1 /*******************************************************************************
3 Intel(R) Gigabit Ethernet Linux driver
4 Copyright(c) 2007-2009 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #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>
48 #include <linux/dca.h>
52 #define DRV_VERSION "2.1.0-k2"
53 char igb_driver_name[] = "igb";
54 char igb_driver_version[] = DRV_VERSION;
55 static const char igb_driver_string[] =
56 "Intel(R) Gigabit Ethernet Network Driver";
57 static const char igb_copyright[] = "Copyright (c) 2007-2009 Intel Corporation.";
59 static const struct e1000_info *igb_info_tbl[] = {
60 [board_82575] = &e1000_82575_info,
63 static struct pci_device_id igb_pci_tbl[] = {
64 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
65 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
66 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
67 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
68 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
69 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
70 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
71 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
72 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
73 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
74 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
75 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
76 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
77 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
79 /* required last entry */
83 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
85 void igb_reset(struct igb_adapter *);
86 static int igb_setup_all_tx_resources(struct igb_adapter *);
87 static int igb_setup_all_rx_resources(struct igb_adapter *);
88 static void igb_free_all_tx_resources(struct igb_adapter *);
89 static void igb_free_all_rx_resources(struct igb_adapter *);
90 static void igb_setup_mrqc(struct igb_adapter *);
91 void igb_update_stats(struct igb_adapter *);
92 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
93 static void __devexit igb_remove(struct pci_dev *pdev);
94 static int igb_sw_init(struct igb_adapter *);
95 static int igb_open(struct net_device *);
96 static int igb_close(struct net_device *);
97 static void igb_configure_tx(struct igb_adapter *);
98 static void igb_configure_rx(struct igb_adapter *);
99 static void igb_clean_all_tx_rings(struct igb_adapter *);
100 static void igb_clean_all_rx_rings(struct igb_adapter *);
101 static void igb_clean_tx_ring(struct igb_ring *);
102 static void igb_clean_rx_ring(struct igb_ring *);
103 static void igb_set_rx_mode(struct net_device *);
104 static void igb_update_phy_info(unsigned long);
105 static void igb_watchdog(unsigned long);
106 static void igb_watchdog_task(struct work_struct *);
107 static netdev_tx_t igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
108 static struct net_device_stats *igb_get_stats(struct net_device *);
109 static int igb_change_mtu(struct net_device *, int);
110 static int igb_set_mac(struct net_device *, void *);
111 static void igb_set_uta(struct igb_adapter *adapter);
112 static irqreturn_t igb_intr(int irq, void *);
113 static irqreturn_t igb_intr_msi(int irq, void *);
114 static irqreturn_t igb_msix_other(int irq, void *);
115 static irqreturn_t igb_msix_ring(int irq, void *);
116 #ifdef CONFIG_IGB_DCA
117 static void igb_update_dca(struct igb_q_vector *);
118 static void igb_setup_dca(struct igb_adapter *);
119 #endif /* CONFIG_IGB_DCA */
120 static bool igb_clean_tx_irq(struct igb_q_vector *);
121 static int igb_poll(struct napi_struct *, int);
122 static bool igb_clean_rx_irq_adv(struct igb_q_vector *, int *, int);
123 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
124 static void igb_tx_timeout(struct net_device *);
125 static void igb_reset_task(struct work_struct *);
126 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
127 static void igb_vlan_rx_add_vid(struct net_device *, u16);
128 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
129 static void igb_restore_vlan(struct igb_adapter *);
130 static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
131 static void igb_ping_all_vfs(struct igb_adapter *);
132 static void igb_msg_task(struct igb_adapter *);
133 static void igb_vmm_control(struct igb_adapter *);
134 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
135 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
138 static int igb_suspend(struct pci_dev *, pm_message_t);
139 static int igb_resume(struct pci_dev *);
141 static void igb_shutdown(struct pci_dev *);
142 #ifdef CONFIG_IGB_DCA
143 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
144 static struct notifier_block dca_notifier = {
145 .notifier_call = igb_notify_dca,
150 #ifdef CONFIG_NET_POLL_CONTROLLER
151 /* for netdump / net console */
152 static void igb_netpoll(struct net_device *);
154 #ifdef CONFIG_PCI_IOV
155 static unsigned int max_vfs = 0;
156 module_param(max_vfs, uint, 0);
157 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
158 "per physical function");
159 #endif /* CONFIG_PCI_IOV */
161 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
162 pci_channel_state_t);
163 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
164 static void igb_io_resume(struct pci_dev *);
166 static struct pci_error_handlers igb_err_handler = {
167 .error_detected = igb_io_error_detected,
168 .slot_reset = igb_io_slot_reset,
169 .resume = igb_io_resume,
173 static struct pci_driver igb_driver = {
174 .name = igb_driver_name,
175 .id_table = igb_pci_tbl,
177 .remove = __devexit_p(igb_remove),
179 /* Power Managment Hooks */
180 .suspend = igb_suspend,
181 .resume = igb_resume,
183 .shutdown = igb_shutdown,
184 .err_handler = &igb_err_handler
187 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
188 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
189 MODULE_LICENSE("GPL");
190 MODULE_VERSION(DRV_VERSION);
193 * igb_read_clock - read raw cycle counter (to be used by time counter)
195 static cycle_t igb_read_clock(const struct cyclecounter *tc)
197 struct igb_adapter *adapter =
198 container_of(tc, struct igb_adapter, cycles);
199 struct e1000_hw *hw = &adapter->hw;
204 * The timestamp latches on lowest register read. For the 82580
205 * the lowest register is SYSTIMR instead of SYSTIML. However we never
206 * adjusted TIMINCA so SYSTIMR will just read as all 0s so ignore it.
208 if (hw->mac.type == e1000_82580) {
209 stamp = rd32(E1000_SYSTIMR) >> 8;
210 shift = IGB_82580_TSYNC_SHIFT;
213 stamp |= (u64)rd32(E1000_SYSTIML) << shift;
214 stamp |= (u64)rd32(E1000_SYSTIMH) << (shift + 32);
220 * igb_get_hw_dev_name - return device name string
221 * used by hardware layer to print debugging information
223 char *igb_get_hw_dev_name(struct e1000_hw *hw)
225 struct igb_adapter *adapter = hw->back;
226 return adapter->netdev->name;
230 * igb_get_time_str - format current NIC and system time as string
232 static char *igb_get_time_str(struct igb_adapter *adapter,
235 cycle_t hw = adapter->cycles.read(&adapter->cycles);
236 struct timespec nic = ns_to_timespec(timecounter_read(&adapter->clock));
238 struct timespec delta;
239 getnstimeofday(&sys);
241 delta = timespec_sub(nic, sys);
244 "HW %llu, NIC %ld.%09lus, SYS %ld.%09lus, NIC-SYS %lds + %09luns",
246 (long)nic.tv_sec, nic.tv_nsec,
247 (long)sys.tv_sec, sys.tv_nsec,
248 (long)delta.tv_sec, delta.tv_nsec);
255 * igb_init_module - Driver Registration Routine
257 * igb_init_module is the first routine called when the driver is
258 * loaded. All it does is register with the PCI subsystem.
260 static int __init igb_init_module(void)
263 printk(KERN_INFO "%s - version %s\n",
264 igb_driver_string, igb_driver_version);
266 printk(KERN_INFO "%s\n", igb_copyright);
268 #ifdef CONFIG_IGB_DCA
269 dca_register_notify(&dca_notifier);
271 ret = pci_register_driver(&igb_driver);
275 module_init(igb_init_module);
278 * igb_exit_module - Driver Exit Cleanup Routine
280 * igb_exit_module is called just before the driver is removed
283 static void __exit igb_exit_module(void)
285 #ifdef CONFIG_IGB_DCA
286 dca_unregister_notify(&dca_notifier);
288 pci_unregister_driver(&igb_driver);
291 module_exit(igb_exit_module);
293 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
295 * igb_cache_ring_register - Descriptor ring to register mapping
296 * @adapter: board private structure to initialize
298 * Once we know the feature-set enabled for the device, we'll cache
299 * the register offset the descriptor ring is assigned to.
301 static void igb_cache_ring_register(struct igb_adapter *adapter)
304 u32 rbase_offset = adapter->vfs_allocated_count;
306 switch (adapter->hw.mac.type) {
308 /* The queues are allocated for virtualization such that VF 0
309 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
310 * In order to avoid collision we start at the first free queue
311 * and continue consuming queues in the same sequence
313 if (adapter->vfs_allocated_count) {
314 for (; i < adapter->rss_queues; i++)
315 adapter->rx_ring[i].reg_idx = rbase_offset +
317 for (; j < adapter->rss_queues; j++)
318 adapter->tx_ring[j].reg_idx = rbase_offset +
324 for (; i < adapter->num_rx_queues; i++)
325 adapter->rx_ring[i].reg_idx = rbase_offset + i;
326 for (; j < adapter->num_tx_queues; j++)
327 adapter->tx_ring[j].reg_idx = rbase_offset + j;
332 static void igb_free_queues(struct igb_adapter *adapter)
334 kfree(adapter->tx_ring);
335 kfree(adapter->rx_ring);
337 adapter->tx_ring = NULL;
338 adapter->rx_ring = NULL;
340 adapter->num_rx_queues = 0;
341 adapter->num_tx_queues = 0;
345 * igb_alloc_queues - Allocate memory for all rings
346 * @adapter: board private structure to initialize
348 * We allocate one ring per queue at run-time since we don't know the
349 * number of queues at compile-time.
351 static int igb_alloc_queues(struct igb_adapter *adapter)
355 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
356 sizeof(struct igb_ring), GFP_KERNEL);
357 if (!adapter->tx_ring)
360 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
361 sizeof(struct igb_ring), GFP_KERNEL);
362 if (!adapter->rx_ring)
365 for (i = 0; i < adapter->num_tx_queues; i++) {
366 struct igb_ring *ring = &(adapter->tx_ring[i]);
367 ring->count = adapter->tx_ring_count;
368 ring->queue_index = i;
369 ring->pdev = adapter->pdev;
370 ring->netdev = adapter->netdev;
371 /* For 82575, context index must be unique per ring. */
372 if (adapter->hw.mac.type == e1000_82575)
373 ring->flags = IGB_RING_FLAG_TX_CTX_IDX;
376 for (i = 0; i < adapter->num_rx_queues; i++) {
377 struct igb_ring *ring = &(adapter->rx_ring[i]);
378 ring->count = adapter->rx_ring_count;
379 ring->queue_index = i;
380 ring->pdev = adapter->pdev;
381 ring->netdev = adapter->netdev;
382 ring->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
383 ring->flags = IGB_RING_FLAG_RX_CSUM; /* enable rx checksum */
384 /* set flag indicating ring supports SCTP checksum offload */
385 if (adapter->hw.mac.type >= e1000_82576)
386 ring->flags |= IGB_RING_FLAG_RX_SCTP_CSUM;
389 igb_cache_ring_register(adapter);
394 igb_free_queues(adapter);
399 #define IGB_N0_QUEUE -1
400 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
403 struct igb_adapter *adapter = q_vector->adapter;
404 struct e1000_hw *hw = &adapter->hw;
406 int rx_queue = IGB_N0_QUEUE;
407 int tx_queue = IGB_N0_QUEUE;
409 if (q_vector->rx_ring)
410 rx_queue = q_vector->rx_ring->reg_idx;
411 if (q_vector->tx_ring)
412 tx_queue = q_vector->tx_ring->reg_idx;
414 switch (hw->mac.type) {
416 /* The 82575 assigns vectors using a bitmask, which matches the
417 bitmask for the EICR/EIMS/EIMC registers. To assign one
418 or more queues to a vector, we write the appropriate bits
419 into the MSIXBM register for that vector. */
420 if (rx_queue > IGB_N0_QUEUE)
421 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
422 if (tx_queue > IGB_N0_QUEUE)
423 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
424 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
425 q_vector->eims_value = msixbm;
428 /* 82576 uses a table-based method for assigning vectors.
429 Each queue has a single entry in the table to which we write
430 a vector number along with a "valid" bit. Sadly, the layout
431 of the table is somewhat counterintuitive. */
432 if (rx_queue > IGB_N0_QUEUE) {
433 index = (rx_queue & 0x7);
434 ivar = array_rd32(E1000_IVAR0, index);
436 /* vector goes into low byte of register */
437 ivar = ivar & 0xFFFFFF00;
438 ivar |= msix_vector | E1000_IVAR_VALID;
440 /* vector goes into third byte of register */
441 ivar = ivar & 0xFF00FFFF;
442 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
444 array_wr32(E1000_IVAR0, index, ivar);
446 if (tx_queue > IGB_N0_QUEUE) {
447 index = (tx_queue & 0x7);
448 ivar = array_rd32(E1000_IVAR0, index);
450 /* vector goes into second byte of register */
451 ivar = ivar & 0xFFFF00FF;
452 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
454 /* vector goes into high byte of register */
455 ivar = ivar & 0x00FFFFFF;
456 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
458 array_wr32(E1000_IVAR0, index, ivar);
460 q_vector->eims_value = 1 << msix_vector;
463 /* 82580 uses the same table-based approach as 82576 but has fewer
464 entries as a result we carry over for queues greater than 4. */
465 if (rx_queue > IGB_N0_QUEUE) {
466 index = (rx_queue >> 1);
467 ivar = array_rd32(E1000_IVAR0, index);
468 if (rx_queue & 0x1) {
469 /* vector goes into third byte of register */
470 ivar = ivar & 0xFF00FFFF;
471 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
473 /* vector goes into low byte of register */
474 ivar = ivar & 0xFFFFFF00;
475 ivar |= msix_vector | E1000_IVAR_VALID;
477 array_wr32(E1000_IVAR0, index, ivar);
479 if (tx_queue > IGB_N0_QUEUE) {
480 index = (tx_queue >> 1);
481 ivar = array_rd32(E1000_IVAR0, index);
482 if (tx_queue & 0x1) {
483 /* vector goes into high byte of register */
484 ivar = ivar & 0x00FFFFFF;
485 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
487 /* vector goes into second byte of register */
488 ivar = ivar & 0xFFFF00FF;
489 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
491 array_wr32(E1000_IVAR0, index, ivar);
493 q_vector->eims_value = 1 << msix_vector;
502 * igb_configure_msix - Configure MSI-X hardware
504 * igb_configure_msix sets up the hardware to properly
505 * generate MSI-X interrupts.
507 static void igb_configure_msix(struct igb_adapter *adapter)
511 struct e1000_hw *hw = &adapter->hw;
513 adapter->eims_enable_mask = 0;
515 /* set vector for other causes, i.e. link changes */
516 switch (hw->mac.type) {
518 tmp = rd32(E1000_CTRL_EXT);
519 /* enable MSI-X PBA support*/
520 tmp |= E1000_CTRL_EXT_PBA_CLR;
522 /* Auto-Mask interrupts upon ICR read. */
523 tmp |= E1000_CTRL_EXT_EIAME;
524 tmp |= E1000_CTRL_EXT_IRCA;
526 wr32(E1000_CTRL_EXT, tmp);
528 /* enable msix_other interrupt */
529 array_wr32(E1000_MSIXBM(0), vector++,
531 adapter->eims_other = E1000_EIMS_OTHER;
537 /* Turn on MSI-X capability first, or our settings
538 * won't stick. And it will take days to debug. */
539 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
540 E1000_GPIE_PBA | E1000_GPIE_EIAME |
543 /* enable msix_other interrupt */
544 adapter->eims_other = 1 << vector;
545 tmp = (vector++ | E1000_IVAR_VALID) << 8;
547 wr32(E1000_IVAR_MISC, tmp);
550 /* do nothing, since nothing else supports MSI-X */
552 } /* switch (hw->mac.type) */
554 adapter->eims_enable_mask |= adapter->eims_other;
556 for (i = 0; i < adapter->num_q_vectors; i++) {
557 struct igb_q_vector *q_vector = adapter->q_vector[i];
558 igb_assign_vector(q_vector, vector++);
559 adapter->eims_enable_mask |= q_vector->eims_value;
566 * igb_request_msix - Initialize MSI-X interrupts
568 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
571 static int igb_request_msix(struct igb_adapter *adapter)
573 struct net_device *netdev = adapter->netdev;
574 struct e1000_hw *hw = &adapter->hw;
575 int i, err = 0, vector = 0;
577 err = request_irq(adapter->msix_entries[vector].vector,
578 igb_msix_other, 0, netdev->name, adapter);
583 for (i = 0; i < adapter->num_q_vectors; i++) {
584 struct igb_q_vector *q_vector = adapter->q_vector[i];
586 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
588 if (q_vector->rx_ring && q_vector->tx_ring)
589 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
590 q_vector->rx_ring->queue_index);
591 else if (q_vector->tx_ring)
592 sprintf(q_vector->name, "%s-tx-%u", netdev->name,
593 q_vector->tx_ring->queue_index);
594 else if (q_vector->rx_ring)
595 sprintf(q_vector->name, "%s-rx-%u", netdev->name,
596 q_vector->rx_ring->queue_index);
598 sprintf(q_vector->name, "%s-unused", netdev->name);
600 err = request_irq(adapter->msix_entries[vector].vector,
601 igb_msix_ring, 0, q_vector->name,
608 igb_configure_msix(adapter);
614 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
616 if (adapter->msix_entries) {
617 pci_disable_msix(adapter->pdev);
618 kfree(adapter->msix_entries);
619 adapter->msix_entries = NULL;
620 } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
621 pci_disable_msi(adapter->pdev);
626 * igb_free_q_vectors - Free memory allocated for interrupt vectors
627 * @adapter: board private structure to initialize
629 * This function frees the memory allocated to the q_vectors. In addition if
630 * NAPI is enabled it will delete any references to the NAPI struct prior
631 * to freeing the q_vector.
633 static void igb_free_q_vectors(struct igb_adapter *adapter)
637 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
638 struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
639 adapter->q_vector[v_idx] = NULL;
640 netif_napi_del(&q_vector->napi);
643 adapter->num_q_vectors = 0;
647 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
649 * This function resets the device so that it has 0 rx queues, tx queues, and
650 * MSI-X interrupts allocated.
652 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
654 igb_free_queues(adapter);
655 igb_free_q_vectors(adapter);
656 igb_reset_interrupt_capability(adapter);
660 * igb_set_interrupt_capability - set MSI or MSI-X if supported
662 * Attempt to configure interrupts using the best available
663 * capabilities of the hardware and kernel.
665 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
670 /* Number of supported queues. */
671 adapter->num_rx_queues = adapter->rss_queues;
672 adapter->num_tx_queues = adapter->rss_queues;
674 /* start with one vector for every rx queue */
675 numvecs = adapter->num_rx_queues;
677 /* if tx handler is seperate add 1 for every tx queue */
678 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
679 numvecs += adapter->num_tx_queues;
681 /* store the number of vectors reserved for queues */
682 adapter->num_q_vectors = numvecs;
684 /* add 1 vector for link status interrupts */
686 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
688 if (!adapter->msix_entries)
691 for (i = 0; i < numvecs; i++)
692 adapter->msix_entries[i].entry = i;
694 err = pci_enable_msix(adapter->pdev,
695 adapter->msix_entries,
700 igb_reset_interrupt_capability(adapter);
702 /* If we can't do MSI-X, try MSI */
704 #ifdef CONFIG_PCI_IOV
705 /* disable SR-IOV for non MSI-X configurations */
706 if (adapter->vf_data) {
707 struct e1000_hw *hw = &adapter->hw;
708 /* disable iov and allow time for transactions to clear */
709 pci_disable_sriov(adapter->pdev);
712 kfree(adapter->vf_data);
713 adapter->vf_data = NULL;
714 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
716 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
719 adapter->vfs_allocated_count = 0;
720 adapter->rss_queues = 1;
721 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
722 adapter->num_rx_queues = 1;
723 adapter->num_tx_queues = 1;
724 adapter->num_q_vectors = 1;
725 if (!pci_enable_msi(adapter->pdev))
726 adapter->flags |= IGB_FLAG_HAS_MSI;
728 /* Notify the stack of the (possibly) reduced Tx Queue count. */
729 adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
734 * igb_alloc_q_vectors - Allocate memory for interrupt vectors
735 * @adapter: board private structure to initialize
737 * We allocate one q_vector per queue interrupt. If allocation fails we
740 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
742 struct igb_q_vector *q_vector;
743 struct e1000_hw *hw = &adapter->hw;
746 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
747 q_vector = kzalloc(sizeof(struct igb_q_vector), GFP_KERNEL);
750 q_vector->adapter = adapter;
751 q_vector->itr_shift = (hw->mac.type == e1000_82575) ? 16 : 0;
752 q_vector->itr_register = hw->hw_addr + E1000_EITR(0);
753 q_vector->itr_val = IGB_START_ITR;
754 q_vector->set_itr = 1;
755 netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll, 64);
756 adapter->q_vector[v_idx] = q_vector;
763 q_vector = adapter->q_vector[v_idx];
764 netif_napi_del(&q_vector->napi);
766 adapter->q_vector[v_idx] = NULL;
771 static void igb_map_rx_ring_to_vector(struct igb_adapter *adapter,
772 int ring_idx, int v_idx)
774 struct igb_q_vector *q_vector;
776 q_vector = adapter->q_vector[v_idx];
777 q_vector->rx_ring = &adapter->rx_ring[ring_idx];
778 q_vector->rx_ring->q_vector = q_vector;
779 q_vector->itr_val = adapter->rx_itr_setting;
780 if (q_vector->itr_val && q_vector->itr_val <= 3)
781 q_vector->itr_val = IGB_START_ITR;
784 static void igb_map_tx_ring_to_vector(struct igb_adapter *adapter,
785 int ring_idx, int v_idx)
787 struct igb_q_vector *q_vector;
789 q_vector = adapter->q_vector[v_idx];
790 q_vector->tx_ring = &adapter->tx_ring[ring_idx];
791 q_vector->tx_ring->q_vector = q_vector;
792 q_vector->itr_val = adapter->tx_itr_setting;
793 if (q_vector->itr_val && q_vector->itr_val <= 3)
794 q_vector->itr_val = IGB_START_ITR;
798 * igb_map_ring_to_vector - maps allocated queues to vectors
800 * This function maps the recently allocated queues to vectors.
