1 /* Intel(R) Gigabit Ethernet Linux driver
2 * Copyright(c) 2007-2014 Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
16 * The full GNU General Public License is included in this distribution in
17 * the file called "COPYING".
19 * Contact Information:
20 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
21 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/init.h>
29 #include <linux/bitops.h>
30 #include <linux/vmalloc.h>
31 #include <linux/pagemap.h>
32 #include <linux/netdevice.h>
33 #include <linux/ipv6.h>
34 #include <linux/slab.h>
35 #include <net/checksum.h>
36 #include <net/ip6_checksum.h>
37 #include <linux/net_tstamp.h>
38 #include <linux/mii.h>
39 #include <linux/ethtool.h>
41 #include <linux/if_vlan.h>
42 #include <linux/pci.h>
43 #include <linux/pci-aspm.h>
44 #include <linux/delay.h>
45 #include <linux/interrupt.h>
47 #include <linux/tcp.h>
48 #include <linux/sctp.h>
49 #include <linux/if_ether.h>
50 #include <linux/aer.h>
51 #include <linux/prefetch.h>
52 #include <linux/pm_runtime.h>
54 #include <linux/dca.h>
56 #include <linux/i2c.h>
62 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
63 __stringify(BUILD) "-k"
64 char igb_driver_name
[] = "igb";
65 char igb_driver_version
[] = DRV_VERSION
;
66 static const char igb_driver_string
[] =
67 "Intel(R) Gigabit Ethernet Network Driver";
68 static const char igb_copyright
[] =
69 "Copyright (c) 2007-2014 Intel Corporation.";
71 static const struct e1000_info
*igb_info_tbl
[] = {
72 [board_82575
] = &e1000_82575_info
,
75 static const struct pci_device_id igb_pci_tbl
[] = {
76 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I354_BACKPLANE_1GBPS
) },
77 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I354_SGMII
) },
78 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS
) },
79 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I211_COPPER
), board_82575
},
80 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_COPPER
), board_82575
},
81 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_FIBER
), board_82575
},
82 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_SERDES
), board_82575
},
83 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_SGMII
), board_82575
},
84 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_COPPER_FLASHLESS
), board_82575
},
85 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_SERDES_FLASHLESS
), board_82575
},
86 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I350_COPPER
), board_82575
},
87 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I350_FIBER
), board_82575
},
88 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I350_SERDES
), board_82575
},
89 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I350_SGMII
), board_82575
},
90 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_COPPER
), board_82575
},
91 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_FIBER
), board_82575
},
92 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_QUAD_FIBER
), board_82575
},
93 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_SERDES
), board_82575
},
94 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_SGMII
), board_82575
},
95 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_COPPER_DUAL
), board_82575
},
96 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_DH89XXCC_SGMII
), board_82575
},
97 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_DH89XXCC_SERDES
), board_82575
},
98 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_DH89XXCC_BACKPLANE
), board_82575
},
99 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_DH89XXCC_SFP
), board_82575
},
100 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576
), board_82575
},
101 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_NS
), board_82575
},
102 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_NS_SERDES
), board_82575
},
103 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_FIBER
), board_82575
},
104 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_SERDES
), board_82575
},
105 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_SERDES_QUAD
), board_82575
},
106 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_QUAD_COPPER_ET2
), board_82575
},
107 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_QUAD_COPPER
), board_82575
},
108 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82575EB_COPPER
), board_82575
},
109 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82575EB_FIBER_SERDES
), board_82575
},
110 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82575GB_QUAD_COPPER
), board_82575
},
111 /* required last entry */
115 MODULE_DEVICE_TABLE(pci
, igb_pci_tbl
);
117 static int igb_setup_all_tx_resources(struct igb_adapter
*);
118 static int igb_setup_all_rx_resources(struct igb_adapter
*);
119 static void igb_free_all_tx_resources(struct igb_adapter
*);
120 static void igb_free_all_rx_resources(struct igb_adapter
*);
121 static void igb_setup_mrqc(struct igb_adapter
*);
122 static int igb_probe(struct pci_dev
*, const struct pci_device_id
*);
123 static void igb_remove(struct pci_dev
*pdev
);
124 static int igb_sw_init(struct igb_adapter
*);
125 static int igb_open(struct net_device
*);
126 static int igb_close(struct net_device
*);
127 static void igb_configure(struct igb_adapter
*);
128 static void igb_configure_tx(struct igb_adapter
*);
129 static void igb_configure_rx(struct igb_adapter
*);
130 static void igb_clean_all_tx_rings(struct igb_adapter
*);
131 static void igb_clean_all_rx_rings(struct igb_adapter
*);
132 static void igb_clean_tx_ring(struct igb_ring
*);
133 static void igb_clean_rx_ring(struct igb_ring
*);
134 static void igb_set_rx_mode(struct net_device
*);
135 static void igb_update_phy_info(unsigned long);
136 static void igb_watchdog(unsigned long);
137 static void igb_watchdog_task(struct work_struct
*);
138 static netdev_tx_t
igb_xmit_frame(struct sk_buff
*skb
, struct net_device
*);
139 static struct rtnl_link_stats64
*igb_get_stats64(struct net_device
*dev
,
140 struct rtnl_link_stats64
*stats
);
141 static int igb_change_mtu(struct net_device
*, int);
142 static int igb_set_mac(struct net_device
*, void *);
143 static void igb_set_uta(struct igb_adapter
*adapter
);
144 static irqreturn_t
igb_intr(int irq
, void *);
145 static irqreturn_t
igb_intr_msi(int irq
, void *);
146 static irqreturn_t
igb_msix_other(int irq
, void *);
147 static irqreturn_t
igb_msix_ring(int irq
, void *);
148 #ifdef CONFIG_IGB_DCA
149 static void igb_update_dca(struct igb_q_vector
*);
150 static void igb_setup_dca(struct igb_adapter
*);
151 #endif /* CONFIG_IGB_DCA */
152 static int igb_poll(struct napi_struct
*, int);
153 static bool igb_clean_tx_irq(struct igb_q_vector
*);
154 static bool igb_clean_rx_irq(struct igb_q_vector
*, int);
155 static int igb_ioctl(struct net_device
*, struct ifreq
*, int cmd
);
156 static void igb_tx_timeout(struct net_device
*);
157 static void igb_reset_task(struct work_struct
*);
158 static void igb_vlan_mode(struct net_device
*netdev
,
159 netdev_features_t features
);
160 static int igb_vlan_rx_add_vid(struct net_device
*, __be16
, u16
);
161 static int igb_vlan_rx_kill_vid(struct net_device
*, __be16
, u16
);
162 static void igb_restore_vlan(struct igb_adapter
*);
163 static void igb_rar_set_qsel(struct igb_adapter
*, u8
*, u32
, u8
);
164 static void igb_ping_all_vfs(struct igb_adapter
*);
165 static void igb_msg_task(struct igb_adapter
*);
166 static void igb_vmm_control(struct igb_adapter
*);
167 static int igb_set_vf_mac(struct igb_adapter
*, int, unsigned char *);
168 static void igb_restore_vf_multicasts(struct igb_adapter
*adapter
);
169 static int igb_ndo_set_vf_mac(struct net_device
*netdev
, int vf
, u8
*mac
);
170 static int igb_ndo_set_vf_vlan(struct net_device
*netdev
,
171 int vf
, u16 vlan
, u8 qos
);
172 static int igb_ndo_set_vf_bw(struct net_device
*, int, int, int);
173 static int igb_ndo_set_vf_spoofchk(struct net_device
*netdev
, int vf
,
175 static int igb_ndo_get_vf_config(struct net_device
*netdev
, int vf
,
176 struct ifla_vf_info
*ivi
);
177 static void igb_check_vf_rate_limit(struct igb_adapter
*);
179 #ifdef CONFIG_PCI_IOV
180 static int igb_vf_configure(struct igb_adapter
*adapter
, int vf
);
181 static int igb_pci_enable_sriov(struct pci_dev
*dev
, int num_vfs
);
185 #ifdef CONFIG_PM_SLEEP
186 static int igb_suspend(struct device
*);
188 static int igb_resume(struct device
*);
189 static int igb_runtime_suspend(struct device
*dev
);
190 static int igb_runtime_resume(struct device
*dev
);
191 static int igb_runtime_idle(struct device
*dev
);
192 static const struct dev_pm_ops igb_pm_ops
= {
193 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend
, igb_resume
)
194 SET_RUNTIME_PM_OPS(igb_runtime_suspend
, igb_runtime_resume
,
198 static void igb_shutdown(struct pci_dev
*);
199 static int igb_pci_sriov_configure(struct pci_dev
*dev
, int num_vfs
);
200 #ifdef CONFIG_IGB_DCA
201 static int igb_notify_dca(struct notifier_block
*, unsigned long, void *);
202 static struct notifier_block dca_notifier
= {
203 .notifier_call
= igb_notify_dca
,
208 #ifdef CONFIG_NET_POLL_CONTROLLER
209 /* for netdump / net console */
210 static void igb_netpoll(struct net_device
*);
212 #ifdef CONFIG_PCI_IOV
213 static unsigned int max_vfs
;
214 module_param(max_vfs
, uint
, 0);
215 MODULE_PARM_DESC(max_vfs
, "Maximum number of virtual functions to allocate per physical function");
216 #endif /* CONFIG_PCI_IOV */
218 static pci_ers_result_t
igb_io_error_detected(struct pci_dev
*,
219 pci_channel_state_t
);
220 static pci_ers_result_t
igb_io_slot_reset(struct pci_dev
*);
221 static void igb_io_resume(struct pci_dev
*);
223 static const struct pci_error_handlers igb_err_handler
= {
224 .error_detected
= igb_io_error_detected
,
225 .slot_reset
= igb_io_slot_reset
,
226 .resume
= igb_io_resume
,
229 static void igb_init_dmac(struct igb_adapter
*adapter
, u32 pba
);
231 static struct pci_driver igb_driver
= {
232 .name
= igb_driver_name
,
233 .id_table
= igb_pci_tbl
,
235 .remove
= igb_remove
,
237 .driver
.pm
= &igb_pm_ops
,
239 .shutdown
= igb_shutdown
,
240 .sriov_configure
= igb_pci_sriov_configure
,
241 .err_handler
= &igb_err_handler
244 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
245 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
246 MODULE_LICENSE("GPL");
247 MODULE_VERSION(DRV_VERSION
);
249 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
250 static int debug
= -1;
251 module_param(debug
, int, 0);
252 MODULE_PARM_DESC(debug
, "Debug level (0=none,...,16=all)");
254 struct igb_reg_info
{
259 static const struct igb_reg_info igb_reg_info_tbl
[] = {
261 /* General Registers */
262 {E1000_CTRL
, "CTRL"},
263 {E1000_STATUS
, "STATUS"},
264 {E1000_CTRL_EXT
, "CTRL_EXT"},
266 /* Interrupt Registers */
270 {E1000_RCTL
, "RCTL"},
271 {E1000_RDLEN(0), "RDLEN"},
272 {E1000_RDH(0), "RDH"},
273 {E1000_RDT(0), "RDT"},
274 {E1000_RXDCTL(0), "RXDCTL"},
275 {E1000_RDBAL(0), "RDBAL"},
276 {E1000_RDBAH(0), "RDBAH"},
279 {E1000_TCTL
, "TCTL"},
280 {E1000_TDBAL(0), "TDBAL"},
281 {E1000_TDBAH(0), "TDBAH"},
282 {E1000_TDLEN(0), "TDLEN"},
283 {E1000_TDH(0), "TDH"},
284 {E1000_TDT(0), "TDT"},
285 {E1000_TXDCTL(0), "TXDCTL"},
286 {E1000_TDFH
, "TDFH"},
287 {E1000_TDFT
, "TDFT"},
288 {E1000_TDFHS
, "TDFHS"},
289 {E1000_TDFPC
, "TDFPC"},
291 /* List Terminator */
295 /* igb_regdump - register printout routine */
296 static void igb_regdump(struct e1000_hw
*hw
, struct igb_reg_info
*reginfo
)
302 switch (reginfo
->ofs
) {
304 for (n
= 0; n
< 4; n
++)
305 regs
[n
] = rd32(E1000_RDLEN(n
));
308 for (n
= 0; n
< 4; n
++)
309 regs
[n
] = rd32(E1000_RDH(n
));
312 for (n
= 0; n
< 4; n
++)
313 regs
[n
] = rd32(E1000_RDT(n
));
315 case E1000_RXDCTL(0):
316 for (n
= 0; n
< 4; n
++)
317 regs
[n
] = rd32(E1000_RXDCTL(n
));
320 for (n
= 0; n
< 4; n
++)
321 regs
[n
] = rd32(E1000_RDBAL(n
));
324 for (n
= 0; n
< 4; n
++)
325 regs
[n
] = rd32(E1000_RDBAH(n
));
328 for (n
= 0; n
< 4; n
++)
329 regs
[n
] = rd32(E1000_RDBAL(n
));
332 for (n
= 0; n
< 4; n
++)
333 regs
[n
] = rd32(E1000_TDBAH(n
));
336 for (n
= 0; n
< 4; n
++)
337 regs
[n
] = rd32(E1000_TDLEN(n
));
340 for (n
= 0; n
< 4; n
++)
341 regs
[n
] = rd32(E1000_TDH(n
));
344 for (n
= 0; n
< 4; n
++)
345 regs
[n
] = rd32(E1000_TDT(n
));
347 case E1000_TXDCTL(0):
348 for (n
= 0; n
< 4; n
++)
349 regs
[n
] = rd32(E1000_TXDCTL(n
));
352 pr_info("%-15s %08x\n", reginfo
->name
, rd32(reginfo
->ofs
));
356 snprintf(rname
, 16, "%s%s", reginfo
->name
, "[0-3]");
357 pr_info("%-15s %08x %08x %08x %08x\n", rname
, regs
[0], regs
[1],
361 /* igb_dump - Print registers, Tx-rings and Rx-rings */
362 static void igb_dump(struct igb_adapter
*adapter
)
364 struct net_device
*netdev
= adapter
->netdev
;
365 struct e1000_hw
*hw
= &adapter
->hw
;
366 struct igb_reg_info
*reginfo
;
367 struct igb_ring
*tx_ring
;
368 union e1000_adv_tx_desc
*tx_desc
;
369 struct my_u0
{ u64 a
; u64 b
; } *u0
;
370 struct igb_ring
*rx_ring
;
371 union e1000_adv_rx_desc
*rx_desc
;
375 if (!netif_msg_hw(adapter
))
378 /* Print netdevice Info */
380 dev_info(&adapter
->pdev
->dev
, "Net device Info\n");
381 pr_info("Device Name state trans_start last_rx\n");
382 pr_info("%-15s %016lX %016lX %016lX\n", netdev
->name
,
383 netdev
->state
, netdev
->trans_start
, netdev
->last_rx
);
386 /* Print Registers */
387 dev_info(&adapter
->pdev
->dev
, "Register Dump\n");
388 pr_info(" Register Name Value\n");
389 for (reginfo
= (struct igb_reg_info
*)igb_reg_info_tbl
;
390 reginfo
->name
; reginfo
++) {
391 igb_regdump(hw
, reginfo
);
394 /* Print TX Ring Summary */
395 if (!netdev
|| !netif_running(netdev
))
398 dev_info(&adapter
->pdev
->dev
, "TX Rings Summary\n");
399 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
400 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
401 struct igb_tx_buffer
*buffer_info
;
402 tx_ring
= adapter
->tx_ring
[n
];
403 buffer_info
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_clean
];
404 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
405 n
, tx_ring
->next_to_use
, tx_ring
->next_to_clean
,
406 (u64
)dma_unmap_addr(buffer_info
, dma
),
407 dma_unmap_len(buffer_info
, len
),
408 buffer_info
->next_to_watch
,
409 (u64
)buffer_info
->time_stamp
);
413 if (!netif_msg_tx_done(adapter
))
414 goto rx_ring_summary
;
416 dev_info(&adapter
->pdev
->dev
, "TX Rings Dump\n");
418 /* Transmit Descriptor Formats
420 * Advanced Transmit Descriptor
421 * +--------------------------------------------------------------+
422 * 0 | Buffer Address [63:0] |
423 * +--------------------------------------------------------------+
424 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
425 * +--------------------------------------------------------------+
426 * 63 46 45 40 39 38 36 35 32 31 24 15 0
429 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
430 tx_ring
= adapter
->tx_ring
[n
];
431 pr_info("------------------------------------\n");
432 pr_info("TX QUEUE INDEX = %d\n", tx_ring
->queue_index
);
433 pr_info("------------------------------------\n");
434 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] [bi->dma ] leng ntw timestamp bi->skb\n");
436 for (i
= 0; tx_ring
->desc
&& (i
< tx_ring
->count
); i
++) {
437 const char *next_desc
;
438 struct igb_tx_buffer
*buffer_info
;
439 tx_desc
= IGB_TX_DESC(tx_ring
, i
);
440 buffer_info
= &tx_ring
->tx_buffer_info
[i
];
441 u0
= (struct my_u0
*)tx_desc
;
442 if (i
== tx_ring
->next_to_use
&&
443 i
== tx_ring
->next_to_clean
)
444 next_desc
= " NTC/U";
445 else if (i
== tx_ring
->next_to_use
)
447 else if (i
== tx_ring
->next_to_clean
)
452 pr_info("T [0x%03X] %016llX %016llX %016llX %04X %p %016llX %p%s\n",
453 i
, le64_to_cpu(u0
->a
),
455 (u64
)dma_unmap_addr(buffer_info
, dma
),
456 dma_unmap_len(buffer_info
, len
),
457 buffer_info
->next_to_watch
,
458 (u64
)buffer_info
->time_stamp
,
459 buffer_info
->skb
, next_desc
);
461 if (netif_msg_pktdata(adapter
) && buffer_info
->skb
)
462 print_hex_dump(KERN_INFO
, "",
464 16, 1, buffer_info
->skb
->data
,
465 dma_unmap_len(buffer_info
, len
),
470 /* Print RX Rings Summary */
472 dev_info(&adapter
->pdev
->dev
, "RX Rings Summary\n");
473 pr_info("Queue [NTU] [NTC]\n");
474 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
475 rx_ring
= adapter
->rx_ring
[n
];
476 pr_info(" %5d %5X %5X\n",
477 n
, rx_ring
->next_to_use
, rx_ring
->next_to_clean
);
481 if (!netif_msg_rx_status(adapter
))
484 dev_info(&adapter
->pdev
->dev
, "RX Rings Dump\n");
486 /* Advanced Receive Descriptor (Read) Format
488 * +-----------------------------------------------------+
489 * 0 | Packet Buffer Address [63:1] |A0/NSE|
490 * +----------------------------------------------+------+
491 * 8 | Header Buffer Address [63:1] | DD |
492 * +-----------------------------------------------------+
495 * Advanced Receive Descriptor (Write-Back) Format
497 * 63 48 47 32 31 30 21 20 17 16 4 3 0
498 * +------------------------------------------------------+
499 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
500 * | Checksum Ident | | | | Type | Type |
501 * +------------------------------------------------------+
502 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
503 * +------------------------------------------------------+
504 * 63 48 47 32 31 20 19 0
507 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
508 rx_ring
= adapter
->rx_ring
[n
];
509 pr_info("------------------------------------\n");
510 pr_info("RX QUEUE INDEX = %d\n", rx_ring
->queue_index
);
511 pr_info("------------------------------------\n");
512 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] [bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
513 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n");
515 for (i
= 0; i
< rx_ring
->count
; i
++) {
516 const char *next_desc
;
517 struct igb_rx_buffer
*buffer_info
;
518 buffer_info
= &rx_ring
->rx_buffer_info
[i
];
519 rx_desc
= IGB_RX_DESC(rx_ring
, i
);
520 u0
= (struct my_u0
*)rx_desc
;
521 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
523 if (i
== rx_ring
->next_to_use
)
525 else if (i
== rx_ring
->next_to_clean
)
530 if (staterr
& E1000_RXD_STAT_DD
) {
531 /* Descriptor Done */
532 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n",
538 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
542 (u64
)buffer_info
->dma
,
545 if (netif_msg_pktdata(adapter
) &&
546 buffer_info
->dma
&& buffer_info
->page
) {
547 print_hex_dump(KERN_INFO
, "",
550 page_address(buffer_info
->page
) +
551 buffer_info
->page_offset
,
563 * igb_get_i2c_data - Reads the I2C SDA data bit
564 * @hw: pointer to hardware structure
565 * @i2cctl: Current value of I2CCTL register
567 * Returns the I2C data bit value
569 static int igb_get_i2c_data(void *data
)
571 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
572 struct e1000_hw
*hw
= &adapter
->hw
;
573 s32 i2cctl
= rd32(E1000_I2CPARAMS
);
575 return !!(i2cctl
& E1000_I2C_DATA_IN
);
579 * igb_set_i2c_data - Sets the I2C data bit
580 * @data: pointer to hardware structure
581 * @state: I2C data value (0 or 1) to set
583 * Sets the I2C data bit
585 static void igb_set_i2c_data(void *data
, int state
)
587 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
588 struct e1000_hw
*hw
= &adapter
->hw
;
589 s32 i2cctl
= rd32(E1000_I2CPARAMS
);
592 i2cctl
|= E1000_I2C_DATA_OUT
;
594 i2cctl
&= ~E1000_I2C_DATA_OUT
;
596 i2cctl
&= ~E1000_I2C_DATA_OE_N
;
597 i2cctl
|= E1000_I2C_CLK_OE_N
;
598 wr32(E1000_I2CPARAMS
, i2cctl
);
604 * igb_set_i2c_clk - Sets the I2C SCL clock
605 * @data: pointer to hardware structure
606 * @state: state to set clock
608 * Sets the I2C clock line to state
610 static void igb_set_i2c_clk(void *data
, int state
)
612 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
613 struct e1000_hw
*hw
= &adapter
->hw
;
614 s32 i2cctl
= rd32(E1000_I2CPARAMS
);
617 i2cctl
|= E1000_I2C_CLK_OUT
;
618 i2cctl
&= ~E1000_I2C_CLK_OE_N
;
620 i2cctl
&= ~E1000_I2C_CLK_OUT
;
621 i2cctl
&= ~E1000_I2C_CLK_OE_N
;
623 wr32(E1000_I2CPARAMS
, i2cctl
);
628 * igb_get_i2c_clk - Gets the I2C SCL clock state
629 * @data: pointer to hardware structure
631 * Gets the I2C clock state
633 static int igb_get_i2c_clk(void *data
)
635 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
636 struct e1000_hw
*hw
= &adapter
->hw
;
637 s32 i2cctl
= rd32(E1000_I2CPARAMS
);
639 return !!(i2cctl
& E1000_I2C_CLK_IN
);
642 static const struct i2c_algo_bit_data igb_i2c_algo
= {
643 .setsda
= igb_set_i2c_data
,
644 .setscl
= igb_set_i2c_clk
,
645 .getsda
= igb_get_i2c_data
,
646 .getscl
= igb_get_i2c_clk
,
652 * igb_get_hw_dev - return device
653 * @hw: pointer to hardware structure
655 * used by hardware layer to print debugging information
657 struct net_device
*igb_get_hw_dev(struct e1000_hw
*hw
)
659 struct igb_adapter
*adapter
= hw
->back
;
660 return adapter
->netdev
;
664 * igb_init_module - Driver Registration Routine
666 * igb_init_module is the first routine called when the driver is
667 * loaded. All it does is register with the PCI subsystem.
669 static int __init
igb_init_module(void)
673 pr_info("%s - version %s\n",
674 igb_driver_string
, igb_driver_version
);
675 pr_info("%s\n", igb_copyright
);
677 #ifdef CONFIG_IGB_DCA
678 dca_register_notify(&dca_notifier
);
680 ret
= pci_register_driver(&igb_driver
);
684 module_init(igb_init_module
);
687 * igb_exit_module - Driver Exit Cleanup Routine
689 * igb_exit_module is called just before the driver is removed
692 static void __exit
igb_exit_module(void)
694 #ifdef CONFIG_IGB_DCA
695 dca_unregister_notify(&dca_notifier
);
697 pci_unregister_driver(&igb_driver
);
700 module_exit(igb_exit_module
);
702 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
704 * igb_cache_ring_register - Descriptor ring to register mapping
705 * @adapter: board private structure to initialize
707 * Once we know the feature-set enabled for the device, we'll cache
708 * the register offset the descriptor ring is assigned to.
710 static void igb_cache_ring_register(struct igb_adapter
*adapter
)
713 u32 rbase_offset
= adapter
->vfs_allocated_count
;
715 switch (adapter
->hw
.mac
.type
) {
717 /* The queues are allocated for virtualization such that VF 0
718 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
719 * In order to avoid collision we start at the first free queue
720 * and continue consuming queues in the same sequence
722 if (adapter
->vfs_allocated_count
) {
723 for (; i
< adapter
->rss_queues
; i
++)
724 adapter
->rx_ring
[i
]->reg_idx
= rbase_offset
+
736 for (; i
< adapter
->num_rx_queues
; i
++)
737 adapter
->rx_ring
[i
]->reg_idx
= rbase_offset
+ i
;
738 for (; j
< adapter
->num_tx_queues
; j
++)
739 adapter
->tx_ring
[j
]->reg_idx
= rbase_offset
+ j
;
744 u32
igb_rd32(struct e1000_hw
*hw
, u32 reg
)
746 struct igb_adapter
*igb
= container_of(hw
, struct igb_adapter
, hw
);
747 u8 __iomem
*hw_addr
= ACCESS_ONCE(hw
->hw_addr
);
750 if (E1000_REMOVED(hw_addr
))
753 value
= readl(&hw_addr
[reg
]);
755 /* reads should not return all F's */
756 if (!(~value
) && (!reg
|| !(~readl(hw_addr
)))) {
757 struct net_device
*netdev
= igb
->netdev
;
759 netif_device_detach(netdev
);
760 netdev_err(netdev
, "PCIe link lost, device now detached\n");
767 * igb_write_ivar - configure ivar for given MSI-X vector
768 * @hw: pointer to the HW structure
769 * @msix_vector: vector number we are allocating to a given ring
770 * @index: row index of IVAR register to write within IVAR table
771 * @offset: column offset of in IVAR, should be multiple of 8
773 * This function is intended to handle the writing of the IVAR register
774 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
775 * each containing an cause allocation for an Rx and Tx ring, and a
776 * variable number of rows depending on the number of queues supported.
778 static void igb_write_ivar(struct e1000_hw
*hw
, int msix_vector
,
779 int index
, int offset
)
781 u32 ivar
= array_rd32(E1000_IVAR0
, index
);
783 /* clear any bits that are currently set */
784 ivar
&= ~((u32
)0xFF << offset
);
786 /* write vector and valid bit */
787 ivar
|= (msix_vector
| E1000_IVAR_VALID
) << offset
;
789 array_wr32(E1000_IVAR0
, index
, ivar
);
792 #define IGB_N0_QUEUE -1
793 static void igb_assign_vector(struct igb_q_vector
*q_vector
, int msix_vector
)
795 struct igb_adapter
*adapter
= q_vector
->adapter
;
796 struct e1000_hw
*hw
= &adapter
->hw
;
797 int rx_queue
= IGB_N0_QUEUE
;
798 int tx_queue
= IGB_N0_QUEUE
;
801 if (q_vector
->rx
.ring
)
802 rx_queue
= q_vector
->rx
.ring
->reg_idx
;
803 if (q_vector
->tx
.ring
)
804 tx_queue
= q_vector
->tx
.ring
->reg_idx
;
806 switch (hw
->mac
.type
) {
808 /* The 82575 assigns vectors using a bitmask, which matches the
809 * bitmask for the EICR/EIMS/EIMC registers. To assign one
810 * or more queues to a vector, we write the appropriate bits
811 * into the MSIXBM register for that vector.
813 if (rx_queue
> IGB_N0_QUEUE
)
814 msixbm
= E1000_EICR_RX_QUEUE0
<< rx_queue
;
815 if (tx_queue
> IGB_N0_QUEUE
)
816 msixbm
|= E1000_EICR_TX_QUEUE0
<< tx_queue
;
817 if (!(adapter
->flags
& IGB_FLAG_HAS_MSIX
) && msix_vector
== 0)
818 msixbm
|= E1000_EIMS_OTHER
;
819 array_wr32(E1000_MSIXBM(0), msix_vector
, msixbm
);
820 q_vector
->eims_value
= msixbm
;
823 /* 82576 uses a table that essentially consists of 2 columns
824 * with 8 rows. The ordering is column-major so we use the
825 * lower 3 bits as the row index, and the 4th bit as the
828 if (rx_queue
> IGB_N0_QUEUE
)
829 igb_write_ivar(hw
, msix_vector
,
831 (rx_queue
& 0x8) << 1);
832 if (tx_queue
> IGB_N0_QUEUE
)
833 igb_write_ivar(hw
, msix_vector
,
835 ((tx_queue
& 0x8) << 1) + 8);
836 q_vector
->eims_value
= 1 << msix_vector
;
843 /* On 82580 and newer adapters the scheme is similar to 82576
844 * however instead of ordering column-major we have things
845 * ordered row-major. So we traverse the table by using
846 * bit 0 as the column offset, and the remaining bits as the
849 if (rx_queue
> IGB_N0_QUEUE
)
850 igb_write_ivar(hw
, msix_vector
,
852 (rx_queue
& 0x1) << 4);
853 if (tx_queue
> IGB_N0_QUEUE
)
854 igb_write_ivar(hw
, msix_vector
,
856 ((tx_queue
& 0x1) << 4) + 8);
857 q_vector
->eims_value
= 1 << msix_vector
;
864 /* add q_vector eims value to global eims_enable_mask */
865 adapter
->eims_enable_mask
|= q_vector
->eims_value
;
867 /* configure q_vector to set itr on first interrupt */
868 q_vector
->set_itr
= 1;
872 * igb_configure_msix - Configure MSI-X hardware
873 * @adapter: board private structure to initialize
875 * igb_configure_msix sets up the hardware to properly
876 * generate MSI-X interrupts.
878 static void igb_configure_msix(struct igb_adapter
*adapter
)
882 struct e1000_hw
*hw
= &adapter
->hw
;
884 adapter
->eims_enable_mask
= 0;
886 /* set vector for other causes, i.e. link changes */
887 switch (hw
->mac
.type
) {
889 tmp
= rd32(E1000_CTRL_EXT
);
890 /* enable MSI-X PBA support*/
891 tmp
|= E1000_CTRL_EXT_PBA_CLR
;
893 /* Auto-Mask interrupts upon ICR read. */
894 tmp
|= E1000_CTRL_EXT_EIAME
;
895 tmp
|= E1000_CTRL_EXT_IRCA
;
897 wr32(E1000_CTRL_EXT
, tmp
);
899 /* enable msix_other interrupt */
900 array_wr32(E1000_MSIXBM(0), vector
++, E1000_EIMS_OTHER
);
901 adapter
->eims_other
= E1000_EIMS_OTHER
;
911 /* Turn on MSI-X capability first, or our settings
912 * won't stick. And it will take days to debug.
914 wr32(E1000_GPIE
, E1000_GPIE_MSIX_MODE
|
915 E1000_GPIE_PBA
| E1000_GPIE_EIAME
|
918 /* enable msix_other interrupt */
919 adapter
->eims_other
= 1 << vector
;
920 tmp
= (vector
++ | E1000_IVAR_VALID
) << 8;
922 wr32(E1000_IVAR_MISC
, tmp
);
925 /* do nothing, since nothing else supports MSI-X */
927 } /* switch (hw->mac.type) */
929 adapter
->eims_enable_mask
|= adapter
->eims_other
;
931 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
932 igb_assign_vector(adapter
->q_vector
[i
], vector
++);
938 * igb_request_msix - Initialize MSI-X interrupts
939 * @adapter: board private structure to initialize
941 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
944 static int igb_request_msix(struct igb_adapter
*adapter
)
946 struct net_device
*netdev
= adapter
->netdev
;
947 struct e1000_hw
*hw
= &adapter
->hw
;
948 int i
, err
= 0, vector
= 0, free_vector
= 0;
950 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
951 igb_msix_other
, 0, netdev
->name
, adapter
);
955 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
956 struct igb_q_vector
*q_vector
= adapter
->q_vector
[i
];
960 q_vector
->itr_register
= hw
->hw_addr
+ E1000_EITR(vector
);
962 if (q_vector
->rx
.ring
&& q_vector
->tx
.ring
)
963 sprintf(q_vector
->name
, "%s-TxRx-%u", netdev
->name
,
964 q_vector
->rx
.ring
->queue_index
);
965 else if (q_vector
->tx
.ring
)
966 sprintf(q_vector
->name
, "%s-tx-%u", netdev
->name
,
967 q_vector
->tx
.ring
->queue_index
);
968 else if (q_vector
->rx
.ring
)
969 sprintf(q_vector
->name
, "%s-rx-%u", netdev
->name
,
970 q_vector
->rx
.ring
->queue_index
);
972 sprintf(q_vector
->name
, "%s-unused", netdev
->name
);
974 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
975 igb_msix_ring
, 0, q_vector
->name
,
981 igb_configure_msix(adapter
);
985 /* free already assigned IRQs */
986 free_irq(adapter
->msix_entries
[free_vector
++].vector
, adapter
);
989 for (i
= 0; i
< vector
; i
++) {
990 free_irq(adapter
->msix_entries
[free_vector
++].vector
,
991 adapter
->q_vector
[i
]);
998 * igb_free_q_vector - Free memory allocated for specific interrupt vector
999 * @adapter: board private structure to initialize
1000 * @v_idx: Index of vector to be freed
1002 * This function frees the memory allocated to the q_vector.
1004 static void igb_free_q_vector(struct igb_adapter
*adapter
, int v_idx
)
1006 struct igb_q_vector
*q_vector
= adapter
->q_vector
[v_idx
];
1008 adapter
->q_vector
[v_idx
] = NULL
;
1010 /* igb_get_stats64() might access the rings on this vector,
1011 * we must wait a grace period before freeing it.
1013 kfree_rcu(q_vector
, rcu
);
1017 * igb_reset_q_vector - Reset config for interrupt vector
1018 * @adapter: board private structure to initialize
1019 * @v_idx: Index of vector to be reset
1021 * If NAPI is enabled it will delete any references to the
1022 * NAPI struct. This is preparation for igb_free_q_vector.
