ring_container
->total_packets
;
2036 u8 itr_setting
= ring_container
->itr
;
2041 /* simple throttlerate management
2042 * 0-10MB/s lowest (100000 ints/s)
2043 * 10-20MB/s low (20000 ints/s)
2044 * 20-1249MB/s bulk (8000 ints/s)
2046 /* what was last interrupt timeslice? */
2047 timepassed_us
= q_vector
->itr
>> 2;
2048 bytes_perint
= bytes
/ timepassed_us
; /* bytes/usec */
2050 switch (itr_setting
) {
2051 case lowest_latency
:
2052 if (bytes_perint
> 10)
2053 itr_setting
= low_latency
;
2056 if (bytes_perint
> 20)
2057 itr_setting
= bulk_latency
;
2058 else if (bytes_perint
<= 10)
2059 itr_setting
= lowest_latency
;
2062 if (bytes_perint
<= 20)
2063 itr_setting
= low_latency
;
2067 /* clear work counters since we have the values we need */
2068 ring_container
->total_bytes
= 0;
2069 ring_container
->total_packets
= 0;
2071 /* write updated itr to ring container */
2072 ring_container
->itr
= itr_setting
;
2076 * ixgbe_write_eitr - write EITR register in hardware specific way
2077 * @q_vector: structure containing interrupt and ring information
2079 * This function is made to be called by ethtool and by the driver
2080 * when it needs to update EITR registers at runtime. Hardware
2081 * specific quirks/differences are taken care of here.
2083 void ixgbe_write_eitr(struct ixgbe_q_vector
*q_vector
)
2085 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
2086 struct ixgbe_hw
*hw
= &adapter
->hw
;
2087 int v_idx
= q_vector
->v_idx
;
2088 u32 itr_reg
= q_vector
->itr
& IXGBE_MAX_EITR
;
2090 switch (adapter
->hw
.mac
.type
) {
2091 case ixgbe_mac_82598EB
:
2092 /* must write high and low 16 bits to reset counter */
2093 itr_reg
|= (itr_reg
<< 16);
2095 case ixgbe_mac_82599EB
:
2096 case ixgbe_mac_X540
:
2098 * set the WDIS bit to not clear the timer bits and cause an
2099 * immediate assertion of the interrupt
2101 itr_reg
|= IXGBE_EITR_CNT_WDIS
;
2106 IXGBE_WRITE_REG(hw
, IXGBE_EITR(v_idx
), itr_reg
);
2109 static void ixgbe_set_itr(struct ixgbe_q_vector
*q_vector
)
2111 u32 new_itr
= q_vector
->itr
;
2114 ixgbe_update_itr(q_vector
, &q_vector
->tx
);
2115 ixgbe_update_itr(q_vector
, &q_vector
->rx
);
2117 current_itr
= max(q_vector
->rx
.itr
, q_vector
->tx
.itr
);
2119 switch (current_itr
) {
2120 /* counts and packets in update_itr are dependent on these numbers */
2121 case lowest_latency
:
2122 new_itr
= IXGBE_100K_ITR
;
2125 new_itr
= IXGBE_20K_ITR
;
2128 new_itr
= IXGBE_8K_ITR
;
2134 if (new_itr
!= q_vector
->itr
) {
2135 /* do an exponential smoothing */
2136 new_itr
= (10 * new_itr
* q_vector
->itr
) /
2137 ((9 * new_itr
) + q_vector
->itr
);
2139 /* save the algorithm value here */
2140 q_vector
->itr
= new_itr
;
2142 ixgbe_write_eitr(q_vector
);
2147 * ixgbe_check_overtemp_subtask - check for over temperature
2148 * @adapter: pointer to adapter
2150 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter
*adapter
)
2152 struct ixgbe_hw
*hw
= &adapter
->hw
;
2153 u32 eicr
= adapter
->interrupt_event
;
2155 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
2158 if (!(adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
) &&
2159 !(adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_EVENT
))
2162 adapter
->flags2
&= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT
;
2164 switch (hw
->device_id
) {
2165 case IXGBE_DEV_ID_82599_T3_LOM
:
2167 * Since the warning interrupt is for both ports
2168 * we don't have to check if:
2169 * - This interrupt wasn't for our port.
2170 * - We may have missed the interrupt so always have to
2171 * check if we got a LSC
2173 if (!(eicr
& IXGBE_EICR_GPI_SDP0
) &&
2174 !(eicr
& IXGBE_EICR_LSC
))
2177 if (!(eicr
& IXGBE_EICR_LSC
) && hw
->mac
.ops
.check_link
) {
2179 bool link_up
= false;
2181 hw
->mac
.ops
.check_link(hw
, &autoneg
, &link_up
, false);
2187 /* Check if this is not due to overtemp */
2188 if (hw
->phy
.ops
.check_overtemp(hw
) != IXGBE_ERR_OVERTEMP
)
2193 if (!(eicr
& IXGBE_EICR_GPI_SDP0
))
2198 "Network adapter has been stopped because it has over heated. "
2199 "Restart the computer. If the problem persists, "
2200 "power off the system and replace the adapter\n");
2202 adapter
->interrupt_event
= 0;
2205 static void ixgbe_check_fan_failure(struct ixgbe_adapter
*adapter
, u32 eicr
)
2207 struct ixgbe_hw
*hw
= &adapter
->hw
;
2209 if ((adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) &&
2210 (eicr
& IXGBE_EICR_GPI_SDP1
)) {
2211 e_crit(probe
, "Fan has stopped, replace the adapter\n");
2212 /* write to clear the interrupt */
2213 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_GPI_SDP1
);
2217 static void ixgbe_check_overtemp_event(struct ixgbe_adapter
*adapter
, u32 eicr
)
2219 if (!(adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
))
2222 switch (adapter
->hw
.mac
.type
) {
2223 case ixgbe_mac_82599EB
:
2225 * Need to check link state so complete overtemp check
2228 if (((eicr
& IXGBE_EICR_GPI_SDP0
) || (eicr
& IXGBE_EICR_LSC
)) &&
2229 (!test_bit(__IXGBE_DOWN
, &adapter
->state
))) {
2230 adapter
->interrupt_event
= eicr
;
2231 adapter
->flags2
|= IXGBE_FLAG2_TEMP_SENSOR_EVENT
;
2232 ixgbe_service_event_schedule(adapter
);
2236 case ixgbe_mac_X540
:
2237 if (!(eicr
& IXGBE_EICR_TS
))
2245 "Network adapter has been stopped because it has over heated. "
2246 "Restart the computer. If the problem persists, "
2247 "power off the system and replace the adapter\n");
2250 static void ixgbe_check_sfp_event(struct ixgbe_adapter
*adapter
, u32 eicr
)
2252 struct ixgbe_hw
*hw
= &adapter
->hw
;
2254 if (eicr
& IXGBE_EICR_GPI_SDP2
) {
2255 /* Clear the interrupt */
2256 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_GPI_SDP2
);
2257 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
2258 adapter
->flags2
|= IXGBE_FLAG2_SFP_NEEDS_RESET
;
2259 ixgbe_service_event_schedule(adapter
);
2263 if (eicr
& IXGBE_EICR_GPI_SDP1
) {
2264 /* Clear the interrupt */
2265 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_GPI_SDP1
);
2266 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
2267 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_CONFIG
;
2268 ixgbe_service_event_schedule(adapter
);
2273 static void ixgbe_check_lsc(struct ixgbe_adapter
*adapter
)
2275 struct ixgbe_hw
*hw
= &adapter
->hw
;
2278 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_UPDATE
;
2279 adapter
->link_check_timeout
= jiffies
;
2280 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
2281 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, IXGBE_EIMC_LSC
);
2282 IXGBE_WRITE_FLUSH(hw
);
2283 ixgbe_service_event_schedule(adapter
);
2287 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter
*adapter
,
2291 struct ixgbe_hw
*hw
= &adapter
->hw
;
2293 switch (hw
->mac
.type
) {
2294 case ixgbe_mac_82598EB
:
2295 mask
= (IXGBE_EIMS_RTX_QUEUE
& qmask
);
2296 IXGBE_WRITE_REG(hw
, IXGBE_EIMS
, mask
);
2298 case ixgbe_mac_82599EB
:
2299 case ixgbe_mac_X540
:
2300 mask
= (qmask
& 0xFFFFFFFF);
2302 IXGBE_WRITE_REG(hw
, IXGBE_EIMS_EX(0), mask
);
2303 mask
= (qmask
>> 32);
2305 IXGBE_WRITE_REG(hw
, IXGBE_EIMS_EX(1), mask
);
2310 /* skip the flush */
2313 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter
*adapter
,
2317 struct ixgbe_hw
*hw
= &adapter
->hw
;
2319 switch (hw
->mac
.type
) {
2320 case ixgbe_mac_82598EB
:
2321 mask
= (IXGBE_EIMS_RTX_QUEUE
& qmask
);
2322 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, mask
);
2324 case ixgbe_mac_82599EB
:
2325 case ixgbe_mac_X540
:
2326 mask
= (qmask
& 0xFFFFFFFF);
2328 IXGBE_WRITE_REG(hw
, IXGBE_EIMC_EX(0), mask
);
2329 mask
= (qmask
>> 32);
2331 IXGBE_WRITE_REG(hw
, IXGBE_EIMC_EX(1), mask
);
2336 /* skip the flush */
2340 * ixgbe_irq_enable - Enable default interrupt generation settings
2341 * @adapter: board private structure
2343 static inline void ixgbe_irq_enable(struct ixgbe_adapter
*adapter
, bool queues
,
2346 u32 mask
= (IXGBE_EIMS_ENABLE_MASK
& ~IXGBE_EIMS_RTX_QUEUE
);
2348 /* don't reenable LSC while waiting for link */
2349 if (adapter
->flags
& IXGBE_FLAG_NEED_LINK_UPDATE
)
2350 mask
&= ~IXGBE_EIMS_LSC
;
2352 if (adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
)
2353 switch (adapter
->hw
.mac
.type
) {
2354 case ixgbe_mac_82599EB
:
2355 mask
|= IXGBE_EIMS_GPI_SDP0
;
2357 case ixgbe_mac_X540
:
2358 mask
|= IXGBE_EIMS_TS
;
2363 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
)
2364 mask
|= IXGBE_EIMS_GPI_SDP1
;
2365 switch (adapter
->hw
.mac
.type
) {
2366 case ixgbe_mac_82599EB
:
2367 mask
|= IXGBE_EIMS_GPI_SDP1
;
2368 mask
|= IXGBE_EIMS_GPI_SDP2
;
2369 case ixgbe_mac_X540
:
2370 mask
|= IXGBE_EIMS_ECC
;
2371 mask
|= IXGBE_EIMS_MAILBOX
;
2377 if (adapter
->hw
.mac
.type
== ixgbe_mac_X540
)
2378 mask
|= IXGBE_EIMS_TIMESYNC
;
2380 if ((adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) &&
2381 !(adapter
->flags2
& IXGBE_FLAG2_FDIR_REQUIRES_REINIT
))
2382 mask
|= IXGBE_EIMS_FLOW_DIR
;
2384 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMS
, mask
);
2386 ixgbe_irq_enable_queues(adapter
, ~0);
2388 IXGBE_WRITE_FLUSH(&adapter
->hw
);
2391 static irqreturn_t
ixgbe_msix_other(int irq
, void *data
)
2393 struct ixgbe_adapter
*adapter
= data
;
2394 struct ixgbe_hw
*hw
= &adapter
->hw
;
2398 * Workaround for Silicon errata. Use clear-by-write instead
2399 * of clear-by-read. Reading with EICS will return the
2400 * interrupt causes without clearing, which later be done
2401 * with the write to EICR.
2403 eicr
= IXGBE_READ_REG(hw
, IXGBE_EICS
);
2404 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, eicr
);
2406 if (eicr
& IXGBE_EICR_LSC
)
2407 ixgbe_check_lsc(adapter
);
2409 if (eicr
& IXGBE_EICR_MAILBOX
)
2410 ixgbe_msg_task(adapter
);
2412 switch (hw
->mac
.type
) {
2413 case ixgbe_mac_82599EB
:
2414 case ixgbe_mac_X540
:
2415 if (eicr
& IXGBE_EICR_ECC
)
2416 e_info(link
, "Received unrecoverable ECC Err, please "
2418 /* Handle Flow Director Full threshold interrupt */
2419 if (eicr
& IXGBE_EICR_FLOW_DIR
) {
2420 int reinit_count
= 0;
2422 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
2423 struct ixgbe_ring
*ring
= adapter
->tx_ring
[i
];
2424 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE
,
2429 /* no more flow director interrupts until after init */
2430 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, IXGBE_EIMC_FLOW_DIR
);
2431 adapter
->flags2
|= IXGBE_FLAG2_FDIR_REQUIRES_REINIT
;
2432 ixgbe_service_event_schedule(adapter
);
2435 ixgbe_check_sfp_event(adapter
, eicr
);
2436 ixgbe_check_overtemp_event(adapter
, eicr
);
2442 ixgbe_check_fan_failure(adapter
, eicr
);
2444 if (unlikely(eicr
& IXGBE_EICR_TIMESYNC
))
2445 ixgbe_ptp_check_pps_event(adapter
, eicr
);
2447 /* re-enable the original interrupt state, no lsc, no queues */
2448 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2449 ixgbe_irq_enable(adapter
, false, false);
2454 static irqreturn_t
ixgbe_msix_clean_rings(int irq
, void *data
)
2456 struct ixgbe_q_vector
*q_vector
= data
;
2458 /* EIAM disabled interrupts (on this vector) for us */
2460 if (q_vector
->rx
.ring
|| q_vector
->tx
.ring
)
2461 napi_schedule(&q_vector
->napi
);
2467 * ixgbe_poll - NAPI Rx polling callback
2468 * @napi: structure for representing this polling device
2469 * @budget: how many packets driver is allowed to clean
2471 * This function is used for legacy and MSI, NAPI mode
2473 int ixgbe_poll(struct napi_struct
*napi
, int budget
)
2475 struct ixgbe_q_vector
*q_vector
=
2476 container_of(napi
, struct ixgbe_q_vector
, napi
);
2477 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
2478 struct ixgbe_ring
*ring
;
2479 int per_ring_budget
;
2480 bool clean_complete
= true;
2482 #ifdef CONFIG_IXGBE_DCA
2483 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
2484 ixgbe_update_dca(q_vector
);
2487 ixgbe_for_each_ring(ring
, q_vector
->tx
)
2488 clean_complete
&= !!ixgbe_clean_tx_irq(q_vector
, ring
);
2490 /* attempt to distribute budget to each queue fairly, but don't allow
2491 * the budget to go below 1 because we'll exit polling */
2492 if (q_vector
->rx
.count
> 1)
2493 per_ring_budget
= max(budget
/q_vector
->rx
.count
, 1);
2495 per_ring_budget
= budget
;
2497 ixgbe_for_each_ring(ring
, q_vector
->rx
)
2498 clean_complete
&= ixgbe_clean_rx_irq(q_vector
, ring
,
2501 /* If all work not completed, return budget and keep polling */
2502 if (!clean_complete
)
2505 /* all work done, exit the polling mode */
2506 napi_complete(napi
);
2507 if (adapter
->rx_itr_setting
& 1)
2508 ixgbe_set_itr(q_vector
);
2509 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2510 ixgbe_irq_enable_queues(adapter
, ((u64
)1 << q_vector
->v_idx
));
2516 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2517 * @adapter: board private structure
2519 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2520 * interrupts from the kernel.
2522 static int ixgbe_request_msix_irqs(struct ixgbe_adapter
*adapter
)
2524 struct net_device
*netdev
= adapter
->netdev
;
2528 for (vector
= 0; vector
< adapter
->num_q_vectors
; vector
++) {
2529 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[vector
];
2530 struct msix_entry
*entry
= &adapter
->msix_entries
[vector
];
2532 if (q_vector
->tx
.ring
&& q_vector
->rx
.ring
) {
2533 snprintf(q_vector
->name
, sizeof(q_vector
->name
) - 1,
2534 "%s-%s-%d", netdev
->name
, "TxRx", ri
++);
2536 } else if (q_vector
->rx
.ring
) {
2537 snprintf(q_vector
->name
, sizeof(q_vector
->name
) - 1,
2538 "%s-%s-%d", netdev
->name
, "rx", ri
++);
2539 } else if (q_vector
->tx
.ring
) {
2540 snprintf(q_vector
->name
, sizeof(q_vector
->name
) - 1,
2541 "%s-%s-%d", netdev
->name
, "tx", ti
++);
2543 /* skip this unused q_vector */
2546 err
= request_irq(entry
->vector
, &ixgbe_msix_clean_rings
, 0,
2547 q_vector
->name
, q_vector
);
2549 e_err(probe
, "request_irq failed for MSIX interrupt "
2550 "Error: %d\n", err
);
2551 goto free_queue_irqs
;
2553 /* If Flow Director is enabled, set interrupt affinity */
2554 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) {
2555 /* assign the mask for this irq */
2556 irq_set_affinity_hint(entry
->vector
,
2557 &q_vector
->affinity_mask
);
2561 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
2562 ixgbe_msix_other
, 0, netdev
->name
, adapter
);
2564 e_err(probe
, "request_irq for msix_other failed: %d\n", err
);
2565 goto free_queue_irqs
;
2573 irq_set_affinity_hint(adapter
->msix_entries
[vector
].vector
,
2575 free_irq(adapter
->msix_entries
[vector
].vector
,
2576 adapter
->q_vector
[vector
]);
2578 adapter
->flags
&= ~IXGBE_FLAG_MSIX_ENABLED
;
2579 pci_disable_msix(adapter
->pdev
);
2580 kfree(adapter
->msix_entries
);
2581 adapter
->msix_entries
= NULL
;
2586 * ixgbe_intr - legacy mode Interrupt Handler
2587 * @irq: interrupt number
2588 * @data: pointer to a network interface device structure
2590 static irqreturn_t
ixgbe_intr(int irq
, void *data
)
2592 struct ixgbe_adapter
*adapter
= data
;
2593 struct ixgbe_hw
*hw
= &adapter
->hw
;
2594 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[0];
2598 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2599 * before the read of EICR.
2601 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, IXGBE_IRQ_CLEAR_MASK
);
2603 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2604 * therefore no explicit interrupt disable is necessary */
2605 eicr
= IXGBE_READ_REG(hw
, IXGBE_EICR
);
2608 * shared interrupt alert!
2609 * make sure interrupts are enabled because the read will
2610 * have disabled interrupts due to EIAM
2611 * finish the workaround of silicon errata on 82598. Unmask
2612 * the interrupt that we masked before the EICR read.
2614 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2615 ixgbe_irq_enable(adapter
, true, true);
2616 return IRQ_NONE
; /* Not our interrupt */
2619 if (eicr
& IXGBE_EICR_LSC
)
2620 ixgbe_check_lsc(adapter
);
2622 switch (hw
->mac
.type
) {
2623 case ixgbe_mac_82599EB
:
2624 ixgbe_check_sfp_event(adapter
, eicr
);
2626 case ixgbe_mac_X540
:
2627 if (eicr
& IXGBE_EICR_ECC
)
2628 e_info(link
, "Received unrecoverable ECC err, please "
2630 ixgbe_check_overtemp_event(adapter
, eicr
);
2636 ixgbe_check_fan_failure(adapter
, eicr
);
2637 if (unlikely(eicr
& IXGBE_EICR_TIMESYNC
))
2638 ixgbe_ptp_check_pps_event(adapter
, eicr
);
2640 /* would disable interrupts here but EIAM disabled it */
2641 napi_schedule(&q_vector
->napi
);
2644 * re-enable link(maybe) and non-queue interrupts, no flush.
2645 * ixgbe_poll will re-enable the queue interrupts
2647 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2648 ixgbe_irq_enable(adapter
, false, false);
2654 * ixgbe_request_irq - initialize interrupts
2655 * @adapter: board private structure
2657 * Attempts to configure interrupts using the best available
2658 * capabilities of the hardware and kernel.
2660 static int ixgbe_request_irq(struct ixgbe_adapter
*adapter
)
2662 struct net_device
*netdev
= adapter
->netdev
;
2665 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)
2666 err
= ixgbe_request_msix_irqs(adapter
);
2667 else if (adapter
->flags
& IXGBE_FLAG_MSI_ENABLED
)
2668 err
= request_irq(adapter
->pdev
->irq
, ixgbe_intr
, 0,
2669 netdev
->name
, adapter
);
2671 err
= request_irq(adapter
->pdev
->irq
, ixgbe_intr
, IRQF_SHARED
,
2672 netdev
->name
, adapter
);
2675 e_err(probe
, "request_irq failed, Error %d\n", err
);
2680 static void ixgbe_free_irq(struct ixgbe_adapter
*adapter
)
2684 if (!(adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)) {
2685 free_irq(adapter
->pdev
->irq
, adapter
);
2689 for (vector
= 0; vector
< adapter
->num_q_vectors
; vector
++) {
2690 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[vector
];
2691 struct msix_entry
*entry
= &adapter
->msix_entries
[vector
];
2693 /* free only the irqs that were actually requested */
2694 if (!q_vector
->rx
.ring
&& !q_vector
->tx
.ring
)
2697 /* clear the affinity_mask in the IRQ descriptor */
2698 irq_set_affinity_hint(entry
->vector
, NULL
);
2700 free_irq(entry
->vector
, q_vector
);
2703 free_irq(adapter
->msix_entries
[vector
++].vector
, adapter
);
2707 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2708 * @adapter: board private structure
2710 static inline void ixgbe_irq_disable(struct ixgbe_adapter
*adapter
)
2712 switch (adapter
->hw
.mac
.type
) {
2713 case ixgbe_mac_82598EB
:
2714 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC
, ~0);
2716 case ixgbe_mac_82599EB
:
2717 case ixgbe_mac_X540
:
2718 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC
, 0xFFFF0000);
2719 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC_EX(0), ~0);
2720 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC_EX(1), ~0);
2725 IXGBE_WRITE_FLUSH(&adapter
->hw
);
2726 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
2729 for (vector
= 0; vector
< adapter
->num_q_vectors
; vector
++)
2730 synchronize_irq(adapter
->msix_entries
[vector
].vector
);
2732 synchronize_irq(adapter
->msix_entries
[vector
++].vector
);
2734 synchronize_irq(adapter
->pdev
->irq
);
2739 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2742 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter
*adapter
)
2744 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[0];
2746 ixgbe_write_eitr(q_vector
);
2748 ixgbe_set_ivar(adapter
, 0, 0, 0);
2749 ixgbe_set_ivar(adapter
, 1, 0, 0);
2751 e_info(hw
, "Legacy interrupt IVAR setup done\n");
2755 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2756 * @adapter: board private structure
2757 * @ring: structure containing ring specific data
2759 * Configure the Tx descriptor ring after a reset.
2761 void ixgbe_configure_tx_ring(struct ixgbe_adapter
*adapter
,
2762 struct ixgbe_ring
*ring
)
2764 struct ixgbe_hw
*hw
= &adapter
->hw
;
2765 u64 tdba
= ring
->dma
;
2767 u32 txdctl
= IXGBE_TXDCTL_ENABLE
;
2768 u8 reg_idx
= ring
->reg_idx
;
2770 /* disable queue to avoid issues while updating state */
2771 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(reg_idx
), 0);
2772 IXGBE_WRITE_FLUSH(hw
);
2774 IXGBE_WRITE_REG(hw
, IXGBE_TDBAL(reg_idx
),
2775 (tdba
& DMA_BIT_MASK(32)));
2776 IXGBE_WRITE_REG(hw
, IXGBE_TDBAH(reg_idx
), (tdba
>> 32));
2777 IXGBE_WRITE_REG(hw
, IXGBE_TDLEN(reg_idx
),
2778 ring
->count
* sizeof(union ixgbe_adv_tx_desc
));
2779 IXGBE_WRITE_REG(hw
, IXGBE_TDH(reg_idx
), 0);
2780 IXGBE_WRITE_REG(hw
, IXGBE_TDT(reg_idx
), 0);
2781 ring
->tail
= hw
->hw_addr
+ IXGBE_TDT(reg_idx
);
2784 * set WTHRESH to encourage burst writeback, it should not be set
2785 * higher than 1 when ITR is 0 as it could cause false TX hangs
2787 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2788 * to or less than the number of on chip descriptors, which is
2791 if (!ring
->q_vector
|| (ring
->q_vector
->itr
< 8))
2792 txdctl
|= (1 << 16); /* WTHRESH = 1 */
2794 txdctl
|= (8 << 16); /* WTHRESH = 8 */
2797 * Setting PTHRESH to 32 both improves performance
2798 * and avoids a TX hang with DFP enabled
2800 txdctl
|= (1 << 8) | /* HTHRESH = 1 */
2801 32; /* PTHRESH = 32 */
2803 /* reinitialize flowdirector state */
2804 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) {
2805 ring
->atr_sample_rate
= adapter
->atr_sample_rate
;
2806 ring
->atr_count
= 0;
2807 set_bit(__IXGBE_TX_FDIR_INIT_DONE
, &ring
->state
);
2809 ring
->atr_sample_rate
= 0;
2812 clear_bit(__IXGBE_HANG_CHECK_ARMED
, &ring
->state
);
2815 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(reg_idx
), txdctl
);
2817 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2818 if (hw
->mac
.type
== ixgbe_mac_82598EB
&&
2819 !(IXGBE_READ_REG(hw
, IXGBE_LINKS
) & IXGBE_LINKS_UP
))
2822 /* poll to verify queue is enabled */
2824 usleep_range(1000, 2000);
2825 txdctl
= IXGBE_READ_REG(hw
, IXGBE_TXDCTL(reg_idx
));
2826 } while (--wait_loop
&& !(txdctl
& IXGBE_TXDCTL_ENABLE
));
2828 e_err(drv
, "Could not enable Tx Queue %d\n", reg_idx
);
2831 static void ixgbe_setup_mtqc(struct ixgbe_adapter
*adapter
)
2833 struct ixgbe_hw
*hw
= &adapter
->hw
;
2835 u8 tcs
= netdev_get_num_tc(adapter
->netdev
);
2837 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
2840 /* disable the arbiter while setting MTQC */
2841 rttdcs
= IXGBE_READ_REG(hw
, IXGBE_RTTDCS
);
2842 rttdcs
|= IXGBE_RTTDCS_ARBDIS
;
2843 IXGBE_WRITE_REG(hw
, IXGBE_RTTDCS
, rttdcs
);
2845 /* set transmit pool layout */
2846 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
2847 mtqc
= IXGBE_MTQC_VT_ENA
;
2849 mtqc
|= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_8TC_8TQ
;
2851 mtqc
|= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_4TC_4TQ
;
2852 else if (adapter
->ring_feature
[RING_F_RSS
].indices
== 4)
2853 mtqc
|= IXGBE_MTQC_32VF
;
2855 mtqc
|= IXGBE_MTQC_64VF
;
2858 mtqc
= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_8TC_8TQ
;
2860 mtqc
= IXGBE_MTQC_RT_ENA
| IXGBE_MTQC_4TC_4TQ
;
2862 mtqc
= IXGBE_MTQC_64Q_1PB
;
2865 IXGBE_WRITE_REG(hw
, IXGBE_MTQC
, mtqc
);
2867 /* Enable Security TX Buffer IFG for multiple pb */
2869 u32 sectx
= IXGBE_READ_REG(hw
, IXGBE_SECTXMINIFG
);
2870 sectx
|= IXGBE_SECTX_DCB
;
2871 IXGBE_WRITE_REG(hw
, IXGBE_SECTXMINIFG
, sectx
);
2874 /* re-enable the arbiter */
2875 rttdcs
&= ~IXGBE_RTTDCS_ARBDIS
;
2876 IXGBE_WRITE_REG(hw
, IXGBE_RTTDCS
, rttdcs
);
2880 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2881 * @adapter: board private structure
2883 * Configure the Tx unit of the MAC after a reset.