802 static int igb_map_ring_to_vector(struct igb_adapter *adapter)
807 if ((adapter->num_q_vectors < adapter->num_rx_queues) ||
808 (adapter->num_q_vectors < adapter->num_tx_queues))
811 if (adapter->num_q_vectors >=
812 (adapter->num_rx_queues + adapter->num_tx_queues)) {
813 for (i = 0; i < adapter->num_rx_queues; i++)
814 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
815 for (i = 0; i < adapter->num_tx_queues; i++)
816 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
818 for (i = 0; i < adapter->num_rx_queues; i++) {
819 if (i < adapter->num_tx_queues)
820 igb_map_tx_ring_to_vector(adapter, i, v_idx);
821 igb_map_rx_ring_to_vector(adapter, i, v_idx++);
823 for (; i < adapter->num_tx_queues; i++)
824 igb_map_tx_ring_to_vector(adapter, i, v_idx++);
830 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
832 * This function initializes the interrupts and allocates all of the queues.
834 static int igb_init_interrupt_scheme(struct igb_adapter *adapter)
836 struct pci_dev *pdev = adapter->pdev;
839 igb_set_interrupt_capability(adapter);
841 err = igb_alloc_q_vectors(adapter);
843 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
844 goto err_alloc_q_vectors;
847 err = igb_alloc_queues(adapter);
849 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
850 goto err_alloc_queues;
853 err = igb_map_ring_to_vector(adapter);
855 dev_err(&pdev->dev, "Invalid q_vector to ring mapping\n");
862 igb_free_queues(adapter);
864 igb_free_q_vectors(adapter);
866 igb_reset_interrupt_capability(adapter);
871 * igb_request_irq - initialize interrupts
873 * Attempts to configure interrupts using the best available
874 * capabilities of the hardware and kernel.
876 static int igb_request_irq(struct igb_adapter *adapter)
878 struct net_device *netdev = adapter->netdev;
879 struct pci_dev *pdev = adapter->pdev;
880 struct e1000_hw *hw = &adapter->hw;
883 if (adapter->msix_entries) {
884 err = igb_request_msix(adapter);
887 /* fall back to MSI */
888 igb_clear_interrupt_scheme(adapter);
889 if (!pci_enable_msi(adapter->pdev))
890 adapter->flags |= IGB_FLAG_HAS_MSI;
891 igb_free_all_tx_resources(adapter);
892 igb_free_all_rx_resources(adapter);
893 adapter->num_tx_queues = 1;
894 adapter->num_rx_queues = 1;
895 adapter->num_q_vectors = 1;
896 err = igb_alloc_q_vectors(adapter);
899 "Unable to allocate memory for vectors\n");
902 err = igb_alloc_queues(adapter);
905 "Unable to allocate memory for queues\n");
906 igb_free_q_vectors(adapter);
909 igb_setup_all_tx_resources(adapter);
910 igb_setup_all_rx_resources(adapter);
912 switch (hw->mac.type) {
914 wr32(E1000_MSIXBM(0),
915 (E1000_EICR_RX_QUEUE0 |
916 E1000_EICR_TX_QUEUE0 |
921 wr32(E1000_IVAR0, E1000_IVAR_VALID);
928 if (adapter->flags & IGB_FLAG_HAS_MSI) {
929 err = request_irq(adapter->pdev->irq, igb_intr_msi, 0,
930 netdev->name, adapter);
934 /* fall back to legacy interrupts */
935 igb_reset_interrupt_capability(adapter);
936 adapter->flags &= ~IGB_FLAG_HAS_MSI;
939 err = request_irq(adapter->pdev->irq, igb_intr, IRQF_SHARED,
940 netdev->name, adapter);
943 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
950 static void igb_free_irq(struct igb_adapter *adapter)
952 if (adapter->msix_entries) {
955 free_irq(adapter->msix_entries[vector++].vector, adapter);
957 for (i = 0; i < adapter->num_q_vectors; i++) {
958 struct igb_q_vector *q_vector = adapter->q_vector[i];
959 free_irq(adapter->msix_entries[vector++].vector,
963 free_irq(adapter->pdev->irq, adapter);
968 * igb_irq_disable - Mask off interrupt generation on the NIC
969 * @adapter: board private structure
971 static void igb_irq_disable(struct igb_adapter *adapter)
973 struct e1000_hw *hw = &adapter->hw;
976 * we need to be careful when disabling interrupts. The VFs are also
977 * mapped into these registers and so clearing the bits can cause
978 * issues on the VF drivers so we only need to clear what we set
980 if (adapter->msix_entries) {
981 u32 regval = rd32(E1000_EIAM);
982 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
983 wr32(E1000_EIMC, adapter->eims_enable_mask);
984 regval = rd32(E1000_EIAC);
985 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
991 synchronize_irq(adapter->pdev->irq);
995 * igb_irq_enable - Enable default interrupt generation settings
996 * @adapter: board private structure
998 static void igb_irq_enable(struct igb_adapter *adapter)
1000 struct e1000_hw *hw = &adapter->hw;
1002 if (adapter->msix_entries) {
1003 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC;
1004 u32 regval = rd32(E1000_EIAC);
1005 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1006 regval = rd32(E1000_EIAM);
1007 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1008 wr32(E1000_EIMS, adapter->eims_enable_mask);
1009 if (adapter->vfs_allocated_count) {
1010 wr32(E1000_MBVFIMR, 0xFF);
1011 ims |= E1000_IMS_VMMB;
1013 if (adapter->hw.mac.type == e1000_82580)
1014 ims |= E1000_IMS_DRSTA;
1016 wr32(E1000_IMS, ims);
1018 wr32(E1000_IMS, IMS_ENABLE_MASK |
1020 wr32(E1000_IAM, IMS_ENABLE_MASK |
1025 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1027 struct e1000_hw *hw = &adapter->hw;
1028 u16 vid = adapter->hw.mng_cookie.vlan_id;
1029 u16 old_vid = adapter->mng_vlan_id;
1031 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1032 /* add VID to filter table */
1033 igb_vfta_set(hw, vid, true);
1034 adapter->mng_vlan_id = vid;
1036 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1039 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1041 !vlan_group_get_device(adapter->vlgrp, old_vid)) {
1042 /* remove VID from filter table */
1043 igb_vfta_set(hw, old_vid, false);
1048 * igb_release_hw_control - release control of the h/w to f/w
1049 * @adapter: address of board private structure
1051 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1052 * For ASF and Pass Through versions of f/w this means that the
1053 * driver is no longer loaded.
1056 static void igb_release_hw_control(struct igb_adapter *adapter)
1058 struct e1000_hw *hw = &adapter->hw;
1061 /* Let firmware take over control of h/w */
1062 ctrl_ext = rd32(E1000_CTRL_EXT);
1063 wr32(E1000_CTRL_EXT,
1064 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1068 * igb_get_hw_control - get control of the h/w from f/w
1069 * @adapter: address of board private structure
1071 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1072 * For ASF and Pass Through versions of f/w this means that
1073 * the driver is loaded.
1076 static void igb_get_hw_control(struct igb_adapter *adapter)
1078 struct e1000_hw *hw = &adapter->hw;
1081 /* Let firmware know the driver has taken over */
1082 ctrl_ext = rd32(E1000_CTRL_EXT);
1083 wr32(E1000_CTRL_EXT,
1084 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1088 * igb_configure - configure the hardware for RX and TX
1089 * @adapter: private board structure
1091 static void igb_configure(struct igb_adapter *adapter)
1093 struct net_device *netdev = adapter->netdev;
1096 igb_get_hw_control(adapter);
1097 igb_set_rx_mode(netdev);
1099 igb_restore_vlan(adapter);
1101 igb_setup_tctl(adapter);
1102 igb_setup_mrqc(adapter);
1103 igb_setup_rctl(adapter);
1105 igb_configure_tx(adapter);
1106 igb_configure_rx(adapter);
1108 igb_rx_fifo_flush_82575(&adapter->hw);
1110 /* call igb_desc_unused which always leaves
1111 * at least 1 descriptor unused to make sure
1112 * next_to_use != next_to_clean */
1113 for (i = 0; i < adapter->num_rx_queues; i++) {
1114 struct igb_ring *ring = &adapter->rx_ring[i];
1115 igb_alloc_rx_buffers_adv(ring, igb_desc_unused(ring));
1119 adapter->tx_queue_len = netdev->tx_queue_len;
1124 * igb_up - Open the interface and prepare it to handle traffic
1125 * @adapter: board private structure
1127 int igb_up(struct igb_adapter *adapter)
1129 struct e1000_hw *hw = &adapter->hw;
1132 /* hardware has been reset, we need to reload some things */
1133 igb_configure(adapter);
1135 clear_bit(__IGB_DOWN, &adapter->state);
1137 for (i = 0; i < adapter->num_q_vectors; i++) {
1138 struct igb_q_vector *q_vector = adapter->q_vector[i];
1139 napi_enable(&q_vector->napi);
1141 if (adapter->msix_entries)
1142 igb_configure_msix(adapter);
1144 /* Clear any pending interrupts. */
1146 igb_irq_enable(adapter);
1148 /* notify VFs that reset has been completed */
1149 if (adapter->vfs_allocated_count) {
1150 u32 reg_data = rd32(E1000_CTRL_EXT);
1151 reg_data |= E1000_CTRL_EXT_PFRSTD;
1152 wr32(E1000_CTRL_EXT, reg_data);
1155 netif_tx_start_all_queues(adapter->netdev);
1157 /* start the watchdog. */
1158 hw->mac.get_link_status = 1;
1159 schedule_work(&adapter->watchdog_task);
1164 void igb_down(struct igb_adapter *adapter)
1166 struct net_device *netdev = adapter->netdev;
1167 struct e1000_hw *hw = &adapter->hw;
1171 /* signal that we're down so the interrupt handler does not
1172 * reschedule our watchdog timer */
1173 set_bit(__IGB_DOWN, &adapter->state);
1175 /* disable receives in the hardware */
1176 rctl = rd32(E1000_RCTL);
1177 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1178 /* flush and sleep below */
1180 netif_tx_stop_all_queues(netdev);
1182 /* disable transmits in the hardware */
1183 tctl = rd32(E1000_TCTL);
1184 tctl &= ~E1000_TCTL_EN;
1185 wr32(E1000_TCTL, tctl);
1186 /* flush both disables and wait for them to finish */
1190 for (i = 0; i < adapter->num_q_vectors; i++) {
1191 struct igb_q_vector *q_vector = adapter->q_vector[i];
1192 napi_disable(&q_vector->napi);
1195 igb_irq_disable(adapter);
1197 del_timer_sync(&adapter->watchdog_timer);
1198 del_timer_sync(&adapter->phy_info_timer);
1200 netdev->tx_queue_len = adapter->tx_queue_len;
1201 netif_carrier_off(netdev);
1203 /* record the stats before reset*/
1204 igb_update_stats(adapter);
1206 adapter->link_speed = 0;
1207 adapter->link_duplex = 0;
1209 if (!pci_channel_offline(adapter->pdev))
1211 igb_clean_all_tx_rings(adapter);
1212 igb_clean_all_rx_rings(adapter);
1213 #ifdef CONFIG_IGB_DCA
1215 /* since we reset the hardware DCA settings were cleared */
1216 igb_setup_dca(adapter);
1220 void igb_reinit_locked(struct igb_adapter *adapter)
1222 WARN_ON(in_interrupt());
1223 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1227 clear_bit(__IGB_RESETTING, &adapter->state);
1230 void igb_reset(struct igb_adapter *adapter)
1232 struct pci_dev *pdev = adapter->pdev;
1233 struct e1000_hw *hw = &adapter->hw;
1234 struct e1000_mac_info *mac = &hw->mac;
1235 struct e1000_fc_info *fc = &hw->fc;
1236 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
1239 /* Repartition Pba for greater than 9k mtu
1240 * To take effect CTRL.RST is required.
1242 switch (mac->type) {
1244 pba = rd32(E1000_RXPBS);
1245 pba = igb_rxpbs_adjust_82580(pba);
1248 pba = rd32(E1000_RXPBS);
1249 pba &= E1000_RXPBS_SIZE_MASK_82576;
1253 pba = E1000_PBA_34K;
1257 if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1258 (mac->type < e1000_82576)) {
1259 /* adjust PBA for jumbo frames */
1260 wr32(E1000_PBA, pba);
1262 /* To maintain wire speed transmits, the Tx FIFO should be
1263 * large enough to accommodate two full transmit packets,
1264 * rounded up to the next 1KB and expressed in KB. Likewise,
1265 * the Rx FIFO should be large enough to accommodate at least
1266 * one full receive packet and is similarly rounded up and
1267 * expressed in KB. */
1268 pba = rd32(E1000_PBA);
1269 /* upper 16 bits has Tx packet buffer allocation size in KB */
1270 tx_space = pba >> 16;
1271 /* lower 16 bits has Rx packet buffer allocation size in KB */
1273 /* the tx fifo also stores 16 bytes of information about the tx
1274 * but don't include ethernet FCS because hardware appends it */
1275 min_tx_space = (adapter->max_frame_size +
1276 sizeof(union e1000_adv_tx_desc) -
1278 min_tx_space = ALIGN(min_tx_space, 1024);
1279 min_tx_space >>= 10;
1280 /* software strips receive CRC, so leave room for it */
1281 min_rx_space = adapter->max_frame_size;
1282 min_rx_space = ALIGN(min_rx_space, 1024);
1283 min_rx_space >>= 10;
1285 /* If current Tx allocation is less than the min Tx FIFO size,
1286 * and the min Tx FIFO size is less than the current Rx FIFO
1287 * allocation, take space away from current Rx allocation */
1288 if (tx_space < min_tx_space &&
1289 ((min_tx_space - tx_space) < pba)) {
1290 pba = pba - (min_tx_space - tx_space);
1292 /* if short on rx space, rx wins and must trump tx
1294 if (pba < min_rx_space)
1297 wr32(E1000_PBA, pba);
1300 /* flow control settings */
1301 /* The high water mark must be low enough to fit one full frame
1302 * (or the size used for early receive) above it in the Rx FIFO.
1303 * Set it to the lower of:
1304 * - 90% of the Rx FIFO size, or
1305 * - the full Rx FIFO size minus one full frame */
1306 hwm = min(((pba << 10) * 9 / 10),
1307 ((pba << 10) - 2 * adapter->max_frame_size));
1309 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1310 fc->low_water = fc->high_water - 16;
1311 fc->pause_time = 0xFFFF;
1313 fc->current_mode = fc->requested_mode;
1315 /* disable receive for all VFs and wait one second */
1316 if (adapter->vfs_allocated_count) {
1318 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1319 adapter->vf_data[i].flags = 0;
1321 /* ping all the active vfs to let them know we are going down */
1322 igb_ping_all_vfs(adapter);
1324 /* disable transmits and receives */
1325 wr32(E1000_VFRE, 0);
1326 wr32(E1000_VFTE, 0);
1329 /* Allow time for pending master requests to run */
1330 hw->mac.ops.reset_hw(hw);
1333 if (hw->mac.ops.init_hw(hw))
1334 dev_err(&pdev->dev, "Hardware Error\n");
1336 if (hw->mac.type == e1000_82580) {
1337 u32 reg = rd32(E1000_PCIEMISC);
1338 wr32(E1000_PCIEMISC,
1339 reg & ~E1000_PCIEMISC_LX_DECISION);
1341 igb_update_mng_vlan(adapter);
1343 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1344 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1346 igb_reset_adaptive(hw);
1347 igb_get_phy_info(hw);
1350 static const struct net_device_ops igb_netdev_ops = {
1351 .ndo_open = igb_open,
1352 .ndo_stop = igb_close,
1353 .ndo_start_xmit = igb_xmit_frame_adv,
1354 .ndo_get_stats = igb_get_stats,
1355 .ndo_set_rx_mode = igb_set_rx_mode,
1356 .ndo_set_multicast_list = igb_set_rx_mode,
1357 .ndo_set_mac_address = igb_set_mac,
1358 .ndo_change_mtu = igb_change_mtu,
1359 .ndo_do_ioctl = igb_ioctl,
1360 .ndo_tx_timeout = igb_tx_timeout,
1361 .ndo_validate_addr = eth_validate_addr,
1362 .ndo_vlan_rx_register = igb_vlan_rx_register,
1363 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid,
1364 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid,
1365 #ifdef CONFIG_NET_POLL_CONTROLLER
1366 .ndo_poll_controller = igb_netpoll,
1371 * igb_probe - Device Initialization Routine
1372 * @pdev: PCI device information struct
1373 * @ent: entry in igb_pci_tbl
1375 * Returns 0 on success, negative on failure
1377 * igb_probe initializes an adapter identified by a pci_dev structure.
1378 * The OS initialization, configuring of the adapter private structure,
1379 * and a hardware reset occur.