1024 static void igb_reset_q_vector(struct igb_adapter
*adapter
, int v_idx
)
1026 struct igb_q_vector
*q_vector
= adapter
->q_vector
[v_idx
];
1028 /* Coming from igb_set_interrupt_capability, the vectors are not yet
1029 * allocated. So, q_vector is NULL so we should stop here.
1034 if (q_vector
->tx
.ring
)
1035 adapter
->tx_ring
[q_vector
->tx
.ring
->queue_index
] = NULL
;
1037 if (q_vector
->rx
.ring
)
1038 adapter
->tx_ring
[q_vector
->rx
.ring
->queue_index
] = NULL
;
1040 netif_napi_del(&q_vector
->napi
);
1044 static void igb_reset_interrupt_capability(struct igb_adapter
*adapter
)
1046 int v_idx
= adapter
->num_q_vectors
;
1048 if (adapter
->flags
& IGB_FLAG_HAS_MSIX
)
1049 pci_disable_msix(adapter
->pdev
);
1050 else if (adapter
->flags
& IGB_FLAG_HAS_MSI
)
1051 pci_disable_msi(adapter
->pdev
);
1054 igb_reset_q_vector(adapter
, v_idx
);
1058 * igb_free_q_vectors - Free memory allocated for interrupt vectors
1059 * @adapter: board private structure to initialize
1061 * This function frees the memory allocated to the q_vectors. In addition if
1062 * NAPI is enabled it will delete any references to the NAPI struct prior
1063 * to freeing the q_vector.
1065 static void igb_free_q_vectors(struct igb_adapter
*adapter
)
1067 int v_idx
= adapter
->num_q_vectors
;
1069 adapter
->num_tx_queues
= 0;
1070 adapter
->num_rx_queues
= 0;
1071 adapter
->num_q_vectors
= 0;
1074 igb_reset_q_vector(adapter
, v_idx
);
1075 igb_free_q_vector(adapter
, v_idx
);
1080 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1081 * @adapter: board private structure to initialize
1083 * This function resets the device so that it has 0 Rx queues, Tx queues, and
1084 * MSI-X interrupts allocated.
1086 static void igb_clear_interrupt_scheme(struct igb_adapter
*adapter
)
1088 igb_free_q_vectors(adapter
);
1089 igb_reset_interrupt_capability(adapter
);
1093 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1094 * @adapter: board private structure to initialize
1095 * @msix: boolean value of MSIX capability
1097 * Attempt to configure interrupts using the best available
1098 * capabilities of the hardware and kernel.
1100 static void igb_set_interrupt_capability(struct igb_adapter
*adapter
, bool msix
)
1107 adapter
->flags
|= IGB_FLAG_HAS_MSIX
;
1109 /* Number of supported queues. */
1110 adapter
->num_rx_queues
= adapter
->rss_queues
;
1111 if (adapter
->vfs_allocated_count
)
1112 adapter
->num_tx_queues
= 1;
1114 adapter
->num_tx_queues
= adapter
->rss_queues
;
1116 /* start with one vector for every Rx queue */
1117 numvecs
= adapter
->num_rx_queues
;
1119 /* if Tx handler is separate add 1 for every Tx queue */
1120 if (!(adapter
->flags
& IGB_FLAG_QUEUE_PAIRS
))
1121 numvecs
+= adapter
->num_tx_queues
;
1123 /* store the number of vectors reserved for queues */
1124 adapter
->num_q_vectors
= numvecs
;
1126 /* add 1 vector for link status interrupts */
1128 for (i
= 0; i
< numvecs
; i
++)
1129 adapter
->msix_entries
[i
].entry
= i
;
1131 err
= pci_enable_msix_range(adapter
->pdev
,
1132 adapter
->msix_entries
,
1138 igb_reset_interrupt_capability(adapter
);
1140 /* If we can't do MSI-X, try MSI */
1142 adapter
->flags
&= ~IGB_FLAG_HAS_MSIX
;
1143 #ifdef CONFIG_PCI_IOV
1144 /* disable SR-IOV for non MSI-X configurations */
1145 if (adapter
->vf_data
) {
1146 struct e1000_hw
*hw
= &adapter
->hw
;
1147 /* disable iov and allow time for transactions to clear */
1148 pci_disable_sriov(adapter
->pdev
);
1151 kfree(adapter
->vf_data
);
1152 adapter
->vf_data
= NULL
;
1153 wr32(E1000_IOVCTL
, E1000_IOVCTL_REUSE_VFQ
);
1156 dev_info(&adapter
->pdev
->dev
, "IOV Disabled\n");
1159 adapter
->vfs_allocated_count
= 0;
1160 adapter
->rss_queues
= 1;
1161 adapter
->flags
|= IGB_FLAG_QUEUE_PAIRS
;
1162 adapter
->num_rx_queues
= 1;
1163 adapter
->num_tx_queues
= 1;
1164 adapter
->num_q_vectors
= 1;
1165 if (!pci_enable_msi(adapter
->pdev
))
1166 adapter
->flags
|= IGB_FLAG_HAS_MSI
;
1169 static void igb_add_ring(struct igb_ring
*ring
,
1170 struct igb_ring_container
*head
)
1177 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1178 * @adapter: board private structure to initialize
1179 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1180 * @v_idx: index of vector in adapter struct
1181 * @txr_count: total number of Tx rings to allocate
1182 * @txr_idx: index of first Tx ring to allocate
1183 * @rxr_count: total number of Rx rings to allocate
1184 * @rxr_idx: index of first Rx ring to allocate
1186 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1188 static int igb_alloc_q_vector(struct igb_adapter
*adapter
,
1189 int v_count
, int v_idx
,
1190 int txr_count
, int txr_idx
,
1191 int rxr_count
, int rxr_idx
)
1193 struct igb_q_vector
*q_vector
;
1194 struct igb_ring
*ring
;
1195 int ring_count
, size
;
1197 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1198 if (txr_count
> 1 || rxr_count
> 1)
1201 ring_count
= txr_count
+ rxr_count
;
1202 size
= sizeof(struct igb_q_vector
) +
1203 (sizeof(struct igb_ring
) * ring_count
);
1205 /* allocate q_vector and rings */
1206 q_vector
= adapter
->q_vector
[v_idx
];
1208 q_vector
= kzalloc(size
, GFP_KERNEL
);
1212 /* initialize NAPI */
1213 netif_napi_add(adapter
->netdev
, &q_vector
->napi
,
1216 /* tie q_vector and adapter together */
1217 adapter
->q_vector
[v_idx
] = q_vector
;
1218 q_vector
->adapter
= adapter
;
1220 /* initialize work limits */
1221 q_vector
->tx
.work_limit
= adapter
->tx_work_limit
;
1223 /* initialize ITR configuration */
1224 q_vector
->itr_register
= adapter
->hw
.hw_addr
+ E1000_EITR(0);
1225 q_vector
->itr_val
= IGB_START_ITR
;
1227 /* initialize pointer to rings */
1228 ring
= q_vector
->ring
;
1232 /* rx or rx/tx vector */
1233 if (!adapter
->rx_itr_setting
|| adapter
->rx_itr_setting
> 3)
1234 q_vector
->itr_val
= adapter
->rx_itr_setting
;
1236 /* tx only vector */
1237 if (!adapter
->tx_itr_setting
|| adapter
->tx_itr_setting
> 3)
1238 q_vector
->itr_val
= adapter
->tx_itr_setting
;
1242 /* assign generic ring traits */
1243 ring
->dev
= &adapter
->pdev
->dev
;
1244 ring
->netdev
= adapter
->netdev
;
1246 /* configure backlink on ring */
1247 ring
->q_vector
= q_vector
;
1249 /* update q_vector Tx values */
1250 igb_add_ring(ring
, &q_vector
->tx
);
1252 /* For 82575, context index must be unique per ring. */
1253 if (adapter
->hw
.mac
.type
== e1000_82575
)
1254 set_bit(IGB_RING_FLAG_TX_CTX_IDX
, &ring
->flags
);
1256 /* apply Tx specific ring traits */
1257 ring
->count
= adapter
->tx_ring_count
;
1258 ring
->queue_index
= txr_idx
;
1260 u64_stats_init(&ring
->tx_syncp
);
1261 u64_stats_init(&ring
->tx_syncp2
);
1263 /* assign ring to adapter */
1264 adapter
->tx_ring
[txr_idx
] = ring
;
1266 /* push pointer to next ring */
1271 /* assign generic ring traits */
1272 ring
->dev
= &adapter
->pdev
->dev
;
1273 ring
->netdev
= adapter
->netdev
;
1275 /* configure backlink on ring */
1276 ring
->q_vector
= q_vector
;
1278 /* update q_vector Rx values */
1279 igb_add_ring(ring
, &q_vector
->rx
);
1281 /* set flag indicating ring supports SCTP checksum offload */
1282 if (adapter
->hw
.mac
.type
>= e1000_82576
)
1283 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM
, &ring
->flags
);
1285 /* On i350, i354, i210, and i211, loopback VLAN packets
1286 * have the tag byte-swapped.
1288 if (adapter
->hw
.mac
.type
>= e1000_i350
)
1289 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP
, &ring
->flags
);
1291 /* apply Rx specific ring traits */
1292 ring
->count
= adapter
->rx_ring_count
;
1293 ring
->queue_index
= rxr_idx
;
1295 u64_stats_init(&ring
->rx_syncp
);
1297 /* assign ring to adapter */
1298 adapter
->rx_ring
[rxr_idx
] = ring
;
1306 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1307 * @adapter: board private structure to initialize
1309 * We allocate one q_vector per queue interrupt. If allocation fails we
1312 static int igb_alloc_q_vectors(struct igb_adapter
*adapter
)
1314 int q_vectors
= adapter
->num_q_vectors
;
1315 int rxr_remaining
= adapter
->num_rx_queues
;
1316 int txr_remaining
= adapter
->num_tx_queues
;
1317 int rxr_idx
= 0, txr_idx
= 0, v_idx
= 0;
1320 if (q_vectors
>= (rxr_remaining
+ txr_remaining
)) {
1321 for (; rxr_remaining
; v_idx
++) {
1322 err
= igb_alloc_q_vector(adapter
, q_vectors
, v_idx
,
1328 /* update counts and index */
1334 for (; v_idx
< q_vectors
; v_idx
++) {
1335 int rqpv
= DIV_ROUND_UP(rxr_remaining
, q_vectors
- v_idx
);
1336 int tqpv
= DIV_ROUND_UP(txr_remaining
, q_vectors
- v_idx
);
1338 err
= igb_alloc_q_vector(adapter
, q_vectors
, v_idx
,
1339 tqpv
, txr_idx
, rqpv
, rxr_idx
);
1344 /* update counts and index */
1345 rxr_remaining
-= rqpv
;
1346 txr_remaining
-= tqpv
;
1354 adapter
->num_tx_queues
= 0;
1355 adapter
->num_rx_queues
= 0;
1356 adapter
->num_q_vectors
= 0;
1359 igb_free_q_vector(adapter
, v_idx
);
1365 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1366 * @adapter: board private structure to initialize
1367 * @msix: boolean value of MSIX capability
1369 * This function initializes the interrupts and allocates all of the queues.
1371 static int igb_init_interrupt_scheme(struct igb_adapter
*adapter
, bool msix
)
1373 struct pci_dev
*pdev
= adapter
->pdev
;
1376 igb_set_interrupt_capability(adapter
, msix
);
1378 err
= igb_alloc_q_vectors(adapter
);
1380 dev_err(&pdev
->dev
, "Unable to allocate memory for vectors\n");
1381 goto err_alloc_q_vectors
;
1384 igb_cache_ring_register(adapter
);
1388 err_alloc_q_vectors
:
1389 igb_reset_interrupt_capability(adapter
);
1394 * igb_request_irq - initialize interrupts
1395 * @adapter: board private structure to initialize
1397 * Attempts to configure interrupts using the best available
1398 * capabilities of the hardware and kernel.
1400 static int igb_request_irq(struct igb_adapter
*adapter
)
1402 struct net_device
*netdev
= adapter
->netdev
;
1403 struct pci_dev
*pdev
= adapter
->pdev
;
1406 if (adapter
->flags
& IGB_FLAG_HAS_MSIX
) {
1407 err
= igb_request_msix(adapter
);
1410 /* fall back to MSI */
1411 igb_free_all_tx_resources(adapter
);
1412 igb_free_all_rx_resources(adapter
);
1414 igb_clear_interrupt_scheme(adapter
);
1415 err
= igb_init_interrupt_scheme(adapter
, false);
1419 igb_setup_all_tx_resources(adapter
);
1420 igb_setup_all_rx_resources(adapter
);
1421 igb_configure(adapter
);
1424 igb_assign_vector(adapter
->q_vector
[0], 0);
1426 if (adapter
->flags
& IGB_FLAG_HAS_MSI
) {
1427 err
= request_irq(pdev
->irq
, igb_intr_msi
, 0,
1428 netdev
->name
, adapter
);
1432 /* fall back to legacy interrupts */
1433 igb_reset_interrupt_capability(adapter
);
1434 adapter
->flags
&= ~IGB_FLAG_HAS_MSI
;
1437 err
= request_irq(pdev
->irq
, igb_intr
, IRQF_SHARED
,
1438 netdev
->name
, adapter
);
1441 dev_err(&pdev
->dev
, "Error %d getting interrupt\n",
1448 static void igb_free_irq(struct igb_adapter
*adapter
)
1450 if (adapter
->flags
& IGB_FLAG_HAS_MSIX
) {
1453 free_irq(adapter
->msix_entries
[vector
++].vector
, adapter
);
1455 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1456 free_irq(adapter
->msix_entries
[vector
++].vector
,
1457 adapter
->q_vector
[i
]);
1459 free_irq(adapter
->pdev
->irq
, adapter
);
1464 * igb_irq_disable - Mask off interrupt generation on the NIC
1465 * @adapter: board private structure
1467 static void igb_irq_disable(struct igb_adapter
*adapter
)
1469 struct e1000_hw
*hw
= &adapter
->hw
;
1471 /* we need to be careful when disabling interrupts. The VFs are also
1472 * mapped into these registers and so clearing the bits can cause
1473 * issues on the VF drivers so we only need to clear what we set
1475 if (adapter
->flags
& IGB_FLAG_HAS_MSIX
) {
1476 u32 regval
= rd32(E1000_EIAM
);
1478 wr32(E1000_EIAM
, regval
& ~adapter
->eims_enable_mask
);
1479 wr32(E1000_EIMC
, adapter
->eims_enable_mask
);
1480 regval
= rd32(E1000_EIAC
);
1481 wr32(E1000_EIAC
, regval
& ~adapter
->eims_enable_mask
);
1485 wr32(E1000_IMC
, ~0);
1487 if (adapter
->flags
& IGB_FLAG_HAS_MSIX
) {
1490 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1491 synchronize_irq(adapter
->msix_entries
[i
].vector
);
1493 synchronize_irq(adapter
->pdev
->irq
);
1498 * igb_irq_enable - Enable default interrupt generation settings
1499 * @adapter: board private structure
1501 static void igb_irq_enable(struct igb_adapter
*adapter
)
1503 struct e1000_hw
*hw
= &adapter
->hw
;
1505 if (adapter
->flags
& IGB_FLAG_HAS_MSIX
) {
1506 u32 ims
= E1000_IMS_LSC
| E1000_IMS_DOUTSYNC
| E1000_IMS_DRSTA
;
1507 u32 regval
= rd32(E1000_EIAC
);
1509 wr32(E1000_EIAC
, regval
| adapter
->eims_enable_mask
);
1510 regval
= rd32(E1000_EIAM
);
1511 wr32(E1000_EIAM
, regval
| adapter
->eims_enable_mask
);
1512 wr32(E1000_EIMS
, adapter
->eims_enable_mask
);
1513 if (adapter
->vfs_allocated_count
) {
1514 wr32(E1000_MBVFIMR
, 0xFF);
1515 ims
|= E1000_IMS_VMMB
;
1517 wr32(E1000_IMS
, ims
);
1519 wr32(E1000_IMS
, IMS_ENABLE_MASK
|
1521 wr32(E1000_IAM
, IMS_ENABLE_MASK
|
1526 static void igb_update_mng_vlan(struct igb_adapter
*adapter
)
1528 struct e1000_hw
*hw
= &adapter
->hw
;
1529 u16 vid
= adapter
->hw
.mng_cookie
.vlan_id
;
1530 u16 old_vid
= adapter
->mng_vlan_id
;
1532 if (hw
->mng_cookie
.status
& E1000_MNG_DHCP_COOKIE_STATUS_VLAN
) {
1533 /* add VID to filter table */
1534 igb_vfta_set(hw
, vid
, true);
1535 adapter
->mng_vlan_id
= vid
;
1537 adapter
->mng_vlan_id
= IGB_MNG_VLAN_NONE
;
1540 if ((old_vid
!= (u16
)IGB_MNG_VLAN_NONE
) &&
1542 !test_bit(old_vid
, adapter
->active_vlans
)) {
1543 /* remove VID from filter table */
1544 igb_vfta_set(hw
, old_vid
, false);
1549 * igb_release_hw_control - release control of the h/w to f/w
1550 * @adapter: address of board private structure
1552 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1553 * For ASF and Pass Through versions of f/w this means that the
1554 * driver is no longer loaded.
1556 static void igb_release_hw_control(struct igb_adapter
*adapter
)
1558 struct e1000_hw
*hw
= &adapter
->hw
;
1561 /* Let firmware take over control of h/w */
1562 ctrl_ext
= rd32(E1000_CTRL_EXT
);
1563 wr32(E1000_CTRL_EXT
,
1564 ctrl_ext
& ~E1000_CTRL_EXT_DRV_LOAD
);
1568 * igb_get_hw_control - get control of the h/w from f/w
1569 * @adapter: address of board private structure
1571 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1572 * For ASF and Pass Through versions of f/w this means that
1573 * the driver is loaded.
1575 static void igb_get_hw_control(struct igb_adapter
*adapter
)
1577 struct e1000_hw
*hw
= &adapter
->hw
;
1580 /* Let firmware know the driver has taken over */
1581 ctrl_ext
= rd32(E1000_CTRL_EXT
);
1582 wr32(E1000_CTRL_EXT
,
1583 ctrl_ext
| E1000_CTRL_EXT_DRV_LOAD
);
1587 * igb_configure - configure the hardware for RX and TX
1588 * @adapter: private board structure
1590 static void igb_configure(struct igb_adapter
*adapter
)
1592 struct net_device
*netdev
= adapter
->netdev
;
1595 igb_get_hw_control(adapter
);
1596 igb_set_rx_mode(netdev
);
1598 igb_restore_vlan(adapter
);
1600 igb_setup_tctl(adapter
);
1601 igb_setup_mrqc(adapter
);
1602 igb_setup_rctl(adapter
);
1604 igb_configure_tx(adapter
);
1605 igb_configure_rx(adapter
);
1607 igb_rx_fifo_flush_82575(&adapter
->hw
);
1609 /* call igb_desc_unused which always leaves
1610 * at least 1 descriptor unused to make sure
1611 * next_to_use != next_to_clean
1613 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
1614 struct igb_ring
*ring
= adapter
->rx_ring
[i
];
1615 igb_alloc_rx_buffers(ring
, igb_desc_unused(ring
));
1620 * igb_power_up_link - Power up the phy/serdes link
1621 * @adapter: address of board private structure
1623 void igb_power_up_link(struct igb_adapter
*adapter
)
1625 igb_reset_phy(&adapter
->hw
);
1627 if (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
)
1628 igb_power_up_phy_copper(&adapter
->hw
);
1630 igb_power_up_serdes_link_82575(&adapter
->hw
);
1632 igb_setup_link(&adapter
->hw
);
1636 * igb_power_down_link - Power down the phy/serdes link
1637 * @adapter: address of board private structure
1639 static void igb_power_down_link(struct igb_adapter
*adapter
)
1641 if (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
)
1642 igb_power_down_phy_copper_82575(&adapter
->hw
);
1644 igb_shutdown_serdes_link_82575(&adapter
->hw
);
1648 * Detect and switch function for Media Auto Sense
1649 * @adapter: address of the board private structure
1651 static void igb_check_swap_media(struct igb_adapter
*adapter
)
1653 struct e1000_hw
*hw
= &adapter
->hw
;
1654 u32 ctrl_ext
, connsw
;
1655 bool swap_now
= false;
1657 ctrl_ext
= rd32(E1000_CTRL_EXT
);
1658 connsw
= rd32(E1000_CONNSW
);
1660 /* need to live swap if current media is copper and we have fiber/serdes
1664 if ((hw
->phy
.media_type
== e1000_media_type_copper
) &&
1665 (!(connsw
& E1000_CONNSW_AUTOSENSE_EN
))) {
1667 } else if (!(connsw
& E1000_CONNSW_SERDESD
)) {
1668 /* copper signal takes time to appear */
1669 if (adapter
->copper_tries
< 4) {
1670 adapter
->copper_tries
++;
1671 connsw
|= E1000_CONNSW_AUTOSENSE_CONF
;
1672 wr32(E1000_CONNSW
, connsw
);
1675 adapter
->copper_tries
= 0;
1676 if ((connsw
& E1000_CONNSW_PHYSD
) &&
1677 (!(connsw
& E1000_CONNSW_PHY_PDN
))) {
1679 connsw
&= ~E1000_CONNSW_AUTOSENSE_CONF
;
1680 wr32(E1000_CONNSW
, connsw
);
1688 switch (hw
->phy
.media_type
) {
1689 case e1000_media_type_copper
:
1690 netdev_info(adapter
->netdev
,
1691 "MAS: changing media to fiber/serdes\n");
1693 E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES
;
1694 adapter
->flags
|= IGB_FLAG_MEDIA_RESET
;
1695 adapter
->copper_tries
= 0;
1697 case e1000_media_type_internal_serdes
:
1698 case e1000_media_type_fiber
:
1699 netdev_info(adapter
->netdev
,
1700 "MAS: changing media to copper\n");
1702 ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES
;
1703 adapter
->flags
|= IGB_FLAG_MEDIA_RESET
;
1706 /* shouldn't get here during regular operation */
1707 netdev_err(adapter
->netdev
,
1708 "AMS: Invalid media type found, returning\n");
1711 wr32(E1000_CTRL_EXT
, ctrl_ext
);
1715 * igb_up - Open the interface and prepare it to handle traffic
1716 * @adapter: board private structure
1718 int igb_up(struct igb_adapter
*adapter
)
1720 struct e1000_hw
*hw
= &adapter
->hw
;
1723 /* hardware has been reset, we need to reload some things */
1724 igb_configure(adapter
);
1726 clear_bit(__IGB_DOWN
, &adapter
->state
);
1728 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1729 napi_enable(&(adapter
->q_vector
[i
]->napi
));
1731 if (adapter
->flags
& IGB_FLAG_HAS_MSIX
)
1732 igb_configure_msix(adapter
);
1734 igb_assign_vector(adapter
->q_vector
[0], 0);
1736 /* Clear any pending interrupts. */
1738 igb_irq_enable(adapter
);
1740 /* notify VFs that reset has been completed */
1741 if (adapter
->vfs_allocated_count
) {
1742 u32 reg_data
= rd32(E1000_CTRL_EXT
);
1744 reg_data
|= E1000_CTRL_EXT_PFRSTD
;
1745 wr32(E1000_CTRL_EXT
, reg_data
);
1748 netif_tx_start_all_queues(adapter
->netdev
);
1750 /* start the watchdog. */
1751 hw
->mac
.get_link_status
= 1;
1752 schedule_work(&adapter
->watchdog_task
);
1754 if ((adapter
->flags
& IGB_FLAG_EEE
) &&
1755 (!hw
->dev_spec
._82575
.eee_disable
))
1756 adapter
->eee_advert
= MDIO_EEE_100TX
| MDIO_EEE_1000T
;
1761 void igb_down(struct igb_adapter
*adapter
)
1763 struct net_device
*netdev
= adapter
->netdev
;
1764 struct e1000_hw
*hw
= &adapter
->hw
;
1768 /* signal that we're down so the interrupt handler does not
1769 * reschedule our watchdog timer
1771 set_bit(__IGB_DOWN
, &adapter
->state
);
1773 /* disable receives in the hardware */
1774 rctl
= rd32(E1000_RCTL
);
1775 wr32(E1000_RCTL
, rctl
& ~E1000_RCTL_EN
);
1776 /* flush and sleep below */
1778 netif_tx_stop_all_queues(netdev
);
1780 /* disable transmits in the hardware */
1781 tctl
= rd32(E1000_TCTL
);
1782 tctl
&= ~E1000_TCTL_EN
;
1783 wr32(E1000_TCTL
, tctl
);
1784 /* flush both disables and wait for them to finish */
1786 usleep_range(10000, 11000);
1788 igb_irq_disable(adapter
);
1790 adapter
->flags
&= ~IGB_FLAG_NEED_LINK_UPDATE
;
1792 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
1793 napi_synchronize(&(adapter
->q_vector
[i
]->napi
));
1794 napi_disable(&(adapter
->q_vector
[i
]->napi
));
1798 del_timer_sync(&adapter
->watchdog_timer
);
1799 del_timer_sync(&adapter
->phy_info_timer
);
1801 netif_carrier_off(netdev
);
1803 /* record the stats before reset*/
1804 spin_lock(&adapter
->stats64_lock
);
1805 igb_update_stats(adapter
, &adapter
->stats64
);
1806 spin_unlock(&adapter
->stats64_lock
);
1808 adapter
->link_speed
= 0;
1809 adapter
->link_duplex
= 0;
1811 if (!pci_channel_offline(adapter
->pdev
))
1813 igb_clean_all_tx_rings(adapter
);
1814 igb_clean_all_rx_rings(adapter
);
1815 #ifdef CONFIG_IGB_DCA
1817 /* since we reset the hardware DCA settings were cleared */
1818 igb_setup_dca(adapter
);
1822 void igb_reinit_locked(struct igb_adapter
*adapter
)
1824 WARN_ON(in_interrupt());
1825 while (test_and_set_bit(__IGB_RESETTING
, &adapter
->state
))
1826 usleep_range(1000, 2000);
1829 clear_bit(__IGB_RESETTING
, &adapter
->state
);
1832 /** igb_enable_mas - Media Autosense re-enable after swap
1834 * @adapter: adapter struct
1836 static s32
igb_enable_mas(struct igb_adapter
*adapter
)
1838 struct e1000_hw
*hw
= &adapter
->hw
;
1842 connsw
= rd32(E1000_CONNSW
);
1843 if (!(hw
->phy
.media_type
== e1000_media_type_copper
))
1846 /* configure for SerDes media detect */
1847 if (!(connsw
& E1000_CONNSW_SERDESD
)) {
1848 connsw
|= E1000_CONNSW_ENRGSRC
;
1849 connsw
|= E1000_CONNSW_AUTOSENSE_EN
;
1850 wr32(E1000_CONNSW
, connsw
);
1852 } else if (connsw
& E1000_CONNSW_SERDESD
) {
1853 /* already SerDes, no need to enable anything */
1856 netdev_info(adapter
->netdev
,
1857 "MAS: Unable to configure feature, disabling..\n");
1858 adapter
->flags
&= ~IGB_FLAG_MAS_ENABLE
;
1863 void igb_reset(struct igb_adapter
*adapter
)
1865 struct pci_dev
*pdev
= adapter
->pdev
;
1866 struct e1000_hw
*hw
= &adapter
->hw
;
1867 struct e1000_mac_info
*mac
= &hw
->mac
;
1868 struct e1000_fc_info
*fc
= &hw
->fc
;
1869 u32 pba
= 0, tx_space
, min_tx_space
, min_rx_space
, hwm
;
1871 /* Repartition Pba for greater than 9k mtu
1872 * To take effect CTRL.RST is required.
1874 switch (mac
->type
) {
1878 pba
= rd32(E1000_RXPBS
);
1879 pba
= igb_rxpbs_adjust_82580(pba
);
1882 pba
= rd32(E1000_RXPBS
);
1883 pba
&= E1000_RXPBS_SIZE_MASK_82576
;
1889 pba
= E1000_PBA_34K
;
1893 if ((adapter
->max_frame_size
> ETH_FRAME_LEN
+ ETH_FCS_LEN
) &&
1894 (mac
->type
< e1000_82576
)) {
1895 /* adjust PBA for jumbo frames */
1896 wr32(E1000_PBA
, pba
);
1898 /* To maintain wire speed transmits, the Tx FIFO should be
1899 * large enough to accommodate two full transmit packets,
1900 * rounded up to the next 1KB and expressed in KB. Likewise,
1901 * the Rx FIFO should be large enough to accommodate at least
1902 * one full receive packet and is similarly rounded up and
1905 pba
= rd32(E1000_PBA
);
1906 /* upper 16 bits has Tx packet buffer allocation size in KB */
1907 tx_space
= pba
>> 16;
1908 /* lower 16 bits has Rx packet buffer allocation size in KB */
1910 /* the Tx fifo also stores 16 bytes of information about the Tx
1911 * but don't include ethernet FCS because hardware appends it
1913 min_tx_space
= (adapter
->max_frame_size
+
1914 sizeof(union e1000_adv_tx_desc
) -
1916 min_tx_space
= ALIGN(min_tx_space
, 1024);
1917 min_tx_space
>>= 10;
1918 /* software strips receive CRC, so leave room for it */
1919 min_rx_space
= adapter
->max_frame_size
;
1920 min_rx_space
= ALIGN(min_rx_space
, 1024);
1921 min_rx_space
>>= 10;
1923 /* If current Tx allocation is less than the min Tx FIFO size,
1924 * and the min Tx FIFO size is less than the current Rx FIFO
1925 * allocation, take space away from current Rx allocation
1927 if (tx_space
< min_tx_space
&&
1928 ((min_tx_space
- tx_space
) < pba
)) {
1929 pba
= pba
- (min_tx_space
- tx_space
);
1931 /* if short on Rx space, Rx wins and must trump Tx
1934 if (pba
< min_rx_space
)
1937 wr32(E1000_PBA
, pba
);
1940 /* flow control settings */
1941 /* The high water mark must be low enough to fit one full frame
1942 * (or the size used for early receive) above it in the Rx FIFO.
1943 * Set it to the lower of:
1944 * - 90% of the Rx FIFO size, or
1945 * - the full Rx FIFO size minus one full frame
1947 hwm
= min(((pba
<< 10) * 9 / 10),
1948 ((pba
<< 10) - 2 * adapter
->max_frame_size
));
1950 fc
->high_water
= hwm
& 0xFFFFFFF0; /* 16-byte granularity */
1951 fc
->low_water
= fc
->high_water
- 16;
1952 fc
->pause_time
= 0xFFFF;
1954 fc
->current_mode
= fc
->requested_mode
;
1956 /* disable receive for all VFs and wait one second */
1957 if (adapter
->vfs_allocated_count
) {
1960 for (i
= 0 ; i
< adapter
->vfs_allocated_count
; i
++)
1961 adapter
->vf_data
[i
].flags
&= IGB_VF_FLAG_PF_SET_MAC
;
1963 /* ping all the active vfs to let them know we are going down */
1964 igb_ping_all_vfs(adapter
);
1966 /* disable transmits and receives */
1967 wr32(E1000_VFRE
, 0);
1968 wr32(E1000_VFTE
, 0);
1971 /* Allow time for pending master requests to run */
1972 hw
->mac
.ops
.reset_hw(hw
);
1975 if (adapter
->flags
& IGB_FLAG_MEDIA_RESET
) {
1976 /* need to resetup here after media swap */
1977 adapter
->ei
.get_invariants(hw
);
1978 adapter
->flags
&= ~IGB_FLAG_MEDIA_RESET
;
1980 if (adapter
->flags
& IGB_FLAG_MAS_ENABLE
) {
1981 if (igb_enable_mas(adapter
))
1983 "Error enabling Media Auto Sense\n");
1985 if (hw
->mac
.ops
.init_hw(hw
))
1986 dev_err(&pdev
->dev
, "Hardware Error\n");
1988 /* Flow control settings reset on hardware reset, so guarantee flow
1989 * control is off when forcing speed.
1991 if (!hw
->mac
.autoneg
)
1992 igb_force_mac_fc(hw
);
1994 igb_init_dmac(adapter
, pba
);
1995 #ifdef CONFIG_IGB_HWMON
1996 /* Re-initialize the thermal sensor on i350 devices. */
1997 if (!test_bit(__IGB_DOWN
, &adapter
->state
)) {
1998 if (mac
->type
== e1000_i350
&& hw
->bus
.func
== 0) {
1999 /* If present, re-initialize the external thermal sensor
2003 mac
->ops
.init_thermal_sensor_thresh(hw
);
2007 /* Re-establish EEE setting */
2008 if (hw
->phy
.media_type
== e1000_media_type_copper
) {
2009 switch (mac
->type
) {
2013 igb_set_eee_i350(hw
, true, true);
2016 igb_set_eee_i354(hw
, true, true);
2022 if (!netif_running(adapter
->netdev
))
2023 igb_power_down_link(adapter
);
2025 igb_update_mng_vlan(adapter
);
2027 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2028 wr32(E1000_VET
, ETHERNET_IEEE_VLAN_TYPE
);
2030 /* Re-enable PTP, where applicable. */
2031 igb_ptp_reset(adapter
);
2033 igb_get_phy_info(hw
);
2036 static netdev_features_t
igb_fix_features(struct net_device
*netdev
,
2037 netdev_features_t features
)
2039 /* Since there is no support for separate Rx/Tx vlan accel
2040 * enable/disable make sure Tx flag is always in same state as Rx.
