?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
igb: Map entire page and sync half instead of mapping and unmapping half pages
[linux-2.6/cjktty.git] / drivers / net / ethernet / intel / igb / igb_main.c
blob18ad18fb42679d124d5137c1cde38134ce68cbeb
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 4
63 #define MIN 0
64 #define BUILD 1
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 bool igb_vfs_are_assigned(struct igb_adapter *adapter);
176 #endif
177
178 #ifdef CONFIG_PM
179 #ifdef CONFIG_PM_SLEEP
180 static int igb_suspend(struct device *);
181 #endif
182 static int igb_resume(struct device *);
183 #ifdef CONFIG_PM_RUNTIME
184 static int igb_runtime_suspend(struct device *dev);
185 static int igb_runtime_resume(struct device *dev);
186 static int igb_runtime_idle(struct device *dev);
187 #endif
188 static const struct dev_pm_ops igb_pm_ops = {
189 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
190 SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
191 igb_runtime_idle)
192 };
193 #endif
194 static void igb_shutdown(struct pci_dev *);
195 #ifdef CONFIG_IGB_DCA
196 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
197 static struct notifier_block dca_notifier = {
198 .notifier_call = igb_notify_dca,
199 .next = NULL,
200 .priority = 0
201 };
202 #endif
203 #ifdef CONFIG_NET_POLL_CONTROLLER
204 /* for netdump / net console */
205 static void igb_netpoll(struct net_device *);
206 #endif
207 #ifdef CONFIG_PCI_IOV
208 static unsigned int max_vfs = 0;
209 module_param(max_vfs, uint, 0);
210 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
211 "per physical function");
212 #endif /* CONFIG_PCI_IOV */
213
214 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
215 pci_channel_state_t);
216 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
217 static void igb_io_resume(struct pci_dev *);
218
219 static const struct pci_error_handlers igb_err_handler = {
220 .error_detected = igb_io_error_detected,
221 .slot_reset = igb_io_slot_reset,
222 .resume = igb_io_resume,
223 };
224
225 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
226
227 static struct pci_driver igb_driver = {
228 .name = igb_driver_name,
229 .id_table = igb_pci_tbl,
230 .probe = igb_probe,
231 .remove = __devexit_p(igb_remove),
232 #ifdef CONFIG_PM
233 .driver.pm = &igb_pm_ops,
234 #endif
235 .shutdown = igb_shutdown,
236 .err_handler = &igb_err_handler
237 };
238
239 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
240 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
241 MODULE_LICENSE("GPL");
242 MODULE_VERSION(DRV_VERSION);
243
244 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
245 static int debug = -1;
246 module_param(debug, int, 0);
247 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
248
249 struct igb_reg_info {
250 u32 ofs;
251 char *name;
252 };
253
254 static const struct igb_reg_info igb_reg_info_tbl[] = {
255
256 /* General Registers */
257 {E1000_CTRL, "CTRL"},
258 {E1000_STATUS, "STATUS"},
259 {E1000_CTRL_EXT, "CTRL_EXT"},
260
261 /* Interrupt Registers */
262 {E1000_ICR, "ICR"},
263
264 /* RX Registers */
265 {E1000_RCTL, "RCTL"},
266 {E1000_RDLEN(0), "RDLEN"},
267 {E1000_RDH(0), "RDH"},
268 {E1000_RDT(0), "RDT"},
269 {E1000_RXDCTL(0), "RXDCTL"},
270 {E1000_RDBAL(0), "RDBAL"},
271 {E1000_RDBAH(0), "RDBAH"},
272
273 /* TX Registers */
274 {E1000_TCTL, "TCTL"},
275 {E1000_TDBAL(0), "TDBAL"},
276 {E1000_TDBAH(0), "TDBAH"},
277 {E1000_TDLEN(0), "TDLEN"},
278 {E1000_TDH(0), "TDH"},
279 {E1000_TDT(0), "TDT"},
280 {E1000_TXDCTL(0), "TXDCTL"},
281 {E1000_TDFH, "TDFH"},
282 {E1000_TDFT, "TDFT"},
283 {E1000_TDFHS, "TDFHS"},
284 {E1000_TDFPC, "TDFPC"},
285
286 /* List Terminator */
287 {}
288 };
289
290 /*
291 * igb_regdump - register printout routine
292 */
293 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
294 {
295 int n = 0;
296 char rname[16];
297 u32 regs[8];
298
299 switch (reginfo->ofs) {
300 case E1000_RDLEN(0):
301 for (n = 0; n < 4; n++)
302 regs[n] = rd32(E1000_RDLEN(n));
303 break;
304 case E1000_RDH(0):
305 for (n = 0; n < 4; n++)
306 regs[n] = rd32(E1000_RDH(n));
307 break;
308 case E1000_RDT(0):
309 for (n = 0; n < 4; n++)
310 regs[n] = rd32(E1000_RDT(n));
311 break;
312 case E1000_RXDCTL(0):
313 for (n = 0; n < 4; n++)
314 regs[n] = rd32(E1000_RXDCTL(n));
315 break;
316 case E1000_RDBAL(0):
317 for (n = 0; n < 4; n++)
318 regs[n] = rd32(E1000_RDBAL(n));
319 break;
320 case E1000_RDBAH(0):
321 for (n = 0; n < 4; n++)
322 regs[n] = rd32(E1000_RDBAH(n));
323 break;
324 case E1000_TDBAL(0):
325 for (n = 0; n < 4; n++)
326 regs[n] = rd32(E1000_RDBAL(n));
327 break;
328 case E1000_TDBAH(0):
329 for (n = 0; n < 4; n++)
330 regs[n] = rd32(E1000_TDBAH(n));
331 break;
332 case E1000_TDLEN(0):
333 for (n = 0; n < 4; n++)
334 regs[n] = rd32(E1000_TDLEN(n));
335 break;
336 case E1000_TDH(0):
337 for (n = 0; n < 4; n++)
338 regs[n] = rd32(E1000_TDH(n));
339 break;
340 case E1000_TDT(0):
341 for (n = 0; n < 4; n++)
342 regs[n] = rd32(E1000_TDT(n));
343 break;
344 case E1000_TXDCTL(0):
345 for (n = 0; n < 4; n++)
346 regs[n] = rd32(E1000_TXDCTL(n));
347 break;
348 default:
349 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
350 return;
351 }
352
353 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
354 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
355 regs[2], regs[3]);
356 }
357
358 /*
359 * igb_dump - Print registers, tx-rings and rx-rings
360 */
361 static void igb_dump(struct igb_adapter *adapter)
362 {
363 struct net_device *netdev = adapter->netdev;
364 struct e1000_hw *hw = &adapter->hw;
365 struct igb_reg_info *reginfo;
366 struct igb_ring *tx_ring;
367 union e1000_adv_tx_desc *tx_desc;
368 struct my_u0 { u64 a; u64 b; } *u0;
369 struct igb_ring *rx_ring;
370 union e1000_adv_rx_desc *rx_desc;
371 u32 staterr;
372 u16 i, n;
373
374 if (!netif_msg_hw(adapter))
375 return;
376
377 /* Print netdevice Info */
378 if (netdev) {
379 dev_info(&adapter->pdev->dev, "Net device Info\n");
380 pr_info("Device Name state trans_start "
381 "last_rx\n");
382 pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
383 netdev->state, netdev->trans_start, netdev->last_rx);
384 }
385
386 /* Print Registers */
387 dev_info(&adapter->pdev->dev, "Register Dump\n");
388 pr_info(" Register Name Value\n");
389 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
390 reginfo->name; reginfo++) {
391 igb_regdump(hw, reginfo);
392 }
393
394 /* Print TX Ring Summary */
395 if (!netdev || !netif_running(netdev))
396 goto exit;
397
398 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
399 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
400 for (n = 0; n < adapter->num_tx_queues; n++) {
401 struct igb_tx_buffer *buffer_info;
402 tx_ring = adapter->tx_ring[n];
403 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
404 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
405 n, tx_ring->next_to_use, tx_ring->next_to_clean,
406 (u64)dma_unmap_addr(buffer_info, dma),
407 dma_unmap_len(buffer_info, len),
408 buffer_info->next_to_watch,
409 (u64)buffer_info->time_stamp);
410 }
411
412 /* Print TX Rings */
413 if (!netif_msg_tx_done(adapter))
414 goto rx_ring_summary;
415
416 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
417
418 /* Transmit Descriptor Formats
419 *
420 * Advanced Transmit Descriptor
421 * +--------------------------------------------------------------+
422 * 0 | Buffer Address [63:0] |
423 * +--------------------------------------------------------------+
424 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN |
425 * +--------------------------------------------------------------+
426 * 63 46 45 40 39 38 36 35 32 31 24 15 0
427 */
428
429 for (n = 0; n < adapter->num_tx_queues; n++) {
430 tx_ring = adapter->tx_ring[n];
431 pr_info("------------------------------------\n");
432 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
433 pr_info("------------------------------------\n");
434 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] "
435 "[bi->dma ] leng ntw timestamp "
436 "bi->skb\n");
437
438 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
439 const char *next_desc;
440 struct igb_tx_buffer *buffer_info;
441 tx_desc = IGB_TX_DESC(tx_ring, i);
442 buffer_info = &tx_ring->tx_buffer_info[i];
443 u0 = (struct my_u0 *)tx_desc;
444 if (i == tx_ring->next_to_use &&
445 i == tx_ring->next_to_clean)
446 next_desc = " NTC/U";
447 else if (i == tx_ring->next_to_use)
448 next_desc = " NTU";
449 else if (i == tx_ring->next_to_clean)
450 next_desc = " NTC";
451 else
452 next_desc = "";
453
454 pr_info("T [0x%03X] %016llX %016llX %016llX"
455 " %04X %p %016llX %p%s\n", i,
456 le64_to_cpu(u0->a),
457 le64_to_cpu(u0->b),
458 (u64)dma_unmap_addr(buffer_info, dma),
459 dma_unmap_len(buffer_info, len),
460 buffer_info->next_to_watch,
461 (u64)buffer_info->time_stamp,
462 buffer_info->skb, next_desc);
463
464 if (netif_msg_pktdata(adapter) && buffer_info->skb)
465 print_hex_dump(KERN_INFO, "",
466 DUMP_PREFIX_ADDRESS,
467 16, 1, buffer_info->skb->data,
468 dma_unmap_len(buffer_info, len),
469 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 ---------------- %s\n",
538 "RWB", i,
539 le64_to_cpu(u0->a),
540 le64_to_cpu(u0->b),
541 next_desc);
542 } else {
543 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n",
544 "R ", i,
545 le64_to_cpu(u0->a),
546 le64_to_cpu(u0->b),
547 (u64)buffer_info->dma,
548 next_desc);
549
550 if (netif_msg_pktdata(adapter) &&
551 buffer_info->dma && buffer_info->page) {
552 print_hex_dump(KERN_INFO, "",
553 DUMP_PREFIX_ADDRESS,
554 16, 1,
555 page_address(buffer_info->page) +
556 buffer_info->page_offset,
557 PAGE_SIZE/2, true);
558 }
559 }
560 }
561 }
562
563 exit:
564 return;
565 }
566
567 /**
568 * igb_get_hw_dev - return device
569 * used by hardware layer to print debugging information
570 **/
571 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
572 {
573 struct igb_adapter *adapter = hw->back;
574 return adapter->netdev;
575 }
576
577 /**
578 * igb_init_module - Driver Registration Routine
579 *
580 * igb_init_module is the first routine called when the driver is
581 * loaded. All it does is register with the PCI subsystem.
582 **/
583 static int __init igb_init_module(void)
584 {
585 int ret;
586 pr_info("%s - version %s\n",
587 igb_driver_string, igb_driver_version);
588
589 pr_info("%s\n", igb_copyright);
590
591 #ifdef CONFIG_IGB_DCA
592 dca_register_notify(&dca_notifier);
593 #endif
594 ret = pci_register_driver(&igb_driver);
595 return ret;
596 }
597
598 module_init(igb_init_module);
599
600 /**
601 * igb_exit_module - Driver Exit Cleanup Routine
602 *
603 * igb_exit_module is called just before the driver is removed
604 * from memory.
605 **/
606 static void __exit igb_exit_module(void)
607 {
608 #ifdef CONFIG_IGB_DCA
609 dca_unregister_notify(&dca_notifier);
610 #endif
611 pci_unregister_driver(&igb_driver);
612 }
613
614 module_exit(igb_exit_module);
615
616 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
617 /**
618 * igb_cache_ring_register - Descriptor ring to register mapping
619 * @adapter: board private structure to initialize
620 *
621 * Once we know the feature-set enabled for the device, we'll cache
622 * the register offset the descriptor ring is assigned to.
623 **/
624 static void igb_cache_ring_register(struct igb_adapter *adapter)
625 {
626 int i = 0, j = 0;
627 u32 rbase_offset = adapter->vfs_allocated_count;
628
629 switch (adapter->hw.mac.type) {
630 case e1000_82576:
631 /* The queues are allocated for virtualization such that VF 0
632 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
633 * In order to avoid collision we start at the first free queue
634 * and continue consuming queues in the same sequence
635 */
636 if (adapter->vfs_allocated_count) {
637 for (; i < adapter->rss_queues; i++)
638 adapter->rx_ring[i]->reg_idx = rbase_offset +
639 Q_IDX_82576(i);
640 }
641 case e1000_82575:
642 case e1000_82580:
643 case e1000_i350:
644 case e1000_i210:
645 case e1000_i211:
646 default:
647 for (; i < adapter->num_rx_queues; i++)
648 adapter->rx_ring[i]->reg_idx = rbase_offset + i;
649 for (; j < adapter->num_tx_queues; j++)
650 adapter->tx_ring[j]->reg_idx = rbase_offset + j;
651 break;
652 }
653 }
654
655 static void igb_free_queues(struct igb_adapter *adapter)
656 {
657 int i;
658
659 for (i = 0; i < adapter->num_tx_queues; i++) {
660 kfree(adapter->tx_ring[i]);
661 adapter->tx_ring[i] = NULL;
662 }
663 for (i = 0; i < adapter->num_rx_queues; i++) {
664 kfree(adapter->rx_ring[i]);
665 adapter->rx_ring[i] = NULL;
666 }
667 adapter->num_rx_queues = 0;
668 adapter->num_tx_queues = 0;
669 }
670
671 /**
672 * igb_alloc_queues - Allocate memory for all rings
673 * @adapter: board private structure to initialize
674 *
675 * We allocate one ring per queue at run-time since we don't know the
676 * number of queues at compile-time.
677 **/
678 static int igb_alloc_queues(struct igb_adapter *adapter)
679 {
680 struct igb_ring *ring;
681 int i;
682
683 for (i = 0; i < adapter->num_tx_queues; i++) {
684 ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL);
685 if (!ring)
686 goto err;
687 ring->count = adapter->tx_ring_count;
688 ring->queue_index = i;
689 ring->dev = &adapter->pdev->dev;
690 ring->netdev = adapter->netdev;
691 /* For 82575, context index must be unique per ring. */
692 if (adapter->hw.mac.type == e1000_82575)
693 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
694 adapter->tx_ring[i] = ring;
695 }
696
697 for (i = 0; i < adapter->num_rx_queues; i++) {
698 ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL);
699 if (!ring)
700 goto err;
701 ring->count = adapter->rx_ring_count;
702 ring->queue_index = i;
703 ring->dev = &adapter->pdev->dev;
704 ring->netdev = adapter->netdev;
705 /* set flag indicating ring supports SCTP checksum offload */
706 if (adapter->hw.mac.type >= e1000_82576)
707 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
708
709 /*
710 * On i350, i210, and i211, loopback VLAN packets
711 * have the tag byte-swapped.
712 * */
713 if (adapter->hw.mac.type >= e1000_i350)
714 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
715
716 adapter->rx_ring[i] = ring;
717 }
718
719 igb_cache_ring_register(adapter);
720
721 return 0;
722
723 err:
724 igb_free_queues(adapter);
725
726 return -ENOMEM;
727 }
728
729 /**
730 * igb_write_ivar - configure ivar for given MSI-X vector
731 * @hw: pointer to the HW structure
732 * @msix_vector: vector number we are allocating to a given ring
733 * @index: row index of IVAR register to write within IVAR table
734 * @offset: column offset of in IVAR, should be multiple of 8
735 *
736 * This function is intended to handle the writing of the IVAR register
737 * for adapters 82576 and newer. The IVAR table consists of 2 columns,
738 * each containing an cause allocation for an Rx and Tx ring, and a
739 * variable number of rows depending on the number of queues supported.
740 **/
741 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
742 int index, int offset)
743 {
744 u32 ivar = array_rd32(E1000_IVAR0, index);
745
746 /* clear any bits that are currently set */
747 ivar &= ~((u32)0xFF << offset);
748
749 /* write vector and valid bit */
750 ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
751
752 array_wr32(E1000_IVAR0, index, ivar);
753 }
754
755 #define IGB_N0_QUEUE -1
756 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
757 {
758 struct igb_adapter *adapter = q_vector->adapter;
759 struct e1000_hw *hw = &adapter->hw;
760 int rx_queue = IGB_N0_QUEUE;
761 int tx_queue = IGB_N0_QUEUE;
762 u32 msixbm = 0;
763
764 if (q_vector->rx.ring)
765 rx_queue = q_vector->rx.ring->reg_idx;
766 if (q_vector->tx.ring)
767 tx_queue = q_vector->tx.ring->reg_idx;
768
769 switch (hw->mac.type) {
770 case e1000_82575:
771 /* The 82575 assigns vectors using a bitmask, which matches the
772 bitmask for the EICR/EIMS/EIMC registers. To assign one
773 or more queues to a vector, we write the appropriate bits
774 into the MSIXBM register for that vector. */
775 if (rx_queue > IGB_N0_QUEUE)
776 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
777 if (tx_queue > IGB_N0_QUEUE)
778 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
779 if (!adapter->msix_entries && msix_vector == 0)
780 msixbm |= E1000_EIMS_OTHER;
781 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
782 q_vector->eims_value = msixbm;
783 break;
784 case e1000_82576:
785 /*
786 * 82576 uses a table that essentially consists of 2 columns
787 * with 8 rows. The ordering is column-major so we use the
788 * lower 3 bits as the row index, and the 4th bit as the
789 * column offset.
790 */
791 if (rx_queue > IGB_N0_QUEUE)
792 igb_write_ivar(hw, msix_vector,
793 rx_queue & 0x7,
794 (rx_queue & 0x8) << 1);
795 if (tx_queue > IGB_N0_QUEUE)
796 igb_write_ivar(hw, msix_vector,
797 tx_queue & 0x7,
798 ((tx_queue & 0x8) << 1) + 8);
799 q_vector->eims_value = 1 << msix_vector;
800 break;
801 case e1000_82580:
802 case e1000_i350:
803 case e1000_i210:
804 case e1000_i211:
805 /*
806 * On 82580 and newer adapters the scheme is similar to 82576
807 * however instead of ordering column-major we have things
808 * ordered row-major. So we traverse the table by using
809 * bit 0 as the column offset, and the remaining bits as the
810 * row index.
811 */
812 if (rx_queue > IGB_N0_QUEUE)
813 igb_write_ivar(hw, msix_vector,
814 rx_queue >> 1,
815 (rx_queue & 0x1) << 4);
816 if (tx_queue > IGB_N0_QUEUE)
817 igb_write_ivar(hw, msix_vector,
818 tx_queue >> 1,
819 ((tx_queue & 0x1) << 4) + 8);
820 q_vector->eims_value = 1 << msix_vector;
821 break;
822 default:
823 BUG();
824 break;
825 }
826
827 /* add q_vector eims value to global eims_enable_mask */
828 adapter->eims_enable_mask |= q_vector->eims_value;
829
830 /* configure q_vector to set itr on first interrupt */
831 q_vector->set_itr = 1;
832 }
833
834 /**
835 * igb_configure_msix - Configure MSI-X hardware
836 *
837 * igb_configure_msix sets up the hardware to properly
838 * generate MSI-X interrupts.
839 **/
840 static void igb_configure_msix(struct igb_adapter *adapter)
841 {
842 u32 tmp;
843 int i, vector = 0;
844 struct e1000_hw *hw = &adapter->hw;
845
846 adapter->eims_enable_mask = 0;
847
848 /* set vector for other causes, i.e. link changes */
849 switch (hw->mac.type) {
850 case e1000_82575:
851 tmp = rd32(E1000_CTRL_EXT);
852 /* enable MSI-X PBA support*/
853 tmp |= E1000_CTRL_EXT_PBA_CLR;
854
855 /* Auto-Mask interrupts upon ICR read. */
856 tmp |= E1000_CTRL_EXT_EIAME;
857 tmp |= E1000_CTRL_EXT_IRCA;
858
859 wr32(E1000_CTRL_EXT, tmp);
860
861 /* enable msix_other interrupt */
862 array_wr32(E1000_MSIXBM(0), vector++,
863 E1000_EIMS_OTHER);
864 adapter->eims_other = E1000_EIMS_OTHER;
865
866 break;
867
868 case e1000_82576:
869 case e1000_82580:
870 case e1000_i350:
871 case e1000_i210:
872 case e1000_i211:
873 /* Turn on MSI-X capability first, or our settings
874 * won't stick. And it will take days to debug. */
875 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
876 E1000_GPIE_PBA | E1000_GPIE_EIAME |
877 E1000_GPIE_NSICR);
878
879 /* enable msix_other interrupt */
880 adapter->eims_other = 1 << vector;
881 tmp = (vector++ | E1000_IVAR_VALID) << 8;
882
883 wr32(E1000_IVAR_MISC, tmp);
884 break;
885 default:
886 /* do nothing, since nothing else supports MSI-X */
887 break;
888 } /* switch (hw->mac.type) */
889
890 adapter->eims_enable_mask |= adapter->eims_other;
891
892 for (i = 0; i < adapter->num_q_vectors; i++)
893 igb_assign_vector(adapter->q_vector[i], vector++);
894
895 wrfl();
896 }
897
898 /**
899 * igb_request_msix - Initialize MSI-X interrupts
900 *
901 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
902 * kernel.
903 **/
904 static int igb_request_msix(struct igb_adapter *adapter)
905 {
906 struct net_device *netdev = adapter->netdev;
907 struct e1000_hw *hw = &adapter->hw;
908 int i, err = 0, vector = 0;
909
910 err = request_irq(adapter->msix_entries[vector].vector,
911 igb_msix_other, 0, netdev->name, adapter);
912 if (err)
913 goto out;
914 vector++;
915
916 for (i = 0; i < adapter->num_q_vectors; i++) {
917 struct igb_q_vector *q_vector = adapter->q_vector[i];
918
919 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
920
921 if (q_vector->rx.ring && q_vector->tx.ring)
922 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
923 q_vector->rx.ring->queue_index);
924 else if (q_vector->tx.ring)
925 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
926 q_vector->tx.ring->queue_index);
927 else if (q_vector->rx.ring)
928 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
929 q_vector->rx.ring->queue_index);
930 else
931 sprintf(q_vector->name, "%s-unused", netdev->name);
932
933 err = request_irq(adapter->msix_entries[vector].vector,
934 igb_msix_ring, 0, q_vector->name,
935 q_vector);
936 if (err)
937 goto out;
938 vector++;
939 }
940
941 igb_configure_msix(adapter);
942 return 0;
943 out:
944 return err;
945 }
946
947 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
948 {
949 if (adapter->msix_entries) {
950 pci_disable_msix(adapter->pdev);
951 kfree(adapter->msix_entries);
952 adapter->msix_entries = NULL;
953 } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
954 pci_disable_msi(adapter->pdev);
955 }
956 }
957
958 /**
959 * igb_free_q_vectors - Free memory allocated for interrupt vectors
960 * @adapter: board private structure to initialize
961 *
962 * This function frees the memory allocated to the q_vectors. In addition if
963 * NAPI is enabled it will delete any references to the NAPI struct prior
964 * to freeing the q_vector.
965 **/
966 static void igb_free_q_vectors(struct igb_adapter *adapter)
967 {
968 int v_idx;
969
970 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
971 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
972 adapter->q_vector[v_idx] = NULL;
973 if (!q_vector)
974 continue;
975 netif_napi_del(&q_vector->napi);
976 kfree(q_vector);
977 }
978 adapter->num_q_vectors = 0;
979 }
980
981 /**
982 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
983 *
984 * This function resets the device so that it has 0 rx queues, tx queues, and
985 * MSI-X interrupts allocated.
986 */
987 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
988 {
989 igb_free_queues(adapter);
990 igb_free_q_vectors(adapter);
991 igb_reset_interrupt_capability(adapter);
992 }
993
994 /**
995 * igb_set_interrupt_capability - set MSI or MSI-X if supported
996 *
997 * Attempt to configure interrupts using the best available
998 * capabilities of the hardware and kernel.
999 **/
1000 static int igb_set_interrupt_capability(struct igb_adapter *adapter)
1001 {
1002 int err;
1003 int numvecs, i;
1004
1005 /* Number of supported queues. */
1006 adapter->num_rx_queues = adapter->rss_queues;
1007 if (adapter->vfs_allocated_count)
1008 adapter->num_tx_queues = 1;
1009 else
1010 adapter->num_tx_queues = adapter->rss_queues;
1011
1012 /* start with one vector for every rx queue */
1013 numvecs = adapter->num_rx_queues;
1014
1015 /* if tx handler is separate add 1 for every tx queue */
1016 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1017 numvecs += adapter->num_tx_queues;
1018
1019 /* store the number of vectors reserved for queues */
1020 adapter->num_q_vectors = numvecs;
1021
1022 /* add 1 vector for link status interrupts */
1023 numvecs++;
1024 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
1025 GFP_KERNEL);
1026
1027 if (!adapter->msix_entries)
1028 goto msi_only;
1029
1030 for (i = 0; i < numvecs; i++)
1031 adapter->msix_entries[i].entry = i;
1032
1033 err = pci_enable_msix(adapter->pdev,
1034 adapter->msix_entries,
1035 numvecs);
1036 if (err == 0)
1037 goto out;
1038
1039 igb_reset_interrupt_capability(adapter);
1040
1041 /* If we can't do MSI-X, try MSI */
1042 msi_only:
1043 #ifdef CONFIG_PCI_IOV
1044 /* disable SR-IOV for non MSI-X configurations */
1045 if (adapter->vf_data) {
1046 struct e1000_hw *hw = &adapter->hw;
1047 /* disable iov and allow time for transactions to clear */
1048 pci_disable_sriov(adapter->pdev);
1049 msleep(500);
1050
1051 kfree(adapter->vf_data);
1052 adapter->vf_data = NULL;
1053 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1054 wrfl();
1055 msleep(100);
1056 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1057 }
1058 #endif
1059 adapter->vfs_allocated_count = 0;
1060 adapter->rss_queues = 1;
1061 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1062 adapter->num_rx_queues = 1;
1063 adapter->num_tx_queues = 1;
1064 adapter->num_q_vectors = 1;
1065 if (!pci_enable_msi(adapter->pdev))
1066 adapter->flags |= IGB_FLAG_HAS_MSI;
1067 out:
1068 /* Notify the stack of the (possibly) reduced queue counts. */
1069 rtnl_lock();
1070 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
1071 err = netif_set_real_num_rx_queues(adapter->netdev,
1072 adapter->num_rx_queues);
1073 rtnl_unlock();
1074 return err;
1075 }
1076
1077 /**
1078 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1079 * @adapter: board private structure to initialize
1080 *
1081 * We allocate one q_vector per queue interrupt. If allocation fails we
1082 * return -ENOMEM.
1083 **/
1084 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1085 {
1086 struct igb_q_vector *q_vector;
1087 struct e1000_hw *hw = &adapter->hw;
1088 int v_idx;
1089
1090 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
1091 q_vector = kzalloc(sizeof(struct igb_q_vector),
1092 GFP_KERNEL);
1093 if (!q_vector)
1094 goto err_out;
1095 q_vector->adapter = adapter;
1096 q_vector->itr_register = hw->hw_addr + E1000_EITR(0);
1097 q_vector->itr_val = IGB_START_ITR;
1098 netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll, 64);
1099 adapter->q_vector[v_idx] = q_vector;
1100 }
1101
1102 return 0;
1103
1104 err_out:
1105 igb_free_q_vectors(adapter);
1106 return -ENOMEM;
1107 }
1108
1109 static void igb_map_rx_ring_to_vector(struct igb_adapter *adapter,
1110 int ring_idx, int v_idx)
1111 {
1112 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1113
1114 q_vector->rx.ring = adapter->rx_ring[ring_idx];
1115 q_vector->rx.ring->q_vector = q_vector;
1116 q_vector->rx.count++;
1117 q_vector->itr_val = adapter->rx_itr_setting;
1118 if (q_vector->itr_val && q_vector->itr_val <= 3)
1119 q_vector->itr_val = IGB_START_ITR;
1120 }
1121
1122 static void igb_map_tx_ring_to_vector(struct igb_adapter *adapter,
1123 int ring_idx, int v_idx)
1124 {
1125 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1126
1127 q_vector->tx.ring = adapter->tx_ring[ring_idx];
1128 q_vector->tx.ring->q_vector = q_vector;
1129 q_vector->tx.count++;
1130 q_vector->itr_val = adapter->tx_itr_setting;
1131 q_vector->tx.work_limit = adapter->tx_work_limit;
1132 if (q_vector->itr_val && q_vector->itr_val <= 3)
1133 q_vector->itr_val = IGB_START_ITR;
1134 }
1135
1136 /**
1137 * igb_map_ring_to_vector - maps allocated queues to vectors
1138 *
1139 * This function maps the recently allocated queues to vectors.