1381 static int __devinit igb_probe(struct pci_dev *pdev,
1382 const struct pci_device_id *ent)
1384 struct net_device *netdev;
1385 struct igb_adapter *adapter;
1386 struct e1000_hw *hw;
1387 u16 eeprom_data = 0;
1388 static int global_quad_port_a; /* global quad port a indication */
1389 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1390 unsigned long mmio_start, mmio_len;
1391 int err, pci_using_dac;
1392 u16 eeprom_apme_mask = IGB_EEPROM_APME;
1395 err = pci_enable_device_mem(pdev);
1400 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1402 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1406 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1408 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1410 dev_err(&pdev->dev, "No usable DMA "
1411 "configuration, aborting\n");
1417 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1423 pci_enable_pcie_error_reporting(pdev);
1425 pci_set_master(pdev);
1426 pci_save_state(pdev);
1429 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1430 IGB_ABS_MAX_TX_QUEUES);
1432 goto err_alloc_etherdev;
1434 SET_NETDEV_DEV(netdev, &pdev->dev);
1436 pci_set_drvdata(pdev, netdev);
1437 adapter = netdev_priv(netdev);
1438 adapter->netdev = netdev;
1439 adapter->pdev = pdev;
1442 adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1444 mmio_start = pci_resource_start(pdev, 0);
1445 mmio_len = pci_resource_len(pdev, 0);
1448 hw->hw_addr = ioremap(mmio_start, mmio_len);
1452 netdev->netdev_ops = &igb_netdev_ops;
1453 igb_set_ethtool_ops(netdev);
1454 netdev->watchdog_timeo = 5 * HZ;
1456 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1458 netdev->mem_start = mmio_start;
1459 netdev->mem_end = mmio_start + mmio_len;
1461 /* PCI config space info */
1462 hw->vendor_id = pdev->vendor;
1463 hw->device_id = pdev->device;
1464 hw->revision_id = pdev->revision;
1465 hw->subsystem_vendor_id = pdev->subsystem_vendor;
1466 hw->subsystem_device_id = pdev->subsystem_device;
1468 /* Copy the default MAC, PHY and NVM function pointers */
1469 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1470 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1471 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1472 /* Initialize skew-specific constants */
1473 err = ei->get_invariants(hw);
1477 /* setup the private structure */
1478 err = igb_sw_init(adapter);
1482 igb_get_bus_info_pcie(hw);
1484 hw->phy.autoneg_wait_to_complete = false;
1485 hw->mac.adaptive_ifs = true;
1487 /* Copper options */
1488 if (hw->phy.media_type == e1000_media_type_copper) {
1489 hw->phy.mdix = AUTO_ALL_MODES;
1490 hw->phy.disable_polarity_correction = false;
1491 hw->phy.ms_type = e1000_ms_hw_default;
1494 if (igb_check_reset_block(hw))
1495 dev_info(&pdev->dev,
1496 "PHY reset is blocked due to SOL/IDER session.\n");
1498 netdev->features = NETIF_F_SG |
1500 NETIF_F_HW_VLAN_TX |
1501 NETIF_F_HW_VLAN_RX |
1502 NETIF_F_HW_VLAN_FILTER;
1504 netdev->features |= NETIF_F_IPV6_CSUM;
1505 netdev->features |= NETIF_F_TSO;
1506 netdev->features |= NETIF_F_TSO6;
1507 netdev->features |= NETIF_F_GRO;
1509 netdev->vlan_features |= NETIF_F_TSO;
1510 netdev->vlan_features |= NETIF_F_TSO6;
1511 netdev->vlan_features |= NETIF_F_IP_CSUM;
1512 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
1513 netdev->vlan_features |= NETIF_F_SG;
1516 netdev->features |= NETIF_F_HIGHDMA;
1518 if (hw->mac.type >= e1000_82576)
1519 netdev->features |= NETIF_F_SCTP_CSUM;
1521 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
1523 /* before reading the NVM, reset the controller to put the device in a
1524 * known good starting state */
1525 hw->mac.ops.reset_hw(hw);
1527 /* make sure the NVM is good */
1528 if (igb_validate_nvm_checksum(hw) < 0) {
1529 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1534 /* copy the MAC address out of the NVM */
1535 if (hw->mac.ops.read_mac_addr(hw))
1536 dev_err(&pdev->dev, "NVM Read Error\n");
1538 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1539 memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1541 if (!is_valid_ether_addr(netdev->perm_addr)) {
1542 dev_err(&pdev->dev, "Invalid MAC Address\n");
1547 setup_timer(&adapter->watchdog_timer, &igb_watchdog,
1548 (unsigned long) adapter);
1549 setup_timer(&adapter->phy_info_timer, &igb_update_phy_info,
1550 (unsigned long) adapter);
1552 INIT_WORK(&adapter->reset_task, igb_reset_task);
1553 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1555 /* Initialize link properties that are user-changeable */
1556 adapter->fc_autoneg = true;
1557 hw->mac.autoneg = true;
1558 hw->phy.autoneg_advertised = 0x2f;
1560 hw->fc.requested_mode = e1000_fc_default;
1561 hw->fc.current_mode = e1000_fc_default;
1563 igb_validate_mdi_setting(hw);
1565 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1566 * enable the ACPI Magic Packet filter
1569 if (hw->bus.func == 0)
1570 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1571 else if (hw->mac.type == e1000_82580)
1572 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
1573 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
1575 else if (hw->bus.func == 1)
1576 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1578 if (eeprom_data & eeprom_apme_mask)
1579 adapter->eeprom_wol |= E1000_WUFC_MAG;
1581 /* now that we have the eeprom settings, apply the special cases where
1582 * the eeprom may be wrong or the board simply won't support wake on
1583 * lan on a particular port */
1584 switch (pdev->device) {
1585 case E1000_DEV_ID_82575GB_QUAD_COPPER:
1586 adapter->eeprom_wol = 0;
1588 case E1000_DEV_ID_82575EB_FIBER_SERDES:
1589 case E1000_DEV_ID_82576_FIBER:
1590 case E1000_DEV_ID_82576_SERDES:
1591 /* Wake events only supported on port A for dual fiber
1592 * regardless of eeprom setting */
1593 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1594 adapter->eeprom_wol = 0;
1596 case E1000_DEV_ID_82576_QUAD_COPPER:
1597 /* if quad port adapter, disable WoL on all but port A */
1598 if (global_quad_port_a != 0)
1599 adapter->eeprom_wol = 0;
1601 adapter->flags |= IGB_FLAG_QUAD_PORT_A;
1602 /* Reset for multiple quad port adapters */
1603 if (++global_quad_port_a == 4)
1604 global_quad_port_a = 0;
1608 /* initialize the wol settings based on the eeprom settings */
1609 adapter->wol = adapter->eeprom_wol;
1610 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1612 /* reset the hardware with the new settings */
1615 /* let the f/w know that the h/w is now under the control of the
1617 igb_get_hw_control(adapter);
1619 strcpy(netdev->name, "eth%d");
1620 err = register_netdev(netdev);
1624 /* carrier off reporting is important to ethtool even BEFORE open */
1625 netif_carrier_off(netdev);
1627 #ifdef CONFIG_IGB_DCA
1628 if (dca_add_requester(&pdev->dev) == 0) {
1629 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1630 dev_info(&pdev->dev, "DCA enabled\n");
1631 igb_setup_dca(adapter);
1635 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1636 /* print bus type/speed/width info */
1637 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
1639 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
1641 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
1642 (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
1643 (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
1647 igb_read_part_num(hw, &part_num);
1648 dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1649 (part_num >> 8), (part_num & 0xff));
1651 dev_info(&pdev->dev,
1652 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1653 adapter->msix_entries ? "MSI-X" :
1654 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1655 adapter->num_rx_queues, adapter->num_tx_queues);
1660 igb_release_hw_control(adapter);
1662 if (!igb_check_reset_block(hw))
1665 if (hw->flash_address)
1666 iounmap(hw->flash_address);
1668 igb_clear_interrupt_scheme(adapter);
1669 iounmap(hw->hw_addr);
1671 free_netdev(netdev);
1673 pci_release_selected_regions(pdev,
1674 pci_select_bars(pdev, IORESOURCE_MEM));
1677 pci_disable_device(pdev);
1682 * igb_remove - Device Removal Routine
1683 * @pdev: PCI device information struct
1685 * igb_remove is called by the PCI subsystem to alert the driver
1686 * that it should release a PCI device. The could be caused by a
1687 * Hot-Plug event, or because the driver is going to be removed from
1690 static void __devexit igb_remove(struct pci_dev *pdev)
1692 struct net_device *netdev = pci_get_drvdata(pdev);
1693 struct igb_adapter *adapter = netdev_priv(netdev);
1694 struct e1000_hw *hw = &adapter->hw;
1696 /* flush_scheduled work may reschedule our watchdog task, so
1697 * explicitly disable watchdog tasks from being rescheduled */
1698 set_bit(__IGB_DOWN, &adapter->state);
1699 del_timer_sync(&adapter->watchdog_timer);
1700 del_timer_sync(&adapter->phy_info_timer);
1702 flush_scheduled_work();
1704 #ifdef CONFIG_IGB_DCA
1705 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1706 dev_info(&pdev->dev, "DCA disabled\n");
1707 dca_remove_requester(&pdev->dev);
1708 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1709 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
1713 /* Release control of h/w to f/w. If f/w is AMT enabled, this
1714 * would have already happened in close and is redundant. */
1715 igb_release_hw_control(adapter);
1717 unregister_netdev(netdev);
1719 if (!igb_check_reset_block(hw))
1722 igb_clear_interrupt_scheme(adapter);
1724 #ifdef CONFIG_PCI_IOV
1725 /* reclaim resources allocated to VFs */
1726 if (adapter->vf_data) {
1727 /* disable iov and allow time for transactions to clear */
1728 pci_disable_sriov(pdev);
1731 kfree(adapter->vf_data);
1732 adapter->vf_data = NULL;
1733 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1735 dev_info(&pdev->dev, "IOV Disabled\n");
1739 iounmap(hw->hw_addr);
1740 if (hw->flash_address)
1741 iounmap(hw->flash_address);
1742 pci_release_selected_regions(pdev,
1743 pci_select_bars(pdev, IORESOURCE_MEM));
1745 free_netdev(netdev);
1747 pci_disable_pcie_error_reporting(pdev);
1749 pci_disable_device(pdev);
1753 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
1754 * @adapter: board private structure to initialize
1756 * This function initializes the vf specific data storage and then attempts to
1757 * allocate the VFs. The reason for ordering it this way is because it is much
1758 * mor expensive time wise to disable SR-IOV than it is to allocate and free
1759 * the memory for the VFs.
1761 static void __devinit igb_probe_vfs(struct igb_adapter * adapter)
1763 #ifdef CONFIG_PCI_IOV
1764 struct pci_dev *pdev = adapter->pdev;
1766 if (adapter->vfs_allocated_count > 7)
1767 adapter->vfs_allocated_count = 7;
1769 if (adapter->vfs_allocated_count) {
1770 adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
1771 sizeof(struct vf_data_storage),
1773 /* if allocation failed then we do not support SR-IOV */
1774 if (!adapter->vf_data) {
1775 adapter->vfs_allocated_count = 0;
1776 dev_err(&pdev->dev, "Unable to allocate memory for VF "
1781 if (pci_enable_sriov(pdev, adapter->vfs_allocated_count)) {
1782 kfree(adapter->vf_data);
1783 adapter->vf_data = NULL;
1784 #endif /* CONFIG_PCI_IOV */
1785 adapter->vfs_allocated_count = 0;
1786 #ifdef CONFIG_PCI_IOV
1788 unsigned char mac_addr[ETH_ALEN];
1790 dev_info(&pdev->dev, "%d vfs allocated\n",
1791 adapter->vfs_allocated_count);
1792 for (i = 0; i < adapter->vfs_allocated_count; i++) {
1793 random_ether_addr(mac_addr);
1794 igb_set_vf_mac(adapter, i, mac_addr);
1797 #endif /* CONFIG_PCI_IOV */
1802 * igb_init_hw_timer - Initialize hardware timer used with IEEE 1588 timestamp
1803 * @adapter: board private structure to initialize
1805 * igb_init_hw_timer initializes the function pointer and values for the hw
1806 * timer found in hardware.
1808 static void igb_init_hw_timer(struct igb_adapter *adapter)
1810 struct e1000_hw *hw = &adapter->hw;
1812 switch (hw->mac.type) {
1814 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1815 adapter->cycles.read = igb_read_clock;
1816 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1817 adapter->cycles.mult = 1;
1819 * The 82580 timesync updates the system timer every 8ns by 8ns
1820 * and the value cannot be shifted. Instead we need to shift
1821 * the registers to generate a 64bit timer value. As a result
1822 * SYSTIMR/L/H, TXSTMPL/H, RXSTMPL/H all have to be shifted by
1823 * 24 in order to generate a larger value for synchronization.
1825 adapter->cycles.shift = IGB_82580_TSYNC_SHIFT;
1826 /* disable system timer temporarily by setting bit 31 */
1827 wr32(E1000_TSAUXC, 0x80000000);
1830 /* Set registers so that rollover occurs soon to test this. */
1831 wr32(E1000_SYSTIMR, 0x00000000);
1832 wr32(E1000_SYSTIML, 0x80000000);
1833 wr32(E1000_SYSTIMH, 0x000000FF);
1836 /* enable system timer by clearing bit 31 */
1837 wr32(E1000_TSAUXC, 0x0);
1840 timecounter_init(&adapter->clock,
1842 ktime_to_ns(ktime_get_real()));
1844 * Synchronize our NIC clock against system wall clock. NIC
1845 * time stamp reading requires ~3us per sample, each sample
1846 * was pretty stable even under load => only require 10
1847 * samples for each offset comparison.
1849 memset(&adapter->compare, 0, sizeof(adapter->compare));
1850 adapter->compare.source = &adapter->clock;
1851 adapter->compare.target = ktime_get_real;
1852 adapter->compare.num_samples = 10;
1853 timecompare_update(&adapter->compare, 0);
1857 * Initialize hardware timer: we keep it running just in case
1858 * that some program needs it later on.
1860 memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1861 adapter->cycles.read = igb_read_clock;
1862 adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1863 adapter->cycles.mult = 1;
1865 * Scale the NIC clock cycle by a large factor so that
1866 * relatively small clock corrections can be added or
1867 * substracted at each clock tick. The drawbacks of a large
1868 * factor are a) that the clock register overflows more quickly
1869 * (not such a big deal) and b) that the increment per tick has
1870 * to fit into 24 bits. As a result we need to use a shift of
1871 * 19 so we can fit a value of 16 into the TIMINCA register.
1873 adapter->cycles.shift = IGB_82576_TSYNC_SHIFT;
1875 (1 << E1000_TIMINCA_16NS_SHIFT) |
1876 (16 << IGB_82576_TSYNC_SHIFT));
1878 /* Set registers so that rollover occurs soon to test this. */
1879 wr32(E1000_SYSTIML, 0x00000000);
1880 wr32(E1000_SYSTIMH, 0xFF800000);
1883 timecounter_init(&adapter->clock,
1885 ktime_to_ns(ktime_get_real()));
1887 * Synchronize our NIC clock against system wall clock. NIC
1888 * time stamp reading requires ~3us per sample, each sample
1889 * was pretty stable even under load => only require 10
1890 * samples for each offset comparison.
1892 memset(&adapter->compare, 0, sizeof(adapter->compare));
1893 adapter->compare.source = &adapter->clock;
1894 adapter->compare.target = ktime_get_real;
1895 adapter->compare.num_samples = 10;
1896 timecompare_update(&adapter->compare, 0);
1899 /* 82575 does not support timesync */
1907 * igb_sw_init - Initialize general software structures (struct igb_adapter)
1908 * @adapter: board private structure to initialize
1910 * igb_sw_init initializes the Adapter private data structure.
1911 * Fields are initialized based on PCI device information and
1912 * OS network device settings (MTU size).
1914 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1916 struct e1000_hw *hw = &adapter->hw;
1917 struct net_device *netdev = adapter->netdev;
1918 struct pci_dev *pdev = adapter->pdev;
1920 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1922 adapter->tx_ring_count = IGB_DEFAULT_TXD;
1923 adapter->rx_ring_count = IGB_DEFAULT_RXD;
1924 adapter->rx_itr_setting = IGB_DEFAULT_ITR;
1925 adapter->tx_itr_setting = IGB_DEFAULT_ITR;
1927 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1928 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1930 #ifdef CONFIG_PCI_IOV
1931 if (hw->mac.type == e1000_82576)
1932 adapter->vfs_allocated_count = max_vfs;
1934 #endif /* CONFIG_PCI_IOV */
1935 adapter->rss_queues = min_t(u32, IGB_MAX_RX_QUEUES, num_online_cpus());
1938 * if rss_queues > 4 or vfs are going to be allocated with rss_queues
1939 * then we should combine the queues into a queue pair in order to
1940 * conserve interrupts due to limited supply
1942 if ((adapter->rss_queues > 4) ||
1943 ((adapter->rss_queues > 1) && (adapter->vfs_allocated_count > 6)))
1944 adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1946 /* This call may decrease the number of queues */
1947 if (igb_init_interrupt_scheme(adapter)) {
1948 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1952 igb_init_hw_timer(adapter);
1953 igb_probe_vfs(adapter);
1955 /* Explicitly disable IRQ since the NIC can be in any state. */
1956 igb_irq_disable(adapter);
1958 set_bit(__IGB_DOWN, &adapter->state);
1963 * igb_open - Called when a network interface is made active
1964 * @netdev: network interface device structure
1966 * Returns 0 on success, negative value on failure
1968 * The open entry point is called when a network interface is made
1969 * active by the system (IFF_UP). At this point all resources needed
1970 * for transmit and receive operations are allocated, the interrupt
1971 * handler is registered with the OS, the watchdog timer is started,
1972 * and the stack is notified that the interface is ready.
1974 static int igb_open(struct net_device *netdev)
1976 struct igb_adapter *adapter = netdev_priv(netdev);
1977 struct e1000_hw *hw = &adapter->hw;
1981 /* disallow open during test */
1982 if (test_bit(__IGB_TESTING, &adapter->state))
1985 netif_carrier_off(netdev);
1987 /* allocate transmit descriptors */
1988 err = igb_setup_all_tx_resources(adapter);
1992 /* allocate receive descriptors */
1993 err = igb_setup_all_rx_resources(adapter);
1997 /* e1000_power_up_phy(adapter); */
1999 /* before we allocate an interrupt, we must be ready to handle it.
2000 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2001 * as soon as we call pci_request_irq, so we have to setup our
2002 * clean_rx handler before we do so. */
2003 igb_configure(adapter);
2005 err = igb_request_irq(adapter);
2009 /* From here on the code is the same as igb_up() */
2010 clear_bit(__IGB_DOWN, &adapter->state);
2012 for (i = 0; i < adapter->num_q_vectors; i++) {
2013 struct igb_q_vector *q_vector = adapter->q_vector[i];
2014 napi_enable(&q_vector->napi);
2017 /* Clear any pending interrupts. */
2020 igb_irq_enable(adapter);
2022 /* notify VFs that reset has been completed */
2023 if (adapter->vfs_allocated_count) {
2024 u32 reg_data = rd32(E1000_CTRL_EXT);
2025 reg_data |= E1000_CTRL_EXT_PFRSTD;
2026 wr32(E1000_CTRL_EXT, reg_data);
2029 netif_tx_start_all_queues(netdev);
2031 /* start the watchdog. */
2032 hw->mac.get_link_status = 1;
2033 schedule_work(&adapter->watchdog_task);
2038 igb_release_hw_control(adapter);
2039 /* e1000_power_down_phy(adapter); */
2040 igb_free_all_rx_resources(adapter);
2042 igb_free_all_tx_resources(adapter);
2050 * igb_close - Disables a network interface
2051 * @netdev: network interface device structure
2053 * Returns 0, this is not allowed to fail
2055 * The close entry point is called when an interface is de-activated
2056 * by the OS. The hardware is still under the driver's control, but
2057 * needs to be disabled. A global MAC reset is issued to stop the
2058 * hardware, and all transmit and receive resources are freed.
2060 static int igb_close(struct net_device *netdev)
2062 struct igb_adapter *adapter = netdev_priv(netdev);
2064 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
2067 igb_free_irq(adapter);
2069 igb_free_all_tx_resources(adapter);
2070 igb_free_all_rx_resources(adapter);
2076 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2077 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2079 * Return 0 on success, negative on failure
2081 int igb_setup_tx_resources(struct igb_ring *tx_ring)
2083 struct pci_dev *pdev = tx_ring->pdev;
2086 size = sizeof(struct igb_buffer) * tx_ring->count;
2087 tx_ring->buffer_info = vmalloc(size);
2088 if (!tx_ring->buffer_info)
2090 memset(tx_ring->buffer_info, 0, size);
2092 /* round up to nearest 4K */
2093 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
2094 tx_ring->size = ALIGN(tx_ring->size, 4096);
2096 tx_ring->desc = pci_alloc_consistent(pdev,
2103 tx_ring->next_to_use = 0;
2104 tx_ring->next_to_clean = 0;
2108 vfree(tx_ring->buffer_info);
2110 "Unable to allocate memory for the transmit descriptor ring\n");
2115 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2116 * (Descriptors) for all queues
2117 * @adapter: board private structure
2119 * Return 0 on success, negative on failure
2121 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2123 struct pci_dev *pdev = adapter->pdev;
2126 for (i = 0; i < adapter->num_tx_queues; i++) {
2127 err = igb_setup_tx_resources(&adapter->tx_ring[i]);
2130 "Allocation for Tx Queue %u failed\n", i);
2131 for (i--; i >= 0; i--)
2132 igb_free_tx_resources(&adapter->tx_ring[i]);
2137 for (i = 0; i < IGB_ABS_MAX_TX_QUEUES; i++) {
2138 int r_idx = i % adapter->num_tx_queues;
2139 adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
2145 * igb_setup_tctl - configure the transmit control registers
2146 * @adapter: Board private structure
2148 void igb_setup_tctl(struct igb_adapter *adapter)
2150 struct e1000_hw *hw = &adapter->hw;
2153 /* disable queue 0 which is enabled by default on 82575 and 82576 */
2154 wr32(E1000_TXDCTL(0), 0);
2156 /* Program the Transmit Control Register */
2157 tctl = rd32(E1000_TCTL);
2158 tctl &= ~E1000_TCTL_CT;
2159 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2160 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2162 igb_config_collision_dist(hw);
2164 /* Enable transmits */
2165 tctl |= E1000_TCTL_EN;
2167 wr32(E1000_TCTL, tctl);
2171 * igb_configure_tx_ring - Configure transmit ring after Reset
2172 * @adapter: board private structure
2173 * @ring: tx ring to configure
2175 * Configure a transmit ring after a reset.
2177 void igb_configure_tx_ring(struct igb_adapter *adapter,
2178 struct igb_ring *ring)
2180 struct e1000_hw *hw = &adapter->hw;
2182 u64 tdba = ring->dma;
2183 int reg_idx = ring->reg_idx;
2185 /* disable the queue */
2186 txdctl = rd32(E1000_TXDCTL(reg_idx));
2187 wr32(E1000_TXDCTL(reg_idx),
2188 txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
2192 wr32(E1000_TDLEN(reg_idx),
2193 ring->count * sizeof(union e1000_adv_tx_desc));
2194 wr32(E1000_TDBAL(reg_idx),
2195 tdba & 0x00000000ffffffffULL);
2196 wr32(E1000_TDBAH(reg_idx), tdba >> 32);
2198 ring->head = hw->hw_addr + E1000_TDH(reg_idx);
2199 ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
2200 writel(0, ring->head);
2201 writel(0, ring->tail);
2203 txdctl |= IGB_TX_PTHRESH;
2204 txdctl |= IGB_TX_HTHRESH << 8;
2205 txdctl |= IGB_TX_WTHRESH << 16;
2207 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
2208 wr32(E1000_TXDCTL(reg_idx), txdctl);
2212 * igb_configure_tx - Configure transmit Unit after Reset
2213 * @adapter: board private structure
2215 * Configure the Tx unit of the MAC after a reset.
