?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6.git] / drivers / net / igb / igb_main.c
blobbb1a6eeade06c0c001b2feaf6f2d80857313a704
1 /*******************************************************************************
2
3 Intel(R) Gigabit Ethernet Linux driver
4 Copyright(c) 2007-2009 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 #include <linux/module.h>
29 #include <linux/types.h>
30 #include <linux/init.h>
31 #include <linux/vmalloc.h>
32 #include <linux/pagemap.h>
33 #include <linux/netdevice.h>
34 #include <linux/ipv6.h>
35 #include <net/checksum.h>
36 #include <net/ip6_checksum.h>
37 #include <linux/net_tstamp.h>
38 #include <linux/mii.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_vlan.h>
41 #include <linux/pci.h>
42 #include <linux/pci-aspm.h>
43 #include <linux/delay.h>
44 #include <linux/interrupt.h>
45 #include <linux/if_ether.h>
46 #include <linux/aer.h>
47 #ifdef CONFIG_IGB_DCA
48 #include <linux/dca.h>
49 #endif
50 #include "igb.h"
51
52 #define DRV_VERSION "2.1.0-k2"
53 char igb_driver_name[] = "igb";
54 char igb_driver_version[] = DRV_VERSION;
55 static const char igb_driver_string[] =
56 "Intel(R) Gigabit Ethernet Network Driver";
57 static const char igb_copyright[] = "Copyright (c) 2007-2009 Intel Corporation.";
58
59 static const struct e1000_info *igb_info_tbl[] = {
60 [board_82575] = &e1000_82575_info,
61 };
62
63 static struct pci_device_id igb_pci_tbl[] = {
64 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
65 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
66 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
67 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
68 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
69 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
70 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
71 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
72 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
73 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
74 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
75 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
76 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
77 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
79 /* required last entry */
80 {0, }
81 };
82
83 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
84
85 void igb_reset(struct igb_adapter *);
86 static int igb_setup_all_tx_resources(struct igb_adapter *);
87 static int igb_setup_all_rx_resources(struct igb_adapter *);
88 static void igb_free_all_tx_resources(struct igb_adapter *);
89 static void igb_free_all_rx_resources(struct igb_adapter *);
90 static void igb_setup_mrqc(struct igb_adapter *);
91 void igb_update_stats(struct igb_adapter *);
92 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
93 static void __devexit igb_remove(struct pci_dev *pdev);
94 static int igb_sw_init(struct igb_adapter *);
95 static int igb_open(struct net_device *);
96 static int igb_close(struct net_device *);
97 static void igb_configure_tx(struct igb_adapter *);
98 static void igb_configure_rx(struct igb_adapter *);
99 static void igb_clean_all_tx_rings(struct igb_adapter *);
100 static void igb_clean_all_rx_rings(struct igb_adapter *);
101 static void igb_clean_tx_ring(struct igb_ring *);
102 static void igb_clean_rx_ring(struct igb_ring *);
103 static void igb_set_rx_mode(struct net_device *);
104 static void igb_update_phy_info(unsigned long);
105 static void igb_watchdog(unsigned long);
106 static void igb_watchdog_task(struct work_struct *);
107 static netdev_tx_t igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
108 static struct net_device_stats *igb_get_stats(struct net_device *);
109 static int igb_change_mtu(struct net_device *, int);
110 static int igb_set_mac(struct net_device *, void *);
111 static void igb_set_uta(struct igb_adapter *adapter);
112 static irqreturn_t igb_intr(int irq, void *);
113 static irqreturn_t igb_intr_msi(int irq, void *);
114 static irqreturn_t igb_msix_other(int irq, void *);
115 static irqreturn_t igb_msix_ring(int irq, void *);
116 #ifdef CONFIG_IGB_DCA
117 static void igb_update_dca(struct igb_q_vector *);
118 static void igb_setup_dca(struct igb_adapter *);
119 #endif /* CONFIG_IGB_DCA */
120 static bool igb_clean_tx_irq(struct igb_q_vector *);
121 static int igb_poll(struct napi_struct *, int);
122 static bool igb_clean_rx_irq_adv(struct igb_q_vector *, int *, int);
123 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
124 static void igb_tx_timeout(struct net_device *);
125 static void igb_reset_task(struct work_struct *);
126 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
127 static void igb_vlan_rx_add_vid(struct net_device *, u16);
128 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
129 static void igb_restore_vlan(struct igb_adapter *);
130 static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
131 static void igb_ping_all_vfs(struct igb_adapter *);
132 static void igb_msg_task(struct igb_adapter *);
133 static void igb_vmm_control(struct igb_adapter *);
134 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
135 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
136
137 #ifdef CONFIG_PM
138 static int igb_suspend(struct pci_dev *, pm_message_t);
139 static int igb_resume(struct pci_dev *);
140 #endif
141 static void igb_shutdown(struct pci_dev *);
142 #ifdef CONFIG_IGB_DCA
143 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
144 static struct notifier_block dca_notifier = {
145 .notifier_call = igb_notify_dca,
146 .next = NULL,
147 .priority = 0
148 };
149 #endif
150 #ifdef CONFIG_NET_POLL_CONTROLLER
151 /* for netdump / net console */
152 static void igb_netpoll(struct net_device *);
153 #endif
154 #ifdef CONFIG_PCI_IOV
155 static unsigned int max_vfs = 0;
156 module_param(max_vfs, uint, 0);
157 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
158 "per physical function");
159 #endif /* CONFIG_PCI_IOV */
160
161 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
162 pci_channel_state_t);
163 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
164 static void igb_io_resume(struct pci_dev *);
165
166 static struct pci_error_handlers igb_err_handler = {
167 .error_detected = igb_io_error_detected,
168 .slot_reset = igb_io_slot_reset,
169 .resume = igb_io_resume,
170 };
171
172
173 static struct pci_driver igb_driver = {
174 .name = igb_driver_name,
175 .id_table = igb_pci_tbl,
176 .probe = igb_probe,
177 .remove = __devexit_p(igb_remove),
178 #ifdef CONFIG_PM
179 /* Power Managment Hooks */
180 .suspend = igb_suspend,
181 .resume = igb_resume,
182 #endif
183 .shutdown = igb_shutdown,
184 .err_handler = &igb_err_handler
185 };
186
187 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
188 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
189 MODULE_LICENSE("GPL");
190 MODULE_VERSION(DRV_VERSION);
191
192 /**
193 * igb_read_clock - read raw cycle counter (to be used by time counter)
194 */
195 static cycle_t igb_read_clock(const struct cyclecounter *tc)
196 {
197 struct igb_adapter *adapter =
198 container_of(tc, struct igb_adapter, cycles);
199 struct e1000_hw *hw = &adapter->hw;
200 u64 stamp = 0;
201 int shift = 0;
202
203 /*
204 * The timestamp latches on lowest register read. For the 82580
205 * the lowest register is SYSTIMR instead of SYSTIML. However we never
206 * adjusted TIMINCA so SYSTIMR will just read as all 0s so ignore it.
207 */
208 if (hw->mac.type == e1000_82580) {
209 stamp = rd32(E1000_SYSTIMR) >> 8;
210 shift = IGB_82580_TSYNC_SHIFT;
211 }
212
213 stamp |= (u64)rd32(E1000_SYSTIML) << shift;
214 stamp |= (u64)rd32(E1000_SYSTIMH) << (shift + 32);
215 return stamp;
216 }
217
218 #ifdef DEBUG
219 /**
220 * igb_get_hw_dev_name - return device name string
221 * used by hardware layer to print debugging information
222 **/
223 char *igb_get_hw_dev_name(struct e1000_hw *hw)
224 {
225 struct igb_adapter *adapter = hw->back;
226 return adapter->netdev->name;
227 }
228
229 /**
230 * igb_get_time_str - format current NIC and system time as string
231 */
232 static char *igb_get_time_str(struct igb_adapter *adapter,
233 char buffer[160])
234 {
235 cycle_t hw = adapter->cycles.read(&adapter->cycles);
236 struct timespec nic = ns_to_timespec(timecounter_read(&adapter->clock));
237 struct timespec sys;
238 struct timespec delta;
239 getnstimeofday(&sys);
240
241 delta = timespec_sub(nic, sys);
242
243 sprintf(buffer,
244 "HW %llu, NIC %ld.%09lus, SYS %ld.%09lus, NIC-SYS %lds + %09luns",
245 hw,
246 (long)nic.tv_sec, nic.tv_nsec,
247 (long)sys.tv_sec, sys.tv_nsec,
248 (long)delta.tv_sec, delta.tv_nsec);
249
250 return buffer;
251 }
252 #endif
253
254 /**
255 * igb_init_module - Driver Registration Routine
256 *
257 * igb_init_module is the first routine called when the driver is
258 * loaded. All it does is register with the PCI subsystem.
259 **/
260 static int __init igb_init_module(void)
261 {
262 int ret;
263 printk(KERN_INFO "%s - version %s\n",
264 igb_driver_string, igb_driver_version);
265
266 printk(KERN_INFO "%s\n", igb_copyright);
267
268 #ifdef CONFIG_IGB_DCA
269 dca_register_notify(&dca_notifier);
270 #endif
271 ret = pci_register_driver(&igb_driver);
272 return ret;
273 }
274
275 module_init(igb_init_module);
276
277 /**
278 * igb_exit_module - Driver Exit Cleanup Routine
279 *
280 * igb_exit_module is called just before the driver is removed
281 * from memory.
282 **/
283 static void __exit igb_exit_module(void)
284 {
285 #ifdef CONFIG_IGB_DCA
286 dca_unregister_notify(&dca_notifier);
287 #endif
288 pci_unregister_driver(&igb_driver);
289 }
290
291 module_exit(igb_exit_module);
292
293 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
294 /**
295 * igb_cache_ring_register - Descriptor ring to register mapping
296 * @adapter: board private structure to initialize
297 *
298 * Once we know the feature-set enabled for the device, we'll cache
299 * the register offset the descriptor ring is assigned to.
300 **/
301 static void igb_cache_ring_register(struct igb_adapter *adapter)
302 {
303 int i = 0, j = 0;
304 u32 rbase_offset = adapter->vfs_allocated_count;
305
306 switch (adapter->hw.mac.type) {
307 case e1000_82576:
308 /* The queues are allocated for virtualization such that VF 0
309 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
310 * In order to avoid collision we start at the first free queue
311 * and continue consuming queues in the same sequence
312 */
313 if (adapter->vfs_allocated_count) {
314 for (; i < adapter->rss_queues; i++)
315 adapter->rx_ring[i].reg_idx = rbase_offset +
316 Q_IDX_82576(i);
317 for (; j < adapter->rss_queues; j++)
318 adapter->tx_ring[j].reg_idx = rbase_offset +
319 Q_IDX_82576(j);
320 }
321 case e1000_82575:
322 case e1000_82580:
323 default:
324 for (; i < adapter->num_rx_queues; i++)
325 adapter->rx_ring[i].reg_idx = rbase_offset + i;
326 for (; j < adapter->num_tx_queues; j++)
327 adapter->tx_ring[j].reg_idx = rbase_offset + j;
328 break;
329 }
330 }
331
332 static void igb_free_queues(struct igb_adapter *adapter)
333 {
334 kfree(adapter->tx_ring);
335 kfree(adapter->rx_ring);
336
337 adapter->tx_ring = NULL;
338 adapter->rx_ring = NULL;
339
340 adapter->num_rx_queues = 0;
341 adapter->num_tx_queues = 0;
342 }
343
344 /**
345 * igb_alloc_queues - Allocate memory for all rings
346 * @adapter: board private structure to initialize
347 *
348 * We allocate one ring per queue at run-time since we don't know the
349 * number of queues at compile-time.
350 **/
351 static int igb_alloc_queues(struct igb_adapter *adapter)
352 {
353 int i;
354
355 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
356 sizeof(struct igb_ring), GFP_KERNEL);
357 if (!adapter->tx_ring)
358 goto err;
359
360 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
361 sizeof(struct igb_ring), GFP_KERNEL);
362 if (!adapter->rx_ring)
363 goto err;
364
365 for (i = 0; i < adapter->num_tx_queues; i++) {
366 struct igb_ring *ring = &(adapter->tx_ring[i]);
367 ring->count = adapter->tx_ring_count;
368 ring->queue_index = i;
369 ring->pdev = adapter->pdev;
370 ring->netdev = adapter->netdev;
371 /* For 82575, context index must be unique per ring. */
372 if (adapter->hw.mac.type == e1000_82575)
373 ring->flags = IGB_RING_FLAG_TX_CTX_IDX;
374 }
375
376 for (i = 0; i < adapter->num_rx_queues; i++) {
377 struct igb_ring *ring = &(adapter->rx_ring[i]);
378 ring->count = adapter->rx_ring_count;
379 ring->queue_index = i;
380 ring->pdev = adapter->pdev;
381 ring->netdev = adapter->netdev;
382 ring->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
383 ring->flags = IGB_RING_FLAG_RX_CSUM; /* enable rx checksum */
384 /* set flag indicating ring supports SCTP checksum offload */
385 if (adapter->hw.mac.type >= e1000_82576)
386 ring->flags |= IGB_RING_FLAG_RX_SCTP_CSUM;
387 }
388
389 igb_cache_ring_register(adapter);
390
391 return 0;
392
393 err:
394 igb_free_queues(adapter);
395
396 return -ENOMEM;
397 }
398
399 #define IGB_N0_QUEUE -1
400 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
401 {
402 u32 msixbm = 0;
403 struct igb_adapter *adapter = q_vector->adapter;
404 struct e1000_hw *hw = &adapter->hw;
405 u32 ivar, index;
406 int rx_queue = IGB_N0_QUEUE;
407 int tx_queue = IGB_N0_QUEUE;
408
409 if (q_vector->rx_ring)
410 rx_queue = q_vector->rx_ring->reg_idx;
411 if (q_vector->tx_ring)
412 tx_queue = q_vector->tx_ring->reg_idx;
413
414 switch (hw->mac.type) {
415 case e1000_82575:
416 /* The 82575 assigns vectors using a bitmask, which matches the
417 bitmask for the EICR/EIMS/EIMC registers. To assign one
418 or more queues to a vector, we write the appropriate bits
419 into the MSIXBM register for that vector. */
420 if (rx_queue > IGB_N0_QUEUE)
421 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
422 if (tx_queue > IGB_N0_QUEUE)
423 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
424 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
425 q_vector->eims_value = msixbm;
426 break;
427 case e1000_82576:
428 /* 82576 uses a table-based method for assigning vectors.
429 Each queue has a single entry in the table to which we write
430 a vector number along with a "valid" bit. Sadly, the layout
431 of the table is somewhat counterintuitive. */
432 if (rx_queue > IGB_N0_QUEUE) {
433 index = (rx_queue & 0x7);
434 ivar = array_rd32(E1000_IVAR0, index);
435 if (rx_queue < 8) {
436 /* vector goes into low byte of register */
437 ivar = ivar & 0xFFFFFF00;
438 ivar |= msix_vector | E1000_IVAR_VALID;
439 } else {
440 /* vector goes into third byte of register */
441 ivar = ivar & 0xFF00FFFF;
442 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
443 }
444 array_wr32(E1000_IVAR0, index, ivar);
445 }
446 if (tx_queue > IGB_N0_QUEUE) {
447 index = (tx_queue & 0x7);
448 ivar = array_rd32(E1000_IVAR0, index);
449 if (tx_queue < 8) {
450 /* vector goes into second byte of register */
451 ivar = ivar & 0xFFFF00FF;
452 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
453 } else {
454 /* vector goes into high byte of register */
455 ivar = ivar & 0x00FFFFFF;
456 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
457 }
458 array_wr32(E1000_IVAR0, index, ivar);
459 }
460 q_vector->eims_value = 1 << msix_vector;
461 break;
462 case e1000_82580:
463 /* 82580 uses the same table-based approach as 82576 but has fewer
464 entries as a result we carry over for queues greater than 4. */
465 if (rx_queue > IGB_N0_QUEUE) {
466 index = (rx_queue >> 1);
467 ivar = array_rd32(E1000_IVAR0, index);
468 if (rx_queue & 0x1) {
469 /* vector goes into third byte of register */
470 ivar = ivar & 0xFF00FFFF;
471 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
472 } else {
473 /* vector goes into low byte of register */
474 ivar = ivar & 0xFFFFFF00;
475 ivar |= msix_vector | E1000_IVAR_VALID;
476 }
477 array_wr32(E1000_IVAR0, index, ivar);
478 }
479 if (tx_queue > IGB_N0_QUEUE) {
480 index = (tx_queue >> 1);
481 ivar = array_rd32(E1000_IVAR0, index);
482 if (tx_queue & 0x1) {
483 /* vector goes into high byte of register */
484 ivar = ivar & 0x00FFFFFF;
485 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
486 } else {
487 /* vector goes into second byte of register */
488 ivar = ivar & 0xFFFF00FF;
489 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
490 }
491 array_wr32(E1000_IVAR0, index, ivar);
492 }
493 q_vector->eims_value = 1 << msix_vector;
494 break;
495 default:
496 BUG();
497 break;
498 }
499 }
500
501 /**
502 * igb_configure_msix - Configure MSI-X hardware
503 *
504 * igb_configure_msix sets up the hardware to properly
505 * generate MSI-X interrupts.
506 **/
507 static void igb_configure_msix(struct igb_adapter *adapter)
508 {
509 u32 tmp;
510 int i, vector = 0;
511 struct e1000_hw *hw = &adapter->hw;
512
513 adapter->eims_enable_mask = 0;
514
515 /* set vector for other causes, i.e. link changes */
516 switch (hw->mac.type) {
517 case e1000_82575:
518 tmp = rd32(E1000_CTRL_EXT);
519 /* enable MSI-X PBA support*/
520 tmp |= E1000_CTRL_EXT_PBA_CLR;
521
522 /* Auto-Mask interrupts upon ICR read. */
523 tmp |= E1000_CTRL_EXT_EIAME;
524 tmp |= E1000_CTRL_EXT_IRCA;
525
526 wr32(E1000_CTRL_EXT, tmp);
527
528 /* enable msix_other interrupt */
529 array_wr32(E1000_MSIXBM(0), vector++,
530 E1000_EIMS_OTHER);
531 adapter->eims_other = E1000_EIMS_OTHER;
532
533 break;
534
535 case e1000_82576:
536 case e1000_82580:
537 /* Turn on MSI-X capability first, or our settings
538 * won't stick. And it will take days to debug. */
539 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
540 E1000_GPIE_PBA | E1000_GPIE_EIAME |
541 E1000_GPIE_NSICR);
542
543 /* enable msix_other interrupt */
544 adapter->eims_other = 1 << vector;
545 tmp = (vector++ | E1000_IVAR_VALID) << 8;
546
547 wr32(E1000_IVAR_MISC, tmp);
548 break;
549 default:
550 /* do nothing, since nothing else supports MSI-X */
551 break;
552 } /* switch (hw->mac.type) */
553
554 adapter->eims_enable_mask |= adapter->eims_other;
555
556 for (i = 0; i < adapter->num_q_vectors; i++) {
557 struct igb_q_vector *q_vector = adapter->q_vector[i];
558 igb_assign_vector(q_vector, vector++);
559 adapter->eims_enable_mask |= q_vector->eims_value;
560 }
561
562 wrfl();
563 }
564
565 /**
566 * igb_request_msix - Initialize MSI-X interrupts
567 *
568 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
569 * kernel.
570 **/
571 static int igb_request_msix(struct igb_adapter *adapter)
572 {
573 struct net_device *netdev = adapter->netdev;
574 struct e1000_hw *hw = &adapter->hw;
575 int i, err = 0, vector = 0;
576
577 err = request_irq(adapter->msix_entries[vector].vector,
578 igb_msix_other, 0, netdev->name, adapter);
579 if (err)
580 goto out;
581 vector++;
582
583 for (i = 0; i < adapter->num_q_vectors; i++) {
584 struct igb_q_vector *q_vector = adapter->q_vector[i];
585
586 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
587
588 if (q_vector->rx_ring && q_vector->tx_ring)
589 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
590 q_vector->rx_ring->queue_index);
591 else if (q_vector->tx_ring)
592 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
593 q_vector->tx_ring->queue_index);
594 else if (q_vector->rx_ring)
595 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
596 q_vector->rx_ring->queue_index);
597 else
598 sprintf(q_vector->name, "%s-unused", netdev->name);
599
600 err = request_irq(adapter->msix_entries[vector].vector,
601 igb_msix_ring, 0, q_vector->name,
602 q_vector);
603 if (err)
604 goto out;
605 vector++;
606 }
607
608 igb_configure_msix(adapter);
609 return 0;
610 out:
611 return err;
612 }
613
614 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
615 {
616 if (adapter->msix_entries) {
617 pci_disable_msix(adapter->pdev);
618 kfree(adapter->msix_entries);
619 adapter->msix_entries = NULL;
620 } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
621 pci_disable_msi(adapter->pdev);
622 }
623 }
624
625 /**
626 * igb_free_q_vectors - Free memory allocated for interrupt vectors
627 * @adapter: board private structure to initialize
628 *
629 * This function frees the memory allocated to the q_vectors. In addition if
630 * NAPI is enabled it will delete any references to the NAPI struct prior
631 * to freeing the q_vector.
632 **/
633 static void igb_free_q_vectors(struct igb_adapter *adapter)
634 {
635 int v_idx;
636
637 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
638 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
639 adapter->q_vector[v_idx] = NULL;
640 netif_napi_del(&q_vector->napi);
641 kfree(q_vector);
642 }
643 adapter->num_q_vectors = 0;
644 }
645
646 /**
647 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
648 *
649 * This function resets the device so that it has 0 rx queues, tx queues, and
650 * MSI-X interrupts allocated.
651 */
652 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
653 {
654 igb_free_queues(adapter);
655 igb_free_q_vectors(adapter);
656 igb_reset_interrupt_capability(adapter);
657 }
658
659 /**
660 * igb_set_interrupt_capability - set MSI or MSI-X if supported
661 *
662 * Attempt to configure interrupts using the best available
663 * capabilities of the hardware and kernel.
664 **/
665 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
666 {
667 int err;
668 int numvecs, i;
669
670 /* Number of supported queues. */
671 adapter->num_rx_queues = adapter->rss_queues;
672 adapter->num_tx_queues = adapter->rss_queues;
673
674 /* start with one vector for every rx queue */
675 numvecs = adapter->num_rx_queues;
676
677 /* if tx handler is seperate add 1 for every tx queue */
678 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
679 numvecs += adapter->num_tx_queues;
680
681 /* store the number of vectors reserved for queues */
682 adapter->num_q_vectors = numvecs;
683
684 /* add 1 vector for link status interrupts */
685 numvecs++;
686 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
687 GFP_KERNEL);
688 if (!adapter->msix_entries)
689 goto msi_only;
690
691 for (i = 0; i < numvecs; i++)
692 adapter->msix_entries[i].entry = i;
693
694 err = pci_enable_msix(adapter->pdev,
695 adapter->msix_entries,
696 numvecs);
697 if (err == 0)
698 goto out;
699
700 igb_reset_interrupt_capability(adapter);
701
702 /* If we can't do MSI-X, try MSI */
703 msi_only:
704 #ifdef CONFIG_PCI_IOV
705 /* disable SR-IOV for non MSI-X configurations */
706 if (adapter->vf_data) {
707 struct e1000_hw *hw = &adapter->hw;
708 /* disable iov and allow time for transactions to clear */
709 pci_disable_sriov(adapter->pdev);
710 msleep(500);
711
712 kfree(adapter->vf_data);
713 adapter->vf_data = NULL;
714 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
715 msleep(100);
716 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
717 }
718 #endif
719 adapter->vfs_allocated_count = 0;
720 adapter->rss_queues = 1;
721 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
722 adapter->num_rx_queues = 1;
723 adapter->num_tx_queues = 1;
724 adapter->num_q_vectors = 1;
725 if (!pci_enable_msi(adapter->pdev))
726 adapter->flags |= IGB_FLAG_HAS_MSI;
727 out:
728 /* Notify the stack of the (possibly) reduced Tx Queue count. */
729 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
730 return;
731 }
732
733 /**
734 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
735 * @adapter: board private structure to initialize
736 *
737 * We allocate one q_vector per queue interrupt. If allocation fails we
738 * return -ENOMEM.
739 **/
740 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
741 {
742 struct igb_q_vector *q_vector;
743 struct e1000_hw *hw = &adapter->hw;
744 int v_idx;
745
746 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
747 q_vector = kzalloc(sizeof(struct igb_q_vector), GFP_KERNEL);
748 if (!q_vector)
749 goto err_out;
750 q_vector->adapter = adapter;
751 q_vector->itr_shift = (hw->mac.type == e1000_82575) ? 16 : 0;
752 q_vector->itr_register = hw->hw_addr + E1000_EITR(0);
753 q_vector->itr_val = IGB_START_ITR;
754 q_vector->set_itr = 1;
755 netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll, 64);
756 adapter->q_vector[v_idx] = q_vector;
757 }
758 return 0;
759
760 err_out:
761 while (v_idx) {
762 v_idx--;
763 q_vector = adapter->q_vector[v_idx];
764 netif_napi_del(&q_vector->napi);
765 kfree(q_vector);
766 adapter->q_vector[v_idx] = NULL;
767 }
768 return -ENOMEM;
769 }
770
771 static void igb_map_rx_ring_to_vector(struct igb_adapter *adapter,
772 int ring_idx, int v_idx)
773 {
774 struct igb_q_vector *q_vector;
775
776 q_vector = adapter->q_vector[v_idx];
777 q_vector->rx_ring = &adapter->rx_ring[ring_idx];
778 q_vector->rx_ring->q_vector = q_vector;
779 q_vector->itr_val = adapter->rx_itr_setting;
780 if (q_vector->itr_val && q_vector->itr_val <= 3)
781 q_vector->itr_val = IGB_START_ITR;
782 }
783
784 static void igb_map_tx_ring_to_vector(struct igb_adapter *adapter,
785 int ring_idx, int v_idx)
786 {
787 struct igb_q_vector *q_vector;
788
789 q_vector = adapter->q_vector[v_idx];
790 q_vector->tx_ring = &adapter->tx_ring[ring_idx];
791 q_vector->tx_ring->q_vector = q_vector;
792 q_vector->itr_val = adapter->tx_itr_setting;
793 if (q_vector->itr_val && q_vector->itr_val <= 3)
794 q_vector->itr_val = IGB_START_ITR;
795 }
796
797 /**
798 * igb_map_ring_to_vector - maps allocated queues to vectors
799 *
800 * This function maps the recently allocated queues to vectors.
