1 /*******************************************************************************
3 Intel(R) Gigabit Ethernet Linux driver
4 Copyright(c) 2007-2013 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>
60 #include <linux/i2c.h>
66 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
67 __stringify(BUILD) "-k"
68 char igb_driver_name
[] = "igb";
69 char igb_driver_version
[] = DRV_VERSION
;
70 static const char igb_driver_string
[] =
71 "Intel(R) Gigabit Ethernet Network Driver";
72 static const char igb_copyright
[] =
73 "Copyright (c) 2007-2013 Intel Corporation.";
75 static const struct e1000_info
*igb_info_tbl
[] = {
76 [board_82575
] = &e1000_82575_info
,
79 static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl
) = {
80 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I354_BACKPLANE_1GBPS
) },
81 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I354_SGMII
) },
82 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS
) },
83 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I211_COPPER
), board_82575
},
84 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_COPPER
), board_82575
},
85 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_FIBER
), board_82575
},
86 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_SERDES
), board_82575
},
87 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_SGMII
), board_82575
},
88 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_COPPER_FLASHLESS
), board_82575
},
89 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I210_SERDES_FLASHLESS
), board_82575
},
90 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I350_COPPER
), board_82575
},
91 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I350_FIBER
), board_82575
},
92 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I350_SERDES
), board_82575
},
93 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_I350_SGMII
), board_82575
},
94 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_COPPER
), board_82575
},
95 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_FIBER
), board_82575
},
96 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_QUAD_FIBER
), board_82575
},
97 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_SERDES
), board_82575
},
98 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_SGMII
), board_82575
},
99 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82580_COPPER_DUAL
), board_82575
},
100 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_DH89XXCC_SGMII
), board_82575
},
101 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_DH89XXCC_SERDES
), board_82575
},
102 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_DH89XXCC_BACKPLANE
), board_82575
},
103 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_DH89XXCC_SFP
), board_82575
},
104 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576
), board_82575
},
105 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_NS
), board_82575
},
106 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_NS_SERDES
), board_82575
},
107 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_FIBER
), board_82575
},
108 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_SERDES
), board_82575
},
109 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_SERDES_QUAD
), board_82575
},
110 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_QUAD_COPPER_ET2
), board_82575
},
111 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82576_QUAD_COPPER
), board_82575
},
112 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82575EB_COPPER
), board_82575
},
113 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82575EB_FIBER_SERDES
), board_82575
},
114 { PCI_VDEVICE(INTEL
, E1000_DEV_ID_82575GB_QUAD_COPPER
), board_82575
},
115 /* required last entry */
119 MODULE_DEVICE_TABLE(pci
, igb_pci_tbl
);
121 void igb_reset(struct igb_adapter
*);
122 static int igb_setup_all_tx_resources(struct igb_adapter
*);
123 static int igb_setup_all_rx_resources(struct igb_adapter
*);
124 static void igb_free_all_tx_resources(struct igb_adapter
*);
125 static void igb_free_all_rx_resources(struct igb_adapter
*);
126 static void igb_setup_mrqc(struct igb_adapter
*);
127 static int igb_probe(struct pci_dev
*, const struct pci_device_id
*);
128 static void igb_remove(struct pci_dev
*pdev
);
129 static int igb_sw_init(struct igb_adapter
*);
130 static int igb_open(struct net_device
*);
131 static int igb_close(struct net_device
*);
132 static void igb_configure(struct igb_adapter
*);
133 static void igb_configure_tx(struct igb_adapter
*);
134 static void igb_configure_rx(struct igb_adapter
*);
135 static void igb_clean_all_tx_rings(struct igb_adapter
*);
136 static void igb_clean_all_rx_rings(struct igb_adapter
*);
137 static void igb_clean_tx_ring(struct igb_ring
*);
138 static void igb_clean_rx_ring(struct igb_ring
*);
139 static void igb_set_rx_mode(struct net_device
*);
140 static void igb_update_phy_info(unsigned long);
141 static void igb_watchdog(unsigned long);
142 static void igb_watchdog_task(struct work_struct
*);
143 static netdev_tx_t
igb_xmit_frame(struct sk_buff
*skb
, struct net_device
*);
144 static struct rtnl_link_stats64
*igb_get_stats64(struct net_device
*dev
,
145 struct rtnl_link_stats64
*stats
);
146 static int igb_change_mtu(struct net_device
*, int);
147 static int igb_set_mac(struct net_device
*, void *);
148 static void igb_set_uta(struct igb_adapter
*adapter
);
149 static irqreturn_t
igb_intr(int irq
, void *);
150 static irqreturn_t
igb_intr_msi(int irq
, void *);
151 static irqreturn_t
igb_msix_other(int irq
, void *);
152 static irqreturn_t
igb_msix_ring(int irq
, void *);
153 #ifdef CONFIG_IGB_DCA
154 static void igb_update_dca(struct igb_q_vector
*);
155 static void igb_setup_dca(struct igb_adapter
*);
156 #endif /* CONFIG_IGB_DCA */
157 static int igb_poll(struct napi_struct
*, int);
158 static bool igb_clean_tx_irq(struct igb_q_vector
*);
159 static bool igb_clean_rx_irq(struct igb_q_vector
*, int);
160 static int igb_ioctl(struct net_device
*, struct ifreq
*, int cmd
);
161 static void igb_tx_timeout(struct net_device
*);
162 static void igb_reset_task(struct work_struct
*);
163 static void igb_vlan_mode(struct net_device
*netdev
, netdev_features_t features
);
164 static int igb_vlan_rx_add_vid(struct net_device
*, __be16
, u16
);
165 static int igb_vlan_rx_kill_vid(struct net_device
*, __be16
, u16
);
166 static void igb_restore_vlan(struct igb_adapter
*);
167 static void igb_rar_set_qsel(struct igb_adapter
*, u8
*, u32
, u8
);
168 static void igb_ping_all_vfs(struct igb_adapter
*);
169 static void igb_msg_task(struct igb_adapter
*);
170 static void igb_vmm_control(struct igb_adapter
*);
171 static int igb_set_vf_mac(struct igb_adapter
*, int, unsigned char *);
172 static void igb_restore_vf_multicasts(struct igb_adapter
*adapter
);
173 static int igb_ndo_set_vf_mac(struct net_device
*netdev
, int vf
, u8
*mac
);
174 static int igb_ndo_set_vf_vlan(struct net_device
*netdev
,
175 int vf
, u16 vlan
, u8 qos
);
176 static int igb_ndo_set_vf_bw(struct net_device
*netdev
, int vf
, int tx_rate
);
177 static int igb_ndo_set_vf_spoofchk(struct net_device
*netdev
, int vf
,
179 static int igb_ndo_get_vf_config(struct net_device
*netdev
, int vf
,
180 struct ifla_vf_info
*ivi
);
181 static void igb_check_vf_rate_limit(struct igb_adapter
*);
183 #ifdef CONFIG_PCI_IOV
184 static int igb_vf_configure(struct igb_adapter
*adapter
, int vf
);
185 static int igb_pci_enable_sriov(struct pci_dev
*dev
, int num_vfs
);
189 #ifdef CONFIG_PM_SLEEP
190 static int igb_suspend(struct device
*);
192 static int igb_resume(struct device
*);
193 #ifdef CONFIG_PM_RUNTIME
194 static int igb_runtime_suspend(struct device
*dev
);
195 static int igb_runtime_resume(struct device
*dev
);
196 static int igb_runtime_idle(struct device
*dev
);
198 static const struct dev_pm_ops igb_pm_ops
= {
199 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend
, igb_resume
)
200 SET_RUNTIME_PM_OPS(igb_runtime_suspend
, igb_runtime_resume
,
204 static void igb_shutdown(struct pci_dev
*);
205 static int igb_pci_sriov_configure(struct pci_dev
*dev
, int num_vfs
);
206 #ifdef CONFIG_IGB_DCA
207 static int igb_notify_dca(struct notifier_block
*, unsigned long, void *);
208 static struct notifier_block dca_notifier
= {
209 .notifier_call
= igb_notify_dca
,
214 #ifdef CONFIG_NET_POLL_CONTROLLER
215 /* for netdump / net console */
216 static void igb_netpoll(struct net_device
*);
218 #ifdef CONFIG_PCI_IOV
219 static unsigned int max_vfs
= 0;
220 module_param(max_vfs
, uint
, 0);
221 MODULE_PARM_DESC(max_vfs
, "Maximum number of virtual functions to allocate "
222 "per physical function");
223 #endif /* CONFIG_PCI_IOV */
225 static pci_ers_result_t
igb_io_error_detected(struct pci_dev
*,
226 pci_channel_state_t
);
227 static pci_ers_result_t
igb_io_slot_reset(struct pci_dev
*);
228 static void igb_io_resume(struct pci_dev
*);
230 static const struct pci_error_handlers igb_err_handler
= {
231 .error_detected
= igb_io_error_detected
,
232 .slot_reset
= igb_io_slot_reset
,
233 .resume
= igb_io_resume
,
236 static void igb_init_dmac(struct igb_adapter
*adapter
, u32 pba
);
238 static struct pci_driver igb_driver
= {
239 .name
= igb_driver_name
,
240 .id_table
= igb_pci_tbl
,
242 .remove
= igb_remove
,
244 .driver
.pm
= &igb_pm_ops
,
246 .shutdown
= igb_shutdown
,
247 .sriov_configure
= igb_pci_sriov_configure
,
248 .err_handler
= &igb_err_handler
251 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
252 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
253 MODULE_LICENSE("GPL");
254 MODULE_VERSION(DRV_VERSION
);
256 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
257 static int debug
= -1;
258 module_param(debug
, int, 0);
259 MODULE_PARM_DESC(debug
, "Debug level (0=none,...,16=all)");
261 struct igb_reg_info
{
266 static const struct igb_reg_info igb_reg_info_tbl
[] = {
268 /* General Registers */
269 {E1000_CTRL
, "CTRL"},
270 {E1000_STATUS
, "STATUS"},
271 {E1000_CTRL_EXT
, "CTRL_EXT"},
273 /* Interrupt Registers */
277 {E1000_RCTL
, "RCTL"},
278 {E1000_RDLEN(0), "RDLEN"},
279 {E1000_RDH(0), "RDH"},
280 {E1000_RDT(0), "RDT"},
281 {E1000_RXDCTL(0), "RXDCTL"},
282 {E1000_RDBAL(0), "RDBAL"},
283 {E1000_RDBAH(0), "RDBAH"},
286 {E1000_TCTL
, "TCTL"},
287 {E1000_TDBAL(0), "TDBAL"},
288 {E1000_TDBAH(0), "TDBAH"},
289 {E1000_TDLEN(0), "TDLEN"},
290 {E1000_TDH(0), "TDH"},
291 {E1000_TDT(0), "TDT"},
292 {E1000_TXDCTL(0), "TXDCTL"},
293 {E1000_TDFH
, "TDFH"},
294 {E1000_TDFT
, "TDFT"},
295 {E1000_TDFHS
, "TDFHS"},
296 {E1000_TDFPC
, "TDFPC"},
298 /* List Terminator */
302 /* igb_regdump - register printout routine */
303 static void igb_regdump(struct e1000_hw
*hw
, struct igb_reg_info
*reginfo
)
309 switch (reginfo
->ofs
) {
311 for (n
= 0; n
< 4; n
++)
312 regs
[n
] = rd32(E1000_RDLEN(n
));
315 for (n
= 0; n
< 4; n
++)
316 regs
[n
] = rd32(E1000_RDH(n
));
319 for (n
= 0; n
< 4; n
++)
320 regs
[n
] = rd32(E1000_RDT(n
));
322 case E1000_RXDCTL(0):
323 for (n
= 0; n
< 4; n
++)
324 regs
[n
] = rd32(E1000_RXDCTL(n
));
327 for (n
= 0; n
< 4; n
++)
328 regs
[n
] = rd32(E1000_RDBAL(n
));
331 for (n
= 0; n
< 4; n
++)
332 regs
[n
] = rd32(E1000_RDBAH(n
));
335 for (n
= 0; n
< 4; n
++)
336 regs
[n
] = rd32(E1000_RDBAL(n
));
339 for (n
= 0; n
< 4; n
++)
340 regs
[n
] = rd32(E1000_TDBAH(n
));
343 for (n
= 0; n
< 4; n
++)
344 regs
[n
] = rd32(E1000_TDLEN(n
));
347 for (n
= 0; n
< 4; n
++)
348 regs
[n
] = rd32(E1000_TDH(n
));
351 for (n
= 0; n
< 4; n
++)
352 regs
[n
] = rd32(E1000_TDT(n
));
354 case E1000_TXDCTL(0):
355 for (n
= 0; n
< 4; n
++)
356 regs
[n
] = rd32(E1000_TXDCTL(n
));
359 pr_info("%-15s %08x\n", reginfo
->name
, rd32(reginfo
->ofs
));
363 snprintf(rname
, 16, "%s%s", reginfo
->name
, "[0-3]");
364 pr_info("%-15s %08x %08x %08x %08x\n", rname
, regs
[0], regs
[1],
368 /* igb_dump - Print registers, Tx-rings and Rx-rings */
369 static void igb_dump(struct igb_adapter
*adapter
)
371 struct net_device
*netdev
= adapter
->netdev
;
372 struct e1000_hw
*hw
= &adapter
->hw
;
373 struct igb_reg_info
*reginfo
;
374 struct igb_ring
*tx_ring
;
375 union e1000_adv_tx_desc
*tx_desc
;
376 struct my_u0
{ u64 a
; u64 b
; } *u0
;
377 struct igb_ring
*rx_ring
;
378 union e1000_adv_rx_desc
*rx_desc
;
382 if (!netif_msg_hw(adapter
))
385 /* Print netdevice Info */
387 dev_info(&adapter
->pdev
->dev
, "Net device Info\n");
388 pr_info("Device Name state trans_start "
390 pr_info("%-15s %016lX %016lX %016lX\n", netdev
->name
,
391 netdev
->state
, netdev
->trans_start
, netdev
->last_rx
);
394 /* Print Registers */
395 dev_info(&adapter
->pdev
->dev
, "Register Dump\n");
396 pr_info(" Register Name Value\n");
397 for (reginfo
= (struct igb_reg_info
*)igb_reg_info_tbl
;
398 reginfo
->name
; reginfo
++) {
399 igb_regdump(hw
, reginfo
);
402 /* Print TX Ring Summary */
403 if (!netdev
|| !netif_running(netdev
))
406 dev_info(&adapter
->pdev
->dev
, "TX Rings Summary\n");
407 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
408 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
409 struct igb_tx_buffer
*buffer_info
;
410 tx_ring
= adapter
->tx_ring
[n
];
411 buffer_info
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_clean
];
412 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
413 n
, tx_ring
->next_to_use
, tx_ring
->next_to_clean
,
414 (u64
)dma_unmap_addr(buffer_info
, dma
),
415 dma_unmap_len(buffer_info
, len
),
416 buffer_info
->next_to_watch
,
417 (u64
)buffer_info
->time_stamp
);
421 if (!netif_msg_tx_done(adapter
))
422 goto rx_ring_summary
;
424 dev_info(&adapter
->pdev
->dev
, "TX Rings Dump\n");
426 /* Transmit Descriptor Formats
428 * Advanced Transmit Descriptor
429 * +--------------------------------------------------------------+
430 * 0 | Buffer Address [63:0] |
431 * +--------------------------------------------------------------+
432 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
433 * +--------------------------------------------------------------+
434 * 63 46 45 40 39 38 36 35 32 31 24 15 0
437 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
438 tx_ring
= adapter
->tx_ring
[n
];
439 pr_info("------------------------------------\n");
440 pr_info("TX QUEUE INDEX = %d\n", tx_ring
->queue_index
);
441 pr_info("------------------------------------\n");
442 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] "
443 "[bi->dma ] leng ntw timestamp "
446 for (i
= 0; tx_ring
->desc
&& (i
< tx_ring
->count
); i
++) {
447 const char *next_desc
;
448 struct igb_tx_buffer
*buffer_info
;
449 tx_desc
= IGB_TX_DESC(tx_ring
, i
);
450 buffer_info
= &tx_ring
->tx_buffer_info
[i
];
451 u0
= (struct my_u0
*)tx_desc
;
452 if (i
== tx_ring
->next_to_use
&&
453 i
== tx_ring
->next_to_clean
)
454 next_desc
= " NTC/U";
455 else if (i
== tx_ring
->next_to_use
)
457 else if (i
== tx_ring
->next_to_clean
)
462 pr_info("T [0x%03X] %016llX %016llX %016llX"
463 " %04X %p %016llX %p%s\n", i
,
466 (u64
)dma_unmap_addr(buffer_info
, dma
),
467 dma_unmap_len(buffer_info
, len
),
468 buffer_info
->next_to_watch
,
469 (u64
)buffer_info
->time_stamp
,
470 buffer_info
->skb
, next_desc
);
472 if (netif_msg_pktdata(adapter
) && buffer_info
->skb
)
473 print_hex_dump(KERN_INFO
, "",
475 16, 1, buffer_info
->skb
->data
,
476 dma_unmap_len(buffer_info
, len
),
481 /* Print RX Rings Summary */
483 dev_info(&adapter
->pdev
->dev
, "RX Rings Summary\n");
484 pr_info("Queue [NTU] [NTC]\n");
485 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
486 rx_ring
= adapter
->rx_ring
[n
];
487 pr_info(" %5d %5X %5X\n",
488 n
, rx_ring
->next_to_use
, rx_ring
->next_to_clean
);
492 if (!netif_msg_rx_status(adapter
))
495 dev_info(&adapter
->pdev
->dev
, "RX Rings Dump\n");
497 /* Advanced Receive Descriptor (Read) Format
499 * +-----------------------------------------------------+
500 * 0 | Packet Buffer Address [63:1] |A0/NSE|
501 * +----------------------------------------------+------+
502 * 8 | Header Buffer Address [63:1] | DD |
503 * +-----------------------------------------------------+
506 * Advanced Receive Descriptor (Write-Back) Format
508 * 63 48 47 32 31 30 21 20 17 16 4 3 0
509 * +------------------------------------------------------+
510 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
511 * | Checksum Ident | | | | Type | Type |
512 * +------------------------------------------------------+
513 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
514 * +------------------------------------------------------+
515 * 63 48 47 32 31 20 19 0
518 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
519 rx_ring
= adapter
->rx_ring
[n
];
520 pr_info("------------------------------------\n");
521 pr_info("RX QUEUE INDEX = %d\n", rx_ring
->queue_index
);
522 pr_info("------------------------------------\n");
523 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] "
524 "[bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
525 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] -----"
526 "----------- [bi->skb] <-- Adv Rx Write-Back format\n");
528 for (i
= 0; i
< rx_ring
->count
; i
++) {
529 const char *next_desc
;
530 struct igb_rx_buffer
*buffer_info
;
531 buffer_info
= &rx_ring
->rx_buffer_info
[i
];
532 rx_desc
= IGB_RX_DESC(rx_ring
, i
);
533 u0
= (struct my_u0
*)rx_desc
;
534 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
536 if (i
== rx_ring
->next_to_use
)
538 else if (i
== rx_ring
->next_to_clean
)
543 if (staterr
& E1000_RXD_STAT_DD
) {
544 /* Descriptor Done */
545 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n",
551 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
555 (u64
)buffer_info
->dma
,
558 if (netif_msg_pktdata(adapter
) &&
559 buffer_info
->dma
&& buffer_info
->page
) {
560 print_hex_dump(KERN_INFO
, "",
563 page_address(buffer_info
->page
) +
564 buffer_info
->page_offset
,
576 * igb_get_i2c_data - Reads the I2C SDA data bit
577 * @hw: pointer to hardware structure
578 * @i2cctl: Current value of I2CCTL register
580 * Returns the I2C data bit value
582 static int igb_get_i2c_data(void *data
)
584 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
585 struct e1000_hw
*hw
= &adapter
->hw
;
586 s32 i2cctl
= rd32(E1000_I2CPARAMS
);
588 return ((i2cctl
& E1000_I2C_DATA_IN
) != 0);
592 * igb_set_i2c_data - Sets the I2C data bit
593 * @data: pointer to hardware structure
594 * @state: I2C data value (0 or 1) to set
596 * Sets the I2C data bit
598 static void igb_set_i2c_data(void *data
, int state
)
600 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
601 struct e1000_hw
*hw
= &adapter
->hw
;
602 s32 i2cctl
= rd32(E1000_I2CPARAMS
);
605 i2cctl
|= E1000_I2C_DATA_OUT
;
607 i2cctl
&= ~E1000_I2C_DATA_OUT
;
609 i2cctl
&= ~E1000_I2C_DATA_OE_N
;
610 i2cctl
|= E1000_I2C_CLK_OE_N
;
611 wr32(E1000_I2CPARAMS
, i2cctl
);
617 * igb_set_i2c_clk - Sets the I2C SCL clock
618 * @data: pointer to hardware structure
619 * @state: state to set clock
621 * Sets the I2C clock line to state
623 static void igb_set_i2c_clk(void *data
, int state
)
625 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
626 struct e1000_hw
*hw
= &adapter
->hw
;
627 s32 i2cctl
= rd32(E1000_I2CPARAMS
);
630 i2cctl
|= E1000_I2C_CLK_OUT
;
631 i2cctl
&= ~E1000_I2C_CLK_OE_N
;
633 i2cctl
&= ~E1000_I2C_CLK_OUT
;
634 i2cctl
&= ~E1000_I2C_CLK_OE_N
;
636 wr32(E1000_I2CPARAMS
, i2cctl
);
641 * igb_get_i2c_clk - Gets the I2C SCL clock state
642 * @data: pointer to hardware structure
644 * Gets the I2C clock state
646 static int igb_get_i2c_clk(void *data
)
648 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
649 struct e1000_hw
*hw
= &adapter
->hw
;
650 s32 i2cctl
= rd32(E1000_I2CPARAMS
);
652 return ((i2cctl
& E1000_I2C_CLK_IN
) != 0);
655 static const struct i2c_algo_bit_data igb_i2c_algo
= {
656 .setsda
= igb_set_i2c_data
,
657 .setscl
= igb_set_i2c_clk
,
658 .getsda
= igb_get_i2c_data
,
659 .getscl
= igb_get_i2c_clk
,
665 * igb_get_hw_dev - return device
666 * @hw: pointer to hardware structure
668 * used by hardware layer to print debugging information
670 struct net_device
*igb_get_hw_dev(struct e1000_hw
*hw
)
672 struct igb_adapter
*adapter
= hw
->back
;
673 return adapter
->netdev
;
677 * igb_init_module - Driver Registration Routine
679 * igb_init_module is the first routine called when the driver is
680 * loaded. All it does is register with the PCI subsystem.
682 static int __init
igb_init_module(void)
685 pr_info("%s - version %s\n",
686 igb_driver_string
, igb_driver_version
);
688 pr_info("%s\n", igb_copyright
);
690 #ifdef CONFIG_IGB_DCA
691 dca_register_notify(&dca_notifier
);
693 ret
= pci_register_driver(&igb_driver
);
697 module_init(igb_init_module
);
700 * igb_exit_module - Driver Exit Cleanup Routine
702 * igb_exit_module is called just before the driver is removed
705 static void __exit
igb_exit_module(void)
707 #ifdef CONFIG_IGB_DCA
708 dca_unregister_notify(&dca_notifier
);
710 pci_unregister_driver(&igb_driver
);
713 module_exit(igb_exit_module
);
715 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
717 * igb_cache_ring_register - Descriptor ring to register mapping
718 * @adapter: board private structure to initialize
720 * Once we know the feature-set enabled for the device, we'll cache
721 * the register offset the descriptor ring is assigned to.
723 static void igb_cache_ring_register(struct igb_adapter
*adapter
)
726 u32 rbase_offset
= adapter
->vfs_allocated_count
;
728 switch (adapter
->hw
.mac
.type
) {
730 /* The queues are allocated for virtualization such that VF 0
731 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
732 * In order to avoid collision we start at the first free queue
733 * and continue consuming queues in the same sequence
735 if (adapter
->vfs_allocated_count
) {
736 for (; i
< adapter
->rss_queues
; i
++)
737 adapter
->rx_ring
[i
]->reg_idx
= rbase_offset
+
747 for (; i
< adapter
->num_rx_queues
; i
++)
748 adapter
->rx_ring
[i
]->reg_idx
= rbase_offset
+ i
;
749 for (; j
< adapter
->num_tx_queues
; j
++)
750 adapter
->tx_ring
[j
]->reg_idx
= rbase_offset
+ j
;
756 * igb_write_ivar - configure ivar for given MSI-X vector
757 * @hw: pointer to the HW structure
758 * @msix_vector: vector number we are allocating to a given ring
759 * @index: row index of IVAR register to write within IVAR table
760 * @offset: column offset of in IVAR, should be multiple of 8
762 * This function is intended to handle the writing of the IVAR register
763 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
764 * each containing an cause allocation for an Rx and Tx ring, and a
765 * variable number of rows depending on the number of queues supported.
767 static void igb_write_ivar(struct e1000_hw
*hw
, int msix_vector
,
768 int index
, int offset
)
770 u32 ivar
= array_rd32(E1000_IVAR0
, index
);
772 /* clear any bits that are currently set */
773 ivar
&= ~((u32
)0xFF << offset
);
775 /* write vector and valid bit */
776 ivar
|= (msix_vector
| E1000_IVAR_VALID
) << offset
;
778 array_wr32(E1000_IVAR0
, index
, ivar
);
781 #define IGB_N0_QUEUE -1
782 static void igb_assign_vector(struct igb_q_vector
*q_vector
, int msix_vector
)
784 struct igb_adapter
*adapter
= q_vector
->adapter
;
785 struct e1000_hw
*hw
= &adapter
->hw
;
786 int rx_queue
= IGB_N0_QUEUE
;
787 int tx_queue
= IGB_N0_QUEUE
;
790 if (q_vector
->rx
.ring
)
791 rx_queue
= q_vector
->rx
.ring
->reg_idx
;
792 if (q_vector
->tx
.ring
)
793 tx_queue
= q_vector
->tx
.ring
->reg_idx
;
795 switch (hw
->mac
.type
) {
797 /* The 82575 assigns vectors using a bitmask, which matches the
798 * bitmask for the EICR/EIMS/EIMC registers. To assign one
799 * or more queues to a vector, we write the appropriate bits
800 * into the MSIXBM register for that vector.
802 if (rx_queue
> IGB_N0_QUEUE
)
803 msixbm
= E1000_EICR_RX_QUEUE0
<< rx_queue
;
804 if (tx_queue
> IGB_N0_QUEUE
)
805 msixbm
|= E1000_EICR_TX_QUEUE0
<< tx_queue
;
806 if (!adapter
->msix_entries
&& msix_vector
== 0)
807 msixbm
|= E1000_EIMS_OTHER
;
808 array_wr32(E1000_MSIXBM(0), msix_vector
, msixbm
);
809 q_vector
->eims_value
= msixbm
;
812 /* 82576 uses a table that essentially consists of 2 columns
813 * with 8 rows. The ordering is column-major so we use the
814 * lower 3 bits as the row index, and the 4th bit as the
817 if (rx_queue
> IGB_N0_QUEUE
)
818 igb_write_ivar(hw
, msix_vector
,
820 (rx_queue
& 0x8) << 1);
821 if (tx_queue
> IGB_N0_QUEUE
)
822 igb_write_ivar(hw
, msix_vector
,
824 ((tx_queue
& 0x8) << 1) + 8);
825 q_vector
->eims_value
= 1 << msix_vector
;
832 /* On 82580 and newer adapters the scheme is similar to 82576
833 * however instead of ordering column-major we have things
834 * ordered row-major. So we traverse the table by using
835 * bit 0 as the column offset, and the remaining bits as the
838 if (rx_queue
> IGB_N0_QUEUE
)
839 igb_write_ivar(hw
, msix_vector
,
841 (rx_queue
& 0x1) << 4);
842 if (tx_queue
> IGB_N0_QUEUE
)
843 igb_write_ivar(hw
, msix_vector
,
845 ((tx_queue
& 0x1) << 4) + 8);
846 q_vector
->eims_value
= 1 << msix_vector
;
853 /* add q_vector eims value to global eims_enable_mask */
854 adapter
->eims_enable_mask
|= q_vector
->eims_value
;
856 /* configure q_vector to set itr on first interrupt */
857 q_vector
->set_itr
= 1;
861 * igb_configure_msix - Configure MSI-X hardware
862 * @adapter: board private structure to initialize
864 * igb_configure_msix sets up the hardware to properly
865 * generate MSI-X interrupts.
867 static void igb_configure_msix(struct igb_adapter
*adapter
)
871 struct e1000_hw
*hw
= &adapter
->hw
;
873 adapter
->eims_enable_mask
= 0;
875 /* set vector for other causes, i.e. link changes */
876 switch (hw
->mac
.type
) {
878 tmp
= rd32(E1000_CTRL_EXT
);
879 /* enable MSI-X PBA support*/
880 tmp
|= E1000_CTRL_EXT_PBA_CLR
;
882 /* Auto-Mask interrupts upon ICR read. */
883 tmp
|= E1000_CTRL_EXT_EIAME
;
884 tmp
|= E1000_CTRL_EXT_IRCA
;
886 wr32(E1000_CTRL_EXT
, tmp
);
888 /* enable msix_other interrupt */
889 array_wr32(E1000_MSIXBM(0), vector
++, E1000_EIMS_OTHER
);
890 adapter
->eims_other
= E1000_EIMS_OTHER
;
900 /* Turn on MSI-X capability first, or our settings
901 * won't stick. And it will take days to debug.
903 wr32(E1000_GPIE
, E1000_GPIE_MSIX_MODE
|
904 E1000_GPIE_PBA
| E1000_GPIE_EIAME
|
907 /* enable msix_other interrupt */
908 adapter
->eims_other
= 1 << vector
;
909 tmp
= (vector
++ | E1000_IVAR_VALID
) << 8;
911 wr32(E1000_IVAR_MISC
, tmp
);
914 /* do nothing, since nothing else supports MSI-X */
916 } /* switch (hw->mac.type) */
918 adapter
->eims_enable_mask
|= adapter
->eims_other
;
920 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
921 igb_assign_vector(adapter
->q_vector
[i
], vector
++);
927 * igb_request_msix - Initialize MSI-X interrupts
928 * @adapter: board private structure to initialize
930 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
933 static int igb_request_msix(struct igb_adapter
*adapter
)
935 struct net_device
*netdev
= adapter
->netdev
;
936 struct e1000_hw
*hw
= &adapter
->hw
;
937 int i
, err
= 0, vector
= 0, free_vector
= 0;
939 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
940 igb_msix_other
, 0, netdev
->name
, adapter
);
944 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
945 struct igb_q_vector
*q_vector
= adapter
->q_vector
[i
];
949 q_vector
->itr_register
= hw
->hw_addr
+ E1000_EITR(vector
);
951 if (q_vector
->rx
.ring
&& q_vector
->tx
.ring
)
952 sprintf(q_vector
->name
, "%s-TxRx-%u", netdev
->name
,
953 q_vector
->rx
.ring
->queue_index
);
954 else if (q_vector
->tx
.ring
)
955 sprintf(q_vector
->name
, "%s-tx-%u", netdev
->name
,
956 q_vector
->tx
.ring
->queue_index
);
957 else if (q_vector
->rx
.ring
)
958 sprintf(q_vector
->name
, "%s-rx-%u", netdev
->name
,
959 q_vector
->rx
.ring
->queue_index
);
961 sprintf(q_vector
->name
, "%s-unused", netdev
->name
);
963 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
964 igb_msix_ring
, 0, q_vector
->name
,
970 igb_configure_msix(adapter
);
974 /* free already assigned IRQs */
975 free_irq(adapter
->msix_entries
[free_vector
++].vector
, adapter
);
978 for (i
= 0; i
< vector
; i
++) {
979 free_irq(adapter
->msix_entries
[free_vector
++].vector
,
980 adapter
->q_vector
[i
]);
986 static void igb_reset_interrupt_capability(struct igb_adapter
*adapter
)
988 if (adapter
->msix_entries
) {
989 pci_disable_msix(adapter
->pdev
);
990 kfree(adapter
->msix_entries
);
991 adapter
->msix_entries
= NULL
;
992 } else if (adapter
->flags
& IGB_FLAG_HAS_MSI
) {
993 pci_disable_msi(adapter
->pdev
);
998 * igb_free_q_vector - Free memory allocated for specific interrupt vector
999 * @adapter: board private structure to initialize
1000 * @v_idx: Index of vector to be freed
1002 * This function frees the memory allocated to the q_vector. In addition if
1003 * NAPI is enabled it will delete any references to the NAPI struct prior
1004 * to freeing the q_vector.
1006 static void igb_free_q_vector(struct igb_adapter
*adapter
, int v_idx
)
1008 struct igb_q_vector
*q_vector
= adapter
->q_vector
[v_idx
];
1010 if (q_vector
->tx
.ring
)
1011 adapter
->tx_ring
[q_vector
->tx
.ring
->queue_index
] = NULL
;
1013 if (q_vector
->rx
.ring
)
1014 adapter
->tx_ring
[q_vector
->rx
.ring
->queue_index
] = NULL
;
1016 adapter
->q_vector
[v_idx
] = NULL
;
1017 netif_napi_del(&q_vector
->napi
);
1019 /* igb_get_stats64() might access the rings on this vector,
1020 * we must wait a grace period before freeing it.