2217 static void igb_configure_tx(struct igb_adapter *adapter)
2221 for (i = 0; i < adapter->num_tx_queues; i++)
2222 igb_configure_tx_ring(adapter, &adapter->tx_ring[i]);
2226 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
2227 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2229 * Returns 0 on success, negative on failure
2231 int igb_setup_rx_resources(struct igb_ring *rx_ring)
2233 struct pci_dev *pdev = rx_ring->pdev;
2236 size = sizeof(struct igb_buffer) * rx_ring->count;
2237 rx_ring->buffer_info = vmalloc(size);
2238 if (!rx_ring->buffer_info)
2240 memset(rx_ring->buffer_info, 0, size);
2242 desc_len = sizeof(union e1000_adv_rx_desc);
2244 /* Round up to nearest 4K */
2245 rx_ring->size = rx_ring->count * desc_len;
2246 rx_ring->size = ALIGN(rx_ring->size, 4096);
2248 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
2254 rx_ring->next_to_clean = 0;
2255 rx_ring->next_to_use = 0;
2260 vfree(rx_ring->buffer_info);
2261 rx_ring->buffer_info = NULL;
2262 dev_err(&pdev->dev, "Unable to allocate memory for "
2263 "the receive descriptor ring\n");
2268 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
2269 * (Descriptors) for all queues
2270 * @adapter: board private structure
2272 * Return 0 on success, negative on failure
2274 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
2276 struct pci_dev *pdev = adapter->pdev;
2279 for (i = 0; i < adapter->num_rx_queues; i++) {
2280 err = igb_setup_rx_resources(&adapter->rx_ring[i]);
2283 "Allocation for Rx Queue %u failed\n", i);
2284 for (i--; i >= 0; i--)
2285 igb_free_rx_resources(&adapter->rx_ring[i]);
2294 * igb_setup_mrqc - configure the multiple receive queue control registers
2295 * @adapter: Board private structure
2297 static void igb_setup_mrqc(struct igb_adapter *adapter)
2299 struct e1000_hw *hw = &adapter->hw;
2301 u32 j, num_rx_queues, shift = 0, shift2 = 0;
2306 static const u8 rsshash[40] = {
2307 0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2, 0x41, 0x67,
2308 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0, 0xd0, 0xca, 0x2b, 0xcb,
2309 0xae, 0x7b, 0x30, 0xb4, 0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30,
2310 0xf2, 0x0c, 0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa };
2312 /* Fill out hash function seeds */
2313 for (j = 0; j < 10; j++) {
2314 u32 rsskey = rsshash[(j * 4)];
2315 rsskey |= rsshash[(j * 4) + 1] << 8;
2316 rsskey |= rsshash[(j * 4) + 2] << 16;
2317 rsskey |= rsshash[(j * 4) + 3] << 24;
2318 array_wr32(E1000_RSSRK(0), j, rsskey);
2321 num_rx_queues = adapter->rss_queues;
2323 if (adapter->vfs_allocated_count) {
2324 /* 82575 and 82576 supports 2 RSS queues for VMDq */
2325 switch (hw->mac.type) {
2341 if (hw->mac.type == e1000_82575)
2345 for (j = 0; j < (32 * 4); j++) {
2346 reta.bytes[j & 3] = (j % num_rx_queues) << shift;
2348 reta.bytes[j & 3] |= num_rx_queues << shift2;
2350 wr32(E1000_RETA(j >> 2), reta.dword);
2354 * Disable raw packet checksumming so that RSS hash is placed in
2355 * descriptor on writeback. No need to enable TCP/UDP/IP checksum
2356 * offloads as they are enabled by default
2358 rxcsum = rd32(E1000_RXCSUM);
2359 rxcsum |= E1000_RXCSUM_PCSD;
2361 if (adapter->hw.mac.type >= e1000_82576)
2362 /* Enable Receive Checksum Offload for SCTP */
2363 rxcsum |= E1000_RXCSUM_CRCOFL;
2365 /* Don't need to set TUOFL or IPOFL, they default to 1 */
2366 wr32(E1000_RXCSUM, rxcsum);
2368 /* If VMDq is enabled then we set the appropriate mode for that, else
2369 * we default to RSS so that an RSS hash is calculated per packet even
2370 * if we are only using one queue */
2371 if (adapter->vfs_allocated_count) {
2372 if (hw->mac.type > e1000_82575) {
2373 /* Set the default pool for the PF's first queue */
2374 u32 vtctl = rd32(E1000_VT_CTL);
2375 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2376 E1000_VT_CTL_DISABLE_DEF_POOL);
2377 vtctl |= adapter->vfs_allocated_count <<
2378 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2379 wr32(E1000_VT_CTL, vtctl);
2381 if (adapter->rss_queues > 1)
2382 mrqc = E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2384 mrqc = E1000_MRQC_ENABLE_VMDQ;
2386 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2388 igb_vmm_control(adapter);
2390 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2391 E1000_MRQC_RSS_FIELD_IPV4_TCP);
2392 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
2393 E1000_MRQC_RSS_FIELD_IPV6_TCP);
2394 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
2395 E1000_MRQC_RSS_FIELD_IPV6_UDP);
2396 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2397 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2399 wr32(E1000_MRQC, mrqc);
2403 * igb_setup_rctl - configure the receive control registers
2404 * @adapter: Board private structure
2406 void igb_setup_rctl(struct igb_adapter *adapter)
2408 struct e1000_hw *hw = &adapter->hw;
2411 rctl = rd32(E1000_RCTL);
2413 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2414 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2416 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
2417 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2420 * enable stripping of CRC. It's unlikely this will break BMC
2421 * redirection as it did with e1000. Newer features require
2422 * that the HW strips the CRC.
2424 rctl |= E1000_RCTL_SECRC;
2426 /* disable store bad packets and clear size bits. */
2427 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
2429 /* enable LPE to prevent packets larger than max_frame_size */
2430 rctl |= E1000_RCTL_LPE;
2432 /* disable queue 0 to prevent tail write w/o re-config */
2433 wr32(E1000_RXDCTL(0), 0);
2435 /* Attention!!! For SR-IOV PF driver operations you must enable
2436 * queue drop for all VF and PF queues to prevent head of line blocking
2437 * if an un-trusted VF does not provide descriptors to hardware.
2439 if (adapter->vfs_allocated_count) {
2440 /* set all queue drop enable bits */
2441 wr32(E1000_QDE, ALL_QUEUES);
2444 wr32(E1000_RCTL, rctl);
2447 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
2450 struct e1000_hw *hw = &adapter->hw;
2453 /* if it isn't the PF check to see if VFs are enabled and
2454 * increase the size to support vlan tags */
2455 if (vfn < adapter->vfs_allocated_count &&
2456 adapter->vf_data[vfn].vlans_enabled)
2457 size += VLAN_TAG_SIZE;
2459 vmolr = rd32(E1000_VMOLR(vfn));
2460 vmolr &= ~E1000_VMOLR_RLPML_MASK;
2461 vmolr |= size | E1000_VMOLR_LPE;
2462 wr32(E1000_VMOLR(vfn), vmolr);
2468 * igb_rlpml_set - set maximum receive packet size
2469 * @adapter: board private structure
2471 * Configure maximum receivable packet size.
2473 static void igb_rlpml_set(struct igb_adapter *adapter)
2475 u32 max_frame_size = adapter->max_frame_size;
2476 struct e1000_hw *hw = &adapter->hw;
2477 u16 pf_id = adapter->vfs_allocated_count;
2480 max_frame_size += VLAN_TAG_SIZE;
2482 /* if vfs are enabled we set RLPML to the largest possible request
2483 * size and set the VMOLR RLPML to the size we need */
2485 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
2486 max_frame_size = MAX_JUMBO_FRAME_SIZE;
2489 wr32(E1000_RLPML, max_frame_size);
2492 static inline void igb_set_vmolr(struct igb_adapter *adapter, int vfn)
2494 struct e1000_hw *hw = &adapter->hw;
2498 * This register exists only on 82576 and newer so if we are older then
2499 * we should exit and do nothing
2501 if (hw->mac.type < e1000_82576)
2504 vmolr = rd32(E1000_VMOLR(vfn));
2505 vmolr |= E1000_VMOLR_AUPE | /* Accept untagged packets */
2506 E1000_VMOLR_STRVLAN; /* Strip vlan tags */
2508 /* clear all bits that might not be set */
2509 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
2511 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
2512 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
2514 * for VMDq only allow the VFs and pool 0 to accept broadcast and
2517 if (vfn <= adapter->vfs_allocated_count)
2518 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
2520 wr32(E1000_VMOLR(vfn), vmolr);
2524 * igb_configure_rx_ring - Configure a receive ring after Reset
2525 * @adapter: board private structure
2526 * @ring: receive ring to be configured
2528 * Configure the Rx unit of the MAC after a reset.
2530 void igb_configure_rx_ring(struct igb_adapter *adapter,
2531 struct igb_ring *ring)
2533 struct e1000_hw *hw = &adapter->hw;
2534 u64 rdba = ring->dma;
2535 int reg_idx = ring->reg_idx;
2538 /* disable the queue */
2539 rxdctl = rd32(E1000_RXDCTL(reg_idx));
2540 wr32(E1000_RXDCTL(reg_idx),
2541 rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
2543 /* Set DMA base address registers */
2544 wr32(E1000_RDBAL(reg_idx),
2545 rdba & 0x00000000ffffffffULL);
2546 wr32(E1000_RDBAH(reg_idx), rdba >> 32);
2547 wr32(E1000_RDLEN(reg_idx),
2548 ring->count * sizeof(union e1000_adv_rx_desc));
2550 /* initialize head and tail */
2551 ring->head = hw->hw_addr + E1000_RDH(reg_idx);
2552 ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
2553 writel(0, ring->head);
2554 writel(0, ring->tail);
2556 /* set descriptor configuration */
2557 if (ring->rx_buffer_len < IGB_RXBUFFER_1024) {
2558 srrctl = ALIGN(ring->rx_buffer_len, 64) <<
2559 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2560 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
2561 srrctl |= IGB_RXBUFFER_16384 >>
2562 E1000_SRRCTL_BSIZEPKT_SHIFT;
2564 srrctl |= (PAGE_SIZE / 2) >>
2565 E1000_SRRCTL_BSIZEPKT_SHIFT;
2567 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2569 srrctl = ALIGN(ring->rx_buffer_len, 1024) >>
2570 E1000_SRRCTL_BSIZEPKT_SHIFT;
2571 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2574 wr32(E1000_SRRCTL(reg_idx), srrctl);
2576 /* set filtering for VMDQ pools */
2577 igb_set_vmolr(adapter, reg_idx & 0x7);
2579 /* enable receive descriptor fetching */
2580 rxdctl = rd32(E1000_RXDCTL(reg_idx));
2581 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2582 rxdctl &= 0xFFF00000;
2583 rxdctl |= IGB_RX_PTHRESH;
2584 rxdctl |= IGB_RX_HTHRESH << 8;
2585 rxdctl |= IGB_RX_WTHRESH << 16;
2586 wr32(E1000_RXDCTL(reg_idx), rxdctl);
2590 * igb_configure_rx - Configure receive Unit after Reset
2591 * @adapter: board private structure
2593 * Configure the Rx unit of the MAC after a reset.
2595 static void igb_configure_rx(struct igb_adapter *adapter)
2599 /* set UTA to appropriate mode */
2600 igb_set_uta(adapter);
2602 /* set the correct pool for the PF default MAC address in entry 0 */
2603 igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
2604 adapter->vfs_allocated_count);
2606 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2607 * the Base and Length of the Rx Descriptor Ring */
2608 for (i = 0; i < adapter->num_rx_queues; i++)
2609 igb_configure_rx_ring(adapter, &adapter->rx_ring[i]);
2613 * igb_free_tx_resources - Free Tx Resources per Queue
2614 * @tx_ring: Tx descriptor ring for a specific queue
2616 * Free all transmit software resources
2618 void igb_free_tx_resources(struct igb_ring *tx_ring)
2620 igb_clean_tx_ring(tx_ring);
2622 vfree(tx_ring->buffer_info);
2623 tx_ring->buffer_info = NULL;
2625 /* if not set, then don't free */
2629 pci_free_consistent(tx_ring->pdev, tx_ring->size,
2630 tx_ring->desc, tx_ring->dma);
2632 tx_ring->desc = NULL;
2636 * igb_free_all_tx_resources - Free Tx Resources for All Queues
2637 * @adapter: board private structure
2639 * Free all transmit software resources
2641 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2645 for (i = 0; i < adapter->num_tx_queues; i++)
2646 igb_free_tx_resources(&adapter->tx_ring[i]);
2649 void igb_unmap_and_free_tx_resource(struct igb_ring *tx_ring,
2650 struct igb_buffer *buffer_info)
2652 if (buffer_info->dma) {
2653 if (buffer_info->mapped_as_page)
2654 pci_unmap_page(tx_ring->pdev,
2656 buffer_info->length,
2659 pci_unmap_single(tx_ring->pdev,
2661 buffer_info->length,
2663 buffer_info->dma = 0;
2665 if (buffer_info->skb) {
2666 dev_kfree_skb_any(buffer_info->skb);
2667 buffer_info->skb = NULL;
2669 buffer_info->time_stamp = 0;
2670 buffer_info->length = 0;
2671 buffer_info->next_to_watch = 0;
2672 buffer_info->mapped_as_page = false;
2676 * igb_clean_tx_ring - Free Tx Buffers
2677 * @tx_ring: ring to be cleaned
2679 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2681 struct igb_buffer *buffer_info;
2685 if (!tx_ring->buffer_info)
2687 /* Free all the Tx ring sk_buffs */
2689 for (i = 0; i < tx_ring->count; i++) {
2690 buffer_info = &tx_ring->buffer_info[i];
2691 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
2694 size = sizeof(struct igb_buffer) * tx_ring->count;
2695 memset(tx_ring->buffer_info, 0, size);
2697 /* Zero out the descriptor ring */
2698 memset(tx_ring->desc, 0, tx_ring->size);
2700 tx_ring->next_to_use = 0;
2701 tx_ring->next_to_clean = 0;
2705 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2706 * @adapter: board private structure
2708 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2712 for (i = 0; i < adapter->num_tx_queues; i++)
2713 igb_clean_tx_ring(&adapter->tx_ring[i]);
2717 * igb_free_rx_resources - Free Rx Resources
2718 * @rx_ring: ring to clean the resources from
2720 * Free all receive software resources
2722 void igb_free_rx_resources(struct igb_ring *rx_ring)
2724 igb_clean_rx_ring(rx_ring);
2726 vfree(rx_ring->buffer_info);
2727 rx_ring->buffer_info = NULL;
2729 /* if not set, then don't free */
2733 pci_free_consistent(rx_ring->pdev, rx_ring->size,
2734 rx_ring->desc, rx_ring->dma);
2736 rx_ring->desc = NULL;
2740 * igb_free_all_rx_resources - Free Rx Resources for All Queues
2741 * @adapter: board private structure
2743 * Free all receive software resources
2745 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2749 for (i = 0; i < adapter->num_rx_queues; i++)
2750 igb_free_rx_resources(&adapter->rx_ring[i]);
2754 * igb_clean_rx_ring - Free Rx Buffers per Queue
2755 * @rx_ring: ring to free buffers from
2757 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2759 struct igb_buffer *buffer_info;
2763 if (!rx_ring->buffer_info)
2766 /* Free all the Rx ring sk_buffs */
2767 for (i = 0; i < rx_ring->count; i++) {
2768 buffer_info = &rx_ring->buffer_info[i];
2769 if (buffer_info->dma) {
2770 pci_unmap_single(rx_ring->pdev,
2772 rx_ring->rx_buffer_len,
2773 PCI_DMA_FROMDEVICE);
2774 buffer_info->dma = 0;
2777 if (buffer_info->skb) {
2778 dev_kfree_skb(buffer_info->skb);
2779 buffer_info->skb = NULL;
2781 if (buffer_info->page_dma) {
2782 pci_unmap_page(rx_ring->pdev,
2783 buffer_info->page_dma,
2785 PCI_DMA_FROMDEVICE);
2786 buffer_info->page_dma = 0;
2788 if (buffer_info->page) {
2789 put_page(buffer_info->page);
2790 buffer_info->page = NULL;
2791 buffer_info->page_offset = 0;
2795 size = sizeof(struct igb_buffer) * rx_ring->count;
2796 memset(rx_ring->buffer_info, 0, size);
2798 /* Zero out the descriptor ring */
2799 memset(rx_ring->desc, 0, rx_ring->size);
2801 rx_ring->next_to_clean = 0;
2802 rx_ring->next_to_use = 0;
2806 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2807 * @adapter: board private structure
2809 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2813 for (i = 0; i < adapter->num_rx_queues; i++)
2814 igb_clean_rx_ring(&adapter->rx_ring[i]);
2818 * igb_set_mac - Change the Ethernet Address of the NIC
2819 * @netdev: network interface device structure
2820 * @p: pointer to an address structure
2822 * Returns 0 on success, negative on failure
2824 static int igb_set_mac(struct net_device *netdev, void *p)
2826 struct igb_adapter *adapter = netdev_priv(netdev);
2827 struct e1000_hw *hw = &adapter->hw;
2828 struct sockaddr *addr = p;
2830 if (!is_valid_ether_addr(addr->sa_data))
2831 return -EADDRNOTAVAIL;
2833 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2834 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
2836 /* set the correct pool for the new PF MAC address in entry 0 */
2837 igb_rar_set_qsel(adapter, hw->mac.addr, 0,
2838 adapter->vfs_allocated_count);
2844 * igb_write_mc_addr_list - write multicast addresses to MTA
2845 * @netdev: network interface device structure
2847 * Writes multicast address list to the MTA hash table.
2848 * Returns: -ENOMEM on failure
2849 * 0 on no addresses written
2850 * X on writing X addresses to MTA
2852 static int igb_write_mc_addr_list(struct net_device *netdev)
2854 struct igb_adapter *adapter = netdev_priv(netdev);
2855 struct e1000_hw *hw = &adapter->hw;
2856 struct dev_mc_list *mc_ptr = netdev->mc_list;
2861 if (!netdev->mc_count) {
2862 /* nothing to program, so clear mc list */
2863 igb_update_mc_addr_list(hw, NULL, 0);
2864 igb_restore_vf_multicasts(adapter);
2868 mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2872 /* set vmolr receive overflow multicast bit */
2873 vmolr |= E1000_VMOLR_ROMPE;
2875 /* The shared function expects a packed array of only addresses. */
2876 mc_ptr = netdev->mc_list;
2878 for (i = 0; i < netdev->mc_count; i++) {
2881 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2882 mc_ptr = mc_ptr->next;
2884 igb_update_mc_addr_list(hw, mta_list, i);
2887 return netdev->mc_count;
2891 * igb_write_uc_addr_list - write unicast addresses to RAR table
2892 * @netdev: network interface device structure
2894 * Writes unicast address list to the RAR table.
2895 * Returns: -ENOMEM on failure/insufficient address space
2896 * 0 on no addresses written
2897 * X on writing X addresses to the RAR table
2899 static int igb_write_uc_addr_list(struct net_device *netdev)
2901 struct igb_adapter *adapter = netdev_priv(netdev);
2902 struct e1000_hw *hw = &adapter->hw;
2903 unsigned int vfn = adapter->vfs_allocated_count;
2904 unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
2907 /* return ENOMEM indicating insufficient memory for addresses */
2908 if (netdev->uc.count > rar_entries)
2911 if (netdev->uc.count && rar_entries) {
2912 struct netdev_hw_addr *ha;
2913 list_for_each_entry(ha, &netdev->uc.list, list) {
2916 igb_rar_set_qsel(adapter, ha->addr,
2922 /* write the addresses in reverse order to avoid write combining */
2923 for (; rar_entries > 0 ; rar_entries--) {
2924 wr32(E1000_RAH(rar_entries), 0);
2925 wr32(E1000_RAL(rar_entries), 0);
2933 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2934 * @netdev: network interface device structure
2936 * The set_rx_mode entry point is called whenever the unicast or multicast
2937 * address lists or the network interface flags are updated. This routine is
2938 * responsible for configuring the hardware for proper unicast, multicast,
2939 * promiscuous mode, and all-multi behavior.
2941 static void igb_set_rx_mode(struct net_device *netdev)
2943 struct igb_adapter *adapter = netdev_priv(netdev);
2944 struct e1000_hw *hw = &adapter->hw;
2945 unsigned int vfn = adapter->vfs_allocated_count;
2946 u32 rctl, vmolr = 0;
2949 /* Check for Promiscuous and All Multicast modes */
2950 rctl = rd32(E1000_RCTL);
2952 /* clear the effected bits */
2953 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
2955 if (netdev->flags & IFF_PROMISC) {
2956 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2957 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
2959 if (netdev->flags & IFF_ALLMULTI) {
2960 rctl |= E1000_RCTL_MPE;
2961 vmolr |= E1000_VMOLR_MPME;
2964 * Write addresses to the MTA, if the attempt fails
2965 * then we should just turn on promiscous mode so
2966 * that we can at least receive multicast traffic
2968 count = igb_write_mc_addr_list(netdev);
2970 rctl |= E1000_RCTL_MPE;
2971 vmolr |= E1000_VMOLR_MPME;
2973 vmolr |= E1000_VMOLR_ROMPE;
2977 * Write addresses to available RAR registers, if there is not
2978 * sufficient space to store all the addresses then enable
2979 * unicast promiscous mode
2981 count = igb_write_uc_addr_list(netdev);
2983 rctl |= E1000_RCTL_UPE;
2984 vmolr |= E1000_VMOLR_ROPE;
2986 rctl |= E1000_RCTL_VFE;
2988 wr32(E1000_RCTL, rctl);
2991 * In order to support SR-IOV and eventually VMDq it is necessary to set
2992 * the VMOLR to enable the appropriate modes. Without this workaround
2993 * we will have issues with VLAN tag stripping not being done for frames
2994 * that are only arriving because we are the default pool
2996 if (hw->mac.type < e1000_82576)
2999 vmolr |= rd32(E1000_VMOLR(vfn)) &
3000 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3001 wr32(E1000_VMOLR(vfn), vmolr);
3002 igb_restore_vf_multicasts(adapter);
3005 /* Need to wait a few seconds after link up to get diagnostic information from
3007 static void igb_update_phy_info(unsigned long data)
3009 struct igb_adapter *adapter = (struct igb_adapter *) data;
3010 igb_get_phy_info(&adapter->hw);
3014 * igb_has_link - check shared code for link and determine up/down
3015 * @adapter: pointer to driver private info
3017 static bool igb_has_link(struct igb_adapter *adapter)
3019 struct e1000_hw *hw = &adapter->hw;
3020 bool link_active = false;
3023 /* get_link_status is set on LSC (link status) interrupt or
3024 * rx sequence error interrupt. get_link_status will stay
3025 * false until the e1000_check_for_link establishes link
3026 * for copper adapters ONLY
3028 switch (hw->phy.media_type) {
3029 case e1000_media_type_copper:
3030 if (hw->mac.get_link_status) {
3031 ret_val = hw->mac.ops.check_for_link(hw);
3032 link_active = !hw->mac.get_link_status;
3037 case e1000_media_type_internal_serdes:
3038 ret_val = hw->mac.ops.check_for_link(hw);
3039 link_active = hw->mac.serdes_has_link;
3042 case e1000_media_type_unknown:
3050 * igb_watchdog - Timer Call-back
3051 * @data: pointer to adapter cast into an unsigned long
3053 static void igb_watchdog(unsigned long data)
3055 struct igb_adapter *adapter = (struct igb_adapter *)data;
3056 /* Do the rest outside of interrupt context */
3057 schedule_work(&adapter->watchdog_task);
3060 static void igb_watchdog_task(struct work_struct *work)
3062 struct igb_adapter *adapter = container_of(work,
3065 struct e1000_hw *hw = &adapter->hw;
3066 struct net_device *netdev = adapter->netdev;
3070 link = igb_has_link(adapter);
3072 if (!netif_carrier_ok(netdev)) {
3074 hw->mac.ops.get_speed_and_duplex(hw,
3075 &adapter->link_speed,
3076 &adapter->link_duplex);
3078 ctrl = rd32(E1000_CTRL);
3079 /* Links status message must follow this format */
3080 printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
3081 "Flow Control: %s\n",
3083 adapter->link_speed,
3084 adapter->link_duplex == FULL_DUPLEX ?
