1 /*******************************************************************************
3 Intel(R) Gigabit Ethernet Linux driver
4 Copyright(c) 2007-2012 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30 #include <linux/module.h>
31 #include <linux/types.h>
32 #include <linux/init.h>
33 #include <linux/bitops.h>
34 #include <linux/vmalloc.h>
35 #include <linux/pagemap.h>
36 #include <linux/netdevice.h>
37 #include <linux/ipv6.h>
38 #include <linux/slab.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/net_tstamp.h>
42 #include <linux/mii.h>
43 #include <linux/ethtool.h>
45 #include <linux/if_vlan.h>
46 #include <linux/pci.h>
47 #include <linux/pci-aspm.h>
48 #include <linux/delay.h>
49 #include <linux/interrupt.h>
51 #include <linux/tcp.h>
52 #include <linux/sctp.h>
53 #include <linux/if_ether.h>
54 #include <linux/aer.h>
55 #include <linux/prefetch.h>
56 #include <linux/pm_runtime.h>
58 #include <linux/dca.h>
65 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
66 __stringify(BUILD) "-k"
67 char igb_driver_name
[] = "igb";
68 char igb_driver_version
[] = DRV_VERSION
;
69 static const char igb_driver_string
[] =
70 "Intel(R) Gigabit Ethernet Network Driver";
71 static const char igb_copyright
[] = "Copyright (c) 2007-2012 Intel Corporation.";
73 static const struct e1000_info
*igb_info_tbl
[] = {
74 [board_82575
] = &e1000_82575_info
,
77 static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl
) = {
78 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I211_COPPER
), board_82575
},
79 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_COPPER
), board_82575
},
80 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_FIBER
), board_82575
},
81 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_SERDES
), board_82575
},
82 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_SGMII
), board_82575
},
83 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I350_COPPER
), board_82575
},
84 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I350_FIBER
), board_82575
},
85 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I350_SERDES
), board_82575
},
86 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I350_SGMII
), board_82575
},
87 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_COPPER
), board_82575
},
88 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_FIBER
), board_82575
},
89 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_QUAD_FIBER
), board_82575
},
90 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_SERDES
), board_82575
},
91 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_SGMII
), board_82575
},
92 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_COPPER_DUAL
), board_82575
},
93 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_DH89XXCC_SGMII
), board_82575
},
94 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_DH89XXCC_SERDES
), board_82575
},
95 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_DH89XXCC_BACKPLANE
), board_82575
},
96 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_DH89XXCC_SFP
), board_82575
},
97 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576
), board_82575
},
98 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_NS
), board_82575
},
99 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_NS_SERDES
), board_82575
},
100 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_FIBER
), board_82575
},
101 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_SERDES
), board_82575
},
102 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_SERDES_QUAD
), board_82575
},
103 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_QUAD_COPPER_ET2
), board_82575
},
104 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_QUAD_COPPER
), board_82575
},
105 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82575EB_COPPER
), board_82575
},
106 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82575EB_FIBER_SERDES
), board_82575
},
107 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82575GB_QUAD_COPPER
), board_82575
},
108 /* required last entry */
112 MODULE_DEVICE_TABLE(pci
, igb_pci_tbl
);
114 void igb_reset(struct igb_adapter
*);
115 static int igb_setup_all_tx_resources(struct igb_adapter
*);
116 static int igb_setup_all_rx_resources(struct igb_adapter
*);
117 static void igb_free_all_tx_resources(struct igb_adapter
*);
118 static void igb_free_all_rx_resources(struct igb_adapter
*);
119 static void igb_setup_mrqc(struct igb_adapter
*);
120 static int igb_probe(struct pci_dev
*, const struct pci_device_id
*);
121 static void __devexit
igb_remove(struct pci_dev
*pdev
);
122 static int igb_sw_init(struct igb_adapter
*);
123 static int igb_open(struct net_device
*);
124 static int igb_close(struct net_device
*);
125 static void igb_configure_tx(struct igb_adapter
*);
126 static void igb_configure_rx(struct igb_adapter
*);
127 static void igb_clean_all_tx_rings(struct igb_adapter
*);
128 static void igb_clean_all_rx_rings(struct igb_adapter
*);
129 static void igb_clean_tx_ring(struct igb_ring
*);
130 static void igb_clean_rx_ring(struct igb_ring
*);
131 static void igb_set_rx_mode(struct net_device
*);
132 static void igb_update_phy_info(unsigned long);
133 static void igb_watchdog(unsigned long);
134 static void igb_watchdog_task(struct work_struct
*);
135 static netdev_tx_t
igb_xmit_frame(struct sk_buff
*skb
, struct net_device
*);
136 static struct rtnl_link_stats64
*igb_get_stats64(struct net_device
*dev
,
137 struct rtnl_link_stats64
*stats
);
138 static int igb_change_mtu(struct net_device
*, int);
139 static int igb_set_mac(struct net_device
*, void *);
140 static void igb_set_uta(struct igb_adapter
*adapter
);
141 static irqreturn_t
igb_intr(int irq
, void *);
142 static irqreturn_t
igb_intr_msi(int irq
, void *);
143 static irqreturn_t
igb_msix_other(int irq
, void *);
144 static irqreturn_t
igb_msix_ring(int irq
, void *);
145 #ifdef CONFIG_IGB_DCA
146 static void igb_update_dca(struct igb_q_vector
*);
147 static void igb_setup_dca(struct igb_adapter
*);
148 #endif /* CONFIG_IGB_DCA */
149 static int igb_poll(struct napi_struct
*, int);
150 static bool igb_clean_tx_irq(struct igb_q_vector
*);
151 static bool igb_clean_rx_irq(struct igb_q_vector
*, int);
152 static int igb_ioctl(struct net_device
*, struct ifreq
*, int cmd
);
153 static void igb_tx_timeout(struct net_device
*);
154 static void igb_reset_task(struct work_struct
*);
155 static void igb_vlan_mode(struct net_device
*netdev
, netdev_features_t features
);
156 static int igb_vlan_rx_add_vid(struct net_device
*, u16
);
157 static int igb_vlan_rx_kill_vid(struct net_device
*, u16
);
158 static void igb_restore_vlan(struct igb_adapter
*);
159 static void igb_rar_set_qsel(struct igb_adapter
*, u8
*, u32
, u8
);
160 static void igb_ping_all_vfs(struct igb_adapter
*);
161 static void igb_msg_task(struct igb_adapter
*);
162 static void igb_vmm_control(struct igb_adapter
*);
163 static int igb_set_vf_mac(struct igb_adapter
*, int, unsigned char *);
164 static void igb_restore_vf_multicasts(struct igb_adapter
*adapter
);
165 static int igb_ndo_set_vf_mac(struct net_device
*netdev
, int vf
, u8
*mac
);
166 static int igb_ndo_set_vf_vlan(struct net_device
*netdev
,
167 int vf
, u16 vlan
, u8 qos
);
168 static int igb_ndo_set_vf_bw(struct net_device
*netdev
, int vf
, int tx_rate
);
169 static int igb_ndo_get_vf_config(struct net_device
*netdev
, int vf
,
170 struct ifla_vf_info
*ivi
);
171 static void igb_check_vf_rate_limit(struct igb_adapter
*);
173 #ifdef CONFIG_PCI_IOV
174 static int igb_vf_configure(struct igb_adapter
*adapter
, int vf
);
175 static int igb_find_enabled_vfs(struct igb_adapter
*adapter
);
176 static int igb_check_vf_assignment(struct igb_adapter
*adapter
);
180 #ifdef CONFIG_PM_SLEEP
181 static int igb_suspend(struct device
*);
183 static int igb_resume(struct device
*);
184 #ifdef CONFIG_PM_RUNTIME
185 static int igb_runtime_suspend(struct device
*dev
);
186 static int igb_runtime_resume(struct device
*dev
);
187 static int igb_runtime_idle(struct device
*dev
);
189 static const struct dev_pm_ops igb_pm_ops
= {
190 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend
, igb_resume
)
191 SET_RUNTIME_PM_OPS(igb_runtime_suspend
, igb_runtime_resume
,
195 static void igb_shutdown(struct pci_dev
*);
196 #ifdef CONFIG_IGB_DCA
197 static int igb_notify_dca(struct notifier_block
*, unsigned long, void *);
198 static struct notifier_block dca_notifier
= {
199 .notifier_call
= igb_notify_dca
,
204 #ifdef CONFIG_NET_POLL_CONTROLLER
205 /* for netdump / net console */
206 static void igb_netpoll(struct net_device
*);
208 #ifdef CONFIG_PCI_IOV
209 static unsigned int max_vfs
= 0;
210 module_param(max_vfs
, uint
, 0);
211 MODULE_PARM_DESC(max_vfs
, "Maximum number of virtual functions to allocate "
212 "per physical function");
213 #endif /* CONFIG_PCI_IOV */
215 static pci_ers_result_t
igb_io_error_detected(struct pci_dev
*,
216 pci_channel_state_t
);
217 static pci_ers_result_t
igb_io_slot_reset(struct pci_dev
*);
218 static void igb_io_resume(struct pci_dev
*);
220 static struct pci_error_handlers igb_err_handler
= {
221 .error_detected
= igb_io_error_detected
,
222 .slot_reset
= igb_io_slot_reset
,
223 .resume
= igb_io_resume
,
226 static void igb_init_dmac(struct igb_adapter
*adapter
, u32 pba
);
228 static struct pci_driver igb_driver
= {
229 .name
= igb_driver_name
,
230 .id_table
= igb_pci_tbl
,
232 .remove
= __devexit_p(igb_remove
),
234 .driver
.pm
= &igb_pm_ops
,
236 .shutdown
= igb_shutdown
,
237 .err_handler
= &igb_err_handler
240 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
241 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
242 MODULE_LICENSE("GPL");
243 MODULE_VERSION(DRV_VERSION
);
245 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
246 static int debug
= -1;
247 module_param(debug
, int, 0);
248 MODULE_PARM_DESC(debug
, "Debug level (0=none,...,16=all)");
250 struct igb_reg_info
{
255 static const struct igb_reg_info igb_reg_info_tbl
[] = {
257 /* General Registers */
258 {E1000_CTRL
, "CTRL"},
259 {E1000_STATUS
, "STATUS"},
260 {E1000_CTRL_EXT
, "CTRL_EXT"},
262 /* Interrupt Registers */
266 {E1000_RCTL
, "RCTL"},
267 {E1000_RDLEN(0), "RDLEN"},
268 {E1000_RDH(0), "RDH"},
269 {E1000_RDT(0), "RDT"},
270 {E1000_RXDCTL(0), "RXDCTL"},
271 {E1000_RDBAL(0), "RDBAL"},
272 {E1000_RDBAH(0), "RDBAH"},
275 {E1000_TCTL
, "TCTL"},
276 {E1000_TDBAL(0), "TDBAL"},
277 {E1000_TDBAH(0), "TDBAH"},
278 {E1000_TDLEN(0), "TDLEN"},
279 {E1000_TDH(0), "TDH"},
280 {E1000_TDT(0), "TDT"},
281 {E1000_TXDCTL(0), "TXDCTL"},
282 {E1000_TDFH
, "TDFH"},
283 {E1000_TDFT
, "TDFT"},
284 {E1000_TDFHS
, "TDFHS"},
285 {E1000_TDFPC
, "TDFPC"},
287 /* List Terminator */
292 * igb_regdump - register printout routine
294 static void igb_regdump(struct e1000_hw
*hw
, struct igb_reg_info
*reginfo
)
300 switch (reginfo
->ofs
) {
302 for (n
= 0; n
< 4; n
++)
303 regs
[n
] = rd32(E1000_RDLEN(n
));
306 for (n
= 0; n
< 4; n
++)
307 regs
[n
] = rd32(E1000_RDH(n
));
310 for (n
= 0; n
< 4; n
++)
311 regs
[n
] = rd32(E1000_RDT(n
));
313 case E1000_RXDCTL(0):
314 for (n
= 0; n
< 4; n
++)
315 regs
[n
] = rd32(E1000_RXDCTL(n
));
318 for (n
= 0; n
< 4; n
++)
319 regs
[n
] = rd32(E1000_RDBAL(n
));
322 for (n
= 0; n
< 4; n
++)
323 regs
[n
] = rd32(E1000_RDBAH(n
));
326 for (n
= 0; n
< 4; n
++)
327 regs
[n
] = rd32(E1000_RDBAL(n
));
330 for (n
= 0; n
< 4; n
++)
331 regs
[n
] = rd32(E1000_TDBAH(n
));
334 for (n
= 0; n
< 4; n
++)
335 regs
[n
] = rd32(E1000_TDLEN(n
));
338 for (n
= 0; n
< 4; n
++)
339 regs
[n
] = rd32(E1000_TDH(n
));
342 for (n
= 0; n
< 4; n
++)
343 regs
[n
] = rd32(E1000_TDT(n
));
345 case E1000_TXDCTL(0):
346 for (n
= 0; n
< 4; n
++)
347 regs
[n
] = rd32(E1000_TXDCTL(n
));
350 pr_info("%-15s %08x\n", reginfo
->name
, rd32(reginfo
->ofs
));
354 snprintf(rname
, 16, "%s%s", reginfo
->name
, "[0-3]");
355 pr_info("%-15s %08x %08x %08x %08x\n", rname
, regs
[0], regs
[1],
360 * 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 "
383 pr_info("%-15s %016lX %016lX %016lX\n", netdev
->name
,
384 netdev
->state
, netdev
->trans_start
, netdev
->last_rx
);
387 /* Print Registers */
388 dev_info(&adapter
->pdev
->dev
, "Register Dump\n");
389 pr_info(" Register Name Value\n");
390 for (reginfo
= (struct igb_reg_info
*)igb_reg_info_tbl
;
391 reginfo
->name
; reginfo
++) {
392 igb_regdump(hw
, reginfo
);
395 /* Print TX Ring Summary */
396 if (!netdev
|| !netif_running(netdev
))
399 dev_info(&adapter
->pdev
->dev
, "TX Rings Summary\n");
400 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
401 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
402 struct igb_tx_buffer
*buffer_info
;
403 tx_ring
= adapter
->tx_ring
[n
];
404 buffer_info
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_clean
];
405 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
406 n
, tx_ring
->next_to_use
, tx_ring
->next_to_clean
,
407 (u64
)buffer_info
->dma
,
409 buffer_info
->next_to_watch
,
410 (u64
)buffer_info
->time_stamp
);
414 if (!netif_msg_tx_done(adapter
))
415 goto rx_ring_summary
;
417 dev_info(&adapter
->pdev
->dev
, "TX Rings Dump\n");
419 /* Transmit Descriptor Formats
421 * Advanced Transmit Descriptor
422 * +--------------------------------------------------------------+
423 * 0 | Buffer Address [63:0] |
424 * +--------------------------------------------------------------+
425 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
426 * +--------------------------------------------------------------+
427 * 63 46 45 40 39 38 36 35 32 31 24 15 0
430 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
431 tx_ring
= adapter
->tx_ring
[n
];
432 pr_info("------------------------------------\n");
433 pr_info("TX QUEUE INDEX = %d\n", tx_ring
->queue_index
);
434 pr_info("------------------------------------\n");
435 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] "
436 "[bi->dma ] leng ntw timestamp "
439 for (i
= 0; tx_ring
->desc
&& (i
< tx_ring
->count
); i
++) {
440 const char *next_desc
;
441 struct igb_tx_buffer
*buffer_info
;
442 tx_desc
= IGB_TX_DESC(tx_ring
, i
);
443 buffer_info
= &tx_ring
->tx_buffer_info
[i
];
444 u0
= (struct my_u0
*)tx_desc
;
445 if (i
== tx_ring
->next_to_use
&&
446 i
== tx_ring
->next_to_clean
)
447 next_desc
= " NTC/U";
448 else if (i
== tx_ring
->next_to_use
)
450 else if (i
== tx_ring
->next_to_clean
)
455 pr_info("T [0x%03X] %016llX %016llX %016llX"
456 " %04X %p %016llX %p%s\n", i
,
459 (u64
)buffer_info
->dma
,
461 buffer_info
->next_to_watch
,
462 (u64
)buffer_info
->time_stamp
,
463 buffer_info
->skb
, next_desc
);
465 if (netif_msg_pktdata(adapter
) && buffer_info
->skb
)
466 print_hex_dump(KERN_INFO
, "",
468 16, 1, buffer_info
->skb
->data
,
469 buffer_info
->length
, true);
473 /* Print RX Rings Summary */
475 dev_info(&adapter
->pdev
->dev
, "RX Rings Summary\n");
476 pr_info("Queue [NTU] [NTC]\n");
477 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
478 rx_ring
= adapter
->rx_ring
[n
];
479 pr_info(" %5d %5X %5X\n",
480 n
, rx_ring
->next_to_use
, rx_ring
->next_to_clean
);
484 if (!netif_msg_rx_status(adapter
))
487 dev_info(&adapter
->pdev
->dev
, "RX Rings Dump\n");
489 /* Advanced Receive Descriptor (Read) Format
491 * +-----------------------------------------------------+
492 * 0 | Packet Buffer Address [63:1] |A0/NSE|
493 * +----------------------------------------------+------+
494 * 8 | Header Buffer Address [63:1] | DD |
495 * +-----------------------------------------------------+
498 * Advanced Receive Descriptor (Write-Back) Format
500 * 63 48 47 32 31 30 21 20 17 16 4 3 0
501 * +------------------------------------------------------+
502 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
503 * | Checksum Ident | | | | Type | Type |
504 * +------------------------------------------------------+
505 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
506 * +------------------------------------------------------+
507 * 63 48 47 32 31 20 19 0
510 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
511 rx_ring
= adapter
->rx_ring
[n
];
512 pr_info("------------------------------------\n");
513 pr_info("RX QUEUE INDEX = %d\n", rx_ring
->queue_index
);
514 pr_info("------------------------------------\n");
515 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] "
516 "[bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
517 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] -----"
518 "----------- [bi->skb] <-- Adv Rx Write-Back format\n");
520 for (i
= 0; i
< rx_ring
->count
; i
++) {
521 const char *next_desc
;
522 struct igb_rx_buffer
*buffer_info
;
523 buffer_info
= &rx_ring
->rx_buffer_info
[i
];
524 rx_desc
= IGB_RX_DESC(rx_ring
, i
);
525 u0
= (struct my_u0
*)rx_desc
;
526 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
528 if (i
== rx_ring
->next_to_use
)
530 else if (i
== rx_ring
->next_to_clean
)
535 if (staterr
& E1000_RXD_STAT_DD
) {
536 /* Descriptor Done */
537 pr_info("%s[0x%03X] %016llX %016llX -------"
538 "--------- %p%s\n", "RWB", i
,
541 buffer_info
->skb
, next_desc
);
543 pr_info("%s[0x%03X] %016llX %016llX %016llX"
547 (u64
)buffer_info
->dma
,
548 buffer_info
->skb
, next_desc
);
550 if (netif_msg_pktdata(adapter
) &&
551 buffer_info
->dma
&& buffer_info
->skb
) {
552 print_hex_dump(KERN_INFO
, "",
554 16, 1, buffer_info
->skb
->data
,
555 IGB_RX_HDR_LEN
, true);
556 print_hex_dump(KERN_INFO
, "",
559 page_address(buffer_info
->page
) +
560 buffer_info
->page_offset
,
572 * igb_get_hw_dev - return device
573 * used by hardware layer to print debugging information
575 struct net_device
*igb_get_hw_dev(struct e1000_hw
*hw
)
577 struct igb_adapter
*adapter
= hw
->back
;
578 return adapter
->netdev
;
582 * igb_init_module - Driver Registration Routine
584 * igb_init_module is the first routine called when the driver is
585 * loaded. All it does is register with the PCI subsystem.
587 static int __init
igb_init_module(void)
590 pr_info("%s - version %s\n",
591 igb_driver_string
, igb_driver_version
);
593 pr_info("%s\n", igb_copyright
);
595 #ifdef CONFIG_IGB_DCA
596 dca_register_notify(&dca_notifier
);
598 ret
= pci_register_driver(&igb_driver
);
602 module_init(igb_init_module
);
605 * igb_exit_module - Driver Exit Cleanup Routine
607 * igb_exit_module is called just before the driver is removed
610 static void __exit
igb_exit_module(void)
612 #ifdef CONFIG_IGB_DCA
613 dca_unregister_notify(&dca_notifier
);
615 pci_unregister_driver(&igb_driver
);
618 module_exit(igb_exit_module
);
620 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
622 * igb_cache_ring_register - Descriptor ring to register mapping
623 * @adapter: board private structure to initialize
625 * Once we know the feature-set enabled for the device, we'll cache
626 * the register offset the descriptor ring is assigned to.
628 static void igb_cache_ring_register(struct igb_adapter
*adapter
)
631 u32 rbase_offset
= adapter
->vfs_allocated_count
;
633 switch (adapter
->hw
.mac
.type
) {
635 /* The queues are allocated for virtualization such that VF 0
636 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
637 * In order to avoid collision we start at the first free queue
638 * and continue consuming queues in the same sequence
640 if (adapter
->vfs_allocated_count
) {
641 for (; i
< adapter
->rss_queues
; i
++)
642 adapter
->rx_ring
[i
]->reg_idx
= rbase_offset
+
651 for (; i
< adapter
->num_rx_queues
; i
++)
652 adapter
->rx_ring
[i
]->reg_idx
= rbase_offset
+ i
;
653 for (; j
< adapter
->num_tx_queues
; j
++)
654 adapter
->tx_ring
[j
]->reg_idx
= rbase_offset
+ j
;
659 static void igb_free_queues(struct igb_adapter
*adapter
)
663 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
664 kfree(adapter
->tx_ring
[i
]);
665 adapter
->tx_ring
[i
] = NULL
;
667 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
668 kfree(adapter
->rx_ring
[i
]);
669 adapter
->rx_ring
[i
] = NULL
;
671 adapter
->num_rx_queues
= 0;
672 adapter
->num_tx_queues
= 0;
676 * igb_alloc_queues - Allocate memory for all rings
677 * @adapter: board private structure to initialize
679 * We allocate one ring per queue at run-time since we don't know the
680 * number of queues at compile-time.
682 static int igb_alloc_queues(struct igb_adapter
*adapter
)
684 struct igb_ring
*ring
;
686 int orig_node
= adapter
->node
;
688 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
689 if (orig_node
== -1) {
690 int cur_node
= next_online_node(adapter
->node
);
691 if (cur_node
== MAX_NUMNODES
)
692 cur_node
= first_online_node
;
693 adapter
->node
= cur_node
;
695 ring
= kzalloc_node(sizeof(struct igb_ring
), GFP_KERNEL
,
698 ring
= kzalloc(sizeof(struct igb_ring
), GFP_KERNEL
);
701 ring
->count
= adapter
->tx_ring_count
;
702 ring
->queue_index
= i
;
703 ring
->dev
= &adapter
->pdev
->dev
;
704 ring
->netdev
= adapter
->netdev
;
705 ring
->numa_node
= adapter
->node
;
706 /* For 82575, context index must be unique per ring. */
707 if (adapter
->hw
.mac
.type
== e1000_82575
)
708 set_bit(IGB_RING_FLAG_TX_CTX_IDX
, &ring
->flags
);
709 adapter
->tx_ring
[i
] = ring
;
711 /* Restore the adapter's original node */
712 adapter
->node
= orig_node
;
714 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
715 if (orig_node
== -1) {
716 int cur_node
= next_online_node(adapter
->node
);
717 if (cur_node
== MAX_NUMNODES
)
718 cur_node
= first_online_node
;
719 adapter
->node
= cur_node
;
721 ring
= kzalloc_node(sizeof(struct igb_ring
), GFP_KERNEL
,
724 ring
= kzalloc(sizeof(struct igb_ring
), GFP_KERNEL
);
727 ring
->count
= adapter
->rx_ring_count
;
728 ring
->queue_index
= i
;
729 ring
->dev
= &adapter
->pdev
->dev
;
730 ring
->netdev
= adapter
->netdev
;
731 ring
->numa_node
= adapter
->node
;
732 /* set flag indicating ring supports SCTP checksum offload */
733 if (adapter
->hw
.mac
.type
>= e1000_82576
)
734 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM
, &ring
->flags
);
737 * On i350, i210, and i211, loopback VLAN packets
738 * have the tag byte-swapped.
740 if (adapter
->hw
.mac
.type
>= e1000_i350
)
741 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP
, &ring
->flags
);
743 adapter
->rx_ring
[i
] = ring
;
745 /* Restore the adapter's original node */
746 adapter
->node
= orig_node
;
748 igb_cache_ring_register(adapter
);
753 /* Restore the adapter's original node */
754 adapter
->node
= orig_node
;
755 igb_free_queues(adapter
);
761 * igb_write_ivar - configure ivar for given MSI-X vector
762 * @hw: pointer to the HW structure
763 * @msix_vector: vector number we are allocating to a given ring
764 * @index: row index of IVAR register to write within IVAR table
765 * @offset: column offset of in IVAR, should be multiple of 8
767 * This function is intended to handle the writing of the IVAR register
768 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
769 * each containing an cause allocation for an Rx and Tx ring, and a
770 * variable number of rows depending on the number of queues supported.
772 static void igb_write_ivar(struct e1000_hw
*hw
, int msix_vector
,
773 int index
, int offset
)
775 u32 ivar
= array_rd32(E1000_IVAR0
, index
);
777 /* clear any bits that are currently set */
778 ivar
&= ~((u32
)0xFF << offset
);
780 /* write vector and valid bit */
781 ivar
|= (msix_vector
| E1000_IVAR_VALID
) << offset
;
783 array_wr32(E1000_IVAR0
, index
, ivar
);
786 #define IGB_N0_QUEUE -1
787 static void igb_assign_vector(struct igb_q_vector
*q_vector
, int msix_vector
)
789 struct igb_adapter
*adapter
= q_vector
->adapter
;
790 struct e1000_hw
*hw
= &adapter
->hw
;
791 int rx_queue
= IGB_N0_QUEUE
;
792 int tx_queue
= IGB_N0_QUEUE
;
795 if (q_vector
->rx
.ring
)
796 rx_queue
= q_vector
->rx
.ring
->reg_idx
;
797 if (q_vector
->tx
.ring
)
798 tx_queue
= q_vector
->tx
.ring
->reg_idx
;
800 switch (hw
->mac
.type
) {
802 /* The 82575 assigns vectors using a bitmask, which matches the
803 bitmask for the EICR/EIMS/EIMC registers. To assign one
804 or more queues to a vector, we write the appropriate bits
805 into the MSIXBM register for that vector. */
806 if (rx_queue
> IGB_N0_QUEUE
)
807 msixbm
= E1000_EICR_RX_QUEUE0
<< rx_queue
;
808 if (tx_queue
> IGB_N0_QUEUE
)
809 msixbm
|= E1000_EICR_TX_QUEUE0
<< tx_queue
;
810 if (!adapter
->msix_entries
&& msix_vector
== 0)
811 msixbm
|= E1000_EIMS_OTHER
;
812 array_wr32(E1000_MSIXBM(0), msix_vector
, msixbm
);
813 q_vector
->eims_value
= msixbm
;
817 * 82576 uses a table that essentially consists of 2 columns
818 * with 8 rows. The ordering is column-major so we use the
819 * lower 3 bits as the row index, and the 4th bit as the
822 if (rx_queue
> IGB_N0_QUEUE
)
823 igb_write_ivar(hw
, msix_vector
,
825 (rx_queue
& 0x8) << 1);
826 if (tx_queue
> IGB_N0_QUEUE
)
827 igb_write_ivar(hw
, msix_vector
,
829 ((tx_queue
& 0x8) << 1) + 8);
830 q_vector
->eims_value
= 1 << msix_vector
;
837 * On 82580 and newer adapters the scheme is similar to 82576
838 * however instead of ordering column-major we have things
839 * ordered row-major. So we traverse the table by using
840 * bit 0 as the column offset, and the remaining bits as the
843 if (rx_queue
> IGB_N0_QUEUE
)
844 igb_write_ivar(hw
, msix_vector
,
846 (rx_queue
& 0x1) << 4);
847 if (tx_queue
> IGB_N0_QUEUE
)
848 igb_write_ivar(hw
, msix_vector
,
850 ((tx_queue
& 0x1) << 4) + 8);
851 q_vector
->eims_value
= 1 << msix_vector
;
858 /* add q_vector eims value to global eims_enable_mask */
859 adapter
->eims_enable_mask
|= q_vector
->eims_value
;
861 /* configure q_vector to set itr on first interrupt */
862 q_vector
->set_itr
= 1;
866 * igb_configure_msix - Configure MSI-X hardware
868 * igb_configure_msix sets up the hardware to properly
869 * generate MSI-X interrupts.