1022 kfree_rcu(q_vector
, rcu
);
1026 * igb_free_q_vectors - Free memory allocated for interrupt vectors
1027 * @adapter: board private structure to initialize
1029 * This function frees the memory allocated to the q_vectors. In addition if
1030 * NAPI is enabled it will delete any references to the NAPI struct prior
1031 * to freeing the q_vector.
1033 static void igb_free_q_vectors(struct igb_adapter
*adapter
)
1035 int v_idx
= adapter
->num_q_vectors
;
1037 adapter
->num_tx_queues
= 0;
1038 adapter
->num_rx_queues
= 0;
1039 adapter
->num_q_vectors
= 0;
1042 igb_free_q_vector(adapter
, v_idx
);
1046 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1047 * @adapter: board private structure to initialize
1049 * This function resets the device so that it has 0 Rx queues, Tx queues, and
1050 * MSI-X interrupts allocated.
1052 static void igb_clear_interrupt_scheme(struct igb_adapter
*adapter
)
1054 igb_free_q_vectors(adapter
);
1055 igb_reset_interrupt_capability(adapter
);
1059 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1060 * @adapter: board private structure to initialize
1061 * @msix: boolean value of MSIX capability
1063 * Attempt to configure interrupts using the best available
1064 * capabilities of the hardware and kernel.
1066 static void igb_set_interrupt_capability(struct igb_adapter
*adapter
, bool msix
)
1074 /* Number of supported queues. */
1075 adapter
->num_rx_queues
= adapter
->rss_queues
;
1076 if (adapter
->vfs_allocated_count
)
1077 adapter
->num_tx_queues
= 1;
1079 adapter
->num_tx_queues
= adapter
->rss_queues
;
1081 /* start with one vector for every Rx queue */
1082 numvecs
= adapter
->num_rx_queues
;
1084 /* if Tx handler is separate add 1 for every Tx queue */
1085 if (!(adapter
->flags
& IGB_FLAG_QUEUE_PAIRS
))
1086 numvecs
+= adapter
->num_tx_queues
;
1088 /* store the number of vectors reserved for queues */
1089 adapter
->num_q_vectors
= numvecs
;
1091 /* add 1 vector for link status interrupts */
1093 adapter
->msix_entries
= kcalloc(numvecs
, sizeof(struct msix_entry
),
1096 if (!adapter
->msix_entries
)
1099 for (i
= 0; i
< numvecs
; i
++)
1100 adapter
->msix_entries
[i
].entry
= i
;
1102 err
= pci_enable_msix(adapter
->pdev
,
1103 adapter
->msix_entries
,
1108 igb_reset_interrupt_capability(adapter
);
1110 /* If we can't do MSI-X, try MSI */
1112 #ifdef CONFIG_PCI_IOV
1113 /* disable SR-IOV for non MSI-X configurations */
1114 if (adapter
->vf_data
) {
1115 struct e1000_hw
*hw
= &adapter
->hw
;
1116 /* disable iov and allow time for transactions to clear */
1117 pci_disable_sriov(adapter
->pdev
);
1120 kfree(adapter
->vf_data
);
1121 adapter
->vf_data
= NULL
;
1122 wr32(E1000_IOVCTL
, E1000_IOVCTL_REUSE_VFQ
);
1125 dev_info(&adapter
->pdev
->dev
, "IOV Disabled\n");
1128 adapter
->vfs_allocated_count
= 0;
1129 adapter
->rss_queues
= 1;
1130 adapter
->flags
|= IGB_FLAG_QUEUE_PAIRS
;
1131 adapter
->num_rx_queues
= 1;
1132 adapter
->num_tx_queues
= 1;
1133 adapter
->num_q_vectors
= 1;
1134 if (!pci_enable_msi(adapter
->pdev
))
1135 adapter
->flags
|= IGB_FLAG_HAS_MSI
;
1138 static void igb_add_ring(struct igb_ring
*ring
,
1139 struct igb_ring_container
*head
)
1146 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1147 * @adapter: board private structure to initialize
1148 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1149 * @v_idx: index of vector in adapter struct
1150 * @txr_count: total number of Tx rings to allocate
1151 * @txr_idx: index of first Tx ring to allocate
1152 * @rxr_count: total number of Rx rings to allocate
1153 * @rxr_idx: index of first Rx ring to allocate
1155 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1157 static int igb_alloc_q_vector(struct igb_adapter
*adapter
,
1158 int v_count
, int v_idx
,
1159 int txr_count
, int txr_idx
,
1160 int rxr_count
, int rxr_idx
)
1162 struct igb_q_vector
*q_vector
;
1163 struct igb_ring
*ring
;
1164 int ring_count
, size
;
1166 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1167 if (txr_count
> 1 || rxr_count
> 1)
1170 ring_count
= txr_count
+ rxr_count
;
1171 size
= sizeof(struct igb_q_vector
) +
1172 (sizeof(struct igb_ring
) * ring_count
);
1174 /* allocate q_vector and rings */
1175 q_vector
= kzalloc(size
, GFP_KERNEL
);
1179 /* initialize NAPI */
1180 netif_napi_add(adapter
->netdev
, &q_vector
->napi
,
1183 /* tie q_vector and adapter together */
1184 adapter
->q_vector
[v_idx
] = q_vector
;
1185 q_vector
->adapter
= adapter
;
1187 /* initialize work limits */
1188 q_vector
->tx
.work_limit
= adapter
->tx_work_limit
;
1190 /* initialize ITR configuration */
1191 q_vector
->itr_register
= adapter
->hw
.hw_addr
+ E1000_EITR(0);
1192 q_vector
->itr_val
= IGB_START_ITR
;
1194 /* initialize pointer to rings */
1195 ring
= q_vector
->ring
;
1199 /* rx or rx/tx vector */
1200 if (!adapter
->rx_itr_setting
|| adapter
->rx_itr_setting
> 3)
1201 q_vector
->itr_val
= adapter
->rx_itr_setting
;
1203 /* tx only vector */
1204 if (!adapter
->tx_itr_setting
|| adapter
->tx_itr_setting
> 3)
1205 q_vector
->itr_val
= adapter
->tx_itr_setting
;
1209 /* assign generic ring traits */
1210 ring
->dev
= &adapter
->pdev
->dev
;
1211 ring
->netdev
= adapter
->netdev
;
1213 /* configure backlink on ring */
1214 ring
->q_vector
= q_vector
;
1216 /* update q_vector Tx values */
1217 igb_add_ring(ring
, &q_vector
->tx
);
1219 /* For 82575, context index must be unique per ring. */
1220 if (adapter
->hw
.mac
.type
== e1000_82575
)
1221 set_bit(IGB_RING_FLAG_TX_CTX_IDX
, &ring
->flags
);
1223 /* apply Tx specific ring traits */
1224 ring
->count
= adapter
->tx_ring_count
;
1225 ring
->queue_index
= txr_idx
;
1227 u64_stats_init(&ring
->tx_syncp
);
1228 u64_stats_init(&ring
->tx_syncp2
);
1230 /* assign ring to adapter */
1231 adapter
->tx_ring
[txr_idx
] = ring
;
1233 /* push pointer to next ring */
1238 /* assign generic ring traits */
1239 ring
->dev
= &adapter
->pdev
->dev
;
1240 ring
->netdev
= adapter
->netdev
;
1242 /* configure backlink on ring */
1243 ring
->q_vector
= q_vector
;
1245 /* update q_vector Rx values */
1246 igb_add_ring(ring
, &q_vector
->rx
);
1248 /* set flag indicating ring supports SCTP checksum offload */
1249 if (adapter
->hw
.mac
.type
>= e1000_82576
)
1250 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM
, &ring
->flags
);
1253 * On i350, i354, i210, and i211, loopback VLAN packets
1254 * have the tag byte-swapped.
1256 if (adapter
->hw
.mac
.type
>= e1000_i350
)
1257 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP
, &ring
->flags
);
1259 /* apply Rx specific ring traits */
1260 ring
->count
= adapter
->rx_ring_count
;
1261 ring
->queue_index
= rxr_idx
;
1263 u64_stats_init(&ring
->rx_syncp
);
1265 /* assign ring to adapter */
1266 adapter
->rx_ring
[rxr_idx
] = ring
;
1274 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1275 * @adapter: board private structure to initialize
1277 * We allocate one q_vector per queue interrupt. If allocation fails we
1280 static int igb_alloc_q_vectors(struct igb_adapter
*adapter
)
1282 int q_vectors
= adapter
->num_q_vectors
;
1283 int rxr_remaining
= adapter
->num_rx_queues
;
1284 int txr_remaining
= adapter
->num_tx_queues
;
1285 int rxr_idx
= 0, txr_idx
= 0, v_idx
= 0;
1288 if (q_vectors
>= (rxr_remaining
+ txr_remaining
)) {
1289 for (; rxr_remaining
; v_idx
++) {
1290 err
= igb_alloc_q_vector(adapter
, q_vectors
, v_idx
,
1296 /* update counts and index */
1302 for (; v_idx
< q_vectors
; v_idx
++) {
1303 int rqpv
= DIV_ROUND_UP(rxr_remaining
, q_vectors
- v_idx
);
1304 int tqpv
= DIV_ROUND_UP(txr_remaining
, q_vectors
- v_idx
);
1305 err
= igb_alloc_q_vector(adapter
, q_vectors
, v_idx
,
1306 tqpv
, txr_idx
, rqpv
, rxr_idx
);
1311 /* update counts and index */
1312 rxr_remaining
-= rqpv
;
1313 txr_remaining
-= tqpv
;
1321 adapter
->num_tx_queues
= 0;
1322 adapter
->num_rx_queues
= 0;
1323 adapter
->num_q_vectors
= 0;
1326 igb_free_q_vector(adapter
, v_idx
);
1332 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1333 * @adapter: board private structure to initialize
1334 * @msix: boolean value of MSIX capability
1336 * This function initializes the interrupts and allocates all of the queues.
1338 static int igb_init_interrupt_scheme(struct igb_adapter
*adapter
, bool msix
)
1340 struct pci_dev
*pdev
= adapter
->pdev
;
1343 igb_set_interrupt_capability(adapter
, msix
);
1345 err
= igb_alloc_q_vectors(adapter
);
1347 dev_err(&pdev
->dev
, "Unable to allocate memory for vectors\n");
1348 goto err_alloc_q_vectors
;
1351 igb_cache_ring_register(adapter
);
1355 err_alloc_q_vectors
:
1356 igb_reset_interrupt_capability(adapter
);
1361 * igb_request_irq - initialize interrupts
1362 * @adapter: board private structure to initialize
1364 * Attempts to configure interrupts using the best available
1365 * capabilities of the hardware and kernel.
1367 static int igb_request_irq(struct igb_adapter
*adapter
)
1369 struct net_device
*netdev
= adapter
->netdev
;
1370 struct pci_dev
*pdev
= adapter
->pdev
;
1373 if (adapter
->msix_entries
) {
1374 err
= igb_request_msix(adapter
);
1377 /* fall back to MSI */
1378 igb_free_all_tx_resources(adapter
);
1379 igb_free_all_rx_resources(adapter
);
1381 igb_clear_interrupt_scheme(adapter
);
1382 err
= igb_init_interrupt_scheme(adapter
, false);
1386 igb_setup_all_tx_resources(adapter
);
1387 igb_setup_all_rx_resources(adapter
);
1388 igb_configure(adapter
);
1391 igb_assign_vector(adapter
->q_vector
[0], 0);
1393 if (adapter
->flags
& IGB_FLAG_HAS_MSI
) {
1394 err
= request_irq(pdev
->irq
, igb_intr_msi
, 0,
1395 netdev
->name
, adapter
);
1399 /* fall back to legacy interrupts */
1400 igb_reset_interrupt_capability(adapter
);
1401 adapter
->flags
&= ~IGB_FLAG_HAS_MSI
;
1404 err
= request_irq(pdev
->irq
, igb_intr
, IRQF_SHARED
,
1405 netdev
->name
, adapter
);
1408 dev_err(&pdev
->dev
, "Error %d getting interrupt\n",
1415 static void igb_free_irq(struct igb_adapter
*adapter
)
1417 if (adapter
->msix_entries
) {
1420 free_irq(adapter
->msix_entries
[vector
++].vector
, adapter
);
1422 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1423 free_irq(adapter
->msix_entries
[vector
++].vector
,
1424 adapter
->q_vector
[i
]);
1426 free_irq(adapter
->pdev
->irq
, adapter
);
1431 * igb_irq_disable - Mask off interrupt generation on the NIC
1432 * @adapter: board private structure
1434 static void igb_irq_disable(struct igb_adapter
*adapter
)
1436 struct e1000_hw
*hw
= &adapter
->hw
;
1438 /* we need to be careful when disabling interrupts. The VFs are also
1439 * mapped into these registers and so clearing the bits can cause
1440 * issues on the VF drivers so we only need to clear what we set
1442 if (adapter
->msix_entries
) {
1443 u32 regval
= rd32(E1000_EIAM
);
1444 wr32(E1000_EIAM
, regval
& ~adapter
->eims_enable_mask
);
1445 wr32(E1000_EIMC
, adapter
->eims_enable_mask
);
1446 regval
= rd32(E1000_EIAC
);
1447 wr32(E1000_EIAC
, regval
& ~adapter
->eims_enable_mask
);
1451 wr32(E1000_IMC
, ~0);
1453 if (adapter
->msix_entries
) {
1455 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1456 synchronize_irq(adapter
->msix_entries
[i
].vector
);
1458 synchronize_irq(adapter
->pdev
->irq
);
1463 * igb_irq_enable - Enable default interrupt generation settings
1464 * @adapter: board private structure
1466 static void igb_irq_enable(struct igb_adapter
*adapter
)
1468 struct e1000_hw
*hw
= &adapter
->hw
;
1470 if (adapter
->msix_entries
) {
1471 u32 ims
= E1000_IMS_LSC
| E1000_IMS_DOUTSYNC
| E1000_IMS_DRSTA
;
1472 u32 regval
= rd32(E1000_EIAC
);
1473 wr32(E1000_EIAC
, regval
| adapter
->eims_enable_mask
);
1474 regval
= rd32(E1000_EIAM
);
1475 wr32(E1000_EIAM
, regval
| adapter
->eims_enable_mask
);
1476 wr32(E1000_EIMS
, adapter
->eims_enable_mask
);
1477 if (adapter
->vfs_allocated_count
) {
1478 wr32(E1000_MBVFIMR
, 0xFF);
1479 ims
|= E1000_IMS_VMMB
;
1481 wr32(E1000_IMS
, ims
);
1483 wr32(E1000_IMS
, IMS_ENABLE_MASK
|
1485 wr32(E1000_IAM
, IMS_ENABLE_MASK
|
1490 static void igb_update_mng_vlan(struct igb_adapter
*adapter
)
1492 struct e1000_hw
*hw
= &adapter
->hw
;
1493 u16 vid
= adapter
->hw
.mng_cookie
.vlan_id
;
1494 u16 old_vid
= adapter
->mng_vlan_id
;
1496 if (hw
->mng_cookie
.status
& E1000_MNG_DHCP_COOKIE_STATUS_VLAN
) {
1497 /* add VID to filter table */
1498 igb_vfta_set(hw
, vid
, true);
1499 adapter
->mng_vlan_id
= vid
;
1501 adapter
->mng_vlan_id
= IGB_MNG_VLAN_NONE
;
1504 if ((old_vid
!= (u16
)IGB_MNG_VLAN_NONE
) &&
1506 !test_bit(old_vid
, adapter
->active_vlans
)) {
1507 /* remove VID from filter table */
1508 igb_vfta_set(hw
, old_vid
, false);
1513 * igb_release_hw_control - release control of the h/w to f/w
1514 * @adapter: address of board private structure
1516 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1517 * For ASF and Pass Through versions of f/w this means that the
1518 * driver is no longer loaded.
1520 static void igb_release_hw_control(struct igb_adapter
*adapter
)
1522 struct e1000_hw
*hw
= &adapter
->hw
;
1525 /* Let firmware take over control of h/w */
1526 ctrl_ext
= rd32(E1000_CTRL_EXT
);
1527 wr32(E1000_CTRL_EXT
,
1528 ctrl_ext
& ~E1000_CTRL_EXT_DRV_LOAD
);
1532 * igb_get_hw_control - get control of the h/w from f/w
1533 * @adapter: address of board private structure
1535 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1536 * For ASF and Pass Through versions of f/w this means that
1537 * the driver is loaded.
1539 static void igb_get_hw_control(struct igb_adapter
*adapter
)
1541 struct e1000_hw
*hw
= &adapter
->hw
;
1544 /* Let firmware know the driver has taken over */
1545 ctrl_ext
= rd32(E1000_CTRL_EXT
);
1546 wr32(E1000_CTRL_EXT
,
1547 ctrl_ext
| E1000_CTRL_EXT_DRV_LOAD
);
1551 * igb_configure - configure the hardware for RX and TX
1552 * @adapter: private board structure
1554 static void igb_configure(struct igb_adapter
*adapter
)
1556 struct net_device
*netdev
= adapter
->netdev
;
1559 igb_get_hw_control(adapter
);
1560 igb_set_rx_mode(netdev
);
1562 igb_restore_vlan(adapter
);
1564 igb_setup_tctl(adapter
);
1565 igb_setup_mrqc(adapter
);
1566 igb_setup_rctl(adapter
);
1568 igb_configure_tx(adapter
);
1569 igb_configure_rx(adapter
);
1571 igb_rx_fifo_flush_82575(&adapter
->hw
);
1573 /* call igb_desc_unused which always leaves
1574 * at least 1 descriptor unused to make sure
1575 * next_to_use != next_to_clean
1577 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
1578 struct igb_ring
*ring
= adapter
->rx_ring
[i
];
1579 igb_alloc_rx_buffers(ring
, igb_desc_unused(ring
));
1584 * igb_power_up_link - Power up the phy/serdes link
1585 * @adapter: address of board private structure
1587 void igb_power_up_link(struct igb_adapter
*adapter
)
1589 igb_reset_phy(&adapter
->hw
);
1591 if (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
)
1592 igb_power_up_phy_copper(&adapter
->hw
);
1594 igb_power_up_serdes_link_82575(&adapter
->hw
);
1598 * igb_power_down_link - Power down the phy/serdes link
1599 * @adapter: address of board private structure
1601 static void igb_power_down_link(struct igb_adapter
*adapter
)
1603 if (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
)
1604 igb_power_down_phy_copper_82575(&adapter
->hw
);
1606 igb_shutdown_serdes_link_82575(&adapter
->hw
);
1610 * igb_up - Open the interface and prepare it to handle traffic
1611 * @adapter: board private structure
1613 int igb_up(struct igb_adapter
*adapter
)
1615 struct e1000_hw
*hw
= &adapter
->hw
;
1618 /* hardware has been reset, we need to reload some things */
1619 igb_configure(adapter
);
1621 clear_bit(__IGB_DOWN
, &adapter
->state
);
1623 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1624 napi_enable(&(adapter
->q_vector
[i
]->napi
));
1626 if (adapter
->msix_entries
)
1627 igb_configure_msix(adapter
);
1629 igb_assign_vector(adapter
->q_vector
[0], 0);
1631 /* Clear any pending interrupts. */
1633 igb_irq_enable(adapter
);
1635 /* notify VFs that reset has been completed */
1636 if (adapter
->vfs_allocated_count
) {
1637 u32 reg_data
= rd32(E1000_CTRL_EXT
);
1638 reg_data
|= E1000_CTRL_EXT_PFRSTD
;
1639 wr32(E1000_CTRL_EXT
, reg_data
);
1642 netif_tx_start_all_queues(adapter
->netdev
);
1644 /* start the watchdog. */
1645 hw
->mac
.get_link_status
= 1;
1646 schedule_work(&adapter
->watchdog_task
);
1651 void igb_down(struct igb_adapter
*adapter
)
1653 struct net_device
*netdev
= adapter
->netdev
;
1654 struct e1000_hw
*hw
= &adapter
->hw
;
1658 /* signal that we're down so the interrupt handler does not
1659 * reschedule our watchdog timer
1661 set_bit(__IGB_DOWN
, &adapter
->state
);
1663 /* disable receives in the hardware */
1664 rctl
= rd32(E1000_RCTL
);
1665 wr32(E1000_RCTL
, rctl
& ~E1000_RCTL_EN
);
1666 /* flush and sleep below */
1668 netif_tx_stop_all_queues(netdev
);
1670 /* disable transmits in the hardware */
1671 tctl
= rd32(E1000_TCTL
);
1672 tctl
&= ~E1000_TCTL_EN
;
1673 wr32(E1000_TCTL
, tctl
);
1674 /* flush both disables and wait for them to finish */
1678 igb_irq_disable(adapter
);
1680 adapter
->flags
&= ~IGB_FLAG_NEED_LINK_UPDATE
;
1682 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
1683 napi_synchronize(&(adapter
->q_vector
[i
]->napi
));
1684 napi_disable(&(adapter
->q_vector
[i
]->napi
));
1688 del_timer_sync(&adapter
->watchdog_timer
);
1689 del_timer_sync(&adapter
->phy_info_timer
);
1691 netif_carrier_off(netdev
);
1693 /* record the stats before reset*/
1694 spin_lock(&adapter
->stats64_lock
);
1695 igb_update_stats(adapter
, &adapter
->stats64
);
1696 spin_unlock(&adapter
->stats64_lock
);
1698 adapter
->link_speed
= 0;
1699 adapter
->link_duplex
= 0;
1701 if (!pci_channel_offline(adapter
->pdev
))
1703 igb_clean_all_tx_rings(adapter
);
1704 igb_clean_all_rx_rings(adapter
);
1705 #ifdef CONFIG_IGB_DCA
1707 /* since we reset the hardware DCA settings were cleared */
1708 igb_setup_dca(adapter
);
1712 void igb_reinit_locked(struct igb_adapter
*adapter
)
1714 WARN_ON(in_interrupt());
1715 while (test_and_set_bit(__IGB_RESETTING
, &adapter
->state
))
1719 clear_bit(__IGB_RESETTING
, &adapter
->state
);
1722 void igb_reset(struct igb_adapter
*adapter
)
1724 struct pci_dev
*pdev
= adapter
->pdev
;
1725 struct e1000_hw
*hw
= &adapter
->hw
;
1726 struct e1000_mac_info
*mac
= &hw
->mac
;
1727 struct e1000_fc_info
*fc
= &hw
->fc
;
1728 u32 pba
= 0, tx_space
, min_tx_space
, min_rx_space
, hwm
;
1730 /* Repartition Pba for greater than 9k mtu
1731 * To take effect CTRL.RST is required.
1733 switch (mac
->type
) {
1737 pba
= rd32(E1000_RXPBS
);
1738 pba
= igb_rxpbs_adjust_82580(pba
);
1741 pba
= rd32(E1000_RXPBS
);
1742 pba
&= E1000_RXPBS_SIZE_MASK_82576
;
1748 pba
= E1000_PBA_34K
;
1752 if ((adapter
->max_frame_size
> ETH_FRAME_LEN
+ ETH_FCS_LEN
) &&
1753 (mac
->type
< e1000_82576
)) {
1754 /* adjust PBA for jumbo frames */
1755 wr32(E1000_PBA
, pba
);
1757 /* To maintain wire speed transmits, the Tx FIFO should be
1758 * large enough to accommodate two full transmit packets,
1759 * rounded up to the next 1KB and expressed in KB. Likewise,
1760 * the Rx FIFO should be large enough to accommodate at least
1761 * one full receive packet and is similarly rounded up and
1764 pba
= rd32(E1000_PBA
);
1765 /* upper 16 bits has Tx packet buffer allocation size in KB */
1766 tx_space
= pba
>> 16;
1767 /* lower 16 bits has Rx packet buffer allocation size in KB */
1769 /* the Tx fifo also stores 16 bytes of information about the Tx
1770 * but don't include ethernet FCS because hardware appends it
1772 min_tx_space
= (adapter
->max_frame_size
+
1773 sizeof(union e1000_adv_tx_desc
) -
1775 min_tx_space
= ALIGN(min_tx_space
, 1024);
1776 min_tx_space
>>= 10;
1777 /* software strips receive CRC, so leave room for it */
1778 min_rx_space
= adapter
->max_frame_size
;
1779 min_rx_space
= ALIGN(min_rx_space
, 1024);
1780 min_rx_space
>>= 10;
1782 /* If current Tx allocation is less than the min Tx FIFO size,
1783 * and the min Tx FIFO size is less than the current Rx FIFO
1784 * allocation, take space away from current Rx allocation
1786 if (tx_space
< min_tx_space
&&
1787 ((min_tx_space
- tx_space
) < pba
)) {
1788 pba
= pba
- (min_tx_space
- tx_space
);
1790 /* if short on Rx space, Rx wins and must trump Tx
1793 if (pba
< min_rx_space
)
1796 wr32(E1000_PBA
, pba
);
1799 /* flow control settings */
1800 /* The high water mark must be low enough to fit one full frame
1801 * (or the size used for early receive) above it in the Rx FIFO.
1802 * Set it to the lower of:
1803 * - 90% of the Rx FIFO size, or
1804 * - the full Rx FIFO size minus one full frame
1806 hwm
= min(((pba
<< 10) * 9 / 10),
1807 ((pba
<< 10) - 2 * adapter
->max_frame_size
));
1809 fc
->high_water
= hwm
& 0xFFFFFFF0; /* 16-byte granularity */
1810 fc
->low_water
= fc
->high_water
- 16;
1811 fc
->pause_time
= 0xFFFF;
1813 fc
->current_mode
= fc
->requested_mode
;
1815 /* disable receive for all VFs and wait one second */
1816 if (adapter
->vfs_allocated_count
) {
1818 for (i
= 0 ; i
< adapter
->vfs_allocated_count
; i
++)
1819 adapter
->vf_data
[i
].flags
&= IGB_VF_FLAG_PF_SET_MAC
;
1821 /* ping all the active vfs to let them know we are going down */
1822 igb_ping_all_vfs(adapter
);
1824 /* disable transmits and receives */
1825 wr32(E1000_VFRE
, 0);
1826 wr32(E1000_VFTE
, 0);
1829 /* Allow time for pending master requests to run */
1830 hw
->mac
.ops
.reset_hw(hw
);
1833 if (hw
->mac
.ops
.init_hw(hw
))
1834 dev_err(&pdev
->dev
, "Hardware Error\n");
1836 /* Flow control settings reset on hardware reset, so guarantee flow
1837 * control is off when forcing speed.
1839 if (!hw
->mac
.autoneg
)
1840 igb_force_mac_fc(hw
);
1842 igb_init_dmac(adapter
, pba
);
1843 #ifdef CONFIG_IGB_HWMON
1844 /* Re-initialize the thermal sensor on i350 devices. */
1845 if (!test_bit(__IGB_DOWN
, &adapter
->state
)) {
1846 if (mac
->type
== e1000_i350
&& hw
->bus
.func
== 0) {
1847 /* If present, re-initialize the external thermal sensor
1851 mac
->ops
.init_thermal_sensor_thresh(hw
);
1855 if (!netif_running(adapter
->netdev
))
1856 igb_power_down_link(adapter
);
1858 igb_update_mng_vlan(adapter
);
1860 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1861 wr32(E1000_VET
, ETHERNET_IEEE_VLAN_TYPE
);
1863 /* Re-enable PTP, where applicable. */
1864 igb_ptp_reset(adapter
);
1866 igb_get_phy_info(hw
);
1869 static netdev_features_t
igb_fix_features(struct net_device
*netdev
,
1870 netdev_features_t features
)
1872 /* Since there is no support for separate Rx/Tx vlan accel
1873 * enable/disable make sure Tx flag is always in same state as Rx.
1875 if (features
& NETIF_F_HW_VLAN_CTAG_RX
)
1876 features
|= NETIF_F_HW_VLAN_CTAG_TX
;
1878 features
&= ~NETIF_F_HW_VLAN_CTAG_TX
;
1883 static int igb_set_features(struct net_device
*netdev
,
1884 netdev_features_t features
)
1886 netdev_features_t changed
= netdev
->features
^ features
;
1887 struct igb_adapter
*adapter
= netdev_priv(netdev
);
1889 if (changed
& NETIF_F_HW_VLAN_CTAG_RX
)
1890 igb_vlan_mode(netdev
, features
);
1892 if (!(changed
& NETIF_F_RXALL
))
1895 netdev
->features
= features
;
1897 if (netif_running(netdev
))
1898 igb_reinit_locked(adapter
);
1905 static const struct net_device_ops igb_netdev_ops
= {
1906 .ndo_open
= igb_open
,
1907 .ndo_stop
= igb_close
,
1908 .ndo_start_xmit
= igb_xmit_frame
,
1909 .ndo_get_stats64
= igb_get_stats64
,
1910 .ndo_set_rx_mode
= igb_set_rx_mode
,
1911 .ndo_set_mac_address
= igb_set_mac
,
1912 .ndo_change_mtu
= igb_change_mtu
,
1913 .ndo_do_ioctl
= igb_ioctl
,
1914 .ndo_tx_timeout
= igb_tx_timeout
,
1915 .ndo_validate_addr
= eth_validate_addr
,
1916 .ndo_vlan_rx_add_vid
= igb_vlan_rx_add_vid
,
1917 .ndo_vlan_rx_kill_vid
= igb_vlan_rx_kill_vid
,
1918 .ndo_set_vf_mac
= igb_ndo_set_vf_mac
,
1919 .ndo_set_vf_vlan
= igb_ndo_set_vf_vlan
,
1920 .ndo_set_vf_tx_rate
= igb_ndo_set_vf_bw
,
1921 .ndo_set_vf_spoofchk
= igb_ndo_set_vf_spoofchk
,
1922 .ndo_get_vf_config
= igb_ndo_get_vf_config
,
1923 #ifdef CONFIG_NET_POLL_CONTROLLER
1924 .ndo_poll_controller
= igb_netpoll
,
1926 .ndo_fix_features
= igb_fix_features
,
1927 .ndo_set_features
= igb_set_features
,
1931 * igb_set_fw_version - Configure version string for ethtool
1932 * @adapter: adapter struct
1934 void igb_set_fw_version(struct igb_adapter
*adapter
)
1936 struct e1000_hw
*hw
= &adapter
->hw
;
1937 struct e1000_fw_version fw
;
1939 igb_get_fw_version(hw
, &fw
);
1941 switch (hw
->mac
.type
) {
1944 if (!(igb_get_flash_presence_i210(hw
))) {
1945 snprintf(adapter
->fw_version
,
1946 sizeof(adapter
->fw_version
),
1948 fw
.invm_major
, fw
.invm_minor
,
1954 /* if option is rom valid, display its version too */
1956 snprintf(adapter
->fw_version
,
1957 sizeof(adapter
->fw_version
),
1958 "%d.%d, 0x%08x, %d.%d.%d",
1959 fw
.eep_major
, fw
.eep_minor
, fw
.etrack_id
,
1960 fw
.or_major
, fw
.or_build
, fw
.or_patch
);
1962 } else if (fw
.etrack_id
!= 0X0000) {
1963 snprintf(adapter
->fw_version
,
1964 sizeof(adapter
->fw_version
),
1966 fw
.eep_major
, fw
.eep_minor
, fw
.etrack_id
);
1968 snprintf(adapter
->fw_version
,
1969 sizeof(adapter
->fw_version
),
1971 fw
.eep_major
, fw
.eep_minor
, fw
.eep_build
);
1979 * igb_init_i2c - Init I2C interface
1980 * @adapter: pointer to adapter structure
1982 static s32
igb_init_i2c(struct igb_adapter
*adapter
)
1984 s32 status
= E1000_SUCCESS
;
1986 /* I2C interface supported on i350 devices */
1987 if (adapter
->hw
.mac
.type
!= e1000_i350
)
1988 return E1000_SUCCESS
;
1990 /* Initialize the i2c bus which is controlled by the registers.
1991 * This bus will use the i2c_algo_bit structue that implements
1992 * the protocol through toggling of the 4 bits in the register.
1994 adapter
->i2c_adap
.owner
= THIS_MODULE
;
1995 adapter
->i2c_algo
= igb_i2c_algo
;
1996 adapter
->i2c_algo
.data
= adapter
;
1997 adapter
->i2c_adap
.algo_data
= &adapter
->i2c_algo
;
1998 adapter
->i2c_adap
.dev
.parent
= &adapter
->pdev
->dev
;
1999 strlcpy(adapter
->i2c_adap
.name
, "igb BB",
2000 sizeof(adapter
->i2c_adap
.name
));
2001 status
= i2c_bit_add_bus(&adapter
->i2c_adap
);
2006 * igb_probe - Device Initialization Routine
2007 * @pdev: PCI device information struct
2008 * @ent: entry in igb_pci_tbl
2010 * Returns 0 on success, negative on failure
2012 * igb_probe initializes an adapter identified by a pci_dev structure.
2013 * The OS initialization, configuring of the adapter private structure,
2014 * and a hardware reset occur.