3085 "Full Duplex" : "Half Duplex",
3086 ((ctrl & E1000_CTRL_TFCE) &&
3087 (ctrl & E1000_CTRL_RFCE)) ? "RX/TX" :
3088 ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3089 ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None")));
3091 /* tweak tx_queue_len according to speed/duplex and
3092 * adjust the timeout factor */
3093 netdev->tx_queue_len = adapter->tx_queue_len;
3094 adapter->tx_timeout_factor = 1;
3095 switch (adapter->link_speed) {
3097 netdev->tx_queue_len = 10;
3098 adapter->tx_timeout_factor = 14;
3101 netdev->tx_queue_len = 100;
3102 /* maybe add some timeout factor ? */
3106 netif_carrier_on(netdev);
3108 igb_ping_all_vfs(adapter);
3110 /* link state has changed, schedule phy info update */
3111 if (!test_bit(__IGB_DOWN, &adapter->state))
3112 mod_timer(&adapter->phy_info_timer,
3113 round_jiffies(jiffies + 2 * HZ));
3116 if (netif_carrier_ok(netdev)) {
3117 adapter->link_speed = 0;
3118 adapter->link_duplex = 0;
3119 /* Links status message must follow this format */
3120 printk(KERN_INFO "igb: %s NIC Link is Down\n",
3122 netif_carrier_off(netdev);
3124 igb_ping_all_vfs(adapter);
3126 /* link state has changed, schedule phy info update */
3127 if (!test_bit(__IGB_DOWN, &adapter->state))
3128 mod_timer(&adapter->phy_info_timer,
3129 round_jiffies(jiffies + 2 * HZ));
3133 igb_update_stats(adapter);
3134 igb_update_adaptive(hw);
3136 for (i = 0; i < adapter->num_tx_queues; i++) {
3137 struct igb_ring *tx_ring = &adapter->tx_ring[i];
3138 if (!netif_carrier_ok(netdev)) {
3139 /* We've lost link, so the controller stops DMA,
3140 * but we've got queued Tx work that's never going
3141 * to get done, so reset controller to flush Tx.
3142 * (Do the reset outside of interrupt context). */
3143 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
3144 adapter->tx_timeout_count++;
3145 schedule_work(&adapter->reset_task);
3146 /* return immediately since reset is imminent */
3151 /* Force detection of hung controller every watchdog period */
3152 tx_ring->detect_tx_hung = true;
3155 /* Cause software interrupt to ensure rx ring is cleaned */
3156 if (adapter->msix_entries) {
3158 for (i = 0; i < adapter->num_q_vectors; i++) {
3159 struct igb_q_vector *q_vector = adapter->q_vector[i];
3160 eics |= q_vector->eims_value;
3162 wr32(E1000_EICS, eics);
3164 wr32(E1000_ICS, E1000_ICS_RXDMT0);
3167 /* Reset the timer */
3168 if (!test_bit(__IGB_DOWN, &adapter->state))
3169 mod_timer(&adapter->watchdog_timer,
3170 round_jiffies(jiffies + 2 * HZ));
3173 enum latency_range {
3177 latency_invalid = 255
3181 * igb_update_ring_itr - update the dynamic ITR value based on packet size
3183 * Stores a new ITR value based on strictly on packet size. This
3184 * algorithm is less sophisticated than that used in igb_update_itr,
3185 * due to the difficulty of synchronizing statistics across multiple
3186 * receive rings. The divisors and thresholds used by this fuction
3187 * were determined based on theoretical maximum wire speed and testing
3188 * data, in order to minimize response time while increasing bulk
3190 * This functionality is controlled by the InterruptThrottleRate module
3191 * parameter (see igb_param.c)
3192 * NOTE: This function is called only when operating in a multiqueue
3193 * receive environment.
3194 * @q_vector: pointer to q_vector
3196 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
3198 int new_val = q_vector->itr_val;
3199 int avg_wire_size = 0;
3200 struct igb_adapter *adapter = q_vector->adapter;
3202 /* For non-gigabit speeds, just fix the interrupt rate at 4000
3203 * ints/sec - ITR timer value of 120 ticks.
3205 if (adapter->link_speed != SPEED_1000) {
3210 if (q_vector->rx_ring && q_vector->rx_ring->total_packets) {
3211 struct igb_ring *ring = q_vector->rx_ring;
3212 avg_wire_size = ring->total_bytes / ring->total_packets;
3215 if (q_vector->tx_ring && q_vector->tx_ring->total_packets) {
3216 struct igb_ring *ring = q_vector->tx_ring;
3217 avg_wire_size = max_t(u32, avg_wire_size,
3218 (ring->total_bytes /
3219 ring->total_packets));
3222 /* if avg_wire_size isn't set no work was done */
3226 /* Add 24 bytes to size to account for CRC, preamble, and gap */
3227 avg_wire_size += 24;
3229 /* Don't starve jumbo frames */
3230 avg_wire_size = min(avg_wire_size, 3000);
3232 /* Give a little boost to mid-size frames */
3233 if ((avg_wire_size > 300) && (avg_wire_size < 1200))
3234 new_val = avg_wire_size / 3;
3236 new_val = avg_wire_size / 2;
3239 if (new_val != q_vector->itr_val) {
3240 q_vector->itr_val = new_val;
3241 q_vector->set_itr = 1;
3244 if (q_vector->rx_ring) {
3245 q_vector->rx_ring->total_bytes = 0;
3246 q_vector->rx_ring->total_packets = 0;
3248 if (q_vector->tx_ring) {
3249 q_vector->tx_ring->total_bytes = 0;
3250 q_vector->tx_ring->total_packets = 0;
3255 * igb_update_itr - update the dynamic ITR value based on statistics
3256 * Stores a new ITR value based on packets and byte
3257 * counts during the last interrupt. The advantage of per interrupt
3258 * computation is faster updates and more accurate ITR for the current
3259 * traffic pattern. Constants in this function were computed
3260 * based on theoretical maximum wire speed and thresholds were set based
3261 * on testing data as well as attempting to minimize response time
3262 * while increasing bulk throughput.
3263 * this functionality is controlled by the InterruptThrottleRate module
3264 * parameter (see igb_param.c)
3265 * NOTE: These calculations are only valid when operating in a single-
3266 * queue environment.
3267 * @adapter: pointer to adapter
3268 * @itr_setting: current q_vector->itr_val
3269 * @packets: the number of packets during this measurement interval
3270 * @bytes: the number of bytes during this measurement interval
3272 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
3273 int packets, int bytes)
3275 unsigned int retval = itr_setting;
3278 goto update_itr_done;
3280 switch (itr_setting) {
3281 case lowest_latency:
3282 /* handle TSO and jumbo frames */
3283 if (bytes/packets > 8000)
3284 retval = bulk_latency;
3285 else if ((packets < 5) && (bytes > 512))
3286 retval = low_latency;
3288 case low_latency: /* 50 usec aka 20000 ints/s */
3289 if (bytes > 10000) {
3290 /* this if handles the TSO accounting */
3291 if (bytes/packets > 8000) {
3292 retval = bulk_latency;
3293 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
3294 retval = bulk_latency;
3295 } else if ((packets > 35)) {
3296 retval = lowest_latency;
3298 } else if (bytes/packets > 2000) {
3299 retval = bulk_latency;
3300 } else if (packets <= 2 && bytes < 512) {
3301 retval = lowest_latency;
3304 case bulk_latency: /* 250 usec aka 4000 ints/s */
3305 if (bytes > 25000) {
3307 retval = low_latency;
3308 } else if (bytes < 1500) {
3309 retval = low_latency;
3318 static void igb_set_itr(struct igb_adapter *adapter)
3320 struct igb_q_vector *q_vector = adapter->q_vector[0];
3322 u32 new_itr = q_vector->itr_val;
3324 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
3325 if (adapter->link_speed != SPEED_1000) {
3331 adapter->rx_itr = igb_update_itr(adapter,
3333 adapter->rx_ring->total_packets,
3334 adapter->rx_ring->total_bytes);
3336 adapter->tx_itr = igb_update_itr(adapter,
3338 adapter->tx_ring->total_packets,
3339 adapter->tx_ring->total_bytes);
3340 current_itr = max(adapter->rx_itr, adapter->tx_itr);
3342 /* conservative mode (itr 3) eliminates the lowest_latency setting */
3343 if (adapter->rx_itr_setting == 3 && current_itr == lowest_latency)
3344 current_itr = low_latency;
3346 switch (current_itr) {
3347 /* counts and packets in update_itr are dependent on these numbers */
3348 case lowest_latency:
3349 new_itr = 56; /* aka 70,000 ints/sec */
3352 new_itr = 196; /* aka 20,000 ints/sec */
3355 new_itr = 980; /* aka 4,000 ints/sec */
3362 adapter->rx_ring->total_bytes = 0;
3363 adapter->rx_ring->total_packets = 0;
3364 adapter->tx_ring->total_bytes = 0;
3365 adapter->tx_ring->total_packets = 0;
3367 if (new_itr != q_vector->itr_val) {
3368 /* this attempts to bias the interrupt rate towards Bulk
3369 * by adding intermediate steps when interrupt rate is
3371 new_itr = new_itr > q_vector->itr_val ?
3372 max((new_itr * q_vector->itr_val) /
3373 (new_itr + (q_vector->itr_val >> 2)),
3376 /* Don't write the value here; it resets the adapter's
3377 * internal timer, and causes us to delay far longer than
3378 * we should between interrupts. Instead, we write the ITR
3379 * value at the beginning of the next interrupt so the timing
3380 * ends up being correct.
3382 q_vector->itr_val = new_itr;
3383 q_vector->set_itr = 1;
3389 #define IGB_TX_FLAGS_CSUM 0x00000001
3390 #define IGB_TX_FLAGS_VLAN 0x00000002
3391 #define IGB_TX_FLAGS_TSO 0x00000004
3392 #define IGB_TX_FLAGS_IPV4 0x00000008
3393 #define IGB_TX_FLAGS_TSTAMP 0x00000010
3394 #define IGB_TX_FLAGS_VLAN_MASK 0xffff0000
3395 #define IGB_TX_FLAGS_VLAN_SHIFT 16
3397 static inline int igb_tso_adv(struct igb_ring *tx_ring,
3398 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
3400 struct e1000_adv_tx_context_desc *context_desc;
3403 struct igb_buffer *buffer_info;
3404 u32 info = 0, tu_cmd = 0;
3405 u32 mss_l4len_idx, l4len;
3408 if (skb_header_cloned(skb)) {
3409 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3414 l4len = tcp_hdrlen(skb);
3417 if (skb->protocol == htons(ETH_P_IP)) {
3418 struct iphdr *iph = ip_hdr(skb);
3421 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3425 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3426 ipv6_hdr(skb)->payload_len = 0;
3427 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3428 &ipv6_hdr(skb)->daddr,
3432 i = tx_ring->next_to_use;
3434 buffer_info = &tx_ring->buffer_info[i];
3435 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3436 /* VLAN MACLEN IPLEN */
3437 if (tx_flags & IGB_TX_FLAGS_VLAN)
3438 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3439 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3440 *hdr_len += skb_network_offset(skb);
3441 info |= skb_network_header_len(skb);
3442 *hdr_len += skb_network_header_len(skb);
3443 context_desc->vlan_macip_lens = cpu_to_le32(info);
3445 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3446 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3448 if (skb->protocol == htons(ETH_P_IP))
3449 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3450 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3452 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3455 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
3456 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
3458 /* For 82575, context index must be unique per ring. */
3459 if (tx_ring->flags & IGB_RING_FLAG_TX_CTX_IDX)
3460 mss_l4len_idx |= tx_ring->reg_idx << 4;
3462 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3463 context_desc->seqnum_seed = 0;
3465 buffer_info->time_stamp = jiffies;
3466 buffer_info->next_to_watch = i;
3467 buffer_info->dma = 0;
3469 if (i == tx_ring->count)
3472 tx_ring->next_to_use = i;
3477 static inline bool igb_tx_csum_adv(struct igb_ring *tx_ring,
3478 struct sk_buff *skb, u32 tx_flags)
3480 struct e1000_adv_tx_context_desc *context_desc;
3481 struct pci_dev *pdev = tx_ring->pdev;
3482 struct igb_buffer *buffer_info;
3483 u32 info = 0, tu_cmd = 0;
3486 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
3487 (tx_flags & IGB_TX_FLAGS_VLAN)) {
3488 i = tx_ring->next_to_use;
3489 buffer_info = &tx_ring->buffer_info[i];
3490 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3492 if (tx_flags & IGB_TX_FLAGS_VLAN)
3493 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3495 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3496 if (skb->ip_summed == CHECKSUM_PARTIAL)
3497 info |= skb_network_header_len(skb);
3499 context_desc->vlan_macip_lens = cpu_to_le32(info);
3501 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3503 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3506 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3507 const struct vlan_ethhdr *vhdr =
3508 (const struct vlan_ethhdr*)skb->data;
3510 protocol = vhdr->h_vlan_encapsulated_proto;
3512 protocol = skb->protocol;
3516 case cpu_to_be16(ETH_P_IP):
3517 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3518 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3519 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3520 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
3521 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
3523 case cpu_to_be16(ETH_P_IPV6):
3524 /* XXX what about other V6 headers?? */
3525 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3526 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3527 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
3528 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
3531 if (unlikely(net_ratelimit()))
3532 dev_warn(&pdev->dev,
3533 "partial checksum but proto=%x!\n",
3539 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3540 context_desc->seqnum_seed = 0;
3541 if (tx_ring->flags & IGB_RING_FLAG_TX_CTX_IDX)
3542 context_desc->mss_l4len_idx =
3543 cpu_to_le32(tx_ring->reg_idx << 4);
3545 buffer_info->time_stamp = jiffies;
3546 buffer_info->next_to_watch = i;
3547 buffer_info->dma = 0;
3550 if (i == tx_ring->count)
3552 tx_ring->next_to_use = i;
3559 #define IGB_MAX_TXD_PWR 16
3560 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
3562 static inline int igb_tx_map_adv(struct igb_ring *tx_ring, struct sk_buff *skb,
3565 struct igb_buffer *buffer_info;
3566 struct pci_dev *pdev = tx_ring->pdev;
3567 unsigned int len = skb_headlen(skb);
3568 unsigned int count = 0, i;
3571 i = tx_ring->next_to_use;
3573 buffer_info = &tx_ring->buffer_info[i];
3574 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3575 buffer_info->length = len;
3576 /* set time_stamp *before* dma to help avoid a possible race */
3577 buffer_info->time_stamp = jiffies;
3578 buffer_info->next_to_watch = i;
3579 buffer_info->dma = pci_map_single(pdev, skb->data, len,
3581 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3584 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
3585 struct skb_frag_struct *frag;
3588 if (i == tx_ring->count)
3591 frag = &skb_shinfo(skb)->frags[f];
3594 buffer_info = &tx_ring->buffer_info[i];
3595 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3596 buffer_info->length = len;
3597 buffer_info->time_stamp = jiffies;
3598 buffer_info->next_to_watch = i;
3599 buffer_info->mapped_as_page = true;
3600 buffer_info->dma = pci_map_page(pdev,
3605 if (pci_dma_mapping_error(pdev, buffer_info->dma))
3611 tx_ring->buffer_info[i].skb = skb;
3612 tx_ring->buffer_info[first].next_to_watch = i;
3617 dev_err(&pdev->dev, "TX DMA map failed\n");
3619 /* clear timestamp and dma mappings for failed buffer_info mapping */
3620 buffer_info->dma = 0;
3621 buffer_info->time_stamp = 0;
3622 buffer_info->length = 0;
3623 buffer_info->next_to_watch = 0;
3624 buffer_info->mapped_as_page = false;
3627 /* clear timestamp and dma mappings for remaining portion of packet */
3628 while (count >= 0) {
3632 i += tx_ring->count;
3633 buffer_info = &tx_ring->buffer_info[i];
3634 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3640 static inline void igb_tx_queue_adv(struct igb_ring *tx_ring,
3641 int tx_flags, int count, u32 paylen,
3644 union e1000_adv_tx_desc *tx_desc;
3645 struct igb_buffer *buffer_info;
3646 u32 olinfo_status = 0, cmd_type_len;
3647 unsigned int i = tx_ring->next_to_use;
3649 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
3650 E1000_ADVTXD_DCMD_DEXT);
3652 if (tx_flags & IGB_TX_FLAGS_VLAN)
3653 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3655 if (tx_flags & IGB_TX_FLAGS_TSTAMP)
3656 cmd_type_len |= E1000_ADVTXD_MAC_TSTAMP;
3658 if (tx_flags & IGB_TX_FLAGS_TSO) {
3659 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3661 /* insert tcp checksum */
3662 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3664 /* insert ip checksum */
3665 if (tx_flags & IGB_TX_FLAGS_IPV4)
3666 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3668 } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
3669 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3672 if ((tx_ring->flags & IGB_RING_FLAG_TX_CTX_IDX) &&
3673 (tx_flags & (IGB_TX_FLAGS_CSUM |
3675 IGB_TX_FLAGS_VLAN)))
3676 olinfo_status |= tx_ring->reg_idx << 4;
3678 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
3681 buffer_info = &tx_ring->buffer_info[i];
3682 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
3683 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
3684 tx_desc->read.cmd_type_len =
3685 cpu_to_le32(cmd_type_len | buffer_info->length);
3686 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3689 if (i == tx_ring->count)
3691 } while (count > 0);
3693 tx_desc->read.cmd_type_len |= cpu_to_le32(IGB_ADVTXD_DCMD);
3694 /* Force memory writes to complete before letting h/w
3695 * know there are new descriptors to fetch. (Only
3696 * applicable for weak-ordered memory model archs,
3697 * such as IA-64). */
3700 tx_ring->next_to_use = i;
3701 writel(i, tx_ring->tail);
3702 /* we need this if more than one processor can write to our tail
3703 * at a time, it syncronizes IO on IA64/Altix systems */
3707 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, int size)
3709 struct net_device *netdev = tx_ring->netdev;
3711 netif_stop_subqueue(netdev, tx_ring->queue_index);
3713 /* Herbert's original patch had:
3714 * smp_mb__after_netif_stop_queue();
3715 * but since that doesn't exist yet, just open code it. */
3718 /* We need to check again in a case another CPU has just
3719 * made room available. */
3720 if (igb_desc_unused(tx_ring) < size)
3724 netif_wake_subqueue(netdev, tx_ring->queue_index);
3725 tx_ring->tx_stats.restart_queue++;
3729 static int igb_maybe_stop_tx(struct igb_ring *tx_ring, int size)
3731 if (igb_desc_unused(tx_ring) >= size)
3733 return __igb_maybe_stop_tx(tx_ring, size);
3736 netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *skb,
3737 struct igb_ring *tx_ring)
3739 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
3741 unsigned int tx_flags = 0;
3744 union skb_shared_tx *shtx = skb_tx(skb);
3746 /* need: 1 descriptor per page,
3747 * + 2 desc gap to keep tail from touching head,
3748 * + 1 desc for skb->data,
3749 * + 1 desc for context descriptor,
3750 * otherwise try next time */
3751 if (igb_maybe_stop_tx(tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
3752 /* this is a hard error */
3753 return NETDEV_TX_BUSY;
3756 if (unlikely(shtx->hardware)) {
3757 shtx->in_progress = 1;
3758 tx_flags |= IGB_TX_FLAGS_TSTAMP;
3761 if (vlan_tx_tag_present(skb) && adapter->vlgrp) {
3762 tx_flags |= IGB_TX_FLAGS_VLAN;
3763 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
3766 if (skb->protocol == htons(ETH_P_IP))
3767 tx_flags |= IGB_TX_FLAGS_IPV4;
3769 first = tx_ring->next_to_use;
3770 if (skb_is_gso(skb)) {
3771 tso = igb_tso_adv(tx_ring, skb, tx_flags, &hdr_len);
3774 dev_kfree_skb_any(skb);
3775 return NETDEV_TX_OK;
3780 tx_flags |= IGB_TX_FLAGS_TSO;
3781 else if (igb_tx_csum_adv(tx_ring, skb, tx_flags) &&
3782 (skb->ip_summed == CHECKSUM_PARTIAL))
3783 tx_flags |= IGB_TX_FLAGS_CSUM;
3786 * count reflects descriptors mapped, if 0 or less then mapping error
3787 * has occured and we need to rewind the descriptor queue
3789 count = igb_tx_map_adv(tx_ring, skb, first);
3791 dev_kfree_skb_any(skb);
3792 tx_ring->buffer_info[first].time_stamp = 0;
3793 tx_ring->next_to_use = first;
3794 return NETDEV_TX_OK;
3797 igb_tx_queue_adv(tx_ring, tx_flags, count, skb->len, hdr_len);
3799 /* Make sure there is space in the ring for the next send. */
3800 igb_maybe_stop_tx(tx_ring, MAX_SKB_FRAGS + 4);
3802 return NETDEV_TX_OK;
3805 static netdev_tx_t igb_xmit_frame_adv(struct sk_buff *skb,
3806 struct net_device *netdev)
3808 struct igb_adapter *adapter = netdev_priv(netdev);
3809 struct igb_ring *tx_ring;
3812 if (test_bit(__IGB_DOWN, &adapter->state)) {
3813 dev_kfree_skb_any(skb);
3814 return NETDEV_TX_OK;
3817 if (skb->len <= 0) {
3818 dev_kfree_skb_any(skb);
3819 return NETDEV_TX_OK;
3822 r_idx = skb->queue_mapping & (IGB_ABS_MAX_TX_QUEUES - 1);
3823 tx_ring = adapter->multi_tx_table[r_idx];
3825 /* This goes back to the question of how to logically map a tx queue
3826 * to a flow. Right now, performance is impacted slightly negatively
3827 * if using multiple tx queues. If the stack breaks away from a
3828 * single qdisc implementation, we can look at this again. */
3829 return igb_xmit_frame_ring_adv(skb, tx_ring);
3833 * igb_tx_timeout - Respond to a Tx Hang
3834 * @netdev: network interface device structure
3836 static void igb_tx_timeout(struct net_device *netdev)
3838 struct igb_adapter *adapter = netdev_priv(netdev);
3839 struct e1000_hw *hw = &adapter->hw;
3841 /* Do the reset outside of interrupt context */
3842 adapter->tx_timeout_count++;
3844 if (hw->mac.type == e1000_82580)
3845 hw->dev_spec._82575.global_device_reset = true;
3847 schedule_work(&adapter->reset_task);
3849 (adapter->eims_enable_mask & ~adapter->eims_other));
3852 static void igb_reset_task(struct work_struct *work)
3854 struct igb_adapter *adapter;
3855 adapter = container_of(work, struct igb_adapter, reset_task);
3857 igb_reinit_locked(adapter);
3861 * igb_get_stats - Get System Network Statistics
3862 * @netdev: network interface device structure
3864 * Returns the address of the device statistics structure.