2042 if (features
& NETIF_F_HW_VLAN_CTAG_RX
)
2043 features
|= NETIF_F_HW_VLAN_CTAG_TX
;
2045 features
&= ~NETIF_F_HW_VLAN_CTAG_TX
;
2050 static int igb_set_features(struct net_device
*netdev
,
2051 netdev_features_t features
)
2053 netdev_features_t changed
= netdev
->features
^ features
;
2054 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2056 if (changed
& NETIF_F_HW_VLAN_CTAG_RX
)
2057 igb_vlan_mode(netdev
, features
);
2059 if (!(changed
& NETIF_F_RXALL
))
2062 netdev
->features
= features
;
2064 if (netif_running(netdev
))
2065 igb_reinit_locked(adapter
);
2072 static const struct net_device_ops igb_netdev_ops
= {
2073 .ndo_open
= igb_open
,
2074 .ndo_stop
= igb_close
,
2075 .ndo_start_xmit
= igb_xmit_frame
,
2076 .ndo_get_stats64
= igb_get_stats64
,
2077 .ndo_set_rx_mode
= igb_set_rx_mode
,
2078 .ndo_set_mac_address
= igb_set_mac
,
2079 .ndo_change_mtu
= igb_change_mtu
,
2080 .ndo_do_ioctl
= igb_ioctl
,
2081 .ndo_tx_timeout
= igb_tx_timeout
,
2082 .ndo_validate_addr
= eth_validate_addr
,
2083 .ndo_vlan_rx_add_vid
= igb_vlan_rx_add_vid
,
2084 .ndo_vlan_rx_kill_vid
= igb_vlan_rx_kill_vid
,
2085 .ndo_set_vf_mac
= igb_ndo_set_vf_mac
,
2086 .ndo_set_vf_vlan
= igb_ndo_set_vf_vlan
,
2087 .ndo_set_vf_rate
= igb_ndo_set_vf_bw
,
2088 .ndo_set_vf_spoofchk
= igb_ndo_set_vf_spoofchk
,
2089 .ndo_get_vf_config
= igb_ndo_get_vf_config
,
2090 #ifdef CONFIG_NET_POLL_CONTROLLER
2091 .ndo_poll_controller
= igb_netpoll
,
2093 .ndo_fix_features
= igb_fix_features
,
2094 .ndo_set_features
= igb_set_features
,
2098 * igb_set_fw_version - Configure version string for ethtool
2099 * @adapter: adapter struct
2101 void igb_set_fw_version(struct igb_adapter
*adapter
)
2103 struct e1000_hw
*hw
= &adapter
->hw
;
2104 struct e1000_fw_version fw
;
2106 igb_get_fw_version(hw
, &fw
);
2108 switch (hw
->mac
.type
) {
2111 if (!(igb_get_flash_presence_i210(hw
))) {
2112 snprintf(adapter
->fw_version
,
2113 sizeof(adapter
->fw_version
),
2115 fw
.invm_major
, fw
.invm_minor
,
2121 /* if option is rom valid, display its version too */
2123 snprintf(adapter
->fw_version
,
2124 sizeof(adapter
->fw_version
),
2125 "%d.%d, 0x%08x, %d.%d.%d",
2126 fw
.eep_major
, fw
.eep_minor
, fw
.etrack_id
,
2127 fw
.or_major
, fw
.or_build
, fw
.or_patch
);
2129 } else if (fw
.etrack_id
!= 0X0000) {
2130 snprintf(adapter
->fw_version
,
2131 sizeof(adapter
->fw_version
),
2133 fw
.eep_major
, fw
.eep_minor
, fw
.etrack_id
);
2135 snprintf(adapter
->fw_version
,
2136 sizeof(adapter
->fw_version
),
2138 fw
.eep_major
, fw
.eep_minor
, fw
.eep_build
);
2145 * igb_init_mas - init Media Autosense feature if enabled in the NVM
2147 * @adapter: adapter struct
2149 static void igb_init_mas(struct igb_adapter
*adapter
)
2151 struct e1000_hw
*hw
= &adapter
->hw
;
2154 hw
->nvm
.ops
.read(hw
, NVM_COMPAT
, 1, &eeprom_data
);
2155 switch (hw
->bus
.func
) {
2157 if (eeprom_data
& IGB_MAS_ENABLE_0
) {
2158 adapter
->flags
|= IGB_FLAG_MAS_ENABLE
;
2159 netdev_info(adapter
->netdev
,
2160 "MAS: Enabling Media Autosense for port %d\n",
2165 if (eeprom_data
& IGB_MAS_ENABLE_1
) {
2166 adapter
->flags
|= IGB_FLAG_MAS_ENABLE
;
2167 netdev_info(adapter
->netdev
,
2168 "MAS: Enabling Media Autosense for port %d\n",
2173 if (eeprom_data
& IGB_MAS_ENABLE_2
) {
2174 adapter
->flags
|= IGB_FLAG_MAS_ENABLE
;
2175 netdev_info(adapter
->netdev
,
2176 "MAS: Enabling Media Autosense for port %d\n",
2181 if (eeprom_data
& IGB_MAS_ENABLE_3
) {
2182 adapter
->flags
|= IGB_FLAG_MAS_ENABLE
;
2183 netdev_info(adapter
->netdev
,
2184 "MAS: Enabling Media Autosense for port %d\n",
2189 /* Shouldn't get here */
2190 netdev_err(adapter
->netdev
,
2191 "MAS: Invalid port configuration, returning\n");
2197 * igb_init_i2c - Init I2C interface
2198 * @adapter: pointer to adapter structure
2200 static s32
igb_init_i2c(struct igb_adapter
*adapter
)
2204 /* I2C interface supported on i350 devices */
2205 if (adapter
->hw
.mac
.type
!= e1000_i350
)
2208 /* Initialize the i2c bus which is controlled by the registers.
2209 * This bus will use the i2c_algo_bit structue that implements
2210 * the protocol through toggling of the 4 bits in the register.
2212 adapter
->i2c_adap
.owner
= THIS_MODULE
;
2213 adapter
->i2c_algo
= igb_i2c_algo
;
2214 adapter
->i2c_algo
.data
= adapter
;
2215 adapter
->i2c_adap
.algo_data
= &adapter
->i2c_algo
;
2216 adapter
->i2c_adap
.dev
.parent
= &adapter
->pdev
->dev
;
2217 strlcpy(adapter
->i2c_adap
.name
, "igb BB",
2218 sizeof(adapter
->i2c_adap
.name
));
2219 status
= i2c_bit_add_bus(&adapter
->i2c_adap
);
2224 * igb_probe - Device Initialization Routine
2225 * @pdev: PCI device information struct
2226 * @ent: entry in igb_pci_tbl
2228 * Returns 0 on success, negative on failure
2230 * igb_probe initializes an adapter identified by a pci_dev structure.
2231 * The OS initialization, configuring of the adapter private structure,
2232 * and a hardware reset occur.
2234 static int igb_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
2236 struct net_device
*netdev
;
2237 struct igb_adapter
*adapter
;
2238 struct e1000_hw
*hw
;
2239 u16 eeprom_data
= 0;
2241 static int global_quad_port_a
; /* global quad port a indication */
2242 const struct e1000_info
*ei
= igb_info_tbl
[ent
->driver_data
];
2243 int err
, pci_using_dac
;
2244 u8 part_str
[E1000_PBANUM_LENGTH
];
2246 /* Catch broken hardware that put the wrong VF device ID in
2247 * the PCIe SR-IOV capability.
2249 if (pdev
->is_virtfn
) {
2250 WARN(1, KERN_ERR
"%s (%hx:%hx) should not be a VF!\n",
2251 pci_name(pdev
), pdev
->vendor
, pdev
->device
);
2255 err
= pci_enable_device_mem(pdev
);
2260 err
= dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(64));
2264 err
= dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(32));
2267 "No usable DMA configuration, aborting\n");
2272 err
= pci_request_selected_regions(pdev
, pci_select_bars(pdev
,
2278 pci_enable_pcie_error_reporting(pdev
);
2280 pci_set_master(pdev
);
2281 pci_save_state(pdev
);
2284 netdev
= alloc_etherdev_mq(sizeof(struct igb_adapter
),
2287 goto err_alloc_etherdev
;
2289 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
2291 pci_set_drvdata(pdev
, netdev
);
2292 adapter
= netdev_priv(netdev
);
2293 adapter
->netdev
= netdev
;
2294 adapter
->pdev
= pdev
;
2297 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
2300 hw
->hw_addr
= pci_iomap(pdev
, 0, 0);
2304 netdev
->netdev_ops
= &igb_netdev_ops
;
2305 igb_set_ethtool_ops(netdev
);
2306 netdev
->watchdog_timeo
= 5 * HZ
;
2308 strncpy(netdev
->name
, pci_name(pdev
), sizeof(netdev
->name
) - 1);
2310 netdev
->mem_start
= pci_resource_start(pdev
, 0);
2311 netdev
->mem_end
= pci_resource_end(pdev
, 0);
2313 /* PCI config space info */
2314 hw
->vendor_id
= pdev
->vendor
;
2315 hw
->device_id
= pdev
->device
;
2316 hw
->revision_id
= pdev
->revision
;
2317 hw
->subsystem_vendor_id
= pdev
->subsystem_vendor
;
2318 hw
->subsystem_device_id
= pdev
->subsystem_device
;
2320 /* Copy the default MAC, PHY and NVM function pointers */
2321 memcpy(&hw
->mac
.ops
, ei
->mac_ops
, sizeof(hw
->mac
.ops
));
2322 memcpy(&hw
->phy
.ops
, ei
->phy_ops
, sizeof(hw
->phy
.ops
));
2323 memcpy(&hw
->nvm
.ops
, ei
->nvm_ops
, sizeof(hw
->nvm
.ops
));
2324 /* Initialize skew-specific constants */
2325 err
= ei
->get_invariants(hw
);
2329 /* setup the private structure */
2330 err
= igb_sw_init(adapter
);
2334 igb_get_bus_info_pcie(hw
);
2336 hw
->phy
.autoneg_wait_to_complete
= false;
2338 /* Copper options */
2339 if (hw
->phy
.media_type
== e1000_media_type_copper
) {
2340 hw
->phy
.mdix
= AUTO_ALL_MODES
;
2341 hw
->phy
.disable_polarity_correction
= false;
2342 hw
->phy
.ms_type
= e1000_ms_hw_default
;
2345 if (igb_check_reset_block(hw
))
2346 dev_info(&pdev
->dev
,
2347 "PHY reset is blocked due to SOL/IDER session.\n");
2349 /* features is initialized to 0 in allocation, it might have bits
2350 * set by igb_sw_init so we should use an or instead of an
2353 netdev
->features
|= NETIF_F_SG
|
2360 NETIF_F_HW_VLAN_CTAG_RX
|
2361 NETIF_F_HW_VLAN_CTAG_TX
;
2363 /* copy netdev features into list of user selectable features */
2364 netdev
->hw_features
|= netdev
->features
;
2365 netdev
->hw_features
|= NETIF_F_RXALL
;
2367 /* set this bit last since it cannot be part of hw_features */
2368 netdev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
2370 netdev
->vlan_features
|= NETIF_F_TSO
|
2376 netdev
->priv_flags
|= IFF_SUPP_NOFCS
;
2378 if (pci_using_dac
) {
2379 netdev
->features
|= NETIF_F_HIGHDMA
;
2380 netdev
->vlan_features
|= NETIF_F_HIGHDMA
;
2383 if (hw
->mac
.type
>= e1000_82576
) {
2384 netdev
->hw_features
|= NETIF_F_SCTP_CSUM
;
2385 netdev
->features
|= NETIF_F_SCTP_CSUM
;
2388 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
2390 adapter
->en_mng_pt
= igb_enable_mng_pass_thru(hw
);
2392 /* before reading the NVM, reset the controller to put the device in a
2393 * known good starting state
2395 hw
->mac
.ops
.reset_hw(hw
);
2397 /* make sure the NVM is good , i211/i210 parts can have special NVM
2398 * that doesn't contain a checksum
2400 switch (hw
->mac
.type
) {
2403 if (igb_get_flash_presence_i210(hw
)) {
2404 if (hw
->nvm
.ops
.validate(hw
) < 0) {
2406 "The NVM Checksum Is Not Valid\n");
2413 if (hw
->nvm
.ops
.validate(hw
) < 0) {
2414 dev_err(&pdev
->dev
, "The NVM Checksum Is Not Valid\n");
2421 /* copy the MAC address out of the NVM */
2422 if (hw
->mac
.ops
.read_mac_addr(hw
))
2423 dev_err(&pdev
->dev
, "NVM Read Error\n");
2425 memcpy(netdev
->dev_addr
, hw
->mac
.addr
, netdev
->addr_len
);
2427 if (!is_valid_ether_addr(netdev
->dev_addr
)) {
2428 dev_err(&pdev
->dev
, "Invalid MAC Address\n");
2433 /* get firmware version for ethtool -i */
2434 igb_set_fw_version(adapter
);
2436 /* configure RXPBSIZE and TXPBSIZE */
2437 if (hw
->mac
.type
== e1000_i210
) {
2438 wr32(E1000_RXPBS
, I210_RXPBSIZE_DEFAULT
);
2439 wr32(E1000_TXPBS
, I210_TXPBSIZE_DEFAULT
);
2442 setup_timer(&adapter
->watchdog_timer
, igb_watchdog
,
2443 (unsigned long) adapter
);
2444 setup_timer(&adapter
->phy_info_timer
, igb_update_phy_info
,
2445 (unsigned long) adapter
);
2447 INIT_WORK(&adapter
->reset_task
, igb_reset_task
);
2448 INIT_WORK(&adapter
->watchdog_task
, igb_watchdog_task
);
2450 /* Initialize link properties that are user-changeable */
2451 adapter
->fc_autoneg
= true;
2452 hw
->mac
.autoneg
= true;
2453 hw
->phy
.autoneg_advertised
= 0x2f;
2455 hw
->fc
.requested_mode
= e1000_fc_default
;
2456 hw
->fc
.current_mode
= e1000_fc_default
;
2458 igb_validate_mdi_setting(hw
);
2460 /* By default, support wake on port A */
2461 if (hw
->bus
.func
== 0)
2462 adapter
->flags
|= IGB_FLAG_WOL_SUPPORTED
;
2464 /* Check the NVM for wake support on non-port A ports */
2465 if (hw
->mac
.type
>= e1000_82580
)
2466 hw
->nvm
.ops
.read(hw
, NVM_INIT_CONTROL3_PORT_A
+
2467 NVM_82580_LAN_FUNC_OFFSET(hw
->bus
.func
), 1,
2469 else if (hw
->bus
.func
== 1)
2470 hw
->nvm
.ops
.read(hw
, NVM_INIT_CONTROL3_PORT_B
, 1, &eeprom_data
);
2472 if (eeprom_data
& IGB_EEPROM_APME
)
2473 adapter
->flags
|= IGB_FLAG_WOL_SUPPORTED
;
2475 /* now that we have the eeprom settings, apply the special cases where
2476 * the eeprom may be wrong or the board simply won't support wake on
2477 * lan on a particular port
2479 switch (pdev
->device
) {
2480 case E1000_DEV_ID_82575GB_QUAD_COPPER
:
2481 adapter
->flags
&= ~IGB_FLAG_WOL_SUPPORTED
;
2483 case E1000_DEV_ID_82575EB_FIBER_SERDES
:
2484 case E1000_DEV_ID_82576_FIBER
:
2485 case E1000_DEV_ID_82576_SERDES
:
2486 /* Wake events only supported on port A for dual fiber
2487 * regardless of eeprom setting
2489 if (rd32(E1000_STATUS
) & E1000_STATUS_FUNC_1
)
2490 adapter
->flags
&= ~IGB_FLAG_WOL_SUPPORTED
;
2492 case E1000_DEV_ID_82576_QUAD_COPPER
:
2493 case E1000_DEV_ID_82576_QUAD_COPPER_ET2
:
2494 /* if quad port adapter, disable WoL on all but port A */
2495 if (global_quad_port_a
!= 0)
2496 adapter
->flags
&= ~IGB_FLAG_WOL_SUPPORTED
;
2498 adapter
->flags
|= IGB_FLAG_QUAD_PORT_A
;
2499 /* Reset for multiple quad port adapters */
2500 if (++global_quad_port_a
== 4)
2501 global_quad_port_a
= 0;
2504 /* If the device can't wake, don't set software support */
2505 if (!device_can_wakeup(&adapter
->pdev
->dev
))
2506 adapter
->flags
&= ~IGB_FLAG_WOL_SUPPORTED
;
2509 /* initialize the wol settings based on the eeprom settings */
2510 if (adapter
->flags
& IGB_FLAG_WOL_SUPPORTED
)
2511 adapter
->wol
|= E1000_WUFC_MAG
;
2513 /* Some vendors want WoL disabled by default, but still supported */
2514 if ((hw
->mac
.type
== e1000_i350
) &&
2515 (pdev
->subsystem_vendor
== PCI_VENDOR_ID_HP
)) {
2516 adapter
->flags
|= IGB_FLAG_WOL_SUPPORTED
;
2520 device_set_wakeup_enable(&adapter
->pdev
->dev
,
2521 adapter
->flags
& IGB_FLAG_WOL_SUPPORTED
);
2523 /* reset the hardware with the new settings */
2526 /* Init the I2C interface */
2527 err
= igb_init_i2c(adapter
);
2529 dev_err(&pdev
->dev
, "failed to init i2c interface\n");
2533 /* let the f/w know that the h/w is now under the control of the
2536 igb_get_hw_control(adapter
);
2538 strcpy(netdev
->name
, "eth%d");
2539 err
= register_netdev(netdev
);
2543 /* carrier off reporting is important to ethtool even BEFORE open */
2544 netif_carrier_off(netdev
);
2546 #ifdef CONFIG_IGB_DCA
2547 if (dca_add_requester(&pdev
->dev
) == 0) {
2548 adapter
->flags
|= IGB_FLAG_DCA_ENABLED
;
2549 dev_info(&pdev
->dev
, "DCA enabled\n");
2550 igb_setup_dca(adapter
);
2554 #ifdef CONFIG_IGB_HWMON
2555 /* Initialize the thermal sensor on i350 devices. */
2556 if (hw
->mac
.type
== e1000_i350
&& hw
->bus
.func
== 0) {
2559 /* Read the NVM to determine if this i350 device supports an
2560 * external thermal sensor.
2562 hw
->nvm
.ops
.read(hw
, NVM_ETS_CFG
, 1, &ets_word
);
2563 if (ets_word
!= 0x0000 && ets_word
!= 0xFFFF)
2564 adapter
->ets
= true;
2566 adapter
->ets
= false;
2567 if (igb_sysfs_init(adapter
))
2569 "failed to allocate sysfs resources\n");
2571 adapter
->ets
= false;
2574 /* Check if Media Autosense is enabled */
2576 if (hw
->dev_spec
._82575
.mas_capable
)
2577 igb_init_mas(adapter
);
2579 /* do hw tstamp init after resetting */
2580 igb_ptp_init(adapter
);
2582 dev_info(&pdev
->dev
, "Intel(R) Gigabit Ethernet Network Connection\n");
2583 /* print bus type/speed/width info, not applicable to i354 */
2584 if (hw
->mac
.type
!= e1000_i354
) {
2585 dev_info(&pdev
->dev
, "%s: (PCIe:%s:%s) %pM\n",
2587 ((hw
->bus
.speed
== e1000_bus_speed_2500
) ? "2.5Gb/s" :
2588 (hw
->bus
.speed
== e1000_bus_speed_5000
) ? "5.0Gb/s" :
2590 ((hw
->bus
.width
== e1000_bus_width_pcie_x4
) ?
2592 (hw
->bus
.width
== e1000_bus_width_pcie_x2
) ?
2594 (hw
->bus
.width
== e1000_bus_width_pcie_x1
) ?
2595 "Width x1" : "unknown"), netdev
->dev_addr
);
2598 if ((hw
->mac
.type
>= e1000_i210
||
2599 igb_get_flash_presence_i210(hw
))) {
2600 ret_val
= igb_read_part_string(hw
, part_str
,
2601 E1000_PBANUM_LENGTH
);
2603 ret_val
= -E1000_ERR_INVM_VALUE_NOT_FOUND
;
2607 strcpy(part_str
, "Unknown");
2608 dev_info(&pdev
->dev
, "%s: PBA No: %s\n", netdev
->name
, part_str
);
2609 dev_info(&pdev
->dev
,
2610 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2611 (adapter
->flags
& IGB_FLAG_HAS_MSIX
) ? "MSI-X" :
2612 (adapter
->flags
& IGB_FLAG_HAS_MSI
) ? "MSI" : "legacy",
2613 adapter
->num_rx_queues
, adapter
->num_tx_queues
);
2614 if (hw
->phy
.media_type
== e1000_media_type_copper
) {
2615 switch (hw
->mac
.type
) {
2619 /* Enable EEE for internal copper PHY devices */
2620 err
= igb_set_eee_i350(hw
, true, true);
2622 (!hw
->dev_spec
._82575
.eee_disable
)) {
2623 adapter
->eee_advert
=
2624 MDIO_EEE_100TX
| MDIO_EEE_1000T
;
2625 adapter
->flags
|= IGB_FLAG_EEE
;
2629 if ((rd32(E1000_CTRL_EXT
) &
2630 E1000_CTRL_EXT_LINK_MODE_SGMII
)) {
2631 err
= igb_set_eee_i354(hw
, true, true);
2633 (!hw
->dev_spec
._82575
.eee_disable
)) {
2634 adapter
->eee_advert
=
2635 MDIO_EEE_100TX
| MDIO_EEE_1000T
;
2636 adapter
->flags
|= IGB_FLAG_EEE
;
2644 pm_runtime_put_noidle(&pdev
->dev
);
2648 igb_release_hw_control(adapter
);
2649 memset(&adapter
->i2c_adap
, 0, sizeof(adapter
->i2c_adap
));
2651 if (!igb_check_reset_block(hw
))
2654 if (hw
->flash_address
)
2655 iounmap(hw
->flash_address
);
2657 igb_clear_interrupt_scheme(adapter
);
2658 pci_iounmap(pdev
, hw
->hw_addr
);
2660 free_netdev(netdev
);
2662 pci_release_selected_regions(pdev
,
2663 pci_select_bars(pdev
, IORESOURCE_MEM
));
2666 pci_disable_device(pdev
);
2670 #ifdef CONFIG_PCI_IOV
2671 static int igb_disable_sriov(struct pci_dev
*pdev
)
2673 struct net_device
*netdev
= pci_get_drvdata(pdev
);
2674 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2675 struct e1000_hw
*hw
= &adapter
->hw
;
2677 /* reclaim resources allocated to VFs */
2678 if (adapter
->vf_data
) {
2679 /* disable iov and allow time for transactions to clear */
2680 if (pci_vfs_assigned(pdev
)) {
2681 dev_warn(&pdev
->dev
,
2682 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2685 pci_disable_sriov(pdev
);
2689 kfree(adapter
->vf_data
);
2690 adapter
->vf_data
= NULL
;
2691 adapter
->vfs_allocated_count
= 0;
2692 wr32(E1000_IOVCTL
, E1000_IOVCTL_REUSE_VFQ
);
2695 dev_info(&pdev
->dev
, "IOV Disabled\n");
2697 /* Re-enable DMA Coalescing flag since IOV is turned off */
2698 adapter
->flags
|= IGB_FLAG_DMAC
;
2704 static int igb_enable_sriov(struct pci_dev
*pdev
, int num_vfs
)
2706 struct net_device
*netdev
= pci_get_drvdata(pdev
);
2707 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2708 int old_vfs
= pci_num_vf(pdev
);
2712 if (!(adapter
->flags
& IGB_FLAG_HAS_MSIX
) || num_vfs
> 7) {
2720 dev_info(&pdev
->dev
, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
2722 adapter
->vfs_allocated_count
= old_vfs
;
2724 adapter
->vfs_allocated_count
= num_vfs
;
2726 adapter
->vf_data
= kcalloc(adapter
->vfs_allocated_count
,
2727 sizeof(struct vf_data_storage
), GFP_KERNEL
);
2729 /* if allocation failed then we do not support SR-IOV */
2730 if (!adapter
->vf_data
) {
2731 adapter
->vfs_allocated_count
= 0;
2733 "Unable to allocate memory for VF Data Storage\n");
2738 /* only call pci_enable_sriov() if no VFs are allocated already */
2740 err
= pci_enable_sriov(pdev
, adapter
->vfs_allocated_count
);
2744 dev_info(&pdev
->dev
, "%d VFs allocated\n",
2745 adapter
->vfs_allocated_count
);
2746 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++)
2747 igb_vf_configure(adapter
, i
);
2749 /* DMA Coalescing is not supported in IOV mode. */
2750 adapter
->flags
&= ~IGB_FLAG_DMAC
;
2754 kfree(adapter
->vf_data
);
2755 adapter
->vf_data
= NULL
;
2756 adapter
->vfs_allocated_count
= 0;
2763 * igb_remove_i2c - Cleanup I2C interface
2764 * @adapter: pointer to adapter structure
2766 static void igb_remove_i2c(struct igb_adapter
*adapter
)
2768 /* free the adapter bus structure */
2769 i2c_del_adapter(&adapter
->i2c_adap
);
2773 * igb_remove - Device Removal Routine
2774 * @pdev: PCI device information struct
2776 * igb_remove is called by the PCI subsystem to alert the driver
2777 * that it should release a PCI device. The could be caused by a
2778 * Hot-Plug event, or because the driver is going to be removed from
2781 static void igb_remove(struct pci_dev
*pdev
)
2783 struct net_device
*netdev
= pci_get_drvdata(pdev
);
2784 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2785 struct e1000_hw
*hw
= &adapter
->hw
;
2787 pm_runtime_get_noresume(&pdev
->dev
);
2788 #ifdef CONFIG_IGB_HWMON
2789 igb_sysfs_exit(adapter
);
2791 igb_remove_i2c(adapter
);
2792 igb_ptp_stop(adapter
);
2793 /* The watchdog timer may be rescheduled, so explicitly
2794 * disable watchdog from being rescheduled.
2796 set_bit(__IGB_DOWN
, &adapter
->state
);
2797 del_timer_sync(&adapter
->watchdog_timer
);
2798 del_timer_sync(&adapter
->phy_info_timer
);
2800 cancel_work_sync(&adapter
->reset_task
);
2801 cancel_work_sync(&adapter
->watchdog_task
);
2803 #ifdef CONFIG_IGB_DCA
2804 if (adapter
->flags
& IGB_FLAG_DCA_ENABLED
) {
2805 dev_info(&pdev
->dev
, "DCA disabled\n");
2806 dca_remove_requester(&pdev
->dev
);
2807 adapter
->flags
&= ~IGB_FLAG_DCA_ENABLED
;
2808 wr32(E1000_DCA_CTRL
, E1000_DCA_CTRL_DCA_MODE_DISABLE
);
2812 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2813 * would have already happened in close and is redundant.
2815 igb_release_hw_control(adapter
);
2817 unregister_netdev(netdev
);
2819 igb_clear_interrupt_scheme(adapter
);
2821 #ifdef CONFIG_PCI_IOV
2822 igb_disable_sriov(pdev
);
2825 pci_iounmap(pdev
, hw
->hw_addr
);
2826 if (hw
->flash_address
)
2827 iounmap(hw
->flash_address
);
2828 pci_release_selected_regions(pdev
,
2829 pci_select_bars(pdev
, IORESOURCE_MEM
));
2831 kfree(adapter
->shadow_vfta
);
2832 free_netdev(netdev
);
2834 pci_disable_pcie_error_reporting(pdev
);
2836 pci_disable_device(pdev
);
2840 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2841 * @adapter: board private structure to initialize
2843 * This function initializes the vf specific data storage and then attempts to
2844 * allocate the VFs. The reason for ordering it this way is because it is much
2845 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2846 * the memory for the VFs.
2848 static void igb_probe_vfs(struct igb_adapter
*adapter
)
2850 #ifdef CONFIG_PCI_IOV
2851 struct pci_dev
*pdev
= adapter
->pdev
;
2852 struct e1000_hw
*hw
= &adapter
->hw
;
2854 /* Virtualization features not supported on i210 family. */
2855 if ((hw
->mac
.type
== e1000_i210
) || (hw
->mac
.type
== e1000_i211
))
2858 pci_sriov_set_totalvfs(pdev
, 7);
2859 igb_pci_enable_sriov(pdev
, max_vfs
);
2861 #endif /* CONFIG_PCI_IOV */
2864 static void igb_init_queue_configuration(struct igb_adapter
*adapter
)
2866 struct e1000_hw
*hw
= &adapter
->hw
;
2869 /* Determine the maximum number of RSS queues supported. */
2870 switch (hw
->mac
.type
) {
2872 max_rss_queues
= IGB_MAX_RX_QUEUES_I211
;
2876 max_rss_queues
= IGB_MAX_RX_QUEUES_82575
;
2879 /* I350 cannot do RSS and SR-IOV at the same time */
2880 if (!!adapter
->vfs_allocated_count
) {
2886 if (!!adapter
->vfs_allocated_count
) {
2894 max_rss_queues
= IGB_MAX_RX_QUEUES
;
2898 adapter
->rss_queues
= min_t(u32
, max_rss_queues
, num_online_cpus());
2900 /* Determine if we need to pair queues. */
2901 switch (hw
->mac
.type
) {
2904 /* Device supports enough interrupts without queue pairing. */
2907 /* If VFs are going to be allocated with RSS queues then we
2908 * should pair the queues in order to conserve interrupts due
2909 * to limited supply.
2911 if ((adapter
->rss_queues
> 1) &&
2912 (adapter
->vfs_allocated_count
> 6))
2913 adapter
->flags
|= IGB_FLAG_QUEUE_PAIRS
;
2920 /* If rss_queues > half of max_rss_queues, pair the queues in
2921 * order to conserve interrupts due to limited supply.
2923 if (adapter
->rss_queues
> (max_rss_queues
/ 2))
2924 adapter
->flags
|= IGB_FLAG_QUEUE_PAIRS
;
2930 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2931 * @adapter: board private structure to initialize
2933 * igb_sw_init initializes the Adapter private data structure.
2934 * Fields are initialized based on PCI device information and
2935 * OS network device settings (MTU size).
2937 static int igb_sw_init(struct igb_adapter
*adapter
)
2939 struct e1000_hw
*hw
= &adapter
->hw
;
2940 struct net_device
*netdev
= adapter
->netdev
;
2941 struct pci_dev
*pdev
= adapter
->pdev
;
2943 pci_read_config_word(pdev
, PCI_COMMAND
, &hw
->bus
.pci_cmd_word
);
2945 /* set default ring sizes */
2946 adapter
->tx_ring_count
= IGB_DEFAULT_TXD
;
2947 adapter
->rx_ring_count
= IGB_DEFAULT_RXD
;
2949 /* set default ITR values */
2950 adapter
->rx_itr_setting
= IGB_DEFAULT_ITR
;
2951 adapter
->tx_itr_setting
= IGB_DEFAULT_ITR
;
2953 /* set default work limits */
2954 adapter
->tx_work_limit
= IGB_DEFAULT_TX_WORK
;
2956 adapter
->max_frame_size
= netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+
2958 adapter
->min_frame_size
= ETH_ZLEN
+ ETH_FCS_LEN
;
2960 spin_lock_init(&adapter
->stats64_lock
);
2961 #ifdef CONFIG_PCI_IOV
2962 switch (hw
->mac
.type
) {
2966 dev_warn(&pdev
->dev
,
2967 "Maximum of 7 VFs per PF, using max\n");
2968 max_vfs
= adapter
->vfs_allocated_count
= 7;
2970 adapter
->vfs_allocated_count
= max_vfs
;
2971 if (adapter
->vfs_allocated_count
)
2972 dev_warn(&pdev
->dev
,
2973 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
2978 #endif /* CONFIG_PCI_IOV */
2980 igb_init_queue_configuration(adapter
);
2982 /* Setup and initialize a copy of the hw vlan table array */
2983 adapter
->shadow_vfta
= kcalloc(E1000_VLAN_FILTER_TBL_SIZE
, sizeof(u32
),
2986 /* This call may decrease the number of queues */
2987 if (igb_init_interrupt_scheme(adapter
, true)) {
2988 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
2992 igb_probe_vfs(adapter
);
2994 /* Explicitly disable IRQ since the NIC can be in any state. */
2995 igb_irq_disable(adapter
);
2997 if (hw
->mac
.type
>= e1000_i350
)
2998 adapter
->flags
&= ~IGB_FLAG_DMAC
;
3000 set_bit(__IGB_DOWN
, &adapter
->state
);
3005 * igb_open - Called when a network interface is made active
3006 * @netdev: network interface device structure
3008 * Returns 0 on success, negative value on failure
3010 * The open entry point is called when a network interface is made
3011 * active by the system (IFF_UP). At this point all resources needed
3012 * for transmit and receive operations are allocated, the interrupt
3013 * handler is registered with the OS, the watchdog timer is started,
3014 * and the stack is notified that the interface is ready.
3016 static int __igb_open(struct net_device
*netdev
, bool resuming
)
3018 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3019 struct e1000_hw
*hw
= &adapter
->hw
;
3020 struct pci_dev
*pdev
= adapter
->pdev
;
3024 /* disallow open during test */
3025 if (test_bit(__IGB_TESTING
, &adapter
->state
)) {
3031 pm_runtime_get_sync(&pdev
->dev
);
3033 netif_carrier_off(netdev
);
3035 /* allocate transmit descriptors */
3036 err
= igb_setup_all_tx_resources(adapter
);
3040 /* allocate receive descriptors */
3041 err
= igb_setup_all_rx_resources(adapter
);
3045 igb_power_up_link(adapter
);
3047 /* before we allocate an interrupt, we must be ready to handle it.
3048 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3049 * as soon as we call pci_request_irq, so we have to setup our
3050 * clean_rx handler before we do so.
3052 igb_configure(adapter
);
3054 err
= igb_request_irq(adapter
);
3058 /* Notify the stack of the actual queue counts. */
3059 err
= netif_set_real_num_tx_queues(adapter
->netdev
,
3060 adapter
->num_tx_queues
);
3062 goto err_set_queues
;
3064 err
= netif_set_real_num_rx_queues(adapter
->netdev
,
3065 adapter
->num_rx_queues
);
3067 goto err_set_queues
;
3069 /* From here on the code is the same as igb_up() */
3070 clear_bit(__IGB_DOWN
, &adapter
->state
);
3072 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
3073 napi_enable(&(adapter
->q_vector
[i
]->napi
));
3075 /* Clear any pending interrupts. */
3078 igb_irq_enable(adapter
);
3080 /* notify VFs that reset has been completed */
3081 if (adapter
->vfs_allocated_count
) {
3082 u32 reg_data
= rd32(E1000_CTRL_EXT
);
3084 reg_data
|= E1000_CTRL_EXT_PFRSTD
;
3085 wr32(E1000_CTRL_EXT
, reg_data
);
3088 netif_tx_start_all_queues(netdev
);
3091 pm_runtime_put(&pdev
->dev
);
3093 /* start the watchdog. */
3094 hw
->mac
.get_link_status
= 1;
3095 schedule_work(&adapter
->watchdog_task
);
3100 igb_free_irq(adapter
);
3102 igb_release_hw_control(adapter
);
3103 igb_power_down_link(adapter
);
3104 igb_free_all_rx_resources(adapter
);
3106 igb_free_all_tx_resources(adapter
);
3110 pm_runtime_put(&pdev
->dev
);
3115 static int igb_open(struct net_device
*netdev
)
3117 return __igb_open(netdev
, false);
3121 * igb_close - Disables a network interface
3122 * @netdev: network interface device structure
3124 * Returns 0, this is not allowed to fail
3126 * The close entry point is called when an interface is de-activated
3127 * by the OS. The hardware is still under the driver's control, but
3128 * needs to be disabled. A global MAC reset is issued to stop the
3129 * hardware, and all transmit and receive resources are freed.
