projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
igb: A fix to VF TX rate limit
[deliverable/linux.git] / drivers / net / ethernet / intel / igb / igb_main.c
1 /*******************************************************************************
2
3 Intel(R) Gigabit Ethernet Linux driver
4 Copyright(c) 2007-2012 Intel Corporation.
5
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.
9
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
13 more details.
14
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.
18
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
21
22 Contact Information:
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29
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>
44 #include <linux/if.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>
50 #include <linux/ip.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>
57 #ifdef CONFIG_IGB_DCA
58 #include <linux/dca.h>
59 #endif
60 #include "igb.h"
61
62 #define MAJ 3
63 #define MIN 4
64 #define BUILD 7
65 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
66 __stringify(BUILD) "-k"
67 char igb_driver_name[] = "igb";
68 char igb_driver_version[] = DRV_VERSION;
69 static const char igb_driver_string[] =
70 "Intel(R) Gigabit Ethernet Network Driver";
71 static const char igb_copyright[] = "Copyright (c) 2007-2012 Intel Corporation.";
72
73 static const struct e1000_info *igb_info_tbl[] = {
74 [board_82575] = &e1000_82575_info,
75 };
76
77 static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl) = {
78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
79 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
98 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
99 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
100 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
101 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
102 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
103 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
104 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
105 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
106 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
107 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
108 /* required last entry */
109 {0, }
110 };
111
112 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
113
114 void igb_reset(struct igb_adapter *);
115 static int igb_setup_all_tx_resources(struct igb_adapter *);
116 static int igb_setup_all_rx_resources(struct igb_adapter *);
117 static void igb_free_all_tx_resources(struct igb_adapter *);
118 static void igb_free_all_rx_resources(struct igb_adapter *);
119 static void igb_setup_mrqc(struct igb_adapter *);
120 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
121 static void __devexit igb_remove(struct pci_dev *pdev);
122 static int igb_sw_init(struct igb_adapter *);
123 static int igb_open(struct net_device *);
124 static int igb_close(struct net_device *);
125 static void igb_configure_tx(struct igb_adapter *);
126 static void igb_configure_rx(struct igb_adapter *);
127 static void igb_clean_all_tx_rings(struct igb_adapter *);
128 static void igb_clean_all_rx_rings(struct igb_adapter *);
129 static void igb_clean_tx_ring(struct igb_ring *);
130 static void igb_clean_rx_ring(struct igb_ring *);
131 static void igb_set_rx_mode(struct net_device *);
132 static void igb_update_phy_info(unsigned long);
133 static void igb_watchdog(unsigned long);
134 static void igb_watchdog_task(struct work_struct *);
135 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
136 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *dev,
137 struct rtnl_link_stats64 *stats);
138 static int igb_change_mtu(struct net_device *, int);
139 static int igb_set_mac(struct net_device *, void *);
140 static void igb_set_uta(struct igb_adapter *adapter);
141 static irqreturn_t igb_intr(int irq, void *);
142 static irqreturn_t igb_intr_msi(int irq, void *);
143 static irqreturn_t igb_msix_other(int irq, void *);
144 static irqreturn_t igb_msix_ring(int irq, void *);
145 #ifdef CONFIG_IGB_DCA
146 static void igb_update_dca(struct igb_q_vector *);
147 static void igb_setup_dca(struct igb_adapter *);
148 #endif /* CONFIG_IGB_DCA */
149 static int igb_poll(struct napi_struct *, int);
150 static bool igb_clean_tx_irq(struct igb_q_vector *);
151 static bool igb_clean_rx_irq(struct igb_q_vector *, int);
152 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
153 static void igb_tx_timeout(struct net_device *);
154 static void igb_reset_task(struct work_struct *);
155 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features);
156 static int igb_vlan_rx_add_vid(struct net_device *, u16);
157 static int igb_vlan_rx_kill_vid(struct net_device *, u16);
158 static void igb_restore_vlan(struct igb_adapter *);
159 static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
160 static void igb_ping_all_vfs(struct igb_adapter *);
161 static void igb_msg_task(struct igb_adapter *);
162 static void igb_vmm_control(struct igb_adapter *);
163 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
164 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
165 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
166 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
167 int vf, u16 vlan, u8 qos);
168 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate);
169 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
170 struct ifla_vf_info *ivi);
171 static void igb_check_vf_rate_limit(struct igb_adapter *);
172
173 #ifdef CONFIG_PCI_IOV
174 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
175 static int igb_find_enabled_vfs(struct igb_adapter *adapter);
176 static int igb_check_vf_assignment(struct igb_adapter *adapter);
177 #endif
178
179 #ifdef CONFIG_PM
180 #ifdef CONFIG_PM_SLEEP
181 static int igb_suspend(struct device *);
182 #endif
183 static int igb_resume(struct device *);
184 #ifdef CONFIG_PM_RUNTIME
185 static int igb_runtime_suspend(struct device *dev);
186 static int igb_runtime_resume(struct device *dev);
187 static int igb_runtime_idle(struct device *dev);
188 #endif
189 static const struct dev_pm_ops igb_pm_ops = {
190 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
191 SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
192 igb_runtime_idle)
193 };
194 #endif
195 static void igb_shutdown(struct pci_dev *);
196 #ifdef CONFIG_IGB_DCA
197 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
198 static struct notifier_block dca_notifier = {
199 .notifier_call = igb_notify_dca,
200 .next = NULL,
201 .priority = 0
202 };
203 #endif
204 #ifdef CONFIG_NET_POLL_CONTROLLER
205 /* for netdump / net console */
206 static void igb_netpoll(struct net_device *);
207 #endif
208 #ifdef CONFIG_PCI_IOV
209 static unsigned int max_vfs = 0;
210 module_param(max_vfs, uint, 0);
211 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
212 "per physical function");
213 #endif /* CONFIG_PCI_IOV */
214
215 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
216 pci_channel_state_t);
217 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
218 static void igb_io_resume(struct pci_dev *);
219
220 static struct pci_error_handlers igb_err_handler = {
221 .error_detected = igb_io_error_detected,
222 .slot_reset = igb_io_slot_reset,
223 .resume = igb_io_resume,
224 };
225
226 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
227
228 static struct pci_driver igb_driver = {
229 .name = igb_driver_name,
230 .id_table = igb_pci_tbl,
231 .probe = igb_probe,
232 .remove = __devexit_p(igb_remove),
233 #ifdef CONFIG_PM
234 .driver.pm = &igb_pm_ops,
235 #endif
236 .shutdown = igb_shutdown,
237 .err_handler = &igb_err_handler
238 };
239
240 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
241 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
242 MODULE_LICENSE("GPL");
243 MODULE_VERSION(DRV_VERSION);
244
245 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
246 static int debug = -1;
247 module_param(debug, int, 0);
248 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
249
250 struct igb_reg_info {
251 u32 ofs;
252 char *name;
253 };
254
255 static const struct igb_reg_info igb_reg_info_tbl[] = {
256
257 /* General Registers */
258 {E1000_CTRL, "CTRL"},
259 {E1000_STATUS, "STATUS"},
260 {E1000_CTRL_EXT, "CTRL_EXT"},
261
262 /* Interrupt Registers */
263 {E1000_ICR, "ICR"},
264
265 /* RX Registers */
266 {E1000_RCTL, "RCTL"},
267 {E1000_RDLEN(0), "RDLEN"},
268 {E1000_RDH(0), "RDH"},
269 {E1000_RDT(0), "RDT"},
270 {E1000_RXDCTL(0), "RXDCTL"},
271 {E1000_RDBAL(0), "RDBAL"},
272 {E1000_RDBAH(0), "RDBAH"},
273
274 /* TX Registers */
275 {E1000_TCTL, "TCTL"},
276 {E1000_TDBAL(0), "TDBAL"},
277 {E1000_TDBAH(0), "TDBAH"},
278 {E1000_TDLEN(0), "TDLEN"},
279 {E1000_TDH(0), "TDH"},
280 {E1000_TDT(0), "TDT"},
281 {E1000_TXDCTL(0), "TXDCTL"},
282 {E1000_TDFH, "TDFH"},
283 {E1000_TDFT, "TDFT"},
284 {E1000_TDFHS, "TDFHS"},
285 {E1000_TDFPC, "TDFPC"},
286
287 /* List Terminator */
288 {}
289 };
290
291 /*
292 * igb_regdump - register printout routine
293 */
294 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
295 {
296 int n = 0;
297 char rname[16];
298 u32 regs[8];
299
300 switch (reginfo->ofs) {
301 case E1000_RDLEN(0):
302 for (n = 0; n < 4; n++)
303 regs[n] = rd32(E1000_RDLEN(n));
304 break;
305 case E1000_RDH(0):
306 for (n = 0; n < 4; n++)
307 regs[n] = rd32(E1000_RDH(n));
308 break;
309 case E1000_RDT(0):
310 for (n = 0; n < 4; n++)
311 regs[n] = rd32(E1000_RDT(n));
312 break;
313 case E1000_RXDCTL(0):
314 for (n = 0; n < 4; n++)
315 regs[n] = rd32(E1000_RXDCTL(n));
316 break;
317 case E1000_RDBAL(0):
318 for (n = 0; n < 4; n++)
319 regs[n] = rd32(E1000_RDBAL(n));
320 break;
321 case E1000_RDBAH(0):
322 for (n = 0; n < 4; n++)
323 regs[n] = rd32(E1000_RDBAH(n));
324 break;
325 case E1000_TDBAL(0):
326 for (n = 0; n < 4; n++)
327 regs[n] = rd32(E1000_RDBAL(n));
328 break;
329 case E1000_TDBAH(0):
330 for (n = 0; n < 4; n++)
331 regs[n] = rd32(E1000_TDBAH(n));
332 break;
333 case E1000_TDLEN(0):
334 for (n = 0; n < 4; n++)
335 regs[n] = rd32(E1000_TDLEN(n));
336 break;
337 case E1000_TDH(0):
338 for (n = 0; n < 4; n++)
339 regs[n] = rd32(E1000_TDH(n));
340 break;
341 case E1000_TDT(0):
342 for (n = 0; n < 4; n++)
343 regs[n] = rd32(E1000_TDT(n));
344 break;
345 case E1000_TXDCTL(0):
346 for (n = 0; n < 4; n++)
347 regs[n] = rd32(E1000_TXDCTL(n));
348 break;
349 default:
350 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
351 return;
352 }
353
354 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
355 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
356 regs[2], regs[3]);
357 }
358
359 /*
360 * igb_dump - Print registers, tx-rings and rx-rings
361 */
362 static void igb_dump(struct igb_adapter *adapter)
363 {
364 struct net_device *netdev = adapter->netdev;
365 struct e1000_hw *hw = &adapter->hw;
366 struct igb_reg_info *reginfo;
367 struct igb_ring *tx_ring;
368 union e1000_adv_tx_desc *tx_desc;
369 struct my_u0 { u64 a; u64 b; } *u0;
370 struct igb_ring *rx_ring;
371 union e1000_adv_rx_desc *rx_desc;
372 u32 staterr;
373 u16 i, n;
374
375 if (!netif_msg_hw(adapter))
376 return;
377
378 /* Print netdevice Info */
379 if (netdev) {
380 dev_info(&adapter->pdev->dev, "Net device Info\n");
381 pr_info("Device Name state trans_start "
382 "last_rx\n");
383 pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
384 netdev->state, netdev->trans_start, netdev->last_rx);
385 }
386
387 /* Print Registers */
388 dev_info(&adapter->pdev->dev, "Register Dump\n");
389 pr_info(" Register Name Value\n");
390 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
391 reginfo->name; reginfo++) {
392 igb_regdump(hw, reginfo);
393 }
394
395 /* Print TX Ring Summary */
396 if (!netdev || !netif_running(netdev))
397 goto exit;
398
399 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
400 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
401 for (n = 0; n < adapter->num_tx_queues; n++) {
402 struct igb_tx_buffer *buffer_info;
403 tx_ring = adapter->tx_ring[n];
404 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
405 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
406 n, tx_ring->next_to_use, tx_ring->next_to_clean,
407 (u64)buffer_info->dma,
408 buffer_info->length,
409 buffer_info->next_to_watch,
410 (u64)buffer_info->time_stamp);
411 }
412
413 /* Print TX Rings */
414 if (!netif_msg_tx_done(adapter))
415 goto rx_ring_summary;
416
417 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
418
419 /* Transmit Descriptor Formats
420 *
421 * Advanced Transmit Descriptor
422 * +--------------------------------------------------------------+
423 * 0 | Buffer Address [63:0] |
424 * +--------------------------------------------------------------+
425 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
426 * +--------------------------------------------------------------+
427 * 63 46 45 40 39 38 36 35 32 31 24 15 0
428 */
429
430 for (n = 0; n < adapter->num_tx_queues; n++) {
431 tx_ring = adapter->tx_ring[n];
432 pr_info("------------------------------------\n");
433 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
434 pr_info("------------------------------------\n");
435 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] "
436 "[bi->dma ] leng ntw timestamp "
437 "bi->skb\n");
438
439 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
440 const char *next_desc;
441 struct igb_tx_buffer *buffer_info;
442 tx_desc = IGB_TX_DESC(tx_ring, i);
443 buffer_info = &tx_ring->tx_buffer_info[i];
444 u0 = (struct my_u0 *)tx_desc;
445 if (i == tx_ring->next_to_use &&
446 i == tx_ring->next_to_clean)
447 next_desc = " NTC/U";
448 else if (i == tx_ring->next_to_use)
449 next_desc = " NTU";
450 else if (i == tx_ring->next_to_clean)
451 next_desc = " NTC";
452 else
453 next_desc = "";
454
455 pr_info("T [0x%03X] %016llX %016llX %016llX"
456 " %04X %p %016llX %p%s\n", i,
457 le64_to_cpu(u0->a),
458 le64_to_cpu(u0->b),
459 (u64)buffer_info->dma,
460 buffer_info->length,
461 buffer_info->next_to_watch,
462 (u64)buffer_info->time_stamp,
463 buffer_info->skb, next_desc);
464
465 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
466 print_hex_dump(KERN_INFO, "",
467 DUMP_PREFIX_ADDRESS,
468 16, 1, phys_to_virt(buffer_info->dma),
469 buffer_info->length, true);
470 }
471 }
472
473 /* Print RX Rings Summary */
474 rx_ring_summary:
475 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
476 pr_info("Queue [NTU] [NTC]\n");
477 for (n = 0; n < adapter->num_rx_queues; n++) {
478 rx_ring = adapter->rx_ring[n];
479 pr_info(" %5d %5X %5X\n",
480 n, rx_ring->next_to_use, rx_ring->next_to_clean);
481 }
482
483 /* Print RX Rings */
484 if (!netif_msg_rx_status(adapter))
485 goto exit;
486
487 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
488
489 /* Advanced Receive Descriptor (Read) Format
490 * 63 1 0
491 * +-----------------------------------------------------+
492 * 0 | Packet Buffer Address [63:1] |A0/NSE|
493 * +----------------------------------------------+------+
494 * 8 | Header Buffer Address [63:1] | DD |
495 * +-----------------------------------------------------+
496 *
497 *
498 * Advanced Receive Descriptor (Write-Back) Format
499 *
500 * 63 48 47 32 31 30 21 20 17 16 4 3 0
501 * +------------------------------------------------------+
502 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
503 * | Checksum Ident | | | | Type | Type |
504 * +------------------------------------------------------+
505 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
506 * +------------------------------------------------------+
507 * 63 48 47 32 31 20 19 0
508 */
509
510 for (n = 0; n < adapter->num_rx_queues; n++) {
511 rx_ring = adapter->rx_ring[n];
512 pr_info("------------------------------------\n");
513 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
514 pr_info("------------------------------------\n");
515 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] "
516 "[bi->dma ] [bi->skb] <-- Adv Rx Read format\n");
517 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] -----"
518 "----------- [bi->skb] <-- Adv Rx Write-Back format\n");
519
520 for (i = 0; i < rx_ring->count; i++) {
521 const char *next_desc;
522 struct igb_rx_buffer *buffer_info;
523 buffer_info = &rx_ring->rx_buffer_info[i];
524 rx_desc = IGB_RX_DESC(rx_ring, i);
525 u0 = (struct my_u0 *)rx_desc;
526 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
527
528 if (i == rx_ring->next_to_use)
529 next_desc = " NTU";
530 else if (i == rx_ring->next_to_clean)
531 next_desc = " NTC";
532 else
533 next_desc = "";
534
535 if (staterr & E1000_RXD_STAT_DD) {
536 /* Descriptor Done */
537 pr_info("%s[0x%03X] %016llX %016llX -------"
538 "--------- %p%s\n", "RWB", i,
539 le64_to_cpu(u0->a),
540 le64_to_cpu(u0->b),
541 buffer_info->skb, next_desc);
542 } else {
543 pr_info("%s[0x%03X] %016llX %016llX %016llX"
544 " %p%s\n", "R ", i,
545 le64_to_cpu(u0->a),
546 le64_to_cpu(u0->b),
547 (u64)buffer_info->dma,
548 buffer_info->skb, next_desc);
549
550 if (netif_msg_pktdata(adapter)) {
551 print_hex_dump(KERN_INFO, "",
552 DUMP_PREFIX_ADDRESS,
553 16, 1,
554 phys_to_virt(buffer_info->dma),
555 IGB_RX_HDR_LEN, true);
556 print_hex_dump(KERN_INFO, "",
557 DUMP_PREFIX_ADDRESS,
558 16, 1,
559 phys_to_virt(
560 buffer_info->page_dma +
561 buffer_info->page_offset),
562 PAGE_SIZE/2, true);
563 }
564 }
565 }
566 }
567
568 exit:
569 return;
570 }
571
572 /**
573 * igb_get_hw_dev - return device
574 * used by hardware layer to print debugging information
575 **/
576 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
577 {
578 struct igb_adapter *adapter = hw->back;
579 return adapter->netdev;
580 }
581
582 /**
583 * igb_init_module - Driver Registration Routine
584 *
585 * igb_init_module is the first routine called when the driver is
586 * loaded. All it does is register with the PCI subsystem.
587 **/
588 static int __init igb_init_module(void)
589 {
590 int ret;
591 pr_info("%s - version %s\n",
592 igb_driver_string, igb_driver_version);
593
594 pr_info("%s\n", igb_copyright);
595
596 #ifdef CONFIG_IGB_DCA
597 dca_register_notify(&dca_notifier);
598 #endif
599 ret = pci_register_driver(&igb_driver);
600 return ret;
601 }
602
603 module_init(igb_init_module);
604
605 /**
606 * igb_exit_module - Driver Exit Cleanup Routine
607 *
608 * igb_exit_module is called just before the driver is removed
609 * from memory.
610 **/
611 static void __exit igb_exit_module(void)
612 {
613 #ifdef CONFIG_IGB_DCA
614 dca_unregister_notify(&dca_notifier);
615 #endif
616 pci_unregister_driver(&igb_driver);
617 }
618
619 module_exit(igb_exit_module);
620
621 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
622 /**
623 * igb_cache_ring_register - Descriptor ring to register mapping
624 * @adapter: board private structure to initialize
625 *
626 * Once we know the feature-set enabled for the device, we'll cache
627 * the register offset the descriptor ring is assigned to.
628 **/
629 static void igb_cache_ring_register(struct igb_adapter *adapter)
630 {
631 int i = 0, j = 0;
632 u32 rbase_offset = adapter->vfs_allocated_count;
633
634 switch (adapter->hw.mac.type) {
635 case e1000_82576:
636 /* The queues are allocated for virtualization such that VF 0
637 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
638 * In order to avoid collision we start at the first free queue
639 * and continue consuming queues in the same sequence
640 */
641 if (adapter->vfs_allocated_count) {
642 for (; i < adapter->rss_queues; i++)
643 adapter->rx_ring[i]->reg_idx = rbase_offset +
644 Q_IDX_82576(i);
645 }
646 case e1000_82575:
647 case e1000_82580:
648 case e1000_i350:
649 case e1000_i210:
650 case e1000_i211:
651 default:
652 for (; i < adapter->num_rx_queues; i++)
653 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
654 for (; j < adapter->num_tx_queues; j++)
655 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
656 break;
657 }
658 }
659
660 static void igb_free_queues(struct igb_adapter *adapter)
661 {
662 int i;
663
664 for (i = 0; i < adapter->num_tx_queues; i++) {
665 kfree(adapter->tx_ring[i]);
666 adapter->tx_ring[i] = NULL;
667 }
668 for (i = 0; i < adapter->num_rx_queues; i++) {
669 kfree(adapter->rx_ring[i]);
670 adapter->rx_ring[i] = NULL;
671 }
672 adapter->num_rx_queues = 0;
673 adapter->num_tx_queues = 0;
674 }
675
676 /**
677 * igb_alloc_queues - Allocate memory for all rings
678 * @adapter: board private structure to initialize
679 *
680 * We allocate one ring per queue at run-time since we don't know the
681 * number of queues at compile-time.
682 **/
683 static int igb_alloc_queues(struct igb_adapter *adapter)
684 {
685 struct igb_ring *ring;
686 int i;
687 int orig_node = adapter->node;
688
689 for (i = 0; i < adapter->num_tx_queues; i++) {
690 if (orig_node == -1) {
691 int cur_node = next_online_node(adapter->node);
692 if (cur_node == MAX_NUMNODES)
693 cur_node = first_online_node;
694 adapter->node = cur_node;
695 }
696 ring = kzalloc_node(sizeof(struct igb_ring), GFP_KERNEL,
697 adapter->node);
698 if (!ring)
699 ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL);
700 if (!ring)
701 goto err;
702 ring->count = adapter->tx_ring_count;
703 ring->queue_index = i;
704 ring->dev = &adapter->pdev->dev;
705 ring->netdev = adapter->netdev;
706 ring->numa_node = adapter->node;
707 /* For 82575, context index must be unique per ring. */
708 if (adapter->hw.mac.type == e1000_82575)
709 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
710 adapter->tx_ring[i] = ring;
711 }
712 /* Restore the adapter's original node */
713 adapter->node = orig_node;
714
715 for (i = 0; i < adapter->num_rx_queues; i++) {
716 if (orig_node == -1) {
717 int cur_node = next_online_node(adapter->node);
718 if (cur_node == MAX_NUMNODES)
719 cur_node = first_online_node;
720 adapter->node = cur_node;
721 }
722 ring = kzalloc_node(sizeof(struct igb_ring), GFP_KERNEL,
723 adapter->node);
724 if (!ring)
725 ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL);
726 if (!ring)
727 goto err;
728 ring->count = adapter->rx_ring_count;
729 ring->queue_index = i;
730 ring->dev = &adapter->pdev->dev;
731 ring->netdev = adapter->netdev;
732 ring->numa_node = adapter->node;
733 /* set flag indicating ring supports SCTP checksum offload */
734 if (adapter->hw.mac.type >= e1000_82576)
735 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
736
737 /*
738 * On i350, i210, and i211, loopback VLAN packets
739 * have the tag byte-swapped.
740 * */
741 if (adapter->hw.mac.type >= e1000_i350)
742 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
743
744 adapter->rx_ring[i] = ring;
745 }
746 /* Restore the adapter's original node */
747 adapter->node = orig_node;
748
749 igb_cache_ring_register(adapter);
750
751 return 0;
752
753 err:
754 /* Restore the adapter's original node */
755 adapter->node = orig_node;
756 igb_free_queues(adapter);
757
758 return -ENOMEM;
759 }
760
761 /**
762 * igb_write_ivar - configure ivar for given MSI-X vector
763 * @hw: pointer to the HW structure
764 * @msix_vector: vector number we are allocating to a given ring
765 * @index: row index of IVAR register to write within IVAR table
766 * @offset: column offset of in IVAR, should be multiple of 8
767 *
768 * This function is intended to handle the writing of the IVAR register
769 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
770 * each containing an cause allocation for an Rx and Tx ring, and a
771 * variable number of rows depending on the number of queues supported.
772 **/
773 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
774 int index, int offset)
775 {
776 u32 ivar = array_rd32(E1000_IVAR0, index);
777
778 /* clear any bits that are currently set */
779 ivar &= ~((u32)0xFF << offset);
780
781 /* write vector and valid bit */
782 ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
783
784 array_wr32(E1000_IVAR0, index, ivar);
785 }
786
787 #define IGB_N0_QUEUE -1
788 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
789 {
790 struct igb_adapter *adapter = q_vector->adapter;
791 struct e1000_hw *hw = &adapter->hw;
792 int rx_queue = IGB_N0_QUEUE;
793 int tx_queue = IGB_N0_QUEUE;
794 u32 msixbm = 0;
795
796 if (q_vector->rx.ring)
797 rx_queue = q_vector->rx.ring->reg_idx;
798 if (q_vector->tx.ring)
799 tx_queue = q_vector->tx.ring->reg_idx;
800
801 switch (hw->mac.type) {
802 case e1000_82575:
803 /* The 82575 assigns vectors using a bitmask, which matches the
804 bitmask for the EICR/EIMS/EIMC registers. To assign one
805 or more queues to a vector, we write the appropriate bits
806 into the MSIXBM register for that vector. */
807 if (rx_queue > IGB_N0_QUEUE)
808 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
809 if (tx_queue > IGB_N0_QUEUE)
810 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
811 if (!adapter->msix_entries && msix_vector == 0)
812 msixbm |= E1000_EIMS_OTHER;
813 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
814 q_vector->eims_value = msixbm;
815 break;
816 case e1000_82576:
817 /*
818 * 82576 uses a table that essentially consists of 2 columns
819 * with 8 rows. The ordering is column-major so we use the
820 * lower 3 bits as the row index, and the 4th bit as the
821 * column offset.
822 */
823 if (rx_queue > IGB_N0_QUEUE)
824 igb_write_ivar(hw, msix_vector,
825 rx_queue & 0x7,
826 (rx_queue & 0x8) << 1);
827 if (tx_queue > IGB_N0_QUEUE)
828 igb_write_ivar(hw, msix_vector,
829 tx_queue & 0x7,
830 ((tx_queue & 0x8) << 1) + 8);
831 q_vector->eims_value = 1 << msix_vector;
832 break;
833 case e1000_82580:
834 case e1000_i350:
835 case e1000_i210:
836 case e1000_i211:
837 /*
838 * On 82580 and newer adapters the scheme is similar to 82576
839 * however instead of ordering column-major we have things
840 * ordered row-major. So we traverse the table by using
841 * bit 0 as the column offset, and the remaining bits as the
842 * row index.
843 */
844 if (rx_queue > IGB_N0_QUEUE)
845 igb_write_ivar(hw, msix_vector,
846 rx_queue >> 1,
847 (rx_queue & 0x1) << 4);
848 if (tx_queue > IGB_N0_QUEUE)
849 igb_write_ivar(hw, msix_vector,
850 tx_queue >> 1,
851 ((tx_queue & 0x1) << 4) + 8);
852 q_vector->eims_value = 1 << msix_vector;
853 break;
854 default:
855 BUG();
856 break;
857 }
858
859 /* add q_vector eims value to global eims_enable_mask */
860 adapter->eims_enable_mask |= q_vector->eims_value;
861
862 /* configure q_vector to set itr on first interrupt */
863 q_vector->set_itr = 1;
864 }
865
866 /**
867 * igb_configure_msix - Configure MSI-X hardware
868 *
869 * igb_configure_msix sets up the hardware to properly
870 * generate MSI-X interrupts.
871 **/
872 static void igb_configure_msix(struct igb_adapter *adapter)
873 {
874 u32 tmp;
875 int i, vector = 0;
876 struct e1000_hw *hw = &adapter->hw;
877
878 adapter->eims_enable_mask = 0;
879
880 /* set vector for other causes, i.e. link changes */
881 switch (hw->mac.type) {
882 case e1000_82575:
883 tmp = rd32(E1000_CTRL_EXT);
884 /* enable MSI-X PBA support*/
885 tmp |= E1000_CTRL_EXT_PBA_CLR;
886
887 /* Auto-Mask interrupts upon ICR read. */
888 tmp |= E1000_CTRL_EXT_EIAME;
889 tmp |= E1000_CTRL_EXT_IRCA;
890
891 wr32(E1000_CTRL_EXT, tmp);
892
893 /* enable msix_other interrupt */
894 array_wr32(E1000_MSIXBM(0), vector++,
895 E1000_EIMS_OTHER);
896 adapter->eims_other = E1000_EIMS_OTHER;
897
898 break;
899
900 case e1000_82576:
901 case e1000_82580:
902 case e1000_i350:
903 case e1000_i210:
904 case e1000_i211:
905 /* Turn on MSI-X capability first, or our settings
906 * won't stick. And it will take days to debug. */
907 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
908 E1000_GPIE_PBA | E1000_GPIE_EIAME |
909 E1000_GPIE_NSICR);
910
911 /* enable msix_other interrupt */
912 adapter->eims_other = 1 << vector;
913 tmp = (vector++ | E1000_IVAR_VALID) << 8;
914
915 wr32(E1000_IVAR_MISC, tmp);
916 break;
917 default:
918 /* do nothing, since nothing else supports MSI-X */
919 break;
920 } /* switch (hw->mac.type) */
921
922 adapter->eims_enable_mask |= adapter->eims_other;
923
924 for (i = 0; i < adapter->num_q_vectors; i++)
925 igb_assign_vector(adapter->q_vector[i], vector++);
926
927 wrfl();
928 }
929
930 /**
931 * igb_request_msix - Initialize MSI-X interrupts
932 *
933 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
934 * kernel.