2016 static int igb_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
2018 struct net_device
*netdev
;
2019 struct igb_adapter
*adapter
;
2020 struct e1000_hw
*hw
;
2021 u16 eeprom_data
= 0;
2023 static int global_quad_port_a
; /* global quad port a indication */
2024 const struct e1000_info
*ei
= igb_info_tbl
[ent
->driver_data
];
2025 unsigned long mmio_start
, mmio_len
;
2026 int err
, pci_using_dac
;
2027 u8 part_str
[E1000_PBANUM_LENGTH
];
2029 /* Catch broken hardware that put the wrong VF device ID in
2030 * the PCIe SR-IOV capability.
2032 if (pdev
->is_virtfn
) {
2033 WARN(1, KERN_ERR
"%s (%hx:%hx) should not be a VF!\n",
2034 pci_name(pdev
), pdev
->vendor
, pdev
->device
);
2038 err
= pci_enable_device_mem(pdev
);
2043 err
= dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(64));
2047 err
= dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(32));
2050 "No usable DMA configuration, aborting\n");
2055 err
= pci_request_selected_regions(pdev
, pci_select_bars(pdev
,
2061 pci_enable_pcie_error_reporting(pdev
);
2063 pci_set_master(pdev
);
2064 pci_save_state(pdev
);
2067 netdev
= alloc_etherdev_mq(sizeof(struct igb_adapter
),
2070 goto err_alloc_etherdev
;
2072 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
2074 pci_set_drvdata(pdev
, netdev
);
2075 adapter
= netdev_priv(netdev
);
2076 adapter
->netdev
= netdev
;
2077 adapter
->pdev
= pdev
;
2080 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
2082 mmio_start
= pci_resource_start(pdev
, 0);
2083 mmio_len
= pci_resource_len(pdev
, 0);
2086 hw
->hw_addr
= ioremap(mmio_start
, mmio_len
);
2090 netdev
->netdev_ops
= &igb_netdev_ops
;
2091 igb_set_ethtool_ops(netdev
);
2092 netdev
->watchdog_timeo
= 5 * HZ
;
2094 strncpy(netdev
->name
, pci_name(pdev
), sizeof(netdev
->name
) - 1);
2096 netdev
->mem_start
= mmio_start
;
2097 netdev
->mem_end
= mmio_start
+ mmio_len
;
2099 /* PCI config space info */
2100 hw
->vendor_id
= pdev
->vendor
;
2101 hw
->device_id
= pdev
->device
;
2102 hw
->revision_id
= pdev
->revision
;
2103 hw
->subsystem_vendor_id
= pdev
->subsystem_vendor
;
2104 hw
->subsystem_device_id
= pdev
->subsystem_device
;
2106 /* Copy the default MAC, PHY and NVM function pointers */
2107 memcpy(&hw
->mac
.ops
, ei
->mac_ops
, sizeof(hw
->mac
.ops
));
2108 memcpy(&hw
->phy
.ops
, ei
->phy_ops
, sizeof(hw
->phy
.ops
));
2109 memcpy(&hw
->nvm
.ops
, ei
->nvm_ops
, sizeof(hw
->nvm
.ops
));
2110 /* Initialize skew-specific constants */
2111 err
= ei
->get_invariants(hw
);
2115 /* setup the private structure */
2116 err
= igb_sw_init(adapter
);
2120 igb_get_bus_info_pcie(hw
);
2122 hw
->phy
.autoneg_wait_to_complete
= false;
2124 /* Copper options */
2125 if (hw
->phy
.media_type
== e1000_media_type_copper
) {
2126 hw
->phy
.mdix
= AUTO_ALL_MODES
;
2127 hw
->phy
.disable_polarity_correction
= false;
2128 hw
->phy
.ms_type
= e1000_ms_hw_default
;
2131 if (igb_check_reset_block(hw
))
2132 dev_info(&pdev
->dev
,
2133 "PHY reset is blocked due to SOL/IDER session.\n");
2135 /* features is initialized to 0 in allocation, it might have bits
2136 * set by igb_sw_init so we should use an or instead of an
2139 netdev
->features
|= NETIF_F_SG
|
2146 NETIF_F_HW_VLAN_CTAG_RX
|
2147 NETIF_F_HW_VLAN_CTAG_TX
;
2149 /* copy netdev features into list of user selectable features */
2150 netdev
->hw_features
|= netdev
->features
;
2151 netdev
->hw_features
|= NETIF_F_RXALL
;
2153 /* set this bit last since it cannot be part of hw_features */
2154 netdev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
2156 netdev
->vlan_features
|= NETIF_F_TSO
|
2162 netdev
->priv_flags
|= IFF_SUPP_NOFCS
;
2164 if (pci_using_dac
) {
2165 netdev
->features
|= NETIF_F_HIGHDMA
;
2166 netdev
->vlan_features
|= NETIF_F_HIGHDMA
;
2169 if (hw
->mac
.type
>= e1000_82576
) {
2170 netdev
->hw_features
|= NETIF_F_SCTP_CSUM
;
2171 netdev
->features
|= NETIF_F_SCTP_CSUM
;
2174 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
2176 adapter
->en_mng_pt
= igb_enable_mng_pass_thru(hw
);
2178 /* before reading the NVM, reset the controller to put the device in a
2179 * known good starting state
2181 hw
->mac
.ops
.reset_hw(hw
);
2183 /* make sure the NVM is good , i211/i210 parts can have special NVM
2184 * that doesn't contain a checksum
2186 switch (hw
->mac
.type
) {
2189 if (igb_get_flash_presence_i210(hw
)) {
2190 if (hw
->nvm
.ops
.validate(hw
) < 0) {
2192 "The NVM Checksum Is Not Valid\n");
2199 if (hw
->nvm
.ops
.validate(hw
) < 0) {
2200 dev_err(&pdev
->dev
, "The NVM Checksum Is Not Valid\n");
2207 /* copy the MAC address out of the NVM */
2208 if (hw
->mac
.ops
.read_mac_addr(hw
))
2209 dev_err(&pdev
->dev
, "NVM Read Error\n");
2211 memcpy(netdev
->dev_addr
, hw
->mac
.addr
, netdev
->addr_len
);
2213 if (!is_valid_ether_addr(netdev
->dev_addr
)) {
2214 dev_err(&pdev
->dev
, "Invalid MAC Address\n");
2219 /* get firmware version for ethtool -i */
2220 igb_set_fw_version(adapter
);
2222 setup_timer(&adapter
->watchdog_timer
, igb_watchdog
,
2223 (unsigned long) adapter
);
2224 setup_timer(&adapter
->phy_info_timer
, igb_update_phy_info
,
2225 (unsigned long) adapter
);
2227 INIT_WORK(&adapter
->reset_task
, igb_reset_task
);
2228 INIT_WORK(&adapter
->watchdog_task
, igb_watchdog_task
);
2230 /* Initialize link properties that are user-changeable */
2231 adapter
->fc_autoneg
= true;
2232 hw
->mac
.autoneg
= true;
2233 hw
->phy
.autoneg_advertised
= 0x2f;
2235 hw
->fc
.requested_mode
= e1000_fc_default
;
2236 hw
->fc
.current_mode
= e1000_fc_default
;
2238 igb_validate_mdi_setting(hw
);
2240 /* By default, support wake on port A */
2241 if (hw
->bus
.func
== 0)
2242 adapter
->flags
|= IGB_FLAG_WOL_SUPPORTED
;
2244 /* Check the NVM for wake support on non-port A ports */
2245 if (hw
->mac
.type
>= e1000_82580
)
2246 hw
->nvm
.ops
.read(hw
, NVM_INIT_CONTROL3_PORT_A
+
2247 NVM_82580_LAN_FUNC_OFFSET(hw
->bus
.func
), 1,
2249 else if (hw
->bus
.func
== 1)
2250 hw
->nvm
.ops
.read(hw
, NVM_INIT_CONTROL3_PORT_B
, 1, &eeprom_data
);
2252 if (eeprom_data
& IGB_EEPROM_APME
)
2253 adapter
->flags
|= IGB_FLAG_WOL_SUPPORTED
;
2255 /* now that we have the eeprom settings, apply the special cases where
2256 * the eeprom may be wrong or the board simply won't support wake on
2257 * lan on a particular port
2259 switch (pdev
->device
) {
2260 case E1000_DEV_ID_82575GB_QUAD_COPPER
:
2261 adapter
->flags
&= ~IGB_FLAG_WOL_SUPPORTED
;
2263 case E1000_DEV_ID_82575EB_FIBER_SERDES
:
2264 case E1000_DEV_ID_82576_FIBER
:
2265 case E1000_DEV_ID_82576_SERDES
:
2266 /* Wake events only supported on port A for dual fiber
2267 * regardless of eeprom setting
2269 if (rd32(E1000_STATUS
) & E1000_STATUS_FUNC_1
)
2270 adapter
->flags
&= ~IGB_FLAG_WOL_SUPPORTED
;
2272 case E1000_DEV_ID_82576_QUAD_COPPER
:
2273 case E1000_DEV_ID_82576_QUAD_COPPER_ET2
:
2274 /* if quad port adapter, disable WoL on all but port A */
2275 if (global_quad_port_a
!= 0)
2276 adapter
->flags
&= ~IGB_FLAG_WOL_SUPPORTED
;
2278 adapter
->flags
|= IGB_FLAG_QUAD_PORT_A
;
2279 /* Reset for multiple quad port adapters */
2280 if (++global_quad_port_a
== 4)
2281 global_quad_port_a
= 0;
2284 /* If the device can't wake, don't set software support */
2285 if (!device_can_wakeup(&adapter
->pdev
->dev
))
2286 adapter
->flags
&= ~IGB_FLAG_WOL_SUPPORTED
;
2289 /* initialize the wol settings based on the eeprom settings */
2290 if (adapter
->flags
& IGB_FLAG_WOL_SUPPORTED
)
2291 adapter
->wol
|= E1000_WUFC_MAG
;
2293 /* Some vendors want WoL disabled by default, but still supported */
2294 if ((hw
->mac
.type
== e1000_i350
) &&
2295 (pdev
->subsystem_vendor
== PCI_VENDOR_ID_HP
)) {
2296 adapter
->flags
|= IGB_FLAG_WOL_SUPPORTED
;
2300 device_set_wakeup_enable(&adapter
->pdev
->dev
,
2301 adapter
->flags
& IGB_FLAG_WOL_SUPPORTED
);
2303 /* reset the hardware with the new settings */
2306 /* Init the I2C interface */
2307 err
= igb_init_i2c(adapter
);
2309 dev_err(&pdev
->dev
, "failed to init i2c interface\n");
2313 /* let the f/w know that the h/w is now under the control of the
2315 igb_get_hw_control(adapter
);
2317 strcpy(netdev
->name
, "eth%d");
2318 err
= register_netdev(netdev
);
2322 /* carrier off reporting is important to ethtool even BEFORE open */
2323 netif_carrier_off(netdev
);
2325 #ifdef CONFIG_IGB_DCA
2326 if (dca_add_requester(&pdev
->dev
) == 0) {
2327 adapter
->flags
|= IGB_FLAG_DCA_ENABLED
;
2328 dev_info(&pdev
->dev
, "DCA enabled\n");
2329 igb_setup_dca(adapter
);
2333 #ifdef CONFIG_IGB_HWMON
2334 /* Initialize the thermal sensor on i350 devices. */
2335 if (hw
->mac
.type
== e1000_i350
&& hw
->bus
.func
== 0) {
2338 /* Read the NVM to determine if this i350 device supports an
2339 * external thermal sensor.
2341 hw
->nvm
.ops
.read(hw
, NVM_ETS_CFG
, 1, &ets_word
);
2342 if (ets_word
!= 0x0000 && ets_word
!= 0xFFFF)
2343 adapter
->ets
= true;
2345 adapter
->ets
= false;
2346 if (igb_sysfs_init(adapter
))
2348 "failed to allocate sysfs resources\n");
2350 adapter
->ets
= false;
2353 /* do hw tstamp init after resetting */
2354 igb_ptp_init(adapter
);
2356 dev_info(&pdev
->dev
, "Intel(R) Gigabit Ethernet Network Connection\n");
2357 /* print bus type/speed/width info, not applicable to i354 */
2358 if (hw
->mac
.type
!= e1000_i354
) {
2359 dev_info(&pdev
->dev
, "%s: (PCIe:%s:%s) %pM\n",
2361 ((hw
->bus
.speed
== e1000_bus_speed_2500
) ? "2.5Gb/s" :
2362 (hw
->bus
.speed
== e1000_bus_speed_5000
) ? "5.0Gb/s" :
2364 ((hw
->bus
.width
== e1000_bus_width_pcie_x4
) ?
2366 (hw
->bus
.width
== e1000_bus_width_pcie_x2
) ?
2368 (hw
->bus
.width
== e1000_bus_width_pcie_x1
) ?
2369 "Width x1" : "unknown"), netdev
->dev_addr
);
2372 if ((hw
->mac
.type
>= e1000_i210
||
2373 igb_get_flash_presence_i210(hw
))) {
2374 ret_val
= igb_read_part_string(hw
, part_str
,
2375 E1000_PBANUM_LENGTH
);
2377 ret_val
= -E1000_ERR_INVM_VALUE_NOT_FOUND
;
2381 strcpy(part_str
, "Unknown");
2382 dev_info(&pdev
->dev
, "%s: PBA No: %s\n", netdev
->name
, part_str
);
2383 dev_info(&pdev
->dev
,
2384 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2385 adapter
->msix_entries
? "MSI-X" :
2386 (adapter
->flags
& IGB_FLAG_HAS_MSI
) ? "MSI" : "legacy",
2387 adapter
->num_rx_queues
, adapter
->num_tx_queues
);
2388 switch (hw
->mac
.type
) {
2392 igb_set_eee_i350(hw
);
2395 if (hw
->phy
.media_type
== e1000_media_type_copper
) {
2396 if ((rd32(E1000_CTRL_EXT
) &
2397 E1000_CTRL_EXT_LINK_MODE_SGMII
))
2398 igb_set_eee_i354(hw
);
2405 pm_runtime_put_noidle(&pdev
->dev
);
2409 igb_release_hw_control(adapter
);
2410 memset(&adapter
->i2c_adap
, 0, sizeof(adapter
->i2c_adap
));
2412 if (!igb_check_reset_block(hw
))
2415 if (hw
->flash_address
)
2416 iounmap(hw
->flash_address
);
2418 igb_clear_interrupt_scheme(adapter
);
2419 iounmap(hw
->hw_addr
);
2421 free_netdev(netdev
);
2423 pci_release_selected_regions(pdev
,
2424 pci_select_bars(pdev
, IORESOURCE_MEM
));
2427 pci_disable_device(pdev
);
2431 #ifdef CONFIG_PCI_IOV
2432 static int igb_disable_sriov(struct pci_dev
*pdev
)
2434 struct net_device
*netdev
= pci_get_drvdata(pdev
);
2435 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2436 struct e1000_hw
*hw
= &adapter
->hw
;
2438 /* reclaim resources allocated to VFs */
2439 if (adapter
->vf_data
) {
2440 /* disable iov and allow time for transactions to clear */
2441 if (pci_vfs_assigned(pdev
)) {
2442 dev_warn(&pdev
->dev
,
2443 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2446 pci_disable_sriov(pdev
);
2450 kfree(adapter
->vf_data
);
2451 adapter
->vf_data
= NULL
;
2452 adapter
->vfs_allocated_count
= 0;
2453 wr32(E1000_IOVCTL
, E1000_IOVCTL_REUSE_VFQ
);
2456 dev_info(&pdev
->dev
, "IOV Disabled\n");
2458 /* Re-enable DMA Coalescing flag since IOV is turned off */
2459 adapter
->flags
|= IGB_FLAG_DMAC
;
2465 static int igb_enable_sriov(struct pci_dev
*pdev
, int num_vfs
)
2467 struct net_device
*netdev
= pci_get_drvdata(pdev
);
2468 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2469 int old_vfs
= pci_num_vf(pdev
);
2473 if (!adapter
->msix_entries
|| num_vfs
> 7) {
2481 dev_info(&pdev
->dev
, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
2483 adapter
->vfs_allocated_count
= old_vfs
;
2485 adapter
->vfs_allocated_count
= num_vfs
;
2487 adapter
->vf_data
= kcalloc(adapter
->vfs_allocated_count
,
2488 sizeof(struct vf_data_storage
), GFP_KERNEL
);
2490 /* if allocation failed then we do not support SR-IOV */
2491 if (!adapter
->vf_data
) {
2492 adapter
->vfs_allocated_count
= 0;
2494 "Unable to allocate memory for VF Data Storage\n");
2499 /* only call pci_enable_sriov() if no VFs are allocated already */
2501 err
= pci_enable_sriov(pdev
, adapter
->vfs_allocated_count
);
2505 dev_info(&pdev
->dev
, "%d VFs allocated\n",
2506 adapter
->vfs_allocated_count
);
2507 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++)
2508 igb_vf_configure(adapter
, i
);
2510 /* DMA Coalescing is not supported in IOV mode. */
2511 adapter
->flags
&= ~IGB_FLAG_DMAC
;
2515 kfree(adapter
->vf_data
);
2516 adapter
->vf_data
= NULL
;
2517 adapter
->vfs_allocated_count
= 0;
2524 * igb_remove_i2c - Cleanup I2C interface
2525 * @adapter: pointer to adapter structure
2527 static void igb_remove_i2c(struct igb_adapter
*adapter
)
2529 /* free the adapter bus structure */
2530 i2c_del_adapter(&adapter
->i2c_adap
);
2534 * igb_remove - Device Removal Routine
2535 * @pdev: PCI device information struct
2537 * igb_remove is called by the PCI subsystem to alert the driver
2538 * that it should release a PCI device. The could be caused by a
2539 * Hot-Plug event, or because the driver is going to be removed from
2542 static void igb_remove(struct pci_dev
*pdev
)
2544 struct net_device
*netdev
= pci_get_drvdata(pdev
);
2545 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2546 struct e1000_hw
*hw
= &adapter
->hw
;
2548 pm_runtime_get_noresume(&pdev
->dev
);
2549 #ifdef CONFIG_IGB_HWMON
2550 igb_sysfs_exit(adapter
);
2552 igb_remove_i2c(adapter
);
2553 igb_ptp_stop(adapter
);
2554 /* The watchdog timer may be rescheduled, so explicitly
2555 * disable watchdog from being rescheduled.
2557 set_bit(__IGB_DOWN
, &adapter
->state
);
2558 del_timer_sync(&adapter
->watchdog_timer
);
2559 del_timer_sync(&adapter
->phy_info_timer
);
2561 cancel_work_sync(&adapter
->reset_task
);
2562 cancel_work_sync(&adapter
->watchdog_task
);
2564 #ifdef CONFIG_IGB_DCA
2565 if (adapter
->flags
& IGB_FLAG_DCA_ENABLED
) {
2566 dev_info(&pdev
->dev
, "DCA disabled\n");
2567 dca_remove_requester(&pdev
->dev
);
2568 adapter
->flags
&= ~IGB_FLAG_DCA_ENABLED
;
2569 wr32(E1000_DCA_CTRL
, E1000_DCA_CTRL_DCA_MODE_DISABLE
);
2573 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2574 * would have already happened in close and is redundant.
2576 igb_release_hw_control(adapter
);
2578 unregister_netdev(netdev
);
2580 igb_clear_interrupt_scheme(adapter
);
2582 #ifdef CONFIG_PCI_IOV
2583 igb_disable_sriov(pdev
);
2586 iounmap(hw
->hw_addr
);
2587 if (hw
->flash_address
)
2588 iounmap(hw
->flash_address
);
2589 pci_release_selected_regions(pdev
,
2590 pci_select_bars(pdev
, IORESOURCE_MEM
));
2592 kfree(adapter
->shadow_vfta
);
2593 free_netdev(netdev
);
2595 pci_disable_pcie_error_reporting(pdev
);
2597 pci_disable_device(pdev
);
2601 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2602 * @adapter: board private structure to initialize
2604 * This function initializes the vf specific data storage and then attempts to
2605 * allocate the VFs. The reason for ordering it this way is because it is much
2606 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2607 * the memory for the VFs.
2609 static void igb_probe_vfs(struct igb_adapter
*adapter
)
2611 #ifdef CONFIG_PCI_IOV
2612 struct pci_dev
*pdev
= adapter
->pdev
;
2613 struct e1000_hw
*hw
= &adapter
->hw
;
2615 /* Virtualization features not supported on i210 family. */
2616 if ((hw
->mac
.type
== e1000_i210
) || (hw
->mac
.type
== e1000_i211
))
2619 pci_sriov_set_totalvfs(pdev
, 7);
2620 igb_pci_enable_sriov(pdev
, max_vfs
);
2622 #endif /* CONFIG_PCI_IOV */
2625 static void igb_init_queue_configuration(struct igb_adapter
*adapter
)
2627 struct e1000_hw
*hw
= &adapter
->hw
;
2630 /* Determine the maximum number of RSS queues supported. */
2631 switch (hw
->mac
.type
) {
2633 max_rss_queues
= IGB_MAX_RX_QUEUES_I211
;
2637 max_rss_queues
= IGB_MAX_RX_QUEUES_82575
;
2640 /* I350 cannot do RSS and SR-IOV at the same time */
2641 if (!!adapter
->vfs_allocated_count
) {
2647 if (!!adapter
->vfs_allocated_count
) {
2655 max_rss_queues
= IGB_MAX_RX_QUEUES
;
2659 adapter
->rss_queues
= min_t(u32
, max_rss_queues
, num_online_cpus());
2661 /* Determine if we need to pair queues. */
2662 switch (hw
->mac
.type
) {
2665 /* Device supports enough interrupts without queue pairing. */
2668 /* If VFs are going to be allocated with RSS queues then we
2669 * should pair the queues in order to conserve interrupts due
2670 * to limited supply.
2672 if ((adapter
->rss_queues
> 1) &&
2673 (adapter
->vfs_allocated_count
> 6))
2674 adapter
->flags
|= IGB_FLAG_QUEUE_PAIRS
;
2681 /* If rss_queues > half of max_rss_queues, pair the queues in
2682 * order to conserve interrupts due to limited supply.
2684 if (adapter
->rss_queues
> (max_rss_queues
/ 2))
2685 adapter
->flags
|= IGB_FLAG_QUEUE_PAIRS
;
2691 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2692 * @adapter: board private structure to initialize
2694 * igb_sw_init initializes the Adapter private data structure.
2695 * Fields are initialized based on PCI device information and
2696 * OS network device settings (MTU size).
2698 static int igb_sw_init(struct igb_adapter
*adapter
)
2700 struct e1000_hw
*hw
= &adapter
->hw
;
2701 struct net_device
*netdev
= adapter
->netdev
;
2702 struct pci_dev
*pdev
= adapter
->pdev
;
2704 pci_read_config_word(pdev
, PCI_COMMAND
, &hw
->bus
.pci_cmd_word
);
2706 /* set default ring sizes */
2707 adapter
->tx_ring_count
= IGB_DEFAULT_TXD
;
2708 adapter
->rx_ring_count
= IGB_DEFAULT_RXD
;
2710 /* set default ITR values */
2711 adapter
->rx_itr_setting
= IGB_DEFAULT_ITR
;
2712 adapter
->tx_itr_setting
= IGB_DEFAULT_ITR
;
2714 /* set default work limits */
2715 adapter
->tx_work_limit
= IGB_DEFAULT_TX_WORK
;
2717 adapter
->max_frame_size
= netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+
2719 adapter
->min_frame_size
= ETH_ZLEN
+ ETH_FCS_LEN
;
2721 spin_lock_init(&adapter
->stats64_lock
);
2722 #ifdef CONFIG_PCI_IOV
2723 switch (hw
->mac
.type
) {
2727 dev_warn(&pdev
->dev
,
2728 "Maximum of 7 VFs per PF, using max\n");
2729 max_vfs
= adapter
->vfs_allocated_count
= 7;
2731 adapter
->vfs_allocated_count
= max_vfs
;
2732 if (adapter
->vfs_allocated_count
)
2733 dev_warn(&pdev
->dev
,
2734 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
2739 #endif /* CONFIG_PCI_IOV */
2741 igb_init_queue_configuration(adapter
);
2743 /* Setup and initialize a copy of the hw vlan table array */
2744 adapter
->shadow_vfta
= kcalloc(E1000_VLAN_FILTER_TBL_SIZE
, sizeof(u32
),
2747 /* This call may decrease the number of queues */
2748 if (igb_init_interrupt_scheme(adapter
, true)) {
2749 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
2753 igb_probe_vfs(adapter
);
2755 /* Explicitly disable IRQ since the NIC can be in any state. */
2756 igb_irq_disable(adapter
);
2758 if (hw
->mac
.type
>= e1000_i350
)
2759 adapter
->flags
&= ~IGB_FLAG_DMAC
;
2761 set_bit(__IGB_DOWN
, &adapter
->state
);
2766 * igb_open - Called when a network interface is made active
2767 * @netdev: network interface device structure
2769 * Returns 0 on success, negative value on failure
2771 * The open entry point is called when a network interface is made
2772 * active by the system (IFF_UP). At this point all resources needed
2773 * for transmit and receive operations are allocated, the interrupt
2774 * handler is registered with the OS, the watchdog timer is started,
2775 * and the stack is notified that the interface is ready.
2777 static int __igb_open(struct net_device
*netdev
, bool resuming
)
2779 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2780 struct e1000_hw
*hw
= &adapter
->hw
;
2781 struct pci_dev
*pdev
= adapter
->pdev
;
2785 /* disallow open during test */
2786 if (test_bit(__IGB_TESTING
, &adapter
->state
)) {
2792 pm_runtime_get_sync(&pdev
->dev
);
2794 netif_carrier_off(netdev
);
2796 /* allocate transmit descriptors */
2797 err
= igb_setup_all_tx_resources(adapter
);
2801 /* allocate receive descriptors */
2802 err
= igb_setup_all_rx_resources(adapter
);
2806 igb_power_up_link(adapter
);
2808 /* before we allocate an interrupt, we must be ready to handle it.
2809 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2810 * as soon as we call pci_request_irq, so we have to setup our
2811 * clean_rx handler before we do so.
2813 igb_configure(adapter
);
2815 err
= igb_request_irq(adapter
);
2819 /* Notify the stack of the actual queue counts. */
2820 err
= netif_set_real_num_tx_queues(adapter
->netdev
,
2821 adapter
->num_tx_queues
);
2823 goto err_set_queues
;
2825 err
= netif_set_real_num_rx_queues(adapter
->netdev
,
2826 adapter
->num_rx_queues
);
2828 goto err_set_queues
;
2830 /* From here on the code is the same as igb_up() */
2831 clear_bit(__IGB_DOWN
, &adapter
->state
);
2833 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
2834 napi_enable(&(adapter
->q_vector
[i
]->napi
));
2836 /* Clear any pending interrupts. */
2839 igb_irq_enable(adapter
);
2841 /* notify VFs that reset has been completed */
2842 if (adapter
->vfs_allocated_count
) {
2843 u32 reg_data
= rd32(E1000_CTRL_EXT
);
2844 reg_data
|= E1000_CTRL_EXT_PFRSTD
;
2845 wr32(E1000_CTRL_EXT
, reg_data
);
2848 netif_tx_start_all_queues(netdev
);
2851 pm_runtime_put(&pdev
->dev
);
2853 /* start the watchdog. */
2854 hw
->mac
.get_link_status
= 1;
2855 schedule_work(&adapter
->watchdog_task
);
2860 igb_free_irq(adapter
);
2862 igb_release_hw_control(adapter
);
2863 igb_power_down_link(adapter
);
2864 igb_free_all_rx_resources(adapter
);
2866 igb_free_all_tx_resources(adapter
);
2870 pm_runtime_put(&pdev
->dev
);
2875 static int igb_open(struct net_device
*netdev
)
2877 return __igb_open(netdev
, false);
2881 * igb_close - Disables a network interface
2882 * @netdev: network interface device structure
2884 * Returns 0, this is not allowed to fail
2886 * The close entry point is called when an interface is de-activated
2887 * by the OS. The hardware is still under the driver's control, but
2888 * needs to be disabled. A global MAC reset is issued to stop the
2889 * hardware, and all transmit and receive resources are freed.
2891 static int __igb_close(struct net_device
*netdev
, bool suspending
)
2893 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2894 struct pci_dev
*pdev
= adapter
->pdev
;
2896 WARN_ON(test_bit(__IGB_RESETTING
, &adapter
->state
));
2899 pm_runtime_get_sync(&pdev
->dev
);
2902 igb_free_irq(adapter
);
2904 igb_free_all_tx_resources(adapter
);
2905 igb_free_all_rx_resources(adapter
);
2908 pm_runtime_put_sync(&pdev
->dev
);
2912 static int igb_close(struct net_device
*netdev
)
2914 return __igb_close(netdev
, false);
2918 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2919 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2921 * Return 0 on success, negative on failure
2923 int igb_setup_tx_resources(struct igb_ring
*tx_ring
)
2925 struct device
*dev
= tx_ring
->dev
;
2928 size
= sizeof(struct igb_tx_buffer
) * tx_ring
->count
;
2930 tx_ring
->tx_buffer_info
= vzalloc(size
);
2931 if (!tx_ring
->tx_buffer_info
)
2934 /* round up to nearest 4K */
2935 tx_ring
->size
= tx_ring
->count
* sizeof(union e1000_adv_tx_desc
);
2936 tx_ring
->size
= ALIGN(tx_ring
->size
, 4096);
2938 tx_ring
->desc
= dma_alloc_coherent(dev
, tx_ring
->size
,
2939 &tx_ring
->dma
, GFP_KERNEL
);
2943 tx_ring
->next_to_use
= 0;
2944 tx_ring
->next_to_clean
= 0;
2949 vfree(tx_ring
->tx_buffer_info
);
2950 tx_ring
->tx_buffer_info
= NULL
;
2951 dev_err(dev
, "Unable to allocate memory for the Tx descriptor ring\n");
2956 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2957 * (Descriptors) for all queues
2958 * @adapter: board private structure
2960 * Return 0 on success, negative on failure
2962 static int igb_setup_all_tx_resources(struct igb_adapter
*adapter
)
2964 struct pci_dev
*pdev
= adapter
->pdev
;
2967 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
2968 err
= igb_setup_tx_resources(adapter
->tx_ring
[i
]);
2971 "Allocation for Tx Queue %u failed\n", i
);
2972 for (i
--; i
>= 0; i
--)
2973 igb_free_tx_resources(adapter
->tx_ring
[i
]);
2982 * igb_setup_tctl - configure the transmit control registers
2983 * @adapter: Board private structure
2985 void igb_setup_tctl(struct igb_adapter
*adapter
)
2987 struct e1000_hw
*hw
= &adapter
->hw
;
2990 /* disable queue 0 which is enabled by default on 82575 and 82576 */
2991 wr32(E1000_TXDCTL(0), 0);
2993 /* Program the Transmit Control Register */
2994 tctl
= rd32(E1000_TCTL
);
2995 tctl
&= ~E1000_TCTL_CT
;
2996 tctl
|= E1000_TCTL_PSP
| E1000_TCTL_RTLC
|
2997 (E1000_COLLISION_THRESHOLD
<< E1000_CT_SHIFT
);
2999 igb_config_collision_dist(hw
);
3001 /* Enable transmits */
3002 tctl
|= E1000_TCTL_EN
;
3004 wr32(E1000_TCTL
, tctl
);
3008 * igb_configure_tx_ring - Configure transmit ring after Reset
3009 * @adapter: board private structure
3010 * @ring: tx ring to configure
3012 * Configure a transmit ring after a reset.
3014 void igb_configure_tx_ring(struct igb_adapter
*adapter
,
3015 struct igb_ring
*ring
)
3017 struct e1000_hw
*hw
= &adapter
->hw
;
3019 u64 tdba
= ring
->dma
;
3020 int reg_idx
= ring
->reg_idx
;
3022 /* disable the queue */
3023 wr32(E1000_TXDCTL(reg_idx
), 0);
3027 wr32(E1000_TDLEN(reg_idx
),
3028 ring
->count
* sizeof(union e1000_adv_tx_desc
));
3029 wr32(E1000_TDBAL(reg_idx
),
3030 tdba
& 0x00000000ffffffffULL
);
3031 wr32(E1000_TDBAH(reg_idx
), tdba
>> 32);
3033 ring
->tail
= hw
->hw_addr
+ E1000_TDT(reg_idx
);
3034 wr32(E1000_TDH(reg_idx
), 0);
3035 writel(0, ring
->tail
);
3037 txdctl
|= IGB_TX_PTHRESH
;
3038 txdctl
|= IGB_TX_HTHRESH
<< 8;
3039 txdctl
|= IGB_TX_WTHRESH
<< 16;
3041 txdctl
|= E1000_TXDCTL_QUEUE_ENABLE
;
3042 wr32(E1000_TXDCTL(reg_idx
), txdctl
);
3046 * igb_configure_tx - Configure transmit Unit after Reset
3047 * @adapter: board private structure
3049 * Configure the Tx unit of the MAC after a reset.