801 **/
802 static int igb_map_ring_to_vector(struct igb_adapter *adapter)
803 {
804 int i;
805 int v_idx = 0;
806
807 if ((adapter->num_q_vectors < adapter->num_rx_queues) ||
808 (adapter->num_q_vectors < adapter->num_tx_queues))
809 return -ENOMEM;
810
811 if (adapter->num_q_vectors >=
812 (adapter->num_rx_queues + adapter->num_tx_queues)) {
813 for (i = 0; i < adapter->num_rx_queues; i++)
814 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
815 for (i = 0; i < adapter->num_tx_queues; i++)
816 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
817 } else {
818 for (i = 0; i < adapter->num_rx_queues; i++) {
819 if (i < adapter->num_tx_queues)
820 igb_map_tx_ring_to_vector(adapter, i, v_idx);
821 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
822 }
823 for (; i < adapter->num_tx_queues; i++)
824 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
825 }
826 return 0;
827 }
828
829 /**
830 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
831 *
832 * This function initializes the interrupts and allocates all of the queues.
833 **/
834 static int igb_init_interrupt_scheme(struct igb_adapter *adapter)
835 {
836 struct pci_dev *pdev = adapter->pdev;
837 int err;
838
839 igb_set_interrupt_capability(adapter);
840
841 err = igb_alloc_q_vectors(adapter);
842 if (err) {
843 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
844 goto err_alloc_q_vectors;
845 }
846
847 err = igb_alloc_queues(adapter);
848 if (err) {
849 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
850 goto err_alloc_queues;
851 }
852
853 err = igb_map_ring_to_vector(adapter);
854 if (err) {
855 dev_err(&pdev->dev, "Invalid q_vector to ring mapping\n");
856 goto err_map_queues;
857 }
858
859
860 return 0;
861 err_map_queues:
862 igb_free_queues(adapter);
863 err_alloc_queues:
864 igb_free_q_vectors(adapter);
865 err_alloc_q_vectors:
866 igb_reset_interrupt_capability(adapter);
867 return err;
868 }
869
870 /**
871 * igb_request_irq - initialize interrupts
872 *
873 * Attempts to configure interrupts using the best available
874 * capabilities of the hardware and kernel.
875 **/
876 static int igb_request_irq(struct igb_adapter *adapter)
877 {
878 struct net_device *netdev = adapter->netdev;
879 struct pci_dev *pdev = adapter->pdev;
880 struct e1000_hw *hw = &adapter->hw;
881 int err = 0;
882
883 if (adapter->msix_entries) {
884 err = igb_request_msix(adapter);
885 if (!err)
886 goto request_done;
887 /* fall back to MSI */
888 igb_clear_interrupt_scheme(adapter);
889 if (!pci_enable_msi(adapter->pdev))
890 adapter->flags |= IGB_FLAG_HAS_MSI;
891 igb_free_all_tx_resources(adapter);
892 igb_free_all_rx_resources(adapter);
893 adapter->num_tx_queues = 1;
894 adapter->num_rx_queues = 1;
895 adapter->num_q_vectors = 1;
896 err = igb_alloc_q_vectors(adapter);
897 if (err) {
898 dev_err(&pdev->dev,
899 "Unable to allocate memory for vectors\n");
900 goto request_done;
901 }
902 err = igb_alloc_queues(adapter);
903 if (err) {
904 dev_err(&pdev->dev,
905 "Unable to allocate memory for queues\n");
906 igb_free_q_vectors(adapter);
907 goto request_done;
908 }
909 igb_setup_all_tx_resources(adapter);
910 igb_setup_all_rx_resources(adapter);
911 } else {
912 switch (hw->mac.type) {
913 case e1000_82575:
914 wr32(E1000_MSIXBM(0),
915 (E1000_EICR_RX_QUEUE0 |
916 E1000_EICR_TX_QUEUE0 |
917 E1000_EIMS_OTHER));
918 break;
919 case e1000_82580:
920 case e1000_82576:
921 wr32(E1000_IVAR0, E1000_IVAR_VALID);
922 break;
923 default:
924 break;
925 }
926 }
927
928 if (adapter->flags & IGB_FLAG_HAS_MSI) {
929 err = request_irq(adapter->pdev->irq, igb_intr_msi, 0,
930 netdev->name, adapter);
931 if (!err)
932 goto request_done;
933
934 /* fall back to legacy interrupts */
935 igb_reset_interrupt_capability(adapter);
936 adapter->flags &= ~IGB_FLAG_HAS_MSI;
937 }
938
939 err = request_irq(adapter->pdev->irq, igb_intr, IRQF_SHARED,
940 netdev->name, adapter);
941
942 if (err)
943 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
944 err);
945
946 request_done:
947 return err;
948 }
949
950 static void igb_free_irq(struct igb_adapter *adapter)
951 {
952 if (adapter->msix_entries) {
953 int vector = 0, i;
954
955 free_irq(adapter->msix_entries[vector++].vector, adapter);
956
957 for (i = 0; i < adapter->num_q_vectors; i++) {
958 struct igb_q_vector *q_vector = adapter->q_vector[i];
959 free_irq(adapter->msix_entries[vector++].vector,
960 q_vector);
961 }
962 } else {
963 free_irq(adapter->pdev->irq, adapter);
964 }
965 }
966
967 /**
968 * igb_irq_disable - Mask off interrupt generation on the NIC
969 * @adapter: board private structure
970 **/
971 static void igb_irq_disable(struct igb_adapter *adapter)
972 {
973 struct e1000_hw *hw = &adapter->hw;
974
975 /*
976 * we need to be careful when disabling interrupts. The VFs are also
977 * mapped into these registers and so clearing the bits can cause
978 * issues on the VF drivers so we only need to clear what we set
979 */
980 if (adapter->msix_entries) {
981 u32 regval = rd32(E1000_EIAM);
982 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
983 wr32(E1000_EIMC, adapter->eims_enable_mask);
984 regval = rd32(E1000_EIAC);
985 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
986 }
987
988 wr32(E1000_IAM, 0);
989 wr32(E1000_IMC, ~0);
990 wrfl();
991 synchronize_irq(adapter->pdev->irq);
992 }
993
994 /**
995 * igb_irq_enable - Enable default interrupt generation settings
996 * @adapter: board private structure
997 **/
998 static void igb_irq_enable(struct igb_adapter *adapter)
999 {
1000 struct e1000_hw *hw = &adapter->hw;
1001
1002 if (adapter->msix_entries) {
1003 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC;
1004 u32 regval = rd32(E1000_EIAC);
1005 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1006 regval = rd32(E1000_EIAM);
1007 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1008 wr32(E1000_EIMS, adapter->eims_enable_mask);
1009 if (adapter->vfs_allocated_count) {
1010 wr32(E1000_MBVFIMR, 0xFF);
1011 ims |= E1000_IMS_VMMB;
1012 }
1013 if (adapter->hw.mac.type == e1000_82580)
1014 ims |= E1000_IMS_DRSTA;
1015
1016 wr32(E1000_IMS, ims);
1017 } else {
1018 wr32(E1000_IMS, IMS_ENABLE_MASK |
1019 E1000_IMS_DRSTA);
1020 wr32(E1000_IAM, IMS_ENABLE_MASK |
1021 E1000_IMS_DRSTA);
1022 }
1023 }
1024
1025 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1026 {
1027 struct e1000_hw *hw = &adapter->hw;
1028 u16 vid = adapter->hw.mng_cookie.vlan_id;
1029 u16 old_vid = adapter->mng_vlan_id;
1030
1031 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1032 /* add VID to filter table */
1033 igb_vfta_set(hw, vid, true);
1034 adapter->mng_vlan_id = vid;
1035 } else {
1036 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1037 }
1038
1039 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1040 (vid != old_vid) &&
1041 !vlan_group_get_device(adapter->vlgrp, old_vid)) {
1042 /* remove VID from filter table */
1043 igb_vfta_set(hw, old_vid, false);
1044 }
1045 }
1046
1047 /**
1048 * igb_release_hw_control - release control of the h/w to f/w
1049 * @adapter: address of board private structure
1050 *
1051 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1052 * For ASF and Pass Through versions of f/w this means that the
1053 * driver is no longer loaded.
1054 *
1055 **/
1056 static void igb_release_hw_control(struct igb_adapter *adapter)
1057 {
1058 struct e1000_hw *hw = &adapter->hw;
1059 u32 ctrl_ext;
1060
1061 /* Let firmware take over control of h/w */
1062 ctrl_ext = rd32(E1000_CTRL_EXT);
1063 wr32(E1000_CTRL_EXT,
1064 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1065 }
1066
1067 /**
1068 * igb_get_hw_control - get control of the h/w from f/w
1069 * @adapter: address of board private structure
1070 *
1071 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1072 * For ASF and Pass Through versions of f/w this means that
1073 * the driver is loaded.
1074 *
1075 **/
1076 static void igb_get_hw_control(struct igb_adapter *adapter)
1077 {
1078 struct e1000_hw *hw = &adapter->hw;
1079 u32 ctrl_ext;
1080
1081 /* Let firmware know the driver has taken over */
1082 ctrl_ext = rd32(E1000_CTRL_EXT);
1083 wr32(E1000_CTRL_EXT,
1084 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1085 }
1086
1087 /**
1088 * igb_configure - configure the hardware for RX and TX
1089 * @adapter: private board structure
1090 **/
1091 static void igb_configure(struct igb_adapter *adapter)
1092 {
1093 struct net_device *netdev = adapter->netdev;
1094 int i;
1095
1096 igb_get_hw_control(adapter);
1097 igb_set_rx_mode(netdev);
1098
1099 igb_restore_vlan(adapter);
1100
1101 igb_setup_tctl(adapter);
1102 igb_setup_mrqc(adapter);
1103 igb_setup_rctl(adapter);
1104
1105 igb_configure_tx(adapter);
1106 igb_configure_rx(adapter);
1107
1108 igb_rx_fifo_flush_82575(&adapter->hw);
1109
1110 /* call igb_desc_unused which always leaves
1111 * at least 1 descriptor unused to make sure
1112 * next_to_use != next_to_clean */
1113 for (i = 0; i < adapter->num_rx_queues; i++) {
1114 struct igb_ring *ring = &adapter->rx_ring[i];
1115 igb_alloc_rx_buffers_adv(ring, igb_desc_unused(ring));
1116 }
1117
1118
1119 adapter->tx_queue_len = netdev->tx_queue_len;
1120 }
1121
1122
1123 /**
1124 * igb_up - Open the interface and prepare it to handle traffic
1125 * @adapter: board private structure
1126 **/
1127 int igb_up(struct igb_adapter *adapter)
1128 {
1129 struct e1000_hw *hw = &adapter->hw;
1130 int i;
1131
1132 /* hardware has been reset, we need to reload some things */
1133 igb_configure(adapter);
1134
1135 clear_bit(__IGB_DOWN, &adapter->state);
1136
1137 for (i = 0; i < adapter->num_q_vectors; i++) {
1138 struct igb_q_vector *q_vector = adapter->q_vector[i];
1139 napi_enable(&q_vector->napi);
1140 }
1141 if (adapter->msix_entries)
1142 igb_configure_msix(adapter);
1143
1144 /* Clear any pending interrupts. */
1145 rd32(E1000_ICR);
1146 igb_irq_enable(adapter);
1147
1148 /* notify VFs that reset has been completed */
1149 if (adapter->vfs_allocated_count) {
1150 u32 reg_data = rd32(E1000_CTRL_EXT);
1151 reg_data |= E1000_CTRL_EXT_PFRSTD;
1152 wr32(E1000_CTRL_EXT, reg_data);
1153 }
1154
1155 netif_tx_start_all_queues(adapter->netdev);
1156
1157 /* start the watchdog. */
1158 hw->mac.get_link_status = 1;
1159 schedule_work(&adapter->watchdog_task);
1160
1161 return 0;
1162 }
1163
1164 void igb_down(struct igb_adapter *adapter)
1165 {
1166 struct net_device *netdev = adapter->netdev;
1167 struct e1000_hw *hw = &adapter->hw;
1168 u32 tctl, rctl;
1169 int i;
1170
1171 /* signal that we're down so the interrupt handler does not
1172 * reschedule our watchdog timer */
1173 set_bit(__IGB_DOWN, &adapter->state);
1174
1175 /* disable receives in the hardware */
1176 rctl = rd32(E1000_RCTL);
1177 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1178 /* flush and sleep below */
1179
1180 netif_tx_stop_all_queues(netdev);
1181
1182 /* disable transmits in the hardware */
1183 tctl = rd32(E1000_TCTL);
1184 tctl &= ~E1000_TCTL_EN;
1185 wr32(E1000_TCTL, tctl);
1186 /* flush both disables and wait for them to finish */
1187 wrfl();
1188 msleep(10);
1189
1190 for (i = 0; i < adapter->num_q_vectors; i++) {
1191 struct igb_q_vector *q_vector = adapter->q_vector[i];
1192 napi_disable(&q_vector->napi);
1193 }
1194
1195 igb_irq_disable(adapter);
1196
1197 del_timer_sync(&adapter->watchdog_timer);
1198 del_timer_sync(&adapter->phy_info_timer);
1199
1200 netdev->tx_queue_len = adapter->tx_queue_len;
1201 netif_carrier_off(netdev);
1202
1203 /* record the stats before reset*/
1204 igb_update_stats(adapter);
1205
1206 adapter->link_speed = 0;
1207 adapter->link_duplex = 0;
1208
1209 if (!pci_channel_offline(adapter->pdev))
1210 igb_reset(adapter);
1211 igb_clean_all_tx_rings(adapter);
1212 igb_clean_all_rx_rings(adapter);
1213 #ifdef CONFIG_IGB_DCA
1214
1215 /* since we reset the hardware DCA settings were cleared */
1216 igb_setup_dca(adapter);
1217 #endif
1218 }
1219
1220 void igb_reinit_locked(struct igb_adapter *adapter)
1221 {
1222 WARN_ON(in_interrupt());
1223 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1224 msleep(1);
1225 igb_down(adapter);
1226 igb_up(adapter);
1227 clear_bit(__IGB_RESETTING, &adapter->state);
1228 }
1229
1230 void igb_reset(struct igb_adapter *adapter)
1231 {
1232 struct pci_dev *pdev = adapter->pdev;
1233 struct e1000_hw *hw = &adapter->hw;
1234 struct e1000_mac_info *mac = &hw->mac;
1235 struct e1000_fc_info *fc = &hw->fc;
1236 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
1237 u16 hwm;
1238
1239 /* Repartition Pba for greater than 9k mtu
1240 * To take effect CTRL.RST is required.
1241 */
1242 switch (mac->type) {
1243 case e1000_82580:
1244 pba = rd32(E1000_RXPBS);
1245 pba = igb_rxpbs_adjust_82580(pba);
1246 break;
1247 case e1000_82576:
1248 pba = rd32(E1000_RXPBS);
1249 pba &= E1000_RXPBS_SIZE_MASK_82576;
1250 break;
1251 case e1000_82575:
1252 default:
1253 pba = E1000_PBA_34K;
1254 break;
1255 }
1256
1257 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1258 (mac->type < e1000_82576)) {
1259 /* adjust PBA for jumbo frames */
1260 wr32(E1000_PBA, pba);
1261
1262 /* To maintain wire speed transmits, the Tx FIFO should be
1263 * large enough to accommodate two full transmit packets,
1264 * rounded up to the next 1KB and expressed in KB. Likewise,
1265 * the Rx FIFO should be large enough to accommodate at least
1266 * one full receive packet and is similarly rounded up and
1267 * expressed in KB. */
1268 pba = rd32(E1000_PBA);
1269 /* upper 16 bits has Tx packet buffer allocation size in KB */
1270 tx_space = pba >> 16;
1271 /* lower 16 bits has Rx packet buffer allocation size in KB */
1272 pba &= 0xffff;
1273 /* the tx fifo also stores 16 bytes of information about the tx
1274 * but don't include ethernet FCS because hardware appends it */
1275 min_tx_space = (adapter->max_frame_size +
1276 sizeof(union e1000_adv_tx_desc) -
1277 ETH_FCS_LEN) * 2;
1278 min_tx_space = ALIGN(min_tx_space, 1024);
1279 min_tx_space >>= 10;
1280 /* software strips receive CRC, so leave room for it */
1281 min_rx_space = adapter->max_frame_size;
1282 min_rx_space = ALIGN(min_rx_space, 1024);
1283 min_rx_space >>= 10;
1284
1285 /* If current Tx allocation is less than the min Tx FIFO size,
1286 * and the min Tx FIFO size is less than the current Rx FIFO
1287 * allocation, take space away from current Rx allocation */
1288 if (tx_space < min_tx_space &&
1289 ((min_tx_space - tx_space) < pba)) {
1290 pba = pba - (min_tx_space - tx_space);
1291
1292 /* if short on rx space, rx wins and must trump tx
1293 * adjustment */
1294 if (pba < min_rx_space)
1295 pba = min_rx_space;
1296 }
1297 wr32(E1000_PBA, pba);
1298 }
1299
1300 /* flow control settings */
1301 /* The high water mark must be low enough to fit one full frame
1302 * (or the size used for early receive) above it in the Rx FIFO.
1303 * Set it to the lower of:
1304 * - 90% of the Rx FIFO size, or
1305 * - the full Rx FIFO size minus one full frame */
1306 hwm = min(((pba << 10) * 9 / 10),
1307 ((pba << 10) - 2 * adapter->max_frame_size));
1308
1309 if (mac->type < e1000_82576) {
1310 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
1311 fc->low_water = fc->high_water - 8;
1312 } else {
1313 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1314 fc->low_water = fc->high_water - 16;
1315 }
1316 fc->pause_time = 0xFFFF;
1317 fc->send_xon = 1;
1318 fc->current_mode = fc->requested_mode;
1319
1320 /* disable receive for all VFs and wait one second */
1321 if (adapter->vfs_allocated_count) {
1322 int i;
1323 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1324 adapter->vf_data[i].flags = 0;
1325
1326 /* ping all the active vfs to let them know we are going down */
1327 igb_ping_all_vfs(adapter);
1328
1329 /* disable transmits and receives */
1330 wr32(E1000_VFRE, 0);
1331 wr32(E1000_VFTE, 0);
1332 }
1333
1334 /* Allow time for pending master requests to run */
1335 hw->mac.ops.reset_hw(hw);
1336 wr32(E1000_WUC, 0);
1337
1338 if (hw->mac.ops.init_hw(hw))
1339 dev_err(&pdev->dev, "Hardware Error\n");
1340
1341 if (hw->mac.type == e1000_82580) {
1342 u32 reg = rd32(E1000_PCIEMISC);
1343 wr32(E1000_PCIEMISC,
1344 reg & ~E1000_PCIEMISC_LX_DECISION);
1345 }
1346 igb_update_mng_vlan(adapter);
1347
1348 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1349 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1350
1351 igb_reset_adaptive(hw);
1352 igb_get_phy_info(hw);
1353 }
1354
1355 static const struct net_device_ops igb_netdev_ops = {
1356 .ndo_open = igb_open,
1357 .ndo_stop = igb_close,
1358 .ndo_start_xmit = igb_xmit_frame_adv,
1359 .ndo_get_stats = igb_get_stats,
1360 .ndo_set_rx_mode = igb_set_rx_mode,
1361 .ndo_set_multicast_list = igb_set_rx_mode,
1362 .ndo_set_mac_address = igb_set_mac,
1363 .ndo_change_mtu = igb_change_mtu,
1364 .ndo_do_ioctl = igb_ioctl,
1365 .ndo_tx_timeout = igb_tx_timeout,
1366 .ndo_validate_addr = eth_validate_addr,
1367 .ndo_vlan_rx_register = igb_vlan_rx_register,
1368 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1369 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1370 #ifdef CONFIG_NET_POLL_CONTROLLER
1371 .ndo_poll_controller = igb_netpoll,
1372 #endif
1373 };
1374
1375 /**
1376 * igb_probe - Device Initialization Routine
1377 * @pdev: PCI device information struct
1378 * @ent: entry in igb_pci_tbl
1379 *
1380 * Returns 0 on success, negative on failure
1381 *
1382 * igb_probe initializes an adapter identified by a pci_dev structure.
1383 * The OS initialization, configuring of the adapter private structure,
1384 * and a hardware reset occur.