935 **/
936 static int igb_request_msix(struct igb_adapter *adapter)
937 {
938 struct net_device *netdev = adapter->netdev;
939 struct e1000_hw *hw = &adapter->hw;
940 int i, err = 0, vector = 0;
941
942 err = request_irq(adapter->msix_entries[vector].vector,
943 igb_msix_other, 0, netdev->name, adapter);
944 if (err)
945 goto out;
946 vector++;
947
948 for (i = 0; i < adapter->num_q_vectors; i++) {
949 struct igb_q_vector *q_vector = adapter->q_vector[i];
950
951 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
952
953 if (q_vector->rx.ring && q_vector->tx.ring)
954 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
955 q_vector->rx.ring->queue_index);
956 else if (q_vector->tx.ring)
957 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
958 q_vector->tx.ring->queue_index);
959 else if (q_vector->rx.ring)
960 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
961 q_vector->rx.ring->queue_index);
962 else
963 sprintf(q_vector->name, "%s-unused", netdev->name);
964
965 err = request_irq(adapter->msix_entries[vector].vector,
966 igb_msix_ring, 0, q_vector->name,
967 q_vector);
968 if (err)
969 goto out;
970 vector++;
971 }
972
973 igb_configure_msix(adapter);
974 return 0;
975 out:
976 return err;
977 }
978
979 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
980 {
981 if (adapter->msix_entries) {
982 pci_disable_msix(adapter->pdev);
983 kfree(adapter->msix_entries);
984 adapter->msix_entries = NULL;
985 } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
986 pci_disable_msi(adapter->pdev);
987 }
988 }
989
990 /**
991 * igb_free_q_vectors - Free memory allocated for interrupt vectors
992 * @adapter: board private structure to initialize
993 *
994 * This function frees the memory allocated to the q_vectors. In addition if
995 * NAPI is enabled it will delete any references to the NAPI struct prior
996 * to freeing the q_vector.
997 **/
998 static void igb_free_q_vectors(struct igb_adapter *adapter)
999 {
1000 int v_idx;
1001
1002 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
1003 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1004 adapter->q_vector[v_idx] = NULL;
1005 if (!q_vector)
1006 continue;
1007 netif_napi_del(&q_vector->napi);
1008 kfree(q_vector);
1009 }
1010 adapter->num_q_vectors = 0;
1011 }
1012
1013 /**
1014 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1015 *
1016 * This function resets the device so that it has 0 rx queues, tx queues, and
1017 * MSI-X interrupts allocated.
1018 */
1019 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1020 {
1021 igb_free_queues(adapter);
1022 igb_free_q_vectors(adapter);
1023 igb_reset_interrupt_capability(adapter);
1024 }
1025
1026 /**
1027 * igb_set_interrupt_capability - set MSI or MSI-X if supported
1028 *
1029 * Attempt to configure interrupts using the best available
1030 * capabilities of the hardware and kernel.
1031 **/
1032 static int igb_set_interrupt_capability(struct igb_adapter *adapter)
1033 {
1034 int err;
1035 int numvecs, i;
1036
1037 /* Number of supported queues. */
1038 adapter->num_rx_queues = adapter->rss_queues;
1039 if (adapter->vfs_allocated_count)
1040 adapter->num_tx_queues = 1;
1041 else
1042 adapter->num_tx_queues = adapter->rss_queues;
1043
1044 /* start with one vector for every rx queue */
1045 numvecs = adapter->num_rx_queues;
1046
1047 /* if tx handler is separate add 1 for every tx queue */
1048 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1049 numvecs += adapter->num_tx_queues;
1050
1051 /* i210 and i211 can only have 4 MSIX vectors for rx/tx queues. */
1052 if ((adapter->hw.mac.type == e1000_i210)
1053 || (adapter->hw.mac.type == e1000_i211))
1054 numvecs = 4;
1055
1056 /* store the number of vectors reserved for queues */
1057 adapter->num_q_vectors = numvecs;
1058
1059 /* add 1 vector for link status interrupts */
1060 numvecs++;
1061 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
1062 GFP_KERNEL);
1063
1064 if (!adapter->msix_entries)
1065 goto msi_only;
1066
1067 for (i = 0; i < numvecs; i++)
1068 adapter->msix_entries[i].entry = i;
1069
1070 err = pci_enable_msix(adapter->pdev,
1071 adapter->msix_entries,
1072 numvecs);
1073 if (err == 0)
1074 goto out;
1075
1076 igb_reset_interrupt_capability(adapter);
1077
1078 /* If we can't do MSI-X, try MSI */
1079 msi_only:
1080 #ifdef CONFIG_PCI_IOV
1081 /* disable SR-IOV for non MSI-X configurations */
1082 if (adapter->vf_data) {
1083 struct e1000_hw *hw = &adapter->hw;
1084 /* disable iov and allow time for transactions to clear */
1085 pci_disable_sriov(adapter->pdev);
1086 msleep(500);
1087
1088 kfree(adapter->vf_data);
1089 adapter->vf_data = NULL;
1090 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1091 wrfl();
1092 msleep(100);
1093 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1094 }
1095 #endif
1096 adapter->vfs_allocated_count = 0;
1097 adapter->rss_queues = 1;
1098 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1099 adapter->num_rx_queues = 1;
1100 adapter->num_tx_queues = 1;
1101 adapter->num_q_vectors = 1;
1102 if (!pci_enable_msi(adapter->pdev))
1103 adapter->flags |= IGB_FLAG_HAS_MSI;
1104 out:
1105 /* Notify the stack of the (possibly) reduced queue counts. */
1106 rtnl_lock();
1107 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
1108 err = netif_set_real_num_rx_queues(adapter->netdev,
1109 adapter->num_rx_queues);
1110 rtnl_unlock();
1111 return err;
1112 }
1113
1114 /**
1115 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1116 * @adapter: board private structure to initialize
1117 *
1118 * We allocate one q_vector per queue interrupt. If allocation fails we
1119 * return -ENOMEM.
1120 **/
1121 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1122 {
1123 struct igb_q_vector *q_vector;
1124 struct e1000_hw *hw = &adapter->hw;
1125 int v_idx;
1126 int orig_node = adapter->node;
1127
1128 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
1129 if ((adapter->num_q_vectors == (adapter->num_rx_queues +
1130 adapter->num_tx_queues)) &&
1131 (adapter->num_rx_queues == v_idx))
1132 adapter->node = orig_node;
1133 if (orig_node == -1) {
1134 int cur_node = next_online_node(adapter->node);
1135 if (cur_node == MAX_NUMNODES)
1136 cur_node = first_online_node;
1137 adapter->node = cur_node;
1138 }
1139 q_vector = kzalloc_node(sizeof(struct igb_q_vector), GFP_KERNEL,
1140 adapter->node);
1141 if (!q_vector)
1142 q_vector = kzalloc(sizeof(struct igb_q_vector),
1143 GFP_KERNEL);
1144 if (!q_vector)
1145 goto err_out;
1146 q_vector->adapter = adapter;
1147 q_vector->itr_register = hw->hw_addr + E1000_EITR(0);
1148 q_vector->itr_val = IGB_START_ITR;
1149 netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll, 64);
1150 adapter->q_vector[v_idx] = q_vector;
1151 }
1152 /* Restore the adapter's original node */
1153 adapter->node = orig_node;
1154
1155 return 0;
1156
1157 err_out:
1158 /* Restore the adapter's original node */
1159 adapter->node = orig_node;
1160 igb_free_q_vectors(adapter);
1161 return -ENOMEM;
1162 }
1163
1164 static void igb_map_rx_ring_to_vector(struct igb_adapter *adapter,
1165 int ring_idx, int v_idx)
1166 {
1167 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1168
1169 q_vector->rx.ring = adapter->rx_ring[ring_idx];
1170 q_vector->rx.ring->q_vector = q_vector;
1171 q_vector->rx.count++;
1172 q_vector->itr_val = adapter->rx_itr_setting;
1173 if (q_vector->itr_val && q_vector->itr_val <= 3)
1174 q_vector->itr_val = IGB_START_ITR;
1175 }
1176
1177 static void igb_map_tx_ring_to_vector(struct igb_adapter *adapter,
1178 int ring_idx, int v_idx)
1179 {
1180 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1181
1182 q_vector->tx.ring = adapter->tx_ring[ring_idx];
1183 q_vector->tx.ring->q_vector = q_vector;
1184 q_vector->tx.count++;
1185 q_vector->itr_val = adapter->tx_itr_setting;
1186 q_vector->tx.work_limit = adapter->tx_work_limit;
1187 if (q_vector->itr_val && q_vector->itr_val <= 3)
1188 q_vector->itr_val = IGB_START_ITR;
1189 }
1190
1191 /**
1192 * igb_map_ring_to_vector - maps allocated queues to vectors
1193 *
1194 * This function maps the recently allocated queues to vectors.
1195 **/
1196 static int igb_map_ring_to_vector(struct igb_adapter *adapter)
1197 {
1198 int i;
1199 int v_idx = 0;
1200
1201 if ((adapter->num_q_vectors < adapter->num_rx_queues) ||
1202 (adapter->num_q_vectors < adapter->num_tx_queues))
1203 return -ENOMEM;
1204
1205 if (adapter->num_q_vectors >=
1206 (adapter->num_rx_queues + adapter->num_tx_queues)) {
1207 for (i = 0; i < adapter->num_rx_queues; i++)
1208 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
1209 for (i = 0; i < adapter->num_tx_queues; i++)
1210 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
1211 } else {
1212 for (i = 0; i < adapter->num_rx_queues; i++) {
1213 if (i < adapter->num_tx_queues)
1214 igb_map_tx_ring_to_vector(adapter, i, v_idx);
1215 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
1216 }
1217 for (; i < adapter->num_tx_queues; i++)
1218 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
1219 }
1220 return 0;
1221 }
1222
1223 /**
1224 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1225 *
1226 * This function initializes the interrupts and allocates all of the queues.
1227 **/
1228 static int igb_init_interrupt_scheme(struct igb_adapter *adapter)
1229 {
1230 struct pci_dev *pdev = adapter->pdev;
1231 int err;
1232
1233 err = igb_set_interrupt_capability(adapter);
1234 if (err)
1235 return err;
1236
1237 err = igb_alloc_q_vectors(adapter);
1238 if (err) {
1239 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1240 goto err_alloc_q_vectors;
1241 }
1242
1243 err = igb_alloc_queues(adapter);
1244 if (err) {
1245 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1246 goto err_alloc_queues;
1247 }
1248
1249 err = igb_map_ring_to_vector(adapter);
1250 if (err) {
1251 dev_err(&pdev->dev, "Invalid q_vector to ring mapping\n");
1252 goto err_map_queues;
1253 }
1254
1255
1256 return 0;
1257 err_map_queues:
1258 igb_free_queues(adapter);
1259 err_alloc_queues:
1260 igb_free_q_vectors(adapter);
1261 err_alloc_q_vectors:
1262 igb_reset_interrupt_capability(adapter);
1263 return err;
1264 }
1265
1266 /**
1267 * igb_request_irq - initialize interrupts
1268 *
1269 * Attempts to configure interrupts using the best available
1270 * capabilities of the hardware and kernel.
1271 **/
1272 static int igb_request_irq(struct igb_adapter *adapter)
1273 {
1274 struct net_device *netdev = adapter->netdev;
1275 struct pci_dev *pdev = adapter->pdev;
1276 int err = 0;
1277
1278 if (adapter->msix_entries) {
1279 err = igb_request_msix(adapter);
1280 if (!err)
1281 goto request_done;
1282 /* fall back to MSI */
1283 igb_clear_interrupt_scheme(adapter);
1284 if (!pci_enable_msi(pdev))
1285 adapter->flags |= IGB_FLAG_HAS_MSI;
1286 igb_free_all_tx_resources(adapter);
1287 igb_free_all_rx_resources(adapter);
1288 adapter->num_tx_queues = 1;
1289 adapter->num_rx_queues = 1;
1290 adapter->num_q_vectors = 1;
1291 err = igb_alloc_q_vectors(adapter);
1292 if (err) {
1293 dev_err(&pdev->dev,
1294 "Unable to allocate memory for vectors\n");
1295 goto request_done;
1296 }
1297 err = igb_alloc_queues(adapter);
1298 if (err) {
1299 dev_err(&pdev->dev,
1300 "Unable to allocate memory for queues\n");
1301 igb_free_q_vectors(adapter);
1302 goto request_done;
1303 }
1304 igb_setup_all_tx_resources(adapter);
1305 igb_setup_all_rx_resources(adapter);
1306 }
1307
1308 igb_assign_vector(adapter->q_vector[0], 0);
1309
1310 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1311 err = request_irq(pdev->irq, igb_intr_msi, 0,
1312 netdev->name, adapter);
1313 if (!err)
1314 goto request_done;
1315
1316 /* fall back to legacy interrupts */
1317 igb_reset_interrupt_capability(adapter);
1318 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1319 }
1320
1321 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1322 netdev->name, adapter);
1323
1324 if (err)
1325 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1326 err);
1327
1328 request_done:
1329 return err;
1330 }
1331
1332 static void igb_free_irq(struct igb_adapter *adapter)
1333 {
1334 if (adapter->msix_entries) {
1335 int vector = 0, i;
1336
1337 free_irq(adapter->msix_entries[vector++].vector, adapter);
1338
1339 for (i = 0; i < adapter->num_q_vectors; i++)
1340 free_irq(adapter->msix_entries[vector++].vector,
1341 adapter->q_vector[i]);
1342 } else {
1343 free_irq(adapter->pdev->irq, adapter);
1344 }
1345 }
1346
1347 /**
1348 * igb_irq_disable - Mask off interrupt generation on the NIC
1349 * @adapter: board private structure
1350 **/
1351 static void igb_irq_disable(struct igb_adapter *adapter)
1352 {
1353 struct e1000_hw *hw = &adapter->hw;
1354
1355 /*
1356 * we need to be careful when disabling interrupts. The VFs are also
1357 * mapped into these registers and so clearing the bits can cause
1358 * issues on the VF drivers so we only need to clear what we set
1359 */
1360 if (adapter->msix_entries) {
1361 u32 regval = rd32(E1000_EIAM);
1362 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1363 wr32(E1000_EIMC, adapter->eims_enable_mask);
1364 regval = rd32(E1000_EIAC);
1365 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1366 }
1367
1368 wr32(E1000_IAM, 0);
1369 wr32(E1000_IMC, ~0);
1370 wrfl();
1371 if (adapter->msix_entries) {
1372 int i;
1373 for (i = 0; i < adapter->num_q_vectors; i++)
1374 synchronize_irq(adapter->msix_entries[i].vector);
1375 } else {
1376 synchronize_irq(adapter->pdev->irq);
1377 }
1378 }
1379
1380 /**
1381 * igb_irq_enable - Enable default interrupt generation settings
1382 * @adapter: board private structure
1383 **/
1384 static void igb_irq_enable(struct igb_adapter *adapter)
1385 {
1386 struct e1000_hw *hw = &adapter->hw;
1387
1388 if (adapter->msix_entries) {
1389 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1390 u32 regval = rd32(E1000_EIAC);
1391 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1392 regval = rd32(E1000_EIAM);
1393 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1394 wr32(E1000_EIMS, adapter->eims_enable_mask);
1395 if (adapter->vfs_allocated_count) {
1396 wr32(E1000_MBVFIMR, 0xFF);
1397 ims |= E1000_IMS_VMMB;
1398 }
1399 wr32(E1000_IMS, ims);
1400 } else {
1401 wr32(E1000_IMS, IMS_ENABLE_MASK |
1402 E1000_IMS_DRSTA);
1403 wr32(E1000_IAM, IMS_ENABLE_MASK |
1404 E1000_IMS_DRSTA);
1405 }
1406 }
1407
1408 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1409 {
1410 struct e1000_hw *hw = &adapter->hw;
1411 u16 vid = adapter->hw.mng_cookie.vlan_id;
1412 u16 old_vid = adapter->mng_vlan_id;
1413
1414 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1415 /* add VID to filter table */
1416 igb_vfta_set(hw, vid, true);
1417 adapter->mng_vlan_id = vid;
1418 } else {
1419 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1420 }
1421
1422 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1423 (vid != old_vid) &&
1424 !test_bit(old_vid, adapter->active_vlans)) {
1425 /* remove VID from filter table */
1426 igb_vfta_set(hw, old_vid, false);
1427 }
1428 }
1429
1430 /**
1431 * igb_release_hw_control - release control of the h/w to f/w
1432 * @adapter: address of board private structure
1433 *
1434 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1435 * For ASF and Pass Through versions of f/w this means that the
1436 * driver is no longer loaded.
1437 *
1438 **/
1439 static void igb_release_hw_control(struct igb_adapter *adapter)
1440 {
1441 struct e1000_hw *hw = &adapter->hw;
1442 u32 ctrl_ext;
1443
1444 /* Let firmware take over control of h/w */
1445 ctrl_ext = rd32(E1000_CTRL_EXT);
1446 wr32(E1000_CTRL_EXT,
1447 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1448 }
1449
1450 /**
1451 * igb_get_hw_control - get control of the h/w from f/w
1452 * @adapter: address of board private structure
1453 *
1454 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1455 * For ASF and Pass Through versions of f/w this means that
1456 * the driver is loaded.
1457 *
1458 **/
1459 static void igb_get_hw_control(struct igb_adapter *adapter)
1460 {
1461 struct e1000_hw *hw = &adapter->hw;
1462 u32 ctrl_ext;
1463
1464 /* Let firmware know the driver has taken over */
1465 ctrl_ext = rd32(E1000_CTRL_EXT);
1466 wr32(E1000_CTRL_EXT,
1467 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1468 }
1469
1470 /**
1471 * igb_configure - configure the hardware for RX and TX
1472 * @adapter: private board structure
1473 **/
1474 static void igb_configure(struct igb_adapter *adapter)
1475 {
1476 struct net_device *netdev = adapter->netdev;
1477 int i;
1478
1479 igb_get_hw_control(adapter);
1480 igb_set_rx_mode(netdev);
1481
1482 igb_restore_vlan(adapter);
1483
1484 igb_setup_tctl(adapter);
1485 igb_setup_mrqc(adapter);
1486 igb_setup_rctl(adapter);
1487
1488 igb_configure_tx(adapter);
1489 igb_configure_rx(adapter);
1490
1491 igb_rx_fifo_flush_82575(&adapter->hw);
1492
1493 /* call igb_desc_unused which always leaves
1494 * at least 1 descriptor unused to make sure
1495 * next_to_use != next_to_clean */
1496 for (i = 0; i < adapter->num_rx_queues; i++) {
1497 struct igb_ring *ring = adapter->rx_ring[i];
1498 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1499 }
1500 }
1501
1502 /**
1503 * igb_power_up_link - Power up the phy/serdes link
1504 * @adapter: address of board private structure
1505 **/
1506 void igb_power_up_link(struct igb_adapter *adapter)
1507 {
1508 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1509 igb_power_up_phy_copper(&adapter->hw);
1510 else
1511 igb_power_up_serdes_link_82575(&adapter->hw);
1512 igb_reset_phy(&adapter->hw);
1513 }
1514
1515 /**
1516 * igb_power_down_link - Power down the phy/serdes link
1517 * @adapter: address of board private structure
1518 */
1519 static void igb_power_down_link(struct igb_adapter *adapter)
1520 {
1521 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1522 igb_power_down_phy_copper_82575(&adapter->hw);
1523 else
1524 igb_shutdown_serdes_link_82575(&adapter->hw);
1525 }
1526
1527 /**
1528 * igb_up - Open the interface and prepare it to handle traffic
1529 * @adapter: board private structure
1530 **/
1531 int igb_up(struct igb_adapter *adapter)
1532 {
1533 struct e1000_hw *hw = &adapter->hw;
1534 int i;
1535
1536 /* hardware has been reset, we need to reload some things */
1537 igb_configure(adapter);
1538
1539 clear_bit(__IGB_DOWN, &adapter->state);
1540
1541 for (i = 0; i < adapter->num_q_vectors; i++)
1542 napi_enable(&(adapter->q_vector[i]->napi));
1543
1544 if (adapter->msix_entries)
1545 igb_configure_msix(adapter);
1546 else
1547 igb_assign_vector(adapter->q_vector[0], 0);
1548
1549 /* Clear any pending interrupts. */
1550 rd32(E1000_ICR);
1551 igb_irq_enable(adapter);
1552
1553 /* notify VFs that reset has been completed */
1554 if (adapter->vfs_allocated_count) {
1555 u32 reg_data = rd32(E1000_CTRL_EXT);
1556 reg_data |= E1000_CTRL_EXT_PFRSTD;
1557 wr32(E1000_CTRL_EXT, reg_data);
1558 }
1559
1560 netif_tx_start_all_queues(adapter->netdev);
1561
1562 /* start the watchdog. */
1563 hw->mac.get_link_status = 1;
1564 schedule_work(&adapter->watchdog_task);
1565
1566 return 0;
1567 }
1568
1569 void igb_down(struct igb_adapter *adapter)
1570 {
1571 struct net_device *netdev = adapter->netdev;
1572 struct e1000_hw *hw = &adapter->hw;
1573 u32 tctl, rctl;
1574 int i;
1575
1576 /* signal that we're down so the interrupt handler does not
1577 * reschedule our watchdog timer */
1578 set_bit(__IGB_DOWN, &adapter->state);
1579
1580 /* disable receives in the hardware */
1581 rctl = rd32(E1000_RCTL);
1582 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1583 /* flush and sleep below */
1584
1585 netif_tx_stop_all_queues(netdev);
1586
1587 /* disable transmits in the hardware */
1588 tctl = rd32(E1000_TCTL);
1589 tctl &= ~E1000_TCTL_EN;
1590 wr32(E1000_TCTL, tctl);
1591 /* flush both disables and wait for them to finish */
1592 wrfl();
1593 msleep(10);
1594
1595 for (i = 0; i < adapter->num_q_vectors; i++)
1596 napi_disable(&(adapter->q_vector[i]->napi));
1597
1598 igb_irq_disable(adapter);
1599
1600 del_timer_sync(&adapter->watchdog_timer);
1601 del_timer_sync(&adapter->phy_info_timer);
1602
1603 netif_carrier_off(netdev);
1604
1605 /* record the stats before reset*/
1606 spin_lock(&adapter->stats64_lock);
1607 igb_update_stats(adapter, &adapter->stats64);
1608 spin_unlock(&adapter->stats64_lock);
1609
1610 adapter->link_speed = 0;
1611 adapter->link_duplex = 0;
1612
1613 if (!pci_channel_offline(adapter->pdev))
1614 igb_reset(adapter);
1615 igb_clean_all_tx_rings(adapter);
1616 igb_clean_all_rx_rings(adapter);
1617 #ifdef CONFIG_IGB_DCA
1618
1619 /* since we reset the hardware DCA settings were cleared */
1620 igb_setup_dca(adapter);
1621 #endif
1622 }
1623
1624 void igb_reinit_locked(struct igb_adapter *adapter)
1625 {
1626 WARN_ON(in_interrupt());
1627 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1628 msleep(1);
1629 igb_down(adapter);
1630 igb_up(adapter);
1631 clear_bit(__IGB_RESETTING, &adapter->state);
1632 }
1633
1634 void igb_reset(struct igb_adapter *adapter)
1635 {
1636 struct pci_dev *pdev = adapter->pdev;
1637 struct e1000_hw *hw = &adapter->hw;
1638 struct e1000_mac_info *mac = &hw->mac;
1639 struct e1000_fc_info *fc = &hw->fc;
1640 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
1641 u16 hwm;
1642
1643 /* Repartition Pba for greater than 9k mtu
1644 * To take effect CTRL.RST is required.
1645 */
1646 switch (mac->type) {
1647 case e1000_i350:
1648 case e1000_82580:
1649 pba = rd32(E1000_RXPBS);
1650 pba = igb_rxpbs_adjust_82580(pba);
1651 break;
1652 case e1000_82576:
1653 pba = rd32(E1000_RXPBS);
1654 pba &= E1000_RXPBS_SIZE_MASK_82576;
1655 break;
1656 case e1000_82575:
1657 case e1000_i210:
1658 case e1000_i211:
1659 default:
1660 pba = E1000_PBA_34K;
1661 break;
1662 }
1663
1664 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1665 (mac->type < e1000_82576)) {
1666 /* adjust PBA for jumbo frames */
1667 wr32(E1000_PBA, pba);
1668
1669 /* To maintain wire speed transmits, the Tx FIFO should be
1670 * large enough to accommodate two full transmit packets,
1671 * rounded up to the next 1KB and expressed in KB. Likewise,
1672 * the Rx FIFO should be large enough to accommodate at least
1673 * one full receive packet and is similarly rounded up and
1674 * expressed in KB. */
1675 pba = rd32(E1000_PBA);
1676 /* upper 16 bits has Tx packet buffer allocation size in KB */
1677 tx_space = pba >> 16;
1678 /* lower 16 bits has Rx packet buffer allocation size in KB */
1679 pba &= 0xffff;
1680 /* the tx fifo also stores 16 bytes of information about the tx
1681 * but don't include ethernet FCS because hardware appends it */
1682 min_tx_space = (adapter->max_frame_size +
1683 sizeof(union e1000_adv_tx_desc) -
1684 ETH_FCS_LEN) * 2;
1685 min_tx_space = ALIGN(min_tx_space, 1024);
1686 min_tx_space >>= 10;
1687 /* software strips receive CRC, so leave room for it */
1688 min_rx_space = adapter->max_frame_size;
1689 min_rx_space = ALIGN(min_rx_space, 1024);
1690 min_rx_space >>= 10;
1691
1692 /* If current Tx allocation is less than the min Tx FIFO size,
1693 * and the min Tx FIFO size is less than the current Rx FIFO
1694 * allocation, take space away from current Rx allocation */
1695 if (tx_space < min_tx_space &&
1696 ((min_tx_space - tx_space) < pba)) {
1697 pba = pba - (min_tx_space - tx_space);
1698
1699 /* if short on rx space, rx wins and must trump tx
1700 * adjustment */
1701 if (pba < min_rx_space)
1702 pba = min_rx_space;
1703 }
1704 wr32(E1000_PBA, pba);
1705 }
1706
1707 /* flow control settings */
1708 /* The high water mark must be low enough to fit one full frame
1709 * (or the size used for early receive) above it in the Rx FIFO.
1710 * Set it to the lower of:
1711 * - 90% of the Rx FIFO size, or
1712 * - the full Rx FIFO size minus one full frame */
1713 hwm = min(((pba << 10) * 9 / 10),
1714 ((pba << 10) - 2 * adapter->max_frame_size));
1715
1716 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1717 fc->low_water = fc->high_water - 16;
1718 fc->pause_time = 0xFFFF;
1719 fc->send_xon = 1;
1720 fc->current_mode = fc->requested_mode;
1721
1722 /* disable receive for all VFs and wait one second */
1723 if (adapter->vfs_allocated_count) {
1724 int i;
1725 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1726 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1727
1728 /* ping all the active vfs to let them know we are going down */
1729 igb_ping_all_vfs(adapter);
1730
1731 /* disable transmits and receives */
1732 wr32(E1000_VFRE, 0);
1733 wr32(E1000_VFTE, 0);
1734 }
1735
1736 /* Allow time for pending master requests to run */
1737 hw->mac.ops.reset_hw(hw);
1738 wr32(E1000_WUC, 0);
1739
1740 if (hw->mac.ops.init_hw(hw))
1741 dev_err(&pdev->dev, "Hardware Error\n");
1742
1743 /*
1744 * Flow control settings reset on hardware reset, so guarantee flow
1745 * control is off when forcing speed.
1746 */
1747 if (!hw->mac.autoneg)
1748 igb_force_mac_fc(hw);
1749
1750 igb_init_dmac(adapter, pba);
1751 if (!netif_running(adapter->netdev))
1752 igb_power_down_link(adapter);
1753
1754 igb_update_mng_vlan(adapter);
1755
1756 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1757 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1758
1759 igb_get_phy_info(hw);
1760 }
1761
1762 static netdev_features_t igb_fix_features(struct net_device *netdev,
1763 netdev_features_t features)
1764 {
1765 /*
1766 * Since there is no support for separate rx/tx vlan accel
1767 * enable/disable make sure tx flag is always in same state as rx.
1768 */
1769 if (features & NETIF_F_HW_VLAN_RX)
1770 features |= NETIF_F_HW_VLAN_TX;
1771 else
1772 features &= ~NETIF_F_HW_VLAN_TX;
1773
1774 return features;
1775 }
1776
1777 static int igb_set_features(struct net_device *netdev,
1778 netdev_features_t features)
1779 {
1780 netdev_features_t changed = netdev->features ^ features;
1781 struct igb_adapter *adapter = netdev_priv(netdev);
1782
1783 if (changed & NETIF_F_HW_VLAN_RX)
1784 igb_vlan_mode(netdev, features);
1785
1786 if (!(changed & NETIF_F_RXALL))
1787 return 0;
1788
1789 netdev->features = features;
1790
1791 if (netif_running(netdev))
1792 igb_reinit_locked(adapter);
1793 else
1794 igb_reset(adapter);
1795
1796 return 0;
1797 }
1798
1799 static const struct net_device_ops igb_netdev_ops = {
1800 .ndo_open = igb_open,
1801 .ndo_stop = igb_close,
1802 .ndo_start_xmit = igb_xmit_frame,
1803 .ndo_get_stats64 = igb_get_stats64,
1804 .ndo_set_rx_mode = igb_set_rx_mode,
1805 .ndo_set_mac_address = igb_set_mac,
1806 .ndo_change_mtu = igb_change_mtu,
1807 .ndo_do_ioctl = igb_ioctl,
1808 .ndo_tx_timeout = igb_tx_timeout,
1809 .ndo_validate_addr = eth_validate_addr,
1810 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1811 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1812 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
1813 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
1814 .ndo_set_vf_tx_rate = igb_ndo_set_vf_bw,
1815 .ndo_get_vf_config = igb_ndo_get_vf_config,
1816 #ifdef CONFIG_NET_POLL_CONTROLLER
1817 .ndo_poll_controller = igb_netpoll,
1818 #endif
1819 .ndo_fix_features = igb_fix_features,
1820 .ndo_set_features = igb_set_features,
1821 };
1822
1823 /**
1824 * igb_probe - Device Initialization Routine
1825 * @pdev: PCI device information struct
1826 * @ent: entry in igb_pci_tbl
1827 *
1828 * Returns 0 on success, negative on failure
1829 *
1830 * igb_probe initializes an adapter identified by a pci_dev structure.