1140 **/
1141 static int igb_map_ring_to_vector(struct igb_adapter *adapter)
1142 {
1143 int i;
1144 int v_idx = 0;
1145
1146 if ((adapter->num_q_vectors < adapter->num_rx_queues) ||
1147 (adapter->num_q_vectors < adapter->num_tx_queues))
1148 return -ENOMEM;
1149
1150 if (adapter->num_q_vectors >=
1151 (adapter->num_rx_queues + adapter->num_tx_queues)) {
1152 for (i = 0; i < adapter->num_rx_queues; i++)
1153 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
1154 for (i = 0; i < adapter->num_tx_queues; i++)
1155 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
1156 } else {
1157 for (i = 0; i < adapter->num_rx_queues; i++) {
1158 if (i < adapter->num_tx_queues)
1159 igb_map_tx_ring_to_vector(adapter, i, v_idx);
1160 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
1161 }
1162 for (; i < adapter->num_tx_queues; i++)
1163 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
1164 }
1165 return 0;
1166 }
1167
1168 /**
1169 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1170 *
1171 * This function initializes the interrupts and allocates all of the queues.
1172 **/
1173 static int igb_init_interrupt_scheme(struct igb_adapter *adapter)
1174 {
1175 struct pci_dev *pdev = adapter->pdev;
1176 int err;
1177
1178 err = igb_set_interrupt_capability(adapter);
1179 if (err)
1180 return err;
1181
1182 err = igb_alloc_q_vectors(adapter);
1183 if (err) {
1184 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1185 goto err_alloc_q_vectors;
1186 }
1187
1188 err = igb_alloc_queues(adapter);
1189 if (err) {
1190 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1191 goto err_alloc_queues;
1192 }
1193
1194 err = igb_map_ring_to_vector(adapter);
1195 if (err) {
1196 dev_err(&pdev->dev, "Invalid q_vector to ring mapping\n");
1197 goto err_map_queues;
1198 }
1199
1200
1201 return 0;
1202 err_map_queues:
1203 igb_free_queues(adapter);
1204 err_alloc_queues:
1205 igb_free_q_vectors(adapter);
1206 err_alloc_q_vectors:
1207 igb_reset_interrupt_capability(adapter);
1208 return err;
1209 }
1210
1211 /**
1212 * igb_request_irq - initialize interrupts
1213 *
1214 * Attempts to configure interrupts using the best available
1215 * capabilities of the hardware and kernel.
1216 **/
1217 static int igb_request_irq(struct igb_adapter *adapter)
1218 {
1219 struct net_device *netdev = adapter->netdev;
1220 struct pci_dev *pdev = adapter->pdev;
1221 int err = 0;
1222
1223 if (adapter->msix_entries) {
1224 err = igb_request_msix(adapter);
1225 if (!err)
1226 goto request_done;
1227 /* fall back to MSI */
1228 igb_clear_interrupt_scheme(adapter);
1229 if (!pci_enable_msi(pdev))
1230 adapter->flags |= IGB_FLAG_HAS_MSI;
1231 igb_free_all_tx_resources(adapter);
1232 igb_free_all_rx_resources(adapter);
1233 adapter->num_tx_queues = 1;
1234 adapter->num_rx_queues = 1;
1235 adapter->num_q_vectors = 1;
1236 err = igb_alloc_q_vectors(adapter);
1237 if (err) {
1238 dev_err(&pdev->dev,
1239 "Unable to allocate memory for vectors\n");
1240 goto request_done;
1241 }
1242 err = igb_alloc_queues(adapter);
1243 if (err) {
1244 dev_err(&pdev->dev,
1245 "Unable to allocate memory for queues\n");
1246 igb_free_q_vectors(adapter);
1247 goto request_done;
1248 }
1249 igb_setup_all_tx_resources(adapter);
1250 igb_setup_all_rx_resources(adapter);
1251 }
1252
1253 igb_assign_vector(adapter->q_vector[0], 0);
1254
1255 if (adapter->flags & IGB_FLAG_HAS_MSI) {
1256 err = request_irq(pdev->irq, igb_intr_msi, 0,
1257 netdev->name, adapter);
1258 if (!err)
1259 goto request_done;
1260
1261 /* fall back to legacy interrupts */
1262 igb_reset_interrupt_capability(adapter);
1263 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1264 }
1265
1266 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1267 netdev->name, adapter);
1268
1269 if (err)
1270 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1271 err);
1272
1273 request_done:
1274 return err;
1275 }
1276
1277 static void igb_free_irq(struct igb_adapter *adapter)
1278 {
1279 if (adapter->msix_entries) {
1280 int vector = 0, i;
1281
1282 free_irq(adapter->msix_entries[vector++].vector, adapter);
1283
1284 for (i = 0; i < adapter->num_q_vectors; i++)
1285 free_irq(adapter->msix_entries[vector++].vector,
1286 adapter->q_vector[i]);
1287 } else {
1288 free_irq(adapter->pdev->irq, adapter);
1289 }
1290 }
1291
1292 /**
1293 * igb_irq_disable - Mask off interrupt generation on the NIC
1294 * @adapter: board private structure
1295 **/
1296 static void igb_irq_disable(struct igb_adapter *adapter)
1297 {
1298 struct e1000_hw *hw = &adapter->hw;
1299
1300 /*
1301 * we need to be careful when disabling interrupts. The VFs are also
1302 * mapped into these registers and so clearing the bits can cause
1303 * issues on the VF drivers so we only need to clear what we set
1304 */
1305 if (adapter->msix_entries) {
1306 u32 regval = rd32(E1000_EIAM);
1307 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1308 wr32(E1000_EIMC, adapter->eims_enable_mask);
1309 regval = rd32(E1000_EIAC);
1310 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1311 }
1312
1313 wr32(E1000_IAM, 0);
1314 wr32(E1000_IMC, ~0);
1315 wrfl();
1316 if (adapter->msix_entries) {
1317 int i;
1318 for (i = 0; i < adapter->num_q_vectors; i++)
1319 synchronize_irq(adapter->msix_entries[i].vector);
1320 } else {
1321 synchronize_irq(adapter->pdev->irq);
1322 }
1323 }
1324
1325 /**
1326 * igb_irq_enable - Enable default interrupt generation settings
1327 * @adapter: board private structure
1328 **/
1329 static void igb_irq_enable(struct igb_adapter *adapter)
1330 {
1331 struct e1000_hw *hw = &adapter->hw;
1332
1333 if (adapter->msix_entries) {
1334 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1335 u32 regval = rd32(E1000_EIAC);
1336 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1337 regval = rd32(E1000_EIAM);
1338 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1339 wr32(E1000_EIMS, adapter->eims_enable_mask);
1340 if (adapter->vfs_allocated_count) {
1341 wr32(E1000_MBVFIMR, 0xFF);
1342 ims |= E1000_IMS_VMMB;
1343 }
1344 wr32(E1000_IMS, ims);
1345 } else {
1346 wr32(E1000_IMS, IMS_ENABLE_MASK |
1347 E1000_IMS_DRSTA);
1348 wr32(E1000_IAM, IMS_ENABLE_MASK |
1349 E1000_IMS_DRSTA);
1350 }
1351 }
1352
1353 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1354 {
1355 struct e1000_hw *hw = &adapter->hw;
1356 u16 vid = adapter->hw.mng_cookie.vlan_id;
1357 u16 old_vid = adapter->mng_vlan_id;
1358
1359 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1360 /* add VID to filter table */
1361 igb_vfta_set(hw, vid, true);
1362 adapter->mng_vlan_id = vid;
1363 } else {
1364 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1365 }
1366
1367 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1368 (vid != old_vid) &&
1369 !test_bit(old_vid, adapter->active_vlans)) {
1370 /* remove VID from filter table */
1371 igb_vfta_set(hw, old_vid, false);
1372 }
1373 }
1374
1375 /**
1376 * igb_release_hw_control - release control of the h/w to f/w
1377 * @adapter: address of board private structure
1378 *
1379 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1380 * For ASF and Pass Through versions of f/w this means that the
1381 * driver is no longer loaded.
1382 *
1383 **/
1384 static void igb_release_hw_control(struct igb_adapter *adapter)
1385 {
1386 struct e1000_hw *hw = &adapter->hw;
1387 u32 ctrl_ext;
1388
1389 /* Let firmware take over control of h/w */
1390 ctrl_ext = rd32(E1000_CTRL_EXT);
1391 wr32(E1000_CTRL_EXT,
1392 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1393 }
1394
1395 /**
1396 * igb_get_hw_control - get control of the h/w from f/w
1397 * @adapter: address of board private structure
1398 *
1399 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1400 * For ASF and Pass Through versions of f/w this means that
1401 * the driver is loaded.
1402 *
1403 **/
1404 static void igb_get_hw_control(struct igb_adapter *adapter)
1405 {
1406 struct e1000_hw *hw = &adapter->hw;
1407 u32 ctrl_ext;
1408
1409 /* Let firmware know the driver has taken over */
1410 ctrl_ext = rd32(E1000_CTRL_EXT);
1411 wr32(E1000_CTRL_EXT,
1412 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1413 }
1414
1415 /**
1416 * igb_configure - configure the hardware for RX and TX
1417 * @adapter: private board structure
1418 **/
1419 static void igb_configure(struct igb_adapter *adapter)
1420 {
1421 struct net_device *netdev = adapter->netdev;
1422 int i;
1423
1424 igb_get_hw_control(adapter);
1425 igb_set_rx_mode(netdev);
1426
1427 igb_restore_vlan(adapter);
1428
1429 igb_setup_tctl(adapter);
1430 igb_setup_mrqc(adapter);
1431 igb_setup_rctl(adapter);
1432
1433 igb_configure_tx(adapter);
1434 igb_configure_rx(adapter);
1435
1436 igb_rx_fifo_flush_82575(&adapter->hw);
1437
1438 /* call igb_desc_unused which always leaves
1439 * at least 1 descriptor unused to make sure
1440 * next_to_use != next_to_clean */
1441 for (i = 0; i < adapter->num_rx_queues; i++) {
1442 struct igb_ring *ring = adapter->rx_ring[i];
1443 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1444 }
1445 }
1446
1447 /**
1448 * igb_power_up_link - Power up the phy/serdes link
1449 * @adapter: address of board private structure
1450 **/
1451 void igb_power_up_link(struct igb_adapter *adapter)
1452 {
1453 igb_reset_phy(&adapter->hw);
1454
1455 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1456 igb_power_up_phy_copper(&adapter->hw);
1457 else
1458 igb_power_up_serdes_link_82575(&adapter->hw);
1459 }
1460
1461 /**
1462 * igb_power_down_link - Power down the phy/serdes link
1463 * @adapter: address of board private structure
1464 */
1465 static void igb_power_down_link(struct igb_adapter *adapter)
1466 {
1467 if (adapter->hw.phy.media_type == e1000_media_type_copper)
1468 igb_power_down_phy_copper_82575(&adapter->hw);
1469 else
1470 igb_shutdown_serdes_link_82575(&adapter->hw);
1471 }
1472
1473 /**
1474 * igb_up - Open the interface and prepare it to handle traffic
1475 * @adapter: board private structure
1476 **/
1477 int igb_up(struct igb_adapter *adapter)
1478 {
1479 struct e1000_hw *hw = &adapter->hw;
1480 int i;
1481
1482 /* hardware has been reset, we need to reload some things */
1483 igb_configure(adapter);
1484
1485 clear_bit(__IGB_DOWN, &adapter->state);
1486
1487 for (i = 0; i < adapter->num_q_vectors; i++)
1488 napi_enable(&(adapter->q_vector[i]->napi));
1489
1490 if (adapter->msix_entries)
1491 igb_configure_msix(adapter);
1492 else
1493 igb_assign_vector(adapter->q_vector[0], 0);
1494
1495 /* Clear any pending interrupts. */
1496 rd32(E1000_ICR);
1497 igb_irq_enable(adapter);
1498
1499 /* notify VFs that reset has been completed */
1500 if (adapter->vfs_allocated_count) {
1501 u32 reg_data = rd32(E1000_CTRL_EXT);
1502 reg_data |= E1000_CTRL_EXT_PFRSTD;
1503 wr32(E1000_CTRL_EXT, reg_data);
1504 }
1505
1506 netif_tx_start_all_queues(adapter->netdev);
1507
1508 /* start the watchdog. */
1509 hw->mac.get_link_status = 1;
1510 schedule_work(&adapter->watchdog_task);
1511
1512 return 0;
1513 }
1514
1515 void igb_down(struct igb_adapter *adapter)
1516 {
1517 struct net_device *netdev = adapter->netdev;
1518 struct e1000_hw *hw = &adapter->hw;
1519 u32 tctl, rctl;
1520 int i;
1521
1522 /* signal that we're down so the interrupt handler does not
1523 * reschedule our watchdog timer */
1524 set_bit(__IGB_DOWN, &adapter->state);
1525
1526 /* disable receives in the hardware */
1527 rctl = rd32(E1000_RCTL);
1528 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1529 /* flush and sleep below */
1530
1531 netif_tx_stop_all_queues(netdev);
1532
1533 /* disable transmits in the hardware */
1534 tctl = rd32(E1000_TCTL);
1535 tctl &= ~E1000_TCTL_EN;
1536 wr32(E1000_TCTL, tctl);
1537 /* flush both disables and wait for them to finish */
1538 wrfl();
1539 msleep(10);
1540
1541 for (i = 0; i < adapter->num_q_vectors; i++)
1542 napi_disable(&(adapter->q_vector[i]->napi));
1543
1544 igb_irq_disable(adapter);
1545
1546 del_timer_sync(&adapter->watchdog_timer);
1547 del_timer_sync(&adapter->phy_info_timer);
1548
1549 netif_carrier_off(netdev);
1550
1551 /* record the stats before reset*/
1552 spin_lock(&adapter->stats64_lock);
1553 igb_update_stats(adapter, &adapter->stats64);
1554 spin_unlock(&adapter->stats64_lock);
1555
1556 adapter->link_speed = 0;
1557 adapter->link_duplex = 0;
1558
1559 if (!pci_channel_offline(adapter->pdev))
1560 igb_reset(adapter);
1561 igb_clean_all_tx_rings(adapter);
1562 igb_clean_all_rx_rings(adapter);
1563 #ifdef CONFIG_IGB_DCA
1564
1565 /* since we reset the hardware DCA settings were cleared */
1566 igb_setup_dca(adapter);
1567 #endif
1568 }
1569
1570 void igb_reinit_locked(struct igb_adapter *adapter)
1571 {
1572 WARN_ON(in_interrupt());
1573 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1574 msleep(1);
1575 igb_down(adapter);
1576 igb_up(adapter);
1577 clear_bit(__IGB_RESETTING, &adapter->state);
1578 }
1579
1580 void igb_reset(struct igb_adapter *adapter)
1581 {
1582 struct pci_dev *pdev = adapter->pdev;
1583 struct e1000_hw *hw = &adapter->hw;
1584 struct e1000_mac_info *mac = &hw->mac;
1585 struct e1000_fc_info *fc = &hw->fc;
1586 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
1587 u16 hwm;
1588
1589 /* Repartition Pba for greater than 9k mtu
1590 * To take effect CTRL.RST is required.
1591 */
1592 switch (mac->type) {
1593 case e1000_i350:
1594 case e1000_82580:
1595 pba = rd32(E1000_RXPBS);
1596 pba = igb_rxpbs_adjust_82580(pba);
1597 break;
1598 case e1000_82576:
1599 pba = rd32(E1000_RXPBS);
1600 pba &= E1000_RXPBS_SIZE_MASK_82576;
1601 break;
1602 case e1000_82575:
1603 case e1000_i210:
1604 case e1000_i211:
1605 default:
1606 pba = E1000_PBA_34K;
1607 break;
1608 }
1609
1610 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1611 (mac->type < e1000_82576)) {
1612 /* adjust PBA for jumbo frames */
1613 wr32(E1000_PBA, pba);
1614
1615 /* To maintain wire speed transmits, the Tx FIFO should be
1616 * large enough to accommodate two full transmit packets,
1617 * rounded up to the next 1KB and expressed in KB. Likewise,
1618 * the Rx FIFO should be large enough to accommodate at least
1619 * one full receive packet and is similarly rounded up and
1620 * expressed in KB. */
1621 pba = rd32(E1000_PBA);
1622 /* upper 16 bits has Tx packet buffer allocation size in KB */
1623 tx_space = pba >> 16;
1624 /* lower 16 bits has Rx packet buffer allocation size in KB */
1625 pba &= 0xffff;
1626 /* the tx fifo also stores 16 bytes of information about the tx
1627 * but don't include ethernet FCS because hardware appends it */
1628 min_tx_space = (adapter->max_frame_size +
1629 sizeof(union e1000_adv_tx_desc) -
1630 ETH_FCS_LEN) * 2;
1631 min_tx_space = ALIGN(min_tx_space, 1024);
1632 min_tx_space >>= 10;
1633 /* software strips receive CRC, so leave room for it */
1634 min_rx_space = adapter->max_frame_size;
1635 min_rx_space = ALIGN(min_rx_space, 1024);
1636 min_rx_space >>= 10;
1637
1638 /* If current Tx allocation is less than the min Tx FIFO size,
1639 * and the min Tx FIFO size is less than the current Rx FIFO
1640 * allocation, take space away from current Rx allocation */
1641 if (tx_space < min_tx_space &&
1642 ((min_tx_space - tx_space) < pba)) {
1643 pba = pba - (min_tx_space - tx_space);
1644
1645 /* if short on rx space, rx wins and must trump tx
1646 * adjustment */
1647 if (pba < min_rx_space)
1648 pba = min_rx_space;
1649 }
1650 wr32(E1000_PBA, pba);
1651 }
1652
1653 /* flow control settings */
1654 /* The high water mark must be low enough to fit one full frame
1655 * (or the size used for early receive) above it in the Rx FIFO.
1656 * Set it to the lower of:
1657 * - 90% of the Rx FIFO size, or
1658 * - the full Rx FIFO size minus one full frame */
1659 hwm = min(((pba << 10) * 9 / 10),
1660 ((pba << 10) - 2 * adapter->max_frame_size));
1661
1662 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1663 fc->low_water = fc->high_water - 16;
1664 fc->pause_time = 0xFFFF;
1665 fc->send_xon = 1;
1666 fc->current_mode = fc->requested_mode;
1667
1668 /* disable receive for all VFs and wait one second */
1669 if (adapter->vfs_allocated_count) {
1670 int i;
1671 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1672 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1673
1674 /* ping all the active vfs to let them know we are going down */
1675 igb_ping_all_vfs(adapter);
1676
1677 /* disable transmits and receives */
1678 wr32(E1000_VFRE, 0);
1679 wr32(E1000_VFTE, 0);
1680 }
1681
1682 /* Allow time for pending master requests to run */
1683 hw->mac.ops.reset_hw(hw);
1684 wr32(E1000_WUC, 0);
1685
1686 if (hw->mac.ops.init_hw(hw))
1687 dev_err(&pdev->dev, "Hardware Error\n");
1688
1689 /*
1690 * Flow control settings reset on hardware reset, so guarantee flow
1691 * control is off when forcing speed.
1692 */
1693 if (!hw->mac.autoneg)
1694 igb_force_mac_fc(hw);
1695
1696 igb_init_dmac(adapter, pba);
1697 if (!netif_running(adapter->netdev))
1698 igb_power_down_link(adapter);
1699
1700 igb_update_mng_vlan(adapter);
1701
1702 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1703 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1704
1705 #ifdef CONFIG_IGB_PTP
1706 /* Re-enable PTP, where applicable. */
1707 igb_ptp_reset(adapter);
1708 #endif /* CONFIG_IGB_PTP */
1709
1710 igb_get_phy_info(hw);
1711 }
1712
1713 static netdev_features_t igb_fix_features(struct net_device *netdev,
1714 netdev_features_t features)
1715 {
1716 /*
1717 * Since there is no support for separate rx/tx vlan accel
1718 * enable/disable make sure tx flag is always in same state as rx.
1719 */
1720 if (features & NETIF_F_HW_VLAN_RX)
1721 features |= NETIF_F_HW_VLAN_TX;
1722 else
1723 features &= ~NETIF_F_HW_VLAN_TX;
1724
1725 return features;
1726 }
1727
1728 static int igb_set_features(struct net_device *netdev,
1729 netdev_features_t features)
1730 {
1731 netdev_features_t changed = netdev->features ^ features;
1732 struct igb_adapter *adapter = netdev_priv(netdev);
1733
1734 if (changed & NETIF_F_HW_VLAN_RX)
1735 igb_vlan_mode(netdev, features);
1736
1737 if (!(changed & NETIF_F_RXALL))
1738 return 0;
1739
1740 netdev->features = features;
1741
1742 if (netif_running(netdev))
1743 igb_reinit_locked(adapter);
1744 else
1745 igb_reset(adapter);
1746
1747 return 0;
1748 }
1749
1750 static const struct net_device_ops igb_netdev_ops = {
1751 .ndo_open = igb_open,
1752 .ndo_stop = igb_close,
1753 .ndo_start_xmit = igb_xmit_frame,
1754 .ndo_get_stats64 = igb_get_stats64,
1755 .ndo_set_rx_mode = igb_set_rx_mode,
1756 .ndo_set_mac_address = igb_set_mac,
1757 .ndo_change_mtu = igb_change_mtu,
1758 .ndo_do_ioctl = igb_ioctl,
1759 .ndo_tx_timeout = igb_tx_timeout,
1760 .ndo_validate_addr = eth_validate_addr,
1761 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1762 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1763 .ndo_set_vf_mac = igb_ndo_set_vf_mac,
1764 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan,
1765 .ndo_set_vf_tx_rate = igb_ndo_set_vf_bw,
1766 .ndo_get_vf_config = igb_ndo_get_vf_config,
1767 #ifdef CONFIG_NET_POLL_CONTROLLER
1768 .ndo_poll_controller = igb_netpoll,
1769 #endif
1770 .ndo_fix_features = igb_fix_features,
1771 .ndo_set_features = igb_set_features,
1772 };
1773
1774 /**
1775 * igb_set_fw_version - Configure version string for ethtool
1776 * @adapter: adapter struct
1777 *
1778 **/
1779 void igb_set_fw_version(struct igb_adapter *adapter)
1780 {
1781 struct e1000_hw *hw = &adapter->hw;
1782 u16 eeprom_verh, eeprom_verl, comb_verh, comb_verl, comb_offset;
1783 u16 major, build, patch, fw_version;
1784 u32 etrack_id;
1785
1786 hw->nvm.ops.read(hw, 5, 1, &fw_version);
1787 if (adapter->hw.mac.type != e1000_i211) {
1788 hw->nvm.ops.read(hw, NVM_ETRACK_WORD, 1, &eeprom_verh);
1789 hw->nvm.ops.read(hw, (NVM_ETRACK_WORD + 1), 1, &eeprom_verl);
1790 etrack_id = (eeprom_verh << IGB_ETRACK_SHIFT) | eeprom_verl;
1791
1792 /* combo image version needs to be found */
1793 hw->nvm.ops.read(hw, NVM_COMB_VER_PTR, 1, &comb_offset);
1794 if ((comb_offset != 0x0) &&
1795 (comb_offset != IGB_NVM_VER_INVALID)) {
1796 hw->nvm.ops.read(hw, (NVM_COMB_VER_OFF + comb_offset
1797 + 1), 1, &comb_verh);
1798 hw->nvm.ops.read(hw, (NVM_COMB_VER_OFF + comb_offset),
1799 1, &comb_verl);
1800
1801 /* Only display Option Rom if it exists and is valid */
1802 if ((comb_verh && comb_verl) &&
1803 ((comb_verh != IGB_NVM_VER_INVALID) &&
1804 (comb_verl != IGB_NVM_VER_INVALID))) {
1805 major = comb_verl >> IGB_COMB_VER_SHFT;
1806 build = (comb_verl << IGB_COMB_VER_SHFT) |
1807 (comb_verh >> IGB_COMB_VER_SHFT);
1808 patch = comb_verh & IGB_COMB_VER_MASK;
1809 snprintf(adapter->fw_version,
1810 sizeof(adapter->fw_version),
1811 "%d.%d%d, 0x%08x, %d.%d.%d",
1812 (fw_version & IGB_MAJOR_MASK) >>
1813 IGB_MAJOR_SHIFT,
1814 (fw_version & IGB_MINOR_MASK) >>
1815 IGB_MINOR_SHIFT,
1816 (fw_version & IGB_BUILD_MASK),
1817 etrack_id, major, build, patch);
1818 goto out;
1819 }
1820 }
1821 snprintf(adapter->fw_version, sizeof(adapter->fw_version),
1822 "%d.%d%d, 0x%08x",
1823 (fw_version & IGB_MAJOR_MASK) >> IGB_MAJOR_SHIFT,
1824 (fw_version & IGB_MINOR_MASK) >> IGB_MINOR_SHIFT,
1825 (fw_version & IGB_BUILD_MASK), etrack_id);
1826 } else {
1827 snprintf(adapter->fw_version, sizeof(adapter->fw_version),
1828 "%d.%d%d",
1829 (fw_version & IGB_MAJOR_MASK) >> IGB_MAJOR_SHIFT,
1830 (fw_version & IGB_MINOR_MASK) >> IGB_MINOR_SHIFT,
1831 (fw_version & IGB_BUILD_MASK));
1832 }
1833 out:
1834 return;
1835 }
1836
1837 /**
1838 * igb_probe - Device Initialization Routine
1839 * @pdev: PCI device information struct
1840 * @ent: entry in igb_pci_tbl
1841 *
1842 * Returns 0 on success, negative on failure
1843 *
1844 * igb_probe initializes an adapter identified by a pci_dev structure.
1845 * The OS initialization, configuring of the adapter private structure,
1846 * and a hardware reset occur.
1847 **/
1848 static int __devinit igb_probe(struct pci_dev *pdev,
1849 const struct pci_device_id *ent)
1850 {
1851 struct net_device *netdev;
1852 struct igb_adapter *adapter;
1853 struct e1000_hw *hw;
1854 u16 eeprom_data = 0;
1855 s32 ret_val;
1856 static int global_quad_port_a; /* global quad port a indication */
1857 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1858 unsigned long mmio_start, mmio_len;
1859 int err, pci_using_dac;
1860 u16 eeprom_apme_mask = IGB_EEPROM_APME;
1861 u8 part_str[E1000_PBANUM_LENGTH];
1862
1863 /* Catch broken hardware that put the wrong VF device ID in
1864 * the PCIe SR-IOV capability.
1865 */
1866 if (pdev->is_virtfn) {
1867 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
1868 pci_name(pdev), pdev->vendor, pdev->device);
1869 return -EINVAL;
1870 }
1871
1872 err = pci_enable_device_mem(pdev);
1873 if (err)
1874 return err;
1875
1876 pci_using_dac = 0;
1877 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
1878 if (!err) {
1879 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
1880 if (!err)
1881 pci_using_dac = 1;
1882 } else {
1883 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
1884 if (err) {
1885 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
1886 if (err) {
1887 dev_err(&pdev->dev, "No usable DMA "
1888 "configuration, aborting\n");
1889 goto err_dma;
1890 }
1891 }
1892 }
1893
1894 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1895 IORESOURCE_MEM),
1896 igb_driver_name);
1897 if (err)
1898 goto err_pci_reg;
1899
1900 pci_enable_pcie_error_reporting(pdev);
1901
1902 pci_set_master(pdev);
1903 pci_save_state(pdev);
1904
1905 err = -ENOMEM;
1906 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1907 IGB_MAX_TX_QUEUES);
1908 if (!netdev)
1909 goto err_alloc_etherdev;
1910
1911 SET_NETDEV_DEV(netdev, &pdev->dev);
1912
1913 pci_set_drvdata(pdev, netdev);
1914 adapter = netdev_priv(netdev);
1915 adapter->netdev = netdev;
1916 adapter->pdev = pdev;
1917 hw = &adapter->hw;
1918 hw->back = adapter;
1919 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
1920
1921 mmio_start = pci_resource_start(pdev, 0);
1922 mmio_len = pci_resource_len(pdev, 0);
1923
1924 err = -EIO;
1925 hw->hw_addr = ioremap(mmio_start, mmio_len);
1926 if (!hw->hw_addr)
1927 goto err_ioremap;
1928
1929 netdev->netdev_ops = &igb_netdev_ops;
1930 igb_set_ethtool_ops(netdev);
1931 netdev->watchdog_timeo = 5 * HZ;
1932
1933 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1934
1935 netdev->mem_start = mmio_start;
1936 netdev->mem_end = mmio_start + mmio_len;
1937
1938 /* PCI config space info */
1939 hw->vendor_id = pdev->vendor;
1940 hw->device_id = pdev->device;
1941 hw->revision_id = pdev->revision;
1942 hw->subsystem_vendor_id = pdev->subsystem_vendor;
1943 hw->subsystem_device_id = pdev->subsystem_device;
1944
1945 /* Copy the default MAC, PHY and NVM function pointers */
1946 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1947 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1948 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1949 /* Initialize skew-specific constants */
1950 err = ei->get_invariants(hw);
1951 if (err)
1952 goto err_sw_init;
1953
1954 /* setup the private structure */
1955 err = igb_sw_init(adapter);
1956 if (err)
1957 goto err_sw_init;
1958
1959 igb_get_bus_info_pcie(hw);
1960
1961 hw->phy.autoneg_wait_to_complete = false;
1962
1963 /* Copper options */
1964 if (hw->phy.media_type == e1000_media_type_copper) {
1965 hw->phy.mdix = AUTO_ALL_MODES;
1966 hw->phy.disable_polarity_correction = false;
1967 hw->phy.ms_type = e1000_ms_hw_default;
1968 }
1969
1970 if (igb_check_reset_block(hw))
1971 dev_info(&pdev->dev,
1972 "PHY reset is blocked due to SOL/IDER session.\n");
1973
1974 /*
1975 * features is initialized to 0 in allocation, it might have bits
1976 * set by igb_sw_init so we should use an or instead of an
1977 * assignment.