1385 **/
1386 static int __devinit igb_probe(struct pci_dev *pdev,
1387 const struct pci_device_id *ent)
1388 {
1389 struct net_device *netdev;
1390 struct igb_adapter *adapter;
1391 struct e1000_hw *hw;
1392 u16 eeprom_data = 0;
1393 static int global_quad_port_a; /* global quad port a indication */
1394 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1395 unsigned long mmio_start, mmio_len;
1396 int err, pci_using_dac;
1397 u16 eeprom_apme_mask = IGB_EEPROM_APME;
1398 u32 part_num;
1399
1400 err = pci_enable_device_mem(pdev);
1401 if (err)
1402 return err;
1403
1404 pci_using_dac = 0;
1405 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1406 if (!err) {
1407 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1408 if (!err)
1409 pci_using_dac = 1;
1410 } else {
1411 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1412 if (err) {
1413 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1414 if (err) {
1415 dev_err(&pdev->dev, "No usable DMA "
1416 "configuration, aborting\n");
1417 goto err_dma;
1418 }
1419 }
1420 }
1421
1422 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1423 IORESOURCE_MEM),
1424 igb_driver_name);
1425 if (err)
1426 goto err_pci_reg;
1427
1428 pci_enable_pcie_error_reporting(pdev);
1429
1430 pci_set_master(pdev);
1431 pci_save_state(pdev);
1432
1433 err = -ENOMEM;
1434 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1435 IGB_ABS_MAX_TX_QUEUES);
1436 if (!netdev)
1437 goto err_alloc_etherdev;
1438
1439 SET_NETDEV_DEV(netdev, &pdev->dev);
1440
1441 pci_set_drvdata(pdev, netdev);
1442 adapter = netdev_priv(netdev);
1443 adapter->netdev = netdev;
1444 adapter->pdev = pdev;
1445 hw = &adapter->hw;
1446 hw->back = adapter;
1447 adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1448
1449 mmio_start = pci_resource_start(pdev, 0);
1450 mmio_len = pci_resource_len(pdev, 0);
1451
1452 err = -EIO;
1453 hw->hw_addr = ioremap(mmio_start, mmio_len);
1454 if (!hw->hw_addr)
1455 goto err_ioremap;
1456
1457 netdev->netdev_ops = &igb_netdev_ops;
1458 igb_set_ethtool_ops(netdev);
1459 netdev->watchdog_timeo = 5 * HZ;
1460
1461 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1462
1463 netdev->mem_start = mmio_start;
1464 netdev->mem_end = mmio_start + mmio_len;
1465
1466 /* PCI config space info */
1467 hw->vendor_id = pdev->vendor;
1468 hw->device_id = pdev->device;
1469 hw->revision_id = pdev->revision;
1470 hw->subsystem_vendor_id = pdev->subsystem_vendor;
1471 hw->subsystem_device_id = pdev->subsystem_device;
1472
1473 /* Copy the default MAC, PHY and NVM function pointers */
1474 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1475 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1476 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1477 /* Initialize skew-specific constants */
1478 err = ei->get_invariants(hw);
1479 if (err)
1480 goto err_sw_init;
1481
1482 /* setup the private structure */
1483 err = igb_sw_init(adapter);
1484 if (err)
1485 goto err_sw_init;
1486
1487 igb_get_bus_info_pcie(hw);
1488
1489 hw->phy.autoneg_wait_to_complete = false;
1490 hw->mac.adaptive_ifs = true;
1491
1492 /* Copper options */
1493 if (hw->phy.media_type == e1000_media_type_copper) {
1494 hw->phy.mdix = AUTO_ALL_MODES;
1495 hw->phy.disable_polarity_correction = false;
1496 hw->phy.ms_type = e1000_ms_hw_default;
1497 }
1498
1499 if (igb_check_reset_block(hw))
1500 dev_info(&pdev->dev,
1501 "PHY reset is blocked due to SOL/IDER session.\n");
1502
1503 netdev->features = NETIF_F_SG |
1504 NETIF_F_IP_CSUM |
1505 NETIF_F_HW_VLAN_TX |
1506 NETIF_F_HW_VLAN_RX |
1507 NETIF_F_HW_VLAN_FILTER;
1508
1509 netdev->features |= NETIF_F_IPV6_CSUM;
1510 netdev->features |= NETIF_F_TSO;
1511 netdev->features |= NETIF_F_TSO6;
1512 netdev->features |= NETIF_F_GRO;
1513
1514 netdev->vlan_features |= NETIF_F_TSO;
1515 netdev->vlan_features |= NETIF_F_TSO6;
1516 netdev->vlan_features |= NETIF_F_IP_CSUM;
1517 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
1518 netdev->vlan_features |= NETIF_F_SG;
1519
1520 if (pci_using_dac)
1521 netdev->features |= NETIF_F_HIGHDMA;
1522
1523 if (hw->mac.type >= e1000_82576)
1524 netdev->features |= NETIF_F_SCTP_CSUM;
1525
1526 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
1527
1528 /* before reading the NVM, reset the controller to put the device in a
1529 * known good starting state */
1530 hw->mac.ops.reset_hw(hw);
1531
1532 /* make sure the NVM is good */
1533 if (igb_validate_nvm_checksum(hw) < 0) {
1534 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1535 err = -EIO;
1536 goto err_eeprom;
1537 }
1538
1539 /* copy the MAC address out of the NVM */
1540 if (hw->mac.ops.read_mac_addr(hw))
1541 dev_err(&pdev->dev, "NVM Read Error\n");
1542
1543 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1544 memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1545
1546 if (!is_valid_ether_addr(netdev->perm_addr)) {
1547 dev_err(&pdev->dev, "Invalid MAC Address\n");
1548 err = -EIO;
1549 goto err_eeprom;
1550 }
1551
1552 setup_timer(&adapter->watchdog_timer, &igb_watchdog,
1553 (unsigned long) adapter);
1554 setup_timer(&adapter->phy_info_timer, &igb_update_phy_info,
1555 (unsigned long) adapter);
1556
1557 INIT_WORK(&adapter->reset_task, igb_reset_task);
1558 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1559
1560 /* Initialize link properties that are user-changeable */
1561 adapter->fc_autoneg = true;
1562 hw->mac.autoneg = true;
1563 hw->phy.autoneg_advertised = 0x2f;
1564
1565 hw->fc.requested_mode = e1000_fc_default;
1566 hw->fc.current_mode = e1000_fc_default;
1567
1568 igb_validate_mdi_setting(hw);
1569
1570 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1571 * enable the ACPI Magic Packet filter
1572 */
1573
1574 if (hw->bus.func == 0)
1575 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1576 else if (hw->mac.type == e1000_82580)
1577 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
1578 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
1579 &eeprom_data);
1580 else if (hw->bus.func == 1)
1581 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1582
1583 if (eeprom_data & eeprom_apme_mask)
1584 adapter->eeprom_wol |= E1000_WUFC_MAG;
1585
1586 /* now that we have the eeprom settings, apply the special cases where
1587 * the eeprom may be wrong or the board simply won't support wake on
1588 * lan on a particular port */
1589 switch (pdev->device) {
1590 case E1000_DEV_ID_82575GB_QUAD_COPPER:
1591 adapter->eeprom_wol = 0;
1592 break;
1593 case E1000_DEV_ID_82575EB_FIBER_SERDES:
1594 case E1000_DEV_ID_82576_FIBER:
1595 case E1000_DEV_ID_82576_SERDES:
1596 /* Wake events only supported on port A for dual fiber
1597 * regardless of eeprom setting */
1598 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1599 adapter->eeprom_wol = 0;
1600 break;
1601 case E1000_DEV_ID_82576_QUAD_COPPER:
1602 /* if quad port adapter, disable WoL on all but port A */
1603 if (global_quad_port_a != 0)
1604 adapter->eeprom_wol = 0;
1605 else
1606 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
1607 /* Reset for multiple quad port adapters */
1608 if (++global_quad_port_a == 4)
1609 global_quad_port_a = 0;
1610 break;
1611 }
1612
1613 /* initialize the wol settings based on the eeprom settings */
1614 adapter->wol = adapter->eeprom_wol;
1615 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1616
1617 /* reset the hardware with the new settings */
1618 igb_reset(adapter);
1619
1620 /* let the f/w know that the h/w is now under the control of the
1621 * driver. */
1622 igb_get_hw_control(adapter);
1623
1624 strcpy(netdev->name, "eth%d");
1625 err = register_netdev(netdev);
1626 if (err)
1627 goto err_register;
1628
1629 /* carrier off reporting is important to ethtool even BEFORE open */
1630 netif_carrier_off(netdev);
1631
1632 #ifdef CONFIG_IGB_DCA
1633 if (dca_add_requester(&pdev->dev) == 0) {
1634 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1635 dev_info(&pdev->dev, "DCA enabled\n");
1636 igb_setup_dca(adapter);
1637 }
1638
1639 #endif
1640 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1641 /* print bus type/speed/width info */
1642 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
1643 netdev->name,
1644 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
1645 "unknown"),
1646 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
1647 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
1648 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
1649 "unknown"),
1650 netdev->dev_addr);
1651
1652 igb_read_part_num(hw, &part_num);
1653 dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1654 (part_num >> 8), (part_num & 0xff));
1655
1656 dev_info(&pdev->dev,
1657 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1658 adapter->msix_entries ? "MSI-X" :
1659 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1660 adapter->num_rx_queues, adapter->num_tx_queues);
1661
1662 return 0;
1663
1664 err_register:
1665 igb_release_hw_control(adapter);
1666 err_eeprom:
1667 if (!igb_check_reset_block(hw))
1668 igb_reset_phy(hw);
1669
1670 if (hw->flash_address)
1671 iounmap(hw->flash_address);
1672 err_sw_init:
1673 igb_clear_interrupt_scheme(adapter);
1674 iounmap(hw->hw_addr);
1675 err_ioremap:
1676 free_netdev(netdev);
1677 err_alloc_etherdev:
1678 pci_release_selected_regions(pdev,
1679 pci_select_bars(pdev, IORESOURCE_MEM));
1680 err_pci_reg:
1681 err_dma:
1682 pci_disable_device(pdev);
1683 return err;
1684 }
1685
1686 /**
1687 * igb_remove - Device Removal Routine
1688 * @pdev: PCI device information struct
1689 *
1690 * igb_remove is called by the PCI subsystem to alert the driver
1691 * that it should release a PCI device. The could be caused by a
1692 * Hot-Plug event, or because the driver is going to be removed from
1693 * memory.
1694 **/
1695 static void __devexit igb_remove(struct pci_dev *pdev)
1696 {
1697 struct net_device *netdev = pci_get_drvdata(pdev);
1698 struct igb_adapter *adapter = netdev_priv(netdev);
1699 struct e1000_hw *hw = &adapter->hw;
1700
1701 /* flush_scheduled work may reschedule our watchdog task, so
1702 * explicitly disable watchdog tasks from being rescheduled */
1703 set_bit(__IGB_DOWN, &adapter->state);
1704 del_timer_sync(&adapter->watchdog_timer);
1705 del_timer_sync(&adapter->phy_info_timer);
1706
1707 flush_scheduled_work();
1708
1709 #ifdef CONFIG_IGB_DCA
1710 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1711 dev_info(&pdev->dev, "DCA disabled\n");
1712 dca_remove_requester(&pdev->dev);
1713 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1714 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
1715 }
1716 #endif
1717
1718 /* Release control of h/w to f/w. If f/w is AMT enabled, this
1719 * would have already happened in close and is redundant. */
1720 igb_release_hw_control(adapter);
1721
1722 unregister_netdev(netdev);
1723
1724 if (!igb_check_reset_block(hw))
1725 igb_reset_phy(hw);
1726
1727 igb_clear_interrupt_scheme(adapter);
1728
1729 #ifdef CONFIG_PCI_IOV
1730 /* reclaim resources allocated to VFs */
1731 if (adapter->vf_data) {
1732 /* disable iov and allow time for transactions to clear */
1733 pci_disable_sriov(pdev);
1734 msleep(500);
1735
1736 kfree(adapter->vf_data);
1737 adapter->vf_data = NULL;
1738 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1739 msleep(100);
1740 dev_info(&pdev->dev, "IOV Disabled\n");
1741 }
1742 #endif
1743
1744 iounmap(hw->hw_addr);
1745 if (hw->flash_address)
1746 iounmap(hw->flash_address);
1747 pci_release_selected_regions(pdev,
1748 pci_select_bars(pdev, IORESOURCE_MEM));
1749
1750 free_netdev(netdev);
1751
1752 pci_disable_pcie_error_reporting(pdev);
1753
1754 pci_disable_device(pdev);
1755 }
1756
1757 /**
1758 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
1759 * @adapter: board private structure to initialize
1760 *
1761 * This function initializes the vf specific data storage and then attempts to
1762 * allocate the VFs. The reason for ordering it this way is because it is much
1763 * mor expensive time wise to disable SR-IOV than it is to allocate and free
1764 * the memory for the VFs.
1765 **/
1766 static void __devinit igb_probe_vfs(struct igb_adapter * adapter)
1767 {
1768 #ifdef CONFIG_PCI_IOV
1769 struct pci_dev *pdev = adapter->pdev;
1770
1771 if (adapter->vfs_allocated_count > 7)
1772 adapter->vfs_allocated_count = 7;
1773
1774 if (adapter->vfs_allocated_count) {
1775 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
1776 sizeof(struct vf_data_storage),
1777 GFP_KERNEL);
1778 /* if allocation failed then we do not support SR-IOV */
1779 if (!adapter->vf_data) {
1780 adapter->vfs_allocated_count = 0;
1781 dev_err(&pdev->dev, "Unable to allocate memory for VF "
1782 "Data Storage\n");
1783 }
1784 }
1785
1786 if (pci_enable_sriov(pdev, adapter->vfs_allocated_count)) {
1787 kfree(adapter->vf_data);
1788 adapter->vf_data = NULL;
1789 #endif /* CONFIG_PCI_IOV */
1790 adapter->vfs_allocated_count = 0;
1791 #ifdef CONFIG_PCI_IOV
1792 } else {
1793 unsigned char mac_addr[ETH_ALEN];
1794 int i;
1795 dev_info(&pdev->dev, "%d vfs allocated\n",
1796 adapter->vfs_allocated_count);
1797 for (i = 0; i < adapter->vfs_allocated_count; i++) {
1798 random_ether_addr(mac_addr);
1799 igb_set_vf_mac(adapter, i, mac_addr);
1800 }
1801 }
1802 #endif /* CONFIG_PCI_IOV */
1803 }
1804
1805
1806 /**
1807 * igb_init_hw_timer - Initialize hardware timer used with IEEE 1588 timestamp
1808 * @adapter: board private structure to initialize
1809 *
1810 * igb_init_hw_timer initializes the function pointer and values for the hw
1811 * timer found in hardware.
1812 **/
1813 static void igb_init_hw_timer(struct igb_adapter *adapter)
1814 {
1815 struct e1000_hw *hw = &adapter->hw;
1816
1817 switch (hw->mac.type) {
1818 case e1000_82580:
1819 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1820 adapter->cycles.read = igb_read_clock;
1821 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1822 adapter->cycles.mult = 1;
1823 /*
1824 * The 82580 timesync updates the system timer every 8ns by 8ns
1825 * and the value cannot be shifted. Instead we need to shift
1826 * the registers to generate a 64bit timer value. As a result
1827 * SYSTIMR/L/H, TXSTMPL/H, RXSTMPL/H all have to be shifted by
1828 * 24 in order to generate a larger value for synchronization.
1829 */
1830 adapter->cycles.shift = IGB_82580_TSYNC_SHIFT;
1831 /* disable system timer temporarily by setting bit 31 */
1832 wr32(E1000_TSAUXC, 0x80000000);
1833 wrfl();
1834
1835 /* Set registers so that rollover occurs soon to test this. */
1836 wr32(E1000_SYSTIMR, 0x00000000);
1837 wr32(E1000_SYSTIML, 0x80000000);
1838 wr32(E1000_SYSTIMH, 0x000000FF);
1839 wrfl();
1840
1841 /* enable system timer by clearing bit 31 */
1842 wr32(E1000_TSAUXC, 0x0);
1843 wrfl();
1844
1845 timecounter_init(&adapter->clock,
1846 &adapter->cycles,
1847 ktime_to_ns(ktime_get_real()));
1848 /*
1849 * Synchronize our NIC clock against system wall clock. NIC
1850 * time stamp reading requires ~3us per sample, each sample
1851 * was pretty stable even under load => only require 10
1852 * samples for each offset comparison.
1853 */
1854 memset(&adapter->compare, 0, sizeof(adapter->compare));
1855 adapter->compare.source = &adapter->clock;
1856 adapter->compare.target = ktime_get_real;
1857 adapter->compare.num_samples = 10;
1858 timecompare_update(&adapter->compare, 0);
1859 break;
1860 case e1000_82576:
1861 /*
1862 * Initialize hardware timer: we keep it running just in case
1863 * that some program needs it later on.
1864 */
1865 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1866 adapter->cycles.read = igb_read_clock;
1867 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1868 adapter->cycles.mult = 1;
1869 /**
1870 * Scale the NIC clock cycle by a large factor so that
1871 * relatively small clock corrections can be added or
1872 * substracted at each clock tick. The drawbacks of a large
1873 * factor are a) that the clock register overflows more quickly
1874 * (not such a big deal) and b) that the increment per tick has
1875 * to fit into 24 bits. As a result we need to use a shift of
1876 * 19 so we can fit a value of 16 into the TIMINCA register.
1877 */
1878 adapter->cycles.shift = IGB_82576_TSYNC_SHIFT;
1879 wr32(E1000_TIMINCA,
1880 (1 << E1000_TIMINCA_16NS_SHIFT) |
1881 (16 << IGB_82576_TSYNC_SHIFT));
1882
1883 /* Set registers so that rollover occurs soon to test this. */
1884 wr32(E1000_SYSTIML, 0x00000000);
1885 wr32(E1000_SYSTIMH, 0xFF800000);
1886 wrfl();
1887
1888 timecounter_init(&adapter->clock,
1889 &adapter->cycles,
1890 ktime_to_ns(ktime_get_real()));
1891 /*
1892 * Synchronize our NIC clock against system wall clock. NIC
1893 * time stamp reading requires ~3us per sample, each sample
1894 * was pretty stable even under load => only require 10
1895 * samples for each offset comparison.
1896 */
1897 memset(&adapter->compare, 0, sizeof(adapter->compare));
1898 adapter->compare.source = &adapter->clock;
1899 adapter->compare.target = ktime_get_real;
1900 adapter->compare.num_samples = 10;
1901 timecompare_update(&adapter->compare, 0);
1902 break;
1903 case e1000_82575:
1904 /* 82575 does not support timesync */
1905 default:
1906 break;
1907 }
1908
1909 }
1910
1911 /**
1912 * igb_sw_init - Initialize general software structures (struct igb_adapter)
1913 * @adapter: board private structure to initialize
1914 *
1915 * igb_sw_init initializes the Adapter private data structure.
1916 * Fields are initialized based on PCI device information and
1917 * OS network device settings (MTU size).
1918 **/
1919 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1920 {
1921 struct e1000_hw *hw = &adapter->hw;
1922 struct net_device *netdev = adapter->netdev;
1923 struct pci_dev *pdev = adapter->pdev;
1924
1925 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1926
1927 adapter->tx_ring_count = IGB_DEFAULT_TXD;
1928 adapter->rx_ring_count = IGB_DEFAULT_RXD;
1929 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
1930 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
1931
1932 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1933 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1934
1935 #ifdef CONFIG_PCI_IOV
1936 if (hw->mac.type == e1000_82576)
1937 adapter->vfs_allocated_count = max_vfs;
1938
1939 #endif /* CONFIG_PCI_IOV */
1940 adapter->rss_queues = min_t(u32, IGB_MAX_RX_QUEUES, num_online_cpus());
1941
1942 /*
1943 * if rss_queues > 4 or vfs are going to be allocated with rss_queues
1944 * then we should combine the queues into a queue pair in order to
1945 * conserve interrupts due to limited supply
1946 */
1947 if ((adapter->rss_queues > 4) ||
1948 ((adapter->rss_queues > 1) && (adapter->vfs_allocated_count > 6)))
1949 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1950
1951 /* This call may decrease the number of queues */
1952 if (igb_init_interrupt_scheme(adapter)) {
1953 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1954 return -ENOMEM;
1955 }
1956
1957 igb_init_hw_timer(adapter);
1958 igb_probe_vfs(adapter);
1959
1960 /* Explicitly disable IRQ since the NIC can be in any state. */
1961 igb_irq_disable(adapter);
1962
1963 set_bit(__IGB_DOWN, &adapter->state);
1964 return 0;
1965 }
1966
1967 /**
1968 * igb_open - Called when a network interface is made active
1969 * @netdev: network interface device structure
1970 *
1971 * Returns 0 on success, negative value on failure
1972 *
1973 * The open entry point is called when a network interface is made
1974 * active by the system (IFF_UP). At this point all resources needed
1975 * for transmit and receive operations are allocated, the interrupt
1976 * handler is registered with the OS, the watchdog timer is started,
1977 * and the stack is notified that the interface is ready.
1978 **/
1979 static int igb_open(struct net_device *netdev)
1980 {
1981 struct igb_adapter *adapter = netdev_priv(netdev);
1982 struct e1000_hw *hw = &adapter->hw;
1983 int err;
1984 int i;
1985
1986 /* disallow open during test */
1987 if (test_bit(__IGB_TESTING, &adapter->state))
1988 return -EBUSY;
1989
1990 netif_carrier_off(netdev);
1991
1992 /* allocate transmit descriptors */
1993 err = igb_setup_all_tx_resources(adapter);
1994 if (err)
1995 goto err_setup_tx;
1996
1997 /* allocate receive descriptors */
1998 err = igb_setup_all_rx_resources(adapter);
1999 if (err)
2000 goto err_setup_rx;
2001
2002 /* e1000_power_up_phy(adapter); */
2003
2004 /* before we allocate an interrupt, we must be ready to handle it.
2005 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2006 * as soon as we call pci_request_irq, so we have to setup our
2007 * clean_rx handler before we do so. */
2008 igb_configure(adapter);
2009
2010 err = igb_request_irq(adapter);
2011 if (err)
2012 goto err_req_irq;
2013
2014 /* From here on the code is the same as igb_up() */
2015 clear_bit(__IGB_DOWN, &adapter->state);
2016
2017 for (i = 0; i < adapter->num_q_vectors; i++) {
2018 struct igb_q_vector *q_vector = adapter->q_vector[i];
2019 napi_enable(&q_vector->napi);
2020 }
2021
2022 /* Clear any pending interrupts. */
2023 rd32(E1000_ICR);
2024
2025 igb_irq_enable(adapter);
2026
2027 /* notify VFs that reset has been completed */
2028 if (adapter->vfs_allocated_count) {
2029 u32 reg_data = rd32(E1000_CTRL_EXT);
2030 reg_data |= E1000_CTRL_EXT_PFRSTD;
2031 wr32(E1000_CTRL_EXT, reg_data);
2032 }
2033
2034 netif_tx_start_all_queues(netdev);
2035
2036 /* start the watchdog. */
2037 hw->mac.get_link_status = 1;
2038 schedule_work(&adapter->watchdog_task);
2039
2040 return 0;
2041
2042 err_req_irq:
2043 igb_release_hw_control(adapter);
2044 /* e1000_power_down_phy(adapter); */
2045 igb_free_all_rx_resources(adapter);
2046 err_setup_rx:
2047 igb_free_all_tx_resources(adapter);
2048 err_setup_tx:
2049 igb_reset(adapter);
2050
2051 return err;
2052 }
2053
2054 /**
2055 * igb_close - Disables a network interface
2056 * @netdev: network interface device structure
2057 *
2058 * Returns 0, this is not allowed to fail
2059 *
2060 * The close entry point is called when an interface is de-activated
2061 * by the OS. The hardware is still under the driver's control, but
2062 * needs to be disabled. A global MAC reset is issued to stop the
2063 * hardware, and all transmit and receive resources are freed.
2064 **/
2065 static int igb_close(struct net_device *netdev)
2066 {
2067 struct igb_adapter *adapter = netdev_priv(netdev);
2068
2069 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
2070 igb_down(adapter);
2071
2072 igb_free_irq(adapter);
2073
2074 igb_free_all_tx_resources(adapter);
2075 igb_free_all_rx_resources(adapter);
2076
2077 return 0;
2078 }
2079
2080 /**
2081 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2082 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2083 *
2084 * Return 0 on success, negative on failure
2085 **/
2086 int igb_setup_tx_resources(struct igb_ring *tx_ring)
2087 {
2088 struct pci_dev *pdev = tx_ring->pdev;
2089 int size;
2090
2091 size = sizeof(struct igb_buffer) * tx_ring->count;
2092 tx_ring->buffer_info = vmalloc(size);
2093 if (!tx_ring->buffer_info)
2094 goto err;
2095 memset(tx_ring->buffer_info, 0, size);
2096
2097 /* round up to nearest 4K */
2098 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
2099 tx_ring->size = ALIGN(tx_ring->size, 4096);
2100
2101 tx_ring->desc = pci_alloc_consistent(pdev,
2102 tx_ring->size,
2103 &tx_ring->dma);
2104
2105 if (!tx_ring->desc)
2106 goto err;
2107
2108 tx_ring->next_to_use = 0;
2109 tx_ring->next_to_clean = 0;
2110 return 0;
2111
2112 err:
2113 vfree(tx_ring->buffer_info);
2114 dev_err(&pdev->dev,
2115 "Unable to allocate memory for the transmit descriptor ring\n");
2116 return -ENOMEM;
2117 }
2118
2119 /**
2120 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2121 * (Descriptors) for all queues
2122 * @adapter: board private structure
2123 *
2124 * Return 0 on success, negative on failure
2125 **/
2126 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2127 {
2128 struct pci_dev *pdev = adapter->pdev;
2129 int i, err = 0;
2130
2131 for (i = 0; i < adapter->num_tx_queues; i++) {
2132 err = igb_setup_tx_resources(&adapter->tx_ring[i]);
2133 if (err) {
2134 dev_err(&pdev->dev,
2135 "Allocation for Tx Queue %u failed\n", i);
2136 for (i--; i >= 0; i--)
2137 igb_free_tx_resources(&adapter->tx_ring[i]);
2138 break;
2139 }
2140 }
2141
2142 for (i = 0; i < IGB_ABS_MAX_TX_QUEUES; i++) {
2143 int r_idx = i % adapter->num_tx_queues;
2144 adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
2145 }
2146 return err;
2147 }
2148
2149 /**
2150 * igb_setup_tctl - configure the transmit control registers
2151 * @adapter: Board private structure
2152 **/
2153 void igb_setup_tctl(struct igb_adapter *adapter)
2154 {
2155 struct e1000_hw *hw = &adapter->hw;
2156 u32 tctl;
2157
2158 /* disable queue 0 which is enabled by default on 82575 and 82576 */
2159 wr32(E1000_TXDCTL(0), 0);
2160
2161 /* Program the Transmit Control Register */
2162 tctl = rd32(E1000_TCTL);
2163 tctl &= ~E1000_TCTL_CT;
2164 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2165 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2166
2167 igb_config_collision_dist(hw);
2168
2169 /* Enable transmits */
2170 tctl |= E1000_TCTL_EN;
2171
2172 wr32(E1000_TCTL, tctl);
2173 }
2174
2175 /**
2176 * igb_configure_tx_ring - Configure transmit ring after Reset
2177 * @adapter: board private structure
2178 * @ring: tx ring to configure
2179 *
2180 * Configure a transmit ring after a reset.
2181 **/
2182 void igb_configure_tx_ring(struct igb_adapter *adapter,
2183 struct igb_ring *ring)
2184 {
2185 struct e1000_hw *hw = &adapter->hw;
2186 u32 txdctl;
2187 u64 tdba = ring->dma;
2188 int reg_idx = ring->reg_idx;
2189
2190 /* disable the queue */
2191 txdctl = rd32(E1000_TXDCTL(reg_idx));
2192 wr32(E1000_TXDCTL(reg_idx),
2193 txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
2194 wrfl();
2195 mdelay(10);
2196
2197 wr32(E1000_TDLEN(reg_idx),
2198 ring->count * sizeof(union e1000_adv_tx_desc));
2199 wr32(E1000_TDBAL(reg_idx),
2200 tdba & 0x00000000ffffffffULL);
2201 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
2202
2203 ring->head = hw->hw_addr + E1000_TDH(reg_idx);
2204 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
2205 writel(0, ring->head);
2206 writel(0, ring->tail);
2207
2208 txdctl |= IGB_TX_PTHRESH;
2209 txdctl |= IGB_TX_HTHRESH << 8;
2210 txdctl |= IGB_TX_WTHRESH << 16;
2211
2212 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
2213 wr32(E1000_TXDCTL(reg_idx), txdctl);
2214 }
2215
2216 /**
2217 * igb_configure_tx - Configure transmit Unit after Reset
2218 * @adapter: board private structure
2219 *
2220 * Configure the Tx unit of the MAC after a reset.