871 static void igb_configure_msix(struct igb_adapter
*adapter
)
875 struct e1000_hw
*hw
= &adapter
->hw
;
877 adapter
->eims_enable_mask
= 0;
879 /* set vector for other causes, i.e. link changes */
880 switch (hw
->mac
.type
) {
882 tmp
= rd32(E1000_CTRL_EXT
);
883 /* enable MSI-X PBA support*/
884 tmp
|= E1000_CTRL_EXT_PBA_CLR
;
886 /* Auto-Mask interrupts upon ICR read. */
887 tmp
|= E1000_CTRL_EXT_EIAME
;
888 tmp
|= E1000_CTRL_EXT_IRCA
;
890 wr32(E1000_CTRL_EXT
, tmp
);
892 /* enable msix_other interrupt */
893 array_wr32(E1000_MSIXBM(0), vector
++,
895 adapter
->eims_other
= E1000_EIMS_OTHER
;
904 /* Turn on MSI-X capability first, or our settings
905 * won't stick. And it will take days to debug. */
906 wr32(E1000_GPIE
, E1000_GPIE_MSIX_MODE
|
907 E1000_GPIE_PBA
| E1000_GPIE_EIAME
|
910 /* enable msix_other interrupt */
911 adapter
->eims_other
= 1 << vector
;
912 tmp
= (vector
++ | E1000_IVAR_VALID
) << 8;
914 wr32(E1000_IVAR_MISC
, tmp
);
917 /* do nothing, since nothing else supports MSI-X */
919 } /* switch (hw->mac.type) */
921 adapter
->eims_enable_mask
|= adapter
->eims_other
;
923 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
924 igb_assign_vector(adapter
->q_vector
[i
], vector
++);
930 * igb_request_msix - Initialize MSI-X interrupts
932 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
935 static int igb_request_msix(struct igb_adapter
*adapter
)
937 struct net_device
*netdev
= adapter
->netdev
;
938 struct e1000_hw
*hw
= &adapter
->hw
;
939 int i
, err
= 0, vector
= 0;
941 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
942 igb_msix_other
, 0, netdev
->name
, adapter
);
947 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
948 struct igb_q_vector
*q_vector
= adapter
->q_vector
[i
];
950 q_vector
->itr_register
= hw
->hw_addr
+ E1000_EITR(vector
);
952 if (q_vector
->rx
.ring
&& q_vector
->tx
.ring
)
953 sprintf(q_vector
->name
, "%s-TxRx-%u", netdev
->name
,
954 q_vector
->rx
.ring
->queue_index
);
955 else if (q_vector
->tx
.ring
)
956 sprintf(q_vector
->name
, "%s-tx-%u", netdev
->name
,
957 q_vector
->tx
.ring
->queue_index
);
958 else if (q_vector
->rx
.ring
)
959 sprintf(q_vector
->name
, "%s-rx-%u", netdev
->name
,
960 q_vector
->rx
.ring
->queue_index
);
962 sprintf(q_vector
->name
, "%s-unused", netdev
->name
);
964 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
965 igb_msix_ring
, 0, q_vector
->name
,
972 igb_configure_msix(adapter
);
978 static void igb_reset_interrupt_capability(struct igb_adapter
*adapter
)
980 if (adapter
->msix_entries
) {
981 pci_disable_msix(adapter
->pdev
);
982 kfree(adapter
->msix_entries
);
983 adapter
->msix_entries
= NULL
;
984 } else if (adapter
->flags
& IGB_FLAG_HAS_MSI
) {
985 pci_disable_msi(adapter
->pdev
);
990 * igb_free_q_vectors - Free memory allocated for interrupt vectors
991 * @adapter: board private structure to initialize
993 * This function frees the memory allocated to the q_vectors. In addition if
994 * NAPI is enabled it will delete any references to the NAPI struct prior
995 * to freeing the q_vector.
997 static void igb_free_q_vectors(struct igb_adapter
*adapter
)
1001 for (v_idx
= 0; v_idx
< adapter
->num_q_vectors
; v_idx
++) {
1002 struct igb_q_vector
*q_vector
= adapter
->q_vector
[v_idx
];
1003 adapter
->q_vector
[v_idx
] = NULL
;
1006 netif_napi_del(&q_vector
->napi
);
1009 adapter
->num_q_vectors
= 0;
1013 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1015 * This function resets the device so that it has 0 rx queues, tx queues, and
1016 * MSI-X interrupts allocated.
1018 static void igb_clear_interrupt_scheme(struct igb_adapter
*adapter
)
1020 igb_free_queues(adapter
);
1021 igb_free_q_vectors(adapter
);
1022 igb_reset_interrupt_capability(adapter
);
1026 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1028 * Attempt to configure interrupts using the best available
1029 * capabilities of the hardware and kernel.
1031 static int igb_set_interrupt_capability(struct igb_adapter
*adapter
)
1036 /* Number of supported queues. */
1037 adapter
->num_rx_queues
= adapter
->rss_queues
;
1038 if (adapter
->vfs_allocated_count
)
1039 adapter
->num_tx_queues
= 1;
1041 adapter
->num_tx_queues
= adapter
->rss_queues
;
1043 /* start with one vector for every rx queue */
1044 numvecs
= adapter
->num_rx_queues
;
1046 /* if tx handler is separate add 1 for every tx queue */
1047 if (!(adapter
->flags
& IGB_FLAG_QUEUE_PAIRS
))
1048 numvecs
+= adapter
->num_tx_queues
;
1050 /* store the number of vectors reserved for queues */
1051 adapter
->num_q_vectors
= numvecs
;
1053 /* add 1 vector for link status interrupts */
1055 adapter
->msix_entries
= kcalloc(numvecs
, sizeof(struct msix_entry
),
1058 if (!adapter
->msix_entries
)
1061 for (i
= 0; i
< numvecs
; i
++)
1062 adapter
->msix_entries
[i
].entry
= i
;
1064 err
= pci_enable_msix(adapter
->pdev
,
1065 adapter
->msix_entries
,
1070 igb_reset_interrupt_capability(adapter
);
1072 /* If we can't do MSI-X, try MSI */
1074 #ifdef CONFIG_PCI_IOV
1075 /* disable SR-IOV for non MSI-X configurations */
1076 if (adapter
->vf_data
) {
1077 struct e1000_hw
*hw
= &adapter
->hw
;
1078 /* disable iov and allow time for transactions to clear */
1079 pci_disable_sriov(adapter
->pdev
);
1082 kfree(adapter
->vf_data
);
1083 adapter
->vf_data
= NULL
;
1084 wr32(E1000_IOVCTL
, E1000_IOVCTL_REUSE_VFQ
);
1087 dev_info(&adapter
->pdev
->dev
, "IOV Disabled\n");
1090 adapter
->vfs_allocated_count
= 0;
1091 adapter
->rss_queues
= 1;
1092 adapter
->flags
|= IGB_FLAG_QUEUE_PAIRS
;
1093 adapter
->num_rx_queues
= 1;
1094 adapter
->num_tx_queues
= 1;
1095 adapter
->num_q_vectors
= 1;
1096 if (!pci_enable_msi(adapter
->pdev
))
1097 adapter
->flags
|= IGB_FLAG_HAS_MSI
;
1099 /* Notify the stack of the (possibly) reduced queue counts. */
1101 netif_set_real_num_tx_queues(adapter
->netdev
, adapter
->num_tx_queues
);
1102 err
= netif_set_real_num_rx_queues(adapter
->netdev
,
1103 adapter
->num_rx_queues
);
1109 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1110 * @adapter: board private structure to initialize
1112 * We allocate one q_vector per queue interrupt. If allocation fails we
1115 static int igb_alloc_q_vectors(struct igb_adapter
*adapter
)
1117 struct igb_q_vector
*q_vector
;
1118 struct e1000_hw
*hw
= &adapter
->hw
;
1120 int orig_node
= adapter
->node
;
1122 for (v_idx
= 0; v_idx
< adapter
->num_q_vectors
; v_idx
++) {
1123 if ((adapter
->num_q_vectors
== (adapter
->num_rx_queues
+
1124 adapter
->num_tx_queues
)) &&
1125 (adapter
->num_rx_queues
== v_idx
))
1126 adapter
->node
= orig_node
;
1127 if (orig_node
== -1) {
1128 int cur_node
= next_online_node(adapter
->node
);
1129 if (cur_node
== MAX_NUMNODES
)
1130 cur_node
= first_online_node
;
1131 adapter
->node
= cur_node
;
1133 q_vector
= kzalloc_node(sizeof(struct igb_q_vector
), GFP_KERNEL
,
1136 q_vector
= kzalloc(sizeof(struct igb_q_vector
),
1140 q_vector
->adapter
= adapter
;
1141 q_vector
->itr_register
= hw
->hw_addr
+ E1000_EITR(0);
1142 q_vector
->itr_val
= IGB_START_ITR
;
1143 netif_napi_add(adapter
->netdev
, &q_vector
->napi
, igb_poll
, 64);
1144 adapter
->q_vector
[v_idx
] = q_vector
;
1146 /* Restore the adapter's original node */
1147 adapter
->node
= orig_node
;
1152 /* Restore the adapter's original node */
1153 adapter
->node
= orig_node
;
1154 igb_free_q_vectors(adapter
);
1158 static void igb_map_rx_ring_to_vector(struct igb_adapter
*adapter
,
1159 int ring_idx
, int v_idx
)
1161 struct igb_q_vector
*q_vector
= adapter
->q_vector
[v_idx
];
1163 q_vector
->rx
.ring
= adapter
->rx_ring
[ring_idx
];
1164 q_vector
->rx
.ring
->q_vector
= q_vector
;
1165 q_vector
->rx
.count
++;
1166 q_vector
->itr_val
= adapter
->rx_itr_setting
;
1167 if (q_vector
->itr_val
&& q_vector
->itr_val
<= 3)
1168 q_vector
->itr_val
= IGB_START_ITR
;
1171 static void igb_map_tx_ring_to_vector(struct igb_adapter
*adapter
,
1172 int ring_idx
, int v_idx
)
1174 struct igb_q_vector
*q_vector
= adapter
->q_vector
[v_idx
];
1176 q_vector
->tx
.ring
= adapter
->tx_ring
[ring_idx
];
1177 q_vector
->tx
.ring
->q_vector
= q_vector
;
1178 q_vector
->tx
.count
++;
1179 q_vector
->itr_val
= adapter
->tx_itr_setting
;
1180 q_vector
->tx
.work_limit
= adapter
->tx_work_limit
;
1181 if (q_vector
->itr_val
&& q_vector
->itr_val
<= 3)
1182 q_vector
->itr_val
= IGB_START_ITR
;
1186 * igb_map_ring_to_vector - maps allocated queues to vectors
1188 * This function maps the recently allocated queues to vectors.
1190 static int igb_map_ring_to_vector(struct igb_adapter
*adapter
)
1195 if ((adapter
->num_q_vectors
< adapter
->num_rx_queues
) ||
1196 (adapter
->num_q_vectors
< adapter
->num_tx_queues
))
1199 if (adapter
->num_q_vectors
>=
1200 (adapter
->num_rx_queues
+ adapter
->num_tx_queues
)) {
1201 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
1202 igb_map_rx_ring_to_vector(adapter
, i
, v_idx
++);
1203 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
1204 igb_map_tx_ring_to_vector(adapter
, i
, v_idx
++);
1206 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
1207 if (i
< adapter
->num_tx_queues
)
1208 igb_map_tx_ring_to_vector(adapter
, i
, v_idx
);
1209 igb_map_rx_ring_to_vector(adapter
, i
, v_idx
++);
1211 for (; i
< adapter
->num_tx_queues
; i
++)
1212 igb_map_tx_ring_to_vector(adapter
, i
, v_idx
++);
1218 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1220 * This function initializes the interrupts and allocates all of the queues.
1222 static int igb_init_interrupt_scheme(struct igb_adapter
*adapter
)
1224 struct pci_dev
*pdev
= adapter
->pdev
;
1227 err
= igb_set_interrupt_capability(adapter
);
1231 err
= igb_alloc_q_vectors(adapter
);
1233 dev_err(&pdev
->dev
, "Unable to allocate memory for vectors\n");
1234 goto err_alloc_q_vectors
;
1237 err
= igb_alloc_queues(adapter
);
1239 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
1240 goto err_alloc_queues
;
1243 err
= igb_map_ring_to_vector(adapter
);
1245 dev_err(&pdev
->dev
, "Invalid q_vector to ring mapping\n");
1246 goto err_map_queues
;
1252 igb_free_queues(adapter
);
1254 igb_free_q_vectors(adapter
);
1255 err_alloc_q_vectors
:
1256 igb_reset_interrupt_capability(adapter
);
1261 * igb_request_irq - initialize interrupts
1263 * Attempts to configure interrupts using the best available
1264 * capabilities of the hardware and kernel.
1266 static int igb_request_irq(struct igb_adapter
*adapter
)
1268 struct net_device
*netdev
= adapter
->netdev
;
1269 struct pci_dev
*pdev
= adapter
->pdev
;
1272 if (adapter
->msix_entries
) {
1273 err
= igb_request_msix(adapter
);
1276 /* fall back to MSI */
1277 igb_clear_interrupt_scheme(adapter
);
1278 if (!pci_enable_msi(pdev
))
1279 adapter
->flags
|= IGB_FLAG_HAS_MSI
;
1280 igb_free_all_tx_resources(adapter
);
1281 igb_free_all_rx_resources(adapter
);
1282 adapter
->num_tx_queues
= 1;
1283 adapter
->num_rx_queues
= 1;
1284 adapter
->num_q_vectors
= 1;
1285 err
= igb_alloc_q_vectors(adapter
);
1288 "Unable to allocate memory for vectors\n");
1291 err
= igb_alloc_queues(adapter
);
1294 "Unable to allocate memory for queues\n");
1295 igb_free_q_vectors(adapter
);
1298 igb_setup_all_tx_resources(adapter
);
1299 igb_setup_all_rx_resources(adapter
);
1302 igb_assign_vector(adapter
->q_vector
[0], 0);
1304 if (adapter
->flags
& IGB_FLAG_HAS_MSI
) {
1305 err
= request_irq(pdev
->irq
, igb_intr_msi
, 0,
1306 netdev
->name
, adapter
);
1310 /* fall back to legacy interrupts */
1311 igb_reset_interrupt_capability(adapter
);
1312 adapter
->flags
&= ~IGB_FLAG_HAS_MSI
;
1315 err
= request_irq(pdev
->irq
, igb_intr
, IRQF_SHARED
,
1316 netdev
->name
, adapter
);
1319 dev_err(&pdev
->dev
, "Error %d getting interrupt\n",
1326 static void igb_free_irq(struct igb_adapter
*adapter
)
1328 if (adapter
->msix_entries
) {
1331 free_irq(adapter
->msix_entries
[vector
++].vector
, adapter
);
1333 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1334 free_irq(adapter
->msix_entries
[vector
++].vector
,
1335 adapter
->q_vector
[i
]);
1337 free_irq(adapter
->pdev
->irq
, adapter
);
1342 * igb_irq_disable - Mask off interrupt generation on the NIC
1343 * @adapter: board private structure
1345 static void igb_irq_disable(struct igb_adapter
*adapter
)
1347 struct e1000_hw
*hw
= &adapter
->hw
;
1350 * we need to be careful when disabling interrupts. The VFs are also
1351 * mapped into these registers and so clearing the bits can cause
1352 * issues on the VF drivers so we only need to clear what we set
1354 if (adapter
->msix_entries
) {
1355 u32 regval
= rd32(E1000_EIAM
);
1356 wr32(E1000_EIAM
, regval
& ~adapter
->eims_enable_mask
);
1357 wr32(E1000_EIMC
, adapter
->eims_enable_mask
);
1358 regval
= rd32(E1000_EIAC
);
1359 wr32(E1000_EIAC
, regval
& ~adapter
->eims_enable_mask
);
1363 wr32(E1000_IMC
, ~0);
1365 if (adapter
->msix_entries
) {
1367 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1368 synchronize_irq(adapter
->msix_entries
[i
].vector
);
1370 synchronize_irq(adapter
->pdev
->irq
);
1375 * igb_irq_enable - Enable default interrupt generation settings
1376 * @adapter: board private structure
1378 static void igb_irq_enable(struct igb_adapter
*adapter
)
1380 struct e1000_hw
*hw
= &adapter
->hw
;
1382 if (adapter
->msix_entries
) {
1383 u32 ims
= E1000_IMS_LSC
| E1000_IMS_DOUTSYNC
| E1000_IMS_DRSTA
;
1384 u32 regval
= rd32(E1000_EIAC
);
1385 wr32(E1000_EIAC
, regval
| adapter
->eims_enable_mask
);
1386 regval
= rd32(E1000_EIAM
);
1387 wr32(E1000_EIAM
, regval
| adapter
->eims_enable_mask
);
1388 wr32(E1000_EIMS
, adapter
->eims_enable_mask
);
1389 if (adapter
->vfs_allocated_count
) {
1390 wr32(E1000_MBVFIMR
, 0xFF);
1391 ims
|= E1000_IMS_VMMB
;
1393 wr32(E1000_IMS
, ims
);
1395 wr32(E1000_IMS
, IMS_ENABLE_MASK
|
1397 wr32(E1000_IAM
, IMS_ENABLE_MASK
|
1402 static void igb_update_mng_vlan(struct igb_adapter
*adapter
)
1404 struct e1000_hw
*hw
= &adapter
->hw
;
1405 u16 vid
= adapter
->hw
.mng_cookie
.vlan_id
;
1406 u16 old_vid
= adapter
->mng_vlan_id
;
1408 if (hw
->mng_cookie
.status
& E1000_MNG_DHCP_COOKIE_STATUS_VLAN
) {
1409 /* add VID to filter table */
1410 igb_vfta_set(hw
, vid
, true);
1411 adapter
->mng_vlan_id
= vid
;
1413 adapter
->mng_vlan_id
= IGB_MNG_VLAN_NONE
;
1416 if ((old_vid
!= (u16
)IGB_MNG_VLAN_NONE
) &&
1418 !test_bit(old_vid
, adapter
->active_vlans
)) {
1419 /* remove VID from filter table */
1420 igb_vfta_set(hw
, old_vid
, false);
1425 * igb_release_hw_control - release control of the h/w to f/w
1426 * @adapter: address of board private structure
1428 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1429 * For ASF and Pass Through versions of f/w this means that the
1430 * driver is no longer loaded.
1433 static void igb_release_hw_control(struct igb_adapter
*adapter
)
1435 struct e1000_hw
*hw
= &adapter
->hw
;
1438 /* Let firmware take over control of h/w */
1439 ctrl_ext
= rd32(E1000_CTRL_EXT
);
1440 wr32(E1000_CTRL_EXT
,
1441 ctrl_ext
& ~E1000_CTRL_EXT_DRV_LOAD
);
1445 * igb_get_hw_control - get control of the h/w from f/w
1446 * @adapter: address of board private structure
1448 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1449 * For ASF and Pass Through versions of f/w this means that
1450 * the driver is loaded.
1453 static void igb_get_hw_control(struct igb_adapter
*adapter
)
1455 struct e1000_hw
*hw
= &adapter
->hw
;
1458 /* Let firmware know the driver has taken over */
1459 ctrl_ext
= rd32(E1000_CTRL_EXT
);
1460 wr32(E1000_CTRL_EXT
,
1461 ctrl_ext
| E1000_CTRL_EXT_DRV_LOAD
);
1465 * igb_configure - configure the hardware for RX and TX
1466 * @adapter: private board structure
1468 static void igb_configure(struct igb_adapter
*adapter
)
1470 struct net_device
*netdev
= adapter
->netdev
;
1473 igb_get_hw_control(adapter
);
1474 igb_set_rx_mode(netdev
);
1476 igb_restore_vlan(adapter
);
1478 igb_setup_tctl(adapter
);
1479 igb_setup_mrqc(adapter
);
1480 igb_setup_rctl(adapter
);
1482 igb_configure_tx(adapter
);
1483 igb_configure_rx(adapter
);
1485 igb_rx_fifo_flush_82575(&adapter
->hw
);
1487 /* call igb_desc_unused which always leaves
1488 * at least 1 descriptor unused to make sure
1489 * next_to_use != next_to_clean */
1490 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
1491 struct igb_ring
*ring
= adapter
->rx_ring
[i
];
1492 igb_alloc_rx_buffers(ring
, igb_desc_unused(ring
));
1497 * igb_power_up_link - Power up the phy/serdes link
1498 * @adapter: address of board private structure
1500 void igb_power_up_link(struct igb_adapter
*adapter
)
1502 igb_reset_phy(&adapter
->hw
);
1504 if (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
)
1505 igb_power_up_phy_copper(&adapter
->hw
);
1507 igb_power_up_serdes_link_82575(&adapter
->hw
);
1511 * igb_power_down_link - Power down the phy/serdes link
1512 * @adapter: address of board private structure
1514 static void igb_power_down_link(struct igb_adapter
*adapter
)
1516 if (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
)
1517 igb_power_down_phy_copper_82575(&adapter
->hw
);
1519 igb_shutdown_serdes_link_82575(&adapter
->hw
);
1523 * igb_up - Open the interface and prepare it to handle traffic
1524 * @adapter: board private structure
1526 int igb_up(struct igb_adapter
*adapter
)
1528 struct e1000_hw
*hw
= &adapter
->hw
;
1531 /* hardware has been reset, we need to reload some things */
1532 igb_configure(adapter
);
1534 clear_bit(__IGB_DOWN
, &adapter
->state
);
1536 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1537 napi_enable(&(adapter
->q_vector
[i
]->napi
));
1539 if (adapter
->msix_entries
)
1540 igb_configure_msix(adapter
);
1542 igb_assign_vector(adapter
->q_vector
[0], 0);
1544 /* Clear any pending interrupts. */
1546 igb_irq_enable(adapter
);
1548 /* notify VFs that reset has been completed */
1549 if (adapter
->vfs_allocated_count
) {
1550 u32 reg_data
= rd32(E1000_CTRL_EXT
);
1551 reg_data
|= E1000_CTRL_EXT_PFRSTD
;
1552 wr32(E1000_CTRL_EXT
, reg_data
);
1555 netif_tx_start_all_queues(adapter
->netdev
);
1557 /* start the watchdog. */
1558 hw
->mac
.get_link_status
= 1;
1559 schedule_work(&adapter
->watchdog_task
);
1564 void igb_down(struct igb_adapter
*adapter
)
1566 struct net_device
*netdev
= adapter
->netdev
;
1567 struct e1000_hw
*hw
= &adapter
->hw
;
1571 /* signal that we're down so the interrupt handler does not
1572 * reschedule our watchdog timer */
1573 set_bit(__IGB_DOWN
, &adapter
->state
);
1575 /* disable receives in the hardware */
1576 rctl
= rd32(E1000_RCTL
);
1577 wr32(E1000_RCTL
, rctl
& ~E1000_RCTL_EN
);
1578 /* flush and sleep below */
1580 netif_tx_stop_all_queues(netdev
);
1582 /* disable transmits in the hardware */
1583 tctl
= rd32(E1000_TCTL
);
1584 tctl
&= ~E1000_TCTL_EN
;
1585 wr32(E1000_TCTL
, tctl
);
1586 /* flush both disables and wait for them to finish */
1590 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1591 napi_disable(&(adapter
->q_vector
[i
]->napi
));
1593 igb_irq_disable(adapter
);
1595 del_timer_sync(&adapter
->watchdog_timer
);
1596 del_timer_sync(&adapter
->phy_info_timer
);
1598 netif_carrier_off(netdev
);
1600 /* record the stats before reset*/
1601 spin_lock(&adapter
->stats64_lock
);
1602 igb_update_stats(adapter
, &adapter
->stats64
);
1603 spin_unlock(&adapter
->stats64_lock
);
1605 adapter
->link_speed
= 0;
1606 adapter
->link_duplex
= 0;
1608 if (!pci_channel_offline(adapter
->pdev
))
1610 igb_clean_all_tx_rings(adapter
);
1611 igb_clean_all_rx_rings(adapter
);
1612 #ifdef CONFIG_IGB_DCA
1614 /* since we reset the hardware DCA settings were cleared */
1615 igb_setup_dca(adapter
);
1619 void igb_reinit_locked(struct igb_adapter
*adapter
)
1621 WARN_ON(in_interrupt());
1622 while (test_and_set_bit(__IGB_RESETTING
, &adapter
->state
))
1626 clear_bit(__IGB_RESETTING
, &adapter
->state
);
1629 void igb_reset(struct igb_adapter
*adapter
)
1631 struct pci_dev
*pdev
= adapter
->pdev
;
1632 struct e1000_hw
*hw
= &adapter
->hw
;
1633 struct e1000_mac_info
*mac
= &hw
->mac
;
1634 struct e1000_fc_info
*fc
= &hw
->fc
;
1635 u32 pba
= 0, tx_space
, min_tx_space
, min_rx_space
;
1638 /* Repartition Pba for greater than 9k mtu
1639 * To take effect CTRL.RST is required.
1641 switch (mac
->type
) {
1644 pba
= rd32(E1000_RXPBS
);
1645 pba
= igb_rxpbs_adjust_82580(pba
);
1648 pba
= rd32(E1000_RXPBS
);
1649 pba
&= E1000_RXPBS_SIZE_MASK_82576
;
1655 pba
= E1000_PBA_34K
;
1659 if ((adapter
->max_frame_size
> ETH_FRAME_LEN
+ ETH_FCS_LEN
) &&
1660 (mac
->type
< e1000_82576
)) {
1661 /* adjust PBA for jumbo frames */
1662 wr32(E1000_PBA
, pba
);
1664 /* To maintain wire speed transmits, the Tx FIFO should be
1665 * large enough to accommodate two full transmit packets,
1666 * rounded up to the next 1KB and expressed in KB. Likewise,
1667 * the Rx FIFO should be large enough to accommodate at least
1668 * one full receive packet and is similarly rounded up and
1669 * expressed in KB. */
1670 pba
= rd32(E1000_PBA
);
1671 /* upper 16 bits has Tx packet buffer allocation size in KB */
1672 tx_space
= pba
>> 16;
1673 /* lower 16 bits has Rx packet buffer allocation size in KB */
1675 /* the tx fifo also stores 16 bytes of information about the tx
1676 * but don't include ethernet FCS because hardware appends it */
1677 min_tx_space
= (adapter
->max_frame_size
+
1678 sizeof(union e1000_adv_tx_desc
) -
1680 min_tx_space
= ALIGN(min_tx_space
, 1024);
1681 min_tx_space
>>= 10;
1682 /* software strips receive CRC, so leave room for it */
1683 min_rx_space
= adapter
->max_frame_size
;
1684 min_rx_space
= ALIGN(min_rx_space
, 1024);
1685 min_rx_space
>>= 10;
1687 /* If current Tx allocation is less than the min Tx FIFO size,
1688 * and the min Tx FIFO size is less than the current Rx FIFO
1689 * allocation, take space away from current Rx allocation */
1690 if (tx_space
< min_tx_space
&&
1691 ((min_tx_space
- tx_space
) < pba
)) {
1692 pba
= pba
- (min_tx_space
- tx_space
);
1694 /* if short on rx space, rx wins and must trump tx
1696 if (pba
< min_rx_space
)
1699 wr32(E1000_PBA
, pba
);
1702 /* flow control settings */
1703 /* The high water mark must be low enough to fit one full frame
1704 * (or the size used for early receive) above it in the Rx FIFO.
1705 * Set it to the lower of:
1706 * - 90% of the Rx FIFO size, or
1707 * - the full Rx FIFO size minus one full frame */
1708 hwm
= min(((pba
<< 10) * 9 / 10),
1709 ((pba
<< 10) - 2 * adapter
->max_frame_size
));
1711 fc
->high_water
= hwm
& 0xFFF0; /* 16-byte granularity */
1712 fc
->low_water
= fc
->high_water
- 16;
1713 fc
->pause_time
= 0xFFFF;
1715 fc
->current_mode
= fc
->requested_mode
;
1717 /* disable receive for all VFs and wait one second */
1718 if (adapter
->vfs_allocated_count
) {
1720 for (i
= 0 ; i
< adapter
->vfs_allocated_count
; i
++)
1721 adapter
->vf_data
[i
].flags
&= IGB_VF_FLAG_PF_SET_MAC
;
1723 /* ping all the active vfs to let them know we are going down */
1724 igb_ping_all_vfs(adapter
);
1726 /* disable transmits and receives */
1727 wr32(E1000_VFRE
, 0);
1728 wr32(E1000_VFTE
, 0);
1731 /* Allow time for pending master requests to run */
1732 hw
->mac
.ops
.reset_hw(hw
);
1735 if (hw
->mac
.ops
.init_hw(hw
))
1736 dev_err(&pdev
->dev
, "Hardware Error\n");
1739 * Flow control settings reset on hardware reset, so guarantee flow
1740 * control is off when forcing speed.
1742 if (!hw
->mac
.autoneg
)
1743 igb_force_mac_fc(hw
);
1745 igb_init_dmac(adapter
, pba
);
1746 if (!netif_running(adapter
->netdev
))
1747 igb_power_down_link(adapter
);
1749 igb_update_mng_vlan(adapter
);
1751 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1752 wr32(E1000_VET
, ETHERNET_IEEE_VLAN_TYPE
);
1754 #ifdef CONFIG_IGB_PTP
1755 /* Re-enable PTP, where applicable. */
1756 igb_ptp_reset(adapter
);
1757 #endif /* CONFIG_IGB_PTP */
1759 igb_get_phy_info(hw
);
1762 static netdev_features_t
igb_fix_features(struct net_device
*netdev
,
1763 netdev_features_t features
)
1766 * Since there is no support for separate rx/tx vlan accel
1767 * enable/disable make sure tx flag is always in same state as rx.