1978 */
1979 netdev->features |= NETIF_F_SG |
1980 NETIF_F_IP_CSUM |
1981 NETIF_F_IPV6_CSUM |
1982 NETIF_F_TSO |
1983 NETIF_F_TSO6 |
1984 NETIF_F_RXHASH |
1985 NETIF_F_RXCSUM |
1986 NETIF_F_HW_VLAN_RX |
1987 NETIF_F_HW_VLAN_TX;
1988
1989 /* copy netdev features into list of user selectable features */
1990 netdev->hw_features |= netdev->features;
1991 netdev->hw_features |= NETIF_F_RXALL;
1992
1993 /* set this bit last since it cannot be part of hw_features */
1994 netdev->features |= NETIF_F_HW_VLAN_FILTER;
1995
1996 netdev->vlan_features |= NETIF_F_TSO |
1997 NETIF_F_TSO6 |
1998 NETIF_F_IP_CSUM |
1999 NETIF_F_IPV6_CSUM |
2000 NETIF_F_SG;
2001
2002 netdev->priv_flags |= IFF_SUPP_NOFCS;
2003
2004 if (pci_using_dac) {
2005 netdev->features |= NETIF_F_HIGHDMA;
2006 netdev->vlan_features |= NETIF_F_HIGHDMA;
2007 }
2008
2009 if (hw->mac.type >= e1000_82576) {
2010 netdev->hw_features |= NETIF_F_SCTP_CSUM;
2011 netdev->features |= NETIF_F_SCTP_CSUM;
2012 }
2013
2014 netdev->priv_flags |= IFF_UNICAST_FLT;
2015
2016 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2017
2018 /* before reading the NVM, reset the controller to put the device in a
2019 * known good starting state */
2020 hw->mac.ops.reset_hw(hw);
2021
2022 /*
2023 * make sure the NVM is good , i211 parts have special NVM that
2024 * doesn't contain a checksum
2025 */
2026 if (hw->mac.type != e1000_i211) {
2027 if (hw->nvm.ops.validate(hw) < 0) {
2028 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2029 err = -EIO;
2030 goto err_eeprom;
2031 }
2032 }
2033
2034 /* copy the MAC address out of the NVM */
2035 if (hw->mac.ops.read_mac_addr(hw))
2036 dev_err(&pdev->dev, "NVM Read Error\n");
2037
2038 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2039 memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
2040
2041 if (!is_valid_ether_addr(netdev->perm_addr)) {
2042 dev_err(&pdev->dev, "Invalid MAC Address\n");
2043 err = -EIO;
2044 goto err_eeprom;
2045 }
2046
2047 /* get firmware version for ethtool -i */
2048 igb_set_fw_version(adapter);
2049
2050 setup_timer(&adapter->watchdog_timer, igb_watchdog,
2051 (unsigned long) adapter);
2052 setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2053 (unsigned long) adapter);
2054
2055 INIT_WORK(&adapter->reset_task, igb_reset_task);
2056 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2057
2058 /* Initialize link properties that are user-changeable */
2059 adapter->fc_autoneg = true;
2060 hw->mac.autoneg = true;
2061 hw->phy.autoneg_advertised = 0x2f;
2062
2063 hw->fc.requested_mode = e1000_fc_default;
2064 hw->fc.current_mode = e1000_fc_default;
2065
2066 igb_validate_mdi_setting(hw);
2067
2068 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
2069 * enable the ACPI Magic Packet filter
2070 */
2071
2072 if (hw->bus.func == 0)
2073 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
2074 else if (hw->mac.type >= e1000_82580)
2075 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2076 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2077 &eeprom_data);
2078 else if (hw->bus.func == 1)
2079 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2080
2081 if (eeprom_data & eeprom_apme_mask)
2082 adapter->eeprom_wol |= E1000_WUFC_MAG;
2083
2084 /* now that we have the eeprom settings, apply the special cases where
2085 * the eeprom may be wrong or the board simply won't support wake on
2086 * lan on a particular port */
2087 switch (pdev->device) {
2088 case E1000_DEV_ID_82575GB_QUAD_COPPER:
2089 adapter->eeprom_wol = 0;
2090 break;
2091 case E1000_DEV_ID_82575EB_FIBER_SERDES:
2092 case E1000_DEV_ID_82576_FIBER:
2093 case E1000_DEV_ID_82576_SERDES:
2094 /* Wake events only supported on port A for dual fiber
2095 * regardless of eeprom setting */
2096 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2097 adapter->eeprom_wol = 0;
2098 break;
2099 case E1000_DEV_ID_82576_QUAD_COPPER:
2100 case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2101 /* if quad port adapter, disable WoL on all but port A */
2102 if (global_quad_port_a != 0)
2103 adapter->eeprom_wol = 0;
2104 else
2105 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2106 /* Reset for multiple quad port adapters */
2107 if (++global_quad_port_a == 4)
2108 global_quad_port_a = 0;
2109 break;
2110 }
2111
2112 /* initialize the wol settings based on the eeprom settings */
2113 adapter->wol = adapter->eeprom_wol;
2114 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
2115
2116 /* reset the hardware with the new settings */
2117 igb_reset(adapter);
2118
2119 /* let the f/w know that the h/w is now under the control of the
2120 * driver. */
2121 igb_get_hw_control(adapter);
2122
2123 strcpy(netdev->name, "eth%d");
2124 err = register_netdev(netdev);
2125 if (err)
2126 goto err_register;
2127
2128 /* carrier off reporting is important to ethtool even BEFORE open */
2129 netif_carrier_off(netdev);
2130
2131 #ifdef CONFIG_IGB_DCA
2132 if (dca_add_requester(&pdev->dev) == 0) {
2133 adapter->flags |= IGB_FLAG_DCA_ENABLED;
2134 dev_info(&pdev->dev, "DCA enabled\n");
2135 igb_setup_dca(adapter);
2136 }
2137
2138 #endif
2139
2140 #ifdef CONFIG_IGB_PTP
2141 /* do hw tstamp init after resetting */
2142 igb_ptp_init(adapter);
2143 #endif /* CONFIG_IGB_PTP */
2144
2145 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2146 /* print bus type/speed/width info */
2147 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2148 netdev->name,
2149 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2150 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2151 "unknown"),
2152 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
2153 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
2154 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
2155 "unknown"),
2156 netdev->dev_addr);
2157
2158 ret_val = igb_read_part_string(hw, part_str, E1000_PBANUM_LENGTH);
2159 if (ret_val)
2160 strcpy(part_str, "Unknown");
2161 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2162 dev_info(&pdev->dev,
2163 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2164 adapter->msix_entries ? "MSI-X" :
2165 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2166 adapter->num_rx_queues, adapter->num_tx_queues);
2167 switch (hw->mac.type) {
2168 case e1000_i350:
2169 case e1000_i210:
2170 case e1000_i211:
2171 igb_set_eee_i350(hw);
2172 break;
2173 default:
2174 break;
2175 }
2176
2177 pm_runtime_put_noidle(&pdev->dev);
2178 return 0;
2179
2180 err_register:
2181 igb_release_hw_control(adapter);
2182 err_eeprom:
2183 if (!igb_check_reset_block(hw))
2184 igb_reset_phy(hw);
2185
2186 if (hw->flash_address)
2187 iounmap(hw->flash_address);
2188 err_sw_init:
2189 igb_clear_interrupt_scheme(adapter);
2190 iounmap(hw->hw_addr);
2191 err_ioremap:
2192 free_netdev(netdev);
2193 err_alloc_etherdev:
2194 pci_release_selected_regions(pdev,
2195 pci_select_bars(pdev, IORESOURCE_MEM));
2196 err_pci_reg:
2197 err_dma:
2198 pci_disable_device(pdev);
2199 return err;
2200 }
2201
2202 /**
2203 * igb_remove - Device Removal Routine
2204 * @pdev: PCI device information struct
2205 *
2206 * igb_remove is called by the PCI subsystem to alert the driver
2207 * that it should release a PCI device. The could be caused by a
2208 * Hot-Plug event, or because the driver is going to be removed from
2209 * memory.
2210 **/
2211 static void __devexit igb_remove(struct pci_dev *pdev)
2212 {
2213 struct net_device *netdev = pci_get_drvdata(pdev);
2214 struct igb_adapter *adapter = netdev_priv(netdev);
2215 struct e1000_hw *hw = &adapter->hw;
2216
2217 pm_runtime_get_noresume(&pdev->dev);
2218 #ifdef CONFIG_IGB_PTP
2219 igb_ptp_stop(adapter);
2220 #endif /* CONFIG_IGB_PTP */
2221
2222 /*
2223 * The watchdog timer may be rescheduled, so explicitly
2224 * disable watchdog from being rescheduled.
2225 */
2226 set_bit(__IGB_DOWN, &adapter->state);
2227 del_timer_sync(&adapter->watchdog_timer);
2228 del_timer_sync(&adapter->phy_info_timer);
2229
2230 cancel_work_sync(&adapter->reset_task);
2231 cancel_work_sync(&adapter->watchdog_task);
2232
2233 #ifdef CONFIG_IGB_DCA
2234 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2235 dev_info(&pdev->dev, "DCA disabled\n");
2236 dca_remove_requester(&pdev->dev);
2237 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2238 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2239 }
2240 #endif
2241
2242 /* Release control of h/w to f/w. If f/w is AMT enabled, this
2243 * would have already happened in close and is redundant. */
2244 igb_release_hw_control(adapter);
2245
2246 unregister_netdev(netdev);
2247
2248 igb_clear_interrupt_scheme(adapter);
2249
2250 #ifdef CONFIG_PCI_IOV
2251 /* reclaim resources allocated to VFs */
2252 if (adapter->vf_data) {
2253 /* disable iov and allow time for transactions to clear */
2254 if (igb_vfs_are_assigned(adapter)) {
2255 dev_info(&pdev->dev, "Unloading driver while VFs are assigned - VFs will not be deallocated\n");
2256 } else {
2257 pci_disable_sriov(pdev);
2258 msleep(500);
2259 }
2260
2261 kfree(adapter->vf_data);
2262 adapter->vf_data = NULL;
2263 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2264 wrfl();
2265 msleep(100);
2266 dev_info(&pdev->dev, "IOV Disabled\n");
2267 }
2268 #endif
2269
2270 iounmap(hw->hw_addr);
2271 if (hw->flash_address)
2272 iounmap(hw->flash_address);
2273 pci_release_selected_regions(pdev,
2274 pci_select_bars(pdev, IORESOURCE_MEM));
2275
2276 kfree(adapter->shadow_vfta);
2277 free_netdev(netdev);
2278
2279 pci_disable_pcie_error_reporting(pdev);
2280
2281 pci_disable_device(pdev);
2282 }
2283
2284 /**
2285 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2286 * @adapter: board private structure to initialize
2287 *
2288 * This function initializes the vf specific data storage and then attempts to
2289 * allocate the VFs. The reason for ordering it this way is because it is much
2290 * mor expensive time wise to disable SR-IOV than it is to allocate and free
2291 * the memory for the VFs.
2292 **/
2293 static void __devinit igb_probe_vfs(struct igb_adapter * adapter)
2294 {
2295 #ifdef CONFIG_PCI_IOV
2296 struct pci_dev *pdev = adapter->pdev;
2297 struct e1000_hw *hw = &adapter->hw;
2298 int old_vfs = pci_num_vf(adapter->pdev);
2299 int i;
2300
2301 /* Virtualization features not supported on i210 family. */
2302 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
2303 return;
2304
2305 if (old_vfs) {
2306 dev_info(&pdev->dev, "%d pre-allocated VFs found - override "
2307 "max_vfs setting of %d\n", old_vfs, max_vfs);
2308 adapter->vfs_allocated_count = old_vfs;
2309 }
2310
2311 if (!adapter->vfs_allocated_count)
2312 return;
2313
2314 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2315 sizeof(struct vf_data_storage), GFP_KERNEL);
2316
2317 /* if allocation failed then we do not support SR-IOV */
2318 if (!adapter->vf_data) {
2319 adapter->vfs_allocated_count = 0;
2320 dev_err(&pdev->dev, "Unable to allocate memory for VF "
2321 "Data Storage\n");
2322 goto out;
2323 }
2324
2325 if (!old_vfs) {
2326 if (pci_enable_sriov(pdev, adapter->vfs_allocated_count))
2327 goto err_out;
2328 }
2329 dev_info(&pdev->dev, "%d VFs allocated\n",
2330 adapter->vfs_allocated_count);
2331 for (i = 0; i < adapter->vfs_allocated_count; i++)
2332 igb_vf_configure(adapter, i);
2333
2334 /* DMA Coalescing is not supported in IOV mode. */
2335 adapter->flags &= ~IGB_FLAG_DMAC;
2336 goto out;
2337 err_out:
2338 kfree(adapter->vf_data);
2339 adapter->vf_data = NULL;
2340 adapter->vfs_allocated_count = 0;
2341 out:
2342 return;
2343 #endif /* CONFIG_PCI_IOV */
2344 }
2345
2346 /**
2347 * igb_sw_init - Initialize general software structures (struct igb_adapter)
2348 * @adapter: board private structure to initialize
2349 *
2350 * igb_sw_init initializes the Adapter private data structure.
2351 * Fields are initialized based on PCI device information and
2352 * OS network device settings (MTU size).
2353 **/
2354 static int __devinit igb_sw_init(struct igb_adapter *adapter)
2355 {
2356 struct e1000_hw *hw = &adapter->hw;
2357 struct net_device *netdev = adapter->netdev;
2358 struct pci_dev *pdev = adapter->pdev;
2359 u32 max_rss_queues;
2360
2361 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2362
2363 /* set default ring sizes */
2364 adapter->tx_ring_count = IGB_DEFAULT_TXD;
2365 adapter->rx_ring_count = IGB_DEFAULT_RXD;
2366
2367 /* set default ITR values */
2368 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2369 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2370
2371 /* set default work limits */
2372 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2373
2374 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2375 VLAN_HLEN;
2376 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2377
2378 spin_lock_init(&adapter->stats64_lock);
2379 #ifdef CONFIG_PCI_IOV
2380 switch (hw->mac.type) {
2381 case e1000_82576:
2382 case e1000_i350:
2383 if (max_vfs > 7) {
2384 dev_warn(&pdev->dev,
2385 "Maximum of 7 VFs per PF, using max\n");
2386 adapter->vfs_allocated_count = 7;
2387 } else
2388 adapter->vfs_allocated_count = max_vfs;
2389 break;
2390 default:
2391 break;
2392 }
2393 #endif /* CONFIG_PCI_IOV */
2394
2395 /* Determine the maximum number of RSS queues supported. */
2396 switch (hw->mac.type) {
2397 case e1000_i211:
2398 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
2399 break;
2400 case e1000_82575:
2401 case e1000_i210:
2402 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
2403 break;
2404 case e1000_i350:
2405 /* I350 cannot do RSS and SR-IOV at the same time */
2406 if (!!adapter->vfs_allocated_count) {
2407 max_rss_queues = 1;
2408 break;
2409 }
2410 /* fall through */
2411 case e1000_82576:
2412 if (!!adapter->vfs_allocated_count) {
2413 max_rss_queues = 2;
2414 break;
2415 }
2416 /* fall through */
2417 case e1000_82580:
2418 default:
2419 max_rss_queues = IGB_MAX_RX_QUEUES;
2420 break;
2421 }
2422
2423 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2424
2425 /* Determine if we need to pair queues. */
2426 switch (hw->mac.type) {
2427 case e1000_82575:
2428 case e1000_i211:
2429 /* Device supports enough interrupts without queue pairing. */
2430 break;
2431 case e1000_82576:
2432 /*
2433 * If VFs are going to be allocated with RSS queues then we
2434 * should pair the queues in order to conserve interrupts due
2435 * to limited supply.
2436 */
2437 if ((adapter->rss_queues > 1) &&
2438 (adapter->vfs_allocated_count > 6))
2439 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2440 /* fall through */
2441 case e1000_82580:
2442 case e1000_i350:
2443 case e1000_i210:
2444 default:
2445 /*
2446 * If rss_queues > half of max_rss_queues, pair the queues in
2447 * order to conserve interrupts due to limited supply.
2448 */
2449 if (adapter->rss_queues > (max_rss_queues / 2))
2450 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2451 break;
2452 }
2453
2454 /* Setup and initialize a copy of the hw vlan table array */
2455 adapter->shadow_vfta = kzalloc(sizeof(u32) *
2456 E1000_VLAN_FILTER_TBL_SIZE,
2457 GFP_ATOMIC);
2458
2459 /* This call may decrease the number of queues */
2460 if (igb_init_interrupt_scheme(adapter)) {
2461 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2462 return -ENOMEM;
2463 }
2464
2465 igb_probe_vfs(adapter);
2466
2467 /* Explicitly disable IRQ since the NIC can be in any state. */
2468 igb_irq_disable(adapter);
2469
2470 if (hw->mac.type >= e1000_i350)
2471 adapter->flags &= ~IGB_FLAG_DMAC;
2472
2473 set_bit(__IGB_DOWN, &adapter->state);
2474 return 0;
2475 }
2476
2477 /**
2478 * igb_open - Called when a network interface is made active
2479 * @netdev: network interface device structure
2480 *
2481 * Returns 0 on success, negative value on failure
2482 *
2483 * The open entry point is called when a network interface is made
2484 * active by the system (IFF_UP). At this point all resources needed
2485 * for transmit and receive operations are allocated, the interrupt
2486 * handler is registered with the OS, the watchdog timer is started,
2487 * and the stack is notified that the interface is ready.
2488 **/
2489 static int __igb_open(struct net_device *netdev, bool resuming)
2490 {
2491 struct igb_adapter *adapter = netdev_priv(netdev);
2492 struct e1000_hw *hw = &adapter->hw;
2493 struct pci_dev *pdev = adapter->pdev;
2494 int err;
2495 int i;
2496
2497 /* disallow open during test */
2498 if (test_bit(__IGB_TESTING, &adapter->state)) {
2499 WARN_ON(resuming);
2500 return -EBUSY;
2501 }
2502
2503 if (!resuming)
2504 pm_runtime_get_sync(&pdev->dev);
2505
2506 netif_carrier_off(netdev);
2507
2508 /* allocate transmit descriptors */
2509 err = igb_setup_all_tx_resources(adapter);
2510 if (err)
2511 goto err_setup_tx;
2512
2513 /* allocate receive descriptors */
2514 err = igb_setup_all_rx_resources(adapter);
2515 if (err)
2516 goto err_setup_rx;
2517
2518 igb_power_up_link(adapter);
2519
2520 /* before we allocate an interrupt, we must be ready to handle it.
2521 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2522 * as soon as we call pci_request_irq, so we have to setup our
2523 * clean_rx handler before we do so. */
2524 igb_configure(adapter);
2525
2526 err = igb_request_irq(adapter);
2527 if (err)
2528 goto err_req_irq;
2529
2530 /* From here on the code is the same as igb_up() */
2531 clear_bit(__IGB_DOWN, &adapter->state);
2532
2533 for (i = 0; i < adapter->num_q_vectors; i++)
2534 napi_enable(&(adapter->q_vector[i]->napi));
2535
2536 /* Clear any pending interrupts. */
2537 rd32(E1000_ICR);
2538
2539 igb_irq_enable(adapter);
2540
2541 /* notify VFs that reset has been completed */
2542 if (adapter->vfs_allocated_count) {
2543 u32 reg_data = rd32(E1000_CTRL_EXT);
2544 reg_data |= E1000_CTRL_EXT_PFRSTD;
2545 wr32(E1000_CTRL_EXT, reg_data);
2546 }
2547
2548 netif_tx_start_all_queues(netdev);
2549
2550 if (!resuming)
2551 pm_runtime_put(&pdev->dev);
2552
2553 /* start the watchdog. */
2554 hw->mac.get_link_status = 1;
2555 schedule_work(&adapter->watchdog_task);
2556
2557 return 0;
2558
2559 err_req_irq:
2560 igb_release_hw_control(adapter);
2561 igb_power_down_link(adapter);
2562 igb_free_all_rx_resources(adapter);
2563 err_setup_rx:
2564 igb_free_all_tx_resources(adapter);
2565 err_setup_tx:
2566 igb_reset(adapter);
2567 if (!resuming)
2568 pm_runtime_put(&pdev->dev);
2569
2570 return err;
2571 }
2572
2573 static int igb_open(struct net_device *netdev)
2574 {
2575 return __igb_open(netdev, false);
2576 }
2577
2578 /**
2579 * igb_close - Disables a network interface
2580 * @netdev: network interface device structure
2581 *
2582 * Returns 0, this is not allowed to fail
2583 *
2584 * The close entry point is called when an interface is de-activated
2585 * by the OS. The hardware is still under the driver's control, but
2586 * needs to be disabled. A global MAC reset is issued to stop the
2587 * hardware, and all transmit and receive resources are freed.
2588 **/
2589 static int __igb_close(struct net_device *netdev, bool suspending)
2590 {
2591 struct igb_adapter *adapter = netdev_priv(netdev);
2592 struct pci_dev *pdev = adapter->pdev;
2593
2594 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
2595
2596 if (!suspending)
2597 pm_runtime_get_sync(&pdev->dev);
2598
2599 igb_down(adapter);
2600 igb_free_irq(adapter);
2601
2602 igb_free_all_tx_resources(adapter);
2603 igb_free_all_rx_resources(adapter);
2604
2605 if (!suspending)
2606 pm_runtime_put_sync(&pdev->dev);
2607 return 0;
2608 }
2609
2610 static int igb_close(struct net_device *netdev)
2611 {
2612 return __igb_close(netdev, false);
2613 }
2614
2615 /**
2616 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2617 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2618 *
2619 * Return 0 on success, negative on failure
2620 **/
2621 int igb_setup_tx_resources(struct igb_ring *tx_ring)
2622 {
2623 struct device *dev = tx_ring->dev;
2624 int size;
2625
2626 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
2627
2628 tx_ring->tx_buffer_info = vzalloc(size);
2629 if (!tx_ring->tx_buffer_info)
2630 goto err;
2631
2632 /* round up to nearest 4K */
2633 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
2634 tx_ring->size = ALIGN(tx_ring->size, 4096);
2635
2636 tx_ring->desc = dma_alloc_coherent(dev,
2637 tx_ring->size,
2638 &tx_ring->dma,
2639 GFP_KERNEL);
2640 if (!tx_ring->desc)
2641 goto err;
2642
2643 tx_ring->next_to_use = 0;
2644 tx_ring->next_to_clean = 0;
2645
2646 return 0;
2647
2648 err:
2649 vfree(tx_ring->tx_buffer_info);
2650 tx_ring->tx_buffer_info = NULL;
2651 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
2652 return -ENOMEM;
2653 }
2654
2655 /**
2656 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2657 * (Descriptors) for all queues
2658 * @adapter: board private structure
2659 *
2660 * Return 0 on success, negative on failure
2661 **/
2662 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2663 {
2664 struct pci_dev *pdev = adapter->pdev;
2665 int i, err = 0;
2666
2667 for (i = 0; i < adapter->num_tx_queues; i++) {
2668 err = igb_setup_tx_resources(adapter->tx_ring[i]);
2669 if (err) {
2670 dev_err(&pdev->dev,
2671 "Allocation for Tx Queue %u failed\n", i);
2672 for (i--; i >= 0; i--)
2673 igb_free_tx_resources(adapter->tx_ring[i]);
2674 break;
2675 }
2676 }
2677
2678 return err;
2679 }
2680
2681 /**
2682 * igb_setup_tctl - configure the transmit control registers
2683 * @adapter: Board private structure
2684 **/
2685 void igb_setup_tctl(struct igb_adapter *adapter)
2686 {
2687 struct e1000_hw *hw = &adapter->hw;
2688 u32 tctl;
2689
2690 /* disable queue 0 which is enabled by default on 82575 and 82576 */
2691 wr32(E1000_TXDCTL(0), 0);
2692
2693 /* Program the Transmit Control Register */
2694 tctl = rd32(E1000_TCTL);
2695 tctl &= ~E1000_TCTL_CT;
2696 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2697 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2698
2699 igb_config_collision_dist(hw);
2700
2701 /* Enable transmits */
2702 tctl |= E1000_TCTL_EN;
2703
2704 wr32(E1000_TCTL, tctl);
2705 }
2706
2707 /**
2708 * igb_configure_tx_ring - Configure transmit ring after Reset
2709 * @adapter: board private structure
2710 * @ring: tx ring to configure
2711 *
2712 * Configure a transmit ring after a reset.
2713 **/
2714 void igb_configure_tx_ring(struct igb_adapter *adapter,
2715 struct igb_ring *ring)
2716 {
2717 struct e1000_hw *hw = &adapter->hw;
2718 u32 txdctl = 0;
2719 u64 tdba = ring->dma;
2720 int reg_idx = ring->reg_idx;
2721
2722 /* disable the queue */
2723 wr32(E1000_TXDCTL(reg_idx), 0);
2724 wrfl();
2725 mdelay(10);
2726
2727 wr32(E1000_TDLEN(reg_idx),
2728 ring->count * sizeof(union e1000_adv_tx_desc));
2729 wr32(E1000_TDBAL(reg_idx),
2730 tdba & 0x00000000ffffffffULL);
2731 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
2732
2733 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
2734 wr32(E1000_TDH(reg_idx), 0);
2735 writel(0, ring->tail);
2736
2737 txdctl |= IGB_TX_PTHRESH;
2738 txdctl |= IGB_TX_HTHRESH << 8;
2739 txdctl |= IGB_TX_WTHRESH << 16;
2740
2741 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
2742 wr32(E1000_TXDCTL(reg_idx), txdctl);
2743 }
2744
2745 /**
2746 * igb_configure_tx - Configure transmit Unit after Reset
2747 * @adapter: board private structure
2748 *
2749 * Configure the Tx unit of the MAC after a reset.
2750 **/
2751 static void igb_configure_tx(struct igb_adapter *adapter)
2752 {
2753 int i;
2754
2755 for (i = 0; i < adapter->num_tx_queues; i++)
2756 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
2757 }
2758
2759 /**
2760 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
2761 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2762 *
2763 * Returns 0 on success, negative on failure
2764 **/
2765 int igb_setup_rx_resources(struct igb_ring *rx_ring)
2766 {
2767 struct device *dev = rx_ring->dev;
2768 int size;
2769
2770 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
2771
2772 rx_ring->rx_buffer_info = vzalloc(size);
2773 if (!rx_ring->rx_buffer_info)
2774 goto err;
2775
2776
2777 /* Round up to nearest 4K */
2778 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
2779 rx_ring->size = ALIGN(rx_ring->size, 4096);
2780
2781 rx_ring->desc = dma_alloc_coherent(dev,
2782 rx_ring->size,
2783 &rx_ring->dma,
2784 GFP_KERNEL);
2785 if (!rx_ring->desc)
2786 goto err;
2787
2788 rx_ring->next_to_alloc = 0;
2789 rx_ring->next_to_clean = 0;
2790 rx_ring->next_to_use = 0;
2791
2792 return 0;
2793
2794 err:
2795 vfree(rx_ring->rx_buffer_info);
2796 rx_ring->rx_buffer_info = NULL;
2797 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
2798 return -ENOMEM;
2799 }
2800
2801 /**
2802 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
2803 * (Descriptors) for all queues
2804 * @adapter: board private structure
2805 *
2806 * Return 0 on success, negative on failure
2807 **/
2808 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
2809 {
2810 struct pci_dev *pdev = adapter->pdev;
2811 int i, err = 0;
2812
2813 for (i = 0; i < adapter->num_rx_queues; i++) {
2814 err = igb_setup_rx_resources(adapter->rx_ring[i]);
2815 if (err) {
2816 dev_err(&pdev->dev,
2817 "Allocation for Rx Queue %u failed\n", i);
2818 for (i--; i >= 0; i--)
2819 igb_free_rx_resources(adapter->rx_ring[i]);
2820 break;
2821 }
2822 }
2823
2824 return err;
2825 }
2826
2827 /**
2828 * igb_setup_mrqc - configure the multiple receive queue control registers
2829 * @adapter: Board private structure
2830 **/
2831 static void igb_setup_mrqc(struct igb_adapter *adapter)
2832 {
2833 struct e1000_hw *hw = &adapter->hw;
2834 u32 mrqc, rxcsum;
2835 u32 j, num_rx_queues, shift = 0;
2836 static const u32 rsskey[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
2837 0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
2838 0xA32DCB77, 0x0CF23080, 0x3BB7426A,
2839 0xFA01ACBE };
2840
2841 /* Fill out hash function seeds */
2842 for (j = 0; j < 10; j++)
2843 wr32(E1000_RSSRK(j), rsskey[j]);
2844
2845 num_rx_queues = adapter->rss_queues;
2846
2847 switch (hw->mac.type) {
2848 case e1000_82575:
2849 shift = 6;
2850 break;
2851 case e1000_82576:
2852 /* 82576 supports 2 RSS queues for SR-IOV */
2853 if (adapter->vfs_allocated_count) {
2854 shift = 3;
2855 num_rx_queues = 2;
2856 }
2857 break;
2858 default:
2859 break;
2860 }
2861
2862 /*
2863 * Populate the indirection table 4 entries at a time. To do this
2864 * we are generating the results for n and n+2 and then interleaving
2865 * those with the results with n+1 and n+3.
