igb: do not check for vf_data if we didn't enable vfs
[linux-2.6.git] / drivers / net / igb / igb_main.c
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 "1.3.16-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_82576), board_82575 },
65         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
66         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
67         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
68         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
69         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
70         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
71         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
72         /* required last entry */
73         {0, }
74 };
75
76 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
77
78 void igb_reset(struct igb_adapter *);
79 static int igb_setup_all_tx_resources(struct igb_adapter *);
80 static int igb_setup_all_rx_resources(struct igb_adapter *);
81 static void igb_free_all_tx_resources(struct igb_adapter *);
82 static void igb_free_all_rx_resources(struct igb_adapter *);
83 void igb_update_stats(struct igb_adapter *);
84 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
85 static void __devexit igb_remove(struct pci_dev *pdev);
86 static int igb_sw_init(struct igb_adapter *);
87 static int igb_open(struct net_device *);
88 static int igb_close(struct net_device *);
89 static void igb_configure_tx(struct igb_adapter *);
90 static void igb_configure_rx(struct igb_adapter *);
91 static void igb_setup_rctl(struct igb_adapter *);
92 static void igb_clean_all_tx_rings(struct igb_adapter *);
93 static void igb_clean_all_rx_rings(struct igb_adapter *);
94 static void igb_clean_tx_ring(struct igb_ring *);
95 static void igb_clean_rx_ring(struct igb_ring *);
96 static void igb_set_multi(struct net_device *);
97 static void igb_update_phy_info(unsigned long);
98 static void igb_watchdog(unsigned long);
99 static void igb_watchdog_task(struct work_struct *);
100 static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
101                                   struct igb_ring *);
102 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
103 static struct net_device_stats *igb_get_stats(struct net_device *);
104 static int igb_change_mtu(struct net_device *, int);
105 static int igb_set_mac(struct net_device *, void *);
106 static irqreturn_t igb_intr(int irq, void *);
107 static irqreturn_t igb_intr_msi(int irq, void *);
108 static irqreturn_t igb_msix_other(int irq, void *);
109 static irqreturn_t igb_msix_rx(int irq, void *);
110 static irqreturn_t igb_msix_tx(int irq, void *);
111 #ifdef CONFIG_IGB_DCA
112 static void igb_update_rx_dca(struct igb_ring *);
113 static void igb_update_tx_dca(struct igb_ring *);
114 static void igb_setup_dca(struct igb_adapter *);
115 #endif /* CONFIG_IGB_DCA */
116 static bool igb_clean_tx_irq(struct igb_ring *);
117 static int igb_poll(struct napi_struct *, int);
118 static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
119 static void igb_alloc_rx_buffers_adv(struct igb_ring *, int);
120 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
121 static void igb_tx_timeout(struct net_device *);
122 static void igb_reset_task(struct work_struct *);
123 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
124 static void igb_vlan_rx_add_vid(struct net_device *, u16);
125 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
126 static void igb_restore_vlan(struct igb_adapter *);
127 static void igb_ping_all_vfs(struct igb_adapter *);
128 static void igb_msg_task(struct igb_adapter *);
129 static int igb_rcv_msg_from_vf(struct igb_adapter *, u32);
130 static inline void igb_set_rah_pool(struct e1000_hw *, int , int);
131 static void igb_set_mc_list_pools(struct igb_adapter *, int, u16);
132 static void igb_vmm_control(struct igb_adapter *);
133 static inline void igb_set_vmolr(struct e1000_hw *, int);
134 static inline int igb_set_vf_rlpml(struct igb_adapter *, int, int);
135 static int igb_set_vf_mac(struct igb_adapter *adapter, int, unsigned char *);
136 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
137
138 #ifdef CONFIG_PM
139 static int igb_suspend(struct pci_dev *, pm_message_t);
140 static int igb_resume(struct pci_dev *);
141 #endif
142 static void igb_shutdown(struct pci_dev *);
143 #ifdef CONFIG_IGB_DCA
144 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
145 static struct notifier_block dca_notifier = {
146         .notifier_call  = igb_notify_dca,
147         .next           = NULL,
148         .priority       = 0
149 };
150 #endif
151 #ifdef CONFIG_NET_POLL_CONTROLLER
152 /* for netdump / net console */
153 static void igb_netpoll(struct net_device *);
154 #endif
155 #ifdef CONFIG_PCI_IOV
156 static unsigned int max_vfs = 0;
157 module_param(max_vfs, uint, 0);
158 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
159                  "per physical function");
160 #endif /* CONFIG_PCI_IOV */
161
162 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
163                      pci_channel_state_t);
164 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
165 static void igb_io_resume(struct pci_dev *);
166
167 static struct pci_error_handlers igb_err_handler = {
168         .error_detected = igb_io_error_detected,
169         .slot_reset = igb_io_slot_reset,
170         .resume = igb_io_resume,
171 };
172
173
174 static struct pci_driver igb_driver = {
175         .name     = igb_driver_name,
176         .id_table = igb_pci_tbl,
177         .probe    = igb_probe,
178         .remove   = __devexit_p(igb_remove),
179 #ifdef CONFIG_PM
180         /* Power Managment Hooks */
181         .suspend  = igb_suspend,
182         .resume   = igb_resume,
183 #endif
184         .shutdown = igb_shutdown,
185         .err_handler = &igb_err_handler
186 };
187
188 static int global_quad_port_a; /* global quad port a indication */
189
190 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
191 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
192 MODULE_LICENSE("GPL");
193 MODULE_VERSION(DRV_VERSION);
194
195 /**
196  * Scale the NIC clock cycle by a large factor so that
197  * relatively small clock corrections can be added or
198  * substracted at each clock tick. The drawbacks of a
199  * large factor are a) that the clock register overflows
200  * more quickly (not such a big deal) and b) that the
201  * increment per tick has to fit into 24 bits.
202  *
203  * Note that
204  *   TIMINCA = IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS *
205  *             IGB_TSYNC_SCALE
206  *   TIMINCA += TIMINCA * adjustment [ppm] / 1e9
207  *
208  * The base scale factor is intentionally a power of two
209  * so that the division in %struct timecounter can be done with
210  * a shift.
211  */
212 #define IGB_TSYNC_SHIFT (19)
213 #define IGB_TSYNC_SCALE (1<<IGB_TSYNC_SHIFT)
214
215 /**
216  * The duration of one clock cycle of the NIC.
217  *
218  * @todo This hard-coded value is part of the specification and might change
219  * in future hardware revisions. Add revision check.
220  */
221 #define IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS 16
222
223 #if (IGB_TSYNC_SCALE * IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS) >= (1<<24)
224 # error IGB_TSYNC_SCALE and/or IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS are too large to fit into TIMINCA
225 #endif
226
227 /**
228  * igb_read_clock - read raw cycle counter (to be used by time counter)
229  */
230 static cycle_t igb_read_clock(const struct cyclecounter *tc)
231 {
232         struct igb_adapter *adapter =
233                 container_of(tc, struct igb_adapter, cycles);
234         struct e1000_hw *hw = &adapter->hw;
235         u64 stamp;
236
237         stamp =  rd32(E1000_SYSTIML);
238         stamp |= (u64)rd32(E1000_SYSTIMH) << 32ULL;
239
240         return stamp;
241 }
242
243 #ifdef DEBUG
244 /**
245  * igb_get_hw_dev_name - return device name string
246  * used by hardware layer to print debugging information
247  **/
248 char *igb_get_hw_dev_name(struct e1000_hw *hw)
249 {
250         struct igb_adapter *adapter = hw->back;
251         return adapter->netdev->name;
252 }
253
254 /**
255  * igb_get_time_str - format current NIC and system time as string
256  */
257 static char *igb_get_time_str(struct igb_adapter *adapter,
258                               char buffer[160])
259 {
260         cycle_t hw = adapter->cycles.read(&adapter->cycles);
261         struct timespec nic = ns_to_timespec(timecounter_read(&adapter->clock));
262         struct timespec sys;
263         struct timespec delta;
264         getnstimeofday(&sys);
265
266         delta = timespec_sub(nic, sys);
267
268         sprintf(buffer,
269                 "HW %llu, NIC %ld.%09lus, SYS %ld.%09lus, NIC-SYS %lds + %09luns",
270                 hw,
271                 (long)nic.tv_sec, nic.tv_nsec,
272                 (long)sys.tv_sec, sys.tv_nsec,
273                 (long)delta.tv_sec, delta.tv_nsec);
274
275         return buffer;
276 }
277 #endif
278
279 /**
280  * igb_desc_unused - calculate if we have unused descriptors
281  **/
282 static int igb_desc_unused(struct igb_ring *ring)
283 {
284         if (ring->next_to_clean > ring->next_to_use)
285                 return ring->next_to_clean - ring->next_to_use - 1;
286
287         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
288 }
289
290 /**
291  * igb_init_module - Driver Registration Routine
292  *
293  * igb_init_module is the first routine called when the driver is
294  * loaded. All it does is register with the PCI subsystem.
295  **/
296 static int __init igb_init_module(void)
297 {
298         int ret;
299         printk(KERN_INFO "%s - version %s\n",
300                igb_driver_string, igb_driver_version);
301
302         printk(KERN_INFO "%s\n", igb_copyright);
303
304         global_quad_port_a = 0;
305
306 #ifdef CONFIG_IGB_DCA
307         dca_register_notify(&dca_notifier);
308 #endif
309
310         ret = pci_register_driver(&igb_driver);
311         return ret;
312 }
313
314 module_init(igb_init_module);
315
316 /**
317  * igb_exit_module - Driver Exit Cleanup Routine
318  *
319  * igb_exit_module is called just before the driver is removed
320  * from memory.
321  **/
322 static void __exit igb_exit_module(void)
323 {
324 #ifdef CONFIG_IGB_DCA
325         dca_unregister_notify(&dca_notifier);
326 #endif
327         pci_unregister_driver(&igb_driver);
328 }
329
330 module_exit(igb_exit_module);
331
332 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
333 /**
334  * igb_cache_ring_register - Descriptor ring to register mapping
335  * @adapter: board private structure to initialize
336  *
337  * Once we know the feature-set enabled for the device, we'll cache
338  * the register offset the descriptor ring is assigned to.
339  **/
340 static void igb_cache_ring_register(struct igb_adapter *adapter)
341 {
342         int i;
343         unsigned int rbase_offset = adapter->vfs_allocated_count;
344
345         switch (adapter->hw.mac.type) {
346         case e1000_82576:
347                 /* The queues are allocated for virtualization such that VF 0
348                  * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
349                  * In order to avoid collision we start at the first free queue
350                  * and continue consuming queues in the same sequence
351                  */
352                 for (i = 0; i < adapter->num_rx_queues; i++)
353                         adapter->rx_ring[i].reg_idx = rbase_offset +
354                                                       Q_IDX_82576(i);
355                 for (i = 0; i < adapter->num_tx_queues; i++)
356                         adapter->tx_ring[i].reg_idx = rbase_offset +
357                                                       Q_IDX_82576(i);
358                 break;
359         case e1000_82575:
360         default:
361                 for (i = 0; i < adapter->num_rx_queues; i++)
362                         adapter->rx_ring[i].reg_idx = i;
363                 for (i = 0; i < adapter->num_tx_queues; i++)
364                         adapter->tx_ring[i].reg_idx = i;
365                 break;
366         }
367 }
368
369 /**
370  * igb_alloc_queues - Allocate memory for all rings
371  * @adapter: board private structure to initialize
372  *
373  * We allocate one ring per queue at run-time since we don't know the
374  * number of queues at compile-time.
375  **/
376 static int igb_alloc_queues(struct igb_adapter *adapter)
377 {
378         int i;
379
380         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
381                                    sizeof(struct igb_ring), GFP_KERNEL);
382         if (!adapter->tx_ring)
383                 return -ENOMEM;
384
385         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
386                                    sizeof(struct igb_ring), GFP_KERNEL);
387         if (!adapter->rx_ring) {
388                 kfree(adapter->tx_ring);
389                 return -ENOMEM;
390         }
391
392         adapter->rx_ring->buddy = adapter->tx_ring;
393
394         for (i = 0; i < adapter->num_tx_queues; i++) {
395                 struct igb_ring *ring = &(adapter->tx_ring[i]);
396                 ring->count = adapter->tx_ring_count;
397                 ring->adapter = adapter;
398                 ring->queue_index = i;
399         }
400         for (i = 0; i < adapter->num_rx_queues; i++) {
401                 struct igb_ring *ring = &(adapter->rx_ring[i]);
402                 ring->count = adapter->rx_ring_count;
403                 ring->adapter = adapter;
404                 ring->queue_index = i;
405                 ring->itr_register = E1000_ITR;
406
407                 /* set a default napi handler for each rx_ring */
408                 netif_napi_add(adapter->netdev, &ring->napi, igb_poll, 64);
409         }
410
411         igb_cache_ring_register(adapter);
412         return 0;
413 }
414
415 static void igb_free_queues(struct igb_adapter *adapter)
416 {
417         int i;
418
419         for (i = 0; i < adapter->num_rx_queues; i++)
420                 netif_napi_del(&adapter->rx_ring[i].napi);
421
422         adapter->num_rx_queues = 0;
423         adapter->num_tx_queues = 0;
424
425         kfree(adapter->tx_ring);
426         kfree(adapter->rx_ring);
427 }
428
429 #define IGB_N0_QUEUE -1
430 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
431                               int tx_queue, int msix_vector)
432 {
433         u32 msixbm = 0;
434         struct e1000_hw *hw = &adapter->hw;
435         u32 ivar, index;
436
437         switch (hw->mac.type) {
438         case e1000_82575:
439                 /* The 82575 assigns vectors using a bitmask, which matches the
440                    bitmask for the EICR/EIMS/EIMC registers.  To assign one
441                    or more queues to a vector, we write the appropriate bits
442                    into the MSIXBM register for that vector. */
443                 if (rx_queue > IGB_N0_QUEUE) {
444                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
445                         adapter->rx_ring[rx_queue].eims_value = msixbm;
446                 }
447                 if (tx_queue > IGB_N0_QUEUE) {
448                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
449                         adapter->tx_ring[tx_queue].eims_value =
450                                   E1000_EICR_TX_QUEUE0 << tx_queue;
451                 }
452                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
453                 break;
454         case e1000_82576:
455                 /* 82576 uses a table-based method for assigning vectors.
456                    Each queue has a single entry in the table to which we write
457                    a vector number along with a "valid" bit.  Sadly, the layout
458                    of the table is somewhat counterintuitive. */
459                 if (rx_queue > IGB_N0_QUEUE) {
460                         index = (rx_queue >> 1) + adapter->vfs_allocated_count;
461                         ivar = array_rd32(E1000_IVAR0, index);
462                         if (rx_queue & 0x1) {
463                                 /* vector goes into third byte of register */
464                                 ivar = ivar & 0xFF00FFFF;
465                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
466                         } else {
467                                 /* vector goes into low byte of register */
468                                 ivar = ivar & 0xFFFFFF00;
469                                 ivar |= msix_vector | E1000_IVAR_VALID;
470                         }
471                         adapter->rx_ring[rx_queue].eims_value= 1 << msix_vector;
472                         array_wr32(E1000_IVAR0, index, ivar);
473                 }
474                 if (tx_queue > IGB_N0_QUEUE) {
475                         index = (tx_queue >> 1) + adapter->vfs_allocated_count;
476                         ivar = array_rd32(E1000_IVAR0, index);
477                         if (tx_queue & 0x1) {
478                                 /* vector goes into high byte of register */
479                                 ivar = ivar & 0x00FFFFFF;
480                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
481                         } else {
482                                 /* vector goes into second byte of register */
483                                 ivar = ivar & 0xFFFF00FF;
484                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
485                         }
486                         adapter->tx_ring[tx_queue].eims_value= 1 << msix_vector;
487                         array_wr32(E1000_IVAR0, index, ivar);
488                 }
489                 break;
490         default:
491                 BUG();
492                 break;
493         }
494 }
495
496 /**
497  * igb_configure_msix - Configure MSI-X hardware
498  *
499  * igb_configure_msix sets up the hardware to properly
500  * generate MSI-X interrupts.
501  **/
502 static void igb_configure_msix(struct igb_adapter *adapter)
503 {
504         u32 tmp;
505         int i, vector = 0;
506         struct e1000_hw *hw = &adapter->hw;
507
508         adapter->eims_enable_mask = 0;
509         if (hw->mac.type == e1000_82576)
510                 /* Turn on MSI-X capability first, or our settings
511                  * won't stick.  And it will take days to debug. */
512                 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
513                                    E1000_GPIE_PBA | E1000_GPIE_EIAME |
514                                    E1000_GPIE_NSICR);
515
516         for (i = 0; i < adapter->num_tx_queues; i++) {
517                 struct igb_ring *tx_ring = &adapter->tx_ring[i];
518                 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
519                 adapter->eims_enable_mask |= tx_ring->eims_value;
520                 if (tx_ring->itr_val)
521                         writel(tx_ring->itr_val,
522                                hw->hw_addr + tx_ring->itr_register);
523                 else
524                         writel(1, hw->hw_addr + tx_ring->itr_register);
525         }
526
527         for (i = 0; i < adapter->num_rx_queues; i++) {
528                 struct igb_ring *rx_ring = &adapter->rx_ring[i];
529                 rx_ring->buddy = NULL;
530                 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
531                 adapter->eims_enable_mask |= rx_ring->eims_value;
532                 if (rx_ring->itr_val)
533                         writel(rx_ring->itr_val,
534                                hw->hw_addr + rx_ring->itr_register);
535                 else
536                         writel(1, hw->hw_addr + rx_ring->itr_register);
537         }
538
539
540         /* set vector for other causes, i.e. link changes */
541         switch (hw->mac.type) {
542         case e1000_82575:
543                 array_wr32(E1000_MSIXBM(0), vector++,
544                                       E1000_EIMS_OTHER);
545
546                 tmp = rd32(E1000_CTRL_EXT);
547                 /* enable MSI-X PBA support*/
548                 tmp |= E1000_CTRL_EXT_PBA_CLR;
549
550                 /* Auto-Mask interrupts upon ICR read. */
551                 tmp |= E1000_CTRL_EXT_EIAME;
552                 tmp |= E1000_CTRL_EXT_IRCA;
553
554                 wr32(E1000_CTRL_EXT, tmp);
555                 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
556                 adapter->eims_other = E1000_EIMS_OTHER;
557
558                 break;
559
560         case e1000_82576:
561                 tmp = (vector++ | E1000_IVAR_VALID) << 8;
562                 wr32(E1000_IVAR_MISC, tmp);
563
564                 adapter->eims_enable_mask = (1 << (vector)) - 1;
565                 adapter->eims_other = 1 << (vector - 1);
566                 break;
567         default:
568                 /* do nothing, since nothing else supports MSI-X */
569                 break;
570         } /* switch (hw->mac.type) */
571         wrfl();
572 }
573
574 /**
575  * igb_request_msix - Initialize MSI-X interrupts
576  *
577  * igb_request_msix allocates MSI-X vectors and requests interrupts from the
578  * kernel.
579  **/
580 static int igb_request_msix(struct igb_adapter *adapter)
581 {
582         struct net_device *netdev = adapter->netdev;
583         int i, err = 0, vector = 0;
584
585         vector = 0;
586
587         for (i = 0; i < adapter->num_tx_queues; i++) {
588                 struct igb_ring *ring = &(adapter->tx_ring[i]);
589                 sprintf(ring->name, "%s-tx-%d", netdev->name, i);
590                 err = request_irq(adapter->msix_entries[vector].vector,
591                                   &igb_msix_tx, 0, ring->name,
592                                   &(adapter->tx_ring[i]));
593                 if (err)
594                         goto out;
595                 ring->itr_register = E1000_EITR(0) + (vector << 2);
596                 ring->itr_val = 976; /* ~4000 ints/sec */
597                 vector++;
598         }
599         for (i = 0; i < adapter->num_rx_queues; i++) {
600                 struct igb_ring *ring = &(adapter->rx_ring[i]);
601                 if (strlen(netdev->name) < (IFNAMSIZ - 5))
602                         sprintf(ring->name, "%s-rx-%d", netdev->name, i);
603                 else
604                         memcpy(ring->name, netdev->name, IFNAMSIZ);
605                 err = request_irq(adapter->msix_entries[vector].vector,
606                                   &igb_msix_rx, 0, ring->name,
607                                   &(adapter->rx_ring[i]));
608                 if (err)
609                         goto out;
610                 ring->itr_register = E1000_EITR(0) + (vector << 2);
611                 ring->itr_val = adapter->itr;
612                 vector++;
613         }
614
615         err = request_irq(adapter->msix_entries[vector].vector,
616                           &igb_msix_other, 0, netdev->name, netdev);
617         if (err)
618                 goto out;
619
620         igb_configure_msix(adapter);
621         return 0;
622 out:
623         return err;
624 }
625
626 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
627 {
628         if (adapter->msix_entries) {
629                 pci_disable_msix(adapter->pdev);
630                 kfree(adapter->msix_entries);
631                 adapter->msix_entries = NULL;
632         } else if (adapter->flags & IGB_FLAG_HAS_MSI)
633                 pci_disable_msi(adapter->pdev);
634         return;
635 }
636
637
638 /**
639  * igb_set_interrupt_capability - set MSI or MSI-X if supported
640  *
641  * Attempt to configure interrupts using the best available
642  * capabilities of the hardware and kernel.
643  **/
644 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
645 {
646         int err;
647         int numvecs, i;
648
649         /* Number of supported queues. */
650         /* Having more queues than CPUs doesn't make sense. */
651         adapter->num_rx_queues = min_t(u32, IGB_MAX_RX_QUEUES, num_online_cpus());
652         adapter->num_tx_queues = min_t(u32, IGB_MAX_TX_QUEUES, num_online_cpus());
653
654         numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
655         adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
656                                         GFP_KERNEL);
657         if (!adapter->msix_entries)
658                 goto msi_only;
659
660         for (i = 0; i < numvecs; i++)
661                 adapter->msix_entries[i].entry = i;
662
663         err = pci_enable_msix(adapter->pdev,
664                               adapter->msix_entries,
665                               numvecs);
666         if (err == 0)
667                 goto out;
668
669         igb_reset_interrupt_capability(adapter);
670
671         /* If we can't do MSI-X, try MSI */
672 msi_only:
673 #ifdef CONFIG_PCI_IOV
674         /* disable SR-IOV for non MSI-X configurations */
675         if (adapter->vf_data) {
676                 struct e1000_hw *hw = &adapter->hw;
677                 /* disable iov and allow time for transactions to clear */
678                 pci_disable_sriov(adapter->pdev);
679                 msleep(500);
680
681                 kfree(adapter->vf_data);
682                 adapter->vf_data = NULL;
683                 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
684                 msleep(100);
685                 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
686         }
687 #endif
688         adapter->num_rx_queues = 1;
689         adapter->num_tx_queues = 1;
690         if (!pci_enable_msi(adapter->pdev))
691                 adapter->flags |= IGB_FLAG_HAS_MSI;
692 out:
693         /* Notify the stack of the (possibly) reduced Tx Queue count. */
694         adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
695         return;
696 }
697
698 /**
699  * igb_request_irq - initialize interrupts
700  *
701  * Attempts to configure interrupts using the best available
702  * capabilities of the hardware and kernel.