3051 static void igb_configure_tx(struct igb_adapter
*adapter
)
3055 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3056 igb_configure_tx_ring(adapter
, adapter
->tx_ring
[i
]);
3060 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
3061 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3063 * Returns 0 on success, negative on failure
3065 int igb_setup_rx_resources(struct igb_ring
*rx_ring
)
3067 struct device
*dev
= rx_ring
->dev
;
3070 size
= sizeof(struct igb_rx_buffer
) * rx_ring
->count
;
3072 rx_ring
->rx_buffer_info
= vzalloc(size
);
3073 if (!rx_ring
->rx_buffer_info
)
3076 /* Round up to nearest 4K */
3077 rx_ring
->size
= rx_ring
->count
* sizeof(union e1000_adv_rx_desc
);
3078 rx_ring
->size
= ALIGN(rx_ring
->size
, 4096);
3080 rx_ring
->desc
= dma_alloc_coherent(dev
, rx_ring
->size
,
3081 &rx_ring
->dma
, GFP_KERNEL
);
3085 rx_ring
->next_to_alloc
= 0;
3086 rx_ring
->next_to_clean
= 0;
3087 rx_ring
->next_to_use
= 0;
3092 vfree(rx_ring
->rx_buffer_info
);
3093 rx_ring
->rx_buffer_info
= NULL
;
3094 dev_err(dev
, "Unable to allocate memory for the Rx descriptor ring\n");
3099 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
3100 * (Descriptors) for all queues
3101 * @adapter: board private structure
3103 * Return 0 on success, negative on failure
3105 static int igb_setup_all_rx_resources(struct igb_adapter
*adapter
)
3107 struct pci_dev
*pdev
= adapter
->pdev
;
3110 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3111 err
= igb_setup_rx_resources(adapter
->rx_ring
[i
]);
3114 "Allocation for Rx Queue %u failed\n", i
);
3115 for (i
--; i
>= 0; i
--)
3116 igb_free_rx_resources(adapter
->rx_ring
[i
]);
3125 * igb_setup_mrqc - configure the multiple receive queue control registers
3126 * @adapter: Board private structure
3128 static void igb_setup_mrqc(struct igb_adapter
*adapter
)
3130 struct e1000_hw
*hw
= &adapter
->hw
;
3132 u32 j
, num_rx_queues
;
3133 static const u32 rsskey
[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
3134 0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
3135 0xA32DCB77, 0x0CF23080, 0x3BB7426A,
3138 /* Fill out hash function seeds */
3139 for (j
= 0; j
< 10; j
++)
3140 wr32(E1000_RSSRK(j
), rsskey
[j
]);
3142 num_rx_queues
= adapter
->rss_queues
;
3144 switch (hw
->mac
.type
) {
3146 /* 82576 supports 2 RSS queues for SR-IOV */
3147 if (adapter
->vfs_allocated_count
)
3154 if (adapter
->rss_indir_tbl_init
!= num_rx_queues
) {
3155 for (j
= 0; j
< IGB_RETA_SIZE
; j
++)
3156 adapter
->rss_indir_tbl
[j
] = (j
* num_rx_queues
) / IGB_RETA_SIZE
;
3157 adapter
->rss_indir_tbl_init
= num_rx_queues
;
3159 igb_write_rss_indir_tbl(adapter
);
3161 /* Disable raw packet checksumming so that RSS hash is placed in
3162 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
3163 * offloads as they are enabled by default
3165 rxcsum
= rd32(E1000_RXCSUM
);
3166 rxcsum
|= E1000_RXCSUM_PCSD
;
3168 if (adapter
->hw
.mac
.type
>= e1000_82576
)
3169 /* Enable Receive Checksum Offload for SCTP */
3170 rxcsum
|= E1000_RXCSUM_CRCOFL
;
3172 /* Don't need to set TUOFL or IPOFL, they default to 1 */
3173 wr32(E1000_RXCSUM
, rxcsum
);
3175 /* Generate RSS hash based on packet types, TCP/UDP
3176 * port numbers and/or IPv4/v6 src and dst addresses
3178 mrqc
= E1000_MRQC_RSS_FIELD_IPV4
|
3179 E1000_MRQC_RSS_FIELD_IPV4_TCP
|
3180 E1000_MRQC_RSS_FIELD_IPV6
|
3181 E1000_MRQC_RSS_FIELD_IPV6_TCP
|
3182 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX
;
3184 if (adapter
->flags
& IGB_FLAG_RSS_FIELD_IPV4_UDP
)
3185 mrqc
|= E1000_MRQC_RSS_FIELD_IPV4_UDP
;
3186 if (adapter
->flags
& IGB_FLAG_RSS_FIELD_IPV6_UDP
)
3187 mrqc
|= E1000_MRQC_RSS_FIELD_IPV6_UDP
;
3189 /* If VMDq is enabled then we set the appropriate mode for that, else
3190 * we default to RSS so that an RSS hash is calculated per packet even
3191 * if we are only using one queue
3193 if (adapter
->vfs_allocated_count
) {
3194 if (hw
->mac
.type
> e1000_82575
) {
3195 /* Set the default pool for the PF's first queue */
3196 u32 vtctl
= rd32(E1000_VT_CTL
);
3197 vtctl
&= ~(E1000_VT_CTL_DEFAULT_POOL_MASK
|
3198 E1000_VT_CTL_DISABLE_DEF_POOL
);
3199 vtctl
|= adapter
->vfs_allocated_count
<<
3200 E1000_VT_CTL_DEFAULT_POOL_SHIFT
;
3201 wr32(E1000_VT_CTL
, vtctl
);
3203 if (adapter
->rss_queues
> 1)
3204 mrqc
|= E1000_MRQC_ENABLE_VMDQ_RSS_2Q
;
3206 mrqc
|= E1000_MRQC_ENABLE_VMDQ
;
3208 if (hw
->mac
.type
!= e1000_i211
)
3209 mrqc
|= E1000_MRQC_ENABLE_RSS_4Q
;
3211 igb_vmm_control(adapter
);
3213 wr32(E1000_MRQC
, mrqc
);
3217 * igb_setup_rctl - configure the receive control registers
3218 * @adapter: Board private structure
3220 void igb_setup_rctl(struct igb_adapter
*adapter
)
3222 struct e1000_hw
*hw
= &adapter
->hw
;
3225 rctl
= rd32(E1000_RCTL
);
3227 rctl
&= ~(3 << E1000_RCTL_MO_SHIFT
);
3228 rctl
&= ~(E1000_RCTL_LBM_TCVR
| E1000_RCTL_LBM_MAC
);
3230 rctl
|= E1000_RCTL_EN
| E1000_RCTL_BAM
| E1000_RCTL_RDMTS_HALF
|
3231 (hw
->mac
.mc_filter_type
<< E1000_RCTL_MO_SHIFT
);
3233 /* enable stripping of CRC. It's unlikely this will break BMC
3234 * redirection as it did with e1000. Newer features require
3235 * that the HW strips the CRC.
3237 rctl
|= E1000_RCTL_SECRC
;
3239 /* disable store bad packets and clear size bits. */
3240 rctl
&= ~(E1000_RCTL_SBP
| E1000_RCTL_SZ_256
);
3242 /* enable LPE to prevent packets larger than max_frame_size */
3243 rctl
|= E1000_RCTL_LPE
;
3245 /* disable queue 0 to prevent tail write w/o re-config */
3246 wr32(E1000_RXDCTL(0), 0);
3248 /* Attention!!! For SR-IOV PF driver operations you must enable
3249 * queue drop for all VF and PF queues to prevent head of line blocking
3250 * if an un-trusted VF does not provide descriptors to hardware.
3252 if (adapter
->vfs_allocated_count
) {
3253 /* set all queue drop enable bits */
3254 wr32(E1000_QDE
, ALL_QUEUES
);
3257 /* This is useful for sniffing bad packets. */
3258 if (adapter
->netdev
->features
& NETIF_F_RXALL
) {
3259 /* UPE and MPE will be handled by normal PROMISC logic
3260 * in e1000e_set_rx_mode
3262 rctl
|= (E1000_RCTL_SBP
| /* Receive bad packets */
3263 E1000_RCTL_BAM
| /* RX All Bcast Pkts */
3264 E1000_RCTL_PMCF
); /* RX All MAC Ctrl Pkts */
3266 rctl
&= ~(E1000_RCTL_VFE
| /* Disable VLAN filter */
3267 E1000_RCTL_DPF
| /* Allow filtered pause */
3268 E1000_RCTL_CFIEN
); /* Dis VLAN CFIEN Filter */
3269 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3270 * and that breaks VLANs.
3274 wr32(E1000_RCTL
, rctl
);
3277 static inline int igb_set_vf_rlpml(struct igb_adapter
*adapter
, int size
,
3280 struct e1000_hw
*hw
= &adapter
->hw
;
3283 /* if it isn't the PF check to see if VFs are enabled and
3284 * increase the size to support vlan tags
3286 if (vfn
< adapter
->vfs_allocated_count
&&
3287 adapter
->vf_data
[vfn
].vlans_enabled
)
3288 size
+= VLAN_TAG_SIZE
;
3290 vmolr
= rd32(E1000_VMOLR(vfn
));
3291 vmolr
&= ~E1000_VMOLR_RLPML_MASK
;
3292 vmolr
|= size
| E1000_VMOLR_LPE
;
3293 wr32(E1000_VMOLR(vfn
), vmolr
);
3299 * igb_rlpml_set - set maximum receive packet size
3300 * @adapter: board private structure
3302 * Configure maximum receivable packet size.
3304 static void igb_rlpml_set(struct igb_adapter
*adapter
)
3306 u32 max_frame_size
= adapter
->max_frame_size
;
3307 struct e1000_hw
*hw
= &adapter
->hw
;
3308 u16 pf_id
= adapter
->vfs_allocated_count
;
3311 igb_set_vf_rlpml(adapter
, max_frame_size
, pf_id
);
3312 /* If we're in VMDQ or SR-IOV mode, then set global RLPML
3313 * to our max jumbo frame size, in case we need to enable
3314 * jumbo frames on one of the rings later.
3315 * This will not pass over-length frames into the default
3316 * queue because it's gated by the VMOLR.RLPML.
3318 max_frame_size
= MAX_JUMBO_FRAME_SIZE
;
3321 wr32(E1000_RLPML
, max_frame_size
);
3324 static inline void igb_set_vmolr(struct igb_adapter
*adapter
,
3327 struct e1000_hw
*hw
= &adapter
->hw
;
3330 /* This register exists only on 82576 and newer so if we are older then
3331 * we should exit and do nothing
3333 if (hw
->mac
.type
< e1000_82576
)
3336 vmolr
= rd32(E1000_VMOLR(vfn
));
3337 vmolr
|= E1000_VMOLR_STRVLAN
; /* Strip vlan tags */
3339 vmolr
|= E1000_VMOLR_AUPE
; /* Accept untagged packets */
3341 vmolr
&= ~(E1000_VMOLR_AUPE
); /* Tagged packets ONLY */
3343 /* clear all bits that might not be set */
3344 vmolr
&= ~(E1000_VMOLR_BAM
| E1000_VMOLR_RSSE
);
3346 if (adapter
->rss_queues
> 1 && vfn
== adapter
->vfs_allocated_count
)
3347 vmolr
|= E1000_VMOLR_RSSE
; /* enable RSS */
3348 /* for VMDq only allow the VFs and pool 0 to accept broadcast and
3351 if (vfn
<= adapter
->vfs_allocated_count
)
3352 vmolr
|= E1000_VMOLR_BAM
; /* Accept broadcast */
3354 wr32(E1000_VMOLR(vfn
), vmolr
);
3358 * igb_configure_rx_ring - Configure a receive ring after Reset
3359 * @adapter: board private structure
3360 * @ring: receive ring to be configured
3362 * Configure the Rx unit of the MAC after a reset.
3364 void igb_configure_rx_ring(struct igb_adapter
*adapter
,
3365 struct igb_ring
*ring
)
3367 struct e1000_hw
*hw
= &adapter
->hw
;
3368 u64 rdba
= ring
->dma
;
3369 int reg_idx
= ring
->reg_idx
;
3370 u32 srrctl
= 0, rxdctl
= 0;
3372 /* disable the queue */
3373 wr32(E1000_RXDCTL(reg_idx
), 0);
3375 /* Set DMA base address registers */
3376 wr32(E1000_RDBAL(reg_idx
),
3377 rdba
& 0x00000000ffffffffULL
);
3378 wr32(E1000_RDBAH(reg_idx
), rdba
>> 32);
3379 wr32(E1000_RDLEN(reg_idx
),
3380 ring
->count
* sizeof(union e1000_adv_rx_desc
));
3382 /* initialize head and tail */
3383 ring
->tail
= hw
->hw_addr
+ E1000_RDT(reg_idx
);
3384 wr32(E1000_RDH(reg_idx
), 0);
3385 writel(0, ring
->tail
);
3387 /* set descriptor configuration */
3388 srrctl
= IGB_RX_HDR_LEN
<< E1000_SRRCTL_BSIZEHDRSIZE_SHIFT
;
3389 srrctl
|= IGB_RX_BUFSZ
>> E1000_SRRCTL_BSIZEPKT_SHIFT
;
3390 srrctl
|= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF
;
3391 if (hw
->mac
.type
>= e1000_82580
)
3392 srrctl
|= E1000_SRRCTL_TIMESTAMP
;
3393 /* Only set Drop Enable if we are supporting multiple queues */
3394 if (adapter
->vfs_allocated_count
|| adapter
->num_rx_queues
> 1)
3395 srrctl
|= E1000_SRRCTL_DROP_EN
;
3397 wr32(E1000_SRRCTL(reg_idx
), srrctl
);
3399 /* set filtering for VMDQ pools */
3400 igb_set_vmolr(adapter
, reg_idx
& 0x7, true);
3402 rxdctl
|= IGB_RX_PTHRESH
;
3403 rxdctl
|= IGB_RX_HTHRESH
<< 8;
3404 rxdctl
|= IGB_RX_WTHRESH
<< 16;
3406 /* enable receive descriptor fetching */
3407 rxdctl
|= E1000_RXDCTL_QUEUE_ENABLE
;
3408 wr32(E1000_RXDCTL(reg_idx
), rxdctl
);
3412 * igb_configure_rx - Configure receive Unit after Reset
3413 * @adapter: board private structure
3415 * Configure the Rx unit of the MAC after a reset.
3417 static void igb_configure_rx(struct igb_adapter
*adapter
)
3421 /* set UTA to appropriate mode */
3422 igb_set_uta(adapter
);
3424 /* set the correct pool for the PF default MAC address in entry 0 */
3425 igb_rar_set_qsel(adapter
, adapter
->hw
.mac
.addr
, 0,
3426 adapter
->vfs_allocated_count
);
3428 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3429 * the Base and Length of the Rx Descriptor Ring
3431 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3432 igb_configure_rx_ring(adapter
, adapter
->rx_ring
[i
]);
3436 * igb_free_tx_resources - Free Tx Resources per Queue
3437 * @tx_ring: Tx descriptor ring for a specific queue
3439 * Free all transmit software resources
3441 void igb_free_tx_resources(struct igb_ring
*tx_ring
)
3443 igb_clean_tx_ring(tx_ring
);
3445 vfree(tx_ring
->tx_buffer_info
);
3446 tx_ring
->tx_buffer_info
= NULL
;
3448 /* if not set, then don't free */
3452 dma_free_coherent(tx_ring
->dev
, tx_ring
->size
,
3453 tx_ring
->desc
, tx_ring
->dma
);
3455 tx_ring
->desc
= NULL
;
3459 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3460 * @adapter: board private structure
3462 * Free all transmit software resources
3464 static void igb_free_all_tx_resources(struct igb_adapter
*adapter
)
3468 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3469 igb_free_tx_resources(adapter
->tx_ring
[i
]);
3472 void igb_unmap_and_free_tx_resource(struct igb_ring
*ring
,
3473 struct igb_tx_buffer
*tx_buffer
)
3475 if (tx_buffer
->skb
) {
3476 dev_kfree_skb_any(tx_buffer
->skb
);
3477 if (dma_unmap_len(tx_buffer
, len
))
3478 dma_unmap_single(ring
->dev
,
3479 dma_unmap_addr(tx_buffer
, dma
),
3480 dma_unmap_len(tx_buffer
, len
),
3482 } else if (dma_unmap_len(tx_buffer
, len
)) {
3483 dma_unmap_page(ring
->dev
,
3484 dma_unmap_addr(tx_buffer
, dma
),
3485 dma_unmap_len(tx_buffer
, len
),
3488 tx_buffer
->next_to_watch
= NULL
;
3489 tx_buffer
->skb
= NULL
;
3490 dma_unmap_len_set(tx_buffer
, len
, 0);
3491 /* buffer_info must be completely set up in the transmit path */
3495 * igb_clean_tx_ring - Free Tx Buffers
3496 * @tx_ring: ring to be cleaned
3498 static void igb_clean_tx_ring(struct igb_ring
*tx_ring
)
3500 struct igb_tx_buffer
*buffer_info
;
3504 if (!tx_ring
->tx_buffer_info
)
3506 /* Free all the Tx ring sk_buffs */
3508 for (i
= 0; i
< tx_ring
->count
; i
++) {
3509 buffer_info
= &tx_ring
->tx_buffer_info
[i
];
3510 igb_unmap_and_free_tx_resource(tx_ring
, buffer_info
);
3513 netdev_tx_reset_queue(txring_txq(tx_ring
));
3515 size
= sizeof(struct igb_tx_buffer
) * tx_ring
->count
;
3516 memset(tx_ring
->tx_buffer_info
, 0, size
);
3518 /* Zero out the descriptor ring */
3519 memset(tx_ring
->desc
, 0, tx_ring
->size
);
3521 tx_ring
->next_to_use
= 0;
3522 tx_ring
->next_to_clean
= 0;
3526 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3527 * @adapter: board private structure
3529 static void igb_clean_all_tx_rings(struct igb_adapter
*adapter
)
3533 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3534 igb_clean_tx_ring(adapter
->tx_ring
[i
]);
3538 * igb_free_rx_resources - Free Rx Resources
3539 * @rx_ring: ring to clean the resources from
3541 * Free all receive software resources
3543 void igb_free_rx_resources(struct igb_ring
*rx_ring
)
3545 igb_clean_rx_ring(rx_ring
);
3547 vfree(rx_ring
->rx_buffer_info
);
3548 rx_ring
->rx_buffer_info
= NULL
;
3550 /* if not set, then don't free */
3554 dma_free_coherent(rx_ring
->dev
, rx_ring
->size
,
3555 rx_ring
->desc
, rx_ring
->dma
);
3557 rx_ring
->desc
= NULL
;
3561 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3562 * @adapter: board private structure
3564 * Free all receive software resources
3566 static void igb_free_all_rx_resources(struct igb_adapter
*adapter
)
3570 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3571 igb_free_rx_resources(adapter
->rx_ring
[i
]);
3575 * igb_clean_rx_ring - Free Rx Buffers per Queue
3576 * @rx_ring: ring to free buffers from
3578 static void igb_clean_rx_ring(struct igb_ring
*rx_ring
)
3584 dev_kfree_skb(rx_ring
->skb
);
3585 rx_ring
->skb
= NULL
;
3587 if (!rx_ring
->rx_buffer_info
)
3590 /* Free all the Rx ring sk_buffs */
3591 for (i
= 0; i
< rx_ring
->count
; i
++) {
3592 struct igb_rx_buffer
*buffer_info
= &rx_ring
->rx_buffer_info
[i
];
3594 if (!buffer_info
->page
)
3597 dma_unmap_page(rx_ring
->dev
,
3601 __free_page(buffer_info
->page
);
3603 buffer_info
->page
= NULL
;
3606 size
= sizeof(struct igb_rx_buffer
) * rx_ring
->count
;
3607 memset(rx_ring
->rx_buffer_info
, 0, size
);
3609 /* Zero out the descriptor ring */
3610 memset(rx_ring
->desc
, 0, rx_ring
->size
);
3612 rx_ring
->next_to_alloc
= 0;
3613 rx_ring
->next_to_clean
= 0;
3614 rx_ring
->next_to_use
= 0;
3618 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3619 * @adapter: board private structure
3621 static void igb_clean_all_rx_rings(struct igb_adapter
*adapter
)
3625 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3626 igb_clean_rx_ring(adapter
->rx_ring
[i
]);
3630 * igb_set_mac - Change the Ethernet Address of the NIC
3631 * @netdev: network interface device structure
3632 * @p: pointer to an address structure
3634 * Returns 0 on success, negative on failure
3636 static int igb_set_mac(struct net_device
*netdev
, void *p
)
3638 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3639 struct e1000_hw
*hw
= &adapter
->hw
;
3640 struct sockaddr
*addr
= p
;
3642 if (!is_valid_ether_addr(addr
->sa_data
))
3643 return -EADDRNOTAVAIL
;
3645 memcpy(netdev
->dev_addr
, addr
->sa_data
, netdev
->addr_len
);
3646 memcpy(hw
->mac
.addr
, addr
->sa_data
, netdev
->addr_len
);
3648 /* set the correct pool for the new PF MAC address in entry 0 */
3649 igb_rar_set_qsel(adapter
, hw
->mac
.addr
, 0,
3650 adapter
->vfs_allocated_count
);
3656 * igb_write_mc_addr_list - write multicast addresses to MTA
3657 * @netdev: network interface device structure
3659 * Writes multicast address list to the MTA hash table.
3660 * Returns: -ENOMEM on failure
3661 * 0 on no addresses written
3662 * X on writing X addresses to MTA
3664 static int igb_write_mc_addr_list(struct net_device
*netdev
)
3666 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3667 struct e1000_hw
*hw
= &adapter
->hw
;
3668 struct netdev_hw_addr
*ha
;
3672 if (netdev_mc_empty(netdev
)) {
3673 /* nothing to program, so clear mc list */
3674 igb_update_mc_addr_list(hw
, NULL
, 0);
3675 igb_restore_vf_multicasts(adapter
);
3679 mta_list
= kzalloc(netdev_mc_count(netdev
) * 6, GFP_ATOMIC
);
3683 /* The shared function expects a packed array of only addresses. */
3685 netdev_for_each_mc_addr(ha
, netdev
)
3686 memcpy(mta_list
+ (i
++ * ETH_ALEN
), ha
->addr
, ETH_ALEN
);
3688 igb_update_mc_addr_list(hw
, mta_list
, i
);
3691 return netdev_mc_count(netdev
);
3695 * igb_write_uc_addr_list - write unicast addresses to RAR table
3696 * @netdev: network interface device structure
3698 * Writes unicast address list to the RAR table.
3699 * Returns: -ENOMEM on failure/insufficient address space
3700 * 0 on no addresses written
3701 * X on writing X addresses to the RAR table
3703 static int igb_write_uc_addr_list(struct net_device
*netdev
)
3705 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3706 struct e1000_hw
*hw
= &adapter
->hw
;
3707 unsigned int vfn
= adapter
->vfs_allocated_count
;
3708 unsigned int rar_entries
= hw
->mac
.rar_entry_count
- (vfn
+ 1);
3711 /* return ENOMEM indicating insufficient memory for addresses */
3712 if (netdev_uc_count(netdev
) > rar_entries
)
3715 if (!netdev_uc_empty(netdev
) && rar_entries
) {
3716 struct netdev_hw_addr
*ha
;
3718 netdev_for_each_uc_addr(ha
, netdev
) {
3721 igb_rar_set_qsel(adapter
, ha
->addr
,
3727 /* write the addresses in reverse order to avoid write combining */
3728 for (; rar_entries
> 0 ; rar_entries
--) {
3729 wr32(E1000_RAH(rar_entries
), 0);
3730 wr32(E1000_RAL(rar_entries
), 0);
3738 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3739 * @netdev: network interface device structure
3741 * The set_rx_mode entry point is called whenever the unicast or multicast
3742 * address lists or the network interface flags are updated. This routine is
3743 * responsible for configuring the hardware for proper unicast, multicast,
3744 * promiscuous mode, and all-multi behavior.
3746 static void igb_set_rx_mode(struct net_device
*netdev
)
3748 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3749 struct e1000_hw
*hw
= &adapter
->hw
;
3750 unsigned int vfn
= adapter
->vfs_allocated_count
;
3751 u32 rctl
, vmolr
= 0;
3754 /* Check for Promiscuous and All Multicast modes */
3755 rctl
= rd32(E1000_RCTL
);
3757 /* clear the effected bits */
3758 rctl
&= ~(E1000_RCTL_UPE
| E1000_RCTL_MPE
| E1000_RCTL_VFE
);
3760 if (netdev
->flags
& IFF_PROMISC
) {
3761 /* retain VLAN HW filtering if in VT mode */
3762 if (adapter
->vfs_allocated_count
)
3763 rctl
|= E1000_RCTL_VFE
;
3764 rctl
|= (E1000_RCTL_UPE
| E1000_RCTL_MPE
);
3765 vmolr
|= (E1000_VMOLR_ROPE
| E1000_VMOLR_MPME
);
3767 if (netdev
->flags
& IFF_ALLMULTI
) {
3768 rctl
|= E1000_RCTL_MPE
;
3769 vmolr
|= E1000_VMOLR_MPME
;
3771 /* Write addresses to the MTA, if the attempt fails
3772 * then we should just turn on promiscuous mode so
3773 * that we can at least receive multicast traffic
3775 count
= igb_write_mc_addr_list(netdev
);
3777 rctl
|= E1000_RCTL_MPE
;
3778 vmolr
|= E1000_VMOLR_MPME
;
3780 vmolr
|= E1000_VMOLR_ROMPE
;
3783 /* Write addresses to available RAR registers, if there is not
3784 * sufficient space to store all the addresses then enable
3785 * unicast promiscuous mode
3787 count
= igb_write_uc_addr_list(netdev
);
3789 rctl
|= E1000_RCTL_UPE
;
3790 vmolr
|= E1000_VMOLR_ROPE
;
3792 rctl
|= E1000_RCTL_VFE
;
3794 wr32(E1000_RCTL
, rctl
);
3796 /* In order to support SR-IOV and eventually VMDq it is necessary to set
3797 * the VMOLR to enable the appropriate modes. Without this workaround
3798 * we will have issues with VLAN tag stripping not being done for frames
3799 * that are only arriving because we are the default pool
3801 if ((hw
->mac
.type
< e1000_82576
) || (hw
->mac
.type
> e1000_i350
))
3804 vmolr
|= rd32(E1000_VMOLR(vfn
)) &
3805 ~(E1000_VMOLR_ROPE
| E1000_VMOLR_MPME
| E1000_VMOLR_ROMPE
);
3806 wr32(E1000_VMOLR(vfn
), vmolr
);
3807 igb_restore_vf_multicasts(adapter
);
3810 static void igb_check_wvbr(struct igb_adapter
*adapter
)
3812 struct e1000_hw
*hw
= &adapter
->hw
;
3815 switch (hw
->mac
.type
) {
3818 if (!(wvbr
= rd32(E1000_WVBR
)))
3825 adapter
->wvbr
|= wvbr
;
3828 #define IGB_STAGGERED_QUEUE_OFFSET 8
3830 static void igb_spoof_check(struct igb_adapter
*adapter
)
3837 for(j
= 0; j
< adapter
->vfs_allocated_count
; j
++) {
3838 if (adapter
->wvbr
& (1 << j
) ||
3839 adapter
->wvbr
& (1 << (j
+ IGB_STAGGERED_QUEUE_OFFSET
))) {
3840 dev_warn(&adapter
->pdev
->dev
,
3841 "Spoof event(s) detected on VF %d\n", j
);
3844 (1 << (j
+ IGB_STAGGERED_QUEUE_OFFSET
)));
3849 /* Need to wait a few seconds after link up to get diagnostic information from
3852 static void igb_update_phy_info(unsigned long data
)
3854 struct igb_adapter
*adapter
= (struct igb_adapter
*) data
;
3855 igb_get_phy_info(&adapter
->hw
);
3859 * igb_has_link - check shared code for link and determine up/down
3860 * @adapter: pointer to driver private info
3862 bool igb_has_link(struct igb_adapter
*adapter
)
3864 struct e1000_hw
*hw
= &adapter
->hw
;
3865 bool link_active
= false;
3867 /* get_link_status is set on LSC (link status) interrupt or
3868 * rx sequence error interrupt. get_link_status will stay
3869 * false until the e1000_check_for_link establishes link
3870 * for copper adapters ONLY
3872 switch (hw
->phy
.media_type
) {
3873 case e1000_media_type_copper
:
3874 if (!hw
->mac
.get_link_status
)
3876 case e1000_media_type_internal_serdes
:
3877 hw
->mac
.ops
.check_for_link(hw
);
3878 link_active
= !hw
->mac
.get_link_status
;
3881 case e1000_media_type_unknown
:
3885 if (((hw
->mac
.type
== e1000_i210
) ||
3886 (hw
->mac
.type
== e1000_i211
)) &&
3887 (hw
->phy
.id
== I210_I_PHY_ID
)) {
3888 if (!netif_carrier_ok(adapter
->netdev
)) {
3889 adapter
->flags
&= ~IGB_FLAG_NEED_LINK_UPDATE
;
3890 } else if (!(adapter
->flags
& IGB_FLAG_NEED_LINK_UPDATE
)) {
3891 adapter
->flags
|= IGB_FLAG_NEED_LINK_UPDATE
;
3892 adapter
->link_check_timeout
= jiffies
;
3899 static bool igb_thermal_sensor_event(struct e1000_hw
*hw
, u32 event
)
3902 u32 ctrl_ext
, thstat
;
3904 /* check for thermal sensor event on i350 copper only */
3905 if (hw
->mac
.type
== e1000_i350
) {
3906 thstat
= rd32(E1000_THSTAT
);
3907 ctrl_ext
= rd32(E1000_CTRL_EXT
);
3909 if ((hw
->phy
.media_type
== e1000_media_type_copper
) &&
3910 !(ctrl_ext
& E1000_CTRL_EXT_LINK_MODE_SGMII
))
3911 ret
= !!(thstat
& event
);
3918 * igb_watchdog - Timer Call-back
3919 * @data: pointer to adapter cast into an unsigned long
3921 static void igb_watchdog(unsigned long data
)
3923 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
3924 /* Do the rest outside of interrupt context */
3925 schedule_work(&adapter
->watchdog_task
);
3928 static void igb_watchdog_task(struct work_struct
*work
)
3930 struct igb_adapter
*adapter
= container_of(work
,
3933 struct e1000_hw
*hw
= &adapter
->hw
;
3934 struct e1000_phy_info
*phy
= &hw
->phy
;
3935 struct net_device
*netdev
= adapter
->netdev
;
3939 link
= igb_has_link(adapter
);
3941 if (adapter
->flags
& IGB_FLAG_NEED_LINK_UPDATE
) {
3942 if (time_after(jiffies
, (adapter
->link_check_timeout
+ HZ
)))
3943 adapter
->flags
&= ~IGB_FLAG_NEED_LINK_UPDATE
;
3949 /* Perform a reset if the media type changed. */
3950 if (hw
->dev_spec
._82575
.media_changed
) {
3951 hw
->dev_spec
._82575
.media_changed
= false;
3952 adapter
->flags
|= IGB_FLAG_MEDIA_RESET
;
3955 /* Cancel scheduled suspend requests. */
3956 pm_runtime_resume(netdev
->dev
.parent
);
3958 if (!netif_carrier_ok(netdev
)) {
3960 hw
->mac
.ops
.get_speed_and_duplex(hw
,
3961 &adapter
->link_speed
,
3962 &adapter
->link_duplex
);
3964 ctrl
= rd32(E1000_CTRL
);
3965 /* Links status message must follow this format */
3966 printk(KERN_INFO
"igb: %s NIC Link is Up %d Mbps %s "
3967 "Duplex, Flow Control: %s\n",
3969 adapter
->link_speed
,
3970 adapter
->link_duplex
== FULL_DUPLEX
?
3972 (ctrl
& E1000_CTRL_TFCE
) &&
3973 (ctrl
& E1000_CTRL_RFCE
) ? "RX/TX" :
3974 (ctrl
& E1000_CTRL_RFCE
) ? "RX" :
3975 (ctrl
& E1000_CTRL_TFCE
) ? "TX" : "None");
3977 /* check if SmartSpeed worked */
3978 igb_check_downshift(hw
);
3979 if (phy
->speed_downgraded
)
3980 netdev_warn(netdev
, "Link Speed was downgraded by SmartSpeed\n");
3982 /* check for thermal sensor event */
3983 if (igb_thermal_sensor_event(hw
,
3984 E1000_THSTAT_LINK_THROTTLE
)) {
3985 netdev_info(netdev
, "The network adapter link "
3986 "speed was downshifted because it "
3990 /* adjust timeout factor according to speed/duplex */
3991 adapter
->tx_timeout_factor
= 1;
3992 switch (adapter
->link_speed
) {
3994 adapter
->tx_timeout_factor
= 14;
3997 /* maybe add some timeout factor ? */
4001 netif_carrier_on(netdev
);
4003 igb_ping_all_vfs(adapter
);
4004 igb_check_vf_rate_limit(adapter
);
4006 /* link state has changed, schedule phy info update */
4007 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
4008 mod_timer(&adapter
->phy_info_timer
,
4009 round_jiffies(jiffies
+ 2 * HZ
));
4012 if (netif_carrier_ok(netdev
)) {
4013 adapter
->link_speed
= 0;
4014 adapter
->link_duplex
= 0;
4016 /* check for thermal sensor event */
4017 if (igb_thermal_sensor_event(hw
,
4018 E1000_THSTAT_PWR_DOWN
)) {
4019 netdev_err(netdev
, "The network adapter was "
4020 "stopped because it overheated\n");
4023 /* Links status message must follow this format */
4024 printk(KERN_INFO
"igb: %s NIC Link is Down\n",
4026 netif_carrier_off(netdev
);
4028 igb_ping_all_vfs(adapter
);
4030 /* link state has changed, schedule phy info update */
4031 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
4032 mod_timer(&adapter
->phy_info_timer
,
4033 round_jiffies(jiffies
+ 2 * HZ
));
4035 pm_schedule_suspend(netdev
->dev
.parent
,
4040 spin_lock(&adapter
->stats64_lock
);
4041 igb_update_stats(adapter
, &adapter
->stats64
);
4042 spin_unlock(&adapter
->stats64_lock
);
4044 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
4045 struct igb_ring
*tx_ring
= adapter
->tx_ring
[i
];
4046 if (!netif_carrier_ok(netdev
)) {
4047 /* We've lost link, so the controller stops DMA,
4048 * but we've got queued Tx work that's never going
4049 * to get done, so reset controller to flush Tx.