2221 **/
2222 static void igb_configure_tx(struct igb_adapter *adapter)
2223 {
2224 int i;
2225
2226 for (i = 0; i < adapter->num_tx_queues; i++)
2227 igb_configure_tx_ring(adapter, &adapter->tx_ring[i]);
2228 }
2229
2230 /**
2231 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
2232 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2233 *
2234 * Returns 0 on success, negative on failure
2235 **/
2236 int igb_setup_rx_resources(struct igb_ring *rx_ring)
2237 {
2238 struct pci_dev *pdev = rx_ring->pdev;
2239 int size, desc_len;
2240
2241 size = sizeof(struct igb_buffer) * rx_ring->count;
2242 rx_ring->buffer_info = vmalloc(size);
2243 if (!rx_ring->buffer_info)
2244 goto err;
2245 memset(rx_ring->buffer_info, 0, size);
2246
2247 desc_len = sizeof(union e1000_adv_rx_desc);
2248
2249 /* Round up to nearest 4K */
2250 rx_ring->size = rx_ring->count * desc_len;
2251 rx_ring->size = ALIGN(rx_ring->size, 4096);
2252
2253 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
2254 &rx_ring->dma);
2255
2256 if (!rx_ring->desc)
2257 goto err;
2258
2259 rx_ring->next_to_clean = 0;
2260 rx_ring->next_to_use = 0;
2261
2262 return 0;
2263
2264 err:
2265 vfree(rx_ring->buffer_info);
2266 rx_ring->buffer_info = NULL;
2267 dev_err(&pdev->dev, "Unable to allocate memory for "
2268 "the receive descriptor ring\n");
2269 return -ENOMEM;
2270 }
2271
2272 /**
2273 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
2274 * (Descriptors) for all queues
2275 * @adapter: board private structure
2276 *
2277 * Return 0 on success, negative on failure
2278 **/
2279 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
2280 {
2281 struct pci_dev *pdev = adapter->pdev;
2282 int i, err = 0;
2283
2284 for (i = 0; i < adapter->num_rx_queues; i++) {
2285 err = igb_setup_rx_resources(&adapter->rx_ring[i]);
2286 if (err) {
2287 dev_err(&pdev->dev,
2288 "Allocation for Rx Queue %u failed\n", i);
2289 for (i--; i >= 0; i--)
2290 igb_free_rx_resources(&adapter->rx_ring[i]);
2291 break;
2292 }
2293 }
2294
2295 return err;
2296 }
2297
2298 /**
2299 * igb_setup_mrqc - configure the multiple receive queue control registers
2300 * @adapter: Board private structure
2301 **/
2302 static void igb_setup_mrqc(struct igb_adapter *adapter)
2303 {
2304 struct e1000_hw *hw = &adapter->hw;
2305 u32 mrqc, rxcsum;
2306 u32 j, num_rx_queues, shift = 0, shift2 = 0;
2307 union e1000_reta {
2308 u32 dword;
2309 u8 bytes[4];
2310 } reta;
2311 static const u8 rsshash[40] = {
2312 0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2, 0x41, 0x67,
2313 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0, 0xd0, 0xca, 0x2b, 0xcb,
2314 0xae, 0x7b, 0x30, 0xb4, 0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30,
2315 0xf2, 0x0c, 0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa };
2316
2317 /* Fill out hash function seeds */
2318 for (j = 0; j < 10; j++) {
2319 u32 rsskey = rsshash[(j * 4)];
2320 rsskey |= rsshash[(j * 4) + 1] << 8;
2321 rsskey |= rsshash[(j * 4) + 2] << 16;
2322 rsskey |= rsshash[(j * 4) + 3] << 24;
2323 array_wr32(E1000_RSSRK(0), j, rsskey);
2324 }
2325
2326 num_rx_queues = adapter->rss_queues;
2327
2328 if (adapter->vfs_allocated_count) {
2329 /* 82575 and 82576 supports 2 RSS queues for VMDq */
2330 switch (hw->mac.type) {
2331 case e1000_82580:
2332 num_rx_queues = 1;
2333 shift = 0;
2334 break;
2335 case e1000_82576:
2336 shift = 3;
2337 num_rx_queues = 2;
2338 break;
2339 case e1000_82575:
2340 shift = 2;
2341 shift2 = 6;
2342 default:
2343 break;
2344 }
2345 } else {
2346 if (hw->mac.type == e1000_82575)
2347 shift = 6;
2348 }
2349
2350 for (j = 0; j < (32 * 4); j++) {
2351 reta.bytes[j & 3] = (j % num_rx_queues) << shift;
2352 if (shift2)
2353 reta.bytes[j & 3] |= num_rx_queues << shift2;
2354 if ((j & 3) == 3)
2355 wr32(E1000_RETA(j >> 2), reta.dword);
2356 }
2357
2358 /*
2359 * Disable raw packet checksumming so that RSS hash is placed in
2360 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
2361 * offloads as they are enabled by default
2362 */
2363 rxcsum = rd32(E1000_RXCSUM);
2364 rxcsum |= E1000_RXCSUM_PCSD;
2365
2366 if (adapter->hw.mac.type >= e1000_82576)
2367 /* Enable Receive Checksum Offload for SCTP */
2368 rxcsum |= E1000_RXCSUM_CRCOFL;
2369
2370 /* Don't need to set TUOFL or IPOFL, they default to 1 */
2371 wr32(E1000_RXCSUM, rxcsum);
2372
2373 /* If VMDq is enabled then we set the appropriate mode for that, else
2374 * we default to RSS so that an RSS hash is calculated per packet even
2375 * if we are only using one queue */
2376 if (adapter->vfs_allocated_count) {
2377 if (hw->mac.type > e1000_82575) {
2378 /* Set the default pool for the PF's first queue */
2379 u32 vtctl = rd32(E1000_VT_CTL);
2380 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2381 E1000_VT_CTL_DISABLE_DEF_POOL);
2382 vtctl |= adapter->vfs_allocated_count <<
2383 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2384 wr32(E1000_VT_CTL, vtctl);
2385 }
2386 if (adapter->rss_queues > 1)
2387 mrqc = E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2388 else
2389 mrqc = E1000_MRQC_ENABLE_VMDQ;
2390 } else {
2391 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2392 }
2393 igb_vmm_control(adapter);
2394
2395 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2396 E1000_MRQC_RSS_FIELD_IPV4_TCP);
2397 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
2398 E1000_MRQC_RSS_FIELD_IPV6_TCP);
2399 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
2400 E1000_MRQC_RSS_FIELD_IPV6_UDP);
2401 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2402 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2403
2404 wr32(E1000_MRQC, mrqc);
2405 }
2406
2407 /**
2408 * igb_setup_rctl - configure the receive control registers
2409 * @adapter: Board private structure
2410 **/
2411 void igb_setup_rctl(struct igb_adapter *adapter)
2412 {
2413 struct e1000_hw *hw = &adapter->hw;
2414 u32 rctl;
2415
2416 rctl = rd32(E1000_RCTL);
2417
2418 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2419 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2420
2421 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2422 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2423
2424 /*
2425 * enable stripping of CRC. It's unlikely this will break BMC
2426 * redirection as it did with e1000. Newer features require
2427 * that the HW strips the CRC.
2428 */
2429 rctl |= E1000_RCTL_SECRC;
2430
2431 /* disable store bad packets and clear size bits. */
2432 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2433
2434 /* enable LPE to prevent packets larger than max_frame_size */
2435 rctl |= E1000_RCTL_LPE;
2436
2437 /* disable queue 0 to prevent tail write w/o re-config */
2438 wr32(E1000_RXDCTL(0), 0);
2439
2440 /* Attention!!! For SR-IOV PF driver operations you must enable
2441 * queue drop for all VF and PF queues to prevent head of line blocking
2442 * if an un-trusted VF does not provide descriptors to hardware.
2443 */
2444 if (adapter->vfs_allocated_count) {
2445 /* set all queue drop enable bits */
2446 wr32(E1000_QDE, ALL_QUEUES);
2447 }
2448
2449 wr32(E1000_RCTL, rctl);
2450 }
2451
2452 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
2453 int vfn)
2454 {
2455 struct e1000_hw *hw = &adapter->hw;
2456 u32 vmolr;
2457
2458 /* if it isn't the PF check to see if VFs are enabled and
2459 * increase the size to support vlan tags */
2460 if (vfn < adapter->vfs_allocated_count &&
2461 adapter->vf_data[vfn].vlans_enabled)
2462 size += VLAN_TAG_SIZE;
2463
2464 vmolr = rd32(E1000_VMOLR(vfn));
2465 vmolr &= ~E1000_VMOLR_RLPML_MASK;
2466 vmolr |= size | E1000_VMOLR_LPE;
2467 wr32(E1000_VMOLR(vfn), vmolr);
2468
2469 return 0;
2470 }
2471
2472 /**
2473 * igb_rlpml_set - set maximum receive packet size
2474 * @adapter: board private structure
2475 *
2476 * Configure maximum receivable packet size.
2477 **/
2478 static void igb_rlpml_set(struct igb_adapter *adapter)
2479 {
2480 u32 max_frame_size = adapter->max_frame_size;
2481 struct e1000_hw *hw = &adapter->hw;
2482 u16 pf_id = adapter->vfs_allocated_count;
2483
2484 if (adapter->vlgrp)
2485 max_frame_size += VLAN_TAG_SIZE;
2486
2487 /* if vfs are enabled we set RLPML to the largest possible request
2488 * size and set the VMOLR RLPML to the size we need */
2489 if (pf_id) {
2490 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
2491 max_frame_size = MAX_JUMBO_FRAME_SIZE;
2492 }
2493
2494 wr32(E1000_RLPML, max_frame_size);
2495 }
2496
2497 static inline void igb_set_vmolr(struct igb_adapter *adapter, int vfn)
2498 {
2499 struct e1000_hw *hw = &adapter->hw;
2500 u32 vmolr;
2501
2502 /*
2503 * This register exists only on 82576 and newer so if we are older then
2504 * we should exit and do nothing
2505 */
2506 if (hw->mac.type < e1000_82576)
2507 return;
2508
2509 vmolr = rd32(E1000_VMOLR(vfn));
2510 vmolr |= E1000_VMOLR_AUPE | /* Accept untagged packets */
2511 E1000_VMOLR_STRVLAN; /* Strip vlan tags */
2512
2513 /* clear all bits that might not be set */
2514 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
2515
2516 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
2517 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
2518 /*
2519 * for VMDq only allow the VFs and pool 0 to accept broadcast and
2520 * multicast packets
2521 */
2522 if (vfn <= adapter->vfs_allocated_count)
2523 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
2524
2525 wr32(E1000_VMOLR(vfn), vmolr);
2526 }
2527
2528 /**
2529 * igb_configure_rx_ring - Configure a receive ring after Reset
2530 * @adapter: board private structure
2531 * @ring: receive ring to be configured
2532 *
2533 * Configure the Rx unit of the MAC after a reset.
2534 **/
2535 void igb_configure_rx_ring(struct igb_adapter *adapter,
2536 struct igb_ring *ring)
2537 {
2538 struct e1000_hw *hw = &adapter->hw;
2539 u64 rdba = ring->dma;
2540 int reg_idx = ring->reg_idx;
2541 u32 srrctl, rxdctl;
2542
2543 /* disable the queue */
2544 rxdctl = rd32(E1000_RXDCTL(reg_idx));
2545 wr32(E1000_RXDCTL(reg_idx),
2546 rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
2547
2548 /* Set DMA base address registers */
2549 wr32(E1000_RDBAL(reg_idx),
2550 rdba & 0x00000000ffffffffULL);
2551 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
2552 wr32(E1000_RDLEN(reg_idx),
2553 ring->count * sizeof(union e1000_adv_rx_desc));
2554
2555 /* initialize head and tail */
2556 ring->head = hw->hw_addr + E1000_RDH(reg_idx);
2557 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
2558 writel(0, ring->head);
2559 writel(0, ring->tail);
2560
2561 /* set descriptor configuration */
2562 if (ring->rx_buffer_len < IGB_RXBUFFER_1024) {
2563 srrctl = ALIGN(ring->rx_buffer_len, 64) <<
2564 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2565 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
2566 srrctl |= IGB_RXBUFFER_16384 >>
2567 E1000_SRRCTL_BSIZEPKT_SHIFT;
2568 #else
2569 srrctl |= (PAGE_SIZE / 2) >>
2570 E1000_SRRCTL_BSIZEPKT_SHIFT;
2571 #endif
2572 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2573 } else {
2574 srrctl = ALIGN(ring->rx_buffer_len, 1024) >>
2575 E1000_SRRCTL_BSIZEPKT_SHIFT;
2576 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2577 }
2578
2579 wr32(E1000_SRRCTL(reg_idx), srrctl);
2580
2581 /* set filtering for VMDQ pools */
2582 igb_set_vmolr(adapter, reg_idx & 0x7);
2583
2584 /* enable receive descriptor fetching */
2585 rxdctl = rd32(E1000_RXDCTL(reg_idx));
2586 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2587 rxdctl &= 0xFFF00000;
2588 rxdctl |= IGB_RX_PTHRESH;
2589 rxdctl |= IGB_RX_HTHRESH << 8;
2590 rxdctl |= IGB_RX_WTHRESH << 16;
2591 wr32(E1000_RXDCTL(reg_idx), rxdctl);
2592 }
2593
2594 /**
2595 * igb_configure_rx - Configure receive Unit after Reset
2596 * @adapter: board private structure
2597 *
2598 * Configure the Rx unit of the MAC after a reset.
2599 **/
2600 static void igb_configure_rx(struct igb_adapter *adapter)
2601 {
2602 int i;
2603
2604 /* set UTA to appropriate mode */
2605 igb_set_uta(adapter);
2606
2607 /* set the correct pool for the PF default MAC address in entry 0 */
2608 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
2609 adapter->vfs_allocated_count);
2610
2611 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2612 * the Base and Length of the Rx Descriptor Ring */
2613 for (i = 0; i < adapter->num_rx_queues; i++)
2614 igb_configure_rx_ring(adapter, &adapter->rx_ring[i]);
2615 }
2616
2617 /**
2618 * igb_free_tx_resources - Free Tx Resources per Queue
2619 * @tx_ring: Tx descriptor ring for a specific queue
2620 *
2621 * Free all transmit software resources
2622 **/
2623 void igb_free_tx_resources(struct igb_ring *tx_ring)
2624 {
2625 igb_clean_tx_ring(tx_ring);
2626
2627 vfree(tx_ring->buffer_info);
2628 tx_ring->buffer_info = NULL;
2629
2630 /* if not set, then don't free */
2631 if (!tx_ring->desc)
2632 return;
2633
2634 pci_free_consistent(tx_ring->pdev, tx_ring->size,
2635 tx_ring->desc, tx_ring->dma);
2636
2637 tx_ring->desc = NULL;
2638 }
2639
2640 /**
2641 * igb_free_all_tx_resources - Free Tx Resources for All Queues
2642 * @adapter: board private structure
2643 *
2644 * Free all transmit software resources
2645 **/
2646 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2647 {
2648 int i;
2649
2650 for (i = 0; i < adapter->num_tx_queues; i++)
2651 igb_free_tx_resources(&adapter->tx_ring[i]);
2652 }
2653
2654 void igb_unmap_and_free_tx_resource(struct igb_ring *tx_ring,
2655 struct igb_buffer *buffer_info)
2656 {
2657 buffer_info->dma = 0;
2658 if (buffer_info->skb) {
2659 skb_dma_unmap(&tx_ring->pdev->dev,
2660 buffer_info->skb,
2661 DMA_TO_DEVICE);
2662 dev_kfree_skb_any(buffer_info->skb);
2663 buffer_info->skb = NULL;
2664 }
2665 buffer_info->time_stamp = 0;
2666 /* buffer_info must be completely set up in the transmit path */
2667 }
2668
2669 /**
2670 * igb_clean_tx_ring - Free Tx Buffers
2671 * @tx_ring: ring to be cleaned
2672 **/
2673 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2674 {
2675 struct igb_buffer *buffer_info;
2676 unsigned long size;
2677 unsigned int i;
2678
2679 if (!tx_ring->buffer_info)
2680 return;
2681 /* Free all the Tx ring sk_buffs */
2682
2683 for (i = 0; i < tx_ring->count; i++) {
2684 buffer_info = &tx_ring->buffer_info[i];
2685 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
2686 }
2687
2688 size = sizeof(struct igb_buffer) * tx_ring->count;
2689 memset(tx_ring->buffer_info, 0, size);
2690
2691 /* Zero out the descriptor ring */
2692 memset(tx_ring->desc, 0, tx_ring->size);
2693
2694 tx_ring->next_to_use = 0;
2695 tx_ring->next_to_clean = 0;
2696 }
2697
2698 /**
2699 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2700 * @adapter: board private structure
2701 **/
2702 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2703 {
2704 int i;
2705
2706 for (i = 0; i < adapter->num_tx_queues; i++)
2707 igb_clean_tx_ring(&adapter->tx_ring[i]);
2708 }
2709
2710 /**
2711 * igb_free_rx_resources - Free Rx Resources
2712 * @rx_ring: ring to clean the resources from
2713 *
2714 * Free all receive software resources
2715 **/
2716 void igb_free_rx_resources(struct igb_ring *rx_ring)
2717 {
2718 igb_clean_rx_ring(rx_ring);
2719
2720 vfree(rx_ring->buffer_info);
2721 rx_ring->buffer_info = NULL;
2722
2723 /* if not set, then don't free */
2724 if (!rx_ring->desc)
2725 return;
2726
2727 pci_free_consistent(rx_ring->pdev, rx_ring->size,
2728 rx_ring->desc, rx_ring->dma);
2729
2730 rx_ring->desc = NULL;
2731 }
2732
2733 /**
2734 * igb_free_all_rx_resources - Free Rx Resources for All Queues
2735 * @adapter: board private structure
2736 *
2737 * Free all receive software resources
2738 **/
2739 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2740 {
2741 int i;
2742
2743 for (i = 0; i < adapter->num_rx_queues; i++)
2744 igb_free_rx_resources(&adapter->rx_ring[i]);
2745 }
2746
2747 /**
2748 * igb_clean_rx_ring - Free Rx Buffers per Queue
2749 * @rx_ring: ring to free buffers from
2750 **/
2751 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2752 {
2753 struct igb_buffer *buffer_info;
2754 unsigned long size;
2755 unsigned int i;
2756
2757 if (!rx_ring->buffer_info)
2758 return;
2759
2760 /* Free all the Rx ring sk_buffs */
2761 for (i = 0; i < rx_ring->count; i++) {
2762 buffer_info = &rx_ring->buffer_info[i];
2763 if (buffer_info->dma) {
2764 pci_unmap_single(rx_ring->pdev,
2765 buffer_info->dma,
2766 rx_ring->rx_buffer_len,
2767 PCI_DMA_FROMDEVICE);
2768 buffer_info->dma = 0;
2769 }
2770
2771 if (buffer_info->skb) {
2772 dev_kfree_skb(buffer_info->skb);
2773 buffer_info->skb = NULL;
2774 }
2775 if (buffer_info->page_dma) {
2776 pci_unmap_page(rx_ring->pdev,
2777 buffer_info->page_dma,
2778 PAGE_SIZE / 2,
2779 PCI_DMA_FROMDEVICE);
2780 buffer_info->page_dma = 0;
2781 }
2782 if (buffer_info->page) {
2783 put_page(buffer_info->page);
2784 buffer_info->page = NULL;
2785 buffer_info->page_offset = 0;
2786 }
2787 }
2788
2789 size = sizeof(struct igb_buffer) * rx_ring->count;
2790 memset(rx_ring->buffer_info, 0, size);
2791
2792 /* Zero out the descriptor ring */
2793 memset(rx_ring->desc, 0, rx_ring->size);
2794
2795 rx_ring->next_to_clean = 0;
2796 rx_ring->next_to_use = 0;
2797 }
2798
2799 /**
2800 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2801 * @adapter: board private structure
2802 **/
2803 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2804 {
2805 int i;
2806
2807 for (i = 0; i < adapter->num_rx_queues; i++)
2808 igb_clean_rx_ring(&adapter->rx_ring[i]);
2809 }
2810
2811 /**
2812 * igb_set_mac - Change the Ethernet Address of the NIC
2813 * @netdev: network interface device structure
2814 * @p: pointer to an address structure
2815 *
2816 * Returns 0 on success, negative on failure
2817 **/
2818 static int igb_set_mac(struct net_device *netdev, void *p)
2819 {
2820 struct igb_adapter *adapter = netdev_priv(netdev);
2821 struct e1000_hw *hw = &adapter->hw;
2822 struct sockaddr *addr = p;
2823
2824 if (!is_valid_ether_addr(addr->sa_data))
2825 return -EADDRNOTAVAIL;
2826
2827 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2828 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
2829
2830 /* set the correct pool for the new PF MAC address in entry 0 */
2831 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
2832 adapter->vfs_allocated_count);
2833
2834 return 0;
2835 }
2836
2837 /**
2838 * igb_write_mc_addr_list - write multicast addresses to MTA
2839 * @netdev: network interface device structure
2840 *
2841 * Writes multicast address list to the MTA hash table.
2842 * Returns: -ENOMEM on failure
2843 * 0 on no addresses written
2844 * X on writing X addresses to MTA
2845 **/
2846 static int igb_write_mc_addr_list(struct net_device *netdev)
2847 {
2848 struct igb_adapter *adapter = netdev_priv(netdev);
2849 struct e1000_hw *hw = &adapter->hw;
2850 struct dev_mc_list *mc_ptr = netdev->mc_list;
2851 u8 *mta_list;
2852 u32 vmolr = 0;
2853 int i;
2854
2855 if (!netdev->mc_count) {
2856 /* nothing to program, so clear mc list */
2857 igb_update_mc_addr_list(hw, NULL, 0);
2858 igb_restore_vf_multicasts(adapter);
2859 return 0;
2860 }
2861
2862 mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2863 if (!mta_list)
2864 return -ENOMEM;
2865
2866 /* set vmolr receive overflow multicast bit */
2867 vmolr |= E1000_VMOLR_ROMPE;
2868
2869 /* The shared function expects a packed array of only addresses. */
2870 mc_ptr = netdev->mc_list;
2871
2872 for (i = 0; i < netdev->mc_count; i++) {
2873 if (!mc_ptr)
2874 break;
2875 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2876 mc_ptr = mc_ptr->next;
2877 }
2878 igb_update_mc_addr_list(hw, mta_list, i);
2879 kfree(mta_list);
2880
2881 return netdev->mc_count;
2882 }
2883
2884 /**
2885 * igb_write_uc_addr_list - write unicast addresses to RAR table
2886 * @netdev: network interface device structure
2887 *
2888 * Writes unicast address list to the RAR table.
2889 * Returns: -ENOMEM on failure/insufficient address space
2890 * 0 on no addresses written
2891 * X on writing X addresses to the RAR table
2892 **/
2893 static int igb_write_uc_addr_list(struct net_device *netdev)
2894 {
2895 struct igb_adapter *adapter = netdev_priv(netdev);
2896 struct e1000_hw *hw = &adapter->hw;
2897 unsigned int vfn = adapter->vfs_allocated_count;
2898 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
2899 int count = 0;
2900
2901 /* return ENOMEM indicating insufficient memory for addresses */
2902 if (netdev->uc.count > rar_entries)
2903 return -ENOMEM;
2904
2905 if (netdev->uc.count && rar_entries) {
2906 struct netdev_hw_addr *ha;
2907 list_for_each_entry(ha, &netdev->uc.list, list) {
2908 if (!rar_entries)
2909 break;
2910 igb_rar_set_qsel(adapter, ha->addr,
2911 rar_entries--,
2912 vfn);
2913 count++;
2914 }
2915 }
2916 /* write the addresses in reverse order to avoid write combining */
2917 for (; rar_entries > 0 ; rar_entries--) {
2918 wr32(E1000_RAH(rar_entries), 0);
2919 wr32(E1000_RAL(rar_entries), 0);
2920 }
2921 wrfl();
2922
2923 return count;
2924 }
2925
2926 /**
2927 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2928 * @netdev: network interface device structure
2929 *
2930 * The set_rx_mode entry point is called whenever the unicast or multicast
2931 * address lists or the network interface flags are updated. This routine is
2932 * responsible for configuring the hardware for proper unicast, multicast,
2933 * promiscuous mode, and all-multi behavior.
2934 **/
2935 static void igb_set_rx_mode(struct net_device *netdev)
2936 {
2937 struct igb_adapter *adapter = netdev_priv(netdev);
2938 struct e1000_hw *hw = &adapter->hw;
2939 unsigned int vfn = adapter->vfs_allocated_count;
2940 u32 rctl, vmolr = 0;
2941 int count;
2942
2943 /* Check for Promiscuous and All Multicast modes */
2944 rctl = rd32(E1000_RCTL);
2945
2946 /* clear the effected bits */
2947 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
2948
2949 if (netdev->flags & IFF_PROMISC) {
2950 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2951 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
2952 } else {
2953 if (netdev->flags & IFF_ALLMULTI) {
2954 rctl |= E1000_RCTL_MPE;
2955 vmolr |= E1000_VMOLR_MPME;
2956 } else {
2957 /*
2958 * Write addresses to the MTA, if the attempt fails
2959 * then we should just turn on promiscous mode so
2960 * that we can at least receive multicast traffic
2961 */
2962 count = igb_write_mc_addr_list(netdev);
2963 if (count < 0) {
2964 rctl |= E1000_RCTL_MPE;
2965 vmolr |= E1000_VMOLR_MPME;
2966 } else if (count) {
2967 vmolr |= E1000_VMOLR_ROMPE;
2968 }
2969 }
2970 /*
2971 * Write addresses to available RAR registers, if there is not
2972 * sufficient space to store all the addresses then enable
2973 * unicast promiscous mode
2974 */
2975 count = igb_write_uc_addr_list(netdev);
2976 if (count < 0) {
2977 rctl |= E1000_RCTL_UPE;
2978 vmolr |= E1000_VMOLR_ROPE;
2979 }
2980 rctl |= E1000_RCTL_VFE;
2981 }
2982 wr32(E1000_RCTL, rctl);
2983
2984 /*
2985 * In order to support SR-IOV and eventually VMDq it is necessary to set
2986 * the VMOLR to enable the appropriate modes. Without this workaround
2987 * we will have issues with VLAN tag stripping not being done for frames
2988 * that are only arriving because we are the default pool
2989 */
2990 if (hw->mac.type < e1000_82576)
2991 return;
2992
2993 vmolr |= rd32(E1000_VMOLR(vfn)) &
2994 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
2995 wr32(E1000_VMOLR(vfn), vmolr);
2996 igb_restore_vf_multicasts(adapter);
2997 }
2998
2999 /* Need to wait a few seconds after link up to get diagnostic information from
3000 * the phy */
3001 static void igb_update_phy_info(unsigned long data)
3002 {
3003 struct igb_adapter *adapter = (struct igb_adapter *) data;
3004 igb_get_phy_info(&adapter->hw);
3005 }
3006
3007 /**
3008 * igb_has_link - check shared code for link and determine up/down
3009 * @adapter: pointer to driver private info
3010 **/
3011 static bool igb_has_link(struct igb_adapter *adapter)
3012 {
3013 struct e1000_hw *hw = &adapter->hw;
3014 bool link_active = false;
3015 s32 ret_val = 0;
3016
3017 /* get_link_status is set on LSC (link status) interrupt or
3018 * rx sequence error interrupt. get_link_status will stay
3019 * false until the e1000_check_for_link establishes link
3020 * for copper adapters ONLY
3021 */
3022 switch (hw->phy.media_type) {
3023 case e1000_media_type_copper:
3024 if (hw->mac.get_link_status) {
3025 ret_val = hw->mac.ops.check_for_link(hw);
3026 link_active = !hw->mac.get_link_status;
3027 } else {
3028 link_active = true;
3029 }
3030 break;
3031 case e1000_media_type_internal_serdes:
3032 ret_val = hw->mac.ops.check_for_link(hw);
3033 link_active = hw->mac.serdes_has_link;
3034 break;
3035 default:
3036 case e1000_media_type_unknown:
3037 break;
3038 }
3039
3040 return link_active;
3041 }
3042
3043 /**
3044 * igb_watchdog - Timer Call-back
3045 * @data: pointer to adapter cast into an unsigned long
3046 **/
3047 static void igb_watchdog(unsigned long data)
3048 {
3049 struct igb_adapter *adapter = (struct igb_adapter *)data;
3050 /* Do the rest outside of interrupt context */
3051 schedule_work(&adapter->watchdog_task);
3052 }
3053
3054 static void igb_watchdog_task(struct work_struct *work)
3055 {
3056 struct igb_adapter *adapter = container_of(work,
3057 struct igb_adapter,
3058 watchdog_task);
3059 struct e1000_hw *hw = &adapter->hw;
3060 struct net_device *netdev = adapter->netdev;
3061 u32 link;
3062 int i;
3063
3064 link = igb_has_link(adapter);
3065 if (link) {
3066 if (!netif_carrier_ok(netdev)) {
3067 u32 ctrl;
3068 hw->mac.ops.get_speed_and_duplex(hw,
3069 &adapter->link_speed,
3070 &adapter->link_duplex);
3071
3072 ctrl = rd32(E1000_CTRL);
3073 /* Links status message must follow this format */
3074 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
3075 "Flow Control: %s\n",
3076 netdev->name,
3077 adapter->link_speed,
3078 adapter->link_duplex == FULL_DUPLEX ?
