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
);
188 #ifdef CONFIG_PM_SLEEP
189 static int igb_suspend(struct device
*);
191 static int igb_resume(struct device
*);
192 #ifdef CONFIG_PM_RUNTIME
193 static int igb_runtime_suspend(struct device
*dev
);
194 static int igb_runtime_resume(struct device
*dev
);
195 static int igb_runtime_idle(struct device
*dev
);
197 static const struct dev_pm_ops igb_pm_ops
= {
198 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend
, igb_resume
)
199 SET_RUNTIME_PM_OPS(igb_runtime_suspend
, igb_runtime_resume
,
203 static void igb_shutdown(struct pci_dev
*);
204 static int igb_pci_sriov_configure(struct pci_dev
*dev
, int num_vfs
);
205 #ifdef CONFIG_IGB_DCA
206 static int igb_notify_dca(struct notifier_block
*, unsigned long, void *);
207 static struct notifier_block dca_notifier
= {
208 .notifier_call
= igb_notify_dca
,
213 #ifdef CONFIG_NET_POLL_CONTROLLER
214 /* for netdump / net console */
215 static void igb_netpoll(struct net_device
*);
217 #ifdef CONFIG_PCI_IOV
218 static unsigned int max_vfs
= 0;
219 module_param(max_vfs
, uint
, 0);
220 MODULE_PARM_DESC(max_vfs
, "Maximum number of virtual functions to allocate "
221 "per physical function");
222 #endif /* CONFIG_PCI_IOV */
224 static pci_ers_result_t
igb_io_error_detected(struct pci_dev
*,
225 pci_channel_state_t
);
226 static pci_ers_result_t
igb_io_slot_reset(struct pci_dev
*);
227 static void igb_io_resume(struct pci_dev
*);
229 static const struct pci_error_handlers igb_err_handler
= {
230 .error_detected
= igb_io_error_detected
,
231 .slot_reset
= igb_io_slot_reset
,
232 .resume
= igb_io_resume
,
235 static void igb_init_dmac(struct igb_adapter
*adapter
, u32 pba
);
237 static struct pci_driver igb_driver
= {
238 .name
= igb_driver_name
,
239 .id_table
= igb_pci_tbl
,
241 .remove
= igb_remove
,
243 .driver
.pm
= &igb_pm_ops
,
245 .shutdown
= igb_shutdown
,
246 .sriov_configure
= igb_pci_sriov_configure
,
247 .err_handler
= &igb_err_handler
250 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
251 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
252 MODULE_LICENSE("GPL");
253 MODULE_VERSION(DRV_VERSION
);
255 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
256 static int debug
= -1;
257 module_param(debug
, int, 0);
258 MODULE_PARM_DESC(debug
, "Debug level (0=none,...,16=all)");
260 struct igb_reg_info
{
265 static const struct igb_reg_info igb_reg_info_tbl
[] = {
267 /* General Registers */
268 {E1000_CTRL
, "CTRL"},
269 {E1000_STATUS
, "STATUS"},
270 {E1000_CTRL_EXT
, "CTRL_EXT"},
272 /* Interrupt Registers */
276 {E1000_RCTL
, "RCTL"},
277 {E1000_RDLEN(0), "RDLEN"},
278 {E1000_RDH(0), "RDH"},
279 {E1000_RDT(0), "RDT"},
280 {E1000_RXDCTL(0), "RXDCTL"},
281 {E1000_RDBAL(0), "RDBAL"},
282 {E1000_RDBAH(0), "RDBAH"},
285 {E1000_TCTL
, "TCTL"},
286 {E1000_TDBAL(0), "TDBAL"},
287 {E1000_TDBAH(0), "TDBAH"},
288 {E1000_TDLEN(0), "TDLEN"},
289 {E1000_TDH(0), "TDH"},
290 {E1000_TDT(0), "TDT"},
291 {E1000_TXDCTL(0), "TXDCTL"},
292 {E1000_TDFH
, "TDFH"},
293 {E1000_TDFT
, "TDFT"},
294 {E1000_TDFHS
, "TDFHS"},
295 {E1000_TDFPC
, "TDFPC"},
297 /* List Terminator */
301 /* igb_regdump - register printout routine */
302 static void igb_regdump(struct e1000_hw
*hw
, struct igb_reg_info
*reginfo
)
308 switch (reginfo
->ofs
) {
310 for (n
= 0; n
< 4; n
++)
311 regs
[n
] = rd32(E1000_RDLEN(n
));
314 for (n
= 0; n
< 4; n
++)
315 regs
[n
] = rd32(E1000_RDH(n
));
318 for (n
= 0; n
< 4; n
++)
319 regs
[n
] = rd32(E1000_RDT(n
));
321 case E1000_RXDCTL(0):
322 for (n
= 0; n
< 4; n
++)
323 regs
[n
] = rd32(E1000_RXDCTL(n
));
326 for (n
= 0; n
< 4; n
++)
327 regs
[n
] = rd32(E1000_RDBAL(n
));
330 for (n
= 0; n
< 4; n
++)
331 regs
[n
] = rd32(E1000_RDBAH(n
));
334 for (n
= 0; n
< 4; n
++)
335 regs
[n
] = rd32(E1000_RDBAL(n
));
338 for (n
= 0; n
< 4; n
++)
339 regs
[n
] = rd32(E1000_TDBAH(n
));
342 for (n
= 0; n
< 4; n
++)
343 regs
[n
] = rd32(E1000_TDLEN(n
));
346 for (n
= 0; n
< 4; n
++)
347 regs
[n
] = rd32(E1000_TDH(n
));
350 for (n
= 0; n
< 4; n
++)
351 regs
[n
] = rd32(E1000_TDT(n
));
353 case E1000_TXDCTL(0):
354 for (n
= 0; n
< 4; n
++)
355 regs
[n
] = rd32(E1000_TXDCTL(n
));
358 pr_info("%-15s %08x\n", reginfo
->name
, rd32(reginfo
->ofs
));
362 snprintf(rname
, 16, "%s%s", reginfo
->name
, "[0-3]");
363 pr_info("%-15s %08x %08x %08x %08x\n", rname
, regs
[0], regs
[1],
367 /* igb_dump - Print registers, Tx-rings and Rx-rings */
368 static void igb_dump(struct igb_adapter
*adapter
)
370 struct net_device
*netdev
= adapter
->netdev
;
371 struct e1000_hw
*hw
= &adapter
->hw
;
372 struct igb_reg_info
*reginfo
;
373 struct igb_ring
*tx_ring
;
374 union e1000_adv_tx_desc
*tx_desc
;
375 struct my_u0
{ u64 a
; u64 b
; } *u0
;
376 struct igb_ring
*rx_ring
;
377 union e1000_adv_rx_desc
*rx_desc
;
381 if (!netif_msg_hw(adapter
))
384 /* Print netdevice Info */
386 dev_info(&adapter
->pdev
->dev
, "Net device Info\n");
387 pr_info("Device Name state trans_start "
389 pr_info("%-15s %016lX %016lX %016lX\n", netdev
->name
,
390 netdev
->state
, netdev
->trans_start
, netdev
->last_rx
);
393 /* Print Registers */
394 dev_info(&adapter
->pdev
->dev
, "Register Dump\n");
395 pr_info(" Register Name Value\n");
396 for (reginfo
= (struct igb_reg_info
*)igb_reg_info_tbl
;
397 reginfo
->name
; reginfo
++) {
398 igb_regdump(hw
, reginfo
);
401 /* Print TX Ring Summary */
402 if (!netdev
|| !netif_running(netdev
))
405 dev_info(&adapter
->pdev
->dev
, "TX Rings Summary\n");
406 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
407 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
408 struct igb_tx_buffer
*buffer_info
;
409 tx_ring
= adapter
->tx_ring
[n
];
410 buffer_info
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_clean
];
411 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
412 n
, tx_ring
->next_to_use
, tx_ring
->next_to_clean
,
413 (u64
)dma_unmap_addr(buffer_info
, dma
),
414 dma_unmap_len(buffer_info
, len
),
415 buffer_info
->next_to_watch
,
416 (u64
)buffer_info
->time_stamp
);
420 if (!netif_msg_tx_done(adapter
))
421 goto rx_ring_summary
;
423 dev_info(&adapter
->pdev
->dev
, "TX Rings Dump\n");
425 /* Transmit Descriptor Formats
427 * Advanced Transmit Descriptor
428 * +--------------------------------------------------------------+
429 * 0 | Buffer Address [63:0] |
430 * +--------------------------------------------------------------+
431 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
432 * +--------------------------------------------------------------+
433 * 63 46 45 40 39 38 36 35 32 31 24 15 0
436 for (n
= 0; n
< adapter
->num_tx_queues
; n
++) {
437 tx_ring
= adapter
->tx_ring
[n
];
438 pr_info("------------------------------------\n");
439 pr_info("TX QUEUE INDEX = %d\n", tx_ring
->queue_index
);
440 pr_info("------------------------------------\n");
441 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] "
442 "[bi->dma ] leng ntw timestamp "
445 for (i
= 0; tx_ring
->desc
&& (i
< tx_ring
->count
); i
++) {
446 const char *next_desc
;
447 struct igb_tx_buffer
*buffer_info
;
448 tx_desc
= IGB_TX_DESC(tx_ring
, i
);
449 buffer_info
= &tx_ring
->tx_buffer_info
[i
];
450 u0
= (struct my_u0
*)tx_desc
;
451 if (i
== tx_ring
->next_to_use
&&
452 i
== tx_ring
->next_to_clean
)
453 next_desc
= " NTC/U";
454 else if (i
== tx_ring
->next_to_use
)
456 else if (i
== tx_ring
->next_to_clean
)
461 pr_info("T [0x%03X] %016llX %016llX %016llX"
462 " %04X %p %016llX %p%s\n", i
,
465 (u64
)dma_unmap_addr(buffer_info
, dma
),
466 dma_unmap_len(buffer_info
, len
),
467 buffer_info
->next_to_watch
,
468 (u64
)buffer_info
->time_stamp
,
469 buffer_info
->skb
, next_desc
);
471 if (netif_msg_pktdata(adapter
) && buffer_info
->skb
)
472 print_hex_dump(KERN_INFO
, "",
474 16, 1, buffer_info
->skb
->data
,
475 dma_unmap_len(buffer_info
, len
),
480 /* Print RX Rings Summary */
482 dev_info(&adapter
->pdev
->dev
, "RX Rings Summary\n");
483 pr_info("Queue [NTU] [NTC]\n");
484 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
485 rx_ring
= adapter
->rx_ring
[n
];
486 pr_info(" %5d %5X %5X\n",
487 n
, rx_ring
->next_to_use
, rx_ring
->next_to_clean
);
491 if (!netif_msg_rx_status(adapter
))
494 dev_info(&adapter
->pdev
->dev
, "RX Rings Dump\n");
496 /* Advanced Receive Descriptor (Read) Format
498 * +-----------------------------------------------------+
499 * 0 | Packet Buffer Address [63:1] |A0/NSE|
500 * +----------------------------------------------+------+
501 * 8 | Header Buffer Address [63:1] | DD |
502 * +-----------------------------------------------------+
505 * Advanced Receive Descriptor (Write-Back) Format
507 * 63 48 47 32 31 30 21 20 17 16 4 3 0
508 * +------------------------------------------------------+
509 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
510 * | Checksum Ident | | | | Type | Type |
511 * +------------------------------------------------------+
512 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
513 * +------------------------------------------------------+
514 * 63 48 47 32 31 20 19 0
517 for (n
= 0; n
< adapter
->num_rx_queues
; n
++) {
518 rx_ring
= adapter
->rx_ring
[n
];
519 pr_info("------------------------------------\n");
520 pr_info("RX QUEUE INDEX = %d\n", rx_ring
->queue_index
);
521 pr_info("------------------------------------\n");
522 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] "
523 "[bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
524 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] -----"
525 "----------- [bi->skb] <-- Adv Rx Write-Back format\n");
527 for (i
= 0; i
< rx_ring
->count
; i
++) {
528 const char *next_desc
;
529 struct igb_rx_buffer
*buffer_info
;
530 buffer_info
= &rx_ring
->rx_buffer_info
[i
];
531 rx_desc
= IGB_RX_DESC(rx_ring
, i
);
532 u0
= (struct my_u0
*)rx_desc
;
533 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
535 if (i
== rx_ring
->next_to_use
)
537 else if (i
== rx_ring
->next_to_clean
)
542 if (staterr
& E1000_RXD_STAT_DD
) {
543 /* Descriptor Done */
544 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n",
550 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
554 (u64
)buffer_info
->dma
,
557 if (netif_msg_pktdata(adapter
) &&
558 buffer_info
->dma
&& buffer_info
->page
) {
559 print_hex_dump(KERN_INFO
, "",
562 page_address(buffer_info
->page
) +
563 buffer_info
->page_offset
,
575 * igb_get_i2c_data - Reads the I2C SDA data bit
576 * @hw: pointer to hardware structure
577 * @i2cctl: Current value of I2CCTL register
579 * Returns the I2C data bit value
581 static int igb_get_i2c_data(void *data
)
583 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
584 struct e1000_hw
*hw
= &adapter
->hw
;
585 s32 i2cctl
= rd32(E1000_I2CPARAMS
);
587 return ((i2cctl
& E1000_I2C_DATA_IN
) != 0);
591 * igb_set_i2c_data - Sets the I2C data bit
592 * @data: pointer to hardware structure
593 * @state: I2C data value (0 or 1) to set
595 * Sets the I2C data bit
597 static void igb_set_i2c_data(void *data
, int state
)
599 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
600 struct e1000_hw
*hw
= &adapter
->hw
;
601 s32 i2cctl
= rd32(E1000_I2CPARAMS
);
604 i2cctl
|= E1000_I2C_DATA_OUT
;
606 i2cctl
&= ~E1000_I2C_DATA_OUT
;
608 i2cctl
&= ~E1000_I2C_DATA_OE_N
;
609 i2cctl
|= E1000_I2C_CLK_OE_N
;
610 wr32(E1000_I2CPARAMS
, i2cctl
);
616 * igb_set_i2c_clk - Sets the I2C SCL clock
617 * @data: pointer to hardware structure
618 * @state: state to set clock
620 * Sets the I2C clock line to state
622 static void igb_set_i2c_clk(void *data
, int state
)
624 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
625 struct e1000_hw
*hw
= &adapter
->hw
;
626 s32 i2cctl
= rd32(E1000_I2CPARAMS
);
629 i2cctl
|= E1000_I2C_CLK_OUT
;
630 i2cctl
&= ~E1000_I2C_CLK_OE_N
;
632 i2cctl
&= ~E1000_I2C_CLK_OUT
;
633 i2cctl
&= ~E1000_I2C_CLK_OE_N
;
635 wr32(E1000_I2CPARAMS
, i2cctl
);
640 * igb_get_i2c_clk - Gets the I2C SCL clock state
641 * @data: pointer to hardware structure
643 * Gets the I2C clock state
645 static int igb_get_i2c_clk(void *data
)
647 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
648 struct e1000_hw
*hw
= &adapter
->hw
;
649 s32 i2cctl
= rd32(E1000_I2CPARAMS
);
651 return ((i2cctl
& E1000_I2C_CLK_IN
) != 0);
654 static const struct i2c_algo_bit_data igb_i2c_algo
= {
655 .setsda
= igb_set_i2c_data
,
656 .setscl
= igb_set_i2c_clk
,
657 .getsda
= igb_get_i2c_data
,
658 .getscl
= igb_get_i2c_clk
,
664 * igb_get_hw_dev - return device
665 * @hw: pointer to hardware structure
667 * used by hardware layer to print debugging information
669 struct net_device
*igb_get_hw_dev(struct e1000_hw
*hw
)
671 struct igb_adapter
*adapter
= hw
->back
;
672 return adapter
->netdev
;
676 * igb_init_module - Driver Registration Routine
678 * igb_init_module is the first routine called when the driver is
679 * loaded. All it does is register with the PCI subsystem.
681 static int __init
igb_init_module(void)
684 pr_info("%s - version %s\n",
685 igb_driver_string
, igb_driver_version
);
687 pr_info("%s\n", igb_copyright
);
689 #ifdef CONFIG_IGB_DCA
690 dca_register_notify(&dca_notifier
);
692 ret
= pci_register_driver(&igb_driver
);
696 module_init(igb_init_module
);
699 * igb_exit_module - Driver Exit Cleanup Routine
701 * igb_exit_module is called just before the driver is removed
704 static void __exit
igb_exit_module(void)
706 #ifdef CONFIG_IGB_DCA
707 dca_unregister_notify(&dca_notifier
);
709 pci_unregister_driver(&igb_driver
);
712 module_exit(igb_exit_module
);
714 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
716 * igb_cache_ring_register - Descriptor ring to register mapping
717 * @adapter: board private structure to initialize
719 * Once we know the feature-set enabled for the device, we'll cache
720 * the register offset the descriptor ring is assigned to.
722 static void igb_cache_ring_register(struct igb_adapter
*adapter
)
725 u32 rbase_offset
= adapter
->vfs_allocated_count
;
727 switch (adapter
->hw
.mac
.type
) {
729 /* The queues are allocated for virtualization such that VF 0
730 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
731 * In order to avoid collision we start at the first free queue
732 * and continue consuming queues in the same sequence
734 if (adapter
->vfs_allocated_count
) {
735 for (; i
< adapter
->rss_queues
; i
++)
736 adapter
->rx_ring
[i
]->reg_idx
= rbase_offset
+
746 for (; i
< adapter
->num_rx_queues
; i
++)
747 adapter
->rx_ring
[i
]->reg_idx
= rbase_offset
+ i
;
748 for (; j
< adapter
->num_tx_queues
; j
++)
749 adapter
->tx_ring
[j
]->reg_idx
= rbase_offset
+ j
;
755 * igb_write_ivar - configure ivar for given MSI-X vector
756 * @hw: pointer to the HW structure
757 * @msix_vector: vector number we are allocating to a given ring
758 * @index: row index of IVAR register to write within IVAR table
759 * @offset: column offset of in IVAR, should be multiple of 8
761 * This function is intended to handle the writing of the IVAR register
762 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
763 * each containing an cause allocation for an Rx and Tx ring, and a
764 * variable number of rows depending on the number of queues supported.
766 static void igb_write_ivar(struct e1000_hw
*hw
, int msix_vector
,
767 int index
, int offset
)
769 u32 ivar
= array_rd32(E1000_IVAR0
, index
);
771 /* clear any bits that are currently set */
772 ivar
&= ~((u32
)0xFF << offset
);
774 /* write vector and valid bit */
775 ivar
|= (msix_vector
| E1000_IVAR_VALID
) << offset
;
777 array_wr32(E1000_IVAR0
, index
, ivar
);
780 #define IGB_N0_QUEUE -1
781 static void igb_assign_vector(struct igb_q_vector
*q_vector
, int msix_vector
)
783 struct igb_adapter
*adapter
= q_vector
->adapter
;
784 struct e1000_hw
*hw
= &adapter
->hw
;
785 int rx_queue
= IGB_N0_QUEUE
;
786 int tx_queue
= IGB_N0_QUEUE
;
789 if (q_vector
->rx
.ring
)
790 rx_queue
= q_vector
->rx
.ring
->reg_idx
;
791 if (q_vector
->tx
.ring
)
792 tx_queue
= q_vector
->tx
.ring
->reg_idx
;
794 switch (hw
->mac
.type
) {
796 /* The 82575 assigns vectors using a bitmask, which matches the
797 * bitmask for the EICR/EIMS/EIMC registers. To assign one
798 * or more queues to a vector, we write the appropriate bits
799 * into the MSIXBM register for that vector.
801 if (rx_queue
> IGB_N0_QUEUE
)
802 msixbm
= E1000_EICR_RX_QUEUE0
<< rx_queue
;
803 if (tx_queue
> IGB_N0_QUEUE
)
804 msixbm
|= E1000_EICR_TX_QUEUE0
<< tx_queue
;
805 if (!adapter
->msix_entries
&& msix_vector
== 0)
806 msixbm
|= E1000_EIMS_OTHER
;
807 array_wr32(E1000_MSIXBM(0), msix_vector
, msixbm
);
808 q_vector
->eims_value
= msixbm
;
811 /* 82576 uses a table that essentially consists of 2 columns
812 * with 8 rows. The ordering is column-major so we use the
813 * lower 3 bits as the row index, and the 4th bit as the
816 if (rx_queue
> IGB_N0_QUEUE
)
817 igb_write_ivar(hw
, msix_vector
,
819 (rx_queue
& 0x8) << 1);
820 if (tx_queue
> IGB_N0_QUEUE
)
821 igb_write_ivar(hw
, msix_vector
,
823 ((tx_queue
& 0x8) << 1) + 8);
824 q_vector
->eims_value
= 1 << msix_vector
;
831 /* On 82580 and newer adapters the scheme is similar to 82576
832 * however instead of ordering column-major we have things
833 * ordered row-major. So we traverse the table by using
834 * bit 0 as the column offset, and the remaining bits as the
837 if (rx_queue
> IGB_N0_QUEUE
)
838 igb_write_ivar(hw
, msix_vector
,
840 (rx_queue
& 0x1) << 4);
841 if (tx_queue
> IGB_N0_QUEUE
)
842 igb_write_ivar(hw
, msix_vector
,
844 ((tx_queue
& 0x1) << 4) + 8);
845 q_vector
->eims_value
= 1 << msix_vector
;
852 /* add q_vector eims value to global eims_enable_mask */
853 adapter
->eims_enable_mask
|= q_vector
->eims_value
;
855 /* configure q_vector to set itr on first interrupt */
856 q_vector
->set_itr
= 1;
860 * igb_configure_msix - Configure MSI-X hardware
861 * @adapter: board private structure to initialize
863 * igb_configure_msix sets up the hardware to properly
864 * generate MSI-X interrupts.
866 static void igb_configure_msix(struct igb_adapter
*adapter
)
870 struct e1000_hw
*hw
= &adapter
->hw
;
872 adapter
->eims_enable_mask
= 0;
874 /* set vector for other causes, i.e. link changes */
875 switch (hw
->mac
.type
) {
877 tmp
= rd32(E1000_CTRL_EXT
);
878 /* enable MSI-X PBA support*/
879 tmp
|= E1000_CTRL_EXT_PBA_CLR
;
881 /* Auto-Mask interrupts upon ICR read. */
882 tmp
|= E1000_CTRL_EXT_EIAME
;
883 tmp
|= E1000_CTRL_EXT_IRCA
;
885 wr32(E1000_CTRL_EXT
, tmp
);
887 /* enable msix_other interrupt */
888 array_wr32(E1000_MSIXBM(0), vector
++, E1000_EIMS_OTHER
);
889 adapter
->eims_other
= E1000_EIMS_OTHER
;
899 /* Turn on MSI-X capability first, or our settings
900 * won't stick. And it will take days to debug.
902 wr32(E1000_GPIE
, E1000_GPIE_MSIX_MODE
|
903 E1000_GPIE_PBA
| E1000_GPIE_EIAME
|
906 /* enable msix_other interrupt */
907 adapter
->eims_other
= 1 << vector
;
908 tmp
= (vector
++ | E1000_IVAR_VALID
) << 8;
910 wr32(E1000_IVAR_MISC
, tmp
);
913 /* do nothing, since nothing else supports MSI-X */
915 } /* switch (hw->mac.type) */
917 adapter
->eims_enable_mask
|= adapter
->eims_other
;
919 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
920 igb_assign_vector(adapter
->q_vector
[i
], vector
++);
926 * igb_request_msix - Initialize MSI-X interrupts
927 * @adapter: board private structure to initialize
929 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
932 static int igb_request_msix(struct igb_adapter
*adapter
)
934 struct net_device
*netdev
= adapter
->netdev
;
935 struct e1000_hw
*hw
= &adapter
->hw
;
936 int i
, err
= 0, vector
= 0, free_vector
= 0;
938 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
939 igb_msix_other
, 0, netdev
->name
, adapter
);
943 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
944 struct igb_q_vector
*q_vector
= adapter
->q_vector
[i
];
948 q_vector
->itr_register
= hw
->hw_addr
+ E1000_EITR(vector
);
950 if (q_vector
->rx
.ring
&& q_vector
->tx
.ring
)
951 sprintf(q_vector
->name
, "%s-TxRx-%u", netdev
->name
,
952 q_vector
->rx
.ring
->queue_index
);
953 else if (q_vector
->tx
.ring
)
954 sprintf(q_vector
->name
, "%s-tx-%u", netdev
->name
,
955 q_vector
->tx
.ring
->queue_index
);
956 else if (q_vector
->rx
.ring
)
957 sprintf(q_vector
->name
, "%s-rx-%u", netdev
->name
,
958 q_vector
->rx
.ring
->queue_index
);
960 sprintf(q_vector
->name
, "%s-unused", netdev
->name
);
962 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
963 igb_msix_ring
, 0, q_vector
->name
,
969 igb_configure_msix(adapter
);
973 /* free already assigned IRQs */
974 free_irq(adapter
->msix_entries
[free_vector
++].vector
, adapter
);
977 for (i
= 0; i
< vector
; i
++) {
978 free_irq(adapter
->msix_entries
[free_vector
++].vector
,
979 adapter
->q_vector
[i
]);
985 static void igb_reset_interrupt_capability(struct igb_adapter
*adapter
)
987 if (adapter
->msix_entries
) {
988 pci_disable_msix(adapter
->pdev
);
989 kfree(adapter
->msix_entries
);
990 adapter
->msix_entries
= NULL
;
991 } else if (adapter
->flags
& IGB_FLAG_HAS_MSI
) {
992 pci_disable_msi(adapter
->pdev
);
997 * igb_free_q_vector - Free memory allocated for specific interrupt vector
998 * @adapter: board private structure to initialize
999 * @v_idx: Index of vector to be freed
1001 * This function frees the memory allocated to the q_vector. In addition if
1002 * NAPI is enabled it will delete any references to the NAPI struct prior
1003 * to freeing the q_vector.
1005 static void igb_free_q_vector(struct igb_adapter
*adapter
, int v_idx
)
1007 struct igb_q_vector
*q_vector
= adapter
->q_vector
[v_idx
];
1009 if (q_vector
->tx
.ring
)
1010 adapter
->tx_ring
[q_vector
->tx
.ring
->queue_index
] = NULL
;
1012 if (q_vector
->rx
.ring
)
1013 adapter
->tx_ring
[q_vector
->rx
.ring
->queue_index
] = NULL
;
1015 adapter
->q_vector
[v_idx
] = NULL
;
1016 netif_napi_del(&q_vector
->napi
);
1018 /* igb_get_stats64() might access the rings on this vector,
1019 * we must wait a grace period before freeing it.
1021 kfree_rcu(q_vector
, rcu
);
1025 * igb_free_q_vectors - Free memory allocated for interrupt vectors
1026 * @adapter: board private structure to initialize
1028 * This function frees the memory allocated to the q_vectors. In addition if
1029 * NAPI is enabled it will delete any references to the NAPI struct prior
1030 * to freeing the q_vector.
1032 static void igb_free_q_vectors(struct igb_adapter
*adapter
)
1034 int v_idx
= adapter
->num_q_vectors
;
1036 adapter
->num_tx_queues
= 0;
1037 adapter
->num_rx_queues
= 0;
1038 adapter
->num_q_vectors
= 0;
1041 igb_free_q_vector(adapter
, v_idx
);
1045 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1046 * @adapter: board private structure to initialize
1048 * This function resets the device so that it has 0 Rx queues, Tx queues, and
1049 * MSI-X interrupts allocated.
1051 static void igb_clear_interrupt_scheme(struct igb_adapter
*adapter
)
1053 igb_free_q_vectors(adapter
);
1054 igb_reset_interrupt_capability(adapter
);
1058 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1059 * @adapter: board private structure to initialize
1060 * @msix: boolean value of MSIX capability
1062 * Attempt to configure interrupts using the best available
1063 * capabilities of the hardware and kernel.
1065 static void igb_set_interrupt_capability(struct igb_adapter
*adapter
, bool msix
)
1073 /* Number of supported queues. */
1074 adapter
->num_rx_queues
= adapter
->rss_queues
;
1075 if (adapter
->vfs_allocated_count
)
1076 adapter
->num_tx_queues
= 1;
1078 adapter
->num_tx_queues
= adapter
->rss_queues
;
1080 /* start with one vector for every Rx queue */
1081 numvecs
= adapter
->num_rx_queues
;
1083 /* if Tx handler is separate add 1 for every Tx queue */
1084 if (!(adapter
->flags
& IGB_FLAG_QUEUE_PAIRS
))
1085 numvecs
+= adapter
->num_tx_queues
;
1087 /* store the number of vectors reserved for queues */
1088 adapter
->num_q_vectors
= numvecs
;
1090 /* add 1 vector for link status interrupts */
1092 adapter
->msix_entries
= kcalloc(numvecs
, sizeof(struct msix_entry
),
1095 if (!adapter
->msix_entries
)
1098 for (i
= 0; i
< numvecs
; i
++)
1099 adapter
->msix_entries
[i
].entry
= i
;
1101 err
= pci_enable_msix(adapter
->pdev
,
1102 adapter
->msix_entries
,
1107 igb_reset_interrupt_capability(adapter
);
1109 /* If we can't do MSI-X, try MSI */
1111 #ifdef CONFIG_PCI_IOV
1112 /* disable SR-IOV for non MSI-X configurations */
1113 if (adapter
->vf_data
) {
1114 struct e1000_hw
*hw
= &adapter
->hw
;
1115 /* disable iov and allow time for transactions to clear */
1116 pci_disable_sriov(adapter
->pdev
);
1119 kfree(adapter
->vf_data
);
1120 adapter
->vf_data
= NULL
;
1121 wr32(E1000_IOVCTL
, E1000_IOVCTL_REUSE_VFQ
);
1124 dev_info(&adapter
->pdev
->dev
, "IOV Disabled\n");
1127 adapter
->vfs_allocated_count
= 0;
1128 adapter
->rss_queues
= 1;
1129 adapter
->flags
|= IGB_FLAG_QUEUE_PAIRS
;
1130 adapter
->num_rx_queues
= 1;
1131 adapter
->num_tx_queues
= 1;
1132 adapter
->num_q_vectors
= 1;
1133 if (!pci_enable_msi(adapter
->pdev
))
1134 adapter
->flags
|= IGB_FLAG_HAS_MSI
;
1137 static void igb_add_ring(struct igb_ring
*ring
,
1138 struct igb_ring_container
*head
)
1145 * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1146 * @adapter: board private structure to initialize
1147 * @v_count: q_vectors allocated on adapter, used for ring interleaving
1148 * @v_idx: index of vector in adapter struct
1149 * @txr_count: total number of Tx rings to allocate
1150 * @txr_idx: index of first Tx ring to allocate
1151 * @rxr_count: total number of Rx rings to allocate
1152 * @rxr_idx: index of first Rx ring to allocate
1154 * We allocate one q_vector. If allocation fails we return -ENOMEM.