1769 if (features
& NETIF_F_HW_VLAN_RX
)
1770 features
|= NETIF_F_HW_VLAN_TX
;
1772 features
&= ~NETIF_F_HW_VLAN_TX
;
1777 static int igb_set_features(struct net_device
*netdev
,
1778 netdev_features_t features
)
1780 netdev_features_t changed
= netdev
->features
^ features
;
1781 struct igb_adapter
*adapter
= netdev_priv(netdev
);
1783 if (changed
& NETIF_F_HW_VLAN_RX
)
1784 igb_vlan_mode(netdev
, features
);
1786 if (!(changed
& NETIF_F_RXALL
))
1789 netdev
->features
= features
;
1791 if (netif_running(netdev
))
1792 igb_reinit_locked(adapter
);
1799 static const struct net_device_ops igb_netdev_ops
= {
1800 .ndo_open
= igb_open
,
1801 .ndo_stop
= igb_close
,
1802 .ndo_start_xmit
= igb_xmit_frame
,
1803 .ndo_get_stats64
= igb_get_stats64
,
1804 .ndo_set_rx_mode
= igb_set_rx_mode
,
1805 .ndo_set_mac_address
= igb_set_mac
,
1806 .ndo_change_mtu
= igb_change_mtu
,
1807 .ndo_do_ioctl
= igb_ioctl
,
1808 .ndo_tx_timeout
= igb_tx_timeout
,
1809 .ndo_validate_addr
= eth_validate_addr
,
1810 .ndo_vlan_rx_add_vid
= igb_vlan_rx_add_vid
,
1811 .ndo_vlan_rx_kill_vid
= igb_vlan_rx_kill_vid
,
1812 .ndo_set_vf_mac
= igb_ndo_set_vf_mac
,
1813 .ndo_set_vf_vlan
= igb_ndo_set_vf_vlan
,
1814 .ndo_set_vf_tx_rate
= igb_ndo_set_vf_bw
,
1815 .ndo_get_vf_config
= igb_ndo_get_vf_config
,
1816 #ifdef CONFIG_NET_POLL_CONTROLLER
1817 .ndo_poll_controller
= igb_netpoll
,
1819 .ndo_fix_features
= igb_fix_features
,
1820 .ndo_set_features
= igb_set_features
,
1824 * igb_set_fw_version - Configure version string for ethtool
1825 * @adapter: adapter struct
1828 void igb_set_fw_version(struct igb_adapter
*adapter
)
1830 struct e1000_hw
*hw
= &adapter
->hw
;
1831 u16 eeprom_verh
, eeprom_verl
, comb_verh
, comb_verl
, comb_offset
;
1832 u16 major
, build
, patch
, fw_version
;
1835 hw
->nvm
.ops
.read(hw
, 5, 1, &fw_version
);
1836 if (adapter
->hw
.mac
.type
!= e1000_i211
) {
1837 hw
->nvm
.ops
.read(hw
, NVM_ETRACK_WORD
, 1, &eeprom_verh
);
1838 hw
->nvm
.ops
.read(hw
, (NVM_ETRACK_WORD
+ 1), 1, &eeprom_verl
);
1839 etrack_id
= (eeprom_verh
<< IGB_ETRACK_SHIFT
) | eeprom_verl
;
1841 /* combo image version needs to be found */
1842 hw
->nvm
.ops
.read(hw
, NVM_COMB_VER_PTR
, 1, &comb_offset
);
1843 if ((comb_offset
!= 0x0) &&
1844 (comb_offset
!= IGB_NVM_VER_INVALID
)) {
1845 hw
->nvm
.ops
.read(hw
, (NVM_COMB_VER_OFF
+ comb_offset
1846 + 1), 1, &comb_verh
);
1847 hw
->nvm
.ops
.read(hw
, (NVM_COMB_VER_OFF
+ comb_offset
),
1850 /* Only display Option Rom if it exists and is valid */
1851 if ((comb_verh
&& comb_verl
) &&
1852 ((comb_verh
!= IGB_NVM_VER_INVALID
) &&
1853 (comb_verl
!= IGB_NVM_VER_INVALID
))) {
1854 major
= comb_verl
>> IGB_COMB_VER_SHFT
;
1855 build
= (comb_verl
<< IGB_COMB_VER_SHFT
) |
1856 (comb_verh
>> IGB_COMB_VER_SHFT
);
1857 patch
= comb_verh
& IGB_COMB_VER_MASK
;
1858 snprintf(adapter
->fw_version
,
1859 sizeof(adapter
->fw_version
),
1860 "%d.%d%d, 0x%08x, %d.%d.%d",
1861 (fw_version
& IGB_MAJOR_MASK
) >>
1863 (fw_version
& IGB_MINOR_MASK
) >>
1865 (fw_version
& IGB_BUILD_MASK
),
1866 etrack_id
, major
, build
, patch
);
1870 snprintf(adapter
->fw_version
, sizeof(adapter
->fw_version
),
1872 (fw_version
& IGB_MAJOR_MASK
) >> IGB_MAJOR_SHIFT
,
1873 (fw_version
& IGB_MINOR_MASK
) >> IGB_MINOR_SHIFT
,
1874 (fw_version
& IGB_BUILD_MASK
), etrack_id
);
1876 snprintf(adapter
->fw_version
, sizeof(adapter
->fw_version
),
1878 (fw_version
& IGB_MAJOR_MASK
) >> IGB_MAJOR_SHIFT
,
1879 (fw_version
& IGB_MINOR_MASK
) >> IGB_MINOR_SHIFT
,
1880 (fw_version
& IGB_BUILD_MASK
));
1887 * igb_probe - Device Initialization Routine
1888 * @pdev: PCI device information struct
1889 * @ent: entry in igb_pci_tbl
1891 * Returns 0 on success, negative on failure
1893 * igb_probe initializes an adapter identified by a pci_dev structure.
1894 * The OS initialization, configuring of the adapter private structure,
1895 * and a hardware reset occur.
1897 static int __devinit
igb_probe(struct pci_dev
*pdev
,
1898 const struct pci_device_id
*ent
)
1900 struct net_device
*netdev
;
1901 struct igb_adapter
*adapter
;
1902 struct e1000_hw
*hw
;
1903 u16 eeprom_data
= 0;
1905 static int global_quad_port_a
; /* global quad port a indication */
1906 const struct e1000_info
*ei
= igb_info_tbl
[ent
->driver_data
];
1907 unsigned long mmio_start
, mmio_len
;
1908 int err
, pci_using_dac
;
1909 u16 eeprom_apme_mask
= IGB_EEPROM_APME
;
1910 u8 part_str
[E1000_PBANUM_LENGTH
];
1912 /* Catch broken hardware that put the wrong VF device ID in
1913 * the PCIe SR-IOV capability.
1915 if (pdev
->is_virtfn
) {
1916 WARN(1, KERN_ERR
"%s (%hx:%hx) should not be a VF!\n",
1917 pci_name(pdev
), pdev
->vendor
, pdev
->device
);
1921 err
= pci_enable_device_mem(pdev
);
1926 err
= dma_set_mask(&pdev
->dev
, DMA_BIT_MASK(64));
1928 err
= dma_set_coherent_mask(&pdev
->dev
, DMA_BIT_MASK(64));
1932 err
= dma_set_mask(&pdev
->dev
, DMA_BIT_MASK(32));
1934 err
= dma_set_coherent_mask(&pdev
->dev
, DMA_BIT_MASK(32));
1936 dev_err(&pdev
->dev
, "No usable DMA "
1937 "configuration, aborting\n");
1943 err
= pci_request_selected_regions(pdev
, pci_select_bars(pdev
,
1949 pci_enable_pcie_error_reporting(pdev
);
1951 pci_set_master(pdev
);
1952 pci_save_state(pdev
);
1955 netdev
= alloc_etherdev_mq(sizeof(struct igb_adapter
),
1958 goto err_alloc_etherdev
;
1960 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
1962 pci_set_drvdata(pdev
, netdev
);
1963 adapter
= netdev_priv(netdev
);
1964 adapter
->netdev
= netdev
;
1965 adapter
->pdev
= pdev
;
1968 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
1970 mmio_start
= pci_resource_start(pdev
, 0);
1971 mmio_len
= pci_resource_len(pdev
, 0);
1974 hw
->hw_addr
= ioremap(mmio_start
, mmio_len
);
1978 netdev
->netdev_ops
= &igb_netdev_ops
;
1979 igb_set_ethtool_ops(netdev
);
1980 netdev
->watchdog_timeo
= 5 * HZ
;
1982 strncpy(netdev
->name
, pci_name(pdev
), sizeof(netdev
->name
) - 1);
1984 netdev
->mem_start
= mmio_start
;
1985 netdev
->mem_end
= mmio_start
+ mmio_len
;
1987 /* PCI config space info */
1988 hw
->vendor_id
= pdev
->vendor
;
1989 hw
->device_id
= pdev
->device
;
1990 hw
->revision_id
= pdev
->revision
;
1991 hw
->subsystem_vendor_id
= pdev
->subsystem_vendor
;
1992 hw
->subsystem_device_id
= pdev
->subsystem_device
;
1994 /* Copy the default MAC, PHY and NVM function pointers */
1995 memcpy(&hw
->mac
.ops
, ei
->mac_ops
, sizeof(hw
->mac
.ops
));
1996 memcpy(&hw
->phy
.ops
, ei
->phy_ops
, sizeof(hw
->phy
.ops
));
1997 memcpy(&hw
->nvm
.ops
, ei
->nvm_ops
, sizeof(hw
->nvm
.ops
));
1998 /* Initialize skew-specific constants */
1999 err
= ei
->get_invariants(hw
);
2003 /* setup the private structure */
2004 err
= igb_sw_init(adapter
);
2008 igb_get_bus_info_pcie(hw
);
2010 hw
->phy
.autoneg_wait_to_complete
= false;
2012 /* Copper options */
2013 if (hw
->phy
.media_type
== e1000_media_type_copper
) {
2014 hw
->phy
.mdix
= AUTO_ALL_MODES
;
2015 hw
->phy
.disable_polarity_correction
= false;
2016 hw
->phy
.ms_type
= e1000_ms_hw_default
;
2019 if (igb_check_reset_block(hw
))
2020 dev_info(&pdev
->dev
,
2021 "PHY reset is blocked due to SOL/IDER session.\n");
2024 * features is initialized to 0 in allocation, it might have bits
2025 * set by igb_sw_init so we should use an or instead of an
2028 netdev
->features
|= NETIF_F_SG
|
2035 NETIF_F_HW_VLAN_RX
|
2038 /* copy netdev features into list of user selectable features */
2039 netdev
->hw_features
|= netdev
->features
;
2040 netdev
->hw_features
|= NETIF_F_RXALL
;
2042 /* set this bit last since it cannot be part of hw_features */
2043 netdev
->features
|= NETIF_F_HW_VLAN_FILTER
;
2045 netdev
->vlan_features
|= NETIF_F_TSO
|
2051 netdev
->priv_flags
|= IFF_SUPP_NOFCS
;
2053 if (pci_using_dac
) {
2054 netdev
->features
|= NETIF_F_HIGHDMA
;
2055 netdev
->vlan_features
|= NETIF_F_HIGHDMA
;
2058 if (hw
->mac
.type
>= e1000_82576
) {
2059 netdev
->hw_features
|= NETIF_F_SCTP_CSUM
;
2060 netdev
->features
|= NETIF_F_SCTP_CSUM
;
2063 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
2065 adapter
->en_mng_pt
= igb_enable_mng_pass_thru(hw
);
2067 /* before reading the NVM, reset the controller to put the device in a
2068 * known good starting state */
2069 hw
->mac
.ops
.reset_hw(hw
);
2072 * make sure the NVM is good , i211 parts have special NVM that
2073 * doesn't contain a checksum
2075 if (hw
->mac
.type
!= e1000_i211
) {
2076 if (hw
->nvm
.ops
.validate(hw
) < 0) {
2077 dev_err(&pdev
->dev
, "The NVM Checksum Is Not Valid\n");
2083 /* copy the MAC address out of the NVM */
2084 if (hw
->mac
.ops
.read_mac_addr(hw
))
2085 dev_err(&pdev
->dev
, "NVM Read Error\n");
2087 memcpy(netdev
->dev_addr
, hw
->mac
.addr
, netdev
->addr_len
);
2088 memcpy(netdev
->perm_addr
, hw
->mac
.addr
, netdev
->addr_len
);
2090 if (!is_valid_ether_addr(netdev
->perm_addr
)) {
2091 dev_err(&pdev
->dev
, "Invalid MAC Address\n");
2096 /* get firmware version for ethtool -i */
2097 igb_set_fw_version(adapter
);
2099 setup_timer(&adapter
->watchdog_timer
, igb_watchdog
,
2100 (unsigned long) adapter
);
2101 setup_timer(&adapter
->phy_info_timer
, igb_update_phy_info
,
2102 (unsigned long) adapter
);
2104 INIT_WORK(&adapter
->reset_task
, igb_reset_task
);
2105 INIT_WORK(&adapter
->watchdog_task
, igb_watchdog_task
);
2107 /* Initialize link properties that are user-changeable */
2108 adapter
->fc_autoneg
= true;
2109 hw
->mac
.autoneg
= true;
2110 hw
->phy
.autoneg_advertised
= 0x2f;
2112 hw
->fc
.requested_mode
= e1000_fc_default
;
2113 hw
->fc
.current_mode
= e1000_fc_default
;
2115 igb_validate_mdi_setting(hw
);
2117 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
2118 * enable the ACPI Magic Packet filter
2121 if (hw
->bus
.func
== 0)
2122 hw
->nvm
.ops
.read(hw
, NVM_INIT_CONTROL3_PORT_A
, 1, &eeprom_data
);
2123 else if (hw
->mac
.type
>= e1000_82580
)
2124 hw
->nvm
.ops
.read(hw
, NVM_INIT_CONTROL3_PORT_A
+
2125 NVM_82580_LAN_FUNC_OFFSET(hw
->bus
.func
), 1,
2127 else if (hw
->bus
.func
== 1)
2128 hw
->nvm
.ops
.read(hw
, NVM_INIT_CONTROL3_PORT_B
, 1, &eeprom_data
);
2130 if (eeprom_data
& eeprom_apme_mask
)
2131 adapter
->eeprom_wol
|= E1000_WUFC_MAG
;
2133 /* now that we have the eeprom settings, apply the special cases where
2134 * the eeprom may be wrong or the board simply won't support wake on
2135 * lan on a particular port */
2136 switch (pdev
->device
) {
2137 case E1000_DEV_ID_82575GB_QUAD_COPPER
:
2138 adapter
->eeprom_wol
= 0;
2140 case E1000_DEV_ID_82575EB_FIBER_SERDES
:
2141 case E1000_DEV_ID_82576_FIBER
:
2142 case E1000_DEV_ID_82576_SERDES
:
2143 /* Wake events only supported on port A for dual fiber
2144 * regardless of eeprom setting */
2145 if (rd32(E1000_STATUS
) & E1000_STATUS_FUNC_1
)
2146 adapter
->eeprom_wol
= 0;
2148 case E1000_DEV_ID_82576_QUAD_COPPER
:
2149 case E1000_DEV_ID_82576_QUAD_COPPER_ET2
:
2150 /* if quad port adapter, disable WoL on all but port A */
2151 if (global_quad_port_a
!= 0)
2152 adapter
->eeprom_wol
= 0;
2154 adapter
->flags
|= IGB_FLAG_QUAD_PORT_A
;
2155 /* Reset for multiple quad port adapters */
2156 if (++global_quad_port_a
== 4)
2157 global_quad_port_a
= 0;
2161 /* initialize the wol settings based on the eeprom settings */
2162 adapter
->wol
= adapter
->eeprom_wol
;
2163 device_set_wakeup_enable(&adapter
->pdev
->dev
, adapter
->wol
);
2165 /* reset the hardware with the new settings */
2168 /* let the f/w know that the h/w is now under the control of the
2170 igb_get_hw_control(adapter
);
2172 strcpy(netdev
->name
, "eth%d");
2173 err
= register_netdev(netdev
);
2177 /* carrier off reporting is important to ethtool even BEFORE open */
2178 netif_carrier_off(netdev
);
2180 #ifdef CONFIG_IGB_DCA
2181 if (dca_add_requester(&pdev
->dev
) == 0) {
2182 adapter
->flags
|= IGB_FLAG_DCA_ENABLED
;
2183 dev_info(&pdev
->dev
, "DCA enabled\n");
2184 igb_setup_dca(adapter
);
2189 #ifdef CONFIG_IGB_PTP
2190 /* do hw tstamp init after resetting */
2191 igb_ptp_init(adapter
);
2192 #endif /* CONFIG_IGB_PTP */
2194 dev_info(&pdev
->dev
, "Intel(R) Gigabit Ethernet Network Connection\n");
2195 /* print bus type/speed/width info */
2196 dev_info(&pdev
->dev
, "%s: (PCIe:%s:%s) %pM\n",
2198 ((hw
->bus
.speed
== e1000_bus_speed_2500
) ? "2.5Gb/s" :
2199 (hw
->bus
.speed
== e1000_bus_speed_5000
) ? "5.0Gb/s" :
2201 ((hw
->bus
.width
== e1000_bus_width_pcie_x4
) ? "Width x4" :
2202 (hw
->bus
.width
== e1000_bus_width_pcie_x2
) ? "Width x2" :
2203 (hw
->bus
.width
== e1000_bus_width_pcie_x1
) ? "Width x1" :
2207 ret_val
= igb_read_part_string(hw
, part_str
, E1000_PBANUM_LENGTH
);
2209 strcpy(part_str
, "Unknown");
2210 dev_info(&pdev
->dev
, "%s: PBA No: %s\n", netdev
->name
, part_str
);
2211 dev_info(&pdev
->dev
,
2212 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2213 adapter
->msix_entries
? "MSI-X" :
2214 (adapter
->flags
& IGB_FLAG_HAS_MSI
) ? "MSI" : "legacy",
2215 adapter
->num_rx_queues
, adapter
->num_tx_queues
);
2216 switch (hw
->mac
.type
) {
2220 igb_set_eee_i350(hw
);
2226 pm_runtime_put_noidle(&pdev
->dev
);
2230 igb_release_hw_control(adapter
);
2232 if (!igb_check_reset_block(hw
))
2235 if (hw
->flash_address
)
2236 iounmap(hw
->flash_address
);
2238 igb_clear_interrupt_scheme(adapter
);
2239 iounmap(hw
->hw_addr
);
2241 free_netdev(netdev
);
2243 pci_release_selected_regions(pdev
,
2244 pci_select_bars(pdev
, IORESOURCE_MEM
));
2247 pci_disable_device(pdev
);
2252 * igb_remove - Device Removal Routine
2253 * @pdev: PCI device information struct
2255 * igb_remove is called by the PCI subsystem to alert the driver
2256 * that it should release a PCI device. The could be caused by a
2257 * Hot-Plug event, or because the driver is going to be removed from
2260 static void __devexit
igb_remove(struct pci_dev
*pdev
)
2262 struct net_device
*netdev
= pci_get_drvdata(pdev
);
2263 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2264 struct e1000_hw
*hw
= &adapter
->hw
;
2266 pm_runtime_get_noresume(&pdev
->dev
);
2267 #ifdef CONFIG_IGB_PTP
2268 igb_ptp_stop(adapter
);
2269 #endif /* CONFIG_IGB_PTP */
2272 * The watchdog timer may be rescheduled, so explicitly
2273 * disable watchdog from being rescheduled.
2275 set_bit(__IGB_DOWN
, &adapter
->state
);
2276 del_timer_sync(&adapter
->watchdog_timer
);
2277 del_timer_sync(&adapter
->phy_info_timer
);
2279 cancel_work_sync(&adapter
->reset_task
);
2280 cancel_work_sync(&adapter
->watchdog_task
);
2282 #ifdef CONFIG_IGB_DCA
2283 if (adapter
->flags
& IGB_FLAG_DCA_ENABLED
) {
2284 dev_info(&pdev
->dev
, "DCA disabled\n");
2285 dca_remove_requester(&pdev
->dev
);
2286 adapter
->flags
&= ~IGB_FLAG_DCA_ENABLED
;
2287 wr32(E1000_DCA_CTRL
, E1000_DCA_CTRL_DCA_MODE_DISABLE
);
2291 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2292 * would have already happened in close and is redundant. */
2293 igb_release_hw_control(adapter
);
2295 unregister_netdev(netdev
);
2297 igb_clear_interrupt_scheme(adapter
);
2299 #ifdef CONFIG_PCI_IOV
2300 /* reclaim resources allocated to VFs */
2301 if (adapter
->vf_data
) {
2302 /* disable iov and allow time for transactions to clear */
2303 if (!igb_check_vf_assignment(adapter
)) {
2304 pci_disable_sriov(pdev
);
2307 dev_info(&pdev
->dev
, "VF(s) assigned to guests!\n");
2310 kfree(adapter
->vf_data
);
2311 adapter
->vf_data
= NULL
;
2312 wr32(E1000_IOVCTL
, E1000_IOVCTL_REUSE_VFQ
);
2315 dev_info(&pdev
->dev
, "IOV Disabled\n");
2319 iounmap(hw
->hw_addr
);
2320 if (hw
->flash_address
)
2321 iounmap(hw
->flash_address
);
2322 pci_release_selected_regions(pdev
,
2323 pci_select_bars(pdev
, IORESOURCE_MEM
));
2325 kfree(adapter
->shadow_vfta
);
2326 free_netdev(netdev
);
2328 pci_disable_pcie_error_reporting(pdev
);
2330 pci_disable_device(pdev
);
2334 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2335 * @adapter: board private structure to initialize
2337 * This function initializes the vf specific data storage and then attempts to
2338 * allocate the VFs. The reason for ordering it this way is because it is much
2339 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2340 * the memory for the VFs.
2342 static void __devinit
igb_probe_vfs(struct igb_adapter
* adapter
)
2344 #ifdef CONFIG_PCI_IOV
2345 struct pci_dev
*pdev
= adapter
->pdev
;
2346 struct e1000_hw
*hw
= &adapter
->hw
;
2347 int old_vfs
= igb_find_enabled_vfs(adapter
);
2350 /* Virtualization features not supported on i210 family. */
2351 if ((hw
->mac
.type
== e1000_i210
) || (hw
->mac
.type
== e1000_i211
))
2355 dev_info(&pdev
->dev
, "%d pre-allocated VFs found - override "
2356 "max_vfs setting of %d\n", old_vfs
, max_vfs
);
2357 adapter
->vfs_allocated_count
= old_vfs
;
2360 if (!adapter
->vfs_allocated_count
)
2363 adapter
->vf_data
= kcalloc(adapter
->vfs_allocated_count
,
2364 sizeof(struct vf_data_storage
), GFP_KERNEL
);
2366 /* if allocation failed then we do not support SR-IOV */
2367 if (!adapter
->vf_data
) {
2368 adapter
->vfs_allocated_count
= 0;
2369 dev_err(&pdev
->dev
, "Unable to allocate memory for VF "
2375 if (pci_enable_sriov(pdev
, adapter
->vfs_allocated_count
))
2378 dev_info(&pdev
->dev
, "%d VFs allocated\n",
2379 adapter
->vfs_allocated_count
);
2380 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++)
2381 igb_vf_configure(adapter
, i
);
2383 /* DMA Coalescing is not supported in IOV mode. */
2384 adapter
->flags
&= ~IGB_FLAG_DMAC
;
2387 kfree(adapter
->vf_data
);
2388 adapter
->vf_data
= NULL
;
2389 adapter
->vfs_allocated_count
= 0;
2392 #endif /* CONFIG_PCI_IOV */
2396 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2397 * @adapter: board private structure to initialize
2399 * igb_sw_init initializes the Adapter private data structure.
2400 * Fields are initialized based on PCI device information and
2401 * OS network device settings (MTU size).
2403 static int __devinit
igb_sw_init(struct igb_adapter
*adapter
)
2405 struct e1000_hw
*hw
= &adapter
->hw
;
2406 struct net_device
*netdev
= adapter
->netdev
;
2407 struct pci_dev
*pdev
= adapter
->pdev
;
2410 pci_read_config_word(pdev
, PCI_COMMAND
, &hw
->bus
.pci_cmd_word
);
2412 /* set default ring sizes */
2413 adapter
->tx_ring_count
= IGB_DEFAULT_TXD
;
2414 adapter
->rx_ring_count
= IGB_DEFAULT_RXD
;
2416 /* set default ITR values */
2417 adapter
->rx_itr_setting
= IGB_DEFAULT_ITR
;
2418 adapter
->tx_itr_setting
= IGB_DEFAULT_ITR
;
2420 /* set default work limits */
2421 adapter
->tx_work_limit
= IGB_DEFAULT_TX_WORK
;
2423 adapter
->max_frame_size
= netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+
2425 adapter
->min_frame_size
= ETH_ZLEN
+ ETH_FCS_LEN
;
2429 spin_lock_init(&adapter
->stats64_lock
);
2430 #ifdef CONFIG_PCI_IOV
2431 switch (hw
->mac
.type
) {
2435 dev_warn(&pdev
->dev
,
2436 "Maximum of 7 VFs per PF, using max\n");
2437 adapter
->vfs_allocated_count
= 7;
2439 adapter
->vfs_allocated_count
= max_vfs
;
2444 #endif /* CONFIG_PCI_IOV */
2446 /* Determine the maximum number of RSS queues supported. */
2447 switch (hw
->mac
.type
) {
2449 max_rss_queues
= IGB_MAX_RX_QUEUES_I211
;
2453 max_rss_queues
= IGB_MAX_RX_QUEUES_82575
;
2456 /* I350 cannot do RSS and SR-IOV at the same time */
2457 if (!!adapter
->vfs_allocated_count
) {
2463 if (!!adapter
->vfs_allocated_count
) {
2470 max_rss_queues
= IGB_MAX_RX_QUEUES
;
2474 adapter
->rss_queues
= min_t(u32
, max_rss_queues
, num_online_cpus());
2476 /* Determine if we need to pair queues. */
2477 switch (hw
->mac
.type
) {
2480 /* Device supports enough interrupts without queue pairing. */
2484 * If VFs are going to be allocated with RSS queues then we
2485 * should pair the queues in order to conserve interrupts due
2486 * to limited supply.
2488 if ((adapter
->rss_queues
> 1) &&
2489 (adapter
->vfs_allocated_count
> 6))
2490 adapter
->flags
|= IGB_FLAG_QUEUE_PAIRS
;
2497 * If rss_queues > half of max_rss_queues, pair the queues in
2498 * order to conserve interrupts due to limited supply.
2500 if (adapter
->rss_queues
> (max_rss_queues
/ 2))
2501 adapter
->flags
|= IGB_FLAG_QUEUE_PAIRS
;
2505 /* Setup and initialize a copy of the hw vlan table array */
2506 adapter
->shadow_vfta
= kzalloc(sizeof(u32
) *
2507 E1000_VLAN_FILTER_TBL_SIZE
,
2510 /* This call may decrease the number of queues */
2511 if (igb_init_interrupt_scheme(adapter
)) {
2512 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
2516 igb_probe_vfs(adapter
);
2518 /* Explicitly disable IRQ since the NIC can be in any state. */
2519 igb_irq_disable(adapter
);
2521 if (hw
->mac
.type
>= e1000_i350
)
2522 adapter
->flags
&= ~IGB_FLAG_DMAC
;
2524 set_bit(__IGB_DOWN
, &adapter
->state
);
2529 * igb_open - Called when a network interface is made active
2530 * @netdev: network interface device structure
2532 * Returns 0 on success, negative value on failure
2534 * The open entry point is called when a network interface is made
2535 * active by the system (IFF_UP). At this point all resources needed
2536 * for transmit and receive operations are allocated, the interrupt
2537 * handler is registered with the OS, the watchdog timer is started,
2538 * and the stack is notified that the interface is ready.
2540 static int __igb_open(struct net_device
*netdev
, bool resuming
)
2542 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2543 struct e1000_hw
*hw
= &adapter
->hw
;
2544 struct pci_dev
*pdev
= adapter
->pdev
;
2548 /* disallow open during test */
2549 if (test_bit(__IGB_TESTING
, &adapter
->state
)) {
2555 pm_runtime_get_sync(&pdev
->dev
);
2557 netif_carrier_off(netdev
);
2559 /* allocate transmit descriptors */
2560 err
= igb_setup_all_tx_resources(adapter
);
2564 /* allocate receive descriptors */
2565 err
= igb_setup_all_rx_resources(adapter
);
2569 igb_power_up_link(adapter
);
2571 /* before we allocate an interrupt, we must be ready to handle it.
2572 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2573 * as soon as we call pci_request_irq, so we have to setup our
2574 * clean_rx handler before we do so. */
2575 igb_configure(adapter
);
2577 err
= igb_request_irq(adapter
);
2581 /* From here on the code is the same as igb_up() */
2582 clear_bit(__IGB_DOWN
, &adapter
->state
);
2584 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
2585 napi_enable(&(adapter
->q_vector
[i
]->napi
));
2587 /* Clear any pending interrupts. */
2590 igb_irq_enable(adapter
);
2592 /* notify VFs that reset has been completed */
2593 if (adapter
->vfs_allocated_count
) {
2594 u32 reg_data
= rd32(E1000_CTRL_EXT
);
2595 reg_data
|= E1000_CTRL_EXT_PFRSTD
;
2596 wr32(E1000_CTRL_EXT
, reg_data
);
2599 netif_tx_start_all_queues(netdev
);
2602 pm_runtime_put(&pdev
->dev
);
2604 /* start the watchdog. */
2605 hw
->mac
.get_link_status
= 1;
2606 schedule_work(&adapter
->watchdog_task
);
2611 igb_release_hw_control(adapter
);
2612 igb_power_down_link(adapter
);
2613 igb_free_all_rx_resources(adapter
);
2615 igb_free_all_tx_resources(adapter
);
2619 pm_runtime_put(&pdev
->dev
);
2624 static int igb_open(struct net_device
*netdev
)
2626 return __igb_open(netdev
, false);
2630 * igb_close - Disables a network interface
2631 * @netdev: network interface device structure
2633 * Returns 0, this is not allowed to fail
2635 * The close entry point is called when an interface is de-activated
2636 * by the OS. The hardware is still under the driver's control, but
2637 * needs to be disabled. A global MAC reset is issued to stop the
2638 * hardware, and all transmit and receive resources are freed.
2640 static int __igb_close(struct net_device
*netdev
, bool suspending
)
2642 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2643 struct pci_dev
*pdev
= adapter
->pdev
;
2645 WARN_ON(test_bit(__IGB_RESETTING
, &adapter
->state
));
2648 pm_runtime_get_sync(&pdev
->dev
);
2651 igb_free_irq(adapter
);
2653 igb_free_all_tx_resources(adapter
);
2654 igb_free_all_rx_resources(adapter
);
2657 pm_runtime_put_sync(&pdev
->dev
);
2661 static int igb_close(struct net_device
*netdev
)
2663 return __igb_close(netdev
, false);
2667 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2668 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2670 * Return 0 on success, negative on failure
2672 int igb_setup_tx_resources(struct igb_ring
*tx_ring
)
2674 struct device
*dev
= tx_ring
->dev
;
2675 int orig_node
= dev_to_node(dev
);
2678 size
= sizeof(struct igb_tx_buffer
) * tx_ring
->count
;
2679 tx_ring
->tx_buffer_info
= vzalloc_node(size
, tx_ring
->numa_node
);
2680 if (!tx_ring
->tx_buffer_info
)
2681 tx_ring
->tx_buffer_info
= vzalloc(size
);
2682 if (!tx_ring
->tx_buffer_info
)
2685 /* round up to nearest 4K */
2686 tx_ring
->size
= tx_ring
->count
* sizeof(union e1000_adv_tx_desc
);
2687 tx_ring
->size
= ALIGN(tx_ring
->size
, 4096);
2689 set_dev_node(dev
, tx_ring
->numa_node
);
2690 tx_ring
->desc
= dma_alloc_coherent(dev
,
2694 set_dev_node(dev
, orig_node
);
2696 tx_ring
->desc
= dma_alloc_coherent(dev
,
2704 tx_ring
->next_to_use
= 0;
2705 tx_ring
->next_to_clean
= 0;
2710 vfree(tx_ring
->tx_buffer_info
);
2712 "Unable to allocate memory for the transmit descriptor ring\n");
2717 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2718 * (Descriptors) for all queues
2719 * @adapter: board private structure
2721 * Return 0 on success, negative on failure
2723 static int igb_setup_all_tx_resources(struct igb_adapter
*adapter
)
2725 struct pci_dev
*pdev
= adapter
->pdev
;
2728 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
2729 err
= igb_setup_tx_resources(adapter
->tx_ring
[i
]);
2732 "Allocation for Tx Queue %u failed\n", i
);
2733 for (i
--; i
>= 0; i
--)
2734 igb_free_tx_resources(adapter
->tx_ring
[i
]);
2743 * igb_setup_tctl - configure the transmit control registers
2744 * @adapter: Board private structure
2746 void igb_setup_tctl(struct igb_adapter
*adapter
)
2748 struct e1000_hw
*hw
= &adapter
->hw
;
2751 /* disable queue 0 which is enabled by default on 82575 and 82576 */
2752 wr32(E1000_TXDCTL(0), 0);
2754 /* Program the Transmit Control Register */
2755 tctl
= rd32(E1000_TCTL
);
2756 tctl
&= ~E1000_TCTL_CT
;
2757 tctl
|= E1000_TCTL_PSP
| E1000_TCTL_RTLC
|
2758 (E1000_COLLISION_THRESHOLD
<< E1000_CT_SHIFT
);
2760 igb_config_collision_dist(hw
);
2762 /* Enable transmits */
2763 tctl
|= E1000_TCTL_EN
;
2765 wr32(E1000_TCTL
, tctl
);
2769 * igb_configure_tx_ring - Configure transmit ring after Reset
2770 * @adapter: board private structure
2771 * @ring: tx ring to configure
2773 * Configure a transmit ring after a reset.