2885 static void ixgbe_configure_tx(struct ixgbe_adapter
*adapter
)
2887 struct ixgbe_hw
*hw
= &adapter
->hw
;
2891 ixgbe_setup_mtqc(adapter
);
2893 if (hw
->mac
.type
!= ixgbe_mac_82598EB
) {
2894 /* DMATXCTL.EN must be before Tx queues are enabled */
2895 dmatxctl
= IXGBE_READ_REG(hw
, IXGBE_DMATXCTL
);
2896 dmatxctl
|= IXGBE_DMATXCTL_TE
;
2897 IXGBE_WRITE_REG(hw
, IXGBE_DMATXCTL
, dmatxctl
);
2900 /* Setup the HW Tx Head and Tail descriptor pointers */
2901 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
2902 ixgbe_configure_tx_ring(adapter
, adapter
->tx_ring
[i
]);
2905 static void ixgbe_enable_rx_drop(struct ixgbe_adapter
*adapter
,
2906 struct ixgbe_ring
*ring
)
2908 struct ixgbe_hw
*hw
= &adapter
->hw
;
2909 u8 reg_idx
= ring
->reg_idx
;
2910 u32 srrctl
= IXGBE_READ_REG(hw
, IXGBE_SRRCTL(reg_idx
));
2912 srrctl
|= IXGBE_SRRCTL_DROP_EN
;
2914 IXGBE_WRITE_REG(hw
, IXGBE_SRRCTL(reg_idx
), srrctl
);
2917 static void ixgbe_disable_rx_drop(struct ixgbe_adapter
*adapter
,
2918 struct ixgbe_ring
*ring
)
2920 struct ixgbe_hw
*hw
= &adapter
->hw
;
2921 u8 reg_idx
= ring
->reg_idx
;
2922 u32 srrctl
= IXGBE_READ_REG(hw
, IXGBE_SRRCTL(reg_idx
));
2924 srrctl
&= ~IXGBE_SRRCTL_DROP_EN
;
2926 IXGBE_WRITE_REG(hw
, IXGBE_SRRCTL(reg_idx
), srrctl
);
2929 #ifdef CONFIG_IXGBE_DCB
2930 void ixgbe_set_rx_drop_en(struct ixgbe_adapter
*adapter
)
2932 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter
*adapter
)
2936 bool pfc_en
= adapter
->dcb_cfg
.pfc_mode_enable
;
2938 if (adapter
->ixgbe_ieee_pfc
)
2939 pfc_en
|= !!(adapter
->ixgbe_ieee_pfc
->pfc_en
);
2942 * We should set the drop enable bit if:
2945 * Number of Rx queues > 1 and flow control is disabled
2947 * This allows us to avoid head of line blocking for security
2948 * and performance reasons.
2950 if (adapter
->num_vfs
|| (adapter
->num_rx_queues
> 1 &&
2951 !(adapter
->hw
.fc
.current_mode
& ixgbe_fc_tx_pause
) && !pfc_en
)) {
2952 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
2953 ixgbe_enable_rx_drop(adapter
, adapter
->rx_ring
[i
]);
2955 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
2956 ixgbe_disable_rx_drop(adapter
, adapter
->rx_ring
[i
]);
2960 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2962 static void ixgbe_configure_srrctl(struct ixgbe_adapter
*adapter
,
2963 struct ixgbe_ring
*rx_ring
)
2965 struct ixgbe_hw
*hw
= &adapter
->hw
;
2967 u8 reg_idx
= rx_ring
->reg_idx
;
2969 if (hw
->mac
.type
== ixgbe_mac_82598EB
) {
2970 u16 mask
= adapter
->ring_feature
[RING_F_RSS
].mask
;
2973 * if VMDq is not active we must program one srrctl register
2974 * per RSS queue since we have enabled RDRXCTL.MVMEN
2979 /* configure header buffer length, needed for RSC */
2980 srrctl
= IXGBE_RX_HDR_SIZE
<< IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT
;
2982 /* configure the packet buffer length */
2983 srrctl
|= ixgbe_rx_bufsz(rx_ring
) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT
;
2985 /* configure descriptor type */
2986 srrctl
|= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF
;
2988 IXGBE_WRITE_REG(hw
, IXGBE_SRRCTL(reg_idx
), srrctl
);
2991 static void ixgbe_setup_mrqc(struct ixgbe_adapter
*adapter
)
2993 struct ixgbe_hw
*hw
= &adapter
->hw
;
2994 static const u32 seed
[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2995 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2996 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2997 u32 mrqc
= 0, reta
= 0;
3000 u16 rss_i
= adapter
->ring_feature
[RING_F_RSS
].indices
;
3003 * Program table for at least 2 queues w/ SR-IOV so that VFs can
3004 * make full use of any rings they may have. We will use the
3005 * PSRTYPE register to control how many rings we use within the PF.
3007 if ((adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) && (rss_i
< 2))
3010 /* Fill out hash function seeds */
3011 for (i
= 0; i
< 10; i
++)
3012 IXGBE_WRITE_REG(hw
, IXGBE_RSSRK(i
), seed
[i
]);
3014 /* Fill out redirection table */
3015 for (i
= 0, j
= 0; i
< 128; i
++, j
++) {
3018 /* reta = 4-byte sliding window of
3019 * 0x00..(indices-1)(indices-1)00..etc. */
3020 reta
= (reta
<< 8) | (j
* 0x11);
3022 IXGBE_WRITE_REG(hw
, IXGBE_RETA(i
>> 2), reta
);
3025 /* Disable indicating checksum in descriptor, enables RSS hash */
3026 rxcsum
= IXGBE_READ_REG(hw
, IXGBE_RXCSUM
);
3027 rxcsum
|= IXGBE_RXCSUM_PCSD
;
3028 IXGBE_WRITE_REG(hw
, IXGBE_RXCSUM
, rxcsum
);
3030 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
3031 if (adapter
->ring_feature
[RING_F_RSS
].mask
)
3032 mrqc
= IXGBE_MRQC_RSSEN
;
3034 u8 tcs
= netdev_get_num_tc(adapter
->netdev
);
3036 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
3038 mrqc
= IXGBE_MRQC_VMDQRT8TCEN
; /* 8 TCs */
3040 mrqc
= IXGBE_MRQC_VMDQRT4TCEN
; /* 4 TCs */
3041 else if (adapter
->ring_feature
[RING_F_RSS
].indices
== 4)
3042 mrqc
= IXGBE_MRQC_VMDQRSS32EN
;
3044 mrqc
= IXGBE_MRQC_VMDQRSS64EN
;
3047 mrqc
= IXGBE_MRQC_RTRSS8TCEN
;
3049 mrqc
= IXGBE_MRQC_RTRSS4TCEN
;
3051 mrqc
= IXGBE_MRQC_RSSEN
;
3055 /* Perform hash on these packet types */
3056 mrqc
|= IXGBE_MRQC_RSS_FIELD_IPV4
|
3057 IXGBE_MRQC_RSS_FIELD_IPV4_TCP
|
3058 IXGBE_MRQC_RSS_FIELD_IPV6
|
3059 IXGBE_MRQC_RSS_FIELD_IPV6_TCP
;
3061 if (adapter
->flags2
& IXGBE_FLAG2_RSS_FIELD_IPV4_UDP
)
3062 mrqc
|= IXGBE_MRQC_RSS_FIELD_IPV4_UDP
;
3063 if (adapter
->flags2
& IXGBE_FLAG2_RSS_FIELD_IPV6_UDP
)
3064 mrqc
|= IXGBE_MRQC_RSS_FIELD_IPV6_UDP
;
3066 IXGBE_WRITE_REG(hw
, IXGBE_MRQC
, mrqc
);
3070 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3071 * @adapter: address of board private structure
3072 * @index: index of ring to set
3074 static void ixgbe_configure_rscctl(struct ixgbe_adapter
*adapter
,
3075 struct ixgbe_ring
*ring
)
3077 struct ixgbe_hw
*hw
= &adapter
->hw
;
3079 u8 reg_idx
= ring
->reg_idx
;
3081 if (!ring_is_rsc_enabled(ring
))
3084 rscctrl
= IXGBE_READ_REG(hw
, IXGBE_RSCCTL(reg_idx
));
3085 rscctrl
|= IXGBE_RSCCTL_RSCEN
;
3087 * we must limit the number of descriptors so that the
3088 * total size of max desc * buf_len is not greater
3091 rscctrl
|= IXGBE_RSCCTL_MAXDESC_16
;
3092 IXGBE_WRITE_REG(hw
, IXGBE_RSCCTL(reg_idx
), rscctrl
);
3095 #define IXGBE_MAX_RX_DESC_POLL 10
3096 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter
*adapter
,
3097 struct ixgbe_ring
*ring
)
3099 struct ixgbe_hw
*hw
= &adapter
->hw
;
3100 int wait_loop
= IXGBE_MAX_RX_DESC_POLL
;
3102 u8 reg_idx
= ring
->reg_idx
;
3104 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3105 if (hw
->mac
.type
== ixgbe_mac_82598EB
&&
3106 !(IXGBE_READ_REG(hw
, IXGBE_LINKS
) & IXGBE_LINKS_UP
))
3110 usleep_range(1000, 2000);
3111 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
3112 } while (--wait_loop
&& !(rxdctl
& IXGBE_RXDCTL_ENABLE
));
3115 e_err(drv
, "RXDCTL.ENABLE on Rx queue %d not set within "
3116 "the polling period\n", reg_idx
);
3120 void ixgbe_disable_rx_queue(struct ixgbe_adapter
*adapter
,
3121 struct ixgbe_ring
*ring
)
3123 struct ixgbe_hw
*hw
= &adapter
->hw
;
3124 int wait_loop
= IXGBE_MAX_RX_DESC_POLL
;
3126 u8 reg_idx
= ring
->reg_idx
;
3128 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
3129 rxdctl
&= ~IXGBE_RXDCTL_ENABLE
;
3131 /* write value back with RXDCTL.ENABLE bit cleared */
3132 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(reg_idx
), rxdctl
);
3134 if (hw
->mac
.type
== ixgbe_mac_82598EB
&&
3135 !(IXGBE_READ_REG(hw
, IXGBE_LINKS
) & IXGBE_LINKS_UP
))
3138 /* the hardware may take up to 100us to really disable the rx queue */
3141 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
3142 } while (--wait_loop
&& (rxdctl
& IXGBE_RXDCTL_ENABLE
));
3145 e_err(drv
, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3146 "the polling period\n", reg_idx
);
3150 void ixgbe_configure_rx_ring(struct ixgbe_adapter
*adapter
,
3151 struct ixgbe_ring
*ring
)
3153 struct ixgbe_hw
*hw
= &adapter
->hw
;
3154 u64 rdba
= ring
->dma
;
3156 u8 reg_idx
= ring
->reg_idx
;
3158 /* disable queue to avoid issues while updating state */
3159 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(reg_idx
));
3160 ixgbe_disable_rx_queue(adapter
, ring
);
3162 IXGBE_WRITE_REG(hw
, IXGBE_RDBAL(reg_idx
), (rdba
& DMA_BIT_MASK(32)));
3163 IXGBE_WRITE_REG(hw
, IXGBE_RDBAH(reg_idx
), (rdba
>> 32));
3164 IXGBE_WRITE_REG(hw
, IXGBE_RDLEN(reg_idx
),
3165 ring
->count
* sizeof(union ixgbe_adv_rx_desc
));
3166 IXGBE_WRITE_REG(hw
, IXGBE_RDH(reg_idx
), 0);
3167 IXGBE_WRITE_REG(hw
, IXGBE_RDT(reg_idx
), 0);
3168 ring
->tail
= hw
->hw_addr
+ IXGBE_RDT(reg_idx
);
3170 ixgbe_configure_srrctl(adapter
, ring
);
3171 ixgbe_configure_rscctl(adapter
, ring
);
3173 if (hw
->mac
.type
== ixgbe_mac_82598EB
) {
3175 * enable cache line friendly hardware writes:
3176 * PTHRESH=32 descriptors (half the internal cache),
3177 * this also removes ugly rx_no_buffer_count increment
3178 * HTHRESH=4 descriptors (to minimize latency on fetch)
3179 * WTHRESH=8 burst writeback up to two cache lines
3181 rxdctl
&= ~0x3FFFFF;
3185 /* enable receive descriptor ring */
3186 rxdctl
|= IXGBE_RXDCTL_ENABLE
;
3187 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(reg_idx
), rxdctl
);
3189 ixgbe_rx_desc_queue_enable(adapter
, ring
);
3190 ixgbe_alloc_rx_buffers(ring
, ixgbe_desc_unused(ring
));
3193 static void ixgbe_setup_psrtype(struct ixgbe_adapter
*adapter
)
3195 struct ixgbe_hw
*hw
= &adapter
->hw
;
3196 int rss_i
= adapter
->ring_feature
[RING_F_RSS
].indices
;
3199 /* PSRTYPE must be initialized in non 82598 adapters */
3200 u32 psrtype
= IXGBE_PSRTYPE_TCPHDR
|
3201 IXGBE_PSRTYPE_UDPHDR
|
3202 IXGBE_PSRTYPE_IPV4HDR
|
3203 IXGBE_PSRTYPE_L2HDR
|
3204 IXGBE_PSRTYPE_IPV6HDR
;
3206 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
3214 for (p
= 0; p
< adapter
->num_rx_pools
; p
++)
3215 IXGBE_WRITE_REG(hw
, IXGBE_PSRTYPE(VMDQ_P(p
)),
3219 static void ixgbe_configure_virtualization(struct ixgbe_adapter
*adapter
)
3221 struct ixgbe_hw
*hw
= &adapter
->hw
;
3222 u32 reg_offset
, vf_shift
;
3223 u32 gcr_ext
, vmdctl
;
3226 if (!(adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
))
3229 vmdctl
= IXGBE_READ_REG(hw
, IXGBE_VT_CTL
);
3230 vmdctl
|= IXGBE_VMD_CTL_VMDQ_EN
;
3231 vmdctl
&= ~IXGBE_VT_CTL_POOL_MASK
;
3232 vmdctl
|= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT
;
3233 vmdctl
|= IXGBE_VT_CTL_REPLEN
;
3234 IXGBE_WRITE_REG(hw
, IXGBE_VT_CTL
, vmdctl
);
3236 vf_shift
= VMDQ_P(0) % 32;
3237 reg_offset
= (VMDQ_P(0) >= 32) ? 1 : 0;
3239 /* Enable only the PF's pool for Tx/Rx */
3240 IXGBE_WRITE_REG(hw
, IXGBE_VFRE(reg_offset
), (~0) << vf_shift
);
3241 IXGBE_WRITE_REG(hw
, IXGBE_VFRE(reg_offset
^ 1), reg_offset
- 1);
3242 IXGBE_WRITE_REG(hw
, IXGBE_VFTE(reg_offset
), (~0) << vf_shift
);
3243 IXGBE_WRITE_REG(hw
, IXGBE_VFTE(reg_offset
^ 1), reg_offset
- 1);
3245 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3246 hw
->mac
.ops
.set_vmdq(hw
, 0, VMDQ_P(0));
3249 * Set up VF register offsets for selected VT Mode,
3250 * i.e. 32 or 64 VFs for SR-IOV
3252 switch (adapter
->ring_feature
[RING_F_VMDQ
].mask
) {
3253 case IXGBE_82599_VMDQ_8Q_MASK
:
3254 gcr_ext
= IXGBE_GCR_EXT_VT_MODE_16
;
3256 case IXGBE_82599_VMDQ_4Q_MASK
:
3257 gcr_ext
= IXGBE_GCR_EXT_VT_MODE_32
;
3260 gcr_ext
= IXGBE_GCR_EXT_VT_MODE_64
;
3264 IXGBE_WRITE_REG(hw
, IXGBE_GCR_EXT
, gcr_ext
);
3267 /* Enable MAC Anti-Spoofing */
3268 hw
->mac
.ops
.set_mac_anti_spoofing(hw
, (adapter
->num_vfs
!= 0),
3270 /* For VFs that have spoof checking turned off */
3271 for (i
= 0; i
< adapter
->num_vfs
; i
++) {
3272 if (!adapter
->vfinfo
[i
].spoofchk_enabled
)
3273 ixgbe_ndo_set_vf_spoofchk(adapter
->netdev
, i
, false);
3277 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter
*adapter
)
3279 struct ixgbe_hw
*hw
= &adapter
->hw
;
3280 struct net_device
*netdev
= adapter
->netdev
;
3281 int max_frame
= netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
3282 struct ixgbe_ring
*rx_ring
;
3287 /* adjust max frame to be able to do baby jumbo for FCoE */
3288 if ((adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
) &&
3289 (max_frame
< IXGBE_FCOE_JUMBO_FRAME_SIZE
))
3290 max_frame
= IXGBE_FCOE_JUMBO_FRAME_SIZE
;
3292 #endif /* IXGBE_FCOE */
3294 /* adjust max frame to be at least the size of a standard frame */
3295 if (max_frame
< (ETH_FRAME_LEN
+ ETH_FCS_LEN
))
3296 max_frame
= (ETH_FRAME_LEN
+ ETH_FCS_LEN
);
3298 mhadd
= IXGBE_READ_REG(hw
, IXGBE_MHADD
);
3299 if (max_frame
!= (mhadd
>> IXGBE_MHADD_MFS_SHIFT
)) {
3300 mhadd
&= ~IXGBE_MHADD_MFS_MASK
;
3301 mhadd
|= max_frame
<< IXGBE_MHADD_MFS_SHIFT
;
3303 IXGBE_WRITE_REG(hw
, IXGBE_MHADD
, mhadd
);
3306 hlreg0
= IXGBE_READ_REG(hw
, IXGBE_HLREG0
);
3307 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3308 hlreg0
|= IXGBE_HLREG0_JUMBOEN
;
3309 IXGBE_WRITE_REG(hw
, IXGBE_HLREG0
, hlreg0
);
3312 * Setup the HW Rx Head and Tail Descriptor Pointers and
3313 * the Base and Length of the Rx Descriptor Ring
3315 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3316 rx_ring
= adapter
->rx_ring
[i
];
3317 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)
3318 set_ring_rsc_enabled(rx_ring
);
3320 clear_ring_rsc_enabled(rx_ring
);
3324 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter
*adapter
)
3326 struct ixgbe_hw
*hw
= &adapter
->hw
;
3327 u32 rdrxctl
= IXGBE_READ_REG(hw
, IXGBE_RDRXCTL
);
3329 switch (hw
->mac
.type
) {
3330 case ixgbe_mac_82598EB
:
3332 * For VMDq support of different descriptor types or
3333 * buffer sizes through the use of multiple SRRCTL
3334 * registers, RDRXCTL.MVMEN must be set to 1
3336 * also, the manual doesn't mention it clearly but DCA hints
3337 * will only use queue 0's tags unless this bit is set. Side
3338 * effects of setting this bit are only that SRRCTL must be
3339 * fully programmed [0..15]
3341 rdrxctl
|= IXGBE_RDRXCTL_MVMEN
;
3343 case ixgbe_mac_82599EB
:
3344 case ixgbe_mac_X540
:
3345 /* Disable RSC for ACK packets */
3346 IXGBE_WRITE_REG(hw
, IXGBE_RSCDBU
,
3347 (IXGBE_RSCDBU_RSCACKDIS
| IXGBE_READ_REG(hw
, IXGBE_RSCDBU
)));
3348 rdrxctl
&= ~IXGBE_RDRXCTL_RSCFRSTSIZE
;
3349 /* hardware requires some bits to be set by default */
3350 rdrxctl
|= (IXGBE_RDRXCTL_RSCACKC
| IXGBE_RDRXCTL_FCOE_WRFIX
);
3351 rdrxctl
|= IXGBE_RDRXCTL_CRCSTRIP
;
3354 /* We should do nothing since we don't know this hardware */
3358 IXGBE_WRITE_REG(hw
, IXGBE_RDRXCTL
, rdrxctl
);
3362 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3363 * @adapter: board private structure
3365 * Configure the Rx unit of the MAC after a reset.
3367 static void ixgbe_configure_rx(struct ixgbe_adapter
*adapter
)
3369 struct ixgbe_hw
*hw
= &adapter
->hw
;
3373 /* disable receives while setting up the descriptors */
3374 rxctrl
= IXGBE_READ_REG(hw
, IXGBE_RXCTRL
);
3375 IXGBE_WRITE_REG(hw
, IXGBE_RXCTRL
, rxctrl
& ~IXGBE_RXCTRL_RXEN
);
3377 ixgbe_setup_psrtype(adapter
);
3378 ixgbe_setup_rdrxctl(adapter
);
3380 /* Program registers for the distribution of queues */
3381 ixgbe_setup_mrqc(adapter
);
3383 /* set_rx_buffer_len must be called before ring initialization */
3384 ixgbe_set_rx_buffer_len(adapter
);
3387 * Setup the HW Rx Head and Tail Descriptor Pointers and
3388 * the Base and Length of the Rx Descriptor Ring
3390 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3391 ixgbe_configure_rx_ring(adapter
, adapter
->rx_ring
[i
]);
3393 /* disable drop enable for 82598 parts */
3394 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
3395 rxctrl
|= IXGBE_RXCTRL_DMBYPS
;
3397 /* enable all receives */
3398 rxctrl
|= IXGBE_RXCTRL_RXEN
;
3399 hw
->mac
.ops
.enable_rx_dma(hw
, rxctrl
);
3402 static int ixgbe_vlan_rx_add_vid(struct net_device
*netdev
, u16 vid
)
3404 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
3405 struct ixgbe_hw
*hw
= &adapter
->hw
;
3407 /* add VID to filter table */
3408 hw
->mac
.ops
.set_vfta(&adapter
->hw
, vid
, VMDQ_P(0), true);
3409 set_bit(vid
, adapter
->active_vlans
);
3414 static int ixgbe_vlan_rx_kill_vid(struct net_device
*netdev
, u16 vid
)
3416 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
3417 struct ixgbe_hw
*hw
= &adapter
->hw
;
3419 /* remove VID from filter table */
3420 hw
->mac
.ops
.set_vfta(&adapter
->hw
, vid
, VMDQ_P(0), false);
3421 clear_bit(vid
, adapter
->active_vlans
);
3427 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3428 * @adapter: driver data
3430 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter
*adapter
)
3432 struct ixgbe_hw
*hw
= &adapter
->hw
;
3435 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
3436 vlnctrl
&= ~(IXGBE_VLNCTRL_VFE
| IXGBE_VLNCTRL_CFIEN
);
3437 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
3441 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3442 * @adapter: driver data
3444 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter
*adapter
)
3446 struct ixgbe_hw
*hw
= &adapter
->hw
;
3449 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
3450 vlnctrl
|= IXGBE_VLNCTRL_VFE
;
3451 vlnctrl
&= ~IXGBE_VLNCTRL_CFIEN
;
3452 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
3456 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3457 * @adapter: driver data
3459 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter
*adapter
)
3461 struct ixgbe_hw
*hw
= &adapter
->hw
;
3465 switch (hw
->mac
.type
) {
3466 case ixgbe_mac_82598EB
:
3467 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
3468 vlnctrl
&= ~IXGBE_VLNCTRL_VME
;
3469 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
3471 case ixgbe_mac_82599EB
:
3472 case ixgbe_mac_X540
:
3473 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3474 j
= adapter
->rx_ring
[i
]->reg_idx
;
3475 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(j
));
3476 vlnctrl
&= ~IXGBE_RXDCTL_VME
;
3477 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(j
), vlnctrl
);
3486 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3487 * @adapter: driver data
3489 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter
*adapter
)
3491 struct ixgbe_hw
*hw
= &adapter
->hw
;
3495 switch (hw
->mac
.type
) {
3496 case ixgbe_mac_82598EB
:
3497 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
3498 vlnctrl
|= IXGBE_VLNCTRL_VME
;
3499 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
3501 case ixgbe_mac_82599EB
:
3502 case ixgbe_mac_X540
:
3503 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3504 j
= adapter
->rx_ring
[i
]->reg_idx
;
3505 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(j
));
3506 vlnctrl
|= IXGBE_RXDCTL_VME
;
3507 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(j
), vlnctrl
);
3515 static void ixgbe_restore_vlan(struct ixgbe_adapter
*adapter
)
3519 ixgbe_vlan_rx_add_vid(adapter
->netdev
, 0);
3521 for_each_set_bit(vid
, adapter
->active_vlans
, VLAN_N_VID
)
3522 ixgbe_vlan_rx_add_vid(adapter
->netdev
, vid
);
3526 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3527 * @netdev: network interface device structure
3529 * Writes unicast address list to the RAR table.