703  **/
704 static int igb_request_irq(struct igb_adapter *adapter)
705 {
706         struct net_device *netdev = adapter->netdev;
707         struct e1000_hw *hw = &adapter->hw;
708         int err = 0;
709
710         if (adapter->msix_entries) {
711                 err = igb_request_msix(adapter);
712                 if (!err)
713                         goto request_done;
714                 /* fall back to MSI */
715                 igb_reset_interrupt_capability(adapter);
716                 if (!pci_enable_msi(adapter->pdev))
717                         adapter->flags |= IGB_FLAG_HAS_MSI;
718                 igb_free_all_tx_resources(adapter);
719                 igb_free_all_rx_resources(adapter);
720                 adapter->num_rx_queues = 1;
721                 igb_alloc_queues(adapter);
722         } else {
723                 switch (hw->mac.type) {
724                 case e1000_82575:
725                         wr32(E1000_MSIXBM(0),
726                              (E1000_EICR_RX_QUEUE0 | E1000_EIMS_OTHER));
727                         break;
728                 case e1000_82576:
729                         wr32(E1000_IVAR0, E1000_IVAR_VALID);
730                         break;
731                 default:
732                         break;
733                 }
734         }
735
736         if (adapter->flags & IGB_FLAG_HAS_MSI) {
737                 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
738                                   netdev->name, netdev);
739                 if (!err)
740                         goto request_done;
741                 /* fall back to legacy interrupts */
742                 igb_reset_interrupt_capability(adapter);
743                 adapter->flags &= ~IGB_FLAG_HAS_MSI;
744         }
745
746         err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
747                           netdev->name, netdev);
748
749         if (err)
750                 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
751                         err);
752
753 request_done:
754         return err;
755 }
756
757 static void igb_free_irq(struct igb_adapter *adapter)
758 {
759         struct net_device *netdev = adapter->netdev;
760
761         if (adapter->msix_entries) {
762                 int vector = 0, i;
763
764                 for (i = 0; i < adapter->num_tx_queues; i++)
765                         free_irq(adapter->msix_entries[vector++].vector,
766                                 &(adapter->tx_ring[i]));
767                 for (i = 0; i < adapter->num_rx_queues; i++)
768                         free_irq(adapter->msix_entries[vector++].vector,
769                                 &(adapter->rx_ring[i]));
770
771                 free_irq(adapter->msix_entries[vector++].vector, netdev);
772                 return;
773         }
774
775         free_irq(adapter->pdev->irq, netdev);
776 }
777
778 /**
779  * igb_irq_disable - Mask off interrupt generation on the NIC
780  * @adapter: board private structure
781  **/
782 static void igb_irq_disable(struct igb_adapter *adapter)
783 {
784         struct e1000_hw *hw = &adapter->hw;
785
786         if (adapter->msix_entries) {
787                 wr32(E1000_EIAM, 0);
788                 wr32(E1000_EIMC, ~0);
789                 wr32(E1000_EIAC, 0);
790         }
791
792         wr32(E1000_IAM, 0);
793         wr32(E1000_IMC, ~0);
794         wrfl();
795         synchronize_irq(adapter->pdev->irq);
796 }
797
798 /**
799  * igb_irq_enable - Enable default interrupt generation settings
800  * @adapter: board private structure
801  **/
802 static void igb_irq_enable(struct igb_adapter *adapter)
803 {
804         struct e1000_hw *hw = &adapter->hw;
805
806         if (adapter->msix_entries) {
807                 wr32(E1000_EIAC, adapter->eims_enable_mask);
808                 wr32(E1000_EIAM, adapter->eims_enable_mask);
809                 wr32(E1000_EIMS, adapter->eims_enable_mask);
810                 if (adapter->vfs_allocated_count)
811                         wr32(E1000_MBVFIMR, 0xFF);
812                 wr32(E1000_IMS, (E1000_IMS_LSC | E1000_IMS_VMMB |
813                                  E1000_IMS_DOUTSYNC));
814         } else {
815                 wr32(E1000_IMS, IMS_ENABLE_MASK);
816                 wr32(E1000_IAM, IMS_ENABLE_MASK);
817         }
818 }
819
820 static void igb_update_mng_vlan(struct igb_adapter *adapter)
821 {
822         struct net_device *netdev = adapter->netdev;
823         u16 vid = adapter->hw.mng_cookie.vlan_id;
824         u16 old_vid = adapter->mng_vlan_id;
825         if (adapter->vlgrp) {
826                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
827                         if (adapter->hw.mng_cookie.status &
828                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
829                                 igb_vlan_rx_add_vid(netdev, vid);
830                                 adapter->mng_vlan_id = vid;
831                         } else
832                                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
833
834                         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
835                                         (vid != old_vid) &&
836                             !vlan_group_get_device(adapter->vlgrp, old_vid))
837                                 igb_vlan_rx_kill_vid(netdev, old_vid);
838                 } else
839                         adapter->mng_vlan_id = vid;
840         }
841 }
842
843 /**
844  * igb_release_hw_control - release control of the h/w to f/w
845  * @adapter: address of board private structure
846  *
847  * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
848  * For ASF and Pass Through versions of f/w this means that the
849  * driver is no longer loaded.
850  *
851  **/
852 static void igb_release_hw_control(struct igb_adapter *adapter)
853 {
854         struct e1000_hw *hw = &adapter->hw;
855         u32 ctrl_ext;
856
857         /* Let firmware take over control of h/w */
858         ctrl_ext = rd32(E1000_CTRL_EXT);
859         wr32(E1000_CTRL_EXT,
860                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
861 }
862
863
864 /**
865  * igb_get_hw_control - get control of the h/w from f/w
866  * @adapter: address of board private structure
867  *
868  * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
869  * For ASF and Pass Through versions of f/w this means that
870  * the driver is loaded.
871  *
872  **/
873 static void igb_get_hw_control(struct igb_adapter *adapter)
874 {
875         struct e1000_hw *hw = &adapter->hw;
876         u32 ctrl_ext;
877
878         /* Let firmware know the driver has taken over */
879         ctrl_ext = rd32(E1000_CTRL_EXT);
880         wr32(E1000_CTRL_EXT,
881                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
882 }
883
884 /**
885  * igb_configure - configure the hardware for RX and TX
886  * @adapter: private board structure
887  **/
888 static void igb_configure(struct igb_adapter *adapter)
889 {
890         struct net_device *netdev = adapter->netdev;
891         int i;
892
893         igb_get_hw_control(adapter);
894         igb_set_multi(netdev);
895
896         igb_restore_vlan(adapter);
897
898         igb_configure_tx(adapter);
899         igb_setup_rctl(adapter);
900         igb_configure_rx(adapter);
901
902         igb_rx_fifo_flush_82575(&adapter->hw);
903
904         /* call igb_desc_unused which always leaves
905          * at least 1 descriptor unused to make sure
906          * next_to_use != next_to_clean */
907         for (i = 0; i < adapter->num_rx_queues; i++) {
908                 struct igb_ring *ring = &adapter->rx_ring[i];
909                 igb_alloc_rx_buffers_adv(ring, igb_desc_unused(ring));
910         }
911
912
913         adapter->tx_queue_len = netdev->tx_queue_len;
914 }
915
916
917 /**
918  * igb_up - Open the interface and prepare it to handle traffic
919  * @adapter: board private structure
920  **/
921
922 int igb_up(struct igb_adapter *adapter)
923 {
924         struct e1000_hw *hw = &adapter->hw;
925         int i;
926
927         /* hardware has been reset, we need to reload some things */
928         igb_configure(adapter);
929
930         clear_bit(__IGB_DOWN, &adapter->state);
931
932         for (i = 0; i < adapter->num_rx_queues; i++)
933                 napi_enable(&adapter->rx_ring[i].napi);
934         if (adapter->msix_entries)
935                 igb_configure_msix(adapter);
936
937         igb_vmm_control(adapter);
938         igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
939         igb_set_vmolr(hw, adapter->vfs_allocated_count);
940
941         /* Clear any pending interrupts. */
942         rd32(E1000_ICR);
943         igb_irq_enable(adapter);
944
945         /* Fire a link change interrupt to start the watchdog. */
946         wr32(E1000_ICS, E1000_ICS_LSC);
947         return 0;
948 }
949
950 void igb_down(struct igb_adapter *adapter)
951 {
952         struct e1000_hw *hw = &adapter->hw;
953         struct net_device *netdev = adapter->netdev;
954         u32 tctl, rctl;
955         int i;
956
957         /* signal that we're down so the interrupt handler does not
958          * reschedule our watchdog timer */
959         set_bit(__IGB_DOWN, &adapter->state);
960
961         /* disable receives in the hardware */
962         rctl = rd32(E1000_RCTL);
963         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
964         /* flush and sleep below */
965
966         netif_tx_stop_all_queues(netdev);
967
968         /* disable transmits in the hardware */
969         tctl = rd32(E1000_TCTL);
970         tctl &= ~E1000_TCTL_EN;
971         wr32(E1000_TCTL, tctl);
972         /* flush both disables and wait for them to finish */
973         wrfl();
974         msleep(10);
975
976         for (i = 0; i < adapter->num_rx_queues; i++)
977                 napi_disable(&adapter->rx_ring[i].napi);
978
979         igb_irq_disable(adapter);
980
981         del_timer_sync(&adapter->watchdog_timer);
982         del_timer_sync(&adapter->phy_info_timer);
983
984         netdev->tx_queue_len = adapter->tx_queue_len;
985         netif_carrier_off(netdev);
986
987         /* record the stats before reset*/
988         igb_update_stats(adapter);
989
990         adapter->link_speed = 0;
991         adapter->link_duplex = 0;
992
993         if (!pci_channel_offline(adapter->pdev))
994                 igb_reset(adapter);
995         igb_clean_all_tx_rings(adapter);
996         igb_clean_all_rx_rings(adapter);
997 }
998
999 void igb_reinit_locked(struct igb_adapter *adapter)
1000 {
1001         WARN_ON(in_interrupt());
1002         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1003                 msleep(1);
1004         igb_down(adapter);
1005         igb_up(adapter);
1006         clear_bit(__IGB_RESETTING, &adapter->state);
1007 }
1008
1009 void igb_reset(struct igb_adapter *adapter)
1010 {
1011         struct e1000_hw *hw = &adapter->hw;
1012         struct e1000_mac_info *mac = &hw->mac;
1013         struct e1000_fc_info *fc = &hw->fc;
1014         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
1015         u16 hwm;
1016
1017         /* Repartition Pba for greater than 9k mtu
1018          * To take effect CTRL.RST is required.
1019          */
1020         switch (mac->type) {
1021         case e1000_82576:
1022                 pba = E1000_PBA_64K;
1023                 break;
1024         case e1000_82575:
1025         default:
1026                 pba = E1000_PBA_34K;
1027                 break;
1028         }
1029
1030         if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1031             (mac->type < e1000_82576)) {
1032                 /* adjust PBA for jumbo frames */
1033                 wr32(E1000_PBA, pba);
1034
1035                 /* To maintain wire speed transmits, the Tx FIFO should be
1036                  * large enough to accommodate two full transmit packets,
1037                  * rounded up to the next 1KB and expressed in KB.  Likewise,
1038                  * the Rx FIFO should be large enough to accommodate at least
1039                  * one full receive packet and is similarly rounded up and
1040                  * expressed in KB. */
1041                 pba = rd32(E1000_PBA);
1042                 /* upper 16 bits has Tx packet buffer allocation size in KB */
1043                 tx_space = pba >> 16;
1044                 /* lower 16 bits has Rx packet buffer allocation size in KB */
1045                 pba &= 0xffff;
1046                 /* the tx fifo also stores 16 bytes of information about the tx
1047                  * but don't include ethernet FCS because hardware appends it */
1048                 min_tx_space = (adapter->max_frame_size +
1049                                 sizeof(union e1000_adv_tx_desc) -
1050                                 ETH_FCS_LEN) * 2;
1051                 min_tx_space = ALIGN(min_tx_space, 1024);
1052                 min_tx_space >>= 10;
1053                 /* software strips receive CRC, so leave room for it */
1054                 min_rx_space = adapter->max_frame_size;
1055                 min_rx_space = ALIGN(min_rx_space, 1024);
1056                 min_rx_space >>= 10;
1057
1058                 /* If current Tx allocation is less than the min Tx FIFO size,
1059                  * and the min Tx FIFO size is less than the current Rx FIFO
1060                  * allocation, take space away from current Rx allocation */
1061                 if (tx_space < min_tx_space &&
1062                     ((min_tx_space - tx_space) < pba)) {
1063                         pba = pba - (min_tx_space - tx_space);
1064
1065                         /* if short on rx space, rx wins and must trump tx
1066                          * adjustment */
1067                         if (pba < min_rx_space)
1068                                 pba = min_rx_space;
1069                 }
1070                 wr32(E1000_PBA, pba);
1071         }
1072
1073         /* flow control settings */
1074         /* The high water mark must be low enough to fit one full frame
1075          * (or the size used for early receive) above it in the Rx FIFO.
1076          * Set it to the lower of:
1077          * - 90% of the Rx FIFO size, or
1078          * - the full Rx FIFO size minus one full frame */
1079         hwm = min(((pba << 10) * 9 / 10),
1080                         ((pba << 10) - 2 * adapter->max_frame_size));
1081
1082         if (mac->type < e1000_82576) {
1083                 fc->high_water = hwm & 0xFFF8;  /* 8-byte granularity */
1084                 fc->low_water = fc->high_water - 8;
1085         } else {
1086                 fc->high_water = hwm & 0xFFF0;  /* 16-byte granularity */
1087                 fc->low_water = fc->high_water - 16;
1088         }
1089         fc->pause_time = 0xFFFF;
1090         fc->send_xon = 1;
1091         fc->type = fc->original_type;
1092
1093         /* disable receive for all VFs and wait one second */
1094         if (adapter->vfs_allocated_count) {
1095                 int i;
1096                 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1097                         adapter->vf_data[i].clear_to_send = false;
1098
1099                 /* ping all the active vfs to let them know we are going down */
1100                         igb_ping_all_vfs(adapter);
1101
1102                 /* disable transmits and receives */
1103                 wr32(E1000_VFRE, 0);
1104                 wr32(E1000_VFTE, 0);
1105         }
1106
1107         /* Allow time for pending master requests to run */
1108         adapter->hw.mac.ops.reset_hw(&adapter->hw);
1109         wr32(E1000_WUC, 0);
1110
1111         if (adapter->hw.mac.ops.init_hw(&adapter->hw))
1112                 dev_err(&adapter->pdev->dev, "Hardware Error\n");
1113
1114         igb_update_mng_vlan(adapter);
1115
1116         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1117         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1118
1119         igb_reset_adaptive(&adapter->hw);
1120         igb_get_phy_info(&adapter->hw);
1121 }
1122
1123 static const struct net_device_ops igb_netdev_ops = {
1124         .ndo_open               = igb_open,
1125         .ndo_stop               = igb_close,
1126         .ndo_start_xmit         = igb_xmit_frame_adv,
1127         .ndo_get_stats          = igb_get_stats,
1128         .ndo_set_multicast_list = igb_set_multi,
1129         .ndo_set_mac_address    = igb_set_mac,
1130         .ndo_change_mtu         = igb_change_mtu,
1131         .ndo_do_ioctl           = igb_ioctl,
1132         .ndo_tx_timeout         = igb_tx_timeout,
1133         .ndo_validate_addr      = eth_validate_addr,
1134         .ndo_vlan_rx_register   = igb_vlan_rx_register,
1135         .ndo_vlan_rx_add_vid    = igb_vlan_rx_add_vid,
1136         .ndo_vlan_rx_kill_vid   = igb_vlan_rx_kill_vid,
1137 #ifdef CONFIG_NET_POLL_CONTROLLER
1138         .ndo_poll_controller    = igb_netpoll,
1139 #endif
1140 };
1141
1142 /**
1143  * igb_probe - Device Initialization Routine
1144  * @pdev: PCI device information struct
1145  * @ent: entry in igb_pci_tbl
1146  *
1147  * Returns 0 on success, negative on failure
1148  *
1149  * igb_probe initializes an adapter identified by a pci_dev structure.
1150  * The OS initialization, configuring of the adapter private structure,
1151  * and a hardware reset occur.
1152  **/
1153 static int __devinit igb_probe(struct pci_dev *pdev,
1154                                const struct pci_device_id *ent)
1155 {
1156         struct net_device *netdev;
1157         struct igb_adapter *adapter;
1158         struct e1000_hw *hw;
1159         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1160         unsigned long mmio_start, mmio_len;
1161         int err, pci_using_dac;
1162         u16 eeprom_data = 0;
1163         u16 eeprom_apme_mask = IGB_EEPROM_APME;
1164         u32 part_num;
1165
1166         err = pci_enable_device_mem(pdev);
1167         if (err)
1168                 return err;
1169
1170         pci_using_dac = 0;
1171         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1172         if (!err) {
1173                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1174                 if (!err)
1175                         pci_using_dac = 1;
1176         } else {
1177                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1178                 if (err) {
1179                         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1180                         if (err) {
1181                                 dev_err(&pdev->dev, "No usable DMA "
1182                                         "configuration, aborting\n");
1183                                 goto err_dma;
1184                         }
1185                 }
1186         }
1187
1188         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1189                                            IORESOURCE_MEM),
1190                                            igb_driver_name);
1191         if (err)
1192                 goto err_pci_reg;
1193
1194         err = pci_enable_pcie_error_reporting(pdev);
1195         if (err) {
1196                 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
1197                         "0x%x\n", err);
1198                 /* non-fatal, continue */
1199         }
1200
1201         pci_set_master(pdev);
1202         pci_save_state(pdev);
1203
1204         err = -ENOMEM;
1205         netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1206                                    IGB_ABS_MAX_TX_QUEUES);
1207         if (!netdev)
1208                 goto err_alloc_etherdev;
1209
1210         SET_NETDEV_DEV(netdev, &pdev->dev);
1211
1212         pci_set_drvdata(pdev, netdev);
1213         adapter = netdev_priv(netdev);
1214         adapter->netdev = netdev;
1215         adapter->pdev = pdev;
1216         hw = &adapter->hw;
1217         hw->back = adapter;
1218         adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1219
1220         mmio_start = pci_resource_start(pdev, 0);
1221         mmio_len = pci_resource_len(pdev, 0);
1222
1223         err = -EIO;
1224         hw->hw_addr = ioremap(mmio_start, mmio_len);
1225         if (!hw->hw_addr)
1226                 goto err_ioremap;
1227
1228         netdev->netdev_ops = &igb_netdev_ops;
1229         igb_set_ethtool_ops(netdev);
1230         netdev->watchdog_timeo = 5 * HZ;
1231
1232         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1233
1234         netdev->mem_start = mmio_start;
1235         netdev->mem_end = mmio_start + mmio_len;
1236
1237         /* PCI config space info */
1238         hw->vendor_id = pdev->vendor;
1239         hw->device_id = pdev->device;
1240         hw->revision_id = pdev->revision;
1241         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1242         hw->subsystem_device_id = pdev->subsystem_device;
1243
1244         /* setup the private structure */
1245         hw->back = adapter;
1246         /* Copy the default MAC, PHY and NVM function pointers */
1247         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1248         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1249         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1250         /* Initialize skew-specific constants */
1251         err = ei->get_invariants(hw);
1252         if (err)
1253                 goto err_sw_init;
1254
1255 #ifdef CONFIG_PCI_IOV
1256         /* since iov functionality isn't critical to base device function we
1257          * can accept failure.  If it fails we don't allow iov to be enabled */
1258         if (hw->mac.type == e1000_82576) {
1259                 /* 82576 supports a maximum of 7 VFs in addition to the PF */
1260                 unsigned int num_vfs = (max_vfs > 7) ? 7 : max_vfs;
1261                 int i;
1262                 unsigned char mac_addr[ETH_ALEN];
1263
1264                 if (num_vfs) {
1265                         adapter->vf_data = kcalloc(num_vfs,
1266                                                 sizeof(struct vf_data_storage),
1267                                                 GFP_KERNEL);
1268                         if (!adapter->vf_data) {
1269                                 dev_err(&pdev->dev,
1270                                         "Could not allocate VF private data - "
1271                                         "IOV enable failed\n");
1272                         } else {
1273                                 err = pci_enable_sriov(pdev, num_vfs);
1274                                 if (!err) {
1275                                         adapter->vfs_allocated_count = num_vfs;
1276                                         dev_info(&pdev->dev,
1277                                                  "%d vfs allocated\n",
1278                                                  num_vfs);
1279                                         for (i = 0;
1280                                              i < adapter->vfs_allocated_count;
1281                                              i++) {
1282                                                 random_ether_addr(mac_addr);
1283                                                 igb_set_vf_mac(adapter, i,
1284                                                                mac_addr);
1285                                         }
1286                                 } else {
1287                                         kfree(adapter->vf_data);
1288                                         adapter->vf_data = NULL;
1289                                 }
1290                         }
1291                 }
1292         }
1293
1294 #endif
1295         /* setup the private structure */
1296         err = igb_sw_init(adapter);
1297         if (err)
1298                 goto err_sw_init;
1299
1300         igb_get_bus_info_pcie(hw);
1301
1302         /* set flags */
1303         switch (hw->mac.type) {
1304         case e1000_82575:
1305                 adapter->flags |= IGB_FLAG_NEED_CTX_IDX;
1306                 break;
1307         case e1000_82576:
1308         default:
1309                 break;
1310         }
1311
1312         hw->phy.autoneg_wait_to_complete = false;
1313         hw->mac.adaptive_ifs = true;
1314
1315         /* Copper options */
1316         if (hw->phy.media_type == e1000_media_type_copper) {
1317                 hw->phy.mdix = AUTO_ALL_MODES;
1318                 hw->phy.disable_polarity_correction = false;
1319                 hw->phy.ms_type = e1000_ms_hw_default;
1320         }
1321
1322         if (igb_check_reset_block(hw))
1323                 dev_info(&pdev->dev,
1324                         "PHY reset is blocked due to SOL/IDER session.\n");
1325
1326         netdev->features = NETIF_F_SG |
1327                            NETIF_F_IP_CSUM |
1328                            NETIF_F_HW_VLAN_TX |
1329                            NETIF_F_HW_VLAN_RX |
1330                            NETIF_F_HW_VLAN_FILTER;
1331
1332         netdev->features |= NETIF_F_IPV6_CSUM;
1333         netdev->features |= NETIF_F_TSO;
1334         netdev->features |= NETIF_F_TSO6;
1335
1336         netdev->features |= NETIF_F_GRO;
1337
1338         netdev->vlan_features |= NETIF_F_TSO;
1339         netdev->vlan_features |= NETIF_F_TSO6;
1340         netdev->vlan_features |= NETIF_F_IP_CSUM;
1341         netdev->vlan_features |= NETIF_F_SG;
1342
1343         if (pci_using_dac)
1344                 netdev->features |= NETIF_F_HIGHDMA;
1345
1346         adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1347
1348         /* before reading the NVM, reset the controller to put the device in a
1349          * known good starting state */
1350         hw->mac.ops.reset_hw(hw);
1351
1352         /* make sure the NVM is good */
1353         if (igb_validate_nvm_checksum(hw) < 0) {
1354                 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1355                 err = -EIO;
1356                 goto err_eeprom;
1357         }
1358
1359         /* copy the MAC address out of the NVM */
1360         if (hw->mac.ops.read_mac_addr(hw))
1361                 dev_err(&pdev->dev, "NVM Read Error\n");
1362
1363         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1364         memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1365
1366         if (!is_valid_ether_addr(netdev->perm_addr)) {
1367                 dev_err(&pdev->dev, "Invalid MAC Address\n");
1368                 err = -EIO;
1369                 goto err_eeprom;
1370         }
1371
1372         setup_timer(&adapter->watchdog_timer, &igb_watchdog,
1373                     (unsigned long) adapter);
1374         setup_timer(&adapter->phy_info_timer, &igb_update_phy_info,
1375                     (unsigned long) adapter);
1376
1377         INIT_WORK(&adapter->reset_task, igb_reset_task);
1378         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1379
1380         /* Initialize link properties that are user-changeable */
1381         adapter->fc_autoneg = true;
1382         hw->mac.autoneg = true;
1383         hw->phy.autoneg_advertised = 0x2f;
1384
1385         hw->fc.original_type = e1000_fc_default;
1386         hw->fc.type = e1000_fc_default;
1387
1388         adapter->itr_setting = IGB_DEFAULT_ITR;
1389         adapter->itr = IGB_START_ITR;
1390
1391         igb_validate_mdi_setting(hw);
1392
1393         adapter->rx_csum = 1;
1394
1395         /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1396          * enable the ACPI Magic Packet filter
1397          */
1398
1399         if (hw->bus.func == 0)
1400                 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1401         else if (hw->bus.func == 1)
1402                 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1403
1404         if (eeprom_data & eeprom_apme_mask)
1405                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1406
1407         /* now that we have the eeprom settings, apply the special cases where
1408          * the eeprom may be wrong or the board simply won't support wake on
1409          * lan on a particular port */
1410         switch (pdev->device) {
1411         case E1000_DEV_ID_82575GB_QUAD_COPPER:
1412                 adapter->eeprom_wol = 0;
1413                 break;
1414         case E1000_DEV_ID_82575EB_FIBER_SERDES:
1415         case E1000_DEV_ID_82576_FIBER:
1416         case E1000_DEV_ID_82576_SERDES:
1417                 /* Wake events only supported on port A for dual fiber
1418                  * regardless of eeprom setting */
1419                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1420                         adapter->eeprom_wol = 0;
1421                 break;
1422         case E1000_DEV_ID_82576_QUAD_COPPER:
1423                 /* if quad port adapter, disable WoL on all but port A */
1424                 if (global_quad_port_a != 0)
1425                         adapter->eeprom_wol = 0;
1426                 else
1427                         adapter->flags |= IGB_FLAG_QUAD_PORT_A;
1428                 /* Reset for multiple quad port adapters */
1429                 if (++global_quad_port_a == 4)
1430                         global_quad_port_a = 0;
1431                 break;
1432         }
1433
1434         /* initialize the wol settings based on the eeprom settings */
1435         adapter->wol = adapter->eeprom_wol;
1436         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1437
1438         /* reset the hardware with the new settings */
1439         igb_reset(adapter);
1440
1441         /* let the f/w know that the h/w is now under the control of the
1442          * driver. */
1443         igb_get_hw_control(adapter);
1444
1445         /* tell the stack to leave us alone until igb_open() is called */
1446         netif_carrier_off(netdev);
1447         netif_tx_stop_all_queues(netdev);
1448
1449         strcpy(netdev->name, "eth%d");
1450         err = register_netdev(netdev);
1451         if (err)
1452                 goto err_register;
1453
1454 #ifdef CONFIG_IGB_DCA
1455         if (dca_add_requester(&pdev->dev) == 0) {
1456                 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1457                 dev_info(&pdev->dev, "DCA enabled\n");
1458                 /* Always use CB2 mode, difference is masked
1459                  * in the CB driver. */
1460                 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
1461                 igb_setup_dca(adapter);
1462         }
1463 #endif
1464
1465         /*
1466          * Initialize hardware timer: we keep it running just in case
1467          * that some program needs it later on.