1831 * The OS initialization, configuring of the adapter private structure,
1832 * and a hardware reset occur.
1833 **/
1834 static int __devinit igb_probe(struct pci_dev *pdev,
1835 const struct pci_device_id *ent)
1836 {
1837 struct net_device *netdev;
1838 struct igb_adapter *adapter;
1839 struct e1000_hw *hw;
1840 u16 eeprom_data = 0;
1841 s32 ret_val;
1842 static int global_quad_port_a; /* global quad port a indication */
1843 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1844 unsigned long mmio_start, mmio_len;
1845 int err, pci_using_dac;
1846 u16 eeprom_apme_mask = IGB_EEPROM_APME;
1847 u8 part_str[E1000_PBANUM_LENGTH];
1848
1849 /* Catch broken hardware that put the wrong VF device ID in
1850 * the PCIe SR-IOV capability.
1851 */
1852 if (pdev->is_virtfn) {
1853 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
1854 pci_name(pdev), pdev->vendor, pdev->device);
1855 return -EINVAL;
1856 }
1857
1858 err = pci_enable_device_mem(pdev);
1859 if (err)
1860 return err;
1861
1862 pci_using_dac = 0;
1863 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
1864 if (!err) {
1865 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
1866 if (!err)
1867 pci_using_dac = 1;
1868 } else {
1869 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
1870 if (err) {
1871 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
1872 if (err) {
1873 dev_err(&pdev->dev, "No usable DMA "
1874 "configuration, aborting\n");
1875 goto err_dma;
1876 }
1877 }
1878 }
1879
1880 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1881 IORESOURCE_MEM),
1882 igb_driver_name);
1883 if (err)
1884 goto err_pci_reg;
1885
1886 pci_enable_pcie_error_reporting(pdev);
1887
1888 pci_set_master(pdev);
1889 pci_save_state(pdev);
1890
1891 err = -ENOMEM;
1892 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1893 IGB_MAX_TX_QUEUES);
1894 if (!netdev)
1895 goto err_alloc_etherdev;
1896
1897 SET_NETDEV_DEV(netdev, &pdev->dev);
1898
1899 pci_set_drvdata(pdev, netdev);
1900 adapter = netdev_priv(netdev);
1901 adapter->netdev = netdev;
1902 adapter->pdev = pdev;
1903 hw = &adapter->hw;
1904 hw->back = adapter;
1905 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
1906
1907 mmio_start = pci_resource_start(pdev, 0);
1908 mmio_len = pci_resource_len(pdev, 0);
1909
1910 err = -EIO;
1911 hw->hw_addr = ioremap(mmio_start, mmio_len);
1912 if (!hw->hw_addr)
1913 goto err_ioremap;
1914
1915 netdev->netdev_ops = &igb_netdev_ops;
1916 igb_set_ethtool_ops(netdev);
1917 netdev->watchdog_timeo = 5 * HZ;
1918
1919 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1920
1921 netdev->mem_start = mmio_start;
1922 netdev->mem_end = mmio_start + mmio_len;
1923
1924 /* PCI config space info */
1925 hw->vendor_id = pdev->vendor;
1926 hw->device_id = pdev->device;
1927 hw->revision_id = pdev->revision;
1928 hw->subsystem_vendor_id = pdev->subsystem_vendor;
1929 hw->subsystem_device_id = pdev->subsystem_device;
1930
1931 /* Copy the default MAC, PHY and NVM function pointers */
1932 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1933 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1934 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1935 /* Initialize skew-specific constants */
1936 err = ei->get_invariants(hw);
1937 if (err)
1938 goto err_sw_init;
1939
1940 /* setup the private structure */
1941 err = igb_sw_init(adapter);
1942 if (err)
1943 goto err_sw_init;
1944
1945 igb_get_bus_info_pcie(hw);
1946
1947 hw->phy.autoneg_wait_to_complete = false;
1948
1949 /* Copper options */
1950 if (hw->phy.media_type == e1000_media_type_copper) {
1951 hw->phy.mdix = AUTO_ALL_MODES;
1952 hw->phy.disable_polarity_correction = false;
1953 hw->phy.ms_type = e1000_ms_hw_default;
1954 }
1955
1956 if (igb_check_reset_block(hw))
1957 dev_info(&pdev->dev,
1958 "PHY reset is blocked due to SOL/IDER session.\n");
1959
1960 /*
1961 * features is initialized to 0 in allocation, it might have bits
1962 * set by igb_sw_init so we should use an or instead of an
1963 * assignment.
1964 */
1965 netdev->features |= NETIF_F_SG |
1966 NETIF_F_IP_CSUM |
1967 NETIF_F_IPV6_CSUM |
1968 NETIF_F_TSO |
1969 NETIF_F_TSO6 |
1970 NETIF_F_RXHASH |
1971 NETIF_F_RXCSUM |
1972 NETIF_F_HW_VLAN_RX |
1973 NETIF_F_HW_VLAN_TX;
1974
1975 /* copy netdev features into list of user selectable features */
1976 netdev->hw_features |= netdev->features;
1977 netdev->hw_features |= NETIF_F_RXALL;
1978
1979 /* set this bit last since it cannot be part of hw_features */
1980 netdev->features |= NETIF_F_HW_VLAN_FILTER;
1981
1982 netdev->vlan_features |= NETIF_F_TSO |
1983 NETIF_F_TSO6 |
1984 NETIF_F_IP_CSUM |
1985 NETIF_F_IPV6_CSUM |
1986 NETIF_F_SG;
1987
1988 netdev->priv_flags |= IFF_SUPP_NOFCS;
1989
1990 if (pci_using_dac) {
1991 netdev->features |= NETIF_F_HIGHDMA;
1992 netdev->vlan_features |= NETIF_F_HIGHDMA;
1993 }
1994
1995 if (hw->mac.type >= e1000_82576) {
1996 netdev->hw_features |= NETIF_F_SCTP_CSUM;
1997 netdev->features |= NETIF_F_SCTP_CSUM;
1998 }
1999
2000 netdev->priv_flags |= IFF_UNICAST_FLT;
2001
2002 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2003
2004 /* before reading the NVM, reset the controller to put the device in a
2005 * known good starting state */
2006 hw->mac.ops.reset_hw(hw);
2007
2008 /*
2009 * make sure the NVM is good , i211 parts have special NVM that
2010 * doesn't contain a checksum
2011 */
2012 if (hw->mac.type != e1000_i211) {
2013 if (hw->nvm.ops.validate(hw) < 0) {
2014 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2015 err = -EIO;
2016 goto err_eeprom;
2017 }
2018 }
2019
2020 /* copy the MAC address out of the NVM */
2021 if (hw->mac.ops.read_mac_addr(hw))
2022 dev_err(&pdev->dev, "NVM Read Error\n");
2023
2024 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2025 memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
2026
2027 if (!is_valid_ether_addr(netdev->perm_addr)) {
2028 dev_err(&pdev->dev, "Invalid MAC Address\n");
2029 err = -EIO;
2030 goto err_eeprom;
2031 }
2032
2033 setup_timer(&adapter->watchdog_timer, igb_watchdog,
2034 (unsigned long) adapter);
2035 setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2036 (unsigned long) adapter);
2037
2038 INIT_WORK(&adapter->reset_task, igb_reset_task);
2039 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2040
2041 /* Initialize link properties that are user-changeable */
2042 adapter->fc_autoneg = true;
2043 hw->mac.autoneg = true;
2044 hw->phy.autoneg_advertised = 0x2f;
2045
2046 hw->fc.requested_mode = e1000_fc_default;
2047 hw->fc.current_mode = e1000_fc_default;
2048
2049 igb_validate_mdi_setting(hw);
2050
2051 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
2052 * enable the ACPI Magic Packet filter
2053 */
2054
2055 if (hw->bus.func == 0)
2056 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
2057 else if (hw->mac.type >= e1000_82580)
2058 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2059 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2060 &eeprom_data);
2061 else if (hw->bus.func == 1)
2062 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2063
2064 if (eeprom_data & eeprom_apme_mask)
2065 adapter->eeprom_wol |= E1000_WUFC_MAG;
2066
2067 /* now that we have the eeprom settings, apply the special cases where
2068 * the eeprom may be wrong or the board simply won't support wake on
2069 * lan on a particular port */
2070 switch (pdev->device) {
2071 case E1000_DEV_ID_82575GB_QUAD_COPPER:
2072 adapter->eeprom_wol = 0;
2073 break;
2074 case E1000_DEV_ID_82575EB_FIBER_SERDES:
2075 case E1000_DEV_ID_82576_FIBER:
2076 case E1000_DEV_ID_82576_SERDES:
2077 /* Wake events only supported on port A for dual fiber
2078 * regardless of eeprom setting */
2079 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2080 adapter->eeprom_wol = 0;
2081 break;
2082 case E1000_DEV_ID_82576_QUAD_COPPER:
2083 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2084 /* if quad port adapter, disable WoL on all but port A */
2085 if (global_quad_port_a != 0)
2086 adapter->eeprom_wol = 0;
2087 else
2088 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2089 /* Reset for multiple quad port adapters */
2090 if (++global_quad_port_a == 4)
2091 global_quad_port_a = 0;
2092 break;
2093 }
2094
2095 /* initialize the wol settings based on the eeprom settings */
2096 adapter->wol = adapter->eeprom_wol;
2097 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
2098
2099 /* reset the hardware with the new settings */
2100 igb_reset(adapter);
2101
2102 /* let the f/w know that the h/w is now under the control of the
2103 * driver. */
2104 igb_get_hw_control(adapter);
2105
2106 strcpy(netdev->name, "eth%d");
2107 err = register_netdev(netdev);
2108 if (err)
2109 goto err_register;
2110
2111 /* carrier off reporting is important to ethtool even BEFORE open */
2112 netif_carrier_off(netdev);
2113
2114 #ifdef CONFIG_IGB_DCA
2115 if (dca_add_requester(&pdev->dev) == 0) {
2116 adapter->flags |= IGB_FLAG_DCA_ENABLED;
2117 dev_info(&pdev->dev, "DCA enabled\n");
2118 igb_setup_dca(adapter);
2119 }
2120
2121 #endif
2122 #ifdef CONFIG_IGB_PTP
2123 /* do hw tstamp init after resetting */
2124 igb_ptp_init(adapter);
2125
2126 #endif
2127 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2128 /* print bus type/speed/width info */
2129 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2130 netdev->name,
2131 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2132 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2133 "unknown"),
2134 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
2135 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
2136 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
2137 "unknown"),
2138 netdev->dev_addr);
2139
2140 ret_val = igb_read_part_string(hw, part_str, E1000_PBANUM_LENGTH);
2141 if (ret_val)
2142 strcpy(part_str, "Unknown");
2143 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2144 dev_info(&pdev->dev,
2145 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2146 adapter->msix_entries ? "MSI-X" :
2147 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2148 adapter->num_rx_queues, adapter->num_tx_queues);
2149 switch (hw->mac.type) {
2150 case e1000_i350:
2151 case e1000_i210:
2152 case e1000_i211:
2153 igb_set_eee_i350(hw);
2154 break;
2155 default:
2156 break;
2157 }
2158
2159 pm_runtime_put_noidle(&pdev->dev);
2160 return 0;
2161
2162 err_register:
2163 igb_release_hw_control(adapter);
2164 err_eeprom:
2165 if (!igb_check_reset_block(hw))
2166 igb_reset_phy(hw);
2167
2168 if (hw->flash_address)
2169 iounmap(hw->flash_address);
2170 err_sw_init:
2171 igb_clear_interrupt_scheme(adapter);
2172 iounmap(hw->hw_addr);
2173 err_ioremap:
2174 free_netdev(netdev);
2175 err_alloc_etherdev:
2176 pci_release_selected_regions(pdev,
2177 pci_select_bars(pdev, IORESOURCE_MEM));
2178 err_pci_reg:
2179 err_dma:
2180 pci_disable_device(pdev);
2181 return err;
2182 }
2183
2184 /**
2185 * igb_remove - Device Removal Routine
2186 * @pdev: PCI device information struct
2187 *
2188 * igb_remove is called by the PCI subsystem to alert the driver
2189 * that it should release a PCI device. The could be caused by a
2190 * Hot-Plug event, or because the driver is going to be removed from
2191 * memory.
2192 **/
2193 static void __devexit igb_remove(struct pci_dev *pdev)
2194 {
2195 struct net_device *netdev = pci_get_drvdata(pdev);
2196 struct igb_adapter *adapter = netdev_priv(netdev);
2197 struct e1000_hw *hw = &adapter->hw;
2198
2199 pm_runtime_get_noresume(&pdev->dev);
2200 #ifdef CONFIG_IGB_PTP
2201 igb_ptp_remove(adapter);
2202
2203 #endif
2204 /*
2205 * The watchdog timer may be rescheduled, so explicitly
2206 * disable watchdog from being rescheduled.
2207 */
2208 set_bit(__IGB_DOWN, &adapter->state);
2209 del_timer_sync(&adapter->watchdog_timer);
2210 del_timer_sync(&adapter->phy_info_timer);
2211
2212 cancel_work_sync(&adapter->reset_task);
2213 cancel_work_sync(&adapter->watchdog_task);
2214
2215 #ifdef CONFIG_IGB_DCA
2216 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2217 dev_info(&pdev->dev, "DCA disabled\n");
2218 dca_remove_requester(&pdev->dev);
2219 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2220 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2221 }
2222 #endif
2223
2224 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2225 * would have already happened in close and is redundant. */
2226 igb_release_hw_control(adapter);
2227
2228 unregister_netdev(netdev);
2229
2230 igb_clear_interrupt_scheme(adapter);
2231
2232 #ifdef CONFIG_PCI_IOV
2233 /* reclaim resources allocated to VFs */
2234 if (adapter->vf_data) {
2235 /* disable iov and allow time for transactions to clear */
2236 if (!igb_check_vf_assignment(adapter)) {
2237 pci_disable_sriov(pdev);
2238 msleep(500);
2239 } else {
2240 dev_info(&pdev->dev, "VF(s) assigned to guests!\n");
2241 }
2242
2243 kfree(adapter->vf_data);
2244 adapter->vf_data = NULL;
2245 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2246 wrfl();
2247 msleep(100);
2248 dev_info(&pdev->dev, "IOV Disabled\n");
2249 }
2250 #endif
2251
2252 iounmap(hw->hw_addr);
2253 if (hw->flash_address)
2254 iounmap(hw->flash_address);
2255 pci_release_selected_regions(pdev,
2256 pci_select_bars(pdev, IORESOURCE_MEM));
2257
2258 kfree(adapter->shadow_vfta);
2259 free_netdev(netdev);
2260
2261 pci_disable_pcie_error_reporting(pdev);
2262
2263 pci_disable_device(pdev);
2264 }
2265
2266 /**
2267 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2268 * @adapter: board private structure to initialize
2269 *
2270 * This function initializes the vf specific data storage and then attempts to
2271 * allocate the VFs. The reason for ordering it this way is because it is much
2272 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2273 * the memory for the VFs.
2274 **/
2275 static void __devinit igb_probe_vfs(struct igb_adapter * adapter)
2276 {
2277 #ifdef CONFIG_PCI_IOV
2278 struct pci_dev *pdev = adapter->pdev;
2279 struct e1000_hw *hw = &adapter->hw;
2280 int old_vfs = igb_find_enabled_vfs(adapter);
2281 int i;
2282
2283 /* Virtualization features not supported on i210 family. */
2284 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
2285 return;
2286
2287 if (old_vfs) {
2288 dev_info(&pdev->dev, "%d pre-allocated VFs found - override "
2289 "max_vfs setting of %d\n", old_vfs, max_vfs);
2290 adapter->vfs_allocated_count = old_vfs;
2291 }
2292
2293 if (!adapter->vfs_allocated_count)
2294 return;
2295
2296 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2297 sizeof(struct vf_data_storage), GFP_KERNEL);
2298
2299 /* if allocation failed then we do not support SR-IOV */
2300 if (!adapter->vf_data) {
2301 adapter->vfs_allocated_count = 0;
2302 dev_err(&pdev->dev, "Unable to allocate memory for VF "
2303 "Data Storage\n");
2304 goto out;
2305 }
2306
2307 if (!old_vfs) {
2308 if (pci_enable_sriov(pdev, adapter->vfs_allocated_count))
2309 goto err_out;
2310 }
2311 dev_info(&pdev->dev, "%d VFs allocated\n",
2312 adapter->vfs_allocated_count);
2313 for (i = 0; i < adapter->vfs_allocated_count; i++)
2314 igb_vf_configure(adapter, i);
2315
2316 /* DMA Coalescing is not supported in IOV mode. */
2317 adapter->flags &= ~IGB_FLAG_DMAC;
2318 goto out;
2319 err_out:
2320 kfree(adapter->vf_data);
2321 adapter->vf_data = NULL;
2322 adapter->vfs_allocated_count = 0;
2323 out:
2324 return;
2325 #endif /* CONFIG_PCI_IOV */
2326 }
2327
2328 /**
2329 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2330 * @adapter: board private structure to initialize
2331 *
2332 * igb_sw_init initializes the Adapter private data structure.
2333 * Fields are initialized based on PCI device information and
2334 * OS network device settings (MTU size).
2335 **/
2336 static int __devinit igb_sw_init(struct igb_adapter *adapter)
2337 {
2338 struct e1000_hw *hw = &adapter->hw;
2339 struct net_device *netdev = adapter->netdev;
2340 struct pci_dev *pdev = adapter->pdev;
2341
2342 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2343
2344 /* set default ring sizes */
2345 adapter->tx_ring_count = IGB_DEFAULT_TXD;
2346 adapter->rx_ring_count = IGB_DEFAULT_RXD;
2347
2348 /* set default ITR values */
2349 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2350 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2351
2352 /* set default work limits */
2353 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2354
2355 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2356 VLAN_HLEN;
2357 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2358
2359 adapter->node = -1;
2360
2361 spin_lock_init(&adapter->stats64_lock);
2362 #ifdef CONFIG_PCI_IOV
2363 switch (hw->mac.type) {
2364 case e1000_82576:
2365 case e1000_i350:
2366 if (max_vfs > 7) {
2367 dev_warn(&pdev->dev,
2368 "Maximum of 7 VFs per PF, using max\n");
2369 adapter->vfs_allocated_count = 7;
2370 } else
2371 adapter->vfs_allocated_count = max_vfs;
2372 break;
2373 case e1000_i210:
2374 case e1000_i211:
2375 adapter->vfs_allocated_count = 0;
2376 break;
2377 default:
2378 break;
2379 }
2380 #endif /* CONFIG_PCI_IOV */
2381 switch (hw->mac.type) {
2382 case e1000_i210:
2383 adapter->rss_queues = min_t(u32, IGB_MAX_RX_QUEUES_I210,
2384 num_online_cpus());
2385 break;
2386 case e1000_i211:
2387 adapter->rss_queues = min_t(u32, IGB_MAX_RX_QUEUES_I211,
2388 num_online_cpus());
2389 break;
2390 default:
2391 adapter->rss_queues = min_t(u32, IGB_MAX_RX_QUEUES,
2392 num_online_cpus());
2393 break;
2394 }
2395 /* i350 cannot do RSS and SR-IOV at the same time */
2396 if (hw->mac.type == e1000_i350 && adapter->vfs_allocated_count)
2397 adapter->rss_queues = 1;
2398
2399 /*
2400 * if rss_queues > 4 or vfs are going to be allocated with rss_queues
2401 * then we should combine the queues into a queue pair in order to
2402 * conserve interrupts due to limited supply
2403 */
2404 if ((adapter->rss_queues > 4) ||
2405 ((adapter->rss_queues > 1) && (adapter->vfs_allocated_count > 6)))
2406 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2407
2408 /* Setup and initialize a copy of the hw vlan table array */
2409 adapter->shadow_vfta = kzalloc(sizeof(u32) *
2410 E1000_VLAN_FILTER_TBL_SIZE,
2411 GFP_ATOMIC);
2412
2413 /* This call may decrease the number of queues */
2414 if (igb_init_interrupt_scheme(adapter)) {
2415 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2416 return -ENOMEM;
2417 }
2418
2419 igb_probe_vfs(adapter);
2420
2421 /* Explicitly disable IRQ since the NIC can be in any state. */
2422 igb_irq_disable(adapter);
2423
2424 if (hw->mac.type >= e1000_i350)
2425 adapter->flags &= ~IGB_FLAG_DMAC;
2426
2427 set_bit(__IGB_DOWN, &adapter->state);
2428 return 0;
2429 }
2430
2431 /**
2432 * igb_open - Called when a network interface is made active
2433 * @netdev: network interface device structure
2434 *
2435 * Returns 0 on success, negative value on failure
2436 *
2437 * The open entry point is called when a network interface is made
2438 * active by the system (IFF_UP). At this point all resources needed
2439 * for transmit and receive operations are allocated, the interrupt
2440 * handler is registered with the OS, the watchdog timer is started,
2441 * and the stack is notified that the interface is ready.
2442 **/
2443 static int __igb_open(struct net_device *netdev, bool resuming)
2444 {
2445 struct igb_adapter *adapter = netdev_priv(netdev);
2446 struct e1000_hw *hw = &adapter->hw;
2447 struct pci_dev *pdev = adapter->pdev;
2448 int err;
2449 int i;
2450
2451 /* disallow open during test */
2452 if (test_bit(__IGB_TESTING, &adapter->state)) {
2453 WARN_ON(resuming);
2454 return -EBUSY;
2455 }
2456
2457 if (!resuming)
2458 pm_runtime_get_sync(&pdev->dev);
2459
2460 netif_carrier_off(netdev);
2461
2462 /* allocate transmit descriptors */
2463 err = igb_setup_all_tx_resources(adapter);
2464 if (err)
2465 goto err_setup_tx;
2466
2467 /* allocate receive descriptors */
2468 err = igb_setup_all_rx_resources(adapter);
2469 if (err)
2470 goto err_setup_rx;
2471
2472 igb_power_up_link(adapter);
2473
2474 /* before we allocate an interrupt, we must be ready to handle it.
2475 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2476 * as soon as we call pci_request_irq, so we have to setup our
2477 * clean_rx handler before we do so. */
2478 igb_configure(adapter);
2479
2480 err = igb_request_irq(adapter);
2481 if (err)
2482 goto err_req_irq;
2483
2484 /* From here on the code is the same as igb_up() */
2485 clear_bit(__IGB_DOWN, &adapter->state);
2486
2487 for (i = 0; i < adapter->num_q_vectors; i++)
2488 napi_enable(&(adapter->q_vector[i]->napi));
2489
2490 /* Clear any pending interrupts. */
2491 rd32(E1000_ICR);
2492
2493 igb_irq_enable(adapter);
2494
2495 /* notify VFs that reset has been completed */
2496 if (adapter->vfs_allocated_count) {
2497 u32 reg_data = rd32(E1000_CTRL_EXT);
2498 reg_data |= E1000_CTRL_EXT_PFRSTD;
2499 wr32(E1000_CTRL_EXT, reg_data);
2500 }
2501
2502 netif_tx_start_all_queues(netdev);
2503
2504 if (!resuming)
2505 pm_runtime_put(&pdev->dev);
2506
2507 /* start the watchdog. */
2508 hw->mac.get_link_status = 1;
2509 schedule_work(&adapter->watchdog_task);
2510
2511 return 0;
2512
2513 err_req_irq:
2514 igb_release_hw_control(adapter);
2515 igb_power_down_link(adapter);
2516 igb_free_all_rx_resources(adapter);
2517 err_setup_rx:
2518 igb_free_all_tx_resources(adapter);
2519 err_setup_tx:
2520 igb_reset(adapter);
2521 if (!resuming)
2522 pm_runtime_put(&pdev->dev);
2523
2524 return err;
2525 }
2526
2527 static int igb_open(struct net_device *netdev)
2528 {
2529 return __igb_open(netdev, false);
2530 }
2531
2532 /**
2533 * igb_close - Disables a network interface
2534 * @netdev: network interface device structure
2535 *
2536 * Returns 0, this is not allowed to fail
2537 *
2538 * The close entry point is called when an interface is de-activated
2539 * by the OS. The hardware is still under the driver's control, but
2540 * needs to be disabled. A global MAC reset is issued to stop the
2541 * hardware, and all transmit and receive resources are freed.
2542 **/
2543 static int __igb_close(struct net_device *netdev, bool suspending)
2544 {
2545 struct igb_adapter *adapter = netdev_priv(netdev);
2546 struct pci_dev *pdev = adapter->pdev;
2547
2548 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
2549
2550 if (!suspending)
2551 pm_runtime_get_sync(&pdev->dev);
2552
2553 igb_down(adapter);
2554 igb_free_irq(adapter);
2555
2556 igb_free_all_tx_resources(adapter);
2557 igb_free_all_rx_resources(adapter);
2558
2559 if (!suspending)
2560 pm_runtime_put_sync(&pdev->dev);
2561 return 0;
2562 }
2563
2564 static int igb_close(struct net_device *netdev)
2565 {
2566 return __igb_close(netdev, false);
2567 }
2568
2569 /**
2570 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2571 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2572 *
2573 * Return 0 on success, negative on failure
2574 **/
2575 int igb_setup_tx_resources(struct igb_ring *tx_ring)
2576 {
2577 struct device *dev = tx_ring->dev;
2578 int orig_node = dev_to_node(dev);
2579 int size;
2580
2581 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
2582 tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node);
2583 if (!tx_ring->tx_buffer_info)
2584 tx_ring->tx_buffer_info = vzalloc(size);
2585 if (!tx_ring->tx_buffer_info)
2586 goto err;
2587
2588 /* round up to nearest 4K */
2589 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
2590 tx_ring->size = ALIGN(tx_ring->size, 4096);
2591
2592 set_dev_node(dev, tx_ring->numa_node);
2593 tx_ring->desc = dma_alloc_coherent(dev,
2594 tx_ring->size,
2595 &tx_ring->dma,
2596 GFP_KERNEL);
2597 set_dev_node(dev, orig_node);
2598 if (!tx_ring->desc)
2599 tx_ring->desc = dma_alloc_coherent(dev,
2600 tx_ring->size,
2601 &tx_ring->dma,
2602 GFP_KERNEL);
2603
2604 if (!tx_ring->desc)
2605 goto err;
2606
2607 tx_ring->next_to_use = 0;
2608 tx_ring->next_to_clean = 0;
2609
2610 return 0;
2611
2612 err:
2613 vfree(tx_ring->tx_buffer_info);
2614 dev_err(dev,
2615 "Unable to allocate memory for the transmit descriptor ring\n");
2616 return -ENOMEM;
2617 }
2618
2619 /**
2620 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2621 * (Descriptors) for all queues
2622 * @adapter: board private structure
2623 *
2624 * Return 0 on success, negative on failure
2625 **/
2626 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2627 {
2628 struct pci_dev *pdev = adapter->pdev;
2629 int i, err = 0;
2630
2631 for (i = 0; i < adapter->num_tx_queues; i++) {
2632 err = igb_setup_tx_resources(adapter->tx_ring[i]);
2633 if (err) {
2634 dev_err(&pdev->dev,
2635 "Allocation for Tx Queue %u failed\n", i);
2636 for (i--; i >= 0; i--)
2637 igb_free_tx_resources(adapter->tx_ring[i]);
2638 break;
2639 }
2640 }
2641
2642 return err;
2643 }
2644
2645 /**
2646 * igb_setup_tctl - configure the transmit control registers
2647 * @adapter: Board private structure
2648 **/
2649 void igb_setup_tctl(struct igb_adapter *adapter)
2650 {
2651 struct e1000_hw *hw = &adapter->hw;
2652 u32 tctl;
2653
2654 /* disable queue 0 which is enabled by default on 82575 and 82576 */
2655 wr32(E1000_TXDCTL(0), 0);
2656
2657 /* Program the Transmit Control Register */
2658 tctl = rd32(E1000_TCTL);
2659 tctl &= ~E1000_TCTL_CT;
2660 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2661 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2662
2663 igb_config_collision_dist(hw);
2664
2665 /* Enable transmits */
2666 tctl |= E1000_TCTL_EN;
2667
2668 wr32(E1000_TCTL, tctl);
2669 }
2670
2671 /**
2672 * igb_configure_tx_ring - Configure transmit ring after Reset
2673 * @adapter: board private structure
2674 * @ring: tx ring to configure
2675 *
2676 * Configure a transmit ring after a reset.
2677 **/
2678 void igb_configure_tx_ring(struct igb_adapter *adapter,
2679 struct igb_ring *ring)
2680 {
2681 struct e1000_hw *hw = &adapter->hw;
2682 u32 txdctl = 0;
2683 u64 tdba = ring->dma;
2684 int reg_idx = ring->reg_idx;
2685
2686 /* disable the queue */
2687 wr32(E1000_TXDCTL(reg_idx), 0);
2688 wrfl();
2689 mdelay(10);
2690
2691 wr32(E1000_TDLEN(reg_idx),
2692 ring->count * sizeof(union e1000_adv_tx_desc));
2693 wr32(E1000_TDBAL(reg_idx),
2694 tdba & 0x00000000ffffffffULL);
2695 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
2696
2697 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
2698 wr32(E1000_TDH(reg_idx), 0);
2699 writel(0, ring->tail);
2700
2701 txdctl |= IGB_TX_PTHRESH;
2702 txdctl |= IGB_TX_HTHRESH << 8;
2703 txdctl |= IGB_TX_WTHRESH << 16;
2704
2705 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
2706 wr32(E1000_TXDCTL(reg_idx), txdctl);
2707 }
2708
2709 /**
2710 * igb_configure_tx - Configure transmit Unit after Reset
2711 * @adapter: board private structure
2712 *
2713 * Configure the Tx unit of the MAC after a reset.