2866 */
2867 for (j = 0; j < 32; j++) {
2868 /* first pass generates n and n+2 */
2869 u32 base = ((j * 0x00040004) + 0x00020000) * num_rx_queues;
2870 u32 reta = (base & 0x07800780) >> (7 - shift);
2871
2872 /* second pass generates n+1 and n+3 */
2873 base += 0x00010001 * num_rx_queues;
2874 reta |= (base & 0x07800780) << (1 + shift);
2875
2876 wr32(E1000_RETA(j), reta);
2877 }
2878
2879 /*
2880 * Disable raw packet checksumming so that RSS hash is placed in
2881 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
2882 * offloads as they are enabled by default
2883 */
2884 rxcsum = rd32(E1000_RXCSUM);
2885 rxcsum |= E1000_RXCSUM_PCSD;
2886
2887 if (adapter->hw.mac.type >= e1000_82576)
2888 /* Enable Receive Checksum Offload for SCTP */
2889 rxcsum |= E1000_RXCSUM_CRCOFL;
2890
2891 /* Don't need to set TUOFL or IPOFL, they default to 1 */
2892 wr32(E1000_RXCSUM, rxcsum);
2893 /*
2894 * Generate RSS hash based on TCP port numbers and/or
2895 * IPv4/v6 src and dst addresses since UDP cannot be
2896 * hashed reliably due to IP fragmentation
2897 */
2898
2899 mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
2900 E1000_MRQC_RSS_FIELD_IPV4_TCP |
2901 E1000_MRQC_RSS_FIELD_IPV6 |
2902 E1000_MRQC_RSS_FIELD_IPV6_TCP |
2903 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
2904
2905 /* If VMDq is enabled then we set the appropriate mode for that, else
2906 * we default to RSS so that an RSS hash is calculated per packet even
2907 * if we are only using one queue */
2908 if (adapter->vfs_allocated_count) {
2909 if (hw->mac.type > e1000_82575) {
2910 /* Set the default pool for the PF's first queue */
2911 u32 vtctl = rd32(E1000_VT_CTL);
2912 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2913 E1000_VT_CTL_DISABLE_DEF_POOL);
2914 vtctl |= adapter->vfs_allocated_count <<
2915 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2916 wr32(E1000_VT_CTL, vtctl);
2917 }
2918 if (adapter->rss_queues > 1)
2919 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2920 else
2921 mrqc |= E1000_MRQC_ENABLE_VMDQ;
2922 } else {
2923 if (hw->mac.type != e1000_i211)
2924 mrqc |= E1000_MRQC_ENABLE_RSS_4Q;
2925 }
2926 igb_vmm_control(adapter);
2927
2928 wr32(E1000_MRQC, mrqc);
2929 }
2930
2931 /**
2932 * igb_setup_rctl - configure the receive control registers
2933 * @adapter: Board private structure
2934 **/
2935 void igb_setup_rctl(struct igb_adapter *adapter)
2936 {
2937 struct e1000_hw *hw = &adapter->hw;
2938 u32 rctl;
2939
2940 rctl = rd32(E1000_RCTL);
2941
2942 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2943 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2944
2945 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2946 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2947
2948 /*
2949 * enable stripping of CRC. It's unlikely this will break BMC
2950 * redirection as it did with e1000. Newer features require
2951 * that the HW strips the CRC.
2952 */
2953 rctl |= E1000_RCTL_SECRC;
2954
2955 /* disable store bad packets and clear size bits. */
2956 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2957
2958 /* enable LPE to prevent packets larger than max_frame_size */
2959 rctl |= E1000_RCTL_LPE;
2960
2961 /* disable queue 0 to prevent tail write w/o re-config */
2962 wr32(E1000_RXDCTL(0), 0);
2963
2964 /* Attention!!! For SR-IOV PF driver operations you must enable
2965 * queue drop for all VF and PF queues to prevent head of line blocking
2966 * if an un-trusted VF does not provide descriptors to hardware.
2967 */
2968 if (adapter->vfs_allocated_count) {
2969 /* set all queue drop enable bits */
2970 wr32(E1000_QDE, ALL_QUEUES);
2971 }
2972
2973 /* This is useful for sniffing bad packets. */
2974 if (adapter->netdev->features & NETIF_F_RXALL) {
2975 /* UPE and MPE will be handled by normal PROMISC logic
2976 * in e1000e_set_rx_mode */
2977 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
2978 E1000_RCTL_BAM | /* RX All Bcast Pkts */
2979 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
2980
2981 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
2982 E1000_RCTL_DPF | /* Allow filtered pause */
2983 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
2984 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
2985 * and that breaks VLANs.
2986 */
2987 }
2988
2989 wr32(E1000_RCTL, rctl);
2990 }
2991
2992 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
2993 int vfn)
2994 {
2995 struct e1000_hw *hw = &adapter->hw;
2996 u32 vmolr;
2997
2998 /* if it isn't the PF check to see if VFs are enabled and
2999 * increase the size to support vlan tags */
3000 if (vfn < adapter->vfs_allocated_count &&
3001 adapter->vf_data[vfn].vlans_enabled)
3002 size += VLAN_TAG_SIZE;
3003
3004 vmolr = rd32(E1000_VMOLR(vfn));
3005 vmolr &= ~E1000_VMOLR_RLPML_MASK;
3006 vmolr |= size | E1000_VMOLR_LPE;
3007 wr32(E1000_VMOLR(vfn), vmolr);
3008
3009 return 0;
3010 }
3011
3012 /**
3013 * igb_rlpml_set - set maximum receive packet size
3014 * @adapter: board private structure
3015 *
3016 * Configure maximum receivable packet size.
3017 **/
3018 static void igb_rlpml_set(struct igb_adapter *adapter)
3019 {
3020 u32 max_frame_size = adapter->max_frame_size;
3021 struct e1000_hw *hw = &adapter->hw;
3022 u16 pf_id = adapter->vfs_allocated_count;
3023
3024 if (pf_id) {
3025 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3026 /*
3027 * If we're in VMDQ or SR-IOV mode, then set global RLPML
3028 * to our max jumbo frame size, in case we need to enable
3029 * jumbo frames on one of the rings later.
3030 * This will not pass over-length frames into the default
3031 * queue because it's gated by the VMOLR.RLPML.
3032 */
3033 max_frame_size = MAX_JUMBO_FRAME_SIZE;
3034 }
3035
3036 wr32(E1000_RLPML, max_frame_size);
3037 }
3038
3039 static inline void igb_set_vmolr(struct igb_adapter *adapter,
3040 int vfn, bool aupe)
3041 {
3042 struct e1000_hw *hw = &adapter->hw;
3043 u32 vmolr;
3044
3045 /*
3046 * This register exists only on 82576 and newer so if we are older then
3047 * we should exit and do nothing
3048 */
3049 if (hw->mac.type < e1000_82576)
3050 return;
3051
3052 vmolr = rd32(E1000_VMOLR(vfn));
3053 vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
3054 if (aupe)
3055 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3056 else
3057 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3058
3059 /* clear all bits that might not be set */
3060 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3061
3062 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3063 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3064 /*
3065 * for VMDq only allow the VFs and pool 0 to accept broadcast and
3066 * multicast packets
3067 */
3068 if (vfn <= adapter->vfs_allocated_count)
3069 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3070
3071 wr32(E1000_VMOLR(vfn), vmolr);
3072 }
3073
3074 /**
3075 * igb_configure_rx_ring - Configure a receive ring after Reset
3076 * @adapter: board private structure
3077 * @ring: receive ring to be configured
3078 *
3079 * Configure the Rx unit of the MAC after a reset.
3080 **/
3081 void igb_configure_rx_ring(struct igb_adapter *adapter,
3082 struct igb_ring *ring)
3083 {
3084 struct e1000_hw *hw = &adapter->hw;
3085 u64 rdba = ring->dma;
3086 int reg_idx = ring->reg_idx;
3087 u32 srrctl = 0, rxdctl = 0;
3088
3089 /* disable the queue */
3090 wr32(E1000_RXDCTL(reg_idx), 0);
3091
3092 /* Set DMA base address registers */
3093 wr32(E1000_RDBAL(reg_idx),
3094 rdba & 0x00000000ffffffffULL);
3095 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3096 wr32(E1000_RDLEN(reg_idx),
3097 ring->count * sizeof(union e1000_adv_rx_desc));
3098
3099 /* initialize head and tail */
3100 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3101 wr32(E1000_RDH(reg_idx), 0);
3102 writel(0, ring->tail);
3103
3104 /* set descriptor configuration */
3105 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3106 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3107 srrctl |= IGB_RXBUFFER_16384 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3108 #else
3109 srrctl |= (PAGE_SIZE / 2) >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3110 #endif
3111 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3112 #ifdef CONFIG_IGB_PTP
3113 if (hw->mac.type >= e1000_82580)
3114 srrctl |= E1000_SRRCTL_TIMESTAMP;
3115 #endif /* CONFIG_IGB_PTP */
3116 /* Only set Drop Enable if we are supporting multiple queues */
3117 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3118 srrctl |= E1000_SRRCTL_DROP_EN;
3119
3120 wr32(E1000_SRRCTL(reg_idx), srrctl);
3121
3122 /* set filtering for VMDQ pools */
3123 igb_set_vmolr(adapter, reg_idx & 0x7, true);
3124
3125 rxdctl |= IGB_RX_PTHRESH;
3126 rxdctl |= IGB_RX_HTHRESH << 8;
3127 rxdctl |= IGB_RX_WTHRESH << 16;
3128
3129 /* enable receive descriptor fetching */
3130 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3131 wr32(E1000_RXDCTL(reg_idx), rxdctl);
3132 }
3133
3134 /**
3135 * igb_configure_rx - Configure receive Unit after Reset
3136 * @adapter: board private structure
3137 *
3138 * Configure the Rx unit of the MAC after a reset.
3139 **/
3140 static void igb_configure_rx(struct igb_adapter *adapter)
3141 {
3142 int i;
3143
3144 /* set UTA to appropriate mode */
3145 igb_set_uta(adapter);
3146
3147 /* set the correct pool for the PF default MAC address in entry 0 */
3148 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3149 adapter->vfs_allocated_count);
3150
3151 /* Setup the HW Rx Head and Tail Descriptor Pointers and
3152 * the Base and Length of the Rx Descriptor Ring */
3153 for (i = 0; i < adapter->num_rx_queues; i++)
3154 igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
3155 }
3156
3157 /**
3158 * igb_free_tx_resources - Free Tx Resources per Queue
3159 * @tx_ring: Tx descriptor ring for a specific queue
3160 *
3161 * Free all transmit software resources
3162 **/
3163 void igb_free_tx_resources(struct igb_ring *tx_ring)
3164 {
3165 igb_clean_tx_ring(tx_ring);
3166
3167 vfree(tx_ring->tx_buffer_info);
3168 tx_ring->tx_buffer_info = NULL;
3169
3170 /* if not set, then don't free */
3171 if (!tx_ring->desc)
3172 return;
3173
3174 dma_free_coherent(tx_ring->dev, tx_ring->size,
3175 tx_ring->desc, tx_ring->dma);
3176
3177 tx_ring->desc = NULL;
3178 }
3179
3180 /**
3181 * igb_free_all_tx_resources - Free Tx Resources for All Queues
3182 * @adapter: board private structure
3183 *
3184 * Free all transmit software resources
3185 **/
3186 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3187 {
3188 int i;
3189
3190 for (i = 0; i < adapter->num_tx_queues; i++)
3191 igb_free_tx_resources(adapter->tx_ring[i]);
3192 }
3193
3194 void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3195 struct igb_tx_buffer *tx_buffer)
3196 {
3197 if (tx_buffer->skb) {
3198 dev_kfree_skb_any(tx_buffer->skb);
3199 if (dma_unmap_len(tx_buffer, len))
3200 dma_unmap_single(ring->dev,
3201 dma_unmap_addr(tx_buffer, dma),
3202 dma_unmap_len(tx_buffer, len),
3203 DMA_TO_DEVICE);
3204 } else if (dma_unmap_len(tx_buffer, len)) {
3205 dma_unmap_page(ring->dev,
3206 dma_unmap_addr(tx_buffer, dma),
3207 dma_unmap_len(tx_buffer, len),
3208 DMA_TO_DEVICE);
3209 }
3210 tx_buffer->next_to_watch = NULL;
3211 tx_buffer->skb = NULL;
3212 dma_unmap_len_set(tx_buffer, len, 0);
3213 /* buffer_info must be completely set up in the transmit path */
3214 }
3215
3216 /**
3217 * igb_clean_tx_ring - Free Tx Buffers
3218 * @tx_ring: ring to be cleaned
3219 **/
3220 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3221 {
3222 struct igb_tx_buffer *buffer_info;
3223 unsigned long size;
3224 u16 i;
3225
3226 if (!tx_ring->tx_buffer_info)
3227 return;
3228 /* Free all the Tx ring sk_buffs */
3229
3230 for (i = 0; i < tx_ring->count; i++) {
3231 buffer_info = &tx_ring->tx_buffer_info[i];
3232 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3233 }
3234
3235 netdev_tx_reset_queue(txring_txq(tx_ring));
3236
3237 size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3238 memset(tx_ring->tx_buffer_info, 0, size);
3239
3240 /* Zero out the descriptor ring */
3241 memset(tx_ring->desc, 0, tx_ring->size);
3242
3243 tx_ring->next_to_use = 0;
3244 tx_ring->next_to_clean = 0;
3245 }
3246
3247 /**
3248 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3249 * @adapter: board private structure
3250 **/
3251 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3252 {
3253 int i;
3254
3255 for (i = 0; i < adapter->num_tx_queues; i++)
3256 igb_clean_tx_ring(adapter->tx_ring[i]);
3257 }
3258
3259 /**
3260 * igb_free_rx_resources - Free Rx Resources
3261 * @rx_ring: ring to clean the resources from
3262 *
3263 * Free all receive software resources
3264 **/
3265 void igb_free_rx_resources(struct igb_ring *rx_ring)
3266 {
3267 igb_clean_rx_ring(rx_ring);
3268
3269 vfree(rx_ring->rx_buffer_info);
3270 rx_ring->rx_buffer_info = NULL;
3271
3272 /* if not set, then don't free */
3273 if (!rx_ring->desc)
3274 return;
3275
3276 dma_free_coherent(rx_ring->dev, rx_ring->size,
3277 rx_ring->desc, rx_ring->dma);
3278
3279 rx_ring->desc = NULL;
3280 }
3281
3282 /**
3283 * igb_free_all_rx_resources - Free Rx Resources for All Queues
3284 * @adapter: board private structure
3285 *
3286 * Free all receive software resources
3287 **/
3288 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3289 {
3290 int i;
3291
3292 for (i = 0; i < adapter->num_rx_queues; i++)
3293 igb_free_rx_resources(adapter->rx_ring[i]);
3294 }
3295
3296 /**
3297 * igb_clean_rx_ring - Free Rx Buffers per Queue
3298 * @rx_ring: ring to free buffers from
3299 **/
3300 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3301 {
3302 unsigned long size;
3303 u16 i;
3304
3305 if (rx_ring->skb)
3306 dev_kfree_skb(rx_ring->skb);
3307 rx_ring->skb = NULL;
3308
3309 if (!rx_ring->rx_buffer_info)
3310 return;
3311
3312 /* Free all the Rx ring sk_buffs */
3313 for (i = 0; i < rx_ring->count; i++) {
3314 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3315
3316 if (!buffer_info->page)
3317 continue;
3318
3319 dma_unmap_page(rx_ring->dev,
3320 buffer_info->dma,
3321 PAGE_SIZE,
3322 DMA_FROM_DEVICE);
3323 __free_page(buffer_info->page);
3324
3325 buffer_info->page = NULL;
3326 }
3327
3328 size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3329 memset(rx_ring->rx_buffer_info, 0, size);
3330
3331 /* Zero out the descriptor ring */
3332 memset(rx_ring->desc, 0, rx_ring->size);
3333
3334 rx_ring->next_to_alloc = 0;
3335 rx_ring->next_to_clean = 0;
3336 rx_ring->next_to_use = 0;
3337 }
3338
3339 /**
3340 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3341 * @adapter: board private structure
3342 **/
3343 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3344 {
3345 int i;
3346
3347 for (i = 0; i < adapter->num_rx_queues; i++)
3348 igb_clean_rx_ring(adapter->rx_ring[i]);
3349 }
3350
3351 /**
3352 * igb_set_mac - Change the Ethernet Address of the NIC
3353 * @netdev: network interface device structure
3354 * @p: pointer to an address structure
3355 *
3356 * Returns 0 on success, negative on failure
3357 **/
3358 static int igb_set_mac(struct net_device *netdev, void *p)
3359 {
3360 struct igb_adapter *adapter = netdev_priv(netdev);
3361 struct e1000_hw *hw = &adapter->hw;
3362 struct sockaddr *addr = p;
3363
3364 if (!is_valid_ether_addr(addr->sa_data))
3365 return -EADDRNOTAVAIL;
3366
3367 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3368 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3369
3370 /* set the correct pool for the new PF MAC address in entry 0 */
3371 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3372 adapter->vfs_allocated_count);
3373
3374 return 0;
3375 }
3376
3377 /**
3378 * igb_write_mc_addr_list - write multicast addresses to MTA
3379 * @netdev: network interface device structure
3380 *
3381 * Writes multicast address list to the MTA hash table.
3382 * Returns: -ENOMEM on failure
3383 * 0 on no addresses written
3384 * X on writing X addresses to MTA
3385 **/
3386 static int igb_write_mc_addr_list(struct net_device *netdev)
3387 {
3388 struct igb_adapter *adapter = netdev_priv(netdev);
3389 struct e1000_hw *hw = &adapter->hw;
3390 struct netdev_hw_addr *ha;
3391 u8 *mta_list;
3392 int i;
3393
3394 if (netdev_mc_empty(netdev)) {
3395 /* nothing to program, so clear mc list */
3396 igb_update_mc_addr_list(hw, NULL, 0);
3397 igb_restore_vf_multicasts(adapter);
3398 return 0;
3399 }
3400
3401 mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3402 if (!mta_list)
3403 return -ENOMEM;
3404
3405 /* The shared function expects a packed array of only addresses. */
3406 i = 0;
3407 netdev_for_each_mc_addr(ha, netdev)
3408 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3409
3410 igb_update_mc_addr_list(hw, mta_list, i);
3411 kfree(mta_list);
3412
3413 return netdev_mc_count(netdev);
3414 }
3415
3416 /**
3417 * igb_write_uc_addr_list - write unicast addresses to RAR table
3418 * @netdev: network interface device structure
3419 *
3420 * Writes unicast address list to the RAR table.
3421 * Returns: -ENOMEM on failure/insufficient address space
3422 * 0 on no addresses written
3423 * X on writing X addresses to the RAR table
3424 **/
3425 static int igb_write_uc_addr_list(struct net_device *netdev)
3426 {
3427 struct igb_adapter *adapter = netdev_priv(netdev);
3428 struct e1000_hw *hw = &adapter->hw;
3429 unsigned int vfn = adapter->vfs_allocated_count;
3430 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3431 int count = 0;
3432
3433 /* return ENOMEM indicating insufficient memory for addresses */
3434 if (netdev_uc_count(netdev) > rar_entries)
3435 return -ENOMEM;
3436
3437 if (!netdev_uc_empty(netdev) && rar_entries) {
3438 struct netdev_hw_addr *ha;
3439
3440 netdev_for_each_uc_addr(ha, netdev) {
3441 if (!rar_entries)
3442 break;
3443 igb_rar_set_qsel(adapter, ha->addr,
3444 rar_entries--,
3445 vfn);
3446 count++;
3447 }
3448 }
3449 /* write the addresses in reverse order to avoid write combining */
3450 for (; rar_entries > 0 ; rar_entries--) {
3451 wr32(E1000_RAH(rar_entries), 0);
3452 wr32(E1000_RAL(rar_entries), 0);
3453 }
3454 wrfl();
3455
3456 return count;
3457 }
3458
3459 /**
3460 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3461 * @netdev: network interface device structure
3462 *
3463 * The set_rx_mode entry point is called whenever the unicast or multicast
3464 * address lists or the network interface flags are updated. This routine is
3465 * responsible for configuring the hardware for proper unicast, multicast,
3466 * promiscuous mode, and all-multi behavior.
3467 **/
3468 static void igb_set_rx_mode(struct net_device *netdev)
3469 {
3470 struct igb_adapter *adapter = netdev_priv(netdev);
3471 struct e1000_hw *hw = &adapter->hw;
3472 unsigned int vfn = adapter->vfs_allocated_count;
3473 u32 rctl, vmolr = 0;
3474 int count;
3475
3476 /* Check for Promiscuous and All Multicast modes */
3477 rctl = rd32(E1000_RCTL);
3478
3479 /* clear the effected bits */
3480 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
3481
3482 if (netdev->flags & IFF_PROMISC) {
3483 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3484 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
3485 } else {
3486 if (netdev->flags & IFF_ALLMULTI) {
3487 rctl |= E1000_RCTL_MPE;
3488 vmolr |= E1000_VMOLR_MPME;
3489 } else {
3490 /*
3491 * Write addresses to the MTA, if the attempt fails
3492 * then we should just turn on promiscuous mode so
3493 * that we can at least receive multicast traffic
3494 */
3495 count = igb_write_mc_addr_list(netdev);
3496 if (count < 0) {
3497 rctl |= E1000_RCTL_MPE;
3498 vmolr |= E1000_VMOLR_MPME;
3499 } else if (count) {
3500 vmolr |= E1000_VMOLR_ROMPE;
3501 }
3502 }
3503 /*
3504 * Write addresses to available RAR registers, if there is not
3505 * sufficient space to store all the addresses then enable
3506 * unicast promiscuous mode
3507 */
3508 count = igb_write_uc_addr_list(netdev);
3509 if (count < 0) {
3510 rctl |= E1000_RCTL_UPE;
3511 vmolr |= E1000_VMOLR_ROPE;
3512 }
3513 rctl |= E1000_RCTL_VFE;
3514 }
3515 wr32(E1000_RCTL, rctl);
3516
3517 /*
3518 * In order to support SR-IOV and eventually VMDq it is necessary to set
3519 * the VMOLR to enable the appropriate modes. Without this workaround
3520 * we will have issues with VLAN tag stripping not being done for frames
3521 * that are only arriving because we are the default pool
3522 */
3523 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
3524 return;
3525
3526 vmolr |= rd32(E1000_VMOLR(vfn)) &
3527 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3528 wr32(E1000_VMOLR(vfn), vmolr);
3529 igb_restore_vf_multicasts(adapter);
3530 }
3531
3532 static void igb_check_wvbr(struct igb_adapter *adapter)
3533 {
3534 struct e1000_hw *hw = &adapter->hw;
3535 u32 wvbr = 0;
3536
3537 switch (hw->mac.type) {
3538 case e1000_82576:
3539 case e1000_i350:
3540 if (!(wvbr = rd32(E1000_WVBR)))
3541 return;
3542 break;
3543 default:
3544 break;
3545 }
3546
3547 adapter->wvbr |= wvbr;
3548 }
3549
3550 #define IGB_STAGGERED_QUEUE_OFFSET 8
3551
3552 static void igb_spoof_check(struct igb_adapter *adapter)
3553 {
3554 int j;
3555
3556 if (!adapter->wvbr)
3557 return;
3558
3559 for(j = 0; j < adapter->vfs_allocated_count; j++) {
3560 if (adapter->wvbr & (1 << j) ||
3561 adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
3562 dev_warn(&adapter->pdev->dev,
3563 "Spoof event(s) detected on VF %d\n", j);
3564 adapter->wvbr &=
3565 ~((1 << j) |
3566 (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
3567 }
3568 }
3569 }
3570
3571 /* Need to wait a few seconds after link up to get diagnostic information from
3572 * the phy */
3573 static void igb_update_phy_info(unsigned long data)
3574 {
3575 struct igb_adapter *adapter = (struct igb_adapter *) data;
3576 igb_get_phy_info(&adapter->hw);
3577 }
3578
3579 /**
3580 * igb_has_link - check shared code for link and determine up/down
3581 * @adapter: pointer to driver private info
3582 **/
3583 bool igb_has_link(struct igb_adapter *adapter)
3584 {
3585 struct e1000_hw *hw = &adapter->hw;
3586 bool link_active = false;
3587 s32 ret_val = 0;
3588
3589 /* get_link_status is set on LSC (link status) interrupt or
3590 * rx sequence error interrupt. get_link_status will stay
3591 * false until the e1000_check_for_link establishes link
3592 * for copper adapters ONLY
3593 */
3594 switch (hw->phy.media_type) {
3595 case e1000_media_type_copper:
3596 if (hw->mac.get_link_status) {
3597 ret_val = hw->mac.ops.check_for_link(hw);
3598 link_active = !hw->mac.get_link_status;
3599 } else {
3600 link_active = true;
3601 }
3602 break;
3603 case e1000_media_type_internal_serdes:
3604 ret_val = hw->mac.ops.check_for_link(hw);
3605 link_active = hw->mac.serdes_has_link;
3606 break;
3607 default:
3608 case e1000_media_type_unknown:
3609 break;
3610 }
3611
3612 return link_active;
3613 }
3614
3615 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
3616 {
3617 bool ret = false;
3618 u32 ctrl_ext, thstat;
3619
3620 /* check for thermal sensor event on i350 copper only */
3621 if (hw->mac.type == e1000_i350) {
3622 thstat = rd32(E1000_THSTAT);
3623 ctrl_ext = rd32(E1000_CTRL_EXT);
3624
3625 if ((hw->phy.media_type == e1000_media_type_copper) &&
3626 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3627 ret = !!(thstat & event);
3628 }
3629 }
3630
3631 return ret;
3632 }
3633
3634 /**
3635 * igb_watchdog - Timer Call-back
3636 * @data: pointer to adapter cast into an unsigned long
3637 **/
3638 static void igb_watchdog(unsigned long data)
3639 {
3640 struct igb_adapter *adapter = (struct igb_adapter *)data;
3641 /* Do the rest outside of interrupt context */
3642 schedule_work(&adapter->watchdog_task);
3643 }
3644
3645 static void igb_watchdog_task(struct work_struct *work)
3646 {
3647 struct igb_adapter *adapter = container_of(work,
3648 struct igb_adapter,
3649 watchdog_task);
3650 struct e1000_hw *hw = &adapter->hw;
3651 struct net_device *netdev = adapter->netdev;
3652 u32 link;
3653 int i;
3654
3655 link = igb_has_link(adapter);
3656 if (link) {
3657 /* Cancel scheduled suspend requests. */
3658 pm_runtime_resume(netdev->dev.parent);
3659
3660 if (!netif_carrier_ok(netdev)) {
3661 u32 ctrl;
3662 hw->mac.ops.get_speed_and_duplex(hw,
3663 &adapter->link_speed,
3664 &adapter->link_duplex);
3665
3666 ctrl = rd32(E1000_CTRL);
3667 /* Links status message must follow this format */
3668 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s "
3669 "Duplex, Flow Control: %s\n",
3670 netdev->name,
3671 adapter->link_speed,
3672 adapter->link_duplex == FULL_DUPLEX ?
3673 "Full" : "Half",
3674 (ctrl & E1000_CTRL_TFCE) &&
3675 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
3676 (ctrl & E1000_CTRL_RFCE) ? "RX" :
3677 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None");
3678
3679 /* check for thermal sensor event */
3680 if (igb_thermal_sensor_event(hw,
3681 E1000_THSTAT_LINK_THROTTLE)) {
3682 netdev_info(netdev, "The network adapter link "
3683 "speed was downshifted because it "
3684 "overheated\n");
3685 }
3686
3687 /* adjust timeout factor according to speed/duplex */
3688 adapter->tx_timeout_factor = 1;
3689 switch (adapter->link_speed) {
3690 case SPEED_10:
3691 adapter->tx_timeout_factor = 14;
3692 break;
3693 case SPEED_100:
3694 /* maybe add some timeout factor ? */
3695 break;
3696 }
3697
3698 netif_carrier_on(netdev);
3699
3700 igb_ping_all_vfs(adapter);
3701 igb_check_vf_rate_limit(adapter);
3702
3703 /* link state has changed, schedule phy info update */
3704 if (!test_bit(__IGB_DOWN, &adapter->state))
3705 mod_timer(&adapter->phy_info_timer,
3706 round_jiffies(jiffies + 2 * HZ));
3707 }
3708 } else {
3709 if (netif_carrier_ok(netdev)) {
3710 adapter->link_speed = 0;
3711 adapter->link_duplex = 0;
3712
3713 /* check for thermal sensor event */
3714 if (igb_thermal_sensor_event(hw,
3715 E1000_THSTAT_PWR_DOWN)) {
3716 netdev_err(netdev, "The network adapter was "
3717 "stopped because it overheated\n");
3718 }
3719
3720 /* Links status message must follow this format */
3721 printk(KERN_INFO "igb: %s NIC Link is Down\n",
3722 netdev->name);
3723 netif_carrier_off(netdev);
3724
3725 igb_ping_all_vfs(adapter);
3726
3727 /* link state has changed, schedule phy info update */
3728 if (!test_bit(__IGB_DOWN, &adapter->state))
3729 mod_timer(&adapter->phy_info_timer,
3730 round_jiffies(jiffies + 2 * HZ));
3731
3732 pm_schedule_suspend(netdev->dev.parent,
3733 MSEC_PER_SEC * 5);
3734 }
3735 }
3736
3737 spin_lock(&adapter->stats64_lock);
3738 igb_update_stats(adapter, &adapter->stats64);
3739 spin_unlock(&adapter->stats64_lock);
3740
3741 for (i = 0; i < adapter->num_tx_queues; i++) {
3742 struct igb_ring *tx_ring = adapter->tx_ring[i];
3743 if (!netif_carrier_ok(netdev)) {
3744 /* We've lost link, so the controller stops DMA,
3745 * but we've got queued Tx work that's never going
3746 * to get done, so reset controller to flush Tx.