2775 void igb_configure_tx_ring(struct igb_adapter
*adapter
,
2776 struct igb_ring
*ring
)
2778 struct e1000_hw
*hw
= &adapter
->hw
;
2780 u64 tdba
= ring
->dma
;
2781 int reg_idx
= ring
->reg_idx
;
2783 /* disable the queue */
2784 wr32(E1000_TXDCTL(reg_idx
), 0);
2788 wr32(E1000_TDLEN(reg_idx
),
2789 ring
->count
* sizeof(union e1000_adv_tx_desc
));
2790 wr32(E1000_TDBAL(reg_idx
),
2791 tdba
& 0x00000000ffffffffULL
);
2792 wr32(E1000_TDBAH(reg_idx
), tdba
>> 32);
2794 ring
->tail
= hw
->hw_addr
+ E1000_TDT(reg_idx
);
2795 wr32(E1000_TDH(reg_idx
), 0);
2796 writel(0, ring
->tail
);
2798 txdctl
|= IGB_TX_PTHRESH
;
2799 txdctl
|= IGB_TX_HTHRESH
<< 8;
2800 txdctl
|= IGB_TX_WTHRESH
<< 16;
2802 txdctl
|= E1000_TXDCTL_QUEUE_ENABLE
;
2803 wr32(E1000_TXDCTL(reg_idx
), txdctl
);
2807 * igb_configure_tx - Configure transmit Unit after Reset
2808 * @adapter: board private structure
2810 * Configure the Tx unit of the MAC after a reset.
2812 static void igb_configure_tx(struct igb_adapter
*adapter
)
2816 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
2817 igb_configure_tx_ring(adapter
, adapter
->tx_ring
[i
]);
2821 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
2822 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2824 * Returns 0 on success, negative on failure
2826 int igb_setup_rx_resources(struct igb_ring
*rx_ring
)
2828 struct device
*dev
= rx_ring
->dev
;
2829 int orig_node
= dev_to_node(dev
);
2832 size
= sizeof(struct igb_rx_buffer
) * rx_ring
->count
;
2833 rx_ring
->rx_buffer_info
= vzalloc_node(size
, rx_ring
->numa_node
);
2834 if (!rx_ring
->rx_buffer_info
)
2835 rx_ring
->rx_buffer_info
= vzalloc(size
);
2836 if (!rx_ring
->rx_buffer_info
)
2839 desc_len
= sizeof(union e1000_adv_rx_desc
);
2841 /* Round up to nearest 4K */
2842 rx_ring
->size
= rx_ring
->count
* desc_len
;
2843 rx_ring
->size
= ALIGN(rx_ring
->size
, 4096);
2845 set_dev_node(dev
, rx_ring
->numa_node
);
2846 rx_ring
->desc
= dma_alloc_coherent(dev
,
2850 set_dev_node(dev
, orig_node
);
2852 rx_ring
->desc
= dma_alloc_coherent(dev
,
2860 rx_ring
->next_to_clean
= 0;
2861 rx_ring
->next_to_use
= 0;
2866 vfree(rx_ring
->rx_buffer_info
);
2867 rx_ring
->rx_buffer_info
= NULL
;
2868 dev_err(dev
, "Unable to allocate memory for the receive descriptor"
2874 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
2875 * (Descriptors) for all queues
2876 * @adapter: board private structure
2878 * Return 0 on success, negative on failure
2880 static int igb_setup_all_rx_resources(struct igb_adapter
*adapter
)
2882 struct pci_dev
*pdev
= adapter
->pdev
;
2885 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
2886 err
= igb_setup_rx_resources(adapter
->rx_ring
[i
]);
2889 "Allocation for Rx Queue %u failed\n", i
);
2890 for (i
--; i
>= 0; i
--)
2891 igb_free_rx_resources(adapter
->rx_ring
[i
]);
2900 * igb_setup_mrqc - configure the multiple receive queue control registers
2901 * @adapter: Board private structure
2903 static void igb_setup_mrqc(struct igb_adapter
*adapter
)
2905 struct e1000_hw
*hw
= &adapter
->hw
;
2907 u32 j
, num_rx_queues
, shift
= 0, shift2
= 0;
2912 static const u8 rsshash
[40] = {
2913 0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2, 0x41, 0x67,
2914 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0, 0xd0, 0xca, 0x2b, 0xcb,
2915 0xae, 0x7b, 0x30, 0xb4, 0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30,
2916 0xf2, 0x0c, 0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa };
2918 /* Fill out hash function seeds */
2919 for (j
= 0; j
< 10; j
++) {
2920 u32 rsskey
= rsshash
[(j
* 4)];
2921 rsskey
|= rsshash
[(j
* 4) + 1] << 8;
2922 rsskey
|= rsshash
[(j
* 4) + 2] << 16;
2923 rsskey
|= rsshash
[(j
* 4) + 3] << 24;
2924 array_wr32(E1000_RSSRK(0), j
, rsskey
);
2927 num_rx_queues
= adapter
->rss_queues
;
2929 if (adapter
->vfs_allocated_count
) {
2930 /* 82575 and 82576 supports 2 RSS queues for VMDq */
2931 switch (hw
->mac
.type
) {
2948 if (hw
->mac
.type
== e1000_82575
)
2952 for (j
= 0; j
< (32 * 4); j
++) {
2953 reta
.bytes
[j
& 3] = (j
% num_rx_queues
) << shift
;
2955 reta
.bytes
[j
& 3] |= num_rx_queues
<< shift2
;
2957 wr32(E1000_RETA(j
>> 2), reta
.dword
);
2961 * Disable raw packet checksumming so that RSS hash is placed in
2962 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
2963 * offloads as they are enabled by default
2965 rxcsum
= rd32(E1000_RXCSUM
);
2966 rxcsum
|= E1000_RXCSUM_PCSD
;
2968 if (adapter
->hw
.mac
.type
>= e1000_82576
)
2969 /* Enable Receive Checksum Offload for SCTP */
2970 rxcsum
|= E1000_RXCSUM_CRCOFL
;
2972 /* Don't need to set TUOFL or IPOFL, they default to 1 */
2973 wr32(E1000_RXCSUM
, rxcsum
);
2975 * Generate RSS hash based on TCP port numbers and/or
2976 * IPv4/v6 src and dst addresses since UDP cannot be
2977 * hashed reliably due to IP fragmentation
2980 mrqc
= E1000_MRQC_RSS_FIELD_IPV4
|
2981 E1000_MRQC_RSS_FIELD_IPV4_TCP
|
2982 E1000_MRQC_RSS_FIELD_IPV6
|
2983 E1000_MRQC_RSS_FIELD_IPV6_TCP
|
2984 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX
;
2986 /* If VMDq is enabled then we set the appropriate mode for that, else
2987 * we default to RSS so that an RSS hash is calculated per packet even
2988 * if we are only using one queue */
2989 if (adapter
->vfs_allocated_count
) {
2990 if (hw
->mac
.type
> e1000_82575
) {
2991 /* Set the default pool for the PF's first queue */
2992 u32 vtctl
= rd32(E1000_VT_CTL
);
2993 vtctl
&= ~(E1000_VT_CTL_DEFAULT_POOL_MASK
|
2994 E1000_VT_CTL_DISABLE_DEF_POOL
);
2995 vtctl
|= adapter
->vfs_allocated_count
<<
2996 E1000_VT_CTL_DEFAULT_POOL_SHIFT
;
2997 wr32(E1000_VT_CTL
, vtctl
);
2999 if (adapter
->rss_queues
> 1)
3000 mrqc
|= E1000_MRQC_ENABLE_VMDQ_RSS_2Q
;
3002 mrqc
|= E1000_MRQC_ENABLE_VMDQ
;
3004 if (hw
->mac
.type
!= e1000_i211
)
3005 mrqc
|= E1000_MRQC_ENABLE_RSS_4Q
;
3007 igb_vmm_control(adapter
);
3009 wr32(E1000_MRQC
, mrqc
);
3013 * igb_setup_rctl - configure the receive control registers
3014 * @adapter: Board private structure
3016 void igb_setup_rctl(struct igb_adapter
*adapter
)
3018 struct e1000_hw
*hw
= &adapter
->hw
;
3021 rctl
= rd32(E1000_RCTL
);
3023 rctl
&= ~(3 << E1000_RCTL_MO_SHIFT
);
3024 rctl
&= ~(E1000_RCTL_LBM_TCVR
| E1000_RCTL_LBM_MAC
);
3026 rctl
|= E1000_RCTL_EN
| E1000_RCTL_BAM
| E1000_RCTL_RDMTS_HALF
|
3027 (hw
->mac
.mc_filter_type
<< E1000_RCTL_MO_SHIFT
);
3030 * enable stripping of CRC. It's unlikely this will break BMC
3031 * redirection as it did with e1000. Newer features require
3032 * that the HW strips the CRC.
3034 rctl
|= E1000_RCTL_SECRC
;
3036 /* disable store bad packets and clear size bits. */
3037 rctl
&= ~(E1000_RCTL_SBP
| E1000_RCTL_SZ_256
);
3039 /* enable LPE to prevent packets larger than max_frame_size */
3040 rctl
|= E1000_RCTL_LPE
;
3042 /* disable queue 0 to prevent tail write w/o re-config */
3043 wr32(E1000_RXDCTL(0), 0);
3045 /* Attention!!! For SR-IOV PF driver operations you must enable
3046 * queue drop for all VF and PF queues to prevent head of line blocking
3047 * if an un-trusted VF does not provide descriptors to hardware.
3049 if (adapter
->vfs_allocated_count
) {
3050 /* set all queue drop enable bits */
3051 wr32(E1000_QDE
, ALL_QUEUES
);
3054 /* This is useful for sniffing bad packets. */
3055 if (adapter
->netdev
->features
& NETIF_F_RXALL
) {
3056 /* UPE and MPE will be handled by normal PROMISC logic
3057 * in e1000e_set_rx_mode */
3058 rctl
|= (E1000_RCTL_SBP
| /* Receive bad packets */
3059 E1000_RCTL_BAM
| /* RX All Bcast Pkts */
3060 E1000_RCTL_PMCF
); /* RX All MAC Ctrl Pkts */
3062 rctl
&= ~(E1000_RCTL_VFE
| /* Disable VLAN filter */
3063 E1000_RCTL_DPF
| /* Allow filtered pause */
3064 E1000_RCTL_CFIEN
); /* Dis VLAN CFIEN Filter */
3065 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3066 * and that breaks VLANs.
3070 wr32(E1000_RCTL
, rctl
);
3073 static inline int igb_set_vf_rlpml(struct igb_adapter
*adapter
, int size
,
3076 struct e1000_hw
*hw
= &adapter
->hw
;
3079 /* if it isn't the PF check to see if VFs are enabled and
3080 * increase the size to support vlan tags */
3081 if (vfn
< adapter
->vfs_allocated_count
&&
3082 adapter
->vf_data
[vfn
].vlans_enabled
)
3083 size
+= VLAN_TAG_SIZE
;
3085 vmolr
= rd32(E1000_VMOLR(vfn
));
3086 vmolr
&= ~E1000_VMOLR_RLPML_MASK
;
3087 vmolr
|= size
| E1000_VMOLR_LPE
;
3088 wr32(E1000_VMOLR(vfn
), vmolr
);
3094 * igb_rlpml_set - set maximum receive packet size
3095 * @adapter: board private structure
3097 * Configure maximum receivable packet size.
3099 static void igb_rlpml_set(struct igb_adapter
*adapter
)
3101 u32 max_frame_size
= adapter
->max_frame_size
;
3102 struct e1000_hw
*hw
= &adapter
->hw
;
3103 u16 pf_id
= adapter
->vfs_allocated_count
;
3106 igb_set_vf_rlpml(adapter
, max_frame_size
, pf_id
);
3108 * If we're in VMDQ or SR-IOV mode, then set global RLPML
3109 * to our max jumbo frame size, in case we need to enable
3110 * jumbo frames on one of the rings later.
3111 * This will not pass over-length frames into the default
3112 * queue because it's gated by the VMOLR.RLPML.
3114 max_frame_size
= MAX_JUMBO_FRAME_SIZE
;
3117 wr32(E1000_RLPML
, max_frame_size
);
3120 static inline void igb_set_vmolr(struct igb_adapter
*adapter
,
3123 struct e1000_hw
*hw
= &adapter
->hw
;
3127 * This register exists only on 82576 and newer so if we are older then
3128 * we should exit and do nothing
3130 if (hw
->mac
.type
< e1000_82576
)
3133 vmolr
= rd32(E1000_VMOLR(vfn
));
3134 vmolr
|= E1000_VMOLR_STRVLAN
; /* Strip vlan tags */
3136 vmolr
|= E1000_VMOLR_AUPE
; /* Accept untagged packets */
3138 vmolr
&= ~(E1000_VMOLR_AUPE
); /* Tagged packets ONLY */
3140 /* clear all bits that might not be set */
3141 vmolr
&= ~(E1000_VMOLR_BAM
| E1000_VMOLR_RSSE
);
3143 if (adapter
->rss_queues
> 1 && vfn
== adapter
->vfs_allocated_count
)
3144 vmolr
|= E1000_VMOLR_RSSE
; /* enable RSS */
3146 * for VMDq only allow the VFs and pool 0 to accept broadcast and
3149 if (vfn
<= adapter
->vfs_allocated_count
)
3150 vmolr
|= E1000_VMOLR_BAM
; /* Accept broadcast */
3152 wr32(E1000_VMOLR(vfn
), vmolr
);
3156 * igb_configure_rx_ring - Configure a receive ring after Reset
3157 * @adapter: board private structure
3158 * @ring: receive ring to be configured
3160 * Configure the Rx unit of the MAC after a reset.
3162 void igb_configure_rx_ring(struct igb_adapter
*adapter
,
3163 struct igb_ring
*ring
)
3165 struct e1000_hw
*hw
= &adapter
->hw
;
3166 u64 rdba
= ring
->dma
;
3167 int reg_idx
= ring
->reg_idx
;
3168 u32 srrctl
= 0, rxdctl
= 0;
3170 /* disable the queue */
3171 wr32(E1000_RXDCTL(reg_idx
), 0);
3173 /* Set DMA base address registers */
3174 wr32(E1000_RDBAL(reg_idx
),
3175 rdba
& 0x00000000ffffffffULL
);
3176 wr32(E1000_RDBAH(reg_idx
), rdba
>> 32);
3177 wr32(E1000_RDLEN(reg_idx
),
3178 ring
->count
* sizeof(union e1000_adv_rx_desc
));
3180 /* initialize head and tail */
3181 ring
->tail
= hw
->hw_addr
+ E1000_RDT(reg_idx
);
3182 wr32(E1000_RDH(reg_idx
), 0);
3183 writel(0, ring
->tail
);
3185 /* set descriptor configuration */
3186 srrctl
= IGB_RX_HDR_LEN
<< E1000_SRRCTL_BSIZEHDRSIZE_SHIFT
;
3187 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3188 srrctl
|= IGB_RXBUFFER_16384
>> E1000_SRRCTL_BSIZEPKT_SHIFT
;
3190 srrctl
|= (PAGE_SIZE
/ 2) >> E1000_SRRCTL_BSIZEPKT_SHIFT
;
3192 srrctl
|= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS
;
3193 #ifdef CONFIG_IGB_PTP
3194 if (hw
->mac
.type
>= e1000_82580
)
3195 srrctl
|= E1000_SRRCTL_TIMESTAMP
;
3196 #endif /* CONFIG_IGB_PTP */
3197 /* Only set Drop Enable if we are supporting multiple queues */
3198 if (adapter
->vfs_allocated_count
|| adapter
->num_rx_queues
> 1)
3199 srrctl
|= E1000_SRRCTL_DROP_EN
;
3201 wr32(E1000_SRRCTL(reg_idx
), srrctl
);
3203 /* set filtering for VMDQ pools */
3204 igb_set_vmolr(adapter
, reg_idx
& 0x7, true);
3206 rxdctl
|= IGB_RX_PTHRESH
;
3207 rxdctl
|= IGB_RX_HTHRESH
<< 8;
3208 rxdctl
|= IGB_RX_WTHRESH
<< 16;
3210 /* enable receive descriptor fetching */
3211 rxdctl
|= E1000_RXDCTL_QUEUE_ENABLE
;
3212 wr32(E1000_RXDCTL(reg_idx
), rxdctl
);
3216 * igb_configure_rx - Configure receive Unit after Reset
3217 * @adapter: board private structure
3219 * Configure the Rx unit of the MAC after a reset.
3221 static void igb_configure_rx(struct igb_adapter
*adapter
)
3225 /* set UTA to appropriate mode */
3226 igb_set_uta(adapter
);
3228 /* set the correct pool for the PF default MAC address in entry 0 */
3229 igb_rar_set_qsel(adapter
, adapter
->hw
.mac
.addr
, 0,
3230 adapter
->vfs_allocated_count
);
3232 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3233 * the Base and Length of the Rx Descriptor Ring */
3234 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3235 igb_configure_rx_ring(adapter
, adapter
->rx_ring
[i
]);
3239 * igb_free_tx_resources - Free Tx Resources per Queue
3240 * @tx_ring: Tx descriptor ring for a specific queue
3242 * Free all transmit software resources
3244 void igb_free_tx_resources(struct igb_ring
*tx_ring
)
3246 igb_clean_tx_ring(tx_ring
);
3248 vfree(tx_ring
->tx_buffer_info
);
3249 tx_ring
->tx_buffer_info
= NULL
;
3251 /* if not set, then don't free */
3255 dma_free_coherent(tx_ring
->dev
, tx_ring
->size
,
3256 tx_ring
->desc
, tx_ring
->dma
);
3258 tx_ring
->desc
= NULL
;
3262 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3263 * @adapter: board private structure
3265 * Free all transmit software resources
3267 static void igb_free_all_tx_resources(struct igb_adapter
*adapter
)
3271 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3272 igb_free_tx_resources(adapter
->tx_ring
[i
]);
3275 void igb_unmap_and_free_tx_resource(struct igb_ring
*ring
,
3276 struct igb_tx_buffer
*tx_buffer
)
3278 if (tx_buffer
->skb
) {
3279 dev_kfree_skb_any(tx_buffer
->skb
);
3281 dma_unmap_single(ring
->dev
,
3285 } else if (tx_buffer
->dma
) {
3286 dma_unmap_page(ring
->dev
,
3291 tx_buffer
->next_to_watch
= NULL
;
3292 tx_buffer
->skb
= NULL
;
3294 /* buffer_info must be completely set up in the transmit path */
3298 * igb_clean_tx_ring - Free Tx Buffers
3299 * @tx_ring: ring to be cleaned
3301 static void igb_clean_tx_ring(struct igb_ring
*tx_ring
)
3303 struct igb_tx_buffer
*buffer_info
;
3307 if (!tx_ring
->tx_buffer_info
)
3309 /* Free all the Tx ring sk_buffs */
3311 for (i
= 0; i
< tx_ring
->count
; i
++) {
3312 buffer_info
= &tx_ring
->tx_buffer_info
[i
];
3313 igb_unmap_and_free_tx_resource(tx_ring
, buffer_info
);
3316 netdev_tx_reset_queue(txring_txq(tx_ring
));
3318 size
= sizeof(struct igb_tx_buffer
) * tx_ring
->count
;
3319 memset(tx_ring
->tx_buffer_info
, 0, size
);
3321 /* Zero out the descriptor ring */
3322 memset(tx_ring
->desc
, 0, tx_ring
->size
);
3324 tx_ring
->next_to_use
= 0;
3325 tx_ring
->next_to_clean
= 0;
3329 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3330 * @adapter: board private structure
3332 static void igb_clean_all_tx_rings(struct igb_adapter
*adapter
)
3336 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3337 igb_clean_tx_ring(adapter
->tx_ring
[i
]);
3341 * igb_free_rx_resources - Free Rx Resources
3342 * @rx_ring: ring to clean the resources from
3344 * Free all receive software resources
3346 void igb_free_rx_resources(struct igb_ring
*rx_ring
)
3348 igb_clean_rx_ring(rx_ring
);
3350 vfree(rx_ring
->rx_buffer_info
);
3351 rx_ring
->rx_buffer_info
= NULL
;
3353 /* if not set, then don't free */
3357 dma_free_coherent(rx_ring
->dev
, rx_ring
->size
,
3358 rx_ring
->desc
, rx_ring
->dma
);
3360 rx_ring
->desc
= NULL
;
3364 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3365 * @adapter: board private structure
3367 * Free all receive software resources
3369 static void igb_free_all_rx_resources(struct igb_adapter
*adapter
)
3373 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3374 igb_free_rx_resources(adapter
->rx_ring
[i
]);
3378 * igb_clean_rx_ring - Free Rx Buffers per Queue
3379 * @rx_ring: ring to free buffers from
3381 static void igb_clean_rx_ring(struct igb_ring
*rx_ring
)
3386 if (!rx_ring
->rx_buffer_info
)
3389 /* Free all the Rx ring sk_buffs */
3390 for (i
= 0; i
< rx_ring
->count
; i
++) {
3391 struct igb_rx_buffer
*buffer_info
= &rx_ring
->rx_buffer_info
[i
];
3392 if (buffer_info
->dma
) {
3393 dma_unmap_single(rx_ring
->dev
,
3397 buffer_info
->dma
= 0;
3400 if (buffer_info
->skb
) {
3401 dev_kfree_skb(buffer_info
->skb
);
3402 buffer_info
->skb
= NULL
;
3404 if (buffer_info
->page_dma
) {
3405 dma_unmap_page(rx_ring
->dev
,
3406 buffer_info
->page_dma
,
3409 buffer_info
->page_dma
= 0;
3411 if (buffer_info
->page
) {
3412 put_page(buffer_info
->page
);
3413 buffer_info
->page
= NULL
;
3414 buffer_info
->page_offset
= 0;
3418 size
= sizeof(struct igb_rx_buffer
) * rx_ring
->count
;
3419 memset(rx_ring
->rx_buffer_info
, 0, size
);
3421 /* Zero out the descriptor ring */
3422 memset(rx_ring
->desc
, 0, rx_ring
->size
);
3424 rx_ring
->next_to_clean
= 0;
3425 rx_ring
->next_to_use
= 0;
3429 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3430 * @adapter: board private structure
3432 static void igb_clean_all_rx_rings(struct igb_adapter
*adapter
)
3436 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3437 igb_clean_rx_ring(adapter
->rx_ring
[i
]);
3441 * igb_set_mac - Change the Ethernet Address of the NIC
3442 * @netdev: network interface device structure
3443 * @p: pointer to an address structure
3445 * Returns 0 on success, negative on failure
3447 static int igb_set_mac(struct net_device
*netdev
, void *p
)
3449 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3450 struct e1000_hw
*hw
= &adapter
->hw
;
3451 struct sockaddr
*addr
= p
;
3453 if (!is_valid_ether_addr(addr
->sa_data
))
3454 return -EADDRNOTAVAIL
;
3456 memcpy(netdev
->dev_addr
, addr
->sa_data
, netdev
->addr_len
);
3457 memcpy(hw
->mac
.addr
, addr
->sa_data
, netdev
->addr_len
);
3459 /* set the correct pool for the new PF MAC address in entry 0 */
3460 igb_rar_set_qsel(adapter
, hw
->mac
.addr
, 0,
3461 adapter
->vfs_allocated_count
);
3467 * igb_write_mc_addr_list - write multicast addresses to MTA
3468 * @netdev: network interface device structure
3470 * Writes multicast address list to the MTA hash table.
3471 * Returns: -ENOMEM on failure
3472 * 0 on no addresses written
3473 * X on writing X addresses to MTA
3475 static int igb_write_mc_addr_list(struct net_device
*netdev
)
3477 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3478 struct e1000_hw
*hw
= &adapter
->hw
;
3479 struct netdev_hw_addr
*ha
;
3483 if (netdev_mc_empty(netdev
)) {
3484 /* nothing to program, so clear mc list */
3485 igb_update_mc_addr_list(hw
, NULL
, 0);
3486 igb_restore_vf_multicasts(adapter
);
3490 mta_list
= kzalloc(netdev_mc_count(netdev
) * 6, GFP_ATOMIC
);
3494 /* The shared function expects a packed array of only addresses. */
3496 netdev_for_each_mc_addr(ha
, netdev
)
3497 memcpy(mta_list
+ (i
++ * ETH_ALEN
), ha
->addr
, ETH_ALEN
);
3499 igb_update_mc_addr_list(hw
, mta_list
, i
);
3502 return netdev_mc_count(netdev
);
3506 * igb_write_uc_addr_list - write unicast addresses to RAR table
3507 * @netdev: network interface device structure
3509 * Writes unicast address list to the RAR table.
3510 * Returns: -ENOMEM on failure/insufficient address space
3511 * 0 on no addresses written
3512 * X on writing X addresses to the RAR table
3514 static int igb_write_uc_addr_list(struct net_device
*netdev
)
3516 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3517 struct e1000_hw
*hw
= &adapter
->hw
;
3518 unsigned int vfn
= adapter
->vfs_allocated_count
;
3519 unsigned int rar_entries
= hw
->mac
.rar_entry_count
- (vfn
+ 1);
3522 /* return ENOMEM indicating insufficient memory for addresses */
3523 if (netdev_uc_count(netdev
) > rar_entries
)
3526 if (!netdev_uc_empty(netdev
) && rar_entries
) {
3527 struct netdev_hw_addr
*ha
;
3529 netdev_for_each_uc_addr(ha
, netdev
) {
3532 igb_rar_set_qsel(adapter
, ha
->addr
,
3538 /* write the addresses in reverse order to avoid write combining */
3539 for (; rar_entries
> 0 ; rar_entries
--) {
3540 wr32(E1000_RAH(rar_entries
), 0);
3541 wr32(E1000_RAL(rar_entries
), 0);
3549 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3550 * @netdev: network interface device structure
3552 * The set_rx_mode entry point is called whenever the unicast or multicast
3553 * address lists or the network interface flags are updated. This routine is
3554 * responsible for configuring the hardware for proper unicast, multicast,
3555 * promiscuous mode, and all-multi behavior.