3530 * Returns: -ENOMEM on failure/insufficient address space
3531 * 0 on no addresses written
3532 * X on writing X addresses to the RAR table
3534 static int ixgbe_write_uc_addr_list(struct net_device
*netdev
)
3536 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
3537 struct ixgbe_hw
*hw
= &adapter
->hw
;
3538 unsigned int rar_entries
= hw
->mac
.num_rar_entries
- 1;
3541 /* In SR-IOV mode significantly less RAR entries are available */
3542 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
3543 rar_entries
= IXGBE_MAX_PF_MACVLANS
- 1;
3545 /* return ENOMEM indicating insufficient memory for addresses */
3546 if (netdev_uc_count(netdev
) > rar_entries
)
3549 if (!netdev_uc_empty(netdev
)) {
3550 struct netdev_hw_addr
*ha
;
3551 /* return error if we do not support writing to RAR table */
3552 if (!hw
->mac
.ops
.set_rar
)
3555 netdev_for_each_uc_addr(ha
, netdev
) {
3558 hw
->mac
.ops
.set_rar(hw
, rar_entries
--, ha
->addr
,
3559 VMDQ_P(0), IXGBE_RAH_AV
);
3563 /* write the addresses in reverse order to avoid write combining */
3564 for (; rar_entries
> 0 ; rar_entries
--)
3565 hw
->mac
.ops
.clear_rar(hw
, rar_entries
);
3571 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3572 * @netdev: network interface device structure
3574 * The set_rx_method entry point is called whenever the unicast/multicast
3575 * address list or the network interface flags are updated. This routine is
3576 * responsible for configuring the hardware for proper unicast, multicast and
3579 void ixgbe_set_rx_mode(struct net_device
*netdev
)
3581 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
3582 struct ixgbe_hw
*hw
= &adapter
->hw
;
3583 u32 fctrl
, vmolr
= IXGBE_VMOLR_BAM
| IXGBE_VMOLR_AUPE
;
3586 /* Check for Promiscuous and All Multicast modes */
3588 fctrl
= IXGBE_READ_REG(hw
, IXGBE_FCTRL
);
3590 /* set all bits that we expect to always be set */
3591 fctrl
&= ~IXGBE_FCTRL_SBP
; /* disable store-bad-packets */
3592 fctrl
|= IXGBE_FCTRL_BAM
;
3593 fctrl
|= IXGBE_FCTRL_DPF
; /* discard pause frames when FC enabled */
3594 fctrl
|= IXGBE_FCTRL_PMCF
;
3596 /* clear the bits we are changing the status of */
3597 fctrl
&= ~(IXGBE_FCTRL_UPE
| IXGBE_FCTRL_MPE
);
3599 if (netdev
->flags
& IFF_PROMISC
) {
3600 hw
->addr_ctrl
.user_set_promisc
= true;
3601 fctrl
|= (IXGBE_FCTRL_UPE
| IXGBE_FCTRL_MPE
);
3602 vmolr
|= (IXGBE_VMOLR_ROPE
| IXGBE_VMOLR_MPE
);
3603 /* don't hardware filter vlans in promisc mode */
3604 ixgbe_vlan_filter_disable(adapter
);
3606 if (netdev
->flags
& IFF_ALLMULTI
) {
3607 fctrl
|= IXGBE_FCTRL_MPE
;
3608 vmolr
|= IXGBE_VMOLR_MPE
;
3611 * Write addresses to the MTA, if the attempt fails
3612 * then we should just turn on promiscuous mode so
3613 * that we can at least receive multicast traffic
3615 hw
->mac
.ops
.update_mc_addr_list(hw
, netdev
);
3616 vmolr
|= IXGBE_VMOLR_ROMPE
;
3618 ixgbe_vlan_filter_enable(adapter
);
3619 hw
->addr_ctrl
.user_set_promisc
= false;
3623 * Write addresses to available RAR registers, if there is not
3624 * sufficient space to store all the addresses then enable
3625 * unicast promiscuous mode
3627 count
= ixgbe_write_uc_addr_list(netdev
);
3629 fctrl
|= IXGBE_FCTRL_UPE
;
3630 vmolr
|= IXGBE_VMOLR_ROPE
;
3633 if (adapter
->num_vfs
)
3634 ixgbe_restore_vf_multicasts(adapter
);
3636 if (hw
->mac
.type
!= ixgbe_mac_82598EB
) {
3637 vmolr
|= IXGBE_READ_REG(hw
, IXGBE_VMOLR(VMDQ_P(0))) &
3638 ~(IXGBE_VMOLR_MPE
| IXGBE_VMOLR_ROMPE
|
3640 IXGBE_WRITE_REG(hw
, IXGBE_VMOLR(VMDQ_P(0)), vmolr
);
3643 /* This is useful for sniffing bad packets. */
3644 if (adapter
->netdev
->features
& NETIF_F_RXALL
) {
3645 /* UPE and MPE will be handled by normal PROMISC logic
3646 * in e1000e_set_rx_mode */
3647 fctrl
|= (IXGBE_FCTRL_SBP
| /* Receive bad packets */
3648 IXGBE_FCTRL_BAM
| /* RX All Bcast Pkts */
3649 IXGBE_FCTRL_PMCF
); /* RX All MAC Ctrl Pkts */
3651 fctrl
&= ~(IXGBE_FCTRL_DPF
);
3652 /* NOTE: VLAN filtering is disabled by setting PROMISC */
3655 IXGBE_WRITE_REG(hw
, IXGBE_FCTRL
, fctrl
);
3657 if (netdev
->features
& NETIF_F_HW_VLAN_RX
)
3658 ixgbe_vlan_strip_enable(adapter
);
3660 ixgbe_vlan_strip_disable(adapter
);
3663 static void ixgbe_napi_enable_all(struct ixgbe_adapter
*adapter
)
3667 for (q_idx
= 0; q_idx
< adapter
->num_q_vectors
; q_idx
++)
3668 napi_enable(&adapter
->q_vector
[q_idx
]->napi
);
3671 static void ixgbe_napi_disable_all(struct ixgbe_adapter
*adapter
)
3675 for (q_idx
= 0; q_idx
< adapter
->num_q_vectors
; q_idx
++)
3676 napi_disable(&adapter
->q_vector
[q_idx
]->napi
);
3679 #ifdef CONFIG_IXGBE_DCB
3681 * ixgbe_configure_dcb - Configure DCB hardware
3682 * @adapter: ixgbe adapter struct
3684 * This is called by the driver on open to configure the DCB hardware.
3685 * This is also called by the gennetlink interface when reconfiguring
3688 static void ixgbe_configure_dcb(struct ixgbe_adapter
*adapter
)
3690 struct ixgbe_hw
*hw
= &adapter
->hw
;
3691 int max_frame
= adapter
->netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
3693 if (!(adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
)) {
3694 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
3695 netif_set_gso_max_size(adapter
->netdev
, 65536);
3699 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
3700 netif_set_gso_max_size(adapter
->netdev
, 32768);
3703 if (adapter
->netdev
->features
& NETIF_F_FCOE_MTU
)
3704 max_frame
= max(max_frame
, IXGBE_FCOE_JUMBO_FRAME_SIZE
);
3707 /* reconfigure the hardware */
3708 if (adapter
->dcbx_cap
& DCB_CAP_DCBX_VER_CEE
) {
3709 ixgbe_dcb_calculate_tc_credits(hw
, &adapter
->dcb_cfg
, max_frame
,
3711 ixgbe_dcb_calculate_tc_credits(hw
, &adapter
->dcb_cfg
, max_frame
,
3713 ixgbe_dcb_hw_config(hw
, &adapter
->dcb_cfg
);
3714 } else if (adapter
->ixgbe_ieee_ets
&& adapter
->ixgbe_ieee_pfc
) {
3715 ixgbe_dcb_hw_ets(&adapter
->hw
,
3716 adapter
->ixgbe_ieee_ets
,
3718 ixgbe_dcb_hw_pfc_config(&adapter
->hw
,
3719 adapter
->ixgbe_ieee_pfc
->pfc_en
,
3720 adapter
->ixgbe_ieee_ets
->prio_tc
);
3723 /* Enable RSS Hash per TC */
3724 if (hw
->mac
.type
!= ixgbe_mac_82598EB
) {
3726 u16 rss_i
= adapter
->ring_feature
[RING_F_RSS
].indices
- 1;
3733 /* write msb to all 8 TCs in one write */
3734 IXGBE_WRITE_REG(hw
, IXGBE_RQTC
, msb
* 0x11111111);
3739 /* Additional bittime to account for IXGBE framing */
3740 #define IXGBE_ETH_FRAMING 20
3743 * ixgbe_hpbthresh - calculate high water mark for flow control
3745 * @adapter: board private structure to calculate for
3746 * @pb: packet buffer to calculate
3748 static int ixgbe_hpbthresh(struct ixgbe_adapter
*adapter
, int pb
)
3750 struct ixgbe_hw
*hw
= &adapter
->hw
;
3751 struct net_device
*dev
= adapter
->netdev
;
3752 int link
, tc
, kb
, marker
;
3755 /* Calculate max LAN frame size */
3756 tc
= link
= dev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+ IXGBE_ETH_FRAMING
;
3759 /* FCoE traffic class uses FCOE jumbo frames */
3760 if ((dev
->features
& NETIF_F_FCOE_MTU
) &&
3761 (tc
< IXGBE_FCOE_JUMBO_FRAME_SIZE
) &&
3762 (pb
== ixgbe_fcoe_get_tc(adapter
)))
3763 tc
= IXGBE_FCOE_JUMBO_FRAME_SIZE
;
3766 /* Calculate delay value for device */
3767 switch (hw
->mac
.type
) {
3768 case ixgbe_mac_X540
:
3769 dv_id
= IXGBE_DV_X540(link
, tc
);
3772 dv_id
= IXGBE_DV(link
, tc
);
3776 /* Loopback switch introduces additional latency */
3777 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
3778 dv_id
+= IXGBE_B2BT(tc
);
3780 /* Delay value is calculated in bit times convert to KB */
3781 kb
= IXGBE_BT2KB(dv_id
);
3782 rx_pba
= IXGBE_READ_REG(hw
, IXGBE_RXPBSIZE(pb
)) >> 10;
3784 marker
= rx_pba
- kb
;
3786 /* It is possible that the packet buffer is not large enough
3787 * to provide required headroom. In this case throw an error
3788 * to user and a do the best we can.
3791 e_warn(drv
, "Packet Buffer(%i) can not provide enough"
3792 "headroom to support flow control."
3793 "Decrease MTU or number of traffic classes\n", pb
);
3801 * ixgbe_lpbthresh - calculate low water mark for for flow control
3803 * @adapter: board private structure to calculate for
3804 * @pb: packet buffer to calculate
3806 static int ixgbe_lpbthresh(struct ixgbe_adapter
*adapter
)
3808 struct ixgbe_hw
*hw
= &adapter
->hw
;
3809 struct net_device
*dev
= adapter
->netdev
;
3813 /* Calculate max LAN frame size */
3814 tc
= dev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
3816 /* Calculate delay value for device */
3817 switch (hw
->mac
.type
) {
3818 case ixgbe_mac_X540
:
3819 dv_id
= IXGBE_LOW_DV_X540(tc
);
3822 dv_id
= IXGBE_LOW_DV(tc
);
3826 /* Delay value is calculated in bit times convert to KB */
3827 return IXGBE_BT2KB(dv_id
);
3831 * ixgbe_pbthresh_setup - calculate and setup high low water marks
3833 static void ixgbe_pbthresh_setup(struct ixgbe_adapter
*adapter
)
3835 struct ixgbe_hw
*hw
= &adapter
->hw
;
3836 int num_tc
= netdev_get_num_tc(adapter
->netdev
);
3842 hw
->fc
.low_water
= ixgbe_lpbthresh(adapter
);
3844 for (i
= 0; i
< num_tc
; i
++) {
3845 hw
->fc
.high_water
[i
] = ixgbe_hpbthresh(adapter
, i
);
3847 /* Low water marks must not be larger than high water marks */
3848 if (hw
->fc
.low_water
> hw
->fc
.high_water
[i
])
3849 hw
->fc
.low_water
= 0;
3853 static void ixgbe_configure_pb(struct ixgbe_adapter
*adapter
)
3855 struct ixgbe_hw
*hw
= &adapter
->hw
;
3857 u8 tc
= netdev_get_num_tc(adapter
->netdev
);
3859 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
||
3860 adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
)
3861 hdrm
= 32 << adapter
->fdir_pballoc
;
3865 hw
->mac
.ops
.set_rxpba(hw
, tc
, hdrm
, PBA_STRATEGY_EQUAL
);
3866 ixgbe_pbthresh_setup(adapter
);
3869 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter
*adapter
)
3871 struct ixgbe_hw
*hw
= &adapter
->hw
;
3872 struct hlist_node
*node
, *node2
;
3873 struct ixgbe_fdir_filter
*filter
;
3875 spin_lock(&adapter
->fdir_perfect_lock
);
3877 if (!hlist_empty(&adapter
->fdir_filter_list
))
3878 ixgbe_fdir_set_input_mask_82599(hw
, &adapter
->fdir_mask
);
3880 hlist_for_each_entry_safe(filter
, node
, node2
,
3881 &adapter
->fdir_filter_list
, fdir_node
) {
3882 ixgbe_fdir_write_perfect_filter_82599(hw
,
3885 (filter
->action
== IXGBE_FDIR_DROP_QUEUE
) ?
3886 IXGBE_FDIR_DROP_QUEUE
:
3887 adapter
->rx_ring
[filter
->action
]->reg_idx
);
3890 spin_unlock(&adapter
->fdir_perfect_lock
);
3893 static void ixgbe_configure(struct ixgbe_adapter
*adapter
)
3895 struct ixgbe_hw
*hw
= &adapter
->hw
;
3897 ixgbe_configure_pb(adapter
);
3898 #ifdef CONFIG_IXGBE_DCB
3899 ixgbe_configure_dcb(adapter
);
3902 * We must restore virtualization before VLANs or else
3903 * the VLVF registers will not be populated
3905 ixgbe_configure_virtualization(adapter
);
3907 ixgbe_set_rx_mode(adapter
->netdev
);
3908 ixgbe_restore_vlan(adapter
);
3910 switch (hw
->mac
.type
) {
3911 case ixgbe_mac_82599EB
:
3912 case ixgbe_mac_X540
:
3913 hw
->mac
.ops
.disable_rx_buff(hw
);
3919 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) {
3920 ixgbe_init_fdir_signature_82599(&adapter
->hw
,
3921 adapter
->fdir_pballoc
);
3922 } else if (adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
) {
3923 ixgbe_init_fdir_perfect_82599(&adapter
->hw
,
3924 adapter
->fdir_pballoc
);
3925 ixgbe_fdir_filter_restore(adapter
);
3928 switch (hw
->mac
.type
) {
3929 case ixgbe_mac_82599EB
:
3930 case ixgbe_mac_X540
:
3931 hw
->mac
.ops
.enable_rx_buff(hw
);
3938 /* configure FCoE L2 filters, redirection table, and Rx control */
3939 ixgbe_configure_fcoe(adapter
);
3941 #endif /* IXGBE_FCOE */
3942 ixgbe_configure_tx(adapter
);
3943 ixgbe_configure_rx(adapter
);
3946 static inline bool ixgbe_is_sfp(struct ixgbe_hw
*hw
)
3948 switch (hw
->phy
.type
) {
3949 case ixgbe_phy_sfp_avago
:
3950 case ixgbe_phy_sfp_ftl
:
3951 case ixgbe_phy_sfp_intel
:
3952 case ixgbe_phy_sfp_unknown
:
3953 case ixgbe_phy_sfp_passive_tyco
:
3954 case ixgbe_phy_sfp_passive_unknown
:
3955 case ixgbe_phy_sfp_active_unknown
:
3956 case ixgbe_phy_sfp_ftl_active
:
3959 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
3967 * ixgbe_sfp_link_config - set up SFP+ link
3968 * @adapter: pointer to private adapter struct
3970 static void ixgbe_sfp_link_config(struct ixgbe_adapter
*adapter
)
3973 * We are assuming the worst case scenario here, and that
3974 * is that an SFP was inserted/removed after the reset
3975 * but before SFP detection was enabled. As such the best
3976 * solution is to just start searching as soon as we start
3978 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
)
3979 adapter
->flags2
|= IXGBE_FLAG2_SEARCH_FOR_SFP
;
3981 adapter
->flags2
|= IXGBE_FLAG2_SFP_NEEDS_RESET
;
3985 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3986 * @hw: pointer to private hardware struct
3988 * Returns 0 on success, negative on failure
3990 static int ixgbe_non_sfp_link_config(struct ixgbe_hw
*hw
)
3993 bool negotiation
, link_up
= false;
3994 u32 ret
= IXGBE_ERR_LINK_SETUP
;
3996 if (hw
->mac
.ops
.check_link
)
3997 ret
= hw
->mac
.ops
.check_link(hw
, &autoneg
, &link_up
, false);
4002 autoneg
= hw
->phy
.autoneg_advertised
;
4003 if ((!autoneg
) && (hw
->mac
.ops
.get_link_capabilities
))
4004 ret
= hw
->mac
.ops
.get_link_capabilities(hw
, &autoneg
,
4009 if (hw
->mac
.ops
.setup_link
)
4010 ret
= hw
->mac
.ops
.setup_link(hw
, autoneg
, negotiation
, link_up
);
4015 static void ixgbe_setup_gpie(struct ixgbe_adapter
*adapter
)
4017 struct ixgbe_hw
*hw
= &adapter
->hw
;
4020 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
4021 gpie
= IXGBE_GPIE_MSIX_MODE
| IXGBE_GPIE_PBA_SUPPORT
|
4023 gpie
|= IXGBE_GPIE_EIAME
;
4025 * use EIAM to auto-mask when MSI-X interrupt is asserted
4026 * this saves a register write for every interrupt
4028 switch (hw
->mac
.type
) {
4029 case ixgbe_mac_82598EB
:
4030 IXGBE_WRITE_REG(hw
, IXGBE_EIAM
, IXGBE_EICS_RTX_QUEUE
);
4032 case ixgbe_mac_82599EB
:
4033 case ixgbe_mac_X540
:
4035 IXGBE_WRITE_REG(hw
, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4036 IXGBE_WRITE_REG(hw
, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4040 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
4041 * specifically only auto mask tx and rx interrupts */
4042 IXGBE_WRITE_REG(hw
, IXGBE_EIAM
, IXGBE_EICS_RTX_QUEUE
);
4045 /* XXX: to interrupt immediately for EICS writes, enable this */
4046 /* gpie |= IXGBE_GPIE_EIMEN; */
4048 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
4049 gpie
&= ~IXGBE_GPIE_VTMODE_MASK
;
4051 switch (adapter
->ring_feature
[RING_F_VMDQ
].mask
) {
4052 case IXGBE_82599_VMDQ_8Q_MASK
:
4053 gpie
|= IXGBE_GPIE_VTMODE_16
;
4055 case IXGBE_82599_VMDQ_4Q_MASK
:
4056 gpie
|= IXGBE_GPIE_VTMODE_32
;
4059 gpie
|= IXGBE_GPIE_VTMODE_64
;
4064 /* Enable Thermal over heat sensor interrupt */
4065 if (adapter
->flags2
& IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
) {
4066 switch (adapter
->hw
.mac
.type
) {
4067 case ixgbe_mac_82599EB
:
4068 gpie
|= IXGBE_SDP0_GPIEN
;
4070 case ixgbe_mac_X540
:
4071 gpie
|= IXGBE_EIMS_TS
;
4078 /* Enable fan failure interrupt */
4079 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
)
4080 gpie
|= IXGBE_SDP1_GPIEN
;
4082 if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
4083 gpie
|= IXGBE_SDP1_GPIEN
;
4084 gpie
|= IXGBE_SDP2_GPIEN
;
4087 IXGBE_WRITE_REG(hw
, IXGBE_GPIE
, gpie
);
4090 static void ixgbe_up_complete(struct ixgbe_adapter
*adapter
)
4092 struct ixgbe_hw
*hw
= &adapter
->hw
;
4096 ixgbe_get_hw_control(adapter
);
4097 ixgbe_setup_gpie(adapter
);
4099 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)
4100 ixgbe_configure_msix(adapter
);
4102 ixgbe_configure_msi_and_legacy(adapter
);
4104 /* enable the optics for 82599 SFP+ fiber */
4105 if (hw
->mac
.ops
.enable_tx_laser
)
4106 hw
->mac
.ops
.enable_tx_laser(hw
);
4108 clear_bit(__IXGBE_DOWN
, &adapter
->state
);
4109 ixgbe_napi_enable_all(adapter
);
4111 if (ixgbe_is_sfp(hw
)) {
4112 ixgbe_sfp_link_config(adapter
);
4114 err
= ixgbe_non_sfp_link_config(hw
);
4116 e_err(probe
, "link_config FAILED %d\n", err
);
4119 /* clear any pending interrupts, may auto mask */
4120 IXGBE_READ_REG(hw
, IXGBE_EICR
);
4121 ixgbe_irq_enable(adapter
, true, true);
4124 * If this adapter has a fan, check to see if we had a failure
4125 * before we enabled the interrupt.