3131 static int __igb_close(struct net_device
*netdev
, bool suspending
)
3133 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3134 struct pci_dev
*pdev
= adapter
->pdev
;
3136 WARN_ON(test_bit(__IGB_RESETTING
, &adapter
->state
));
3139 pm_runtime_get_sync(&pdev
->dev
);
3142 igb_free_irq(adapter
);
3144 igb_free_all_tx_resources(adapter
);
3145 igb_free_all_rx_resources(adapter
);
3148 pm_runtime_put_sync(&pdev
->dev
);
3152 static int igb_close(struct net_device
*netdev
)
3154 return __igb_close(netdev
, false);
3158 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
3159 * @tx_ring: tx descriptor ring (for a specific queue) to setup
3161 * Return 0 on success, negative on failure
3163 int igb_setup_tx_resources(struct igb_ring
*tx_ring
)
3165 struct device
*dev
= tx_ring
->dev
;
3168 size
= sizeof(struct igb_tx_buffer
) * tx_ring
->count
;
3170 tx_ring
->tx_buffer_info
= vzalloc(size
);
3171 if (!tx_ring
->tx_buffer_info
)
3174 /* round up to nearest 4K */
3175 tx_ring
->size
= tx_ring
->count
* sizeof(union e1000_adv_tx_desc
);
3176 tx_ring
->size
= ALIGN(tx_ring
->size
, 4096);
3178 tx_ring
->desc
= dma_alloc_coherent(dev
, tx_ring
->size
,
3179 &tx_ring
->dma
, GFP_KERNEL
);
3183 tx_ring
->next_to_use
= 0;
3184 tx_ring
->next_to_clean
= 0;
3189 vfree(tx_ring
->tx_buffer_info
);
3190 tx_ring
->tx_buffer_info
= NULL
;
3191 dev_err(dev
, "Unable to allocate memory for the Tx descriptor ring\n");
3196 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
3197 * (Descriptors) for all queues
3198 * @adapter: board private structure
3200 * Return 0 on success, negative on failure
3202 static int igb_setup_all_tx_resources(struct igb_adapter
*adapter
)
3204 struct pci_dev
*pdev
= adapter
->pdev
;
3207 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
3208 err
= igb_setup_tx_resources(adapter
->tx_ring
[i
]);
3211 "Allocation for Tx Queue %u failed\n", i
);
3212 for (i
--; i
>= 0; i
--)
3213 igb_free_tx_resources(adapter
->tx_ring
[i
]);
3222 * igb_setup_tctl - configure the transmit control registers
3223 * @adapter: Board private structure
3225 void igb_setup_tctl(struct igb_adapter
*adapter
)
3227 struct e1000_hw
*hw
= &adapter
->hw
;
3230 /* disable queue 0 which is enabled by default on 82575 and 82576 */
3231 wr32(E1000_TXDCTL(0), 0);
3233 /* Program the Transmit Control Register */
3234 tctl
= rd32(E1000_TCTL
);
3235 tctl
&= ~E1000_TCTL_CT
;
3236 tctl
|= E1000_TCTL_PSP
| E1000_TCTL_RTLC
|
3237 (E1000_COLLISION_THRESHOLD
<< E1000_CT_SHIFT
);
3239 igb_config_collision_dist(hw
);
3241 /* Enable transmits */
3242 tctl
|= E1000_TCTL_EN
;
3244 wr32(E1000_TCTL
, tctl
);
3248 * igb_configure_tx_ring - Configure transmit ring after Reset
3249 * @adapter: board private structure
3250 * @ring: tx ring to configure
3252 * Configure a transmit ring after a reset.
3254 void igb_configure_tx_ring(struct igb_adapter
*adapter
,
3255 struct igb_ring
*ring
)
3257 struct e1000_hw
*hw
= &adapter
->hw
;
3259 u64 tdba
= ring
->dma
;
3260 int reg_idx
= ring
->reg_idx
;
3262 /* disable the queue */
3263 wr32(E1000_TXDCTL(reg_idx
), 0);
3267 wr32(E1000_TDLEN(reg_idx
),
3268 ring
->count
* sizeof(union e1000_adv_tx_desc
));
3269 wr32(E1000_TDBAL(reg_idx
),
3270 tdba
& 0x00000000ffffffffULL
);
3271 wr32(E1000_TDBAH(reg_idx
), tdba
>> 32);
3273 ring
->tail
= hw
->hw_addr
+ E1000_TDT(reg_idx
);
3274 wr32(E1000_TDH(reg_idx
), 0);
3275 writel(0, ring
->tail
);
3277 txdctl
|= IGB_TX_PTHRESH
;
3278 txdctl
|= IGB_TX_HTHRESH
<< 8;
3279 txdctl
|= IGB_TX_WTHRESH
<< 16;
3281 txdctl
|= E1000_TXDCTL_QUEUE_ENABLE
;
3282 wr32(E1000_TXDCTL(reg_idx
), txdctl
);
3286 * igb_configure_tx - Configure transmit Unit after Reset
3287 * @adapter: board private structure
3289 * Configure the Tx unit of the MAC after a reset.
3291 static void igb_configure_tx(struct igb_adapter
*adapter
)
3295 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3296 igb_configure_tx_ring(adapter
, adapter
->tx_ring
[i
]);
3300 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
3301 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3303 * Returns 0 on success, negative on failure
3305 int igb_setup_rx_resources(struct igb_ring
*rx_ring
)
3307 struct device
*dev
= rx_ring
->dev
;
3310 size
= sizeof(struct igb_rx_buffer
) * rx_ring
->count
;
3312 rx_ring
->rx_buffer_info
= vzalloc(size
);
3313 if (!rx_ring
->rx_buffer_info
)
3316 /* Round up to nearest 4K */
3317 rx_ring
->size
= rx_ring
->count
* sizeof(union e1000_adv_rx_desc
);
3318 rx_ring
->size
= ALIGN(rx_ring
->size
, 4096);
3320 rx_ring
->desc
= dma_alloc_coherent(dev
, rx_ring
->size
,
3321 &rx_ring
->dma
, GFP_KERNEL
);
3325 rx_ring
->next_to_alloc
= 0;
3326 rx_ring
->next_to_clean
= 0;
3327 rx_ring
->next_to_use
= 0;
3332 vfree(rx_ring
->rx_buffer_info
);
3333 rx_ring
->rx_buffer_info
= NULL
;
3334 dev_err(dev
, "Unable to allocate memory for the Rx descriptor ring\n");
3339 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
3340 * (Descriptors) for all queues
3341 * @adapter: board private structure
3343 * Return 0 on success, negative on failure
3345 static int igb_setup_all_rx_resources(struct igb_adapter
*adapter
)
3347 struct pci_dev
*pdev
= adapter
->pdev
;
3350 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3351 err
= igb_setup_rx_resources(adapter
->rx_ring
[i
]);
3354 "Allocation for Rx Queue %u failed\n", i
);
3355 for (i
--; i
>= 0; i
--)
3356 igb_free_rx_resources(adapter
->rx_ring
[i
]);
3365 * igb_setup_mrqc - configure the multiple receive queue control registers
3366 * @adapter: Board private structure
3368 static void igb_setup_mrqc(struct igb_adapter
*adapter
)
3370 struct e1000_hw
*hw
= &adapter
->hw
;
3372 u32 j
, num_rx_queues
;
3373 static const u32 rsskey
[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
3374 0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
3375 0xA32DCB77, 0x0CF23080, 0x3BB7426A,
3378 /* Fill out hash function seeds */
3379 for (j
= 0; j
< 10; j
++)
3380 wr32(E1000_RSSRK(j
), rsskey
[j
]);
3382 num_rx_queues
= adapter
->rss_queues
;
3384 switch (hw
->mac
.type
) {
3386 /* 82576 supports 2 RSS queues for SR-IOV */
3387 if (adapter
->vfs_allocated_count
)
3394 if (adapter
->rss_indir_tbl_init
!= num_rx_queues
) {
3395 for (j
= 0; j
< IGB_RETA_SIZE
; j
++)
3396 adapter
->rss_indir_tbl
[j
] =
3397 (j
* num_rx_queues
) / IGB_RETA_SIZE
;
3398 adapter
->rss_indir_tbl_init
= num_rx_queues
;
3400 igb_write_rss_indir_tbl(adapter
);
3402 /* Disable raw packet checksumming so that RSS hash is placed in
3403 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
3404 * offloads as they are enabled by default
3406 rxcsum
= rd32(E1000_RXCSUM
);
3407 rxcsum
|= E1000_RXCSUM_PCSD
;
3409 if (adapter
->hw
.mac
.type
>= e1000_82576
)
3410 /* Enable Receive Checksum Offload for SCTP */
3411 rxcsum
|= E1000_RXCSUM_CRCOFL
;
3413 /* Don't need to set TUOFL or IPOFL, they default to 1 */
3414 wr32(E1000_RXCSUM
, rxcsum
);
3416 /* Generate RSS hash based on packet types, TCP/UDP
3417 * port numbers and/or IPv4/v6 src and dst addresses
3419 mrqc
= E1000_MRQC_RSS_FIELD_IPV4
|
3420 E1000_MRQC_RSS_FIELD_IPV4_TCP
|
3421 E1000_MRQC_RSS_FIELD_IPV6
|
3422 E1000_MRQC_RSS_FIELD_IPV6_TCP
|
3423 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX
;
3425 if (adapter
->flags
& IGB_FLAG_RSS_FIELD_IPV4_UDP
)
3426 mrqc
|= E1000_MRQC_RSS_FIELD_IPV4_UDP
;
3427 if (adapter
->flags
& IGB_FLAG_RSS_FIELD_IPV6_UDP
)
3428 mrqc
|= E1000_MRQC_RSS_FIELD_IPV6_UDP
;
3430 /* If VMDq is enabled then we set the appropriate mode for that, else
3431 * we default to RSS so that an RSS hash is calculated per packet even
3432 * if we are only using one queue
3434 if (adapter
->vfs_allocated_count
) {
3435 if (hw
->mac
.type
> e1000_82575
) {
3436 /* Set the default pool for the PF's first queue */
3437 u32 vtctl
= rd32(E1000_VT_CTL
);
3439 vtctl
&= ~(E1000_VT_CTL_DEFAULT_POOL_MASK
|
3440 E1000_VT_CTL_DISABLE_DEF_POOL
);
3441 vtctl
|= adapter
->vfs_allocated_count
<<
3442 E1000_VT_CTL_DEFAULT_POOL_SHIFT
;
3443 wr32(E1000_VT_CTL
, vtctl
);
3445 if (adapter
->rss_queues
> 1)
3446 mrqc
|= E1000_MRQC_ENABLE_VMDQ_RSS_2Q
;
3448 mrqc
|= E1000_MRQC_ENABLE_VMDQ
;
3450 if (hw
->mac
.type
!= e1000_i211
)
3451 mrqc
|= E1000_MRQC_ENABLE_RSS_4Q
;
3453 igb_vmm_control(adapter
);
3455 wr32(E1000_MRQC
, mrqc
);
3459 * igb_setup_rctl - configure the receive control registers
3460 * @adapter: Board private structure
3462 void igb_setup_rctl(struct igb_adapter
*adapter
)
3464 struct e1000_hw
*hw
= &adapter
->hw
;
3467 rctl
= rd32(E1000_RCTL
);
3469 rctl
&= ~(3 << E1000_RCTL_MO_SHIFT
);
3470 rctl
&= ~(E1000_RCTL_LBM_TCVR
| E1000_RCTL_LBM_MAC
);
3472 rctl
|= E1000_RCTL_EN
| E1000_RCTL_BAM
| E1000_RCTL_RDMTS_HALF
|
3473 (hw
->mac
.mc_filter_type
<< E1000_RCTL_MO_SHIFT
);
3475 /* enable stripping of CRC. It's unlikely this will break BMC
3476 * redirection as it did with e1000. Newer features require
3477 * that the HW strips the CRC.
3479 rctl
|= E1000_RCTL_SECRC
;
3481 /* disable store bad packets and clear size bits. */
3482 rctl
&= ~(E1000_RCTL_SBP
| E1000_RCTL_SZ_256
);
3484 /* enable LPE to prevent packets larger than max_frame_size */
3485 rctl
|= E1000_RCTL_LPE
;
3487 /* disable queue 0 to prevent tail write w/o re-config */
3488 wr32(E1000_RXDCTL(0), 0);
3490 /* Attention!!! For SR-IOV PF driver operations you must enable
3491 * queue drop for all VF and PF queues to prevent head of line blocking
3492 * if an un-trusted VF does not provide descriptors to hardware.
3494 if (adapter
->vfs_allocated_count
) {
3495 /* set all queue drop enable bits */
3496 wr32(E1000_QDE
, ALL_QUEUES
);
3499 /* This is useful for sniffing bad packets. */
3500 if (adapter
->netdev
->features
& NETIF_F_RXALL
) {
3501 /* UPE and MPE will be handled by normal PROMISC logic
3502 * in e1000e_set_rx_mode
3504 rctl
|= (E1000_RCTL_SBP
| /* Receive bad packets */
3505 E1000_RCTL_BAM
| /* RX All Bcast Pkts */
3506 E1000_RCTL_PMCF
); /* RX All MAC Ctrl Pkts */
3508 rctl
&= ~(E1000_RCTL_VFE
| /* Disable VLAN filter */
3509 E1000_RCTL_DPF
| /* Allow filtered pause */
3510 E1000_RCTL_CFIEN
); /* Dis VLAN CFIEN Filter */
3511 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3512 * and that breaks VLANs.
3516 wr32(E1000_RCTL
, rctl
);
3519 static inline int igb_set_vf_rlpml(struct igb_adapter
*adapter
, int size
,
3522 struct e1000_hw
*hw
= &adapter
->hw
;
3525 /* if it isn't the PF check to see if VFs are enabled and
3526 * increase the size to support vlan tags
3528 if (vfn
< adapter
->vfs_allocated_count
&&
3529 adapter
->vf_data
[vfn
].vlans_enabled
)
3530 size
+= VLAN_TAG_SIZE
;
3532 vmolr
= rd32(E1000_VMOLR(vfn
));
3533 vmolr
&= ~E1000_VMOLR_RLPML_MASK
;
3534 vmolr
|= size
| E1000_VMOLR_LPE
;
3535 wr32(E1000_VMOLR(vfn
), vmolr
);
3541 * igb_rlpml_set - set maximum receive packet size
3542 * @adapter: board private structure
3544 * Configure maximum receivable packet size.
3546 static void igb_rlpml_set(struct igb_adapter
*adapter
)
3548 u32 max_frame_size
= adapter
->max_frame_size
;
3549 struct e1000_hw
*hw
= &adapter
->hw
;
3550 u16 pf_id
= adapter
->vfs_allocated_count
;
3553 igb_set_vf_rlpml(adapter
, max_frame_size
, pf_id
);
3554 /* If we're in VMDQ or SR-IOV mode, then set global RLPML
3555 * to our max jumbo frame size, in case we need to enable
3556 * jumbo frames on one of the rings later.
3557 * This will not pass over-length frames into the default
3558 * queue because it's gated by the VMOLR.RLPML.
3560 max_frame_size
= MAX_JUMBO_FRAME_SIZE
;
3563 wr32(E1000_RLPML
, max_frame_size
);
3566 static inline void igb_set_vmolr(struct igb_adapter
*adapter
,
3569 struct e1000_hw
*hw
= &adapter
->hw
;
3572 /* This register exists only on 82576 and newer so if we are older then
3573 * we should exit and do nothing
3575 if (hw
->mac
.type
< e1000_82576
)
3578 vmolr
= rd32(E1000_VMOLR(vfn
));
3579 vmolr
|= E1000_VMOLR_STRVLAN
; /* Strip vlan tags */
3580 if (hw
->mac
.type
== e1000_i350
) {
3583 dvmolr
= rd32(E1000_DVMOLR(vfn
));
3584 dvmolr
|= E1000_DVMOLR_STRVLAN
;
3585 wr32(E1000_DVMOLR(vfn
), dvmolr
);
3588 vmolr
|= E1000_VMOLR_AUPE
; /* Accept untagged packets */
3590 vmolr
&= ~(E1000_VMOLR_AUPE
); /* Tagged packets ONLY */
3592 /* clear all bits that might not be set */
3593 vmolr
&= ~(E1000_VMOLR_BAM
| E1000_VMOLR_RSSE
);
3595 if (adapter
->rss_queues
> 1 && vfn
== adapter
->vfs_allocated_count
)
3596 vmolr
|= E1000_VMOLR_RSSE
; /* enable RSS */
3597 /* for VMDq only allow the VFs and pool 0 to accept broadcast and
3600 if (vfn
<= adapter
->vfs_allocated_count
)
3601 vmolr
|= E1000_VMOLR_BAM
; /* Accept broadcast */
3603 wr32(E1000_VMOLR(vfn
), vmolr
);
3607 * igb_configure_rx_ring - Configure a receive ring after Reset
3608 * @adapter: board private structure
3609 * @ring: receive ring to be configured
3611 * Configure the Rx unit of the MAC after a reset.
3613 void igb_configure_rx_ring(struct igb_adapter
*adapter
,
3614 struct igb_ring
*ring
)
3616 struct e1000_hw
*hw
= &adapter
->hw
;
3617 u64 rdba
= ring
->dma
;
3618 int reg_idx
= ring
->reg_idx
;
3619 u32 srrctl
= 0, rxdctl
= 0;
3621 /* disable the queue */
3622 wr32(E1000_RXDCTL(reg_idx
), 0);
3624 /* Set DMA base address registers */
3625 wr32(E1000_RDBAL(reg_idx
),
3626 rdba
& 0x00000000ffffffffULL
);
3627 wr32(E1000_RDBAH(reg_idx
), rdba
>> 32);
3628 wr32(E1000_RDLEN(reg_idx
),
3629 ring
->count
* sizeof(union e1000_adv_rx_desc
));
3631 /* initialize head and tail */
3632 ring
->tail
= hw
->hw_addr
+ E1000_RDT(reg_idx
);
3633 wr32(E1000_RDH(reg_idx
), 0);
3634 writel(0, ring
->tail
);
3636 /* set descriptor configuration */
3637 srrctl
= IGB_RX_HDR_LEN
<< E1000_SRRCTL_BSIZEHDRSIZE_SHIFT
;
3638 srrctl
|= IGB_RX_BUFSZ
>> E1000_SRRCTL_BSIZEPKT_SHIFT
;
3639 srrctl
|= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF
;
3640 if (hw
->mac
.type
>= e1000_82580
)
3641 srrctl
|= E1000_SRRCTL_TIMESTAMP
;
3642 /* Only set Drop Enable if we are supporting multiple queues */
3643 if (adapter
->vfs_allocated_count
|| adapter
->num_rx_queues
> 1)
3644 srrctl
|= E1000_SRRCTL_DROP_EN
;
3646 wr32(E1000_SRRCTL(reg_idx
), srrctl
);
3648 /* set filtering for VMDQ pools */
3649 igb_set_vmolr(adapter
, reg_idx
& 0x7, true);
3651 rxdctl
|= IGB_RX_PTHRESH
;
3652 rxdctl
|= IGB_RX_HTHRESH
<< 8;
3653 rxdctl
|= IGB_RX_WTHRESH
<< 16;
3655 /* enable receive descriptor fetching */
3656 rxdctl
|= E1000_RXDCTL_QUEUE_ENABLE
;
3657 wr32(E1000_RXDCTL(reg_idx
), rxdctl
);
3661 * igb_configure_rx - Configure receive Unit after Reset
3662 * @adapter: board private structure
3664 * Configure the Rx unit of the MAC after a reset.
3666 static void igb_configure_rx(struct igb_adapter
*adapter
)
3670 /* set UTA to appropriate mode */
3671 igb_set_uta(adapter
);
3673 /* set the correct pool for the PF default MAC address in entry 0 */
3674 igb_rar_set_qsel(adapter
, adapter
->hw
.mac
.addr
, 0,
3675 adapter
->vfs_allocated_count
);
3677 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3678 * the Base and Length of the Rx Descriptor Ring
3680 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3681 igb_configure_rx_ring(adapter
, adapter
->rx_ring
[i
]);
3685 * igb_free_tx_resources - Free Tx Resources per Queue
3686 * @tx_ring: Tx descriptor ring for a specific queue
3688 * Free all transmit software resources
3690 void igb_free_tx_resources(struct igb_ring
*tx_ring
)
3692 igb_clean_tx_ring(tx_ring
);
3694 vfree(tx_ring
->tx_buffer_info
);
3695 tx_ring
->tx_buffer_info
= NULL
;
3697 /* if not set, then don't free */
3701 dma_free_coherent(tx_ring
->dev
, tx_ring
->size
,
3702 tx_ring
->desc
, tx_ring
->dma
);
3704 tx_ring
->desc
= NULL
;
3708 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3709 * @adapter: board private structure
3711 * Free all transmit software resources
3713 static void igb_free_all_tx_resources(struct igb_adapter
*adapter
)
3717 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3718 igb_free_tx_resources(adapter
->tx_ring
[i
]);
3721 void igb_unmap_and_free_tx_resource(struct igb_ring
*ring
,
3722 struct igb_tx_buffer
*tx_buffer
)
3724 if (tx_buffer
->skb
) {
3725 dev_kfree_skb_any(tx_buffer
->skb
);
3726 if (dma_unmap_len(tx_buffer
, len
))
3727 dma_unmap_single(ring
->dev
,
3728 dma_unmap_addr(tx_buffer
, dma
),
3729 dma_unmap_len(tx_buffer
, len
),
3731 } else if (dma_unmap_len(tx_buffer
, len
)) {
3732 dma_unmap_page(ring
->dev
,
3733 dma_unmap_addr(tx_buffer
, dma
),
3734 dma_unmap_len(tx_buffer
, len
),
3737 tx_buffer
->next_to_watch
= NULL
;
3738 tx_buffer
->skb
= NULL
;
3739 dma_unmap_len_set(tx_buffer
, len
, 0);
3740 /* buffer_info must be completely set up in the transmit path */
3744 * igb_clean_tx_ring - Free Tx Buffers
3745 * @tx_ring: ring to be cleaned
3747 static void igb_clean_tx_ring(struct igb_ring
*tx_ring
)
3749 struct igb_tx_buffer
*buffer_info
;
3753 if (!tx_ring
->tx_buffer_info
)
3755 /* Free all the Tx ring sk_buffs */
3757 for (i
= 0; i
< tx_ring
->count
; i
++) {
3758 buffer_info
= &tx_ring
->tx_buffer_info
[i
];
3759 igb_unmap_and_free_tx_resource(tx_ring
, buffer_info
);
3762 netdev_tx_reset_queue(txring_txq(tx_ring
));
3764 size
= sizeof(struct igb_tx_buffer
) * tx_ring
->count
;
3765 memset(tx_ring
->tx_buffer_info
, 0, size
);
3767 /* Zero out the descriptor ring */
3768 memset(tx_ring
->desc
, 0, tx_ring
->size
);
3770 tx_ring
->next_to_use
= 0;
3771 tx_ring
->next_to_clean
= 0;
3775 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3776 * @adapter: board private structure
3778 static void igb_clean_all_tx_rings(struct igb_adapter
*adapter
)
3782 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3783 igb_clean_tx_ring(adapter
->tx_ring
[i
]);
3787 * igb_free_rx_resources - Free Rx Resources
3788 * @rx_ring: ring to clean the resources from
3790 * Free all receive software resources
3792 void igb_free_rx_resources(struct igb_ring
*rx_ring
)
3794 igb_clean_rx_ring(rx_ring
);
3796 vfree(rx_ring
->rx_buffer_info
);
3797 rx_ring
->rx_buffer_info
= NULL
;
3799 /* if not set, then don't free */
3803 dma_free_coherent(rx_ring
->dev
, rx_ring
->size
,
3804 rx_ring
->desc
, rx_ring
->dma
);
3806 rx_ring
->desc
= NULL
;
3810 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3811 * @adapter: board private structure
3813 * Free all receive software resources
3815 static void igb_free_all_rx_resources(struct igb_adapter
*adapter
)
3819 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3820 igb_free_rx_resources(adapter
->rx_ring
[i
]);
3824 * igb_clean_rx_ring - Free Rx Buffers per Queue
3825 * @rx_ring: ring to free buffers from
3827 static void igb_clean_rx_ring(struct igb_ring
*rx_ring
)
3833 dev_kfree_skb(rx_ring
->skb
);
3834 rx_ring
->skb
= NULL
;
3836 if (!rx_ring
->rx_buffer_info
)
3839 /* Free all the Rx ring sk_buffs */
3840 for (i
= 0; i
< rx_ring
->count
; i
++) {
3841 struct igb_rx_buffer
*buffer_info
= &rx_ring
->rx_buffer_info
[i
];
3843 if (!buffer_info
->page
)
3846 dma_unmap_page(rx_ring
->dev
,
3850 __free_page(buffer_info
->page
);
3852 buffer_info
->page
= NULL
;
3855 size
= sizeof(struct igb_rx_buffer
) * rx_ring
->count
;
3856 memset(rx_ring
->rx_buffer_info
, 0, size
);
3858 /* Zero out the descriptor ring */
3859 memset(rx_ring
->desc
, 0, rx_ring
->size
);
3861 rx_ring
->next_to_alloc
= 0;
3862 rx_ring
->next_to_clean
= 0;
3863 rx_ring
->next_to_use
= 0;
3867 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3868 * @adapter: board private structure
3870 static void igb_clean_all_rx_rings(struct igb_adapter
*adapter
)
3874 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3875 igb_clean_rx_ring(adapter
->rx_ring
[i
]);
3879 * igb_set_mac - Change the Ethernet Address of the NIC
3880 * @netdev: network interface device structure
3881 * @p: pointer to an address structure
3883 * Returns 0 on success, negative on failure
3885 static int igb_set_mac(struct net_device
*netdev
, void *p
)
3887 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3888 struct e1000_hw
*hw
= &adapter
->hw
;
3889 struct sockaddr
*addr
= p
;
3891 if (!is_valid_ether_addr(addr
->sa_data
))
3892 return -EADDRNOTAVAIL
;
3894 memcpy(netdev
->dev_addr
, addr
->sa_data
, netdev
->addr_len
);
3895 memcpy(hw
->mac
.addr
, addr
->sa_data
, netdev
->addr_len
);
3897 /* set the correct pool for the new PF MAC address in entry 0 */
3898 igb_rar_set_qsel(adapter
, hw
->mac
.addr
, 0,
3899 adapter
->vfs_allocated_count
);
3905 * igb_write_mc_addr_list - write multicast addresses to MTA
3906 * @netdev: network interface device structure
3908 * Writes multicast address list to the MTA hash table.
3909 * Returns: -ENOMEM on failure
3910 * 0 on no addresses written
3911 * X on writing X addresses to MTA
3913 static int igb_write_mc_addr_list(struct net_device
*netdev
)
3915 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3916 struct e1000_hw
*hw
= &adapter
->hw
;
3917 struct netdev_hw_addr
*ha
;
3921 if (netdev_mc_empty(netdev
)) {
3922 /* nothing to program, so clear mc list */
3923 igb_update_mc_addr_list(hw
, NULL
, 0);
3924 igb_restore_vf_multicasts(adapter
);
3928 mta_list
= kzalloc(netdev_mc_count(netdev
) * 6, GFP_ATOMIC
);
3932 /* The shared function expects a packed array of only addresses. */
3934 netdev_for_each_mc_addr(ha
, netdev
)
3935 memcpy(mta_list
+ (i
++ * ETH_ALEN
), ha
->addr
, ETH_ALEN
);
3937 igb_update_mc_addr_list(hw
, mta_list
, i
);
3940 return netdev_mc_count(netdev
);
3944 * igb_write_uc_addr_list - write unicast addresses to RAR table
3945 * @netdev: network interface device structure
3947 * Writes unicast address list to the RAR table.
3948 * Returns: -ENOMEM on failure/insufficient address space
3949 * 0 on no addresses written
3950 * X on writing X addresses to the RAR table
3952 static int igb_write_uc_addr_list(struct net_device
*netdev
)
3954 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3955 struct e1000_hw
*hw
= &adapter
->hw
;
3956 unsigned int vfn
= adapter
->vfs_allocated_count
;
3957 unsigned int rar_entries
= hw
->mac
.rar_entry_count
- (vfn
+ 1);
3960 /* return ENOMEM indicating insufficient memory for addresses */
3961 if (netdev_uc_count(netdev
) > rar_entries
)
3964 if (!netdev_uc_empty(netdev
) && rar_entries
) {
3965 struct netdev_hw_addr
*ha
;
3967 netdev_for_each_uc_addr(ha
, netdev
) {
3970 igb_rar_set_qsel(adapter
, ha
->addr
,
3976 /* write the addresses in reverse order to avoid write combining */
3977 for (; rar_entries
> 0 ; rar_entries
--) {
3978 wr32(E1000_RAH(rar_entries
), 0);
3979 wr32(E1000_RAL(rar_entries
), 0);
3987 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3988 * @netdev: network interface device structure
3990 * The set_rx_mode entry point is called whenever the unicast or multicast
3991 * address lists or the network interface flags are updated. This routine is
3992 * responsible for configuring the hardware for proper unicast, multicast,
3993 * promiscuous mode, and all-multi behavior.
3995 static void igb_set_rx_mode(struct net_device
*netdev
)
3997 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3998 struct e1000_hw
*hw
= &adapter
->hw
;
3999 unsigned int vfn
= adapter
->vfs_allocated_count
;
4000 u32 rctl
, vmolr
= 0;
4003 /* Check for Promiscuous and All Multicast modes */
4004 rctl
= rd32(E1000_RCTL
);
4006 /* clear the effected bits */
4007 rctl
&= ~(E1000_RCTL_UPE
| E1000_RCTL_MPE
| E1000_RCTL_VFE
);
4009 if (netdev
->flags
& IFF_PROMISC
) {
4010 /* retain VLAN HW filtering if in VT mode */
4011 if (adapter
->vfs_allocated_count
)
4012 rctl
|= E1000_RCTL_VFE
;
4013 rctl
|= (E1000_RCTL_UPE
| E1000_RCTL_MPE
);
4014 vmolr
|= (E1000_VMOLR_ROPE
| E1000_VMOLR_MPME
);
4016 if (netdev
->flags
& IFF_ALLMULTI
) {
4017 rctl
|= E1000_RCTL_MPE
;
4018 vmolr
|= E1000_VMOLR_MPME
;
4020 /* Write addresses to the MTA, if the attempt fails
4021 * then we should just turn on promiscuous mode so
4022 * that we can at least receive multicast traffic
4024 count
= igb_write_mc_addr_list(netdev
);
4026 rctl
|= E1000_RCTL_MPE
;
4027 vmolr
|= E1000_VMOLR_MPME
;
4029 vmolr
|= E1000_VMOLR_ROMPE
;
4032 /* Write addresses to available RAR registers, if there is not
4033 * sufficient space to store all the addresses then enable
4034 * unicast promiscuous mode
4036 count
= igb_write_uc_addr_list(netdev
);
4038 rctl
|= E1000_RCTL_UPE
;
4039 vmolr
|= E1000_VMOLR_ROPE
;
4041 rctl
|= E1000_RCTL_VFE
;
4043 wr32(E1000_RCTL
, rctl
);
4045 /* In order to support SR-IOV and eventually VMDq it is necessary to set
4046 * the VMOLR to enable the appropriate modes. Without this workaround
4047 * we will have issues with VLAN tag stripping not being done for frames
4048 * that are only arriving because we are the default pool
4050 if ((hw
->mac
.type
< e1000_82576
) || (hw
->mac
.type
> e1000_i350
))
4053 vmolr
|= rd32(E1000_VMOLR(vfn
)) &
4054 ~(E1000_VMOLR_ROPE
| E1000_VMOLR_MPME
| E1000_VMOLR_ROMPE
);
4055 wr32(E1000_VMOLR(vfn
), vmolr
);
4056 igb_restore_vf_multicasts(adapter
);
4059 static void igb_check_wvbr(struct igb_adapter
*adapter
)
4061 struct e1000_hw
*hw
= &adapter
->hw
;
4064 switch (hw
->mac
.type
) {
4067 wvbr
= rd32(E1000_WVBR
);
4075 adapter
->wvbr
|= wvbr
;
4078 #define IGB_STAGGERED_QUEUE_OFFSET 8
4080 static void igb_spoof_check(struct igb_adapter
*adapter
)
4087 for (j
= 0; j
< adapter
->vfs_allocated_count
; j
++) {
4088 if (adapter
->wvbr
& (1 << j
) ||
4089 adapter
->wvbr
& (1 << (j
+ IGB_STAGGERED_QUEUE_OFFSET
))) {
4090 dev_warn(&adapter
->pdev
->dev
,
4091 "Spoof event(s) detected on VF %d\n", j
);
4094 (1 << (j
+ IGB_STAGGERED_QUEUE_OFFSET
)));
4099 /* Need to wait a few seconds after link up to get diagnostic information from
4102 static void igb_update_phy_info(unsigned long data
)
4104 struct igb_adapter
*adapter
= (struct igb_adapter
*) data
;
4105 igb_get_phy_info(&adapter
->hw
);
4109 * igb_has_link - check shared code for link and determine up/down
4110 * @adapter: pointer to driver private info
4112 bool igb_has_link(struct igb_adapter
*adapter
)
4114 struct e1000_hw
*hw
= &adapter
->hw
;
4115 bool link_active
= false;
4117 /* get_link_status is set on LSC (link status) interrupt or
4118 * rx sequence error interrupt. get_link_status will stay
4119 * false until the e1000_check_for_link establishes link
4120 * for copper adapters ONLY
4122 switch (hw
->phy
.media_type
) {
4123 case e1000_media_type_copper
:
4124 if (!hw
->mac
.get_link_status
)
4126 case e1000_media_type_internal_serdes
:
4127 hw
->mac
.ops
.check_for_link(hw
);
4128 link_active
= !hw
->mac
.get_link_status
;
4131 case e1000_media_type_unknown
:
4135 if (((hw
->mac
.type
== e1000_i210
) ||
4136 (hw
->mac
.type
== e1000_i211
)) &&
4137 (hw
->phy
.id
== I210_I_PHY_ID
)) {
4138 if (!netif_carrier_ok(adapter
->netdev
)) {
4139 adapter
->flags
&= ~IGB_FLAG_NEED_LINK_UPDATE
;
4140 } else if (!(adapter
->flags
& IGB_FLAG_NEED_LINK_UPDATE
)) {
4141 adapter
->flags
|= IGB_FLAG_NEED_LINK_UPDATE
;
4142 adapter
->link_check_timeout
= jiffies
;
4149 static bool igb_thermal_sensor_event(struct e1000_hw
*hw
, u32 event
)
4152 u32 ctrl_ext
, thstat
;
4154 /* check for thermal sensor event on i350 copper only */
4155 if (hw
->mac
.type
== e1000_i350
) {
4156 thstat
= rd32(E1000_THSTAT
);
4157 ctrl_ext
= rd32(E1000_CTRL_EXT
);
4159 if ((hw
->phy
.media_type
== e1000_media_type_copper
) &&
4160 !(ctrl_ext
& E1000_CTRL_EXT_LINK_MODE_SGMII
))
4161 ret
= !!(thstat
& event
);
4168 * igb_check_lvmmc - check for malformed packets received
4169 * and indicated in LVMMC register
4170 * @adapter: pointer to adapter
4172 static void igb_check_lvmmc(struct igb_adapter
*adapter
)
4174 struct e1000_hw
*hw
= &adapter
->hw
;
4177 lvmmc
= rd32(E1000_LVMMC
);
4179 if (unlikely(net_ratelimit())) {
4180 netdev_warn(adapter
->netdev
,
4181 "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
4188 * igb_watchdog - Timer Call-back
4189 * @data: pointer to adapter cast into an unsigned long
4191 static void igb_watchdog(unsigned long data
)
4193 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
4194 /* Do the rest outside of interrupt context */
4195 schedule_work(&adapter
->watchdog_task
);
4198 static void igb_watchdog_task(struct work_struct
*work
)
4200 struct igb_adapter
*adapter
= container_of(work
,
4203 struct e1000_hw
*hw
= &adapter
->hw
;
4204 struct e1000_phy_info
*phy
= &hw
->phy
;
4205 struct net_device
*netdev
= adapter
->netdev
;
4210 link
= igb_has_link(adapter
);
4212 if (adapter
->flags
& IGB_FLAG_NEED_LINK_UPDATE
) {
4213 if (time_after(jiffies
, (adapter
->link_check_timeout
+ HZ
)))
4214 adapter
->flags
&= ~IGB_FLAG_NEED_LINK_UPDATE
;
4219 /* Force link down if we have fiber to swap to */
4220 if (adapter
->flags
& IGB_FLAG_MAS_ENABLE
) {
4221 if (hw
->phy
.media_type
== e1000_media_type_copper
) {
4222 connsw
= rd32(E1000_CONNSW
);
4223 if (!(connsw
& E1000_CONNSW_AUTOSENSE_EN
))
4228 /* Perform a reset if the media type changed. */
4229 if (hw
->dev_spec
._82575
.media_changed
) {
4230 hw
->dev_spec
._82575
.media_changed
= false;
4231 adapter
->flags
|= IGB_FLAG_MEDIA_RESET
;
4234 /* Cancel scheduled suspend requests. */
4235 pm_runtime_resume(netdev
->dev
.parent
);
4237 if (!netif_carrier_ok(netdev
)) {
4240 hw
->mac
.ops
.get_speed_and_duplex(hw
,
4241 &adapter
->link_speed
,
4242 &adapter
->link_duplex
);
4244 ctrl
= rd32(E1000_CTRL
);
4245 /* Links status message must follow this format */
4247 "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4249 adapter
->link_speed
,
4250 adapter
->link_duplex
== FULL_DUPLEX
?