3865 * The statistics are actually updated from the timer callback.
3867 static struct net_device_stats *igb_get_stats(struct net_device *netdev)
3869 /* only return the current stats */
3870 return &netdev->stats;
3874 * igb_change_mtu - Change the Maximum Transfer Unit
3875 * @netdev: network interface device structure
3876 * @new_mtu: new value for maximum frame size
3878 * Returns 0 on success, negative on failure
3880 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3882 struct igb_adapter *adapter = netdev_priv(netdev);
3883 struct pci_dev *pdev = adapter->pdev;
3884 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3885 u32 rx_buffer_len, i;
3887 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3888 dev_err(&pdev->dev, "Invalid MTU setting\n");
3892 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3893 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
3897 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3900 /* igb_down has a dependency on max_frame_size */
3901 adapter->max_frame_size = max_frame;
3903 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3904 * means we reserve 2 more, this pushes us to allocate from the next
3906 * i.e. RXBUFFER_2048 --> size-4096 slab
3909 if (max_frame <= IGB_RXBUFFER_1024)
3910 rx_buffer_len = IGB_RXBUFFER_1024;
3911 else if (max_frame <= MAXIMUM_ETHERNET_VLAN_SIZE)
3912 rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3914 rx_buffer_len = IGB_RXBUFFER_128;
3916 if (netif_running(netdev))
3919 dev_info(&pdev->dev, "changing MTU from %d to %d\n",
3920 netdev->mtu, new_mtu);
3921 netdev->mtu = new_mtu;
3923 for (i = 0; i < adapter->num_rx_queues; i++)
3924 adapter->rx_ring[i].rx_buffer_len = rx_buffer_len;
3926 if (netif_running(netdev))
3931 clear_bit(__IGB_RESETTING, &adapter->state);
3937 * igb_update_stats - Update the board statistics counters
3938 * @adapter: board private structure
3941 void igb_update_stats(struct igb_adapter *adapter)
3943 struct net_device_stats *net_stats = igb_get_stats(adapter->netdev);
3944 struct e1000_hw *hw = &adapter->hw;
3945 struct pci_dev *pdev = adapter->pdev;
3951 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3954 * Prevent stats update while adapter is being reset, or if the pci
3955 * connection is down.
3957 if (adapter->link_speed == 0)
3959 if (pci_channel_offline(pdev))
3964 for (i = 0; i < adapter->num_rx_queues; i++) {
3965 u32 rqdpc_tmp = rd32(E1000_RQDPC(i)) & 0x0FFF;
3966 adapter->rx_ring[i].rx_stats.drops += rqdpc_tmp;
3967 net_stats->rx_fifo_errors += rqdpc_tmp;
3968 bytes += adapter->rx_ring[i].rx_stats.bytes;
3969 packets += adapter->rx_ring[i].rx_stats.packets;
3972 net_stats->rx_bytes = bytes;
3973 net_stats->rx_packets = packets;
3977 for (i = 0; i < adapter->num_tx_queues; i++) {
3978 bytes += adapter->tx_ring[i].tx_stats.bytes;
3979 packets += adapter->tx_ring[i].tx_stats.packets;
3981 net_stats->tx_bytes = bytes;
3982 net_stats->tx_packets = packets;
3984 /* read stats registers */
3985 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
3986 adapter->stats.gprc += rd32(E1000_GPRC);
3987 adapter->stats.gorc += rd32(E1000_GORCL);
3988 rd32(E1000_GORCH); /* clear GORCL */
3989 adapter->stats.bprc += rd32(E1000_BPRC);
3990 adapter->stats.mprc += rd32(E1000_MPRC);
3991 adapter->stats.roc += rd32(E1000_ROC);
3993 adapter->stats.prc64 += rd32(E1000_PRC64);
3994 adapter->stats.prc127 += rd32(E1000_PRC127);
3995 adapter->stats.prc255 += rd32(E1000_PRC255);
3996 adapter->stats.prc511 += rd32(E1000_PRC511);
3997 adapter->stats.prc1023 += rd32(E1000_PRC1023);
3998 adapter->stats.prc1522 += rd32(E1000_PRC1522);
3999 adapter->stats.symerrs += rd32(E1000_SYMERRS);
4000 adapter->stats.sec += rd32(E1000_SEC);
4002 adapter->stats.mpc += rd32(E1000_MPC);
4003 adapter->stats.scc += rd32(E1000_SCC);
4004 adapter->stats.ecol += rd32(E1000_ECOL);
4005 adapter->stats.mcc += rd32(E1000_MCC);
4006 adapter->stats.latecol += rd32(E1000_LATECOL);
4007 adapter->stats.dc += rd32(E1000_DC);
4008 adapter->stats.rlec += rd32(E1000_RLEC);
4009 adapter->stats.xonrxc += rd32(E1000_XONRXC);
4010 adapter->stats.xontxc += rd32(E1000_XONTXC);
4011 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
4012 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
4013 adapter->stats.fcruc += rd32(E1000_FCRUC);
4014 adapter->stats.gptc += rd32(E1000_GPTC);
4015 adapter->stats.gotc += rd32(E1000_GOTCL);
4016 rd32(E1000_GOTCH); /* clear GOTCL */
4017 rnbc = rd32(E1000_RNBC);
4018 adapter->stats.rnbc += rnbc;
4019 net_stats->rx_fifo_errors += rnbc;
4020 adapter->stats.ruc += rd32(E1000_RUC);
4021 adapter->stats.rfc += rd32(E1000_RFC);
4022 adapter->stats.rjc += rd32(E1000_RJC);
4023 adapter->stats.tor += rd32(E1000_TORH);
4024 adapter->stats.tot += rd32(E1000_TOTH);
4025 adapter->stats.tpr += rd32(E1000_TPR);
4027 adapter->stats.ptc64 += rd32(E1000_PTC64);
4028 adapter->stats.ptc127 += rd32(E1000_PTC127);
4029 adapter->stats.ptc255 += rd32(E1000_PTC255);
4030 adapter->stats.ptc511 += rd32(E1000_PTC511);
4031 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
4032 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
4034 adapter->stats.mptc += rd32(E1000_MPTC);
4035 adapter->stats.bptc += rd32(E1000_BPTC);
4037 /* used for adaptive IFS */
4038 hw->mac.tx_packet_delta = rd32(E1000_TPT);
4039 adapter->stats.tpt += hw->mac.tx_packet_delta;
4040 hw->mac.collision_delta = rd32(E1000_COLC);
4041 adapter->stats.colc += hw->mac.collision_delta;
4043 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
4044 adapter->stats.rxerrc += rd32(E1000_RXERRC);
4045 adapter->stats.tncrs += rd32(E1000_TNCRS);
4046 adapter->stats.tsctc += rd32(E1000_TSCTC);
4047 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
4049 adapter->stats.iac += rd32(E1000_IAC);
4050 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
4051 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
4052 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
4053 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
4054 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
4055 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
4056 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
4057 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
4059 /* Fill out the OS statistics structure */
4060 net_stats->multicast = adapter->stats.mprc;
4061 net_stats->collisions = adapter->stats.colc;
4065 /* RLEC on some newer hardware can be incorrect so build
4066 * our own version based on RUC and ROC */
4067 net_stats->rx_errors = adapter->stats.rxerrc +
4068 adapter->stats.crcerrs + adapter->stats.algnerrc +
4069 adapter->stats.ruc + adapter->stats.roc +
4070 adapter->stats.cexterr;
4071 net_stats->rx_length_errors = adapter->stats.ruc +
4073 net_stats->rx_crc_errors = adapter->stats.crcerrs;
4074 net_stats->rx_frame_errors = adapter->stats.algnerrc;
4075 net_stats->rx_missed_errors = adapter->stats.mpc;
4078 net_stats->tx_errors = adapter->stats.ecol +
4079 adapter->stats.latecol;
4080 net_stats->tx_aborted_errors = adapter->stats.ecol;
4081 net_stats->tx_window_errors = adapter->stats.latecol;
4082 net_stats->tx_carrier_errors = adapter->stats.tncrs;
4084 /* Tx Dropped needs to be maintained elsewhere */
4087 if (hw->phy.media_type == e1000_media_type_copper) {
4088 if ((adapter->link_speed == SPEED_1000) &&
4089 (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
4090 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
4091 adapter->phy_stats.idle_errors += phy_tmp;
4095 /* Management Stats */
4096 adapter->stats.mgptc += rd32(E1000_MGTPTC);
4097 adapter->stats.mgprc += rd32(E1000_MGTPRC);
4098 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
4101 static irqreturn_t igb_msix_other(int irq, void *data)
4103 struct igb_adapter *adapter = data;
4104 struct e1000_hw *hw = &adapter->hw;
4105 u32 icr = rd32(E1000_ICR);
4106 /* reading ICR causes bit 31 of EICR to be cleared */
4108 if (icr & E1000_ICR_DOUTSYNC) {
4109 /* HW is reporting DMA is out of sync */
4110 adapter->stats.doosync++;
4113 /* Check for a mailbox event */
4114 if (icr & E1000_ICR_VMMB)
4115 igb_msg_task(adapter);
4117 if (icr & E1000_ICR_LSC) {
4118 hw->mac.get_link_status = 1;
4119 /* guard against interrupt when we're going down */
4120 if (!test_bit(__IGB_DOWN, &adapter->state))
4121 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4124 if (adapter->vfs_allocated_count)
4125 wr32(E1000_IMS, E1000_IMS_LSC |
4127 E1000_IMS_DOUTSYNC);
4129 wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC);
4130 wr32(E1000_EIMS, adapter->eims_other);
4135 static void igb_write_itr(struct igb_q_vector *q_vector)
4137 u32 itr_val = q_vector->itr_val & 0x7FFC;
4139 if (!q_vector->set_itr)
4145 if (q_vector->itr_shift)
4146 itr_val |= itr_val << q_vector->itr_shift;
4148 itr_val |= 0x8000000;
4150 writel(itr_val, q_vector->itr_register);
4151 q_vector->set_itr = 0;
4154 static irqreturn_t igb_msix_ring(int irq, void *data)
4156 struct igb_q_vector *q_vector = data;
4158 /* Write the ITR value calculated from the previous interrupt. */
4159 igb_write_itr(q_vector);
4161 napi_schedule(&q_vector->napi);
4166 #ifdef CONFIG_IGB_DCA
4167 static void igb_update_dca(struct igb_q_vector *q_vector)
4169 struct igb_adapter *adapter = q_vector->adapter;
4170 struct e1000_hw *hw = &adapter->hw;
4171 int cpu = get_cpu();
4173 if (q_vector->cpu == cpu)
4176 if (q_vector->tx_ring) {
4177 int q = q_vector->tx_ring->reg_idx;
4178 u32 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
4179 if (hw->mac.type == e1000_82575) {
4180 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
4181 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4183 dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
4184 dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4185 E1000_DCA_TXCTRL_CPUID_SHIFT;
4187 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
4188 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
4190 if (q_vector->rx_ring) {
4191 int q = q_vector->rx_ring->reg_idx;
4192 u32 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
4193 if (hw->mac.type == e1000_82575) {
4194 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
4195 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
4197 dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
4198 dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
4199 E1000_DCA_RXCTRL_CPUID_SHIFT;
4201 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
4202 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
4203 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
4204 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
4206 q_vector->cpu = cpu;
4211 static void igb_setup_dca(struct igb_adapter *adapter)
4213 struct e1000_hw *hw = &adapter->hw;
4216 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
4219 /* Always use CB2 mode, difference is masked in the CB driver. */
4220 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
4222 for (i = 0; i < adapter->num_q_vectors; i++) {
4223 struct igb_q_vector *q_vector = adapter->q_vector[i];
4225 igb_update_dca(q_vector);
4229 static int __igb_notify_dca(struct device *dev, void *data)
4231 struct net_device *netdev = dev_get_drvdata(dev);
4232 struct igb_adapter *adapter = netdev_priv(netdev);
4233 struct pci_dev *pdev = adapter->pdev;
4234 struct e1000_hw *hw = &adapter->hw;
4235 unsigned long event = *(unsigned long *)data;
4238 case DCA_PROVIDER_ADD:
4239 /* if already enabled, don't do it again */
4240 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
4242 if (dca_add_requester(dev) == 0) {
4243 adapter->flags |= IGB_FLAG_DCA_ENABLED;
4244 dev_info(&pdev->dev, "DCA enabled\n");
4245 igb_setup_dca(adapter);
4248 /* Fall Through since DCA is disabled. */
4249 case DCA_PROVIDER_REMOVE:
4250 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
4251 /* without this a class_device is left
4252 * hanging around in the sysfs model */
4253 dca_remove_requester(dev);
4254 dev_info(&pdev->dev, "DCA disabled\n");
4255 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
4256 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
4264 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
4269 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
4272 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4274 #endif /* CONFIG_IGB_DCA */
4276 static void igb_ping_all_vfs(struct igb_adapter *adapter)
4278 struct e1000_hw *hw = &adapter->hw;
4282 for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
4283 ping = E1000_PF_CONTROL_MSG;
4284 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
4285 ping |= E1000_VT_MSGTYPE_CTS;
4286 igb_write_mbx(hw, &ping, 1, i);
4290 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
4292 struct e1000_hw *hw = &adapter->hw;
4293 u32 vmolr = rd32(E1000_VMOLR(vf));
4294 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4296 vf_data->flags |= ~(IGB_VF_FLAG_UNI_PROMISC |
4297 IGB_VF_FLAG_MULTI_PROMISC);
4298 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
4300 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
4301 vmolr |= E1000_VMOLR_MPME;
4302 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
4305 * if we have hashes and we are clearing a multicast promisc
4306 * flag we need to write the hashes to the MTA as this step
4307 * was previously skipped
4309 if (vf_data->num_vf_mc_hashes > 30) {
4310 vmolr |= E1000_VMOLR_MPME;
4311 } else if (vf_data->num_vf_mc_hashes) {
4313 vmolr |= E1000_VMOLR_ROMPE;
4314 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
4315 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
4319 wr32(E1000_VMOLR(vf), vmolr);
4321 /* there are flags left unprocessed, likely not supported */
4322 if (*msgbuf & E1000_VT_MSGINFO_MASK)
4329 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
4330 u32 *msgbuf, u32 vf)
4332 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
4333 u16 *hash_list = (u16 *)&msgbuf[1];
4334 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4337 /* salt away the number of multicast addresses assigned
4338 * to this VF for later use to restore when the PF multi cast
4341 vf_data->num_vf_mc_hashes = n;
4343 /* only up to 30 hash values supported */
4347 /* store the hashes for later use */
4348 for (i = 0; i < n; i++)
4349 vf_data->vf_mc_hashes[i] = hash_list[i];
4351 /* Flush and reset the mta with the new values */
4352 igb_set_rx_mode(adapter->netdev);
4357 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
4359 struct e1000_hw *hw = &adapter->hw;
4360 struct vf_data_storage *vf_data;
4363 for (i = 0; i < adapter->vfs_allocated_count; i++) {
4364 u32 vmolr = rd32(E1000_VMOLR(i));
4365 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
4367 vf_data = &adapter->vf_data[i];
4369 if ((vf_data->num_vf_mc_hashes > 30) ||
4370 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
4371 vmolr |= E1000_VMOLR_MPME;
4372 } else if (vf_data->num_vf_mc_hashes) {
4373 vmolr |= E1000_VMOLR_ROMPE;
4374 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
4375 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
4377 wr32(E1000_VMOLR(i), vmolr);
4381 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
4383 struct e1000_hw *hw = &adapter->hw;
4384 u32 pool_mask, reg, vid;
4387 pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
4389 /* Find the vlan filter for this id */
4390 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4391 reg = rd32(E1000_VLVF(i));
4393 /* remove the vf from the pool */
4396 /* if pool is empty then remove entry from vfta */
4397 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
4398 (reg & E1000_VLVF_VLANID_ENABLE)) {
4400 vid = reg & E1000_VLVF_VLANID_MASK;
4401 igb_vfta_set(hw, vid, false);
4404 wr32(E1000_VLVF(i), reg);
4407 adapter->vf_data[vf].vlans_enabled = 0;
4410 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
4412 struct e1000_hw *hw = &adapter->hw;
4415 /* The vlvf table only exists on 82576 hardware and newer */
4416 if (hw->mac.type < e1000_82576)
4419 /* we only need to do this if VMDq is enabled */
4420 if (!adapter->vfs_allocated_count)
4423 /* Find the vlan filter for this id */
4424 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4425 reg = rd32(E1000_VLVF(i));
4426 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
4427 vid == (reg & E1000_VLVF_VLANID_MASK))
4432 if (i == E1000_VLVF_ARRAY_SIZE) {
4433 /* Did not find a matching VLAN ID entry that was
4434 * enabled. Search for a free filter entry, i.e.