4127 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) {
4128 u32 esdp
= IXGBE_READ_REG(hw
, IXGBE_ESDP
);
4129 if (esdp
& IXGBE_ESDP_SDP1
)
4130 e_crit(drv
, "Fan has stopped, replace the adapter\n");
4133 /* enable transmits */
4134 netif_tx_start_all_queues(adapter
->netdev
);
4136 /* bring the link up in the watchdog, this could race with our first
4137 * link up interrupt but shouldn't be a problem */
4138 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_UPDATE
;
4139 adapter
->link_check_timeout
= jiffies
;
4140 mod_timer(&adapter
->service_timer
, jiffies
);
4142 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4143 ctrl_ext
= IXGBE_READ_REG(hw
, IXGBE_CTRL_EXT
);
4144 ctrl_ext
|= IXGBE_CTRL_EXT_PFRSTD
;
4145 IXGBE_WRITE_REG(hw
, IXGBE_CTRL_EXT
, ctrl_ext
);
4148 void ixgbe_reinit_locked(struct ixgbe_adapter
*adapter
)
4150 WARN_ON(in_interrupt());
4151 /* put off any impending NetWatchDogTimeout */
4152 adapter
->netdev
->trans_start
= jiffies
;
4154 while (test_and_set_bit(__IXGBE_RESETTING
, &adapter
->state
))
4155 usleep_range(1000, 2000);
4156 ixgbe_down(adapter
);
4158 * If SR-IOV enabled then wait a bit before bringing the adapter
4159 * back up to give the VFs time to respond to the reset. The
4160 * two second wait is based upon the watchdog timer cycle in
4163 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
4166 clear_bit(__IXGBE_RESETTING
, &adapter
->state
);
4169 void ixgbe_up(struct ixgbe_adapter
*adapter
)
4171 /* hardware has been reset, we need to reload some things */
4172 ixgbe_configure(adapter
);
4174 ixgbe_up_complete(adapter
);
4177 void ixgbe_reset(struct ixgbe_adapter
*adapter
)
4179 struct ixgbe_hw
*hw
= &adapter
->hw
;
4182 /* lock SFP init bit to prevent race conditions with the watchdog */
4183 while (test_and_set_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
))
4184 usleep_range(1000, 2000);
4186 /* clear all SFP and link config related flags while holding SFP_INIT */
4187 adapter
->flags2
&= ~(IXGBE_FLAG2_SEARCH_FOR_SFP
|
4188 IXGBE_FLAG2_SFP_NEEDS_RESET
);
4189 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_CONFIG
;
4191 err
= hw
->mac
.ops
.init_hw(hw
);
4194 case IXGBE_ERR_SFP_NOT_PRESENT
:
4195 case IXGBE_ERR_SFP_NOT_SUPPORTED
:
4197 case IXGBE_ERR_MASTER_REQUESTS_PENDING
:
4198 e_dev_err("master disable timed out\n");
4200 case IXGBE_ERR_EEPROM_VERSION
:
4201 /* We are running on a pre-production device, log a warning */
4202 e_dev_warn("This device is a pre-production adapter/LOM. "
4203 "Please be aware there may be issues associated with "
4204 "your hardware. If you are experiencing problems "
4205 "please contact your Intel or hardware "
4206 "representative who provided you with this "
4210 e_dev_err("Hardware Error: %d\n", err
);
4213 clear_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
);
4215 /* reprogram the RAR[0] in case user changed it. */
4216 hw
->mac
.ops
.set_rar(hw
, 0, hw
->mac
.addr
, VMDQ_P(0), IXGBE_RAH_AV
);
4218 /* update SAN MAC vmdq pool selection */
4219 if (hw
->mac
.san_mac_rar_index
)
4220 hw
->mac
.ops
.set_vmdq_san_mac(hw
, VMDQ_P(0));
4222 if (adapter
->flags2
& IXGBE_FLAG2_PTP_ENABLED
)
4223 ixgbe_ptp_reset(adapter
);
4227 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4228 * @rx_ring: ring to free buffers from
4230 static void ixgbe_clean_rx_ring(struct ixgbe_ring
*rx_ring
)
4232 struct device
*dev
= rx_ring
->dev
;
4236 /* ring already cleared, nothing to do */
4237 if (!rx_ring
->rx_buffer_info
)
4240 /* Free all the Rx ring sk_buffs */
4241 for (i
= 0; i
< rx_ring
->count
; i
++) {
4242 struct ixgbe_rx_buffer
*rx_buffer
;
4244 rx_buffer
= &rx_ring
->rx_buffer_info
[i
];
4245 if (rx_buffer
->skb
) {
4246 struct sk_buff
*skb
= rx_buffer
->skb
;
4247 if (IXGBE_CB(skb
)->page_released
) {
4250 ixgbe_rx_bufsz(rx_ring
),
4252 IXGBE_CB(skb
)->page_released
= false;
4256 rx_buffer
->skb
= NULL
;
4258 dma_unmap_page(dev
, rx_buffer
->dma
,
4259 ixgbe_rx_pg_size(rx_ring
),
4262 if (rx_buffer
->page
)
4263 __free_pages(rx_buffer
->page
,
4264 ixgbe_rx_pg_order(rx_ring
));
4265 rx_buffer
->page
= NULL
;
4268 size
= sizeof(struct ixgbe_rx_buffer
) * rx_ring
->count
;
4269 memset(rx_ring
->rx_buffer_info
, 0, size
);
4271 /* Zero out the descriptor ring */
4272 memset(rx_ring
->desc
, 0, rx_ring
->size
);
4274 rx_ring
->next_to_alloc
= 0;
4275 rx_ring
->next_to_clean
= 0;
4276 rx_ring
->next_to_use
= 0;
4280 * ixgbe_clean_tx_ring - Free Tx Buffers
4281 * @tx_ring: ring to be cleaned
4283 static void ixgbe_clean_tx_ring(struct ixgbe_ring
*tx_ring
)
4285 struct ixgbe_tx_buffer
*tx_buffer_info
;
4289 /* ring already cleared, nothing to do */
4290 if (!tx_ring
->tx_buffer_info
)
4293 /* Free all the Tx ring sk_buffs */
4294 for (i
= 0; i
< tx_ring
->count
; i
++) {
4295 tx_buffer_info
= &tx_ring
->tx_buffer_info
[i
];
4296 ixgbe_unmap_and_free_tx_resource(tx_ring
, tx_buffer_info
);
4299 netdev_tx_reset_queue(txring_txq(tx_ring
));
4301 size
= sizeof(struct ixgbe_tx_buffer
) * tx_ring
->count
;
4302 memset(tx_ring
->tx_buffer_info
, 0, size
);
4304 /* Zero out the descriptor ring */
4305 memset(tx_ring
->desc
, 0, tx_ring
->size
);
4307 tx_ring
->next_to_use
= 0;
4308 tx_ring
->next_to_clean
= 0;
4312 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4313 * @adapter: board private structure
4315 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter
*adapter
)
4319 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
4320 ixgbe_clean_rx_ring(adapter
->rx_ring
[i
]);
4324 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4325 * @adapter: board private structure
4327 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter
*adapter
)
4331 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
4332 ixgbe_clean_tx_ring(adapter
->tx_ring
[i
]);
4335 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter
*adapter
)
4337 struct hlist_node
*node
, *node2
;
4338 struct ixgbe_fdir_filter
*filter
;
4340 spin_lock(&adapter
->fdir_perfect_lock
);
4342 hlist_for_each_entry_safe(filter
, node
, node2
,
4343 &adapter
->fdir_filter_list
, fdir_node
) {
4344 hlist_del(&filter
->fdir_node
);
4347 adapter
->fdir_filter_count
= 0;
4349 spin_unlock(&adapter
->fdir_perfect_lock
);
4352 void ixgbe_down(struct ixgbe_adapter
*adapter
)
4354 struct net_device
*netdev
= adapter
->netdev
;
4355 struct ixgbe_hw
*hw
= &adapter
->hw
;
4359 /* signal that we are down to the interrupt handler */
4360 set_bit(__IXGBE_DOWN
, &adapter
->state
);
4362 /* disable receives */
4363 rxctrl
= IXGBE_READ_REG(hw
, IXGBE_RXCTRL
);
4364 IXGBE_WRITE_REG(hw
, IXGBE_RXCTRL
, rxctrl
& ~IXGBE_RXCTRL_RXEN
);
4366 /* disable all enabled rx queues */
4367 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
4368 /* this call also flushes the previous write */
4369 ixgbe_disable_rx_queue(adapter
, adapter
->rx_ring
[i
]);
4371 usleep_range(10000, 20000);
4373 netif_tx_stop_all_queues(netdev
);
4375 /* call carrier off first to avoid false dev_watchdog timeouts */
4376 netif_carrier_off(netdev
);
4377 netif_tx_disable(netdev
);
4379 ixgbe_irq_disable(adapter
);
4381 ixgbe_napi_disable_all(adapter
);
4383 adapter
->flags2
&= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT
|
4384 IXGBE_FLAG2_RESET_REQUESTED
);
4385 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_UPDATE
;
4387 del_timer_sync(&adapter
->service_timer
);
4389 if (adapter
->num_vfs
) {
4390 /* Clear EITR Select mapping */
4391 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EITRSEL
, 0);
4393 /* Mark all the VFs as inactive */
4394 for (i
= 0 ; i
< adapter
->num_vfs
; i
++)
4395 adapter
->vfinfo
[i
].clear_to_send
= false;
4397 /* ping all the active vfs to let them know we are going down */
4398 ixgbe_ping_all_vfs(adapter
);
4400 /* Disable all VFTE/VFRE TX/RX */
4401 ixgbe_disable_tx_rx(adapter
);
4404 /* disable transmits in the hardware now that interrupts are off */
4405 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
4406 u8 reg_idx
= adapter
->tx_ring
[i
]->reg_idx
;
4407 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(reg_idx
), IXGBE_TXDCTL_SWFLSH
);
4410 /* Disable the Tx DMA engine on 82599 and X540 */
4411 switch (hw
->mac
.type
) {
4412 case ixgbe_mac_82599EB
:
4413 case ixgbe_mac_X540
:
4414 IXGBE_WRITE_REG(hw
, IXGBE_DMATXCTL
,
4415 (IXGBE_READ_REG(hw
, IXGBE_DMATXCTL
) &
4416 ~IXGBE_DMATXCTL_TE
));
4422 if (!pci_channel_offline(adapter
->pdev
))
4423 ixgbe_reset(adapter
);
4425 /* power down the optics for 82599 SFP+ fiber */
4426 if (hw
->mac
.ops
.disable_tx_laser
)
4427 hw
->mac
.ops
.disable_tx_laser(hw
);
4429 ixgbe_clean_all_tx_rings(adapter
);
4430 ixgbe_clean_all_rx_rings(adapter
);
4432 #ifdef CONFIG_IXGBE_DCA
4433 /* since we reset the hardware DCA settings were cleared */
4434 ixgbe_setup_dca(adapter
);
4439 * ixgbe_tx_timeout - Respond to a Tx Hang
4440 * @netdev: network interface device structure
4442 static void ixgbe_tx_timeout(struct net_device
*netdev
)
4444 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4446 /* Do the reset outside of interrupt context */
4447 ixgbe_tx_timeout_reset(adapter
);
4451 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4452 * @adapter: board private structure to initialize
4454 * ixgbe_sw_init initializes the Adapter private data structure.
4455 * Fields are initialized based on PCI device information and
4456 * OS network device settings (MTU size).
4458 static int __devinit
ixgbe_sw_init(struct ixgbe_adapter
*adapter
)
4460 struct ixgbe_hw
*hw
= &adapter
->hw
;
4461 struct pci_dev
*pdev
= adapter
->pdev
;
4463 #ifdef CONFIG_IXGBE_DCB
4465 struct tc_configuration
*tc
;
4468 /* PCI config space info */
4470 hw
->vendor_id
= pdev
->vendor
;
4471 hw
->device_id
= pdev
->device
;
4472 hw
->revision_id
= pdev
->revision
;
4473 hw
->subsystem_vendor_id
= pdev
->subsystem_vendor
;
4474 hw
->subsystem_device_id
= pdev
->subsystem_device
;
4476 /* Set capability flags */
4477 rss
= min_t(int, IXGBE_MAX_RSS_INDICES
, num_online_cpus());
4478 adapter
->ring_feature
[RING_F_RSS
].limit
= rss
;
4479 switch (hw
->mac
.type
) {
4480 case ixgbe_mac_82598EB
:
4481 if (hw
->device_id
== IXGBE_DEV_ID_82598AT
)
4482 adapter
->flags
|= IXGBE_FLAG_FAN_FAIL_CAPABLE
;
4483 adapter
->max_q_vectors
= MAX_Q_VECTORS_82598
;
4485 case ixgbe_mac_X540
:
4486 adapter
->flags2
|= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
;
4487 case ixgbe_mac_82599EB
:
4488 adapter
->max_q_vectors
= MAX_Q_VECTORS_82599
;
4489 adapter
->flags2
|= IXGBE_FLAG2_RSC_CAPABLE
;
4490 adapter
->flags2
|= IXGBE_FLAG2_RSC_ENABLED
;
4491 if (hw
->device_id
== IXGBE_DEV_ID_82599_T3_LOM
)
4492 adapter
->flags2
|= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE
;
4493 /* Flow Director hash filters enabled */
4494 adapter
->atr_sample_rate
= 20;
4495 adapter
->ring_feature
[RING_F_FDIR
].limit
=
4496 IXGBE_MAX_FDIR_INDICES
;
4497 adapter
->fdir_pballoc
= IXGBE_FDIR_PBALLOC_64K
;
4499 adapter
->flags
|= IXGBE_FLAG_FCOE_CAPABLE
;
4500 adapter
->flags
&= ~IXGBE_FLAG_FCOE_ENABLED
;
4501 #ifdef CONFIG_IXGBE_DCB
4502 /* Default traffic class to use for FCoE */
4503 adapter
->fcoe
.up
= IXGBE_FCOE_DEFTC
;
4505 #endif /* IXGBE_FCOE */
4512 /* FCoE support exists, always init the FCoE lock */
4513 spin_lock_init(&adapter
->fcoe
.lock
);
4516 /* n-tuple support exists, always init our spinlock */
4517 spin_lock_init(&adapter
->fdir_perfect_lock
);
4519 #ifdef CONFIG_IXGBE_DCB
4520 switch (hw
->mac
.type
) {
4521 case ixgbe_mac_X540
:
4522 adapter
->dcb_cfg
.num_tcs
.pg_tcs
= X540_TRAFFIC_CLASS
;
4523 adapter
->dcb_cfg
.num_tcs
.pfc_tcs
= X540_TRAFFIC_CLASS
;
4526 adapter
->dcb_cfg
.num_tcs
.pg_tcs
= MAX_TRAFFIC_CLASS
;
4527 adapter
->dcb_cfg
.num_tcs
.pfc_tcs
= MAX_TRAFFIC_CLASS
;
4531 /* Configure DCB traffic classes */
4532 for (j
= 0; j
< MAX_TRAFFIC_CLASS
; j
++) {
4533 tc
= &adapter
->dcb_cfg
.tc_config
[j
];
4534 tc
->path
[DCB_TX_CONFIG
].bwg_id
= 0;
4535 tc
->path
[DCB_TX_CONFIG
].bwg_percent
= 12 + (j
& 1);
4536 tc
->path
[DCB_RX_CONFIG
].bwg_id
= 0;
4537 tc
->path
[DCB_RX_CONFIG
].bwg_percent
= 12 + (j
& 1);
4538 tc
->dcb_pfc
= pfc_disabled
;
4541 /* Initialize default user to priority mapping, UPx->TC0 */
4542 tc
= &adapter
->dcb_cfg
.tc_config
[0];
4543 tc
->path
[DCB_TX_CONFIG
].up_to_tc_bitmap
= 0xFF;
4544 tc
->path
[DCB_RX_CONFIG
].up_to_tc_bitmap
= 0xFF;
4546 adapter
->dcb_cfg
.bw_percentage
[DCB_TX_CONFIG
][0] = 100;
4547 adapter
->dcb_cfg
.bw_percentage
[DCB_RX_CONFIG
][0] = 100;
4548 adapter
->dcb_cfg
.pfc_mode_enable
= false;
4549 adapter
->dcb_set_bitmap
= 0x00;
4550 adapter
->dcbx_cap
= DCB_CAP_DCBX_HOST
| DCB_CAP_DCBX_VER_CEE
;
4551 memcpy(&adapter
->temp_dcb_cfg
, &adapter
->dcb_cfg
,
4552 sizeof(adapter
->temp_dcb_cfg
));
4556 /* default flow control settings */
4557 hw
->fc
.requested_mode
= ixgbe_fc_full
;
4558 hw
->fc
.current_mode
= ixgbe_fc_full
; /* init for ethtool output */
4559 ixgbe_pbthresh_setup(adapter
);
4560 hw
->fc
.pause_time
= IXGBE_DEFAULT_FCPAUSE
;
4561 hw
->fc
.send_xon
= true;
4562 hw
->fc
.disable_fc_autoneg
=
4563 (ixgbe_device_supports_autoneg_fc(hw
) == 0) ? false : true;
4565 #ifdef CONFIG_PCI_IOV
4566 /* assign number of SR-IOV VFs */
4567 if (hw
->mac
.type
!= ixgbe_mac_82598EB
)
4568 adapter
->num_vfs
= (max_vfs
> 63) ? 0 : max_vfs
;
4571 /* enable itr by default in dynamic mode */
4572 adapter
->rx_itr_setting
= 1;
4573 adapter
->tx_itr_setting
= 1;
4575 /* set default ring sizes */
4576 adapter
->tx_ring_count
= IXGBE_DEFAULT_TXD
;
4577 adapter
->rx_ring_count
= IXGBE_DEFAULT_RXD
;
4579 /* set default work limits */
4580 adapter
->tx_work_limit
= IXGBE_DEFAULT_TX_WORK
;
4582 /* initialize eeprom parameters */
4583 if (ixgbe_init_eeprom_params_generic(hw
)) {
4584 e_dev_err("EEPROM initialization failed\n");
4588 set_bit(__IXGBE_DOWN
, &adapter
->state
);
4594 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4595 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4597 * Return 0 on success, negative on failure
4599 int ixgbe_setup_tx_resources(struct ixgbe_ring
*tx_ring
)
4601 struct device
*dev
= tx_ring
->dev
;
4602 int orig_node
= dev_to_node(dev
);
4606 size
= sizeof(struct ixgbe_tx_buffer
) * tx_ring
->count
;
4608 if (tx_ring
->q_vector
)
4609 numa_node
= tx_ring
->q_vector
->numa_node
;
4611 tx_ring
->tx_buffer_info
= vzalloc_node(size
, numa_node
);
4612 if (!tx_ring
->tx_buffer_info
)
4613 tx_ring
->tx_buffer_info
= vzalloc(size
);
4614 if (!tx_ring
->tx_buffer_info
)
4617 /* round up to nearest 4K */
4618 tx_ring
->size
= tx_ring
->count
* sizeof(union ixgbe_adv_tx_desc
);
4619 tx_ring
->size
= ALIGN(tx_ring
->size
, 4096);
4621 set_dev_node(dev
, numa_node
);
4622 tx_ring
->desc
= dma_alloc_coherent(dev
,
4626 set_dev_node(dev
, orig_node
);
4628 tx_ring
->desc
= dma_alloc_coherent(dev
, tx_ring
->size
,
4629 &tx_ring
->dma
, GFP_KERNEL
);
4633 tx_ring
->next_to_use
= 0;
4634 tx_ring
->next_to_clean
= 0;
4638 vfree(tx_ring
->tx_buffer_info
);
4639 tx_ring
->tx_buffer_info
= NULL
;
4640 dev_err(dev
, "Unable to allocate memory for the Tx descriptor ring\n");
4645 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4646 * @adapter: board private structure
4648 * If this function returns with an error, then it's possible one or
4649 * more of the rings is populated (while the rest are not). It is the
4650 * callers duty to clean those orphaned rings.
4652 * Return 0 on success, negative on failure
4654 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter
*adapter
)
4658 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
4659 err
= ixgbe_setup_tx_resources(adapter
->tx_ring
[i
]);
4663 e_err(probe
, "Allocation for Tx Queue %u failed\n", i
);
4669 /* rewind the index freeing the rings as we go */
4671 ixgbe_free_tx_resources(adapter
->tx_ring
[i
]);
4676 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4677 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4679 * Returns 0 on success, negative on failure
4681 int ixgbe_setup_rx_resources(struct ixgbe_ring
*rx_ring
)
4683 struct device
*dev
= rx_ring
->dev
;
4684 int orig_node
= dev_to_node(dev
);
4688 size
= sizeof(struct ixgbe_rx_buffer
) * rx_ring
->count
;
4690 if (rx_ring
->q_vector
)
4691 numa_node
= rx_ring
->q_vector
->numa_node
;
4693 rx_ring
->rx_buffer_info
= vzalloc_node(size
, numa_node
);
4694 if (!rx_ring
->rx_buffer_info
)
4695 rx_ring
->rx_buffer_info
= vzalloc(size
);
4696 if (!rx_ring
->rx_buffer_info
)
4699 /* Round up to nearest 4K */
4700 rx_ring
->size
= rx_ring
->count
* sizeof(union ixgbe_adv_rx_desc
);
4701 rx_ring
->size
= ALIGN(rx_ring
->size
, 4096);
4703 set_dev_node(dev
, numa_node
);
4704 rx_ring
->desc
= dma_alloc_coherent(dev
,
4708 set_dev_node(dev
, orig_node
);
4710 rx_ring
->desc
= dma_alloc_coherent(dev
, rx_ring
->size
,
4711 &rx_ring
->dma
, GFP_KERNEL
);
4715 rx_ring
->next_to_clean
= 0;
4716 rx_ring
->next_to_use
= 0;
4720 vfree(rx_ring
->rx_buffer_info
);
4721 rx_ring
->rx_buffer_info
= NULL
;
4722 dev_err(dev
, "Unable to allocate memory for the Rx descriptor ring\n");
4727 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4728 * @adapter: board private structure
4730 * If this function returns with an error, then it's possible one or
4731 * more of the rings is populated (while the rest are not). It is the
4732 * callers duty to clean those orphaned rings.
4734 * Return 0 on success, negative on failure
4736 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter
*adapter
)
4740 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
4741 err
= ixgbe_setup_rx_resources(adapter
->rx_ring
[i
]);
4745 e_err(probe
, "Allocation for Rx Queue %u failed\n", i
);
4750 err
= ixgbe_setup_fcoe_ddp_resources(adapter
);
4755 /* rewind the index freeing the rings as we go */
4757 ixgbe_free_rx_resources(adapter
->rx_ring
[i
]);
4762 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4763 * @tx_ring: Tx descriptor ring for a specific queue
4765 * Free all transmit software resources
4767 void ixgbe_free_tx_resources(struct ixgbe_ring
*tx_ring
)
4769 ixgbe_clean_tx_ring(tx_ring
);
4771 vfree(tx_ring
->tx_buffer_info
);
4772 tx_ring
->tx_buffer_info
= NULL
;
4774 /* if not set, then don't free */
4778 dma_free_coherent(tx_ring
->dev
, tx_ring
->size
,
4779 tx_ring
->desc
, tx_ring
->dma
);
4781 tx_ring
->desc
= NULL
;
4785 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4786 * @adapter: board private structure
4788 * Free all transmit software resources
4790 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter
*adapter
)
4794 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
4795 if (adapter
->tx_ring
[i
]->desc
)
4796 ixgbe_free_tx_resources(adapter
->tx_ring
[i
]);
4800 * ixgbe_free_rx_resources - Free Rx Resources
4801 * @rx_ring: ring to clean the resources from
4803 * Free all receive software resources
4805 void ixgbe_free_rx_resources(struct ixgbe_ring
*rx_ring
)
4807 ixgbe_clean_rx_ring(rx_ring
);
4809 vfree(rx_ring
->rx_buffer_info
);
4810 rx_ring
->rx_buffer_info
= NULL
;
4812 /* if not set, then don't free */
4816 dma_free_coherent(rx_ring
->dev
, rx_ring
->size
,
4817 rx_ring
->desc
, rx_ring
->dma
);
4819 rx_ring
->desc
= NULL
;
4823 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4824 * @adapter: board private structure
4826 * Free all receive software resources
4828 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter
*adapter
)
4833 ixgbe_free_fcoe_ddp_resources(adapter
);
4836 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
4837 if (adapter
->rx_ring
[i
]->desc
)
4838 ixgbe_free_rx_resources(adapter
->rx_ring
[i
]);
4842 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4843 * @netdev: network interface device structure
4844 * @new_mtu: new value for maximum frame size
4846 * Returns 0 on success, negative on failure
4848 static int ixgbe_change_mtu(struct net_device
*netdev
, int new_mtu
)
4850 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4851 int max_frame
= new_mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
4853 /* MTU < 68 is an error and causes problems on some kernels */
4854 if ((new_mtu
< 68) || (max_frame
> IXGBE_MAX_JUMBO_FRAME_SIZE
))
4858 * For 82599EB we cannot allow legacy VFs to enable their receive
4859 * paths when MTU greater than 1500 is configured. So display a
4860 * warning that legacy VFs will be disabled.
4862 if ((adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) &&
4863 (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
) &&
4864 (max_frame
> MAXIMUM_ETHERNET_VLAN_SIZE
))
4865 e_warn(probe
, "Setting MTU > 1500 will disable legacy VFs\n");
4867 e_info(probe
, "changing MTU from %d to %d\n", netdev
->mtu
, new_mtu
);
4869 /* must set new MTU before calling down or up */
4870 netdev
->mtu
= new_mtu
;
4872 if (netif_running(netdev
))
4873 ixgbe_reinit_locked(adapter
);
4879 * ixgbe_open - Called when a network interface is made active
4880 * @netdev: network interface device structure
4882 * Returns 0 on success, negative value on failure
4884 * The open entry point is called when a network interface is made
4885 * active by the system (IFF_UP). At this point all resources needed
4886 * for transmit and receive operations are allocated, the interrupt
4887 * handler is registered with the OS, the watchdog timer is started,
4888 * and the stack is notified that the interface is ready.
4890 static int ixgbe_open(struct net_device
*netdev
)
4892 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4895 /* disallow open during test */
4896 if (test_bit(__IXGBE_TESTING
, &adapter
->state
))
4899 netif_carrier_off(netdev
);
4901 /* allocate transmit descriptors */
4902 err
= ixgbe_setup_all_tx_resources(adapter
);
4906 /* allocate receive descriptors */
4907 err
= ixgbe_setup_all_rx_resources(adapter
);
4911 ixgbe_configure(adapter
);
4913 err
= ixgbe_request_irq(adapter
);
4917 /* Notify the stack of the actual queue counts. */
4918 err
= netif_set_real_num_tx_queues(netdev
,
4919 adapter
->num_rx_pools
> 1 ? 1 :
4920 adapter
->num_tx_queues
);
4922 goto err_set_queues
;
4925 err
= netif_set_real_num_rx_queues(netdev
,
4926 adapter
->num_rx_pools
> 1 ? 1 :
4927 adapter
->num_rx_queues
);
4929 goto err_set_queues
;
4931 ixgbe_ptp_init(adapter
);
4933 ixgbe_up_complete(adapter
);
4938 ixgbe_free_irq(adapter
);
4940 ixgbe_free_all_rx_resources(adapter
);
4942 ixgbe_free_all_tx_resources(adapter
);
4944 ixgbe_reset(adapter
);
4950 * ixgbe_close - Disables a network interface
4951 * @netdev: network interface device structure
4953 * Returns 0, this is not allowed to fail
4955 * The close entry point is called when an interface is de-activated
4956 * by the OS. The hardware is still under the drivers control, but
4957 * needs to be disabled. A global MAC reset is issued to stop the
4958 * hardware, and all transmit and receive resources are freed.
4960 static int ixgbe_close(struct net_device
*netdev
)
4962 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4964 ixgbe_ptp_stop(adapter
);
4966 ixgbe_down(adapter
);
4967 ixgbe_free_irq(adapter
);
4969 ixgbe_fdir_filter_exit(adapter
);
4971 ixgbe_free_all_tx_resources(adapter
);
4972 ixgbe_free_all_rx_resources(adapter
);
4974 ixgbe_release_hw_control(adapter
);
4980 static int ixgbe_resume(struct pci_dev
*pdev
)
4982 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
4983 struct net_device
*netdev
= adapter
->netdev
;
4986 pci_set_power_state(pdev
, PCI_D0
);
4987 pci_restore_state(pdev
);
4989 * pci_restore_state clears dev->state_saved so call
4990 * pci_save_state to restore it.