1468          */
1469         memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1470         adapter->cycles.read = igb_read_clock;
1471         adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1472         adapter->cycles.mult = 1;
1473         adapter->cycles.shift = IGB_TSYNC_SHIFT;
1474         wr32(E1000_TIMINCA,
1475              (1<<24) |
1476              IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS * IGB_TSYNC_SCALE);
1477 #if 0
1478         /*
1479          * Avoid rollover while we initialize by resetting the time counter.
1480          */
1481         wr32(E1000_SYSTIML, 0x00000000);
1482         wr32(E1000_SYSTIMH, 0x00000000);
1483 #else
1484         /*
1485          * Set registers so that rollover occurs soon to test this.
1486          */
1487         wr32(E1000_SYSTIML, 0x00000000);
1488         wr32(E1000_SYSTIMH, 0xFF800000);
1489 #endif
1490         wrfl();
1491         timecounter_init(&adapter->clock,
1492                          &adapter->cycles,
1493                          ktime_to_ns(ktime_get_real()));
1494
1495         /*
1496          * Synchronize our NIC clock against system wall clock. NIC
1497          * time stamp reading requires ~3us per sample, each sample
1498          * was pretty stable even under load => only require 10
1499          * samples for each offset comparison.
1500          */
1501         memset(&adapter->compare, 0, sizeof(adapter->compare));
1502         adapter->compare.source = &adapter->clock;
1503         adapter->compare.target = ktime_get_real;
1504         adapter->compare.num_samples = 10;
1505         timecompare_update(&adapter->compare, 0);
1506
1507 #ifdef DEBUG
1508         {
1509                 char buffer[160];
1510                 printk(KERN_DEBUG
1511                         "igb: %s: hw %p initialized timer\n",
1512                         igb_get_time_str(adapter, buffer),
1513                         &adapter->hw);
1514         }
1515 #endif
1516
1517         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1518         /* print bus type/speed/width info */
1519         dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
1520                  netdev->name,
1521                  ((hw->bus.speed == e1000_bus_speed_2500)
1522                   ? "2.5Gb/s" : "unknown"),
1523                  ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
1524                   (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
1525                   (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
1526                    "unknown"),
1527                  netdev->dev_addr);
1528
1529         igb_read_part_num(hw, &part_num);
1530         dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1531                 (part_num >> 8), (part_num & 0xff));
1532
1533         dev_info(&pdev->dev,
1534                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1535                 adapter->msix_entries ? "MSI-X" :
1536                 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1537                 adapter->num_rx_queues, adapter->num_tx_queues);
1538
1539         return 0;
1540
1541 err_register:
1542         igb_release_hw_control(adapter);
1543 err_eeprom:
1544         if (!igb_check_reset_block(hw))
1545                 igb_reset_phy(hw);
1546
1547         if (hw->flash_address)
1548                 iounmap(hw->flash_address);
1549
1550         igb_free_queues(adapter);
1551 err_sw_init:
1552         iounmap(hw->hw_addr);
1553 err_ioremap:
1554         free_netdev(netdev);
1555 err_alloc_etherdev:
1556         pci_release_selected_regions(pdev, pci_select_bars(pdev,
1557                                      IORESOURCE_MEM));
1558 err_pci_reg:
1559 err_dma:
1560         pci_disable_device(pdev);
1561         return err;
1562 }
1563
1564 /**
1565  * igb_remove - Device Removal Routine
1566  * @pdev: PCI device information struct
1567  *
1568  * igb_remove is called by the PCI subsystem to alert the driver
1569  * that it should release a PCI device.  The could be caused by a
1570  * Hot-Plug event, or because the driver is going to be removed from
1571  * memory.
1572  **/
1573 static void __devexit igb_remove(struct pci_dev *pdev)
1574 {
1575         struct net_device *netdev = pci_get_drvdata(pdev);
1576         struct igb_adapter *adapter = netdev_priv(netdev);
1577         struct e1000_hw *hw = &adapter->hw;
1578         int err;
1579
1580         /* flush_scheduled work may reschedule our watchdog task, so
1581          * explicitly disable watchdog tasks from being rescheduled  */
1582         set_bit(__IGB_DOWN, &adapter->state);
1583         del_timer_sync(&adapter->watchdog_timer);
1584         del_timer_sync(&adapter->phy_info_timer);
1585
1586         flush_scheduled_work();
1587
1588 #ifdef CONFIG_IGB_DCA
1589         if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1590                 dev_info(&pdev->dev, "DCA disabled\n");
1591                 dca_remove_requester(&pdev->dev);
1592                 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1593                 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
1594         }
1595 #endif
1596
1597         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1598          * would have already happened in close and is redundant. */
1599         igb_release_hw_control(adapter);
1600
1601         unregister_netdev(netdev);
1602
1603         if (!igb_check_reset_block(&adapter->hw))
1604                 igb_reset_phy(&adapter->hw);
1605
1606         igb_reset_interrupt_capability(adapter);
1607
1608         igb_free_queues(adapter);
1609
1610 #ifdef CONFIG_PCI_IOV
1611         /* reclaim resources allocated to VFs */
1612         if (adapter->vf_data) {
1613                 /* disable iov and allow time for transactions to clear */
1614                 pci_disable_sriov(pdev);
1615                 msleep(500);
1616
1617                 kfree(adapter->vf_data);
1618                 adapter->vf_data = NULL;
1619                 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1620                 msleep(100);
1621                 dev_info(&pdev->dev, "IOV Disabled\n");
1622         }
1623 #endif
1624         iounmap(hw->hw_addr);
1625         if (hw->flash_address)
1626                 iounmap(hw->flash_address);
1627         pci_release_selected_regions(pdev, pci_select_bars(pdev,
1628                                      IORESOURCE_MEM));
1629
1630         free_netdev(netdev);
1631
1632         err = pci_disable_pcie_error_reporting(pdev);
1633         if (err)
1634                 dev_err(&pdev->dev,
1635                         "pci_disable_pcie_error_reporting failed 0x%x\n", err);
1636
1637         pci_disable_device(pdev);
1638 }
1639
1640 /**
1641  * igb_sw_init - Initialize general software structures (struct igb_adapter)
1642  * @adapter: board private structure to initialize
1643  *
1644  * igb_sw_init initializes the Adapter private data structure.
1645  * Fields are initialized based on PCI device information and
1646  * OS network device settings (MTU size).
1647  **/
1648 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1649 {
1650         struct e1000_hw *hw = &adapter->hw;
1651         struct net_device *netdev = adapter->netdev;
1652         struct pci_dev *pdev = adapter->pdev;
1653
1654         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1655
1656         adapter->tx_ring_count = IGB_DEFAULT_TXD;
1657         adapter->rx_ring_count = IGB_DEFAULT_RXD;
1658         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1659         adapter->rx_ps_hdr_size = 0; /* disable packet split */
1660         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1661         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1662
1663         /* This call may decrease the number of queues depending on
1664          * interrupt mode. */
1665         igb_set_interrupt_capability(adapter);
1666
1667         if (igb_alloc_queues(adapter)) {
1668                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1669                 return -ENOMEM;
1670         }
1671
1672         /* Explicitly disable IRQ since the NIC can be in any state. */
1673         igb_irq_disable(adapter);
1674
1675         set_bit(__IGB_DOWN, &adapter->state);
1676         return 0;
1677 }
1678
1679 /**
1680  * igb_open - Called when a network interface is made active
1681  * @netdev: network interface device structure
1682  *
1683  * Returns 0 on success, negative value on failure
1684  *
1685  * The open entry point is called when a network interface is made
1686  * active by the system (IFF_UP).  At this point all resources needed
1687  * for transmit and receive operations are allocated, the interrupt
1688  * handler is registered with the OS, the watchdog timer is started,
1689  * and the stack is notified that the interface is ready.
1690  **/
1691 static int igb_open(struct net_device *netdev)
1692 {
1693         struct igb_adapter *adapter = netdev_priv(netdev);
1694         struct e1000_hw *hw = &adapter->hw;
1695         int err;
1696         int i;
1697
1698         /* disallow open during test */
1699         if (test_bit(__IGB_TESTING, &adapter->state))
1700                 return -EBUSY;
1701
1702         /* allocate transmit descriptors */
1703         err = igb_setup_all_tx_resources(adapter);
1704         if (err)
1705                 goto err_setup_tx;
1706
1707         /* allocate receive descriptors */
1708         err = igb_setup_all_rx_resources(adapter);
1709         if (err)
1710                 goto err_setup_rx;
1711
1712         /* e1000_power_up_phy(adapter); */
1713
1714         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1715         if ((adapter->hw.mng_cookie.status &
1716              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1717                 igb_update_mng_vlan(adapter);
1718
1719         /* before we allocate an interrupt, we must be ready to handle it.
1720          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1721          * as soon as we call pci_request_irq, so we have to setup our
1722          * clean_rx handler before we do so.  */
1723         igb_configure(adapter);
1724
1725         igb_vmm_control(adapter);
1726         igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
1727         igb_set_vmolr(hw, adapter->vfs_allocated_count);
1728
1729         err = igb_request_irq(adapter);
1730         if (err)
1731                 goto err_req_irq;
1732
1733         /* From here on the code is the same as igb_up() */
1734         clear_bit(__IGB_DOWN, &adapter->state);
1735
1736         for (i = 0; i < adapter->num_rx_queues; i++)
1737                 napi_enable(&adapter->rx_ring[i].napi);
1738
1739         /* Clear any pending interrupts. */
1740         rd32(E1000_ICR);
1741
1742         igb_irq_enable(adapter);
1743
1744         netif_tx_start_all_queues(netdev);
1745
1746         /* Fire a link status change interrupt to start the watchdog. */
1747         wr32(E1000_ICS, E1000_ICS_LSC);
1748
1749         return 0;
1750
1751 err_req_irq:
1752         igb_release_hw_control(adapter);
1753         /* e1000_power_down_phy(adapter); */
1754         igb_free_all_rx_resources(adapter);
1755 err_setup_rx:
1756         igb_free_all_tx_resources(adapter);
1757 err_setup_tx:
1758         igb_reset(adapter);
1759
1760         return err;
1761 }
1762
1763 /**
1764  * igb_close - Disables a network interface
1765  * @netdev: network interface device structure
1766  *
1767  * Returns 0, this is not allowed to fail
1768  *
1769  * The close entry point is called when an interface is de-activated
1770  * by the OS.  The hardware is still under the driver's control, but
1771  * needs to be disabled.  A global MAC reset is issued to stop the
1772  * hardware, and all transmit and receive resources are freed.
1773  **/
1774 static int igb_close(struct net_device *netdev)
1775 {
1776         struct igb_adapter *adapter = netdev_priv(netdev);
1777
1778         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1779         igb_down(adapter);
1780
1781         igb_free_irq(adapter);
1782
1783         igb_free_all_tx_resources(adapter);
1784         igb_free_all_rx_resources(adapter);
1785
1786         /* kill manageability vlan ID if supported, but not if a vlan with
1787          * the same ID is registered on the host OS (let 8021q kill it) */
1788         if ((adapter->hw.mng_cookie.status &
1789                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1790              !(adapter->vlgrp &&
1791                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1792                 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1793
1794         return 0;
1795 }
1796
1797 /**
1798  * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1799  * @adapter: board private structure
1800  * @tx_ring: tx descriptor ring (for a specific queue) to setup
1801  *
1802  * Return 0 on success, negative on failure
1803  **/
1804 int igb_setup_tx_resources(struct igb_adapter *adapter,
1805                            struct igb_ring *tx_ring)
1806 {
1807         struct pci_dev *pdev = adapter->pdev;
1808         int size;
1809
1810         size = sizeof(struct igb_buffer) * tx_ring->count;
1811         tx_ring->buffer_info = vmalloc(size);
1812         if (!tx_ring->buffer_info)
1813                 goto err;
1814         memset(tx_ring->buffer_info, 0, size);
1815
1816         /* round up to nearest 4K */
1817         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
1818         tx_ring->size = ALIGN(tx_ring->size, 4096);
1819
1820         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1821                                              &tx_ring->dma);
1822
1823         if (!tx_ring->desc)
1824                 goto err;
1825
1826         tx_ring->adapter = adapter;
1827         tx_ring->next_to_use = 0;
1828         tx_ring->next_to_clean = 0;
1829         return 0;
1830
1831 err:
1832         vfree(tx_ring->buffer_info);
1833         dev_err(&adapter->pdev->dev,
1834                 "Unable to allocate memory for the transmit descriptor ring\n");
1835         return -ENOMEM;
1836 }
1837
1838 /**
1839  * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1840  *                                (Descriptors) for all queues
1841  * @adapter: board private structure
1842  *
1843  * Return 0 on success, negative on failure
1844  **/
1845 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1846 {
1847         int i, err = 0;
1848         int r_idx;
1849
1850         for (i = 0; i < adapter->num_tx_queues; i++) {
1851                 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1852                 if (err) {
1853                         dev_err(&adapter->pdev->dev,
1854                                 "Allocation for Tx Queue %u failed\n", i);
1855                         for (i--; i >= 0; i--)
1856                                 igb_free_tx_resources(&adapter->tx_ring[i]);
1857                         break;
1858                 }
1859         }
1860
1861         for (i = 0; i < IGB_MAX_TX_QUEUES; i++) {
1862                 r_idx = i % adapter->num_tx_queues;
1863                 adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
1864         }
1865         return err;
1866 }
1867
1868 /**
1869  * igb_configure_tx - Configure transmit Unit after Reset
1870  * @adapter: board private structure
1871  *
1872  * Configure the Tx unit of the MAC after a reset.
1873  **/
1874 static void igb_configure_tx(struct igb_adapter *adapter)
1875 {
1876         u64 tdba;
1877         struct e1000_hw *hw = &adapter->hw;
1878         u32 tctl;
1879         u32 txdctl, txctrl;
1880         int i, j;
1881
1882         for (i = 0; i < adapter->num_tx_queues; i++) {
1883                 struct igb_ring *ring = &adapter->tx_ring[i];
1884                 j = ring->reg_idx;
1885                 wr32(E1000_TDLEN(j),
1886                      ring->count * sizeof(union e1000_adv_tx_desc));
1887                 tdba = ring->dma;
1888                 wr32(E1000_TDBAL(j),
1889                      tdba & 0x00000000ffffffffULL);
1890                 wr32(E1000_TDBAH(j), tdba >> 32);
1891
1892                 ring->head = E1000_TDH(j);
1893                 ring->tail = E1000_TDT(j);
1894                 writel(0, hw->hw_addr + ring->tail);
1895                 writel(0, hw->hw_addr + ring->head);
1896                 txdctl = rd32(E1000_TXDCTL(j));
1897                 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1898                 wr32(E1000_TXDCTL(j), txdctl);
1899
1900                 /* Turn off Relaxed Ordering on head write-backs.  The
1901                  * writebacks MUST be delivered in order or it will
1902                  * completely screw up our bookeeping.
1903                  */
1904                 txctrl = rd32(E1000_DCA_TXCTRL(j));
1905                 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1906                 wr32(E1000_DCA_TXCTRL(j), txctrl);
1907         }
1908
1909         /* disable queue 0 to prevent tail bump w/o re-configuration */
1910         if (adapter->vfs_allocated_count)
1911                 wr32(E1000_TXDCTL(0), 0);
1912
1913         /* Program the Transmit Control Register */
1914         tctl = rd32(E1000_TCTL);
1915         tctl &= ~E1000_TCTL_CT;
1916         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1917                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1918
1919         igb_config_collision_dist(hw);
1920
1921         /* Setup Transmit Descriptor Settings for eop descriptor */
1922         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1923
1924         /* Enable transmits */
1925         tctl |= E1000_TCTL_EN;
1926
1927         wr32(E1000_TCTL, tctl);
1928 }
1929
1930 /**
1931  * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1932  * @adapter: board private structure
1933  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
1934  *
1935  * Returns 0 on success, negative on failure
1936  **/
1937 int igb_setup_rx_resources(struct igb_adapter *adapter,
1938                            struct igb_ring *rx_ring)
1939 {
1940         struct pci_dev *pdev = adapter->pdev;
1941         int size, desc_len;
1942
1943         size = sizeof(struct igb_buffer) * rx_ring->count;
1944         rx_ring->buffer_info = vmalloc(size);
1945         if (!rx_ring->buffer_info)
1946                 goto err;
1947         memset(rx_ring->buffer_info, 0, size);
1948
1949         desc_len = sizeof(union e1000_adv_rx_desc);
1950
1951         /* Round up to nearest 4K */
1952         rx_ring->size = rx_ring->count * desc_len;
1953         rx_ring->size = ALIGN(rx_ring->size, 4096);
1954
1955         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1956                                              &rx_ring->dma);
1957
1958         if (!rx_ring->desc)
1959                 goto err;
1960
1961         rx_ring->next_to_clean = 0;
1962         rx_ring->next_to_use = 0;
1963
1964         rx_ring->adapter = adapter;
1965
1966         return 0;
1967
1968 err:
1969         vfree(rx_ring->buffer_info);
1970         dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1971                 "the receive descriptor ring\n");
1972         return -ENOMEM;
1973 }
1974
1975 /**
1976  * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1977  *                                (Descriptors) for all queues
1978  * @adapter: board private structure
1979  *
1980  * Return 0 on success, negative on failure
1981  **/
1982 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1983 {
1984         int i, err = 0;
1985
1986         for (i = 0; i < adapter->num_rx_queues; i++) {
1987                 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1988                 if (err) {
1989                         dev_err(&adapter->pdev->dev,
1990                                 "Allocation for Rx Queue %u failed\n", i);
1991                         for (i--; i >= 0; i--)
1992                                 igb_free_rx_resources(&adapter->rx_ring[i]);
1993                         break;
1994                 }
1995         }
1996
1997         return err;
1998 }
1999
2000 /**
2001  * igb_setup_rctl - configure the receive control registers
2002  * @adapter: Board private structure
2003  **/
2004 static void igb_setup_rctl(struct igb_adapter *adapter)
2005 {
2006         struct e1000_hw *hw = &adapter->hw;
2007         u32 rctl;
2008         u32 srrctl = 0;
2009         int i, j;
2010
2011         rctl = rd32(E1000_RCTL);
2012
2013         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2014         rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2015
2016         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2017                 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2018
2019         /*
2020          * enable stripping of CRC. It's unlikely this will break BMC
2021          * redirection as it did with e1000. Newer features require
2022          * that the HW strips the CRC.
2023          */
2024         rctl |= E1000_RCTL_SECRC;
2025
2026         /*
2027          * disable store bad packets and clear size bits.
2028          */
2029         rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2030
2031         /* enable LPE when to prevent packets larger than max_frame_size */
2032                 rctl |= E1000_RCTL_LPE;
2033
2034         /* Setup buffer sizes */
2035         switch (adapter->rx_buffer_len) {
2036         case IGB_RXBUFFER_256:
2037                 rctl |= E1000_RCTL_SZ_256;
2038                 break;
2039         case IGB_RXBUFFER_512:
2040                 rctl |= E1000_RCTL_SZ_512;
2041                 break;
2042         default:
2043                 srrctl = ALIGN(adapter->rx_buffer_len, 1024)
2044                          >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2045                 break;
2046         }
2047
2048         /* 82575 and greater support packet-split where the protocol
2049          * header is placed in skb->data and the packet data is
2050          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2051          * In the case of a non-split, skb->data is linearly filled,
2052          * followed by the page buffers.  Therefore, skb->data is
2053          * sized to hold the largest protocol header.
2054          */
2055         /* allocations using alloc_page take too long for regular MTU
2056          * so only enable packet split for jumbo frames */
2057         if (adapter->netdev->mtu > ETH_DATA_LEN) {
2058                 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
2059                 srrctl |= adapter->rx_ps_hdr_size <<
2060                          E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2061                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2062         } else {
2063                 adapter->rx_ps_hdr_size = 0;
2064                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2065         }
2066
2067         /* Attention!!!  For SR-IOV PF driver operations you must enable
2068          * queue drop for all VF and PF queues to prevent head of line blocking
2069          * if an un-trusted VF does not provide descriptors to hardware.