4050 * (Do the reset outside of interrupt context).
4052 if (igb_desc_unused(tx_ring
) + 1 < tx_ring
->count
) {
4053 adapter
->tx_timeout_count
++;
4054 schedule_work(&adapter
->reset_task
);
4055 /* return immediately since reset is imminent */
4060 /* Force detection of hung controller every watchdog period */
4061 set_bit(IGB_RING_FLAG_TX_DETECT_HANG
, &tx_ring
->flags
);
4064 /* Cause software interrupt to ensure Rx ring is cleaned */
4065 if (adapter
->msix_entries
) {
4067 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
4068 eics
|= adapter
->q_vector
[i
]->eims_value
;
4069 wr32(E1000_EICS
, eics
);
4071 wr32(E1000_ICS
, E1000_ICS_RXDMT0
);
4074 igb_spoof_check(adapter
);
4075 igb_ptp_rx_hang(adapter
);
4077 /* Reset the timer */
4078 if (!test_bit(__IGB_DOWN
, &adapter
->state
)) {
4079 if (adapter
->flags
& IGB_FLAG_NEED_LINK_UPDATE
)
4080 mod_timer(&adapter
->watchdog_timer
,
4081 round_jiffies(jiffies
+ HZ
));
4083 mod_timer(&adapter
->watchdog_timer
,
4084 round_jiffies(jiffies
+ 2 * HZ
));
4088 enum latency_range
{
4092 latency_invalid
= 255
4096 * igb_update_ring_itr - update the dynamic ITR value based on packet size
4097 * @q_vector: pointer to q_vector
4099 * Stores a new ITR value based on strictly on packet size. This
4100 * algorithm is less sophisticated than that used in igb_update_itr,
4101 * due to the difficulty of synchronizing statistics across multiple
4102 * receive rings. The divisors and thresholds used by this function
4103 * were determined based on theoretical maximum wire speed and testing
4104 * data, in order to minimize response time while increasing bulk
4106 * This functionality is controlled by the InterruptThrottleRate module
4107 * parameter (see igb_param.c)
4108 * NOTE: This function is called only when operating in a multiqueue
4109 * receive environment.
4111 static void igb_update_ring_itr(struct igb_q_vector
*q_vector
)
4113 int new_val
= q_vector
->itr_val
;
4114 int avg_wire_size
= 0;
4115 struct igb_adapter
*adapter
= q_vector
->adapter
;
4116 unsigned int packets
;
4118 /* For non-gigabit speeds, just fix the interrupt rate at 4000
4119 * ints/sec - ITR timer value of 120 ticks.
4121 if (adapter
->link_speed
!= SPEED_1000
) {
4122 new_val
= IGB_4K_ITR
;
4126 packets
= q_vector
->rx
.total_packets
;
4128 avg_wire_size
= q_vector
->rx
.total_bytes
/ packets
;
4130 packets
= q_vector
->tx
.total_packets
;
4132 avg_wire_size
= max_t(u32
, avg_wire_size
,
4133 q_vector
->tx
.total_bytes
/ packets
);
4135 /* if avg_wire_size isn't set no work was done */
4139 /* Add 24 bytes to size to account for CRC, preamble, and gap */
4140 avg_wire_size
+= 24;
4142 /* Don't starve jumbo frames */
4143 avg_wire_size
= min(avg_wire_size
, 3000);
4145 /* Give a little boost to mid-size frames */
4146 if ((avg_wire_size
> 300) && (avg_wire_size
< 1200))
4147 new_val
= avg_wire_size
/ 3;
4149 new_val
= avg_wire_size
/ 2;
4151 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4152 if (new_val
< IGB_20K_ITR
&&
4153 ((q_vector
->rx
.ring
&& adapter
->rx_itr_setting
== 3) ||
4154 (!q_vector
->rx
.ring
&& adapter
->tx_itr_setting
== 3)))
4155 new_val
= IGB_20K_ITR
;
4158 if (new_val
!= q_vector
->itr_val
) {
4159 q_vector
->itr_val
= new_val
;
4160 q_vector
->set_itr
= 1;
4163 q_vector
->rx
.total_bytes
= 0;
4164 q_vector
->rx
.total_packets
= 0;
4165 q_vector
->tx
.total_bytes
= 0;
4166 q_vector
->tx
.total_packets
= 0;
4170 * igb_update_itr - update the dynamic ITR value based on statistics
4171 * @q_vector: pointer to q_vector
4172 * @ring_container: ring info to update the itr for
4174 * Stores a new ITR value based on packets and byte
4175 * counts during the last interrupt. The advantage of per interrupt
4176 * computation is faster updates and more accurate ITR for the current
4177 * traffic pattern. Constants in this function were computed
4178 * based on theoretical maximum wire speed and thresholds were set based
4179 * on testing data as well as attempting to minimize response time
4180 * while increasing bulk throughput.
4181 * this functionality is controlled by the InterruptThrottleRate module
4182 * parameter (see igb_param.c)
4183 * NOTE: These calculations are only valid when operating in a single-
4184 * queue environment.
4186 static void igb_update_itr(struct igb_q_vector
*q_vector
,
4187 struct igb_ring_container
*ring_container
)
4189 unsigned int packets
= ring_container
->total_packets
;
4190 unsigned int bytes
= ring_container
->total_bytes
;
4191 u8 itrval
= ring_container
->itr
;
4193 /* no packets, exit with status unchanged */
4198 case lowest_latency
:
4199 /* handle TSO and jumbo frames */
4200 if (bytes
/packets
> 8000)
4201 itrval
= bulk_latency
;
4202 else if ((packets
< 5) && (bytes
> 512))
4203 itrval
= low_latency
;
4205 case low_latency
: /* 50 usec aka 20000 ints/s */
4206 if (bytes
> 10000) {
4207 /* this if handles the TSO accounting */
4208 if (bytes
/packets
> 8000) {
4209 itrval
= bulk_latency
;
4210 } else if ((packets
< 10) || ((bytes
/packets
) > 1200)) {
4211 itrval
= bulk_latency
;
4212 } else if ((packets
> 35)) {
4213 itrval
= lowest_latency
;
4215 } else if (bytes
/packets
> 2000) {
4216 itrval
= bulk_latency
;
4217 } else if (packets
<= 2 && bytes
< 512) {
4218 itrval
= lowest_latency
;
4221 case bulk_latency
: /* 250 usec aka 4000 ints/s */
4222 if (bytes
> 25000) {
4224 itrval
= low_latency
;
4225 } else if (bytes
< 1500) {
4226 itrval
= low_latency
;
4231 /* clear work counters since we have the values we need */
4232 ring_container
->total_bytes
= 0;
4233 ring_container
->total_packets
= 0;
4235 /* write updated itr to ring container */
4236 ring_container
->itr
= itrval
;
4239 static void igb_set_itr(struct igb_q_vector
*q_vector
)
4241 struct igb_adapter
*adapter
= q_vector
->adapter
;
4242 u32 new_itr
= q_vector
->itr_val
;
4245 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4246 if (adapter
->link_speed
!= SPEED_1000
) {
4248 new_itr
= IGB_4K_ITR
;
4252 igb_update_itr(q_vector
, &q_vector
->tx
);
4253 igb_update_itr(q_vector
, &q_vector
->rx
);
4255 current_itr
= max(q_vector
->rx
.itr
, q_vector
->tx
.itr
);
4257 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4258 if (current_itr
== lowest_latency
&&
4259 ((q_vector
->rx
.ring
&& adapter
->rx_itr_setting
== 3) ||
4260 (!q_vector
->rx
.ring
&& adapter
->tx_itr_setting
== 3)))
4261 current_itr
= low_latency
;
4263 switch (current_itr
) {
4264 /* counts and packets in update_itr are dependent on these numbers */
4265 case lowest_latency
:
4266 new_itr
= IGB_70K_ITR
; /* 70,000 ints/sec */
4269 new_itr
= IGB_20K_ITR
; /* 20,000 ints/sec */
4272 new_itr
= IGB_4K_ITR
; /* 4,000 ints/sec */
4279 if (new_itr
!= q_vector
->itr_val
) {
4280 /* this attempts to bias the interrupt rate towards Bulk
4281 * by adding intermediate steps when interrupt rate is
4284 new_itr
= new_itr
> q_vector
->itr_val
?
4285 max((new_itr
* q_vector
->itr_val
) /
4286 (new_itr
+ (q_vector
->itr_val
>> 2)),
4288 /* Don't write the value here; it resets the adapter's
4289 * internal timer, and causes us to delay far longer than
4290 * we should between interrupts. Instead, we write the ITR
4291 * value at the beginning of the next interrupt so the timing
4292 * ends up being correct.
4294 q_vector
->itr_val
= new_itr
;
4295 q_vector
->set_itr
= 1;
4299 static void igb_tx_ctxtdesc(struct igb_ring
*tx_ring
, u32 vlan_macip_lens
,
4300 u32 type_tucmd
, u32 mss_l4len_idx
)
4302 struct e1000_adv_tx_context_desc
*context_desc
;
4303 u16 i
= tx_ring
->next_to_use
;
4305 context_desc
= IGB_TX_CTXTDESC(tx_ring
, i
);
4308 tx_ring
->next_to_use
= (i
< tx_ring
->count
) ? i
: 0;
4310 /* set bits to identify this as an advanced context descriptor */
4311 type_tucmd
|= E1000_TXD_CMD_DEXT
| E1000_ADVTXD_DTYP_CTXT
;
4313 /* For 82575, context index must be unique per ring. */
4314 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX
, &tx_ring
->flags
))
4315 mss_l4len_idx
|= tx_ring
->reg_idx
<< 4;
4317 context_desc
->vlan_macip_lens
= cpu_to_le32(vlan_macip_lens
);
4318 context_desc
->seqnum_seed
= 0;
4319 context_desc
->type_tucmd_mlhl
= cpu_to_le32(type_tucmd
);
4320 context_desc
->mss_l4len_idx
= cpu_to_le32(mss_l4len_idx
);
4323 static int igb_tso(struct igb_ring
*tx_ring
,
4324 struct igb_tx_buffer
*first
,
4327 struct sk_buff
*skb
= first
->skb
;
4328 u32 vlan_macip_lens
, type_tucmd
;
4329 u32 mss_l4len_idx
, l4len
;
4331 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
4334 if (!skb_is_gso(skb
))
4337 if (skb_header_cloned(skb
)) {
4338 int err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
4343 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4344 type_tucmd
= E1000_ADVTXD_TUCMD_L4T_TCP
;
4346 if (first
->protocol
== __constant_htons(ETH_P_IP
)) {
4347 struct iphdr
*iph
= ip_hdr(skb
);
4350 tcp_hdr(skb
)->check
= ~csum_tcpudp_magic(iph
->saddr
,
4354 type_tucmd
|= E1000_ADVTXD_TUCMD_IPV4
;
4355 first
->tx_flags
|= IGB_TX_FLAGS_TSO
|
4358 } else if (skb_is_gso_v6(skb
)) {
4359 ipv6_hdr(skb
)->payload_len
= 0;
4360 tcp_hdr(skb
)->check
= ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
4361 &ipv6_hdr(skb
)->daddr
,
4363 first
->tx_flags
|= IGB_TX_FLAGS_TSO
|
4367 /* compute header lengths */
4368 l4len
= tcp_hdrlen(skb
);
4369 *hdr_len
= skb_transport_offset(skb
) + l4len
;
4371 /* update gso size and bytecount with header size */
4372 first
->gso_segs
= skb_shinfo(skb
)->gso_segs
;
4373 first
->bytecount
+= (first
->gso_segs
- 1) * *hdr_len
;
4376 mss_l4len_idx
= l4len
<< E1000_ADVTXD_L4LEN_SHIFT
;
4377 mss_l4len_idx
|= skb_shinfo(skb
)->gso_size
<< E1000_ADVTXD_MSS_SHIFT
;
4379 /* VLAN MACLEN IPLEN */
4380 vlan_macip_lens
= skb_network_header_len(skb
);
4381 vlan_macip_lens
|= skb_network_offset(skb
) << E1000_ADVTXD_MACLEN_SHIFT
;
4382 vlan_macip_lens
|= first
->tx_flags
& IGB_TX_FLAGS_VLAN_MASK
;
4384 igb_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, type_tucmd
, mss_l4len_idx
);
4389 static void igb_tx_csum(struct igb_ring
*tx_ring
, struct igb_tx_buffer
*first
)
4391 struct sk_buff
*skb
= first
->skb
;
4392 u32 vlan_macip_lens
= 0;
4393 u32 mss_l4len_idx
= 0;
4396 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
) {
4397 if (!(first
->tx_flags
& IGB_TX_FLAGS_VLAN
))
4401 switch (first
->protocol
) {
4402 case __constant_htons(ETH_P_IP
):
4403 vlan_macip_lens
|= skb_network_header_len(skb
);
4404 type_tucmd
|= E1000_ADVTXD_TUCMD_IPV4
;
4405 l4_hdr
= ip_hdr(skb
)->protocol
;
4407 case __constant_htons(ETH_P_IPV6
):
4408 vlan_macip_lens
|= skb_network_header_len(skb
);
4409 l4_hdr
= ipv6_hdr(skb
)->nexthdr
;
4412 if (unlikely(net_ratelimit())) {
4413 dev_warn(tx_ring
->dev
,
4414 "partial checksum but proto=%x!\n",
4422 type_tucmd
|= E1000_ADVTXD_TUCMD_L4T_TCP
;
4423 mss_l4len_idx
= tcp_hdrlen(skb
) <<
4424 E1000_ADVTXD_L4LEN_SHIFT
;
4427 type_tucmd
|= E1000_ADVTXD_TUCMD_L4T_SCTP
;
4428 mss_l4len_idx
= sizeof(struct sctphdr
) <<
4429 E1000_ADVTXD_L4LEN_SHIFT
;
4432 mss_l4len_idx
= sizeof(struct udphdr
) <<
4433 E1000_ADVTXD_L4LEN_SHIFT
;
4436 if (unlikely(net_ratelimit())) {
4437 dev_warn(tx_ring
->dev
,
4438 "partial checksum but l4 proto=%x!\n",
4444 /* update TX checksum flag */
4445 first
->tx_flags
|= IGB_TX_FLAGS_CSUM
;
4448 vlan_macip_lens
|= skb_network_offset(skb
) << E1000_ADVTXD_MACLEN_SHIFT
;
4449 vlan_macip_lens
|= first
->tx_flags
& IGB_TX_FLAGS_VLAN_MASK
;
4451 igb_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, type_tucmd
, mss_l4len_idx
);
4454 #define IGB_SET_FLAG(_input, _flag, _result) \
4455 ((_flag <= _result) ? \
4456 ((u32)(_input & _flag) * (_result / _flag)) : \
4457 ((u32)(_input & _flag) / (_flag / _result)))
4459 static u32
igb_tx_cmd_type(struct sk_buff
*skb
, u32 tx_flags
)
4461 /* set type for advanced descriptor with frame checksum insertion */
4462 u32 cmd_type
= E1000_ADVTXD_DTYP_DATA
|
4463 E1000_ADVTXD_DCMD_DEXT
|
4464 E1000_ADVTXD_DCMD_IFCS
;
4466 /* set HW vlan bit if vlan is present */
4467 cmd_type
|= IGB_SET_FLAG(tx_flags
, IGB_TX_FLAGS_VLAN
,
4468 (E1000_ADVTXD_DCMD_VLE
));
4470 /* set segmentation bits for TSO */
4471 cmd_type
|= IGB_SET_FLAG(tx_flags
, IGB_TX_FLAGS_TSO
,
4472 (E1000_ADVTXD_DCMD_TSE
));
4474 /* set timestamp bit if present */
4475 cmd_type
|= IGB_SET_FLAG(tx_flags
, IGB_TX_FLAGS_TSTAMP
,
4476 (E1000_ADVTXD_MAC_TSTAMP
));
4478 /* insert frame checksum */
4479 cmd_type
^= IGB_SET_FLAG(skb
->no_fcs
, 1, E1000_ADVTXD_DCMD_IFCS
);
4484 static void igb_tx_olinfo_status(struct igb_ring
*tx_ring
,
4485 union e1000_adv_tx_desc
*tx_desc
,
4486 u32 tx_flags
, unsigned int paylen
)
4488 u32 olinfo_status
= paylen
<< E1000_ADVTXD_PAYLEN_SHIFT
;
4490 /* 82575 requires a unique index per ring */
4491 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX
, &tx_ring
->flags
))
4492 olinfo_status
|= tx_ring
->reg_idx
<< 4;
4494 /* insert L4 checksum */
4495 olinfo_status
|= IGB_SET_FLAG(tx_flags
,
4497 (E1000_TXD_POPTS_TXSM
<< 8));
4499 /* insert IPv4 checksum */
4500 olinfo_status
|= IGB_SET_FLAG(tx_flags
,
4502 (E1000_TXD_POPTS_IXSM
<< 8));
4504 tx_desc
->read
.olinfo_status
= cpu_to_le32(olinfo_status
);
4507 static void igb_tx_map(struct igb_ring
*tx_ring
,
4508 struct igb_tx_buffer
*first
,
4511 struct sk_buff
*skb
= first
->skb
;
4512 struct igb_tx_buffer
*tx_buffer
;
4513 union e1000_adv_tx_desc
*tx_desc
;
4514 struct skb_frag_struct
*frag
;
4516 unsigned int data_len
, size
;
4517 u32 tx_flags
= first
->tx_flags
;
4518 u32 cmd_type
= igb_tx_cmd_type(skb
, tx_flags
);
4519 u16 i
= tx_ring
->next_to_use
;
4521 tx_desc
= IGB_TX_DESC(tx_ring
, i
);
4523 igb_tx_olinfo_status(tx_ring
, tx_desc
, tx_flags
, skb
->len
- hdr_len
);
4525 size
= skb_headlen(skb
);
4526 data_len
= skb
->data_len
;
4528 dma
= dma_map_single(tx_ring
->dev
, skb
->data
, size
, DMA_TO_DEVICE
);
4532 for (frag
= &skb_shinfo(skb
)->frags
[0];; frag
++) {
4533 if (dma_mapping_error(tx_ring
->dev
, dma
))
4536 /* record length, and DMA address */
4537 dma_unmap_len_set(tx_buffer
, len
, size
);
4538 dma_unmap_addr_set(tx_buffer
, dma
, dma
);
4540 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
4542 while (unlikely(size
> IGB_MAX_DATA_PER_TXD
)) {
4543 tx_desc
->read
.cmd_type_len
=
4544 cpu_to_le32(cmd_type
^ IGB_MAX_DATA_PER_TXD
);
4548 if (i
== tx_ring
->count
) {
4549 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
4552 tx_desc
->read
.olinfo_status
= 0;
4554 dma
+= IGB_MAX_DATA_PER_TXD
;
4555 size
-= IGB_MAX_DATA_PER_TXD
;
4557 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
4560 if (likely(!data_len
))
4563 tx_desc
->read
.cmd_type_len
= cpu_to_le32(cmd_type
^ size
);
4567 if (i
== tx_ring
->count
) {
4568 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
4571 tx_desc
->read
.olinfo_status
= 0;
4573 size
= skb_frag_size(frag
);
4576 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0,
4577 size
, DMA_TO_DEVICE
);
4579 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
4582 /* write last descriptor with RS and EOP bits */
4583 cmd_type
|= size
| IGB_TXD_DCMD
;
4584 tx_desc
->read
.cmd_type_len
= cpu_to_le32(cmd_type
);
4586 netdev_tx_sent_queue(txring_txq(tx_ring
), first
->bytecount
);
4588 /* set the timestamp */
4589 first
->time_stamp
= jiffies
;
4591 /* Force memory writes to complete before letting h/w know there
4592 * are new descriptors to fetch. (Only applicable for weak-ordered
4593 * memory model archs, such as IA-64).
4595 * We also need this memory barrier to make certain all of the
4596 * status bits have been updated before next_to_watch is written.
4600 /* set next_to_watch value indicating a packet is present */
4601 first
->next_to_watch
= tx_desc
;
4604 if (i
== tx_ring
->count
)
4607 tx_ring
->next_to_use
= i
;
4609 writel(i
, tx_ring
->tail
);
4611 /* we need this if more than one processor can write to our tail
4612 * at a time, it synchronizes IO on IA64/Altix systems
4619 dev_err(tx_ring
->dev
, "TX DMA map failed\n");
4621 /* clear dma mappings for failed tx_buffer_info map */
4623 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
4624 igb_unmap_and_free_tx_resource(tx_ring
, tx_buffer
);
4625 if (tx_buffer
== first
)
4632 tx_ring
->next_to_use
= i
;
4635 static int __igb_maybe_stop_tx(struct igb_ring
*tx_ring
, const u16 size
)
4637 struct net_device
*netdev
= tx_ring
->netdev
;
4639 netif_stop_subqueue(netdev
, tx_ring
->queue_index
);
4641 /* Herbert's original patch had:
4642 * smp_mb__after_netif_stop_queue();
4643 * but since that doesn't exist yet, just open code it.
4647 /* We need to check again in a case another CPU has just
4648 * made room available.
4650 if (igb_desc_unused(tx_ring
) < size
)
4654 netif_wake_subqueue(netdev
, tx_ring
->queue_index
);
4656 u64_stats_update_begin(&tx_ring
->tx_syncp2
);
4657 tx_ring
->tx_stats
.restart_queue2
++;
4658 u64_stats_update_end(&tx_ring
->tx_syncp2
);
4663 static inline int igb_maybe_stop_tx(struct igb_ring
*tx_ring
, const u16 size
)
4665 if (igb_desc_unused(tx_ring
) >= size
)
4667 return __igb_maybe_stop_tx(tx_ring
, size
);
4670 netdev_tx_t
igb_xmit_frame_ring(struct sk_buff
*skb
,
4671 struct igb_ring
*tx_ring
)
4673 struct igb_tx_buffer
*first
;
4676 u16 count
= TXD_USE_COUNT(skb_headlen(skb
));
4677 __be16 protocol
= vlan_get_protocol(skb
);
4680 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
4681 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
4682 * + 2 desc gap to keep tail from touching head,
4683 * + 1 desc for context descriptor,
4684 * otherwise try next time
4686 if (NETDEV_FRAG_PAGE_MAX_SIZE
> IGB_MAX_DATA_PER_TXD
) {
4688 for (f
= 0; f
< skb_shinfo(skb
)->nr_frags
; f
++)
4689 count
+= TXD_USE_COUNT(skb_shinfo(skb
)->frags
[f
].size
);
4691 count
+= skb_shinfo(skb
)->nr_frags
;
4694 if (igb_maybe_stop_tx(tx_ring
, count
+ 3)) {
4695 /* this is a hard error */
4696 return NETDEV_TX_BUSY
;
4699 /* record the location of the first descriptor for this packet */
4700 first
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_use
];
4702 first
->bytecount
= skb
->len
;
4703 first
->gso_segs
= 1;
4705 skb_tx_timestamp(skb
);
4707 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_HW_TSTAMP
)) {
4708 struct igb_adapter
*adapter
= netdev_priv(tx_ring
->netdev
);
4710 if (!(adapter
->ptp_tx_skb
)) {
4711 skb_shinfo(skb
)->tx_flags
|= SKBTX_IN_PROGRESS
;
4712 tx_flags
|= IGB_TX_FLAGS_TSTAMP
;
4714 adapter
->ptp_tx_skb
= skb_get(skb
);
4715 adapter
->ptp_tx_start
= jiffies
;
4716 if (adapter
->hw
.mac
.type
== e1000_82576
)
4717 schedule_work(&adapter
->ptp_tx_work
);
4721 if (vlan_tx_tag_present(skb
)) {
4722 tx_flags
|= IGB_TX_FLAGS_VLAN
;
4723 tx_flags
|= (vlan_tx_tag_get(skb
) << IGB_TX_FLAGS_VLAN_SHIFT
);
4726 /* record initial flags and protocol */
4727 first
->tx_flags
= tx_flags
;
4728 first
->protocol
= protocol
;
4730 tso
= igb_tso(tx_ring
, first
, &hdr_len
);
4734 igb_tx_csum(tx_ring
, first
);
4736 igb_tx_map(tx_ring
, first
, hdr_len
);
4738 /* Make sure there is space in the ring for the next send. */
4739 igb_maybe_stop_tx(tx_ring
, DESC_NEEDED
);
4741 return NETDEV_TX_OK
;
4744 igb_unmap_and_free_tx_resource(tx_ring
, first
);
4746 return NETDEV_TX_OK
;
4749 static inline struct igb_ring
*igb_tx_queue_mapping(struct igb_adapter
*adapter
,
4750 struct sk_buff
*skb
)
4752 unsigned int r_idx
= skb
->queue_mapping
;
4754 if (r_idx
>= adapter
->num_tx_queues
)
4755 r_idx
= r_idx
% adapter
->num_tx_queues
;
4757 return adapter
->tx_ring
[r_idx
];
4760 static netdev_tx_t
igb_xmit_frame(struct sk_buff
*skb
,
4761 struct net_device
*netdev
)
4763 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4765 if (test_bit(__IGB_DOWN
, &adapter
->state
)) {
4766 dev_kfree_skb_any(skb
);
4767 return NETDEV_TX_OK
;
4770 if (skb
->len
<= 0) {
4771 dev_kfree_skb_any(skb
);
4772 return NETDEV_TX_OK
;
4775 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
4776 * in order to meet this minimum size requirement.
4778 if (unlikely(skb
->len
< 17)) {
4779 if (skb_pad(skb
, 17 - skb
->len
))
4780 return NETDEV_TX_OK
;
4782 skb_set_tail_pointer(skb
, 17);
4785 return igb_xmit_frame_ring(skb
, igb_tx_queue_mapping(adapter
, skb
));
4789 * igb_tx_timeout - Respond to a Tx Hang
4790 * @netdev: network interface device structure
4792 static void igb_tx_timeout(struct net_device
*netdev
)
4794 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4795 struct e1000_hw
*hw
= &adapter
->hw
;
4797 /* Do the reset outside of interrupt context */
4798 adapter
->tx_timeout_count
++;
4800 if (hw
->mac
.type
>= e1000_82580
)
4801 hw
->dev_spec
._82575
.global_device_reset
= true;
4803 schedule_work(&adapter
->reset_task
);
4805 (adapter
->eims_enable_mask
& ~adapter
->eims_other
));
4808 static void igb_reset_task(struct work_struct
*work
)
4810 struct igb_adapter
*adapter
;
4811 adapter
= container_of(work
, struct igb_adapter
, reset_task
);
4814 netdev_err(adapter
->netdev
, "Reset adapter\n");
4815 igb_reinit_locked(adapter
);
4819 * igb_get_stats64 - Get System Network Statistics
4820 * @netdev: network interface device structure
4821 * @stats: rtnl_link_stats64 pointer
4823 static struct rtnl_link_stats64
*igb_get_stats64(struct net_device
*netdev
,
4824 struct rtnl_link_stats64
*stats
)
4826 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4828 spin_lock(&adapter
->stats64_lock
);
4829 igb_update_stats(adapter
, &adapter
->stats64
);
4830 memcpy(stats
, &adapter
->stats64
, sizeof(*stats
));
4831 spin_unlock(&adapter
->stats64_lock
);
4837 * igb_change_mtu - Change the Maximum Transfer Unit
4838 * @netdev: network interface device structure
4839 * @new_mtu: new value for maximum frame size
4841 * Returns 0 on success, negative on failure
4843 static int igb_change_mtu(struct net_device
*netdev
, int new_mtu
)
4845 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4846 struct pci_dev
*pdev
= adapter
->pdev
;
4847 int max_frame
= new_mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+ VLAN_HLEN
;
4849 if ((new_mtu
< 68) || (max_frame
> MAX_JUMBO_FRAME_SIZE
)) {
4850 dev_err(&pdev
->dev
, "Invalid MTU setting\n");
4854 #define MAX_STD_JUMBO_FRAME_SIZE 9238
4855 if (max_frame
> MAX_STD_JUMBO_FRAME_SIZE
) {
4856 dev_err(&pdev
->dev
, "MTU > 9216 not supported.\n");
4860 /* adjust max frame to be at least the size of a standard frame */
4861 if (max_frame
< (ETH_FRAME_LEN
+ ETH_FCS_LEN
))
4862 max_frame
= ETH_FRAME_LEN
+ ETH_FCS_LEN
;
4864 while (test_and_set_bit(__IGB_RESETTING
, &adapter
->state
))
4867 /* igb_down has a dependency on max_frame_size */
4868 adapter
->max_frame_size
= max_frame
;
4870 if (netif_running(netdev
))
4873 dev_info(&pdev
->dev
, "changing MTU from %d to %d\n",
4874 netdev
->mtu
, new_mtu
);
4875 netdev
->mtu
= new_mtu
;
4877 if (netif_running(netdev
))
4882 clear_bit(__IGB_RESETTING
, &adapter
->state
);
4888 * igb_update_stats - Update the board statistics counters
4889 * @adapter: board private structure
4891 void igb_update_stats(struct igb_adapter
*adapter
,
4892 struct rtnl_link_stats64
*net_stats
)
4894 struct e1000_hw
*hw
= &adapter
->hw
;
4895 struct pci_dev
*pdev
= adapter
->pdev
;
4901 u64 _bytes
, _packets
;
4903 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
4905 /* Prevent stats update while adapter is being reset, or if the pci
4906 * connection is down.