3747 * (Do the reset outside of interrupt context). */
3748 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
3749 adapter->tx_timeout_count++;
3750 schedule_work(&adapter->reset_task);
3751 /* return immediately since reset is imminent */
3752 return;
3753 }
3754 }
3755
3756 /* Force detection of hung controller every watchdog period */
3757 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
3758 }
3759
3760 /* Cause software interrupt to ensure rx ring is cleaned */
3761 if (adapter->msix_entries) {
3762 u32 eics = 0;
3763 for (i = 0; i < adapter->num_q_vectors; i++)
3764 eics |= adapter->q_vector[i]->eims_value;
3765 wr32(E1000_EICS, eics);
3766 } else {
3767 wr32(E1000_ICS, E1000_ICS_RXDMT0);
3768 }
3769
3770 igb_spoof_check(adapter);
3771
3772 /* Reset the timer */
3773 if (!test_bit(__IGB_DOWN, &adapter->state))
3774 mod_timer(&adapter->watchdog_timer,
3775 round_jiffies(jiffies + 2 * HZ));
3776 }
3777
3778 enum latency_range {
3779 lowest_latency = 0,
3780 low_latency = 1,
3781 bulk_latency = 2,
3782 latency_invalid = 255
3783 };
3784
3785 /**
3786 * igb_update_ring_itr - update the dynamic ITR value based on packet size
3787 *
3788 * Stores a new ITR value based on strictly on packet size. This
3789 * algorithm is less sophisticated than that used in igb_update_itr,
3790 * due to the difficulty of synchronizing statistics across multiple
3791 * receive rings. The divisors and thresholds used by this function
3792 * were determined based on theoretical maximum wire speed and testing
3793 * data, in order to minimize response time while increasing bulk
3794 * throughput.
3795 * This functionality is controlled by the InterruptThrottleRate module
3796 * parameter (see igb_param.c)
3797 * NOTE: This function is called only when operating in a multiqueue
3798 * receive environment.
3799 * @q_vector: pointer to q_vector
3800 **/
3801 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
3802 {
3803 int new_val = q_vector->itr_val;
3804 int avg_wire_size = 0;
3805 struct igb_adapter *adapter = q_vector->adapter;
3806 unsigned int packets;
3807
3808 /* For non-gigabit speeds, just fix the interrupt rate at 4000
3809 * ints/sec - ITR timer value of 120 ticks.
3810 */
3811 if (adapter->link_speed != SPEED_1000) {
3812 new_val = IGB_4K_ITR;
3813 goto set_itr_val;
3814 }
3815
3816 packets = q_vector->rx.total_packets;
3817 if (packets)
3818 avg_wire_size = q_vector->rx.total_bytes / packets;
3819
3820 packets = q_vector->tx.total_packets;
3821 if (packets)
3822 avg_wire_size = max_t(u32, avg_wire_size,
3823 q_vector->tx.total_bytes / packets);
3824
3825 /* if avg_wire_size isn't set no work was done */
3826 if (!avg_wire_size)
3827 goto clear_counts;
3828
3829 /* Add 24 bytes to size to account for CRC, preamble, and gap */
3830 avg_wire_size += 24;
3831
3832 /* Don't starve jumbo frames */
3833 avg_wire_size = min(avg_wire_size, 3000);
3834
3835 /* Give a little boost to mid-size frames */
3836 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
3837 new_val = avg_wire_size / 3;
3838 else
3839 new_val = avg_wire_size / 2;
3840
3841 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3842 if (new_val < IGB_20K_ITR &&
3843 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
3844 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
3845 new_val = IGB_20K_ITR;
3846
3847 set_itr_val:
3848 if (new_val != q_vector->itr_val) {
3849 q_vector->itr_val = new_val;
3850 q_vector->set_itr = 1;
3851 }
3852 clear_counts:
3853 q_vector->rx.total_bytes = 0;
3854 q_vector->rx.total_packets = 0;
3855 q_vector->tx.total_bytes = 0;
3856 q_vector->tx.total_packets = 0;
3857 }
3858
3859 /**
3860 * igb_update_itr - update the dynamic ITR value based on statistics
3861 * Stores a new ITR value based on packets and byte
3862 * counts during the last interrupt. The advantage of per interrupt
3863 * computation is faster updates and more accurate ITR for the current
3864 * traffic pattern. Constants in this function were computed
3865 * based on theoretical maximum wire speed and thresholds were set based
3866 * on testing data as well as attempting to minimize response time
3867 * while increasing bulk throughput.
3868 * this functionality is controlled by the InterruptThrottleRate module
3869 * parameter (see igb_param.c)
3870 * NOTE: These calculations are only valid when operating in a single-
3871 * queue environment.
3872 * @q_vector: pointer to q_vector
3873 * @ring_container: ring info to update the itr for
3874 **/
3875 static void igb_update_itr(struct igb_q_vector *q_vector,
3876 struct igb_ring_container *ring_container)
3877 {
3878 unsigned int packets = ring_container->total_packets;
3879 unsigned int bytes = ring_container->total_bytes;
3880 u8 itrval = ring_container->itr;
3881
3882 /* no packets, exit with status unchanged */
3883 if (packets == 0)
3884 return;
3885
3886 switch (itrval) {
3887 case lowest_latency:
3888 /* handle TSO and jumbo frames */
3889 if (bytes/packets > 8000)
3890 itrval = bulk_latency;
3891 else if ((packets < 5) && (bytes > 512))
3892 itrval = low_latency;
3893 break;
3894 case low_latency: /* 50 usec aka 20000 ints/s */
3895 if (bytes > 10000) {
3896 /* this if handles the TSO accounting */
3897 if (bytes/packets > 8000) {
3898 itrval = bulk_latency;
3899 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
3900 itrval = bulk_latency;
3901 } else if ((packets > 35)) {
3902 itrval = lowest_latency;
3903 }
3904 } else if (bytes/packets > 2000) {
3905 itrval = bulk_latency;
3906 } else if (packets <= 2 && bytes < 512) {
3907 itrval = lowest_latency;
3908 }
3909 break;
3910 case bulk_latency: /* 250 usec aka 4000 ints/s */
3911 if (bytes > 25000) {
3912 if (packets > 35)
3913 itrval = low_latency;
3914 } else if (bytes < 1500) {
3915 itrval = low_latency;
3916 }
3917 break;
3918 }
3919
3920 /* clear work counters since we have the values we need */
3921 ring_container->total_bytes = 0;
3922 ring_container->total_packets = 0;
3923
3924 /* write updated itr to ring container */
3925 ring_container->itr = itrval;
3926 }
3927
3928 static void igb_set_itr(struct igb_q_vector *q_vector)
3929 {
3930 struct igb_adapter *adapter = q_vector->adapter;
3931 u32 new_itr = q_vector->itr_val;
3932 u8 current_itr = 0;
3933
3934 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
3935 if (adapter->link_speed != SPEED_1000) {
3936 current_itr = 0;
3937 new_itr = IGB_4K_ITR;
3938 goto set_itr_now;
3939 }
3940
3941 igb_update_itr(q_vector, &q_vector->tx);
3942 igb_update_itr(q_vector, &q_vector->rx);
3943
3944 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
3945
3946 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3947 if (current_itr == lowest_latency &&
3948 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
3949 (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
3950 current_itr = low_latency;
3951
3952 switch (current_itr) {
3953 /* counts and packets in update_itr are dependent on these numbers */
3954 case lowest_latency:
3955 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
3956 break;
3957 case low_latency:
3958 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
3959 break;
3960 case bulk_latency:
3961 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
3962 break;
3963 default:
3964 break;
3965 }
3966
3967 set_itr_now:
3968 if (new_itr != q_vector->itr_val) {
3969 /* this attempts to bias the interrupt rate towards Bulk
3970 * by adding intermediate steps when interrupt rate is
3971 * increasing */
3972 new_itr = new_itr > q_vector->itr_val ?
3973 max((new_itr * q_vector->itr_val) /
3974 (new_itr + (q_vector->itr_val >> 2)),
3975 new_itr) :
3976 new_itr;
3977 /* Don't write the value here; it resets the adapter's
3978 * internal timer, and causes us to delay far longer than
3979 * we should between interrupts. Instead, we write the ITR
3980 * value at the beginning of the next interrupt so the timing
3981 * ends up being correct.
3982 */
3983 q_vector->itr_val = new_itr;
3984 q_vector->set_itr = 1;
3985 }
3986 }
3987
3988 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
3989 u32 type_tucmd, u32 mss_l4len_idx)
3990 {
3991 struct e1000_adv_tx_context_desc *context_desc;
3992 u16 i = tx_ring->next_to_use;
3993
3994 context_desc = IGB_TX_CTXTDESC(tx_ring, i);
3995
3996 i++;
3997 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3998
3999 /* set bits to identify this as an advanced context descriptor */
4000 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4001
4002 /* For 82575, context index must be unique per ring. */
4003 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4004 mss_l4len_idx |= tx_ring->reg_idx << 4;
4005
4006 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4007 context_desc->seqnum_seed = 0;
4008 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
4009 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4010 }
4011
4012 static int igb_tso(struct igb_ring *tx_ring,
4013 struct igb_tx_buffer *first,
4014 u8 *hdr_len)
4015 {
4016 struct sk_buff *skb = first->skb;
4017 u32 vlan_macip_lens, type_tucmd;
4018 u32 mss_l4len_idx, l4len;
4019
4020 if (!skb_is_gso(skb))
4021 return 0;
4022
4023 if (skb_header_cloned(skb)) {
4024 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4025 if (err)
4026 return err;
4027 }
4028
4029 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4030 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4031
4032 if (first->protocol == __constant_htons(ETH_P_IP)) {
4033 struct iphdr *iph = ip_hdr(skb);
4034 iph->tot_len = 0;
4035 iph->check = 0;
4036 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4037 iph->daddr, 0,
4038 IPPROTO_TCP,
4039 0);
4040 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4041 first->tx_flags |= IGB_TX_FLAGS_TSO |
4042 IGB_TX_FLAGS_CSUM |
4043 IGB_TX_FLAGS_IPV4;
4044 } else if (skb_is_gso_v6(skb)) {
4045 ipv6_hdr(skb)->payload_len = 0;
4046 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4047 &ipv6_hdr(skb)->daddr,
4048 0, IPPROTO_TCP, 0);
4049 first->tx_flags |= IGB_TX_FLAGS_TSO |
4050 IGB_TX_FLAGS_CSUM;
4051 }
4052
4053 /* compute header lengths */
4054 l4len = tcp_hdrlen(skb);
4055 *hdr_len = skb_transport_offset(skb) + l4len;
4056
4057 /* update gso size and bytecount with header size */
4058 first->gso_segs = skb_shinfo(skb)->gso_segs;
4059 first->bytecount += (first->gso_segs - 1) * *hdr_len;
4060
4061 /* MSS L4LEN IDX */
4062 mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4063 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4064
4065 /* VLAN MACLEN IPLEN */
4066 vlan_macip_lens = skb_network_header_len(skb);
4067 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4068 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4069
4070 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4071
4072 return 1;
4073 }
4074
4075 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4076 {
4077 struct sk_buff *skb = first->skb;
4078 u32 vlan_macip_lens = 0;
4079 u32 mss_l4len_idx = 0;
4080 u32 type_tucmd = 0;
4081
4082 if (skb->ip_summed != CHECKSUM_PARTIAL) {
4083 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4084 return;
4085 } else {
4086 u8 l4_hdr = 0;
4087 switch (first->protocol) {
4088 case __constant_htons(ETH_P_IP):
4089 vlan_macip_lens |= skb_network_header_len(skb);
4090 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4091 l4_hdr = ip_hdr(skb)->protocol;
4092 break;
4093 case __constant_htons(ETH_P_IPV6):
4094 vlan_macip_lens |= skb_network_header_len(skb);
4095 l4_hdr = ipv6_hdr(skb)->nexthdr;
4096 break;
4097 default:
4098 if (unlikely(net_ratelimit())) {
4099 dev_warn(tx_ring->dev,
4100 "partial checksum but proto=%x!\n",
4101 first->protocol);
4102 }
4103 break;
4104 }
4105
4106 switch (l4_hdr) {
4107 case IPPROTO_TCP:
4108 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4109 mss_l4len_idx = tcp_hdrlen(skb) <<
4110 E1000_ADVTXD_L4LEN_SHIFT;
4111 break;
4112 case IPPROTO_SCTP:
4113 type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4114 mss_l4len_idx = sizeof(struct sctphdr) <<
4115 E1000_ADVTXD_L4LEN_SHIFT;
4116 break;
4117 case IPPROTO_UDP:
4118 mss_l4len_idx = sizeof(struct udphdr) <<
4119 E1000_ADVTXD_L4LEN_SHIFT;
4120 break;
4121 default:
4122 if (unlikely(net_ratelimit())) {
4123 dev_warn(tx_ring->dev,
4124 "partial checksum but l4 proto=%x!\n",
4125 l4_hdr);
4126 }
4127 break;
4128 }
4129
4130 /* update TX checksum flag */
4131 first->tx_flags |= IGB_TX_FLAGS_CSUM;
4132 }
4133
4134 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4135 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4136
4137 igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4138 }
4139
4140 static __le32 igb_tx_cmd_type(u32 tx_flags)
4141 {
4142 /* set type for advanced descriptor with frame checksum insertion */
4143 __le32 cmd_type = cpu_to_le32(E1000_ADVTXD_DTYP_DATA |
4144 E1000_ADVTXD_DCMD_IFCS |
4145 E1000_ADVTXD_DCMD_DEXT);
4146
4147 /* set HW vlan bit if vlan is present */
4148 if (tx_flags & IGB_TX_FLAGS_VLAN)
4149 cmd_type |= cpu_to_le32(E1000_ADVTXD_DCMD_VLE);
4150
4151 #ifdef CONFIG_IGB_PTP
4152 /* set timestamp bit if present */
4153 if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP))
4154 cmd_type |= cpu_to_le32(E1000_ADVTXD_MAC_TSTAMP);
4155 #endif /* CONFIG_IGB_PTP */
4156
4157 /* set segmentation bits for TSO */
4158 if (tx_flags & IGB_TX_FLAGS_TSO)
4159 cmd_type |= cpu_to_le32(E1000_ADVTXD_DCMD_TSE);
4160
4161 return cmd_type;
4162 }
4163
4164 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4165 union e1000_adv_tx_desc *tx_desc,
4166 u32 tx_flags, unsigned int paylen)
4167 {
4168 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4169
4170 /* 82575 requires a unique index per ring if any offload is enabled */
4171 if ((tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_VLAN)) &&
4172 test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4173 olinfo_status |= tx_ring->reg_idx << 4;
4174
4175 /* insert L4 checksum */
4176 if (tx_flags & IGB_TX_FLAGS_CSUM) {
4177 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
4178
4179 /* insert IPv4 checksum */
4180 if (tx_flags & IGB_TX_FLAGS_IPV4)
4181 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
4182 }
4183
4184 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4185 }
4186
4187 /*
4188 * The largest size we can write to the descriptor is 65535. In order to
4189 * maintain a power of two alignment we have to limit ourselves to 32K.
4190 */
4191 #define IGB_MAX_TXD_PWR 15
4192 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
4193
4194 static void igb_tx_map(struct igb_ring *tx_ring,
4195 struct igb_tx_buffer *first,
4196 const u8 hdr_len)
4197 {
4198 struct sk_buff *skb = first->skb;
4199 struct igb_tx_buffer *tx_buffer;
4200 union e1000_adv_tx_desc *tx_desc;
4201 dma_addr_t dma;
4202 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
4203 unsigned int data_len = skb->data_len;
4204 unsigned int size = skb_headlen(skb);
4205 unsigned int paylen = skb->len - hdr_len;
4206 __le32 cmd_type;
4207 u32 tx_flags = first->tx_flags;
4208 u16 i = tx_ring->next_to_use;
4209
4210 tx_desc = IGB_TX_DESC(tx_ring, i);
4211
4212 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, paylen);
4213 cmd_type = igb_tx_cmd_type(tx_flags);
4214
4215 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4216 if (dma_mapping_error(tx_ring->dev, dma))
4217 goto dma_error;
4218
4219 /* record length, and DMA address */
4220 dma_unmap_len_set(first, len, size);
4221 dma_unmap_addr_set(first, dma, dma);
4222 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4223
4224 for (;;) {
4225 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4226 tx_desc->read.cmd_type_len =
4227 cmd_type | cpu_to_le32(IGB_MAX_DATA_PER_TXD);
4228
4229 i++;
4230 tx_desc++;
4231 if (i == tx_ring->count) {
4232 tx_desc = IGB_TX_DESC(tx_ring, 0);
4233 i = 0;
4234 }
4235
4236 dma += IGB_MAX_DATA_PER_TXD;
4237 size -= IGB_MAX_DATA_PER_TXD;
4238
4239 tx_desc->read.olinfo_status = 0;
4240 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4241 }
4242
4243 if (likely(!data_len))
4244 break;
4245
4246 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
4247
4248 i++;
4249 tx_desc++;
4250 if (i == tx_ring->count) {
4251 tx_desc = IGB_TX_DESC(tx_ring, 0);
4252 i = 0;
4253 }
4254
4255 size = skb_frag_size(frag);
4256 data_len -= size;
4257
4258 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4259 size, DMA_TO_DEVICE);
4260 if (dma_mapping_error(tx_ring->dev, dma))
4261 goto dma_error;
4262
4263 tx_buffer = &tx_ring->tx_buffer_info[i];
4264 dma_unmap_len_set(tx_buffer, len, size);
4265 dma_unmap_addr_set(tx_buffer, dma, dma);
4266
4267 tx_desc->read.olinfo_status = 0;
4268 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4269
4270 frag++;
4271 }
4272
4273 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
4274
4275 /* write last descriptor with RS and EOP bits */
4276 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IGB_TXD_DCMD);
4277 if (unlikely(skb->no_fcs))
4278 cmd_type &= ~(cpu_to_le32(E1000_ADVTXD_DCMD_IFCS));
4279 tx_desc->read.cmd_type_len = cmd_type;
4280
4281 /* set the timestamp */
4282 first->time_stamp = jiffies;
4283
4284 /*
4285 * Force memory writes to complete before letting h/w know there
4286 * are new descriptors to fetch. (Only applicable for weak-ordered
4287 * memory model archs, such as IA-64).
4288 *
4289 * We also need this memory barrier to make certain all of the
4290 * status bits have been updated before next_to_watch is written.
4291 */
4292 wmb();
4293
4294 /* set next_to_watch value indicating a packet is present */
4295 first->next_to_watch = tx_desc;
4296
4297 i++;
4298 if (i == tx_ring->count)
4299 i = 0;
4300
4301 tx_ring->next_to_use = i;
4302
4303 writel(i, tx_ring->tail);
4304
4305 /* we need this if more than one processor can write to our tail
4306 * at a time, it syncronizes IO on IA64/Altix systems */
4307 mmiowb();
4308
4309 return;
4310
4311 dma_error:
4312 dev_err(tx_ring->dev, "TX DMA map failed\n");
4313
4314 /* clear dma mappings for failed tx_buffer_info map */
4315 for (;;) {
4316 tx_buffer = &tx_ring->tx_buffer_info[i];
4317 igb_unmap_and_free_tx_resource(tx_ring, tx_buffer);
4318 if (tx_buffer == first)
4319 break;
4320 if (i == 0)
4321 i = tx_ring->count;
4322 i--;
4323 }
4324
4325 tx_ring->next_to_use = i;
4326 }
4327
4328 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4329 {
4330 struct net_device *netdev = tx_ring->netdev;
4331
4332 netif_stop_subqueue(netdev, tx_ring->queue_index);
4333
4334 /* Herbert's original patch had:
4335 * smp_mb__after_netif_stop_queue();
4336 * but since that doesn't exist yet, just open code it. */
4337 smp_mb();
4338
4339 /* We need to check again in a case another CPU has just
4340 * made room available. */
4341 if (igb_desc_unused(tx_ring) < size)
4342 return -EBUSY;
4343
4344 /* A reprieve! */
4345 netif_wake_subqueue(netdev, tx_ring->queue_index);
4346
4347 u64_stats_update_begin(&tx_ring->tx_syncp2);
4348 tx_ring->tx_stats.restart_queue2++;
4349 u64_stats_update_end(&tx_ring->tx_syncp2);
4350
4351 return 0;
4352 }
4353
4354 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4355 {
4356 if (igb_desc_unused(tx_ring) >= size)
4357 return 0;
4358 return __igb_maybe_stop_tx(tx_ring, size);
4359 }
4360
4361 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4362 struct igb_ring *tx_ring)
4363 {
4364 #ifdef CONFIG_IGB_PTP
4365 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
4366 #endif /* CONFIG_IGB_PTP */
4367 struct igb_tx_buffer *first;
4368 int tso;
4369 u32 tx_flags = 0;
4370 __be16 protocol = vlan_get_protocol(skb);
4371 u8 hdr_len = 0;
4372
4373 /* need: 1 descriptor per page,
4374 * + 2 desc gap to keep tail from touching head,
4375 * + 1 desc for skb->data,
4376 * + 1 desc for context descriptor,
4377 * otherwise try next time */
4378 if (igb_maybe_stop_tx(tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
4379 /* this is a hard error */
4380 return NETDEV_TX_BUSY;
4381 }
4382
4383 /* record the location of the first descriptor for this packet */
4384 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4385 first->skb = skb;
4386 first->bytecount = skb->len;
4387 first->gso_segs = 1;
4388
4389 #ifdef CONFIG_IGB_PTP
4390 if (unlikely((skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
4391 !(adapter->ptp_tx_skb))) {
4392 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
4393 tx_flags |= IGB_TX_FLAGS_TSTAMP;
4394
4395 adapter->ptp_tx_skb = skb_get(skb);
4396 if (adapter->hw.mac.type == e1000_82576)
4397 schedule_work(&adapter->ptp_tx_work);
4398 }
4399 #endif /* CONFIG_IGB_PTP */
4400
4401 if (vlan_tx_tag_present(skb)) {
4402 tx_flags |= IGB_TX_FLAGS_VLAN;
4403 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
4404 }
4405
4406 /* record initial flags and protocol */
4407 first->tx_flags = tx_flags;
4408 first->protocol = protocol;
4409
4410 tso = igb_tso(tx_ring, first, &hdr_len);
4411 if (tso < 0)
4412 goto out_drop;
4413 else if (!tso)
4414 igb_tx_csum(tx_ring, first);
4415
4416 igb_tx_map(tx_ring, first, hdr_len);
4417
4418 /* Make sure there is space in the ring for the next send. */
4419 igb_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 4);
4420
4421 return NETDEV_TX_OK;
4422
4423 out_drop:
4424 igb_unmap_and_free_tx_resource(tx_ring, first);
4425
4426 return NETDEV_TX_OK;
4427 }
4428
4429 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
4430 struct sk_buff *skb)
4431 {
4432 unsigned int r_idx = skb->queue_mapping;
4433
4434 if (r_idx >= adapter->num_tx_queues)
4435 r_idx = r_idx % adapter->num_tx_queues;
4436
4437 return adapter->tx_ring[r_idx];
4438 }
4439
4440 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
4441 struct net_device *netdev)
4442 {
4443 struct igb_adapter *adapter = netdev_priv(netdev);
4444
4445 if (test_bit(__IGB_DOWN, &adapter->state)) {
4446 dev_kfree_skb_any(skb);
4447 return NETDEV_TX_OK;
4448 }
4449
4450 if (skb->len <= 0) {
4451 dev_kfree_skb_any(skb);
4452 return NETDEV_TX_OK;
4453 }
4454
4455 /*
4456 * The minimum packet size with TCTL.PSP set is 17 so pad the skb
4457 * in order to meet this minimum size requirement.
4458 */
4459 if (unlikely(skb->len < 17)) {
4460 if (skb_pad(skb, 17 - skb->len))
4461 return NETDEV_TX_OK;
4462 skb->len = 17;
4463 skb_set_tail_pointer(skb, 17);
4464 }
4465
4466 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
4467 }
4468
4469 /**
4470 * igb_tx_timeout - Respond to a Tx Hang
4471 * @netdev: network interface device structure
4472 **/
4473 static void igb_tx_timeout(struct net_device *netdev)
4474 {
4475 struct igb_adapter *adapter = netdev_priv(netdev);
4476 struct e1000_hw *hw = &adapter->hw;
4477
4478 /* Do the reset outside of interrupt context */
4479 adapter->tx_timeout_count++;
4480
4481 if (hw->mac.type >= e1000_82580)
4482 hw->dev_spec._82575.global_device_reset = true;
4483
4484 schedule_work(&adapter->reset_task);
4485 wr32(E1000_EICS,
4486 (adapter->eims_enable_mask & ~adapter->eims_other));
4487 }
4488
4489 static void igb_reset_task(struct work_struct *work)
4490 {
4491 struct igb_adapter *adapter;
4492 adapter = container_of(work, struct igb_adapter, reset_task);
4493
4494 igb_dump(adapter);
4495 netdev_err(adapter->netdev, "Reset adapter\n");
4496 igb_reinit_locked(adapter);
4497 }
4498
4499 /**
4500 * igb_get_stats64 - Get System Network Statistics
4501 * @netdev: network interface device structure
4502 * @stats: rtnl_link_stats64 pointer
4503 *
4504 **/
4505 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
4506 struct rtnl_link_stats64 *stats)
4507 {
4508 struct igb_adapter *adapter = netdev_priv(netdev);
4509
4510 spin_lock(&adapter->stats64_lock);
4511 igb_update_stats(adapter, &adapter->stats64);
4512 memcpy(stats, &adapter->stats64, sizeof(*stats));
4513 spin_unlock(&adapter->stats64_lock);
4514
4515 return stats;
4516 }
4517
4518 /**
4519 * igb_change_mtu - Change the Maximum Transfer Unit
4520 * @netdev: network interface device structure
4521 * @new_mtu: new value for maximum frame size
4522 *
4523 * Returns 0 on success, negative on failure
4524 **/
4525 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
4526 {
4527 struct igb_adapter *adapter = netdev_priv(netdev);
4528 struct pci_dev *pdev = adapter->pdev;
4529 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4530
4531 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4532 dev_err(&pdev->dev, "Invalid MTU setting\n");
4533 return -EINVAL;
4534 }
4535
4536 #define MAX_STD_JUMBO_FRAME_SIZE 9238
4537 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4538 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
4539 return -EINVAL;
4540 }
4541
4542 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
4543 msleep(1);
4544
4545 /* igb_down has a dependency on max_frame_size */
4546 adapter->max_frame_size = max_frame;
4547
4548 if (netif_running(netdev))
4549 igb_down(adapter);
4550
4551 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
4552 netdev->mtu, new_mtu);
4553 netdev->mtu = new_mtu;
4554
4555 if (netif_running(netdev))
4556 igb_up(adapter);
4557 else
4558 igb_reset(adapter);
4559
4560 clear_bit(__IGB_RESETTING, &adapter->state);
4561
4562 return 0;
4563 }
4564
4565 /**
4566 * igb_update_stats - Update the board statistics counters
4567 * @adapter: board private structure
4568 **/
4569
4570 void igb_update_stats(struct igb_adapter *adapter,
4571 struct rtnl_link_stats64 *net_stats)
4572 {
4573 struct e1000_hw *hw = &adapter->hw;
4574 struct pci_dev *pdev = adapter->pdev;
4575 u32 reg, mpc;
4576 u16 phy_tmp;
4577 int i;
4578 u64 bytes, packets;
4579 unsigned int start;
4580 u64 _bytes, _packets;
4581
4582 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
4583
4584 /*
4585 * Prevent stats update while adapter is being reset, or if the pci
4586 * connection is down.