4992 pci_save_state(pdev
);
4994 err
= pci_enable_device_mem(pdev
);
4996 e_dev_err("Cannot enable PCI device from suspend\n");
4999 pci_set_master(pdev
);
5001 pci_wake_from_d3(pdev
, false);
5003 ixgbe_reset(adapter
);
5005 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_WUS
, ~0);
5008 err
= ixgbe_init_interrupt_scheme(adapter
);
5009 if (!err
&& netif_running(netdev
))
5010 err
= ixgbe_open(netdev
);
5017 netif_device_attach(netdev
);
5021 #endif /* CONFIG_PM */
5023 static int __ixgbe_shutdown(struct pci_dev
*pdev
, bool *enable_wake
)
5025 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
5026 struct net_device
*netdev
= adapter
->netdev
;
5027 struct ixgbe_hw
*hw
= &adapter
->hw
;
5029 u32 wufc
= adapter
->wol
;
5034 netif_device_detach(netdev
);
5036 if (netif_running(netdev
)) {
5038 ixgbe_down(adapter
);
5039 ixgbe_free_irq(adapter
);
5040 ixgbe_free_all_tx_resources(adapter
);
5041 ixgbe_free_all_rx_resources(adapter
);
5045 ixgbe_clear_interrupt_scheme(adapter
);
5048 retval
= pci_save_state(pdev
);
5054 ixgbe_set_rx_mode(netdev
);
5056 /* enable the optics for 82599 SFP+ fiber as we can WoL */
5057 if (hw
->mac
.ops
.enable_tx_laser
)
5058 hw
->mac
.ops
.enable_tx_laser(hw
);
5060 /* turn on all-multi mode if wake on multicast is enabled */
5061 if (wufc
& IXGBE_WUFC_MC
) {
5062 fctrl
= IXGBE_READ_REG(hw
, IXGBE_FCTRL
);
5063 fctrl
|= IXGBE_FCTRL_MPE
;
5064 IXGBE_WRITE_REG(hw
, IXGBE_FCTRL
, fctrl
);
5067 ctrl
= IXGBE_READ_REG(hw
, IXGBE_CTRL
);
5068 ctrl
|= IXGBE_CTRL_GIO_DIS
;
5069 IXGBE_WRITE_REG(hw
, IXGBE_CTRL
, ctrl
);
5071 IXGBE_WRITE_REG(hw
, IXGBE_WUFC
, wufc
);
5073 IXGBE_WRITE_REG(hw
, IXGBE_WUC
, 0);
5074 IXGBE_WRITE_REG(hw
, IXGBE_WUFC
, 0);
5077 switch (hw
->mac
.type
) {
5078 case ixgbe_mac_82598EB
:
5079 pci_wake_from_d3(pdev
, false);
5081 case ixgbe_mac_82599EB
:
5082 case ixgbe_mac_X540
:
5083 pci_wake_from_d3(pdev
, !!wufc
);
5089 *enable_wake
= !!wufc
;
5091 ixgbe_release_hw_control(adapter
);
5093 pci_disable_device(pdev
);
5099 static int ixgbe_suspend(struct pci_dev
*pdev
, pm_message_t state
)
5104 retval
= __ixgbe_shutdown(pdev
, &wake
);
5109 pci_prepare_to_sleep(pdev
);
5111 pci_wake_from_d3(pdev
, false);
5112 pci_set_power_state(pdev
, PCI_D3hot
);
5117 #endif /* CONFIG_PM */
5119 static void ixgbe_shutdown(struct pci_dev
*pdev
)
5123 __ixgbe_shutdown(pdev
, &wake
);
5125 if (system_state
== SYSTEM_POWER_OFF
) {
5126 pci_wake_from_d3(pdev
, wake
);
5127 pci_set_power_state(pdev
, PCI_D3hot
);
5132 * ixgbe_update_stats - Update the board statistics counters.
5133 * @adapter: board private structure
5135 void ixgbe_update_stats(struct ixgbe_adapter
*adapter
)
5137 struct net_device
*netdev
= adapter
->netdev
;
5138 struct ixgbe_hw
*hw
= &adapter
->hw
;
5139 struct ixgbe_hw_stats
*hwstats
= &adapter
->stats
;
5141 u32 i
, missed_rx
= 0, mpc
, bprc
, lxon
, lxoff
, xon_off_tot
;
5142 u64 non_eop_descs
= 0, restart_queue
= 0, tx_busy
= 0;
5143 u64 alloc_rx_page_failed
= 0, alloc_rx_buff_failed
= 0;
5144 u64 bytes
= 0, packets
= 0, hw_csum_rx_error
= 0;
5146 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
5147 test_bit(__IXGBE_RESETTING
, &adapter
->state
))
5150 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
) {
5153 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
5154 rsc_count
+= adapter
->rx_ring
[i
]->rx_stats
.rsc_count
;
5155 rsc_flush
+= adapter
->rx_ring
[i
]->rx_stats
.rsc_flush
;
5157 adapter
->rsc_total_count
= rsc_count
;
5158 adapter
->rsc_total_flush
= rsc_flush
;
5161 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
5162 struct ixgbe_ring
*rx_ring
= adapter
->rx_ring
[i
];
5163 non_eop_descs
+= rx_ring
->rx_stats
.non_eop_descs
;
5164 alloc_rx_page_failed
+= rx_ring
->rx_stats
.alloc_rx_page_failed
;
5165 alloc_rx_buff_failed
+= rx_ring
->rx_stats
.alloc_rx_buff_failed
;
5166 hw_csum_rx_error
+= rx_ring
->rx_stats
.csum_err
;
5167 bytes
+= rx_ring
->stats
.bytes
;
5168 packets
+= rx_ring
->stats
.packets
;
5170 adapter
->non_eop_descs
= non_eop_descs
;
5171 adapter
->alloc_rx_page_failed
= alloc_rx_page_failed
;
5172 adapter
->alloc_rx_buff_failed
= alloc_rx_buff_failed
;
5173 adapter
->hw_csum_rx_error
= hw_csum_rx_error
;
5174 netdev
->stats
.rx_bytes
= bytes
;
5175 netdev
->stats
.rx_packets
= packets
;
5179 /* gather some stats to the adapter struct that are per queue */
5180 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
5181 struct ixgbe_ring
*tx_ring
= adapter
->tx_ring
[i
];
5182 restart_queue
+= tx_ring
->tx_stats
.restart_queue
;
5183 tx_busy
+= tx_ring
->tx_stats
.tx_busy
;
5184 bytes
+= tx_ring
->stats
.bytes
;
5185 packets
+= tx_ring
->stats
.packets
;
5187 adapter
->restart_queue
= restart_queue
;
5188 adapter
->tx_busy
= tx_busy
;
5189 netdev
->stats
.tx_bytes
= bytes
;
5190 netdev
->stats
.tx_packets
= packets
;
5192 hwstats
->crcerrs
+= IXGBE_READ_REG(hw
, IXGBE_CRCERRS
);
5194 /* 8 register reads */
5195 for (i
= 0; i
< 8; i
++) {
5196 /* for packet buffers not used, the register should read 0 */
5197 mpc
= IXGBE_READ_REG(hw
, IXGBE_MPC(i
));
5199 hwstats
->mpc
[i
] += mpc
;
5200 total_mpc
+= hwstats
->mpc
[i
];
5201 hwstats
->pxontxc
[i
] += IXGBE_READ_REG(hw
, IXGBE_PXONTXC(i
));
5202 hwstats
->pxofftxc
[i
] += IXGBE_READ_REG(hw
, IXGBE_PXOFFTXC(i
));
5203 switch (hw
->mac
.type
) {
5204 case ixgbe_mac_82598EB
:
5205 hwstats
->rnbc
[i
] += IXGBE_READ_REG(hw
, IXGBE_RNBC(i
));
5206 hwstats
->qbtc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBTC(i
));
5207 hwstats
->qbrc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBRC(i
));
5208 hwstats
->pxonrxc
[i
] +=
5209 IXGBE_READ_REG(hw
, IXGBE_PXONRXC(i
));
5211 case ixgbe_mac_82599EB
:
5212 case ixgbe_mac_X540
:
5213 hwstats
->pxonrxc
[i
] +=
5214 IXGBE_READ_REG(hw
, IXGBE_PXONRXCNT(i
));
5221 /*16 register reads */
5222 for (i
= 0; i
< 16; i
++) {
5223 hwstats
->qptc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QPTC(i
));
5224 hwstats
->qprc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QPRC(i
));
5225 if ((hw
->mac
.type
== ixgbe_mac_82599EB
) ||
5226 (hw
->mac
.type
== ixgbe_mac_X540
)) {
5227 hwstats
->qbtc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBTC_L(i
));
5228 IXGBE_READ_REG(hw
, IXGBE_QBTC_H(i
)); /* to clear */
5229 hwstats
->qbrc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBRC_L(i
));
5230 IXGBE_READ_REG(hw
, IXGBE_QBRC_H(i
)); /* to clear */
5234 hwstats
->gprc
+= IXGBE_READ_REG(hw
, IXGBE_GPRC
);
5235 /* work around hardware counting issue */
5236 hwstats
->gprc
-= missed_rx
;
5238 ixgbe_update_xoff_received(adapter
);
5240 /* 82598 hardware only has a 32 bit counter in the high register */
5241 switch (hw
->mac
.type
) {
5242 case ixgbe_mac_82598EB
:
5243 hwstats
->lxonrxc
+= IXGBE_READ_REG(hw
, IXGBE_LXONRXC
);
5244 hwstats
->gorc
+= IXGBE_READ_REG(hw
, IXGBE_GORCH
);
5245 hwstats
->gotc
+= IXGBE_READ_REG(hw
, IXGBE_GOTCH
);
5246 hwstats
->tor
+= IXGBE_READ_REG(hw
, IXGBE_TORH
);
5248 case ixgbe_mac_X540
:
5249 /* OS2BMC stats are X540 only*/
5250 hwstats
->o2bgptc
+= IXGBE_READ_REG(hw
, IXGBE_O2BGPTC
);
5251 hwstats
->o2bspc
+= IXGBE_READ_REG(hw
, IXGBE_O2BSPC
);
5252 hwstats
->b2ospc
+= IXGBE_READ_REG(hw
, IXGBE_B2OSPC
);
5253 hwstats
->b2ogprc
+= IXGBE_READ_REG(hw
, IXGBE_B2OGPRC
);
5254 case ixgbe_mac_82599EB
:
5255 for (i
= 0; i
< 16; i
++)
5256 adapter
->hw_rx_no_dma_resources
+=
5257 IXGBE_READ_REG(hw
, IXGBE_QPRDC(i
));
5258 hwstats
->gorc
+= IXGBE_READ_REG(hw
, IXGBE_GORCL
);
5259 IXGBE_READ_REG(hw
, IXGBE_GORCH
); /* to clear */
5260 hwstats
->gotc
+= IXGBE_READ_REG(hw
, IXGBE_GOTCL
);
5261 IXGBE_READ_REG(hw
, IXGBE_GOTCH
); /* to clear */
5262 hwstats
->tor
+= IXGBE_READ_REG(hw
, IXGBE_TORL
);
5263 IXGBE_READ_REG(hw
, IXGBE_TORH
); /* to clear */
5264 hwstats
->lxonrxc
+= IXGBE_READ_REG(hw
, IXGBE_LXONRXCNT
);
5265 hwstats
->fdirmatch
+= IXGBE_READ_REG(hw
, IXGBE_FDIRMATCH
);
5266 hwstats
->fdirmiss
+= IXGBE_READ_REG(hw
, IXGBE_FDIRMISS
);
5268 hwstats
->fccrc
+= IXGBE_READ_REG(hw
, IXGBE_FCCRC
);
5269 hwstats
->fcoerpdc
+= IXGBE_READ_REG(hw
, IXGBE_FCOERPDC
);
5270 hwstats
->fcoeprc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEPRC
);
5271 hwstats
->fcoeptc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEPTC
);
5272 hwstats
->fcoedwrc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEDWRC
);
5273 hwstats
->fcoedwtc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEDWTC
);
5274 /* Add up per cpu counters for total ddp aloc fail */
5275 if (adapter
->fcoe
.ddp_pool
) {
5276 struct ixgbe_fcoe
*fcoe
= &adapter
->fcoe
;
5277 struct ixgbe_fcoe_ddp_pool
*ddp_pool
;
5279 u64 noddp
= 0, noddp_ext_buff
= 0;
5280 for_each_possible_cpu(cpu
) {
5281 ddp_pool
= per_cpu_ptr(fcoe
->ddp_pool
, cpu
);
5282 noddp
+= ddp_pool
->noddp
;
5283 noddp_ext_buff
+= ddp_pool
->noddp_ext_buff
;
5285 hwstats
->fcoe_noddp
= noddp
;
5286 hwstats
->fcoe_noddp_ext_buff
= noddp_ext_buff
;
5288 #endif /* IXGBE_FCOE */
5293 bprc
= IXGBE_READ_REG(hw
, IXGBE_BPRC
);
5294 hwstats
->bprc
+= bprc
;
5295 hwstats
->mprc
+= IXGBE_READ_REG(hw
, IXGBE_MPRC
);
5296 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
5297 hwstats
->mprc
-= bprc
;
5298 hwstats
->roc
+= IXGBE_READ_REG(hw
, IXGBE_ROC
);
5299 hwstats
->prc64
+= IXGBE_READ_REG(hw
, IXGBE_PRC64
);
5300 hwstats
->prc127
+= IXGBE_READ_REG(hw
, IXGBE_PRC127
);
5301 hwstats
->prc255
+= IXGBE_READ_REG(hw
, IXGBE_PRC255
);
5302 hwstats
->prc511
+= IXGBE_READ_REG(hw
, IXGBE_PRC511
);
5303 hwstats
->prc1023
+= IXGBE_READ_REG(hw
, IXGBE_PRC1023
);
5304 hwstats
->prc1522
+= IXGBE_READ_REG(hw
, IXGBE_PRC1522
);
5305 hwstats
->rlec
+= IXGBE_READ_REG(hw
, IXGBE_RLEC
);
5306 lxon
= IXGBE_READ_REG(hw
, IXGBE_LXONTXC
);
5307 hwstats
->lxontxc
+= lxon
;
5308 lxoff
= IXGBE_READ_REG(hw
, IXGBE_LXOFFTXC
);
5309 hwstats
->lxofftxc
+= lxoff
;
5310 hwstats
->gptc
+= IXGBE_READ_REG(hw
, IXGBE_GPTC
);
5311 hwstats
->mptc
+= IXGBE_READ_REG(hw
, IXGBE_MPTC
);
5313 * 82598 errata - tx of flow control packets is included in tx counters
5315 xon_off_tot
= lxon
+ lxoff
;
5316 hwstats
->gptc
-= xon_off_tot
;
5317 hwstats
->mptc
-= xon_off_tot
;
5318 hwstats
->gotc
-= (xon_off_tot
* (ETH_ZLEN
+ ETH_FCS_LEN
));
5319 hwstats
->ruc
+= IXGBE_READ_REG(hw
, IXGBE_RUC
);
5320 hwstats
->rfc
+= IXGBE_READ_REG(hw
, IXGBE_RFC
);
5321 hwstats
->rjc
+= IXGBE_READ_REG(hw
, IXGBE_RJC
);
5322 hwstats
->tpr
+= IXGBE_READ_REG(hw
, IXGBE_TPR
);
5323 hwstats
->ptc64
+= IXGBE_READ_REG(hw
, IXGBE_PTC64
);
5324 hwstats
->ptc64
-= xon_off_tot
;
5325 hwstats
->ptc127
+= IXGBE_READ_REG(hw
, IXGBE_PTC127
);
5326 hwstats
->ptc255
+= IXGBE_READ_REG(hw
, IXGBE_PTC255
);
5327 hwstats
->ptc511
+= IXGBE_READ_REG(hw
, IXGBE_PTC511
);
5328 hwstats
->ptc1023
+= IXGBE_READ_REG(hw
, IXGBE_PTC1023
);
5329 hwstats
->ptc1522
+= IXGBE_READ_REG(hw
, IXGBE_PTC1522
);
5330 hwstats
->bptc
+= IXGBE_READ_REG(hw
, IXGBE_BPTC
);
5332 /* Fill out the OS statistics structure */
5333 netdev
->stats
.multicast
= hwstats
->mprc
;
5336 netdev
->stats
.rx_errors
= hwstats
->crcerrs
+ hwstats
->rlec
;
5337 netdev
->stats
.rx_dropped
= 0;
5338 netdev
->stats
.rx_length_errors
= hwstats
->rlec
;
5339 netdev
->stats
.rx_crc_errors
= hwstats
->crcerrs
;
5340 netdev
->stats
.rx_missed_errors
= total_mpc
;
5344 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5345 * @adapter: pointer to the device adapter structure
5347 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter
*adapter
)
5349 struct ixgbe_hw
*hw
= &adapter
->hw
;
5352 if (!(adapter
->flags2
& IXGBE_FLAG2_FDIR_REQUIRES_REINIT
))
5355 adapter
->flags2
&= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT
;
5357 /* if interface is down do nothing */
5358 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
5361 /* do nothing if we are not using signature filters */
5362 if (!(adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
))
5365 adapter
->fdir_overflow
++;
5367 if (ixgbe_reinit_fdir_tables_82599(hw
) == 0) {
5368 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
5369 set_bit(__IXGBE_TX_FDIR_INIT_DONE
,
5370 &(adapter
->tx_ring
[i
]->state
));
5371 /* re-enable flow director interrupts */
5372 IXGBE_WRITE_REG(hw
, IXGBE_EIMS
, IXGBE_EIMS_FLOW_DIR
);
5374 e_err(probe
, "failed to finish FDIR re-initialization, "
5375 "ignored adding FDIR ATR filters\n");
5380 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5381 * @adapter: pointer to the device adapter structure
5383 * This function serves two purposes. First it strobes the interrupt lines
5384 * in order to make certain interrupts are occurring. Secondly it sets the
5385 * bits needed to check for TX hangs. As a result we should immediately
5386 * determine if a hang has occurred.
5388 static void ixgbe_check_hang_subtask(struct ixgbe_adapter
*adapter
)
5390 struct ixgbe_hw
*hw
= &adapter
->hw
;
5394 /* If we're down or resetting, just bail */
5395 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
5396 test_bit(__IXGBE_RESETTING
, &adapter
->state
))
5399 /* Force detection of hung controller */
5400 if (netif_carrier_ok(adapter
->netdev
)) {
5401 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
5402 set_check_for_tx_hang(adapter
->tx_ring
[i
]);
5405 if (!(adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)) {
5407 * for legacy and MSI interrupts don't set any bits
5408 * that are enabled for EIAM, because this operation
5409 * would set *both* EIMS and EICS for any bit in EIAM
5411 IXGBE_WRITE_REG(hw
, IXGBE_EICS
,
5412 (IXGBE_EICS_TCP_TIMER
| IXGBE_EICS_OTHER
));
5414 /* get one bit for every active tx/rx interrupt vector */
5415 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
5416 struct ixgbe_q_vector
*qv
= adapter
->q_vector
[i
];
5417 if (qv
->rx
.ring
|| qv
->tx
.ring
)
5418 eics
|= ((u64
)1 << i
);
5422 /* Cause software interrupt to ensure rings are cleaned */
5423 ixgbe_irq_rearm_queues(adapter
, eics
);
5428 * ixgbe_watchdog_update_link - update the link status
5429 * @adapter: pointer to the device adapter structure
5430 * @link_speed: pointer to a u32 to store the link_speed
5432 static void ixgbe_watchdog_update_link(struct ixgbe_adapter
*adapter
)
5434 struct ixgbe_hw
*hw
= &adapter
->hw
;
5435 u32 link_speed
= adapter
->link_speed
;
5436 bool link_up
= adapter
->link_up
;
5437 bool pfc_en
= adapter
->dcb_cfg
.pfc_mode_enable
;
5439 if (!(adapter
->flags
& IXGBE_FLAG_NEED_LINK_UPDATE
))
5442 if (hw
->mac
.ops
.check_link
) {
5443 hw
->mac
.ops
.check_link(hw
, &link_speed
, &link_up
, false);
5445 /* always assume link is up, if no check link function */
5446 link_speed
= IXGBE_LINK_SPEED_10GB_FULL
;
5450 if (adapter
->ixgbe_ieee_pfc
)
5451 pfc_en
|= !!(adapter
->ixgbe_ieee_pfc
->pfc_en
);
5453 if (link_up
&& !((adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) && pfc_en
)) {
5454 hw
->mac
.ops
.fc_enable(hw
);
5455 ixgbe_set_rx_drop_en(adapter
);
5459 time_after(jiffies
, (adapter
->link_check_timeout
+
5460 IXGBE_TRY_LINK_TIMEOUT
))) {
5461 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_UPDATE
;
5462 IXGBE_WRITE_REG(hw
, IXGBE_EIMS
, IXGBE_EIMC_LSC
);
5463 IXGBE_WRITE_FLUSH(hw
);
5466 adapter
->link_up
= link_up
;
5467 adapter
->link_speed
= link_speed
;
5470 static void ixgbe_update_default_up(struct ixgbe_adapter
*adapter
)
5472 #ifdef CONFIG_IXGBE_DCB
5473 struct net_device
*netdev
= adapter
->netdev
;
5474 struct dcb_app app
= {
5475 .selector
= IEEE_8021QAZ_APP_SEL_ETHERTYPE
,
5480 if (adapter
->dcbx_cap
& DCB_CAP_DCBX_VER_IEEE
)
5481 up
= dcb_ieee_getapp_mask(netdev
, &app
);
5483 adapter
->default_up
= (up
> 1) ? (ffs(up
) - 1) : 0;
5488 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5489 * print link up message
5490 * @adapter: pointer to the device adapter structure
5492 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter
*adapter
)
5494 struct net_device
*netdev
= adapter
->netdev
;
5495 struct ixgbe_hw
*hw
= &adapter
->hw
;
5496 u32 link_speed
= adapter
->link_speed
;
5497 bool flow_rx
, flow_tx
;
5499 /* only continue if link was previously down */
5500 if (netif_carrier_ok(netdev
))
5503 adapter
->flags2
&= ~IXGBE_FLAG2_SEARCH_FOR_SFP
;
5505 switch (hw
->mac
.type
) {
5506 case ixgbe_mac_82598EB
: {
5507 u32 frctl
= IXGBE_READ_REG(hw
, IXGBE_FCTRL
);
5508 u32 rmcs
= IXGBE_READ_REG(hw
, IXGBE_RMCS
);
5509 flow_rx
= !!(frctl
& IXGBE_FCTRL_RFCE
);
5510 flow_tx
= !!(rmcs
& IXGBE_RMCS_TFCE_802_3X
);
5513 case ixgbe_mac_X540
:
5514 case ixgbe_mac_82599EB
: {
5515 u32 mflcn
= IXGBE_READ_REG(hw
, IXGBE_MFLCN
);
5516 u32 fccfg
= IXGBE_READ_REG(hw
, IXGBE_FCCFG
);
5517 flow_rx
= !!(mflcn
& IXGBE_MFLCN_RFCE
);
5518 flow_tx
= !!(fccfg
& IXGBE_FCCFG_TFCE_802_3X
);
5527 if (adapter
->flags2
& IXGBE_FLAG2_PTP_ENABLED
)
5528 ixgbe_ptp_start_cyclecounter(adapter
);
5530 e_info(drv
, "NIC Link is Up %s, Flow Control: %s\n",
5531 (link_speed
== IXGBE_LINK_SPEED_10GB_FULL
?
5533 (link_speed
== IXGBE_LINK_SPEED_1GB_FULL
?
5535 (link_speed
== IXGBE_LINK_SPEED_100_FULL
?
5538 ((flow_rx
&& flow_tx
) ? "RX/TX" :
5540 (flow_tx
? "TX" : "None"))));
5542 netif_carrier_on(netdev
);
5543 ixgbe_check_vf_rate_limit(adapter
);
5545 /* update the default user priority for VFs */
5546 ixgbe_update_default_up(adapter
);
5548 /* ping all the active vfs to let them know link has changed */
5549 ixgbe_ping_all_vfs(adapter
);
5553 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5554 * print link down message
5555 * @adapter: pointer to the adapter structure
5557 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter
*adapter
)
5559 struct net_device
*netdev
= adapter
->netdev
;
5560 struct ixgbe_hw
*hw
= &adapter
->hw
;
5562 adapter
->link_up
= false;
5563 adapter
->link_speed
= 0;
5565 /* only continue if link was up previously */
5566 if (!netif_carrier_ok(netdev
))
5569 /* poll for SFP+ cable when link is down */
5570 if (ixgbe_is_sfp(hw
) && hw
->mac
.type
== ixgbe_mac_82598EB
)
5571 adapter
->flags2
|= IXGBE_FLAG2_SEARCH_FOR_SFP
;
5573 if (adapter
->flags2
& IXGBE_FLAG2_PTP_ENABLED
)
5574 ixgbe_ptp_start_cyclecounter(adapter
);
5576 e_info(drv
, "NIC Link is Down\n");
5577 netif_carrier_off(netdev
);
5579 /* ping all the active vfs to let them know link has changed */
5580 ixgbe_ping_all_vfs(adapter
);
5584 * ixgbe_watchdog_flush_tx - flush queues on link down
5585 * @adapter: pointer to the device adapter structure
5587 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter
*adapter
)
5590 int some_tx_pending
= 0;
5592 if (!netif_carrier_ok(adapter
->netdev
)) {
5593 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
5594 struct ixgbe_ring
*tx_ring
= adapter
->tx_ring
[i
];
5595 if (tx_ring
->next_to_use
!= tx_ring
->next_to_clean
) {
5596 some_tx_pending
= 1;
5601 if (some_tx_pending
) {
5602 /* We've lost link, so the controller stops DMA,
5603 * but we've got queued Tx work that's never going
5604 * to get done, so reset controller to flush Tx.
5605 * (Do the reset outside of interrupt context).
5607 adapter
->flags2
|= IXGBE_FLAG2_RESET_REQUESTED
;
5612 static void ixgbe_spoof_check(struct ixgbe_adapter
*adapter
)
5616 /* Do not perform spoof check for 82598 or if not in IOV mode */
5617 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
||
5618 adapter
->num_vfs
== 0)
5621 ssvpc
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_SSVPC
);
5624 * ssvpc register is cleared on read, if zero then no
5625 * spoofed packets in the last interval.