2714 **/
2715 static void igb_configure_tx(struct igb_adapter *adapter)
2716 {
2717 int i;
2718
2719 for (i = 0; i < adapter->num_tx_queues; i++)
2720 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
2721 }
2722
2723 /**
2724 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
2725 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2726 *
2727 * Returns 0 on success, negative on failure
2728 **/
2729 int igb_setup_rx_resources(struct igb_ring *rx_ring)
2730 {
2731 struct device *dev = rx_ring->dev;
2732 int orig_node = dev_to_node(dev);
2733 int size, desc_len;
2734
2735 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
2736 rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node);
2737 if (!rx_ring->rx_buffer_info)
2738 rx_ring->rx_buffer_info = vzalloc(size);
2739 if (!rx_ring->rx_buffer_info)
2740 goto err;
2741
2742 desc_len = sizeof(union e1000_adv_rx_desc);
2743
2744 /* Round up to nearest 4K */
2745 rx_ring->size = rx_ring->count * desc_len;
2746 rx_ring->size = ALIGN(rx_ring->size, 4096);
2747
2748 set_dev_node(dev, rx_ring->numa_node);
2749 rx_ring->desc = dma_alloc_coherent(dev,
2750 rx_ring->size,
2751 &rx_ring->dma,
2752 GFP_KERNEL);
2753 set_dev_node(dev, orig_node);
2754 if (!rx_ring->desc)
2755 rx_ring->desc = dma_alloc_coherent(dev,
2756 rx_ring->size,
2757 &rx_ring->dma,
2758 GFP_KERNEL);
2759
2760 if (!rx_ring->desc)
2761 goto err;
2762
2763 rx_ring->next_to_clean = 0;
2764 rx_ring->next_to_use = 0;
2765
2766 return 0;
2767
2768 err:
2769 vfree(rx_ring->rx_buffer_info);
2770 rx_ring->rx_buffer_info = NULL;
2771 dev_err(dev, "Unable to allocate memory for the receive descriptor"
2772 " ring\n");
2773 return -ENOMEM;
2774 }
2775
2776 /**
2777 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
2778 * (Descriptors) for all queues
2779 * @adapter: board private structure
2780 *
2781 * Return 0 on success, negative on failure
2782 **/
2783 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
2784 {
2785 struct pci_dev *pdev = adapter->pdev;
2786 int i, err = 0;
2787
2788 for (i = 0; i < adapter->num_rx_queues; i++) {
2789 err = igb_setup_rx_resources(adapter->rx_ring[i]);
2790 if (err) {
2791 dev_err(&pdev->dev,
2792 "Allocation for Rx Queue %u failed\n", i);
2793 for (i--; i >= 0; i--)
2794 igb_free_rx_resources(adapter->rx_ring[i]);
2795 break;
2796 }
2797 }
2798
2799 return err;
2800 }
2801
2802 /**
2803 * igb_setup_mrqc - configure the multiple receive queue control registers
2804 * @adapter: Board private structure
2805 **/
2806 static void igb_setup_mrqc(struct igb_adapter *adapter)
2807 {
2808 struct e1000_hw *hw = &adapter->hw;
2809 u32 mrqc, rxcsum;
2810 u32 j, num_rx_queues, shift = 0, shift2 = 0;
2811 union e1000_reta {
2812 u32 dword;
2813 u8 bytes[4];
2814 } reta;
2815 static const u8 rsshash[40] = {
2816 0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2, 0x41, 0x67,
2817 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0, 0xd0, 0xca, 0x2b, 0xcb,
2818 0xae, 0x7b, 0x30, 0xb4, 0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30,
2819 0xf2, 0x0c, 0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa };
2820
2821 /* Fill out hash function seeds */
2822 for (j = 0; j < 10; j++) {
2823 u32 rsskey = rsshash[(j * 4)];
2824 rsskey |= rsshash[(j * 4) + 1] << 8;
2825 rsskey |= rsshash[(j * 4) + 2] << 16;
2826 rsskey |= rsshash[(j * 4) + 3] << 24;
2827 array_wr32(E1000_RSSRK(0), j, rsskey);
2828 }
2829
2830 num_rx_queues = adapter->rss_queues;
2831
2832 if (adapter->vfs_allocated_count) {
2833 /* 82575 and 82576 supports 2 RSS queues for VMDq */
2834 switch (hw->mac.type) {
2835 case e1000_i350:
2836 case e1000_82580:
2837 num_rx_queues = 1;
2838 shift = 0;
2839 break;
2840 case e1000_82576:
2841 shift = 3;
2842 num_rx_queues = 2;
2843 break;
2844 case e1000_82575:
2845 shift = 2;
2846 shift2 = 6;
2847 default:
2848 break;
2849 }
2850 } else {
2851 if (hw->mac.type == e1000_82575)
2852 shift = 6;
2853 }
2854
2855 for (j = 0; j < (32 * 4); j++) {
2856 reta.bytes[j & 3] = (j % num_rx_queues) << shift;
2857 if (shift2)
2858 reta.bytes[j & 3] |= num_rx_queues << shift2;
2859 if ((j & 3) == 3)
2860 wr32(E1000_RETA(j >> 2), reta.dword);
2861 }
2862
2863 /*
2864 * Disable raw packet checksumming so that RSS hash is placed in
2865 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
2866 * offloads as they are enabled by default
2867 */
2868 rxcsum = rd32(E1000_RXCSUM);
2869 rxcsum |= E1000_RXCSUM_PCSD;
2870
2871 if (adapter->hw.mac.type >= e1000_82576)
2872 /* Enable Receive Checksum Offload for SCTP */
2873 rxcsum |= E1000_RXCSUM_CRCOFL;
2874
2875 /* Don't need to set TUOFL or IPOFL, they default to 1 */
2876 wr32(E1000_RXCSUM, rxcsum);
2877 /*
2878 * Generate RSS hash based on TCP port numbers and/or
2879 * IPv4/v6 src and dst addresses since UDP cannot be
2880 * hashed reliably due to IP fragmentation
2881 */
2882
2883 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
2884 E1000_MRQC_RSS_FIELD_IPV4_TCP |
2885 E1000_MRQC_RSS_FIELD_IPV6 |
2886 E1000_MRQC_RSS_FIELD_IPV6_TCP |
2887 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
2888
2889 /* If VMDq is enabled then we set the appropriate mode for that, else
2890 * we default to RSS so that an RSS hash is calculated per packet even
2891 * if we are only using one queue */
2892 if (adapter->vfs_allocated_count) {
2893 if (hw->mac.type > e1000_82575) {
2894 /* Set the default pool for the PF's first queue */
2895 u32 vtctl = rd32(E1000_VT_CTL);
2896 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2897 E1000_VT_CTL_DISABLE_DEF_POOL);
2898 vtctl |= adapter->vfs_allocated_count <<
2899 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2900 wr32(E1000_VT_CTL, vtctl);
2901 }
2902 if (adapter->rss_queues > 1)
2903 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2904 else
2905 mrqc |= E1000_MRQC_ENABLE_VMDQ;
2906 } else {
2907 if (hw->mac.type != e1000_i211)
2908 mrqc |= E1000_MRQC_ENABLE_RSS_4Q;
2909 }
2910 igb_vmm_control(adapter);
2911
2912 wr32(E1000_MRQC, mrqc);
2913 }
2914
2915 /**
2916 * igb_setup_rctl - configure the receive control registers
2917 * @adapter: Board private structure
2918 **/
2919 void igb_setup_rctl(struct igb_adapter *adapter)
2920 {
2921 struct e1000_hw *hw = &adapter->hw;
2922 u32 rctl;
2923
2924 rctl = rd32(E1000_RCTL);
2925
2926 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2927 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2928
2929 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2930 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2931
2932 /*
2933 * enable stripping of CRC. It's unlikely this will break BMC
2934 * redirection as it did with e1000. Newer features require
2935 * that the HW strips the CRC.
2936 */
2937 rctl |= E1000_RCTL_SECRC;
2938
2939 /* disable store bad packets and clear size bits. */
2940 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2941
2942 /* enable LPE to prevent packets larger than max_frame_size */
2943 rctl |= E1000_RCTL_LPE;
2944
2945 /* disable queue 0 to prevent tail write w/o re-config */
2946 wr32(E1000_RXDCTL(0), 0);
2947
2948 /* Attention!!! For SR-IOV PF driver operations you must enable
2949 * queue drop for all VF and PF queues to prevent head of line blocking
2950 * if an un-trusted VF does not provide descriptors to hardware.
2951 */
2952 if (adapter->vfs_allocated_count) {
2953 /* set all queue drop enable bits */
2954 wr32(E1000_QDE, ALL_QUEUES);
2955 }
2956
2957 /* This is useful for sniffing bad packets. */
2958 if (adapter->netdev->features & NETIF_F_RXALL) {
2959 /* UPE and MPE will be handled by normal PROMISC logic
2960 * in e1000e_set_rx_mode */
2961 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
2962 E1000_RCTL_BAM | /* RX All Bcast Pkts */
2963 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
2964
2965 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
2966 E1000_RCTL_DPF | /* Allow filtered pause */
2967 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
2968 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
2969 * and that breaks VLANs.
2970 */
2971 }
2972
2973 wr32(E1000_RCTL, rctl);
2974 }
2975
2976 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
2977 int vfn)
2978 {
2979 struct e1000_hw *hw = &adapter->hw;
2980 u32 vmolr;
2981
2982 /* if it isn't the PF check to see if VFs are enabled and
2983 * increase the size to support vlan tags */
2984 if (vfn < adapter->vfs_allocated_count &&
2985 adapter->vf_data[vfn].vlans_enabled)
2986 size += VLAN_TAG_SIZE;
2987
2988 vmolr = rd32(E1000_VMOLR(vfn));
2989 vmolr &= ~E1000_VMOLR_RLPML_MASK;
2990 vmolr |= size | E1000_VMOLR_LPE;
2991 wr32(E1000_VMOLR(vfn), vmolr);
2992
2993 return 0;
2994 }
2995
2996 /**
2997 * igb_rlpml_set - set maximum receive packet size
2998 * @adapter: board private structure
2999 *
3000 * Configure maximum receivable packet size.
3001 **/
3002 static void igb_rlpml_set(struct igb_adapter *adapter)
3003 {
3004 u32 max_frame_size = adapter->max_frame_size;
3005 struct e1000_hw *hw = &adapter->hw;
3006 u16 pf_id = adapter->vfs_allocated_count;
3007
3008 if (pf_id) {
3009 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3010 /*
3011 * If we're in VMDQ or SR-IOV mode, then set global RLPML
3012 * to our max jumbo frame size, in case we need to enable
3013 * jumbo frames on one of the rings later.
3014 * This will not pass over-length frames into the default
3015 * queue because it's gated by the VMOLR.RLPML.
3016 */
3017 max_frame_size = MAX_JUMBO_FRAME_SIZE;
3018 }
3019
3020 wr32(E1000_RLPML, max_frame_size);
3021 }
3022
3023 static inline void igb_set_vmolr(struct igb_adapter *adapter,
3024 int vfn, bool aupe)
3025 {
3026 struct e1000_hw *hw = &adapter->hw;
3027 u32 vmolr;
3028
3029 /*
3030 * This register exists only on 82576 and newer so if we are older then
3031 * we should exit and do nothing
3032 */
3033 if (hw->mac.type < e1000_82576)
3034 return;
3035
3036 vmolr = rd32(E1000_VMOLR(vfn));
3037 vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
3038 if (aupe)
3039 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3040 else
3041 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3042
3043 /* clear all bits that might not be set */
3044 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3045
3046 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3047 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3048 /*
3049 * for VMDq only allow the VFs and pool 0 to accept broadcast and
3050 * multicast packets
3051 */
3052 if (vfn <= adapter->vfs_allocated_count)
3053 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3054
3055 wr32(E1000_VMOLR(vfn), vmolr);
3056 }
3057
3058 /**
3059 * igb_configure_rx_ring - Configure a receive ring after Reset
3060 * @adapter: board private structure
3061 * @ring: receive ring to be configured
3062 *
3063 * Configure the Rx unit of the MAC after a reset.
3064 **/
3065 void igb_configure_rx_ring(struct igb_adapter *adapter,
3066 struct igb_ring *ring)
3067 {
3068 struct e1000_hw *hw = &adapter->hw;
3069 u64 rdba = ring->dma;
3070 int reg_idx = ring->reg_idx;
3071 u32 srrctl = 0, rxdctl = 0;
3072
3073 /* disable the queue */
3074 wr32(E1000_RXDCTL(reg_idx), 0);
3075
3076 /* Set DMA base address registers */
3077 wr32(E1000_RDBAL(reg_idx),
3078 rdba & 0x00000000ffffffffULL);
3079 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3080 wr32(E1000_RDLEN(reg_idx),
3081 ring->count * sizeof(union e1000_adv_rx_desc));
3082
3083 /* initialize head and tail */
3084 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3085 wr32(E1000_RDH(reg_idx), 0);
3086 writel(0, ring->tail);
3087
3088 /* set descriptor configuration */
3089 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3090 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3091 srrctl |= IGB_RXBUFFER_16384 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3092 #else
3093 srrctl |= (PAGE_SIZE / 2) >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3094 #endif
3095 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
3096 if (hw->mac.type >= e1000_82580)
3097 srrctl |= E1000_SRRCTL_TIMESTAMP;
3098 /* Only set Drop Enable if we are supporting multiple queues */
3099 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3100 srrctl |= E1000_SRRCTL_DROP_EN;
3101
3102 wr32(E1000_SRRCTL(reg_idx), srrctl);
3103
3104 /* set filtering for VMDQ pools */
3105 igb_set_vmolr(adapter, reg_idx & 0x7, true);
3106
3107 rxdctl |= IGB_RX_PTHRESH;
3108 rxdctl |= IGB_RX_HTHRESH << 8;
3109 rxdctl |= IGB_RX_WTHRESH << 16;
3110
3111 /* enable receive descriptor fetching */
3112 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3113 wr32(E1000_RXDCTL(reg_idx), rxdctl);
3114 }
3115
3116 /**
3117 * igb_configure_rx - Configure receive Unit after Reset
3118 * @adapter: board private structure
3119 *
3120 * Configure the Rx unit of the MAC after a reset.
3121 **/
3122 static void igb_configure_rx(struct igb_adapter *adapter)
3123 {
3124 int i;
3125
3126 /* set UTA to appropriate mode */
3127 igb_set_uta(adapter);
3128
3129 /* set the correct pool for the PF default MAC address in entry 0 */
3130 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3131 adapter->vfs_allocated_count);
3132
3133 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3134 * the Base and Length of the Rx Descriptor Ring */
3135 for (i = 0; i < adapter->num_rx_queues; i++)
3136 igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
3137 }
3138
3139 /**
3140 * igb_free_tx_resources - Free Tx Resources per Queue
3141 * @tx_ring: Tx descriptor ring for a specific queue
3142 *
3143 * Free all transmit software resources
3144 **/
3145 void igb_free_tx_resources(struct igb_ring *tx_ring)
3146 {
3147 igb_clean_tx_ring(tx_ring);
3148
3149 vfree(tx_ring->tx_buffer_info);
3150 tx_ring->tx_buffer_info = NULL;
3151
3152 /* if not set, then don't free */
3153 if (!tx_ring->desc)
3154 return;
3155
3156 dma_free_coherent(tx_ring->dev, tx_ring->size,
3157 tx_ring->desc, tx_ring->dma);
3158
3159 tx_ring->desc = NULL;
3160 }
3161
3162 /**
3163 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3164 * @adapter: board private structure
3165 *
3166 * Free all transmit software resources
3167 **/
3168 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3169 {
3170 int i;
3171
3172 for (i = 0; i < adapter->num_tx_queues; i++)
3173 igb_free_tx_resources(adapter->tx_ring[i]);
3174 }
3175
3176 void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3177 struct igb_tx_buffer *tx_buffer)
3178 {
3179 if (tx_buffer->skb) {
3180 dev_kfree_skb_any(tx_buffer->skb);
3181 if (tx_buffer->dma)
3182 dma_unmap_single(ring->dev,
3183 tx_buffer->dma,
3184 tx_buffer->length,
3185 DMA_TO_DEVICE);
3186 } else if (tx_buffer->dma) {
3187 dma_unmap_page(ring->dev,
3188 tx_buffer->dma,
3189 tx_buffer->length,
3190 DMA_TO_DEVICE);
3191 }
3192 tx_buffer->next_to_watch = NULL;
3193 tx_buffer->skb = NULL;
3194 tx_buffer->dma = 0;
3195 /* buffer_info must be completely set up in the transmit path */
3196 }
3197
3198 /**
3199 * igb_clean_tx_ring - Free Tx Buffers
3200 * @tx_ring: ring to be cleaned
3201 **/
3202 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3203 {
3204 struct igb_tx_buffer *buffer_info;
3205 unsigned long size;
3206 u16 i;
3207
3208 if (!tx_ring->tx_buffer_info)
3209 return;
3210 /* Free all the Tx ring sk_buffs */
3211
3212 for (i = 0; i < tx_ring->count; i++) {
3213 buffer_info = &tx_ring->tx_buffer_info[i];
3214 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3215 }
3216
3217 netdev_tx_reset_queue(txring_txq(tx_ring));
3218
3219 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3220 memset(tx_ring->tx_buffer_info, 0, size);
3221
3222 /* Zero out the descriptor ring */
3223 memset(tx_ring->desc, 0, tx_ring->size);
3224
3225 tx_ring->next_to_use = 0;
3226 tx_ring->next_to_clean = 0;
3227 }
3228
3229 /**
3230 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3231 * @adapter: board private structure
3232 **/
3233 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3234 {
3235 int i;
3236
3237 for (i = 0; i < adapter->num_tx_queues; i++)
3238 igb_clean_tx_ring(adapter->tx_ring[i]);
3239 }
3240
3241 /**
3242 * igb_free_rx_resources - Free Rx Resources
3243 * @rx_ring: ring to clean the resources from
3244 *
3245 * Free all receive software resources
3246 **/
3247 void igb_free_rx_resources(struct igb_ring *rx_ring)
3248 {
3249 igb_clean_rx_ring(rx_ring);
3250
3251 vfree(rx_ring->rx_buffer_info);
3252 rx_ring->rx_buffer_info = NULL;
3253
3254 /* if not set, then don't free */
3255 if (!rx_ring->desc)
3256 return;
3257
3258 dma_free_coherent(rx_ring->dev, rx_ring->size,
3259 rx_ring->desc, rx_ring->dma);
3260
3261 rx_ring->desc = NULL;
3262 }
3263
3264 /**
3265 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3266 * @adapter: board private structure
3267 *
3268 * Free all receive software resources
3269 **/
3270 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3271 {
3272 int i;
3273
3274 for (i = 0; i < adapter->num_rx_queues; i++)
3275 igb_free_rx_resources(adapter->rx_ring[i]);
3276 }
3277
3278 /**
3279 * igb_clean_rx_ring - Free Rx Buffers per Queue
3280 * @rx_ring: ring to free buffers from
3281 **/
3282 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3283 {
3284 unsigned long size;
3285 u16 i;
3286
3287 if (!rx_ring->rx_buffer_info)
3288 return;
3289
3290 /* Free all the Rx ring sk_buffs */
3291 for (i = 0; i < rx_ring->count; i++) {
3292 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3293 if (buffer_info->dma) {
3294 dma_unmap_single(rx_ring->dev,
3295 buffer_info->dma,
3296 IGB_RX_HDR_LEN,
3297 DMA_FROM_DEVICE);
3298 buffer_info->dma = 0;
3299 }
3300
3301 if (buffer_info->skb) {
3302 dev_kfree_skb(buffer_info->skb);
3303 buffer_info->skb = NULL;
3304 }
3305 if (buffer_info->page_dma) {
3306 dma_unmap_page(rx_ring->dev,
3307 buffer_info->page_dma,
3308 PAGE_SIZE / 2,
3309 DMA_FROM_DEVICE);
3310 buffer_info->page_dma = 0;
3311 }
3312 if (buffer_info->page) {
3313 put_page(buffer_info->page);
3314 buffer_info->page = NULL;
3315 buffer_info->page_offset = 0;
3316 }
3317 }
3318
3319 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3320 memset(rx_ring->rx_buffer_info, 0, size);
3321
3322 /* Zero out the descriptor ring */
3323 memset(rx_ring->desc, 0, rx_ring->size);
3324
3325 rx_ring->next_to_clean = 0;
3326 rx_ring->next_to_use = 0;
3327 }
3328
3329 /**
3330 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3331 * @adapter: board private structure
3332 **/
3333 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3334 {
3335 int i;
3336
3337 for (i = 0; i < adapter->num_rx_queues; i++)
3338 igb_clean_rx_ring(adapter->rx_ring[i]);
3339 }
3340
3341 /**
3342 * igb_set_mac - Change the Ethernet Address of the NIC
3343 * @netdev: network interface device structure
3344 * @p: pointer to an address structure
3345 *
3346 * Returns 0 on success, negative on failure
3347 **/
3348 static int igb_set_mac(struct net_device *netdev, void *p)
3349 {
3350 struct igb_adapter *adapter = netdev_priv(netdev);
3351 struct e1000_hw *hw = &adapter->hw;
3352 struct sockaddr *addr = p;
3353
3354 if (!is_valid_ether_addr(addr->sa_data))
3355 return -EADDRNOTAVAIL;
3356
3357 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3358 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3359
3360 /* set the correct pool for the new PF MAC address in entry 0 */
3361 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3362 adapter->vfs_allocated_count);
3363
3364 return 0;
3365 }
3366
3367 /**
3368 * igb_write_mc_addr_list - write multicast addresses to MTA
3369 * @netdev: network interface device structure
3370 *
3371 * Writes multicast address list to the MTA hash table.
3372 * Returns: -ENOMEM on failure
3373 * 0 on no addresses written
3374 * X on writing X addresses to MTA
3375 **/
3376 static int igb_write_mc_addr_list(struct net_device *netdev)
3377 {
3378 struct igb_adapter *adapter = netdev_priv(netdev);
3379 struct e1000_hw *hw = &adapter->hw;
3380 struct netdev_hw_addr *ha;
3381 u8 *mta_list;
3382 int i;
3383
3384 if (netdev_mc_empty(netdev)) {
3385 /* nothing to program, so clear mc list */
3386 igb_update_mc_addr_list(hw, NULL, 0);
3387 igb_restore_vf_multicasts(adapter);
3388 return 0;
3389 }
3390
3391 mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3392 if (!mta_list)
3393 return -ENOMEM;
3394
3395 /* The shared function expects a packed array of only addresses. */
3396 i = 0;
3397 netdev_for_each_mc_addr(ha, netdev)
3398 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3399
3400 igb_update_mc_addr_list(hw, mta_list, i);
3401 kfree(mta_list);
3402
3403 return netdev_mc_count(netdev);
3404 }
3405
3406 /**
3407 * igb_write_uc_addr_list - write unicast addresses to RAR table
3408 * @netdev: network interface device structure
3409 *
3410 * Writes unicast address list to the RAR table.
3411 * Returns: -ENOMEM on failure/insufficient address space
3412 * 0 on no addresses written
3413 * X on writing X addresses to the RAR table
3414 **/
3415 static int igb_write_uc_addr_list(struct net_device *netdev)
3416 {
3417 struct igb_adapter *adapter = netdev_priv(netdev);
3418 struct e1000_hw *hw = &adapter->hw;
3419 unsigned int vfn = adapter->vfs_allocated_count;
3420 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3421 int count = 0;
3422
3423 /* return ENOMEM indicating insufficient memory for addresses */
3424 if (netdev_uc_count(netdev) > rar_entries)
3425 return -ENOMEM;
3426
3427 if (!netdev_uc_empty(netdev) && rar_entries) {
3428 struct netdev_hw_addr *ha;
3429
3430 netdev_for_each_uc_addr(ha, netdev) {
3431 if (!rar_entries)
3432 break;
3433 igb_rar_set_qsel(adapter, ha->addr,
3434 rar_entries--,
3435 vfn);
3436 count++;
3437 }
3438 }
3439 /* write the addresses in reverse order to avoid write combining */
3440 for (; rar_entries > 0 ; rar_entries--) {
3441 wr32(E1000_RAH(rar_entries), 0);
3442 wr32(E1000_RAL(rar_entries), 0);
3443 }
3444 wrfl();
3445
3446 return count;
3447 }
3448
3449 /**
3450 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3451 * @netdev: network interface device structure
3452 *
3453 * The set_rx_mode entry point is called whenever the unicast or multicast
3454 * address lists or the network interface flags are updated. This routine is
3455 * responsible for configuring the hardware for proper unicast, multicast,
3456 * promiscuous mode, and all-multi behavior.
3457 **/
3458 static void igb_set_rx_mode(struct net_device *netdev)
3459 {
3460 struct igb_adapter *adapter = netdev_priv(netdev);
3461 struct e1000_hw *hw = &adapter->hw;
3462 unsigned int vfn = adapter->vfs_allocated_count;
3463 u32 rctl, vmolr = 0;
3464 int count;
3465
3466 /* Check for Promiscuous and All Multicast modes */
3467 rctl = rd32(E1000_RCTL);
3468
3469 /* clear the effected bits */
3470 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
3471
3472 if (netdev->flags & IFF_PROMISC) {
3473 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3474 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
3475 } else {
3476 if (netdev->flags & IFF_ALLMULTI) {
3477 rctl |= E1000_RCTL_MPE;
3478 vmolr |= E1000_VMOLR_MPME;
3479 } else {
3480 /*
3481 * Write addresses to the MTA, if the attempt fails
3482 * then we should just turn on promiscuous mode so
3483 * that we can at least receive multicast traffic
3484 */
3485 count = igb_write_mc_addr_list(netdev);
3486 if (count < 0) {
3487 rctl |= E1000_RCTL_MPE;
3488 vmolr |= E1000_VMOLR_MPME;
3489 } else if (count) {
3490 vmolr |= E1000_VMOLR_ROMPE;
3491 }
3492 }
3493 /*
3494 * Write addresses to available RAR registers, if there is not
3495 * sufficient space to store all the addresses then enable
3496 * unicast promiscuous mode
3497 */
3498 count = igb_write_uc_addr_list(netdev);
3499 if (count < 0) {
3500 rctl |= E1000_RCTL_UPE;
3501 vmolr |= E1000_VMOLR_ROPE;
3502 }
3503 rctl |= E1000_RCTL_VFE;
3504 }
3505 wr32(E1000_RCTL, rctl);
3506
3507 /*
3508 * In order to support SR-IOV and eventually VMDq it is necessary to set
3509 * the VMOLR to enable the appropriate modes. Without this workaround
3510 * we will have issues with VLAN tag stripping not being done for frames
3511 * that are only arriving because we are the default pool
3512 */
3513 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
3514 return;
3515
3516 vmolr |= rd32(E1000_VMOLR(vfn)) &
3517 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3518 wr32(E1000_VMOLR(vfn), vmolr);
3519 igb_restore_vf_multicasts(adapter);
3520 }
3521
3522 static void igb_check_wvbr(struct igb_adapter *adapter)
3523 {
3524 struct e1000_hw *hw = &adapter->hw;
3525 u32 wvbr = 0;
3526
3527 switch (hw->mac.type) {
3528 case e1000_82576:
3529 case e1000_i350:
3530 if (!(wvbr = rd32(E1000_WVBR)))
3531 return;
3532 break;
3533 default:
3534 break;
3535 }
3536
3537 adapter->wvbr |= wvbr;
3538 }
3539
3540 #define IGB_STAGGERED_QUEUE_OFFSET 8
3541
3542 static void igb_spoof_check(struct igb_adapter *adapter)
3543 {
3544 int j;
3545
3546 if (!adapter->wvbr)
3547 return;
3548
3549 for(j = 0; j < adapter->vfs_allocated_count; j++) {
3550 if (adapter->wvbr & (1 << j) ||
3551 adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
3552 dev_warn(&adapter->pdev->dev,
3553 "Spoof event(s) detected on VF %d\n", j);
3554 adapter->wvbr &=
3555 ~((1 << j) |
3556 (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
3557 }
3558 }
3559 }
3560
3561 /* Need to wait a few seconds after link up to get diagnostic information from
3562 * the phy */
3563 static void igb_update_phy_info(unsigned long data)
3564 {
3565 struct igb_adapter *adapter = (struct igb_adapter *) data;
3566 igb_get_phy_info(&adapter->hw);
3567 }
3568
3569 /**
3570 * igb_has_link - check shared code for link and determine up/down
3571 * @adapter: pointer to driver private info
3572 **/
3573 bool igb_has_link(struct igb_adapter *adapter)
3574 {
3575 struct e1000_hw *hw = &adapter->hw;
3576 bool link_active = false;
3577 s32 ret_val = 0;
3578
3579 /* get_link_status is set on LSC (link status) interrupt or
3580 * rx sequence error interrupt. get_link_status will stay
3581 * false until the e1000_check_for_link establishes link
3582 * for copper adapters ONLY
3583 */
3584 switch (hw->phy.media_type) {
3585 case e1000_media_type_copper:
3586 if (hw->mac.get_link_status) {
3587 ret_val = hw->mac.ops.check_for_link(hw);
3588 link_active = !hw->mac.get_link_status;
3589 } else {
3590 link_active = true;
3591 }
3592 break;
3593 case e1000_media_type_internal_serdes:
3594 ret_val = hw->mac.ops.check_for_link(hw);
3595 link_active = hw->mac.serdes_has_link;
3596 break;
3597 default:
3598 case e1000_media_type_unknown:
3599 break;
3600 }
3601
3602 return link_active;
3603 }
3604
3605 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
3606 {
3607 bool ret = false;
3608 u32 ctrl_ext, thstat;
3609
3610 /* check for thermal sensor event on i350 copper only */
3611 if (hw->mac.type == e1000_i350) {
3612 thstat = rd32(E1000_THSTAT);
3613 ctrl_ext = rd32(E1000_CTRL_EXT);
3614
3615 if ((hw->phy.media_type == e1000_media_type_copper) &&
3616 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3617 ret = !!(thstat & event);
3618 }
3619 }
3620
3621 return ret;
3622 }
3623
3624 /**
3625 * igb_watchdog - Timer Call-back
3626 * @data: pointer to adapter cast into an unsigned long
3627 **/
3628 static void igb_watchdog(unsigned long data)
3629 {
3630 struct igb_adapter *adapter = (struct igb_adapter *)data;
3631 /* Do the rest outside of interrupt context */
3632 schedule_work(&adapter->watchdog_task);
3633 }
3634
3635 static void igb_watchdog_task(struct work_struct *work)
3636 {
3637 struct igb_adapter *adapter = container_of(work,
3638 struct igb_adapter,
3639 watchdog_task);
3640 struct e1000_hw *hw = &adapter->hw;
3641 struct net_device *netdev = adapter->netdev;
3642 u32 link;
3643 int i;
3644
3645 link = igb_has_link(adapter);
3646 if (link) {
3647 /* Cancel scheduled suspend requests. */
3648 pm_runtime_resume(netdev->dev.parent);
3649
3650 if (!netif_carrier_ok(netdev)) {
3651 u32 ctrl;
3652 hw->mac.ops.get_speed_and_duplex(hw,
3653 &adapter->link_speed,
3654 &adapter->link_duplex);
3655
3656 ctrl = rd32(E1000_CTRL);
3657 /* Links status message must follow this format */
3658 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s "
3659 "Duplex, Flow Control: %s\n",
3660 netdev->name,
3661 adapter->link_speed,
3662 adapter->link_duplex == FULL_DUPLEX ?