1156 static int igb_alloc_q_vector(struct igb_adapter
*adapter
,
1157 int v_count
, int v_idx
,
1158 int txr_count
, int txr_idx
,
1159 int rxr_count
, int rxr_idx
)
1161 struct igb_q_vector
*q_vector
;
1162 struct igb_ring
*ring
;
1163 int ring_count
, size
;
1165 /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1166 if (txr_count
> 1 || rxr_count
> 1)
1169 ring_count
= txr_count
+ rxr_count
;
1170 size
= sizeof(struct igb_q_vector
) +
1171 (sizeof(struct igb_ring
) * ring_count
);
1173 /* allocate q_vector and rings */
1174 q_vector
= kzalloc(size
, GFP_KERNEL
);
1178 /* initialize NAPI */
1179 netif_napi_add(adapter
->netdev
, &q_vector
->napi
,
1182 /* tie q_vector and adapter together */
1183 adapter
->q_vector
[v_idx
] = q_vector
;
1184 q_vector
->adapter
= adapter
;
1186 /* initialize work limits */
1187 q_vector
->tx
.work_limit
= adapter
->tx_work_limit
;
1189 /* initialize ITR configuration */
1190 q_vector
->itr_register
= adapter
->hw
.hw_addr
+ E1000_EITR(0);
1191 q_vector
->itr_val
= IGB_START_ITR
;
1193 /* initialize pointer to rings */
1194 ring
= q_vector
->ring
;
1198 /* rx or rx/tx vector */
1199 if (!adapter
->rx_itr_setting
|| adapter
->rx_itr_setting
> 3)
1200 q_vector
->itr_val
= adapter
->rx_itr_setting
;
1202 /* tx only vector */
1203 if (!adapter
->tx_itr_setting
|| adapter
->tx_itr_setting
> 3)
1204 q_vector
->itr_val
= adapter
->tx_itr_setting
;
1208 /* assign generic ring traits */
1209 ring
->dev
= &adapter
->pdev
->dev
;
1210 ring
->netdev
= adapter
->netdev
;
1212 /* configure backlink on ring */
1213 ring
->q_vector
= q_vector
;
1215 /* update q_vector Tx values */
1216 igb_add_ring(ring
, &q_vector
->tx
);
1218 /* For 82575, context index must be unique per ring. */
1219 if (adapter
->hw
.mac
.type
== e1000_82575
)
1220 set_bit(IGB_RING_FLAG_TX_CTX_IDX
, &ring
->flags
);
1222 /* apply Tx specific ring traits */
1223 ring
->count
= adapter
->tx_ring_count
;
1224 ring
->queue_index
= txr_idx
;
1226 /* assign ring to adapter */
1227 adapter
->tx_ring
[txr_idx
] = ring
;
1229 /* push pointer to next ring */
1234 /* assign generic ring traits */
1235 ring
->dev
= &adapter
->pdev
->dev
;
1236 ring
->netdev
= adapter
->netdev
;
1238 /* configure backlink on ring */
1239 ring
->q_vector
= q_vector
;
1241 /* update q_vector Rx values */
1242 igb_add_ring(ring
, &q_vector
->rx
);
1244 /* set flag indicating ring supports SCTP checksum offload */
1245 if (adapter
->hw
.mac
.type
>= e1000_82576
)
1246 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM
, &ring
->flags
);
1249 * On i350, i354, i210, and i211, loopback VLAN packets
1250 * have the tag byte-swapped.
1252 if (adapter
->hw
.mac
.type
>= e1000_i350
)
1253 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP
, &ring
->flags
);
1255 /* apply Rx specific ring traits */
1256 ring
->count
= adapter
->rx_ring_count
;
1257 ring
->queue_index
= rxr_idx
;
1259 /* assign ring to adapter */
1260 adapter
->rx_ring
[rxr_idx
] = ring
;
1268 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1269 * @adapter: board private structure to initialize
1271 * We allocate one q_vector per queue interrupt. If allocation fails we
1274 static int igb_alloc_q_vectors(struct igb_adapter
*adapter
)
1276 int q_vectors
= adapter
->num_q_vectors
;
1277 int rxr_remaining
= adapter
->num_rx_queues
;
1278 int txr_remaining
= adapter
->num_tx_queues
;
1279 int rxr_idx
= 0, txr_idx
= 0, v_idx
= 0;
1282 if (q_vectors
>= (rxr_remaining
+ txr_remaining
)) {
1283 for (; rxr_remaining
; v_idx
++) {
1284 err
= igb_alloc_q_vector(adapter
, q_vectors
, v_idx
,
1290 /* update counts and index */
1296 for (; v_idx
< q_vectors
; v_idx
++) {
1297 int rqpv
= DIV_ROUND_UP(rxr_remaining
, q_vectors
- v_idx
);
1298 int tqpv
= DIV_ROUND_UP(txr_remaining
, q_vectors
- v_idx
);
1299 err
= igb_alloc_q_vector(adapter
, q_vectors
, v_idx
,
1300 tqpv
, txr_idx
, rqpv
, rxr_idx
);
1305 /* update counts and index */
1306 rxr_remaining
-= rqpv
;
1307 txr_remaining
-= tqpv
;
1315 adapter
->num_tx_queues
= 0;
1316 adapter
->num_rx_queues
= 0;
1317 adapter
->num_q_vectors
= 0;
1320 igb_free_q_vector(adapter
, v_idx
);
1326 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1327 * @adapter: board private structure to initialize
1328 * @msix: boolean value of MSIX capability
1330 * This function initializes the interrupts and allocates all of the queues.
1332 static int igb_init_interrupt_scheme(struct igb_adapter
*adapter
, bool msix
)
1334 struct pci_dev
*pdev
= adapter
->pdev
;
1337 igb_set_interrupt_capability(adapter
, msix
);
1339 err
= igb_alloc_q_vectors(adapter
);
1341 dev_err(&pdev
->dev
, "Unable to allocate memory for vectors\n");
1342 goto err_alloc_q_vectors
;
1345 igb_cache_ring_register(adapter
);
1349 err_alloc_q_vectors
:
1350 igb_reset_interrupt_capability(adapter
);
1355 * igb_request_irq - initialize interrupts
1356 * @adapter: board private structure to initialize
1358 * Attempts to configure interrupts using the best available
1359 * capabilities of the hardware and kernel.
1361 static int igb_request_irq(struct igb_adapter
*adapter
)
1363 struct net_device
*netdev
= adapter
->netdev
;
1364 struct pci_dev
*pdev
= adapter
->pdev
;
1367 if (adapter
->msix_entries
) {
1368 err
= igb_request_msix(adapter
);
1371 /* fall back to MSI */
1372 igb_free_all_tx_resources(adapter
);
1373 igb_free_all_rx_resources(adapter
);
1375 igb_clear_interrupt_scheme(adapter
);
1376 err
= igb_init_interrupt_scheme(adapter
, false);
1380 igb_setup_all_tx_resources(adapter
);
1381 igb_setup_all_rx_resources(adapter
);
1382 igb_configure(adapter
);
1385 igb_assign_vector(adapter
->q_vector
[0], 0);
1387 if (adapter
->flags
& IGB_FLAG_HAS_MSI
) {
1388 err
= request_irq(pdev
->irq
, igb_intr_msi
, 0,
1389 netdev
->name
, adapter
);
1393 /* fall back to legacy interrupts */
1394 igb_reset_interrupt_capability(adapter
);
1395 adapter
->flags
&= ~IGB_FLAG_HAS_MSI
;
1398 err
= request_irq(pdev
->irq
, igb_intr
, IRQF_SHARED
,
1399 netdev
->name
, adapter
);
1402 dev_err(&pdev
->dev
, "Error %d getting interrupt\n",
1409 static void igb_free_irq(struct igb_adapter
*adapter
)
1411 if (adapter
->msix_entries
) {
1414 free_irq(adapter
->msix_entries
[vector
++].vector
, adapter
);
1416 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1417 free_irq(adapter
->msix_entries
[vector
++].vector
,
1418 adapter
->q_vector
[i
]);
1420 free_irq(adapter
->pdev
->irq
, adapter
);
1425 * igb_irq_disable - Mask off interrupt generation on the NIC
1426 * @adapter: board private structure
1428 static void igb_irq_disable(struct igb_adapter
*adapter
)
1430 struct e1000_hw
*hw
= &adapter
->hw
;
1432 /* we need to be careful when disabling interrupts. The VFs are also
1433 * mapped into these registers and so clearing the bits can cause
1434 * issues on the VF drivers so we only need to clear what we set
1436 if (adapter
->msix_entries
) {
1437 u32 regval
= rd32(E1000_EIAM
);
1438 wr32(E1000_EIAM
, regval
& ~adapter
->eims_enable_mask
);
1439 wr32(E1000_EIMC
, adapter
->eims_enable_mask
);
1440 regval
= rd32(E1000_EIAC
);
1441 wr32(E1000_EIAC
, regval
& ~adapter
->eims_enable_mask
);
1445 wr32(E1000_IMC
, ~0);
1447 if (adapter
->msix_entries
) {
1449 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1450 synchronize_irq(adapter
->msix_entries
[i
].vector
);
1452 synchronize_irq(adapter
->pdev
->irq
);
1457 * igb_irq_enable - Enable default interrupt generation settings
1458 * @adapter: board private structure
1460 static void igb_irq_enable(struct igb_adapter
*adapter
)
1462 struct e1000_hw
*hw
= &adapter
->hw
;
1464 if (adapter
->msix_entries
) {
1465 u32 ims
= E1000_IMS_LSC
| E1000_IMS_DOUTSYNC
| E1000_IMS_DRSTA
;
1466 u32 regval
= rd32(E1000_EIAC
);
1467 wr32(E1000_EIAC
, regval
| adapter
->eims_enable_mask
);
1468 regval
= rd32(E1000_EIAM
);
1469 wr32(E1000_EIAM
, regval
| adapter
->eims_enable_mask
);
1470 wr32(E1000_EIMS
, adapter
->eims_enable_mask
);
1471 if (adapter
->vfs_allocated_count
) {
1472 wr32(E1000_MBVFIMR
, 0xFF);
1473 ims
|= E1000_IMS_VMMB
;
1475 wr32(E1000_IMS
, ims
);
1477 wr32(E1000_IMS
, IMS_ENABLE_MASK
|
1479 wr32(E1000_IAM
, IMS_ENABLE_MASK
|
1484 static void igb_update_mng_vlan(struct igb_adapter
*adapter
)
1486 struct e1000_hw
*hw
= &adapter
->hw
;
1487 u16 vid
= adapter
->hw
.mng_cookie
.vlan_id
;
1488 u16 old_vid
= adapter
->mng_vlan_id
;
1490 if (hw
->mng_cookie
.status
& E1000_MNG_DHCP_COOKIE_STATUS_VLAN
) {
1491 /* add VID to filter table */
1492 igb_vfta_set(hw
, vid
, true);
1493 adapter
->mng_vlan_id
= vid
;
1495 adapter
->mng_vlan_id
= IGB_MNG_VLAN_NONE
;
1498 if ((old_vid
!= (u16
)IGB_MNG_VLAN_NONE
) &&
1500 !test_bit(old_vid
, adapter
->active_vlans
)) {
1501 /* remove VID from filter table */
1502 igb_vfta_set(hw
, old_vid
, false);
1507 * igb_release_hw_control - release control of the h/w to f/w
1508 * @adapter: address of board private structure
1510 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1511 * For ASF and Pass Through versions of f/w this means that the
1512 * driver is no longer loaded.
1514 static void igb_release_hw_control(struct igb_adapter
*adapter
)
1516 struct e1000_hw
*hw
= &adapter
->hw
;
1519 /* Let firmware take over control of h/w */
1520 ctrl_ext
= rd32(E1000_CTRL_EXT
);
1521 wr32(E1000_CTRL_EXT
,
1522 ctrl_ext
& ~E1000_CTRL_EXT_DRV_LOAD
);
1526 * igb_get_hw_control - get control of the h/w from f/w
1527 * @adapter: address of board private structure
1529 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1530 * For ASF and Pass Through versions of f/w this means that
1531 * the driver is loaded.
1533 static void igb_get_hw_control(struct igb_adapter
*adapter
)
1535 struct e1000_hw
*hw
= &adapter
->hw
;
1538 /* Let firmware know the driver has taken over */
1539 ctrl_ext
= rd32(E1000_CTRL_EXT
);
1540 wr32(E1000_CTRL_EXT
,
1541 ctrl_ext
| E1000_CTRL_EXT_DRV_LOAD
);
1545 * igb_configure - configure the hardware for RX and TX
1546 * @adapter: private board structure
1548 static void igb_configure(struct igb_adapter
*adapter
)
1550 struct net_device
*netdev
= adapter
->netdev
;
1553 igb_get_hw_control(adapter
);
1554 igb_set_rx_mode(netdev
);
1556 igb_restore_vlan(adapter
);
1558 igb_setup_tctl(adapter
);
1559 igb_setup_mrqc(adapter
);
1560 igb_setup_rctl(adapter
);
1562 igb_configure_tx(adapter
);
1563 igb_configure_rx(adapter
);
1565 igb_rx_fifo_flush_82575(&adapter
->hw
);
1567 /* call igb_desc_unused which always leaves
1568 * at least 1 descriptor unused to make sure
1569 * next_to_use != next_to_clean
1571 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
1572 struct igb_ring
*ring
= adapter
->rx_ring
[i
];
1573 igb_alloc_rx_buffers(ring
, igb_desc_unused(ring
));
1578 * igb_power_up_link - Power up the phy/serdes link
1579 * @adapter: address of board private structure
1581 void igb_power_up_link(struct igb_adapter
*adapter
)
1583 igb_reset_phy(&adapter
->hw
);
1585 if (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
)
1586 igb_power_up_phy_copper(&adapter
->hw
);
1588 igb_power_up_serdes_link_82575(&adapter
->hw
);
1592 * igb_power_down_link - Power down the phy/serdes link
1593 * @adapter: address of board private structure
1595 static void igb_power_down_link(struct igb_adapter
*adapter
)
1597 if (adapter
->hw
.phy
.media_type
== e1000_media_type_copper
)
1598 igb_power_down_phy_copper_82575(&adapter
->hw
);
1600 igb_shutdown_serdes_link_82575(&adapter
->hw
);
1604 * igb_up - Open the interface and prepare it to handle traffic
1605 * @adapter: board private structure
1607 int igb_up(struct igb_adapter
*adapter
)
1609 struct e1000_hw
*hw
= &adapter
->hw
;
1612 /* hardware has been reset, we need to reload some things */
1613 igb_configure(adapter
);
1615 clear_bit(__IGB_DOWN
, &adapter
->state
);
1617 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
1618 napi_enable(&(adapter
->q_vector
[i
]->napi
));
1620 if (adapter
->msix_entries
)
1621 igb_configure_msix(adapter
);
1623 igb_assign_vector(adapter
->q_vector
[0], 0);
1625 /* Clear any pending interrupts. */
1627 igb_irq_enable(adapter
);
1629 /* notify VFs that reset has been completed */
1630 if (adapter
->vfs_allocated_count
) {
1631 u32 reg_data
= rd32(E1000_CTRL_EXT
);
1632 reg_data
|= E1000_CTRL_EXT_PFRSTD
;
1633 wr32(E1000_CTRL_EXT
, reg_data
);
1636 netif_tx_start_all_queues(adapter
->netdev
);
1638 /* start the watchdog. */
1639 hw
->mac
.get_link_status
= 1;
1640 schedule_work(&adapter
->watchdog_task
);
1645 void igb_down(struct igb_adapter
*adapter
)
1647 struct net_device
*netdev
= adapter
->netdev
;
1648 struct e1000_hw
*hw
= &adapter
->hw
;
1652 /* signal that we're down so the interrupt handler does not
1653 * reschedule our watchdog timer
1655 set_bit(__IGB_DOWN
, &adapter
->state
);
1657 /* disable receives in the hardware */
1658 rctl
= rd32(E1000_RCTL
);
1659 wr32(E1000_RCTL
, rctl
& ~E1000_RCTL_EN
);
1660 /* flush and sleep below */
1662 netif_tx_stop_all_queues(netdev
);
1664 /* disable transmits in the hardware */
1665 tctl
= rd32(E1000_TCTL
);
1666 tctl
&= ~E1000_TCTL_EN
;
1667 wr32(E1000_TCTL
, tctl
);
1668 /* flush both disables and wait for them to finish */
1672 igb_irq_disable(adapter
);
1674 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
1675 napi_synchronize(&(adapter
->q_vector
[i
]->napi
));
1676 napi_disable(&(adapter
->q_vector
[i
]->napi
));
1680 del_timer_sync(&adapter
->watchdog_timer
);
1681 del_timer_sync(&adapter
->phy_info_timer
);
1683 netif_carrier_off(netdev
);
1685 /* record the stats before reset*/
1686 spin_lock(&adapter
->stats64_lock
);
1687 igb_update_stats(adapter
, &adapter
->stats64
);
1688 spin_unlock(&adapter
->stats64_lock
);
1690 adapter
->link_speed
= 0;
1691 adapter
->link_duplex
= 0;
1693 if (!pci_channel_offline(adapter
->pdev
))
1695 igb_clean_all_tx_rings(adapter
);
1696 igb_clean_all_rx_rings(adapter
);
1697 #ifdef CONFIG_IGB_DCA
1699 /* since we reset the hardware DCA settings were cleared */
1700 igb_setup_dca(adapter
);
1704 void igb_reinit_locked(struct igb_adapter
*adapter
)
1706 WARN_ON(in_interrupt());
1707 while (test_and_set_bit(__IGB_RESETTING
, &adapter
->state
))
1711 clear_bit(__IGB_RESETTING
, &adapter
->state
);
1714 void igb_reset(struct igb_adapter
*adapter
)
1716 struct pci_dev
*pdev
= adapter
->pdev
;
1717 struct e1000_hw
*hw
= &adapter
->hw
;
1718 struct e1000_mac_info
*mac
= &hw
->mac
;
1719 struct e1000_fc_info
*fc
= &hw
->fc
;
1720 u32 pba
= 0, tx_space
, min_tx_space
, min_rx_space
, hwm
;
1722 /* Repartition Pba for greater than 9k mtu
1723 * To take effect CTRL.RST is required.
1725 switch (mac
->type
) {
1729 pba
= rd32(E1000_RXPBS
);
1730 pba
= igb_rxpbs_adjust_82580(pba
);
1733 pba
= rd32(E1000_RXPBS
);
1734 pba
&= E1000_RXPBS_SIZE_MASK_82576
;
1740 pba
= E1000_PBA_34K
;
1744 if ((adapter
->max_frame_size
> ETH_FRAME_LEN
+ ETH_FCS_LEN
) &&
1745 (mac
->type
< e1000_82576
)) {
1746 /* adjust PBA for jumbo frames */
1747 wr32(E1000_PBA
, pba
);
1749 /* To maintain wire speed transmits, the Tx FIFO should be
1750 * large enough to accommodate two full transmit packets,
1751 * rounded up to the next 1KB and expressed in KB. Likewise,
1752 * the Rx FIFO should be large enough to accommodate at least
1753 * one full receive packet and is similarly rounded up and
1756 pba
= rd32(E1000_PBA
);
1757 /* upper 16 bits has Tx packet buffer allocation size in KB */
1758 tx_space
= pba
>> 16;
1759 /* lower 16 bits has Rx packet buffer allocation size in KB */
1761 /* the Tx fifo also stores 16 bytes of information about the Tx
1762 * but don't include ethernet FCS because hardware appends it
1764 min_tx_space
= (adapter
->max_frame_size
+
1765 sizeof(union e1000_adv_tx_desc
) -
1767 min_tx_space
= ALIGN(min_tx_space
, 1024);
1768 min_tx_space
>>= 10;
1769 /* software strips receive CRC, so leave room for it */
1770 min_rx_space
= adapter
->max_frame_size
;
1771 min_rx_space
= ALIGN(min_rx_space
, 1024);
1772 min_rx_space
>>= 10;
1774 /* If current Tx allocation is less than the min Tx FIFO size,
1775 * and the min Tx FIFO size is less than the current Rx FIFO
1776 * allocation, take space away from current Rx allocation
1778 if (tx_space
< min_tx_space
&&
1779 ((min_tx_space
- tx_space
) < pba
)) {
1780 pba
= pba
- (min_tx_space
- tx_space
);
1782 /* if short on Rx space, Rx wins and must trump Tx
1785 if (pba
< min_rx_space
)
1788 wr32(E1000_PBA
, pba
);
1791 /* flow control settings */
1792 /* The high water mark must be low enough to fit one full frame
1793 * (or the size used for early receive) above it in the Rx FIFO.
1794 * Set it to the lower of:
1795 * - 90% of the Rx FIFO size, or
1796 * - the full Rx FIFO size minus one full frame
1798 hwm
= min(((pba
<< 10) * 9 / 10),
1799 ((pba
<< 10) - 2 * adapter
->max_frame_size
));
1801 fc
->high_water
= hwm
& 0xFFFFFFF0; /* 16-byte granularity */
1802 fc
->low_water
= fc
->high_water
- 16;
1803 fc
->pause_time
= 0xFFFF;
1805 fc
->current_mode
= fc
->requested_mode
;
1807 /* disable receive for all VFs and wait one second */
1808 if (adapter
->vfs_allocated_count
) {
1810 for (i
= 0 ; i
< adapter
->vfs_allocated_count
; i
++)
1811 adapter
->vf_data
[i
].flags
&= IGB_VF_FLAG_PF_SET_MAC
;
1813 /* ping all the active vfs to let them know we are going down */
1814 igb_ping_all_vfs(adapter
);
1816 /* disable transmits and receives */
1817 wr32(E1000_VFRE
, 0);
1818 wr32(E1000_VFTE
, 0);
1821 /* Allow time for pending master requests to run */
1822 hw
->mac
.ops
.reset_hw(hw
);
1825 if (hw
->mac
.ops
.init_hw(hw
))
1826 dev_err(&pdev
->dev
, "Hardware Error\n");
1828 /* Flow control settings reset on hardware reset, so guarantee flow
1829 * control is off when forcing speed.
1831 if (!hw
->mac
.autoneg
)
1832 igb_force_mac_fc(hw
);
1834 igb_init_dmac(adapter
, pba
);
1835 #ifdef CONFIG_IGB_HWMON
1836 /* Re-initialize the thermal sensor on i350 devices. */
1837 if (!test_bit(__IGB_DOWN
, &adapter
->state
)) {
1838 if (mac
->type
== e1000_i350
&& hw
->bus
.func
== 0) {
1839 /* If present, re-initialize the external thermal sensor
1843 mac
->ops
.init_thermal_sensor_thresh(hw
);
1847 if (!netif_running(adapter
->netdev
))
1848 igb_power_down_link(adapter
);
1850 igb_update_mng_vlan(adapter
);
1852 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1853 wr32(E1000_VET
, ETHERNET_IEEE_VLAN_TYPE
);
1855 /* Re-enable PTP, where applicable. */
1856 igb_ptp_reset(adapter
);
1858 igb_get_phy_info(hw
);
1861 static netdev_features_t
igb_fix_features(struct net_device
*netdev
,
1862 netdev_features_t features
)
1864 /* Since there is no support for separate Rx/Tx vlan accel
1865 * enable/disable make sure Tx flag is always in same state as Rx.
1867 if (features
& NETIF_F_HW_VLAN_CTAG_RX
)
1868 features
|= NETIF_F_HW_VLAN_CTAG_TX
;
1870 features
&= ~NETIF_F_HW_VLAN_CTAG_TX
;
1875 static int igb_set_features(struct net_device
*netdev
,
1876 netdev_features_t features
)
1878 netdev_features_t changed
= netdev
->features
^ features
;
1879 struct igb_adapter
*adapter
= netdev_priv(netdev
);
1881 if (changed
& NETIF_F_HW_VLAN_CTAG_RX
)
1882 igb_vlan_mode(netdev
, features
);
1884 if (!(changed
& NETIF_F_RXALL
))
1887 netdev
->features
= features
;
1889 if (netif_running(netdev
))
1890 igb_reinit_locked(adapter
);
1897 static const struct net_device_ops igb_netdev_ops
= {
1898 .ndo_open
= igb_open
,
1899 .ndo_stop
= igb_close
,
1900 .ndo_start_xmit
= igb_xmit_frame
,
1901 .ndo_get_stats64
= igb_get_stats64
,
1902 .ndo_set_rx_mode
= igb_set_rx_mode
,
1903 .ndo_set_mac_address
= igb_set_mac
,
1904 .ndo_change_mtu
= igb_change_mtu
,
1905 .ndo_do_ioctl
= igb_ioctl
,
1906 .ndo_tx_timeout
= igb_tx_timeout
,
1907 .ndo_validate_addr
= eth_validate_addr
,
1908 .ndo_vlan_rx_add_vid
= igb_vlan_rx_add_vid
,
1909 .ndo_vlan_rx_kill_vid
= igb_vlan_rx_kill_vid
,
1910 .ndo_set_vf_mac
= igb_ndo_set_vf_mac
,
1911 .ndo_set_vf_vlan
= igb_ndo_set_vf_vlan
,
1912 .ndo_set_vf_tx_rate
= igb_ndo_set_vf_bw
,
1913 .ndo_set_vf_spoofchk
= igb_ndo_set_vf_spoofchk
,
1914 .ndo_get_vf_config
= igb_ndo_get_vf_config
,
1915 #ifdef CONFIG_NET_POLL_CONTROLLER
1916 .ndo_poll_controller
= igb_netpoll
,
1918 .ndo_fix_features
= igb_fix_features
,
1919 .ndo_set_features
= igb_set_features
,
1923 * igb_set_fw_version - Configure version string for ethtool
1924 * @adapter: adapter struct
1926 void igb_set_fw_version(struct igb_adapter
*adapter
)
1928 struct e1000_hw
*hw
= &adapter
->hw
;
1929 struct e1000_fw_version fw
;
1931 igb_get_fw_version(hw
, &fw
);
1933 switch (hw
->mac
.type
) {
1935 snprintf(adapter
->fw_version
, sizeof(adapter
->fw_version
),
1937 fw
.invm_major
, fw
.invm_minor
, fw
.invm_img_type
);
1941 /* if option is rom valid, display its version too */
1943 snprintf(adapter
->fw_version
,
1944 sizeof(adapter
->fw_version
),
1945 "%d.%d, 0x%08x, %d.%d.%d",
1946 fw
.eep_major
, fw
.eep_minor
, fw
.etrack_id
,
1947 fw
.or_major
, fw
.or_build
, fw
.or_patch
);
1950 snprintf(adapter
->fw_version
,
1951 sizeof(adapter
->fw_version
),
1953 fw
.eep_major
, fw
.eep_minor
, fw
.etrack_id
);
1961 * igb_init_i2c - Init I2C interface
1962 * @adapter: pointer to adapter structure
1964 static s32
igb_init_i2c(struct igb_adapter
*adapter
)
1966 s32 status
= E1000_SUCCESS
;
1968 /* I2C interface supported on i350 devices */
1969 if (adapter
->hw
.mac
.type
!= e1000_i350
)
1970 return E1000_SUCCESS
;
1972 /* Initialize the i2c bus which is controlled by the registers.
1973 * This bus will use the i2c_algo_bit structue that implements
1974 * the protocol through toggling of the 4 bits in the register.
1976 adapter
->i2c_adap
.owner
= THIS_MODULE
;
1977 adapter
->i2c_algo
= igb_i2c_algo
;
1978 adapter
->i2c_algo
.data
= adapter
;
1979 adapter
->i2c_adap
.algo_data
= &adapter
->i2c_algo
;
1980 adapter
->i2c_adap
.dev
.parent
= &adapter
->pdev
->dev
;
1981 strlcpy(adapter
->i2c_adap
.name
, "igb BB",
1982 sizeof(adapter
->i2c_adap
.name
));
1983 status
= i2c_bit_add_bus(&adapter
->i2c_adap
);
1988 * igb_probe - Device Initialization Routine
1989 * @pdev: PCI device information struct
1990 * @ent: entry in igb_pci_tbl
1992 * Returns 0 on success, negative on failure
1994 * igb_probe initializes an adapter identified by a pci_dev structure.
1995 * The OS initialization, configuring of the adapter private structure,
1996 * and a hardware reset occur.
1998 static int igb_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
2000 struct net_device
*netdev
;
2001 struct igb_adapter
*adapter
;
2002 struct e1000_hw
*hw
;
2003 u16 eeprom_data
= 0;
2005 static int global_quad_port_a
; /* global quad port a indication */
2006 const struct e1000_info
*ei
= igb_info_tbl
[ent
->driver_data
];
2007 unsigned long mmio_start
, mmio_len
;
2008 int err
, pci_using_dac
;
2009 u8 part_str
[E1000_PBANUM_LENGTH
];
2011 /* Catch broken hardware that put the wrong VF device ID in
2012 * the PCIe SR-IOV capability.