3557 static void igb_set_rx_mode(struct net_device
*netdev
)
3559 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3560 struct e1000_hw
*hw
= &adapter
->hw
;
3561 unsigned int vfn
= adapter
->vfs_allocated_count
;
3562 u32 rctl
, vmolr
= 0;
3565 /* Check for Promiscuous and All Multicast modes */
3566 rctl
= rd32(E1000_RCTL
);
3568 /* clear the effected bits */
3569 rctl
&= ~(E1000_RCTL_UPE
| E1000_RCTL_MPE
| E1000_RCTL_VFE
);
3571 if (netdev
->flags
& IFF_PROMISC
) {
3572 rctl
|= (E1000_RCTL_UPE
| E1000_RCTL_MPE
);
3573 vmolr
|= (E1000_VMOLR_ROPE
| E1000_VMOLR_MPME
);
3575 if (netdev
->flags
& IFF_ALLMULTI
) {
3576 rctl
|= E1000_RCTL_MPE
;
3577 vmolr
|= E1000_VMOLR_MPME
;
3580 * Write addresses to the MTA, if the attempt fails
3581 * then we should just turn on promiscuous mode so
3582 * that we can at least receive multicast traffic
3584 count
= igb_write_mc_addr_list(netdev
);
3586 rctl
|= E1000_RCTL_MPE
;
3587 vmolr
|= E1000_VMOLR_MPME
;
3589 vmolr
|= E1000_VMOLR_ROMPE
;
3593 * Write addresses to available RAR registers, if there is not
3594 * sufficient space to store all the addresses then enable
3595 * unicast promiscuous mode
3597 count
= igb_write_uc_addr_list(netdev
);
3599 rctl
|= E1000_RCTL_UPE
;
3600 vmolr
|= E1000_VMOLR_ROPE
;
3602 rctl
|= E1000_RCTL_VFE
;
3604 wr32(E1000_RCTL
, rctl
);
3607 * In order to support SR-IOV and eventually VMDq it is necessary to set
3608 * the VMOLR to enable the appropriate modes. Without this workaround
3609 * we will have issues with VLAN tag stripping not being done for frames
3610 * that are only arriving because we are the default pool
3612 if ((hw
->mac
.type
< e1000_82576
) || (hw
->mac
.type
> e1000_i350
))
3615 vmolr
|= rd32(E1000_VMOLR(vfn
)) &
3616 ~(E1000_VMOLR_ROPE
| E1000_VMOLR_MPME
| E1000_VMOLR_ROMPE
);
3617 wr32(E1000_VMOLR(vfn
), vmolr
);
3618 igb_restore_vf_multicasts(adapter
);
3621 static void igb_check_wvbr(struct igb_adapter
*adapter
)
3623 struct e1000_hw
*hw
= &adapter
->hw
;
3626 switch (hw
->mac
.type
) {
3629 if (!(wvbr
= rd32(E1000_WVBR
)))
3636 adapter
->wvbr
|= wvbr
;
3639 #define IGB_STAGGERED_QUEUE_OFFSET 8
3641 static void igb_spoof_check(struct igb_adapter
*adapter
)
3648 for(j
= 0; j
< adapter
->vfs_allocated_count
; j
++) {
3649 if (adapter
->wvbr
& (1 << j
) ||
3650 adapter
->wvbr
& (1 << (j
+ IGB_STAGGERED_QUEUE_OFFSET
))) {
3651 dev_warn(&adapter
->pdev
->dev
,
3652 "Spoof event(s) detected on VF %d\n", j
);
3655 (1 << (j
+ IGB_STAGGERED_QUEUE_OFFSET
)));
3660 /* Need to wait a few seconds after link up to get diagnostic information from
3662 static void igb_update_phy_info(unsigned long data
)
3664 struct igb_adapter
*adapter
= (struct igb_adapter
*) data
;
3665 igb_get_phy_info(&adapter
->hw
);
3669 * igb_has_link - check shared code for link and determine up/down
3670 * @adapter: pointer to driver private info
3672 bool igb_has_link(struct igb_adapter
*adapter
)
3674 struct e1000_hw
*hw
= &adapter
->hw
;
3675 bool link_active
= false;
3678 /* get_link_status is set on LSC (link status) interrupt or
3679 * rx sequence error interrupt. get_link_status will stay
3680 * false until the e1000_check_for_link establishes link
3681 * for copper adapters ONLY
3683 switch (hw
->phy
.media_type
) {
3684 case e1000_media_type_copper
:
3685 if (hw
->mac
.get_link_status
) {
3686 ret_val
= hw
->mac
.ops
.check_for_link(hw
);
3687 link_active
= !hw
->mac
.get_link_status
;
3692 case e1000_media_type_internal_serdes
:
3693 ret_val
= hw
->mac
.ops
.check_for_link(hw
);
3694 link_active
= hw
->mac
.serdes_has_link
;
3697 case e1000_media_type_unknown
:
3704 static bool igb_thermal_sensor_event(struct e1000_hw
*hw
, u32 event
)
3707 u32 ctrl_ext
, thstat
;
3709 /* check for thermal sensor event on i350 copper only */
3710 if (hw
->mac
.type
== e1000_i350
) {
3711 thstat
= rd32(E1000_THSTAT
);
3712 ctrl_ext
= rd32(E1000_CTRL_EXT
);
3714 if ((hw
->phy
.media_type
== e1000_media_type_copper
) &&
3715 !(ctrl_ext
& E1000_CTRL_EXT_LINK_MODE_SGMII
)) {
3716 ret
= !!(thstat
& event
);
3724 * igb_watchdog - Timer Call-back
3725 * @data: pointer to adapter cast into an unsigned long
3727 static void igb_watchdog(unsigned long data
)
3729 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
3730 /* Do the rest outside of interrupt context */
3731 schedule_work(&adapter
->watchdog_task
);
3734 static void igb_watchdog_task(struct work_struct
*work
)
3736 struct igb_adapter
*adapter
= container_of(work
,
3739 struct e1000_hw
*hw
= &adapter
->hw
;
3740 struct net_device
*netdev
= adapter
->netdev
;
3744 link
= igb_has_link(adapter
);
3746 /* Cancel scheduled suspend requests. */
3747 pm_runtime_resume(netdev
->dev
.parent
);
3749 if (!netif_carrier_ok(netdev
)) {
3751 hw
->mac
.ops
.get_speed_and_duplex(hw
,
3752 &adapter
->link_speed
,
3753 &adapter
->link_duplex
);
3755 ctrl
= rd32(E1000_CTRL
);
3756 /* Links status message must follow this format */
3757 printk(KERN_INFO
"igb: %s NIC Link is Up %d Mbps %s "
3758 "Duplex, Flow Control: %s\n",
3760 adapter
->link_speed
,
3761 adapter
->link_duplex
== FULL_DUPLEX
?
3763 (ctrl
& E1000_CTRL_TFCE
) &&
3764 (ctrl
& E1000_CTRL_RFCE
) ? "RX/TX" :
3765 (ctrl
& E1000_CTRL_RFCE
) ? "RX" :
3766 (ctrl
& E1000_CTRL_TFCE
) ? "TX" : "None");
3768 /* check for thermal sensor event */
3769 if (igb_thermal_sensor_event(hw
,
3770 E1000_THSTAT_LINK_THROTTLE
)) {
3771 netdev_info(netdev
, "The network adapter link "
3772 "speed was downshifted because it "
3776 /* adjust timeout factor according to speed/duplex */
3777 adapter
->tx_timeout_factor
= 1;
3778 switch (adapter
->link_speed
) {
3780 adapter
->tx_timeout_factor
= 14;
3783 /* maybe add some timeout factor ? */
3787 netif_carrier_on(netdev
);
3789 igb_ping_all_vfs(adapter
);
3790 igb_check_vf_rate_limit(adapter
);
3792 /* link state has changed, schedule phy info update */
3793 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
3794 mod_timer(&adapter
->phy_info_timer
,
3795 round_jiffies(jiffies
+ 2 * HZ
));
3798 if (netif_carrier_ok(netdev
)) {
3799 adapter
->link_speed
= 0;
3800 adapter
->link_duplex
= 0;
3802 /* check for thermal sensor event */
3803 if (igb_thermal_sensor_event(hw
,
3804 E1000_THSTAT_PWR_DOWN
)) {
3805 netdev_err(netdev
, "The network adapter was "
3806 "stopped because it overheated\n");
3809 /* Links status message must follow this format */
3810 printk(KERN_INFO
"igb: %s NIC Link is Down\n",
3812 netif_carrier_off(netdev
);
3814 igb_ping_all_vfs(adapter
);
3816 /* link state has changed, schedule phy info update */
3817 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
3818 mod_timer(&adapter
->phy_info_timer
,
3819 round_jiffies(jiffies
+ 2 * HZ
));
3821 pm_schedule_suspend(netdev
->dev
.parent
,
3826 spin_lock(&adapter
->stats64_lock
);
3827 igb_update_stats(adapter
, &adapter
->stats64
);
3828 spin_unlock(&adapter
->stats64_lock
);
3830 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
3831 struct igb_ring
*tx_ring
= adapter
->tx_ring
[i
];
3832 if (!netif_carrier_ok(netdev
)) {
3833 /* We've lost link, so the controller stops DMA,
3834 * but we've got queued Tx work that's never going
3835 * to get done, so reset controller to flush Tx.
3836 * (Do the reset outside of interrupt context). */
3837 if (igb_desc_unused(tx_ring
) + 1 < tx_ring
->count
) {
3838 adapter
->tx_timeout_count
++;
3839 schedule_work(&adapter
->reset_task
);
3840 /* return immediately since reset is imminent */
3845 /* Force detection of hung controller every watchdog period */
3846 set_bit(IGB_RING_FLAG_TX_DETECT_HANG
, &tx_ring
->flags
);
3849 /* Cause software interrupt to ensure rx ring is cleaned */
3850 if (adapter
->msix_entries
) {
3852 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
3853 eics
|= adapter
->q_vector
[i
]->eims_value
;
3854 wr32(E1000_EICS
, eics
);
3856 wr32(E1000_ICS
, E1000_ICS_RXDMT0
);
3859 igb_spoof_check(adapter
);
3861 /* Reset the timer */
3862 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
3863 mod_timer(&adapter
->watchdog_timer
,
3864 round_jiffies(jiffies
+ 2 * HZ
));
3867 enum latency_range
{
3871 latency_invalid
= 255
3875 * igb_update_ring_itr - update the dynamic ITR value based on packet size
3877 * Stores a new ITR value based on strictly on packet size. This
3878 * algorithm is less sophisticated than that used in igb_update_itr,
3879 * due to the difficulty of synchronizing statistics across multiple
3880 * receive rings. The divisors and thresholds used by this function
3881 * were determined based on theoretical maximum wire speed and testing
3882 * data, in order to minimize response time while increasing bulk
3884 * This functionality is controlled by the InterruptThrottleRate module
3885 * parameter (see igb_param.c)
3886 * NOTE: This function is called only when operating in a multiqueue
3887 * receive environment.
3888 * @q_vector: pointer to q_vector
3890 static void igb_update_ring_itr(struct igb_q_vector
*q_vector
)
3892 int new_val
= q_vector
->itr_val
;
3893 int avg_wire_size
= 0;
3894 struct igb_adapter
*adapter
= q_vector
->adapter
;
3895 unsigned int packets
;
3897 /* For non-gigabit speeds, just fix the interrupt rate at 4000
3898 * ints/sec - ITR timer value of 120 ticks.
3900 if (adapter
->link_speed
!= SPEED_1000
) {
3901 new_val
= IGB_4K_ITR
;
3905 packets
= q_vector
->rx
.total_packets
;
3907 avg_wire_size
= q_vector
->rx
.total_bytes
/ packets
;
3909 packets
= q_vector
->tx
.total_packets
;
3911 avg_wire_size
= max_t(u32
, avg_wire_size
,
3912 q_vector
->tx
.total_bytes
/ packets
);
3914 /* if avg_wire_size isn't set no work was done */
3918 /* Add 24 bytes to size to account for CRC, preamble, and gap */
3919 avg_wire_size
+= 24;
3921 /* Don't starve jumbo frames */
3922 avg_wire_size
= min(avg_wire_size
, 3000);
3924 /* Give a little boost to mid-size frames */
3925 if ((avg_wire_size
> 300) && (avg_wire_size
< 1200))
3926 new_val
= avg_wire_size
/ 3;
3928 new_val
= avg_wire_size
/ 2;
3930 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3931 if (new_val
< IGB_20K_ITR
&&
3932 ((q_vector
->rx
.ring
&& adapter
->rx_itr_setting
== 3) ||
3933 (!q_vector
->rx
.ring
&& adapter
->tx_itr_setting
== 3)))
3934 new_val
= IGB_20K_ITR
;
3937 if (new_val
!= q_vector
->itr_val
) {
3938 q_vector
->itr_val
= new_val
;
3939 q_vector
->set_itr
= 1;
3942 q_vector
->rx
.total_bytes
= 0;
3943 q_vector
->rx
.total_packets
= 0;
3944 q_vector
->tx
.total_bytes
= 0;
3945 q_vector
->tx
.total_packets
= 0;
3949 * igb_update_itr - update the dynamic ITR value based on statistics
3950 * Stores a new ITR value based on packets and byte
3951 * counts during the last interrupt. The advantage of per interrupt
3952 * computation is faster updates and more accurate ITR for the current
3953 * traffic pattern. Constants in this function were computed
3954 * based on theoretical maximum wire speed and thresholds were set based
3955 * on testing data as well as attempting to minimize response time
3956 * while increasing bulk throughput.
3957 * this functionality is controlled by the InterruptThrottleRate module
3958 * parameter (see igb_param.c)
3959 * NOTE: These calculations are only valid when operating in a single-
3960 * queue environment.
3961 * @q_vector: pointer to q_vector
3962 * @ring_container: ring info to update the itr for
3964 static void igb_update_itr(struct igb_q_vector
*q_vector
,
3965 struct igb_ring_container
*ring_container
)
3967 unsigned int packets
= ring_container
->total_packets
;
3968 unsigned int bytes
= ring_container
->total_bytes
;
3969 u8 itrval
= ring_container
->itr
;
3971 /* no packets, exit with status unchanged */
3976 case lowest_latency
:
3977 /* handle TSO and jumbo frames */
3978 if (bytes
/packets
> 8000)
3979 itrval
= bulk_latency
;
3980 else if ((packets
< 5) && (bytes
> 512))
3981 itrval
= low_latency
;
3983 case low_latency
: /* 50 usec aka 20000 ints/s */
3984 if (bytes
> 10000) {
3985 /* this if handles the TSO accounting */
3986 if (bytes
/packets
> 8000) {
3987 itrval
= bulk_latency
;
3988 } else if ((packets
< 10) || ((bytes
/packets
) > 1200)) {
3989 itrval
= bulk_latency
;
3990 } else if ((packets
> 35)) {
3991 itrval
= lowest_latency
;
3993 } else if (bytes
/packets
> 2000) {
3994 itrval
= bulk_latency
;
3995 } else if (packets
<= 2 && bytes
< 512) {
3996 itrval
= lowest_latency
;
3999 case bulk_latency
: /* 250 usec aka 4000 ints/s */
4000 if (bytes
> 25000) {
4002 itrval
= low_latency
;
4003 } else if (bytes
< 1500) {
4004 itrval
= low_latency
;
4009 /* clear work counters since we have the values we need */
4010 ring_container
->total_bytes
= 0;
4011 ring_container
->total_packets
= 0;
4013 /* write updated itr to ring container */
4014 ring_container
->itr
= itrval
;
4017 static void igb_set_itr(struct igb_q_vector
*q_vector
)
4019 struct igb_adapter
*adapter
= q_vector
->adapter
;
4020 u32 new_itr
= q_vector
->itr_val
;
4023 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4024 if (adapter
->link_speed
!= SPEED_1000
) {
4026 new_itr
= IGB_4K_ITR
;
4030 igb_update_itr(q_vector
, &q_vector
->tx
);
4031 igb_update_itr(q_vector
, &q_vector
->rx
);
4033 current_itr
= max(q_vector
->rx
.itr
, q_vector
->tx
.itr
);
4035 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4036 if (current_itr
== lowest_latency
&&
4037 ((q_vector
->rx
.ring
&& adapter
->rx_itr_setting
== 3) ||
4038 (!q_vector
->rx
.ring
&& adapter
->tx_itr_setting
== 3)))
4039 current_itr
= low_latency
;
4041 switch (current_itr
) {
4042 /* counts and packets in update_itr are dependent on these numbers */
4043 case lowest_latency
:
4044 new_itr
= IGB_70K_ITR
; /* 70,000 ints/sec */
4047 new_itr
= IGB_20K_ITR
; /* 20,000 ints/sec */
4050 new_itr
= IGB_4K_ITR
; /* 4,000 ints/sec */
4057 if (new_itr
!= q_vector
->itr_val
) {
4058 /* this attempts to bias the interrupt rate towards Bulk
4059 * by adding intermediate steps when interrupt rate is
4061 new_itr
= new_itr
> q_vector
->itr_val
?
4062 max((new_itr
* q_vector
->itr_val
) /
4063 (new_itr
+ (q_vector
->itr_val
>> 2)),
4066 /* Don't write the value here; it resets the adapter's
4067 * internal timer, and causes us to delay far longer than
4068 * we should between interrupts. Instead, we write the ITR
4069 * value at the beginning of the next interrupt so the timing
4070 * ends up being correct.
4072 q_vector
->itr_val
= new_itr
;
4073 q_vector
->set_itr
= 1;
4077 static void igb_tx_ctxtdesc(struct igb_ring
*tx_ring
, u32 vlan_macip_lens
,
4078 u32 type_tucmd
, u32 mss_l4len_idx
)
4080 struct e1000_adv_tx_context_desc
*context_desc
;
4081 u16 i
= tx_ring
->next_to_use
;
4083 context_desc
= IGB_TX_CTXTDESC(tx_ring
, i
);
4086 tx_ring
->next_to_use
= (i
< tx_ring
->count
) ? i
: 0;
4088 /* set bits to identify this as an advanced context descriptor */
4089 type_tucmd
|= E1000_TXD_CMD_DEXT
| E1000_ADVTXD_DTYP_CTXT
;
4091 /* For 82575, context index must be unique per ring. */
4092 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX
, &tx_ring
->flags
))
4093 mss_l4len_idx
|= tx_ring
->reg_idx
<< 4;
4095 context_desc
->vlan_macip_lens
= cpu_to_le32(vlan_macip_lens
);
4096 context_desc
->seqnum_seed
= 0;
4097 context_desc
->type_tucmd_mlhl
= cpu_to_le32(type_tucmd
);
4098 context_desc
->mss_l4len_idx
= cpu_to_le32(mss_l4len_idx
);
4101 static int igb_tso(struct igb_ring
*tx_ring
,
4102 struct igb_tx_buffer
*first
,
4105 struct sk_buff
*skb
= first
->skb
;
4106 u32 vlan_macip_lens
, type_tucmd
;
4107 u32 mss_l4len_idx
, l4len
;
4109 if (!skb_is_gso(skb
))
4112 if (skb_header_cloned(skb
)) {
4113 int err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
4118 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4119 type_tucmd
= E1000_ADVTXD_TUCMD_L4T_TCP
;
4121 if (first
->protocol
== __constant_htons(ETH_P_IP
)) {
4122 struct iphdr
*iph
= ip_hdr(skb
);
4125 tcp_hdr(skb
)->check
= ~csum_tcpudp_magic(iph
->saddr
,
4129 type_tucmd
|= E1000_ADVTXD_TUCMD_IPV4
;
4130 first
->tx_flags
|= IGB_TX_FLAGS_TSO
|
4133 } else if (skb_is_gso_v6(skb
)) {
4134 ipv6_hdr(skb
)->payload_len
= 0;
4135 tcp_hdr(skb
)->check
= ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
4136 &ipv6_hdr(skb
)->daddr
,
4138 first
->tx_flags
|= IGB_TX_FLAGS_TSO
|
4142 /* compute header lengths */
4143 l4len
= tcp_hdrlen(skb
);
4144 *hdr_len
= skb_transport_offset(skb
) + l4len
;
4146 /* update gso size and bytecount with header size */
4147 first
->gso_segs
= skb_shinfo(skb
)->gso_segs
;
4148 first
->bytecount
+= (first
->gso_segs
- 1) * *hdr_len
;
4151 mss_l4len_idx
= l4len
<< E1000_ADVTXD_L4LEN_SHIFT
;
4152 mss_l4len_idx
|= skb_shinfo(skb
)->gso_size
<< E1000_ADVTXD_MSS_SHIFT
;
4154 /* VLAN MACLEN IPLEN */
4155 vlan_macip_lens
= skb_network_header_len(skb
);
4156 vlan_macip_lens
|= skb_network_offset(skb
) << E1000_ADVTXD_MACLEN_SHIFT
;
4157 vlan_macip_lens
|= first
->tx_flags
& IGB_TX_FLAGS_VLAN_MASK
;
4159 igb_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, type_tucmd
, mss_l4len_idx
);
4164 static void igb_tx_csum(struct igb_ring
*tx_ring
, struct igb_tx_buffer
*first
)
4166 struct sk_buff
*skb
= first
->skb
;
4167 u32 vlan_macip_lens
= 0;
4168 u32 mss_l4len_idx
= 0;
4171 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
) {
4172 if (!(first
->tx_flags
& IGB_TX_FLAGS_VLAN
))
4176 switch (first
->protocol
) {
4177 case __constant_htons(ETH_P_IP
):
4178 vlan_macip_lens
|= skb_network_header_len(skb
);
4179 type_tucmd
|= E1000_ADVTXD_TUCMD_IPV4
;
4180 l4_hdr
= ip_hdr(skb
)->protocol
;
4182 case __constant_htons(ETH_P_IPV6
):
4183 vlan_macip_lens
|= skb_network_header_len(skb
);
4184 l4_hdr
= ipv6_hdr(skb
)->nexthdr
;
4187 if (unlikely(net_ratelimit())) {
4188 dev_warn(tx_ring
->dev
,
4189 "partial checksum but proto=%x!\n",
4197 type_tucmd
|= E1000_ADVTXD_TUCMD_L4T_TCP
;
4198 mss_l4len_idx
= tcp_hdrlen(skb
) <<
4199 E1000_ADVTXD_L4LEN_SHIFT
;
4202 type_tucmd
|= E1000_ADVTXD_TUCMD_L4T_SCTP
;
4203 mss_l4len_idx
= sizeof(struct sctphdr
) <<
4204 E1000_ADVTXD_L4LEN_SHIFT
;
4207 mss_l4len_idx
= sizeof(struct udphdr
) <<
4208 E1000_ADVTXD_L4LEN_SHIFT
;
4211 if (unlikely(net_ratelimit())) {
4212 dev_warn(tx_ring
->dev
,
4213 "partial checksum but l4 proto=%x!\n",
4219 /* update TX checksum flag */
4220 first
->tx_flags
|= IGB_TX_FLAGS_CSUM
;
4223 vlan_macip_lens
|= skb_network_offset(skb
) << E1000_ADVTXD_MACLEN_SHIFT
;
4224 vlan_macip_lens
|= first
->tx_flags
& IGB_TX_FLAGS_VLAN_MASK
;
4226 igb_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, type_tucmd
, mss_l4len_idx
);
4229 static __le32
igb_tx_cmd_type(u32 tx_flags
)
4231 /* set type for advanced descriptor with frame checksum insertion */
4232 __le32 cmd_type
= cpu_to_le32(E1000_ADVTXD_DTYP_DATA
|
4233 E1000_ADVTXD_DCMD_IFCS
|
4234 E1000_ADVTXD_DCMD_DEXT
);
4236 /* set HW vlan bit if vlan is present */
4237 if (tx_flags
& IGB_TX_FLAGS_VLAN
)
4238 cmd_type
|= cpu_to_le32(E1000_ADVTXD_DCMD_VLE
);
4240 #ifdef CONFIG_IGB_PTP
4241 /* set timestamp bit if present */
4242 if (unlikely(tx_flags
& IGB_TX_FLAGS_TSTAMP
))
4243 cmd_type
|= cpu_to_le32(E1000_ADVTXD_MAC_TSTAMP
);
4244 #endif /* CONFIG_IGB_PTP */
4246 /* set segmentation bits for TSO */
4247 if (tx_flags
& IGB_TX_FLAGS_TSO
)
4248 cmd_type
|= cpu_to_le32(E1000_ADVTXD_DCMD_TSE
);
4253 static void igb_tx_olinfo_status(struct igb_ring
*tx_ring
,
4254 union e1000_adv_tx_desc
*tx_desc
,
4255 u32 tx_flags
, unsigned int paylen
)
4257 u32 olinfo_status
= paylen
<< E1000_ADVTXD_PAYLEN_SHIFT
;
4259 /* 82575 requires a unique index per ring if any offload is enabled */
4260 if ((tx_flags
& (IGB_TX_FLAGS_CSUM
| IGB_TX_FLAGS_VLAN
)) &&
4261 test_bit(IGB_RING_FLAG_TX_CTX_IDX
, &tx_ring
->flags
))
4262 olinfo_status
|= tx_ring
->reg_idx
<< 4;
4264 /* insert L4 checksum */
4265 if (tx_flags
& IGB_TX_FLAGS_CSUM
) {
4266 olinfo_status
|= E1000_TXD_POPTS_TXSM
<< 8;
4268 /* insert IPv4 checksum */
4269 if (tx_flags
& IGB_TX_FLAGS_IPV4
)
4270 olinfo_status
|= E1000_TXD_POPTS_IXSM
<< 8;
4273 tx_desc
->read
.olinfo_status
= cpu_to_le32(olinfo_status
);
4277 * The largest size we can write to the descriptor is 65535. In order to
4278 * maintain a power of two alignment we have to limit ourselves to 32K.
4280 #define IGB_MAX_TXD_PWR 15
4281 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
4283 static void igb_tx_map(struct igb_ring
*tx_ring
,
4284 struct igb_tx_buffer
*first
,
4287 struct sk_buff
*skb
= first
->skb
;
4288 struct igb_tx_buffer
*tx_buffer_info
;
4289 union e1000_adv_tx_desc
*tx_desc
;
4291 struct skb_frag_struct
*frag
= &skb_shinfo(skb
)->frags
[0];
4292 unsigned int data_len
= skb
->data_len
;
4293 unsigned int size
= skb_headlen(skb
);
4294 unsigned int paylen
= skb
->len
- hdr_len
;
4296 u32 tx_flags
= first
->tx_flags
;
4297 u16 i
= tx_ring
->next_to_use
;
4299 tx_desc
= IGB_TX_DESC(tx_ring
, i
);
4301 igb_tx_olinfo_status(tx_ring
, tx_desc
, tx_flags
, paylen
);
4302 cmd_type
= igb_tx_cmd_type(tx_flags
);
4304 dma
= dma_map_single(tx_ring
->dev
, skb
->data
, size
, DMA_TO_DEVICE
);
4305 if (dma_mapping_error(tx_ring
->dev
, dma
))
4308 /* record length, and DMA address */
4309 first
->length
= size
;
4311 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
4314 while (unlikely(size
> IGB_MAX_DATA_PER_TXD
)) {
4315 tx_desc
->read
.cmd_type_len
=
4316 cmd_type
| cpu_to_le32(IGB_MAX_DATA_PER_TXD
);
4320 if (i
== tx_ring
->count
) {
4321 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
4325 dma
+= IGB_MAX_DATA_PER_TXD
;
4326 size
-= IGB_MAX_DATA_PER_TXD
;
4328 tx_desc
->read
.olinfo_status
= 0;
4329 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
4332 if (likely(!data_len
))
4335 tx_desc
->read
.cmd_type_len
= cmd_type
| cpu_to_le32(size
);
4339 if (i
== tx_ring
->count
) {
4340 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
4344 size
= skb_frag_size(frag
);
4347 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0,
4348 size
, DMA_TO_DEVICE
);
4349 if (dma_mapping_error(tx_ring
->dev
, dma
))
4352 tx_buffer_info
= &tx_ring
->tx_buffer_info
[i
];
4353 tx_buffer_info
->length
= size
;
4354 tx_buffer_info
->dma
= dma
;
4356 tx_desc
->read
.olinfo_status
= 0;
4357 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
4362 netdev_tx_sent_queue(txring_txq(tx_ring
), first
->bytecount
);
4364 /* write last descriptor with RS and EOP bits */
4365 cmd_type
|= cpu_to_le32(size
) | cpu_to_le32(IGB_TXD_DCMD
);
4366 if (unlikely(skb
->no_fcs
))
4367 cmd_type
&= ~(cpu_to_le32(E1000_ADVTXD_DCMD_IFCS
));
4368 tx_desc
->read
.cmd_type_len
= cmd_type
;
4370 /* set the timestamp */
4371 first
->time_stamp
= jiffies
;
4374 * Force memory writes to complete before letting h/w know there
4375 * are new descriptors to fetch. (Only applicable for weak-ordered
4376 * memory model archs, such as IA-64).
4378 * We also need this memory barrier to make certain all of the
4379 * status bits have been updated before next_to_watch is written.
4383 /* set next_to_watch value indicating a packet is present */
4384 first
->next_to_watch
= tx_desc
;
4387 if (i
== tx_ring
->count
)
4390 tx_ring
->next_to_use
= i
;
4392 writel(i
, tx_ring
->tail
);
4394 /* we need this if more than one processor can write to our tail
4395 * at a time, it syncronizes IO on IA64/Altix systems */
4401 dev_err(tx_ring
->dev
, "TX DMA map failed\n");
4403 /* clear dma mappings for failed tx_buffer_info map */
4405 tx_buffer_info
= &tx_ring
->tx_buffer_info
[i
];
4406 igb_unmap_and_free_tx_resource(tx_ring
, tx_buffer_info
);
4407 if (tx_buffer_info
== first
)
4414 tx_ring
->next_to_use
= i
;
4417 static int __igb_maybe_stop_tx(struct igb_ring
*tx_ring
, const u16 size
)
4419 struct net_device
*netdev
= tx_ring
->netdev
;
4421 netif_stop_subqueue(netdev
, tx_ring
->queue_index
);
4423 /* Herbert's original patch had:
4424 * smp_mb__after_netif_stop_queue();
4425 * but since that doesn't exist yet, just open code it. */
4428 /* We need to check again in a case another CPU has just
4429 * made room available. */
4430 if (igb_desc_unused(tx_ring
) < size
)
4434 netif_wake_subqueue(netdev
, tx_ring
->queue_index
);
4436 u64_stats_update_begin(&tx_ring
->tx_syncp2
);
4437 tx_ring
->tx_stats
.restart_queue2
++;
4438 u64_stats_update_end(&tx_ring
->tx_syncp2
);
4443 static inline int igb_maybe_stop_tx(struct igb_ring
*tx_ring
, const u16 size
)
4445 if (igb_desc_unused(tx_ring
) >= size
)
4447 return __igb_maybe_stop_tx(tx_ring
, size
);
4450 netdev_tx_t
igb_xmit_frame_ring(struct sk_buff
*skb
,
4451 struct igb_ring
*tx_ring
)
4453 #ifdef CONFIG_IGB_PTP
4454 struct igb_adapter
*adapter
= netdev_priv(tx_ring
->netdev
);
4455 #endif /* CONFIG_IGB_PTP */
4456 struct igb_tx_buffer
*first
;
4459 __be16 protocol
= vlan_get_protocol(skb
);
4462 /* need: 1 descriptor per page,
4463 * + 2 desc gap to keep tail from touching head,
4464 * + 1 desc for skb->data,
4465 * + 1 desc for context descriptor,
4466 * otherwise try next time */
4467 if (igb_maybe_stop_tx(tx_ring
, skb_shinfo(skb
)->nr_frags
+ 4)) {
4468 /* this is a hard error */
4469 return NETDEV_TX_BUSY
;
4472 /* record the location of the first descriptor for this packet */
4473 first
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_use
];
4475 first
->bytecount
= skb
->len
;
4476 first
->gso_segs
= 1;
4478 #ifdef CONFIG_IGB_PTP
4479 if (unlikely((skb_shinfo(skb
)->tx_flags
& SKBTX_HW_TSTAMP
) &&
4480 !(adapter
->ptp_tx_skb
))) {
4481 skb_shinfo(skb
)->tx_flags
|= SKBTX_IN_PROGRESS
;
4482 tx_flags
|= IGB_TX_FLAGS_TSTAMP
;
4484 adapter
->ptp_tx_skb
= skb_get(skb
);
4485 if (adapter
->hw
.mac
.type
== e1000_82576
)
4486 schedule_work(&adapter
->ptp_tx_work
);
4488 #endif /* CONFIG_IGB_PTP */
4490 if (vlan_tx_tag_present(skb
)) {
4491 tx_flags
|= IGB_TX_FLAGS_VLAN
;
4492 tx_flags
|= (vlan_tx_tag_get(skb
) << IGB_TX_FLAGS_VLAN_SHIFT
);
4495 /* record initial flags and protocol */
4496 first
->tx_flags
= tx_flags
;
4497 first
->protocol
= protocol
;
4499 tso
= igb_tso(tx_ring
, first
, &hdr_len
);
4503 igb_tx_csum(tx_ring
, first
);
4505 igb_tx_map(tx_ring
, first
, hdr_len
);
4507 /* Make sure there is space in the ring for the next send. */
4508 igb_maybe_stop_tx(tx_ring
, MAX_SKB_FRAGS
+ 4);
4510 return NETDEV_TX_OK
;
4513 igb_unmap_and_free_tx_resource(tx_ring
, first
);
4515 return NETDEV_TX_OK
;
4518 static inline struct igb_ring
*igb_tx_queue_mapping(struct igb_adapter
*adapter
,
4519 struct sk_buff
*skb
)
4521 unsigned int r_idx
= skb
->queue_mapping
;
4523 if (r_idx
>= adapter
->num_tx_queues
)
4524 r_idx
= r_idx
% adapter
->num_tx_queues
;
4526 return adapter
->tx_ring
[r_idx
];
4529 static netdev_tx_t
igb_xmit_frame(struct sk_buff
*skb
,
4530 struct net_device
*netdev
)
4532 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4534 if (test_bit(__IGB_DOWN
, &adapter
->state
)) {
4535 dev_kfree_skb_any(skb
);
4536 return NETDEV_TX_OK
;
4539 if (skb
->len
<= 0) {
4540 dev_kfree_skb_any(skb
);
4541 return NETDEV_TX_OK
;
4545 * The minimum packet size with TCTL.PSP set is 17 so pad the skb
4546 * in order to meet this minimum size requirement.