5630 e_warn(drv
, "%u Spoofed packets detected\n", ssvpc
);
5634 * ixgbe_watchdog_subtask - check and bring link up
5635 * @adapter: pointer to the device adapter structure
5637 static void ixgbe_watchdog_subtask(struct ixgbe_adapter
*adapter
)
5639 /* if interface is down do nothing */
5640 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
5641 test_bit(__IXGBE_RESETTING
, &adapter
->state
))
5644 ixgbe_watchdog_update_link(adapter
);
5646 if (adapter
->link_up
)
5647 ixgbe_watchdog_link_is_up(adapter
);
5649 ixgbe_watchdog_link_is_down(adapter
);
5651 ixgbe_spoof_check(adapter
);
5652 ixgbe_update_stats(adapter
);
5654 ixgbe_watchdog_flush_tx(adapter
);
5658 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5659 * @adapter: the ixgbe adapter structure
5661 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter
*adapter
)
5663 struct ixgbe_hw
*hw
= &adapter
->hw
;
5666 /* not searching for SFP so there is nothing to do here */
5667 if (!(adapter
->flags2
& IXGBE_FLAG2_SEARCH_FOR_SFP
) &&
5668 !(adapter
->flags2
& IXGBE_FLAG2_SFP_NEEDS_RESET
))
5671 /* someone else is in init, wait until next service event */
5672 if (test_and_set_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
))
5675 err
= hw
->phy
.ops
.identify_sfp(hw
);
5676 if (err
== IXGBE_ERR_SFP_NOT_SUPPORTED
)
5679 if (err
== IXGBE_ERR_SFP_NOT_PRESENT
) {
5680 /* If no cable is present, then we need to reset
5681 * the next time we find a good cable. */
5682 adapter
->flags2
|= IXGBE_FLAG2_SFP_NEEDS_RESET
;
5689 /* exit if reset not needed */
5690 if (!(adapter
->flags2
& IXGBE_FLAG2_SFP_NEEDS_RESET
))
5693 adapter
->flags2
&= ~IXGBE_FLAG2_SFP_NEEDS_RESET
;
5696 * A module may be identified correctly, but the EEPROM may not have
5697 * support for that module. setup_sfp() will fail in that case, so
5698 * we should not allow that module to load.
5700 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
5701 err
= hw
->phy
.ops
.reset(hw
);
5703 err
= hw
->mac
.ops
.setup_sfp(hw
);
5705 if (err
== IXGBE_ERR_SFP_NOT_SUPPORTED
)
5708 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_CONFIG
;
5709 e_info(probe
, "detected SFP+: %d\n", hw
->phy
.sfp_type
);
5712 clear_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
);
5714 if ((err
== IXGBE_ERR_SFP_NOT_SUPPORTED
) &&
5715 (adapter
->netdev
->reg_state
== NETREG_REGISTERED
)) {
5716 e_dev_err("failed to initialize because an unsupported "
5717 "SFP+ module type was detected.\n");
5718 e_dev_err("Reload the driver after installing a "
5719 "supported module.\n");
5720 unregister_netdev(adapter
->netdev
);
5725 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5726 * @adapter: the ixgbe adapter structure
5728 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter
*adapter
)
5730 struct ixgbe_hw
*hw
= &adapter
->hw
;
5734 if (!(adapter
->flags
& IXGBE_FLAG_NEED_LINK_CONFIG
))
5737 /* someone else is in init, wait until next service event */
5738 if (test_and_set_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
))
5741 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_CONFIG
;
5743 autoneg
= hw
->phy
.autoneg_advertised
;
5744 if ((!autoneg
) && (hw
->mac
.ops
.get_link_capabilities
))
5745 hw
->mac
.ops
.get_link_capabilities(hw
, &autoneg
, &negotiation
);
5746 if (hw
->mac
.ops
.setup_link
)
5747 hw
->mac
.ops
.setup_link(hw
, autoneg
, negotiation
, true);
5749 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_UPDATE
;
5750 adapter
->link_check_timeout
= jiffies
;
5751 clear_bit(__IXGBE_IN_SFP_INIT
, &adapter
->state
);
5754 #ifdef CONFIG_PCI_IOV
5755 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter
*adapter
)
5758 struct ixgbe_hw
*hw
= &adapter
->hw
;
5759 struct net_device
*netdev
= adapter
->netdev
;
5763 gpc
= IXGBE_READ_REG(hw
, IXGBE_TXDGPC
);
5764 if (gpc
) /* If incrementing then no need for the check below */
5767 * Check to see if a bad DMA write target from an errant or
5768 * malicious VF has caused a PCIe error. If so then we can
5769 * issue a VFLR to the offending VF(s) and then resume without
5770 * requesting a full slot reset.
5773 for (vf
= 0; vf
< adapter
->num_vfs
; vf
++) {
5774 ciaa
= (vf
<< 16) | 0x80000000;
5775 /* 32 bit read so align, we really want status at offset 6 */
5776 ciaa
|= PCI_COMMAND
;
5777 IXGBE_WRITE_REG(hw
, IXGBE_CIAA_82599
, ciaa
);
5778 ciad
= IXGBE_READ_REG(hw
, IXGBE_CIAD_82599
);
5780 /* disable debug mode asap after reading data */
5781 IXGBE_WRITE_REG(hw
, IXGBE_CIAA_82599
, ciaa
);
5782 /* Get the upper 16 bits which will be the PCI status reg */
5784 if (ciad
& PCI_STATUS_REC_MASTER_ABORT
) {
5785 netdev_err(netdev
, "VF %d Hung DMA\n", vf
);
5787 ciaa
= (vf
<< 16) | 0x80000000;
5789 IXGBE_WRITE_REG(hw
, IXGBE_CIAA_82599
, ciaa
);
5790 ciad
= 0x00008000; /* VFLR */
5791 IXGBE_WRITE_REG(hw
, IXGBE_CIAD_82599
, ciad
);
5793 IXGBE_WRITE_REG(hw
, IXGBE_CIAA_82599
, ciaa
);
5800 * ixgbe_service_timer - Timer Call-back
5801 * @data: pointer to adapter cast into an unsigned long
5803 static void ixgbe_service_timer(unsigned long data
)
5805 struct ixgbe_adapter
*adapter
= (struct ixgbe_adapter
*)data
;
5806 unsigned long next_event_offset
;
5809 /* poll faster when waiting for link */
5810 if (adapter
->flags
& IXGBE_FLAG_NEED_LINK_UPDATE
)
5811 next_event_offset
= HZ
/ 10;
5813 next_event_offset
= HZ
* 2;
5815 #ifdef CONFIG_PCI_IOV
5817 * don't bother with SR-IOV VF DMA hang check if there are
5818 * no VFs or the link is down
5820 if (!adapter
->num_vfs
||
5821 (adapter
->flags
& IXGBE_FLAG_NEED_LINK_UPDATE
))
5822 goto normal_timer_service
;
5824 /* If we have VFs allocated then we must check for DMA hangs */
5825 ixgbe_check_for_bad_vf(adapter
);
5826 next_event_offset
= HZ
/ 50;
5827 adapter
->timer_event_accumulator
++;
5829 if (adapter
->timer_event_accumulator
>= 100)
5830 adapter
->timer_event_accumulator
= 0;
5834 normal_timer_service
:
5836 /* Reset the timer */
5837 mod_timer(&adapter
->service_timer
, next_event_offset
+ jiffies
);
5840 ixgbe_service_event_schedule(adapter
);
5843 static void ixgbe_reset_subtask(struct ixgbe_adapter
*adapter
)
5845 if (!(adapter
->flags2
& IXGBE_FLAG2_RESET_REQUESTED
))
5848 adapter
->flags2
&= ~IXGBE_FLAG2_RESET_REQUESTED
;
5850 /* If we're already down or resetting, just bail */
5851 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
5852 test_bit(__IXGBE_RESETTING
, &adapter
->state
))
5855 ixgbe_dump(adapter
);
5856 netdev_err(adapter
->netdev
, "Reset adapter\n");
5857 adapter
->tx_timeout_count
++;
5859 ixgbe_reinit_locked(adapter
);
5863 * ixgbe_service_task - manages and runs subtasks
5864 * @work: pointer to work_struct containing our data
5866 static void ixgbe_service_task(struct work_struct
*work
)
5868 struct ixgbe_adapter
*adapter
= container_of(work
,
5869 struct ixgbe_adapter
,
5872 ixgbe_reset_subtask(adapter
);
5873 ixgbe_sfp_detection_subtask(adapter
);
5874 ixgbe_sfp_link_config_subtask(adapter
);
5875 ixgbe_check_overtemp_subtask(adapter
);
5876 ixgbe_watchdog_subtask(adapter
);
5877 ixgbe_fdir_reinit_subtask(adapter
);
5878 ixgbe_check_hang_subtask(adapter
);
5879 ixgbe_ptp_overflow_check(adapter
);
5881 ixgbe_service_event_complete(adapter
);
5884 static int ixgbe_tso(struct ixgbe_ring
*tx_ring
,
5885 struct ixgbe_tx_buffer
*first
,
5888 struct sk_buff
*skb
= first
->skb
;
5889 u32 vlan_macip_lens
, type_tucmd
;
5890 u32 mss_l4len_idx
, l4len
;
5892 if (!skb_is_gso(skb
))
5895 if (skb_header_cloned(skb
)) {
5896 int err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
5901 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5902 type_tucmd
= IXGBE_ADVTXD_TUCMD_L4T_TCP
;
5904 if (first
->protocol
== __constant_htons(ETH_P_IP
)) {
5905 struct iphdr
*iph
= ip_hdr(skb
);
5908 tcp_hdr(skb
)->check
= ~csum_tcpudp_magic(iph
->saddr
,
5912 type_tucmd
|= IXGBE_ADVTXD_TUCMD_IPV4
;
5913 first
->tx_flags
|= IXGBE_TX_FLAGS_TSO
|
5914 IXGBE_TX_FLAGS_CSUM
|
5915 IXGBE_TX_FLAGS_IPV4
;
5916 } else if (skb_is_gso_v6(skb
)) {
5917 ipv6_hdr(skb
)->payload_len
= 0;
5918 tcp_hdr(skb
)->check
=
5919 ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
5920 &ipv6_hdr(skb
)->daddr
,
5922 first
->tx_flags
|= IXGBE_TX_FLAGS_TSO
|
5923 IXGBE_TX_FLAGS_CSUM
;
5926 /* compute header lengths */
5927 l4len
= tcp_hdrlen(skb
);
5928 *hdr_len
= skb_transport_offset(skb
) + l4len
;
5930 /* update gso size and bytecount with header size */
5931 first
->gso_segs
= skb_shinfo(skb
)->gso_segs
;
5932 first
->bytecount
+= (first
->gso_segs
- 1) * *hdr_len
;
5934 /* mss_l4len_id: use 1 as index for TSO */
5935 mss_l4len_idx
= l4len
<< IXGBE_ADVTXD_L4LEN_SHIFT
;
5936 mss_l4len_idx
|= skb_shinfo(skb
)->gso_size
<< IXGBE_ADVTXD_MSS_SHIFT
;
5937 mss_l4len_idx
|= 1 << IXGBE_ADVTXD_IDX_SHIFT
;
5939 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
5940 vlan_macip_lens
= skb_network_header_len(skb
);
5941 vlan_macip_lens
|= skb_network_offset(skb
) << IXGBE_ADVTXD_MACLEN_SHIFT
;
5942 vlan_macip_lens
|= first
->tx_flags
& IXGBE_TX_FLAGS_VLAN_MASK
;
5944 ixgbe_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, 0, type_tucmd
,
5950 static void ixgbe_tx_csum(struct ixgbe_ring
*tx_ring
,
5951 struct ixgbe_tx_buffer
*first
)
5953 struct sk_buff
*skb
= first
->skb
;
5954 u32 vlan_macip_lens
= 0;
5955 u32 mss_l4len_idx
= 0;
5958 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
) {
5959 if (!(first
->tx_flags
& IXGBE_TX_FLAGS_HW_VLAN
)) {
5960 if (unlikely(skb
->no_fcs
))
5961 first
->tx_flags
|= IXGBE_TX_FLAGS_NO_IFCS
;
5962 if (!(first
->tx_flags
& IXGBE_TX_FLAGS_TXSW
))
5967 switch (first
->protocol
) {
5968 case __constant_htons(ETH_P_IP
):
5969 vlan_macip_lens
|= skb_network_header_len(skb
);
5970 type_tucmd
|= IXGBE_ADVTXD_TUCMD_IPV4
;
5971 l4_hdr
= ip_hdr(skb
)->protocol
;
5973 case __constant_htons(ETH_P_IPV6
):
5974 vlan_macip_lens
|= skb_network_header_len(skb
);
5975 l4_hdr
= ipv6_hdr(skb
)->nexthdr
;
5978 if (unlikely(net_ratelimit())) {
5979 dev_warn(tx_ring
->dev
,
5980 "partial checksum but proto=%x!\n",
5988 type_tucmd
|= IXGBE_ADVTXD_TUCMD_L4T_TCP
;
5989 mss_l4len_idx
= tcp_hdrlen(skb
) <<
5990 IXGBE_ADVTXD_L4LEN_SHIFT
;
5993 type_tucmd
|= IXGBE_ADVTXD_TUCMD_L4T_SCTP
;
5994 mss_l4len_idx
= sizeof(struct sctphdr
) <<
5995 IXGBE_ADVTXD_L4LEN_SHIFT
;
5998 mss_l4len_idx
= sizeof(struct udphdr
) <<
5999 IXGBE_ADVTXD_L4LEN_SHIFT
;
6002 if (unlikely(net_ratelimit())) {
6003 dev_warn(tx_ring
->dev
,
6004 "partial checksum but l4 proto=%x!\n",
6010 /* update TX checksum flag */
6011 first
->tx_flags
|= IXGBE_TX_FLAGS_CSUM
;
6014 /* vlan_macip_lens: MACLEN, VLAN tag */
6015 vlan_macip_lens
|= skb_network_offset(skb
) << IXGBE_ADVTXD_MACLEN_SHIFT
;
6016 vlan_macip_lens
|= first
->tx_flags
& IXGBE_TX_FLAGS_VLAN_MASK
;
6018 ixgbe_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, 0,
6019 type_tucmd
, mss_l4len_idx
);
6022 static __le32
ixgbe_tx_cmd_type(u32 tx_flags
)
6024 /* set type for advanced descriptor with frame checksum insertion */
6025 __le32 cmd_type
= cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA
|
6026 IXGBE_ADVTXD_DCMD_DEXT
);
6028 /* set HW vlan bit if vlan is present */
6029 if (tx_flags
& IXGBE_TX_FLAGS_HW_VLAN
)
6030 cmd_type
|= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE
);
6032 if (tx_flags
& IXGBE_TX_FLAGS_TSTAMP
)
6033 cmd_type
|= cpu_to_le32(IXGBE_ADVTXD_MAC_TSTAMP
);
6035 /* set segmentation enable bits for TSO/FSO */
6037 if (tx_flags
& (IXGBE_TX_FLAGS_TSO
| IXGBE_TX_FLAGS_FSO
))
6039 if (tx_flags
& IXGBE_TX_FLAGS_TSO
)
6041 cmd_type
|= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE
);
6043 /* insert frame checksum */
6044 if (!(tx_flags
& IXGBE_TX_FLAGS_NO_IFCS
))
6045 cmd_type
|= cpu_to_le32(IXGBE_ADVTXD_DCMD_IFCS
);
6050 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc
*tx_desc
,
6051 u32 tx_flags
, unsigned int paylen
)
6053 __le32 olinfo_status
= cpu_to_le32(paylen
<< IXGBE_ADVTXD_PAYLEN_SHIFT
);
6055 /* enable L4 checksum for TSO and TX checksum offload */
6056 if (tx_flags
& IXGBE_TX_FLAGS_CSUM
)
6057 olinfo_status
|= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM
);
6059 /* enble IPv4 checksum for TSO */
6060 if (tx_flags
& IXGBE_TX_FLAGS_IPV4
)
6061 olinfo_status
|= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM
);
6063 /* use index 1 context for TSO/FSO/FCOE */
6065 if (tx_flags
& (IXGBE_TX_FLAGS_TSO
| IXGBE_TX_FLAGS_FCOE
))
6067 if (tx_flags
& IXGBE_TX_FLAGS_TSO
)
6069 olinfo_status
|= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT
);
6072 * Check Context must be set if Tx switch is enabled, which it
6073 * always is for case where virtual functions are running
6076 if (tx_flags
& (IXGBE_TX_FLAGS_TXSW
| IXGBE_TX_FLAGS_FCOE
))
6078 if (tx_flags
& IXGBE_TX_FLAGS_TXSW
)
6080 olinfo_status
|= cpu_to_le32(IXGBE_ADVTXD_CC
);
6082 tx_desc
->read
.olinfo_status
= olinfo_status
;
6085 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
6088 static void ixgbe_tx_map(struct ixgbe_ring
*tx_ring
,
6089 struct ixgbe_tx_buffer
*first
,
6093 struct sk_buff
*skb
= first
->skb
;
6094 struct ixgbe_tx_buffer
*tx_buffer
;
6095 union ixgbe_adv_tx_desc
*tx_desc
;
6096 struct skb_frag_struct
*frag
= &skb_shinfo(skb
)->frags
[0];
6097 unsigned int data_len
= skb
->data_len
;
6098 unsigned int size
= skb_headlen(skb
);
6099 unsigned int paylen
= skb
->len
- hdr_len
;
6100 u32 tx_flags
= first
->tx_flags
;
6102 u16 i
= tx_ring
->next_to_use
;
6104 tx_desc
= IXGBE_TX_DESC(tx_ring
, i
);
6106 ixgbe_tx_olinfo_status(tx_desc
, tx_flags
, paylen
);
6107 cmd_type
= ixgbe_tx_cmd_type(tx_flags
);
6110 if (tx_flags
& IXGBE_TX_FLAGS_FCOE
) {
6111 if (data_len
< sizeof(struct fcoe_crc_eof
)) {
6112 size
-= sizeof(struct fcoe_crc_eof
) - data_len
;
6115 data_len
-= sizeof(struct fcoe_crc_eof
);
6120 dma
= dma_map_single(tx_ring
->dev
, skb
->data
, size
, DMA_TO_DEVICE
);
6121 if (dma_mapping_error(tx_ring
->dev
, dma
))
6124 /* record length, and DMA address */
6125 dma_unmap_len_set(first
, len
, size
);
6126 dma_unmap_addr_set(first
, dma
, dma
);
6128 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
6131 while (unlikely(size
> IXGBE_MAX_DATA_PER_TXD
)) {
6132 tx_desc
->read
.cmd_type_len
=
6133 cmd_type
| cpu_to_le32(IXGBE_MAX_DATA_PER_TXD
);
6137 if (i
== tx_ring
->count
) {
6138 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
6142 dma
+= IXGBE_MAX_DATA_PER_TXD
;
6143 size
-= IXGBE_MAX_DATA_PER_TXD
;
6145 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
6146 tx_desc
->read
.olinfo_status
= 0;
6149 if (likely(!data_len
))
6152 tx_desc
->read
.cmd_type_len
= cmd_type
| cpu_to_le32(size
);
6156 if (i
== tx_ring
->count
) {
6157 tx_desc
= IXGBE_TX_DESC(tx_ring
, 0);
6162 size
= min_t(unsigned int, data_len
, skb_frag_size(frag
));
6164 size
= skb_frag_size(frag
);
6168 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0, size
,
6170 if (dma_mapping_error(tx_ring
->dev
, dma
))
6173 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
6174 dma_unmap_len_set(tx_buffer
, len
, size
);
6175 dma_unmap_addr_set(tx_buffer
, dma
, dma
);
6177 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
6178 tx_desc
->read
.olinfo_status
= 0;
6183 /* write last descriptor with RS and EOP bits */
6184 cmd_type
|= cpu_to_le32(size
) | cpu_to_le32(IXGBE_TXD_CMD
);
6185 tx_desc
->read
.cmd_type_len
= cmd_type
;
6187 netdev_tx_sent_queue(txring_txq(tx_ring
), first
->bytecount
);
6189 /* set the timestamp */
6190 first
->time_stamp
= jiffies
;
6193 * Force memory writes to complete before letting h/w know there
6194 * are new descriptors to fetch. (Only applicable for weak-ordered
6195 * memory model archs, such as IA-64).
6197 * We also need this memory barrier to make certain all of the
6198 * status bits have been updated before next_to_watch is written.
6202 /* set next_to_watch value indicating a packet is present */
6203 first
->next_to_watch
= tx_desc
;
6206 if (i
== tx_ring
->count
)
6209 tx_ring
->next_to_use
= i
;
6211 /* notify HW of packet */
6212 writel(i
, tx_ring
->tail
);
6216 dev_err(tx_ring
->dev
, "TX DMA map failed\n");
6218 /* clear dma mappings for failed tx_buffer_info map */
6220 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
6221 ixgbe_unmap_and_free_tx_resource(tx_ring
, tx_buffer
);
6222 if (tx_buffer
== first
)
6229 tx_ring
->next_to_use
= i
;
6232 static void ixgbe_atr(struct ixgbe_ring
*ring
,
6233 struct ixgbe_tx_buffer
*first
)
6235 struct ixgbe_q_vector
*q_vector
= ring
->q_vector
;
6236 union ixgbe_atr_hash_dword input
= { .dword
= 0 };
6237 union ixgbe_atr_hash_dword common
= { .dword
= 0 };
6239 unsigned char *network
;
6241 struct ipv6hdr
*ipv6
;
6246 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6250 /* do nothing if sampling is disabled */
6251 if (!ring
->atr_sample_rate
)
6256 /* snag network header to get L4 type and address */
6257 hdr
.network
= skb_network_header(first
->skb
);
6259 /* Currently only IPv4/IPv6 with TCP is supported */
6260 if ((first
->protocol
!= __constant_htons(ETH_P_IPV6
) ||
6261 hdr
.ipv6
->nexthdr
!= IPPROTO_TCP
) &&
6262 (first
->protocol
!= __constant_htons(ETH_P_IP
) ||
6263 hdr
.ipv4
->protocol
!= IPPROTO_TCP
))
6266 th
= tcp_hdr(first
->skb
);
6268 /* skip this packet since it is invalid or the socket is closing */
6272 /* sample on all syn packets or once every atr sample count */
6273 if (!th
->syn
&& (ring
->atr_count
< ring
->atr_sample_rate
))
6276 /* reset sample count */
6277 ring
->atr_count
= 0;
6279 vlan_id
= htons(first
->tx_flags
>> IXGBE_TX_FLAGS_VLAN_SHIFT
);
6282 * src and dst are inverted, think how the receiver sees them
6284 * The input is broken into two sections, a non-compressed section
6285 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6286 * is XORed together and stored in the compressed dword.
6288 input
.formatted
.vlan_id
= vlan_id
;
6291 * since src port and flex bytes occupy the same word XOR them together
6292 * and write the value to source port portion of compressed dword
6294 if (first
->tx_flags
& (IXGBE_TX_FLAGS_SW_VLAN
| IXGBE_TX_FLAGS_HW_VLAN
))
6295 common
.port
.src
^= th
->dest
^ __constant_htons(ETH_P_8021Q
);
6297 common
.port
.src
^= th
->dest
^ first
->protocol
;
6298 common
.port
.dst
^= th
->source
;
6300 if (first
->protocol
== __constant_htons(ETH_P_IP
)) {
6301 input
.formatted
.flow_type
= IXGBE_ATR_FLOW_TYPE_TCPV4
;
6302 common
.ip
^= hdr
.ipv4
->saddr
^ hdr
.ipv4
->daddr
;
6304 input
.formatted
.flow_type
= IXGBE_ATR_FLOW_TYPE_TCPV6
;
6305 common
.ip
^= hdr
.ipv6
->saddr
.s6_addr32
[0] ^
6306 hdr
.ipv6
->saddr
.s6_addr32
[1] ^
6307 hdr
.ipv6
->saddr
.s6_addr32
[2] ^
6308 hdr
.ipv6
->saddr
.s6_addr32
[3] ^
6309 hdr
.ipv6
->daddr
.s6_addr32
[0] ^
6310 hdr
.ipv6
->daddr
.s6_addr32
[1] ^
6311 hdr
.ipv6
->daddr
.s6_addr32
[2] ^
6312 hdr
.ipv6
->daddr
.s6_addr32
[3];
6315 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6316 ixgbe_fdir_add_signature_filter_82599(&q_vector
->adapter
->hw
,
6317 input
, common
, ring
->queue_index
);
6320 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring
*tx_ring
, u16 size
)
6322 netif_stop_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
6323 /* Herbert's original patch had:
6324 * smp_mb__after_netif_stop_queue();
6325 * but since that doesn't exist yet, just open code it. */
6328 /* We need to check again in a case another CPU has just
6329 * made room available. */
6330 if (likely(ixgbe_desc_unused(tx_ring
) < size
))
6333 /* A reprieve! - use start_queue because it doesn't call schedule */
6334 netif_start_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
6335 ++tx_ring
->tx_stats
.restart_queue
;
6339 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring
*tx_ring
, u16 size
)
6341 if (likely(ixgbe_desc_unused(tx_ring
) >= size
))
6343 return __ixgbe_maybe_stop_tx(tx_ring
, size
);
6346 static u16
ixgbe_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
6348 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
6349 int txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) :
6352 __be16 protocol
= vlan_get_protocol(skb
);
6354 if (((protocol
== htons(ETH_P_FCOE
)) ||
6355 (protocol
== htons(ETH_P_FIP
))) &&
6356 (adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
)) {
6357 struct ixgbe_ring_feature
*f
;
6359 f
= &adapter
->ring_feature
[RING_F_FCOE
];
6361 while (txq
>= f
->indices
)
6363 txq
+= adapter
->ring_feature
[RING_F_FCOE
].offset
;
6369 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) {
6370 while (unlikely(txq
>= dev
->real_num_tx_queues
))
6371 txq
-= dev
->real_num_tx_queues
;
6375 return skb_tx_hash(dev
, skb
);
6378 netdev_tx_t
ixgbe_xmit_frame_ring(struct sk_buff
*skb
,
6379 struct ixgbe_adapter
*adapter
,
6380 struct ixgbe_ring
*tx_ring
)
6382 struct ixgbe_tx_buffer
*first
;
6385 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6388 u16 count
= TXD_USE_COUNT(skb_headlen(skb
));
6389 __be16 protocol
= skb
->protocol
;
6393 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6394 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6395 * + 2 desc gap to keep tail from touching head,
6396 * + 1 desc for context descriptor,
6397 * otherwise try next time
6399 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6400 for (f
= 0; f
< skb_shinfo(skb
)->nr_frags
; f
++)
6401 count
+= TXD_USE_COUNT(skb_shinfo(skb
)->frags
[f
].size
);
6403 count
+= skb_shinfo(skb
)->nr_frags
;
6405 if (ixgbe_maybe_stop_tx(tx_ring
, count
+ 3)) {
6406 tx_ring
->tx_stats
.tx_busy
++;
6407 return NETDEV_TX_BUSY
;
6410 /* record the location of the first descriptor for this packet */
6411 first
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_use
];
6413 first
->bytecount
= skb
->len
;
6414 first
->gso_segs
= 1;
6416 /* if we have a HW VLAN tag being added default to the HW one */
6417 if (vlan_tx_tag_present(skb
)) {
6418 tx_flags
|= vlan_tx_tag_get(skb
) << IXGBE_TX_FLAGS_VLAN_SHIFT
;
6419 tx_flags
|= IXGBE_TX_FLAGS_HW_VLAN
;
6420 /* else if it is a SW VLAN check the next protocol and store the tag */
6421 } else if (protocol
== __constant_htons(ETH_P_8021Q
)) {
6422 struct vlan_hdr
*vhdr
, _vhdr
;
6423 vhdr
= skb_header_pointer(skb
, ETH_HLEN
, sizeof(_vhdr
), &_vhdr
);
6427 protocol
= vhdr
->h_vlan_encapsulated_proto
;
6428 tx_flags
|= ntohs(vhdr
->h_vlan_TCI
) <<
6429 IXGBE_TX_FLAGS_VLAN_SHIFT
;
6430 tx_flags
|= IXGBE_TX_FLAGS_SW_VLAN
;
6433 skb_tx_timestamp(skb
);
6435 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_HW_TSTAMP
)) {
6436 skb_shinfo(skb
)->tx_flags
|= SKBTX_IN_PROGRESS
;
6437 tx_flags
|= IXGBE_TX_FLAGS_TSTAMP
;
6440 #ifdef CONFIG_PCI_IOV
6442 * Use the l2switch_enable flag - would be false if the DMA
6443 * Tx switch had been disabled.