2014 if (pdev
->is_virtfn
) {
2015 WARN(1, KERN_ERR
"%s (%hx:%hx) should not be a VF!\n",
2016 pci_name(pdev
), pdev
->vendor
, pdev
->device
);
2020 err
= pci_enable_device_mem(pdev
);
2025 err
= dma_set_mask(&pdev
->dev
, DMA_BIT_MASK(64));
2027 err
= dma_set_coherent_mask(&pdev
->dev
, DMA_BIT_MASK(64));
2031 err
= dma_set_mask(&pdev
->dev
, DMA_BIT_MASK(32));
2033 err
= dma_set_coherent_mask(&pdev
->dev
,
2037 "No usable DMA configuration, aborting\n");
2043 err
= pci_request_selected_regions(pdev
, pci_select_bars(pdev
,
2049 pci_enable_pcie_error_reporting(pdev
);
2051 pci_set_master(pdev
);
2052 pci_save_state(pdev
);
2055 netdev
= alloc_etherdev_mq(sizeof(struct igb_adapter
),
2058 goto err_alloc_etherdev
;
2060 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
2062 pci_set_drvdata(pdev
, netdev
);
2063 adapter
= netdev_priv(netdev
);
2064 adapter
->netdev
= netdev
;
2065 adapter
->pdev
= pdev
;
2068 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
2070 mmio_start
= pci_resource_start(pdev
, 0);
2071 mmio_len
= pci_resource_len(pdev
, 0);
2074 hw
->hw_addr
= ioremap(mmio_start
, mmio_len
);
2078 netdev
->netdev_ops
= &igb_netdev_ops
;
2079 igb_set_ethtool_ops(netdev
);
2080 netdev
->watchdog_timeo
= 5 * HZ
;
2082 strncpy(netdev
->name
, pci_name(pdev
), sizeof(netdev
->name
) - 1);
2084 netdev
->mem_start
= mmio_start
;
2085 netdev
->mem_end
= mmio_start
+ mmio_len
;
2087 /* PCI config space info */
2088 hw
->vendor_id
= pdev
->vendor
;
2089 hw
->device_id
= pdev
->device
;
2090 hw
->revision_id
= pdev
->revision
;
2091 hw
->subsystem_vendor_id
= pdev
->subsystem_vendor
;
2092 hw
->subsystem_device_id
= pdev
->subsystem_device
;
2094 /* Copy the default MAC, PHY and NVM function pointers */
2095 memcpy(&hw
->mac
.ops
, ei
->mac_ops
, sizeof(hw
->mac
.ops
));
2096 memcpy(&hw
->phy
.ops
, ei
->phy_ops
, sizeof(hw
->phy
.ops
));
2097 memcpy(&hw
->nvm
.ops
, ei
->nvm_ops
, sizeof(hw
->nvm
.ops
));
2098 /* Initialize skew-specific constants */
2099 err
= ei
->get_invariants(hw
);
2103 /* setup the private structure */
2104 err
= igb_sw_init(adapter
);
2108 igb_get_bus_info_pcie(hw
);
2110 hw
->phy
.autoneg_wait_to_complete
= false;
2112 /* Copper options */
2113 if (hw
->phy
.media_type
== e1000_media_type_copper
) {
2114 hw
->phy
.mdix
= AUTO_ALL_MODES
;
2115 hw
->phy
.disable_polarity_correction
= false;
2116 hw
->phy
.ms_type
= e1000_ms_hw_default
;
2119 if (igb_check_reset_block(hw
))
2120 dev_info(&pdev
->dev
,
2121 "PHY reset is blocked due to SOL/IDER session.\n");
2123 /* features is initialized to 0 in allocation, it might have bits
2124 * set by igb_sw_init so we should use an or instead of an
2127 netdev
->features
|= NETIF_F_SG
|
2134 NETIF_F_HW_VLAN_CTAG_RX
|
2135 NETIF_F_HW_VLAN_CTAG_TX
;
2137 /* copy netdev features into list of user selectable features */
2138 netdev
->hw_features
|= netdev
->features
;
2139 netdev
->hw_features
|= NETIF_F_RXALL
;
2141 /* set this bit last since it cannot be part of hw_features */
2142 netdev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
2144 netdev
->vlan_features
|= NETIF_F_TSO
|
2150 netdev
->priv_flags
|= IFF_SUPP_NOFCS
;
2152 if (pci_using_dac
) {
2153 netdev
->features
|= NETIF_F_HIGHDMA
;
2154 netdev
->vlan_features
|= NETIF_F_HIGHDMA
;
2157 if (hw
->mac
.type
>= e1000_82576
) {
2158 netdev
->hw_features
|= NETIF_F_SCTP_CSUM
;
2159 netdev
->features
|= NETIF_F_SCTP_CSUM
;
2162 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
2164 adapter
->en_mng_pt
= igb_enable_mng_pass_thru(hw
);
2166 /* before reading the NVM, reset the controller to put the device in a
2167 * known good starting state
2169 hw
->mac
.ops
.reset_hw(hw
);
2171 /* make sure the NVM is good , i211/i210 parts can have special NVM
2172 * that doesn't contain a checksum
2174 switch (hw
->mac
.type
) {
2177 if (igb_get_flash_presence_i210(hw
)) {
2178 if (hw
->nvm
.ops
.validate(hw
) < 0) {
2180 "The NVM Checksum Is Not Valid\n");
2187 if (hw
->nvm
.ops
.validate(hw
) < 0) {
2188 dev_err(&pdev
->dev
, "The NVM Checksum Is Not Valid\n");
2195 /* copy the MAC address out of the NVM */
2196 if (hw
->mac
.ops
.read_mac_addr(hw
))
2197 dev_err(&pdev
->dev
, "NVM Read Error\n");
2199 memcpy(netdev
->dev_addr
, hw
->mac
.addr
, netdev
->addr_len
);
2201 if (!is_valid_ether_addr(netdev
->dev_addr
)) {
2202 dev_err(&pdev
->dev
, "Invalid MAC Address\n");
2207 /* get firmware version for ethtool -i */
2208 igb_set_fw_version(adapter
);
2210 setup_timer(&adapter
->watchdog_timer
, igb_watchdog
,
2211 (unsigned long) adapter
);
2212 setup_timer(&adapter
->phy_info_timer
, igb_update_phy_info
,
2213 (unsigned long) adapter
);
2215 INIT_WORK(&adapter
->reset_task
, igb_reset_task
);
2216 INIT_WORK(&adapter
->watchdog_task
, igb_watchdog_task
);
2218 /* Initialize link properties that are user-changeable */
2219 adapter
->fc_autoneg
= true;
2220 hw
->mac
.autoneg
= true;
2221 hw
->phy
.autoneg_advertised
= 0x2f;
2223 hw
->fc
.requested_mode
= e1000_fc_default
;
2224 hw
->fc
.current_mode
= e1000_fc_default
;
2226 igb_validate_mdi_setting(hw
);
2228 /* By default, support wake on port A */
2229 if (hw
->bus
.func
== 0)
2230 adapter
->flags
|= IGB_FLAG_WOL_SUPPORTED
;
2232 /* Check the NVM for wake support on non-port A ports */
2233 if (hw
->mac
.type
>= e1000_82580
)
2234 hw
->nvm
.ops
.read(hw
, NVM_INIT_CONTROL3_PORT_A
+
2235 NVM_82580_LAN_FUNC_OFFSET(hw
->bus
.func
), 1,
2237 else if (hw
->bus
.func
== 1)
2238 hw
->nvm
.ops
.read(hw
, NVM_INIT_CONTROL3_PORT_B
, 1, &eeprom_data
);
2240 if (eeprom_data
& IGB_EEPROM_APME
)
2241 adapter
->flags
|= IGB_FLAG_WOL_SUPPORTED
;
2243 /* now that we have the eeprom settings, apply the special cases where
2244 * the eeprom may be wrong or the board simply won't support wake on
2245 * lan on a particular port
2247 switch (pdev
->device
) {
2248 case E1000_DEV_ID_82575GB_QUAD_COPPER
:
2249 adapter
->flags
&= ~IGB_FLAG_WOL_SUPPORTED
;
2251 case E1000_DEV_ID_82575EB_FIBER_SERDES
:
2252 case E1000_DEV_ID_82576_FIBER
:
2253 case E1000_DEV_ID_82576_SERDES
:
2254 /* Wake events only supported on port A for dual fiber
2255 * regardless of eeprom setting
2257 if (rd32(E1000_STATUS
) & E1000_STATUS_FUNC_1
)
2258 adapter
->flags
&= ~IGB_FLAG_WOL_SUPPORTED
;
2260 case E1000_DEV_ID_82576_QUAD_COPPER
:
2261 case E1000_DEV_ID_82576_QUAD_COPPER_ET2
:
2262 /* if quad port adapter, disable WoL on all but port A */
2263 if (global_quad_port_a
!= 0)
2264 adapter
->flags
&= ~IGB_FLAG_WOL_SUPPORTED
;
2266 adapter
->flags
|= IGB_FLAG_QUAD_PORT_A
;
2267 /* Reset for multiple quad port adapters */
2268 if (++global_quad_port_a
== 4)
2269 global_quad_port_a
= 0;
2272 /* If the device can't wake, don't set software support */
2273 if (!device_can_wakeup(&adapter
->pdev
->dev
))
2274 adapter
->flags
&= ~IGB_FLAG_WOL_SUPPORTED
;
2277 /* initialize the wol settings based on the eeprom settings */
2278 if (adapter
->flags
& IGB_FLAG_WOL_SUPPORTED
)
2279 adapter
->wol
|= E1000_WUFC_MAG
;
2281 /* Some vendors want WoL disabled by default, but still supported */
2282 if ((hw
->mac
.type
== e1000_i350
) &&
2283 (pdev
->subsystem_vendor
== PCI_VENDOR_ID_HP
)) {
2284 adapter
->flags
|= IGB_FLAG_WOL_SUPPORTED
;
2288 device_set_wakeup_enable(&adapter
->pdev
->dev
,
2289 adapter
->flags
& IGB_FLAG_WOL_SUPPORTED
);
2291 /* reset the hardware with the new settings */
2294 /* Init the I2C interface */
2295 err
= igb_init_i2c(adapter
);
2297 dev_err(&pdev
->dev
, "failed to init i2c interface\n");
2301 /* let the f/w know that the h/w is now under the control of the
2303 igb_get_hw_control(adapter
);
2305 strcpy(netdev
->name
, "eth%d");
2306 err
= register_netdev(netdev
);
2310 /* carrier off reporting is important to ethtool even BEFORE open */
2311 netif_carrier_off(netdev
);
2313 #ifdef CONFIG_IGB_DCA
2314 if (dca_add_requester(&pdev
->dev
) == 0) {
2315 adapter
->flags
|= IGB_FLAG_DCA_ENABLED
;
2316 dev_info(&pdev
->dev
, "DCA enabled\n");
2317 igb_setup_dca(adapter
);
2321 #ifdef CONFIG_IGB_HWMON
2322 /* Initialize the thermal sensor on i350 devices. */
2323 if (hw
->mac
.type
== e1000_i350
&& hw
->bus
.func
== 0) {
2326 /* Read the NVM to determine if this i350 device supports an
2327 * external thermal sensor.
2329 hw
->nvm
.ops
.read(hw
, NVM_ETS_CFG
, 1, &ets_word
);
2330 if (ets_word
!= 0x0000 && ets_word
!= 0xFFFF)
2331 adapter
->ets
= true;
2333 adapter
->ets
= false;
2334 if (igb_sysfs_init(adapter
))
2336 "failed to allocate sysfs resources\n");
2338 adapter
->ets
= false;
2341 /* do hw tstamp init after resetting */
2342 igb_ptp_init(adapter
);
2344 dev_info(&pdev
->dev
, "Intel(R) Gigabit Ethernet Network Connection\n");
2345 /* print bus type/speed/width info, not applicable to i354 */
2346 if (hw
->mac
.type
!= e1000_i354
) {
2347 dev_info(&pdev
->dev
, "%s: (PCIe:%s:%s) %pM\n",
2349 ((hw
->bus
.speed
== e1000_bus_speed_2500
) ? "2.5Gb/s" :
2350 (hw
->bus
.speed
== e1000_bus_speed_5000
) ? "5.0Gb/s" :
2352 ((hw
->bus
.width
== e1000_bus_width_pcie_x4
) ?
2354 (hw
->bus
.width
== e1000_bus_width_pcie_x2
) ?
2356 (hw
->bus
.width
== e1000_bus_width_pcie_x1
) ?
2357 "Width x1" : "unknown"), netdev
->dev_addr
);
2360 ret_val
= igb_read_part_string(hw
, part_str
, E1000_PBANUM_LENGTH
);
2362 strcpy(part_str
, "Unknown");
2363 dev_info(&pdev
->dev
, "%s: PBA No: %s\n", netdev
->name
, part_str
);
2364 dev_info(&pdev
->dev
,
2365 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2366 adapter
->msix_entries
? "MSI-X" :
2367 (adapter
->flags
& IGB_FLAG_HAS_MSI
) ? "MSI" : "legacy",
2368 adapter
->num_rx_queues
, adapter
->num_tx_queues
);
2369 switch (hw
->mac
.type
) {
2373 igb_set_eee_i350(hw
);
2376 if (hw
->phy
.media_type
== e1000_media_type_copper
) {
2377 if ((rd32(E1000_CTRL_EXT
) &
2378 E1000_CTRL_EXT_LINK_MODE_SGMII
))
2379 igb_set_eee_i354(hw
);
2386 pm_runtime_put_noidle(&pdev
->dev
);
2390 igb_release_hw_control(adapter
);
2391 memset(&adapter
->i2c_adap
, 0, sizeof(adapter
->i2c_adap
));
2393 if (!igb_check_reset_block(hw
))
2396 if (hw
->flash_address
)
2397 iounmap(hw
->flash_address
);
2399 igb_clear_interrupt_scheme(adapter
);
2400 iounmap(hw
->hw_addr
);
2402 free_netdev(netdev
);
2404 pci_release_selected_regions(pdev
,
2405 pci_select_bars(pdev
, IORESOURCE_MEM
));
2408 pci_disable_device(pdev
);
2412 #ifdef CONFIG_PCI_IOV
2413 static int igb_disable_sriov(struct pci_dev
*pdev
)
2415 struct net_device
*netdev
= pci_get_drvdata(pdev
);
2416 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2417 struct e1000_hw
*hw
= &adapter
->hw
;
2419 /* reclaim resources allocated to VFs */
2420 if (adapter
->vf_data
) {
2421 /* disable iov and allow time for transactions to clear */
2422 if (pci_vfs_assigned(pdev
)) {
2423 dev_warn(&pdev
->dev
,
2424 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2427 pci_disable_sriov(pdev
);
2431 kfree(adapter
->vf_data
);
2432 adapter
->vf_data
= NULL
;
2433 adapter
->vfs_allocated_count
= 0;
2434 wr32(E1000_IOVCTL
, E1000_IOVCTL_REUSE_VFQ
);
2437 dev_info(&pdev
->dev
, "IOV Disabled\n");
2439 /* Re-enable DMA Coalescing flag since IOV is turned off */
2440 adapter
->flags
|= IGB_FLAG_DMAC
;
2446 static int igb_enable_sriov(struct pci_dev
*pdev
, int num_vfs
)
2448 struct net_device
*netdev
= pci_get_drvdata(pdev
);
2449 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2450 int old_vfs
= pci_num_vf(pdev
);
2454 if (!adapter
->msix_entries
) {
2461 else if (old_vfs
&& old_vfs
== num_vfs
)
2463 else if (old_vfs
&& old_vfs
!= num_vfs
)
2464 err
= igb_disable_sriov(pdev
);
2474 adapter
->vfs_allocated_count
= num_vfs
;
2476 adapter
->vf_data
= kcalloc(adapter
->vfs_allocated_count
,
2477 sizeof(struct vf_data_storage
), GFP_KERNEL
);
2479 /* if allocation failed then we do not support SR-IOV */
2480 if (!adapter
->vf_data
) {
2481 adapter
->vfs_allocated_count
= 0;
2483 "Unable to allocate memory for VF Data Storage\n");
2488 err
= pci_enable_sriov(pdev
, adapter
->vfs_allocated_count
);
2492 dev_info(&pdev
->dev
, "%d VFs allocated\n",
2493 adapter
->vfs_allocated_count
);
2494 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++)
2495 igb_vf_configure(adapter
, i
);
2497 /* DMA Coalescing is not supported in IOV mode. */
2498 adapter
->flags
&= ~IGB_FLAG_DMAC
;
2502 kfree(adapter
->vf_data
);
2503 adapter
->vf_data
= NULL
;
2504 adapter
->vfs_allocated_count
= 0;
2511 * igb_remove_i2c - Cleanup I2C interface
2512 * @adapter: pointer to adapter structure
2514 static void igb_remove_i2c(struct igb_adapter
*adapter
)
2516 /* free the adapter bus structure */
2517 i2c_del_adapter(&adapter
->i2c_adap
);
2521 * igb_remove - Device Removal Routine
2522 * @pdev: PCI device information struct
2524 * igb_remove is called by the PCI subsystem to alert the driver
2525 * that it should release a PCI device. The could be caused by a
2526 * Hot-Plug event, or because the driver is going to be removed from
2529 static void igb_remove(struct pci_dev
*pdev
)
2531 struct net_device
*netdev
= pci_get_drvdata(pdev
);
2532 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2533 struct e1000_hw
*hw
= &adapter
->hw
;
2535 pm_runtime_get_noresume(&pdev
->dev
);
2536 #ifdef CONFIG_IGB_HWMON
2537 igb_sysfs_exit(adapter
);
2539 igb_remove_i2c(adapter
);
2540 igb_ptp_stop(adapter
);
2541 /* The watchdog timer may be rescheduled, so explicitly
2542 * disable watchdog from being rescheduled.
2544 set_bit(__IGB_DOWN
, &adapter
->state
);
2545 del_timer_sync(&adapter
->watchdog_timer
);
2546 del_timer_sync(&adapter
->phy_info_timer
);
2548 cancel_work_sync(&adapter
->reset_task
);
2549 cancel_work_sync(&adapter
->watchdog_task
);
2551 #ifdef CONFIG_IGB_DCA
2552 if (adapter
->flags
& IGB_FLAG_DCA_ENABLED
) {
2553 dev_info(&pdev
->dev
, "DCA disabled\n");
2554 dca_remove_requester(&pdev
->dev
);
2555 adapter
->flags
&= ~IGB_FLAG_DCA_ENABLED
;
2556 wr32(E1000_DCA_CTRL
, E1000_DCA_CTRL_DCA_MODE_DISABLE
);
2560 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2561 * would have already happened in close and is redundant.
2563 igb_release_hw_control(adapter
);
2565 unregister_netdev(netdev
);
2567 igb_clear_interrupt_scheme(adapter
);
2569 #ifdef CONFIG_PCI_IOV
2570 igb_disable_sriov(pdev
);
2573 iounmap(hw
->hw_addr
);
2574 if (hw
->flash_address
)
2575 iounmap(hw
->flash_address
);
2576 pci_release_selected_regions(pdev
,
2577 pci_select_bars(pdev
, IORESOURCE_MEM
));
2579 kfree(adapter
->shadow_vfta
);
2580 free_netdev(netdev
);
2582 pci_disable_pcie_error_reporting(pdev
);
2584 pci_disable_device(pdev
);
2588 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2589 * @adapter: board private structure to initialize
2591 * This function initializes the vf specific data storage and then attempts to
2592 * allocate the VFs. The reason for ordering it this way is because it is much
2593 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2594 * the memory for the VFs.
2596 static void igb_probe_vfs(struct igb_adapter
*adapter
)
2598 #ifdef CONFIG_PCI_IOV
2599 struct pci_dev
*pdev
= adapter
->pdev
;
2600 struct e1000_hw
*hw
= &adapter
->hw
;
2602 /* Virtualization features not supported on i210 family. */
2603 if ((hw
->mac
.type
== e1000_i210
) || (hw
->mac
.type
== e1000_i211
))
2606 pci_sriov_set_totalvfs(pdev
, 7);
2607 igb_enable_sriov(pdev
, max_vfs
);
2609 #endif /* CONFIG_PCI_IOV */
2612 static void igb_init_queue_configuration(struct igb_adapter
*adapter
)
2614 struct e1000_hw
*hw
= &adapter
->hw
;
2617 /* Determine the maximum number of RSS queues supported. */
2618 switch (hw
->mac
.type
) {
2620 max_rss_queues
= IGB_MAX_RX_QUEUES_I211
;
2624 max_rss_queues
= IGB_MAX_RX_QUEUES_82575
;
2627 /* I350 cannot do RSS and SR-IOV at the same time */
2628 if (!!adapter
->vfs_allocated_count
) {
2634 if (!!adapter
->vfs_allocated_count
) {
2642 max_rss_queues
= IGB_MAX_RX_QUEUES
;
2646 adapter
->rss_queues
= min_t(u32
, max_rss_queues
, num_online_cpus());
2648 /* Determine if we need to pair queues. */
2649 switch (hw
->mac
.type
) {
2652 /* Device supports enough interrupts without queue pairing. */
2655 /* If VFs are going to be allocated with RSS queues then we
2656 * should pair the queues in order to conserve interrupts due
2657 * to limited supply.
2659 if ((adapter
->rss_queues
> 1) &&
2660 (adapter
->vfs_allocated_count
> 6))
2661 adapter
->flags
|= IGB_FLAG_QUEUE_PAIRS
;
2668 /* If rss_queues > half of max_rss_queues, pair the queues in
2669 * order to conserve interrupts due to limited supply.
2671 if (adapter
->rss_queues
> (max_rss_queues
/ 2))
2672 adapter
->flags
|= IGB_FLAG_QUEUE_PAIRS
;
2678 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2679 * @adapter: board private structure to initialize
2681 * igb_sw_init initializes the Adapter private data structure.
2682 * Fields are initialized based on PCI device information and
2683 * OS network device settings (MTU size).
2685 static int igb_sw_init(struct igb_adapter
*adapter
)
2687 struct e1000_hw
*hw
= &adapter
->hw
;
2688 struct net_device
*netdev
= adapter
->netdev
;
2689 struct pci_dev
*pdev
= adapter
->pdev
;
2691 pci_read_config_word(pdev
, PCI_COMMAND
, &hw
->bus
.pci_cmd_word
);
2693 /* set default ring sizes */
2694 adapter
->tx_ring_count
= IGB_DEFAULT_TXD
;
2695 adapter
->rx_ring_count
= IGB_DEFAULT_RXD
;
2697 /* set default ITR values */
2698 adapter
->rx_itr_setting
= IGB_DEFAULT_ITR
;
2699 adapter
->tx_itr_setting
= IGB_DEFAULT_ITR
;
2701 /* set default work limits */
2702 adapter
->tx_work_limit
= IGB_DEFAULT_TX_WORK
;
2704 adapter
->max_frame_size
= netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+
2706 adapter
->min_frame_size
= ETH_ZLEN
+ ETH_FCS_LEN
;
2708 spin_lock_init(&adapter
->stats64_lock
);
2709 #ifdef CONFIG_PCI_IOV
2710 switch (hw
->mac
.type
) {
2714 dev_warn(&pdev
->dev
,
2715 "Maximum of 7 VFs per PF, using max\n");
2716 max_vfs
= adapter
->vfs_allocated_count
= 7;
2718 adapter
->vfs_allocated_count
= max_vfs
;
2719 if (adapter
->vfs_allocated_count
)
2720 dev_warn(&pdev
->dev
,
2721 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
2726 #endif /* CONFIG_PCI_IOV */
2728 igb_init_queue_configuration(adapter
);
2730 /* Setup and initialize a copy of the hw vlan table array */
2731 adapter
->shadow_vfta
= kcalloc(E1000_VLAN_FILTER_TBL_SIZE
, sizeof(u32
),
2734 /* This call may decrease the number of queues */
2735 if (igb_init_interrupt_scheme(adapter
, true)) {
2736 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
2740 igb_probe_vfs(adapter
);
2742 /* Explicitly disable IRQ since the NIC can be in any state. */
2743 igb_irq_disable(adapter
);
2745 if (hw
->mac
.type
>= e1000_i350
)
2746 adapter
->flags
&= ~IGB_FLAG_DMAC
;
2748 set_bit(__IGB_DOWN
, &adapter
->state
);
2753 * igb_open - Called when a network interface is made active
2754 * @netdev: network interface device structure
2756 * Returns 0 on success, negative value on failure
2758 * The open entry point is called when a network interface is made
2759 * active by the system (IFF_UP). At this point all resources needed
2760 * for transmit and receive operations are allocated, the interrupt
2761 * handler is registered with the OS, the watchdog timer is started,
2762 * and the stack is notified that the interface is ready.
2764 static int __igb_open(struct net_device
*netdev
, bool resuming
)
2766 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2767 struct e1000_hw
*hw
= &adapter
->hw
;
2768 struct pci_dev
*pdev
= adapter
->pdev
;
2772 /* disallow open during test */
2773 if (test_bit(__IGB_TESTING
, &adapter
->state
)) {
2779 pm_runtime_get_sync(&pdev
->dev
);
2781 netif_carrier_off(netdev
);
2783 /* allocate transmit descriptors */
2784 err
= igb_setup_all_tx_resources(adapter
);
2788 /* allocate receive descriptors */
2789 err
= igb_setup_all_rx_resources(adapter
);
2793 igb_power_up_link(adapter
);
2795 /* before we allocate an interrupt, we must be ready to handle it.
2796 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2797 * as soon as we call pci_request_irq, so we have to setup our
2798 * clean_rx handler before we do so.
2800 igb_configure(adapter
);
2802 err
= igb_request_irq(adapter
);
2806 /* Notify the stack of the actual queue counts. */
2807 err
= netif_set_real_num_tx_queues(adapter
->netdev
,
2808 adapter
->num_tx_queues
);
2810 goto err_set_queues
;
2812 err
= netif_set_real_num_rx_queues(adapter
->netdev
,
2813 adapter
->num_rx_queues
);
2815 goto err_set_queues
;
2817 /* From here on the code is the same as igb_up() */
2818 clear_bit(__IGB_DOWN
, &adapter
->state
);
2820 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
2821 napi_enable(&(adapter
->q_vector
[i
]->napi
));
2823 /* Clear any pending interrupts. */
2826 igb_irq_enable(adapter
);
2828 /* notify VFs that reset has been completed */
2829 if (adapter
->vfs_allocated_count
) {
2830 u32 reg_data
= rd32(E1000_CTRL_EXT
);
2831 reg_data
|= E1000_CTRL_EXT_PFRSTD
;
2832 wr32(E1000_CTRL_EXT
, reg_data
);
2835 netif_tx_start_all_queues(netdev
);
2838 pm_runtime_put(&pdev
->dev
);
2840 /* start the watchdog. */
2841 hw
->mac
.get_link_status
= 1;
2842 schedule_work(&adapter
->watchdog_task
);
2847 igb_free_irq(adapter
);
2849 igb_release_hw_control(adapter
);
2850 igb_power_down_link(adapter
);
2851 igb_free_all_rx_resources(adapter
);
2853 igb_free_all_tx_resources(adapter
);
2857 pm_runtime_put(&pdev
->dev
);
2862 static int igb_open(struct net_device
*netdev
)
2864 return __igb_open(netdev
, false);
2868 * igb_close - Disables a network interface
2869 * @netdev: network interface device structure
2871 * Returns 0, this is not allowed to fail
2873 * The close entry point is called when an interface is de-activated
2874 * by the OS. The hardware is still under the driver's control, but
2875 * needs to be disabled. A global MAC reset is issued to stop the
2876 * hardware, and all transmit and receive resources are freed.
2878 static int __igb_close(struct net_device
*netdev
, bool suspending
)
2880 struct igb_adapter
*adapter
= netdev_priv(netdev
);
2881 struct pci_dev
*pdev
= adapter
->pdev
;
2883 WARN_ON(test_bit(__IGB_RESETTING
, &adapter
->state
));
2886 pm_runtime_get_sync(&pdev
->dev
);
2889 igb_free_irq(adapter
);
2891 igb_free_all_tx_resources(adapter
);
2892 igb_free_all_rx_resources(adapter
);
2895 pm_runtime_put_sync(&pdev
->dev
);
2899 static int igb_close(struct net_device
*netdev
)
2901 return __igb_close(netdev
, false);
2905 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2906 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2908 * Return 0 on success, negative on failure
2910 int igb_setup_tx_resources(struct igb_ring
*tx_ring
)
2912 struct device
*dev
= tx_ring
->dev
;
2915 size
= sizeof(struct igb_tx_buffer
) * tx_ring
->count
;
2917 tx_ring
->tx_buffer_info
= vzalloc(size
);
2918 if (!tx_ring
->tx_buffer_info
)
2921 /* round up to nearest 4K */
2922 tx_ring
->size
= tx_ring
->count
* sizeof(union e1000_adv_tx_desc
);
2923 tx_ring
->size
= ALIGN(tx_ring
->size
, 4096);
2925 tx_ring
->desc
= dma_alloc_coherent(dev
, tx_ring
->size
,
2926 &tx_ring
->dma
, GFP_KERNEL
);
2930 tx_ring
->next_to_use
= 0;
2931 tx_ring
->next_to_clean
= 0;
2936 vfree(tx_ring
->tx_buffer_info
);
2937 tx_ring
->tx_buffer_info
= NULL
;
2938 dev_err(dev
, "Unable to allocate memory for the Tx descriptor ring\n");
2943 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2944 * (Descriptors) for all queues
2945 * @adapter: board private structure
2947 * Return 0 on success, negative on failure
2949 static int igb_setup_all_tx_resources(struct igb_adapter
*adapter
)
2951 struct pci_dev
*pdev
= adapter
->pdev
;
2954 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
2955 err
= igb_setup_tx_resources(adapter
->tx_ring
[i
]);
2958 "Allocation for Tx Queue %u failed\n", i
);
2959 for (i
--; i
>= 0; i
--)
2960 igb_free_tx_resources(adapter
->tx_ring
[i
]);
2969 * igb_setup_tctl - configure the transmit control registers
2970 * @adapter: Board private structure
2972 void igb_setup_tctl(struct igb_adapter
*adapter
)
2974 struct e1000_hw
*hw
= &adapter
->hw
;
2977 /* disable queue 0 which is enabled by default on 82575 and 82576 */
2978 wr32(E1000_TXDCTL(0), 0);
2980 /* Program the Transmit Control Register */
2981 tctl
= rd32(E1000_TCTL
);
2982 tctl
&= ~E1000_TCTL_CT
;
2983 tctl
|= E1000_TCTL_PSP
| E1000_TCTL_RTLC
|
2984 (E1000_COLLISION_THRESHOLD
<< E1000_CT_SHIFT
);
2986 igb_config_collision_dist(hw
);
2988 /* Enable transmits */
2989 tctl
|= E1000_TCTL_EN
;
2991 wr32(E1000_TCTL
, tctl
);
2995 * igb_configure_tx_ring - Configure transmit ring after Reset
2996 * @adapter: board private structure
2997 * @ring: tx ring to configure
2999 * Configure a transmit ring after a reset.
3001 void igb_configure_tx_ring(struct igb_adapter
*adapter
,
3002 struct igb_ring
*ring
)
3004 struct e1000_hw
*hw
= &adapter
->hw
;
3006 u64 tdba
= ring
->dma
;
3007 int reg_idx
= ring
->reg_idx
;
3009 /* disable the queue */
3010 wr32(E1000_TXDCTL(reg_idx
), 0);
3014 wr32(E1000_TDLEN(reg_idx
),
3015 ring
->count
* sizeof(union e1000_adv_tx_desc
));
3016 wr32(E1000_TDBAL(reg_idx
),
3017 tdba
& 0x00000000ffffffffULL
);
3018 wr32(E1000_TDBAH(reg_idx
), tdba
>> 32);
3020 ring
->tail
= hw
->hw_addr
+ E1000_TDT(reg_idx
);
3021 wr32(E1000_TDH(reg_idx
), 0);
3022 writel(0, ring
->tail
);
3024 txdctl
|= IGB_TX_PTHRESH
;
3025 txdctl
|= IGB_TX_HTHRESH
<< 8;
3026 txdctl
|= IGB_TX_WTHRESH
<< 16;
3028 txdctl
|= E1000_TXDCTL_QUEUE_ENABLE
;
3029 wr32(E1000_TXDCTL(reg_idx
), txdctl
);
3033 * igb_configure_tx - Configure transmit Unit after Reset
3034 * @adapter: board private structure
3036 * Configure the Tx unit of the MAC after a reset.
3038 static void igb_configure_tx(struct igb_adapter
*adapter
)
3042 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3043 igb_configure_tx_ring(adapter
, adapter
->tx_ring
[i
]);
3047 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
3048 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3050 * Returns 0 on success, negative on failure
3052 int igb_setup_rx_resources(struct igb_ring
*rx_ring
)
3054 struct device
*dev
= rx_ring
->dev
;
3057 size
= sizeof(struct igb_rx_buffer
) * rx_ring
->count
;
3059 rx_ring
->rx_buffer_info
= vzalloc(size
);
3060 if (!rx_ring
->rx_buffer_info
)
3063 /* Round up to nearest 4K */
3064 rx_ring
->size
= rx_ring
->count
* sizeof(union e1000_adv_rx_desc
);
3065 rx_ring
->size
= ALIGN(rx_ring
->size
, 4096);
3067 rx_ring
->desc
= dma_alloc_coherent(dev
, rx_ring
->size
,
3068 &rx_ring
->dma
, GFP_KERNEL
);
3072 rx_ring
->next_to_alloc
= 0;
3073 rx_ring
->next_to_clean
= 0;
3074 rx_ring
->next_to_use
= 0;
3079 vfree(rx_ring
->rx_buffer_info
);
3080 rx_ring
->rx_buffer_info
= NULL
;
3081 dev_err(dev
, "Unable to allocate memory for the Rx descriptor ring\n");
3086 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
3087 * (Descriptors) for all queues
3088 * @adapter: board private structure
3090 * Return 0 on success, negative on failure
3092 static int igb_setup_all_rx_resources(struct igb_adapter
*adapter
)
3094 struct pci_dev
*pdev
= adapter
->pdev
;
3097 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3098 err
= igb_setup_rx_resources(adapter
->rx_ring
[i
]);
3101 "Allocation for Rx Queue %u failed\n", i
);
3102 for (i
--; i
>= 0; i
--)
3103 igb_free_rx_resources(adapter
->rx_ring
[i
]);
3112 * igb_setup_mrqc - configure the multiple receive queue control registers
3113 * @adapter: Board private structure
3115 static void igb_setup_mrqc(struct igb_adapter
*adapter
)
3117 struct e1000_hw
*hw
= &adapter
->hw
;
3119 u32 j
, num_rx_queues
, shift
= 0;
3120 static const u32 rsskey
[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
3121 0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
3122 0xA32DCB77, 0x0CF23080, 0x3BB7426A,
3125 /* Fill out hash function seeds */
3126 for (j
= 0; j
< 10; j
++)
3127 wr32(E1000_RSSRK(j
), rsskey
[j
]);
3129 num_rx_queues
= adapter
->rss_queues
;
3131 switch (hw
->mac
.type
) {
3136 /* 82576 supports 2 RSS queues for SR-IOV */
3137 if (adapter
->vfs_allocated_count
) {
3146 /* Populate the indirection table 4 entries at a time. To do this
3147 * we are generating the results for n and n+2 and then interleaving
3148 * those with the results with n+1 and n+3.