3663 "Full" : "Half",
3664 (ctrl & E1000_CTRL_TFCE) &&
3665 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
3666 (ctrl & E1000_CTRL_RFCE) ? "RX" :
3667 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
3668
3669 /* check for thermal sensor event */
3670 if (igb_thermal_sensor_event(hw,
3671 E1000_THSTAT_LINK_THROTTLE)) {
3672 netdev_info(netdev, "The network adapter link "
3673 "speed was downshifted because it "
3674 "overheated\n");
3675 }
3676
3677 /* adjust timeout factor according to speed/duplex */
3678 adapter->tx_timeout_factor = 1;
3679 switch (adapter->link_speed) {
3680 case SPEED_10:
3681 adapter->tx_timeout_factor = 14;
3682 break;
3683 case SPEED_100:
3684 /* maybe add some timeout factor ? */
3685 break;
3686 }
3687
3688 netif_carrier_on(netdev);
3689
3690 igb_ping_all_vfs(adapter);
3691 igb_check_vf_rate_limit(adapter);
3692
3693 /* link state has changed, schedule phy info update */
3694 if (!test_bit(__IGB_DOWN, &adapter->state))
3695 mod_timer(&adapter->phy_info_timer,
3696 round_jiffies(jiffies + 2 * HZ));
3697 }
3698 } else {
3699 if (netif_carrier_ok(netdev)) {
3700 adapter->link_speed = 0;
3701 adapter->link_duplex = 0;
3702
3703 /* check for thermal sensor event */
3704 if (igb_thermal_sensor_event(hw,
3705 E1000_THSTAT_PWR_DOWN)) {
3706 netdev_err(netdev, "The network adapter was "
3707 "stopped because it overheated\n");
3708 }
3709
3710 /* Links status message must follow this format */
3711 printk(KERN_INFO "igb: %s NIC Link is Down\n",
3712 netdev->name);
3713 netif_carrier_off(netdev);
3714
3715 igb_ping_all_vfs(adapter);
3716
3717 /* link state has changed, schedule phy info update */
3718 if (!test_bit(__IGB_DOWN, &adapter->state))
3719 mod_timer(&adapter->phy_info_timer,
3720 round_jiffies(jiffies + 2 * HZ));
3721
3722 pm_schedule_suspend(netdev->dev.parent,
3723 MSEC_PER_SEC * 5);
3724 }
3725 }
3726
3727 spin_lock(&adapter->stats64_lock);
3728 igb_update_stats(adapter, &adapter->stats64);
3729 spin_unlock(&adapter->stats64_lock);
3730
3731 for (i = 0; i < adapter->num_tx_queues; i++) {
3732 struct igb_ring *tx_ring = adapter->tx_ring[i];
3733 if (!netif_carrier_ok(netdev)) {
3734 /* We've lost link, so the controller stops DMA,
3735 * but we've got queued Tx work that's never going
3736 * to get done, so reset controller to flush Tx.
3737 * (Do the reset outside of interrupt context). */
3738 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
3739 adapter->tx_timeout_count++;
3740 schedule_work(&adapter->reset_task);
3741 /* return immediately since reset is imminent */
3742 return;
3743 }
3744 }
3745
3746 /* Force detection of hung controller every watchdog period */
3747 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
3748 }
3749
3750 /* Cause software interrupt to ensure rx ring is cleaned */
3751 if (adapter->msix_entries) {
3752 u32 eics = 0;
3753 for (i = 0; i < adapter->num_q_vectors; i++)
3754 eics |= adapter->q_vector[i]->eims_value;
3755 wr32(E1000_EICS, eics);
3756 } else {
3757 wr32(E1000_ICS, E1000_ICS_RXDMT0);
3758 }
3759
3760 igb_spoof_check(adapter);
3761
3762 /* Reset the timer */
3763 if (!test_bit(__IGB_DOWN, &adapter->state))
3764 mod_timer(&adapter->watchdog_timer,
3765 round_jiffies(jiffies + 2 * HZ));
3766 }
3767
3768 enum latency_range {
3769 lowest_latency = 0,
3770 low_latency = 1,
3771 bulk_latency = 2,
3772 latency_invalid = 255
3773 };
3774
3775 /**
3776 * igb_update_ring_itr - update the dynamic ITR value based on packet size
3777 *
3778 * Stores a new ITR value based on strictly on packet size. This
3779 * algorithm is less sophisticated than that used in igb_update_itr,
3780 * due to the difficulty of synchronizing statistics across multiple
3781 * receive rings. The divisors and thresholds used by this function
3782 * were determined based on theoretical maximum wire speed and testing
3783 * data, in order to minimize response time while increasing bulk
3784 * throughput.
3785 * This functionality is controlled by the InterruptThrottleRate module
3786 * parameter (see igb_param.c)
3787 * NOTE: This function is called only when operating in a multiqueue
3788 * receive environment.
3789 * @q_vector: pointer to q_vector
3790 **/
3791 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
3792 {
3793 int new_val = q_vector->itr_val;
3794 int avg_wire_size = 0;
3795 struct igb_adapter *adapter = q_vector->adapter;
3796 unsigned int packets;
3797
3798 /* For non-gigabit speeds, just fix the interrupt rate at 4000
3799 * ints/sec - ITR timer value of 120 ticks.
3800 */
3801 if (adapter->link_speed != SPEED_1000) {
3802 new_val = IGB_4K_ITR;
3803 goto set_itr_val;
3804 }
3805
3806 packets = q_vector->rx.total_packets;
3807 if (packets)
3808 avg_wire_size = q_vector->rx.total_bytes / packets;
3809
3810 packets = q_vector->tx.total_packets;
3811 if (packets)
3812 avg_wire_size = max_t(u32, avg_wire_size,
3813 q_vector->tx.total_bytes / packets);
3814
3815 /* if avg_wire_size isn't set no work was done */
3816 if (!avg_wire_size)
3817 goto clear_counts;
3818
3819 /* Add 24 bytes to size to account for CRC, preamble, and gap */
3820 avg_wire_size += 24;
3821
3822 /* Don't starve jumbo frames */
3823 avg_wire_size = min(avg_wire_size, 3000);
3824
3825 /* Give a little boost to mid-size frames */
3826 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
3827 new_val = avg_wire_size / 3;
3828 else
3829 new_val = avg_wire_size / 2;
3830
3831 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3832 if (new_val < IGB_20K_ITR &&
3833 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
3834 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
3835 new_val = IGB_20K_ITR;
3836
3837 set_itr_val:
3838 if (new_val != q_vector->itr_val) {
3839 q_vector->itr_val = new_val;
3840 q_vector->set_itr = 1;
3841 }
3842 clear_counts:
3843 q_vector->rx.total_bytes = 0;
3844 q_vector->rx.total_packets = 0;
3845 q_vector->tx.total_bytes = 0;
3846 q_vector->tx.total_packets = 0;
3847 }
3848
3849 /**
3850 * igb_update_itr - update the dynamic ITR value based on statistics
3851 * Stores a new ITR value based on packets and byte
3852 * counts during the last interrupt. The advantage of per interrupt
3853 * computation is faster updates and more accurate ITR for the current
3854 * traffic pattern. Constants in this function were computed
3855 * based on theoretical maximum wire speed and thresholds were set based
3856 * on testing data as well as attempting to minimize response time
3857 * while increasing bulk throughput.
3858 * this functionality is controlled by the InterruptThrottleRate module
3859 * parameter (see igb_param.c)
3860 * NOTE: These calculations are only valid when operating in a single-
3861 * queue environment.
3862 * @q_vector: pointer to q_vector
3863 * @ring_container: ring info to update the itr for
3864 **/
3865 static void igb_update_itr(struct igb_q_vector *q_vector,
3866 struct igb_ring_container *ring_container)
3867 {
3868 unsigned int packets = ring_container->total_packets;
3869 unsigned int bytes = ring_container->total_bytes;
3870 u8 itrval = ring_container->itr;
3871
3872 /* no packets, exit with status unchanged */
3873 if (packets == 0)
3874 return;
3875
3876 switch (itrval) {
3877 case lowest_latency:
3878 /* handle TSO and jumbo frames */
3879 if (bytes/packets > 8000)
3880 itrval = bulk_latency;
3881 else if ((packets < 5) && (bytes > 512))
3882 itrval = low_latency;
3883 break;
3884 case low_latency: /* 50 usec aka 20000 ints/s */
3885 if (bytes > 10000) {
3886 /* this if handles the TSO accounting */
3887 if (bytes/packets > 8000) {
3888 itrval = bulk_latency;
3889 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
3890 itrval = bulk_latency;
3891 } else if ((packets > 35)) {
3892 itrval = lowest_latency;
3893 }
3894 } else if (bytes/packets > 2000) {
3895 itrval = bulk_latency;
3896 } else if (packets <= 2 && bytes < 512) {
3897 itrval = lowest_latency;
3898 }
3899 break;
3900 case bulk_latency: /* 250 usec aka 4000 ints/s */
3901 if (bytes > 25000) {
3902 if (packets > 35)
3903 itrval = low_latency;
3904 } else if (bytes < 1500) {
3905 itrval = low_latency;
3906 }
3907 break;
3908 }
3909
3910 /* clear work counters since we have the values we need */
3911 ring_container->total_bytes = 0;
3912 ring_container->total_packets = 0;
3913
3914 /* write updated itr to ring container */
3915 ring_container->itr = itrval;
3916 }
3917
3918 static void igb_set_itr(struct igb_q_vector *q_vector)
3919 {
3920 struct igb_adapter *adapter = q_vector->adapter;
3921 u32 new_itr = q_vector->itr_val;
3922 u8 current_itr = 0;
3923
3924 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
3925 if (adapter->link_speed != SPEED_1000) {
3926 current_itr = 0;
3927 new_itr = IGB_4K_ITR;
3928 goto set_itr_now;
3929 }
3930
3931 igb_update_itr(q_vector, &q_vector->tx);
3932 igb_update_itr(q_vector, &q_vector->rx);
3933
3934 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
3935
3936 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3937 if (current_itr == lowest_latency &&
3938 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
3939 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
3940 current_itr = low_latency;
3941
3942 switch (current_itr) {
3943 /* counts and packets in update_itr are dependent on these numbers */
3944 case lowest_latency:
3945 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
3946 break;
3947 case low_latency:
3948 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
3949 break;
3950 case bulk_latency:
3951 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
3952 break;
3953 default:
3954 break;
3955 }
3956
3957 set_itr_now:
3958 if (new_itr != q_vector->itr_val) {
3959 /* this attempts to bias the interrupt rate towards Bulk
3960 * by adding intermediate steps when interrupt rate is
3961 * increasing */
3962 new_itr = new_itr > q_vector->itr_val ?
3963 max((new_itr * q_vector->itr_val) /
3964 (new_itr + (q_vector->itr_val >> 2)),
3965 new_itr) :
3966 new_itr;
3967 /* Don't write the value here; it resets the adapter's
3968 * internal timer, and causes us to delay far longer than
3969 * we should between interrupts. Instead, we write the ITR
3970 * value at the beginning of the next interrupt so the timing
3971 * ends up being correct.
3972 */
3973 q_vector->itr_val = new_itr;
3974 q_vector->set_itr = 1;
3975 }
3976 }
3977
3978 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
3979 u32 type_tucmd, u32 mss_l4len_idx)
3980 {
3981 struct e1000_adv_tx_context_desc *context_desc;
3982 u16 i = tx_ring->next_to_use;
3983
3984 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
3985
3986 i++;
3987 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3988
3989 /* set bits to identify this as an advanced context descriptor */
3990 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
3991
3992 /* For 82575, context index must be unique per ring. */
3993 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
3994 mss_l4len_idx |= tx_ring->reg_idx << 4;
3995
3996 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3997 context_desc->seqnum_seed = 0;
3998 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
3999 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4000 }
4001
4002 static int igb_tso(struct igb_ring *tx_ring,
4003 struct igb_tx_buffer *first,
4004 u8 *hdr_len)
4005 {
4006 struct sk_buff *skb = first->skb;
4007 u32 vlan_macip_lens, type_tucmd;
4008 u32 mss_l4len_idx, l4len;
4009
4010 if (!skb_is_gso(skb))
4011 return 0;
4012
4013 if (skb_header_cloned(skb)) {
4014 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4015 if (err)
4016 return err;
4017 }
4018
4019 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4020 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4021
4022 if (first->protocol == __constant_htons(ETH_P_IP)) {
4023 struct iphdr *iph = ip_hdr(skb);
4024 iph->tot_len = 0;
4025 iph->check = 0;
4026 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4027 iph->daddr, 0,
4028 IPPROTO_TCP,
4029 0);
4030 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4031 first->tx_flags |= IGB_TX_FLAGS_TSO |
4032 IGB_TX_FLAGS_CSUM |
4033 IGB_TX_FLAGS_IPV4;
4034 } else if (skb_is_gso_v6(skb)) {
4035 ipv6_hdr(skb)->payload_len = 0;
4036 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4037 &ipv6_hdr(skb)->daddr,
4038 0, IPPROTO_TCP, 0);
4039 first->tx_flags |= IGB_TX_FLAGS_TSO |
4040 IGB_TX_FLAGS_CSUM;
4041 }
4042
4043 /* compute header lengths */
4044 l4len = tcp_hdrlen(skb);
4045 *hdr_len = skb_transport_offset(skb) + l4len;
4046
4047 /* update gso size and bytecount with header size */
4048 first->gso_segs = skb_shinfo(skb)->gso_segs;
4049 first->bytecount += (first->gso_segs - 1) * *hdr_len;
4050
4051 /* MSS L4LEN IDX */
4052 mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4053 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4054
4055 /* VLAN MACLEN IPLEN */
4056 vlan_macip_lens = skb_network_header_len(skb);
4057 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4058 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4059
4060 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4061
4062 return 1;
4063 }
4064
4065 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4066 {
4067 struct sk_buff *skb = first->skb;
4068 u32 vlan_macip_lens = 0;
4069 u32 mss_l4len_idx = 0;
4070 u32 type_tucmd = 0;
4071
4072 if (skb->ip_summed != CHECKSUM_PARTIAL) {
4073 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4074 return;
4075 } else {
4076 u8 l4_hdr = 0;
4077 switch (first->protocol) {
4078 case __constant_htons(ETH_P_IP):
4079 vlan_macip_lens |= skb_network_header_len(skb);
4080 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4081 l4_hdr = ip_hdr(skb)->protocol;
4082 break;
4083 case __constant_htons(ETH_P_IPV6):
4084 vlan_macip_lens |= skb_network_header_len(skb);
4085 l4_hdr = ipv6_hdr(skb)->nexthdr;
4086 break;
4087 default:
4088 if (unlikely(net_ratelimit())) {
4089 dev_warn(tx_ring->dev,
4090 "partial checksum but proto=%x!\n",
4091 first->protocol);
4092 }
4093 break;
4094 }
4095
4096 switch (l4_hdr) {
4097 case IPPROTO_TCP:
4098 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4099 mss_l4len_idx = tcp_hdrlen(skb) <<
4100 E1000_ADVTXD_L4LEN_SHIFT;
4101 break;
4102 case IPPROTO_SCTP:
4103 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4104 mss_l4len_idx = sizeof(struct sctphdr) <<
4105 E1000_ADVTXD_L4LEN_SHIFT;
4106 break;
4107 case IPPROTO_UDP:
4108 mss_l4len_idx = sizeof(struct udphdr) <<
4109 E1000_ADVTXD_L4LEN_SHIFT;
4110 break;
4111 default:
4112 if (unlikely(net_ratelimit())) {
4113 dev_warn(tx_ring->dev,
4114 "partial checksum but l4 proto=%x!\n",
4115 l4_hdr);
4116 }
4117 break;
4118 }
4119
4120 /* update TX checksum flag */
4121 first->tx_flags |= IGB_TX_FLAGS_CSUM;
4122 }
4123
4124 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4125 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4126
4127 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4128 }
4129
4130 static __le32 igb_tx_cmd_type(u32 tx_flags)
4131 {
4132 /* set type for advanced descriptor with frame checksum insertion */
4133 __le32 cmd_type = cpu_to_le32(E1000_ADVTXD_DTYP_DATA |
4134 E1000_ADVTXD_DCMD_IFCS |
4135 E1000_ADVTXD_DCMD_DEXT);
4136
4137 /* set HW vlan bit if vlan is present */
4138 if (tx_flags & IGB_TX_FLAGS_VLAN)
4139 cmd_type |= cpu_to_le32(E1000_ADVTXD_DCMD_VLE);
4140
4141 /* set timestamp bit if present */
4142 if (tx_flags & IGB_TX_FLAGS_TSTAMP)
4143 cmd_type |= cpu_to_le32(E1000_ADVTXD_MAC_TSTAMP);
4144
4145 /* set segmentation bits for TSO */
4146 if (tx_flags & IGB_TX_FLAGS_TSO)
4147 cmd_type |= cpu_to_le32(E1000_ADVTXD_DCMD_TSE);
4148
4149 return cmd_type;
4150 }
4151
4152 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4153 union e1000_adv_tx_desc *tx_desc,
4154 u32 tx_flags, unsigned int paylen)
4155 {
4156 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4157
4158 /* 82575 requires a unique index per ring if any offload is enabled */
4159 if ((tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_VLAN)) &&
4160 test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4161 olinfo_status |= tx_ring->reg_idx << 4;
4162
4163 /* insert L4 checksum */
4164 if (tx_flags & IGB_TX_FLAGS_CSUM) {
4165 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
4166
4167 /* insert IPv4 checksum */
4168 if (tx_flags & IGB_TX_FLAGS_IPV4)
4169 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
4170 }
4171
4172 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4173 }
4174
4175 /*
4176 * The largest size we can write to the descriptor is 65535. In order to
4177 * maintain a power of two alignment we have to limit ourselves to 32K.
4178 */
4179 #define IGB_MAX_TXD_PWR 15
4180 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
4181
4182 static void igb_tx_map(struct igb_ring *tx_ring,
4183 struct igb_tx_buffer *first,
4184 const u8 hdr_len)
4185 {
4186 struct sk_buff *skb = first->skb;
4187 struct igb_tx_buffer *tx_buffer_info;
4188 union e1000_adv_tx_desc *tx_desc;
4189 dma_addr_t dma;
4190 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
4191 unsigned int data_len = skb->data_len;
4192 unsigned int size = skb_headlen(skb);
4193 unsigned int paylen = skb->len - hdr_len;
4194 __le32 cmd_type;
4195 u32 tx_flags = first->tx_flags;
4196 u16 i = tx_ring->next_to_use;
4197
4198 tx_desc = IGB_TX_DESC(tx_ring, i);
4199
4200 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, paylen);
4201 cmd_type = igb_tx_cmd_type(tx_flags);
4202
4203 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4204 if (dma_mapping_error(tx_ring->dev, dma))
4205 goto dma_error;
4206
4207 /* record length, and DMA address */
4208 first->length = size;
4209 first->dma = dma;
4210 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4211
4212 for (;;) {
4213 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4214 tx_desc->read.cmd_type_len =
4215 cmd_type | cpu_to_le32(IGB_MAX_DATA_PER_TXD);
4216
4217 i++;
4218 tx_desc++;
4219 if (i == tx_ring->count) {
4220 tx_desc = IGB_TX_DESC(tx_ring, 0);
4221 i = 0;
4222 }
4223
4224 dma += IGB_MAX_DATA_PER_TXD;
4225 size -= IGB_MAX_DATA_PER_TXD;
4226
4227 tx_desc->read.olinfo_status = 0;
4228 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4229 }
4230
4231 if (likely(!data_len))
4232 break;
4233
4234 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
4235
4236 i++;
4237 tx_desc++;
4238 if (i == tx_ring->count) {
4239 tx_desc = IGB_TX_DESC(tx_ring, 0);
4240 i = 0;
4241 }
4242
4243 size = skb_frag_size(frag);
4244 data_len -= size;
4245
4246 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4247 size, DMA_TO_DEVICE);
4248 if (dma_mapping_error(tx_ring->dev, dma))
4249 goto dma_error;
4250
4251 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4252 tx_buffer_info->length = size;
4253 tx_buffer_info->dma = dma;
4254
4255 tx_desc->read.olinfo_status = 0;
4256 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4257
4258 frag++;
4259 }
4260
4261 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
4262
4263 /* write last descriptor with RS and EOP bits */
4264 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IGB_TXD_DCMD);
4265 if (unlikely(skb->no_fcs))
4266 cmd_type &= ~(cpu_to_le32(E1000_ADVTXD_DCMD_IFCS));
4267 tx_desc->read.cmd_type_len = cmd_type;
4268
4269 /* set the timestamp */
4270 first->time_stamp = jiffies;
4271
4272 /*
4273 * Force memory writes to complete before letting h/w know there
4274 * are new descriptors to fetch. (Only applicable for weak-ordered
4275 * memory model archs, such as IA-64).
4276 *
4277 * We also need this memory barrier to make certain all of the
4278 * status bits have been updated before next_to_watch is written.
4279 */
4280 wmb();
4281
4282 /* set next_to_watch value indicating a packet is present */
4283 first->next_to_watch = tx_desc;
4284
4285 i++;
4286 if (i == tx_ring->count)
4287 i = 0;
4288
4289 tx_ring->next_to_use = i;
4290
4291 writel(i, tx_ring->tail);
4292
4293 /* we need this if more than one processor can write to our tail
4294 * at a time, it syncronizes IO on IA64/Altix systems */
4295 mmiowb();
4296
4297 return;
4298
4299 dma_error:
4300 dev_err(tx_ring->dev, "TX DMA map failed\n");
4301
4302 /* clear dma mappings for failed tx_buffer_info map */
4303 for (;;) {
4304 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4305 igb_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4306 if (tx_buffer_info == first)
4307 break;
4308 if (i == 0)
4309 i = tx_ring->count;
4310 i--;
4311 }
4312
4313 tx_ring->next_to_use = i;
4314 }
4315
4316 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4317 {
4318 struct net_device *netdev = tx_ring->netdev;
4319
4320 netif_stop_subqueue(netdev, tx_ring->queue_index);
4321
4322 /* Herbert's original patch had:
4323 * smp_mb__after_netif_stop_queue();
4324 * but since that doesn't exist yet, just open code it. */
4325 smp_mb();
4326
4327 /* We need to check again in a case another CPU has just
4328 * made room available. */
4329 if (igb_desc_unused(tx_ring) < size)
4330 return -EBUSY;
4331
4332 /* A reprieve! */
4333 netif_wake_subqueue(netdev, tx_ring->queue_index);
4334
4335 u64_stats_update_begin(&tx_ring->tx_syncp2);
4336 tx_ring->tx_stats.restart_queue2++;
4337 u64_stats_update_end(&tx_ring->tx_syncp2);
4338
4339 return 0;
4340 }
4341
4342 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4343 {
4344 if (igb_desc_unused(tx_ring) >= size)
4345 return 0;
4346 return __igb_maybe_stop_tx(tx_ring, size);
4347 }
4348
4349 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4350 struct igb_ring *tx_ring)
4351 {
4352 struct igb_tx_buffer *first;
4353 int tso;
4354 u32 tx_flags = 0;
4355 __be16 protocol = vlan_get_protocol(skb);
4356 u8 hdr_len = 0;
4357
4358 /* need: 1 descriptor per page,
4359 * + 2 desc gap to keep tail from touching head,
4360 * + 1 desc for skb->data,
4361 * + 1 desc for context descriptor,
4362 * otherwise try next time */
4363 if (igb_maybe_stop_tx(tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
4364 /* this is a hard error */
4365 return NETDEV_TX_BUSY;
4366 }
4367
4368 /* record the location of the first descriptor for this packet */
4369 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4370 first->skb = skb;
4371 first->bytecount = skb->len;
4372 first->gso_segs = 1;
4373
4374 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
4375 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
4376 tx_flags |= IGB_TX_FLAGS_TSTAMP;
4377 }
4378
4379 if (vlan_tx_tag_present(skb)) {
4380 tx_flags |= IGB_TX_FLAGS_VLAN;
4381 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
4382 }
4383
4384 /* record initial flags and protocol */
4385 first->tx_flags = tx_flags;
4386 first->protocol = protocol;
4387
4388 tso = igb_tso(tx_ring, first, &hdr_len);
4389 if (tso < 0)
4390 goto out_drop;
4391 else if (!tso)
4392 igb_tx_csum(tx_ring, first);
4393
4394 igb_tx_map(tx_ring, first, hdr_len);
4395
4396 /* Make sure there is space in the ring for the next send. */
4397 igb_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 4);
4398
4399 return NETDEV_TX_OK;
4400
4401 out_drop:
4402 igb_unmap_and_free_tx_resource(tx_ring, first);
4403
4404 return NETDEV_TX_OK;
4405 }
4406
4407 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
4408 struct sk_buff *skb)
4409 {
4410 unsigned int r_idx = skb->queue_mapping;
4411
4412 if (r_idx >= adapter->num_tx_queues)
4413 r_idx = r_idx % adapter->num_tx_queues;
4414
4415 return adapter->tx_ring[r_idx];
4416 }
4417
4418 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
4419 struct net_device *netdev)
4420 {
4421 struct igb_adapter *adapter = netdev_priv(netdev);
4422
4423 if (test_bit(__IGB_DOWN, &adapter->state)) {
4424 dev_kfree_skb_any(skb);
4425 return NETDEV_TX_OK;
4426 }
4427
4428 if (skb->len <= 0) {
4429 dev_kfree_skb_any(skb);
4430 return NETDEV_TX_OK;
4431 }
4432
4433 /*
4434 * The minimum packet size with TCTL.PSP set is 17 so pad the skb
4435 * in order to meet this minimum size requirement.
4436 */
4437 if (skb->len < 17) {
4438 if (skb_padto(skb, 17))
4439 return NETDEV_TX_OK;
4440 skb->len = 17;
4441 }
4442
4443 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
4444 }
4445
4446 /**
4447 * igb_tx_timeout - Respond to a Tx Hang
4448 * @netdev: network interface device structure
4449 **/
4450 static void igb_tx_timeout(struct net_device *netdev)
4451 {
4452 struct igb_adapter *adapter = netdev_priv(netdev);
4453 struct e1000_hw *hw = &adapter->hw;
4454
4455 /* Do the reset outside of interrupt context */
4456 adapter->tx_timeout_count++;
4457
4458 if (hw->mac.type >= e1000_82580)
4459 hw->dev_spec._82575.global_device_reset = true;
4460
4461 schedule_work(&adapter->reset_task);
4462 wr32(E1000_EICS,
4463 (adapter->eims_enable_mask & ~adapter->eims_other));
4464 }
4465
4466 static void igb_reset_task(struct work_struct *work)
4467 {
4468 struct igb_adapter *adapter;
4469 adapter = container_of(work, struct igb_adapter, reset_task);
4470
4471 igb_dump(adapter);
4472 netdev_err(adapter->netdev, "Reset adapter\n");
4473 igb_reinit_locked(adapter);
4474 }
4475
4476 /**
4477 * igb_get_stats64 - Get System Network Statistics
4478 * @netdev: network interface device structure
4479 * @stats: rtnl_link_stats64 pointer
4480 *
4481 **/
4482 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
4483 struct rtnl_link_stats64 *stats)
4484 {
4485 struct igb_adapter *adapter = netdev_priv(netdev);
4486
4487 spin_lock(&adapter->stats64_lock);
4488 igb_update_stats(adapter, &adapter->stats64);
4489 memcpy(stats, &adapter->stats64, sizeof(*stats));
4490 spin_unlock(&adapter->stats64_lock);
4491
4492 return stats;
4493 }
4494
4495 /**
4496 * igb_change_mtu - Change the Maximum Transfer Unit
4497 * @netdev: network interface device structure
4498 * @new_mtu: new value for maximum frame size
4499 *
4500 * Returns 0 on success, negative on failure
4501 **/
4502 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
4503 {
4504 struct igb_adapter *adapter = netdev_priv(netdev);
4505 struct pci_dev *pdev = adapter->pdev;
4506 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4507
4508 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4509 dev_err(&pdev->dev, "Invalid MTU setting\n");
4510 return -EINVAL;
4511 }
4512
4513 #define MAX_STD_JUMBO_FRAME_SIZE 9238
4514 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4515 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
4516 return -EINVAL;
4517 }
4518
4519 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
4520 msleep(1);
4521
4522 /* igb_down has a dependency on max_frame_size */
4523 adapter->max_frame_size = max_frame;
4524
4525 if (netif_running(netdev))
4526 igb_down(adapter);
4527
4528 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
4529 netdev->mtu, new_mtu);
4530 netdev->mtu = new_mtu;
4531
4532 if (netif_running(netdev))
4533 igb_up(adapter);
4534 else
4535 igb_reset(adapter);
4536
4537 clear_bit(__IGB_RESETTING, &adapter->state);
4538
4539 return 0;
4540 }
4541
4542 /**
4543 * igb_update_stats - Update the board statistics counters
4544 * @adapter: board private structure
4545 **/
4546
4547 void igb_update_stats(struct igb_adapter *adapter,
4548 struct rtnl_link_stats64 *net_stats)
4549 {
4550 struct e1000_hw *hw = &adapter->hw;
4551 struct pci_dev *pdev = adapter->pdev;
4552 u32 reg, mpc;
4553 u16 phy_tmp;
4554 int i;
4555 u64 bytes, packets;
4556 unsigned int start;
4557 u64 _bytes, _packets;
4558
4559 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
4560
4561 /*
4562 * Prevent stats update while adapter is being reset, or if the pci
4563 * connection is down.