4908 if (adapter
->link_speed
== 0)
4910 if (pci_channel_offline(pdev
))
4917 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
4918 u32 rqdpc
= rd32(E1000_RQDPC(i
));
4919 struct igb_ring
*ring
= adapter
->rx_ring
[i
];
4922 ring
->rx_stats
.drops
+= rqdpc
;
4923 net_stats
->rx_fifo_errors
+= rqdpc
;
4927 start
= u64_stats_fetch_begin_bh(&ring
->rx_syncp
);
4928 _bytes
= ring
->rx_stats
.bytes
;
4929 _packets
= ring
->rx_stats
.packets
;
4930 } while (u64_stats_fetch_retry_bh(&ring
->rx_syncp
, start
));
4932 packets
+= _packets
;
4935 net_stats
->rx_bytes
= bytes
;
4936 net_stats
->rx_packets
= packets
;
4940 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
4941 struct igb_ring
*ring
= adapter
->tx_ring
[i
];
4943 start
= u64_stats_fetch_begin_bh(&ring
->tx_syncp
);
4944 _bytes
= ring
->tx_stats
.bytes
;
4945 _packets
= ring
->tx_stats
.packets
;
4946 } while (u64_stats_fetch_retry_bh(&ring
->tx_syncp
, start
));
4948 packets
+= _packets
;
4950 net_stats
->tx_bytes
= bytes
;
4951 net_stats
->tx_packets
= packets
;
4954 /* read stats registers */
4955 adapter
->stats
.crcerrs
+= rd32(E1000_CRCERRS
);
4956 adapter
->stats
.gprc
+= rd32(E1000_GPRC
);
4957 adapter
->stats
.gorc
+= rd32(E1000_GORCL
);
4958 rd32(E1000_GORCH
); /* clear GORCL */
4959 adapter
->stats
.bprc
+= rd32(E1000_BPRC
);
4960 adapter
->stats
.mprc
+= rd32(E1000_MPRC
);
4961 adapter
->stats
.roc
+= rd32(E1000_ROC
);
4963 adapter
->stats
.prc64
+= rd32(E1000_PRC64
);
4964 adapter
->stats
.prc127
+= rd32(E1000_PRC127
);
4965 adapter
->stats
.prc255
+= rd32(E1000_PRC255
);
4966 adapter
->stats
.prc511
+= rd32(E1000_PRC511
);
4967 adapter
->stats
.prc1023
+= rd32(E1000_PRC1023
);
4968 adapter
->stats
.prc1522
+= rd32(E1000_PRC1522
);
4969 adapter
->stats
.symerrs
+= rd32(E1000_SYMERRS
);
4970 adapter
->stats
.sec
+= rd32(E1000_SEC
);
4972 mpc
= rd32(E1000_MPC
);
4973 adapter
->stats
.mpc
+= mpc
;
4974 net_stats
->rx_fifo_errors
+= mpc
;
4975 adapter
->stats
.scc
+= rd32(E1000_SCC
);
4976 adapter
->stats
.ecol
+= rd32(E1000_ECOL
);
4977 adapter
->stats
.mcc
+= rd32(E1000_MCC
);
4978 adapter
->stats
.latecol
+= rd32(E1000_LATECOL
);
4979 adapter
->stats
.dc
+= rd32(E1000_DC
);
4980 adapter
->stats
.rlec
+= rd32(E1000_RLEC
);
4981 adapter
->stats
.xonrxc
+= rd32(E1000_XONRXC
);
4982 adapter
->stats
.xontxc
+= rd32(E1000_XONTXC
);
4983 adapter
->stats
.xoffrxc
+= rd32(E1000_XOFFRXC
);
4984 adapter
->stats
.xofftxc
+= rd32(E1000_XOFFTXC
);
4985 adapter
->stats
.fcruc
+= rd32(E1000_FCRUC
);
4986 adapter
->stats
.gptc
+= rd32(E1000_GPTC
);
4987 adapter
->stats
.gotc
+= rd32(E1000_GOTCL
);
4988 rd32(E1000_GOTCH
); /* clear GOTCL */
4989 adapter
->stats
.rnbc
+= rd32(E1000_RNBC
);
4990 adapter
->stats
.ruc
+= rd32(E1000_RUC
);
4991 adapter
->stats
.rfc
+= rd32(E1000_RFC
);
4992 adapter
->stats
.rjc
+= rd32(E1000_RJC
);
4993 adapter
->stats
.tor
+= rd32(E1000_TORH
);
4994 adapter
->stats
.tot
+= rd32(E1000_TOTH
);
4995 adapter
->stats
.tpr
+= rd32(E1000_TPR
);
4997 adapter
->stats
.ptc64
+= rd32(E1000_PTC64
);
4998 adapter
->stats
.ptc127
+= rd32(E1000_PTC127
);
4999 adapter
->stats
.ptc255
+= rd32(E1000_PTC255
);
5000 adapter
->stats
.ptc511
+= rd32(E1000_PTC511
);
5001 adapter
->stats
.ptc1023
+= rd32(E1000_PTC1023
);
5002 adapter
->stats
.ptc1522
+= rd32(E1000_PTC1522
);
5004 adapter
->stats
.mptc
+= rd32(E1000_MPTC
);
5005 adapter
->stats
.bptc
+= rd32(E1000_BPTC
);
5007 adapter
->stats
.tpt
+= rd32(E1000_TPT
);
5008 adapter
->stats
.colc
+= rd32(E1000_COLC
);
5010 adapter
->stats
.algnerrc
+= rd32(E1000_ALGNERRC
);
5011 /* read internal phy specific stats */
5012 reg
= rd32(E1000_CTRL_EXT
);
5013 if (!(reg
& E1000_CTRL_EXT_LINK_MODE_MASK
)) {
5014 adapter
->stats
.rxerrc
+= rd32(E1000_RXERRC
);
5016 /* this stat has invalid values on i210/i211 */
5017 if ((hw
->mac
.type
!= e1000_i210
) &&
5018 (hw
->mac
.type
!= e1000_i211
))
5019 adapter
->stats
.tncrs
+= rd32(E1000_TNCRS
);
5022 adapter
->stats
.tsctc
+= rd32(E1000_TSCTC
);
5023 adapter
->stats
.tsctfc
+= rd32(E1000_TSCTFC
);
5025 adapter
->stats
.iac
+= rd32(E1000_IAC
);
5026 adapter
->stats
.icrxoc
+= rd32(E1000_ICRXOC
);
5027 adapter
->stats
.icrxptc
+= rd32(E1000_ICRXPTC
);
5028 adapter
->stats
.icrxatc
+= rd32(E1000_ICRXATC
);
5029 adapter
->stats
.ictxptc
+= rd32(E1000_ICTXPTC
);
5030 adapter
->stats
.ictxatc
+= rd32(E1000_ICTXATC
);
5031 adapter
->stats
.ictxqec
+= rd32(E1000_ICTXQEC
);
5032 adapter
->stats
.ictxqmtc
+= rd32(E1000_ICTXQMTC
);
5033 adapter
->stats
.icrxdmtc
+= rd32(E1000_ICRXDMTC
);
5035 /* Fill out the OS statistics structure */
5036 net_stats
->multicast
= adapter
->stats
.mprc
;
5037 net_stats
->collisions
= adapter
->stats
.colc
;
5041 /* RLEC on some newer hardware can be incorrect so build
5042 * our own version based on RUC and ROC
5044 net_stats
->rx_errors
= adapter
->stats
.rxerrc
+
5045 adapter
->stats
.crcerrs
+ adapter
->stats
.algnerrc
+
5046 adapter
->stats
.ruc
+ adapter
->stats
.roc
+
5047 adapter
->stats
.cexterr
;
5048 net_stats
->rx_length_errors
= adapter
->stats
.ruc
+
5050 net_stats
->rx_crc_errors
= adapter
->stats
.crcerrs
;
5051 net_stats
->rx_frame_errors
= adapter
->stats
.algnerrc
;
5052 net_stats
->rx_missed_errors
= adapter
->stats
.mpc
;
5055 net_stats
->tx_errors
= adapter
->stats
.ecol
+
5056 adapter
->stats
.latecol
;
5057 net_stats
->tx_aborted_errors
= adapter
->stats
.ecol
;
5058 net_stats
->tx_window_errors
= adapter
->stats
.latecol
;
5059 net_stats
->tx_carrier_errors
= adapter
->stats
.tncrs
;
5061 /* Tx Dropped needs to be maintained elsewhere */
5064 if (hw
->phy
.media_type
== e1000_media_type_copper
) {
5065 if ((adapter
->link_speed
== SPEED_1000
) &&
5066 (!igb_read_phy_reg(hw
, PHY_1000T_STATUS
, &phy_tmp
))) {
5067 phy_tmp
&= PHY_IDLE_ERROR_COUNT_MASK
;
5068 adapter
->phy_stats
.idle_errors
+= phy_tmp
;
5072 /* Management Stats */
5073 adapter
->stats
.mgptc
+= rd32(E1000_MGTPTC
);
5074 adapter
->stats
.mgprc
+= rd32(E1000_MGTPRC
);
5075 adapter
->stats
.mgpdc
+= rd32(E1000_MGTPDC
);
5078 reg
= rd32(E1000_MANC
);
5079 if (reg
& E1000_MANC_EN_BMC2OS
) {
5080 adapter
->stats
.o2bgptc
+= rd32(E1000_O2BGPTC
);
5081 adapter
->stats
.o2bspc
+= rd32(E1000_O2BSPC
);
5082 adapter
->stats
.b2ospc
+= rd32(E1000_B2OSPC
);
5083 adapter
->stats
.b2ogprc
+= rd32(E1000_B2OGPRC
);
5087 static irqreturn_t
igb_msix_other(int irq
, void *data
)
5089 struct igb_adapter
*adapter
= data
;
5090 struct e1000_hw
*hw
= &adapter
->hw
;
5091 u32 icr
= rd32(E1000_ICR
);
5092 /* reading ICR causes bit 31 of EICR to be cleared */
5094 if (icr
& E1000_ICR_DRSTA
)
5095 schedule_work(&adapter
->reset_task
);
5097 if (icr
& E1000_ICR_DOUTSYNC
) {
5098 /* HW is reporting DMA is out of sync */
5099 adapter
->stats
.doosync
++;
5100 /* The DMA Out of Sync is also indication of a spoof event
5101 * in IOV mode. Check the Wrong VM Behavior register to
5102 * see if it is really a spoof event.
5104 igb_check_wvbr(adapter
);
5107 /* Check for a mailbox event */
5108 if (icr
& E1000_ICR_VMMB
)
5109 igb_msg_task(adapter
);
5111 if (icr
& E1000_ICR_LSC
) {
5112 hw
->mac
.get_link_status
= 1;
5113 /* guard against interrupt when we're going down */
5114 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
5115 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
5118 if (icr
& E1000_ICR_TS
) {
5119 u32 tsicr
= rd32(E1000_TSICR
);
5121 if (tsicr
& E1000_TSICR_TXTS
) {
5122 /* acknowledge the interrupt */
5123 wr32(E1000_TSICR
, E1000_TSICR_TXTS
);
5124 /* retrieve hardware timestamp */
5125 schedule_work(&adapter
->ptp_tx_work
);
5129 wr32(E1000_EIMS
, adapter
->eims_other
);
5134 static void igb_write_itr(struct igb_q_vector
*q_vector
)
5136 struct igb_adapter
*adapter
= q_vector
->adapter
;
5137 u32 itr_val
= q_vector
->itr_val
& 0x7FFC;
5139 if (!q_vector
->set_itr
)
5145 if (adapter
->hw
.mac
.type
== e1000_82575
)
5146 itr_val
|= itr_val
<< 16;
5148 itr_val
|= E1000_EITR_CNT_IGNR
;
5150 writel(itr_val
, q_vector
->itr_register
);
5151 q_vector
->set_itr
= 0;
5154 static irqreturn_t
igb_msix_ring(int irq
, void *data
)
5156 struct igb_q_vector
*q_vector
= data
;
5158 /* Write the ITR value calculated from the previous interrupt. */
5159 igb_write_itr(q_vector
);
5161 napi_schedule(&q_vector
->napi
);
5166 #ifdef CONFIG_IGB_DCA
5167 static void igb_update_tx_dca(struct igb_adapter
*adapter
,
5168 struct igb_ring
*tx_ring
,
5171 struct e1000_hw
*hw
= &adapter
->hw
;
5172 u32 txctrl
= dca3_get_tag(tx_ring
->dev
, cpu
);
5174 if (hw
->mac
.type
!= e1000_82575
)
5175 txctrl
<<= E1000_DCA_TXCTRL_CPUID_SHIFT
;
5177 /* We can enable relaxed ordering for reads, but not writes when
5178 * DCA is enabled. This is due to a known issue in some chipsets
5179 * which will cause the DCA tag to be cleared.
5181 txctrl
|= E1000_DCA_TXCTRL_DESC_RRO_EN
|
5182 E1000_DCA_TXCTRL_DATA_RRO_EN
|
5183 E1000_DCA_TXCTRL_DESC_DCA_EN
;
5185 wr32(E1000_DCA_TXCTRL(tx_ring
->reg_idx
), txctrl
);
5188 static void igb_update_rx_dca(struct igb_adapter
*adapter
,
5189 struct igb_ring
*rx_ring
,
5192 struct e1000_hw
*hw
= &adapter
->hw
;
5193 u32 rxctrl
= dca3_get_tag(&adapter
->pdev
->dev
, cpu
);
5195 if (hw
->mac
.type
!= e1000_82575
)
5196 rxctrl
<<= E1000_DCA_RXCTRL_CPUID_SHIFT
;
5198 /* We can enable relaxed ordering for reads, but not writes when
5199 * DCA is enabled. This is due to a known issue in some chipsets
5200 * which will cause the DCA tag to be cleared.
5202 rxctrl
|= E1000_DCA_RXCTRL_DESC_RRO_EN
|
5203 E1000_DCA_RXCTRL_DESC_DCA_EN
;
5205 wr32(E1000_DCA_RXCTRL(rx_ring
->reg_idx
), rxctrl
);
5208 static void igb_update_dca(struct igb_q_vector
*q_vector
)
5210 struct igb_adapter
*adapter
= q_vector
->adapter
;
5211 int cpu
= get_cpu();
5213 if (q_vector
->cpu
== cpu
)
5216 if (q_vector
->tx
.ring
)
5217 igb_update_tx_dca(adapter
, q_vector
->tx
.ring
, cpu
);
5219 if (q_vector
->rx
.ring
)
5220 igb_update_rx_dca(adapter
, q_vector
->rx
.ring
, cpu
);
5222 q_vector
->cpu
= cpu
;
5227 static void igb_setup_dca(struct igb_adapter
*adapter
)
5229 struct e1000_hw
*hw
= &adapter
->hw
;
5232 if (!(adapter
->flags
& IGB_FLAG_DCA_ENABLED
))
5235 /* Always use CB2 mode, difference is masked in the CB driver. */
5236 wr32(E1000_DCA_CTRL
, E1000_DCA_CTRL_DCA_MODE_CB2
);
5238 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
5239 adapter
->q_vector
[i
]->cpu
= -1;
5240 igb_update_dca(adapter
->q_vector
[i
]);
5244 static int __igb_notify_dca(struct device
*dev
, void *data
)
5246 struct net_device
*netdev
= dev_get_drvdata(dev
);
5247 struct igb_adapter
*adapter
= netdev_priv(netdev
);
5248 struct pci_dev
*pdev
= adapter
->pdev
;
5249 struct e1000_hw
*hw
= &adapter
->hw
;
5250 unsigned long event
= *(unsigned long *)data
;
5253 case DCA_PROVIDER_ADD
:
5254 /* if already enabled, don't do it again */
5255 if (adapter
->flags
& IGB_FLAG_DCA_ENABLED
)
5257 if (dca_add_requester(dev
) == 0) {
5258 adapter
->flags
|= IGB_FLAG_DCA_ENABLED
;
5259 dev_info(&pdev
->dev
, "DCA enabled\n");
5260 igb_setup_dca(adapter
);
5263 /* Fall Through since DCA is disabled. */
5264 case DCA_PROVIDER_REMOVE
:
5265 if (adapter
->flags
& IGB_FLAG_DCA_ENABLED
) {
5266 /* without this a class_device is left
5267 * hanging around in the sysfs model
5269 dca_remove_requester(dev
);
5270 dev_info(&pdev
->dev
, "DCA disabled\n");
5271 adapter
->flags
&= ~IGB_FLAG_DCA_ENABLED
;
5272 wr32(E1000_DCA_CTRL
, E1000_DCA_CTRL_DCA_MODE_DISABLE
);
5280 static int igb_notify_dca(struct notifier_block
*nb
, unsigned long event
,
5285 ret_val
= driver_for_each_device(&igb_driver
.driver
, NULL
, &event
,
5288 return ret_val
? NOTIFY_BAD
: NOTIFY_DONE
;
5290 #endif /* CONFIG_IGB_DCA */
5292 #ifdef CONFIG_PCI_IOV
5293 static int igb_vf_configure(struct igb_adapter
*adapter
, int vf
)
5295 unsigned char mac_addr
[ETH_ALEN
];
5297 eth_zero_addr(mac_addr
);
5298 igb_set_vf_mac(adapter
, vf
, mac_addr
);
5300 /* By default spoof check is enabled for all VFs */
5301 adapter
->vf_data
[vf
].spoofchk_enabled
= true;
5307 static void igb_ping_all_vfs(struct igb_adapter
*adapter
)
5309 struct e1000_hw
*hw
= &adapter
->hw
;
5313 for (i
= 0 ; i
< adapter
->vfs_allocated_count
; i
++) {
5314 ping
= E1000_PF_CONTROL_MSG
;
5315 if (adapter
->vf_data
[i
].flags
& IGB_VF_FLAG_CTS
)
5316 ping
|= E1000_VT_MSGTYPE_CTS
;
5317 igb_write_mbx(hw
, &ping
, 1, i
);
5321 static int igb_set_vf_promisc(struct igb_adapter
*adapter
, u32
*msgbuf
, u32 vf
)
5323 struct e1000_hw
*hw
= &adapter
->hw
;
5324 u32 vmolr
= rd32(E1000_VMOLR(vf
));
5325 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5327 vf_data
->flags
&= ~(IGB_VF_FLAG_UNI_PROMISC
|
5328 IGB_VF_FLAG_MULTI_PROMISC
);
5329 vmolr
&= ~(E1000_VMOLR_ROPE
| E1000_VMOLR_ROMPE
| E1000_VMOLR_MPME
);
5331 if (*msgbuf
& E1000_VF_SET_PROMISC_MULTICAST
) {
5332 vmolr
|= E1000_VMOLR_MPME
;
5333 vf_data
->flags
|= IGB_VF_FLAG_MULTI_PROMISC
;
5334 *msgbuf
&= ~E1000_VF_SET_PROMISC_MULTICAST
;
5336 /* if we have hashes and we are clearing a multicast promisc
5337 * flag we need to write the hashes to the MTA as this step
5338 * was previously skipped
5340 if (vf_data
->num_vf_mc_hashes
> 30) {
5341 vmolr
|= E1000_VMOLR_MPME
;
5342 } else if (vf_data
->num_vf_mc_hashes
) {
5344 vmolr
|= E1000_VMOLR_ROMPE
;
5345 for (j
= 0; j
< vf_data
->num_vf_mc_hashes
; j
++)
5346 igb_mta_set(hw
, vf_data
->vf_mc_hashes
[j
]);
5350 wr32(E1000_VMOLR(vf
), vmolr
);
5352 /* there are flags left unprocessed, likely not supported */
5353 if (*msgbuf
& E1000_VT_MSGINFO_MASK
)
5359 static int igb_set_vf_multicasts(struct igb_adapter
*adapter
,
5360 u32
*msgbuf
, u32 vf
)
5362 int n
= (msgbuf
[0] & E1000_VT_MSGINFO_MASK
) >> E1000_VT_MSGINFO_SHIFT
;
5363 u16
*hash_list
= (u16
*)&msgbuf
[1];
5364 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5367 /* salt away the number of multicast addresses assigned
5368 * to this VF for later use to restore when the PF multi cast
5371 vf_data
->num_vf_mc_hashes
= n
;
5373 /* only up to 30 hash values supported */
5377 /* store the hashes for later use */
5378 for (i
= 0; i
< n
; i
++)
5379 vf_data
->vf_mc_hashes
[i
] = hash_list
[i
];
5381 /* Flush and reset the mta with the new values */
5382 igb_set_rx_mode(adapter
->netdev
);
5387 static void igb_restore_vf_multicasts(struct igb_adapter
*adapter
)
5389 struct e1000_hw
*hw
= &adapter
->hw
;
5390 struct vf_data_storage
*vf_data
;
5393 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++) {
5394 u32 vmolr
= rd32(E1000_VMOLR(i
));
5395 vmolr
&= ~(E1000_VMOLR_ROMPE
| E1000_VMOLR_MPME
);
5397 vf_data
= &adapter
->vf_data
[i
];
5399 if ((vf_data
->num_vf_mc_hashes
> 30) ||
5400 (vf_data
->flags
& IGB_VF_FLAG_MULTI_PROMISC
)) {
5401 vmolr
|= E1000_VMOLR_MPME
;
5402 } else if (vf_data
->num_vf_mc_hashes
) {
5403 vmolr
|= E1000_VMOLR_ROMPE
;
5404 for (j
= 0; j
< vf_data
->num_vf_mc_hashes
; j
++)
5405 igb_mta_set(hw
, vf_data
->vf_mc_hashes
[j
]);
5407 wr32(E1000_VMOLR(i
), vmolr
);
5411 static void igb_clear_vf_vfta(struct igb_adapter
*adapter
, u32 vf
)
5413 struct e1000_hw
*hw
= &adapter
->hw
;
5414 u32 pool_mask
, reg
, vid
;
5417 pool_mask
= 1 << (E1000_VLVF_POOLSEL_SHIFT
+ vf
);
5419 /* Find the vlan filter for this id */
5420 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5421 reg
= rd32(E1000_VLVF(i
));
5423 /* remove the vf from the pool */
5426 /* if pool is empty then remove entry from vfta */
5427 if (!(reg
& E1000_VLVF_POOLSEL_MASK
) &&
5428 (reg
& E1000_VLVF_VLANID_ENABLE
)) {
5430 vid
= reg
& E1000_VLVF_VLANID_MASK
;
5431 igb_vfta_set(hw
, vid
, false);
5434 wr32(E1000_VLVF(i
), reg
);
5437 adapter
->vf_data
[vf
].vlans_enabled
= 0;
5440 static s32
igb_vlvf_set(struct igb_adapter
*adapter
, u32 vid
, bool add
, u32 vf
)
5442 struct e1000_hw
*hw
= &adapter
->hw
;
5445 /* The vlvf table only exists on 82576 hardware and newer */
5446 if (hw
->mac
.type
< e1000_82576
)
5449 /* we only need to do this if VMDq is enabled */
5450 if (!adapter
->vfs_allocated_count
)
5453 /* Find the vlan filter for this id */
5454 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5455 reg
= rd32(E1000_VLVF(i
));
5456 if ((reg
& E1000_VLVF_VLANID_ENABLE
) &&
5457 vid
== (reg
& E1000_VLVF_VLANID_MASK
))
5462 if (i
== E1000_VLVF_ARRAY_SIZE
) {
5463 /* Did not find a matching VLAN ID entry that was
5464 * enabled. Search for a free filter entry, i.e.
5465 * one without the enable bit set
5467 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5468 reg
= rd32(E1000_VLVF(i
));
5469 if (!(reg
& E1000_VLVF_VLANID_ENABLE
))
5473 if (i
< E1000_VLVF_ARRAY_SIZE
) {
5474 /* Found an enabled/available entry */
5475 reg
|= 1 << (E1000_VLVF_POOLSEL_SHIFT
+ vf
);
5477 /* if !enabled we need to set this up in vfta */
5478 if (!(reg
& E1000_VLVF_VLANID_ENABLE
)) {
5479 /* add VID to filter table */
5480 igb_vfta_set(hw
, vid
, true);
5481 reg
|= E1000_VLVF_VLANID_ENABLE
;
5483 reg
&= ~E1000_VLVF_VLANID_MASK
;
5485 wr32(E1000_VLVF(i
), reg
);
5487 /* do not modify RLPML for PF devices */
5488 if (vf
>= adapter
->vfs_allocated_count
)
5491 if (!adapter
->vf_data
[vf
].vlans_enabled
) {
5493 reg
= rd32(E1000_VMOLR(vf
));
5494 size
= reg
& E1000_VMOLR_RLPML_MASK
;
5496 reg
&= ~E1000_VMOLR_RLPML_MASK
;
5498 wr32(E1000_VMOLR(vf
), reg
);
5501 adapter
->vf_data
[vf
].vlans_enabled
++;
5504 if (i
< E1000_VLVF_ARRAY_SIZE
) {
5505 /* remove vf from the pool */
5506 reg
&= ~(1 << (E1000_VLVF_POOLSEL_SHIFT
+ vf
));
5507 /* if pool is empty then remove entry from vfta */
5508 if (!(reg
& E1000_VLVF_POOLSEL_MASK
)) {
5510 igb_vfta_set(hw
, vid
, false);
5512 wr32(E1000_VLVF(i
), reg
);
5514 /* do not modify RLPML for PF devices */
5515 if (vf
>= adapter
->vfs_allocated_count
)
5518 adapter
->vf_data
[vf
].vlans_enabled
--;
5519 if (!adapter
->vf_data
[vf
].vlans_enabled
) {
5521 reg
= rd32(E1000_VMOLR(vf
));
5522 size
= reg
& E1000_VMOLR_RLPML_MASK
;
5524 reg
&= ~E1000_VMOLR_RLPML_MASK
;
5526 wr32(E1000_VMOLR(vf
), reg
);
5533 static void igb_set_vmvir(struct igb_adapter
*adapter
, u32 vid
, u32 vf
)
5535 struct e1000_hw
*hw
= &adapter
->hw
;
5538 wr32(E1000_VMVIR(vf
), (vid
| E1000_VMVIR_VLANA_DEFAULT
));
5540 wr32(E1000_VMVIR(vf
), 0);
5543 static int igb_ndo_set_vf_vlan(struct net_device
*netdev
,
5544 int vf
, u16 vlan
, u8 qos
)
5547 struct igb_adapter
*adapter
= netdev_priv(netdev
);
5549 if ((vf
>= adapter
->vfs_allocated_count
) || (vlan
> 4095) || (qos
> 7))
5552 err
= igb_vlvf_set(adapter
, vlan
, !!vlan
, vf
);
5555 igb_set_vmvir(adapter
, vlan
| (qos
<< VLAN_PRIO_SHIFT
), vf
);
5556 igb_set_vmolr(adapter
, vf
, !vlan
);
5557 adapter
->vf_data
[vf
].pf_vlan
= vlan
;
5558 adapter
->vf_data
[vf
].pf_qos
= qos
;
5559 dev_info(&adapter
->pdev
->dev
,
5560 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan
, qos
, vf
);
5561 if (test_bit(__IGB_DOWN
, &adapter
->state
)) {
5562 dev_warn(&adapter
->pdev
->dev
,
5563 "The VF VLAN has been set, but the PF device is not up.\n");
5564 dev_warn(&adapter
->pdev
->dev
,
5565 "Bring the PF device up before attempting to use the VF device.\n");
5568 igb_vlvf_set(adapter
, adapter
->vf_data
[vf
].pf_vlan
,
5570 igb_set_vmvir(adapter
, vlan
, vf
);
5571 igb_set_vmolr(adapter
, vf
, true);
5572 adapter
->vf_data
[vf
].pf_vlan
= 0;
5573 adapter
->vf_data
[vf
].pf_qos
= 0;
5579 static int igb_find_vlvf_entry(struct igb_adapter
*adapter
, int vid
)
5581 struct e1000_hw
*hw
= &adapter
->hw
;
5585 /* Find the vlan filter for this id */
5586 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5587 reg
= rd32(E1000_VLVF(i
));
5588 if ((reg
& E1000_VLVF_VLANID_ENABLE
) &&
5589 vid
== (reg
& E1000_VLVF_VLANID_MASK
))
5593 if (i
>= E1000_VLVF_ARRAY_SIZE
)
5599 static int igb_set_vf_vlan(struct igb_adapter
*adapter
, u32
*msgbuf
, u32 vf
)
5601 struct e1000_hw
*hw
= &adapter
->hw
;
5602 int add
= (msgbuf
[0] & E1000_VT_MSGINFO_MASK
) >> E1000_VT_MSGINFO_SHIFT
;
5603 int vid
= (msgbuf
[1] & E1000_VLVF_VLANID_MASK
);
5606 /* If in promiscuous mode we need to make sure the PF also has
5607 * the VLAN filter set.
5609 if (add
&& (adapter
->netdev
->flags
& IFF_PROMISC
))
5610 err
= igb_vlvf_set(adapter
, vid
, add
,
5611 adapter
->vfs_allocated_count
);
5615 err
= igb_vlvf_set(adapter
, vid
, add
, vf
);
5620 /* Go through all the checks to see if the VLAN filter should
5621 * be wiped completely.
5623 if (!add
&& (adapter
->netdev
->flags
& IFF_PROMISC
)) {
5626 int regndx
= igb_find_vlvf_entry(adapter
, vid
);
5629 /* See if any other pools are set for this VLAN filter
5630 * entry other than the PF.
5632 vlvf
= bits
= rd32(E1000_VLVF(regndx
));
5633 bits
&= 1 << (E1000_VLVF_POOLSEL_SHIFT
+
5634 adapter
->vfs_allocated_count
);
5635 /* If the filter was removed then ensure PF pool bit
5636 * is cleared if the PF only added itself to the pool
5637 * because the PF is in promiscuous mode.
5639 if ((vlvf
& VLAN_VID_MASK
) == vid
&&
5640 !test_bit(vid
, adapter
->active_vlans
) &&
5642 igb_vlvf_set(adapter
, vid
, add
,
5643 adapter
->vfs_allocated_count
);
5650 static inline void igb_vf_reset(struct igb_adapter
*adapter
, u32 vf
)
5652 /* clear flags - except flag that indicates PF has set the MAC */
5653 adapter
->vf_data
[vf
].flags
&= IGB_VF_FLAG_PF_SET_MAC
;
5654 adapter
->vf_data
[vf
].last_nack
= jiffies
;
5656 /* reset offloads to defaults */
5657 igb_set_vmolr(adapter
, vf
, true);
5659 /* reset vlans for device */
5660 igb_clear_vf_vfta(adapter
, vf
);
5661 if (adapter
->vf_data
[vf
].pf_vlan
)
5662 igb_ndo_set_vf_vlan(adapter
->netdev
, vf
,
5663 adapter
->vf_data
[vf
].pf_vlan
,
5664 adapter
->vf_data
[vf
].pf_qos
);
5666 igb_clear_vf_vfta(adapter
, vf
);
5668 /* reset multicast table array for vf */
5669 adapter
->vf_data
[vf
].num_vf_mc_hashes
= 0;
5671 /* Flush and reset the mta with the new values */
5672 igb_set_rx_mode(adapter
->netdev
);
5675 static void igb_vf_reset_event(struct igb_adapter
*adapter
, u32 vf
)
5677 unsigned char *vf_mac
= adapter
->vf_data
[vf
].vf_mac_addresses
;
5679 /* clear mac address as we were hotplug removed/added */
5680 if (!(adapter
->vf_data
[vf
].flags
& IGB_VF_FLAG_PF_SET_MAC
))
5681 eth_zero_addr(vf_mac
);
5683 /* process remaining reset events */
5684 igb_vf_reset(adapter
, vf
);
5687 static void igb_vf_reset_msg(struct igb_adapter
*adapter
, u32 vf
)
5689 struct e1000_hw
*hw
= &adapter
->hw
;
5690 unsigned char *vf_mac
= adapter
->vf_data
[vf
].vf_mac_addresses
;
5691 int rar_entry
= hw
->mac
.rar_entry_count
- (vf
+ 1);
5693 u8
*addr
= (u8
*)(&msgbuf
[1]);
5695 /* process all the same items cleared in a function level reset */
5696 igb_vf_reset(adapter
, vf
);
5698 /* set vf mac address */
5699 igb_rar_set_qsel(adapter
, vf_mac
, rar_entry
, vf
);
5701 /* enable transmit and receive for vf */
5702 reg
= rd32(E1000_VFTE
);
5703 wr32(E1000_VFTE
, reg
| (1 << vf
));
5704 reg
= rd32(E1000_VFRE
);
5705 wr32(E1000_VFRE
, reg
| (1 << vf
));
5707 adapter
->vf_data
[vf
].flags
|= IGB_VF_FLAG_CTS
;
5709 /* reply to reset with ack and vf mac address */
5710 msgbuf
[0] = E1000_VF_RESET
| E1000_VT_MSGTYPE_ACK
;
5711 memcpy(addr
, vf_mac
, ETH_ALEN
);
5712 igb_write_mbx(hw
, msgbuf
, 3, vf
);
5715 static int igb_set_vf_mac_addr(struct igb_adapter
*adapter
, u32
*msg
, int vf
)
5717 /* The VF MAC Address is stored in a packed array of bytes
5718 * starting at the second 32 bit word of the msg array
5720 unsigned char *addr
= (char *)&msg
[1];
5723 if (is_valid_ether_addr(addr
))
5724 err
= igb_set_vf_mac(adapter
, vf
, addr
);
5729 static void igb_rcv_ack_from_vf(struct igb_adapter
*adapter
, u32 vf
)
5731 struct e1000_hw
*hw
= &adapter
->hw
;
5732 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5733 u32 msg
= E1000_VT_MSGTYPE_NACK
;
5735 /* if device isn't clear to send it shouldn't be reading either */
5736 if (!(vf_data
->flags
& IGB_VF_FLAG_CTS
) &&
5737 time_after(jiffies
, vf_data
->last_nack
+ (2 * HZ
))) {
5738 igb_write_mbx(hw
, &msg
, 1, vf
);
5739 vf_data
->last_nack
= jiffies
;
5743 static void igb_rcv_msg_from_vf(struct igb_adapter
*adapter
, u32 vf
)
5745 struct pci_dev
*pdev
= adapter
->pdev
;
5746 u32 msgbuf
[E1000_VFMAILBOX_SIZE
];
5747 struct e1000_hw
*hw
= &adapter
->hw
;
5748 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5751 retval
= igb_read_mbx(hw
, msgbuf
, E1000_VFMAILBOX_SIZE
, vf
);
5754 /* if receive failed revoke VF CTS stats and restart init */
5755 dev_err(&pdev
->dev
, "Error receiving message from VF\n");
5756 vf_data
->flags
&= ~IGB_VF_FLAG_CTS
;
5757 if (!time_after(jiffies
, vf_data
->last_nack
+ (2 * HZ
)))
5762 /* this is a message we already processed, do nothing */
5763 if (msgbuf
[0] & (E1000_VT_MSGTYPE_ACK
| E1000_VT_MSGTYPE_NACK
))
5766 /* until the vf completes a reset it should not be
5767 * allowed to start any configuration.