3150 for (j
= 0; j
< 32; j
++) {
3151 /* first pass generates n and n+2 */
3152 u32 base
= ((j
* 0x00040004) + 0x00020000) * num_rx_queues
;
3153 u32 reta
= (base
& 0x07800780) >> (7 - shift
);
3155 /* second pass generates n+1 and n+3 */
3156 base
+= 0x00010001 * num_rx_queues
;
3157 reta
|= (base
& 0x07800780) << (1 + shift
);
3159 wr32(E1000_RETA(j
), reta
);
3162 /* Disable raw packet checksumming so that RSS hash is placed in
3163 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
3164 * offloads as they are enabled by default
3166 rxcsum
= rd32(E1000_RXCSUM
);
3167 rxcsum
|= E1000_RXCSUM_PCSD
;
3169 if (adapter
->hw
.mac
.type
>= e1000_82576
)
3170 /* Enable Receive Checksum Offload for SCTP */
3171 rxcsum
|= E1000_RXCSUM_CRCOFL
;
3173 /* Don't need to set TUOFL or IPOFL, they default to 1 */
3174 wr32(E1000_RXCSUM
, rxcsum
);
3176 /* Generate RSS hash based on packet types, TCP/UDP
3177 * port numbers and/or IPv4/v6 src and dst addresses
3179 mrqc
= E1000_MRQC_RSS_FIELD_IPV4
|
3180 E1000_MRQC_RSS_FIELD_IPV4_TCP
|
3181 E1000_MRQC_RSS_FIELD_IPV6
|
3182 E1000_MRQC_RSS_FIELD_IPV6_TCP
|
3183 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX
;
3185 if (adapter
->flags
& IGB_FLAG_RSS_FIELD_IPV4_UDP
)
3186 mrqc
|= E1000_MRQC_RSS_FIELD_IPV4_UDP
;
3187 if (adapter
->flags
& IGB_FLAG_RSS_FIELD_IPV6_UDP
)
3188 mrqc
|= E1000_MRQC_RSS_FIELD_IPV6_UDP
;
3190 /* If VMDq is enabled then we set the appropriate mode for that, else
3191 * we default to RSS so that an RSS hash is calculated per packet even
3192 * if we are only using one queue
3194 if (adapter
->vfs_allocated_count
) {
3195 if (hw
->mac
.type
> e1000_82575
) {
3196 /* Set the default pool for the PF's first queue */
3197 u32 vtctl
= rd32(E1000_VT_CTL
);
3198 vtctl
&= ~(E1000_VT_CTL_DEFAULT_POOL_MASK
|
3199 E1000_VT_CTL_DISABLE_DEF_POOL
);
3200 vtctl
|= adapter
->vfs_allocated_count
<<
3201 E1000_VT_CTL_DEFAULT_POOL_SHIFT
;
3202 wr32(E1000_VT_CTL
, vtctl
);
3204 if (adapter
->rss_queues
> 1)
3205 mrqc
|= E1000_MRQC_ENABLE_VMDQ_RSS_2Q
;
3207 mrqc
|= E1000_MRQC_ENABLE_VMDQ
;
3209 if (hw
->mac
.type
!= e1000_i211
)
3210 mrqc
|= E1000_MRQC_ENABLE_RSS_4Q
;
3212 igb_vmm_control(adapter
);
3214 wr32(E1000_MRQC
, mrqc
);
3218 * igb_setup_rctl - configure the receive control registers
3219 * @adapter: Board private structure
3221 void igb_setup_rctl(struct igb_adapter
*adapter
)
3223 struct e1000_hw
*hw
= &adapter
->hw
;
3226 rctl
= rd32(E1000_RCTL
);
3228 rctl
&= ~(3 << E1000_RCTL_MO_SHIFT
);
3229 rctl
&= ~(E1000_RCTL_LBM_TCVR
| E1000_RCTL_LBM_MAC
);
3231 rctl
|= E1000_RCTL_EN
| E1000_RCTL_BAM
| E1000_RCTL_RDMTS_HALF
|
3232 (hw
->mac
.mc_filter_type
<< E1000_RCTL_MO_SHIFT
);
3234 /* enable stripping of CRC. It's unlikely this will break BMC
3235 * redirection as it did with e1000. Newer features require
3236 * that the HW strips the CRC.
3238 rctl
|= E1000_RCTL_SECRC
;
3240 /* disable store bad packets and clear size bits. */
3241 rctl
&= ~(E1000_RCTL_SBP
| E1000_RCTL_SZ_256
);
3243 /* enable LPE to prevent packets larger than max_frame_size */
3244 rctl
|= E1000_RCTL_LPE
;
3246 /* disable queue 0 to prevent tail write w/o re-config */
3247 wr32(E1000_RXDCTL(0), 0);
3249 /* Attention!!! For SR-IOV PF driver operations you must enable
3250 * queue drop for all VF and PF queues to prevent head of line blocking
3251 * if an un-trusted VF does not provide descriptors to hardware.
3253 if (adapter
->vfs_allocated_count
) {
3254 /* set all queue drop enable bits */
3255 wr32(E1000_QDE
, ALL_QUEUES
);
3258 /* This is useful for sniffing bad packets. */
3259 if (adapter
->netdev
->features
& NETIF_F_RXALL
) {
3260 /* UPE and MPE will be handled by normal PROMISC logic
3261 * in e1000e_set_rx_mode
3263 rctl
|= (E1000_RCTL_SBP
| /* Receive bad packets */
3264 E1000_RCTL_BAM
| /* RX All Bcast Pkts */
3265 E1000_RCTL_PMCF
); /* RX All MAC Ctrl Pkts */
3267 rctl
&= ~(E1000_RCTL_VFE
| /* Disable VLAN filter */
3268 E1000_RCTL_DPF
| /* Allow filtered pause */
3269 E1000_RCTL_CFIEN
); /* Dis VLAN CFIEN Filter */
3270 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3271 * and that breaks VLANs.
3275 wr32(E1000_RCTL
, rctl
);
3278 static inline int igb_set_vf_rlpml(struct igb_adapter
*adapter
, int size
,
3281 struct e1000_hw
*hw
= &adapter
->hw
;
3284 /* if it isn't the PF check to see if VFs are enabled and
3285 * increase the size to support vlan tags
3287 if (vfn
< adapter
->vfs_allocated_count
&&
3288 adapter
->vf_data
[vfn
].vlans_enabled
)
3289 size
+= VLAN_TAG_SIZE
;
3291 vmolr
= rd32(E1000_VMOLR(vfn
));
3292 vmolr
&= ~E1000_VMOLR_RLPML_MASK
;
3293 vmolr
|= size
| E1000_VMOLR_LPE
;
3294 wr32(E1000_VMOLR(vfn
), vmolr
);
3300 * igb_rlpml_set - set maximum receive packet size
3301 * @adapter: board private structure
3303 * Configure maximum receivable packet size.
3305 static void igb_rlpml_set(struct igb_adapter
*adapter
)
3307 u32 max_frame_size
= adapter
->max_frame_size
;
3308 struct e1000_hw
*hw
= &adapter
->hw
;
3309 u16 pf_id
= adapter
->vfs_allocated_count
;
3312 igb_set_vf_rlpml(adapter
, max_frame_size
, pf_id
);
3313 /* If we're in VMDQ or SR-IOV mode, then set global RLPML
3314 * to our max jumbo frame size, in case we need to enable
3315 * jumbo frames on one of the rings later.
3316 * This will not pass over-length frames into the default
3317 * queue because it's gated by the VMOLR.RLPML.
3319 max_frame_size
= MAX_JUMBO_FRAME_SIZE
;
3322 wr32(E1000_RLPML
, max_frame_size
);
3325 static inline void igb_set_vmolr(struct igb_adapter
*adapter
,
3328 struct e1000_hw
*hw
= &adapter
->hw
;
3331 /* This register exists only on 82576 and newer so if we are older then
3332 * we should exit and do nothing
3334 if (hw
->mac
.type
< e1000_82576
)
3337 vmolr
= rd32(E1000_VMOLR(vfn
));
3338 vmolr
|= E1000_VMOLR_STRVLAN
; /* Strip vlan tags */
3340 vmolr
|= E1000_VMOLR_AUPE
; /* Accept untagged packets */
3342 vmolr
&= ~(E1000_VMOLR_AUPE
); /* Tagged packets ONLY */
3344 /* clear all bits that might not be set */
3345 vmolr
&= ~(E1000_VMOLR_BAM
| E1000_VMOLR_RSSE
);
3347 if (adapter
->rss_queues
> 1 && vfn
== adapter
->vfs_allocated_count
)
3348 vmolr
|= E1000_VMOLR_RSSE
; /* enable RSS */
3349 /* for VMDq only allow the VFs and pool 0 to accept broadcast and
3352 if (vfn
<= adapter
->vfs_allocated_count
)
3353 vmolr
|= E1000_VMOLR_BAM
; /* Accept broadcast */
3355 wr32(E1000_VMOLR(vfn
), vmolr
);
3359 * igb_configure_rx_ring - Configure a receive ring after Reset
3360 * @adapter: board private structure
3361 * @ring: receive ring to be configured
3363 * Configure the Rx unit of the MAC after a reset.
3365 void igb_configure_rx_ring(struct igb_adapter
*adapter
,
3366 struct igb_ring
*ring
)
3368 struct e1000_hw
*hw
= &adapter
->hw
;
3369 u64 rdba
= ring
->dma
;
3370 int reg_idx
= ring
->reg_idx
;
3371 u32 srrctl
= 0, rxdctl
= 0;
3373 /* disable the queue */
3374 wr32(E1000_RXDCTL(reg_idx
), 0);
3376 /* Set DMA base address registers */
3377 wr32(E1000_RDBAL(reg_idx
),
3378 rdba
& 0x00000000ffffffffULL
);
3379 wr32(E1000_RDBAH(reg_idx
), rdba
>> 32);
3380 wr32(E1000_RDLEN(reg_idx
),
3381 ring
->count
* sizeof(union e1000_adv_rx_desc
));
3383 /* initialize head and tail */
3384 ring
->tail
= hw
->hw_addr
+ E1000_RDT(reg_idx
);
3385 wr32(E1000_RDH(reg_idx
), 0);
3386 writel(0, ring
->tail
);
3388 /* set descriptor configuration */
3389 srrctl
= IGB_RX_HDR_LEN
<< E1000_SRRCTL_BSIZEHDRSIZE_SHIFT
;
3390 srrctl
|= IGB_RX_BUFSZ
>> E1000_SRRCTL_BSIZEPKT_SHIFT
;
3391 srrctl
|= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF
;
3392 if (hw
->mac
.type
>= e1000_82580
)
3393 srrctl
|= E1000_SRRCTL_TIMESTAMP
;
3394 /* Only set Drop Enable if we are supporting multiple queues */
3395 if (adapter
->vfs_allocated_count
|| adapter
->num_rx_queues
> 1)
3396 srrctl
|= E1000_SRRCTL_DROP_EN
;
3398 wr32(E1000_SRRCTL(reg_idx
), srrctl
);
3400 /* set filtering for VMDQ pools */
3401 igb_set_vmolr(adapter
, reg_idx
& 0x7, true);
3403 rxdctl
|= IGB_RX_PTHRESH
;
3404 rxdctl
|= IGB_RX_HTHRESH
<< 8;
3405 rxdctl
|= IGB_RX_WTHRESH
<< 16;
3407 /* enable receive descriptor fetching */
3408 rxdctl
|= E1000_RXDCTL_QUEUE_ENABLE
;
3409 wr32(E1000_RXDCTL(reg_idx
), rxdctl
);
3413 * igb_configure_rx - Configure receive Unit after Reset
3414 * @adapter: board private structure
3416 * Configure the Rx unit of the MAC after a reset.
3418 static void igb_configure_rx(struct igb_adapter
*adapter
)
3422 /* set UTA to appropriate mode */
3423 igb_set_uta(adapter
);
3425 /* set the correct pool for the PF default MAC address in entry 0 */
3426 igb_rar_set_qsel(adapter
, adapter
->hw
.mac
.addr
, 0,
3427 adapter
->vfs_allocated_count
);
3429 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3430 * the Base and Length of the Rx Descriptor Ring
3432 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3433 igb_configure_rx_ring(adapter
, adapter
->rx_ring
[i
]);
3437 * igb_free_tx_resources - Free Tx Resources per Queue
3438 * @tx_ring: Tx descriptor ring for a specific queue
3440 * Free all transmit software resources
3442 void igb_free_tx_resources(struct igb_ring
*tx_ring
)
3444 igb_clean_tx_ring(tx_ring
);
3446 vfree(tx_ring
->tx_buffer_info
);
3447 tx_ring
->tx_buffer_info
= NULL
;
3449 /* if not set, then don't free */
3453 dma_free_coherent(tx_ring
->dev
, tx_ring
->size
,
3454 tx_ring
->desc
, tx_ring
->dma
);
3456 tx_ring
->desc
= NULL
;
3460 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3461 * @adapter: board private structure
3463 * Free all transmit software resources
3465 static void igb_free_all_tx_resources(struct igb_adapter
*adapter
)
3469 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3470 igb_free_tx_resources(adapter
->tx_ring
[i
]);
3473 void igb_unmap_and_free_tx_resource(struct igb_ring
*ring
,
3474 struct igb_tx_buffer
*tx_buffer
)
3476 if (tx_buffer
->skb
) {
3477 dev_kfree_skb_any(tx_buffer
->skb
);
3478 if (dma_unmap_len(tx_buffer
, len
))
3479 dma_unmap_single(ring
->dev
,
3480 dma_unmap_addr(tx_buffer
, dma
),
3481 dma_unmap_len(tx_buffer
, len
),
3483 } else if (dma_unmap_len(tx_buffer
, len
)) {
3484 dma_unmap_page(ring
->dev
,
3485 dma_unmap_addr(tx_buffer
, dma
),
3486 dma_unmap_len(tx_buffer
, len
),
3489 tx_buffer
->next_to_watch
= NULL
;
3490 tx_buffer
->skb
= NULL
;
3491 dma_unmap_len_set(tx_buffer
, len
, 0);
3492 /* buffer_info must be completely set up in the transmit path */
3496 * igb_clean_tx_ring - Free Tx Buffers
3497 * @tx_ring: ring to be cleaned
3499 static void igb_clean_tx_ring(struct igb_ring
*tx_ring
)
3501 struct igb_tx_buffer
*buffer_info
;
3505 if (!tx_ring
->tx_buffer_info
)
3507 /* Free all the Tx ring sk_buffs */
3509 for (i
= 0; i
< tx_ring
->count
; i
++) {
3510 buffer_info
= &tx_ring
->tx_buffer_info
[i
];
3511 igb_unmap_and_free_tx_resource(tx_ring
, buffer_info
);
3514 netdev_tx_reset_queue(txring_txq(tx_ring
));
3516 size
= sizeof(struct igb_tx_buffer
) * tx_ring
->count
;
3517 memset(tx_ring
->tx_buffer_info
, 0, size
);
3519 /* Zero out the descriptor ring */
3520 memset(tx_ring
->desc
, 0, tx_ring
->size
);
3522 tx_ring
->next_to_use
= 0;
3523 tx_ring
->next_to_clean
= 0;
3527 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3528 * @adapter: board private structure
3530 static void igb_clean_all_tx_rings(struct igb_adapter
*adapter
)
3534 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3535 igb_clean_tx_ring(adapter
->tx_ring
[i
]);
3539 * igb_free_rx_resources - Free Rx Resources
3540 * @rx_ring: ring to clean the resources from
3542 * Free all receive software resources
3544 void igb_free_rx_resources(struct igb_ring
*rx_ring
)
3546 igb_clean_rx_ring(rx_ring
);
3548 vfree(rx_ring
->rx_buffer_info
);
3549 rx_ring
->rx_buffer_info
= NULL
;
3551 /* if not set, then don't free */
3555 dma_free_coherent(rx_ring
->dev
, rx_ring
->size
,
3556 rx_ring
->desc
, rx_ring
->dma
);
3558 rx_ring
->desc
= NULL
;
3562 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3563 * @adapter: board private structure
3565 * Free all receive software resources
3567 static void igb_free_all_rx_resources(struct igb_adapter
*adapter
)
3571 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3572 igb_free_rx_resources(adapter
->rx_ring
[i
]);
3576 * igb_clean_rx_ring - Free Rx Buffers per Queue
3577 * @rx_ring: ring to free buffers from
3579 static void igb_clean_rx_ring(struct igb_ring
*rx_ring
)
3585 dev_kfree_skb(rx_ring
->skb
);
3586 rx_ring
->skb
= NULL
;
3588 if (!rx_ring
->rx_buffer_info
)
3591 /* Free all the Rx ring sk_buffs */
3592 for (i
= 0; i
< rx_ring
->count
; i
++) {
3593 struct igb_rx_buffer
*buffer_info
= &rx_ring
->rx_buffer_info
[i
];
3595 if (!buffer_info
->page
)
3598 dma_unmap_page(rx_ring
->dev
,
3602 __free_page(buffer_info
->page
);
3604 buffer_info
->page
= NULL
;
3607 size
= sizeof(struct igb_rx_buffer
) * rx_ring
->count
;
3608 memset(rx_ring
->rx_buffer_info
, 0, size
);
3610 /* Zero out the descriptor ring */
3611 memset(rx_ring
->desc
, 0, rx_ring
->size
);
3613 rx_ring
->next_to_alloc
= 0;
3614 rx_ring
->next_to_clean
= 0;
3615 rx_ring
->next_to_use
= 0;
3619 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3620 * @adapter: board private structure
3622 static void igb_clean_all_rx_rings(struct igb_adapter
*adapter
)
3626 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3627 igb_clean_rx_ring(adapter
->rx_ring
[i
]);
3631 * igb_set_mac - Change the Ethernet Address of the NIC
3632 * @netdev: network interface device structure
3633 * @p: pointer to an address structure
3635 * Returns 0 on success, negative on failure
3637 static int igb_set_mac(struct net_device
*netdev
, void *p
)
3639 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3640 struct e1000_hw
*hw
= &adapter
->hw
;
3641 struct sockaddr
*addr
= p
;
3643 if (!is_valid_ether_addr(addr
->sa_data
))
3644 return -EADDRNOTAVAIL
;
3646 memcpy(netdev
->dev_addr
, addr
->sa_data
, netdev
->addr_len
);
3647 memcpy(hw
->mac
.addr
, addr
->sa_data
, netdev
->addr_len
);
3649 /* set the correct pool for the new PF MAC address in entry 0 */
3650 igb_rar_set_qsel(adapter
, hw
->mac
.addr
, 0,
3651 adapter
->vfs_allocated_count
);
3657 * igb_write_mc_addr_list - write multicast addresses to MTA
3658 * @netdev: network interface device structure
3660 * Writes multicast address list to the MTA hash table.
3661 * Returns: -ENOMEM on failure
3662 * 0 on no addresses written
3663 * X on writing X addresses to MTA
3665 static int igb_write_mc_addr_list(struct net_device
*netdev
)
3667 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3668 struct e1000_hw
*hw
= &adapter
->hw
;
3669 struct netdev_hw_addr
*ha
;
3673 if (netdev_mc_empty(netdev
)) {
3674 /* nothing to program, so clear mc list */
3675 igb_update_mc_addr_list(hw
, NULL
, 0);
3676 igb_restore_vf_multicasts(adapter
);
3680 mta_list
= kzalloc(netdev_mc_count(netdev
) * 6, GFP_ATOMIC
);
3684 /* The shared function expects a packed array of only addresses. */
3686 netdev_for_each_mc_addr(ha
, netdev
)
3687 memcpy(mta_list
+ (i
++ * ETH_ALEN
), ha
->addr
, ETH_ALEN
);
3689 igb_update_mc_addr_list(hw
, mta_list
, i
);
3692 return netdev_mc_count(netdev
);
3696 * igb_write_uc_addr_list - write unicast addresses to RAR table
3697 * @netdev: network interface device structure
3699 * Writes unicast address list to the RAR table.
3700 * Returns: -ENOMEM on failure/insufficient address space
3701 * 0 on no addresses written
3702 * X on writing X addresses to the RAR table
3704 static int igb_write_uc_addr_list(struct net_device
*netdev
)
3706 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3707 struct e1000_hw
*hw
= &adapter
->hw
;
3708 unsigned int vfn
= adapter
->vfs_allocated_count
;
3709 unsigned int rar_entries
= hw
->mac
.rar_entry_count
- (vfn
+ 1);
3712 /* return ENOMEM indicating insufficient memory for addresses */
3713 if (netdev_uc_count(netdev
) > rar_entries
)
3716 if (!netdev_uc_empty(netdev
) && rar_entries
) {
3717 struct netdev_hw_addr
*ha
;
3719 netdev_for_each_uc_addr(ha
, netdev
) {
3722 igb_rar_set_qsel(adapter
, ha
->addr
,
3728 /* write the addresses in reverse order to avoid write combining */
3729 for (; rar_entries
> 0 ; rar_entries
--) {
3730 wr32(E1000_RAH(rar_entries
), 0);
3731 wr32(E1000_RAL(rar_entries
), 0);
3739 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3740 * @netdev: network interface device structure
3742 * The set_rx_mode entry point is called whenever the unicast or multicast
3743 * address lists or the network interface flags are updated. This routine is
3744 * responsible for configuring the hardware for proper unicast, multicast,
3745 * promiscuous mode, and all-multi behavior.
3747 static void igb_set_rx_mode(struct net_device
*netdev
)
3749 struct igb_adapter
*adapter
= netdev_priv(netdev
);
3750 struct e1000_hw
*hw
= &adapter
->hw
;
3751 unsigned int vfn
= adapter
->vfs_allocated_count
;
3752 u32 rctl
, vmolr
= 0;
3755 /* Check for Promiscuous and All Multicast modes */
3756 rctl
= rd32(E1000_RCTL
);
3758 /* clear the effected bits */
3759 rctl
&= ~(E1000_RCTL_UPE
| E1000_RCTL_MPE
| E1000_RCTL_VFE
);
3761 if (netdev
->flags
& IFF_PROMISC
) {
3762 /* retain VLAN HW filtering if in VT mode */
3763 if (adapter
->vfs_allocated_count
)
3764 rctl
|= E1000_RCTL_VFE
;
3765 rctl
|= (E1000_RCTL_UPE
| E1000_RCTL_MPE
);
3766 vmolr
|= (E1000_VMOLR_ROPE
| E1000_VMOLR_MPME
);
3768 if (netdev
->flags
& IFF_ALLMULTI
) {
3769 rctl
|= E1000_RCTL_MPE
;
3770 vmolr
|= E1000_VMOLR_MPME
;
3772 /* Write addresses to the MTA, if the attempt fails
3773 * then we should just turn on promiscuous mode so
3774 * that we can at least receive multicast traffic
3776 count
= igb_write_mc_addr_list(netdev
);
3778 rctl
|= E1000_RCTL_MPE
;
3779 vmolr
|= E1000_VMOLR_MPME
;
3781 vmolr
|= E1000_VMOLR_ROMPE
;
3784 /* Write addresses to available RAR registers, if there is not
3785 * sufficient space to store all the addresses then enable
3786 * unicast promiscuous mode
3788 count
= igb_write_uc_addr_list(netdev
);
3790 rctl
|= E1000_RCTL_UPE
;
3791 vmolr
|= E1000_VMOLR_ROPE
;
3793 rctl
|= E1000_RCTL_VFE
;
3795 wr32(E1000_RCTL
, rctl
);
3797 /* In order to support SR-IOV and eventually VMDq it is necessary to set
3798 * the VMOLR to enable the appropriate modes. Without this workaround
3799 * we will have issues with VLAN tag stripping not being done for frames
3800 * that are only arriving because we are the default pool
3802 if ((hw
->mac
.type
< e1000_82576
) || (hw
->mac
.type
> e1000_i350
))
3805 vmolr
|= rd32(E1000_VMOLR(vfn
)) &
3806 ~(E1000_VMOLR_ROPE
| E1000_VMOLR_MPME
| E1000_VMOLR_ROMPE
);
3807 wr32(E1000_VMOLR(vfn
), vmolr
);
3808 igb_restore_vf_multicasts(adapter
);
3811 static void igb_check_wvbr(struct igb_adapter
*adapter
)
3813 struct e1000_hw
*hw
= &adapter
->hw
;
3816 switch (hw
->mac
.type
) {
3819 if (!(wvbr
= rd32(E1000_WVBR
)))
3826 adapter
->wvbr
|= wvbr
;
3829 #define IGB_STAGGERED_QUEUE_OFFSET 8
3831 static void igb_spoof_check(struct igb_adapter
*adapter
)
3838 for(j
= 0; j
< adapter
->vfs_allocated_count
; j
++) {
3839 if (adapter
->wvbr
& (1 << j
) ||
3840 adapter
->wvbr
& (1 << (j
+ IGB_STAGGERED_QUEUE_OFFSET
))) {
3841 dev_warn(&adapter
->pdev
->dev
,
3842 "Spoof event(s) detected on VF %d\n", j
);
3845 (1 << (j
+ IGB_STAGGERED_QUEUE_OFFSET
)));
3850 /* Need to wait a few seconds after link up to get diagnostic information from
3853 static void igb_update_phy_info(unsigned long data
)
3855 struct igb_adapter
*adapter
= (struct igb_adapter
*) data
;
3856 igb_get_phy_info(&adapter
->hw
);
3860 * igb_has_link - check shared code for link and determine up/down
3861 * @adapter: pointer to driver private info
3863 bool igb_has_link(struct igb_adapter
*adapter
)
3865 struct e1000_hw
*hw
= &adapter
->hw
;
3866 bool link_active
= false;
3868 /* get_link_status is set on LSC (link status) interrupt or
3869 * rx sequence error interrupt. get_link_status will stay
3870 * false until the e1000_check_for_link establishes link
3871 * for copper adapters ONLY
3873 switch (hw
->phy
.media_type
) {
3874 case e1000_media_type_copper
:
3875 if (!hw
->mac
.get_link_status
)
3877 case e1000_media_type_internal_serdes
:
3878 hw
->mac
.ops
.check_for_link(hw
);
3879 link_active
= !hw
->mac
.get_link_status
;
3882 case e1000_media_type_unknown
:
3889 static bool igb_thermal_sensor_event(struct e1000_hw
*hw
, u32 event
)
3892 u32 ctrl_ext
, thstat
;
3894 /* check for thermal sensor event on i350 copper only */
3895 if (hw
->mac
.type
== e1000_i350
) {
3896 thstat
= rd32(E1000_THSTAT
);
3897 ctrl_ext
= rd32(E1000_CTRL_EXT
);
3899 if ((hw
->phy
.media_type
== e1000_media_type_copper
) &&
3900 !(ctrl_ext
& E1000_CTRL_EXT_LINK_MODE_SGMII
))
3901 ret
= !!(thstat
& event
);
3908 * igb_watchdog - Timer Call-back
3909 * @data: pointer to adapter cast into an unsigned long
3911 static void igb_watchdog(unsigned long data
)
3913 struct igb_adapter
*adapter
= (struct igb_adapter
*)data
;
3914 /* Do the rest outside of interrupt context */
3915 schedule_work(&adapter
->watchdog_task
);
3918 static void igb_watchdog_task(struct work_struct
*work
)
3920 struct igb_adapter
*adapter
= container_of(work
,
3923 struct e1000_hw
*hw
= &adapter
->hw
;
3924 struct e1000_phy_info
*phy
= &hw
->phy
;
3925 struct net_device
*netdev
= adapter
->netdev
;
3929 link
= igb_has_link(adapter
);
3931 /* Cancel scheduled suspend requests. */
3932 pm_runtime_resume(netdev
->dev
.parent
);
3934 if (!netif_carrier_ok(netdev
)) {
3936 hw
->mac
.ops
.get_speed_and_duplex(hw
,
3937 &adapter
->link_speed
,
3938 &adapter
->link_duplex
);
3940 ctrl
= rd32(E1000_CTRL
);
3941 /* Links status message must follow this format */
3942 printk(KERN_INFO
"igb: %s NIC Link is Up %d Mbps %s "
3943 "Duplex, Flow Control: %s\n",
3945 adapter
->link_speed
,
3946 adapter
->link_duplex
== FULL_DUPLEX
?
3948 (ctrl
& E1000_CTRL_TFCE
) &&
3949 (ctrl
& E1000_CTRL_RFCE
) ? "RX/TX" :
3950 (ctrl
& E1000_CTRL_RFCE
) ? "RX" :
3951 (ctrl
& E1000_CTRL_TFCE
) ? "TX" : "None");
3953 /* check if SmartSpeed worked */
3954 igb_check_downshift(hw
);
3955 if (phy
->speed_downgraded
)
3956 netdev_warn(netdev
, "Link Speed was downgraded by SmartSpeed\n");
3958 /* check for thermal sensor event */
3959 if (igb_thermal_sensor_event(hw
,
3960 E1000_THSTAT_LINK_THROTTLE
)) {
3961 netdev_info(netdev
, "The network adapter link "
3962 "speed was downshifted because it "
3966 /* adjust timeout factor according to speed/duplex */
3967 adapter
->tx_timeout_factor
= 1;
3968 switch (adapter
->link_speed
) {
3970 adapter
->tx_timeout_factor
= 14;
3973 /* maybe add some timeout factor ? */
3977 netif_carrier_on(netdev
);
3979 igb_ping_all_vfs(adapter
);
3980 igb_check_vf_rate_limit(adapter
);
3982 /* link state has changed, schedule phy info update */
3983 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
3984 mod_timer(&adapter
->phy_info_timer
,
3985 round_jiffies(jiffies
+ 2 * HZ
));
3988 if (netif_carrier_ok(netdev
)) {
3989 adapter
->link_speed
= 0;
3990 adapter
->link_duplex
= 0;
3992 /* check for thermal sensor event */
3993 if (igb_thermal_sensor_event(hw
,
3994 E1000_THSTAT_PWR_DOWN
)) {
3995 netdev_err(netdev
, "The network adapter was "
3996 "stopped because it overheated\n");
3999 /* Links status message must follow this format */
4000 printk(KERN_INFO
"igb: %s NIC Link is Down\n",
4002 netif_carrier_off(netdev
);
4004 igb_ping_all_vfs(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
));
4011 pm_schedule_suspend(netdev
->dev
.parent
,
4016 spin_lock(&adapter
->stats64_lock
);
4017 igb_update_stats(adapter
, &adapter
->stats64
);
4018 spin_unlock(&adapter
->stats64_lock
);
4020 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
4021 struct igb_ring
*tx_ring
= adapter
->tx_ring
[i
];
4022 if (!netif_carrier_ok(netdev
)) {
4023 /* We've lost link, so the controller stops DMA,
4024 * but we've got queued Tx work that's never going
4025 * to get done, so reset controller to flush Tx.
4026 * (Do the reset outside of interrupt context).
4028 if (igb_desc_unused(tx_ring
) + 1 < tx_ring
->count
) {
4029 adapter
->tx_timeout_count
++;
4030 schedule_work(&adapter
->reset_task
);
4031 /* return immediately since reset is imminent */
4036 /* Force detection of hung controller every watchdog period */
4037 set_bit(IGB_RING_FLAG_TX_DETECT_HANG
, &tx_ring
->flags
);
4040 /* Cause software interrupt to ensure Rx ring is cleaned */
4041 if (adapter
->msix_entries
) {
4043 for (i
= 0; i
< adapter
->num_q_vectors
; i
++)
4044 eics
|= adapter
->q_vector
[i
]->eims_value
;
4045 wr32(E1000_EICS
, eics
);
4047 wr32(E1000_ICS
, E1000_ICS_RXDMT0
);
4050 igb_spoof_check(adapter
);
4051 igb_ptp_rx_hang(adapter
);
4053 /* Reset the timer */
4054 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
4055 mod_timer(&adapter
->watchdog_timer
,
4056 round_jiffies(jiffies
+ 2 * HZ
));
4059 enum latency_range
{
4063 latency_invalid
= 255
4067 * igb_update_ring_itr - update the dynamic ITR value based on packet size
4068 * @q_vector: pointer to q_vector
4070 * Stores a new ITR value based on strictly on packet size. This
4071 * algorithm is less sophisticated than that used in igb_update_itr,
4072 * due to the difficulty of synchronizing statistics across multiple
4073 * receive rings. The divisors and thresholds used by this function
4074 * were determined based on theoretical maximum wire speed and testing
4075 * data, in order to minimize response time while increasing bulk
4077 * This functionality is controlled by the InterruptThrottleRate module
4078 * parameter (see igb_param.c)
4079 * NOTE: This function is called only when operating in a multiqueue
4080 * receive environment.
4082 static void igb_update_ring_itr(struct igb_q_vector
*q_vector
)
4084 int new_val
= q_vector
->itr_val
;
4085 int avg_wire_size
= 0;
4086 struct igb_adapter
*adapter
= q_vector
->adapter
;
4087 unsigned int packets
;
4089 /* For non-gigabit speeds, just fix the interrupt rate at 4000
4090 * ints/sec - ITR timer value of 120 ticks.