4548 if (skb
->len
< 17) {
4549 if (skb_padto(skb
, 17))
4550 return NETDEV_TX_OK
;
4554 return igb_xmit_frame_ring(skb
, igb_tx_queue_mapping(adapter
, skb
));
4558 * igb_tx_timeout - Respond to a Tx Hang
4559 * @netdev: network interface device structure
4561 static void igb_tx_timeout(struct net_device
*netdev
)
4563 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4564 struct e1000_hw
*hw
= &adapter
->hw
;
4566 /* Do the reset outside of interrupt context */
4567 adapter
->tx_timeout_count
++;
4569 if (hw
->mac
.type
>= e1000_82580
)
4570 hw
->dev_spec
._82575
.global_device_reset
= true;
4572 schedule_work(&adapter
->reset_task
);
4574 (adapter
->eims_enable_mask
& ~adapter
->eims_other
));
4577 static void igb_reset_task(struct work_struct
*work
)
4579 struct igb_adapter
*adapter
;
4580 adapter
= container_of(work
, struct igb_adapter
, reset_task
);
4583 netdev_err(adapter
->netdev
, "Reset adapter\n");
4584 igb_reinit_locked(adapter
);
4588 * igb_get_stats64 - Get System Network Statistics
4589 * @netdev: network interface device structure
4590 * @stats: rtnl_link_stats64 pointer
4593 static struct rtnl_link_stats64
*igb_get_stats64(struct net_device
*netdev
,
4594 struct rtnl_link_stats64
*stats
)
4596 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4598 spin_lock(&adapter
->stats64_lock
);
4599 igb_update_stats(adapter
, &adapter
->stats64
);
4600 memcpy(stats
, &adapter
->stats64
, sizeof(*stats
));
4601 spin_unlock(&adapter
->stats64_lock
);
4607 * igb_change_mtu - Change the Maximum Transfer Unit
4608 * @netdev: network interface device structure
4609 * @new_mtu: new value for maximum frame size
4611 * Returns 0 on success, negative on failure
4613 static int igb_change_mtu(struct net_device
*netdev
, int new_mtu
)
4615 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4616 struct pci_dev
*pdev
= adapter
->pdev
;
4617 int max_frame
= new_mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+ VLAN_HLEN
;
4619 if ((new_mtu
< 68) || (max_frame
> MAX_JUMBO_FRAME_SIZE
)) {
4620 dev_err(&pdev
->dev
, "Invalid MTU setting\n");
4624 #define MAX_STD_JUMBO_FRAME_SIZE 9238
4625 if (max_frame
> MAX_STD_JUMBO_FRAME_SIZE
) {
4626 dev_err(&pdev
->dev
, "MTU > 9216 not supported.\n");
4630 while (test_and_set_bit(__IGB_RESETTING
, &adapter
->state
))
4633 /* igb_down has a dependency on max_frame_size */
4634 adapter
->max_frame_size
= max_frame
;
4636 if (netif_running(netdev
))
4639 dev_info(&pdev
->dev
, "changing MTU from %d to %d\n",
4640 netdev
->mtu
, new_mtu
);
4641 netdev
->mtu
= new_mtu
;
4643 if (netif_running(netdev
))
4648 clear_bit(__IGB_RESETTING
, &adapter
->state
);
4654 * igb_update_stats - Update the board statistics counters
4655 * @adapter: board private structure
4658 void igb_update_stats(struct igb_adapter
*adapter
,
4659 struct rtnl_link_stats64
*net_stats
)
4661 struct e1000_hw
*hw
= &adapter
->hw
;
4662 struct pci_dev
*pdev
= adapter
->pdev
;
4668 u64 _bytes
, _packets
;
4670 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
4673 * Prevent stats update while adapter is being reset, or if the pci
4674 * connection is down.
4676 if (adapter
->link_speed
== 0)
4678 if (pci_channel_offline(pdev
))
4683 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
4684 u32 rqdpc_tmp
= rd32(E1000_RQDPC(i
)) & 0x0FFF;
4685 struct igb_ring
*ring
= adapter
->rx_ring
[i
];
4687 ring
->rx_stats
.drops
+= rqdpc_tmp
;
4688 net_stats
->rx_fifo_errors
+= rqdpc_tmp
;
4691 start
= u64_stats_fetch_begin_bh(&ring
->rx_syncp
);
4692 _bytes
= ring
->rx_stats
.bytes
;
4693 _packets
= ring
->rx_stats
.packets
;
4694 } while (u64_stats_fetch_retry_bh(&ring
->rx_syncp
, start
));
4696 packets
+= _packets
;
4699 net_stats
->rx_bytes
= bytes
;
4700 net_stats
->rx_packets
= packets
;
4704 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
4705 struct igb_ring
*ring
= adapter
->tx_ring
[i
];
4707 start
= u64_stats_fetch_begin_bh(&ring
->tx_syncp
);
4708 _bytes
= ring
->tx_stats
.bytes
;
4709 _packets
= ring
->tx_stats
.packets
;
4710 } while (u64_stats_fetch_retry_bh(&ring
->tx_syncp
, start
));
4712 packets
+= _packets
;
4714 net_stats
->tx_bytes
= bytes
;
4715 net_stats
->tx_packets
= packets
;
4717 /* read stats registers */
4718 adapter
->stats
.crcerrs
+= rd32(E1000_CRCERRS
);
4719 adapter
->stats
.gprc
+= rd32(E1000_GPRC
);
4720 adapter
->stats
.gorc
+= rd32(E1000_GORCL
);
4721 rd32(E1000_GORCH
); /* clear GORCL */
4722 adapter
->stats
.bprc
+= rd32(E1000_BPRC
);
4723 adapter
->stats
.mprc
+= rd32(E1000_MPRC
);
4724 adapter
->stats
.roc
+= rd32(E1000_ROC
);
4726 adapter
->stats
.prc64
+= rd32(E1000_PRC64
);
4727 adapter
->stats
.prc127
+= rd32(E1000_PRC127
);
4728 adapter
->stats
.prc255
+= rd32(E1000_PRC255
);
4729 adapter
->stats
.prc511
+= rd32(E1000_PRC511
);
4730 adapter
->stats
.prc1023
+= rd32(E1000_PRC1023
);
4731 adapter
->stats
.prc1522
+= rd32(E1000_PRC1522
);
4732 adapter
->stats
.symerrs
+= rd32(E1000_SYMERRS
);
4733 adapter
->stats
.sec
+= rd32(E1000_SEC
);
4735 mpc
= rd32(E1000_MPC
);
4736 adapter
->stats
.mpc
+= mpc
;
4737 net_stats
->rx_fifo_errors
+= mpc
;
4738 adapter
->stats
.scc
+= rd32(E1000_SCC
);
4739 adapter
->stats
.ecol
+= rd32(E1000_ECOL
);
4740 adapter
->stats
.mcc
+= rd32(E1000_MCC
);
4741 adapter
->stats
.latecol
+= rd32(E1000_LATECOL
);
4742 adapter
->stats
.dc
+= rd32(E1000_DC
);
4743 adapter
->stats
.rlec
+= rd32(E1000_RLEC
);
4744 adapter
->stats
.xonrxc
+= rd32(E1000_XONRXC
);
4745 adapter
->stats
.xontxc
+= rd32(E1000_XONTXC
);
4746 adapter
->stats
.xoffrxc
+= rd32(E1000_XOFFRXC
);
4747 adapter
->stats
.xofftxc
+= rd32(E1000_XOFFTXC
);
4748 adapter
->stats
.fcruc
+= rd32(E1000_FCRUC
);
4749 adapter
->stats
.gptc
+= rd32(E1000_GPTC
);
4750 adapter
->stats
.gotc
+= rd32(E1000_GOTCL
);
4751 rd32(E1000_GOTCH
); /* clear GOTCL */
4752 adapter
->stats
.rnbc
+= rd32(E1000_RNBC
);
4753 adapter
->stats
.ruc
+= rd32(E1000_RUC
);
4754 adapter
->stats
.rfc
+= rd32(E1000_RFC
);
4755 adapter
->stats
.rjc
+= rd32(E1000_RJC
);
4756 adapter
->stats
.tor
+= rd32(E1000_TORH
);
4757 adapter
->stats
.tot
+= rd32(E1000_TOTH
);
4758 adapter
->stats
.tpr
+= rd32(E1000_TPR
);
4760 adapter
->stats
.ptc64
+= rd32(E1000_PTC64
);
4761 adapter
->stats
.ptc127
+= rd32(E1000_PTC127
);
4762 adapter
->stats
.ptc255
+= rd32(E1000_PTC255
);
4763 adapter
->stats
.ptc511
+= rd32(E1000_PTC511
);
4764 adapter
->stats
.ptc1023
+= rd32(E1000_PTC1023
);
4765 adapter
->stats
.ptc1522
+= rd32(E1000_PTC1522
);
4767 adapter
->stats
.mptc
+= rd32(E1000_MPTC
);
4768 adapter
->stats
.bptc
+= rd32(E1000_BPTC
);
4770 adapter
->stats
.tpt
+= rd32(E1000_TPT
);
4771 adapter
->stats
.colc
+= rd32(E1000_COLC
);
4773 adapter
->stats
.algnerrc
+= rd32(E1000_ALGNERRC
);
4774 /* read internal phy specific stats */
4775 reg
= rd32(E1000_CTRL_EXT
);
4776 if (!(reg
& E1000_CTRL_EXT_LINK_MODE_MASK
)) {
4777 adapter
->stats
.rxerrc
+= rd32(E1000_RXERRC
);
4778 adapter
->stats
.tncrs
+= rd32(E1000_TNCRS
);
4781 adapter
->stats
.tsctc
+= rd32(E1000_TSCTC
);
4782 adapter
->stats
.tsctfc
+= rd32(E1000_TSCTFC
);
4784 adapter
->stats
.iac
+= rd32(E1000_IAC
);
4785 adapter
->stats
.icrxoc
+= rd32(E1000_ICRXOC
);
4786 adapter
->stats
.icrxptc
+= rd32(E1000_ICRXPTC
);
4787 adapter
->stats
.icrxatc
+= rd32(E1000_ICRXATC
);
4788 adapter
->stats
.ictxptc
+= rd32(E1000_ICTXPTC
);
4789 adapter
->stats
.ictxatc
+= rd32(E1000_ICTXATC
);
4790 adapter
->stats
.ictxqec
+= rd32(E1000_ICTXQEC
);
4791 adapter
->stats
.ictxqmtc
+= rd32(E1000_ICTXQMTC
);
4792 adapter
->stats
.icrxdmtc
+= rd32(E1000_ICRXDMTC
);
4794 /* Fill out the OS statistics structure */
4795 net_stats
->multicast
= adapter
->stats
.mprc
;
4796 net_stats
->collisions
= adapter
->stats
.colc
;
4800 /* RLEC on some newer hardware can be incorrect so build
4801 * our own version based on RUC and ROC */
4802 net_stats
->rx_errors
= adapter
->stats
.rxerrc
+
4803 adapter
->stats
.crcerrs
+ adapter
->stats
.algnerrc
+
4804 adapter
->stats
.ruc
+ adapter
->stats
.roc
+
4805 adapter
->stats
.cexterr
;
4806 net_stats
->rx_length_errors
= adapter
->stats
.ruc
+
4808 net_stats
->rx_crc_errors
= adapter
->stats
.crcerrs
;
4809 net_stats
->rx_frame_errors
= adapter
->stats
.algnerrc
;
4810 net_stats
->rx_missed_errors
= adapter
->stats
.mpc
;
4813 net_stats
->tx_errors
= adapter
->stats
.ecol
+
4814 adapter
->stats
.latecol
;
4815 net_stats
->tx_aborted_errors
= adapter
->stats
.ecol
;
4816 net_stats
->tx_window_errors
= adapter
->stats
.latecol
;
4817 net_stats
->tx_carrier_errors
= adapter
->stats
.tncrs
;
4819 /* Tx Dropped needs to be maintained elsewhere */
4822 if (hw
->phy
.media_type
== e1000_media_type_copper
) {
4823 if ((adapter
->link_speed
== SPEED_1000
) &&
4824 (!igb_read_phy_reg(hw
, PHY_1000T_STATUS
, &phy_tmp
))) {
4825 phy_tmp
&= PHY_IDLE_ERROR_COUNT_MASK
;
4826 adapter
->phy_stats
.idle_errors
+= phy_tmp
;
4830 /* Management Stats */
4831 adapter
->stats
.mgptc
+= rd32(E1000_MGTPTC
);
4832 adapter
->stats
.mgprc
+= rd32(E1000_MGTPRC
);
4833 adapter
->stats
.mgpdc
+= rd32(E1000_MGTPDC
);
4836 reg
= rd32(E1000_MANC
);
4837 if (reg
& E1000_MANC_EN_BMC2OS
) {
4838 adapter
->stats
.o2bgptc
+= rd32(E1000_O2BGPTC
);
4839 adapter
->stats
.o2bspc
+= rd32(E1000_O2BSPC
);
4840 adapter
->stats
.b2ospc
+= rd32(E1000_B2OSPC
);
4841 adapter
->stats
.b2ogprc
+= rd32(E1000_B2OGPRC
);
4845 static irqreturn_t
igb_msix_other(int irq
, void *data
)
4847 struct igb_adapter
*adapter
= data
;
4848 struct e1000_hw
*hw
= &adapter
->hw
;
4849 u32 icr
= rd32(E1000_ICR
);
4850 /* reading ICR causes bit 31 of EICR to be cleared */
4852 if (icr
& E1000_ICR_DRSTA
)
4853 schedule_work(&adapter
->reset_task
);
4855 if (icr
& E1000_ICR_DOUTSYNC
) {
4856 /* HW is reporting DMA is out of sync */
4857 adapter
->stats
.doosync
++;
4858 /* The DMA Out of Sync is also indication of a spoof event
4859 * in IOV mode. Check the Wrong VM Behavior register to
4860 * see if it is really a spoof event. */
4861 igb_check_wvbr(adapter
);
4864 /* Check for a mailbox event */
4865 if (icr
& E1000_ICR_VMMB
)
4866 igb_msg_task(adapter
);
4868 if (icr
& E1000_ICR_LSC
) {
4869 hw
->mac
.get_link_status
= 1;
4870 /* guard against interrupt when we're going down */
4871 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
4872 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
4875 #ifdef CONFIG_IGB_PTP
4876 if (icr
& E1000_ICR_TS
) {
4877 u32 tsicr
= rd32(E1000_TSICR
);
4879 if (tsicr
& E1000_TSICR_TXTS
) {
4880 /* acknowledge the interrupt */
4881 wr32(E1000_TSICR
, E1000_TSICR_TXTS
);
4882 /* retrieve hardware timestamp */
4883 schedule_work(&adapter
->ptp_tx_work
);
4886 #endif /* CONFIG_IGB_PTP */
4888 wr32(E1000_EIMS
, adapter
->eims_other
);
4893 static void igb_write_itr(struct igb_q_vector
*q_vector
)
4895 struct igb_adapter
*adapter
= q_vector
->adapter
;
4896 u32 itr_val
= q_vector
->itr_val
& 0x7FFC;
4898 if (!q_vector
->set_itr
)
4904 if (adapter
->hw
.mac
.type
== e1000_82575
)
4905 itr_val
|= itr_val
<< 16;
4907 itr_val
|= E1000_EITR_CNT_IGNR
;
4909 writel(itr_val
, q_vector
->itr_register
);
4910 q_vector
->set_itr
= 0;
4913 static irqreturn_t
igb_msix_ring(int irq
, void *data
)
4915 struct igb_q_vector
*q_vector
= data
;
4917 /* Write the ITR value calculated from the previous interrupt. */
4918 igb_write_itr(q_vector
);
4920 napi_schedule(&q_vector
->napi
);
4925 #ifdef CONFIG_IGB_DCA
4926 static void igb_update_dca(struct igb_q_vector
*q_vector
)
4928 struct igb_adapter
*adapter
= q_vector
->adapter
;
4929 struct e1000_hw
*hw
= &adapter
->hw
;
4930 int cpu
= get_cpu();
4932 if (q_vector
->cpu
== cpu
)
4935 if (q_vector
->tx
.ring
) {
4936 int q
= q_vector
->tx
.ring
->reg_idx
;
4937 u32 dca_txctrl
= rd32(E1000_DCA_TXCTRL(q
));
4938 if (hw
->mac
.type
== e1000_82575
) {
4939 dca_txctrl
&= ~E1000_DCA_TXCTRL_CPUID_MASK
;
4940 dca_txctrl
|= dca3_get_tag(&adapter
->pdev
->dev
, cpu
);
4942 dca_txctrl
&= ~E1000_DCA_TXCTRL_CPUID_MASK_82576
;
4943 dca_txctrl
|= dca3_get_tag(&adapter
->pdev
->dev
, cpu
) <<
4944 E1000_DCA_TXCTRL_CPUID_SHIFT
;
4946 dca_txctrl
|= E1000_DCA_TXCTRL_DESC_DCA_EN
;
4947 wr32(E1000_DCA_TXCTRL(q
), dca_txctrl
);
4949 if (q_vector
->rx
.ring
) {
4950 int q
= q_vector
->rx
.ring
->reg_idx
;
4951 u32 dca_rxctrl
= rd32(E1000_DCA_RXCTRL(q
));
4952 if (hw
->mac
.type
== e1000_82575
) {
4953 dca_rxctrl
&= ~E1000_DCA_RXCTRL_CPUID_MASK
;
4954 dca_rxctrl
|= dca3_get_tag(&adapter
->pdev
->dev
, cpu
);
4956 dca_rxctrl
&= ~E1000_DCA_RXCTRL_CPUID_MASK_82576
;
4957 dca_rxctrl
|= dca3_get_tag(&adapter
->pdev
->dev
, cpu
) <<
4958 E1000_DCA_RXCTRL_CPUID_SHIFT
;
4960 dca_rxctrl
|= E1000_DCA_RXCTRL_DESC_DCA_EN
;
4961 dca_rxctrl
|= E1000_DCA_RXCTRL_HEAD_DCA_EN
;
4962 dca_rxctrl
|= E1000_DCA_RXCTRL_DATA_DCA_EN
;
4963 wr32(E1000_DCA_RXCTRL(q
), dca_rxctrl
);
4965 q_vector
->cpu
= cpu
;
4970 static void igb_setup_dca(struct igb_adapter
*adapter
)
4972 struct e1000_hw
*hw
= &adapter
->hw
;
4975 if (!(adapter
->flags
& IGB_FLAG_DCA_ENABLED
))
4978 /* Always use CB2 mode, difference is masked in the CB driver. */
4979 wr32(E1000_DCA_CTRL
, E1000_DCA_CTRL_DCA_MODE_CB2
);
4981 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
4982 adapter
->q_vector
[i
]->cpu
= -1;
4983 igb_update_dca(adapter
->q_vector
[i
]);
4987 static int __igb_notify_dca(struct device
*dev
, void *data
)
4989 struct net_device
*netdev
= dev_get_drvdata(dev
);
4990 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4991 struct pci_dev
*pdev
= adapter
->pdev
;
4992 struct e1000_hw
*hw
= &adapter
->hw
;
4993 unsigned long event
= *(unsigned long *)data
;
4996 case DCA_PROVIDER_ADD
:
4997 /* if already enabled, don't do it again */
4998 if (adapter
->flags
& IGB_FLAG_DCA_ENABLED
)
5000 if (dca_add_requester(dev
) == 0) {
5001 adapter
->flags
|= IGB_FLAG_DCA_ENABLED
;
5002 dev_info(&pdev
->dev
, "DCA enabled\n");
5003 igb_setup_dca(adapter
);
5006 /* Fall Through since DCA is disabled. */
5007 case DCA_PROVIDER_REMOVE
:
5008 if (adapter
->flags
& IGB_FLAG_DCA_ENABLED
) {
5009 /* without this a class_device is left
5010 * hanging around in the sysfs model */
5011 dca_remove_requester(dev
);
5012 dev_info(&pdev
->dev
, "DCA disabled\n");
5013 adapter
->flags
&= ~IGB_FLAG_DCA_ENABLED
;
5014 wr32(E1000_DCA_CTRL
, E1000_DCA_CTRL_DCA_MODE_DISABLE
);
5022 static int igb_notify_dca(struct notifier_block
*nb
, unsigned long event
,
5027 ret_val
= driver_for_each_device(&igb_driver
.driver
, NULL
, &event
,
5030 return ret_val
? NOTIFY_BAD
: NOTIFY_DONE
;
5032 #endif /* CONFIG_IGB_DCA */
5034 #ifdef CONFIG_PCI_IOV
5035 static int igb_vf_configure(struct igb_adapter
*adapter
, int vf
)
5037 unsigned char mac_addr
[ETH_ALEN
];
5038 struct pci_dev
*pdev
= adapter
->pdev
;
5039 struct e1000_hw
*hw
= &adapter
->hw
;
5040 struct pci_dev
*pvfdev
;
5041 unsigned int device_id
;
5044 eth_random_addr(mac_addr
);
5045 igb_set_vf_mac(adapter
, vf
, mac_addr
);
5047 switch (adapter
->hw
.mac
.type
) {
5049 device_id
= IGB_82576_VF_DEV_ID
;
5050 /* VF Stride for 82576 is 2 */
5051 thisvf_devfn
= (pdev
->devfn
+ 0x80 + (vf
<< 1)) |
5055 device_id
= IGB_I350_VF_DEV_ID
;
5056 /* VF Stride for I350 is 4 */
5057 thisvf_devfn
= (pdev
->devfn
+ 0x80 + (vf
<< 2)) |
5066 pvfdev
= pci_get_device(hw
->vendor_id
, device_id
, NULL
);
5068 if (pvfdev
->devfn
== thisvf_devfn
)
5070 pvfdev
= pci_get_device(hw
->vendor_id
,
5075 adapter
->vf_data
[vf
].vfdev
= pvfdev
;
5078 "Couldn't find pci dev ptr for VF %4.4x\n",
5080 return pvfdev
!= NULL
;
5083 static int igb_find_enabled_vfs(struct igb_adapter
*adapter
)
5085 struct e1000_hw
*hw
= &adapter
->hw
;
5086 struct pci_dev
*pdev
= adapter
->pdev
;
5087 struct pci_dev
*pvfdev
;
5090 unsigned int device_id
;
5093 switch (adapter
->hw
.mac
.type
) {
5095 device_id
= IGB_82576_VF_DEV_ID
;
5096 /* VF Stride for 82576 is 2 */
5100 device_id
= IGB_I350_VF_DEV_ID
;
5101 /* VF Stride for I350 is 4 */
5110 vf_devfn
= pdev
->devfn
+ 0x80;
5111 pvfdev
= pci_get_device(hw
->vendor_id
, device_id
, NULL
);
5113 if (pvfdev
->devfn
== vf_devfn
&&
5114 (pvfdev
->bus
->number
>= pdev
->bus
->number
))
5116 vf_devfn
+= vf_stride
;
5117 pvfdev
= pci_get_device(hw
->vendor_id
,
5124 static int igb_check_vf_assignment(struct igb_adapter
*adapter
)
5127 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++) {
5128 if (adapter
->vf_data
[i
].vfdev
) {
5129 if (adapter
->vf_data
[i
].vfdev
->dev_flags
&
5130 PCI_DEV_FLAGS_ASSIGNED
)
5138 static void igb_ping_all_vfs(struct igb_adapter
*adapter
)
5140 struct e1000_hw
*hw
= &adapter
->hw
;
5144 for (i
= 0 ; i
< adapter
->vfs_allocated_count
; i
++) {
5145 ping
= E1000_PF_CONTROL_MSG
;
5146 if (adapter
->vf_data
[i
].flags
& IGB_VF_FLAG_CTS
)
5147 ping
|= E1000_VT_MSGTYPE_CTS
;
5148 igb_write_mbx(hw
, &ping
, 1, i
);
5152 static int igb_set_vf_promisc(struct igb_adapter
*adapter
, u32
*msgbuf
, u32 vf
)
5154 struct e1000_hw
*hw
= &adapter
->hw
;
5155 u32 vmolr
= rd32(E1000_VMOLR(vf
));
5156 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5158 vf_data
->flags
&= ~(IGB_VF_FLAG_UNI_PROMISC
|
5159 IGB_VF_FLAG_MULTI_PROMISC
);
5160 vmolr
&= ~(E1000_VMOLR_ROPE
| E1000_VMOLR_ROMPE
| E1000_VMOLR_MPME
);
5162 if (*msgbuf
& E1000_VF_SET_PROMISC_MULTICAST
) {
5163 vmolr
|= E1000_VMOLR_MPME
;
5164 vf_data
->flags
|= IGB_VF_FLAG_MULTI_PROMISC
;
5165 *msgbuf
&= ~E1000_VF_SET_PROMISC_MULTICAST
;
5168 * if we have hashes and we are clearing a multicast promisc
5169 * flag we need to write the hashes to the MTA as this step
5170 * was previously skipped
5172 if (vf_data
->num_vf_mc_hashes
> 30) {
5173 vmolr
|= E1000_VMOLR_MPME
;
5174 } else if (vf_data
->num_vf_mc_hashes
) {
5176 vmolr
|= E1000_VMOLR_ROMPE
;
5177 for (j
= 0; j
< vf_data
->num_vf_mc_hashes
; j
++)
5178 igb_mta_set(hw
, vf_data
->vf_mc_hashes
[j
]);
5182 wr32(E1000_VMOLR(vf
), vmolr
);
5184 /* there are flags left unprocessed, likely not supported */
5185 if (*msgbuf
& E1000_VT_MSGINFO_MASK
)
5192 static int igb_set_vf_multicasts(struct igb_adapter
*adapter
,
5193 u32
*msgbuf
, u32 vf
)
5195 int n
= (msgbuf
[0] & E1000_VT_MSGINFO_MASK
) >> E1000_VT_MSGINFO_SHIFT
;
5196 u16
*hash_list
= (u16
*)&msgbuf
[1];
5197 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5200 /* salt away the number of multicast addresses assigned
5201 * to this VF for later use to restore when the PF multi cast
5204 vf_data
->num_vf_mc_hashes
= n
;
5206 /* only up to 30 hash values supported */
5210 /* store the hashes for later use */
5211 for (i
= 0; i
< n
; i
++)
5212 vf_data
->vf_mc_hashes
[i
] = hash_list
[i
];
5214 /* Flush and reset the mta with the new values */
5215 igb_set_rx_mode(adapter
->netdev
);
5220 static void igb_restore_vf_multicasts(struct igb_adapter
*adapter
)
5222 struct e1000_hw
*hw
= &adapter
->hw
;
5223 struct vf_data_storage
*vf_data
;
5226 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++) {
5227 u32 vmolr
= rd32(E1000_VMOLR(i
));
5228 vmolr
&= ~(E1000_VMOLR_ROMPE
| E1000_VMOLR_MPME
);
5230 vf_data
= &adapter
->vf_data
[i
];
5232 if ((vf_data
->num_vf_mc_hashes
> 30) ||
5233 (vf_data
->flags
& IGB_VF_FLAG_MULTI_PROMISC
)) {
5234 vmolr
|= E1000_VMOLR_MPME
;
5235 } else if (vf_data
->num_vf_mc_hashes
) {
5236 vmolr
|= E1000_VMOLR_ROMPE
;
5237 for (j
= 0; j
< vf_data
->num_vf_mc_hashes
; j
++)
5238 igb_mta_set(hw
, vf_data
->vf_mc_hashes
[j
]);
5240 wr32(E1000_VMOLR(i
), vmolr
);
5244 static void igb_clear_vf_vfta(struct igb_adapter
*adapter
, u32 vf
)
5246 struct e1000_hw
*hw
= &adapter
->hw
;
5247 u32 pool_mask
, reg
, vid
;
5250 pool_mask
= 1 << (E1000_VLVF_POOLSEL_SHIFT
+ vf
);
5252 /* Find the vlan filter for this id */
5253 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5254 reg
= rd32(E1000_VLVF(i
));
5256 /* remove the vf from the pool */
5259 /* if pool is empty then remove entry from vfta */
5260 if (!(reg
& E1000_VLVF_POOLSEL_MASK
) &&
5261 (reg
& E1000_VLVF_VLANID_ENABLE
)) {
5263 vid
= reg
& E1000_VLVF_VLANID_MASK
;
5264 igb_vfta_set(hw
, vid
, false);
5267 wr32(E1000_VLVF(i
), reg
);
5270 adapter
->vf_data
[vf
].vlans_enabled
= 0;
5273 static s32
igb_vlvf_set(struct igb_adapter
*adapter
, u32 vid
, bool add
, u32 vf
)
5275 struct e1000_hw
*hw
= &adapter
->hw
;
5278 /* The vlvf table only exists on 82576 hardware and newer */
5279 if (hw
->mac
.type
< e1000_82576
)
5282 /* we only need to do this if VMDq is enabled */
5283 if (!adapter
->vfs_allocated_count
)
5286 /* Find the vlan filter for this id */
5287 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5288 reg
= rd32(E1000_VLVF(i
));
5289 if ((reg
& E1000_VLVF_VLANID_ENABLE
) &&
5290 vid
== (reg
& E1000_VLVF_VLANID_MASK
))
5295 if (i
== E1000_VLVF_ARRAY_SIZE
) {
5296 /* Did not find a matching VLAN ID entry that was
5297 * enabled. Search for a free filter entry, i.e.