4564 */
4565 if (adapter->link_speed == 0)
4566 return;
4567 if (pci_channel_offline(pdev))
4568 return;
4569
4570 bytes = 0;
4571 packets = 0;
4572 for (i = 0; i < adapter->num_rx_queues; i++) {
4573 u32 rqdpc_tmp = rd32(E1000_RQDPC(i)) & 0x0FFF;
4574 struct igb_ring *ring = adapter->rx_ring[i];
4575
4576 ring->rx_stats.drops += rqdpc_tmp;
4577 net_stats->rx_fifo_errors += rqdpc_tmp;
4578
4579 do {
4580 start = u64_stats_fetch_begin_bh(&ring->rx_syncp);
4581 _bytes = ring->rx_stats.bytes;
4582 _packets = ring->rx_stats.packets;
4583 } while (u64_stats_fetch_retry_bh(&ring->rx_syncp, start));
4584 bytes += _bytes;
4585 packets += _packets;
4586 }
4587
4588 net_stats->rx_bytes = bytes;
4589 net_stats->rx_packets = packets;
4590
4591 bytes = 0;
4592 packets = 0;
4593 for (i = 0; i < adapter->num_tx_queues; i++) {
4594 struct igb_ring *ring = adapter->tx_ring[i];
4595 do {
4596 start = u64_stats_fetch_begin_bh(&ring->tx_syncp);
4597 _bytes = ring->tx_stats.bytes;
4598 _packets = ring->tx_stats.packets;
4599 } while (u64_stats_fetch_retry_bh(&ring->tx_syncp, start));
4600 bytes += _bytes;
4601 packets += _packets;
4602 }
4603 net_stats->tx_bytes = bytes;
4604 net_stats->tx_packets = packets;
4605
4606 /* read stats registers */
4607 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
4608 adapter->stats.gprc += rd32(E1000_GPRC);
4609 adapter->stats.gorc += rd32(E1000_GORCL);
4610 rd32(E1000_GORCH); /* clear GORCL */
4611 adapter->stats.bprc += rd32(E1000_BPRC);
4612 adapter->stats.mprc += rd32(E1000_MPRC);
4613 adapter->stats.roc += rd32(E1000_ROC);
4614
4615 adapter->stats.prc64 += rd32(E1000_PRC64);
4616 adapter->stats.prc127 += rd32(E1000_PRC127);
4617 adapter->stats.prc255 += rd32(E1000_PRC255);
4618 adapter->stats.prc511 += rd32(E1000_PRC511);
4619 adapter->stats.prc1023 += rd32(E1000_PRC1023);
4620 adapter->stats.prc1522 += rd32(E1000_PRC1522);
4621 adapter->stats.symerrs += rd32(E1000_SYMERRS);
4622 adapter->stats.sec += rd32(E1000_SEC);
4623
4624 mpc = rd32(E1000_MPC);
4625 adapter->stats.mpc += mpc;
4626 net_stats->rx_fifo_errors += mpc;
4627 adapter->stats.scc += rd32(E1000_SCC);
4628 adapter->stats.ecol += rd32(E1000_ECOL);
4629 adapter->stats.mcc += rd32(E1000_MCC);
4630 adapter->stats.latecol += rd32(E1000_LATECOL);
4631 adapter->stats.dc += rd32(E1000_DC);
4632 adapter->stats.rlec += rd32(E1000_RLEC);
4633 adapter->stats.xonrxc += rd32(E1000_XONRXC);
4634 adapter->stats.xontxc += rd32(E1000_XONTXC);
4635 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
4636 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
4637 adapter->stats.fcruc += rd32(E1000_FCRUC);
4638 adapter->stats.gptc += rd32(E1000_GPTC);
4639 adapter->stats.gotc += rd32(E1000_GOTCL);
4640 rd32(E1000_GOTCH); /* clear GOTCL */
4641 adapter->stats.rnbc += rd32(E1000_RNBC);
4642 adapter->stats.ruc += rd32(E1000_RUC);
4643 adapter->stats.rfc += rd32(E1000_RFC);
4644 adapter->stats.rjc += rd32(E1000_RJC);
4645 adapter->stats.tor += rd32(E1000_TORH);
4646 adapter->stats.tot += rd32(E1000_TOTH);
4647 adapter->stats.tpr += rd32(E1000_TPR);
4648
4649 adapter->stats.ptc64 += rd32(E1000_PTC64);
4650 adapter->stats.ptc127 += rd32(E1000_PTC127);
4651 adapter->stats.ptc255 += rd32(E1000_PTC255);
4652 adapter->stats.ptc511 += rd32(E1000_PTC511);
4653 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
4654 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
4655
4656 adapter->stats.mptc += rd32(E1000_MPTC);
4657 adapter->stats.bptc += rd32(E1000_BPTC);
4658
4659 adapter->stats.tpt += rd32(E1000_TPT);
4660 adapter->stats.colc += rd32(E1000_COLC);
4661
4662 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
4663 /* read internal phy specific stats */
4664 reg = rd32(E1000_CTRL_EXT);
4665 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
4666 adapter->stats.rxerrc += rd32(E1000_RXERRC);
4667 adapter->stats.tncrs += rd32(E1000_TNCRS);
4668 }
4669
4670 adapter->stats.tsctc += rd32(E1000_TSCTC);
4671 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
4672
4673 adapter->stats.iac += rd32(E1000_IAC);
4674 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
4675 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
4676 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
4677 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
4678 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
4679 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
4680 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
4681 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
4682
4683 /* Fill out the OS statistics structure */
4684 net_stats->multicast = adapter->stats.mprc;
4685 net_stats->collisions = adapter->stats.colc;
4686
4687 /* Rx Errors */
4688
4689 /* RLEC on some newer hardware can be incorrect so build
4690 * our own version based on RUC and ROC */
4691 net_stats->rx_errors = adapter->stats.rxerrc +
4692 adapter->stats.crcerrs + adapter->stats.algnerrc +
4693 adapter->stats.ruc + adapter->stats.roc +
4694 adapter->stats.cexterr;
4695 net_stats->rx_length_errors = adapter->stats.ruc +
4696 adapter->stats.roc;
4697 net_stats->rx_crc_errors = adapter->stats.crcerrs;
4698 net_stats->rx_frame_errors = adapter->stats.algnerrc;
4699 net_stats->rx_missed_errors = adapter->stats.mpc;
4700
4701 /* Tx Errors */
4702 net_stats->tx_errors = adapter->stats.ecol +
4703 adapter->stats.latecol;
4704 net_stats->tx_aborted_errors = adapter->stats.ecol;
4705 net_stats->tx_window_errors = adapter->stats.latecol;
4706 net_stats->tx_carrier_errors = adapter->stats.tncrs;
4707
4708 /* Tx Dropped needs to be maintained elsewhere */
4709
4710 /* Phy Stats */
4711 if (hw->phy.media_type == e1000_media_type_copper) {
4712 if ((adapter->link_speed == SPEED_1000) &&
4713 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
4714 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
4715 adapter->phy_stats.idle_errors += phy_tmp;
4716 }
4717 }
4718
4719 /* Management Stats */
4720 adapter->stats.mgptc += rd32(E1000_MGTPTC);
4721 adapter->stats.mgprc += rd32(E1000_MGTPRC);
4722 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
4723
4724 /* OS2BMC Stats */
4725 reg = rd32(E1000_MANC);
4726 if (reg & E1000_MANC_EN_BMC2OS) {
4727 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
4728 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
4729 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
4730 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
4731 }
4732 }
4733
4734 static irqreturn_t igb_msix_other(int irq, void *data)
4735 {
4736 struct igb_adapter *adapter = data;
4737 struct e1000_hw *hw = &adapter->hw;
4738 u32 icr = rd32(E1000_ICR);
4739 /* reading ICR causes bit 31 of EICR to be cleared */
4740
4741 if (icr & E1000_ICR_DRSTA)
4742 schedule_work(&adapter->reset_task);
4743
4744 if (icr & E1000_ICR_DOUTSYNC) {
4745 /* HW is reporting DMA is out of sync */
4746 adapter->stats.doosync++;
4747 /* The DMA Out of Sync is also indication of a spoof event
4748 * in IOV mode. Check the Wrong VM Behavior register to
4749 * see if it is really a spoof event. */
4750 igb_check_wvbr(adapter);
4751 }
4752
4753 /* Check for a mailbox event */
4754 if (icr & E1000_ICR_VMMB)
4755 igb_msg_task(adapter);
4756
4757 if (icr & E1000_ICR_LSC) {
4758 hw->mac.get_link_status = 1;
4759 /* guard against interrupt when we're going down */
4760 if (!test_bit(__IGB_DOWN, &adapter->state))
4761 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4762 }
4763
4764 wr32(E1000_EIMS, adapter->eims_other);
4765
4766 return IRQ_HANDLED;
4767 }
4768
4769 static void igb_write_itr(struct igb_q_vector *q_vector)
4770 {
4771 struct igb_adapter *adapter = q_vector->adapter;
4772 u32 itr_val = q_vector->itr_val & 0x7FFC;
4773
4774 if (!q_vector->set_itr)
4775 return;
4776
4777 if (!itr_val)
4778 itr_val = 0x4;
4779
4780 if (adapter->hw.mac.type == e1000_82575)
4781 itr_val |= itr_val << 16;
4782 else
4783 itr_val |= E1000_EITR_CNT_IGNR;
4784
4785 writel(itr_val, q_vector->itr_register);
4786 q_vector->set_itr = 0;
4787 }
4788
4789 static irqreturn_t igb_msix_ring(int irq, void *data)
4790 {
4791 struct igb_q_vector *q_vector = data;
4792
4793 /* Write the ITR value calculated from the previous interrupt. */
4794 igb_write_itr(q_vector);
4795
4796 napi_schedule(&q_vector->napi);
4797
4798 return IRQ_HANDLED;
4799 }
4800
4801 #ifdef CONFIG_IGB_DCA
4802 static void igb_update_dca(struct igb_q_vector *q_vector)
4803 {
4804 struct igb_adapter *adapter = q_vector->adapter;
4805 struct e1000_hw *hw = &adapter->hw;
4806 int cpu = get_cpu();
4807
4808 if (q_vector->cpu == cpu)
4809 goto out_no_update;
4810
4811 if (q_vector->tx.ring) {
4812 int q = q_vector->tx.ring->reg_idx;
4813 u32 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
4814 if (hw->mac.type == e1000_82575) {
4815 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
4816 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4817 } else {
4818 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
4819 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4820 E1000_DCA_TXCTRL_CPUID_SHIFT;
4821 }
4822 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
4823 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
4824 }
4825 if (q_vector->rx.ring) {
4826 int q = q_vector->rx.ring->reg_idx;
4827 u32 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
4828 if (hw->mac.type == e1000_82575) {
4829 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
4830 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4831 } else {
4832 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
4833 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4834 E1000_DCA_RXCTRL_CPUID_SHIFT;
4835 }
4836 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
4837 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
4838 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
4839 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
4840 }
4841 q_vector->cpu = cpu;
4842 out_no_update:
4843 put_cpu();
4844 }
4845
4846 static void igb_setup_dca(struct igb_adapter *adapter)
4847 {
4848 struct e1000_hw *hw = &adapter->hw;
4849 int i;
4850
4851 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
4852 return;
4853
4854 /* Always use CB2 mode, difference is masked in the CB driver. */
4855 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
4856
4857 for (i = 0; i < adapter->num_q_vectors; i++) {
4858 adapter->q_vector[i]->cpu = -1;
4859 igb_update_dca(adapter->q_vector[i]);
4860 }
4861 }
4862
4863 static int __igb_notify_dca(struct device *dev, void *data)
4864 {
4865 struct net_device *netdev = dev_get_drvdata(dev);
4866 struct igb_adapter *adapter = netdev_priv(netdev);
4867 struct pci_dev *pdev = adapter->pdev;
4868 struct e1000_hw *hw = &adapter->hw;
4869 unsigned long event = *(unsigned long *)data;
4870
4871 switch (event) {
4872 case DCA_PROVIDER_ADD:
4873 /* if already enabled, don't do it again */
4874 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
4875 break;
4876 if (dca_add_requester(dev) == 0) {
4877 adapter->flags |= IGB_FLAG_DCA_ENABLED;
4878 dev_info(&pdev->dev, "DCA enabled\n");
4879 igb_setup_dca(adapter);
4880 break;
4881 }
4882 /* Fall Through since DCA is disabled. */
4883 case DCA_PROVIDER_REMOVE:
4884 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
4885 /* without this a class_device is left
4886 * hanging around in the sysfs model */
4887 dca_remove_requester(dev);
4888 dev_info(&pdev->dev, "DCA disabled\n");
4889 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
4890 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
4891 }
4892 break;
4893 }
4894
4895 return 0;
4896 }
4897
4898 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
4899 void *p)
4900 {
4901 int ret_val;
4902
4903 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
4904 __igb_notify_dca);
4905
4906 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4907 }
4908 #endif /* CONFIG_IGB_DCA */
4909
4910 #ifdef CONFIG_PCI_IOV
4911 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
4912 {
4913 unsigned char mac_addr[ETH_ALEN];
4914 struct pci_dev *pdev = adapter->pdev;
4915 struct e1000_hw *hw = &adapter->hw;
4916 struct pci_dev *pvfdev;
4917 unsigned int device_id;
4918 u16 thisvf_devfn;
4919
4920 random_ether_addr(mac_addr);
4921 igb_set_vf_mac(adapter, vf, mac_addr);
4922
4923 switch (adapter->hw.mac.type) {
4924 case e1000_82576:
4925 device_id = IGB_82576_VF_DEV_ID;
4926 /* VF Stride for 82576 is 2 */
4927 thisvf_devfn = (pdev->devfn + 0x80 + (vf << 1)) |
4928 (pdev->devfn & 1);
4929 break;
4930 case e1000_i350:
4931 device_id = IGB_I350_VF_DEV_ID;
4932 /* VF Stride for I350 is 4 */
4933 thisvf_devfn = (pdev->devfn + 0x80 + (vf << 2)) |
4934 (pdev->devfn & 3);
4935 break;
4936 default:
4937 device_id = 0;
4938 thisvf_devfn = 0;
4939 break;
4940 }
4941
4942 pvfdev = pci_get_device(hw->vendor_id, device_id, NULL);
4943 while (pvfdev) {
4944 if (pvfdev->devfn == thisvf_devfn)
4945 break;
4946 pvfdev = pci_get_device(hw->vendor_id,
4947 device_id, pvfdev);
4948 }
4949
4950 if (pvfdev)
4951 adapter->vf_data[vf].vfdev = pvfdev;
4952 else
4953 dev_err(&pdev->dev,
4954 "Couldn't find pci dev ptr for VF %4.4x\n",
4955 thisvf_devfn);
4956 return pvfdev != NULL;
4957 }
4958
4959 static int igb_find_enabled_vfs(struct igb_adapter *adapter)
4960 {
4961 struct e1000_hw *hw = &adapter->hw;
4962 struct pci_dev *pdev = adapter->pdev;
4963 struct pci_dev *pvfdev;
4964 u16 vf_devfn = 0;
4965 u16 vf_stride;
4966 unsigned int device_id;
4967 int vfs_found = 0;
4968
4969 switch (adapter->hw.mac.type) {
4970 case e1000_82576:
4971 device_id = IGB_82576_VF_DEV_ID;
4972 /* VF Stride for 82576 is 2 */
4973 vf_stride = 2;
4974 break;
4975 case e1000_i350:
4976 device_id = IGB_I350_VF_DEV_ID;
4977 /* VF Stride for I350 is 4 */
4978 vf_stride = 4;
4979 break;
4980 default:
4981 device_id = 0;
4982 vf_stride = 0;
4983 break;
4984 }
4985
4986 vf_devfn = pdev->devfn + 0x80;
4987 pvfdev = pci_get_device(hw->vendor_id, device_id, NULL);
4988 while (pvfdev) {
4989 if (pvfdev->devfn == vf_devfn &&
4990 (pvfdev->bus->number >= pdev->bus->number))
4991 vfs_found++;
4992 vf_devfn += vf_stride;
4993 pvfdev = pci_get_device(hw->vendor_id,
4994 device_id, pvfdev);
4995 }
4996
4997 return vfs_found;
4998 }
4999
5000 static int igb_check_vf_assignment(struct igb_adapter *adapter)
5001 {
5002 int i;
5003 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5004 if (adapter->vf_data[i].vfdev) {
5005 if (adapter->vf_data[i].vfdev->dev_flags &
5006 PCI_DEV_FLAGS_ASSIGNED)
5007 return true;
5008 }
5009 }
5010 return false;
5011 }
5012
5013 #endif
5014 static void igb_ping_all_vfs(struct igb_adapter *adapter)
5015 {
5016 struct e1000_hw *hw = &adapter->hw;
5017 u32 ping;
5018 int i;
5019
5020 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5021 ping = E1000_PF_CONTROL_MSG;
5022 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5023 ping |= E1000_VT_MSGTYPE_CTS;
5024 igb_write_mbx(hw, &ping, 1, i);
5025 }
5026 }
5027
5028 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5029 {
5030 struct e1000_hw *hw = &adapter->hw;
5031 u32 vmolr = rd32(E1000_VMOLR(vf));
5032 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5033
5034 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5035 IGB_VF_FLAG_MULTI_PROMISC);
5036 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5037
5038 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5039 vmolr |= E1000_VMOLR_MPME;
5040 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5041 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5042 } else {
5043 /*
5044 * if we have hashes and we are clearing a multicast promisc
5045 * flag we need to write the hashes to the MTA as this step
5046 * was previously skipped
5047 */
5048 if (vf_data->num_vf_mc_hashes > 30) {
5049 vmolr |= E1000_VMOLR_MPME;
5050 } else if (vf_data->num_vf_mc_hashes) {
5051 int j;
5052 vmolr |= E1000_VMOLR_ROMPE;
5053 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5054 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5055 }
5056 }
5057
5058 wr32(E1000_VMOLR(vf), vmolr);
5059
5060 /* there are flags left unprocessed, likely not supported */
5061 if (*msgbuf & E1000_VT_MSGINFO_MASK)
5062 return -EINVAL;
5063
5064 return 0;
5065
5066 }
5067
5068 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5069 u32 *msgbuf, u32 vf)
5070 {
5071 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5072 u16 *hash_list = (u16 *)&msgbuf[1];
5073 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5074 int i;
5075
5076 /* salt away the number of multicast addresses assigned
5077 * to this VF for later use to restore when the PF multi cast
5078 * list changes
5079 */
5080 vf_data->num_vf_mc_hashes = n;
5081
5082 /* only up to 30 hash values supported */
5083 if (n > 30)
5084 n = 30;
5085
5086 /* store the hashes for later use */
5087 for (i = 0; i < n; i++)
5088 vf_data->vf_mc_hashes[i] = hash_list[i];
5089
5090 /* Flush and reset the mta with the new values */
5091 igb_set_rx_mode(adapter->netdev);
5092
5093 return 0;
5094 }
5095
5096 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5097 {
5098 struct e1000_hw *hw = &adapter->hw;
5099 struct vf_data_storage *vf_data;
5100 int i, j;
5101
5102 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5103 u32 vmolr = rd32(E1000_VMOLR(i));
5104 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5105
5106 vf_data = &adapter->vf_data[i];
5107
5108 if ((vf_data->num_vf_mc_hashes > 30) ||
5109 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5110 vmolr |= E1000_VMOLR_MPME;
5111 } else if (vf_data->num_vf_mc_hashes) {
5112 vmolr |= E1000_VMOLR_ROMPE;
5113 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5114 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5115 }
5116 wr32(E1000_VMOLR(i), vmolr);
5117 }
5118 }
5119
5120 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5121 {
5122 struct e1000_hw *hw = &adapter->hw;
5123 u32 pool_mask, reg, vid;
5124 int i;
5125
5126 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5127
5128 /* Find the vlan filter for this id */
5129 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5130 reg = rd32(E1000_VLVF(i));
5131
5132 /* remove the vf from the pool */
5133 reg &= ~pool_mask;
5134
5135 /* if pool is empty then remove entry from vfta */
5136 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5137 (reg & E1000_VLVF_VLANID_ENABLE)) {
5138 reg = 0;
5139 vid = reg & E1000_VLVF_VLANID_MASK;
5140 igb_vfta_set(hw, vid, false);
5141 }
5142
5143 wr32(E1000_VLVF(i), reg);
5144 }
5145
5146 adapter->vf_data[vf].vlans_enabled = 0;
5147 }
5148
5149 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5150 {
5151 struct e1000_hw *hw = &adapter->hw;
5152 u32 reg, i;
5153
5154 /* The vlvf table only exists on 82576 hardware and newer */
5155 if (hw->mac.type < e1000_82576)
5156 return -1;
5157
5158 /* we only need to do this if VMDq is enabled */
5159 if (!adapter->vfs_allocated_count)
5160 return -1;
5161
5162 /* Find the vlan filter for this id */
5163 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5164 reg = rd32(E1000_VLVF(i));
5165 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5166 vid == (reg & E1000_VLVF_VLANID_MASK))
5167 break;
5168 }
5169
5170 if (add) {
5171 if (i == E1000_VLVF_ARRAY_SIZE) {
5172 /* Did not find a matching VLAN ID entry that was
5173 * enabled. Search for a free filter entry, i.e.
5174 * one without the enable bit set
5175 */
5176 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5177 reg = rd32(E1000_VLVF(i));
5178 if (!(reg & E1000_VLVF_VLANID_ENABLE))
5179 break;
5180 }
5181 }
5182 if (i < E1000_VLVF_ARRAY_SIZE) {
5183 /* Found an enabled/available entry */
5184 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5185
5186 /* if !enabled we need to set this up in vfta */
5187 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5188 /* add VID to filter table */
5189 igb_vfta_set(hw, vid, true);
5190 reg |= E1000_VLVF_VLANID_ENABLE;
5191 }
5192 reg &= ~E1000_VLVF_VLANID_MASK;
5193 reg |= vid;
5194 wr32(E1000_VLVF(i), reg);
5195
5196 /* do not modify RLPML for PF devices */
5197 if (vf >= adapter->vfs_allocated_count)
5198 return 0;
5199
5200 if (!adapter->vf_data[vf].vlans_enabled) {
5201 u32 size;
5202 reg = rd32(E1000_VMOLR(vf));
5203 size = reg & E1000_VMOLR_RLPML_MASK;
5204 size += 4;
5205 reg &= ~E1000_VMOLR_RLPML_MASK;
5206 reg |= size;
5207 wr32(E1000_VMOLR(vf), reg);
5208 }
5209
5210 adapter->vf_data[vf].vlans_enabled++;
5211 }
5212 } else {
5213 if (i < E1000_VLVF_ARRAY_SIZE) {
5214 /* remove vf from the pool */
5215 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5216 /* if pool is empty then remove entry from vfta */
5217 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5218 reg = 0;
5219 igb_vfta_set(hw, vid, false);
5220 }
5221 wr32(E1000_VLVF(i), reg);
5222
5223 /* do not modify RLPML for PF devices */
5224 if (vf >= adapter->vfs_allocated_count)
5225 return 0;
5226
5227 adapter->vf_data[vf].vlans_enabled--;
5228 if (!adapter->vf_data[vf].vlans_enabled) {
5229 u32 size;
5230 reg = rd32(E1000_VMOLR(vf));
5231 size = reg & E1000_VMOLR_RLPML_MASK;
5232 size -= 4;
5233 reg &= ~E1000_VMOLR_RLPML_MASK;
5234 reg |= size;
5235 wr32(E1000_VMOLR(vf), reg);
5236 }
5237 }
5238 }
5239 return 0;
5240 }
5241
5242 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5243 {
5244 struct e1000_hw *hw = &adapter->hw;
5245
5246 if (vid)
5247 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5248 else
5249 wr32(E1000_VMVIR(vf), 0);
5250 }
5251
5252 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5253 int vf, u16 vlan, u8 qos)
5254 {
5255 int err = 0;
5256 struct igb_adapter *adapter = netdev_priv(netdev);
5257
5258 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5259 return -EINVAL;
5260 if (vlan || qos) {
5261 err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5262 if (err)
5263 goto out;
5264 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5265 igb_set_vmolr(adapter, vf, !vlan);
5266 adapter->vf_data[vf].pf_vlan = vlan;
5267 adapter->vf_data[vf].pf_qos = qos;
5268 dev_info(&adapter->pdev->dev,
5269 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5270 if (test_bit(__IGB_DOWN, &adapter->state)) {
5271 dev_warn(&adapter->pdev->dev,
5272 "The VF VLAN has been set,"
5273 " but the PF device is not up.\n");
5274 dev_warn(&adapter->pdev->dev,
5275 "Bring the PF device up before"
5276 " attempting to use the VF device.\n");
5277 }
5278 } else {
5279 igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5280 false, vf);
5281 igb_set_vmvir(adapter, vlan, vf);
5282 igb_set_vmolr(adapter, vf, true);
5283 adapter->vf_data[vf].pf_vlan = 0;
5284 adapter->vf_data[vf].pf_qos = 0;
5285 }
5286 out:
5287 return err;
5288 }
5289
5290 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5291 {
5292 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5293 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5294
5295 return igb_vlvf_set(adapter, vid, add, vf);
5296 }
5297
5298 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5299 {
5300 /* clear flags - except flag that indicates PF has set the MAC */
5301 adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5302 adapter->vf_data[vf].last_nack = jiffies;
5303
5304 /* reset offloads to defaults */
5305 igb_set_vmolr(adapter, vf, true);
5306
5307 /* reset vlans for device */
5308 igb_clear_vf_vfta(adapter, vf);
5309 if (adapter->vf_data[vf].pf_vlan)
5310 igb_ndo_set_vf_vlan(adapter->netdev, vf,
5311 adapter->vf_data[vf].pf_vlan,
5312 adapter->vf_data[vf].pf_qos);
5313 else
5314 igb_clear_vf_vfta(adapter, vf);
5315
5316 /* reset multicast table array for vf */
5317 adapter->vf_data[vf].num_vf_mc_hashes = 0;
5318
5319 /* Flush and reset the mta with the new values */
5320 igb_set_rx_mode(adapter->netdev);
5321 }
5322
5323 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5324 {
5325 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5326
5327 /* generate a new mac address as we were hotplug removed/added */
5328 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5329 random_ether_addr(vf_mac);
5330
5331 /* process remaining reset events */
5332 igb_vf_reset(adapter, vf);
5333 }
5334
5335 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5336 {
5337 struct e1000_hw *hw = &adapter->hw;
5338 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5339 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5340 u32 reg, msgbuf[3];
5341 u8 *addr = (u8 *)(&msgbuf[1]);
5342
5343 /* process all the same items cleared in a function level reset */
5344 igb_vf_reset(adapter, vf);
5345
5346 /* set vf mac address */
5347 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
5348
5349 /* enable transmit and receive for vf */
5350 reg = rd32(E1000_VFTE);
5351 wr32(E1000_VFTE, reg | (1 << vf));
5352 reg = rd32(E1000_VFRE);
5353 wr32(E1000_VFRE, reg | (1 << vf));
5354
5355 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
5356
5357 /* reply to reset with ack and vf mac address */
5358 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
5359 memcpy(addr, vf_mac, 6);
5360 igb_write_mbx(hw, msgbuf, 3, vf);
5361 }
5362
5363 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
5364 {
5365 /*
5366 * The VF MAC Address is stored in a packed array of bytes
5367 * starting at the second 32 bit word of the msg array
5368 */
5369 unsigned char *addr = (char *)&msg[1];
5370 int err = -1;
5371
5372 if (is_valid_ether_addr(addr))
5373 err = igb_set_vf_mac(adapter, vf, addr);
5374
5375 return err;
5376 }
5377
5378 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
5379 {
5380 struct e1000_hw *hw = &adapter->hw;
5381 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5382 u32 msg = E1000_VT_MSGTYPE_NACK;
5383
5384 /* if device isn't clear to send it shouldn't be reading either */
5385 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
5386 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
5387 igb_write_mbx(hw, &msg, 1, vf);
5388 vf_data->last_nack = jiffies;
5389 }
5390 }
5391
5392 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
5393 {
5394 struct pci_dev *pdev = adapter->pdev;
5395 u32 msgbuf[E1000_VFMAILBOX_SIZE];
5396 struct e1000_hw *hw = &adapter->hw;
5397 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5398 s32 retval;
5399
5400 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
5401
5402 if (retval) {
5403 /* if receive failed revoke VF CTS stats and restart init */
5404 dev_err(&pdev->dev, "Error receiving message from VF\n");
5405 vf_data->flags &= ~IGB_VF_FLAG_CTS;
5406 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5407 return;
5408 goto out;
5409 }
5410
5411 /* this is a message we already processed, do nothing */
5412 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
5413 return;
5414
5415 /*
5416 * until the vf completes a reset it should not be
5417 * allowed to start any configuration.