6445 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
6446 tx_flags
|= IXGBE_TX_FLAGS_TXSW
;
6449 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6450 if ((adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) &&
6451 ((tx_flags
& (IXGBE_TX_FLAGS_HW_VLAN
| IXGBE_TX_FLAGS_SW_VLAN
)) ||
6452 (skb
->priority
!= TC_PRIO_CONTROL
))) {
6453 tx_flags
&= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK
;
6454 tx_flags
|= (skb
->priority
& 0x7) <<
6455 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT
;
6456 if (tx_flags
& IXGBE_TX_FLAGS_SW_VLAN
) {
6457 struct vlan_ethhdr
*vhdr
;
6458 if (skb_header_cloned(skb
) &&
6459 pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))
6461 vhdr
= (struct vlan_ethhdr
*)skb
->data
;
6462 vhdr
->h_vlan_TCI
= htons(tx_flags
>>
6463 IXGBE_TX_FLAGS_VLAN_SHIFT
);
6465 tx_flags
|= IXGBE_TX_FLAGS_HW_VLAN
;
6469 /* record initial flags and protocol */
6470 first
->tx_flags
= tx_flags
;
6471 first
->protocol
= protocol
;
6474 /* setup tx offload for FCoE */
6475 if ((protocol
== __constant_htons(ETH_P_FCOE
)) &&
6476 (tx_ring
->netdev
->features
& (NETIF_F_FSO
| NETIF_F_FCOE_CRC
))) {
6477 tso
= ixgbe_fso(tx_ring
, first
, &hdr_len
);
6484 #endif /* IXGBE_FCOE */
6485 tso
= ixgbe_tso(tx_ring
, first
, &hdr_len
);
6489 ixgbe_tx_csum(tx_ring
, first
);
6491 /* add the ATR filter if ATR is on */
6492 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE
, &tx_ring
->state
))
6493 ixgbe_atr(tx_ring
, first
);
6497 #endif /* IXGBE_FCOE */
6498 ixgbe_tx_map(tx_ring
, first
, hdr_len
);
6500 ixgbe_maybe_stop_tx(tx_ring
, DESC_NEEDED
);
6502 return NETDEV_TX_OK
;
6505 dev_kfree_skb_any(first
->skb
);
6508 return NETDEV_TX_OK
;
6511 static netdev_tx_t
ixgbe_xmit_frame(struct sk_buff
*skb
,
6512 struct net_device
*netdev
)
6514 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6515 struct ixgbe_ring
*tx_ring
;
6518 * The minimum packet size for olinfo paylen is 17 so pad the skb
6519 * in order to meet this minimum size requirement.
6521 if (unlikely(skb
->len
< 17)) {
6522 if (skb_pad(skb
, 17 - skb
->len
))
6523 return NETDEV_TX_OK
;
6525 skb_set_tail_pointer(skb
, 17);
6528 tx_ring
= adapter
->tx_ring
[skb
->queue_mapping
];
6529 return ixgbe_xmit_frame_ring(skb
, adapter
, tx_ring
);
6533 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6534 * @netdev: network interface device structure
6535 * @p: pointer to an address structure
6537 * Returns 0 on success, negative on failure
6539 static int ixgbe_set_mac(struct net_device
*netdev
, void *p
)
6541 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6542 struct ixgbe_hw
*hw
= &adapter
->hw
;
6543 struct sockaddr
*addr
= p
;
6545 if (!is_valid_ether_addr(addr
->sa_data
))
6546 return -EADDRNOTAVAIL
;
6548 memcpy(netdev
->dev_addr
, addr
->sa_data
, netdev
->addr_len
);
6549 memcpy(hw
->mac
.addr
, addr
->sa_data
, netdev
->addr_len
);
6551 hw
->mac
.ops
.set_rar(hw
, 0, hw
->mac
.addr
, VMDQ_P(0), IXGBE_RAH_AV
);
6557 ixgbe_mdio_read(struct net_device
*netdev
, int prtad
, int devad
, u16 addr
)
6559 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6560 struct ixgbe_hw
*hw
= &adapter
->hw
;
6564 if (prtad
!= hw
->phy
.mdio
.prtad
)
6566 rc
= hw
->phy
.ops
.read_reg(hw
, addr
, devad
, &value
);
6572 static int ixgbe_mdio_write(struct net_device
*netdev
, int prtad
, int devad
,
6573 u16 addr
, u16 value
)
6575 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6576 struct ixgbe_hw
*hw
= &adapter
->hw
;
6578 if (prtad
!= hw
->phy
.mdio
.prtad
)
6580 return hw
->phy
.ops
.write_reg(hw
, addr
, devad
, value
);
6583 static int ixgbe_ioctl(struct net_device
*netdev
, struct ifreq
*req
, int cmd
)
6585 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6589 return ixgbe_ptp_hwtstamp_ioctl(adapter
, req
, cmd
);
6591 return mdio_mii_ioctl(&adapter
->hw
.phy
.mdio
, if_mii(req
), cmd
);
6596 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6598 * @netdev: network interface device structure
6600 * Returns non-zero on failure
6602 static int ixgbe_add_sanmac_netdev(struct net_device
*dev
)
6605 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
6606 struct ixgbe_hw
*hw
= &adapter
->hw
;
6608 if (is_valid_ether_addr(hw
->mac
.san_addr
)) {
6610 err
= dev_addr_add(dev
, hw
->mac
.san_addr
, NETDEV_HW_ADDR_T_SAN
);
6613 /* update SAN MAC vmdq pool selection */
6614 hw
->mac
.ops
.set_vmdq_san_mac(hw
, VMDQ_P(0));
6620 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6622 * @netdev: network interface device structure
6624 * Returns non-zero on failure
6626 static int ixgbe_del_sanmac_netdev(struct net_device
*dev
)
6629 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
6630 struct ixgbe_mac_info
*mac
= &adapter
->hw
.mac
;
6632 if (is_valid_ether_addr(mac
->san_addr
)) {
6634 err
= dev_addr_del(dev
, mac
->san_addr
, NETDEV_HW_ADDR_T_SAN
);
6640 #ifdef CONFIG_NET_POLL_CONTROLLER
6642 * Polling 'interrupt' - used by things like netconsole to send skbs
6643 * without having to re-enable interrupts. It's not called while
6644 * the interrupt routine is executing.
6646 static void ixgbe_netpoll(struct net_device
*netdev
)
6648 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6651 /* if interface is down do nothing */
6652 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
6655 adapter
->flags
|= IXGBE_FLAG_IN_NETPOLL
;
6656 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
6657 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
6658 ixgbe_msix_clean_rings(0, adapter
->q_vector
[i
]);
6660 ixgbe_intr(adapter
->pdev
->irq
, netdev
);
6662 adapter
->flags
&= ~IXGBE_FLAG_IN_NETPOLL
;
6666 static struct rtnl_link_stats64
*ixgbe_get_stats64(struct net_device
*netdev
,
6667 struct rtnl_link_stats64
*stats
)
6669 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6673 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
6674 struct ixgbe_ring
*ring
= ACCESS_ONCE(adapter
->rx_ring
[i
]);
6680 start
= u64_stats_fetch_begin_bh(&ring
->syncp
);
6681 packets
= ring
->stats
.packets
;
6682 bytes
= ring
->stats
.bytes
;
6683 } while (u64_stats_fetch_retry_bh(&ring
->syncp
, start
));
6684 stats
->rx_packets
+= packets
;
6685 stats
->rx_bytes
+= bytes
;
6689 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
6690 struct ixgbe_ring
*ring
= ACCESS_ONCE(adapter
->tx_ring
[i
]);
6696 start
= u64_stats_fetch_begin_bh(&ring
->syncp
);
6697 packets
= ring
->stats
.packets
;
6698 bytes
= ring
->stats
.bytes
;
6699 } while (u64_stats_fetch_retry_bh(&ring
->syncp
, start
));
6700 stats
->tx_packets
+= packets
;
6701 stats
->tx_bytes
+= bytes
;
6705 /* following stats updated by ixgbe_watchdog_task() */
6706 stats
->multicast
= netdev
->stats
.multicast
;
6707 stats
->rx_errors
= netdev
->stats
.rx_errors
;
6708 stats
->rx_length_errors
= netdev
->stats
.rx_length_errors
;
6709 stats
->rx_crc_errors
= netdev
->stats
.rx_crc_errors
;
6710 stats
->rx_missed_errors
= netdev
->stats
.rx_missed_errors
;
6714 #ifdef CONFIG_IXGBE_DCB
6716 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6717 * @adapter: pointer to ixgbe_adapter
6718 * @tc: number of traffic classes currently enabled
6720 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6721 * 802.1Q priority maps to a packet buffer that exists.
6723 static void ixgbe_validate_rtr(struct ixgbe_adapter
*adapter
, u8 tc
)
6725 struct ixgbe_hw
*hw
= &adapter
->hw
;
6729 /* 82598 have a static priority to TC mapping that can not
6730 * be changed so no validation is needed.
6732 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
6735 reg
= IXGBE_READ_REG(hw
, IXGBE_RTRUP2TC
);
6738 for (i
= 0; i
< MAX_TRAFFIC_CLASS
; i
++) {
6739 u8 up2tc
= reg
>> (i
* IXGBE_RTRUP2TC_UP_SHIFT
);
6741 /* If up2tc is out of bounds default to zero */
6743 reg
&= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT
);
6747 IXGBE_WRITE_REG(hw
, IXGBE_RTRUP2TC
, reg
);
6753 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
6754 * @adapter: Pointer to adapter struct
6756 * Populate the netdev user priority to tc map
6758 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter
*adapter
)
6760 struct net_device
*dev
= adapter
->netdev
;
6761 struct ixgbe_dcb_config
*dcb_cfg
= &adapter
->dcb_cfg
;
6762 struct ieee_ets
*ets
= adapter
->ixgbe_ieee_ets
;
6765 for (prio
= 0; prio
< MAX_USER_PRIORITY
; prio
++) {
6768 if (adapter
->dcbx_cap
& DCB_CAP_DCBX_VER_CEE
)
6769 tc
= ixgbe_dcb_get_tc_from_up(dcb_cfg
, 0, prio
);
6771 tc
= ets
->prio_tc
[prio
];
6773 netdev_set_prio_tc_map(dev
, prio
, tc
);
6778 * ixgbe_setup_tc - configure net_device for multiple traffic classes
6780 * @netdev: net device to configure
6781 * @tc: number of traffic classes to enable
6783 int ixgbe_setup_tc(struct net_device
*dev
, u8 tc
)
6785 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
6786 struct ixgbe_hw
*hw
= &adapter
->hw
;
6788 /* Hardware supports up to 8 traffic classes */
6789 if (tc
> adapter
->dcb_cfg
.num_tcs
.pg_tcs
||
6790 (hw
->mac
.type
== ixgbe_mac_82598EB
&&
6791 tc
< MAX_TRAFFIC_CLASS
))
6794 /* Hardware has to reinitialize queues and interrupts to
6795 * match packet buffer alignment. Unfortunately, the
6796 * hardware is not flexible enough to do this dynamically.
6798 if (netif_running(dev
))
6800 ixgbe_clear_interrupt_scheme(adapter
);
6803 netdev_set_num_tc(dev
, tc
);
6804 ixgbe_set_prio_tc_map(adapter
);
6806 adapter
->flags
|= IXGBE_FLAG_DCB_ENABLED
;
6808 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
6809 adapter
->last_lfc_mode
= adapter
->hw
.fc
.requested_mode
;
6810 adapter
->hw
.fc
.requested_mode
= ixgbe_fc_none
;
6813 netdev_reset_tc(dev
);
6815 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
)
6816 adapter
->hw
.fc
.requested_mode
= adapter
->last_lfc_mode
;
6818 adapter
->flags
&= ~IXGBE_FLAG_DCB_ENABLED
;
6820 adapter
->temp_dcb_cfg
.pfc_mode_enable
= false;
6821 adapter
->dcb_cfg
.pfc_mode_enable
= false;
6824 ixgbe_init_interrupt_scheme(adapter
);
6825 ixgbe_validate_rtr(adapter
, tc
);
6826 if (netif_running(dev
))
6832 #endif /* CONFIG_IXGBE_DCB */
6833 void ixgbe_do_reset(struct net_device
*netdev
)
6835 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6837 if (netif_running(netdev
))
6838 ixgbe_reinit_locked(adapter
);
6840 ixgbe_reset(adapter
);
6843 static netdev_features_t
ixgbe_fix_features(struct net_device
*netdev
,
6844 netdev_features_t features
)
6846 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6848 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6849 if (!(features
& NETIF_F_RXCSUM
))
6850 features
&= ~NETIF_F_LRO
;
6852 /* Turn off LRO if not RSC capable */
6853 if (!(adapter
->flags2
& IXGBE_FLAG2_RSC_CAPABLE
))
6854 features
&= ~NETIF_F_LRO
;
6859 static int ixgbe_set_features(struct net_device
*netdev
,
6860 netdev_features_t features
)
6862 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
6863 netdev_features_t changed
= netdev
->features
^ features
;
6864 bool need_reset
= false;
6866 /* Make sure RSC matches LRO, reset if change */
6867 if (!(features
& NETIF_F_LRO
)) {
6868 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)
6870 adapter
->flags2
&= ~IXGBE_FLAG2_RSC_ENABLED
;
6871 } else if ((adapter
->flags2
& IXGBE_FLAG2_RSC_CAPABLE
) &&
6872 !(adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)) {
6873 if (adapter
->rx_itr_setting
== 1 ||
6874 adapter
->rx_itr_setting
> IXGBE_MIN_RSC_ITR
) {
6875 adapter
->flags2
|= IXGBE_FLAG2_RSC_ENABLED
;
6877 } else if ((changed
^ features
) & NETIF_F_LRO
) {
6878 e_info(probe
, "rx-usecs set too low, "
6884 * Check if Flow Director n-tuple support was enabled or disabled. If
6885 * the state changed, we need to reset.
6887 switch (features
& NETIF_F_NTUPLE
) {
6888 case NETIF_F_NTUPLE
:
6889 /* turn off ATR, enable perfect filters and reset */
6890 if (!(adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
))
6893 adapter
->flags
&= ~IXGBE_FLAG_FDIR_HASH_CAPABLE
;
6894 adapter
->flags
|= IXGBE_FLAG_FDIR_PERFECT_CAPABLE
;
6897 /* turn off perfect filters, enable ATR and reset */
6898 if (adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
)
6901 adapter
->flags
&= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE
;
6903 /* We cannot enable ATR if SR-IOV is enabled */
6904 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
6907 /* We cannot enable ATR if we have 2 or more traffic classes */
6908 if (netdev_get_num_tc(netdev
) > 1)
6911 /* We cannot enable ATR if RSS is disabled */
6912 if (adapter
->ring_feature
[RING_F_RSS
].limit
<= 1)
6915 /* A sample rate of 0 indicates ATR disabled */
6916 if (!adapter
->atr_sample_rate
)
6919 adapter
->flags
|= IXGBE_FLAG_FDIR_HASH_CAPABLE
;
6923 if (features
& NETIF_F_HW_VLAN_RX
)
6924 ixgbe_vlan_strip_enable(adapter
);
6926 ixgbe_vlan_strip_disable(adapter
);
6928 if (changed
& NETIF_F_RXALL
)
6931 netdev
->features
= features
;
6933 ixgbe_do_reset(netdev
);
6938 static int ixgbe_ndo_fdb_add(struct ndmsg
*ndm
, struct nlattr
*tb
[],
6939 struct net_device
*dev
,
6940 const unsigned char *addr
,
6943 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
6946 if (!(adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
))
6949 /* Hardware does not support aging addresses so if a
6950 * ndm_state is given only allow permanent addresses
6952 if (ndm
->ndm_state
&& !(ndm
->ndm_state
& NUD_PERMANENT
)) {
6953 pr_info("%s: FDB only supports static addresses\n",
6958 if (is_unicast_ether_addr(addr
) || is_link_local_ether_addr(addr
)) {
6959 u32 rar_uc_entries
= IXGBE_MAX_PF_MACVLANS
;
6961 if (netdev_uc_count(dev
) < rar_uc_entries
)
6962 err
= dev_uc_add_excl(dev
, addr
);
6965 } else if (is_multicast_ether_addr(addr
)) {
6966 err
= dev_mc_add_excl(dev
, addr
);
6971 /* Only return duplicate errors if NLM_F_EXCL is set */
6972 if (err
== -EEXIST
&& !(flags
& NLM_F_EXCL
))
6978 static int ixgbe_ndo_fdb_del(struct ndmsg
*ndm
,
6979 struct net_device
*dev
,
6980 const unsigned char *addr
)
6982 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
6983 int err
= -EOPNOTSUPP
;
6985 if (ndm
->ndm_state
& NUD_PERMANENT
) {
6986 pr_info("%s: FDB only supports static addresses\n",
6991 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
6992 if (is_unicast_ether_addr(addr
))
6993 err
= dev_uc_del(dev
, addr
);
6994 else if (is_multicast_ether_addr(addr
))
6995 err
= dev_mc_del(dev
, addr
);
7003 static int ixgbe_ndo_fdb_dump(struct sk_buff
*skb
,
7004 struct netlink_callback
*cb
,
7005 struct net_device
*dev
,
7008 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
7010 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
)
7011 idx
= ndo_dflt_fdb_dump(skb
, cb
, dev
, idx
);
7016 static int ixgbe_ndo_bridge_setlink(struct net_device
*dev
,
7017 struct nlmsghdr
*nlh
)
7019 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
7020 struct nlattr
*attr
, *br_spec
;
7023 if (!(adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
))
7026 br_spec
= nlmsg_find_attr(nlh
, sizeof(struct ifinfomsg
), IFLA_AF_SPEC
);
7028 nla_for_each_nested(attr
, br_spec
, rem
) {
7032 if (nla_type(attr
) != IFLA_BRIDGE_MODE
)
7035 mode
= nla_get_u16(attr
);
7036 if (mode
== BRIDGE_MODE_VEPA
)
7038 else if (mode
== BRIDGE_MODE_VEB
)
7039 reg
= IXGBE_PFDTXGSWC_VT_LBEN
;
7043 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_PFDTXGSWC
, reg
);
7045 e_info(drv
, "enabling bridge mode: %s\n",
7046 mode
== BRIDGE_MODE_VEPA
? "VEPA" : "VEB");
7052 static int ixgbe_ndo_bridge_getlink(struct sk_buff
*skb
, u32 pid
, u32 seq
,
7053 struct net_device
*dev
)
7055 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
7058 if (!(adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
))
7061 if (IXGBE_READ_REG(&adapter
->hw
, IXGBE_PFDTXGSWC
) & 1)
7062 mode
= BRIDGE_MODE_VEB
;
7064 mode
= BRIDGE_MODE_VEPA
;
7066 return ndo_dflt_bridge_getlink(skb
, pid
, seq
, dev
, mode
);
7069 static const struct net_device_ops ixgbe_netdev_ops
= {
7070 .ndo_open
= ixgbe_open
,
7071 .ndo_stop
= ixgbe_close
,
7072 .ndo_start_xmit
= ixgbe_xmit_frame
,
7073 .ndo_select_queue
= ixgbe_select_queue
,
7074 .ndo_set_rx_mode
= ixgbe_set_rx_mode
,
7075 .ndo_validate_addr
= eth_validate_addr
,
7076 .ndo_set_mac_address
= ixgbe_set_mac
,
7077 .ndo_change_mtu
= ixgbe_change_mtu
,
7078 .ndo_tx_timeout
= ixgbe_tx_timeout
,
7079 .ndo_vlan_rx_add_vid
= ixgbe_vlan_rx_add_vid
,
7080 .ndo_vlan_rx_kill_vid
= ixgbe_vlan_rx_kill_vid
,
7081 .ndo_do_ioctl
= ixgbe_ioctl
,
7082 .ndo_set_vf_mac
= ixgbe_ndo_set_vf_mac
,
7083 .ndo_set_vf_vlan
= ixgbe_ndo_set_vf_vlan
,
7084 .ndo_set_vf_tx_rate
= ixgbe_ndo_set_vf_bw
,
7085 .ndo_set_vf_spoofchk
= ixgbe_ndo_set_vf_spoofchk
,
7086 .ndo_get_vf_config
= ixgbe_ndo_get_vf_config
,
7087 .ndo_get_stats64
= ixgbe_get_stats64
,
7088 #ifdef CONFIG_IXGBE_DCB
7089 .ndo_setup_tc
= ixgbe_setup_tc
,
7091 #ifdef CONFIG_NET_POLL_CONTROLLER
7092 .ndo_poll_controller
= ixgbe_netpoll
,
7095 .ndo_fcoe_ddp_setup
= ixgbe_fcoe_ddp_get
,
7096 .ndo_fcoe_ddp_target
= ixgbe_fcoe_ddp_target
,
7097 .ndo_fcoe_ddp_done
= ixgbe_fcoe_ddp_put
,
7098 .ndo_fcoe_enable
= ixgbe_fcoe_enable
,
7099 .ndo_fcoe_disable
= ixgbe_fcoe_disable
,
7100 .ndo_fcoe_get_wwn
= ixgbe_fcoe_get_wwn
,
7101 .ndo_fcoe_get_hbainfo
= ixgbe_fcoe_get_hbainfo
,
7102 #endif /* IXGBE_FCOE */
7103 .ndo_set_features
= ixgbe_set_features
,
7104 .ndo_fix_features
= ixgbe_fix_features
,
7105 .ndo_fdb_add
= ixgbe_ndo_fdb_add
,
7106 .ndo_fdb_del
= ixgbe_ndo_fdb_del
,
7107 .ndo_fdb_dump
= ixgbe_ndo_fdb_dump
,
7108 .ndo_bridge_setlink
= ixgbe_ndo_bridge_setlink
,
7109 .ndo_bridge_getlink
= ixgbe_ndo_bridge_getlink
,
7113 * ixgbe_wol_supported - Check whether device supports WoL
7114 * @hw: hw specific details
7115 * @device_id: the device ID
7116 * @subdev_id: the subsystem device ID
7118 * This function is used by probe and ethtool to determine
7119 * which devices have WoL support
7122 int ixgbe_wol_supported(struct ixgbe_adapter
*adapter
, u16 device_id
,
7125 struct ixgbe_hw
*hw
= &adapter
->hw
;
7126 u16 wol_cap
= adapter
->eeprom_cap
& IXGBE_DEVICE_CAPS_WOL_MASK
;
7127 int is_wol_supported
= 0;
7129 switch (device_id
) {
7130 case IXGBE_DEV_ID_82599_SFP
:
7131 /* Only these subdevices could supports WOL */
7132 switch (subdevice_id
) {
7133 case IXGBE_SUBDEV_ID_82599_560FLR
:
7134 /* only support first port */
7135 if (hw
->bus
.func
!= 0)
7137 case IXGBE_SUBDEV_ID_82599_SFP
:
7138 case IXGBE_SUBDEV_ID_82599_RNDC
:
7139 case IXGBE_SUBDEV_ID_82599_ECNA_DP
:
7140 is_wol_supported
= 1;
7144 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE
:
7145 /* All except this subdevice support WOL */
7146 if (subdevice_id
!= IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ
)
7147 is_wol_supported
= 1;
7149 case IXGBE_DEV_ID_82599_KX4
:
7150 is_wol_supported
= 1;
7152 case IXGBE_DEV_ID_X540T
:
7153 case IXGBE_DEV_ID_X540T1
:
7154 /* check eeprom to see if enabled wol */
7155 if ((wol_cap
== IXGBE_DEVICE_CAPS_WOL_PORT0_1
) ||
7156 ((wol_cap
== IXGBE_DEVICE_CAPS_WOL_PORT0
) &&
7157 (hw
->bus
.func
== 0))) {
7158 is_wol_supported
= 1;
7163 return is_wol_supported
;
7167 * ixgbe_probe - Device Initialization Routine
7168 * @pdev: PCI device information struct
7169 * @ent: entry in ixgbe_pci_tbl
7171 * Returns 0 on success, negative on failure
7173 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
7174 * The OS initialization, configuring of the adapter private structure,
7175 * and a hardware reset occur.