3079 "Full Duplex" : "Half Duplex",
3080 ((ctrl & E1000_CTRL_TFCE) &&
3081 (ctrl & E1000_CTRL_RFCE)) ? "RX/TX" :
3082 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3083 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None")));
3084
3085 /* tweak tx_queue_len according to speed/duplex and
3086 * adjust the timeout factor */
3087 netdev->tx_queue_len = adapter->tx_queue_len;
3088 adapter->tx_timeout_factor = 1;
3089 switch (adapter->link_speed) {
3090 case SPEED_10:
3091 netdev->tx_queue_len = 10;
3092 adapter->tx_timeout_factor = 14;
3093 break;
3094 case SPEED_100:
3095 netdev->tx_queue_len = 100;
3096 /* maybe add some timeout factor ? */
3097 break;
3098 }
3099
3100 netif_carrier_on(netdev);
3101
3102 igb_ping_all_vfs(adapter);
3103
3104 /* link state has changed, schedule phy info update */
3105 if (!test_bit(__IGB_DOWN, &adapter->state))
3106 mod_timer(&adapter->phy_info_timer,
3107 round_jiffies(jiffies + 2 * HZ));
3108 }
3109 } else {
3110 if (netif_carrier_ok(netdev)) {
3111 adapter->link_speed = 0;
3112 adapter->link_duplex = 0;
3113 /* Links status message must follow this format */
3114 printk(KERN_INFO "igb: %s NIC Link is Down\n",
3115 netdev->name);
3116 netif_carrier_off(netdev);
3117
3118 igb_ping_all_vfs(adapter);
3119
3120 /* link state has changed, schedule phy info update */
3121 if (!test_bit(__IGB_DOWN, &adapter->state))
3122 mod_timer(&adapter->phy_info_timer,
3123 round_jiffies(jiffies + 2 * HZ));
3124 }
3125 }
3126
3127 igb_update_stats(adapter);
3128 igb_update_adaptive(hw);
3129
3130 for (i = 0; i < adapter->num_tx_queues; i++) {
3131 struct igb_ring *tx_ring = &adapter->tx_ring[i];
3132 if (!netif_carrier_ok(netdev)) {
3133 /* We've lost link, so the controller stops DMA,
3134 * but we've got queued Tx work that's never going
3135 * to get done, so reset controller to flush Tx.
3136 * (Do the reset outside of interrupt context). */
3137 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
3138 adapter->tx_timeout_count++;
3139 schedule_work(&adapter->reset_task);
3140 /* return immediately since reset is imminent */
3141 return;
3142 }
3143 }
3144
3145 /* Force detection of hung controller every watchdog period */
3146 tx_ring->detect_tx_hung = true;
3147 }
3148
3149 /* Cause software interrupt to ensure rx ring is cleaned */
3150 if (adapter->msix_entries) {
3151 u32 eics = 0;
3152 for (i = 0; i < adapter->num_q_vectors; i++) {
3153 struct igb_q_vector *q_vector = adapter->q_vector[i];
3154 eics |= q_vector->eims_value;
3155 }
3156 wr32(E1000_EICS, eics);
3157 } else {
3158 wr32(E1000_ICS, E1000_ICS_RXDMT0);
3159 }
3160
3161 /* Reset the timer */
3162 if (!test_bit(__IGB_DOWN, &adapter->state))
3163 mod_timer(&adapter->watchdog_timer,
3164 round_jiffies(jiffies + 2 * HZ));
3165 }
3166
3167 enum latency_range {
3168 lowest_latency = 0,
3169 low_latency = 1,
3170 bulk_latency = 2,
3171 latency_invalid = 255
3172 };
3173
3174 /**
3175 * igb_update_ring_itr - update the dynamic ITR value based on packet size
3176 *
3177 * Stores a new ITR value based on strictly on packet size. This
3178 * algorithm is less sophisticated than that used in igb_update_itr,
3179 * due to the difficulty of synchronizing statistics across multiple
3180 * receive rings. The divisors and thresholds used by this fuction
3181 * were determined based on theoretical maximum wire speed and testing
3182 * data, in order to minimize response time while increasing bulk
3183 * throughput.
3184 * This functionality is controlled by the InterruptThrottleRate module
3185 * parameter (see igb_param.c)
3186 * NOTE: This function is called only when operating in a multiqueue
3187 * receive environment.
3188 * @q_vector: pointer to q_vector
3189 **/
3190 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
3191 {
3192 int new_val = q_vector->itr_val;
3193 int avg_wire_size = 0;
3194 struct igb_adapter *adapter = q_vector->adapter;
3195
3196 /* For non-gigabit speeds, just fix the interrupt rate at 4000
3197 * ints/sec - ITR timer value of 120 ticks.
3198 */
3199 if (adapter->link_speed != SPEED_1000) {
3200 new_val = 976;
3201 goto set_itr_val;
3202 }
3203
3204 if (q_vector->rx_ring && q_vector->rx_ring->total_packets) {
3205 struct igb_ring *ring = q_vector->rx_ring;
3206 avg_wire_size = ring->total_bytes / ring->total_packets;
3207 }
3208
3209 if (q_vector->tx_ring && q_vector->tx_ring->total_packets) {
3210 struct igb_ring *ring = q_vector->tx_ring;
3211 avg_wire_size = max_t(u32, avg_wire_size,
3212 (ring->total_bytes /
3213 ring->total_packets));
3214 }
3215
3216 /* if avg_wire_size isn't set no work was done */
3217 if (!avg_wire_size)
3218 goto clear_counts;
3219
3220 /* Add 24 bytes to size to account for CRC, preamble, and gap */
3221 avg_wire_size += 24;
3222
3223 /* Don't starve jumbo frames */
3224 avg_wire_size = min(avg_wire_size, 3000);
3225
3226 /* Give a little boost to mid-size frames */
3227 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
3228 new_val = avg_wire_size / 3;
3229 else
3230 new_val = avg_wire_size / 2;
3231
3232 set_itr_val:
3233 if (new_val != q_vector->itr_val) {
3234 q_vector->itr_val = new_val;
3235 q_vector->set_itr = 1;
3236 }
3237 clear_counts:
3238 if (q_vector->rx_ring) {
3239 q_vector->rx_ring->total_bytes = 0;
3240 q_vector->rx_ring->total_packets = 0;
3241 }
3242 if (q_vector->tx_ring) {
3243 q_vector->tx_ring->total_bytes = 0;
3244 q_vector->tx_ring->total_packets = 0;
3245 }
3246 }
3247
3248 /**
3249 * igb_update_itr - update the dynamic ITR value based on statistics
3250 * Stores a new ITR value based on packets and byte
3251 * counts during the last interrupt. The advantage of per interrupt
3252 * computation is faster updates and more accurate ITR for the current
3253 * traffic pattern. Constants in this function were computed
3254 * based on theoretical maximum wire speed and thresholds were set based
3255 * on testing data as well as attempting to minimize response time
3256 * while increasing bulk throughput.
3257 * this functionality is controlled by the InterruptThrottleRate module
3258 * parameter (see igb_param.c)
3259 * NOTE: These calculations are only valid when operating in a single-
3260 * queue environment.
3261 * @adapter: pointer to adapter
3262 * @itr_setting: current q_vector->itr_val
3263 * @packets: the number of packets during this measurement interval
3264 * @bytes: the number of bytes during this measurement interval
3265 **/
3266 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
3267 int packets, int bytes)
3268 {
3269 unsigned int retval = itr_setting;
3270
3271 if (packets == 0)
3272 goto update_itr_done;
3273
3274 switch (itr_setting) {
3275 case lowest_latency:
3276 /* handle TSO and jumbo frames */
3277 if (bytes/packets > 8000)
3278 retval = bulk_latency;
3279 else if ((packets < 5) && (bytes > 512))
3280 retval = low_latency;
3281 break;
3282 case low_latency: /* 50 usec aka 20000 ints/s */
3283 if (bytes > 10000) {
3284 /* this if handles the TSO accounting */
3285 if (bytes/packets > 8000) {
3286 retval = bulk_latency;
3287 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
3288 retval = bulk_latency;
3289 } else if ((packets > 35)) {
3290 retval = lowest_latency;
3291 }
3292 } else if (bytes/packets > 2000) {
3293 retval = bulk_latency;
3294 } else if (packets <= 2 && bytes < 512) {
3295 retval = lowest_latency;
3296 }
3297 break;
3298 case bulk_latency: /* 250 usec aka 4000 ints/s */
3299 if (bytes > 25000) {
3300 if (packets > 35)
3301 retval = low_latency;
3302 } else if (bytes < 1500) {
3303 retval = low_latency;
3304 }
3305 break;
3306 }
3307
3308 update_itr_done:
3309 return retval;
3310 }
3311
3312 static void igb_set_itr(struct igb_adapter *adapter)
3313 {
3314 struct igb_q_vector *q_vector = adapter->q_vector[0];
3315 u16 current_itr;
3316 u32 new_itr = q_vector->itr_val;
3317
3318 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
3319 if (adapter->link_speed != SPEED_1000) {
3320 current_itr = 0;
3321 new_itr = 4000;
3322 goto set_itr_now;
3323 }
3324
3325 adapter->rx_itr = igb_update_itr(adapter,
3326 adapter->rx_itr,
3327 adapter->rx_ring->total_packets,
3328 adapter->rx_ring->total_bytes);
3329
3330 adapter->tx_itr = igb_update_itr(adapter,
3331 adapter->tx_itr,
3332 adapter->tx_ring->total_packets,
3333 adapter->tx_ring->total_bytes);
3334 current_itr = max(adapter->rx_itr, adapter->tx_itr);
3335
3336 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3337 if (adapter->rx_itr_setting == 3 && current_itr == lowest_latency)
3338 current_itr = low_latency;
3339
3340 switch (current_itr) {
3341 /* counts and packets in update_itr are dependent on these numbers */
3342 case lowest_latency:
3343 new_itr = 56; /* aka 70,000 ints/sec */
3344 break;
3345 case low_latency:
3346 new_itr = 196; /* aka 20,000 ints/sec */
3347 break;
3348 case bulk_latency:
3349 new_itr = 980; /* aka 4,000 ints/sec */
3350 break;
3351 default:
3352 break;
3353 }
3354
3355 set_itr_now:
3356 adapter->rx_ring->total_bytes = 0;
3357 adapter->rx_ring->total_packets = 0;
3358 adapter->tx_ring->total_bytes = 0;
3359 adapter->tx_ring->total_packets = 0;
3360
3361 if (new_itr != q_vector->itr_val) {
3362 /* this attempts to bias the interrupt rate towards Bulk
3363 * by adding intermediate steps when interrupt rate is
3364 * increasing */
3365 new_itr = new_itr > q_vector->itr_val ?
3366 max((new_itr * q_vector->itr_val) /
3367 (new_itr + (q_vector->itr_val >> 2)),
3368 new_itr) :
3369 new_itr;
3370 /* Don't write the value here; it resets the adapter's
3371 * internal timer, and causes us to delay far longer than
3372 * we should between interrupts. Instead, we write the ITR
3373 * value at the beginning of the next interrupt so the timing
3374 * ends up being correct.
3375 */
3376 q_vector->itr_val = new_itr;
3377 q_vector->set_itr = 1;
3378 }
3379
3380 return;
3381 }
3382
3383 #define IGB_TX_FLAGS_CSUM 0x00000001
3384 #define IGB_TX_FLAGS_VLAN 0x00000002
3385 #define IGB_TX_FLAGS_TSO 0x00000004
3386 #define IGB_TX_FLAGS_IPV4 0x00000008
3387 #define IGB_TX_FLAGS_TSTAMP 0x00000010
3388 #define IGB_TX_FLAGS_VLAN_MASK 0xffff0000
3389 #define IGB_TX_FLAGS_VLAN_SHIFT 16
3390
3391 static inline int igb_tso_adv(struct igb_ring *tx_ring,
3392 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
3393 {
3394 struct e1000_adv_tx_context_desc *context_desc;
3395 unsigned int i;
3396 int err;
3397 struct igb_buffer *buffer_info;
3398 u32 info = 0, tu_cmd = 0;
3399 u32 mss_l4len_idx, l4len;
3400 *hdr_len = 0;
3401
3402 if (skb_header_cloned(skb)) {
3403 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3404 if (err)
3405 return err;
3406 }
3407
3408 l4len = tcp_hdrlen(skb);
3409 *hdr_len += l4len;
3410
3411 if (skb->protocol == htons(ETH_P_IP)) {
3412 struct iphdr *iph = ip_hdr(skb);
3413 iph->tot_len = 0;
3414 iph->check = 0;
3415 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3416 iph->daddr, 0,
3417 IPPROTO_TCP,
3418 0);
3419 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3420 ipv6_hdr(skb)->payload_len = 0;
3421 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3422 &ipv6_hdr(skb)->daddr,
3423 0, IPPROTO_TCP, 0);
3424 }
3425
3426 i = tx_ring->next_to_use;
3427
3428 buffer_info = &tx_ring->buffer_info[i];
3429 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3430 /* VLAN MACLEN IPLEN */
3431 if (tx_flags & IGB_TX_FLAGS_VLAN)
3432 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3433 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3434 *hdr_len += skb_network_offset(skb);
3435 info |= skb_network_header_len(skb);
3436 *hdr_len += skb_network_header_len(skb);
3437 context_desc->vlan_macip_lens = cpu_to_le32(info);
3438
3439 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3440 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3441
3442 if (skb->protocol == htons(ETH_P_IP))
3443 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3444 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3445
3446 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3447
3448 /* MSS L4LEN IDX */
3449 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
3450 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
3451
3452 /* For 82575, context index must be unique per ring. */
3453 if (tx_ring->flags & IGB_RING_FLAG_TX_CTX_IDX)
3454 mss_l4len_idx |= tx_ring->reg_idx << 4;
3455
3456 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3457 context_desc->seqnum_seed = 0;
3458
3459 buffer_info->time_stamp = jiffies;
3460 buffer_info->next_to_watch = i;
3461 buffer_info->dma = 0;
3462 i++;
3463 if (i == tx_ring->count)
3464 i = 0;
3465
3466 tx_ring->next_to_use = i;
3467
3468 return true;
3469 }
3470
3471 static inline bool igb_tx_csum_adv(struct igb_ring *tx_ring,
3472 struct sk_buff *skb, u32 tx_flags)
3473 {
3474 struct e1000_adv_tx_context_desc *context_desc;
3475 struct pci_dev *pdev = tx_ring->pdev;
3476 struct igb_buffer *buffer_info;
3477 u32 info = 0, tu_cmd = 0;
3478 unsigned int i;
3479
3480 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
3481 (tx_flags & IGB_TX_FLAGS_VLAN)) {
3482 i = tx_ring->next_to_use;
3483 buffer_info = &tx_ring->buffer_info[i];
3484 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3485
3486 if (tx_flags & IGB_TX_FLAGS_VLAN)
3487 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3488
3489 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3490 if (skb->ip_summed == CHECKSUM_PARTIAL)
3491 info |= skb_network_header_len(skb);
3492
3493 context_desc->vlan_macip_lens = cpu_to_le32(info);
3494
3495 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3496
3497 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3498 __be16 protocol;
3499
3500 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3501 const struct vlan_ethhdr *vhdr =
3502 (const struct vlan_ethhdr*)skb->data;
3503
3504 protocol = vhdr->h_vlan_encapsulated_proto;
3505 } else {
3506 protocol = skb->protocol;
3507 }
3508
3509 switch (protocol) {
3510 case cpu_to_be16(ETH_P_IP):
3511 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3512 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3513 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3514 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
3515 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
3516 break;
3517 case cpu_to_be16(ETH_P_IPV6):
3518 /* XXX what about other V6 headers?? */
3519 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3520 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3521 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
3522 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
3523 break;
3524 default:
3525 if (unlikely(net_ratelimit()))
3526 dev_warn(&pdev->dev,
3527 "partial checksum but proto=%x!\n",
3528 skb->protocol);
3529 break;
3530 }
3531 }
3532
3533 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3534 context_desc->seqnum_seed = 0;
3535 if (tx_ring->flags & IGB_RING_FLAG_TX_CTX_IDX)
3536 context_desc->mss_l4len_idx =
3537 cpu_to_le32(tx_ring->reg_idx << 4);
3538
3539 buffer_info->time_stamp = jiffies;
3540 buffer_info->next_to_watch = i;
3541 buffer_info->dma = 0;
3542
3543 i++;
3544 if (i == tx_ring->count)
3545 i = 0;
3546 tx_ring->next_to_use = i;
3547
3548 return true;
3549 }
3550 return false;
3551 }
3552
3553 #define IGB_MAX_TXD_PWR 16
3554 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
3555
3556 static inline int igb_tx_map_adv(struct igb_ring *tx_ring, struct sk_buff *skb,
3557 unsigned int first)
3558 {
3559 struct igb_buffer *buffer_info;
3560 struct pci_dev *pdev = tx_ring->pdev;
3561 unsigned int len = skb_headlen(skb);
3562 unsigned int count = 0, i;
3563 unsigned int f;
3564 dma_addr_t *map;
3565
3566 i = tx_ring->next_to_use;
3567
3568 if (skb_dma_map(&pdev->dev, skb, DMA_TO_DEVICE)) {
3569 dev_err(&pdev->dev, "TX DMA map failed\n");
3570 return 0;
3571 }
3572
3573 map = skb_shinfo(skb)->dma_maps;
3574
3575 buffer_info = &tx_ring->buffer_info[i];
3576 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3577 buffer_info->length = len;
3578 /* set time_stamp *before* dma to help avoid a possible race */
3579 buffer_info->time_stamp = jiffies;
3580 buffer_info->next_to_watch = i;
3581 buffer_info->dma = skb_shinfo(skb)->dma_head;
3582
3583 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
3584 struct skb_frag_struct *frag;
3585
3586 i++;
3587 if (i == tx_ring->count)
3588 i = 0;
3589
3590 frag = &skb_shinfo(skb)->frags[f];
3591 len = frag->size;
3592
3593 buffer_info = &tx_ring->buffer_info[i];
3594 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3595 buffer_info->length = len;
3596 buffer_info->time_stamp = jiffies;
3597 buffer_info->next_to_watch = i;
3598 buffer_info->dma = map[count];
3599 count++;
3600 }
3601
3602 tx_ring->buffer_info[i].skb = skb;
3603 tx_ring->buffer_info[first].next_to_watch = i;
3604
3605 return ++count;
3606 }
3607
3608 static inline void igb_tx_queue_adv(struct igb_ring *tx_ring,
3609 int tx_flags, int count, u32 paylen,
3610 u8 hdr_len)
3611 {
3612 union e1000_adv_tx_desc *tx_desc;
3613 struct igb_buffer *buffer_info;
3614 u32 olinfo_status = 0, cmd_type_len;
3615 unsigned int i = tx_ring->next_to_use;
3616
3617 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
3618 E1000_ADVTXD_DCMD_DEXT);
3619
3620 if (tx_flags & IGB_TX_FLAGS_VLAN)
3621 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3622
3623 if (tx_flags & IGB_TX_FLAGS_TSTAMP)
3624 cmd_type_len |= E1000_ADVTXD_MAC_TSTAMP;
3625
3626 if (tx_flags & IGB_TX_FLAGS_TSO) {
3627 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3628
3629 /* insert tcp checksum */
3630 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3631
3632 /* insert ip checksum */
3633 if (tx_flags & IGB_TX_FLAGS_IPV4)
3634 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3635
3636 } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
3637 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3638 }
3639
3640 if ((tx_ring->flags & IGB_RING_FLAG_TX_CTX_IDX) &&
3641 (tx_flags & (IGB_TX_FLAGS_CSUM |
3642 IGB_TX_FLAGS_TSO |
3643 IGB_TX_FLAGS_VLAN)))
3644 olinfo_status |= tx_ring->reg_idx << 4;
3645
3646 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
3647
3648 do {
3649 buffer_info = &tx_ring->buffer_info[i];
3650 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
3651 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
3652 tx_desc->read.cmd_type_len =
3653 cpu_to_le32(cmd_type_len | buffer_info->length);
3654 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3655 count--;
3656 i++;
3657 if (i == tx_ring->count)
3658 i = 0;
3659 } while (count > 0);
3660
3661 tx_desc->read.cmd_type_len |= cpu_to_le32(IGB_ADVTXD_DCMD);
3662 /* Force memory writes to complete before letting h/w
3663 * know there are new descriptors to fetch. (Only
3664 * applicable for weak-ordered memory model archs,
3665 * such as IA-64). */
3666 wmb();
3667
3668 tx_ring->next_to_use = i;
3669 writel(i, tx_ring->tail);
3670 /* we need this if more than one processor can write to our tail
3671 * at a time, it syncronizes IO on IA64/Altix systems */
3672 mmiowb();
3673 }
3674
3675 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, int size)
3676 {
3677 struct net_device *netdev = tx_ring->netdev;
3678
3679 netif_stop_subqueue(netdev, tx_ring->queue_index);
3680
3681 /* Herbert's original patch had:
3682 * smp_mb__after_netif_stop_queue();
3683 * but since that doesn't exist yet, just open code it. */
3684 smp_mb();
3685
3686 /* We need to check again in a case another CPU has just
3687 * made room available. */
3688 if (igb_desc_unused(tx_ring) < size)
3689 return -EBUSY;
3690
3691 /* A reprieve! */
3692 netif_wake_subqueue(netdev, tx_ring->queue_index);
3693 tx_ring->tx_stats.restart_queue++;
3694 return 0;
3695 }
3696
3697 static int igb_maybe_stop_tx(struct igb_ring *tx_ring, int size)
3698 {
3699 if (igb_desc_unused(tx_ring) >= size)
3700 return 0;
3701 return __igb_maybe_stop_tx(tx_ring, size);
3702 }
3703
3704 netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *skb,
3705 struct igb_ring *tx_ring)
3706 {
3707 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
3708 unsigned int first;
3709 unsigned int tx_flags = 0;
3710 u8 hdr_len = 0;
3711 int tso = 0, count;
3712 union skb_shared_tx *shtx = skb_tx(skb);
3713
3714 /* need: 1 descriptor per page,
3715 * + 2 desc gap to keep tail from touching head,
3716 * + 1 desc for skb->data,
3717 * + 1 desc for context descriptor,
3718 * otherwise try next time */
3719 if (igb_maybe_stop_tx(tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
3720 /* this is a hard error */
3721 return NETDEV_TX_BUSY;
3722 }
3723
3724 if (unlikely(shtx->hardware)) {
3725 shtx->in_progress = 1;
3726 tx_flags |= IGB_TX_FLAGS_TSTAMP;
3727 }
3728
3729 if (vlan_tx_tag_present(skb) && adapter->vlgrp) {
3730 tx_flags |= IGB_TX_FLAGS_VLAN;
3731 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
3732 }
3733
3734 if (skb->protocol == htons(ETH_P_IP))
3735 tx_flags |= IGB_TX_FLAGS_IPV4;
3736
3737 first = tx_ring->next_to_use;
3738 if (skb_is_gso(skb)) {
3739 tso = igb_tso_adv(tx_ring, skb, tx_flags, &hdr_len);
3740
3741 if (tso < 0) {
3742 dev_kfree_skb_any(skb);
3743 return NETDEV_TX_OK;
3744 }
3745 }
3746
3747 if (tso)
3748 tx_flags |= IGB_TX_FLAGS_TSO;
3749 else if (igb_tx_csum_adv(tx_ring, skb, tx_flags) &&
3750 (skb->ip_summed == CHECKSUM_PARTIAL))
3751 tx_flags |= IGB_TX_FLAGS_CSUM;
3752
3753 /*
3754 * count reflects descriptors mapped, if 0 or less then mapping error
3755 * has occured and we need to rewind the descriptor queue
3756 */
3757 count = igb_tx_map_adv(tx_ring, skb, first);
3758 if (count <= 0) {
3759 dev_kfree_skb_any(skb);
3760 tx_ring->buffer_info[first].time_stamp = 0;
3761 tx_ring->next_to_use = first;
3762 return NETDEV_TX_OK;
3763 }
3764
3765 igb_tx_queue_adv(tx_ring, tx_flags, count, skb->len, hdr_len);
3766
3767 /* Make sure there is space in the ring for the next send. */
3768 igb_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 4);
3769
3770 return NETDEV_TX_OK;
3771 }
3772
3773 static netdev_tx_t igb_xmit_frame_adv(struct sk_buff *skb,
3774 struct net_device *netdev)
3775 {
3776 struct igb_adapter *adapter = netdev_priv(netdev);
3777 struct igb_ring *tx_ring;
3778 int r_idx = 0;
3779
3780 if (test_bit(__IGB_DOWN, &adapter->state)) {
3781 dev_kfree_skb_any(skb);
3782 return NETDEV_TX_OK;
3783 }
3784
3785 if (skb->len <= 0) {
3786 dev_kfree_skb_any(skb);
3787 return NETDEV_TX_OK;
3788 }
3789
3790 r_idx = skb->queue_mapping & (IGB_ABS_MAX_TX_QUEUES - 1);
3791 tx_ring = adapter->multi_tx_table[r_idx];
3792
3793 /* This goes back to the question of how to logically map a tx queue
3794 * to a flow. Right now, performance is impacted slightly negatively
3795 * if using multiple tx queues. If the stack breaks away from a
3796 * single qdisc implementation, we can look at this again. */
3797 return igb_xmit_frame_ring_adv(skb, tx_ring);
3798 }
3799
3800 /**
3801 * igb_tx_timeout - Respond to a Tx Hang
3802 * @netdev: network interface device structure
3803 **/
3804 static void igb_tx_timeout(struct net_device *netdev)
3805 {
3806 struct igb_adapter *adapter = netdev_priv(netdev);
3807 struct e1000_hw *hw = &adapter->hw;
3808
3809 /* Do the reset outside of interrupt context */
3810 adapter->tx_timeout_count++;
3811
3812 if (hw->mac.type == e1000_82580)
3813 hw->dev_spec._82575.global_device_reset = true;
3814
3815 schedule_work(&adapter->reset_task);
3816 wr32(E1000_EICS,
3817 (adapter->eims_enable_mask & ~adapter->eims_other));
3818 }
3819
3820 static void igb_reset_task(struct work_struct *work)
3821 {
3822 struct igb_adapter *adapter;
3823 adapter = container_of(work, struct igb_adapter, reset_task);
3824
3825 igb_reinit_locked(adapter);
3826 }
3827
3828 /**
3829 * igb_get_stats - Get System Network Statistics
3830 * @netdev: network interface device structure
3831 *
3832 * Returns the address of the device statistics structure.