5418 */
5419
5420 if (msgbuf[0] == E1000_VF_RESET) {
5421 igb_vf_reset_msg(adapter, vf);
5422 return;
5423 }
5424
5425 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
5426 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5427 return;
5428 retval = -1;
5429 goto out;
5430 }
5431
5432 switch ((msgbuf[0] & 0xFFFF)) {
5433 case E1000_VF_SET_MAC_ADDR:
5434 retval = -EINVAL;
5435 if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
5436 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
5437 else
5438 dev_warn(&pdev->dev,
5439 "VF %d attempted to override administratively "
5440 "set MAC address\nReload the VF driver to "
5441 "resume operations\n", vf);
5442 break;
5443 case E1000_VF_SET_PROMISC:
5444 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
5445 break;
5446 case E1000_VF_SET_MULTICAST:
5447 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
5448 break;
5449 case E1000_VF_SET_LPE:
5450 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
5451 break;
5452 case E1000_VF_SET_VLAN:
5453 retval = -1;
5454 if (vf_data->pf_vlan)
5455 dev_warn(&pdev->dev,
5456 "VF %d attempted to override administratively "
5457 "set VLAN tag\nReload the VF driver to "
5458 "resume operations\n", vf);
5459 else
5460 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
5461 break;
5462 default:
5463 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
5464 retval = -1;
5465 break;
5466 }
5467
5468 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
5469 out:
5470 /* notify the VF of the results of what it sent us */
5471 if (retval)
5472 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
5473 else
5474 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
5475
5476 igb_write_mbx(hw, msgbuf, 1, vf);
5477 }
5478
5479 static void igb_msg_task(struct igb_adapter *adapter)
5480 {
5481 struct e1000_hw *hw = &adapter->hw;
5482 u32 vf;
5483
5484 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
5485 /* process any reset requests */
5486 if (!igb_check_for_rst(hw, vf))
5487 igb_vf_reset_event(adapter, vf);
5488
5489 /* process any messages pending */
5490 if (!igb_check_for_msg(hw, vf))
5491 igb_rcv_msg_from_vf(adapter, vf);
5492
5493 /* process any acks */
5494 if (!igb_check_for_ack(hw, vf))
5495 igb_rcv_ack_from_vf(adapter, vf);
5496 }
5497 }
5498
5499 /**
5500 * igb_set_uta - Set unicast filter table address
5501 * @adapter: board private structure
5502 *
5503 * The unicast table address is a register array of 32-bit registers.
5504 * The table is meant to be used in a way similar to how the MTA is used
5505 * however due to certain limitations in the hardware it is necessary to
5506 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
5507 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
5508 **/
5509 static void igb_set_uta(struct igb_adapter *adapter)
5510 {
5511 struct e1000_hw *hw = &adapter->hw;
5512 int i;
5513
5514 /* The UTA table only exists on 82576 hardware and newer */
5515 if (hw->mac.type < e1000_82576)
5516 return;
5517
5518 /* we only need to do this if VMDq is enabled */
5519 if (!adapter->vfs_allocated_count)
5520 return;
5521
5522 for (i = 0; i < hw->mac.uta_reg_count; i++)
5523 array_wr32(E1000_UTA, i, ~0);
5524 }
5525
5526 /**
5527 * igb_intr_msi - Interrupt Handler
5528 * @irq: interrupt number
5529 * @data: pointer to a network interface device structure
5530 **/
5531 static irqreturn_t igb_intr_msi(int irq, void *data)
5532 {
5533 struct igb_adapter *adapter = data;
5534 struct igb_q_vector *q_vector = adapter->q_vector[0];
5535 struct e1000_hw *hw = &adapter->hw;
5536 /* read ICR disables interrupts using IAM */
5537 u32 icr = rd32(E1000_ICR);
5538
5539 igb_write_itr(q_vector);
5540
5541 if (icr & E1000_ICR_DRSTA)
5542 schedule_work(&adapter->reset_task);
5543
5544 if (icr & E1000_ICR_DOUTSYNC) {
5545 /* HW is reporting DMA is out of sync */
5546 adapter->stats.doosync++;
5547 }
5548
5549 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5550 hw->mac.get_link_status = 1;
5551 if (!test_bit(__IGB_DOWN, &adapter->state))
5552 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5553 }
5554
5555 napi_schedule(&q_vector->napi);
5556
5557 return IRQ_HANDLED;
5558 }
5559
5560 /**
5561 * igb_intr - Legacy Interrupt Handler
5562 * @irq: interrupt number
5563 * @data: pointer to a network interface device structure
5564 **/
5565 static irqreturn_t igb_intr(int irq, void *data)
5566 {
5567 struct igb_adapter *adapter = data;
5568 struct igb_q_vector *q_vector = adapter->q_vector[0];
5569 struct e1000_hw *hw = &adapter->hw;
5570 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
5571 * need for the IMC write */
5572 u32 icr = rd32(E1000_ICR);
5573
5574 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5575 * not set, then the adapter didn't send an interrupt */
5576 if (!(icr & E1000_ICR_INT_ASSERTED))
5577 return IRQ_NONE;
5578
5579 igb_write_itr(q_vector);
5580
5581 if (icr & E1000_ICR_DRSTA)
5582 schedule_work(&adapter->reset_task);
5583
5584 if (icr & E1000_ICR_DOUTSYNC) {
5585 /* HW is reporting DMA is out of sync */
5586 adapter->stats.doosync++;
5587 }
5588
5589 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5590 hw->mac.get_link_status = 1;
5591 /* guard against interrupt when we're going down */
5592 if (!test_bit(__IGB_DOWN, &adapter->state))
5593 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5594 }
5595
5596 napi_schedule(&q_vector->napi);
5597
5598 return IRQ_HANDLED;
5599 }
5600
5601 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
5602 {
5603 struct igb_adapter *adapter = q_vector->adapter;
5604 struct e1000_hw *hw = &adapter->hw;
5605
5606 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
5607 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
5608 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
5609 igb_set_itr(q_vector);
5610 else
5611 igb_update_ring_itr(q_vector);
5612 }
5613
5614 if (!test_bit(__IGB_DOWN, &adapter->state)) {
5615 if (adapter->msix_entries)
5616 wr32(E1000_EIMS, q_vector->eims_value);
5617 else
5618 igb_irq_enable(adapter);
5619 }
5620 }
5621
5622 /**
5623 * igb_poll - NAPI Rx polling callback
5624 * @napi: napi polling structure
5625 * @budget: count of how many packets we should handle
5626 **/
5627 static int igb_poll(struct napi_struct *napi, int budget)
5628 {
5629 struct igb_q_vector *q_vector = container_of(napi,
5630 struct igb_q_vector,
5631 napi);
5632 bool clean_complete = true;
5633
5634 #ifdef CONFIG_IGB_DCA
5635 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
5636 igb_update_dca(q_vector);
5637 #endif
5638 if (q_vector->tx.ring)
5639 clean_complete = igb_clean_tx_irq(q_vector);
5640
5641 if (q_vector->rx.ring)
5642 clean_complete &= igb_clean_rx_irq(q_vector, budget);
5643
5644 /* If all work not completed, return budget and keep polling */
5645 if (!clean_complete)
5646 return budget;
5647
5648 /* If not enough Rx work done, exit the polling mode */
5649 napi_complete(napi);
5650 igb_ring_irq_enable(q_vector);
5651
5652 return 0;
5653 }
5654
5655 #ifdef CONFIG_IGB_PTP
5656 /**
5657 * igb_tx_hwtstamp - utility function which checks for TX time stamp
5658 * @q_vector: pointer to q_vector containing needed info
5659 * @buffer: pointer to igb_tx_buffer structure
5660 *
5661 * If we were asked to do hardware stamping and such a time stamp is
5662 * available, then it must have been for this skb here because we only
5663 * allow only one such packet into the queue.
5664 */
5665 static void igb_tx_hwtstamp(struct igb_q_vector *q_vector,
5666 struct igb_tx_buffer *buffer_info)
5667 {
5668 struct igb_adapter *adapter = q_vector->adapter;
5669 struct e1000_hw *hw = &adapter->hw;
5670 struct skb_shared_hwtstamps shhwtstamps;
5671 u64 regval;
5672
5673 /* if skb does not support hw timestamp or TX stamp not valid exit */
5674 if (likely(!(buffer_info->tx_flags & IGB_TX_FLAGS_TSTAMP)) ||
5675 !(rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID))
5676 return;
5677
5678 regval = rd32(E1000_TXSTMPL);
5679 regval |= (u64)rd32(E1000_TXSTMPH) << 32;
5680
5681 igb_systim_to_hwtstamp(adapter, &shhwtstamps, regval);
5682 skb_tstamp_tx(buffer_info->skb, &shhwtstamps);
5683 }
5684
5685 #endif
5686 /**
5687 * igb_clean_tx_irq - Reclaim resources after transmit completes
5688 * @q_vector: pointer to q_vector containing needed info
5689 * returns true if ring is completely cleaned
5690 **/
5691 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
5692 {
5693 struct igb_adapter *adapter = q_vector->adapter;
5694 struct igb_ring *tx_ring = q_vector->tx.ring;
5695 struct igb_tx_buffer *tx_buffer;
5696 union e1000_adv_tx_desc *tx_desc, *eop_desc;
5697 unsigned int total_bytes = 0, total_packets = 0;
5698 unsigned int budget = q_vector->tx.work_limit;
5699 unsigned int i = tx_ring->next_to_clean;
5700
5701 if (test_bit(__IGB_DOWN, &adapter->state))
5702 return true;
5703
5704 tx_buffer = &tx_ring->tx_buffer_info[i];
5705 tx_desc = IGB_TX_DESC(tx_ring, i);
5706 i -= tx_ring->count;
5707
5708 for (; budget; budget--) {
5709 eop_desc = tx_buffer->next_to_watch;
5710
5711 /* prevent any other reads prior to eop_desc */
5712 rmb();
5713
5714 /* if next_to_watch is not set then there is no work pending */
5715 if (!eop_desc)
5716 break;
5717
5718 /* if DD is not set pending work has not been completed */
5719 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
5720 break;
5721
5722 /* clear next_to_watch to prevent false hangs */
5723 tx_buffer->next_to_watch = NULL;
5724
5725 /* update the statistics for this packet */
5726 total_bytes += tx_buffer->bytecount;
5727 total_packets += tx_buffer->gso_segs;
5728
5729 #ifdef CONFIG_IGB_PTP
5730 /* retrieve hardware timestamp */
5731 igb_tx_hwtstamp(q_vector, tx_buffer);
5732
5733 #endif
5734 /* free the skb */
5735 dev_kfree_skb_any(tx_buffer->skb);
5736 tx_buffer->skb = NULL;
5737
5738 /* unmap skb header data */
5739 dma_unmap_single(tx_ring->dev,
5740 tx_buffer->dma,
5741 tx_buffer->length,
5742 DMA_TO_DEVICE);
5743
5744 /* clear last DMA location and unmap remaining buffers */
5745 while (tx_desc != eop_desc) {
5746 tx_buffer->dma = 0;
5747
5748 tx_buffer++;
5749 tx_desc++;
5750 i++;
5751 if (unlikely(!i)) {
5752 i -= tx_ring->count;
5753 tx_buffer = tx_ring->tx_buffer_info;
5754 tx_desc = IGB_TX_DESC(tx_ring, 0);
5755 }
5756
5757 /* unmap any remaining paged data */
5758 if (tx_buffer->dma) {
5759 dma_unmap_page(tx_ring->dev,
5760 tx_buffer->dma,
5761 tx_buffer->length,
5762 DMA_TO_DEVICE);
5763 }
5764 }
5765
5766 /* clear last DMA location */
5767 tx_buffer->dma = 0;
5768
5769 /* move us one more past the eop_desc for start of next pkt */
5770 tx_buffer++;
5771 tx_desc++;
5772 i++;
5773 if (unlikely(!i)) {
5774 i -= tx_ring->count;
5775 tx_buffer = tx_ring->tx_buffer_info;
5776 tx_desc = IGB_TX_DESC(tx_ring, 0);
5777 }
5778 }
5779
5780 netdev_tx_completed_queue(txring_txq(tx_ring),
5781 total_packets, total_bytes);
5782 i += tx_ring->count;
5783 tx_ring->next_to_clean = i;
5784 u64_stats_update_begin(&tx_ring->tx_syncp);
5785 tx_ring->tx_stats.bytes += total_bytes;
5786 tx_ring->tx_stats.packets += total_packets;
5787 u64_stats_update_end(&tx_ring->tx_syncp);
5788 q_vector->tx.total_bytes += total_bytes;
5789 q_vector->tx.total_packets += total_packets;
5790
5791 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
5792 struct e1000_hw *hw = &adapter->hw;
5793
5794 eop_desc = tx_buffer->next_to_watch;
5795
5796 /* Detect a transmit hang in hardware, this serializes the
5797 * check with the clearing of time_stamp and movement of i */
5798 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5799 if (eop_desc &&
5800 time_after(jiffies, tx_buffer->time_stamp +
5801 (adapter->tx_timeout_factor * HZ)) &&
5802 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
5803
5804 /* detected Tx unit hang */
5805 dev_err(tx_ring->dev,
5806 "Detected Tx Unit Hang\n"
5807 " Tx Queue <%d>\n"
5808 " TDH <%x>\n"
5809 " TDT <%x>\n"
5810 " next_to_use <%x>\n"
5811 " next_to_clean <%x>\n"
5812 "buffer_info[next_to_clean]\n"
5813 " time_stamp <%lx>\n"
5814 " next_to_watch <%p>\n"
5815 " jiffies <%lx>\n"
5816 " desc.status <%x>\n",
5817 tx_ring->queue_index,
5818 rd32(E1000_TDH(tx_ring->reg_idx)),
5819 readl(tx_ring->tail),
5820 tx_ring->next_to_use,
5821 tx_ring->next_to_clean,
5822 tx_buffer->time_stamp,
5823 eop_desc,
5824 jiffies,
5825 eop_desc->wb.status);
5826 netif_stop_subqueue(tx_ring->netdev,
5827 tx_ring->queue_index);
5828
5829 /* we are about to reset, no point in enabling stuff */
5830 return true;
5831 }
5832 }
5833
5834 if (unlikely(total_packets &&
5835 netif_carrier_ok(tx_ring->netdev) &&
5836 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
5837 /* Make sure that anybody stopping the queue after this
5838 * sees the new next_to_clean.
5839 */
5840 smp_mb();
5841 if (__netif_subqueue_stopped(tx_ring->netdev,
5842 tx_ring->queue_index) &&
5843 !(test_bit(__IGB_DOWN, &adapter->state))) {
5844 netif_wake_subqueue(tx_ring->netdev,
5845 tx_ring->queue_index);
5846
5847 u64_stats_update_begin(&tx_ring->tx_syncp);
5848 tx_ring->tx_stats.restart_queue++;
5849 u64_stats_update_end(&tx_ring->tx_syncp);
5850 }
5851 }
5852
5853 return !!budget;
5854 }
5855
5856 static inline void igb_rx_checksum(struct igb_ring *ring,
5857 union e1000_adv_rx_desc *rx_desc,
5858 struct sk_buff *skb)
5859 {
5860 skb_checksum_none_assert(skb);
5861
5862 /* Ignore Checksum bit is set */
5863 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
5864 return;
5865
5866 /* Rx checksum disabled via ethtool */
5867 if (!(ring->netdev->features & NETIF_F_RXCSUM))
5868 return;
5869
5870 /* TCP/UDP checksum error bit is set */
5871 if (igb_test_staterr(rx_desc,
5872 E1000_RXDEXT_STATERR_TCPE |
5873 E1000_RXDEXT_STATERR_IPE)) {
5874 /*
5875 * work around errata with sctp packets where the TCPE aka
5876 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
5877 * packets, (aka let the stack check the crc32c)
5878 */
5879 if (!((skb->len == 60) &&
5880 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
5881 u64_stats_update_begin(&ring->rx_syncp);
5882 ring->rx_stats.csum_err++;
5883 u64_stats_update_end(&ring->rx_syncp);
5884 }
5885 /* let the stack verify checksum errors */
5886 return;
5887 }
5888 /* It must be a TCP or UDP packet with a valid checksum */
5889 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
5890 E1000_RXD_STAT_UDPCS))
5891 skb->ip_summed = CHECKSUM_UNNECESSARY;
5892
5893 dev_dbg(ring->dev, "cksum success: bits %08X\n",
5894 le32_to_cpu(rx_desc->wb.upper.status_error));
5895 }
5896
5897 static inline void igb_rx_hash(struct igb_ring *ring,
5898 union e1000_adv_rx_desc *rx_desc,
5899 struct sk_buff *skb)
5900 {
5901 if (ring->netdev->features & NETIF_F_RXHASH)
5902 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
5903 }
5904
5905 #ifdef CONFIG_IGB_PTP
5906 static void igb_rx_hwtstamp(struct igb_q_vector *q_vector,
5907 union e1000_adv_rx_desc *rx_desc,
5908 struct sk_buff *skb)
5909 {
5910 struct igb_adapter *adapter = q_vector->adapter;
5911 struct e1000_hw *hw = &adapter->hw;
5912 u64 regval;
5913
5914 if (!igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP |
5915 E1000_RXDADV_STAT_TS))
5916 return;
5917
5918 /*
5919 * If this bit is set, then the RX registers contain the time stamp. No
5920 * other packet will be time stamped until we read these registers, so
5921 * read the registers to make them available again. Because only one
5922 * packet can be time stamped at a time, we know that the register
5923 * values must belong to this one here and therefore we don't need to
5924 * compare any of the additional attributes stored for it.
5925 *
5926 * If nothing went wrong, then it should have a shared tx_flags that we
5927 * can turn into a skb_shared_hwtstamps.
5928 */
5929 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
5930 u32 *stamp = (u32 *)skb->data;
5931 regval = le32_to_cpu(*(stamp + 2));
5932 regval |= (u64)le32_to_cpu(*(stamp + 3)) << 32;
5933 skb_pull(skb, IGB_TS_HDR_LEN);
5934 } else {
5935 if(!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
5936 return;
5937
5938 regval = rd32(E1000_RXSTMPL);
5939 regval |= (u64)rd32(E1000_RXSTMPH) << 32;
5940 }
5941
5942 igb_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), regval);
5943 }
5944
5945 #endif
5946 static void igb_rx_vlan(struct igb_ring *ring,
5947 union e1000_adv_rx_desc *rx_desc,
5948 struct sk_buff *skb)
5949 {
5950 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
5951 u16 vid;
5952 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
5953 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags))
5954 vid = be16_to_cpu(rx_desc->wb.upper.vlan);
5955 else
5956 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
5957
5958 __vlan_hwaccel_put_tag(skb, vid);
5959 }
5960 }
5961
5962 static inline u16 igb_get_hlen(union e1000_adv_rx_desc *rx_desc)
5963 {
5964 /* HW will not DMA in data larger than the given buffer, even if it
5965 * parses the (NFS, of course) header to be larger. In that case, it
5966 * fills the header buffer and spills the rest into the page.
5967 */
5968 u16 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
5969 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
5970 if (hlen > IGB_RX_HDR_LEN)
5971 hlen = IGB_RX_HDR_LEN;
5972 return hlen;
5973 }
5974
5975 static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, int budget)
5976 {
5977 struct igb_ring *rx_ring = q_vector->rx.ring;
5978 union e1000_adv_rx_desc *rx_desc;
5979 const int current_node = numa_node_id();
5980 unsigned int total_bytes = 0, total_packets = 0;
5981 u16 cleaned_count = igb_desc_unused(rx_ring);
5982 u16 i = rx_ring->next_to_clean;
5983
5984 rx_desc = IGB_RX_DESC(rx_ring, i);
5985
5986 while (igb_test_staterr(rx_desc, E1000_RXD_STAT_DD)) {
5987 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
5988 struct sk_buff *skb = buffer_info->skb;
5989 union e1000_adv_rx_desc *next_rxd;
5990
5991 buffer_info->skb = NULL;
5992 prefetch(skb->data);
5993
5994 i++;
5995 if (i == rx_ring->count)
5996 i = 0;
5997
5998 next_rxd = IGB_RX_DESC(rx_ring, i);
5999 prefetch(next_rxd);
6000
6001 /*
6002 * This memory barrier is needed to keep us from reading
6003 * any other fields out of the rx_desc until we know the
6004 * RXD_STAT_DD bit is set
6005 */
6006 rmb();
6007
6008 if (!skb_is_nonlinear(skb)) {
6009 __skb_put(skb, igb_get_hlen(rx_desc));
6010 dma_unmap_single(rx_ring->dev, buffer_info->dma,
6011 IGB_RX_HDR_LEN,
6012 DMA_FROM_DEVICE);
6013 buffer_info->dma = 0;
6014 }
6015
6016 if (rx_desc->wb.upper.length) {
6017 u16 length = le16_to_cpu(rx_desc->wb.upper.length);
6018
6019 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
6020 buffer_info->page,
6021 buffer_info->page_offset,
6022 length);
6023
6024 skb->len += length;
6025 skb->data_len += length;
6026 skb->truesize += PAGE_SIZE / 2;
6027
6028 if ((page_count(buffer_info->page) != 1) ||
6029 (page_to_nid(buffer_info->page) != current_node))
6030 buffer_info->page = NULL;
6031 else
6032 get_page(buffer_info->page);
6033
6034 dma_unmap_page(rx_ring->dev, buffer_info->page_dma,
6035 PAGE_SIZE / 2, DMA_FROM_DEVICE);
6036 buffer_info->page_dma = 0;
6037 }
6038
6039 if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)) {
6040 struct igb_rx_buffer *next_buffer;
6041 next_buffer = &rx_ring->rx_buffer_info[i];
6042 buffer_info->skb = next_buffer->skb;
6043 buffer_info->dma = next_buffer->dma;
6044 next_buffer->skb = skb;
6045 next_buffer->dma = 0;
6046 goto next_desc;
6047 }
6048
6049 if (unlikely((igb_test_staterr(rx_desc,
6050 E1000_RXDEXT_ERR_FRAME_ERR_MASK))
6051 && !(rx_ring->netdev->features & NETIF_F_RXALL))) {
6052 dev_kfree_skb_any(skb);
6053 goto next_desc;
6054 }
6055
6056 #ifdef CONFIG_IGB_PTP
6057 igb_rx_hwtstamp(q_vector, rx_desc, skb);
6058 #endif
6059 igb_rx_hash(rx_ring, rx_desc, skb);
6060 igb_rx_checksum(rx_ring, rx_desc, skb);
6061 igb_rx_vlan(rx_ring, rx_desc, skb);
6062
6063 total_bytes += skb->len;
6064 total_packets++;
6065
6066 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6067
6068 napi_gro_receive(&q_vector->napi, skb);
6069
6070 budget--;
6071 next_desc:
6072 if (!budget)
6073 break;
6074
6075 cleaned_count++;
6076 /* return some buffers to hardware, one at a time is too slow */
6077 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6078 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6079 cleaned_count = 0;
6080 }
6081
6082 /* use prefetched values */
6083 rx_desc = next_rxd;
6084 }
6085
6086 rx_ring->next_to_clean = i;
6087 u64_stats_update_begin(&rx_ring->rx_syncp);
6088 rx_ring->rx_stats.packets += total_packets;
6089 rx_ring->rx_stats.bytes += total_bytes;
6090 u64_stats_update_end(&rx_ring->rx_syncp);
6091 q_vector->rx.total_packets += total_packets;
6092 q_vector->rx.total_bytes += total_bytes;
6093
6094 if (cleaned_count)
6095 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6096
6097 return !!budget;
6098 }
6099
6100 static bool igb_alloc_mapped_skb(struct igb_ring *rx_ring,
6101 struct igb_rx_buffer *bi)
6102 {
6103 struct sk_buff *skb = bi->skb;
6104 dma_addr_t dma = bi->dma;
6105
6106 if (dma)
6107 return true;
6108
6109 if (likely(!skb)) {
6110 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
6111 IGB_RX_HDR_LEN);
6112 bi->skb = skb;
6113 if (!skb) {
6114 rx_ring->rx_stats.alloc_failed++;
6115 return false;
6116 }
6117
6118 /* initialize skb for ring */
6119 skb_record_rx_queue(skb, rx_ring->queue_index);
6120 }
6121
6122 dma = dma_map_single(rx_ring->dev, skb->data,
6123 IGB_RX_HDR_LEN, DMA_FROM_DEVICE);
6124
6125 if (dma_mapping_error(rx_ring->dev, dma)) {
6126 rx_ring->rx_stats.alloc_failed++;
6127 return false;
6128 }
6129
6130 bi->dma = dma;
6131 return true;
6132 }
6133
6134 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
6135 struct igb_rx_buffer *bi)
6136 {
6137 struct page *page = bi->page;
6138 dma_addr_t page_dma = bi->page_dma;
6139 unsigned int page_offset = bi->page_offset ^ (PAGE_SIZE / 2);
6140
6141 if (page_dma)
6142 return true;
6143
6144 if (!page) {
6145 page = alloc_page(GFP_ATOMIC | __GFP_COLD);
6146 bi->page = page;
6147 if (unlikely(!page)) {
6148 rx_ring->rx_stats.alloc_failed++;
6149 return false;
6150 }
6151 }
6152
6153 page_dma = dma_map_page(rx_ring->dev, page,
6154 page_offset, PAGE_SIZE / 2,
6155 DMA_FROM_DEVICE);
6156
6157 if (dma_mapping_error(rx_ring->dev, page_dma)) {
6158 rx_ring->rx_stats.alloc_failed++;
6159 return false;
6160 }
6161
6162 bi->page_dma = page_dma;
6163 bi->page_offset = page_offset;
6164 return true;
6165 }
6166
6167 /**
6168 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
6169 * @adapter: address of board private structure
6170 **/
6171 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
6172 {
6173 union e1000_adv_rx_desc *rx_desc;
6174 struct igb_rx_buffer *bi;
6175 u16 i = rx_ring->next_to_use;
6176
6177 rx_desc = IGB_RX_DESC(rx_ring, i);
6178 bi = &rx_ring->rx_buffer_info[i];
6179 i -= rx_ring->count;
6180
6181 while (cleaned_count--) {
6182 if (!igb_alloc_mapped_skb(rx_ring, bi))
6183 break;
6184
6185 /* Refresh the desc even if buffer_addrs didn't change
6186 * because each write-back erases this info. */
6187 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
6188
6189 if (!igb_alloc_mapped_page(rx_ring, bi))
6190 break;
6191
6192 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
6193
6194 rx_desc++;
6195 bi++;
6196 i++;
6197 if (unlikely(!i)) {
6198 rx_desc = IGB_RX_DESC(rx_ring, 0);
6199 bi = rx_ring->rx_buffer_info;
6200 i -= rx_ring->count;
6201 }
6202
6203 /* clear the hdr_addr for the next_to_use descriptor */
6204 rx_desc->read.hdr_addr = 0;
6205 }
6206
6207 i += rx_ring->count;
6208
6209 if (rx_ring->next_to_use != i) {
6210 rx_ring->next_to_use = i;
6211
6212 /* Force memory writes to complete before letting h/w
6213 * know there are new descriptors to fetch. (Only
6214 * applicable for weak-ordered memory model archs,
6215 * such as IA-64). */
6216 wmb();
6217 writel(i, rx_ring->tail);
6218 }
6219 }
6220
6221 /**
6222 * igb_mii_ioctl -
6223 * @netdev:
6224 * @ifreq:
6225 * @cmd:
6226 **/
6227 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6228 {
6229 struct igb_adapter *adapter = netdev_priv(netdev);
6230 struct mii_ioctl_data *data = if_mii(ifr);
6231
6232 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6233 return -EOPNOTSUPP;
6234
6235 switch (cmd) {
6236 case SIOCGMIIPHY:
6237 data->phy_id = adapter->hw.phy.addr;
6238 break;
6239 case SIOCGMIIREG:
6240 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
6241 &data->val_out))
6242 return -EIO;
6243 break;
6244 case SIOCSMIIREG:
6245 default:
6246 return -EOPNOTSUPP;
6247 }
6248 return 0;
6249 }
6250
6251 /**
6252 * igb_hwtstamp_ioctl - control hardware time stamping
6253 * @netdev:
6254 * @ifreq:
6255 * @cmd:
6256 *
6257 * Outgoing time stamping can be enabled and disabled. Play nice and
6258 * disable it when requested, although it shouldn't case any overhead
6259 * when no packet needs it. At most one packet in the queue may be
6260 * marked for time stamping, otherwise it would be impossible to tell
6261 * for sure to which packet the hardware time stamp belongs.
6262 *
6263 * Incoming time stamping has to be configured via the hardware
6264 * filters. Not all combinations are supported, in particular event
6265 * type has to be specified. Matching the kind of event packet is
6266 * not supported, with the exception of "all V2 events regardless of
6267 * level 2 or 4".