2070          */
2071         if (adapter->vfs_allocated_count) {
2072                 u32 vmolr;
2073
2074                 j = adapter->rx_ring[0].reg_idx;
2075
2076                 /* set all queue drop enable bits */
2077                 wr32(E1000_QDE, ALL_QUEUES);
2078                 srrctl |= E1000_SRRCTL_DROP_EN;
2079
2080                 /* disable queue 0 to prevent tail write w/o re-config */
2081                 wr32(E1000_RXDCTL(0), 0);
2082
2083                 vmolr = rd32(E1000_VMOLR(j));
2084                 if (rctl & E1000_RCTL_LPE)
2085                         vmolr |= E1000_VMOLR_LPE;
2086                 if (adapter->num_rx_queues > 0)
2087                         vmolr |= E1000_VMOLR_RSSE;
2088                 wr32(E1000_VMOLR(j), vmolr);
2089         }
2090
2091         for (i = 0; i < adapter->num_rx_queues; i++) {
2092                 j = adapter->rx_ring[i].reg_idx;
2093                 wr32(E1000_SRRCTL(j), srrctl);
2094         }
2095
2096         wr32(E1000_RCTL, rctl);
2097 }
2098
2099 /**
2100  * igb_rlpml_set - set maximum receive packet size
2101  * @adapter: board private structure
2102  *
2103  * Configure maximum receivable packet size.
2104  **/
2105 static void igb_rlpml_set(struct igb_adapter *adapter)
2106 {
2107         u32 max_frame_size = adapter->max_frame_size;
2108         struct e1000_hw *hw = &adapter->hw;
2109         u16 pf_id = adapter->vfs_allocated_count;
2110
2111         if (adapter->vlgrp)
2112                 max_frame_size += VLAN_TAG_SIZE;
2113
2114         /* if vfs are enabled we set RLPML to the largest possible request
2115          * size and set the VMOLR RLPML to the size we need */
2116         if (pf_id) {
2117                 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
2118                 max_frame_size = MAX_STD_JUMBO_FRAME_SIZE + VLAN_TAG_SIZE;
2119         }
2120
2121         wr32(E1000_RLPML, max_frame_size);
2122 }
2123
2124 /**
2125  * igb_configure_vt_default_pool - Configure VT default pool
2126  * @adapter: board private structure
2127  *
2128  * Configure the default pool
2129  **/
2130 static void igb_configure_vt_default_pool(struct igb_adapter *adapter)
2131 {
2132         struct e1000_hw *hw = &adapter->hw;
2133         u16 pf_id = adapter->vfs_allocated_count;
2134         u32 vtctl;
2135
2136         /* not in sr-iov mode - do nothing */
2137         if (!pf_id)
2138                 return;
2139
2140         vtctl = rd32(E1000_VT_CTL);
2141         vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2142                    E1000_VT_CTL_DISABLE_DEF_POOL);
2143         vtctl |= pf_id << E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2144         wr32(E1000_VT_CTL, vtctl);
2145 }
2146
2147 /**
2148  * igb_configure_rx - Configure receive Unit after Reset
2149  * @adapter: board private structure
2150  *
2151  * Configure the Rx unit of the MAC after a reset.
2152  **/
2153 static void igb_configure_rx(struct igb_adapter *adapter)
2154 {
2155         u64 rdba;
2156         struct e1000_hw *hw = &adapter->hw;
2157         u32 rctl, rxcsum;
2158         u32 rxdctl;
2159         int i;
2160
2161         /* disable receives while setting up the descriptors */
2162         rctl = rd32(E1000_RCTL);
2163         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2164         wrfl();
2165         mdelay(10);
2166
2167         if (adapter->itr_setting > 3)
2168                 wr32(E1000_ITR, adapter->itr);
2169
2170         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2171          * the Base and Length of the Rx Descriptor Ring */
2172         for (i = 0; i < adapter->num_rx_queues; i++) {
2173                 struct igb_ring *ring = &adapter->rx_ring[i];
2174                 int j = ring->reg_idx;
2175                 rdba = ring->dma;
2176                 wr32(E1000_RDBAL(j),
2177                      rdba & 0x00000000ffffffffULL);
2178                 wr32(E1000_RDBAH(j), rdba >> 32);
2179                 wr32(E1000_RDLEN(j),
2180                      ring->count * sizeof(union e1000_adv_rx_desc));
2181
2182                 ring->head = E1000_RDH(j);
2183                 ring->tail = E1000_RDT(j);
2184                 writel(0, hw->hw_addr + ring->tail);
2185                 writel(0, hw->hw_addr + ring->head);
2186
2187                 rxdctl = rd32(E1000_RXDCTL(j));
2188                 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2189                 rxdctl &= 0xFFF00000;
2190                 rxdctl |= IGB_RX_PTHRESH;
2191                 rxdctl |= IGB_RX_HTHRESH << 8;
2192                 rxdctl |= IGB_RX_WTHRESH << 16;
2193                 wr32(E1000_RXDCTL(j), rxdctl);
2194         }
2195
2196         if (adapter->num_rx_queues > 1) {
2197                 u32 random[10];
2198                 u32 mrqc;
2199                 u32 j, shift;
2200                 union e1000_reta {
2201                         u32 dword;
2202                         u8  bytes[4];
2203                 } reta;
2204
2205                 get_random_bytes(&random[0], 40);
2206
2207                 if (hw->mac.type >= e1000_82576)
2208                         shift = 0;
2209                 else
2210                         shift = 6;
2211                 for (j = 0; j < (32 * 4); j++) {
2212                         reta.bytes[j & 3] =
2213                                 adapter->rx_ring[(j % adapter->num_rx_queues)].reg_idx << shift;
2214                         if ((j & 3) == 3)
2215                                 writel(reta.dword,
2216                                        hw->hw_addr + E1000_RETA(0) + (j & ~3));
2217                 }
2218                 if (adapter->vfs_allocated_count)
2219                         mrqc = E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2220                 else
2221                         mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2222
2223                 /* Fill out hash function seeds */
2224                 for (j = 0; j < 10; j++)
2225                         array_wr32(E1000_RSSRK(0), j, random[j]);
2226
2227                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2228                          E1000_MRQC_RSS_FIELD_IPV4_TCP);
2229                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
2230                          E1000_MRQC_RSS_FIELD_IPV6_TCP);
2231                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
2232                          E1000_MRQC_RSS_FIELD_IPV6_UDP);
2233                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2234                          E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2235
2236
2237                 wr32(E1000_MRQC, mrqc);
2238
2239                 /* Multiqueue and raw packet checksumming are mutually
2240                  * exclusive.  Note that this not the same as TCP/IP
2241                  * checksumming, which works fine. */
2242                 rxcsum = rd32(E1000_RXCSUM);
2243                 rxcsum |= E1000_RXCSUM_PCSD;
2244                 wr32(E1000_RXCSUM, rxcsum);
2245         } else {
2246                 /* Enable multi-queue for sr-iov */
2247                 if (adapter->vfs_allocated_count)
2248                         wr32(E1000_MRQC, E1000_MRQC_ENABLE_VMDQ);
2249                 /* Enable Receive Checksum Offload for TCP and UDP */
2250                 rxcsum = rd32(E1000_RXCSUM);
2251                 if (adapter->rx_csum)
2252                         rxcsum |= E1000_RXCSUM_TUOFL | E1000_RXCSUM_IPPCSE;
2253                 else
2254                         rxcsum &= ~(E1000_RXCSUM_TUOFL | E1000_RXCSUM_IPPCSE);
2255
2256                 wr32(E1000_RXCSUM, rxcsum);
2257         }
2258
2259         /* Set the default pool for the PF's first queue */
2260         igb_configure_vt_default_pool(adapter);
2261
2262         igb_rlpml_set(adapter);
2263
2264         /* Enable Receives */
2265         wr32(E1000_RCTL, rctl);
2266 }
2267
2268 /**
2269  * igb_free_tx_resources - Free Tx Resources per Queue
2270  * @tx_ring: Tx descriptor ring for a specific queue
2271  *
2272  * Free all transmit software resources
2273  **/
2274 void igb_free_tx_resources(struct igb_ring *tx_ring)
2275 {
2276         struct pci_dev *pdev = tx_ring->adapter->pdev;
2277
2278         igb_clean_tx_ring(tx_ring);
2279
2280         vfree(tx_ring->buffer_info);
2281         tx_ring->buffer_info = NULL;
2282
2283         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2284
2285         tx_ring->desc = NULL;
2286 }
2287
2288 /**
2289  * igb_free_all_tx_resources - Free Tx Resources for All Queues
2290  * @adapter: board private structure
2291  *
2292  * Free all transmit software resources
2293  **/
2294 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2295 {
2296         int i;
2297
2298         for (i = 0; i < adapter->num_tx_queues; i++)
2299                 igb_free_tx_resources(&adapter->tx_ring[i]);
2300 }
2301
2302 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
2303                                            struct igb_buffer *buffer_info)
2304 {
2305         buffer_info->dma = 0;
2306         if (buffer_info->skb) {
2307                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
2308                               DMA_TO_DEVICE);
2309                 dev_kfree_skb_any(buffer_info->skb);
2310                 buffer_info->skb = NULL;
2311         }
2312         buffer_info->time_stamp = 0;
2313         /* buffer_info must be completely set up in the transmit path */
2314 }
2315
2316 /**
2317  * igb_clean_tx_ring - Free Tx Buffers
2318  * @tx_ring: ring to be cleaned
2319  **/
2320 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2321 {
2322         struct igb_adapter *adapter = tx_ring->adapter;
2323         struct igb_buffer *buffer_info;
2324         unsigned long size;
2325         unsigned int i;
2326
2327         if (!tx_ring->buffer_info)
2328                 return;
2329         /* Free all the Tx ring sk_buffs */
2330
2331         for (i = 0; i < tx_ring->count; i++) {
2332                 buffer_info = &tx_ring->buffer_info[i];
2333                 igb_unmap_and_free_tx_resource(adapter, buffer_info);
2334         }
2335
2336         size = sizeof(struct igb_buffer) * tx_ring->count;
2337         memset(tx_ring->buffer_info, 0, size);
2338
2339         /* Zero out the descriptor ring */
2340
2341         memset(tx_ring->desc, 0, tx_ring->size);
2342
2343         tx_ring->next_to_use = 0;
2344         tx_ring->next_to_clean = 0;
2345
2346         writel(0, adapter->hw.hw_addr + tx_ring->head);
2347         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2348 }
2349
2350 /**
2351  * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2352  * @adapter: board private structure
2353  **/
2354 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2355 {
2356         int i;
2357
2358         for (i = 0; i < adapter->num_tx_queues; i++)
2359                 igb_clean_tx_ring(&adapter->tx_ring[i]);
2360 }
2361
2362 /**
2363  * igb_free_rx_resources - Free Rx Resources
2364  * @rx_ring: ring to clean the resources from
2365  *
2366  * Free all receive software resources
2367  **/
2368 void igb_free_rx_resources(struct igb_ring *rx_ring)
2369 {
2370         struct pci_dev *pdev = rx_ring->adapter->pdev;
2371
2372         igb_clean_rx_ring(rx_ring);
2373
2374         vfree(rx_ring->buffer_info);
2375         rx_ring->buffer_info = NULL;
2376
2377         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2378
2379         rx_ring->desc = NULL;
2380 }
2381
2382 /**
2383  * igb_free_all_rx_resources - Free Rx Resources for All Queues
2384  * @adapter: board private structure
2385  *
2386  * Free all receive software resources
2387  **/
2388 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2389 {
2390         int i;
2391
2392         for (i = 0; i < adapter->num_rx_queues; i++)
2393                 igb_free_rx_resources(&adapter->rx_ring[i]);
2394 }
2395
2396 /**
2397  * igb_clean_rx_ring - Free Rx Buffers per Queue
2398  * @rx_ring: ring to free buffers from
2399  **/
2400 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2401 {
2402         struct igb_adapter *adapter = rx_ring->adapter;
2403         struct igb_buffer *buffer_info;
2404         struct pci_dev *pdev = adapter->pdev;
2405         unsigned long size;
2406         unsigned int i;
2407
2408         if (!rx_ring->buffer_info)
2409                 return;
2410         /* Free all the Rx ring sk_buffs */
2411         for (i = 0; i < rx_ring->count; i++) {
2412                 buffer_info = &rx_ring->buffer_info[i];
2413                 if (buffer_info->dma) {
2414                         if (adapter->rx_ps_hdr_size)
2415                                 pci_unmap_single(pdev, buffer_info->dma,
2416                                                  adapter->rx_ps_hdr_size,
2417                                                  PCI_DMA_FROMDEVICE);
2418                         else
2419                                 pci_unmap_single(pdev, buffer_info->dma,
2420                                                  adapter->rx_buffer_len,
2421                                                  PCI_DMA_FROMDEVICE);
2422                         buffer_info->dma = 0;
2423                 }
2424
2425                 if (buffer_info->skb) {
2426                         dev_kfree_skb(buffer_info->skb);
2427                         buffer_info->skb = NULL;
2428                 }
2429                 if (buffer_info->page) {
2430                         if (buffer_info->page_dma)
2431                                 pci_unmap_page(pdev, buffer_info->page_dma,
2432                                                PAGE_SIZE / 2,
2433                                                PCI_DMA_FROMDEVICE);
2434                         put_page(buffer_info->page);
2435                         buffer_info->page = NULL;
2436                         buffer_info->page_dma = 0;
2437                         buffer_info->page_offset = 0;
2438                 }
2439         }
2440
2441         size = sizeof(struct igb_buffer) * rx_ring->count;
2442         memset(rx_ring->buffer_info, 0, size);
2443
2444         /* Zero out the descriptor ring */
2445         memset(rx_ring->desc, 0, rx_ring->size);
2446
2447         rx_ring->next_to_clean = 0;
2448         rx_ring->next_to_use = 0;
2449
2450         writel(0, adapter->hw.hw_addr + rx_ring->head);
2451         writel(0, adapter->hw.hw_addr + rx_ring->tail);
2452 }
2453
2454 /**
2455  * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2456  * @adapter: board private structure
2457  **/
2458 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2459 {
2460         int i;
2461
2462         for (i = 0; i < adapter->num_rx_queues; i++)
2463                 igb_clean_rx_ring(&adapter->rx_ring[i]);
2464 }
2465
2466 /**
2467  * igb_set_mac - Change the Ethernet Address of the NIC
2468  * @netdev: network interface device structure
2469  * @p: pointer to an address structure
2470  *
2471  * Returns 0 on success, negative on failure
2472  **/
2473 static int igb_set_mac(struct net_device *netdev, void *p)
2474 {
2475         struct igb_adapter *adapter = netdev_priv(netdev);
2476         struct e1000_hw *hw = &adapter->hw;
2477         struct sockaddr *addr = p;
2478
2479         if (!is_valid_ether_addr(addr->sa_data))
2480                 return -EADDRNOTAVAIL;
2481
2482         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2483         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
2484
2485         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
2486
2487         igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
2488
2489         return 0;
2490 }
2491
2492 /**
2493  * igb_set_multi - Multicast and Promiscuous mode set
2494  * @netdev: network interface device structure
2495  *
2496  * The set_multi entry point is called whenever the multicast address
2497  * list or the network interface flags are updated.  This routine is
2498  * responsible for configuring the hardware for proper multicast,
2499  * promiscuous mode, and all-multi behavior.
2500  **/
2501 static void igb_set_multi(struct net_device *netdev)
2502 {
2503         struct igb_adapter *adapter = netdev_priv(netdev);
2504         struct e1000_hw *hw = &adapter->hw;
2505         struct e1000_mac_info *mac = &hw->mac;
2506         struct dev_mc_list *mc_ptr;
2507         u8  *mta_list = NULL;
2508         u32 rctl;
2509         int i;
2510
2511         /* Check for Promiscuous and All Multicast modes */
2512
2513         rctl = rd32(E1000_RCTL);
2514
2515         if (netdev->flags & IFF_PROMISC) {
2516                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2517                 rctl &= ~E1000_RCTL_VFE;
2518         } else {
2519                 if (netdev->flags & IFF_ALLMULTI) {
2520                         rctl |= E1000_RCTL_MPE;
2521                         rctl &= ~E1000_RCTL_UPE;
2522                 } else
2523                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2524                 rctl |= E1000_RCTL_VFE;
2525         }
2526         wr32(E1000_RCTL, rctl);
2527
2528         if (netdev->mc_count) {
2529                 mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2530                 if (!mta_list) {
2531                         dev_err(&adapter->pdev->dev,
2532                                 "failed to allocate multicast filter list\n");
2533                         return;
2534                 }
2535         }
2536
2537         /* The shared function expects a packed array of only addresses. */
2538         mc_ptr = netdev->mc_list;
2539
2540         for (i = 0; i < netdev->mc_count; i++) {
2541                 if (!mc_ptr)
2542                         break;
2543                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2544                 mc_ptr = mc_ptr->next;
2545         }
2546         igb_update_mc_addr_list(hw, mta_list, i,
2547                                 adapter->vfs_allocated_count + 1,
2548                                 mac->rar_entry_count);
2549
2550         igb_set_mc_list_pools(adapter, i, mac->rar_entry_count);
2551         igb_restore_vf_multicasts(adapter);
2552
2553         kfree(mta_list);
2554 }
2555
2556 /* Need to wait a few seconds after link up to get diagnostic information from
2557  * the phy */
2558 static void igb_update_phy_info(unsigned long data)
2559 {
2560         struct igb_adapter *adapter = (struct igb_adapter *) data;
2561         igb_get_phy_info(&adapter->hw);
2562 }
2563
2564 /**
2565  * igb_has_link - check shared code for link and determine up/down
2566  * @adapter: pointer to driver private info
2567  **/
2568 static bool igb_has_link(struct igb_adapter *adapter)
2569 {
2570         struct e1000_hw *hw = &adapter->hw;
2571         bool link_active = false;
2572         s32 ret_val = 0;
2573
2574         /* get_link_status is set on LSC (link status) interrupt or
2575          * rx sequence error interrupt.  get_link_status will stay
2576          * false until the e1000_check_for_link establishes link
2577          * for copper adapters ONLY
2578          */
2579         switch (hw->phy.media_type) {
2580         case e1000_media_type_copper:
2581                 if (hw->mac.get_link_status) {
2582                         ret_val = hw->mac.ops.check_for_link(hw);
2583                         link_active = !hw->mac.get_link_status;
2584                 } else {
2585                         link_active = true;
2586                 }
2587                 break;
2588         case e1000_media_type_fiber:
2589                 ret_val = hw->mac.ops.check_for_link(hw);
2590                 link_active = !!(rd32(E1000_STATUS) & E1000_STATUS_LU);
2591                 break;
2592         case e1000_media_type_internal_serdes:
2593                 ret_val = hw->mac.ops.check_for_link(hw);
2594                 link_active = hw->mac.serdes_has_link;
2595                 break;
2596         default:
2597         case e1000_media_type_unknown:
2598                 break;
2599         }
2600
2601         return link_active;
2602 }
2603
2604 /**
2605  * igb_watchdog - Timer Call-back
2606  * @data: pointer to adapter cast into an unsigned long
2607  **/
2608 static void igb_watchdog(unsigned long data)
2609 {
2610         struct igb_adapter *adapter = (struct igb_adapter *)data;
2611         /* Do the rest outside of interrupt context */
2612         schedule_work(&adapter->watchdog_task);
2613 }
2614
2615 static void igb_watchdog_task(struct work_struct *work)
2616 {
2617         struct igb_adapter *adapter = container_of(work,
2618                                         struct igb_adapter, watchdog_task);
2619         struct e1000_hw *hw = &adapter->hw;
2620         struct net_device *netdev = adapter->netdev;
2621         struct igb_ring *tx_ring = adapter->tx_ring;
2622         u32 link;
2623         u32 eics = 0;
2624         int i;
2625
2626         link = igb_has_link(adapter);
2627         if ((netif_carrier_ok(netdev)) && link)
2628                 goto link_up;
2629
2630         if (link) {
2631                 if (!netif_carrier_ok(netdev)) {
2632                         u32 ctrl;
2633                         hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2634                                                    &adapter->link_speed,
2635                                                    &adapter->link_duplex);
2636
2637                         ctrl = rd32(E1000_CTRL);
2638                         /* Links status message must follow this format */
2639                         printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
2640                                  "Flow Control: %s\n",
2641                                  netdev->name,
2642                                  adapter->link_speed,
2643                                  adapter->link_duplex == FULL_DUPLEX ?
2644                                  "Full Duplex" : "Half Duplex",
2645                                  ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2646                                  E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2647                                  E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2648                                  E1000_CTRL_TFCE) ? "TX" : "None")));
2649
2650                         /* tweak tx_queue_len according to speed/duplex and
2651                          * adjust the timeout factor */
2652                         netdev->tx_queue_len = adapter->tx_queue_len;
2653                         adapter->tx_timeout_factor = 1;
2654                         switch (adapter->link_speed) {
2655                         case SPEED_10:
2656                                 netdev->tx_queue_len = 10;
2657                                 adapter->tx_timeout_factor = 14;
2658                                 break;
2659                         case SPEED_100:
2660                                 netdev->tx_queue_len = 100;
2661                                 /* maybe add some timeout factor ? */
2662                                 break;
2663                         }
2664
2665                         netif_carrier_on(netdev);
2666                         netif_tx_wake_all_queues(netdev);
2667
2668                         igb_ping_all_vfs(adapter);
2669
2670                         /* link state has changed, schedule phy info update */
2671                         if (!test_bit(__IGB_DOWN, &adapter->state))
2672                                 mod_timer(&adapter->phy_info_timer,
2673                                           round_jiffies(jiffies + 2 * HZ));
2674                 }
2675         } else {
2676                 if (netif_carrier_ok(netdev)) {
2677                         adapter->link_speed = 0;
2678                         adapter->link_duplex = 0;
2679                         /* Links status message must follow this format */
2680                         printk(KERN_INFO "igb: %s NIC Link is Down\n",
2681                                netdev->name);
2682                         netif_carrier_off(netdev);
2683                         netif_tx_stop_all_queues(netdev);
2684
2685                         igb_ping_all_vfs(adapter);
2686
2687                         /* link state has changed, schedule phy info update */
2688                         if (!test_bit(__IGB_DOWN, &adapter->state))
2689                                 mod_timer(&adapter->phy_info_timer,
2690                                           round_jiffies(jiffies + 2 * HZ));
2691                 }
2692         }
2693
2694 link_up:
2695         igb_update_stats(adapter);
2696
2697         hw->mac.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2698         adapter->tpt_old = adapter->stats.tpt;
2699         hw->mac.collision_delta = adapter->stats.colc - adapter->colc_old;
2700         adapter->colc_old = adapter->stats.colc;
2701
2702         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2703         adapter->gorc_old = adapter->stats.gorc;
2704         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2705         adapter->gotc_old = adapter->stats.gotc;
2706
2707         igb_update_adaptive(&adapter->hw);
2708
2709         if (!netif_carrier_ok(netdev)) {
2710                 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
2711                         /* We've lost link, so the controller stops DMA,
2712                          * but we've got queued Tx work that's never going
2713                          * to get done, so reset controller to flush Tx.