4435 * one without the enable bit set
4437 for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
4438 reg = rd32(E1000_VLVF(i));
4439 if (!(reg & E1000_VLVF_VLANID_ENABLE))
4443 if (i < E1000_VLVF_ARRAY_SIZE) {
4444 /* Found an enabled/available entry */
4445 reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
4447 /* if !enabled we need to set this up in vfta */
4448 if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
4449 /* add VID to filter table */
4450 igb_vfta_set(hw, vid, true);
4451 reg |= E1000_VLVF_VLANID_ENABLE;
4453 reg &= ~E1000_VLVF_VLANID_MASK;
4455 wr32(E1000_VLVF(i), reg);
4457 /* do not modify RLPML for PF devices */
4458 if (vf >= adapter->vfs_allocated_count)
4461 if (!adapter->vf_data[vf].vlans_enabled) {
4463 reg = rd32(E1000_VMOLR(vf));
4464 size = reg & E1000_VMOLR_RLPML_MASK;
4466 reg &= ~E1000_VMOLR_RLPML_MASK;
4468 wr32(E1000_VMOLR(vf), reg);
4471 adapter->vf_data[vf].vlans_enabled++;
4475 if (i < E1000_VLVF_ARRAY_SIZE) {
4476 /* remove vf from the pool */
4477 reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
4478 /* if pool is empty then remove entry from vfta */
4479 if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
4481 igb_vfta_set(hw, vid, false);
4483 wr32(E1000_VLVF(i), reg);
4485 /* do not modify RLPML for PF devices */
4486 if (vf >= adapter->vfs_allocated_count)
4489 adapter->vf_data[vf].vlans_enabled--;
4490 if (!adapter->vf_data[vf].vlans_enabled) {
4492 reg = rd32(E1000_VMOLR(vf));
4493 size = reg & E1000_VMOLR_RLPML_MASK;
4495 reg &= ~E1000_VMOLR_RLPML_MASK;
4497 wr32(E1000_VMOLR(vf), reg);
4505 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
4507 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
4508 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
4510 return igb_vlvf_set(adapter, vid, add, vf);
4513 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
4515 /* clear all flags */
4516 adapter->vf_data[vf].flags = 0;
4517 adapter->vf_data[vf].last_nack = jiffies;
4519 /* reset offloads to defaults */
4520 igb_set_vmolr(adapter, vf);
4522 /* reset vlans for device */
4523 igb_clear_vf_vfta(adapter, vf);
4525 /* reset multicast table array for vf */
4526 adapter->vf_data[vf].num_vf_mc_hashes = 0;
4528 /* Flush and reset the mta with the new values */
4529 igb_set_rx_mode(adapter->netdev);
4532 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
4534 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4536 /* generate a new mac address as we were hotplug removed/added */
4537 random_ether_addr(vf_mac);
4539 /* process remaining reset events */
4540 igb_vf_reset(adapter, vf);
4543 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
4545 struct e1000_hw *hw = &adapter->hw;
4546 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
4547 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
4549 u8 *addr = (u8 *)(&msgbuf[1]);
4551 /* process all the same items cleared in a function level reset */
4552 igb_vf_reset(adapter, vf);
4554 /* set vf mac address */
4555 igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
4557 /* enable transmit and receive for vf */
4558 reg = rd32(E1000_VFTE);
4559 wr32(E1000_VFTE, reg | (1 << vf));
4560 reg = rd32(E1000_VFRE);
4561 wr32(E1000_VFRE, reg | (1 << vf));
4563 adapter->vf_data[vf].flags = IGB_VF_FLAG_CTS;
4565 /* reply to reset with ack and vf mac address */
4566 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
4567 memcpy(addr, vf_mac, 6);
4568 igb_write_mbx(hw, msgbuf, 3, vf);
4571 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
4573 unsigned char *addr = (char *)&msg[1];
4576 if (is_valid_ether_addr(addr))
4577 err = igb_set_vf_mac(adapter, vf, addr);
4582 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
4584 struct e1000_hw *hw = &adapter->hw;
4585 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4586 u32 msg = E1000_VT_MSGTYPE_NACK;
4588 /* if device isn't clear to send it shouldn't be reading either */
4589 if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
4590 time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
4591 igb_write_mbx(hw, &msg, 1, vf);
4592 vf_data->last_nack = jiffies;
4596 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
4598 struct pci_dev *pdev = adapter->pdev;
4599 u32 msgbuf[E1000_VFMAILBOX_SIZE];
4600 struct e1000_hw *hw = &adapter->hw;
4601 struct vf_data_storage *vf_data = &adapter->vf_data[vf];
4604 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
4607 /* if receive failed revoke VF CTS stats and restart init */
4608 dev_err(&pdev->dev, "Error receiving message from VF\n");
4609 vf_data->flags &= ~IGB_VF_FLAG_CTS;
4610 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
4615 /* this is a message we already processed, do nothing */
4616 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
4620 * until the vf completes a reset it should not be
4621 * allowed to start any configuration.
4624 if (msgbuf[0] == E1000_VF_RESET) {
4625 igb_vf_reset_msg(adapter, vf);
4629 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
4630 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
4636 switch ((msgbuf[0] & 0xFFFF)) {
4637 case E1000_VF_SET_MAC_ADDR:
4638 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
4640 case E1000_VF_SET_PROMISC:
4641 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
4643 case E1000_VF_SET_MULTICAST:
4644 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
4646 case E1000_VF_SET_LPE:
4647 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
4649 case E1000_VF_SET_VLAN:
4650 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
4653 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
4658 msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
4660 /* notify the VF of the results of what it sent us */
4662 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4664 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
4666 igb_write_mbx(hw, msgbuf, 1, vf);
4669 static void igb_msg_task(struct igb_adapter *adapter)
4671 struct e1000_hw *hw = &adapter->hw;
4674 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
4675 /* process any reset requests */
4676 if (!igb_check_for_rst(hw, vf))
4677 igb_vf_reset_event(adapter, vf);
4679 /* process any messages pending */
4680 if (!igb_check_for_msg(hw, vf))
4681 igb_rcv_msg_from_vf(adapter, vf);
4683 /* process any acks */
4684 if (!igb_check_for_ack(hw, vf))
4685 igb_rcv_ack_from_vf(adapter, vf);
4690 * igb_set_uta - Set unicast filter table address
4691 * @adapter: board private structure
4693 * The unicast table address is a register array of 32-bit registers.
4694 * The table is meant to be used in a way similar to how the MTA is used
4695 * however due to certain limitations in the hardware it is necessary to
4696 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscous
4697 * enable bit to allow vlan tag stripping when promiscous mode is enabled
4699 static void igb_set_uta(struct igb_adapter *adapter)
4701 struct e1000_hw *hw = &adapter->hw;
4704 /* The UTA table only exists on 82576 hardware and newer */
4705 if (hw->mac.type < e1000_82576)
4708 /* we only need to do this if VMDq is enabled */
4709 if (!adapter->vfs_allocated_count)
4712 for (i = 0; i < hw->mac.uta_reg_count; i++)
4713 array_wr32(E1000_UTA, i, ~0);
4717 * igb_intr_msi - Interrupt Handler
4718 * @irq: interrupt number
4719 * @data: pointer to a network interface device structure
4721 static irqreturn_t igb_intr_msi(int irq, void *data)
4723 struct igb_adapter *adapter = data;
4724 struct igb_q_vector *q_vector = adapter->q_vector[0];
4725 struct e1000_hw *hw = &adapter->hw;
4726 /* read ICR disables interrupts using IAM */
4727 u32 icr = rd32(E1000_ICR);
4729 igb_write_itr(q_vector);
4731 if (icr & E1000_ICR_DOUTSYNC) {
4732 /* HW is reporting DMA is out of sync */
4733 adapter->stats.doosync++;
4736 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4737 hw->mac.get_link_status = 1;
4738 if (!test_bit(__IGB_DOWN, &adapter->state))
4739 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4742 napi_schedule(&q_vector->napi);
4748 * igb_intr - Legacy Interrupt Handler
4749 * @irq: interrupt number
4750 * @data: pointer to a network interface device structure
4752 static irqreturn_t igb_intr(int irq, void *data)
4754 struct igb_adapter *adapter = data;
4755 struct igb_q_vector *q_vector = adapter->q_vector[0];
4756 struct e1000_hw *hw = &adapter->hw;
4757 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
4758 * need for the IMC write */
4759 u32 icr = rd32(E1000_ICR);
4761 return IRQ_NONE; /* Not our interrupt */
4763 igb_write_itr(q_vector);
4765 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
4766 * not set, then the adapter didn't send an interrupt */
4767 if (!(icr & E1000_ICR_INT_ASSERTED))
4770 if (icr & E1000_ICR_DOUTSYNC) {
4771 /* HW is reporting DMA is out of sync */
4772 adapter->stats.doosync++;
4775 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4776 hw->mac.get_link_status = 1;
4777 /* guard against interrupt when we're going down */
4778 if (!test_bit(__IGB_DOWN, &adapter->state))
4779 mod_timer(&adapter->watchdog_timer, jiffies + 1);
4782 napi_schedule(&q_vector->napi);
4787 static inline void igb_ring_irq_enable(struct igb_q_vector *q_vector)
4789 struct igb_adapter *adapter = q_vector->adapter;
4790 struct e1000_hw *hw = &adapter->hw;
4792 if ((q_vector->rx_ring && (adapter->rx_itr_setting & 3)) ||
4793 (!q_vector->rx_ring && (adapter->tx_itr_setting & 3))) {
4794 if (!adapter->msix_entries)
4795 igb_set_itr(adapter);
4797 igb_update_ring_itr(q_vector);
4800 if (!test_bit(__IGB_DOWN, &adapter->state)) {
4801 if (adapter->msix_entries)
4802 wr32(E1000_EIMS, q_vector->eims_value);
4804 igb_irq_enable(adapter);
4809 * igb_poll - NAPI Rx polling callback
4810 * @napi: napi polling structure
4811 * @budget: count of how many packets we should handle
4813 static int igb_poll(struct napi_struct *napi, int budget)
4815 struct igb_q_vector *q_vector = container_of(napi,
4816 struct igb_q_vector,
4818 int tx_clean_complete = 1, work_done = 0;
4820 #ifdef CONFIG_IGB_DCA
4821 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
4822 igb_update_dca(q_vector);
4824 if (q_vector->tx_ring)
4825 tx_clean_complete = igb_clean_tx_irq(q_vector);
4827 if (q_vector->rx_ring)
4828 igb_clean_rx_irq_adv(q_vector, &work_done, budget);
4830 if (!tx_clean_complete)
4833 /* If not enough Rx work done, exit the polling mode */
4834 if (work_done < budget) {
4835 napi_complete(napi);
4836 igb_ring_irq_enable(q_vector);
4843 * igb_systim_to_hwtstamp - convert system time value to hw timestamp
4844 * @adapter: board private structure
4845 * @shhwtstamps: timestamp structure to update
4846 * @regval: unsigned 64bit system time value.
4848 * We need to convert the system time value stored in the RX/TXSTMP registers
4849 * into a hwtstamp which can be used by the upper level timestamping functions
4851 static void igb_systim_to_hwtstamp(struct igb_adapter *adapter,
4852 struct skb_shared_hwtstamps *shhwtstamps,
4858 * The 82580 starts with 1ns at bit 0 in RX/TXSTMPL, shift this up to
4859 * 24 to match clock shift we setup earlier.
4861 if (adapter->hw.mac.type == e1000_82580)
4862 regval <<= IGB_82580_TSYNC_SHIFT;
4864 ns = timecounter_cyc2time(&adapter->clock, regval);
4865 timecompare_update(&adapter->compare, ns);
4866 memset(shhwtstamps, 0, sizeof(struct skb_shared_hwtstamps));
4867 shhwtstamps->hwtstamp = ns_to_ktime(ns);
4868 shhwtstamps->syststamp = timecompare_transform(&adapter->compare, ns);
4872 * igb_tx_hwtstamp - utility function which checks for TX time stamp
4873 * @q_vector: pointer to q_vector containing needed info
4874 * @skb: packet that was just sent
4876 * If we were asked to do hardware stamping and such a time stamp is
4877 * available, then it must have been for this skb here because we only
4878 * allow only one such packet into the queue.
4880 static void igb_tx_hwtstamp(struct igb_q_vector *q_vector, struct sk_buff *skb)
4882 struct igb_adapter *adapter = q_vector->adapter;
4883 union skb_shared_tx *shtx = skb_tx(skb);
4884 struct e1000_hw *hw = &adapter->hw;
4885 struct skb_shared_hwtstamps shhwtstamps;
4888 /* if skb does not support hw timestamp or TX stamp not valid exit */
4889 if (likely(!shtx->hardware) ||
4890 !(rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID))
4893 regval = rd32(E1000_TXSTMPL);
4894 regval |= (u64)rd32(E1000_TXSTMPH) << 32;
4896 igb_systim_to_hwtstamp(adapter, &shhwtstamps, regval);
4897 skb_tstamp_tx(skb, &shhwtstamps);
4901 * igb_clean_tx_irq - Reclaim resources after transmit completes
4902 * @q_vector: pointer to q_vector containing needed info
4903 * returns true if ring is completely cleaned
4905 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
4907 struct igb_adapter *adapter = q_vector->adapter;
4908 struct igb_ring *tx_ring = q_vector->tx_ring;
4909 struct net_device *netdev = tx_ring->netdev;
4910 struct e1000_hw *hw = &adapter->hw;
4911 struct igb_buffer *buffer_info;
4912 struct sk_buff *skb;
4913 union e1000_adv_tx_desc *tx_desc, *eop_desc;
4914 unsigned int total_bytes = 0, total_packets = 0;
4915 unsigned int i, eop, count = 0;
4916 bool cleaned = false;
4918 i = tx_ring->next_to_clean;
4919 eop = tx_ring->buffer_info[i].next_to_watch;
4920 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4922 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
4923 (count < tx_ring->count)) {
4924 for (cleaned = false; !cleaned; count++) {
4925 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
4926 buffer_info = &tx_ring->buffer_info[i];
4927 cleaned = (i == eop);
4928 skb = buffer_info->skb;
4931 unsigned int segs, bytecount;
4932 /* gso_segs is currently only valid for tcp */
4933 segs = skb_shinfo(skb)->gso_segs ?: 1;
4934 /* multiply data chunks by size of headers */
4935 bytecount = ((segs - 1) * skb_headlen(skb)) +
4937 total_packets += segs;
4938 total_bytes += bytecount;
4940 igb_tx_hwtstamp(q_vector, skb);
4943 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
4944 tx_desc->wb.status = 0;
4947 if (i == tx_ring->count)
4950 eop = tx_ring->buffer_info[i].next_to_watch;
4951 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4954 tx_ring->next_to_clean = i;
4956 if (unlikely(count &&
4957 netif_carrier_ok(netdev) &&
4958 igb_desc_unused(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
4959 /* Make sure that anybody stopping the queue after this
4960 * sees the new next_to_clean.
4963 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
4964 !(test_bit(__IGB_DOWN, &adapter->state))) {
4965 netif_wake_subqueue(netdev, tx_ring->queue_index);
4966 tx_ring->tx_stats.restart_queue++;
4970 if (tx_ring->detect_tx_hung) {
4971 /* Detect a transmit hang in hardware, this serializes the
4972 * check with the clearing of time_stamp and movement of i */
4973 tx_ring->detect_tx_hung = false;
4974 if (tx_ring->buffer_info[i].time_stamp &&
4975 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
4976 (adapter->tx_timeout_factor * HZ)) &&
4977 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
4979 /* detected Tx unit hang */
4980 dev_err(&tx_ring->pdev->dev,
4981 "Detected Tx Unit Hang\n"
4985 " next_to_use <%x>\n"
4986 " next_to_clean <%x>\n"
4987 "buffer_info[next_to_clean]\n"
4988 " time_stamp <%lx>\n"
4989 " next_to_watch <%x>\n"
4991 " desc.status <%x>\n",
4992 tx_ring->queue_index,
4993 readl(tx_ring->head),
4994 readl(tx_ring->tail),
4995 tx_ring->next_to_use,
4996 tx_ring->next_to_clean,
4997 tx_ring->buffer_info[eop].time_stamp,
5000 eop_desc->wb.status);
5001 netif_stop_subqueue(netdev, tx_ring->queue_index);
5004 tx_ring->total_bytes += total_bytes;
5005 tx_ring->total_packets += total_packets;
5006 tx_ring->tx_stats.bytes += total_bytes;
5007 tx_ring->tx_stats.packets += total_packets;
5008 return (count < tx_ring->count);
5012 * igb_receive_skb - helper function to handle rx indications
5013 * @q_vector: structure containing interrupt and ring information
5014 * @skb: packet to send up
5015 * @vlan_tag: vlan tag for packet
5017 static void igb_receive_skb(struct igb_q_vector *q_vector,
5018 struct sk_buff *skb,
5021 struct igb_adapter *adapter = q_vector->adapter;
5024 vlan_gro_receive(&q_vector->napi, adapter->vlgrp,
5027 napi_gro_receive(&q_vector->napi, skb);
5030 static inline void igb_rx_checksum_adv(struct igb_ring *ring,
5031 u32 status_err, struct sk_buff *skb)
5033 skb->ip_summed = CHECKSUM_NONE;
5035 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
5036 if (!(ring->flags & IGB_RING_FLAG_RX_CSUM) ||
5037 (status_err & E1000_RXD_STAT_IXSM))
5040 /* TCP/UDP checksum error bit is set */
5042 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
5044 * work around errata with sctp packets where the TCPE aka
5045 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
5046 * packets, (aka let the stack check the crc32c)
5048 if ((skb->len == 60) &&
5049 (ring->flags & IGB_RING_FLAG_RX_SCTP_CSUM))
5050 ring->rx_stats.csum_err++;
5052 /* let the stack verify checksum errors */
5055 /* It must be a TCP or UDP packet with a valid checksum */
5056 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
5057 skb->ip_summed = CHECKSUM_UNNECESSARY;
5059 dev_dbg(&ring->pdev->dev, "cksum success: bits %08X\n", status_err);
5062 static inline void igb_rx_hwtstamp(struct igb_q_vector *q_vector, u32 staterr,
5063 struct sk_buff *skb)
5065 struct igb_adapter *adapter = q_vector->adapter;
5066 struct e1000_hw *hw = &adapter->hw;
5070 * If this bit is set, then the RX registers contain the time stamp. No
5071 * other packet will be time stamped until we read these registers, so
5072 * read the registers to make them available again. Because only one
5073 * packet can be time stamped at a time, we know that the register
5074 * values must belong to this one here and therefore we don't need to
5075 * compare any of the additional attributes stored for it.
5077 * If nothing went wrong, then it should have a skb_shared_tx that we
5078 * can turn into a skb_shared_hwtstamps.
5080 if (likely(!(staterr & E1000_RXDADV_STAT_TS)))
5082 if (!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID))
5085 regval = rd32(E1000_RXSTMPL);
5086 regval |= (u64)rd32(E1000_RXSTMPH) << 32;
5088 igb_systim_to_hwtstamp(adapter, skb_hwtstamps(skb), regval);
5090 static inline u16 igb_get_hlen(struct igb_ring *rx_ring,
5091 union e1000_adv_rx_desc *rx_desc)
5093 /* HW will not DMA in data larger than the given buffer, even if it
5094 * parses the (NFS, of course) header to be larger. In that case, it
5095 * fills the header buffer and spills the rest into the page.
5097 u16 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
5098 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
5099 if (hlen > rx_ring->rx_buffer_len)
5100 hlen = rx_ring->rx_buffer_len;
5104 static bool igb_clean_rx_irq_adv(struct igb_q_vector *q_vector,
5105 int *work_done, int budget)
5107 struct igb_ring *rx_ring = q_vector->rx_ring;
5108 struct net_device *netdev = rx_ring->netdev;
5109 struct pci_dev *pdev = rx_ring->pdev;
5110 union e1000_adv_rx_desc *rx_desc , *next_rxd;
5111 struct igb_buffer *buffer_info , *next_buffer;
5112 struct sk_buff *skb;
5113 bool cleaned = false;
5114 int cleaned_count = 0;
5115 int current_node = numa_node_id();
5116 unsigned int total_bytes = 0, total_packets = 0;
5122 i = rx_ring->next_to_clean;
5123 buffer_info = &rx_ring->buffer_info[i];
5124 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
5125 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
5127 while (staterr & E1000_RXD_STAT_DD) {
5128 if (*work_done >= budget)
5132 skb = buffer_info->skb;
5133 prefetch(skb->data - NET_IP_ALIGN);
5134 buffer_info->skb = NULL;
5137 if (i == rx_ring->count)
5140 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
5142 next_buffer = &rx_ring->buffer_info[i];
5144 length = le16_to_cpu(rx_desc->wb.upper.length);
5148 if (buffer_info->dma) {
5149 pci_unmap_single(pdev, buffer_info->dma,
5150 rx_ring->rx_buffer_len,
5151 PCI_DMA_FROMDEVICE);
5152 buffer_info->dma = 0;
5153 if (rx_ring->rx_buffer_len >= IGB_RXBUFFER_1024) {
5154 skb_put(skb, length);
5157 skb_put(skb, igb_get_hlen(rx_ring, rx_desc));
5161 pci_unmap_page(pdev, buffer_info->page_dma,
5162 PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
5163 buffer_info->page_dma = 0;
5165 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
5167 buffer_info->page_offset,
5170 if ((page_count(buffer_info->page) != 1) ||
5171 (page_to_nid(buffer_info->page) != current_node))
5172 buffer_info->page = NULL;
5174 get_page(buffer_info->page);
5177 skb->data_len += length;
5178 skb->truesize += length;
5181 if (!(staterr & E1000_RXD_STAT_EOP)) {
5182 buffer_info->skb = next_buffer->skb;
5183 buffer_info->dma = next_buffer->dma;
5184 next_buffer->skb = skb;
5185 next_buffer->dma = 0;
5189 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
5190 dev_kfree_skb_irq(skb);
5194 igb_rx_hwtstamp(q_vector, staterr, skb);
5195 total_bytes += skb->len;
5198 igb_rx_checksum_adv(rx_ring, staterr, skb);
5200 skb->protocol = eth_type_trans(skb, netdev);
5201 skb_record_rx_queue(skb, rx_ring->queue_index);
5203 vlan_tag = ((staterr & E1000_RXD_STAT_VP) ?