3833 * The statistics are actually updated from the timer callback.
3834 **/
3835 static struct net_device_stats *igb_get_stats(struct net_device *netdev)
3836 {
3837 /* only return the current stats */
3838 return &netdev->stats;
3839 }
3840
3841 /**
3842 * igb_change_mtu - Change the Maximum Transfer Unit
3843 * @netdev: network interface device structure
3844 * @new_mtu: new value for maximum frame size
3845 *
3846 * Returns 0 on success, negative on failure
3847 **/
3848 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3849 {
3850 struct igb_adapter *adapter = netdev_priv(netdev);
3851 struct pci_dev *pdev = adapter->pdev;
3852 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3853 u32 rx_buffer_len, i;
3854
3855 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3856 dev_err(&pdev->dev, "Invalid MTU setting\n");
3857 return -EINVAL;
3858 }
3859
3860 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3861 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
3862 return -EINVAL;
3863 }
3864
3865 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3866 msleep(1);
3867
3868 /* igb_down has a dependency on max_frame_size */
3869 adapter->max_frame_size = max_frame;
3870
3871 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3872 * means we reserve 2 more, this pushes us to allocate from the next
3873 * larger slab size.
3874 * i.e. RXBUFFER_2048 --> size-4096 slab
3875 */
3876
3877 if (max_frame <= IGB_RXBUFFER_1024)
3878 rx_buffer_len = IGB_RXBUFFER_1024;
3879 else if (max_frame <= MAXIMUM_ETHERNET_VLAN_SIZE)
3880 rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3881 else
3882 rx_buffer_len = IGB_RXBUFFER_128;
3883
3884 if (netif_running(netdev))
3885 igb_down(adapter);
3886
3887 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
3888 netdev->mtu, new_mtu);
3889 netdev->mtu = new_mtu;
3890
3891 for (i = 0; i < adapter->num_rx_queues; i++)
3892 adapter->rx_ring[i].rx_buffer_len = rx_buffer_len;
3893
3894 if (netif_running(netdev))
3895 igb_up(adapter);
3896 else
3897 igb_reset(adapter);
3898
3899 clear_bit(__IGB_RESETTING, &adapter->state);
3900
3901 return 0;
3902 }
3903
3904 /**
3905 * igb_update_stats - Update the board statistics counters
3906 * @adapter: board private structure
3907 **/
3908
3909 void igb_update_stats(struct igb_adapter *adapter)
3910 {
3911 struct net_device_stats *net_stats = igb_get_stats(adapter->netdev);
3912 struct e1000_hw *hw = &adapter->hw;
3913 struct pci_dev *pdev = adapter->pdev;
3914 u32 rnbc;
3915 u16 phy_tmp;
3916 int i;
3917 u64 bytes, packets;
3918
3919 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3920
3921 /*
3922 * Prevent stats update while adapter is being reset, or if the pci
3923 * connection is down.
3924 */
3925 if (adapter->link_speed == 0)
3926 return;
3927 if (pci_channel_offline(pdev))
3928 return;
3929
3930 bytes = 0;
3931 packets = 0;
3932 for (i = 0; i < adapter->num_rx_queues; i++) {
3933 u32 rqdpc_tmp = rd32(E1000_RQDPC(i)) & 0x0FFF;
3934 adapter->rx_ring[i].rx_stats.drops += rqdpc_tmp;
3935 net_stats->rx_fifo_errors += rqdpc_tmp;
3936 bytes += adapter->rx_ring[i].rx_stats.bytes;
3937 packets += adapter->rx_ring[i].rx_stats.packets;
3938 }
3939
3940 net_stats->rx_bytes = bytes;
3941 net_stats->rx_packets = packets;
3942
3943 bytes = 0;
3944 packets = 0;
3945 for (i = 0; i < adapter->num_tx_queues; i++) {
3946 bytes += adapter->tx_ring[i].tx_stats.bytes;
3947 packets += adapter->tx_ring[i].tx_stats.packets;
3948 }
3949 net_stats->tx_bytes = bytes;
3950 net_stats->tx_packets = packets;
3951
3952 /* read stats registers */
3953 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
3954 adapter->stats.gprc += rd32(E1000_GPRC);
3955 adapter->stats.gorc += rd32(E1000_GORCL);
3956 rd32(E1000_GORCH); /* clear GORCL */
3957 adapter->stats.bprc += rd32(E1000_BPRC);
3958 adapter->stats.mprc += rd32(E1000_MPRC);
3959 adapter->stats.roc += rd32(E1000_ROC);
3960
3961 adapter->stats.prc64 += rd32(E1000_PRC64);
3962 adapter->stats.prc127 += rd32(E1000_PRC127);
3963 adapter->stats.prc255 += rd32(E1000_PRC255);
3964 adapter->stats.prc511 += rd32(E1000_PRC511);
3965 adapter->stats.prc1023 += rd32(E1000_PRC1023);
3966 adapter->stats.prc1522 += rd32(E1000_PRC1522);
3967 adapter->stats.symerrs += rd32(E1000_SYMERRS);
3968 adapter->stats.sec += rd32(E1000_SEC);
3969
3970 adapter->stats.mpc += rd32(E1000_MPC);
3971 adapter->stats.scc += rd32(E1000_SCC);
3972 adapter->stats.ecol += rd32(E1000_ECOL);
3973 adapter->stats.mcc += rd32(E1000_MCC);
3974 adapter->stats.latecol += rd32(E1000_LATECOL);
3975 adapter->stats.dc += rd32(E1000_DC);
3976 adapter->stats.rlec += rd32(E1000_RLEC);
3977 adapter->stats.xonrxc += rd32(E1000_XONRXC);
3978 adapter->stats.xontxc += rd32(E1000_XONTXC);
3979 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
3980 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
3981 adapter->stats.fcruc += rd32(E1000_FCRUC);
3982 adapter->stats.gptc += rd32(E1000_GPTC);
3983 adapter->stats.gotc += rd32(E1000_GOTCL);
3984 rd32(E1000_GOTCH); /* clear GOTCL */
3985 rnbc = rd32(E1000_RNBC);
3986 adapter->stats.rnbc += rnbc;
3987 net_stats->rx_fifo_errors += rnbc;
3988 adapter->stats.ruc += rd32(E1000_RUC);
3989 adapter->stats.rfc += rd32(E1000_RFC);
3990 adapter->stats.rjc += rd32(E1000_RJC);
3991 adapter->stats.tor += rd32(E1000_TORH);
3992 adapter->stats.tot += rd32(E1000_TOTH);
3993 adapter->stats.tpr += rd32(E1000_TPR);
3994
3995 adapter->stats.ptc64 += rd32(E1000_PTC64);
3996 adapter->stats.ptc127 += rd32(E1000_PTC127);
3997 adapter->stats.ptc255 += rd32(E1000_PTC255);
3998 adapter->stats.ptc511 += rd32(E1000_PTC511);
3999 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
4000 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
4001
4002 adapter->stats.mptc += rd32(E1000_MPTC);
4003 adapter->stats.bptc += rd32(E1000_BPTC);
4004
4005 /* used for adaptive IFS */
4006 hw->mac.tx_packet_delta = rd32(E1000_TPT);
4007 adapter->stats.tpt += hw->mac.tx_packet_delta;
4008 hw->mac.collision_delta = rd32(E1000_COLC);
4009 adapter->stats.colc += hw->mac.collision_delta;
4010
4011 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
4012 adapter->stats.rxerrc += rd32(E1000_RXERRC);
4013 adapter->stats.tncrs += rd32(E1000_TNCRS);
4014 adapter->stats.tsctc += rd32(E1000_TSCTC);
4015 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
4016
4017 adapter->stats.iac += rd32(E1000_IAC);
4018 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
4019 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
4020 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
4021 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
4022 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
4023 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
4024 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
4025 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
4026
4027 /* Fill out the OS statistics structure */
4028 net_stats->multicast = adapter->stats.mprc;
4029 net_stats->collisions = adapter->stats.colc;
4030
4031 /* Rx Errors */
4032
4033 /* RLEC on some newer hardware can be incorrect so build
4034 * our own version based on RUC and ROC */
4035 net_stats->rx_errors = adapter->stats.rxerrc +
4036 adapter->stats.crcerrs + adapter->stats.algnerrc +
4037 adapter->stats.ruc + adapter->stats.roc +
4038 adapter->stats.cexterr;
4039 net_stats->rx_length_errors = adapter->stats.ruc +
4040 adapter->stats.roc;
4041 net_stats->rx_crc_errors = adapter->stats.crcerrs;
4042 net_stats->rx_frame_errors = adapter->stats.algnerrc;
4043 net_stats->rx_missed_errors = adapter->stats.mpc;
4044
4045 /* Tx Errors */
4046 net_stats->tx_errors = adapter->stats.ecol +
4047 adapter->stats.latecol;
4048 net_stats->tx_aborted_errors = adapter->stats.ecol;
4049 net_stats->tx_window_errors = adapter->stats.latecol;
4050 net_stats->tx_carrier_errors = adapter->stats.tncrs;
4051
4052 /* Tx Dropped needs to be maintained elsewhere */
4053
4054 /* Phy Stats */
4055 if (hw->phy.media_type == e1000_media_type_copper) {
4056 if ((adapter->link_speed == SPEED_1000) &&
4057 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
4058 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
4059 adapter->phy_stats.idle_errors += phy_tmp;
4060 }
4061 }
4062
4063 /* Management Stats */
4064 adapter->stats.mgptc += rd32(E1000_MGTPTC);
4065 adapter->stats.mgprc += rd32(E1000_MGTPRC);
4066 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
4067 }
4068
4069 static irqreturn_t igb_msix_other(int irq, void *data)
4070 {
4071 struct igb_adapter *adapter = data;
4072 struct e1000_hw *hw = &adapter->hw;
4073 u32 icr = rd32(E1000_ICR);
4074 /* reading ICR causes bit 31 of EICR to be cleared */
4075
4076 if (icr & E1000_ICR_DOUTSYNC) {
4077 /* HW is reporting DMA is out of sync */
4078 adapter->stats.doosync++;
4079 }
4080
4081 /* Check for a mailbox event */
4082 if (icr & E1000_ICR_VMMB)
4083 igb_msg_task(adapter);
4084
4085 if (icr & E1000_ICR_LSC) {
4086 hw->mac.get_link_status = 1;
4087 /* guard against interrupt when we're going down */
4088 if (!test_bit(__IGB_DOWN, &adapter->state))
4089 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4090 }
4091
4092 if (adapter->vfs_allocated_count)
4093 wr32(E1000_IMS, E1000_IMS_LSC |
4094 E1000_IMS_VMMB |
4095 E1000_IMS_DOUTSYNC);
4096 else
4097 wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC);
4098 wr32(E1000_EIMS, adapter->eims_other);
4099
4100 return IRQ_HANDLED;
4101 }
4102
4103 static void igb_write_itr(struct igb_q_vector *q_vector)
4104 {
4105 u32 itr_val = q_vector->itr_val & 0x7FFC;
4106
4107 if (!q_vector->set_itr)
4108 return;
4109
4110 if (!itr_val)
4111 itr_val = 0x4;
4112
4113 if (q_vector->itr_shift)
4114 itr_val |= itr_val << q_vector->itr_shift;
4115 else
4116 itr_val |= 0x8000000;
4117
4118 writel(itr_val, q_vector->itr_register);
4119 q_vector->set_itr = 0;
4120 }
4121
4122 static irqreturn_t igb_msix_ring(int irq, void *data)
4123 {
4124 struct igb_q_vector *q_vector = data;
4125
4126 /* Write the ITR value calculated from the previous interrupt. */
4127 igb_write_itr(q_vector);
4128
4129 napi_schedule(&q_vector->napi);
4130
4131 return IRQ_HANDLED;
4132 }
4133
4134 #ifdef CONFIG_IGB_DCA
4135 static void igb_update_dca(struct igb_q_vector *q_vector)
4136 {
4137 struct igb_adapter *adapter = q_vector->adapter;
4138 struct e1000_hw *hw = &adapter->hw;
4139 int cpu = get_cpu();
4140
4141 if (q_vector->cpu == cpu)
4142 goto out_no_update;
4143
4144 if (q_vector->tx_ring) {
4145 int q = q_vector->tx_ring->reg_idx;
4146 u32 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
4147 if (hw->mac.type == e1000_82575) {
4148 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
4149 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4150 } else {
4151 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
4152 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4153 E1000_DCA_TXCTRL_CPUID_SHIFT;
4154 }
4155 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
4156 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
4157 }
4158 if (q_vector->rx_ring) {
4159 int q = q_vector->rx_ring->reg_idx;
4160 u32 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
4161 if (hw->mac.type == e1000_82575) {
4162 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
4163 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4164 } else {
4165 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
4166 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4167 E1000_DCA_RXCTRL_CPUID_SHIFT;
4168 }
4169 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
4170 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
4171 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
4172 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
4173 }
4174 q_vector->cpu = cpu;
4175 out_no_update:
4176 put_cpu();
4177 }
4178
4179 static void igb_setup_dca(struct igb_adapter *adapter)
4180 {
4181 struct e1000_hw *hw = &adapter->hw;
4182 int i;
4183
4184 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
4185 return;
4186
4187 /* Always use CB2 mode, difference is masked in the CB driver. */
4188 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
4189
4190 for (i = 0; i < adapter->num_q_vectors; i++) {
4191 struct igb_q_vector *q_vector = adapter->q_vector[i];
4192 q_vector->cpu = -1;
4193 igb_update_dca(q_vector);
4194 }
4195 }
4196
4197 static int __igb_notify_dca(struct device *dev, void *data)
4198 {
4199 struct net_device *netdev = dev_get_drvdata(dev);
4200 struct igb_adapter *adapter = netdev_priv(netdev);
4201 struct pci_dev *pdev = adapter->pdev;
4202 struct e1000_hw *hw = &adapter->hw;
4203 unsigned long event = *(unsigned long *)data;
4204
4205 switch (event) {
4206 case DCA_PROVIDER_ADD:
4207 /* if already enabled, don't do it again */
4208 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
4209 break;
4210 if (dca_add_requester(dev) == 0) {
4211 adapter->flags |= IGB_FLAG_DCA_ENABLED;
4212 dev_info(&pdev->dev, "DCA enabled\n");
4213 igb_setup_dca(adapter);
4214 break;
4215 }
4216 /* Fall Through since DCA is disabled. */
4217 case DCA_PROVIDER_REMOVE:
4218 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
4219 /* without this a class_device is left
4220 * hanging around in the sysfs model */
4221 dca_remove_requester(dev);
4222 dev_info(&pdev->dev, "DCA disabled\n");
4223 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
4224 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
4225 }
4226 break;
4227 }
4228
4229 return 0;
4230 }
4231
4232 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
4233 void *p)
4234 {
4235 int ret_val;
4236
4237 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
4238 __igb_notify_dca);
4239
4240 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4241 }
4242 #endif /* CONFIG_IGB_DCA */
4243
4244 static void igb_ping_all_vfs(struct igb_adapter *adapter)
4245 {
4246 struct e1000_hw *hw = &adapter->hw;
4247 u32 ping;
4248 int i;
4249
4250 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
4251 ping = E1000_PF_CONTROL_MSG;
4252 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
4253 ping |= E1000_VT_MSGTYPE_CTS;
4254 igb_write_mbx(hw, &ping, 1, i);
4255 }
4256 }
4257
4258 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
4259 {
4260 struct e1000_hw *hw = &adapter->hw;
4261 u32 vmolr = rd32(E1000_VMOLR(vf));
4262 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4263
4264 vf_data->flags |= ~(IGB_VF_FLAG_UNI_PROMISC |
4265 IGB_VF_FLAG_MULTI_PROMISC);
4266 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
4267
4268 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
4269 vmolr |= E1000_VMOLR_MPME;
4270 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
4271 } else {
4272 /*
4273 * if we have hashes and we are clearing a multicast promisc
4274 * flag we need to write the hashes to the MTA as this step
4275 * was previously skipped
4276 */
4277 if (vf_data->num_vf_mc_hashes > 30) {
4278 vmolr |= E1000_VMOLR_MPME;
4279 } else if (vf_data->num_vf_mc_hashes) {
4280 int j;
4281 vmolr |= E1000_VMOLR_ROMPE;
4282 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
4283 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
4284 }
4285 }
4286
4287 wr32(E1000_VMOLR(vf), vmolr);
4288
4289 /* there are flags left unprocessed, likely not supported */
4290 if (*msgbuf & E1000_VT_MSGINFO_MASK)
4291 return -EINVAL;
4292
4293 return 0;
4294
4295 }
4296
4297 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
4298 u32 *msgbuf, u32 vf)
4299 {
4300 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
4301 u16 *hash_list = (u16 *)&msgbuf[1];
4302 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4303 int i;
4304
4305 /* salt away the number of multicast addresses assigned
4306 * to this VF for later use to restore when the PF multi cast
4307 * list changes
4308 */
4309 vf_data->num_vf_mc_hashes = n;
4310
4311 /* only up to 30 hash values supported */
4312 if (n > 30)
4313 n = 30;
4314
4315 /* store the hashes for later use */
4316 for (i = 0; i < n; i++)
4317 vf_data->vf_mc_hashes[i] = hash_list[i];
4318
4319 /* Flush and reset the mta with the new values */
4320 igb_set_rx_mode(adapter->netdev);
4321
4322 return 0;
4323 }
4324
4325 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
4326 {
4327 struct e1000_hw *hw = &adapter->hw;
4328 struct vf_data_storage *vf_data;
4329 int i, j;
4330
4331 for (i = 0; i < adapter->vfs_allocated_count; i++) {
4332 u32 vmolr = rd32(E1000_VMOLR(i));
4333 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
4334
4335 vf_data = &adapter->vf_data[i];
4336
4337 if ((vf_data->num_vf_mc_hashes > 30) ||
4338 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
4339 vmolr |= E1000_VMOLR_MPME;
4340 } else if (vf_data->num_vf_mc_hashes) {
4341 vmolr |= E1000_VMOLR_ROMPE;
4342 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
4343 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
4344 }
4345 wr32(E1000_VMOLR(i), vmolr);
4346 }
4347 }
4348
4349 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
4350 {
4351 struct e1000_hw *hw = &adapter->hw;
4352 u32 pool_mask, reg, vid;
4353 int i;
4354
4355 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
4356
4357 /* Find the vlan filter for this id */
4358 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4359 reg = rd32(E1000_VLVF(i));
4360
4361 /* remove the vf from the pool */
4362 reg &= ~pool_mask;
4363
4364 /* if pool is empty then remove entry from vfta */
4365 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
4366 (reg & E1000_VLVF_VLANID_ENABLE)) {
4367 reg = 0;
4368 vid = reg & E1000_VLVF_VLANID_MASK;
4369 igb_vfta_set(hw, vid, false);
4370 }
4371
4372 wr32(E1000_VLVF(i), reg);
4373 }
4374
4375 adapter->vf_data[vf].vlans_enabled = 0;
4376 }
4377
4378 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
4379 {
4380 struct e1000_hw *hw = &adapter->hw;
4381 u32 reg, i;
4382
4383 /* The vlvf table only exists on 82576 hardware and newer */
4384 if (hw->mac.type < e1000_82576)
4385 return -1;
4386
4387 /* we only need to do this if VMDq is enabled */
4388 if (!adapter->vfs_allocated_count)
4389 return -1;
4390
4391 /* Find the vlan filter for this id */
4392 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4393 reg = rd32(E1000_VLVF(i));
4394 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
4395 vid == (reg & E1000_VLVF_VLANID_MASK))
4396 break;
4397 }
4398
4399 if (add) {
4400 if (i == E1000_VLVF_ARRAY_SIZE) {
4401 /* Did not find a matching VLAN ID entry that was
4402 * enabled. Search for a free filter entry, i.e.