2714                          * (Do the reset outside of interrupt context). */
2715                         adapter->tx_timeout_count++;
2716                         schedule_work(&adapter->reset_task);
2717                 }
2718         }
2719
2720         /* Cause software interrupt to ensure rx ring is cleaned */
2721         if (adapter->msix_entries) {
2722                 for (i = 0; i < adapter->num_rx_queues; i++)
2723                         eics |= adapter->rx_ring[i].eims_value;
2724                 wr32(E1000_EICS, eics);
2725         } else {
2726                 wr32(E1000_ICS, E1000_ICS_RXDMT0);
2727         }
2728
2729         /* Force detection of hung controller every watchdog period */
2730         tx_ring->detect_tx_hung = true;
2731
2732         /* Reset the timer */
2733         if (!test_bit(__IGB_DOWN, &adapter->state))
2734                 mod_timer(&adapter->watchdog_timer,
2735                           round_jiffies(jiffies + 2 * HZ));
2736 }
2737
2738 enum latency_range {
2739         lowest_latency = 0,
2740         low_latency = 1,
2741         bulk_latency = 2,
2742         latency_invalid = 255
2743 };
2744
2745
2746 /**
2747  * igb_update_ring_itr - update the dynamic ITR value based on packet size
2748  *
2749  *      Stores a new ITR value based on strictly on packet size.  This
2750  *      algorithm is less sophisticated than that used in igb_update_itr,
2751  *      due to the difficulty of synchronizing statistics across multiple
2752  *      receive rings.  The divisors and thresholds used by this fuction
2753  *      were determined based on theoretical maximum wire speed and testing
2754  *      data, in order to minimize response time while increasing bulk
2755  *      throughput.
2756  *      This functionality is controlled by the InterruptThrottleRate module
2757  *      parameter (see igb_param.c)
2758  *      NOTE:  This function is called only when operating in a multiqueue
2759  *             receive environment.
2760  * @rx_ring: pointer to ring
2761  **/
2762 static void igb_update_ring_itr(struct igb_ring *rx_ring)
2763 {
2764         int new_val = rx_ring->itr_val;
2765         int avg_wire_size = 0;
2766         struct igb_adapter *adapter = rx_ring->adapter;
2767
2768         if (!rx_ring->total_packets)
2769                 goto clear_counts; /* no packets, so don't do anything */
2770
2771         /* For non-gigabit speeds, just fix the interrupt rate at 4000
2772          * ints/sec - ITR timer value of 120 ticks.
2773          */
2774         if (adapter->link_speed != SPEED_1000) {
2775                 new_val = 120;
2776                 goto set_itr_val;
2777         }
2778         avg_wire_size = rx_ring->total_bytes / rx_ring->total_packets;
2779
2780         /* Add 24 bytes to size to account for CRC, preamble, and gap */
2781         avg_wire_size += 24;
2782
2783         /* Don't starve jumbo frames */
2784         avg_wire_size = min(avg_wire_size, 3000);
2785
2786         /* Give a little boost to mid-size frames */
2787         if ((avg_wire_size > 300) && (avg_wire_size < 1200))
2788                 new_val = avg_wire_size / 3;
2789         else
2790                 new_val = avg_wire_size / 2;
2791
2792 set_itr_val:
2793         if (new_val != rx_ring->itr_val) {
2794                 rx_ring->itr_val = new_val;
2795                 rx_ring->set_itr = 1;
2796         }
2797 clear_counts:
2798         rx_ring->total_bytes = 0;
2799         rx_ring->total_packets = 0;
2800 }
2801
2802 /**
2803  * igb_update_itr - update the dynamic ITR value based on statistics
2804  *      Stores a new ITR value based on packets and byte
2805  *      counts during the last interrupt.  The advantage of per interrupt
2806  *      computation is faster updates and more accurate ITR for the current
2807  *      traffic pattern.  Constants in this function were computed
2808  *      based on theoretical maximum wire speed and thresholds were set based
2809  *      on testing data as well as attempting to minimize response time
2810  *      while increasing bulk throughput.
2811  *      this functionality is controlled by the InterruptThrottleRate module
2812  *      parameter (see igb_param.c)
2813  *      NOTE:  These calculations are only valid when operating in a single-
2814  *             queue environment.
2815  * @adapter: pointer to adapter
2816  * @itr_setting: current adapter->itr
2817  * @packets: the number of packets during this measurement interval
2818  * @bytes: the number of bytes during this measurement interval
2819  **/
2820 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2821                                    int packets, int bytes)
2822 {
2823         unsigned int retval = itr_setting;
2824
2825         if (packets == 0)
2826                 goto update_itr_done;
2827
2828         switch (itr_setting) {
2829         case lowest_latency:
2830                 /* handle TSO and jumbo frames */
2831                 if (bytes/packets > 8000)
2832                         retval = bulk_latency;
2833                 else if ((packets < 5) && (bytes > 512))
2834                         retval = low_latency;
2835                 break;
2836         case low_latency:  /* 50 usec aka 20000 ints/s */
2837                 if (bytes > 10000) {
2838                         /* this if handles the TSO accounting */
2839                         if (bytes/packets > 8000) {
2840                                 retval = bulk_latency;
2841                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2842                                 retval = bulk_latency;
2843                         } else if ((packets > 35)) {
2844                                 retval = lowest_latency;
2845                         }
2846                 } else if (bytes/packets > 2000) {
2847                         retval = bulk_latency;
2848                 } else if (packets <= 2 && bytes < 512) {
2849                         retval = lowest_latency;
2850                 }
2851                 break;
2852         case bulk_latency: /* 250 usec aka 4000 ints/s */
2853                 if (bytes > 25000) {
2854                         if (packets > 35)
2855                                 retval = low_latency;
2856                 } else if (bytes < 1500) {
2857                         retval = low_latency;
2858                 }
2859                 break;
2860         }
2861
2862 update_itr_done:
2863         return retval;
2864 }
2865
2866 static void igb_set_itr(struct igb_adapter *adapter)
2867 {
2868         u16 current_itr;
2869         u32 new_itr = adapter->itr;
2870
2871         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2872         if (adapter->link_speed != SPEED_1000) {
2873                 current_itr = 0;
2874                 new_itr = 4000;
2875                 goto set_itr_now;
2876         }
2877
2878         adapter->rx_itr = igb_update_itr(adapter,
2879                                     adapter->rx_itr,
2880                                     adapter->rx_ring->total_packets,
2881                                     adapter->rx_ring->total_bytes);
2882
2883         if (adapter->rx_ring->buddy) {
2884                 adapter->tx_itr = igb_update_itr(adapter,
2885                                             adapter->tx_itr,
2886                                             adapter->tx_ring->total_packets,
2887                                             adapter->tx_ring->total_bytes);
2888                 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2889         } else {
2890                 current_itr = adapter->rx_itr;
2891         }
2892
2893         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2894         if (adapter->itr_setting == 3 && current_itr == lowest_latency)
2895                 current_itr = low_latency;
2896
2897         switch (current_itr) {
2898         /* counts and packets in update_itr are dependent on these numbers */
2899         case lowest_latency:
2900                 new_itr = 70000;
2901                 break;
2902         case low_latency:
2903                 new_itr = 20000; /* aka hwitr = ~200 */
2904                 break;
2905         case bulk_latency:
2906                 new_itr = 4000;
2907                 break;
2908         default:
2909                 break;
2910         }
2911
2912 set_itr_now:
2913         adapter->rx_ring->total_bytes = 0;
2914         adapter->rx_ring->total_packets = 0;
2915         if (adapter->rx_ring->buddy) {
2916                 adapter->rx_ring->buddy->total_bytes = 0;
2917                 adapter->rx_ring->buddy->total_packets = 0;
2918         }
2919
2920         if (new_itr != adapter->itr) {
2921                 /* this attempts to bias the interrupt rate towards Bulk
2922                  * by adding intermediate steps when interrupt rate is
2923                  * increasing */
2924                 new_itr = new_itr > adapter->itr ?
2925                              min(adapter->itr + (new_itr >> 2), new_itr) :
2926                              new_itr;
2927                 /* Don't write the value here; it resets the adapter's
2928                  * internal timer, and causes us to delay far longer than
2929                  * we should between interrupts.  Instead, we write the ITR
2930                  * value at the beginning of the next interrupt so the timing
2931                  * ends up being correct.
2932                  */
2933                 adapter->itr = new_itr;
2934                 adapter->rx_ring->itr_val = 1000000000 / (new_itr * 256);
2935                 adapter->rx_ring->set_itr = 1;
2936         }
2937
2938         return;
2939 }
2940
2941
2942 #define IGB_TX_FLAGS_CSUM               0x00000001
2943 #define IGB_TX_FLAGS_VLAN               0x00000002
2944 #define IGB_TX_FLAGS_TSO                0x00000004
2945 #define IGB_TX_FLAGS_IPV4               0x00000008
2946 #define IGB_TX_FLAGS_TSTAMP             0x00000010
2947 #define IGB_TX_FLAGS_VLAN_MASK  0xffff0000
2948 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2949
2950 static inline int igb_tso_adv(struct igb_adapter *adapter,
2951                               struct igb_ring *tx_ring,
2952                               struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2953 {
2954         struct e1000_adv_tx_context_desc *context_desc;
2955         unsigned int i;
2956         int err;
2957         struct igb_buffer *buffer_info;
2958         u32 info = 0, tu_cmd = 0;
2959         u32 mss_l4len_idx, l4len;
2960         *hdr_len = 0;
2961
2962         if (skb_header_cloned(skb)) {
2963                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2964                 if (err)
2965                         return err;
2966         }
2967
2968         l4len = tcp_hdrlen(skb);
2969         *hdr_len += l4len;
2970
2971         if (skb->protocol == htons(ETH_P_IP)) {
2972                 struct iphdr *iph = ip_hdr(skb);
2973                 iph->tot_len = 0;
2974                 iph->check = 0;
2975                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2976                                                          iph->daddr, 0,
2977                                                          IPPROTO_TCP,
2978                                                          0);
2979         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2980                 ipv6_hdr(skb)->payload_len = 0;
2981                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2982                                                        &ipv6_hdr(skb)->daddr,
2983                                                        0, IPPROTO_TCP, 0);
2984         }
2985
2986         i = tx_ring->next_to_use;
2987
2988         buffer_info = &tx_ring->buffer_info[i];
2989         context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2990         /* VLAN MACLEN IPLEN */
2991         if (tx_flags & IGB_TX_FLAGS_VLAN)
2992                 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2993         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2994         *hdr_len += skb_network_offset(skb);
2995         info |= skb_network_header_len(skb);
2996         *hdr_len += skb_network_header_len(skb);
2997         context_desc->vlan_macip_lens = cpu_to_le32(info);
2998
2999         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3000         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3001
3002         if (skb->protocol == htons(ETH_P_IP))
3003                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3004         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3005
3006         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3007
3008         /* MSS L4LEN IDX */
3009         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
3010         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
3011
3012         /* For 82575, context index must be unique per ring. */
3013         if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
3014                 mss_l4len_idx |= tx_ring->queue_index << 4;
3015
3016         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3017         context_desc->seqnum_seed = 0;
3018
3019         buffer_info->time_stamp = jiffies;
3020         buffer_info->next_to_watch = i;
3021         buffer_info->dma = 0;
3022         i++;
3023         if (i == tx_ring->count)
3024                 i = 0;
3025
3026         tx_ring->next_to_use = i;
3027
3028         return true;
3029 }
3030
3031 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
3032                                         struct igb_ring *tx_ring,
3033                                         struct sk_buff *skb, u32 tx_flags)
3034 {
3035         struct e1000_adv_tx_context_desc *context_desc;
3036         unsigned int i;
3037         struct igb_buffer *buffer_info;
3038         u32 info = 0, tu_cmd = 0;
3039
3040         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
3041             (tx_flags & IGB_TX_FLAGS_VLAN)) {
3042                 i = tx_ring->next_to_use;
3043                 buffer_info = &tx_ring->buffer_info[i];
3044                 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3045
3046                 if (tx_flags & IGB_TX_FLAGS_VLAN)
3047                         info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3048                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3049                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3050                         info |= skb_network_header_len(skb);
3051
3052                 context_desc->vlan_macip_lens = cpu_to_le32(info);
3053
3054                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3055
3056                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3057                         __be16 protocol;
3058
3059                         if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3060                                 const struct vlan_ethhdr *vhdr =
3061                                           (const struct vlan_ethhdr*)skb->data;
3062
3063                                 protocol = vhdr->h_vlan_encapsulated_proto;
3064                         } else {
3065                                 protocol = skb->protocol;
3066                         }
3067
3068                         switch (protocol) {
3069                         case cpu_to_be16(ETH_P_IP):
3070                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3071                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3072                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3073                                 break;
3074                         case cpu_to_be16(ETH_P_IPV6):
3075                                 /* XXX what about other V6 headers?? */
3076                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3077                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3078                                 break;
3079                         default:
3080                                 if (unlikely(net_ratelimit()))
3081                                         dev_warn(&adapter->pdev->dev,
3082                                             "partial checksum but proto=%x!\n",
3083                                             skb->protocol);
3084                                 break;
3085                         }
3086                 }
3087
3088                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3089                 context_desc->seqnum_seed = 0;
3090                 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
3091                         context_desc->mss_l4len_idx =
3092                                 cpu_to_le32(tx_ring->queue_index << 4);
3093                 else
3094                         context_desc->mss_l4len_idx = 0;
3095
3096                 buffer_info->time_stamp = jiffies;
3097                 buffer_info->next_to_watch = i;
3098                 buffer_info->dma = 0;
3099
3100                 i++;
3101                 if (i == tx_ring->count)
3102                         i = 0;
3103                 tx_ring->next_to_use = i;
3104
3105                 return true;
3106         }
3107         return false;
3108 }
3109
3110 #define IGB_MAX_TXD_PWR 16
3111 #define IGB_MAX_DATA_PER_TXD    (1<<IGB_MAX_TXD_PWR)
3112
3113 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
3114                                  struct igb_ring *tx_ring, struct sk_buff *skb,
3115                                  unsigned int first)
3116 {
3117         struct igb_buffer *buffer_info;
3118         unsigned int len = skb_headlen(skb);
3119         unsigned int count = 0, i;
3120         unsigned int f;
3121         dma_addr_t *map;
3122
3123         i = tx_ring->next_to_use;
3124
3125         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
3126                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
3127                 return 0;
3128         }
3129
3130         map = skb_shinfo(skb)->dma_maps;
3131
3132         buffer_info = &tx_ring->buffer_info[i];
3133         BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3134         buffer_info->length = len;
3135         /* set time_stamp *before* dma to help avoid a possible race */
3136         buffer_info->time_stamp = jiffies;
3137         buffer_info->next_to_watch = i;
3138         buffer_info->dma = map[count];
3139         count++;
3140
3141         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
3142                 struct skb_frag_struct *frag;
3143
3144                 i++;
3145                 if (i == tx_ring->count)
3146                         i = 0;
3147
3148                 frag = &skb_shinfo(skb)->frags[f];
3149                 len = frag->size;
3150
3151                 buffer_info = &tx_ring->buffer_info[i];
3152                 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3153                 buffer_info->length = len;
3154                 buffer_info->time_stamp = jiffies;
3155                 buffer_info->next_to_watch = i;
3156                 buffer_info->dma = map[count];
3157                 count++;
3158         }
3159
3160         tx_ring->buffer_info[i].skb = skb;
3161         tx_ring->buffer_info[first].next_to_watch = i;
3162
3163         return count;
3164 }
3165
3166 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
3167                                     struct igb_ring *tx_ring,
3168                                     int tx_flags, int count, u32 paylen,
3169                                     u8 hdr_len)
3170 {
3171         union e1000_adv_tx_desc *tx_desc = NULL;
3172         struct igb_buffer *buffer_info;
3173         u32 olinfo_status = 0, cmd_type_len;
3174         unsigned int i;
3175
3176         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
3177                         E1000_ADVTXD_DCMD_DEXT);
3178
3179         if (tx_flags & IGB_TX_FLAGS_VLAN)
3180                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3181
3182         if (tx_flags & IGB_TX_FLAGS_TSTAMP)
3183                 cmd_type_len |= E1000_ADVTXD_MAC_TSTAMP;
3184
3185         if (tx_flags & IGB_TX_FLAGS_TSO) {
3186                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3187
3188                 /* insert tcp checksum */
3189                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3190
3191                 /* insert ip checksum */
3192                 if (tx_flags & IGB_TX_FLAGS_IPV4)
3193                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3194
3195         } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
3196                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3197         }
3198
3199         if ((adapter->flags & IGB_FLAG_NEED_CTX_IDX) &&
3200             (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
3201                          IGB_TX_FLAGS_VLAN)))
3202                 olinfo_status |= tx_ring->queue_index << 4;
3203
3204         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
3205
3206         i = tx_ring->next_to_use;
3207         while (count--) {
3208                 buffer_info = &tx_ring->buffer_info[i];
3209                 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
3210                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
3211                 tx_desc->read.cmd_type_len =
3212                         cpu_to_le32(cmd_type_len | buffer_info->length);
3213                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3214                 i++;
3215                 if (i == tx_ring->count)
3216                         i = 0;
3217         }
3218
3219         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
3220         /* Force memory writes to complete before letting h/w
3221          * know there are new descriptors to fetch.  (Only
3222          * applicable for weak-ordered memory model archs,
3223          * such as IA-64). */
3224         wmb();
3225
3226         tx_ring->next_to_use = i;
3227         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3228         /* we need this if more than one processor can write to our tail
3229          * at a time, it syncronizes IO on IA64/Altix systems */
3230         mmiowb();
3231 }
3232
3233 static int __igb_maybe_stop_tx(struct net_device *netdev,
3234                                struct igb_ring *tx_ring, int size)
3235 {
3236         struct igb_adapter *adapter = netdev_priv(netdev);
3237
3238         netif_stop_subqueue(netdev, tx_ring->queue_index);
3239
3240         /* Herbert's original patch had:
3241          *  smp_mb__after_netif_stop_queue();
3242          * but since that doesn't exist yet, just open code it. */
3243         smp_mb();
3244
3245         /* We need to check again in a case another CPU has just
3246          * made room available. */
3247         if (igb_desc_unused(tx_ring) < size)
3248                 return -EBUSY;
3249
3250         /* A reprieve! */
3251         netif_wake_subqueue(netdev, tx_ring->queue_index);
3252         ++adapter->restart_queue;
3253         return 0;
3254 }
3255
3256 static int igb_maybe_stop_tx(struct net_device *netdev,
3257                              struct igb_ring *tx_ring, int size)
3258 {
3259         if (igb_desc_unused(tx_ring) >= size)
3260                 return 0;
3261         return __igb_maybe_stop_tx(netdev, tx_ring, size);
3262 }
3263
3264 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
3265                                    struct net_device *netdev,
3266                                    struct igb_ring *tx_ring)
3267 {
3268         struct igb_adapter *adapter = netdev_priv(netdev);
3269         unsigned int first;
3270         unsigned int tx_flags = 0;
3271         u8 hdr_len = 0;
3272         int count = 0;
3273         int tso = 0;
3274         union skb_shared_tx *shtx;
3275
3276         if (test_bit(__IGB_DOWN, &adapter->state)) {
3277                 dev_kfree_skb_any(skb);
3278                 return NETDEV_TX_OK;
3279         }
3280
3281         if (skb->len <= 0) {
3282                 dev_kfree_skb_any(skb);
3283                 return NETDEV_TX_OK;
3284         }
3285
3286         /* need: 1 descriptor per page,
3287          *       + 2 desc gap to keep tail from touching head,
3288          *       + 1 desc for skb->data,
3289          *       + 1 desc for context descriptor,
3290          * otherwise try next time */
3291         if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
3292                 /* this is a hard error */
3293                 return NETDEV_TX_BUSY;
3294         }
3295
3296         /*
3297          * TODO: check that there currently is no other packet with
3298          * time stamping in the queue
3299          *
3300          * When doing time stamping, keep the connection to the socket
3301          * a while longer: it is still needed by skb_hwtstamp_tx(),
3302          * called either in igb_tx_hwtstamp() or by our caller when
3303          * doing software time stamping.
3304          */
3305         shtx = skb_tx(skb);
3306         if (unlikely(shtx->hardware)) {
3307                 shtx->in_progress = 1;
3308                 tx_flags |= IGB_TX_FLAGS_TSTAMP;
3309         }
3310
3311         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3312                 tx_flags |= IGB_TX_FLAGS_VLAN;
3313                 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
3314         }
3315
3316         if (skb->protocol == htons(ETH_P_IP))
3317                 tx_flags |= IGB_TX_FLAGS_IPV4;
3318
3319         first = tx_ring->next_to_use;
3320         tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
3321                                               &hdr_len) : 0;
3322
3323         if (tso < 0) {
3324                 dev_kfree_skb_any(skb);
3325                 return NETDEV_TX_OK;
3326         }
3327
3328         if (tso)
3329                 tx_flags |= IGB_TX_FLAGS_TSO;
3330         else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags) &&
3331                  (skb->ip_summed == CHECKSUM_PARTIAL))
3332                 tx_flags |= IGB_TX_FLAGS_CSUM;
3333
3334         /*
3335          * count reflects descriptors mapped, if 0 then mapping error
3336          * has occured and we need to rewind the descriptor queue
3337          */
3338         count = igb_tx_map_adv(adapter, tx_ring, skb, first);
3339
3340         if (count) {
3341                 igb_tx_queue_adv(adapter, tx_ring, tx_flags, count,
3342                                  skb->len, hdr_len);
3343                 netdev->trans_start = jiffies;
3344                 /* Make sure there is space in the ring for the next send. */
3345                 igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
3346         } else {
3347                 dev_kfree_skb_any(skb);
3348                 tx_ring->buffer_info[first].time_stamp = 0;
3349                 tx_ring->next_to_use = first;
3350         }
3351
3352         return NETDEV_TX_OK;
3353 }
3354
3355 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
3356 {
3357         struct igb_adapter *adapter = netdev_priv(netdev);
3358         struct igb_ring *tx_ring;
3359
3360         int r_idx = 0;
3361         r_idx = skb->queue_mapping & (IGB_ABS_MAX_TX_QUEUES - 1);
3362         tx_ring = adapter->multi_tx_table[r_idx];
3363
3364         /* This goes back to the question of how to logically map a tx queue
3365          * to a flow.  Right now, performance is impacted slightly negatively
3366          * if using multiple tx queues.  If the stack breaks away from a
3367          * single qdisc implementation, we can look at this again. */
3368         return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
3369 }
3370
3371 /**
3372  * igb_tx_timeout - Respond to a Tx Hang
3373  * @netdev: network interface device structure
3374  **/
3375 static void igb_tx_timeout(struct net_device *netdev)
3376 {
3377         struct igb_adapter *adapter = netdev_priv(netdev);
3378         struct e1000_hw *hw = &adapter->hw;
3379
3380         /* Do the reset outside of interrupt context */
3381         adapter->tx_timeout_count++;
3382         schedule_work(&adapter->reset_task);
3383         wr32(E1000_EICS,
3384              (adapter->eims_enable_mask & ~adapter->eims_other));
3385 }
3386
3387 static void igb_reset_task(struct work_struct *work)
3388 {
3389         struct igb_adapter *adapter;
3390         adapter = container_of(work, struct igb_adapter, reset_task);
3391
3392         igb_reinit_locked(adapter);
3393 }
3394
3395 /**
3396  * igb_get_stats - Get System Network Statistics
3397  * @netdev: network interface device structure
3398  *
3399  * Returns the address of the device statistics structure.
3400  * The statistics are actually updated from the timer callback.
3401  **/
3402 static struct net_device_stats *igb_get_stats(struct net_device *netdev)
3403 {
3404         struct igb_adapter *adapter = netdev_priv(netdev);
3405
3406         /* only return the current stats */
3407         return &adapter->net_stats;
3408 }
3409
3410 /**
3411  * igb_change_mtu - Change the Maximum Transfer Unit
3412  * @netdev: network interface device structure
3413  * @new_mtu: new value for maximum frame size
3414  *
3415  * Returns 0 on success, negative on failure
3416  **/
3417 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3418 {
3419         struct igb_adapter *adapter = netdev_priv(netdev);
3420         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3421
3422         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3423             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3424                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
3425                 return -EINVAL;
3426         }
3427
3428         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3429                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
3430                 return -EINVAL;
3431         }
3432
3433         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3434                 msleep(1);
3435
3436         /* igb_down has a dependency on max_frame_size */
3437         adapter->max_frame_size = max_frame;
3438         if (netif_running(netdev))
3439                 igb_down(adapter);
3440
3441         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3442          * means we reserve 2 more, this pushes us to allocate from the next
3443          * larger slab size.