4092 if (adapter
->link_speed
!= SPEED_1000
) {
4093 new_val
= IGB_4K_ITR
;
4097 packets
= q_vector
->rx
.total_packets
;
4099 avg_wire_size
= q_vector
->rx
.total_bytes
/ packets
;
4101 packets
= q_vector
->tx
.total_packets
;
4103 avg_wire_size
= max_t(u32
, avg_wire_size
,
4104 q_vector
->tx
.total_bytes
/ packets
);
4106 /* if avg_wire_size isn't set no work was done */
4110 /* Add 24 bytes to size to account for CRC, preamble, and gap */
4111 avg_wire_size
+= 24;
4113 /* Don't starve jumbo frames */
4114 avg_wire_size
= min(avg_wire_size
, 3000);
4116 /* Give a little boost to mid-size frames */
4117 if ((avg_wire_size
> 300) && (avg_wire_size
< 1200))
4118 new_val
= avg_wire_size
/ 3;
4120 new_val
= avg_wire_size
/ 2;
4122 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4123 if (new_val
< IGB_20K_ITR
&&
4124 ((q_vector
->rx
.ring
&& adapter
->rx_itr_setting
== 3) ||
4125 (!q_vector
->rx
.ring
&& adapter
->tx_itr_setting
== 3)))
4126 new_val
= IGB_20K_ITR
;
4129 if (new_val
!= q_vector
->itr_val
) {
4130 q_vector
->itr_val
= new_val
;
4131 q_vector
->set_itr
= 1;
4134 q_vector
->rx
.total_bytes
= 0;
4135 q_vector
->rx
.total_packets
= 0;
4136 q_vector
->tx
.total_bytes
= 0;
4137 q_vector
->tx
.total_packets
= 0;
4141 * igb_update_itr - update the dynamic ITR value based on statistics
4142 * @q_vector: pointer to q_vector
4143 * @ring_container: ring info to update the itr for
4145 * Stores a new ITR value based on packets and byte
4146 * counts during the last interrupt. The advantage of per interrupt
4147 * computation is faster updates and more accurate ITR for the current
4148 * traffic pattern. Constants in this function were computed
4149 * based on theoretical maximum wire speed and thresholds were set based
4150 * on testing data as well as attempting to minimize response time
4151 * while increasing bulk throughput.
4152 * this functionality is controlled by the InterruptThrottleRate module
4153 * parameter (see igb_param.c)
4154 * NOTE: These calculations are only valid when operating in a single-
4155 * queue environment.
4157 static void igb_update_itr(struct igb_q_vector
*q_vector
,
4158 struct igb_ring_container
*ring_container
)
4160 unsigned int packets
= ring_container
->total_packets
;
4161 unsigned int bytes
= ring_container
->total_bytes
;
4162 u8 itrval
= ring_container
->itr
;
4164 /* no packets, exit with status unchanged */
4169 case lowest_latency
:
4170 /* handle TSO and jumbo frames */
4171 if (bytes
/packets
> 8000)
4172 itrval
= bulk_latency
;
4173 else if ((packets
< 5) && (bytes
> 512))
4174 itrval
= low_latency
;
4176 case low_latency
: /* 50 usec aka 20000 ints/s */
4177 if (bytes
> 10000) {
4178 /* this if handles the TSO accounting */
4179 if (bytes
/packets
> 8000) {
4180 itrval
= bulk_latency
;
4181 } else if ((packets
< 10) || ((bytes
/packets
) > 1200)) {
4182 itrval
= bulk_latency
;
4183 } else if ((packets
> 35)) {
4184 itrval
= lowest_latency
;
4186 } else if (bytes
/packets
> 2000) {
4187 itrval
= bulk_latency
;
4188 } else if (packets
<= 2 && bytes
< 512) {
4189 itrval
= lowest_latency
;
4192 case bulk_latency
: /* 250 usec aka 4000 ints/s */
4193 if (bytes
> 25000) {
4195 itrval
= low_latency
;
4196 } else if (bytes
< 1500) {
4197 itrval
= low_latency
;
4202 /* clear work counters since we have the values we need */
4203 ring_container
->total_bytes
= 0;
4204 ring_container
->total_packets
= 0;
4206 /* write updated itr to ring container */
4207 ring_container
->itr
= itrval
;
4210 static void igb_set_itr(struct igb_q_vector
*q_vector
)
4212 struct igb_adapter
*adapter
= q_vector
->adapter
;
4213 u32 new_itr
= q_vector
->itr_val
;
4216 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4217 if (adapter
->link_speed
!= SPEED_1000
) {
4219 new_itr
= IGB_4K_ITR
;
4223 igb_update_itr(q_vector
, &q_vector
->tx
);
4224 igb_update_itr(q_vector
, &q_vector
->rx
);
4226 current_itr
= max(q_vector
->rx
.itr
, q_vector
->tx
.itr
);
4228 /* conservative mode (itr 3) eliminates the lowest_latency setting */
4229 if (current_itr
== lowest_latency
&&
4230 ((q_vector
->rx
.ring
&& adapter
->rx_itr_setting
== 3) ||
4231 (!q_vector
->rx
.ring
&& adapter
->tx_itr_setting
== 3)))
4232 current_itr
= low_latency
;
4234 switch (current_itr
) {
4235 /* counts and packets in update_itr are dependent on these numbers */
4236 case lowest_latency
:
4237 new_itr
= IGB_70K_ITR
; /* 70,000 ints/sec */
4240 new_itr
= IGB_20K_ITR
; /* 20,000 ints/sec */
4243 new_itr
= IGB_4K_ITR
; /* 4,000 ints/sec */
4250 if (new_itr
!= q_vector
->itr_val
) {
4251 /* this attempts to bias the interrupt rate towards Bulk
4252 * by adding intermediate steps when interrupt rate is
4255 new_itr
= new_itr
> q_vector
->itr_val
?
4256 max((new_itr
* q_vector
->itr_val
) /
4257 (new_itr
+ (q_vector
->itr_val
>> 2)),
4259 /* Don't write the value here; it resets the adapter's
4260 * internal timer, and causes us to delay far longer than
4261 * we should between interrupts. Instead, we write the ITR
4262 * value at the beginning of the next interrupt so the timing
4263 * ends up being correct.
4265 q_vector
->itr_val
= new_itr
;
4266 q_vector
->set_itr
= 1;
4270 static void igb_tx_ctxtdesc(struct igb_ring
*tx_ring
, u32 vlan_macip_lens
,
4271 u32 type_tucmd
, u32 mss_l4len_idx
)
4273 struct e1000_adv_tx_context_desc
*context_desc
;
4274 u16 i
= tx_ring
->next_to_use
;
4276 context_desc
= IGB_TX_CTXTDESC(tx_ring
, i
);
4279 tx_ring
->next_to_use
= (i
< tx_ring
->count
) ? i
: 0;
4281 /* set bits to identify this as an advanced context descriptor */
4282 type_tucmd
|= E1000_TXD_CMD_DEXT
| E1000_ADVTXD_DTYP_CTXT
;
4284 /* For 82575, context index must be unique per ring. */
4285 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX
, &tx_ring
->flags
))
4286 mss_l4len_idx
|= tx_ring
->reg_idx
<< 4;
4288 context_desc
->vlan_macip_lens
= cpu_to_le32(vlan_macip_lens
);
4289 context_desc
->seqnum_seed
= 0;
4290 context_desc
->type_tucmd_mlhl
= cpu_to_le32(type_tucmd
);
4291 context_desc
->mss_l4len_idx
= cpu_to_le32(mss_l4len_idx
);
4294 static int igb_tso(struct igb_ring
*tx_ring
,
4295 struct igb_tx_buffer
*first
,
4298 struct sk_buff
*skb
= first
->skb
;
4299 u32 vlan_macip_lens
, type_tucmd
;
4300 u32 mss_l4len_idx
, l4len
;
4302 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
4305 if (!skb_is_gso(skb
))
4308 if (skb_header_cloned(skb
)) {
4309 int err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
4314 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4315 type_tucmd
= E1000_ADVTXD_TUCMD_L4T_TCP
;
4317 if (first
->protocol
== __constant_htons(ETH_P_IP
)) {
4318 struct iphdr
*iph
= ip_hdr(skb
);
4321 tcp_hdr(skb
)->check
= ~csum_tcpudp_magic(iph
->saddr
,
4325 type_tucmd
|= E1000_ADVTXD_TUCMD_IPV4
;
4326 first
->tx_flags
|= IGB_TX_FLAGS_TSO
|
4329 } else if (skb_is_gso_v6(skb
)) {
4330 ipv6_hdr(skb
)->payload_len
= 0;
4331 tcp_hdr(skb
)->check
= ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
4332 &ipv6_hdr(skb
)->daddr
,
4334 first
->tx_flags
|= IGB_TX_FLAGS_TSO
|
4338 /* compute header lengths */
4339 l4len
= tcp_hdrlen(skb
);
4340 *hdr_len
= skb_transport_offset(skb
) + l4len
;
4342 /* update gso size and bytecount with header size */
4343 first
->gso_segs
= skb_shinfo(skb
)->gso_segs
;
4344 first
->bytecount
+= (first
->gso_segs
- 1) * *hdr_len
;
4347 mss_l4len_idx
= l4len
<< E1000_ADVTXD_L4LEN_SHIFT
;
4348 mss_l4len_idx
|= skb_shinfo(skb
)->gso_size
<< E1000_ADVTXD_MSS_SHIFT
;
4350 /* VLAN MACLEN IPLEN */
4351 vlan_macip_lens
= skb_network_header_len(skb
);
4352 vlan_macip_lens
|= skb_network_offset(skb
) << E1000_ADVTXD_MACLEN_SHIFT
;
4353 vlan_macip_lens
|= first
->tx_flags
& IGB_TX_FLAGS_VLAN_MASK
;
4355 igb_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, type_tucmd
, mss_l4len_idx
);
4360 static void igb_tx_csum(struct igb_ring
*tx_ring
, struct igb_tx_buffer
*first
)
4362 struct sk_buff
*skb
= first
->skb
;
4363 u32 vlan_macip_lens
= 0;
4364 u32 mss_l4len_idx
= 0;
4367 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
) {
4368 if (!(first
->tx_flags
& IGB_TX_FLAGS_VLAN
))
4372 switch (first
->protocol
) {
4373 case __constant_htons(ETH_P_IP
):
4374 vlan_macip_lens
|= skb_network_header_len(skb
);
4375 type_tucmd
|= E1000_ADVTXD_TUCMD_IPV4
;
4376 l4_hdr
= ip_hdr(skb
)->protocol
;
4378 case __constant_htons(ETH_P_IPV6
):
4379 vlan_macip_lens
|= skb_network_header_len(skb
);
4380 l4_hdr
= ipv6_hdr(skb
)->nexthdr
;
4383 if (unlikely(net_ratelimit())) {
4384 dev_warn(tx_ring
->dev
,
4385 "partial checksum but proto=%x!\n",
4393 type_tucmd
|= E1000_ADVTXD_TUCMD_L4T_TCP
;
4394 mss_l4len_idx
= tcp_hdrlen(skb
) <<
4395 E1000_ADVTXD_L4LEN_SHIFT
;
4398 type_tucmd
|= E1000_ADVTXD_TUCMD_L4T_SCTP
;
4399 mss_l4len_idx
= sizeof(struct sctphdr
) <<
4400 E1000_ADVTXD_L4LEN_SHIFT
;
4403 mss_l4len_idx
= sizeof(struct udphdr
) <<
4404 E1000_ADVTXD_L4LEN_SHIFT
;
4407 if (unlikely(net_ratelimit())) {
4408 dev_warn(tx_ring
->dev
,
4409 "partial checksum but l4 proto=%x!\n",
4415 /* update TX checksum flag */
4416 first
->tx_flags
|= IGB_TX_FLAGS_CSUM
;
4419 vlan_macip_lens
|= skb_network_offset(skb
) << E1000_ADVTXD_MACLEN_SHIFT
;
4420 vlan_macip_lens
|= first
->tx_flags
& IGB_TX_FLAGS_VLAN_MASK
;
4422 igb_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, type_tucmd
, mss_l4len_idx
);
4425 #define IGB_SET_FLAG(_input, _flag, _result) \
4426 ((_flag <= _result) ? \
4427 ((u32)(_input & _flag) * (_result / _flag)) : \
4428 ((u32)(_input & _flag) / (_flag / _result)))
4430 static u32
igb_tx_cmd_type(struct sk_buff
*skb
, u32 tx_flags
)
4432 /* set type for advanced descriptor with frame checksum insertion */
4433 u32 cmd_type
= E1000_ADVTXD_DTYP_DATA
|
4434 E1000_ADVTXD_DCMD_DEXT
|
4435 E1000_ADVTXD_DCMD_IFCS
;
4437 /* set HW vlan bit if vlan is present */
4438 cmd_type
|= IGB_SET_FLAG(tx_flags
, IGB_TX_FLAGS_VLAN
,
4439 (E1000_ADVTXD_DCMD_VLE
));
4441 /* set segmentation bits for TSO */
4442 cmd_type
|= IGB_SET_FLAG(tx_flags
, IGB_TX_FLAGS_TSO
,
4443 (E1000_ADVTXD_DCMD_TSE
));
4445 /* set timestamp bit if present */
4446 cmd_type
|= IGB_SET_FLAG(tx_flags
, IGB_TX_FLAGS_TSTAMP
,
4447 (E1000_ADVTXD_MAC_TSTAMP
));
4449 /* insert frame checksum */
4450 cmd_type
^= IGB_SET_FLAG(skb
->no_fcs
, 1, E1000_ADVTXD_DCMD_IFCS
);
4455 static void igb_tx_olinfo_status(struct igb_ring
*tx_ring
,
4456 union e1000_adv_tx_desc
*tx_desc
,
4457 u32 tx_flags
, unsigned int paylen
)
4459 u32 olinfo_status
= paylen
<< E1000_ADVTXD_PAYLEN_SHIFT
;
4461 /* 82575 requires a unique index per ring */
4462 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX
, &tx_ring
->flags
))
4463 olinfo_status
|= tx_ring
->reg_idx
<< 4;
4465 /* insert L4 checksum */
4466 olinfo_status
|= IGB_SET_FLAG(tx_flags
,
4468 (E1000_TXD_POPTS_TXSM
<< 8));
4470 /* insert IPv4 checksum */
4471 olinfo_status
|= IGB_SET_FLAG(tx_flags
,
4473 (E1000_TXD_POPTS_IXSM
<< 8));
4475 tx_desc
->read
.olinfo_status
= cpu_to_le32(olinfo_status
);
4478 static void igb_tx_map(struct igb_ring
*tx_ring
,
4479 struct igb_tx_buffer
*first
,
4482 struct sk_buff
*skb
= first
->skb
;
4483 struct igb_tx_buffer
*tx_buffer
;
4484 union e1000_adv_tx_desc
*tx_desc
;
4485 struct skb_frag_struct
*frag
;
4487 unsigned int data_len
, size
;
4488 u32 tx_flags
= first
->tx_flags
;
4489 u32 cmd_type
= igb_tx_cmd_type(skb
, tx_flags
);
4490 u16 i
= tx_ring
->next_to_use
;
4492 tx_desc
= IGB_TX_DESC(tx_ring
, i
);
4494 igb_tx_olinfo_status(tx_ring
, tx_desc
, tx_flags
, skb
->len
- hdr_len
);
4496 size
= skb_headlen(skb
);
4497 data_len
= skb
->data_len
;
4499 dma
= dma_map_single(tx_ring
->dev
, skb
->data
, size
, DMA_TO_DEVICE
);
4503 for (frag
= &skb_shinfo(skb
)->frags
[0];; frag
++) {
4504 if (dma_mapping_error(tx_ring
->dev
, dma
))
4507 /* record length, and DMA address */
4508 dma_unmap_len_set(tx_buffer
, len
, size
);
4509 dma_unmap_addr_set(tx_buffer
, dma
, dma
);
4511 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
4513 while (unlikely(size
> IGB_MAX_DATA_PER_TXD
)) {
4514 tx_desc
->read
.cmd_type_len
=
4515 cpu_to_le32(cmd_type
^ IGB_MAX_DATA_PER_TXD
);
4519 if (i
== tx_ring
->count
) {
4520 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
4523 tx_desc
->read
.olinfo_status
= 0;
4525 dma
+= IGB_MAX_DATA_PER_TXD
;
4526 size
-= IGB_MAX_DATA_PER_TXD
;
4528 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
4531 if (likely(!data_len
))
4534 tx_desc
->read
.cmd_type_len
= cpu_to_le32(cmd_type
^ size
);
4538 if (i
== tx_ring
->count
) {
4539 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
4542 tx_desc
->read
.olinfo_status
= 0;
4544 size
= skb_frag_size(frag
);
4547 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0,
4548 size
, DMA_TO_DEVICE
);
4550 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
4553 /* write last descriptor with RS and EOP bits */
4554 cmd_type
|= size
| IGB_TXD_DCMD
;
4555 tx_desc
->read
.cmd_type_len
= cpu_to_le32(cmd_type
);
4557 netdev_tx_sent_queue(txring_txq(tx_ring
), first
->bytecount
);
4559 /* set the timestamp */
4560 first
->time_stamp
= jiffies
;
4562 /* Force memory writes to complete before letting h/w know there
4563 * are new descriptors to fetch. (Only applicable for weak-ordered
4564 * memory model archs, such as IA-64).
4566 * We also need this memory barrier to make certain all of the
4567 * status bits have been updated before next_to_watch is written.
4571 /* set next_to_watch value indicating a packet is present */
4572 first
->next_to_watch
= tx_desc
;
4575 if (i
== tx_ring
->count
)
4578 tx_ring
->next_to_use
= i
;
4580 writel(i
, tx_ring
->tail
);
4582 /* we need this if more than one processor can write to our tail
4583 * at a time, it synchronizes IO on IA64/Altix systems
4590 dev_err(tx_ring
->dev
, "TX DMA map failed\n");
4592 /* clear dma mappings for failed tx_buffer_info map */
4594 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
4595 igb_unmap_and_free_tx_resource(tx_ring
, tx_buffer
);
4596 if (tx_buffer
== first
)
4603 tx_ring
->next_to_use
= i
;
4606 static int __igb_maybe_stop_tx(struct igb_ring
*tx_ring
, const u16 size
)
4608 struct net_device
*netdev
= tx_ring
->netdev
;
4610 netif_stop_subqueue(netdev
, tx_ring
->queue_index
);
4612 /* Herbert's original patch had:
4613 * smp_mb__after_netif_stop_queue();
4614 * but since that doesn't exist yet, just open code it.
4618 /* We need to check again in a case another CPU has just
4619 * made room available.
4621 if (igb_desc_unused(tx_ring
) < size
)
4625 netif_wake_subqueue(netdev
, tx_ring
->queue_index
);
4627 u64_stats_update_begin(&tx_ring
->tx_syncp2
);
4628 tx_ring
->tx_stats
.restart_queue2
++;
4629 u64_stats_update_end(&tx_ring
->tx_syncp2
);
4634 static inline int igb_maybe_stop_tx(struct igb_ring
*tx_ring
, const u16 size
)
4636 if (igb_desc_unused(tx_ring
) >= size
)
4638 return __igb_maybe_stop_tx(tx_ring
, size
);
4641 netdev_tx_t
igb_xmit_frame_ring(struct sk_buff
*skb
,
4642 struct igb_ring
*tx_ring
)
4644 struct igb_tx_buffer
*first
;
4647 u16 count
= TXD_USE_COUNT(skb_headlen(skb
));
4648 __be16 protocol
= vlan_get_protocol(skb
);
4651 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
4652 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
4653 * + 2 desc gap to keep tail from touching head,
4654 * + 1 desc for context descriptor,
4655 * otherwise try next time
4657 if (NETDEV_FRAG_PAGE_MAX_SIZE
> IGB_MAX_DATA_PER_TXD
) {
4659 for (f
= 0; f
< skb_shinfo(skb
)->nr_frags
; f
++)
4660 count
+= TXD_USE_COUNT(skb_shinfo(skb
)->frags
[f
].size
);
4662 count
+= skb_shinfo(skb
)->nr_frags
;
4665 if (igb_maybe_stop_tx(tx_ring
, count
+ 3)) {
4666 /* this is a hard error */
4667 return NETDEV_TX_BUSY
;
4670 /* record the location of the first descriptor for this packet */
4671 first
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_use
];
4673 first
->bytecount
= skb
->len
;
4674 first
->gso_segs
= 1;
4676 skb_tx_timestamp(skb
);
4678 if (unlikely(skb_shinfo(skb
)->tx_flags
& SKBTX_HW_TSTAMP
)) {
4679 struct igb_adapter
*adapter
= netdev_priv(tx_ring
->netdev
);
4681 if (!(adapter
->ptp_tx_skb
)) {
4682 skb_shinfo(skb
)->tx_flags
|= SKBTX_IN_PROGRESS
;
4683 tx_flags
|= IGB_TX_FLAGS_TSTAMP
;
4685 adapter
->ptp_tx_skb
= skb_get(skb
);
4686 adapter
->ptp_tx_start
= jiffies
;
4687 if (adapter
->hw
.mac
.type
== e1000_82576
)
4688 schedule_work(&adapter
->ptp_tx_work
);
4692 if (vlan_tx_tag_present(skb
)) {
4693 tx_flags
|= IGB_TX_FLAGS_VLAN
;
4694 tx_flags
|= (vlan_tx_tag_get(skb
) << IGB_TX_FLAGS_VLAN_SHIFT
);
4697 /* record initial flags and protocol */
4698 first
->tx_flags
= tx_flags
;
4699 first
->protocol
= protocol
;
4701 tso
= igb_tso(tx_ring
, first
, &hdr_len
);
4705 igb_tx_csum(tx_ring
, first
);
4707 igb_tx_map(tx_ring
, first
, hdr_len
);
4709 /* Make sure there is space in the ring for the next send. */
4710 igb_maybe_stop_tx(tx_ring
, DESC_NEEDED
);
4712 return NETDEV_TX_OK
;
4715 igb_unmap_and_free_tx_resource(tx_ring
, first
);
4717 return NETDEV_TX_OK
;
4720 static inline struct igb_ring
*igb_tx_queue_mapping(struct igb_adapter
*adapter
,
4721 struct sk_buff
*skb
)
4723 unsigned int r_idx
= skb
->queue_mapping
;
4725 if (r_idx
>= adapter
->num_tx_queues
)
4726 r_idx
= r_idx
% adapter
->num_tx_queues
;
4728 return adapter
->tx_ring
[r_idx
];
4731 static netdev_tx_t
igb_xmit_frame(struct sk_buff
*skb
,
4732 struct net_device
*netdev
)
4734 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4736 if (test_bit(__IGB_DOWN
, &adapter
->state
)) {
4737 dev_kfree_skb_any(skb
);
4738 return NETDEV_TX_OK
;
4741 if (skb
->len
<= 0) {
4742 dev_kfree_skb_any(skb
);
4743 return NETDEV_TX_OK
;
4746 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
4747 * in order to meet this minimum size requirement.
4749 if (unlikely(skb
->len
< 17)) {
4750 if (skb_pad(skb
, 17 - skb
->len
))
4751 return NETDEV_TX_OK
;
4753 skb_set_tail_pointer(skb
, 17);
4756 return igb_xmit_frame_ring(skb
, igb_tx_queue_mapping(adapter
, skb
));
4760 * igb_tx_timeout - Respond to a Tx Hang
4761 * @netdev: network interface device structure
4763 static void igb_tx_timeout(struct net_device
*netdev
)
4765 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4766 struct e1000_hw
*hw
= &adapter
->hw
;
4768 /* Do the reset outside of interrupt context */
4769 adapter
->tx_timeout_count
++;
4771 if (hw
->mac
.type
>= e1000_82580
)
4772 hw
->dev_spec
._82575
.global_device_reset
= true;
4774 schedule_work(&adapter
->reset_task
);
4776 (adapter
->eims_enable_mask
& ~adapter
->eims_other
));
4779 static void igb_reset_task(struct work_struct
*work
)
4781 struct igb_adapter
*adapter
;
4782 adapter
= container_of(work
, struct igb_adapter
, reset_task
);
4785 netdev_err(adapter
->netdev
, "Reset adapter\n");
4786 igb_reinit_locked(adapter
);
4790 * igb_get_stats64 - Get System Network Statistics
4791 * @netdev: network interface device structure
4792 * @stats: rtnl_link_stats64 pointer
4794 static struct rtnl_link_stats64
*igb_get_stats64(struct net_device
*netdev
,
4795 struct rtnl_link_stats64
*stats
)
4797 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4799 spin_lock(&adapter
->stats64_lock
);
4800 igb_update_stats(adapter
, &adapter
->stats64
);
4801 memcpy(stats
, &adapter
->stats64
, sizeof(*stats
));
4802 spin_unlock(&adapter
->stats64_lock
);
4808 * igb_change_mtu - Change the Maximum Transfer Unit
4809 * @netdev: network interface device structure
4810 * @new_mtu: new value for maximum frame size
4812 * Returns 0 on success, negative on failure
4814 static int igb_change_mtu(struct net_device
*netdev
, int new_mtu
)
4816 struct igb_adapter
*adapter
= netdev_priv(netdev
);
4817 struct pci_dev
*pdev
= adapter
->pdev
;
4818 int max_frame
= new_mtu
+ ETH_HLEN
+ ETH_FCS_LEN
+ VLAN_HLEN
;
4820 if ((new_mtu
< 68) || (max_frame
> MAX_JUMBO_FRAME_SIZE
)) {
4821 dev_err(&pdev
->dev
, "Invalid MTU setting\n");
4825 #define MAX_STD_JUMBO_FRAME_SIZE 9238
4826 if (max_frame
> MAX_STD_JUMBO_FRAME_SIZE
) {
4827 dev_err(&pdev
->dev
, "MTU > 9216 not supported.\n");
4831 /* adjust max frame to be at least the size of a standard frame */
4832 if (max_frame
< (ETH_FRAME_LEN
+ ETH_FCS_LEN
))
4833 max_frame
= ETH_FRAME_LEN
+ ETH_FCS_LEN
;
4835 while (test_and_set_bit(__IGB_RESETTING
, &adapter
->state
))
4838 /* igb_down has a dependency on max_frame_size */
4839 adapter
->max_frame_size
= max_frame
;
4841 if (netif_running(netdev
))
4844 dev_info(&pdev
->dev
, "changing MTU from %d to %d\n",
4845 netdev
->mtu
, new_mtu
);
4846 netdev
->mtu
= new_mtu
;
4848 if (netif_running(netdev
))
4853 clear_bit(__IGB_RESETTING
, &adapter
->state
);
4859 * igb_update_stats - Update the board statistics counters
4860 * @adapter: board private structure
4862 void igb_update_stats(struct igb_adapter
*adapter
,
4863 struct rtnl_link_stats64
*net_stats
)
4865 struct e1000_hw
*hw
= &adapter
->hw
;
4866 struct pci_dev
*pdev
= adapter
->pdev
;
4872 u64 _bytes
, _packets
;
4874 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
4876 /* Prevent stats update while adapter is being reset, or if the pci
4877 * connection is down.
4879 if (adapter
->link_speed
== 0)
4881 if (pci_channel_offline(pdev
))
4888 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
4889 u32 rqdpc
= rd32(E1000_RQDPC(i
));
4890 struct igb_ring
*ring
= adapter
->rx_ring
[i
];
4893 ring
->rx_stats
.drops
+= rqdpc
;
4894 net_stats
->rx_fifo_errors
+= rqdpc
;
4898 start
= u64_stats_fetch_begin_bh(&ring
->rx_syncp
);
4899 _bytes
= ring
->rx_stats
.bytes
;
4900 _packets
= ring
->rx_stats
.packets
;
4901 } while (u64_stats_fetch_retry_bh(&ring
->rx_syncp
, start
));
4903 packets
+= _packets
;
4906 net_stats
->rx_bytes
= bytes
;
4907 net_stats
->rx_packets
= packets
;
4911 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
4912 struct igb_ring
*ring
= adapter
->tx_ring
[i
];
4914 start
= u64_stats_fetch_begin_bh(&ring
->tx_syncp
);
4915 _bytes
= ring
->tx_stats
.bytes
;
4916 _packets
= ring
->tx_stats
.packets
;
4917 } while (u64_stats_fetch_retry_bh(&ring
->tx_syncp
, start
));
4919 packets
+= _packets
;
4921 net_stats
->tx_bytes
= bytes
;
4922 net_stats
->tx_packets
= packets
;
4925 /* read stats registers */
4926 adapter
->stats
.crcerrs
+= rd32(E1000_CRCERRS
);
4927 adapter
->stats
.gprc
+= rd32(E1000_GPRC
);
4928 adapter
->stats
.gorc
+= rd32(E1000_GORCL
);
4929 rd32(E1000_GORCH
); /* clear GORCL */
4930 adapter
->stats
.bprc
+= rd32(E1000_BPRC
);
4931 adapter
->stats
.mprc
+= rd32(E1000_MPRC
);
4932 adapter
->stats
.roc
+= rd32(E1000_ROC
);
4934 adapter
->stats
.prc64
+= rd32(E1000_PRC64
);
4935 adapter
->stats
.prc127
+= rd32(E1000_PRC127
);
4936 adapter
->stats
.prc255
+= rd32(E1000_PRC255
);
4937 adapter
->stats
.prc511
+= rd32(E1000_PRC511
);
4938 adapter
->stats
.prc1023
+= rd32(E1000_PRC1023
);
4939 adapter
->stats
.prc1522
+= rd32(E1000_PRC1522
);
4940 adapter
->stats
.symerrs
+= rd32(E1000_SYMERRS
);
4941 adapter
->stats
.sec
+= rd32(E1000_SEC
);
4943 mpc
= rd32(E1000_MPC
);
4944 adapter
->stats
.mpc
+= mpc
;
4945 net_stats
->rx_fifo_errors
+= mpc
;
4946 adapter
->stats
.scc
+= rd32(E1000_SCC
);
4947 adapter
->stats
.ecol
+= rd32(E1000_ECOL
);
4948 adapter
->stats
.mcc
+= rd32(E1000_MCC
);
4949 adapter
->stats
.latecol
+= rd32(E1000_LATECOL
);
4950 adapter
->stats
.dc
+= rd32(E1000_DC
);
4951 adapter
->stats
.rlec
+= rd32(E1000_RLEC
);
4952 adapter
->stats
.xonrxc
+= rd32(E1000_XONRXC
);
4953 adapter
->stats
.xontxc
+= rd32(E1000_XONTXC
);
4954 adapter
->stats
.xoffrxc
+= rd32(E1000_XOFFRXC
);
4955 adapter
->stats
.xofftxc
+= rd32(E1000_XOFFTXC
);
4956 adapter
->stats
.fcruc
+= rd32(E1000_FCRUC
);
4957 adapter
->stats
.gptc
+= rd32(E1000_GPTC
);
4958 adapter
->stats
.gotc
+= rd32(E1000_GOTCL
);
4959 rd32(E1000_GOTCH
); /* clear GOTCL */
4960 adapter
->stats
.rnbc
+= rd32(E1000_RNBC
);
4961 adapter
->stats
.ruc
+= rd32(E1000_RUC
);
4962 adapter
->stats
.rfc
+= rd32(E1000_RFC
);
4963 adapter
->stats
.rjc
+= rd32(E1000_RJC
);
4964 adapter
->stats
.tor
+= rd32(E1000_TORH
);
4965 adapter
->stats
.tot
+= rd32(E1000_TOTH
);
4966 adapter
->stats
.tpr
+= rd32(E1000_TPR
);
4968 adapter
->stats
.ptc64
+= rd32(E1000_PTC64
);
4969 adapter
->stats
.ptc127
+= rd32(E1000_PTC127
);
4970 adapter
->stats
.ptc255
+= rd32(E1000_PTC255
);
4971 adapter
->stats
.ptc511
+= rd32(E1000_PTC511
);
4972 adapter
->stats
.ptc1023
+= rd32(E1000_PTC1023
);
4973 adapter
->stats
.ptc1522
+= rd32(E1000_PTC1522
);
4975 adapter
->stats
.mptc
+= rd32(E1000_MPTC
);
4976 adapter
->stats
.bptc
+= rd32(E1000_BPTC
);
4978 adapter
->stats
.tpt
+= rd32(E1000_TPT
);
4979 adapter
->stats
.colc
+= rd32(E1000_COLC
);
4981 adapter
->stats
.algnerrc
+= rd32(E1000_ALGNERRC
);
4982 /* read internal phy specific stats */
4983 reg
= rd32(E1000_CTRL_EXT
);
4984 if (!(reg
& E1000_CTRL_EXT_LINK_MODE_MASK
)) {
4985 adapter
->stats
.rxerrc
+= rd32(E1000_RXERRC
);
4987 /* this stat has invalid values on i210/i211 */
4988 if ((hw
->mac
.type
!= e1000_i210
) &&
4989 (hw
->mac
.type
!= e1000_i211
))
4990 adapter
->stats
.tncrs
+= rd32(E1000_TNCRS
);
4993 adapter
->stats
.tsctc
+= rd32(E1000_TSCTC
);
4994 adapter
->stats
.tsctfc
+= rd32(E1000_TSCTFC
);
4996 adapter
->stats
.iac
+= rd32(E1000_IAC
);
4997 adapter
->stats
.icrxoc
+= rd32(E1000_ICRXOC
);
4998 adapter
->stats
.icrxptc
+= rd32(E1000_ICRXPTC
);
4999 adapter
->stats
.icrxatc
+= rd32(E1000_ICRXATC
);
5000 adapter
->stats
.ictxptc
+= rd32(E1000_ICTXPTC
);
5001 adapter
->stats
.ictxatc
+= rd32(E1000_ICTXATC
);
5002 adapter
->stats
.ictxqec
+= rd32(E1000_ICTXQEC
);
5003 adapter
->stats
.ictxqmtc
+= rd32(E1000_ICTXQMTC
);
5004 adapter
->stats
.icrxdmtc
+= rd32(E1000_ICRXDMTC
);
5006 /* Fill out the OS statistics structure */
5007 net_stats
->multicast
= adapter
->stats
.mprc
;
5008 net_stats
->collisions
= adapter
->stats
.colc
;
5012 /* RLEC on some newer hardware can be incorrect so build
5013 * our own version based on RUC and ROC
5015 net_stats
->rx_errors
= adapter
->stats
.rxerrc
+
5016 adapter
->stats
.crcerrs
+ adapter
->stats
.algnerrc
+
5017 adapter
->stats
.ruc
+ adapter
->stats
.roc
+
5018 adapter
->stats
.cexterr
;
5019 net_stats
->rx_length_errors
= adapter
->stats
.ruc
+
5021 net_stats
->rx_crc_errors
= adapter
->stats
.crcerrs
;
5022 net_stats
->rx_frame_errors
= adapter
->stats
.algnerrc
;
5023 net_stats
->rx_missed_errors
= adapter
->stats
.mpc
;
5026 net_stats
->tx_errors
= adapter
->stats
.ecol
+
5027 adapter
->stats
.latecol
;
5028 net_stats
->tx_aborted_errors
= adapter
->stats
.ecol
;
5029 net_stats
->tx_window_errors
= adapter
->stats
.latecol
;
5030 net_stats
->tx_carrier_errors
= adapter
->stats
.tncrs
;
5032 /* Tx Dropped needs to be maintained elsewhere */
5035 if (hw
->phy
.media_type
== e1000_media_type_copper
) {
5036 if ((adapter
->link_speed
== SPEED_1000
) &&
5037 (!igb_read_phy_reg(hw
, PHY_1000T_STATUS
, &phy_tmp
))) {
5038 phy_tmp
&= PHY_IDLE_ERROR_COUNT_MASK
;
5039 adapter
->phy_stats
.idle_errors
+= phy_tmp
;
5043 /* Management Stats */
5044 adapter
->stats
.mgptc
+= rd32(E1000_MGTPTC
);
5045 adapter
->stats
.mgprc
+= rd32(E1000_MGTPRC
);
5046 adapter
->stats
.mgpdc
+= rd32(E1000_MGTPDC
);
5049 reg
= rd32(E1000_MANC
);
5050 if (reg
& E1000_MANC_EN_BMC2OS
) {
5051 adapter
->stats
.o2bgptc
+= rd32(E1000_O2BGPTC
);
5052 adapter
->stats
.o2bspc
+= rd32(E1000_O2BSPC
);
5053 adapter
->stats
.b2ospc
+= rd32(E1000_B2OSPC
);
5054 adapter
->stats
.b2ogprc
+= rd32(E1000_B2OGPRC
);
5058 static irqreturn_t
igb_msix_other(int irq
, void *data
)
5060 struct igb_adapter
*adapter
= data
;
5061 struct e1000_hw
*hw
= &adapter
->hw
;
5062 u32 icr
= rd32(E1000_ICR
);
5063 /* reading ICR causes bit 31 of EICR to be cleared */
5065 if (icr
& E1000_ICR_DRSTA
)
5066 schedule_work(&adapter
->reset_task
);
5068 if (icr
& E1000_ICR_DOUTSYNC
) {
5069 /* HW is reporting DMA is out of sync */
5070 adapter
->stats
.doosync
++;
5071 /* The DMA Out of Sync is also indication of a spoof event
5072 * in IOV mode. Check the Wrong VM Behavior register to
5073 * see if it is really a spoof event.