4252 (ctrl
& E1000_CTRL_TFCE
) &&
4253 (ctrl
& E1000_CTRL_RFCE
) ? "RX/TX" :
4254 (ctrl
& E1000_CTRL_RFCE
) ? "RX" :
4255 (ctrl
& E1000_CTRL_TFCE
) ? "TX" : "None");
4257 /* disable EEE if enabled */
4258 if ((adapter
->flags
& IGB_FLAG_EEE
) &&
4259 (adapter
->link_duplex
== HALF_DUPLEX
)) {
4260 dev_info(&adapter
->pdev
->dev
,
4261 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
4262 adapter
->hw
.dev_spec
._82575
.eee_disable
= true;
4263 adapter
->flags
&= ~IGB_FLAG_EEE
;
4266 /* check if SmartSpeed worked */
4267 igb_check_downshift(hw
);
4268 if (phy
->speed_downgraded
)
4269 netdev_warn(netdev
, "Link Speed was downgraded by SmartSpeed\n");
4271 /* check for thermal sensor event */
4272 if (igb_thermal_sensor_event(hw
,
4273 E1000_THSTAT_LINK_THROTTLE
))
4274 netdev_info(netdev
, "The network adapter link speed was downshifted because it overheated\n");
4276 /* adjust timeout factor according to speed/duplex */
4277 adapter
->tx_timeout_factor
= 1;
4278 switch (adapter
->link_speed
) {
4280 adapter
->tx_timeout_factor
= 14;
4283 /* maybe add some timeout factor ? */
4287 netif_carrier_on(netdev
);
4289 igb_ping_all_vfs(adapter
);
4290 igb_check_vf_rate_limit(adapter
);
4292 /* link state has changed, schedule phy info update */
4293 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
4294 mod_timer(&adapter
->phy_info_timer
,
4295 round_jiffies(jiffies
+ 2 * HZ
));
4298 if (netif_carrier_ok(netdev
)) {
4299 adapter
->link_speed
= 0;
4300 adapter
->link_duplex
= 0;
4302 /* check for thermal sensor event */
4303 if (igb_thermal_sensor_event(hw
,
4304 E1000_THSTAT_PWR_DOWN
)) {
4305 netdev_err(netdev
, "The network adapter was stopped because it overheated\n");
4308 /* Links status message must follow this format */
4309 netdev_info(netdev
, "igb: %s NIC Link is Down\n",
4311 netif_carrier_off(netdev
);
4313 igb_ping_all_vfs(adapter
);
4315 /* link state has changed, schedule phy info update */
4316 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
4317 mod_timer(&adapter
->phy_info_timer
,
4318 round_jiffies(jiffies
+ 2 * HZ
));
4320 /* link is down, time to check for alternate media */
4321 if (adapter
->flags
& IGB_FLAG_MAS_ENABLE
) {
4322 igb_check_swap_media(adapter
);
4323 if (adapter
->flags
& IGB_FLAG_MEDIA_RESET
) {
4324 schedule_work(&adapter
->reset_task
);
4325 /* return immediately */
4329 pm_schedule_suspend(netdev
->dev
.parent
,
4332 /* also check for alternate media here */
4333 } else if (!netif_carrier_ok(netdev
) &&
4334 (adapter
->flags
& IGB_FLAG_MAS_ENABLE
)) {
4335 igb_check_swap_media(adapter
);
4336 if (adapter
->flags
& IGB_FLAG_MEDIA_RESET
) {
4337 schedule_work(&adapter
->reset_task
);
4338 /* return immediately */
4344 spin_lock(&adapter
->stats64_lock
);
4345 igb_update_stats(adapter
, &adapter
->stats64
);
4346 spin_unlock(&adapter
->stats64_lock
);
4348 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
4349 struct igb_ring
*tx_ring
= adapter
->tx_ring
[i
];
4350 if (!netif_carrier_ok(netdev
)) {
4351 /* We've lost link, so the controller stops DMA,
4352 * but we've got queued Tx work that's never going
4353 * to get done, so reset controller to flush Tx.
4354 * (Do the reset outside of interrupt context).
4356 if (igb_desc_unused(tx_ring
) + 1 < tx_ring
->count
) {
4357 adapter
->tx_timeout_count
++;
4358 schedule_work(&adapter
->reset_task
);
4359 /* return immediately since reset is imminent */
4364 /* Force detection of hung controller every watchdog period */
4365 set_bit(IGB_RING_FLAG_TX_DETECT_HANG
, &tx_ring
->flags
);
4368 /* Cause software interrupt to ensure Rx ring is cleaned */
4369 if (adapter
->flags
& IGB_FLAG_HAS_MSIX
) {
4372 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
4373 eics
|= adapter
->q_vector
[i
]->eims_value
;
4374 wr32(E1000_EICS
, eics
);
4376 wr32(E1000_ICS
, E1000_ICS_RXDMT0
);
4379 igb_spoof_check(adapter
);
4380 igb_ptp_rx_hang(adapter
);
4382 /* Check LVMMC register on i350/i354 only */
4383 if ((adapter
->hw
.mac
.type
== e1000_i350
) ||
4384 (adapter
->hw
.mac
.type
== e1000_i354
))
4385 igb_check_lvmmc(adapter
);
4387 /* Reset the timer */
4388 if (!test_bit(__IGB_DOWN
, &adapter
->state
)) {
4389 if (adapter
->flags
& IGB_FLAG_NEED_LINK_UPDATE
)
4390 mod_timer(&adapter
->watchdog_timer
,
4391 round_jiffies(jiffies
+ HZ
));
4393 mod_timer(&adapter
->watchdog_timer
,
4394 round_jiffies(jiffies
+ 2 * HZ
));
4398 enum latency_range
{
4402 latency_invalid
= 255
4406 * igb_update_ring_itr - update the dynamic ITR value based on packet size
4407 * @q_vector: pointer to q_vector
4409 * Stores a new ITR value based on strictly on packet size. This
4410 * algorithm is less sophisticated than that used in igb_update_itr,
4411 * due to the difficulty of synchronizing statistics across multiple
4412 * receive rings. The divisors and thresholds used by this function
4413 * were determined based on theoretical maximum wire speed and testing
4414 * data, in order to minimize response time while increasing bulk
4416 * This functionality is controlled by ethtool's coalescing settings.
4417 * NOTE: This function is called only when operating in a multiqueue
4418 * receive environment.
4420 static void igb_update_ring_itr(struct igb_q_vector
*q_vector
)
4422 int new_val
= q_vector
->itr_val
;
4423 int avg_wire_size
= 0;
4424 struct igb_adapter
*adapter
= q_vector
->adapter
;
4425 unsigned int packets
;
4427 /* For non-gigabit speeds, just fix the interrupt rate at 4000
4428 * ints/sec - ITR timer value of 120 ticks.
4430 if (adapter
->link_speed
!= SPEED_1000
) {
4431 new_val
= IGB_4K_ITR
;
4435 packets
= q_vector
->rx
.total_packets
;
4437 avg_wire_size
= q_vector
->rx
.total_bytes
/ packets
;
4439 packets
= q_vector
->tx
.total_packets
;
4441 avg_wire_size
= max_t(u32
, avg_wire_size
,
4442 q_vector
->tx
.total_bytes
/ packets
);
4444 /* if avg_wire_size isn't set no work was done */
4448 /* Add 24 bytes to size to account for CRC, preamble, and gap */
4449 avg_wire_size
+= 24;
4451 /* Don't starve jumbo frames */
4452 avg_wire_size
= min(avg_wire_size
, 3000);
4454 /* Give a little boost to mid-size frames */
4455 if ((avg_wire_size
> 300) && (avg_wire_size
< 1200))
4456 new_val
= avg_wire_size
/ 3;
4458 new_val
= avg_wire_size
/ 2;
4460 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4461 if (new_val
< IGB_20K_ITR
&&
4462 ((q_vector
->rx
.ring
&& adapter
->rx_itr_setting
== 3) ||
4463 (!q_vector
->rx
.ring
&& adapter
->tx_itr_setting
== 3)))
4464 new_val
= IGB_20K_ITR
;
4467 if (new_val
!= q_vector
->itr_val
) {
4468 q_vector
->itr_val
= new_val
;
4469 q_vector
->set_itr
= 1;
4472 q_vector
->rx
.total_bytes
= 0;
4473 q_vector
->rx
.total_packets
= 0;
4474 q_vector
->tx
.total_bytes
= 0;
4475 q_vector
->tx
.total_packets
= 0;
4479 * igb_update_itr - update the dynamic ITR value based on statistics
4480 * @q_vector: pointer to q_vector
4481 * @ring_container: ring info to update the itr for
4483 * Stores a new ITR value based on packets and byte
4484 * counts during the last interrupt. The advantage of per interrupt
4485 * computation is faster updates and more accurate ITR for the current
4486 * traffic pattern. Constants in this function were computed
4487 * based on theoretical maximum wire speed and thresholds were set based
4488 * on testing data as well as attempting to minimize response time
4489 * while increasing bulk throughput.
4490 * This functionality is controlled by ethtool's coalescing settings.
4491 * NOTE: These calculations are only valid when operating in a single-
4492 * queue environment.
4494 static void igb_update_itr(struct igb_q_vector
*q_vector
,
4495 struct igb_ring_container
*ring_container
)
4497 unsigned int packets
= ring_container
->total_packets
;
4498 unsigned int bytes
= ring_container
->total_bytes
;
4499 u8 itrval
= ring_container
->itr
;
4501 /* no packets, exit with status unchanged */
4506 case lowest_latency
:
4507 /* handle TSO and jumbo frames */
4508 if (bytes
/packets
> 8000)
4509 itrval
= bulk_latency
;
4510 else if ((packets
< 5) && (bytes
> 512))
4511 itrval
= low_latency
;
4513 case low_latency
: /* 50 usec aka 20000 ints/s */
4514 if (bytes
> 10000) {
4515 /* this if handles the TSO accounting */
4516 if (bytes
/packets
> 8000)
4517 itrval
= bulk_latency
;
4518 else if ((packets
< 10) || ((bytes
/packets
) > 1200))
4519 itrval
= bulk_latency
;
4520 else if ((packets
> 35))
4521 itrval
= lowest_latency
;
4522 } else if (bytes
/packets
> 2000) {
4523 itrval
= bulk_latency
;
4524 } else if (packets
<= 2 && bytes
< 512) {
4525 itrval
= lowest_latency
;
4528 case bulk_latency
: /* 250 usec aka 4000 ints/s */
4529 if (bytes
> 25000) {
4531 itrval
= low_latency
;
4532 } else if (bytes
< 1500) {
4533 itrval
= low_latency
;
4538 /* clear work counters since we have the values we need */
4539 ring_container
->total_bytes
= 0;
4540 ring_container
->total_packets
= 0;
4542 /* write updated itr to ring container */
4543 ring_container
->itr
= itrval
;
4546 static void igb_set_itr(struct igb_q_vector
*q_vector
)
4548 struct igb_adapter
*adapter
= q_vector
->adapter
;
4549 u32 new_itr
= q_vector
->itr_val
;
4552 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4553 if (adapter
->link_speed
!= SPEED_1000
) {
4555 new_itr
= IGB_4K_ITR
;
4559 igb_update_itr(q_vector
, &q_vector
->tx
);
4560 igb_update_itr(q_vector
, &q_vector
->rx
);
4562 current_itr
= max(q_vector
->rx
.itr
, q_vector
->tx
.itr
);
4564 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4565 if (current_itr
== lowest_latency
&&
4566 ((q_vector
->rx
.ring
&& adapter
->rx_itr_setting
== 3) ||
4567 (!q_vector
->rx
.ring
&& adapter
->tx_itr_setting
== 3)))
4568 current_itr
= low_latency
;
4570 switch (current_itr
) {
4571 /* counts and packets in update_itr are dependent on these numbers */
4572 case lowest_latency
:
4573 new_itr
= IGB_70K_ITR
; /* 70,000 ints/sec */
4576 new_itr
= IGB_20K_ITR
; /* 20,000 ints/sec */
4579 new_itr
= IGB_4K_ITR
; /* 4,000 ints/sec */
4586 if (new_itr
!= q_vector
->itr_val
) {
4587 /* this attempts to bias the interrupt rate towards Bulk
4588 * by adding intermediate steps when interrupt rate is
4591 new_itr
= new_itr
> q_vector
->itr_val
?
4592 max((new_itr
* q_vector
->itr_val
) /
4593 (new_itr
+ (q_vector
->itr_val
>> 2)),
4595 /* Don't write the value here; it resets the adapter's
4596 * internal timer, and causes us to delay far longer than
4597 * we should between interrupts. Instead, we write the ITR
4598 * value at the beginning of the next interrupt so the timing
4599 * ends up being correct.
4601 q_vector
->itr_val
= new_itr
;
4602 q_vector
->set_itr
= 1;
4606 static void igb_tx_ctxtdesc(struct igb_ring
*tx_ring
, u32 vlan_macip_lens
,
4607 u32 type_tucmd
, u32 mss_l4len_idx
)
4609 struct e1000_adv_tx_context_desc
*context_desc
;
4610 u16 i
= tx_ring
->next_to_use
;
4612 context_desc
= IGB_TX_CTXTDESC(tx_ring
, i
);
4615 tx_ring
->next_to_use
= (i
< tx_ring
->count
) ? i
: 0;
4617 /* set bits to identify this as an advanced context descriptor */
4618 type_tucmd
|= E1000_TXD_CMD_DEXT
| E1000_ADVTXD_DTYP_CTXT
;
4620 /* For 82575, context index must be unique per ring. */
4621 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX
, &tx_ring
->flags
))
4622 mss_l4len_idx
|= tx_ring
->reg_idx
<< 4;
4624 context_desc
->vlan_macip_lens
= cpu_to_le32(vlan_macip_lens
);
4625 context_desc
->seqnum_seed
= 0;
4626 context_desc
->type_tucmd_mlhl
= cpu_to_le32(type_tucmd
);
4627 context_desc
->mss_l4len_idx
= cpu_to_le32(mss_l4len_idx
);
4630 static int igb_tso(struct igb_ring
*tx_ring
,
4631 struct igb_tx_buffer
*first
,
4634 struct sk_buff
*skb
= first
->skb
;
4635 u32 vlan_macip_lens
, type_tucmd
;
4636 u32 mss_l4len_idx
, l4len
;
4639 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
4642 if (!skb_is_gso(skb
))
4645 err
= skb_cow_head(skb
, 0);
4649 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4650 type_tucmd
= E1000_ADVTXD_TUCMD_L4T_TCP
;
4652 if (first
->protocol
== htons(ETH_P_IP
)) {
4653 struct iphdr
*iph
= ip_hdr(skb
);
4656 tcp_hdr(skb
)->check
= ~csum_tcpudp_magic(iph
->saddr
,
4660 type_tucmd
|= E1000_ADVTXD_TUCMD_IPV4
;
4661 first
->tx_flags
|= IGB_TX_FLAGS_TSO
|
4664 } else if (skb_is_gso_v6(skb
)) {
4665 ipv6_hdr(skb
)->payload_len
= 0;
4666 tcp_hdr(skb
)->check
= ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
4667 &ipv6_hdr(skb
)->daddr
,
4669 first
->tx_flags
|= IGB_TX_FLAGS_TSO
|
4673 /* compute header lengths */
4674 l4len
= tcp_hdrlen(skb
);
4675 *hdr_len
= skb_transport_offset(skb
) + l4len
;
4677 /* update gso size and bytecount with header size */
4678 first
->gso_segs
= skb_shinfo(skb
)->gso_segs
;
4679 first
->bytecount
+= (first
->gso_segs
- 1) * *hdr_len
;
4682 mss_l4len_idx
= l4len
<< E1000_ADVTXD_L4LEN_SHIFT
;
4683 mss_l4len_idx
|= skb_shinfo(skb
)->gso_size
<< E1000_ADVTXD_MSS_SHIFT
;
4685 /* VLAN MACLEN IPLEN */
4686 vlan_macip_lens
= skb_network_header_len(skb
);
4687 vlan_macip_lens
|= skb_network_offset(skb
) << E1000_ADVTXD_MACLEN_SHIFT
;
4688 vlan_macip_lens
|= first
->tx_flags
& IGB_TX_FLAGS_VLAN_MASK
;
4690 igb_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, type_tucmd
, mss_l4len_idx
);
4695 static void igb_tx_csum(struct igb_ring
*tx_ring
, struct igb_tx_buffer
*first
)
4697 struct sk_buff
*skb
= first
->skb
;
4698 u32 vlan_macip_lens
= 0;
4699 u32 mss_l4len_idx
= 0;
4702 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
) {
4703 if (!(first
->tx_flags
& IGB_TX_FLAGS_VLAN
))
4708 switch (first
->protocol
) {
4709 case htons(ETH_P_IP
):
4710 vlan_macip_lens
|= skb_network_header_len(skb
);
4711 type_tucmd
|= E1000_ADVTXD_TUCMD_IPV4
;
4712 l4_hdr
= ip_hdr(skb
)->protocol
;
4714 case htons(ETH_P_IPV6
):
4715 vlan_macip_lens
|= skb_network_header_len(skb
);
4716 l4_hdr
= ipv6_hdr(skb
)->nexthdr
;
4719 if (unlikely(net_ratelimit())) {
4720 dev_warn(tx_ring
->dev
,
4721 "partial checksum but proto=%x!\n",
4729 type_tucmd
|= E1000_ADVTXD_TUCMD_L4T_TCP
;
4730 mss_l4len_idx
= tcp_hdrlen(skb
) <<
4731 E1000_ADVTXD_L4LEN_SHIFT
;
4734 type_tucmd
|= E1000_ADVTXD_TUCMD_L4T_SCTP
;
4735 mss_l4len_idx
= sizeof(struct sctphdr
) <<
4736 E1000_ADVTXD_L4LEN_SHIFT
;
4739 mss_l4len_idx
= sizeof(struct udphdr
) <<
4740 E1000_ADVTXD_L4LEN_SHIFT
;
4743 if (unlikely(net_ratelimit())) {
4744 dev_warn(tx_ring
->dev
,
4745 "partial checksum but l4 proto=%x!\n",
4751 /* update TX checksum flag */
4752 first
->tx_flags
|= IGB_TX_FLAGS_CSUM
;
4755 vlan_macip_lens
|= skb_network_offset(skb
) << E1000_ADVTXD_MACLEN_SHIFT
;
4756 vlan_macip_lens
|= first
->tx_flags
& IGB_TX_FLAGS_VLAN_MASK
;
4758 igb_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, type_tucmd
, mss_l4len_idx
);
4761 #define IGB_SET_FLAG(_input, _flag, _result) \
4762 ((_flag <= _result) ? \
4763 ((u32)(_input & _flag) * (_result / _flag)) : \
4764 ((u32)(_input & _flag) / (_flag / _result)))
4766 static u32
igb_tx_cmd_type(struct sk_buff
*skb
, u32 tx_flags
)
4768 /* set type for advanced descriptor with frame checksum insertion */
4769 u32 cmd_type
= E1000_ADVTXD_DTYP_DATA
|
4770 E1000_ADVTXD_DCMD_DEXT
|
4771 E1000_ADVTXD_DCMD_IFCS
;
4773 /* set HW vlan bit if vlan is present */
4774 cmd_type
|= IGB_SET_FLAG(tx_flags
, IGB_TX_FLAGS_VLAN
,
4775 (E1000_ADVTXD_DCMD_VLE
));
4777 /* set segmentation bits for TSO */
4778 cmd_type
|= IGB_SET_FLAG(tx_flags
, IGB_TX_FLAGS_TSO
,
4779 (E1000_ADVTXD_DCMD_TSE
));
4781 /* set timestamp bit if present */
4782 cmd_type
|= IGB_SET_FLAG(tx_flags
, IGB_TX_FLAGS_TSTAMP
,
4783 (E1000_ADVTXD_MAC_TSTAMP
));
4785 /* insert frame checksum */
4786 cmd_type
^= IGB_SET_FLAG(skb
->no_fcs
, 1, E1000_ADVTXD_DCMD_IFCS
);
4791 static void igb_tx_olinfo_status(struct igb_ring
*tx_ring
,
4792 union e1000_adv_tx_desc
*tx_desc
,
4793 u32 tx_flags
, unsigned int paylen
)
4795 u32 olinfo_status
= paylen
<< E1000_ADVTXD_PAYLEN_SHIFT
;
4797 /* 82575 requires a unique index per ring */
4798 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX
, &tx_ring
->flags
))
4799 olinfo_status
|= tx_ring
->reg_idx
<< 4;
4801 /* insert L4 checksum */
4802 olinfo_status
|= IGB_SET_FLAG(tx_flags
,
4804 (E1000_TXD_POPTS_TXSM
<< 8));
4806 /* insert IPv4 checksum */
4807 olinfo_status
|= IGB_SET_FLAG(tx_flags
,
4809 (E1000_TXD_POPTS_IXSM
<< 8));
4811 tx_desc
->read
.olinfo_status
= cpu_to_le32(olinfo_status
);
4814 static int __igb_maybe_stop_tx(struct igb_ring
*tx_ring
, const u16 size
)
4816 struct net_device
*netdev
= tx_ring
->netdev
;
4818 netif_stop_subqueue(netdev
, tx_ring
->queue_index
);
4820 /* Herbert's original patch had:
4821 * smp_mb__after_netif_stop_queue();
4822 * but since that doesn't exist yet, just open code it.
4826 /* We need to check again in a case another CPU has just
4827 * made room available.
4829 if (igb_desc_unused(tx_ring
) < size
)
4833 netif_wake_subqueue(netdev
, tx_ring
->queue_index
);
4835 u64_stats_update_begin(&tx_ring
->tx_syncp2
);
4836 tx_ring
->tx_stats
.restart_queue2
++;
4837 u64_stats_update_end(&tx_ring
->tx_syncp2
);
4842 static inline int igb_maybe_stop_tx(struct igb_ring
*tx_ring
, const u16 size
)
4844 if (igb_desc_unused(tx_ring
) >= size
)
4846 return __igb_maybe_stop_tx(tx_ring
, size
);
4849 static void igb_tx_map(struct igb_ring
*tx_ring
,
4850 struct igb_tx_buffer
*first
,
4853 struct sk_buff
*skb
= first
->skb
;
4854 struct igb_tx_buffer
*tx_buffer
;
4855 union e1000_adv_tx_desc
*tx_desc
;
4856 struct skb_frag_struct
*frag
;
4858 unsigned int data_len
, size
;
4859 u32 tx_flags
= first
->tx_flags
;
4860 u32 cmd_type
= igb_tx_cmd_type(skb
, tx_flags
);
4861 u16 i
= tx_ring
->next_to_use
;
4863 tx_desc
= IGB_TX_DESC(tx_ring
, i
);
4865 igb_tx_olinfo_status(tx_ring
, tx_desc
, tx_flags
, skb
->len
- hdr_len
);
4867 size
= skb_headlen(skb
);
4868 data_len
= skb
->data_len
;
4870 dma
= dma_map_single(tx_ring
->dev
, skb
->data
, size
, DMA_TO_DEVICE
);
4874 for (frag
= &skb_shinfo(skb
)->frags
[0];; frag
++) {
4875 if (dma_mapping_error(tx_ring
->dev
, dma
))
4878 /* record length, and DMA address */
4879 dma_unmap_len_set(tx_buffer
, len
, size
);
4880 dma_unmap_addr_set(tx_buffer
, dma
, dma
);
4882 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
4884 while (unlikely(size
> IGB_MAX_DATA_PER_TXD
)) {
4885 tx_desc
->read
.cmd_type_len
=
4886 cpu_to_le32(cmd_type
^ IGB_MAX_DATA_PER_TXD
);
4890 if (i
== tx_ring
->count
) {
4891 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
4894 tx_desc
->read
.olinfo_status
= 0;
4896 dma
+= IGB_MAX_DATA_PER_TXD
;
4897 size
-= IGB_MAX_DATA_PER_TXD
;
4899 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
4902 if (likely(!data_len
))
4905 tx_desc
->read
.cmd_type_len
= cpu_to_le32(cmd_type
^ size
);
4909 if (i
== tx_ring
->count
) {
4910 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
4913 tx_desc
->read
.olinfo_status
= 0;
4915 size
= skb_frag_size(frag
);
4918 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0,
4919 size
, DMA_TO_DEVICE
);
4921 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
4924 /* write last descriptor with RS and EOP bits */
4925 cmd_type
|= size
| IGB_TXD_DCMD
;
4926 tx_desc
->read
.cmd_type_len
= cpu_to_le32(cmd_type
);
4928 netdev_tx_sent_queue(txring_txq(tx_ring
), first
->bytecount
);
4930 /* set the timestamp */
4931 first
->time_stamp
= jiffies
;
4933 /* Force memory writes to complete before letting h/w know there
4934 * are new descriptors to fetch. (Only applicable for weak-ordered
4935 * memory model archs, such as IA-64).
4937 * We also need this memory barrier to make certain all of the
4938 * status bits have been updated before next_to_watch is written.
4942 /* set next_to_watch value indicating a packet is present */
4943 first
->next_to_watch
= tx_desc
;
4946 if (i
== tx_ring
->count
)
4949 tx_ring
->next_to_use
= i
;
4951 /* Make sure there is space in the ring for the next send. */
4952 igb_maybe_stop_tx(tx_ring
, DESC_NEEDED
);
4954 if (netif_xmit_stopped(txring_txq(tx_ring
)) || !skb
->xmit_more
) {
4955 writel(i
, tx_ring
->tail
);
4957 /* we need this if more than one processor can write to our tail
4958 * at a time, it synchronizes IO on IA64/Altix systems
4965 dev_err(tx_ring
->dev
, "TX DMA map failed\n");
4967 /* clear dma mappings for failed tx_buffer_info map */
4969 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
4970 igb_unmap_and_free_tx_resource(tx_ring
, tx_buffer
);
4971 if (tx_buffer
== first
)
4978 tx_ring
->next_to_use
= i
;
4981 netdev_tx_t
igb_xmit_frame_ring(struct sk_buff
*skb
,
4982 struct igb_ring
*tx_ring
)
4984 struct igb_tx_buffer
*first
;
4987 u16 count
= TXD_USE_COUNT(skb_headlen(skb
));
4988 __be16 protocol
= vlan_get_protocol(skb
);
4991 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
4992 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
4993 * + 2 desc gap to keep tail from touching head,
4994 * + 1 desc for context descriptor,
4995 * otherwise try next time
4997 if (NETDEV_FRAG_PAGE_MAX_SIZE
> IGB_MAX_DATA_PER_TXD
) {
5000 for (f
= 0; f
< skb_shinfo(skb
)->nr_frags
; f
++)
5001 count
+= TXD_USE_COUNT(skb_shinfo(skb
)->frags
[f
].size
);
5003 count
+= skb_shinfo(skb
)->nr_frags
;
5006 if (igb_maybe_stop_tx(tx_ring
, count
+ 3)) {
5007 /* this is a hard error */
5008 return NETDEV_TX_BUSY
;
5011 /* record the location of the first descriptor for this packet */
5012 first
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_use
];
5014 first
->bytecount
= skb
->len
;
5015 first
->gso_segs
= 1;
5017 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_HW_TSTAMP
)) {
5018 struct igb_adapter
*adapter
= netdev_priv(tx_ring
->netdev
);
5020 if (!test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS
,
5022 skb_shinfo(skb
)->tx_flags
|= SKBTX_IN_PROGRESS
;
5023 tx_flags
|= IGB_TX_FLAGS_TSTAMP
;
5025 adapter
->ptp_tx_skb
= skb_get(skb
);
5026 adapter
->ptp_tx_start
= jiffies
;
5027 if (adapter
->hw
.mac
.type
== e1000_82576
)
5028 schedule_work(&adapter
->ptp_tx_work
);
5032 skb_tx_timestamp(skb
);
5034 if (vlan_tx_tag_present(skb
)) {
5035 tx_flags
|= IGB_TX_FLAGS_VLAN
;
5036 tx_flags
|= (vlan_tx_tag_get(skb
) << IGB_TX_FLAGS_VLAN_SHIFT
);
5039 /* record initial flags and protocol */
5040 first
->tx_flags
= tx_flags
;
5041 first
->protocol
= protocol
;
5043 tso
= igb_tso(tx_ring
, first
, &hdr_len
);
5047 igb_tx_csum(tx_ring
, first
);
5049 igb_tx_map(tx_ring
, first
, hdr_len
);
5051 return NETDEV_TX_OK
;
5054 igb_unmap_and_free_tx_resource(tx_ring
, first
);
5056 return NETDEV_TX_OK
;
5059 static inline struct igb_ring
*igb_tx_queue_mapping(struct igb_adapter
*adapter
,
5060 struct sk_buff
*skb
)
5062 unsigned int r_idx
= skb
->queue_mapping
;
5064 if (r_idx
>= adapter
->num_tx_queues
)
5065 r_idx
= r_idx
% adapter
->num_tx_queues
;
5067 return adapter
->tx_ring
[r_idx
];
5070 static netdev_tx_t
igb_xmit_frame(struct sk_buff
*skb
,
5071 struct net_device
*netdev
)
5073 struct igb_adapter
*adapter
= netdev_priv(netdev
);
5075 if (test_bit(__IGB_DOWN
, &adapter
->state
)) {
5076 dev_kfree_skb_any(skb
);
5077 return NETDEV_TX_OK
;
5080 if (skb
->len
<= 0) {
5081 dev_kfree_skb_any(skb
);
5082 return NETDEV_TX_OK
;
5085 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
5086 * in order to meet this minimum size requirement.