4403 * one without the enable bit set
4404 */
4405 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4406 reg = rd32(E1000_VLVF(i));
4407 if (!(reg & E1000_VLVF_VLANID_ENABLE))
4408 break;
4409 }
4410 }
4411 if (i < E1000_VLVF_ARRAY_SIZE) {
4412 /* Found an enabled/available entry */
4413 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
4414
4415 /* if !enabled we need to set this up in vfta */
4416 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
4417 /* add VID to filter table */
4418 igb_vfta_set(hw, vid, true);
4419 reg |= E1000_VLVF_VLANID_ENABLE;
4420 }
4421 reg &= ~E1000_VLVF_VLANID_MASK;
4422 reg |= vid;
4423 wr32(E1000_VLVF(i), reg);
4424
4425 /* do not modify RLPML for PF devices */
4426 if (vf >= adapter->vfs_allocated_count)
4427 return 0;
4428
4429 if (!adapter->vf_data[vf].vlans_enabled) {
4430 u32 size;
4431 reg = rd32(E1000_VMOLR(vf));
4432 size = reg & E1000_VMOLR_RLPML_MASK;
4433 size += 4;
4434 reg &= ~E1000_VMOLR_RLPML_MASK;
4435 reg |= size;
4436 wr32(E1000_VMOLR(vf), reg);
4437 }
4438
4439 adapter->vf_data[vf].vlans_enabled++;
4440 return 0;
4441 }
4442 } else {
4443 if (i < E1000_VLVF_ARRAY_SIZE) {
4444 /* remove vf from the pool */
4445 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
4446 /* if pool is empty then remove entry from vfta */
4447 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
4448 reg = 0;
4449 igb_vfta_set(hw, vid, false);
4450 }
4451 wr32(E1000_VLVF(i), reg);
4452
4453 /* do not modify RLPML for PF devices */
4454 if (vf >= adapter->vfs_allocated_count)
4455 return 0;
4456
4457 adapter->vf_data[vf].vlans_enabled--;
4458 if (!adapter->vf_data[vf].vlans_enabled) {
4459 u32 size;
4460 reg = rd32(E1000_VMOLR(vf));
4461 size = reg & E1000_VMOLR_RLPML_MASK;
4462 size -= 4;
4463 reg &= ~E1000_VMOLR_RLPML_MASK;
4464 reg |= size;
4465 wr32(E1000_VMOLR(vf), reg);
4466 }
4467 return 0;
4468 }
4469 }
4470 return -1;
4471 }
4472
4473 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
4474 {
4475 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
4476 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
4477
4478 return igb_vlvf_set(adapter, vid, add, vf);
4479 }
4480
4481 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
4482 {
4483 /* clear all flags */
4484 adapter->vf_data[vf].flags = 0;
4485 adapter->vf_data[vf].last_nack = jiffies;
4486
4487 /* reset offloads to defaults */
4488 igb_set_vmolr(adapter, vf);
4489
4490 /* reset vlans for device */
4491 igb_clear_vf_vfta(adapter, vf);
4492
4493 /* reset multicast table array for vf */
4494 adapter->vf_data[vf].num_vf_mc_hashes = 0;
4495
4496 /* Flush and reset the mta with the new values */
4497 igb_set_rx_mode(adapter->netdev);
4498 }
4499
4500 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
4501 {
4502 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4503
4504 /* generate a new mac address as we were hotplug removed/added */
4505 random_ether_addr(vf_mac);
4506
4507 /* process remaining reset events */
4508 igb_vf_reset(adapter, vf);
4509 }
4510
4511 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
4512 {
4513 struct e1000_hw *hw = &adapter->hw;
4514 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4515 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
4516 u32 reg, msgbuf[3];
4517 u8 *addr = (u8 *)(&msgbuf[1]);
4518
4519 /* process all the same items cleared in a function level reset */
4520 igb_vf_reset(adapter, vf);
4521
4522 /* set vf mac address */
4523 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
4524
4525 /* enable transmit and receive for vf */
4526 reg = rd32(E1000_VFTE);
4527 wr32(E1000_VFTE, reg | (1 << vf));
4528 reg = rd32(E1000_VFRE);
4529 wr32(E1000_VFRE, reg | (1 << vf));
4530
4531 adapter->vf_data[vf].flags = IGB_VF_FLAG_CTS;
4532
4533 /* reply to reset with ack and vf mac address */
4534 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
4535 memcpy(addr, vf_mac, 6);
4536 igb_write_mbx(hw, msgbuf, 3, vf);
4537 }
4538
4539 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
4540 {
4541 unsigned char *addr = (char *)&msg[1];
4542 int err = -1;
4543
4544 if (is_valid_ether_addr(addr))
4545 err = igb_set_vf_mac(adapter, vf, addr);
4546
4547 return err;
4548 }
4549
4550 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
4551 {
4552 struct e1000_hw *hw = &adapter->hw;
4553 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4554 u32 msg = E1000_VT_MSGTYPE_NACK;
4555
4556 /* if device isn't clear to send it shouldn't be reading either */
4557 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
4558 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
4559 igb_write_mbx(hw, &msg, 1, vf);
4560 vf_data->last_nack = jiffies;
4561 }
4562 }
4563
4564 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
4565 {
4566 struct pci_dev *pdev = adapter->pdev;
4567 u32 msgbuf[E1000_VFMAILBOX_SIZE];
4568 struct e1000_hw *hw = &adapter->hw;
4569 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4570 s32 retval;
4571
4572 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
4573
4574 if (retval)
4575 dev_err(&pdev->dev, "Error receiving message from VF\n");
4576
4577 /* this is a message we already processed, do nothing */
4578 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
4579 return;
4580
4581 /*
4582 * until the vf completes a reset it should not be
4583 * allowed to start any configuration.
4584 */
4585
4586 if (msgbuf[0] == E1000_VF_RESET) {
4587 igb_vf_reset_msg(adapter, vf);
4588 return;
4589 }
4590
4591 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
4592 msgbuf[0] = E1000_VT_MSGTYPE_NACK;
4593 if (time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
4594 igb_write_mbx(hw, msgbuf, 1, vf);
4595 vf_data->last_nack = jiffies;
4596 }
4597 return;
4598 }
4599
4600 switch ((msgbuf[0] & 0xFFFF)) {
4601 case E1000_VF_SET_MAC_ADDR:
4602 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
4603 break;
4604 case E1000_VF_SET_PROMISC:
4605 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
4606 break;
4607 case E1000_VF_SET_MULTICAST:
4608 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
4609 break;
4610 case E1000_VF_SET_LPE:
4611 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
4612 break;
4613 case E1000_VF_SET_VLAN:
4614 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
4615 break;
4616 default:
4617 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
4618 retval = -1;
4619 break;
4620 }
4621
4622 /* notify the VF of the results of what it sent us */
4623 if (retval)
4624 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4625 else
4626 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
4627
4628 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
4629
4630 igb_write_mbx(hw, msgbuf, 1, vf);
4631 }
4632
4633 static void igb_msg_task(struct igb_adapter *adapter)
4634 {
4635 struct e1000_hw *hw = &adapter->hw;
4636 u32 vf;
4637
4638 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
4639 /* process any reset requests */
4640 if (!igb_check_for_rst(hw, vf))
4641 igb_vf_reset_event(adapter, vf);
4642
4643 /* process any messages pending */
4644 if (!igb_check_for_msg(hw, vf))
4645 igb_rcv_msg_from_vf(adapter, vf);
4646
4647 /* process any acks */
4648 if (!igb_check_for_ack(hw, vf))
4649 igb_rcv_ack_from_vf(adapter, vf);
4650 }
4651 }
4652
4653 /**
4654 * igb_set_uta - Set unicast filter table address
4655 * @adapter: board private structure
4656 *
4657 * The unicast table address is a register array of 32-bit registers.
4658 * The table is meant to be used in a way similar to how the MTA is used
4659 * however due to certain limitations in the hardware it is necessary to
4660 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscous
4661 * enable bit to allow vlan tag stripping when promiscous mode is enabled
4662 **/
4663 static void igb_set_uta(struct igb_adapter *adapter)
4664 {
4665 struct e1000_hw *hw = &adapter->hw;
4666 int i;
4667
4668 /* The UTA table only exists on 82576 hardware and newer */
4669 if (hw->mac.type < e1000_82576)
4670 return;
4671
4672 /* we only need to do this if VMDq is enabled */
4673 if (!adapter->vfs_allocated_count)
4674 return;
4675
4676 for (i = 0; i < hw->mac.uta_reg_count; i++)
4677 array_wr32(E1000_UTA, i, ~0);
4678 }
4679
4680 /**
4681 * igb_intr_msi - Interrupt Handler
4682 * @irq: interrupt number
4683 * @data: pointer to a network interface device structure
4684 **/
4685 static irqreturn_t igb_intr_msi(int irq, void *data)
4686 {
4687 struct igb_adapter *adapter = data;
4688 struct igb_q_vector *q_vector = adapter->q_vector[0];
4689 struct e1000_hw *hw = &adapter->hw;
4690 /* read ICR disables interrupts using IAM */
4691 u32 icr = rd32(E1000_ICR);
4692
4693 igb_write_itr(q_vector);
4694
4695 if (icr & E1000_ICR_DOUTSYNC) {
4696 /* HW is reporting DMA is out of sync */
4697 adapter->stats.doosync++;
4698 }
4699
4700 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4701 hw->mac.get_link_status = 1;
4702 if (!test_bit(__IGB_DOWN, &adapter->state))
4703 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4704 }
4705
4706 napi_schedule(&q_vector->napi);
4707
4708 return IRQ_HANDLED;
4709 }
4710
4711 /**
4712 * igb_intr - Legacy Interrupt Handler
4713 * @irq: interrupt number
4714 * @data: pointer to a network interface device structure
4715 **/
4716 static irqreturn_t igb_intr(int irq, void *data)
4717 {
4718 struct igb_adapter *adapter = data;
4719 struct igb_q_vector *q_vector = adapter->q_vector[0];
4720 struct e1000_hw *hw = &adapter->hw;
4721 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
4722 * need for the IMC write */
4723 u32 icr = rd32(E1000_ICR);
4724 if (!icr)
4725 return IRQ_NONE; /* Not our interrupt */
4726
4727 igb_write_itr(q_vector);
4728
4729 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
4730 * not set, then the adapter didn't send an interrupt */
4731 if (!(icr & E1000_ICR_INT_ASSERTED))
4732 return IRQ_NONE;
4733
4734 if (icr & E1000_ICR_DOUTSYNC) {
4735 /* HW is reporting DMA is out of sync */
4736 adapter->stats.doosync++;
4737 }
4738
4739 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4740 hw->mac.get_link_status = 1;
4741 /* guard against interrupt when we're going down */
4742 if (!test_bit(__IGB_DOWN, &adapter->state))
4743 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4744 }
4745
4746 napi_schedule(&q_vector->napi);
4747
4748 return IRQ_HANDLED;
4749 }
4750
4751 static inline void igb_ring_irq_enable(struct igb_q_vector *q_vector)
4752 {
4753 struct igb_adapter *adapter = q_vector->adapter;
4754 struct e1000_hw *hw = &adapter->hw;
4755
4756 if ((q_vector->rx_ring && (adapter->rx_itr_setting & 3)) ||
4757 (!q_vector->rx_ring && (adapter->tx_itr_setting & 3))) {
4758 if (!adapter->msix_entries)
4759 igb_set_itr(adapter);
4760 else
4761 igb_update_ring_itr(q_vector);
4762 }
4763
4764 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4765 if (adapter->msix_entries)
4766 wr32(E1000_EIMS, q_vector->eims_value);
4767 else
4768 igb_irq_enable(adapter);
4769 }
4770 }
4771
4772 /**
4773 * igb_poll - NAPI Rx polling callback
4774 * @napi: napi polling structure
4775 * @budget: count of how many packets we should handle
4776 **/
4777 static int igb_poll(struct napi_struct *napi, int budget)
4778 {
4779 struct igb_q_vector *q_vector = container_of(napi,
4780 struct igb_q_vector,
4781 napi);
4782 int tx_clean_complete = 1, work_done = 0;
4783
4784 #ifdef CONFIG_IGB_DCA
4785 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
4786 igb_update_dca(q_vector);
4787 #endif
4788 if (q_vector->tx_ring)
4789 tx_clean_complete = igb_clean_tx_irq(q_vector);
4790
4791 if (q_vector->rx_ring)
4792 igb_clean_rx_irq_adv(q_vector, &work_done, budget);
4793
4794 if (!tx_clean_complete)
4795 work_done = budget;
4796
4797 /* If not enough Rx work done, exit the polling mode */
4798 if (work_done < budget) {
4799 napi_complete(napi);
4800 igb_ring_irq_enable(q_vector);
4801 }
4802
4803 return work_done;
4804 }
4805
4806 /**
4807 * igb_systim_to_hwtstamp - convert system time value to hw timestamp
4808 * @adapter: board private structure
4809 * @shhwtstamps: timestamp structure to update
4810 * @regval: unsigned 64bit system time value.
4811 *
4812 * We need to convert the system time value stored in the RX/TXSTMP registers
4813 * into a hwtstamp which can be used by the upper level timestamping functions
4814 */
4815 static void igb_systim_to_hwtstamp(struct igb_adapter *adapter,
4816 struct skb_shared_hwtstamps *shhwtstamps,
4817 u64 regval)
4818 {
4819 u64 ns;
4820
4821 /*
4822 * The 82580 starts with 1ns at bit 0 in RX/TXSTMPL, shift this up to
4823 * 24 to match clock shift we setup earlier.
4824 */
4825 if (adapter->hw.mac.type == e1000_82580)
4826 regval <<= IGB_82580_TSYNC_SHIFT;
4827
4828 ns = timecounter_cyc2time(&adapter->clock, regval);
4829 timecompare_update(&adapter->compare, ns);
4830 memset(shhwtstamps, 0, sizeof(struct skb_shared_hwtstamps));
4831 shhwtstamps->hwtstamp = ns_to_ktime(ns);
4832 shhwtstamps->syststamp = timecompare_transform(&adapter->compare, ns);
4833 }
4834
4835 /**
4836 * igb_tx_hwtstamp - utility function which checks for TX time stamp
4837 * @q_vector: pointer to q_vector containing needed info
4838 * @skb: packet that was just sent
4839 *
4840 * If we were asked to do hardware stamping and such a time stamp is
4841 * available, then it must have been for this skb here because we only
4842 * allow only one such packet into the queue.
4843 */
4844 static void igb_tx_hwtstamp(struct igb_q_vector *q_vector, struct sk_buff *skb)
4845 {
4846 struct igb_adapter *adapter = q_vector->adapter;
4847 union skb_shared_tx *shtx = skb_tx(skb);
4848 struct e1000_hw *hw = &adapter->hw;
4849 struct skb_shared_hwtstamps shhwtstamps;
4850 u64 regval;
4851
4852 /* if skb does not support hw timestamp or TX stamp not valid exit */
4853 if (likely(!shtx->hardware) ||
4854 !(rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID))
4855 return;
4856
4857 regval = rd32(E1000_TXSTMPL);
4858 regval |= (u64)rd32(E1000_TXSTMPH) << 32;
4859
4860 igb_systim_to_hwtstamp(adapter, &shhwtstamps, regval);
4861 skb_tstamp_tx(skb, &shhwtstamps);
4862 }
4863
4864 /**
4865 * igb_clean_tx_irq - Reclaim resources after transmit completes
4866 * @q_vector: pointer to q_vector containing needed info
4867 * returns true if ring is completely cleaned
4868 **/
4869 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
4870 {
4871 struct igb_adapter *adapter = q_vector->adapter;
4872 struct igb_ring *tx_ring = q_vector->tx_ring;
4873 struct net_device *netdev = tx_ring->netdev;
4874 struct e1000_hw *hw = &adapter->hw;
4875 struct igb_buffer *buffer_info;
4876 struct sk_buff *skb;
4877 union e1000_adv_tx_desc *tx_desc, *eop_desc;
4878 unsigned int total_bytes = 0, total_packets = 0;
4879 unsigned int i, eop, count = 0;
4880 bool cleaned = false;
4881
4882 i = tx_ring->next_to_clean;
4883 eop = tx_ring->buffer_info[i].next_to_watch;
4884 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4885
4886 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
4887 (count < tx_ring->count)) {
4888 for (cleaned = false; !cleaned; count++) {
4889 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
4890 buffer_info = &tx_ring->buffer_info[i];
4891 cleaned = (i == eop);
4892 skb = buffer_info->skb;
4893
4894 if (skb) {
4895 unsigned int segs, bytecount;
4896 /* gso_segs is currently only valid for tcp */
4897 segs = skb_shinfo(skb)->gso_segs ?: 1;
4898 /* multiply data chunks by size of headers */
4899 bytecount = ((segs - 1) * skb_headlen(skb)) +
4900 skb->len;
4901 total_packets += segs;
4902 total_bytes += bytecount;
4903
4904 igb_tx_hwtstamp(q_vector, skb);
4905 }
4906
4907 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
4908 tx_desc->wb.status = 0;
4909
4910 i++;
4911 if (i == tx_ring->count)
4912 i = 0;
4913 }
4914 eop = tx_ring->buffer_info[i].next_to_watch;
4915 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4916 }
4917
4918 tx_ring->next_to_clean = i;
4919
4920 if (unlikely(count &&
4921 netif_carrier_ok(netdev) &&
4922 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
4923 /* Make sure that anybody stopping the queue after this
4924 * sees the new next_to_clean.
4925 */
4926 smp_mb();
4927 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
4928 !(test_bit(__IGB_DOWN, &adapter->state))) {
4929 netif_wake_subqueue(netdev, tx_ring->queue_index);
4930 tx_ring->tx_stats.restart_queue++;
4931 }
4932 }
4933
4934 if (tx_ring->detect_tx_hung) {
4935 /* Detect a transmit hang in hardware, this serializes the
4936 * check with the clearing of time_stamp and movement of i */
4937 tx_ring->detect_tx_hung = false;
4938 if (tx_ring->buffer_info[i].time_stamp &&
4939 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
4940 (adapter->tx_timeout_factor * HZ))
4941 && !(rd32(E1000_STATUS) &
4942 E1000_STATUS_TXOFF)) {
4943
4944 /* detected Tx unit hang */
4945 dev_err(&tx_ring->pdev->dev,
4946 "Detected Tx Unit Hang\n"
4947 " Tx Queue <%d>\n"
4948 " TDH <%x>\n"
4949 " TDT <%x>\n"
4950 " next_to_use <%x>\n"
4951 " next_to_clean <%x>\n"
4952 "buffer_info[next_to_clean]\n"
4953 " time_stamp <%lx>\n"
4954 " next_to_watch <%x>\n"
4955 " jiffies <%lx>\n"
4956 " desc.status <%x>\n",
4957 tx_ring->queue_index,
4958 readl(tx_ring->head),
4959 readl(tx_ring->tail),
4960 tx_ring->next_to_use,
4961 tx_ring->next_to_clean,
4962 tx_ring->buffer_info[eop].time_stamp,
4963 eop,
4964 jiffies,
4965 eop_desc->wb.status);
4966 netif_stop_subqueue(netdev, tx_ring->queue_index);
4967 }
4968 }
4969 tx_ring->total_bytes += total_bytes;
4970 tx_ring->total_packets += total_packets;
4971 tx_ring->tx_stats.bytes += total_bytes;
4972 tx_ring->tx_stats.packets += total_packets;
4973 return (count < tx_ring->count);
4974 }
4975
4976 /**
4977 * igb_receive_skb - helper function to handle rx indications
4978 * @q_vector: structure containing interrupt and ring information
4979 * @skb: packet to send up
4980 * @vlan_tag: vlan tag for packet
4981 **/
4982 static void igb_receive_skb(struct igb_q_vector *q_vector,
4983 struct sk_buff *skb,
4984 u16 vlan_tag)
4985 {
4986 struct igb_adapter *adapter = q_vector->adapter;
4987
4988 if (vlan_tag)
4989 vlan_gro_receive(&q_vector->napi, adapter->vlgrp,
4990 vlan_tag, skb);
4991 else
4992 napi_gro_receive(&q_vector->napi, skb);
4993 }
4994
4995 static inline void igb_rx_checksum_adv(struct igb_ring *ring,
4996 u32 status_err, struct sk_buff *skb)
4997 {
4998 skb->ip_summed = CHECKSUM_NONE;
4999
5000 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
5001 if (!(ring->flags & IGB_RING_FLAG_RX_CSUM) ||
5002 (status_err & E1000_RXD_STAT_IXSM))
5003 return;
5004
5005 /* TCP/UDP checksum error bit is set */
5006 if (status_err &
5007 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
5008 /*
5009 * work around errata with sctp packets where the TCPE aka
5010 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
5011 * packets, (aka let the stack check the crc32c)
5012 */
5013 if ((skb->len == 60) &&
5014 (ring->flags & IGB_RING_FLAG_RX_SCTP_CSUM))
5015 ring->rx_stats.csum_err++;
5016
5017 /* let the stack verify checksum errors */
5018 return;
5019 }
5020 /* It must be a TCP or UDP packet with a valid checksum */
5021 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
5022 skb->ip_summed = CHECKSUM_UNNECESSARY;
5023
5024 dev_dbg(&ring->pdev->dev, "cksum success: bits %08X\n", status_err);
5025 }
5026
5027 static inline void igb_rx_hwtstamp(struct igb_q_vector *q_vector, u32 staterr,
5028 struct sk_buff *skb)
5029 {
5030 struct igb_adapter *adapter = q_vector->adapter;
5031 struct e1000_hw *hw = &adapter->hw;
5032 u64 regval;
5033
5034 /*
5035 * If this bit is set, then the RX registers contain the time stamp. No
5036 * other packet will be time stamped until we read these registers, so
5037 * read the registers to make them available again. Because only one
5038 * packet can be time stamped at a time, we know that the register
5039 * values must belong to this one here and therefore we don't need to
5040 * compare any of the additional attributes stored for it.
5041 *
5042 * If nothing went wrong, then it should have a skb_shared_tx that we
5043 * can turn into a skb_shared_hwtstamps.
5044 */
5045 if (likely(!(staterr & E1000_RXDADV_STAT_TS)))
5046 return;
5047 if (!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
5048 return;
5049
5050 regval = rd32(E1000_RXSTMPL);
5051 regval |= (u64)rd32(E1000_RXSTMPH) << 32;
5052
5053 igb_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), regval);
5054 }
5055 static inline u16 igb_get_hlen(struct igb_ring *rx_ring,
5056 union e1000_adv_rx_desc *rx_desc)
5057 {
5058 /* HW will not DMA in data larger than the given buffer, even if it
5059 * parses the (NFS, of course) header to be larger. In that case, it
5060 * fills the header buffer and spills the rest into the page.
5061 */
5062 u16 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
5063 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
5064 if (hlen > rx_ring->rx_buffer_len)
5065 hlen = rx_ring->rx_buffer_len;
5066 return hlen;
5067 }
5068
5069 static bool igb_clean_rx_irq_adv(struct igb_q_vector *q_vector,
5070 int *work_done, int budget)
5071 {
5072 struct igb_ring *rx_ring = q_vector->rx_ring;
5073 struct net_device *netdev = rx_ring->netdev;
5074 struct pci_dev *pdev = rx_ring->pdev;
5075 union e1000_adv_rx_desc *rx_desc , *next_rxd;
5076 struct igb_buffer *buffer_info , *next_buffer;
5077 struct sk_buff *skb;
5078 bool cleaned = false;
5079 int cleaned_count = 0;
5080 int current_node = numa_node_id();
5081 unsigned int total_bytes = 0, total_packets = 0;
5082 unsigned int i;
5083 u32 staterr;
5084 u16 length;
5085 u16 vlan_tag;
5086
5087 i = rx_ring->next_to_clean;
5088 buffer_info = &rx_ring->buffer_info[i];
5089 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
5090 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
5091
5092 while (staterr & E1000_RXD_STAT_DD) {
5093 if (*work_done >= budget)
5094 break;
5095 (*work_done)++;
5096
5097 skb = buffer_info->skb;
5098 prefetch(skb->data - NET_IP_ALIGN);
5099 buffer_info->skb = NULL;
5100
5101 i++;
5102 if (i == rx_ring->count)
5103 i = 0;
5104
5105 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
5106 prefetch(next_rxd);
5107 next_buffer = &rx_ring->buffer_info[i];
5108
5109 length = le16_to_cpu(rx_desc->wb.upper.length);
5110 cleaned = true;
5111 cleaned_count++;
5112
5113 if (buffer_info->dma) {
5114 pci_unmap_single(pdev, buffer_info->dma,
5115 rx_ring->rx_buffer_len,
5116 PCI_DMA_FROMDEVICE);
5117 buffer_info->dma = 0;
5118 if (rx_ring->rx_buffer_len >= IGB_RXBUFFER_1024) {
5119 skb_put(skb, length);
5120 goto send_up;
5121 }
5122 skb_put(skb, igb_get_hlen(rx_ring, rx_desc));
5123 }
5124
5125 if (length) {
5126 pci_unmap_page(pdev, buffer_info->page_dma,
5127 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
5128 buffer_info->page_dma = 0;
5129
5130 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
5131 buffer_info->page,
5132 buffer_info->page_offset,
5133 length);
5134
5135 if ((page_count(buffer_info->page) != 1) ||
5136 (page_to_nid(buffer_info->page) != current_node))
5137 buffer_info->page = NULL;
5138 else
5139 get_page(buffer_info->page);
5140
5141 skb->len += length;
5142 skb->data_len += length;
5143 skb->truesize += length;
5144 }
5145
5146 if (!(staterr & E1000_RXD_STAT_EOP)) {
5147 buffer_info->skb = next_buffer->skb;
5148 buffer_info->dma = next_buffer->dma;
5149 next_buffer->skb = skb;
5150 next_buffer->dma = 0;
5151 goto next_desc;
5152 }
5153 send_up:
5154 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
5155 dev_kfree_skb_irq(skb);
5156 goto next_desc;
5157 }
5158
5159 igb_rx_hwtstamp(q_vector, staterr, skb);
5160 total_bytes += skb->len;
5161 total_packets++;
5162
5163 igb_rx_checksum_adv(rx_ring, staterr, skb);
5164
5165 skb->protocol = eth_type_trans(skb, netdev);
5166 skb_record_rx_queue(skb, rx_ring->queue_index);
5167
5168 vlan_tag = ((staterr & E1000_RXD_STAT_VP) ?