7177 static int __devinit
ixgbe_probe(struct pci_dev
*pdev
,
7178 const struct pci_device_id
*ent
)
7180 struct net_device
*netdev
;
7181 struct ixgbe_adapter
*adapter
= NULL
;
7182 struct ixgbe_hw
*hw
;
7183 const struct ixgbe_info
*ii
= ixgbe_info_tbl
[ent
->driver_data
];
7184 static int cards_found
;
7185 int i
, err
, pci_using_dac
;
7186 u8 part_str
[IXGBE_PBANUM_LENGTH
];
7187 unsigned int indices
= num_possible_cpus();
7188 unsigned int dcb_max
= 0;
7194 /* Catch broken hardware that put the wrong VF device ID in
7195 * the PCIe SR-IOV capability.
7197 if (pdev
->is_virtfn
) {
7198 WARN(1, KERN_ERR
"%s (%hx:%hx) should not be a VF!\n",
7199 pci_name(pdev
), pdev
->vendor
, pdev
->device
);
7203 err
= pci_enable_device_mem(pdev
);
7207 if (!dma_set_mask(&pdev
->dev
, DMA_BIT_MASK(64)) &&
7208 !dma_set_coherent_mask(&pdev
->dev
, DMA_BIT_MASK(64))) {
7211 err
= dma_set_mask(&pdev
->dev
, DMA_BIT_MASK(32));
7213 err
= dma_set_coherent_mask(&pdev
->dev
,
7217 "No usable DMA configuration, aborting\n");
7224 err
= pci_request_selected_regions(pdev
, pci_select_bars(pdev
,
7225 IORESOURCE_MEM
), ixgbe_driver_name
);
7228 "pci_request_selected_regions failed 0x%x\n", err
);
7232 pci_enable_pcie_error_reporting(pdev
);
7234 pci_set_master(pdev
);
7235 pci_save_state(pdev
);
7237 #ifdef CONFIG_IXGBE_DCB
7238 if (ii
->mac
== ixgbe_mac_82598EB
)
7239 dcb_max
= min_t(unsigned int, indices
* MAX_TRAFFIC_CLASS
,
7240 IXGBE_MAX_RSS_INDICES
);
7242 dcb_max
= min_t(unsigned int, indices
* MAX_TRAFFIC_CLASS
,
7243 IXGBE_MAX_FDIR_INDICES
);
7246 if (ii
->mac
== ixgbe_mac_82598EB
)
7247 indices
= min_t(unsigned int, indices
, IXGBE_MAX_RSS_INDICES
);
7249 indices
= min_t(unsigned int, indices
, IXGBE_MAX_FDIR_INDICES
);
7252 indices
+= min_t(unsigned int, num_possible_cpus(),
7253 IXGBE_MAX_FCOE_INDICES
);
7255 indices
= max_t(unsigned int, dcb_max
, indices
);
7256 netdev
= alloc_etherdev_mq(sizeof(struct ixgbe_adapter
), indices
);
7259 goto err_alloc_etherdev
;
7262 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
7264 adapter
= netdev_priv(netdev
);
7265 pci_set_drvdata(pdev
, adapter
);
7267 adapter
->netdev
= netdev
;
7268 adapter
->pdev
= pdev
;
7271 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
7273 hw
->hw_addr
= ioremap(pci_resource_start(pdev
, 0),
7274 pci_resource_len(pdev
, 0));
7280 netdev
->netdev_ops
= &ixgbe_netdev_ops
;
7281 ixgbe_set_ethtool_ops(netdev
);
7282 netdev
->watchdog_timeo
= 5 * HZ
;
7283 strncpy(netdev
->name
, pci_name(pdev
), sizeof(netdev
->name
) - 1);
7285 adapter
->bd_number
= cards_found
;
7288 memcpy(&hw
->mac
.ops
, ii
->mac_ops
, sizeof(hw
->mac
.ops
));
7289 hw
->mac
.type
= ii
->mac
;
7292 memcpy(&hw
->eeprom
.ops
, ii
->eeprom_ops
, sizeof(hw
->eeprom
.ops
));
7293 eec
= IXGBE_READ_REG(hw
, IXGBE_EEC
);
7294 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7295 if (!(eec
& (1 << 8)))
7296 hw
->eeprom
.ops
.read
= &ixgbe_read_eeprom_bit_bang_generic
;
7299 memcpy(&hw
->phy
.ops
, ii
->phy_ops
, sizeof(hw
->phy
.ops
));
7300 hw
->phy
.sfp_type
= ixgbe_sfp_type_unknown
;
7301 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7302 hw
->phy
.mdio
.prtad
= MDIO_PRTAD_NONE
;
7303 hw
->phy
.mdio
.mmds
= 0;
7304 hw
->phy
.mdio
.mode_support
= MDIO_SUPPORTS_C45
| MDIO_EMULATE_C22
;
7305 hw
->phy
.mdio
.dev
= netdev
;
7306 hw
->phy
.mdio
.mdio_read
= ixgbe_mdio_read
;
7307 hw
->phy
.mdio
.mdio_write
= ixgbe_mdio_write
;
7309 ii
->get_invariants(hw
);
7311 /* setup the private structure */
7312 err
= ixgbe_sw_init(adapter
);
7316 /* Make it possible the adapter to be woken up via WOL */
7317 switch (adapter
->hw
.mac
.type
) {
7318 case ixgbe_mac_82599EB
:
7319 case ixgbe_mac_X540
:
7320 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_WUS
, ~0);
7327 * If there is a fan on this device and it has failed log the
7330 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) {
7331 u32 esdp
= IXGBE_READ_REG(hw
, IXGBE_ESDP
);
7332 if (esdp
& IXGBE_ESDP_SDP1
)
7333 e_crit(probe
, "Fan has stopped, replace the adapter\n");
7336 if (allow_unsupported_sfp
)
7337 hw
->allow_unsupported_sfp
= allow_unsupported_sfp
;
7339 /* reset_hw fills in the perm_addr as well */
7340 hw
->phy
.reset_if_overtemp
= true;
7341 err
= hw
->mac
.ops
.reset_hw(hw
);
7342 hw
->phy
.reset_if_overtemp
= false;
7343 if (err
== IXGBE_ERR_SFP_NOT_PRESENT
&&
7344 hw
->mac
.type
== ixgbe_mac_82598EB
) {
7346 } else if (err
== IXGBE_ERR_SFP_NOT_SUPPORTED
) {
7347 e_dev_err("failed to load because an unsupported SFP+ "
7348 "module type was detected.\n");
7349 e_dev_err("Reload the driver after installing a supported "
7353 e_dev_err("HW Init failed: %d\n", err
);
7357 #ifdef CONFIG_PCI_IOV
7358 ixgbe_enable_sriov(adapter
, ii
);
7361 netdev
->features
= NETIF_F_SG
|
7364 NETIF_F_HW_VLAN_TX
|
7365 NETIF_F_HW_VLAN_RX
|
7366 NETIF_F_HW_VLAN_FILTER
|
7372 netdev
->hw_features
= netdev
->features
;
7374 switch (adapter
->hw
.mac
.type
) {
7375 case ixgbe_mac_82599EB
:
7376 case ixgbe_mac_X540
:
7377 netdev
->features
|= NETIF_F_SCTP_CSUM
;
7378 netdev
->hw_features
|= NETIF_F_SCTP_CSUM
|
7385 netdev
->hw_features
|= NETIF_F_RXALL
;
7387 netdev
->vlan_features
|= NETIF_F_TSO
;
7388 netdev
->vlan_features
|= NETIF_F_TSO6
;
7389 netdev
->vlan_features
|= NETIF_F_IP_CSUM
;
7390 netdev
->vlan_features
|= NETIF_F_IPV6_CSUM
;
7391 netdev
->vlan_features
|= NETIF_F_SG
;
7393 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
7394 netdev
->priv_flags
|= IFF_SUPP_NOFCS
;
7396 #ifdef CONFIG_IXGBE_DCB
7397 netdev
->dcbnl_ops
= &dcbnl_ops
;
7401 if (adapter
->flags
& IXGBE_FLAG_FCOE_CAPABLE
) {
7402 if (hw
->mac
.ops
.get_device_caps
) {
7403 hw
->mac
.ops
.get_device_caps(hw
, &device_caps
);
7404 if (device_caps
& IXGBE_DEVICE_CAPS_FCOE_OFFLOADS
)
7405 adapter
->flags
&= ~IXGBE_FLAG_FCOE_CAPABLE
;
7408 adapter
->ring_feature
[RING_F_FCOE
].limit
= IXGBE_FCRETA_SIZE
;
7410 netdev
->features
|= NETIF_F_FSO
|
7413 netdev
->vlan_features
|= NETIF_F_FSO
|
7417 #endif /* IXGBE_FCOE */
7418 if (pci_using_dac
) {
7419 netdev
->features
|= NETIF_F_HIGHDMA
;
7420 netdev
->vlan_features
|= NETIF_F_HIGHDMA
;
7423 if (adapter
->flags2
& IXGBE_FLAG2_RSC_CAPABLE
)
7424 netdev
->hw_features
|= NETIF_F_LRO
;
7425 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)
7426 netdev
->features
|= NETIF_F_LRO
;
7428 /* make sure the EEPROM is good */
7429 if (hw
->eeprom
.ops
.validate_checksum(hw
, NULL
) < 0) {
7430 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7435 memcpy(netdev
->dev_addr
, hw
->mac
.perm_addr
, netdev
->addr_len
);
7436 memcpy(netdev
->perm_addr
, hw
->mac
.perm_addr
, netdev
->addr_len
);
7438 if (ixgbe_validate_mac_addr(netdev
->perm_addr
)) {
7439 e_dev_err("invalid MAC address\n");
7444 setup_timer(&adapter
->service_timer
, &ixgbe_service_timer
,
7445 (unsigned long) adapter
);
7447 INIT_WORK(&adapter
->service_task
, ixgbe_service_task
);
7448 clear_bit(__IXGBE_SERVICE_SCHED
, &adapter
->state
);
7450 err
= ixgbe_init_interrupt_scheme(adapter
);
7454 /* WOL not supported for all devices */
7456 hw
->eeprom
.ops
.read(hw
, 0x2c, &adapter
->eeprom_cap
);
7457 if (ixgbe_wol_supported(adapter
, pdev
->device
, pdev
->subsystem_device
))
7458 adapter
->wol
= IXGBE_WUFC_MAG
;
7460 device_set_wakeup_enable(&adapter
->pdev
->dev
, adapter
->wol
);
7462 /* save off EEPROM version number */
7463 hw
->eeprom
.ops
.read(hw
, 0x2e, &adapter
->eeprom_verh
);
7464 hw
->eeprom
.ops
.read(hw
, 0x2d, &adapter
->eeprom_verl
);
7466 /* pick up the PCI bus settings for reporting later */
7467 hw
->mac
.ops
.get_bus_info(hw
);
7469 /* print bus type/speed/width info */
7470 e_dev_info("(PCI Express:%s:%s) %pM\n",
7471 (hw
->bus
.speed
== ixgbe_bus_speed_5000
? "5.0GT/s" :
7472 hw
->bus
.speed
== ixgbe_bus_speed_2500
? "2.5GT/s" :
7474 (hw
->bus
.width
== ixgbe_bus_width_pcie_x8
? "Width x8" :
7475 hw
->bus
.width
== ixgbe_bus_width_pcie_x4
? "Width x4" :
7476 hw
->bus
.width
== ixgbe_bus_width_pcie_x1
? "Width x1" :
7480 err
= ixgbe_read_pba_string_generic(hw
, part_str
, IXGBE_PBANUM_LENGTH
);
7482 strncpy(part_str
, "Unknown", IXGBE_PBANUM_LENGTH
);
7483 if (ixgbe_is_sfp(hw
) && hw
->phy
.sfp_type
!= ixgbe_sfp_type_not_present
)
7484 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7485 hw
->mac
.type
, hw
->phy
.type
, hw
->phy
.sfp_type
,
7488 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7489 hw
->mac
.type
, hw
->phy
.type
, part_str
);
7491 if (hw
->bus
.width
<= ixgbe_bus_width_pcie_x4
) {
7492 e_dev_warn("PCI-Express bandwidth available for this card is "
7493 "not sufficient for optimal performance.\n");
7494 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7498 /* reset the hardware with the new settings */
7499 err
= hw
->mac
.ops
.start_hw(hw
);
7500 if (err
== IXGBE_ERR_EEPROM_VERSION
) {
7501 /* We are running on a pre-production device, log a warning */
7502 e_dev_warn("This device is a pre-production adapter/LOM. "
7503 "Please be aware there may be issues associated "
7504 "with your hardware. If you are experiencing "
7505 "problems please contact your Intel or hardware "
7506 "representative who provided you with this "
7509 strcpy(netdev
->name
, "eth%d");
7510 err
= register_netdev(netdev
);
7514 /* power down the optics for 82599 SFP+ fiber */
7515 if (hw
->mac
.ops
.disable_tx_laser
)
7516 hw
->mac
.ops
.disable_tx_laser(hw
);
7518 /* carrier off reporting is important to ethtool even BEFORE open */
7519 netif_carrier_off(netdev
);
7521 #ifdef CONFIG_IXGBE_DCA
7522 if (dca_add_requester(&pdev
->dev
) == 0) {
7523 adapter
->flags
|= IXGBE_FLAG_DCA_ENABLED
;
7524 ixgbe_setup_dca(adapter
);
7527 if (adapter
->flags
& IXGBE_FLAG_SRIOV_ENABLED
) {
7528 e_info(probe
, "IOV is enabled with %d VFs\n", adapter
->num_vfs
);
7529 for (i
= 0; i
< adapter
->num_vfs
; i
++)
7530 ixgbe_vf_configuration(pdev
, (i
| 0x10000000));
7533 /* firmware requires driver version to be 0xFFFFFFFF
7534 * since os does not support feature
7536 if (hw
->mac
.ops
.set_fw_drv_ver
)
7537 hw
->mac
.ops
.set_fw_drv_ver(hw
, 0xFF, 0xFF, 0xFF,
7540 /* add san mac addr to netdev */
7541 ixgbe_add_sanmac_netdev(netdev
);
7543 e_dev_info("%s\n", ixgbe_default_device_descr
);
7546 #ifdef CONFIG_IXGBE_HWMON
7547 if (ixgbe_sysfs_init(adapter
))
7548 e_err(probe
, "failed to allocate sysfs resources\n");
7549 #endif /* CONFIG_IXGBE_HWMON */
7551 #ifdef CONFIG_DEBUG_FS
7552 ixgbe_dbg_adapter_init(adapter
);
7553 #endif /* CONFIG_DEBUG_FS */
7558 ixgbe_release_hw_control(adapter
);
7559 ixgbe_clear_interrupt_scheme(adapter
);
7561 ixgbe_disable_sriov(adapter
);
7562 adapter
->flags2
&= ~IXGBE_FLAG2_SEARCH_FOR_SFP
;
7563 iounmap(hw
->hw_addr
);
7565 free_netdev(netdev
);
7567 pci_release_selected_regions(pdev
,
7568 pci_select_bars(pdev
, IORESOURCE_MEM
));
7571 pci_disable_device(pdev
);
7576 * ixgbe_remove - Device Removal Routine
7577 * @pdev: PCI device information struct
7579 * ixgbe_remove is called by the PCI subsystem to alert the driver
7580 * that it should release a PCI device. The could be caused by a
7581 * Hot-Plug event, or because the driver is going to be removed from
7584 static void __devexit
ixgbe_remove(struct pci_dev
*pdev
)
7586 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
7587 struct net_device
*netdev
= adapter
->netdev
;
7589 #ifdef CONFIG_DEBUG_FS
7590 ixgbe_dbg_adapter_exit(adapter
);
7591 #endif /*CONFIG_DEBUG_FS */
7593 set_bit(__IXGBE_DOWN
, &adapter
->state
);
7594 cancel_work_sync(&adapter
->service_task
);
7597 #ifdef CONFIG_IXGBE_DCA
7598 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
) {
7599 adapter
->flags
&= ~IXGBE_FLAG_DCA_ENABLED
;
7600 dca_remove_requester(&pdev
->dev
);
7601 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
, 1);
7605 #ifdef CONFIG_IXGBE_HWMON
7606 ixgbe_sysfs_exit(adapter
);
7607 #endif /* CONFIG_IXGBE_HWMON */
7609 /* remove the added san mac */
7610 ixgbe_del_sanmac_netdev(netdev
);
7612 if (netdev
->reg_state
== NETREG_REGISTERED
)
7613 unregister_netdev(netdev
);
7615 ixgbe_disable_sriov(adapter
);
7617 ixgbe_clear_interrupt_scheme(adapter
);
7619 ixgbe_release_hw_control(adapter
);
7622 kfree(adapter
->ixgbe_ieee_pfc
);
7623 kfree(adapter
->ixgbe_ieee_ets
);
7626 iounmap(adapter
->hw
.hw_addr
);
7627 pci_release_selected_regions(pdev
, pci_select_bars(pdev
,
7630 e_dev_info("complete\n");
7632 free_netdev(netdev
);
7634 pci_disable_pcie_error_reporting(pdev
);
7636 pci_disable_device(pdev
);
7640 * ixgbe_io_error_detected - called when PCI error is detected
7641 * @pdev: Pointer to PCI device
7642 * @state: The current pci connection state
7644 * This function is called after a PCI bus error affecting
7645 * this device has been detected.
7647 static pci_ers_result_t
ixgbe_io_error_detected(struct pci_dev
*pdev
,
7648 pci_channel_state_t state
)
7650 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
7651 struct net_device
*netdev
= adapter
->netdev
;
7653 #ifdef CONFIG_PCI_IOV
7654 struct pci_dev
*bdev
, *vfdev
;
7655 u32 dw0
, dw1
, dw2
, dw3
;
7657 u16 req_id
, pf_func
;
7659 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
||
7660 adapter
->num_vfs
== 0)
7661 goto skip_bad_vf_detection
;
7663 bdev
= pdev
->bus
->self
;
7664 while (bdev
&& (pci_pcie_type(bdev
) != PCI_EXP_TYPE_ROOT_PORT
))
7665 bdev
= bdev
->bus
->self
;
7668 goto skip_bad_vf_detection
;
7670 pos
= pci_find_ext_capability(bdev
, PCI_EXT_CAP_ID_ERR
);
7672 goto skip_bad_vf_detection
;
7674 pci_read_config_dword(bdev
, pos
+ PCI_ERR_HEADER_LOG
, &dw0
);
7675 pci_read_config_dword(bdev
, pos
+ PCI_ERR_HEADER_LOG
+ 4, &dw1
);
7676 pci_read_config_dword(bdev
, pos
+ PCI_ERR_HEADER_LOG
+ 8, &dw2
);
7677 pci_read_config_dword(bdev
, pos
+ PCI_ERR_HEADER_LOG
+ 12, &dw3
);
7680 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7681 if (!(req_id
& 0x0080))
7682 goto skip_bad_vf_detection
;
7684 pf_func
= req_id
& 0x01;
7685 if ((pf_func
& 1) == (pdev
->devfn
& 1)) {
7686 unsigned int device_id
;
7688 vf
= (req_id
& 0x7F) >> 1;
7689 e_dev_err("VF %d has caused a PCIe error\n", vf
);
7690 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7691 "%8.8x\tdw3: %8.8x\n",
7692 dw0
, dw1
, dw2
, dw3
);
7693 switch (adapter
->hw
.mac
.type
) {
7694 case ixgbe_mac_82599EB
:
7695 device_id
= IXGBE_82599_VF_DEVICE_ID
;
7697 case ixgbe_mac_X540
:
7698 device_id
= IXGBE_X540_VF_DEVICE_ID
;
7705 /* Find the pci device of the offending VF */
7706 vfdev
= pci_get_device(PCI_VENDOR_ID_INTEL
, device_id
, NULL
);
7708 if (vfdev
->devfn
== (req_id
& 0xFF))
7710 vfdev
= pci_get_device(PCI_VENDOR_ID_INTEL
,
7714 * There's a slim chance the VF could have been hot plugged,
7715 * so if it is no longer present we don't need to issue the
7716 * VFLR. Just clean up the AER in that case.
7719 e_dev_err("Issuing VFLR to VF %d\n", vf
);
7720 pci_write_config_dword(vfdev
, 0xA8, 0x00008000);
7723 pci_cleanup_aer_uncorrect_error_status(pdev
);
7727 * Even though the error may have occurred on the other port
7728 * we still need to increment the vf error reference count for
7729 * both ports because the I/O resume function will be called
7732 adapter
->vferr_refcount
++;
7734 return PCI_ERS_RESULT_RECOVERED
;
7736 skip_bad_vf_detection
:
7737 #endif /* CONFIG_PCI_IOV */
7738 netif_device_detach(netdev
);
7740 if (state
== pci_channel_io_perm_failure
)
7741 return PCI_ERS_RESULT_DISCONNECT
;
7743 if (netif_running(netdev
))
7744 ixgbe_down(adapter
);
7745 pci_disable_device(pdev
);
7747 /* Request a slot reset. */
7748 return PCI_ERS_RESULT_NEED_RESET
;
7752 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7753 * @pdev: Pointer to PCI device
7755 * Restart the card from scratch, as if from a cold-boot.
7757 static pci_ers_result_t
ixgbe_io_slot_reset(struct pci_dev
*pdev
)
7759 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
7760 pci_ers_result_t result
;
7763 if (pci_enable_device_mem(pdev
)) {
7764 e_err(probe
, "Cannot re-enable PCI device after reset.\n");
7765 result
= PCI_ERS_RESULT_DISCONNECT
;
7767 pci_set_master(pdev
);
7768 pci_restore_state(pdev
);
7769 pci_save_state(pdev
);
7771 pci_wake_from_d3(pdev
, false);
7773 ixgbe_reset(adapter
);
7774 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_WUS
, ~0);
7775 result
= PCI_ERS_RESULT_RECOVERED
;
7778 err
= pci_cleanup_aer_uncorrect_error_status(pdev
);
7780 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7781 "failed 0x%0x\n", err
);
7782 /* non-fatal, continue */
7789 * ixgbe_io_resume - called when traffic can start flowing again.
7790 * @pdev: Pointer to PCI device
7792 * This callback is called when the error recovery driver tells us that
7793 * its OK to resume normal operation.
7795 static void ixgbe_io_resume(struct pci_dev
*pdev
)
7797 struct ixgbe_adapter
*adapter
= pci_get_drvdata(pdev
);
7798 struct net_device
*netdev
= adapter
->netdev
;
7800 #ifdef CONFIG_PCI_IOV
7801 if (adapter
->vferr_refcount
) {
7802 e_info(drv
, "Resuming after VF err\n");
7803 adapter
->vferr_refcount
--;
7808 if (netif_running(netdev
))
7811 netif_device_attach(netdev
);
7814 static const struct pci_error_handlers ixgbe_err_handler
= {
7815 .error_detected
= ixgbe_io_error_detected
,
7816 .slot_reset
= ixgbe_io_slot_reset
,
7817 .resume
= ixgbe_io_resume
,
7820 static struct pci_driver ixgbe_driver
= {
7821 .name
= ixgbe_driver_name
,
7822 .id_table
= ixgbe_pci_tbl
,
7823 .probe
= ixgbe_probe
,
7824 .remove
= __devexit_p(ixgbe_remove
),
7826 .suspend
= ixgbe_suspend
,
7827 .resume
= ixgbe_resume
,
7829 .shutdown
= ixgbe_shutdown
,
7830 .err_handler
= &ixgbe_err_handler
7834 * ixgbe_init_module - Driver Registration Routine
7836 * ixgbe_init_module is the first routine called when the driver is
7837 * loaded. All it does is register with the PCI subsystem.
7839 static int __init
ixgbe_init_module(void)
7842 pr_info("%s - version %s\n", ixgbe_driver_string
, ixgbe_driver_version
);
7843 pr_info("%s\n", ixgbe_copyright
);
7845 #ifdef CONFIG_DEBUG_FS
7847 #endif /* CONFIG_DEBUG_FS */
7849 #ifdef CONFIG_IXGBE_DCA
7850 dca_register_notify(&dca_notifier
);
7853 ret
= pci_register_driver(&ixgbe_driver
);
7857 module_init(ixgbe_init_module
);
7860 * ixgbe_exit_module - Driver Exit Cleanup Routine
7862 * ixgbe_exit_module is called just before the driver is removed
7865 static void __exit
ixgbe_exit_module(void)
7867 #ifdef CONFIG_IXGBE_DCA
7868 dca_unregister_notify(&dca_notifier
);
7870 pci_unregister_driver(&ixgbe_driver
);
7872 #ifdef CONFIG_DEBUG_FS
7874 #endif /* CONFIG_DEBUG_FS */
7876 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7879 #ifdef CONFIG_IXGBE_DCA
7880 static int ixgbe_notify_dca(struct notifier_block
*nb
, unsigned long event
,
7885 ret_val
= driver_for_each_device(&ixgbe_driver
.driver
, NULL
, &event
,
7886 __ixgbe_notify_dca
);
7888 return ret_val
? NOTIFY_BAD
: NOTIFY_DONE
;
7891 #endif /* CONFIG_IXGBE_DCA */
7893 module_exit(ixgbe_exit_module
);