5769 if (msgbuf
[0] == E1000_VF_RESET
) {
5770 igb_vf_reset_msg(adapter
, vf
);
5774 if (!(vf_data
->flags
& IGB_VF_FLAG_CTS
)) {
5775 if (!time_after(jiffies
, vf_data
->last_nack
+ (2 * HZ
)))
5781 switch ((msgbuf
[0] & 0xFFFF)) {
5782 case E1000_VF_SET_MAC_ADDR
:
5784 if (!(vf_data
->flags
& IGB_VF_FLAG_PF_SET_MAC
))
5785 retval
= igb_set_vf_mac_addr(adapter
, msgbuf
, vf
);
5787 dev_warn(&pdev
->dev
,
5788 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
5791 case E1000_VF_SET_PROMISC
:
5792 retval
= igb_set_vf_promisc(adapter
, msgbuf
, vf
);
5794 case E1000_VF_SET_MULTICAST
:
5795 retval
= igb_set_vf_multicasts(adapter
, msgbuf
, vf
);
5797 case E1000_VF_SET_LPE
:
5798 retval
= igb_set_vf_rlpml(adapter
, msgbuf
[1], vf
);
5800 case E1000_VF_SET_VLAN
:
5802 if (vf_data
->pf_vlan
)
5803 dev_warn(&pdev
->dev
,
5804 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
5807 retval
= igb_set_vf_vlan(adapter
, msgbuf
, vf
);
5810 dev_err(&pdev
->dev
, "Unhandled Msg %08x\n", msgbuf
[0]);
5815 msgbuf
[0] |= E1000_VT_MSGTYPE_CTS
;
5817 /* notify the VF of the results of what it sent us */
5819 msgbuf
[0] |= E1000_VT_MSGTYPE_NACK
;
5821 msgbuf
[0] |= E1000_VT_MSGTYPE_ACK
;
5823 igb_write_mbx(hw
, msgbuf
, 1, vf
);
5826 static void igb_msg_task(struct igb_adapter
*adapter
)
5828 struct e1000_hw
*hw
= &adapter
->hw
;
5831 for (vf
= 0; vf
< adapter
->vfs_allocated_count
; vf
++) {
5832 /* process any reset requests */
5833 if (!igb_check_for_rst(hw
, vf
))
5834 igb_vf_reset_event(adapter
, vf
);
5836 /* process any messages pending */
5837 if (!igb_check_for_msg(hw
, vf
))
5838 igb_rcv_msg_from_vf(adapter
, vf
);
5840 /* process any acks */
5841 if (!igb_check_for_ack(hw
, vf
))
5842 igb_rcv_ack_from_vf(adapter
, vf
);
5847 * igb_set_uta - Set unicast filter table address
5848 * @adapter: board private structure
5850 * The unicast table address is a register array of 32-bit registers.
5851 * The table is meant to be used in a way similar to how the MTA is used
5852 * however due to certain limitations in the hardware it is necessary to
5853 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
5854 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
5856 static void igb_set_uta(struct igb_adapter
*adapter
)
5858 struct e1000_hw
*hw
= &adapter
->hw
;
5861 /* The UTA table only exists on 82576 hardware and newer */
5862 if (hw
->mac
.type
< e1000_82576
)
5865 /* we only need to do this if VMDq is enabled */
5866 if (!adapter
->vfs_allocated_count
)
5869 for (i
= 0; i
< hw
->mac
.uta_reg_count
; i
++)
5870 array_wr32(E1000_UTA
, i
, ~0);
5874 * igb_intr_msi - Interrupt Handler
5875 * @irq: interrupt number
5876 * @data: pointer to a network interface device structure
5878 static irqreturn_t
igb_intr_msi(int irq
, void *data
)
5880 struct igb_adapter
*adapter
= data
;
5881 struct igb_q_vector
*q_vector
= adapter
->q_vector
[0];
5882 struct e1000_hw
*hw
= &adapter
->hw
;
5883 /* read ICR disables interrupts using IAM */
5884 u32 icr
= rd32(E1000_ICR
);
5886 igb_write_itr(q_vector
);
5888 if (icr
& E1000_ICR_DRSTA
)
5889 schedule_work(&adapter
->reset_task
);
5891 if (icr
& E1000_ICR_DOUTSYNC
) {
5892 /* HW is reporting DMA is out of sync */
5893 adapter
->stats
.doosync
++;
5896 if (icr
& (E1000_ICR_RXSEQ
| E1000_ICR_LSC
)) {
5897 hw
->mac
.get_link_status
= 1;
5898 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
5899 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
5902 if (icr
& E1000_ICR_TS
) {
5903 u32 tsicr
= rd32(E1000_TSICR
);
5905 if (tsicr
& E1000_TSICR_TXTS
) {
5906 /* acknowledge the interrupt */
5907 wr32(E1000_TSICR
, E1000_TSICR_TXTS
);
5908 /* retrieve hardware timestamp */
5909 schedule_work(&adapter
->ptp_tx_work
);
5913 napi_schedule(&q_vector
->napi
);
5919 * igb_intr - Legacy Interrupt Handler
5920 * @irq: interrupt number
5921 * @data: pointer to a network interface device structure
5923 static irqreturn_t
igb_intr(int irq
, void *data
)
5925 struct igb_adapter
*adapter
= data
;
5926 struct igb_q_vector
*q_vector
= adapter
->q_vector
[0];
5927 struct e1000_hw
*hw
= &adapter
->hw
;
5928 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
5929 * need for the IMC write
5931 u32 icr
= rd32(E1000_ICR
);
5933 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5934 * not set, then the adapter didn't send an interrupt
5936 if (!(icr
& E1000_ICR_INT_ASSERTED
))
5939 igb_write_itr(q_vector
);
5941 if (icr
& E1000_ICR_DRSTA
)
5942 schedule_work(&adapter
->reset_task
);
5944 if (icr
& E1000_ICR_DOUTSYNC
) {
5945 /* HW is reporting DMA is out of sync */
5946 adapter
->stats
.doosync
++;
5949 if (icr
& (E1000_ICR_RXSEQ
| E1000_ICR_LSC
)) {
5950 hw
->mac
.get_link_status
= 1;
5951 /* guard against interrupt when we're going down */
5952 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
5953 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
5956 if (icr
& E1000_ICR_TS
) {
5957 u32 tsicr
= rd32(E1000_TSICR
);
5959 if (tsicr
& E1000_TSICR_TXTS
) {
5960 /* acknowledge the interrupt */
5961 wr32(E1000_TSICR
, E1000_TSICR_TXTS
);
5962 /* retrieve hardware timestamp */
5963 schedule_work(&adapter
->ptp_tx_work
);
5967 napi_schedule(&q_vector
->napi
);
5972 static void igb_ring_irq_enable(struct igb_q_vector
*q_vector
)
5974 struct igb_adapter
*adapter
= q_vector
->adapter
;
5975 struct e1000_hw
*hw
= &adapter
->hw
;
5977 if ((q_vector
->rx
.ring
&& (adapter
->rx_itr_setting
& 3)) ||
5978 (!q_vector
->rx
.ring
&& (adapter
->tx_itr_setting
& 3))) {
5979 if ((adapter
->num_q_vectors
== 1) && !adapter
->vf_data
)
5980 igb_set_itr(q_vector
);
5982 igb_update_ring_itr(q_vector
);
5985 if (!test_bit(__IGB_DOWN
, &adapter
->state
)) {
5986 if (adapter
->msix_entries
)
5987 wr32(E1000_EIMS
, q_vector
->eims_value
);
5989 igb_irq_enable(adapter
);
5994 * igb_poll - NAPI Rx polling callback
5995 * @napi: napi polling structure
5996 * @budget: count of how many packets we should handle
5998 static int igb_poll(struct napi_struct
*napi
, int budget
)
6000 struct igb_q_vector
*q_vector
= container_of(napi
,
6001 struct igb_q_vector
,
6003 bool clean_complete
= true;
6005 #ifdef CONFIG_IGB_DCA
6006 if (q_vector
->adapter
->flags
& IGB_FLAG_DCA_ENABLED
)
6007 igb_update_dca(q_vector
);
6009 if (q_vector
->tx
.ring
)
6010 clean_complete
= igb_clean_tx_irq(q_vector
);
6012 if (q_vector
->rx
.ring
)
6013 clean_complete
&= igb_clean_rx_irq(q_vector
, budget
);
6015 /* If all work not completed, return budget and keep polling */
6016 if (!clean_complete
)
6019 /* If not enough Rx work done, exit the polling mode */
6020 napi_complete(napi
);
6021 igb_ring_irq_enable(q_vector
);
6027 * igb_clean_tx_irq - Reclaim resources after transmit completes
6028 * @q_vector: pointer to q_vector containing needed info
6030 * returns true if ring is completely cleaned
6032 static bool igb_clean_tx_irq(struct igb_q_vector
*q_vector
)
6034 struct igb_adapter
*adapter
= q_vector
->adapter
;
6035 struct igb_ring
*tx_ring
= q_vector
->tx
.ring
;
6036 struct igb_tx_buffer
*tx_buffer
;
6037 union e1000_adv_tx_desc
*tx_desc
;
6038 unsigned int total_bytes
= 0, total_packets
= 0;
6039 unsigned int budget
= q_vector
->tx
.work_limit
;
6040 unsigned int i
= tx_ring
->next_to_clean
;
6042 if (test_bit(__IGB_DOWN
, &adapter
->state
))
6045 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
6046 tx_desc
= IGB_TX_DESC(tx_ring
, i
);
6047 i
-= tx_ring
->count
;
6050 union e1000_adv_tx_desc
*eop_desc
= tx_buffer
->next_to_watch
;
6052 /* if next_to_watch is not set then there is no work pending */
6056 /* prevent any other reads prior to eop_desc */
6057 read_barrier_depends();
6059 /* if DD is not set pending work has not been completed */
6060 if (!(eop_desc
->wb
.status
& cpu_to_le32(E1000_TXD_STAT_DD
)))
6063 /* clear next_to_watch to prevent false hangs */
6064 tx_buffer
->next_to_watch
= NULL
;
6066 /* update the statistics for this packet */
6067 total_bytes
+= tx_buffer
->bytecount
;
6068 total_packets
+= tx_buffer
->gso_segs
;
6071 dev_kfree_skb_any(tx_buffer
->skb
);
6073 /* unmap skb header data */
6074 dma_unmap_single(tx_ring
->dev
,
6075 dma_unmap_addr(tx_buffer
, dma
),
6076 dma_unmap_len(tx_buffer
, len
),
6079 /* clear tx_buffer data */
6080 tx_buffer
->skb
= NULL
;
6081 dma_unmap_len_set(tx_buffer
, len
, 0);
6083 /* clear last DMA location and unmap remaining buffers */
6084 while (tx_desc
!= eop_desc
) {
6089 i
-= tx_ring
->count
;
6090 tx_buffer
= tx_ring
->tx_buffer_info
;
6091 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
6094 /* unmap any remaining paged data */
6095 if (dma_unmap_len(tx_buffer
, len
)) {
6096 dma_unmap_page(tx_ring
->dev
,
6097 dma_unmap_addr(tx_buffer
, dma
),
6098 dma_unmap_len(tx_buffer
, len
),
6100 dma_unmap_len_set(tx_buffer
, len
, 0);
6104 /* move us one more past the eop_desc for start of next pkt */
6109 i
-= tx_ring
->count
;
6110 tx_buffer
= tx_ring
->tx_buffer_info
;
6111 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
6114 /* issue prefetch for next Tx descriptor */
6117 /* update budget accounting */
6119 } while (likely(budget
));
6121 netdev_tx_completed_queue(txring_txq(tx_ring
),
6122 total_packets
, total_bytes
);
6123 i
+= tx_ring
->count
;
6124 tx_ring
->next_to_clean
= i
;
6125 u64_stats_update_begin(&tx_ring
->tx_syncp
);
6126 tx_ring
->tx_stats
.bytes
+= total_bytes
;
6127 tx_ring
->tx_stats
.packets
+= total_packets
;
6128 u64_stats_update_end(&tx_ring
->tx_syncp
);
6129 q_vector
->tx
.total_bytes
+= total_bytes
;
6130 q_vector
->tx
.total_packets
+= total_packets
;
6132 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG
, &tx_ring
->flags
)) {
6133 struct e1000_hw
*hw
= &adapter
->hw
;
6135 /* Detect a transmit hang in hardware, this serializes the
6136 * check with the clearing of time_stamp and movement of i
6138 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG
, &tx_ring
->flags
);
6139 if (tx_buffer
->next_to_watch
&&
6140 time_after(jiffies
, tx_buffer
->time_stamp
+
6141 (adapter
->tx_timeout_factor
* HZ
)) &&
6142 !(rd32(E1000_STATUS
) & E1000_STATUS_TXOFF
)) {
6144 /* detected Tx unit hang */
6145 dev_err(tx_ring
->dev
,
6146 "Detected Tx Unit Hang\n"
6150 " next_to_use <%x>\n"
6151 " next_to_clean <%x>\n"
6152 "buffer_info[next_to_clean]\n"
6153 " time_stamp <%lx>\n"
6154 " next_to_watch <%p>\n"
6156 " desc.status <%x>\n",
6157 tx_ring
->queue_index
,
6158 rd32(E1000_TDH(tx_ring
->reg_idx
)),
6159 readl(tx_ring
->tail
),
6160 tx_ring
->next_to_use
,
6161 tx_ring
->next_to_clean
,
6162 tx_buffer
->time_stamp
,
6163 tx_buffer
->next_to_watch
,
6165 tx_buffer
->next_to_watch
->wb
.status
);
6166 netif_stop_subqueue(tx_ring
->netdev
,
6167 tx_ring
->queue_index
);
6169 /* we are about to reset, no point in enabling stuff */
6174 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6175 if (unlikely(total_packets
&&
6176 netif_carrier_ok(tx_ring
->netdev
) &&
6177 igb_desc_unused(tx_ring
) >= TX_WAKE_THRESHOLD
)) {
6178 /* Make sure that anybody stopping the queue after this
6179 * sees the new next_to_clean.
6182 if (__netif_subqueue_stopped(tx_ring
->netdev
,
6183 tx_ring
->queue_index
) &&
6184 !(test_bit(__IGB_DOWN
, &adapter
->state
))) {
6185 netif_wake_subqueue(tx_ring
->netdev
,
6186 tx_ring
->queue_index
);
6188 u64_stats_update_begin(&tx_ring
->tx_syncp
);
6189 tx_ring
->tx_stats
.restart_queue
++;
6190 u64_stats_update_end(&tx_ring
->tx_syncp
);
6198 * igb_reuse_rx_page - page flip buffer and store it back on the ring
6199 * @rx_ring: rx descriptor ring to store buffers on
6200 * @old_buff: donor buffer to have page reused
6202 * Synchronizes page for reuse by the adapter
6204 static void igb_reuse_rx_page(struct igb_ring
*rx_ring
,
6205 struct igb_rx_buffer
*old_buff
)
6207 struct igb_rx_buffer
*new_buff
;
6208 u16 nta
= rx_ring
->next_to_alloc
;
6210 new_buff
= &rx_ring
->rx_buffer_info
[nta
];
6212 /* update, and store next to alloc */
6214 rx_ring
->next_to_alloc
= (nta
< rx_ring
->count
) ? nta
: 0;
6216 /* transfer page from old buffer to new buffer */
6217 memcpy(new_buff
, old_buff
, sizeof(struct igb_rx_buffer
));
6219 /* sync the buffer for use by the device */
6220 dma_sync_single_range_for_device(rx_ring
->dev
, old_buff
->dma
,
6221 old_buff
->page_offset
,
6226 static bool igb_can_reuse_rx_page(struct igb_rx_buffer
*rx_buffer
,
6228 unsigned int truesize
)
6230 /* avoid re-using remote pages */
6231 if (unlikely(page_to_nid(page
) != numa_node_id()))
6234 #if (PAGE_SIZE < 8192)
6235 /* if we are only owner of page we can reuse it */
6236 if (unlikely(page_count(page
) != 1))
6239 /* flip page offset to other buffer */
6240 rx_buffer
->page_offset
^= IGB_RX_BUFSZ
;
6242 /* since we are the only owner of the page and we need to
6243 * increment it, just set the value to 2 in order to avoid
6244 * an unnecessary locked operation
6246 atomic_set(&page
->_count
, 2);
6248 /* move offset up to the next cache line */
6249 rx_buffer
->page_offset
+= truesize
;
6251 if (rx_buffer
->page_offset
> (PAGE_SIZE
- IGB_RX_BUFSZ
))
6254 /* bump ref count on page before it is given to the stack */
6262 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6263 * @rx_ring: rx descriptor ring to transact packets on
6264 * @rx_buffer: buffer containing page to add
6265 * @rx_desc: descriptor containing length of buffer written by hardware
6266 * @skb: sk_buff to place the data into
6268 * This function will add the data contained in rx_buffer->page to the skb.
6269 * This is done either through a direct copy if the data in the buffer is
6270 * less than the skb header size, otherwise it will just attach the page as
6271 * a frag to the skb.
6273 * The function will then update the page offset if necessary and return
6274 * true if the buffer can be reused by the adapter.
6276 static bool igb_add_rx_frag(struct igb_ring
*rx_ring
,
6277 struct igb_rx_buffer
*rx_buffer
,
6278 union e1000_adv_rx_desc
*rx_desc
,
6279 struct sk_buff
*skb
)
6281 struct page
*page
= rx_buffer
->page
;
6282 unsigned int size
= le16_to_cpu(rx_desc
->wb
.upper
.length
);
6283 #if (PAGE_SIZE < 8192)
6284 unsigned int truesize
= IGB_RX_BUFSZ
;
6286 unsigned int truesize
= ALIGN(size
, L1_CACHE_BYTES
);
6289 if ((size
<= IGB_RX_HDR_LEN
) && !skb_is_nonlinear(skb
)) {
6290 unsigned char *va
= page_address(page
) + rx_buffer
->page_offset
;
6292 if (igb_test_staterr(rx_desc
, E1000_RXDADV_STAT_TSIP
)) {
6293 igb_ptp_rx_pktstamp(rx_ring
->q_vector
, va
, skb
);
6294 va
+= IGB_TS_HDR_LEN
;
6295 size
-= IGB_TS_HDR_LEN
;
6298 memcpy(__skb_put(skb
, size
), va
, ALIGN(size
, sizeof(long)));
6300 /* we can reuse buffer as-is, just make sure it is local */
6301 if (likely(page_to_nid(page
) == numa_node_id()))
6304 /* this page cannot be reused so discard it */
6309 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
,
6310 rx_buffer
->page_offset
, size
, truesize
);
6312 return igb_can_reuse_rx_page(rx_buffer
, page
, truesize
);
6315 static struct sk_buff
*igb_fetch_rx_buffer(struct igb_ring
*rx_ring
,
6316 union e1000_adv_rx_desc
*rx_desc
,
6317 struct sk_buff
*skb
)
6319 struct igb_rx_buffer
*rx_buffer
;
6322 rx_buffer
= &rx_ring
->rx_buffer_info
[rx_ring
->next_to_clean
];
6324 page
= rx_buffer
->page
;
6328 void *page_addr
= page_address(page
) +
6329 rx_buffer
->page_offset
;
6331 /* prefetch first cache line of first page */
6332 prefetch(page_addr
);
6333 #if L1_CACHE_BYTES < 128
6334 prefetch(page_addr
+ L1_CACHE_BYTES
);
6337 /* allocate a skb to store the frags */
6338 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
6340 if (unlikely(!skb
)) {
6341 rx_ring
->rx_stats
.alloc_failed
++;
6345 /* we will be copying header into skb->data in
6346 * pskb_may_pull so it is in our interest to prefetch
6347 * it now to avoid a possible cache miss
6349 prefetchw(skb
->data
);
6352 /* we are reusing so sync this buffer for CPU use */
6353 dma_sync_single_range_for_cpu(rx_ring
->dev
,
6355 rx_buffer
->page_offset
,
6359 /* pull page into skb */
6360 if (igb_add_rx_frag(rx_ring
, rx_buffer
, rx_desc
, skb
)) {
6361 /* hand second half of page back to the ring */
6362 igb_reuse_rx_page(rx_ring
, rx_buffer
);
6364 /* we are not reusing the buffer so unmap it */
6365 dma_unmap_page(rx_ring
->dev
, rx_buffer
->dma
,
6366 PAGE_SIZE
, DMA_FROM_DEVICE
);
6369 /* clear contents of rx_buffer */
6370 rx_buffer
->page
= NULL
;
6375 static inline void igb_rx_checksum(struct igb_ring
*ring
,
6376 union e1000_adv_rx_desc
*rx_desc
,
6377 struct sk_buff
*skb
)
6379 skb_checksum_none_assert(skb
);
6381 /* Ignore Checksum bit is set */
6382 if (igb_test_staterr(rx_desc
, E1000_RXD_STAT_IXSM
))
6385 /* Rx checksum disabled via ethtool */
6386 if (!(ring
->netdev
->features
& NETIF_F_RXCSUM
))
6389 /* TCP/UDP checksum error bit is set */
6390 if (igb_test_staterr(rx_desc
,
6391 E1000_RXDEXT_STATERR_TCPE
|
6392 E1000_RXDEXT_STATERR_IPE
)) {
6393 /* work around errata with sctp packets where the TCPE aka
6394 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6395 * packets, (aka let the stack check the crc32c)
6397 if (!((skb
->len
== 60) &&
6398 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM
, &ring
->flags
))) {
6399 u64_stats_update_begin(&ring
->rx_syncp
);
6400 ring
->rx_stats
.csum_err
++;
6401 u64_stats_update_end(&ring
->rx_syncp
);
6403 /* let the stack verify checksum errors */
6406 /* It must be a TCP or UDP packet with a valid checksum */
6407 if (igb_test_staterr(rx_desc
, E1000_RXD_STAT_TCPCS
|
6408 E1000_RXD_STAT_UDPCS
))
6409 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
6411 dev_dbg(ring
->dev
, "cksum success: bits %08X\n",
6412 le32_to_cpu(rx_desc
->wb
.upper
.status_error
));
6415 static inline void igb_rx_hash(struct igb_ring
*ring
,
6416 union e1000_adv_rx_desc
*rx_desc
,
6417 struct sk_buff
*skb
)
6419 if (ring
->netdev
->features
& NETIF_F_RXHASH
)
6420 skb
->rxhash
= le32_to_cpu(rx_desc
->wb
.lower
.hi_dword
.rss
);
6424 * igb_is_non_eop - process handling of non-EOP buffers
6425 * @rx_ring: Rx ring being processed
6426 * @rx_desc: Rx descriptor for current buffer
6427 * @skb: current socket buffer containing buffer in progress
6429 * This function updates next to clean. If the buffer is an EOP buffer
6430 * this function exits returning false, otherwise it will place the
6431 * sk_buff in the next buffer to be chained and return true indicating
6432 * that this is in fact a non-EOP buffer.
6434 static bool igb_is_non_eop(struct igb_ring
*rx_ring
,
6435 union e1000_adv_rx_desc
*rx_desc
)
6437 u32 ntc
= rx_ring
->next_to_clean
+ 1;
6439 /* fetch, update, and store next to clean */
6440 ntc
= (ntc
< rx_ring
->count
) ? ntc
: 0;
6441 rx_ring
->next_to_clean
= ntc
;
6443 prefetch(IGB_RX_DESC(rx_ring
, ntc
));
6445 if (likely(igb_test_staterr(rx_desc
, E1000_RXD_STAT_EOP
)))
6452 * igb_get_headlen - determine size of header for LRO/GRO
6453 * @data: pointer to the start of the headers
6454 * @max_len: total length of section to find headers in
6456 * This function is meant to determine the length of headers that will
6457 * be recognized by hardware for LRO, and GRO offloads. The main
6458 * motivation of doing this is to only perform one pull for IPv4 TCP
6459 * packets so that we can do basic things like calculating the gso_size
6460 * based on the average data per packet.
6462 static unsigned int igb_get_headlen(unsigned char *data
,
6463 unsigned int max_len
)
6466 unsigned char *network
;
6469 struct vlan_hdr
*vlan
;
6472 struct ipv6hdr
*ipv6
;
6475 u8 nexthdr
= 0; /* default to not TCP */
6478 /* this should never happen, but better safe than sorry */
6479 if (max_len
< ETH_HLEN
)
6482 /* initialize network frame pointer */
6485 /* set first protocol and move network header forward */
6486 protocol
= hdr
.eth
->h_proto
;
6487 hdr
.network
+= ETH_HLEN
;
6489 /* handle any vlan tag if present */
6490 if (protocol
== __constant_htons(ETH_P_8021Q
)) {
6491 if ((hdr
.network
- data
) > (max_len
- VLAN_HLEN
))
6494 protocol
= hdr
.vlan
->h_vlan_encapsulated_proto
;
6495 hdr
.network
+= VLAN_HLEN
;
6498 /* handle L3 protocols */
6499 if (protocol
== __constant_htons(ETH_P_IP
)) {
6500 if ((hdr
.network
- data
) > (max_len
- sizeof(struct iphdr
)))
6503 /* access ihl as a u8 to avoid unaligned access on ia64 */
6504 hlen
= (hdr
.network
[0] & 0x0F) << 2;
6506 /* verify hlen meets minimum size requirements */
6507 if (hlen
< sizeof(struct iphdr
))
6508 return hdr
.network
- data
;
6510 /* record next protocol if header is present */
6511 if (!(hdr
.ipv4
->frag_off
& htons(IP_OFFSET
)))
6512 nexthdr
= hdr
.ipv4
->protocol
;
6513 } else if (protocol
== __constant_htons(ETH_P_IPV6
)) {
6514 if ((hdr
.network
- data
) > (max_len
- sizeof(struct ipv6hdr
)))
6517 /* record next protocol */
6518 nexthdr
= hdr
.ipv6
->nexthdr
;
6519 hlen
= sizeof(struct ipv6hdr
);
6521 return hdr
.network
- data
;
6524 /* relocate pointer to start of L4 header */
6525 hdr
.network
+= hlen
;
6527 /* finally sort out TCP */
6528 if (nexthdr
== IPPROTO_TCP
) {
6529 if ((hdr
.network
- data
) > (max_len
- sizeof(struct tcphdr
)))
6532 /* access doff as a u8 to avoid unaligned access on ia64 */
6533 hlen
= (hdr
.network
[12] & 0xF0) >> 2;
6535 /* verify hlen meets minimum size requirements */
6536 if (hlen
< sizeof(struct tcphdr
))
6537 return hdr
.network
- data
;
6539 hdr
.network
+= hlen
;
6540 } else if (nexthdr
== IPPROTO_UDP
) {
6541 if ((hdr
.network
- data
) > (max_len
- sizeof(struct udphdr
)))
6544 hdr
.network
+= sizeof(struct udphdr
);
6547 /* If everything has gone correctly hdr.network should be the
6548 * data section of the packet and will be the end of the header.
6549 * If not then it probably represents the end of the last recognized
6552 if ((hdr
.network
- data
) < max_len
)
6553 return hdr
.network
- data
;
6559 * igb_pull_tail - igb specific version of skb_pull_tail
6560 * @rx_ring: rx descriptor ring packet is being transacted on
6561 * @rx_desc: pointer to the EOP Rx descriptor
6562 * @skb: pointer to current skb being adjusted
6564 * This function is an igb specific version of __pskb_pull_tail. The
6565 * main difference between this version and the original function is that
6566 * this function can make several assumptions about the state of things
6567 * that allow for significant optimizations versus the standard function.
6568 * As a result we can do things like drop a frag and maintain an accurate
6569 * truesize for the skb.
6571 static void igb_pull_tail(struct igb_ring
*rx_ring
,
6572 union e1000_adv_rx_desc
*rx_desc
,
6573 struct sk_buff
*skb
)
6575 struct skb_frag_struct
*frag
= &skb_shinfo(skb
)->frags
[0];
6577 unsigned int pull_len
;
6579 /* it is valid to use page_address instead of kmap since we are
6580 * working with pages allocated out of the lomem pool per
6581 * alloc_page(GFP_ATOMIC)
6583 va
= skb_frag_address(frag
);
6585 if (igb_test_staterr(rx_desc
, E1000_RXDADV_STAT_TSIP
)) {
6586 /* retrieve timestamp from buffer */
6587 igb_ptp_rx_pktstamp(rx_ring
->q_vector
, va
, skb
);
6589 /* update pointers to remove timestamp header */
6590 skb_frag_size_sub(frag
, IGB_TS_HDR_LEN
);
6591 frag
->page_offset
+= IGB_TS_HDR_LEN
;
6592 skb
->data_len
-= IGB_TS_HDR_LEN
;
6593 skb
->len
-= IGB_TS_HDR_LEN
;
6595 /* move va to start of packet data */
6596 va
+= IGB_TS_HDR_LEN
;
6599 /* we need the header to contain the greater of either ETH_HLEN or
6600 * 60 bytes if the skb->len is less than 60 for skb_pad.
6602 pull_len
= igb_get_headlen(va
, IGB_RX_HDR_LEN
);
6604 /* align pull length to size of long to optimize memcpy performance */
6605 skb_copy_to_linear_data(skb
, va
, ALIGN(pull_len
, sizeof(long)));
6607 /* update all of the pointers */
6608 skb_frag_size_sub(frag
, pull_len
);
6609 frag
->page_offset
+= pull_len
;
6610 skb
->data_len
-= pull_len
;
6611 skb
->tail
+= pull_len
;
6615 * igb_cleanup_headers - Correct corrupted or empty headers
6616 * @rx_ring: rx descriptor ring packet is being transacted on
6617 * @rx_desc: pointer to the EOP Rx descriptor
6618 * @skb: pointer to current skb being fixed
6620 * Address the case where we are pulling data in on pages only
6621 * and as such no data is present in the skb header.
6623 * In addition if skb is not at least 60 bytes we need to pad it so that
6624 * it is large enough to qualify as a valid Ethernet frame.
6626 * Returns true if an error was encountered and skb was freed.
6628 static bool igb_cleanup_headers(struct igb_ring
*rx_ring
,
6629 union e1000_adv_rx_desc
*rx_desc
,
6630 struct sk_buff
*skb
)
6632 if (unlikely((igb_test_staterr(rx_desc
,
6633 E1000_RXDEXT_ERR_FRAME_ERR_MASK
)))) {
6634 struct net_device
*netdev
= rx_ring
->netdev
;
6635 if (!(netdev
->features
& NETIF_F_RXALL
)) {
6636 dev_kfree_skb_any(skb
);
6641 /* place header in linear portion of buffer */
6642 if (skb_is_nonlinear(skb
))
6643 igb_pull_tail(rx_ring
, rx_desc
, skb
);
6645 /* if skb_pad returns an error the skb was freed */
6646 if (unlikely(skb
->len
< 60)) {
6647 int pad_len
= 60 - skb
->len
;
6649 if (skb_pad(skb
, pad_len
))
6651 __skb_put(skb
, pad_len
);
6658 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
6659 * @rx_ring: rx descriptor ring packet is being transacted on
6660 * @rx_desc: pointer to the EOP Rx descriptor
6661 * @skb: pointer to current skb being populated
6663 * This function checks the ring, descriptor, and packet information in
6664 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
6665 * other fields within the skb.
6667 static void igb_process_skb_fields(struct igb_ring
*rx_ring
,
6668 union e1000_adv_rx_desc
*rx_desc
,
6669 struct sk_buff
*skb
)
6671 struct net_device
*dev
= rx_ring
->netdev
;
6673 igb_rx_hash(rx_ring
, rx_desc
, skb
);
6675 igb_rx_checksum(rx_ring
, rx_desc
, skb
);
6677 igb_ptp_rx_hwtstamp(rx_ring
, rx_desc
, skb
);
6679 if ((dev
->features
& NETIF_F_HW_VLAN_CTAG_RX
) &&
6680 igb_test_staterr(rx_desc
, E1000_RXD_STAT_VP
)) {
6682 if (igb_test_staterr(rx_desc
, E1000_RXDEXT_STATERR_LB
) &&
6683 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP
, &rx_ring
->flags
))
6684 vid
= be16_to_cpu(rx_desc
->wb
.upper
.vlan
);
6686 vid
= le16_to_cpu(rx_desc
->wb
.upper
.vlan
);
6688 __vlan_hwaccel_put_tag(skb
, htons(ETH_P_8021Q
), vid
);
6691 skb_record_rx_queue(skb
, rx_ring
->queue_index
);
6693 skb
->protocol
= eth_type_trans(skb
, rx_ring
->netdev
);
6696 static bool igb_clean_rx_irq(struct igb_q_vector
*q_vector
, const int budget
)
6698 struct igb_ring
*rx_ring
= q_vector
->rx
.ring
;
6699 struct sk_buff
*skb
= rx_ring
->skb
;
6700 unsigned int total_bytes
= 0, total_packets
= 0;
6701 u16 cleaned_count
= igb_desc_unused(rx_ring
);
6704 union e1000_adv_rx_desc
*rx_desc
;
6706 /* return some buffers to hardware, one at a time is too slow */
6707 if (cleaned_count
>= IGB_RX_BUFFER_WRITE
) {
6708 igb_alloc_rx_buffers(rx_ring
, cleaned_count
);
6712 rx_desc
= IGB_RX_DESC(rx_ring
, rx_ring
->next_to_clean
);
6714 if (!igb_test_staterr(rx_desc
, E1000_RXD_STAT_DD
))
6717 /* This memory barrier is needed to keep us from reading
6718 * any other fields out of the rx_desc until we know the
6719 * RXD_STAT_DD bit is set
6723 /* retrieve a buffer from the ring */
6724 skb
= igb_fetch_rx_buffer(rx_ring
, rx_desc
, skb
);
6726 /* exit if we failed to retrieve a buffer */
6732 /* fetch next buffer in frame if non-eop */
6733 if (igb_is_non_eop(rx_ring
, rx_desc
))
6736 /* verify the packet layout is correct */
6737 if (igb_cleanup_headers(rx_ring
, rx_desc
, skb
)) {
6742 /* probably a little skewed due to removing CRC */
6743 total_bytes
+= skb
->len
;
6745 /* populate checksum, timestamp, VLAN, and protocol */
6746 igb_process_skb_fields(rx_ring
, rx_desc
, skb
);
6748 napi_gro_receive(&q_vector
->napi
, skb
);
6750 /* reset skb pointer */
6753 /* update budget accounting */
6755 } while (likely(total_packets
< budget
));
6757 /* place incomplete frames back on ring for completion */
6760 u64_stats_update_begin(&rx_ring
->rx_syncp
);
6761 rx_ring
->rx_stats
.packets
+= total_packets
;
6762 rx_ring
->rx_stats
.bytes
+= total_bytes
;
6763 u64_stats_update_end(&rx_ring
->rx_syncp
);
6764 q_vector
->rx
.total_packets
+= total_packets
;
6765 q_vector
->rx
.total_bytes
+= total_bytes
;
6768 igb_alloc_rx_buffers(rx_ring
, cleaned_count
);
6770 return (total_packets
< budget
);
6773 static bool igb_alloc_mapped_page(struct igb_ring
*rx_ring
,
6774 struct igb_rx_buffer
*bi
)
6776 struct page
*page
= bi
->page
;
6779 /* since we are recycling buffers we should seldom need to alloc */
6783 /* alloc new page for storage */
6784 page
= __skb_alloc_page(GFP_ATOMIC
| __GFP_COLD
, NULL
);
6785 if (unlikely(!page
)) {
6786 rx_ring
->rx_stats
.alloc_failed
++;
6790 /* map page for use */
6791 dma
= dma_map_page(rx_ring
->dev
, page
, 0, PAGE_SIZE
, DMA_FROM_DEVICE
);
6793 /* if mapping failed free memory back to system since
6794 * there isn't much point in holding memory we can't use
6796 if (dma_mapping_error(rx_ring
->dev
, dma
)) {
6799 rx_ring
->rx_stats
.alloc_failed
++;
6805 bi
->page_offset
= 0;
6811 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
6812 * @adapter: address of board private structure
6814 void igb_alloc_rx_buffers(struct igb_ring
*rx_ring
, u16 cleaned_count
)
6816 union e1000_adv_rx_desc
*rx_desc
;
6817 struct igb_rx_buffer
*bi
;
6818 u16 i
= rx_ring
->next_to_use
;
6824 rx_desc
= IGB_RX_DESC(rx_ring
, i
);
6825 bi
= &rx_ring
->rx_buffer_info
[i
];
6826 i
-= rx_ring
->count
;
6829 if (!igb_alloc_mapped_page(rx_ring
, bi
))
6832 /* Refresh the desc even if buffer_addrs didn't change
6833 * because each write-back erases this info.