5169 le16_to_cpu(rx_desc->wb.upper.vlan) : 0);
5170
5171 igb_receive_skb(q_vector, skb, vlan_tag);
5172
5173 next_desc:
5174 rx_desc->wb.upper.status_error = 0;
5175
5176 /* return some buffers to hardware, one at a time is too slow */
5177 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
5178 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
5179 cleaned_count = 0;
5180 }
5181
5182 /* use prefetched values */
5183 rx_desc = next_rxd;
5184 buffer_info = next_buffer;
5185 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
5186 }
5187
5188 rx_ring->next_to_clean = i;
5189 cleaned_count = igb_desc_unused(rx_ring);
5190
5191 if (cleaned_count)
5192 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
5193
5194 rx_ring->total_packets += total_packets;
5195 rx_ring->total_bytes += total_bytes;
5196 rx_ring->rx_stats.packets += total_packets;
5197 rx_ring->rx_stats.bytes += total_bytes;
5198 return cleaned;
5199 }
5200
5201 /**
5202 * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
5203 * @adapter: address of board private structure
5204 **/
5205 void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring, int cleaned_count)
5206 {
5207 struct net_device *netdev = rx_ring->netdev;
5208 union e1000_adv_rx_desc *rx_desc;
5209 struct igb_buffer *buffer_info;
5210 struct sk_buff *skb;
5211 unsigned int i;
5212 int bufsz;
5213
5214 i = rx_ring->next_to_use;
5215 buffer_info = &rx_ring->buffer_info[i];
5216
5217 bufsz = rx_ring->rx_buffer_len;
5218
5219 while (cleaned_count--) {
5220 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
5221
5222 if ((bufsz < IGB_RXBUFFER_1024) && !buffer_info->page_dma) {
5223 if (!buffer_info->page) {
5224 buffer_info->page = netdev_alloc_page(netdev);
5225 if (!buffer_info->page) {
5226 rx_ring->rx_stats.alloc_failed++;
5227 goto no_buffers;
5228 }
5229 buffer_info->page_offset = 0;
5230 } else {
5231 buffer_info->page_offset ^= PAGE_SIZE / 2;
5232 }
5233 buffer_info->page_dma =
5234 pci_map_page(rx_ring->pdev, buffer_info->page,
5235 buffer_info->page_offset,
5236 PAGE_SIZE / 2,
5237 PCI_DMA_FROMDEVICE);
5238 if (pci_dma_mapping_error(rx_ring->pdev,
5239 buffer_info->page_dma)) {
5240 buffer_info->page_dma = 0;
5241 rx_ring->rx_stats.alloc_failed++;
5242 goto no_buffers;
5243 }
5244 }
5245
5246 skb = buffer_info->skb;
5247 if (!skb) {
5248 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
5249 if (!skb) {
5250 rx_ring->rx_stats.alloc_failed++;
5251 goto no_buffers;
5252 }
5253
5254 buffer_info->skb = skb;
5255 }
5256 if (!buffer_info->dma) {
5257 buffer_info->dma = pci_map_single(rx_ring->pdev,
5258 skb->data,
5259 bufsz,
5260 PCI_DMA_FROMDEVICE);
5261 if (pci_dma_mapping_error(rx_ring->pdev,
5262 buffer_info->dma)) {
5263 buffer_info->dma = 0;
5264 rx_ring->rx_stats.alloc_failed++;
5265 goto no_buffers;
5266 }
5267 }
5268 /* Refresh the desc even if buffer_addrs didn't change because
5269 * each write-back erases this info. */
5270 if (bufsz < IGB_RXBUFFER_1024) {
5271 rx_desc->read.pkt_addr =
5272 cpu_to_le64(buffer_info->page_dma);
5273 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
5274 } else {
5275 rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma);
5276 rx_desc->read.hdr_addr = 0;
5277 }
5278
5279 i++;
5280 if (i == rx_ring->count)
5281 i = 0;
5282 buffer_info = &rx_ring->buffer_info[i];
5283 }
5284
5285 no_buffers:
5286 if (rx_ring->next_to_use != i) {
5287 rx_ring->next_to_use = i;
5288 if (i == 0)
5289 i = (rx_ring->count - 1);
5290 else
5291 i--;
5292
5293 /* Force memory writes to complete before letting h/w
5294 * know there are new descriptors to fetch. (Only
5295 * applicable for weak-ordered memory model archs,
5296 * such as IA-64). */
5297 wmb();
5298 writel(i, rx_ring->tail);
5299 }
5300 }
5301
5302 /**
5303 * igb_mii_ioctl -
5304 * @netdev:
5305 * @ifreq:
5306 * @cmd:
5307 **/
5308 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5309 {
5310 struct igb_adapter *adapter = netdev_priv(netdev);
5311 struct mii_ioctl_data *data = if_mii(ifr);
5312
5313 if (adapter->hw.phy.media_type != e1000_media_type_copper)
5314 return -EOPNOTSUPP;
5315
5316 switch (cmd) {
5317 case SIOCGMIIPHY:
5318 data->phy_id = adapter->hw.phy.addr;
5319 break;
5320 case SIOCGMIIREG:
5321 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
5322 &data->val_out))
5323 return -EIO;
5324 break;
5325 case SIOCSMIIREG:
5326 default:
5327 return -EOPNOTSUPP;
5328 }
5329 return 0;
5330 }
5331
5332 /**
5333 * igb_hwtstamp_ioctl - control hardware time stamping
5334 * @netdev:
5335 * @ifreq:
5336 * @cmd:
5337 *
5338 * Outgoing time stamping can be enabled and disabled. Play nice and
5339 * disable it when requested, although it shouldn't case any overhead
5340 * when no packet needs it. At most one packet in the queue may be
5341 * marked for time stamping, otherwise it would be impossible to tell
5342 * for sure to which packet the hardware time stamp belongs.
5343 *
5344 * Incoming time stamping has to be configured via the hardware
5345 * filters. Not all combinations are supported, in particular event
5346 * type has to be specified. Matching the kind of event packet is
5347 * not supported, with the exception of "all V2 events regardless of
5348 * level 2 or 4".
5349 *
5350 **/
5351 static int igb_hwtstamp_ioctl(struct net_device *netdev,
5352 struct ifreq *ifr, int cmd)
5353 {
5354 struct igb_adapter *adapter = netdev_priv(netdev);
5355 struct e1000_hw *hw = &adapter->hw;
5356 struct hwtstamp_config config;
5357 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
5358 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
5359 u32 tsync_rx_cfg = 0;
5360 bool is_l4 = false;
5361 bool is_l2 = false;
5362 u32 regval;
5363
5364 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
5365 return -EFAULT;
5366
5367 /* reserved for future extensions */
5368 if (config.flags)
5369 return -EINVAL;
5370
5371 switch (config.tx_type) {
5372 case HWTSTAMP_TX_OFF:
5373 tsync_tx_ctl = 0;
5374 case HWTSTAMP_TX_ON:
5375 break;
5376 default:
5377 return -ERANGE;
5378 }
5379
5380 switch (config.rx_filter) {
5381 case HWTSTAMP_FILTER_NONE:
5382 tsync_rx_ctl = 0;
5383 break;
5384 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
5385 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
5386 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
5387 case HWTSTAMP_FILTER_ALL:
5388 /*
5389 * register TSYNCRXCFG must be set, therefore it is not
5390 * possible to time stamp both Sync and Delay_Req messages
5391 * => fall back to time stamping all packets
5392 */
5393 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
5394 config.rx_filter = HWTSTAMP_FILTER_ALL;
5395 break;
5396 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
5397 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
5398 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
5399 is_l4 = true;
5400 break;
5401 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
5402 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
5403 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
5404 is_l4 = true;
5405 break;
5406 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
5407 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
5408 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
5409 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
5410 is_l2 = true;
5411 is_l4 = true;
5412 config.rx_filter = HWTSTAMP_FILTER_SOME;
5413 break;
5414 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
5415 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
5416 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
5417 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
5418 is_l2 = true;
5419 is_l4 = true;
5420 config.rx_filter = HWTSTAMP_FILTER_SOME;
5421 break;
5422 case HWTSTAMP_FILTER_PTP_V2_EVENT:
5423 case HWTSTAMP_FILTER_PTP_V2_SYNC:
5424 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
5425 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
5426 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
5427 is_l2 = true;
5428 break;
5429 default:
5430 return -ERANGE;
5431 }
5432
5433 if (hw->mac.type == e1000_82575) {
5434 if (tsync_rx_ctl | tsync_tx_ctl)
5435 return -EINVAL;
5436 return 0;
5437 }
5438
5439 /* enable/disable TX */
5440 regval = rd32(E1000_TSYNCTXCTL);
5441 regval &= ~E1000_TSYNCTXCTL_ENABLED;
5442 regval |= tsync_tx_ctl;
5443 wr32(E1000_TSYNCTXCTL, regval);
5444
5445 /* enable/disable RX */
5446 regval = rd32(E1000_TSYNCRXCTL);
5447 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
5448 regval |= tsync_rx_ctl;
5449 wr32(E1000_TSYNCRXCTL, regval);
5450
5451 /* define which PTP packets are time stamped */
5452 wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
5453
5454 /* define ethertype filter for timestamped packets */
5455 if (is_l2)
5456 wr32(E1000_ETQF(3),
5457 (E1000_ETQF_FILTER_ENABLE | /* enable filter */
5458 E1000_ETQF_1588 | /* enable timestamping */
5459 ETH_P_1588)); /* 1588 eth protocol type */
5460 else
5461 wr32(E1000_ETQF(3), 0);
5462
5463 #define PTP_PORT 319
5464 /* L4 Queue Filter[3]: filter by destination port and protocol */
5465 if (is_l4) {
5466 u32 ftqf = (IPPROTO_UDP /* UDP */
5467 | E1000_FTQF_VF_BP /* VF not compared */
5468 | E1000_FTQF_1588_TIME_STAMP /* Enable Timestamping */
5469 | E1000_FTQF_MASK); /* mask all inputs */
5470 ftqf &= ~E1000_FTQF_MASK_PROTO_BP; /* enable protocol check */
5471
5472 wr32(E1000_IMIR(3), htons(PTP_PORT));
5473 wr32(E1000_IMIREXT(3),
5474 (E1000_IMIREXT_SIZE_BP | E1000_IMIREXT_CTRL_BP));
5475 if (hw->mac.type == e1000_82576) {
5476 /* enable source port check */
5477 wr32(E1000_SPQF(3), htons(PTP_PORT));
5478 ftqf &= ~E1000_FTQF_MASK_SOURCE_PORT_BP;
5479 }
5480 wr32(E1000_FTQF(3), ftqf);
5481 } else {
5482 wr32(E1000_FTQF(3), E1000_FTQF_MASK);
5483 }
5484 wrfl();
5485
5486 adapter->hwtstamp_config = config;
5487
5488 /* clear TX/RX time stamp registers, just to be sure */
5489 regval = rd32(E1000_TXSTMPH);
5490 regval = rd32(E1000_RXSTMPH);
5491
5492 return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
5493 -EFAULT : 0;
5494 }
5495
5496 /**
5497 * igb_ioctl -
5498 * @netdev:
5499 * @ifreq:
5500 * @cmd:
5501 **/
5502 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5503 {
5504 switch (cmd) {
5505 case SIOCGMIIPHY:
5506 case SIOCGMIIREG:
5507 case SIOCSMIIREG:
5508 return igb_mii_ioctl(netdev, ifr, cmd);
5509 case SIOCSHWTSTAMP:
5510 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
5511 default:
5512 return -EOPNOTSUPP;
5513 }
5514 }
5515
5516 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
5517 {
5518 struct igb_adapter *adapter = hw->back;
5519 u16 cap_offset;
5520
5521 cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
5522 if (!cap_offset)
5523 return -E1000_ERR_CONFIG;
5524
5525 pci_read_config_word(adapter->pdev, cap_offset + reg, value);
5526
5527 return 0;
5528 }
5529
5530 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
5531 {
5532 struct igb_adapter *adapter = hw->back;
5533 u16 cap_offset;
5534
5535 cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
5536 if (!cap_offset)
5537 return -E1000_ERR_CONFIG;
5538
5539 pci_write_config_word(adapter->pdev, cap_offset + reg, *value);
5540
5541 return 0;
5542 }
5543
5544 static void igb_vlan_rx_register(struct net_device *netdev,
5545 struct vlan_group *grp)
5546 {
5547 struct igb_adapter *adapter = netdev_priv(netdev);
5548 struct e1000_hw *hw = &adapter->hw;
5549 u32 ctrl, rctl;
5550
5551 igb_irq_disable(adapter);
5552 adapter->vlgrp = grp;
5553
5554 if (grp) {
5555 /* enable VLAN tag insert/strip */
5556 ctrl = rd32(E1000_CTRL);
5557 ctrl |= E1000_CTRL_VME;
5558 wr32(E1000_CTRL, ctrl);
5559
5560 /* Disable CFI check */
5561 rctl = rd32(E1000_RCTL);
5562 rctl &= ~E1000_RCTL_CFIEN;
5563 wr32(E1000_RCTL, rctl);
5564 } else {
5565 /* disable VLAN tag insert/strip */
5566 ctrl = rd32(E1000_CTRL);
5567 ctrl &= ~E1000_CTRL_VME;
5568 wr32(E1000_CTRL, ctrl);
5569 }
5570
5571 igb_rlpml_set(adapter);
5572
5573 if (!test_bit(__IGB_DOWN, &adapter->state))
5574 igb_irq_enable(adapter);
5575 }
5576
5577 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
5578 {
5579 struct igb_adapter *adapter = netdev_priv(netdev);
5580 struct e1000_hw *hw = &adapter->hw;
5581 int pf_id = adapter->vfs_allocated_count;
5582
5583 /* attempt to add filter to vlvf array */
5584 igb_vlvf_set(adapter, vid, true, pf_id);
5585
5586 /* add the filter since PF can receive vlans w/o entry in vlvf */
5587 igb_vfta_set(hw, vid, true);
5588 }
5589
5590 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
5591 {
5592 struct igb_adapter *adapter = netdev_priv(netdev);
5593 struct e1000_hw *hw = &adapter->hw;
5594 int pf_id = adapter->vfs_allocated_count;
5595 s32 err;
5596
5597 igb_irq_disable(adapter);
5598 vlan_group_set_device(adapter->vlgrp, vid, NULL);
5599
5600 if (!test_bit(__IGB_DOWN, &adapter->state))
5601 igb_irq_enable(adapter);
5602
5603 /* remove vlan from VLVF table array */
5604 err = igb_vlvf_set(adapter, vid, false, pf_id);
5605
5606 /* if vid was not present in VLVF just remove it from table */
5607 if (err)
5608 igb_vfta_set(hw, vid, false);
5609 }
5610
5611 static void igb_restore_vlan(struct igb_adapter *adapter)
5612 {
5613 igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5614
5615 if (adapter->vlgrp) {
5616 u16 vid;
5617 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5618 if (!vlan_group_get_device(adapter->vlgrp, vid))
5619 continue;
5620 igb_vlan_rx_add_vid(adapter->netdev, vid);
5621 }
5622 }
5623 }
5624
5625 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
5626 {
5627 struct pci_dev *pdev = adapter->pdev;
5628 struct e1000_mac_info *mac = &adapter->hw.mac;
5629
5630 mac->autoneg = 0;
5631
5632 switch (spddplx) {
5633 case SPEED_10 + DUPLEX_HALF:
5634 mac->forced_speed_duplex = ADVERTISE_10_HALF;
5635 break;
5636 case SPEED_10 + DUPLEX_FULL:
5637 mac->forced_speed_duplex = ADVERTISE_10_FULL;
5638 break;
5639 case SPEED_100 + DUPLEX_HALF:
5640 mac->forced_speed_duplex = ADVERTISE_100_HALF;
5641 break;
5642 case SPEED_100 + DUPLEX_FULL:
5643 mac->forced_speed_duplex = ADVERTISE_100_FULL;
5644 break;
5645 case SPEED_1000 + DUPLEX_FULL:
5646 mac->autoneg = 1;
5647 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
5648 break;
5649 case SPEED_1000 + DUPLEX_HALF: /* not supported */
5650 default:
5651 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
5652 return -EINVAL;
5653 }
5654 return 0;
5655 }
5656
5657 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
5658 {
5659 struct net_device *netdev = pci_get_drvdata(pdev);
5660 struct igb_adapter *adapter = netdev_priv(netdev);
5661 struct e1000_hw *hw = &adapter->hw;
5662 u32 ctrl, rctl, status;
5663 u32 wufc = adapter->wol;
5664 #ifdef CONFIG_PM
5665 int retval = 0;
5666 #endif
5667
5668 netif_device_detach(netdev);
5669
5670 if (netif_running(netdev))
5671 igb_close(netdev);
5672
5673 igb_clear_interrupt_scheme(adapter);
5674
5675 #ifdef CONFIG_PM
5676 retval = pci_save_state(pdev);
5677 if (retval)
5678 return retval;
5679 #endif
5680
5681 status = rd32(E1000_STATUS);
5682 if (status & E1000_STATUS_LU)
5683 wufc &= ~E1000_WUFC_LNKC;
5684
5685 if (wufc) {
5686 igb_setup_rctl(adapter);
5687 igb_set_rx_mode(netdev);
5688
5689 /* turn on all-multi mode if wake on multicast is enabled */
5690 if (wufc & E1000_WUFC_MC) {
5691 rctl = rd32(E1000_RCTL);
5692 rctl |= E1000_RCTL_MPE;
5693 wr32(E1000_RCTL, rctl);
5694 }
5695
5696 ctrl = rd32(E1000_CTRL);
5697 /* advertise wake from D3Cold */
5698 #define E1000_CTRL_ADVD3WUC 0x00100000
5699 /* phy power management enable */
5700 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5701 ctrl |= E1000_CTRL_ADVD3WUC;
5702 wr32(E1000_CTRL, ctrl);
5703
5704 /* Allow time for pending master requests to run */
5705 igb_disable_pcie_master(hw);
5706
5707 wr32(E1000_WUC, E1000_WUC_PME_EN);
5708 wr32(E1000_WUFC, wufc);
5709 } else {
5710 wr32(E1000_WUC, 0);
5711 wr32(E1000_WUFC, 0);
5712 }
5713
5714 *enable_wake = wufc || adapter->en_mng_pt;
5715 if (!*enable_wake)
5716 igb_shutdown_serdes_link_82575(hw);
5717
5718 /* Release control of h/w to f/w. If f/w is AMT enabled, this
5719 * would have already happened in close and is redundant. */
5720 igb_release_hw_control(adapter);
5721
5722 pci_disable_device(pdev);
5723
5724 return 0;
5725 }
5726
5727 #ifdef CONFIG_PM
5728 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
5729 {
5730 int retval;
5731 bool wake;
5732
5733 retval = __igb_shutdown(pdev, &wake);
5734 if (retval)
5735 return retval;
5736
5737 if (wake) {
5738 pci_prepare_to_sleep(pdev);
5739 } else {
5740 pci_wake_from_d3(pdev, false);
5741 pci_set_power_state(pdev, PCI_D3hot);
5742 }
5743
5744 return 0;
5745 }
5746
5747 static int igb_resume(struct pci_dev *pdev)
5748 {
5749 struct net_device *netdev = pci_get_drvdata(pdev);
5750 struct igb_adapter *adapter = netdev_priv(netdev);
5751 struct e1000_hw *hw = &adapter->hw;
5752 u32 err;
5753
5754 pci_set_power_state(pdev, PCI_D0);
5755 pci_restore_state(pdev);
5756
5757 err = pci_enable_device_mem(pdev);
5758 if (err) {
5759 dev_err(&pdev->dev,
5760 "igb: Cannot enable PCI device from suspend\n");
5761 return err;
5762 }
5763 pci_set_master(pdev);
5764
5765 pci_enable_wake(pdev, PCI_D3hot, 0);
5766 pci_enable_wake(pdev, PCI_D3cold, 0);
5767
5768 if (igb_init_interrupt_scheme(adapter)) {
5769 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5770 return -ENOMEM;
5771 }
5772
5773 /* e1000_power_up_phy(adapter); */
5774
5775 igb_reset(adapter);
5776
5777 /* let the f/w know that the h/w is now under the control of the
5778 * driver. */
5779 igb_get_hw_control(adapter);
5780
5781 wr32(E1000_WUS, ~0);
5782
5783 if (netif_running(netdev)) {
5784 err = igb_open(netdev);
5785 if (err)
5786 return err;
5787 }
5788
5789 netif_device_attach(netdev);
5790
5791 return 0;
5792 }
5793 #endif
5794
5795 static void igb_shutdown(struct pci_dev *pdev)
5796 {
5797 bool wake;
5798
5799 __igb_shutdown(pdev, &wake);
5800
5801 if (system_state == SYSTEM_POWER_OFF) {
5802 pci_wake_from_d3(pdev, wake);
5803 pci_set_power_state(pdev, PCI_D3hot);
5804 }
5805 }
5806
5807 #ifdef CONFIG_NET_POLL_CONTROLLER
5808 /*
5809 * Polling 'interrupt' - used by things like netconsole to send skbs
5810 * without having to re-enable interrupts. It's not called while
5811 * the interrupt routine is executing.
5812 */
5813 static void igb_netpoll(struct net_device *netdev)
5814 {
5815 struct igb_adapter *adapter = netdev_priv(netdev);
5816 struct e1000_hw *hw = &adapter->hw;
5817 int i;
5818
5819 if (!adapter->msix_entries) {
5820 struct igb_q_vector *q_vector = adapter->q_vector[0];
5821 igb_irq_disable(adapter);
5822 napi_schedule(&q_vector->napi);
5823 return;
5824 }
5825
5826 for (i = 0; i < adapter->num_q_vectors; i++) {
5827 struct igb_q_vector *q_vector = adapter->q_vector[i];
5828 wr32(E1000_EIMC, q_vector->eims_value);
5829 napi_schedule(&q_vector->napi);
5830 }
5831 }
5832 #endif /* CONFIG_NET_POLL_CONTROLLER */
5833
5834 /**
5835 * igb_io_error_detected - called when PCI error is detected
5836 * @pdev: Pointer to PCI device
5837 * @state: The current pci connection state
5838 *
5839 * This function is called after a PCI bus error affecting
5840 * this device has been detected.
5841 */
5842 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
5843 pci_channel_state_t state)
5844 {
5845 struct net_device *netdev = pci_get_drvdata(pdev);
5846 struct igb_adapter *adapter = netdev_priv(netdev);
5847
5848 netif_device_detach(netdev);
5849
5850 if (state == pci_channel_io_perm_failure)
5851 return PCI_ERS_RESULT_DISCONNECT;
5852
5853 if (netif_running(netdev))
5854 igb_down(adapter);
5855 pci_disable_device(pdev);
5856
5857 /* Request a slot slot reset. */
5858 return PCI_ERS_RESULT_NEED_RESET;
5859 }
5860
5861 /**
5862 * igb_io_slot_reset - called after the pci bus has been reset.
5863 * @pdev: Pointer to PCI device
5864 *
5865 * Restart the card from scratch, as if from a cold-boot. Implementation
5866 * resembles the first-half of the igb_resume routine.
5867 */
5868 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
5869 {
5870 struct net_device *netdev = pci_get_drvdata(pdev);
5871 struct igb_adapter *adapter = netdev_priv(netdev);
5872 struct e1000_hw *hw = &adapter->hw;
5873 pci_ers_result_t result;
5874 int err;
5875
5876 if (pci_enable_device_mem(pdev)) {
5877 dev_err(&pdev->dev,
5878 "Cannot re-enable PCI device after reset.\n");
5879 result = PCI_ERS_RESULT_DISCONNECT;
5880 } else {
5881 pci_set_master(pdev);
5882 pci_restore_state(pdev);
5883
5884 pci_enable_wake(pdev, PCI_D3hot, 0);
5885 pci_enable_wake(pdev, PCI_D3cold, 0);
5886
5887 igb_reset(adapter);
5888 wr32(E1000_WUS, ~0);
5889 result = PCI_ERS_RESULT_RECOVERED;
5890 }
5891
5892 err = pci_cleanup_aer_uncorrect_error_status(pdev);
5893 if (err) {
5894 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
5895 "failed 0x%0x\n", err);
5896 /* non-fatal, continue */
5897 }
5898
5899 return result;
5900 }
5901
5902 /**
5903 * igb_io_resume - called when traffic can start flowing again.
5904 * @pdev: Pointer to PCI device
5905 *
5906 * This callback is called when the error recovery driver tells us that
5907 * its OK to resume normal operation. Implementation resembles the
5908 * second-half of the igb_resume routine.
5909 */
5910 static void igb_io_resume(struct pci_dev *pdev)
5911 {
5912 struct net_device *netdev = pci_get_drvdata(pdev);
5913 struct igb_adapter *adapter = netdev_priv(netdev);
5914
5915 if (netif_running(netdev)) {
5916 if (igb_up(adapter)) {
5917 dev_err(&pdev->dev, "igb_up failed after reset\n");
5918 return;
5919 }
5920 }
5921
5922 netif_device_attach(netdev);
5923
5924 /* let the f/w know that the h/w is now under the control of the
5925 * driver. */
5926 igb_get_hw_control(adapter);
5927 }
5928
5929 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
5930 u8 qsel)
5931 {
5932 u32 rar_low, rar_high;
5933 struct e1000_hw *hw = &adapter->hw;
5934
5935 /* HW expects these in little endian so we reverse the byte order
5936 * from network order (big endian) to little endian
5937 */
5938 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
5939 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
5940 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
5941
5942 /* Indicate to hardware the Address is Valid. */
5943 rar_high |= E1000_RAH_AV;
5944
5945 if (hw->mac.type == e1000_82575)
5946 rar_high |= E1000_RAH_POOL_1 * qsel;
5947 else
5948 rar_high |= E1000_RAH_POOL_1 << qsel;
5949
5950 wr32(E1000_RAL(index), rar_low);
5951 wrfl();
5952 wr32(E1000_RAH(index), rar_high);
5953 wrfl();
5954 }
5955
5956 static int igb_set_vf_mac(struct igb_adapter *adapter,
5957 int vf, unsigned char *mac_addr)
5958 {
5959 struct e1000_hw *hw = &adapter->hw;
5960 /* VF MAC addresses start at end of receive addresses and moves
5961 * torwards the first, as a result a collision should not be possible */
5962 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5963
5964 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
5965
5966 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
5967
5968 return 0;
5969 }
5970
5971 static void igb_vmm_control(struct igb_adapter *adapter)
5972 {
5973 struct e1000_hw *hw = &adapter->hw;
5974 u32 reg;
5975
5976 /* replication is not supported for 82575 */
5977 if (hw->mac.type == e1000_82575)
5978 return;
5979
5980 /* enable replication vlan tag stripping */
5981 reg = rd32(E1000_RPLOLR);
5982 reg |= E1000_RPLOLR_STRVLAN;
5983 wr32(E1000_RPLOLR, reg);
5984
5985 /* notify HW that the MAC is adding vlan tags */
5986 reg = rd32(E1000_DTXCTL);
5987 reg |= E1000_DTXCTL_VLAN_ADDED;
5988 wr32(E1000_DTXCTL, reg);
5989
5990 if (adapter->vfs_allocated_count) {
5991 igb_vmdq_set_loopback_pf(hw, true);
5992 igb_vmdq_set_replication_pf(hw, true);
5993 } else {
5994 igb_vmdq_set_loopback_pf(hw, false);
5995 igb_vmdq_set_replication_pf(hw, false);
5996 }
5997 }
5998
5999 /* igb_main.c */
Linus' kernel tree