5088 if (unlikely(skb
->len
< 17)) {
5089 if (skb_pad(skb
, 17 - skb
->len
))
5090 return NETDEV_TX_OK
;
5092 skb_set_tail_pointer(skb
, 17);
5095 return igb_xmit_frame_ring(skb
, igb_tx_queue_mapping(adapter
, skb
));
5099 * igb_tx_timeout - Respond to a Tx Hang
5100 * @netdev: network interface device structure
5102 static void igb_tx_timeout(struct net_device
*netdev
)
5104 struct igb_adapter
*adapter
= netdev_priv(netdev
);
5105 struct e1000_hw
*hw
= &adapter
->hw
;
5107 /* Do the reset outside of interrupt context */
5108 adapter
->tx_timeout_count
++;
5110 if (hw
->mac
.type
>= e1000_82580
)
5111 hw
->dev_spec
._82575
.global_device_reset
= true;
5113 schedule_work(&adapter
->reset_task
);
5115 (adapter
->eims_enable_mask
& ~adapter
->eims_other
));
5118 static void igb_reset_task(struct work_struct
*work
)
5120 struct igb_adapter
*adapter
;
5121 adapter
= container_of(work
, struct igb_adapter
, reset_task
);
5124 netdev_err(adapter
->netdev
, "Reset adapter\n");
5125 igb_reinit_locked(adapter
);
5129 * igb_get_stats64 - Get System Network Statistics
5130 * @netdev: network interface device structure
5131 * @stats: rtnl_link_stats64 pointer
5133 static struct rtnl_link_stats64
*igb_get_stats64(struct net_device
*netdev
,
5134 struct rtnl_link_stats64
*stats
)
5136 struct igb_adapter
*adapter
= netdev_priv(netdev
);
5138 spin_lock(&adapter
->stats64_lock
);
5139 igb_update_stats(adapter
, &adapter
->stats64
);
5140 memcpy(stats
, &adapter
->stats64
, sizeof(*stats
));
5141 spin_unlock(&adapter
->stats64_lock
);
5147 * igb_change_mtu - Change the Maximum Transfer Unit
5148 * @netdev: network interface device structure
5149 * @new_mtu: new value for maximum frame size
5151 * Returns 0 on success, negative on failure
5153 static int igb_change_mtu(struct net_device
*netdev
, int new_mtu
)
5155 struct igb_adapter
*adapter
= netdev_priv(netdev
);
5156 struct pci_dev
*pdev
= adapter
->pdev
;
5157 int max_frame
= new_mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+ VLAN_HLEN
;
5159 if ((new_mtu
< 68) || (max_frame
> MAX_JUMBO_FRAME_SIZE
)) {
5160 dev_err(&pdev
->dev
, "Invalid MTU setting\n");
5164 #define MAX_STD_JUMBO_FRAME_SIZE 9238
5165 if (max_frame
> MAX_STD_JUMBO_FRAME_SIZE
) {
5166 dev_err(&pdev
->dev
, "MTU > 9216 not supported.\n");
5170 /* adjust max frame to be at least the size of a standard frame */
5171 if (max_frame
< (ETH_FRAME_LEN
+ ETH_FCS_LEN
))
5172 max_frame
= ETH_FRAME_LEN
+ ETH_FCS_LEN
;
5174 while (test_and_set_bit(__IGB_RESETTING
, &adapter
->state
))
5175 usleep_range(1000, 2000);
5177 /* igb_down has a dependency on max_frame_size */
5178 adapter
->max_frame_size
= max_frame
;
5180 if (netif_running(netdev
))
5183 dev_info(&pdev
->dev
, "changing MTU from %d to %d\n",
5184 netdev
->mtu
, new_mtu
);
5185 netdev
->mtu
= new_mtu
;
5187 if (netif_running(netdev
))
5192 clear_bit(__IGB_RESETTING
, &adapter
->state
);
5198 * igb_update_stats - Update the board statistics counters
5199 * @adapter: board private structure
5201 void igb_update_stats(struct igb_adapter
*adapter
,
5202 struct rtnl_link_stats64
*net_stats
)
5204 struct e1000_hw
*hw
= &adapter
->hw
;
5205 struct pci_dev
*pdev
= adapter
->pdev
;
5210 u64 _bytes
, _packets
;
5212 /* Prevent stats update while adapter is being reset, or if the pci
5213 * connection is down.
5215 if (adapter
->link_speed
== 0)
5217 if (pci_channel_offline(pdev
))
5224 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
5225 struct igb_ring
*ring
= adapter
->rx_ring
[i
];
5226 u32 rqdpc
= rd32(E1000_RQDPC(i
));
5227 if (hw
->mac
.type
>= e1000_i210
)
5228 wr32(E1000_RQDPC(i
), 0);
5231 ring
->rx_stats
.drops
+= rqdpc
;
5232 net_stats
->rx_fifo_errors
+= rqdpc
;
5236 start
= u64_stats_fetch_begin_irq(&ring
->rx_syncp
);
5237 _bytes
= ring
->rx_stats
.bytes
;
5238 _packets
= ring
->rx_stats
.packets
;
5239 } while (u64_stats_fetch_retry_irq(&ring
->rx_syncp
, start
));
5241 packets
+= _packets
;
5244 net_stats
->rx_bytes
= bytes
;
5245 net_stats
->rx_packets
= packets
;
5249 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
5250 struct igb_ring
*ring
= adapter
->tx_ring
[i
];
5252 start
= u64_stats_fetch_begin_irq(&ring
->tx_syncp
);
5253 _bytes
= ring
->tx_stats
.bytes
;
5254 _packets
= ring
->tx_stats
.packets
;
5255 } while (u64_stats_fetch_retry_irq(&ring
->tx_syncp
, start
));
5257 packets
+= _packets
;
5259 net_stats
->tx_bytes
= bytes
;
5260 net_stats
->tx_packets
= packets
;
5263 /* read stats registers */
5264 adapter
->stats
.crcerrs
+= rd32(E1000_CRCERRS
);
5265 adapter
->stats
.gprc
+= rd32(E1000_GPRC
);
5266 adapter
->stats
.gorc
+= rd32(E1000_GORCL
);
5267 rd32(E1000_GORCH
); /* clear GORCL */
5268 adapter
->stats
.bprc
+= rd32(E1000_BPRC
);
5269 adapter
->stats
.mprc
+= rd32(E1000_MPRC
);
5270 adapter
->stats
.roc
+= rd32(E1000_ROC
);
5272 adapter
->stats
.prc64
+= rd32(E1000_PRC64
);
5273 adapter
->stats
.prc127
+= rd32(E1000_PRC127
);
5274 adapter
->stats
.prc255
+= rd32(E1000_PRC255
);
5275 adapter
->stats
.prc511
+= rd32(E1000_PRC511
);
5276 adapter
->stats
.prc1023
+= rd32(E1000_PRC1023
);
5277 adapter
->stats
.prc1522
+= rd32(E1000_PRC1522
);
5278 adapter
->stats
.symerrs
+= rd32(E1000_SYMERRS
);
5279 adapter
->stats
.sec
+= rd32(E1000_SEC
);
5281 mpc
= rd32(E1000_MPC
);
5282 adapter
->stats
.mpc
+= mpc
;
5283 net_stats
->rx_fifo_errors
+= mpc
;
5284 adapter
->stats
.scc
+= rd32(E1000_SCC
);
5285 adapter
->stats
.ecol
+= rd32(E1000_ECOL
);
5286 adapter
->stats
.mcc
+= rd32(E1000_MCC
);
5287 adapter
->stats
.latecol
+= rd32(E1000_LATECOL
);
5288 adapter
->stats
.dc
+= rd32(E1000_DC
);
5289 adapter
->stats
.rlec
+= rd32(E1000_RLEC
);
5290 adapter
->stats
.xonrxc
+= rd32(E1000_XONRXC
);
5291 adapter
->stats
.xontxc
+= rd32(E1000_XONTXC
);
5292 adapter
->stats
.xoffrxc
+= rd32(E1000_XOFFRXC
);
5293 adapter
->stats
.xofftxc
+= rd32(E1000_XOFFTXC
);
5294 adapter
->stats
.fcruc
+= rd32(E1000_FCRUC
);
5295 adapter
->stats
.gptc
+= rd32(E1000_GPTC
);
5296 adapter
->stats
.gotc
+= rd32(E1000_GOTCL
);
5297 rd32(E1000_GOTCH
); /* clear GOTCL */
5298 adapter
->stats
.rnbc
+= rd32(E1000_RNBC
);
5299 adapter
->stats
.ruc
+= rd32(E1000_RUC
);
5300 adapter
->stats
.rfc
+= rd32(E1000_RFC
);
5301 adapter
->stats
.rjc
+= rd32(E1000_RJC
);
5302 adapter
->stats
.tor
+= rd32(E1000_TORH
);
5303 adapter
->stats
.tot
+= rd32(E1000_TOTH
);
5304 adapter
->stats
.tpr
+= rd32(E1000_TPR
);
5306 adapter
->stats
.ptc64
+= rd32(E1000_PTC64
);
5307 adapter
->stats
.ptc127
+= rd32(E1000_PTC127
);
5308 adapter
->stats
.ptc255
+= rd32(E1000_PTC255
);
5309 adapter
->stats
.ptc511
+= rd32(E1000_PTC511
);
5310 adapter
->stats
.ptc1023
+= rd32(E1000_PTC1023
);
5311 adapter
->stats
.ptc1522
+= rd32(E1000_PTC1522
);
5313 adapter
->stats
.mptc
+= rd32(E1000_MPTC
);
5314 adapter
->stats
.bptc
+= rd32(E1000_BPTC
);
5316 adapter
->stats
.tpt
+= rd32(E1000_TPT
);
5317 adapter
->stats
.colc
+= rd32(E1000_COLC
);
5319 adapter
->stats
.algnerrc
+= rd32(E1000_ALGNERRC
);
5320 /* read internal phy specific stats */
5321 reg
= rd32(E1000_CTRL_EXT
);
5322 if (!(reg
& E1000_CTRL_EXT_LINK_MODE_MASK
)) {
5323 adapter
->stats
.rxerrc
+= rd32(E1000_RXERRC
);
5325 /* this stat has invalid values on i210/i211 */
5326 if ((hw
->mac
.type
!= e1000_i210
) &&
5327 (hw
->mac
.type
!= e1000_i211
))
5328 adapter
->stats
.tncrs
+= rd32(E1000_TNCRS
);
5331 adapter
->stats
.tsctc
+= rd32(E1000_TSCTC
);
5332 adapter
->stats
.tsctfc
+= rd32(E1000_TSCTFC
);
5334 adapter
->stats
.iac
+= rd32(E1000_IAC
);
5335 adapter
->stats
.icrxoc
+= rd32(E1000_ICRXOC
);
5336 adapter
->stats
.icrxptc
+= rd32(E1000_ICRXPTC
);
5337 adapter
->stats
.icrxatc
+= rd32(E1000_ICRXATC
);
5338 adapter
->stats
.ictxptc
+= rd32(E1000_ICTXPTC
);
5339 adapter
->stats
.ictxatc
+= rd32(E1000_ICTXATC
);
5340 adapter
->stats
.ictxqec
+= rd32(E1000_ICTXQEC
);
5341 adapter
->stats
.ictxqmtc
+= rd32(E1000_ICTXQMTC
);
5342 adapter
->stats
.icrxdmtc
+= rd32(E1000_ICRXDMTC
);
5344 /* Fill out the OS statistics structure */
5345 net_stats
->multicast
= adapter
->stats
.mprc
;
5346 net_stats
->collisions
= adapter
->stats
.colc
;
5350 /* RLEC on some newer hardware can be incorrect so build
5351 * our own version based on RUC and ROC
5353 net_stats
->rx_errors
= adapter
->stats
.rxerrc
+
5354 adapter
->stats
.crcerrs
+ adapter
->stats
.algnerrc
+
5355 adapter
->stats
.ruc
+ adapter
->stats
.roc
+
5356 adapter
->stats
.cexterr
;
5357 net_stats
->rx_length_errors
= adapter
->stats
.ruc
+
5359 net_stats
->rx_crc_errors
= adapter
->stats
.crcerrs
;
5360 net_stats
->rx_frame_errors
= adapter
->stats
.algnerrc
;
5361 net_stats
->rx_missed_errors
= adapter
->stats
.mpc
;
5364 net_stats
->tx_errors
= adapter
->stats
.ecol
+
5365 adapter
->stats
.latecol
;
5366 net_stats
->tx_aborted_errors
= adapter
->stats
.ecol
;
5367 net_stats
->tx_window_errors
= adapter
->stats
.latecol
;
5368 net_stats
->tx_carrier_errors
= adapter
->stats
.tncrs
;
5370 /* Tx Dropped needs to be maintained elsewhere */
5372 /* Management Stats */
5373 adapter
->stats
.mgptc
+= rd32(E1000_MGTPTC
);
5374 adapter
->stats
.mgprc
+= rd32(E1000_MGTPRC
);
5375 adapter
->stats
.mgpdc
+= rd32(E1000_MGTPDC
);
5378 reg
= rd32(E1000_MANC
);
5379 if (reg
& E1000_MANC_EN_BMC2OS
) {
5380 adapter
->stats
.o2bgptc
+= rd32(E1000_O2BGPTC
);
5381 adapter
->stats
.o2bspc
+= rd32(E1000_O2BSPC
);
5382 adapter
->stats
.b2ospc
+= rd32(E1000_B2OSPC
);
5383 adapter
->stats
.b2ogprc
+= rd32(E1000_B2OGPRC
);
5387 static irqreturn_t
igb_msix_other(int irq
, void *data
)
5389 struct igb_adapter
*adapter
= data
;
5390 struct e1000_hw
*hw
= &adapter
->hw
;
5391 u32 icr
= rd32(E1000_ICR
);
5392 /* reading ICR causes bit 31 of EICR to be cleared */
5394 if (icr
& E1000_ICR_DRSTA
)
5395 schedule_work(&adapter
->reset_task
);
5397 if (icr
& E1000_ICR_DOUTSYNC
) {
5398 /* HW is reporting DMA is out of sync */
5399 adapter
->stats
.doosync
++;
5400 /* The DMA Out of Sync is also indication of a spoof event
5401 * in IOV mode. Check the Wrong VM Behavior register to
5402 * see if it is really a spoof event.
5404 igb_check_wvbr(adapter
);
5407 /* Check for a mailbox event */
5408 if (icr
& E1000_ICR_VMMB
)
5409 igb_msg_task(adapter
);
5411 if (icr
& E1000_ICR_LSC
) {
5412 hw
->mac
.get_link_status
= 1;
5413 /* guard against interrupt when we're going down */
5414 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
5415 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
5418 if (icr
& E1000_ICR_TS
) {
5419 u32 tsicr
= rd32(E1000_TSICR
);
5421 if (tsicr
& E1000_TSICR_TXTS
) {
5422 /* acknowledge the interrupt */
5423 wr32(E1000_TSICR
, E1000_TSICR_TXTS
);
5424 /* retrieve hardware timestamp */
5425 schedule_work(&adapter
->ptp_tx_work
);
5429 wr32(E1000_EIMS
, adapter
->eims_other
);
5434 static void igb_write_itr(struct igb_q_vector
*q_vector
)
5436 struct igb_adapter
*adapter
= q_vector
->adapter
;
5437 u32 itr_val
= q_vector
->itr_val
& 0x7FFC;
5439 if (!q_vector
->set_itr
)
5445 if (adapter
->hw
.mac
.type
== e1000_82575
)
5446 itr_val
|= itr_val
<< 16;
5448 itr_val
|= E1000_EITR_CNT_IGNR
;
5450 writel(itr_val
, q_vector
->itr_register
);
5451 q_vector
->set_itr
= 0;
5454 static irqreturn_t
igb_msix_ring(int irq
, void *data
)
5456 struct igb_q_vector
*q_vector
= data
;
5458 /* Write the ITR value calculated from the previous interrupt. */
5459 igb_write_itr(q_vector
);
5461 napi_schedule(&q_vector
->napi
);
5466 #ifdef CONFIG_IGB_DCA
5467 static void igb_update_tx_dca(struct igb_adapter
*adapter
,
5468 struct igb_ring
*tx_ring
,
5471 struct e1000_hw
*hw
= &adapter
->hw
;
5472 u32 txctrl
= dca3_get_tag(tx_ring
->dev
, cpu
);
5474 if (hw
->mac
.type
!= e1000_82575
)
5475 txctrl
<<= E1000_DCA_TXCTRL_CPUID_SHIFT
;
5477 /* We can enable relaxed ordering for reads, but not writes when
5478 * DCA is enabled. This is due to a known issue in some chipsets
5479 * which will cause the DCA tag to be cleared.
5481 txctrl
|= E1000_DCA_TXCTRL_DESC_RRO_EN
|
5482 E1000_DCA_TXCTRL_DATA_RRO_EN
|
5483 E1000_DCA_TXCTRL_DESC_DCA_EN
;
5485 wr32(E1000_DCA_TXCTRL(tx_ring
->reg_idx
), txctrl
);
5488 static void igb_update_rx_dca(struct igb_adapter
*adapter
,
5489 struct igb_ring
*rx_ring
,
5492 struct e1000_hw
*hw
= &adapter
->hw
;
5493 u32 rxctrl
= dca3_get_tag(&adapter
->pdev
->dev
, cpu
);
5495 if (hw
->mac
.type
!= e1000_82575
)
5496 rxctrl
<<= E1000_DCA_RXCTRL_CPUID_SHIFT
;
5498 /* We can enable relaxed ordering for reads, but not writes when
5499 * DCA is enabled. This is due to a known issue in some chipsets
5500 * which will cause the DCA tag to be cleared.
5502 rxctrl
|= E1000_DCA_RXCTRL_DESC_RRO_EN
|
5503 E1000_DCA_RXCTRL_DESC_DCA_EN
;
5505 wr32(E1000_DCA_RXCTRL(rx_ring
->reg_idx
), rxctrl
);
5508 static void igb_update_dca(struct igb_q_vector
*q_vector
)
5510 struct igb_adapter
*adapter
= q_vector
->adapter
;
5511 int cpu
= get_cpu();
5513 if (q_vector
->cpu
== cpu
)
5516 if (q_vector
->tx
.ring
)
5517 igb_update_tx_dca(adapter
, q_vector
->tx
.ring
, cpu
);
5519 if (q_vector
->rx
.ring
)
5520 igb_update_rx_dca(adapter
, q_vector
->rx
.ring
, cpu
);
5522 q_vector
->cpu
= cpu
;
5527 static void igb_setup_dca(struct igb_adapter
*adapter
)
5529 struct e1000_hw
*hw
= &adapter
->hw
;
5532 if (!(adapter
->flags
& IGB_FLAG_DCA_ENABLED
))
5535 /* Always use CB2 mode, difference is masked in the CB driver. */
5536 wr32(E1000_DCA_CTRL
, E1000_DCA_CTRL_DCA_MODE_CB2
);
5538 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
5539 adapter
->q_vector
[i
]->cpu
= -1;
5540 igb_update_dca(adapter
->q_vector
[i
]);
5544 static int __igb_notify_dca(struct device
*dev
, void *data
)
5546 struct net_device
*netdev
= dev_get_drvdata(dev
);
5547 struct igb_adapter
*adapter
= netdev_priv(netdev
);
5548 struct pci_dev
*pdev
= adapter
->pdev
;
5549 struct e1000_hw
*hw
= &adapter
->hw
;
5550 unsigned long event
= *(unsigned long *)data
;
5553 case DCA_PROVIDER_ADD
:
5554 /* if already enabled, don't do it again */
5555 if (adapter
->flags
& IGB_FLAG_DCA_ENABLED
)
5557 if (dca_add_requester(dev
) == 0) {
5558 adapter
->flags
|= IGB_FLAG_DCA_ENABLED
;
5559 dev_info(&pdev
->dev
, "DCA enabled\n");
5560 igb_setup_dca(adapter
);
5563 /* Fall Through since DCA is disabled. */
5564 case DCA_PROVIDER_REMOVE
:
5565 if (adapter
->flags
& IGB_FLAG_DCA_ENABLED
) {
5566 /* without this a class_device is left
5567 * hanging around in the sysfs model
5569 dca_remove_requester(dev
);
5570 dev_info(&pdev
->dev
, "DCA disabled\n");
5571 adapter
->flags
&= ~IGB_FLAG_DCA_ENABLED
;
5572 wr32(E1000_DCA_CTRL
, E1000_DCA_CTRL_DCA_MODE_DISABLE
);
5580 static int igb_notify_dca(struct notifier_block
*nb
, unsigned long event
,
5585 ret_val
= driver_for_each_device(&igb_driver
.driver
, NULL
, &event
,
5588 return ret_val
? NOTIFY_BAD
: NOTIFY_DONE
;
5590 #endif /* CONFIG_IGB_DCA */
5592 #ifdef CONFIG_PCI_IOV
5593 static int igb_vf_configure(struct igb_adapter
*adapter
, int vf
)
5595 unsigned char mac_addr
[ETH_ALEN
];
5597 eth_zero_addr(mac_addr
);
5598 igb_set_vf_mac(adapter
, vf
, mac_addr
);
5600 /* By default spoof check is enabled for all VFs */
5601 adapter
->vf_data
[vf
].spoofchk_enabled
= true;
5607 static void igb_ping_all_vfs(struct igb_adapter
*adapter
)
5609 struct e1000_hw
*hw
= &adapter
->hw
;
5613 for (i
= 0 ; i
< adapter
->vfs_allocated_count
; i
++) {
5614 ping
= E1000_PF_CONTROL_MSG
;
5615 if (adapter
->vf_data
[i
].flags
& IGB_VF_FLAG_CTS
)
5616 ping
|= E1000_VT_MSGTYPE_CTS
;
5617 igb_write_mbx(hw
, &ping
, 1, i
);
5621 static int igb_set_vf_promisc(struct igb_adapter
*adapter
, u32
*msgbuf
, u32 vf
)
5623 struct e1000_hw
*hw
= &adapter
->hw
;
5624 u32 vmolr
= rd32(E1000_VMOLR(vf
));
5625 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5627 vf_data
->flags
&= ~(IGB_VF_FLAG_UNI_PROMISC
|
5628 IGB_VF_FLAG_MULTI_PROMISC
);
5629 vmolr
&= ~(E1000_VMOLR_ROPE
| E1000_VMOLR_ROMPE
| E1000_VMOLR_MPME
);
5631 if (*msgbuf
& E1000_VF_SET_PROMISC_MULTICAST
) {
5632 vmolr
|= E1000_VMOLR_MPME
;
5633 vf_data
->flags
|= IGB_VF_FLAG_MULTI_PROMISC
;
5634 *msgbuf
&= ~E1000_VF_SET_PROMISC_MULTICAST
;
5636 /* if we have hashes and we are clearing a multicast promisc
5637 * flag we need to write the hashes to the MTA as this step
5638 * was previously skipped
5640 if (vf_data
->num_vf_mc_hashes
> 30) {
5641 vmolr
|= E1000_VMOLR_MPME
;
5642 } else if (vf_data
->num_vf_mc_hashes
) {
5645 vmolr
|= E1000_VMOLR_ROMPE
;
5646 for (j
= 0; j
< vf_data
->num_vf_mc_hashes
; j
++)
5647 igb_mta_set(hw
, vf_data
->vf_mc_hashes
[j
]);
5651 wr32(E1000_VMOLR(vf
), vmolr
);
5653 /* there are flags left unprocessed, likely not supported */
5654 if (*msgbuf
& E1000_VT_MSGINFO_MASK
)
5660 static int igb_set_vf_multicasts(struct igb_adapter
*adapter
,
5661 u32
*msgbuf
, u32 vf
)
5663 int n
= (msgbuf
[0] & E1000_VT_MSGINFO_MASK
) >> E1000_VT_MSGINFO_SHIFT
;
5664 u16
*hash_list
= (u16
*)&msgbuf
[1];
5665 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5668 /* salt away the number of multicast addresses assigned
5669 * to this VF for later use to restore when the PF multi cast
5672 vf_data
->num_vf_mc_hashes
= n
;
5674 /* only up to 30 hash values supported */
5678 /* store the hashes for later use */
5679 for (i
= 0; i
< n
; i
++)
5680 vf_data
->vf_mc_hashes
[i
] = hash_list
[i
];
5682 /* Flush and reset the mta with the new values */
5683 igb_set_rx_mode(adapter
->netdev
);
5688 static void igb_restore_vf_multicasts(struct igb_adapter
*adapter
)
5690 struct e1000_hw
*hw
= &adapter
->hw
;
5691 struct vf_data_storage
*vf_data
;
5694 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++) {
5695 u32 vmolr
= rd32(E1000_VMOLR(i
));
5697 vmolr
&= ~(E1000_VMOLR_ROMPE
| E1000_VMOLR_MPME
);
5699 vf_data
= &adapter
->vf_data
[i
];
5701 if ((vf_data
->num_vf_mc_hashes
> 30) ||
5702 (vf_data
->flags
& IGB_VF_FLAG_MULTI_PROMISC
)) {
5703 vmolr
|= E1000_VMOLR_MPME
;
5704 } else if (vf_data
->num_vf_mc_hashes
) {
5705 vmolr
|= E1000_VMOLR_ROMPE
;
5706 for (j
= 0; j
< vf_data
->num_vf_mc_hashes
; j
++)
5707 igb_mta_set(hw
, vf_data
->vf_mc_hashes
[j
]);
5709 wr32(E1000_VMOLR(i
), vmolr
);
5713 static void igb_clear_vf_vfta(struct igb_adapter
*adapter
, u32 vf
)
5715 struct e1000_hw
*hw
= &adapter
->hw
;
5716 u32 pool_mask
, reg
, vid
;
5719 pool_mask
= 1 << (E1000_VLVF_POOLSEL_SHIFT
+ vf
);
5721 /* Find the vlan filter for this id */
5722 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5723 reg
= rd32(E1000_VLVF(i
));
5725 /* remove the vf from the pool */
5728 /* if pool is empty then remove entry from vfta */
5729 if (!(reg
& E1000_VLVF_POOLSEL_MASK
) &&
5730 (reg
& E1000_VLVF_VLANID_ENABLE
)) {
5732 vid
= reg
& E1000_VLVF_VLANID_MASK
;
5733 igb_vfta_set(hw
, vid
, false);
5736 wr32(E1000_VLVF(i
), reg
);
5739 adapter
->vf_data
[vf
].vlans_enabled
= 0;
5742 static s32
igb_vlvf_set(struct igb_adapter
*adapter
, u32 vid
, bool add
, u32 vf
)
5744 struct e1000_hw
*hw
= &adapter
->hw
;
5747 /* The vlvf table only exists on 82576 hardware and newer */
5748 if (hw
->mac
.type
< e1000_82576
)
5751 /* we only need to do this if VMDq is enabled */
5752 if (!adapter
->vfs_allocated_count
)
5755 /* Find the vlan filter for this id */
5756 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5757 reg
= rd32(E1000_VLVF(i
));
5758 if ((reg
& E1000_VLVF_VLANID_ENABLE
) &&
5759 vid
== (reg
& E1000_VLVF_VLANID_MASK
))
5764 if (i
== E1000_VLVF_ARRAY_SIZE
) {
5765 /* Did not find a matching VLAN ID entry that was
5766 * enabled. Search for a free filter entry, i.e.
5767 * one without the enable bit set
5769 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5770 reg
= rd32(E1000_VLVF(i
));
5771 if (!(reg
& E1000_VLVF_VLANID_ENABLE
))
5775 if (i
< E1000_VLVF_ARRAY_SIZE
) {
5776 /* Found an enabled/available entry */
5777 reg
|= 1 << (E1000_VLVF_POOLSEL_SHIFT
+ vf
);
5779 /* if !enabled we need to set this up in vfta */
5780 if (!(reg
& E1000_VLVF_VLANID_ENABLE
)) {
5781 /* add VID to filter table */
5782 igb_vfta_set(hw
, vid
, true);
5783 reg
|= E1000_VLVF_VLANID_ENABLE
;
5785 reg
&= ~E1000_VLVF_VLANID_MASK
;
5787 wr32(E1000_VLVF(i
), reg
);
5789 /* do not modify RLPML for PF devices */
5790 if (vf
>= adapter
->vfs_allocated_count
)
5793 if (!adapter
->vf_data
[vf
].vlans_enabled
) {
5796 reg
= rd32(E1000_VMOLR(vf
));
5797 size
= reg
& E1000_VMOLR_RLPML_MASK
;
5799 reg
&= ~E1000_VMOLR_RLPML_MASK
;
5801 wr32(E1000_VMOLR(vf
), reg
);
5804 adapter
->vf_data
[vf
].vlans_enabled
++;
5807 if (i
< E1000_VLVF_ARRAY_SIZE
) {
5808 /* remove vf from the pool */
5809 reg
&= ~(1 << (E1000_VLVF_POOLSEL_SHIFT
+ vf
));
5810 /* if pool is empty then remove entry from vfta */
5811 if (!(reg
& E1000_VLVF_POOLSEL_MASK
)) {
5813 igb_vfta_set(hw
, vid
, false);
5815 wr32(E1000_VLVF(i
), reg
);
5817 /* do not modify RLPML for PF devices */
5818 if (vf
>= adapter
->vfs_allocated_count
)
5821 adapter
->vf_data
[vf
].vlans_enabled
--;
5822 if (!adapter
->vf_data
[vf
].vlans_enabled
) {
5825 reg
= rd32(E1000_VMOLR(vf
));
5826 size
= reg
& E1000_VMOLR_RLPML_MASK
;
5828 reg
&= ~E1000_VMOLR_RLPML_MASK
;
5830 wr32(E1000_VMOLR(vf
), reg
);
5837 static void igb_set_vmvir(struct igb_adapter
*adapter
, u32 vid
, u32 vf
)
5839 struct e1000_hw
*hw
= &adapter
->hw
;
5842 wr32(E1000_VMVIR(vf
), (vid
| E1000_VMVIR_VLANA_DEFAULT
));
5844 wr32(E1000_VMVIR(vf
), 0);
5847 static int igb_ndo_set_vf_vlan(struct net_device
*netdev
,
5848 int vf
, u16 vlan
, u8 qos
)
5851 struct igb_adapter
*adapter
= netdev_priv(netdev
);
5853 if ((vf
>= adapter
->vfs_allocated_count
) || (vlan
> 4095) || (qos
> 7))
5856 err
= igb_vlvf_set(adapter
, vlan
, !!vlan
, vf
);
5859 igb_set_vmvir(adapter
, vlan
| (qos
<< VLAN_PRIO_SHIFT
), vf
);
5860 igb_set_vmolr(adapter
, vf
, !vlan
);
5861 adapter
->vf_data
[vf
].pf_vlan
= vlan
;
5862 adapter
->vf_data
[vf
].pf_qos
= qos
;
5863 dev_info(&adapter
->pdev
->dev
,
5864 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan
, qos
, vf
);
5865 if (test_bit(__IGB_DOWN
, &adapter
->state
)) {
5866 dev_warn(&adapter
->pdev
->dev
,
5867 "The VF VLAN has been set, but the PF device is not up.\n");
5868 dev_warn(&adapter
->pdev
->dev
,
5869 "Bring the PF device up before attempting to use the VF device.\n");
5872 igb_vlvf_set(adapter
, adapter
->vf_data
[vf
].pf_vlan
,
5874 igb_set_vmvir(adapter
, vlan
, vf
);
5875 igb_set_vmolr(adapter
, vf
, true);
5876 adapter
->vf_data
[vf
].pf_vlan
= 0;
5877 adapter
->vf_data
[vf
].pf_qos
= 0;
5883 static int igb_find_vlvf_entry(struct igb_adapter
*adapter
, int vid
)
5885 struct e1000_hw
*hw
= &adapter
->hw
;
5889 /* Find the vlan filter for this id */
5890 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5891 reg
= rd32(E1000_VLVF(i
));
5892 if ((reg
& E1000_VLVF_VLANID_ENABLE
) &&
5893 vid
== (reg
& E1000_VLVF_VLANID_MASK
))
5897 if (i
>= E1000_VLVF_ARRAY_SIZE
)
5903 static int igb_set_vf_vlan(struct igb_adapter
*adapter
, u32
*msgbuf
, u32 vf
)
5905 struct e1000_hw
*hw
= &adapter
->hw
;
5906 int add
= (msgbuf
[0] & E1000_VT_MSGINFO_MASK
) >> E1000_VT_MSGINFO_SHIFT
;
5907 int vid
= (msgbuf
[1] & E1000_VLVF_VLANID_MASK
);
5910 /* If in promiscuous mode we need to make sure the PF also has
5911 * the VLAN filter set.
5913 if (add
&& (adapter
->netdev
->flags
& IFF_PROMISC
))
5914 err
= igb_vlvf_set(adapter
, vid
, add
,
5915 adapter
->vfs_allocated_count
);
5919 err
= igb_vlvf_set(adapter
, vid
, add
, vf
);
5924 /* Go through all the checks to see if the VLAN filter should
5925 * be wiped completely.
5927 if (!add
&& (adapter
->netdev
->flags
& IFF_PROMISC
)) {
5929 int regndx
= igb_find_vlvf_entry(adapter
, vid
);
5933 /* See if any other pools are set for this VLAN filter
5934 * entry other than the PF.
5936 vlvf
= bits
= rd32(E1000_VLVF(regndx
));
5937 bits
&= 1 << (E1000_VLVF_POOLSEL_SHIFT
+
5938 adapter
->vfs_allocated_count
);
5939 /* If the filter was removed then ensure PF pool bit
5940 * is cleared if the PF only added itself to the pool
5941 * because the PF is in promiscuous mode.