6268 *
6269 **/
6270 static int igb_hwtstamp_ioctl(struct net_device *netdev,
6271 struct ifreq *ifr, int cmd)
6272 {
6273 struct igb_adapter *adapter = netdev_priv(netdev);
6274 struct e1000_hw *hw = &adapter->hw;
6275 struct hwtstamp_config config;
6276 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
6277 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
6278 u32 tsync_rx_cfg = 0;
6279 bool is_l4 = false;
6280 bool is_l2 = false;
6281 u32 regval;
6282
6283 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
6284 return -EFAULT;
6285
6286 /* reserved for future extensions */
6287 if (config.flags)
6288 return -EINVAL;
6289
6290 switch (config.tx_type) {
6291 case HWTSTAMP_TX_OFF:
6292 tsync_tx_ctl = 0;
6293 case HWTSTAMP_TX_ON:
6294 break;
6295 default:
6296 return -ERANGE;
6297 }
6298
6299 switch (config.rx_filter) {
6300 case HWTSTAMP_FILTER_NONE:
6301 tsync_rx_ctl = 0;
6302 break;
6303 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
6304 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
6305 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
6306 case HWTSTAMP_FILTER_ALL:
6307 /*
6308 * register TSYNCRXCFG must be set, therefore it is not
6309 * possible to time stamp both Sync and Delay_Req messages
6310 * => fall back to time stamping all packets
6311 */
6312 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
6313 config.rx_filter = HWTSTAMP_FILTER_ALL;
6314 break;
6315 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
6316 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
6317 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
6318 is_l4 = true;
6319 break;
6320 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
6321 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
6322 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
6323 is_l4 = true;
6324 break;
6325 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6326 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6327 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
6328 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
6329 is_l2 = true;
6330 is_l4 = true;
6331 config.rx_filter = HWTSTAMP_FILTER_SOME;
6332 break;
6333 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6334 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6335 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
6336 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
6337 is_l2 = true;
6338 is_l4 = true;
6339 config.rx_filter = HWTSTAMP_FILTER_SOME;
6340 break;
6341 case HWTSTAMP_FILTER_PTP_V2_EVENT:
6342 case HWTSTAMP_FILTER_PTP_V2_SYNC:
6343 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6344 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
6345 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
6346 is_l2 = true;
6347 is_l4 = true;
6348 break;
6349 default:
6350 return -ERANGE;
6351 }
6352
6353 if (hw->mac.type == e1000_82575) {
6354 if (tsync_rx_ctl | tsync_tx_ctl)
6355 return -EINVAL;
6356 return 0;
6357 }
6358
6359 /*
6360 * Per-packet timestamping only works if all packets are
6361 * timestamped, so enable timestamping in all packets as
6362 * long as one rx filter was configured.
6363 */
6364 if ((hw->mac.type >= e1000_82580) && tsync_rx_ctl) {
6365 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
6366 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
6367 }
6368
6369 /* enable/disable TX */
6370 regval = rd32(E1000_TSYNCTXCTL);
6371 regval &= ~E1000_TSYNCTXCTL_ENABLED;
6372 regval |= tsync_tx_ctl;
6373 wr32(E1000_TSYNCTXCTL, regval);
6374
6375 /* enable/disable RX */
6376 regval = rd32(E1000_TSYNCRXCTL);
6377 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
6378 regval |= tsync_rx_ctl;
6379 wr32(E1000_TSYNCRXCTL, regval);
6380
6381 /* define which PTP packets are time stamped */
6382 wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
6383
6384 /* define ethertype filter for timestamped packets */
6385 if (is_l2)
6386 wr32(E1000_ETQF(3),
6387 (E1000_ETQF_FILTER_ENABLE | /* enable filter */
6388 E1000_ETQF_1588 | /* enable timestamping */
6389 ETH_P_1588)); /* 1588 eth protocol type */
6390 else
6391 wr32(E1000_ETQF(3), 0);
6392
6393 #define PTP_PORT 319
6394 /* L4 Queue Filter[3]: filter by destination port and protocol */
6395 if (is_l4) {
6396 u32 ftqf = (IPPROTO_UDP /* UDP */
6397 | E1000_FTQF_VF_BP /* VF not compared */
6398 | E1000_FTQF_1588_TIME_STAMP /* Enable Timestamping */
6399 | E1000_FTQF_MASK); /* mask all inputs */
6400 ftqf &= ~E1000_FTQF_MASK_PROTO_BP; /* enable protocol check */
6401
6402 wr32(E1000_IMIR(3), htons(PTP_PORT));
6403 wr32(E1000_IMIREXT(3),
6404 (E1000_IMIREXT_SIZE_BP | E1000_IMIREXT_CTRL_BP));
6405 if (hw->mac.type == e1000_82576) {
6406 /* enable source port check */
6407 wr32(E1000_SPQF(3), htons(PTP_PORT));
6408 ftqf &= ~E1000_FTQF_MASK_SOURCE_PORT_BP;
6409 }
6410 wr32(E1000_FTQF(3), ftqf);
6411 } else {
6412 wr32(E1000_FTQF(3), E1000_FTQF_MASK);
6413 }
6414 wrfl();
6415
6416 adapter->hwtstamp_config = config;
6417
6418 /* clear TX/RX time stamp registers, just to be sure */
6419 regval = rd32(E1000_TXSTMPH);
6420 regval = rd32(E1000_RXSTMPH);
6421
6422 return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
6423 -EFAULT : 0;
6424 }
6425
6426 /**
6427 * igb_ioctl -
6428 * @netdev:
6429 * @ifreq:
6430 * @cmd:
6431 **/
6432 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6433 {
6434 switch (cmd) {
6435 case SIOCGMIIPHY:
6436 case SIOCGMIIREG:
6437 case SIOCSMIIREG:
6438 return igb_mii_ioctl(netdev, ifr, cmd);
6439 case SIOCSHWTSTAMP:
6440 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
6441 default:
6442 return -EOPNOTSUPP;
6443 }
6444 }
6445
6446 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6447 {
6448 struct igb_adapter *adapter = hw->back;
6449 u16 cap_offset;
6450
6451 cap_offset = adapter->pdev->pcie_cap;
6452 if (!cap_offset)
6453 return -E1000_ERR_CONFIG;
6454
6455 pci_read_config_word(adapter->pdev, cap_offset + reg, value);
6456
6457 return 0;
6458 }
6459
6460 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6461 {
6462 struct igb_adapter *adapter = hw->back;
6463 u16 cap_offset;
6464
6465 cap_offset = adapter->pdev->pcie_cap;
6466 if (!cap_offset)
6467 return -E1000_ERR_CONFIG;
6468
6469 pci_write_config_word(adapter->pdev, cap_offset + reg, *value);
6470
6471 return 0;
6472 }
6473
6474 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
6475 {
6476 struct igb_adapter *adapter = netdev_priv(netdev);
6477 struct e1000_hw *hw = &adapter->hw;
6478 u32 ctrl, rctl;
6479 bool enable = !!(features & NETIF_F_HW_VLAN_RX);
6480
6481 if (enable) {
6482 /* enable VLAN tag insert/strip */
6483 ctrl = rd32(E1000_CTRL);
6484 ctrl |= E1000_CTRL_VME;
6485 wr32(E1000_CTRL, ctrl);
6486
6487 /* Disable CFI check */
6488 rctl = rd32(E1000_RCTL);
6489 rctl &= ~E1000_RCTL_CFIEN;
6490 wr32(E1000_RCTL, rctl);
6491 } else {
6492 /* disable VLAN tag insert/strip */
6493 ctrl = rd32(E1000_CTRL);
6494 ctrl &= ~E1000_CTRL_VME;
6495 wr32(E1000_CTRL, ctrl);
6496 }
6497
6498 igb_rlpml_set(adapter);
6499 }
6500
6501 static int igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
6502 {
6503 struct igb_adapter *adapter = netdev_priv(netdev);
6504 struct e1000_hw *hw = &adapter->hw;
6505 int pf_id = adapter->vfs_allocated_count;
6506
6507 /* attempt to add filter to vlvf array */
6508 igb_vlvf_set(adapter, vid, true, pf_id);
6509
6510 /* add the filter since PF can receive vlans w/o entry in vlvf */
6511 igb_vfta_set(hw, vid, true);
6512
6513 set_bit(vid, adapter->active_vlans);
6514
6515 return 0;
6516 }
6517
6518 static int igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
6519 {
6520 struct igb_adapter *adapter = netdev_priv(netdev);
6521 struct e1000_hw *hw = &adapter->hw;
6522 int pf_id = adapter->vfs_allocated_count;
6523 s32 err;
6524
6525 /* remove vlan from VLVF table array */
6526 err = igb_vlvf_set(adapter, vid, false, pf_id);
6527
6528 /* if vid was not present in VLVF just remove it from table */
6529 if (err)
6530 igb_vfta_set(hw, vid, false);
6531
6532 clear_bit(vid, adapter->active_vlans);
6533
6534 return 0;
6535 }
6536
6537 static void igb_restore_vlan(struct igb_adapter *adapter)
6538 {
6539 u16 vid;
6540
6541 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
6542
6543 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
6544 igb_vlan_rx_add_vid(adapter->netdev, vid);
6545 }
6546
6547 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
6548 {
6549 struct pci_dev *pdev = adapter->pdev;
6550 struct e1000_mac_info *mac = &adapter->hw.mac;
6551
6552 mac->autoneg = 0;
6553
6554 /* Make sure dplx is at most 1 bit and lsb of speed is not set
6555 * for the switch() below to work */
6556 if ((spd & 1) || (dplx & ~1))
6557 goto err_inval;
6558
6559 /* Fiber NIC's only allow 1000 Gbps Full duplex */
6560 if ((adapter->hw.phy.media_type == e1000_media_type_internal_serdes) &&
6561 spd != SPEED_1000 &&
6562 dplx != DUPLEX_FULL)
6563 goto err_inval;
6564
6565 switch (spd + dplx) {
6566 case SPEED_10 + DUPLEX_HALF:
6567 mac->forced_speed_duplex = ADVERTISE_10_HALF;
6568 break;
6569 case SPEED_10 + DUPLEX_FULL:
6570 mac->forced_speed_duplex = ADVERTISE_10_FULL;
6571 break;
6572 case SPEED_100 + DUPLEX_HALF:
6573 mac->forced_speed_duplex = ADVERTISE_100_HALF;
6574 break;
6575 case SPEED_100 + DUPLEX_FULL:
6576 mac->forced_speed_duplex = ADVERTISE_100_FULL;
6577 break;
6578 case SPEED_1000 + DUPLEX_FULL:
6579 mac->autoneg = 1;
6580 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
6581 break;
6582 case SPEED_1000 + DUPLEX_HALF: /* not supported */
6583 default:
6584 goto err_inval;
6585 }
6586 return 0;
6587
6588 err_inval:
6589 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
6590 return -EINVAL;
6591 }
6592
6593 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
6594 bool runtime)
6595 {
6596 struct net_device *netdev = pci_get_drvdata(pdev);
6597 struct igb_adapter *adapter = netdev_priv(netdev);
6598 struct e1000_hw *hw = &adapter->hw;
6599 u32 ctrl, rctl, status;
6600 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
6601 #ifdef CONFIG_PM
6602 int retval = 0;
6603 #endif
6604
6605 netif_device_detach(netdev);
6606
6607 if (netif_running(netdev))
6608 __igb_close(netdev, true);
6609
6610 igb_clear_interrupt_scheme(adapter);
6611
6612 #ifdef CONFIG_PM
6613 retval = pci_save_state(pdev);
6614 if (retval)
6615 return retval;
6616 #endif
6617
6618 status = rd32(E1000_STATUS);
6619 if (status & E1000_STATUS_LU)
6620 wufc &= ~E1000_WUFC_LNKC;
6621
6622 if (wufc) {
6623 igb_setup_rctl(adapter);
6624 igb_set_rx_mode(netdev);
6625
6626 /* turn on all-multi mode if wake on multicast is enabled */
6627 if (wufc & E1000_WUFC_MC) {
6628 rctl = rd32(E1000_RCTL);
6629 rctl |= E1000_RCTL_MPE;
6630 wr32(E1000_RCTL, rctl);
6631 }
6632
6633 ctrl = rd32(E1000_CTRL);
6634 /* advertise wake from D3Cold */
6635 #define E1000_CTRL_ADVD3WUC 0x00100000
6636 /* phy power management enable */
6637 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
6638 ctrl |= E1000_CTRL_ADVD3WUC;
6639 wr32(E1000_CTRL, ctrl);
6640
6641 /* Allow time for pending master requests to run */
6642 igb_disable_pcie_master(hw);
6643
6644 wr32(E1000_WUC, E1000_WUC_PME_EN);
6645 wr32(E1000_WUFC, wufc);
6646 } else {
6647 wr32(E1000_WUC, 0);
6648 wr32(E1000_WUFC, 0);
6649 }
6650
6651 *enable_wake = wufc || adapter->en_mng_pt;
6652 if (!*enable_wake)
6653 igb_power_down_link(adapter);
6654 else
6655 igb_power_up_link(adapter);
6656
6657 /* Release control of h/w to f/w. If f/w is AMT enabled, this
6658 * would have already happened in close and is redundant. */
6659 igb_release_hw_control(adapter);
6660
6661 pci_disable_device(pdev);
6662
6663 return 0;
6664 }
6665
6666 #ifdef CONFIG_PM
6667 #ifdef CONFIG_PM_SLEEP
6668 static int igb_suspend(struct device *dev)
6669 {
6670 int retval;
6671 bool wake;
6672 struct pci_dev *pdev = to_pci_dev(dev);
6673
6674 retval = __igb_shutdown(pdev, &wake, 0);
6675 if (retval)
6676 return retval;
6677
6678 if (wake) {
6679 pci_prepare_to_sleep(pdev);
6680 } else {
6681 pci_wake_from_d3(pdev, false);
6682 pci_set_power_state(pdev, PCI_D3hot);
6683 }
6684
6685 return 0;
6686 }
6687 #endif /* CONFIG_PM_SLEEP */
6688
6689 static int igb_resume(struct device *dev)
6690 {
6691 struct pci_dev *pdev = to_pci_dev(dev);
6692 struct net_device *netdev = pci_get_drvdata(pdev);
6693 struct igb_adapter *adapter = netdev_priv(netdev);
6694 struct e1000_hw *hw = &adapter->hw;
6695 u32 err;
6696
6697 pci_set_power_state(pdev, PCI_D0);
6698 pci_restore_state(pdev);
6699 pci_save_state(pdev);
6700
6701 err = pci_enable_device_mem(pdev);
6702 if (err) {
6703 dev_err(&pdev->dev,
6704 "igb: Cannot enable PCI device from suspend\n");
6705 return err;
6706 }
6707 pci_set_master(pdev);
6708
6709 pci_enable_wake(pdev, PCI_D3hot, 0);
6710 pci_enable_wake(pdev, PCI_D3cold, 0);
6711
6712 if (igb_init_interrupt_scheme(adapter)) {
6713 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
6714 return -ENOMEM;
6715 }
6716
6717 igb_reset(adapter);
6718
6719 /* let the f/w know that the h/w is now under the control of the
6720 * driver. */
6721 igb_get_hw_control(adapter);
6722
6723 wr32(E1000_WUS, ~0);
6724
6725 if (netdev->flags & IFF_UP) {
6726 err = __igb_open(netdev, true);
6727 if (err)
6728 return err;
6729 }
6730
6731 netif_device_attach(netdev);
6732 return 0;
6733 }
6734
6735 #ifdef CONFIG_PM_RUNTIME
6736 static int igb_runtime_idle(struct device *dev)
6737 {
6738 struct pci_dev *pdev = to_pci_dev(dev);
6739 struct net_device *netdev = pci_get_drvdata(pdev);
6740 struct igb_adapter *adapter = netdev_priv(netdev);
6741
6742 if (!igb_has_link(adapter))
6743 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
6744
6745 return -EBUSY;
6746 }
6747
6748 static int igb_runtime_suspend(struct device *dev)
6749 {
6750 struct pci_dev *pdev = to_pci_dev(dev);
6751 int retval;
6752 bool wake;
6753
6754 retval = __igb_shutdown(pdev, &wake, 1);
6755 if (retval)
6756 return retval;
6757
6758 if (wake) {
6759 pci_prepare_to_sleep(pdev);
6760 } else {
6761 pci_wake_from_d3(pdev, false);
6762 pci_set_power_state(pdev, PCI_D3hot);
6763 }
6764
6765 return 0;
6766 }
6767
6768 static int igb_runtime_resume(struct device *dev)
6769 {
6770 return igb_resume(dev);
6771 }
6772 #endif /* CONFIG_PM_RUNTIME */
6773 #endif
6774
6775 static void igb_shutdown(struct pci_dev *pdev)
6776 {
6777 bool wake;
6778
6779 __igb_shutdown(pdev, &wake, 0);
6780
6781 if (system_state == SYSTEM_POWER_OFF) {
6782 pci_wake_from_d3(pdev, wake);
6783 pci_set_power_state(pdev, PCI_D3hot);
6784 }
6785 }
6786
6787 #ifdef CONFIG_NET_POLL_CONTROLLER
6788 /*
6789 * Polling 'interrupt' - used by things like netconsole to send skbs
6790 * without having to re-enable interrupts. It's not called while
6791 * the interrupt routine is executing.
6792 */
6793 static void igb_netpoll(struct net_device *netdev)
6794 {
6795 struct igb_adapter *adapter = netdev_priv(netdev);
6796 struct e1000_hw *hw = &adapter->hw;
6797 struct igb_q_vector *q_vector;
6798 int i;
6799
6800 for (i = 0; i < adapter->num_q_vectors; i++) {
6801 q_vector = adapter->q_vector[i];
6802 if (adapter->msix_entries)
6803 wr32(E1000_EIMC, q_vector->eims_value);
6804 else
6805 igb_irq_disable(adapter);
6806 napi_schedule(&q_vector->napi);
6807 }
6808 }
6809 #endif /* CONFIG_NET_POLL_CONTROLLER */
6810
6811 /**
6812 * igb_io_error_detected - called when PCI error is detected
6813 * @pdev: Pointer to PCI device
6814 * @state: The current pci connection state
6815 *
6816 * This function is called after a PCI bus error affecting
6817 * this device has been detected.
6818 */
6819 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
6820 pci_channel_state_t state)
6821 {
6822 struct net_device *netdev = pci_get_drvdata(pdev);
6823 struct igb_adapter *adapter = netdev_priv(netdev);
6824
6825 netif_device_detach(netdev);
6826
6827 if (state == pci_channel_io_perm_failure)
6828 return PCI_ERS_RESULT_DISCONNECT;
6829
6830 if (netif_running(netdev))
6831 igb_down(adapter);
6832 pci_disable_device(pdev);
6833
6834 /* Request a slot slot reset. */
6835 return PCI_ERS_RESULT_NEED_RESET;
6836 }
6837
6838 /**
6839 * igb_io_slot_reset - called after the pci bus has been reset.
6840 * @pdev: Pointer to PCI device
6841 *
6842 * Restart the card from scratch, as if from a cold-boot. Implementation
6843 * resembles the first-half of the igb_resume routine.
6844 */
6845 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
6846 {
6847 struct net_device *netdev = pci_get_drvdata(pdev);
6848 struct igb_adapter *adapter = netdev_priv(netdev);
6849 struct e1000_hw *hw = &adapter->hw;
6850 pci_ers_result_t result;
6851 int err;
6852
6853 if (pci_enable_device_mem(pdev)) {
6854 dev_err(&pdev->dev,
6855 "Cannot re-enable PCI device after reset.\n");
6856 result = PCI_ERS_RESULT_DISCONNECT;
6857 } else {
6858 pci_set_master(pdev);
6859 pci_restore_state(pdev);
6860 pci_save_state(pdev);
6861
6862 pci_enable_wake(pdev, PCI_D3hot, 0);
6863 pci_enable_wake(pdev, PCI_D3cold, 0);
6864
6865 igb_reset(adapter);
6866 wr32(E1000_WUS, ~0);
6867 result = PCI_ERS_RESULT_RECOVERED;
6868 }
6869
6870 err = pci_cleanup_aer_uncorrect_error_status(pdev);
6871 if (err) {
6872 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
6873 "failed 0x%0x\n", err);
6874 /* non-fatal, continue */
6875 }
6876
6877 return result;
6878 }
6879
6880 /**
6881 * igb_io_resume - called when traffic can start flowing again.
6882 * @pdev: Pointer to PCI device
6883 *
6884 * This callback is called when the error recovery driver tells us that
6885 * its OK to resume normal operation. Implementation resembles the
6886 * second-half of the igb_resume routine.
6887 */
6888 static void igb_io_resume(struct pci_dev *pdev)
6889 {
6890 struct net_device *netdev = pci_get_drvdata(pdev);
6891 struct igb_adapter *adapter = netdev_priv(netdev);
6892
6893 if (netif_running(netdev)) {
6894 if (igb_up(adapter)) {
6895 dev_err(&pdev->dev, "igb_up failed after reset\n");
6896 return;
6897 }
6898 }
6899
6900 netif_device_attach(netdev);
6901
6902 /* let the f/w know that the h/w is now under the control of the
6903 * driver. */
6904 igb_get_hw_control(adapter);
6905 }
6906
6907 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
6908 u8 qsel)
6909 {
6910 u32 rar_low, rar_high;
6911 struct e1000_hw *hw = &adapter->hw;
6912
6913 /* HW expects these in little endian so we reverse the byte order
6914 * from network order (big endian) to little endian
6915 */
6916 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
6917 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
6918 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
6919
6920 /* Indicate to hardware the Address is Valid. */
6921 rar_high |= E1000_RAH_AV;
6922
6923 if (hw->mac.type == e1000_82575)
6924 rar_high |= E1000_RAH_POOL_1 * qsel;
6925 else
6926 rar_high |= E1000_RAH_POOL_1 << qsel;
6927
6928 wr32(E1000_RAL(index), rar_low);
6929 wrfl();
6930 wr32(E1000_RAH(index), rar_high);
6931 wrfl();
6932 }
6933
6934 static int igb_set_vf_mac(struct igb_adapter *adapter,
6935 int vf, unsigned char *mac_addr)
6936 {
6937 struct e1000_hw *hw = &adapter->hw;
6938 /* VF MAC addresses start at end of receive addresses and moves
6939 * torwards the first, as a result a collision should not be possible */
6940 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
6941
6942 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
6943
6944 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
6945
6946 return 0;
6947 }
6948
6949 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
6950 {
6951 struct igb_adapter *adapter = netdev_priv(netdev);
6952 if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
6953 return -EINVAL;
6954 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
6955 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
6956 dev_info(&adapter->pdev->dev, "Reload the VF driver to make this"
6957 " change effective.");
6958 if (test_bit(__IGB_DOWN, &adapter->state)) {
6959 dev_warn(&adapter->pdev->dev, "The VF MAC address has been set,"
6960 " but the PF device is not up.\n");
6961 dev_warn(&adapter->pdev->dev, "Bring the PF device up before"
6962 " attempting to use the VF device.\n");
6963 }
6964 return igb_set_vf_mac(adapter, vf, mac);
6965 }
6966
6967 static int igb_link_mbps(int internal_link_speed)
6968 {
6969 switch (internal_link_speed) {
6970 case SPEED_100:
6971 return 100;
6972 case SPEED_1000:
6973 return 1000;
6974 default:
6975 return 0;
6976 }
6977 }
6978
6979 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
6980 int link_speed)
6981 {
6982 int rf_dec, rf_int;
6983 u32 bcnrc_val;
6984
6985 if (tx_rate != 0) {
6986 /* Calculate the rate factor values to set */
6987 rf_int = link_speed / tx_rate;
6988 rf_dec = (link_speed - (rf_int * tx_rate));
6989 rf_dec = (rf_dec * (1<<E1000_RTTBCNRC_RF_INT_SHIFT)) / tx_rate;
6990
6991 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
6992 bcnrc_val |= ((rf_int<<E1000_RTTBCNRC_RF_INT_SHIFT) &
6993 E1000_RTTBCNRC_RF_INT_MASK);
6994 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
6995 } else {
6996 bcnrc_val = 0;
6997 }
6998
6999 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7000 /*
7001 * Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7002 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7003 */
7004 wr32(E1000_RTTBCNRM, 0x14);
7005 wr32(E1000_RTTBCNRC, bcnrc_val);
7006 }
7007
7008 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7009 {
7010 int actual_link_speed, i;
7011 bool reset_rate = false;
7012
7013 /* VF TX rate limit was not set or not supported */
7014 if ((adapter->vf_rate_link_speed == 0) ||
7015 (adapter->hw.mac.type != e1000_82576))
7016 return;
7017
7018 actual_link_speed = igb_link_mbps(adapter->link_speed);
7019 if (actual_link_speed != adapter->vf_rate_link_speed) {
7020 reset_rate = true;
7021 adapter->vf_rate_link_speed = 0;
7022 dev_info(&adapter->pdev->dev,
7023 "Link speed has been changed. VF Transmit "
7024 "rate is disabled\n");
7025 }
7026
7027 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7028 if (reset_rate)
7029 adapter->vf_data[i].tx_rate = 0;
7030
7031 igb_set_vf_rate_limit(&adapter->hw, i,
7032 adapter->vf_data[i].tx_rate,
7033 actual_link_speed);
7034 }
7035 }
7036
7037 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
7038 {
7039 struct igb_adapter *adapter = netdev_priv(netdev);
7040 struct e1000_hw *hw = &adapter->hw;
7041 int actual_link_speed;
7042
7043 if (hw->mac.type != e1000_82576)
7044 return -EOPNOTSUPP;
7045
7046 actual_link_speed = igb_link_mbps(adapter->link_speed);
7047 if ((vf >= adapter->vfs_allocated_count) ||
7048 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7049 (tx_rate < 0) || (tx_rate > actual_link_speed))
7050 return -EINVAL;
7051
7052 adapter->vf_rate_link_speed = actual_link_speed;
7053 adapter->vf_data[vf].tx_rate = (u16)tx_rate;
7054 igb_set_vf_rate_limit(hw, vf, tx_rate, actual_link_speed);
7055
7056 return 0;
7057 }
7058
7059 static int igb_ndo_get_vf_config(struct net_device *netdev,
7060 int vf, struct ifla_vf_info *ivi)
7061 {
7062 struct igb_adapter *adapter = netdev_priv(netdev);
7063 if (vf >= adapter->vfs_allocated_count)
7064 return -EINVAL;
7065 ivi->vf = vf;
7066 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7067 ivi->tx_rate = adapter->vf_data[vf].tx_rate;
7068 ivi->vlan = adapter->vf_data[vf].pf_vlan;
7069 ivi->qos = adapter->vf_data[vf].pf_qos;
7070 return 0;
7071 }
7072
7073 static void igb_vmm_control(struct igb_adapter *adapter)
7074 {
7075 struct e1000_hw *hw = &adapter->hw;
7076 u32 reg;
7077
7078 switch (hw->mac.type) {
7079 case e1000_82575:
7080 case e1000_i210:
7081 case e1000_i211:
7082 default:
7083 /* replication is not supported for 82575 */
7084 return;
7085 case e1000_82576:
7086 /* notify HW that the MAC is adding vlan tags */
7087 reg = rd32(E1000_DTXCTL);
7088 reg |= E1000_DTXCTL_VLAN_ADDED;
7089 wr32(E1000_DTXCTL, reg);
7090 case e1000_82580:
7091 /* enable replication vlan tag stripping */
7092 reg = rd32(E1000_RPLOLR);
7093 reg |= E1000_RPLOLR_STRVLAN;
7094 wr32(E1000_RPLOLR, reg);
7095 case e1000_i350:
7096 /* none of the above registers are supported by i350 */
7097 break;
7098 }
7099
7100 if (adapter->vfs_allocated_count) {
7101 igb_vmdq_set_loopback_pf(hw, true);
7102 igb_vmdq_set_replication_pf(hw, true);
7103 igb_vmdq_set_anti_spoofing_pf(hw, true,
7104 adapter->vfs_allocated_count);
7105 } else {
7106 igb_vmdq_set_loopback_pf(hw, false);
7107 igb_vmdq_set_replication_pf(hw, false);
7108 }
7109 }
7110
7111 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7112 {
7113 struct e1000_hw *hw = &adapter->hw;
7114 u32 dmac_thr;
7115 u16 hwm;
7116
7117 if (hw->mac.type > e1000_82580) {
7118 if (adapter->flags & IGB_FLAG_DMAC) {
7119 u32 reg;
7120
7121 /* force threshold to 0. */
7122 wr32(E1000_DMCTXTH, 0);
7123
7124 /*
7125 * DMA Coalescing high water mark needs to be greater
7126 * than the Rx threshold. Set hwm to PBA - max frame
7127 * size in 16B units, capping it at PBA - 6KB.
7128 */
7129 hwm = 64 * pba - adapter->max_frame_size / 16;
7130 if (hwm < 64 * (pba - 6))
7131 hwm = 64 * (pba - 6);
7132 reg = rd32(E1000_FCRTC);
7133 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
7134 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
7135 & E1000_FCRTC_RTH_COAL_MASK);
7136 wr32(E1000_FCRTC, reg);
7137
7138 /*
7139 * Set the DMA Coalescing Rx threshold to PBA - 2 * max
7140 * frame size, capping it at PBA - 10KB.
7141 */
7142 dmac_thr = pba - adapter->max_frame_size / 512;
7143 if (dmac_thr < pba - 10)
7144 dmac_thr = pba - 10;
7145 reg = rd32(E1000_DMACR);
7146 reg &= ~E1000_DMACR_DMACTHR_MASK;
7147 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
7148 & E1000_DMACR_DMACTHR_MASK);
7149
7150 /* transition to L0x or L1 if available..*/
7151 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
7152
7153 /* watchdog timer= +-1000 usec in 32usec intervals */
7154 reg |= (1000 >> 5);
7155
7156 /* Disable BMC-to-OS Watchdog Enable */
7157 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
7158 wr32(E1000_DMACR, reg);
7159
7160 /*
7161 * no lower threshold to disable
7162 * coalescing(smart fifb)-UTRESH=0
7163 */
7164 wr32(E1000_DMCRTRH, 0);
7165
7166 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
7167
7168 wr32(E1000_DMCTLX, reg);
7169
7170 /*
7171 * free space in tx packet buffer to wake from
7172 * DMA coal
7173 */
7174 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
7175 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
7176
7177 /*
7178 * make low power state decision controlled
7179 * by DMA coal
7180 */
7181 reg = rd32(E1000_PCIEMISC);
7182 reg &= ~E1000_PCIEMISC_LX_DECISION;
7183 wr32(E1000_PCIEMISC, reg);
7184 } /* endif adapter->dmac is not disabled */
7185 } else if (hw->mac.type == e1000_82580) {
7186 u32 reg = rd32(E1000_PCIEMISC);
7187 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
7188 wr32(E1000_DMACR, 0);
7189 }
7190 }
7191
7192 /* igb_main.c */
Linux kernel - Deliverables
This page took 0.255218 seconds and 6 git commands to generate.