6835 rx_desc
->read
.pkt_addr
= cpu_to_le64(bi
->dma
+ bi
->page_offset
);
6841 rx_desc
= IGB_RX_DESC(rx_ring
, 0);
6842 bi
= rx_ring
->rx_buffer_info
;
6843 i
-= rx_ring
->count
;
6846 /* clear the hdr_addr for the next_to_use descriptor */
6847 rx_desc
->read
.hdr_addr
= 0;
6850 } while (cleaned_count
);
6852 i
+= rx_ring
->count
;
6854 if (rx_ring
->next_to_use
!= i
) {
6855 /* record the next descriptor to use */
6856 rx_ring
->next_to_use
= i
;
6858 /* update next to alloc since we have filled the ring */
6859 rx_ring
->next_to_alloc
= i
;
6861 /* Force memory writes to complete before letting h/w
6862 * know there are new descriptors to fetch. (Only
6863 * applicable for weak-ordered memory model archs,
6867 writel(i
, rx_ring
->tail
);
6877 static int igb_mii_ioctl(struct net_device
*netdev
, struct ifreq
*ifr
, int cmd
)
6879 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6880 struct mii_ioctl_data
*data
= if_mii(ifr
);
6882 if (adapter
->hw
.phy
.media_type
!= e1000_media_type_copper
)
6887 data
->phy_id
= adapter
->hw
.phy
.addr
;
6890 if (igb_read_phy_reg(&adapter
->hw
, data
->reg_num
& 0x1F,
6907 static int igb_ioctl(struct net_device
*netdev
, struct ifreq
*ifr
, int cmd
)
6913 return igb_mii_ioctl(netdev
, ifr
, cmd
);
6915 return igb_ptp_hwtstamp_ioctl(netdev
, ifr
, cmd
);
6921 s32
igb_read_pcie_cap_reg(struct e1000_hw
*hw
, u32 reg
, u16
*value
)
6923 struct igb_adapter
*adapter
= hw
->back
;
6925 if (pcie_capability_read_word(adapter
->pdev
, reg
, value
))
6926 return -E1000_ERR_CONFIG
;
6931 s32
igb_write_pcie_cap_reg(struct e1000_hw
*hw
, u32 reg
, u16
*value
)
6933 struct igb_adapter
*adapter
= hw
->back
;
6935 if (pcie_capability_write_word(adapter
->pdev
, reg
, *value
))
6936 return -E1000_ERR_CONFIG
;
6941 static void igb_vlan_mode(struct net_device
*netdev
, netdev_features_t features
)
6943 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6944 struct e1000_hw
*hw
= &adapter
->hw
;
6946 bool enable
= !!(features
& NETIF_F_HW_VLAN_CTAG_RX
);
6949 /* enable VLAN tag insert/strip */
6950 ctrl
= rd32(E1000_CTRL
);
6951 ctrl
|= E1000_CTRL_VME
;
6952 wr32(E1000_CTRL
, ctrl
);
6954 /* Disable CFI check */
6955 rctl
= rd32(E1000_RCTL
);
6956 rctl
&= ~E1000_RCTL_CFIEN
;
6957 wr32(E1000_RCTL
, rctl
);
6959 /* disable VLAN tag insert/strip */
6960 ctrl
= rd32(E1000_CTRL
);
6961 ctrl
&= ~E1000_CTRL_VME
;
6962 wr32(E1000_CTRL
, ctrl
);
6965 igb_rlpml_set(adapter
);
6968 static int igb_vlan_rx_add_vid(struct net_device
*netdev
,
6969 __be16 proto
, u16 vid
)
6971 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6972 struct e1000_hw
*hw
= &adapter
->hw
;
6973 int pf_id
= adapter
->vfs_allocated_count
;
6975 /* attempt to add filter to vlvf array */
6976 igb_vlvf_set(adapter
, vid
, true, pf_id
);
6978 /* add the filter since PF can receive vlans w/o entry in vlvf */
6979 igb_vfta_set(hw
, vid
, true);
6981 set_bit(vid
, adapter
->active_vlans
);
6986 static int igb_vlan_rx_kill_vid(struct net_device
*netdev
,
6987 __be16 proto
, u16 vid
)
6989 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6990 struct e1000_hw
*hw
= &adapter
->hw
;
6991 int pf_id
= adapter
->vfs_allocated_count
;
6994 /* remove vlan from VLVF table array */
6995 err
= igb_vlvf_set(adapter
, vid
, false, pf_id
);
6997 /* if vid was not present in VLVF just remove it from table */
6999 igb_vfta_set(hw
, vid
, false);
7001 clear_bit(vid
, adapter
->active_vlans
);
7006 static void igb_restore_vlan(struct igb_adapter
*adapter
)
7010 igb_vlan_mode(adapter
->netdev
, adapter
->netdev
->features
);
7012 for_each_set_bit(vid
, adapter
->active_vlans
, VLAN_N_VID
)
7013 igb_vlan_rx_add_vid(adapter
->netdev
, htons(ETH_P_8021Q
), vid
);
7016 int igb_set_spd_dplx(struct igb_adapter
*adapter
, u32 spd
, u8 dplx
)
7018 struct pci_dev
*pdev
= adapter
->pdev
;
7019 struct e1000_mac_info
*mac
= &adapter
->hw
.mac
;
7023 /* Make sure dplx is at most 1 bit and lsb of speed is not set
7024 * for the switch() below to work
7026 if ((spd
& 1) || (dplx
& ~1))
7029 /* Fiber NIC's only allow 1000 gbps Full duplex
7030 * and 100Mbps Full duplex for 100baseFx sfp
7032 if (adapter
->hw
.phy
.media_type
== e1000_media_type_internal_serdes
) {
7033 switch (spd
+ dplx
) {
7034 case SPEED_10
+ DUPLEX_HALF
:
7035 case SPEED_10
+ DUPLEX_FULL
:
7036 case SPEED_100
+ DUPLEX_HALF
:
7043 switch (spd
+ dplx
) {
7044 case SPEED_10
+ DUPLEX_HALF
:
7045 mac
->forced_speed_duplex
= ADVERTISE_10_HALF
;
7047 case SPEED_10
+ DUPLEX_FULL
:
7048 mac
->forced_speed_duplex
= ADVERTISE_10_FULL
;
7050 case SPEED_100
+ DUPLEX_HALF
:
7051 mac
->forced_speed_duplex
= ADVERTISE_100_HALF
;
7053 case SPEED_100
+ DUPLEX_FULL
:
7054 mac
->forced_speed_duplex
= ADVERTISE_100_FULL
;
7056 case SPEED_1000
+ DUPLEX_FULL
:
7058 adapter
->hw
.phy
.autoneg_advertised
= ADVERTISE_1000_FULL
;
7060 case SPEED_1000
+ DUPLEX_HALF
: /* not supported */
7065 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7066 adapter
->hw
.phy
.mdix
= AUTO_ALL_MODES
;
7071 dev_err(&pdev
->dev
, "Unsupported Speed/Duplex configuration\n");
7075 static int __igb_shutdown(struct pci_dev
*pdev
, bool *enable_wake
,
7078 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7079 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7080 struct e1000_hw
*hw
= &adapter
->hw
;
7081 u32 ctrl
, rctl
, status
;
7082 u32 wufc
= runtime
? E1000_WUFC_LNKC
: adapter
->wol
;
7087 netif_device_detach(netdev
);
7089 if (netif_running(netdev
))
7090 __igb_close(netdev
, true);
7092 igb_clear_interrupt_scheme(adapter
);
7095 retval
= pci_save_state(pdev
);
7100 status
= rd32(E1000_STATUS
);
7101 if (status
& E1000_STATUS_LU
)
7102 wufc
&= ~E1000_WUFC_LNKC
;
7105 igb_setup_rctl(adapter
);
7106 igb_set_rx_mode(netdev
);
7108 /* turn on all-multi mode if wake on multicast is enabled */
7109 if (wufc
& E1000_WUFC_MC
) {
7110 rctl
= rd32(E1000_RCTL
);
7111 rctl
|= E1000_RCTL_MPE
;
7112 wr32(E1000_RCTL
, rctl
);
7115 ctrl
= rd32(E1000_CTRL
);
7116 /* advertise wake from D3Cold */
7117 #define E1000_CTRL_ADVD3WUC 0x00100000
7118 /* phy power management enable */
7119 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7120 ctrl
|= E1000_CTRL_ADVD3WUC
;
7121 wr32(E1000_CTRL
, ctrl
);
7123 /* Allow time for pending master requests to run */
7124 igb_disable_pcie_master(hw
);
7126 wr32(E1000_WUC
, E1000_WUC_PME_EN
);
7127 wr32(E1000_WUFC
, wufc
);
7130 wr32(E1000_WUFC
, 0);
7133 *enable_wake
= wufc
|| adapter
->en_mng_pt
;
7135 igb_power_down_link(adapter
);
7137 igb_power_up_link(adapter
);
7139 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7140 * would have already happened in close and is redundant.
7142 igb_release_hw_control(adapter
);
7144 pci_disable_device(pdev
);
7150 #ifdef CONFIG_PM_SLEEP
7151 static int igb_suspend(struct device
*dev
)
7155 struct pci_dev
*pdev
= to_pci_dev(dev
);
7157 retval
= __igb_shutdown(pdev
, &wake
, 0);
7162 pci_prepare_to_sleep(pdev
);
7164 pci_wake_from_d3(pdev
, false);
7165 pci_set_power_state(pdev
, PCI_D3hot
);
7170 #endif /* CONFIG_PM_SLEEP */
7172 static int igb_resume(struct device
*dev
)
7174 struct pci_dev
*pdev
= to_pci_dev(dev
);
7175 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7176 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7177 struct e1000_hw
*hw
= &adapter
->hw
;
7180 pci_set_power_state(pdev
, PCI_D0
);
7181 pci_restore_state(pdev
);
7182 pci_save_state(pdev
);
7184 err
= pci_enable_device_mem(pdev
);
7187 "igb: Cannot enable PCI device from suspend\n");
7190 pci_set_master(pdev
);
7192 pci_enable_wake(pdev
, PCI_D3hot
, 0);
7193 pci_enable_wake(pdev
, PCI_D3cold
, 0);
7195 if (igb_init_interrupt_scheme(adapter
, true)) {
7196 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
7202 /* let the f/w know that the h/w is now under the control of the
7205 igb_get_hw_control(adapter
);
7207 wr32(E1000_WUS
, ~0);
7209 if (netdev
->flags
& IFF_UP
) {
7211 err
= __igb_open(netdev
, true);
7217 netif_device_attach(netdev
);
7221 #ifdef CONFIG_PM_RUNTIME
7222 static int igb_runtime_idle(struct device
*dev
)
7224 struct pci_dev
*pdev
= to_pci_dev(dev
);
7225 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7226 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7228 if (!igb_has_link(adapter
))
7229 pm_schedule_suspend(dev
, MSEC_PER_SEC
* 5);
7234 static int igb_runtime_suspend(struct device
*dev
)
7236 struct pci_dev
*pdev
= to_pci_dev(dev
);
7240 retval
= __igb_shutdown(pdev
, &wake
, 1);
7245 pci_prepare_to_sleep(pdev
);
7247 pci_wake_from_d3(pdev
, false);
7248 pci_set_power_state(pdev
, PCI_D3hot
);
7254 static int igb_runtime_resume(struct device
*dev
)
7256 return igb_resume(dev
);
7258 #endif /* CONFIG_PM_RUNTIME */
7261 static void igb_shutdown(struct pci_dev
*pdev
)
7265 __igb_shutdown(pdev
, &wake
, 0);
7267 if (system_state
== SYSTEM_POWER_OFF
) {
7268 pci_wake_from_d3(pdev
, wake
);
7269 pci_set_power_state(pdev
, PCI_D3hot
);
7273 #ifdef CONFIG_PCI_IOV
7274 static int igb_sriov_reinit(struct pci_dev
*dev
)
7276 struct net_device
*netdev
= pci_get_drvdata(dev
);
7277 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7278 struct pci_dev
*pdev
= adapter
->pdev
;
7282 if (netif_running(netdev
))
7285 igb_clear_interrupt_scheme(adapter
);
7287 igb_init_queue_configuration(adapter
);
7289 if (igb_init_interrupt_scheme(adapter
, true)) {
7290 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
7294 if (netif_running(netdev
))
7302 static int igb_pci_disable_sriov(struct pci_dev
*dev
)
7304 int err
= igb_disable_sriov(dev
);
7307 err
= igb_sriov_reinit(dev
);
7312 static int igb_pci_enable_sriov(struct pci_dev
*dev
, int num_vfs
)
7314 int err
= igb_enable_sriov(dev
, num_vfs
);
7319 err
= igb_sriov_reinit(dev
);
7328 static int igb_pci_sriov_configure(struct pci_dev
*dev
, int num_vfs
)
7330 #ifdef CONFIG_PCI_IOV
7332 return igb_pci_disable_sriov(dev
);
7334 return igb_pci_enable_sriov(dev
, num_vfs
);
7339 #ifdef CONFIG_NET_POLL_CONTROLLER
7340 /* Polling 'interrupt' - used by things like netconsole to send skbs
7341 * without having to re-enable interrupts. It's not called while
7342 * the interrupt routine is executing.
7344 static void igb_netpoll(struct net_device
*netdev
)
7346 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7347 struct e1000_hw
*hw
= &adapter
->hw
;
7348 struct igb_q_vector
*q_vector
;
7351 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
7352 q_vector
= adapter
->q_vector
[i
];
7353 if (adapter
->msix_entries
)
7354 wr32(E1000_EIMC
, q_vector
->eims_value
);
7356 igb_irq_disable(adapter
);
7357 napi_schedule(&q_vector
->napi
);
7360 #endif /* CONFIG_NET_POLL_CONTROLLER */
7363 * igb_io_error_detected - called when PCI error is detected
7364 * @pdev: Pointer to PCI device
7365 * @state: The current pci connection state
7367 * This function is called after a PCI bus error affecting
7368 * this device has been detected.
7370 static pci_ers_result_t
igb_io_error_detected(struct pci_dev
*pdev
,
7371 pci_channel_state_t state
)
7373 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7374 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7376 netif_device_detach(netdev
);
7378 if (state
== pci_channel_io_perm_failure
)
7379 return PCI_ERS_RESULT_DISCONNECT
;
7381 if (netif_running(netdev
))
7383 pci_disable_device(pdev
);
7385 /* Request a slot slot reset. */
7386 return PCI_ERS_RESULT_NEED_RESET
;
7390 * igb_io_slot_reset - called after the pci bus has been reset.
7391 * @pdev: Pointer to PCI device
7393 * Restart the card from scratch, as if from a cold-boot. Implementation
7394 * resembles the first-half of the igb_resume routine.
7396 static pci_ers_result_t
igb_io_slot_reset(struct pci_dev
*pdev
)
7398 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7399 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7400 struct e1000_hw
*hw
= &adapter
->hw
;
7401 pci_ers_result_t result
;
7404 if (pci_enable_device_mem(pdev
)) {
7406 "Cannot re-enable PCI device after reset.\n");
7407 result
= PCI_ERS_RESULT_DISCONNECT
;
7409 pci_set_master(pdev
);
7410 pci_restore_state(pdev
);
7411 pci_save_state(pdev
);
7413 pci_enable_wake(pdev
, PCI_D3hot
, 0);
7414 pci_enable_wake(pdev
, PCI_D3cold
, 0);
7417 wr32(E1000_WUS
, ~0);
7418 result
= PCI_ERS_RESULT_RECOVERED
;
7421 err
= pci_cleanup_aer_uncorrect_error_status(pdev
);
7424 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
7426 /* non-fatal, continue */
7433 * igb_io_resume - called when traffic can start flowing again.
7434 * @pdev: Pointer to PCI device
7436 * This callback is called when the error recovery driver tells us that
7437 * its OK to resume normal operation. Implementation resembles the
7438 * second-half of the igb_resume routine.
7440 static void igb_io_resume(struct pci_dev
*pdev
)
7442 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7443 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7445 if (netif_running(netdev
)) {
7446 if (igb_up(adapter
)) {
7447 dev_err(&pdev
->dev
, "igb_up failed after reset\n");
7452 netif_device_attach(netdev
);
7454 /* let the f/w know that the h/w is now under the control of the
7457 igb_get_hw_control(adapter
);
7460 static void igb_rar_set_qsel(struct igb_adapter
*adapter
, u8
*addr
, u32 index
,
7463 u32 rar_low
, rar_high
;
7464 struct e1000_hw
*hw
= &adapter
->hw
;
7466 /* HW expects these in little endian so we reverse the byte order
7467 * from network order (big endian) to little endian
7469 rar_low
= ((u32
) addr
[0] | ((u32
) addr
[1] << 8) |
7470 ((u32
) addr
[2] << 16) | ((u32
) addr
[3] << 24));
7471 rar_high
= ((u32
) addr
[4] | ((u32
) addr
[5] << 8));
7473 /* Indicate to hardware the Address is Valid. */
7474 rar_high
|= E1000_RAH_AV
;
7476 if (hw
->mac
.type
== e1000_82575
)
7477 rar_high
|= E1000_RAH_POOL_1
* qsel
;
7479 rar_high
|= E1000_RAH_POOL_1
<< qsel
;
7481 wr32(E1000_RAL(index
), rar_low
);
7483 wr32(E1000_RAH(index
), rar_high
);
7487 static int igb_set_vf_mac(struct igb_adapter
*adapter
,
7488 int vf
, unsigned char *mac_addr
)
7490 struct e1000_hw
*hw
= &adapter
->hw
;
7491 /* VF MAC addresses start at end of receive addresses and moves
7492 * towards the first, as a result a collision should not be possible
7494 int rar_entry
= hw
->mac
.rar_entry_count
- (vf
+ 1);
7496 memcpy(adapter
->vf_data
[vf
].vf_mac_addresses
, mac_addr
, ETH_ALEN
);
7498 igb_rar_set_qsel(adapter
, mac_addr
, rar_entry
, vf
);
7503 static int igb_ndo_set_vf_mac(struct net_device
*netdev
, int vf
, u8
*mac
)
7505 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7506 if (!is_valid_ether_addr(mac
) || (vf
>= adapter
->vfs_allocated_count
))
7508 adapter
->vf_data
[vf
].flags
|= IGB_VF_FLAG_PF_SET_MAC
;
7509 dev_info(&adapter
->pdev
->dev
, "setting MAC %pM on VF %d\n", mac
, vf
);
7510 dev_info(&adapter
->pdev
->dev
,
7511 "Reload the VF driver to make this change effective.");
7512 if (test_bit(__IGB_DOWN
, &adapter
->state
)) {
7513 dev_warn(&adapter
->pdev
->dev
,
7514 "The VF MAC address has been set, but the PF device is not up.\n");
7515 dev_warn(&adapter
->pdev
->dev
,
7516 "Bring the PF device up before attempting to use the VF device.\n");
7518 return igb_set_vf_mac(adapter
, vf
, mac
);
7521 static int igb_link_mbps(int internal_link_speed
)
7523 switch (internal_link_speed
) {
7533 static void igb_set_vf_rate_limit(struct e1000_hw
*hw
, int vf
, int tx_rate
,
7540 /* Calculate the rate factor values to set */
7541 rf_int
= link_speed
/ tx_rate
;
7542 rf_dec
= (link_speed
- (rf_int
* tx_rate
));
7543 rf_dec
= (rf_dec
* (1 << E1000_RTTBCNRC_RF_INT_SHIFT
)) /
7546 bcnrc_val
= E1000_RTTBCNRC_RS_ENA
;
7547 bcnrc_val
|= ((rf_int
<< E1000_RTTBCNRC_RF_INT_SHIFT
) &
7548 E1000_RTTBCNRC_RF_INT_MASK
);
7549 bcnrc_val
|= (rf_dec
& E1000_RTTBCNRC_RF_DEC_MASK
);
7554 wr32(E1000_RTTDQSEL
, vf
); /* vf X uses queue X */
7555 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7556 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7558 wr32(E1000_RTTBCNRM
, 0x14);
7559 wr32(E1000_RTTBCNRC
, bcnrc_val
);
7562 static void igb_check_vf_rate_limit(struct igb_adapter
*adapter
)
7564 int actual_link_speed
, i
;
7565 bool reset_rate
= false;
7567 /* VF TX rate limit was not set or not supported */
7568 if ((adapter
->vf_rate_link_speed
== 0) ||
7569 (adapter
->hw
.mac
.type
!= e1000_82576
))
7572 actual_link_speed
= igb_link_mbps(adapter
->link_speed
);
7573 if (actual_link_speed
!= adapter
->vf_rate_link_speed
) {
7575 adapter
->vf_rate_link_speed
= 0;
7576 dev_info(&adapter
->pdev
->dev
,
7577 "Link speed has been changed. VF Transmit rate is disabled\n");
7580 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++) {
7582 adapter
->vf_data
[i
].tx_rate
= 0;
7584 igb_set_vf_rate_limit(&adapter
->hw
, i
,
7585 adapter
->vf_data
[i
].tx_rate
,
7590 static int igb_ndo_set_vf_bw(struct net_device
*netdev
, int vf
, int tx_rate
)
7592 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7593 struct e1000_hw
*hw
= &adapter
->hw
;
7594 int actual_link_speed
;
7596 if (hw
->mac
.type
!= e1000_82576
)
7599 actual_link_speed
= igb_link_mbps(adapter
->link_speed
);
7600 if ((vf
>= adapter
->vfs_allocated_count
) ||
7601 (!(rd32(E1000_STATUS
) & E1000_STATUS_LU
)) ||
7602 (tx_rate
< 0) || (tx_rate
> actual_link_speed
))
7605 adapter
->vf_rate_link_speed
= actual_link_speed
;
7606 adapter
->vf_data
[vf
].tx_rate
= (u16
)tx_rate
;
7607 igb_set_vf_rate_limit(hw
, vf
, tx_rate
, actual_link_speed
);
7612 static int igb_ndo_set_vf_spoofchk(struct net_device
*netdev
, int vf
,
7615 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7616 struct e1000_hw
*hw
= &adapter
->hw
;
7617 u32 reg_val
, reg_offset
;
7619 if (!adapter
->vfs_allocated_count
)
7622 if (vf
>= adapter
->vfs_allocated_count
)
7625 reg_offset
= (hw
->mac
.type
== e1000_82576
) ? E1000_DTXSWC
: E1000_TXSWC
;
7626 reg_val
= rd32(reg_offset
);
7628 reg_val
|= ((1 << vf
) |
7629 (1 << (vf
+ E1000_DTXSWC_VLAN_SPOOF_SHIFT
)));
7631 reg_val
&= ~((1 << vf
) |
7632 (1 << (vf
+ E1000_DTXSWC_VLAN_SPOOF_SHIFT
)));
7633 wr32(reg_offset
, reg_val
);
7635 adapter
->vf_data
[vf
].spoofchk_enabled
= setting
;
7636 return E1000_SUCCESS
;
7639 static int igb_ndo_get_vf_config(struct net_device
*netdev
,
7640 int vf
, struct ifla_vf_info
*ivi
)
7642 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7643 if (vf
>= adapter
->vfs_allocated_count
)
7646 memcpy(&ivi
->mac
, adapter
->vf_data
[vf
].vf_mac_addresses
, ETH_ALEN
);
7647 ivi
->tx_rate
= adapter
->vf_data
[vf
].tx_rate
;
7648 ivi
->vlan
= adapter
->vf_data
[vf
].pf_vlan
;
7649 ivi
->qos
= adapter
->vf_data
[vf
].pf_qos
;
7650 ivi
->spoofchk
= adapter
->vf_data
[vf
].spoofchk_enabled
;
7654 static void igb_vmm_control(struct igb_adapter
*adapter
)
7656 struct e1000_hw
*hw
= &adapter
->hw
;
7659 switch (hw
->mac
.type
) {
7665 /* replication is not supported for 82575 */
7668 /* notify HW that the MAC is adding vlan tags */
7669 reg
= rd32(E1000_DTXCTL
);
7670 reg
|= E1000_DTXCTL_VLAN_ADDED
;
7671 wr32(E1000_DTXCTL
, reg
);
7673 /* enable replication vlan tag stripping */
7674 reg
= rd32(E1000_RPLOLR
);
7675 reg
|= E1000_RPLOLR_STRVLAN
;
7676 wr32(E1000_RPLOLR
, reg
);
7678 /* none of the above registers are supported by i350 */
7682 if (adapter
->vfs_allocated_count
) {
7683 igb_vmdq_set_loopback_pf(hw
, true);
7684 igb_vmdq_set_replication_pf(hw
, true);
7685 igb_vmdq_set_anti_spoofing_pf(hw
, true,
7686 adapter
->vfs_allocated_count
);
7688 igb_vmdq_set_loopback_pf(hw
, false);
7689 igb_vmdq_set_replication_pf(hw
, false);
7693 static void igb_init_dmac(struct igb_adapter
*adapter
, u32 pba
)
7695 struct e1000_hw
*hw
= &adapter
->hw
;
7699 if (hw
->mac
.type
> e1000_82580
) {
7700 if (adapter
->flags
& IGB_FLAG_DMAC
) {
7703 /* force threshold to 0. */
7704 wr32(E1000_DMCTXTH
, 0);
7706 /* DMA Coalescing high water mark needs to be greater
7707 * than the Rx threshold. Set hwm to PBA - max frame
7708 * size in 16B units, capping it at PBA - 6KB.
7710 hwm
= 64 * pba
- adapter
->max_frame_size
/ 16;
7711 if (hwm
< 64 * (pba
- 6))
7712 hwm
= 64 * (pba
- 6);
7713 reg
= rd32(E1000_FCRTC
);
7714 reg
&= ~E1000_FCRTC_RTH_COAL_MASK
;
7715 reg
|= ((hwm
<< E1000_FCRTC_RTH_COAL_SHIFT
)
7716 & E1000_FCRTC_RTH_COAL_MASK
);
7717 wr32(E1000_FCRTC
, reg
);
7719 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
7720 * frame size, capping it at PBA - 10KB.
7722 dmac_thr
= pba
- adapter
->max_frame_size
/ 512;
7723 if (dmac_thr
< pba
- 10)
7724 dmac_thr
= pba
- 10;
7725 reg
= rd32(E1000_DMACR
);
7726 reg
&= ~E1000_DMACR_DMACTHR_MASK
;
7727 reg
|= ((dmac_thr
<< E1000_DMACR_DMACTHR_SHIFT
)
7728 & E1000_DMACR_DMACTHR_MASK
);
7730 /* transition to L0x or L1 if available..*/
7731 reg
|= (E1000_DMACR_DMAC_EN
| E1000_DMACR_DMAC_LX_MASK
);
7733 /* watchdog timer= +-1000 usec in 32usec intervals */
7736 /* Disable BMC-to-OS Watchdog Enable */
7737 if (hw
->mac
.type
!= e1000_i354
)
7738 reg
&= ~E1000_DMACR_DC_BMC2OSW_EN
;
7740 wr32(E1000_DMACR
, reg
);
7742 /* no lower threshold to disable
7743 * coalescing(smart fifb)-UTRESH=0
7745 wr32(E1000_DMCRTRH
, 0);
7747 reg
= (IGB_DMCTLX_DCFLUSH_DIS
| 0x4);
7749 wr32(E1000_DMCTLX
, reg
);
7751 /* free space in tx packet buffer to wake from
7754 wr32(E1000_DMCTXTH
, (IGB_MIN_TXPBSIZE
-
7755 (IGB_TX_BUF_4096
+ adapter
->max_frame_size
)) >> 6);
7757 /* make low power state decision controlled
7760 reg
= rd32(E1000_PCIEMISC
);
7761 reg
&= ~E1000_PCIEMISC_LX_DECISION
;
7762 wr32(E1000_PCIEMISC
, reg
);
7763 } /* endif adapter->dmac is not disabled */
7764 } else if (hw
->mac
.type
== e1000_82580
) {
7765 u32 reg
= rd32(E1000_PCIEMISC
);
7766 wr32(E1000_PCIEMISC
, reg
& ~E1000_PCIEMISC_LX_DECISION
);
7767 wr32(E1000_DMACR
, 0);
7772 * igb_read_i2c_byte - Reads 8 bit word over I2C
7773 * @hw: pointer to hardware structure
7774 * @byte_offset: byte offset to read
7775 * @dev_addr: device address
7778 * Performs byte read operation over I2C interface at
7779 * a specified device address.
7781 s32
igb_read_i2c_byte(struct e1000_hw
*hw
, u8 byte_offset
,
7782 u8 dev_addr
, u8
*data
)
7784 struct igb_adapter
*adapter
= container_of(hw
, struct igb_adapter
, hw
);
7785 struct i2c_client
*this_client
= adapter
->i2c_client
;
7790 return E1000_ERR_I2C
;
7792 swfw_mask
= E1000_SWFW_PHY0_SM
;
7794 if (hw
->mac
.ops
.acquire_swfw_sync(hw
, swfw_mask
)
7796 return E1000_ERR_SWFW_SYNC
;
7798 status
= i2c_smbus_read_byte_data(this_client
, byte_offset
);
7799 hw
->mac
.ops
.release_swfw_sync(hw
, swfw_mask
);
7802 return E1000_ERR_I2C
;
7805 return E1000_SUCCESS
;
7810 * igb_write_i2c_byte - Writes 8 bit word over I2C
7811 * @hw: pointer to hardware structure
7812 * @byte_offset: byte offset to write
7813 * @dev_addr: device address
7814 * @data: value to write
7816 * Performs byte write operation over I2C interface at
7817 * a specified device address.
7819 s32
igb_write_i2c_byte(struct e1000_hw
*hw
, u8 byte_offset
,
7820 u8 dev_addr
, u8 data
)
7822 struct igb_adapter
*adapter
= container_of(hw
, struct igb_adapter
, hw
);
7823 struct i2c_client
*this_client
= adapter
->i2c_client
;
7825 u16 swfw_mask
= E1000_SWFW_PHY0_SM
;
7828 return E1000_ERR_I2C
;
7830 if (hw
->mac
.ops
.acquire_swfw_sync(hw
, swfw_mask
) != E1000_SUCCESS
)
7831 return E1000_ERR_SWFW_SYNC
;
7832 status
= i2c_smbus_write_byte_data(this_client
, byte_offset
, data
);
7833 hw
->mac
.ops
.release_swfw_sync(hw
, swfw_mask
);
7836 return E1000_ERR_I2C
;
7838 return E1000_SUCCESS
;
7842 int igb_reinit_queues(struct igb_adapter
*adapter
)
7844 struct net_device
*netdev
= adapter
->netdev
;
7845 struct pci_dev
*pdev
= adapter
->pdev
;
7848 if (netif_running(netdev
))
7851 igb_clear_interrupt_scheme(adapter
);
7853 if (igb_init_interrupt_scheme(adapter
, true)) {
7854 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
7858 if (netif_running(netdev
))
7859 err
= igb_open(netdev
);