5204 le16_to_cpu(rx_desc->wb.upper.vlan) : 0);
5206 igb_receive_skb(q_vector, skb, vlan_tag);
5209 rx_desc->wb.upper.status_error = 0;
5211 /* return some buffers to hardware, one at a time is too slow */
5212 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
5213 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
5217 /* use prefetched values */
5219 buffer_info = next_buffer;
5220 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
5223 rx_ring->next_to_clean = i;
5224 cleaned_count = igb_desc_unused(rx_ring);
5227 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
5229 rx_ring->total_packets += total_packets;
5230 rx_ring->total_bytes += total_bytes;
5231 rx_ring->rx_stats.packets += total_packets;
5232 rx_ring->rx_stats.bytes += total_bytes;
5237 * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
5238 * @adapter: address of board private structure
5240 void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring, int cleaned_count)
5242 struct net_device *netdev = rx_ring->netdev;
5243 union e1000_adv_rx_desc *rx_desc;
5244 struct igb_buffer *buffer_info;
5245 struct sk_buff *skb;
5249 i = rx_ring->next_to_use;
5250 buffer_info = &rx_ring->buffer_info[i];
5252 bufsz = rx_ring->rx_buffer_len;
5254 while (cleaned_count--) {
5255 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
5257 if ((bufsz < IGB_RXBUFFER_1024) && !buffer_info->page_dma) {
5258 if (!buffer_info->page) {
5259 buffer_info->page = netdev_alloc_page(netdev);
5260 if (!buffer_info->page) {
5261 rx_ring->rx_stats.alloc_failed++;
5264 buffer_info->page_offset = 0;
5266 buffer_info->page_offset ^= PAGE_SIZE / 2;
5268 buffer_info->page_dma =
5269 pci_map_page(rx_ring->pdev, buffer_info->page,
5270 buffer_info->page_offset,
5272 PCI_DMA_FROMDEVICE);
5273 if (pci_dma_mapping_error(rx_ring->pdev,
5274 buffer_info->page_dma)) {
5275 buffer_info->page_dma = 0;
5276 rx_ring->rx_stats.alloc_failed++;
5281 skb = buffer_info->skb;
5283 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
5285 rx_ring->rx_stats.alloc_failed++;
5289 buffer_info->skb = skb;
5291 if (!buffer_info->dma) {
5292 buffer_info->dma = pci_map_single(rx_ring->pdev,
5295 PCI_DMA_FROMDEVICE);
5296 if (pci_dma_mapping_error(rx_ring->pdev,
5297 buffer_info->dma)) {
5298 buffer_info->dma = 0;
5299 rx_ring->rx_stats.alloc_failed++;
5303 /* Refresh the desc even if buffer_addrs didn't change because
5304 * each write-back erases this info. */
5305 if (bufsz < IGB_RXBUFFER_1024) {
5306 rx_desc->read.pkt_addr =
5307 cpu_to_le64(buffer_info->page_dma);
5308 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
5310 rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma);
5311 rx_desc->read.hdr_addr = 0;
5315 if (i == rx_ring->count)
5317 buffer_info = &rx_ring->buffer_info[i];
5321 if (rx_ring->next_to_use != i) {
5322 rx_ring->next_to_use = i;
5324 i = (rx_ring->count - 1);
5328 /* Force memory writes to complete before letting h/w
5329 * know there are new descriptors to fetch. (Only
5330 * applicable for weak-ordered memory model archs,
5331 * such as IA-64). */
5333 writel(i, rx_ring->tail);
5343 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5345 struct igb_adapter *adapter = netdev_priv(netdev);
5346 struct mii_ioctl_data *data = if_mii(ifr);
5348 if (adapter->hw.phy.media_type != e1000_media_type_copper)
5353 data->phy_id = adapter->hw.phy.addr;
5356 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
5368 * igb_hwtstamp_ioctl - control hardware time stamping
5373 * Outgoing time stamping can be enabled and disabled. Play nice and
5374 * disable it when requested, although it shouldn't case any overhead
5375 * when no packet needs it. At most one packet in the queue may be
5376 * marked for time stamping, otherwise it would be impossible to tell
5377 * for sure to which packet the hardware time stamp belongs.
5379 * Incoming time stamping has to be configured via the hardware
5380 * filters. Not all combinations are supported, in particular event
5381 * type has to be specified. Matching the kind of event packet is
5382 * not supported, with the exception of "all V2 events regardless of
5386 static int igb_hwtstamp_ioctl(struct net_device *netdev,
5387 struct ifreq *ifr, int cmd)
5389 struct igb_adapter *adapter = netdev_priv(netdev);
5390 struct e1000_hw *hw = &adapter->hw;
5391 struct hwtstamp_config config;
5392 u32 tsync_tx_ctl = E1000_TSYNCTXCTL_ENABLED;
5393 u32 tsync_rx_ctl = E1000_TSYNCRXCTL_ENABLED;
5394 u32 tsync_rx_cfg = 0;
5399 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
5402 /* reserved for future extensions */
5406 switch (config.tx_type) {
5407 case HWTSTAMP_TX_OFF:
5409 case HWTSTAMP_TX_ON:
5415 switch (config.rx_filter) {
5416 case HWTSTAMP_FILTER_NONE:
5419 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
5420 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
5421 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
5422 case HWTSTAMP_FILTER_ALL:
5424 * register TSYNCRXCFG must be set, therefore it is not
5425 * possible to time stamp both Sync and Delay_Req messages
5426 * => fall back to time stamping all packets
5428 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_ALL;
5429 config.rx_filter = HWTSTAMP_FILTER_ALL;
5431 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
5432 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
5433 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
5436 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
5437 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L4_V1;
5438 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
5441 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
5442 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
5443 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
5444 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
5447 config.rx_filter = HWTSTAMP_FILTER_SOME;
5449 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
5450 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
5451 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
5452 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
5455 config.rx_filter = HWTSTAMP_FILTER_SOME;
5457 case HWTSTAMP_FILTER_PTP_V2_EVENT:
5458 case HWTSTAMP_FILTER_PTP_V2_SYNC:
5459 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
5460 tsync_rx_ctl |= E1000_TSYNCRXCTL_TYPE_EVENT_V2;
5461 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
5468 if (hw->mac.type == e1000_82575) {
5469 if (tsync_rx_ctl | tsync_tx_ctl)
5474 /* enable/disable TX */
5475 regval = rd32(E1000_TSYNCTXCTL);
5476 regval &= ~E1000_TSYNCTXCTL_ENABLED;
5477 regval |= tsync_tx_ctl;
5478 wr32(E1000_TSYNCTXCTL, regval);
5480 /* enable/disable RX */
5481 regval = rd32(E1000_TSYNCRXCTL);
5482 regval &= ~(E1000_TSYNCRXCTL_ENABLED | E1000_TSYNCRXCTL_TYPE_MASK);
5483 regval |= tsync_rx_ctl;
5484 wr32(E1000_TSYNCRXCTL, regval);
5486 /* define which PTP packets are time stamped */
5487 wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
5489 /* define ethertype filter for timestamped packets */
5492 (E1000_ETQF_FILTER_ENABLE | /* enable filter */
5493 E1000_ETQF_1588 | /* enable timestamping */
5494 ETH_P_1588)); /* 1588 eth protocol type */
5496 wr32(E1000_ETQF(3), 0);
5498 #define PTP_PORT 319
5499 /* L4 Queue Filter[3]: filter by destination port and protocol */
5501 u32 ftqf = (IPPROTO_UDP /* UDP */
5502 | E1000_FTQF_VF_BP /* VF not compared */
5503 | E1000_FTQF_1588_TIME_STAMP /* Enable Timestamping */
5504 | E1000_FTQF_MASK); /* mask all inputs */
5505 ftqf &= ~E1000_FTQF_MASK_PROTO_BP; /* enable protocol check */
5507 wr32(E1000_IMIR(3), htons(PTP_PORT));
5508 wr32(E1000_IMIREXT(3),
5509 (E1000_IMIREXT_SIZE_BP | E1000_IMIREXT_CTRL_BP));
5510 if (hw->mac.type == e1000_82576) {
5511 /* enable source port check */
5512 wr32(E1000_SPQF(3), htons(PTP_PORT));
5513 ftqf &= ~E1000_FTQF_MASK_SOURCE_PORT_BP;
5515 wr32(E1000_FTQF(3), ftqf);
5517 wr32(E1000_FTQF(3), E1000_FTQF_MASK);
5521 adapter->hwtstamp_config = config;
5523 /* clear TX/RX time stamp registers, just to be sure */
5524 regval = rd32(E1000_TXSTMPH);
5525 regval = rd32(E1000_RXSTMPH);
5527 return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
5537 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5543 return igb_mii_ioctl(netdev, ifr, cmd);
5545 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
5551 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
5553 struct igb_adapter *adapter = hw->back;
5556 cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
5558 return -E1000_ERR_CONFIG;
5560 pci_read_config_word(adapter->pdev, cap_offset + reg, value);
5565 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
5567 struct igb_adapter *adapter = hw->back;
5570 cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
5572 return -E1000_ERR_CONFIG;
5574 pci_write_config_word(adapter->pdev, cap_offset + reg, *value);
5579 static void igb_vlan_rx_register(struct net_device *netdev,
5580 struct vlan_group *grp)
5582 struct igb_adapter *adapter = netdev_priv(netdev);
5583 struct e1000_hw *hw = &adapter->hw;
5586 igb_irq_disable(adapter);
5587 adapter->vlgrp = grp;
5590 /* enable VLAN tag insert/strip */
5591 ctrl = rd32(E1000_CTRL);
5592 ctrl |= E1000_CTRL_VME;
5593 wr32(E1000_CTRL, ctrl);
5595 /* Disable CFI check */
5596 rctl = rd32(E1000_RCTL);
5597 rctl &= ~E1000_RCTL_CFIEN;
5598 wr32(E1000_RCTL, rctl);
5600 /* disable VLAN tag insert/strip */
5601 ctrl = rd32(E1000_CTRL);
5602 ctrl &= ~E1000_CTRL_VME;
5603 wr32(E1000_CTRL, ctrl);
5606 igb_rlpml_set(adapter);
5608 if (!test_bit(__IGB_DOWN, &adapter->state))
5609 igb_irq_enable(adapter);
5612 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
5614 struct igb_adapter *adapter = netdev_priv(netdev);
5615 struct e1000_hw *hw = &adapter->hw;
5616 int pf_id = adapter->vfs_allocated_count;
5618 /* attempt to add filter to vlvf array */
5619 igb_vlvf_set(adapter, vid, true, pf_id);
5621 /* add the filter since PF can receive vlans w/o entry in vlvf */
5622 igb_vfta_set(hw, vid, true);
5625 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
5627 struct igb_adapter *adapter = netdev_priv(netdev);
5628 struct e1000_hw *hw = &adapter->hw;
5629 int pf_id = adapter->vfs_allocated_count;
5632 igb_irq_disable(adapter);
5633 vlan_group_set_device(adapter->vlgrp, vid, NULL);
5635 if (!test_bit(__IGB_DOWN, &adapter->state))
5636 igb_irq_enable(adapter);
5638 /* remove vlan from VLVF table array */
5639 err = igb_vlvf_set(adapter, vid, false, pf_id);
5641 /* if vid was not present in VLVF just remove it from table */
5643 igb_vfta_set(hw, vid, false);
5646 static void igb_restore_vlan(struct igb_adapter *adapter)
5648 igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5650 if (adapter->vlgrp) {
5652 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5653 if (!vlan_group_get_device(adapter->vlgrp, vid))
5655 igb_vlan_rx_add_vid(adapter->netdev, vid);
5660 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
5662 struct pci_dev *pdev = adapter->pdev;
5663 struct e1000_mac_info *mac = &adapter->hw.mac;
5668 case SPEED_10 + DUPLEX_HALF:
5669 mac->forced_speed_duplex = ADVERTISE_10_HALF;
5671 case SPEED_10 + DUPLEX_FULL:
5672 mac->forced_speed_duplex = ADVERTISE_10_FULL;
5674 case SPEED_100 + DUPLEX_HALF:
5675 mac->forced_speed_duplex = ADVERTISE_100_HALF;
5677 case SPEED_100 + DUPLEX_FULL:
5678 mac->forced_speed_duplex = ADVERTISE_100_FULL;
5680 case SPEED_1000 + DUPLEX_FULL:
5682 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
5684 case SPEED_1000 + DUPLEX_HALF: /* not supported */
5686 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
5692 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
5694 struct net_device *netdev = pci_get_drvdata(pdev);
5695 struct igb_adapter *adapter = netdev_priv(netdev);
5696 struct e1000_hw *hw = &adapter->hw;
5697 u32 ctrl, rctl, status;
5698 u32 wufc = adapter->wol;
5703 netif_device_detach(netdev);
5705 if (netif_running(netdev))
5708 igb_clear_interrupt_scheme(adapter);
5711 retval = pci_save_state(pdev);
5716 status = rd32(E1000_STATUS);
5717 if (status & E1000_STATUS_LU)
5718 wufc &= ~E1000_WUFC_LNKC;
5721 igb_setup_rctl(adapter);
5722 igb_set_rx_mode(netdev);
5724 /* turn on all-multi mode if wake on multicast is enabled */
5725 if (wufc & E1000_WUFC_MC) {
5726 rctl = rd32(E1000_RCTL);
5727 rctl |= E1000_RCTL_MPE;
5728 wr32(E1000_RCTL, rctl);
5731 ctrl = rd32(E1000_CTRL);
5732 /* advertise wake from D3Cold */
5733 #define E1000_CTRL_ADVD3WUC 0x00100000
5734 /* phy power management enable */
5735 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5736 ctrl |= E1000_CTRL_ADVD3WUC;
5737 wr32(E1000_CTRL, ctrl);
5739 /* Allow time for pending master requests to run */
5740 igb_disable_pcie_master(hw);
5742 wr32(E1000_WUC, E1000_WUC_PME_EN);
5743 wr32(E1000_WUFC, wufc);
5746 wr32(E1000_WUFC, 0);
5749 *enable_wake = wufc || adapter->en_mng_pt;
5751 igb_shutdown_serdes_link_82575(hw);
5753 /* Release control of h/w to f/w. If f/w is AMT enabled, this
5754 * would have already happened in close and is redundant. */
5755 igb_release_hw_control(adapter);
5757 pci_disable_device(pdev);
5763 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
5768 retval = __igb_shutdown(pdev, &wake);
5773 pci_prepare_to_sleep(pdev);
5775 pci_wake_from_d3(pdev, false);
5776 pci_set_power_state(pdev, PCI_D3hot);
5782 static int igb_resume(struct pci_dev *pdev)
5784 struct net_device *netdev = pci_get_drvdata(pdev);
5785 struct igb_adapter *adapter = netdev_priv(netdev);
5786 struct e1000_hw *hw = &adapter->hw;
5789 pci_set_power_state(pdev, PCI_D0);
5790 pci_restore_state(pdev);
5792 err = pci_enable_device_mem(pdev);
5795 "igb: Cannot enable PCI device from suspend\n");
5798 pci_set_master(pdev);
5800 pci_enable_wake(pdev, PCI_D3hot, 0);
5801 pci_enable_wake(pdev, PCI_D3cold, 0);
5803 if (igb_init_interrupt_scheme(adapter)) {
5804 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5808 /* e1000_power_up_phy(adapter); */
5812 /* let the f/w know that the h/w is now under the control of the
5814 igb_get_hw_control(adapter);
5816 wr32(E1000_WUS, ~0);
5818 if (netif_running(netdev)) {
5819 err = igb_open(netdev);
5824 netif_device_attach(netdev);
5830 static void igb_shutdown(struct pci_dev *pdev)
5834 __igb_shutdown(pdev, &wake);
5836 if (system_state == SYSTEM_POWER_OFF) {
5837 pci_wake_from_d3(pdev, wake);
5838 pci_set_power_state(pdev, PCI_D3hot);
5842 #ifdef CONFIG_NET_POLL_CONTROLLER
5844 * Polling 'interrupt' - used by things like netconsole to send skbs
5845 * without having to re-enable interrupts. It's not called while
5846 * the interrupt routine is executing.
5848 static void igb_netpoll(struct net_device *netdev)
5850 struct igb_adapter *adapter = netdev_priv(netdev);
5851 struct e1000_hw *hw = &adapter->hw;
5854 if (!adapter->msix_entries) {
5855 struct igb_q_vector *q_vector = adapter->q_vector[0];
5856 igb_irq_disable(adapter);
5857 napi_schedule(&q_vector->napi);
5861 for (i = 0; i < adapter->num_q_vectors; i++) {
5862 struct igb_q_vector *q_vector = adapter->q_vector[i];
5863 wr32(E1000_EIMC, q_vector->eims_value);
5864 napi_schedule(&q_vector->napi);
5867 #endif /* CONFIG_NET_POLL_CONTROLLER */
5870 * igb_io_error_detected - called when PCI error is detected
5871 * @pdev: Pointer to PCI device
5872 * @state: The current pci connection state
5874 * This function is called after a PCI bus error affecting
5875 * this device has been detected.
5877 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
5878 pci_channel_state_t state)
5880 struct net_device *netdev = pci_get_drvdata(pdev);
5881 struct igb_adapter *adapter = netdev_priv(netdev);
5883 netif_device_detach(netdev);
5885 if (state == pci_channel_io_perm_failure)
5886 return PCI_ERS_RESULT_DISCONNECT;
5888 if (netif_running(netdev))
5890 pci_disable_device(pdev);
5892 /* Request a slot slot reset. */
5893 return PCI_ERS_RESULT_NEED_RESET;
5897 * igb_io_slot_reset - called after the pci bus has been reset.
5898 * @pdev: Pointer to PCI device
5900 * Restart the card from scratch, as if from a cold-boot. Implementation
5901 * resembles the first-half of the igb_resume routine.
5903 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
5905 struct net_device *netdev = pci_get_drvdata(pdev);
5906 struct igb_adapter *adapter = netdev_priv(netdev);
5907 struct e1000_hw *hw = &adapter->hw;
5908 pci_ers_result_t result;
5911 if (pci_enable_device_mem(pdev)) {
5913 "Cannot re-enable PCI device after reset.\n");
5914 result = PCI_ERS_RESULT_DISCONNECT;
5916 pci_set_master(pdev);
5917 pci_restore_state(pdev);
5919 pci_enable_wake(pdev, PCI_D3hot, 0);
5920 pci_enable_wake(pdev, PCI_D3cold, 0);
5923 wr32(E1000_WUS, ~0);
5924 result = PCI_ERS_RESULT_RECOVERED;
5927 err = pci_cleanup_aer_uncorrect_error_status(pdev);
5929 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
5930 "failed 0x%0x\n", err);
5931 /* non-fatal, continue */
5938 * igb_io_resume - called when traffic can start flowing again.
5939 * @pdev: Pointer to PCI device
5941 * This callback is called when the error recovery driver tells us that
5942 * its OK to resume normal operation. Implementation resembles the
5943 * second-half of the igb_resume routine.
5945 static void igb_io_resume(struct pci_dev *pdev)
5947 struct net_device *netdev = pci_get_drvdata(pdev);
5948 struct igb_adapter *adapter = netdev_priv(netdev);
5950 if (netif_running(netdev)) {
5951 if (igb_up(adapter)) {
5952 dev_err(&pdev->dev, "igb_up failed after reset\n");
5957 netif_device_attach(netdev);
5959 /* let the f/w know that the h/w is now under the control of the
5961 igb_get_hw_control(adapter);
5964 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
5967 u32 rar_low, rar_high;
5968 struct e1000_hw *hw = &adapter->hw;
5970 /* HW expects these in little endian so we reverse the byte order
5971 * from network order (big endian) to little endian
5973 rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
5974 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
5975 rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
5977 /* Indicate to hardware the Address is Valid. */
5978 rar_high |= E1000_RAH_AV;
5980 if (hw->mac.type == e1000_82575)
5981 rar_high |= E1000_RAH_POOL_1 * qsel;
5983 rar_high |= E1000_RAH_POOL_1 << qsel;
5985 wr32(E1000_RAL(index), rar_low);
5987 wr32(E1000_RAH(index), rar_high);
5991 static int igb_set_vf_mac(struct igb_adapter *adapter,
5992 int vf, unsigned char *mac_addr)
5994 struct e1000_hw *hw = &adapter->hw;
5995 /* VF MAC addresses start at end of receive addresses and moves
5996 * torwards the first, as a result a collision should not be possible */
5997 int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5999 memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
6001 igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
6006 static void igb_vmm_control(struct igb_adapter *adapter)
6008 struct e1000_hw *hw = &adapter->hw;
6011 /* replication is not supported for 82575 */
6012 if (hw->mac.type == e1000_82575)
6015 /* enable replication vlan tag stripping */
6016 reg = rd32(E1000_RPLOLR);
6017 reg |= E1000_RPLOLR_STRVLAN;
6018 wr32(E1000_RPLOLR, reg);
6020 /* notify HW that the MAC is adding vlan tags */
6021 reg = rd32(E1000_DTXCTL);
6022 reg |= E1000_DTXCTL_VLAN_ADDED;
6023 wr32(E1000_DTXCTL, reg);
6025 if (adapter->vfs_allocated_count) {
6026 igb_vmdq_set_loopback_pf(hw, true);
6027 igb_vmdq_set_replication_pf(hw, true);
6029 igb_vmdq_set_loopback_pf(hw, false);
6030 igb_vmdq_set_replication_pf(hw, false);