5298 * one without the enable bit set
5300 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5301 reg
= rd32(E1000_VLVF(i
));
5302 if (!(reg
& E1000_VLVF_VLANID_ENABLE
))
5306 if (i
< E1000_VLVF_ARRAY_SIZE
) {
5307 /* Found an enabled/available entry */
5308 reg
|= 1 << (E1000_VLVF_POOLSEL_SHIFT
+ vf
);
5310 /* if !enabled we need to set this up in vfta */
5311 if (!(reg
& E1000_VLVF_VLANID_ENABLE
)) {
5312 /* add VID to filter table */
5313 igb_vfta_set(hw
, vid
, true);
5314 reg
|= E1000_VLVF_VLANID_ENABLE
;
5316 reg
&= ~E1000_VLVF_VLANID_MASK
;
5318 wr32(E1000_VLVF(i
), reg
);
5320 /* do not modify RLPML for PF devices */
5321 if (vf
>= adapter
->vfs_allocated_count
)
5324 if (!adapter
->vf_data
[vf
].vlans_enabled
) {
5326 reg
= rd32(E1000_VMOLR(vf
));
5327 size
= reg
& E1000_VMOLR_RLPML_MASK
;
5329 reg
&= ~E1000_VMOLR_RLPML_MASK
;
5331 wr32(E1000_VMOLR(vf
), reg
);
5334 adapter
->vf_data
[vf
].vlans_enabled
++;
5337 if (i
< E1000_VLVF_ARRAY_SIZE
) {
5338 /* remove vf from the pool */
5339 reg
&= ~(1 << (E1000_VLVF_POOLSEL_SHIFT
+ vf
));
5340 /* if pool is empty then remove entry from vfta */
5341 if (!(reg
& E1000_VLVF_POOLSEL_MASK
)) {
5343 igb_vfta_set(hw
, vid
, false);
5345 wr32(E1000_VLVF(i
), reg
);
5347 /* do not modify RLPML for PF devices */
5348 if (vf
>= adapter
->vfs_allocated_count
)
5351 adapter
->vf_data
[vf
].vlans_enabled
--;
5352 if (!adapter
->vf_data
[vf
].vlans_enabled
) {
5354 reg
= rd32(E1000_VMOLR(vf
));
5355 size
= reg
& E1000_VMOLR_RLPML_MASK
;
5357 reg
&= ~E1000_VMOLR_RLPML_MASK
;
5359 wr32(E1000_VMOLR(vf
), reg
);
5366 static void igb_set_vmvir(struct igb_adapter
*adapter
, u32 vid
, u32 vf
)
5368 struct e1000_hw
*hw
= &adapter
->hw
;
5371 wr32(E1000_VMVIR(vf
), (vid
| E1000_VMVIR_VLANA_DEFAULT
));
5373 wr32(E1000_VMVIR(vf
), 0);
5376 static int igb_ndo_set_vf_vlan(struct net_device
*netdev
,
5377 int vf
, u16 vlan
, u8 qos
)
5380 struct igb_adapter
*adapter
= netdev_priv(netdev
);
5382 if ((vf
>= adapter
->vfs_allocated_count
) || (vlan
> 4095) || (qos
> 7))
5385 err
= igb_vlvf_set(adapter
, vlan
, !!vlan
, vf
);
5388 igb_set_vmvir(adapter
, vlan
| (qos
<< VLAN_PRIO_SHIFT
), vf
);
5389 igb_set_vmolr(adapter
, vf
, !vlan
);
5390 adapter
->vf_data
[vf
].pf_vlan
= vlan
;
5391 adapter
->vf_data
[vf
].pf_qos
= qos
;
5392 dev_info(&adapter
->pdev
->dev
,
5393 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan
, qos
, vf
);
5394 if (test_bit(__IGB_DOWN
, &adapter
->state
)) {
5395 dev_warn(&adapter
->pdev
->dev
,
5396 "The VF VLAN has been set,"
5397 " but the PF device is not up.\n");
5398 dev_warn(&adapter
->pdev
->dev
,
5399 "Bring the PF device up before"
5400 " attempting to use the VF device.\n");
5403 igb_vlvf_set(adapter
, adapter
->vf_data
[vf
].pf_vlan
,
5405 igb_set_vmvir(adapter
, vlan
, vf
);
5406 igb_set_vmolr(adapter
, vf
, true);
5407 adapter
->vf_data
[vf
].pf_vlan
= 0;
5408 adapter
->vf_data
[vf
].pf_qos
= 0;
5414 static int igb_set_vf_vlan(struct igb_adapter
*adapter
, u32
*msgbuf
, u32 vf
)
5416 int add
= (msgbuf
[0] & E1000_VT_MSGINFO_MASK
) >> E1000_VT_MSGINFO_SHIFT
;
5417 int vid
= (msgbuf
[1] & E1000_VLVF_VLANID_MASK
);
5419 return igb_vlvf_set(adapter
, vid
, add
, vf
);
5422 static inline void igb_vf_reset(struct igb_adapter
*adapter
, u32 vf
)
5424 /* clear flags - except flag that indicates PF has set the MAC */
5425 adapter
->vf_data
[vf
].flags
&= IGB_VF_FLAG_PF_SET_MAC
;
5426 adapter
->vf_data
[vf
].last_nack
= jiffies
;
5428 /* reset offloads to defaults */
5429 igb_set_vmolr(adapter
, vf
, true);
5431 /* reset vlans for device */
5432 igb_clear_vf_vfta(adapter
, vf
);
5433 if (adapter
->vf_data
[vf
].pf_vlan
)
5434 igb_ndo_set_vf_vlan(adapter
->netdev
, vf
,
5435 adapter
->vf_data
[vf
].pf_vlan
,
5436 adapter
->vf_data
[vf
].pf_qos
);
5438 igb_clear_vf_vfta(adapter
, vf
);
5440 /* reset multicast table array for vf */
5441 adapter
->vf_data
[vf
].num_vf_mc_hashes
= 0;
5443 /* Flush and reset the mta with the new values */
5444 igb_set_rx_mode(adapter
->netdev
);
5447 static void igb_vf_reset_event(struct igb_adapter
*adapter
, u32 vf
)
5449 unsigned char *vf_mac
= adapter
->vf_data
[vf
].vf_mac_addresses
;
5451 /* generate a new mac address as we were hotplug removed/added */
5452 if (!(adapter
->vf_data
[vf
].flags
& IGB_VF_FLAG_PF_SET_MAC
))
5453 eth_random_addr(vf_mac
);
5455 /* process remaining reset events */
5456 igb_vf_reset(adapter
, vf
);
5459 static void igb_vf_reset_msg(struct igb_adapter
*adapter
, u32 vf
)
5461 struct e1000_hw
*hw
= &adapter
->hw
;
5462 unsigned char *vf_mac
= adapter
->vf_data
[vf
].vf_mac_addresses
;
5463 int rar_entry
= hw
->mac
.rar_entry_count
- (vf
+ 1);
5465 u8
*addr
= (u8
*)(&msgbuf
[1]);
5467 /* process all the same items cleared in a function level reset */
5468 igb_vf_reset(adapter
, vf
);
5470 /* set vf mac address */
5471 igb_rar_set_qsel(adapter
, vf_mac
, rar_entry
, vf
);
5473 /* enable transmit and receive for vf */
5474 reg
= rd32(E1000_VFTE
);
5475 wr32(E1000_VFTE
, reg
| (1 << vf
));
5476 reg
= rd32(E1000_VFRE
);
5477 wr32(E1000_VFRE
, reg
| (1 << vf
));
5479 adapter
->vf_data
[vf
].flags
|= IGB_VF_FLAG_CTS
;
5481 /* reply to reset with ack and vf mac address */
5482 msgbuf
[0] = E1000_VF_RESET
| E1000_VT_MSGTYPE_ACK
;
5483 memcpy(addr
, vf_mac
, 6);
5484 igb_write_mbx(hw
, msgbuf
, 3, vf
);
5487 static int igb_set_vf_mac_addr(struct igb_adapter
*adapter
, u32
*msg
, int vf
)
5490 * The VF MAC Address is stored in a packed array of bytes
5491 * starting at the second 32 bit word of the msg array
5493 unsigned char *addr
= (char *)&msg
[1];
5496 if (is_valid_ether_addr(addr
))
5497 err
= igb_set_vf_mac(adapter
, vf
, addr
);
5502 static void igb_rcv_ack_from_vf(struct igb_adapter
*adapter
, u32 vf
)
5504 struct e1000_hw
*hw
= &adapter
->hw
;
5505 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5506 u32 msg
= E1000_VT_MSGTYPE_NACK
;
5508 /* if device isn't clear to send it shouldn't be reading either */
5509 if (!(vf_data
->flags
& IGB_VF_FLAG_CTS
) &&
5510 time_after(jiffies
, vf_data
->last_nack
+ (2 * HZ
))) {
5511 igb_write_mbx(hw
, &msg
, 1, vf
);
5512 vf_data
->last_nack
= jiffies
;
5516 static void igb_rcv_msg_from_vf(struct igb_adapter
*adapter
, u32 vf
)
5518 struct pci_dev
*pdev
= adapter
->pdev
;
5519 u32 msgbuf
[E1000_VFMAILBOX_SIZE
];
5520 struct e1000_hw
*hw
= &adapter
->hw
;
5521 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5524 retval
= igb_read_mbx(hw
, msgbuf
, E1000_VFMAILBOX_SIZE
, vf
);
5527 /* if receive failed revoke VF CTS stats and restart init */
5528 dev_err(&pdev
->dev
, "Error receiving message from VF\n");
5529 vf_data
->flags
&= ~IGB_VF_FLAG_CTS
;
5530 if (!time_after(jiffies
, vf_data
->last_nack
+ (2 * HZ
)))
5535 /* this is a message we already processed, do nothing */
5536 if (msgbuf
[0] & (E1000_VT_MSGTYPE_ACK
| E1000_VT_MSGTYPE_NACK
))
5540 * until the vf completes a reset it should not be
5541 * allowed to start any configuration.
5544 if (msgbuf
[0] == E1000_VF_RESET
) {
5545 igb_vf_reset_msg(adapter
, vf
);
5549 if (!(vf_data
->flags
& IGB_VF_FLAG_CTS
)) {
5550 if (!time_after(jiffies
, vf_data
->last_nack
+ (2 * HZ
)))
5556 switch ((msgbuf
[0] & 0xFFFF)) {
5557 case E1000_VF_SET_MAC_ADDR
:
5559 if (!(vf_data
->flags
& IGB_VF_FLAG_PF_SET_MAC
))
5560 retval
= igb_set_vf_mac_addr(adapter
, msgbuf
, vf
);
5562 dev_warn(&pdev
->dev
,
5563 "VF %d attempted to override administratively "
5564 "set MAC address\nReload the VF driver to "
5565 "resume operations\n", vf
);
5567 case E1000_VF_SET_PROMISC
:
5568 retval
= igb_set_vf_promisc(adapter
, msgbuf
, vf
);
5570 case E1000_VF_SET_MULTICAST
:
5571 retval
= igb_set_vf_multicasts(adapter
, msgbuf
, vf
);
5573 case E1000_VF_SET_LPE
:
5574 retval
= igb_set_vf_rlpml(adapter
, msgbuf
[1], vf
);
5576 case E1000_VF_SET_VLAN
:
5578 if (vf_data
->pf_vlan
)
5579 dev_warn(&pdev
->dev
,
5580 "VF %d attempted to override administratively "
5581 "set VLAN tag\nReload the VF driver to "
5582 "resume operations\n", vf
);
5584 retval
= igb_set_vf_vlan(adapter
, msgbuf
, vf
);
5587 dev_err(&pdev
->dev
, "Unhandled Msg %08x\n", msgbuf
[0]);
5592 msgbuf
[0] |= E1000_VT_MSGTYPE_CTS
;
5594 /* notify the VF of the results of what it sent us */
5596 msgbuf
[0] |= E1000_VT_MSGTYPE_NACK
;
5598 msgbuf
[0] |= E1000_VT_MSGTYPE_ACK
;
5600 igb_write_mbx(hw
, msgbuf
, 1, vf
);
5603 static void igb_msg_task(struct igb_adapter
*adapter
)
5605 struct e1000_hw
*hw
= &adapter
->hw
;
5608 for (vf
= 0; vf
< adapter
->vfs_allocated_count
; vf
++) {
5609 /* process any reset requests */
5610 if (!igb_check_for_rst(hw
, vf
))
5611 igb_vf_reset_event(adapter
, vf
);
5613 /* process any messages pending */
5614 if (!igb_check_for_msg(hw
, vf
))
5615 igb_rcv_msg_from_vf(adapter
, vf
);
5617 /* process any acks */
5618 if (!igb_check_for_ack(hw
, vf
))
5619 igb_rcv_ack_from_vf(adapter
, vf
);
5624 * igb_set_uta - Set unicast filter table address
5625 * @adapter: board private structure
5627 * The unicast table address is a register array of 32-bit registers.
5628 * The table is meant to be used in a way similar to how the MTA is used
5629 * however due to certain limitations in the hardware it is necessary to
5630 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
5631 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
5633 static void igb_set_uta(struct igb_adapter
*adapter
)
5635 struct e1000_hw
*hw
= &adapter
->hw
;
5638 /* The UTA table only exists on 82576 hardware and newer */
5639 if (hw
->mac
.type
< e1000_82576
)
5642 /* we only need to do this if VMDq is enabled */
5643 if (!adapter
->vfs_allocated_count
)
5646 for (i
= 0; i
< hw
->mac
.uta_reg_count
; i
++)
5647 array_wr32(E1000_UTA
, i
, ~0);
5651 * igb_intr_msi - Interrupt Handler
5652 * @irq: interrupt number
5653 * @data: pointer to a network interface device structure
5655 static irqreturn_t
igb_intr_msi(int irq
, void *data
)
5657 struct igb_adapter
*adapter
= data
;
5658 struct igb_q_vector
*q_vector
= adapter
->q_vector
[0];
5659 struct e1000_hw
*hw
= &adapter
->hw
;
5660 /* read ICR disables interrupts using IAM */
5661 u32 icr
= rd32(E1000_ICR
);
5663 igb_write_itr(q_vector
);
5665 if (icr
& E1000_ICR_DRSTA
)
5666 schedule_work(&adapter
->reset_task
);
5668 if (icr
& E1000_ICR_DOUTSYNC
) {
5669 /* HW is reporting DMA is out of sync */
5670 adapter
->stats
.doosync
++;
5673 if (icr
& (E1000_ICR_RXSEQ
| E1000_ICR_LSC
)) {
5674 hw
->mac
.get_link_status
= 1;
5675 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
5676 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
5679 #ifdef CONFIG_IGB_PTP
5680 if (icr
& E1000_ICR_TS
) {
5681 u32 tsicr
= rd32(E1000_TSICR
);
5683 if (tsicr
& E1000_TSICR_TXTS
) {
5684 /* acknowledge the interrupt */
5685 wr32(E1000_TSICR
, E1000_TSICR_TXTS
);
5686 /* retrieve hardware timestamp */
5687 schedule_work(&adapter
->ptp_tx_work
);
5690 #endif /* CONFIG_IGB_PTP */
5692 napi_schedule(&q_vector
->napi
);
5698 * igb_intr - Legacy Interrupt Handler
5699 * @irq: interrupt number
5700 * @data: pointer to a network interface device structure
5702 static irqreturn_t
igb_intr(int irq
, void *data
)
5704 struct igb_adapter
*adapter
= data
;
5705 struct igb_q_vector
*q_vector
= adapter
->q_vector
[0];
5706 struct e1000_hw
*hw
= &adapter
->hw
;
5707 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
5708 * need for the IMC write */
5709 u32 icr
= rd32(E1000_ICR
);
5711 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5712 * not set, then the adapter didn't send an interrupt */
5713 if (!(icr
& E1000_ICR_INT_ASSERTED
))
5716 igb_write_itr(q_vector
);
5718 if (icr
& E1000_ICR_DRSTA
)
5719 schedule_work(&adapter
->reset_task
);
5721 if (icr
& E1000_ICR_DOUTSYNC
) {
5722 /* HW is reporting DMA is out of sync */
5723 adapter
->stats
.doosync
++;
5726 if (icr
& (E1000_ICR_RXSEQ
| E1000_ICR_LSC
)) {
5727 hw
->mac
.get_link_status
= 1;
5728 /* guard against interrupt when we're going down */
5729 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
5730 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
5733 #ifdef CONFIG_IGB_PTP
5734 if (icr
& E1000_ICR_TS
) {
5735 u32 tsicr
= rd32(E1000_TSICR
);
5737 if (tsicr
& E1000_TSICR_TXTS
) {
5738 /* acknowledge the interrupt */
5739 wr32(E1000_TSICR
, E1000_TSICR_TXTS
);
5740 /* retrieve hardware timestamp */
5741 schedule_work(&adapter
->ptp_tx_work
);
5744 #endif /* CONFIG_IGB_PTP */
5746 napi_schedule(&q_vector
->napi
);
5751 static void igb_ring_irq_enable(struct igb_q_vector
*q_vector
)
5753 struct igb_adapter
*adapter
= q_vector
->adapter
;
5754 struct e1000_hw
*hw
= &adapter
->hw
;
5756 if ((q_vector
->rx
.ring
&& (adapter
->rx_itr_setting
& 3)) ||
5757 (!q_vector
->rx
.ring
&& (adapter
->tx_itr_setting
& 3))) {
5758 if ((adapter
->num_q_vectors
== 1) && !adapter
->vf_data
)
5759 igb_set_itr(q_vector
);
5761 igb_update_ring_itr(q_vector
);
5764 if (!test_bit(__IGB_DOWN
, &adapter
->state
)) {
5765 if (adapter
->msix_entries
)
5766 wr32(E1000_EIMS
, q_vector
->eims_value
);
5768 igb_irq_enable(adapter
);
5773 * igb_poll - NAPI Rx polling callback
5774 * @napi: napi polling structure
5775 * @budget: count of how many packets we should handle
5777 static int igb_poll(struct napi_struct
*napi
, int budget
)
5779 struct igb_q_vector
*q_vector
= container_of(napi
,
5780 struct igb_q_vector
,
5782 bool clean_complete
= true;
5784 #ifdef CONFIG_IGB_DCA
5785 if (q_vector
->adapter
->flags
& IGB_FLAG_DCA_ENABLED
)
5786 igb_update_dca(q_vector
);
5788 if (q_vector
->tx
.ring
)
5789 clean_complete
= igb_clean_tx_irq(q_vector
);
5791 if (q_vector
->rx
.ring
)
5792 clean_complete
&= igb_clean_rx_irq(q_vector
, budget
);
5794 /* If all work not completed, return budget and keep polling */
5795 if (!clean_complete
)
5798 /* If not enough Rx work done, exit the polling mode */
5799 napi_complete(napi
);
5800 igb_ring_irq_enable(q_vector
);
5806 * igb_clean_tx_irq - Reclaim resources after transmit completes
5807 * @q_vector: pointer to q_vector containing needed info
5809 * returns true if ring is completely cleaned
5811 static bool igb_clean_tx_irq(struct igb_q_vector
*q_vector
)
5813 struct igb_adapter
*adapter
= q_vector
->adapter
;
5814 struct igb_ring
*tx_ring
= q_vector
->tx
.ring
;
5815 struct igb_tx_buffer
*tx_buffer
;
5816 union e1000_adv_tx_desc
*tx_desc
, *eop_desc
;
5817 unsigned int total_bytes
= 0, total_packets
= 0;
5818 unsigned int budget
= q_vector
->tx
.work_limit
;
5819 unsigned int i
= tx_ring
->next_to_clean
;
5821 if (test_bit(__IGB_DOWN
, &adapter
->state
))
5824 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
5825 tx_desc
= IGB_TX_DESC(tx_ring
, i
);
5826 i
-= tx_ring
->count
;
5828 for (; budget
; budget
--) {
5829 eop_desc
= tx_buffer
->next_to_watch
;
5831 /* prevent any other reads prior to eop_desc */
5834 /* if next_to_watch is not set then there is no work pending */
5838 /* if DD is not set pending work has not been completed */
5839 if (!(eop_desc
->wb
.status
& cpu_to_le32(E1000_TXD_STAT_DD
)))
5842 /* clear next_to_watch to prevent false hangs */
5843 tx_buffer
->next_to_watch
= NULL
;
5845 /* update the statistics for this packet */
5846 total_bytes
+= tx_buffer
->bytecount
;
5847 total_packets
+= tx_buffer
->gso_segs
;
5850 dev_kfree_skb_any(tx_buffer
->skb
);
5851 tx_buffer
->skb
= NULL
;
5853 /* unmap skb header data */
5854 dma_unmap_single(tx_ring
->dev
,
5859 /* clear last DMA location and unmap remaining buffers */
5860 while (tx_desc
!= eop_desc
) {
5867 i
-= tx_ring
->count
;
5868 tx_buffer
= tx_ring
->tx_buffer_info
;
5869 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
5872 /* unmap any remaining paged data */
5873 if (tx_buffer
->dma
) {
5874 dma_unmap_page(tx_ring
->dev
,
5881 /* clear last DMA location */
5884 /* move us one more past the eop_desc for start of next pkt */
5889 i
-= tx_ring
->count
;
5890 tx_buffer
= tx_ring
->tx_buffer_info
;
5891 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
5895 netdev_tx_completed_queue(txring_txq(tx_ring
),
5896 total_packets
, total_bytes
);
5897 i
+= tx_ring
->count
;
5898 tx_ring
->next_to_clean
= i
;
5899 u64_stats_update_begin(&tx_ring
->tx_syncp
);
5900 tx_ring
->tx_stats
.bytes
+= total_bytes
;
5901 tx_ring
->tx_stats
.packets
+= total_packets
;
5902 u64_stats_update_end(&tx_ring
->tx_syncp
);
5903 q_vector
->tx
.total_bytes
+= total_bytes
;
5904 q_vector
->tx
.total_packets
+= total_packets
;
5906 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG
, &tx_ring
->flags
)) {
5907 struct e1000_hw
*hw
= &adapter
->hw
;
5909 eop_desc
= tx_buffer
->next_to_watch
;
5911 /* Detect a transmit hang in hardware, this serializes the
5912 * check with the clearing of time_stamp and movement of i */
5913 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG
, &tx_ring
->flags
);
5915 time_after(jiffies
, tx_buffer
->time_stamp
+
5916 (adapter
->tx_timeout_factor
* HZ
)) &&
5917 !(rd32(E1000_STATUS
) & E1000_STATUS_TXOFF
)) {
5919 /* detected Tx unit hang */
5920 dev_err(tx_ring
->dev
,
5921 "Detected Tx Unit Hang\n"
5925 " next_to_use <%x>\n"
5926 " next_to_clean <%x>\n"
5927 "buffer_info[next_to_clean]\n"
5928 " time_stamp <%lx>\n"
5929 " next_to_watch <%p>\n"
5931 " desc.status <%x>\n",
5932 tx_ring
->queue_index
,
5933 rd32(E1000_TDH(tx_ring
->reg_idx
)),
5934 readl(tx_ring
->tail
),
5935 tx_ring
->next_to_use
,
5936 tx_ring
->next_to_clean
,
5937 tx_buffer
->time_stamp
,
5940 eop_desc
->wb
.status
);
5941 netif_stop_subqueue(tx_ring
->netdev
,
5942 tx_ring
->queue_index
);
5944 /* we are about to reset, no point in enabling stuff */
5949 if (unlikely(total_packets
&&
5950 netif_carrier_ok(tx_ring
->netdev
) &&
5951 igb_desc_unused(tx_ring
) >= IGB_TX_QUEUE_WAKE
)) {
5952 /* Make sure that anybody stopping the queue after this
5953 * sees the new next_to_clean.
5956 if (__netif_subqueue_stopped(tx_ring
->netdev
,
5957 tx_ring
->queue_index
) &&
5958 !(test_bit(__IGB_DOWN
, &adapter
->state
))) {
5959 netif_wake_subqueue(tx_ring
->netdev
,
5960 tx_ring
->queue_index
);
5962 u64_stats_update_begin(&tx_ring
->tx_syncp
);
5963 tx_ring
->tx_stats
.restart_queue
++;
5964 u64_stats_update_end(&tx_ring
->tx_syncp
);
5971 static inline void igb_rx_checksum(struct igb_ring
*ring
,
5972 union e1000_adv_rx_desc
*rx_desc
,
5973 struct sk_buff
*skb
)
5975 skb_checksum_none_assert(skb
);
5977 /* Ignore Checksum bit is set */
5978 if (igb_test_staterr(rx_desc
, E1000_RXD_STAT_IXSM
))
5981 /* Rx checksum disabled via ethtool */
5982 if (!(ring
->netdev
->features
& NETIF_F_RXCSUM
))
5985 /* TCP/UDP checksum error bit is set */
5986 if (igb_test_staterr(rx_desc
,
5987 E1000_RXDEXT_STATERR_TCPE
|
5988 E1000_RXDEXT_STATERR_IPE
)) {
5990 * work around errata with sctp packets where the TCPE aka
5991 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
5992 * packets, (aka let the stack check the crc32c)
5994 if (!((skb
->len
== 60) &&
5995 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM
, &ring
->flags
))) {
5996 u64_stats_update_begin(&ring
->rx_syncp
);
5997 ring
->rx_stats
.csum_err
++;
5998 u64_stats_update_end(&ring
->rx_syncp
);
6000 /* let the stack verify checksum errors */
6003 /* It must be a TCP or UDP packet with a valid checksum */
6004 if (igb_test_staterr(rx_desc
, E1000_RXD_STAT_TCPCS
|
6005 E1000_RXD_STAT_UDPCS
))
6006 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
6008 dev_dbg(ring
->dev
, "cksum success: bits %08X\n",
6009 le32_to_cpu(rx_desc
->wb
.upper
.status_error
));
6012 static inline void igb_rx_hash(struct igb_ring
*ring
,
6013 union e1000_adv_rx_desc
*rx_desc
,
6014 struct sk_buff
*skb
)
6016 if (ring
->netdev
->features
& NETIF_F_RXHASH
)
6017 skb
->rxhash
= le32_to_cpu(rx_desc
->wb
.lower
.hi_dword
.rss
);
6020 static void igb_rx_vlan(struct igb_ring
*ring
,
6021 union e1000_adv_rx_desc
*rx_desc
,
6022 struct sk_buff
*skb
)
6024 if (igb_test_staterr(rx_desc
, E1000_RXD_STAT_VP
)) {
6026 if (igb_test_staterr(rx_desc
, E1000_RXDEXT_STATERR_LB
) &&
6027 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP
, &ring
->flags
))
6028 vid
= be16_to_cpu(rx_desc
->wb
.upper
.vlan
);
6030 vid
= le16_to_cpu(rx_desc
->wb
.upper
.vlan
);
6032 __vlan_hwaccel_put_tag(skb
, vid
);
6036 static inline u16
igb_get_hlen(union e1000_adv_rx_desc
*rx_desc
)
6038 /* HW will not DMA in data larger than the given buffer, even if it
6039 * parses the (NFS, of course) header to be larger. In that case, it
6040 * fills the header buffer and spills the rest into the page.