5075 igb_check_wvbr(adapter
);
5078 /* Check for a mailbox event */
5079 if (icr
& E1000_ICR_VMMB
)
5080 igb_msg_task(adapter
);
5082 if (icr
& E1000_ICR_LSC
) {
5083 hw
->mac
.get_link_status
= 1;
5084 /* guard against interrupt when we're going down */
5085 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
5086 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
5089 if (icr
& E1000_ICR_TS
) {
5090 u32 tsicr
= rd32(E1000_TSICR
);
5092 if (tsicr
& E1000_TSICR_TXTS
) {
5093 /* acknowledge the interrupt */
5094 wr32(E1000_TSICR
, E1000_TSICR_TXTS
);
5095 /* retrieve hardware timestamp */
5096 schedule_work(&adapter
->ptp_tx_work
);
5100 wr32(E1000_EIMS
, adapter
->eims_other
);
5105 static void igb_write_itr(struct igb_q_vector
*q_vector
)
5107 struct igb_adapter
*adapter
= q_vector
->adapter
;
5108 u32 itr_val
= q_vector
->itr_val
& 0x7FFC;
5110 if (!q_vector
->set_itr
)
5116 if (adapter
->hw
.mac
.type
== e1000_82575
)
5117 itr_val
|= itr_val
<< 16;
5119 itr_val
|= E1000_EITR_CNT_IGNR
;
5121 writel(itr_val
, q_vector
->itr_register
);
5122 q_vector
->set_itr
= 0;
5125 static irqreturn_t
igb_msix_ring(int irq
, void *data
)
5127 struct igb_q_vector
*q_vector
= data
;
5129 /* Write the ITR value calculated from the previous interrupt. */
5130 igb_write_itr(q_vector
);
5132 napi_schedule(&q_vector
->napi
);
5137 #ifdef CONFIG_IGB_DCA
5138 static void igb_update_tx_dca(struct igb_adapter
*adapter
,
5139 struct igb_ring
*tx_ring
,
5142 struct e1000_hw
*hw
= &adapter
->hw
;
5143 u32 txctrl
= dca3_get_tag(tx_ring
->dev
, cpu
);
5145 if (hw
->mac
.type
!= e1000_82575
)
5146 txctrl
<<= E1000_DCA_TXCTRL_CPUID_SHIFT
;
5148 /* We can enable relaxed ordering for reads, but not writes when
5149 * DCA is enabled. This is due to a known issue in some chipsets
5150 * which will cause the DCA tag to be cleared.
5152 txctrl
|= E1000_DCA_TXCTRL_DESC_RRO_EN
|
5153 E1000_DCA_TXCTRL_DATA_RRO_EN
|
5154 E1000_DCA_TXCTRL_DESC_DCA_EN
;
5156 wr32(E1000_DCA_TXCTRL(tx_ring
->reg_idx
), txctrl
);
5159 static void igb_update_rx_dca(struct igb_adapter
*adapter
,
5160 struct igb_ring
*rx_ring
,
5163 struct e1000_hw
*hw
= &adapter
->hw
;
5164 u32 rxctrl
= dca3_get_tag(&adapter
->pdev
->dev
, cpu
);
5166 if (hw
->mac
.type
!= e1000_82575
)
5167 rxctrl
<<= E1000_DCA_RXCTRL_CPUID_SHIFT
;
5169 /* We can enable relaxed ordering for reads, but not writes when
5170 * DCA is enabled. This is due to a known issue in some chipsets
5171 * which will cause the DCA tag to be cleared.
5173 rxctrl
|= E1000_DCA_RXCTRL_DESC_RRO_EN
|
5174 E1000_DCA_RXCTRL_DESC_DCA_EN
;
5176 wr32(E1000_DCA_RXCTRL(rx_ring
->reg_idx
), rxctrl
);
5179 static void igb_update_dca(struct igb_q_vector
*q_vector
)
5181 struct igb_adapter
*adapter
= q_vector
->adapter
;
5182 int cpu
= get_cpu();
5184 if (q_vector
->cpu
== cpu
)
5187 if (q_vector
->tx
.ring
)
5188 igb_update_tx_dca(adapter
, q_vector
->tx
.ring
, cpu
);
5190 if (q_vector
->rx
.ring
)
5191 igb_update_rx_dca(adapter
, q_vector
->rx
.ring
, cpu
);
5193 q_vector
->cpu
= cpu
;
5198 static void igb_setup_dca(struct igb_adapter
*adapter
)
5200 struct e1000_hw
*hw
= &adapter
->hw
;
5203 if (!(adapter
->flags
& IGB_FLAG_DCA_ENABLED
))
5206 /* Always use CB2 mode, difference is masked in the CB driver. */
5207 wr32(E1000_DCA_CTRL
, E1000_DCA_CTRL_DCA_MODE_CB2
);
5209 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
5210 adapter
->q_vector
[i
]->cpu
= -1;
5211 igb_update_dca(adapter
->q_vector
[i
]);
5215 static int __igb_notify_dca(struct device
*dev
, void *data
)
5217 struct net_device
*netdev
= dev_get_drvdata(dev
);
5218 struct igb_adapter
*adapter
= netdev_priv(netdev
);
5219 struct pci_dev
*pdev
= adapter
->pdev
;
5220 struct e1000_hw
*hw
= &adapter
->hw
;
5221 unsigned long event
= *(unsigned long *)data
;
5224 case DCA_PROVIDER_ADD
:
5225 /* if already enabled, don't do it again */
5226 if (adapter
->flags
& IGB_FLAG_DCA_ENABLED
)
5228 if (dca_add_requester(dev
) == 0) {
5229 adapter
->flags
|= IGB_FLAG_DCA_ENABLED
;
5230 dev_info(&pdev
->dev
, "DCA enabled\n");
5231 igb_setup_dca(adapter
);
5234 /* Fall Through since DCA is disabled. */
5235 case DCA_PROVIDER_REMOVE
:
5236 if (adapter
->flags
& IGB_FLAG_DCA_ENABLED
) {
5237 /* without this a class_device is left
5238 * hanging around in the sysfs model
5240 dca_remove_requester(dev
);
5241 dev_info(&pdev
->dev
, "DCA disabled\n");
5242 adapter
->flags
&= ~IGB_FLAG_DCA_ENABLED
;
5243 wr32(E1000_DCA_CTRL
, E1000_DCA_CTRL_DCA_MODE_DISABLE
);
5251 static int igb_notify_dca(struct notifier_block
*nb
, unsigned long event
,
5256 ret_val
= driver_for_each_device(&igb_driver
.driver
, NULL
, &event
,
5259 return ret_val
? NOTIFY_BAD
: NOTIFY_DONE
;
5261 #endif /* CONFIG_IGB_DCA */
5263 #ifdef CONFIG_PCI_IOV
5264 static int igb_vf_configure(struct igb_adapter
*adapter
, int vf
)
5266 unsigned char mac_addr
[ETH_ALEN
];
5268 eth_zero_addr(mac_addr
);
5269 igb_set_vf_mac(adapter
, vf
, mac_addr
);
5271 /* By default spoof check is enabled for all VFs */
5272 adapter
->vf_data
[vf
].spoofchk_enabled
= true;
5278 static void igb_ping_all_vfs(struct igb_adapter
*adapter
)
5280 struct e1000_hw
*hw
= &adapter
->hw
;
5284 for (i
= 0 ; i
< adapter
->vfs_allocated_count
; i
++) {
5285 ping
= E1000_PF_CONTROL_MSG
;
5286 if (adapter
->vf_data
[i
].flags
& IGB_VF_FLAG_CTS
)
5287 ping
|= E1000_VT_MSGTYPE_CTS
;
5288 igb_write_mbx(hw
, &ping
, 1, i
);
5292 static int igb_set_vf_promisc(struct igb_adapter
*adapter
, u32
*msgbuf
, u32 vf
)
5294 struct e1000_hw
*hw
= &adapter
->hw
;
5295 u32 vmolr
= rd32(E1000_VMOLR(vf
));
5296 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5298 vf_data
->flags
&= ~(IGB_VF_FLAG_UNI_PROMISC
|
5299 IGB_VF_FLAG_MULTI_PROMISC
);
5300 vmolr
&= ~(E1000_VMOLR_ROPE
| E1000_VMOLR_ROMPE
| E1000_VMOLR_MPME
);
5302 if (*msgbuf
& E1000_VF_SET_PROMISC_MULTICAST
) {
5303 vmolr
|= E1000_VMOLR_MPME
;
5304 vf_data
->flags
|= IGB_VF_FLAG_MULTI_PROMISC
;
5305 *msgbuf
&= ~E1000_VF_SET_PROMISC_MULTICAST
;
5307 /* if we have hashes and we are clearing a multicast promisc
5308 * flag we need to write the hashes to the MTA as this step
5309 * was previously skipped
5311 if (vf_data
->num_vf_mc_hashes
> 30) {
5312 vmolr
|= E1000_VMOLR_MPME
;
5313 } else if (vf_data
->num_vf_mc_hashes
) {
5315 vmolr
|= E1000_VMOLR_ROMPE
;
5316 for (j
= 0; j
< vf_data
->num_vf_mc_hashes
; j
++)
5317 igb_mta_set(hw
, vf_data
->vf_mc_hashes
[j
]);
5321 wr32(E1000_VMOLR(vf
), vmolr
);
5323 /* there are flags left unprocessed, likely not supported */
5324 if (*msgbuf
& E1000_VT_MSGINFO_MASK
)
5330 static int igb_set_vf_multicasts(struct igb_adapter
*adapter
,
5331 u32
*msgbuf
, u32 vf
)
5333 int n
= (msgbuf
[0] & E1000_VT_MSGINFO_MASK
) >> E1000_VT_MSGINFO_SHIFT
;
5334 u16
*hash_list
= (u16
*)&msgbuf
[1];
5335 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5338 /* salt away the number of multicast addresses assigned
5339 * to this VF for later use to restore when the PF multi cast
5342 vf_data
->num_vf_mc_hashes
= n
;
5344 /* only up to 30 hash values supported */
5348 /* store the hashes for later use */
5349 for (i
= 0; i
< n
; i
++)
5350 vf_data
->vf_mc_hashes
[i
] = hash_list
[i
];
5352 /* Flush and reset the mta with the new values */
5353 igb_set_rx_mode(adapter
->netdev
);
5358 static void igb_restore_vf_multicasts(struct igb_adapter
*adapter
)
5360 struct e1000_hw
*hw
= &adapter
->hw
;
5361 struct vf_data_storage
*vf_data
;
5364 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++) {
5365 u32 vmolr
= rd32(E1000_VMOLR(i
));
5366 vmolr
&= ~(E1000_VMOLR_ROMPE
| E1000_VMOLR_MPME
);
5368 vf_data
= &adapter
->vf_data
[i
];
5370 if ((vf_data
->num_vf_mc_hashes
> 30) ||
5371 (vf_data
->flags
& IGB_VF_FLAG_MULTI_PROMISC
)) {
5372 vmolr
|= E1000_VMOLR_MPME
;
5373 } else if (vf_data
->num_vf_mc_hashes
) {
5374 vmolr
|= E1000_VMOLR_ROMPE
;
5375 for (j
= 0; j
< vf_data
->num_vf_mc_hashes
; j
++)
5376 igb_mta_set(hw
, vf_data
->vf_mc_hashes
[j
]);
5378 wr32(E1000_VMOLR(i
), vmolr
);
5382 static void igb_clear_vf_vfta(struct igb_adapter
*adapter
, u32 vf
)
5384 struct e1000_hw
*hw
= &adapter
->hw
;
5385 u32 pool_mask
, reg
, vid
;
5388 pool_mask
= 1 << (E1000_VLVF_POOLSEL_SHIFT
+ vf
);
5390 /* Find the vlan filter for this id */
5391 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5392 reg
= rd32(E1000_VLVF(i
));
5394 /* remove the vf from the pool */
5397 /* if pool is empty then remove entry from vfta */
5398 if (!(reg
& E1000_VLVF_POOLSEL_MASK
) &&
5399 (reg
& E1000_VLVF_VLANID_ENABLE
)) {
5401 vid
= reg
& E1000_VLVF_VLANID_MASK
;
5402 igb_vfta_set(hw
, vid
, false);
5405 wr32(E1000_VLVF(i
), reg
);
5408 adapter
->vf_data
[vf
].vlans_enabled
= 0;
5411 static s32
igb_vlvf_set(struct igb_adapter
*adapter
, u32 vid
, bool add
, u32 vf
)
5413 struct e1000_hw
*hw
= &adapter
->hw
;
5416 /* The vlvf table only exists on 82576 hardware and newer */
5417 if (hw
->mac
.type
< e1000_82576
)
5420 /* we only need to do this if VMDq is enabled */
5421 if (!adapter
->vfs_allocated_count
)
5424 /* Find the vlan filter for this id */
5425 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5426 reg
= rd32(E1000_VLVF(i
));
5427 if ((reg
& E1000_VLVF_VLANID_ENABLE
) &&
5428 vid
== (reg
& E1000_VLVF_VLANID_MASK
))
5433 if (i
== E1000_VLVF_ARRAY_SIZE
) {
5434 /* Did not find a matching VLAN ID entry that was
5435 * enabled. Search for a free filter entry, i.e.
5436 * one without the enable bit set
5438 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5439 reg
= rd32(E1000_VLVF(i
));
5440 if (!(reg
& E1000_VLVF_VLANID_ENABLE
))
5444 if (i
< E1000_VLVF_ARRAY_SIZE
) {
5445 /* Found an enabled/available entry */
5446 reg
|= 1 << (E1000_VLVF_POOLSEL_SHIFT
+ vf
);
5448 /* if !enabled we need to set this up in vfta */
5449 if (!(reg
& E1000_VLVF_VLANID_ENABLE
)) {
5450 /* add VID to filter table */
5451 igb_vfta_set(hw
, vid
, true);
5452 reg
|= E1000_VLVF_VLANID_ENABLE
;
5454 reg
&= ~E1000_VLVF_VLANID_MASK
;
5456 wr32(E1000_VLVF(i
), reg
);
5458 /* do not modify RLPML for PF devices */
5459 if (vf
>= adapter
->vfs_allocated_count
)
5462 if (!adapter
->vf_data
[vf
].vlans_enabled
) {
5464 reg
= rd32(E1000_VMOLR(vf
));
5465 size
= reg
& E1000_VMOLR_RLPML_MASK
;
5467 reg
&= ~E1000_VMOLR_RLPML_MASK
;
5469 wr32(E1000_VMOLR(vf
), reg
);
5472 adapter
->vf_data
[vf
].vlans_enabled
++;
5475 if (i
< E1000_VLVF_ARRAY_SIZE
) {
5476 /* remove vf from the pool */
5477 reg
&= ~(1 << (E1000_VLVF_POOLSEL_SHIFT
+ vf
));
5478 /* if pool is empty then remove entry from vfta */
5479 if (!(reg
& E1000_VLVF_POOLSEL_MASK
)) {
5481 igb_vfta_set(hw
, vid
, false);
5483 wr32(E1000_VLVF(i
), reg
);
5485 /* do not modify RLPML for PF devices */
5486 if (vf
>= adapter
->vfs_allocated_count
)
5489 adapter
->vf_data
[vf
].vlans_enabled
--;
5490 if (!adapter
->vf_data
[vf
].vlans_enabled
) {
5492 reg
= rd32(E1000_VMOLR(vf
));
5493 size
= reg
& E1000_VMOLR_RLPML_MASK
;
5495 reg
&= ~E1000_VMOLR_RLPML_MASK
;
5497 wr32(E1000_VMOLR(vf
), reg
);
5504 static void igb_set_vmvir(struct igb_adapter
*adapter
, u32 vid
, u32 vf
)
5506 struct e1000_hw
*hw
= &adapter
->hw
;
5509 wr32(E1000_VMVIR(vf
), (vid
| E1000_VMVIR_VLANA_DEFAULT
));
5511 wr32(E1000_VMVIR(vf
), 0);
5514 static int igb_ndo_set_vf_vlan(struct net_device
*netdev
,
5515 int vf
, u16 vlan
, u8 qos
)
5518 struct igb_adapter
*adapter
= netdev_priv(netdev
);
5520 if ((vf
>= adapter
->vfs_allocated_count
) || (vlan
> 4095) || (qos
> 7))
5523 err
= igb_vlvf_set(adapter
, vlan
, !!vlan
, vf
);
5526 igb_set_vmvir(adapter
, vlan
| (qos
<< VLAN_PRIO_SHIFT
), vf
);
5527 igb_set_vmolr(adapter
, vf
, !vlan
);
5528 adapter
->vf_data
[vf
].pf_vlan
= vlan
;
5529 adapter
->vf_data
[vf
].pf_qos
= qos
;
5530 dev_info(&adapter
->pdev
->dev
,
5531 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan
, qos
, vf
);
5532 if (test_bit(__IGB_DOWN
, &adapter
->state
)) {
5533 dev_warn(&adapter
->pdev
->dev
,
5534 "The VF VLAN has been set, but the PF device is not up.\n");
5535 dev_warn(&adapter
->pdev
->dev
,
5536 "Bring the PF device up before attempting to use the VF device.\n");
5539 igb_vlvf_set(adapter
, adapter
->vf_data
[vf
].pf_vlan
,
5541 igb_set_vmvir(adapter
, vlan
, vf
);
5542 igb_set_vmolr(adapter
, vf
, true);
5543 adapter
->vf_data
[vf
].pf_vlan
= 0;
5544 adapter
->vf_data
[vf
].pf_qos
= 0;
5550 static int igb_find_vlvf_entry(struct igb_adapter
*adapter
, int vid
)
5552 struct e1000_hw
*hw
= &adapter
->hw
;
5556 /* Find the vlan filter for this id */
5557 for (i
= 0; i
< E1000_VLVF_ARRAY_SIZE
; i
++) {
5558 reg
= rd32(E1000_VLVF(i
));
5559 if ((reg
& E1000_VLVF_VLANID_ENABLE
) &&
5560 vid
== (reg
& E1000_VLVF_VLANID_MASK
))
5564 if (i
>= E1000_VLVF_ARRAY_SIZE
)
5570 static int igb_set_vf_vlan(struct igb_adapter
*adapter
, u32
*msgbuf
, u32 vf
)
5572 struct e1000_hw
*hw
= &adapter
->hw
;
5573 int add
= (msgbuf
[0] & E1000_VT_MSGINFO_MASK
) >> E1000_VT_MSGINFO_SHIFT
;
5574 int vid
= (msgbuf
[1] & E1000_VLVF_VLANID_MASK
);
5577 /* If in promiscuous mode we need to make sure the PF also has
5578 * the VLAN filter set.
5580 if (add
&& (adapter
->netdev
->flags
& IFF_PROMISC
))
5581 err
= igb_vlvf_set(adapter
, vid
, add
,
5582 adapter
->vfs_allocated_count
);
5586 err
= igb_vlvf_set(adapter
, vid
, add
, vf
);
5591 /* Go through all the checks to see if the VLAN filter should
5592 * be wiped completely.
5594 if (!add
&& (adapter
->netdev
->flags
& IFF_PROMISC
)) {
5597 int regndx
= igb_find_vlvf_entry(adapter
, vid
);
5600 /* See if any other pools are set for this VLAN filter
5601 * entry other than the PF.
5603 vlvf
= bits
= rd32(E1000_VLVF(regndx
));
5604 bits
&= 1 << (E1000_VLVF_POOLSEL_SHIFT
+
5605 adapter
->vfs_allocated_count
);
5606 /* If the filter was removed then ensure PF pool bit
5607 * is cleared if the PF only added itself to the pool
5608 * because the PF is in promiscuous mode.
5610 if ((vlvf
& VLAN_VID_MASK
) == vid
&&
5611 !test_bit(vid
, adapter
->active_vlans
) &&
5613 igb_vlvf_set(adapter
, vid
, add
,
5614 adapter
->vfs_allocated_count
);
5621 static inline void igb_vf_reset(struct igb_adapter
*adapter
, u32 vf
)
5623 /* clear flags - except flag that indicates PF has set the MAC */
5624 adapter
->vf_data
[vf
].flags
&= IGB_VF_FLAG_PF_SET_MAC
;
5625 adapter
->vf_data
[vf
].last_nack
= jiffies
;
5627 /* reset offloads to defaults */
5628 igb_set_vmolr(adapter
, vf
, true);
5630 /* reset vlans for device */
5631 igb_clear_vf_vfta(adapter
, vf
);
5632 if (adapter
->vf_data
[vf
].pf_vlan
)
5633 igb_ndo_set_vf_vlan(adapter
->netdev
, vf
,
5634 adapter
->vf_data
[vf
].pf_vlan
,
5635 adapter
->vf_data
[vf
].pf_qos
);
5637 igb_clear_vf_vfta(adapter
, vf
);
5639 /* reset multicast table array for vf */
5640 adapter
->vf_data
[vf
].num_vf_mc_hashes
= 0;
5642 /* Flush and reset the mta with the new values */
5643 igb_set_rx_mode(adapter
->netdev
);
5646 static void igb_vf_reset_event(struct igb_adapter
*adapter
, u32 vf
)
5648 unsigned char *vf_mac
= adapter
->vf_data
[vf
].vf_mac_addresses
;
5650 /* clear mac address as we were hotplug removed/added */
5651 if (!(adapter
->vf_data
[vf
].flags
& IGB_VF_FLAG_PF_SET_MAC
))
5652 eth_zero_addr(vf_mac
);
5654 /* process remaining reset events */
5655 igb_vf_reset(adapter
, vf
);
5658 static void igb_vf_reset_msg(struct igb_adapter
*adapter
, u32 vf
)
5660 struct e1000_hw
*hw
= &adapter
->hw
;
5661 unsigned char *vf_mac
= adapter
->vf_data
[vf
].vf_mac_addresses
;
5662 int rar_entry
= hw
->mac
.rar_entry_count
- (vf
+ 1);
5664 u8
*addr
= (u8
*)(&msgbuf
[1]);
5666 /* process all the same items cleared in a function level reset */
5667 igb_vf_reset(adapter
, vf
);
5669 /* set vf mac address */
5670 igb_rar_set_qsel(adapter
, vf_mac
, rar_entry
, vf
);
5672 /* enable transmit and receive for vf */
5673 reg
= rd32(E1000_VFTE
);
5674 wr32(E1000_VFTE
, reg
| (1 << vf
));
5675 reg
= rd32(E1000_VFRE
);
5676 wr32(E1000_VFRE
, reg
| (1 << vf
));
5678 adapter
->vf_data
[vf
].flags
|= IGB_VF_FLAG_CTS
;
5680 /* reply to reset with ack and vf mac address */
5681 msgbuf
[0] = E1000_VF_RESET
| E1000_VT_MSGTYPE_ACK
;
5682 memcpy(addr
, vf_mac
, 6);
5683 igb_write_mbx(hw
, msgbuf
, 3, vf
);
5686 static int igb_set_vf_mac_addr(struct igb_adapter
*adapter
, u32
*msg
, int vf
)
5688 /* The VF MAC Address is stored in a packed array of bytes
5689 * starting at the second 32 bit word of the msg array
5691 unsigned char *addr
= (char *)&msg
[1];
5694 if (is_valid_ether_addr(addr
))
5695 err
= igb_set_vf_mac(adapter
, vf
, addr
);
5700 static void igb_rcv_ack_from_vf(struct igb_adapter
*adapter
, u32 vf
)
5702 struct e1000_hw
*hw
= &adapter
->hw
;
5703 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5704 u32 msg
= E1000_VT_MSGTYPE_NACK
;
5706 /* if device isn't clear to send it shouldn't be reading either */
5707 if (!(vf_data
->flags
& IGB_VF_FLAG_CTS
) &&
5708 time_after(jiffies
, vf_data
->last_nack
+ (2 * HZ
))) {
5709 igb_write_mbx(hw
, &msg
, 1, vf
);
5710 vf_data
->last_nack
= jiffies
;
5714 static void igb_rcv_msg_from_vf(struct igb_adapter
*adapter
, u32 vf
)
5716 struct pci_dev
*pdev
= adapter
->pdev
;
5717 u32 msgbuf
[E1000_VFMAILBOX_SIZE
];
5718 struct e1000_hw
*hw
= &adapter
->hw
;
5719 struct vf_data_storage
*vf_data
= &adapter
->vf_data
[vf
];
5722 retval
= igb_read_mbx(hw
, msgbuf
, E1000_VFMAILBOX_SIZE
, vf
);
5725 /* if receive failed revoke VF CTS stats and restart init */
5726 dev_err(&pdev
->dev
, "Error receiving message from VF\n");
5727 vf_data
->flags
&= ~IGB_VF_FLAG_CTS
;
5728 if (!time_after(jiffies
, vf_data
->last_nack
+ (2 * HZ
)))
5733 /* this is a message we already processed, do nothing */
5734 if (msgbuf
[0] & (E1000_VT_MSGTYPE_ACK
| E1000_VT_MSGTYPE_NACK
))
5737 /* until the vf completes a reset it should not be
5738 * allowed to start any configuration.
5740 if (msgbuf
[0] == E1000_VF_RESET
) {
5741 igb_vf_reset_msg(adapter
, vf
);
5745 if (!(vf_data
->flags
& IGB_VF_FLAG_CTS
)) {
5746 if (!time_after(jiffies
, vf_data
->last_nack
+ (2 * HZ
)))
5752 switch ((msgbuf
[0] & 0xFFFF)) {
5753 case E1000_VF_SET_MAC_ADDR
:
5755 if (!(vf_data
->flags
& IGB_VF_FLAG_PF_SET_MAC
))
5756 retval
= igb_set_vf_mac_addr(adapter
, msgbuf
, vf
);
5758 dev_warn(&pdev
->dev
,
5759 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
5762 case E1000_VF_SET_PROMISC
:
5763 retval
= igb_set_vf_promisc(adapter
, msgbuf
, vf
);
5765 case E1000_VF_SET_MULTICAST
:
5766 retval
= igb_set_vf_multicasts(adapter
, msgbuf
, vf
);
5768 case E1000_VF_SET_LPE
:
5769 retval
= igb_set_vf_rlpml(adapter
, msgbuf
[1], vf
);
5771 case E1000_VF_SET_VLAN
:
5773 if (vf_data
->pf_vlan
)
5774 dev_warn(&pdev
->dev
,
5775 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
5778 retval
= igb_set_vf_vlan(adapter
, msgbuf
, vf
);
5781 dev_err(&pdev
->dev
, "Unhandled Msg %08x\n", msgbuf
[0]);
5786 msgbuf
[0] |= E1000_VT_MSGTYPE_CTS
;
5788 /* notify the VF of the results of what it sent us */
5790 msgbuf
[0] |= E1000_VT_MSGTYPE_NACK
;
5792 msgbuf
[0] |= E1000_VT_MSGTYPE_ACK
;
5794 igb_write_mbx(hw
, msgbuf
, 1, vf
);
5797 static void igb_msg_task(struct igb_adapter
*adapter
)
5799 struct e1000_hw
*hw
= &adapter
->hw
;
5802 for (vf
= 0; vf
< adapter
->vfs_allocated_count
; vf
++) {
5803 /* process any reset requests */
5804 if (!igb_check_for_rst(hw
, vf
))
5805 igb_vf_reset_event(adapter
, vf
);
5807 /* process any messages pending */
5808 if (!igb_check_for_msg(hw
, vf
))
5809 igb_rcv_msg_from_vf(adapter
, vf
);
5811 /* process any acks */
5812 if (!igb_check_for_ack(hw
, vf
))
5813 igb_rcv_ack_from_vf(adapter
, vf
);
5818 * igb_set_uta - Set unicast filter table address
5819 * @adapter: board private structure
5821 * The unicast table address is a register array of 32-bit registers.