3444          * i.e. RXBUFFER_2048 --> size-4096 slab
3445          */
3446
3447         if (max_frame <= IGB_RXBUFFER_256)
3448                 adapter->rx_buffer_len = IGB_RXBUFFER_256;
3449         else if (max_frame <= IGB_RXBUFFER_512)
3450                 adapter->rx_buffer_len = IGB_RXBUFFER_512;
3451         else if (max_frame <= IGB_RXBUFFER_1024)
3452                 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3453         else if (max_frame <= IGB_RXBUFFER_2048)
3454                 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
3455         else
3456 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3457                 adapter->rx_buffer_len = IGB_RXBUFFER_16384;
3458 #else
3459                 adapter->rx_buffer_len = PAGE_SIZE / 2;
3460 #endif
3461
3462         /* if sr-iov is enabled we need to force buffer size to 1K or larger */
3463         if (adapter->vfs_allocated_count &&
3464             (adapter->rx_buffer_len < IGB_RXBUFFER_1024))
3465                 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3466
3467         /* adjust allocation if LPE protects us, and we aren't using SBP */
3468         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3469              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
3470                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3471
3472         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
3473                  netdev->mtu, new_mtu);
3474         netdev->mtu = new_mtu;
3475
3476         if (netif_running(netdev))
3477                 igb_up(adapter);
3478         else
3479                 igb_reset(adapter);
3480
3481         clear_bit(__IGB_RESETTING, &adapter->state);
3482
3483         return 0;
3484 }
3485
3486 /**
3487  * igb_update_stats - Update the board statistics counters
3488  * @adapter: board private structure
3489  **/
3490
3491 void igb_update_stats(struct igb_adapter *adapter)
3492 {
3493         struct e1000_hw *hw = &adapter->hw;
3494         struct pci_dev *pdev = adapter->pdev;
3495         u16 phy_tmp;
3496
3497 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3498
3499         /*
3500          * Prevent stats update while adapter is being reset, or if the pci
3501          * connection is down.
3502          */
3503         if (adapter->link_speed == 0)
3504                 return;
3505         if (pci_channel_offline(pdev))
3506                 return;
3507
3508         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
3509         adapter->stats.gprc += rd32(E1000_GPRC);
3510         adapter->stats.gorc += rd32(E1000_GORCL);
3511         rd32(E1000_GORCH); /* clear GORCL */
3512         adapter->stats.bprc += rd32(E1000_BPRC);
3513         adapter->stats.mprc += rd32(E1000_MPRC);
3514         adapter->stats.roc += rd32(E1000_ROC);
3515
3516         adapter->stats.prc64 += rd32(E1000_PRC64);
3517         adapter->stats.prc127 += rd32(E1000_PRC127);
3518         adapter->stats.prc255 += rd32(E1000_PRC255);
3519         adapter->stats.prc511 += rd32(E1000_PRC511);
3520         adapter->stats.prc1023 += rd32(E1000_PRC1023);
3521         adapter->stats.prc1522 += rd32(E1000_PRC1522);
3522         adapter->stats.symerrs += rd32(E1000_SYMERRS);
3523         adapter->stats.sec += rd32(E1000_SEC);
3524
3525         adapter->stats.mpc += rd32(E1000_MPC);
3526         adapter->stats.scc += rd32(E1000_SCC);
3527         adapter->stats.ecol += rd32(E1000_ECOL);
3528         adapter->stats.mcc += rd32(E1000_MCC);
3529         adapter->stats.latecol += rd32(E1000_LATECOL);
3530         adapter->stats.dc += rd32(E1000_DC);
3531         adapter->stats.rlec += rd32(E1000_RLEC);
3532         adapter->stats.xonrxc += rd32(E1000_XONRXC);
3533         adapter->stats.xontxc += rd32(E1000_XONTXC);
3534         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
3535         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
3536         adapter->stats.fcruc += rd32(E1000_FCRUC);
3537         adapter->stats.gptc += rd32(E1000_GPTC);
3538         adapter->stats.gotc += rd32(E1000_GOTCL);
3539         rd32(E1000_GOTCH); /* clear GOTCL */
3540         adapter->stats.rnbc += rd32(E1000_RNBC);
3541         adapter->stats.ruc += rd32(E1000_RUC);
3542         adapter->stats.rfc += rd32(E1000_RFC);
3543         adapter->stats.rjc += rd32(E1000_RJC);
3544         adapter->stats.tor += rd32(E1000_TORH);
3545         adapter->stats.tot += rd32(E1000_TOTH);
3546         adapter->stats.tpr += rd32(E1000_TPR);
3547
3548         adapter->stats.ptc64 += rd32(E1000_PTC64);
3549         adapter->stats.ptc127 += rd32(E1000_PTC127);
3550         adapter->stats.ptc255 += rd32(E1000_PTC255);
3551         adapter->stats.ptc511 += rd32(E1000_PTC511);
3552         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
3553         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
3554
3555         adapter->stats.mptc += rd32(E1000_MPTC);
3556         adapter->stats.bptc += rd32(E1000_BPTC);
3557
3558         /* used for adaptive IFS */
3559
3560         hw->mac.tx_packet_delta = rd32(E1000_TPT);
3561         adapter->stats.tpt += hw->mac.tx_packet_delta;
3562         hw->mac.collision_delta = rd32(E1000_COLC);
3563         adapter->stats.colc += hw->mac.collision_delta;
3564
3565         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
3566         adapter->stats.rxerrc += rd32(E1000_RXERRC);
3567         adapter->stats.tncrs += rd32(E1000_TNCRS);
3568         adapter->stats.tsctc += rd32(E1000_TSCTC);
3569         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
3570
3571         adapter->stats.iac += rd32(E1000_IAC);
3572         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
3573         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
3574         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
3575         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
3576         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
3577         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
3578         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
3579         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
3580
3581         /* Fill out the OS statistics structure */
3582         adapter->net_stats.multicast = adapter->stats.mprc;
3583         adapter->net_stats.collisions = adapter->stats.colc;
3584
3585         /* Rx Errors */
3586
3587         /* RLEC on some newer hardware can be incorrect so build
3588         * our own version based on RUC and ROC */
3589         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3590                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3591                 adapter->stats.ruc + adapter->stats.roc +
3592                 adapter->stats.cexterr;
3593         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3594                                               adapter->stats.roc;
3595         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3596         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3597         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3598
3599         /* Tx Errors */
3600         adapter->net_stats.tx_errors = adapter->stats.ecol +
3601                                        adapter->stats.latecol;
3602         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3603         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3604         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3605
3606         /* Tx Dropped needs to be maintained elsewhere */
3607
3608         /* Phy Stats */
3609         if (hw->phy.media_type == e1000_media_type_copper) {
3610                 if ((adapter->link_speed == SPEED_1000) &&
3611                    (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3612                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3613                         adapter->phy_stats.idle_errors += phy_tmp;
3614                 }
3615         }
3616
3617         /* Management Stats */
3618         adapter->stats.mgptc += rd32(E1000_MGTPTC);
3619         adapter->stats.mgprc += rd32(E1000_MGTPRC);
3620         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3621 }
3622
3623 static irqreturn_t igb_msix_other(int irq, void *data)
3624 {
3625         struct net_device *netdev = data;
3626         struct igb_adapter *adapter = netdev_priv(netdev);
3627         struct e1000_hw *hw = &adapter->hw;
3628         u32 icr = rd32(E1000_ICR);
3629
3630         /* reading ICR causes bit 31 of EICR to be cleared */
3631
3632         if(icr & E1000_ICR_DOUTSYNC) {
3633                 /* HW is reporting DMA is out of sync */
3634                 adapter->stats.doosync++;
3635         }
3636
3637         /* Check for a mailbox event */
3638         if (icr & E1000_ICR_VMMB)
3639                 igb_msg_task(adapter);
3640
3641         if (icr & E1000_ICR_LSC) {
3642                 hw->mac.get_link_status = 1;
3643                 /* guard against interrupt when we're going down */
3644                 if (!test_bit(__IGB_DOWN, &adapter->state))
3645                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3646         }
3647
3648         wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_VMMB);
3649         wr32(E1000_EIMS, adapter->eims_other);
3650
3651         return IRQ_HANDLED;
3652 }
3653
3654 static irqreturn_t igb_msix_tx(int irq, void *data)
3655 {
3656         struct igb_ring *tx_ring = data;
3657         struct igb_adapter *adapter = tx_ring->adapter;
3658         struct e1000_hw *hw = &adapter->hw;
3659
3660 #ifdef CONFIG_IGB_DCA
3661         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3662                 igb_update_tx_dca(tx_ring);
3663 #endif
3664
3665         tx_ring->total_bytes = 0;
3666         tx_ring->total_packets = 0;
3667
3668         /* auto mask will automatically reenable the interrupt when we write
3669          * EICS */
3670         if (!igb_clean_tx_irq(tx_ring))
3671                 /* Ring was not completely cleaned, so fire another interrupt */
3672                 wr32(E1000_EICS, tx_ring->eims_value);
3673         else
3674                 wr32(E1000_EIMS, tx_ring->eims_value);
3675
3676         return IRQ_HANDLED;
3677 }
3678
3679 static void igb_write_itr(struct igb_ring *ring)
3680 {
3681         struct e1000_hw *hw = &ring->adapter->hw;
3682         if ((ring->adapter->itr_setting & 3) && ring->set_itr) {
3683                 switch (hw->mac.type) {
3684                 case e1000_82576:
3685                         wr32(ring->itr_register, ring->itr_val |
3686                              0x80000000);
3687                         break;
3688                 default:
3689                         wr32(ring->itr_register, ring->itr_val |
3690                              (ring->itr_val << 16));
3691                         break;
3692                 }
3693                 ring->set_itr = 0;
3694         }
3695 }
3696
3697 static irqreturn_t igb_msix_rx(int irq, void *data)
3698 {
3699         struct igb_ring *rx_ring = data;
3700
3701         /* Write the ITR value calculated at the end of the
3702          * previous interrupt.
3703          */
3704
3705         igb_write_itr(rx_ring);
3706
3707         if (napi_schedule_prep(&rx_ring->napi))
3708                 __napi_schedule(&rx_ring->napi);
3709
3710 #ifdef CONFIG_IGB_DCA
3711         if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
3712                 igb_update_rx_dca(rx_ring);
3713 #endif
3714                 return IRQ_HANDLED;
3715 }
3716
3717 #ifdef CONFIG_IGB_DCA
3718 static void igb_update_rx_dca(struct igb_ring *rx_ring)
3719 {
3720         u32 dca_rxctrl;
3721         struct igb_adapter *adapter = rx_ring->adapter;
3722         struct e1000_hw *hw = &adapter->hw;
3723         int cpu = get_cpu();
3724         int q = rx_ring->reg_idx;
3725
3726         if (rx_ring->cpu != cpu) {
3727                 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
3728                 if (hw->mac.type == e1000_82576) {
3729                         dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
3730                         dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
3731                                       E1000_DCA_RXCTRL_CPUID_SHIFT;
3732                 } else {
3733                         dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
3734                         dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
3735                 }
3736                 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
3737                 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
3738                 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
3739                 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
3740                 rx_ring->cpu = cpu;
3741         }
3742         put_cpu();
3743 }
3744
3745 static void igb_update_tx_dca(struct igb_ring *tx_ring)
3746 {
3747         u32 dca_txctrl;
3748         struct igb_adapter *adapter = tx_ring->adapter;
3749         struct e1000_hw *hw = &adapter->hw;
3750         int cpu = get_cpu();
3751         int q = tx_ring->reg_idx;
3752
3753         if (tx_ring->cpu != cpu) {
3754                 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
3755                 if (hw->mac.type == e1000_82576) {
3756                         dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
3757                         dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
3758                                       E1000_DCA_TXCTRL_CPUID_SHIFT;
3759                 } else {
3760                         dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
3761                         dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
3762                 }
3763                 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
3764                 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
3765                 tx_ring->cpu = cpu;
3766         }
3767         put_cpu();
3768 }
3769
3770 static void igb_setup_dca(struct igb_adapter *adapter)
3771 {
3772         int i;
3773
3774         if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
3775                 return;
3776
3777         for (i = 0; i < adapter->num_tx_queues; i++) {
3778                 adapter->tx_ring[i].cpu = -1;
3779                 igb_update_tx_dca(&adapter->tx_ring[i]);
3780         }
3781         for (i = 0; i < adapter->num_rx_queues; i++) {
3782                 adapter->rx_ring[i].cpu = -1;
3783                 igb_update_rx_dca(&adapter->rx_ring[i]);
3784         }
3785 }
3786
3787 static int __igb_notify_dca(struct device *dev, void *data)
3788 {
3789         struct net_device *netdev = dev_get_drvdata(dev);
3790         struct igb_adapter *adapter = netdev_priv(netdev);
3791         struct e1000_hw *hw = &adapter->hw;
3792         unsigned long event = *(unsigned long *)data;
3793
3794         switch (event) {
3795         case DCA_PROVIDER_ADD:
3796                 /* if already enabled, don't do it again */
3797                 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3798                         break;
3799                 /* Always use CB2 mode, difference is masked
3800                  * in the CB driver. */
3801                 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
3802                 if (dca_add_requester(dev) == 0) {
3803                         adapter->flags |= IGB_FLAG_DCA_ENABLED;
3804                         dev_info(&adapter->pdev->dev, "DCA enabled\n");
3805                         igb_setup_dca(adapter);
3806                         break;
3807                 }
3808                 /* Fall Through since DCA is disabled. */
3809         case DCA_PROVIDER_REMOVE:
3810                 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3811                         /* without this a class_device is left
3812                          * hanging around in the sysfs model */
3813                         dca_remove_requester(dev);
3814                         dev_info(&adapter->pdev->dev, "DCA disabled\n");
3815                         adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3816                         wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3817                 }
3818                 break;
3819         }
3820
3821         return 0;
3822 }
3823
3824 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
3825                           void *p)
3826 {
3827         int ret_val;
3828
3829         ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
3830                                          __igb_notify_dca);
3831
3832         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
3833 }
3834 #endif /* CONFIG_IGB_DCA */
3835
3836 static void igb_ping_all_vfs(struct igb_adapter *adapter)
3837 {
3838         struct e1000_hw *hw = &adapter->hw;
3839         u32 ping;
3840         int i;
3841
3842         for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
3843                 ping = E1000_PF_CONTROL_MSG;
3844                 if (adapter->vf_data[i].clear_to_send)
3845                         ping |= E1000_VT_MSGTYPE_CTS;
3846                 igb_write_mbx(hw, &ping, 1, i);
3847         }
3848 }
3849
3850 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
3851                                   u32 *msgbuf, u32 vf)
3852 {
3853         int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3854         u16 *hash_list = (u16 *)&msgbuf[1];
3855         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
3856         int i;
3857
3858         /* only up to 30 hash values supported */
3859         if (n > 30)
3860                 n = 30;
3861
3862         /* salt away the number of multi cast addresses assigned
3863          * to this VF for later use to restore when the PF multi cast
3864          * list changes
3865          */
3866         vf_data->num_vf_mc_hashes = n;
3867
3868         /* VFs are limited to using the MTA hash table for their multicast
3869          * addresses */
3870         for (i = 0; i < n; i++)
3871                 vf_data->vf_mc_hashes[i] = hash_list[i];;
3872
3873         /* Flush and reset the mta with the new values */
3874         igb_set_multi(adapter->netdev);
3875
3876         return 0;
3877 }
3878
3879 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
3880 {
3881         struct e1000_hw *hw = &adapter->hw;
3882         struct vf_data_storage *vf_data;
3883         int i, j;
3884
3885         for (i = 0; i < adapter->vfs_allocated_count; i++) {
3886                 vf_data = &adapter->vf_data[i];
3887                 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
3888                         igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
3889         }
3890 }
3891
3892 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
3893 {
3894         struct e1000_hw *hw = &adapter->hw;
3895         u32 pool_mask, reg, vid;
3896         int i;
3897
3898         pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3899
3900         /* Find the vlan filter for this id */
3901         for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3902                 reg = rd32(E1000_VLVF(i));
3903
3904                 /* remove the vf from the pool */
3905                 reg &= ~pool_mask;
3906
3907                 /* if pool is empty then remove entry from vfta */
3908                 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
3909                     (reg & E1000_VLVF_VLANID_ENABLE)) {
3910                         reg = 0;
3911                         vid = reg & E1000_VLVF_VLANID_MASK;
3912                         igb_vfta_set(hw, vid, false);
3913                 }
3914
3915                 wr32(E1000_VLVF(i), reg);
3916         }
3917 }
3918
3919 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
3920 {
3921         struct e1000_hw *hw = &adapter->hw;
3922         u32 reg, i;
3923
3924         /* It is an error to call this function when VFs are not enabled */
3925         if (!adapter->vfs_allocated_count)
3926                 return -1;
3927
3928         /* Find the vlan filter for this id */
3929         for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3930                 reg = rd32(E1000_VLVF(i));
3931                 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
3932                     vid == (reg & E1000_VLVF_VLANID_MASK))
3933                         break;
3934         }
3935
3936         if (add) {
3937                 if (i == E1000_VLVF_ARRAY_SIZE) {
3938                         /* Did not find a matching VLAN ID entry that was
3939                          * enabled.  Search for a free filter entry, i.e.
3940                          * one without the enable bit set
3941                          */
3942                         for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3943                                 reg = rd32(E1000_VLVF(i));
3944                                 if (!(reg & E1000_VLVF_VLANID_ENABLE))
3945                                         break;
3946                         }
3947                 }
3948                 if (i < E1000_VLVF_ARRAY_SIZE) {
3949                         /* Found an enabled/available entry */
3950                         reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3951
3952                         /* if !enabled we need to set this up in vfta */
3953                         if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
3954                                 /* add VID to filter table, if bit already set
3955                                  * PF must have added it outside of table */
3956                                 if (igb_vfta_set(hw, vid, true))
3957                                         reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT +
3958                                                 adapter->vfs_allocated_count);
3959                                 reg |= E1000_VLVF_VLANID_ENABLE;
3960                         }
3961                         reg &= ~E1000_VLVF_VLANID_MASK;
3962                         reg |= vid;
3963
3964                         wr32(E1000_VLVF(i), reg);
3965                         return 0;
3966                 }
3967         } else {
3968                 if (i < E1000_VLVF_ARRAY_SIZE) {
3969                         /* remove vf from the pool */
3970                         reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
3971                         /* if pool is empty then remove entry from vfta */
3972                         if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
3973                                 reg = 0;
3974                                 igb_vfta_set(hw, vid, false);
3975                         }
3976                         wr32(E1000_VLVF(i), reg);
3977                         return 0;
3978                 }
3979         }
3980         return -1;
3981 }
3982
3983 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
3984 {
3985         int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3986         int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
3987
3988         return igb_vlvf_set(adapter, vid, add, vf);
3989 }
3990
3991 static inline void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
3992 {
3993         struct e1000_hw *hw = &adapter->hw;
3994
3995         /* disable mailbox functionality for vf */
3996         adapter->vf_data[vf].clear_to_send = false;
3997
3998         /* reset offloads to defaults */
3999         igb_set_vmolr(hw, vf);
4000
4001         /* reset vlans for device */
4002         igb_clear_vf_vfta(adapter, vf);
4003
4004         /* reset multicast table array for vf */
4005         adapter->vf_data[vf].num_vf_mc_hashes = 0;
4006
4007         /* Flush and reset the mta with the new values */
4008         igb_set_multi(adapter->netdev);
4009 }
4010
4011 static inline void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
4012 {
4013         struct e1000_hw *hw = &adapter->hw;
4014         unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4015         u32 reg, msgbuf[3];
4016         u8 *addr = (u8 *)(&msgbuf[1]);
4017
4018         /* process all the same items cleared in a function level reset */
4019         igb_vf_reset_event(adapter, vf);
4020
4021         /* set vf mac address */
4022         igb_rar_set(hw, vf_mac, vf + 1);
4023         igb_set_rah_pool(hw, vf, vf + 1);
4024
4025         /* enable transmit and receive for vf */
4026         reg = rd32(E1000_VFTE);
4027         wr32(E1000_VFTE, reg | (1 << vf));
4028         reg = rd32(E1000_VFRE);
4029         wr32(E1000_VFRE, reg | (1 << vf));
4030
4031         /* enable mailbox functionality for vf */
4032         adapter->vf_data[vf].clear_to_send = true;
4033
4034         /* reply to reset with ack and vf mac address */
4035         msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
4036         memcpy(addr, vf_mac, 6);
4037         igb_write_mbx(hw, msgbuf, 3, vf);
4038 }
4039
4040 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
4041 {
4042                 unsigned char *addr = (char *)&msg[1];
4043                 int err = -1;
4044
4045                 if (is_valid_ether_addr(addr))
4046                         err = igb_set_vf_mac(adapter, vf, addr);
4047
4048                 return err;
4049
4050 }
4051
4052 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
4053 {
4054         struct e1000_hw *hw = &adapter->hw;
4055         u32 msg = E1000_VT_MSGTYPE_NACK;
4056
4057         /* if device isn't clear to send it shouldn't be reading either */
4058         if (!adapter->vf_data[vf].clear_to_send)
4059                 igb_write_mbx(hw, &msg, 1, vf);
4060 }
4061
4062
4063 static void igb_msg_task(struct igb_adapter *adapter)
4064 {
4065         struct e1000_hw *hw = &adapter->hw;
4066         u32 vf;
4067
4068         for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
4069                 /* process any reset requests */
4070                 if (!igb_check_for_rst(hw, vf)) {
4071                         adapter->vf_data[vf].clear_to_send = false;
4072                         igb_vf_reset_event(adapter, vf);
4073                 }
4074
4075                 /* process any messages pending */
4076                 if (!igb_check_for_msg(hw, vf))
4077                         igb_rcv_msg_from_vf(adapter, vf);
4078
4079                 /* process any acks */
4080                 if (!igb_check_for_ack(hw, vf))
4081                         igb_rcv_ack_from_vf(adapter, vf);
4082
4083         }
4084 }
4085
4086 static int igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
4087 {
4088         u32 mbx_size = E1000_VFMAILBOX_SIZE;
4089         u32 msgbuf[mbx_size];
4090         struct e1000_hw *hw = &adapter->hw;
4091         s32 retval;
4092
4093         retval = igb_read_mbx(hw, msgbuf, mbx_size, vf);
4094
4095         if (retval)
4096                 dev_err(&adapter->pdev->dev,
4097                         "Error receiving message from VF\n");
4098
4099         /* this is a message we already processed, do nothing */
4100         if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
4101                 return retval;
4102
4103         /*
4104          * until the vf completes a reset it should not be
4105          * allowed to start any configuration.