4587 */
4588 if (adapter->link_speed == 0)
4589 return;
4590 if (pci_channel_offline(pdev))
4591 return;
4592
4593 bytes = 0;
4594 packets = 0;
4595 for (i = 0; i < adapter->num_rx_queues; i++) {
4596 u32 rqdpc = rd32(E1000_RQDPC(i));
4597 struct igb_ring *ring = adapter->rx_ring[i];
4598
4599 if (rqdpc) {
4600 ring->rx_stats.drops += rqdpc;
4601 net_stats->rx_fifo_errors += rqdpc;
4602 }
4603
4604 do {
4605 start = u64_stats_fetch_begin_bh(&ring->rx_syncp);
4606 _bytes = ring->rx_stats.bytes;
4607 _packets = ring->rx_stats.packets;
4608 } while (u64_stats_fetch_retry_bh(&ring->rx_syncp, start));
4609 bytes += _bytes;
4610 packets += _packets;
4611 }
4612
4613 net_stats->rx_bytes = bytes;
4614 net_stats->rx_packets = packets;
4615
4616 bytes = 0;
4617 packets = 0;
4618 for (i = 0; i < adapter->num_tx_queues; i++) {
4619 struct igb_ring *ring = adapter->tx_ring[i];
4620 do {
4621 start = u64_stats_fetch_begin_bh(&ring->tx_syncp);
4622 _bytes = ring->tx_stats.bytes;
4623 _packets = ring->tx_stats.packets;
4624 } while (u64_stats_fetch_retry_bh(&ring->tx_syncp, start));
4625 bytes += _bytes;
4626 packets += _packets;
4627 }
4628 net_stats->tx_bytes = bytes;
4629 net_stats->tx_packets = packets;
4630
4631 /* read stats registers */
4632 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
4633 adapter->stats.gprc += rd32(E1000_GPRC);
4634 adapter->stats.gorc += rd32(E1000_GORCL);
4635 rd32(E1000_GORCH); /* clear GORCL */
4636 adapter->stats.bprc += rd32(E1000_BPRC);
4637 adapter->stats.mprc += rd32(E1000_MPRC);
4638 adapter->stats.roc += rd32(E1000_ROC);
4639
4640 adapter->stats.prc64 += rd32(E1000_PRC64);
4641 adapter->stats.prc127 += rd32(E1000_PRC127);
4642 adapter->stats.prc255 += rd32(E1000_PRC255);
4643 adapter->stats.prc511 += rd32(E1000_PRC511);
4644 adapter->stats.prc1023 += rd32(E1000_PRC1023);
4645 adapter->stats.prc1522 += rd32(E1000_PRC1522);
4646 adapter->stats.symerrs += rd32(E1000_SYMERRS);
4647 adapter->stats.sec += rd32(E1000_SEC);
4648
4649 mpc = rd32(E1000_MPC);
4650 adapter->stats.mpc += mpc;
4651 net_stats->rx_fifo_errors += mpc;
4652 adapter->stats.scc += rd32(E1000_SCC);
4653 adapter->stats.ecol += rd32(E1000_ECOL);
4654 adapter->stats.mcc += rd32(E1000_MCC);
4655 adapter->stats.latecol += rd32(E1000_LATECOL);
4656 adapter->stats.dc += rd32(E1000_DC);
4657 adapter->stats.rlec += rd32(E1000_RLEC);
4658 adapter->stats.xonrxc += rd32(E1000_XONRXC);
4659 adapter->stats.xontxc += rd32(E1000_XONTXC);
4660 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
4661 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
4662 adapter->stats.fcruc += rd32(E1000_FCRUC);
4663 adapter->stats.gptc += rd32(E1000_GPTC);
4664 adapter->stats.gotc += rd32(E1000_GOTCL);
4665 rd32(E1000_GOTCH); /* clear GOTCL */
4666 adapter->stats.rnbc += rd32(E1000_RNBC);
4667 adapter->stats.ruc += rd32(E1000_RUC);
4668 adapter->stats.rfc += rd32(E1000_RFC);
4669 adapter->stats.rjc += rd32(E1000_RJC);
4670 adapter->stats.tor += rd32(E1000_TORH);
4671 adapter->stats.tot += rd32(E1000_TOTH);
4672 adapter->stats.tpr += rd32(E1000_TPR);
4673
4674 adapter->stats.ptc64 += rd32(E1000_PTC64);
4675 adapter->stats.ptc127 += rd32(E1000_PTC127);
4676 adapter->stats.ptc255 += rd32(E1000_PTC255);
4677 adapter->stats.ptc511 += rd32(E1000_PTC511);
4678 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
4679 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
4680
4681 adapter->stats.mptc += rd32(E1000_MPTC);
4682 adapter->stats.bptc += rd32(E1000_BPTC);
4683
4684 adapter->stats.tpt += rd32(E1000_TPT);
4685 adapter->stats.colc += rd32(E1000_COLC);
4686
4687 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
4688 /* read internal phy specific stats */
4689 reg = rd32(E1000_CTRL_EXT);
4690 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
4691 adapter->stats.rxerrc += rd32(E1000_RXERRC);
4692
4693 /* this stat has invalid values on i210/i211 */
4694 if ((hw->mac.type != e1000_i210) &&
4695 (hw->mac.type != e1000_i211))
4696 adapter->stats.tncrs += rd32(E1000_TNCRS);
4697 }
4698
4699 adapter->stats.tsctc += rd32(E1000_TSCTC);
4700 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
4701
4702 adapter->stats.iac += rd32(E1000_IAC);
4703 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
4704 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
4705 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
4706 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
4707 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
4708 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
4709 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
4710 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
4711
4712 /* Fill out the OS statistics structure */
4713 net_stats->multicast = adapter->stats.mprc;
4714 net_stats->collisions = adapter->stats.colc;
4715
4716 /* Rx Errors */
4717
4718 /* RLEC on some newer hardware can be incorrect so build
4719 * our own version based on RUC and ROC */
4720 net_stats->rx_errors = adapter->stats.rxerrc +
4721 adapter->stats.crcerrs + adapter->stats.algnerrc +
4722 adapter->stats.ruc + adapter->stats.roc +
4723 adapter->stats.cexterr;
4724 net_stats->rx_length_errors = adapter->stats.ruc +
4725 adapter->stats.roc;
4726 net_stats->rx_crc_errors = adapter->stats.crcerrs;
4727 net_stats->rx_frame_errors = adapter->stats.algnerrc;
4728 net_stats->rx_missed_errors = adapter->stats.mpc;
4729
4730 /* Tx Errors */
4731 net_stats->tx_errors = adapter->stats.ecol +
4732 adapter->stats.latecol;
4733 net_stats->tx_aborted_errors = adapter->stats.ecol;
4734 net_stats->tx_window_errors = adapter->stats.latecol;
4735 net_stats->tx_carrier_errors = adapter->stats.tncrs;
4736
4737 /* Tx Dropped needs to be maintained elsewhere */
4738
4739 /* Phy Stats */
4740 if (hw->phy.media_type == e1000_media_type_copper) {
4741 if ((adapter->link_speed == SPEED_1000) &&
4742 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
4743 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
4744 adapter->phy_stats.idle_errors += phy_tmp;
4745 }
4746 }
4747
4748 /* Management Stats */
4749 adapter->stats.mgptc += rd32(E1000_MGTPTC);
4750 adapter->stats.mgprc += rd32(E1000_MGTPRC);
4751 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
4752
4753 /* OS2BMC Stats */
4754 reg = rd32(E1000_MANC);
4755 if (reg & E1000_MANC_EN_BMC2OS) {
4756 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
4757 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
4758 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
4759 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
4760 }
4761 }
4762
4763 static irqreturn_t igb_msix_other(int irq, void *data)
4764 {
4765 struct igb_adapter *adapter = data;
4766 struct e1000_hw *hw = &adapter->hw;
4767 u32 icr = rd32(E1000_ICR);
4768 /* reading ICR causes bit 31 of EICR to be cleared */
4769
4770 if (icr & E1000_ICR_DRSTA)
4771 schedule_work(&adapter->reset_task);
4772
4773 if (icr & E1000_ICR_DOUTSYNC) {
4774 /* HW is reporting DMA is out of sync */
4775 adapter->stats.doosync++;
4776 /* The DMA Out of Sync is also indication of a spoof event
4777 * in IOV mode. Check the Wrong VM Behavior register to
4778 * see if it is really a spoof event. */
4779 igb_check_wvbr(adapter);
4780 }
4781
4782 /* Check for a mailbox event */
4783 if (icr & E1000_ICR_VMMB)
4784 igb_msg_task(adapter);
4785
4786 if (icr & E1000_ICR_LSC) {
4787 hw->mac.get_link_status = 1;
4788 /* guard against interrupt when we're going down */
4789 if (!test_bit(__IGB_DOWN, &adapter->state))
4790 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4791 }
4792
4793 #ifdef CONFIG_IGB_PTP
4794 if (icr & E1000_ICR_TS) {
4795 u32 tsicr = rd32(E1000_TSICR);
4796
4797 if (tsicr & E1000_TSICR_TXTS) {
4798 /* acknowledge the interrupt */
4799 wr32(E1000_TSICR, E1000_TSICR_TXTS);
4800 /* retrieve hardware timestamp */
4801 schedule_work(&adapter->ptp_tx_work);
4802 }
4803 }
4804 #endif /* CONFIG_IGB_PTP */
4805
4806 wr32(E1000_EIMS, adapter->eims_other);
4807
4808 return IRQ_HANDLED;
4809 }
4810
4811 static void igb_write_itr(struct igb_q_vector *q_vector)
4812 {
4813 struct igb_adapter *adapter = q_vector->adapter;
4814 u32 itr_val = q_vector->itr_val & 0x7FFC;
4815
4816 if (!q_vector->set_itr)
4817 return;
4818
4819 if (!itr_val)
4820 itr_val = 0x4;
4821
4822 if (adapter->hw.mac.type == e1000_82575)
4823 itr_val |= itr_val << 16;
4824 else
4825 itr_val |= E1000_EITR_CNT_IGNR;
4826
4827 writel(itr_val, q_vector->itr_register);
4828 q_vector->set_itr = 0;
4829 }
4830
4831 static irqreturn_t igb_msix_ring(int irq, void *data)
4832 {
4833 struct igb_q_vector *q_vector = data;
4834
4835 /* Write the ITR value calculated from the previous interrupt. */
4836 igb_write_itr(q_vector);
4837
4838 napi_schedule(&q_vector->napi);
4839
4840 return IRQ_HANDLED;
4841 }
4842
4843 #ifdef CONFIG_IGB_DCA
4844 static void igb_update_dca(struct igb_q_vector *q_vector)
4845 {
4846 struct igb_adapter *adapter = q_vector->adapter;
4847 struct e1000_hw *hw = &adapter->hw;
4848 int cpu = get_cpu();
4849
4850 if (q_vector->cpu == cpu)
4851 goto out_no_update;
4852
4853 if (q_vector->tx.ring) {
4854 int q = q_vector->tx.ring->reg_idx;
4855 u32 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
4856 if (hw->mac.type == e1000_82575) {
4857 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
4858 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4859 } else {
4860 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
4861 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4862 E1000_DCA_TXCTRL_CPUID_SHIFT;
4863 }
4864 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
4865 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
4866 }
4867 if (q_vector->rx.ring) {
4868 int q = q_vector->rx.ring->reg_idx;
4869 u32 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
4870 if (hw->mac.type == e1000_82575) {
4871 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
4872 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4873 } else {
4874 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
4875 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4876 E1000_DCA_RXCTRL_CPUID_SHIFT;
4877 }
4878 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
4879 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
4880 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
4881 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
4882 }
4883 q_vector->cpu = cpu;
4884 out_no_update:
4885 put_cpu();
4886 }
4887
4888 static void igb_setup_dca(struct igb_adapter *adapter)
4889 {
4890 struct e1000_hw *hw = &adapter->hw;
4891 int i;
4892
4893 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
4894 return;
4895
4896 /* Always use CB2 mode, difference is masked in the CB driver. */
4897 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
4898
4899 for (i = 0; i < adapter->num_q_vectors; i++) {
4900 adapter->q_vector[i]->cpu = -1;
4901 igb_update_dca(adapter->q_vector[i]);
4902 }
4903 }
4904
4905 static int __igb_notify_dca(struct device *dev, void *data)
4906 {
4907 struct net_device *netdev = dev_get_drvdata(dev);
4908 struct igb_adapter *adapter = netdev_priv(netdev);
4909 struct pci_dev *pdev = adapter->pdev;
4910 struct e1000_hw *hw = &adapter->hw;
4911 unsigned long event = *(unsigned long *)data;
4912
4913 switch (event) {
4914 case DCA_PROVIDER_ADD:
4915 /* if already enabled, don't do it again */
4916 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
4917 break;
4918 if (dca_add_requester(dev) == 0) {
4919 adapter->flags |= IGB_FLAG_DCA_ENABLED;
4920 dev_info(&pdev->dev, "DCA enabled\n");
4921 igb_setup_dca(adapter);
4922 break;
4923 }
4924 /* Fall Through since DCA is disabled. */
4925 case DCA_PROVIDER_REMOVE:
4926 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
4927 /* without this a class_device is left
4928 * hanging around in the sysfs model */
4929 dca_remove_requester(dev);
4930 dev_info(&pdev->dev, "DCA disabled\n");
4931 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
4932 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
4933 }
4934 break;
4935 }
4936
4937 return 0;
4938 }
4939
4940 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
4941 void *p)
4942 {
4943 int ret_val;
4944
4945 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
4946 __igb_notify_dca);
4947
4948 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4949 }
4950 #endif /* CONFIG_IGB_DCA */
4951
4952 #ifdef CONFIG_PCI_IOV
4953 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
4954 {
4955 unsigned char mac_addr[ETH_ALEN];
4956
4957 eth_random_addr(mac_addr);
4958 igb_set_vf_mac(adapter, vf, mac_addr);
4959
4960 return 0;
4961 }
4962
4963 static bool igb_vfs_are_assigned(struct igb_adapter *adapter)
4964 {
4965 struct pci_dev *pdev = adapter->pdev;
4966 struct pci_dev *vfdev;
4967 int dev_id;
4968
4969 switch (adapter->hw.mac.type) {
4970 case e1000_82576:
4971 dev_id = IGB_82576_VF_DEV_ID;
4972 break;
4973 case e1000_i350:
4974 dev_id = IGB_I350_VF_DEV_ID;
4975 break;
4976 default:
4977 return false;
4978 }
4979
4980 /* loop through all the VFs to see if we own any that are assigned */
4981 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, dev_id, NULL);
4982 while (vfdev) {
4983 /* if we don't own it we don't care */
4984 if (vfdev->is_virtfn && vfdev->physfn == pdev) {
4985 /* if it is assigned we cannot release it */
4986 if (vfdev->dev_flags & PCI_DEV_FLAGS_ASSIGNED)
4987 return true;
4988 }
4989
4990 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, dev_id, vfdev);
4991 }
4992
4993 return false;
4994 }
4995
4996 #endif
4997 static void igb_ping_all_vfs(struct igb_adapter *adapter)
4998 {
4999 struct e1000_hw *hw = &adapter->hw;
5000 u32 ping;
5001 int i;
5002
5003 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5004 ping = E1000_PF_CONTROL_MSG;
5005 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5006 ping |= E1000_VT_MSGTYPE_CTS;
5007 igb_write_mbx(hw, &ping, 1, i);
5008 }
5009 }
5010
5011 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5012 {
5013 struct e1000_hw *hw = &adapter->hw;
5014 u32 vmolr = rd32(E1000_VMOLR(vf));
5015 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5016
5017 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5018 IGB_VF_FLAG_MULTI_PROMISC);
5019 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5020
5021 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5022 vmolr |= E1000_VMOLR_MPME;
5023 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5024 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5025 } else {
5026 /*
5027 * if we have hashes and we are clearing a multicast promisc
5028 * flag we need to write the hashes to the MTA as this step
5029 * was previously skipped
5030 */
5031 if (vf_data->num_vf_mc_hashes > 30) {
5032 vmolr |= E1000_VMOLR_MPME;
5033 } else if (vf_data->num_vf_mc_hashes) {
5034 int j;
5035 vmolr |= E1000_VMOLR_ROMPE;
5036 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5037 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5038 }
5039 }
5040
5041 wr32(E1000_VMOLR(vf), vmolr);
5042
5043 /* there are flags left unprocessed, likely not supported */
5044 if (*msgbuf & E1000_VT_MSGINFO_MASK)
5045 return -EINVAL;
5046
5047 return 0;
5048
5049 }
5050
5051 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5052 u32 *msgbuf, u32 vf)
5053 {
5054 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5055 u16 *hash_list = (u16 *)&msgbuf[1];
5056 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5057 int i;
5058
5059 /* salt away the number of multicast addresses assigned
5060 * to this VF for later use to restore when the PF multi cast
5061 * list changes
5062 */
5063 vf_data->num_vf_mc_hashes = n;
5064
5065 /* only up to 30 hash values supported */
5066 if (n > 30)
5067 n = 30;
5068
5069 /* store the hashes for later use */
5070 for (i = 0; i < n; i++)
5071 vf_data->vf_mc_hashes[i] = hash_list[i];
5072
5073 /* Flush and reset the mta with the new values */
5074 igb_set_rx_mode(adapter->netdev);
5075
5076 return 0;
5077 }
5078
5079 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5080 {
5081 struct e1000_hw *hw = &adapter->hw;
5082 struct vf_data_storage *vf_data;
5083 int i, j;
5084
5085 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5086 u32 vmolr = rd32(E1000_VMOLR(i));
5087 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5088
5089 vf_data = &adapter->vf_data[i];
5090
5091 if ((vf_data->num_vf_mc_hashes > 30) ||
5092 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5093 vmolr |= E1000_VMOLR_MPME;
5094 } else if (vf_data->num_vf_mc_hashes) {
5095 vmolr |= E1000_VMOLR_ROMPE;
5096 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5097 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5098 }
5099 wr32(E1000_VMOLR(i), vmolr);
5100 }
5101 }
5102
5103 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5104 {
5105 struct e1000_hw *hw = &adapter->hw;
5106 u32 pool_mask, reg, vid;
5107 int i;
5108
5109 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5110
5111 /* Find the vlan filter for this id */
5112 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5113 reg = rd32(E1000_VLVF(i));
5114
5115 /* remove the vf from the pool */
5116 reg &= ~pool_mask;
5117
5118 /* if pool is empty then remove entry from vfta */
5119 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5120 (reg & E1000_VLVF_VLANID_ENABLE)) {
5121 reg = 0;
5122 vid = reg & E1000_VLVF_VLANID_MASK;
5123 igb_vfta_set(hw, vid, false);
5124 }
5125
5126 wr32(E1000_VLVF(i), reg);
5127 }
5128
5129 adapter->vf_data[vf].vlans_enabled = 0;
5130 }
5131
5132 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5133 {
5134 struct e1000_hw *hw = &adapter->hw;
5135 u32 reg, i;
5136
5137 /* The vlvf table only exists on 82576 hardware and newer */
5138 if (hw->mac.type < e1000_82576)
5139 return -1;
5140
5141 /* we only need to do this if VMDq is enabled */
5142 if (!adapter->vfs_allocated_count)
5143 return -1;
5144
5145 /* Find the vlan filter for this id */
5146 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5147 reg = rd32(E1000_VLVF(i));
5148 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5149 vid == (reg & E1000_VLVF_VLANID_MASK))
5150 break;
5151 }
5152
5153 if (add) {
5154 if (i == E1000_VLVF_ARRAY_SIZE) {
5155 /* Did not find a matching VLAN ID entry that was
5156 * enabled. Search for a free filter entry, i.e.
5157 * one without the enable bit set
5158 */
5159 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5160 reg = rd32(E1000_VLVF(i));
5161 if (!(reg & E1000_VLVF_VLANID_ENABLE))
5162 break;
5163 }
5164 }
5165 if (i < E1000_VLVF_ARRAY_SIZE) {
5166 /* Found an enabled/available entry */
5167 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5168
5169 /* if !enabled we need to set this up in vfta */
5170 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5171 /* add VID to filter table */
5172 igb_vfta_set(hw, vid, true);
5173 reg |= E1000_VLVF_VLANID_ENABLE;
5174 }
5175 reg &= ~E1000_VLVF_VLANID_MASK;
5176 reg |= vid;
5177 wr32(E1000_VLVF(i), reg);
5178
5179 /* do not modify RLPML for PF devices */
5180 if (vf >= adapter->vfs_allocated_count)
5181 return 0;
5182
5183 if (!adapter->vf_data[vf].vlans_enabled) {
5184 u32 size;
5185 reg = rd32(E1000_VMOLR(vf));
5186 size = reg & E1000_VMOLR_RLPML_MASK;
5187 size += 4;
5188 reg &= ~E1000_VMOLR_RLPML_MASK;
5189 reg |= size;
5190 wr32(E1000_VMOLR(vf), reg);
5191 }
5192
5193 adapter->vf_data[vf].vlans_enabled++;
5194 }
5195 } else {
5196 if (i < E1000_VLVF_ARRAY_SIZE) {
5197 /* remove vf from the pool */
5198 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5199 /* if pool is empty then remove entry from vfta */
5200 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5201 reg = 0;
5202 igb_vfta_set(hw, vid, false);
5203 }
5204 wr32(E1000_VLVF(i), reg);
5205
5206 /* do not modify RLPML for PF devices */
5207 if (vf >= adapter->vfs_allocated_count)
5208 return 0;
5209
5210 adapter->vf_data[vf].vlans_enabled--;
5211 if (!adapter->vf_data[vf].vlans_enabled) {
5212 u32 size;
5213 reg = rd32(E1000_VMOLR(vf));
5214 size = reg & E1000_VMOLR_RLPML_MASK;
5215 size -= 4;
5216 reg &= ~E1000_VMOLR_RLPML_MASK;
5217 reg |= size;
5218 wr32(E1000_VMOLR(vf), reg);
5219 }
5220 }
5221 }
5222 return 0;
5223 }
5224
5225 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5226 {
5227 struct e1000_hw *hw = &adapter->hw;
5228
5229 if (vid)
5230 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5231 else
5232 wr32(E1000_VMVIR(vf), 0);
5233 }
5234
5235 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5236 int vf, u16 vlan, u8 qos)
5237 {
5238 int err = 0;
5239 struct igb_adapter *adapter = netdev_priv(netdev);
5240
5241 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5242 return -EINVAL;
5243 if (vlan || qos) {
5244 err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5245 if (err)
5246 goto out;
5247 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5248 igb_set_vmolr(adapter, vf, !vlan);
5249 adapter->vf_data[vf].pf_vlan = vlan;
5250 adapter->vf_data[vf].pf_qos = qos;
5251 dev_info(&adapter->pdev->dev,
5252 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5253 if (test_bit(__IGB_DOWN, &adapter->state)) {
5254 dev_warn(&adapter->pdev->dev,
5255 "The VF VLAN has been set,"
5256 " but the PF device is not up.\n");
5257 dev_warn(&adapter->pdev->dev,
5258 "Bring the PF device up before"
5259 " attempting to use the VF device.\n");
5260 }
5261 } else {
5262 igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5263 false, vf);
5264 igb_set_vmvir(adapter, vlan, vf);
5265 igb_set_vmolr(adapter, vf, true);
5266 adapter->vf_data[vf].pf_vlan = 0;
5267 adapter->vf_data[vf].pf_qos = 0;
5268 }
5269 out:
5270 return err;
5271 }
5272
5273 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5274 {
5275 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5276 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5277
5278 return igb_vlvf_set(adapter, vid, add, vf);
5279 }
5280
5281 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5282 {
5283 /* clear flags - except flag that indicates PF has set the MAC */
5284 adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5285 adapter->vf_data[vf].last_nack = jiffies;
5286
5287 /* reset offloads to defaults */
5288 igb_set_vmolr(adapter, vf, true);
5289
5290 /* reset vlans for device */
5291 igb_clear_vf_vfta(adapter, vf);
5292 if (adapter->vf_data[vf].pf_vlan)
5293 igb_ndo_set_vf_vlan(adapter->netdev, vf,
5294 adapter->vf_data[vf].pf_vlan,
5295 adapter->vf_data[vf].pf_qos);
5296 else
5297 igb_clear_vf_vfta(adapter, vf);
5298
5299 /* reset multicast table array for vf */
5300 adapter->vf_data[vf].num_vf_mc_hashes = 0;
5301
5302 /* Flush and reset the mta with the new values */
5303 igb_set_rx_mode(adapter->netdev);
5304 }
5305
5306 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5307 {
5308 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5309
5310 /* generate a new mac address as we were hotplug removed/added */
5311 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5312 eth_random_addr(vf_mac);
5313
5314 /* process remaining reset events */
5315 igb_vf_reset(adapter, vf);
5316 }
5317
5318 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5319 {
5320 struct e1000_hw *hw = &adapter->hw;
5321 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5322 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5323 u32 reg, msgbuf[3];
5324 u8 *addr = (u8 *)(&msgbuf[1]);
5325
5326 /* process all the same items cleared in a function level reset */
5327 igb_vf_reset(adapter, vf);
5328
5329 /* set vf mac address */
5330 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
5331
5332 /* enable transmit and receive for vf */
5333 reg = rd32(E1000_VFTE);
5334 wr32(E1000_VFTE, reg | (1 << vf));
5335 reg = rd32(E1000_VFRE);
5336 wr32(E1000_VFRE, reg | (1 << vf));
5337
5338 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
5339
5340 /* reply to reset with ack and vf mac address */
5341 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
5342 memcpy(addr, vf_mac, 6);
5343 igb_write_mbx(hw, msgbuf, 3, vf);
5344 }
5345
5346 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
5347 {
5348 /*
5349 * The VF MAC Address is stored in a packed array of bytes
5350 * starting at the second 32 bit word of the msg array
5351 */
5352 unsigned char *addr = (char *)&msg[1];
5353 int err = -1;
5354
5355 if (is_valid_ether_addr(addr))
5356 err = igb_set_vf_mac(adapter, vf, addr);
5357
5358 return err;
5359 }
5360
5361 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
5362 {
5363 struct e1000_hw *hw = &adapter->hw;
5364 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5365 u32 msg = E1000_VT_MSGTYPE_NACK;
5366
5367 /* if device isn't clear to send it shouldn't be reading either */
5368 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
5369 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
5370 igb_write_mbx(hw, &msg, 1, vf);
5371 vf_data->last_nack = jiffies;
5372 }
5373 }
5374
5375 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
5376 {
5377 struct pci_dev *pdev = adapter->pdev;
5378 u32 msgbuf[E1000_VFMAILBOX_SIZE];
5379 struct e1000_hw *hw = &adapter->hw;
5380 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5381 s32 retval;
5382
5383 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
5384
5385 if (retval) {
5386 /* if receive failed revoke VF CTS stats and restart init */
5387 dev_err(&pdev->dev, "Error receiving message from VF\n");
5388 vf_data->flags &= ~IGB_VF_FLAG_CTS;
5389 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5390 return;
5391 goto out;
5392 }
5393
5394 /* this is a message we already processed, do nothing */
5395 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
5396 return;
5397
5398 /*
5399 * until the vf completes a reset it should not be
5400 * allowed to start any configuration.
5401 */
5402
5403 if (msgbuf[0] == E1000_VF_RESET) {
5404 igb_vf_reset_msg(adapter, vf);
5405 return;
5406 }
5407
5408 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
5409 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5410 return;
5411 retval = -1;
5412 goto out;
5413 }
5414
5415 switch ((msgbuf[0] & 0xFFFF)) {
5416 case E1000_VF_SET_MAC_ADDR:
5417 retval = -EINVAL;
5418 if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
5419 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
5420 else
5421 dev_warn(&pdev->dev,
5422 "VF %d attempted to override administratively "
5423 "set MAC address\nReload the VF driver to "
5424 "resume operations\n", vf);
5425 break;
5426 case E1000_VF_SET_PROMISC:
5427 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
5428 break;
5429 case E1000_VF_SET_MULTICAST:
5430 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
5431 break;
5432 case E1000_VF_SET_LPE:
5433 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
5434 break;
5435 case E1000_VF_SET_VLAN:
5436 retval = -1;
5437 if (vf_data->pf_vlan)
5438 dev_warn(&pdev->dev,
5439 "VF %d attempted to override administratively "
5440 "set VLAN tag\nReload the VF driver to "
5441 "resume operations\n", vf);
5442 else
5443 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
5444 break;
5445 default:
5446 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
5447 retval = -1;
5448 break;
5449 }
5450
5451 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
5452 out:
5453 /* notify the VF of the results of what it sent us */
5454 if (retval)
5455 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
5456 else
5457 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
5458
5459 igb_write_mbx(hw, msgbuf, 1, vf);
5460 }
5461
5462 static void igb_msg_task(struct igb_adapter *adapter)
5463 {
5464 struct e1000_hw *hw = &adapter->hw;
5465 u32 vf;
5466
5467 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
5468 /* process any reset requests */
5469 if (!igb_check_for_rst(hw, vf))
5470 igb_vf_reset_event(adapter, vf);
5471
5472 /* process any messages pending */
5473 if (!igb_check_for_msg(hw, vf))
5474 igb_rcv_msg_from_vf(adapter, vf);
5475
5476 /* process any acks */
5477 if (!igb_check_for_ack(hw, vf))
5478 igb_rcv_ack_from_vf(adapter, vf);
5479 }
5480 }
5481
5482 /**
5483 * igb_set_uta - Set unicast filter table address
5484 * @adapter: board private structure
5485 *
5486 * The unicast table address is a register array of 32-bit registers.