5822 * The table is meant to be used in a way similar to how the MTA is used
5823 * however due to certain limitations in the hardware it is necessary to
5824 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
5825 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
5827 static void igb_set_uta(struct igb_adapter
*adapter
)
5829 struct e1000_hw
*hw
= &adapter
->hw
;
5832 /* The UTA table only exists on 82576 hardware and newer */
5833 if (hw
->mac
.type
< e1000_82576
)
5836 /* we only need to do this if VMDq is enabled */
5837 if (!adapter
->vfs_allocated_count
)
5840 for (i
= 0; i
< hw
->mac
.uta_reg_count
; i
++)
5841 array_wr32(E1000_UTA
, i
, ~0);
5845 * igb_intr_msi - Interrupt Handler
5846 * @irq: interrupt number
5847 * @data: pointer to a network interface device structure
5849 static irqreturn_t
igb_intr_msi(int irq
, void *data
)
5851 struct igb_adapter
*adapter
= data
;
5852 struct igb_q_vector
*q_vector
= adapter
->q_vector
[0];
5853 struct e1000_hw
*hw
= &adapter
->hw
;
5854 /* read ICR disables interrupts using IAM */
5855 u32 icr
= rd32(E1000_ICR
);
5857 igb_write_itr(q_vector
);
5859 if (icr
& E1000_ICR_DRSTA
)
5860 schedule_work(&adapter
->reset_task
);
5862 if (icr
& E1000_ICR_DOUTSYNC
) {
5863 /* HW is reporting DMA is out of sync */
5864 adapter
->stats
.doosync
++;
5867 if (icr
& (E1000_ICR_RXSEQ
| E1000_ICR_LSC
)) {
5868 hw
->mac
.get_link_status
= 1;
5869 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
5870 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
5873 if (icr
& E1000_ICR_TS
) {
5874 u32 tsicr
= rd32(E1000_TSICR
);
5876 if (tsicr
& E1000_TSICR_TXTS
) {
5877 /* acknowledge the interrupt */
5878 wr32(E1000_TSICR
, E1000_TSICR_TXTS
);
5879 /* retrieve hardware timestamp */
5880 schedule_work(&adapter
->ptp_tx_work
);
5884 napi_schedule(&q_vector
->napi
);
5890 * igb_intr - Legacy Interrupt Handler
5891 * @irq: interrupt number
5892 * @data: pointer to a network interface device structure
5894 static irqreturn_t
igb_intr(int irq
, void *data
)
5896 struct igb_adapter
*adapter
= data
;
5897 struct igb_q_vector
*q_vector
= adapter
->q_vector
[0];
5898 struct e1000_hw
*hw
= &adapter
->hw
;
5899 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
5900 * need for the IMC write
5902 u32 icr
= rd32(E1000_ICR
);
5904 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5905 * not set, then the adapter didn't send an interrupt
5907 if (!(icr
& E1000_ICR_INT_ASSERTED
))
5910 igb_write_itr(q_vector
);
5912 if (icr
& E1000_ICR_DRSTA
)
5913 schedule_work(&adapter
->reset_task
);
5915 if (icr
& E1000_ICR_DOUTSYNC
) {
5916 /* HW is reporting DMA is out of sync */
5917 adapter
->stats
.doosync
++;
5920 if (icr
& (E1000_ICR_RXSEQ
| E1000_ICR_LSC
)) {
5921 hw
->mac
.get_link_status
= 1;
5922 /* guard against interrupt when we're going down */
5923 if (!test_bit(__IGB_DOWN
, &adapter
->state
))
5924 mod_timer(&adapter
->watchdog_timer
, jiffies
+ 1);
5927 if (icr
& E1000_ICR_TS
) {
5928 u32 tsicr
= rd32(E1000_TSICR
);
5930 if (tsicr
& E1000_TSICR_TXTS
) {
5931 /* acknowledge the interrupt */
5932 wr32(E1000_TSICR
, E1000_TSICR_TXTS
);
5933 /* retrieve hardware timestamp */
5934 schedule_work(&adapter
->ptp_tx_work
);
5938 napi_schedule(&q_vector
->napi
);
5943 static void igb_ring_irq_enable(struct igb_q_vector
*q_vector
)
5945 struct igb_adapter
*adapter
= q_vector
->adapter
;
5946 struct e1000_hw
*hw
= &adapter
->hw
;
5948 if ((q_vector
->rx
.ring
&& (adapter
->rx_itr_setting
& 3)) ||
5949 (!q_vector
->rx
.ring
&& (adapter
->tx_itr_setting
& 3))) {
5950 if ((adapter
->num_q_vectors
== 1) && !adapter
->vf_data
)
5951 igb_set_itr(q_vector
);
5953 igb_update_ring_itr(q_vector
);
5956 if (!test_bit(__IGB_DOWN
, &adapter
->state
)) {
5957 if (adapter
->msix_entries
)
5958 wr32(E1000_EIMS
, q_vector
->eims_value
);
5960 igb_irq_enable(adapter
);
5965 * igb_poll - NAPI Rx polling callback
5966 * @napi: napi polling structure
5967 * @budget: count of how many packets we should handle
5969 static int igb_poll(struct napi_struct
*napi
, int budget
)
5971 struct igb_q_vector
*q_vector
= container_of(napi
,
5972 struct igb_q_vector
,
5974 bool clean_complete
= true;
5976 #ifdef CONFIG_IGB_DCA
5977 if (q_vector
->adapter
->flags
& IGB_FLAG_DCA_ENABLED
)
5978 igb_update_dca(q_vector
);
5980 if (q_vector
->tx
.ring
)
5981 clean_complete
= igb_clean_tx_irq(q_vector
);
5983 if (q_vector
->rx
.ring
)
5984 clean_complete
&= igb_clean_rx_irq(q_vector
, budget
);
5986 /* If all work not completed, return budget and keep polling */
5987 if (!clean_complete
)
5990 /* If not enough Rx work done, exit the polling mode */
5991 napi_complete(napi
);
5992 igb_ring_irq_enable(q_vector
);
5998 * igb_clean_tx_irq - Reclaim resources after transmit completes
5999 * @q_vector: pointer to q_vector containing needed info
6001 * returns true if ring is completely cleaned
6003 static bool igb_clean_tx_irq(struct igb_q_vector
*q_vector
)
6005 struct igb_adapter
*adapter
= q_vector
->adapter
;
6006 struct igb_ring
*tx_ring
= q_vector
->tx
.ring
;
6007 struct igb_tx_buffer
*tx_buffer
;
6008 union e1000_adv_tx_desc
*tx_desc
;
6009 unsigned int total_bytes
= 0, total_packets
= 0;
6010 unsigned int budget
= q_vector
->tx
.work_limit
;
6011 unsigned int i
= tx_ring
->next_to_clean
;
6013 if (test_bit(__IGB_DOWN
, &adapter
->state
))
6016 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
6017 tx_desc
= IGB_TX_DESC(tx_ring
, i
);
6018 i
-= tx_ring
->count
;
6021 union e1000_adv_tx_desc
*eop_desc
= tx_buffer
->next_to_watch
;
6023 /* if next_to_watch is not set then there is no work pending */
6027 /* prevent any other reads prior to eop_desc */
6028 read_barrier_depends();
6030 /* if DD is not set pending work has not been completed */
6031 if (!(eop_desc
->wb
.status
& cpu_to_le32(E1000_TXD_STAT_DD
)))
6034 /* clear next_to_watch to prevent false hangs */
6035 tx_buffer
->next_to_watch
= NULL
;
6037 /* update the statistics for this packet */
6038 total_bytes
+= tx_buffer
->bytecount
;
6039 total_packets
+= tx_buffer
->gso_segs
;
6042 dev_kfree_skb_any(tx_buffer
->skb
);
6044 /* unmap skb header data */
6045 dma_unmap_single(tx_ring
->dev
,
6046 dma_unmap_addr(tx_buffer
, dma
),
6047 dma_unmap_len(tx_buffer
, len
),
6050 /* clear tx_buffer data */
6051 tx_buffer
->skb
= NULL
;
6052 dma_unmap_len_set(tx_buffer
, len
, 0);
6054 /* clear last DMA location and unmap remaining buffers */
6055 while (tx_desc
!= eop_desc
) {
6060 i
-= tx_ring
->count
;
6061 tx_buffer
= tx_ring
->tx_buffer_info
;
6062 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
6065 /* unmap any remaining paged data */
6066 if (dma_unmap_len(tx_buffer
, len
)) {
6067 dma_unmap_page(tx_ring
->dev
,
6068 dma_unmap_addr(tx_buffer
, dma
),
6069 dma_unmap_len(tx_buffer
, len
),
6071 dma_unmap_len_set(tx_buffer
, len
, 0);
6075 /* move us one more past the eop_desc for start of next pkt */
6080 i
-= tx_ring
->count
;
6081 tx_buffer
= tx_ring
->tx_buffer_info
;
6082 tx_desc
= IGB_TX_DESC(tx_ring
, 0);
6085 /* issue prefetch for next Tx descriptor */
6088 /* update budget accounting */
6090 } while (likely(budget
));
6092 netdev_tx_completed_queue(txring_txq(tx_ring
),
6093 total_packets
, total_bytes
);
6094 i
+= tx_ring
->count
;
6095 tx_ring
->next_to_clean
= i
;
6096 u64_stats_update_begin(&tx_ring
->tx_syncp
);
6097 tx_ring
->tx_stats
.bytes
+= total_bytes
;
6098 tx_ring
->tx_stats
.packets
+= total_packets
;
6099 u64_stats_update_end(&tx_ring
->tx_syncp
);
6100 q_vector
->tx
.total_bytes
+= total_bytes
;
6101 q_vector
->tx
.total_packets
+= total_packets
;
6103 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG
, &tx_ring
->flags
)) {
6104 struct e1000_hw
*hw
= &adapter
->hw
;
6106 /* Detect a transmit hang in hardware, this serializes the
6107 * check with the clearing of time_stamp and movement of i
6109 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG
, &tx_ring
->flags
);
6110 if (tx_buffer
->next_to_watch
&&
6111 time_after(jiffies
, tx_buffer
->time_stamp
+
6112 (adapter
->tx_timeout_factor
* HZ
)) &&
6113 !(rd32(E1000_STATUS
) & E1000_STATUS_TXOFF
)) {
6115 /* detected Tx unit hang */
6116 dev_err(tx_ring
->dev
,
6117 "Detected Tx Unit Hang\n"
6121 " next_to_use <%x>\n"
6122 " next_to_clean <%x>\n"
6123 "buffer_info[next_to_clean]\n"
6124 " time_stamp <%lx>\n"
6125 " next_to_watch <%p>\n"
6127 " desc.status <%x>\n",
6128 tx_ring
->queue_index
,
6129 rd32(E1000_TDH(tx_ring
->reg_idx
)),
6130 readl(tx_ring
->tail
),
6131 tx_ring
->next_to_use
,
6132 tx_ring
->next_to_clean
,
6133 tx_buffer
->time_stamp
,
6134 tx_buffer
->next_to_watch
,
6136 tx_buffer
->next_to_watch
->wb
.status
);
6137 netif_stop_subqueue(tx_ring
->netdev
,
6138 tx_ring
->queue_index
);
6140 /* we are about to reset, no point in enabling stuff */
6145 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6146 if (unlikely(total_packets
&&
6147 netif_carrier_ok(tx_ring
->netdev
) &&
6148 igb_desc_unused(tx_ring
) >= TX_WAKE_THRESHOLD
)) {
6149 /* Make sure that anybody stopping the queue after this
6150 * sees the new next_to_clean.
6153 if (__netif_subqueue_stopped(tx_ring
->netdev
,
6154 tx_ring
->queue_index
) &&
6155 !(test_bit(__IGB_DOWN
, &adapter
->state
))) {
6156 netif_wake_subqueue(tx_ring
->netdev
,
6157 tx_ring
->queue_index
);
6159 u64_stats_update_begin(&tx_ring
->tx_syncp
);
6160 tx_ring
->tx_stats
.restart_queue
++;
6161 u64_stats_update_end(&tx_ring
->tx_syncp
);
6169 * igb_reuse_rx_page - page flip buffer and store it back on the ring
6170 * @rx_ring: rx descriptor ring to store buffers on
6171 * @old_buff: donor buffer to have page reused
6173 * Synchronizes page for reuse by the adapter
6175 static void igb_reuse_rx_page(struct igb_ring
*rx_ring
,
6176 struct igb_rx_buffer
*old_buff
)
6178 struct igb_rx_buffer
*new_buff
;
6179 u16 nta
= rx_ring
->next_to_alloc
;
6181 new_buff
= &rx_ring
->rx_buffer_info
[nta
];
6183 /* update, and store next to alloc */
6185 rx_ring
->next_to_alloc
= (nta
< rx_ring
->count
) ? nta
: 0;
6187 /* transfer page from old buffer to new buffer */
6188 memcpy(new_buff
, old_buff
, sizeof(struct igb_rx_buffer
));
6190 /* sync the buffer for use by the device */
6191 dma_sync_single_range_for_device(rx_ring
->dev
, old_buff
->dma
,
6192 old_buff
->page_offset
,
6197 static bool igb_can_reuse_rx_page(struct igb_rx_buffer
*rx_buffer
,
6199 unsigned int truesize
)
6201 /* avoid re-using remote pages */
6202 if (unlikely(page_to_nid(page
) != numa_node_id()))
6205 #if (PAGE_SIZE < 8192)
6206 /* if we are only owner of page we can reuse it */
6207 if (unlikely(page_count(page
) != 1))
6210 /* flip page offset to other buffer */
6211 rx_buffer
->page_offset
^= IGB_RX_BUFSZ
;
6213 /* since we are the only owner of the page and we need to
6214 * increment it, just set the value to 2 in order to avoid
6215 * an unnecessary locked operation
6217 atomic_set(&page
->_count
, 2);
6219 /* move offset up to the next cache line */
6220 rx_buffer
->page_offset
+= truesize
;
6222 if (rx_buffer
->page_offset
> (PAGE_SIZE
- IGB_RX_BUFSZ
))
6225 /* bump ref count on page before it is given to the stack */
6233 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6234 * @rx_ring: rx descriptor ring to transact packets on
6235 * @rx_buffer: buffer containing page to add
6236 * @rx_desc: descriptor containing length of buffer written by hardware
6237 * @skb: sk_buff to place the data into
6239 * This function will add the data contained in rx_buffer->page to the skb.
6240 * This is done either through a direct copy if the data in the buffer is
6241 * less than the skb header size, otherwise it will just attach the page as
6242 * a frag to the skb.
6244 * The function will then update the page offset if necessary and return
6245 * true if the buffer can be reused by the adapter.
6247 static bool igb_add_rx_frag(struct igb_ring
*rx_ring
,
6248 struct igb_rx_buffer
*rx_buffer
,
6249 union e1000_adv_rx_desc
*rx_desc
,
6250 struct sk_buff
*skb
)
6252 struct page
*page
= rx_buffer
->page
;
6253 unsigned int size
= le16_to_cpu(rx_desc
->wb
.upper
.length
);
6254 #if (PAGE_SIZE < 8192)
6255 unsigned int truesize
= IGB_RX_BUFSZ
;
6257 unsigned int truesize
= ALIGN(size
, L1_CACHE_BYTES
);
6260 if ((size
<= IGB_RX_HDR_LEN
) && !skb_is_nonlinear(skb
)) {
6261 unsigned char *va
= page_address(page
) + rx_buffer
->page_offset
;
6263 if (igb_test_staterr(rx_desc
, E1000_RXDADV_STAT_TSIP
)) {
6264 igb_ptp_rx_pktstamp(rx_ring
->q_vector
, va
, skb
);
6265 va
+= IGB_TS_HDR_LEN
;
6266 size
-= IGB_TS_HDR_LEN
;
6269 memcpy(__skb_put(skb
, size
), va
, ALIGN(size
, sizeof(long)));
6271 /* we can reuse buffer as-is, just make sure it is local */
6272 if (likely(page_to_nid(page
) == numa_node_id()))
6275 /* this page cannot be reused so discard it */
6280 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
,
6281 rx_buffer
->page_offset
, size
, truesize
);
6283 return igb_can_reuse_rx_page(rx_buffer
, page
, truesize
);
6286 static struct sk_buff
*igb_fetch_rx_buffer(struct igb_ring
*rx_ring
,
6287 union e1000_adv_rx_desc
*rx_desc
,
6288 struct sk_buff
*skb
)
6290 struct igb_rx_buffer
*rx_buffer
;
6293 rx_buffer
= &rx_ring
->rx_buffer_info
[rx_ring
->next_to_clean
];
6295 page
= rx_buffer
->page
;
6299 void *page_addr
= page_address(page
) +
6300 rx_buffer
->page_offset
;
6302 /* prefetch first cache line of first page */
6303 prefetch(page_addr
);
6304 #if L1_CACHE_BYTES < 128
6305 prefetch(page_addr
+ L1_CACHE_BYTES
);
6308 /* allocate a skb to store the frags */
6309 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
6311 if (unlikely(!skb
)) {
6312 rx_ring
->rx_stats
.alloc_failed
++;
6316 /* we will be copying header into skb->data in
6317 * pskb_may_pull so it is in our interest to prefetch
6318 * it now to avoid a possible cache miss
6320 prefetchw(skb
->data
);
6323 /* we are reusing so sync this buffer for CPU use */
6324 dma_sync_single_range_for_cpu(rx_ring
->dev
,
6326 rx_buffer
->page_offset
,
6330 /* pull page into skb */
6331 if (igb_add_rx_frag(rx_ring
, rx_buffer
, rx_desc
, skb
)) {
6332 /* hand second half of page back to the ring */
6333 igb_reuse_rx_page(rx_ring
, rx_buffer
);
6335 /* we are not reusing the buffer so unmap it */
6336 dma_unmap_page(rx_ring
->dev
, rx_buffer
->dma
,
6337 PAGE_SIZE
, DMA_FROM_DEVICE
);
6340 /* clear contents of rx_buffer */
6341 rx_buffer
->page
= NULL
;
6346 static inline void igb_rx_checksum(struct igb_ring
*ring
,
6347 union e1000_adv_rx_desc
*rx_desc
,
6348 struct sk_buff
*skb
)
6350 skb_checksum_none_assert(skb
);
6352 /* Ignore Checksum bit is set */
6353 if (igb_test_staterr(rx_desc
, E1000_RXD_STAT_IXSM
))
6356 /* Rx checksum disabled via ethtool */
6357 if (!(ring
->netdev
->features
& NETIF_F_RXCSUM
))
6360 /* TCP/UDP checksum error bit is set */
6361 if (igb_test_staterr(rx_desc
,
6362 E1000_RXDEXT_STATERR_TCPE
|
6363 E1000_RXDEXT_STATERR_IPE
)) {
6364 /* work around errata with sctp packets where the TCPE aka
6365 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6366 * packets, (aka let the stack check the crc32c)
6368 if (!((skb
->len
== 60) &&
6369 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM
, &ring
->flags
))) {
6370 u64_stats_update_begin(&ring
->rx_syncp
);
6371 ring
->rx_stats
.csum_err
++;
6372 u64_stats_update_end(&ring
->rx_syncp
);
6374 /* let the stack verify checksum errors */
6377 /* It must be a TCP or UDP packet with a valid checksum */
6378 if (igb_test_staterr(rx_desc
, E1000_RXD_STAT_TCPCS
|
6379 E1000_RXD_STAT_UDPCS
))
6380 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
6382 dev_dbg(ring
->dev
, "cksum success: bits %08X\n",
6383 le32_to_cpu(rx_desc
->wb
.upper
.status_error
));
6386 static inline void igb_rx_hash(struct igb_ring
*ring
,
6387 union e1000_adv_rx_desc
*rx_desc
,
6388 struct sk_buff
*skb
)
6390 if (ring
->netdev
->features
& NETIF_F_RXHASH
)
6391 skb
->rxhash
= le32_to_cpu(rx_desc
->wb
.lower
.hi_dword
.rss
);
6395 * igb_is_non_eop - process handling of non-EOP buffers
6396 * @rx_ring: Rx ring being processed
6397 * @rx_desc: Rx descriptor for current buffer
6398 * @skb: current socket buffer containing buffer in progress
6400 * This function updates next to clean. If the buffer is an EOP buffer
6401 * this function exits returning false, otherwise it will place the
6402 * sk_buff in the next buffer to be chained and return true indicating
6403 * that this is in fact a non-EOP buffer.
6405 static bool igb_is_non_eop(struct igb_ring
*rx_ring
,
6406 union e1000_adv_rx_desc
*rx_desc
)
6408 u32 ntc
= rx_ring
->next_to_clean
+ 1;
6410 /* fetch, update, and store next to clean */
6411 ntc
= (ntc
< rx_ring
->count
) ? ntc
: 0;
6412 rx_ring
->next_to_clean
= ntc
;
6414 prefetch(IGB_RX_DESC(rx_ring
, ntc
));
6416 if (likely(igb_test_staterr(rx_desc
, E1000_RXD_STAT_EOP
)))
6423 * igb_get_headlen - determine size of header for LRO/GRO
6424 * @data: pointer to the start of the headers
6425 * @max_len: total length of section to find headers in
6427 * This function is meant to determine the length of headers that will
6428 * be recognized by hardware for LRO, and GRO offloads. The main
6429 * motivation of doing this is to only perform one pull for IPv4 TCP
6430 * packets so that we can do basic things like calculating the gso_size
6431 * based on the average data per packet.
6433 static unsigned int igb_get_headlen(unsigned char *data
,
6434 unsigned int max_len
)
6437 unsigned char *network
;
6440 struct vlan_hdr
*vlan
;
6443 struct ipv6hdr
*ipv6
;
6446 u8 nexthdr
= 0; /* default to not TCP */
6449 /* this should never happen, but better safe than sorry */
6450 if (max_len
< ETH_HLEN
)
6453 /* initialize network frame pointer */
6456 /* set first protocol and move network header forward */
6457 protocol
= hdr
.eth
->h_proto
;
6458 hdr
.network
+= ETH_HLEN
;
6460 /* handle any vlan tag if present */
6461 if (protocol
== __constant_htons(ETH_P_8021Q
)) {
6462 if ((hdr
.network
- data
) > (max_len
- VLAN_HLEN
))
6465 protocol
= hdr
.vlan
->h_vlan_encapsulated_proto
;
6466 hdr
.network
+= VLAN_HLEN
;
6469 /* handle L3 protocols */
6470 if (protocol
== __constant_htons(ETH_P_IP
)) {
6471 if ((hdr
.network
- data
) > (max_len
- sizeof(struct iphdr
)))
6474 /* access ihl as a u8 to avoid unaligned access on ia64 */
6475 hlen
= (hdr
.network
[0] & 0x0F) << 2;
6477 /* verify hlen meets minimum size requirements */
6478 if (hlen
< sizeof(struct iphdr
))
6479 return hdr
.network
- data
;
6481 /* record next protocol if header is present */
6482 if (!(hdr
.ipv4
->frag_off
& htons(IP_OFFSET
)))
6483 nexthdr
= hdr
.ipv4
->protocol
;
6484 } else if (protocol
== __constant_htons(ETH_P_IPV6
)) {
6485 if ((hdr
.network
- data
) > (max_len
- sizeof(struct ipv6hdr
)))
6488 /* record next protocol */
6489 nexthdr
= hdr
.ipv6
->nexthdr
;
6490 hlen
= sizeof(struct ipv6hdr
);
6492 return hdr
.network
- data
;
6495 /* relocate pointer to start of L4 header */
6496 hdr
.network
+= hlen
;
6498 /* finally sort out TCP */
6499 if (nexthdr
== IPPROTO_TCP
) {
6500 if ((hdr
.network
- data
) > (max_len
- sizeof(struct tcphdr
)))
6503 /* access doff as a u8 to avoid unaligned access on ia64 */
6504 hlen
= (hdr
.network
[12] & 0xF0) >> 2;
6506 /* verify hlen meets minimum size requirements */
6507 if (hlen
< sizeof(struct tcphdr
))
6508 return hdr
.network
- data
;
6510 hdr
.network
+= hlen
;
6511 } else if (nexthdr
== IPPROTO_UDP
) {
6512 if ((hdr
.network
- data
) > (max_len
- sizeof(struct udphdr
)))
6515 hdr
.network
+= sizeof(struct udphdr
);
6518 /* If everything has gone correctly hdr.network should be the
6519 * data section of the packet and will be the end of the header.
6520 * If not then it probably represents the end of the last recognized
6523 if ((hdr
.network
- data
) < max_len
)
6524 return hdr
.network
- data
;
6530 * igb_pull_tail - igb specific version of skb_pull_tail
6531 * @rx_ring: rx descriptor ring packet is being transacted on
6532 * @rx_desc: pointer to the EOP Rx descriptor
6533 * @skb: pointer to current skb being adjusted
6535 * This function is an igb specific version of __pskb_pull_tail. The
6536 * main difference between this version and the original function is that
6537 * this function can make several assumptions about the state of things
6538 * that allow for significant optimizations versus the standard function.
6539 * As a result we can do things like drop a frag and maintain an accurate
6540 * truesize for the skb.
6542 static void igb_pull_tail(struct igb_ring
*rx_ring
,
6543 union e1000_adv_rx_desc
*rx_desc
,
6544 struct sk_buff
*skb
)
6546 struct skb_frag_struct
*frag
= &skb_shinfo(skb
)->frags
[0];
6548 unsigned int pull_len
;
6550 /* it is valid to use page_address instead of kmap since we are
6551 * working with pages allocated out of the lomem pool per
6552 * alloc_page(GFP_ATOMIC)
6554 va
= skb_frag_address(frag
);
6556 if (igb_test_staterr(rx_desc
, E1000_RXDADV_STAT_TSIP
)) {
6557 /* retrieve timestamp from buffer */
6558 igb_ptp_rx_pktstamp(rx_ring
->q_vector
, va
, skb
);
6560 /* update pointers to remove timestamp header */
6561 skb_frag_size_sub(frag
, IGB_TS_HDR_LEN
);
6562 frag
->page_offset
+= IGB_TS_HDR_LEN
;
6563 skb
->data_len
-= IGB_TS_HDR_LEN
;
6564 skb
->len
-= IGB_TS_HDR_LEN
;
6566 /* move va to start of packet data */
6567 va
+= IGB_TS_HDR_LEN
;
6570 /* we need the header to contain the greater of either ETH_HLEN or
6571 * 60 bytes if the skb->len is less than 60 for skb_pad.
6573 pull_len
= igb_get_headlen(va
, IGB_RX_HDR_LEN
);
6575 /* align pull length to size of long to optimize memcpy performance */
6576 skb_copy_to_linear_data(skb
, va
, ALIGN(pull_len
, sizeof(long)));
6578 /* update all of the pointers */
6579 skb_frag_size_sub(frag
, pull_len
);
6580 frag
->page_offset
+= pull_len
;
6581 skb
->data_len
-= pull_len
;
6582 skb
->tail
+= pull_len
;
6586 * igb_cleanup_headers - Correct corrupted or empty headers
6587 * @rx_ring: rx descriptor ring packet is being transacted on
6588 * @rx_desc: pointer to the EOP Rx descriptor
6589 * @skb: pointer to current skb being fixed
6591 * Address the case where we are pulling data in on pages only
6592 * and as such no data is present in the skb header.
6594 * In addition if skb is not at least 60 bytes we need to pad it so that
6595 * it is large enough to qualify as a valid Ethernet frame.
6597 * Returns true if an error was encountered and skb was freed.
6599 static bool igb_cleanup_headers(struct igb_ring
*rx_ring
,
6600 union e1000_adv_rx_desc
*rx_desc
,
6601 struct sk_buff
*skb
)
6603 if (unlikely((igb_test_staterr(rx_desc
,
6604 E1000_RXDEXT_ERR_FRAME_ERR_MASK
)))) {
6605 struct net_device
*netdev
= rx_ring
->netdev
;
6606 if (!(netdev
->features
& NETIF_F_RXALL
)) {
6607 dev_kfree_skb_any(skb
);
6612 /* place header in linear portion of buffer */
6613 if (skb_is_nonlinear(skb
))
6614 igb_pull_tail(rx_ring
, rx_desc
, skb
);
6616 /* if skb_pad returns an error the skb was freed */
6617 if (unlikely(skb
->len
< 60)) {
6618 int pad_len
= 60 - skb
->len
;
6620 if (skb_pad(skb
, pad_len
))
6622 __skb_put(skb
, pad_len
);
6629 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
6630 * @rx_ring: rx descriptor ring packet is being transacted on
6631 * @rx_desc: pointer to the EOP Rx descriptor
6632 * @skb: pointer to current skb being populated
6634 * This function checks the ring, descriptor, and packet information in
6635 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
6636 * other fields within the skb.
6638 static void igb_process_skb_fields(struct igb_ring
*rx_ring
,
6639 union e1000_adv_rx_desc
*rx_desc
,
6640 struct sk_buff
*skb
)
6642 struct net_device
*dev
= rx_ring
->netdev
;
6644 igb_rx_hash(rx_ring
, rx_desc
, skb
);
6646 igb_rx_checksum(rx_ring
, rx_desc
, skb
);
6648 igb_ptp_rx_hwtstamp(rx_ring
, rx_desc
, skb
);
6650 if ((dev
->features
& NETIF_F_HW_VLAN_CTAG_RX
) &&
6651 igb_test_staterr(rx_desc
, E1000_RXD_STAT_VP
)) {
6653 if (igb_test_staterr(rx_desc
, E1000_RXDEXT_STATERR_LB
) &&
6654 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP
, &rx_ring
->flags
))
6655 vid
= be16_to_cpu(rx_desc
->wb
.upper
.vlan
);
6657 vid
= le16_to_cpu(rx_desc
->wb
.upper
.vlan
);
6659 __vlan_hwaccel_put_tag(skb
, htons(ETH_P_8021Q
), vid
);
6662 skb_record_rx_queue(skb
, rx_ring
->queue_index
);
6664 skb
->protocol
= eth_type_trans(skb
, rx_ring
->netdev
);
6667 static bool igb_clean_rx_irq(struct igb_q_vector
*q_vector
, const int budget
)
6669 struct igb_ring
*rx_ring
= q_vector
->rx
.ring
;
6670 struct sk_buff
*skb
= rx_ring
->skb
;
6671 unsigned int total_bytes
= 0, total_packets
= 0;
6672 u16 cleaned_count
= igb_desc_unused(rx_ring
);
6675 union e1000_adv_rx_desc
*rx_desc
;
6677 /* return some buffers to hardware, one at a time is too slow */
6678 if (cleaned_count
>= IGB_RX_BUFFER_WRITE
) {
6679 igb_alloc_rx_buffers(rx_ring
, cleaned_count
);
6683 rx_desc
= IGB_RX_DESC(rx_ring
, rx_ring
->next_to_clean
);
6685 if (!igb_test_staterr(rx_desc
, E1000_RXD_STAT_DD
))
6688 /* This memory barrier is needed to keep us from reading
6689 * any other fields out of the rx_desc until we know the
6690 * RXD_STAT_DD bit is set
6694 /* retrieve a buffer from the ring */
6695 skb
= igb_fetch_rx_buffer(rx_ring
, rx_desc
, skb
);
6697 /* exit if we failed to retrieve a buffer */
6703 /* fetch next buffer in frame if non-eop */
6704 if (igb_is_non_eop(rx_ring
, rx_desc
))
6707 /* verify the packet layout is correct */
6708 if (igb_cleanup_headers(rx_ring
, rx_desc
, skb
)) {
6713 /* probably a little skewed due to removing CRC */
6714 total_bytes
+= skb
->len
;
6716 /* populate checksum, timestamp, VLAN, and protocol */
6717 igb_process_skb_fields(rx_ring
, rx_desc
, skb
);
6719 napi_gro_receive(&q_vector
->napi
, skb
);
6721 /* reset skb pointer */
6724 /* update budget accounting */
6726 } while (likely(total_packets
< budget
));
6728 /* place incomplete frames back on ring for completion */
6731 u64_stats_update_begin(&rx_ring
->rx_syncp
);
6732 rx_ring
->rx_stats
.packets
+= total_packets
;
6733 rx_ring
->rx_stats
.bytes
+= total_bytes
;
6734 u64_stats_update_end(&rx_ring
->rx_syncp
);
6735 q_vector
->rx
.total_packets
+= total_packets
;
6736 q_vector
->rx
.total_bytes
+= total_bytes
;
6739 igb_alloc_rx_buffers(rx_ring
, cleaned_count
);
6741 return (total_packets
< budget
);
6744 static bool igb_alloc_mapped_page(struct igb_ring
*rx_ring
,
6745 struct igb_rx_buffer
*bi
)
6747 struct page
*page
= bi
->page
;
6750 /* since we are recycling buffers we should seldom need to alloc */
6754 /* alloc new page for storage */
6755 page
= __skb_alloc_page(GFP_ATOMIC
| __GFP_COLD
, NULL
);
6756 if (unlikely(!page
)) {
6757 rx_ring
->rx_stats
.alloc_failed
++;
6761 /* map page for use */
6762 dma
= dma_map_page(rx_ring
->dev
, page
, 0, PAGE_SIZE
, DMA_FROM_DEVICE
);
6764 /* if mapping failed free memory back to system since
6765 * there isn't much point in holding memory we can't use
6767 if (dma_mapping_error(rx_ring
->dev
, dma
)) {
6770 rx_ring
->rx_stats
.alloc_failed
++;
6776 bi
->page_offset
= 0;
6782 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
6783 * @adapter: address of board private structure
6785 void igb_alloc_rx_buffers(struct igb_ring
*rx_ring
, u16 cleaned_count
)
6787 union e1000_adv_rx_desc
*rx_desc
;
6788 struct igb_rx_buffer
*bi
;
6789 u16 i
= rx_ring
->next_to_use
;
6795 rx_desc
= IGB_RX_DESC(rx_ring
, i
);
6796 bi
= &rx_ring
->rx_buffer_info
[i
];
6797 i
-= rx_ring
->count
;
6800 if (!igb_alloc_mapped_page(rx_ring
, bi
))
6803 /* Refresh the desc even if buffer_addrs didn't change
6804 * because each write-back erases this info.