5943 if ((vlvf
& VLAN_VID_MASK
) == vid
&&
5944 !test_bit(vid
, adapter
->active_vlans
) &&
5946 igb_vlvf_set(adapter
, vid
, add
,
5947 adapter
->vfs_allocated_count
);
5954 static inline void igb_vf_reset(struct igb_adapter
*adapter
, u32 vf
)
5956 /* clear flags - except flag that indicates PF has set the MAC */
5957 adapter
->vf_data
[vf
].flags
&= IGB_VF_FLAG_PF_SET_MAC
;
5958 adapter
->vf_data
[vf
].last_nack
= jiffies
;
5960 /* reset offloads to defaults */
5961 igb_set_vmolr(adapter
, vf
, true);
5963 /* reset vlans for device */
5964 igb_clear_vf_vfta(adapter
, vf
);
5965 if (adapter
->vf_data
[vf
].pf_vlan
)
5966 igb_ndo_set_vf_vlan(adapter
->netdev
, vf
,
5967 adapter
->vf_data
[vf
].pf_vlan
,
5968 adapter
->vf_data
[vf
].pf_qos
);
5970 igb_clear_vf_vfta(adapter
, vf
);
5972 /* reset multicast table array for vf */
5973 adapter
->vf_data
[vf
].num_vf_mc_hashes
= 0;
5975 /* Flush and reset the mta with the new values */
5976 igb_set_rx_mode(adapter
->netdev
);
5979 static void igb_vf_reset_event(struct igb_adapter
*adapter
, u32 vf
)
5981 unsigned char *vf_mac
= adapter
->vf_data
[vf
].vf_mac_addresses
;
5983 /* clear mac address as we were hotplug removed/added */
5984 if (!(adapter
->vf_data
[vf
].flags
& IGB_VF_FLAG_PF_SET_MAC
))
5985 eth_zero_addr(vf_mac
);
5987 /* process remaining reset events */
5988 igb_vf_reset(adapter
, vf
);
5991 static void igb_vf_reset_msg(struct igb_adapter
*adapter
, u32 vf
)
5993 struct e1000_hw
*hw
= &adapter
->hw
;
5994 unsigned char *vf_mac
= adapter
->vf_data
[vf
].vf_mac_addresses
;
5995 int rar_entry
= hw
->mac
.rar_entry_count
- (vf
+ 1);
5997 u8
*addr
= (u8
*)(&msgbuf
[1]);
5999 /* process all the same items cleared in a function level reset */
6000 igb_vf_reset(adapter
, vf
);
6002 /* set vf mac address */
6003 igb_rar_set_qsel(adapter
, vf_mac
, rar_entry
, vf
);
6005 /* enable transmit and receive for vf */
6006 reg
= rd32(E1000_VFTE
);
6007 wr32(E1000_VFTE
, reg
| (1 << vf
));
6008 reg
= rd32(E1000_VFRE
);
6009 wr32(E1000_VFRE
, reg
| (1 << vf
));
6011 adapter
->vf_data
[vf
].flags
|= IGB_VF_FLAG_CTS
;
6013 /* reply to reset with ack and vf mac address */
6014 msgbuf
[0] = E1000_VF_RESET
| E1000_VT_MSGTYPE_ACK
;
6015 memcpy(addr
, vf_mac
, ETH_ALEN
);
6016 igb_write_mbx(hw
, msgbuf
, 3, vf
);
6019 static int igb_set_vf_mac_addr(struct igb_adapter
*adapter
, u32
*msg
, int vf
)
6021 /* The VF MAC Address is stored in a packed array of bytes
6022 * starting at the second 32 bit word of the msg array
6024 unsigned char *addr
= (char *)&msg
[1];
6027 if (is_valid_ether_addr(addr
))
6028 err
= igb_set_vf_mac(adapter
, vf
, addr
);
6033 static void igb_rcv_ack_from_vf(struct igb_adapter
*adapter
, u32 vf
)
6035 struct e1000_hw
*hw
= &adapter
->hw
;
6036 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
6037 u32 msg
= E1000_VT_MSGTYPE_NACK
;
6039 /* if device isn't clear to send it shouldn't be reading either */
6040 if (!(vf_data
->flags
& IGB_VF_FLAG_CTS
) &&
6041 time_after(jiffies
, vf_data
->last_nack
+ (2 * HZ
))) {
6042 igb_write_mbx(hw
, &msg
, 1, vf
);
6043 vf_data
->last_nack
= jiffies
;
6047 static void igb_rcv_msg_from_vf(struct igb_adapter
*adapter
, u32 vf
)
6049 struct pci_dev
*pdev
= adapter
->pdev
;
6050 u32 msgbuf
[E1000_VFMAILBOX_SIZE
];
6051 struct e1000_hw
*hw
= &adapter
->hw
;
6052 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
6055 retval
= igb_read_mbx(hw
, msgbuf
, E1000_VFMAILBOX_SIZE
, vf
);
6058 /* if receive failed revoke VF CTS stats and restart init */
6059 dev_err(&pdev
->dev
, "Error receiving message from VF\n");
6060 vf_data
->flags
&= ~IGB_VF_FLAG_CTS
;
6061 if (!time_after(jiffies
, vf_data
->last_nack
+ (2 * HZ
)))
6066 /* this is a message we already processed, do nothing */
6067 if (msgbuf
[0] & (E1000_VT_MSGTYPE_ACK
| E1000_VT_MSGTYPE_NACK
))
6070 /* until the vf completes a reset it should not be
6071 * allowed to start any configuration.
6073 if (msgbuf
[0] == E1000_VF_RESET
) {
6074 igb_vf_reset_msg(adapter
, vf
);
6078 if (!(vf_data
->flags
& IGB_VF_FLAG_CTS
)) {
6079 if (!time_after(jiffies
, vf_data
->last_nack
+ (2 * HZ
)))
6085 switch ((msgbuf
[0] & 0xFFFF)) {
6086 case E1000_VF_SET_MAC_ADDR
:
6088 if (!(vf_data
->flags
& IGB_VF_FLAG_PF_SET_MAC
))
6089 retval
= igb_set_vf_mac_addr(adapter
, msgbuf
, vf
);
6091 dev_warn(&pdev
->dev
,
6092 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
6095 case E1000_VF_SET_PROMISC
:
6096 retval
= igb_set_vf_promisc(adapter
, msgbuf
, vf
);
6098 case E1000_VF_SET_MULTICAST
:
6099 retval
= igb_set_vf_multicasts(adapter
, msgbuf
, vf
);
6101 case E1000_VF_SET_LPE
:
6102 retval
= igb_set_vf_rlpml(adapter
, msgbuf
[1], vf
);
6104 case E1000_VF_SET_VLAN
:
6106 if (vf_data
->pf_vlan
)
6107 dev_warn(&pdev
->dev
,
6108 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
6111 retval
= igb_set_vf_vlan(adapter
, msgbuf
, vf
);
6114 dev_err(&pdev
->dev
, "Unhandled Msg %08x\n", msgbuf
[0]);
6119 msgbuf
[0] |= E1000_VT_MSGTYPE_CTS
;
6121 /* notify the VF of the results of what it sent us */
6123 msgbuf
[0] |= E1000_VT_MSGTYPE_NACK
;
6125 msgbuf
[0] |= E1000_VT_MSGTYPE_ACK
;
6127 igb_write_mbx(hw
, msgbuf
, 1, vf
);
6130 static void igb_msg_task(struct igb_adapter
*adapter
)
6132 struct e1000_hw
*hw
= &adapter
->hw
;
6135 for (vf
= 0; vf
< adapter
->vfs_allocated_count
; vf
++) {
6136 /* process any reset requests */
6137 if (!igb_check_for_rst(hw
, vf
))
6138 igb_vf_reset_event(adapter
, vf
);
6140 /* process any messages pending */
6141 if (!igb_check_for_msg(hw
, vf
))
6142 igb_rcv_msg_from_vf(adapter
, vf
);
6144 /* process any acks */
6145 if (!igb_check_for_ack(hw
, vf
))
6146 igb_rcv_ack_from_vf(adapter
, vf
);
6151 * igb_set_uta - Set unicast filter table address
6152 * @adapter: board private structure
6154 * The unicast table address is a register array of 32-bit registers.
6155 * The table is meant to be used in a way similar to how the MTA is used
6156 * however due to certain limitations in the hardware it is necessary to
6157 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
6158 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
6160 static void igb_set_uta(struct igb_adapter
*adapter
)
6162 struct e1000_hw
*hw
= &adapter
->hw
;
6165 /* The UTA table only exists on 82576 hardware and newer */
6166 if (hw
->mac
.type
< e1000_82576
)
6169 /* we only need to do this if VMDq is enabled */
6170 if (!adapter
->vfs_allocated_count
)
6173 for (i
= 0; i
< hw
->mac
.uta_reg_count
; i
++)
6174 array_wr32(E1000_UTA
, i
, ~0);
6178 * igb_intr_msi - Interrupt Handler
6179 * @irq: interrupt number
6180 * @data: pointer to a network interface device structure
6182 static irqreturn_t
igb_intr_msi(int irq
, void *data
)
6184 struct igb_adapter
*adapter
= data
;
6185 struct igb_q_vector
*q_vector
= adapter
->q_vector
[0];
6186 struct e1000_hw
*hw
= &adapter
->hw
;
6187 /* read ICR disables interrupts using IAM */
6188 u32 icr
= rd32(E1000_ICR
);
6190 igb_write_itr(q_vector
);
6192 if (icr
& E1000_ICR_DRSTA
)
6193 schedule_work(&adapter
->reset_task
);
6195 if (icr
& E1000_ICR_DOUTSYNC
) {
6196 /* HW is reporting DMA is out of sync */
6197 adapter
->stats
.doosync
++;
6200 if (icr
& (E1000_ICR_RXSEQ
| E1000_ICR_LSC
)) {
6201 hw
->mac
.get_link_status
= 1;
6202 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
6203 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
6206 if (icr
& E1000_ICR_TS
) {
6207 u32 tsicr
= rd32(E1000_TSICR
);
6209 if (tsicr
& E1000_TSICR_TXTS
) {
6210 /* acknowledge the interrupt */
6211 wr32(E1000_TSICR
, E1000_TSICR_TXTS
);
6212 /* retrieve hardware timestamp */
6213 schedule_work(&adapter
->ptp_tx_work
);
6217 napi_schedule(&q_vector
->napi
);
6223 * igb_intr - Legacy Interrupt Handler
6224 * @irq: interrupt number
6225 * @data: pointer to a network interface device structure
6227 static irqreturn_t
igb_intr(int irq
, void *data
)
6229 struct igb_adapter
*adapter
= data
;
6230 struct igb_q_vector
*q_vector
= adapter
->q_vector
[0];
6231 struct e1000_hw
*hw
= &adapter
->hw
;
6232 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
6233 * need for the IMC write
6235 u32 icr
= rd32(E1000_ICR
);
6237 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
6238 * not set, then the adapter didn't send an interrupt
6240 if (!(icr
& E1000_ICR_INT_ASSERTED
))
6243 igb_write_itr(q_vector
);
6245 if (icr
& E1000_ICR_DRSTA
)
6246 schedule_work(&adapter
->reset_task
);
6248 if (icr
& E1000_ICR_DOUTSYNC
) {
6249 /* HW is reporting DMA is out of sync */
6250 adapter
->stats
.doosync
++;
6253 if (icr
& (E1000_ICR_RXSEQ
| E1000_ICR_LSC
)) {
6254 hw
->mac
.get_link_status
= 1;
6255 /* guard against interrupt when we're going down */
6256 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
6257 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
6260 if (icr
& E1000_ICR_TS
) {
6261 u32 tsicr
= rd32(E1000_TSICR
);
6263 if (tsicr
& E1000_TSICR_TXTS
) {
6264 /* acknowledge the interrupt */
6265 wr32(E1000_TSICR
, E1000_TSICR_TXTS
);
6266 /* retrieve hardware timestamp */
6267 schedule_work(&adapter
->ptp_tx_work
);
6271 napi_schedule(&q_vector
->napi
);
6276 static void igb_ring_irq_enable(struct igb_q_vector
*q_vector
)
6278 struct igb_adapter
*adapter
= q_vector
->adapter
;
6279 struct e1000_hw
*hw
= &adapter
->hw
;
6281 if ((q_vector
->rx
.ring
&& (adapter
->rx_itr_setting
& 3)) ||
6282 (!q_vector
->rx
.ring
&& (adapter
->tx_itr_setting
& 3))) {
6283 if ((adapter
->num_q_vectors
== 1) && !adapter
->vf_data
)
6284 igb_set_itr(q_vector
);
6286 igb_update_ring_itr(q_vector
);
6289 if (!test_bit(__IGB_DOWN
, &adapter
->state
)) {
6290 if (adapter
->flags
& IGB_FLAG_HAS_MSIX
)
6291 wr32(E1000_EIMS
, q_vector
->eims_value
);
6293 igb_irq_enable(adapter
);
6298 * igb_poll - NAPI Rx polling callback
6299 * @napi: napi polling structure
6300 * @budget: count of how many packets we should handle
6302 static int igb_poll(struct napi_struct
*napi
, int budget
)
6304 struct igb_q_vector
*q_vector
= container_of(napi
,
6305 struct igb_q_vector
,
6307 bool clean_complete
= true;
6309 #ifdef CONFIG_IGB_DCA
6310 if (q_vector
->adapter
->flags
& IGB_FLAG_DCA_ENABLED
)
6311 igb_update_dca(q_vector
);
6313 if (q_vector
->tx
.ring
)
6314 clean_complete
= igb_clean_tx_irq(q_vector
);
6316 if (q_vector
->rx
.ring
)
6317 clean_complete
&= igb_clean_rx_irq(q_vector
, budget
);
6319 /* If all work not completed, return budget and keep polling */
6320 if (!clean_complete
)
6323 /* If not enough Rx work done, exit the polling mode */
6324 napi_complete(napi
);
6325 igb_ring_irq_enable(q_vector
);
6331 * igb_clean_tx_irq - Reclaim resources after transmit completes
6332 * @q_vector: pointer to q_vector containing needed info
6334 * returns true if ring is completely cleaned
6336 static bool igb_clean_tx_irq(struct igb_q_vector
*q_vector
)
6338 struct igb_adapter
*adapter
= q_vector
->adapter
;
6339 struct igb_ring
*tx_ring
= q_vector
->tx
.ring
;
6340 struct igb_tx_buffer
*tx_buffer
;
6341 union e1000_adv_tx_desc
*tx_desc
;
6342 unsigned int total_bytes
= 0, total_packets
= 0;
6343 unsigned int budget
= q_vector
->tx
.work_limit
;
6344 unsigned int i
= tx_ring
->next_to_clean
;
6346 if (test_bit(__IGB_DOWN
, &adapter
->state
))
6349 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
6350 tx_desc
= IGB_TX_DESC(tx_ring
, i
);
6351 i
-= tx_ring
->count
;
6354 union e1000_adv_tx_desc
*eop_desc
= tx_buffer
->next_to_watch
;
6356 /* if next_to_watch is not set then there is no work pending */
6360 /* prevent any other reads prior to eop_desc */
6361 read_barrier_depends();
6363 /* if DD is not set pending work has not been completed */
6364 if (!(eop_desc
->wb
.status
& cpu_to_le32(E1000_TXD_STAT_DD
)))
6367 /* clear next_to_watch to prevent false hangs */
6368 tx_buffer
->next_to_watch
= NULL
;
6370 /* update the statistics for this packet */
6371 total_bytes
+= tx_buffer
->bytecount
;
6372 total_packets
+= tx_buffer
->gso_segs
;
6375 dev_consume_skb_any(tx_buffer
->skb
);
6377 /* unmap skb header data */
6378 dma_unmap_single(tx_ring
->dev
,
6379 dma_unmap_addr(tx_buffer
, dma
),
6380 dma_unmap_len(tx_buffer
, len
),
6383 /* clear tx_buffer data */
6384 tx_buffer
->skb
= NULL
;
6385 dma_unmap_len_set(tx_buffer
, len
, 0);
6387 /* clear last DMA location and unmap remaining buffers */
6388 while (tx_desc
!= eop_desc
) {
6393 i
-= tx_ring
->count
;
6394 tx_buffer
= tx_ring
->tx_buffer_info
;
6395 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
6398 /* unmap any remaining paged data */
6399 if (dma_unmap_len(tx_buffer
, len
)) {
6400 dma_unmap_page(tx_ring
->dev
,
6401 dma_unmap_addr(tx_buffer
, dma
),
6402 dma_unmap_len(tx_buffer
, len
),
6404 dma_unmap_len_set(tx_buffer
, len
, 0);
6408 /* move us one more past the eop_desc for start of next pkt */
6413 i
-= tx_ring
->count
;
6414 tx_buffer
= tx_ring
->tx_buffer_info
;
6415 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
6418 /* issue prefetch for next Tx descriptor */
6421 /* update budget accounting */
6423 } while (likely(budget
));
6425 netdev_tx_completed_queue(txring_txq(tx_ring
),
6426 total_packets
, total_bytes
);
6427 i
+= tx_ring
->count
;
6428 tx_ring
->next_to_clean
= i
;
6429 u64_stats_update_begin(&tx_ring
->tx_syncp
);
6430 tx_ring
->tx_stats
.bytes
+= total_bytes
;
6431 tx_ring
->tx_stats
.packets
+= total_packets
;
6432 u64_stats_update_end(&tx_ring
->tx_syncp
);
6433 q_vector
->tx
.total_bytes
+= total_bytes
;
6434 q_vector
->tx
.total_packets
+= total_packets
;
6436 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG
, &tx_ring
->flags
)) {
6437 struct e1000_hw
*hw
= &adapter
->hw
;
6439 /* Detect a transmit hang in hardware, this serializes the
6440 * check with the clearing of time_stamp and movement of i
6442 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG
, &tx_ring
->flags
);
6443 if (tx_buffer
->next_to_watch
&&
6444 time_after(jiffies
, tx_buffer
->time_stamp
+
6445 (adapter
->tx_timeout_factor
* HZ
)) &&
6446 !(rd32(E1000_STATUS
) & E1000_STATUS_TXOFF
)) {
6448 /* detected Tx unit hang */
6449 dev_err(tx_ring
->dev
,
6450 "Detected Tx Unit Hang\n"
6454 " next_to_use <%x>\n"
6455 " next_to_clean <%x>\n"
6456 "buffer_info[next_to_clean]\n"
6457 " time_stamp <%lx>\n"
6458 " next_to_watch <%p>\n"
6460 " desc.status <%x>\n",
6461 tx_ring
->queue_index
,
6462 rd32(E1000_TDH(tx_ring
->reg_idx
)),
6463 readl(tx_ring
->tail
),
6464 tx_ring
->next_to_use
,
6465 tx_ring
->next_to_clean
,
6466 tx_buffer
->time_stamp
,
6467 tx_buffer
->next_to_watch
,
6469 tx_buffer
->next_to_watch
->wb
.status
);
6470 netif_stop_subqueue(tx_ring
->netdev
,
6471 tx_ring
->queue_index
);
6473 /* we are about to reset, no point in enabling stuff */
6478 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6479 if (unlikely(total_packets
&&
6480 netif_carrier_ok(tx_ring
->netdev
) &&
6481 igb_desc_unused(tx_ring
) >= TX_WAKE_THRESHOLD
)) {
6482 /* Make sure that anybody stopping the queue after this
6483 * sees the new next_to_clean.
6486 if (__netif_subqueue_stopped(tx_ring
->netdev
,
6487 tx_ring
->queue_index
) &&
6488 !(test_bit(__IGB_DOWN
, &adapter
->state
))) {
6489 netif_wake_subqueue(tx_ring
->netdev
,
6490 tx_ring
->queue_index
);
6492 u64_stats_update_begin(&tx_ring
->tx_syncp
);
6493 tx_ring
->tx_stats
.restart_queue
++;
6494 u64_stats_update_end(&tx_ring
->tx_syncp
);
6502 * igb_reuse_rx_page - page flip buffer and store it back on the ring
6503 * @rx_ring: rx descriptor ring to store buffers on
6504 * @old_buff: donor buffer to have page reused
6506 * Synchronizes page for reuse by the adapter
6508 static void igb_reuse_rx_page(struct igb_ring
*rx_ring
,
6509 struct igb_rx_buffer
*old_buff
)
6511 struct igb_rx_buffer
*new_buff
;
6512 u16 nta
= rx_ring
->next_to_alloc
;
6514 new_buff
= &rx_ring
->rx_buffer_info
[nta
];
6516 /* update, and store next to alloc */
6518 rx_ring
->next_to_alloc
= (nta
< rx_ring
->count
) ? nta
: 0;
6520 /* transfer page from old buffer to new buffer */
6521 *new_buff
= *old_buff
;
6523 /* sync the buffer for use by the device */
6524 dma_sync_single_range_for_device(rx_ring
->dev
, old_buff
->dma
,
6525 old_buff
->page_offset
,
6530 static bool igb_can_reuse_rx_page(struct igb_rx_buffer
*rx_buffer
,
6532 unsigned int truesize
)
6534 /* avoid re-using remote pages */
6535 if (unlikely(page_to_nid(page
) != numa_node_id()))
6538 if (unlikely(page
->pfmemalloc
))
6541 #if (PAGE_SIZE < 8192)
6542 /* if we are only owner of page we can reuse it */
6543 if (unlikely(page_count(page
) != 1))
6546 /* flip page offset to other buffer */
6547 rx_buffer
->page_offset
^= IGB_RX_BUFSZ
;
6549 /* Even if we own the page, we are not allowed to use atomic_set()
6550 * This would break get_page_unless_zero() users.
6552 atomic_inc(&page
->_count
);
6554 /* move offset up to the next cache line */
6555 rx_buffer
->page_offset
+= truesize
;
6557 if (rx_buffer
->page_offset
> (PAGE_SIZE
- IGB_RX_BUFSZ
))
6560 /* bump ref count on page before it is given to the stack */
6568 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6569 * @rx_ring: rx descriptor ring to transact packets on
6570 * @rx_buffer: buffer containing page to add
6571 * @rx_desc: descriptor containing length of buffer written by hardware
6572 * @skb: sk_buff to place the data into
6574 * This function will add the data contained in rx_buffer->page to the skb.
6575 * This is done either through a direct copy if the data in the buffer is
6576 * less than the skb header size, otherwise it will just attach the page as
6577 * a frag to the skb.
6579 * The function will then update the page offset if necessary and return
6580 * true if the buffer can be reused by the adapter.
6582 static bool igb_add_rx_frag(struct igb_ring
*rx_ring
,
6583 struct igb_rx_buffer
*rx_buffer
,
6584 union e1000_adv_rx_desc
*rx_desc
,
6585 struct sk_buff
*skb
)
6587 struct page
*page
= rx_buffer
->page
;
6588 unsigned int size
= le16_to_cpu(rx_desc
->wb
.upper
.length
);
6589 #if (PAGE_SIZE < 8192)
6590 unsigned int truesize
= IGB_RX_BUFSZ
;
6592 unsigned int truesize
= ALIGN(size
, L1_CACHE_BYTES
);
6595 if ((size
<= IGB_RX_HDR_LEN
) && !skb_is_nonlinear(skb
)) {
6596 unsigned char *va
= page_address(page
) + rx_buffer
->page_offset
;
6598 if (igb_test_staterr(rx_desc
, E1000_RXDADV_STAT_TSIP
)) {
6599 igb_ptp_rx_pktstamp(rx_ring
->q_vector
, va
, skb
);
6600 va
+= IGB_TS_HDR_LEN
;
6601 size
-= IGB_TS_HDR_LEN
;
6604 memcpy(__skb_put(skb
, size
), va
, ALIGN(size
, sizeof(long)));
6606 /* we can reuse buffer as-is, just make sure it is local */
6607 if (likely((page_to_nid(page
) == numa_node_id()) &&
6611 /* this page cannot be reused so discard it */
6616 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
,
6617 rx_buffer
->page_offset
, size
, truesize
);
6619 return igb_can_reuse_rx_page(rx_buffer
, page
, truesize
);
6622 static struct sk_buff
*igb_fetch_rx_buffer(struct igb_ring
*rx_ring
,
6623 union e1000_adv_rx_desc
*rx_desc
,
6624 struct sk_buff
*skb
)
6626 struct igb_rx_buffer
*rx_buffer
;
6629 rx_buffer
= &rx_ring
->rx_buffer_info
[rx_ring
->next_to_clean
];
6631 page
= rx_buffer
->page
;
6635 void *page_addr
= page_address(page
) +
6636 rx_buffer
->page_offset
;
6638 /* prefetch first cache line of first page */
6639 prefetch(page_addr
);
6640 #if L1_CACHE_BYTES < 128
6641 prefetch(page_addr
+ L1_CACHE_BYTES
);
6644 /* allocate a skb to store the frags */
6645 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
6647 if (unlikely(!skb
)) {
6648 rx_ring
->rx_stats
.alloc_failed
++;
6652 /* we will be copying header into skb->data in
6653 * pskb_may_pull so it is in our interest to prefetch
6654 * it now to avoid a possible cache miss
6656 prefetchw(skb
->data
);
6659 /* we are reusing so sync this buffer for CPU use */
6660 dma_sync_single_range_for_cpu(rx_ring
->dev
,
6662 rx_buffer
->page_offset
,
6666 /* pull page into skb */
6667 if (igb_add_rx_frag(rx_ring
, rx_buffer
, rx_desc
, skb
)) {
6668 /* hand second half of page back to the ring */
6669 igb_reuse_rx_page(rx_ring
, rx_buffer
);
6671 /* we are not reusing the buffer so unmap it */
6672 dma_unmap_page(rx_ring
->dev
, rx_buffer
->dma
,
6673 PAGE_SIZE
, DMA_FROM_DEVICE
);
6676 /* clear contents of rx_buffer */
6677 rx_buffer
->page
= NULL
;
6682 static inline void igb_rx_checksum(struct igb_ring
*ring
,
6683 union e1000_adv_rx_desc
*rx_desc
,
6684 struct sk_buff
*skb
)
6686 skb_checksum_none_assert(skb
);
6688 /* Ignore Checksum bit is set */
6689 if (igb_test_staterr(rx_desc
, E1000_RXD_STAT_IXSM
))
6692 /* Rx checksum disabled via ethtool */
6693 if (!(ring
->netdev
->features
& NETIF_F_RXCSUM
))
6696 /* TCP/UDP checksum error bit is set */
6697 if (igb_test_staterr(rx_desc
,
6698 E1000_RXDEXT_STATERR_TCPE
|
6699 E1000_RXDEXT_STATERR_IPE
)) {
6700 /* work around errata with sctp packets where the TCPE aka
6701 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6702 * packets, (aka let the stack check the crc32c)
6704 if (!((skb
->len
== 60) &&
6705 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM
, &ring
->flags
))) {
6706 u64_stats_update_begin(&ring
->rx_syncp
);
6707 ring
->rx_stats
.csum_err
++;
6708 u64_stats_update_end(&ring
->rx_syncp
);
6710 /* let the stack verify checksum errors */
6713 /* It must be a TCP or UDP packet with a valid checksum */
6714 if (igb_test_staterr(rx_desc
, E1000_RXD_STAT_TCPCS
|
6715 E1000_RXD_STAT_UDPCS
))
6716 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
6718 dev_dbg(ring
->dev
, "cksum success: bits %08X\n",
6719 le32_to_cpu(rx_desc
->wb
.upper
.status_error
));
6722 static inline void igb_rx_hash(struct igb_ring
*ring
,
6723 union e1000_adv_rx_desc
*rx_desc
,
6724 struct sk_buff
*skb
)
6726 if (ring
->netdev
->features
& NETIF_F_RXHASH
)
6728 le32_to_cpu(rx_desc
->wb
.lower
.hi_dword
.rss
),
6733 * igb_is_non_eop - process handling of non-EOP buffers
6734 * @rx_ring: Rx ring being processed
6735 * @rx_desc: Rx descriptor for current buffer
6736 * @skb: current socket buffer containing buffer in progress
6738 * This function updates next to clean. If the buffer is an EOP buffer
6739 * this function exits returning false, otherwise it will place the
6740 * sk_buff in the next buffer to be chained and return true indicating
6741 * that this is in fact a non-EOP buffer.
6743 static bool igb_is_non_eop(struct igb_ring
*rx_ring
,
6744 union e1000_adv_rx_desc
*rx_desc
)
6746 u32 ntc
= rx_ring
->next_to_clean
+ 1;
6748 /* fetch, update, and store next to clean */
6749 ntc
= (ntc
< rx_ring
->count
) ? ntc
: 0;
6750 rx_ring
->next_to_clean
= ntc
;
6752 prefetch(IGB_RX_DESC(rx_ring
, ntc
));
6754 if (likely(igb_test_staterr(rx_desc
, E1000_RXD_STAT_EOP
)))
6761 * igb_pull_tail - igb specific version of skb_pull_tail
6762 * @rx_ring: rx descriptor ring packet is being transacted on
6763 * @rx_desc: pointer to the EOP Rx descriptor
6764 * @skb: pointer to current skb being adjusted
6766 * This function is an igb specific version of __pskb_pull_tail. The
6767 * main difference between this version and the original function is that
6768 * this function can make several assumptions about the state of things
6769 * that allow for significant optimizations versus the standard function.
6770 * As a result we can do things like drop a frag and maintain an accurate
6771 * truesize for the skb.
6773 static void igb_pull_tail(struct igb_ring
*rx_ring
,
6774 union e1000_adv_rx_desc
*rx_desc
,
6775 struct sk_buff
*skb
)
6777 struct skb_frag_struct
*frag
= &skb_shinfo(skb
)->frags
[0];
6779 unsigned int pull_len
;
6781 /* it is valid to use page_address instead of kmap since we are
6782 * working with pages allocated out of the lomem pool per
6783 * alloc_page(GFP_ATOMIC)
6785 va
= skb_frag_address(frag
);
6787 if (igb_test_staterr(rx_desc
, E1000_RXDADV_STAT_TSIP
)) {
6788 /* retrieve timestamp from buffer */
6789 igb_ptp_rx_pktstamp(rx_ring
->q_vector
, va
, skb
);
6791 /* update pointers to remove timestamp header */
6792 skb_frag_size_sub(frag
, IGB_TS_HDR_LEN
);
6793 frag
->page_offset
+= IGB_TS_HDR_LEN
;
6794 skb
->data_len
-= IGB_TS_HDR_LEN
;
6795 skb
->len
-= IGB_TS_HDR_LEN
;
6797 /* move va to start of packet data */
6798 va
+= IGB_TS_HDR_LEN
;
6801 /* we need the header to contain the greater of either ETH_HLEN or
6802 * 60 bytes if the skb->len is less than 60 for skb_pad.
6804 pull_len
= eth_get_headlen(va
, IGB_RX_HDR_LEN
);
6806 /* align pull length to size of long to optimize memcpy performance */
6807 skb_copy_to_linear_data(skb
, va
, ALIGN(pull_len
, sizeof(long)));
6809 /* update all of the pointers */
6810 skb_frag_size_sub(frag
, pull_len
);
6811 frag
->page_offset
+= pull_len
;
6812 skb
->data_len
-= pull_len
;
6813 skb
->tail
+= pull_len
;
6817 * igb_cleanup_headers - Correct corrupted or empty headers
6818 * @rx_ring: rx descriptor ring packet is being transacted on
6819 * @rx_desc: pointer to the EOP Rx descriptor
6820 * @skb: pointer to current skb being fixed
6822 * Address the case where we are pulling data in on pages only
6823 * and as such no data is present in the skb header.
6825 * In addition if skb is not at least 60 bytes we need to pad it so that
6826 * it is large enough to qualify as a valid Ethernet frame.
6828 * Returns true if an error was encountered and skb was freed.
6830 static bool igb_cleanup_headers(struct igb_ring
*rx_ring
,
6831 union e1000_adv_rx_desc
*rx_desc
,
6832 struct sk_buff
*skb
)
6834 if (unlikely((igb_test_staterr(rx_desc
,
6835 E1000_RXDEXT_ERR_FRAME_ERR_MASK
)))) {
6836 struct net_device
*netdev
= rx_ring
->netdev
;
6837 if (!(netdev
->features
& NETIF_F_RXALL
)) {
6838 dev_kfree_skb_any(skb
);
6843 /* place header in linear portion of buffer */
6844 if (skb_is_nonlinear(skb
))
6845 igb_pull_tail(rx_ring
, rx_desc
, skb
);
6847 /* if skb_pad returns an error the skb was freed */
6848 if (unlikely(skb
->len
< 60)) {
6849 int pad_len
= 60 - skb
->len
;
6851 if (skb_pad(skb
, pad_len
))
6853 __skb_put(skb
, pad_len
);
6860 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
6861 * @rx_ring: rx descriptor ring packet is being transacted on
6862 * @rx_desc: pointer to the EOP Rx descriptor
6863 * @skb: pointer to current skb being populated
6865 * This function checks the ring, descriptor, and packet information in
6866 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
6867 * other fields within the skb.