5487 * The table is meant to be used in a way similar to how the MTA is used
5488 * however due to certain limitations in the hardware it is necessary to
5489 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
5490 * enable bit to allow vlan tag stripping when promiscuous mode is enabled
5491 **/
5492 static void igb_set_uta(struct igb_adapter *adapter)
5493 {
5494 struct e1000_hw *hw = &adapter->hw;
5495 int i;
5496
5497 /* The UTA table only exists on 82576 hardware and newer */
5498 if (hw->mac.type < e1000_82576)
5499 return;
5500
5501 /* we only need to do this if VMDq is enabled */
5502 if (!adapter->vfs_allocated_count)
5503 return;
5504
5505 for (i = 0; i < hw->mac.uta_reg_count; i++)
5506 array_wr32(E1000_UTA, i, ~0);
5507 }
5508
5509 /**
5510 * igb_intr_msi - Interrupt Handler
5511 * @irq: interrupt number
5512 * @data: pointer to a network interface device structure
5513 **/
5514 static irqreturn_t igb_intr_msi(int irq, void *data)
5515 {
5516 struct igb_adapter *adapter = data;
5517 struct igb_q_vector *q_vector = adapter->q_vector[0];
5518 struct e1000_hw *hw = &adapter->hw;
5519 /* read ICR disables interrupts using IAM */
5520 u32 icr = rd32(E1000_ICR);
5521
5522 igb_write_itr(q_vector);
5523
5524 if (icr & E1000_ICR_DRSTA)
5525 schedule_work(&adapter->reset_task);
5526
5527 if (icr & E1000_ICR_DOUTSYNC) {
5528 /* HW is reporting DMA is out of sync */
5529 adapter->stats.doosync++;
5530 }
5531
5532 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5533 hw->mac.get_link_status = 1;
5534 if (!test_bit(__IGB_DOWN, &adapter->state))
5535 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5536 }
5537
5538 #ifdef CONFIG_IGB_PTP
5539 if (icr & E1000_ICR_TS) {
5540 u32 tsicr = rd32(E1000_TSICR);
5541
5542 if (tsicr & E1000_TSICR_TXTS) {
5543 /* acknowledge the interrupt */
5544 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5545 /* retrieve hardware timestamp */
5546 schedule_work(&adapter->ptp_tx_work);
5547 }
5548 }
5549 #endif /* CONFIG_IGB_PTP */
5550
5551 napi_schedule(&q_vector->napi);
5552
5553 return IRQ_HANDLED;
5554 }
5555
5556 /**
5557 * igb_intr - Legacy Interrupt Handler
5558 * @irq: interrupt number
5559 * @data: pointer to a network interface device structure
5560 **/
5561 static irqreturn_t igb_intr(int irq, void *data)
5562 {
5563 struct igb_adapter *adapter = data;
5564 struct igb_q_vector *q_vector = adapter->q_vector[0];
5565 struct e1000_hw *hw = &adapter->hw;
5566 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
5567 * need for the IMC write */
5568 u32 icr = rd32(E1000_ICR);
5569
5570 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5571 * not set, then the adapter didn't send an interrupt */
5572 if (!(icr & E1000_ICR_INT_ASSERTED))
5573 return IRQ_NONE;
5574
5575 igb_write_itr(q_vector);
5576
5577 if (icr & E1000_ICR_DRSTA)
5578 schedule_work(&adapter->reset_task);
5579
5580 if (icr & E1000_ICR_DOUTSYNC) {
5581 /* HW is reporting DMA is out of sync */
5582 adapter->stats.doosync++;
5583 }
5584
5585 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5586 hw->mac.get_link_status = 1;
5587 /* guard against interrupt when we're going down */
5588 if (!test_bit(__IGB_DOWN, &adapter->state))
5589 mod_timer(&adapter->watchdog_timer, jiffies + 1);
5590 }
5591
5592 #ifdef CONFIG_IGB_PTP
5593 if (icr & E1000_ICR_TS) {
5594 u32 tsicr = rd32(E1000_TSICR);
5595
5596 if (tsicr & E1000_TSICR_TXTS) {
5597 /* acknowledge the interrupt */
5598 wr32(E1000_TSICR, E1000_TSICR_TXTS);
5599 /* retrieve hardware timestamp */
5600 schedule_work(&adapter->ptp_tx_work);
5601 }
5602 }
5603 #endif /* CONFIG_IGB_PTP */
5604
5605 napi_schedule(&q_vector->napi);
5606
5607 return IRQ_HANDLED;
5608 }
5609
5610 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
5611 {
5612 struct igb_adapter *adapter = q_vector->adapter;
5613 struct e1000_hw *hw = &adapter->hw;
5614
5615 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
5616 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
5617 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
5618 igb_set_itr(q_vector);
5619 else
5620 igb_update_ring_itr(q_vector);
5621 }
5622
5623 if (!test_bit(__IGB_DOWN, &adapter->state)) {
5624 if (adapter->msix_entries)
5625 wr32(E1000_EIMS, q_vector->eims_value);
5626 else
5627 igb_irq_enable(adapter);
5628 }
5629 }
5630
5631 /**
5632 * igb_poll - NAPI Rx polling callback
5633 * @napi: napi polling structure
5634 * @budget: count of how many packets we should handle
5635 **/
5636 static int igb_poll(struct napi_struct *napi, int budget)
5637 {
5638 struct igb_q_vector *q_vector = container_of(napi,
5639 struct igb_q_vector,
5640 napi);
5641 bool clean_complete = true;
5642
5643 #ifdef CONFIG_IGB_DCA
5644 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
5645 igb_update_dca(q_vector);
5646 #endif
5647 if (q_vector->tx.ring)
5648 clean_complete = igb_clean_tx_irq(q_vector);
5649
5650 if (q_vector->rx.ring)
5651 clean_complete &= igb_clean_rx_irq(q_vector, budget);
5652
5653 /* If all work not completed, return budget and keep polling */
5654 if (!clean_complete)
5655 return budget;
5656
5657 /* If not enough Rx work done, exit the polling mode */
5658 napi_complete(napi);
5659 igb_ring_irq_enable(q_vector);
5660
5661 return 0;
5662 }
5663
5664 /**
5665 * igb_clean_tx_irq - Reclaim resources after transmit completes
5666 * @q_vector: pointer to q_vector containing needed info
5667 *
5668 * returns true if ring is completely cleaned
5669 **/
5670 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
5671 {
5672 struct igb_adapter *adapter = q_vector->adapter;
5673 struct igb_ring *tx_ring = q_vector->tx.ring;
5674 struct igb_tx_buffer *tx_buffer;
5675 union e1000_adv_tx_desc *tx_desc;
5676 unsigned int total_bytes = 0, total_packets = 0;
5677 unsigned int budget = q_vector->tx.work_limit;
5678 unsigned int i = tx_ring->next_to_clean;
5679
5680 if (test_bit(__IGB_DOWN, &adapter->state))
5681 return true;
5682
5683 tx_buffer = &tx_ring->tx_buffer_info[i];
5684 tx_desc = IGB_TX_DESC(tx_ring, i);
5685 i -= tx_ring->count;
5686
5687 do {
5688 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
5689
5690 /* if next_to_watch is not set then there is no work pending */
5691 if (!eop_desc)
5692 break;
5693
5694 /* prevent any other reads prior to eop_desc */
5695 rmb();
5696
5697 /* if DD is not set pending work has not been completed */
5698 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
5699 break;
5700
5701 /* clear next_to_watch to prevent false hangs */
5702 tx_buffer->next_to_watch = NULL;
5703
5704 /* update the statistics for this packet */
5705 total_bytes += tx_buffer->bytecount;
5706 total_packets += tx_buffer->gso_segs;
5707
5708 /* free the skb */
5709 dev_kfree_skb_any(tx_buffer->skb);
5710
5711 /* unmap skb header data */
5712 dma_unmap_single(tx_ring->dev,
5713 dma_unmap_addr(tx_buffer, dma),
5714 dma_unmap_len(tx_buffer, len),
5715 DMA_TO_DEVICE);
5716
5717 /* clear tx_buffer data */
5718 tx_buffer->skb = NULL;
5719 dma_unmap_len_set(tx_buffer, len, 0);
5720
5721 /* clear last DMA location and unmap remaining buffers */
5722 while (tx_desc != eop_desc) {
5723 tx_buffer++;
5724 tx_desc++;
5725 i++;
5726 if (unlikely(!i)) {
5727 i -= tx_ring->count;
5728 tx_buffer = tx_ring->tx_buffer_info;
5729 tx_desc = IGB_TX_DESC(tx_ring, 0);
5730 }
5731
5732 /* unmap any remaining paged data */
5733 if (dma_unmap_len(tx_buffer, len)) {
5734 dma_unmap_page(tx_ring->dev,
5735 dma_unmap_addr(tx_buffer, dma),
5736 dma_unmap_len(tx_buffer, len),
5737 DMA_TO_DEVICE);
5738 dma_unmap_len_set(tx_buffer, len, 0);
5739 }
5740 }
5741
5742 /* move us one more past the eop_desc for start of next pkt */
5743 tx_buffer++;
5744 tx_desc++;
5745 i++;
5746 if (unlikely(!i)) {
5747 i -= tx_ring->count;
5748 tx_buffer = tx_ring->tx_buffer_info;
5749 tx_desc = IGB_TX_DESC(tx_ring, 0);
5750 }
5751
5752 /* issue prefetch for next Tx descriptor */
5753 prefetch(tx_desc);
5754
5755 /* update budget accounting */
5756 budget--;
5757 } while (likely(budget));
5758
5759 netdev_tx_completed_queue(txring_txq(tx_ring),
5760 total_packets, total_bytes);
5761 i += tx_ring->count;
5762 tx_ring->next_to_clean = i;
5763 u64_stats_update_begin(&tx_ring->tx_syncp);
5764 tx_ring->tx_stats.bytes += total_bytes;
5765 tx_ring->tx_stats.packets += total_packets;
5766 u64_stats_update_end(&tx_ring->tx_syncp);
5767 q_vector->tx.total_bytes += total_bytes;
5768 q_vector->tx.total_packets += total_packets;
5769
5770 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
5771 struct e1000_hw *hw = &adapter->hw;
5772
5773 /* Detect a transmit hang in hardware, this serializes the
5774 * check with the clearing of time_stamp and movement of i */
5775 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5776 if (tx_buffer->next_to_watch &&
5777 time_after(jiffies, tx_buffer->time_stamp +
5778 (adapter->tx_timeout_factor * HZ)) &&
5779 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
5780
5781 /* detected Tx unit hang */
5782 dev_err(tx_ring->dev,
5783 "Detected Tx Unit Hang\n"
5784 " Tx Queue <%d>\n"
5785 " TDH <%x>\n"
5786 " TDT <%x>\n"
5787 " next_to_use <%x>\n"
5788 " next_to_clean <%x>\n"
5789 "buffer_info[next_to_clean]\n"
5790 " time_stamp <%lx>\n"
5791 " next_to_watch <%p>\n"
5792 " jiffies <%lx>\n"
5793 " desc.status <%x>\n",
5794 tx_ring->queue_index,
5795 rd32(E1000_TDH(tx_ring->reg_idx)),
5796 readl(tx_ring->tail),
5797 tx_ring->next_to_use,
5798 tx_ring->next_to_clean,
5799 tx_buffer->time_stamp,
5800 tx_buffer->next_to_watch,
5801 jiffies,
5802 tx_buffer->next_to_watch->wb.status);
5803 netif_stop_subqueue(tx_ring->netdev,
5804 tx_ring->queue_index);
5805
5806 /* we are about to reset, no point in enabling stuff */
5807 return true;
5808 }
5809 }
5810
5811 if (unlikely(total_packets &&
5812 netif_carrier_ok(tx_ring->netdev) &&
5813 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
5814 /* Make sure that anybody stopping the queue after this
5815 * sees the new next_to_clean.
5816 */
5817 smp_mb();
5818 if (__netif_subqueue_stopped(tx_ring->netdev,
5819 tx_ring->queue_index) &&
5820 !(test_bit(__IGB_DOWN, &adapter->state))) {
5821 netif_wake_subqueue(tx_ring->netdev,
5822 tx_ring->queue_index);
5823
5824 u64_stats_update_begin(&tx_ring->tx_syncp);
5825 tx_ring->tx_stats.restart_queue++;
5826 u64_stats_update_end(&tx_ring->tx_syncp);
5827 }
5828 }
5829
5830 return !!budget;
5831 }
5832
5833 /**
5834 * igb_reuse_rx_page - page flip buffer and store it back on the ring
5835 * @rx_ring: rx descriptor ring to store buffers on
5836 * @old_buff: donor buffer to have page reused
5837 *
5838 * Synchronizes page for reuse by the adapter
5839 **/
5840 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
5841 struct igb_rx_buffer *old_buff)
5842 {
5843 struct igb_rx_buffer *new_buff;
5844 u16 nta = rx_ring->next_to_alloc;
5845
5846 new_buff = &rx_ring->rx_buffer_info[nta];
5847
5848 /* update, and store next to alloc */
5849 nta++;
5850 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
5851
5852 /* transfer page from old buffer to new buffer */
5853 memcpy(new_buff, old_buff, sizeof(struct igb_rx_buffer));
5854
5855 /* sync the buffer for use by the device */
5856 dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
5857 old_buff->page_offset,
5858 PAGE_SIZE / 2,
5859 DMA_FROM_DEVICE);
5860 }
5861
5862 /**
5863 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
5864 * @rx_ring: rx descriptor ring to transact packets on
5865 * @rx_buffer: buffer containing page to add
5866 * @rx_desc: descriptor containing length of buffer written by hardware
5867 * @skb: sk_buff to place the data into
5868 *
5869 * This function will add the data contained in rx_buffer->page to the skb.
5870 * This is done either through a direct copy if the data in the buffer is
5871 * less than the skb header size, otherwise it will just attach the page as
5872 * a frag to the skb.
5873 *
5874 * The function will then update the page offset if necessary and return
5875 * true if the buffer can be reused by the adapter.
5876 **/
5877 static bool igb_add_rx_frag(struct igb_ring *rx_ring,
5878 struct igb_rx_buffer *rx_buffer,
5879 union e1000_adv_rx_desc *rx_desc,
5880 struct sk_buff *skb)
5881 {
5882 struct page *page = rx_buffer->page;
5883 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
5884
5885 if ((size <= IGB_RX_HDR_LEN) && !skb_is_nonlinear(skb)) {
5886 unsigned char *va = page_address(page) + rx_buffer->page_offset;
5887
5888 #ifdef CONFIG_IGB_PTP
5889 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
5890 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
5891 va += IGB_TS_HDR_LEN;
5892 size -= IGB_TS_HDR_LEN;
5893 }
5894
5895 #endif
5896 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
5897
5898 /* we can reuse buffer as-is, just make sure it is local */
5899 if (likely(page_to_nid(page) == numa_node_id()))
5900 return true;
5901
5902 /* this page cannot be reused so discard it */
5903 put_page(page);
5904 return false;
5905 }
5906
5907 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
5908 rx_buffer->page_offset, size, PAGE_SIZE / 2);
5909
5910 /* avoid re-using remote pages */
5911 if (unlikely(page_to_nid(page) != numa_node_id()))
5912 return false;
5913
5914 /* if we are only owner of page we can reuse it */
5915 if (unlikely(page_count(page) != 1))
5916 return false;
5917
5918 /* flip page offset to other buffer */
5919 rx_buffer->page_offset ^= PAGE_SIZE / 2;
5920
5921 /*
5922 * since we are the only owner of the page and we need to
5923 * increment it, just set the value to 2 in order to avoid
5924 * an unnecessary locked operation
5925 */
5926 atomic_set(&page->_count, 2);
5927
5928 return true;
5929 }
5930
5931 static inline void igb_rx_checksum(struct igb_ring *ring,
5932 union e1000_adv_rx_desc *rx_desc,
5933 struct sk_buff *skb)
5934 {
5935 skb_checksum_none_assert(skb);
5936
5937 /* Ignore Checksum bit is set */
5938 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
5939 return;
5940
5941 /* Rx checksum disabled via ethtool */
5942 if (!(ring->netdev->features & NETIF_F_RXCSUM))
5943 return;
5944
5945 /* TCP/UDP checksum error bit is set */
5946 if (igb_test_staterr(rx_desc,
5947 E1000_RXDEXT_STATERR_TCPE |
5948 E1000_RXDEXT_STATERR_IPE)) {
5949 /*
5950 * work around errata with sctp packets where the TCPE aka
5951 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
5952 * packets, (aka let the stack check the crc32c)
5953 */
5954 if (!((skb->len == 60) &&
5955 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
5956 u64_stats_update_begin(&ring->rx_syncp);
5957 ring->rx_stats.csum_err++;
5958 u64_stats_update_end(&ring->rx_syncp);
5959 }
5960 /* let the stack verify checksum errors */
5961 return;
5962 }
5963 /* It must be a TCP or UDP packet with a valid checksum */
5964 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
5965 E1000_RXD_STAT_UDPCS))
5966 skb->ip_summed = CHECKSUM_UNNECESSARY;
5967
5968 dev_dbg(ring->dev, "cksum success: bits %08X\n",
5969 le32_to_cpu(rx_desc->wb.upper.status_error));
5970 }
5971
5972 static inline void igb_rx_hash(struct igb_ring *ring,
5973 union e1000_adv_rx_desc *rx_desc,
5974 struct sk_buff *skb)
5975 {
5976 if (ring->netdev->features & NETIF_F_RXHASH)
5977 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
5978 }
5979
5980 /**
5981 * igb_get_headlen - determine size of header for LRO/GRO
5982 * @data: pointer to the start of the headers
5983 * @max_len: total length of section to find headers in
5984 *
5985 * This function is meant to determine the length of headers that will
5986 * be recognized by hardware for LRO, and GRO offloads. The main
5987 * motivation of doing this is to only perform one pull for IPv4 TCP
5988 * packets so that we can do basic things like calculating the gso_size
5989 * based on the average data per packet.
5990 **/
5991 static unsigned int igb_get_headlen(unsigned char *data,
5992 unsigned int max_len)
5993 {
5994 union {
5995 unsigned char *network;
5996 /* l2 headers */
5997 struct ethhdr *eth;
5998 struct vlan_hdr *vlan;
5999 /* l3 headers */
6000 struct iphdr *ipv4;
6001 struct ipv6hdr *ipv6;
6002 } hdr;
6003 __be16 protocol;
6004 u8 nexthdr = 0; /* default to not TCP */
6005 u8 hlen;
6006
6007 /* this should never happen, but better safe than sorry */
6008 if (max_len < ETH_HLEN)
6009 return max_len;
6010
6011 /* initialize network frame pointer */
6012 hdr.network = data;
6013
6014 /* set first protocol and move network header forward */
6015 protocol = hdr.eth->h_proto;
6016 hdr.network += ETH_HLEN;
6017
6018 /* handle any vlan tag if present */
6019 if (protocol == __constant_htons(ETH_P_8021Q)) {
6020 if ((hdr.network - data) > (max_len - VLAN_HLEN))
6021 return max_len;
6022
6023 protocol = hdr.vlan->h_vlan_encapsulated_proto;
6024 hdr.network += VLAN_HLEN;
6025 }
6026
6027 /* handle L3 protocols */
6028 if (protocol == __constant_htons(ETH_P_IP)) {
6029 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
6030 return max_len;
6031
6032 /* access ihl as a u8 to avoid unaligned access on ia64 */
6033 hlen = (hdr.network[0] & 0x0F) << 2;
6034
6035 /* verify hlen meets minimum size requirements */
6036 if (hlen < sizeof(struct iphdr))
6037 return hdr.network - data;
6038
6039 /* record next protocol */
6040 nexthdr = hdr.ipv4->protocol;
6041 hdr.network += hlen;
6042 } else if (protocol == __constant_htons(ETH_P_IPV6)) {
6043 if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
6044 return max_len;
6045
6046 /* record next protocol */
6047 nexthdr = hdr.ipv6->nexthdr;
6048 hdr.network += sizeof(struct ipv6hdr);
6049 } else {
6050 return hdr.network - data;
6051 }
6052
6053 /* finally sort out TCP */
6054 if (nexthdr == IPPROTO_TCP) {
6055 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
6056 return max_len;
6057
6058 /* access doff as a u8 to avoid unaligned access on ia64 */
6059 hlen = (hdr.network[12] & 0xF0) >> 2;
6060
6061 /* verify hlen meets minimum size requirements */
6062 if (hlen < sizeof(struct tcphdr))
6063 return hdr.network - data;
6064
6065 hdr.network += hlen;
6066 } else if (nexthdr == IPPROTO_UDP) {
6067 if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
6068 return max_len;
6069
6070 hdr.network += sizeof(struct udphdr);
6071 }
6072
6073 /*
6074 * If everything has gone correctly hdr.network should be the
6075 * data section of the packet and will be the end of the header.
6076 * If not then it probably represents the end of the last recognized
6077 * header.
6078 */
6079 if ((hdr.network - data) < max_len)
6080 return hdr.network - data;
6081 else
6082 return max_len;
6083 }
6084
6085 /**
6086 * igb_pull_tail - igb specific version of skb_pull_tail
6087 * @rx_ring: rx descriptor ring packet is being transacted on
6088 * @rx_desc: pointer to the EOP Rx descriptor
6089 * @skb: pointer to current skb being adjusted
6090 *
6091 * This function is an igb specific version of __pskb_pull_tail. The
6092 * main difference between this version and the original function is that
6093 * this function can make several assumptions about the state of things
6094 * that allow for significant optimizations versus the standard function.
6095 * As a result we can do things like drop a frag and maintain an accurate
6096 * truesize for the skb.
6097 */
6098 static void igb_pull_tail(struct igb_ring *rx_ring,
6099 union e1000_adv_rx_desc *rx_desc,
6100 struct sk_buff *skb)
6101 {
6102 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6103 unsigned char *va;
6104 unsigned int pull_len;
6105
6106 /*
6107 * it is valid to use page_address instead of kmap since we are
6108 * working with pages allocated out of the lomem pool per
6109 * alloc_page(GFP_ATOMIC)
6110 */
6111 va = skb_frag_address(frag);
6112
6113 #ifdef CONFIG_IGB_PTP
6114 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6115 /* retrieve timestamp from buffer */
6116 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6117
6118 /* update pointers to remove timestamp header */
6119 skb_frag_size_sub(frag, IGB_TS_HDR_LEN);
6120 frag->page_offset += IGB_TS_HDR_LEN;
6121 skb->data_len -= IGB_TS_HDR_LEN;
6122 skb->len -= IGB_TS_HDR_LEN;
6123
6124 /* move va to start of packet data */
6125 va += IGB_TS_HDR_LEN;
6126 }
6127
6128 #endif
6129 /*
6130 * we need the header to contain the greater of either ETH_HLEN or
6131 * 60 bytes if the skb->len is less than 60 for skb_pad.
6132 */
6133 pull_len = igb_get_headlen(va, IGB_RX_HDR_LEN);
6134
6135 /* align pull length to size of long to optimize memcpy performance */
6136 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
6137
6138 /* update all of the pointers */
6139 skb_frag_size_sub(frag, pull_len);
6140 frag->page_offset += pull_len;
6141 skb->data_len -= pull_len;
6142 skb->tail += pull_len;
6143 }
6144
6145 /**
6146 * igb_cleanup_headers - Correct corrupted or empty headers
6147 * @rx_ring: rx descriptor ring packet is being transacted on
6148 * @rx_desc: pointer to the EOP Rx descriptor
6149 * @skb: pointer to current skb being fixed
6150 *
6151 * Address the case where we are pulling data in on pages only
6152 * and as such no data is present in the skb header.
6153 *
6154 * In addition if skb is not at least 60 bytes we need to pad it so that
6155 * it is large enough to qualify as a valid Ethernet frame.
6156 *
6157 * Returns true if an error was encountered and skb was freed.
6158 **/
6159 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
6160 union e1000_adv_rx_desc *rx_desc,
6161 struct sk_buff *skb)
6162 {
6163
6164 if (unlikely((igb_test_staterr(rx_desc,
6165 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
6166 struct net_device *netdev = rx_ring->netdev;
6167 if (!(netdev->features & NETIF_F_RXALL)) {
6168 dev_kfree_skb_any(skb);
6169 return true;
6170 }
6171 }
6172
6173 /* place header in linear portion of buffer */
6174 if (skb_is_nonlinear(skb))
6175 igb_pull_tail(rx_ring, rx_desc, skb);
6176
6177 /* if skb_pad returns an error the skb was freed */
6178 if (unlikely(skb->len < 60)) {
6179 int pad_len = 60 - skb->len;
6180
6181 if (skb_pad(skb, pad_len))
6182 return true;
6183 __skb_put(skb, pad_len);
6184 }
6185
6186 return false;
6187 }
6188
6189 /**
6190 * igb_process_skb_fields - Populate skb header fields from Rx descriptor
6191 * @rx_ring: rx descriptor ring packet is being transacted on
6192 * @rx_desc: pointer to the EOP Rx descriptor
6193 * @skb: pointer to current skb being populated
6194 *
6195 * This function checks the ring, descriptor, and packet information in
6196 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
6197 * other fields within the skb.
6198 **/
6199 static void igb_process_skb_fields(struct igb_ring *rx_ring,
6200 union e1000_adv_rx_desc *rx_desc,
6201 struct sk_buff *skb)
6202 {
6203 struct net_device *dev = rx_ring->netdev;
6204
6205 igb_rx_hash(rx_ring, rx_desc, skb);
6206
6207 igb_rx_checksum(rx_ring, rx_desc, skb);
6208
6209 #ifdef CONFIG_IGB_PTP
6210 igb_ptp_rx_hwtstamp(rx_ring->q_vector, rx_desc, skb);
6211 #endif /* CONFIG_IGB_PTP */
6212
6213 if ((dev->features & NETIF_F_HW_VLAN_RX) &&
6214 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
6215 u16 vid;
6216 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
6217 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
6218 vid = be16_to_cpu(rx_desc->wb.upper.vlan);
6219 else
6220 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
6221
6222 __vlan_hwaccel_put_tag(skb, vid);
6223 }
6224
6225 skb_record_rx_queue(skb, rx_ring->queue_index);
6226
6227 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6228 }
6229
6230 static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, int budget)
6231 {
6232 struct igb_ring *rx_ring = q_vector->rx.ring;
6233 union e1000_adv_rx_desc *rx_desc;
6234 struct sk_buff *skb = rx_ring->skb;
6235 unsigned int total_bytes = 0, total_packets = 0;
6236 u16 cleaned_count = igb_desc_unused(rx_ring);
6237 u16 i = rx_ring->next_to_clean;
6238
6239 rx_desc = IGB_RX_DESC(rx_ring, i);
6240
6241 while (igb_test_staterr(rx_desc, E1000_RXD_STAT_DD)) {
6242 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
6243 struct page *page;
6244 union e1000_adv_rx_desc *next_rxd;
6245
6246 i++;
6247 if (i == rx_ring->count)
6248 i = 0;
6249
6250 next_rxd = IGB_RX_DESC(rx_ring, i);
6251 prefetch(next_rxd);
6252
6253 /*
6254 * This memory barrier is needed to keep us from reading
6255 * any other fields out of the rx_desc until we know the
6256 * RXD_STAT_DD bit is set
6257 */
6258 rmb();
6259
6260 page = buffer_info->page;
6261 prefetchw(page);
6262
6263 if (likely(!skb)) {
6264 void *page_addr = page_address(page) +
6265 buffer_info->page_offset;
6266
6267 /* prefetch first cache line of first page */
6268 prefetch(page_addr);
6269 #if L1_CACHE_BYTES < 128
6270 prefetch(page_addr + L1_CACHE_BYTES);
6271 #endif
6272
6273 /* allocate a skb to store the frags */
6274 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
6275 IGB_RX_HDR_LEN);
6276 if (unlikely(!skb)) {
6277 rx_ring->rx_stats.alloc_failed++;
6278 break;
6279 }
6280
6281 /*
6282 * we will be copying header into skb->data in
6283 * pskb_may_pull so it is in our interest to prefetch
6284 * it now to avoid a possible cache miss
6285 */
6286 prefetchw(skb->data);
6287 }
6288
6289 /* we are reusing so sync this buffer for CPU use */
6290 dma_sync_single_range_for_cpu(rx_ring->dev,
6291 buffer_info->dma,
6292 buffer_info->page_offset,
6293 PAGE_SIZE / 2,
6294 DMA_FROM_DEVICE);
6295
6296 /* pull page into skb */
6297 if (igb_add_rx_frag(rx_ring, buffer_info, rx_desc, skb)) {
6298 /* hand second half of page back to the ring */
6299 igb_reuse_rx_page(rx_ring, buffer_info);
6300 } else {
6301 /* we are not reusing the buffer so unmap it */
6302 dma_unmap_page(rx_ring->dev, buffer_info->dma,
6303 PAGE_SIZE, DMA_FROM_DEVICE);
6304 }
6305
6306 /* clear contents of buffer_info */
6307 buffer_info->page = NULL;
6308
6309 if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP))
6310 goto next_desc;
6311
6312 /* verify the packet layout is correct */
6313 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
6314 skb = NULL;
6315 continue;
6316 }
6317
6318 /* probably a little skewed due to removing CRC */
6319 total_bytes += skb->len;
6320 total_packets++;
6321
6322 /* populate checksum, timestamp, VLAN, and protocol */
6323 igb_process_skb_fields(rx_ring, rx_desc, skb);
6324
6325 napi_gro_receive(&q_vector->napi, skb);
6326
6327 /* reset skb pointer */
6328 skb = NULL;
6329
6330 budget--;
6331 next_desc:
6332 if (!budget)
6333 break;
6334
6335 cleaned_count++;
6336 /* return some buffers to hardware, one at a time is too slow */
6337 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6338 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6339 cleaned_count = 0;
6340 }
6341
6342 /* use prefetched values */
6343 rx_desc = next_rxd;
6344 }
6345
6346 /* place incomplete frames back on ring for completion */
6347 rx_ring->skb = skb;
6348
6349 rx_ring->next_to_clean = i;
6350 u64_stats_update_begin(&rx_ring->rx_syncp);
6351 rx_ring->rx_stats.packets += total_packets;
6352 rx_ring->rx_stats.bytes += total_bytes;
6353 u64_stats_update_end(&rx_ring->rx_syncp);
6354 q_vector->rx.total_packets += total_packets;
6355 q_vector->rx.total_bytes += total_bytes;
6356
6357 if (cleaned_count)
6358 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6359
6360 return !!budget;
6361 }
6362
6363 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
6364 struct igb_rx_buffer *bi)
6365 {
6366 struct page *page = bi->page;
6367 dma_addr_t dma;
6368
6369 /* since we are recycling buffers we should seldom need to alloc */
6370 if (likely(page))
6371 return true;
6372
6373 /* alloc new page for storage */
6374 page = __skb_alloc_page(GFP_ATOMIC | __GFP_COLD, NULL);
6375 if (unlikely(!page)) {
6376 rx_ring->rx_stats.alloc_failed++;
6377 return false;
6378 }
6379
6380 /* map page for use */
6381 dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
6382
6383 /*
6384 * if mapping failed free memory back to system since
6385 * there isn't much point in holding memory we can't use
6386 */
6387 if (dma_mapping_error(rx_ring->dev, dma)) {
6388 __free_page(page);
6389
6390 rx_ring->rx_stats.alloc_failed++;
6391 return false;
6392 }
6393
6394 bi->dma = dma;
6395 bi->page = page;
6396 bi->page_offset = 0;
6397
6398 return true;
6399 }
6400
6401 /**
6402 * igb_alloc_rx_buffers - Replace used receive buffers; packet split
6403 * @adapter: address of board private structure
6404 **/
6405 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
6406 {
6407 union e1000_adv_rx_desc *rx_desc;
6408 struct igb_rx_buffer *bi;
6409 u16 i = rx_ring->next_to_use;
6410
6411 /* nothing to do */
6412 if (!cleaned_count)
6413 return;
6414
6415 rx_desc = IGB_RX_DESC(rx_ring, i);
6416 bi = &rx_ring->rx_buffer_info[i];
6417 i -= rx_ring->count;
6418
6419 do {
6420 if (!igb_alloc_mapped_page(rx_ring, bi))
6421 break;
6422
6423 /*
6424 * Refresh the desc even if buffer_addrs didn't change
6425 * because each write-back erases this info.