6042 u16 hlen
= (le16_to_cpu(rx_desc
->wb
.lower
.lo_dword
.hdr_info
) &
6043 E1000_RXDADV_HDRBUFLEN_MASK
) >> E1000_RXDADV_HDRBUFLEN_SHIFT
;
6044 if (hlen
> IGB_RX_HDR_LEN
)
6045 hlen
= IGB_RX_HDR_LEN
;
6049 static bool igb_clean_rx_irq(struct igb_q_vector
*q_vector
, int budget
)
6051 struct igb_ring
*rx_ring
= q_vector
->rx
.ring
;
6052 union e1000_adv_rx_desc
*rx_desc
;
6053 const int current_node
= numa_node_id();
6054 unsigned int total_bytes
= 0, total_packets
= 0;
6055 u16 cleaned_count
= igb_desc_unused(rx_ring
);
6056 u16 i
= rx_ring
->next_to_clean
;
6058 rx_desc
= IGB_RX_DESC(rx_ring
, i
);
6060 while (igb_test_staterr(rx_desc
, E1000_RXD_STAT_DD
)) {
6061 struct igb_rx_buffer
*buffer_info
= &rx_ring
->rx_buffer_info
[i
];
6062 struct sk_buff
*skb
= buffer_info
->skb
;
6063 union e1000_adv_rx_desc
*next_rxd
;
6065 buffer_info
->skb
= NULL
;
6066 prefetch(skb
->data
);
6069 if (i
== rx_ring
->count
)
6072 next_rxd
= IGB_RX_DESC(rx_ring
, i
);
6076 * This memory barrier is needed to keep us from reading
6077 * any other fields out of the rx_desc until we know the
6078 * RXD_STAT_DD bit is set
6082 if (!skb_is_nonlinear(skb
)) {
6083 __skb_put(skb
, igb_get_hlen(rx_desc
));
6084 dma_unmap_single(rx_ring
->dev
, buffer_info
->dma
,
6087 buffer_info
->dma
= 0;
6090 if (rx_desc
->wb
.upper
.length
) {
6091 u16 length
= le16_to_cpu(rx_desc
->wb
.upper
.length
);
6093 skb_fill_page_desc(skb
, skb_shinfo(skb
)->nr_frags
,
6095 buffer_info
->page_offset
,
6099 skb
->data_len
+= length
;
6100 skb
->truesize
+= PAGE_SIZE
/ 2;
6102 if ((page_count(buffer_info
->page
) != 1) ||
6103 (page_to_nid(buffer_info
->page
) != current_node
))
6104 buffer_info
->page
= NULL
;
6106 get_page(buffer_info
->page
);
6108 dma_unmap_page(rx_ring
->dev
, buffer_info
->page_dma
,
6109 PAGE_SIZE
/ 2, DMA_FROM_DEVICE
);
6110 buffer_info
->page_dma
= 0;
6113 if (!igb_test_staterr(rx_desc
, E1000_RXD_STAT_EOP
)) {
6114 struct igb_rx_buffer
*next_buffer
;
6115 next_buffer
= &rx_ring
->rx_buffer_info
[i
];
6116 buffer_info
->skb
= next_buffer
->skb
;
6117 buffer_info
->dma
= next_buffer
->dma
;
6118 next_buffer
->skb
= skb
;
6119 next_buffer
->dma
= 0;
6123 if (unlikely((igb_test_staterr(rx_desc
,
6124 E1000_RXDEXT_ERR_FRAME_ERR_MASK
))
6125 && !(rx_ring
->netdev
->features
& NETIF_F_RXALL
))) {
6126 dev_kfree_skb_any(skb
);
6130 #ifdef CONFIG_IGB_PTP
6131 igb_ptp_rx_hwtstamp(q_vector
, rx_desc
, skb
);
6132 #endif /* CONFIG_IGB_PTP */
6133 igb_rx_hash(rx_ring
, rx_desc
, skb
);
6134 igb_rx_checksum(rx_ring
, rx_desc
, skb
);
6135 igb_rx_vlan(rx_ring
, rx_desc
, skb
);
6137 total_bytes
+= skb
->len
;
6140 skb
->protocol
= eth_type_trans(skb
, rx_ring
->netdev
);
6142 napi_gro_receive(&q_vector
->napi
, skb
);
6150 /* return some buffers to hardware, one at a time is too slow */
6151 if (cleaned_count
>= IGB_RX_BUFFER_WRITE
) {
6152 igb_alloc_rx_buffers(rx_ring
, cleaned_count
);
6156 /* use prefetched values */
6160 rx_ring
->next_to_clean
= i
;
6161 u64_stats_update_begin(&rx_ring
->rx_syncp
);
6162 rx_ring
->rx_stats
.packets
+= total_packets
;
6163 rx_ring
->rx_stats
.bytes
+= total_bytes
;
6164 u64_stats_update_end(&rx_ring
->rx_syncp
);
6165 q_vector
->rx
.total_packets
+= total_packets
;
6166 q_vector
->rx
.total_bytes
+= total_bytes
;
6169 igb_alloc_rx_buffers(rx_ring
, cleaned_count
);
6174 static bool igb_alloc_mapped_skb(struct igb_ring
*rx_ring
,
6175 struct igb_rx_buffer
*bi
)
6177 struct sk_buff
*skb
= bi
->skb
;
6178 dma_addr_t dma
= bi
->dma
;
6184 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
6188 rx_ring
->rx_stats
.alloc_failed
++;
6192 /* initialize skb for ring */
6193 skb_record_rx_queue(skb
, rx_ring
->queue_index
);
6196 dma
= dma_map_single(rx_ring
->dev
, skb
->data
,
6197 IGB_RX_HDR_LEN
, DMA_FROM_DEVICE
);
6199 if (dma_mapping_error(rx_ring
->dev
, dma
)) {
6200 rx_ring
->rx_stats
.alloc_failed
++;
6208 static bool igb_alloc_mapped_page(struct igb_ring
*rx_ring
,
6209 struct igb_rx_buffer
*bi
)
6211 struct page
*page
= bi
->page
;
6212 dma_addr_t page_dma
= bi
->page_dma
;
6213 unsigned int page_offset
= bi
->page_offset
^ (PAGE_SIZE
/ 2);
6219 page
= __skb_alloc_page(GFP_ATOMIC
, bi
->skb
);
6221 if (unlikely(!page
)) {
6222 rx_ring
->rx_stats
.alloc_failed
++;
6227 page_dma
= dma_map_page(rx_ring
->dev
, page
,
6228 page_offset
, PAGE_SIZE
/ 2,
6231 if (dma_mapping_error(rx_ring
->dev
, page_dma
)) {
6232 rx_ring
->rx_stats
.alloc_failed
++;
6236 bi
->page_dma
= page_dma
;
6237 bi
->page_offset
= page_offset
;
6242 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
6243 * @adapter: address of board private structure
6245 void igb_alloc_rx_buffers(struct igb_ring
*rx_ring
, u16 cleaned_count
)
6247 union e1000_adv_rx_desc
*rx_desc
;
6248 struct igb_rx_buffer
*bi
;
6249 u16 i
= rx_ring
->next_to_use
;
6251 rx_desc
= IGB_RX_DESC(rx_ring
, i
);
6252 bi
= &rx_ring
->rx_buffer_info
[i
];
6253 i
-= rx_ring
->count
;
6255 while (cleaned_count
--) {
6256 if (!igb_alloc_mapped_skb(rx_ring
, bi
))
6259 /* Refresh the desc even if buffer_addrs didn't change
6260 * because each write-back erases this info. */
6261 rx_desc
->read
.hdr_addr
= cpu_to_le64(bi
->dma
);
6263 if (!igb_alloc_mapped_page(rx_ring
, bi
))
6266 rx_desc
->read
.pkt_addr
= cpu_to_le64(bi
->page_dma
);
6272 rx_desc
= IGB_RX_DESC(rx_ring
, 0);
6273 bi
= rx_ring
->rx_buffer_info
;
6274 i
-= rx_ring
->count
;
6277 /* clear the hdr_addr for the next_to_use descriptor */
6278 rx_desc
->read
.hdr_addr
= 0;
6281 i
+= rx_ring
->count
;
6283 if (rx_ring
->next_to_use
!= i
) {
6284 rx_ring
->next_to_use
= i
;
6286 /* Force memory writes to complete before letting h/w
6287 * know there are new descriptors to fetch. (Only
6288 * applicable for weak-ordered memory model archs,
6289 * such as IA-64). */
6291 writel(i
, rx_ring
->tail
);
6301 static int igb_mii_ioctl(struct net_device
*netdev
, struct ifreq
*ifr
, int cmd
)
6303 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6304 struct mii_ioctl_data
*data
= if_mii(ifr
);
6306 if (adapter
->hw
.phy
.media_type
!= e1000_media_type_copper
)
6311 data
->phy_id
= adapter
->hw
.phy
.addr
;
6314 if (igb_read_phy_reg(&adapter
->hw
, data
->reg_num
& 0x1F,
6331 static int igb_ioctl(struct net_device
*netdev
, struct ifreq
*ifr
, int cmd
)
6337 return igb_mii_ioctl(netdev
, ifr
, cmd
);
6338 #ifdef CONFIG_IGB_PTP
6340 return igb_ptp_hwtstamp_ioctl(netdev
, ifr
, cmd
);
6341 #endif /* CONFIG_IGB_PTP */
6347 s32
igb_read_pcie_cap_reg(struct e1000_hw
*hw
, u32 reg
, u16
*value
)
6349 struct igb_adapter
*adapter
= hw
->back
;
6352 cap_offset
= adapter
->pdev
->pcie_cap
;
6354 return -E1000_ERR_CONFIG
;
6356 pci_read_config_word(adapter
->pdev
, cap_offset
+ reg
, value
);
6361 s32
igb_write_pcie_cap_reg(struct e1000_hw
*hw
, u32 reg
, u16
*value
)
6363 struct igb_adapter
*adapter
= hw
->back
;
6366 cap_offset
= adapter
->pdev
->pcie_cap
;
6368 return -E1000_ERR_CONFIG
;
6370 pci_write_config_word(adapter
->pdev
, cap_offset
+ reg
, *value
);
6375 static void igb_vlan_mode(struct net_device
*netdev
, netdev_features_t features
)
6377 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6378 struct e1000_hw
*hw
= &adapter
->hw
;
6380 bool enable
= !!(features
& NETIF_F_HW_VLAN_RX
);
6383 /* enable VLAN tag insert/strip */
6384 ctrl
= rd32(E1000_CTRL
);
6385 ctrl
|= E1000_CTRL_VME
;
6386 wr32(E1000_CTRL
, ctrl
);
6388 /* Disable CFI check */
6389 rctl
= rd32(E1000_RCTL
);
6390 rctl
&= ~E1000_RCTL_CFIEN
;
6391 wr32(E1000_RCTL
, rctl
);
6393 /* disable VLAN tag insert/strip */
6394 ctrl
= rd32(E1000_CTRL
);
6395 ctrl
&= ~E1000_CTRL_VME
;
6396 wr32(E1000_CTRL
, ctrl
);
6399 igb_rlpml_set(adapter
);
6402 static int igb_vlan_rx_add_vid(struct net_device
*netdev
, u16 vid
)
6404 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6405 struct e1000_hw
*hw
= &adapter
->hw
;
6406 int pf_id
= adapter
->vfs_allocated_count
;
6408 /* attempt to add filter to vlvf array */
6409 igb_vlvf_set(adapter
, vid
, true, pf_id
);
6411 /* add the filter since PF can receive vlans w/o entry in vlvf */
6412 igb_vfta_set(hw
, vid
, true);
6414 set_bit(vid
, adapter
->active_vlans
);
6419 static int igb_vlan_rx_kill_vid(struct net_device
*netdev
, u16 vid
)
6421 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6422 struct e1000_hw
*hw
= &adapter
->hw
;
6423 int pf_id
= adapter
->vfs_allocated_count
;
6426 /* remove vlan from VLVF table array */
6427 err
= igb_vlvf_set(adapter
, vid
, false, pf_id
);
6429 /* if vid was not present in VLVF just remove it from table */
6431 igb_vfta_set(hw
, vid
, false);
6433 clear_bit(vid
, adapter
->active_vlans
);
6438 static void igb_restore_vlan(struct igb_adapter
*adapter
)
6442 igb_vlan_mode(adapter
->netdev
, adapter
->netdev
->features
);
6444 for_each_set_bit(vid
, adapter
->active_vlans
, VLAN_N_VID
)
6445 igb_vlan_rx_add_vid(adapter
->netdev
, vid
);
6448 int igb_set_spd_dplx(struct igb_adapter
*adapter
, u32 spd
, u8 dplx
)
6450 struct pci_dev
*pdev
= adapter
->pdev
;
6451 struct e1000_mac_info
*mac
= &adapter
->hw
.mac
;
6455 /* Make sure dplx is at most 1 bit and lsb of speed is not set
6456 * for the switch() below to work */
6457 if ((spd
& 1) || (dplx
& ~1))
6460 /* Fiber NIC's only allow 1000 Gbps Full duplex */
6461 if ((adapter
->hw
.phy
.media_type
== e1000_media_type_internal_serdes
) &&
6462 spd
!= SPEED_1000
&&
6463 dplx
!= DUPLEX_FULL
)
6466 switch (spd
+ dplx
) {
6467 case SPEED_10
+ DUPLEX_HALF
:
6468 mac
->forced_speed_duplex
= ADVERTISE_10_HALF
;
6470 case SPEED_10
+ DUPLEX_FULL
:
6471 mac
->forced_speed_duplex
= ADVERTISE_10_FULL
;
6473 case SPEED_100
+ DUPLEX_HALF
:
6474 mac
->forced_speed_duplex
= ADVERTISE_100_HALF
;
6476 case SPEED_100
+ DUPLEX_FULL
:
6477 mac
->forced_speed_duplex
= ADVERTISE_100_FULL
;
6479 case SPEED_1000
+ DUPLEX_FULL
:
6481 adapter
->hw
.phy
.autoneg_advertised
= ADVERTISE_1000_FULL
;
6483 case SPEED_1000
+ DUPLEX_HALF
: /* not supported */
6488 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
6489 adapter
->hw
.phy
.mdix
= AUTO_ALL_MODES
;
6494 dev_err(&pdev
->dev
, "Unsupported Speed/Duplex configuration\n");
6498 static int __igb_shutdown(struct pci_dev
*pdev
, bool *enable_wake
,
6501 struct net_device
*netdev
= pci_get_drvdata(pdev
);
6502 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6503 struct e1000_hw
*hw
= &adapter
->hw
;
6504 u32 ctrl
, rctl
, status
;
6505 u32 wufc
= runtime
? E1000_WUFC_LNKC
: adapter
->wol
;
6510 netif_device_detach(netdev
);
6512 if (netif_running(netdev
))
6513 __igb_close(netdev
, true);
6515 igb_clear_interrupt_scheme(adapter
);
6518 retval
= pci_save_state(pdev
);
6523 status
= rd32(E1000_STATUS
);
6524 if (status
& E1000_STATUS_LU
)
6525 wufc
&= ~E1000_WUFC_LNKC
;
6528 igb_setup_rctl(adapter
);
6529 igb_set_rx_mode(netdev
);
6531 /* turn on all-multi mode if wake on multicast is enabled */
6532 if (wufc
& E1000_WUFC_MC
) {
6533 rctl
= rd32(E1000_RCTL
);
6534 rctl
|= E1000_RCTL_MPE
;
6535 wr32(E1000_RCTL
, rctl
);
6538 ctrl
= rd32(E1000_CTRL
);
6539 /* advertise wake from D3Cold */
6540 #define E1000_CTRL_ADVD3WUC 0x00100000
6541 /* phy power management enable */
6542 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
6543 ctrl
|= E1000_CTRL_ADVD3WUC
;
6544 wr32(E1000_CTRL
, ctrl
);
6546 /* Allow time for pending master requests to run */
6547 igb_disable_pcie_master(hw
);
6549 wr32(E1000_WUC
, E1000_WUC_PME_EN
);
6550 wr32(E1000_WUFC
, wufc
);
6553 wr32(E1000_WUFC
, 0);
6556 *enable_wake
= wufc
|| adapter
->en_mng_pt
;
6558 igb_power_down_link(adapter
);
6560 igb_power_up_link(adapter
);
6562 /* Release control of h/w to f/w. If f/w is AMT enabled, this
6563 * would have already happened in close and is redundant. */
6564 igb_release_hw_control(adapter
);
6566 pci_disable_device(pdev
);
6572 #ifdef CONFIG_PM_SLEEP
6573 static int igb_suspend(struct device
*dev
)
6577 struct pci_dev
*pdev
= to_pci_dev(dev
);
6579 retval
= __igb_shutdown(pdev
, &wake
, 0);
6584 pci_prepare_to_sleep(pdev
);
6586 pci_wake_from_d3(pdev
, false);
6587 pci_set_power_state(pdev
, PCI_D3hot
);
6592 #endif /* CONFIG_PM_SLEEP */
6594 static int igb_resume(struct device
*dev
)
6596 struct pci_dev
*pdev
= to_pci_dev(dev
);
6597 struct net_device
*netdev
= pci_get_drvdata(pdev
);
6598 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6599 struct e1000_hw
*hw
= &adapter
->hw
;
6602 pci_set_power_state(pdev
, PCI_D0
);
6603 pci_restore_state(pdev
);
6604 pci_save_state(pdev
);
6606 err
= pci_enable_device_mem(pdev
);
6609 "igb: Cannot enable PCI device from suspend\n");
6612 pci_set_master(pdev
);
6614 pci_enable_wake(pdev
, PCI_D3hot
, 0);
6615 pci_enable_wake(pdev
, PCI_D3cold
, 0);
6617 if (igb_init_interrupt_scheme(adapter
)) {
6618 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
6624 /* let the f/w know that the h/w is now under the control of the
6626 igb_get_hw_control(adapter
);
6628 wr32(E1000_WUS
, ~0);
6630 if (netdev
->flags
& IFF_UP
) {
6631 err
= __igb_open(netdev
, true);
6636 netif_device_attach(netdev
);
6640 #ifdef CONFIG_PM_RUNTIME
6641 static int igb_runtime_idle(struct device
*dev
)
6643 struct pci_dev
*pdev
= to_pci_dev(dev
);
6644 struct net_device
*netdev
= pci_get_drvdata(pdev
);
6645 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6647 if (!igb_has_link(adapter
))
6648 pm_schedule_suspend(dev
, MSEC_PER_SEC
* 5);
6653 static int igb_runtime_suspend(struct device
*dev
)
6655 struct pci_dev
*pdev
= to_pci_dev(dev
);
6659 retval
= __igb_shutdown(pdev
, &wake
, 1);
6664 pci_prepare_to_sleep(pdev
);
6666 pci_wake_from_d3(pdev
, false);
6667 pci_set_power_state(pdev
, PCI_D3hot
);
6673 static int igb_runtime_resume(struct device
*dev
)
6675 return igb_resume(dev
);
6677 #endif /* CONFIG_PM_RUNTIME */
6680 static void igb_shutdown(struct pci_dev
*pdev
)
6684 __igb_shutdown(pdev
, &wake
, 0);
6686 if (system_state
== SYSTEM_POWER_OFF
) {
6687 pci_wake_from_d3(pdev
, wake
);
6688 pci_set_power_state(pdev
, PCI_D3hot
);
6692 #ifdef CONFIG_NET_POLL_CONTROLLER
6694 * Polling 'interrupt' - used by things like netconsole to send skbs
6695 * without having to re-enable interrupts. It's not called while
6696 * the interrupt routine is executing.
6698 static void igb_netpoll(struct net_device
*netdev
)
6700 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6701 struct e1000_hw
*hw
= &adapter
->hw
;
6702 struct igb_q_vector
*q_vector
;
6705 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
6706 q_vector
= adapter
->q_vector
[i
];
6707 if (adapter
->msix_entries
)
6708 wr32(E1000_EIMC
, q_vector
->eims_value
);
6710 igb_irq_disable(adapter
);
6711 napi_schedule(&q_vector
->napi
);
6714 #endif /* CONFIG_NET_POLL_CONTROLLER */
6717 * igb_io_error_detected - called when PCI error is detected
6718 * @pdev: Pointer to PCI device
6719 * @state: The current pci connection state
6721 * This function is called after a PCI bus error affecting
6722 * this device has been detected.
6724 static pci_ers_result_t
igb_io_error_detected(struct pci_dev
*pdev
,
6725 pci_channel_state_t state
)
6727 struct net_device
*netdev
= pci_get_drvdata(pdev
);
6728 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6730 netif_device_detach(netdev
);
6732 if (state
== pci_channel_io_perm_failure
)
6733 return PCI_ERS_RESULT_DISCONNECT
;
6735 if (netif_running(netdev
))
6737 pci_disable_device(pdev
);
6739 /* Request a slot slot reset. */
6740 return PCI_ERS_RESULT_NEED_RESET
;
6744 * igb_io_slot_reset - called after the pci bus has been reset.
6745 * @pdev: Pointer to PCI device
6747 * Restart the card from scratch, as if from a cold-boot. Implementation
6748 * resembles the first-half of the igb_resume routine.
6750 static pci_ers_result_t
igb_io_slot_reset(struct pci_dev
*pdev
)
6752 struct net_device
*netdev
= pci_get_drvdata(pdev
);
6753 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6754 struct e1000_hw
*hw
= &adapter
->hw
;
6755 pci_ers_result_t result
;
6758 if (pci_enable_device_mem(pdev
)) {
6760 "Cannot re-enable PCI device after reset.\n");
6761 result
= PCI_ERS_RESULT_DISCONNECT
;
6763 pci_set_master(pdev
);
6764 pci_restore_state(pdev
);
6765 pci_save_state(pdev
);
6767 pci_enable_wake(pdev
, PCI_D3hot
, 0);
6768 pci_enable_wake(pdev
, PCI_D3cold
, 0);
6771 wr32(E1000_WUS
, ~0);
6772 result
= PCI_ERS_RESULT_RECOVERED
;
6775 err
= pci_cleanup_aer_uncorrect_error_status(pdev
);
6777 dev_err(&pdev
->dev
, "pci_cleanup_aer_uncorrect_error_status "
6778 "failed 0x%0x\n", err
);
6779 /* non-fatal, continue */
6786 * igb_io_resume - called when traffic can start flowing again.
6787 * @pdev: Pointer to PCI device
6789 * This callback is called when the error recovery driver tells us that
6790 * its OK to resume normal operation. Implementation resembles the
6791 * second-half of the igb_resume routine.
6793 static void igb_io_resume(struct pci_dev
*pdev
)
6795 struct net_device
*netdev
= pci_get_drvdata(pdev
);
6796 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6798 if (netif_running(netdev
)) {
6799 if (igb_up(adapter
)) {
6800 dev_err(&pdev
->dev
, "igb_up failed after reset\n");
6805 netif_device_attach(netdev
);
6807 /* let the f/w know that the h/w is now under the control of the
6809 igb_get_hw_control(adapter
);
6812 static void igb_rar_set_qsel(struct igb_adapter
*adapter
, u8
*addr
, u32 index
,
6815 u32 rar_low
, rar_high
;
6816 struct e1000_hw
*hw
= &adapter
->hw
;
6818 /* HW expects these in little endian so we reverse the byte order
6819 * from network order (big endian) to little endian
6821 rar_low
= ((u32
) addr
[0] | ((u32
) addr
[1] << 8) |
6822 ((u32
) addr
[2] << 16) | ((u32
) addr
[3] << 24));
6823 rar_high
= ((u32
) addr
[4] | ((u32
) addr
[5] << 8));
6825 /* Indicate to hardware the Address is Valid. */
6826 rar_high
|= E1000_RAH_AV
;
6828 if (hw
->mac
.type
== e1000_82575
)
6829 rar_high
|= E1000_RAH_POOL_1
* qsel
;
6831 rar_high
|= E1000_RAH_POOL_1
<< qsel
;
6833 wr32(E1000_RAL(index
), rar_low
);
6835 wr32(E1000_RAH(index
), rar_high
);
6839 static int igb_set_vf_mac(struct igb_adapter
*adapter
,
6840 int vf
, unsigned char *mac_addr
)
6842 struct e1000_hw
*hw
= &adapter
->hw
;
6843 /* VF MAC addresses start at end of receive addresses and moves
6844 * torwards the first, as a result a collision should not be possible */
6845 int rar_entry
= hw
->mac
.rar_entry_count
- (vf
+ 1);
6847 memcpy(adapter
->vf_data
[vf
].vf_mac_addresses
, mac_addr
, ETH_ALEN
);
6849 igb_rar_set_qsel(adapter
, mac_addr
, rar_entry
, vf
);
6854 static int igb_ndo_set_vf_mac(struct net_device
*netdev
, int vf
, u8
*mac
)
6856 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6857 if (!is_valid_ether_addr(mac
) || (vf
>= adapter
->vfs_allocated_count
))
6859 adapter
->vf_data
[vf
].flags
|= IGB_VF_FLAG_PF_SET_MAC
;
6860 dev_info(&adapter
->pdev
->dev
, "setting MAC %pM on VF %d\n", mac
, vf
);
6861 dev_info(&adapter
->pdev
->dev
, "Reload the VF driver to make this"
6862 " change effective.");
6863 if (test_bit(__IGB_DOWN
, &adapter
->state
)) {
6864 dev_warn(&adapter
->pdev
->dev
, "The VF MAC address has been set,"
6865 " but the PF device is not up.\n");
6866 dev_warn(&adapter
->pdev
->dev
, "Bring the PF device up before"
6867 " attempting to use the VF device.\n");
6869 return igb_set_vf_mac(adapter
, vf
, mac
);
6872 static int igb_link_mbps(int internal_link_speed
)
6874 switch (internal_link_speed
) {
6884 static void igb_set_vf_rate_limit(struct e1000_hw
*hw
, int vf
, int tx_rate
,
6891 /* Calculate the rate factor values to set */
6892 rf_int
= link_speed
/ tx_rate
;
6893 rf_dec
= (link_speed
- (rf_int
* tx_rate
));
6894 rf_dec
= (rf_dec
* (1<<E1000_RTTBCNRC_RF_INT_SHIFT
)) / tx_rate
;
6896 bcnrc_val
= E1000_RTTBCNRC_RS_ENA
;
6897 bcnrc_val
|= ((rf_int
<<E1000_RTTBCNRC_RF_INT_SHIFT
) &
6898 E1000_RTTBCNRC_RF_INT_MASK
);
6899 bcnrc_val
|= (rf_dec
& E1000_RTTBCNRC_RF_DEC_MASK
);
6904 wr32(E1000_RTTDQSEL
, vf
); /* vf X uses queue X */
6906 * Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
6907 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
6909 wr32(E1000_RTTBCNRM
, 0x14);
6910 wr32(E1000_RTTBCNRC
, bcnrc_val
);
6913 static void igb_check_vf_rate_limit(struct igb_adapter
*adapter
)
6915 int actual_link_speed
, i
;
6916 bool reset_rate
= false;
6918 /* VF TX rate limit was not set or not supported */
6919 if ((adapter
->vf_rate_link_speed
== 0) ||
6920 (adapter
->hw
.mac
.type
!= e1000_82576
))
6923 actual_link_speed
= igb_link_mbps(adapter
->link_speed
);
6924 if (actual_link_speed
!= adapter
->vf_rate_link_speed
) {
6926 adapter
->vf_rate_link_speed
= 0;
6927 dev_info(&adapter
->pdev
->dev
,
6928 "Link speed has been changed. VF Transmit "
6929 "rate is disabled\n");
6932 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++) {
6934 adapter
->vf_data
[i
].tx_rate
= 0;
6936 igb_set_vf_rate_limit(&adapter
->hw
, i
,
6937 adapter
->vf_data
[i
].tx_rate
,
6942 static int igb_ndo_set_vf_bw(struct net_device
*netdev
, int vf
, int tx_rate
)
6944 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6945 struct e1000_hw
*hw
= &adapter
->hw
;
6946 int actual_link_speed
;
6948 if (hw
->mac
.type
!= e1000_82576
)
6951 actual_link_speed
= igb_link_mbps(adapter
->link_speed
);
6952 if ((vf
>= adapter
->vfs_allocated_count
) ||
6953 (!(rd32(E1000_STATUS
) & E1000_STATUS_LU
)) ||
6954 (tx_rate
< 0) || (tx_rate
> actual_link_speed
))
6957 adapter
->vf_rate_link_speed
= actual_link_speed
;
6958 adapter
->vf_data
[vf
].tx_rate
= (u16
)tx_rate
;
6959 igb_set_vf_rate_limit(hw
, vf
, tx_rate
, actual_link_speed
);
6964 static int igb_ndo_get_vf_config(struct net_device
*netdev
,
6965 int vf
, struct ifla_vf_info
*ivi
)
6967 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6968 if (vf
>= adapter
->vfs_allocated_count
)
6971 memcpy(&ivi
->mac
, adapter
->vf_data
[vf
].vf_mac_addresses
, ETH_ALEN
);
6972 ivi
->tx_rate
= adapter
->vf_data
[vf
].tx_rate
;
6973 ivi
->vlan
= adapter
->vf_data
[vf
].pf_vlan
;
6974 ivi
->qos
= adapter
->vf_data
[vf
].pf_qos
;
6978 static void igb_vmm_control(struct igb_adapter
*adapter
)
6980 struct e1000_hw
*hw
= &adapter
->hw
;
6983 switch (hw
->mac
.type
) {
6988 /* replication is not supported for 82575 */
6991 /* notify HW that the MAC is adding vlan tags */
6992 reg
= rd32(E1000_DTXCTL
);
6993 reg
|= E1000_DTXCTL_VLAN_ADDED
;
6994 wr32(E1000_DTXCTL
, reg
);
6996 /* enable replication vlan tag stripping */
6997 reg
= rd32(E1000_RPLOLR
);
6998 reg
|= E1000_RPLOLR_STRVLAN
;
6999 wr32(E1000_RPLOLR
, reg
);
7001 /* none of the above registers are supported by i350 */
7005 if (adapter
->vfs_allocated_count
) {
7006 igb_vmdq_set_loopback_pf(hw
, true);
7007 igb_vmdq_set_replication_pf(hw
, true);
7008 igb_vmdq_set_anti_spoofing_pf(hw
, true,
7009 adapter
->vfs_allocated_count
);
7011 igb_vmdq_set_loopback_pf(hw
, false);
7012 igb_vmdq_set_replication_pf(hw
, false);
7016 static void igb_init_dmac(struct igb_adapter
*adapter
, u32 pba
)
7018 struct e1000_hw
*hw
= &adapter
->hw
;
7022 if (hw
->mac
.type
> e1000_82580
) {
7023 if (adapter
->flags
& IGB_FLAG_DMAC
) {
7026 /* force threshold to 0. */
7027 wr32(E1000_DMCTXTH
, 0);
7030 * DMA Coalescing high water mark needs to be greater
7031 * than the Rx threshold. Set hwm to PBA - max frame
7032 * size in 16B units, capping it at PBA - 6KB.
7034 hwm
= 64 * pba
- adapter
->max_frame_size
/ 16;
7035 if (hwm
< 64 * (pba
- 6))
7036 hwm
= 64 * (pba
- 6);
7037 reg
= rd32(E1000_FCRTC
);
7038 reg
&= ~E1000_FCRTC_RTH_COAL_MASK
;
7039 reg
|= ((hwm
<< E1000_FCRTC_RTH_COAL_SHIFT
)
7040 & E1000_FCRTC_RTH_COAL_MASK
);
7041 wr32(E1000_FCRTC
, reg
);
7044 * Set the DMA Coalescing Rx threshold to PBA - 2 * max
7045 * frame size, capping it at PBA - 10KB.
7047 dmac_thr
= pba
- adapter
->max_frame_size
/ 512;
7048 if (dmac_thr
< pba
- 10)
7049 dmac_thr
= pba
- 10;
7050 reg
= rd32(E1000_DMACR
);
7051 reg
&= ~E1000_DMACR_DMACTHR_MASK
;
7052 reg
|= ((dmac_thr
<< E1000_DMACR_DMACTHR_SHIFT
)
7053 & E1000_DMACR_DMACTHR_MASK
);
7055 /* transition to L0x or L1 if available..*/
7056 reg
|= (E1000_DMACR_DMAC_EN
| E1000_DMACR_DMAC_LX_MASK
);
7058 /* watchdog timer= +-1000 usec in 32usec intervals */
7061 /* Disable BMC-to-OS Watchdog Enable */
7062 reg
&= ~E1000_DMACR_DC_BMC2OSW_EN
;
7063 wr32(E1000_DMACR
, reg
);
7066 * no lower threshold to disable
7067 * coalescing(smart fifb)-UTRESH=0
7069 wr32(E1000_DMCRTRH
, 0);
7071 reg
= (IGB_DMCTLX_DCFLUSH_DIS
| 0x4);
7073 wr32(E1000_DMCTLX
, reg
);
7076 * free space in tx packet buffer to wake from
7079 wr32(E1000_DMCTXTH
, (IGB_MIN_TXPBSIZE
-
7080 (IGB_TX_BUF_4096
+ adapter
->max_frame_size
)) >> 6);
7083 * make low power state decision controlled
7086 reg
= rd32(E1000_PCIEMISC
);
7087 reg
&= ~E1000_PCIEMISC_LX_DECISION
;
7088 wr32(E1000_PCIEMISC
, reg
);
7089 } /* endif adapter->dmac is not disabled */
7090 } else if (hw
->mac
.type
== e1000_82580
) {
7091 u32 reg
= rd32(E1000_PCIEMISC
);
7092 wr32(E1000_PCIEMISC
, reg
& ~E1000_PCIEMISC_LX_DECISION
);
7093 wr32(E1000_DMACR
, 0);