6869 static void igb_process_skb_fields(struct igb_ring
*rx_ring
,
6870 union e1000_adv_rx_desc
*rx_desc
,
6871 struct sk_buff
*skb
)
6873 struct net_device
*dev
= rx_ring
->netdev
;
6875 igb_rx_hash(rx_ring
, rx_desc
, skb
);
6877 igb_rx_checksum(rx_ring
, rx_desc
, skb
);
6879 if (igb_test_staterr(rx_desc
, E1000_RXDADV_STAT_TS
) &&
6880 !igb_test_staterr(rx_desc
, E1000_RXDADV_STAT_TSIP
))
6881 igb_ptp_rx_rgtstamp(rx_ring
->q_vector
, skb
);
6883 if ((dev
->features
& NETIF_F_HW_VLAN_CTAG_RX
) &&
6884 igb_test_staterr(rx_desc
, E1000_RXD_STAT_VP
)) {
6887 if (igb_test_staterr(rx_desc
, E1000_RXDEXT_STATERR_LB
) &&
6888 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP
, &rx_ring
->flags
))
6889 vid
= be16_to_cpu(rx_desc
->wb
.upper
.vlan
);
6891 vid
= le16_to_cpu(rx_desc
->wb
.upper
.vlan
);
6893 __vlan_hwaccel_put_tag(skb
, htons(ETH_P_8021Q
), vid
);
6896 skb_record_rx_queue(skb
, rx_ring
->queue_index
);
6898 skb
->protocol
= eth_type_trans(skb
, rx_ring
->netdev
);
6901 static bool igb_clean_rx_irq(struct igb_q_vector
*q_vector
, const int budget
)
6903 struct igb_ring
*rx_ring
= q_vector
->rx
.ring
;
6904 struct sk_buff
*skb
= rx_ring
->skb
;
6905 unsigned int total_bytes
= 0, total_packets
= 0;
6906 u16 cleaned_count
= igb_desc_unused(rx_ring
);
6908 while (likely(total_packets
< budget
)) {
6909 union e1000_adv_rx_desc
*rx_desc
;
6911 /* return some buffers to hardware, one at a time is too slow */
6912 if (cleaned_count
>= IGB_RX_BUFFER_WRITE
) {
6913 igb_alloc_rx_buffers(rx_ring
, cleaned_count
);
6917 rx_desc
= IGB_RX_DESC(rx_ring
, rx_ring
->next_to_clean
);
6919 if (!igb_test_staterr(rx_desc
, E1000_RXD_STAT_DD
))
6922 /* This memory barrier is needed to keep us from reading
6923 * any other fields out of the rx_desc until we know the
6924 * RXD_STAT_DD bit is set
6928 /* retrieve a buffer from the ring */
6929 skb
= igb_fetch_rx_buffer(rx_ring
, rx_desc
, skb
);
6931 /* exit if we failed to retrieve a buffer */
6937 /* fetch next buffer in frame if non-eop */
6938 if (igb_is_non_eop(rx_ring
, rx_desc
))
6941 /* verify the packet layout is correct */
6942 if (igb_cleanup_headers(rx_ring
, rx_desc
, skb
)) {
6947 /* probably a little skewed due to removing CRC */
6948 total_bytes
+= skb
->len
;
6950 /* populate checksum, timestamp, VLAN, and protocol */
6951 igb_process_skb_fields(rx_ring
, rx_desc
, skb
);
6953 napi_gro_receive(&q_vector
->napi
, skb
);
6955 /* reset skb pointer */
6958 /* update budget accounting */
6962 /* place incomplete frames back on ring for completion */
6965 u64_stats_update_begin(&rx_ring
->rx_syncp
);
6966 rx_ring
->rx_stats
.packets
+= total_packets
;
6967 rx_ring
->rx_stats
.bytes
+= total_bytes
;
6968 u64_stats_update_end(&rx_ring
->rx_syncp
);
6969 q_vector
->rx
.total_packets
+= total_packets
;
6970 q_vector
->rx
.total_bytes
+= total_bytes
;
6973 igb_alloc_rx_buffers(rx_ring
, cleaned_count
);
6975 return total_packets
< budget
;
6978 static bool igb_alloc_mapped_page(struct igb_ring
*rx_ring
,
6979 struct igb_rx_buffer
*bi
)
6981 struct page
*page
= bi
->page
;
6984 /* since we are recycling buffers we should seldom need to alloc */
6988 /* alloc new page for storage */
6989 page
= __skb_alloc_page(GFP_ATOMIC
| __GFP_COLD
, NULL
);
6990 if (unlikely(!page
)) {
6991 rx_ring
->rx_stats
.alloc_failed
++;
6995 /* map page for use */
6996 dma
= dma_map_page(rx_ring
->dev
, page
, 0, PAGE_SIZE
, DMA_FROM_DEVICE
);
6998 /* if mapping failed free memory back to system since
6999 * there isn't much point in holding memory we can't use
7001 if (dma_mapping_error(rx_ring
->dev
, dma
)) {
7004 rx_ring
->rx_stats
.alloc_failed
++;
7010 bi
->page_offset
= 0;
7016 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
7017 * @adapter: address of board private structure
7019 void igb_alloc_rx_buffers(struct igb_ring
*rx_ring
, u16 cleaned_count
)
7021 union e1000_adv_rx_desc
*rx_desc
;
7022 struct igb_rx_buffer
*bi
;
7023 u16 i
= rx_ring
->next_to_use
;
7029 rx_desc
= IGB_RX_DESC(rx_ring
, i
);
7030 bi
= &rx_ring
->rx_buffer_info
[i
];
7031 i
-= rx_ring
->count
;
7034 if (!igb_alloc_mapped_page(rx_ring
, bi
))
7037 /* Refresh the desc even if buffer_addrs didn't change
7038 * because each write-back erases this info.
7040 rx_desc
->read
.pkt_addr
= cpu_to_le64(bi
->dma
+ bi
->page_offset
);
7046 rx_desc
= IGB_RX_DESC(rx_ring
, 0);
7047 bi
= rx_ring
->rx_buffer_info
;
7048 i
-= rx_ring
->count
;
7051 /* clear the hdr_addr for the next_to_use descriptor */
7052 rx_desc
->read
.hdr_addr
= 0;
7055 } while (cleaned_count
);
7057 i
+= rx_ring
->count
;
7059 if (rx_ring
->next_to_use
!= i
) {
7060 /* record the next descriptor to use */
7061 rx_ring
->next_to_use
= i
;
7063 /* update next to alloc since we have filled the ring */
7064 rx_ring
->next_to_alloc
= i
;
7066 /* Force memory writes to complete before letting h/w
7067 * know there are new descriptors to fetch. (Only
7068 * applicable for weak-ordered memory model archs,
7072 writel(i
, rx_ring
->tail
);
7082 static int igb_mii_ioctl(struct net_device
*netdev
, struct ifreq
*ifr
, int cmd
)
7084 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7085 struct mii_ioctl_data
*data
= if_mii(ifr
);
7087 if (adapter
->hw
.phy
.media_type
!= e1000_media_type_copper
)
7092 data
->phy_id
= adapter
->hw
.phy
.addr
;
7095 if (igb_read_phy_reg(&adapter
->hw
, data
->reg_num
& 0x1F,
7112 static int igb_ioctl(struct net_device
*netdev
, struct ifreq
*ifr
, int cmd
)
7118 return igb_mii_ioctl(netdev
, ifr
, cmd
);
7120 return igb_ptp_get_ts_config(netdev
, ifr
);
7122 return igb_ptp_set_ts_config(netdev
, ifr
);
7128 void igb_read_pci_cfg(struct e1000_hw
*hw
, u32 reg
, u16
*value
)
7130 struct igb_adapter
*adapter
= hw
->back
;
7132 pci_read_config_word(adapter
->pdev
, reg
, value
);
7135 void igb_write_pci_cfg(struct e1000_hw
*hw
, u32 reg
, u16
*value
)
7137 struct igb_adapter
*adapter
= hw
->back
;
7139 pci_write_config_word(adapter
->pdev
, reg
, *value
);
7142 s32
igb_read_pcie_cap_reg(struct e1000_hw
*hw
, u32 reg
, u16
*value
)
7144 struct igb_adapter
*adapter
= hw
->back
;
7146 if (pcie_capability_read_word(adapter
->pdev
, reg
, value
))
7147 return -E1000_ERR_CONFIG
;
7152 s32
igb_write_pcie_cap_reg(struct e1000_hw
*hw
, u32 reg
, u16
*value
)
7154 struct igb_adapter
*adapter
= hw
->back
;
7156 if (pcie_capability_write_word(adapter
->pdev
, reg
, *value
))
7157 return -E1000_ERR_CONFIG
;
7162 static void igb_vlan_mode(struct net_device
*netdev
, netdev_features_t features
)
7164 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7165 struct e1000_hw
*hw
= &adapter
->hw
;
7167 bool enable
= !!(features
& NETIF_F_HW_VLAN_CTAG_RX
);
7170 /* enable VLAN tag insert/strip */
7171 ctrl
= rd32(E1000_CTRL
);
7172 ctrl
|= E1000_CTRL_VME
;
7173 wr32(E1000_CTRL
, ctrl
);
7175 /* Disable CFI check */
7176 rctl
= rd32(E1000_RCTL
);
7177 rctl
&= ~E1000_RCTL_CFIEN
;
7178 wr32(E1000_RCTL
, rctl
);
7180 /* disable VLAN tag insert/strip */
7181 ctrl
= rd32(E1000_CTRL
);
7182 ctrl
&= ~E1000_CTRL_VME
;
7183 wr32(E1000_CTRL
, ctrl
);
7186 igb_rlpml_set(adapter
);
7189 static int igb_vlan_rx_add_vid(struct net_device
*netdev
,
7190 __be16 proto
, u16 vid
)
7192 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7193 struct e1000_hw
*hw
= &adapter
->hw
;
7194 int pf_id
= adapter
->vfs_allocated_count
;
7196 /* attempt to add filter to vlvf array */
7197 igb_vlvf_set(adapter
, vid
, true, pf_id
);
7199 /* add the filter since PF can receive vlans w/o entry in vlvf */
7200 igb_vfta_set(hw
, vid
, true);
7202 set_bit(vid
, adapter
->active_vlans
);
7207 static int igb_vlan_rx_kill_vid(struct net_device
*netdev
,
7208 __be16 proto
, u16 vid
)
7210 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7211 struct e1000_hw
*hw
= &adapter
->hw
;
7212 int pf_id
= adapter
->vfs_allocated_count
;
7215 /* remove vlan from VLVF table array */
7216 err
= igb_vlvf_set(adapter
, vid
, false, pf_id
);
7218 /* if vid was not present in VLVF just remove it from table */
7220 igb_vfta_set(hw
, vid
, false);
7222 clear_bit(vid
, adapter
->active_vlans
);
7227 static void igb_restore_vlan(struct igb_adapter
*adapter
)
7231 igb_vlan_mode(adapter
->netdev
, adapter
->netdev
->features
);
7233 for_each_set_bit(vid
, adapter
->active_vlans
, VLAN_N_VID
)
7234 igb_vlan_rx_add_vid(adapter
->netdev
, htons(ETH_P_8021Q
), vid
);
7237 int igb_set_spd_dplx(struct igb_adapter
*adapter
, u32 spd
, u8 dplx
)
7239 struct pci_dev
*pdev
= adapter
->pdev
;
7240 struct e1000_mac_info
*mac
= &adapter
->hw
.mac
;
7244 /* Make sure dplx is at most 1 bit and lsb of speed is not set
7245 * for the switch() below to work
7247 if ((spd
& 1) || (dplx
& ~1))
7250 /* Fiber NIC's only allow 1000 gbps Full duplex
7251 * and 100Mbps Full duplex for 100baseFx sfp
7253 if (adapter
->hw
.phy
.media_type
== e1000_media_type_internal_serdes
) {
7254 switch (spd
+ dplx
) {
7255 case SPEED_10
+ DUPLEX_HALF
:
7256 case SPEED_10
+ DUPLEX_FULL
:
7257 case SPEED_100
+ DUPLEX_HALF
:
7264 switch (spd
+ dplx
) {
7265 case SPEED_10
+ DUPLEX_HALF
:
7266 mac
->forced_speed_duplex
= ADVERTISE_10_HALF
;
7268 case SPEED_10
+ DUPLEX_FULL
:
7269 mac
->forced_speed_duplex
= ADVERTISE_10_FULL
;
7271 case SPEED_100
+ DUPLEX_HALF
:
7272 mac
->forced_speed_duplex
= ADVERTISE_100_HALF
;
7274 case SPEED_100
+ DUPLEX_FULL
:
7275 mac
->forced_speed_duplex
= ADVERTISE_100_FULL
;
7277 case SPEED_1000
+ DUPLEX_FULL
:
7279 adapter
->hw
.phy
.autoneg_advertised
= ADVERTISE_1000_FULL
;
7281 case SPEED_1000
+ DUPLEX_HALF
: /* not supported */
7286 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7287 adapter
->hw
.phy
.mdix
= AUTO_ALL_MODES
;
7292 dev_err(&pdev
->dev
, "Unsupported Speed/Duplex configuration\n");
7296 static int __igb_shutdown(struct pci_dev
*pdev
, bool *enable_wake
,
7299 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7300 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7301 struct e1000_hw
*hw
= &adapter
->hw
;
7302 u32 ctrl
, rctl
, status
;
7303 u32 wufc
= runtime
? E1000_WUFC_LNKC
: adapter
->wol
;
7308 netif_device_detach(netdev
);
7310 if (netif_running(netdev
))
7311 __igb_close(netdev
, true);
7313 igb_clear_interrupt_scheme(adapter
);
7316 retval
= pci_save_state(pdev
);
7321 status
= rd32(E1000_STATUS
);
7322 if (status
& E1000_STATUS_LU
)
7323 wufc
&= ~E1000_WUFC_LNKC
;
7326 igb_setup_rctl(adapter
);
7327 igb_set_rx_mode(netdev
);
7329 /* turn on all-multi mode if wake on multicast is enabled */
7330 if (wufc
& E1000_WUFC_MC
) {
7331 rctl
= rd32(E1000_RCTL
);
7332 rctl
|= E1000_RCTL_MPE
;
7333 wr32(E1000_RCTL
, rctl
);
7336 ctrl
= rd32(E1000_CTRL
);
7337 /* advertise wake from D3Cold */
7338 #define E1000_CTRL_ADVD3WUC 0x00100000
7339 /* phy power management enable */
7340 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7341 ctrl
|= E1000_CTRL_ADVD3WUC
;
7342 wr32(E1000_CTRL
, ctrl
);
7344 /* Allow time for pending master requests to run */
7345 igb_disable_pcie_master(hw
);
7347 wr32(E1000_WUC
, E1000_WUC_PME_EN
);
7348 wr32(E1000_WUFC
, wufc
);
7351 wr32(E1000_WUFC
, 0);
7354 *enable_wake
= wufc
|| adapter
->en_mng_pt
;
7356 igb_power_down_link(adapter
);
7358 igb_power_up_link(adapter
);
7360 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7361 * would have already happened in close and is redundant.
7363 igb_release_hw_control(adapter
);
7365 pci_disable_device(pdev
);
7371 #ifdef CONFIG_PM_SLEEP
7372 static int igb_suspend(struct device
*dev
)
7376 struct pci_dev
*pdev
= to_pci_dev(dev
);
7378 retval
= __igb_shutdown(pdev
, &wake
, 0);
7383 pci_prepare_to_sleep(pdev
);
7385 pci_wake_from_d3(pdev
, false);
7386 pci_set_power_state(pdev
, PCI_D3hot
);
7391 #endif /* CONFIG_PM_SLEEP */
7393 static int igb_resume(struct device
*dev
)
7395 struct pci_dev
*pdev
= to_pci_dev(dev
);
7396 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7397 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7398 struct e1000_hw
*hw
= &adapter
->hw
;
7401 pci_set_power_state(pdev
, PCI_D0
);
7402 pci_restore_state(pdev
);
7403 pci_save_state(pdev
);
7405 err
= pci_enable_device_mem(pdev
);
7408 "igb: Cannot enable PCI device from suspend\n");
7411 pci_set_master(pdev
);
7413 pci_enable_wake(pdev
, PCI_D3hot
, 0);
7414 pci_enable_wake(pdev
, PCI_D3cold
, 0);
7416 if (igb_init_interrupt_scheme(adapter
, true)) {
7417 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
7423 /* let the f/w know that the h/w is now under the control of the
7426 igb_get_hw_control(adapter
);
7428 wr32(E1000_WUS
, ~0);
7430 if (netdev
->flags
& IFF_UP
) {
7432 err
= __igb_open(netdev
, true);
7438 netif_device_attach(netdev
);
7442 static int igb_runtime_idle(struct device
*dev
)
7444 struct pci_dev
*pdev
= to_pci_dev(dev
);
7445 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7446 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7448 if (!igb_has_link(adapter
))
7449 pm_schedule_suspend(dev
, MSEC_PER_SEC
* 5);
7454 static int igb_runtime_suspend(struct device
*dev
)
7456 struct pci_dev
*pdev
= to_pci_dev(dev
);
7460 retval
= __igb_shutdown(pdev
, &wake
, 1);
7465 pci_prepare_to_sleep(pdev
);
7467 pci_wake_from_d3(pdev
, false);
7468 pci_set_power_state(pdev
, PCI_D3hot
);
7474 static int igb_runtime_resume(struct device
*dev
)
7476 return igb_resume(dev
);
7478 #endif /* CONFIG_PM */
7480 static void igb_shutdown(struct pci_dev
*pdev
)
7484 __igb_shutdown(pdev
, &wake
, 0);
7486 if (system_state
== SYSTEM_POWER_OFF
) {
7487 pci_wake_from_d3(pdev
, wake
);
7488 pci_set_power_state(pdev
, PCI_D3hot
);
7492 #ifdef CONFIG_PCI_IOV
7493 static int igb_sriov_reinit(struct pci_dev
*dev
)
7495 struct net_device
*netdev
= pci_get_drvdata(dev
);
7496 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7497 struct pci_dev
*pdev
= adapter
->pdev
;
7501 if (netif_running(netdev
))
7506 igb_clear_interrupt_scheme(adapter
);
7508 igb_init_queue_configuration(adapter
);
7510 if (igb_init_interrupt_scheme(adapter
, true)) {
7511 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
7515 if (netif_running(netdev
))
7523 static int igb_pci_disable_sriov(struct pci_dev
*dev
)
7525 int err
= igb_disable_sriov(dev
);
7528 err
= igb_sriov_reinit(dev
);
7533 static int igb_pci_enable_sriov(struct pci_dev
*dev
, int num_vfs
)
7535 int err
= igb_enable_sriov(dev
, num_vfs
);
7540 err
= igb_sriov_reinit(dev
);
7549 static int igb_pci_sriov_configure(struct pci_dev
*dev
, int num_vfs
)
7551 #ifdef CONFIG_PCI_IOV
7553 return igb_pci_disable_sriov(dev
);
7555 return igb_pci_enable_sriov(dev
, num_vfs
);
7560 #ifdef CONFIG_NET_POLL_CONTROLLER
7561 /* Polling 'interrupt' - used by things like netconsole to send skbs
7562 * without having to re-enable interrupts. It's not called while
7563 * the interrupt routine is executing.
7565 static void igb_netpoll(struct net_device
*netdev
)
7567 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7568 struct e1000_hw
*hw
= &adapter
->hw
;
7569 struct igb_q_vector
*q_vector
;
7572 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
7573 q_vector
= adapter
->q_vector
[i
];
7574 if (adapter
->flags
& IGB_FLAG_HAS_MSIX
)
7575 wr32(E1000_EIMC
, q_vector
->eims_value
);
7577 igb_irq_disable(adapter
);
7578 napi_schedule(&q_vector
->napi
);
7581 #endif /* CONFIG_NET_POLL_CONTROLLER */
7584 * igb_io_error_detected - called when PCI error is detected
7585 * @pdev: Pointer to PCI device
7586 * @state: The current pci connection state
7588 * This function is called after a PCI bus error affecting
7589 * this device has been detected.
7591 static pci_ers_result_t
igb_io_error_detected(struct pci_dev
*pdev
,
7592 pci_channel_state_t state
)
7594 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7595 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7597 netif_device_detach(netdev
);
7599 if (state
== pci_channel_io_perm_failure
)
7600 return PCI_ERS_RESULT_DISCONNECT
;
7602 if (netif_running(netdev
))
7604 pci_disable_device(pdev
);
7606 /* Request a slot slot reset. */
7607 return PCI_ERS_RESULT_NEED_RESET
;
7611 * igb_io_slot_reset - called after the pci bus has been reset.
7612 * @pdev: Pointer to PCI device
7614 * Restart the card from scratch, as if from a cold-boot. Implementation
7615 * resembles the first-half of the igb_resume routine.
7617 static pci_ers_result_t
igb_io_slot_reset(struct pci_dev
*pdev
)
7619 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7620 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7621 struct e1000_hw
*hw
= &adapter
->hw
;
7622 pci_ers_result_t result
;
7625 if (pci_enable_device_mem(pdev
)) {
7627 "Cannot re-enable PCI device after reset.\n");
7628 result
= PCI_ERS_RESULT_DISCONNECT
;
7630 pci_set_master(pdev
);
7631 pci_restore_state(pdev
);
7632 pci_save_state(pdev
);
7634 pci_enable_wake(pdev
, PCI_D3hot
, 0);
7635 pci_enable_wake(pdev
, PCI_D3cold
, 0);
7638 wr32(E1000_WUS
, ~0);
7639 result
= PCI_ERS_RESULT_RECOVERED
;
7642 err
= pci_cleanup_aer_uncorrect_error_status(pdev
);
7645 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
7647 /* non-fatal, continue */
7654 * igb_io_resume - called when traffic can start flowing again.
7655 * @pdev: Pointer to PCI device
7657 * This callback is called when the error recovery driver tells us that
7658 * its OK to resume normal operation. Implementation resembles the
7659 * second-half of the igb_resume routine.
7661 static void igb_io_resume(struct pci_dev
*pdev
)
7663 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7664 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7666 if (netif_running(netdev
)) {
7667 if (igb_up(adapter
)) {
7668 dev_err(&pdev
->dev
, "igb_up failed after reset\n");
7673 netif_device_attach(netdev
);
7675 /* let the f/w know that the h/w is now under the control of the
7678 igb_get_hw_control(adapter
);
7681 static void igb_rar_set_qsel(struct igb_adapter
*adapter
, u8
*addr
, u32 index
,
7684 u32 rar_low
, rar_high
;
7685 struct e1000_hw
*hw
= &adapter
->hw
;
7687 /* HW expects these in little endian so we reverse the byte order
7688 * from network order (big endian) to little endian
7690 rar_low
= ((u32
) addr
[0] | ((u32
) addr
[1] << 8) |
7691 ((u32
) addr
[2] << 16) | ((u32
) addr
[3] << 24));
7692 rar_high
= ((u32
) addr
[4] | ((u32
) addr
[5] << 8));
7694 /* Indicate to hardware the Address is Valid. */
7695 rar_high
|= E1000_RAH_AV
;
7697 if (hw
->mac
.type
== e1000_82575
)
7698 rar_high
|= E1000_RAH_POOL_1
* qsel
;
7700 rar_high
|= E1000_RAH_POOL_1
<< qsel
;
7702 wr32(E1000_RAL(index
), rar_low
);
7704 wr32(E1000_RAH(index
), rar_high
);
7708 static int igb_set_vf_mac(struct igb_adapter
*adapter
,
7709 int vf
, unsigned char *mac_addr
)
7711 struct e1000_hw
*hw
= &adapter
->hw
;
7712 /* VF MAC addresses start at end of receive addresses and moves
7713 * towards the first, as a result a collision should not be possible
7715 int rar_entry
= hw
->mac
.rar_entry_count
- (vf
+ 1);
7717 memcpy(adapter
->vf_data
[vf
].vf_mac_addresses
, mac_addr
, ETH_ALEN
);
7719 igb_rar_set_qsel(adapter
, mac_addr
, rar_entry
, vf
);
7724 static int igb_ndo_set_vf_mac(struct net_device
*netdev
, int vf
, u8
*mac
)
7726 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7727 if (!is_valid_ether_addr(mac
) || (vf
>= adapter
->vfs_allocated_count
))
7729 adapter
->vf_data
[vf
].flags
|= IGB_VF_FLAG_PF_SET_MAC
;
7730 dev_info(&adapter
->pdev
->dev
, "setting MAC %pM on VF %d\n", mac
, vf
);
7731 dev_info(&adapter
->pdev
->dev
,
7732 "Reload the VF driver to make this change effective.");
7733 if (test_bit(__IGB_DOWN
, &adapter
->state
)) {
7734 dev_warn(&adapter
->pdev
->dev
,
7735 "The VF MAC address has been set, but the PF device is not up.\n");
7736 dev_warn(&adapter
->pdev
->dev
,
7737 "Bring the PF device up before attempting to use the VF device.\n");
7739 return igb_set_vf_mac(adapter
, vf
, mac
);
7742 static int igb_link_mbps(int internal_link_speed
)
7744 switch (internal_link_speed
) {
7754 static void igb_set_vf_rate_limit(struct e1000_hw
*hw
, int vf
, int tx_rate
,
7761 /* Calculate the rate factor values to set */
7762 rf_int
= link_speed
/ tx_rate
;
7763 rf_dec
= (link_speed
- (rf_int
* tx_rate
));
7764 rf_dec
= (rf_dec
* (1 << E1000_RTTBCNRC_RF_INT_SHIFT
)) /
7767 bcnrc_val
= E1000_RTTBCNRC_RS_ENA
;
7768 bcnrc_val
|= ((rf_int
<< E1000_RTTBCNRC_RF_INT_SHIFT
) &
7769 E1000_RTTBCNRC_RF_INT_MASK
);
7770 bcnrc_val
|= (rf_dec
& E1000_RTTBCNRC_RF_DEC_MASK
);
7775 wr32(E1000_RTTDQSEL
, vf
); /* vf X uses queue X */
7776 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7777 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7779 wr32(E1000_RTTBCNRM
, 0x14);
7780 wr32(E1000_RTTBCNRC
, bcnrc_val
);
7783 static void igb_check_vf_rate_limit(struct igb_adapter
*adapter
)
7785 int actual_link_speed
, i
;
7786 bool reset_rate
= false;
7788 /* VF TX rate limit was not set or not supported */
7789 if ((adapter
->vf_rate_link_speed
== 0) ||
7790 (adapter
->hw
.mac
.type
!= e1000_82576
))
7793 actual_link_speed
= igb_link_mbps(adapter
->link_speed
);
7794 if (actual_link_speed
!= adapter
->vf_rate_link_speed
) {
7796 adapter
->vf_rate_link_speed
= 0;
7797 dev_info(&adapter
->pdev
->dev
,
7798 "Link speed has been changed. VF Transmit rate is disabled\n");
7801 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++) {
7803 adapter
->vf_data
[i
].tx_rate
= 0;
7805 igb_set_vf_rate_limit(&adapter
->hw
, i
,
7806 adapter
->vf_data
[i
].tx_rate
,
7811 static int igb_ndo_set_vf_bw(struct net_device
*netdev
, int vf
,
7812 int min_tx_rate
, int max_tx_rate
)
7814 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7815 struct e1000_hw
*hw
= &adapter
->hw
;
7816 int actual_link_speed
;
7818 if (hw
->mac
.type
!= e1000_82576
)
7824 actual_link_speed
= igb_link_mbps(adapter
->link_speed
);
7825 if ((vf
>= adapter
->vfs_allocated_count
) ||
7826 (!(rd32(E1000_STATUS
) & E1000_STATUS_LU
)) ||
7827 (max_tx_rate
< 0) ||
7828 (max_tx_rate
> actual_link_speed
))
7831 adapter
->vf_rate_link_speed
= actual_link_speed
;
7832 adapter
->vf_data
[vf
].tx_rate
= (u16
)max_tx_rate
;
7833 igb_set_vf_rate_limit(hw
, vf
, max_tx_rate
, actual_link_speed
);
7838 static int igb_ndo_set_vf_spoofchk(struct net_device
*netdev
, int vf
,
7841 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7842 struct e1000_hw
*hw
= &adapter
->hw
;
7843 u32 reg_val
, reg_offset
;
7845 if (!adapter
->vfs_allocated_count
)
7848 if (vf
>= adapter
->vfs_allocated_count
)
7851 reg_offset
= (hw
->mac
.type
== e1000_82576
) ? E1000_DTXSWC
: E1000_TXSWC
;
7852 reg_val
= rd32(reg_offset
);
7854 reg_val
|= ((1 << vf
) |
7855 (1 << (vf
+ E1000_DTXSWC_VLAN_SPOOF_SHIFT
)));
7857 reg_val
&= ~((1 << vf
) |
7858 (1 << (vf
+ E1000_DTXSWC_VLAN_SPOOF_SHIFT
)));
7859 wr32(reg_offset
, reg_val
);
7861 adapter
->vf_data
[vf
].spoofchk_enabled
= setting
;
7865 static int igb_ndo_get_vf_config(struct net_device
*netdev
,
7866 int vf
, struct ifla_vf_info
*ivi
)
7868 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7869 if (vf
>= adapter
->vfs_allocated_count
)
7872 memcpy(&ivi
->mac
, adapter
->vf_data
[vf
].vf_mac_addresses
, ETH_ALEN
);
7873 ivi
->max_tx_rate
= adapter
->vf_data
[vf
].tx_rate
;
7874 ivi
->min_tx_rate
= 0;
7875 ivi
->vlan
= adapter
->vf_data
[vf
].pf_vlan
;
7876 ivi
->qos
= adapter
->vf_data
[vf
].pf_qos
;
7877 ivi
->spoofchk
= adapter
->vf_data
[vf
].spoofchk_enabled
;
7881 static void igb_vmm_control(struct igb_adapter
*adapter
)
7883 struct e1000_hw
*hw
= &adapter
->hw
;
7886 switch (hw
->mac
.type
) {
7892 /* replication is not supported for 82575 */
7895 /* notify HW that the MAC is adding vlan tags */
7896 reg
= rd32(E1000_DTXCTL
);
7897 reg
|= E1000_DTXCTL_VLAN_ADDED
;
7898 wr32(E1000_DTXCTL
, reg
);
7901 /* enable replication vlan tag stripping */
7902 reg
= rd32(E1000_RPLOLR
);
7903 reg
|= E1000_RPLOLR_STRVLAN
;
7904 wr32(E1000_RPLOLR
, reg
);
7907 /* none of the above registers are supported by i350 */
7911 if (adapter
->vfs_allocated_count
) {
7912 igb_vmdq_set_loopback_pf(hw
, true);
7913 igb_vmdq_set_replication_pf(hw
, true);
7914 igb_vmdq_set_anti_spoofing_pf(hw
, true,
7915 adapter
->vfs_allocated_count
);
7917 igb_vmdq_set_loopback_pf(hw
, false);
7918 igb_vmdq_set_replication_pf(hw
, false);
7922 static void igb_init_dmac(struct igb_adapter
*adapter
, u32 pba
)
7924 struct e1000_hw
*hw
= &adapter
->hw
;
7928 if (hw
->mac
.type
> e1000_82580
) {
7929 if (adapter
->flags
& IGB_FLAG_DMAC
) {
7932 /* force threshold to 0. */
7933 wr32(E1000_DMCTXTH
, 0);
7935 /* DMA Coalescing high water mark needs to be greater
7936 * than the Rx threshold. Set hwm to PBA - max frame
7937 * size in 16B units, capping it at PBA - 6KB.
7939 hwm
= 64 * pba
- adapter
->max_frame_size
/ 16;
7940 if (hwm
< 64 * (pba
- 6))
7941 hwm
= 64 * (pba
- 6);
7942 reg
= rd32(E1000_FCRTC
);
7943 reg
&= ~E1000_FCRTC_RTH_COAL_MASK
;
7944 reg
|= ((hwm
<< E1000_FCRTC_RTH_COAL_SHIFT
)
7945 & E1000_FCRTC_RTH_COAL_MASK
);
7946 wr32(E1000_FCRTC
, reg
);
7948 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
7949 * frame size, capping it at PBA - 10KB.
7951 dmac_thr
= pba
- adapter
->max_frame_size
/ 512;
7952 if (dmac_thr
< pba
- 10)
7953 dmac_thr
= pba
- 10;
7954 reg
= rd32(E1000_DMACR
);
7955 reg
&= ~E1000_DMACR_DMACTHR_MASK
;
7956 reg
|= ((dmac_thr
<< E1000_DMACR_DMACTHR_SHIFT
)
7957 & E1000_DMACR_DMACTHR_MASK
);
7959 /* transition to L0x or L1 if available..*/
7960 reg
|= (E1000_DMACR_DMAC_EN
| E1000_DMACR_DMAC_LX_MASK
);
7962 /* watchdog timer= +-1000 usec in 32usec intervals */
7965 /* Disable BMC-to-OS Watchdog Enable */
7966 if (hw
->mac
.type
!= e1000_i354
)
7967 reg
&= ~E1000_DMACR_DC_BMC2OSW_EN
;
7969 wr32(E1000_DMACR
, reg
);
7971 /* no lower threshold to disable
7972 * coalescing(smart fifb)-UTRESH=0
7974 wr32(E1000_DMCRTRH
, 0);
7976 reg
= (IGB_DMCTLX_DCFLUSH_DIS
| 0x4);
7978 wr32(E1000_DMCTLX
, reg
);
7980 /* free space in tx packet buffer to wake from
7983 wr32(E1000_DMCTXTH
, (IGB_MIN_TXPBSIZE
-
7984 (IGB_TX_BUF_4096
+ adapter
->max_frame_size
)) >> 6);
7986 /* make low power state decision controlled
7989 reg
= rd32(E1000_PCIEMISC
);
7990 reg
&= ~E1000_PCIEMISC_LX_DECISION
;
7991 wr32(E1000_PCIEMISC
, reg
);
7992 } /* endif adapter->dmac is not disabled */
7993 } else if (hw
->mac
.type
== e1000_82580
) {
7994 u32 reg
= rd32(E1000_PCIEMISC
);
7996 wr32(E1000_PCIEMISC
, reg
& ~E1000_PCIEMISC_LX_DECISION
);
7997 wr32(E1000_DMACR
, 0);
8002 * igb_read_i2c_byte - Reads 8 bit word over I2C
8003 * @hw: pointer to hardware structure
8004 * @byte_offset: byte offset to read
8005 * @dev_addr: device address
8008 * Performs byte read operation over I2C interface at
8009 * a specified device address.
8011 s32
igb_read_i2c_byte(struct e1000_hw
*hw
, u8 byte_offset
,
8012 u8 dev_addr
, u8
*data
)
8014 struct igb_adapter
*adapter
= container_of(hw
, struct igb_adapter
, hw
);
8015 struct i2c_client
*this_client
= adapter
->i2c_client
;
8020 return E1000_ERR_I2C
;
8022 swfw_mask
= E1000_SWFW_PHY0_SM
;
8024 if (hw
->mac
.ops
.acquire_swfw_sync(hw
, swfw_mask
))
8025 return E1000_ERR_SWFW_SYNC
;
8027 status
= i2c_smbus_read_byte_data(this_client
, byte_offset
);
8028 hw
->mac
.ops
.release_swfw_sync(hw
, swfw_mask
);
8031 return E1000_ERR_I2C
;
8039 * igb_write_i2c_byte - Writes 8 bit word over I2C
8040 * @hw: pointer to hardware structure
8041 * @byte_offset: byte offset to write
8042 * @dev_addr: device address
8043 * @data: value to write
8045 * Performs byte write operation over I2C interface at
8046 * a specified device address.
8048 s32
igb_write_i2c_byte(struct e1000_hw
*hw
, u8 byte_offset
,
8049 u8 dev_addr
, u8 data
)
8051 struct igb_adapter
*adapter
= container_of(hw
, struct igb_adapter
, hw
);
8052 struct i2c_client
*this_client
= adapter
->i2c_client
;
8054 u16 swfw_mask
= E1000_SWFW_PHY0_SM
;
8057 return E1000_ERR_I2C
;
8059 if (hw
->mac
.ops
.acquire_swfw_sync(hw
, swfw_mask
))
8060 return E1000_ERR_SWFW_SYNC
;
8061 status
= i2c_smbus_write_byte_data(this_client
, byte_offset
, data
);
8062 hw
->mac
.ops
.release_swfw_sync(hw
, swfw_mask
);
8065 return E1000_ERR_I2C
;
8071 int igb_reinit_queues(struct igb_adapter
*adapter
)
8073 struct net_device
*netdev
= adapter
->netdev
;
8074 struct pci_dev
*pdev
= adapter
->pdev
;
8077 if (netif_running(netdev
))
8080 igb_reset_interrupt_capability(adapter
);
8082 if (igb_init_interrupt_scheme(adapter
, true)) {
8083 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
8087 if (netif_running(netdev
))
8088 err
= igb_open(netdev
);