6426 */
6427 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
6428
6429 rx_desc++;
6430 bi++;
6431 i++;
6432 if (unlikely(!i)) {
6433 rx_desc = IGB_RX_DESC(rx_ring, 0);
6434 bi = rx_ring->rx_buffer_info;
6435 i -= rx_ring->count;
6436 }
6437
6438 /* clear the hdr_addr for the next_to_use descriptor */
6439 rx_desc->read.hdr_addr = 0;
6440
6441 cleaned_count--;
6442 } while (cleaned_count);
6443
6444 i += rx_ring->count;
6445
6446 if (rx_ring->next_to_use != i) {
6447 /* record the next descriptor to use */
6448 rx_ring->next_to_use = i;
6449
6450 /* update next to alloc since we have filled the ring */
6451 rx_ring->next_to_alloc = i;
6452
6453 /*
6454 * Force memory writes to complete before letting h/w
6455 * know there are new descriptors to fetch. (Only
6456 * applicable for weak-ordered memory model archs,
6457 * such as IA-64).
6458 */
6459 wmb();
6460 writel(i, rx_ring->tail);
6461 }
6462 }
6463
6464 /**
6465 * igb_mii_ioctl -
6466 * @netdev:
6467 * @ifreq:
6468 * @cmd:
6469 **/
6470 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6471 {
6472 struct igb_adapter *adapter = netdev_priv(netdev);
6473 struct mii_ioctl_data *data = if_mii(ifr);
6474
6475 if (adapter->hw.phy.media_type != e1000_media_type_copper)
6476 return -EOPNOTSUPP;
6477
6478 switch (cmd) {
6479 case SIOCGMIIPHY:
6480 data->phy_id = adapter->hw.phy.addr;
6481 break;
6482 case SIOCGMIIREG:
6483 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
6484 &data->val_out))
6485 return -EIO;
6486 break;
6487 case SIOCSMIIREG:
6488 default:
6489 return -EOPNOTSUPP;
6490 }
6491 return 0;
6492 }
6493
6494 /**
6495 * igb_ioctl -
6496 * @netdev:
6497 * @ifreq:
6498 * @cmd:
6499 **/
6500 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6501 {
6502 switch (cmd) {
6503 case SIOCGMIIPHY:
6504 case SIOCGMIIREG:
6505 case SIOCSMIIREG:
6506 return igb_mii_ioctl(netdev, ifr, cmd);
6507 #ifdef CONFIG_IGB_PTP
6508 case SIOCSHWTSTAMP:
6509 return igb_ptp_hwtstamp_ioctl(netdev, ifr, cmd);
6510 #endif /* CONFIG_IGB_PTP */
6511 default:
6512 return -EOPNOTSUPP;
6513 }
6514 }
6515
6516 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6517 {
6518 struct igb_adapter *adapter = hw->back;
6519
6520 if (pcie_capability_read_word(adapter->pdev, reg, value))
6521 return -E1000_ERR_CONFIG;
6522
6523 return 0;
6524 }
6525
6526 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6527 {
6528 struct igb_adapter *adapter = hw->back;
6529
6530 if (pcie_capability_write_word(adapter->pdev, reg, *value))
6531 return -E1000_ERR_CONFIG;
6532
6533 return 0;
6534 }
6535
6536 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
6537 {
6538 struct igb_adapter *adapter = netdev_priv(netdev);
6539 struct e1000_hw *hw = &adapter->hw;
6540 u32 ctrl, rctl;
6541 bool enable = !!(features & NETIF_F_HW_VLAN_RX);
6542
6543 if (enable) {
6544 /* enable VLAN tag insert/strip */
6545 ctrl = rd32(E1000_CTRL);
6546 ctrl |= E1000_CTRL_VME;
6547 wr32(E1000_CTRL, ctrl);
6548
6549 /* Disable CFI check */
6550 rctl = rd32(E1000_RCTL);
6551 rctl &= ~E1000_RCTL_CFIEN;
6552 wr32(E1000_RCTL, rctl);
6553 } else {
6554 /* disable VLAN tag insert/strip */
6555 ctrl = rd32(E1000_CTRL);
6556 ctrl &= ~E1000_CTRL_VME;
6557 wr32(E1000_CTRL, ctrl);
6558 }
6559
6560 igb_rlpml_set(adapter);
6561 }
6562
6563 static int igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
6564 {
6565 struct igb_adapter *adapter = netdev_priv(netdev);
6566 struct e1000_hw *hw = &adapter->hw;
6567 int pf_id = adapter->vfs_allocated_count;
6568
6569 /* attempt to add filter to vlvf array */
6570 igb_vlvf_set(adapter, vid, true, pf_id);
6571
6572 /* add the filter since PF can receive vlans w/o entry in vlvf */
6573 igb_vfta_set(hw, vid, true);
6574
6575 set_bit(vid, adapter->active_vlans);
6576
6577 return 0;
6578 }
6579
6580 static int igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
6581 {
6582 struct igb_adapter *adapter = netdev_priv(netdev);
6583 struct e1000_hw *hw = &adapter->hw;
6584 int pf_id = adapter->vfs_allocated_count;
6585 s32 err;
6586
6587 /* remove vlan from VLVF table array */
6588 err = igb_vlvf_set(adapter, vid, false, pf_id);
6589
6590 /* if vid was not present in VLVF just remove it from table */
6591 if (err)
6592 igb_vfta_set(hw, vid, false);
6593
6594 clear_bit(vid, adapter->active_vlans);
6595
6596 return 0;
6597 }
6598
6599 static void igb_restore_vlan(struct igb_adapter *adapter)
6600 {
6601 u16 vid;
6602
6603 igb_vlan_mode(adapter->netdev, adapter->netdev->features);
6604
6605 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
6606 igb_vlan_rx_add_vid(adapter->netdev, vid);
6607 }
6608
6609 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
6610 {
6611 struct pci_dev *pdev = adapter->pdev;
6612 struct e1000_mac_info *mac = &adapter->hw.mac;
6613
6614 mac->autoneg = 0;
6615
6616 /* Make sure dplx is at most 1 bit and lsb of speed is not set
6617 * for the switch() below to work */
6618 if ((spd & 1) || (dplx & ~1))
6619 goto err_inval;
6620
6621 /* Fiber NIC's only allow 1000 Gbps Full duplex */
6622 if ((adapter->hw.phy.media_type == e1000_media_type_internal_serdes) &&
6623 spd != SPEED_1000 &&
6624 dplx != DUPLEX_FULL)
6625 goto err_inval;
6626
6627 switch (spd + dplx) {
6628 case SPEED_10 + DUPLEX_HALF:
6629 mac->forced_speed_duplex = ADVERTISE_10_HALF;
6630 break;
6631 case SPEED_10 + DUPLEX_FULL:
6632 mac->forced_speed_duplex = ADVERTISE_10_FULL;
6633 break;
6634 case SPEED_100 + DUPLEX_HALF:
6635 mac->forced_speed_duplex = ADVERTISE_100_HALF;
6636 break;
6637 case SPEED_100 + DUPLEX_FULL:
6638 mac->forced_speed_duplex = ADVERTISE_100_FULL;
6639 break;
6640 case SPEED_1000 + DUPLEX_FULL:
6641 mac->autoneg = 1;
6642 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
6643 break;
6644 case SPEED_1000 + DUPLEX_HALF: /* not supported */
6645 default:
6646 goto err_inval;
6647 }
6648
6649 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
6650 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6651
6652 return 0;
6653
6654 err_inval:
6655 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
6656 return -EINVAL;
6657 }
6658
6659 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
6660 bool runtime)
6661 {
6662 struct net_device *netdev = pci_get_drvdata(pdev);
6663 struct igb_adapter *adapter = netdev_priv(netdev);
6664 struct e1000_hw *hw = &adapter->hw;
6665 u32 ctrl, rctl, status;
6666 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
6667 #ifdef CONFIG_PM
6668 int retval = 0;
6669 #endif
6670
6671 netif_device_detach(netdev);
6672
6673 if (netif_running(netdev))
6674 __igb_close(netdev, true);
6675
6676 igb_clear_interrupt_scheme(adapter);
6677
6678 #ifdef CONFIG_PM
6679 retval = pci_save_state(pdev);
6680 if (retval)
6681 return retval;
6682 #endif
6683
6684 status = rd32(E1000_STATUS);
6685 if (status & E1000_STATUS_LU)
6686 wufc &= ~E1000_WUFC_LNKC;
6687
6688 if (wufc) {
6689 igb_setup_rctl(adapter);
6690 igb_set_rx_mode(netdev);
6691
6692 /* turn on all-multi mode if wake on multicast is enabled */
6693 if (wufc & E1000_WUFC_MC) {
6694 rctl = rd32(E1000_RCTL);
6695 rctl |= E1000_RCTL_MPE;
6696 wr32(E1000_RCTL, rctl);
6697 }
6698
6699 ctrl = rd32(E1000_CTRL);
6700 /* advertise wake from D3Cold */
6701 #define E1000_CTRL_ADVD3WUC 0x00100000
6702 /* phy power management enable */
6703 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
6704 ctrl |= E1000_CTRL_ADVD3WUC;
6705 wr32(E1000_CTRL, ctrl);
6706
6707 /* Allow time for pending master requests to run */
6708 igb_disable_pcie_master(hw);
6709
6710 wr32(E1000_WUC, E1000_WUC_PME_EN);
6711 wr32(E1000_WUFC, wufc);
6712 } else {
6713 wr32(E1000_WUC, 0);
6714 wr32(E1000_WUFC, 0);
6715 }
6716
6717 *enable_wake = wufc || adapter->en_mng_pt;
6718 if (!*enable_wake)
6719 igb_power_down_link(adapter);
6720 else
6721 igb_power_up_link(adapter);
6722
6723 /* Release control of h/w to f/w. If f/w is AMT enabled, this
6724 * would have already happened in close and is redundant. */
6725 igb_release_hw_control(adapter);
6726
6727 pci_disable_device(pdev);
6728
6729 return 0;
6730 }
6731
6732 #ifdef CONFIG_PM
6733 #ifdef CONFIG_PM_SLEEP
6734 static int igb_suspend(struct device *dev)
6735 {
6736 int retval;
6737 bool wake;
6738 struct pci_dev *pdev = to_pci_dev(dev);
6739
6740 retval = __igb_shutdown(pdev, &wake, 0);
6741 if (retval)
6742 return retval;
6743
6744 if (wake) {
6745 pci_prepare_to_sleep(pdev);
6746 } else {
6747 pci_wake_from_d3(pdev, false);
6748 pci_set_power_state(pdev, PCI_D3hot);
6749 }
6750
6751 return 0;
6752 }
6753 #endif /* CONFIG_PM_SLEEP */
6754
6755 static int igb_resume(struct device *dev)
6756 {
6757 struct pci_dev *pdev = to_pci_dev(dev);
6758 struct net_device *netdev = pci_get_drvdata(pdev);
6759 struct igb_adapter *adapter = netdev_priv(netdev);
6760 struct e1000_hw *hw = &adapter->hw;
6761 u32 err;
6762
6763 pci_set_power_state(pdev, PCI_D0);
6764 pci_restore_state(pdev);
6765 pci_save_state(pdev);
6766
6767 err = pci_enable_device_mem(pdev);
6768 if (err) {
6769 dev_err(&pdev->dev,
6770 "igb: Cannot enable PCI device from suspend\n");
6771 return err;
6772 }
6773 pci_set_master(pdev);
6774
6775 pci_enable_wake(pdev, PCI_D3hot, 0);
6776 pci_enable_wake(pdev, PCI_D3cold, 0);
6777
6778 if (igb_init_interrupt_scheme(adapter)) {
6779 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
6780 return -ENOMEM;
6781 }
6782
6783 igb_reset(adapter);
6784
6785 /* let the f/w know that the h/w is now under the control of the
6786 * driver. */
6787 igb_get_hw_control(adapter);
6788
6789 wr32(E1000_WUS, ~0);
6790
6791 if (netdev->flags & IFF_UP) {
6792 err = __igb_open(netdev, true);
6793 if (err)
6794 return err;
6795 }
6796
6797 netif_device_attach(netdev);
6798 return 0;
6799 }
6800
6801 #ifdef CONFIG_PM_RUNTIME
6802 static int igb_runtime_idle(struct device *dev)
6803 {
6804 struct pci_dev *pdev = to_pci_dev(dev);
6805 struct net_device *netdev = pci_get_drvdata(pdev);
6806 struct igb_adapter *adapter = netdev_priv(netdev);
6807
6808 if (!igb_has_link(adapter))
6809 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
6810
6811 return -EBUSY;
6812 }
6813
6814 static int igb_runtime_suspend(struct device *dev)
6815 {
6816 struct pci_dev *pdev = to_pci_dev(dev);
6817 int retval;
6818 bool wake;
6819
6820 retval = __igb_shutdown(pdev, &wake, 1);
6821 if (retval)
6822 return retval;
6823
6824 if (wake) {
6825 pci_prepare_to_sleep(pdev);
6826 } else {
6827 pci_wake_from_d3(pdev, false);
6828 pci_set_power_state(pdev, PCI_D3hot);
6829 }
6830
6831 return 0;
6832 }
6833
6834 static int igb_runtime_resume(struct device *dev)
6835 {
6836 return igb_resume(dev);
6837 }
6838 #endif /* CONFIG_PM_RUNTIME */
6839 #endif
6840
6841 static void igb_shutdown(struct pci_dev *pdev)
6842 {
6843 bool wake;
6844
6845 __igb_shutdown(pdev, &wake, 0);
6846
6847 if (system_state == SYSTEM_POWER_OFF) {
6848 pci_wake_from_d3(pdev, wake);
6849 pci_set_power_state(pdev, PCI_D3hot);
6850 }
6851 }
6852
6853 #ifdef CONFIG_NET_POLL_CONTROLLER
6854 /*
6855 * Polling 'interrupt' - used by things like netconsole to send skbs
6856 * without having to re-enable interrupts. It's not called while
6857 * the interrupt routine is executing.
6858 */
6859 static void igb_netpoll(struct net_device *netdev)
6860 {
6861 struct igb_adapter *adapter = netdev_priv(netdev);
6862 struct e1000_hw *hw = &adapter->hw;
6863 struct igb_q_vector *q_vector;
6864 int i;
6865
6866 for (i = 0; i < adapter->num_q_vectors; i++) {
6867 q_vector = adapter->q_vector[i];
6868 if (adapter->msix_entries)
6869 wr32(E1000_EIMC, q_vector->eims_value);
6870 else
6871 igb_irq_disable(adapter);
6872 napi_schedule(&q_vector->napi);
6873 }
6874 }
6875 #endif /* CONFIG_NET_POLL_CONTROLLER */
6876
6877 /**
6878 * igb_io_error_detected - called when PCI error is detected
6879 * @pdev: Pointer to PCI device
6880 * @state: The current pci connection state
6881 *
6882 * This function is called after a PCI bus error affecting
6883 * this device has been detected.
6884 */
6885 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
6886 pci_channel_state_t state)
6887 {
6888 struct net_device *netdev = pci_get_drvdata(pdev);
6889 struct igb_adapter *adapter = netdev_priv(netdev);
6890
6891 netif_device_detach(netdev);
6892
6893 if (state == pci_channel_io_perm_failure)
6894 return PCI_ERS_RESULT_DISCONNECT;
6895
6896 if (netif_running(netdev))
6897 igb_down(adapter);
6898 pci_disable_device(pdev);
6899
6900 /* Request a slot slot reset. */
6901 return PCI_ERS_RESULT_NEED_RESET;
6902 }
6903
6904 /**
6905 * igb_io_slot_reset - called after the pci bus has been reset.
6906 * @pdev: Pointer to PCI device
6907 *
6908 * Restart the card from scratch, as if from a cold-boot. Implementation
6909 * resembles the first-half of the igb_resume routine.
6910 */
6911 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
6912 {
6913 struct net_device *netdev = pci_get_drvdata(pdev);
6914 struct igb_adapter *adapter = netdev_priv(netdev);
6915 struct e1000_hw *hw = &adapter->hw;
6916 pci_ers_result_t result;
6917 int err;
6918
6919 if (pci_enable_device_mem(pdev)) {
6920 dev_err(&pdev->dev,
6921 "Cannot re-enable PCI device after reset.\n");
6922 result = PCI_ERS_RESULT_DISCONNECT;
6923 } else {
6924 pci_set_master(pdev);
6925 pci_restore_state(pdev);
6926 pci_save_state(pdev);
6927
6928 pci_enable_wake(pdev, PCI_D3hot, 0);
6929 pci_enable_wake(pdev, PCI_D3cold, 0);
6930
6931 igb_reset(adapter);
6932 wr32(E1000_WUS, ~0);
6933 result = PCI_ERS_RESULT_RECOVERED;
6934 }
6935
6936 err = pci_cleanup_aer_uncorrect_error_status(pdev);
6937 if (err) {
6938 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
6939 "failed 0x%0x\n", err);
6940 /* non-fatal, continue */
6941 }
6942
6943 return result;
6944 }
6945
6946 /**
6947 * igb_io_resume - called when traffic can start flowing again.
6948 * @pdev: Pointer to PCI device
6949 *
6950 * This callback is called when the error recovery driver tells us that
6951 * its OK to resume normal operation. Implementation resembles the
6952 * second-half of the igb_resume routine.
6953 */
6954 static void igb_io_resume(struct pci_dev *pdev)
6955 {
6956 struct net_device *netdev = pci_get_drvdata(pdev);
6957 struct igb_adapter *adapter = netdev_priv(netdev);
6958
6959 if (netif_running(netdev)) {
6960 if (igb_up(adapter)) {
6961 dev_err(&pdev->dev, "igb_up failed after reset\n");
6962 return;
6963 }
6964 }
6965
6966 netif_device_attach(netdev);
6967
6968 /* let the f/w know that the h/w is now under the control of the
6969 * driver. */
6970 igb_get_hw_control(adapter);
6971 }
6972
6973 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
6974 u8 qsel)
6975 {
6976 u32 rar_low, rar_high;
6977 struct e1000_hw *hw = &adapter->hw;
6978
6979 /* HW expects these in little endian so we reverse the byte order
6980 * from network order (big endian) to little endian
6981 */
6982 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
6983 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
6984 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
6985
6986 /* Indicate to hardware the Address is Valid. */
6987 rar_high |= E1000_RAH_AV;
6988
6989 if (hw->mac.type == e1000_82575)
6990 rar_high |= E1000_RAH_POOL_1 * qsel;
6991 else
6992 rar_high |= E1000_RAH_POOL_1 << qsel;
6993
6994 wr32(E1000_RAL(index), rar_low);
6995 wrfl();
6996 wr32(E1000_RAH(index), rar_high);
6997 wrfl();
6998 }
6999
7000 static int igb_set_vf_mac(struct igb_adapter *adapter,
7001 int vf, unsigned char *mac_addr)
7002 {
7003 struct e1000_hw *hw = &adapter->hw;
7004 /* VF MAC addresses start at end of receive addresses and moves
7005 * torwards the first, as a result a collision should not be possible */
7006 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
7007
7008 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
7009
7010 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
7011
7012 return 0;
7013 }
7014
7015 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
7016 {
7017 struct igb_adapter *adapter = netdev_priv(netdev);
7018 if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
7019 return -EINVAL;
7020 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
7021 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
7022 dev_info(&adapter->pdev->dev, "Reload the VF driver to make this"
7023 " change effective.");
7024 if (test_bit(__IGB_DOWN, &adapter->state)) {
7025 dev_warn(&adapter->pdev->dev, "The VF MAC address has been set,"
7026 " but the PF device is not up.\n");
7027 dev_warn(&adapter->pdev->dev, "Bring the PF device up before"
7028 " attempting to use the VF device.\n");
7029 }
7030 return igb_set_vf_mac(adapter, vf, mac);
7031 }
7032
7033 static int igb_link_mbps(int internal_link_speed)
7034 {
7035 switch (internal_link_speed) {
7036 case SPEED_100:
7037 return 100;
7038 case SPEED_1000:
7039 return 1000;
7040 default:
7041 return 0;
7042 }
7043 }
7044
7045 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
7046 int link_speed)
7047 {
7048 int rf_dec, rf_int;
7049 u32 bcnrc_val;
7050
7051 if (tx_rate != 0) {
7052 /* Calculate the rate factor values to set */
7053 rf_int = link_speed / tx_rate;
7054 rf_dec = (link_speed - (rf_int * tx_rate));
7055 rf_dec = (rf_dec * (1<<E1000_RTTBCNRC_RF_INT_SHIFT)) / tx_rate;
7056
7057 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
7058 bcnrc_val |= ((rf_int<<E1000_RTTBCNRC_RF_INT_SHIFT) &
7059 E1000_RTTBCNRC_RF_INT_MASK);
7060 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
7061 } else {
7062 bcnrc_val = 0;
7063 }
7064
7065 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7066 /*
7067 * Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7068 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7069 */
7070 wr32(E1000_RTTBCNRM, 0x14);
7071 wr32(E1000_RTTBCNRC, bcnrc_val);
7072 }
7073
7074 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7075 {
7076 int actual_link_speed, i;
7077 bool reset_rate = false;
7078
7079 /* VF TX rate limit was not set or not supported */
7080 if ((adapter->vf_rate_link_speed == 0) ||
7081 (adapter->hw.mac.type != e1000_82576))
7082 return;
7083
7084 actual_link_speed = igb_link_mbps(adapter->link_speed);
7085 if (actual_link_speed != adapter->vf_rate_link_speed) {
7086 reset_rate = true;
7087 adapter->vf_rate_link_speed = 0;
7088 dev_info(&adapter->pdev->dev,
7089 "Link speed has been changed. VF Transmit "
7090 "rate is disabled\n");
7091 }
7092
7093 for (i = 0; i < adapter->vfs_allocated_count; i++) {
7094 if (reset_rate)
7095 adapter->vf_data[i].tx_rate = 0;
7096
7097 igb_set_vf_rate_limit(&adapter->hw, i,
7098 adapter->vf_data[i].tx_rate,
7099 actual_link_speed);
7100 }
7101 }
7102
7103 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
7104 {
7105 struct igb_adapter *adapter = netdev_priv(netdev);
7106 struct e1000_hw *hw = &adapter->hw;
7107 int actual_link_speed;
7108
7109 if (hw->mac.type != e1000_82576)
7110 return -EOPNOTSUPP;
7111
7112 actual_link_speed = igb_link_mbps(adapter->link_speed);
7113 if ((vf >= adapter->vfs_allocated_count) ||
7114 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7115 (tx_rate < 0) || (tx_rate > actual_link_speed))
7116 return -EINVAL;
7117
7118 adapter->vf_rate_link_speed = actual_link_speed;
7119 adapter->vf_data[vf].tx_rate = (u16)tx_rate;
7120 igb_set_vf_rate_limit(hw, vf, tx_rate, actual_link_speed);
7121
7122 return 0;
7123 }
7124
7125 static int igb_ndo_get_vf_config(struct net_device *netdev,
7126 int vf, struct ifla_vf_info *ivi)
7127 {
7128 struct igb_adapter *adapter = netdev_priv(netdev);
7129 if (vf >= adapter->vfs_allocated_count)
7130 return -EINVAL;
7131 ivi->vf = vf;
7132 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7133 ivi->tx_rate = adapter->vf_data[vf].tx_rate;
7134 ivi->vlan = adapter->vf_data[vf].pf_vlan;
7135 ivi->qos = adapter->vf_data[vf].pf_qos;
7136 return 0;
7137 }
7138
7139 static void igb_vmm_control(struct igb_adapter *adapter)
7140 {
7141 struct e1000_hw *hw = &adapter->hw;
7142 u32 reg;
7143
7144 switch (hw->mac.type) {
7145 case e1000_82575:
7146 case e1000_i210:
7147 case e1000_i211:
7148 default:
7149 /* replication is not supported for 82575 */
7150 return;
7151 case e1000_82576:
7152 /* notify HW that the MAC is adding vlan tags */
7153 reg = rd32(E1000_DTXCTL);
7154 reg |= E1000_DTXCTL_VLAN_ADDED;
7155 wr32(E1000_DTXCTL, reg);
7156 case e1000_82580:
7157 /* enable replication vlan tag stripping */
7158 reg = rd32(E1000_RPLOLR);
7159 reg |= E1000_RPLOLR_STRVLAN;
7160 wr32(E1000_RPLOLR, reg);
7161 case e1000_i350:
7162 /* none of the above registers are supported by i350 */
7163 break;
7164 }
7165
7166 if (adapter->vfs_allocated_count) {
7167 igb_vmdq_set_loopback_pf(hw, true);
7168 igb_vmdq_set_replication_pf(hw, true);
7169 igb_vmdq_set_anti_spoofing_pf(hw, true,
7170 adapter->vfs_allocated_count);
7171 } else {
7172 igb_vmdq_set_loopback_pf(hw, false);
7173 igb_vmdq_set_replication_pf(hw, false);
7174 }
7175 }
7176
7177 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7178 {
7179 struct e1000_hw *hw = &adapter->hw;
7180 u32 dmac_thr;
7181 u16 hwm;
7182
7183 if (hw->mac.type > e1000_82580) {
7184 if (adapter->flags & IGB_FLAG_DMAC) {
7185 u32 reg;
7186
7187 /* force threshold to 0. */
7188 wr32(E1000_DMCTXTH, 0);
7189
7190 /*
7191 * DMA Coalescing high water mark needs to be greater
7192 * than the Rx threshold. Set hwm to PBA - max frame
7193 * size in 16B units, capping it at PBA - 6KB.
7194 */
7195 hwm = 64 * pba - adapter->max_frame_size / 16;
7196 if (hwm < 64 * (pba - 6))
7197 hwm = 64 * (pba - 6);
7198 reg = rd32(E1000_FCRTC);
7199 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
7200 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
7201 & E1000_FCRTC_RTH_COAL_MASK);
7202 wr32(E1000_FCRTC, reg);
7203
7204 /*
7205 * Set the DMA Coalescing Rx threshold to PBA - 2 * max
7206 * frame size, capping it at PBA - 10KB.
7207 */
7208 dmac_thr = pba - adapter->max_frame_size / 512;
7209 if (dmac_thr < pba - 10)
7210 dmac_thr = pba - 10;
7211 reg = rd32(E1000_DMACR);
7212 reg &= ~E1000_DMACR_DMACTHR_MASK;
7213 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
7214 & E1000_DMACR_DMACTHR_MASK);
7215
7216 /* transition to L0x or L1 if available..*/
7217 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
7218
7219 /* watchdog timer= +-1000 usec in 32usec intervals */
7220 reg |= (1000 >> 5);
7221
7222 /* Disable BMC-to-OS Watchdog Enable */
7223 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
7224 wr32(E1000_DMACR, reg);
7225
7226 /*
7227 * no lower threshold to disable
7228 * coalescing(smart fifb)-UTRESH=0
7229 */
7230 wr32(E1000_DMCRTRH, 0);
7231
7232 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
7233
7234 wr32(E1000_DMCTLX, reg);
7235
7236 /*
7237 * free space in tx packet buffer to wake from
7238 * DMA coal
7239 */
7240 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
7241 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
7242
7243 /*
7244 * make low power state decision controlled
7245 * by DMA coal
7246 */
7247 reg = rd32(E1000_PCIEMISC);
7248 reg &= ~E1000_PCIEMISC_LX_DECISION;
7249 wr32(E1000_PCIEMISC, reg);
7250 } /* endif adapter->dmac is not disabled */
7251 } else if (hw->mac.type == e1000_82580) {
7252 u32 reg = rd32(E1000_PCIEMISC);
7253 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
7254 wr32(E1000_DMACR, 0);
7255 }
7256 }
7257
7258 /* igb_main.c */
CJK support for tty framebuffer vt