6806 rx_desc
->read
.pkt_addr
= cpu_to_le64(bi
->dma
+ bi
->page_offset
);
6812 rx_desc
= IGB_RX_DESC(rx_ring
, 0);
6813 bi
= rx_ring
->rx_buffer_info
;
6814 i
-= rx_ring
->count
;
6817 /* clear the hdr_addr for the next_to_use descriptor */
6818 rx_desc
->read
.hdr_addr
= 0;
6821 } while (cleaned_count
);
6823 i
+= rx_ring
->count
;
6825 if (rx_ring
->next_to_use
!= i
) {
6826 /* record the next descriptor to use */
6827 rx_ring
->next_to_use
= i
;
6829 /* update next to alloc since we have filled the ring */
6830 rx_ring
->next_to_alloc
= i
;
6832 /* Force memory writes to complete before letting h/w
6833 * know there are new descriptors to fetch. (Only
6834 * applicable for weak-ordered memory model archs,
6838 writel(i
, rx_ring
->tail
);
6848 static int igb_mii_ioctl(struct net_device
*netdev
, struct ifreq
*ifr
, int cmd
)
6850 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6851 struct mii_ioctl_data
*data
= if_mii(ifr
);
6853 if (adapter
->hw
.phy
.media_type
!= e1000_media_type_copper
)
6858 data
->phy_id
= adapter
->hw
.phy
.addr
;
6861 if (igb_read_phy_reg(&adapter
->hw
, data
->reg_num
& 0x1F,
6878 static int igb_ioctl(struct net_device
*netdev
, struct ifreq
*ifr
, int cmd
)
6884 return igb_mii_ioctl(netdev
, ifr
, cmd
);
6886 return igb_ptp_hwtstamp_ioctl(netdev
, ifr
, cmd
);
6892 s32
igb_read_pcie_cap_reg(struct e1000_hw
*hw
, u32 reg
, u16
*value
)
6894 struct igb_adapter
*adapter
= hw
->back
;
6896 if (pcie_capability_read_word(adapter
->pdev
, reg
, value
))
6897 return -E1000_ERR_CONFIG
;
6902 s32
igb_write_pcie_cap_reg(struct e1000_hw
*hw
, u32 reg
, u16
*value
)
6904 struct igb_adapter
*adapter
= hw
->back
;
6906 if (pcie_capability_write_word(adapter
->pdev
, reg
, *value
))
6907 return -E1000_ERR_CONFIG
;
6912 static void igb_vlan_mode(struct net_device
*netdev
, netdev_features_t features
)
6914 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6915 struct e1000_hw
*hw
= &adapter
->hw
;
6917 bool enable
= !!(features
& NETIF_F_HW_VLAN_CTAG_RX
);
6920 /* enable VLAN tag insert/strip */
6921 ctrl
= rd32(E1000_CTRL
);
6922 ctrl
|= E1000_CTRL_VME
;
6923 wr32(E1000_CTRL
, ctrl
);
6925 /* Disable CFI check */
6926 rctl
= rd32(E1000_RCTL
);
6927 rctl
&= ~E1000_RCTL_CFIEN
;
6928 wr32(E1000_RCTL
, rctl
);
6930 /* disable VLAN tag insert/strip */
6931 ctrl
= rd32(E1000_CTRL
);
6932 ctrl
&= ~E1000_CTRL_VME
;
6933 wr32(E1000_CTRL
, ctrl
);
6936 igb_rlpml_set(adapter
);
6939 static int igb_vlan_rx_add_vid(struct net_device
*netdev
,
6940 __be16 proto
, u16 vid
)
6942 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6943 struct e1000_hw
*hw
= &adapter
->hw
;
6944 int pf_id
= adapter
->vfs_allocated_count
;
6946 /* attempt to add filter to vlvf array */
6947 igb_vlvf_set(adapter
, vid
, true, pf_id
);
6949 /* add the filter since PF can receive vlans w/o entry in vlvf */
6950 igb_vfta_set(hw
, vid
, true);
6952 set_bit(vid
, adapter
->active_vlans
);
6957 static int igb_vlan_rx_kill_vid(struct net_device
*netdev
,
6958 __be16 proto
, u16 vid
)
6960 struct igb_adapter
*adapter
= netdev_priv(netdev
);
6961 struct e1000_hw
*hw
= &adapter
->hw
;
6962 int pf_id
= adapter
->vfs_allocated_count
;
6965 /* remove vlan from VLVF table array */
6966 err
= igb_vlvf_set(adapter
, vid
, false, pf_id
);
6968 /* if vid was not present in VLVF just remove it from table */
6970 igb_vfta_set(hw
, vid
, false);
6972 clear_bit(vid
, adapter
->active_vlans
);
6977 static void igb_restore_vlan(struct igb_adapter
*adapter
)
6981 igb_vlan_mode(adapter
->netdev
, adapter
->netdev
->features
);
6983 for_each_set_bit(vid
, adapter
->active_vlans
, VLAN_N_VID
)
6984 igb_vlan_rx_add_vid(adapter
->netdev
, htons(ETH_P_8021Q
), vid
);
6987 int igb_set_spd_dplx(struct igb_adapter
*adapter
, u32 spd
, u8 dplx
)
6989 struct pci_dev
*pdev
= adapter
->pdev
;
6990 struct e1000_mac_info
*mac
= &adapter
->hw
.mac
;
6994 /* Make sure dplx is at most 1 bit and lsb of speed is not set
6995 * for the switch() below to work
6997 if ((spd
& 1) || (dplx
& ~1))
7000 /* Fiber NIC's only allow 1000 gbps Full duplex
7001 * and 100Mbps Full duplex for 100baseFx sfp
7003 if (adapter
->hw
.phy
.media_type
== e1000_media_type_internal_serdes
) {
7004 switch (spd
+ dplx
) {
7005 case SPEED_10
+ DUPLEX_HALF
:
7006 case SPEED_10
+ DUPLEX_FULL
:
7007 case SPEED_100
+ DUPLEX_HALF
:
7014 switch (spd
+ dplx
) {
7015 case SPEED_10
+ DUPLEX_HALF
:
7016 mac
->forced_speed_duplex
= ADVERTISE_10_HALF
;
7018 case SPEED_10
+ DUPLEX_FULL
:
7019 mac
->forced_speed_duplex
= ADVERTISE_10_FULL
;
7021 case SPEED_100
+ DUPLEX_HALF
:
7022 mac
->forced_speed_duplex
= ADVERTISE_100_HALF
;
7024 case SPEED_100
+ DUPLEX_FULL
:
7025 mac
->forced_speed_duplex
= ADVERTISE_100_FULL
;
7027 case SPEED_1000
+ DUPLEX_FULL
:
7029 adapter
->hw
.phy
.autoneg_advertised
= ADVERTISE_1000_FULL
;
7031 case SPEED_1000
+ DUPLEX_HALF
: /* not supported */
7036 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7037 adapter
->hw
.phy
.mdix
= AUTO_ALL_MODES
;
7042 dev_err(&pdev
->dev
, "Unsupported Speed/Duplex configuration\n");
7046 static int __igb_shutdown(struct pci_dev
*pdev
, bool *enable_wake
,
7049 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7050 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7051 struct e1000_hw
*hw
= &adapter
->hw
;
7052 u32 ctrl
, rctl
, status
;
7053 u32 wufc
= runtime
? E1000_WUFC_LNKC
: adapter
->wol
;
7058 netif_device_detach(netdev
);
7060 if (netif_running(netdev
))
7061 __igb_close(netdev
, true);
7063 igb_clear_interrupt_scheme(adapter
);
7066 retval
= pci_save_state(pdev
);
7071 status
= rd32(E1000_STATUS
);
7072 if (status
& E1000_STATUS_LU
)
7073 wufc
&= ~E1000_WUFC_LNKC
;
7076 igb_setup_rctl(adapter
);
7077 igb_set_rx_mode(netdev
);
7079 /* turn on all-multi mode if wake on multicast is enabled */
7080 if (wufc
& E1000_WUFC_MC
) {
7081 rctl
= rd32(E1000_RCTL
);
7082 rctl
|= E1000_RCTL_MPE
;
7083 wr32(E1000_RCTL
, rctl
);
7086 ctrl
= rd32(E1000_CTRL
);
7087 /* advertise wake from D3Cold */
7088 #define E1000_CTRL_ADVD3WUC 0x00100000
7089 /* phy power management enable */
7090 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7091 ctrl
|= E1000_CTRL_ADVD3WUC
;
7092 wr32(E1000_CTRL
, ctrl
);
7094 /* Allow time for pending master requests to run */
7095 igb_disable_pcie_master(hw
);
7097 wr32(E1000_WUC
, E1000_WUC_PME_EN
);
7098 wr32(E1000_WUFC
, wufc
);
7101 wr32(E1000_WUFC
, 0);
7104 *enable_wake
= wufc
|| adapter
->en_mng_pt
;
7106 igb_power_down_link(adapter
);
7108 igb_power_up_link(adapter
);
7110 /* Release control of h/w to f/w. If f/w is AMT enabled, this
7111 * would have already happened in close and is redundant.
7113 igb_release_hw_control(adapter
);
7115 pci_disable_device(pdev
);
7121 #ifdef CONFIG_PM_SLEEP
7122 static int igb_suspend(struct device
*dev
)
7126 struct pci_dev
*pdev
= to_pci_dev(dev
);
7128 retval
= __igb_shutdown(pdev
, &wake
, 0);
7133 pci_prepare_to_sleep(pdev
);
7135 pci_wake_from_d3(pdev
, false);
7136 pci_set_power_state(pdev
, PCI_D3hot
);
7141 #endif /* CONFIG_PM_SLEEP */
7143 static int igb_resume(struct device
*dev
)
7145 struct pci_dev
*pdev
= to_pci_dev(dev
);
7146 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7147 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7148 struct e1000_hw
*hw
= &adapter
->hw
;
7151 pci_set_power_state(pdev
, PCI_D0
);
7152 pci_restore_state(pdev
);
7153 pci_save_state(pdev
);
7155 err
= pci_enable_device_mem(pdev
);
7158 "igb: Cannot enable PCI device from suspend\n");
7161 pci_set_master(pdev
);
7163 pci_enable_wake(pdev
, PCI_D3hot
, 0);
7164 pci_enable_wake(pdev
, PCI_D3cold
, 0);
7166 if (igb_init_interrupt_scheme(adapter
, true)) {
7167 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
7173 /* let the f/w know that the h/w is now under the control of the
7176 igb_get_hw_control(adapter
);
7178 wr32(E1000_WUS
, ~0);
7180 if (netdev
->flags
& IFF_UP
) {
7182 err
= __igb_open(netdev
, true);
7188 netif_device_attach(netdev
);
7192 #ifdef CONFIG_PM_RUNTIME
7193 static int igb_runtime_idle(struct device
*dev
)
7195 struct pci_dev
*pdev
= to_pci_dev(dev
);
7196 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7197 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7199 if (!igb_has_link(adapter
))
7200 pm_schedule_suspend(dev
, MSEC_PER_SEC
* 5);
7205 static int igb_runtime_suspend(struct device
*dev
)
7207 struct pci_dev
*pdev
= to_pci_dev(dev
);
7211 retval
= __igb_shutdown(pdev
, &wake
, 1);
7216 pci_prepare_to_sleep(pdev
);
7218 pci_wake_from_d3(pdev
, false);
7219 pci_set_power_state(pdev
, PCI_D3hot
);
7225 static int igb_runtime_resume(struct device
*dev
)
7227 return igb_resume(dev
);
7229 #endif /* CONFIG_PM_RUNTIME */
7232 static void igb_shutdown(struct pci_dev
*pdev
)
7236 __igb_shutdown(pdev
, &wake
, 0);
7238 if (system_state
== SYSTEM_POWER_OFF
) {
7239 pci_wake_from_d3(pdev
, wake
);
7240 pci_set_power_state(pdev
, PCI_D3hot
);
7244 #ifdef CONFIG_PCI_IOV
7245 static int igb_sriov_reinit(struct pci_dev
*dev
)
7247 struct net_device
*netdev
= pci_get_drvdata(dev
);
7248 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7249 struct pci_dev
*pdev
= adapter
->pdev
;
7253 if (netif_running(netdev
))
7256 igb_clear_interrupt_scheme(adapter
);
7258 igb_init_queue_configuration(adapter
);
7260 if (igb_init_interrupt_scheme(adapter
, true)) {
7261 dev_err(&pdev
->dev
, "Unable to allocate memory for queues\n");
7265 if (netif_running(netdev
))
7273 static int igb_pci_disable_sriov(struct pci_dev
*dev
)
7275 int err
= igb_disable_sriov(dev
);
7278 err
= igb_sriov_reinit(dev
);
7283 static int igb_pci_enable_sriov(struct pci_dev
*dev
, int num_vfs
)
7285 int err
= igb_enable_sriov(dev
, num_vfs
);
7290 err
= igb_sriov_reinit(dev
);
7299 static int igb_pci_sriov_configure(struct pci_dev
*dev
, int num_vfs
)
7301 #ifdef CONFIG_PCI_IOV
7303 return igb_pci_disable_sriov(dev
);
7305 return igb_pci_enable_sriov(dev
, num_vfs
);
7310 #ifdef CONFIG_NET_POLL_CONTROLLER
7311 /* Polling 'interrupt' - used by things like netconsole to send skbs
7312 * without having to re-enable interrupts. It's not called while
7313 * the interrupt routine is executing.
7315 static void igb_netpoll(struct net_device
*netdev
)
7317 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7318 struct e1000_hw
*hw
= &adapter
->hw
;
7319 struct igb_q_vector
*q_vector
;
7322 for (i
= 0; i
< adapter
->num_q_vectors
; i
++) {
7323 q_vector
= adapter
->q_vector
[i
];
7324 if (adapter
->msix_entries
)
7325 wr32(E1000_EIMC
, q_vector
->eims_value
);
7327 igb_irq_disable(adapter
);
7328 napi_schedule(&q_vector
->napi
);
7331 #endif /* CONFIG_NET_POLL_CONTROLLER */
7334 * igb_io_error_detected - called when PCI error is detected
7335 * @pdev: Pointer to PCI device
7336 * @state: The current pci connection state
7338 * This function is called after a PCI bus error affecting
7339 * this device has been detected.
7341 static pci_ers_result_t
igb_io_error_detected(struct pci_dev
*pdev
,
7342 pci_channel_state_t state
)
7344 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7345 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7347 netif_device_detach(netdev
);
7349 if (state
== pci_channel_io_perm_failure
)
7350 return PCI_ERS_RESULT_DISCONNECT
;
7352 if (netif_running(netdev
))
7354 pci_disable_device(pdev
);
7356 /* Request a slot slot reset. */
7357 return PCI_ERS_RESULT_NEED_RESET
;
7361 * igb_io_slot_reset - called after the pci bus has been reset.
7362 * @pdev: Pointer to PCI device
7364 * Restart the card from scratch, as if from a cold-boot. Implementation
7365 * resembles the first-half of the igb_resume routine.
7367 static pci_ers_result_t
igb_io_slot_reset(struct pci_dev
*pdev
)
7369 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7370 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7371 struct e1000_hw
*hw
= &adapter
->hw
;
7372 pci_ers_result_t result
;
7375 if (pci_enable_device_mem(pdev
)) {
7377 "Cannot re-enable PCI device after reset.\n");
7378 result
= PCI_ERS_RESULT_DISCONNECT
;
7380 pci_set_master(pdev
);
7381 pci_restore_state(pdev
);
7382 pci_save_state(pdev
);
7384 pci_enable_wake(pdev
, PCI_D3hot
, 0);
7385 pci_enable_wake(pdev
, PCI_D3cold
, 0);
7388 wr32(E1000_WUS
, ~0);
7389 result
= PCI_ERS_RESULT_RECOVERED
;
7392 err
= pci_cleanup_aer_uncorrect_error_status(pdev
);
7395 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
7397 /* non-fatal, continue */
7404 * igb_io_resume - called when traffic can start flowing again.
7405 * @pdev: Pointer to PCI device
7407 * This callback is called when the error recovery driver tells us that
7408 * its OK to resume normal operation. Implementation resembles the
7409 * second-half of the igb_resume routine.
7411 static void igb_io_resume(struct pci_dev
*pdev
)
7413 struct net_device
*netdev
= pci_get_drvdata(pdev
);
7414 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7416 if (netif_running(netdev
)) {
7417 if (igb_up(adapter
)) {
7418 dev_err(&pdev
->dev
, "igb_up failed after reset\n");
7423 netif_device_attach(netdev
);
7425 /* let the f/w know that the h/w is now under the control of the
7428 igb_get_hw_control(adapter
);
7431 static void igb_rar_set_qsel(struct igb_adapter
*adapter
, u8
*addr
, u32 index
,
7434 u32 rar_low
, rar_high
;
7435 struct e1000_hw
*hw
= &adapter
->hw
;
7437 /* HW expects these in little endian so we reverse the byte order
7438 * from network order (big endian) to little endian
7440 rar_low
= ((u32
) addr
[0] | ((u32
) addr
[1] << 8) |
7441 ((u32
) addr
[2] << 16) | ((u32
) addr
[3] << 24));
7442 rar_high
= ((u32
) addr
[4] | ((u32
) addr
[5] << 8));
7444 /* Indicate to hardware the Address is Valid. */
7445 rar_high
|= E1000_RAH_AV
;
7447 if (hw
->mac
.type
== e1000_82575
)
7448 rar_high
|= E1000_RAH_POOL_1
* qsel
;
7450 rar_high
|= E1000_RAH_POOL_1
<< qsel
;
7452 wr32(E1000_RAL(index
), rar_low
);
7454 wr32(E1000_RAH(index
), rar_high
);
7458 static int igb_set_vf_mac(struct igb_adapter
*adapter
,
7459 int vf
, unsigned char *mac_addr
)
7461 struct e1000_hw
*hw
= &adapter
->hw
;
7462 /* VF MAC addresses start at end of receive addresses and moves
7463 * towards the first, as a result a collision should not be possible
7465 int rar_entry
= hw
->mac
.rar_entry_count
- (vf
+ 1);
7467 memcpy(adapter
->vf_data
[vf
].vf_mac_addresses
, mac_addr
, ETH_ALEN
);
7469 igb_rar_set_qsel(adapter
, mac_addr
, rar_entry
, vf
);
7474 static int igb_ndo_set_vf_mac(struct net_device
*netdev
, int vf
, u8
*mac
)
7476 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7477 if (!is_valid_ether_addr(mac
) || (vf
>= adapter
->vfs_allocated_count
))
7479 adapter
->vf_data
[vf
].flags
|= IGB_VF_FLAG_PF_SET_MAC
;
7480 dev_info(&adapter
->pdev
->dev
, "setting MAC %pM on VF %d\n", mac
, vf
);
7481 dev_info(&adapter
->pdev
->dev
,
7482 "Reload the VF driver to make this change effective.");
7483 if (test_bit(__IGB_DOWN
, &adapter
->state
)) {
7484 dev_warn(&adapter
->pdev
->dev
,
7485 "The VF MAC address has been set, but the PF device is not up.\n");
7486 dev_warn(&adapter
->pdev
->dev
,
7487 "Bring the PF device up before attempting to use the VF device.\n");
7489 return igb_set_vf_mac(adapter
, vf
, mac
);
7492 static int igb_link_mbps(int internal_link_speed
)
7494 switch (internal_link_speed
) {
7504 static void igb_set_vf_rate_limit(struct e1000_hw
*hw
, int vf
, int tx_rate
,
7511 /* Calculate the rate factor values to set */
7512 rf_int
= link_speed
/ tx_rate
;
7513 rf_dec
= (link_speed
- (rf_int
* tx_rate
));
7514 rf_dec
= (rf_dec
* (1 << E1000_RTTBCNRC_RF_INT_SHIFT
)) /
7517 bcnrc_val
= E1000_RTTBCNRC_RS_ENA
;
7518 bcnrc_val
|= ((rf_int
<< E1000_RTTBCNRC_RF_INT_SHIFT
) &
7519 E1000_RTTBCNRC_RF_INT_MASK
);
7520 bcnrc_val
|= (rf_dec
& E1000_RTTBCNRC_RF_DEC_MASK
);
7525 wr32(E1000_RTTDQSEL
, vf
); /* vf X uses queue X */
7526 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7527 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7529 wr32(E1000_RTTBCNRM
, 0x14);
7530 wr32(E1000_RTTBCNRC
, bcnrc_val
);
7533 static void igb_check_vf_rate_limit(struct igb_adapter
*adapter
)
7535 int actual_link_speed
, i
;
7536 bool reset_rate
= false;
7538 /* VF TX rate limit was not set or not supported */
7539 if ((adapter
->vf_rate_link_speed
== 0) ||
7540 (adapter
->hw
.mac
.type
!= e1000_82576
))
7543 actual_link_speed
= igb_link_mbps(adapter
->link_speed
);
7544 if (actual_link_speed
!= adapter
->vf_rate_link_speed
) {
7546 adapter
->vf_rate_link_speed
= 0;
7547 dev_info(&adapter
->pdev
->dev
,
7548 "Link speed has been changed. VF Transmit rate is disabled\n");
7551 for (i
= 0; i
< adapter
->vfs_allocated_count
; i
++) {
7553 adapter
->vf_data
[i
].tx_rate
= 0;
7555 igb_set_vf_rate_limit(&adapter
->hw
, i
,
7556 adapter
->vf_data
[i
].tx_rate
,
7561 static int igb_ndo_set_vf_bw(struct net_device
*netdev
, int vf
, int tx_rate
)
7563 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7564 struct e1000_hw
*hw
= &adapter
->hw
;
7565 int actual_link_speed
;
7567 if (hw
->mac
.type
!= e1000_82576
)
7570 actual_link_speed
= igb_link_mbps(adapter
->link_speed
);
7571 if ((vf
>= adapter
->vfs_allocated_count
) ||
7572 (!(rd32(E1000_STATUS
) & E1000_STATUS_LU
)) ||
7573 (tx_rate
< 0) || (tx_rate
> actual_link_speed
))
7576 adapter
->vf_rate_link_speed
= actual_link_speed
;
7577 adapter
->vf_data
[vf
].tx_rate
= (u16
)tx_rate
;
7578 igb_set_vf_rate_limit(hw
, vf
, tx_rate
, actual_link_speed
);
7583 static int igb_ndo_set_vf_spoofchk(struct net_device
*netdev
, int vf
,
7586 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7587 struct e1000_hw
*hw
= &adapter
->hw
;
7588 u32 reg_val
, reg_offset
;
7590 if (!adapter
->vfs_allocated_count
)
7593 if (vf
>= adapter
->vfs_allocated_count
)
7596 reg_offset
= (hw
->mac
.type
== e1000_82576
) ? E1000_DTXSWC
: E1000_TXSWC
;
7597 reg_val
= rd32(reg_offset
);
7599 reg_val
|= ((1 << vf
) |
7600 (1 << (vf
+ E1000_DTXSWC_VLAN_SPOOF_SHIFT
)));
7602 reg_val
&= ~((1 << vf
) |
7603 (1 << (vf
+ E1000_DTXSWC_VLAN_SPOOF_SHIFT
)));
7604 wr32(reg_offset
, reg_val
);
7606 adapter
->vf_data
[vf
].spoofchk_enabled
= setting
;
7607 return E1000_SUCCESS
;
7610 static int igb_ndo_get_vf_config(struct net_device
*netdev
,
7611 int vf
, struct ifla_vf_info
*ivi
)
7613 struct igb_adapter
*adapter
= netdev_priv(netdev
);
7614 if (vf
>= adapter
->vfs_allocated_count
)
7617 memcpy(&ivi
->mac
, adapter
->vf_data
[vf
].vf_mac_addresses
, ETH_ALEN
);
7618 ivi
->tx_rate
= adapter
->vf_data
[vf
].tx_rate
;
7619 ivi
->vlan
= adapter
->vf_data
[vf
].pf_vlan
;
7620 ivi
->qos
= adapter
->vf_data
[vf
].pf_qos
;
7621 ivi
->spoofchk
= adapter
->vf_data
[vf
].spoofchk_enabled
;
7625 static void igb_vmm_control(struct igb_adapter
*adapter
)
7627 struct e1000_hw
*hw
= &adapter
->hw
;
7630 switch (hw
->mac
.type
) {
7636 /* replication is not supported for 82575 */
7639 /* notify HW that the MAC is adding vlan tags */
7640 reg
= rd32(E1000_DTXCTL
);
7641 reg
|= E1000_DTXCTL_VLAN_ADDED
;
7642 wr32(E1000_DTXCTL
, reg
);
7644 /* enable replication vlan tag stripping */
7645 reg
= rd32(E1000_RPLOLR
);
7646 reg
|= E1000_RPLOLR_STRVLAN
;
7647 wr32(E1000_RPLOLR
, reg
);
7649 /* none of the above registers are supported by i350 */
7653 if (adapter
->vfs_allocated_count
) {
7654 igb_vmdq_set_loopback_pf(hw
, true);
7655 igb_vmdq_set_replication_pf(hw
, true);
7656 igb_vmdq_set_anti_spoofing_pf(hw
, true,
7657 adapter
->vfs_allocated_count
);
7659 igb_vmdq_set_loopback_pf(hw
, false);
7660 igb_vmdq_set_replication_pf(hw
, false);
7664 static void igb_init_dmac(struct igb_adapter
*adapter
, u32 pba
)
7666 struct e1000_hw
*hw
= &adapter
->hw
;
7670 if (hw
->mac
.type
> e1000_82580
) {
7671 if (adapter
->flags
& IGB_FLAG_DMAC
) {
7674 /* force threshold to 0. */
7675 wr32(E1000_DMCTXTH
, 0);
7677 /* DMA Coalescing high water mark needs to be greater
7678 * than the Rx threshold. Set hwm to PBA - max frame
7679 * size in 16B units, capping it at PBA - 6KB.
7681 hwm
= 64 * pba
- adapter
->max_frame_size
/ 16;
7682 if (hwm
< 64 * (pba
- 6))
7683 hwm
= 64 * (pba
- 6);
7684 reg
= rd32(E1000_FCRTC
);
7685 reg
&= ~E1000_FCRTC_RTH_COAL_MASK
;
7686 reg
|= ((hwm
<< E1000_FCRTC_RTH_COAL_SHIFT
)
7687 & E1000_FCRTC_RTH_COAL_MASK
);
7688 wr32(E1000_FCRTC
, reg
);
7690 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
7691 * frame size, capping it at PBA - 10KB.
7693 dmac_thr
= pba
- adapter
->max_frame_size
/ 512;
7694 if (dmac_thr
< pba
- 10)
7695 dmac_thr
= pba
- 10;
7696 reg
= rd32(E1000_DMACR
);
7697 reg
&= ~E1000_DMACR_DMACTHR_MASK
;
7698 reg
|= ((dmac_thr
<< E1000_DMACR_DMACTHR_SHIFT
)
7699 & E1000_DMACR_DMACTHR_MASK
);
7701 /* transition to L0x or L1 if available..*/
7702 reg
|= (E1000_DMACR_DMAC_EN
| E1000_DMACR_DMAC_LX_MASK
);
7704 /* watchdog timer= +-1000 usec in 32usec intervals */
7707 /* Disable BMC-to-OS Watchdog Enable */
7708 if (hw
->mac
.type
!= e1000_i354
)
7709 reg
&= ~E1000_DMACR_DC_BMC2OSW_EN
;
7711 wr32(E1000_DMACR
, reg
);
7713 /* no lower threshold to disable
7714 * coalescing(smart fifb)-UTRESH=0
7716 wr32(E1000_DMCRTRH
, 0);
7718 reg
= (IGB_DMCTLX_DCFLUSH_DIS
| 0x4);
7720 wr32(E1000_DMCTLX
, reg
);
7722 /* free space in tx packet buffer to wake from
7725 wr32(E1000_DMCTXTH
, (IGB_MIN_TXPBSIZE
-
7726 (IGB_TX_BUF_4096
+ adapter
->max_frame_size
)) >> 6);
7728 /* make low power state decision controlled
7731 reg
= rd32(E1000_PCIEMISC
);
7732 reg
&= ~E1000_PCIEMISC_LX_DECISION
;
7733 wr32(E1000_PCIEMISC
, reg
);
7734 } /* endif adapter->dmac is not disabled */
7735 } else if (hw
->mac
.type
== e1000_82580
) {
7736 u32 reg
= rd32(E1000_PCIEMISC
);
7737 wr32(E1000_PCIEMISC
, reg
& ~E1000_PCIEMISC_LX_DECISION
);
7738 wr32(E1000_DMACR
, 0);
7743 * igb_read_i2c_byte - Reads 8 bit word over I2C
7744 * @hw: pointer to hardware structure
7745 * @byte_offset: byte offset to read
7746 * @dev_addr: device address
7749 * Performs byte read operation over I2C interface at
7750 * a specified device address.
7752 s32
igb_read_i2c_byte(struct e1000_hw
*hw
, u8 byte_offset
,
7753 u8 dev_addr
, u8
*data
)
7755 struct igb_adapter
*adapter
= container_of(hw
, struct igb_adapter
, hw
);
7756 struct i2c_client
*this_client
= adapter
->i2c_client
;
7761 return E1000_ERR_I2C
;
7763 swfw_mask
= E1000_SWFW_PHY0_SM
;
7765 if (hw
->mac
.ops
.acquire_swfw_sync(hw
, swfw_mask
)
7767 return E1000_ERR_SWFW_SYNC
;
7769 status
= i2c_smbus_read_byte_data(this_client
, byte_offset
);
7770 hw
->mac
.ops
.release_swfw_sync(hw
, swfw_mask
);
7773 return E1000_ERR_I2C
;
7776 return E1000_SUCCESS
;
7781 * igb_write_i2c_byte - Writes 8 bit word over I2C
7782 * @hw: pointer to hardware structure
7783 * @byte_offset: byte offset to write
7784 * @dev_addr: device address
7785 * @data: value to write
7787 * Performs byte write operation over I2C interface at
7788 * a specified device address.
7790 s32
igb_write_i2c_byte(struct e1000_hw
*hw
, u8 byte_offset
,
7791 u8 dev_addr
, u8 data
)
7793 struct igb_adapter
*adapter
= container_of(hw
, struct igb_adapter
, hw
);
7794 struct i2c_client
*this_client
= adapter
->i2c_client
;
7796 u16 swfw_mask
= E1000_SWFW_PHY0_SM
;
7799 return E1000_ERR_I2C
;
7801 if (hw
->mac
.ops
.acquire_swfw_sync(hw
, swfw_mask
) != E1000_SUCCESS
)
7802 return E1000_ERR_SWFW_SYNC
;
7803 status
= i2c_smbus_write_byte_data(this_client
, byte_offset
, data
);
7804 hw
->mac
.ops
.release_swfw_sync(hw
, swfw_mask
);
7807 return E1000_ERR_I2C
;
7809 return E1000_SUCCESS
;