4106          */
4107
4108         if (msgbuf[0] == E1000_VF_RESET) {
4109                 igb_vf_reset_msg(adapter, vf);
4110
4111                 return retval;
4112         }
4113
4114         if (!adapter->vf_data[vf].clear_to_send) {
4115                 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4116                 igb_write_mbx(hw, msgbuf, 1, vf);
4117                 return retval;
4118         }
4119
4120         switch ((msgbuf[0] & 0xFFFF)) {
4121         case E1000_VF_SET_MAC_ADDR:
4122                 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
4123                 break;
4124         case E1000_VF_SET_MULTICAST:
4125                 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
4126                 break;
4127         case E1000_VF_SET_LPE:
4128                 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
4129                 break;
4130         case E1000_VF_SET_VLAN:
4131                 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
4132                 break;
4133         default:
4134                 dev_err(&adapter->pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
4135                 retval = -1;
4136                 break;
4137         }
4138
4139         /* notify the VF of the results of what it sent us */
4140         if (retval)
4141                 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4142         else
4143                 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
4144
4145         msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
4146
4147         igb_write_mbx(hw, msgbuf, 1, vf);
4148
4149         return retval;
4150 }
4151
4152 /**
4153  * igb_intr_msi - Interrupt Handler
4154  * @irq: interrupt number
4155  * @data: pointer to a network interface device structure
4156  **/
4157 static irqreturn_t igb_intr_msi(int irq, void *data)
4158 {
4159         struct net_device *netdev = data;
4160         struct igb_adapter *adapter = netdev_priv(netdev);
4161         struct e1000_hw *hw = &adapter->hw;
4162         /* read ICR disables interrupts using IAM */
4163         u32 icr = rd32(E1000_ICR);
4164
4165         igb_write_itr(adapter->rx_ring);
4166
4167         if(icr & E1000_ICR_DOUTSYNC) {
4168                 /* HW is reporting DMA is out of sync */
4169                 adapter->stats.doosync++;
4170         }
4171
4172         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4173                 hw->mac.get_link_status = 1;
4174                 if (!test_bit(__IGB_DOWN, &adapter->state))
4175                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
4176         }
4177
4178         napi_schedule(&adapter->rx_ring[0].napi);
4179
4180         return IRQ_HANDLED;
4181 }
4182
4183 /**
4184  * igb_intr - Legacy Interrupt Handler
4185  * @irq: interrupt number
4186  * @data: pointer to a network interface device structure
4187  **/
4188 static irqreturn_t igb_intr(int irq, void *data)
4189 {
4190         struct net_device *netdev = data;
4191         struct igb_adapter *adapter = netdev_priv(netdev);
4192         struct e1000_hw *hw = &adapter->hw;
4193         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
4194          * need for the IMC write */
4195         u32 icr = rd32(E1000_ICR);
4196         if (!icr)
4197                 return IRQ_NONE;  /* Not our interrupt */
4198
4199         igb_write_itr(adapter->rx_ring);
4200
4201         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
4202          * not set, then the adapter didn't send an interrupt */
4203         if (!(icr & E1000_ICR_INT_ASSERTED))
4204                 return IRQ_NONE;
4205
4206         if(icr & E1000_ICR_DOUTSYNC) {
4207                 /* HW is reporting DMA is out of sync */
4208                 adapter->stats.doosync++;
4209         }
4210
4211         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4212                 hw->mac.get_link_status = 1;
4213                 /* guard against interrupt when we're going down */
4214                 if (!test_bit(__IGB_DOWN, &adapter->state))
4215                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
4216         }
4217
4218         napi_schedule(&adapter->rx_ring[0].napi);
4219
4220         return IRQ_HANDLED;
4221 }
4222
4223 static inline void igb_rx_irq_enable(struct igb_ring *rx_ring)
4224 {
4225         struct igb_adapter *adapter = rx_ring->adapter;
4226         struct e1000_hw *hw = &adapter->hw;
4227
4228         if (adapter->itr_setting & 3) {
4229                 if (adapter->num_rx_queues == 1)
4230                         igb_set_itr(adapter);
4231                 else
4232                         igb_update_ring_itr(rx_ring);
4233         }
4234
4235         if (!test_bit(__IGB_DOWN, &adapter->state)) {
4236                 if (adapter->msix_entries)
4237                         wr32(E1000_EIMS, rx_ring->eims_value);
4238                 else
4239                         igb_irq_enable(adapter);
4240         }
4241 }
4242
4243 /**
4244  * igb_poll - NAPI Rx polling callback
4245  * @napi: napi polling structure
4246  * @budget: count of how many packets we should handle
4247  **/
4248 static int igb_poll(struct napi_struct *napi, int budget)
4249 {
4250         struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
4251         int work_done = 0;
4252
4253 #ifdef CONFIG_IGB_DCA
4254         if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
4255                 igb_update_rx_dca(rx_ring);
4256 #endif
4257         igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
4258
4259         if (rx_ring->buddy) {
4260 #ifdef CONFIG_IGB_DCA
4261                 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
4262                         igb_update_tx_dca(rx_ring->buddy);
4263 #endif
4264                 if (!igb_clean_tx_irq(rx_ring->buddy))
4265                         work_done = budget;
4266         }
4267
4268         /* If not enough Rx work done, exit the polling mode */
4269         if (work_done < budget) {
4270                 napi_complete(napi);
4271                 igb_rx_irq_enable(rx_ring);
4272         }
4273
4274         return work_done;
4275 }
4276
4277 /**
4278  * igb_hwtstamp - utility function which checks for TX time stamp
4279  * @adapter: board private structure
4280  * @skb: packet that was just sent
4281  *
4282  * If we were asked to do hardware stamping and such a time stamp is
4283  * available, then it must have been for this skb here because we only
4284  * allow only one such packet into the queue.
4285  */
4286 static void igb_tx_hwtstamp(struct igb_adapter *adapter, struct sk_buff *skb)
4287 {
4288         union skb_shared_tx *shtx = skb_tx(skb);
4289         struct e1000_hw *hw = &adapter->hw;
4290
4291         if (unlikely(shtx->hardware)) {
4292                 u32 valid = rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID;
4293                 if (valid) {
4294                         u64 regval = rd32(E1000_TXSTMPL);
4295                         u64 ns;
4296                         struct skb_shared_hwtstamps shhwtstamps;
4297
4298                         memset(&shhwtstamps, 0, sizeof(shhwtstamps));
4299                         regval |= (u64)rd32(E1000_TXSTMPH) << 32;
4300                         ns = timecounter_cyc2time(&adapter->clock,
4301                                                   regval);
4302                         timecompare_update(&adapter->compare, ns);
4303                         shhwtstamps.hwtstamp = ns_to_ktime(ns);
4304                         shhwtstamps.syststamp =
4305                                 timecompare_transform(&adapter->compare, ns);
4306                         skb_tstamp_tx(skb, &shhwtstamps);
4307                 }
4308         }
4309 }
4310
4311 /**
4312  * igb_clean_tx_irq - Reclaim resources after transmit completes
4313  * @adapter: board private structure
4314  * returns true if ring is completely cleaned
4315  **/
4316 static bool igb_clean_tx_irq(struct igb_ring *tx_ring)
4317 {
4318         struct igb_adapter *adapter = tx_ring->adapter;
4319         struct net_device *netdev = adapter->netdev;
4320         struct e1000_hw *hw = &adapter->hw;
4321         struct igb_buffer *buffer_info;
4322         struct sk_buff *skb;
4323         union e1000_adv_tx_desc *tx_desc, *eop_desc;
4324         unsigned int total_bytes = 0, total_packets = 0;
4325         unsigned int i, eop, count = 0;
4326         bool cleaned = false;
4327
4328         i = tx_ring->next_to_clean;
4329         eop = tx_ring->buffer_info[i].next_to_watch;
4330         eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4331
4332         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
4333                (count < tx_ring->count)) {
4334                 for (cleaned = false; !cleaned; count++) {
4335                         tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
4336                         buffer_info = &tx_ring->buffer_info[i];
4337                         cleaned = (i == eop);
4338                         skb = buffer_info->skb;
4339
4340                         if (skb) {
4341                                 unsigned int segs, bytecount;
4342                                 /* gso_segs is currently only valid for tcp */
4343                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
4344                                 /* multiply data chunks by size of headers */
4345                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
4346                                             skb->len;
4347                                 total_packets += segs;
4348                                 total_bytes += bytecount;
4349
4350                                 igb_tx_hwtstamp(adapter, skb);
4351                         }
4352
4353                         igb_unmap_and_free_tx_resource(adapter, buffer_info);
4354                         tx_desc->wb.status = 0;
4355
4356                         i++;
4357                         if (i == tx_ring->count)
4358                                 i = 0;
4359                 }
4360                 eop = tx_ring->buffer_info[i].next_to_watch;
4361                 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4362         }
4363
4364         tx_ring->next_to_clean = i;
4365
4366         if (unlikely(count &&
4367                      netif_carrier_ok(netdev) &&
4368                      igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
4369                 /* Make sure that anybody stopping the queue after this
4370                  * sees the new next_to_clean.
4371                  */
4372                 smp_mb();
4373                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
4374                     !(test_bit(__IGB_DOWN, &adapter->state))) {
4375                         netif_wake_subqueue(netdev, tx_ring->queue_index);
4376                         ++adapter->restart_queue;
4377                 }
4378         }
4379
4380         if (tx_ring->detect_tx_hung) {
4381                 /* Detect a transmit hang in hardware, this serializes the
4382                  * check with the clearing of time_stamp and movement of i */
4383                 tx_ring->detect_tx_hung = false;
4384                 if (tx_ring->buffer_info[i].time_stamp &&
4385                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
4386                                (adapter->tx_timeout_factor * HZ))
4387                     && !(rd32(E1000_STATUS) &
4388                          E1000_STATUS_TXOFF)) {
4389
4390                         /* detected Tx unit hang */
4391                         dev_err(&adapter->pdev->dev,
4392                                 "Detected Tx Unit Hang\n"
4393                                 "  Tx Queue             <%d>\n"
4394                                 "  TDH                  <%x>\n"
4395                                 "  TDT                  <%x>\n"
4396                                 "  next_to_use          <%x>\n"
4397                                 "  next_to_clean        <%x>\n"
4398                                 "buffer_info[next_to_clean]\n"
4399                                 "  time_stamp           <%lx>\n"
4400                                 "  next_to_watch        <%x>\n"
4401                                 "  jiffies              <%lx>\n"
4402                                 "  desc.status          <%x>\n",
4403                                 tx_ring->queue_index,
4404                                 readl(adapter->hw.hw_addr + tx_ring->head),
4405                                 readl(adapter->hw.hw_addr + tx_ring->tail),
4406                                 tx_ring->next_to_use,
4407                                 tx_ring->next_to_clean,
4408                                 tx_ring->buffer_info[i].time_stamp,
4409                                 eop,
4410                                 jiffies,
4411                                 eop_desc->wb.status);
4412                         netif_stop_subqueue(netdev, tx_ring->queue_index);
4413                 }
4414         }
4415         tx_ring->total_bytes += total_bytes;
4416         tx_ring->total_packets += total_packets;
4417         tx_ring->tx_stats.bytes += total_bytes;
4418         tx_ring->tx_stats.packets += total_packets;
4419         adapter->net_stats.tx_bytes += total_bytes;
4420         adapter->net_stats.tx_packets += total_packets;
4421         return (count < tx_ring->count);
4422 }
4423
4424 /**
4425  * igb_receive_skb - helper function to handle rx indications
4426  * @ring: pointer to receive ring receving this packet
4427  * @status: descriptor status field as written by hardware
4428  * @rx_desc: receive descriptor containing vlan and type information.
4429  * @skb: pointer to sk_buff to be indicated to stack
4430  **/
4431 static void igb_receive_skb(struct igb_ring *ring, u8 status,
4432                             union e1000_adv_rx_desc * rx_desc,
4433                             struct sk_buff *skb)
4434 {
4435         struct igb_adapter * adapter = ring->adapter;
4436         bool vlan_extracted = (adapter->vlgrp && (status & E1000_RXD_STAT_VP));
4437
4438         skb_record_rx_queue(skb, ring->queue_index);
4439         if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4440                 if (vlan_extracted)
4441                         vlan_gro_receive(&ring->napi, adapter->vlgrp,
4442                                          le16_to_cpu(rx_desc->wb.upper.vlan),
4443                                          skb);
4444                 else
4445                         napi_gro_receive(&ring->napi, skb);
4446         } else {
4447                 if (vlan_extracted)
4448                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4449                                           le16_to_cpu(rx_desc->wb.upper.vlan));
4450                 else
4451                         netif_receive_skb(skb);
4452         }
4453 }
4454
4455 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
4456                                        u32 status_err, struct sk_buff *skb)
4457 {
4458         skb->ip_summed = CHECKSUM_NONE;
4459
4460         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
4461         if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
4462                 return;
4463         /* TCP/UDP checksum error bit is set */
4464         if (status_err &
4465             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
4466                 /* let the stack verify checksum errors */
4467                 adapter->hw_csum_err++;
4468                 return;
4469         }
4470         /* It must be a TCP or UDP packet with a valid checksum */
4471         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
4472                 skb->ip_summed = CHECKSUM_UNNECESSARY;
4473
4474         adapter->hw_csum_good++;
4475 }
4476
4477 static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
4478                                  int *work_done, int budget)
4479 {
4480         struct igb_adapter *adapter = rx_ring->adapter;
4481         struct net_device *netdev = adapter->netdev;
4482         struct e1000_hw *hw = &adapter->hw;
4483         struct pci_dev *pdev = adapter->pdev;
4484         union e1000_adv_rx_desc *rx_desc , *next_rxd;
4485         struct igb_buffer *buffer_info , *next_buffer;
4486         struct sk_buff *skb;
4487         bool cleaned = false;
4488         int cleaned_count = 0;
4489         unsigned int total_bytes = 0, total_packets = 0;
4490         unsigned int i;
4491         u32 length, hlen, staterr;
4492
4493         i = rx_ring->next_to_clean;
4494         buffer_info = &rx_ring->buffer_info[i];
4495         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4496         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4497
4498         while (staterr & E1000_RXD_STAT_DD) {
4499                 if (*work_done >= budget)
4500                         break;
4501                 (*work_done)++;
4502
4503                 skb = buffer_info->skb;
4504                 prefetch(skb->data - NET_IP_ALIGN);
4505                 buffer_info->skb = NULL;
4506
4507                 i++;
4508                 if (i == rx_ring->count)
4509                         i = 0;
4510                 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
4511                 prefetch(next_rxd);
4512                 next_buffer = &rx_ring->buffer_info[i];
4513
4514                 length = le16_to_cpu(rx_desc->wb.upper.length);
4515                 cleaned = true;
4516                 cleaned_count++;
4517
4518                 if (!adapter->rx_ps_hdr_size) {
4519                         pci_unmap_single(pdev, buffer_info->dma,
4520                                          adapter->rx_buffer_len +
4521                                            NET_IP_ALIGN,
4522                                          PCI_DMA_FROMDEVICE);
4523                         skb_put(skb, length);
4524                         goto send_up;
4525                 }
4526
4527                 /* HW will not DMA in data larger than the given buffer, even
4528                  * if it parses the (NFS, of course) header to be larger.  In
4529                  * that case, it fills the header buffer and spills the rest
4530                  * into the page.
4531                  */
4532                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
4533                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
4534                 if (hlen > adapter->rx_ps_hdr_size)
4535                         hlen = adapter->rx_ps_hdr_size;
4536
4537                 if (!skb_shinfo(skb)->nr_frags) {
4538                         pci_unmap_single(pdev, buffer_info->dma,
4539                                          adapter->rx_ps_hdr_size + NET_IP_ALIGN,
4540                                          PCI_DMA_FROMDEVICE);
4541                         skb_put(skb, hlen);
4542                 }
4543
4544                 if (length) {
4545                         pci_unmap_page(pdev, buffer_info->page_dma,
4546                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
4547                         buffer_info->page_dma = 0;
4548
4549                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
4550                                                 buffer_info->page,
4551                                                 buffer_info->page_offset,
4552                                                 length);
4553
4554                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
4555                             (page_count(buffer_info->page) != 1))
4556                                 buffer_info->page = NULL;
4557                         else
4558                                 get_page(buffer_info->page);
4559
4560                         skb->len += length;
4561                         skb->data_len += length;
4562
4563                         skb->truesize += length;
4564                 }
4565
4566                 if (!(staterr & E1000_RXD_STAT_EOP)) {
4567                         buffer_info->skb = next_buffer->skb;
4568                         buffer_info->dma = next_buffer->dma;
4569                         next_buffer->skb = skb;
4570                         next_buffer->dma = 0;
4571                         goto next_desc;
4572                 }
4573 send_up:
4574                 /*
4575                  * If this bit is set, then the RX registers contain
4576                  * the time stamp. No other packet will be time
4577                  * stamped until we read these registers, so read the
4578                  * registers to make them available again. Because
4579                  * only one packet can be time stamped at a time, we
4580                  * know that the register values must belong to this
4581                  * one here and therefore we don't need to compare
4582                  * any of the additional attributes stored for it.
4583                  *
4584                  * If nothing went wrong, then it should have a
4585                  * skb_shared_tx that we can turn into a
4586                  * skb_shared_hwtstamps.
4587                  *
4588                  * TODO: can time stamping be triggered (thus locking
4589                  * the registers) without the packet reaching this point
4590                  * here? In that case RX time stamping would get stuck.
4591                  *
4592                  * TODO: in "time stamp all packets" mode this bit is
4593                  * not set. Need a global flag for this mode and then
4594                  * always read the registers. Cannot be done without
4595                  * a race condition.
4596                  */
4597                 if (unlikely(staterr & E1000_RXD_STAT_TS)) {
4598                         u64 regval;
4599                         u64 ns;
4600                         struct skb_shared_hwtstamps *shhwtstamps =
4601                                 skb_hwtstamps(skb);
4602
4603                         WARN(!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID),
4604                              "igb: no RX time stamp available for time stamped packet");
4605                         regval = rd32(E1000_RXSTMPL);
4606                         regval |= (u64)rd32(E1000_RXSTMPH) << 32;
4607                         ns = timecounter_cyc2time(&adapter->clock, regval);
4608                         timecompare_update(&adapter->compare, ns);
4609                         memset(shhwtstamps, 0, sizeof(*shhwtstamps));
4610                         shhwtstamps->hwtstamp = ns_to_ktime(ns);
4611                         shhwtstamps->syststamp =
4612                                 timecompare_transform(&adapter->compare, ns);
4613                 }
4614
4615                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
4616                         dev_kfree_skb_irq(skb);
4617                         goto next_desc;
4618                 }
4619
4620                 total_bytes += skb->len;
4621                 total_packets++;
4622
4623                 igb_rx_checksum_adv(adapter, staterr, skb);
4624
4625                 skb->protocol = eth_type_trans(skb, netdev);
4626
4627                 igb_receive_skb(rx_ring, staterr, rx_desc, skb);
4628
4629 next_desc:
4630                 rx_desc->wb.upper.status_error = 0;
4631
4632                 /* return some buffers to hardware, one at a time is too slow */
4633                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
4634                         igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
4635                         cleaned_count = 0;
4636                 }
4637
4638                 /* use prefetched values */
4639                 rx_desc = next_rxd;
4640                 buffer_info = next_buffer;
4641                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4642         }
4643
4644         rx_ring->next_to_clean = i;
4645         cleaned_count = igb_desc_unused(rx_ring);
4646
4647         if (cleaned_count)
4648                 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
4649
4650         rx_ring->total_packets += total_packets;
4651         rx_ring->total_bytes += total_bytes;
4652         rx_ring->rx_stats.packets += total_packets;
4653         rx_ring->rx_stats.bytes += total_bytes;
4654         adapter->net_stats.rx_bytes += total_bytes;
4655         adapter->net_stats.rx_packets += total_packets;
4656         return cleaned;
4657 }
4658
4659 /**
4660  * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
4661  * @adapter: address of board private structure
4662  **/
4663 static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
4664                                      int cleaned_count)
4665 {
4666         struct igb_adapter *adapter = rx_ring->adapter;
4667         struct net_device *netdev = adapter->netdev;
4668         struct pci_dev *pdev = adapter->pdev;
4669         union e1000_adv_rx_desc *rx_desc;
4670         struct igb_buffer *buffer_info;
4671         struct sk_buff *skb;
4672         unsigned int i;
4673         int bufsz;
4674
4675         i = rx_ring->next_to_use;
4676         buffer_info = &rx_ring->buffer_info[i];
4677
4678         if (adapter->rx_ps_hdr_size)
4679                 bufsz = adapter->rx_ps_hdr_size;
4680         else
4681                 bufsz = adapter->rx_buffer_len;
4682         bufsz += NET_IP_ALIGN;
4683
4684         while (cleaned_count--) {
4685                 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4686
4687                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
4688                         if (!buffer_info->page) {
4689                                 buffer_info->page = alloc_page(GFP_ATOMIC);
4690                                 if (!buffer_info->page) {
4691                                         adapter->alloc_rx_buff_failed++;
4692                                         goto no_buffers;
4693                                 }
4694                                 buffer_info->page_offset = 0;
4695                         } else {
4696                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
4697                         }
4698                         buffer_info->page_dma =
4699                                 pci_map_page(pdev, buffer_info->page,
4700                                              buffer_info->page_offset,
4701                                              PAGE_SIZE / 2,
4702                                              PCI_DMA_FROMDEVICE);
4703                 }
4704
4705                 if (!buffer_info->skb) {
4706                         skb = netdev_alloc_skb(netdev, bufsz);
4707                         if (!skb) {
4708                                 adapter->alloc_rx_buff_failed++;
4709                                 goto no_buffers;
4710                         }
4711
4712                         /* Make buffer alignment 2 beyond a 16 byte boundary
4713                          * this will result in a 16 byte aligned IP header after
4714                          * the 14 byte MAC header is removed
4715                          */
4716                         skb_reserve(skb, NET_IP_ALIGN);
4717
4718                         buffer_info->skb = skb;
4719                         buffer_info->dma = pci_map_single(pdev, skb->data,
4720                                                           bufsz,
4721                                                           PCI_DMA_FROMDEVICE);
4722                 }
4723                 /* Refresh the desc even if buffer_addrs didn't change because
4724                  * each write-back erases this info. */
4725                 if (adapter->rx_ps_hdr_size) {
4726                         rx_desc->read.pkt_addr =
4727                              cpu_to_le64(buffer_info->page_dma);
4728                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
4729                 } else {
4730                         rx_desc->read.pkt_addr =
4731                              cpu_to_le64(buffer_info->dma);
4732                         rx_desc->read.hdr_addr = 0;
4733                 }
4734
4735                 i++;
4736                 if (i == rx_ring->count)
4737                         i = 0;
4738                 buffer_info = &rx_ring->buffer_info[i];
4739         }
4740
4741 no_buffers:
4742         if (rx_ring->next_to_use != i) {
4743                 rx_ring->next_to_use = i;
4744                 if (i == 0)
4745                         i = (rx_ring->count - 1);
4746                 else
4747                         i--;
4748
4749                 /* Force memory writes to complete before letting h/w
4750                  * know there are new descriptors to fetch.  (Only
4751                  * applicable for weak-ordered memory model archs,
4752                  * such as IA-64). */
4753                 wmb();
4754                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
4755         }
4756 }
4757
4758 /**
4759  * igb_mii_ioctl -
4760  * @netdev:
4761  * @ifreq:
4762  * @cmd:
4763  **/
4764 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4765 {
4766         struct igb_adapter *adapter = netdev_priv(netdev);
4767         struct mii_ioctl_data *data = if_mii(ifr);
4768
4769         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4770                 return -EOPNOTSUPP;
4771
4772         switch (cmd) {
4773         case SIOCGMIIPHY:
4774                 data->phy_id = adapter->hw.phy.addr;
4775                 break;
4776         case SIOCGMIIREG:
4777                 if (!capable(CAP_NET_ADMIN))
4778                         return -EPERM;
4779                 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4780                                      &data->val_out))
4781                         return -EIO;
4782                 break;
4783         case SIOCSMIIREG:
4784         default:
4785                 return -EOPNOTSUPP;
4786         }
4787         return 0;
4788 }
4789
4790 /**
4791  * igb_hwtstamp_ioctl - control hardware time stamping
4792  * @netdev:
4793  * @ifreq:
4794  * @cmd:
4795  *
4796  * Outgoing time stamping can be enabled and disabled. Play nice and
4797  * disable it when requested, although it shouldn't case any overhead
4798  * when no packet needs it. At most one packet in the queue may be
4799  * marked for time stamping, otherwise it would be impossible to tell
4800  * for sure to which packet the hardware time stamp belongs.
4801  *
4802  * Incoming time stamping has to be configured via the hardware
4803  * filters. Not all combinations are supported, in particular event
4804  * type has to be specified. Matching the kind of event packet is
4805  * not supported, with the exception of "all V2 events regardless of
4806  * level 2 or 4".
4807  *
4808  **/
4809 static int igb_hwtstamp_ioctl(struct net_device *netdev,
4810                               struct ifreq *ifr, int cmd)
4811 {
4812         struct igb_adapter *adapter = netdev_priv(netdev);
4813         struct e1000_hw *hw = &adapter->hw;
4814         struct hwtstamp_config config;
4815         u32 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4816         u32 tsync_rx_ctl_bit = E1000_TSYNCRXCTL_ENABLED;
4817         u32 tsync_rx_ctl_type = 0;
4818         u32 tsync_rx_cfg = 0;
4819         int is_l4 = 0;
4820         int is_l2 = 0;
4821         short port = 319; /* PTP */
4822         u32 regval;
4823
4824         if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
4825                 return -EFAULT;
4826
4827         /* reserved for future extensions */
4828         if (config.flags)
4829                 return -EINVAL;
4830
4831         switch (config.tx_type) {
4832         case HWTSTAMP_TX_OFF:
4833                 tsync_tx_ctl_bit = 0;
4834                 break;
4835         case HWTSTAMP_TX_ON:
4836                 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4837                 break;
4838         default:
4839                 return -ERANGE;
4840         }
4841
4842         switch (config.rx_filter) {
4843         case HWTSTAMP_FILTER_NONE:
4844                 tsync_rx_ctl_bit = 0;
4845                 break;
4846         case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4847         case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4848         case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4849         case HWTSTAMP_FILTER_ALL:
4850                 /*
4851                  * register TSYNCRXCFG must be set, therefore it is not
4852                  * possible to time stamp both Sync and Delay_Req messages
4853                  * => fall back to time stamping all packets
4854                  */
4855                 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_ALL;
4856                 config.rx_filter = HWTSTAMP_FILTER_ALL;
4857                 break;
4858         case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4859                 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4860                 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
4861                 is_l4 = 1;
4862                 break;
4863         case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4864                 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4865                 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
4866                 is_l4 = 1;
4867                 break;
4868         case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4869         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4870                 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4871                 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
4872                 is_l2 = 1;
4873                 is_l4 = 1;
4874                 config.rx_filter = HWTSTAMP_FILTER_SOME;
4875                 break;
4876         case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4877         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4878                 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4879                 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
4880                 is_l2 = 1;
4881                 is_l4 = 1;
4882                 config.rx_filter = HWTSTAMP_FILTER_SOME;
4883                 break;
4884         case HWTSTAMP_FILTER_PTP_V2_EVENT:
4885         case HWTSTAMP_FILTER_PTP_V2_SYNC:
4886         case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4887                 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_EVENT_V2;
4888                 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
4889                 is_l2 = 1;
4890                 break;
4891         default:
4892                 return -ERANGE;
4893         }
4894
4895         /* enable/disable TX */
4896         regval = rd32(E1000_TSYNCTXCTL);
4897         regval = (regval & ~E1000_TSYNCTXCTL_ENABLED) | tsync_tx_ctl_bit;
4898         wr32(E1000_TSYNCTXCTL, regval);
4899
4900         /* enable/disable RX, define which PTP packets are time stamped */
4901         regval = rd32(E1000_TSYNCRXCTL);
4902         regval = (regval & ~E1000_TSYNCRXCTL_ENABLED) | tsync_rx_ctl_bit;
4903         regval = (regval & ~0xE) | tsync_rx_ctl_type;
4904         wr32(E1000_TSYNCRXCTL, regval);
4905         wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
4906
4907         /*
4908          * Ethertype Filter Queue Filter[0][15:0] = 0x88F7
4909          *                                          (Ethertype to filter on)
4910          * Ethertype Filter Queue Filter[0][26] = 0x1 (Enable filter)
4911          * Ethertype Filter Queue Filter[0][30] = 0x1 (Enable Timestamping)
4912          */
4913         wr32(E1000_ETQF0, is_l2 ? 0x440088f7 : 0);
4914
4915         /* L4 Queue Filter[0]: only filter by source and destination port */
4916         wr32(E1000_SPQF0, htons(port));
4917         wr32(E1000_IMIREXT(0), is_l4 ?
4918              ((1<<12) | (1<<19) /* bypass size and control flags */) : 0);
4919         wr32(E1000_IMIR(0), is_l4 ?
4920              (htons(port)
4921               | (0<<16) /* immediate interrupt disabled */
4922               | 0 /* (1<<17) bit cleared: do not bypass
4923                      destination port check */)
4924                 : 0);
4925         wr32(E1000_FTQF0, is_l4 ?
4926              (0x11 /* UDP */
4927               | (1<<15) /* VF not compared */
4928               | (1<<27) /* Enable Timestamping */
4929               | (7<<28) /* only source port filter enabled,
4930                            source/target address and protocol
4931                            masked */)
4932              : ((1<<15) | (15<<28) /* all mask bits set = filter not
4933                                       enabled */));
4934
4935         wrfl();
4936
4937         adapter->hwtstamp_config = config;
4938
4939         /* clear TX/RX time stamp registers, just to be sure */
4940         regval = rd32(E1000_TXSTMPH);
4941         regval = rd32(E1000_RXSTMPH);
4942
4943         return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
4944                 -EFAULT : 0;
4945 }
4946
4947 /**
4948  * igb_ioctl -
4949  * @netdev:
4950  * @ifreq:
4951  * @cmd:
4952  **/
4953 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4954 {
4955         switch (cmd) {
4956         case SIOCGMIIPHY:
4957         case SIOCGMIIREG:
4958         case SIOCSMIIREG:
4959                 return igb_mii_ioctl(netdev, ifr, cmd);
4960         case SIOCSHWTSTAMP:
4961                 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
4962         default:
4963                 return -EOPNOTSUPP;
4964         }
4965 }
4966
4967 static void igb_vlan_rx_register(struct net_device *netdev,
4968                                  struct vlan_group *grp)
4969 {
4970         struct igb_adapter *adapter = netdev_priv(netdev);
4971         struct e1000_hw *hw = &adapter->hw;
4972         u32 ctrl, rctl;
4973
4974         igb_irq_disable(adapter);
4975         adapter->vlgrp = grp;
4976
4977         if (grp) {
4978                 /* enable VLAN tag insert/strip */
4979                 ctrl = rd32(E1000_CTRL);
4980                 ctrl |= E1000_CTRL_VME;
4981                 wr32(E1000_CTRL, ctrl);
4982
4983                 /* enable VLAN receive filtering */
4984                 rctl = rd32(E1000_RCTL);
4985                 rctl &= ~E1000_RCTL_CFIEN;
4986                 wr32(E1000_RCTL, rctl);
4987                 igb_update_mng_vlan(adapter);
4988         } else {
4989                 /* disable VLAN tag insert/strip */
4990                 ctrl = rd32(E1000_CTRL);
4991                 ctrl &= ~E1000_CTRL_VME;
4992                 wr32(E1000_CTRL, ctrl);
4993
4994                 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
4995                         igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4996                         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
4997                 }
4998         }
4999
5000         igb_rlpml_set(adapter);
5001
5002         if (!test_bit(__IGB_DOWN, &adapter->state))
5003                 igb_irq_enable(adapter);
5004 }
5005
5006 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
5007 {
5008         struct igb_adapter *adapter = netdev_priv(netdev);
5009         struct e1000_hw *hw = &adapter->hw;
5010         int pf_id = adapter->vfs_allocated_count;
5011
5012         if ((hw->mng_cookie.status &
5013              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
5014             (vid == adapter->mng_vlan_id))
5015                 return;
5016
5017         /* add vid to vlvf if sr-iov is enabled,
5018          * if that fails add directly to filter table */
5019         if (igb_vlvf_set(adapter, vid, true, pf_id))
5020                 igb_vfta_set(hw, vid, true);
5021
5022 }
5023
5024 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
5025 {
5026         struct igb_adapter *adapter = netdev_priv(netdev);
5027         struct e1000_hw *hw = &adapter->hw;
5028         int pf_id = adapter->vfs_allocated_count;
5029
5030         igb_irq_disable(adapter);
5031         vlan_group_set_device(adapter->vlgrp, vid, NULL);
5032
5033         if (!test_bit(__IGB_DOWN, &adapter->state))
5034                 igb_irq_enable(adapter);
5035
5036         if ((adapter->hw.mng_cookie.status &
5037              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
5038             (vid == adapter->mng_vlan_id)) {
5039                 /* release control to f/w */
5040                 igb_release_hw_control(adapter);
5041                 return;
5042         }
5043
5044         /* remove vid from vlvf if sr-iov is enabled,
5045          * if not in vlvf remove from vfta */
5046         if (igb_vlvf_set(adapter, vid, false, pf_id))
5047                 igb_vfta_set(hw, vid, false);
5048 }
5049
5050 static void igb_restore_vlan(struct igb_adapter *adapter)
5051 {
5052         igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5053
5054         if (adapter->vlgrp) {
5055                 u16 vid;
5056                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5057                         if (!vlan_group_get_device(adapter->vlgrp, vid))
5058                                 continue;
5059                         igb_vlan_rx_add_vid(adapter->netdev, vid);
5060                 }
5061         }
5062 }
5063
5064 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
5065 {
5066         struct e1000_mac_info *mac = &adapter->hw.mac;
5067
5068         mac->autoneg = 0;
5069
5070         /* Fiber NICs only allow 1000 gbps Full duplex */
5071         if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
5072                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
5073                 dev_err(&adapter->pdev->dev,
5074                         "Unsupported Speed/Duplex configuration\n");
5075                 return -EINVAL;
5076         }
5077
5078         switch (spddplx) {
5079         case SPEED_10 + DUPLEX_HALF:
5080                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
5081                 break;
5082         case SPEED_10 + DUPLEX_FULL:
5083                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
5084                 break;
5085         case SPEED_100 + DUPLEX_HALF:
5086                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
5087                 break;
5088         case SPEED_100 + DUPLEX_FULL:
5089                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
5090                 break;
5091         case SPEED_1000 + DUPLEX_FULL:
5092                 mac->autoneg = 1;
5093                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
5094                 break;
5095         case SPEED_1000 + DUPLEX_HALF: /* not supported */
5096         default:
5097                 dev_err(&adapter->pdev->dev,
5098                         "Unsupported Speed/Duplex configuration\n");
5099                 return -EINVAL;
5100         }
5101         return 0;
5102 }
5103
5104 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
5105 {
5106         struct net_device *netdev = pci_get_drvdata(pdev);
5107         struct igb_adapter *adapter = netdev_priv(netdev);
5108         struct e1000_hw *hw = &adapter->hw;
5109         u32 ctrl, rctl, status;
5110         u32 wufc = adapter->wol;
5111 #ifdef CONFIG_PM
5112         int retval = 0;
5113 #endif
5114
5115         netif_device_detach(netdev);
5116
5117         if (netif_running(netdev))
5118                 igb_close(netdev);
5119
5120         igb_reset_interrupt_capability(adapter);
5121
5122         igb_free_queues(adapter);
5123
5124 #ifdef CONFIG_PM
5125         retval = pci_save_state(pdev);
5126         if (retval)
5127                 return retval;
5128 #endif
5129
5130         status = rd32(E1000_STATUS);
5131         if (status & E1000_STATUS_LU)
5132                 wufc &= ~E1000_WUFC_LNKC;
5133
5134         if (wufc) {
5135                 igb_setup_rctl(adapter);
5136                 igb_set_multi(netdev);
5137
5138                 /* turn on all-multi mode if wake on multicast is enabled */
5139                 if (wufc & E1000_WUFC_MC) {
5140                         rctl = rd32(E1000_RCTL);
5141                         rctl |= E1000_RCTL_MPE;
5142                         wr32(E1000_RCTL, rctl);
5143                 }
5144
5145                 ctrl = rd32(E1000_CTRL);
5146                 /* advertise wake from D3Cold */
5147                 #define E1000_CTRL_ADVD3WUC 0x00100000
5148                 /* phy power management enable */
5149                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5150                 ctrl |= E1000_CTRL_ADVD3WUC;
5151                 wr32(E1000_CTRL, ctrl);
5152
5153                 /* Allow time for pending master requests to run */
5154                 igb_disable_pcie_master(&adapter->hw);
5155
5156                 wr32(E1000_WUC, E1000_WUC_PME_EN);
5157                 wr32(E1000_WUFC, wufc);
5158         } else {
5159                 wr32(E1000_WUC, 0);
5160                 wr32(E1000_WUFC, 0);
5161         }
5162
5163         *enable_wake = wufc || adapter->en_mng_pt;
5164         if (!*enable_wake)
5165                 igb_shutdown_fiber_serdes_link_82575(hw);
5166
5167         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
5168          * would have already happened in close and is redundant. */
5169         igb_release_hw_control(adapter);
5170
5171         pci_disable_device(pdev);
5172
5173         return 0;
5174 }
5175
5176 #ifdef CONFIG_PM
5177 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
5178 {
5179         int retval;
5180         bool wake;
5181
5182         retval = __igb_shutdown(pdev, &wake);
5183         if (retval)
5184                 return retval;
5185
5186         if (wake) {
5187                 pci_prepare_to_sleep(pdev);
5188         } else {
5189                 pci_wake_from_d3(pdev, false);
5190                 pci_set_power_state(pdev, PCI_D3hot);
5191         }
5192
5193         return 0;
5194 }
5195
5196 static int igb_resume(struct pci_dev *pdev)
5197 {
5198         struct net_device *netdev = pci_get_drvdata(pdev);
5199         struct igb_adapter *adapter = netdev_priv(netdev);
5200         struct e1000_hw *hw = &adapter->hw;
5201         u32 err;
5202
5203         pci_set_power_state(pdev, PCI_D0);
5204         pci_restore_state(pdev);
5205
5206         err = pci_enable_device_mem(pdev);
5207         if (err) {
5208                 dev_err(&pdev->dev,
5209                         "igb: Cannot enable PCI device from suspend\n");
5210                 return err;
5211         }
5212         pci_set_master(pdev);
5213
5214         pci_enable_wake(pdev, PCI_D3hot, 0);
5215         pci_enable_wake(pdev, PCI_D3cold, 0);
5216
5217         igb_set_interrupt_capability(adapter);
5218
5219         if (igb_alloc_queues(adapter)) {
5220                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5221                 return -ENOMEM;
5222         }
5223
5224         /* e1000_power_up_phy(adapter); */
5225
5226         igb_reset(adapter);
5227
5228         /* let the f/w know that the h/w is now under the control of the
5229          * driver. */
5230         igb_get_hw_control(adapter);
5231
5232         wr32(E1000_WUS, ~0);
5233
5234         if (netif_running(netdev)) {
5235                 err = igb_open(netdev);
5236                 if (err)
5237                         return err;
5238         }
5239
5240         netif_device_attach(netdev);
5241
5242         return 0;
5243 }
5244 #endif
5245
5246 static void igb_shutdown(struct pci_dev *pdev)
5247 {
5248         bool wake;
5249
5250         __igb_shutdown(pdev, &wake);
5251
5252         if (system_state == SYSTEM_POWER_OFF) {
5253                 pci_wake_from_d3(pdev, wake);
5254                 pci_set_power_state(pdev, PCI_D3hot);
5255         }
5256 }
5257
5258 #ifdef CONFIG_NET_POLL_CONTROLLER
5259 /*
5260  * Polling 'interrupt' - used by things like netconsole to send skbs
5261  * without having to re-enable interrupts. It's not called while
5262  * the interrupt routine is executing.
5263  */
5264 static void igb_netpoll(struct net_device *netdev)
5265 {
5266         struct igb_adapter *adapter = netdev_priv(netdev);
5267         struct e1000_hw *hw = &adapter->hw;
5268         int i;
5269
5270         if (!adapter->msix_entries) {
5271                 igb_irq_disable(adapter);
5272                 napi_schedule(&adapter->rx_ring[0].napi);
5273                 return;
5274         }
5275
5276         for (i = 0; i < adapter->num_tx_queues; i++) {
5277                 struct igb_ring *tx_ring = &adapter->tx_ring[i];
5278                 wr32(E1000_EIMC, tx_ring->eims_value);
5279                 igb_clean_tx_irq(tx_ring);
5280                 wr32(E1000_EIMS, tx_ring->eims_value);
5281         }
5282
5283         for (i = 0; i < adapter->num_rx_queues; i++) {
5284                 struct igb_ring *rx_ring = &adapter->rx_ring[i];
5285                 wr32(E1000_EIMC, rx_ring->eims_value);
5286                 napi_schedule(&rx_ring->napi);
5287         }
5288 }
5289 #endif /* CONFIG_NET_POLL_CONTROLLER */
5290
5291 /**
5292  * igb_io_error_detected - called when PCI error is detected
5293  * @pdev: Pointer to PCI device
5294  * @state: The current pci connection state
5295  *
5296  * This function is called after a PCI bus error affecting
5297  * this device has been detected.
5298  */
5299 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
5300                                               pci_channel_state_t state)
5301 {
5302         struct net_device *netdev = pci_get_drvdata(pdev);
5303         struct igb_adapter *adapter = netdev_priv(netdev);
5304
5305         netif_device_detach(netdev);
5306
5307         if (netif_running(netdev))
5308                 igb_down(adapter);
5309         pci_disable_device(pdev);
5310
5311         /* Request a slot slot reset. */
5312         return PCI_ERS_RESULT_NEED_RESET;
5313 }
5314
5315 /**
5316  * igb_io_slot_reset - called after the pci bus has been reset.
5317  * @pdev: Pointer to PCI device
5318  *
5319  * Restart the card from scratch, as if from a cold-boot. Implementation
5320  * resembles the first-half of the igb_resume routine.
5321  */
5322 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
5323 {
5324         struct net_device *netdev = pci_get_drvdata(pdev);
5325         struct igb_adapter *adapter = netdev_priv(netdev);
5326         struct e1000_hw *hw = &adapter->hw;
5327         pci_ers_result_t result;
5328         int err;
5329
5330         if (pci_enable_device_mem(pdev)) {
5331                 dev_err(&pdev->dev,
5332                         "Cannot re-enable PCI device after reset.\n");
5333                 result = PCI_ERS_RESULT_DISCONNECT;
5334         } else {
5335                 pci_set_master(pdev);
5336                 pci_restore_state(pdev);
5337
5338                 pci_enable_wake(pdev, PCI_D3hot, 0);
5339                 pci_enable_wake(pdev, PCI_D3cold, 0);
5340
5341                 igb_reset(adapter);
5342                 wr32(E1000_WUS, ~0);
5343                 result = PCI_ERS_RESULT_RECOVERED;
5344         }
5345
5346         err = pci_cleanup_aer_uncorrect_error_status(pdev);
5347         if (err) {
5348                 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
5349                         "failed 0x%0x\n", err);
5350                 /* non-fatal, continue */
5351         }
5352
5353         return result;
5354 }
5355
5356 /**
5357  * igb_io_resume - called when traffic can start flowing again.
5358  * @pdev: Pointer to PCI device
5359  *
5360  * This callback is called when the error recovery driver tells us that
5361  * its OK to resume normal operation. Implementation resembles the
5362  * second-half of the igb_resume routine.
5363  */
5364 static void igb_io_resume(struct pci_dev *pdev)
5365 {
5366         struct net_device *netdev = pci_get_drvdata(pdev);
5367         struct igb_adapter *adapter = netdev_priv(netdev);
5368
5369         if (netif_running(netdev)) {
5370                 if (igb_up(adapter)) {
5371                         dev_err(&pdev->dev, "igb_up failed after reset\n");
5372                         return;
5373                 }
5374         }
5375
5376         netif_device_attach(netdev);
5377
5378         /* let the f/w know that the h/w is now under the control of the
5379          * driver. */
5380         igb_get_hw_control(adapter);
5381 }
5382
5383 static inline void igb_set_vmolr(struct e1000_hw *hw, int vfn)
5384 {
5385         u32 reg_data;
5386
5387         reg_data = rd32(E1000_VMOLR(vfn));
5388         reg_data |= E1000_VMOLR_BAM |    /* Accept broadcast */
5389                     E1000_VMOLR_ROPE |   /* Accept packets matched in UTA */
5390                     E1000_VMOLR_ROMPE |  /* Accept packets matched in MTA */
5391                     E1000_VMOLR_AUPE |   /* Accept untagged packets */
5392                     E1000_VMOLR_STRVLAN; /* Strip vlan tags */
5393         wr32(E1000_VMOLR(vfn), reg_data);
5394 }
5395
5396 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
5397                                  int vfn)
5398 {
5399         struct e1000_hw *hw = &adapter->hw;
5400         u32 vmolr;
5401
5402         vmolr = rd32(E1000_VMOLR(vfn));
5403         vmolr &= ~E1000_VMOLR_RLPML_MASK;
5404         vmolr |= size | E1000_VMOLR_LPE;
5405         wr32(E1000_VMOLR(vfn), vmolr);
5406
5407         return 0;
5408 }
5409
5410 static inline void igb_set_rah_pool(struct e1000_hw *hw, int pool, int entry)
5411 {
5412         u32 reg_data;
5413
5414         reg_data = rd32(E1000_RAH(entry));
5415         reg_data &= ~E1000_RAH_POOL_MASK;
5416         reg_data |= E1000_RAH_POOL_1 << pool;;
5417         wr32(E1000_RAH(entry), reg_data);
5418 }
5419
5420 static void igb_set_mc_list_pools(struct igb_adapter *adapter,
5421                                   int entry_count, u16 total_rar_filters)
5422 {
5423         struct e1000_hw *hw = &adapter->hw;
5424         int i = adapter->vfs_allocated_count + 1;
5425
5426         if ((i + entry_count) < total_rar_filters)
5427                 total_rar_filters = i + entry_count;
5428
5429         for (; i < total_rar_filters; i++)
5430                 igb_set_rah_pool(hw, adapter->vfs_allocated_count, i);
5431 }
5432
5433 static int igb_set_vf_mac(struct igb_adapter *adapter,
5434                           int vf, unsigned char *mac_addr)
5435 {
5436         struct e1000_hw *hw = &adapter->hw;
5437         int rar_entry = vf + 1; /* VF MAC addresses start at entry 1 */
5438
5439         igb_rar_set(hw, mac_addr, rar_entry);
5440
5441         memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
5442
5443         igb_set_rah_pool(hw, vf, rar_entry);
5444
5445         return 0;
5446 }
5447
5448 static void igb_vmm_control(struct igb_adapter *adapter)
5449 {
5450         struct e1000_hw *hw = &adapter->hw;
5451         u32 reg_data;
5452
5453         if (!adapter->vfs_allocated_count)
5454                 return;
5455
5456         /* VF's need PF reset indication before they
5457          * can send/receive mail */
5458         reg_data = rd32(E1000_CTRL_EXT);
5459         reg_data |= E1000_CTRL_EXT_PFRSTD;
5460         wr32(E1000_CTRL_EXT, reg_data);
5461
5462         igb_vmdq_set_loopback_pf(hw, true);
5463         igb_vmdq_set_replication_pf(hw, true);
5464 }
5465
5466 /* igb_main.c */